Increase in Pancreatic Proinsulin and Preservation of β-Cell Mass in Autoantibody-Positive Donors Prior to Type 1 Diabetes Onset

Type 1 diabetes is defined as an autoimmune disease in which clinical symptoms arise as a result of β-cell loss. Genetic and environmental factors might render β-cells susceptible to attack by the immune system or could contribute to β-cell dysfunction (1,2). More than three decades ago, Eisenbarth and colleagues (3) described a linear loss of first-phase insulin release after intravenous glucose administration in individuals with islet-cell antibodies who were monitored for 10 years before diagnosis. However, elevations in fasting blood glucose and peak glucose during oral glucose tolerance tests were only seen in the year before onset. This sustained loss of β-cell function in individuals with prediabetes strongly correlated with the time to overt diabetes and led to Eisenbarth’s (4) landmark article in which the stages of type 1 diabetes were presented and the steady decrease in insulin secretion was linked to a linear reduction in β-cell mass that continued after diagnosis.

Although this model remained a reference for many years, new studies have suggested that β-cell mass is not lost in a linear fashion during the prediabetic phase, and a debate about the discrepancy between β-cell mass and function ensued (2). Subsequent studies have also detected a loss of glucose tolerance in the months preceding diagnosis (5,6). β-Cell dysfunction might occur early in the disease process, at the point at which the individual becomes autoantibody positive (Ab+), but an actual decline in β-cell mass might occur later. In the Diabetes Virus Detection (DiViD) study, a transient β-cell dysfunction was detected in live cells obtained at diagnosis, which improved in a nondiabetic culture milieu (7). Increasing dysfunction would prompt an increase in insulin demand (8,9), which could eventually cause a more cataclysmic decline in β-cell mass around the clinical onset of diabetes. However, the cause of the decline in function and the precise time course of events have remained largely undefined.

Studies from the Network for Pancreatic Organ Donors with Diabetes (nPOD) have recently shown that β-cell mass is not diminished in Ab+ donors and that single β-cells and islets containing insulin can be found in donors with long-standing type 1 diabetes (10). The time course from seroconversion to onset of clinical diabetes has been further characterized in longitudinal studies. After autoantibody seroconversion, 14.5% of single Ab+ and 67.9% of multiple Ab+ patients progressed to type 1 diabetes in a 10-year follow-up study in three geographically different cohorts (11). Another study also revealed that 11% of multiple Ab+ children would progress to clinical disease each year (12). However, the exact triggers and progression to clinical onset are not fully understood.

Proinsulin is an important autoantigen in type 1 diabetes in humans and mice (13) because it shapes the autoreactive CD8 T-cell repertoire (14,15). Importantly, recent studies have shown that several epitopes within its precursor (preproinsulin) and proinsulin itself are recognized by islet-infiltrating CD4 and/or CD8 T cells isolated from patients with type 1 diabetes (16–20), suggesting a potential role for this antigen in disease pathogenesis. Preproinsulin is processed into proinsulin and signal peptide (21). Only a marginal fraction of proinsulin is secreted to the circulation, but it accounts for 30–50% of the protein production in β-cells and increases in response to higher insulin demand. Because of this high metabolic demand, β-cells are prone to endoplasmic reticulum (ER) stress and proinsulin misfolding, which could lead to β-cell failure (22). ER stress may also be induced by viral infection (23), which was recently detected in the islets of Langerhans at diagnosis (24). Interestingly, Cianciaruso et al. (25) demonstrated that cytokine-induced ER stress enhances the exosomal release of proinsulin. Several reports present evidence of high circulating proinsulin and proinsulin intermediates, with or without accompanying hyperglycemia, in patients at risk for developing the disease and after diagnosis (26,27). Proinsulin and proinsulin–to–C-peptide (PI-to-C) ratios in combination with autoantibody concentration have been suggested as potential biomarkers for type 1 diabetes, capable of identifying with high sensitivity individuals at risk for developing disease 1 to 40 months before clinical onset (28,29). However, the link between proinsulin levels in serum and their content in the pancreas in β-cells themselves has not been investigated in human specimens in situ.

The objective of this study was to add refinement to the classic model of linear β-cell loss and to fill an important gap in our understanding of the prediabetic phase in human type 1 diabetes. We studied, for the first time in the human pancreas, the distribution of proinsulin in single- and double-Ab+ individuals preceding the onset of disease as well as in patients with recent-onset type 1 diabetes and correlated it with loss of insulin content, β-cell mass, and PI area–to–insulin (PI-to-INS) area ratio. These studies underline the need for development of biomarkers as well as for preventive therapies focusing on normalizing β-cell dysfunction during the prediabetic stage.

Human pancreas sections were collected from cadaveric organ donors through nPOD. Six micrometer formalin-fixed paraffin-embedded sections from the head, body, and tail of the pancreas were obtained from control donors without diabetes who were single Ab+ (n = 8), double Ab+ (n = 5), and Ab− (n = 9), and from one donor at onset of type 1 diabetes. In addition, 4-μm formalin-fixed paraffin-embedded sections from living donors with type 1 diabetes (n = 6) were obtained through the DiViD study by tail resection (30). Overall, 22 sections from the head, 22 from the body, and 29 from the tail of the pancreas were analyzed (n = 73 sections). Table 1 summarizes the demographic information for each group. Detailed donor information can be found in Supplementary Table 1. All experimental procedures were approved by the La Jolla Institute for Allergy and Immunology Institutional Review Board, approved protocol number DI3-054-0216. For the DiViD study, participants provided written informed consent, and more details can be found in Krogvold et al. (30).

Pancreas sections were stained for insulin, proinsulin, and glucagon after a standard triple indirect immunofluorescence staining. After deparaffinization and rehydration in descending ethanol concentrations, sections were exposed to heat-based antigen retrieval (citrate buffer). Staining was performed using a polyclonal guinea pig anti-insulin antibody (1:500; Dako, Carpinteria, CA), monoclonal mouse anti-proinsulin (DSHB clone GS-9A8, 1:50), and monoclonal mouse anti-glucagon (clone K79bB10, 1:300; Abcam) conjugated in-house to Alexa Fluor 647. Secondary antibodies included F(ab′)₂ fragment of goat anti-guinea pig IgG conjugated to Alexa Fluor 488 (1:800; Jackson ImmunoResearch) and goat anti-mouse IgG (H+L) conjugated to Alexa Fluor 555 (1:1,000; Life Technologies, Grand Island, NY) incubated at room temperature for 30 min. Sections were counterstained with Hoechst (1:400; Life Technologies) for 10 min and then mounted with ProLong Gold Antifade Mountant (Life Technologies).

Sections were scanned with an Axio Scan Z.1 slide scanner (Carl Zeiss Microscopy, Thornwood, NY) using a fluorescent Orca Flash 4.0 v2 (HXP 120 V lamp) camera with a ×20/0.8 NA objective for immunofluorescence. Images were acquired with ZEN2 software slidescan module. Acquisition settings were kept constant between specimens to allow for quantitative comparison between samples. ZEN2 software blue edition was used to process the images before analysis. Lower and upper thresholds were defined for each channel. For insulin and proinsulin, similar thresholds were set for comparison. Then, whole-tissue section images were exported and reduced at 30% or 45%, depending on their size, into .tiff files for automatic software analysis. Custom macros were developed to measure tissue area (macro 1) and islet size and count (macro 2) and calculate the percentage of insulin, proinsulin, and glucagon areas (macro 3; only cytoplasmic staining above background levels was measured). To classify and calculate the number of α- and β-cells, the image of the whole tissue section was exported into 20 × 20 .tiff files to improve resolution and then processed by a different custom macro (macro 4) (Supplementary Fig. 1). All of the macros were developed for Fiji, an image-processing software program developed for ImageJ (National Institutes of Health). R software was used to systematically analyze the data (R is available as free software under the terms of the Free Software Foundation’s GNU General Public License in source code form). In addition, three cases per group were analyzed by high-resolution confocal microscopy using an LSM 880 confocal microscope with Airyscan technology (Carl Zeiss, Jena, Germany) and a ×63 objective.

Differences between group pairs were analyzed with a Student t test or Mann-Whitney test. Group differences were analyzed using one-way ANOVA, followed by a Holm-Šídák multiple comparisons test or Kruskal-Wallis test, followed by a Dunn multiple comparisons test. To assess the plausibility of using the PI-to-INS area ratio as a discriminator for disease status, a logistic regression was fitted to the data using R version 3.3.0 (9).

The scores for this model were calculated as the distance from each point (plotted using their percentage of proinsulin and insulin area as coordinates) to the line proinsulin = insulin (i.e., y = x). Briefly:

The significance of the overall model was calculated by comparing it with a model with just the intercept (i.e., a null model). Besides the logistic regression, a receiver operating characteristic curve analysis using the R package ROCR version 1.0-7 (10) was performed to investigate the predictive power of the model.

Statistical analysis was performed using GraphPad Prism version 6 (GraphPad Software, La Jolla, CA). Data in graphs and tables are presented as mean ± SD unless otherwise indicated. Findings were assumed statistically significant at P ≤ 0.05.

We systematically measured insulin, proinsulin, and glucagon staining from the head, body, and tail regions of pancreatic tissue sections. Interestingly, there was a small increase in total insulin area in Ab+ donors that did not reach statistical significance (Fig. 1A–C, left panel). Conversely, a significant increase in the proinsulin area was observed in the head (54%, P = 0.0233), body (57%, P = 0.0143), and tail (60%, P = 0.0087) regions in Ab+ individuals compared with control sections (Fig. 1A–C, central panel). Lastly, there were no major differences in glucagon area between both groups for any of the regions (Fig. 1A–C, right panel). To better understand whether this was caused by a shift in the subcellular localization of proinsulin and/or by an increase in proinsulin content, a super high–resolution confocal microscope with Airyscan technology was used. Proinsulin in control donors was mainly localized close to the nucleus, with a staining pattern consistent with the Golgi apparatus, and was minimally present in other compartments. In multiple Ab+ donors, proinsulin was more widely localized to the juxtanuclear region (Golgi) and vesicular compartment, confirming a change in subcellular localization (Fig. 2 and Supplementary Fig. 2).

The combination of insulin area from the head, body, and tail sections of the pancreas, normalized by the size of the respective tissues, was multiplied by the total weight of the pancreas. Mean β-cell mass was almost identical in control and Ab+ donors (245.0 ± 60.5 mg control donors vs. 267.7 ± 80.2 mg Ab+) (Fig. 3A). However, when proinsulin area was used as reference instead of insulin, a significant increase in β-cell mass in the Ab+ donor group was seen (187.4 ± 49.7 mg control donors vs. 266.9 ± 99.7 mg Ab+) (Fig. 3B). Lastly, α-cell mass, calculated as the adjusted percentage of glucagon area multiplied by the total weight of the pancreas, was similar for Ab+ and control groups (121.5 ± 42.6 mg control donors vs. 141.2 ± 65.7 mg Ab+) (Fig. 3C). β-Cell mass did not correlate with age (r = −0.018; P = 0.9341), BMI (r = 0.377; P = 0.0833), or time in the intensive care unit (r = −0.056; P = 0.8250) (data not shown).

To study the direct relation between insulin and proinsulin area in the pancreas, the PI-to-INS area ratio was calculated for each section, region, and donor. Interestingly, the ratio was increased for Ab+ donors compared with control donors for the head (38%; P = 0.0031), body (40%; P = 0.0005), and tail (32%; P = 0.0004) regions of the pancreas (Fig. 4A). To evaluate the use of the area ratio as a potential indicator of β-cell dysfunction and to directly compare Ab+ donors with control donors, an arbitrary reference value of 1:1 (proinsulin area–to–insulin area) was chosen and graphically represented as a line to separate control donors from “at-risk” donors. The distance from the donor’s area values to the line was used as a score to estimate the risk of disease and to classify and distinguish control donors from Ab+ donors (Fig. 4B). Then, a logistic regression was performed, and the area under the curve (AUC) was calculated (Fig. 4C). High AUC values, significant coefficients, and a good model fit were obtained for the head (AUC 0.85, coefficient P = 0.0348, model P = 0.00225), body (AUC 0.87, coefficient P = 0.0213, model P = 0.00073), and tail (AUC 0.9, coefficient P = 0.0196, model P = 0.00045), indicating that the pancreatic PI-to-INS area ratio could identify individuals at risk for developing disease.

To see whether the insulin area, the proinsulin area, or the PI-to-INS area ratio could correlate with the risk of developing type 1 diabetes, a risk index was calculated for all of the donors based on their age, HLA, and autoantibody status (Supplementary Fig. 3). To calculate the risk index, a score of 0 (low risk), 1 (medium risk), or 2 (high risk) was assigned as follows based on the risk of developing type 1 diabetes:

• Age: >40 (0), 30–40 (1), 0–30 (2).

• HLA: no risk alleles (0), DR4 or DR3 only (1), DR4 DQ8 or DR3 DQ2 or DQ8 (2).

• Autoantibodies: 0 (0), 1 (1), 2 (2).

The risk index was then calculated as the sum of the values obtained in each category for each donor (range 0 [minimum] to 6 [maximum]). A strong positive correlation was found between the risk index and the PI-to-INS area ratio (Supplementary Fig. 3, right panel), but there was a weak correlation with proinsulin area (central panel) and no correlation with insulin area (left panel).

To investigate the timing of the increase in proinsulin area and the inversion of the PI-to-INS area ratio when compared with control donors, pancreas tissue from a set of recently diagnosed patients with type 1 diabetes (0–9 weeks postdiagnosis) was studied (Fig. 5). Only the tail region was analyzed, because all samples but one were obtained by tail resection, as previously described (30). (One section from a donor with type 1 diabetes at onset obtained from nPOD was also included for comparison.) Patients with type 1 diabetes were compared with control donors, and single Ab+ and double Ab+ donors separately (Fig. 5A). The insulin and proinsulin areas were lower than in the rest of the groups owing to a reduction in the islets that contained insulin, whereas glucagon area was increased (Fig. 5C, D, and F). Interestingly, the PI-to-INS area ratio was increased in patients with type 1 diabetes and almost identical to that of at-risk double Ab+ donors (0.77 ± 0.13 control donors vs. 0.99 ± 0.15 single Ab+ vs. 1.06 ± 0.10 double Ab+ vs. 1.07 ± 0.18 type 1 diabetes) (Fig. 5E, left panel).

Small morphological alterations as well as in differences in islet number, size, and distribution might occur many years before diagnosis in individuals at risk for developing disease. First, the number of islets was counted, and the total area of the tissue section was measured (see research design and methods for details and Supplementary Fig. 1). Total islet number and tissue size were variable across the pancreas but were similar in control donors and Ab+ donors (data not shown). There were no major differences in mean islet density between donors without diabetes (head 2.5 ± 0.5 vs. body 1.8 ± 0.2 vs. tail 2.8 ± 0.7 islets/mm²) and Ab+ donors (head 2.6 ± 0.7 vs. body 2.1 ± 0.7 vs. tail 3.1 ± 0.9 islets/mm²) (Fig. 6A). Next, the islet size distribution was analyzed. Significant differences were found between control donors without diabetes and double Ab+ donors for the head, body, and tail (Fig. 6B and Supplementary Fig. 4). Lastly, sections from patients with type 1 diabetes were analyzed. Larger islets were found in these sections (Fig. 5A and Supplementary Fig. 4, tail region only), but the islet density was lower (Fig. 5B), as expected. Interestingly, the tail of the pancreas presented a distinct islet distribution compared with the head and body regions for all groups, with a predominance of large islets and fewer small islets (Fig. 6C and Supplementary Fig. 4), confirming important regional differences within the pancreas.

The number of insulin- and glucagon-expressing cells per islet was counted, and the ratio between both cell populations was calculated (Supplementary Fig. 5A). Although no significant differences were found, double Ab+ donors presented higher β-cell–to–α-cell ratios in the head, body, and tail regions (head median = 3.8 control vs. 5.3 single Ab+ vs. 6.4 double Ab+/body median = 4.0 control vs. 3.8 single Ab+ vs. 7.1 double Ab+/tail median = 3.6 control vs. 3.8 single Ab+ vs. 4.9 double Ab+). Very heterogeneous cell distribution patterns were found among the Ab+ donors, with some individuals having a higher abundance of α-cells per islet (Supplementary Fig. 5B, central panel) and others with a clear predominance of β-cells (Supplementary Fig. 5B, right panel) compared with control sections (Supplementary Fig. 5B, left panel).

Patients with type 1 diabetes had a significantly lower β-cell–to–α-cell ratio (median = 1.1 type 1 diabetes, tail region only) (Fig. 5E, right panel). Lastly, the percentage of islets containing only β-cells was calculated, which again was similar for the control (24.9% ± 9.0), single (28.8% ± 9.9), and double Ab+ group (31.3% ± 16.8) and lower for the group with type 1 diabetes (3.5% ± 4.9) (Fig. 5G, left panel). Islets with only α-cells were not common (control 3.5% ± 3.2 vs. single Ab+ 2.8% ± 0.8 vs. double Ab+ 5.2% ± 7.7). Only one double Ab+ donor (#6267, 18.9%) and the donors with type 1 diabetes (52.2% ± 22.4) presented evident increases in the percentage of islets containing only α-cells compared with the rest of the donor groups (Fig. 5G, right panel).

The recent access to human pancreata for research purposes and the subsequent histological studies have filled important gaps in our understanding of pancreatic pathology (10,31–33). However, many fundamental questions remain to be answered. In this study, we aimed to fully characterize β- and α-cells, investigating insulin, proinsulin, and glucagon content and distribution across the head, body, and tail regions of the human pancreas in healthy individuals as well as in single Ab+ and multiple Ab+ donors. In addition, we had access to a very unique subset of pancreatic tissues from living patients with recent-onset type 1 diabetes who participated in the DiViD study in which a small piece of the tail of the pancreas was surgically removed immediately in the months after diagnosis (30). Our study is the first to show an increase in pancreatic proinsulin area in individuals with prediabetes and confirms the value of the PI-to-INS area ratio as an indicator of early β-cell dysfunction. In addition, we confirm previous findings that β-cell mass is not reduced in Ab+ individuals. We therefore add critical refinement to the original model of β-cell loss described by Eisenbarth (4) three decades ago. We did not find any significant differences in the overall insulin-positive area between healthy control donors and Ab+ donors to support a reduction in insulin content long before the onset of disease. Moreover, β-cell mass was essentially identical in both groups, demonstrating that β-cells are preserved in Ab+ individuals until shortly before diagnosis and in agreement with previous publications that reported no differences in β-cell mass between donors without diabetes and Ab+ donors (34–36).

Increases in the proinsulin area were found in some of the single Ab+ pancreata, which could explain why the first-phase insulin response is already abnormal in some of these individuals (4). Proinsulin primarily accumulates in the Golgi apparatus in resting β-cells and is further processed to insulin and C-peptide in immature secretory granules (37). Our data show significant increases in the proinsulin area in Ab+ individuals in the head, body, and tail regions of the pancreas. Using a high-resolution confocal microscope to study a subset of samples, we observed that the increase in the proinsulin area was partially caused by a shift in proinsulin subcellular localization from the Golgi area in control individuals (juxtanuclear) to secretory vesicles in multiple Ab+ individuals (cytosolic). Although an increase in proinsulin area suggests an increase in the amount of proinsulin protein, additional experiments are needed to accurately measure the protein content. Nevertheless, this increase in area could be caused by an increase in proinsulin production (to cover insulin demand) or by a defect in the processing and maturation of the existing pool of proinsulin to insulin, which is consequently released to the circulation. This is in agreement with studies that have detected proinsulinemia in patients with prediabetes and in patients with type 1 diabetes (38). In addition, an increase in proinsulin synthesis could lead to ER stress, protein misfolding, and loss of glucose-stimulated insulin secretion (39). Other extrinsic factors, for example, recurrent autoimmune attacks or other forms of metabolic stress (40,41), may affect β-cell function and proinsulin processing early in the disease process of prediabetes. In normal β-cells, up to 20% of proinsulin can be misfolded. However, only under pathological conditions and ER dysfunction, and once a certain level of misfolded proinsulin has been accumulated, do β-cell toxicity and death occur (22).

In a case-control study analysis by Sims et al. (42), an elevation of the PI-to-C ratio preceded disease onset in high-risk subjects and could be detected at least 12 months before diagnosis. This is in agreement with the data presented here, in which the pancreatic PI-to-INS area ratio was increased in Ab+ individuals. Among the single Ab+ group, two donors (#6170 and #6184) consistently presented an elevated PI-to-INS area ratio. This suggests that β-cells in these two individuals could have had a functional defect. Conversely, all but one double Ab+ donor (#6080) presented an increased ratio in at least two of the three regions of the pancreas, supporting the notion of an early functional defect in β-cells during the prediabetic phase and before diagnosis that does not necessarily imply β-cell loss. Donor #6080 was positive for GAD and insulin Abs but was 69 years old and therefore unlikely to have developed the disease. In conclusion, our in situ results correlate well with those found in serum and confirm the potential of monitoring PI-to-C or PI-to-INS area ratios as indicators of β-cell dysfunction.

The samples obtained from living individuals with recent-onset type 1 diabetes through the DiViD study (30) showed an expected significant decrease in the insulin and proinsulin area resulting from a reduced number of insulin-containing islets; however, PI-to-INS area ratios were elevated in four of six patients, suggesting that insulin therapy at onset might not fully alleviate the dysfunctional β-cells. Our findings indicate that potential therapies should target β-cells early before onset, when they still have the ability to be functionally rescued and when the immune system has not been fully activated.

We further characterized islet distribution and composition to study whether small differences in islet pathology could be observed early in the prediabetic phase. The tail of the pancreas contained larger islets and higher islet density in all donor groups, which could point to important developmental and architectural differences in vascularization and innervation of the islets in this region (43,44). Interestingly, subtle differences were found in double Ab+ at-risk individuals, in whom larger islets were found in the body and tail region of the pancreas compared with control individuals. Moreover, this tendency was accentuated in individuals with recently diagnosed type 1 diabetes, where even larger islets could be found in the tail region, many of them containing only α-cells. This is in agreement with previous studies in diabetic NOD mice in which small islets were preferentially lost and a subsequent expansion of large islets was seen (45). An increase in the size of the islet and number of β-cells in multiple Ab+ donors could be a compensatory mechanism caused by the chronic increase in insulin demand.

Our observations in the pancreas of individuals at risk for developing type 1 diabetes point to important and very early (at the first sign of autoimmunity) pathological changes in β-cells without an evident loss of β-cell mass. This confirms the potential benefit of estimating the PI-to-C or PI-to-INS area ratios in Ab+ individuals to identify those patients at high risk at a stage where β-cells might still respond to preventive therapies and to enroll patients in clinical trials at a point in the disease when they would benefit the most. Whether a consequence of an increase in insulin demand, a primary cellular defect, or a change in the orchestrated interplay between the immune system and the islet, the higher accumulation of proinsulin in β-cells without a reduction in insulin content might ultimately lead to β-cell exhaustion and death, with the subsequent release of β-cell antigens that initiate the autoimmune process. Whether type 1 diabetes is a primary autoimmune disease or autoimmunity is secondary to metabolic or functional defects that render β-cells susceptible to autoimmune destruction remains unknown. Future studies on proinsulin and insulin dynamics as well as a better characterization of β-cells themselves will provide the necessary answers to fully understand the pathological changes that precede the clinical onset of type 1 diabetes.

Acknowledgments. This research was performed with the support of nPOD, a collaborative type 1 diabetes research project sponsored by JDRF. Organ procurement organizations, partnering with nPOD to provide research resources, are listed at www.jdrfnpod.org/our-partners.php. The proinsulin antibody clone GS-9A8, developed by O.D. Madsen (Hagedorn Research Institute, Gentofte, Denmark), was obtained from the Developmental Studies Hybridoma Bank, created by the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health and maintained at The University of Iowa, Department of Biology, Iowa City, IA. The authors thank the following individuals from La Jolla Institute for Allergy and Immunology: Zbigniew Mikulsky and Bill Kiosses, for their help with image acquisition and analysis, and Priscilla Colby, for administrative assistance.

Funding. This study was supported by National Institutes of Health/National Institute of Allergy and Infectious Diseases grant R01-AI-092453-03.

Duality of Interest. M.G.v.H. is an employee of Novo Nordisk. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. T.R.-C. performed and designed experiments, analyzed and interpreted data, and wrote the manuscript. J.Z.-G. designed the custom macros developed for computer-assisted software analysis, analyzed and interpreted data using bioinformatics tools, and helped with statistical analysis. N.A. performed experiments. E.C. and Y.L. helped with analysis. L.K. and K.D.-J. collected patient material and revised the manuscript. K.D.-J. is Principal Investigator of the DiViD study. M.G.v.H. designed experiments, interpreted data, and wrote the manuscript. M.G.v.H. 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. Preliminary results from this study were presented at the Immunology of Diabetes Society 14th International Congress, Munich, Germany, 12–16 April 2015; the JDRF nPOD 8th Annual Scientific Meeting, Miami, FL, 22–25 February 2016; the 52nd European Association for the Study of Diabetes (EASD) Annual Meeting, Munich, Germany, 12–16 September 2016; and the 19th Inflammation Research Association (IRA) International Meeting, San Diego, CA, 20–21 September 2016.