Metal-triggered conformational reorientation of a self-peptide bound to a disease-associated HLA-B*27 subtype

Conformational changes of major histocompatibility complex (MHC) antigens have the potential to be recognized by T cells and may arise from polymorphic variation of the MHC molecule, the binding of modifying ligands, or both. Here, we investigated whether metal ions could affect allele-dependent structural variation of the two minimally distinct human leukocyte antigen (HLA)-B*27:05 and HLA-B*27:09 subtypes, which exhibit differential association with the rheumatic disease ankylosing spondylitis (AS). We employed NMR spectroscopy and X-ray crystallography coupled with ensemble refinement to study the AS-associated HLA-B*27:05 subtype and the AS-nonassociated HLA-B* 27:09 in complex with the self-peptide pVIPR (RRKWRRWHL). Both techniques revealed that pVIPR exhibits a higher degree of flexibility when complexed with HLA-B*27:05 than with HLA-B*27:09. Furthermore, we found that the binding of the metal ion Cu2+ or Ni2+, but not Mn2+, Zn2+, or Hg2+, affects the structure of a pVIPR-bound HLA-B*27 molecule in a subtype-dependent manner. In HLA-B*27:05, the metals triggered conformational reorientations of pVIPR, but no such structural changes were observed in the HLA-B*27:09 subtype, with or without bound metal ion. These observations provide the first demonstration that not only major histocompatibility complex class II, but also class I, molecules can undergo metal ion–induced conformational alterations. Our findings suggest that metals may have a role in triggering rheumatic diseases such as AS and also have implications for the molecular basis of metal-induced hypersensitivities and allergies.

The association of the human leukocyte antigen (HLA) 4 HLA-B*27 with rheumatic disorders, in particular ankylosing spondylitis (AS), has been known for nearly 50 years, but the underlying reasons have remained enigmatic (1,2). Because only a small fraction of HLA-B*27-positive individuals develop AS, other factors, genetic as well as environmental, have to contribute to initiate disease (3)(4)(5)(6). An understanding could come from functional, biochemical, and biophysical studies of HLA-B*27 subtypes with differential AS association (7,8). We have previously analyzed the subtypes HLA-B* 27 Despite the close subtype relatedness, healthy B*27:05-positive individuals and AS patients in particular possess cytotoxic T lymphocytes (CTL) in abundance that recognize the selfpeptide pVIPR (RRKWRRWHL, derived from vasoactive intestinal peptide type 1 receptor residues 400 -408), whereas such effector cells are very rare in individuals with B*27:09 (9,10). This indicates that the elimination of pVIPR-specific, self-reactive CTL during thymic selection processes is impaired in persons with the AS-associated subtype. Structural studies of the pVIPR-HLA-B*27 complexes revealed that pVIPR is displayed differentially by the two subtypes: B*27:09 presents the peptide conventionally in the canonical conformation (CC), with its middle bulging out of the binding groove (Fig. 1A), whereas the B*27:05 subtype displays pVIPR in a highly unusual dual conformation, about half resembling the CC structure seen in B*27:09 and the other half in a noncanonical conformation (NC), with the middle of the peptide contacting the polymorphic HC residue 116 (Fig. 1B). These results suggested that structural peculiarities were responsible for the differential recognition of the two subtypes by T cells and their difference in disease association (9). However, subsequent analyses by IR spectroscopy and molecular dynamics (MD) simulations revealed the existence of elevated conformational flexibility in the AS-associated subtype B*27:05, whereas the B*27:09 HC was less mobile (11)(12)(13)(14). Furthermore, an analysis of another pair of nearly identical HLA-B*27 subtypes, B*27:04 (AS-associated) and B*27:06 (not AS-associated), demonstrated that the former subtype, presenting pVIPR in a single CC conformation, was more flexible than B*27:06, although this subtype displayed pVIPR in a dual conformation, one of them in CC and the other in NC-binding mode (15). These results imply that AS-associated HLA-B*27 subtypes are characterized by an increased conformational flexibility and that structural peculiarities, such as dual conformations of a presented peptide, may have no influence on initiating the disease.
A detailed understanding of the interdependence of subtype polymorphism, in particular at HC position 116, and conformational flexibility of a bound peptide cannot be obtained by conventional X-ray crystallography alone. We have therefore revisited our previous structural findings using NMR spectroscopy and X-ray crystallography in combination with classical maximum likelihood and ensemble refinement (16,17). The first technique is known to yield very detailed information on the structure as well as the dynamics of a molecule, whereas classical X-ray crystallography provides atomic resolution without in-depth information on conformational flexibility. Ensemble refinement of crystallographic structures, on the other hand, seeks to bridge the gap between the two former techniques by including short, steered MD simulations, resulting in ensembles of structures for highly mobile residues. These novel experiments have led to a greatly improved understanding of differential peptide mobility in the B*27:05 and B*27:09 subtypes.
In addition, we can now demonstrate that the acquisition of the dual pVIPR conformation within the B*27:05 binding groove can be induced by selected metal ions that bind to exposed residues of the complex. The conformation of the same peptide bound to B*27:09 is, however, not affected. These findings have implications not only for the differential disease association of the HLA-B*27 subtypes, but more generally, also for the molecular basis underlying metal-induced hypersensitivities and allergies.

Peptide conformations analyzed by NMR spectroscopy
The conformational plasticity of the pVIPR peptide was initially assessed in both subtypes by NMR spectroscopy. Because the individual components of the MHC molecules are produced separately, the labeling pattern for each can be chosen independently. We have utilized this approach previously in an investigation in which solely ␤ 2 -microglobulin (␤ 2 m) was labeled with 15 N and thus was the only component visible in the spectra. Two-dimensional 1 H, 15 N correlations of good intensity and linewidth were recorded in that investigation in a comparatively short time (i.e. 1.5 h). For the NMR spectra recorded here, we employed samples containing 15 N, 13 C-labeled Arg at positions 1, 2, 5, and 6 of the peptide in otherwise unlabeled complexes. The intensities and linewidth obtained in NMR spectra are independent of the labeling pattern. Thus, we expected to obtain spectra with peak shapes comparable with those of the samples where only ␤ 2 m was labeled. Judging from the previously obtained crystal structures (9), we expected 1 H, 15 N correlations with three resonances for backbone H-N  pairs as well as up to four H-N pairs for arginine side chains for  pVIPR-B*27:09 and a double set of signals for pArg-5 and pArg-6 in pVIPR-B*27:05, due to the double conformation of the peptide (pArg-1 and pArg-2 are identically bound in both subtypes; Fig. 1, A and B). In contrast, we found that the resonance lines were broadened for both complexes, indicating conformational exchange on the microsecond to millisecond time scale. These broad lines made long experiment times necessary (20 h instead of 1.5 h). More importantly, whereas pVIPR-B*27:09 (Fig. 1C) showed the expected pattern in the 1 H, 15 N correlation, the spectrum of pVIPR-B*27:05 (Fig. 1D) did not contain the expected double set of signals, suggesting the presence of a different, most likely single, but highly mobile peptide conformation in this subtype.

Influence of temperature on the pVIPR conformations
This inconsistency between the NMR data, measured at 310 K and the X-ray crystallographic results obtained previously at 100 K prompted us to crystallize and analyze the pVIPR-B*27:05 complex at room temperature (RT) to exclude a possible artifact due to the cryogenic temperature. The crystal structure of the pVIPR-B*27:05 complex (see Table 1 for data collection and refinement statistics) unambiguously reveals that pVIPR does not adopt the expected dual peptide-binding mode, as described by us before (9). Instead, at RT, the peptide is displayed by B*27:05 only in the NC conformation. This is in agreement with the lack of double signals in the recorded NMR spectra but not in line with the crystallographic data obtained previously. To resolve this discrepancy, we re-investigated the peptide binding mode at cryogenic temperature (Table 1) and were able to confirm the single NC conformation as observed at RT ( Fig. 2A). Apart from an alteration of the side chain conformation of pLys-3 and an increased number of defined water molecules at 100 K, both structures are indistinguishable by conventional refinement procedures. Comparable analyses of pVIPR-B*27:09 showed that, irrespective of the temperature, the peptide is retained in the CC conformation, in agreement with our previously published results (9).

Origin of the dual pVIPR conformation in B*27:05
Intrigued by the origin of the dual peptide-binding mode described previously, we performed a detailed comparison of the crystal structures. This revealed that in the formerly obtained complexes of both subtypes, a metal ion was bound to pHis-8 that is missing in the de novo determined structures, suggesting that this metal ion might have induced the dual peptide conformation. We had previously considered this possibility but had rejected it, because the pVIPR-B*27:09 complex had been found to display the metal ion binding to pHis-8 as well.
A potential influence of metal ions on pVIPR conformations was investigated again after an initial selection, taking the possible coordination of metal ions by histidine residues into account. We were guided by the principle of "hard and soft
We soaked crystals with these metal ions and determined the crystal structures of all metal-treated complexes at 100 K to atomic resolution (see "Experimental procedures"; Fig. 2, B and C). Table 1 provides the X-ray data collection and refinement statistics. In addition, to precisely identify the respective metalbinding sites in the complexes, we collected the diffraction data at longer wavelengths to record the anomalous signal of the metal ions. These analyses show that the overall architecture of all complexes is not altered by the treatment with metal ions and that structural investigation of the crystals of pVIPR-B*27:05 and pVIPR-B*27:09 soaked with Mn 2ϩ , Zn 2ϩ , or Hg 2ϩ does not reveal any bound metal ion. However, soaking with Cu 2ϩ or Ni 2ϩ shows that the B*27:05 complex exhibits two metal-binding sites, the first being part of the peptide, where the side chain of pHis-8 is contacted, and the second located outside of the peptide binding groove, near the N terminus of the HC (Fig. 2B). In B*27:09, we observed a third Cu 2ϩ -binding site in the ␣3 domain of the HC (Fig. 2C).
Because the newly solved structure of pVIPR-B*27:09 possesses a different unit cell than the previously determined structure of this subtype (9), we re-investigated the latter structure (PDB code 1OF2) (9) in terms of metal-binding sites. As in B*27:05, two metal ions are indeed present as well. The first, contacting pHis-8, had tentatively been interpreted as Mn 2ϩ , whereas the second binding site at the N terminus of the HC had previously been overlooked. Consequently, together with the newly obtained results, there are two nonisomorphous data sets of the pVIPR-B*27:09 subtype with two identical metal-binding sites. In contrast, the data sets of the metal-soaked pVIPR-B*27:05 crystals and the previously published structure of this complex (PDB code 1OGT) (9) are isomorphous. It is thus possible to rule out any influence ; the appearance of natural abundance signals from the HC and ␤ 2 m in the center of the spectra is due to the long experiment duration. Despite the broad lines, the spectrum for B*27:09-complexed pVIPR appears as expected, with three backbone peaks for pArg-2, pArg-5, and pArg-6 (black) and four peaks for the side chains of the four Arg residues mentioned above (red). In the spectrum of pVIPR-B*27:05, however, only two peaks for backbone resonances are visible (black) instead of the expected five; in addition, only four peaks for the side chains of Arg residues (red) can be observed (six are expected).

Metal-induced HLA class I neo-antigen creation
of a neighboring HC within the crystal lattice on the peptidebinding mode.
This allows us to conclude that, apart from the polymorphic residue 116 and the subtype-specific peptide-binding mode, the pVIPR-B*27:09 complex is a precise structural image of pVIPR-B*27:05, with one remarkable exception: the binding of Cu 2ϩ or Ni 2ϩ induces a partial conformational reorientation of pVIPR (residues pLys-3 to pTrp-7) only in B*27:05 (Fig. 2, compare B and C), in which roughly 60% of the peptide is presented in the CC-binding mode and the remaining 40% in the original NC conformation. It seems likely that the simultaneous presence of two pVIPR conformations upon metal ion contact is due to the existence of energetically roughly equally favored peptidebinding modes in which the CC conformation in pVIPR-B*27:05 requires the presence of the inorganic ligand. The contact to His-197 of a neighboring HLA class I molecule (Fig. 3, A  and B) is almost certainly no option outside of a crystal lattice, so that a metal ion would probably have to be coordinated not only by pHis-8, but in addition by further suitable residues from other proteins to be stabilized. An example of how this could be accomplished is provided by an HLA-DR molecule and an interacting T cell receptor (TCR) whose residues jointly contact a Ni 2ϩ cation (23).
Because identical treatments of the pVIPR-complexed B*27:05 and B*27:09 subtypes with selected metal ions lead to clearly distinguishable structures, we can rule out that crystal-lographic artifacts are responsible for the observations. Instead, the Asp-116 -mediated increased flexibility of the B*27:05 subtype (11)(12)(13)(14) appears as a prerequisite for the peptide conformations reported here (Fig. 1D). Only in the case of the AS-associated subtype will the addition of Ni 2ϩ or Cu 2ϩ favor the acquisition of a dual peptide-binding mode.

Analysis of pVIPR dynamics by ensemble refinement
The results described above not only demonstrate that certain metals can influence the conformation of a displayed peptide, but also indicate that the dynamic behavior of pVIPR is less pronounced when bound to B*27:09 (Fig. 1C) than to B*27:05 (Fig. 1D). We sought to investigate this further by using ensemble refinements of three complexes presenting pVIPR, all at 100 K: 1) B*27:05 without bound metal ion, 2) B*27:09 with bound Cu 2ϩ , and 3) B*27:05 with bound Cu 2ϩ in CC-or in NC-binding mode. As expected from NMR spectroscopy (Fig. 1, C and  D), the dynamic differences between B*27:05-and B*27:09bound pVIPR are pronounced (Fig. 4). Whereas the primary anchor pArg-2 displays only negligible mobility in both subtypes, all other peptide residues show striking plasticity in the B*27:05 complex (peptide in NC-binding mode), in particular pLys-3, pTrp-4, pArg-5, pArg-6, and pHis-8. In B*27:09 (peptide in CC binding mode), only the solvent-exposed pArg-5 exhibits a moderate degree of plasticity. Although the solventaccessible pArg-1 side chain is mobile in both subtypes, the

Metal-induced HLA class I neo-antigen creation
ensemble of structures that is revealed by the improved type of refinement is larger in B*27:05 than in B*27:09. Furthermore, a comparison of pVIPR dynamics in the NCand CC-binding modes to B*27:05 was carried out (Fig. 5). This analysis reveals that pVIPR retains its overall plasticity irrespective of the peptide-binding mode, but pLys-3 and pArg-5 exhibit a diminished degree of conformational dynamics in the NC conformation. The most conspicuous difference following the binding of Cu 2ϩ concerns pHis-8 and pLeu-9; in both cases (Fig. 5), the side chains show very little mobility compared with the metal-free state (Fig. 4, A and B).
Finally, the metal ion-induced CC conformation of the peptide in B*27:05 exhibits a much higher extent of flexibility of nearly all amino acid side chains when compared with the CC-binding mode of pVIPR to B*27:09 (Figs. 4 (B and D) and 5 (B  and D)). This elevated plasticity even encompasses the pArg-1 side chain, at a distance of ϳ20 Å from the polymorphic HC residue Asp/His-116.

On the number of self-peptides permitting a dual binding mode
Using a data bank search, we also addressed the question of how many self-peptides could be identified that would allow metal ions to bind in a manner comparable with that observed here for pVIPR-B*27:05, thereby possibly leading to dual peptide presentation modes by B*27:05 molecules. In this way, a distinct "arthritogenic" peptide (24), currently still an elusive entity, might not be needed as triggering agent for the development of AS. This data bank search (see "Experimental procedures") revealed that 823 human proteins could act as donors for 873 nonamer self-peptides that might be displayed by the B*27:05 subtype, appear potentially able to bind metal ions, and could be subject to conformational reorientation. Because B*27:05 exhibits no absolute requirement for pArg-2 of a bound peptide (25), the number of such self-peptides might be even larger. Interestingly, none of the 26 candidate peptides identified by Schittenhelm and colleagues (26,27) as potentially arthritogenic fulfills the requirements outlined above.

Discussion
Our study reveals three principal experimental findings: 1) the pVIPR peptide is normally presented in a single (NC) conformation by B*27:05, not in a dual (NCϩCC) binding mode as published previously by us (9); 2) the binding of selected metal ions to pHis-8 leads to the acquisition of a dual (NCϩCC) conformation of pVIPR, but only in the AS-associated B*27:05 subtype; 3) both NMR spectroscopy and ensemble refinements of crystallographic structures show that pVIPR exhibits a different

Metal-induced HLA class I neo-antigen creation
dynamics when bound to B*27:05, as compared with B*27:09, where it is less mobile than in B*27:05.
The structure of an MHC class I molecule can thus be affected in an allele-specific fashion by certain metal ions that bind to an exposed histidine residue of a displayed peptide, leading to a gross structural reorientation of the ligand, but only in the B*27:05 subtype. Cu 2ϩ and Ni 2ϩ are known to bind to histidine side chains (20 -22), and it is conceivable that the mobility of pHis-8 is reduced (compare Figs. 4A and 5A) due to metal binding. In what way the dynamics of several further peptide residues is influenced, however, remains currently a matter of speculation. Changes involving the protonation of His resi-

Metal-induced HLA class I neo-antigen creation
dues have already been described for HLA-DR molecules (28) as well as for pVIPR-B*27:09, where a decreased stability of the complex was detected when the pH of the medium was reduced from 7.5 to 5.6 (13). In addition, using fluorescence depolarization and pK a calculations, we have previously described an allosteric interaction between pHis-8 and the HC Glu-45/63 pair of residues, which contribute to anchoring pArg-2 within the B pocket of the binding groove (29). Comparable long-range interactions have also been identified for HC residue 116 with ␤ 2 m (30) or through ␤ 2 m to the most distal region of the ␣3 domain (31). NMR spectroscopic analyses of the B*27:05 and B*27:09 HC might shed new light on the reasons for the far-reaching effects of conformational changes affecting pHis-8. 5 The novel insight into the acquisition of the NC-and CC-pVIPR binding modes explains functional data (9,10) showing the existence of CTL from B*27:05-positive individuals that react with this peptide in the context of both subtypes; it seems very likely that it is the CC conformation that forms the basis for recognition by these CTL, because CTL are expected to be negatively selected in the thymus on B*27:05-positive cells displaying NC-pVIPR. Only a single CTL from a B*27:05-positive donor was found that reacted with B*27:05 but not B*27:09, in line with an NC-pVIPR-specific reactivity (9).
The two conformations of pVIPR in B*27:05 resemble also those found for the viral pLMP2 peptide (RRRWRRLTV, derived from Epstein-Barr virus latent membrane protein 2 (PDB code 1UXS (32)) (Fig. 6C) and for the citrullinated pVIPR (PDB 3B6S) (33) (Fig. 6D), in the same subtype, respectively. In the former case, pLMP2 is displayed solely in the NC confor-mation (32), whereas citrullinated pVIPR is presented exclusively in the CC-binding mode due to the loss of the positively charged guanidinium group of pArg-5 through the replacement of this peptide residue by citrulline (33). The various peptide-binding modes that we describe here for the B*27:05 subtype could be regarded as an example of how structural differences between self-ligands (Fig. 6A), altered self-peptides (also termed neo-antigens) (Fig. 6, B and D), and closely related foreign antigens (Fig. 6C) presented by MHC class I molecules can be blurred (34). The higher degree of flexibility, which characterizes the B*27:05 binding groove, in contrast to that of B*27:09 (11)(12)(13)(14), forms the basis for the differential, Asp/His-116 -dependent display of similar ligands.
In line with a recent ensemble refinement study of several MHC-bound peptides by Fodor et al. (35), our work demonstrates also that the extent of peptide dynamics has been underestimated using conventional X-ray crystallography refinement alone. The results from our NMR spectroscopic experiments and the dynamics-oriented refinements support each other (Figs. 1 (C and D) and 4 (A and B)). They show that ensemble refinement techniques will greatly improve our understanding of the conformational plasticity of peptides presented by MHC molecules and shed new light on their interactions with TCR and other extracellular ligands.
However, plasticity of HC residue 116 and the peptide terminus-binding F pocket in general (see also the subtypes B*44:02 and B*44:05 (36)) is not only crucial for permitting appropriate T-cell responses but affects also interactions with intracellular ligands, most prominently tapasin (37,38) and TAP-binding protein (TAPBPR) (39 -41). The great variety of molecular contacts has recently been reviewed (42). Virtually all of these interactions critically rely on dynamics of the binding partners. Due to the experimental difficulties accompanying NMR spectros-  Fig. 3A. D, pVIPR with pArg-5 replaced by citrulline. This neo-antigen is bound in the CC conformation, closely resembling that found for pVIPR-CC (B), because a contact between the side chain of citrulline and Asp-116 cannot be established.

Metal-induced HLA class I neo-antigen creation
copy, we expect that extracting dynamic information from existing X-ray diffraction data sets by ensemble refinements (16,35) will be very helpful in providing further insight into the conformational flexibility of MHC molecules and their protein ligands.
The results presented here not only illuminate the dynamics of HLA class I molecules but also allow us to draw some conclusions regarding the involvement of metal ions in disease. Although such ions perform essential functions in physiological processes, as in enzyme catalysis, signal transduction, and electron transfer, many are also known to cause grave health problems in humans, leading to hypersensitivities and allergies in more than 10% of the world's population (43). There are several ways by which metals could initiate inappropriate T cell responses (44). Here we describe one possibility, the conformational reorientation of a bound peptide, induced by its interaction with Cu 2ϩ or Ni 2ϩ . Another mechanism is observed in patients with chronic beryllium disease (45), where Be 2ϩ is coordinated by residues of an HLA class II molecule (HLA-DP2) forming an acidic pocket within the binding groove. The metal ion contributes to the selection of peptides that share pAsp-4 and pGlu-7, apparently to improve the stability of the assembly by interacting with the metal ion. A TCR has no opportunity to recognize the bound Be 2ϩ ion but is indirectly affected by conformational changes of the HLA-DP2 molecule's surface and its altered electrostatic potential (34). Much less is known, however, about the molecular basis for nickel-induced contact hypersensitivities, although this metal is the most common occupational and public contact allergen (43). There is evidence that Ni 2ϩ ions can interact with an HLA-DR-bound peptide (46), with an HLA-DR molecule and a bound peptide (47), as well as jointly with an HLA-DR molecule and a TCR (23), but there are no structural data to support any of these findings.
As we show here, metals can influence antigen presentation also in the case of an MHC class I molecule and are therefore potential triggering agents for disease development due to their omnipresence. An example of the role that metals can play in setting off spondyloarthropathies is provided by Brown Norway rats, whose relative genetic resistance to Chlamydia-induced reactive arthritis (an HLA-B*27-associated disease in humans) can be overcome by injections of mercuric chloride, leading to a marked exacerbation of the severity of arthritic symptoms in the animals (48). Explanatory difficulties connected with the fact that only certain HLA alleles are associated with a particular disease could be accounted for by assuming that conformational reorientations of a peptide as described here for the pVIPR-B*27:05 complex following exposure to Cu 2ϩ or Ni 2ϩ are less likely in subtypes that lack a disease association, such as B*27:09. As mentioned before, a particular arthritogenic peptide (24), possibly derived from a microorganism, would be superfluous in this scenario. A higher degree of molecular dynamics in the AS-associated subtypes B*27:05 and B*27:04 than in the nonassociated B*27:09 and B*27:06 molecules (15) supports the idea that HLA-B*27 polymorphisms, molecular flexibility, concomitant peculiarities in peptide repertoire, and presentation and ultimately AS association are intimately connected with each other.
Furthermore, the so far unexplained fact that smoking contributes to a more severe course of AS (49 -51) and other diseases, including rheumatoid arthritis (49,50), is also of interest in the context described here, because tobacco smoke contains not only toxic organic compounds, but also several metals, including nickel (52). In individuals with HLA class I or II alleles predisposing to rheumatic disorders, prophylactic measures can thus be suggested, including the avoidance of tobacco abuse and environmental tobacco smoke exposure as well as minimizing the contact with metals known to cause allergies. Our results might even provide the rationale for a therapeutic intervention, such as a treatment with chelating agents (53). For AS, HLA-B*27:05/human ␤ 2 m-transgenic rats developing AS-like symptoms (54) would be suitable to test several of these assumptions.
It remains to be determined, however, whether conformational reorientations of peptides bound to class I antigens other than HLA-B*27 can provide the basis for neo-antigen creation in further cases of metal-induced hypersensitivities and allergies (55)(56)(57) and whether hypersensitivities and autoimmunity do really represent two sides of the same coin (45,58).

NMR spectroscopy
Samples of the complexes were produced as described previously (30) and contained 17.6 mg/ml pVIPR-B*27:09 and 10.1 mg/ml pVIPR-B*27:05, respectively, with the peptide synthesized at the Core Facility of the University of Leipzig (Germany) using conventional Fmoc-peptide synthesis utilizing 15 N, 13 Clabeled arginine. The buffer used for NMR spectroscopy contained 150 mM NaCl and 10 mM sodium phosphate, pH 7.5. NMR spectra were recorded as SOFAST-1 H, 15 N-HMQC (59) at 310 K in a buffer free from Cu 2ϩ , Hg 2ϩ , Mn 2ϩ , Ni 2ϩ , and Zn 2ϩ on an AV750 Bruker spectrometer (750 MHz 1 H frequency) with identical parameters: 4000 scans, data size 512( 1 H)*64( 15 N) complex points, t Hmax ϭ 440.8 ms, t Nmax ϭ 16.8 ms. A recycle delay of 0.1 s was used, resulting in an experiment duration of 20 h.

Crystallography
For crystallization experiments, the pVIPR-B*27:09 and pVIPR-B*27:05 proteins were prepared as described before (9). One protein preparation from a given subtype was used for all crystallization and soaking experiments. Crystallization trials and cryoprotection were performed according to the protocol described previously (9). Soaking experiments were carried out at a concentration of 50 mM for the chloride salt of the respective metal ion for 2 min. For measurements at RT, the crystals were mounted in MicroRT TM capillaries (MiTeGen, Ithaca, NY). X-ray diffraction data sets were collected at beamline 14.2 of the MX Joint Berlin laboratory at the BESSY II in Berlin, Germany, or at the beamline P14 at PETRA III in Hamburg, Germany. Anomalous diffraction data were collected at the wavelength as indicated in Table 1. Diffraction data were processed with the XDS package (60). Data collection and refine-Metal-induced HLA class I neo-antigen creation ment statistics are given in Table 1. The following complexes with pVIPR were analyzed by X-ray crystallography: B*27:05 at room temperature, B*27:05 at 100 K, B*27:05 at 100 K soaked with Cu 2ϩ , Ni 2ϩ , Mn 2ϩ , Zn 2ϩ , or Hg 2ϩ ; B*27:09 at room temperature, B*27:09 at 100 K soaked with Cu 2ϩ , Ni 2ϩ , Mn 2ϩ , Zn 2ϩ , or Hg 2ϩ . Atomic coordinates and structure factor amplitudes were deposited in the Protein Data Bank under accession codes 5IB1 (pVIPR-B*27:05 at RT), 5IB2 (pVIPR-B*27:05 at 100 K), 5IB3 (pVIPR-B*27:05 Cu 2ϩ -soaked), 5IB4 (pVIPR-B*27:05 Ni 2ϩ -soaked), and 5IB5 (pVIPR-B*27:09 Cu 2ϩ -soaked).

Structure determination, refinement, and analysis
The structures were solved by molecular replacement with pVIPR-B*27:05 (PDB code 1OGT (9)) as search model using the program PHASER (61) and refined by annealing and maximum-likelihood restrained refinement in PHENIX (62) followed by iterative model building cycles in COOT (63). Water molecules were positioned with COOT and manually inspected. Intermediate and final structures were evaluated with MOLPROBITY (64). The starting structures for ensemble refinements as implemented in PHENIX (62) were prepared as described (16). Briefly, alternate conformations of amino acid side chains were removed from the deposited structures in the PDB, and the occupancies were adjusted to 1. For the structure pVIPR-B*27:05-Cu 2ϩ , all double conformations were removed except those of the pVIPR peptide. The occupancy of each of the two peptide conformations was set to 0.5. Explicit hydrogen atoms were generated with phenix.ready_set. The ensemble refinements were executed with the standard settings except that the scale factor for X-ray/stereochemistry weight (wxc_scale) was adjusted to the value used for standard real-space refinement. For glycerol molecules, harmonic restraints were set to avoid stochastic displacement during simulations. Figures were prepared with PyMOL (65).

Data bank search
SwissProt was accessed on September 15, 2015 to obtain an estimate of the number of human protein-derived nonamer peptides that could principally be presented by B*27:05 and allow metal-induced conformational reorientations. The peptides had to meet the following criteria: no Pro at p1 (Pro is not accepted at this position), pArg-2 (a nearly obligatory anchor for HLA-B*27 molecules (25), no Asp or Glu at p9 (these would not bind with high affinity to B*27:05), and also no Arg or Lys at this position (contact with Asp-116 would preclude the interaction between pArg-5 and Asp-116) (66). As a potential contact site for a metal ion, pHis-8 should also be present as well as a basic residue at p5 to allow binding to HC residue Asp-116 in the NC conformation.