Liver-specific deletion of insulin receptor substrate 2 does not impair hepatic glucose and lipid metabolism in mice
- First Online:
- Cite this article as:
- Simmgen, M., Knauf, C., Lopez, M. et al. Diabetologia (2006) 49: 552. doi:10.1007/s00125-005-0084-4
- 475 Downloads
Hepatic insulin resistance is thought to be a critical component in the pathogenesis of type 2 diabetes but the role of intrinsic insulin signalling pathways in the regulation of hepatic metabolism remains controversial. Global gene targeting in mice and in vitro studies have suggested that IRS2 mediates the physiological effects of insulin in the liver. Reduced hepatic production of IRS2 is found in many cases of insulin resistance. To investigate the role of IRS2 in regulating liver function in vivo, we generated mice that specifically lack Irs2 in the liver (LivIrs2KO).
Materials and methods
Hepatic insulin signalling events were examined in LivIrs2KO mice by western blotting. Glucose homeostasis and insulin sensitivity were assessed by glucose tolerance tests and hyperinsulinaemic–euglycaemic clamp studies. The effects of high-fat feeding upon glucose homeostasis were also determined. Liver function tests were performed and expression of key metabolic genes in the liver was determined by RT-PCR.
Proximal insulin signalling events and forkhead box O1 and A2 function were normal in the liver of LivIrs2KO mice, which displayed minimal abnormalities in glucose and lipid homeostasis, hepatic gene expression and liver function. In addition, hepatic lipid homeostasis and the metabolic response to a high-fat diet did not differ between LivIrs2KO and control mice.
Our findings suggest that liver IRS2 signalling, surprisingly, is not required for the long-term maintenance of glucose and lipid homeostasis, and that extra-hepatic IRS2-dependent mechanisms are involved in the regulation of these processes.
KeywordsInsulin receptor substrate proteinInsulin resistanceInsulin signallingLiver
- AlbCre mice
albumin Cre recombinase transgenic mice
fatty acid synthase
forkhead box A2
forkhead box O1
glycogen synthase kinase
hepatic glucose production
mice with a floxed allele of Irs2
mice lacking Insr in the liver
mice lacking Irs2 in liver
mitogen-activated protein kinase
zinc finger and BTB domain containing protein 16
Type 2 diabetes is a complex metabolic disorder characterised by skeletal muscle and adipose tissue insulin resistance coupled with pancreatic beta cell dysfunction [1, 2]. A failure to appropriately regulate hepatic glucose and lipid metabolism are also key components, but the mechanisms underlying these abnormalities are incompletely understood . The liver plays a critical role in fuel homeostasis, integrating the actions of anabolic and catabolic hormones with nutrient availability and requirements. Insulin may alter hepatic glucose production (HGP) and other aspects of metabolism through direct effects upon liver enzyme activity and gene expression, and through diverse indirect effects such as the regulation of glucagon production, suppression of gluconeogenic precursor production and via the central nervous system . However, the contribution of intrinsic hepatic insulin receptor (INSR) signalling to normal physiology and the pathogenesis of type 2 diabetes is an area of ongoing controversy , fuelled in part by observations using mouse models with different manipulations of the insulin signalling pathway [4–7].
Mice lacking Insr only in the liver (LIRKO mice) have marked insulin resistance and defects in hepatic glucose metabolism and gene expression, suggesting that direct hepatic INSR signalling is required for whole-body glucose homeostasis . Hyperinsulinaemic clamp studies in these mice suggest that hepatic Insr is required for both direct and indirect effects of insulin upon HGP . However, LIRKO mice developed significant liver disease and have abnormal insulin clearance, which may complicate this interpretation . In contrast, acute anti-sense oligonucleotide-mediated knockdown of liver Insr expression did not impair insulin action upon HGP . Similarly, transgenic rescue of Insr expression in the liver but not other tissues of Insr global null mice, suggested that indirect effects of insulin signalling predominate in the regulation of hepatic glucose metabolism . However, methodological and physiological questions have been raised about these studies .
Insulin receptor substrate proteins mediate the effects of the insulin receptor upon cellular and whole-body physiology , and therefore hepatic insulin resistance may result from defects in their function [10, 11]. Mice lacking Irs1 display profound growth retardation and insulin resistance, but do not develop diabetes due to beta cell compensation . In contrast, mice lacking Irs2 develop diabetes due to a combination of insulin resistance and pancreatic beta cell dysfunction [12, 13]. Hyperinsulinaemic–euglycaemic clamp analysis of Irs1 and Irs2-null mice demonstrated that Irs2-null mice have markedly impaired suppression of HGP and reduced stimulation of liver glycogen synthesis, while Irs1-null mice have insulin resistance predominantly in skeletal muscle and adipose tissue . Reduced hepatic IRS2 production has also been observed in a number of in vivo models of insulin resistance, suggesting a role in this condition [10, 15]. Studies on transformed hepatic cell lines derived from Irs2- or Insr-null mice have also demonstrated that IRS2 is the predominant IRS protein through which insulin acts [16, 17]. Short-term adenoviral RNA interference (RNAi) strategies to cause acute knockdown of hepatic Irs expression have suggested that IRS2 also mediates the effects of insulin upon hepatic lipid metabolism . Therefore to further understand the complex roles of IRS2 in liver metabolism, we generated a novel mouse line lacking Irs2 specifically in hepatocytes (LivIrs2KO mice) and performed detailed analysis of glucose homeostasis and liver function in these mice.
Materials and methods
Mouse breeding and genotyping strategies for mice with a floxed allele of Irs2 and for albumin Cre recombinase transgenic mice
Mice with a floxed allele of Irs2 (Irs2lox mice)  were intercrossed with albumin Cre recombinase transgenic mice (AlbCre mice) obtained from The Jackson Laboratory (Bar Harbor, ME, USA)  to generate compound heterozygote mice. Double heterozygote mice were crossed with Irs2lox+/− mice to obtain wild-type, Irs2lox+/+, AlbCre and AlbCreIrs2lox+/+ mice. Mice lacking Irs2 in AlbCre-expressing cells were designated LivIrs2KO mice. Mice were maintained on a 12-h light/dark cycle with free access to water and standard mouse chow (4% fat, RM1; Special Diet Services, Witham, Essex, UK) and housed in specific-pathogen-free barrier facilities. For high-fat diet studies, mice were fed with a diet containing 45% fat, 20% protein and 35% carbohydrate (Special Diet Services) for 3 months. Mice were handled and all in vivo studies performed in accordance with the 1986 Home Office Animal Procedures Act (Home Office, London, UK). LivIrs2KO mice were studied on a mixed 129Sv/C57Bl/6 background with appropriate litter-mate controls. Wild-type, Cre transgenic and Irs2lox+/+ mice were phenotypically indistinguishable and balanced numbers of mice of these genotypes were used as controls. Genotyping of the mice was performed by PCR amplification of tail DNA as described previously .
Blood samples were collected from mice via tail vein bleeds or from cardiac puncture on terminally anaesthetised mice. Blood glucose, plasma insulin levels and leptin levels were determined as previously described . Adiponectin was measured using a mouse adiponectin ELISA (R & D Systems, Minneapolis, MN, USA). NEFAs were measured using a NEFA kit (Roche Diagnostics, Lewes, UK). Serum and tissue triglyceride levels from fasted and fed animals were assayed using the GPO Trinder kit (Sigma-Aldrich, Dorset, UK). Albumin and liver function tests were measured using a Dimension RXL multi-channel analyzer (Dade-Behring, Milton Keynes, UK). Glucose and insulin tolerance tests were performed as previously described . Body fat mass was determined by dual emission X-ray absorptiometry using a PIXImus densitometer (GE Healthcare, Chalfont St Giles, UK).
RNA isolation and real-time quantitative RT-PCR
RT-PCR was performed as previously described  using FAM/TAMRA-labelled fluorescent probes. RT primers listed in 5’ to 3’ orientation were: for glucokinase (Gck), forward: GGTGCTTTTGAGACCCGTTTT, reverse: GAGTGCTCAGGATGTTAAGGATCTG, probe: TGTCGCAGGTGGAGAGCGACTCTG; and for glucose-6-phosphatase (G6pc) forward: ACTCTTGCTATCTTTCGAGGAAAGA, reverse: CCAACCACAAGATGACGTTCA, probe: AAAGCCAACGTATGGATTCCGGTGT. The relative amount of mRNA was calculated from an internal standard curve following normalisation to 18 S ribosomal RNA levels.
Hyperinsulinaemic clamp studies, whole-body glucose turnover rate, and tissue glycogen content
Hyperinsulinaemic clamp studies were performed as previously described . In brief, under general anaesthesia, an indwelling catheter was placed into the left femoral vein, subcutaneously tunnelled and externalised in the interscapular region. The animals were allowed to recover for 5 days and fasted for 6 h on the day of the experiment. For 3 h d-[3-3H]-glucose was infused at a rate of 11 mBq kg−1 min−1 and insulin at a rate of 18 mU kg−1 min−1. Euglycaemia was maintained by a variable infusion of 15% glucose. Whole blood was sampled from the tail every 10 min during the last hour for 3H-Glucose enrichment, which was biochemically determined as follows. Samples were de-proteinised by precipitation with equal volumes (125 μl) of 0.1 mol/l Ba(OH)2 and 0.1 mol/l ZnSO4, followed by centrifugation for 5 min. Total radioactivity in the supernatant (from d-[3-3H]-glucose and glycolysis-derived 3H2O) was measured by scintillation counting. A corresponding aliquot was evaporated to dryness to estimate d-[3-3H]-glucose alone. Total glucose concentration was determined in a third aliquot by the glucose oxidase method (BioMerieux, Marcy l’Etoile, France). Glycogen content was determined in liver and quadriceps muscle using the amyloglucosidase method. In brief, tissues underwent alkaline hydrolysis at 55°C for 1 h (30 mg of liver in 200 μl 1 mol/l NaOH, 20 mg of muscle in 50 μl 1 mol/l NaOH). Equivalent volumes of 1 mol/l HCl were added for neutralisation and the samples spun at 340 g for 2 min. Aliquots of the glycogen-containing supernatant were incubated at 37°C for 1 h with α-amyloglucosidase (Roche Diagnostics) at a concentration of 0.01 mg/ml in 0.2 mol/l Na citrate buffer pH 4.8. The released glucose was determined by the glucose oxidase method (BioMerieux). Baseline hepatic glucose content was measured in an undigested aliquot and subtracted from the value obtained after α-amyloglucosidase incubation. Calculations of glucose turnover were made from parameters obtained during the last 60 min of the infusions under steady-state conditions as described previously .
Preparation of tissue extracts and immunoblot assays
Following an overnight fast, a bolus of insulin (5 IU) or vehicle was injected via the inferior vena cava of terminally anaesthetised mice. Tissues were removed and snap-frozen in liquid nitrogen and stored at −80°C until use. Tissues were homogenised in lysis buffer , solubilised for 30 min on ice and clarified by centrifugation. Supernatants were snap-frozen in aliquots and stored at −80°C. For Western blotting, 50 μg of total protein extract was immunoblotted with the indicated antibodies. For analysis of IRS protein levels and tyrosine phosphorylation or INSR β tyrosine phosphorylation, tissue extracts (2 mg of total protein) were immunoprecipitated for 2 h with the indicated antibodies. Immune complexes were collected with 100 μl of 50% slurry of protein-A or protein-G sepharose, washed with lysis buffer and resolved on 7.5% SDS-PAGE and transferred to nitrocellulose. The blots were probed with polyclonal antibodies against IRS2 and IRS1 or monoclonal antibody to antiphosphotyrosine and enhanced chemiluminescence (Amersham, Little Chalfont, UK) detection performed. Nuclear preparations were performed by homogenisation of liver samples in nuclear buffer  followed by centrifugation over a cushion of nuclear buffer with 1.2 mol/l sucrose and 5% glycerol. Nuclear pellets were re-suspended in nuclear buffer with 0.3 mol/l sucrose and 10% glycerol, purity analysed under a haematocytometer and aliquots frozen. Nuclear preparations were subsequently lysed in lysis buffer, solubilised for 5 min on ice, clarified by centrifugation, after which 50 μg of nuclear extract was immunoblotted with the indicated antibodies. Blots were stripped and re-probed with an antibody against zinc finger and BTB domain containing protein 16 (ZBTB16) as a nuclear loading control.
Sheep polyclonal antibodies to IRS1 and IRS2 were used as previously described . Rabbit anti-IRS1, anti-IRS2, anti-phosphotyrosine (PY) (4G10) and anti-p85 antibodies were from Upstate Biotechnology (Dundee, UK). Anti-AKT, anti-pAKTSer473, anti-pp70S6KThr389, anti-glycogen synthase kinase (GSK)-3β, anti-pGSK3α/βSer9/21, anti-mitogen-activated protein kinase (MAPK), anti-pMAPKThr202/Tyr204, anti-forkhead box O1 (FOXO1) antibodies were from Cell Signalling Technology (Beverly, MA, USA) and anti-forkhead box A2 (FOXA2) antibody was a gift from R. I. Altaba (University of Geneva Medical School, Geneva, Switzerland). Anti-INSR β, and anti-p70S6K antibodies were from Santa Cruz (Insight Biotechnology, Wembley, UK). Mouse anti-ZBTB16 was from Calbiochem-Novachem (San Diego, CA, USA). HRP-conjugated goat anti-rabbit antibody was from DAKO (Ely, UK) and HRP-conjugated sheep anti-mouse antibody was from Amersham Bioscience.
All statistics were performed using GraphPad Prism4 software (Graphpad Software, San Diego, CA, USA) and paired and unpaired t-tests and two-way ANOVA with Bonferroni post-tests performed as appropriate. A p value of <0.05 was regarded as significant.
Mice lacking Irs2 in liver (LivIrs2KO mice)
Analysis of insulin signalling pathways in LivIrs2KO mice
INSR β and IRS1 expression and tyrosine phosphorylation in response to in vivo insulin stimulation were equivalent in control and LivIrs2KO mice, suggesting there was no compensation by these events for the loss of IRS2 (Fig. 1c,d). Insulin-stimulated recruitment of the p85 adaptor subunit of phosphatidylinositol 3-kinase (PI3K) to IRS1 immunoprecipitates was also unchanged in LivIrs2KO mice (Fig. 1d). These findings suggest that proximal elements of the IRS1 limb of the insulin signalling cascade were normal in LivIrs2KO mice, but it has been suggested that there is differential signalling via IRS1 and IRS2 to AKT and GSK-3β. We detected no alteration in expression of AKT or GSK3α/β in the LivIrs2KO mice (Fig. 1e,g). Both AKT and GSK3β basal phosphorylation was mildly elevated in LivIrs2KO mice, but there was no defect in insulin-stimulated phosphorylation of either of these kinases (Fig. 1e–h). MAPK expression and phosphorylation were not perturbed in LivIrs2KO mice (Fig. 1i,j), and no alterations in p70S6K and PDK-1 expression and phosphorylation were detected in these mice (data not shown).
IRS1 and IRS2 have been suggested to have different signalling potential to forkhead transcription factors, which are major transcriptional mediators of insulin action . We therefore examined the nuclear localisation of FOXO1 and FOXA2 upon insulin stimulation, as well as the effects of a physiological stimulus, the fast/fed transition, by isolating nuclear protein preparations from liver lysates and subjecting them to Western blotting for both of these factors. In response to insulin stimulation, we detected an equivalent translocation of either FOXO1 or FOXA2 from the nucleus in LivIrs2KO mice compared to controls (Fig. 1k). Likewise, FOXO1 translocated from the nucleus after a 2-h period of feeding following a 16-h overnight fast in both LivIrs2KO and control mice (Fig. 1l). However, we were unable to demonstrate a convincing translocation of FOXA2 under these conditions in mice of either genotype (Fig. 1), while analysis of AKT phosphorylation in the same animals demonstrated an equivalent increase in pAKTSer473 upon re-feeding in LivIrs2KO and control mice (data not shown). These findings suggest that there is no defect in FOXO1 or FOXA2 function in LivIrs2KO mice.
Glucose homeostasis in LivIrs2KO mice
The surprisingly modest abnormalities in glucose metabolism in LivIrs2KO mice prompted us to undertake more detailed analysis by performing hyperinsulinaemic-euglycaemic clamps. Whole-body glucose disposal was mildly reduced in LivIrs2KO mice (Fig. 2e). This impairment was associated with a small reduction in whole-body glycogen synthesis, which accounted for the impaired whole-body glucose utilisation (Fig. 2h). However, endogenous glucose production and glycolysis rates were not different between LivIrs2KO and control mice (Fig. 2f,g). These findings suggest that LivIrs2KO mice have a very mild defect in whole-body insulin sensitivity, which is unlikely to be detected by glucose and insulin tolerance tests, and also does not impact on long-term glucose homeostasis.
Lipid homeostasis and response to high-fat diet in LivIrs2KO mice
Hepatic insulin resistance is involved in the development of the metabolic syndrome induced by increased fat intake . Therefore, we next assessed the effects of a high-fat diet upon glucose homeostasis in LivIrs2KO mice. Control and LivIrs2KO male mice fed a diet consisting of 45% fat for 3 months gained on average 25% more weight than mice on normal chow (data not shown). However, while fasting blood glucose levels in LivIrs2KO mice were mildly higher than in control mice (Fig. 3e), glucose tolerance was not significantly different (Fig. 3f). Fasting insulin levels and serum and liver triglyceride levels were similar in both control and LivIrs2KO mice subjected to high-fat diet (data not shown). These findings suggest that a high-fat diet does not result in greater insulin resistance in LivIrs2KO mice than in control animals.
Liver function in LivIrs2KO mice
Hepatic gene and protein expression in LivIrs2KO mice
Insulin regulates hepatic metabolism partly by regulating the expression of key metabolic genes in glucose and lipid metabolism. RT-PCR analysis of liver mRNA expression revealed that in the fasted state, Gck expression was mildly reduced in LivIrs2KO mice, although in the fed state no difference was found (Fig. 4c). G6pc expression was mildly increased in fasted LivIrs2KO mice, but in the fed state no difference was found (Fig. 4d). In contrast RNA expression of a panel of other hepatic metabolic genes (notably Pck1, Fasn, Srebf1a and 1c, Acaca, Ppara and Pparg1, Cpt1a, Ppargc1b and Igfbp1) were unaltered in LivIrs2KO mice either in the fasted or fed state (data not shown). Therefore deletion of Irs2 in the liver had no significant effect on a number of insulin-regulated transcriptional events.
LivIrs2KO mice displayed no marked alteration in insulin receptor, IRS1 and downstream PI3K-dependent signalling events in the liver. These observations contrast with findings in mice with global deletion of Irs2 and in Irs2-null transformed hepatic cell lines derived from neonatal mice. In both these models there are abnormalities in signalling events downstream of the insulin receptor. From our studies, however, it is clear that the chronic deletion of Irs2 in the liver appears to have minimal effects upon proximal insulin signalling events in vivo. The transcription factors FOXO1 and FOXA2 are thought to play distinct roles in the regulation of hepatic carbohydrate and lipid metabolism. FOXO1, which predominantly regulates gluconeogenesis, has been reported to be phosphorylated only by the IRS2 signalling limb while FOXA2, which regulates fatty acid oxidation, was phosphorylated in response to both IRS1 and IRS2 signalling pathways . However, these in vitro findings were not confirmed by our studies. In the absence of liver Irs2, FOXO1 translocation both in response to insulin and refeeding was normal. FOXA2 translocation in response to insulin was also comparable to that seen in control animals. Therefore, these results do not support the idea that IRS2 preferentially signals to FOXO1.
Consistent with the normal hepatic insulin signalling events, LivIrs2KO mice on a regular diet displayed minimal abnormalities of glucose homeostasis up to and beyond 6 months of age, which contrasts with the marked and progressive diabetes observed in Irs2 global null mice . Although we did detect decreased expression of Gck and increased expression of G6pc in the livers of fasted LivIrs2KO mice, suggesting some minor dysregulation of insulin-regulated gene expression in the fasted state, we did not find abnormalities in the expression of these enzymes in the fed state, suggesting that they are appropriately regulated by insulin. Interestingly, Irs2 expression is high in the fasted state and markedly reduced by feeding and by increased insulin levels , and perhaps IRS2 in the liver serves to predominantly regulate the expression of particular genes in the fasted state. Taken together, these findings suggest that IRS2-dependent pathways are not essential for the long-term regulation of glucose homeostasis.
Recent studies have suggested that hepatic IRS2 is particularly involved in the regulation of lipid metabolism. RNAi knockdown of Irs2 resulted in increased levels of the transcription factor SREBF1c and this was associated with mildly increased serum NEFA and triglyceride levels and mild hepatic steatosis . Others have demonstrated a reciprocal relationship between IRS2 function and the expression of Srebf1c [10, 11]. We found that hepatic deficiency of IRS2 had no effects upon circulating NEFA and triglyceride or on hepatic triglyceride levels. Likewise, we found no alterations in the expression of key hepatic transcription factors and enzymes involved in the regulation of lipid metabolism or of leptin levels or of total body fat stores. We also found no differences in glucose and lipid homeostasis between LivIrs2KO and control mice while on a high-fat diet. These findings suggest that hepatic IRS2-dependent signalling events are not required for the long-term in vivo maintenance of hepatic lipid homeostasis.
The differences between our observations and those of others may be explained by a number of methodological and physiological factors. Differences between global and conditional deletion may reflect developmental events and are potentially related to both the timing of Irs2 deletion as well as to the contribution of other tissues to the phenotype. AlbCre mice are reported to delete floxed alleles completely and this may occur at several weeks of age post-natally . We found complete deletion of Irs2 by 4 weeks of age and our subsequent physiological studies were performed on mice at older ages. Studies in immortalised cell lines do not reflect the in vivo physiological situation; moreover, the hepatocyte cell lines reported were derived from neonatal mice with germ-line deletion of either Insr or Irs2 [16, 17]. For the studies involving the disruption of Irs2 expression by RNAi, the knockdown was both relatively short-term lasting no more than 5 days and was incomplete, with a persistent 40% residual Irs2 expression evident . In contrast, our studies involve long-term complete deletion of Irs2, which may permit compensatory mechanisms to develop. However, we saw no increased expression of insulin signalling components in the liver and no alterations of Irs2 expression in other tissues. Furthermore, in the Irs2 RNAi knock-down there was an alteration in adipose tissue function, which was not apparent in our studies, but may explain the abnormalities seen in hepatic lipid homeostasis. The mechanism for this effect was not determined but it is clear that systemic as opposed to tissue-specific delivery of RNAi may abrogate gene expression in other tissues in addition to the liver . Unfortunately, Irs2 expression in adipose tissue was not reported in the liver Irs2 RNAi model. Since insulin positively regulates lipogenesis, disruption of insulin receptor function via down-regulation of IRS signalling in the liver alone might be expected to reduce rather than increase lipogenesis. Indeed, this is seen in LIRKO mice and in mice with knock-down of liver Insr expression by anti-sense oligonucleotide [5, 7]. The mechanism of increased lipogenesis in Irs2 RNAi knock-down mice is probably due to extra-hepatic effects or to residual insulin signalling in the liver. Our findings also call into question the suggestion that there is a physiologically relevant reciprocal relationship between IRS2 and regulators of lipid homeostasis such as SREBF1c [10, 11]. These findings were largely derived from in vitro models and in vivo models with systemic deletion of Irs2 or with complex multi-tissue metabolic abnormalities such as ob/ob and lipodystrophic mice, and this may explain the differences between our observations and those of others.
Recent data from two distinct mouse models have suggested that insulin regulates hepatic metabolism largely through indirect means rather than through the liver insulin receptor and this provides a physiological explanation why LivIrs2KO mice have a minimal phenotype [4, 5]. While these recent findings are at odds with those generated by studies on the LIRKO mice, this latter model has marked defects in hepatic function and insulin clearance that impinge upon insulin action in other tissues and may explain the differences in phenotype. Our observations in LivIrs2KO mice are consistent with the idea that insulin signalling pathways are not required for the direct regulation of hepatic metabolism. While we have not abrogated other insulin receptor signalling pathways through this manipulation, hepatic Irs2 is not required for the long-term maintenance of many aspects of hepatic function. Recently we showed that mice lacking Irs2 only in the central nervous system develop insulin resistance and impaired glucose homeostasis, and we are exploring hepatic metabolism in this model, as it is likely that central nervous system IRS2 pathways contribute to the regulation of hepatic metabolism . These findings together with those of others suggest that in mice, central nervous system pathways and other indirect effects may be the predominant mechanisms by which insulin signalling regulates hepatic metabolism.
In summary, through the generation of LivIrs2KO mice we have demonstrated that IRS2 signalling events in the liver are not required for the regulation of hepatic metabolism.
This work was supported by grants from the Wellcome Trust (to M. Simmgen, A. Choudhury, M. Charalambous, A. Vidal-Puig and D. J. Withers), from Diabetes UK (to J. Cantley and D. Bedford), from the Medical Research Council (to A. Vidal-Puig and D. J. Withers) and from the Biotechnology and Biological Sciences Research Council (to M. Claret and H. Heffron). The work was in part performed in the Integrative Physiology Consortium for the Study of Common Metabolic Disease, which is supported by the Wellcome Trust Integrative Physiology programme (to A. Vidal-Puig), and in the Biological Atlas of Insulin Resistance (BAIR) consortium, which is supported by the Wellcome Trust Functional Genomics programme (to D. J. Withers). M. Simmgen is a Wellcome Trust Training Fellow.