Allelic association with ankylosing spondylitis fails to correlate with human leukocyte antigen B27 homodimer formation

Expression of human leukocyte antigen (HLA)-B27 is strongly associated with predisposition toward ankylosing spondylitis (AS) and other spondyloarthropathies. However, the exact involvement of HLA-B27 in disease initiation and progression remains unclear. The homodimer theory, which proposes that HLA-B27 heavy chains aberrantly form homodimers, is a central hypothesis that attempts to explain the role of HLA-B27 in disease pathogenesis. Here, we examined the ability of the eight most prevalent HLA-B27 allotypes (HLA-B*27:02 to HLA-B*27:09) to form homodimers. We observed that HLA-B*27:03, a disease-associated HLA-B27 subtype, showed a significantly reduced ability to form homodimers compared with all other allotypes, including the non-disease-associated/protective allotypes HLA-B*27:06 and HLA-B*27:09. We used X-ray crystallography and site-directed mutagenesis to unravel the molecular and structural mechanisms in HLA-B*27:03 that are responsible for its compromised ability to form homodimers. We show that polymorphism at position 59, which differentiates HLA-B*27:03 from all other allotypes, is responsible for its compromised ability to form homodimers. Indeed, histidine 59 in HLA-B*27:03 leads to a series of local conformational changes that act in concert to reduce the accessibility of the nearby cysteine 67, an essential amino acid residue for the formation of HLA-B27 homodimers. Considered together, the ability of both protective and disease-associated HLA-B27 allotypes to form homodimers and the failure of HLA-B*27:03 to form homodimers challenge the role of HLA-B27 homodimers in AS pathoetiology. Rather, this work implicates other features, such as peptide binding and antigen presentation, as pivotal mechanisms for disease pathogenesis.

explain its apparent role in AS (11). One of the central hypotheses is the "HLA-B27 homodimer theory," which proposes that the formation and subsequent cell-surface presentation of HLA-B27 heavy-chain homodimers triggers an aberrant immune response leading to autoimmunity. In support of this theory, HLA-B27 molecules exhibit an atypically high propensity to form homodimers, and stably expressed HLA-B27 homodimers have been observed both in vitro and in vivo (12)(13)(14)(15)(16). In addition, HLA-B27 homodimers have been shown to interact with immunoreceptors, such as killer-cell immunoglobulin-like receptors and leukocyte immunoglobulin-like receptors, although it is unknown whether these interactions are responsible for the onset of autoimmunity (17)(18)(19)(20). At a molecular level, HLA-B27 homodimers are formed by two HLA-B27 heavy chains, which are covalently linked through a disulfide bridge involving a conserved, unpaired cysteine residue (Cys-67) (12,13,21), although other cysteine residues may be involved as well (15,21). ␤ 2 -Microglobulin (␤2m), the auxiliary subunit of all canonical trimeric HLA class I complexes, is almost certainly not retained upon homodimer formation (12,13). Moreover, the addition of oxidizing or apoptosis-inducing agents has been shown to increase the levels of HLA-B*27:05 homodimers on both the EBV-transformed Jesthom B-cell and the leukemic CEM T-cell lines, suggesting that alterations in the cellular redox environment can induce the formation of HLA-B27 homodimers (15).
Here, we examined the ability of the eight most prevalent HLA-B27 allotypes (HLA-B*27:02 to HLA-B*27:09) to form homodimers, which included the non-disease-associated allotypes HLA-B*27:06 and HLA-B*27:09. We showed that the disease-associated allotype HLA-B*27:03 (31)(32)(33)(34) shows a very weak propensity to form homodimers compared with other HLA-B27 allomorphs, independently of their disease association (protection or progression). Using X-ray crystallography in combination with site-directed mutagenesis, we provide the molecular bases for the compromised ability of HLA-B*27:03 homodimer formation. The observation that a disease-associated HLA-B27 subtype is barely able to form homodimers may suggest that HLA-B27 homodimers are not as important in disease initiation and progression as previously anticipated.

HLA-B*27:03 shows a reduced propensity to form homodimers
The addition of hydrogen peroxide (H 2 O 2 ) to the human Epstein-Barr virus-transformed, HLA-B*27:05-positive Jesthom B-cell and the leukemic CEM T-cell lines expressing HLA-B*27:05 can increase the formation of HLA-B*27:05 homodimers in cellulo (15). To confirm and extend this observation to other HLA-B27 allotypes in a different cell type, we incubated human B lymphocyte C1R cells expressing matched levels of the eight most common allotypes (HLA-B*27:02 to :09 (35); Fig. S1A) with 250 M H 2 O 2 for 8 h and analyzed the formation of HLA-B27 homodimers by Western blotting using the HC10 antibody (36,37), which recognizes HLA class I heavy chains ( Fig. 1 (A and B) and Fig.  S2). Of note, any HC-10 reactivity from those C1R transfectants can be solely attributed to HLA-B27, as WT C1R cells do not form homodimers because they only express very low levels of endogenous HLA-B*35:03 and HLA-C*04:01 (38 -41). As expected, a significant increase in homodimer formation was observed for the HLA-B*27:05 allotype expressed on C1R cells upon treatment with H 2 O 2 , indicating that HLA-B*27:05 is able to robustly form homodimers in a cell line-independent manner. In addition, following exposure to H 2 O 2 , we also observed significantly elevated levels of homodimers for the disease-associated subtypes HLA-B*27: 02, HLA-B*27:04, HLA-B*27:07, and HLA-B*27:08 as well as for the non-disease-associated subtypes HLA-B*27:06 and HLA-B*27:09, with no major differences between these various allotypes (Fig. S2). In agreement with previous studies (12,13,21), the addition of the reducing reagent DTT resulted in a complete loss of the homodimers, confirming that HLA-B27 homodimers are covalently linked via disulfide bridges (data not shown). Moreover, we demonstrated that the unpaired Cys-67 plays a major role in homodimer formation, as a C67S point mutation in HLA-B*27:05 abolishes homodimer formation (Fig. S3).
However, the disease-associated subtype HLA-B*27:03 showed severely reduced levels of homodimer formation after the addition of H 2 O 2 ( Fig. 1 (A and B) and Fig. S2). This discrepancy was evident at steady-state conditions, under which HLA-B*27:03 homodimers were barely visible compared with other allotypes, suggesting that HLA-B*27:03 exhibits a severely reduced propensity to form homodimers, which is an intriguing observation, given that HLA-B*27:03 is associated with AS (6, 31-34, 42, 43).

Polymorphism at position 59 is responsible for the lack of HLA-B*27:03 homodimers
To further investigate this finding, we compared the amino acid sequence of HLA-B*27:03 against the other HLA-B27 allotypes (Table 1), which revealed that HLA-B*27:03 differs from other HLA-B27 allotypes at position 59, whereby the conserved tyrosine residue is replaced by a histidine residue (Y59H) ( Table 1). Of note, Y59H is the only polymorphism that differentiates HLA-B*27:03 from the archetypal HLA-B*27:05 allomorph, suggesting that this single-amino acid polymorphism is sufficient to impact the for-mation of HLA-B27 homodimers. Nevertheless, to further test this notion, we introduced the Y59H polymorphism into the HLA-B*27:07 and HLA-B*27:08 allotypes (Fig. S4A) and expressed matching levels of these mutant versions in C1R cells (C1R-HLA-B*27:07-Y59H and C1R-HLA-B*27:08-Y59H; Fig. S1B). We have chosen these particular allomorphs, as they differ from HLA-B*27:05 by several polymorphisms (Table 1) and are more diverged from this archetypal allotype than the other allomorphs. In agreement with our postulate, the presence of this single point mutation in HLA-B*27:07 and HLA-B*27:08 severely reduced the formation of homodimers under both steady-state and H 2 O 2inducing conditions (Fig. 1, C-F). These results demonstrate that the presence of a histidine residue at position 59 is sufficient to impact the ability of HLA-B27 allotypes to form homodimers in a dominant-negative fashion.

Short-range conformational rearrangements govern the accessibility of Cys-67
Due to the three-dimensional proximity of the Y59H polymorphism to Cys-67, which has been shown to be the key cysteine residue involved in the formation of homodimers through disulfide bonds (12,13), we hypothesized that the presence of His-59 causes conformational changes that may impact the environment of Cys-67. To unravel the effects of this polymorphic position on HLA-B*27:03 and HLA-B*27:05, we analyzed the structures of these HLA molecules in complex with two different peptide ligands. We selected the endogenous self-peptide LRNQSVFNF (LRN), which has been identified in our comprehensive HLA-B27 peptide database as a ligand for both HLA-B*27:03 and HLA-B*27:05 allotypes (35), as well as the well-characterized KK10 peptide (KRWIILGLNK) derived from HIV. The structure of HLA-B*27:05 in complex with  Challenging the HLA-B27 homodimer theory KK10 was published previously (44,45), and we determined the structures of the remaining three complexes (of HLA-B*27:03 in complex with both LRN and KK10 peptide and of HLA-B*27:05 in complex with the LRN peptide). All structures were solved at high resolution and in the same space group, independently of the bound peptide or the allomorph crystallized ( Table 2). The HLA-B*27:05 KK10 and HLA-B*27:03 KK10 structures overlaid well, with a root mean square deviation (r.m.s.d.) of 0.19 Å on the antigen-binding cleft and of 0.42 Å for the peptide itself ( Fig. 2A). As expected, the KK10 peptide was observed to bind in a similar extended conformation in both molecules with the characteristic anchor residues P2-Arg and P⍀-Lys binding the B and F pockets, respectively. The central part of the KK10 peptide, P6-Leu and P7-Gly, was poorly defined in the electron density map, indicating flexibility in both complexes. The structures of the HLA-B*27:03 and HLA-B*27:05 in complex with the self-derived LRN peptide also showed a similar conformation of the HLA antigen binding cleft (r.m.s.d. of 0.14 Å) and of the peptide (r.m.s.d. of 0.06 Å) (Fig. 2B). However, in contrast to the KK10 peptide, the electron density for 9-mer LRN peptide was well-defined with all residues clearly observed, indicating a more rigid structure than the KK10 peptide (Fig. S5). Similar thermal stability was observed for four peptide-HLA (pHLA) complexes (Table 3) with a T m ranging from 60 to 63°C, indicating that the polymorphism at position 59 has no impact on the stability of these pHLAs.
The polymorphic residue at position 59 is located at the beginning of the HLA ␣1-helix in close proximity with the P1 residue of the peptide. This residue is a Tyr in HLA-B*27:05 and a His in the HLA-B*27:03 molecule. The side chains of both Tyr-59 and His-59 are pointing down in the antigen-binding cleft. In HLA-B*27:05, the Tyr-59 hydroxyl group is interacting directly or via a water-mediated bond with the Tyr-171 and Tyr-7, respectively, as well as interacting with the hydrophobic residues Ile-52, Phe-33, and Val-34. The aromatic ring of Tyr-59 sits above the Trp-60 and is sandwiched by two negatively charged residues, Glu-55 and Glu-63 (Fig. S6A). In HLA-B*27:03, the His-59 has an environment similar to that of Tyr-59, with some localized movement. The lack of a hydroxyl group allows for the presence of an additional water molecule bonding His-59 and Tyr-7 in HLA-B*27:03 structures (Fig.  S6B). The smaller side chain of the His-59, compared with Tyr-59, allows the Trp-60 side chain to be closer to residue 59 in HLA-B*27:03 (Fig. 2, C and D). The repositioning of the Trp-60 side chain was more obvious in the KK10-bound structures than in the LRN-bound ones, with a maximum displacement of 1.8 and 0.6 Å, respectively. Interestingly, the Trp-60 sits above the Pro-47, which is part of the HLA molecule's third loop (Fig.  S7). This loop is located beneath the Cys-67 residue and spans from Ser-38 (at the end of the third ␤-strand) to Pro-50 (at the start of the ␣1-helix; Fig. 2, E-G). The Cys-67 residue is located within the ␣1-helix and shielded by Phe-36, which is located upstream of the third loop. The third loop itself is stabilized by hydrophobic side chain stacking of Pro-47, Trp-60, and the polymorphic residue Tyr/His-59.
Whereas the overall structures of the HLA-B*27:03 and HLA-B*27:05 molecules were similar irrespective of the bound peptide, we observed major differences in the flexibility of the third loop of the antigen-binding cleft based on the B factor analysis, which is the "temperature factor" representing the dynamic mobility of each atom. Namely, we compared the B factor of each residue in the loop located between Phe-36 and Pro-47 with the overall B factor of the antigen-binding cleft (residues 1-180 (46) Based on these structural rearrangements observed for Tyr/ His-59, Trp-60, and Pro-47 in combination with the apparent differences in the B factor of the third loops, we concluded that Challenging the HLA-B27 homodimer theory the lower flexibility of the third loop in HLA-B*27:03 may limit the accessibility of the buried Cys-67 residue compared with HLA-B*27:05 molecule, which reduces the ability of HLA-B*27:03 to form homodimers.

Proximal loop mobility dictates Cys-67 accessibility and homodimer formation
To test whether the position and mobility of the proximal third loop plays an important role in the formation of homodimers, we introduced specific point mutations into the HLA-B*27:03 and HLA-B*27:05 heavy chains and analyzed their ability to form homodimers using Western blotting with the HC10 antibody (36,37). We first mutated the Pro-47 in HLA-B*27:03 to a glycine residue (P47G). We hypothesized that the P47G mutation would increase the flexibility of the third loop, which should increase the formation of homodimers in HLA-B*27:03. Indeed, HLA-B*27:03-P47G showed a significant increase in homodimer levels compared with WT HLA-B*27:03 (Fig. 3). Second, we mutated the bulky Trp-60 to a small alanine residue (W60A) in HLA-B*27:05. Due to the interaction interplay between His-Tyr-59/Trp-60/Pro-47, the hypothesis was that the lack of side chain at position 60 would help the third loop of HLA-B*27:05 to be more rigid and adopt an "HLA-B*27:03-like" position resulting in reduced levels of homodimers. Consistent with this hypothesis, we were unable to observe any homodimers in HLA-B*27:05-W60A-expressing cells in contrast to WT HLA-B*27:05 (Fig. 3) despite comparable cell-surface expression levels (Fig. 1C). However, by mutating the tryptophan residue to alanine, we destroyed one of the key residues of the HC10 epitope ( 57 PXXWDR 62 (36,37)), which results in suboptimal antibody binding and explains the very faint signals for HLA-B*27:05-W60A. Nonetheless, these observations provide direct evidence for our hypothesis that naturally occurring and intentional substitutions in this loop dictate Cys-67 accessibility and heavy-chain homodimer formation.
To gain molecular insight into the impact of these loop mutations, we solved the structures of the HLA-B*27:03-P47G LRN and HLA-B*27:05-W60A LRN complex and compared them with their WT counterparts ( Table 2). The overall conformation of HLA-B*27:03 was not disturbed by the P47G mutation, with an r.m.s.d. of 0.10 Å on the overall pHLA complex. In addition, the P47G mutation did not change the conformation of the third loop but had a minor effect on the Trp-60 side chain, which was slightly shifted (0.5 Å of the C2 atom) (     (Fig. 3). Our data also show that the Trp-60, which is directly contacted by the polymorphic residue 59, has a greater impact on the third loop mobility than observed for Pro-47, which sits underneath Trp-60.
To validate biochemically that the Cys-67 residue in HLA-B*27:03 is less accessible compared with HLA-B*27:05, we performed Ellman's assays (47) with HLA-B*27:03 and HLA-B*27:05 by adding 5,5Ј-dithiobis-2-nitrobenzoic acid (DTNB; Ellman's reagent) to the KK10-refolded HLA molecules. In brief, DTNB reacts with free thiol groups and stoichiometrically produces a 2-nitro-5-thiobenzoate dianion (TNB 2Ϫ ) of yellow color that can be colorimetrically monitored over time. Given that HLA-B*27:03 and HLA-B*27:05 are identical apart from a single polymorphism at position 59 (Table 1), and the only free thiol group in both molecules is the cysteine residue at position 67, any difference in the reaction rate with DTNB can be directly attributed to variations in the accessibility of the Cys-67 residue. In agreement with our hypothesis, significant differences were observed in the absorbance at 412 nm between HLA-B*27:03 and HLA-B*27:05 (Fig. 5), suggesting that the reduced flexibility of the third loop in HLA-B*27:03 causes the bulky DTNB compound to react slower with its free Cys-67 residue compared with HLA-B*27:05.
Taken together, we demonstrate that the flexibility and conformation of the third loop of the HLA-B27 molecule has a major impact on its ability to form homodimers. The Tyr-59 in HLA-B*27:05-and presumably in all other HLA-B27 allotypes carrying the conserved Tyr-59 at this position-impacts the third loop conformation through its interaction with Trp-60 and Pro-47. These changes impact the mobility of the third loop, which in turn might modulate the accessibility of Cys-67 and homodimer formation. The presence of the smaller His-59 residue in HLA-B*27:03 allows the third loop to adopt a more rigid conformation, which reduces homodimer formation probably by shielding the underlying Cys-67 residue.

Discussion
In this study, we analyzed the ability of the eight most prevalent HLA-B27 allotypes (HLA-B*27:02 to HLA-B*27:09) to form homodimers under oxidizing conditions. With the exception of HLA-B*27:04, HLA-B*27:05, and HLA-B*27:09, none of these HLA-B27 allotypes have been tested previously for their ability to form homodimers, and, to our knowledge, this is the first report that compared homodimer formation of multiple HLA-B27 allotypes in the one parental cell line. In line with previous results (15,18), we observed that HLA-B*27:05 and HLA-B*27:09 are able to form homodimers that can be effectively induced by the addition of H 2 O 2 , and Cys-67 appears to be essential for the formation of such redox-induced dimers (Fig. S3), although Cys-325 has been reported to be involved as well (15). Importantly, we did not observe any major differences in the level of homodimer formation between this disease-associated and the non-disease-associated HLA-B27 allotypes. This observation is in contrast to a previous report (18), which shows that the arthritis-associated HLA-B*27:05 allele forms more cell-surface homodimers than the nonassociated HLA-B*27:09 allele. HLA-B27 expression levels, cell lines, and/or sensitivity of reporter assays might account for these discrepancies, in addition to conflicting behaviors between cell surface-localized and intracellular HLA-B27 homodimers.
All other tested HLA-B27 allotypes-apart from HLA-B*27: 03-were observed to form inducible homodimers (Fig. S2), which are sensitive to the addition of the reducing agent DTT, confirming that they are covalently linked by disulfide bonds. Surprisingly, we observed that HLA-B*27:03, an arthritis-associated HLA-B27 subtype that has been found to be present in many AS patients (6,(31)(32)(33)(34), is barely able to form homodimers and shows significantly reduced homodimer levels under both steady-state and oxidizing conditions (Fig. 1 (A and B) and  Challenging the HLA-B27 homodimer theory S2). These apparent differences are clearly evident compared with other disease-associated subtypes as well as with both non-disease-associated subtypes HLA-B*27:06 and HLA-B*27: 09, and it is difficult to reconcile our observations with the "HLA-B27 homodimer theory." A comparison of the HLA-B*27:03 amino acid sequence against the other HLA-B27 alleles suggested that the polymorphism at position 59 (Y59H) is accountable for its compromised ability to form homodimers. This notion was further confirmed by the observation (i) that Y59H is the only polymorphism differentiating HLA-B*27:03 and HLA-B*27:05 and (ii) that the HLA-B*27:07 and HLA-B*27:08 allotypes, which are able to readily form homodimers, lose this ability after the substitution of the Tyr-59 by His-59. These findings prompted us to use X-ray crystallography to understand the consequences of this polymorphism at the molecular level. As expected-given that HLA-B*27:03 and HLA-B*27:05 differ by only one polymorphism-the antigenbinding cleft conformations are highly similar, but some notable local conformational differences were observed surrounding the polymorphic residue 59. The bulky Tyr-59 in HLA-B*27:05 exerts an effect on the position of the third loop of the antigen-binding cleft (residues 38 -50) via hydrophobic interactions with Trp-60 and Pro-47. In contrast, the presence of the smaller His-59 in HLA-B*27:03 allows the loop to sit closer to the ␣1-helix. Although these conformational changes were modest, the resulting mobility changes of the third loops were significant. In HLA-B*27:03, the third loop was found to be relatively rigid, which in turn reduces the accessibility of the buried Cys-67 residue and leads to decreased levels of homodimer. In contrast, the third loop in HLA-B*27:05 was  Challenging the HLA-B27 homodimer theory found to be relatively flexible, such that the underlying Cys-67 residue becomes more accessible to form homodimers. Somewhat reminiscent changes in mobility have been described for a hinged unit containing the conserved 3 10 helix in MHC class I molecules that shows differential mobility, depending on whether a peptide is bound or not (48). Although highly speculative, the third loop in HLA-B27 might also reflect a peptidedependent region that may be stabilized or destabilized in the context of appropriate peptides. In addition, the protonation level of the His-59 residue might be different under physiological conditions than in the crystals at pH 5.6, but those changes in charge states will have presumably only marginal impact on the structure.
To confirm our observations, we performed biochemical assays (Ellman's assays) with HLA-B*27:03 and HLA-B*27:05, which showed that the Cys-67 residue in HLA-B*27:03 is less accessible compared with HLA-B*27:05. In addition, we introduced specific, site-directed point mutations (W60A and P47G) into the interaction cascade linking the polymorphic residue at position 59. We succeeded in changing homodimer levels by interfering with the cascade, which directly altered the flexibility of the third loop. When analyzing HLA-B*27:05-W60A by Western blotting using the HC10 antibody, we observed very faint signals even for the monomeric bands (Fig.  3), which was not due to decreased expression levels of HLA-B*27:05-W60A, as shown by flow cytometry with the HLA-B27-specific antibody ME1 (Fig. S1C). However, by mutating the tryptophan residue to alanine, we destroyed one of the key residues of the HC10 epitope ( 57 PXXWDR 62 (36,37)), resulting in suboptimal antibody binding.
Taken together, we provide compelling evidence that HLA-B*27:03-a disease-associated subtype-is barely able to form homodimers. These observations are difficult to reconcile with the "HLA-B27 homodimer theory" and strongly suggest that HLA-B27 homodimers are not as important in disease initiation and progression as previously anticipated. Although we cannot entirely rule out the involvement of HLA-B27 homodimers in AS, the polymorphisms underpinning disease association implicate peptide-binding features and classical antigen presentation in disease pathogenesis.
To generate HLA-B27 constructs for bacterial expression, commercially synthesized and Escherichia coli codon-opti-mized HLA-B*27:03 and HLA-B*27:05 DNA fragments with and without the P47G and/or W60A mutations were introduced into the pET30a vector using NdeI and HindIII (Fig.  S4C). Only the extracellular domains of the HLA-B27 alleles have been used for bacterial expression. All DNA constructs were sequenced to confirm their integrity.

Cell lines
The generation of the C1R cell lines expressing HLA-B*27:02 to HLA-B*27:09 has been described previously (35).
Genomic DNA of all established cell lines was extracted at various stages throughout the study. The HLA-B27 locus was PCR-amplified and sequenced to confirm allotype and mutant identities.

Cell lysis, SDS-PAGE, and immunoblotting
Cells were washed twice with cold PBS, resuspended in lysis buffer (1% (v/v) IGEPAL, 50 mM Tris (pH 8.0), 150 mM NaCl, 1 mM phenylmethylsulfonyl fluoride, 10 mM iodoacetamide, and protease inhibitor mixture (Roche Applied Science)), and incubated for 30 min on ice. Insoluble materials were removed by centrifugation (16,000 ϫ g for 15 min at 4°C), and the protein concentration of the supernatant was determined using a Direct Detect spectrometer (Millipore).
Proteins were separated by SDS-PAGE using 12% Mini-PROTEAN TGX precast gels (Bio-Rad), and 0.1 M DTT was added where necessary. The resolved proteins were transferred onto nitrocellulose membranes by Western blotting at 110 mA for 1 h, and the membranes were blocked with 2.5% milk powder dissolved in TBS-T. The blocked membranes were probed with the HC10 antibody (which binds HLA class I heavy chains (36,37)), which was harvested from hybridoma supernatants, followed by an anti-mouse horseradish peroxidase-conjugated secondary antibody (Abcam). For loading control, the membranes were probed with an anti-GAPDH antibody (Abcam) followed by an anti-rat horseradish peroxidase-conjugated secondary antibody (Abcam). Membranes were scanned with a ChemiDoc Touch imaging system (Bio-Rad), using Western Lightning Plus-ECL enhanced chemiluminescence reagent (PerkinElmer Life Sciences). Densitometry analyses were performed using the Bio-Rad ImageLab software (version 5.2.1) in Challenging the HLA-B27 homodimer theory combination with ImageJ (version 1.5i), Microsoft Excel 2016, and GraphPad Prism 7.

Flow cytometry
Cells were washed with ice-cold PBS, resuspended in flow buffer (1% fetal calf serum in PBS), and incubated on ice with the primary antibody ME1 (which binds to HLA-B27 (49)), which was harvested from hybridoma supernatants). Antimouse FITC or PE antibodies were used as secondary antibodies (Abcam). Propidium iodide was added to monitor cell viability, and 1% paraformaldehyde (in PBS) was used as fixative prior to antibody staining when required. Flow cytometric analysis was performed on a BD LSR II flow cytometer (BD Biosciences), and data were analyzed using FlowJo version 10.

Protein expression and purification
Soluble HLA class I-␤2m heterodimers of WT and mutated HLA-B*27:03 and HLA-B*27:05 in complex with the KK10 (KRWIILGLNK) or LRN (LRNQSVFNF) peptide were prepared as described previously (44). Briefly, ␤2m and the extracellular section of HLA class I were expressed as insoluble inclusion bodies in E. coli cells and purified to be refolded with the peptide. The pHLA complexes were further purified via anion-exchange chromatography, and their conformational integrity was confirmed by binding to the mAb W6/32 (50).

Crystallization, data collection, and structure determination
Crystals of each pHLA complex were grown by the hangingdrop, vapor-diffusion method at 20°C with a protein/reservoir drop ratio of 1:1, at a concentration of 10 mg/ml in 10 mM Tris-HCl, pH 8, 150 mM NaCl using 20 -30% PEG 4000, 0.2 M sodium acetate, and 0.1 M sodium citrate, pH 5.6. The pHLA crystals were soaked in mother liquor solution with an increased PEG 4000 concentration to 30% (w/v) prior to being flash-frozen in liquid nitrogen. All data were collected on the MX1 (51) and MX2 beamlines at the Australian Synchrotron, Clayton, using the ADSC-Quantum or Eiger detectors (at 100 K). Data were processed using XDS (52) and scaled with the Aimless software from the CCP4 suite (57). The structures were determined by molecular replacement using the PHASER (53) program with the HLA-B*27:05 KK10 for the MHC model without the peptide (Protein Data Bank accession number 4G9G (44)). Manual model building was conducted using the Coot software (54), followed by maximum-likelihood refinement with the Buster program (55). The final models have been validated using the Protein Data Base validation web site, and the final refinement statistics are summarized in Table 2

Thermal stability assay
To assess the stability of the allomorphs and the impact of the various polymorphisms, a thermal shift assay was performed. The fluorescent dye Sypro orange was used to monitor the protein unfolding. The assay was performed in a real-time detec-tion system (Corbett RotorGene 3000) originally designed for PCR. Each pHLA complex was measured twice in duplicate at two concentrations (0.5 and 1 mg/ml) in 10 mM Tris-HCl, pH 8, 150 mM NaCl and heated from 30 to 95°C with a heating rate of 1°C/min. The fluorescence intensity was measured with excitation at 530 nm and emission at 555 nm. The T m , or thermal melt point, represents the temperature for which 50% of the protein is unfolded (Table 3).

Ellman's assay
500 g of KK10-refolded HLA-B*27:03 or HLA-B*27:05 was resuspended in 500 l of buffer containing 20 mM Tris-HCl, 300 mM NaCl, and 1 mM EDTA. Ellman's reagent (DTNB, Sigma) was added to a final concentration of 100 M. The reaction was monitored at 412 nm for 1 h using a Jasco V-750 spectrophotometer.