The AMPK (AMP-activated protein kinase) is a highly conserved eukaryotic protein serine/threonine kinase. It mediates a nutrient signalling pathway that senses cellular energy status and was appropriately called the fuel gauge of the cell. At the cellular level, AMPK controls energy homoeostasis by switching on catabolic ATP-generating pathways, while switching off anabolic ATP-consuming processes. Its effect on energy balance extends to whole-body energy homoeostasis, because, in the hypothalamus, it integrates nutritional and hormonal signals that control food intake and body weight. The interest in AMPK also stems from the demonstration of its insulin-independent stimulation of glucose transport in skeletal muscle during exercise. Moreover, the potential importance of AMPK in metabolic diseases is supported by the notion that AMPK mediates the anti-diabetic action of biguanides and thiazolidinediones and that it might be involved in the metabolic syndrome. Finally, the more recent demonstration that AMPK activation could occur independently of changes in cellular energy status, suggests that AMPK action extends to the control of non-metabolic functions.
The purpose of this chapter is to summarize the multiple effects of AMPK. Because of space limitations, we have mainly quoted review articles, in which references to the original primary research papers can be found.
AMPK (AMP-activated protein kinase)
AMPK was discovered independently more than 30 years ago as a protein that inactivated ACC (acetyl-CoA carboxylase) and HMG-CoA (hydroxymethylglutaryl-CoA) reductase, two key enzymes involved in lipid metabolism. Almost 15 years later, Hardie’s group [1,2] realized that both activities were catalysed by the same multisubstrate protein kinase, which was activated by AMP . This unique feature led to the name AMPK and made it an ideal candidate to sense the adenine nucleotide content of the cell [1,2].
AMPK is a heterotrimeric complex containing a catalytic (a) and two regulatory (b and g) subunits. Each subunit has multiple isoforms (α1, α2, β1, β2, γ1, γ2, γ3) giving twelve possible combinations of holoenzyme with different tissue distributions. In skeletal muscle, only the α1β2γ1, α2β2γ1 and α2β2γ3 complexes seem to be expressed . The N-terminus of the α-subunit contains a protein kinase domain, with a threonine residue (Thr172) in the activation loop (termed the T-loop), whose phosphorylation by upstream kinases is required for AMPK activation. The β-subunit could act as a scaffold to which the two subunits are bound. It also contains a glycogen-binding domain, whose three-dimensional structure has been determined, but whose physiological role remains to be clarified. The γ-subunit contains four CBS (cystathionine β-synthase) motifs in tandem repeats, which form pairs of so-called Bateman domains that bind AMP. ATP can also bind to these sites, but with much lower affinity. Mutations in the γ2-subunit cause glycogen accumulation, leading to cardiac arrhythmias [1,2].
The intracellular distribution of the two catalytic isoforms of AMPK may differ. When phosphorylated, the α2-subunit translocates to the nucleus during hypoxia or in contracting muscle, whereas the α1-subunit remains cytoplasmic. In specialized tissues such as the carotid body glomus cells, the α1-subunit, which appears as the effector of hypoxic chemotransduction, is targeted to the plasma membrane , and in polarized epithelia, AMPK co-localizes with the CFTR (cystic fibrosis transmembrane conductance regulator) protein . The role of the different subunits in AMPK translocation is not known.
After many years of intensive research, it was only in 2003 that the first upstream AMPK-kinase was identified as LKB1 [1,6], a tumour suppressor mutated in Peutz—Jeghers syndrome, an autosomal dominant disorder characterized by the development of gastrointestinal polyps and elevated risks of cancer. In the presence of AMP, LKB1 activates AMPK by phosphorylating the Thr172 residue of the a-subunit T-loop. AMP does not act directly on LKB1, which is thought to be constitutively active. As a result of its binding to the g-subunits of AMPK, AMP was proposed to allosterically stimulate AMPK activity and to induce a conformational change exposing Thr172 for phosphorylation by LKB1 . The alternative hypothesis is that AMP inhibits Thr172 dephosphorylation via a so-called ‘substrate-mediated effect’ on the inactivating protein phosphatase, PP2C [7,8], whose involvement in the control of AMPK activity in intact cells has been recently demonstrated . Therefore protein phosphatases would be involved in AMPK activation resulting from a rise in AMP. LKB1 is a master kinase, capable of phosphorylating and activating several ARKs (AMPK-related kinases) . The discovery that CaMKKs (calmodulin-dependent protein kinase kinases) could also activate AMPK, but independently of AMP, has broadened the potential roles of AMPK in the control of processes that do not necessarily involve a change in cellular energy status [1,6]. More recently a third AMPK-kinase, the TAK1 (transforming growth factor-β-activated kinase), has been found . TAK1 is part of a G-protein signalling pathway downstream of cytokine receptors, which opens up the possibility that AMPK could be involved in inflammatory responses. Figure 1 summarizes the signalling pathways that activate AMPK.
As stated above, any increase in AMP/ATP ratio results in AMPK activation via phosphorylation of its Thr172 by LKB1. Conditions leading to changes in AMP concentration are directly related to changes in ATP concentrations. At any time, intracellular ATP concentrations reflect the balance between energy supply-and-demand. In most cells, ATP supply relies on mitochondrial oxidative phosphorylation, whereas ATP demand depends on the energy required to perform various cell functions. Any decrease in supply (e.g. hypoxia) or increase in demand (e.g. intense exercise) will decrease ATP concentrations. The equilibrium maintained by adenylate kinase translates a fall in ATP to a relatively larger change in AMP, since the AMP/ATP ratio increases as the square of the ADP/ATP ratio, thus providing an amplification mechanism .
It is worth noting that the cytosolic concentration of free AMP is less than its total cellular concentration (about 0.2 mM in normoxic tissues). AMP is in fact distributed between the cytosol and mitochondria and binds to several abundant proteins such as glycogen phosphorylase, adenylate kinase and also to many other enzymes. AMP concentrations calculated from 31P-NMR spectroscopy vary from 2 mM in normoxic hearts to about 30 μM in the presence of various inhibitors . These values are at least 10 times lower than the total concentrations but are in the range of AMP concentrations giving half-maximal activation of AMPK (2–50 μM, depending on the concentration of the competitor ATP).
In this section we briefly review the numerous conditions leading to increases in AMP concentration and AMPK activation. A decrease in ATP supply is observed under metabolic stresses, e.g. when oxygen supply is lacking, which in most cases corresponds to non-physiological conditions. This is the case in ischaemic heart where AMPK becomes activated within a few minutes. Similar effects are obtained with mitochondrial inhibitors such as oligomycin [1,13]. Increased ATP demand is another situation leading to AMPK activation, especially when combined with decreased ATP supply. This is the case in contracting muscle, particularly during intense exercise, when the concentration of ATP falls because of the massive energy demand and partial hypoxia resulting from contraction. The extent of AMPK activation seems to depend both on the intensity and duration of exercise . By contrast, AMPK activation does not occur in heart subjected to high workloads and this organ adapts its ATP supply to increased energy demand .
In several tissues, AMP and ATP concentrations depend on the availability of glucose, which controls specialized functions. For example, in pancreatic β-cells, ATP concentrations are related to glucose metabolism and insulin secretion . In these cells, AMPK is activated by glucose deprivation, suggesting that AMPK might indirectly control insulin secretion . In certain discrete hypothalamic centres, glucose deprivation also correlates with AMPK activation and with the secretion of various (an)orexigenic peptides that control food intake . Hypothalamic AMPK activation during starvation and its inactivation upon re-feeding are mediated by nutritional and hormonal signals . Indeed, AMPK is inactivated not only by glucose, but also by insulin and leptin [15–17]. Leptin is secreted by adipose tissue and its circulating concentration reflects whole-body fat stores. The systemic concentrations of insulin and leptin decrease during starvation and both inhibit food intake. Over-expression of active forms of AMPK or intra-cerebral injection of AMPK activators stimulates food intake, whereas expression of a dominant-negative mutant of AMPK has the opposite effect. Taken together, these findings support the concept that AMPK participates in the control of whole body energy homoeostasis in the hypothalamus [2,15].
AMP concentrations can increase independently of changes in ATP concentration. The activation of fatty acids to their CoA derivatives by acyl-CoA synthetases produces AMP and PPi. The rise in AMP and ensuing AMPK activation acts as a nutrient sensor for fatty acids. Under these conditions AMPK activation will favour fatty acid oxidation through the AMPK-mediated inactivation of ACC (see below). This elegant feed-forward mechanism, which has been described in the perfused rat heart , ensures that free fatty acids are indeed oxidized rather than esterified.
Several substances and drugs are known to activate AMPK. AICA (5-amino-4-imidazole-carboxamide) riboside is an analogue of adenosine that can be phosphorylated in certain cells to ZMP, which mimics several effects of AMP. Thus AICA riboside has been extensively used in cells or even in vivo to activate AMPK [1,2]. However, several effects of AICA riboside, including the inhibition of liver glucose uptake and of mitochondrial respiration, are independent of AMPK . In addition, intracellular ZMP accumulation has been reported to directly modulate the activity of enzymes which bind AMP, such as glycogen phosphorylase and fructose-1,6-bisphosphatase. Clearly results obtained with AICA riboside should be interpreted with caution. Because of its unwanted side-effects, AICA riboside is not well-suited to investigate the consequences of AMPK activation on cell function. Other more specific tools and/or new pharmacological compounds with AMPK selectivity are required. The recent development of new AMPK activators is of obvious interest .
Metformin, the most prescribed drug for Type 2 diabetes, restores glucose homoeostasis by increasing glucose disposal and decreasing hepatic glucose production. Metformin activates AMPK and the involvement of AMPK in the anti-diabetic effect of this drug is supported by studies in LKB1-deficient mice where the blood glucose lowering effect of the compound was greatly decreased [21,22]. Metformin was initially reported to activate AMPK without affecting adenine nucleotide concentrations. This interpretation had to be revised, because the initial effect of metformin is to partially inhibit respiratory chain complex I, thus decreasing ATP production. The inhibition is readily observed at high concentrations (>1 mM), but is already present at low (0.1 mM) concentrations corresponding to the therapeutic range in the portal vein . Thiazolidinediones, another class of anti-diabetic compounds, activate AMPK through an increase in AMP/ATP ratio, which also seems to result from an inhibition of mitochondrial respiration . In relation to the effect of these anti-diabetic drugs, AMPK deficiency has been proposed to underlie the progression of the metabolic syndrome, which is characterized by insulin resistance . This syndrome corresponds to a combination of various metabolic and haemodynamic factors that represent a high risk of developing Type 2 diabetes and cardiovascular disease. Hyperglycaemia, hypertension, abdominal obesity, low HDL (high-density lipoprotein) cholesterol and high circulating triacylglycerols are characteristic features. Interestingly, mice that are deficient in the AMPK α2-subunit display abnormal control of body mass, insulin sensitivity, glucose and lipid homoeostasis and mimic the metabolic syndrome to some extent . In addition, the circulating levels of the fat-cell derived adiponectin are decreased in obese and insulin-resistant patients and its insulin-sensitizing effect might well be mediated by AMPK activation [2,15,21,25].
Several conditions are known to change AMPK activity without detectable changes in adenine nucleotide concentrations. The effect of insulin to antagonize hypoxia-induced AMPK activation is not related to changes in adenine nucleotides but could result from phosphorylation by PKB (protein kinase B) of Ser485/491 on the a-subunits of AMPK, which inhibits Thr172 phosphorylation by LKB1  (see below). Since CaMKKβ has been reported to be an activating upstream AMPK kinase, AMPK activation could also be triggered by an increase in cytosolic calcium. This could explain AMPK activation by thrombin in endothelial cells  and in response to triggering of the T-cell antigen receptor in lymphocytes . It could also apply to AMPK activation by osmotic stress, which occurs without a change in adenine nucleotide concentrations . Finally, it is worth mentioning that, by contrast with its inactivating effect on hypothalamic AMPK, leptin has been reported to activate AMPK in muscle, as does adiponectin [2,15,21,25]. The precise mechanism of AMPK (in)activation by these adipokines is controversial.
Key enzymes involved in the control of carbohydrate, lipid and protein metabolism have been identified as AMPK substrates, leading to the concept that AMPK acts as a metabolic master switch that conserves ATP. A list of some of the targets of AMPK and their phosphorylation site sequences is given in Table 1. The substrate recognition motif for AMPK is θ(X,β)XXS/TXXXθ, where β is a hydrophobic residue, β a basic residue, X is any amino acid and the parentheses indicate that the order of residues at the P-4 and P-3 positions is not critical. Table 1 shows that AMPK is predominantly a serine-directed protein kinase and that in some targets the consensus is not perfectly adhered to. One interesting possibility is that the various AMPK isoforms might differ in substrate recognition. In the hearts of mice lacking the a2 catalytic subunits, the metabolic response to no-flow ischaemia is abrogated, indicating that the persisting a1-subunit cannot compensate for the lack of the α2 catalytic subunit .
A number of studies have shown that AMPK activation inhibits fatty acid and cholesterol synthesis in liver and glycogen synthesis in muscle through the phosphorylation and inactivation of ACC, HMGCoA reductase and glycogen synthase respectively. The long-term induction of gluconeogenic enzymes by starvation is also inhibited by AMPK, which phosphorylates TORC2 (transducer of regulated CREB activity 2), a transcriptional co-activator that inhibits the expression of these genes [21,22].
AMPK activation also inhibits protein synthesis at several levels (summarized in Figure 2). Control of protein synthesis is exerted via the (de)phosphorylation of various translation factors and ribosomal proteins [21,22,47], including the mTOR (mammalian-target-of-rapamycin). mTOR activation involves the G protein Rheb (Ras homologue enriched in brain) and the complex TSC1–TSC2 (Tuberous Sclerosis Complex 1 and 2, also called hamartin and tuberin respectively). TSC1–TSC2 is mutated in tuberous sclerosis, a genetic disorder characterized by benign tumours (hamartomas), and acts as a GTPase-activating protein favouring the GDP-bound state of Rheb, thus inhibiting mTOR signalling. PKB, which mediates insulin effects on protein synthesis, phosphorylates and inactivates TSC2 thus keeping Rheb in its active form and favouring mTOR activation. AMPK has the opposite effect by phosphorylating TSC2 on sites other than those phosphorylated by PKB, thereby stimulating GTPase activity and keeping Rheb and mTOR inactive . The mTOR pathway controls initiation via the phosphorylation of 4E-BP1 (eukaryotic initiation factor 4E-binding protein-1), thereby relieving its inhibition on eIF4E, which can then bind the mRNA cap. mTOR also controls elongation via a phosphorylation cascade involving p70S6K (p70 ribosomal protein S6 kinase). p70S6K phosphorylates and inactivates eEF2 (eukaryotic elongation factor 2)-kinase, a dedicated Ca2+-calmodulin-dependent kinase that phosphorylates and inactivates eEF2. Therefore mTOR activation decreases eEF2 phosphorylation, a mechanism involved in the insulin-induced stimulation of protein synthesis . By contrast, AMPK activates eEF2-kinase, which in turn phosphorylates and inactivates eEF2, thus inhibiting translation elongation [21,48]. Taken together, these observations support the concept that AMPK activation inhibits anabolic ATP-consuming processes including protein synthesis and cell growth (Figure 2).
Other studies have demonstrated that AMPK activation favours ATP-producing pathways. Inactivation of ACC by AMPK decreases the concentration of its product malonyl-CoA, thereby relieving inhibition of the carnitine-mediated import of long chain acyl-CoAs into mitochondria and eventually stimulating their oxidation [1,21,22]. This control is probably the reason for the presence of ACC in non-lipogenic tissues. Fatty acid oxidation requires oxygen and obviously cannot be stimulated in ischaemic tissues, i.e. when AMPK is activated. However, the beneficial effect of AMPK activation on fatty acid oxidation is expected during heart re-perfusion following a severe ischaemic episode. AMPK activation also stimulates ATP production by increasing glycolytic flux at two levels. In exercising muscle, AMPK activation stimulates glucose uptake by increasing the recruitment of GLUT4 (glucose transporter 4) transporters to the plasma membrane in an insulin-independent manner. This underlies the recommendation of physical exercise for diabetic patients. The molecular mechanism could involve the phosphorylation by AMPK of the Rab GTPase activating protein, AS160 (Akt substrate 160), a target of PKB in the insulin-signalling pathway . By analogy with the control of mTOR by Rheb, the GTPase activity of AS160 is inhibited via phosphorylation by PKB, thereby maintaining Rabs that regulate GLUT4 translocation in their active GTP-bound state . In hypoxic heart and monocytes, AMPK also phosphorylates and activates 6-phosphofructo-2-kinase, thereby increasing the concentration of its product fructose 2,6-bisphosphate [13,35]. Thus protein phosphorylation seems to be part of the mechanism by which oxygen deprivation stimulates glycolysis (the Pasteur effect). The combined increase in fructose 2,6-bisphosphate and AMP, both of which allosterically stimulate 6-phosphofructo-1-kinase activity, contributes to the overall stimulation of glycolysis. Whether AMPK also influences ATP production by mitochondria does not seem to be the case, at least in the short-term. However, AMPK activation stimulates mitochondrial biogenesis, probably through up-regulation of PGC-1α (peroxisome-proliferator-activated receptor γ co-activator-1α) .
The HIF-1 (hypoxia-inducible factor) complex is another oxygen sensor allowing cells to adapt to oxygen deprivation. This transcription factor stimulates the expression of several genes by binding to hypoxia-responsive elements in promoters of key angiogenic and glycolytic genes. The role of AMPK in the regulation of HIF-1 is unclear. Paradoxically, AMPK should inhibit HIF-1 synthesis, by inhibiting the mTOR pathway, which seems to control its synthesis. However, the available evidence suggests that hypoxia independently activates both AMPK and HIF-1 , implying that activation of pre-existing HIF-1 would be enough to trigger the adaptive response to low oxygen. Therefore both AMPK activation and HIF-1 stabilization appear to be parallel responses to oxygen deprivation, which act synergistically to maintain energy homeostasis. Another interesting target is eNOS (endothelial nitric oxide synthase), which can be phosphorylated and activated by AMPK, thereby contributing to vasodilation and improved oxygen supply .
In growing cells, protein synthesis and ion transport together are the most prominent consumers of oxidatively-derived ATP. Since protein synthesis is inhibited by AMPK activation at several levels, one would expect energy-consuming ion transport systems, such as Na+/K+-ATPase, Ca2+-ATPase or other transporters to be shut down as a result of AMPK activation. Accordingly, there is some evidence that several ion channels could be regulated by AMPK [5,52].
Some caution needs to be exercised before a target protein can be considered as a bona fide AMPK substrate. For example, many AMPK targets have been reported on the basis of over-expression of recombinant proteins in cells incubated with pharmacological activators of AMPK, such as AICA riboside, or subjected to treatments that deplete intracellular ATP, conditions which are somewhat artificial. Also in some cases, the sites for AMPK have not been identified formally by direct sequencing or mass spectrometry, but by searching target protein sequences for the AMPK consensus and then mutating the putative phosphorylation site to a negatively charged (to mimic phosphorylation) or non-phosphorylatable residue and looking at effects on function. To verify that AMPK is the physiologically relevant kinase for a particular target, changes in cellular AMPK activity should be correlated with the extent of phosphorylation of the substrate protein at the relevant site. This could be undertaken either by incubating cells with a selective AMPK inhibitor/activator or by modulating AMPK content by genetic approaches. However, these approaches might be inadequate in one way or another, for example either by lack of isoform specificity with the dominant negative approach, by incomplete AMPK inhibition or by compensatory mechanisms such as induction of the expression of the other AMPK catalytic subunit isoform or indeed of other related kinases. The overexpression of a dominant-negative construct might not target endogenous AMPK in the right intracellular compartment or might titrate out signalling molecules necessary for the activation of other kinase cascades that converge on AMPK targets. Clearly, where possible several approaches should be tested in controlled experiments where AMPK activity is monitored in cells, for example by measuring ACC phosphorylation.
Sharing control and cross-talks
Virtually all the known AMPK targets are multi-site phosphorylated proteins either containing more than one AMPK site, or sites for other protein kinases such as PKA (cyclic AMP-dependent protein kinase) and PKB. AMPK phosphorylation sites are often adjacent to sites phosphorylated by other protein kinases, or the AMPK site itself is also phosphorylated by a protein kinase from another signalling pathway (Table 1). There are cases where the AMPK and insulin signalling pathways (via PKB) converge on the same target at the same site and with the same effects, as e.g. heart PFK-2 (6-phosphofructo-2-kinase)  and skeletal muscle GLUT4 transporters .
By contrast, the insulin and AMPK pathways have opposite effects on protein synthesis (Figure 2). Insulin stimulates protein synthesis by activating the mTOR pathway , whereas AMPK activation decreases protein synthesis via eEF2-kinase activation and by inhibiting mTOR signalling [21,54], but in some cases the latter effect was only evident after the mTOR pathway had first been stimulated by growth factors/insulin or amino acids [14,55]. PKB and AMPK have been reported to phosphorylate distinct sites both on TSC2  and mTOR [40,56] with opposing effects. Therefore the possibility of hierarchical phosphorylation of these proteins by PKB and AMPK obviously needs investigation. Indeed the effect of insulin to antagonize AMPK activation involves a hierarchical mechanism whereby phosphorylation of the a-subunit Ser485/491 by PKB reduces subsequent phosphorylation of Thr172 by LKB1 and the resulting AMPK activation .
Anti-proliferative and non-metabolic effects of AMPK
Although LKB1 seems to be the major upstream activating kinase in many tissues and cells, the fact that signalling via CaMKKs can lead to AMPK activation indicates that both changes in intracellular calcium and AMP concentrations may play separate or inter-dependent roles in the regulation of AMPK activity. Moreover, it was recently shown that TAK1 was an authentic AMPK kinase at least in vitro . Therefore although the primary targets for AMPK have previously been presumed to be mainly involved in energy homoeostasis, AMPK activation seems now to control non-metabolic processes, such as cell growth, progression through the cell cycle and the organization of the cytoskeleton.
Sustained AMPK activation suppresses cell proliferation and growth and can even trigger apoptosis [51,57]. This anti-proliferative action of AMPK is mediated by a number of tumour suppressors that are part of the AMPK signalling pathway, such as LKB1 and TSC1–TSC2 . In addition, AMPK activation in epithelial cells has also been shown to control tight junction formation  and cell migration through re-organization of the actin cytoskeleton . Finally, thrombin has been shown to activate AMPK in endothelial cells through a calcium-dependent mechanism mediated by CaMKKβ . It is clear that these new unexpected roles of AMPK extend beyond energy homoeostasis and potentially link AMPK to the many calcium-dependent cell functions.
Early knowledge about the AMPK system led to an understanding of its role as an energy-sensor. Accumulating evidence now indicates that AMPK can be considered as a multi-tasking protein controlling cellular homoeostasis at large. This might be related to the fact that it is encoded by house-keeping genes. One might speculate that the two catalytic subunits of AMPK have different specificities, the α2-subunit being more metabolically orientated and controlled by LKB1, whereas the α1-subunit of AMPK would be involved in various calcium-dependent cell functions and controlled by CaMKKs. Which catalytic subunit is devoted to the control of the cell cycle needs further investigation.
The number of AMPK substrates is still growing, as is the number of papers published on AMPK. However, the demonstration that reported target proteins can be considered as bona fide substrates is often lacking. Many more targets are likely to be identified in view of the fact that AMPK is an ancient eukaryotic kinase and that its role extends beyond that of controlling energy homoeostasis.
• AMPK is a heterotrimeric protein serine/threonine kinase containing one catalytic (α) and two (β, γ) regulatory subunits, which is activated by phosphorylation of its α-subunit T-loop Thr172 residue by upstream kinases
• Three pathways lead to AMPK activation: one sensing energy depletion and mediated by LKB1; the second resulting from changes in intracellular calcium concentrations and mediated by CaMKKs; and the third mediated by TAK1.
• AMPK controls energy homoeostasis: at the cellular level, by phosphorylating key metabolic enzymes resulting in the stimulation of ATP production and inhibition of ATP consumption; at the whole-body level, by controlling food intake in the hypothalamus.
• AMPK activation also controls non-metabolic processes, such as cell growth, progression through the cell cycle and the organization of the cytoskeleton, by mechanisms that are not necessarily linked to energy homoeostasis.
The work carried out in the authors’ laboratory was supported by the Belgian Fund for Medical Scientific Research, the Interuniversity Poles of Attraction Belgian Science Policy (P5/05), the French Community of Belgium (Actions de Recherche Concertées) and the EXGENESIS Integrated Project (LSHM-CT-2004–005272) from the European Commission.
- © The Authors Journal compilation © 2007 Biochemical Society