Identification of Novel Disease-Relevant Genes and Pathways in the Pathogenesis of Type 1 Diabetes: A Potential Defect in Pancreatic Iron Homeostasis Free

Type 1 diabetes (T1D) results from the gradual autoimmune destruction of pancreatic β-cells. The etiology of this disease is complex and involves a strong genetic component and a number of possible environmental triggers. While multiple pathways have been implicated in disease pathogenesis, the exact mechanisms and timing of their involvement are poorly understood due, in part, to the difficulty in obtaining relevant samples from subjects with pre-T1D.

Researchers have used the well-established NOD mouse model to study T1D. NOD mice express a number of susceptibility genes that contribute to their permissive genetic background, including the H2^g7 MHC that is similar to DQ8, a major T1D disease–associated HLA allele. The non–disease-prone congenic NOD.B10 express the H2^b MHC haplotype and are otherwise isogenic to NOD mice outside of the MHC complex. As they do not develop disease, they are suitable controls for translating our findings to human T1D, in which HLA alleles are the strongest genetic/serologic predictor of disease risk. Using NOD versus NOD.B10 mice and differential gene expression analysis, we have previously identified several genes and pathways that are relevant to the pathogenesis of T1D (1–3).

Female NOD mice develop autoimmune diabetes in a spontaneous and highly penetrant manner with a predictable course: T-cell initiation in the pancreatic lymph node can be seen as early as 10 days of age, peri-insulitis is seen starting at 4 weeks of age, and the onset of destructive insulitis occurs at ∼12 weeks of age, followed by β-cell destruction and resultant hyperglycemia (3–6). These distinct phases of disease are reflected in flow cytometric, transcriptomic, and proteomic analysis of NOD tissues over time (7–9).

We propose that similar pathways as those seen in NOD disease may be involved in the development of T1D. Degradation of the islet extracellular matrix (ECM), enhanced glucagon and insulin secretion, and hyperactivity of β-cells have been observed in the islets of patients with prediabetes as well as in NOD mice (10–16). In this study, we have identified genes that may contribute to pathophysiologically relevant pathways over the course of T1D development. Downregulation of genes involved in ECM formation was seen as early as embryonic day 17 (E17) in the pancreata of embryonic NOD mice. In our current study, we identified a significant number of genes involved in ECM development, as well as glucagon and insulin signaling, that were changed as peri-insulitic lesions developed. These include an upregulation of chemosensory olfactory receptor genes. Later, during the period of destructive insulitis, we observed downregulation of genes involved in antimicrobial defense as well as iron homeostasis.

A loss of pancreatic iron homeostasis may be significant in the pathogenesis of T1D, as a number of epidemiological studies have shown an association between high iron exposure and the development of T1D. Use of iron supplements during prenatal development, feeding with iron-fortified formula during infancy, and high red meat consumption during childhood have all been shown to significantly increase the risk of developing T1D (17–20). The mechanisms involved are unclear. Our data suggest that increased iron exposure combined with a loss in pancreatic iron homeostasis may lead to the accumulation of iron in β-cells and resultant oxidative stress. This pathway, along with compromised ECM development, β-cell hyperactivity, and a loss of antimicrobial defense mechanisms, likely contribute to the autoimmune destruction of β-cells. It is possible that these pathways could be targeted at different stages of disease development to prevent or delay the onset of hyperglycemia in T1D.

Female NOD/LtJ (NOD), NOD.B10Sn-H2^b/J (NOD.B10), NOD.CB17-Prkdc^scid/J (NOD.SCID), and NOD.Cg-Tg(TcraBDC2.5)1DoiTg(TcrbBDC2.5)Doi/DoiJ (NOD.BDC2.5) mice were purchased from The Jackson Laboratory (Bar Harbor, ME) and bred at the Stanford School of Medicine animal facility according to institutional guidelines under approved protocols. Based on phenotyping data from The Jackson Laboratory, 80% of NOD mice are expected to develop hyperglycemia by 25 weeks of age. Animals were fed Prolab RMH-5P04 (LabDiet, St. Louis, MO), unless otherwise specified.

NOD mice were fed standard diets, Prolab RMH-5P04 (Standard 5P04; LabDiet) or Teklad 2018 SX (Standard 2018; Envigo, Indianapolis, IN), or custom diets formulated using Teklad 2018 SX, supplemented with an additional 200 ppm (2018 + 200) or 850 ppm (2018 + 850) of iron sulfate. Blood glucose and weights were monitored weekly, and mice were considered hyperglycemic if blood glucose exceeded 250 mg/dL over 2 consecutive weeks. The iron content of each diet was measured by Minnesota Valley Testing Laboratories (New Ulm, MN): Standard 5P04, 369 ppm; Standard 2018, 176 ppm; Custom 2018 + 200, 378 ppm; and Custom 2018 + 850, 1,013 ppm. Mice can maintain normal growth on diets containing as little as 50 ppm iron (i.e., AIN-93G standard mouse chow) (21). Thus, the 2018, 2018 + 200 ppm, and 2018 + 850 ppm diets contain ∼3.5-fold, 7.5-fold, and 20-fold the level of iron required for normal growth, respectively. To examine the effect of iron chelation, equal numbers of NOD mice, maintained on the Standard 5P04 diet, were given either free access to water or water containing 0.2 mg/mL of the iron chelator deferiprone (ApoPharma, Rockville, MD) and monitored for hyperglycemia. Based on daily drinking volumes of 5 mL/25 g mouse, the dose of deferiprone was ∼40 mg/kg/day, well below the toxic dose for non–iron-overloaded mice (22).

Pancreas tissues were extracted from NOD and NOD.B10 mice at various ages. For embryonic samples, matings were established and plugs were checked every morning. Pregnant mice were sacrificed 17 days postconception. Pancreas tissues were collected. RNA was extracted, and samples from females were selected by genotyping using primers for Sex determining region of Chr Y (Sry) (Supplementary Table 2). Islets were isolated from 10-day-, 4-week-, and 12-week-old female NOD and NOD.B10 mice as previously described (1) and used for microarray or quantitative PCR (QPCR) analysis as described below. Islets collected from 4- and 12-week-old NOD mice contain various infiltrating immune cells, as previously described (7,23,24).

Islets were obtained from the University of Alberta Diabetes Institute’s IsletCore through the Stanford Diabetes Research Center (Supplementary Table 1). Islets were cultured overnight in low-glucose DMEM containing 10% FBS at 37°C prior to RNA extraction.

Two-color microarrays were performed in the islets of individual NOD mice against a pool of age-matched NOD.B10 mice (n = 6/group, 10-day-, 4-week-, and 12-week-old mice) using the Whole Mouse Genome Microarray Kit, 4 × 44K two-color arrays (Agilent Technologies, Santa Clara, CA) as previously described (2). Data were analyzed using GeneSpring (version 12.6.1; Agilent Technologies). Samples were filtered for entities with annotated gene symbols and entities detected in at least 4 of 18 NOD islet preparations (n = 6/age). The resultant 26,823 entities representing 16,822 genes were analyzed (data available from the Gene Expression Omnibus, accession number GSE149086). t tests and Benjamini-Hochberg multiple testing correction were performed to identify genes for which expression changed by twofold or more (corrected P < 0.01). Pathway analysis was performed using Advaita Bio’s iPathwayGuide (https://www.advaitabio.com/ipathwayguide). P values shown in Table 1 were determined by assessing the accumulated perturbation of a pathway and overrepresentation of genes in a pathway. Genes belonging to each pathway were curated by the Kyoto Encyclopedia of Genes and Genomes (KEGG; https://www.genome.jp/kegg/). Gene ontology (GO) analysis was performed on genes up- or downregulated by threefold or more using the Gene Ontology enRIchment anaLysis and visuaLizAtion tool [GOrilla] at http://cbl-gorilla.cs.technion.ac.il/). Separate analyses were performed on up- and downregulated genes to identify pathways more relevant to phenotypic differences (Table 1) (25). A total of 15,592 genes in the islets were associated with a GO term.

Total RNA was extracted from islets and pancreas tissues using TRIzol reagent (Invitrogen, Carlsbad, CA) and the Qiagen RNeasy Micro Kit (for islets) and Mini Kit (for pancreas; Qiagen, Germantown, MD) as previously described (1,26). RNA quality was assessed using the Agilent 2100 Bioanalyzer and the RNA 6000 Pico or Nano Reagent Kit (Agilent Technologies). All samples used had an RNA integrity number >8.0. cDNA was synthesized using Superscript III (Invitrogen). RT-PCR for various olfactory receptors was performed using primers listed on Supplementary Table 2. cDNA was preamplified using the TaqMan PreAmp Master Mix (Applied Biosystems, Foster City, CA) prior to QPCR if threshold cycle (Ct) values were >29. QPCR was performed using assays purchased from Applied Biosystems, the 7900HT Fast Real Time PCR System and the TaqMan Gene Expression Master Mix. Gene expression was calculated using the comparative Ct method for relative quantification (ΔΔCt).

Splenocytes from 12-week-old female NOD.BDC2.5 were activated in anti-CD3/anti-CD28–coated plates (2 μg/mL), in the presence of LPS (1 μg/mL; Sigma-Aldrich, St. Louis, MO) and IFN-α (200 U/mL; Sigma-Aldrich) for 24 h, as previously described (1). Splenocytes (5 × 10⁶) were i.p. injected into female 12-week-old NOD.SCID mice (n = 7), and control mice were injected with an equal volume of PBS (200 μL; n = 4). The pancreata were harvested 48 h later. RNA was extracted, and QPCR was performed as described above. BDC.2.5 transgene primers were purchased from The Jackson Laboratory.

Due to the sensitivity of the assays used, iron, hepcidin, and ferritin measurements were performed in the pancreas rather than the islets of individual NOD and NOD.B10 mice. Pancreas tissue from 12-week-old NOD and NOD.B10 mice were homogenized on ice in T-PER Tissue Protein Extraction reagent containing 1× Halt EDTA-free protease inhibitor (Thermo Fisher Scientific, Waltham, MA), and protein concentrations were determined using the Pierce Rapid Gold BCA assay kit (Sigma-Aldrich). Ferrous, ferric, and total iron levels were measured using an Iron Assay kit (ab83366; Abcam, Cambridge, MA). Hepcidin-1 and ferritin levels were measured using the Hepcidin Murine-Compete ELISA Kit (Intrinsic LifeSciences, La Jolla, CA) and the Ferritin Mouse ELISA kit (ab157713; Abcam), respectively. Prussian blue staining for iron was performed in tissue sections of 12-week-old NOD mice using the Abcam iron staining kit (ab150674), according to the manufacturer’s instructions.

Data were analyzed for normal Gaussian distribution using the Kolmogorov-Smirnov test. Based on the outcome, statistical analyses were performed using either the parametric two-tailed unpaired Student t test or the nonparametric Mann-Whitney test (Prism 5; GraphPad Software, San Diego, CA). For diet studies, the incidence of hyperglycemia in each group was compared using the log-rank (Mantel-Cox) test (Prism 5). A P value of ≤0.05 was considered significant.

Microarray data sets are available at the Gene Expression Omnibus under accession number GSE149086. All other data generated are included in the published article. No applicable resources were generated by the current study.

Principal component analysis and hierarchical clustering of normalized NOD versus NOD.B10 islet gene expression data show clustering of samples by age (Fig. 1A and B). The number of differentially expressed entities changed by twofold or more, threefold, and fivefold in the islets of 10-day-, 4-week-, and 12-week-old NOD versus NOD.B10 mice, respectively, are shown in Fig. 1C (P < 0.01), and entities changed by fivefold or more are listed in Supplementary Table 3.

Only 6 genes (Cd4, Suclg2, Glo1, 4930523C07Rik, Rars, and BC023719) were upregulated and 9 genes (represented by 12 entities: Ntrk2, Sdc2, D17H6S56E-5, Eme2, H2-K1, H2-T22, Zfand3, Zfp185, and Zfp871) were downregulated in the islets of NOD versus NOD.B10 mice across all 3 ages (Fig. 1D and E). More than half of these genes reside in chromosome 17 between chr17:25025623 and 36253239, in the “H2^b haplotype” region transferred from C57BL/10SnJ to NOD/Lt to create the NOD.B10Sn-H2b/J strain. Of the genes outside chromosome 17, Cd4, Suclg2, and Sdc2 may be relevant to T1D. CD4 is a marker of T helper cells. Sdc2 encodes syndecan 2, a heparan sulfate proteoglycan (HSPG) in the ECM, and Suclg2 encodes the β subunit of succinyl-CoA ligase, an enzyme in the tricarboxylic acid cycle that converts succinyl-CoA to succinate. Succinate stimulates insulin synthesis and secretion (27) and can enhance inflammation by activating macrophages and dendritic cells (28).

We performed QPCR analysis using islets from a separate cohort of mice to validate the microarray data for Cd4, Suclg2, and Sdc2, along with interferon-γ (Ifng, a marker of islet inflammation) and insulin 2 (Ins2) (Fig. 2A–E). As expected, Ins2 expression was unchanged, while Ifng expression was increased at 12 weeks of age, coincident with the onset of destructive insulitis. The upregulation of Cd4 and Suclg2 was also confirmed at 4 and 12 weeks, while Sdc2 was significantly downregulated in the NOD islets at all ages examined. The same genes were also measured in the pancreas prior to birth at E17 and at 10 days, 4 weeks, and 12 weeks of age. Similar patterns of Ins2, Ifng, Cd4, and Sdc2 gene expression were observed in the pancreas and islets, with no change in Ins2 expression, an upregulation of Ifng (at 4 and 12 weeks), Cd4 (at 4 and 12 weeks), and Suclg2 (at 10 days and 12 weeks), and a downregulation of Sdc2 (all ages studied) (Fig. 2F–J). The gradual upregulation of Cd4 and Ifng from 4 to 12 weeks in NOD islets is consistent with previous studies showing that ∼10% and ∼80% of NOD islets contain CD4⁺ T cells at 4 and 12 weeks of age, respectively (7).

Pathway and GO analysis were performed on differentially expressed genes at each age. Significantly impacted pathways are listed in Supplementary Table 4, and the most affected pathways related to disease progression are shown in Table 1. GO terms significantly associated with upregulated and downregulated genes at each age are shown in Supplementary Table 5, and the top biologically relevant GO terms are shown in Table 1.

At 10 days, we observed changes consistent with activation of glucagon signaling (KEGG pathway 04922) (Fig. 3A and Table 1) and positive regulation of insulin secretion (GO term 32024). Affected genes belonging to this pathway and GO term are shown in Fig. 3A–C. Using a separate cohort of 10-day-old NOD and NOD.B10 mice, we confirmed the upregulation of two genes in this pathway, glucagon receptor (Gcgr) and gastric inhibitory polypeptide receptor (Gipr), by QPCR (Fig. 3D and E).

Downregulated genes at 10 days of age associated with a loss of ECM receptor interaction (KEGG pathway 04512) (Table 1 and Fig. 4) and with GO terms associated with the ECM. A total of 124 genes belonging to the GO term “ECM structural constituent” (GO identification number 0005201) are expressed in the islets. A total of 34 of these genes are significantly changed by twofold or more at 10 days of age (Fig. 4B).

At 4 weeks, a significant upregulation of genes related to olfactory transduction (KEGG 4740) and olfactory receptor pathways in the islets of NOD mice was observed (Table 1 and Supplementary Fig. 1). A total of 300 genes belonging to the GO term “Olfactory receptor activity” (GO identification number 0004984) are expressed in the islets. A total of 66 were upregulated in NOD islets at 4 weeks, while few were changed at 10 days or 12 weeks of age (Fig. 5A–C). Among the 20 most highly expressed olfactory receptor genes in the islets of 4-week-old NOD mice, 13 were upregulated by twofold or more and 7 were upregulated by threefold or more (P < 0.01) (Fig. 5D). Olfactory receptor Olfr544 has been shown to regulate glucagon (29), while Olfr821 and Olfr15 may regulate insulin secretion (30). We observed an increase in Olfr544 but no change in Olfr821 and Olfr15 at 4 weeks of age (Fig. 5E). Olfr1428 and Olfr410 are homologs of human ORD4D6 and OR3A2, respectively, and are significantly upregulated in the islets of 4-week-old NOD mice (Fig. 5D). Using a second cohort of 4-week-old NOD and NOD.B10 islets, we confirmed by QPCR that both genes are indeed upregulated, while Olfr15 was unchanged in NOD versus NOD.B10 islets. By RT-PCR, we showed that OR3A2, OR51S1, OR4D6, OR5B2, OR674, and OR2C1, the human equivalents of Olfr410, Olfr571, Olfr1428, Olfr1444, Olfr821, and Olfr15, were all detected in human islets (Fig. 5G and Supplementary Fig. 1C).

At 12 weeks of age, the islets of NOD mice are infiltrated and highly inflamed (7). As expected, we saw significant upregulation of Ifng and Cd4 (Fig. 2B and C) and other genes and pathways associated with immune activation (Supplementary Tables 4 and 5 and Table 1), including Th17 cell differentiation (KEGG 4659), natural killer–mediated cytotoxicity (KEGG 4650), and T-cell receptor signaling (KEGG 4660) (Fig. 6A–C and Supplementary Fig. 2).

The most downregulated genes at 12 weeks of age were associated with antimicrobial responses (Supplementary Table 3). GO terms include response to biotic stimulus, response to bacterium, antimicrobial humoral response, and defense response (Table 1). Highly downregulated genes include members of the defensin family (Defa1, Defa3, Defa4, Defa6, Defa20, and Defb1) and the hepcidin antimicrobial peptide gene (Hamp), encoding proteins that mediate their antimicrobial functions through distinct mechanisms (Fig. 6D). Using another cohort of 12-week-old mice, we confirmed by QPCR that the expression of Defb1 and Hamp are downregulated in the islets of NOD versus NOD.B10 mice (Fig. 6E and F). To examine if the loss of Hamp and Defb1 expression is due to islet inflammation, we injected NOD.SCID mice with activated splenocytes of NOD.BDC2.5 mice to induce islet inflammation in vivo. We had previously shown that the islet isolation procedure can induce changes in gene expression (1). To avoid this, we measured Hamp and Defb1 expression in whole pancreas samples, with the knowledge that Hamp is expressed in the islets of the pancreas, while Defb1 is expressed in both the endocrine and exocrine regions (31–34). Successful transfer and homing of splenocytes was confirmed by expression of BDC2.5 (Fig. 7A), and inflammation was confirmed by upregulation of Ifng (Fig. 7B). Ins2 and Gcg were not changed (Fig. 7C and D), while Defb1 and Hamp were significantly downregulated (Fig. 7E and F).

To examine if NOD mice exhibit a defect in pancreatic iron metabolism, we compared Hamp, hepcidin-1, ferritin, and iron levels in the pancreas of 12-week-old NOD and NOD.B10 mice. Similar to its expression in the islets, Hamp was significantly lower in the pancreas of NOD mice (Fig. 8A). Pancreatic hepcidin protein levels were also significantly lower, with hepcidin detected in only one of nine NOD samples, but eight of nine NOD.B10 samples (Fig. 8B). Ferrous iron (Fe²⁺) levels were higher in the NOD pancreas. Interestingly, the only NOD sample with detectable hepcidin expression had the lowest Fe²⁺ levels, while the only NOD.B10 pancreas with undetectable hepcidin expression had the highest Fe²⁺ levels (Fig. 8C). The levels of ferric iron (Fe³⁺) and ferritin, which binds Fe³⁺, did not differ between NOD and NOD.B10 mice (Fig. 8C and D). By histology, the highest concentration of iron was found in iron-rich cells in the peri-insulitic lesions surrounding the islets of NOD mice. These are likely macrophages; however, due to the incompatibility of staining procedures, we were not able to confirm this by immunohistochemistry. Macrophages are present in the peri-insulitic lesion (24,35,36) and are the only “immune cells” that store large amounts of iron (37). Iron was not detected in β-cells. However, β-cells exhibit low antioxidative capacity (38) and may perish before iron levels are high enough to be detected by Prussian blue staining (39). Studies in iron-overloaded rats have shown that ferritin deposits could be detected at the ultrastructural level in the plasmalemma and the cytoplasmic surface of secretory granule membranes of β-cells where elevated iron levels could not be detected by Prussian blue staining (39). Changes in the mRNA expression of various iron-responsive genes, including Tfrc (transferrin receptor 1), may also serve as useful surrogates to assess intracellular iron levels in β-cells (40) (Supplementary Fig. 3).

To examine the role of dietary iron on disease progression, we first tested two standard diets routinely used by the Stanford School of Medicine animal facility: Prolab RMH-5P04 (5P04; 369 ppm iron) and Teklad 2018 SX (2018; 176 ppm iron). The 5P04 diet is comparable to the 5K52 diet (containing 360 ppm iron; LabDiet) used by The Jackson Laboratory, where 80% of female NOD mice developed hyperglycemia by 25 weeks of age. We showed that 80% of NOD mice maintained on 5P04 also developed hyperglycemia by 25 weeks (Fig. 8F). Remarkably, disease incidence fell to 38% and 48% in two cohorts of NOD mice maintained on the low-iron 2018 diet (Fig. 8F and G), suggesting that reduced dietary iron may protect against disease development. To test this hypothesis, we maintained mice on custom diets formulated using 2018SX supplemented with iron. The addition of 200 ppm and 850 ppm iron significantly increased the incidence of hyperglycemia to 65% and 72%, respectively (Fig. 8G). No significant difference in disease incidence was observed among the 5P04, 2018SX+200, or 2018 + 850 diets. Since elevated iron exposure exacerbates NOD disease, we also asked if iron chelation could reduce disease. Treatment with the iron chelator, deferiprone, had no significant effect on disease progression (Supplementary Fig. 4).

The NOD mouse model is an invaluable tool for studying the pathogenesis of T1D. We identified novel genes and pathways that are dysregulated in the islets of NOD mice during distinct stages of disease progression and hypothesize that these pathways may act together to create an environment that promotes β-cell death (Fig. 8H). The earliest defect affects the ECM of NOD islets. The ECM consists of a basement membrane that encases and protects the islets and an interstitial matrix that regulates islet survival, differentiation, and proliferation (41,42). In NOD mice and patients with T1D, degradation of the basement membrane occurs, with a significant loss of heparan sulfate and HSPGs (10–13,42). Heparan sulfate is crucial for β-cell survival and protects against oxidative stress (12,13). Reduced expression of Sdc2 and other HSPG genes may contribute to the loss of heparan sulfate in NOD islets. Sdc2 encodes syndecan-2, a cell surface core binding protein for heparan sulfate (43). Sdc2 was reduced in the pancreata and islets of NOD compared with NOD.B10 mice at all ages studied, including prior to birth at E17 (Figs. 1 and 2).

Genes controlling the expression of other HSPGs, including perlecan (Hspg2), laminin subunit α5 (Lama5), collagen type XVIII α 1 (Col18a1), and agrin (Agrn), were also downregulated at 10 days of age, while three ECM genes were upregulated: fibrinogen (Fgb), mucin 4 (Muc4), and hyaluronan and proteoglycan link protein 1 (Hapln1). The proteins encoded by these three genes interact with hyaluronan (44–46), a component of the basement membrane that accumulates in the islets of 4-week-old NOD mice and patients with established T1D (42,47,48). Under inflammatory conditions, hyaluronan is degraded to form proinflammatory fragments that enhance myeloid and lymphoid cell accumulation and promote T-cell adhesion (48–50). A shift in ECM composition, with increased hyaluronan-binding and reduced heparan sulfate–binding proteins in the islets of NOD versus NOD.B10 mice, may contribute to the formation of the peri-insulitic lesion observed at ∼4 weeks of age (Fig. 8H).

Hypersecretion of glucagon has been observed in NOD mice and patients with T1D (1,15,16,51–53). We showed that the changes associated with enhanced glucagon signaling occur prior to islet inflammation (Fig. 3), with an upregulation of glucagon gene (Gcg) expression that is consistent with hyperactive α cells (1), and an upregulation of genes encoding glucagon receptors (Gcgr), glucagon-like peptide 1 receptors (Glp1r), and glucose-dependent insulinotropic polypeptide receptors (Gipr). These receptors are expressed on β-cells and mediate insulin secretion (54,55). Thus, their overexpression may result in β-cell hyperactivity and contribute to β-cell death (14) (Fig. 8H).

Increased olfactory receptor expression may also enhance glucagon and insulin secretion. Olfactory receptors are expressed in α and β-cells of mice, where they can act as chemosensors for food derivatives (29,30,56). Activation of OLFR544 on α cells by azelaic acid (dicarboxylic acid in grains) stimulates glucagon secretion (29), while activation of OLFR15 on β-cells by octanoic acid (carbon fatty acid in milk) potentiates glucose-stimulated insulin secretion (30,56). Olfr544 was among the 66 Olfr genes significantly changed in 4-week-old NOD islets, but Olfr15 was unchanged (Fig. 5E). We showed that human islets express OR3A2 and OR4D6, the orthologs of Olfr410 and Olfr1428 that were significantly upregulated in the islets of 4-week-old NOD mice. It is possible that OR3A2 and OR4D6 may be similarly changed in the islets during the progression of T1D.

During destructive insulitis, multiple antimicrobial defense genes, including Defb1 and Hamp genes, were downregulated in the NOD islets (Fig. 6A–C). Defensins block viral fusion, can form multimeric pores in the cell membrane of pathogens (57,58), and can serve as chemoattractants for immature DCs and T cells (58).

Variations in the DEFB1 gene are associated with autoimmune diseases and conditions, including T1D (59–62), and studies have shown that DEFB1 is regulated by inflammation, insulin, and glucose (59,63). In the NOD pancreas, inflammation reduces Defb1 expression. Reduced levels of this secreted peptide in either the endocrine or exocrine pancreas may lead to a loss of antimicrobial defense against bacteria and viruses that may infect the pancreas (64,65).

Hepcidin regulates iron homeostasis and elicits its antimicrobial function by limiting the availability of iron that is required for the survival of invading pathogens (66). In the pancreas, hepcidin is expressed exclusively in β-cells and is stored and coreleased with insulin (34,67). Once released, hepcidin binds to ferroportin receptors expressed on iron-rich cells such as macrophages to limit the export of ferrous iron (Fe²⁺) (Fig. 8H). Studies show that tissue-resident macrophages are detected in pancreatic islets of NOD mice at birth and in the first week of life (24,36,68,69). They undergo a stepwise activation program in response to inflammation (24) and are crucially involved in the initiation of disease in NOD mice (70). Monocyte-derived macrophages migrate to the islets at later stages of insulitis (24,35,36). We propose that β-cells normally release hepcidin along with insulin as a protective mechanism to maintain low iron levels near the islets. However, in NOD mice, we show that inflammation significantly reduces hepcidin gene and protein expression (Figs. 6F, 7F, 8A, and 8B), without drastically altering the expression of other iron regulatory genes (Supplementary Fig. 3). The loss of hepcidin leads to the uninhibited release of Fe²⁺ from macrophages and supports the survival of microbes and pathogens that require host iron. Extracellular Fe²⁺ is transported through divalent metal transporters expressed on the surface of β-cells. Once inside the β-cell, Fe²⁺ can drive the production of hydroxyl radicals and reactive oxygen species that contribute to oxidative stress and β-cell death (71–73) (Fig. 8H). Studies have shown that the iron chelator, desferrioxamine, can significantly improve islet allograft survival in NOD mice when administered with nicotinamide, an anti-inflammatory agent (74). Thus, reduced iron levels are beneficial to islet survival, and restoration of pancreatic iron homeostasis could prevent or delay β-cell destruction in NOD mice.

Elevated iron levels have been associated with the development of T1D, type 2 diabetes, and gestational diabetes (75–77), suggesting that the pancreas is particularly sensitive to high iron levels. Individuals suffering from systemic iron overload, such as patients with hereditary hemochromatosis, have been shown to accumulate iron in the pancreas (78,79). The deposition of iron in β-cells has been suggested to mediate iron-induced oxidative stress and β-cell death, leading to the development of diabetes in 30–60% of patients with hemochromatosis (80–82). A number of epidemiological studies have also linked the early exposure of high iron levels to the development of T1D in children. Use of iron supplements during prenatal development increases the incidence of T1D in children before the age of 17, while use of iron-fortified formula during the first 4 months of infancy significantly increases the risk of developing T1D before age 6 (19,20). High iron levels in neonatal blood have also been linked to an increased risk for T1D in children under age 16 (77), while high consumption of red meat is associated with the development of T1D in young children (17,18). Increased dietary iron elevates the iron content in macrophages (83) and thus may amplify the damage mediated by reduced hepcidin expression in the islets (Fig. 8H). We showed that the development of NOD disease also correlates with iron exposure. Mice maintained on low-iron diets had the lowest incidence of hyperglycemia, while those maintained on higher-iron diets had an earlier onset and higher incidence of disease.

Our studies provide insight into the pathogenesis of T1D by identifying novel genes/pathways that are altered during distinct stages of disease and by showing how changes in the islets may integrate with environmental factors, such as diet, to trigger the onset of T1D. Our findings present a plausible mechanism to explain the association between high iron exposure and the increased risk of T1D and suggest that iron supplementation should generally be avoided by at-risk children and pregnant mothers of at-risk individuals.

Acknowledgments. The authors thank Drs. M. Bogdani (Benaroya Research Institute) and E. Korpos (University of Münster) for useful discussion related to the role of ECM proteins in T1D.

Funding. Funding was provided by JDRF (grant 1-SRA-2019-807-S-B) and the Myra Reinhard Foundation. This research was performed with the support of the Stanford Diabetes Research Center Islet Research Core, supported by National Institutes of Health grant P30DK116074.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. L.Y. performed the microarray and data analysis, BDC2.5-activated splenocyte experiments, iron diet and iron chelation studies, RT-PCR, QPCR, biological assays, and histology experiments, prepared the manuscript, and composed the figures and tables. R.A. performed RNA extraction, QPCR experiments, and blood glucose measurements. C.T. maintained the NOD and NOD.B10 mice and isolated pancreatic islets. R.F. performed RNA extraction, bioanalyzer, and QPCR analysis. L.Y. and C.G.F. were involved in the planning and direction of this work. All authors reviewed and edited the manuscript. L.Y. and C.G.F. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.