Diabetes Antibody Standardization Program: First Assay Proficiency Evaluation

Sera were obtained from 50 patients with newly diagnosed type 1 diabetes (median age, 17 years; range, 9–31). Of these, 19 were female and 31 were male, 49 were white and 1 was Hispanic. All samples were collected within 14 days of starting insulin treatment (median, 4.5 days). Control sera were obtained from 50 U.S. blood donors (median age 20 years; range 18–28). Sera were prepared in 100-μl aliquots and frozen. Coded sera were distributed to 46 laboratories in 13 countries (see appendix for listing of participating laboratories). Each participating laboratory received 1–3 uniquely coded sets of sera depending on the volume requirements of their assays. Sera from different sets could not be combined, so only those assays that used <100 μl could be included in the evaluation.

Laboratories were asked to test the sera with whatever relevant assays were currently in use for detection of prediabetes. Of the participating laboratories, 45 reported results for GAD antibodies, 43 for IA-2/ICA512 antibodies, and 23 for IAA. One laboratory reported results for antibodies to a GAD₆₅/IA-2_ic fusion protein and two for ICA testing by indirect immunofluorescence.

In addition, each laboratory received a lyophilized aliquot of the new WHO international reference reagent for GAD and IA-2 antibodies (97/500) (9) together with negative diluent serum. Laboratories were asked to test the standard (250 units/ml) undiluted and 1:2, 1:4, 1:8, 1:16, and 1:64 dilutions as well as the negative diluent in every GAD and IA-2 antibody assay run.

Laboratory-defined sensitivity for each assay was calculated as the percentage of sera from patients with newly diagnosed diabetes reported as positive using the laboratory’s own cutoff. Laboratory-defined specificity was calculated as the percentage of healthy control sera reported as negative using the same threshold.

Receiver operator characteristic (ROC) curve analysis was used to evaluate the performance of each assay in discriminating disease from nondisease. The area under the curve (AUC) with 95% CI was calculated assuming a nonparametric distribution. An AUC of 1.00 would indicate that the test achieved 100% accuracy in identifying disease, and an AUC of 0.00 would indicate that all disease and nondisease sera were misclassified on the basis of the test result. Each ROC curve was tested against a null hypothesis that the AUC was 0.5, which would indicate that the test achieved random assignment of disease/nondisease status.

To facilitate additional comparison between laboratories using very different thresholds for positivity, the coordinates of the ROC curve were used to identify the level of sensitivity achieved with 1-specificity of 0.1, which was defined as the adjusted sensitivity₉₀ (AS₉₀). This threshold was selected to minimize the effect of outlier antibody levels in the limited number of control sera tested. It is not suggested that this is a clinically appropriate threshold. The validity of these measures was assessed by relating them to other parameters, including laboratory-defined sensitivity and specificity and concordance of antibody levels.

For comparing relative levels of autoantibody in different assays, the diabetic sera were ranked from highest to lowest for each assay. For each serum, the rank in each assay was then plotted on the y-axis, with the sera arranged in order of ascending median rank along the x-axis. Concordance was assessed by linear regression of individual assay rank versus the median rank for all sera from patients with type 1 diabetes. Differences in the concordance of ranking in subgroups of laboratories were analyzed by comparing the variance of the regression in an F test.

The signal:noise ratio for each GAD and IA-2 antibody assay was calculated as the median cpm in patient samples reported positive in >50% of laboratories divided by the median cpm in control samples reported negative in >75% of laboratories. This parameter could not be calculated for IAA assays, as many laboratories had reported only Δ cpm with and without cold insulin.

The lower limit of detection in relation to the WHO standard was determined for each GAD and IA-2 antibody assay. The threshold was defined from the variability of measurement in control sera and calculated as (median cpm or OD) + 2 * (SD of the difference between duplicates). The limit of detection was defined as the dilution of the WHO standard with median cpm or OD immediately above the threshold. Three control sera that were reported as GAD antibody positive in >25% of laboratories were excluded from the calculation of limit of detection for these markers.

The WHO international reference reagent for GAD and IA-2 antibodies (97/500) has been defined as containing 100 units per vial (250 units/ml) (9). WHO units for GAD and IA-2 antibodies were calculated as: 1) an index related to the reference material using the formula

2) units derived from a logarithmic standard curve constructed from dilutions of the WHO reference reagent. Units assigned to individual dilutions were 125 units/ml for 1:2, 62 units/ml for 1:4, 31 units/ml for 1:8, 16 units/ml for 1:16, and 4 units/ml for 1:64. For comparing the two methods of unit derivation, the median and interquartile range of reported units for each of the 50 patient sera were compared using the Wilcoxon matched pairs test.

Results were reported from 46 assays (Table 1). Laboratory-defined sensitivity using each laboratory’s own cutoff ranged from 58 to 88%, the majority of assays reporting very similar sensitivity (median, 77%; interquartile range, 74–80%). Laboratory-defined specificity ranged from 80 to 100% and again was similar in the majority of assays (median, 94%; interquartile range, 92–98%). Samples from 30 patients and no control subjects were reported positive in ≥95% of assays; an additional 8 patients and 2 control subjects were reported positive in >50% of assays, and another 2 patients and 1 control subject were positive in >25% of assays (Fig. 1A). Agreement in >95% of assays was observed for 77 samples (34 patient samples and 43 control samples). The ranking of the GAD antibody titer in the patient samples was highly concordant between laboratories (r = 0.96, P < 0.0001; Fig. 1B).

For comparing assay performances directly, results from each assay were represented as a ROC curve, and the AUC and the sensitivity achieved with specificity of 90% (AS₉₀) were calculated (Table 1). All laboratories reported significant differences between patients and control subjects (all P < 0.0001), and results were significantly higher in patients than in control subjects (all P < 0.0001; Mann-Whitney U test). The majority of assays had similar performance (median, AUC 0.93; interquartile range, 0.90–0.94; median, AS₉₀ 84%; interquartile range, 79–87). AUC and AS₉₀ correlated positively with the laboratory-defined sensitivity (r = 0.48, P < 0.001; r = 0.45, P < 0.002), assay specificity (r = 0.36, P < 0.02; r = 0.42, P < 0.005), and assay signal:noise ratio (r = 0.47, P < 0.001; r = 0.52, P < 0.001). Nine assays (laboratories 109, 113, 114, 116, 120, 121, 147a, 150, and 153) had both AUC and AS₉₀ values in the upper 25th centile. One of these (laboratory 150) used a commercially produced GAD65 antigen that was labeled in-house with ¹²⁵I. The remaining eight assays with high AUC and AS₉₀ used in-house assays with ³⁵S labeled in vitro transcribed/translated recombinant GAD65 or GAD65/67 chimera. Eleven assays (laboratories 110, 112, 117, 128, 131, 135, 141, 142, 147b, 149, and 151) had both AUC and AS₉₀ values in the lower 25th centile; all but one of these (laboratory 110) also had laboratory-defined sensitivity or specificity in the lower 25th centile. These 11 assays included 8 using in-house methods with ³⁵S-labeled in vitro-transcribed/translated recombinant GAD65, one assay (laboratory 147b) using ¹²⁵I-labeled GAD65_46–585, one (laboratory 112) using Europium-labeled GAD65, and one commercial enzyme-linked immunosorbent assay (ELISA) (laboratory 151). Concordance as assessed by ranking of patient sera was lower between assays with AUC/AS₉₀ performances in the lower 25th centile (r = 0.94, variance = 22.9) than between assays in 25th–75th centiles (r = 0.96, variance = 15.4; P < 0.0001) and between assays in the upper 25th centile (r = 0.97, variance = 13; P < 0.0001 vs. lower 25th centile, and P < 0.02 vs. 25th–75th centiles; F test).

Results were reported from 44 assays (Table 2). Laboratory-defined sensitivity ranged from 40 to 88%, the majority of assays reporting very similar sensitivity (median, 57%; interquartile range, 54–58%). Laboratory-defined specificity ranged from 28 to 100% and again was similar in the majority of assays (median, 99%; interquartile range, 96–100%). Samples from 23 patients and no control subjects were reported positive in ≥95% of assays; an additional 5 patients were reported positive in >50% of assays, and another 1 patient was positive in >25% of assays (Fig. 2A). Agreement in >95% of assays was observed for 81 samples (41 patient samples and 40 control samples). The ranking of the IA-2 antibody titer in the patient samples was highly concordant between laboratories (r = 0.89; P < 0.0001; Fig. 2B). Samples with median rank <22 were those reported negative in the majority of assays, and the expected large scatter was observed. Two patient samples (median ranks 26 and 29) had clearly dichotomous results. These were reported as negative with ranks of 3–19 and 6–15 in all assays using the human IA-2_{256–556/630–979} antigen and positive with ranks of 22–33 and 25–43 in assays using IA-2ic or full-length IA-2 antigen.

Using ROC curves, all laboratories reported significant differences between patients and control subjects (all P < 0.0001), and results were significantly higher in patients than in control subjects (all P < 0.0001, Mann-Whitney U test). Again, the majority of assays had similar performance (median AUC, 0.77; interquartile range, 0.73–0.81; median AS₉₀, 59%; interquartile range, 56–65). Seven assays had both AUC and AS₉₀ values in the upper 25th centile (laboratories 109, 113, 115, 132, 150, and 156). Five of these assays with high AUC and AS₉₀ values were in-house radiobinding assays with ³⁵S-labeled in vitro transcribed/translated recombinant IA-2ic (n = 3), full-length IA-2, or IA-2_404–979. The remaining two assays used radiobinding assays with ¹²⁵I-labeled IA-2ic. Five assays (laboratories 110, 117, 119, 141, and 151) had both AUC and AS₉₀ values below the 25th centile, and all had laboratory-defined sensitivity or specificity in the lower 25th centile. These included four radiobinding assays with in vitro transcribed/translated IA-2_{256–556/630–979} and an ELISA with IA-2ic (laboratory 151). Low performance in IA-2 antibody assays was associated with low performance in GAD antibody assays: of the five IA-2 assays with both AUC and AS₉₀ in the lower quartile, four were in laboratories that also had GAD antibody assays with performances below the 25th centile (P < 0.05). As with GAD antibodies, concordance as assessed by ranking of patient sera was less between IA-2 antibody assays with AUC and AS₉₀ performance in the lower 25th centile (r = 0.85, variance = 59.8) than between assays in the 25th–75th centiles (r = 0.9, variance = 38.4; P < 0.0001) or between assays in the upper 25th centile (r = 0.89, variance = 43.5; P < 0.01).

Results were reported from 23 assays (Table 3). Laboratory-defined sensitivity ranged from 4 to 42% (median, 14%; interquartile range, 7–25%). Laboratory-defined specificity ranged from 89 to 100% (median, 100%; interquartile range, 98–100%). Only four assays reported IAA in >30% of patient samples (all four had laboratory-defined specificity of 98 or 100%). Samples from two patients and no control subjects were reported positive in >95% of assays (Fig. 3A). An additional four patients and no control subjects were reported positive in >50% of assays, and a further three patients were reported positive in >25% of assays. For 74 samples (26 patient samples and 48 control samples), >95% of assays were in agreement. Although the concordance of ranking of the IAA titer in the patient samples between laboratories was significant (r = 0.56; P < 0.0001), it was clearly inferior to that observed for GAD and IA-2 antibodies (P < 0.0001; Fig. 3B).

Using ROC curves, 5 of 23 laboratories did not find significant differences between patients and control subjects (P > 0.05). Performance characteristics varied substantially (median AUC, 0.67; interquartile range, 0.60–0.74; median AS₉₀, 36%; interquartile range, 26–47). AUC and AS₉₀ were positively correlated with the assay-reported sensitivity (r = 0.5, P < 0.02; r = 0.6; P < 0.005). Six assays (laboratories 116, 120, 121, 126, 133, and 153) had both AUC and AS₉₀ values in the upper 25th centile. All six assays used in-house microradiobinding assays with ¹²⁵I-labeled insulin as previously described (10,11,12). Laboratory-defined sensitivity in these six assays ranged from 14 to 36% (median, 31%). Five laboratories (laboratories 105, 117, 132, 135, and 136) had both AUC and AS₉₀ in the lower 25th centile. These included two of the three participating laboratories using a commercially available radiobinding assay with ¹²⁵I insulin. Concordance in ranking of patient samples was markedly higher between the assays with AUC and AS₉₀ in the upper 25th centile (r = 0.8, variance = 70.3) than between the remaining assays (r = 0.56, variance = 130; P < 0.0001; Fig. 3B and C).

The radioimmunoassay for antibodies to GAD₆₅/IA-2_ic fusion protein achieved 88% sensitivity with 100% specificity. The AUC was 0.960 (95% CI, 0.92–1.00) and AS₉₀ 0.90. Of the two ICA assays participating in the workshop, one (laboratory 137) achieved 90% sensitivity with 98% specificity, and the other (laboratory 120) achieved 68% sensitivity with 100% specificity.

The limits of detection of GAD and IA-2 antibodies, expressed as dilutions of the WHO standard, are shown in Tables 1 and 2. The undiluted WHO standard (250 units/ml) was above the limit of detection of all GAD and IA-2 antibody assays. The median limit of detection of the participating GAD antibody assays was the 1:16 dilution (16 units/ml). In seven assays, the limit of detection was below the 1:64 dilution (4 units/ml). The median limit of detection of IA-2 antibody assays was below the 1:64 dilution (4 units).

The range of GAD and IA-2 antibody levels expressed as WHO units/ml for the 50 patient samples is shown in Fig. 4. Units were slightly but significantly higher when calculated as an index than when derived from a standard curve for GAD antibodies (median 181 vs. 179 units/ml; P = 0.0004, Wilcoxon paired test) and IA-2 antibodies (median 78 vs. 67 units/ml; P = 0.0002, Wilcoxon paired test). For GAD antibodies, the interquartile ranges of reported units did not differ significantly between index- and standard curve-derived units (P = 0.065, Wilcoxon paired test). For IA-2 antibodies, the interquartile ranges of index-derived units were significantly lower than those of standard curve-derived units (P = 0.005, Wilcoxon paired test). Antibody levels measured in assays using IA-2_{256–556/630–979} antigen were lower than those from assays using IA-2ic or full-length IA-2 antigens, whether units were derived from an index or a standard curve (P < 0.002, Mann-Whitney U test).

The first DASP proficiency evaluation has demonstrated a remarkable degree of concordance between laboratories in GAD and IA-2 antibody measurement using radiobinding assays. This has been achieved much more quickly than by previous efforts to standardize ICA assays. The format and size of the workshop, together with the methods of analysis used, allowed fuller evaluation of assay performance than has been possible in small-scale proficiency schemes. Concordance between laboratories was directly related to their capacity to distinguish patient and control samples as defined in the ROC analysis. This was in turn related to assay type, antigen, and signal:noise ratio. For both GAD and IA-2 antibodies, only radiobinding assays achieved high sensitivity and specificity. For IA-2 antibodies, radiobinding assays that used the IA-2_{256–556/630–979} antigen had lower sensitivity than assays that used intact intracellular IA-2. Samples from two patients had antibodies against the intracellular IA-2 but not against the IA-2_{256–556/630–979} spliced form. For GAD antibodies but not for IA-2 antibodies, high sensitivity was also associated with higher assay signal:noise ratio. This difference may be attributable to higher signal:noise ratios in IA-2 antibody assays, which makes this variable less likely to be the limiting factor in determining sensitivity. Finally, the laboratories that were least able to distinguish patient and control samples were often the same for both GAD and IA-2 antibody assays.

An important aspect of this study was the evaluation of general introduction of common units for measurement of GAD and IA-2 antibodies, based on the WHO reference preparation. The value of introducing a common standard to allow laboratories to express GAD and IA-2 antibody results in common units was confirmed, and the Immunology of Diabetes Society now recommends that results be expressed as WHO units/ml. Two methods for deriving units were compared: an index calculated from the ratio of test sample to a single standard and units derived from a standard curve of multiple dilutions over high and low antibody levels. Small but statistically significant differences were observed using these two methods. Index-based units were consistently marginally higher than those derived using a standard curve. For GAD antibodies, between laboratory variation, measured as the interquartile range, was not significantly different between units derived from an index or from a standard curve. For IA-2 antibodies, however, the interlaboratory variation in units derived from an index was significantly less than in units derived from a standard curve. Supplies of the WHO reference reagent are limited, and laboratories will not be able to include this as a standard in all assays but will need to calibrate their own laboratory standards to the reference reagent. As comparability between laboratories was not improved by use of a standard curve, except perhaps at very low antibody levels (data not shown), it is concluded that laboratory standards should be calibrated to the undiluted WHO reference reagent using the more simple index method. It should be noted, however, that the use of multiple-point standard curves would be expected to be of greatest benefit in reducing variation within a laboratory over time and is likely to be useful in this context, although it has not been possible to address this issue in this workshop.

In contrast, there was wide variation between IAA assays in this workshop, and although a number of sensitive assays were identified, the overall performance of IAA assays was poor. In previous workshops, the majority of IAA assays were based on polyethylene glycol precipitation of immune complexes and used serum volumes of up to 600 μl (13), whereas this workshop was limited to assays that used a maximum of 100 μl per test sample. As a consequence, the majority of laboratories used a version of the recently introduced IAA microassays using protein A precipitation (10). Of the IAA microassays in the workshop, six had sensitivity >30% and these assays were highly concordant in their ranking of samples. The workshop demonstrated, however, that most laboratories have yet to optimize the IAA assay performance and that significant effort is needed to transfer the IAA microassays successfully to a larger number of laboratories. The introduction of a reference standard preparation, as has been done for ICA, GAD, and IA-2 antibodies, will be important.

In this set of samples, GAD antibody determination by radiobinding assays achieved higher sensitivity than testing for IA-2 antibodies or, as expected in patients of this age, IAA (Fig. 5). Sensitivity could be maximized by testing for both GAD and IA-2 antibodies. Of the patient samples, 88% were defined as GAD and/or IA-2 antibody positive in at least 80% of assays. One laboratory, using a radiobinding assay for antibodies to a GAD₆₅/IA-2_ic fusion protein, achieved 88% sensitivity with 100% specificity and the highest area under the ROC curve overall. The potential of ICA determination by indirect immunofluorescence to be the most sensitive single test was confirmed in this workshop, although this was achieved by only one laboratory. Previous workshops, however, have demonstrated the great problems of standardizing this assay.

Most of the laboratories that participated in this evaluation are primarily involved in research. Islet antibody testing, however, is likely to be increasingly used in clinical practice and may need to be performed in nonspecialist laboratories. It is therefore of interest that, particularly for IA-2 antibodies, some assay kits achieved good discrimination between cases and controls with AUC and AS₉₀ above the 75th centile. The format of this proficiency evaluation allows much fuller blinded evaluation of the performance of a kit than is generally possible. It is therefore hoped that more kit manufacturers will participate in DASP activities in the future.

In summary, the first DASP proficiency evaluation and the format of analysis have shown that, in the majority of participating laboratories, GAD and IA-2 antibody assays perform well. It has been possible to identify GAD antibody, IA-2 antibody, and IAA assays that achieve high sensitivity and specificity so that the characteristics associated with good levels of discrimination between disease and nondisease can be defined. This should allow laboratories that are not yet performing to the highest standards to identify areas for improvement. The validity of using both the index and multiple-point methods for deriving common units from the WHO standard preparation has been demonstrated, and it has been shown that good interlaboratory concordance is achieved when antibody levels are expressed in terms of WHO units. On the basis of these results, the Immunology of Diabetes Society now recommends that all GAD and IA-2 antibody results be expressed in WHO units/ml. The evaluation has also demonstrated that although the recently introduced IAA microassays can achieve high levels of sensitivity and good interlaboratory concordance; the sensitivity of these assays as implemented in most laboratories is low and must be improved. These results will be invaluable in the design, implementation, and interpretation of multicenter trials and studies in type 1 diabetes.

Manou Batstra, Department of Immunology, Diagnostic Center SSDZ, Delft, The Netherlands; Robyn Baume and Charles Verge, South Eastern Sydney Lab Services, Prince of Wales Hospital, Randwick, New South Wales, Australia; Margaret T. Behme, Diabetes Autoimmunity Laboratory, Siebens-Drake Research Institute, London, ON, Canada; Corado Betterle and Renato Zancetto, Autoimmune Endocrine Laboratory, Department of Medical and Surgical Sciences, University of Padova, Italy; Bernd Bierwolf, Medipan Diagnostica, Selchow, Germany; Emanuele Bosi and Ezio Bonifacio, San Raffaele Institute, Milan, Italy; Luis Castano and Ramon Bilbao, Endocrinology and Diabetes Research Group, Hospital de Cruces, Baracaldo, Bizkaia, Spain; Michael Christie, GKT School of Medicine, University of London, United Kingdom; Corrado Cilio and Anita Nilsson, Wallenburg Laboratory, Department of Endocrinology, University of Lund, Sweden; Peter Colman and Shane Gellert, Department of Diabetes and Endocrinology, Royal Melbourne Hospital, Victoria, Australia; Andrew Cotterill and David Cowley, Departments of Biochemistry and Pediatric Endocrinology, Mater Health Services, Brisbane, Queensland, Australia; Terri Daniels and Ake Lernmark, R H Williams Laboratory, University of Washington, Seattle, Washington; Massimo Pietropaolo and Dorothy Becker, Rangos Research Center, University of Pittsburgh, Pittsburgh, Pennsylvania; Francesco Dotta and Claudio Tiberti, Department of Clinical Sciences-Endocrinology, University of Rome “La Sapienza,” Rome, Italy; Toru Egashira, BML Inc. (Biomedical Lab), Saitama, Japan; Tamir Ellis and Clive Wasserfall, Center for Immunology and Transplantation, University of Florida, Gainesville, Florida; Jadwiga Furmaniak, FIRS Labs, RSR Ltd, Llanishen, Cardiff, United Kingdom; Vinod Gaur and Santica Marcovina, Diabetes Endocrinology Research Center Immunoassay Laboratory, University of Washington, Seattle, Washington; Pat Goubert and Frans Gorus, Diabetes Research Center, Vrije Universiteit Brussel, Brussels, Belgium; Mohammed Hawa and David Leslie, Department of Diabetes and Metabolism, St. Bartholomew’s Hospital, London, United Kingdom; Ari Hinkkanen, Department of Biochemistry, Abo Akademi University, Turku, Finland; Akira Kasuga, Department of Internal Medicine, Keio University School of Medicine, Tokyo Denroyoku Hospital, Tokyo, Japan; Eiji Kawasaki, The First Department of Internal Medicine, Nagasaki University School of Medicine, Nagasaki, Japan; Rose Kientsch-Engel, Roche Diagnostics, Penzberg, Germany; Clemens Jaeger, Immunology Laboratory, Med Klinik 3, Giessen, Germany; James La Gasse and William Hagopian, Pacific Northwest Research Institute, Seattle, Washington; Claire Levy-Marchal, Inserm Unit 457, Hôpital Robert Debré, Paris, France; Tihamer Orban, Joslin Diabetes Center, Boston, Massachusetts; Angie Pollard, Endocrine and Diabetes Unit, University Department of Paediatrics, Women’s and Children’s Hospital, North Adelaide, Australia; Santica M. Marcovina, Dyamid Diagnostics, c/o Diabetes Endocrinology Research Center, Immunoassay Laboratory, University of Washington, Seattle, Washington; Didac Mauricio, Endocrine Research Laboratory, Hospital de Sant Pau, Barcelona, Spain; Patricia Mueller, Centers for Disease Control and Prevention, Atlanta, Georgia; Silvia Pinach and Gianfranco Pagano, Laboratorio di Diabetologia e Malattie del Ricambio, Department of Internal Medicine, Molinette Hospital, Turin, Italy; Alberto Pugliese and Markus Zeller, Diabetes Research Institute, University of Miami, Miami, Florida; Wiltrud Richter, Department of Orthopedic Surgery, University of Heidelberg, Heidelberg, Germany; Matti Ronkainen, Petri Kulmala, and Mikael Knip, Department of Paediatrics, Oulu University, Oulu, Finland; Merrill Rowley and Shahnaz Fida, Department of Biochemistry and Molecular Biology, Monash University, Clayton, Victoria, Australia; Carani Sanjeevi, Department of Molecular Medicine, Karolinska Hospital, Stockholm, Sweden; Michael Schlosser and Manfred Ziegler, Institute of Pathophysiology, University of Greifswald, Karlsburg, Germany; Joachim Seissler, Diabetes Research Institute, University of Duesseldorf, Duesseldorf, Germany; Goran Sundkvist and Henrik Borg, Islet Antibody Laboratory, Wallenberg Laboratory, University of Lund, Lund, Sweden; Carina Torn and Mona Landin-Olsson, Diabetes Laboratory, University Hospital, Lund, Sweden; Alistair Williams and Polly Bingley, Division of Medicine, University of Bristol, Bristol, United Kingdom; Ana Maria Yamamoto, Immunology, Necker Hospital, Paris, France; Liping Yu and George Eisenbarth, Barbara Davis Center, University of Colorado, Denver, Colorado; and Anette Ziegler, Diabetes Research Institute of the Academic Hospital Munchen-Schwabing, Munich, Germany.