Misfolding of HLA-B27 as a Result of Its B Pocket Suggests a Novel Mechanism for Its Role in Susceptibility to Spondyloarthropathies

HLA-B27 designates a group of MHC class I molecules or subtypes strongly associated with susceptibility to the spondyloarthropathies (1). Although evidence supports a direct role in pathogenesis (2, 3, 4, 5), the mechanism remains unknown. One hypothesis predicts that HLA-B27, as a result of its particular peptide binding specificity, presents self peptides mimicking pathogen-derived epitopes that then become the target of autoreactive CTL (6). However, no arthritogenic peptides have been identified, and extensive efforts to isolate HLA-B27-restricted autoreactive CD8⁺ T cells from affected individuals have, with few exceptions (7), been unsuccessful. Of interest, some studies imply an important role for CD4⁺ T cells in humans (8, 9) and transgenic rodents (4, 10). Consequently, the arthritogenic peptide hypothesis remains unproven, suggesting a need to consider alternative mechanisms (11).

Peptide binding differences between alleles result from extensive polymorphisms, particularly in amino acids within the peptide binding groove. One region, the B pocket, is especially significant for HLA-B27, as it is conserved among the subtypes unique to this group of alleles (12) and is probably responsible for the large overlap in their peptide repertoires (13, 14, 15, 16, 17, 18). Since many of the subtypes sufficiently prevalent to be assessed are associated with disease (19), the B pocket is thought to be critical for the arthritogenic phenotype (18). Considering the lack of evidence for arthritogenic peptides, we hypothesized that this pocket might confer other unique characteristics to HLA-B27. To address this question we compared peptide binding and assembly characteristics of HLA-B27 with B27.A2B, an HLA-B27 molecule substituted with an HLA-A2-like B pocket (20). Our results indicate that in addition to its role in peptide selection, the B pocket also causes heavy chains to misfold, which suggests a novel mechanism to explain the role of HLA-B27 in disease susceptibility.

B27.A2B was generated by site-directed mutagenesis of B*2705 (20) and has the following amino acid substitutions: H9F (F for H at position 9), T24A, E45M, I66K, C67V, and H70K. Minigenes encoding the HIV-1 gag p24 epitope (sequence KRWIIMGLNK), or position 2 (P2)³-substituted variants, were constructed using overlapping oligonucleotides and ligated into pCEP4 (Invitrogen, Carlsbad, CA) (21, 22). HMy2.C1R (C1R) are HLA-A negative and -B35 low (23), and 0.174 X CEM^R (T2) (24) are TAP deficient. C1R.B*2705, C1R.B27.A2B, T2.B*2705, and T2.B27.A2B were produced as described previously (20). Cells were grown in R-10 (RPMI 1640 supplemented with 10% FCS, 2 mM l-glutamine, and 10 U/ml penicillin/streptomycin) with geneticin (Life Technologies, Gaithersburg, MD) at 0.5 mg of activity/ml. T5-1 is a B cell line expressing HLA-A1, -A2, B8, -B27, and -Cw4 (25). C1R.B*2705 and C1R.B27.A2B were transfected with minigenes, selected with hygromycin (Calbiochem, La Jolla, CA; 1.5 mg/ml) (21, 22), and maintained in R-10 with hygromycin (0.5 mg/ml) and G418 (0.4 mg activity/ml).

The HPLC-purified peptides were obtained from the University of North Carolina MicroProtein Chemistry Facility (Chapel Hill, NC). Purity and expected molecular mass were confirmed by mass spectroscopy.

The CTL lines recognizing the influenza A nucleoprotein peptide (sequence SRYWAIRTR; HF CTL), and the HIV-1 gag epitope (sequence KRWIIMGLNK; 868 CTL) bound to B*2705 have been described previously (21, 22). Standard 4-h ⁵¹Cr release assays were performed using C1R transfectants pulsed with synthetic peptides or expressing minigenes as targets. Specific lysis was calculated using the formula (E − M/T − M) × 100, where E is experimental release, M is release in the presence of medium alone, and T is total release in the presence of 5% Triton X-100 (Sigma, St. Louis, MO).

Peptide binding was measured with a stabilization assay using TAP-deficient T2 cells as described previously (22, 26). Cell surface B*2705 or B27.A2B complexes were measured by staining with ME.1, which recognizes HLA-B7, B27, and Bw22 (27) and affinity-purified FITC-F(ab′)₂ goat anti-mouse IgG (Fc-specific; Organon Technika, West Chester, PA), followed by FACS analysis (FACScan, Becton Dickinson, Palo Alto, CA). The ability of each peptide to stabilize cell surface class I molecules is used as an indirect measure of peptide binding, and was determined as follows: % maximum change in fluorescence = [(MFI_{sample peptide} − MFI_diluent)/(MFI_SXYWAIRTR − MFI_diluent)] × 100.

SRYWAIRTR (for B*2705) or SQYWAIRTR (for B27.A2B; 100 μM) was used in each assay to establish maximum binding. The concentration of peptide required to achieve 50% maximum change in fluorescence is referred to as the EC₅₀ and is used for all comparisons.

Cells (10⁷/time point) were preincubated for 1 h at 37°C in medium lacking Met and Cys, then pulse-labeled with 2 mCi of [³⁵S]Met/Cys (Easytag, New England Nuclear, Boston, MA) and chased with a 10-fold excess of medium supplemented with 2 mM Met/Cys. Protein synthesis was stopped by the addition of 5 vol of ice-cold PBS and cells were harvested by centrifugation. Cells were washed, then lysed in 500 μl of lysis buffer (20 mM Tris (pH 7.8), 100 mM NaCl, 10 mM EDTA, and 1% Triton X-100) supplemented with protease inhibitors (5 mM iodoacetamide, 0.5 mM PMSF, and 0.1 mM N^α-p-tosyl-l-lysine chloromethyl ketone; Sigma). After 20 min on ice, nuclei were removed with a 10,000 × g spin for 5 min at 4°C, and supernatants were used for immunoprecipitations.

Membrane and soluble fractions were prepared as described by Hughes et al. (28). Briefly, pelleted cells (1.5 × 10⁷/condition) were washed and lysed by two cycles of freezing and thawing in dry ice/ethanol, and then lysates were suspended in 1 ml of TS (10 mM Tris (pH 7.4) and 150 mM NaCl) supplemented with protease inhibitors. Nuclei were removed by centrifugation, and postnuclear supernatants were further centrifuged at 100,000 × g for 1 h at 4°C to produce a pellet (membrane fraction) and supernatant (soluble fraction). One-tenth volume of 10% Triton X-100 in TS was added to supernatants, and pellets were dissolved in TS containing protease inhibitors and 1% Triton X-100 for 30 min on ice before immunoprecipitation.

Immunoprecipitations were performed after preclearing samples with washed formalin-fixed Staphylococcus aureus (Sigma), using purified mAbs at a final Ab concentration of 30 μg/ml for 1 h at 4°C. Protein A-Sepharose (Sigma) was then added (100 μl of 50 mg/ml suspension in lysis buffer) for 1 h at 4°C. Protein A-Sepharose pellets were washed and stored at −20°C until electrophoresis.

Isoelectric focusing (IEF) and SDS-10.5% PAGE were performed according to established methods (29). For phosphorimage analysis, dried gels were exposed to plates for 48 h, then ³⁵S-labeled proteins were visualized and quantitated with ImageQuant software (Molecular Dynamics, Sunnyvale, CA).

Proteins were transferred onto a polyvinylidene difluoride membrane (Westran, Schleicher & Schuell, Keene, NH) using methods adapted for IEF gels (29). Following transfer, the polyvinylidene difluoride membrane was blocked for 1 h with 1% blocking powder in PBS (Schleicher & Schuell), then HC10 (1 μg/ml) was added for an additional 30 min at room temperature. After washing, blots were incubated with a 1/1000 dilution of alkaline phosphatase-conjugated goat anti-mouse IgG heavy and light chain (Southern Biotechnology Associates, Birmingham, AL) for 30 min, then protein bands were visualized using 5-bromo-4-chloro-3-indolyl phosphate/nitro blue tetrazolium for color development.

Previously we showed that B27.A2B, which is identical with HLA-B*2705 except for six amino acid substitutions in the B pocket (see Materials and Methods), can be recognized by HLA-B27-restricted flu-specific CTL, provided the flu peptide has Leu instead of Arg at P2 (20). To determine the characteristics of peptides naturally presented by B27.A2B, it was immunopurified from C1R transfectants, and peptides were eluted and sequenced. Gln, Met, and Leu were found to predominate at P2 (data not shown) instead of Arg seen in the majority of peptides presented by various HLA-B27 subtypes (13, 14, 15, 16, 17). This P2 specificity for B27.A2B is similar to HLA-A2 (A*0201 subtype), but not identical, since Gln has not been reported (30). HLA-A*0205, a subtype with Tyr instead of Phe at position 9 in its B pocket, has a similar Leu/Met/Gln P2 preference (31), suggesting that a relatively conservative change in this pocket can produce an environment suitable for Gln. Recognition of B27.A2B presenting two different HLA-B27-restricted epitopes with Leu, Met, or Gln at P2 is shown in Fig. 1. In addition to exogenous peptide loading (Fig. 1, A and B), B27.A2B loads an endogenously synthesized HIV gag peptide with Gln, but not Ala, Arg, or Lys at P2 (Fig. 1 C). This indicates that the antigenic surface of B27.A2B is similar to HLA-B*2705 and suggests that the B pocket substitution does not have a major effect on amino acid preference at peptide positions outside the B pocket.

To compare peptide binding characteristics of HLA-B*2705 with two different B pockets, known ligands containing either an HLA-B*2705 or B27.A2B-specific anchor residue at P2 were tested using a cell surface class I stabilization assay (22). Peptides with Gln, Met, or Leu at P2 did not bind appreciably to HLA-B*2705, and Arg-P2 peptides only showed limited binding to B27.A2B at the highest concentrations (50–100 μM; data not shown), indicating that binding was specific. The most striking finding was that in each case equivalent binding to HLA-B*2705 required much higher peptide concentrations than B27.A2B (Fig. 2), with EC₅₀ ratios ranging from approximately 30 for NRH/NQH to 60 for GRI/GQI. Several other cognate peptides have been compared, and in each case binding to HLA-B*2705 was less efficient (data not shown). It is important to note that EC₅₀ measured by this assay does not represent an affinity constant, and as we have shown previously, this measure better represents peptide loading efficiency than complex stability (26). Thus, these results suggest HLA-B*2705 is relatively inefficient at loading peptide as a result of its B pocket. This may reflect poor formation of peptide-receptive HLA-B*2705:β₂m complexes, and/or a greater dependence on peptide to undergo the conformational change required for Ab recognition measured as binding in the cell surface stabilization assay.

To assess formation of peptide-receptive complexes in vivo, pulse-chase experiments were performed. C1R transfectants were labeled for 5 min with [³⁵S]Met/Cys, and chased in the presence of excess nonradioactive Met/Cys. Class I molecules were immunoprecipitated from cell lysates first using W6/32, which recognizes folded heavy chain-β₂m complexes (32), then HC10, which recognizes unfolded free (non-β₂m-associated) heavy chains (33). Immunoprecipitated heavy chains were then resolved on IEF gels (Fig. 3, A and B). The 0 form contains unsialylated N-linked glucosyl residues, indicating they remain in the ER, while forms 1 and 2 contain increasing amounts of sialic acid, reflecting ER to Golgi transport (33) and thus focus to more acidic positions. There is a striking difference in the rate of conversion of unfolded HC10-reactive heavy chain into folded W6/32-reactive material (Fig. 3, C and D), with HLA-B*2705 requiring ∼30 min for 50% conversion, while for B27.A2B this occurs within the first 5 min of the pulse. Inefficient folding is not due to overexpression, since only slightly higher levels of HLA-B27 mRNA (per microgram of 18S RNA) were found on Northern blots (data not shown), and if anything, there appears to be greater synthesis of B27.A2B on a per cell basis. We also considered the possibility that HC10 might not react as well with B27.A2B, and thus free heavy chains might remain unprecipitated. To test this, a third immunoprecipitation was performed (after W6/32 and HC10) using 5H7, which recognizes a determinant in the α3 domain of the heavy chain (34) and is not dependent on the conformation of the α1 and α2 domains. Only trace amounts of B27.A2B could be recovered (data not shown), confirming that the majority of this molecule folds and becomes recognizable by W6/32 shortly after synthesis.

Typically, heavy chains fold and associate with β₂m shortly after synthesis and then load peptides that have been transported into the ER from the cytosol before being released to traffic to the cell surface (35). However, when synthesized in mutant cell lines deficient in β₂m or peptide transporters (TAP) or even when TAP function is inhibited, heavy chains can misfold and are dislocated into the cytosol and degraded by proteasomes (28). Due to the rapidity of this process, detection of cytosolic heavy chains requires proteasome inhibition. In light of the inefficient folding of HLA-B*2705, we were interested in whether some of the heavy chains might misfold and enter the dislocation pathway. To test this, cells were incubated with the proteasome inhibitor N-acetyl-l-leucyl-l-leucyl-l-norleucinal (LLnL) and lysed in the absence of detergents by freezing and thawing, and soluble (cytosolic) and membrane fractions were prepared (28). Heavy chains were then immunoprecipitated from these fractions using HC10, separated on IEF gels, and visualized by immunoblotting. HLA-B*2705 heavy chains can be found accumulating in the cytosolic fraction in the presence of LLnL, suggesting that they are normally rapidly degraded (Fig. 4). This is even more readily apparent in a B cell line (T5-1) expressing HLA-B*2705 (as part of its full complement of class I molecules) than in C1R transfectants. In contrast, no accumulation of B27.A2B can be detected. The predominant form of cytosolic heavy chain is more acidic than the ER (0) form (Fig. 4), consistent with conversion of Asn to Asp during N-glycanase-mediated carbohydrate removal during heavy chain dislocation (36). Removal of carbohydrate is also expected to reduce the m.w. of the heavy chain. To confirm this, the experiment was repeated, and the size of cytosolic heavy chains was found to be ∼3 kDa less (Fig. 5), consistent with loss of carbohydrate seen following in vitro digestion with N-glycanase (28) or endoglycosidase H (data not shown). Since LLnL may inhibit other proteases (37), including those that may affect peptide trimming in the ER (38), we have performed the same experiment with lactacystin, a more specific proteasome inhibitor (39) and obtained similar results (data not shown). Thus, it is unlikely that HLA-B*2705 heavy chain misfolding is due to a nonspecific action of LLnL.

To determine whether newly synthesized molecules are misfolding, T5-1 cells were pulse-chased with [³⁵S]Met/Cys in the absence or the presence of LLnL. A soluble 40-kDa heavy chain band accumulates and is most notable at 1–2 h of chase (Fig. 6). A portion of the samples run on an IEF gel confirmed the identity of the heavy chain band as HLA-B*2705 (data not shown). These results indicate that HLA-B*2705 heavy chains are dislocated into the cytosol within 1–2 h of synthesis, as evidenced by their accumulation when proteasomes are inhibited.

Formation of stable heavy chain:β₂m:peptide complexes in the ER is necessary before trafficking to the cell surface (35). Following synthesis and glycosylation, free heavy chains are initially stabilized by chaperones such as calnexin (40, 41) until a conformation suitable to bind β₂m and peptide is achieved. In the absence of β₂m, heavy chains can be retained by calnexin (42), but ultimately misfold and are degraded (28). They have a similar fate in the absence of peptide, either in TAP-deficient cells or when TAP is inhibited (28). Our studies demonstrate that α1-domain amino acids in HLA-B27 that ultimately form the B pocket have a dramatic effect on folding efficiency and cause misfolding, even in the presence of an intact Ag presentation pathway. Although allelic differences in rates of folding and assembly have been noted previously (33, 43), and in one case attributed to α2-domain polymorphisms (43), misfolding in the presence of a normal assembly pathway has not been reported. Furthermore, we have not detected misfolding of HLA-B7, -B53, or even -B8, despite its relatively slow folding (data not shown). Thus, although we cannot rule out misfolding of other class I molecules, to date this phenotype appears to be limited to HLA-B27 and is related to its unique B pocket.

The B pocket could affect the formation of heavy chain:β₂m:peptide complexes by directly influencing heavy chain folding or by altering its affinity for β₂m or peptide. HLA-A2 assembles normally in calnexin-deficient cells (44), and when calnexin binding is blocked with glycosylation inhibitors, HLA-B27 is less stable than HLA-A2 (33). Although this could indicate that alleles such as HLA-A2 are inherently better at folding and do not require calnexin, the involvement of other chaperones has not been ruled out. Since β₂m and peptide binding to heavy chains can facilitate release from calnexin (40, 45), increased affinity and/or availability of these components could enhance folding efficiency. In this regard, since the B pocket in HLA-B27 is highly selective for Arg, while in HLA-A2 it accepts Met and Leu (and Gln for B27.A2B), HLA-A2 and B27.A2B may be more promiscuous in peptide interactions. Furthermore, our peptide binding data suggest that even for individual peptides, HLA-B27 stabilization requires higher concentrations. Since HLA-B27 heavy chains tend to misfold even in the presence of TAP, we considered the possibility that this result might be due to a large population of misfolded HLA-B27 heavy chains on the surface of T2 cells. However, while HC10-reactive material is abundant, there is actually less on T2.B*2705 than on T2.B27.A2B (data not shown), and therefore, this does not explain the peptide binding differences we observe. Thus, while the molecular mechanism by which the B pocket affects folding efficiency remains to be determined, it seems likely that peptide interaction is important.

The strong association between several HLA-B27 subtypes containing the same B pocket and disease susceptibility (19) and the lack of evidence implicating arthritogenic peptides suggest that alternative hypotheses incorporating misfolding should be considered. Typically, misfolded ER proteins are degraded as part of a quality control process (46). When a large percentage of the pool misfolds, protein deficiency diseases may occur (47). However, misfolded proteins can also accumulate in the ER and trigger a stress response (48) or escape the quality control process and traffic to other cellular compartments (49) or to the cell surface (50). HLA-B27 expression in mice has been reported to cause arthritis, but only in the absence of mouse β₂m, a condition that eliminates cell surface HLA-B27 complexes but enhances expression of unfolded (HC10-reactive), or perhaps misfolded, free heavy chain (4). Although implicated in pathogenesis (4, 51), the precise structure or role of free heavy chains is not known. Recently, it has been shown that HLA-B27 heavy chains can form homodimers when synthesized in cells deficient in TAP (52). Whether this also occurs in β₂m-deficient mice or perhaps at low frequency in the presence of peptide is not known, but, interestingly, it appears to be dependent on Cys⁶⁷, a B pocket amino acid (52). In HLA-B27 transgenic rats, β₂m deficiency is not required for disease; however, there is a dependence on high copy number and high levels of HLA-B27 expression (3, 53). Although it has been difficult to reconcile results from rats and mice mechanistically, both could be explained by a requirement for misfolded HLA-B27 molecules. It should be noted that coexpression of an HLA-B27-binding peptide in the ER of transgenic rats reduces the incidence of arthritis (54). These findings were interpreted as support for a class I-restricted mechanism, since the spectrum of peptides presented by HLA-B27 was affected. However, an alternative explanation is that overexpression of an HLA-B27-binding peptide may reduce misfolding. In summary, a plausible hypothesis is that as a result of its unique B pocket, HLA-B27 has a tendency to misfold even when all components of the Ag presentation pathway are present, as would be expected in most individuals with HLA-B27. This may result in the formation of neoantigens such as the homodimers found in TAP-deficient cells (52). In the animal models misfolding is exacerbated: in mice due to the absence of endogenous β₂m, and in rats by overexpression. This would lead to the enhanced production of misfolded forms that, in turn may be responsible for the development of high frequency spontaneous arthritis.

Another consequence of misfolding with potential relevance to autoimmune disease is the inadvertent entry of HLA-B27 heavy chains into the cytosol. This can result in cytosolic degradation by proteasomes, which may lead to enhanced presentation of heavy chain-derived peptides by class I. Indeed, a peptide matching part of the α2 domain of HLA-B27 has been eluted from HLA-B27 expressed in C1R (15), and it is likely that different HLA-B27 breakdown products could be presented by alleles with other specificity. If the degree of misfolding varies among different cell types or is affected by physiological changes such as infection with intracellular bacteria known to be triggers of spondyloarthropathies, then tolerance to HLA-B27-derived self-peptides may be lost. Since attempts to find autoreactive CTL have largely focused on recognition of HLA-B27 (7), other class I molecules presenting bits of HLA-B27 could have been missed. Alternatively, misfolded cytosolic heavy chains could interact with Hsc73 and enter the endosomal-lysosomal pathway (55) in a manner different from their usual recycling from the cell surface. It has been argued that increased heavy chain turnover and presentation of HLA-B27 breakdown products by class II may be important in disease (11), in part to account for the apparent role of CD4⁺ T cells in pathogenesis (4, 10). In support of this, one study has shown that T cells from patients have greater reactivity against HLA-B27-derived peptides than T cells from healthy controls (9). However, development of arthritis in HLA-B27 transgenic, β₂m-deficient mice also lacking class II molecules argues against this (56).

In summary, our studies were undertaken to determine whether B pocket amino acids confer more than just peptide binding specificity to HLA-B27, in particular characteristics that might ultimately underlie the ability of this allele to cause disease. As we have demonstrated, this pocket affects folding efficiency and confers a misfolding phenotype on HLA-B27. Although the mechanism of HLA-B27-associated disease remains to be determined, these observations emphasize the need to consider misfolding and its sequelae in pathogenesis, rather than a unique ability to select and present arthritogenic peptides.

We thank J. A. Frelinger, H. Ploegh, P. Cresswell, and E. Mellins for their gifts of reagents or cells, and D. N. Glass for critically reading the manuscript.