AMPK: a nutrient and energy sensor that maintains energy homeostasis

Preface AMP-activated protein kinase (AMPK) is a crucial cellular energy sensor. Once activated by falling energy status, it promotes ATP production by increasing the activity or expression of proteins involved in catabolism, while conserving ATP by switching off biosynthetic pathways. AMPK also regulates metabolic energy balance at the whole body level. For example, it mediates the effects of agents acting on the hypothalamus that promote feeding, and entrains circadian rhythms of metabolism and feeding behavior. Finally, recent studies reveal that AMPK conserves ATP levels through the regulation of processes other than metabolism, such regulation of the cell cycle and neuronal membrane excitability.

Genes encoding AMPK subunits are found in essentially all eukaryotes. AMPK and its orthologs appear to exist universally as heterotrimeric complexes comprising a catalytic α subunit and regulatory β and γ subunits, the domain organization of which are summarized in Fig. 2 . The kinase activity of the αβγ complexes, in animals from nematodes to humans, is instantaneously increased by AMP by as much as 10-fold, although due to technical limitations with the assay 5 the effect observed is usually smaller. It has been argued that this allosteric activation [G] by AMP (which gave the kinase its name) may not be relevant under physiological conditions because of competition at the allosteric site by ADP and ATP 6 . However, as shown in Fig. 1 , an increase in the ADP:ATP ratio of only 10-fold is sufficient for the cellular concentrations of the three adenine nucleotides to become equal. Although this would represent quite a severe stress and might only occur in pathological states (such as ischemia) rather than under more physiological conditions (such as exercise), the ATP:ADP ratio would still be still many orders of magnitude away from equilibrium and could still drive energy-requiring processes. AMPK catalytic subunits contain a conventional serine/threonine kinase domain at the N-terminus. In all species, the activity of the complex increases more than 100-fold 5 when a conserved threonine residue in the activation loop [G] (which is conventionally referred to as Thr172 due to its position in the original rat sequence 7 )is phosphorylated by upstream kinases. In mammals the major upstream kinases are the LKB1:STRAD:MO25 complex [G] 8 – 10 (originally identified genetically as a tumor suppressor) and the Ca 2+ /calmodulin-activated protein kinases kinases, especially CaMKKβ 11 – 13 . The LKB1 complex provides a high basal level of phosphorylation at Thr-172 that is modulated by the binding of AMP to the AMPK-γ subunit, which promotes phosphorylation and inhibits dephosphorylation 14 , 15 . Although allosteric activation is only caused by AMP, it has recently been found that the effects on phosphorylation and dephosphorylation can also be produced by ADP 16 , 17 . The effects of AMP and ADP on phosphorylation require N-terminal myristylation [G] of the β subunit 17 , 18 . Since AMP and ADP bind the γ subunits of AMPK with similar affinity to ATP 16 (which does not cause activation), and ADP is usually present in cells at higher concentrations than AMP ( Fig. 1 ), ADP may be the key activating signal that promotes net Thr-172 phosphorylation during moderate energy stress. However, allosteric activation by AMP would further amplify activation during a more severe stress. This complex mechanism ( Fig. 1 ) allows the system to provide a graded response of AMPK activity over a wide range of stress levels. The alternative activating pathway, involving CaMKKβ, triggers activation of AMPK in response to increases in cell Ca 2+ without necessarily requiring any change in AMP or ADP levels. In tumor cells that have lost the tumor suppressor LKB1 due to somatic mutations, treatments that increase AMP and ADP levels do not normally activate AMPK 8 , because basal CaMKKβ activity is too low for the effects of nucleotide binding on phosphorylation status to become evident. However, these treatments can cause AMPK activation in such cells if intracellular Ca 2+ is also elevated 19 . This emphasizes that the effects of AMP and/or ADP on Thr172 phosphorylation status are a result of their binding to AMPK, and are independent of the upstream kinases and phosphatases that phosphorylate or dephosphorylate Thr172.

In mammalian cells, AMPK is activated by many different types of metabolic stress, drugs and xenobiotics by the mechanisms described above, involving increases in cellular AMP, ADP or Ca 2+ . These can now be regarded as the classical or “canonical” AMPK activation mechanisms. However, recent work suggests that other stimuli activate AMPK via mechanisms that do not involve changes in the levels of AMP, ADP and Ca 2+ , which can therefore be termed “non-canonical” mechanisms. These distinct types of mechanism are addressed below. Activation by metabolic stresses, drugs and xenobiotics The canonical mechanisms of activation of mammalian AMPK, which involve increases in AMP and ADP ( Fig. 1 ) explain why AMPK is activated by stresses that inhibit the catabolic production of ATP, such as starvation for glucose 38 or oxygen 39 or addition of a metabolic poison 40 , as well as by stresses that increase ATP consumption such as muscle contraction 41 . AMPK is also switched on by numerous drugs and xenobiotics, including anti-diabetic drugs (such as metformin, phenformin and thiazolidinediones 42 , 43 ), plant products reputed to have health-promoting properties (resveratrol from grapes and red wine 44 , epigallocatechin gallate from green tea, capsaicin from peppers 45 , curcumin from turmeric 46 , and even garlic 47 ), and plant products used in traditional Chinese medicine (berberine 48 and hispidulin 49 ). Metformin, which is now prescribed to more than 100 million people with type 2 diabetes worldwide, was derived from the natural product galegine extracted from Galega officinalis , a plant reputedly used to treat diabetes-like conditions in medieval Europe. Although metformin activates AMPK, this may not explain all of the therapeutic effects of the drug. In fact, AMPK-α1 knockout mice that also carry a liver-specific deletion of AMPK-α2 display a normal hypoglycemic response to metformin, and the acute effects of metformin on glucose production in isolated hepatocytes are also preserved 4 . The latter effect may be explained by metformin causing an increase in AMP that directly inhibits the gluconeogenic enzyme fructose-1,6-bisphosphatase. One caveat in the interpretation of these experiments is that the fall in ATP (and hence the increase in AMP) caused by metformin was larger in cells lacking AMPK than in wild type cells 4 , suggesting that the inhibition of fructose-1,6-bisphosphatase by AMP may be accentuated in the absence of AMPK. Thus, although not completely ruling out a role for AMPK in metformin action, these results do indicate that other targets, such as fructose-1,6-bisphosphatase, are also important. It has been suggested that metformin activates AMPK in L6 cells by inhibiting AMP deaminase (the enzyme that breaks down AMP) thus causing AMP to accumulate 50 . While this is an interesting proposal, the concentration of metformin used to inhibit AMP deaminase (10 mM) was very high. The same study also reported that the effect of metformin to activate AMPK was not reduced by siRNA knockdown of adenylate kinase. However, only the “cytosolic” AK1 isoform was knocked down, and not the mitochondrial isoform, AK2, which might be the more relevant isoform when studying effects of a mitochondrial inhibitor. Also, we now know that increased phosphorylation of AMPK can be triggered by an increase in ADP alone, obviating any requirement for adenylate kinase to generate AMP. Intriguingly, some of the AMPK activators described above, including resveratrol and metformin, extend healthy lifespan in C. elegans , and genetic studies show that the AMPK ortholog is required for these effects, as well as for the life-extending effects of dietary restriction 51 , 52 . In both C. elegans and mammalian cells, AMPK up-regulates genes involved in oxidative metabolism and oxidative stress resistance by regulating transcription factors of the DAF-16/FOXO family 53 , 54 . This might contribute to its effects on healthy lifespan. An important question is how these drugs and xenobiotics all manage to activate AMPK, despite the fact that their structures are so varied. Most modulate AMPK in intact cells but not in cell-free assays, suggesting that they activate AMPK indirectly. Using a cell line that expresses an AMP- and ADP-insensitive AMPK mutant, it has been shown that many of them, including metformin and resveratrol, activate AMPK indirectly by increasing cellular AMP and ADP, usually by inhibiting mitochondrial ATP synthesis 55 . Many of these natural products appear to be defensive compounds produced by plants to deter infection by pathogens, or grazing by insect or mammalian herbivores. Consistent with this, resveratrol is produced in grapes in response to fungal infection 56 , while Galega officinalis is classed as a noxious weed in the USA because it is poisonous to herbivores. Since the mitochondrial respiratory chain and ATP synthase contain several large multiprotein complexes, they might have many potential bindi

AMPK and its orthologs phosphorylate downstream targets at serine/threonine residues within a characteristic sequence motif, which has hydrophobic residues at the -5 and +4 positions and basic residues at -4 and -3, or both 65 , 66 . Another basic residue at -6 is an additional positive determinant, while the best substrates (such as acetyl-CoA carboxylase-1/-α) have additional hydrophobic residues forming an amphipathic helix N-terminal to the -5 position 67 . The principal effects of AMPK activation on cell metabolism are summarized in Fig. 4 . Consistent with a role in maintaining energy homeostasis, when AMPK is activated by energetic stress it switches on catabolic pathways that generate ATP, while switching off biosynthetic pathways that consume ATP. Regulation of glucose uptake by the glucose transporter GLUT4 Catabolic events mediated by AMPK include enhanced glucose uptake during muscle contraction, when the major metabolic fate of the glucose is catabolism to generate ATP. Muscle glucose uptake is also promoted by the hormone insulin, although this mainly occurs in resting muscle when the major metabolic fate of the glucose is glycogen synthesis, an anabolic process. In both cases, enhanced glucose uptake is mediated through translocation of the glucose transporter GLUT4 from intracellular storage vesicles to the plasma membrane. Fusion of these vesicles with the plasma membrane requires members of the Rab family of G proteins to be in their GTP-bound state 68 . Under basal conditions Rabs are held in an inactive GDP-bound state by Rab-GAPs such as AS160 (also known as TBC1D4) and TBC1D1, which are associated with the GLUT4 storage vesicles. The insulin-activated kinase Akt phosphorylates AS160 in muscle and adipocytes, triggering its association with 14-3-3 proteins and its consequent dissociation from the vesicles 69 – 71 . Similarly, AMPK phosphorylates TBC1D1 in contracting muscle, with similar effects 72 , 73 . In either case, dissociation of these Rab-GAPs triggers the conversion of Rabs to their active, GTP forms and the consequent fusion of the vesicles, carrying their GLUT4 cargo, with the plasma membrane. Consistent with this model, mice with a knock-in mutation (T649A) at a key Akt phosphorylation site in AS160 show impaired glucose disposal in vivo and impaired insulin-stimulated glucose uptake with isolated muscle 71 . It would be very interesting to perform similar experiments with the AMPK site on TBC1D1, to confirm the mechanism by which AMPK regulates glucose uptake in contracting muscle. Whether AMPK, working via phosphorylation of TBC1D1 and perhaps other targets, entirely accounts for the acute effects of contraction on muscle glucose transport has been a matter of some controversy. In mouse knockouts of AMPK-α2 (the major catalytic subunit isoform in muscle), the stimulatory effects of the AMPK activator 5-aminoimidazole-4-carboxamide riboside (AICAR) on glucose uptake were lost, but the effects of contraction were normal. In AMPK-α1 knockout mice, both responses were normal 74 . By contrast, in mice with muscles lacking the upstream kinase LKB1 (in which activation of both AMPK-α1 and AMPK-α2 complexes by contraction was abolished), the effects of both AICAR and contraction on glucose uptake were lost 2 . One explanation for these discrepancies is that AMPK-α1 may be able to compensate for AMPK-α2 when the latter is absent. Recent results with muscle-specific AMPK-β1 -/- AMPK-β2 -/- double knockout mice, which lack any detectable AMPK activity, have supported the idea that AMPK activation is crucial during muscle contraction. The running speed and endurance of these mice was dramatically reduced, and they exhibited blunted muscle glucose uptake in response to treadmill exercise, and markedly impaired contraction-stimulated glucose uptake in isolated muscles 75 . These results are consistent with the idea that AMPK represents the primary signalling pathway responsible for contraction-induced glucose uptake, although TBC1D1 may not be the only downstream target that mediates this effect.

Another critical process activated by AMPK is mitochondrial biogenesis ( Fig. 4 ), which in the longer term generates increased capacity for the oxidative catabolism of both glucose and fatty acids. Repeated daily dosing of rats with AICAR results in increased expression of mitochondrial genes in muscle 80 as well as increased exercise endurance 81 (the latter finding prompted the World Anti-Doping Agency to ban the use of AICAR in competitive sports). The AMPK-β1 -/- AMPK-β2 -/- double knockout mice mentioned above, which lack any AMPK activity in muscle, also had greatly reduced muscle mitochondrial content 75 . The “master regulator” of mitochondrial biogenesis is PGC-1α, a co-activator that enhances the activity of several transcription factors acting on nuclear-encoded mitochondrial genes 82 . AMPK directly phosphorylates PGC-1α, which has been proposed to cause activation of its own transcription via a positive feedback loop 83 . An alternative mechanism by which AMPK has been proposed to activate PGC-1α is through promotion of its deacetylation, by increasing the concentration of NAD, the co-substrate for the deacetylase SIRT1 84 . As well as increasing mitochondrial biogenesis, AMPK is involved in the turnover of mitochondria via the special form of autophagy termed mitophagy ( Fig. 4 ). ULK1 and ULK2, the mammalian orthologs of the yeast Atg1 kinase that initiates the autophagy cascade, form stable complexes with AMPK 85 , and AMPK phosphorylates and activates ULK1, thus triggering autophagy 86 , 87 . In cells in which endogenous ULK1 was replaced by a kinase-inactive mutant, or by a mutant in which the AMPK phosphorylation sites were substituted by alanine, mitochondria with aberrant morphology and reduced membrane potential accumulated following nutrient starvation 86 . This supports the idea that phosphorylation of ULK1 by AMPK is required for the clearance of dysfunctional mitochondria. Mitochondria are the main site of production of reactive oxygen species in the cell, and are particularly susceptible to oxidative damage. By recycling components of damaged mitochondria, mitophagy may be as important in maintaining a healthy cellular ATP-generating capacity as is the production of new mitochondria.

In mammals, AMPK can also influence metabolism and energy balance at the whole body level, particularly through its actions in the hypothalamus of the brain. AMPK and control of appetite The primary appetite control centre is the arcuate nucleus [G] of the hypothalamus, in which increased electrical activity in neuropeptide Y and agouti-related protein-expressing neurons (NPY/AgRP neurons) [G] induces feeding, while increased activity in neurons expressing pro-opiomelanocortin (POMC neurons) [G] inhibits feeding 97 . Kinase assays in dissected hypothalamic regions from rodents show that hormones that inhibit feeding, such as the adipokine leptin, inhibit the α2 isoform of AMPK 98 , whereas those that promote it, such as the hormone ghrelin [G] from the stomach, the adipokine adiponectin, or cannabinoids, activate AMPK (the specific isoform was not determined) 98 – 101 . Direct injection into the hypothalamus of pharmacological activators of AMPK, or DNAs encoding activated mutants, also promotes feeding 98 , 99 . Although leptin inhibits AMPK-α2 in the hypothalamus 98 , mice in which AMPK-α2 was specifically knocked out in NPY/AgRP neurons or POMC neurons displayed only minor changes in food intake, and still exhibited decreased food intake in response to leptin 102 . A possible explanation for this anomaly was provided by a recent report suggesting that it is in presynaptic neurons [G] upstream of NPY/AgRP neurons, rather than in the NPY/AgRP or POMC neurons themselves, that regulation of AMPK is critical 103 . The frequency of miniature excitatory postsynaptic currents [G] in the NPY/AgRP neurons was used as a measure of neurotransmitter release from presynaptic neurons acting upstream. Using various pharmacological agents, evidence was obtained for a model in which ghrelin, which is released from the stomach during fasting, activates AMPK in presynaptic neurons via GHSR1 receptors ( Fig. 5a ). These receptors activate heterotrimeric G proteins containing G q /G 11 , which trigger intracellular Ca 2+ release and therefore lead to the activation of AMPK by the CaMKKβ pathway 104 . Consistent with this, CaMKKβ-deficient mice show reduced expression of NPY and AgRP, but not POMC, in the hypothalamus, and fail to increase their food intake in response to ghrelin 105 . Although the critical downstream target(s) for AMPK in the presynaptic neurons remain unclear, its activation appears to initiate a positive feedback loop in which Ca 2+ release via ryanodine receptors [G] causes sustained activation of AMPK and release of Ca 2+ , and consequent neurotransmitter release onto the NPY/AgRP neurons. This in turn promotes sustained feeding, which due to the positive feedback loop continues even after ghrelin stimulation has ceased. According to this model, feeding only stops when leptin released by adipocytes stimulates release from neighbouring POMC neurons of opioids, which act via µ-opioid receptors in the presynaptic neurons upstream of the NPY/AgRP neurons to inhibit AMPK (the mechanism for this latter effect remains unknown) ( Fig. 5b ). An interesting parallel exists between this proposed neural circuit and the “reset-set” or “flip-flop” memory storage circuits used in electronic devices. These circuits are switched on by signals coming in via the “set” input, and remain switched on even in the absence of further stimulation until a signal is received on the “reset” input. By analogy, hunger signals such as ghrelin represent the “set” inputs that switch on AgRP neurons and trigger feeding, and feeding would continue until the satiety signal, leptin, resets the circuit by acting on the POMC neuron. This model is consistent with observed diurnal patterns of plasma ghrelin and leptin in humans eating normal meals 106 . If excess food is available, it may have survival value to continue eating until the leptin signal coming from adipocytes indicates that fat stores have been replenished, rather than stopping eating as soon as the hunger signal from the stomach (ghrelin) had ceased.

In most mammals, whole body metabolism exhibits circadian rhythms, via which the timing of feeding and metabolism are synchronized with the daily cycle of light and darkness. The suprachiasmatic nucleus [G] of the hypothalamus is the site of the “master clock” that generates these rhythms. However, the proteins involved in establishing these rhythms are also expressed at peripheral sites such as the liver, forming “slave clocks” that can oscillate independently under certain circumstances 110 . These proteins include the transcription factors Bmal1 and Clock, which form a heterodimer that activates the expression of many genes, two of which are the Per ( period ) and Cry ( cryptochrome) genes. Once synthesized, Per and Cry proteins bind to each other, become phosphorylated and then enter the nucleus, where they inhibit the Clock:Bmal1 complex. This delayed negative feedback loop results in the rhythmic expression not only of Per and Cry themselves, but also of numerous other genes driven by Bmal1/Clock. AMPK has been found to phosphorylate Cry1, reducing its association with Per2 and instead increasing its binding to FBXL3, a ubiquitin ligase that promotes Cry1 ubiquitylation and degradation 111 . This would have the effect of extending the period of the endogenous rhythm. In mouse embryo fibroblasts synchronized by serum starvation, treatment with low glucose or the AMPK activator AICAR reduced the amplitude, and increased the period, of the circadian rhythm of a luciferase reporter gene driven by the Bmal1/Clock promoter; these effects were lost in cells lacking LKB1 or AMPK. It has been known for many years that the phosphorylation of AMPK targets such as acetyl-CoA carboxylase follows a circadian rhythm in rodent liver linked to the times of feeding 88 . The more recent results 111 suggest that AMPK may have a crucial role in determining these circadian rhythms.

Remarkably, ATP turnover in the grey matter of brain is comparable with that in leg muscle during marathon running, explaining why an organ contributing only 2% of body weight can account for >20% of resting metabolism 118 . It has been estimated that the firing of action potentials accounts for 25–50% of this energy consumption, with synaptic transmission (which is triggered by action potentials) contributing most of the remainder 118 , 119 . A mechanism that down-regulated the firing of neuronal action potentials would therefore conserve a considerable amount of energy. Recent studies with HEK-293 cells stably expressing the potassium channel Kv2.1 showed that AMPK activation was found to activate these potassium channels by causing a shift in voltage gating to more negative membrane potentials. Identical effects were observed when activated AMPK (thiophosphorylated at Thr172 to make it resistant to phosphatases), but not an inactive mutant, was introduced into the cells via the patch pipette 120 . AMPK phosphorylates purified Kv2.1 at two sites in its cytoplasmic C-terminal tail, and the effect of AMPK activation on voltage gating was lost in cells in which one of these (Ser440) was substituted with a non-phosphorylatable alanine 120 . Kv2.1 accounts for a large proportion of the delayed rectifier K + channels [G] in central neurons, and their activation has been proposed to reduce the firing of action potentials down the axon 121 . Interestingly, introduction of active thiophosphorylated AMPK, but not an inactive mutant, into cultured rat hippocampal neurons via the patch pipette caused a progressive decrease in the frequency of action potentials induced by a current pulse 120 . These results support the idea that AMPK activation may exert a neuroprotective role by limiting the rate of firing of action potentials, thus conserving energy when the energy status of neuronal tissue is compromised.