Orchestration of Glucose Homeostasis

From a Small Acorn to the California Oak

  1. Richard N. Bergman
  1. From the Keck School of Medicine, University of Southern California, Los Angeles, California
  1. Address correspondence and reprint requests to Richard N. Bergman, PhD, Keck Professor of Medicine, and Chair, Department of Physiology and Biophysics, Keck School of Medicine, MMR626, 1333 San Pablo St., Los Angeles, CA 90033. E-mail: rbergman{at}usc.edu

Type 2 diabetes is increasing at alarming rates in the U.S., in Westernized countries, and in the third world. The increasing costs in terms of human suffering as well as economics are well recognized (1). Improved understanding of the pathogenesis of diabetes should lead to better approaches to predict, forestall, or even prevent diabetes and to treat extant cases. Yet, the precise causes of this disease remain to be totally explained. There is little question that obesity per se is a primary contributor, but the causes of the so-called “obesity epidemic” are in debate (2), as are the mechanisms by which obesity itself is linked to diabetes (3,4).

For some years, our laboratory has been engaged in efforts to understand causation of type 2 diabetes in terms of insulin secretion and resistance and to attempt to account for the role of obesity. The Banting Lecture provided me with a rare opportunity to describe our research efforts done over a prolonged history in a single presentation. It was my goal to address the several areas of diabetes research we have done. At first glance these different areas may appear unrelated. But these lines of work are closely linked within an overall effort to understand diabetes. I am deeply grateful to the members of the American Diabetes Association for the great privilege of making this presentation. I hope I will be able to emphasize the wonderful opportunity that I and my colleagues have had to pursue original research in a highly deserving and important cause. One can scarcely ask for more in one's professional life.

For several decades it has been popular to pursue a “reductionist” approach to studying disease (Fig. 1). The underlying concept is that understanding of disease will result from describing events at increasingly more microscopic levels: (organism→organ→receptor→organelle→substrates→pathway→enzyme→gene).

This reductionist approach emerged naturally from the revolutionary development of molecular biology and molecular genetics. A plethora of animal models of obesity and/or hyperglycemia have been produced (58). It is possible to sample tissues from such animal models and study function in vitro. From the reductionist approach has emerged the well-described signaling pathways by which insulin binds to cells and performs multiple functions, including stimulating glucose disposal, reducing lipolysis, enhancing lipogenesis, and enhancing protein synthesis (912). Equally important have been studies of the molecular events leading to the synthesis and secretion of insulin and other important glucoregulatory peptides (glucagon, glucagon-like peptide-1 [GLP-1], glucose-dependent insulinotropic peptide [GIP], and catecholamines). But it is becoming increasingly clear that the massive amount of information provided by the molecular approach may not, in itself, allow us to explain disease. Approaches to integrate this massively increasing body of material using computer models are emerging with the moniker “systems biology” (13), but principles of systems biology have long been applied to physiological regulation. Has it been possible to apply concepts of systems biology to help us understand the causes of diabetes?

Using the systems approach, mathematical or computer models are constructed based upon known intra- and/or inter-organ signaling patterns. The models are tested by experiment, validated, or altered, and the parameters of the model, which represent real physiological processes, can be measured from experimental data. Such an approach can potentially lead to an understanding of the importance of inter-organ communication with blood glucose regulation and can tell us how a breakdown in this communication breeds disease. But how should such a model be constructed?

SYSTEMS ANALYSIS AND DIABETES

At this juncture it remains a daunting challenge to convert the mountain of data emerging from the new technologies to yield deep understanding of diabetes. In the course of dealing with these newer technologies, it will be important to reduce the data to a simplified comprehensive portrait amenable to clinical interpretation. It is critical to remember the dictates of William of Occam (“Occam's Razor”), “entia non sunt multiplicanda praeter necessitatem,” which translates to “entities should not be multiplied beyond necessity” (14). Albert Einstein reminds us to represent physical reality in its simplest form—but not too simple (!!).

Some years ago we attempted to apply the principle of “Occam's Razor” to describe glucose control (15). Having measured glucose and insulin levels in blood minute by minute after intravenous glucose, we selected a deceptively simple, but physiologically based, model that could totally and accurately account for the glucose and insulin kinetics measured after intravenous glucose injection (1619). This “minimal model” cannot be said to be the only representation that could be acceptable, but it has survived several decades of investigation and probing by a significant number of independent investigators, including mathematicians (2022). Every model is a hypothesis and is therefore imperfect by definition. But, given its survival, we may conclude that the minimal model is a reasonably accurate representation of glucose regulation.

The minimal model.

The minimal model is represented in Fig. 2A; Fig. 2B presents a diagrammatic representation of the model revealing the underlying assumptions—the minimal assumptions required to describe glucose and insulin kinetics after injection. The required processes included the following:

  • Under overnight fasting conditions, endogenous hepatic (and renal) glucose production is balanced by basal glucose utilization by brain (50% of total) and other tissues.

  • Prandial elevation in glycemia is renormalized in part by glucose's ability to enhance its own disposal and suppress liver production (16,2325) independent of the plasma insulin response. This “glucose effectiveness” is reduced in type 2 diabetes (26).

  • Intake of carbohydrate elicits a prompt insulin response critical to glucose renormalization. Importantly, the renormalizing effects of insulin on glucose production and uptake are not manifest immediately, but modeling required a significant temporal delay in insulin's actions. Trying to explain this temporal delay in insulin's actions taught us much about insulin action in vivo (see below).

Has the minimal model been useful? Its original publication was hardly a barnstormer in that the model was hardly cited in the literature for 5 years after appearing. What appears to have increased its significance since are the following. 1) The model was validated by several independent laboratories (2729). 2) The model, coupled with experimental work, helped increase understanding of physiology and pathophysiology. 3) The model has led to clinical tools useful in the clinic and in epidemiological and genetic studies (3032).

Emergent research.

We can identify three main research “branches” which represent offspring of the minimal model (Fig. 3):

  1. Applying the model resulted in the simultaneous measurement of insulin sensitivity and insulin response from a single clinically applicable protocol. These measurements led us to define the “disposition index,” an important index of β-cell functionality and predictor of impaired glucose tolerance and diabetes (1618,33).

  2. Studying the aforementioned insulin action delay revealed the importance of insulin movement across the capillary endothelium (and possibly capillary recruitment) in insulin action and insulin resistance.

  3. An unexpected temporal delay in insulin suppression of liver glucose production evoked the role of free fatty acids as a mediator of insulin action on the liver.

The present exposition will follow the branches of the metaphorical tree shown in Fig. 3. I will describe our journeys to follow up the clues that the minimal model has provided us to climb branches that hopefully have increased to a measurable extent our understanding of what causes type 2 diabetes and what may be done to try to predict, prevent, and eventually cure it.

BRANCH #1: Clinical tools and the disposition index

After years of often contentious debate, a consensus has emerged regarding the pathogenesis of type 2 diabetes (online appendix Fig. A1 [available in an online appendix at http://dx.doi.org/10.2337/db07-9903]) (34,35). Insulin resistance is an important risk factor, necessary but not sufficient to cause the disease. There may be a genetic component to resistance (36), but many nongenetic factors can contribute, including obesity (particularly truncal obesity), lack of exercise, onset of puberty, pregnancy, and a group of conditions and therapies such as polycystic ovarian syndrome, infection, and treatment for HIV. We are now aware that a reduced compensatory response of insulin secretion in the face of insulin resistance is necessary for the development of frank type 2 diabetes. To understand the pathogenesis of diabetes, both insulin resistance and β-cell function must be assessed. One approach to assessment is the intravenous glucose tolerance test, which is described as follows.

In a healthy individual, upon glucose injection, glucose rises immediately to an early peak due to distribution in the extracellular fluid (online appendix Fig. A2). Immediately after, glucose begins to return toward basal levels due to glucose's own effect to enhance cellular uptake (due to mass action as well as glucose transporter mobilization to the membrane) (37) and glucose's ability to suppress hepatic glucose output (“autoregulation”) (23,24). We termed these two latter effects of glucose per se “glucose effectiveness” (16). The attendant hyperglycemia simultaneously provokes the β-cells to release insulin; soon the secreted insulin migrates from the plasma to the interstitial space of skeletal muscle, binding to insulin-sensitive cells and enhancing glucose disposal. The latter insulin effects accelerate the renormalization of glucose; in sensitive individuals glucose may slide below basal before renormalization between 3–4 h. Accounting for the measured time courses of glucose and insulin data with a digital computer (online appendix Fig. A2) yields estimates of the physiological coefficients of the model. Such fitting can be done even for a single individual. With Ray Boston and colleagues at the University of Pennsylvania, we have provided software, “MINMOD Millennium” (copyright R.N. Bergman) (38), that allows for this computer “fitting” process in a friendly computer environment. The resulting metabolic profile allows us to distinguish among different individuals, groups, or populations in terms of metabolic function (31,3942).

Physiologic parameters from the minimal model.

What are the physiological coefficients that emerge from fitting the minimal model to real data from an individual? The most significant ones, along with normal glucose tolerance, impaired glucose tolerance, and diabetes values, are listed in Table 1.

Applications of the minimal model.

The model has been widely applied to physiological and pathophysiologic situations, in animals and in human subjects. A cursory search of the literature reveals nearly 900 minimal model publications about widely varying subjects, ranging from arcane mathematical treatments (2022) to studies in animals (23,43,44) and humans (29,31,45). Many of these applications have employed the frequently sampled intravenous glucose tolerance test to yield insulin sensitivity and secretory response. The use of the minimal model is continuing. It yields a useful metabolic profile that can be applied to physiologic and pathophysiologic studies, comparison of therapeutic regimens, population dynamics, and population genetics.

Disposition index.

We hypothesized that in a normal individual, insulin resistance would be compensated by increased insulin secretion and that this compensation explains why normal glucose tolerance can be maintained in the face of even large changes in insulin sensitivity due to environmental or other nongenetic factors. This remarkable ability of the β-cells to precisely compensate is indicated in online appendix Fig. A3, in which a high-fat diet in the normal dog model induces severe insulin resistance and frank hyperinsulinemia but absolutely no change in either fasting or 24-h glucose excursions.

One significant outcome of minimal model studies was an ability to express in quantitative terms the compensatory relationship between insulin resistance and insulin response. Again we were inspired by an engineering concept, that of “closed loop gain.” The regulation of the blood glucose concentration represents a classic closed-loop endocrine feedback system: nutrient ingestion increases glucose, leading to a reflexive stimulation of insulin release, which counters the tendency to hyperglycemia. The ability of this closed-loop system to normalize glucose quickly can be expressed as the product of the release of the signal, plasma insulin response, and the effect of that signal, insulin sensitivity. We therefore defined the disposition index (DI) as a measure of the overall ability of the glucose regulating system to renormalize glycemia after perturbation by nutrient intake (16). The DI was defined as follows: Formula where AIR is the acute insulin response. There is strong evidence for increased plasma insulin in the face of insulin resistance, at least in normal individuals (46). Therefore, the DI can be interpreted as the ability of the glucose regulating system to compensate for insulin resistance by increasing plasma insulin. We hypothesized that in normal individuals, a reduction in SI (insulin resistance) would be compensated by an equivalent increase in plasma insulin response, such that the DI remains relatively constant (Fig. 4, ↓ SI compensated by ↑ in AIRGLUCOSE). The latter concept, implemented as Eq. 1, has been referred to as the “Hyperbolic Law of Glucose Tolerance” (47). Online appendix Fig. A4 shows the constancy of the DI in the face of insulin resistance.

Interpretation of DI.

Confusion regarding the relative importance of β-cell dysfunction to type 2 diabetes resulted from the fact that hyperinsulinemia often accompanies insulin resistance, even in the face of a latent β-cell defect. Thus, an individual with reduced DI may increase plasma insulin in the face of insulin resistance (Fig. 4), but the degree of compensation may be less than expected if the β-cells were healthy. Thus, while impaired glucose tolerance is often associated with increased risk for diabetes, and with β-cell dysfunction, the degree of dysfunction is represented by a reduced DI compared with healthy individuals. Therefore, it is reasonable to suggest that reduced DI is a harbinger of type 2 diabetes. In the Pima Indians, Weyer and colleagues identified the DI as the strongest predictor of conversion from normal glucose tolerance to type 2 diabetes in a longitudinally observed population over a period of 5 years (48), and a similar result emerged from the Malmo Prospective Study of pre-diabetes (49).

There is an emerging consensus that reduced β-cell function as reflected in the DI is the strongest predictor of type 2 diabetes in at-risk populations. But does this reduced islet function have a genetic component?

Genetics of DI.

Type 2 diabetes has higher concordance in monozygotic versus dizygotic twins (50), suggesting a strong genetic basis for this disease. Numerous studies have attempted to identify the specific inherited metabolic function(s) or gene(s) that might be responsible. Heritability calculations have supported indices of β-cell function as being important. The heritability of DI has been estimated to be as high as 0.73 in an African-American group (36). Poulsen and colleagues used state-of-the-art methodology to measure a variety of metabolic parameters in monozygotic and dizygotic Danish twins (50). They reported heritabilities for the DI of 0.75 in younger and identical twins but much lower values in younger fraternal sets (0.30). These data strongly support the inheritance of β-cell function and suggest that it is inheritance of reduced DI, which may contribute to increased genetic risk for type 2 diabetes.

It is not known what gene or genes may explain inheritance of DI. One clue has emerged from two disparate population studies—the FUSION Study of type 2 diabetes in Finland and the IRAS Family Study of several ethnicities in the U.S. (51,52). Both studies have identified a locus on chromosome 11 related to diabetes risk and linked with the DI. Given the heritability of DI and similarity of these loci—logarithm of odds score of 4.80 at 80 cM for linkage to the DI in the IRAS Family Study and predisposition to type 2 diabetes at a similar locus in the FUSION Study—it is tempting to hypothesize that there may be a gene for DI in this region of chromosome 11 that predisposes to diabetes. Until the putative gene for DI and/or diabetes on chromosome 11 is identified, it will remain unknown what is the true significance of this interesting region for diabetes inheritance. Very recently, several groups have reported 10 genes emerging from genome-wide association studies that increase risk for type 2 diabetes (5355). While the function of the genes are not all known, several appear to be related to β-cell function.

What is the mechanistic explanation for the “Hyperbolic Law of Glucose Tolerance”?

It would seem obvious that the increase in insulin response that attends onset of insulin resistance might be explained by the sequence of events of online appendix Fig. A5; that is, insulin resistance leads to a small reduction in glucose tolerance, mild hyperglycemia, and resultant increase in β-cell mass and/or sensitivity to glucose stimulation. The glycemic hypothesis has been strongly supported in a recent publication (56). However, there is evidence that glycemia per se may not play a primary role. In fact, in the face of obesity induced by either elevated fat-eucaloric or elevated fat-hypercaloric diets, we observed the expected increased body weight, insulin resistance, and hyperinsulinemia and relatively constant DI (online appendix Fig. A4) (43,57). Surprisingly, we did not observe any increase in fasting hyperglycemia. Not only were fasting glucose levels either constant or slightly reduced by the fat intake regimen, but there was absolutely no change in 24-h glucose values (online appendix Fig. A6) (58). Neither did we observe increases in other putative signals, such as cortisol or active GLP-1. These data indicate that in the conscious dog model, glucose cannot be the signal responsible for β-cell upregulation in the face of fat-feeding/obesity-induced insulin resistance. Neither can steroids or the gut peptide GLP-1 play this important role. As will be discussed below, there is stronger evidence that free fatty acids (FFAs) may play the central role in accounting for the increase in insulin secretory response in the face of nutrient-induced insulin resistance. However, at this moment in time, despite extensive evidence for the hyperbolic relationship between insulin sensitivity and insulin secretion (17,45,59,60), the precise signaling accounting for this fundamental physiological relationship remains unexplained! Needless to say, we continue the search for the mechanism(s) that may explain the “Hyperbolic Law of Glucose Tolerance.”

BRANCH #2: Transendothelial transport and insulin resistance

As discussed above, one requisite assumption in the minimal model is the sluggish or delayed effect of insulin to stimulate glucose disappearance. This model result recapitulated conclusions from Andres, Sherwin, and colleagues (61) from the first clinical glucose clamp studies, showing that the rapid increase in plasma insulin by injection resulted in a slow enhancement of glucose disposal, which did not reach steady state before 3 h. Online appendix Fig. A7 compares the rapid effect of insulin in vitro to stimulate glucose uptake by adipose cells, with the sluggish effect in vivo. We were interested in understanding the physiologic basis underlying this slow insulin effect, as well as understanding its possible role in insulin resistance and diabetes pathogenesis.

Once secreted, insulin follows a tortuous path to stimulate glucose uptake. Entering the bloodstream from the pancreatic veins, it must survive passage through the liver (at least 50% of secreted insulin does not) and travel from the venous system to muscle or adipose capillaries rather impermeable to large proteins. The large protein molecule must either pass through capillary endothelial cells or between them through paracellular routes to enter the interstitium, diffuse to target cells, bind, and act via the insulin pathway. The delay in insulin action might be due to slow biochemical steps after receptor binding; alternatively, it may be the delivery of insulin to the sensitive cells that is slow (62). To examine the transendothelial transport process, and to compare the time course of plasma insulin with that of interstitial insulin and glucose disposal, we exploited the use of lymph sampling to access the interstitial fluid. Our studies showed clearly that there is a strong “hand-in-glove” temporal relationship between interstitial insulin and glucose disposal but a weak relationship between plasma insulin and glucose uptake (63,64). We concluded that the delay in insulin action observed in the glucose clamp experiments and implemented in the minimal model is due to sluggish movement of insulin across capillary endothelium between plasma and interstitium; but, once at the cell surface, insulin binds and acts almost immediately (as observed in vitro, Fig. 5).

Implications of sluggish transendothelial transport of insulin.

When stimulated by rapid changes in plasma glucose, insulin release from the β-cells is biphasic. This biphasic pattern, originally described by Grodsky and colleagues (65,66), has been much studied and reflects two pools of releasable insulin within the β-cells, which have recently been imaged directly (67). It is interesting to contemplate the possible significance of biphasic release with regards to overall glucose regulation. Is it possible that biphasic secretion evolved to compensate for slow transendothelial insulin transport, which temporally limits insulin's access to skeletal muscle and adipose tissues?

Lisa Getty, when in our laboratory, injected insulin intravenously to simulate the first-phase insulin release (online appendix Fig. A8). A quick but short first-phase insulin pattern obviates the usual delay in insulin action (seen during glucose clamp experiments, for example), resulting in a rapid and profound increase in glucose disposal (68). It is tempting to hypothesize that the first phase of insulin release was naturally selected to accord the most rapid glucose disposal after nutrient ingestion and to limit the postprandial glycemia. Such a supposition can never be proved, but it does provide a potentially satisfying explanation for the evolution of a rapid first-phase insulin response.

A very important but still unanswered question relates to the putative importance of transendothelial transport in pathogenesis of insulin resistance. Clear biochemical defects have been identified in insulin-resistant animal models and humans, including receptor downregulation, phosphatidylinositol 3-kinase defects, reduction in mitochondrial function, and reduced numbers of glucose transporters. But, is there a defect in delivery of insulin to the sensitive cells?

Martin Ellmerer, in our laboratory, worked with Joyce Richey and used compartmental modeling to analyze distribution of insulin in animals rendered obese with a high-fat diet (69). Their analysis suggested that obesity limited access of insulin to skeletal muscle, accounting for about one-half of the insulin resistance caused by high-fat feeding. Very recently, to examine the possible effects of obesity on insulin kinetics more directly, Jenny Chiu worked with Richey to examine the movement of insulin within the interstitial compartment of muscle tissue. These investigators injected insulin directly into muscle and measured insulin in the interstitium (lymph) and glucose disposal by the hindlimb. We found that glucose uptake by skeletal muscle in normal animals is much more sensitive to the hormone than is suggested by systemic measurements and glucose clamps (Vmax ∼22 mg · min−1 · leg−1 and an ED50 of ∼120 pmol/l). The latter result implies that availability of insulin to the sensitive tissue is limited by either transport across endothelium or distribution of blood flow between “nutritive” versus “non-nutritive” tissues, as suggested by Clark, Barrett, and their colleagues (70,71). Even more interesting is the recent result of Chiu and colleagues (69) that infusion of the lipid emulsion Liposyn, which raises peripheral FFA levels, appears to absent insulin from skeletal muscle tissue—when Liposyn is infused, insulin appears to exit skeletal muscle rapidly, as there is virtually no increase in interstitial insulin despite direct injection of the hormone into muscle tissue. The latter result suggests rapid washout of insulin from skeletal muscle, which could suggest a great increase in permeability of capillary endothelium to the insulin molecule. It is clear that availability of insulin to the sensitive tissues (muscle and adipose) is a major factor determining insulin sensitivity. Thus, conclusions regarding insulin action from measurements of cellular and subcellular function alone cannot reproduce insulin's action in vivo, and the availability of insulin at skeletal muscle deserves further study.

The slow action of insulin, revealed by minimal modeling and glucose clamp experiments, have led to interesting and potentially important insights regarding the action of insulin and mechanisms of insulin resistance. Given that insulin resistance is an important factor, albeit not the only factor in pathogenesis of type 2 diabetes, further study of insulin access to skeletal muscle and adipose tissue is justified.

BRANCH #3: The role of FFAs in insulin action and glucose homeostasis

In normal individuals, several factors guarantee that the blood glucose level is normalized rapidly after meals. In dogs in particular, this normalization process is remarkable, as there is little increase in the blood glucose after glucose ingestion, despite the large increase in glucose turnover (72). Rizza and his colleagues (73) have examined the plethora of factors that play a role in this normalization, which involves suppression of endogenous glucose output (primarily by liver) and enhancement of glucose disposal (primarily by skeletal muscle). One of the most interesting factors is the gut peptide GLP-1, which has several effects to normalize glucose: slowing of gastric emptying, enhancement of the plasma insulin response, and suppression of glucagon (74). Viorica Ionut, in our laboratory, and others have exciting results pointing to an additional and potentially potent effect of GLP-1 to increase glucose effectiveness (7577), possibly via GLP-1 receptors in the porta hepatic circulation, via the central nervous system (CNS) to muscle (Fig. 6). Evidence emanating from Rossetti's laboratory that glucoreception in the CNS may control hepatic glucose output by liver directly (78) adds to the concept that the CNS is more important than has been previously thought (in glucoregulation).

Intravenous administration of glucose (as in the frequently sampled intravenous glucose tolerance test, for example) bypasses effects of gastrointestinal peptides, and yet glucose is rapidly normalized. The euglycemic glucose clamp examines effects of insulin per se on normalization and, as discussed, reveals slow activation of glucose uptake (online appendix Fig. A9). Traditional thinking presumed that suppression of glucose output, unlike activation of disposal, would be very rapid, as insulin secreted by the β-cells has immediate access to the liver via the portal vein. In addition, portohepatic vessels have large fenestrations, allowing insulin entering the liver from the portal vein to access and bind to hepatocytes almost immediately after appearance in the liver. Scintigraphy confirms very rapid concentration of injected insulin molecules in liver tissue (79). Yet, in modeling we were able to account for glucose kinetics by making the counterintuitive assumption that insulin action to suppress endogenous glucose output was sluggish, similar to the effect of stimulation of glucose disposal. This latter assumption was ultimately confirmed by David Bradley in our laboratory using euglycemic glucose clamps (80)—stimulation of glucose uptake and suppression of production were equally slow, with almost identical kinetics (i.e., similar t1/2). This latter result led to the counterintuitive hypothesis that the effect of insulin to suppress glucose output by the liver was an indirect effect of the hormone (i.e., mediated by one or more insulin-dependent extrahepatic signals).

Search for the signal mediating insulin action.

Several signals seemed possible candidates for mediating the apparent indirect insulin effect (online appendix Fig. A10). Further investigation utilizing the glucose clamp revealed a remarkable similarity between insulin suppression of glucose output and insulin suppression of lipolysis from adipose tissue (81,82). It was known that 1) the delay in insulin effect on disposal was due to slow transendothelial transport. From early studies of Scow and colleagues (83), it was also known that 2) access of insulin to adipocytes (similar to skeletal muscle access) was relatively slow. We attempted to explain these results by the “single gateway hypothesis” (84) (online appendix Fig. A11). It was posited that the slow effects of insulin on production and disposal were explained by slow transport into muscle and adipose tissue. The former slowed increasing glucose disposal, and the latter slowed suppression of FFA, hence slowing the suppression of glucose output. There is still debate regarding what precise fraction of suppression of glucose output is secondary to suppression of FFA (85). However, there is consensus that FFA suppression accounts for at least part of the reduction in gluconeogenesis and glycogenolysis that accompanies nutrient intake. What was important to our laboratory was the realization that FFAs play a more important role than previously appreciated in glucose homeostasis (7). A similar message was emanating from other laboratories (86). Adding to this understanding was epidemiologic evidence for the importance of adiposity—especially truncal adiposity—in the development of insulin resistance and as a risk factor for type 2 diabetes and cardiovascular disease. To further examine the role of FFA, and adiposity per se, we developed a canine model of obesity.

The role of FFA in pathogenesis of the metabolic syndrome.

It is a truism that obesity is increasing in the U.S. and other Western countries and is also increasing in less–well-developed nations (87). Obesity is an important risk factor for type 2 diabetes due to its close relationship with insulin resistance. Yet, the mechanistic relationships among obesity, insulin resistance, and diabetes are not totally clarified. There is strong epidemiologic evidence that central adiposity, in particular, carries risk. Experimental studies are needed to explain the causal relationships between central adipose depots and risk.

There are particular advantages in studying obesity in large animal models. Insulin resistance and obesity are characterized by communication among different tissues such as adipose and liver, and in the dog model it is possible to access the abdominal portal vein reflecting visceral fat signaling to the liver. In addition, in the canine model it is possible to study the development of obesity longitudinally (or reversal thereof) while making repetitive metabolic measurements. One such study followed the time course of insulin resistance, insulin secretion, and insulin clearance in the canine model fed a diet with elevated fat content (online appendix Fig. A3) (43). It is noted that there was a reduction in insulin sensitivity that was followed by a slow increase in insulin response, which reached a peak at 6 weeks. Complimenting increased insulin response was a reduction in first-pass clearance of insulin by liver, which accounted for at least as much hyperinsulinemia as increased insulin release (43,88). As discussed, despite these significant changes, there was no detectable change in DI. What mechanisms account for this well-orchestrated response to lipid intake without changes in glucose tolerance in the normal animal?

Nuclear magnetic resonance confirmed a significant deposition of lipid in both the central (omental) and subcutaneous fat depots during fat feeding (online appendix Fig. A12). Very recent data obtained by Morvarid Kabir in our laboratory confirms that while there are increases in adipose deposition in visceral as well as subcutaneous depots in fat-fed dogs, the visceral depot is unique. As previously reported (89,90), we noted that visceral fat tissue is more sensitive to adrenergic stimulation of lipolysis (ED50 of 1.31 × 10−7 mol/l compared with 2.77 × 10−7 mol/l for subcutaneous adipose tissue). Very interesting is the appearance in the enlarged omental fat depot of a unique population of “new” smaller adipocytes, suggesting conversion of preadipocytes to fully developed adipocytes preferentially in the visceral depot. Euglycemic clamp results revealed that most of the developing insulin resistance after feeding the eucaloric high-fat diet was due to resistance of the liver (91). We observed failure of insulin to suppress endogenous glucose output before significant changes in peripheral insulin sensitivity, measured as insulin's action to stimulate glucose disposal. Assessment of expression of genes in visceral tissues as well as liver revealed a pattern that favored increased turnover of omental fat, as well as enhanced hepatic gluconeogenesis (92)—changes that favor an effect of FFA, released from the visceral adipose depot, to cause hepatic insulin resistance (online appendix Fig. A13). However, in this model of modest obesity, we did not observe changes in expression of genes for “adipokines” including tumor necrosis factor-α, interleukin-6, leptin, or adiponectin. Therefore, central obesity induced by 6 weeks of elevated fat causes liver insulin resistance and hyperinsulinemia with a pattern supporting increased flux of FFA from visceral depot to liver but without measurable changes in expression of adipokines in visceral adipose tissue. These data support an important role for FFA release from the visceral fat in the pathogenesis of insulin resistance associated with increased truncal lipid deposition.

Access to the portal vein allowed us to measure the rate of release of FFA from the visceral depot. To our surprise, we found that visceral FFA release was oscillatory, with a burst observed about every 9–11 min (online appendix Fig. A14) (93,94). That the release was lipolytic was supported by a similar and coordinated pattern of glycerol release. Pulsatile visceral lipolysis was totally suppressed by bupranolol, a high-affinity antagonist to β-3 adrenergic receptors, located in the visceral fat depot of the dog. Recent data obtained by Isabel Hsu in our laboratory have supported the concept that oscillations imposed in the portal vein of the conscious animal have profound effects to enhance insulin resistance of the liver.

Our data has led us to support the original concept of Landsberg (95) that the sympathetic nervous system plays a central role in the development of insulin resistance, secondary to deposition of visceral fat. We suggest that fat feeding results in deposition of visceral fat. Adrenergic signals from the CNS induce phasic lipolysis of the visceral depot, which bathes the liver with FFA at regular intervals (online appendix Fig. A15). This FFA barrage induces insulin resistance in the liver by upregulating gluconeogenic enzymes. Further studies are being performed to address the source of the central adrenergic signals, how they may depend upon metabolic state, and what role the signals may play in other insulin-resistant conditions.

Hyperinsulinemic compensation for insulin resistance: putative role of nocturnal FFAs.

As discussed, it remains unexplained why in the face of insulin resistance, plasma insulin response increases and insulin clearance decreases in a well-coordinated manner such that, in normal animals, glucose intolerance does not invariably result in the insulin-resistant state. How do the β-cells of the pancreatic islets “know” the appropriate enhancement of insulin release? How does the insulin-degrading mechanism of the liver establish the appropriate downregulation of first-pass liver insulin clearance? As discussed above, careful investigation of glycemia failed to explain insulinemic upregulation. Is there a possibility that FFAs are involved in this highly regulated inter-organ orchestration?

We considered a group of blood-borne signals that could be put forth as mediating hyperinsulinemia in the face of fat-diet–induced insulin resistance (58). Among these candidates were GLP-1, known to cause pancreatic islet-cell proliferation in rodents, and growth hormone and cortisol, each of which can result in increased insulin. After 6 weeks of a high-fat, hypercaloric diet, resulting in significant weight gain in the dog model, we observed zero evidence for any increase in 24-h glycemia (online appendix Fig. A6). Even more surprising, we measured a paradoxical reduction in 24-h active GLP-1 levels and no increases in either 24-h patterns of cortisol or growth hormone. On the contrary, and to our surprise, we observed a striking and powerful increase in plasma FFA levels in the middle of the night. Comparing the fat-fed animal with the lean model, FFAs are increased beginning in the late afternoon and continue to elevate with a maximal elevation at 3 a.m. What is the significance of the nighttime FFA increase, and does it relate to our sympathetic hypothesis of the causation of the metabolic syndrome?

We are attempting to understand the significance of the nighttime FFA increase. We would propose that it represents lipolysis, stimulated by the sympathetic nervous system. If so, the increasing FFA may be oscillatory, and adrenergic blocking agents should suppress them specifically. While it is indeed tedious to do rapid sampling to establish oscillatory lipolysis, students in our laboratory are doing these studies, led by Stella Kim, Isabel Hsu, Jenny Chiu, and Karyn Catalano. We plan to test whether the lipolysis can be blocked with adrenergic blocking agents, and if so, whether this will reverse the putative effects of elevated nocturnal FFA on insulin secretion and action.

Hypothesis integrating the role of FFAs in the metabolic syndrome.

Our working hypothesis for the role of the visceral fat depot in the development of the metabolic syndrome (Fig. 7) is as follows: We suggest that the adrenergic nervous system plays a central role in the development of the syndrome. We suggest that fat feeding increases the signaling from the brain to the adipose depot and that this signal is particularly evident at night, maximized at 3:00 a.m. We further suggest that the sympathetic signal results in bursts of FFA via the portal vein, bathing the liver in lipid and rendering it insulin resistant. The resistance, at least in the short term, does not appear to involve changes in adipokines, although it is certainly likely that adipokines are important over the longer term. The resistance is associated with increase of expression of enzymes that favor lipolysis and enhancement of gluconeogenesis. It is possible that interfering with this pattern, for example by suppressing lipolysis during the nighttime, may break this pattern and reverse the insulin resistance of the metabolic state.

THE TREE, ITS RICH SOIL, AND ITS FERTILIZERS

I have been a very lucky person. With my wonderful colleagues (see acknowledgments), we have been able to begin with an arcane idea—development of a minimal model of glucose kinetics—and to follow this “acorn” where it led us, hopefully to some greater understanding of the pathogenesis of diabetes and associated metabolic diseases. Trees will not flourish in the wilderness without rich soil and fertilizer. First is my wife Ronni and my kids, Doug and Beth, their spouses, Nancy and Guy, and my grandchildren, Emily, Jessica, Samantha, and Hannah.

I am a lousy tree climber. I received very necessary “legs up” from many wonderful people and institutions: My early mentors included Oscar Hechter, I. Arthur Mirsky, and John Urquhart. Dan Porte, Jr., Mladen Vranic, and Larry Phillips were kind enough to appreciate our early research efforts and enabled us to test our ideas in the clinic. My wonderful peers Claudio Cobelli and Diane Finegood were there at the tree's conception, birth, and early life, and Marilyn Ader, Richard Watanabe, Tom Buchanan, Joyce Richey, and Andrea Dunaif were present during the Cambrian adolescence and adulthood. I am tremendously grateful to the wonderful friends/colleagues who chose to work with me on our journey, listed by necessity in the acknowledgments. Finally, I thank the American Diabetes Association, the National Institutes of Health, and the University of Southern California, all of whom translated their confidence in our work to the tangible resources without which our work not have been performed.

Beloved colleagues

Richard Watanabe, Diane Finegood, Giovanni Pacini, Claudio Cobelli, Jay Taborsky, Steve Mittelman, Martin Ellmerer, Dave Bradley, Aage Volund, Joyce Richey, Gianna Toffolo, Kerstin Rebrin, Katrin Hücking, Idit Liberty, Vivi Ionut, Mori Kabir, Chester Ni, Pat Crane, Marianthe Hamilton-Wessler, Lisa Getty, Dave Cohen, Melvin Dea, Jang Youn, Andrea Hevener, Gregg Van Citters, Casey Donovan, Garry Steil, Y. Ziya Ider, Ray Boston, Charlie Bowden, Renee Poulin, Lise Kjems, Stella Kim, Karyn Catalano, Isabel Hsu, Jenny Chiu, Darko Stefanovski, Maya Lottati, Nicki Harrison, Orison Woolcott, Dan Zheng, Elza Demirchyan, Rita Thomas, Ed Zuniga, Erlinda Kirkman, IRAS Investigators, FUSION family, Anne Sumner, Steve Kahn, Michael Goran, Mike Schwartz, and Jim Best.

I am deeply grateful to Dr. Marilyn Ader for helping me with this manuscript and for many years of fruitful collaboration and camaraderie.

FIG. 1.

Reductionist vs. holistic approaches to study carbohydrate homeostasis. Elegant molecular and genetic approaches have elucidated cellular pathways of hormone action, including insulin (left) (ref. 12). Alternative approach (right) is to test the system in vivo and compare predictions with computer model to hone hypotheses of hormone/substrate interactions.

FIG. 2.

The minimal model of glucose kinetics. A: Insulin in plasma [I(t)] crosses the endothelial barrier to enter interstitial fluid (I’). Production of glucose (G) by liver and glucose disposal in periphery are controlled by “remote” (i.e., interstitial) insulin concentration. Later studies confirmed that slow effect of insulin on liver is due to indirect effect of the hormone to suppress glucose production (via free fatty acids; “single gateway hypothesis,” ref. 83). Adapted from ref. 16. B: Diagrammatic representation of the minimal model. Fasting blood glucose level is determined by a balance between liver production and insulin-independent (brain) and insulin-dependent (skeletal muscle) mechanisms. Carbohydrate intake elicits insulin signal; insulin slowly crosses capillary endothelium to enhance glucose disposal to muscle and suppress lipolysis in adipose, which in turn contributes to reduction in liver glucose output.

FIG. 3.

History of research in the Bergman Laboratory. Minimal model assumptions were tested experimentally, leading to the DI (Disposition Index) concept, the role of transendothelial insulin transport, and role of FFAs in the “Metabolic Xyndrome.”

FIG. 4.

The Hyperbolic Law of Glucose Tolerance. Insulin resistance (abscissa) is normally compensated by a reflexive hyperinsulinemia, to maintain glucose tolerance. The law states that in normal individuals, product of insulin sensitivity and secretion is constant—the DI. In at-risk individuals, DI may be reduced (lower hyperbolic curve). In such individuals, hyperinsulinemia for a given reduction in insulin sensitivity may be inadequate (lower DI). Lower DI may portent eventual type 2 diabetes.

FIG. 5.

Proposed events in insulin stimulation of glucose disposal. Insulin crosses from capillary to interstitial compartment slowly; delay may be due to intercellular or transcellular transport. Once in interstitium, insulin binds to receptors and rapidly stimulates glucose uptake.

FIG. 6.

Possible indirect effect of gastrointestinal peptide GLP-1. GLP-1 degrades rapidly in plasma, suggesting that it may have a site of action close to its site of secretion (L-cells of gut). Putative receptors in the portal vein (or liver) are sensitive to GLP-1 and glucose; interaction with receptors may send afferent signal to brain with efferent signals to enhance glucose utilization (e.g., increased insulin secretion, reduced glucagon, and/or increase in insulin-independent glucose uptake). Extant data show enhanced glucose disposal secondary to portal glucose/GLP-1 administration (ref. 73).

FIG. 7.

Events resulting in the “Metabolic Xyndrome.” CNS and other factors result in lipolysis from visceral and subcutaneous fat depots. Visceral fat depots in particular cause flux of FFA to liver and hepatic insulin resistance; more stored fat causes insulin resistance of skeletal muscle. In our experiments, compensation for insulin resistance (hyperinsulinemia) occurs in the absence of elevated glucose. FFAs themselves may be the compensating signal that causes the hyperinsulinemia by stimulating insulin release, increasing β-cell size or number, and reducing liver clearance of insulin. Hyperinsulinemia is adequate to compensate for insulin resistance, except in the face of a latent β-cell defect, wherein hyperglycemia ensues and type 2 diabetes results.

TABLE 1

Physiologic parameters emerging from the minimal model (data from refs. 39,42)

Acknowledgments

R.B.'s work has been supported by the National Institutes of Health (DK 29867 and DK 27619) and the American Diabetes Association (Mentor Award).

Footnotes

  • Additional information for this article can be found in an online appendix at http://dx.doi.org/10.2337/db07-9903.

    • Accepted February 20, 2007.
    • Received February 20, 2007.

REFERENCES

| Table of Contents