Role of a Novel Human Leukocyte Antigen-DQA1*01:02;DRB1*15:01 Mixed Isotype Heterodimer in the Pathogenesis of “Humanized” Multiple Sclerosis-like Disease*

Background: HLA-DR15 haplotype (DRB1*15:01-DQA1*01:02-DQB1*0602-DRB5*01:01) association with multiple sclerosis (MS) is conventionally attributed to effects from HLA-DRB1*15:01, with impact on MS risk from the neighboring HLA-DQ locus unclear. Results: Functional studies show MS-like disease dependent on a novel DQA1*01:02;DRB1*15:01 mixed isotype heterodimer. Conclusion: DQA1*01:02 within a mixed heterodimer may contribute to MS pathogenesis. Significance: HLA class II/MS susceptibility models may require broader reinterpretation. Gene-wide association and candidate gene studies indicate that the greatest effect on multiple sclerosis (MS) risk is driven by the HLA-DRB1*15:01 allele within the HLA-DR15 haplotype (HLA-DRB1*15:01-DQA1*01:02-DQB1*0602-DRB5*01:01). Nevertheless, linkage disequilibrium makes it difficult to define, without functional studies, whether the functionally relevant effect derives from DRB1*15:01 only, from its neighboring DQA1*01:02-DQB1*06:02 or DRB5*01:01 genes of HLA-DR15 haplotype, or from their combinations or epistatic interactions. Here, we analyzed the impact of the different HLA-DR15 haplotype alleles on disease susceptibility in a new “humanized” model of MS induced in HLA-transgenic (Tg) mice by human oligodendrocyte-specific protein (OSP)/claudin-11 (hOSP), one of the bona fide potential primary target antigens in MS. We show that the hOSP-associated MS-like disease is dominated by the DRB1*15:01 allele not only as the DRA1*01:01;DRB1*15:01 isotypic heterodimer but also, unexpectedly, as a functional DQA1*01:02;DRB1*15:01 mixed isotype heterodimer. The contribution of HLA-DQA1/DRB1 mixed isotype heterodimer to OSP pathogenesis was revealed in (DRB1*1501xDQB1*0602)F1 double-Tg mice immunized with hOSP(142–161) peptide, where the encephalitogenic potential of prevalent DRB1*1501/hOSP(142–161)-reactive Th1/Th17 cells is hindered due to a single amino acid difference in the OSP(142–161) region between humans and mice; this impedes binding of DRB1*1501 to the mouse OSP(142–161) epitope in the mouse CNS while exposing functional binding of mouse OSP(142–161) to DQA1*01:02;DRB1*15:01 mixed isotype heterodimer. This study, which shows for the first time a functional HLA-DQA1/DRB1 mixed isotype heterodimer and its potential association with disease susceptibility, provides a rationale for a potential effect on MS risk from DQA1*01:02 through functional DQA1*01:02;DRB1*15:01 antigen presentation. Furthermore, it highlights a potential contribution to MS risk also from interisotypic combination between products of neighboring HLA-DR15 haplotype alleles, in this case the DQA1/DRB1 combination.

MS is a disease with a strong underlying genetic component. Susceptibility has long been linked to the HLA-DR15 haplotype (DRB1*15:01-DQA1*01:02-DQB1*0602-DRB5*01: 01) and, particularly, the HLA-DRB1*1501 locus, dominating MS risk in Caucasians (3,4). Genome-wide association and candidate gene studies have suggested that the effects of the DR15 haplotype on MS risk are driven by the HLA-DRB1*15:01 allele and that effects on MS from neighboring alleles may reflect linkage disequilibrium (5,6). However, linkage disequilibrium, particularly between DRB1*1501 and DQA1*0102-DQB1*0602, makes it difficult for genetic studies to discern, without functional studies, whether the functionally relevant effect on MS derives from DRB1*15:01 only, from its neighboring DQA1*01:02-DQB1*06:02 or DRB5*01:01 genes of the HLA-DR15 haplotype, or from their combinations or epistatic interactions. HLA-transgenic (Tg) mice constitute a valuable resource for dissecting the association of disease susceptibility with the specific gene products of the HLA-DR15 haplotype. The susceptibility of HLA-DRB1*1501-Tg mice to MBP-or MOG-induced EAE (7,8) is consistent with the primary contribution of DRB1*15:01 to MS risk. However, our recent studies showing that the DQB1*0602 allele, rather than DRB1*1501, determines disease susceptibility to MOBP (9) and to PLP (10), one of the most prominent and "MS-implicated" CNS target antigens, are inconsistent with the conclusion that the DR15 haplotype-associated MS risk is driven by the DRB1*15:01 allele only.
OSP/claudin-11, the third most abundant CNS myelin protein, comprising ϳ7% of total CNS myelin proteins (11,12), is also a primary CNS myelin target antigen in MS (2), but, unlike MBP, PLP, MOG, and MOBP, neither the potentially pathogenic OSP epitopes relevant to HLA-DR15 ϩ MS nor the HLA-DR15 haplotype gene product(s) that may determine OSP-related disease susceptibility has been investigated.
In this study, HLA-DR15-Tg mice were employed first to determine immunogenic/immunodominant epitopes of human OSP/claudin-11 (hOSP) and their potential pathogenic relevance to HLA-DR15 ϩ MS and, second, to study whether DRB1*15:01 or DQA1*01:02-DQB1*06:02 alleles or their combinations control pathogenic T-cell autoimmunity against hOSP. We show that the DR15 haplotype-associated T-cell autoimmunity against hOSP is probably focused on a single immunodominant epitope (hOSP(142-161)) and that pathogenic autoimmunity against hOSP(142-161) is controlled by the DRB1*15:01 allele, not only as the isotypic heterodimer but, unexpectedly, also as a functional DQA1*01:02;DRB1*15:01 mixed isotype hybrid heterodimer. Here, we demonstrate for the first time a functional DQA1*01:02;DRB1*15:01 mixed isotype heterodimer and show its potential relevance to MS pathogenesis. The demonstration of functional interaction between DQA1*01:02 and DRB1*15:01 not only further highlights the complex effects of HLA class II genes on susceptibility to MS but also provides a rationale for a potential contribution of the DQA1*01:02 allele to MS risk through antigen presentation via functional DQA1*01:02;DRB1*15:01 mixed isotype heterodimers and not only through epistasis with DRB1*15:01 (13).
The HLA-(DRB1*1501xDQB1*0602)F1 double-Tg mice were bred at the animal facilities of the Weizmann Institute of Science. The Institutional Animal Care and Use Committee of the Weizmann Institute of Science approved the experiments, which were performed in accordance with its relevant guidelines and regulations.
Recombinant Human ⌬OSP and OSP Synthetic Peptides-h⌬OSP was constructed to delete the sequences encompassing the hydrophobic putative transmembrane domains (Fig. 1, a  and b), thus enabling the expression of soluble protein. h⌬OSP was produced by cloning the coding nucleotide sequences into pRSET bacterial expression vectors, as described previously for the preparation of m⌬OSP (17), with the adjustment of some of the primers used for overlapping PCR extension to accommodate the few amino acid differences between mouse and human ⌬OSP shown in Fig. 1 (in boldface type). Sequences were verified, and open reading frames for fusion protein with histidine tags were confirmed. Expression in Escherichia coli and purification of h⌬OSP on Ni 2ϩ -nitrilotriacetic acid-agarose was carried out as detailed in the study by Kerlero de Rosbo et al. (18). A recombinant protein named Y-DMP (Y-diabetes mellitusrelated protein), a recombinant protein encompassing epitopes of several antigens relevant to diabetes 3 that we expressed in E. coli and affinity-purified similar to the recombinant hOSP, was used as a non-relevant control recombinant protein. The amino acid sequences of synthetic overlapping peptides spanning h⌬OSP are listed in Table 1.
Induction of EAE-Mice were injected subcutaneously at one site in the flank with 200 l of emulsion containing 200 g of h⌬OSP or m⌬OSP or 200 g of peptide in complete Freund's adjuvant (CFA) containing 300 g of M. tuberculosis H37Ra (catalog no. 3114-25, Difco). Mice received 300 ng of pertussis toxin (catalog no. P-9452, Sigma) in 500 l of PBS in the tail vein immediately and 48 h after the immunization (Protocol 1). In some cases, as indicated, the mice received an identical booster immunization on the flank, 1 week later, but without the administration of pertussis toxin (Protocol 2). In another protocol, mice were injected intraperitoneally with cyclophosphoamide (25 mg/kg) and 2 days later were inoculated in the flank with 200 l of emulsion containing 200 g of h⌬OSP in CFA containing 300 g of M. tuberculosis H37Ra (Protocol 3). Following the encephalitogenic challenge, mice were observed daily, and clinical manifestations of EAE were scored on a scale of 0 -6 as follows: 0, no clinical signs; 1, loss of tail tonicity; 2, flaccid tail; 3, hind leg paralysis; 4, hind leg paralysis with hind body paresis; 5, hind and fore leg paralysis; 6, death, as previously described (17).
Adoptive Transfer of EAE by T-cell Lines-Selection of T-cell lines and the cell transfer experiments were conducted as described previously (19,20) Briefly, line T-cells were stimulated in vitro with their respective antigen for 3 days, and the activated T-cells were injected into the tail vein of irradiated (4 grays) naive syngeneic mice. Mice were observed and scored daily as described (17).
T-cell Proliferative Response-Mice were immunized with 150 g of h⌬OSP or individual peptides emulsified in CFA containing 150 g of M. tuberculosis H37Ra (catalog no. 3114-25, Difco). 10 or 14 days postimmunization, draining lymph nodes or spleens, respectively, were removed and cultured in vitro in triplicates in microtiter plates, in the absence or presence of relevant antigens, as described previously (21). The cultures were incubated for 48 -72 h at 37°C in humidified air containing 7.5% CO 2 . [ 3 H]Thymidine (1 mCi/well) was added for an additional 16 h of incubation, and the cultures were harvested and counted using a Matrix 96 Direct Beta Counter (Packard Instrument Co.). The results were expressed as stimulation index (S.I.) (mean cpm of antigen-containing cultures/mean cpm of medium-containing cultures). In some experiments, as indicated, mice were immunized as for induction of EAE (Protocol 1).
Affinity Binding Assays-Recombinant HLA-DRB1*1501 (DRA1*01:01;DRB1*15:01) and HLA-DQB1*0602 (DQA1*01: 02;DQB1*06:02) proteins were expressed in Drosophila melanogaster cells (S2) under the control of the metallothionein pro-motor as soluble molecules engineered as described (23,24). Proteins were antibody-affinity-purified from supernatants of HLA class II-transfected S2 cells. Briefly, supernatants were harvested from CuSO 4 -induced HLA class II-transfected S2 cells, centrifuged at 10,000 ϫ g, and filtered through 0.4-m filters before being passed over an anti-HLA-DR (L243) or anti-HLA-DQ (SPV-l3) antibody-coupled column. After washing the column with 25 column volumes of PBS, 0.05% Nonidet P-40, captured HLA class II molecules were eluted with 4 -5 column volumes of 0.15 M NaCl, 50 mM diethylamine, pH 11, directly into one-twentieth volume of a neutralizing buffer (2 M Tris-HCl, pH 6.3). The eluate was then concentrated, and the buffer was changed to PBS, 0.1% NaN 3 by ultracentrifugation in a Centricon p-20 unit (Millipore). The product was analyzed for purity by SDS-PAGE, and total protein concentration was determined by the bicinchoninic acid assay (Sigma) using bovine serum albumin (BSA) as a reference protein. Inhibition assays were performed essentially as described (24). Varying concentrations of competitor peptide were incubated overnight at 25°C with constant concentrations of either HLA-DRB1*1501 (20 nM) plus biotinylated myelin basic protein peptide b-MBP(85-99) (ENPVVHFFKNIVTPR) (25) agonist peptide (10 nM) or HLA-DQB1*0602 protein (100 nM) plus biotinylated tyrosine phosphatase b-IA2(495-509) (10 nM) in PBS containing 0.06% Nonidet P-40 and 0.1 M sodium citratephosphate buffer, pH 7.0. The formed DRB1*1501⅐peptide complexes or the HLA-DQB1*0602⅐peptide complexes were quantified by incubation for 2 h at 4°C in a 96-well microtiter plate (MaxiSorb, Nunc, Roskilde, Denmark) precoated with 1 g/well L243 (anti-DR) or SPV-L3 (anti-DQ) in carbonate buffer, pH 9.0, and blocked with 5% FCS in PBS. The free peptides and the unbound complexes were washed off of the plate with 0.05% Tween 20 in PBS. Europium-labeled streptavidin was added at 100 ng/ml and incubated for 1 h at 4°C. After washing, Eu 3ϩ was released by adding enhancement solution to the wells and measured in a time-resolved fluorometer (Victor, PerkinElmer Life Sciences). IC 50 values were monitored as the concentration of unlabeled peptide that prevented 50% of the biotinylated peptide from binding to the HLA-DRB1*1501 or the HLA-DQB1*0602 protein. For Fig. 6C, horseradish peroxidase-coupled streptavidin (Abcam) was diluted 5000-fold in washing buffer and added to the ELISA wells and incubated for 1 h at 4°C. The substrate, 3Ј5,5Ј-tetrametylbenzidine (DAKO), was added and stopped after 15-30 min. with an equal volume of H 2 SO 4 , and the transformed substrate was measured at 450 nm.
Core Epitope Prediction and Molecular Modeling-Putative core epitopes within the h⌬OSP and m⌬OSP sequences were detected as described previously for I-A s molecules (18). Thus, the experimental structures of HLA-DRB1*1501(DRA;DRB1* 15:01) or -DQB1*0602 (DQA1*01:02;DQB1*06:02) molecules (Protein Data Bank codes 1BX2, 1YMM, and 2WBJ for DRB1*1501 and 1UVQ for DQB1*0602) were analyzed, and 4 ϫ 20 binding preference matrices were constructed, which tabulate estimates of the tendency of an amino acid to bind in one of the specificity-determining pockets of the HLA binding site, P1, P4, P6, and P9. The binding estimates were based on the physical properties of the pockets, their size, hydrophobicity, polar-ity, and charge. Thus, highly preferred residues were given a score of Ϫ2, less preferred residues were scored Ϫ1, residues that would neither contribute nor hinder the binding were scored 0, and residues that were likely to hinder binding weakly or strongly (e.g. by being too large) were given a score of 1 or 2, respectively. The binding preference matrices were used to estimate the binding ability of overlapping 9-amino acid segments of the target sequences to the HLA binding site. Previous experience with this in-house computer program successfully predicted core epitopes (18,21,22,26), all of which scored Ϫ5 or lower. Higher scores were therefore disregarded in this study. An accurate estimate of the binding preferences in pocket P6 was obtained with ANCHORSMAP.2, a modified version of ANCHORSMAP (27), which also considers threonine and proline probes. This procedure accurately maps preferred binding positions and estimates the binding energies of excised amino acid side chains, taking into consideration that the amino acid is part of a protein. Anchoring spots with ⌬G Յ Ϫ4 kcal/mol were shown to correspond to particularly strong experimental hot spot locations (27), and values below Ϫ3 kcal/mol indicate good binding.
Initial model structures of the HLA peptide-binding domains in complex with peptides that are 12 amino acids long, which include a core epitope, were constructed based on the experimental structure of the DR15⅐peptide complex or DQ6⅐peptide complex. Experimental structures of hybrid HLA molecules are not available; therefore, models were constructed by superposing the experimental structures of DR15 and DQ6 using the corresponding C␣ positions of the peptide binding domains and then combining the ␣ chain of one HLA with the ␤ chain of the other. Each initial model underwent a molecular dynamics (MD) simulation in water as follows: (i) initial energy minimization of the complex immersed in a box of water and neutralized; (ii) 200-ps MD simulation of the solvent molecules, restraining the non-hydrogen HLA and peptide atoms; (iii) 200-ps MD simulation of the solvent and peptide, restraining the non-hydrogen atoms of the HLA; and (iv) 1-ns MD simulation of the HLA, peptide, and solvent, restraining only the C␣ atoms of the HLA. This procedure produced plausible models of the HLA⅐peptide complexes in which the HLA molecules mostly retained their initial structures, whereas the peptides could move. Convergence of the final MD simulation was tested by calculating the root mean square difference between steps along the simulation and the starting geometry, using all non-hydrogen atoms of the peptide after superposition of the HLA C␣ atoms. We used the Gromacs package for these computations (28). UCSF Chimera software (29) was used for visualization and figure preparation.
PathologicalExamination-Micewereperfusedwith4%paraformaldehyde in PBS, and the tissues were postfixed for 24 h at 4°C. Histological evaluation was performed on paraffin-embedded sections of brain and spinal cords that were sampled 19 days postimmunization as the experiment was terminated. Paraffin sections were stained with H&E and Luxol fast blue to assess inflammation and demyelination, respectively. In consecutive sections, immunohistochemistry was performed with Abs directed against the following targets: macrophages/activated microglia (MAC3, BD Pharmingen; iba-1, Wako-chem) and T-cells (CD3, Chemicon International) (18). For staining, paraffin sections were pretreated with a steamer for 60 min. Bound primary Ab was detected with a biotin-avidin technique as described in detail previously (18).
Co-Immunoprecipitation-Splenocytes derived from Tg mice were treated with ammonium chloride potassium buffer to lyse red blood cells and washed with ice-cold PBS. The nucleated cells were lysed overnight with shaking (4°C) in 2 ml of lysis buffer (25 mmol/liter Tris-HCl, 2 mmol/liter sodium orthovanadate, 0.5 mmol/liter EGTA, 10 mmol/liter NaF, 10 mmol/liter sodium pyrophosphate, 80 mmol/liter ␤-glycerophosphate, 25 mmol/liter NaCl, 0.5% Nonidet P-40, protease inhibitor mixture (Sigma) diluted 1:500, pH 7.4). Cell extracts were then centrifuged at 20,000 ϫ g for 15 min at 4°C, and protein concentration was determined in supernatants using a Bradford assay. Supernatant fractions were incubated with 40 l of protein G/A-agarose beads (1:1) (Santa Cruz Biotechnology) for a lysate preclearing step. Preclearing was repeated three times to remove endogenous immunoglobulins. Simultaneously, protein G-agarose beads were incubated with 5 g of anti-HLA-DR␤ antibodies (TAL 14.1, Santa Cruz Biotechnology) with shaking for 3 h at 4°C and then washed three times with lysis buffer. The precleared lysates were incubated with 30 l of the antibody-bound (anti-HLA-DR␤) protein G-agarose beads overnight at 4°C, followed by washing four times with lysis buffer. Beads were resuspended in sample buffer and boiled for 5 min. Proteins were resolved by 12% SDS-PAGE. Gels were transferred to a Nitrocellulose membrane, blocked with 2% BSA in TBS-Tween buffer, and incubated with primary antibodies (anti-HLA-DQA1 EPR7300, Abcam) overnight at 4°C. After washing three times with TBS-Tween buffer, blots were incubated with HRP-conjugated secondary antibody. The blots were then washed six times with TBS-Tween buffer and analyzed by standard enhanced chemiluminescence (RPN2232, GE Healthcare).
In further attempts to demonstrate encephalitogenic potential of any of the immunogenic/immunodominant epitopes of h⌬OSP, non-HLA-DR15 haplotype-relevant controls (DRB1* 1502-Tg and DQB1*0601-Tg mice) were also immunized to induce EAE by the hOSP peptides encompassing their relevant immunodominant epitope as well as by the whole h⌬OSP. As shown in Table 2 Table 2) may suggest that HLA-DR15-associated T-cell autoimmunity against OSP is not pathogenic. The DQB1*0602-Tg mice were expected to be refractory to EAE induction by the OSP(55-80) and OSP(142-161) immunodominant regions of h⌬OSP not only because they represent cryptic epitopes in these Tg mice, but also due to their lack of pro-inflammatory T-cell cytokines against these immunodominant epitopes. In contrast, the consistent failure (Table 2) to induce EAE in DRB1*1501-Tg mice by phOSP(142-161) encompassing the sole immu-nodominant region of h⌬OSP for this Tg mice remained enigmatic, particularly in view of the ability of DRB1*1501-Tg mice to generate potentially pathogenic Th1/Th17 cells reactive against phOSP(142-161) (Fig. 5).
Because pathogenic autoimmunity against the hOSP is analyzed in the HLA-Tg mice expressing CNS mOSP, the failure in inducing EAE by phOSP(142-161) in DRB1*1501-Tg mice could be attributed to species differences between human and mouse OSP, although the hOSP(142-161) region (PVCAH-RETTIVSFGYSLYAG) differs from the mouse counterpart (PVCAHREITIVSFGYSLYAG) in only one amino acid residue (underlined). To investigate whether indeed this single residue difference between the human and mouse OSP(142-161) region underlies the inability of phOSP(142-161)-immunized DRB1*1501-Tg mice to develop EAE despite their ability to mount Th1/Th17 reactivity against the immunodominant hOSP(142-161) epitope, structural bioinformatics and functional studies were carried out. Using bioinformatics tools, overlapping 9-amino acid segments spanning h⌬OSP sequence were analyzed for their binding preference to DRB1*1501 and DQB1*0602 (see "Experimental Procedures"). This scan detected a DRB1*1501-binding nonameric core epitope, residues 144 -152, within the hOSP(142-161) region, and two DQB1*0602-binding nonameric core epitopes, one within the hOSP(55-80) region (residues 57-65) and another within the hOSP(142-161) region (residues 152-160). In contrast, a DRB1*1501-binding nonameric core epitope with preferred binding to DRB1*1501 was not detected within the mouse OSP(142-161) region (data not shown). The functional studies shown in Fig. 2 corroborated the computational analyses, which predicted that the hOSP(142-161) region contains two different core epitopes, one for DRB1*1501 and another for DQB1*0602. (The latter epitope appeared to be cryptic (compare Fig. 4C with Fig. 2C).) Notably, the predicted DRB1*1501-binding nonameric core epitope 144 -152 includes one of the very few residues that is not conserved between human and mouse OSP sequences, Thr-149 in human versus Ile-149 in mouse. To estimate the  type is the predicted hOSP(144 -152) nonameric core epitope, which differs from mOSP(144 -152) only in the OSP149 residue (red)). The semitransparent surface of DRB1*15:01 is colored by atom type: gray for carbon, red for oxygen, blue for nitrogen, and yellow for sulfur. Three peptides are shown: hMBP(85-99) (experimental) is shown with the carbon atoms in yellow; the carbon atoms in hOSP(144 -152) peptide are colored in cyan; and in the mOSP(144 -152) peptide, these atoms are colored in green. Nitrogen, oxygen, and sulfur atoms are colored in blue, red, and yellow, respectively. The side chains of residues that bind in pockets P1, P4, P6, and P9 are shown as ball-and-stick models. The side chains of residues p2 and p8 are omitted for clarity. The bottom panel presents a side view of the overlaid hMBP, hOSP, and mOSP peptides, showing the displacement of the C-terminal ends of the hMBP and the mOSP peptides out of pocket P9. Note that Val-152 of hOSP (cyan) binds deep in pocket P9. B, binding preferences in pocket P6 of DRB1*15:01. The ⌬G of binding for Asn, Thr, and Ile in pocket P6 was calculated with ANCHORSMAP.2, as described under "Experimental Procedures," showing that Ile is not a preferred residue for binding in this pocket and highlighting the difference between the human and mouse OSP epitopes. Note that pocket P6 (enlarged) is hydrophilic (red and blue surface), adequate for binding Asn and Thr but not the hydrophobic Ile. C, the OSP149 difference between human and mouse OSP(142-161) region impairs the binding affinity of mOSP(142-161) to DRB1*15:01. The peptide binding competition assay was performed as described in Fig. 3A, except that the amount of formed DRB1*15: 01⅐peptide complexes was monitored with horseradish peroxidase-labeled streptavidin and the substrate, 3Ј5,5Ј-tetrametylbenzidine. The enzymatic consumption of substrate was stopped with 1 M H 2 SO 4 and measured at 450 nm. The IC 50 value was at least 10 times higher for mOSP(142-161) than that for hOSP(142-161) (in three binding competition assays). where DRB1*1501 is bound to the MBP core epitope 89 VHF-FKNIVT 97 (Protein Data Bank codes 1BX2 and 1YMM) or to the microbial core epitope VHFISALHG (Protein Data Bank code 2WBJ), suggests a shift in the relative importance of the four major binding pockets in DRB1*1501, P1, P4, P6, and P9. P1 and P4 are dominant binding sites in the experimental structures, accommodating residues Val and Phe or Ile, respectively. In our model (Fig. 6A), these pockets accommodate OSP residues Cys-144 and Arg-147, respectively. The small side chain of OSP Cys-144 is adequate for the hydrophobic P1 pocket, but it makes fewer contacts than MBP Val-89. The hydrophobic part of the side chain of OSP Arg-147 interacts with the hydrophobic surface of pocket P4, and its positive end makes favorable contacts with the negative Asp 28␤ at the bottom of the pocket. Pocket P6 is polar, and in the experimental structures, it accommodates the MBP residue Asn-90 or Ala from the microbial peptide. Asn-90 interacts favorably with the hydrogen bonding network formed by DRB1*15:01 residues Arg-13␤, Asp-28␤, Tyr-30␤, Glu-11␣, and Gln-9␣ (31); however, its size and the binding directionality affect the conformation of the C-terminal part of the MBP peptide and hinder its binding. Thus, the hydrogen bonds between the side chain of DRB1*1501 Asn-69␣ and the peptide backbone, common to all MHC class II⅐peptide complexes, are exceptionally long in this structure (Ͼ3.5 Å), and MBP residue Val-99 does not bind inside pocket P9. In the DRB1*1501⅐microbial peptide complex, P6 accommodates the small side chain of Ala, allowing formation of hydrogen bonds between DRB1*1501 Asn-69␣ and the peptide. However, Ala cannot participate in the hydrogen bonding network, and the Gly residue in position p9 of the viral peptide makes only minor contacts with pocket P9. In our model, pocket P6 accommodates residue hOSP Thr-149, which can interact with the hydrogen bonding network. Moreover, Thr is smaller than Asn, and its mode of binding allows formation of standard length hydrogen bonds between the peptide backbone and DRB1* 1501 Asn-69␣ and binding of hOSP Val-152 inside pocket P9 (Fig. 6A). In contrast, the mouse OSP has a hydrophobic Ile residue in position 149, which cannot interact with the hydrogen bonding network in pocket 6 of DRB1*1501. Furthermore, Ile is larger than Thr, preventing formation of hydrogen bonds with DRB1*1501 Asn-69␣ and binding of Val-152 inside pocket P9. Computational mapping of anchoring spots (27) was used to determine the binding preferences of specific amino acids within the DRB1*1501 binding pockets ( Table 3). The ANCHORSMAP.2 results for Ile showed that it binds weakly in pocket P6, with ⌬G of Ϫ2.7 kcal/mol compared with ⌬G of Ϫ6.4 kcal/mol for Asn and Ϫ6.0 kcal/mol for Thr (Fig. 6B). In summary, whereas hMBP(85-99) binds strongly in pockets P1, P4, and P6, the human OSP(144 -152) core epitope binds strongly to pockets P4, P6, and P9, and the corresponding mouse core epitope can bind strongly only to P4 (Fig. 6B). Thus, the mutation in position OSP149 (p6 of the OSP(144 -152) core epitope) from Thr in humans to Ile in mice is likely to impede effective interaction with DRB1*1501.

Discussion
MS is a disease with a strong underlying genetic component and in which CNS demyelination presents as heterogeneous clinical and pathological phenotypes. The basis for the divergent patterns of clinical manifestation is unclear, but it may be a reflection of variable modes of CNS tissue damage associated with different pathways of pathogenic anti-CNS autoimmunity directed against diverse CNS myelin components (1). In view of the multiplicity of potential primary CNS target antigens in MS (including MBP, PLP, MOG, MOBP, and OSP) (2), a different major CNS target antigen/epitope against which pathogenic autoimmunity is primarily directed in different patients and/or different patients' HLA genotype could be a major factor underlying the heterogeneous disease phenotype. This highlights the value of characterizing the disease phenotype corre-lates of patterns of peptide/HLA class II antigenic presentation. This, however, requires definition of MS patient target antigens and major pathogenic epitope in the context of patients' HLA genotypes. Such characterization would be of significance also for devising immune effector-specific therapeutics for MS as well as for investigating disease etiology, in view of the possibility that cross-reactive immunity between viral/bacterial components and mimotopes of CNS antigen(s) may lead to MS in individuals with permissive genotypes (1,34,35).
The strong association of HLA-DR15 haplotype with susceptibility to MS has long been recognized. Genome-wide association and candidate gene studies have suggested that the greatest effects on MS risk conferred by the DR15 haplotype are driven by the HLA-DRB1*15:01 allele (5,6,41,42). However, linkage disequilibrium in the HLA-class II region makes it difficult without functional studies to distinguish whether the functionally relevant effect on MS derives only from the DRB1*15:01 or also from the neighboring genes in the HLA-DR15 region, from their combination, or from their epistatic interaction. Many genetic or immune functional studies implicate DRB1*15:01 as the primary risk factor in MS (5,41,43) while suggesting only a disease-modifying role for the DRB5*01:01 or DQB1*06:02 alleles (44,45). In this respect, the studies presented here showing the dominant role of DRB1* 15:01 in OSP autoimmunity in HLA-Tg mice, together with previously reported susceptibility of DRB1*1501-Tg mice to MBP-or MOG-induced EAE (7,8), are consistent with a primary contribution of the DRB1*15:01 allele to MS risk in HLA-DR15 ϩ individuals. However, our recent findings that the DQB1*06:02, rather than the DRB1*15:01 allele, determines susceptibility to EAE induced by MOBP (9) or PLP (10) offer a rationale for a disease-predisposing role also for DQB1*06:02. Thus, the reductionist transgenic models of MS suggest that for HLA-DR15 ϩ MS, the disease susceptibility associated with pathogenic autoimmunity against MBP, PLP, and OSP is determined by DRB1*15:01, whereas that against MOBP or against the PLP (one of the most prominent "MS-implicated" CNS target antigens) is dominated by DQA1*01:02;DQB1*06:02 alleles of the HLA-DR15 haplotype. These data suggest a more complex and differential functional role for HLA class II isotypes from the DR15 haplotype, depending on the primary target myelin antigen under attack. The finding that DRB1*15:01 may determine OSP-associated pathogenic autoimmunity not only as the classical DRA/DRB1 heterodimer, but also as a DQA1* 01:02;DRB1*15:01 mixed isotype heterodimer, introduces further complexity to the potential contribution of the DR15 haplotype alleles to MS risk. Although confined to transgenic models, these findings altogether strongly support the possibility that the functionally relevant genetic impact on MS can be driven by effects from the DRB1*15:01 locus, from its neighboring DQB1*06:02, and also from an interisotypic combination between alleles of DR and DQ loci, depending on the primary target myelin antigen against which the pathogenic autoimmunity is primarily directed.
There has previously been consideration of mixed heterodimers in murine EAE. In MBP-immunized (PL/JxSJL/J)F1 mice, MBP-reactive T-cells restricted to hybrid A␣ s A␤ U or to E␣ u E␤ s heterodimers were identified, in addition to MBP-reactive T-cells restricted by A␣ s A␤ s and E␣ u E␤ u dimers (46). Also, isotype-mismatched E␣ d A␤ d or E␣ k A␤ b hybrid MHC-class II heterodimers that were functional in stimulating antigen-reactive T-cells have been identified (47)(48)(49), but none of these studies reported A␣-E␤ (the DQA-DRB human orthologs) mixed isotype heterodimers. Human HLA class II isotype mixed pairing has also been previously reported (32,33,50,51). Lotteau et al. (32,33) identified the existence of mixed isotype DR␣-DQ␤ hybrid dimers in B-cell lines but could not identify any DQ␣-DR␤ mixed isotype dimers. Recently, mixed DQ␣-DR␤ isotype hybrid molecules (BoLA-DQA/DRB3) that were functional in presenting viral peptide to CD4 ϩ T-cells have been reported in cattle immunized with Anaplasma marginale vaccine (52).
Our demonstration of functional interaction between DQA1*01:02 and DRB1*15:01 in the humanized model of MS provides a rationale for a mechanism by which the DQA1*01:02 allele may contribute to MS risk also as a DQA1*01:02; DRB1*15:01 mixed isotype heterodimer functional in antigen presentation. However, although these observations highlight the principle, further studies will be needed to establish the extent to which this novel mixed isotype heterodimer contributes to the autoimmune repertoire and particularly the contribution of self-OSP epitope presentation by mixed isotype or by isotypic heterodimers to the autoimmune repertoire of human MS patients.
Overall, our transgenic model suggests that the potentially pathogenic autoimmunity against OSP in HLA-DR15 ϩ MS is likely to be focused on a single immunodominant epitope, OSP(142-161) (144 -152 core epitope), that is governed by the DRB1*15:01 allele, not only as the DRB1*15:01 isotypic heterodimer but also as a functional DQA1*01:02;DRB1*15:01 mixed isotype heterodimer. These studies, which demonstrate for the first time, to the best of our knowledge, a functional HLA-DQA1/DRB1 mixed isotype heterodimer, provide a rationale for the effect of DQA1*01:02 on MS through QA1*01: 02;DRB1*15:01 antigen presentation and highlight a potential contribution to MS also from interisotypic combination between HLA-class II monomers of DQ and DR alleles of the HLA-DR15 haplotype.