AMP-Activated Protein Kinase: A Target for Drugs both Ancient and Modern

AMPK appears to exist in all eukaryotes as heterotrimeric complexes composed of a catalytic α subunit and regulatory β and γ subunits ( Fig. 1 ). In mammals all three subunits have multiple isoforms (α1, α2; β1, β2; γ1, γ2, γ3) encoded by distinct genes, and all twelve possible heterotrimeric combinations are able to form when these are co-expressed in cells. The function of the different subunit isoforms remains unclear, although there is tissue-specific expression of some isoforms, and (as discussed below) there is evidence that different isoforms may target complexes to specific subcellular locations. Moreover, some AMPK-activating drugs show selectivity for certain isoform combinations. The catalytic α subunits The catalytic subunits are encoded by two alternate genes in mammals ( PRKAA1 and PRKAA2 in humans). The α1 isoform appears to be universally expressed, while α2 is more abundant in tissues such as skeletal and cardiac muscle, and appears to be absent in cells of the blood and endothelial cell lineages. Both isoforms contain conventional serine/threonine kinase domains at the N-terminus, and their kinase activity is increased >100-fold by phosphorylation of a conserved threonine residue within the “activation loop”, a sequence segment also critically involved in regulation of many other kinases ( Johnson, et al., 1996 ). This threonine residue is usually referred to as Thr-172 due to its position in the original rat sequence ( Hawley, et al., 1996 ), although the exact residue numbering may differ in other species. The upstream kinases that phosphorylate this site were first identified by whole kinome screens in yeast (Saccharomyces cerevisiae) , where three protein kinases with partially redundant functions were identified ( Hong, et al., 2003 ; Nath, et al., 2003 ; Sutherland, et al., 2003 ). Although there were no clear orthologs of these in the human kinome, searching using their kinase domain sequences revealed that the closest matches were to LKB1 and the two isoforms of the calmodulin-dependent kinase kinases, CaMKKα and CaMKKβ. Indeed, all three of these can act as upstream kinases in mammalian cells, although CaMKKα is less active that the other two ( Hawley, et al., 2003 ; Woods, et al., 2003 ; Shaw, et al., 2004 ; Hawley, et al., 2005 ; Hurley, et al., 2005 ; Woods, et al., 2005 ). Thr-172 can also be phosphorylated by transforming growth factor-β activated protein kinase-1 (TAK1) ( Momcilovic, et al., 2006 ; Herrero-Martin, et al., 2009 ), although the exact physiological significance of that mechanism remains uncertain. The discovery of LKB1 as an upstream kinase was particularly interesting because LKB1 had previously been identified as the product of a tumor suppressor gene ( Alessi, et al., 2006 ), and AMPK was its first downstream target to be identified. It was subsequently discovered that, in addition to the α1 and α2 subunits of AMPK, LKB1 also phosphorylates and activates twelve other enzymes with kinase domains closely related to those of AMPK, now known as the AMPK-related kinase or ARK family ( Lizcano, et al., 2004 ). LKB1 is only active in the form of a complex with two other subunits, STRAD and MO25 ( Hawley, et al., 2003 ). This complex appears to be constitutively active ( Sakamoto, et al., 2004 ), with Thr-172 phosphorylation normally being regulated instead by binding of ligands to the substrate, AMPK (see below). By contrast, CaMKKβ appears to have a very low basal activity against AMPK in intact cells, and only phosphorylates Thr-172 in response to treatments that increase intracellular Ca 2+ ( Hawley, et al., 2005 ; Hurley, et al., 2005 ; Woods, et al., 2005 ). The identities of the protein phosphatase(s) dephosphorylating Thr-172 are not fully resolved, although there is genetic evidence in S. cerevisiae that there are two phosphatases dephosphorylating this site. One is a complex between Glc7 (the single PP1 catalytic subunit) and its targeting subunit Reg1 ( Sanz, et al., 2000 ), while the other is Sit4, the ortholog of the mammalian phosphatase PP6 ( Ruiz, et al., 2011 ; Ruiz, et al., 2012 ). In MIN6 cells (a mouse pancreatic β cell line) evidence from siRNA knockdowns suggests that a complex between PP1 and the targeting subunit PPP1R6 dephosphorylates Thr-172 and, despite no clear sequence relationship, DNA encoding PPP1R6 complements a reg1 mutant yeast strain ( Garcia-Haro, et al., 2010 ). It has also been reported that RNAi-based depletion of PPM1E (a member of the PPM family of protein phosphatases, which are not related to the PPP family containing PP1 and PP6) increases Thr-172 phosphorylation in HEK-293 cells ( Voss, et al., 2011 ). The kinase domain (KD) of AMPK is immediately followed by a small region termed the auto-inhibitory domain (AID) ( Fig. 1 ). It was given this name because truncated α subunits containing the KD plus the AID have very low activity even when phosphorylated on Thr-172, while constructs containing the KD alone exhibit ful

The γ1, γ2 and γ3 subunit isoforms are encoded by three alternate genes in mammals ( PRKAG1-3 in humans). The γ1 isoform appears to be expressed ubiquitously, whereas γ2 and γ3 are most abundant in muscle. The γ2 and γ3 isoforms contain N-terminal extensions that are unrelated to each other and are not present in γ1 ( Fig. 1 ); in both cases these extensions can also occur as “long” and “short” forms due to the use of alternate transcriptional start sites ( Lang, et al., 2000 ; Yu, et al., 2004 ). Recently, a third γ2 variant (γ2-3B) has been identified that lacks the amino acids encoded by exons 1-3 in the “long” form but has 32 unique residues encoded by an alternate exon, 3B ( Pinter, et al., 2012 ). Both at the mRNA and the protein level, γ2 is expressed in adults in both cardiac and skeletal muscle, while γ3 appears to be restricted to skeletal muscle ( Cheung, et al., 2000 ; Milan, et al., 2000 ; Yu, et al., 2004 ). However, AMPK activity has been immunoprecipitated from other tissues, such as brain, using an anti-γ3 antibody ( Cheung, et al., 2000 ). Both γ3 and γ2-3B are also expressed in fetal mouse heart, and γ2-3B appears to account for almost half of total γ2 protein in adult mouse heart ( Pinter, et al., 2012 ). All three γ subunit isoforms contain at their C-termini four tandem repeats of a sequence known as a cystathione β-synthase (CBS) repeat; these are numbered CBS1-4 from the N- to the C-terminus ( Fig. 1 ). CBS1 is immediately preceded by a short sequence involved in the interaction with the β subunit ( Viana, et al., 2007 ; Xiao, et al., 2007 ). CBS repeats occur in around 20 other proteins in the human genome, invariably as tandem pairs. Single tandem pairs form structures known as Bateman domains, which are regulatory domains that bind adenosine-containing ligands, either AMP, ADP, ATP or, in the case of cystathione β-synthase itself, S-adenosyl methionine ( Scott, et al., 2004 ). The γ subunits of AMPK are unusual in having four rather than two repeats, and these form two Bateman domains that assemble in a head-to-head manner ( Amodeo, et al., 2007 ; Townley and Shapiro, 2007 ; Xiao, et al., 2007 ). Because of the four-fold symmetry, there are four clefts that could potentially bind adenine nucleotides; these are numbered according to the number of the CBS repeat bearing an aspartate residue that interacts with the nucleotide ribose ring. Site 2 is unoccupied (CBS2 lacks an aspartate), while site 4 contains AMP that is very tightly bound and non-exchangeable, and which may play a structural role. Sites 1 and 3 are the sites where AMP, ADP and ATP bind reversibly in competition with each other. They bind AMP, ADP and ATP with quite similar affinity, although site 1 appears to have a higher affinity for all three nucleotides ( Scott, et al., 2004 ; Xiao, et al., 2011 ).

AMPK phosphorylates serine residues surrounded by a well-defined recognition motif ( Scott, et al., 2002 ; Gwinn, et al., 2008 ). Space constraints do not permit a comprehensive discussion of the entire literature on downstream targets, and the reader is referred to other recent reviews ( Hardie, 2007 ; Steinberg and Kemp, 2009 ; Hardie, 2011 ; Mihaylova and Shaw, 2011 ; Hardie, et al., 2012 ). However, Figs 2 - 4 summarize some well characterized targets that are also briefly discussed below, with one or two representative references provided in each case. Fig. 2 involves targets involved in the acute regulation of metabolism. Progressing clockwise around the spokes of the “wheel”, starting at 12 o’clock: 1) AMPK activates GLUT4-mediated glucose uptake in muscle via phosphorylation of TBC1D1 ( Frosig, et al., 2010 ), a Rab-GAP protein that under basal conditions inhibits fusion of intracellular GLUT4-containing vesicles with the plasma membrane. Phosphorylation by AMPK triggers 14-3-3 binding, causing TBC1D1 to dissociate from the intracellular vesicles. 2) AMPK also activates glucose uptake in other cells by activation of GLUT1 already present on the membrane ( Barnes, et al., 2002 ). However, GLUT1 does not appear to be a direct target for AMPK, and the molecular mechanism remains unknown. 3) AMPK activates glycolysis in cardiac myocytes by phosphorylation and activation of the PFKFB2 isoform of 6-phosphofructo-2-kinase, which catalyzes the synthesis of fructose-2,6-bisphosphate, a key activator of the glycolytic enzyme 6-phosphofructo-1-kinase ( Marsin, et al., 2000 ). 4) AMPK also activates glycolysis in monocytes and macrophages by phosphorylation and activation of the PFKFB3 isoform of 6-phosphofructo-2-kinase. This may allow macrophages to generate ATP efficiently in regions of infection or injury that are hypoxic ( Marsin, et al., 2002 ). 5) AMPK activates fatty acid uptake via translocation of the transporter CD36 to the plasma membrane, although the molecular mechanism remains unknown ( Habets, et al., 2009 ). 6) AMPK activates fatty acid oxidation by phosphorylating and inactivating the mitochondria-associated isoform of acetyl-CoA carboxylase (ACC2), thus lowering malonyl-CoA, an inhibitor of fatty acid uptake into mitochondria via the carnitine:palmityl-CoA transferase system ( Merrill, et al., 1997 ). 7) Lipolysis (hydrolysis of triglyceride stores) in adipose tissue is catalyzed by desnutrin/ATGL, which removes the first fatty acid, and hormone-sensitive lipase (HSL), which removes the second ( Zimmermann, et al., 2004 ). AMPK inhibits lipolysis in rodent and human adipocytes by phosphorylation of HSL, antagonizing its activation by cyclic AMP-dependent protein kinase and its translocation to the surface of the lipid droplet ( Garton, et al., 1989 ; Daval, et al., 2005 ; Bourron, et al., 2010 ). Since lipolysis is a catabolic pathway, it might seem illogical that AMPK should inhibit it. However, this may be a mechanism to limit the production during lipolysis of free fatty acids; if these are allowed to accumulate they recycle into triglycerides, creating a futile cycle that consumes ATP. Consistent with this, AMPK is activated in 3T3-L1 adipocytes by cyclic AMP-elevating agents and this is associated with increases in cellular AMP:ATP ratios, but both effects are blunted by inhibition of lipolysis or of the acyl-CoA synthetase required for recycling of fatty acids back to triglycerides ( Gauthier, et al., 2008 ). Somewhat paradoxically, AMPK has been reported to phosphorylate and activate desnutrin/ATGL in mammalian cells ( Ahmadian, et al., 2011 ), although in nematode worms the enzyme appears to be inhibited by AMPK phosphorylation ( Narbonne and Roy, 2009 ), which is more consistent with the known effects of AMPK on HSL function in mammals. 8) AMPK inhibits fatty acid synthesis by directly phosphorylating and inactivating the cytosolic isoform of acetyl-CoA carboxylase (ACC1) ( Davies, et al., 1992 ); phosphospecific antibodies that recognize the key phosphorylation sites on both ACC1 and ACC2 are widely used as markers for AMPK activation. 9) AMPK inhibits triglyceride and phospholipid synthesis by causing inactivation of the first enzyme involved in their synthesis, glycerol phosphate acyl transferase (GPAT); it remains unclear whether GPAT is a direct substrate for AMPK ( Muoio, et al., 1999 ). 10) AMPK inhibits isoprenoid/cholesterol synthesis by direct phosphorylation and inactivation of 3-hydroxy-3-methylglutaryl CoA reductase (HMGR) ( Clarke and Hardie, 1990 ). 11) AMPK inhibits muscle glycogen synthesis by phosphorylation and inactivation of glycogen synthase-1 ( Jorgensen, et al., 2004 ). 12) AMPK inhibits liver glycogen synthesis by phosphorylation and inactivation of glycogen synthase-2 ( Bultot, et al., 2012 ). 13) AMPK inhibits protein synthesis by phosphorylating tuberous sclerosis complex protein-2 (TSC2), causing Rheb (a small G protein that activates mammalian target-of-

Metformin is currently the first choice drug for treatment of type 2 diabetes. Because of the very high prevalence of this disorder, it is currently prescribed to more than 100 million people, or more than 1% of the world population. Metformin is derived from guanidine which, along with galegine ( Fig. 5 ), is a natural product found in the plant Galega officinalis (also known as Goat’s Rue). Goat’s Rue was described in the 18 th century as a herbal medicine to treat “symptoms such as thirst and frequent urination” ( Hill, 1772 ). Interestingly, it is classed as a noxious weed in the USA because it is poisonous to herbivores, which presumably includes goats! Guanidine is quite toxic, but galegine (an isoprenyl derivative) is a potent AMPK activator ( Mooney, et al., 2008 ) and was tested for treatment of diabetes in humans, with some success, in the 1920s ( Muller and Reinwein, 1927 ). Metformin (dimethylbiguanide) is a biguanide derivative that was synthesized in the 1920s and found to be effective in lowering blood glucose in animals ( Hesse and Taubmann, 1929 ). However, the success of insulin as a therapy for diabetes, developed in the same decade, put further studies of the biguanide drugs on hold. Metformin and its sister drug, phenformin (phenethylbiguanide, Fig. 5 ), were finally introduced into clinical practice in the 1960s, by which time the distinction between type 1 and type 2 diabetes had been made. Phenformin was withdrawn from the US market in 1978 due to a rare but life-threatening side effect of lactic acidosis, but use of metformin has greatly increased, particularly when impressive results were reported in the UK Prospective Diabetes Study ( UKPDS, 1998 ). Although the major effect of metformin was known to be a reduction of the high hepatic glucose production in type 2 diabetics ( Hundal, et al., 2000 ), its molecular mechanism remained obscure until it was reported that it activated AMPK in isolated hepatocytes ( Zhou, et al., 2001 ). The drug did not, however, activate AMPK or affect its phosphorylation by upstream kinases and phosphatases in cell-free systems ( Hawley, et al., 2002 ), suggesting that it acted indirectly. Metformin inhibits Complex I of the respiratory chain ( El-Mir, et al., 2000 ; Owen, et al., 2000 ), suggesting that it might activate AMPK by inhibiting ATP synthesis and thus increasing cellular ADP and AMP. This was confirmed when it was found that RG cells were completely resistant to the effects of metformin to activate AMPK; the same was observed for phenformin and galegine. As expected from this mechanism, all three agents also increased the cellular ADP:ATP ratio and inhibited oxygen uptake ( Hawley, et al., 2010 ). There remains some controversy as to the extent to which the therapeutic actions of metformin are mediated by AMPK. In favor of this idea, mice with a conditional liver knockout of LKB1, where liver AMPK is no longer activated by metformin, failed to display the anti-hyperglycemic effects of metformin ( Shaw, et al., 2005 ). However, mice with a conditional liver knockout of both AMPK catalytic subunits (α1 and α2) still displayed some responses to metformin in vivo , and acute inhibition of glucose production by metformin in hepatocytes isolated from the mice was still observed ( Foretz, et al., 2010 ). A potential explanation of the latter effect is that fructose-1,6-bisphosphatase, one of the key enzymes of gluconeogenesis, is inhibited by AMP. By inhibiting mitochondrial respiration and increasing AMP and ADP, metformin could clearly also modulate AMP- or ADP-sensitive enzymes other than AMPK, such as fructose-1,6-bisphosphatase.

2-Deoxyglucose (2DG) is taken up into cells by glucose transporters and converted by hexokinase to 2-deoxyglucose-6-phosphate, which is not metabolized by glycolysis beyond that point. Although usually described as an inhibitor of glycolysis, it may activate AMPK in part by depleting ATP due to its rapid and uncontrolled phosphorylation by hexokinase. As expected from this mechanism, 2DG caused increases in ADP:ATP and failed to activate AMPK in RG cells, although unlike AMPK activators acting on mitochondrial function it did not inhibit oxygen uptake ( Hawley, et al., 2010 ). Interestingly, 2DG can be metabolized to some extent by the pentose phosphate pathway, and it has recently been suggested that the generation of NADPH by this route could represent an AMPK-independent mechanism by which 2DG protects cells against the effects of glucose deprivation ( Jeon, et al., 2012 ).

Phenobarbital is another drug that was in clinical use long before it was known to activate AMPK. It is a particularly effective inducer of cytochrome P450 enzymes that catalyze oxidation of hydrophobic drugs, a key initial step in the process that renders them more water soluble for excretion. The gene CYP2B6 , encoding one cytochrome P450 in humans, is induced by the nuclear hormone receptor CAR (constitutive androstane receptor). It has been reported that phenobarbital activates AMPK and that other AMPK activators (AICA riboside, metformin) also induce CYP2B6 in a human hepatoma cell line, while in hepatocytes from AMPK-null mice phenobarbital failed to induce Cyp2b10 , the mouse ortholog ( Rencurel, et al., 2005 ; Rencurel, et al., 2006 ). Thus, induction of this P450 enzyme by phenobarbital is mediated by AMPK activation. Experiments with wild type and RG mutant HEK-293 cells confirmed that phenobarbital activates AMPK, like biguanides and thiazolidinediones, because it is an inhibitor of the respiratory chain ( Hawley, et al., 2010 ). Mitochondria contain several very large multiprotein complexes, including the four respiratory chain complexes as well as the ATP synthase. It seems likely that there might be many binding sites within these complexes where hydrophobic, xenobiotic compounds could bind and cause some inhibition of ATP production. An interesting speculation is that AMPK is being used in this case as a general sensor of mitochondrial poisons that might need to be eliminated by cytochrome P450 metabolism.

Another small molecule activator of AMPK, termed PT1, was isolated via a screen of compounds that activated an AMPK-α1 construct containing just the KD and the AID ( Pang, et al., 2008 ). The compound activated the KD-AID construct and a complete α1β1γ1 heterotrimer but failed to activate a construct containing the KD alone, suggesting that its binding site was in the cleft between the KD and the AID. Consistent with this, PT1 caused phosphorylation of ACC in L6 myotubes without altering the AMP:ATP ratio, and also promoted ACC phosphorylation and lowered lipid content in HepG2 cells ( Pang, et al., 2008 ).

The evidence presented in this review suggests that AMPK-activating drugs are not only effective in treating type 2 diabetes, but might also be effective in providing protection against cancer. If this appears “too good to be true”, it does not stop there because it has recently also been suggested that AMPK activators reduce the accumulation of the Aβ peptides in cell culture models of Alzheimer’s disease, and thus might be useful for treatment of that disease ( Salminen, et al., 2011 ). One caveat is that, while there is growing evidence that AMPK provides protection against the development of cancer, it is perhaps an unjustified leap of faith to say that AMPK activators would also be useful in the treatment of cancer once it has arisen. Indeed, there is now evidence that AMPK protect cells against the metabolic stress caused by over-expression of c-Myc ( Liu, et al., 2012 ), while the LKB1-AMPK pathway also protects cultured cells against metabolic stress caused by glucose starvation ( Jeon, et al., 2012 ). Thus, AMPK activators might be both a “friend and foe” in cancer, providing protection against its development, while at the same time making tumor cells harder to kill using cytotoxic agents once tumors have arisen. Although it is generally easier to develop agents that are enzyme inhibitors rather than enzyme activators (as most indications for AMPK demand), the wide variety of agents that do activate AMPK when applied to intact cells show that development of AMPK activators is an achievable objective. Most of the existing agents activate AMPK indirectly by inhibiting ATP production, whether by inhibiting oxidative phosphorylation (metformin, phenformin, galegine, thiazolidinediones, resveratrol, berberine) or glycolysis (2-deoxyglucose), thus increasing cellular ADP:ATP and AMP:ATP ratios. Exceptions to this are: (i) AICA riboside (acadesine), a pro-drug that is converted to the active derivative, ZMP, which binds at the adenine nucleotide-binding sites; (ii) A-769662 and salicylate, which bind to a common site that, while not fully defined, is distinct from the adenine nucleotide-binding sites; and (iii) PT1, which appears to bind between the KD and the AID on the α subunit ( Pang, et al., 2008 ). Some of these, including acadesine and A-769662, have poor oral availability, but salicylate is orally available, and the best bet to develop novel AMPK activators may be to target the salicylate-binding site. Finally, one should not discount the utility of AMPK inhibitors, although the only chemical inhibitor targeting the kinase domain reported to date [compound C ( Zhou, et al., 2001 )] has very poor selectivity for AMPK ( Bain, et al., 2007 ). If AMPK does indeed protect tumor cells against cell death induced by cytotoxic agents, then combinations of AMPK inhibitors with cytotoxic drugs might be worth considering.