Reactivity to N-Terminally Truncated GAD₆₅(96–585) Identifies GAD Autoantibodies That Are More Closely Associated With Diabetes Progression in Relatives of Patients With Type 1 Diabetes

GAD autoantibodies (GADAs) are the most widely used marker of type 1 diabetes. They are a mainstay of diabetes prediction and recruitment to therapeutic intervention trials, as well as being used for disease characterization (1). However, many individuals found to be GADA positive with current assays are unlikely to progress to type 1 diabetes. Improved discrimination of diabetes risk can be achieved by testing for multiple islet autoantibodies, but the development of more disease-specific GADA assays that enable more efficient screening for type 1 diabetes is a high priority (2).

Recent international islet autoantibody workshops revealed systematic differences in reactivity between ELISAs and radiobinding assays (RBAs), which suggested that the performance of many RBAs may be improved by the use of the N-terminally truncated radiolabel ³⁵S-GAD₆₅(96–585) (3). We therefore assessed the ability of an RBA using the truncated GAD label to identify patients with recent-onset type 1 diabetes and to discriminate diabetes progression in first-degree relatives (FDRs) of type 1 diabetes patients in comparison with an assay using full-length GAD₆₅.

To determine the diabetes sensitivity of autoantibodies to the different GAD constructs, sera from 147 patients (89 male and 58 female, median age 11.6 years, age range 1.3–21 years) with recent-onset type 1 diabetes (median diabetes duration 1 day, range −7 to 90 days) were randomly selected from a well-characterized cohort (4), and assayed for GADA(1–585) and GADA(96–585).

Sera were available from 283 relatives (139 male and 144 female) who had previously been found to be GADA positive with a local RBA using ³⁵S-labeled full-length GAD₆₅, after the screening of 4,470 FDRs (2,121 male and 2,349 female) in the Bart’s-Oxford (BOX) family study (5). These GADA-positive relatives were followed up prospectively for disease development by an annual questionnaire (median follow-up to last contact or diabetes diagnosis 14 years, range 0.2–27.7 years, median age 31.4 years, age range 1.3–57.4 years). A subset of 459 of the first available samples (from 227 male and 232 female FDRs) was randomly selected from the 4,187 BOX FDRs who previously had screened negative with the local GADA assay and reassayed for GADA(1–585) and GADA(96–585). This subset was enriched with 31 of 179 screen-negative relatives who developed diabetes (Table 1). Local RBAs were used to measure autoantibodies to IA-2 (IA-2A) and insulin (IAA) (5) on all relatives, and zinc transporter 8 (ZnT8A) (6) on those relatives who initially had positive screening results for GADAs or who subsequently developed diabetes.

Sera from patients and relatives were assayed for GADAs using the National Institute of Diabetes and Digestive and Kidney Diseases harmonized assay protocol (7) with ³⁵S-methionine–labeled antigens made by in vitro transcription and translation of both N-terminally truncated GAD₆₅(96–585) and full-length GAD₆₅ encoded in the pTNT plasmid vector (Promega, Madison, WI) (3). To localize N-terminal epitopes more precisely, a subset of samples from the FDRs who had positive screening results was also reassayed for GADAs using ³⁵S-labeled GAD₆₅(46–585). A methionine residue was added to both truncated antigens to allow transcription. Samples were considered to be positive if they had autoantibody levels ≥97.5th percentile of 222 healthy schoolchildren (8); equivalent to 13.5 digestive and kidney (DK) units/mL for GADA(1–585), 12.8 DK units/mL for GADA(96–585), and 25.4 DK units/mL for GADA(46–585). Using these thresholds, the sensitivity of the GADA(1–585) assay was 72% at a specificity of 92%, and the sensitivity of the GADA(96–585) assay was 70% at 99% specificity in the Islet Autoantibody Standardization Program 2013 workshop. The interassay coefficient of variation of GADA(1–585) was 12.9% at 53 DK units/mL (n = 32). The interassay coefficient of variation of GADA(96–585) was 13.0% at 59 DK units/mL (n = 32).

HLA class II genotyping was available on 209 of the 283 (74%) GADA-positive FDRs. HLA class II DRB1, DQA1, and DQB1 analysis was performed on blood and mouth swab DNA with sequence-specific primers, as previously described (9). Haplotypes were established based on common patterns of linkage disequilibrium.

Categorical variables were compared using the χ² test. Genetic risk was analyzed according to the high-risk haplotypes DRB1*03-DQA1*0501-DQB1*0201 (DR3-DQ2) and DRB1*04-DQA1*0301-DQB1*0302 (DR4-DQ8), as well as the protective haplotype DRB1*02-DQB1*0602 (DR2-DQ6). Other haplotypes were classified as X. Survival analysis was performed using the Kaplan-Meier method, and the Mantel-Cox log-rank test was used to compare survival between groups. For all analyses, a two-tailed P value of 0.05 was considered significant. The area under the curve (AUC) of the receiver operating characteristic (ROC) with 95% CI was calculated assuming a nonparametric distribution of results using R software. Other statistical analyses were performed using the Statistics Package for Social Sciences Version 21 (IBM, New York, NY).

Using thresholds equivalent to the 97.5th percentile in schoolchildren, the sensitivity of GADA(96–585) was identical to GADA(1–585) in patients with recent-onset type 1 diabetes (Fig. 1A). Of 147 patients, 117 (80%) were positive for GADA(96–585), and 117 were positive for GADA(1–585). One hundred sixteen patients (79%) had autoantibodies to both GAD constructs, and there was excellent correlation between the levels of GADA(96–585) and GADA(1–585) (r = 0.99, P < 0.001; Fig. 2A). The ROC-AUCs based on the patients and schoolchildren were also very similar for GADA(1–585) and GADA(96–585) (0.94 [95% CI 0.91–0.97] and 0.93 [95% CI 0.9–0.96], respectively) (Fig. 3).

Of 283 relatives who were previously found to be positive for GADAs using the local RBA, 259 (92%) were positive on reassay using the harmonized assay protocol with ³⁵S-GAD₆₅(1–585), 206 (73%) were positive with ³⁵S-GAD₆₅(96–585), and 195 (69%) were positive with both labels (Fig. 1B). Of 70 relatives who subsequently developed diabetes, 66 (94%) were positive for GADA(1–585), 63 (90%) were positive for GADA(96–585), and 61 (87%) were positive for both specificities. Of 76 relatives previously found to be positive for GADAs who had at least one additional islet autoantibody, 73 (96%) were positive for GADA(1–585), 70 (92%) were positive for GADA(96–585), and 69 were positive for both specificities. Of these multiple antibody-positive relatives, diabetes developed in 39 (53%) with GADA(1–585) and 38 (54%) with GADA(96–585). Of 207 relatives with no additional autoantibodies, 187 (90%) were positive for GADA(1–585) compared with 136 (66%) who were positive for GADA(96–585) (P < 0.001). There was a good correlation between the levels of GADA(1–585) and GADA(96–585) in sera from the GADA-positive relatives (r = 0.96, P < 0.001) (Fig. 2b). Of 64 relatives positive for GADA(1–585), but negative for GADA(96–585), 38 (59%) had GADA(1–585) levels of <50 DK units/mL, a level found in 27 of the 147 (18%) recent-onset patients. The deletion of amino acids 46 to 95 of GAD65 was important in improving specificity with little loss of sensitivity; of the 64 samples with GADA(1–585) alone, 49 (77%) were positive for GADA(46–585), of whom diabetes developed in only 4 (8%).

Kaplan-Meier survival analysis showed that positivity for GADA(96–585) identified relatives positive for full-length GADAs who were at increased risk of diabetes progression (Fig. 4; P < 0.001). Of the 11 relatives who rescreened positive only for GADA(96–585), 1 had additional autoantibodies (IAA and IA-2A), but was lost to follow-up after 4 years, while diabetes developed in 2 others after 5 and 6 years of follow-up. Of 160 relatives carrying at least one HLA risk haplotype who rescreened positive for GADA(1–585), 129 (81%) were positive for GADA(96–585), compared with 18 of 35 (51%) with no HLA risk haplotype (P < 0.001). Furthermore, positive screening results for GADA(96–585) were less common in GADA-positive relatives carrying protective haplotypes; of 13 relatives carrying HLA-DQ6, 12 were positive for GADA(1–585), but only 3 were positive for GADA(96–585) (P = 0.001).

Of the 428 relatives who remained nondiabetic during follow-up who had previously screened negative with the original GADA assay, 7 (1.6%) were positive for GADA(1–585) alone, 2 (0.5%) were positive for GADA(96–585) alone, and 4 (0.9%) were positive for both. Of the 31 relatives whose conditions progressed to diabetes, but who screened negative for GADAs with the original assay, none were positive for GADA(1–585) and 1 was positive for GADA(96–585).

GADA measured using N-terminally truncated antigen achieved the same sensitivity in recent-onset patients as the assay using full-length GAD₆₅, while those relatives having autoantibodies to GAD₆₅(96–585) were at higher risk of disease progression than those with autoantibodies to full-length GAD alone. Survival analysis showed that very few GADA-positive relatives without autoantibodies to GAD₆₅(96–585) developed diabetes within 20 years. Furthermore, only a minority of GADA-positive relatives who carried protective HLA haplotypes were found to be positive when using the N-terminally truncated label.

Birth cohort studies (10,11) of relatives of type 1 diabetes patients have shown that autoantibody epitope reactivity typically spreads from the COOH-terminal and middle (pyridoxal phosphate binding) regions to the N-terminal domains of the molecule. Autoantibodies to the N-terminal region normally constitute a relatively minor component of GAD autoreactivity and in isolation have little association with progression to diabetes (12). Our data would support this observation, since most samples from relatives who developed diabetes showed similar antibody binding and levels with full-length ³⁵S-GAD₆₅ and ³⁵S-GAD₆₅(96–585) radiolabels (Fig. 2B). Furthermore, the majority (59%) of relatives found to be positive for GADA(1–585), but negative for GADA(96–585), had relatively low levels of GADA (<50 DK units/mL), which is consistent with a less vigorous autoimmune response in these individuals. This explains why sensitivity in recent-onset patients was maintained, but most relatives carrying protective haplotypes and only a small proportion of relatives with multiple islet autoantibodies were found to be negative when GADAs were measured using the truncated construct.

Several groups have investigated the effect of N-terminal truncations of GAD₆₅ on the disease sensitivity of GADA. Deletion of the first 194 amino acids did not cause decreased binding by eight prediabetic/diabetic sera (13), while in agreement with our findings an assay using ¹²⁵I-labeled GAD₆₅(46–585) performed similarly to an assay using ³⁵S-labeled full-length GAD₆₅ (14). However, to our knowledge, this is the first study to show improved discrimination of diabetes progression using an N-terminally truncated GAD₆₅ label. The use of truncated antigens is established practice for the measurement of autoantibodies to IA-2 and ZnT8, since the main diabetes-relevant epitopes are located in the intracellular portion of IA-2 and the carboxy terminal region of ZnT8 (15,16). If confirmed in other populations, including young children, our finding suggests that screening strategies to identify individuals at high risk of developing diabetes should use GADA RBAs based on N-terminally truncated protein.

Although fewer autoantibodies to disease-irrelevant GAD₆₅ epitopes were detected using the N-terminally truncated label, the Kaplan-Meier survival curve suggests that diabetes will develop in fewer than half of GADA(96–585)–positive relatives within 25 years (Fig. 4). N-terminally truncated GADAs were associated with autoantibodies to other islet antigens, but could not discriminate the risk of progression within multiple antibody-positive relatives. Further improvements in assay specificity are therefore desirable. This may be achieved by more radical N-terminal deletions, if additional diabetes-irrelevant epitopes are disrupted without affecting binding to diabetes-relevant epitopes. Our addition of an N-terminal methionine to GADA(96–585) to allow protein expression is unlikely to have affected antibody binding as it is neither highly charged nor bulky. The inclusion of affinity measurements may also help to identify GADA-positive individuals who are at increased risk of diabetes progression (12,17). The potential for truncated GAD₆₅ labels to identify patients with slow-onset autoimmune diabetes in adults with a clinical presentation of type 2 diabetes also needs to be investigated. A previous study (18) using GAD₆₅/GAD₆₇ chimeras rather than truncated GAD₆₅ found no difference in the time to insulin requirement between those patients with or without N-terminal autoantibodies.

RBAs are still widely used for the prediction and characterization of type 1 diabetes despite the advent of high-quality alternative assay formats such as the bridging ELISA (19) and electrochemiluminescence assay (20). Advantages of RBAs include their relatively low cost, high sensitivity, good flexibility, small serum volume requirement, and proven track record in diabetes prediction as well as the wide availability of equipment and reagents. However, a major shortcoming of GADA RBAs has been their relative lack of specificity. We have demonstrated that use of an N-terminally truncated GAD₆₅ label can improve the disease specificity of the GADA assay without the loss of sensitivity in patients and can identify GADA-positive relatives who are at higher risk of disease progression. As the recruitment of high-risk relatives to therapeutic intervention trials normally includes initial testing for GADAs, these findings strongly suggest that the adoption of autoantibody assays using N-terminally truncated GAD₆₅ would greatly improve screening efficiency for future studies aimed at preventing type 1 diabetes.

Acknowledgments. The authors thank the diabetes teams, pediatricians, physicians, and families in the Oxford region for participating in the Bart’s-Oxford study.

Funding. The Islet Autoantibody Standardization Program contributed to the costs of this study. The Bart’s-Oxford study is supported by Diabetes UK grant BDA 14/0004869. V.L. and C.B. worked within the framework of the Italian Ministry of Research “Ivascomar project, Cluster Tecnologico Nazionale Scienze della Vita ALISEI” and were supported by the Associazione Italiana per la Ricerca sul Cancro “AIRC bando 5 × 1000 N_12182” grant. P.A. was supported by grants from the German Federal Ministry of Education and Research (BMBF) to the Competence Network for Diabetes mellitus (FKZ 01GI0805) and to the German Center for Diabetes Research (DZD e.V.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.J.K.W., V.L., and P.A. researched the data, contributed to the discussion, and wrote the article. R.W., C.B., and K.M.G. researched the data, and reviewed and edited the article. P.J.B. researched the data, contributed to the discussion, reviewed and edited the article, and coordinated the Bart’s-Oxford study. A.J.K.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 49th Annual Meeting of the European Association for the Study of Diabetes, Barcelona, Spain, 23–27 September 2013; and at the 13th International Congress of the Immunology of Diabetes Society, Mantra Lorne, Victoria, Australia, 7–11 December 2013.