Dual, HLA-B27 Subtype-dependent Conformation of a Self-peptide

MHC class I antigens are cell surface glycoproteins consisting of a polymorphic transmembrane heavy chain (HC) noncovalently bound to β₂-microglobulin (β₂m; reference 1). The HC forms a groove with six pockets (A–F) for binding of proteolytic fragments derived from self- and nonself-proteins (2, 3). These peptides are presented to TCRs located on CTLs and initiate a cellular immune response leading to the destruction of the antigen-presenting cell if the peptide is derived from a nonself-protein. Several MHC class I alleles are associated with autoimmune diseases, and allele-specific binding of foreign or self-peptides by MHC class I molecules may be crucial to disease pathogenesis (4–8).

The association of the HLA class I allele HLA-B27 with ankylosing spondylitis (AS) and related spondyloarthropathies is among the strongest observed for any HLA gene (9, 10). Population analyses have shown that most of the common HLA-B27 subtypes, including the frequent B*2705 allele, are disease associated, whereas two subtypes, B*2706 and B*2709, are not (11). The B*2705 protein differs in only a single HC amino acid (Asp116) from B*2709 (His116) (12). This residue is located at the floor of the F pocket, which accommodates the COOH-terminal side chain of the bound peptide, and both subtypes can be distinguished by the peptides presented in vivo (13, 14). Despite some differences, in particular with regard to the COOH-terminal peptide residue, the majority of the peptides constituting the B*2705 repertoire were found to overlap with those from B*2709 and vice versa (14).

A self-peptide (pVIPR, previously termed VIP1R_400–408, RRKWRRWHL) with potential arthritogenic properties (15) is derived from vasoactive intestinal peptide type 1 receptor and shows high sequence homology to a peptide from the latent membrane protein 2 (pLMP2, residues 236–244, RRRWRRLTV) of EBV (16). pVIPR can be presented by both subtypes, but the resulting T cell repertoires are distinct; B*2709 individuals very rarely possess pVIPR-specific T cells, whereas such T cells are more abundant in peripheral blood of healthy B*2705 persons and are very frequently present in patients suffering from AS (15). This indicates efficient thymic elimination of T cells specific for HLA-B27:pVIPR complexes in B*2709- but not in B*2705-positive individuals. Although these findings do not prove a causative role of pLMP2 and pVIPR in AS, they open the possibility to correlate structural properties of these subtypes with the retention (in B*2705) or elimination (in B*2709) of pVIPR-specific T cells.

The two crystal structures described here show that B*2705 accommodates pVIPR in two different conformations, of which one is identical to that seen when complexed with B*2709, whereas the other diverges substantially. Combined with functional studies in which pVIPR-specific CTLs were used, these findings suggest a molecular explanation for inappropriate T cell selection by individuals with a disease-associated HLA-B27 subtype.

Six patients with AS and one healthy donor were enrolled for this paper. HLA-B27 typing was performed using the HLA-B27 high resolution kit (Dynal). Five patients and the healthy donor were HLA-B*2705 positive, and one patient (LV) was HLA-B*2702 positive. Lymphoblastoid cell lines from these individuals were generated by in vitro transformation of B cells using the standard type 1 EBV isolate B95.8 (17).

PBMCs were isolated on a gradient of lymphoprep and depleted of the CD4⁺ fraction by Dynabeads M-540 CD4 (Dynal). Cells were incubated at 2 × 10⁴ cells/well in 96-well flat-bottom microplates, stimulated at an 0.5:1 stimulator/responder ratio with autologous EBV-transformed B cells prepulsed overnight with 8.5 μM pVIPR, and γ irradiated (200 Gy). They were grown in RPMI 1640 medium containing 10% heat-inactivated pooled human serum, 2 mM l-glutamine, 10 U/ml penicillin, and 100 μg/ml of streptomycin and in the presence of 20–100 U/ml of human rIL-2 (Boehringer). 10 d later, cytotoxic T lymphocyte lines were restimulated as aforementioned. After 1 wk, they were tested for pVIPR specificity in a standard ⁵¹Cr-release assay using as targets T2 cells transfected with either B*2705 cDNA (T2-B*2705) or B*2709 cDNA (T2-B*2709), pulsed with 70 μM of the peptide, or used untreated.

HLA-B27:pVIPR protein complexes were produced as described previously (18, 19). The pVIPR peptide (RRKWRRWHL) was obtained by standard solid phase synthesis and purified by HPLC (Alta Bioscience). The purified protein complexes were used for crystallization at concentrations of 20 mg/ml (B*2705:pVIPR) and 16 mg/ml (B*2709: pVIPR), respectively, in TBS buffer (10 mM Tris HCl, pH 7.5, and 150 mM NaCl).

Using hanging drop vapor diffusion, crystals suitable for X-ray diffraction experiments were obtained from a PEG 8000 pH screen. As observed earlier for similar HLA-B27 complexes, streak seeding was indispensible to produce large single crystals. For B*2705:pVIPR, drops made from 1 μl of precipitant solution (0.1 M Tris HCl, pH 8.0, and 16% PEG 8000) and 1 μl of protein solution produced prismatic crystals with a maximum size of 400 × 250 × 250 μm. Using glycerol as cryoprotectant, the best dataset in terms of resolution (1.47 Å) was obtained at the BL2 beamline at BESSY-II, Berlin, Germany. In the case of B*2709:pVIPR, drops made of 1 μl of precipitant solution (0.1 M Tris HCl, pH 8.5, and 19% PEG 8000) and 1 μl of protein solution yielded platelike crystals with a maximum size of 300 × 200 × 100 μm. Cryo-cooled crystals of this complex (glycerol as cryoprotectant) allowed collection of a 2.2-Å resolution dataset at the X13 beamline at Deutsches Elektronen Synchrotron. Both datasets were collected at 100 K and processed with the HKL package (see Table I; reference 20).

The structure of HLA-B*2709:pVIPR was determined by molecular replacement using peptide-stripped B*2709:m9 (PDB code 1k5n) as a search model and program molrep (CCP4; reference 21). After rigid body refinement using refmac (22), the initial model was improved with ARP/wARP (23), and water molecules were included. Further improvement of the structure was achieved by iterative cycles of manual rebuilding using O (24) and restrained maximum-likelihood refinement with refmac comprising isotropic B factor adjustment. After translation, libration, and screw rotation refinement (25), the R value converged at 0.188 (R_free, 0.244).

As the two HLA-B27:pVIPR complexes crystallized isomorphously, initial phases for B*2705:pVIPR were calculated from peptide-stripped B*2709:pVIPR with His116 replaced by alanine. This initial model was subjected to rigid body refinement followed by simulated annealing and energy minimization using CNS (26). ARP/wARP was used for improvement of the model and incorporation of water molecules and pVIPR in canonical conformation. After manual rebuilding, inclusion of alternate protein conformations, and atom positional refinement (refmac), there was still residual electron density in the peptide binding groove, to which an additional, noncanonical conformation of pVIPR could be fitted with an estimated occupancy of 0.5 for both peptide chains. Subsequent optimization of atomic displacement parameters decreased the R factor by 2.5% (R_free, 2%) compared with isotropic refinement, justifying this procedure. Evaluation of anisotropy by parvati (27) showed the expected statistical distribution of 0.52 ± 0.14 for all atoms of the structure. The R value converged at 0.128 (R_free 0.178). The equimolar distribution of the two pVIPR conformations in the B*2705 molecule was confirmed by an occupancy refinement performed with CNS. With the p6α conformation (see Results) omitted from the atomic coordinates, the occupancy of the central part of the peptide refined to occupancies ∼0.5, whereas that of p1, p2, p8, and p9 stayed with full occupancy. Similar results were obtained when only the p6α conformation was refined.

In both structures, radiation-decarboxylated Asp and Glu that showed negative difference electron density were assigned occupancies <1.0. The final models include all 276 residues of the HC, 100 residues of β₂m (all 99 residues plus NH₂-terminal Met), all nonhydrogen atoms of the pVIPR peptide, water, and glycerol (from the cryoprotectant). All φ/ψ angles lie in allowed regions of the Ramachandran plot (procheck; reference 28). Both structures were validated with whatcheck (29). Statistics are compiled in Table I. Superimpositions were performed using profit (30). Figures were generated using Grasp (31), povray (www.povray.org), molscript (32), povscript1 (33), and raster3D (34) together with a graphical interface (moldraw) developed by N. Sträter (Institut für Kristallographie, Freie Universität Berlin, Berlin, Germany; unpublished data). Molecular surfaces were calculated with MSMS (35).

The atomic coordinates and structure amplitudes have been deposited in the Protein Data Bank (accession codes 1ogt [B*2705:pVIPR] and 1of 2 [B*2709:pVIPR]).

HLA-B*2705:pVIPR and B*2709:pVIPR crystallized isomorphously (same space group and comparable unit cell constants), showing the typical MHC class I immunoglobulin-like folds (1) and refined to values of R_cryst = 12.8% and R_free = 17.8% at 1.47 Å and R_cryst = 18.8% and R_free = 24.4% at 2.20 Å resolution, respectively (Materials and Methods and Table I). In both structures, pVIPR is bound in the common canonical conformation (Fig. 1 A) found in other HLA-B27 molecules, exploiting all six pockets of the peptide binding groove for interaction with the HC (19, 36, 37). However, B*2705:pVIPR features an additional, grossly different noncanonical peptide conformation (Fig. 1 B). Both conformations found in B*2705 are present in a 1:1 ratio as shown by occupancy refinement. Despite this peculiarity in B*2705, the structures of HC and β₂m of the two subtypes are practically indistinguishable (C_α root mean square [rms] deviation 0.2 Å), including almost all side chain atoms. The atoms contributing to the binding groove occupy the same positions in B*2705 as in B*2709, and extreme atomic displacement factors (thermal anisotropy), which could mask multiple conformations were also not detected for B*2705. It has to be emphasized that, due to the isomorphous crystal structures, the intermolecular crystal contacts in both B*2705:pVIPR conformations are comparable, demonstrating that the two different binding modes of pVIPR found only in B*2705 are an intrinsic feature due to the presence of Asp116 and not a crystallographic artifact.

In addition, there is a metal binding site with an octahedral coordination sphere formed by pHis8-Nε, His197-Nε (of a symmetry-related HC), and four water molecules. The cation is probably Mn²⁺ as inferred from coordination distances (38). The electron density and B factors of Mn²⁺ (19.7 Ų in B*2705 and 42.3 Ų in B*2709; the B values are different due to the different resolutions) (Fig. 2, electron densities for Mn²⁺) suggest that the site is fully occupied. The presence of Mn²⁺ is a purification artifact that originates from a lysis buffer.

The different conformations of pVIPR in the binding groove of B*2705 are reflected in the backbone torsion angles of the peptide. In both structures, pVIPR is bound in the canonical conformation found regularly in crystal structures of MHC class I–nonapeptide complexes with main chain torsion angles (ϕ, ψ) of the extended β-strand type and only residue p4 (pTrp4) in right-handed α-helical conformation (Fig. 1 A, ϕ − 85.2° and ψ − 32.6°). The α-helical turn at position p4 results in a prominent kink directing the main chain “away” from the floor of the binding groove toward the solvent. Residues of pVIPR in this conformation (pVIPR-p4α) interact with all six pockets (Fig. 1 E, pockets A–F) provided by the binding groove, as found commonly for nonapeptides (1).

The additional peptide conformation found in B*2705 exhibits a pArg6 in α-form (Fig. 1 B, ϕ − 106.4° and ψ + 34.7° designated pVIPR-p6α) and both pVIPR-p4α and pVIPR-p6α are half occupied (Materials and Methods and Table I). pVIPR-p4α is virtually identical in both subtypes (Fig. 1 A, rms deviation 0.1 Å for nonhydrogen atoms). Between the p4α peptides in B*2705 and B*2709, there is only one minor difference in the orientation of the pArg6 guanidinium group as it is engaged in different hydrogen bonds.

At the peptide NH₂- and COOH-termini, pArg1, pArg2, pHis8, and pLeu9 occupy identical positions in pVIPR-p4α and -p6α (Fig. 1, A, B, and E and Fig. 2 A). The interactions of pArg1 and pArg2 with A and B pocket residues correspond to those observed for the B*2709:s10R complex (PDB code 1JGD; reference 37); the side chain of pArg1 is sandwiched between the side chains of HC Arg62 (α₁-helix) and Trp167 (α₂-helix). In addition, the side chains of Arg62 and Glu163 (α₂-helix) are linked by a water-mediated salt bridge (Fig. 3, A and B, bridge) that covers the deeply embedded pArg2. At the COOH terminus, pHis8 is solvent exposed as well (Figs. 2 and 3), and pLeu9 is accommodated in the F pocket. The carboxy group of pLeu9 forms the common polar interactions with Tyr84-Oη, Thr143-Oγ, and Lys146-Nζ (1), and the aliphatic pLeu9 side chain forms hydrophobic interactions with the side chains of Leu81, Leu95, Tyr123, and Trp147 (19). Because the side chain of pLeu9 is too short to engage in any direct contacts with Asp116 (B*2705) or His116 (B*2709), there are no significant differences in the F pockets between both subtypes.

In contrast to the NH₂- and COOH-terminal residues, the central sections of pVIPR-p4α and -p6α, pLys3 to pTrp7, which are principally accessible for recognition by a TCR, differ drastically with maximal disparity at pArg5, 3.7 Å for C_α, and 16.5 Å for Nη₂ (Table II and Figs. 1, A, B, and E and Fig. 2 A and Figs. 3 and 4). In pVIPR-p4α, side chains of pTrp4 and pArg5 are exposed to solvent, but in pVIPR-p6α, pArg5 is locked within the binding groove, its guanidinium group forming a bidentate salt bridge with the B*2705 subtype-specific Asp116 at the interface of the C, E, and F pockets (Figs. 1 B and 2 C and Figs. 3 and 4). In pVIPR-p6α, residues pTrp4 and pArg6 flanking pArg5 are maximally exposed to solvent and stacked (Fig. 1 B and Fig. 2). As a consequence of the structural differences around pArg5, the solvent-accessible surface areas of pVIPR-p4α and pVIPR-p6α have different size (160 Ų for pVIPR-p4α and 124 Ų for pVIPR-p6α), shape (Fig. 3, A and B), and charge distribution (Fig. 3 D) so that several prerequisites for differential recognition by a TCR are provided.

As both p4α and p6α are roughly half occupied in the B*2705 structure (Materials and Methods), approximate energetic equivalence of the two conformations seems likely. This is suggested by the temperature factors for the peptide in the two conformations (Fig. 1, C and D) that are very similar: in B*2705, the average (isotropic) B factor for all peptide atoms is 14.8 Ų (range 10.0–25.9 Ų) for p4α and 17.4 Ų (range 10.8–31.9 Ų) for p6α. The highest flexibility found for pTrp4-p6α (∼30 Ų) is still relatively low on an absolute scale, and marked atomic displacement as indicated by large B factors is only found in the side chains of pTrp4, pArg6, and pTrp7 for both p4α and p6α (Fig. 1, C and D).

Previous analyses of autoreactive CTL lines from individuals typing as B*2705 or B*2709 have indicated that pVIPR-specific CTLs are frequently observed in patients with AS, although they occur also in healthy B*2705 individuals. In contrast, such CTLs are only rarely found in B*2709 individuals (15). We extended these studies, in particular in an attempt to identify CTLs exclusively reactive against B*2705 that might indicate immunogenicity also of B*2705-p6α. The results shown in Table III are based on 39 CTL lines, 17 of which are already published (15) and 22 were newly obtained from six AS patients (five B*2705 and one B*2702) and one B*2705 healthy control. The CTLs are grouped as preferentially recognizing peptide-transporter–deficient T2-B*2705 or T2-B*2709 cells, both subtypes with approximately equal efficiency, or exclusively T2-B*2705. No CTL recognizing only T2-B*2709 was observed, although pVIPR was used at a high concentration (70 μM). This experimental setup was chosen to maximize the chances to detect reactivity genuinely specific for B*2705:pVIPR complexes. Dose–response curves for the reactivity of pVIPR-specific CTLs have already been determined (15); at lower peptide concentrations, T2-B*2709 target cells tended to be lysed at higher efficiency than T2-B*2705 cells.

The majority of the CTLs lysed both T2-B*2705 and T2-B*2709 equally well (Table III). However, there were exceptions showing a clear preference for T2-B*2709 cells. This was not unexpected because it has previously been found that B*2709 antigens present pVIPR more efficiently than B*2705 (15). Only one CTL line was found (EP76) that selectively reacted with pVIPR presented by T2-B*2705 cells. The complete lack of reactivity against the B*2709:pVIPR complex, even at the high peptide concentration used, indicates that EP76 is directed against a structure different from B*2705:pVIPR-p4α because of the extreme structural similarity of B*2709:pVIPR and the p4α binding mode in B*2705 (Fig. 1 A). In contrast, five further CTLs from the AS-patient EP reacted with both T2-B*2705 and T2-B*2709 cells.

However, it must be pointed out that oligoclonal CTL lines were used in these studies. Therefore, it appears possible that noncross-reactive CTL clones may be concealed in the population of cells exhibiting cross-reactivity, resulting in a lower degree of lysis of B*2709 target cells, and leading to an underestimation of the immunogenicity of the p6α conformation in B*2705. 15.4% of the CTL lines lysed B*2705:pVIPR targets more efficiently than T2-B*2709:pVIPR cells (Table III).

A fundamental deviation from the single binding mode found in all previously determined MHC class I:peptide complexes is the dual conformation of pVIPR in B*2705. This feature is of considerable interest both from structural and immunological points of view. Only one exception to the single binding mode of peptides has been described so far in the crystal structure of the rat RT1-A^a antigen in complex with the exceptionally long 13-mer peptide MTF-E. Two entirely different MTF-E conformations were found that are possibly associated with crystal packing effects (39). Because they could consequently represent nonphysiological artifacts, the two conformations and the reasons why MTF-E does not adopt a single binding mode are different from those reported here.

The simultaneous occurrence of pVIPR-p4α and pVIPR-p6α in B*2705 may be static or dynamic. If static, the analyzed crystals would be composed of approximately equal amounts of two cocrystallizing asymmetric units: one containing pVIPR-p4α and the other pVIPR-p6α. If dynamic, the peptide would change continually from one conformation to the other within the binding groove, using fixed NH₂- and COOH-terminal amino acids with pivot points located at or close to the main chain connections of p2-p3 and p7-p8, respectively (Figs. 2 A and 3 C). Currently, we cannot distinguish between static and dynamic modes of peptide binding, but spectroscopic methods (40) or X-ray analyses of crystals produced at different temperatures may eventually provide an answer.

Thus far, the presentation of a peptide by an MHC molecule in a single, defined conformation has been considered a prerequisite for positive and negative selection of T cells within the thymus (41). The two drastically different binding modes of pVIPR to B*2705 show that this assumption might be an oversimplification. A peptide changing its conformation dynamically within the groove could possibly preclude high-affinity interaction with the two CDR3 regions of a TCR (42, 43), whereas a static conformation might allow a more efficient recognition. The rare occurrence of pVIPR-specific CTL in B*2709 individuals suggests that the B*2709:pVIPR complex can indeed serve in establishing tolerance (15). Conversely, the number of B*2705:pVIPR-p4α and -p6α complexes might fall below the epitope density threshold required for negative selection (44, 45), leading to the frequent presence of autoreactive, pVIPR-specific CTL (15).

The CTL lines analyzed here seem to possess relatively low affinity (15), questioning the presentation of the VIPR-derived peptide under physiological conditions, and suggesting the use of VIPR transfectants in cytotoxicity studies. However, even when testing the reactivity of CTLs against EBV-derived peptides, lysis of target cells is rarely observed (15, 46). At least the viral peptides are surely processed and presented; comparative studies with transfectants are usually unsuccessful. In addition, a COOH-terminal pLeu9 as seen in pVIPR does not present an obstacle for efficient proteasomal processing (47) and transport to the endoplasmic reticulum, as shown by peptide elution studies from B*2705 and B*2709 molecules (14). The high number of pVIPR-specific CTLs in some B*2705 individuals (15) might be explained by selective increase (clonal expansion) due to cross-reaction. This is not unlikely because TCR cross-reactivity is mandatory for the control of infections (48–52). In the case discussed here, the pVIPR cross-reactive nonself antigen could be the EBV-derived pLMP2 (15, 16, 53). Convincing examples of molecular mimicry between viral and self-proteins have already been identified (51, 52).

Unconventional peptide binding modes distantly resembling that observed for the p6α conformation in B*2705:pVIPR have occasionally been found in other MHC:peptide complexes as well (18, 54, 55), but the involvement of a strategic subtype-specific residue at the floor of the binding groove is unprecedented. The detection of p6α in B*2705:pVIPR but not in the closely related B*2709:pVIPR complex is obviously a consequence of the combined presence of the allele-specific residue Asp116 (B*2705) and a specific peptide sequence with Arg at position 5. The larger size and neutral or positive charge of His116 compared with Asp116 would interfere sterically and electrostatically if the pArg5 side chain were bound to the F pocket (Fig. 1 A). The binding of the peptide COOH-terminal pLeu9 to the F pocket brings a further constraint into play: to avoid clashes, it is likely that the side chain of this residue has to be small and hydrophobic, but not positively charged to prevent competition with another basic amino acid (in this instance, pArg5) that contacts Asp116.

The paucity of peptides with pArg5, such as pLMP2, which have been found to be presented by HLA-B27 molecules under natural conditions (14, 16), makes it difficult to provide an estimate of the generality of the p6α conformation, let alone a dual peptide binding mode, in HLA-B27 molecules. We have found a few further peptides derived from self- as well as microbial proteins that have in common not only an Arg residue but even the WRR motif (unpublished data). As it seems conceivable that some of these peptides share the unorthodox structural features of pVIPR when complexed with the B*2705 subtype, analyses with these novel peptides are underway to examine the possibility that the WRR motif represents a sequence with direct involvement in AS pathogenesis. With only very few structures of HLA-B27:peptide complexes solved to date (Results; references 19, 36, 37), it seems premature to engage in a speculation on the prerequisites for a dual peptide conformation.

However, the pVIPR-p6α peptide binding mode and its subtype dependency is not a unique, serendipitous finding. We have observed recently that the sequence-related pLMP2 peptide (16) binds in p4α-like form to B*2709 but in p6α-like conformation to B*2705, with pArg5 again forming a bidentate salt bridge with Asp116 (unpublished data). Therefore, the noncanonical binding mode involving residue 116 as observed for one of the conformations in B*2705:pVIPR and even its subtype dependency can be found in other HLA-B27:peptide complexes as well. Most AS-associated HLA-B27 alleles encode Asp116, whereas this residue is Tyr (B*2706) or His (B*2709) in the two subtypes not associated with AS (7).

A constellation similar to that of the B*2705/B*2709 pair is found in A*68012 (Asp116) and A*6807 (His116; reference 56), or in B*3501 (Ser116) and B*3503 (Phe116; reference 57), but comparative structural studies have not been performed for these proteins. Among further HLA class I genes, approximately half of the HLA-A alleles encode Asp116, and nearly all others encode Tyr116 that occurs also in a smaller number of HLA-B and HLA-C antigens, in which Asp116 is rarely represented (58). However, the mere availability of Asp116 is not sufficient for a basic residue in the middle of a peptide to engage in a p6α-like binding mode. This is exemplified by the HLA-A*0201:MAGE-A4 complex (59) or HLA-B*4402 and B*4403 complexed with an HLA-DP–derived peptide (60): these peptides exhibit Arg either at position 5 or 6 that are solvent exposed and not buried. Other residues on the floor of the binding groove could principally also be involved in noncanonical anchoring modes such as the HC residues 97, 99, and 114, which are, like residue 116, highly polymorphic in HLA class I molecules (61). For example, HLA-B*3501 presents an EBV-derived nonapeptide with pAsp5 contacting Arg97 and Tyr99 (18). Clearly, a nonconventional peptide anchoring could be more frequent than currently envisaged.

Polymorphisms within the MHC class I peptide binding groove as shown here may influence peptide presentation and T cell repertoire selection (1, 4, 7, 15, 62–64), but a dual peptide presentation mode as found in B*2705 that depends on the HLA-B27 subtype-specific amino acid exchange is novel. Distinct pVIPR presentation modes and differential AS-association of B*2705 and B*2709 may be linked in the course of the development of AS (15), although other factors influencing AS initiation and progression have to be taken into consideration, among them possible polymorphisms of HLA-B–linked genes (65–67). Our results support certain models of AS–pathogenesis that can incorporate differential peptide display by HLA-B27 subtypes, such as the arthritogenic peptide theory (4, 7), in particular in the modified form proposed by us (15), or the β₂m-deposition hypothesis (68).

HLA subtype-dependent differential presentation of peptides in noncanonical conformations due to HLA class I HC polymorphisms at residue 116 might also be connected to the pathogenesis of other disease states, such as the rate of the progression to AIDS in HIV-positive patients (57) or the outcome of bone marrow transplants between unrelated individuals (69). In conclusion, our results provide an unexpected twist regarding the mode of peptide display as a single, structurally unaltered MHC molecule can present a given peptide in two entirely different binding modes, depending only on a buried HC amino acid polymorphism.

The authors thank all patients and the healthy proband for their participation in this paper. We are grateful to P.G. Coulie, J.A. López de Castro, and D.J. Schendel for discussions and comments on the manuscript; to M. Rühl for excellent technical assistance; and to U. Müller, S. Popov, and R. Hillig for help at the synchrotron facilities.

This work was financially supported by the Deutsche Forschungsgemeinschaft (SFB 449), Berliner Krebsgesellschaft, Monika Kutzner-Stiftung, Berlin, Sonnenfeld-Stiftung, Berlin, Fonds der Chemischen Industrie, COFIN 2001, and Istituto Pasteur Fondazione Cenci-Bolognetti. Data collection at Deutsches Elektronen Synchrotron was supported by a European Community grant to the European Molecular Biology Laboratory Outstation in Hamburg.