Diabetes in Mice With Selective Impairment of Insulin Action in Glut4-Expressing Tissues

A DNA construct encoding GLUT4-Cre was engineered by cloning a 2.4-kb human GLUT4 promoter fragment (23) into pSP73 vector, containing Cre cDNA preceded by a β-globin intron. The purified linearized DNA fragment was microinjected into fertilized eggs from C57BL/6 × FVB mice. We obtained two founders (535 and 546) that were characterized for transgene transmission and recombination. The transgene showed autosomal transmission in line 546 and X-linked transmission in line 535. Both lines underwent germ line recombination when transmitted through the dam, but not when transmitted through the sire. When transmitted through the sire, both transgenes were liable to stochastic embryonic activation, leading to generation of animals with varying degrees of chimerism. An example is shown in Supplementary Fig. 1, demonstrating chimeric recombination in liver of GIRKO mice. These events were monitored by genotyping of tail DNA, and chimeric mice were excluded from further analyses. When mice were killed, extensive genotyping was performed to ascertain that mice had undergone recombination only in bona fide Glut4-expressing tissues.

Line 546 transgenics were intercrossed with Insr^flox/flox mice (7), and the resulting progeny was intercrossed to yield Insr^flox/flox controls and Insr^{flox/flox, GLUT4-cre} (GIRKO) mice that were used for all subsequent phenotypic analyses. Animals were maintained on a mixed background derived from 129/Sv, C57BL/6, and FVB. GLUT4-Cre was genotyped using primers 5′-TGGGTCCCATCGGGCCCTTAATT-3′ and 5′-TGGCGATCCCTGAACATGTCCAT-3′, amplifying a 350-bp product in the presence of the transgene. Rosa26/Gfp and Rosa26/Rfp mice were obtained from Jackson Laboratories. The Columbia University Institutional Animal Care and Utilization Committee and the Yale University Animal Care and Use Committee approved all animal procedures.

Blood glucose was measured by the One-Touch Ultra meter (LifeScan Inc., Milpitas, CA). Insulin and leptin were measured by enzyme-linked immunosorbent assay, and glucagon was measured by radioimmunoassay (Linco Research Inc., St. Charles, MO). Plasma free fatty acids (FFA) and cholesterol were measured by NEFA-HR and Cholesterol-E test reagents (Wako Chemicals, Richmond, VA). Triglycerides were measured by serum triglyceride determination kit (Sigma-Aldrich, St. Louis, MO). β-Hydroxyl butyrate was measured by a colorimetric assay (Pointe Scientific, Canton, MI). Measurements of hepatic glycogen and triglycerides were performed as previously described (24). Body composition was determined using Bruker Minispec NMR (Bruker Optics, Billerica, MA) (16).

Male mice, 12 to 14 weeks old, were fasted overnight, anesthetized, and injected with saline or insulin (5U) into the inferior vena cava. Three minutes after injection, muscle, adipose, liver, heart, and brain tissues were rapidly removed and flash-frozen in liquid nitrogen (25). Detergent extracts were prepared in buffer containing 20 mmol/L Tris (pH 7.6), 150 mmol/L NaCl, 1 mmol/L dithiothreitol, 10 mmol/L EGTA, 1% NP40, 2.5 mmol/L Na₄P₂O₇, 1 mmol/L NaVO₃, 1 mmol/L β-glycerophosphate, and protease inhibitor cocktail (Roche, Indianapolis, IN). Protein concentration of extracts was determined by bicinchoninic acid assay (Pierce, Rockford, IL). Equal amounts of protein (50–100 μg) were resolved on SDS-PAGE and transferred onto nitrocellulose membranes (Schleicher & Schuell, Keene, NH). Membranes were probed with antibodies against phospho-Akt (Ser473), total Akt (Cell Signaling, Danvers, MA), InsRβ, and β-actin (Santa Cruz Biotechnology, Santa Cruz, CA).

Thirteen- to fifteen-week-old normoglycemic male GIRKO mice (n = 10) and control littermates (n = 9) were studied. Euglycemic-hyperinsulinemic clamps were performed in conscious, minimally stressed mice at the Yale Mouse Metabolic Phenotyping Center (http://mouse.yale.edu). Briefly, after anesthesia, an indwelling catheter was placed into the right internal jugular vein. Studies were carried out 4 to 5 days after surgery in overnight fasted mice. Tail tethered using tape and basal glucose turnover were measured using a primed-continuous infusion of [3-³H]glucose for 2 h before the start of clamp. At the start of the clamp, mice received a primed constant infusion of human insulin (15 pmol/kg/min) and continued to receive [3-³H]glucose (0.1 mCi/min, PerkinElmer, Waltham, MA) for the duration of the clamp. Blood samples were collected from tail. Blood glucose was determined at 10- to 20-min intervals. Glucose (20% solution) was infused at a variable rate as required to maintain euglycemia. 2-Deoxy-D-[1-¹⁴C]glucose (2-[¹⁴C]DG) was administered as a bolus (10 μCi) 75 min after the start of the clamp. Blood samples (20 μL) were taken at −5, 80, 85, 90, 100, 110, and 120 min of clamp for the measurement of plasma [³H]glucose, tritiated water, or 2-[¹⁴C]DG concentrations. Additional blood samples (20 μL) were collected before and at the end of clamp for the measurement of plasma insulin and FFA. At the end of the clamp, mice were killed, and liver, gastrocnemius, epididymal fat, and cerebral cortex were flash-frozen in liquid nitrogen. Tissue samples were stored at –80°C for analysis (16).

Mice were individually housed in a TSE LabMaster system (TSE Systems Inc., Chesterfield, MO) and acclimated to the respiratory chambers for 24–36 h before measurements were taken. Data on gas exchanges, locomotor activity, and food intake were collected every 14 min for 4–5 days (26).

Adipose tissue and liver were fixed in 10% formaldehyde for hematoxylin–eosin and periodic acid-Schiff staining. Frozen liver embedded in OCT was used for Oil red O staining. Pancreata from 12-week-old male mice were fixed in 10% formaldehyde overnight and embedded in paraffin; consecutive sections 5 μm in thickness were mounted on slides. Sections were stained with antibodies against insulin, glucagon (Sigma-Aldrich San Antonio, TX), Pdx1 (27), and Glut2 (Alpha Diagnostics, Owings Mills, MD) (28).

All data represent means ± SEM. Datasets were analyzed for statistical significance with one-way ANOVA followed by post hoc Bonferroni test using SPSS software (SPSS Inc., Chicago, IL).

We generated transgenic mice expressing Cre off a 2.4-kb GLUT4 promoter fragment that has been shown to confer tissue-specific expression of a reporter construct (23). We characterized recombination patterns in two GLUT4-Cre founders (lines 535 and 546) by crossing them with either ROSA26-Gfp or ROSA26-Rfp transgenics. Line 546 faithfully recapitulated Glut4 expression, yielding recombination in skeletal muscle, epididymal and subcutaneous white adipose tissue (WAT) and brown adipose tissue (BAT), heart, and kidney (glomeruli and loop of Henle). We also detected recombination at sites of Glut4 expression heretofore unrecognized, such as ovarian corpora lutea, myometrium, nuclear layers of the retina, and acinar cells in the exocrine pancreas and salivary gland (Supplementary Fig. 1A and B). Line 535 showed more limited recombination in skeletal muscle and brain (not shown).

In view of literature on Glut4 expression in the CNS (17, 20,21,29,30), we analyzed patterns of GLUT4-Cre–mediated recombination in this organ using GLUT4-Cre × Rosa26-Gfp mice. Green fluorescent protein (GFP) immunohistochemistry in brain sections of the resulting progeny revealed Cre-mediated recombination in scattered subsets of cells throughout the CNS, including cortical pyramidal neurons, cerebellar Purkinje-like cells, olfactory bulb, hippocampus, and hypothalamus (Supplementary Fig. 2). Coimmunostaining with either of two different anti-Glut4 antibodies and GFP in the hypothalamus of GLUT4-Cre × Rosa26-Gfp mice showed that >90% of GFP-positive cells (Fig. 1A–D) were also Glut4-positive (Fig. 1E and F), demonstrating that the transgene recapitulates faithfully the expression of the endogenous gene. To characterize Glut4-positive cells, we performed coimmunostaining with GFP and either the neuronal marker HuC/D or the glial marker Gfap throughout the CNS. The former, but not the latter, colocalized with Glut4 (Fig. 2). These data indicate that GLUT4-Cre induces recombination in a subset of neurons characterized by expression of Glut4 protein. We obtained similar data in the CNS of line 535 (data not shown).

To generate a model of insulin resistance in Glut4-cre–expressing tissues, we intercrossed transgenic line 546 with Insr^flox/flox mice (7). The resulting Insr^{flox/flox,GLUT4-cre} mice (Glut4-cre–driven InsR knockout, or GIRKO) showed InsR ablation in skeletal muscle, WAT, and BAT. A measurable decrease of InsR also could be detected in the hippocampus and mediobasal hypothalamus, but not in the cortex and cerebellum (Fig. 3A and Supplementary Fig. 5), consistent with the fact that most cells in the former two sites are Glut4-positive, whereas ∼10% in the latter two sites are Glut4-positive (Supplementary Fig. 2). Accordingly, insulin-stimulated Akt phosphorylation was absent in skeletal muscle, including soleus and gastrocnemius, WAT (subcutaneous and epididymal), and BAT, whereas it was preserved in liver of GIRKO mice (Fig. 3B). We obtained identical results when we measured phospho-GSK-3 levels in response to insulin, indicating that Akt activity is impaired (Supplementary Fig. 3).

GIRKO mice were born in the expected Mendelian ratios and showed no gross abnormality. A subset in both sexes developed hyperglycemia in the fed state (Fig. 4A and B) and hyperinsulinemia in the fed and fasted states (Supplementary Table 1, Fig. 4C). The prevalence of diabetes (defined as glycemia > mean + 2 SD) increased from 20% at 5 weeks to 25% at 12 weeks and 46% at 24 weeks of age in male mice, and hovered at approximately 10% in 5- to 12-week-old female mice. Intraperitoneal glucose tolerance and insulin tolerance tests in 12- to 13-week-old mice revealed modest glucose intolerance in male GIRKO mice and insulin resistance in female GIRKO mice, respectively (Supplementary Fig. 4).

To probe tissue-specific glucose metabolism and insulin sensitivity, we performed euglycemic-hyperinsulinemic clamps on 13- to 15-week-old nondiabetic male GIRKO and control littermates. Under basal conditions, blood glucose was slightly higher in GIRKO mice than in controls (119 vs. 87 mg/dL) (Fig. 5A). During the hyperinsulinemic phase of the clamp, we infused insulin to yield a similar increase in plasma insulin between the two groups (Supplementary Table 2) and achieved similar blood glucose levels at steady state. The rate of glucose infusion required to maintain euglycemia was decreased by 52% in GIRKO mice (Fig. 5B), consistent with impaired glucose disposal. Basal hepatic glucose production (HGP) was increased by 32% in GIRKO mice (Fig. 5C), and hyperinsulinemia was unable to suppress it (Fig. 5D and E), indicating marked hepatic insulin resistance. Moreover, the rate of glucose disappearance was reduced by 28% and glycolysis by 33% in GIRKO mice compared with controls (Fig. 5F and G), whereas glycogen synthesis trended downward (Fig. 5H). Consistent with impaired insulin-dependent glucose uptake, 2-deoxy-glucose incorporation in gastrocnemius muscle was reduced by 34% in GIRKO mice (Fig. 5I). In contrast, 2-deoxy-glucose uptake in the cerebral cortex was not significantly altered (Fig. 5J), consistent with the insulin-independent mode of this process (mediated by Glut1 and Glut3) (22).

Given that hepatic insulin signaling (Fig. 3) and expression of genes encoding gluconeogenic enzymes are normal in GIRKO mice (data not shown), the increase of HGP is likely secondary to changes in humoral/neural factors or increased gluconeogenic precursors, such as FFA. Fasting FFA showed a nonsignificant decrease in GIRKO mice. Clamp hyperinsulinemia suppressed FFA by 43% in controls, but had no effect in GIRKO mice, consistent with reduced sensitivity of adipose tissue to insulin’s antilipolytic actions (Supplementary Table 2). Fasting plasma glucagon and β-hydroxyl butyrate levels were similar between the two groups of mice (Supplementary Table 1), indicating that glucagon tone and fatty acid oxidation are normal in GIRKO mice. Hepatic triglyceride and glycogen content tended to be increased in overnight-fasted GIRKO mice compared with wild-type mice, but these changes did not reach statistical significance (Supplementary Table 1). The data suggest that failure to suppress FFA contributes to the increased HGP observed in hyperinsulinemic conditions, but cannot explain the increased basal HGP.

Morphologic analysis of BAT of 6-month-old GIRKO mice showed increased lipid droplet size and striking disruption of the multilocular structure that were more pronounced in diabetic GIRKO mice (Fig. 6A). Epididymal WAT of diabetic and nondiabetic GIRKO mice had a heterogeneous appearance compared with controls (Fig. 6B), reminiscent of the phenotype of fat-specific InsR knockouts (8). We also examined hepatic lipid and glycogen content by Oil red O and periodic acid-Schiff staining, respectively, and found hepatosteatosis and glycogen depletion only in diabetic GIRKO mice, whereas livers of nondiabetic GIRKO mice appeared normal (Fig. 6C and D).

We examined pancreatic islet morphology in 12-week-old male animals. Euglycemic GIRKO mice showed β-cell hyperplasia but maintained normal islet architecture (Fig. 7A and B). In contrast, diabetic GIRKO mice showed a leopard patterned loss of insulin immunoreactivity (Fig. 7C) (31), accompanied by loss of Pdx1 immunoreactivity (Fig. 7D and E), ectopic α-cells in the islet core (Fig. 7F–H), and loss of membrane localization of Glut2 (Fig. 7J and K).

To determine whether InsR ablation in Glut4-expressing tissues affected energy homeostasis, we assessed respiratory exchanges by indirect calorimetry, coupled with food intake and activity determinations. GIRKO mice had normal body weight as weanlings, but showed persistent growth retardation thereafter, and weight of 16-week-old male mice was reduced by 19% compared with controls (Fig. 8A). The reduction in body weight affected equally lean and fat mass and was associated with normal plasma leptin levels (Supplementary Table 1). Diabetic GIRKO had significantly lower body weights than control or nondiabetic GIRKO mice (data not shown).

Body weight in the cohort that underwent indirect calorimetry determinations showed the same distribution as in the population as a whole (Fig. 8B). Daily food intake was similar between controls and nondiabetic GIRKO mice (Fig. 8C), but the latter displayed altered circadian feeding patterns, consuming only 26% of total food during the light phase compared with 34% in control mice (Fig. 8D). In addition, oxygen consumption and locomotion were largely unperturbed in nondiabetic GIRKO mice (Fig. 8E–G). In diabetic GIRKO mice, food intake was increased when normalized by body weight (Fig. 8C), but total intake per animal was similar to that of control and nondiabetic GIRKO mice (Fig. 8D). In addition, although oxygen consumption in diabetic GIRKO mice showed an increase when normalized by body weight (Fig. 8E), total energy expenditure per animal tended to decrease (Fig. 8F), consistent with reduced lean body mass and unchanged locomotion (Fig. 8G). Furthermore, respiratory exchange ratio and carbohydrate use in diabetic GIRKO mice failed to show circadian variations, with no discernible increase after food ingestion during the dark phase (Fig. 8H and I). Taken together, these data indicate that InsR ablation in Glut4-expressing neurons does not alter energy intake or expenditure, but may affect circadian feeding patterns. The changes seen in hyperglycemic GIRKO mice, including metabolic substrate inflexibility, altered rhythmicity in substrate use, relative hyperphagia, and weight loss from impaired nutrient storage, are presumably secondary to diabetes.

The key finding of this study is that selective impairment of insulin action in Glut4-expressing tissues, in the absence of primary changes to liver and pancreatic β-cell function, results in a murine model of T2D that faithfully recapitulates the human disease. Further, the study identifies a heretofore ill-characterized entity that we propose to name “Glut4 neuron,” as a site of insulin action. Our findings provide a solution to the conundrum on the site of onset of insulin resistance, expand our understanding of the neuroanatomy and cellular physiology of insulin action in the CNS, and integrate the latter into the broader picture of diabetes pathogenesis.

Identifying the sites of onset of insulin resistance is necessary if we are to develop a rational approach to treating this condition—a cornerstone of diabetes therapy (32). Single InsR knockouts in individual tissues have been more notable for what they failed to cause than for what they did cause (33), as was the combined InsR ablation in muscle/fat (6). Why is the InsR knockout in Glut4-expressing tissues able to cause diabetes? Probably because, in addition to the impairment of peripheral glucose disposal caused by InsR knockout in muscle and fat, and unlike other models of impaired InsR signaling in muscle (34) or fat (8), GIRKO mice develop profound hepatic insulin resistance and β-cell dysfunction. We favor the interpretation that the former is due to impaired InsR signaling in Glut4 neurons (16,35,36), with a possible contributory role of other factors, such as FFA. Given that most hypothalamic neurons appear to be Glut4-positive, it is likely that InsR signaling in these neurons contributes to insulin’s demonstrated ability to control HGP via the CNS (16,35,36). Our data do not imply that InsR ablation in Glut4 neurons is the sole cause of diabetes. Rather, they suggest that the combination of muscle, fat, and CNS insulin resistance underlies the transition from euglycemic insulin resistance to diabetes.

It is also intriguing that InsR ablation in Glut4 neurons did not affect total energy intake or expenditure, but rather shifted circadian patterns of feeding and substrate use. A large body of literature indicates that disruption of the circadian clock in the CNS or peripheral tissues can lead to insulin resistance, defects in insulin secretion, and other manifestations of the metabolic syndrome (37,38). Further studies will be necessary to determine whether alterations in the circadian rhythm contribute to the cause of diabetes in GIRKO mice.

Profound abnormalities of β-cell mass represent another distinguishing feature of GIRKO mice from other models of peripheral insulin resistance due to alterations of InsR signaling (25,39). We speculate that the β-cell abnormality is also secondary to impaired CNS control of endocrine pancreatic function. But we cannot rule that other sites of InsR ablation in GIRKO mice (e.g., vessel wall, exocrine pancreas, or other as yet undiscovered locations) contribute to this process. Nonetheless, the combined presence of cell-autonomous mechanisms of insulin resistance (in muscle/fat) and cell-nonautonomous mechanisms (in liver and β-cell) in GIRKO mice indicates that Glut4-expressing tissues play a direct role as metabolic targets for nutrient disposal and storage, and a “sensing” role through which they are able to affect liver and endocrine pancreatic function.

The presence of Glut4 in the CNS has long been a puzzle, partly because of the presence of additional glucose transporters, partly to lingering controversies as to its precise localization (17,20,21,29,30,40–43), and partly to limited ability to probe its function in this organ. Our study conclusively demonstrates that neurons, not glia, are the site of Glut4 expression. Their distribution and morphology bespeak heterogeneous functions, potentially including locomotor and learning behaviors. In the hypothalamus, Glut4 neurons could reasonably partake in glucose sensing, counterregulation, and food intake (44). Their functions are likely to extend beyond the functions of Glut4 itself. But it is interesting that Glut4 ablation in the CNS impairs glucose sensing and glucose tolerance (44). Brain glucose sensing is important for counterregulation in response to hypoglycemia and is altered by obesity and antecedent hypoglycemia (45). Although most, if not all, glucose-sensing neurons express both InsR and Glut4 (46), the general view is that, under euglycemic conditions, glucose sensing occurs independently of insulin, probably through Glut3 and glucokinase (45). It is possible, however, that InsR signaling leads to Glut4 translocation, thus increasing local glucose concentrations at discrete sites within the cell body or neuronal processes to levels that allow for glucokinase activation, ATP generation, and closure of K_ATP channels (47), with increased firing of glucose-excited neurons (48).

Insulin resistance remains a large unmet medical need, and clinically effective agents that partly restore insulin sensitivity remain limited to metformin and pioglitazone. Because insulin signaling is pleiotropic and entails different biological outcomes in different cell types, the rational design of insulin sensitizers must rely on detailed knowledge of the cellular sites and biochemical processes of insulin resistance. GIRKO mice should prove widely useful as a monogenic model of insulin-resistant diabetes, devoid of the confounders generally associated with other commonly used models.

This research was supported by the Berrie Fellowship Award (to H.V.L.); National Institutes of Health Grants DK-58282, DK-57539 (to D.A.), and DK-40936 (to G.I.S.); the Columbia University Diabetes and Endocrinology Research Center (DK-63608); and the Mouse Metabolic Phenotyping Center (U24 DK-076169).

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

H.V.L. designed and conducted experiments, analyzed data, and wrote the article; H.R. designed and conducted experiments, analyzed data, and wrote the article; V.T.S. and H.-Y.L. conducted glucose clamp experiments; T.Y.L. conducted experiments; G.I.S. designed experiments, analyzed data, and wrote the article; and D.A. designed experiments, analyzed data, and wrote the article.

The authors thank C.R. Kahn (Joslin Diabetes Center) for the gift of the floxed InsR mice, K. Aizawa, A. Flete, Y. Dam, V. Lin, and Q. Xu (Columbia University) for technical assistance, and members of the Accili laboratory for helpful discussions and critical reading of the article.