COOH-Terminal Clustering of Autoantibody and T-Cell Determinants on the Structure of GAD65 Provide Insights Into the Molecular Basis of Autoreactivity

The panel of 11 human mAbs to GAD65 included M2, M3, M4, M5, M6, DPA, DPB, DPC, and DPD from two patients with type 1 diabetes (6,7) and b96.11 and b78 from a nondiabetic patient with autoimmune polyendocrine syndrome type 2 (8). Each of these patients shared the HLA DRB1*0401 allele associated with a high risk for diabetes. None of the mAbs reacts with GAD67. Data were available from published studies (4,9–11) on mutagenesis on the putative location of epitopes for these mAbs, as shown in Table 1.

Details of the purification, crystallization, and structural determination of human GAD65 and GAD67 have been published previously (5). We used for crystallization an NH₂-terminally truncated form of each isoform (hereinafter referred to as GAD67 and GAD65) that lacked the first 89 and 83 residues, respectively; the NH₂-terminal truncation facilitated purification because this region is hydrophobic and highly susceptible to proteolysis and did not affect enzymatic properties (5) nor reactivity with specific antisera (12–14). The proteins were expressed in Saccharomyces cerevisiae as fusions to a COOH-terminal hexahistidine tag and purified from the cell lysate by immobilized metal affinity chromatography and size exclusion chromatography in the presence of glutamate and PLP.

Analysis of structural differences between GAD65 and GAD67 and location of mutations shown to affect mAb binding were performed with the program Pymol (15). For each mAb to GAD65, the locations of mutations that either caused reduced reactivity by RIP or did not affect reactivity were mapped on the structure of dimeric GAD65; together, these data were used to delineate likely epitope regions. Structural alignments were performed with the program MUSTANG (16).

Tests for colocalization of epitopes by cross-competition were performed using rFab from the mAbs b96.11, b78, DPA, DPC, DPD, M2, M3, and M4, prepared as previously described (17). The mAbs were tested at half-maximal binding concentrations, and results shown are from at least two separate experiments. Results were expressed as percent inhibition of full-length ³⁵S-labeled GAD65 bound in the presence of rFab where percent inhibition = 100 − (counts per minute in the presence of mAb/counts per minute in the absence of mAb) × 100. Samples were analyzed in triplicate; the average intra-assay coefficient of variation was 5%, and the maximum value was 9%. A negative control rFab D1.3 specific to an irrelevant antigen, hen egg white lysozyme, was included at 5 μg/ml in each assay, and the cutoff for specifying blocking was conservatively set at >20% inhibition.

The 2.3-Å crystal structures of GAD67 and GAD65 have been reported (5). Both isoforms adopt the same fold and formed obligate dimers from two monomeric units, each comprising the NH₂-terminal, PLP-binding, and COOH-terminal domains (Fig. 1A and B). For each monomer, the NH₂-terminal domain contained two parallel α-helices, helices 1 and 2, that were packed against the NH₂-terminal and PLP-binding domains of the partner monomer. The PLP-binding domain contained nine α-helices, 3–11, that were surrounded by a seven-stranded mainly parallel β-sheet that adopted the type I PLP-dependent transferase-like fold. The COOH-terminal domain contained three α-helices, 13–15, and a short, buried, four-stranded anti-parallel β-sheet, s3C. This relative orientation of monomers within the dimer for both GAD65 and GAD67 was such that the domains from monomer A formed a contiguous surface patch with the domains of the partner monomer B (Fig. 1A). In addition, in both cases, the two active sites of the enzyme, located in the center of the PLP domain at the dimer interface, contained the cofactor PLP and the product γ-aminobutyric acid (GABA) (Fig. 1A and B), indicative of the active (holo-) state of the crystallized enzyme. Overall, of 41 nonconservative amino acid differences between the two isoforms, 32 were surface-exposed, and 9 were buried in the region structurally characterized. Differing surface-exposed residues between GAD67 and GAD65 were distributed over the entire structure of the molecule, with no obvious clustering apparent in any region (Fig. 1C).

There are two key differences between the two structures that have enzymatic and potentially antigenic consequences. First, for GAD67, each of the two active sites was substantially covered by a well-ordered catalytic loop comprising residues 432–443 contributed in trans from the partner monomer. Notably, for GAD65, this same catalytic loop, residues 422–433, was highly flexible and thus not visible in the electron density maps. This difference offers a structural rationalization for the contrasting enzymatic characteristics of the GAD isoforms and insight into the requirement of two GAD isoforms in mammals (5). Second, the two molecules differ markedly in their flexibility, with GAD65 being the more flexible, as indicated by the atomic temperature factors (B factors) (Fig. 1D). This difference is particularly pronounced in the COOH-terminal domain, in a surface-exposed loop (residues 518–520) that was too flexible to be resolved in the crystal structure of GAD65, in marked contrast to the well-defined structure of the corresponding sequence for GAD67.

The known sites of point mutations or GAD67/65 sequence exchanges (chimeric molecules) that affected the binding of any of the panel of mAbs were mapped onto the structure of GAD65. The location of epitopes or epitope regions in the COOH-terminal domain, PLP domain, and the NH₂-terminal domain are described below.

Mutated residues that substantially decreased reactivity with five mAbs (M2, M3, M5, DPA, and b78) were heavily clustered in the region of GAD65 that surrounded the COOH-terminal flexible loop, residues 518–520, which differs markedly from the equivalent region of GAD67. Contact residues for the binding of these five mAbs were distributed over all four faces of the two autoantigenic COOH-terminal domains of dimeric GAD65, denoted as the α, β, γ, and δ faces (Fig. 2A and B), but two clusters of COOH-terminal contact residues (ctc1 and ctc2) could be clearly discerned. These were localized predominantly to opposing faces, α and δ, of the COOH-terminal domain, and could be represented by the epitope regions of two exemplary mAbs, M5 and M3 (Fig. 2A and B). Mutated residues in ctc1 that affected binding of mAbs M2, M5, and b78 were located in helix 14 (residues 522–540) and the adjacent COOH-terminal flexible loop in the α and β faces of the COOH-terminus of one GAD65 monomer and close to the catalytic loop of the partner monomer, and mutated residues that affected binding of M3 and DPA were localized to helices 13 and 15. (Fig. 2C and D). Notably, neither point mutation of residues nor insertions of GAD67 sequences on the α face had any effect on the binding of mAbs reactive with ctc2 on the opposite δ face of the COOH-terminal region and vice versa (Table 1).

The reactivity of M5, the prototypic mAb for ctc1, depended strongly on a nonconserved surface-exposed residue V₅₃₂ located in the COOH-terminal end of helix 14 of GAD65 (Fig. 2C), and the binding of M5 required an additional five residues, with two located in the NH₂- terminus of helix 14 (S₅₂₄ and S₅₂₇), two flanking the COOH-terminal unstructured loop (E₅₁₇ and E₅₂₁), and one within this loop (E₅₂₀) (Fig. 2C). Combinations of these residues located in the most structural divergent region between GAD65 and GAD67 (Fig. 2C) affected the binding of the other four mAbs localized to ctc1 (Table 1). The important residues in ctc2 that engage mAbs M3 and DPA, distributed particularly on the δ face, lie close to the PLP domain, but this epitope region is remote from the catalytic loop. Reactivity of mAbs M3 and DPA, as judged by mutations, depended on residues of GAD65 located in helices 13 (N₄₈₃) and 15 (H₅₆₈), noting that mutation of N₄₈₃ ablated reactivity only with M3, whereas mutation of H₅₆₈ ablated reactivity with both mAbs (Fig. 2D).

Epitopes on GAD65 for mAbs M4, M6, b96.11, and DPC have been mapped to the PLP domain by use of GAD65/67 sequence exchanges (Fig. 2E and F), but information on the exact location of contact sites is limited. However, point mutations of GAD65 did decrease reactivity for each of these mAbs. Epitope regions for both M4 and b96.11 were localized within residues 308–365, but the single point mutation shown to affect binding of M4 to GAD65 was that of K₃₅₈, a residue that is surface exposed, but this of itself is insufficient data to define an epitope. For mAb b96.11, no single point mutation ablated reactivity, but several mutations that substantially reduced reactivity, including the PLP domain sequence ₃₀₅DER₃₀₇, were adjacent to the COOH-terminal epitope region ctc2. Thus, contact residues for b96.11 would lie at the junction of the PLP and COOH-terminal domains in the δ/β face of ctc2 (Fig. 2D). For mAb M6, the epitope region was localized within residues 242–282; binding to GAD65 was not affected by mutation of GAD65 E₂₆₄T, but mutation of the equivalent residue of GAD67 T₂₇₃E restored 50% of the reactivity with GAD65 (4). Notably, E₂₆₄ lies on the interface of the dimer (Fig. 2E), within the previously identified ₂₆₀PEVKEK₂₆₅ region of identity with the P2-C protein of Coxsackie B virus (18), and close to residues 271–285, a major T-cell epitope restricted by HLA DR*0401 (19) (see below).

For mAb DPC, two distinct epitope regions were revealed by effects of sequence exchange: one involving residues 366–413 and one involving residues 134–242 overlapping the junction of the PLP and NH₂-terminal domains. Within the crystal structure of GAD65, residues 366–413 in monomer B are positioned directly adjacent to the NH₂-terminal residues 134–242 in the partner monomer A (Fig. 2E); both sequences contain structural and amino acid differences between GAD65 and GAD67. Also, for mAb DPC, an epitope site was defined by mutagenesis of residues P₂₃₁ and S₂₃₄, which are located on a well-defined loop in the PLP domain that differs structurally from the equivalent region of GAD67 (Fig. 2E and F). No mutagenesis data exist to identify the residues responsible for binding of mAb DPC within the NH₂-terminal domain.

Epitope regions for two mAbs, DPD and DPB, have been mapped exclusively to the NH₂-terminal domain but only by use of GAD65/67 sequence exchanges. For mAb DPD, the epitope region is within residues 96–173, which lie precisely adjacent to the COOH-terminal end of helix 14 of the same monomer and include a structural difference between GAD65 and GAD67 created by a deletion of three amino acids in GAD65 (143_Q—E144 vs. 149_EGMEG153) (Fig. 1C). For mAb DPB, the epitope region is located within residues 1–96, which are lacking in the crystal structure of GAD65.

Of rFab derived from mAbs to GAD65, there are three (M3, DPA, and b78) derived from mAbs that reacted with epitope regions at the COOH-terminus, three (b96.11, DPC, and M4) with the PLP domain, and two (DPC and DPD) with the NH₂-terminal domain. When these rFab were tested for their capacity to block the reactivity by RIP between GAD65 and each of the mAbs (Table 2), each rFab (with one exception) strongly blocked the binding with GAD65 of the corresponding mAb. In general, the mAbs that reacted with one or the other of the two distinct COOH-terminal epitope clusters, ctc1 and ctc2, differed in susceptibility to cross-blocking by the rFab. For example, rFab from mAbs M3 and DPA (ctc2 epitopes, δ-face) of the COOH-terminus cross-blocked, but rFab from mAb b78 (ctc1 epitope, α-face) did not. Also, the binding of mAb M3 was blocked by rFab from mAb b96.11, consistent with the localization of the epitope for mAb b96.11 in the β and δ faces close to ctc2, although rFab from mAb M3, which localizes close to mAb b96.11, did not block binding to GAD65 of mAb b96.11. No other rFab tested blocked the binding of either M3 or DPA.

Cross-competition studies provided important evidence that other epitopes that could not be mapped to the COOH-terminal domain by mutagenesis did in fact lie close to either ctc1 or ctc2. Thus two mAbs that reacted with NH₂-terminal epitopes, DPD and DPC, gave reactivity with GAD65 that was strongly blocked by rFab from mAb b78, which engages epitopes in ctc1. With the exception of rFab from DPD, there was strong mutual cross-competition among all three mAbs, suggesting that epitopes for each overlapped, and thus each could be considered as a component of ctc1 (Fig. 2E and F). Also consistent with the location of contact residues, the two mAbs for which epitopes mapped to the PLP domain, DPC and b96.11, were not cross-competitive, and the cross-competition that was demonstrable between mAbs DPC and b78 (see above) indicates that the epitope for DPC lies close to ctc1 epitopes, whereas the epitope for b96.11 lies close to ctc2 epitopes. The single mAb tested by cross-competition for which an epitope could not be localized to either ctc1 or ctc2 was M4, which maps to the PLP domain; rFab from M4 blocked the binding to GAD65 of mAbs DPC, DPD and b78 (ctc1) and also of mAb b96.11 (ctc2). The single amino acid K₃₅₈, which has been shown by mutagenesis to be required for the binding to GAD65 of M4, lies on the face of the GAD65 dimer between ctc1 and ctc2 (Fig. 2E).

To examine human T-cell epitopes in the context of the crystal structure of GAD65, we selected five particular immunodominant T-cell epitopes of GAD65 identified in several studies (19–21), designated T1-T5, each restricted by HLA DRB1*0401 (Table 3). The COOH-terminal domain contained three epitopes, T3 (residues 481–495), T4 (residues 511–525), and T5 (residues 551–585); the PLP domain contained one, T2 (residues 270–286); and the NH₂-terminal domain contained one, T1 (residues 116–130). Positioning of these on the crystal structure of dimeric GAD65 showed that T1, T3, T4, and T5 were closely associated (Fig. 3). T5 comprised part of helix 15 and the buried COOH-terminal β-sheet s3C, with its surface-exposed component located between two other epitopes, T1 and T3, indicative of strong interactions with both regions (Fig. 3B and D). T1 corresponds to helix 1 in the NH₂-terminus, residues 116–130. T3 localizes at the COOH-terminal end of helix 13 and its adjacent loop and also is spatially contiguous with the T4 epitope region (residues 511–525) that corresponds to the highly flexible COOH-terminal loop for which structure is lacking. Taken together, there are four T-cell epitopes that form a contiguous linear patch on the surface of the GAD dimer located within the two major B-cell epitope regions ctc1 and ctc2 and another, T2, with a different location (residues 271–285), close to the ₂₆₀PEVKEK₂₆₅ sequence of similarity with the Coxsackie B virus protein 2C (18) and mainly buried within the PLP domain.

The analysis of crystal structures of both NH₂-terminally truncated GAD65 and GAD67 has provided a molecular basis for epitope locations, assisted by data derived from several sources: binding assays using a unique panel of human mAb to GAD65, point and sequence exchange mutagenesis of GAD65, and cross-competition experiments with GAD65-specific monoclonal rFab. Notably, we identified a major region of antigenic activity for GAD65 predominantly centered on a flexible and structurally undefined region in the COOH-terminal domain that differs strikingly between GAD65 and GAD67 and including also conformationally contiguous parts of the PLP and NH₂-terminal domains. Within this region, two independent clusters of B-cell epitopes, ctc1 and ctc2, were positioned on opposing faces of the COOH-terminal domain as judged by reactivity of mAbs to GAD65 with particular mutants of GAD65, and blocking studies using human rFab. For T-cells, four of five previously recognized DRB1*0401-restricted epitopes on GAD65 formed a contiguous surface-exposed patch between the two B-cell epitope clusters.

The division of the COOH-terminal epitope region of GAD65 into two epitope clusters, ctc1and ctc2, on opposing faces (α and δ) of the molecule was based on cross-reactivity of mAbs, although boundaries between the two regions were not entirely distinct, because rFab from mAb b78, which engaged residues in ctc1, caused partial inhibition (63%) of the binding to GAD65 of mAb b96.11, which engaged residues in ctc2, although not vice versa. One explanation would be that the epitopes partially overlap, with that for mAb b78 located on the α and β face of the COOH-terminal domain and that for mAb b96.11 located on the δ and β faces. The mAbs that engaged ctc1 in the present study included M2, M5, b78, and DPD, and published data indicate that additional human mAbs (M7, M8, and M9) engage epitopes in ctc1, because these mAbs are cross-competitive, and also block binding of M2 and M5 but not binding of mAbs M1 and M3, which engage residues in ctc2 (4). Previous studies have defined epitopes for anti-GAD65 simply according to their location in domains based on the linear sequence, NH₂-terminal, PLP, or COOH-terminal. This is superseded by the structural data, because some epitopes hitherto assigned to the PLP domain are either contiguous with or even part of ctc1, e.g., mAb DPC, or of ctc2, e.g., mAb b96.11, and all mAbs for which epitopes apparently include an NH₂-terminal component, such as mAbs DPD, DPC, M8, and M9, engage epitopes in ctc1.

Although different mAbs from a single anti-GAD⁺ patient could be reactive with either ctc1 or ctc2 epitopes (Table 1), published data suggest that a response biased toward one or the other of the two epitope regions aligns with differing clinical expressions of type 1 diabetes. Thus, anti-GAD65 reactive with ctc1, as defined by blocking by rFab from mAbs b78, DPD, and DPC, was associated with enzyme inhibitory antibodies, whether from patients with Stiff Person Syndrome (22) or a subset of ketosis-prone type 1 diabetes with a higher β-cell functional reserve and a more benign clinical course (23). Also, anti-GAD65 reactive with ctc1/NH₂-terminal epitopes was associated with slowly progressive type 1 diabetes, i.e., latent autoimmune diabetes of adults (24,25). In contrast, anti-GAD65 reactive with ctc2 epitopes as judged by inhibition by rFab from mAbs b96.11 and DPA was associated with high-risk HLA-DQ alleles and rapidly progressive diabetes (3). Thus our data now structurally establish the epitope preference of autoantibodies to GAD65 that may dictate differing clinical expressions of type 1 diabetes.

The identification of an immunodominant B-cell epitope region in the COOH-terminal region of GAD65 conforms with known features of the normal antibody response. Immunodominant regions of antigens comprise sites on the antigenic molecule that are strongly recognized by most immunized individuals, although there are differences among individuals that reflect the summated reactivity of multiple B-cell clones that contribute to the response. Affinity maturation of an ongoing antibody response is associated with both “epitope focusing” to provide for selection of higher affinity antibodies and also “epitope spreading” to enable the immune response to encompass wider regions of the antigenic molecule. In type 1 diabetes specifically, there is evidence that the autoantibody response may first involve the COOH-terminal domain, with later spreading to the PLP and NH₂-terminal domains (3,26).

Notably, four of five major T-cell epitopes restricted by the high-risk HLA allele DRB1*0401 could be localized on the structure of GAD65 within the same region as the immunodominant B-cell epitope regions ctc1 and ctc2, and the fifth epitope was close to E264, which defines the M6 epitope in the PLP domain. Moreover, as CD8⁺ T-cells restricted by HLA class 1 appear to be more directly relevant to the effector stage of pathogenesis, it is of interest that the class II CD4⁺ T-cell epitope (T1, residues 115–130) overlaps a major class I CD8⁺ T-cell epitope (residues 113–124) restricted by HLA A2 (27). The finding that T-cell epitopes often occur close to B-cell epitopes, for both extrinsic and autoantigens (28), directs attention to the role of B-cells in modulating T-cell responses (29–34). In antigen-presenting cells, antigen-antibody complexes can remain intact after the internalization and fragmentation by proteases along the antigen-processing pathway (32,33), thus protecting residues from proteolysis and modulating presentation of peptides and ensuing T-cell responses. Studies on B-cell modulation of the T-cell response to GAD65 indicate that B-cell clones producing mAbs DPA and DPD (but not DPC) can present the T1 (residues 115–130), T2 (residues 270–283), and T5 (residues 556–575) peptide epitope sequences to DRB1*0401-restricted T-cell clones (30).

An enduring question for autoimmunity is why only a limited proportion, ∼2–3% (35), of all human proteins become selected as autoantigens and why only limited sites on such molecules serve as autoepitopes. Among possible explanations, structural characteristics are cited (35), albeit in the absence of knowledge of the actual protein structure of most autoantigenic molecules. This can now be examined in the case of GAD in the light of the contrasting antigenicity of the two isomers, GAD65 and GAD67, together with the substantial information derived from mutants of GAD65 based on sequence exchanges or single point mutations and the reactivity of such mutants with human mAbs exclusively to GAD65. We have already commented on distinctive features of epitope distribution on GAD65, notably the clear association of epitopes with regions of increased flexibility of protein structure of GAD65 versus GAD67 (5). Structural features associated with antigenicity for other proteins include high flexibility, loops, and protrusion from the antigen surface, and charged amino acids (36–39), all features of the COOH-terminal domain of GAD65. Comparably, structural flexibility was described for the immunodominant epitope of the human autoantigen thyroid peroxidase (40), although not for epitope regions on the crystal structure of the type 1 diabetes autoantigen IA-2 (41). Finally, if the engagement of epitopes of ctc2 were to be predictive of progression of individuals at risk to overt type 1 diabetes, knowledge of the clustering of epitopes for anti-GAD65 at the COOH- terminus and the existence of two separate COOH-terminal epitope clusters could allow the development of assay systems to exploit this. Also, if GAD65-based immunotherapies were to appear promising in retarding progression in at risk individuals to overt type 1 diabetes, consideration could be given to vaccine constructs based on epitopes associated with disease progression.

G.F. has received financial support for his PhD candidature from the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) Foundation, subordinated to the Ministry of Education, Brazil. J.C.W. is a National Health and Medical Research Council (NHMRC) Principal Research Fellow and Monash University Senior Logan Fellow. A.M.B. is a NHMRC Senior Research Fellow. This study has received funding and support from the NHMRC of Australia, the Australian Research Council, and the Australian Synchrotron Research Program.

We thank the Advanced Photon Source and Industrial Macromolecular Crystallography Association Collaborative Access Team (IMCA-CAT) from synchrotron facilities.

Part of this work was presented at the 9^th International Congress of the Immunology of Diabetes Society and American Diabetes Association Research Symposium, Nov 14–18 2007.