NOD/ShiLtJ mice were obtained from The Jackson Laboratory (Bar Harbor, ME). All mice were maintained at the NIAID animal facility and cared for in accordance with institutional guidelines. All mice experiments were performed at NIAID in compliance with the guidelines of the NIAID/NIH Institutional Animal Care and Use Committee (IACUC). Breeders were injected with a single dose of 50 μL of complete Freund's Adjuvant (CFA, Sigma-Aldrich, St. Louis, MO) in the hind footpad to prevent them from developing diabetes [⁴⁰]. For the insulitis study, only female NOD mice from the third or later litters of each breeder female were used due to lower diabetes incidence in the first two litters after CFA injection.

Human plasma samples of known RBA IA status (five IA-positive long-standing diabetic individuals; 4 IA-negative long-standing diabetic individuals and 20 IA-negative non-diabetic individuals) were purchased from Bioreclamation, LLC, Hicksville, NY. Ten IA-positive plasma samples from long-standing diabetic individuals were obtained from the UK GRID diabetic cases collection (http://www.childhood-diabetes.org.uk/grid.shtml) and plasma and serum samples from the same ten long-standing diabetic individuals from the Genes and Phenotypes of Type 1 Diabetes collection (http://www.t1-diabetes.org.uk) at the University of Cambridge. These samples were collected with ethical approval from the National Health Service Cambridgeshire Research Ethics Committee and written consent was obtained from all individuals or parents of individuals who were too young to consent. Plasma samples from 59 healthy non-diabetic individuals of unknown IAA status were purchased from George King Biomedical Inc., Overland Park, KS. Thirty-four serum samples were obtained from long-standing diabetic subjects who consented to and enrolled in the NIH protocol, The Natural History of Autoimmune Diabetes and Its Complications (NCT00896610) approved by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) institutional review board.

Initial human IA ECL assay development was conducted at Wellstat Diagnostics in Gaithersburg, MD, USA (Lab1-Wellstat). To examine the reproducibility of the assay, the same samples from the 34 diabetic and 59 healthy non-diabetic individuals were further evaluated at test sites at the Laboratory of Immunology, National Institute of Allergy and Infectious Disease in Bethesda, MD, USA (Lab2-NIAID); and the JDRF/Wellcome Trust Diabetes and Inflammation Laboratory, University of Cambridge in Cambridge, UK (Lab3-DIL).

Mice were monitored for T1D onset by measurement of blood glucose levels weekly using an Accu-Chek Advantage^® meter (Roche, Mannheim, Germany). Non-fasting blood glucose levels ≥ 200 mg/dL for two consecutive weeks indicated T1D onset.

Streptavidin Dynabeads^® M-280 were obtained from Invitrogen (Oslo, Norway). Recombinant human insulin (MBL^® International Corporation, Woburn, MA) was biotinylated using Biotin-LC-Sulfo-NHS Ester (Pierce Biotechnology, Rockford, IL), conjugated to the streptavidin Dynabeads, and stored at 4°C until used. For the IAA ECL assay (Figure 1A), the insulin-conjugated beads were pre-blocked with skim milk (BD Difco™, Sparks, MD) in phosphate-buffered saline (PBS) with 0.3% Tween 20 (Sigma-Aldrich) for 60 minutes and then incubated with 50 μL per well of a 1:100 dilution of mouse serum or mouse monoclonal antibody to insulin (ABCAM, Cambridge, MA) in triplicate for 60 minutes while shaking in a 96-well plate. The beads were washed and retained in the wells by a magnet. Next, the beads were resuspended in ECL Assay Buffer (PBS with 0.5% BSA + 0.3% Tween 20) plus an anti-mouse IgG secondary antibody (BioVeris™, Gaithersburg, MD) conjugated to Tris(bipyridine)ruthenium(II) cation [Ru(bipy)₃]²⁺ and diluted in StabilCoat (Surmodics, Eden Prairie, MN). The plate was then incubated with shaking for 30 minutes at room temperature in an M-SERIES^® M1MR analyzer (BioVeris™) followed by electrochemiluminescence measurement. M1MR analyzer solutions include: GLO Solution, CLEAN Solution, STORE Solution, and DILUENT Solution (provided by Wellstat Diagnostics, LLC, Gaithersburg, MD). Human serum samples tested using this assay format used a ruthenylated anti-human IgG secondary antibody (Wellstat Diagnostics, LLC, Gaithersburg, MD).

Pancreata were removed from female mice at various ages from 5-20 weeks and fixed in 10% formalin. Paraffin-embedded tissue sections were prepared and stained with hematoxylin and eosin (H&E) by Histoserv, Inc. (Germantown, MD). Insulitis was scored as previously described [⁴¹]. At least four sections from each pancreas were examined and on average ≥ 40 islets per pancreas were scored. Only seven of the 58 mice had fewer than 40 islets scored; four mice had > 25 islets scored, and three mice had few identifiable islets with severe insulitis (15-19 islets).

Blood glucose levels in female NOD mice were measured and serum was collected once a week from 4 weeks up to 20 weeks of age. Serum was stored at -80°C until tested. IAA and insulitis in the prediabetic phase were examined, while diabetic mice were excluded. Fifty-eight mice were sacrificed at various ages and their pancreata were removed for histological examination and insulitis scoring as described above. IAA levels were measured in triplicate using the ECL assay and normalized to the signal measured with beads alone (no serum added).

Fifty-four NOD females were monitored for development of T1D from 4 weeks up to 36 weeks of age. Blood glucose levels were assessed and serum was collected every 1.5-2 weeks. Serum was stored frozen at -80°C until tested. Serum IAA values were measured in triplicate for each mouse at each time point by ECL as described above. IAA value is calculated as signal over background signal from negative control sera. ECL signals from the negative control sera were negligible; i.e., IAA value was approximately 1.0-1.2. The negative control sera were from non-diabetic NOD mice from the first litter of the CFA-treated breeders. Receiver operating characteristic (ROC) curves were generated using the IAA values at 7, 8/8.5, or 10 weeks of age for the NOD mice that did or did not develop diabetes by 20 weeks of age. Threshold values for a "positive" predictive (of later diabetes development) IAA measurement were based on IAA values with the highest likelihood ratios calculated from the ROC curve analyses. Since our NOD mouse colony has a diabetes incidence of about 40% by 20 weeks, we also chose the 60^th percentile as an alternative threshold for a "positive" predictive IAA measurement. Using these thresholds, we calculated the sensitivity, specificity, positive predictive value, and negative predictive value for diabetes prediction at 7, 8.5, and 10 weeks of age.

Two diluents manufactured by Wellstat Diagnostics (Gaithersburg, MD) were utilized throughout the assay: Human Sample Diluent (HSD) and Human Assay Diluent (HAD). HSD consists of deoinized water (Millipore, Billercara, MA), sodium phosphate-monobasic (Sigma-Aldrich), sodium phosphate-dibasic (Sigma-Aldrich), sodium chloride (Sigma-Aldrich), bovine serum albumin (SeraCare, Milford, MA), bovine IgG (Millipore), 2-Methyl-4-isothiazoline-3-one hydrochloride (MIT) (Sigma-Aldrich), 30% Brij-35 solution (Sigma-Aldrich), MAK-3 IgG Poly (Roche, Chicago, IL), HBR-1 (Roche) and pH adjusted to 7.2 ± 0.10. The solution is sterile filtered and stored at 2-8°C. The formulation for HAD is the same, with the addition of polyvinylpyrrolidone (Sigma-Aldrich) and polyvinyl alcohol (Sigma-Aldrich) prior to the pH adjustment. Additional diluents used in the operation of the BioVeris M1MR analyzer include: DILUENT Solution, STORE Solution, GLO Solution, and CLEAN Solution.

The assay is based on the same technology as the murine IAA ECL test. Briefly, 30-35 μL of serum or plasma were diluted 1:10 in HSD in 1.5 mL polypropylene straight-walled tubes (VWR). M-270 Streptavidin (SA) Dynabeads (Invitrogen), pre-washed with HSD, were then added to a final concentration of 0.4 mg/mL, incubated at 25°C for 15 minutes with gentle mixing in a pre-clearing step, spun briefly at 1000 × g, and magnetized for 3 minutes to remove the beads. The bead-free supernatant was harvested into a fresh tube and remagnetized for 2 minutes before 50 μL of sample was transferred to each of six wells of a 96-well polypropylene plate. 25 μL/well of either HAD or 50 ng/mL biotin-insulin in HAD was then added to triplicate wells. The plate was sealed and incubated for 120 minutes with constant shaking (1400 RPM at 25°C, throughout protocol). Then 25 μL/well of HAD washed M-270 SA Dynabeads (0.4 mg/mL) were added to all assay wells, followed by a 30 minute incubation with shaking. A plate magnet (LifeSep) was applied for 2 minutes to isolate IA/insulin complexes, which were then washed with HSD twice by shaking the plate for 2 minutes, with plate magnet separations for 2 minutes between each step. One hundred μL of 0.30 μg/mL Wellstat TAG-goat anti-human Fc was then added to each well and incubated for 60 minutes with shaking. The plate was then washed with Assay Run Diluent (ARD) as described above and then the beads were resuspended in 150 μL ARD, and evaluated using the M1MR Analyzer. Specificity was determined by competition in which varying concentrations of non-biotinylated insulin were added to the samples followed by shaking at 1400 rpm at 25°C for 60 minutes prior to incubation with biotinylated insulin.

ROC curves were generated and analyzed using Prism software (GraphPad, San Diego, CA). Spearman's correlation coefficients (r_s) and corresponding p-values were calculated using S-Plus (TIBCO Software Inc., Palo Alto, CA) or R (http://www.r-project.org/) statistical software. Pearson's coefficients were determined using Prism Software (Graphpad) and compared according to the method described by Hanley, J.A. et al. [⁴²]. Linear regression was used to analyze variability in the determined IAA values and sample ranks within cohorts using Prism software.

To develop a murine IAA assay that does not require the use of radioactivity or lengthy incubation times, we took advantage of ECL technology in which a luminescent signal is generated by a reaction involving a ruthenium (Ru) conjugate at the surface of an electrode [⁴³]. The assay is performed in a 96-well format using two microliters of serum per mouse in 2.5 hours (Figure 1A). Magnetic beads coated with insulin were used to capture IAA from serum, and a polyclonal secondary anti-IgG antibody labeled with the ruthenium chelate, Tris(bipyridine)ruthenium(II) cation [Ru(bipy)₃]²⁺ (TAG) was used to detect bead-bound IAA remaining after washing the wells with magnetic retention of the beads (Figure 1A and 1B). The redox reaction between Ru and the substrate tripropylamine (TPA) that occurs near the electrode is a regenerative process that allows for an ECL signal that amplifies over time (Figure 2A). Background signal is very low as photons can be generated only in the electric field near the electrode surface when the Ru is brought into proximity with the electrode by the magnet. To examine the dynamic range and specificity of the assay, we used a mouse monoclonal antibody against insulin and determined that our ECL assay can discriminate antibody concentrations from less than 10 pg/mL to greater than 100 ng/mL (Figure 2B). No appreciable signal was measured with an isotype control primary antibody. We next evaluated the ability of the assay to detect IAA in murine serum. Little to no ECL signal was detected in samples from non-diabetic mouse strains, including the C57BL/6 and BALB/c strains, in contrast to the robust signal in the majority of 8 week-old NOD mice (Figure 2C). These data demonstrate that our IAA ECL assay is specific and can potentially detect serum insulin autoantibody levels over a nearly 5-log concentration range.

We next sought to evaluate the variations in IAA levels in individual mice during the progression to disease in the NOD mouse model for T1D. Female NOD mice were individually followed from 4 weeks of age to up to 36 weeks of age with peripheral blood samples taken every 1.5 weeks to determine IAA and blood glucose levels. We found that IAA levels peaked or achieved a plateau at approximately 8-10 weeks of age in mice that became diabetic by 20 weeks of age (Figure 3A and 3B). The majority of these mice did not develop diabetes until 16 weeks of age or later (Figure 3C), but after reaching 10 weeks of age the serum IAA levels began to drop markedly, but never returned to baseline levels before the mice had to be sacrificed due to disease. The average IAA levels in mice that were not diabetic by 20 weeks of age were lower and peaked later (Figure 3A, 95% confidence interval).

We next determined the physiologic utility of the ECL measurements by comparing IAA levels with insulitis severity in histological sections of the pancreas in individual NOD mice at various ages. Insulitis severity was determined for each mouse using an insulitis index as outlined in Figure 4A. We consistently observed a higher insulitis index in animals with elevated IAA levels. Moreover, insulitis severity was correlated with the maximum IAA value better than with the final IAA measurement taken at the time the pancreas was removed for histology (r_s = 0.793 vs 0.736, respectively, Figure 4C and 4B, top panels). The correlation coefficients were stronger with the 5-10 week old mice (Figure 4B and 4C, middle panels) compared to the data set containing older mice (Figure 4B and 4C, top panels). The age-dependence of the correlation of IAA values with insulitis is further demonstrated by the loss of correlation with increasing age of the analyzed mice (Figure 4B and 4C, bottom panels). These results demonstrate that ECL measured IAA levels provide a reliable indicator of autoimmune destruction of the pancreatic islets accessible in the blood.

As IAA levels measured by the ECL method correlated with insulitis in NOD mice, we hypothesized that we could use IAA measurements to predict diabetes diagnosis. We, therefore, followed 54 mice over 20 weeks, monitoring blood glucose levels and collecting serum samples every 1.5 weeks. Using receiver-operating characteristic (ROC) analysis, we determined that IAA values measured between 8 and 10 weeks of age were most predictive of diabetes (Figure 5). The ROC plots illustrate that the greatest ratio of '% sensitivity' to '100 - % specificity' (maximum likelihood ratio) was obtained when measuring IAA between 8 and 10 weeks of age rather than 7 or 12 weeks of age (Figure 5 and data not shown). Next, we calculated the sensitivity, specificity, positive predictive values, and negative predictive values based on IAA thresholds determined by ROC analysis (Table 1). In our colony the diabetes cumulative frequency for female NOD mice is about 40% by 20 weeks; therefore, a secondary analysis was performed wherein the thresholds corresponded to the 60^th percentile for this cohort of NOD mice (Table 2). IAA values over the threshold calculated from both analyses at 8.5 and 10 weeks were highly predictive of diabetes onset by 20 weeks of age (Tables 1 and 2, and Figure 6). The sensitivity and specificity using the threshold from the ROC analysis to predict diabetes were 80% and 93%, respectively at 8.5 weeks, and 84% and 93%, respectively at 10 weeks. The positive and negative predictive values were 91% and 84%, respectively at 8.5 weeks, and 91% and 86%, respectively at 10 weeks. Thus, the ECL values measured at 8.5 weeks were highly predictive of eventual diagnosis of diabetes by 20 weeks of age in NOD mice.

As the murine IAA ECL assay was capable of measuring IAA levels that were predictive of diabetes diagnosis in NOD mice and also effectively monitored the development of the underlying insulitis, we used this assay to evaluate human samples of known IA status as determined by RBA. The assay was, however, unable to distinguish between the IA⁺ and IA^-donors with many samples producing high ECL signal independent of antibody status (Figure 7). This result was not entirely unexpected as others have reported non-specific interfering factors when assessing IAA in human samples [²⁹]. As ECL-based assays have previously been successfully used in human plasma and serum samples for other analytes, we were confident that the ECL platform could still allow for the sensitive detection of human antibodies against insulin. Therefore, we developed a new assay to reduce the background signal and examined its ability to A) discriminate between samples of unknown IA status and B) distinguish non-diabetic from long-standing diabetic individuals. Figure 8 displays the design of the method utilized to measure IA in human serum or plasma. Absorbed plasma or serum samples are incubated with or without (HSD only) biotinylated-insulin, which is captured by streptavidin (SA)-coupled magnetic beads, and bound antibody is detected by a secondary antibody tagged with the Ruthenium reagent (TAG). Signal is evaluated using the M1MR Analyzer which employs a magnet to capture the bead-IA-Ig:Tag complex into an electric field for the ECL reaction [¹⁴]. The IA value obtained for each sample is calculated by dividing the average signal obtained from wells incubated with biotinylated-insulin by the average of the signal from three wells without biotinylated-insulin. Preliminary data suggested that the background could be reduced and signal enhanced in human plasma or serum samples (Figure 9A).

To confirm the specificity of the ECL signal, we performed a competition assay using non-biotinylated insulin with several samples of known IA status determined by RBA. Increasing amounts of non-biotinylated insulin incrementally reduced the IA value (Figure 9A and 9B). We next carried out ROC analysis with 25 IA negative samples and 15 IA positive samples as previously determined by RBA. We found that within this small cohort, we could accurately distinguish between the two groups with 94% sensitivity at 100% specificity and an area-under-the curve (AUC) value of 0.94 [95% confidence interval (95% CI): 0.98 to 1.01] (Figure 9C). Finally, we examined the lot-to-lot reproducibility of our reagents and determined that IA levels from the same samples were stable across two independently prepared batches of biotinylated insulin, and TAG-secondary antibody (Figure 9D).

Having established that our IA ECL assay is stable, reproducible across batches of reagents, and consistent with IA status determined by RBA, we examined whether it could distinguish between long-standing diabetic and non-diabetic individuals. Plasma samples from 59 non-diabetic donors (30 males, 29 females; mean age: 28.6 years) and serum samples from 34 long-standing diabetic donors (11 males, 21 females; mean age: 38.2 years) were evaluated using our assay with three different instruments in three separate laboratories. IA values in plasma or serum samples obtained from the same individual did not differ significantly (data not shown). The ROC analysis of the IA values revealed no statistically significant difference between the three laboratories (P > 0.5) in determining donor disease status (Figure 10A). Our assay distinguished IA levels in diabetic donors across a 3-log range while the distribution across non-diabetic donors remained tightly clustered around an IA value of 1.5 (95% CI: 1.3-1.6), enabling a sensitivity of 88% at 95% specificity (Figure 10B). Comparison of the intra-cohort rank for each sample by the individual laboratories versus the mean rank across the three laboratories revealed a strong correlation of individual rank versus mean rank (r² = 0.92, 0.87, 0.87 respectively for Lab1-Wellstat, Lab2-NIAID, Lab3-DIL) with no significant differences between laboratories (P = 0.83) (Figure 10C).

The thresholds for a positive (+) predictive IAA measurement were chosen based on IAA values with the maximum likelihood ratio calculated from the ROC analyses for the NOD mice at 7, 8-8.5, and 10 weeks of age. The number of mice that do (+) or do not (-) develop diabetes by 20 weeks of age and had IAA values greater (+) or less (-) than the threshold are displayed in each column. The IAA threshold for mice at 7, 8.5, and 10 weeks are 176, 115, and 190, respectively. Mice not tested at the specified age were considered to have a positive predictive IAA level if the IAA values were greater than the threshold on the week prior and following the specified age.

The thresholds for a positive (+) predictive IAA measurement were chosen as the 60^th percentile in the NOD mice of the specified age. Displayed in each column are numbers of mice that do (+) or do not (-) develop diabetes by 20 weeks of age and had IAA values greater (+) or less (-) than the threshold. IAA thresholds for 7, 8.5, and 10 week old mice are 47, 100, and 252, respectively. Mice not tested at the specified age were considered to have a positive predictive IAA level if the IAA values were greater than the threshold on the week prior and following the specified age.

*All information regarding radiobinding assay performance has been collected from the 2010 report by Schlosser et al. [²⁶]. †ECL, Electrochemiluminescence; ||AS95, Sensitivity at 95% specificity ¶AUC, area under the curve, calculated from the ROC curve used to assess the ability of each assay to distinguish between non-diabetic and long-standing diabetic individuals; #CI95, 95% confidence interval