Divergent Regulation of Energy Expenditure and Hepatic Glucose Production by Insulin Receptor in Agouti-Related Protein and POMC Neurons

Insr^−/− (14), Ttr-INSR transgenic, βac/INSR knock-in (15), Foxo1^flox/flox (16), Agrp-Cre (17), and Pomc-Cre (18) mice have been previously described. Insr^+/−::Ttr-INSR::βac/INSR mice were intercrossed with Agrp-Cre or Pomc-Cre mice to generate Insr^−/−::Ttr-INSR::βac/INSR::Agrp-Cre (Agrp-KI) and Insr^−/−::Ttr-INSR::βac/INSR::Pomc-Cre (Pomc-KI) mice. Wild-type littermates were used as controls. Genotyping was performed as previously described (15,17,–19). Embryonic recombination of the βac/INSR locus was detected by PCR from tail DNA of an ∼200-bp product using primers 5′-GGTTTTCCTTTGAAAAACACG-3′ and 5′-CTTAATCGCCTTGCAGCACAT-3′. Foxo1^ΔPOMC (Foxo1^flox/flox::Pomc-Cre) mice were obtained by mating Foxo1^flox/flox mice with Pomc-Cre mice (18). All animal procedures have been approved by the Columbia University Institutional Animal Care and Utilization Committee.

Metabolites were measured as described (20). Free fatty acids and triglycerides were measured with NEFA-HR test reagents (Wako Chemicals, Richmond, VA) and serum triglyceride determination kit (Sigma, St. Louis, MO). Insulin and leptin were measured by enzyme-linked immunosorbent assay; adiponectin, resistin, tissue plasminogen activator inhibitor 1, interleukin-6 (IL-6), and tumor necrosis factor-α by Luminex (Linco Research, St. Charles, MO); IGF-1 (Alpco Diagnostics, Salem, NH), corticosterone (MP Biomedicals, Solon, OH), glucagon, and C-peptide by radioimmunoassay. Body composition was determined using Bruker Minispec NMR (Bruker Optics, The Woodlands, TX).

Body weight was measured weekly. Food intake was measured biweekly in 6- to 12-week-old mice using manual feeding hoppers. For refeeding studies, after acclimation mice were fasted overnight. Feeding hoppers were placed in the cages 2 h after the start of the light phase, and food intake was measured 6 h after feeding hoppers were placed.

For glucagon measurement, fasting was started 2 h after the onset of the light phase. After a 6-h fast, blood was drawn from the tail vein. One week later, mice were again fasted for 6 h, followed by an intraperitoneal insulin injection (0.75 units/kg). Blood glucose was measured immediately before insulin injection and at 30 and 60 min after the injection. Mice used for analyses of the stress response were familiarized with gentle handling for 6–8 weeks prior to the experiment. For determination of basal plasma corticosterone, blood was drawn from the tail 1–2 h after the start of the light phase.

Four groups of 14- to 16-week-old normoglycemic male mice were studied: wild type (n = 8), L1 (n = 13), Agrp-KI (n = 8), and Pomc-KI (n = 7). We performed euglycemic-hyperinsulinemic clamps in conscious, catheterized mice as described (10).

Indirect calorimetry was performed using a TSE LabMaster system (TSE Systems). Individually housed mice were acclimated to the respiratory chambers for 24–36 h prior to the experiment, and gas exchanges, locomotor activity, and food intake were measured every 14 min for 3–6 days.

We extracted total RNA using RNeasy Mini Kit and RNase-Free DNase Set (Qiagen, Valencia, CA). Quantitative PCRs were performed in triplicate using a DNA Engine Opticon 2 System (Bio-Rad, Hercules, CA) and DyNAmo HS SYBR green Q-PCR kit (New England Biolabs, Ipswich, MA). Primer sequences are available upon request. Relative mRNA levels were normalized to 36B4 (for liver and hypothalamus) or Rps3 (for white adipose tissue).

We prepared detergent extract from liver and muscle as described (20) and probed the membranes with antibodies against phospho-Akt (Ser473), total Akt, phospho-Stat3 (Tyr705) (Cell Signaling, Danvers, MA), total Stat3, InsRβ (C-19), and β-actin (Santa Cruz Biotechnology, Santa Cruz, CA).

We perfused mice transcardially with PBS followed by formalin, fixed and dissected brains, equilibrated them in 30% sucrose, and froze them in OCT (Sakura, Torrance, CA). We used 30-μm–thick coronal sections for green fluorescent protein immunostaining on free-floating sections with a polyclonal rabbit antiserum (Molecular Probes/Invitrogen, Carlsbad, CA).

Phosphatidylinositol 3,4,5-trisphosphate (PIP3) immunostaining was performed as previously described (21) using a fluorescein isothiocyanate–conjugated antibody (catalog no. Z-G345; Echelon Biosciences, Salt Lake City, UT), a horseradish peroxidase–conjugated fluorescein isothiocyanate antibody, and TSA fluorescein tyramide reagent pack (PerkinElmer, Waltham, MA). For quantitative analysis of PIP3 levels, three coronal sections of each mouse at approximately Bregma −1.3 mm, −1.4 mm, and −1.5 mm were examined. A total of 700–850 cells per mouse were examined as previously described (6). Results were quantified and expressed as percentage of ARC cells showing moderate to high PIP3 levels.

The hypothalamus was dissected from 15- to 17-week-old male mice, and cut in 50-μm vibratome sections that were processed and analyzed as described (22). Owing to significant nonhomogeneity of variance among groups, data were analyzed by one-way nonparametric ANOVA (Kruskal-Wallis), and Mann-Whitney U test was used to determine significance (P < 0.05).

We analyzed data with one-way ANOVA followed by post hoc Bonferroni test using SPSS software.

To investigate the contribution of InsR signaling in AgRP and/or POMC neurons to insulin sensitivity and energy homeostasis, we reactivated InsR expression specifically in either of these neuron populations in L1 mice using a conditional locus knock-in approach (supplementary Fig. 1A and B, available in an online appendix at http://diabetes.diabetesjournals.org/cgi/content/full/db09-1303/DC1) (15). L1 mice in which Cre-mediated recombination resulted in InsR expression in AgRP or POMC neurons are denoted as Agrp knock-in (Agrp-KI) or Pomc knock-in (Pomc-KI), respectively. Cre-mediated recombination was assessed by intercrossing Cre transgenic lines with Rosa26-Gfp mice and revealed the expected expression patterns in ARC AgRP and POMC neurons (supplementary Fig. 1C and D), without affecting InsR levels in peripheral tissues (supplementary Fig. 1E and F).

To determine the effects of AgRP neuron– or POMC neuron–specific InsR reconstitution, we measured insulin's ability to activate the PI 3-kinase pathway in wild-type, L1, Agrp-KI, and Pomc-KI mice (21). In wild-type mice, we detected PIP3 immunoreactivity in 15.7% of ARC cells in the fasted state and in 22.5% of cells after insulin stimulation (Fig. 1 A and B) (P < 0.05). Insulin failed to stimulate PIP3 formation in ARC of L1 mice (19.6 fasted vs. 18.9% after insulin, respectively), but was effective in Pomc-KI mice (18.8 vs. 25.8%) and Agrp-KI (17.9 vs. 24.0%) mice, consistent with restoration of insulin/PI 3-kinase signaling in defined subpopulations of ARC neurons.

The expectation of the current study was that, by reconstituting InsR function in AgRP neurons or POMC neurons of L1 mice, insulin suppression of HGP would be restored. Consistent with previous reports (10,15), L1 mice showed hyperinsulinemia in the fasted and fed states, and 13% became diabetic by 12 weeks of age (Fig. 2 A and B). Agrp-KI mice showed significant reductions in both fasted and fed insulin levels compared with L1, and only 9% were diabetic at 12 weeks. Pomc-KI mice showed insulin levels similar to L1 mice, but the frequency of diabetes nearly doubled to 23%. To circumvent potential secondary effects of chronic hyperglycemia on metabolism and energy homeostasis, we excluded diabetic L1, Agrp-KI, and Pomc-KI animals from further analyses. In intraperitoneal glucose tolerance tests, nondiabetic L1, Agrp-KI, and Pomc-KI mice displayed modest glucose intolerance compared with wild type (supplementary Fig. 2A), whereas insulin tolerance tests did not reveal differences among the four genotypes (supplementary Fig. 2B).

To accurately probe hepatic insulin sensitivity, we performed euglycemic hyperinsulinemic clamps in nondiabetic mice (Fig. 2,C–F). Consistent with the results of glucose tolerance tests, rates of whole-body glucose disappearance in L1, Agrp-KI, and Pomc-KI mice were 10, 24, and 28% lower than wild type (Fig. 2,E), respectively, indicating modest reductions in tissue glucose uptake. As previously described (10), the glucose infusion rate (GIR) required to maintain euglycemia was reduced by 36% and HGP was increased fourfold in L1 mice compared with wild type (Fig. 2 D and F), consistent with hepatic insulin resistance. HGP was restored to normal in Agrp-KI mice, indicating that InsR in AgRP neurons is sufficient for insulin inhibition of HGP. In contrast, Pomc-KI mice exhibited a further 64% reduction in GIR and a 78% increase in HGP compared with L1, indicating a marked increase in hepatic insulin resistance.

This unexpected result prompted us to investigate the mechanism whereby InsR reconstitution in POMC neurons leads to a deterioration of hepatic insulin sensitivity. Changes in hepatic insulin signaling and counterregulatory responses failed to explain the phenotype (supplementary Figs. 3–6). We then explored two mutually nonexclusive hypotheses: 1) altered regulation of genes involved in hypothalamic glucose sensing; and 2) impaired neuropeptide regulation with attendant abnormalities of energy balance.

Hypothalamic K_ATP channels link membrane excitability with cellular metabolism, and are regulated by glucose (23,24), insulin (25,26), and leptin (27). Cells in the mediobasal hypothalamus, including ∼75% of POMC neurons, express the K_ATP channel sulfonylurea receptor 1 (SUR1) and Kir6.2 subunits (encoded by Abcc8 and Kcnj11, respectively) (8,21,28). Activation of hypothalamic K_ATP channels lowers HGP, whereas SUR1 deficiency curtails insulin suppression of HGP (8). To determine whether abnormalities of K_ATP channel expression contribute to the increased HGP in Pomc-KI mice, we examined hypothalamic Abcc8 and Kcnj11 levels in fasted and in clamped mice. We detected a significant 43% reduction of Abcc8 expression in clamped Pomc-KI mice compared with L1 and Agrp-KI mice (Fig. 3,A), but no significant difference in Kcnj11 among the four genotypes (Fig. 3 B). The decrease of Abcc8 is consistent with a reduction in insulin's ability to decrease neuronal firing, which in turn may contribute to increased HGP. (Fasted levels were also lower in Pomc-KI versus wild-type mice, but the difference failed to reach statistical significance.)

To further demonstrate that the observed reduction in Abcc8 expression was linked to InsR signaling in POMC neurons, we measured hypothalamic Abcc8 mRNA in mice lacking the insulin-regulated transcription factor forkhead box class O (FoxO1) in POMC neurons (Foxo1^ΔPOMC). The prediction was that Foxo1^ΔPOMC mice should phenocopy the reduced Abcc8 levels of Pomc-KI mice. Indeed, we found a 53% decrease of Abcc8 in Foxo1^ΔPOMC mice (Fig. 3 C), consistent with our hypothesis that InsR signaling decreases Abcc8 expression in POMC neurons.

Owing to the complex genotype of Pomc-KI mice, we could not obtain primary neuronal cultures from newborns to perform electrophysiology. We thus performed a histomorphometric electron microscopy study of synaptic contacts, comparing wild-type, L1, and Pomc-KI mice. We found a significant decrease in the total number of synapses on POMC neurons of L1 and Pomc-KI mice that was entirely accounted for by decreased symmetric (putative inhibitory) contacts, whereas asymmetric (putative stimulatory) contacts were unchanged, resulting in increased ratios of asymmetric/symmetric contacts (Table 1). Pomc-KI mice had the lowest number of inhibitory contacts. This is consistent with the fact that, as insulin inhibits POMC neurons' electrical activity (21), these neurons are likely to recruit fewer inhibitory contacts from other cells. It stands to reason that the lack of inhibitory connections renders these neurons more excitable, leading to increased HGP (29).

We next explored the contribution of altered energy balance and hypothalamic neuropeptide expression to the impairment of hepatic insulin sensitivity in Pomc-KI mice. As weanlings, L1, Agrp-KI, and Pomc-KI showed a 30% reduction of body weight compared with wild-type mice, and exhibited limited catch-up growth thereafter, attaining ∼90% of normal weight at 16 weeks of age (Fig. 4 A). This was associated with a 25% decrease in plasma IGF-1 levels, and resulted in an 11% decrease of BMIs (supplementary Table 1). Most importantly, for the purpose of the ensuing studies, body composition of the four groups of mice was similar (supplementary Table 1).

In view of the melanocortinergic (MCR) effects of insulin in the hypothalamus (30), we predicted that energy expenditure in L1 mice would reflect decreased MCR signaling, and that AgRP and POMC knock-ins would partially restore MCR effects. As reported previously, L1 mice had similar food intake to wild type (10) (Fig. 4,B), and exhibited normal oxygen consumption (Vo₂) and respiratory quotient (Fig. 4,C–F). They also showed a marked decrease in locomotor activity during the dark phase of the light cycle (Fig. 4,G and H). These data indicate that some MCR actions are preserved in L1 mice (food intake, Vo₂), whereas others are impaired (locomotor activity and suppression of HGP) (31,32). InsR reconstitution in AgRP neurons failed to normalize locomotor activity (Fig. 4,G and H), but restored insulin suppression of HGP (Fig. 2). There was a modest increase in Vo₂ during the light phase and no changes in other parameters (Fig. 4,B–F). Reconstitution of InsR in POMC neurons completely reversed the decrease in locomotor activity in L1 mice (Fig. 4,G–H) and increased Vo₂ (Fig. 4 C and D).

AgRP and POMC neurons project to the lateral hypothalamus, the site of melanin-concentrating hormone (Mch)– or hypocretin (Hcrt)–expressing neurons (33,34) that have been implicated in the regulation of locomotor activity (35,36). Fasting Mch and Hcrt levels were similar in all genotypes (supplementary Fig. 7). Under hyperinsulinemic clamp conditions, expression of Mch and Hcrt tended to be elevated in L1, Agrp-KI, and Pomc-KI compared with wild type, but the differences were not statistically significant.

The hypermetabolic state of Pomc-KI mice, in the presence of normal body weight/composition, led us to investigate their food intake. Pomc-KI mice showed modest increases in ad libitum feeding (Fig. 4,B) and a 36% increase in rebound food intake after fasting (Fig. 4,I). This was associated with a 34–87% reduction in plasma leptin levels under ad libitum–fed, overnight-fasted, and 6-h–refed conditions compared with L1 mice (Fig. 4,J) and a parallel reduction in leptin mRNA expression in adipose tissue (Fig. 4 K). The dissociation of leptin's ability to modulate energy expenditure from regulation of food intake in Pomc-KI mice provides indirect evidence that MCR target neurons regulating energy expenditure are distinct from those regulating food intake (12).

We determined the ability of clamp-induced hyperinsulinemia to affect Agrp and Pomc expression by comparing mRNA levels of the respective genes in fasted and in clamped mice. Insulin infusion raised Pomc levels in wild-type mice (Fig. 5,A); the effect was blunted in L1 and Agrp-KI, and normalized in Pomc-KI mice, consistent with cell-autonomous restoration of InsR-dependent Pomc activation. In contrast, Agrp levels were unaffected by clamp hyperinsulinemia in wild-type mice, probably reflecting different time course and dose sensitivity of insulin regulation of this gene. Nonetheless, in L1 mice we observed a trend toward insulin inhibition of Agrp that became statistically significant in both Agrp-KI and Pomc-KI mice (Fig. 5 B). These data provide critical support for the mechanism proposed by Barsh and colleagues (37), according to which AgRP neurons are regulated by insulin in both a cell-autonomous manner (explaining Agrp inhibition in Agrp-KI mice) and a nonautonomous manner that includes POMC neurons (explaining Agrp inhibition in Pomc-KI mice) and other inhibitory postsynaptic neurons (explaining the trend toward lower Agrp levels in L1 mice).

In summary (Fig. 6), hypothalamic insulin signaling controls both glucose metabolism and energy balance. InsR signaling in AgRP neurons restrains HGP independent of energy expenditure, whereas one-sided InsR activation in POMC neurons results in increased HGP. In addition, Pomc-KI mice show increased energy expenditure without attendant hypophagia. The rebound hyperphagia in Pomc-KI mice is probably due to hypoleptinemia and lack of concurrent inhibition of anabolic AgRP neurons.

Our findings support a branched-pathway model of hypothalamic insulin action (Fig. 6). In essence, 1) hypothalamic regulation of HGP by insulin appears to be a cell-autonomous response mediated by AgRP neurons; and 2) regulation of locomotor activity and energy expenditure appears to occur in a cell-autonomous manner via POMC neurons. Based on the findings of Balthasar et al. (12), it can be inferred that this response is mediated by second-order MCR neurons distinct from those in the paraventricular hypothalamus (PVH). Finally, 3) the anorexigenic response to insulin appears to require direct coregulation of POMC and AgRP neurons, as suggested by Barsh and colleagues (37). The physiologic implication of the present work is that different subsets of neurons mediate the pleiotropic actions of insulin in the hypothalamus.

The increase of HGP in Pomc-KI mice has three likely components: 1) a modest increase in hepatic glucose-6-phosphatase expression; 2) hypermetabolic state, acting perhaps to increase substrate flux to the liver (38); and 3) decreased hypothalamic K_ATP channels. Insulin hyperpolarizes AgRP (11) and POMC neurons (21,26) through activation of K_ATP channels in the mediobasal hypothalamus that results in HGP inhibition (8,9). The complex genetic makeup of Agrp-KI and Pomc-KI mice precludes direct electrophysiological studies. But Abcc8 expression was significantly lower in Pomc-KI mice than in any other genotype during hyperinsulinemic clamps. It should be noted that Abcc8 mRNA levels mirror SUR1 protein levels (39). Thus, decreased Abcc8 correlates with insulin's failure to inhibit HGP, and suggests that hypothalamic neurons are constitutively depolarized in these mice, and therefore resistant to insulin inhibition of their electrical activity. Further evidence that the decrease of Abcc8 is linked to InsR knock-in is provided by the observation that FoxO1 ablation in POMC neurons phenocopies this effect. Although the current data do not allow us to pinpoint the site of Abcc8 reduction, the magnitude of the reduction suggests that, in addition to POMC neurons, Abcc8 is also reduced in other arcuate cells, potentially including AgRP neurons, thereby increasing HGP. Together with the observation of decreased inhibitory synaptic contacts on POMC neurons in L1 mice, these data indicate that lack of InsR constitutively depolarizes hypothalamic neurons and that one-sided restoration of InsR signaling in Pomc-KI mice further decreases inhibitory synapses, presumably because it inhibits POMC neuron electrical activity, thus decreasing the need to recruit further inhibitory connections. The decrease of inhibitory contacts, in the presence of normal excitatory contacts, sets the stage for increased HGP. We also considered a potential contribution of neuropeptide alterations to increasing HGP, as intracerebroventricular infusion of α-melanocyte-stimulating hormone (αMSH), a processing product of POMC, was shown to acutely increase HGP (40). However, Pomc expression in Pomc-KI mice was unaltered during the hyperinsulinemic clamp, arguing against a significant change in αMSH. Future studies will be required to determine whether hyperinsulinemia affects αMSH release in Pomc-KI mice, thus contributing to increased HGP.

Our findings significantly expand upon a recent conditional knockout study demonstrating that InsR in AgRP neurons is necessary to suppress HGP (11). Our data go further, and indicate that InsR in AgRP neurons is sufficient to confer insulin inhibition on HGP. However, whereas IL-6 expression was decreased in liver of AgRP-specific InsR knockouts during hyperinsulinemic clamps, Agrp-KI mice showed no improvement in hepatic IL-6/Stat3 signaling compared with L1, possibly reflecting different duration of the fast prior to clamping (3–5 in this study vs. 16 h in the study by Könner et al. [11]). In contrast to our findings, conditional ablation of InsR in POMC neurons did not result in abnormalities of HGP (11). The likeliest explanation for this difference is that, under normal conditions and in the presence of concurrent activation of InsR in AgRP neurons, it would have been difficult to detect an increase in hepatic insulin sensitivity in Pomc-InsR knockouts. The data underscore the need to examine these complex questions using different models.

The reciprocal changes in locomotor activity observed in L1, Agrp-KI, and Pomc-KI provide evidence for a dominant role of hypothalamic InsR signaling in regulating this process. The hypothalamic effects of insulin on energy homeostasis are mediated primarily through the MCR system (32). Our data suggest that the locomotor/energy expenditure branch of MCR signaling is directly controlled via InsR signaling in POMC neurons, whereas the food intake branch of MCR signaling is largely unaffected by it. This is consistent with the demonstration that these two branches of MCR signaling are regulated by different neuronal subpopulations, with the PVH subset being more important for food intake (12). Accordingly, it can be inferred that InsR signaling is coupled primarily to POMC neurons that project to second-order MCR neurons distinct from those residing in the PVH. In addition to the PVH, ARC POMC neurons also project widely throughout the brain, including to the lateral hypothalamus, amygdala, and reticular nucleus (41). Lateral hypothalamus hypocretin and melanin-concentrating hormone neurons have been implicated in the regulation of energy expenditure, but we failed to detect differences in mRNA levels, indicating that the contribution of these pathways to the observed phenotype—if any—does not involve changes in gene expression. A recent study suggested that Stat3 signaling in AgRP neurons controls locomotor activity via secondary dopaminergic neurons in the striatum and frontal cortex (42). Further studies will be required to ascertain the contribution of POMC neurons in the nucleus of the solitary tract (41) and the role of secondary dopaminergic neurons to the observed phenotype.

The dissociation of InsR pathways regulating energy expenditure from those regulating food intake provides a likely explanation for the absence of obesity in the three models of hypothalamic insulin resistance presented in this article, unlike NIRKO mice (4). Although Pomc-KI mice show a trend toward increased rebound feeding, it should be considered that they are relatively hypoleptinemic. It bears emphasizing that Pten ablation in POMC neurons leads to hyperphagia and hyperleptinemia (21), consistent with the concept that the balance of hypothalamic InsR signaling is a key determinant of leptin production. The mechanism linking InsR hypothalamic signaling to leptin expression in adipocytes may involve nutritional and hormonal inputs and sympathetic nervous system cues (43). Hyperinsulinemia can induce sympathoactivation via PI 3-kinase and mitogen-activated protein kinase signaling in the rat mediobasal hypothalamus (44). It will be of interest to determine whether insulin action in POMC neurons regulates the sympathetic nervous system and contributes to the tonic inhibition of leptin production.

In conclusion, we demonstrate that different subsets of hypothalamic neurons mediate different branches of insulin's metabolic and energy expenditure actions. The data delineate an anatomic-functional map of hypothalamic insulin signaling that will be useful to interpret safety and efficacy data of antiobesity medications with a central mechanism of action.

This work was supported by the Berrie Fellowship Award to H.V.L., Deutsche Forschungsgemeinschaft Grant PL542/1-1 to L.P., National Institutes of Health Grants DK-58282 and DK-57539 to D.A., the Columbia University Diabetes and Endocrinology Research Center (DK-63608), and the Einstein Diabetes Research and Training Center (DK-20541).

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

Parts of this study were presented in abstract form at the 68th Scientific Sessions of the American Diabetes Association, San Francisco, California, 6–10 June 2008.

We thank B. Liu, K. Aizawa, Y. Su, R. Sgueglia, Q. Xu, Y. Dam, and B. Sarman for expert technical assistance and members of the Accili, Gutiérrez-Juárez, and Horvath laboratories for helpful discussion of the data.