STAT3 Regulates Mitochondrial Gene Expression in Pancreatic β-Cells and Its Deficiency Induces Glucose Intolerance in Obesity

Obesity is the world’s major metabolic disorder, affecting 1.94 billion people worldwide (World Health Organization, www.who.int/). While physical inactivity, poor dietary habits, and excessive accumulation of lipids due to obesity are the major risk factors of type 2 diabetes (T2D), the pathophysiology of the disease is incompletely understood (1). Chronic hyperglycemia is a feature of T2D due to insulin resistance in the liver, skeletal muscle, and adipose tissue and progressive insulin secretory defects in pancreatic β-cells (2). Although insulin resistance is commonly observed in obesity, most obese individuals balance the enhanced hormonal demand by increasing insulin secretion, without developing T2D (1,3). Thus, insulin-secreting β-cells in the pancreas undergo “β-cell adaptation” in an attempt to maintain normoglycemia despite enhanced insulin demand in obesity (4). Adaptation is characterized by increased insulin production/secretion and expansion of β-cell mass, as well as heightened responsiveness to glucose, fatty acids, and glucagon-like peptide 1 (1). In individuals with diabetes, however, this compensatory response no longer overcomes insulin resistance, leading to deterioration of β-cell mass and decline in insulin secretory function (5). The disease slowly aggravates and progresses from impaired glucose tolerance to overt diabetes (6).

Signal transducer and activator of transcription 3 (STAT3) was initially identified as a nuclear transcription factor (7). It is activated by phosphorylation on tyrosine (Y)705 in response to a large array of cytokines and growth factors regulating the expression of genes that control cell proliferation, differentiation, and survival (8). Emerging evidence implicates STAT3 as a central regulator of metabolism through its noncanonical activity in the mitochondria (9,10), affecting the expression of mitochondrial genes (11–13) and directly regulating oxidative phosphorylation and ATP production by augmenting electron transport chain activity (14).

Developmental overactivation of STAT3 in pancreatic β-cells promotes neonatal diabetes (15–17). In addition, STAT3 has been identified by single β-cell analysis, matching the positive correlation of STAT3 expression with β-cell–mediated control of glucose homeostasis and β-cell dysfunction in T2D (18). Transgenic mice with floxed Stat3 alleles have allowed the dissection of cell-specific activities of STAT3 (19). In three independent studies, mice in which the Stat3 gene was constitutively deleted from β-cells with use of a RIP-Cre transgene displayed glucose intolerance, defects in early-phase insulin secretion, mild obesity on a chow diet, and enhanced β-cell sensitivity to multiple low doses of streptozotocin (20–22). Although these previous studies suggest that STAT3 plays a role in glucose homeostasis, the use of the RIP-Cre mouse model is controversial (23). Firstly, the expression of the Cre transgene by the Ins2 promoter in β-cells can induce glucose intolerance (24). Secondly, unspecific STAT3 deletion in the brain, which disrupts leptin signaling and food intake, was observed in the RIP-Cre;STAT3^lox/lox mice (20,21). Finally, deletion of STAT3 from murine pancreatic epithelium and islets using a Pdx1-Cre transgene did not reveal any metabolic phenotype (25). Thus, the role of STAT3 in vivo in β-cell–mediated control of glucose homeostasis and in the context of diet-induced obesity still remains unclear.

Taken together, previous work suggests that STAT3 can have specific activities in β-cell development and function. We report for the first time a new molecular mechanism whereby obesity induces noncanonical STAT3 expression in β-cells and its deficiency results in glucose intolerance. Our study postulates STAT3 as a novel regulator of mitochondrial DNA gene transcription and electron transport chain function in β-cells.

We studied 15 pancreas specimens from organ donors for immunofluorescence staining. The clinical characteristics of the organ donors are shown in Supplementary Table 1. The use of human tissues was approved by the French regulation body (French Agency of Biomedicine, authorization no. PFS12-013). Isolation of human islets from six organ donors without diabetes (kindly provided by Piero Marchetti, University of Pisa, Pisa, Italy) (clinical characteristics available in Supplementary Table 2) was performed in accordance with the local ethics committee in Pisa, Italy. After arrival in Brussels, islets were dispersed and cultured as previously described (26). The study was approved by the local ethics committees and the work in our laboratory by the Erasme Hospital Ethics Committee.

Mice were maintained at St Vincent’s Institute of Medical Research animal care facility on a 12-h light-dark cycle in a temperature-controlled room and obtained food and water ad libitum. STAT3^lox/lox mice were generated on a C57BL/6 background as previously described (19). Conditional tissue-specific deletion of STAT3 was generated by crossing STAT3^lox/lox mice with mice expressing the CreERT under the mouse Ins1 promoter (MIP-Cre/ERT) (27). Cre-induced inactivation of the Stat3 gene was carried out in young adult mice (8–10 weeks old) via three gavages of 6 mg tamoxifen (Sigma-Aldrich, St. Louis, MO), dissolved in corn oil at 30 mg/mL over a 5-day period with a 1-day break in between. Male mice were kept on regular chow (20% protein, 6% fat, and 3.2% crude fiber) or placed at 8–10 weeks of age on a high-fat diet (SF04-027; Specialty Feeds, Perth, Western Australia) for 14–20 weeks (28,29). The nutritional composition of the high-fat diet was 18.4% protein, 23.5% fat, and 4.7% crude fiber. The calculated composition of fatty acids in the high-fat diet is as follows: 14.31% total saturated fats, 7.54% total monounsaturated fats, and 2.07% total polyunsaturated fats. In this diet, 46% of total energy is from lipids, 20% of total energy from protein, and the remainder from carbohydrates. All animal studies were conducted at St Vincent’s Institute of Medical Research following the guidelines of the Institutional Animal Ethics Committee.

Mice were fasted for 6 h, and glucose (2 g/kg; Sigma-Aldrich) was either injected into the intraperitoneal cavity or delivered into the stomach by a gavage needle for intraperitoneal or oral glucose tolerance tests, respectively. Blood samples were obtained through a tail nick, and blood glucose was measured with a standard glucometer (Accu-Chek Performa; Roche, Basel, Switzerland). Serum insulin concentration was determined with a commercial ELISA kit (EZRMI-13K; Millipore) according to manufacturer’s instructions. Mice were fasted for 4 h before the intraperitoneal insulin tolerance test. Insulin (Actrapid; Novo Nordisk, Bagsværd, Denmark), at a dose of 0.65 mU/g i.p., was injected, and blood glucose was measured after tail bleeding. Body and tissue composition was determined by MRI scanning (EchoMRI, Houston, TX). After 18–20 weeks of chow/high-fat feeding, mice were sacrificed and tissues were weighed and collected for further analysis.

Mouse islets were isolated (28,30), pelleted, washed in PBS, and dispersed into single cells with incubation in soft trypsin (10.32% w/v bovine trypsin, 20 mmol/L EDTA in PBS 1×). This single-cell suspension was allowed to rest in a 37°C incubator containing 5% CO₂ for 1 h prior to staining or sorting. For flow cytometry analysis, cells were permeabilized by resuspending in 4% paraformaldehyde and 0.1% saponin in PBS and incubated for 30 min at room temperature and stained with primary STAT3 (cat. no. 4904; Cell Signaling Technology), insulin (A056401-2; Dako), and/or glucagon (G2654; Sigma-Aldrich) antibodies. Cells were washed and incubated with Alexa Fluor secondary antibodies (Thermo Fisher Scientific, Waltham, MA) for 30 min at room temperature. After staining, cells were washed, fixed in 1% paraformaldehyde, and analyzed with a BD LSRFortessa cell analyzer. Data were analyzed with FlowJo 8 software.

Insulin-producing MIN6 cell line was cultured in DMEM (Life Technologies) supplemented with 10% FCS. The human β-cell line EndoC-β¹H (31) (kindly provided by the INNODIA consortium and Dr. R. Scharfmann, University of Paris, Paris, France) was cultured in Matrigel fibronectin-coated plates. Cells were cultured in DMEM containing 5.6 mmol/L glucose, 2% BSA fraction V (Roche, Manheim, Germany), 50 μmol/L 2-mercaptoethanol (Sigma-Aldrich), 10 mmol/L nicotinamide (Calbiochem, Darmstadt, Germany), 5.5 μg/mL transferrin, 6.7 ng/mL selenite (Sigma-Aldrich), 100 units/mL penicillin, and 100 μg/mL streptomycin (Lonza, Leusden, the Netherlands). EndoC-β¹H and dispersed human islets were transfected with siRNAs targeting STAT3, STAT1, or AllStars Negative Control siRNA (30 nmol/L; QIAGEN, Venlo, the Netherlands) with use of Lipofectamine RNAiMAX (Invitrogen, Carlsbad, CA). siRNA target sequences are provided in Supplementary Table 3.

The induced pluripotent stem cell (iPSC) line HEL46.11 (derived from human neonatal foreskin fibroblast) was reprogrammed using Sendai virus technology as previously described (32,33). Undifferentiated cells were cultured on plates coated with Matrigel (Corning BV, Amsterdam, the Netherlands) in E8 medium (A1517001; Life Technologies) and passaged with 0.5 mmol/L EDTA (15575-038; Life Technologies). The cells were differentiated into β-like cells with use of a 30-day protocol (17,34–36) (Supplementary Table 4).

Cells were lysed with RIPA buffer, and total proteins were extracted and resolved by SDS-PAGE, transferred onto a nitrocellulose membrane, and immunoblotted with antibodies described in (37) Supplementary Table 5. The intensity values for the protein bands were corrected by the values of the housekeeping protein GAPDH, and β-actin or α-tubulin was used as a loading control. Extraction of total proteins or differential cell fractionation in EndoC-β¹H was performed with a Mitochondria/Cytosol Fractionation Kit (ab65320; Abcam, Cambridge, U.K.). MIN6 cells were collected by scraping with a rubber policeman, washed, and resuspended in buffer A (20 mmol/L HEPES, pH 7.6; 220 mmol/L mannitol; 70 mmol/L sucrose; 1 mmol/L EDTA; 0.5 mmol/L phenylmethylsulfonyl fluoride; and 2 mg/mL BSA). Cells were incubated on ice for 15 min to facilitate cell swelling before being subjected to nitrogen cavitation under 200 PSI of pressure (Parr Instrument Company, Moline, IL). Cell homogenates were centrifuged at 800g at 4°C and the mitochondria containing supernatant retained and centrifuged at 10,000g at 4°C. The supernatant representing the crude mitochondrial fraction was resuspended in 1 mL solution B (20 mmol/L HEPES, pH 7.6; 220 mmol/L mannitol; 70 mmol/L sucrose; 1 mmol/L EDTA; and 0.5 mmol/L PMSF) and loaded on top of a stepwise Percoll (GE Healthcare, Chicago, IL) gradient comprised of 1 mL of 80% Percoll/balance solution A, 4.5 mL of 56% Percoll/balance solution A, and 4.5 mL of 23% Percoll/balance solution A. Gradients were centrifuged at 65,000g and mitochondria isolated from the junction of the 56% and 23% layers. Mitochondria were washed twice in solution B, and mitochondrial protein content was detected with addition of 1 μL mitochondrial suspension to 600 μL of 50 mmol/L Tris pH 7.4 and 0.1% w/v SDS and measurement of the absorbance at 280 and 310 nm. Pellets containing nuclei following 800g centrifugation in the initial mitochondrial purification step were washed twice in solution B with centrifugation at 800g after each wash. The final pellet was resuspended in lysis buffer (50 mmol/L Tris, pH 8; 0.5% Triton X-100; 150 mmol/L NaCl; 0.1 mmol/L EDTA; 0.1 mmol/L EGTA; and 10% glycerol at 4°C prior to centrifugation at 18,000g to clarify nuclear protein. Cytosolic fractions were prepared with collection of the supernatant after the initial 10,000g crude mitochondrial isolation centrifugation, which was centrifuged at 100,000g in a benchtop ultracentrifuge (Beckman Coulter, Brea, CA). Protein concentration was characterized by Bradford assay (Thermo Fisher Scientific). Equal amounts of proteins were resolved by 12 or 4–12% (Bio-Rad Laboratories) SDS-PAGE (28–30).

Poly(A)+ mRNA extraction was performed with the Dynabeads mRNA DIRECT Kit (Invitrogen) in accordance with the manufacturer’s instructions; reverse transcription was carried out with a reverse transcriptase kit (Eurogentec, Liège, Belgium). Human quantitative real-time PCR was performed with the Bio-Rad CFX machine (Bio-Rad Laboratories) and the SYBR Green reagent. Mouse quantitative real-time PCR was performed with the Rotor-Gene RG 3000 machine (Corbett Research, QIAGEN, Hilden, Germany) and the TaqMan PCR Master Mix (AmpliTaq Gold with GeneAmp kit; Applied Biosystems). Data were analyzed with a standard curve or calculated with the ddCT method. β-Actin and/or GAPDH were used as internal controls. Probe and primer details are provided in Supplementary Table 6.

RNA quality analysis, library preparation, and sequencing were performed by the BRIGHTcore facility (Brussels, Belgium). The sequencing was performed on an Illumina NovaSeq 6000. An average of 20 million paired-end reads of 100 nucleotides in length were obtained per sample. The lists of up- and downregulated genes/transcripts and association with canonical pathways were determined with the packages limma and EGSEA in RStudio.

Human and murine pancreas paraffin sections (4–5 µm) were dewaxed, and antigen unmasking was realized using heated citrate buffer. Sections were permeabilized with 0.1% triton and unspecific binding sites blocked with 5% normal goat serum (1 h room temperature). Incubation of primary antibodies was performed overnight at 4°C followed by 1-h incubation with the secondary antibody conjugated to the fluorochrome (Supplementary Table 5). Nuclei were counterstained with DAPI (1 µg/mL; MilliporeSigma) before mounting. Images were analyzed on a Nikon A1R-A1 confocal microscope (Nikon Corporation, Tokyo, Japan) or Nikon Eclipse 50i microscope. The percentage of insulin- and glucagon-positive staining per islet and fluorescence intensity quantification within the human islet were analyzed with ImageJ software. Protein colocalization was estimated with JACoP (38).

Cells were fixed overnight in 2.5% glutaraldehyde (Electron Microscopy Sciences, Hatfield, PA) at 4°C and postfixed with 1% osmium tetroxide (Electron Microscopy Sciences) and 1.5% ferrocyanide (Electron Microscopy Sciences) in 0.15 mol/L cacodylate buffer. The cells were further stained with 1% uranyl acetate (Electron Microscopy Sciences), serially dehydrated, and embedded in epoxy resin (Agar 100 resin; Agar Scientific, Stansted, Essex, U.K.). Ultrathin 70-nm sections were produced with a Leica EM UC6 Ultramicrotome and mounted on copper Formvar/carbon grids (Electron Microscopy Sciences). Observations were made on a Tecnai 10 electron microscope (FEI, Eindhoven, Netherlands), and images were captured with a Veleta camera and processed with Olympis Soft Imaging Solutions iTEM software (Hamburg, Germany).

Oxygen consumption rates (OCR) of EndoC-β¹H cells, dispersed human islets, or β-like cells were measured with the Seahorse XFp Extracellular Flux Analyzer with Seahorse XFp FluxPaks (Agilent Technologies, Santa Clara, CA). The values were divided by the mean of the basal recording of the control cells. The basal glucose concentrations were 0 or 0.5 mmol/L for palmitate treatment in EndoC-β¹H cells and 3 mmol/L for dispersed human islets and β-like cells. For pharmacological inhibition of STAT3, cells were treated for 1 h with a STAT3 inhibitor (STAT3 Inhibitor V, Stat three inhibitory compound [Stattic], 10 µmol/L; MilliporeSigma) followed by a 1-h incubation with 0.5 mmol/L sodium palmitate (Sigma-Aldrich) in the presence of 1% weight for volume fatty acid–free BSA (Sigma-Aldrich) based on our previous studies (29,30,37).

The results are presented as the mean ± SEM. Student t test was used for comparisons between two groups. Differences among groups were assessed by two-way ANOVA or repeated-measures ANOVA. Bonferroni post hoc tests were performed to identify differences among means. Statistical analyses were assessed using Prism 8 software (GraphPad Software, La Jolla, CA). Sample size was predetermined based on the variability observed in prior experiments and on preliminary data. All experiments requiring the use of animals or animals to derive cells were subject to randomization based on litter. Differences were regarded as statistically significant if *P < 0.05, **P < 0.01, or ***P < 0.001 (see legends to figures).

The data set generated during the sequencing procedure is deposited in the Gene Expression Omnibus (GEO) database (accession no. GSE162837) and available from the corresponding author upon reasonable request.

As a first step toward elucidating the role of STAT3 in the regulation of glucose homeostasis in β-cells, we determined whether obesity and/or T2D affects the expression of STAT3 in islets/β-cells in human pancreas. Samples from 15 organ donors were processed and stained for STAT3 expression (Supplementary Table 1). We observed a significant increase in the cytoplasmic but not nuclear expression of STAT3 in the islets of severely obese and diabetic individuals (Fig. 1A and Supplementary Fig. 1). Importantly, a correlation of STAT3 intensity staining in the islets/subjects and BMI was observed in the pancreas sections (Fig. 1A) (Pearson r = 0.659, r² = 0.4345, P = 0.0274). Consistent with the human data and development of obesity, mice after 14 weeks of high-fat feeding (23.5% fat; 46% energy from fat) have strong expression of STAT3 in islets and insulin-producing β-cells (Fig. 1B and Supplementary Fig. 2 and 3). The localization of human and rodent STAT3 is cytoplasmic in the islet cells, suggesting a noncanonical (not nuclear) activity. Interestingly, STAT3 strongly colocalizes with COX IV in the mitochondria of the islet cells from obese mice (Fig. 1B and Supplementary Fig. 4). For identification of the normal expression and distribution of STAT3 within different islet cell types, C57BL/6 islets were dissociated into a single-cell suspension; stained for insulin, glucagon, and STAT3; and analyzed by flow cytometry. As shown in Fig. 1C, insulin expression was only observed in β-cells—not in non-β cells. While STAT3 was detected in both cell populations, its expression was higher in β-cells than in non-β cells.

To address the role of STAT3 in obesity in vivo, we generated a tamoxifen-inducible β-cell–specific STAT3-deficient mouse model by crossing the MIP-Cre/ERT mouse model with STAT3^lox/lox mice. The signal from tamoxifen-treated homozygous STAT3^lox/loxCre mice was barely detectable in β-cells (Fig. 1D), demonstrating effective STAT3 deletion. By contrast, heterozygous STAT3^lox/+Cre mice had a 50% reduction in STAT3 expression in β-cells compared with control littermates. Moreover, the expression of STAT3 in non-β cells was unaffected (Fig. 1D).

Next, we tested whether complete deletion of STAT3 in β-cells affects glucose homeostasis in chow-fed lean mice. STAT3^lox/loxCre and STAT3^lox/lox mice were followed for 27 weeks and body weight was measured; no differences were observed between littermates (Supplementary Fig. 5A). Moreover, oral and intraperitoneal tolerance tests did not show differences between the groups (Fig. 1E). These results are in line with previous studies deleting STAT3 with the Pdx1-Cre knockout mouse model (25). Collectively, our findings indicate that STAT3 was partially (in STAT3^lox/+Cre mice) or fully (in STAT3^lox/loxCre mice) deleted in β-cells after tamoxifen treatment but not in non-β cells. The STAT3 gene deletion in β-cells does not affect glucose homeostasis in lean mice fed a chow diet.

For assessment of the impact of β-cell STAT3 deficiency on the development of obesity, 10-week-old male tamoxifen-treated STAT3^lox/loxCre and control mice were fed a high-fat diet for 12 weeks and their body weight and metabolic status were assessed. We previously showed that the high-fat diet with 46% of the total energy from lipids induces obesity, steatosis, and insulin resistance (28,29). High-fat diet–fed STAT3^lox/loxCre mice had weight gain similar to that of STAT3^lox/lox littermates (Fig. 2A and Supplementary Fig. 5B). There were no differences in body composition (Fig. 2B), fat pad, liver, muscle or pancreas weights (Fig. 2C), or glucose levels after fasting or feeding (Supplementary Fig. 5C). Strikingly, STAT3^lox/loxCre mice displayed significant delay in glucose clearance after administration in the intraperitoneal glucose tolerance test (Fig. 2D). No significant difference was observed in the rate of blood glucose reduction between STAT3^lox/loxCre and STAT3^lox/lox control following an insulin injection (Fig. 2E), suggesting that glucose intolerance observed in STAT3^lox/loxCre mice is not caused by an alteration in insulin sensitivity. There are several caveats with β-cell–specific Cre mouse lines, and results should be interpreted carefully with these models (23,39,40). Importantly, in the MIPCre/ERT mouse model glucose homeostasis is not affected with a chow or high-fat diet compared with C57BL/6 control mice (40). We confirmed that the Cre transgene alone does not affect body/tissue weights or glucose levels in our high-fat diet condition (Supplementary Fig. 5D–G). Importantly, glucose tolerance tests were not different between high-fat diet–fed C57BL/6, MIPCre/ERT, and STAT3^lox/lox control mice (Supplementary Fig. 5H and I). Thus, our data demonstrate that STAT3 deficiency is responsible for glucose intolerance in obese STAT3^lox/loxCre mice.

Next, we repeated the high-fat feeding in heterozygous β-cell STAT3 deficiency to assess whether partial (50%) (Fig. 1D) deletion of STAT3 in β-cells is sufficient to disturb glucose homeostasis in obesity. STAT3^lox/+Cre male mice also had weight gain similar to that of littermate controls when fed a high-fat diet (Supplementary Fig. 6A), and there were no differences in fat pad, liver, muscle, or pancreas weights (Supplementary Fig. 6B); glucose levels after fasting or feeding (Supplementary Fig. 6C); or insulin sensitivity (Supplementary Fig. 6D). Moreover, no differences were observed across metabolic parameters (oxygen consumption, respiratory exchange rate, energy expenditure, food intake, and ambulatory activity) (Supplementary Fig. 6E). This finding further supports that STAT3 deletion using the β-cell–specific MIP-Cre/ERT transgene did not affect food intake in STAT3^lox/+Cre mice. In line with obese STAT3^lox/loxCre, partial STAT3 deficiency induced glucose intolerance in obese STAT3^lox/+Cre mice, determined by intraperitoneal glucose tolerance test analysis (Fig. 2F). Administration of glucose orally via gavage is physiological and can induce incretin response, which can potentiate glucose-mediated insulin response (41). Like with the intraperitoneal glucose tolerance test, STAT3^lox/+Cre mice displayed a significant delay in glucose clearance after oral glucose administration compared with that of the STAT3^lox/+ control (Fig. 2G).

To determine whether glucose intolerance was caused by abnormal insulin expression, we performed immunohistochemical analysis on pancreatic sections from high-fat diet–fed STAT3^lox/+ and STAT3^lox/+Cre mice. As shown in Fig. 2H, no difference was observed in percentage insulin-positive area, which suggests that insulin expression was not altered by STAT3 deletion. The murine islet morphology was also normal, as α-cells (glucagon positive) were evenly distributed on the periphery (Fig. 2H). This demonstrated that glucose intolerance in STAT3^lox/+Cre mice fed a high-fat diet is not caused by an alteration in β-cell mass. STAT3^lox/+Cre and STAT3^lox/+ mice received an oral gavage of glucose, and serum insulin concentration was determined by ELISA using serum collected after gavage (Fig. 2I). The serum insulin concentration in STAT3^lox/+Cre mice was also 2.1-fold lower than in STAT3^lox/+ mice 15 min after glucose administration. These results suggest that disruption of STAT3 expression in β-cells impairs early-phase glucose-stimulated insulin secretion, which is responsible for glucose intolerance.

Glucose homeostasis is strictly dependent on adequate and efficient insulin secretory response from pancreatic β-cells. To gain insight into the mechanism of glucose intolerance induced by STAT3 deficiency, we took advantage of siRNA-mediated knockdown of STAT3 in the EndoC-β¹H human β-cell line. Transfection of STAT3 siRNAs leads to 90% knockdown of the protein (Fig. 3A). We performed RNA sequencing (RNA-Seq) analysis of the STAT3 knocked down and control cells in three independent experiments, obtaining high reproducibility between samples (Fig. 3B). To refine the potential list of STAT3 target genes, we analyzed the top 25 upregulated and downregulated genes and pathways affected (Fig. 3C–E). Glucose-6-phosphatase catalytic subunit 2 (G6PC2) and activating transcription factor 6 (ATF6) expression was reduced in STAT3 knockdown EndoC-β¹H cells (Fig. 3C and D). Deficient G6PC2 activity enhances β-cell function (42)—opposite the in vivo phenotype observed in the STAT3^lox/+Cre obese mice. ATF6 plays a key role in the unfolded protein response, which can trigger insulin production in β-cells (43), and regulates β-cell viability (44). Moreover, profiling of the data set using the Ensemble Gene Set Enrichment Analyses (EGSEA) method for gene set testing allowed identification of several modulated pathways by STAT3 knockdown including the unfolded protein response, hypoxia, apoptosis, and pancreatic β-cell development (Fig. 3E). However, we did not observe differences in endoplasmic reticulum stress markers (Supplementary Fig. 7A), Ca²⁺ content in the endoplasmic reticulum (Supplementary Fig. 7B), or cell viability (Supplementary Fig. 7C) after STAT3 silencing. Insulin expression was significantly downregulated in the RNA-Seq analysis and confirmed by quantitative PCR (qPCR) (Fig. 3F), but no differences were observed in insulin levels in STAT3^lox/+Cre obese mice. Differentiated β-cell markers MAF-A, NEUROG3, and PDX1 were not affected in STAT3 knockdown EndoC-β¹H cells (Fig. 3F). Interestingly, the hypoxia pathway was upregulated (Fig. 3E) and most of the mitochondrial-encoded genes were downregulated in STAT3-deficient cells (Fig. 3G). We ran qPCR on mitochondrial genes in EndoC-β¹H cells and confirmed decreased expression of several mitochondrial encoded genes after STAT3 knockdown (Fig. 4A). We performed subcellular protein extraction of EndoC-β¹H cells and observed colocalization of STAT3 and the proteins COX IV and cytochrome c in the mitochondrial enriched fractions (Fig. 4B). To mimic the increased saturated free fatty acid levels in obesity, we have treated insulin-producing MIN6 cells with palmitate and observed mitochondrial phosphorylation of STAT3 in the Ser727 residue (Supplementary Fig. 7D), required for mitochondrial translocation (14). Importantly, FACS-purified β-cells from high-fat diet–fed STAT3^lox/+Cre and control mice showed no difference in the nuclear-encoded UCP2 mRNA levels but reduced expression of mitochondrial encoded genes mt-ND4, mt-ND5, and mt-Cytb (Fig. 4C), further supporting an in vivo role of STAT3 in mitochondrial gene expression in β-cells.

Perturbation of mitochondrial DNA gene expression can lead to dysfunction of the electron transport chain and oxidative phosphorylation system (45). OCR measurements were carried out to determine the mitochondrial function and activity with two different siRNAs targeting STAT3 (but not STAT1) to discard any off-target effect (Fig. 4D). Knockdown of STAT3 induced a significant reduction in basal and maximal respiration compared with control cells (Fig. 4D). We normalized the results by protein content and confirmed that STAT3 knockdown does not affect cell viability (Supplementary Fig. 7C). Moreover, protein expression of the nuclear-encoded mitochondrial transcription factors TFAM and TFB1M (Supplementary Fig. 7E) and electron transport chain COX IV and cytochrome c was not affected by STAT3 knockdown (Supplementary Fig. 7F). STAT1 shares 50% homology with STAT3, similar Tyr/Ser phosphorylation, and DNA binding sequences (46). However, knockdown of STAT1 did not affect mitochondrial activity (Supplementary Fig. 8), indicating nonredundant roles of STAT proteins in mitochondria function. We studied mitochondrial size and architecture by transmission electron microscopy (TEM). STAT3 deficiency did not significantly affect mitochondrial size and numbers (Fig. 4E) but increased mitochondrial cristae destruction, matrix swelling, autophagosome structures, and mitophagy, events characteristic of mitochondrial remodeling, 6.6%, n = 20 cells, in siRNA control and 18%, n = 18 cells, in siRNA STAT3 (Fig. 4E and Supplementary Fig. 9). Importantly, silencing of STAT3 in dispersed human islets, albeit with variable knockdown efficacy and β-cell numbers, decreased mitochondrial activity in five of six different preparations (Fig. 4F and Supplementary Fig. 10). Stattic is a small nonpeptide STAT3 blocker designed to target its phosphorylation and dimerization (47). It was demonstrated that Stattic reduced electron transport chain by inhibiting the mitochondrial activity of STAT3 in different cells (48–50). We have treated EndoC-β¹H cells with palmitate and Stattic. As shown in Fig. 4G, the enhanced ATP production induced by palmitate and maximal respiratory capacity is impaired with pharmacological inhibition of STAT3. Overall, our results demonstrate that STAT3 regulates β-cell function, at least in part, by mitochondrial gene expression and activity.

Our study suggests that STAT3 is dispensable for β-cell function in normal physiology but that deficient STAT3 activity predisposes β-cells to mitochondrial dysfunction under metabolic stress. An activating STAT3 mutation, K392R, was identified as a cause of permanent neonatal diabetes, leading to premature differentiation of pancreatic progenitors (16,17). However, the expression and cellular localization of STAT3 in iPSC differentiation into β-like cells remain unknown.

We used human iPSC derived from dermal fibroblasts (33) to study the cellular localization and activity of STAT3 in the differentiation transition to β-like cells. The iPSC were differentiated into pancreatic progenitors with use of a seven-stage protocol based on previously published reports (17,34–36) (Fig. 5A). OCT4, SOX17, FOXA2, PDX1, NKX6.1, SOX9, and NEUROG3 have been used as stage transition markers (Fig. 5A–C and Supplementary Fig. 11A) to achieve constant efficiency of differentiation into insulin, glucagon, and double-positive cells (Fig. 5D and Supplementary Fig. 11B). STAT3 expression was significantly inhibited in the definitive endoderm and primitive gut tube stages (Fig. 5E and F and Supplementary Fig. 11C). STAT3 is a target of PDX1 (51), and consistent with PDX1 nuclear activation STAT3 is reexpressed in pancreatic endoderm (stage 4) and maintained in the remaining differentiation stages (stages 5–7). STAT3 can activate NEUROG3 expression in the endocrine precursor (stage 5) as previously suggested (17). Indeed, we observed canonical nuclear activity of STAT3 in this stage (Supplementary Fig. 11C). The reprogramming of fibroblasts to a pluripotent state induces a metabolic shift from oxidative phosphorylation to glycolysis allowing to the proliferating cells a supply of amino acids and protection against oxidative stress and helps to generate NADPH (52). Differentiated β-like cells switch back to oxidative phosphorylation (Fig. 5G), albeit with low/no response to glucose and deficient insulin secretion, characteristic of an immature state. Interestingly, STAT3 canonical nuclear activity in iPSC and endocrine precursor stage changes to a cytoplasmic localization in differentiated insulin-producing β-like cells (Fig. 5G). Our results postulate STAT3 inactivation as a novel marker in the definitive endoderm transition and support a noncanonical localization of STAT3 in differentiated insulin-producing β-cells (Figs. 1A and B, 4B, and 5H).

Glucose homeostasis is strictly dependent on adequate and efficient insulin secretory response from pancreatic β-cells. In individuals with diabetes, the compensatory β-cell response no longer overcomes insulin resistance, leading to a deterioration of β-cell mass and decline in insulin secretory function from exhaustion (5). Mitochondria dysfunction is characteristic of islets from T2D organ donors, showing reduced expression of subunits of respiratory chain complexes, a dramatic reduction in basal and glucose-induced respiration, and morphological changes (53,54). We observed that cytoplasmic STAT3 expression is increased in islets from obese and T2D organ donors and its deficiency in β-cells accelerates defective insulin secretion in obesity.

We generated a novel β-cell–specific conditional Stat3 deletion mouse model to eliminate confounding variables from previous mouse models, to prevent the effect of Stat3 deletion in development, and to elucidate the role of STAT3 in β-cells in obesity. Compared with STAT3^lox/+Cre control littermates, heterozygous and homozygous STAT3-deficient mice display no significant differences in body or tissue weights, blood glucose after fasting or feeding, or metabolic activity. Strikingly, both high-fat diet–fed STAT3^lox/loxCre and STAT3^lox/+Cre mice are glucose intolerant compared with controls, suggesting that STAT3 plays an important role in β-cell function and glucose homeostasis in obesity. Impaired glucose tolerance was not due to changes in insulin sensitivity or β-cell mass. Our results indicate that STAT3^lox/+Cre mice have impaired early-phase glucose-stimulated insulin secretion. This finding is in agreement with findings in human patients with prediabetes, in whom insulin secretion is impaired upon glucose stimulation (55). We focused our in vivo work in male mice, but the role of STAT3 in β-cell function should also be clarified in the development of obesity in females (56). In keeping with this, a higher number of human samples will be required to compare STAT3 expression and localization in β-cell/islets from different subgroups, stratified by sex and age.

In comparisons with the STAT3^lox/+Cre control, the mitochondrial-encoded genes mt-ND4, mt-Nd5, and mt-Cytb were significantly reduced in β-cells of STAT3^lox/+Cre mice after 12 weeks of high-fat diet. Consistent with this result, STAT3 localizes in the mitochondria of human EndoC-β¹H cells, and with STAT3 deficiency there was reduced mitochondrial encoded genes and mitochondrial activity, which can result in mitochondria remodeling manifested by cristae destruction and matrix swelling, as observed with TEM. Interestingly, swollen mitochondria are characteristic of T2D patients (53,57). Our findings reinforce STAT3 as a regulator of mitochondrial gene expression in pancreatic β-cells. In support of our data, several studies in other cell types showed that STAT3 deletion leads to a reduction of mitochondrial gene expression (12,58) and respiratory activity of the electron transport chain (12,14,58–60). It is now clear that nonnuclear STAT3 has the propensity to regulate mitochondrial gene expression, electron transport chain activity, and oxidative phosphorylation (10,61).

Mitochondrial oxidation is an early β-cell adaptation event to glucose and metabolic stress (62). The mechanism by which STAT3 is activated in obesity and modulates mitochondrial expression is presently unknown. This may be mediated by direct interaction of STAT3 with mitochondrial DNA and the mitochondrial transcription factor TFAM, which is observed in keratinocytes (13). However, we did not observe changes in TFAM after STAT3 knockdown in EndoC-β¹H cells. Additional experiments are necessary to determine whether a STAT3-TFAM protein relationship exists in β-cells. Our existing data support that STAT3 is a positive regulator of mitochondrial gene expression in β-cells. It is reasonable to propose that β-cells with functional STAT3, which retain normal STAT3 expression and activity, can prevent or delay the effect of obesity on mitochondrial function and preserve β-cell insulin secretory activity. If the expression of the mitochondrial genes and electron transport chain activity is suppressed in STAT3-deficient β-cells, the ATP-to-ADP ratio necessary for enhanced demand of insulin secretion is reduced, leading to impaired β-cell adaptation in obesity. It will be relevant to use the same cohorts of STAT3 mice with partial and full β-cell deficiency at different time points of obesity development to study STAT3-dependent mitochondrial gene expression associated with β-cell function. This is beyond the scope of the current work. It should also be noted that several nuclear-encoded genes, including ATF6, G6PC2, and INSULIN, have also been modulated by STAT3 deficiency in β-cells. Thus, we cannot completely discard a canonical effect of STAT3 regulating β-cell gene transcription and contributing to the deficient insulin secretion in obesity. Interestingly, recently it was shown that the metabolic role of STAT3 within the mitochondria alters the production of α-ketoglutarate, which is essential for epigenetic regulation of gene expression (63).

STAT3 signaling has been demonstrated to be essential for human pancreas development and endocrine differentiation (17). We observed almost complete STAT3 inactivation in the first stages of stem cells reprogrammed into β-like cells, in line with published data (64). STAT3 is a target of PDX1 (51), binds to the Neurog3 promoter (65), and is involved in the activation of NEUROG3 in differentiating human embryonic stem cells (66). This corresponds with STAT3 reactivation and nuclear localization in the pancreatic endoderm stage. iPSC use oxidative phosphorylation once differentiated in β-like cells and STAT3 localization changes from nuclear canonical to cytoplasmic noncanonical in stage 7. Moreover, studies performed in stems cells have emphasized a switch from glycolytic energy production to oxidative phosphorylation during the differentiation process (67). The mitochondrial “activation” is common to differentiation of many cell types, among them hematopoietic (68), neuronal (69), and muscle stem cells (70), but the exact transition mechanism is currently poorly understood. Interestingly, the LIF/STAT3 axis can potentiate mitochondrial respiration in stem cells (12) and STAT3/fam3a promote muscle stem differentiation and mitochondrial respiration (70), further suggesting a role of STAT3 in oxidative phosphorylation transition during differentiation into β-like cells. Future studies will determine whether enhanced mitochondrial STAT3 activity can improve mitochondrial function, maturation, and insulin secretion in β-like cells, a desirable outcome on the road to β-cell replacement in diabetes.

Overall, our study supports a key role of STAT3 in β-cells at least in part by regulating mitochondrial gene expression and activity. Mitochondrial dysfunction in β-cells is critical for T2D development, and currently available treatments cannot prevent disease progression. Thus, modulated subcellular localization of STAT3 in β-cells may potentially offer a new pharmacological target. If the induction of mitochondrial dysfunction in β-cells can be prevented by enhancing noncanonical STAT3 activity, it may be possible to halt disease progression and improve glucose homeostasis in patients.

Acknowledgments. The authors thank Madalina Popa, Javier Negueruela, Michela Miani, Dakai Yang, and personnel from the ULB-Center for Diabetes Research and Helen E. Thomas and Thomas C. Brodnicki (St Vincent’s Institute of Medical Research) for experimental and technical support. The authors also thank Piero Marchetti and the Pancreatic Islet Laboratory in Pisa (University of Pisa) for human islets.

Funding. This work was supported by a Fonds National de la Recherche Scientifique (FNRS)–Mandat d’impulsion scientifique (MIS) grant (33650793), European Research Council Consolidator grant Protein Tyrosine Phosphatases in METAbolic diseases (METAPTPs) (GA817940), and a JDRF Career Development Award (CDA-2019-758-A-N). The St Vincent’s Institute of Medical Research receives support from the Operational Infrastructure Support Scheme of the Government of Victoria. E.N.G. is a Research Associate of the FNRS, Brussels, Belgium.

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

Author Contributions. A.S., L.P., V.V., B.E., S.A.L., B.V., E.M., M.V., F.M.M.P., and D.P-M. researched data, contributed to result analysis, and reviewed and edited the manuscript. D.J.G. provided samples, contributed to data analysis and discussion, and reviewed and edited the manuscript. E.N.G. researched data, contributed to discussion, designed experiments, and reviewed, edited, and wrote the manuscript. E.N.G. 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.