Peripheral Proinsulin Expression Controls Low-Avidity Proinsulin-Reactive CD8 T Cells in Type 1 Diabetes

Type 1 diabetes is a multifactorial immune-mediated disease (1). In both humans and mice, CD8 T cells (CTLs) play an important role in both early events and the final effector phases of diabetes development (2).

Proinsulin (PI), the larger prohormone of the active hormone insulin, is targeted in the autoimmune attack in type 1 diabetes in both mice and humans (3). Self-antigens, including PI, are expressed in the thymus in mice and humans (4–6), controlled by Aire, and PI is not detectable in the thymus in aire^−/− mice (7). In humans, the second most important genetic susceptibility gene after the MHC (Insulin Dependent Diabetes Mellitus2 [IDDM2]), is the insulin 5′ variable number tandem repeat (VNTR) region, controlling expression of thymic and pancreatic PI (8). Mice express PI 1 (PI1) and PI 2 (PI2; homologous to human PI) differing by two amino acids (aas) in the B chain and three aas in the C-peptide. Although PI1 is expressed predominantly in the pancreas (4,9), expression of PI1 in the thymus is much lower than PI2 (10). Because PI2 expression is greater in the thymus, PI2 likely influences central T-cell tolerance to PI.

However, PI-specific T cells escape from thymic negative selection. In the NOD mouse, when PI2 was overexpressed in the thymus on the MHC class II promoter (PI2^tg mice), diabetes incidence was decreased (11,12). Interestingly, NODIns1^−/− mice have reduced diabetes (13), whereas diabetes is accelerated in NODIns2^−/− mice, with 100% developing disease (13,14). PI2 is expressed in pancreatic lymph nodes (PLNs) (15) and is weakly expressed in lymph node (LN) stromal cells (LNSCs) (16). Peripheral Aire-expressing cells in the secondary lymphoid organs also interact with autoreactive T cells (7). These heterogeneous LNSCs, which function in a similar manner to medullary thymic epithelial cells, may be very important in mediating peripheral tolerance (16). Peripheral tissue antigen expression, like PI2, is regulated by deformed epidermal autoregulatory factor (Deaf1) and its splice variants, and the expression of these factors also correlates with severity of disease in NOD mice (17,18).

We have previously cloned highly diabetogenic CTLs (G9C8) from the islets of young prediabetic NOD mice (19), which rapidly transferred diabetes to NOD and NOD.scid mice (2,19). These G9 CTLs recognize insulin B chain, aas 15–23, restricted by MHC class I K^d. In NOD mice, such T cells infiltrate islets at 4 weeks of age (20,21), representing a substantial proportion of the few cells present in the very early infiltrate. We believe that they are very important in the early events leading to diabetes. Interestingly, unlike CTLs recognizing other islet autoantigens, the percentage of the insulin-reactive cells does not increase with time, and T cells responsive to other autoantigens predominate later (22). The insulin B15-23 peptide, common to both mouse PI1 and PI2, binds poorly to K^d, and mutation at position 16 from Y to A abolished the antigen recognition (23).

CTLs and CD4 T cells, both recognizing insulin, have been implicated in pathogenesis of diabetes in mice and in humans (20,24–26). The dominant CD8 epitope (B15-23) overlaps the CD4 epitope of insulin (B9-23) (20). Nakayama et al. (25) showed that when PI1 and 2 were both knocked out and insulin expressing alanine at position B16 instead of tyrosine, designated Y16A, was substituted for the native insulin, the mutated mice (designated Ins1^−/−Ins2^−/−.Ins2*Y16A mice) were protected from diabetes. This mutation removes both the CD4 and CD8 epitopes in the B chain of insulin but is metabolically active, thus preventing death from insulin deficiency.

Our study examines a central question: How is tolerance maintained to autoreactive T cells that respond to low-affinity peptides? We tested the hypothesis that tolerance of insulin-reactive CTLs is influenced by the expression of PI, both centrally and peripherally. We developed model systems to directly study the effect of differing PI expression on the selection and reactivity, in vivo, of the insulin B15-23–reactive CTLs. Using mice that express different levels of PI, we found that decreased PI expression was associated with heightened CTL effector function and diabetes development, related to regulation by LNSC activity.

Mice were housed in microisolators or scantainers in the specific pathogen-free facility at Cardiff University. All procedures were performed in accordance with protocols approved by the U.K. Home Office. NOD/Caj mice were originally obtained from Yale University. NODIns2^−/−, NODIns1^−/−, and NODY16A^tg mice were obtained from The Jackson Laboratory. NOD-PI2^tg mice, with transgenic overexpression of PI2 on the MHC class II promoter (11), were provided by Len Harrison (Walter and Eliza Hall Institute of Medical Research, Parkville VIC, Australia). G9Cα^−/− mice were generated as previously described (2). The mice with mutations in PI genes were bred with T-cell receptor-α (TCR-α) and TCR-β lines generating G9Cα^−/−Ins2^−/− and G9Cα^−/−Ins1^−/−Ins2^−/−Y16A (G9Cα^−/−Y16A) mice (Table 1). The PI2^tg was introduced in the TCR-α line to generate G9Cα^−/−PI2^tg mice. C57BL/6 mice were purchased from Harlan UK Ltd.

Mice were monitored weekly for glycosuria (Bayer Diastix) from 6 weeks of age. Diabetes was diagnosed after two consecutive blood glucose levels above 13.9 mmol/L.

Lymphoid tissues, including thymus, spleen, PLNs, mesenteric LNs (MLNs), and para-aortic LNs (ALNs) were collected from age- and sex-matched G9Cα^−/−, G9Cα^−/−Ins2^−/−, and G9Cα^−/−Y16A mice. Lymphocytes were labeled with antibodies against CD4 (GK1.5), CD8 (53-6.7), CD62L (MEL14), and CD69 (H1.2F3), all from BioLegend; Vβ6 (RR47, BD Biosciences), and live/dead exclusion (eBioscience). Regulatory T cells (Tregs) were identified as CD4, CD25 (ebio3C7), and Foxp3 (FJK-16s, eBioscience) positive. Stromal cells were labeled with CD45 (30-F11), CD31, gp38/podoplanin (BioLegend), CD11c (BD Biosciences), and Ulex europaeus agglutinin 1 (Vector). Flow cytometric samples were collected using Canto II (BD Biosciences) and analyzed with FlowJo 7.6.5 software (Tree Star, Inc.).

For proliferation against islets, splenic CTLs from G9Cα^−/− mice were selected using MACS CD8 microbeads (Miltenyi Biotec). Islets from TCRCα^−/− or Y16A mice were isolated (27,28), trypsinized, and irradiated (4000 Rad); 20 islets were cultured with 10⁵ CTLs for 4 days before overnight pulse with ³H-thymidine. Proliferation (counts per min) was determined using the Microbeta2 Plate counter (PerkinElmer).

For proliferation assays against insulin B15-23 peptide (LYLVCGERG), PLNs were collected from age- and sex-matched mice. Lymphocytes were labeled with 0.5 µmol/L carboxyfluorescein diacetate succinimidyl ester (CFDA-SE; Molecular Probes), following the manufacturer’s protocol. Cells were stimulated with a range (0–4 µg/mL) of peptide for 72 h at 37°C and 5% CO₂. After stimulation, cells were labeled with anti-CD8, anti-CD69, and live/dead exclusion. Flow cytometry was used to analyze proliferation and activation of CD8 lymphocytes.

Splenic CTLs from age- and sex-matched G9Cα^−/−, G9Cα^−/−Ins2^−/−, and G9Cα^−/−Y16A mice were selected using MACS CD8 microbeads (Miltenyi Biotec) and labeled with 2 µmol/L CFDA-SE (Molecular Probes), following the manufacturers’ protocols. Labeled cells (10⁷) were transferred intravenously to 6-week-old female NOD recipients. After 4 days, recipient spleen and LNs were collected, cells were labeled with CD8, CD69, and live/dead exclusion, and were analyzed for proliferation and activation by flow cytometry as above.

Splenic CTL were positively selected from age- and sex-matched G9Cα^−/−, G9Cα^−/−Ins2^−/−, and G9Cα^−/−Y16A mice. Thymocytes were incubated with plate-bound anti-CD4 (GK1.5, BioLegend) for 1 h at 4°C. Nonadherent T cells were collected, and single-positive CTLs were magnetically sorted. P815 cells were labeled with PKH-26 (Sigma-Aldrich), following the manufacturer’s instructions. CTLs were incubated with peptide-loaded P815 cells for 16–18 h at 37°C and labeled with TO-PRO-3 (Molecular Probes) immediately before flow cytometric analysis to identify dead cells and determine specific lysis of target P815 cells (2).

TCR expression levels were compared in age- and sex-matched G9Cα^−/−, G9Cα^−/−Ins2^−/−, and G9Cα^−/−Y16A mice. CTLs were positively selected from PLN and spleens. Lipopolysaccharide (LPS)-activated dendric cells (DCs) from NOD mice were used to stimulate CTLs with a range (0–4 µg/mL) of insulin peptide B15-23 (LYLVEGERG) over 72 h. Cells were labeled with CD4, CD8, Vβ6, and live/dead exclusion. Mean fluorescence intensity of TCR was analyzed by flow cytometry.

LNSCs were prepared from NODCα^−/− (wild type), NODIns1^−/−, NODIns2^−/−, Y16A, and C57BL/6 mice. PLNs were pooled from 18–20 mice and teased open using a 30-gauge needle. Tissue was digested with 0.5 mg/mL Dispase II (Sigma-Aldrich), 0.5 mg/mL collagenase P (Roche), and 100 units/mL DNase I (Roche) in RPMI-1640. Samples were digested at 37°C in a shaking water bath for 5-min intervals, tissue debris was allowed to settle, and the supernatant containing single cells was collected (29). Hematopoietic cells were removed using complement depletion, following the manufacturer’s protocol. Briefly, cells were labeled with purified anti-CD45 for 45 min at 4°C. Cells were washed, incubated for 60 min at 37°C in Low-Tox-M Rabbit Complement (Cedar Lane), and then plated at 2 million cells/well in six-well plates in RPMI-1640 with 10% FBS. Nonadherent T cells were removed after 48 h, and stromal cells were grown at 37°C in 5% CO₂ for 7–10 days.

Bone marrow DCs, from NOD-PI2^tg mice, were grown in medium containing granulocyte-macrophage colony-stimulating factor (1.5 ng/mL) and stimulated overnight with LPS (1 µg/mL). Splenocyte CTLs were negatively selected using the MACS CD8 T-cell isolation kit II (Miltenyi Biotec) and labeled with 0.5 µmol/L CFDA-SE. Pancreatic LNSCs were collected from 6-well plates with 0.25% trypsin and seeded at 10⁴ cells/well in 96-well round-bottomed plates for 24–48 h to allow adhesion. LPS-activated DCs (2 × 10⁵/well) and CFDA-SE–labeled CTLs (1 × 10⁵/well) were added to wells containing LNSCs. T cells were collected after 72 and 96 h and labeled with anti-CD8, anti-CD69, and live/dead exclusion. Proliferation and activation of T cells were analyzed by flow cytometry as above.

Means were compared using one-way ANOVA with P < 0.05 considered significant. Kaplan-Meier survival analysis was used to evaluate onset of diabetes (GraphPad Prism software).

TCR transgenic mice expressing the clonotypic insulin-reactive G9C8 TCR on the NOD.TCRCα^−/− genetic background (2) were designated G9Cα^−/− mice. G9Cα^−/− mice were crossed to NODIns2^−/− mice (14) to generate G9Cα^−/−Ins2^−/− mice; thus, we studied cells that developed in the absence of PI2. We also crossed G9Cα^−/− mice to the Ins1^−/−Ins2^−/−.Ins2*Y16A mice (25) to generate G9Ins1^−/−Ins2^−/−.Ins2*Y16A mice (designated G9Cα^−/−Y16A mice) expressing only mutated insulin that cannot be recognized by the G9 CTLs. The loss of recognition of the insulin expressed in the Ins1^−/−Ins2^−/−.Ins2*Y16A mice is shown in Supplementary Fig. 1, with much reduced G9 CTL proliferation to Y16A islets compared with NOD islets. Finally, the G9Cα^−/− mice were crossed to mice that expressed PI2 as a transgene on the MHC class II promoter (11). These mice were bred in-house and are summarized in Table 1.

Although G9Cα^−/− mice do not develop spontaneous diabetes (2), here we found that in the absence of PI2, 25% of the male, but not female, G9Cα^−/−Ins2^−/− mice developed spontaneous diabetes by 10 weeks of age (Fig. 1). The diabetes incidence contrasts with wild-type male NOD mice, where diabetes only occurs after 18 weeks in our colony. In previous studies, diabetes was accelerated in NODIns2^−/− female mice compared with NOD mice (13,14).

To further investigate the selection of insulin-reactive CTLs, we examined the development of CTLs in the thymus of G9Cα^−/−, G9Cα^−/−Ins2^−/−, and G9Cα^−/−Y16A mice by flow cytometry (Fig. 2A–C). Overall, no significant differences were seen in the CTL frequency (Fig. 2D), in the total numbers of thymocytes (Fig. 2E), or between male and female mice (data not shown). Specifically, neither lack of PI2 nor lack of the specific epitope recognized by the insulin-reactive T cells altered thymic selection. Furthermore, no deletion of the CD8 thymocytes was seen in the G9Cα^−/−PI2^tg mice that overexpressed PI2.

We next studied G9 cells in the peripheral lymphoid tissues, focusing on the spleen, PLN, MLN, and distant ALN. There were clear differences in the frequency of the G9 CTLs between the G9Cα^−/−, G9Cα^−/−Ins2^−/−, and G9Cα^−/−Y16A mice (Fig. 3) in the spleen and the different LNs. Unlike the thymus, the percentage of the G9Cα^−/−PI2^tg CTLs was considerably reduced in the periphery, suggesting that these cells had been deleted (Fig. 3). As there were few G9 cells in the periphery of G9Cα^−/−PI2^tg mice, these mice were not a focus for further analysis. Conversely, G9 cells were increased in the LNs of the G9Cα^−/−Ins2^−/− mice, which may reflect that when Ins2 is deficient, the G9 cells are released from regulation, become activated, and expand. In contrast, the spleen of the G9Cα^−/−Y16A mice was smaller, as described for the original Y16 mice (25), possibly resulting from the metabolic requirement for insulin to maintain immune cells in addition to the need for T cells to recognize antigen for survival. The mutated insulin is metabolically active but may be less efficient or not as highly expressed as native PI2.

To investigate maturation of the insulin-reactive CTLs, G9 CTLs were tested for proliferation to insulin B15-23 peptide in culture. As indicated in Fig. 4A, CFDA-SE–labeled PLN CTLs from G9Cα^−/−Y16A mice proliferated more than cells from G9Cα^−/−Ins2^−/− mice, which in turn had a greater response than the cells from G9Cα^−/− mice. This result was recapitulated in vivo (Fig. 4B) when the CFDA-SE–labeled CTLs were transferred into 6-week-old NOD female mice to test for proliferation to endogenously presented antigen in the PLNs, as described previously (2). CTLs isolated from an environment lacking normal insulin expression exhibited the most reactivity to endogenous insulin.

Furthermore, we examined cytotoxicity of the G9 CTLs toward insulin-peptide–loaded targets. The CTLs from G9Cα^−/−Y16A mice exhibited greater cytotoxicity in response to the target antigenic peptide compared with the CTLs from G9Cα^−/− or G9Cα^−/−Ins2^−/− mice (Fig. 4C), similar to the proliferative response. This was only seen with peripheral lymphoid cells. There was no difference in the cytotoxic capacity of single-positive CD8 thymocytes (Fig. 4D), indicating that the cells acquired this difference in their activation after release from the thymus. Thus, peripheral, but not central, insulin-reactive CTLs from mice lacking their target antigenic epitope were more cytotoxic on encountering their cognate antigen.

To further investigate the phenotype of G9 CTLs in the different lymphoid tissues, we assessed the cells in the PLNs, MLNs, and ALNs, testing for intrinsic differences in activation status. There was a significant reduction in CD62L expression in CTLs of the PLNs in the G9Cα^−/−Ins2^−/− mice (Fig. 5A), indicating a greater shift in the balance of activated memory T cells. In the G9Cα^−/−Ins2^−/− mice, although PI2 is not expressed, PI1 is still present in the mice, and therefore, the cells will be exposed to endogenous antigen (the B15-23 epitope is identical in PI1 and PI2). This alteration in the CD62L expression was not seen in cells from the G9Cα^−/−Y16A mice, which lacked the antigenic epitope recognized by G9CTL and therefore had no cognate antigenic stimulation in the periphery (Fig. 5).

To assess whether the different levels of activation of the G9 T cells shown in Figs. 4 and 5 were associated with TCR expression, we tested the baseline TCR level in the different strains. We found no overall difference in the levels of the clonotypic TCR, measured by Vβ6 monoclonal antibody staining, shown at day 0, which is the baseline before stimulation, in both the spleen and PLNs. Similarly, there were no significant differences overall after stimulation of the cells with peptide (Fig. 6A and B). No differences in the frequency of CD25+FoxP3+ Tregs indicated that altered frequencies of Tregs were not responsible for these differences (Fig. 6C and D).

As ectopic antigen expression in LNSCs has been implicated in peripheral tolerance (7,16,30,31), we tested the G9 T-cell interactions with LNSCs from mice with different levels of PI expression. We extracted LNSCs, which are a heterogeneous group of cells (7,29,32) that are CD45⁻ and CD11c⁻, with most identified as fibroblastic reticular cells, defined as gp38⁺ (podoplanin) and CD31⁻ (Fig. 7A). We then investigated the effects of LNSCs on G9 CTL proliferation. Our earlier experiments suggested that chronic exposure to PI in development influenced the baseline reactivity of the G9 CTLs (Fig. 4). We hypothesized that expression of PI in LNSCs influenced this reactivity. Thus, we tested proliferation of the G9 CTLs from each of the different PI-expressing strains of mice, in response to the specific antigen, in the presence of pancreatic LNSCs derived from each of the different PI-expressing mouse strains.

To ensure that we were testing the effect of endogenous PI in the LNSCs, we stimulated G9 CTLs in the absence of exogenous peptide. For this, we used PI2 transgenic DCs that express PI transgene in antigen-presenting cells and present the insulin peptide, removing any possibility of additional antigen processing of exogenous antigen by the LNSCs. This allowed us to study the cumulative effects of DC stimulation and potential LNSC effect on T-cell activity (Fig. 7B). We showed that the proliferation of each of the insulin-reactive CTLs was inhibited in the presence of wild-type stromal cells that express both PI1 and PI2, or in the absence of either PI1 (from Ins1^−/−) or PI2 (from Ins2^−/−) alone (Fig. 7C). There was no obvious inhibition by the stromal cells from the Ins1^−/−Ins2^−/−.Ins2*Y16A mice, where no PI expressed was recognizable by the G9 CTLs. In addition, there was no obvious inhibition by LNSCs extracted from C57BL/6 mice, which express K^b, an MHC class I molecule not recognized by G9 cells.

We have studied the development and activation of low-avidity insulin-reactive CTLs (23) that become highly diabetogenic when fully activated. We show that, irrespective of the endogenous PI expression in the mice, the thymic CTL frequency and function was not different. However, in the periphery, although similar in number, the insulin-reactive CTLs demonstrated increased proliferative and cytotoxic capacity when they developed in the absence of PI2 (G9Cα^−/−Ins2^−/− mice), and even more when cognate peptide was absent (Y16A mice). Furthermore, male G9Cα^−/−Ins2^−/− mice developed spontaneous autoimmune diabetes. Why this has occurred only in male mice is currently not known, but the male mice may possibly be more sensitive because of hormonal effects. Conversely, where PI2 was overexpressed, the G9 cells were deleted in the periphery. Pancreatic LNSCs from mice expressing normal PI levels had inhibitory effects on insulin-reactive T-cell proliferation, and there was no inhibition by LNSCs from mice expressing mutated PI that the T cells could not recognize. Our results indicate that LNSCs regulate the activation and function of insulin-reactive G9 T cells.

We focused on these low-avidity insulin-reactive T cells, a prevalent population in early insulitis (20,33), that do not expand later in the disease process, as do islet-specific glucose-6-phosphatase catalytic subunit-related protein (IGRP)­–responsive T cells (22,34). In addition, diabetes requires PI autoreactivity, even if responses to IGRP become dominant later on (35). We observed that in the G9Cα^−/− mice, where the vast majority of the cells are G9 insulin-reactive CTLs, immunizing the mice with peptide and adjuvant induced diabetes (2). Furthermore, purified splenic G9 cells, once activated, can transfer diabetes very rapidly. This suggests that tolerance is maintained under normal circumstances and in the absence of other cell specificities and that additional stimuli are required to raise the cells over the threshold of activation. Our current study demonstrates that when expression of PI2 was reduced, G9 cells proliferate more and are more potently cytotoxic both in vitro and in vivo.

How is this tolerance effected? Previous investigators have suggested that thymically expressed insulin modulates insulin autoreactivity (10), and other evidence suggests that this is important for CD4 T cells (11). PI2 is expressed in the thymus, whereas PI1 is expressed at a lower/undetectable level (4–6). However, central tolerance does not play a role for these low-avidity CTLs, because no differences in their frequency were seen in the thymus or in their cytotoxic potential, irrespective of whether PI expression was increased or decreased. For these low-avidity CTLs, peripheral tolerance mechanisms are dominant, and they do not cause diabetes.

Peripheral tolerance mechanisms include induction of ignorance, anergy, and deletion (36) or regulation induced by Tregs. Here, tolerance was not related to the frequency of Tregs. The expression level of TCR (37) or CD8 coreceptor (38) may contribute to CTL responsiveness, including downregulation on activation (37); however, this was not a mechanism that contributed to differences in responsiveness here either, because the levels of TCR were not higher in those cells that responded more to the antigen.

Although DCs play a major role in determining activation or tolerance of peripheral T cells, in recent years, ectopic expression of antigen in peripheral LNSCs has been shown to modulate immune function (39). A comprehensive study by Yip and Fathman (40) showed that PI2 is expressed in PLNs. In this study, we demonstrated that in the absence of PI2, but in the presence of PI1, more CTLs in the PLNs became effector memory cells, with a reduction in CD62L expression. This phenomenon was not seen in the G9Cα^−/−Y16A mice, in which G9 cells are completely unable to target the endogenous insulin expressed. Therefore, the activation of the insulin-reactive CTLs requires not only release from tolerance mechanisms but also peripheral stimulation. The cells unable to recognize endogenous insulin may be released from inhibition; however, they cannot recognize endogenous antigen and thus do not differentiate into effector memory cells.

LNSCs have been shown to induce tolerance to CD4 T cells (15,17) and are able to present antigen (31). We have demonstrated that the level of antigen expressed in these LNSCs may be very important in determining the outcome of tolerance for autoreactive CTLs. In this case, the more PI antigen expressed, the more effective the tolerance exerted by the LNSCs on the peripheral T cells. When the insulin-reactive CTLs were given the same antigenic stimulus, the reduced proliferation was greater in the presence of the LNSCs derived from mice expressing more endogenous PI. Although the methods we used did not show direct antigen presentation by the LNSCs to the T cells, we only observed the inhibitory effect on G9 T cells when MHC-matched stromal cells were used. This suggests that the tolerance is exerted by the LNSCs directly and relates to their ability to interact with the T cells and present antigen because control LNSCs expressing nonmatched MHC had no suppressive effect.

Low-affinity CTLs have also been found in humans (26,41). The expression of PI, controlled by an upstream VNTR, is an important genetic susceptibility factor, second only to the MHC (6,8,42). Humans have only a single PI, unlike mice. Nevertheless, our mice that express varying levels of PI may recapitulate some of the expression differences seen in humans. We suggest that in individuals expressing different levels of PI, there may be ectopically expressed antigen outside the islet β-cells, encoded by the VNTR, and a similar mechanism of tolerance may apply. In susceptible people who express IDDM2 class I VNTR that leads to increased susceptibility to type 1 diabetes, this may lead to increased activity of insulin-reactive T cells because the peripheral tolerance mechanisms may be decreased. This remains to be tested in humans.

In conclusion, our studies have indicated that for low-avidity autoreactive CTLs, mechanisms of central tolerance causing negative selection are insufficient to remove such cells, even when antigen expression is increased. Peripheral tolerance is clearly the main method by which such cells are controlled. Expression of self-antigen recognized by the T cells in the peripheral LNs, specifically in the stromal cells, plays an important role in reducing autoreactivity. Further understanding of critical tolerogenic pathways will be important in identifying how these mechanisms may be boosted for the development and application of immune-modulating therapeutics for type 1 diabetes.

Acknowledgments. The authors thank members of Joint Biological Services of Cardiff University for animal care, and Len Harrison and Andrew Lew, from the Walter and Eliza Hall Institute of Medical Research (Parkville, Victoria, Australia), for gift of PI2^tg mice before they were available commercially.

Funding. This work was supported by the Medical Research Council (grant G0901155). T.C.T. is supported by a postdoctoral fellowship from JDRF (3-PDF-2014-211-A-N). J.A.P. was supported by a Diabetes UK Studentship (08/3767). A.T. and S.A. were funded by CUROP funds supporting summer students at Cardiff University.

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

Author Contributions. T.C.T., J.A.P., E.D.L., S.J.H., J.B., J.D., A.T., S.A., P.E., and L.K.S. carried out experiments. T.C.T., J.A.P., E.D.L., S.J.H., L.W., and F.S.W. analyzed data. T.C.T., J.A.P., S.J.H., L.W., and F.S.W. edited the manuscript. T.C.T. and F.S.W. wrote the manuscript. F.S.W. conceived the project. F.S.W. 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. Parts of this study were presented in abstract form at the Diabetes UK Annual Professional Conference 2011, 30 March–1 April 2011, and the 13th International Congress of the Immunology of Diabetes Society, Lorne, Victoria, Australia, 7­–11 December 2013.