Cellular Energy Sensing and Metabolism—Implications for Treating Diabetes: The 2017 Outstanding Scientific Achievement Award Lecture

Over the past several decades a large body of evidence has found that insulin resistance is an important contributing factor to the development of type 2 diabetes. The causes of insulin resistance are multifactorial but are tightly linked to an energy surplus and subsequent increases in ectopic lipids, mitochondrial derivatives such as reactive oxygen species, and chronic low-grade inflammation (reviewed in Fu et al. [1] and Samuel and Shulman [2]). Collectively, these factors coalesce to induce a wide variety of stress-signaling pathways that impair the ability of insulin to stimulate glucose uptake into the muscle and fat and inhibit glucose production from the liver. Therefore, designing therapies aimed at reducing these cellular stressors may provide a means to improving insulin resistance in target tissues.

The Diabetes Prevention Program (DPP) showed that metformin delays the development of type 2 diabetes in people with insulin resistance (3). This study also showed that lifestyle intervention, involving exercise and caloric restriction that results in 7% weight loss, was also effective (3). On the basis of these findings, we thought that understanding the mechanisms of how metformin, exercise, and weight loss work to prevent type 2 diabetes may lead to new and possibly more effective therapeutic strategies. Interestingly, we know that metformin (4), exercise (5), and caloric restriction (6) all have one thing in common: they all increase the activity of the AMP-activated protein kinase (AMPK), an evolutionarily conserved enzyme found across phyla and in all cells of the body (Fig. 1). AMPK can be thought of as the body's fuel gauge, telling cells when they need to fill up with gas before they run out (7). Of course cells do not use gasoline, but instead the primary fuel or energy currency of the cell are the adenine nucleotides AMP, ADP, and ATP, which are produced from the combustion of fat and carbohydrates. Under conditions of energetic stress, we have greater AMP and ADP levels and lower levels of ATP in the cell. This results in increases in AMPK activity, which then switches on energy-producing processes, such as fatty acid oxidation and glucose uptake while also promoting the production of more mitochondria (8,9). At the same time, AMPK inhibits key energy-consuming enzymes that are not vital for the survival of the cell, such as the synthesis of protein, cholesterol, and fat (8,9). Ultimately, this activation of AMPK restores cellular energy balance. Importantly, metformin, exercise, and caloric restriction all induce a metabolic challenge that can increase AMPK activity, suggesting that it could be a key molecular link, explaining why these agents may delay the development of type 2 diabetes (8,9).

The liver plays a vital role in regulating whole-body glucose and lipid homeostasis. Numerous studies in humans have indicated that metformin primarily exerts its glucose-lowering effects by improving liver insulin sensitivity (10); however, the mechanisms by which this occurs were not fully understood. Nearly 30 years ago AMPK was shown to phosphorylate and inhibit acetyl-CoA carboxylase (ACC) (11). ACC exists as two isoforms that make the common metabolite malonyl-CoA from acetyl-CoA (Fig. 2). Malonyl-CoA is the first step in the synthesis of fat from glucose and is also an inhibitor of fatty acid oxidation. Given ectopic lipid accumulation in the liver is an important cause of insulin resistance (12), we hypothesized that metformin-induced activation of AMPK and subsequent phosphorylation and inhibition of ACC could potentially lower lipid levels and improve insulin resistance. We tested this idea by making mice lacking the key AMPK phosphorylation sites on ACC1 (Ser79) and ACC2 (Ser212) obese and then treating them with metformin. We found that, when given acutely, high doses of metformin lowered blood glucose independently of the AMPK-ACC signaling axis (13), a finding consistent with other reports indicating AMPK is not required for acutely inhibiting hepatic gluconeogenesis (14–16). However, when given chronically, at lower doses, to obese insulin-resistant mice, metformin increased ACC phosphorylation and lowered liver lipid synthesis, liver lipid content, and liver insulin resistance in control mice but not in mice without the AMPK phosphorylation sites on ACC (13). These data indicated that in the context of obesity and insulin resistance, chronic administration of metformin lowers blood glucose by reducing liver lipid synthesis and improving insulin sensitivity, and this is dependent on a single AMPK phosphorylation site on ACC1 and ACC2 (Fig. 2).

Given the importance of the AMPK-ACC signaling pathway for the chronic insulin-sensitizing effects of metformin, we next asked whether there might be other pharmacological agents that could be used to enhance the effects of metformin to exert additional beneficial metabolic effects. Metformin-induced activation of AMPK occurs indirectly through inhibition of mitochondrial function and subsequent alterations in adenine nucleotides that are sensed by the γ subunit of AMPK and results in enhanced phosphorylation by the upstream kinase LKB1 (17,18). In contrast, direct allosteric activators of AMPK, such as A769662 (19), increase AMPK activity through a specific residue (Ser108) with the AMPK β1 subunit (20). We found that by using A769662 and mice lacking the AMPK β1 subunit that AMPK inhibited macrophage inflammation and could reduce high-fat diet–induced insulin resistance (21). Given these findings and the previous observations of Shoelson and colleagues (22) showing that salicylate could reduce inflammation and insulin resistance, we began to wonder whether salicylate might also increase AMPK activity. In collaboration with Grahame Hardie (23), we found that therapeutic concentrations of salicylate, achievable through the delivery of salsalate (a dimer of salicylate which does not cause bleeding [22]), increased the activity of AMPK and that, surprisingly, the mechanisms by which it exerted these effects were identical to the direct AMPK activator A769662 (Fig. 2). In addition to this direct activation of AMPK (23), we also found salsalate could induce mild mitochondrial uncoupling that suppressed lipid synthesis (24) (Fig. 2). Importantly, these mechanisms for activating AMPK and lowering lipid synthesis were distinct from what we had found for metformin (13), suggesting that combining salicylate with metformin could potentially exert synergistic effects toward lipid synthesis and insulin sensitivity. Consistent with this hypothesis, we found that treating primary mouse and human hepatocytes with metformin or salicylate dose-dependently lowered lipid synthesis; however, when combined their effects were enhanced (25). These additive effects were also observed in vivo in obese mice where the effects of metformin and salsalate to lower liver lipids and improve liver insulin sensitivity were heightened by combining both drugs together (25) (Fig. 2).

In addition to liver insulin sensitization, metformin also exerts a wide range of beneficial effects that has prompted clinical trials in multiple diseases outside of type 2 diabetes, but how metformin exerts these systemic effects is not clear, as circulating concentrations and cellular accumulation outside the liver is quite low (26). We hypothesized that one way metformin could potentially induce beneficial systemic effects would be by promoting the production of a circulating factor that in turn might communicate with the rest of the body. To test this hypothesis, we assessed 284 circulating biomarkers in over 8,000 people with hyperglycemia and found that metformin dose-dependently increased serum levels of GDF15 (also known as macrophage inhibiting cytokine 1 [MIC-1]) (27). Interestingly, this effect of metformin on increasing GDF15 was independent of blood glucose and was not seen with other glucose-lowering therapies, and, importantly, the effect size was very high (∼4.0) compared with other biomarkers such as glucagon-like peptide 1 (∼1.5), which were already known to be increased by metformin (27). So what is GDF15 and why might it be important for the beneficial effects of metformin? GDF15 is a member of the TGFβ family whose expression is upregulated in many cancers and is triggered by mitochondrial dysfunction and the subsequent activation of transcription factors such as p53 and the C/EBP homologous protein (CHOP) (28). Studies have indicated that GDF15 suppresses appetite and inflammation, which led to observations that it may be important for treating obesity (28). Future studies are now needed to identify the tissues that metformin targets to increase GDF15, the mechanism by which this occurs, and whether this may be important for mediating some of the beneficial systemic effects of metformin.

In summary, these data indicate that liver energy-sensing mechanisms involving the AMPK-ACC signaling axis and subsequent reductions in lipogenesis are important for mediating the chronic insulin-sensitizing effects of metformin and that these effects can be enhanced when used in combination with salicylate-based therapeutics such as salsalate. Further support for the importance of the liver AMPK-ACC signaling pathway in reducing lipogenesis and insulin resistance has also been established using small molecules that 1) directly activate AMPK β1 subunit containing heterotrimers and increase ACC phosphorylation (13,19), 2) mimic the effects of AMPK to inhibit ACC activity (29), or 3) starve ACC of substrate by blocking the production of acetyl-CoA through inhibition of ATP-citrate lyase (30). Whether these treatments are also effective for treating nonalcoholic fatty liver disease (NAFLD) and improving liver insulin sensitivity in humans remains to be established. Liver energy-sensing mechanisms may also be important for regulating the production of metformin-derived cytokines (metokines), such as GDF15, which suppresses appetite, thereby protecting the liver from nutrient excess in the face of metformin-induced reductions in mitochondrial function and thus linking energy demand with intake. Identification and characterization of other novel metokines may lead to new therapies for a multitude of disease ailments in which metformin has been shown to exert favorable effects.

Skeletal muscle by virtue of its mass is the primary tissue contributing to insulin-stimulated glucose uptake, and while metformin is effective in improving liver insulin sensitivity, it has minimal effects on skeletal muscle insulin action (10). Given exercise is known to improve skeletal muscle insulin sensitivity and activate AMPK (5), we generated mice lacking both AMPK β1 and β2 subunits in muscle, which effectively removed all AMPK activity (31). We found that mice lacking AMPK in muscle were extremely inactive and when these mice performed treadmill exercise, they struggled to maintain even a walking speed compared with control mice that could easily maintain a much faster running speed (31). These findings have also been observed, albeit to a lesser degree, in other models of reduced muscle AMPK activity (32–35).

So why did we generate a couch potato mouse when we removed AMPK from muscle? During intense exercise, an increase in glucose uptake is an important source of fuel for working muscles. We found glucose uptake was increased into glycolytic (extensor digitorium longus) and oxidative (soleus) muscle from control mice with muscle contractions and treadmill exercise. However, if the muscle lacked AMPK, this increase in glucose uptake was blunted (31), effects that have also been observed to varying degrees in other studies (32,34–36). Skeletal muscle AMPK has also recently been shown to be important for improving skeletal muscle insulin sensitivity following muscle contractions (37). However, in contrast to metformin, the effects of exercise training on improving liver insulin sensitivity do not appear to require AMPK phosphorylation of ACC (38). Thus AMPK is important for increasing skeletal muscle glucose uptake and insulin sensitivity, but the mechanisms by which exercise improves insulin sensitivity in other organs remains unresolved.

In addition to increasing skeletal muscle glucose uptake and insulin sensitivity, another hallmark of exercise training is its ability to increase muscle mitochondrial number and function. Consistent with a role for AMPK in regulating this process, mice lacking AMPK in muscle had a lower number of mitochondria that were very large and dysfunctional (31). These mitochondrial defects became exaggerated with aging and accelerated the development of aging-induced sarcopenia, an effect related to impairments in mitophagy (39). These data in combination with those from other laboratories indicate that in response to exercise, skeletal muscle AMPK plays an important role for increasing mitochondrial biogenesis (40) and in removing old and damaged mitochondria through mitophagy (41), thus making the muscle more capable of responding to future energetic challenges.

Despite the beneficial effects of exercise, for a wide variety of reasons, we know most of the population fails to perform sufficient amounts of daily physical activity. So a key question we had given the important role for AMPK during exercise was whether a pharmacological agent that activated skeletal muscle AMPK might mimic some of these beneficial effects. We found that treatment of obese, insulin-resistant mice with a small molecule that activated skeletal muscle AMPK could increase exercise capacity, muscle mitochondrial biogenesis, and insulin sensitivity much like exercise training, but in mice lacking skeletal muscle AMPK, many of these effects were suppressed (42). This suggests that pharmacological activation of skeletal muscle AMPK can mimic some of the beneficial effects of exercise training, findings that have been recently confirmed in nonhuman primates using a new generation of potent direct AMPK activators (43,44). Collectively, this indicates that muscle AMPK is important for increasing exercise capacity, glucose uptake, and insulin sensitivity and in maintaining mitochondrial function and that a pharmacological agent that activates skeletal muscle AMPK may be able to mimic many of these effects, even in the context of obesity and insulin resistance (Fig. 3). Of course, in addition to muscle and AMPK, we know exercise exerts beneficial effects across the entire body, including improving insulin sensitivity in adipose tissue and the liver, which may be impossible to capture by targeting a single molecule. A greater understanding of the mechanisms by which these effects are mediated will be important to fully harness the therapeutic potential of exercise training.

It is known that a negative caloric balance leading to reductions in weight of ∼5–10% can dramatically improve glucose control in individuals with type 2 diabetes primarily through reductions in liver lipid content and insulin resistance (12). However, one of the challenges with maintaining this weight loss is that energy expenditure drops by ∼300 kcal per day (45), findings that have recently been exemplified from a study conducted in contestants from the television show The Biggest Loser (46). Therefore, finding ways to safely maintain basal metabolic rate following weight loss may be effective for both preventing and treating type 2 diabetes.

One of the ways in which we and many others have been interested in increasing energy expenditure has been through brown adipose tissue (BAT) and beige adipose tissue, which is best described as a hybrid of brown and white fat (Fig. 4). Notably, in contrast to classic white adipose tissue, brown and beige adipose tissue can use large amounts of energy due to the presence of abundant mitochondria that contains uncoupling protein 1 (UCP1) (reviewed in Wang and Seale [47]). Over the past several decades, the mechanisms by which cold increases the activity of BAT has been carefully detailed (Fig. 5). In a simplification of the process, the brain senses cold, which then increases sympathetic nervous activity, resulting in the local production of norepinephrine by tyrosine hydroxylase (47). Norephinephrine then binds to β-adrenergic receptors to increase cAMP, PKA, and lipolysis resulting in the release of fatty acids, which enter the mitochondria and increase the activity of UCP1 (47). This induction of mitochondrial uncoupling generates a futile cycle, which enhances metabolic flux and results in the generation of heat (Fig. 5). Importantly, seminal findings over the past few years have indicated that humans also have brown and beige adipose tissue that can be switched on with cold or β3-agonists, much like rodents (47). Interestingly, it appears that the ability of BAT to take up glucose is impaired in people with type 2 diabetes (reviewed in Loh et al. [48]) and that activated BAT is capable of using 200–300 kcal a day (49), which would be equivalent in most people to the reduction in energy expenditure that occurs following weight loss (45).

In rodents, exposure to cold or β3-agonists increases AMPK activity in BAT (50), but whether this activation was important for maintaining the function of BAT was unclear. We subsequently generated a mouse model in which AMPK could be genetically removed from adipose tissue of adult mice and found that when these mice were exposed to cold they were unable to maintain their body temperature compared with wild-type littermates (51). Similarly, if we injected mice with a specific β3-agonist (CL-316,243) that increases BAT activity, we could see that the effects of the compound on oxygen consumption and surface temperature were blunted in mice lacking adipose tissue AMPK (51) (Fig. 5). So what was the reason for this impaired thermogenesis in AMPK adipose tissue knockout mice? By examining electron micrographs of BAT, we found that AMPK adipose tissue knockout mice had a normal number of mitochondria but that the mitochondria that were present were enlarged and had substantial disorganization of the cristae. This resulted in an impairment of isolated mitochondria to oxidize substrates such as palmitoyl-CoA, an effect that was linked to reductions in mitophagy (51). Mice lacking adipose tissue AMPK also had a compromised ability to convert their white adipose tissue to beige adipose tissue (51), a phenotype that became evident after we injected mice with a β3-agonist for 5 days. Importantly, these deficiencies in brown and beige adipose tissue of AMPK knockout mice translated into a tendency for mice to gain more weight when fed a high-fat diet and increased the accumulation of liver lipids, which led to insulin resistance despite similar caloric intake (51). Comparable observations indicating the importance of AMPK in regulating BAT have recently been observed by others (52,53).

Notably, the role of AMPK in adipose tissue does not appear to be just restricted to mice as we found that noradrenaline also increases AMPK activity in BAT cells derived from humans (51). Consistent with a potentially important role for adipose tissue AMPK in maintaining thermogenesis in humans, AMPK activity is lower in a variety of white adipose depots of obese individuals with insulin resistance compared with weight-matched insulin-sensitive control subjects (54), suggesting that activating AMPK may be important for maintaining BAT mitochondrial function and the browning of white adipose, which collectively may help defend against liver lipid accumulation and insulin resistance in the context of a caloric surplus (Fig. 5). Although at first this role for AMPK to increase brown and beige fat thermogenesis seems paradoxical for an enzyme designed to protect against energy wasteful pathways, it is important to note that increasing BAT mitochondrial function and the volume of beige adipose tissue is a much more efficient means to maintain temperature homeostasis compared with shivering thermogenesis and thus, from the perspective of the entire organism, is entirely consistent with the role of AMPK to maximize energy efficiency under times of metabolic stress.

In addition to cell autonomous energy sensors such as AMPK, multicellular organisms have developed a series of hormonal cues that are important for matching energy demand with energy intake. Of these hormonal signals, serotonin is the most ancient having first evolved in plants to coordinate photosynthesis with light availability. In mammals, tryptophan hydroxylase 1 (Tph1) and tryptophan hydroxylase 2 (Tph2) are the key rate-limiting enzymes controlling the synthesis of serotonin.

These enzymes both make 5-hydroxytryptamine from tryptophan, but the different isoforms are differentially expressed, with Tph1 being primarily expressed in cells of the gastrointestinal tract, and Tph2 primarily in neuronal cells (55). Importantly, this rate-limiting reaction produces serotonin, which cannot cross the blood-brain barrier. As such, we have two distinct pools of serotonin, with one pool found in the nervous system and one pool found in the periphery, which is the site of the vast majority of the body’s serotonin (Fig. 6). Interestingly, although a great deal is known about the role of central serotonin in suppressing appetite, surprisingly, much less was known about the role of peripheral serotonin and Tph1 in regulating energy balance, other than it appeared to be increased by a high-fat Western diet (56) and polymorphisms in Tph1 were associated with obesity (57).

We therefore asked the question, what happens if we reduce peripheral serotonin by inhibiting Tph1? What we found, by genetically deleting Tph1 or suppressing its activity by injecting a small molecule chemical inhibitor, was that mice were protected against high-fat diet–induced obesity (58). Importantly, consistent with reductions in adiposity, this inhibition of Tph1 function led to marked reductions in NAFLD and improvements in glucose tolerance and insulin resistance compared with wild-type or vehicle-treated controls (58). Thus inhibiting Tph1 could treat obesity, NAFLD, and insulin resistance, but why?

We found that mice with impaired Tph1 were not losing weight because they were eating less but instead weight loss was associated with an increase in resting energy expenditure (nonmoving VO₂) of approximately ∼10–12% compared with wild-type or vehicle-treated controls, effects that were present before significant difference in body mass were observed (58). We subsequently conducted PET-CT scans and found that glucose uptake was greater in the BAT and white adipose tissue of mice with impaired Tph1 function compared with controls (58). Consistent with these findings, BAT thermogenesis and the expression of UCP1 in BAT and white adipose tissue was increased with the inhibition of Tph1 (58). Importantly, these effects were also observed at thermoneutrality and were largely eliminated in mice deficient for UCP1 (58). These data indicated that in the context of high-fat diet–induced obesity, reducing the synthesis of peripheral serotonin by blocking Tph1 enhances brown and beige fat thermogenesis, protecting mice from obesity, NAFLD, and insulin resistance (58), findings that have also recently been observed by Oh et al. (59).

Collectively, these data support the concept that elevated levels of peripheral serotonin in obesity and type 2 diabetes potentially derived from specific bacteria within the gastrointestinal tract (60) act as a brake against norepinephrine-induced activation of BAT (58). Thus inhibiting serotonin synthesis or signaling will not only generate more brown and beige adipose tissue but also, importantly, will increase the activity of that adipose tissue at lower levels of sympathetic drive. Consistent with this concept, in humans the abundance of a stable metabolite of serotonin (5-hydroxyindole-3-acetic acid [5-HIAA]) is positively correlated with age, obesity, inflammation, insulin resistance, and type 2 diabetes (61), factors known to be associated with reduced metabolic activity of BAT (48). Future studies are needed to determine whether inhibiting peripheral serotonin synthesis or signaling in BAT of people with type 2 diabetes may be effective for lowering liver lipid content and improving insulin sensitivity.

Our studies have identified how energy-sensing mechanisms in the liver, muscle, and adipose tissue can be important for improving insulin sensitivity. In the liver, metformin activation of AMPK lowers liver lipid synthesis through phosphorylation of ACC. Metformin also increases circulating GDF15, which may suppress appetite (Fig. 7). In skeletal muscle, AMPK is important for increasing exercise capacity, glucose uptake, insulin sensitivity and mitochondrial capacity and pharmacological agents targeting this pathway can act as exercise mimetics to replicate some of these beneficial effects (Fig. 7). Last, activating AMPK or inhibiting peripheral serotonin synthesis increases the metabolic activity of brown and beige adipose tissue. Collectively, these actions in liver, muscle and adipose tissue lower liver lipids and improve insulin sensitivity (Fig. 7).

In summary, our research into cellular fuel gauges has revealed how two highly conserved energy sensors, AMPK and serotonin, act to integrate nutrient status across tissues in order to regulate metabolism and energy expenditure. On the one side of the fuel gauge, when cells are running low on energy, AMPK is switched on (Fig. 8). This energetic challenge can be the result of metformin (4), exercise (5), or cold (50) and in liver (13), muscle (31,39), or adipose tissue (51), respectively; it appears that AMPK is important to defend against these acute metabolic insults. On the other side of the gauge, when the tank is full, as occurs with obesity or high-fat diets, peripheral serotonin synthesis is increased (56,60) and at the same time AMPK activity in tissues is suppressed (8,54), which collectively promotes lipid storage and insulin resistance (Fig. 8). Given the ubiquitous expression and multiple roles of AMPK, identifying ways to specifically target AMPK in liver, muscle, and adipose tissue will be important to maximize beneficial metabolic effects while minimizing deleterious side effects such as cardiac hypertrophy (62). Alternatively, ascertaining the endogenous ligands and receptors that regulate AMPK activity and serotonin synthesis and signaling in the presence of nutrient excess may yield additional ways to restore metabolic homeostasis. Last, although our work has emphasized the role of AMPK and serotonin, there are likely to be many other cellular energy sensors and circulating factors that may also be amicable to therapeutic targeting. Ultimately, the identification of new ways to manipulate ancient cellular fuel gauges may lead to the development of new, safe, and effective therapies for treating type 2 diabetes.

Acknowledgments. The author is thankful to all the trainees in the laboratory, both past and present, whose ideas, expertise, and enthusiasm are the basis for the lecture and is honored to have had the opportunity to present their work. He is also thankful to the great number of mentors, collaborators, and colleagues who he has been so fortunate to work with. A special thanks goes to Eric Desjardins (McMaster University) and Alison McElvaine (American Diabetes Association) for assistance with preparing the presentation and figures.

Funding. The research was supported by operating grants, salary support, and infrastructure from the Canadian Institutes of Health Research, Canada Research Chairs Program, Diabetes Canada, Canadian Foundation for Innovation, Natural Sciences and Engineering Research Council of Canada, the National Health and Medical Research Council of Australia, and the Faculty of Health Sciences at McMaster University.

Duality of Interest. G.R.S. has received grant support from Esperion Therapeutics and Rigel Pharmaceuticals; honoraria from Rigel Pharmaceuticals, AstraZeneca, Eli Lilly, Merck, Novo Nordisk, and Pfizer; and consulting fees from Esperion Therapeutics. No other potential conflicts of interest relevant to this article were reported.