Conformational Dimorphism of Self-peptides and Molecular Mimicry in a Disease-associated HLA-B27 Subtype*

An interesting property of certain peptides presented by major histocompatibility complex (MHC) molecules is their acquisition of a dual binding mode within the peptide binding groove. Using x-ray crystallography at 1.4 Å resolution, we show here that the glucagon receptor-derived self-peptide pGR (412RRRWHRWRL420) is presented by the disease-associated human MHC class I subtype HLA-B*2705 in a dual conformation as well, with the middle of the peptide bent toward the floor of the peptide binding groove of the molecule in both binding modes. The conformations of pGR are compared here with those of another self-peptide (pVIPR, RRKWRRWHL) that is also displayed in two binding modes by HLA-B*2705 antigens and with that of the viral peptide pLMP2 (RRRWRRLTV). Conserved structural features suggest that the N-terminal halves of the peptides are crucial in allowing cytotoxic T lymphocyte (CTL) cross-reactivity. In addition, an analysis of T cell receptors (TCRs) from pGR- or pVIPR-directed, HLA-B27-restricted CTL clones demonstrates that TCR from distinct clones but with comparable reactivity may share CDR3α but not CDR3β regions. Therefore, the cross-reactivity of these CTLs depends on TCR-CDR3α, is modulated by TCR-CDR3β sequences, and is ultimately a consequence of the conformational dimorphism that characterizes binding of the self-peptides to HLA-B*2705. These results lend support to the concept that conformational dimorphisms of MHC class I-bound peptides might be connected with the occurrence of self-reactive CTL.

3). MHC class I antigens consist of a highly polymorphic, MHC-encoded heavy chain (HC), that is non-covalently associated with ␤ 2 -microglobulin (␤ 2 m). The HC forms a groove carrying a peptide that is a proteolytic fragment of self-or nonself-proteins within the cell (4). The complex of HC, ␤ 2 m, and peptide is often termed pMHC.
In case of the human MHC class I allele HLA-B27, which is very strongly associated with ankylosing spondylitis (AS) (5,6), autoimmunity and in particular molecular mimicry between foreign and self-proteins or their fragments have long been suspected to play a role in pathogenetic processes (7)(8)(9)(10)(11). Cytotoxic T lymphocytes (CTLs) directed against the self-antigen pVIPR (RRKWRRWHL, derived from vasoactive intestinal peptide type 1 receptor (residues 400 -408)) have been found in healthy individuals with the AS-associated HLA-B27 subtype B*2705, and their number is increased in AS patients (12). In addition, a proportion of these T cells cross-react with the viral pLMP2 peptide (RRRWRRLTV, derived from latent membrane protein 2 (residues 236 -244) of Epstein-Barr virus) (12,13). Extensive structural similarity between these peptides is observed when they are displayed by B*2705, due to a salt bridge between pArg 5 of both peptides and residue Asp 116 at the floor of the peptide binding groove (14,15). Interestingly, pVIPR is presented in an unusual dual conformation, of which only one binding mode permits the formation of the pArg 5 -Asp 116 salt bridge (14). However, a causal relationship between these two peptides, the CTLs recognizing them in the context of B*2705, and AS has not been established.
We were interested whether further peptides with sequences exhibiting similarity to pVIPR or pLMP2 exist that would share the unorthodox conformation found for one of the two pVIPR binding modes and pLMP2 in B*2705 (termed "p6␣", i.e. main chain / torsion angles in ␣-helical conformation at peptide position p6, contrasting with the common "p4␣" conformation (14,15)). A peptide derived from glucagon receptor (pGR, RRRWHRWRL, residues 412-420) that exhibits extensive sequence similarities was found and chosen for further structural and functional studies. In addition, we investigated whether extensive CTL cross-reactivity had consequences for those regions of T cell receptor (TCR) sequences that typically interact with peptide residues of pMHC complexes, i.e. residues belonging to the complementarity determining regions (CDR) 3 of TCR ␣ and ␤ chains (16 -20).

MATERIALS AND METHODS
HLA-B27-positive Patients-Two female (MP and AB) and three male (EP, LV and VP) individuals, fulfilling the modified New York criteria for diagnosis of AS, were recruited for this study. All patients were informed about the aim of the experiments and gave their consent. HLA-B27 subtyping was performed using the Dynal AllSet ϩTM SSP HLA-B27 "High resolution" kit (Dynal Biotech Ltd., United Kingdom). Patients EP, MP, AB, and VP were B*2705-positive, whereas LV was B*2702-positive.
Cell Lines-Autologous B lymphoblastoid cell lines (B-LCLs) from patients with AS were generated by in vitro immortalization of B cells using the standard type 1 Epstein-Barr virus isolate B95.8 (21) and cultured in RPMI (Invitrogen) supplemented with 10% fetal calf serum, 2 mM L-glutamine, 100 units/ml penicillin, 100 g/ml streptomycin. T2B*2705 transfectants described elsewhere (15) were cultured in the same medium supplemented with 200 g/ml hygromycin B (Roche Diagnostics, Mannheim Germany) to maintain the expression of HLA-B27 molecules.
Analysis of TCR Gene Usage-Total RNA extraction from T cell clones, cDNA synthesis, and amplification of TCR ␣ and ␤ chains were performed as described (22). TCR families V␣ 18 -29 were amplified by PCR with the oligonucleotides reported by Kalams and co-workers (23). The products were purified from an agarose gel using a gel band purification kit (Amersham Biosciences). Internal primers upstream to the TCR C␣ and C␤ reverse primers were used for direct sequencing.
Protein Preparation and Crystallization-The pGR peptide was synthesized and purified by Alta Bioscience (Birmingham, UK). B*2705 HC and ␤ 2 m were expressed separately in Escherichia coli. Inclusion bodies containing the proteins were dissolved in aqueous 50% urea. HLA-B27⅐pGR complexes were reconstituted as described previously for other pMHC (24 -26). The complexes were purified by size exclusion chromatography, concentrated, and used for crystallization at concentrations of 13-15 mg/ml in 20 mM Tris/HCl, pH 7.5, 150 mM NaCl, 0.01% sodium azide. Crystals were obtained from drops made of 1.5 l of protein solution and 1.5 l of precipitant solution (12-16% polyethylene glycol 8000, 100 mM Tris/HCl, pH 7.5 or 8.0) in a hanging-drop vapor diffusion setup using streak seeding techniques. Diffraction datasets were collected at European Synchrotron Radiation Facility, Grenoble (ID 14-2) from cryo-cooled crystals at 100 K with glycerol and polyethylene glycol 8000 as cryoprotectants.
Structure Determination of the B*2705⅐pGR Complex-X-ray data were processed with DENZO (27) and scaled with SCALEPACK (27) ( Table 1). The structure of B*2705⅐pGR was determined by molecular replacement with program EPMR (28) using the water-and peptidedepleted B*2705⅐m9 structure as search model (PDB entry 1JGE). Restrained maximum-likelihood refinement was performed using REF-MAC5 (29) comprising isotropic B-factor adjustment followed by iterative manual model building with O (30). Water molecules were positioned with ARP/wARP (31). Data collection and refinement statistics are given in Table 1. Intermediate and final structures were evaluated with PROCHECK (32) and WHATCHECK (33). The figures showing structural details were prepared with DINO (Visualizing Structural Biology (2002), www.dino3d.org), MSMS (34), and DELPHI (35).
Data Deposition-The atomic coordinates and structure amplitudes have been deposited in the Protein Data Bank (accession code 2A83).

Selection of a Peptide
Recognized by HLA-B27-restricted Cross-reactive CTLs-Because previous studies had suggested that the unusual Trp-Arg-Arg motif at positions p4 -p6 of the pVIPR and pLMP2 peptides might be instrumental in leading to CTL cross-reactivity and alleledependent molecular mimicry involving the p6␣ peptide conformation (14,15), peptide sequences related to this motif were selected from public protein databases (www.ncbi.nlm.nih.gov/) using the blastp program (36). In addition, the peptides had to share the HLA-B27-specific anchor residue pArg 2 and preferably an arginine at p1 as well as an aliphatic residue at p9. Two peptides derived from human proteins were identified: one originates from glucagon receptor (pGR, RRRWHR-WRL, residues 412-420), the other from the voltage-dependent calcium channel ␣1 subunit (pCAC, SRRWRRWNR, residues 513-521). Both peptides bound to HLA-B*2705 molecules expressed on T2 cells (results not shown), but only pGR led to stimulation by some pVIPRprimed CTL (Fig. 1). In addition, four nonself-peptides were identified that, however, were not recognized by the HLA-B27⅐pVIPR-restricted CTL, with the possible exception of the CTL line AB5, which weakly recognized the B*2705⅐pHXLF4 target (Fig. 1). Therefore, the pGR peptide was chosen for further structural and functional studies.

Structural Features of the B*2705⅐pGR
Complex-The structural basis for the observed CTL cross-reactivity was then investigated by crystallographic analysis of the B*2705⅐pGR complex ( Table 1). The peptide could be modeled unambiguously to the electron density ( Fig.  2A), revealing two conformations (termed pGR-A and pGR-B) with similar B-factors (Fig. 2B). The only residue with markedly higher B-factors is pArg 6 whose guanidinium group is considerably more flexible in both conformations than the side chains of any of the other residues. Both conformations are present in a 1:1 ratio as judged from the electron density maps. The pGR-A and -B conformations differ from pArg 3 to pArg 8 , with clearly distinguishable C␣ traces from p4 -p7 (Fig. 2, C and D). The torsion angles (,) of the main-chain residues of the peptide are of ␤-strand type except at residue pArg 6 , which exhibits righthanded ␣-helical conformation. Therefore, irrespective of the binding mode, pGR is bound in the p6␣ conformation, which had been detected for HLA-B27⅐peptide complexes so far in case of one of the two peptide binding modes in B*2705⅐pVIPR (14) as well as in B*2705⅐pLMP2 (15). Consequently, the side chain of pHis 5 points toward the interior of the binding groove. Both pGR conformations lead to fully solvent-exposed pTrp 4 and pArg 6 side chains that exhibit relatively few, weak HC contacts, nearly all Ͼ3.5 Å. A notable exception is the short salt bridge (2.80 Å) between pArg6 NH2 of pGR-B and Glu155 OE1 on the ␣2-helix. The two pGR conformations are also distinguished by the interactions that are formed by the buried pHis 5 . In pGR-A, pHis5 ND1 binds to pArg3 NH2 (with the side chain of pArg 3 in extended conformation), and a watermediated hydrogen bond is formed to Asp 77 on the ␣1-helix and Asp 116 on the floor of the peptide binding groove (Fig. 3A). In the B conformation, pHis5 ND1 forms a water-mediated hydrogen bond to pArg3 NE (with the side chain of pArg 3 in bent conformation) and a direct intrapeptide hydrogen bond is found between pHis5 NE2 and pTrp7 O ( Fig. 3B and Table 2). The two water molecules that mediate the contacts between pHis 5 in pGR-A with Asp 77 and Asp 116 are retained in pGR-B (Fig. 3). Table 2 lists the most pronounced differences between the two conformations. With the exception of pArg 1 , all solvent-exposed peptide side chains exhibit conformational dimorphism. The two Trp residues at p4 and p7 show distinct rotamer conformations that lead to different . d R free is the same as R cryst , but calculated on 5% of the data excluded from refinement. e Root mean square deviation from target geometries.
juxtaposition of their indole moieties. The termini of the peptide are bound in the characteristic binding modes that have been observed previously (14,15,37). Most notably, the side chain of pArg 1 is sandwiched between the HC residues Arg 62 and Trp 167 , resulting in an energetically favorable stabilization (38), whereas the side chain of pLeu 9 does not contact the floor of the F pocket (Figs. 2A, 3A, and 3B) but is firmly anchored by numerous hydrophobic interactions.
Structural Comparison of pGR, pVIPR, and pLMP2 Complexed with B*2705-A comparison of the structures of pGR, pVIPR, and pLMP2 in complex with B*2705 is greatly facilitated by the similar, high resolutions obtained and the isomorphous crystallization modes (space group P2 1 , Table 1) (14,15), demonstrating that intermolecular interactions that are associated with crystal packing apply to all structures. We have already pointed out that structural molecular mimicry in the context of HLA-B27 is an allele-and peptide-dependent property, because the similarity between the viral pLMP2 peptide and the self-peptide pVIPR is much more pronounced when both peptides are displayed by the B*2705 than by the B*2709 subtype (14,15). However, only one of the two pVIPR conformations, the unusual p6␣ binding mode, participates in molecular mimicry with pLMP2. The side-chain orientations (Fig. 4) as well as the surface properties (Fig. 5) show that molecular mimicry extends to the sequence-related pGR peptide as well. This mimicry is most obvious when pGR is in the B conformation, owing to the two rotamer conformations of pTrp4 (Fig. 2, C and D): only one of these is  Only peptide residues and selected ␣1-helical residues are shown; the view is through the ␣2-helix (removed), roughly along the length of the peptide binding groove toward the N-terminal peptide residues. B*2705⅐pGR is shown with the peptide in A-conformation (A) and B-conformation (B), which results in distinct contacts with water molecules, peptide residues, and HC atoms.

Comparison of pGR conformations with those of pVIPR and pLMP2 in the B*2705 subtype
Only direct intra-peptide contacts and contacts between the peptides and HC residues (up to 3.50 Å; HB, hydrogen bond; SB, salt bridge) are included; solvent-mediated interactions are omitted, and van der Waals (vdW) contacts are not given explicitly. In pGR, the p3-p8 residues exhibit double conformations. In case of the B*2705⅐pLMP2 complex, pTrp 4 and Asp 116 occur in alternative conformations; only one of the equally occupied pTrp 4 conformations (with higher degree of similarity to the pVIPR-p6␣ conformation) and the higher occupied Asp 116 conformation (occupancy 75%) are considered. The side chain of this residue is buried in all complexes; the contacts formed by pLeu9 (pGR, pVIPR) and pVal9 (pLMP2) are very similar.

Peptide position B*2705⅐pGR conformations A and
congruent with the corresponding residue in the pVIPR-p6␣ and pLMP2 structures. A calculation of the root mean square deviations for the C␣ atoms between the different peptides and their conformations in B*2705 (results not shown) supports this conclusion and demonstrates that pGR-B, pVIPR-p6␣, and pLMP2 exhibit the highest degree of structural similarity. TCR-accessible, exposed side chains of the peptide that exhibit structural equivalence between the three peptides include at least pArg 1 and pTrp 4 but possibly also pArg 6 because of its considerable flexibility in pGR (Fig. 2B). In addition, the surfaces above the peptide binding groove-embedded residues pArg 3 /Lys 3 , pHis 5 /Arg 5 , and pTrp 7 /Leu 7 are comparable (Fig. 5). As observed previously (15), the similarity is most pronounced around the N-terminal half of the pVIPR and pLMP2 peptides, and this is also true for the pGR peptide. The electrostatic surface properties of the three complexes in B*2705 are similar as well and are clearly most pronounced for the regions surrounding the N-terminal halves of the three peptides (Fig. 5).
CTL Cross-reactivity between pGR, pVIPR, and pLMP2 in the HLA-B27 Context-The cross-reactive potential of pGR-stimulated CTL from AS patients typing as B*2705 and B*2702 was then investigated. The fact that the CTL lines are not clones could be relevant in the case of CTL exhibiting cross-reactivity, as pointed out previously (14,15). However, it is unlikely that lack of reactivity is influenced by oligoclonality of the CTL lines, because this feature would be expected to enhance, and not to diminish cross-reactivity. Of 23 CTL lines from four patients (VP, MP, EP, and LV) (Fig. 6), four reacted also with B*2705⅐pVIPR and The orientations of the peptide side chains in the binding pockets are indicated and the primary sequence of the peptides is shown. *, the indole moieties of the exposed pTrp 4 residues of pGR and pLMP2 exhibit conformational dimorphism, and this is also the case for certain other exposed residues of pGR (pArg 6 and pArg 8 ). E, right panel, floor of peptide binding groove indicated by "␤-sheet" and binding region for a TCR by "TCR." B*2705⅐pLMP2 (VP7G, VP78G, VP90G, and LV3G). One CTL line recognized pGR and pLMP2 (VP52G), and another exhibited cross-reactivity between pGR and pVIPR (EP31G). The reactivity with pGR was usually much stronger than with pVIPR or pLMP2. These results demonstrate that pGR-stimulated CTL lines, which can recognize also one or even both other peptides in the context of B*2705, are readily detectable, although some differences between the CTL donors were apparent. pGR-stimulated CTLs derived from patients MP and EP were nearly never cross-reactive, whereas about half of the CTLs from donor VP exhibited cross-reactivity. These results reveal also that the complexes of each of the three peptides with B*2705 must exhibit structural (14,15) or dynamic (39) properties (or both) that lead to the prevention of cross-reactivity in the majority of the CTL lines.
Analysis of TCR ␣ and ␤ Chains from pVIPR-and pGR-stimulated T Cell Clones-TCR gene usage was assessed for 40 clones derived from pVIPR-or pGR-stimulated CTL. 17 of these clones have been reported earlier (15). Four clonotypes reacted only with the peptide employed for stimulation, whereas three exhibited different degrees of cross-reactivity. The clones mimicked the reactivity of the CTL from which they originated (Tables 3 and 4) and revealed that the CDR regions of the TCR␣ and -␤ chains contribute differentially to the cross-reactivity exhibited by these clones.
All 23 cross-reactive clones shared the (D/N)RDDKIIFG motif within their CDR3␣ regions, although they belonged to different TCR␣ chain families, whereas the non-cross-reactive clones lacked this motif (Table 3). However, also the latter exhibited some similarities: the majority shared the motif SSYKLIFG or a closely related sequence. Interestingly, clones from the most highly cross-reactive CTL AB5 and from the mono-specific PM65 shared the SGGSYIPTFG motif in one of their two ␣-chains. This could be connected to the fact that the PM65 CTL gave only borderline reactivity with pLMP2, thus partly resembling the fully cross-reactive AB5 CTL (Fig. 1). All clones derived from patient EP shared also a DSMD motif, just before the previously mentioned SSYKLIFG sequence, which was absent from all other sequences and may thus be connected with the exclusive specificity for B*2705⅐pGR that characterizes these clones.
In marked contrast to these results, no consistent sequence motif could be discerned among the ␤-chains of these clones, irrespective of whether they belonged to the group of non-cross-reactive or to the cross-reactive clones (Table 4). Only the TXXXQXFG motif was present in several of the CDR3␤ sequences but independent of the clonal reactivity. There was also no similarity in the usage of V␤ and J-region sequences. For example, although the cross-reactive clones derived from AB4 and AB5 shared the V␤22 family, their J-region sequences were different. Despite some similarities in CDR3␤ regions with various CTL or clones from patients with AS or reactive arthritis, another HLA-B27-associated disease, no consistent pattern emerges from a comparison of these TCR␤ chains with those of the clones analyzed in the present study (Table 4), except perhaps that the CDR3␤ region of the more stringent TCR tended to be shorter. Either the V␤ families, the J-regions, or both were distinct. Several T cell clones express a dual TCR with two ␣-chains (PM65 and AB5; Table 3) or ␤-chains (EP16G ;  Table 4), sharing common motifs in their CDR3 regions with T cell clones from the same or a different patient. Although this dual expression might explain the recognition of the different peptides in some cases (AB5), some other cross-reactive clones contain only single ␣ or ␤ chains (AB4, MPVPAC7). Moreover, some mono-specific T cells (PM65 and EP16G) express more than one TCR␣ or -␤ chain. EP16G clones in particular display two ␤ chains that share the same CDR3 region but have two related variable regions (3 and 3.1).

DISCUSSION
The present study addresses the question whether the conformational dimorphism observed for a self-peptide (pVIPR) in the context of the B*2705 subtype (14) and the functional and structural mimicry, which it exhibits with the viral pLMP2 peptide, were serendipitous findings or would extend to further peptides. Among several sequencerelated peptides of human, bacterial, or viral origin, the pGR peptide was, apart from the previously identified pLMP2, the only one that gave strong responses with some B*2705⅐pVIPR-restricted CTLs (Fig. 1). The glucagon receptor, from which pGR is derived, is, like the vasoactive intestinal peptide type 1 receptor, a G-protein-coupled receptor. It interacts with the peptide hormone glucagon and performs a crucial physiological role in glucose and insulin metabolism. Expression has been found in many different tissues, but particularly strongly in cells within the liver and the kidney (40). CTLs that were stimulated by pGR in the context of B*2705 exhibited occasional cross-reactivity with pVIPR and pLMP2 as well (Fig. 6), indicating a functional similarity between CTLs that recognize these three peptides.
It must be borne in mind, however, that the three sequence-related peptides are possibly not the peptides that exhibit molecular mimicry which, if at all, may be relevant in the context of autoimmunity (15). Such relevance might be exhibited by further, currently unknown, peptides that show less pronounced sequence similarity (3) or could be the result of post-translational splicing (41), both of which are likely to have gone unnoticed by the type of data bank searches carried out by us. Despite this caveat, pGR, pVIPR, and pLMP2 allow to study the structural basis of TCR cross-reactivity in the HLA-B27 context. This analysis reveals that the AS-associated subtype B*2705 shows a much more pronounced degree of molecular mimicry between the three peptides than the non-AS-associated B*2709 subtype (42), particularly, when the pGR-B conformation is considered (15). 7 The surfaces above the peptide residues p1 to p6 of pGR, pVIPR, and pLMP2 are nearly identical (Fig. 5), despite the different rotamer conformations of pTrp 4 and pArg 6 , which are very likely interconvertible under physiological conditions (Fig. 2). This may provide an explanation for the occasional cross-reactivity observed for B*2705⅐pVIPR-primed CTL (Fig. 1), and the same applies to B*2705⅐pGR-primed CTL (Fig. 6). However, those parts of the complexes that are in the vicinity of the diverging C-terminal peptide sequences are distinct (Figs. 4 and 5) and might be targeted by those CTL that lack cross-reactivity.
The high resolution of the B*2705⅐pGR structure unequivocally demonstrates the existence of two peptide conformations in the binding groove (Fig. 2). As in the case of the B*2705⅐pVIPR complex, in which the peptide occurs in a dual binding mode (14), it is not possible, on the basis of the crystal structure described here, to distinguish between a static and a dynamic peptide binding mode. Because the difference between the two pGR conformations is not as drastic as for pVIPR-p4␣ and -p6␣, it may be more likely that the pGR peptide exhibits a dynamic mode of binding. Spectroscopic methods or NMR studies could pre- sumably resolve this issue. It seems also possible that local changes in pH or ion concentrations might affect the protonation state of amino acids that are part of the peptide binding groove, or of the peptide itself. In case of pGR, pHis 5 could be affected by such changes, possibly leading to altered interactions between the peptide and binding groove residues (Fig. 3). It would be necessary to analyze the B*2705⅐pGR complex at lower pH to uncover such differences.
We have previously pointed out that the dual pVIPR conformation in the B*2705 subtype might influence T cell selection within the thymus, particularly by impairing negative selection, thereby providing an explanation for the frequent presence of pVIPR-directed CTL in patients with AS (14). The same reasoning could apply for pGR, although the differences between the A and B conformations are not as striking as those in case of pVIPR. It seems thus possible that the emergence of autoreactive CTL may be a direct consequence of conformational dimorphisms and dynamic properties of a given peptide within a distinct binding groove (39). The importance of MHC allele-dependent dynamics and different conformational states of a peptide for recognition of a pMHC by T cells is only beginning to be considered (14,19,39,(43)(44)(45). These attributes of peptides bound to MHC molecules may play a role in the context of molecular mimicry and in the differential association of HLA-B27 subtypes such as B*2704 and B*2706 or B*2705 and B*2709 with AS (5,6,12). However, it is likely that the general features of this model do not only apply to HLA-B27 antigens.
The relatively conserved binding modes of human and mouse TCR on pMHC (16 -20) permit the prediction that the conserved part of the three B*2705 complexes is likely to interact with the CDR3␣ loop of a cross-reactive TCR (e.g. those expressed by CTL MPVPAC7 or AB5), whereas the region around the peptide C termini might be responsible primarily for interaction with the CDR3␤ loop of a peptide-selective CTL such as PM45 or EP16G (Figs. 1 and 6 and Tables 3 and 4). It is notable that CTL clones recognizing B*2705 in complex with pVIPR, pGR, or pLMP2 (e.g. MPVPAC7 clones or clones from donor AB; Table  3) tend to exhibit a considerable degree of similarity with regard to their CDR3␣ sequences, mainly through the presence of the (D/N)RDDKI-IFG motif and sharing of the J␣9.4 region. The lack of such similarity in the TCR␤ chain sequences indicates that the TCR␣ chains are primarily responsible for determining whether cross-reactivity between the three peptides can occur at all, whereas the TCR␤ chains seem to modulate its degree (Table 4). Those clones that exhibited no cross-reactivity, e.g. PM45, EP16G, or EP17G, lack the CDR3␣ (D/N)RDDKIIFG motif, and they were also found to possess different J␣ regions and TCR␤ sequences (Tables 3 and 4). In case of the latter two clones, the CDR3␤ loops are extremely short (Table 4), so that it appears doubtful whether this part of these TCR can participate in recognition of the pMHC surface at all.
However, in the absence of a structure of an HLA-B27⅐peptide⅐TCR complex (46), these considerations must currently be regarded as speculative, and it remains also unclear whether they extend to other MHC class I structures. Furthermore, the recently described structure of an autoimmune TCR complexed with an HLA class II antigen and a selfpeptide revealed a novel topography of a pMHC-TCR interaction (47). In this complex, the reactivity of the TCR is nearly exclusively restricted to the N-terminal half of the peptide and its surrounding, and the CDR3 loops of both chains engage in atypical contacts, resulting in a large tilting angle of the TCR on top of the pMHC. It is therefore impossible to predict whether the TCR that interacts with HLA-B27⅐peptide complexes will recognize their epitopes in the conventional or the novel fashion (the latter is unlikely in case of the CTL with the short CDR3␤ loops), or whether they might exhibit an as yet undetected additional binding mode. Consequently, the limited sequence similarities that we found between the CDR3␤ loops of PM65 and CTL from a patient with reactive arthritis, of AB4, or CTL from two further patients (Table 4) might indicate similar HLA-B27⅐TCR binding modes, but this issue

TCR␤-chain sequences of pVIPR-and pGR-stimulated T cell clones from patients with AS
The specificities and numbers of clones are given in Table 3. must presently remain unresolved. The same applies to the possibility that the relatively high content of positively charged residues that shape TCR epitopes of the B*2705⅐pGR/pVIPR/pLMP2 complexes (Fig. 5) may provide docking points for Asp or Glu residues in the CDR3␣ and CDR3␤ loops of TCR on cross-reactive CTL, e.g. in AB4, AB5, and MPVPAC7 clones (Figs. 1 and 6 and Tables 3 and 4).

CTL BV N-D-N BJ
Our finding that clones from different CTL often express two ␣or two ␤-chains could be functionally relevant. In the case of EP16G, these clones share the same CDR3␤ region but possess two germ-line TCR␤ regions. An analogous observation regarding the TCR␣ chain of autoreactive T cell clones in autoimmune diabetes has recently been described (48). The authors show that small amino acid variations distal to the antigen binding site of the TCR may have a profound effect on the avidity of individual clonotypes and are indicative of a pathogenic maturation of the T cell response. Although we do not know whether the two TCR␤ chains expressed in the T cell clones described here have a different affinity for the B*2705⅐pGR complex, the possibility of affinity maturation of TCR via reactivation of the germ line recombinatorial process (49) should be considered. It may occur more often than supposed and might play a role in tuning autoimmune reactions.
Several hypotheses have been put forward to explain the association of HLA-B27 and spondyloarthropathies (2, 6, 50 -54), but molecular mimicry between a foreign, i.e. microbial or viral, peptide and a selfpeptide is a central postulate of the arthritogenic peptide hypothesis (2,6,54). Together with the previous descriptions of the B*2705⅐pVIPR and B*2705⅐pLMP2 structures (14,15), the present study provides a molecular framework that accounts for the observed CTL cross-reactivity between B*2705 molecules in complex with self-peptides (pGR and pVIPR) and a foreign peptide (pLMP2). Therefore, the arthritogenic peptide hypothesis (2, 6) might well be relevant for HLA-B27 and its association with spondyloarthropathies, although other explanations can still not be excluded. Despite the fact that many more autoimmune diseases are associated with HLA class II than with HLA class I alleles (55,56), the class I allele HLA-B27 presents a particularly interesting case. The extremely strong association between certain HLA-B27 subtypes and AS, e.g. B*2704 and B*2705, and its absence in individuals harboring the B*2706 and B*2709 subtypes that differ only minimally from the former (5, 6), warrants further functional, biochemical, structural, and other biophysical investigations. In our opinion, comparative studies involving AS-associated and non-associated subtypes (12,14,15,37,39,57,58) hold the key to achieving an in-depth understanding of the pathogenesis of HLA-B27-associated autoimmune diseases.