A Combined “Omics” Approach Identifies N-Myc Interactor as a Novel Cytokine-induced Regulator of IRE1α Protein and c-Jun N-terminal Kinase in Pancreatic Beta Cells*

Background: Cytokine-induced IRE1α activation regulates adaptive/pro-apoptotic endoplasmic reticulum (ER) stress transition in type 1 diabetes (T1D). Results: The IRE1α-binding protein NMI modulates JNK activation and apoptosis in beta cells. Conclusion: NMI induction is a novel negative feedback mechanism regulating ER stress and apoptosis in cytokine-exposed beta cells. Significance: Clarifying cytokine modulation of ER stress may provide information to protect beta cells in early T1D. Type 1 diabetes is an autoimmune disease with a strong inflammatory component. The cytokines interleukin-1β and interferon-γ contribute to beta cell apoptosis in type 1 diabetes. These cytokines induce endoplasmic reticulum stress and the unfolded protein response (UPR), contributing to the loss of beta cells. IRE1α, one of the UPR mediators, triggers insulin degradation and inflammation in beta cells and is critical for the transition from “physiological” to “pathological” UPR. The mechanisms regulating inositol-requiring protein 1α (IRE1α) activation and its signaling for beta cell “adaptation,” “stress response,” or “apoptosis” remain to be clarified. To address these questions, we combined mammalian protein-protein interaction trap-based IRE1α interactome and functional genomic analysis of human and rodent beta cells exposed to pro-inflammatory cytokines to identify novel cytokine-induced regulators of IRE1α. Based on this approach, we identified N-Myc interactor (NMI) as an IRE1α-interacting/modulator protein in rodent and human pancreatic beta cells. An increased expression of NMI was detected in islets from nonobese diabetic mice with insulitis and in rodent or human beta cells exposed in vitro to the pro-inflammatory cytokines interleukin-1β and interferon-γ. Detailed mechanistic studies demonstrated that NMI negatively modulates IRE1α-dependent activation of JNK and apoptosis in rodent and human pancreatic beta cells. In conclusion, by using a combined omics approach, we identified NMI induction as a novel negative feedback mechanism that decreases IRE1α-dependent activation of JNK and apoptosis in cytokine-exposed beta cells.

Type 1 diabetes (T1D) 6 is a chronic autoimmune disease characterized by pancreatic islet inflammation (insulitis) and progressive beta cell loss (1,2). During this inflammatory process, there is an ongoing "dialogue" between invading immune cells and the target beta cells. This dialogue is mediated by both cytokines/chemokines produced by immune cells and beta cells and by immunogenic signals delivered by dying or modified beta cells (1), and it is modulated by the genetic background of the affected individuals (3).
Recent findings indicate that a mild ER stress amplifies IL-1␤induced nuclear factor B (NF-B) signaling, with consequent increase in local inflammation (9) and beta cell apoptosis (10). Importantly, in vivo modulation of ER stress by use of chemical chaperones prevents autoimmune diabetes in two mouse models of the disease (8).
In rat beta cells, cytokine-dependent activation of the UPR occurs via NO-dependent inhibition of the sarcoendoplasmic reticulum pump Ca 2ϩ -ATPase 2b (SERCA-2b), ER calcium decrease, and unfolded protein accumulation (4). These events, however, seem to differ among species (11), and other unknown mechanisms are implicated in cytokine-induced UPR activation in human pancreatic beta cells.
The three main sensors of the UPR are the transmembrane proteins inositol-requiring protein 1␣ (IRE1␣), protein kinase RNA-like endoplasmic reticulum kinase (PERK), and activating transcription factor 6 (ATF6) (12). These proteins detect the accumulation of unfolded proteins in the ER lumen and activate mechanisms to restore ER function (12)(13)(14). In case of chronic and/or severe ER stress, persistent activation of the UPR triggers apoptosis (15,16), contributing to the loss of beta cells in type 1 (6,7,8,17) and type 2 diabetes (18). What determines the transition from "physiological" to "pathological" UPR remains to be clarified (13), but accumulating evidence indicates that the amplitude and duration of IRE1␣ signaling is critical for this transition (19).
Once IRE1␣ is activated, its cytoplasmic domain is autophosphorylated and gains endoribonuclease activity, cleaving 26 nucleotides from the mRNA encoding X-box binding protein 1 (Xbp1). This generates a spliced variant (Xbp1s) that functions as a potent transcription factor, regulating genes involved in protein folding, maturation, quality control, and protein degradation via the ER-associated degradation pathway (12,18). IRE1␣ degrades ER-targeted mRNAs, such as insulin mRNA in pancreatic beta cells (20,21), thus decreasing the production of new proteins (20 -22). IRE1␣ also activates c-Jun N-terminal kinase (JNK) through the binding of the adaptor protein TNF receptor-associated factor 2 (TRAF2) (23). IRE1␣ activation is regulated and fine-tuned by both signals derived from the ER lumen (12) and by the formation of protein complexes between its cytosolic domain and regulatory proteins (13,19,24). Several cytosolic IRE1␣-regulating proteins have been described (24,25), but little is known regarding IRE1␣ regulation in pancreatic beta cells (13,25,26).
Against this background, the aim of this study was to identify novel cytokine-induced IRE1␣-interacting proteins that modulate UPR activation in pancreatic beta cells. For this goal, we used a high throughput mammalian two-hybrid technology, MAPPIT (mammalian protein-protein interaction trap) that is based on complementation of a modified cytokine receptor complex. This allows screening of a bait of interest against an arrayed prey collection derived from the CCSB Human ORFeome collection (27). The MAPPIT-based IRE1␣ interactome analysis identified N-Myc interactor (NMI) as an IRE1␣ interacting/modulator protein. This novel role for NMI was confirmed in rodent and human pancreatic beta cells, and an increased expression of NMI was detected in islets from NOD mice with insulitis and in rodent and human beta cells exposed in vitro to pro-inflammatory cytokines. Of particular relevance, cytokine-induced NMI modulates IRE1␣-dependent activation of JNK and apoptosis in pancreatic beta cells.

Culture of Human Islet Cells, FACS-purified Rat Beta Cells, INS-1E Cells, the Human Beta Cell Line EndoC-␤H1, and
HEK293T Cells-Human islets from 17 nondiabetic donors were isolated in Pisa using collagenase digestion and density gradient purification (28). The donors (7 women and 10 men) were 67.8 Ϯ 3.1 years old and had a body mass index of 26 Ϯ 1.2 (kg/m 2 ) ( Table 1). Beta cell purity, as evaluated by immunofluorescence for insulin, using a specific anti-insulin antibody ( Table 2), was 58 Ϯ 3.6%. The islets were cultured in M199 culture medium containing 5.5 mM glucose and sent to Brussels, Belgium, within 1-5 days after isolation, where they were dispersed and cultured in Ham's F-10 medium containing 6.1 mM glucose (Invitrogen) as described (29,30).
The human beta cell line EndoC-␤H1, kindly provided by Dr. R. Scharfmann, University of Paris, France (33), was cultured in DMEM containing 5.6 mM glucose, 2% BSA fraction V, 50 M 2-mercaptoethanol (Sigma), 10 mM nicotinamide (Calbiochem), 5.5 g/ml transferrin, 6.7 ng/ml selenite (Sigma), 100 units/ml penicillin, and 100 g/ml streptomycin (Lonza, Leus- den, The Netherlands). The same medium, but with 2% FBS, was used during the cytokine treatment as described (34). The human embryonic kidney cells HEK293T were cultured in DMEM containing 25 mM glucose, 5% FBS, 100 units/ml penicillin, 100 g/ml streptomycin, and sodium pyruvate 100ϫ (Invitrogen). Cell Treatment and Nitric Oxide (NO) Measurement-INS-1E cells were exposed to 12.5 M cyclopiazonic acid (CPA) (Sigma), a concentration previously shown by us to induce ER stress in these cells (10). CPA was dissolved in dimethyl sulfoxide (DMSO), which was used as control condition at the concentration of 0.03%. The following cytokine concentrations were used, based on previous dose-response experiments performed by our group (29,(35)(36)(37) as follows: recombinant human IL-1␤ (R&D Systems, Abingdon, UK) 10 units/ml for INS-1E cells or 50 units/ml for human islet cells, primary rat beta cells, dispersed rat islet cells, and the human beta cell line EndoC-␤H; recombinant rat IFN-␥ (R&D Systems) 100 units/ml for INS-1E cells or 500 units/ml for dispersed rat islet cells and primary rat beta cells; human IFN-␥ (PeproTech, London, UK) 1000 units/ml for human islet cells or the human beta cell line EndoC-␤H; recombinant murine TNF-␣ (Innogenetics, Gent, Belgium) 1000 units/ml for all cell types. Lower cytokine concentration and shorter time points were used in the rodent experiments because rat beta cells are more sensitive to cytokine-induced damage than human islets (38,39). Culture medium was collected for nitrite determination (nitrite is a stable product of NO oxidation) by the Griess method (40).
Array MAPPIT and Binary MAPPIT-To identify IRE1␣interacting proteins, an array MAPPIT screen was performed as described (27), using as MAPPIT bait the cytoplasmic portion of IRE1␣ (amino acids 571-977). Briefly, HEK293T cells expressing the IRE1␣ bait protein were reverse-transfected in 384-well plate format with a collection of ϳ10,000 full ORFs from the Version 5.1 ORFeome collection that were Gatewaycloned as MAPPIT preys. Two independent screens were performed, and after elimination of nonspecific binders (i.e. binders that appear as "positive" also in irrelevant screens), 31 putative IRE1␣-interacting proteins were selected.
In the binary MAPPIT analyses, an IRE1␣ bait containing the cytoplasmic portion of IRE1␣ was generated based on the original pSEL MAPPIT bait construct (41). This bait contains the extracellular part of the human erythropoietin receptor and the transmembrane and intracellular parts of the mouse leptin receptor. MAPPIT preys were cloned in the previously described pMG1 vector (42). Briefly, HEK293T cells were transfected with bait, prey, and STAT3 reporter plasmids (Table 3) by the use of a standard calcium phosphate transfection method, and luciferase activity was measured 48 h after transfection using the Luciferase assay system kit (Promega, Leiden, The Netherlands) on a TopCount luminometer. Cells were stimulated with erythropoietin (5 ng/ml) 24 h after transfection. Luciferase data were represented as average relative light units Ϯ S.E. of three technical transfection replicates. To confirm equal expression of empty and IRE1␣-cyto baits, the MAPPIT signal with the REM2 prey protein (which interacts with the leptin receptor portion of the bait) was measured. Additional selection of IRE1␣-interacting proteins for subsequent study in pancreatic beta cells was done by systematic comparison against previous RNA-seq and array analysis of cytokine-treated human and rodent beta cells, as described under "Results" and Fig. 1.
RNA Interference and Transfection with Plasmids-The siRNAs used in the study are described in Table 4. The optimal concentration of siRNA used for cell transfection (30 nM) was established previously (29,43). Cells were transfected using the Lipofectamine RNAiMAX lipid reagent (Invitrogen) as described previously (43,44). All Stars Negative Control siRNA (Qiagen, Venlo, The Netherlands) was used as negative control (siCTRL). This siCTRL does not affect beta cell gene expression or insulin release, as compared with nontransfected cells (43). After 16 h of transfection, cells were cultured for a 24-or 48-h recovery period before exposure to cytokines.
The plasmids used in the study are described in Table 3. HEK293T were transfected with 1 g of plasmid using Lipo-  Assessment of Cell Viability-The percentage of viable, apoptotic, and necrotic cells was determined after 15 min of incubation with the DNA-binding dyes propidium iodide (5 g/ml; Sigma) and Hoechst dye 33342 (5 g/ml; Sigma). A minimum of 600 cells was counted for each experimental condition by two independent observers, with one of them unaware of sample identity. The agreement between findings obtained by the two observers was always Ͼ90%. This fluorescence assay for single cells is quantitative and has been validated by systematic comparisons against electron microscopy observations, ladder formation, and caspase 3/9 activation (10,29,45,46). mRNA Extraction and Real Time PCR-Poly(A) ϩ mRNA was isolated from cultured cells using the Dynabeads mRNA DIRECT TM kit (Invitrogen) and reverse-transcribed as described previously (31,47). The real time PCR amplification reactions were done using iQ SYBR Green Supermix on a Light-Cycler instrument (Roche Applied Science) and on a Rotor-Gene Q (Qiagen), and the concentration of the gene of interest was calculated as copies per l using the standard curve method (31,48). Gene expression values in human and rat cells were corrected by respectively the housekeeping genes ␤-actin and Gapdh, whose expression is not modified in these cells by cytokine treatment (29,49). The primers used in this study are provided in the Table 5.
Western Blot-After cell culture and treatment, cells were washed with cold PBS and lysed using Laemmli Sample Buffer. Total protein was extracted and resolved by 8 -10% SDS-PAGE, transferred to a nitrocellulose membrane, and immunoblotted with the specific antibodies for the protein of interest (Table 2) as described (29). The intensity values for the proteins were corrected by the values of the housekeeping proteins ␣-tubulin or ␤-actin.
Immunoprecipitation-Proteins from transfected HEK293T cells were collected with cold lysis buffer (50 mM Tris, pH 7.5, 20 mM NaCl, 1% Nonidet P-40, 10 mM NaF, 1 mM Na 3 VO 4 , and Complete Protease inhibitor mixture, Roche Applied Science) and then precleared 1 h at 4°C with protein G-Sepharose 4 Fast Flow (GE Healthcare). The same amount of protein was then incubated overnight at 4°C, either with the anti-FLAG antibody or with the anti-IRE1␣ antibody (Santa Cruz Biotechnology, sc-20790). Normal mouse IgG (Santa Cruz Biotechnology, sc-2025) or normal rabbit IgG (sc-2027) were used as negative controls. After incubation for 1 h with Protein G-Sepharose 4 Fast Flow (GE Healthcare), the complexes were precipitated by centrifugation. The immunoprecipitates were washed five times with cold lysis buffer and then resuspended in 5ϫ Laemmli Sample Buffer. Immunoprecipitates, total proteins, and flow-through were run in 8% SDS-PAGE, transferred to a nitrocellulose membrane, and immunoblotted with the specific antibodies for the protein of interest (Table 2).
Insulin Secretion-For determination of insulin secretion, INS-1E cells were incubated for 1 h in glucose-free RPMI GlutaMAX-I medium and then incubated for 30 min in Krebs-Ringer solution without glucose. Cells were then exposed to 1.7, 16.7, or 16.7 mM glucose ϩ forskolin (20 M) for 30 min. Dispersed rat islet cells were incubated for 30 min in Krebs-Ringer solution containing 1.7 mM glucose. Cells were then exposed to 1.7, 16.7, or 16.7 mM glucose ϩ forskolin (20 M) for 1 h. Total insulin was extracted with acidic ethanol. Insulin was measured by the rat insulin ELISA kit (Mercodia, Uppsala, Sweden).
Histology-Pancreata from BALB/c control mice and diabetes-prone NOD mice were collected at different ages, namely 4, 9, 13, and 17 weeks for BALB/c and 3, 8, 12, and 17 weeks for NOD mice, fixed in 4% formaldehyde, and then paraffinembedded. 5-m-thick sections were subjected to antigen retrieval by boiling the sections in 10 mM citrate buffer, pH 6, for 10 min. After overnight incubation at 4°C with primary antibody against NMI, slides were double-stained with an anti-insulin antibody for 1 h at room temperature. Alexa Fluor fluorescent secondary antibodies (Invitrogen) were applied for 1 h at room temperature (Table 2). After nuclear staining with Hoechst, pancreatic sections were mounted with Fluorescence Mounting Medium (DAKO, Glostrup, Denmark). Immunofluorescence was visualized on a Zeiss microscope (Axio ImagerA1, Zeiss-Vision, Munich, Germany) equipped with a camera (AxioCAM Zeiss), as described previously (6). Images were acquired at room temperature using Zeiss/EC Plan Neofluar objective lenses at ϫ40 magnification and Axio Vision Zeiss software. Ethic Statements-Human islet collection and handling were approved by the local Ethical Committee in Pisa, Italy.
Male Wistar rats were housed and used according to the guidelines of the Belgian Regulations for Animal Care. All experiments were approved by the local Ethical Committee.
NOD mice, originally obtained from Professor Wu (Department of Endocrinology, Peking Union Medical College Hospital, Beijing, China) were housed and inbred in the KULeuven animal facility since 1989. BALB/c mice were purchased from Harlan (Horst, The Netherlands). All experimental procedures in mice were approved and performed in accordance with the Ethics Committee of the KULeuven, Leuven, Belgium.
Statistical Analysis-Data are expressed as mean Ϯ S.E. A significant difference between experimental conditions was assessed by one-way ANOVA followed by a paired Student's t test with Bonferroni correction. p values Ͻ 0.05 were considered statistically significant.

RESULTS
NMI Interacts with IRE1␣-We have presently used Array-MAPPIT to identify proteins that bind to the cytoplasmic domain of IRE1␣. This generated a list of 31 putative IRE1␣interacting proteins. 7 These results were compared against our microarray and RNA sequencing data from insulin-producing INS-1E cells (50), FACS-purified rat beta cells (51), and human islet cells (30) exposed to stress signals that are relevant for type 1 diabetes, namely the pro-inflammatory cytokines IL-1␤ ϩ IFN-␥ (1,29,30).
Five candidate proteins that interact with IRE1␣ and are modified by pro-inflammatory cytokine treatment in pancreatic beta cells were identified. NMI was chosen as the most promising one following additional selection based on literature review (Fig. 1).
To confirm the interaction between NMI and IRE1␣, a binary MAPPIT analysis was done (Fig. 2a). For this purpose HEK293T cells were co-transfected with an erythropoietin (Epo) receptor-based MAPPIT bait protein containing the cytoplasmic portion of IRE1␣ (pSEL-IRE1␣), a FLAG-tagged NMI prey (pMG1-NMI), and a STAT3-responsive luciferase reporter (pXP2d2-rPAP1). After stimulation with Epo, there was a 100-fold increased luciferase signal as compared with cells co-transfected with the empty prey vector. In addition, although the empty MAPPIT bait produced an aspecific signal when combined with the NMI prey, the specific signal with the IRE1␣ bait was nearly 10-fold stronger (Fig. 2a). As a control for bait expression levels, the interaction of empty and IRE1␣ baits with the REM2 (RAD and GEM-like GTP Binding 2) prey, which is known to bind the intracellular cytokine receptor portion of the MAPPIT bait, was investigated and indicated equal bait expression. These data show that when IRE1␣ and NMI are present in the cells, they interact and reconstitute a functional STAT3-based reporter assay leading to the expression of luciferase.
Binary MAPPIT analysis was used to assess the ability of NMI to bind to the kinase-defective K599A IRE1␣ mutant (Fig. 2b) (52). For this purpose, MAPPIT bait containing the kinase-defective K599A IRE1␣ mutant (pSEL-IRE1␣ K599A) was added to the assay described above. After stimulation with Epo, a 50-fold increased luciferase signal was observed in both IRE1␣ WT and K599A IRE1␣ mutant when combined with the NMI prey, as compared with the control (empty prey vector) (Fig.  2b). This result shows similar ability of NMI to interact with 7 S. Gerlo and J. Tavernier, unpublished data.  (50), FACS-purified rat beta cells (51), and human islet cells (30). As a result of this analysis, five candidate proteins that interact with IRE1␣ and are regulated by pro-inflammatory cytokines under all tested conditions were selected. Additional selection, based on literature survey, indicated NMI as the best candidate for further studies.
both IRE1␣ WT and the K599A mutant, suggesting that NMI-IRE1␣ interaction is independent on the phosphorylated state of IRE1␣.
The interaction between NMI and IRE1␣ was further validated by co-immunoprecipitation experiments. HEK293T cells were co-transfected with IRE1␣-V5 and NMI-FLAG-encoding plasmids and proteins collected and processed for immunoprecipitation. When proteins were immunoprecipitated with anti-FLAG antibody (Fig. 2c), IRE1␣ was detected in the immunoprecipitate (IP) as shown by Western blot analysis (Fig. 2c). In a mirror image of this experiment, immunoprecipitation of the proteins with an anti-IRE1␣ antibody (Fig. 2d) showed that the NMI-FLAG protein was present in the IP, whereas use of a nonspecific rabbit IgG (negative control) failed to detect NMI-FLAG protein (Fig. 2d, 3rd lane from left). The lack of NMI-FLAG signal or IRE1␣ signal from the flow-through of the NMI-FLAG IP or IRE1␣ IP, respectively, confirmed the reliability of the analysis.
Inflammation Increases NMI Expression in Pancreatic Islet Cells-We confirmed by real time-PCR (RT-PCR) our previous microarray findings (50,51) indicating that pro-inflammatory cytokines induce NMI expression in rat insulin-producing cells. There was a peak of NMI expression after 24 h in INS-1E cells (Fig. 3a) and at 6 h in FACS-purified rat beta cells (Fig. 3b).
Importantly, these findings were reproduced in human islets by real time PCR (Fig. 3c) and Western blot (Fig. 3d) after a 48-h incubation with IL-1␤ ϩ IFN-␥, confirming our previous RNA sequencing data (30). The increased expression of NMI in the clonal human beta cell line EndoC-␤H1 after treatment with a different combination of pro-inflammatory cytokines (Fig. 3e) indicates that cytokine-induced NMI expression takes place at least in part at the pancreatic beta cell level.
To test whether increased NMI expression occurs during beta cell inflammation in vivo, we evaluated NMI expression in NOD mice, a well known animal model of autoimmune diabetes (7,53). Immunofluorescence staining of pancreatic sections from NOD mice (Fig. 4a) showed NMI expression in insulinpositive cells at 8, 12, and 17 weeks (Fig. 4a, panels E, H, and K, and F, I, and L), whereas NMI was not detected in age-matched BALB/c mice islets (Fig. 4b, panels E, H, and K). Notably, NMI was not detected in NOD mouse islets at 3 weeks of age (Fig. 4a,  panels B and C), when the islets are still free from the immune cell infiltration (7), and in the islets of NOD mice at 8 weeks of age that were insulitis-free (Fig. 4c, panels B and C), whereas islets from the same mice but with insulitis had clear up-regulation of NMI expression (Fig. 4a, panels E and F). Our data suggest that insulitis is associated with increased NMI expression in mouse beta cells in vivo.
NMI Inhibition Does Not Affect IRE1␣ Endonuclease Activity and Insulin Secretion in Pancreatic Beta Cells-To assess the function of cytokine-induced NMI expression on IRE1␣ activ- ity, we knocked down (KD) NMI with a specific siRNA (siNMI). NMI KD in INS-1E cells (Fig. 5, a and b) did not modify cytokine-induced IRE1␣ endonuclease activity as evaluated by Xbp1 splicing (Xbp1s) (Fig. 5c) and Ins-2 mRNA degradation (Fig. 5, d and e). These data were confirmed in FACS-purified rat beta cells, where NMI KD (Ͼ60% inhibition; Fig. 5f) did not change cytokine-induced IRE1␣-dependent Ins-2 mRNA degradation (Fig. 5g).   (panels A, D, G, and J, green), NMI (panels B, E, H, and K, red), and with Hoechst for nuclear staining (panels C, F, I, and L, blue). The immunofluorescence analysis shows NMI expression in insulin-positive cells in NOD mice at 8, 12, and 17 weeks (panels F, I, and L, yellow), but not at 3 weeks (panels B and C). Infiltrated lymphocytes, a sign of insulitis, are indicated by white arrowheads (panels F, I, and L). Insulitis is present at 8, 12, and 17 weeks of age but not at 3 weeks. The images shown are representative of 14 Ϯ 3 islet sections from three different mice per age (a). Pancreatic sections from 4-, 9-, 13-, and 17-week-old BALB/c mice were stained with antibodies specific for insulin (panels A, D, G, and J, green), NMI (panels B, E, H, and K, red), and with Hoechst for nuclear staining (panels C, F, I, and L, blue). Immunofluorescence analysis indicates the absence of NMI expression in insulin-positive cells in BALB/c mice (panels C, F, I, and L, merge). The images shown are representative of 21 Ϯ 1 islet sections from three mice per age (b). Pancreatic sections from 8-week-old NOD mice were stained with antibodies specific for insulin Glucose-and forskolin-stimulated insulin secretion did not change in INS-1E (Fig. 6a) and dispersed rat islet cells (Fig. 6g) after NMI KD (Fig. 6, c and i) both in untreated and cytokine-treated cells (Fig. 6, a and g). As expected, IL-1␤ ϩ IFN-␥ treatment inhibited insulin secretion in both INS-1E and dispersed rat islet cells (Fig. 6, a and g) and decreased total insulin content in both models (Fig. 6, b and h). INS-1E cells had a mild decrease in insulin content and expression in NMI KD compared with siCTRL in the untreated condition (Fig. 6, b and d), but this was not confirmed in primary rat islet cells, where Ins-2, Maf-A expression (Fig. 6, j and k), and total insulin protein content (Fig. 6h) were similar after   -1 (f and l) was assessed by RT-PCR and normalized for the housekeeping gene Gapdh. **, p Ͻ 0.01, and ***, p Ͻ 0.001, versus untreated siCTRL at 1.7 mM glucose; $, p Ͻ 0.05, and $$$, p Ͻ 0.001, versus untreated siNMI at 1.7 mM glucose; §, p Ͻ 0.05, § §, p Ͻ 0.01, and § § §, p Ͻ 0.001 versus siCTRL untreated; ‡, p Ͻ 0.05, ‡ ‡, p Ͻ 0.01, and ‡ ‡ ‡, p Ͻ 0.001, versus siNMI untreated; #, p Ͻ 0.05, ##, p Ͻ 0.01, and ###, p Ͻ 0.001; @, p Ͻ 0.05, @@, p Ͻ 0.01, and @@@, p Ͻ 0.001; ¤, p Ͻ 0.05; ¤¤, p Ͻ 0.01, and ¤¤¤, p Ͻ 0.001 as indicated by bars. One-way ANOVA followed by Student's paired t test with Bonferroni correction. Data shown are mean Ϯ S.E. of 3-5 independent experiments. siCTRL and siNMI transfection. Pdx-1 expression did not change in either INS-1E (Fig. 6f) or in dispersed rat islet cells (Fig. 6l) following NMI KD.

FIGURE 3. IL-1␤؉IFN-␥ induce NMI expression in rodent and human beta cells. The expression of NMI was assessed by RT-PCR (a-c and e) and by Western blot (d) in INS-1E cells (a), FACS-purified rat beta cells (b), human islet cells (c and d), and the human beta cell line EndoC-␤H1 cells (e) and normalized for the housekeeping gene
NMI Modulates IRE1␣-dependent JNK activation-IRE1␣ activation also modifies c-Jun N-terminal kinase (JNK) activity (23), and we next evaluated whether NMI KD interferes with IRE1␣-dependent JNK phosphorylation in beta cells. NMI KD increased cytokine-induced JNK phosphorylation in INS-1E cells after 8 and 12 h of exposure (Fig. 7, a-e). This was confirmed by quantification of the area under the curves (Fig. 7, c). These observations were reproduced using a second siRNA targeting NMI (Fig. 7, f and g) and were confirmed in FACS-purified rat beta cells (Fig. 7, h and i). Double KD of NMI and IRE1␣ reversed the effect of NMI down-regulation on JNK phosphorylation (Fig. 8, a-c), confirming the regulatory function of NMI on IRE1␣-dependent JNK activation in pancreatic beta cells. Of note, the KD of IRE1␣ in cytokine-treated cells had a most marked inhibitory effect on JNK activation at 2 h (Fig. 8b), indicating a major contribution by IRE1␣ in the early JNK activation.
NMI-dependent JNK Modulation Regulates Apoptosis in Pancreatic Beta Cells-JNK has an important role for both cytokine-and ER stress-induced beta cell apoptosis (54 -58). In line with this, NMI KD increased apoptosis induced by cytokines (Fig. 9b) and the ER stressor CPA (Fig. 9d) in INS-1E cells and FACS purified rat beta cells (Fig. 9f), indicating the involvement of the UPR in the NMI-dependent regulation of apoptosis. Furthermore, NMI KD by two independent siRNAs also augmented apoptosis in human islet cells and the human beta cell line EndoC-␤H1, under both basal conditions and follow-   JULY 25, 2014 • VOLUME 289 • NUMBER 30

NMI Inhibits IRE1␣-dependent JNK Activation in Beta Cells
ing exposure to IL-1␤ ϩ IFN-␥ (Fig. 9, i and k). The KD of NMI, in the rat models investigated, was Ն50%, as confirmed by RT-PCR ( Fig. 9, a, c, and e) and between 30 and 40% in the human models, as confirmed by RT-PCR (Fig. 9g) and Western blot (Fig. 9, h and j).
We have previously shown that JNK/c-Jun-dependent DP5 expression contributes to apoptosis in cytokine-treated pancreatic beta cells (58). In agreement with these previous results, NMI KD increased c-Jun phosphorylation (Fig. 10, a and b) and Dp5 mRNA expression (Fig. 10c) in cytokine-treated INS-1E cells. These data suggest that NMI down-regulation in beta cells contributes to apoptosis via the IRE1␣-JNK-c-Jun-DP5 pathway.
A possible mechanism for cytokine-induced apoptosis in beta cells is increased nitric oxide production and consequent ER stress (4,5). NMI KD, however, did not increase nitric oxide production (Fig. 10d), making it unlikely that this is a relevant mechanism for beta cell apoptosis following NMI inhibition.
Effect of NMI on JNK Phosphorylation Is Mediated via Modulation of the TRAF2/MKK7 Pathway-IRE1␣ activates JNK through the binding of the adaptor protein TRAF2 and downstream activation of MKK7 (23). To determine whether TRAF2 is implicated in the observed phospho-JNK increase in NMIinhibited cells, we used an NMI/TRAF2 double KD approach. NMI/TRAF2 double KD partially reversed the effect of NMI down-regulation on JNK phosphorylation (Fig. 11, a-c), suggesting that TRAF2 is partially responsible for the increased JNK phosphorylation observed in NMI-silenced INS-1E cells. MAPK kinase 7 (MKK7) is the main kinase implicated in JNK activation during inflammation (59) and is a target of the apoptosis signal-regulating kinase 1 (ASK1) (60), the kinase downstream of TRAF2 (61). NMI/MKK7 double KD reversed the effect of NMI down-regulation on JNK phosphorylation in cytokine-treated INS-1E cells (Fig. 11, d-f). Importantly, it also reduced cytokine-induced beta cell apoptosis in NMI KD cells to the same level observed in siCTR-transfected cells (Fig. 11g). These results indicate that NMI KD in pancre-atic beta cells increases JNK phosphorylation via the TRAF2/ MKK7 pathway.

DISCUSSION
IRE1␣ is an important regulator of the UPR in pancreatic beta cells, and it may play a key role in the transition from adaptive to pro-apoptotic UPR in T1D (11,13,19). The mechanisms regulating IRE1␣ activation and its signaling for beta cell "adaptation," "stress response," or "apoptosis" remain to be clarified. To address these questions, we presently used a combined "omics" approach based first on an ArrayMAPPIT IRE1␣ interactome analysis and then comparison of these results against functional genomic analysis of human and rodent beta cells exposed to pro-inflammatory cytokines, as described in Fig. 1. This allowed the identification of NMI as a new cytokineinducible protein that interacts with IRE1␣ in pancreatic beta cells. Importantly, increased expression of NMI was also observed in beta cells exposed in vivo to the autoimmune assault in NOD mice, a well established model of T1D. Detailed mechanistic studies demonstrated that NMI negatively modulates IRE1␣-dependent activation of JNK and apoptosis in rodent and human pancreatic beta cells.
Cytokine exposure induces IRE1␣ activation in pancreatic beta cells and subsequent Xbp1 splicing (4), insulin degradation (20,21,62), and JNK phosphorylation (54,58). This early JNK phosphorylation is crucial for the ensuing cytokine-induced apoptosis (54 -57, 63). The present observations indicate that cytokines induce in parallel NMI expression. NMI will bind IRE1␣ and inhibit the IRE1␣/TRAF2/MKK7-dependent activation of JNK, providing a negative feedback for a key pro-apoptotic mechanism in beta cells. Indeed, following NMI KD, IRE1␣ is not turned off, and the persistent activation of JNK aggravates cytokine-induced apoptosis in pancreatic beta cells via the phospho-c-Jun/DP5 activation pathway.
It has been previously shown that IRE1␣-dependent JNK activation is associated with formation of the IRE1␣-TRAF2-ASK1 complex and contributes to cell apoptosis caused by pharmacological induction of ER stress (23,64). We have confirmed these observations in the context of cytokine-induced beta cell apoptosis, and we observed a novel role for NMI in this process. It is conceivable that during cytokine-induced UPR activation, as described for other beta cell stresses (13), cytosolic proteins modulate the induction of IRE1␣, fine-tuning the signals provided by accumulation of misfolded proteins in the ER. We have presently identified in NMI a novel fine-tuned modulator of IRE1␣, able to regulate both CPA-and cytokineinduced apoptosis in rodent and human pancreatic beta cells. Under the present experimental conditions, the KD of IRE1␣ inhibits cytokine-induced JNK phosphorylation after only 2 h of cytokine exposure (Fig. 8, a and b), an early time point that precedes NO production and SERCA-2b inhibition (4). It is thus conceivable that the main role of NMI is to switch off this early cytokine-induced IRE1␣-dependent JNK activation, preventing excessive beta cell apoptosis in the context, for instance, of a limited innate immune response. JNK phosphorylation levels decreased in both siCTRL and siNMI after 24 h of cytokine exposure (Fig. 7, a and b). This maybe due to the cytokine-dependent induction, in parallel with NMI, of other JNK regulators. Our RNA sequencing data of cytokine-treated FIGURE 11. Effect of NMI on JNK phosphorylation and apoptosis is TRAF2/MKK7 pathway-dependent. INS-1E cells were transfected with siCTRL, siNMI, siTRAF2 (siTR2), or co-transfected with siNMI ϩ siTRAF2 (siNMI ϩ siTR2) (a-c). After 48 h, cells were left untreated or incubated with IL-1␤ ϩ IFN-␥ (IL ϩ IFN) for 8 h, and proteins were analyzed by Western blot with phospho-specific JNK, total JNK, and TRAF2 antibodies. Tubulin expression was used as control for protein loading. One representative Western blot of four independent experiments is shown (a). The mean Ϯ S.E. of the optical density measurements of the Western blots for TRAF2 (b) and phospho-JNK (c) are shown. INS-1E cells were transfected with siCTRL, siNMI, siMKK7, or co-transfected with siNMI ϩ siMKK7 (d-g). After 48 h, cells were left untreated or incubated with IL-1␤ ϩ IFN-␥ (IL ϩ IFN) for 8 h, and proteins were analyzed by Western blot with phosphospecific JNK, total JNK and MKK7 antibodies. ␤-Actin expression was used as control for protein loading. One representative Western blot of three independent experiments is shown (d). The mean Ϯ S.E. of the optical density measurements of the Western blots for MKK7 (e) and phospho-JNK (f) are shown. Apoptosis was evaluated after 24 h of incubation with IL-1␤ ϩ IFN-␥ (ILϩIFN) treatment (g). **, p Ͻ 0.01 and ***, p Ͻ 0.001 versus siCTRL untreated; @, p Ͻ 0.05 siMKK7 ILϩIFN versus siMKK7 untreated; ‡, p Ͻ 0.05 siNMI ϩ siMKK7 ILϩIFN versus and siNMI ϩ siMKK7 untreated; ##, p Ͻ 0.01; ###, p Ͻ 0.001; $, p Ͻ 0.05; $$$, p Ͻ 0.001; §, p Ͻ 0.05 and § §, p Ͻ 0.01 as indicated by the bars; one-way ANOVA. Data shown are mean Ϯ S.E. of 3-4 independent experiments. human islet cells (30) support this hypothesis, showing a significant up-regulation of several phosphatases such DUSP1 (65) and DUSP16 (66,67) in IL-1␤ ϩ IFN-␥-treated islets.
NMI has been shown to potentiate STAT-dependent transcription in other cell types (68). Under our experimental conditions, however, activation of STAT-regulated pathways cannot explain the phenotype induced by NMI KD. Thus, we observed that NMI KD increases cytokine-induced apoptosis in INS-1E and human islet cells, whereas STAT1 KD protects beta cells against cytokine-induced apoptosis (34,50,69). Furthermore, STAT1 silencing decreases NO production in cytokinetreated beta cells (50), whereas NMI KD does not change it (Fig. 10d).
In conclusion, the discovery of this new role for NMI adds to our understanding of the regulation of key steps of the UPR in stressed beta cells. The present observations indicate that cytokine-induced NMI provides a negative feedback on IRE1␣-dependent activation of JNK and apoptosis in beta cells, suggesting a protective role for this protein in the early stages of T1D.