MondoA Is an Essential Glucose-Responsive Transcription Factor in Human Pancreatic β-Cells

In eukaryotic cells, glucose uptake and metabolism represent a major source of energy, but are also a strong regulator of gene expression and cellular function. Pancreatic β-cells represent a model system to dissect these processes, because they are responsible for orchestrating the response of the body to rising postprandial glucose levels by secreting insulin to avoid excessive hyperglycemia. Glucose enters β-cells via GLUTs and is first metabolized through the high-K_m glucokinase (GK; hexokinase IV), which is considered to be “glucose sensor” of the β-cell (1). After this, insulin secretion occurs through a process of cellular depolarization via ATP-sensitive potassium channels, calcium entry, vesicle docking, and exocytosis (2). The incretin hormones glucagon-like peptide 1 (GLP-1) and gastric inhibitory polypeptide further amplify insulin secretion. Both hormones act directly on β-cells to elevate intracellular cAMP levels and promote secretion downstream of glucose sensing. Both hormones also activate the transcription factor cAMP-responsive element–binding protein and thereby influence the β-cell transcriptome (3).

Although the critical role of glucose on insulin secretion is now well described both in rodent and human β-cells (2), the effect of glucose on the β-cell transcriptome has been less explored. Long-term hyperglycemic conditions have been shown to be detrimental to β-cell function, leading to decreased insulin transcription, synthesis, and secretion giving rise to the concept of glucolipotoxicity (4). However, there is a limited understanding of the shorter-term effects of glucose on the β-cell transcriptome, particularly in human models.

Carbohydrate-responsive transcription factors have emerged as major mediators of glucose action on gene expression. Adipocytes and hepatocytes express the carbohydrate-responsive element–binding protein (ChREBP), also named MondoB, whereas skeletal muscle cells express its paralog MondoA (5,6). Both transcription factors reside in the cytoplasm in low-glucose conditions and undergo nuclear translocation in high-glucose conditions. They belong to the same family, with ChREBP encoded by the MLX interacting protein–like (MLXIPL) gene and MondoA encoded by the MLX interacting protein (MLXIP). ChREBP and MondoA are multidomain proteins with highly homologous N-terminal and COOH-terminal regions. They contain a bHLHZ (basic helix-loop-helix leucine zipper) region and a COOH-terminal dimerization domain mediating DNA binding and heterodimerization with a common binding protein named MLX. Both ChREBP and MondoA contain transcriptional activation domains, whereas MLX is much shorter and lacks intrinsic transactivation capacity (7,8). The two complexes bind the carbohydrate response element consensus sequence in promoter regions of specific target genes (7). In hepatocytes, ChREBP shifts the cellular state to maximize glucose storage as lipids by upregulating glycolytic and lipogenic genes (9). In skeletal muscle, the activation of MondoA, which is predominant in this cell type, suppresses glucose uptake via increased expression of thioredoxin interacting protein (TXNIP) and the arrestin domain–containing protein 4 (ARRDC4) (10,11).

Previous work focusing on the immediate effects of glucose on the β-cell transcriptome has mainly used rodent models. This research has indicated a role for ChREBP in long-term cellular deterioration via lipotoxicity (12,13) or in promoting cellular proliferation (14). Recently, the effect of glucose in rat pancreatic cells was shown to occur via two transcriptional programs, one directly dependent on ChREBP and the other downstream of target genes of ChREBP (15). Furthermore, only limited insights have been obtained using human models. An early study (16) analyzing the transcriptional change of human islets exposed to high glucose levels for 24 h demonstrated TXNIP as the most highly upregulated gene. Importantly, the contribution of paralog transcription factor of ChREBP, MondoA, to glucose sensing in pancreatic cells has not been investigated.

Here, we probed the effect of glucose using the recently developed glucose-responsive EndoC-βH1 human cell line (17) and human islets. In contrast to previous reports in rodents, we found no role for ChREBP in upregulating glycolytic or lipogenic genes in response to short-term high-glucose treatment (1–24 h). In contrast to ChREBP, we observed that MondoA was responsive to high glucose concentrations in EndoC-βH1 cells and islets, leading to subsequent upregulation of TXNIP and ARRDC4 expression, thereby reducing cellular glucose uptake. Taken together, we propose that MondoA is an essential glucose-responsive transcription factor in human β-cells.

EndoC-βH1 cells (17) were cultured in low-glucose (5.6 mmol/L) DMEM (Sigma-Aldrich) with 2% BSA fraction V (Roche Diagnostics), 50 mmol/L 2-mercaptoethanol, 10 mmol/L nicotinamide (Calbiochem), 5.5 mg/mL transferrin (Sigma-Aldrich), 6.7 ng/mL selenite (Sigma-Aldrich), 100 units/mL penicillin, and 100 mg/mL streptomycin. Cells were seeded at a 40% confluence on plates coated with Matrigel (1%; Sigma-Aldrich)/fibronectin (2 mg/mL; Sigma-Aldrich). Cells were cultured at 37°C and 5% CO₂ in an incubator and passaged once a week when they were 90–95% confluent. The glucose, forskolin, GLP-1, mannoheptulose, H89, U0126, and PD98059 used in the experiment were from Sigma-Aldrich. cAMP-dependent protein kinase (PKA) inhibitor 14–22 amide (PKAi) was from Calbiochem.

Human islets were obtained from seven donors (mean age 55.8 ± 7.5 years; BMI 27.5 ± 1.5 kg/m²). Up to 100 handpicked islets were deposited in each well of a 12-well plate and cultured in the same culture medium as used for EndoC-βH1 cells.

ChREBP^−/− mice were previously described (18). Wild-type and homozygous ChREBP knockout (ChREBP^−/−) mice were used in accordance with the guidelines of the French Animal Care Committee. The mice were bred on a genetic C57BL/6J background and raised in a 12-h light/dark cycle. They were fed a standard laboratory chow diet. Islets were isolated from 12-week-old mice by collagenase digestion (Sigma-Aldrich) followed by direct handpicking. After overnight culture in DMEM containing 0.5 mmol/L glucose, groups of 50 islets in triplicate were preincubated for 8 h in DMEM containing 0.5 or 20 mmol/L glucose.

EndoC-βH1 cells were passaged and transfected using Lipofectamine RNAiMAX (Life Technologies) 24 h later. ON-TARGETplus small interfering RNA (siRNA) SMARTpool for human MLXIPL, MLXIP, or MLX, or ON-TARGETplus nontargeting control pool (siN) were used (Dharmacon). Briefly, siRNA and Lipofectamine RNAiMAX were combined in OptiMEM and applied to the cells. Three hours later, medium was changed for fresh culture medium. Cells were harvested 4–5 days post-transfection, with preliminary experiments showing that siRNA knockdown was consistently sustained for >7 days post-transfection.

On the day of receiving the samples, ∼100 human islets were handpicked for each condition, washed in PBS, and treated with 1 mL of Accutase (PAA Laboratories) for 5 min. Partial dissociation of the human islets was achieved with slow pipetting and Accutase was removed after a centrifugation step. The cell clusters were gently suspended in OptiMEM, and siN, siMlxip, or siMlxipl Lipofectamine RNAiMax complexes were added. Cell clusters were plated on plates coated with Matrigel (1%)/fibronectin (2 mg/mL). After 4–5 h, an equal amount of EndoC-βH1 cell culture medium was added. Three days later, the medium was changed to low-glucose (1 mmol/L) culture medium for 5 h, and then glucose (to 20 mmol/L) was added to half of the wells. After 16 h, the cells were harvested.

RNeasy Micro Kit (Qiagen) was used to extract total RNA from EndoC-βH1 cells and human islets. A First Strand cDNA Kit (Thermo Fisher Scientific) was used to synthesize cDNA. Quantitative RT-PCR was performed using Power SYBR Green mix (Applied Biosystems) with a QuantStudio analyzer. Custom primers were designed with Primer-Quest online software (IDT), and the efficiency was determined for each with a serial dilution of cDNA samples from EndoC-βH1 cells or human islets. Cyclophilin-A transcript levels were used for the normalization of each sample.

Transcriptomic profiles were obtained using GeneChip Human Gene 2.0 ST Array (Affymetrix), following the manufacturer instructions. Microarray data and all experimental details are available in the Gene Expression Omnibus (GEO) database (accession GSE98501).

EndoC-βH1 cells were cultured on Matrigel/fibronectin-coated glass coverslips. Cells were starved in low-glucose (0.5 mmol/L) culture medium overnight and exposed to low-glucose (0.5 mmol/L) or high-glucose (20 mmol/L) culture medium for 3 h the following day. Cells were fixed using 4% paraformaldehyde for 30 min and processed for immunostaining by blocking in 3% BSA, 2% serum, and 0.1% Tween-20 for 30 min. The cells were exposed to a primary antibody against MondoA raised in rabbit (ProteinTech) overnight at 4°C, washed three times, then exposed to a fluorescent anti-rabbit antibody for 2 h. Nuclei were stained with Hoechst 33342 fluorescent stain (Life Technologies). Images were acquired with a Leica Leitz Fluorescent Microscope equipped with cooled three-chip charge-coupled device camera (model C5810; Hamamatsu) and processed using ImageJ software.

EndoC-βH1 cells at 80% confluence were starved in low-glucose (0.5 mmol/L) culture medium overnight and exposed to different test compounds the following day for 3 h. The nuclear and cytoplasmic proteins were extracted using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Fisher Scientific), and protein concentrations were quantified using a Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). Proteins (40 μg) were resolved by SDS-PAGE and transferred to a membrane using an iBlot2 Gel Transfer Device (Thermo Fisher Scientific). Membranes were immunoblotted using antibodies against PDX1 (1:1,000) (19), tubulin (1:2,000; Sigma-Aldrich), ChREBP (1:1,000; Novus Biologicals), or MondoA (1:500; ProteinTech). After washing, membranes were incubated with species-specific horseradish peroxidase–linked secondary antibodies (1:10,000), washed again, and visualized after enhanced chemiluminescence exposure.

We transfected EndoC-βH1 cells with siRNA SMARTpool against human MLXIPL, MXIP, or siN. Five days later, cells were starved in low-glucose medium (0.5 mmol/L) for 6 h. Next, cells were cultured overnight in media containing 20 mmol/L glucose. Glucose uptake was measured using the calorimetric Glucose Uptake Assay Kit (Abcam) as per the instruction manual.

Data were analyzed using GraphPad Prism 6 software and are presented as the mean ± SD. For comparison between two mean values, statistical significance was estimated using a two-tailed Student t test. For comparison among three or more values, one-way ANOVA was used with Bonferroni post hoc test (repeated-measures).

We searched for genes whose expression was induced by glucose in EndoC-βH1. Microarray analysis was used to compare transcriptional profiles of EndoC-βH1 cells after 8 h of exposure to low or high glucose, with or without forskolin. Expression profiles are depicted in Fig. 1 as scatter plots, which show the relative intensities of all probes represented on the microarrays. High glucose exposure upregulated a limited number of genes, and the most prominent was TXNIP (Fig. 1A). The scarce number of upregulated genes observed was apparent when compared with the effects observed after forskolin treatment (Fig. 1B) (6 vs. 75 genes greater than twofold upregulated in each condition, respectively). Surprisingly, analyses of microarray data indicated that in EndoC-βH1 cells, glucose did not significantly upregulate the vast majority of known ChREBP targets (Fig. 1C). Specifically, although glucose treatment increased TXNIP mRNA levels, it did not modulate the expression of ACACA, MLXIPL, PFKL, PKLR, SCD, RGS16, HBEGF, GPD1, and RORC (Fig. 1C).

To further investigate the upregulation of glucose-dependent genes, EndoC-βH1 cells were exposed to different concentrations of glucose (1, 5, 15, or 20 mmol/L) for 8 h. Quantitative PCR (qPCR) analyses indicated that 15 and 20 mmol/L glucose caused a robust induction in TXNIP gene expression, whereas 5 mmol/L did not (Fig. 2A). The glucose-dependent stimulation of TXNIP gene expression has been attributed to ChREBP in rodent β-cells (20). qPCR analyses indicated that, with the exception of TXNIP, all other ChREBP target genes tested (ACACA, MLXIPL, PFKL, PKLR, RGS16, HBEGF, GPD1, and RORC) were not significantly upregulated in EndoC-βH1 cells exposed to 20 mmol/L glucose for 1, 4, 8, or 24 h (Fig. 2B and C and Supplementary Fig. 1). To test whether this was also the case in primary human cells, human islets from donors were exposed to media containing 1 or 20 mmol/L glucose, and, consistently, only TXNIP was found to be significantly upregulated (Fig. 2D and E and Supplementary Fig. 2).

To elucidate whether the glucose-dependent activation of TXNIP required glucose metabolism, EndoC-βH1 cells and human islets were preincubated with 25 mmol/L of GK inhibitor mannoheptulose. This prevented the glucose-dependent upregulation of TXNIP (Supplementary Fig. 3A and B). Furthermore, forskolin and the GLP-1 mimetic Exendin-4, which both activate the cAMP pathway, decreased glucose-induced TXNIP expression in both EndoC-βH1 cells and human islets (Supplementary Fig. 3C and D). Taken together, glucose upregulates TXNIP expression similarly in both human β-cell lines and human islets, and this effect is dependent on glucose metabolism and cAMP signaling.

To establish the necessity of ChREBP in glucose-induced TXNIP expression, islets from wild-type and ChREBP^−/− mice were isolated and exposed to either low (0.5 mmol/L) or high (20 mmol/L) glucose for 8 h. qPCR analyses indicated that txnip induction by glucose was similar in both groups (Fig. 3), demonstrating that ChREBP is not required to drive txnip expression upon glucose stimulation in rodent islets.

To the best of our knowledge, the expression and function of the ChREBP paralog MondoA has not been evaluated in detail in pancreatic β-cells. qPCR indicated that human EndoC-βH1 cells and human islets expressed high levels of ChREBP (MLXIPL) and MondoA (MLXIP) as well as their obligatory binding partner MLX (Fig. 4A and B). Moreover, data from GeneChip (Affymetrix) analyses indicated that ARRDC4, a key target of MondoA in skeletal muscles (10), was significantly upregulated by high glucose concentrations in EndoC-βH1 cells (Fig. 1C). This upregulation was validated by qPCR both in EndoC-βH1 cells (Fig. 4C) and in human islets (Fig. 4D), and was prevented by GK inhibition and cAMP activation (Fig. 4E and F) in a similar fashion to TXNIP (Supplementary Fig. 3).

Both MondoA and ChREBP are localized to the cytoplasm in an inactive state, and a glucose stimulus causes their translocation to the nucleus (7). To determine how glucose signals in β-cells, we studied the cytoplasmic and nuclear localization of MondoA and ChREBP in EndoC-βH1 cells after glucose treatment. Immunoblotting for MondoA revealed that 3 h of 20 mmol/L glucose exposure caused MondoA nuclear translocation (Fig. 5A), which was further confirmed by immunocytochemistry (Fig. 5B and C for quantification). Of note, the nuclear translocation of MondoA was prevented by forskolin, consistent with the inhibition of ARRDC4 and TXNIP transcription by the compound (Fig. 5A). The effects of forskolin were mediated by PKA. Indeed, PKA inhibitors (PKAi and H89) but not mitogen-activated protein kinase/extracellular signal–regulated kinase (ERK) kinase (U0126) or ERK (PD98059) inhibitors blunted the effects of forskolin on glucose-induced TXNIP (Fig. 5D). Surprisingly, we did not observe ChREBP nuclear translocation upon glucose exposure (Fig. 5E), indicating that MondoA, but not ChREBP, is sensitive to short-term glucose in EndoC-βH1 cells.

To establish whether MondoA is the primary glucose-responsive transcription factor in human β-cells, its encoding mRNA, or that of ChREBP or their obligatory binding partner MLX, were knocked down in EndoC-βH1 cells using siRNA (Fig. 6A–C). ChREBP silencing had no effect on the glucose-induced upregulation of TXNIP or ARRDC4 (Fig. 6D and E). On the other hand, knocking down MondoA or MLX significantly compromised the effect of glucose on both ARRDC4 and TXNIP upregulation (Fig. 6D and E).

ARRDC4 and TXNIP both control glucose uptake in skeletal muscle cells (21,22). To determine whether MondoA regulates glucose uptake in human β-cells, glucose uptake was measured in EndoC-βH1 cells. MondoA knockdown selectively suppressed TXNIP and ARRDC4 gene expression in EndoC-βH1 cells, whereas ChREBP silencing had no effect on these genes (Supplementary Fig. 4). Consistent with this observation, glucose uptake was unaltered in ChREBP-silenced cells but was significantly higher in cells with reduced MondoA expression (Fig. 6F).

Finally, we asked whether the glucose-dependent regulation of ARRDC4 and TXNIP is also MondoA dependent in primary human β-cells. For this purpose, we performed siRNA-mediated knockdown of MondoA and ChREBP in human islets. ChREBP siRNA induced a nearly 80% decrease in ChREBP RNA steady-state level, whereas MondoA siRNA induced a nearly 70% decrease in MondoA RNA steady-state levels without any impact on ChREBP mRNA levels (Fig. 7A and B). Importantly, and consistent with data obtained in EndoC-βH1 cells, ChREBP knockdown did not alter glucose-induced ARRDC4 and TXNIP upregulation, whereas such inductions were blunted after MondoA knockdown (Fig. 7C and D).

There have been limited observations of the effects of glucose on the human β-cell transcriptome. Here, using EndoC-βH1 cells and islets from organ donors, we demonstrated that the glucose-responsive transcription factor MondoA is robustly expressed in pancreatic β-cells. MondoA undergoes nuclear translocation in high-glucose conditions where it upregulates the expression of a limited number of genes including TXNIP and ARRDC4, both of which are involved in glucose uptake inhibition. Therefore, we conclude that MondoA is the glucose-responsive transcription factor in human pancreatic β-cells.

Studies have previously analyzed glucose-regulated gene expression in rodent β-cells using either β-cell lines or islet preparations, whereas fewer data have been generated using human β-cells (15). However, improving knowledge of human β-cell physiology is crucial, as, despite many similarities, rodent and human β-cells differ on various specific points, including disparities in β-cell function (23,24). The limited number of studies using human β-cells mainly stems from the difficulty in accessing primary human islet preparations that derive from deceased donors. Available human islets often come from transplantation rejection and therefore are not in an optimal state for further ex vivo studies. Moreover, they contain β-cells in different proportions from one preparation to another, giving rise to data variability (25). The difficulty of generating functional human β-cell lines has also represented a major limitation for decades (26). The present work mainly focuses on glucose-regulated gene expression in pancreatic β-cells in a human context, using both human islets from donors and the recently developed functional human β-cell line EndoC-βH1 (17). In these two models, we observed a very limited number of genes that were efficiently affected by short-term high-glucose treatment. This is in accordance with a previous microarray analysis (16) reporting only 14 genes significantly upregulated in human islets upon 24-h exposure to high-glucose exposure. Of note, only five of these genes were induced more than twofold, with TXNIP being the most upregulated gene. In contrast, it was recently reported that >2,000 genes are significantly induced by glucose after a 12-h treatment in the rat β-cell line INS-1 (15). Although improvements in the sensitivity of recent transcriptomic tools can account for some increase in the number of detectable genes, a clear underlying difference between the two-species transcriptional response to glucose appears to exist. Among known glucose-related species differences, glucose transport in rodent β-cells is mainly dependent on GLUT2 (SLC2A2), a low-K_m GLUT, whereas the high-K_m GLUT1 (SLC2A1) plays a major role in human β-cells (27); Moreover, the panel of voltage-gated ion channels differs between rodent and human β-cells (28), and different set points for glucose-stimulated insulin secretion have been reported in these two species (28). Regarding glucose-induced cell cycle entry, glucose efficiently activates a large set of cell cycle–related genes that activate β-cell proliferation in rodent β-cells (29), a process that is less efficient in the case of human β-cells (30). Altogether, these data suggest that human β-cells could be quantitatively less sensitive to glucose in terms of the regulation of gene expression than rodent β-cells.

Historically, it was first thought that in pancreatic β-cells the upstream stimulatory factors regulated glucose stimulation of gene expression (31). ChREBP was next proposed as the β-cell carbohydrate-responsive transcription factor based on a series of arguments. ChREBP was first reported as expressed in pancreatic islets and in β-cell lines (32) and its overexpression in INS-1 cells using the Tetracycline-On system improved the glucose-dependent induction of its target gene l-pyruvate kinase (LPK) (32). Moreover, overexpression of a constitutively active mutant of ChREBP in INS-1 cells led to the induction of several ChREBP target genes, including TXNIP (33). Finally, glucose was shown to regulate gene expression in pancreatic β-cells through the carbohydrate response element consensus sequence that ChREBP binds (34). However, the direct evidence that ChREBP acts as the primary glucose transcription factor has not always been consistent. Indeed, although studies have reported that chrebp knockdown in INS-1 cells alters glucose-stimulated txnip expression (34), others have found either no or limited effects in both INS-1 cells and rat islets (35). Here, we demonstrate that islets from ChREBP^−/− animals have a similar upregulation of txnip compared with islets from wild-type littermates. Although this result does not preclude a role for ChREBP in pancreatic β-cells, it directly demonstrates that ChREBP is not necessary for glucose-stimulated txnip expression in mouse pancreatic islet cells. Other functions for ChREBP in human islets may yet be discovered, which could relate to its described nuclear translocation and activation under periods of endoplasmic reticulum stress induced by thapsigargin (36).

A key finding of the present work is that MondoA plays a major role in glucose-stimulated gene expression in human pancreatic β-cells. We observed that MondoA is expressed in β-cells. Moreover, glucose induced MondoA nuclear translocation and activated the expression of its target genes, as is the case in muscle cells (10). It was previously postulated, based on data from HEK293T cells expressing different forms of hexokinase, that MondoA activity requires glucose metabolism. We demonstrate here that this is also the case in human β-cells as MondoA activity was inhibited by mannoheptulose, an inhibitor of the GK enzyme, the hexokinase that catalyzes the first reaction in the glycolytic pathway in pancreatic β-cells.

Our experiments performed with either forskolin or the GLP-1 receptor agonist Exendin-4 demonstrated that the translocation of MondoA as well as its transcriptional activity were inhibited by cAMP signaling. Although this type of inhibitory effect has been observed for the regulation of ChREBP activity in the liver (37), cAMP effects of MondoA translocation have not been previously described. ChREBP is phosphorylated by PKA at Ser196 in response to glucagon, leading to a cytoplasmic localization that involves the 14–3–3 protein (37). This specific PKA phosphorylation site has not been described in the MONDOA sequence, and our results may suggest another PKA phosphorylation site that needs to be characterized.

Interestingly, previous data indicated that Exendin-4 acts as an antiapoptotic agent on β-cells by decreasing txnip expression, though the authors did not elucidate the mechanism by which Exendin-4 exerted this inhibitory effect (38). Our data would suggest that this antiapoptotic effect of Exendin-4 occurs through MondoA inhibition. Recently, an inhibitor of MondoA has been reported to have beneficial effects on insulin and glucose handling in high fat–fed mice (39). Based on our findings of MondoA actions in β-cells, it would be interesting to explore whether its specific inhibition here could be in part responsible for the observed improved glucose handling. Of note, glucose, which induces insulin secretion, also activates the expression of TXNIP and ARRDC4, both of which are involved in glucose uptake inhibition. These new data suggest that glucose entrance in human β-cells induces a transcriptional response that triggers a negative regulatory feedback loop on glucose uptake, which might contribute to the transitional glucose signal in these cells.

Our data indicate that MondoA plays an essential role in glucose-stimulated gene expression in human pancreatic β-cells. However, its functional role in β-cells is not yet elucidated. It has been demonstrated that its paralog ChREBP is required for glucose-stimulated β-cell proliferation (14). Indeed, the loss of ChREBP decreases glucose-induced BrdU incorporation in isolated human β-cells and in β-cells isolated from ChREBP-deficient mice. Given the evidence for the involvement of MondoA in cancer cell proliferation through its transcriptional control of TXNIP and subsequent effects on glucose uptake (40), future studies using MondoA-deficient mice will be useful to determine whether MondoA is implicated in glucose-mediated β-cell proliferation in physiological conditions, but also after β-cell injury.

Until now, studies on the regulation of gene expression by glucose in pancreatic β-cells has mainly focused on ChREBP because of the many similarities between hepatocytes and β-cells, including endodermal origins and metabolic functions. However, our study notably demonstrates for the first time that its paralog, MondoA, is an essential glucose-responsive transcription factor in human pancreatic β-cells.

Acknowledgments. The authors thank Dr. B.B. Kahn (Harvard Medical School, Boston, MA) for sharing ChREBP^−/− mice and Dr. D.E. Ayer (University of Utah, Salt Lake City, UT) for suggestions on MondoA antibodies. The authors also thank the transcriptomic platform from the Cochin Institute for performing array hybridizations and N. Glaser (INSERM U1016) for help in further data analyses.

Funding. P.R. was supported by a postdoctoral grant from Agence Nationale de la Recherche (Laboratoire d’Excellence Revive, Investissement d’Avenir, ANR-10-LABX-73). This study was funded by the Cochin Internal Program PIC (to L.R. and S.G.). The R.S. laboratory is supported by Agence Nationale de la Recherche (ANR-10-LABX-73) and the Bettencourt Schueller Foundation. The C.P. and R.S. research groups belong to the Département Hospitalo Universitaire (DHU). This project received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement number 115881 (RHAPSODY). This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation programme and EFPIA. This work is supported by the Swiss State Secretariat for Education Research and Innovation (SERI) under contract number 16.0097.

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

Author Contributions. P.R. conceptualized the work, performed investigations, and wrote the manuscript. L.R. and S.G. conceptualized the work, performed investigations, and reviewed and edited the manuscript. M.O. performed investigations. P.M., M.B., and M.A. provided human islets. C.P. conceptualized the work and reviewed and edited the manuscript. R.S. conceptualized the work, wrote the manuscript, and supervised the work. R.S. 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.