Nuclear factor κB–inducing kinase activation as a mechanism of pancreatic β cell failure in obesity

Aberrant BIRC function has been associated with many human cancers (Fulda and Vucic, 2012), leading to the development of a group of peptide-derived drugs called second mitochondria-derived activator of caspase (SMAC) mimetics, which induce the degradation of BIRC2 and BIRC3, thereby interfering with their ability to inhibit apoptosis (Varfolomeev et al., 2007). Concomitantly, degradation of BIRC2 and BIRC3 elicits NIK accumulation, which leads to proteasome-dependent processing of p100 into p52 and hence activates the noncanonical NF-κB pathway (Varfolomeev et al., 2007).

To test whether systemic elevation of NIK mediated by the SMAC mimetic MV1 would alter glucose homeostasis in vivo, we used a zebrafish model. The main organs required for metabolic control, such as the pancreas, islet, and insulin-sensitive tissue (muscle, liver) are conserved in zebrafish, rendering it an excellent model to study glucose homeostasis (Hesselson et al., 2009; Jurczyk et al., 2011; Seth et al., 2013). First, we showed that synthesized MV1 could induce NIK accumulation and processing of p100 into p52 in MDA-MB-231/Luc breast cancer cells, indicating activation of noncanonical NF-κB signaling (Fig. 1 A), as described previously (Varfolomeev et al., 2007). We next injected MV1 into zebrafish larvae at 6 d postfertilization (dpf) and examined glucose levels after 6 h. WT larvae treated with MV1 exhibited increased glucose levels compared with vehicle (DMSO) controls (Fig. 1 B). This indicates that MV1-induced NIK activation disrupts glucose homeostasis in vivo in WT zebrafish.

To elucidate whether the defect in glucose homeostasis relates to MV1-induced, NIK-mediated β cell dysfunction, we examined the effect of MV1 on insulin secretion of primary C57BL/6 WT islets. Addition of MV1 to WT islets triggered p100 to p52 processing, indicating NIK activation (Fig. 1 C). MV1 had no effect on islet survival in 24-h in vitro cultures as assessed by calcein-AM uptake and ethidium homodimer DNA intercalation (unpublished data). In a glucose-stimulated insulin secretion (GSIS) assay, MV1-treated islets exhibited normal basal insulin output, but insulin secretion was reduced by ∼50% compared with controls when challenged with 20 mM glucose. Potassium chloride (KCl)–stimulated secretion was normal, indicating normal insulin content (Fig. 1 D). To assess the effect of MV1-treatment in vivo, we transplanted either MV1- or vehicle-treated islets into diabetic syngeneic recipients. A glucose tolerance test (GTT) at post-operative day (POD) 3 revealed that, compared with controls, MV1-treated grafts exhibited impaired glucose tolerance (Fig. 1 E). Glucose tolerance was still slightly impaired at POD 7, but had completely resolved by POD 11 such that MV1-treated islets provided equivalent metabolic control to vehicle treated islets (Fig. 1, F and G). Further, MV1-treated grafts exhibited normal morphology and insulin staining, as well as equivalent graft area determined by morphometry (Fig. 1 H). This indicates that acute activation of NIK results in a temporary β cell defect that is reversed upon drug-clearance and suppression of NIK.

We next assessed whether β cell–intrinsic activation of NIK would impair the function of human islets. Human islets treated with MV1 showed normal basal (2.8 mM) but defective glucose-stimulated (25 mM) insulin secretion (Fig. 1 I). Thus, activation of NIK in human islets resulted in impaired β cell function. These data highlight the importance of the NIK pathway and its contribution to β cell dysfunction in human islet physiology.

To elucidate whether NIK plays a role in islets in a model of glucose intolerance, a diet-induced obesity (DIO) model, C57BL/6 mice were placed on a 45% kcal high-fat diet (HFD). After 8 wk of HFD, these mice showed impaired glucose tolerance and insulin secretion (Fig. 2, A and B), increased β cell mass (Fig. 2 C), and reduced insulin output capacity in fasted conditions and in response to an in vivo glucose challenge (Fig. 2 D). These results are supported by previous studies (Winzell and Ahrén, 2004; Peyot et al., 2010). Islets from DIO mice exhibited activation of the noncanonical NF-κB pathway (Fig. 2 E), as indicated by the accumulation of NIK and increased phosphorylation of the NIK downstream target IκB kinase α (IKKα). Activated phospho-IKKα phosphorylates NF-κB p100, leading to its proteasome-dependent cleavage releasing active p52. Islets from DIO mice exhibited a higher processing rate of NF-κB p100 to p52 and increased levels of the p52 binding partner RelB, compared with islets from chow-fed controls. As the antibody for Phospho-IKK not only identifies phospho-IKKα but also phospho-IKKβ, we probed for IKKβ downstream target IκBα, which is the canonical NF-κB inhibitor; IκBα levels were normal, indicating absence of canonical NF-κB activation in DIO islets.

Both TNF and receptor activator of NF-κB ligand (RANKL) activate the noncanonical NF-κB pathway in various cell types, including immune cells (Rauert et al., 2010) and osteoclasts (Novack et al., 2003), and high circulating serum levels of TNF and RANKL are associated with diabetes (Hotamisligil et al., 1995; Kiechl et al., 2013). The effect of TNF and RANKL on noncanonical NF-κB signaling in islets is unknown. TNF induced p100 to p52 processing and RelB accumulation in primary mouse islets, indicating NIK activation (Fig. 2 F). TNF also induced canonical NF-κB activation as shown by degradation of its inhibitor IκBα. RANKL induced p100 to 52 processing and RelB accumulation in primary islets, indicating NIK activation, but did not activate canonical NF-κB signaling (Fig. 2 G). Therefore, both TNF and RANKL activate NIK signaling in primary mouse islets. Additionally, islets isolated from HFD-fed mice show cell-intrinsic activation of NIK and activation of the noncanonical NF-κB pathway with reduced insulin secretion.

The aforementioned data present a causal link between β cell dysfunction and NIK activation in DIO. To test this concept further, we used a genetic approach whereby β cells would express high levels of active NIK in a cell-intrinsic manner. NIK levels, and hence its activation, are controlled by an E3 ligase complex; TRAF2 recruits the E3 ligases BIRC2/3 while interactions with TRAF3 allow delivery of NIK to the complex (Vallabhapurapu et al., 2008; Zarnegar et al., 2008). In absence of a ligand, NIK is ubiquitinated by this complex and targeted for proteasome-dependent degradation thereby silencing the noncanonical NF-κB pathway (Vallabhapurapu et al., 2008; Zarnegar et al., 2008). Deletion of any one component, i.e., TRAF2, TRAF3, or the BIRC proteins, results in cell-intrinsic hyperactivation of noncanonical NF-κB pathway at a physiological level (Gardam et al., 2008; Zarnegar et al., 2008).

To examine how β cell–intrinsic NIK activation would affect glucose homeostasis, we generated β cell–specific TRAF2 knockout (βTRAF2) mice. TRAF2_loxP/loxP were crossed with mice expressing Cre under the control of the rat insulin promoter (RIP Cre). TRAF2 mRNA and protein were expressed in TRAF2_loxP/loxP islets but depleted in islets from βTRAF2 mice (Fig. 3, A and B). βTRAF2 islets showed high cell-intrinsic accumulation of NIK with hyper-phosphorylation of IKKα (Fig. 3 C). βTRAF2 islets exhibited increased p100 processing into p52 and higher levels of RelB. In contrast, IκBα levels were not changed in βTRAF2 islets, indicating that IKKβ and canonical NF-κB signaling were not activated. Thus, loss of TRAF2 results in cell-intrinsic constitutive activation of NIK in pancreatic β cells.

Chow-fed βTRAF2 mice exhibited normal weight gain over the experimental period of 8 wk and normal glucose tolerance at 8 wk of age (not depicted). They also had mildly elevated nonfasted blood glucose levels (Fig. 3 D) compared with floxed littermates and RIP Cre/+ controls. By 16 wk of age, chow-fed βTRAF2 mice showed very mild glucose intolerance (Fig. 3 E) and no significant difference in insulin secretion (Fig. 3 F) compared with WT mice. However, in a DIO model, βTRAF2 mice showed greatly exacerbated glucose intolerance compared with WT mice (Fig. 3 G). We next examined in vivo insulin secretion (i.v. GTT) to test whether this related to a β cell defect. In this experiment, DIO βTRAF2 mice exhibited a reduction in first-phase insulin secretion (Fig. 3 H). These data demonstrate that β cell–intrinsic activation of NIK results in impaired glucose tolerance due to reduced β cell function.

To determine whether the reduction in insulin secretion was due to decline in insulin production, we determined total pancreatic insulin content and β cell area. On a chow diet, pancreata from βTRAF2 mice showed similar total insulin content as WT mice and normal β cell area (Fig. 3, I and J). However, in the DIO model, βTRAF2 mice exhibited increased pancreatic insulin content compared with WT mice (Fig. 3 I). This is accompanied by an increase in β cell area in DIO βTRAF2 mice (Fig. 3 J). Also, β cells in βTRAF2 islets exhibited robust insulin expression as determined by insulin immunostaining (Fig. 3 K) and normal levels of Insulin-1 and Insulin-2 determined by RT-PCR (Fig. 3 L). Thus, we can conclude that β cell–intrinsic NIK activation results in a loss of insulin secretory capacity rather than a defect in insulin production. Additionally, we hypothesize that the increased β cell mass in DIO βTRAF2 mice most likely represents a physiological compensatory response to the reduced insulin secretory capacity of the βTRAF2 islet.

In the DIO model, βTRAF2 mice presented a blunted first-phase insulin secretion compared with controls. A subset of transgenic RIP Cre mouse lines have been reported to exhibit hypothalamic Cre expression (Wicksteed et al., 2010). To discern between hypothalamus-mediated and islet-specific defects, we isolated pancreatic βTRAF2 and control islets, and transplanted 150 islet equivalents into streptozotocin-treated diabetic syngeneic C57BL/6 recipients. Under these conditions both WT and βTRAF2 islet grafts restored normal metabolic control (Fig. 4 A). To test β cell function, we performed an i.p. GTT at POD 11. Compared with mice receiving WT islet grafts, βTRAF2 graft recipients exhibited impaired glucose tolerance (Fig. 4 B). Next, we determined insulin secretion in graft recipients via an i.v. GTT at POD 21. Recipients of βTRAF2 islet grafts exhibited a blunted first-phase insulin secretion compared with mice receiving WT islet grafts (Fig. 4 C). Graft pathology revealed that βTRAF2 grafts exhibited normal morphology and insulin staining, as well as equivalent graft area determined by morphometry (Fig. 4 D). Therefore, we conclude that loss of glucose tolerance was not due to β cell loss.

In a second approach to establish the β cell–intrinsic defect, we challenged βTRAF2 islets in a GSIS assay. βTRAF2 islets exhibited normal basal insulin output at a glucose concentration of 2 mM (Fig. 4 E). However, when challenged with 20 mM glucose, βTRAF2 islets exhibited a 50% decrease in GSIS compared with WT islets. KCl, which induces membrane depolarization to trigger insulin secretion independent of glucose, provoked a normal insulin secretory response in βTRAF2 islets, indicating that insulin content and secretory machinery were intact in these NIK-activated islets. Collectively, the in vivo transplant and GSIS data demonstrate that the metabolic defect observed for βTRAF2 mice is β cell intrinsic.

The β cell secretory defect in βTRAF2 mice suggests that β cell–intrinsic NIK activation results in impaired β cell function. A potential caveat to this interpretation is that TRAF2 is also necessary for molecular control over the canonical NF-κB and JNK/activator protein 1 (AP-1) signaling pathways (Yeh et al., 1997). While TRAF2 and TRAF3 are both required to suppress NIK activity, TRAF3 differs from TRAF2 as it is not involved in NF-κB and JNK/AP-1 signaling downstream of TNFR family members (Vallabhapurapu et al., 2008). Therefore, we examined NIK activation and β cell function in β cell–specific TRAF3 knockout (βTRAF3) mice. βTRAF3 islets showed reduced mRNA and protein levels of TRAF3 compared with WT islets (Fig. 5, A and B). TRAF3-deficient islets showed constitutive, cell-intrinsic hyperaccumulation of NIK, higher levels of phospho-IKKα, an increased p52 to p100 ratio and increased RelB levels (Fig. 5 C). Canonical NF-κB and JNK signaling in response to TNF, however, was normal in pancreatic islets lacking TRAF3 (unpublished data), as has been described in other cell types (Gardam et al., 2011).

Chow-fed βTRAF3 mice exhibited normal weight increase, normal glucose tolerance at 8 wk of age (not depicted), and normal nonfasted blood glucose levels (Fig. 5 D), but they displayed mildly decreased glucose tolerance and insulin secretion at 16 wk of age compared with WT mice (Fig. 5, E and F). In a DIO model, βTRAF3 mice presented with abnormally high nonfasted blood glucose levels (Fig. 5 D), impaired glucose tolerance (Fig. 5 G), and a reduction in GSIS in vivo compared with WT mice (Fig. 5 H). Analysis of total pancreatic insulin content and β cell mass revealed that DIO βTRAF3 mice exhibited increased pancreatic insulin content compared with WT mice due to an increase in their β cell area (Fig. 5, I and J). Further to this, βTRAF3 islets showed reduced GSIS ex vivo but a normal insulin secretory response to KCl (Fig. 5 K). Thus, βTRAF3 mice phenocopy βTRAF2 mice with β cell–intrinsic NIK hyperactivation and impaired β cell function.

The key role of TRAF2 and TRAF3 in regulation of NIK is the recruitment and directed control of the E3 ligase activity of BIRC proteins toward NIK. As MV1-induced NIK activation occurs through BIRC2 and BIRC3 degradation, we hypothesized genetic ablation of BIRC2 and BIRC3 in β cells would trigger a β cell defect. Mice that lack both BIRC2 and BIRC3 in their β cells (βBIRC2/3) exhibit islet-intrinsic NIK hyperactivation, as indicated by increased p100 processing (Fig. 6 A) with impaired glucose tolerance and reduced insulin secretion in a DIO model (Fig. 6, B and C). These data demonstrate that all components of the TRAF2–TRAF3–BIRC2–BIRC3 E3 ligase complex are required to provide the β cell protective circuit that reigns in the deleterious activation of NIK, which suggests that firm regulation of NIK in β cells is essential to control glucose sensing and insulin secretion. Further, these data demonstrate that NIK activation precipitates β cell failure in a DIO model.

To determine the molecular basis for islet dysfunction, we determined genes that were differentially expressed in both βTRAF2 and βTRAF3 islets compared with WT islets using microarray analysis. We identified 26 differentially expressed genes common to the chow-fed βTRAF2 and βTRAF3 islets and 35 differentially expressed genes common to the DIO βTRAF2 and βTRAF3 islets compared with respective WT controls (Fig. 7 A). The 15 most differentially expressed genes within chow and DIO βTRAF2 and βTRAF3 islets comparing logarithmic fold change (logFC) and their function are listed in Table 1 and Table S1. Additionally, nine differentially expressed genes overlapped between the chow and DIO βTRAF2 and βTRAF3 islets (Table 1, top six most differentially expressed genes highlighted in bold). The majority of the differentially expressed genes were up-regulated in βTRAF2 and βTRAF3 islets. This indicates that NIK and the noncanonical NF-κB pathway activate rather than repress a select transcriptional program in pancreatic β cells. The up-regulation of Tph1, Lrrc55, Hcn1, Tnfrsf11b, Rasgrp1, Grem2, and Pde7b in βTRAF2 and βTRAF3 islets was validated by RT-PCR in a series of independent experiments (Fig. 7 B). Gene ontology analysis highlighted regulators of growth and proliferation, metabolic genes, potassium channel components, and vesicle-associated genes, many of which are associated with diabetes in animal models (Table 2). Significantly, many of the differentially expressed genes have been identified in human studies such as GWAS, linking genes such as Bcat1 (Grimm et al., 2003; Rampersaud et al., 2007; Keller et al., 2008; Chen et al., 2013), Gbp4 (Westra et al., 2013), Igfbp5 (Jehle et al., 1998; Ambra et al., 2014), Kcnj5 (Li et al., 2012; Cnop et al., 2014; Dayeh et al., 2014), Matn2 (Ambra et al., 2014; Cnop et al., 2014), Rerg, Rph3a, Slca13, and Pde7b (Dayeh et al., 2014), Sftpd (Ortega et al., 2013), Rasgrp1 (Qu et al., 2009; Taneera et al., 2012; Li et al., 2013), or Tnfrsf11b (Olesen et al., 2005; Bradfield et al., 2011; Cnop et al., 2014) to type 1 and type 2 diabetes. These data identify NIK as a critical new β cell–intrinsic signaling node in diabetes integrating a transcriptional network that controls glucose homeostasis.

We have identified NIK as a novel mechanism of β cell dysfunction in obesity. Acute NIK activation using a pharmacological approach disrupts glucose homeostasis in a zebrafish model, and induces β cell dysfunction in mouse islets, but also in human islets, highlighting the importance of the NIK pathway in human islet physiology as a negative regulator of β cell function. As we have demonstrated, pancreatic β cells in a model of obesity and glucose intolerance present with β cell–intrinsic NIK activation and a decrease in β cell insulin secretory capacity. We could demonstrate that NIK was induced by inflammatory cytokines TNF and RANKL in pancreatic mouse islets. Both cytokines constitute good candidates as NIK activators in a DIO mouse model and in human obesity. Both RANKL and TNF levels are elevated in mouse models of obesity and in T2D subjects (Hotamisligil et al., 1995; Kiechl et al., 2013). Blockade of TNF or RANKL improved glucose tolerance in mouse models of obesity (Uysal et al., 1997; Kiechl et al., 2013). Further, TNF inhibition has been shown to alleviate the metabolic syndrome in several human studies (Bernstein et al., 2006; Marra et al., 2007; Campanati et al., 2013).

Our data furthers studies showing that diet-induced obesity induces NIK in peripheral insulin-dependent tissues such as liver (Sheng et al., 2012) and muscle (Choudhary et al., 2011). Further evidence for a crucial involvement of NIK in whole body glucose homeostasis comes from studies of NIK knockout mice, which are hypoglycemic and exhibit improved glucose tolerance on a chow diet (Sheng et al., 2012). Accumulation of NIK is associated with decreased glucose uptake in muscle (Choudhary et al., 2011) and increased hepatic glucose output by the liver (Sheng et al., 2012). Thus, NIK is emerging as a global molecular factor in obesity, therefore, these data highlight the possibility of systemic NIK inhibition as a potential therapeutic target to reduce hepatic glucose production and improve muscle uptake of glucose, while simultaneously preserving β cell function.

As we showed for pancreatic islets in diet-induced obesity, NIK levels are increased in the livers of genetically (db/db) and diet-induced obese mice (Sheng et al., 2012). Hepatocyte-specific NIK overexpression increased hepatic glucose output, presumably by phosphorylation and concomitant elevation and activation of the transcription factor cAMP response element-binding (CREB) levels, which stimulates gluconeogenesis. This process is controlled by cAMP-dependent PKA phosphorylating and activating CREB. We identified Pde7b as a highly expressed gene in islets with constitutively activated NIK (βTRAF2 and βTRAF3 islets). Pde7b, which encodes for phosphodiesterase 7b (PDE7B), is involved in the conversion of cAMP to AMP, thereby decreasing cAMP secondary messenger activity (Hetman et al., 2000). cAMP regulates glucose sensing and insulin secretion in pancreatic β cells, through both PKA-dependent and-independent mechanisms (Ämmalä et al., 1994; Leclerc and Rutter, 2004). Significantly, Pde7b and Traf2 were both recently shown to be differentially methylated in a genome-wide analysis of human T2D pancreatic islets (Dayeh et al., 2014). Pde7b was found to be differentially expressed in human T2D pancreatic islets as assessed by RT-PCR, and its overexpression caused a reduction in glucose-stimulated insulin secretion in vitro (Dayeh et al., 2014). This suggests that overactivation of NIK may provide a direct link between regulation of cAMP levels and CREB activity, both in liver and in islets.

Inflammation has emerged as a key factor in the pathogenesis of the metabolic syndrome acting at the level of muscle and liver to induce insulin resistance, at adipocytes to drive elaboration of cytokines, and at the level of the β cell. Key signaling pathways include the JNK and canonical NF-κB pathways. Here, we demonstrate NIK levels are elevated in pancreatic islets isolated from diet-induced obese mice, which exhibited decreased insulin secretory capacity. Further, several genetic mouse models of constitutive β cell–intrinsic NIK activation present impaired glucose-stimulated insulin secretion in diet-induced obesity. Thus, NIK acts as a critical signaling node that regulates a transcriptional network to control glucose homeostasis. NIK activation in obesity is a feature of pancreatic β cells, liver and muscle providing a tissue-spanning mechanistic understanding of the global effects of obesity in the metabolic syndrome.

C57BL/6 mice were sourced from Australian BioResource (Mossvale, NSW, Australia). TRAF2_loxP/loxP, TRAF3_loxP/loxP (C57BL/6 background; gift from R. Brink, Garvan Institute, NSW, Australia) and Birc3^−/− mice containing Birc2_loxP/loxP (C57BL/6 background; gift from D. Vaux, Walter and Eliza Hall Institute, VIC, Australia) were crossed with RIP-Cre mice (Cre driven by the rat insulin 2 promoter; Tg[Ins2-cre]25Mgn/J, C57BL/6 background; The Jackson Laboratory) to generate β cell–specific knockout mice (βTRAF2, βTRAF3, and βBIRC2/3).

WT zebrafish larvae were injected pericardially at 6 d after fertilization with 2 nl of 10 mM MV1 in 10% DMSO or 10% DMSO as described (Hesselson et al., 2009). Free glucose levels were analyzed 6 h after injection using a fluorescent glucose assay kit (Biovision; Jurczyk et al., 2011).

For islet transplantation, 150 hand-counted primary islet equivalents isolated from male and female donor mice were transplanted into syngeneic diabetic male C57BL/6 recipients as previously described (Grey et al., 2003). Mice in the DIO model were fed a diet providing 45% calories from fat (lard/safflower oil) ad libitum. Glucose tolerance tests (i.p. 2 g/kg and i.v. 1 g/kg body weight) after a 16-h fast were performed on female βTRAF2, βTRAF3, and βBIRC2/3 mice and respective floxed and RIP Cre controls mice as previously described (Cantley et al., 2013). Blood samples for insulin ELISA (mouse ultra-sensitive kit; Crystal Chem) were collected using 5-µl micro-capillary tubes (Drummond microcaps; Sigma-Aldrich). Insulin secretory capacity was determined by normalizing insulin levels or insulin secretion measured by AUC to total β cell area. Islets were isolated as previously described (Zammit et al., 2013), from male and female donors for GSIS, RNA extraction, and Western blot protein extraction. In vivo mouse physiology experiments comprise data from at least three independent cohorts.

Isolated mRNA (QIAschredder, RNEasy Plus Mini kit; QIAGEN) was reverse transcribed (Quantitect Reverse Transcription kit; QIAGEN). mRNA expression was determined by Real Time-PCRs using FastStart SYBR Green Master Mix (Roche). PCR reactions were performed with the following primers and normalized to the house-keeping gene Cyclophillin A (CpH, sense 5′-TGGACCAAACACAAACGGTTCC-3′ and antisense 5′-ACATTGCGAGCAGATGGGGTAG-3′; TRAF2, sense 5′-TCTGCCTGACCAGCATCCTC-3′ and antisense 5′-AATGCCTTCTTCATACAGGCCTTC-3′; TRAF3, sense 5′-CAGCGTGCCAAGAAAGCATC-3′ and antisense 5′-CGCACAACCTCTGCCTTCAT-3′). Initial denaturation was performed at 95°C for 10 s, followed by a three step cycle consisting of 95°C for 15 s (4.8°C/s, denaturation), 63°C for 30 s (2.5°C/s, annealing), and 72°C for 30 s (4.8°C/s, elongation). A melt-curve was performed after finalization of 45 cycles at 95°C for 2 min, 40°C for 3 min, and gradual increase to 95°C with 25 acquisitions/°C.

Protein was extracted from islets or cell lines using a detergent-based lysis buffer. 30–40 µg of protein per well were separated on an SDS-PAGE and protein levels were determined by Western Blot analysis using the following antibodies (expected protein size [kD] is indicated): TRAF2 (C-20; 50 kD), TRAF3 (H-122; 65 kD; Santa Cruz Biotechnology Inc.), IκBα (39 kD), NIK (125 kD), phospho-IKKα/β (85/87 kD), p100/p52 (120 & 52kD), RelB (70 kD; Cell Signaling Technology), β-actin (42 kD; Sigma-Aldrich), anti–rabbit IgG (horseradish peroxidase-conjugated; Thermo Fisher Scientific), and anti–mouse IgG (and horseradish peroxidase-conjugated; Thermo Fisher Scientific). Signals were visualized using an enhanced chemiluminescence detection kit (Amersham Pharmacia Biotech). Western blot signals were quantified using the ImageJ (NIH) histogram function. Subsequently, the band intensity was normalized to relative protein concentration using β actin as a reference.

Tissues were fixed in 10% buffered formalin and 5-µm paraffin sections were stained with hematoxylin-eosin. Serial consecutive sections in 100-µm intervals were stained for insulin (polyclonal anti–rabbit; Cell Signaling Technology) and counter-stained with hematoxylin. β cell area was assessed by determining total insulin-positive area (ImageJ).

Three human pancreatic islet isolates were prepared at the St. Vincent’s Hospital Centre for Islet Isolation (Melbourne, Australia) from heart-beating deceased donors. GSIS was determined for islets ex vivo as described (Cantley et al., 2013).

To determine whole pancreas insulin content, pancreata were collected and immediately immersed in ice-cold acidified ethanol. The tissue was homogenized with a tissue homogenizer (Heidolph Silent Crusher M) and total pancreatic insulin content was determined by insulin ELISA (mouse ultra-sensitive kit; Crystal Chem).

The synthetic routes of MV1 production are available upon request.

The quality of mRNA isolated from primary islets was evaluated by using an Agilent 2100 Bioanalyzer (Agilent Technologies). Samples were run in triplicates on a MoGene2.0 (GeneChip Mouse Gene 2.0 ST Array, Affymetrix) at the Ramaciotti Center (University of New South Wales, Australia). Raw data were deposited in the Gene Expression Omnibus database: accession no. GSE68317.

Statistical analysis was performed using an unpaired two-tailed Students t test using Prism software (GraphPad Inc.). Error bars in all figures represent SEM as specified.

Animal studies were approved by the Garvan/St Vincent’s Animal Ethics Committee (AEC) fulfilling all the requirements of the NH&MRC and the NSW State Government. All procedures performed complied with the Australian Code of Practice for Care and Use of Animals for Scientific Purposes. This project was reviewed and approved by the St Vincent’s Hospital (Melbourne, Australia) Human Research Ethics Committee: HREC-A 011/04.

Table S1 is a complete list of differentially expressed genes overlapping βTRAF2 and βTRAF3 islets (chow and DIO).

We thank Professor Tom Kay and Dr. Helen Thomas (St Vincent’s Hospital, Melbourne, Australia) for human pancreatic islets. We thank Michael Pickering (Biological Testing Facility, Garvan Institute of Medical Research, Sydney, Australia) for providing valuable technical support and animal husbandry. The authors also thank Rebecca Stokes and Tess Whitworth (Garvan Institute of Medical Research, Darlinghurst, Australia) for technical support for β cell functional assays.

E.K. Malle was supported by an Australian Postgraduate Award (UIPA). S.T. Grey was supported by a National Health and Medical Research Council (NHMRC) Senior Research Fellowship. This work was supported in part by an Australian Research Council Grant and a NHMRC Special Program Grant in Type 1 Diabetes 427695 to S.T. Grey.

The authors declare no competing financial interests.

Author contributions: E.K. Malle and S.T. Grey designed and led the study. E.K. Malle, N.W. Zammit, S.N. Walters, and J. Cantley conducted mouse experiments and analyzed data. J. Wu wrote code and analyzed microarray data. T. Loudovaris conducted experiments with human islets. R. Brink provided critical resources and expertise. S.R. McAlpine and Y.C. Koay designed and synthesized MV1. D. Hesselson designed and conducted zebrafish experiments. E.K. Malle, S.R. McAlpine, D. Hesselson, J. Cantley, and S.T. Grey critically assessed data and provided intellectual input. E.K. Malle and S.T. Grey co-wrote the manuscript.