Granzyme A Deficiency Breaks Immune Tolerance and Promotes Autoimmune Diabetes Through a Type I Interferon–Dependent Pathway

Standard paradigms suggest an external environmental event in early life triggers autoimmune type 1 diabetes in genetically susceptible individuals. These events might be viral or bacterial infections, introduction of dietary constituents, or just reflect the playing out of stochastic events and probability in people who are genetically at risk. Virus infection may induce a bystander immune response through activation of pattern recognition receptors and production of proinflammatory cytokines (1–5). However, convincing evidence is lacking for exogenous triggers, and mechanisms are understood in outline, but with insufficient detail. These early events are virtually impossible to study in humans, but progress can be made with animal models such as the nonobese diabetic (NOD) mouse, which is the best-characterized animal model of type 1 diabetes.

The first 4 weeks of life in NOD mice are critical to the development of autoimmune diabetes. T-cell responses to proinsulin develop during this time, and inducing tolerance to proinsulin prevents development of disease (6,7). The events that trigger these T-cell responses in NOD mice are poorly understood; however, type I interferons (IFNs) have repeatedly been implicated in these early events in humans and mice and are downstream of suggested triggers like viruses. In humans, an enhanced IFN gene signature is evident before the development of autoimmunity in children at risk for developing type 1 diabetes, suggesting that production of type I IFN is an important initiating event in human type 1 diabetes (8,9). Elevated IFNα, correlating with hyperexpression of HLA class I is also observed in pancreata and islets of deceased patients with type 1 diabetes (10). In NOD mice, type I IFN is also produced in the pancreatic islets, peaking at ∼4 weeks of age and coinciding with the infiltration of dendritic cells (DCs) and plasmacytoid DCs (pDCs) into the islets (11,12). However, it remains unclear how production of type I IFN is stimulated. Poor clearance of nucleic acids and type I IFN production have been linked in the pathogenesis of systemic autoimmune diseases such as systemic lupus erythematosus (SLE) and inflammatory syndromes such as Aicardi-Goutières syndrome (13,14), but this connection has not been observed in organ-specific autoimmune diseases.

Granzyme A is a trypsin-like serine protease (“tryptase”) that cleaves selected proteins after the basic residues Arg or Lys. Mouse granzyme A can induce perforin-mediated target cell death in vitro (15); however, the cytotoxic T lymphocytes and NK cells from granzyme A–deficient mice kill cells with normal kinetics and morphology (15,16). This finding, together with the recent demonstration that human granzyme A lacks cyotoxic function (15,17), suggests that granzyme A may have further undefined role(s) in vivo (18). It was recently reported that GZMA transcript and protein expression are strongly downregulated in T cells from children with type 1 diabetes compared with control children (19). In vitro, granzyme A activates components of the cytoplasmic SET complex, an endoplasmic reticulum–associated complex linked to DNA degradation and repair (20). The SET complex comprises the inhibitor protein SET, the endonuclease Ape1, the SET-binding protein HMG2, the DNAse NM23-H1, and the 3′-to-5′ exonuclease TREX1 (20). Loss of TREX1 nuclease activity results in aberrant accumulation of cytosolic DNA (21), and mice lacking TREX1 develop cardiac autoimmunity (22).

The above observations suggest a connection between granzyme A and immune regulation. Here we show that granzyme A–deficient NOD mice display aberrant cytosolic DNA accumulation, an increased IFN gene signature, and enhanced susceptibility to the development of spontaneous diabetes. We conclude that granzyme A has a previously unrecognized role in the control of autoimmunity.

All mice used in the study were bred and maintained at St. Vincent’s Institute, in the same animal facility, with the same environmental and dietary conditions. All experiments were approved by the St. Vincent’s Health animal ethics committee. B6.Gzma^−/− mice made using C57BL/6-III embryonic stem cells (16) were backcrossed onto the NOD/Lt genetic background for 11 generations and intercrossed to produce NOD.Gzma^−/− mice. Wild-type mice were bred independently, in the same animal facility. DNA from the 11th generation was genotyped using the Illumina mouse medium density linkage panel for 1,449 single nucleotide polymorphisms, and strain differences were identified using The Jackson Laboratory Mouse Genome Informatics and National Center for Biotechnology Information databases. Mice were of the NOD/Lt genotype across the whole genome except for the region on chromosome 13 encompassing the Gzma locus. Fine mapping using high-resolution melt analysis identified the congenic interval between and including rs29244316 (∼107 Mb) and rs29246100 (∼116 Mb).

NOD.Gzma^−/− were crossed with NOD.Ifnar1^−/− mice (12) to produce NOD.Gzma/Ifnar1^−/− mice, and with NOD.PI mice (6) to generate NOD.Gzma^−/−/PI mice. NOD/Lt mice were bred in house.

Female mice were monitored for diabetes for 300 days. Mice with two consecutive blood glucose measurements of ≥15 mmol/L were considered diabetic. For immunohistochemistry, tissues were snap-frozen in optimal cutting temperature compound (OCT Compound; Sakura Finetek, Torrance, CA) and stored at −80°C, or fixed in neutral buffered formalin and embedded in paraffin. Preparation of sections, staining, and scoring for insulitis were performed as previously described (23).

Insulin autoantibodies (IAAs) were measured with a 96-well filtration plate micro-IAA assay as described previously (24). Antinuclear antibodies were detected with HEp-2000 ANA slides (Immuno Concepts, Sacramento, CA). Slides were stained with serum diluted 1:100 in PBS according to the manufacturer’s instructions. Secondary antibody was anti-mouse IgG Alexa Fluor 488 (Life Technologies, Eugene, OR). Fluorescence intensity was scored using a 0–4 scale in a blinded manner by two independent investigators.

Islets of Langerhans were isolated from mice using collagenase P (Roche, Basel, Switzerland) and Histopaque-1077 density gradients (Sigma-Aldrich, St. Louis, MO) as previously described (23).

CD4⁺CD25⁻ T-effector cells were purified using magnetic cell separation from the spleens of NOD mice, and 1 × 10⁵ cells were cultured in a 96-well U-bottom plate alone or with CD4⁺CD25⁺ Treg cells purified from the spleens of NOD or NOD.Gzma^−/− mice at 1:0.1, 1:1, and 1:2 ratios for 72 h. CD11c⁺ DCs purified by magnetic cell separation from the spleens of NOD mice (2 × 10⁴ cells) and anti-CD3 antibody (145-2C11) at 5 μg/mL were added into each well. [³H]thymidine was added at 0.5 μCi/well for the last 18 h of culture. The cells were harvested and [³H]thymidine incorporation was measured by scintillation counting.

Single-cell suspensions were prepared from spleen, lymph nodes, and islets (25). Cells were surface stained with CD45 (30-F11), CD4 (GK1.5), CD8 (53-6.7,), B220 (RA3-6B2; all from Biolegend, San Diego, CA), CD11c (HL3), TCRβ (H57-597), CD3 (145-2C11), CD44 (1M7), and CD49 (DX5; all from BD Biosciences, Franklin Lakes, NJ). Because of low numbers of immune cells in islets, CD11c⁺B220⁺ was used to identify pDCs and CD11c⁺B220⁻ for conventional DCs (cDCs) (11). For intracellular staining, surface-stained cells were fixed and permeabilized with BD Cytofix/Cytoperm (BD Biosciences) as described in the manufacturer’s protocol. Subsequently, these cells were stained with antibodies recognizing Foxp3 (FJK-16s) and granzyme A (GzA-3G8.5; both from eBioscience, San Diego, CA). Cultured islet cells or cells from pooled peripheral lymphoid tissues (spleen and lymph nodes, including mesenteric and inguinal lymph nodes, excluding pancreatic lymph node) were stained with phycoerythrin (PE)-conjugated IGRP_206–214 (VYLKTNVFL) or PE-conjugated INS_B15–23 (LYLVCGERL) H-2K^d tetramers (ImmunoID, Parkville, Australia), or PE-conjugated INS_B10–23 (HLVERLYLVCGGEG) I-A^g7 tetramers (National Institutes of Health Tetramer Core) as previously described (26). I-A^g7-CLIP (PVSKMRMATPLLMQA) or H-2K^d-TUM (KYQAVYTTTL) tetramers were used to determine background staining. Data were collected using a BD LSRFortessa (BD Biosciences) and analyzed using FlowJo analysis software (Tree Star Inc., Ashland, OR).

pDCs and cDCs were purified from the spleens of 4-week-old NOD or NOD.Gzma^−/− mice using Nycoprep1.077 gradients and FACS (BD Influx; BD Biosciences) for CD11c⁺CD45RA⁻ (cDCs) and CD11c^intCD45RA⁺ (pDCs). Cells (2 × 10⁵/200 µL pDC) were stimulated with CpG₂₂₁₆ (0.5 μmol/L; InvivoGen) or left unstimulated for 18 h. Supernatants were collected for IFNα concentration by the FlowCytomix analyte detection system (eBioscience). Unstimulated cells were lysed for RNA isolation. IFNα was also measured in serum samples by FlowCytomix.

Total RNA was isolated using Nucleospin RNA XS kits (Macherey-Nagel, Düren, Germany) and reverse transcribed using a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA) according to the manufacturer’s protocol. Real-time PCR was performed using AmpliTaq Gold DNA Polymerase (Applied Biosystems) and a LightCycler 480 (Roche). TaqMan primers for Mx1, Ifit1, Oas1a, Oas1b, Isg15, Tlr9, Ifnb, and Actin-b were purchased from Applied Biosystems. Threshold cycle (Ct) values from test samples were normalized to Ct values from control samples and expressed as delta Ct (dCT).

Tissues were cut into 5-μm sections and fixed in acetone. Staining with primary antibodies was performed for 1 h with the following antibodies: anti-CD3 (KT3; made in house), anti-CD49 (DX5), anti-CD11c (N418), anti-B220 (RA3-6B2; all purchased from BD Biosciences), anti-Siglec H (eBioscience), anti-glucagon (Abcam, Cambridge, U.K.), and anti-insulin (Dako, Carpinteria, CA). Secondary antibodies (purchased from Thermo Fisher Scientific) were as follows: anti-rat Alexa Fluor 488, anti-rat Alexa Fluor 568, anti-rabbit Alexa Fluor 568, and anti-guinea pig Alexa Fluor 488. For detection of ssDNA, sections were stained with anti-ssDNA antibodies (F7–26) and Cy5-conjugated anti-IgM antibodies (both from Merck, Darmstadt, Germany). Nuclei were stained with 1 μg/mL DAPI (Sigma-Aldrich) and mounted with Fluorescent Mounting Medium (Dako). Slides were analyzed using a Nikon A1R Confocal Microscope. Using confocal microscopy, all islets in a section (15–20 islets/mouse) were scored as either positive or negative for ssDNA staining (i.e., containing at least one ssDNA⁺ cell), and the proportion of islets with ssDNA staining was calculated. DAPI staining was used to confirm cytoplasmic ssDNA staining, and insulin staining was used to identify islets. For the spleen, the number of ssDNA⁺ cells was counted in 15 fields/spleen from 4-week-old mice and expressed as the total number of ssDNA⁺ cells. Scoring was done by two separate investigators in a blinded manner. Kidney sections were stained with FITC-labeled anti-C3 Ab.

NK cells were purified by magnetic bead separation using CD49b microbeads (Miltenyi Biotec) from the spleens of NOD and NOD.Gzma^−/− mice and cultured at 0.5 × 10⁶ cells/mL with 1,000 units/mL rhIL-2 (PeproTech) for 7 days. When harvested, the purity of the TCRβ⁻CD49b⁺ NK cells was >97%. Cytoplasmic lysates were extracted from 4 × 10⁶ cells/sample and incubated with ³²P-labeled 3′ flap substrate generated by annealing oligonucleotides (Supplementary Table 1) as previously described (27). Assays were performed in 1× reaction buffer (50 mmol/L PIPES, pH 6.8, 15 mmol/L MgCl₂, 10 mmol/L DTT) at 37°C for 15 min. Stop buffer containing 2% SDS and 10 mg/mL proteinase K was added, and samples were incubated at 37°C for 30 min and then run with bromophenol blue on a 15% Tris/borate/EDTA acrylamide gel. The gel was dried and exposed to X-ray film at −80°C for up to 5 days. At the same time, samples of protein lysate were run on a 4–12% SDS-PAGE gel in MOPS buffer and then stained with Coomassie blue for protein quantification. Bands were quantified on a LI-COR Odyssey (LI-COR Biosciences), and values were normalized by dividing the intensity of the cleaved 3′ oligonucleotide band by the total protein.

Statistical analysis was performed using GraphPad Prism Software (San Diego, CA). All data shown as bar graphs are represented as mean ± SEM. A P < 0.05 was considered significant.

We generated granzyme A–deficient NOD mice by backcrossing C57BL/6.Gzma^−/− mice onto the NOD/Lt strain. Female NOD.Gzma^−/− mice had a higher incidence of diabetes (81%) compared with wild-type NOD mice (57%) and diabetes developed earlier (median onset of diabetes 20.4 weeks in NOD.Gzma^−/− compared with 31.5 weeks in NOD) (Fig. 1A). We did not observe any significant differences in T-cell populations in the thymus, spleen, and lymph nodes (Supplementary Fig. 1A–C). NOD.Gzma^−/− mice also displayed an increased number and proportion of β-cell antigen-specific CD8⁺ T cells in peripheral lymphoid organs (pooled lymph nodes and spleen) (Fig. 1B) and islets (Fig. 1C and D), respectively, consistent with increased diabetes incidence. Four-week-old female NOD.Gzma^−/− mice had increased infiltration of immune cells in the islets (insulitis) compared with wild-type NOD mice (Fig. 1E and F); however, by 10 or 15 weeks of age, there was no significant difference (Fig. 1E). Male NOD.Gzma^−/− mice had increased insulitis at 10 weeks of age (Fig. 1E) and increased diabetes incidence (Fig. 1G). The proportions of immune cell populations in the islets were unchanged at 4 and 10 weeks of age (Supplementary Fig. 1D–F), with the exception of an increase in the proportion of NK cells in the islets of NOD.Gzma^−/− mice (Supplementary Fig. 1G). An increase in the proportion of NK cells in islets has previously been linked to accelerated diabetes (28). The number and function of regulatory T cells (Supplementary Fig. 2) or DCs (Supplementary Fig. 3) were unaffected by granzyme A deficiency.

NOD.Gzma^−/− mice developed antinuclear antibodies with earlier onset compared with wild-type NOD mice (Fig. 1H). The staining had a speckled pattern (not shown), which may indicate autoantibodies to extractable nuclear antigens, a feature of SLE in humans. Deposition of complement C3 was also detected in the kidneys of 14-week-old NOD.Gzma^−/− mice (Fig. 1I). These data indicate an increase in both organ-specific and systemic autoimmunity in mice lacking granzyme A.

The increased insulitis at 4 weeks of age, an early time point that coincides with the onset of autoimmune disease in NOD mice (11), prompted us to examine whether granzyme A deficiency may be able to break tolerance to islet antigens in NOD mice. NOD.PI mice express the β-cell antigen proinsulin in antigen presenting cells under control of the MHC class II I-E promoter (6). This augments tolerance to proinsulin, and NOD.PI mice are robustly protected from islet infiltration and never develop diabetes (6). Remarkably, when we generated NOD.PI mice also lacking granzyme A (NOD.Gzma^−/−/PI), 30% of mice developed diabetes (Fig. 2A). IAAs were present in a small proportion of NOD.Gzma^−/−/PI mice but not in NOD.PI mice or C57BL/6 mice that are not prone to development of autoimmune disease (Fig. 2B) (7). The IAAs detected in the NOD.Gzma^−/−/PI mice were specific, as autoantibody binding was inhibited in the presence of competitive fluid phase insulin (24). Insulitis was also observed in some nondiabetic NOD.Gzma^−/−/PI mice (Fig. 2C and D). We observed robust proliferation of IGRP-specific CD8⁺ T cells in the pancreatic lymph node (PLN) of NOD.Gzma^−/−/PI mice (Fig. 2E) unlike NOD.PI mice where proliferation in the draining node was not observed as previously reported (7). NOD.Gzma^−/−/PI mice showed increased expression of the antigen experience marker CD44 on endogenous insulin_B10–23–specific CD4⁺ T cells (Fig. 2F and Supplementary Fig. 4A) and IGRP_206–214–specific CD8⁺ T cells (Fig. 2G and Supplementary Fig. 4B) in peripheral lymphoid organs when compared with NOD.PI mice. Together these data lead to the conclusion that robust immune tolerance to proinsulin in NOD.PI mice was broken by ablating granzyme A expression.

Granzyme A can cleave, and therefore activate, molecules in the endoplasmic reticulum–associated SET complex. This complex is involved in clearance of cytosolic DNA (21), so we stained pancreas sections with an antibody recognizing ssDNA. We observed cells around the pancreatic islets with cytosolic ssDNA in NOD.Gzma^−/− mice at 4 weeks of age (Fig. 3A and B). Cells with cytosolic ssDNA were also evident in the islets of wild-type NOD mice, but these were far fewer in number than in NOD.Gzma^−/− mice. The ssDNA was not observed in insulin-positive β-cells or other islet endocrine cells but in the islet-infiltrating immune cells (Fig. 3A). Cytosolic ssDNA was also detected at greater frequency in the spleens of NOD.Gzma^−/− compared with NOD mice (Fig. 3C and D). The ssDNA was localized to CD11c⁺ DCs, pDCs, and NK cells in the spleen and islets but was not observed in T cells or B cells (Fig. 3E). The origin of the cytosolic DNA is unknown, but it may arise from endogenous retroelements or inefficient clearance of apoptotic cells, both of which have been described in NOD mice (29,30).

We determined if the cell types that accumulate ssDNA normally produce granzyme A. Granzyme A was detected in DCs, NK cells, and a small proportion of T cells from naïve, wild-type splenocytes (Fig. 3F), and in a small proportion of CD45⁺ cells from islets of wild-type NOD mice (Fig. 3G). Granzyme A became readily detectable upon activation of β-cell antigen-specific CD8⁺ T cells with specific peptide (data not shown).

Consistent with an increase in cytosolic DNA, we observed reduced cytoplasmic 3′-5′ exonuclease activity in NK cells isolated from NOD.Gzma^−/− mice compared with those from wild-type mice (Fig. 3H).

We next determined whether increased ssDNA in islet-infiltrating cells plays a role in the development of anti-islet autoimmunity. Children genetically predisposed to develop type 1 diabetes show an increased expression of type I IFN–regulated genes before the onset of clinical disease (8,9). NOD mice also display a type I IFN gene signature in the islets that peaks at ∼4 weeks of age (11,12). Expression of the type I IFN–regulated genes Mx1, Ifit1, Isg15, Oas1a, and Oas1b, which we have previously shown to be increased in the islets of NOD mice, was further increased in the islets of NOD.Gzma^−/− mice at 4 weeks of age compared with wild-type NOD mice (Fig. 4A and Supplementary Table 2). Two of these genes, Ifit1 and Oas1b, remained upregulated in the islets of 10–12-week-old NOD.Gzma^−/− mice compared with wild-type NOD mice (Fig. 4B and Supplementary Table 2). We did not observe IFNα in the serum of NOD.Gzma^−/− mice (data not shown), and IFNα produced by cultured pDCs from NOD.Gzma^−/− mice was not significantly above that produced by pDCs from NOD mice (Supplementary Fig. 3C).

To determine if type I IFN signaling contributes to the increased diabetes incidence in NOD.Gzma^−/− mice, we crossed these mice with NOD mice lacking receptors for type I IFN (NOD.Ifnar1^−/−). NOD.Gzma/Ifnar1^−/− mice developed diabetes at the same frequency as wild-type NOD mice and NOD.Ifnar1^−/− mice (12), indicating that type I IFN signaling is required for the increased diabetes incidence in NOD.Gzma^−/− mice (Fig. 4C). Although type I IFN has been linked to triggering of autoimmunity (8,9,11), NOD mice lacking IFNAR1 are not protected from autoimmune diabetes (12). Our data therefore suggest that type I IFN can increase autoimmune diabetes in NOD.Gzma^−/− mice but is not essential for diabetes in wild-type NOD mice.

Mechanisms that influence the development of diabetes are of intense interest because of the possibility that disease progression might be slowed if key factors were understood and targeted. Autoimmunity develops as a result of T cells that escape thymic deletion, perhaps because their affinity for self-antigen may be just below the level that inevitably results in deletion. These “borderline” T cells are present in all individuals, to an extent that may be genetically determined by HLA and proinsulin genotypes (31). For example, NOD.PI mice have fewer proinsulin-specific T cells than NOD mice (32), and these may have insufficient affinity to cause disease. Our current findings pinpoint a mechanism through which such low-affinity autoantigen-specific T cells might be activated and cause overt autoimmune disease. Our data suggest that ssDNA may act as an endogenous signal that can stimulate local type I IFN production to activate the residual T cells in NOD.PI mice. In this scenario, NOD.PI mice are genetically programmed not to develop diabetes but do so when the immune microenvironment is skewed to higher type I IFN expression than is usually observed. Viruses can also accelerate diabetes (1,33), and it is possible that virus infection could break tolerance in a similar manner through production of type I IFN.

The production of type I IFN in subjects with diabetes has been recognized for many years. There are examples where type I IFN can trigger islet autoantibody seroconversion and diabetes (34–39). In humans, mutations in enzymes involved in nucleic acid clearance and type I IFN production result in autoimmune and inflammatory diseases such as SLE or Aicardi-Goutiéres syndrome (13,14). Loss-of-function mutations of the RNA sensor IFIH1 are associated with type 1 diabetes resistance in humans (40). Together these findings suggest that a mechanism whereby endogenous DNA may be a trigger for type I IFN production in type 1 diabetes may exist. Further studies are needed to confirm this.

We did not find any evidence of polymorphisms in human type 1 diabetes genome association studies for GZMA or any genes in the SET complex. However, this does not exclude a role for these proteins in diabetes, as genome association studies do not capture expression data. Individuals with type 1 diabetes may have a defect in the removal of aberrant nucleic acids due to reduced granzyme A and/or SET complex expression arising from genetic perturbations in regulatory factors. Indeed reduced granzyme A expression was recently described in children with type 1 diabetes (19).

We identified ssDNA in cells of the innate immune system, namely NK cells, DCs, and pDCs, but not in T or B cells. These cells expressed granzyme A protein, and BioGPS data indicate that they also express Gzma mRNA. These cell types have also been implicated in the early events in autoimmune diabetes. DCs are among the first cells to infiltrate the islets of NOD mice and are important for presenting insulin epitopes to T cells in the pancreatic lymph node (41). pDCs are the major producers of type I IFN, and depletion of this cell type early in diabetes pathogenesis (at 2–4 weeks of age) prevents the development of disease (11). Recently, a subset of cDCs that express perforin and granzyme A have also been implicated in the control of experimental autoimmune encephalomyelitis in mice (42). Although NK cells are unable to directly kill β-cells (43), they produce high levels of IFNγ and have been implicated in the acceleration of autoimmune diabetes (28).

We have identified a noncytolytic role for granzyme A in limiting autoimmunity in vivo. Evidence suggests that unlike granzyme B, granzyme A does not play a dominant role in the induction of target cell death by cytotoxic T lymphocytes or NK cells. Granzyme A–deficient mice on a C57BL/6 background have no overt phenotype unless challenged by infection (44). They develop higher titers of experimentally induced herpes simplex virus infection compared with wild-type mice but are able to clear the virus and survive. Granzyme A appears to play a role in restricting the spread of this dsDNA virus (45), which could be due to the role of granzyme A in the cleavage of aberrant DNA. Recombinant granzyme A induced nuclease activity that resulted in ssDNA nicks, and silencing the nuclease TREX1 inhibited granzyme A–induced DNA damage (20).

Questions still remain about how granzyme A is able to access and activate the SET complex in vivo. We suggest that this is a cell-intrinsic process where granzyme A is released from endolysosomal storage granules into the cytosol, as described for granzyme B (46,47). Alternatively, an intronic promoter that can be activated by dexamethoasone induces transcription of a granzyme A isoform lacking a signal peptide, which would be produced in the cytosol (48). If folded correctly and functional, this form would have direct access to the SET complex. Future experiments will explore these possibilities and determine the trigger for granzyme A–induced nuclease activation.

In conclusion, our studies provide mechanistic insight into organ-specific autoimmune diseases by showing that aberrant cytosolic DNA and type I IFN accelerate diabetes. This is a major advance in the understanding of the triggers of autoimmune diabetes and indicates that exogenous nucleic acids from viruses are not the only way that local production of type I IFN is stimulated. The innate immune system reacts to nucleic acids in the cell because they might represent a virus infection. In doing this, the risk of autoimmunity is increased because the innate response, especially type I IFN, can break immune tolerance in T cells that recognize autoantigens with low affinity that would ordinarily be harmless.

Acknowledgments. The authors thank L. Elkerbout, C. Tan, S. Thorburn, E. Wilson, T. Kay, E. Duff, and M. Joffe (St. Vincent’s Institute) for technical support, genotyping, and animal husbandry, and the National Institutes of Health Tetramer Core for providing tetramers.

Funding. This work was funded by a National Health and Medical Research Council of Australia (NHMRC) program grant (APP1037321), an NHMRC and JDRF joint special program grant in type 1 diabetes (APP466658), and fellowships from the JDRF (Z.U.A.M. and B.K.) and NHMRC (M.M.W.C. and H.E.T.). This work received support from the Operational Infrastructure Support Scheme of the Government of Victoria.

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

Author Contributions. Z.U.A.M. and H.S.Q. designed the study, performed experiments, analyzed data, and wrote and revised the manuscript. K.L.G., G.J., J.F.M.D., J.C., P.M.T., E.G.P., L.M., E.P.F.C., S.A., S.F., and C.H. performed experiments, analyzed data, and revised the manuscript. B.K. contributed to conception, design, and interpretation of this work; provided essential reagents; performed experiments; analyzed data; and critically revised the manuscript. A.J.D., J.A.T., M.M.W.C., P.I.B., and T.C.B. contributed to conception, design, and interpretation of this work; provided essential reagents; and critically revised the manuscript. H.E.T. and T.W.H.K. designed the study and wrote the manuscript. H.E.T. 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.