STAT3 Sensitizes Insulin Signaling by Negatively Regulating Glycogen Synthase Kinase-3β

Details of generating the transthyretin (Ttr) transgenic mice are described in a previous study by Moh et al. (16). Briefly, vector Ttr1ExV3, which contains the upstream transcriptional regulation region, the first exon and intron, and a partial second exon of the Ttr gene (20), was digested with StuI. A fragment containing the open reading frame of the Cre gene was recovered from the mammalian expression vector pBS185 (GIBCO BRL, Life Technologies, Grand Island, NY) by digestion with XhoI and MluI and inserted into the StuI site of Ttr1ExV3 to get the final construct pTtr-Cre. The plasmid pTtr-Cre was digested with HindIII to generate a transgene fragment, which was isolated by agarose gel electrophoresis. The purified transgene fragment was injected into fertilized eggs as described previously (21).

Generation of mice with a conditional Stat3 allele has been described previously (22). Exons 18–20, which contained the SH2 domain of Stat3, were flanked by two loxP sites (Stat3^f/f). Two Ttr-Cre transgenic mouse strains (10-3 and 19-1) were crossed with Stat3^f/f mice. The genotype was determined by PCR as described previously (22). Mice for experiments were produced by the breeding of male Stat3^f/+;Cre and female Stat3^f/+ in most cases or Stat3^f/f in some occasions. Male L-Stat3^−/− and littermate Stat3^f/+ control mice at the age of 8–12 weeks were used for experiments.

Hepatocytes were isolated by in situ perfusion as previously described (23) with minor modifications. Briefly, the liver was perfused with 0.5 mmol/l EGTA and 0.15 g/l collagenase via the inferior cava. After perfusion, the dissociated liver was minced and suspended in Hanks’ balanced salt solution. Cells were centrifuged twice at 500g, and hepatocytes were separated by centrifugation through a Percoll layer. Cells were counted and plated onto 10-cm dishes with 2 × 10⁶ cells. Three cell lines were established by immortalization of hepatocytes from L-Stat3^−/− and littermate wild-type mice with simian virus 40 T antigen. The cell lines were screened for STAT3 and hepatocyte markers. Line 2F4 of STAT3-null hepatocytes was used in most experiments, and results were confirmed with the other two lines. All hepatocytes were cultured in 4.5 g/l Dulbecco's modified Eagle's medium (DMEM; high glucose) enriched with 10% fetal bovine serum (FBS). After serum starvation for 4 h, 100 nmol/l insulin was added to 80% confluent hepatocytes. The cells were scalped and collected at the indicated time point after being washed with PBS three times.

The plasmid constructs of Stat3 were gifts from Dr. JE Darnell of Rockefeller University. The Stat3-CA construct was made by substituting cysteine residues for A661 and N663 of murine Stat3. This renders the STAT3-CA molecule capable of dimerizing without a phosphate on tyrosine 705 (24). Stat3-S727A was made by substituting alanine for serine 727 of STAT3. For transfection, cell lines of hepatocytes were transfected with Lipofectamine 2000 according to manufacture's instruction (Invitrogen, Carlsbad, CA). Cells were cultured for 48–72 h until harvesting.

Homogenized mouse liver (100 mg) and cultured hepatocytes were lysed in a whole-cell extraction buffer (400 mmol/l KCl, 10 mmol/l NaHPO₄,1 mmol/l EDTA, 1 mmol/l dithiothreitol, 10% [w/v] glycerol, 1 μg/ml aprotinin, 1 μg/ml leupeptin, 1 μg/ml pepstatin, 1 mmol/l phenylmethylsulfonyl fluoride, 5 mmol/l NaF, and 1 mmol/l Na₃VO₄). After centrifugation at 14,000 rpm, the supernatant was collected, and protein concentration was quantified with the BCA kit (Pierce, Rockford, IL). Ten micrograms of proteins were separated by SDS-polyacrylamide gel electrophoresis and then transferred to an Immuno-Blot polyvinylidene difluoride membrane (Bio-Rad, Hercules, CA). After blocking with 5% nonfat milk in a washing buffer (10 mmol/l Tris-HCl, 150 mmol/l NaCl, 1 mmol/l EDTA, and 0.1% [w/v] Triton X-100, pH 8.0), the membranes were soaked in the indicated antibodies, followed by soaking in anti-rabbit or anti-mouse IgG coupled with horseradish peroxidase and visualized using SuperSignal Chemiluminescent Substrate (Pierce).

Cell line of STAT3-null hepatocytes (line 2F4) was used for the chromatin immunoprecipitation assay. Different forms of Stat3 (wild-type Stat3, Stat3CA, and Stat3 S727A) were reintroduced to the hepatocyte by transfection. The cells were cultured in DMEM with 10% FBS for 72 h. After incubation with 100 nmol/l insulin for 2 h, the cells were collected for immunoprecipitation assay with standard procedures as described previously (25). Specific antibodies against STAT3 provided by Dr. Zhong Zhong (26) were used for the immunoprecipitation. Normal rabbit IgG served as a negative control. For PCR, 1 μl from 50 μl precipitated DNA fragments was analyzed with 21–25 cycles of amplification using primers P1 (NT_039624, forward, 5′-GATCCTTCCGCCGCTTCCCT-3′) and P2 (NT_039624, backward, 5′-AAGACTTCGTTCTCTTGGCT-3′) directed against the mouse GSK-3β regulation region or a cytoplasmic β-actin coding region of the 1st exon as a negative control (NT_039324, forward, 5′-GTCAGAAGGACTCCTATGTG-3′; backward, 5′-GGCGTGGCTGAGAAGCTG-3′).

All of the different stat3 constructs were subcloned into pcDNA3 plasmid. A total of 400 ng DNA was transfected into STAT3-null hepatocytes (line 2F4). Mouse Gsk-3β gene promoter region from nucleotides −584 to +1,459 was amplified from mouse genomic DNA by PCR using Gsk-3β promoter sequence–specific primers with XhoI and HindIII adaptation sites in 5′ and 3′ primers, respectively (sense, 5′-CCGCTCGAGTCGGCTCGGAAAGCAGATG-3′; antisense, 5′-CCCAAGCTTCGTTCTCTTGGCTTTTCACTC-3′), and subcloned into PGL3-Basic luciferase reporter plasmid (Promega). For transient transfection, 80% confluent cells in a 24-well plate were transfected with Gsk-3β promoter-luciferase construct and pCMV β-galactosidase or in combination with wild-type or different Stat3 expression plasmids, or control vector using Lipofectamine 2000 (Invitrogen). Luciferase activity was assayed 24 h after transfection using the Dual-light luciferase kit (Applied Bioscience). Values were normalized with β-galactosidase activity expressed from pCMV β-galactosidase plasmid (Clontech).

RNA was extracted from the liver or hepatocytes using a TRIzol reagent (Invitrogen Life Technologies, Carlsbad, CA) following the manufacturer's instructions. Total RNA (2 μg) was used to prepare cDNA (iScript cDNA synthesis kit; Bio-Rad) using random hexamers. cDNA was amplified by PCR with a iQ SYBR Green Supermix kit (Bio-Rad). The primers are GSK-3β (NM_019827, forward, 5′-CACTCAAGAACTGTCAAGTAAC-3′, and reverse, 5′-CATTAGTATCTGAGGCTGCTG-3′) located at the two joints of the last three exons; PEPCK (forward, GTGTCATCCGCAAGCTGAAG, and backward, CTTTCGATCCTGGCCACATC) (27); glucose-6-phosphatose (G6Pase) (forward, CTGCAAGGGAGAACTCAGCAA, and backward, GAGGACCAAGGAAGCCACAAT) (27); and peroxisome proliferator–activated receptor-γ coactivator (PGC-1α) (forward, 5′-ATACCGCAAAGAGCACGAGAAG-3′, and backward, 5′-CTCAAGAGCAGCGAAAGCGTCACAG-3′) (18). PCR conditions are as follows: one cycle of 94°C for 2 min; 35 cycles of 94°C for 30 s, 55°C for 30 s, and 72°C for 45 s; followed by one cycle of 72°C for 10 min. β-Actin cDNA was amplified as an internal control with the same condition (X03672, forward, 5′-CCAGAGCAAGAGAGGCATCC-3′, and backward, 5′-AGGTCTTTACGGATGTCAACG-3′; length 704 bp). All data were means of fold change of triplicate analysis and normalized with those of β-actin. The fold change represents the change of transcript levels in the liver/cells relative to Stat3^f/+ at time point 0 in a time course and to STAT3-null, nontreatment control in transfection experiments after normalization with β-actin levels.

For hematoxylin-eosin (H-E), the sections were prepared and sent to Yale Pathological Services for H-E staining. For immunofluorescent staining, the sections were incubated with rabbit antibody against phosphorylated STAT3 (tyrosine 705 or serine 727) overnight at 4°C. The staining was visualized with a Rodamin-labeled donkey anti-rabbit second antibody (Jackson Immunoresearch Laboratories, West Grove, PA) under a fluorescent microscope. The nuclei were counter-stained with 4′,6-diamidino-2-phenylindole (DAPI). Nuclear STAT3 was quantitated by the product of mean color level multiplied by the area in pixel. The ratio of intensity of each group to that of L-Stat3^−/− at point 0 was calculated to normalize different stains and given as fold change in the figure.

Data are presented as mean ± SD. The difference of two genotype groups was analyzed by a Student's t test. Data in a time course experiment or over two groups were tested by one-way ANOVA. P < 0.05 was considered to be significant. Additional information is available in an online appendix at http://dx.doi.org/10.2337/db06-1582.

We have generated three lines of transgenic mice with C57BL/6 background expressing the transgene that encodes the Cre recombinase enzyme under the control of the Ttr promoter (Ttr-Cre) in the liver. As shown in Fig. 1A, PCR amplified a 300-bp DNA band of the transgene encoding the Cre recombinase. We previously created mice carrying a Stat3 allele with exons 18–20 flanked by loxP sites (Stat3^flox/flox mice with genotyping shown in Fig. 1B) (22) and bred them with a line of Ttr-Cre transgenic mice. A deletion of exons 19 and 20, which encode the SH2 domain of Stat3, results in a conditional knockout of STAT3 protein in the liver (L-Stat3^−/−) and pancreatic islet (Fig. 1C). The basal level of STAT3 in the muscle was lower than other organs like the liver, and densitometry analysis did not show significant differences among the different forms of STAT3 in the muscle (Fig. 1C).

The L-Stat3^−/− (Ttr Cre⁺;Stat3^f/f) mice were born at a Mendelian rate and were otherwise normal (Table 1) except for impaired insulin signaling (see below). No significant morphological changes or inflammatory infiltration were noted in the livers of L-Stat3^−/− mice (Fig. 1E). Young L-Stat3^−/− mice did not develop obesity (Fig. 1E; Table 1), and even at 18 months of age (32.78 ± 1.98 vs. 32.66 ± 1.82 g body wt, Stat3^f/+ vs. L-Stat3^−/− male mice, n = 6 for each group), there were no significant differences in fat mass between the two groups. No STAT3 deletion was found in the brain or fat tissue (not shown). The abundances of STAT3 protein in major metabolic organs did not differ between Stat3^f/f and Stat3^f/+ (Fig. 1D), and Stat3^f/+ mice were used as controls for experiments.

Fasting blood glucose results (Table 1) in the L-Stat3^−/− mice (line 10-3) were not significantly different from those of their littermates. However, glucose levels at 60 and 90 min in L-Stat3^−/− mice were significantly higher than in their Stat3^f/+ littermates after 2 g/kg glucose was injected intraperitoneally (Fig. 2A). We delivered exogenous insulin (1 IU/kg i.p.) to examine the response in the L-Stat3^−/− and littermate Stat3^f/+ mice. Glucose levels were significantly higher at 60 and 90 min in L-Stat3^−/− mice than in their Stat3^f/+ littermates after insulin injection (Fig. 2B). The other line (19-1) of mice with STAT3 specifically knocked out in the liver had similar results (not shown). The glucose phenotypes of Ttr-Cre⁺;Stat3^+/+ mice were similar to those of wild-type Stat3 and Stat3^f/+ mice (not shown).

Nuclear translocation of serine phosphorylated STAT3 (serine 727) was observed in vivo (Fig. 3A) and in vitro (Fig. 3B) after insulin treatment. Levels of tyrosine phosphorylated STAT3 (tyrosine 705) were also higher in the nuclei of Stat3^f/+ hepatocytes than those of L-Stat3^−/− cells but did not show a significant increase after insulin treatment (Fig. 3). Bioinfomatic analysis revealed a relatively conserved STAT binding site in the gene promoter region of the Gsk-3β gene (Fig. 4A). Chromatin immunoprecipitation assay with a STAT3 antibody (26) and PCR amplification with primers covering the putative binding site revealed binding of STAT3 to the Gsk-3β regulation region (Fig. 4C). The binding was weakened by mutating serine 727 (Fig. 4C). The effect of STAT3 on Gsk-3β gene promoter activity was examined by luciferase assay. As shown in Fig. 4D, constitutively active STAT3 or STAT3 in the presence of insulin significantly inhibited Gsk-3β gene promoter activity, whereas serine 727 mutation abolished the inhibition (Fig. 4D). Quantitative RT-PCR analysis of mRNA confirmed that STAT3 CA or STAT3 in the presence of insulin suppressed Gsk-3β gene expression and that serine 727 might be important for the suppression (Fig. 4E). The suppression could be observed in the presence of cycloheximide (Fig. 4E).

Time course of GSK-3β mRNA in L-Stat3^−/− and control mice after insulin injection is shown in Fig. 5A. Levels of GSK-3β mRNA were significantly reduced in Stat3^f/+ mice at 30 and 60 min after insulin injection when compared with those of L-Stat3^−/− mice. The reduction of the mRNA levels in Stat3^f/+ mice was stronger in feeding than in fasting condition (Fig. 5A).

GSK-3β protein tended to increase in L-Stat3^−/− mice at randomly feeding (time point 0), but the difference was not statistically significant (P = 0.067) (Fig. 5B). Insulin injection induced a transient increase in both protein and phosphorylation levels of GSK-3β and a decrease after 30 min of insulin injection in the Stat3^f/+ control mice (Fig. 5B). In contrast, the GSK-3β level continuously increased after 30 min of insulin injection in the L-Stat3^−/− mice (Fig. 5B). Similar results was present in fasting with lesser differences between the L-Stat3^−/− and Stat3^f/+ mice than in feeding (Supplemental Fig. 2). After 30 min of insulin treatment, GSK-3β levels also increased in the Stat3^−/− hepatocytes isolated from the liver of L-Stat3^−/− mice (Fig. 5C). Similar pattern of GSK-3β in the L-Stat3^−/− mice was present in glucose loading (Fig. 6A). Consistently, there was a significant reduction in glycogen synthase activity (Fig. 6B), but liver glycogen remained unchanged in the L-Stat3^−/− mice after glucose loading (Fig. 6C).

It was clear that STAT3 deficiency in the liver led to an increase in GSK-3β expression. The next question we asked was whether or not the increased GSK-3β was responsible for the impaired insulin response in L-Stat3^−/− mice. We treated L-Stat3^−/− mice with lithium chloride, a GSK-3β blocker (28), before insulin injection. To avoid generalized effects of lithium and to obtain optimal results of GSK-3β inhibition, different doses and duration were tried. With the protocols given in Fig. 7, there were no significant differences in blood insulin level (Fig. 7B), mass of adipose tissue, and glycogen of skeletal muscle (not shown) between the L-Stat3^−/− and control mice after lithium treatment. The impaired insulin signaling and glucose tolerance in L-Stat3^−/− mice could be rescued by the GSK-3β inhibitor (Fig. 7B and C). Additional results are available in the online appendix.

In the present work, we generated a strain of mice with Stat3 gene being deleted in the liver and pancreatic islets (Fig. 1C). Insulin secretion in the L-Stat3^−/− mice was not significantly affected by the islet deletion (Fig. 2A) as observed in Pdx-1 Cre mice with islet deletion of STAT3 (29), whereas early studies reported impaired insulin secretion in β-cell STAT3 deletion driven by RIP-Cre (30,31). Interestingly, the RIP-cre/Stat3 phenotype results from hypothalamic, not islet, deletion of STAT3. The L-Stat3^−/− mice demonstrated no obesity (Table 1) but significant insulin resistance (Fig. 2). These data clarify that STAT3 deficiency in the liver leads to primary insulin resistance. It is noteworthy that the serum leptin level tended to increase in the L-Stat3^−/− mice (Table 1). This is consistent with previous findings in which insulin resistance in the liver increases the circulating levels of the extracellular domain of the leptin receptor, stabilizing leptin (32). It has been reported that STAT3 could be activated by insulin with some discrepancies as to whether tyrosine 705 or serine 727 or both were activated/phosphorylated (4,5,19). In the present study, the nuclear translocation of STAT3 was observed (Fig. 3A and B), implying that the activated STAT3 may be involved in a transcriptional event. We did a bioinformatic search, and Gsk-3β emerged as one candidate target gene with relatively conserved STAT3 binding site in the gene promoter region (Fig. 4A). Chromatin immunoprecipitation assay revealed that STAT3 was in the transcription complex associated with the STAT3 binding site (Fig. 4C). Luciferase assay demonstrated that constitutively active STAT3, which carries activated/phosphorylated serine 727 (24), inhibited the promoter activity of the Gsk-3β gene (Fig. 4D). The suppression was stronger in feeding than in fasting conditions (Fig. 5A).

The data were surprising because STAT3 is generally believed as a transcriptional activator. In some cases, the action of STAT3 seems to be ascribed to the induction of a set of important target genes, but in others, it may be acting as a repressor (e.g., in the thymic epithelium) (17,33) or as a signaling adaptor without a transcriptional function (e.g., activation of Akt in neurons) (17,34). In the nuclear translocation assay, low level of tyrosine phosphorylated STAT3 was observed in the nuclei but was independent of insulin (Fig. 3), which was consistent with a previous study (18). On the other hand, an early nuclear translocation of serine phosphorylated STAT3 was observed (Fig. 3). The data suggest a role of serine phosphorylated STAT3 in insulin action, which might switch the DNA-bound tyrosine phosphorylated STAT3 to an inhibition complex in the Gsk-3β promoter. Our data showing interaction of STAT3 with Gsk-3β regulator and decreased mRNA and protein levels of GSK-3β without significant changes in the serine 9 phosphorylation, which inactivates Gsk-3β, clearly indicate that STAT3 suppresses Gsk-3β expression (Figs. 4E and 5A). Because Gsk-3β is a constitutively active gene and because expression of Gsk-3β does not require an enhancer, STAT3 plays a relatively logical role in regulating Gsk-3β as a suppressor. In addition, STAT3 may recruit cofactors to determine whether to activate or repress a downstream gene (18,35). Although STAT3 sometimes plays different roles in biological events, the ultimate biochemical effect of activated STAT3 seems to lie in its ability to increase or decrease the transcriptional activity of previously quiescent or active genes. It would be interesting to further investigate whether Stat3 may also regulate GSK-3β mRNA and protein stability.

It is interesting to note that there was a very physiological regulation of GSK-3β in the Stat3^f/+ mice in different conditions: fasting versus feeding. In feeding, because hepatocytes need insulin signaling to process glucose, mRNA levels of GSK-3β were reduced (compared with those in fasting; Fig. 5A) to allow insulin to induce glycogen synthesis. In fasting, because hepatocytes have to release glucose into the circulation, mRNA expression levels of GSK-3β were increased (compared with those in feeding; Fig. 5A) to block insulin to do the opposite work. The L-Stat3^−/− mice lost this control and showed a monophasic response to insulin (Fig. 5A). Protein levels of GSK-3β showed similar patterns in feeding and fasting in the L-Stat3^−/− and control mice (Fig. 5B; Supplemental Fig. 2). These data indicate that STAT3 is critical in regulating glucose and, more generally, in energy metabolism signaled by insulin.

Insulin induced a rapid increase in GSK-3β protein levels in both L-Stat3^−/− and Stat3^f/+ mice 30 min after insulin injection (Fig. 5B). Because there were no significant differences between L-Stat3^−/− and Stat3^f/+ mice within this short period of time, this change could not be related to STAT3. One could hardly interpret the rapid increase as a result of protein synthesis; it is most likely due to a protein-protein interaction and decrease in GSK-3β degradation for the early change (≤30 min) of insulin stimulation.

The loss of STAT3 in the liver resulted in an increase in the GSK-3β protein level (Fig. 5B), whereas the phosphorylation of GSK-3β remained unchanged (Fig. 5B). As a result, there was a net increase in unphosphorylated or active form of GSK-3β in L-Stat3^−/− mice. Consistently, glycogen synthase activity decreased in the L-Stat3^−/− mice (Fig. 6C). On the other hand, because a number of factors contribute to glycogen level in the liver, not just GSK-3β/glycogen synthase, it is not surprising that the glycogen level was not changed in the L-Stat3^−/− mice (Fig. 6B). It is also possible that the low level of glycogen synthase activity (Fig. 6B) could not drive the difference between the Stat3^f/− and L-Stat3^−/− mice. Thus, although both insulin intolerance (Fig. 2) and decreased glycogen synthase activity (Fig. 6B) were present, the main phenotype of the L-Stat3^−/− mice seems to be the impaired insulin action (Fig. 2) probably due to GSK-3β suppression of insulin signaling molecules like IRS-1 (4,5).

We hypothesized that the increased GSK-3β was the cause of impaired insulin signaling in the L-Stat3^−/− mice because GSK-3β is a key negative regulator for insulin. We chose lithium chloride, the most commonly used GSK-3β inhibitor (28,36), to test our hypothesis. The data clearly show that the GSK-3β inhibitor rescued the impaired insulin effects on glucose in the L-Stat3^−/− mice (Fig. 7), indicating that the increased GSK-3β was responsible for the impaired insulin signaling in the L-Stat3^−/− mice. Additional discussion is available in the online appendix.

In conclusion, the present study provides evidence, we believe for the first time, that STAT3 sensitizes insulin signaling by down-regulating the expression of GSK-3β, a negative regulator of insulin. Because the STAT3 binding site in the GSK-3β is highly conserved in mice and humans (Fig. 4A), we speculate that a similar function should exist in humans and that mutations in this regulatory region should have a fundamental effect on insulin signaling. Thus, a novel genetic approach for fighting human type 2 diabetes could be developed based on our work showing that STAT3 negatively regulates GSK-3β.

X.-Y.F. has received a Career Development Award from the National Institutes of Health (NIH), a pilot project from Yale Liver Center (DK-34989), and grants from the NIH (RO1-AI-34522 and RO1-AR-44906).

We deeply thank Dr. Peter Roach and Dr. Jose Irimia-Dominguez for generating the data of liver glycogen and glycogen synthase and for their valuable expertise and discussion. We also thank Lan Ji for excellent technical assistance.