The Peptide Repertoires of HLA-B27 Subtypes Differentially Associated to Spondyloarthropathy (B*2704 and B*2706) Differ by Specific Changes at Three Anchor Positions*

HLA-B*2704 is strongly associated with ankylosing spondylitis. B*2706, which differs from B*2704 by two amino acid changes, is not associated with this disease. A systematic comparison of the B*2704- and B*2706-bound peptide repertoires was carried out to elucidate their overlap and differential features and to correlate them with disease susceptibility. Both subtypes shared about 90% of their peptide repertoires, consisting of peptides with Arg2and C-terminal aliphatic or Phe residues. B*2706 polymorphism influenced specificity at three anchor positions: it favored basic residues at P3 and PΩ-2 and impaired binding of Tyr and Arg at PΩ. Thus, the main structural feature of peptides differentially bound to B*2704 was the presence of C-terminal Tyr or Arg, together with a strong preference for aliphatic/aromatic P3 residues. This is the only known feature of B*2704 and B*2706 that correlates to their differential association with spondyloarthropathy. The concomitant presence of basic P3 and PΩ-2 residues was observed only among peptides differentially bound to B*2706, suggesting that it impairs binding to B*2704. Similarity between peptide overlap and the degree of cross-reaction with alloreactive T lymphocytes suggested that the majority of shared ligands maintain unaltered antigenic features in the context of both subtypes.

The molecular basis for the very strong association of HLA-B27 with ankylosing spondylitis (AS) 1 (1), reactive arthritis (2), and other spondyloarthropathies remains a major unsolved problem. Hypotheses on the pathogenic role of HLA-B27 fall into three main categories. The classical "arthritogenic peptide" hypothesis assumes that a self-peptide presented by HLA-B27 would be the target of autoimmune CTLs activated by external antigen, such as bacteria (3). The occurrence of arthri-tis in HLA-B27 mice lacking ␤ 2 -microglobulin (4) and the fact that the HLA-B27 heavy chain can form homodimers in vitro (5) suggested that HLA-B27 might act as a noncanonical peptide-presenting molecule, perhaps leading to activation of unusual T-cell responses (6,7). It has also been suggested that misfolding of HLA-B27 heavy chains, perhaps exacerbated by infection or other environmental factors, might lead to endoplasmic reticulum stress responses and inflammation independent of antigen presentation (8,9). Other effects of HLA-B27 on modulating bacteria-host interactions have also been proposed (10,11). Supportive evidence for the arthritogenic peptide hypothesis comes from population studies showing differential association of some HLA-B27 subtypes with AS. Like most other class I antigens, HLA-B27 shows extensive polymorphism in human populations. As many as 25 HLA-B27 subtypes have been described thus far (see www.ebi.ac.uk./ imgt/hla). The apparently low frequency of many of these precludes a statistical analysis of their putative association with AS. At least B*2702, B*2704, B*2705, and B*2707 are linked to this disease (12). In contrast, B*2706 and B*2709 are not associated or are weakly associated with AS (13)(14)(15)(16). B*2706 has been found with significant frequency in Southeast Asia and the Pacific and at much lower frequency in continental China. A population study carried out in Thailand initially showed that whereas B*2704 was strongly linked to AS in this population, no B*2706 AS patients could be found, despite the significant frequency of this allele among healthy controls (13,17). This differential association of B*2704 and B*2706 in a same population was subsequently confirmed in two additional studies carried out in Indonesia (14) and among Singapore Chinese (15). Moreover, in segregation studies carried out in families in whom both B*2704 and B*2706 occurred, AS was observed only in B*2704-positive individuals (18). Two B*2706positive AS patients were found in China (12), suggesting that lack of association of this subtype with AS is not absolute and might be modulated to some extent by additional genetic factors. However, the very low frequency of B*2706 in China has thus far precluded case-control or family segregation studies in these populations (17). B*2704 and B*2706 differ by only two amino acid changes: H114D and D116Y (19 -21). Both of these changes are located in the same strand of the ␤-pleated sheet floor of the peptide binding site of HLA-B27 and are therefore not accessible to direct contact by the T-cell antigen receptor (22). However, due to their location, they can influence peptide specificity. Previous studies from our laboratory showed that a major difference between B*2704 and B*2706 is the more restricted specificity of the latter subtype for peptides with nonpolar C-terminal residues, including only aliphatic and Phe residues, whereas B*2704 also binds peptides with C-terminal Tyr (23). Peptide binding studies using poly(Ala) peptide analogs (24) suggested that B*2704/B*2706 polymorphism could have more complex effects than those revealed by pool sequencing by modulating peptide specificity at secondary anchor positions.
These previous studies suggested a direct relationship between the lack of or weak association of B*2706 with AS and its more restricted peptide specificity, relative to B*2704, but they failed to answer some relevant questions. First, to what extent does B*2706 polymorphism change the peptide repertoire of B*2704? That is, how many of the peptides presented by B*2704 fail to bind in vivo to B*2706? Second, besides the known effect on C-terminal residue specificity, are there other differential features between B*2704 and B*2706 ligands? Third, are the antigenic features of shared peptide ligands different when presented in the context of either B*2704 or B*2706? To address these questions, we have carried out a systematic comparison of the B*2704-and B*2706-bound peptide repertoires to determine their degree of overlap. In addition, we have used mass spectrometry (MS) to sequence a sufficiently large set of natural ligands to assess the differential structural features of the peptides bound to B*2704 and B*2706. Finally, we have used an extensive panel of alloreactive CTLs to compare the degree of antigenic similarity between B*2704 and B*2706 with the overlap of their peptide repertoires.

MATERIALS AND METHODS
Cell Lines and Monoclonal Antibodies-HMy2.C1R (referred to hereafter as C1R) is a human lymphoid cell line with low expression of its endogenous class I antigens (25,26). B*2704-and B*2706-C1R transfectant cells were described elsewhere (23). C1R cell lines were cultured in Dulbecco's modified Eagle's medium supplemented with 7.5% fetal bovine serum (both from Invitrogen). RMA-S is a transporter associated with antigen processing-deficient murine cell line (27,28). RMA-S transfectant cells expressing B*2704 or B*2706 and human ␤ 2 -microglobulin have been described previously (29). These cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum.
Isolation of B*2704-and B*2706-bound Peptides-This was carried out using 10 10 C1R transfectant cells lysed in 1% Nonidet P-40 in the presence of a mixture of protease inhibitors, after immunopurification of HLA-B27 with the W6/32 monoclonal antibody and acid extraction, exactly as described elsewhere (32). HLA-B27-bound peptide pools were fractionated by HPLC at a flow rate of 100 l/min as described previously (33), and 50-l fractions were collected.
Mass Spectrometry Analysis and Sequencing-The peptide composition of HPLC fractions was analyzed by matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) MS using a calibrated Kompact Probe instrument (Kratos-Schimadzu) operating in the positive linear mode, as described previously (33). Alternatively, a Bruker Reflex™ III MALDI-TOF mass spectrometer (Bruker-Franzen Analytic GmbH, Bremen, Germany) equipped with the SCOUT™ source in positive ion reflector mode was also used, as described previously (34).
Peptide sequencing was carried out by quadrupole ion trap nanoelectrospray MS/MS in an LCQ instrument (Finnigan ThermoQuest, San Jose, CA), exactly as detailed elsewhere (35,36). In some cases, peptide sequencing was also done by post-source decay (PSD) MALDI-TOF MS, as described previously (34).
In all cases, peptide-containing HPLC fractions were dried and resuspended in 5 l of methanol/water (1:1) containing 0.1% formic acid. Aliquots of 0.5 or 1 l were used for MALDI-TOF or nanoelectrospray MS analyses, respectively.
Synthetic Peptides-Peptides were synthesized using the standard solid-phase Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry and purified by HPLC. The correct composition and molecular mass of purified peptides were confirmed by amino acid analysis using a 6300 Amino Acid Analyzer (Beckman Coulter, Palo Alto, CA), which also allowed their quantification.
Epitope Stabilization Assay-The epitope stabilization assay used to measure peptide binding was performed as described previously (29).
Briefly, B*2704-or B*2706-RMA-S transfectant cells were incubated at 26°C for 22 h in RPMI 1640 medium supplemented with 10% heatinactivated fetal bovine serum. They were then washed three times in serum-free medium, incubated for 1 h at 26°C with various peptide concentrations without fetal bovine serum, incubated at 37°C, and collected for flow cytometry after 2 h (B*2704) or 4 h (B*2706). HLA-B27 expression was measured using 50 l of hybridoma culture supernatant containing the monoclonal antibody ME1. Binding of the RRYQKSTEL peptide, used as a positive control, was expressed as C 50 , which is the molar concentration of the peptide at 50% of the maximum fluorescence obtained at the concentration range used (10 Ϫ4 to 10 Ϫ8 M). Binding of other peptides was assessed as the concentration of peptide required to obtain the fluorescence value at the C 50 of the control peptide. This was designated as EC 50 .
Isolation of HLA-B*2704-specific CTL Clones and Cytotoxicity Assay-B*2704-specific CTL clones were obtained from five unrelated HLA-B27-negative donors as follows. About 10 6 peripheral blood mononuclear cells from each donor were stimulated for a week with a mixture of 10 5 B*2704-positive lymphoblastoid cell lines and 10 6 autologous peripheral blood mononuclear cells irradiated at 80 and 50 grays, respectively. About 300,000 responder cells from the primary mixed lymphocyte cultures were subsequently stimulated weekly under the same conditions in the presence of 30 units/ml recombinant interleukin 2 (a kind gift of Hoffmann-LaRoche). Alternative stimulation of mixed lymphocyte cultures with two different B*2704-positive lymphoblastoid cell lines, KNE (A1, A2, B8, B*2704, DR2, DR3) and WEWAK I (A11, A24, B62, B*2704, Cw2, Cw4, DR2) was used to improve the yield of B27specific CTLs by minimizing restimulation of T cells specific for non-B27 alloantigens. T-cell clones were obtained by limiting dilution, seeding serial dilutions of stimulated T cells in 96-well plates containing 2,000 irradiated stimulator lymphoblastoid cells/well and 20,000 irradiated feeder peripheral blood mononuclear cells/well in the presence of 30 units/ml recombinant interleukin 2. Cells in wells growing below the statistical limit for clonality were screened for HLA-B27 reactivity using a standard 51 Cr release cytotoxicity assay (37) against B*2704-C1R targets, using untransfected C1R cells as a negative control.
Mixed lymphocyte cultures and T-cell clones were grown in Iscove's modified Dulbecco's modified Eagle's medium with glutamax I (Invitrogen), supplemented with 100 units/ml penicillin, 0.1 mg/ml streptomycin sulfate, and 0.05 mg/ml gentamicin (all from Sigma) and 15% of myoclone (Invitrogen). T-cell clones were restimulated weekly, as described above, in the presence of recombinant interleukin 2.
The reactivity of T-cell clones with B*2706 was assessed with B*2706-C1R transfectant cells, usually at an effector:target ratio of 1:1, using the same 51 Cr release cytotoxicity assay as described above.

B*2704 and B*2706 Bind Largely Overlapping Peptide
Repertoires-The B*2704-and B*2706-bound peptide pools were isolated from the corresponding C1R transfectant cells and fractionated by HPLC under identical conditions in consecutive runs. Peptide-containing fractions were analyzed by MALDI-TOF MS. The MS spectrum of each HPLC fraction from one subtype was compared with the MS spectrum of the correlative, previous, and following HPLC fraction from the other subtype. This was done to account for slight shifts in retention time between consecutive chromatographic runs. Ion peaks with the same (Ϯ1) mass/charge (m/z) among the HPLC fractions compared were considered to be identical peptides shared by both subtypes. Ion peaks in one HPLC fraction not found in the counterpart from the other molecule were considered to be peptides differentially bound to one subtype.
Of a total of 969 ion peaks from B*2704 and a total of 943 ion peaks from B*2706, 849 (88% and 90%, respectively) were common to both subtypes, 120 ion peaks from B*2704 (12%) lacked a detectable counterpart in B*2706, and 94 ion peaks from B*2706 (10%) lacked a detectable counterpart in B*2704 (Table I). These results indicate that B*2704 and B*2706 share about 90% of their peptide repertoires, and each subtype binds about 10 -12% of peptides that are not found in the other subtype.
To identify peptides common to both molecules but much more abundant in one of them, we selected ion peaks in each HPLC fraction whose intensity was Ͼ50% of the maximum signal intensity in that fraction. Their amount was measured as the total number of millivolts corresponding to each ion peak in all HPLC fractions in which it was detected. When the total intensity of a given ion peak was more than 10 times higher in one molecule than in the other molecule, the corresponding peptide was assigned as a quantitative difference. Of 218 peptides compared using these criteria, 15 (7%) predominated in B*2704, and 13 (6%) predominated in B*2706. This result suggests that, in addition to determining differential binding of some peptides, B*2704/B*2706 polymorphism also influences the amount of bound peptide in at least an additional 13% of the shared ligands.
The size of B*2704-and B*2706-bound peptide ligands showed a very similar Gaussian distribution, with a mean peptide mass ([M ϩ H] ϩ ) of 1139 and 1128 Da, respectively (Table I). However the size distribution of peptides differentially bound to each subtype showed some significant differences: whereas in the lower molecular mass range (850 -1,100 Da), B*2706-specific peptides predominated over B*2704bound peptides, the opposite was observed in the higher molecular mass range (Fig. 1). Thus, the mean molecular mass ([M ϩ H] ϩ ) of B*2704 and B*2706 peptide differences was 1,280 and 1,200 Da, respectively. The 80-Da difference is compatible with slightly longer length of peptides differentially bound to B*2704 and/or bulkier amino acid side chains at some position(s).

Peptides with Arg 3 Residues Are Disfavored in B*2704 and
Suitable for B*2706 -P3 is an important anchor position for peptide binding to HLA-B27, second only to P2 and P⍀ (22,24,38). To confirm the differential suitability of basic P3 residues in B*2704 and B*2706 suggested by peptide sequencing, three B*2706-specific ligands with Arg 3 and peptide analogs containing Val 3 or Phe 3 were tested for binding to B*2704 and B*2706 in an epitope stabilization assay using RMA-S transfectant cells. For all three ligands, substitution of Val 3 or Phe 3 for Arg 3 improved binding to B*2704, but not to B*2706 ( Fig. 3; Table  II). The magnitude of the effect on B*2704 was somewhat variable depending on each particular ligand, presumably reflecting the contribution of other anchor positions in each peptide. Phe 3 was slightly disfavored in B*2706, relative to Arg or Val in this position. These results indicate that Arg 3 is disfavored in B*2704, but not in B*2706, relative to nonpolar residues.
High Allospecific Epitope Sharing between B*2704 and B*2706 -Because most alloreactive CTLs recognize peptides naturally bound to the alloantigen molecule, another way to assess the overlap between the peptides bound to B*2704 and B*2706 was to test their cross-reactivity with allospecific CTLs. Thus, 56 CTL clones were raised from five unrelated HLA-B27negative donors against B*2704 and tested for their recognition of B*2706-CIR target cells (Table III). Of the CTLs tested, 77% cross-reacted totally (Ͼ60% relative lysis) or partially (30 -60% relative lysis) with B*2706, whereas 23% of the CTLs showed little or no cross-reaction (Ͻ30% relative lysis) with this allotype. The results correlate well with the peptide overlap estimated by direct biochemical analysis. In addition, they suggest that most of the shared ligands between B*2704 and B*2706  are antigenically similar in both contexts, as assessed with allospecific CTLs.

DISCUSSION
Previous studies from our laboratory, based largely on pool sequencing (23), showed that both B*2704 and B*2706 bind peptides with aliphatic or aromatic C-terminal residues. However, whereas B*2704 could accept C-terminal Tyr, B*2706 specificity was restricted to C-terminal aliphatic and Phe residues. This was confirmed in the present study by showing that all shared ligands between both subtypes had C-terminal aliphatic or Phe residues, whereas B*2704-specific ligands had Tyr or Arg. The presence of Arg as a C-terminal peptide motif of B*2704 had gone undetected in previous studies, probably because it is present in a small proportion (in our study, in 3 of 35 sequenced ligands) of the B*2704 peptide repertoire. However, peptides with C-terminal Arg accounted for 33% (3 of 9) of the natural ligands differentially bound to B*2704.
Previous sequencing studies also failed to detect any other differential peptide motif between B*2704 and B*2706 but suggested an increased frequency of Lys 3 and Lys 7 among B*2706 ligands. In vitro binding studies using poly(Ala) peptide analogs also revealed that these two subtypes differed in their P3 residue specificity, with better acceptance of basic P3 residues by B*2706 (24). However the effect of this modulation on subtype-bound peptide repertoires in vivo could not be assessed from these studies.
Our results now clearly establish that B*2706 polymorphism positively selects for peptides with basic P3 and/or P7 residues. Thus, of the five sequenced B*2706-specific peptides, three had basic residues at both P3 and P7, and all five had a basic residue in at least one of these two positions. Among shared ligands, only 5 of 25 peptides had a basic P3 or P7 residue, and none had basic residues at both positions. Similarly, only two of the B*2704-specific peptides had a basic P7 residue, and none had a basic residue at P3.
These differences presumably account for the bigger mean size of B*2704-specific peptides. For instance, the mean residual mass of the C-terminal residues was about 47 Da higher for the nine sequenced B*2704-specific peptides (three with Arg and six with Tyr) than for the five sequenced B*2706-specific ones (all with Leu or Ile). Thus, our data do not support the possibility that B*2704 might have some specific preference for unusually long peptides. Indeed, eight of the nine characterized B*2704 ligands absent from B*2706 had the canonic length of major histocompatibility complex class I ligands: 9 or 10 amino acid residues. However, the mean molecular mass of these differentially bound peptides (1,249 Da) was 188 Da higher than that of the five sequenced peptides differentially bound by B*2706 (1,061 Da). The differential binding of a 13-mer to B*2704 can be explained just by the presence of a C-terminal Arg, which is disfavored in B*2706, rather than by differential size preferences. Recent evidence indicates that B*2704 and B*2706 do not differ from B*2705 or from each other in their tapasin dependence for peptide binding, 2 thus supporting the view that B*2704 does not have a particular preference for suboptimal peptide ligands.
The molecular basis for the restrictions in peptide specificity imposed by B*2706 polymorphism can be deduced from previous crystallographic and peptide binding studies. The greatly increased preference for nonpolar C-terminal residues is explained by the loss of an acidic charge in the F pocket as a consequence of the D116Y change in B*2706. In particular, C-terminal Leu was greatly favored over Arg or Tyr for in vitro binding of peptide analogs to a B*2705 mutant carrying the D116Y mutation (29). Different mutations at this same position in HLA-B27 also increased the preference for nonpolar C-terminal residues (39 -41). In contrast, introducing acidic charges by the H114D mutation in B*2706 has a rather moderate effect on C-terminal residue specificity (29). However, residue 114 takes part in both the D and E pockets, which bind P3 and P7 residues, respectively, an observation that easily explains the 2 A. W. Purcell, personal communication.   Table II. increased allowance for basic residues at these positions in B*2706. This bias does not impair binding of other residues at these positions due to the plasticity of interactions in these pockets, conferred in part through involvement of water molecules (42).
Several questions concerning the peptide specificity of B*2704 and B*2706 may be relevant to the differential association of these subtypes with AS. First, how do the differences in residue specificity translate in the degree of overlap of the peptide repertoires? Second, what structural features of B*2704 ligands impair binding to B*2706? Third, do the ligands common to B*2704 and B*2706 maintain their antigenic features in the context of both subtypes?
The observation that B*2706 binds in vivo about 90% of the B*2704-bound peptide repertoire suggests that putative arthritogenic peptides may be confined to a relatively small portion of B*2704 ligands. Our study indicated that the major feature of B*2704 ligands that impairs binding to B*2706 is the presence of C-terminal Arg or Tyr. In contrast, peptides with basic residues at both P3 and P7 do not bind B*2704, despite appropriate P2 and P⍀ motifs. Moreover, some peptides with Cterminal Leu and a basic residue at only P3 or P7 did not bind B*2704, revealing an additional contribution of other residues.
An important issue when trying to correlate peptide specificity with disease association is whether shared ligands between subtypes differentially associated with AS maintain their antigenic features in the context of both subtypes. It is possible that particular peptides can be differentially recognized by CTLs when presented by either B*2704 or B*2706. However, the level of cross-reaction of alloreactive CTLs raised against B*2704 with B*2706 (77%) was only about 13% lower than the overlap of peptide repertoires, suggesting that a majority of the shared ligands maintain their antigenic features on both subtypes.
It was reported previously that B*2707, a disease-associated subtype, was unable to bind peptides with C-terminal Tyr (43), which questioned the importance of this motif for determining susceptibility to AS. This conclusion was based on the presence of Tyr 116 in B*2707, the absence of a C-terminal Tyr motif by pool sequencing, and limited sequencing of individual ligands. However, none of these features per se rule out the possibility that some peptides containing C-terminal Tyr may bind B*2707. Much more extensive sequencing of individual B*2707 ligands would be required to assess this issue.
In conclusion, this study allowed us to correlate the lack or low association of B*2706 with AS to failure of this allotype to bind a relatively small portion of the peptide repertoire bound by the structurally closest disease-associated allotype B*2704 and to determine the major structural features of the differentially bound peptides. No other known structural or functional feature of B*2704 and B*2706 can be correlated with differential association of these subtypes with AS.