Antibodies against insulin measured by electrochemiluminescence predicts insulitis severity and disease onset in non-obese diabetic mice and can distinguish human type 1 diabetes status

Autoimmunity occurs when the physiologic mechanisms of immune tolerance fail to curtail aberrant activation and effector activity of self-reactive lymphocytes [1, 2]. Type 1 diabetes (T1D) is an autoimmune disease wherein insulin deficiency results from the destruction of insulin-secreting β cells in the pancreas by infiltrating T cells and other cells of the immune system [3]. As a consequence, individuals with diabetes depend on administration of exogenous insulin and are susceptible in the longer term to complications including retinopathy, nephropathy, and cardiovascular disease [3]. The diagnosis and etiology of T1D appears to be widely variable [4], with poorly defined environmental factors acting upon underlying genetic susceptibility to cause disease in humans [5]. Clinical manifestations of T1D occur once a substantial proportion of the insulin-producing β cells are destroyed [6]. The development of autoantibodies against multiple islet cell antigens is a well-established feature of T1D [7, 8]. Although not an active component of the disease process itself, the presence of circulating autoantibodies to two or more islet antigens, namely insulin (IAA), glutamic acid decarboxylase (GADA), islet antigen 2 (IA-2A), and zinc transporter-8 (ZnT8A), is highly predictive when combined with a family history of the disease or genetic risk [7‐13]. IAA are usually the first islet autoantibodies to appear in prediabetic children [14‐16], making it one of the earliest measurable signs of the autoimmune process. Furthermore, evidence suggests that mean IAA levels, but not of IA-2A or GADA, can serve as a predictive marker of diagnosis [17‐19].

In the non-obese diabetic (NOD) mouse, one of the most extensively studied animal models of T1D, it has been reported that IAA levels correlate with both age of disease onset [15, 20] and insulitis across mice in a strain-dependent manner [21]. NOD mice spontaneously develop autoimmune diabetes that shares numerous characteristics with the human form of the disease. In both humans and NOD mice, multiple genetic loci contribute to diabetes susceptibility with the MHC locus being the most prominent susceptibility locus [22]. Typically, leukocytic infiltration of the islets begins around 4 weeks of age in the NOD mouse. This slowly progresses to more severe insulitis with beta cell destruction and ultimately results in frank diabetes including glucose intolerance between 12-16 weeks of age [23]. Approximately 60-80% of the females and 20-30% of the males eventually develop diabetes by 30 weeks of age [24]. No evidence has yet been reported that the levels of IAA in an individual mouse predict its specific risk for T1D onset and insulitis.

The radiobinding assay (RBA) is currently the most widely used method for assessing autoantibody levels including IAA, as enzyme-linked immunosorbent assays (ELISAs) have not equaled or surpassed the conventional RBA in performance for detecting IAA [25‐27]. Although the RBA is the gold standard for measuring IAA, the RBA approach possesses several drawbacks including: i) a requirement for newly synthesized radiolabeled insulin for each set of assays; ii) the need to generate a new standard curve using a confirmed IAA sample; iii) a lengthy procedure spanning numerous steps over multiple days; iv) an inability to distinguish between different IAA immunoglobulin subtypes; v) non-specific interference by soluble factors including anti-bovine serum albumin (BSA) antibodies; and, most importantly, vi) inconsistent results across laboratories worldwide [26, 28‐30]. A more rapid, non-radioactive, and more reproducible assay for measuring IAA to facilitate T1D prediction is urgently needed.

To progress toward a better analytic tool for autoantibody detection in individuals genetically at-risk for T1D development, we investigated electrochemiluminescence (ECL)-based assays for measuring IAA in mice and IA in humans. Unlike mouse IAA, human IA detected in samples from insulin-treated diabetic individuals are likely to consist of both IAA produced during the autoimmune response against endogenous insulin as well as antibodies arising as a consequence of repeated exogenous insulin administration. IA are detectable by both RBA and ELISA, but it has been reported that IA detected by ELISAs using plate-bound insulin are less relevant to the autoimmune process [25, 31]. Nevertheless, plasma or serum samples from established T1D individuals are readily available and, therefore, we have examined IA in insulin-treated diabetic individuals as the first step toward developing an ECL assay for IAA detection in prediabetic samples. ECL has been used in clinical laboratory tests because it is fast, easy to perform, non-radioactive, and highly sensitive with a broad dynamic range [32]. ECL is currently used to detect a wide range of analytes [33, 34] including biomarkers for Alzheimer's disease [35], meningitis [36], and HIV infection [37]. Several studies have also shown that ECL can outperform conventional ELISAs in both speed and sensitivity [38, 39].

In the NOD mouse model, we demonstrate that our IAA ECL method is highly sensitive and that IAA levels show a strong correlation with insulitis and are predictive of diabetes on a mouse-to-mouse basis. This simple murine IAA ECL assay can be completed in as little as 2-3 hours, requires only a few microliters of serum, and has a broad dynamic range with little background. The mouse assay format, however, could not be directly applied to use with human samples, and, therefore, we developed a new assay using the same ECL platform to detect human IA. This human IA ECL method could accurately and reproducibly distinguish samples from diabetic and non-diabetic individuals. These data indicate that the ECL approach described here has promise for further development into a clinical test for T1D risk using samples from prediabetic and newly-diagnosed diabetic individuals who have not been treated with exogenous insulin.

The primacy of the immune response to insulin and its precursors in the etiology of T1D is evident from the fact that IAA is often the earliest autoantibody to appear in children destined to be diagnosed with T1D [16, 44]; in contrast to other autoantibodies that appear before diagnosis of T1D, mean IAA levels predict age-at-diagnosis [17‐19]; and a polymorphism just 5' of the structural gene, insulin, regulates its transcription in thymus and alters risk of T1D [45]. In the NOD mouse model the autoimmune response to insulin is essential for development of the disease [46]. Therefore, optimizing the testing of IAA is highly desirable. In the present study, we describe novel methods of measuring IAA in prediabetic mice and IA in diabetic human samples that overcome some of the limitations of RBAs. The use of non-radioactive insulin and stable TAG-secondary antibody enabled us to utilize the same materials in three separate laboratories without reagent deterioration over the course of multiple assays. This is a significant improvement over the conventional RBA method of detection, which requires radioactive ¹²⁵I-insulin that must be freshly synthesized before each series of assays [30]. Clarification of human samples with SA-beads to remove non-specifically bound materials significantly decreased background while maintaining specific signal. This clearance step overcomes challenges present in the RBA where assay components including BSA [29] reduce the specificity of that IA assay format. The ECL assay requires the same sample volume per well used in RBA but does not require the generation of a standard curve using known positive samples (Table 3) [30]. Furthermore, the combined incubation time achieves a > 30-fold reduction in assay time compared to the RBA [30] and the stable IA values we obtain from multiple batches of reagents demonstrate that the assay is suitable for longitudinal studies where reproducible, reliable, comparable measurements are critical.

In our investigation of NOD mice, we have shown that the murine IAA ECL assay is extremely sensitive, highly correlative with the degree of insulitis, and is predictive of later development of diabetes in individual mice. The dynamic range can span up to five log units (Figure 2B & 10B), which is about three logs greater than the capability of most chromogenic ELISAs. This is enabled by the Ru ECL species that is recycled rather than consumed during the generation of the ECL signal, allowing each ruthenium tag to emit multiple photons which further amplifies the signal [43]. Our ECL technique uses biotinylated-insulin to bind to streptavidin-coated magnetic beads. The use of the streptavidin-biotin system for linkage of antigens in immunosorbent assays has been shown to be superior in preserving antigen conformation and antibody-binding capacity to the adsorption methods of immobilizing antigens used in most ELISAs [47]. The assay also uses very small amounts of serum and is much faster than available RBAs and ELISAs [27] with results obtained in 2.5 hours.

The value of this ECL assay in measuring IAA as a marker of β cell damage is demonstrated by our finding that IAA ECL values from NOD mice directly correlate with the degree of insulitis (Figure 4). The observed decline in IAA levels late in the disease progression (Figure 3) may be due to the decrease in available autoantigen as the beta cells are destroyed [13]. Thus, the correlation between IAA and insulitis would be expected to be more profound when IAA levels are on the rise, which for NOD mice is before 10 weeks of age. In some mice the IAA levels may have peaked prior to the final time point collected; therefore, the maximum IAA level measured over the time course examined for each mouse may be a more accurate estimate of the underlying insulitis severity. This is supported by the stronger correlation between IAA and insulitis when analyzing maximum IAA levels measured for each mouse rather than the IAA levels at the final time point when insulitis was assessed (Figure 4B and 4C). The strong correlation of IAA and insulitis in NOD mice supports the ability of our murine IAA ECL assay to measure disease progression on a mouse-to-mouse basis.

Beyond evaluating the course of the disease, the murine IAA ECL assay was able to successfully predict which 8-10 week old NOD mice would later go on to develop diabetes by 20 weeks of age. The high predictive value achievable by our method (approximately 90%, Tables 1 and 2) effectively allows a researcher to consider an 8-week-old mouse to be inevitably diabetic, eliminating the need to wait an additional 12 or more weeks to determine whether disease will develop. Accurate prediction of T1D onset at an early age in mice will facilitate therapeutic experiments that seek to prevent disease development. Our determined threshold IAA values and the optimal age for diabetes-predictive IAA measurement may be specific to our NOD colony since diabetes incidence and age of onset differs between NOD colonies [48, 49]. However, colony-dependent values could be established and utilized across time due to the stability of signal in our assay format. These features could enable researchers to utilize a murine IAA ECL assay to reliably monitor the autoimmune attack on β cells.

Our ongoing work is aimed at developing an assay sensitive and specific enough to diagnose insulitis based on a positive test, or series of tests, for IAA and its levels and to facilitate prediction of the progression to T1D in a platform that can be used routinely and robustly in any clinical laboratory. Future intervention or prevention trials in T1D could target the early anti-preproinsulin autoimmune response, not least because it is probably the major etiological trigger of the disease in children.

In our studies thus far, we have evaluated IA in plasma and serum samples from long-standing insulin-treated diabetic individuals rather than from prediabetic or newly-diagnosed individuals that have not been treated with insulin. Several reports have identified quantitative and qualitative differences in the IAA detectable in insulin-untreated diabetic individuals compared to IA present in samples from individuals receiving exogenous insulin including varying affinities, epitope preference, and usage of subtypes of IgG isotype IAA [50‐54]. However, each of these studies utilized only a small number (< 15) of samples, were conducted before the widespread use of human recombinant insulin and rapid-acting analogs, and did not examined the change in affinity, isotype utilization, or epitope specificity of antibodies before and after the onset of exogenous insulin therapy in the same individuals. Therefore, it has not been definitively demonstrated that the antibodies examined in insulin-treated diabetic individuals are exclusively or predominantly generated de novo in response to exogenous antigen administration. Moreover, it is likely that the widespread detection of IA in insulin-treated diabetic individuals is due, at least in part, to anamnestic responses of memory B cells and plasma cells generated during the insulitic phase of the autoimmune attack on pancreatic β cells. Consequently, IAA levels would be predicted to decrease as insulin levels decline over the course of disease progression as we and others have observed in the NOD mouse model and prediabetic human samples (Figure 3B) [15, 19]. Therefore, before our assay can be used to examine IAA in prediabetic individuals and to dissect the complexity of the immune response to the renewed presence of autoantigen in the form of exogenous insulin therapy, it will be critical to enhance the sensitivity of our assay while maintaining a very high specificity. We are presently exploring ways to achieve the goal of increased sensitivity for human IAA detection and T1D prediction in prediabetic samples. Our ongoing efforts include exploitation of the regenerative photon emission enabled by ECL technology and the incorporation a competition assay for samples that are near the cutoff threshold to ensure the detection of insulin-specific IgG. We also plan to incorporate multiple autoantibody testing into our ECL assay to improve identification of prediabetic individuals in the general population, including IA-2A and GADA [7, 8, 12, 55, 56].

Based on these results we are optimizing our human IA assay for the detection of IAA in prediabetic individuals, with the ultimate goal of facilitating intervention early in the autoimmune process in prediabetic individuals to limit or prevent β cell loss and delay T1D diagnosis.