A Structural Framework for Deciphering the Link Between I-Ag7 and Autoimmune Diabetes

See allHide authors and affiliations

Science  21 Apr 2000:
Vol. 288, Issue 5465, pp. 505-511
DOI: 10.1126/science.288.5465.505


Susceptibility to murine and human insulin-dependent diabetes mellitus correlates strongly with major histocompatibility complex (MHC) class II I-A or HLA-DQ alleles that lack an aspartic acid at position β57. I-Ag7 lacks this aspartate and is the only class II allele expressed by the nonobese diabetic mouse. The crystal structure of I-Ag7 was determined at 2.6 angstrom resolution as a complex with a high-affinity peptide from the autoantigen glutamic acid decarboxylase (GAD) 65. I-Ag7 has a substantially wider peptide-binding groove around β57, which accounts for distinct peptide preferences compared with other MHC class II alleles. Loss of Aspβ57leads to an oxyanion hole in I-Ag7 that can be filled by peptide carboxyl residues or, perhaps, through interaction with the T cell receptor.

MHC genes have been linked with susceptibility in almost all autoimmune diseases (1). In the case of insulin-dependent diabetes mellitus (IDDM), a role for particular murine I-A alleles and their human homologs, HLA-DQ, has been inferred from numerous studies. Sequence analysis of these alleles has highlighted the importance of a key residue, β57; an Asp at position 57 of the class II β-chain is correlated with IDDM resistance, while neutral residues Ser, Ala, or Val are linked to disease susceptibility (2, 3).

The nonobese diabetic (NOD) mouse provides a model system for the study of IDDM. I-Ag7 is the only MHC class II molecule expressed in NOD mice and is strongly linked to disease susceptibility. I-Ag7 shares the same α chain as the non–IDDM-linked allele I-Ad(4), but contains a unique β chain that differs by 17 residues (2) and, hence, defines the diabetogenic characteristics of I-Ag7. Two polymorphic residues, Hisβ56 and Serβ57, play a pivotal role in linking I-Ag7 with IDDM; reintroduction by trangenesis of either Proβ56 (5) or Aspβ57 (6) into I-Ag7 reduces markedly the incidence of diabetes in NOD mice. The effector mechanisms linking variation at position β57 with IDDM susceptibility remain unclear; however, acquisition of an acidic residue at position 9 (P9) in the bound peptide may compensate for the lost charge on the class II molecule (7,8). On the other hand, most I-A/HLA-DQ molecules do not display a clear peptide-binding motif, such as described for MHC class I and for other class II I-E/HLA-DR molecules, suggesting greater promiscuity in peptide binding (7).

Several hypotheses suggest that I-Ag7 could bear unusual structural features that could translate at the T cell level into high autoreactivity, poor thymic tolerance, and inefficacious peripheral tolerance (9, 10). Alternatively, I-Ag7 may bind and present a unique set of peptides that are not usually presented by other MHC haplotypes (11).

We explored the structural link between I-Ag7 and autoimmunity by expressing I-Ag7 and I-Ad in Drosophila melanogaster cells as “functionally” empty molecules and as peptide-MHC (pMHC) complexes. Empty molecules were used for peptide-binding studies with synthetic peptides and phage-displayed peptide libraries, while single pMHC complexes were crystallized for x-ray structure analysis. Both MHC molecules were expressed by attaching complementary acidic and basic leucine zippers to the α- and the β-chain COOH-termini (12). A single pMHC was expressed by covalently tethering the peptide sequence to the NH2-terminus of the β chain, as described (12, 13). Removal of the leucine zipper by thrombin digestion did not destabilize the I-Ag7 dimer, as tested by gel filtration (14). The recombinant soluble I-Adand I-Ag7 molecules did not exhibit a tendency to aggregate or to bind peptides poorly; however, both were similarly SDS-unstable.

Our previously determined crystal structure of I-Adwith two different peptides (15) showed that relatively small hydrophobic side chains were generally used to insert into the P4, P6, and P9 pockets. We have confirmed the absence of a highly restricted motif for I-Ad by selecting a 12–amino acid random peptide phage library against empty I-Ad(16). The selection (panning) of the same library against empty I-Ag7 again did not reveal an obvious binding motif, but resulted in an increase in the frequency of acidic residues (Glu and Asp) within the last four residues of the peptide compared with the naı̈ve and the I-Ad–selected library. Binding of an 11–amino acid polyalanine peptide to both I-Ad and I-Ag7 (17) confirmed promiscuity in peptide binding, as this peptide displayed median inhibitory concentrations (IC50's) of 1.5 and 0.11 μM, respectively (Fig. 1A). However, substitution of AlaP9 by GluP9 resulted in a loss of binding for I-Ad, but a 10-fold increase for I-Ag7 (Fig. 1A). A nested series of 15–amino acid peptides (overlapping by three residues), encompassing all but the last six residues of GAD65, showed that of the 48 peptides, only one was a strong I-Ad binder [four others were medium binders (Fig. 1B)], whereas for I-Ag7, there were five strong, five medium, and three poor binders (Fig. 1B). These peptide data suggest increased promiscuity of I-Ag7 relative to I-Ad.

Figure 1

Comparison of peptide binding to I-Ad and I-Ag7. (A) Percentage of binding of biotinylated ROI peptide in the presence of increasing concentrations of test peptide (17). An 11–amino acid polyalanine (pAla) peptide binds to both MHC molecules in a similar manner to ROI. Substitution of AlaP9 by Glu (pAla AP9 →E) increases binding to I-Ag7, but abolishes binding to I-Ad. Both alleles bind to the positive control peptide ROI but not to MM67–79 (negative control). (B) At the level of a single protein, such as GAD65, I-Ag7 binds more peptide fragments than I-Ad. The binding of I-Ag7 and I-Ad, to a collection of 48 15–amino acid peptides encompassing all but the last six residues of GAD65, was tested by using an inhibition assay (17). The schematic indicates the approximate IC50 of peptide binding for each nested peptide and is based on an average of two independent experiments. The IC50 of the control peptide, ROI, is 0.5 μM. Peptides used for this analysis were synthesized in a 96-well format with FMOC (9-fluorenyl methoxycarbonyl) chemistry with a multiple pin synthesizer (Chiron Technologies, San Diego, CA). The purity of peptides exceeded 80%. The starting position of each 15–amino acid peptide, relative to GAD65, is displayed underneath each block.

To identify self peptides for I-Ag7, we constructed a 45- to 75-nucleotide random gene fragment phage display library from the cDNA of murine GAD65, an autoantigen that has been implicated in the early pathogenesis of diabetes (18). Selection of the library against empty I-Ag7 resulted in the multiple identification of two main GAD sequences, residues 207 to 217 and 105 to 119. GAD206-220 represents one of the major epitopes recognized by NOD mice, in vivo, after GAD65 immunization (19, 20). Both peptides bound strongly to I-Ag7 in an in vitro binding assay, with GAD207-220 (YEIAPVFVLLEYVT) being the better binder (IC50 = 0.08 μM). Binding of GAD207-220 by I-Ag7 was entirely lost when Glu217(equivalent to P9) was replaced by glutamine.

The crystal structure of I-Ag7 covalently linked to residues 207 to 220 of murine GAD65 was determined at 2.6 Å resolution (Table 1). The electron density identified residues 209 to 217 as the P1 to P9 core of the bound peptide (Table 1 and Fig. 2A); all 14 residues of the GAD peptide had interpretable main-chain density and occupied the P−2 to P12 register [Web table 1 (21)]. The P1, P4, and P6 pockets contained the hydrophobic residues Ile-209, Val-212, and Val-214, respectively, whereas the P9 pocket was occupied by Glu-217. A Cα superimposition (22) of I-Ag7 onto I-Ad(15, 23) and I-Ak(24) gave low root mean square (rms) deviations of 0.26 and 0.31 Å, respectively, as expected from their high sequence identity. The antiparallel β sheet that forms the floor of the peptide groove and the long α helix of the α1 domain overlayed well in all three structures (Fig. 2B). However, the α-helical segments of the β1 domain diverged progressively from the H2b to the H2a and H1 helical segments; the H1 segment containing residue β57 of the I-Ag7 helix exhibited a noticeable displacement away from the binding groove (25) that was partly caused by the polymorphic residues Hisβ9 and Tyrβ61 (Fig. 2, B and C). Tyrβ61 appears to play the major role here as it is unique to I-Ag7 (Trp in I-Ad and I-Ak), whereas Hisβ9 is also found in I-Ak [Val in I-Ad(26)].

Figure 2

Structure analysis of I-Ag7–GAD207-220 and comparison with other murine MHC class II alleles. (A) Cross-validated σA-weighted 2F oF c electron density of the GAD207-220 peptide. The map, contoured at 1σ, represents the final refined structure. Electron density for the peptide main-chain atoms is present for residues P−2 to P12; however, convincing side-chain density for Tyr-207 (P−2), Tyr-218 (P10), and Thr-220 (P12) was absent (59). The side chains for these residues were therefore truncated back to Cβ. Final coordinates are overlaid in the map. (B) Overlay of the murine MHC class II molecules I-Ag7 (red), I-Ad (yellow, green), and I-Ak(blue). The Cα trace (22) shows only the α1 and β1 domain of the respective MHC II molecule. The two I-Ad pMHC complexes are colored green (peptide OVA323-339; PDB code 1IAO) and yellow (peptide HA126-138; PDB code 2IAD), respectively. The peptide COOH- and NH2-termini are respectively labeled N- and C-. The peptide core (P1 to P9) overlays well in all four structures. (C) Key residues that differ in I-Ag7around the P9 pocket, their effect on peptide presentation, and the position of H1 and H2a segments of the β1-chain α helices. The polymorphic residues Hisβ56, Serβ57, and Tyrβ61 (H1) are found in I-Ag7, but not in other I-A alleles. The presence of Serβ57, in place of the highly conserved Aspβ57, substantially alters the specificity of the P9 pocket. The orientation of the Tyrβ61 side chain plays a significant role in the displacement of the H1 helical segment; the side-chain phenyl ring is almost perpendicular to that observed for the indole of the corresponding Trp residue in I-Ad,k. Displacement of the β1 H1 helical segment allows the side chain of Tyrβ61 to be accommodated and dramatically alters the shape of the P9 pocket. (D) Unique character of the I-Ag7 P9 pocket. The presence of Serβ57in I-Ag7, coupled with displacement of the H1 helix away from the peptide groove, increases the lateral freedom for the P9 peptide side chain, as clearly seen from a comparison of I-Ag7 (red) with I-Ad (cyan), I-Ak (green), I-Ek (gold), HLA-DR1 (purple and blue), and HLA-DR3 (yellow). The P9 side chains for all except I-Ag7 point downward in a classical manner. Three other MHC class II molecules are not shown: HLA-DR1 [PDB code 1SEB (60)] was built with a polyalanine peptide. HLA-DR2 is unusual in that the P9 side chain points toward the side of the peptide groove rather than downward in the classical manner. This orientation occurs as a result of an elevated COOH-terminus of the peptide and suboptimal occupancy of the P9 pocket, because the pocket itself is identical to that found in HLA-DR1 (61). HLA-DR4 [PDB code 2SEB (62)] has a glycine at P9. Stereo versions of Fig. 2, B and C, are available from Science Online as Web figure 1 (21).

Table 1

Data collection and refinement statistics. For crystallization, acidic and basic leucine zippers were removed from I-Ag7 by thrombin digestion, and the residual COOH-terminal spacer sequences were digested with carboxypeptidase B. Covalent I-Ag7–peptide complexes of GAD207-220 and GAD221-235 were concentrated to 10 mg ml−1 in 20 mM Hepes, 25 mM NaCl (pH 7.5) and screened for crystallization. Small crystals were obtained for I-Ag7–GAD207-220 and enlarged by streak seeding (42). Crystals were then grown in 16 to 18% polyethylene glycol 4000, 0.2 M LiCl (pH 6.6), 1% 2-methyl-2,4-pentanediol and flash-cooled to 96 K with glycerol as a cryoprotectant (through a sequential soak in 1, 5, 10, and 15% glycerol). A single crystal was used to collect data at beamline 7-1 of the SSRL. Data were integrated in space group C2221(a = 95.1 Å, b = 110.1 Å,c = 96.1 Å) with DENZO (version 1.9.1) and reduced with SCALEPACK [version 1.9.0 (43)]. Crystal mosaicity was refined in batches and ranged from 1.0 to 1.3°. A single I-Ag7 was assumed to be present in the asymmetric unit based on the calculated Matthew's coefficient [V m ∼ 2.9 Å3/Dalton, ∼57.0% solvent (44)]. Molecular replacement was carried out with CCP4's version of AMoRE (45,46) and the MHC class II search model I-Ak [PDB code 1iak (24)], in which the peptide, carbohydrate, and waters had been omitted. Clear rotation (one peak greater than 50% maximum peak height, resolution 8.0 to 3.5 Å, sphere radius 22.5 Å) and translation solutions (correct solution was 6.7σ above the next highest peak, resolution 8.0 to 3.5 Å) were found. After rigid body fitting within AMoRE (resolution 20.0 to 2.7 Å), the correlation coefficient was 68.7% and theR cryst was 41.5%. Further rigid body refinement was carried out with CNS [version 0.9 (47)] by using three domains α11, α2, and β2. TheB value for all atoms was then reset to 20 A2. Cross-validated σA-weighted 2F oF c andF oF celectron density maps (48) were used throughout in order to rebuild and were viewed with the graphics program O (49). Density was observed for the peptide at this stage, but not refined until later. Because density for the helix region Gly β54 to Gln β64 was ambiguous, its occupancy was set to zero for the next refinement cycle (positional and torsion angle dynamics refinement, resolution 10.0 to 2.7 Å, with a starting temperature for the dynamics determined with the slow cool protocol of 4000 K). Density for the helix was noticeably improved and was gradually rebuilt over the next few cycles. R free (50) (∼10% of the reflections) was used from the outset to monitor the course of the refinement. Waters were picked toward the end of the refinement, with an F oF c map and a 3.5 to 4.0σ cutoff. Waters were rejected after refinement if they failed to appear in both 2F oF c andF oF cmaps, were not within hydrogen bonding distance of another protein atom or water molecule, or if their B value exceeded 50 A2. All data from 40 to 2.6 Å were used in the last few refinement cycles, and an anisotropic B correction was applied along with a bulk solvent correction to give the final structure (51).

View this table:

Most murine and human MHC II molecules contain a highly conserved interdomain salt bridge between residues Argα76 and Aspβ57. Absence of this salt bridge in I-Ag7, owing to the presence of Serβ57, has been proposed to cause interdomain instability and loose fitting of the peptide COOH-terminus (27). Our results do not support this view, because the COOH-terminal residues of the peptide are well defined in the electron density until P12 (28). There is some rearrangement of the hydrogen-bonding network, involving COOH-terminal peptide residues P7 to P9, as a result of MHC polymorphic differences in this region; however, the total number of main-chain hydrogen bonds remains unchanged compared with I-Ad[Fig. 3, A to D (29)].

Figure 3

(A to D) Composite highlighting the differences between I-Ag7 and other murine class II alleles. (Left) Changes in hydrogen bonding between I-Ag7and the peptide P7 to P9 positions compared with other alleles. Hydrogen bonds (dotted) were calculated with HBPLUS (63). In I-Ag7, the presence of the polymorphic residues Serβ57, Tyrβ61, and Tyrβ66results in a substantial alteration of the hydrogen bonding pattern to the peptide. Hydrogen bonds mediated by solvent are not shown. The backbone of the β chain is colored pink, the α chain is cyan, and the peptide is dark blue. The MHC side chains are colored yellow-green with oxygens in red and nitrogens in blue. (Right) Altered surface shape and electrostatics of the I-Ag7 binding groove compared with other murine class II alleles. Molecular surfaces were calculated with a probe radius of 1.4 Å (64). Only the surface around the P9 pocket is shown. Electrostatics were calculated with the Delphi module within Insight II (41). Formal charges were assigned to the protein coordinates. Positive charge is contoured blue, while negative charge is red (−5 to +5 kT/e). For each allele, the GAD207 peptide (green) is overlaid in each to allow an equivalent spatial comparison. The “carboxyl” cavity, occupied by the side chain of Glu-217, next to the P9 pocket of I-Ag7 is readily apparent; this oxyanion hole is absent in the other I-A and I-E alleles, as evidenced by the buried terminus of Glu-217. The P9 pocket of I-Ag7 is clearly positively charged unlike the other alleles, which range from neutral to strongly negative. Stereo versions of both panels are available fromScience Online as Web figures 2 and 3 (21).

The absence of the salt bridge between residues Argα76and Serβ57 does not in itself significantly alter the location of the Argα76 side chain (Fig. 2C). However, the Argα76 side chain shifts slightly to optimize hydrogen bonds with the carbonyl oxygen of Leuβ53 and the side-chain hydroxyl of Serβ57. Additional compensation for the lost Argα76-Aspβ57 salt bridge is provided by formation of a new electrostatic interaction between the P9 carboxyl of Glu-217 and Argα76 (30). The Glu-217 carboxyl also forms a hydrogen bond with the substituted Serβ57 [Web table 1 (21)]. Thus, Glu-217 appears to accomplish two roles: (i) It complements the charge of Argα76, and (ii) owing to movement of the β1-chain H1 helix, it fills a new cavity that arises between I-Ag7 and the peptide as a result of the substitutions at β56 and β57 (Fig. 3A).

The displacement of residue β57 away from the peptide-binding groove results in a shallow P9 pocket that is substantially wider than deep and allows the P9 side chain to have a greater degree of lateral freedom than in other class II molecules (Fig. 2D). Thus, two orientations are possible for the P9 side chain in I-Ag7; one points downward into the peptide groove and the other points sideways. The shallowness of the P9 pocket suggests that only small side chains (Gly, Ala, and possibly Ser) can be accommodated in a downward orientation. The unique sideways orientation (Fig. 2D) could accommodate medium to large side chains, although the positively charged environment would favor negatively charged residues. Modeling of a Glu at P9 into either I-Ad, I-Ak, or I-Ek(with the GAD207–220 peptide) (Fig. 3, B to D, right panels) shows that, in contrast to I-Ag7, an acidic residue would be strongly disfavored because of a combination of poor steric and poor electrostatic complementarity.

The class II–associated invariant chain CLIP peptide dissociates rapidly from I-Ag7 primarily because of a poor fit of Met P9; substitution by an Asp greatly enhanced the stability of this complex (31). The I-Ag7 crystal structure suggests that the side chain of Met would have to point sideways, but it would not be favored in the polar P9 side pocket. For I-Ag7 peptides that do not have a negatively charged residue at P9, a nearby negatively charged residue, such as at P11, might substitute. Modeling of a peptide with a P11 Glu (or Asp) shows that the side chain could be positioned to provide electrostatic complementarity with Argα76 after some main-chain rearrangements to P10 and P11.

Hisβ56 , the other key polymorphic residue, is positioned adjacent to the peptide groove and does not interact with the peptide directly. Its presence (and to a lesser extent that of Serβ57) leads to a significant increase (∼44%) in the exposed molecular surface area of the Argα76 side chain relative to I-Ad, I-Ak, and I-Ek [Fig. 2C (32)]; thus, these two substitutions in I-Ag7 make the P9 cavity more accessible and at the same time more attractive to oxyanions, such as carboxylates [Fig. 3, A to D (33)].

The I-Ag7 pockets P1, P4, and P6 are approximately the same size as those found in I-Ad [Fig. 4, A and B (15)]. The P1 pocket of I-Ag7 is the largest and is only partially occupied by an Ile and three buried water molecules (Fig. 4A). The surface of the P1 pocket is formed by hydrophilic residues Hisα24, Asnβ82, Thrβ86, and Gluβ87 and hydrophobic residues Tyrα8, Leuα31, Pheα32, Trpα43, Ileα52, and Pheα54. This diverse set of residues could permit a broad specificity in the type and size of the P1 residue. The P4 pocket is small, hydrophobic in character, and only partially occupied by a Val; the additional space could accommodate larger hydrophobic residues, such as leucine or isoleucine. The P6 pocket, although formed mainly by hydrophobic residues, does have some polar character due to Gluα66 and the backbone carbonyl of Asnα62. This dual character could allow both small hydrophobic or small hydrophilic residues. Thus, on the basis of peptide-binding studies and the crystal structure, we propose the following optimal peptide-binding motif for I-Ag7: P1, degenerate; P4, small to medium hydrophobic residue; P6, small to medium hydrophobic or hydrophilic residue; and P9, negatively charged residue or Gly, Ala, or Ser.

Figure 4

Comparison of the peptide specificity pockets of (A) I-Ag7 and (B) I-Ad. Molecular surfaces were calculated with a probe radius of 1.4 Å (64). The peptides are in spacefilling (CPK) representation (backbone, yellow; side chain, dark blue). Waters, rendered as spheres of one-half van der Waals (VDW) radii, are colored red. The P1, P4, and P6 pockets of I-Ag7 and I-Ad (15) are about the same size, whereas the P9 pocket of I-Ad is larger but lacks the additional side cavity. I-Ag7 appears to maintain some conformational flexibility in defining the shape of the P1 pocket; repositioning of Leuα31 and Trpα43, compared with the analogous residues in I-Ad, allows the P1 pocket to be occupied by the Ile-209 side chain (and three water molecules).

Most of the higher affinity peptides that bind to I-Ag7 can be aligned with this motif (Table 2). The CLIP86-100 and GAD105-119 sequences were fitted by allowing a nonoptimal residue at P9. The inability to align a few peptides, such as human GAD247-266 (34), with this motif is to be expected given its relatively low stringency. Our analysis of I-Ag7–binding peptides from the GAD65 15–amino acid library further supports this structural motif, with only a few exceptions. Of the 13 peptides that bound to I-Ag7, 8 could clearly be aligned; a further 3 required a Phe or Tyr to be positioned in either P4 or P6. Also, the average IC50 of the peptides with a negatively charged residue at P9 is indeed lower (i.e., three good binders and one medium binder) than that of the other peptides.

Table 2

Alignment of peptides with the canonical P1, P4, P6, and P9 pockets of the I-Ag7 peptide-binding groove. Alignment of most of these peptides, within the I-Ag7peptide-binding groove, can be achieved with the optimal motif; the average affinity of these peptides should presumably be higher compared with peptides that bind using nonoptimal residues. Some of the peptides can adopt different registers, as has been observed for peptides that bind to I-Ad (15). Potential anchor residues at P1, P4, P6, and P9 are indicated in bold. Acidic P9 residues are colored red, whereas Gly/Ala or Ser at P9 are colored blue. Abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

View this table:

Our functional and structural analyses therefore do not support the hypothesis of a greater instability for I-Ag7. First, our recombinant I-Ag7 molecules do not exhibit a tendency to dissociate into monomers or to bind peptide to any lesser extent than I-Ad. Second, the crystal structure of I-Ag7 is remarkably similar to that of other class II molecules, particularly I-Ad (15) and I-Ak (24). The peptide promiscuity of I-Ag7 (and I-Ad) arises from a structural motif that is skewed toward selection of small alternating residues, be they hydrophobic or hydrophilic; this motif is commonly found in globular proteins. The motif's low stringency is further accentuated by a redundant P1 pocket (35,36) and, for I-Ag7, by a P9 pocket with broader specificity. Promiscuous binding by a class II MHC, like I-Ag7, can be a double-edged sword because it potentially allows the host to respond more efficiently to antigen challenge, and yet simultaneously increases the susceptibility toward developing an autoimmune response. The actual response will depend, to a large extent, on the host's genetic background; this is clearly seen for the NOD and the Biozzi (H-2dq1) AB/H mouse (37), which both express only the class II MHC I-Ag7. NOD mice develop IDDM, whereas Biozzi mice develop high responses to all T cell–dependent antigens (38).

Structural differences between I-Ag7 and the non–IDDM-associated I-A alleles I-Ad and I-Ak center around the β1 α-helix and the P9 pocket. The P9 pocket is probably directly linked to the biology and pathology of I-Ag7 because it affects peptide repertoire selection and, potentially, the nature of the interaction with the T cell receptor. The ability of I-Ag7, compared with other I-A alleles, to select for a novel subset of peptides has been demonstrated by a comparison of peptide binding to an I-Ag7 where the β56 and β57 positions are changed back to the I-Ad sequence (Proβ56 and Aspβ57, I-Ag7.PD); I-Ag7 preferentially binds peptides with an acidic P9 residue, unlike its mutated counterpart I-Ag7.PD (19).

I-Ag7 peptide complexes that have small nonacidic P9 residues, such as Gly/Ala, will leave an unfilled “oxyanion” hole. These complexes could conceivably bind a set of complementary T cell receptors that can contribute a negatively charged residue to the oxyanion cavity and provide additional stability. In support of this hypothesis, the recently determined I-Ag7- GAD221-235 structure (39) has a Gly at the peptide P9 position; yet the P9 cavity is still occupied, but by a glutamate side chain from a symmetry-related molecule. Alternatively, other acidic residues present in the COOH-terminal extensions (P10 to P12) of the peptide itself could fill the oxyanion hole and hence lead to kinked conformations of the peptide that would appear very different to the T cell receptor.

Thus, the crystal structure of I-Ag7 now provides a structural framework for unraveling the link between a class II molecule and a tissue-specific autoimmune process. The high number of peripheral autoreactive T cells present in NOD mice may be linked with poor central tolerance or may be the consequence of promiscuous or altered peptide binding of a class II MHC, in this case I-Ag7. In terms of therapeutic intervention at the level of MHC, the idea of “blocking peptides” has long been envisaged and tested exhaustively. Previous failures have probably arisen from their short half-lives, low bioavailability, and low specific binding. The crystal structure of I-Ag7should aid in the future design of novel high-affinity ligands for I-Ag7.

  • * These authors contributed equally to this work.

  • Present address: Department of Medicine 0613-C, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92083–0613, USA.

  • Present address: Stanford University School of Medicine, Departments of Microbiology and Immunology and Structural Biology, Fairchild Sciences Building D-319, 299 Campus Drive, Stanford, CA 94305, USA.

  • § Present address: Abgenix Inc., 7601 Dumbarton Circle, Fremont, CA 94555, USA.

  • || To whom correspondence should be addressed. E-mail: wilson{at}; lteyton{at}


Stay Connected to Science

Navigate This Article