Lilly Lecture 2003: The Struggle for Mastery in Insulin Action: From Triumvirate to Republic

Insulin resistance is genetically determined in many individuals. Epidemiological studies indicate that having a diabetic first-degree relative causes a fourfold increase in one’s risk of developing diabetes. Moreover, the concordance rate for type 2 diabetes in monozygotic twins exceeds that of dizygotic twins, suggesting that genes, more than environment or maternal factors, account for the excess risk (10). Studies of monogenic forms of insulin-resistant diabetes (11) as well as maturity-onset diabetes of the young (MODY) (12) have revealed a surprisingly broad cast of diabetes susceptibility genes. However, the common clinical form of insulin resistance has proven largely intractable in human genetic studies. Several factors have concurred to thwart genetic studies of insulin resistance in humans: phenocopies, i.e., the ability of environmental conditions (diet, sedentary lifestyle) to cause insulin resistance; genetic heterogeneity, i.e., the ability of mutations in different genes to bring about the same metabolic trait (for example, fasting hyperinsulinemia); and polygenic inheritance, i.e., interactions among multiple diabetes susceptibility alleles in the same individual (13). To complicate things further, diabetes susceptibility alleles are likely to be common variants, as opposed to rare mutations (14). This adds another level of complexity, as the ability to detect allele-specific risk decreases with increasing allele frequencies (15), and because functional testing of common allelic variants poses an unmet challenge.

It is because of this daunting complexity that investigations have increasingly relied on gene-targeted and transgenic mice to examine the contribution of genetic variants to insulin resistance and identify new diabetes-causing alleles.

I will review experiments in mice carrying heterozygous mutations of Insulin receptor (Insr) as an example of how mouse genetics can provide insight into the genetic interactions leading to type 2 diabetes (16,17). A first step in this direction was the development of an oligogenic model of insulin-resistant diabetes by crossing mice with heterozygous null mutations of Insr and Irs1. Whereas mice heterozygous for Irs1 deficiency are normal, mice heterozygous for both Insr and Irs1 null alleles develop severe hyperinsulinemia and pancreatic β-cell hyperplasia, and, by 4–6 months of age, a sizable fraction of these mice becomes frankly hyperglycemic. This process closely resembles the pathogenesis of human diabetes, in which insulin resistance is a chronic process and leads gradually to β-cell failure (18,19). The main lesson from these studies is that even a major predisposing allele, such as the null Insr mutation, has a modest effect by itself but plays a major role in the context of a predisposing background (20). Indeed, when the double heterozygous knockout mice are bred onto different genetic backgrounds, the prevalence of diabetes can vary from <2 to 85% (21,22). Interestingly, the relative risk of diabetes in double heterozygous offspring (Insr/Irs1^+/–) of single heterozygous Insr or Irs1 parents increases fourfold, similar to the excess risk of diabetes in first-degree relatives of humans with type 2 diabetes (10). The findings in the Insr/Irs1^+/– mouse are consistent with an oligogenic mode of inheritance of type 2 diabetes and are supported by human genetic studies in which the susceptibility to diabetes in carriers of the NIDDM1 mutation (CALPAIN10) is increased by alleles on chromosome 15 (23).

While combined mutations of Insr and Irs1 have provided insight into the polygenic nature of type 2 diabetes, comparisons of double-mutant mice with Insr/Irs1 or Insr/Irs2 null alleles have shed light onto the issue of genetic heterogeneity. Genetic heterogeneity means that mutations in different genes can result in a similar phenotype. Indeed, both Insr/Irs1 and Insr/Irs2 double-mutant mice develop diabetes at similar rates. However, the Insr/Irs1 mice do so because they are primarily insulin resistant in skeletal muscle, whereas the Insr/Irs2 mice do so because they are primarily insulin resistant in liver (19).

Experimental evidence supporting the need for a reassessment of the primary site(s) of insulin resistance came from two studies in genetically modified mice (24,25). In the first study, our group set out to critically test the hypothesis that insulin resistance in adipose tissue and skeletal muscle is necessary and sufficient to cause diabetes. To this end, we generated mice that were resistant to insulin action in these two tissues and asked whether they developed hyperglycemia. This task was achieved by crossing mice haploinsufficient for an Insr null allele (Insr^+/–) with transgenic mice bearing a kinase-inactive Insr transgene (K1030M) expressed in muscle and fat but not in liver and β-cells. Biochemical and metabolic data indicated that the double-mutant mice (Insr^+/– K1030M) had impaired insulin-dependent glucose uptake in muscle and adipocytes and developed impaired glucose tolerance without ever becoming diabetic (Fig. 3). Their β-cell function was preserved throughout life, despite the increased metabolic demands (24). Thus, these findings indicate that mice can overcome the insulin resistance of muscle and adipose tissue as long as insulin action in liver is preserved and there are no additional strains on the β-cell.

The second piece of evidence came from mice bearing a conditional inactivation of Insr in skeletal muscle (25), generated by tissue-specific recombination with the Cre/loxP system (26) to specifically ablate the Insr gene product in skeletal muscle cells. Conditional Insr knockout in skeletal muscle (so-called “MIRKO” mice) leads to impaired insulin signaling and decreased glucose transport in isolated muscle strips in response to insulin, without systemic insulin resistance. This decrease in muscle metabolism is partly compensated for by a concurrent increase in insulin-stimulated glucose transport in adipose tissue (27), which may explain the lack of a more severe impairment of glucose metabolism (Fig. 4).

These observations are unexpected. Whereas it is generally accepted that insulin resistance per se—in muscle or elsewhere—is not sufficient to cause overt diabetes, the prediction was that an impairment of insulin signaling in muscle and/or adipose tissue would lead to systemic insulin resistance. This apparent discrepancy is discussed below.

The failure to develop diabetes in mice with a selective impairment of insulin signaling in muscle can be explained by two compensatory pathways: the IGF1 receptor (Igf1r) pathway and the contraction-activated pathway.

The Igf1r shares extensive structural similarities with Insr. Like Insr, it is composed of two homodimeric subunits (α₂β₂) and is activated upon ligand binding, the main distinguishing feature of Insr-like receptors from other receptor tyrosine kinases (28). Moreover, the Insr and Igf1r engage in heterodimer formation, yielding “hybrid” receptors (29). Although the interactions of Insr and Igf1r have mostly been thought to regulate embryonic growth (30,31), the metabolic role of Igf1r is emphasized by experiments in which a dominant-negative Igf1r expressed in skeletal muscle of transgenic mice was able to cause diabetes, presumably through inhibition of Insr receptor function (32). These data are consistent with the ability of IGF1 to decrease glucose levels when injected in Insr-deficient mice (33).

Muscle contraction is another powerful trigger for glucose metabolism (34, 35) and provides an additional, potential compensatory mechanism whereby mice deficient in muscle Insr signaling can overcome insulin resistance. Indeed, contraction-stimulated glucose uptake in MIRKO mice is normal (36). The conclusion from these different datasets is that alternative biochemical pathways can bypass the requirement for insulin signaling to promote glucose utilization in muscle.

These data should not be construed as dismissive of the muscle’s role in glucose metabolism. For example, selective disruption of the insulin-sensitive glucose transporter Glut4 in muscle results in a profound reduction of both insulin- and contraction-stimulated glucose transport, with early-onset insulin resistance and glucose intolerance (37). In conclusion, these studies do not challenge the fact that muscle glucose disposal is central to fuel metabolism, but indicate that Insr signaling is only one of the pathways leading to Glut4-mediated glucose uptake.

The use of techniques of gene manipulation in mice has breathed new life into our approach to the integrated physiology of insulin action, the metabolic “balance of power” among various organs in determining insulin sensitivity (or lack thereof). There have been several studies in which investigators have genetically inactivated insulin signaling or glucose transport in selected tissues and then asked what the metabolic consequences of these manipulations would be. A detailed analysis of this work (38) is beyond the scope of the present review, but I will summarize some of the conclusions (Fig. 5). A common theme emerging from these investigations is that inactivation of insulin signaling in tissues that exhibit insulin-dependent glucose uptake and utilization, such as muscle and fat, generally results in milder metabolic phenotypes than would be predicted (18,39). On the other hand, inactivation of insulin signaling in tissues that are not dependent on insulin for their glucose uptake, such as brain, liver, and pancreatic β-cells, is associated with significant metabolic abnormalities, suggesting that these tissues play a primary role in insulin resistance (40–42). Let me then turn to insulin action in two of these cell types, the hepatocyte and the pancreatic β-cell.

The liver is an important site of inhibition of gene expression by insulin. There are acute effects of insulin on hepatic glucose production and uptake that may be mediated via both direct and indirect mechanisms. These effects are very rapid and do not require regulation of gene transcription (43). However, there are potent effects of insulin in modulating the gene expression of key metabolic enzymes. These effects may account for the long-term inhibition of hepatic glucose output by insulin (44). Pioneering work by Granner (45) and Hanson (46) has demonstrated that insulin inhibits the expression of the gene encoding phospho-enolpyruvate carboxykinase (Pck1). Similarly, glucose-6-phosphatase (G6pc) gene expression is inhibited by insulin (47). A liver with altered expression of these enzymes has altered responses not only to the acute direct and indirect effects of insulin, but also to the important effects of hyperglycemia per se on glucose production and uptake (48). Thus, alterations in insulin’s ability to regulate hepatic gene expression are likely to play a pivotal role in the pathophysiology of type 2 diabetes.

The mechanism by which insulin regulates gene expression in liver has until recently remained unclear, despite tremendous progress in understanding how insulin works (49). Genetic epistasis experiments in the roundworm C. Elegans identified Daf16, a transcription factor of the forkhead family, as a key effector of insulin signaling (50,51). The prediction from the C. Elegans studies was that the mammalian ortholog of Daf16, FoxO (FOrkhead boX–containing gene, O subfamily), would act as a negative regulator of insulin signaling and that loss-of-function Foxo mutations would rescue diabetes due to Insr mutations. Foxo proteins are Akt substrates in vivo (52). Upon phosphorylation, Foxo1 is inhibited by nuclear exclusion, thus providing a mechanism for negative control of gene expression by insulin and other growth factors that signal through phosphatidylinositol 3-kinase (PI3K) and Akt (Fig. 6).

Foxo1 plays an important role in insulin control of hepatic glucose production. Evidence from studies in cultured cells indicates that Foxo1 can single-handedly confer insulin-inhibitable transcription of G6pc (53). In the fasting state, Foxo1 is largely dephosphorylated and localizes to the nucleus, where it binds the peroxisome proliferator–activated receptor-γ coactivator (PGC)-1α (54) and Cbp/p300 (55) to promote Pck1 and G6pc gene transcription. Insulin-dependent phosphorylation dissociates the Foxo1/PGC-1α complex and causes Foxo1 redistribution to the cytoplasm. Haploinsufficiency of the Foxo1 gene restores insulin sensitivity and rescues the diabetic phenotype in insulin-resistant mice by reducing hepatic expression of glucogenetic genes and increasing adipocyte expression of insulin-sensitizing genes. Conversely, a gain-of-function Foxo1 mutant targeted to liver results in diabetes arising from increased hepatic glucose production (56). In addition to restoring insulin sensitivity in a genetic model of insulin resistance, Foxo1 haploinsufficiency protects from diet-induced diabetes in mice, suggesting that strategies aimed at decreasing Foxo1 levels and/or activity represent a new approach to diabetes treatment (57).

Foxo1 plays an important role in adipocyte development and physiology. In fact, Foxo1 haploinsufficiency has been shown to decrease adipocyte size and increase expression of genes that promote lipid metabolism (57). This effect is not simply secondary to improved insulin sensitivity in liver, but can be explained by a direct action of this transcription factor in adipocytes. In vitro studies of adipocyte differentiation indicate that Foxo1 expression is induced in the early stages of adipogenesis in 3T3-F442A cells. Interestingly, Foxo1 phosphorylation appears to be tightly regulated (57) during differentiation of 3T3-F442A cells. Expression of a phosphorylation-defective Foxo1 prevents adipocyte differentiation by altering expression patterns of genes involved in cell cycle control and adipogenesis. In contrast, a dominant-negative Foxo1 restores adipocyte differentiation of murine embryonic fibroblasts from Insr-deficient mice (57). The emerging model of Foxo1 function in fat cells is that it may control the coupling of extracellular (hormonal) cues that activate adipocyte differentiation with the cell cycle machinery. A similar paradigm appears to be at play in myoblast differentiation (58).

While Foxo1 function in liver was wholly predictable, the finding that Foxo1 immunoreactivity is abundant in pancreatic β-cells and ducts raises the question of its role in these cells (Fig. 7). Mice lacking Irs2 develop β-cell failure, suggesting that insulin signaling is required to maintain an adequate β-cell mass (59,60). Haploinsufficiency of Foxo1 reverses β-cell failure in Irs2^–/– mice through partial restoration of β-cell proliferation and increased expression of the pancreatic transcription factor Pdx1. Interestingly, Foxo1 and Pdx1 exhibit mutually exclusive patterns of nuclear localization in β-cells, and constitutive nuclear expression of Foxo1 is associated with reduced Pdx1 expression (56). Based on the localization of Foxo1 to a subset of insulin-positive cells abutting pancreatic ducts, we have suggested that insulin/IGFs regulate β-cell differentiation by relieving Foxo1 inhibition of Pdx1 expression in these cells (61). The implication of these studies is that Foxo1 expression is a marker of a pancreatic cell subpopulation with the potential to give rise to mature β-cells.

As noted above, β-cell dysfunction is a sine qua non for development of the diabetes phenotype and may also play a role in the insulin-resistant state. In states of metabolic dysregulation, defects are seen in both insulin secretion and compensatory changes in β-cell mass. The maintenance of an adequate β-cell mass is the result of a physiologic balance between β-cell proliferation and apoptosis. Diabetes ultimately results from the failure to maintain β-cell mass (62). Unlike other terminally differentiated cell types, pancreatic islet β-cells maintain the ability to undergo proliferation in both physiologic (growth, gestation) and disease conditions (obesity, insulin resistance). At the same time, there is evidence for de novo generation of β-cells from undifferentiated progenitor cells, possibly arising from the epithelial lining of pancreatic ducts (63–65) or from a nonrenewable pool of Neurogenin3 (Neurog3)-positive endocrine progenitor cells (66). The factors that drive β-cell neogenesis and proliferation under normal or pathological conditions are unknown (67).

The association of peripheral insulin resistance with compensatory β-cell hyperplasia is extremely well documented in animal models and, to a lesser extent, in humans (12). We do not know why β-cells replicate in response to insulin resistance. One possibility is that they are responding to changes in insulin secretory patterns, although insulin resistance initially causes only subtle abnormalities of insulin secretion. Other authors have suggested that β-cells proliferate in response to transient oscillations in glucose levels, occurring before the onset of overt hyperglycemia (62). The alternative and not mutually exclusive view is that they are activated to proliferate by circulating β-cell growth factors (68).

Classically, β-cell function has been viewed as a response to intracellular glucose metabolism triggered by variations in plasma glucose levels (69). In recent years, signaling by receptor tyrosine kinases has been integrated among the β-cell regulatory mechanisms, impinging on such diverse functions as glucose sensing, insulin release, and cellular proliferation (67). Based on observations in genetically engineered mice lacking various components of the insulin/IGF signaling pathway, it has been proposed that insulin and IGF receptors regulate two key processes in the life of a β-cell: proliferation and hormone secretion (Fig. 8). Ablation of Insr in β-cells by site-specific recombination leads to altered glucose sensing and impairs glucose tolerance (40). A similar phenotype results from generalized ablation of Irs1 (70). Similar results have been obtained in mice lacking Igf1 receptors in β-cells (71,72), consistent with the observation that IGF1 is a negative regulator of insulin secretion (73). On the other hand, it has been shown that ablation of Irs2 results in decreased β-cell mass (59,74), and that this phenotype is exacerbated by Igf1r (75) or Insr haploinsufficiency (61). Complex alterations of insulin secretion result also from mutations of PI3K (76,77). Furthermore, ablation of p70^s6k1, an Akt substrate, is associated with decreased β-cell size (78). Conversely, overexpression of a constitutively active mutant Akt increases β-cell mass and protects from streptozotocin-induced diabetes without affecting insulin secretion (79,80).

As stated above, there is convincing evidence for β-cell neogenesis, but the source of new β-cells is unknown. Insulin-producing cells can be commonly visualized in the vicinity of pancreatic ducts (81), and the number of duct-associated, hormone-positive cells increases in insulin resistance or following partial pancreatectomy (82). Increased duct-associated β-cell neogenesis has also been described in mice expressing constitutively active Akt (79,80). Bonner-Weir et al. (83) have shown that the purification procedure commonly used to obtain pancreatic islets from cadaver donors for transplantation purposes yields a mixture of hormone-producing islets and insulin/pancytokeratin double-positive cells, which they have suggested to represent insulinogenic duct cells. The implication of these studies is that terminally differentiated epithelial ducts are physiologically able to yield hormone-producing cells by a transdifferentiation process (84). The identity of the cell type that gives rise to insulin-producing cells within ducts has remained elusive. Several authors have suggested that these progenitor cells are epithelial in nature, based on the expression of various cytokeratin isoforms, carbonic anhydrase II, and lectins. It is uncertain whether they also express Nestin, another potential pancreatic stem cell marker (85–87).

Recent evidence from lineage-tracing studies appears not to substantiate the alleged role of duct epithelia as a source of endocrine progenitor cells (Fig. 9). Gu et al. (66) have used an inducible Cre transgene driven by either the Neurog3 or Pdx1 promoters to indelibly label pancreatic cells. The conclusions of these studies are that, whereas Neurog3-positive cells act as nonrenewable endocrine progenitor cells, Pdx1-positive cells give rise to exocrine and endocrine cells, as well as epithelial duct cells. The latter, however, fail to develop into islets, indicating that duct epithelia do not act as a physiologic source of endocrine cells. These data do not rule out that duct cells may give rise to endocrine cells with a low frequency or in specific situations (for example, in response to increased metabolic demand) (88).

Although it is clear that β-cells can proliferate in response to a host of growth factors, the most parsimonious explanation for insulin resistance–induced β-cell hyperplasia is that insulin itself promotes β-cell neogenesis and proliferation (67). Insulin may have an autocrine effect on β-cell proliferation, but this mechanism would not explain the increase in neogenesis observed in hyperinsulinemic states. The main problem with this explanation is that β-cells are constantly exposed to insulin, so it is unclear how they could discriminate between a physiologic and an abnormal insulin concentration.

To reconcile these seemingly inconsistent observations, we need to reexamine the functional anatomy of the pancreas. It has been suggested that pancreatic hormones are in part released directly into the ductal system, by way of close connections between endocrine cells and ducts (63). It has been known since the late 1970s that sizable quantities of pancreatic hormones, including insulin, glucagons, and somatostatin, can be detected in pancreatic juice (89). It is therefore tempting to speculate that insulin released into the pancreatic ducts acts directly on a selected population of endocrine progenitor cells that are in close proximity to ducts to stimulate β-cell neogenesis. We hypothesize that these cells are characterized by Foxo1 expression and are maintained in an undifferentiated state because Foxo1 acts as a transcriptional brake on the differentiation program. When duct cells are exposed to insulin, Foxo1 translocates to the cytoplasm, relieving the constraint on terminal differentiation and promoting β-cell neogenesis.

Insulin signaling can be viewed as a three-tiered system, in which a metabolic response requires initially tyrosine phosphorylation, then phosphoinositol generation, and ultimately serine/threonine phosphorylation. At each step along the way, checkpoints have been built up to prevent the system from going astray, much like the checkpoints that prevent neoplastic cell transformation. So, we have tyrosine phosphatases, inositol phosphatases, and serine/threonine kinases, the function of which is to dampen insulin signaling (49). Each one of these checkpoints should be viewed as a therapeutic target opportunity to reverse insulin resistance. It should also be pointed out that mouse genetics has enabled us to sift through many different genes with seemingly related tasks to identify those that are relevant to insulin action.

The theme of this review is to emphasize how a panoply of tissues contributes to metabolic control. I have grouped them under the moniker of “the republic of insulin resistance,” and in doing so I probably overstepped the modesty of nature, since some of these interactions may reflect more the experimenter’s ingenuity than real-life physiologic drama. Practitioners of basic research bear an eerie resemblance to the character of Mephistopheles in Goethe’s Faust, as they conjure up in gruesome detail the spirits of spike-headed flies, ageless worms, and knockout mice (Mephistopheles conjures them up as purveyors of mendacity, not a good precedent for scientists), but it is important to emphasize that sorting out gene function in mammals is no easy task, and my generation is fortunate to be active in science at a time when so much has become technically possible.

I would like to thank the American Diabetes Association for their support. I also would like to acknowledge the inspiring guidance of my mentor, Simeon I. Taylor, as well as Jesse Roth, Phillip Gorden, and Derek LeRoith. The work described in this review is the product of the dedication of doctoral and postdoctoral students and fellows who worked in my laboratory: Fredly Bataille, Martha H. Cool, Giovanni Di Cola, David Frescas, Donatella Gniuli, Carla Hanash, Marta L. Hribal, Hiroko Kanno, Yoshiaki Kido, Jane J. Kim, Tadahiro Kitamura, Yukari Kitamura, Davide Lauro, Jun Nakae, Anindita Nandi, Haruka Okamoto, Francesco Oriente, Byung-Chul Park, Kristina I. Rother, and Paola Salvatore. I want to acknowledge colleagues at other institutions who have provided key reagents and advice. Finally, I would like to acknowledge the commitment and generosity of the Naomi Berrie Diabetes Center and especially of the late Gerald Carrus.