International Workshop on Lessons From Animal Models for Human Type 1 Diabetes: Identification of Insulin but Not Glutamic Acid Decarboxylase or IA-2 as Specific Autoantigens of Humoral Autoimmunity in Nonobese Diabetic Mice

Workshop participation was open and in large part subject to investigator registration for the Lessons From Animal Models for Human Type 1 Diabetes meeting (Denver, CO, October 2000) (18) as well as past performance in using assays that monitor humoral immunity in NOD mice. The goals, priorities, and workshop design were established before the workshop by the meeting organizers. As part of this process, it was agreed that participating investigators were allowed to use methods subject to current utilization within their laboratories (shown below) rather than a common method. Serum was voluntarily collected and submitted by individuals with access to various mouse colonies, as shown in Tables 1 and 2.

In accordance with the workshop design, participating laboratories performed autoantibody assays as indicated below.

IAA were measured by RBA similar to that previously described for humans (18). Sera (5 μl) were incubated in duplicate with 1.159 mU of ¹²⁵I-labeled insulin (Hoechst) contained in 25 μl of 50 mmol/l Tris and 1% Tween 20 (pH 8.0) buffer (TBT) for 3 days on ice. Antibody was subsequently captured by adding 50 μl TBT containing 2 mg Protein A Sepharose (Pharmacia) and 6 μl Protein G (γ bind) and incubated at 4°C for 1 h with shaking. Tubes were then washed four times each with 2 ml cold TBT, and pellets were counted in a γ-counter. Results were expressed as an index calculated as follows: (cpm of test sample − cpm of negative control serum)/(cpm of positive standard sample − cpm of negative control serum) × 100.

A 96-well filtration plate micro IAA assay was performed as previously described (19). ¹²⁵I-insulin (Amersham) of 20,000 cpm was incubated with 5 μl serum with and without cold human insulin, respectively, at a 1:5 dilution of serum for 3 days at 4°C in buffer A (20 mmol/l Tris-HCl buffer [pH 7.4] containing 150 mmol/l NaCl, 1% bovine serum albumin [BSA], 0.15% Tween 20, and 0.1% sodium azide). Fifty microliters of 50% Protein A/8% Protein G Sepharose (Pharmacia) were added to the incubation in a MultiScreen-NOB 96-well filtration plate (Millipore), which was precoated with buffer A overnight at 24°C. The plate was then placed on a plate shaker and shaken at low speed for 45 min at 4°C, followed by two cycles of four washes each with cold buffer A containing 0.1% BSA using a Millipore vacuum-operated 96-well plate washer. After washing, 40 μl scintillation liquid (Microscint-20; Packard) was added to each well, and radioactivity was determined directly in the 96-well plate with a TopCount (96-well plate β-counter; Packard) scintillation counter. The results were calculated on the basis of the difference in counts per minute between the well without cold insulin and the well with cold insulin (Δcpm) and expressed as an index: index = (sample Δcpm − negative control Δcpm)/(positive control Δcpm − negative control Δcpm).

Autoantibodies were determined with minor modifications of those previously described (20). Mouse serum (5 μl) was incubated with 20,000 cpm of ¹²⁵I-insulin (Amersham) with and without 8 units/ml cold insulin (Eli Lilly) at a final 1:5 dilution of serum. Dilutions and stocks were made in assay buffer (150 mmol/l NaCl, 20 mmol/l Tris-HCl, 0.1% NaN₃, 1% BSA, and 0.15% Tween 20 [pH 7.4]). Assays were performed in low protein-binding 96-well plates. The plates were incubated for 3 days at 4°C. Microfiltration 96-well plates were blocked for 18 h with assay buffer. The buffer was removed from the filtration plates, and 25 μl of 50% Protein A Sepharose (Sigma) was pipetted into each well. Fifty microliters of serum/antigen mixture from step 1 was added to each well and incubated at 4°C with shaking for 45 min. Each well of the filtration plate was washed (Millipore plate washer) eight times under vacuum with 200 μl wash buffer (150 mmol/l NaCl, 20 mmol/l Tris-HCl, 0.1% NaN₃, 0.1% BSA, and 0.15% Tween 20 [pH 7.4]). After the plates dried, 30 μl Optisafe scintillation fluid (Fisher) was added to each well and then counted in a Wallac LKB trilux β-counter. The results were reported as an index. The Δcpm was first calculated as cpm without cold insulin − cpm with cold insulin. The index was calculated as follows: ([sample Δcpm − negative control Δcpm]/[positive control Δcpm − negative control Δcpm]) × 100.

¹²⁵I-insulin (Amersham) of 20,000 cpm was incubated with 5 μl serum with and without cold human insulin, respectively, at a 1:5 dilution of serum for 7 days at 4°C in buffer A (20 mmol/l Tris-HCl buffer [pH 7.4] containing 150 mmol/l NaCl, 1% BSA, 0.15% Tween 20, and 0.1% sodium azide). Fifty microliters of 50% Protein A/8% Protein G Sepharose was added to the incubation in a MultiScreen-NOB 96-well filtration plate, which was precoated with buffer A overnight at 24°C. The plate was then placed on a plate shaker and shaken at low speed for 45 min at 4°C, followed by two cycles of four washes each with cold buffer A containing 0.1% BSA using a Millipore vacuum-operated 96-well plate washer. After washing, 40 μl scintillation liquid (Microscint-20) was added to each well and radioactivity was determined directly in the 96-well plate with a TopCount (96-well plate β-counter) scintillation counter. The results were calculated on the basis of the difference in counts per minute between the well without cold insulin and the well with cold insulin (Δcpm) and expressed as an index: index = (sample Δcpm − negative control Δcpm)/(positive control Δcpm − negative control Δcpm).

Enzyme-linked immunosorbent assay (ELISA) was performed as previously described (21). Briefly, the ELISA plate wells were coated overnight at 4°C with antigen (insulin, GAD65, and IA-2) in coating buffer (0.05 mol/l NaHCO₃/Na₂CO₃ buffer, pH 9.6) at a concentration of 1 μg/ml. The plates were then washed twice and saturated at 24°C for 2 h with phosphate-buffered saline (PBS)–0.05% Tween 20 (PBS-T). After washing three times with PBS-T, 100 μl diluted serum (1:80, 1:160, and 1:320) was added in each well. Samples were incubated at 24°C for 2 h. Plates were washed five times, and 100 μl alkaline phosphatase–conjugated goat anti–mouse IgG second antibody (1:1,000; BD PharMingen) was added to each well, followed by incubation at 24°C for 2 h. Finally, the plates were washed five times with PBS-T and 100 μl of phosphatase substrate (10 mg p-nitrophenyl phosphate disodium (Sigma) dissolved in 10 ml substrate buffer (1 mol/l diethanolamine, 3 mmol/l sodium azide, and 0.5 mmol/l magnesium chloride [pH 9.8]) added to each well. Plates were read at 405 nm at various time points. The data were expressed as OD (optical density) after subtraction of background absorbance.

Comparisons of antibody levels between control and NOD mice were performed using the Mann-Whitney U test. Receiver operator curves were used to compare sensitivity and specificity of individual assays. Comparisons of antibody levels at different time points in individual mice were performed using the Wilcoxon matched pairs test. Antibody levels between assays were correlated using Spearman’s rank correlation.

Each of the five participating laboratories observed increased IAA signals in serum samples obtained from NOD mice in comparison with control mice (Fig. 1). Using the 99th percentile of signals obtained using control mouse sera as the threshold for positivity, IAA were observed in 0–30% (4-week-old NOD mice), 12–56% (8-week-old NOD mice), 17–56% (12-week-old mice), and 15–75% of NOD mice at diabetes onset. There were substantial differences, however, in the frequency of IAA reported in each laboratory. Laboratories 1 and 2 had relatively high and comparable sensitivities. Both used an RBA but with different formats for measurement (see research design and methods). Laboratory 1 did not compete binding with excess cold insulin and reported low-level nonspecific binding in a minority of sera from control mice. The assay used by laboratory 2 included competition and observed little or no nonspecific binding in control mice. Both laboratories found increased binding in NOD mouse sera taken at each time point compared with age-matched samples from both control mouse strains. The sensitivity in laboratory 3, using an RBA with competition, was somewhat decreased when compared with laboratories 1 and 2 and observed significant differences between NOD sera and age-matched control samples in only 8- and 12-week old mice. Laboratory 4, which also used an RBA with competition, had a very low sensitivity and very low binding, as indicated by the cpm of the signals reported. Laboratory 5 measured antibodies using an ELISA. A significant difference in insulin binding was observed between control sera from BALB/c mice as compared with C57/B6 mice (P < 0.0001), and binding was higher in 12-week-old mice compared with 4- and 8-week-old mice in both of the control strains (P < 0.001). The relatively large range of binding seen in sera from control mice resulted in a low sensitivity for samples from 4- (0%), 8- (14%), and 12-week-old (19%) NOD mice. The sensitivity of the ELISA for sera from diabetic NOD mice (42%) was less than that of laboratories 1 (75%) and 2 (69%).

For these studies, samples were collected at 4, 8, and 12 weeks of age from 20 NOD mice. The first detection of IAA in individual mice varied between laboratories, with laboratories providing high sensitivity subjects for presentation (Fig. 2). Laboratories 1 and 3 detected IAA in mice early at age 4 weeks (note: laboratory 2 did not test 4-week-old mice). Other mice were negative at 4 weeks and became positive at 8 weeks, and some mice became IAA-positive in only their 12-week sample. Some individual mice showed large differences between IAA levels at 8 and 12 weeks of age. Only a few mice showed major decreases in IAA titer in their 12-week sample. Overall, the highest IAA levels were observed in 12-week samples (P < 0.05 for each laboratory). IAA at any time point were found in 60% (laboratory 1), 65% (laboratory 2), and 55% (laboratory 3) of NOD mice.

IAA levels in NOD mice were highly correlated between laboratories 1 and 2 (r = 0.77) but much less so between laboratories 1 and 5, which used an ELISA for measurement (r = 0.36). For examination of concordance, IAA levels in NOD mice were expressed as scores of 0 (negative), 1 (doubtful), 2 (positive), and 3 (strong positive) corresponding to less than the 97th percentile of control mice samples, between the 97th and 99th percentiles of control sera, greater than the 99th percentile and less than 10 times the 99th percentile of control sera, and greater than 10 times the 99th percentile of control sera, respectively (Fig. 3). Of 151 NOD mouse sera tested in at least three laboratories, 51 (33%) were scored negative in all laboratories and 8 (5%) were scored positive or strong positive in all laboratories. The major discrepancies were due to laboratory 4, which had a very low sensitivity, and laboratory 5, which used ELISA. When only laboratories 1, 2, and 3 were examined, concordance was improved with 53 (35%) sera scored as negative and 42 (28%) scored as positive or strong positive in all laboratories. Discrepant sera among these three laboratories were often accounted for by positivity in a significant number of 8-week-old NOD mouse sera in laboratory 2 only and the decreased sensitivity in laboratory 3.

Of the participating laboratories, three tested GADA and IA-2A using workshop sera (Figs. 4 and 5). Laboratories 1 and 2 used RBA with recombinant human GAD65 or IA-2, and laboratory 5 used an ELISA with recombinant mouse GAD65 and human IA-2. All laboratories observed increased signals in NOD mouse samples obtained at diabetes onset in comparison with control mouse samples, with sensitivities ranging from 29 to 50% for GADA and from 36 to 50% for IA-2A compared with 1% of control mice. Laboratories 1 and 5 also observed increased signals in 8- and 12-week-old NOD mice. Although binding for NOD mouse sera in the RBA was increased relative to control mice, it was extremely low when compared with binding levels commonly observed with human prediabetic and diabetic sera (data not shown). Signals obtained in NOD mice by laboratories 1 (RBA) and 5 (ELISA) were significantly correlated for both GADA (r = 0.59; P < 0.001) and IA-2A (r = 0.6; P < 0.001). A significant correlation between signals obtained for GADA and IA-2A was observed by laboratories 1 (r = 0.57; P < 0.001), 2 (r = 0.57; P < 0.001), and 5 (r = 0.29; P < 0.01). GADA signals did not correlate with IAA signals in all three laboratories (all r < 0.1). IA-2A signals were correlated with IAA signals in laboratory 5 (r = 0.37; P < 0.001) but not in laboratories 1 and 2 (both r < 0.1).

As previously indicated, sera were provided by six different centers. The IAA, GADA, and IA-2A results obtained for laboratories 1 (RBA) and 5 (ELISA) relative to the center providing sera are shown in Fig. 6. For IAA, no differences in results between NOD colonies were observed. However, for both GADA and IA-2A, substantial differences were noted. Specifically, increased GADA or IA-2A signals were observed in only 2 of the 60 NOD mouse samples provided by center A, whereas sera from diabetic NOD mice from centers D and E were often positive for both GAD and IA-2 antibodies.

The observations that GADA and IA-2A signals were weak, that they correlated between RBA and ELISA, that they were colony-dependent, and that positive GADA and IA-2A signals were often observed in the same samples raised the question of whether the signals were due to specific autoantigen binding. As a first step, antibodies to the protein myelin oligodendrocyte glycoprotein (MOG), unrelated to type 1 diabetes, were determined in selected positive and negative NOD mouse sera using the same assay system used by laboratory 1. Antibody binding to MOG did not correlate with IAA (r = −0.15) but remarkably did correlate with both GADA (r = 0.63; P < 0.01) and IA-2A (r = 0.77; P < 0.01) signals measured by that laboratory. Next, the GADA assay was performed with and without competition with an excess of cold GAD65 (Fig. 7). Competition of binding in a serum from a mouse immunized with GAD65 was complete, whereas the binding from the unimmunized NOD mice was unaffected by competition with GAD65, suggesting that the binding was not antigen-specific. Binding in the IAA RBA was completely inhibited by competition with excess insulin.

A significant portion of our knowledge regarding the pathogenesis of type 1 diabetes has been derived from studies of animal models of the disease and in particular, the NOD mouse. However, one of the more disappointing features of studies involving NOD mice has been a relative lack of consensus in terms of investigator agreement of target antigens of β-cell autoimmunity. Indeed, although much enthusiasm has been generated toward the potential pathogenic significance of immunities to the autoantigens insulin, GAD, and heat shock protein in NOD mice, significant controversies remain. The outcomes of this workshop should, in part, bring partial resolution to these questions and again, in part, work to resolve existing controversies.

The data presented gave unanimous support to the concept of the autoantigenic role for insulin molecule as a target of humoral autoimmunity in NOD mice. The ability of every laboratory to detect IAA lends support to this notion as do the facets of relative disease specificity (i.e., presence in NOD vs. non-NOD strains), good concordance in sample identity (again, despite significant differences in assay formats), and the observations of frequencies of IAA following those predicted for the natural history of diabetes. RBA for IAA in NOD mice seemed superior to ELISA when viewed as a function of sensitivity. Furthermore, unlike the RBA, the ELISA format provided results that were both age- and control strain–dependent and that were relatively discordant with those obtained by RBA. These findings comparing assay format for detecting IAA are consistent with those for detection involving human sera, where IAA by RBA represent a clearly superior choice over that of ELISA. Indeed, the ability of investigators even to detect these autoantibodies using NOD sera has become possible in large part only as a result of efforts involving the development of a microassay (22). Despite these opinions, the workshop studies clearly would have been aided by the inclusion of additional laboratories using ELISA formats for IAA detection, a factor that would have allowed for improved comparisons as were performed with the RBA.

As to the two other markers of anti-islet autoimmunity tested, GADA and IA-2A, the results of this workshop call into question their role as a target of humoral autoimmunity. Such a conclusion came only after a series of initial observations that gave some credence and potentially interesting values. Approximately half of diabetic NOD mice were found to have increased GADA and IA-2A signals compared with control mice by both ELISA and RBA. Although increased signals were also detected before diabetes onset, they were infrequent, suggesting that they were secondary to the appearance of IAA. However, a series of observations led us to question whether the increased GADA and IA-2A binding were specific for autoantigen in the first place. First, unlike the results obtained with IAA and unlike results in sera from humans with type 1 diabetes, binding in NOD mouse sera was extremely low (<2.5% of total antigen). Second, whereas IAA were consistently detected in all NOD mouse colonies, GADA and IA-2A elevations were determined to be colony-variant (e.g., observed in sera obtained from colonies D and E). Third, neither GADA nor IA-2A signals correlated with IAA signals but were significantly correlated with each other, and sera with elevated GADA almost always had elevated IA-2A. Testing sera against a fourth unrelated antigen (mouse MOG) showed elevated binding in the same sera with GADA and IA-2A, further suggesting nonspecific interactions with labeled antigen. Finally, attempts to compete binding to GAD autoantigen with excess cold GAD using a protocol that effectively competes 100% of the binding of sera from humans with type 1 diabetes and from GAD-immunized mice was ineffective in the NOD mouse sera, strongly suggesting that the weak elevations seen in the GADA and IA-2A were not specific binding of autoantibody. These observations call for a more critical assessment of data in which small quantitative differences between disease and control groups are found.

This workshop validated insulin as a major target of autoimmunity also in the NOD mouse model of autoimmune diabetes, identified sensitive RBA as the method of choice for IAA measurement in mice, and brought into question GAD and IA-2 as targets of humoral autoimmunity in the NOD mouse. By doing so, the workshop demonstrated similarities and differences between disease characteristics of the NOD mouse and human, in which insulin is also a major early target of humoral autoimmunity but high levels of GADA and IA-2A are detected in the majority of patients before and at diabetes onset. This and previous workshops in humans using large numbers of sera from control and disease-related cohorts tested in a blinded manner have now proved to be invaluable in validating antigens and assays in type 1 diabetes. The findings of these workshops should also provide important lessons for other autoimmune diseases. As to the question of future workshops, such efforts seem necessary and should focus on determination of IAA isotypes in NOD mice, the inclusion of more laboratories that use ELISA formats, and analysis of autoantibodies in BB rats.

These studies were performed, in part, with funding provided by the Juvenile Diabetes Research Foundation International. The authors thank LiPing Yu, Clive Wasserfall, Tamir Ellis, Kerstin Koczwara, Heike Naserke, Vito Lampasona, Chantal Coehlho, and Shane Gellert for technical and scientific assistance in performing these assays.