Protection of Islet Grafts Through Transforming Growth Factor-β–Induced Tolerogenic Dendritic Cells

C57Bl/6 mice, rat insulin promoter (RIP)–tumor necrosis factor (TNF) NOD (29), Tet-TGF NOD (28), G9 T-cell receptor (TCR) transgenic NOD mice (30,31), OT-II TCR transgenic mice with OVA-specific CD4⁺ T cells (32), OT-I TCR transgenic mice with OVA-specific CD8⁺ T cells (33), NOD-scid mice, and Tet-TGF-β–NOD-scid mice (Tet-TGF NOD bred to a NOD scid background) were bred and maintained under specific pathogen-free barrier conditions. In the Tet-TGF NOD mice, islet-specific TGF-β expression is controlled by the doxycycline-controlled on/off switch (34). Littermates of these mice that were exposed to the same levels of doxycycline but did not express the Tet-TGF-β transgene were used to isolate islets for control transplants. Diabetes was detected using Diastix reagent strips (Bayer Diagnostics, Basingstoke, U.K.) and confirmed by a blood glucose measurement of >13.3 mmol/L, using a Breeze2 blood glucose meter (Bayer). All animal work was conducted according to U.K. Home Office project license regulations after approval by the Ethical Review Committee of the University of Cambridge.

CD4 allophycocyanin (APC; clone L3T4), CD8 APC or eFluor 450 (clone M1/70), CD44 phycoerythrin (PE) or eFluor 450 (clone IM7), CD80 fluorescein isothiocyanate (FITC) (clone 16–10A1), CD86 PE (clone GL-1), CD103 FITC (clone 2E7), CD40 PE (clone 1C10), interferon (IFN)-γ PE or APC (cloneXMG1.2), interleukin (IL)-10 PE (JES5–16E3), Foxp3 PE (clone FJK-16s), and H2K^b APC (clone AF6–88.5.5.3) antibodies (eBioscience); H2K^d APC (clone SF/111) and CD45 PerCp-Cy5.5 (clone 30-F11) antibodies (BD Bioscience); and major histocompatibility complex (MHC) II PE or FITC (clone 7FT) antibody (Serotec) were used. Intracellular staining was performed using the fixation/permeabilization buffer kit from eBioscience. Flow cytometry was performed on a CyAn ADP (Dako), and data were analyzed using FlowJo software (Tree Star).

Pancreatic islets were isolated through inflation of the pancreas via the bile duct (35), and islet transplantation was performed according to standard protocols (36). Confirmed diabetic recipient mice received between 300 and 500 islets, resulting in ∼15 islets per gram of body weight. For cotransplantation with DCs, 5 × 10⁵ in vitro generated BMDCs were incubated overnight with or without 5 ng/mL TGF-β (Peprotech) and then were washed twice and mixed with the islets.

For intracellular cytokine detection using flow cytometry, cells were resuspended in Dulbecco's modified Eagle's medium supplemented with 10% FCS, 50 μmol/L β-mercaptoethanol, L-glutamine, and sodium pyruvate and 50 units/mL penicillin and streptomycin (GIBCO). Cells were seeded into the wells of a 96-well plate and stimulated with phorbol myristic acid (12.5 ng/mL) and ionomycin (125 ng/mL) (both Sigma-Aldrich) for 5 h at 37°C with 5% CO₂. Monensin (BD Bioscience) was added during the last 4 h of culture at a concentration of 0.13 µL/200 µL of culture. Cytokines in BMDC culture supernatants were detected by electrochemical luminescence immunoassay using a mouse seven-plex proinflammatory cytokine kit (MesoScale Discovery).

Bone marrow was harvested from the femurs of NOD or C57Bl/6 mice and cultured in 50 mL Dulbecco's modified Eagle's medium supplemented as described plus 10 ng/mL granulocyte–macrophage colony-stimulating factor (Peprotech) for 10 days.

For bead uptake assays and analysis of surface markers, the BMDCs were conditioned for 24 h with or without 5 ng/mL TGF-β (Peprotech), activated overnight with 100 ng/mL lipopolysaccharide (LPS), and then washed twice. Fluorescent microbeads (Bangs Laboratories) were coated with insulin β−chain amino acids 15-23 peptide (insB15-23) according to the manufacturer’s instructions. The beads were added to the BMDCs at a concentration of 25 μg beads/10⁶ cells and incubated for 2 h at 37°C; afterwards, bead uptake was assessed using flow cytometry.

For proliferation assays, OVA-specific CD4⁺ T cells from OT-II TCR transgenic mice, OVA-specific CD8⁺ T cells from OT-I TCR transgenic mice, or insB15-23–specific CD8⁺ T cells from G9 TCR transgenic mice were sorted using negative-selection magnetic bead kits from MACS Miltenyi. The purity was always >95%. For OVA-specific cultures, BMDCs were pulsed with 10 μg/mL OVA for 24 h with or without 5 ng/mL TGF-β (Peprotech) and then were washed and incubated with 100 ng/mL LPS overnight. The next day, the BMDCs were washed again, and 5(6)-carboxyfluorescein diacetate succinimidyl ester (CFSE)-labeled (Invitrogen) T cells were added at a concentration of 2 × 10⁵ cells per well. For insulin β-chain–specific cultures, BMDCs were treated as described, but 1 μg/mL insB15-23 peptide was added after activation with or without TGF-β treatment and was present in the cultures during incubation.

After 72 h, the cells were stained for surface markers and 7-amino-actinomycin D (7AAD) to exclude dead cells (BD Bioscience) and were assessed by flow cytometry.

Graft-bearing kidneys were snap-frozen in liquid nitrogen. Guinea pig anti-insulin antibody (DAKO) was detected with anti-guinea pig Alexa 546 (Molecular Probes). Antibodies against Foxp3 (eBioscience), CD4, CD8, B220, and CD11b (BD Bioscience) were detected with anti-rat Alexa 488 (Molecular Probes). Nuclei were visualized with DAPI (Molecular Probes). The sections were viewed with the Axioskop 2 plus fluorescence microscope (Zeiss), and three fields of view were counted for each graft.

Grafts were removed 5 or 7 days after transplantation as indicated in the text, and mRNA was prepared using the Qiagen miniprep kit according to the manufacturer’s instructions. Real-time PCR was performed in technical duplicates using an Applied Biosystems 7500 fast thermocycler and QuantiFAST fast cycling SYBRGreen master mix (Qiagen). Real-time PCR primers for genes of interest were supplied by Qiagen. Relative gene expression was normalized to the reference gene glyceraldehyde 3-phosphate dehydrogenase using the Δ cycle threshold (ΔCT) method.

As transplant recipients we used either wild-type (WT) NOD mice or RIP-TNF NOD mice that had been genetically modified to express TNF under the influence of RIP, resulting in a 100% incidence of diabetes (29) on the NOD mouse genetic background. Diabetes develops in RIP-TNF NOD mice in a manner similar to that of NOD mice, but disease progression is more rapid and synchronized (29). Diabetic NOD mice (Fig. 1A) or RIP-TNF NOD mice (Fig. 1B) received an islet graft from TGF-β–NOD-scid mice on day 0. After 21 days, TGF-β expression was switched off by administration of doxycycline-supplemented water. Diabetic mice receiving control NOD-scid islets (normal NOD islets that are not affected by preexisting immune infiltration) rapidly rejected these, and the mice were all diabetic again within 7 days. In contrast, mice receiving the TGF-β–expressing islets maintained normal blood glucose for significantly longer. Although recipients eventually rejected the graft and became diabetic after TGF-β expression was switched off, graft failure in all recipients was not apparent until ∼40 days after TGF-β silencing. There was a possibility that the transgenic expression of TNF in any remaining β-cells in the RIP-TNF NOD mouse recipients led to overturning of tolerance to the graft; therefore, we also investigated graft survival in diabetic WT NOD recipients. There was no statistically significant difference in graft rejection between WT NOD mouse recipients and RIP-TNF NOD mouse recipients (Fig. 1A and B), indicating that the preexisting immune response to islet antigens is sufficient to precipitate rejection after TGF-β expression is silenced, even when there is no additional immunogenicity provided by residual TNF-expressing β-cells. Because the expression of TNF in the endogenous recipient islets before progression to diabetes made no difference for graft survival, the subsequent experiments all have been performed on RIP-TNF NOD mouse recipients. The TGF-β–expressing grafts displayed less infiltration of immune cells than control grafts as measured by flow cytometry of whole grafts and gating on CD45⁺ cells (Fig. 1C). There were no significant differences in the graft CD4⁺Foxp3⁺ Treg cell populations between the two recipient groups (Fig. 1D). Immunofluorescence microscopy of sections prepared from the grafts 5 days after transplantation also showed reduced infiltration of T cells in the TGF-β–expressing grafts (bottom, Fig. 1E) compared with WT grafts (top, Fig. 1E), but infiltration with CD11b⁺ and CD11c⁺ antigen-presenting cells was similar between the two groups. Few B cells could be detected in the grafts at this time point.

Statistical analysis of several grafts demonstrated that TGF-β–expressing grafts had less infiltration of T cells but a similar presence of antigen-presenting cells expressing CD11b and CD11c (Fig. 1F).

The lower infiltration of T cells into the TGF-β–expressing grafts could be caused either by reduced migration to the graft or by decreased activation in the draining lymph node. To test this, we assessed responses in the graft-draining lymph nodes. Total cell numbers in the draining renal lymph nodes (RLNs) of TGF-β–expressing grafts were lower than in those draining control grafts (Fig. 2A), whereas the cell numbers in draining RLNs from both groups were higher than in control RLNs (Fig. 2A). Within these populations, the percentages of CD4⁺ and CD8⁺ T cells were similar (Fig. 2B and C), and the percentages of Foxp3⁺ Treg cells also were similar within the CD4⁺ population (Fig. 2D). In accordance with these results, the absolute numbers of both CD4⁺ and CD8⁺ T cells were significantly lower in draining RLNs of TGF-β–expressing grafts than in those draining control grafts, whereas there was no significant difference detected in the absolute numbers of Foxp3⁺ Treg cells (Supplementary Fig. 1). The activation status of the T cells in the draining RLNs of TGF-β–expressing grafts also was decreased, as demonstrated by the lower production of proinflammatory cytokine IFN-γ as well as immunosuppressive cytokine IL-10 (Fig. 3A and B) after ex vivo restimulation with phorbol myristic acid and ionomycin. Examining the ratios of IL-10 produced compared with IFN-γ, it appeared that there was a skewing toward a more IL-10–dominant and less inflammatory profile in the draining RLNs of TGF-β–expressing grafts (Fig. 3C).

The results depicted in Fig. 1E and F showed that although T-cell infiltration into grafts was reduced in the presence of TGF-β, the presence of antigen-presenting cells expressing CD11c and CD11b was not affected. We hypothesized that exposure to TGF-β would change the capacity of these antigen-presenting cells to efficiently present antigen to islet-specific cells and thus prevent recurring islet destruction. To test this, we used the well-described OVA-TCR system (32,33) and we performed in vitro experiments using BMDCs that we exposed to TGF-β. Exposure to TGF-β during pulsing with whole OVA made BMDC less efficient at activating both CD8⁺ OVA–specific OT-I T cells (Fig. 4A) and CD4⁺ OVA–specific OT-II T cells (Fig. 4B). This decrease in activation was not caused by decreased antigen uptake because TGF-β–exposed BMDCs took up fluorescent beads as well as control BMDCs (Fig. 4C). To find out why the TGF-β–exposed BMDCs induced weaker proliferation, we assessed expression of costimulatory molecules on their surface. We found that after LPS stimulation, TGF-β–exposed BMDCs displayed a small but significant decrease of surface expression of costimulatory molecules CD80 and CD86 (Fig. 5A, top), which are crucial for the second activating signal needed in addition to TCR-MHC interaction, mediated through CD28 on the T cell (37). The TGF-β–exposed BMDCs did not show any decrease in surface expression of MHC molecules (Fig. 5A, bottom). TGF-β–exposed BMDCs also produced lower levels of inflammatory cytokines after LPS stimulation, showing three-fold reduced production of IL-12 p70 as well as reduced TNF, IL-6, and IL-1β. In contrast to the reduction observed in these proinflammatory cytokines, production of the anti-inflammatory cytokine IL-10 was preserved in TGF-β–treated DCs. KC/GRO levels (also known as CXCL1) were not different between TGF-β–treated DCs and controls. Levels of IFN-γ were very low in both (Fig. 5B). To further investigate whether the effect of TGF-β was primarily on antigen uptake or processing or on surface costimulatory molecule expression, we tested whether the suppressive effect of TGF-β was still apparent when bypassing antigen uptake through use of a short peptide for T-cell stimulation. We used the diabetes-relevant insulin peptide–specific TCR transgenic system G9 TCR transgenic NOD mice (30,31), which were created using an islet-specific CD8⁺ T-cell clone isolated from infiltrated NOD mouse islets. BMDCs were exposed to TGF-β and then pulsed with the insB15-23. The pulsed BMDCs were used as APC in culture with the insB15-23–specific G9C8 TCR transgenic CD8⁺ T cells; although no antigen uptake or processing is required for T cell stimulation in this culture, there was a marked decrease in T-cell proliferation in cultures in which BMDCs had been exposed to TGF-β (Fig. 5C).

To investigate whether islet graft expression of TGF-β affected DCs in vivo as seen in the in vitro experiments, we isolated CD11c⁺ cells from the draining RLNs of graft recipients and assessed surface expression of CD80, CD86, and MHC II. We found that CD80 expression was significantly reduced in DCs from draining RLNs of TGF-β–expressing grafts (Fig. 6A, top). There was a trend toward lower CD86 expression in DCs from draining RLNs of TGF-β–expressing grafts (Fig. 6A, middle), but this was not statistically significant. Just as in the in vitro experiments, MHC II expression was similar between groups (Fig. 6A, bottom panels). To further assess expression of immunosuppressive molecules in TGF-β–expressing grafts, we performed PCR analysis of grafts removed from the recipients 5 or 7 days after transplantation (Supplementary Fig. 2). We investigated expression of proinflammatory cytokines TNF and IFN-γ, anti-inflammatory TGF-β and IL-10, and also markers of either alternative or classical activation of APC such as arginase 1 and inducible nitric oxide synthase (38,39). Of all of these markers, only TGF-β was consistently elevated as expected in all TGF-β–expressing grafts. There was a trend toward higher expression of arginase 1, which was higher in three out of four TGF-β–expressing grafts, but the mixed nature of the sampled tissue and variability between samples make it difficult to draw any conclusions about the gene expression of DCs and macrophages within the grafts. To investigate whether islet-specific T-cell proliferation was affected by the reduced DC activation in graft-draining RLNs, we performed adoptive transfer of CFSE-labeled insB15-23–specific G9 transgenic T cells (30,31) into graft recipients and determined their proliferation. These cells proliferated less well in the RLN-draining TGF-β–expressing grafts (Fig. 6B), indicating that DCs that have been exposed to TGF-β are less effective in activating the β-cell–specific T-cell response necessary for islet destruction and recurrence of type 1 diabetes.

Finally, we wished to determine whether in vitro generated BMDCs exposed to TGF-β, impaired in their expression of costimulatory molecules and their production of proinflammatory cytokines, could actually prevent or delay rejection of syngeneic islets transplanted into a diabetic NOD mouse with an already established anti-islet antigen immune memory. We mixed 5 × 10⁵ control BMDCs or BMDCs that had been exposed to TGF-β for 24 h with the islets for grafting. Recipients of islets mixed with any type of BMDC retained their grafts for longer than recipients that received syngeneic islets without any BMDCs. The diabetic recipients that received the TGF-β–exposed BMDCs along with the islet graft retained graft function significantly longer than recipients of grafts with control BMDCs, demonstrating that TGF-β–generated BMDCs can induce long-lasting graft acceptance (Fig. 7). Recipients of WT syngeneic NOD islets without any BMDCs mixed in rejected the grafts within 10 days, usually sooner (Fig. 1A). Although mice that received control BMDCs mixed with the islet graft eventually rejected their grafts, they survived more than twice as long (≥20 days) as WT NOD islet grafts, indicating that even BMDCs without any TGF-β conditioning can affect graft survival. Mice that received islet grafts mixed with TGF-β–exposed DCs displayed considerable infiltration of CD8⁺ T cells 10 days after grafting, at levels similar to those receiving islets mixed with control DCs (Supplementary Fig. 3A). In mice that received TGF-β–exposed DCs and control DCs mixed with their islet grafts, there was a considerable infiltration of Foxp3⁺ cells around the grafted islets, indicating no added recruitment of such cells by the TGF-β–exposed DCs (Supplementary Fig. 3A). At later time points ∼3 weeks after transplantation, the recipients of control DCs with mixed islet grafts developed recurrent diabetes, and at that time point there was, as expected, a large infiltration of CD8⁺ T cells (Supplementary Fig. 3B). Perhaps less expected was the considerable presence of Foxp3⁺ regulatory cells that, despite evidently homing to the graft, failed to prevent rejection (Supplementary Fig. 3B, top right). The long-term protected graft in a recipient of islets mixed with TGF-β–exposed DCs, still euglycemic 4 months after transplantation, displayed considerable infiltration of both CD8⁺ cells (Supplementary Fig. 3B, bottom left) and Foxp3⁺ cells (Supplementary Fig. 3B, bottom right).

The immunosuppressive qualities of TGF-β make it an attractive candidate for achieving suppression of autoimmune responses. However, the cytokine has effects on many pathophysiological processes including cancer and fibrosis (40,41). For TGF-β to be used therapeutically, it is crucial to control the site and duration of expression to harness the power of TGF-β for immunotherapy while avoiding undesired side effects.

In our present study, we found that expression of TGF-β–protected islet grafts from memory islet–specific responses in autoimmune diabetic recipients, and that the resulting reduced infiltration of T cells into the grafts corresponded with reduced activation of T cells in the graft-draining lymph nodes. The fact that the TGF-β–expressing grafts displayed similar infiltration of APCs indicated that the APCs were not inhibited in homing to the grafts but that they were affected by the local TGF-β expression in ways that reduced their antigen-presenting abilities. In vitro experiments demonstrated that TGF-β–exposed BMDCs displayed lower levels of costimulatory molecules on their surface; produced less IL-12 p70, TNF, and IL-6; and were impaired in activating antigen-specific T cells. This was also true in vivo because adoptively transferred islet antigen–specific T cells proliferated less well in the RLNs draining TGF-β–expressing grafts, and DCs isolated from these lymph nodes displayed lower levels of costimulatory molecules. Cotransplantation of islets with in vitro generated BMDCs conditioned with TGF-β resulted in prolonged graft survival.

Tolerogenic DCs are immature, maturation-resistant, or, alternatively, activated DCs that have a low ratio of costimulatory to inhibitory signals and produce low levels of IL-12p70. Tolerogenic DCs can be generated in vitro by coculture with various immunosuppressive and anti-inflammatory agents such as IL-10, prostaglandin E2, and vitamin D3. Immunosuppressive drugs such as cyclosporine, tacrolimus, rapamycin, mycophenolate mofetil, and corticosteroids also have been used to promote tolerogenic DC differentiation in vitro, because all these molecules prevent maturation of DCs and impair their capacity to produce IL-12 (42). Tolerogenic DCs generated through culture with dexamethasone and vitamin D3 can inhibit experimental collagen-induced arthritis severity and progression in an antigen-dependent manner (43). Injection of alternatively activated macrophages can reduce and prevent experimental autoimmune encephalomyelitis (44,45). Prevention and significant delay of development of diabetes in NOD mice have been achieved by administration of tolerogenic DCs differentiated in the presence of IL-10 (46), IL-4 (47,48), or flt-3L (49) and also through administration of alternatively activated macrophages (50).

The mechanisms through which these tolerogenic DCs suppress immune responses vary depending on the protocol. Because the tolerogenic DCs express low levels of IL-12 p70, they are less effective at promoting Th1 T-cell responses, and their lower levels of costimulatory molecules reduce their capacity to induce T-cell activation. Many groups also report that tolerogenic DCs induce preferential proliferation and differentiation of Treg (51–53) or Th2 cells (47,48), whereas others report that tolerogenic DCs prevent activation of naïve CD8⁺ T cells and induce depletion of memory CD8⁺ T cells (54). A compelling explanation has been offered by Cobbold et al., suggesting that the capacity to deplete essential amino acids or ATP makes certain DCs tolerogenic. Treatment of DC with vitamin D3, IL-10, or TGF-β leads to upregulation of essential amino acid–depleting enzymes (55), and such enzymes are upregulated in tolerated skin grafts (38). In this study, we observed a trend toward higher expression of the amino acid–depleting enzyme arginase 1 in TGF-β–expressing grafts, but this was not consistent in all grafts assessed. This could be caused by technical limitations attributable to the presence of many cell populations other than DCs in the analyzed whole grafts. Studies of protein content in vitamin D3–induced tolerogenic DCs demonstrated changes in metabolic pathways involving lipids, glucose, and oxidative phosphorylation, which may affect the production of reactive oxygen species as well as response to stimulation by the modulated DCs (56).

In our study, the BMDCs we added to the islet grafts were not pulsed with any antigen. In several studies, antigen pulsing has been necessary for transferred tolerogenic DCs to have an effect (43,57). This may be because in our experiments the DCs are directly mixed with the islets, allowing them to take up antigen during the process rather than injected systemically. Additionally, in our experiments the BMDCs were not matured with LPS in the in vivo experiments, which allows greater antigen uptake capacity, so they can take up antigen during and after transplantation. Other studies using immature, non-LPS-pulsed, tolerogenic DCs to alleviate experimental collagen-induced arthritis found no need for antigen pulsing to achieve protection (50,58). A surprising finding was that addition of immature BMDCs could delay rejection of islet grafts even when not exposed to TGF-β. Our findings complement previous studies showing that immature autologous DCs prolong heart graft survival in Lewis rats (59), even when not further differentiated toward a tolerogenic phenotype, indicating that a state of immaturity can be enough to at least delay immunity. We identified the induction of tolerogenic DCs as a potential mechanism for the protection seen with expression of TGF-β, but, interestingly, TGF-β–exposed BMDCs were more efficient at inducing long-term protection of islet grafts than the direct exposure to TGF-β within the graft. Because the protection afforded by cotransplantation with BMDCs was long-lived, it could be interpreted as active immunological tolerance; however, the short-lived effects of direct expression of TGF-β in the graft show the hallmarks of reversible immunosuppression. We found no evidence of increased presence of Foxp3⁺ regulatory T cells in the grafts cotransplanted with TGF-β–exposed BMDCs (Supplementary Fig. 3) and no decrease in CD8⁺ T cells even in long-term accepted grafts at 125 days after transplantation, indicating that despite a heavy infiltrate of autoreactive T cells, these cells retain the pattern of respectful insulitis and do not damage the islets. This pattern is different from the reduced infiltration of T cells seen in TGF-β–expressing islet grafts (Fig. 1) and indicates that the mechanisms for suppression between direct expression of TGF-β and coadministration of TGF-β–exposed BMDCs are qualitatively different.

DC-based cell therapy has demonstrated considerable promise for the promotion of transplant tolerance in animal models, and now considerable efforts are being made to investigate whether it can be safely adapted for use in humans. Repetitive stimulation with immature DCs pulsed with an HLA-A2–derived allopeptide can expand alloantigen-specific Treg cells ex vivo (60), and subcutaneous injection of immature DCs pulsed with influenza-derived peptide induced a severely impaired cytolytic response (61). Human DCs, conditioned by exposure to vitamin D3 and dexamethasone, induce anergy in alloreactive CD4⁺ T cells (62). A clinical study evaluating safety for treatment of rheumatoid arthritis with autologous DCs generated with dexamethasone and vitamin D3 and loaded with synovial fluid (http://www.clinicaltrials.gov/ct2/show/NCT01352858) is ongoing in Newcastle, U.K. At the University of Pittsburgh (Pittsburgh, PA), the first trial for the treatment of type 1 diabetic patients with autologous DCs treated ex vivo with antisense phosphorothioate-modified oligonucleotides targeting the primary transcripts of the CD40, CD80, and CD86 has passed an initial phase I trial for safety (63). The results from these studies will be of great interest to the scientific community and will instruct further adaptation of tolerogenic DC–based therapies for use in human disease and transplantation tolerance.

In summary, our results demonstrate that short-term local exposure to TGF-β in vivo or exposure of DCs to TGF-β ex vivo can prolong the life of islet grafts. In the case of tolerogenic DC transfer, this protection appears to be long-lasting and sustained. The use of cell-mediated immunotherapy is a rapidly expanding field (64), and we propose that conditioned APC administered at the time of transplantation may become a useful and invaluable tool in promoting graft tolerance along with conventional immunosuppressive agents.

D.C.T. is a National Institute for Health Research Clinical Lecturer funded by Cambridge Institute for Medical Research/Wellcome Trust Next Generation Fellowship. The authors acknowledge funding from the Medical Research Council (to F.S.W.). M.W. is an RD Lawrence Fellow funded by Diabetes U.K. The Cambridge Institute for Medical Research is supported by a Wellcome Trust Strategic Award (079895).

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

D.C.T. researched data and wrote the manuscript. F.S.W. contributed the G9 mice, contributed to discussion of the manuscript, and reviewed and edited the manuscript. P.Z. researched and discussed data. E.A.G. contributed to discussion of the manuscript and reviewed and edited the manuscript. M.W. researched data and wrote the manuscript. M.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.

The authors thank Professor Anne Cooke and her laboratory (University of Cambridge) and Dr. Sarah Howlett (University of Cambridge) for generous sharing of reagents and constructive discussions. The authors also thank Christien Lien (University of Cambridge), Orval Lyle (University of Cambridge), Charlotte Ferguson (University of Cambridge), and Martin Rice (University of Cambridge) for excellent technical assistance.