HLA-B27 Misfolding Is Associated with Aberrant Intermolecular Disulfide Bond Formation (Dimerization) in the Endoplasmic Reticulum*

The class I protein HLA-B27 confers susceptibility to inflammatory arthritis in humans and when overexpressed in rodents for reasons that remain unclear. We demonstrated previously that HLA-B27 heavy chains (HC) undergo endoplasmic reticulum (ER)-associated degradation. We report here that HLA-B27 HC also forms two types of aberrant disulfide-linked complexes (dimers) during the folding and assembly process that can be distinguished by conformation-sensitive antibodies W6/32 and HC10. HC10-reactive dimers form immediately after HC synthesis in the ER and constitute at least 25% of the HC pool, whereas W6/32-reactive dimers appear several hours later and represent less than 10% of the folded HC. HC10-reactive dimers accumulate in the absence of tapasin or β2-microglobulin, whereas W6/32-reactive dimers are not detected. Efficient formation of W6/32-reactive dimers appears to depend on the transporter associated with antigen processing, tapasin, and β2-microglobulin. The unpaired Cys67 and residues at the base of the B pocket that dramatically impair HLA-B27 HC folding are critical for the formation of HC10-reactive ER dimers. Although certain other alleles also form dimers late in the assembly pathway, ER dimerization of HLA-B27 may be unique. These results demonstrate that residues comprising the HLA-B27 B pocket result in aberrant HC folding and disulfide bond formation, and thus confer unusual properties on this molecule that are unrelated to peptide selection per se, yet may be important in disease pathogenesis.

The proper folding and assembly of proteins in the endoplasmic reticulum (ER) 1 is dependent on a number of chaperones, and is subject to stringent quality control measures to ensure that improperly folded proteins are destroyed (1). For major histocompatibility complex (MHC) class I complexes, newly synthesized unfolded heavy chains (HC) bind to calnexin, with subsequent folding, release, and calreticulin binding coincident with ␤ 2 -microglobulin (␤ 2 m) association (2). Further retention of partially folded HC/␤ 2 m heterodimers is mediated by tapasin, an MHC class I-specific chaperone which forms a bridge between HC and TAP (transporter associated with antigen processing) in the peptide loading complex. Tapasin-mediated retention serves to optimize peptide loading and thus stability, so that complexes can traffic to the cell surface and display peptides without dissociating prematurely. An important component of HC folding and assembly is the formation of correct intrachain disulfide bonds (3), as evidenced by mutations in critical Cys residues that interfere with class I expression and antigen presentation (4). For MHC class I proteins this appears to be mediated by ERp57, a thiol oxidoreductase that catalyzes disulfide bond formation and rearrangement. ERp57 has been found associated with calnexin and calreticulin (5), and is also part of the peptide-loading complex (6 -8) where it is linked to tapasin via a disulfide bond (9).
When class I HC are synthesized in the absence of ␤ 2 m or peptide, they misfold and are dislocated from the ER into the cytosol where they are degraded in a process known as ERassociated degradation (10). When certain mouse class I alleles are expressed in ␤ 2 m-deficient cells, misfolding is associated with the formation of HC-HC homodimers in the ER via an unpaired Cys in the cytoplasmic tail (11). In these studies HC dimerization was also observed in cells expressing ␤ 2 m, but only after ␤ 2 m had dissociated from previously assembled complexes. This led to the suggestion that HC dimerization might be a mechanism by which dysfunctional molecules are removed (11).
HLA-B27 is an unusual allele in that it misfolds even in the presence of a normal supply of ␤ 2 m and peptide (12). This characteristic is related to the composition of its B pocket, a region of the peptide-binding groove that plays a dominant role in peptide selection (13,14), but also has a dramatic influence on HC folding rate (12). The rim of the B pocket also contains an unpaired and reactive Cys residue (Cys 67 ) (15,16), which is relatively uncommon among MHC class I alleles (HLA-B27, -B38, -B39, -B14, -B15, and B73) (17). Interestingly, Cys 67 appears to allow HLA-B27 HC to dimerize when they are refolded in vitro in the absence of ␤ 2 m, a phenomenon that was also observed in TAP-deficient T2 cells expressing HLA-B27 (18). This residue also has an important affect on the stability of HLA-B27 HC⅐peptide⅐␤ 2 m complexes (19).
These findings have been of interest particularly because HLA-B27 is highly associated with susceptibility to ankylosing spondylitis and other human spondyloarthropathies (20), and can cause spondyloarthropathy-like inflammatory conditions when overexpressed in rodents (21,22). Several hypotheses have been proposed to explain these findings, but as yet the mechanism is unclear. Whereas arthritogenic peptide presentation has been suspected for some time, there is limited support for this hypothesis (23), and in instances where autoreactivity has been found, it is not clear whether this is either specific for, or initiated by, HLA-B27 (24,25). In HLA-B27 transgenic mice, spontaneous arthritis develops in the absence of ␤ 2 m (22), indicating that cell surface expression of "classical," trimeric HLA-B27 HC⅐peptide⅐␤ 2 m complexes is not required. Interestingly, the same spontaneous arthritis phenotype has been reported in the absence of HLA-B27, in either ␤ 2 m or TAP-deficient mice with a mixed genetic background (26). These studies raise the possibility that class I HC misfolding in general, and intracellular sequelae might be a critical component of the disease mechanism, perhaps through the generation of ER stress (27). Along these lines, effects of HLA-B27 expression in monocytes and epithelial cells that cannot be explained by its function as an antigen presenting molecule have been described (28 -30). Other ideas focusing on T cell recognition of dimers as aberrant structures have also been proposed (31,32). These observations raise several questions, in particular whether HLA-B27 dimerization occurs in cells with an intact antigen presentation pathway and is related to events occurring in the ER. The results presented here indicate that aberrant disulfide-linked HLA-B27 HC include both "folded" and "unfolded" structures based on differential recognition by W6/32 and HC10. Dimerization of HC in the ER requires Cys 67 as well as other B pocket residues that contribute to their prolonged retention. Our studies emphasize abnormalities in the folding of HLA-B27 that may be unique to this allele, and could be involved in the pathogenesis of spondyloarthropathies.

EXPERIMENTAL PROCEDURES
DNA and Cell Lines-B27.A23 and B27.E45M were constructed from a genomic clone of HLA-B*2705 by site-directed mutagenesis (Altered Sites, Promega, Madison, WI). B27.A23 has three amino acid substitutions: H9F (F for H at position 9), T24A, and E45M. B27.C67A was kindly provided by Joel Taurog, and was also produced from genomic HLA-B*2705. B27.A2B contains six amino acid substitutions (H9F, T24A, E45M, I66K, C67V, and K70H) and has been described previously (12,14). The tapasin cDNA was also made by reverse transcriptase-PCR using oligonucleotides designed from the published sequence (33), and is the allele with Thr at position 240 of the mature protein (34).
Metabolic Labeling-Cells (2 ϫ 10 7 /time point) were preincubated in Met/Cys-deficient R-10 media for 1 h at 37°C, pulse-labeled with 2 mCi of [ 35 S]Met/Cys (Easytag, PerkinElmer Life Sciences), and then chased in a 10-fold excess of media containing 2 mM Met/Cys. The length of pulses for various experiments is indicated in the figure legends. At the end of each time point, cells were harvested by centrifugation after the addition of 5 volumes of ice-cold PBS. Cell pellets were lysed in 500 l of lysis buffer (PLB; 20 mM Tris (pH 7.8), 100 mM NaCl, 10 mM EDTA, and 1% Triton X-100, supplemented with 5 mM iodoacetamide, 0.5 mM phenylmethylsulfonyl fluoride, and 0.1 mM N ␣ -p-tosyl-L-lysine chloromethyl ketone (Sigma)) for 20 min on ice. Post-lysis nuclei were removed with a 10,000 ϫ g spin for 5 min at 4°C, and the supernatants were used for immunoprecipitations.
Biotinylation-Cells were rinsed in ice-cold PBS (pH 7.2), then resuspended at 5 ϫ 10 7 per ml. To 800 l of cell suspension was added 400 l of freshly made Sulfo-NHS-biotin (Pierce) solution (1.5 mg/ml in PBS at pH 8.0). This solution was incubated for 30 min at room temperature with rocking. Unconjugated reagent was removed with several washes using ice-cold PBS, and the labeled cells were subsequently lysed and used for immunopreciptations.
Immunoprecipitations-Immunoprecipitations were performed in the following manner unless otherwise stated. Cell lysates were precleared with washed formalin-fixed Staphylococcus aureus (Sigma), and incubated with purified monoclonal antibody (15 g per 2 ϫ 10 7 cells in 0.5 ml) for 1 h at 4°C, followed by the addition of Protein A-Sepharose (100 l of a 50 mg/ml suspension in PLB) (Sigma) and another hour at 4°C. Protein A-Sepharose pellets were washed consecutively with PLB containing 0.1% SDS and 1% bovine serum albumin, a 10-fold dilution of PLB supplemented to 80 mM NaCl, and PLB alone, and stored at Ϫ20°C until electrophoresis.
Gel Electrophoresis and Phosphorimaging-Isoelectric focusing (IEF) and SDS-PAGE (10.5% gels) were performed as described previously (12). For reducing and nonreducing SDS-PAGE, samples were boiled for 5 min in an equal volume of 2 ϫ sample buffer with or without dithiothreitol (200 mM), respectively, prior to electrophoresis. For phosphorimage analysis, gels were fixed in 10% acetic acid for 30 min, then dried. Dried gels were exposed to PhosphorImager plates for 72 h, and 35 S-labeled proteins were visualized and quantitated using Image-Quant software (Molecular Dynamics, Sunnyvale, CA).
Immunoblotting-Proteins were transferred from SDS-polyacrylamide gels to polyvinylidene difluoride membranes (Schleicher & Schuell) for 1 h at 500 mA in buffer containing 25 mM Tris, 200 mM glycine, and 10% methanol. Post-transfer, nonspecific binding was blocked with 1% casein (Sigma) in PBS overnight. Membranes were then incubated for 1 h with 3B10.7 at a final concentration of 1 g/ml. After washing, alkaline phosphatase-conjugated species-specific secondary antibody (Southern BioTechnology Associates, Birmingham, AL) was added for 1 h at a 1:1000 dilution. For detection of biotinylated proteins, alkaline phosphatase-conjugated streptavidin was used at a dilution of 1:3000. Proteins were visualized using 5-bromo-4-chloro-3indolyl phosphate/nitro blue tetrazolium for color development (Sigma). Mock immunoprecipitations were performed with primary antibody in PLB (no cells) followed by Protein A-Sepharose to identify bands arising from cross-reactivity between the alkaline phosphatase-conjugated secondary antibody and the primary antibody used for immunoprecipitation. Unless otherwise noted, material immunoprecipitated from 2 ϫ 10 7 cells is represented in each lane.

RESULTS
HLA-B27 HC Form W6/32 and HC10-reactive Disulfidelinked Complexes in Cells-To determine whether HLA-B27 HC form aberrant disulfide bound complexes in cells with an intact class I assembly pathway, class I molecules were immunoprecipitated from C1R.B27, separated on SDS-PAGE under nonreducing and reducing conditions, and visualized by immunoblotting. High molecular weight (hMW) HC complexes are seen migrating at ϳ90 kDa in addition to monomers at ϳ45 kDa ( Fig. 1A, NRed). The absence of the hMW bands in untransfected C1R and their elimination by sample reduction prior to electrophoresis (Fig. 1A, red) indicates they are indeed HLA-B27 HC. These hMW complexes are also seen when cell lysates are applied immediately to SDS-PAGE, indicating that they are not an artifact of immunoprecipitation (data not shown). To prevent post-lysis disulfide bond formation which can occur with HLA class I molecules, particularly during prolonged purification protocols (44,45), iodoacetamide is included in the cell lysis and sample buffers for all experiments reported here.
W6/32 and HC10 recognize largely distinct populations of HLA class I molecules (38,40). W6/32 is conformation dependent, recognizing folded complexes containing peptide. In contrast, HC10 recognizes free HC not associated with ␤ 2 m. Thus we refer to W6/32 and HC10-reactive HC as folded and unfolded, respectively, although the degree to which HC10-reactive HC are unfolded is not known (see "Discussion"). The heterogeneity of the hMW complexes recognized by HC10 in comparison to W6/32 ( Fig. 1A) together with sequential immunoprecipitations where W6/32-or HC10-reactive material remains after pre-clearing with the reciprocal antibody ( Fig. 1B), reveals that these complexes also contain distinct groups of HLA-B27 molecules. 5H7, which recognizes a folded ␣3 epitope, does not react with the more heterogeneous HC10-reactive material (Fig. 1A), raising the possibility that the ␣3 domain is not properly folded or may be blocked by another protein interacting with free HC. Based on the apparent M r (ϳ90,000) of the homogeneous band immunoprecipitated by W6/32 and 5H7, its disappearance with sample reduction, and the previous report of HLA-B27 dimerization in vitro (18), we conclude that it is a HC-HC dimer linked by a disulfide bond. Consistent with this, purified W6/32-reactive material reveals only two protein bands (HC and ␤ 2 m) following sample reduction (data not shown). However, we have not completely ruled out the possibility that the hMW complexes, particularly those recognized by HC10, might contain an additional protein (or proteins) (see "Discussion").
A small percentage of cell surface HLA-B27 complexes are recognized by the monoclonal antibody MARB4. When purified from whole cell lysates, these were found to contain peptides of a much broader size range than the canonical 8 -11 residues, including peptides up to 33 amino acids in length (43). Immunoprecipitations with MARB4 reveal that this antibody recognizes dimers as well as monomers (Fig. 1C). It is worth noting that the relative intensity of the 90-kDa band compared with the monomer is greater with MARB4 than for other monoclonal antibodies, suggesting that a higher proportion of MARB4reactive molecules are dimers. The smaller quantity of HC immunoprecipitated with MARB4 (per cell equivalent) is consistent with the lower expression of MARB4-reactive molecules (43) (and data not shown). The immunoreactive material in the mock lanes (no cell lysate) is a result of cross-reactivity with the rabbit anti-mouse IgG used to enhance immunoprecipitation. MARB4-reactive dimers can be immunoprecipitated without addition of this cross-linking antibody if more cells are used, whereas MARB4 immunoprecipitations of untransfected C1R reveal no dimer band (data not shown). Recognition of HLA-B27 dimers by MARB4 raises the possibility that some may contain long peptides, although this will need to be addressed experimentally.
Taken together, these data indicate that the W6/32-reactive HLA-B27 HC dimers reported previously to form in the absence of ␤ 2 m or peptide (18)  Each lane represents material immunoprecipitated from 2 ϫ 10 7 cells. B, sequential immunoprecipitations were performed on C1R.B27 lysates (2 ϫ 10 7 cells). For each round (1-4), lysates were incubated with primary antibody for 1 h, followed by protein A-Sepharose for another hour. Immunoprecipitated material was separated by SDS-PAGE under nonreducing conditions, and immunoblotted with 3B10.7. C, immunoprecipitations were performed on lysis buffer alone (Mock) or lysate from 6 ϫ 10 7 C1R.B27 cells using MARB4 followed by a cross-linking antibody (rabbit anti-mouse IgG) and then protein A-Sepharose. SDS-PAGE of immunoprecipitates was performed under nonreducing (NRed) and reducing (Red) conditions, and immunoblotted with 3B10.7. Prestained molecular weight markers are shown (Stds). assembly pathway. Furthermore, there appears to be a distinct pool of dimers that are reactive with HC10 but not W6/32, and thus are unfolded or possibly misfolded.
Kinetics of HLA-B27 HC Dimerization-To assess the kinetics of HLA-B27 dimerization cells were pulsed with [ 35 S]Met/ Cys, then chased in media containing excess nonradioactive Met/Cys, and lysed in buffer containing Triton X-100 to dissociate HC from weakly interacting proteins including the peptide loading complex (46). HC10-reactive dimers are present after a 30-min labeling period (0 h chase), and account for ϳ30 -40% of the total HC10-reactive HC ( Fig. 2A) based on PhosphorImager quantitation (data not shown). It should be noted that this is a minimum estimate, as some of the HC monomer band in C1R.B27 derives from HLA-C (35), whereas this allele does not contribute to the dimer bands ( Fig. 2A, C1R  lane). Dimers and monomers decay at approximately the same rate during the 4-h chase, and are initially sensitive to Endo H (Fig. 2B). However, after longer chase periods (Ͼ6 h) Endo H-resistant HC10-reactive dimers can be detected (data not shown). There is a faint band migrating near the bottom of the hMW HC (4 h lane, bracket) that is not seen in untransfected C1R and remains after sample reduction (1.5 h lane, red). This band is not HC as it is not reactive with anti-HC antibody in immunoblots (Fig. 1A). It has been tentatively identified as the ER chaperone BiP/GRP78. 2 The rapid formation and initial sensitivity of HC10-reactive HLA-B27 dimers to Endo H indicates they form in the ER shortly after HC synthesis.
W6/32-reactive dimers were not detected during the time course shown in Fig. 2A (data not shown), so cells were labeled and chased for a longer period to enhance detection. Dimers representing up to ϳ10% of the total W6/32-precipitable HC were detected after a 12-h labeling period (Fig. 2C) that decayed with kinetics similar to folded monomer HC. Folded dimers were found to be Endo H-resistant in immunoblotting experiments (Fig. 2B), indicating they have been sialylated. Both folded and unfolded dimers can be labeled with a cellimpermeable biotinylating reagent (Fig. 2D), indicating that at least some are on the cell surface. It should be noted that W6/32 has been shown to react poorly with biotinylated HLA-B27 (47) (and Fig. 2D) so that 5H7 was used for this experiment. These results indicate that the kinetics of formation of HC10-and W6/32-reactive HC dimers are dramatically different, and thus provide further evidence that they represent distinct pools of aberrant disulfide-linked HLA-B27 complexes.
The Role of TAP, Tapasin, and ␤ 2 -Microglobulin in the Formation of HLA-B27 Dimers-To determine whether the presence of an intact peptide loading complex influences HLA-B27 dimerization, we examined HLA-B27 expression in TAP1/ TAP2-deficient (T2), tapasin-deficient (721.220), and ␤ 2 m-deficient cells. Neither W6/32-nor HC10-reactive dimers are detected in immunoprecipitates from T2 cells that have been transfected with HLA-B27 (T2.B27) or cDNAs encoding TAP1 and TAP2 (T2.TAP) (Fig. 3A). However, when HLA-B27 is expressed in T2.TAP, a faint W6/32-reactive dimer band is detected, whereas HC10-reactive dimers are not seen. This observation was confirmed in independent transfectants using FLAG-tagged HLA-B27 (data not shown). It should be noted that the overall recovery of HC from these transfectants reflects the effect of TAP on the formation of stable peptideloaded complexes. We considered the possibility that loading material derived from more cells might allow the detection of dimers. Although no W6/32-reactive dimer band could be detected even when 4 times as many cells were used, HC10reactive material from T2.B27 and T2.TAP.B27 became apparent (data not shown).
The TAP dependence for formation of W6/32-reactive dimers is in contrast to a previous report (18), so we sought to confirm these findings in a separate cell line. T5-1 cells which are a B-lymphoblastoid line expressing HLA-B27 as well as -A1, -A2, and -B8 alleles were examined (48). This cell line has been mutagenized, and derivatives lacking MHC genes have been selected in a fashion analogous to the generation of T2 (36). 8.1.6 are deficient in the MHC region encoding TAP1 and TAP2 on one chromosome, whereas 5.2.4 lack both copies of TAP1/ TAP2 as well as HLA-A1 and -B8, but still express HLA-A2 and -B27. W6/32-and HC10-reactive dimers are found in T5-1 and 8.1.6, but not 5.2.4 (Fig. 3B). However, similar to T2.B27, HC10-reactive dimers can be detected if more 5.2.4 cells are used (data not shown). ME.1 immunoprecipitations proved that the folded dimers are indeed HLA-B27 HC (data not shown), because this antibody does not recognize the other alleles expressed in T5-1. The results with W6/32 are consistent with those obtained in T2 transfectants, and suggest that formation of folded HLA-B27 dimers requires the transporter associated with antigen processing. HC10-reactive dimers are more readily detected in T5-1 and 8.1.6 cells for reasons that are unclear.
When HLA-B27 is expressed in 721.220 cells (.220.B27) which express only small amounts of a truncated form of tapasin (34), W6/32-reactive dimers are not detected, but in contrast to T2, HC10-reactive dimers accumulate (Fig. 3C). Reexpression of tapasin (.220.Tsn.B27) enables folded dimers to be recovered and substantially reduces the pool of unfolded dimers (Fig. 3C). As expected it also enhances the recovery of W6/32-reactive monomers (i.e. single-HC⅐peptide⅐␤ 2 m complexes), and reduces the accumulation of HC10-reactive material. The proportion of dimerized HLA-B27 HC appears to be greater in cells lacking tapasin (Fig. 3C) than in TAP-deficient cells (Fig. 3, A and B), TAP-transported peptides may actually play a role in the formation and/or stabilization of HC10-reacitve dimers.
Expression of HLA-B27 HC in the absence of ␤ 2 m eliminates the formation of W6/32-reactive dimers, and only a very faint monomer band can be detected (Fig. 3D). The absence of folded monomers is expected because ␤ 2 m is critical for proper HC folding and formation of the epitope recognized by W6/32. In striking contrast, HC10-reactive dimers and monomers are recovered from ␤ 2 m-deficient cells in proportions similar to what is found when HLA-B27 is expressed in the absence of tapasin (Fig. 3C). Taken together, these results suggest that ␤ 2 m, tapasin, and TAP are all required for the formation of W6/32-reactive HLA-B27 dimers, and at least for ␤ 2 m and tapasin, unfolded dimers tend to accumulate in their absence.
ER Retention and Cys 67 Contribute to HLA-B27 Dimerization-Previously we demonstrated that substituting 6 amino acids in HLA-B27 to create an HLA-A2-like B pocket (referred to as B27.A2B) prevents misfolding and ER-associated degradation of HC (12). To determine whether these structural changes affect HC dimerization, we examined B27.A2B and three other mutants with B pocket substitutions. Replacing the entire B pocket (B27.A2B) prevents the accumulation of both unfolded and folded dimers (Fig. 4A). This can also be achieved by replacing only Cys 67 with Ala (B27.C67A), which confirms previous results showing that this residue is necessary for HLA-B27 HC dimerization in vitro (18). Surprisingly, substituting residues at the base of the B pocket without replacing Cys 67 also affects dimerization. For B27.E45M no dimers are detected with immunoblotting, whereas for B27.A23 a faint dimer band can be seen.
To further investigate why dimerization is affected in the mutants that still contain Cys 67 HC folding efficiency was compared. C1R transfectants were pulsed for 5 min with [ 35 S]Met/Cys and chased for up to 2 h. At each time point, W6/32-and HC10-reactive HC was immunoprecipitated se-quentially from cell lysates, then visualized and quantitated following separation on IEF gels run under reducing conditions. Gel regions from a representative experiment comparing HLA-B27 with B27.E45M, along with quantitative results averaged from two experiments, show that the conversion of HC from an HC10-to W6/32-reactive state is much more rapid for B27.E45M than for HLA-B27 (Fig. 4B). This analysis was applied to each B pocket mutant, and the time required for 75% of newly synthesized HC to fold (become W6/32-reactive) was determined from each graph. Results averaged from two experiments are depicted in Fig. 4C. The data showing HLA-B27 and B27.A2B folding have been reported previously (12) but are included for purposes of comparison. B27.A23 and B27.E45M fold dramatically faster than wild type HLA-B27 or B27.C67A, and are similar to B27.A2B. The rapid folding molecules have in common the substitution of Met for Glu at position 45 at the base of the B pocket. The Glu substitution removes an acidic residue that is expected to be predominantly negatively charged at neutral pH, and is neutralized when HLA-B27 binds a peptide with Arg at P2 (49). These results indicate that at least in the context of the HLA-B27, Glu 45 has a dramatic effect on HC folding rate, increasing the amount of time in which HC exists in the ER in an unfolded state.
Although immunoblotting experiments do not reveal dimers in cells expressing B27.E45M (Fig. 4), it was possible that they might form but fail to accumulate because of rapid degradation. To test this, cells expressing HLA-B27 and B27.E45M were labeled with [ 35 S]Met/Cys for 2 h, then chased for up to 19 h, and the presence of unfolded and folded dimers analyzed by nonreducing SDS-PAGE. Prominent unfolded hMW HC bands representing ϳ25% of the total radioactive HC (unfolded and folded) are seen in HC10 immunoprecipitations of HLA-B27 at time 0 (Fig. 5) (top panel, left bracket), whereas only a faint band is observed for B27.E45M (bottom panel). A band representing folded W6/32-reactive dimers that contains ϳ7% of the total labeled HC is seen after 6 h of chase for HLA-B27, whereas this is not apparent in B27.E45M immunoprecipitates. We have used this assay to examine B27.A23 and find that like B27.E45M, it does not form HC10-reactive dimers in the ER, however, W6/32-reactive dimers can be detected after 6 h of chase (data not shown). Considered together, the results shown in Figs. 4 and 5 demonstrate that there are two characteristics contributing to the formation of HLA-B27 dimers in the ER. Cys 67 is necessary but not sufficient, as B27.E45M does not dimerize. Impaired folding is necessary but not sufficient, as B27.C67A does not dimerize. It appears that the formation of aberrant disulfide bonds between HLA-B27 HC in the ER results from the combination of impaired folding with ER retention of HC in the presence of the unpaired Cys 67 .
Dimer Formation in B-LCL and by Other HLA Class I Molecules-In addition to T5-1, we find dimers in other B cell lines (Fig. 6, HF and BL). Because these cells express other class I alleles in addition to HLA-B27, we confirmed that the dimers can be immunoprecipitated by ME.1 (data not shown). However, because BL express HLA-B8 and HF express HLA-B7, the latter of which is recognized by ME.1, we tested these alleles along with HLA-A2 and -B53 in C1R cells. Whereas we find no evidence of dimers for HLA-B8, -A2, and -B53, a dimer-sized band can be detected with HLA-B7 (Fig. 6). Thus, the dimer band in HF could represent HLA-B27 or -B7 (or both). We have tested several additional cell lines including Jurkat, U937, and HeLa, and find no evidence of HLA class I HC dimerization (data not shown). Jurkat express HLA-A3, A9, B16, and B35 (50), although other references indicate they express HLA-B7 (51,52). U937 express HLA-A3, -A26, -B18, and -B51 (53). Thus, whereas class I HC dimerization is not unique to HLA-B27, it does not appear to be a generalized phenomenon.
HLA-B7 does not contain an unpaired Cys in its extracellular domain, so we considered the possibility that it might differ from HLA-B27 in the formation of dimers. Indeed, when HLA-B7 HC are labeled and chased in an experiment identical to that shown in Fig. 5, less than 5% of total labeled HC appear in the hMW region of the HC10 gel (Fig. 6B). However, folded W6/32-reactive dimers appear with kinetics similar to HLA-B27 and the mutant B27.A23, and constitute ϳ8% of the total HC at 6 h of chase. It is also worth noting that approximately two-thirds of HLA-B7 HC synthesized in the 2-h labeling period are folded, which is similar to B27.E45M and dramatically different from HLA-B27 where only one-third (34%) are folded (Fig. 5). The formation of W6/32-reactive dimers late in the assembly process (perhaps at the cell surface), by HLA-B7 and B27.A23 that do not first form HC10-reactive dimers in the ER, emphasizes that ER dimerization is not required. Background-subtracted radioactivity in monomer (Mon) and dimer (hMW) bands was quantitated, and is expressed below each gel as a percentage of the total HC (W6/32 and HC10). Brackets and arrows indicate the regions that were quantitated. Note that for purposes of quantitation only the upper two hMW HC bands were assessed, as the lowest band does not always separate from a co-precipitating protein (see Fig. 2A). Asterisks indicate nonspecific bands. Ϫ indicates that radioactivity was Ͻ5% of total.
FIG. 6. Specificity of dimer formation. A, HLA class I complexes were immunoprecipitated with W6/32 from B lymphoblastoid cell lines (B-LCL: HF and BL) expressing HLA-B27 as well as other alleles, and from C1R transfected with different HLA alleles, separated by SDS-PAGE under nonreducing conditions, and HC visualized by immunoblotting with 3B10.7. Each lane represents material immunoprecipitated from 2 ϫ 10 7 cells. B, C1R.B7 dimerization was analyzed exactly as described in the legend to Fig. 5 for HLA-B27 and B27.E45M. Ϫ indicates that radioactivity was Ͻ5% of total.

DISCUSSION
Our results indicate that HLA-B27 HC form two types of aberrant disulfide-linked complexes (dimers) that are distinguishable by several criteria, including kinetics and site of formation, migration on nonreducing SDS-PAGE, dependence on ␤ 2 m, TAP, and tapasin, and recognition by monoclonal antibodies HC10 and W6/32. HC10 recognizes newly synthesized HC before they are completely folded and have bound ␤ 2 m (54), free HC on the cell surface that have lost ␤ 2 m (55), and completely denatured HC separated on SDS-polyacrylamide gels (40). In contrast, W6/32 recognizes a conformational epitope on HC⅐peptide⅐␤ 2 m complexes that is for the most part dependent on the presence of ␤ 2 m and peptide. W6/32-reactive complexes are typically referred to as folded, whereas HC10-reactive HC are referred to as free or unfolded (56). The two types of dimeric HLA-B27 HC complexes described here that are distinguished by W6/32 and HC10 have the same characteristics as monomeric HC, and thus we refer to them as folded and unfolded, respectively. Sequential immunoprecipitations indicate that for the most part W6/32 and HC10 are recognizing different pools of HLA-B27 HC dimers, however, we cannot rule out the possibility that a small proportion of dimers may contain both epitopes. Furthermore, HC in W6/32-reactive dimers may be somewhat unfolded relative to W6/32-reactive monomeric HC, given that they have a disulfide bond between two ␣1 domains. Although free HC recognized by HC10 are likely to be more unfolded, we do not intend to imply a specific conformation, nor that HC dimers formed in the ER necessarily have the same conformation as HC10-reactive dimers on the cell surface.
HC10-reactive dimers form in the ER immediately after synthesis and constitute a substantial portion (25%) of the newly synthesized HC pool. In contrast, W6/32-reactive dimers appear several hours after a 2-h pulse, and constitute less than 10% of the total HC pool. Because unfolded dimers decay more rapidly than folded dimers accumulate, it seems likely that a significant fraction is subject to quality control processes where HC dimers are reduced to re-enter the monomer pool and/or degraded. In previous studies where we demonstrated that HLA-B27 HC undergo ER-associated degradation (12), the fraction of newly synthesized HC being degraded was much less than 25%, consistent with the idea that some unfolded dimers re-enter the monomer pool where proper folding can still occur. Whether unfolded dimers can fold without being reduced to monomers is not clear. Although we do not detect folded dimers in the ER (e.g. Endo H-sensitive W6/32-reactive molecules), we cannot rule out the possibility that they form in this compartment at a low rate but then exit rapidly. Even if HC10-reactive dimers can fold productively, our results suggest this is an inefficient process.
The unfolded hMW HC complexes recognized by HC10 typically migrate as three bands between ϳ85 and 105 kDa on nonreducing SDS-PAGE, and these forms accumulate in the absence of ␤ 2 m or tapasin. In other experiments co-expressing wild type and FLAG-tagged HLA-B27 HC, we find wild type HC in anti-FLAG immunoprecipitates, clearly indicating that HC-HC(FLAG) homodimers account for at least part of the hMW complexes. 2 Their molecular weight heterogeneity may be because of a variable number or location of intra-or intermolecular disulfide bonds. However, we have also considered the possibility that one or more of the three bands is an HLA-B27 HC (ϳ45 kDa) disulfide bound to an ER protein of ϳ40 -60 kDa. Indeed, protein folding intermediates have been shown to form transient mixed disulfides with members of the proteindisulfide isomerase family, including protein-disulfide isomerase and ERp57 (57). This has also been demonstrated for MHC class I HC in two studies, one using an in vitro translation system with microsomes (58), and in the other using a highly sensitive immunoblotting technique with 125 I-labeled anti-ER60 (ERp57) (59). In the latter studies, HC⅐ERp57 complexes were estimated to comprise Ͻ1% of the total HC pool, consistent with the notion that they are transient folding intermediates. We have immunoblotted nonreduced HC10 immunoprecipitates with anti-ERp57 antibodies but do not detect ERp57 in any of the three hMW HC bands. (Blots for calreticulin (ϳ54 kDa) and tapasin (ϳ48 kDa) were also negative.) Because the HC10⅐reactive dimer complexes we observe constitute a large portion of the newly synthesized HC pool, decay slowly over several hours, do not appear to contain ERp57, and depend on Cys 67 and impaired folding to form, we conclude that they are not likely to represent intermediates in a normal folding process. It seems more likely that they represent HC-HC homodimers of HLA-B27 with variable disulfide bonds, although further experiments will be necessary to determine their precise composition.
Class I HC dimerization is not a generalized phenomenon, yet it is not unique to HLA-B27. Mouse class I alleles H-2L d , D d , and D b have been shown to dimerize (11), and we demonstrate that this can occur with HLA-B7 HC consistent with a previous report (60). However, there are important differences in that the majority of HLA-B27 HC dimerization occurs in the ER, and ER dimerization of mouse alleles was observed only when ␤ 2 m was absent, a condition that induces HC misfolding (10). In the presence of ␤ 2 m, mouse HC dimerization occurred in a post-ER compartment and was associated with loss of ␤ 2 m. Rapidly folding molecules such as HLA-B7 and the HLA-B27 mutant B27.A23 appear to be more similar to mouse class I in that ER dimerization does not occur in cells expressing ␤ 2 m. Neither HLA-B7 nor the mouse alleles have unpaired Cys residues in their extracellular domain, and indeed mouse HC dimerization appeared to occur through a cytoplasmic tail Cys (11). Taken together these results suggest that the dimerization of HLA-B27 HC in the ER in the presence of an intact class I assembly pathway fundamentally distinguishes HLA-B27 dimerization from what has been observed with other alleles, and may contribute to the unusual tendency of this allele to misfold (12). Our mutagenesis studies identify two characteristics of the HC that are critical for ER dimerization, Cys 67 and impaired folding with ER retention. The requirement for Cys 67 is consistent with previous data suggesting that this residue forms a disulfide link between two HC (18). The importance of impaired HC folding is best exemplified by B27.E45M, which contains a Cys 67 but folds very rapidly and fails to form disulfide-linked HC complexes. These data are consistent with a model where impaired folding of newly synthesized HLA-B27 HC leaves the Cys 67 exposed, and thus prone to form aberrant links to other HC particularly in the oxidizing environment of the ER. Folded dimers could emerge from the unfolded dimer pool, and/or form later after folded monomers (HC/peptide/ ␤ 2 m) lose ␤ 2 m during transport or at the cell surface. The tendency of B27.A23, HLA-B7, and mouse alleles to display the late dimerization phenotype without significant ER dimerization indicates that this pool does not have to arise from ER dimers. B pocket mutants like B27.E45M that fold rapidly and avoid ER dimerization might also avoid late dimerization if the mutation results in greater HC⅐peptide⅐␤ 2 m complex stability compared with wild type HLA-B27 or the B27.A23 mutant.
The reduced accumulation of unfolded HLA-B27 dimers in the absence of TAP (T2.B27 and 5.2.4 cells), and their prominence in the absence of tapasin (.220.B27 cells) or ␤ 2 m, suggests that TAP-transported peptides may contribute to their formation and/or stabilization, whereas tapasin facilitates their elimination. In support of this, it is clear from several studies that TAP-transported peptides can stabilize and be presented by monomeric HLA-A2 and -B27 complexes in the absence of tapasin (61)(62)(63)(64). Recent studies have shown that tapasin forms a disulfide link with ERp57 in the peptide loading complex (9). Disruption of this link by mutating the Cys 95 in tapasin prevents complete oxidation of the class I HC at the conserved Cys 101 /Cys 164 disulfide, and results in poor peptide loading of HC/␤ 2 m heterodimers (9). Thus it is possible that the effect of tapasin on reducing the accumulation of unfolded HLA-B27 HC dimers that we observe may be via recruitment into the peptide-loading complex where they are subject to further quality control. This is consistent with the effect of ␤ 2 m deficiency on the accumulation of HC10-reactive dimers, because ␤ 2 m is also important for the formation of the ERp57tapasin complex (9). This would also be consistent with the heterogeneity of the HC10-reactive dimers being a result of variable and incomplete HC oxidation. Further experiments will be necessary to test these possibilities.
Both TAP and tapasin appear necessary for the recovery of W6/32-reactive folded dimers. This could reflect a specific effect on peptide loading of dimeric HC and/or an overall increase in the expression of stable monomeric HC⅐peptide⅐␤ 2 m complexes on the cell surface that are then available to dimerize. The latter explanation seems unlikely to account for the tapasin effect, because the expression of W6/32-reactive HLA-B27 HC in .220.B27 cells without tapasin is quite high, yet no folded dimers are detected. Folded HLA-B27 dimers are readily detected in T5-1 and other B-LCL with a single copy of the HLA-B27 allele, indicating that overexpression is not required. The lack of B27.E45M dimerization even when overexpressed in C1R argues strongly that a specific characteristic of wild type HLA-B27 other than simply the presence of Cys 67 is necessary.
Our experiments demonstrating a requirement for TAP in T2 cells are in contrast with previous studies by Allen et al. (18). The reason for this discrepancy is not clear. One explanation could be that more cells were used for immunoprecipitation, or that HLA-B27 expression in their T2.B27 cells is greater. Another possibility is that the high molecular weight material previously reported as tetramer and dimer bands in immunoprecipitates from T2.B27 cells is actually the unreduced W6/32 antibody used for immunoprecipitation. Under nonreducing conditions the antibody migrates in this region of the gel and is reactive with the secondary antibody (goat anti-mouse IgG) used to detect the HC10 used for immunoblotting. 2 We have avoided this by immunoblotting with a rat antibody (3B10.7) and using a secondary antibody with minimal cross-reactivity against mouse IgG.
It is interesting that MARB4 recognizes HLA-B27 dimers, because this monoclonal antibody is known to recognize complexes with long peptides. The formation of stable complexes with long peptides may require a more relaxed conformation, which might designate a group of molecules that are more prone to dimerize. However, it should be noted that dimerization is not required for MARB4 recognition. This antibody recognizes cell surface complexes in C1R.B27.C67A and can immunoprecipitate monomers from these cells (data not shown). Rapid folding molecules such as B27.A2B, B27.A23, and B27.E45M are not recognized by MARB4, 3 however, this could also reflect a need for a B pocket with Arg specificity. Whether long peptides utilize the B pocket is unknown, although there is some support for this (65).
We have hypothesized that misfolding may contribute to disease pathogenesis via the generation of ER stress and subsequent deviation of the immune response (27,66). Others have proposed that recognition of homodimers by T cells (31) or KIR/ILT class I receptors (67) may be involved. The relative merits of these and other ideas remain to be determined. It is of interest that Boyle et al. (32) have isolated CD4 T cells from patients with ankylosing spondylitis that appear to recognize forms of HLA-B27 preferentially expressed on T2.B27 and .220.B27 cells, although present perhaps at lower levels on C1R.B27 (32). These responses could be blocked with W6/32 and ME.1, but not HC10, leading to the suggestion that the T cells might be recognizing dimers. Our results suggest that W6/32-reactive HLA-B27 homodimers are not responsible, as their expression is precisely opposite the observed recognition pattern. The expression of HC10-reactive dimers on 220.B27 and C1R.B27 correlates better, but not completely as they are much less abundant on T2.B27. It is also worth mentioning that transgenic rats expressing HLA-B27 with Ser substituted for Cys 67 (B27.C67S) develop the spondyloarthropathy-like phenotype (spontaneous inflammatory disease) with somewhat less arthritis than controls expressing wild type HLA-B27 (21), implying that dimers per se are not required for disease. However, it should be noted that these rats have a much higher ratio of B27.C67S to human ␤ 2 m transgene (3:1), whereas in the 21-3 rats the transgene ratio is 1.3:1. The transgene ratio in the B27.C67S rats is thus more than twice as high, and also considerably higher than other disease-prone HLA-B27 lines (0.83 and 1.7) (21). Therefore, the B27.C67S rats are likely to have greater expression of HC relative to human ␤ 2 m. Because rodent ␤ 2 m is less efficient at promoting HLA class I HC folding and expression (68), 2 this is likely to exacerbate HLA-B27 misfolding. To the extent that Cys 67 might normally contribute to the complete HLA-B27 misfolding phenotype, HC overexpression relative to ␤ 2 m in B27.C67S rats could compensate for the loss of Cys 67 by increasing HLA-B27 misfolding. It is therefore possible that although Cys 67 is not required in this situation, it might still contribute to disease pathogenesis under more physiological conditions.
The results presented here support the idea that HLA-B27 has aberrant immunobiological characteristics related to amino acid residues that form the B pocket, but are unlikely to be related to the role of this pocket in peptide selection per se. Our data suggest that the combination of delayed HC folding and ER retention, together with an unpaired Cys that is exposed when the molecule is unfolded, may create a complete misfolding phenotype which includes aberrant disulfide bonding in the ER. Indeed, there are few alleles that fit these two criteria, and only one that also contains a Lys residue at position 70 (HLA-B73) (17), which could be important as the local chemical environment can affect the reactivity of unpaired Cys residues (15).