The AMP (adenosine 5′-monophosphate)-activated protein kinase (AMPK) is a key regulator of cellular and whole-body energy homeostasis that co-ordinates metabolic processes to ensure energy supply meets demand. At the cellular level, AMPK is activated by metabolic stresses that increase AMP or adenosine 5′-diphosphate (ADP) coupled with falling adenosine 5′-triphosphate (ATP) and acts to restore energy balance by choreographing a shift in metabolism in favour of energy-producing catabolic pathways while inhibiting non-essential anabolic processes. AMPK also regulates systemic energy balance and is activated by hormones and nutritional signals in the hypothalamus to control appetite and body weight. Failure to maintain energy balance plays an important role in chronic diseases such as obesity, type 2 diabetes and inflammatory disorders, which has prompted a major drive to develop pharmacological activators of AMPK. An array of small-molecule allosteric activators has now been developed, several of which can activate AMPK by direct allosteric activation, independently of Thr172 phosphorylation, which was previously regarded as indispensable for AMPK activity. In this review, we summarise the state-of-the-art regarding our understanding of the molecular mechanisms that govern direct allosteric activation of AMPK by adenylate nucleotides and small-molecule drugs.
Critical to the survival of all organisms is the ability to offset the supply of environmental nutrients and energy production with metabolism, growth and proliferation. In eukaryotes, this ability is largely conferred by the AMP (adenosine 5′-monophosphate)-activated protein kinase (AMPK), an evolutionarily conserved serine/threonine protein kinase that functions as a metabolic sensor and regulator of cellular and whole-organism energy homeostasis. AMPK directly senses shortfalls in energy supply, manifested as either elevated AMP:ATP and ADP:ATP ratios or a fall in intracellular glucose, and modulates multiple metabolic pathways to restore optimal adenylate energy charge [1–3]. Consequently, AMPK targets numerous key enzymes involved in metabolic processes, including synthesis of fatty acids, cholesterol and proteins, mitochondrial biogenesis, fat oxidation, autophagy, and skeletal muscle glucose uptake, with the overarching effect of down-regulating anabolic pathways, while up-regulating catabolic pathways . Roles for AMPK have more recently emerged in anti-inflammation, onco-suppression, appetite regulation, circadian rhythm and longevity, extending AMPK's influence beyond its traditional metabolic scope. The therapeutic potential of AMPK is thus firmly established for treatment of insulin resistance, type 2 diabetes (T2D), obesity and cardiovascular disease, with AMPK now also gaining traction as a promising drug target for cancer, non-alcoholic fatty liver disease (NAFLD), neurodegeneration and pathogenic infections [5–8]. Recent advances in our understanding of allosteric control of AMPK signalling by adenylate nucleotides and small-molecule drugs, in conjunction with insight into regulation by upstream kinases, will be of great benefit in driving the realization of this potential (Figure 1).
Allosteric regulation of AMPK.
AMPK complexes and domains
Human AMPK heterotrimeric complexes are composed of a catalytic α subunit (isoforms α1 and α2) and two regulatory β (isoforms β1 and β2) and γ (isoforms γ1, γ2 and γ3) subunits, giving rise to 12 possible combinations (α1β1γ1 through α2β2γ3) that possess markedly varied expression profiles between tissues and organs [9,10]. Each α-subunit contains an N-terminal kinase domain, followed by an autoinhibitory domain (AID), regulatory subunit-interacting motifs (RIM1/2) and a C-terminal region that tethers the α-subunit to the AMPK regulatory core (Figure 2) . The N-termini of both β isoforms are myristoylated at position Gly2, a modification important for temporospatial regulation of AMPK [12,13], followed by a non-conserved region of ∼60 residues of unknown function, a mid-molecule carbohydrate-binding module (CBM) and a C-terminal scaffolding domain that anchors α- and γ-subunits . Each γ isoform contains four conserved tandem cystathionine-β-synthase (CBS) motifs, which function as energy-sensing adenylate nucleotide-binding sites . AMPK complexes containing different γ-subunit isoforms respond distinctively to changes in adenylate charge, which may be an evolutionary adaptation to allow AMPK to respond to energy stress over a wider range of conditions . The γ2 and γ3 isoforms contain different N-terminal extensions of ∼270 and ∼190 residues, respectively, with unknown function ; the γ2 extension contains numerous phosphorylation sites, hinting at possible dynamic regulatory roles.
AMPK allosteric sites
The molecular understanding of AMPK allosteric regulation has been significantly advanced by a combination of biochemical studies and resolution of the crystallographic structures of AMPK (yeast and mammalian) complexed to adenylate nucleotides and small-molecule effectors, which together provide a detailed architectural overview of the domains and binding determinants that contribute to allosteric binding pockets [17–20]. The allosteric sites can be broadly separated into three groups — those encapsulated entirely by the γ-subunit (γ-sites, occupied by adenylate nucleotides or the furan-2-phosphonic derivative C2), a carbohydrate-binding surface formed entirely by the β-CBM, and a binding pocket formed between the β-CBM and N-lobe of the α kinase domain (targeted by synthetic drugs, e.g. A-769662, 991, PF-739, SC4 and R734) [19–22]. The latter has been termed the Allosteric Drug and Metabolite (ADaM) site , in recognition of speculation, this may also be a binding site for an endogenous metabolite [20,24] (Figure 2).
The direct bioenergetic sensing capacity of AMPK is imparted by partially exchangeable adenylate nucleotide-binding sites located in the γ-subunit. Structural information is currently only available for the γ1 isoform, however, conservation of hydrophobic and positively charged residues that contribute to nucleotide binding by contacting the adenine ring or ribose phosphates, respectively , combined with mutational analysis, suggests that interpretation can credibly be extrapolated to isoforms-2 and -3. Each CBS motif forms a potential nucleotide-binding site, providing the γ-subunit with a total of four possible sites [26,27]. These sites (1–4) have been numbered according to the CBS motif that contributes the Asp residue that hydrogen bonds to the 2′- and 3′-hydroxyls attached to the ribose sugar of the adenylate nucleotide [12,28] (Figure 1). The structures of the AMPK regulatory core obtained from crystals soaked with either AMP or ATP  suggest that only three sites (1, 3 and 4, respectively) contribute to nucleotide regulation. Site 2 is unoccupied in all mammalian AMPK structures solved to date, and collectively all data at our disposal indicates no regulatory role for this site, possibly because Arg replaces Asp at the corresponding hydroxyl-coordinating position. AMP bound to γ-site 4 of mammalian AMPK has been assigned as non-exchangeable because the nucleotide co-purified with the regulatory fragment ; however, an ATP co-crystallized structure subsequently revealed ATP can occupy this site, coincident with ATP binding to site 1 and vacancy of site 3 . ATP, ADP (adenosine 5′-diphosphate) and AMP are exchangeable at the other two sites (1 [AMP low-affinity site] and 3 [AMP high-affinity site]), implicating them as key mediators of AMPK regulation [2,11]. In addition, co-operativity between nucleotide sites has been shown by mutational analyses ; AMP binding to one site promotes binding to another, whereas disruptive mutation of site 4 abolished AMP allosteric activation, indicating AMP occupancy of this site improves energy-sensing efficiency by stabilizing AMP binding to site 3 [25,29]. Besides adenine nucleotides, both NADH and NADPH (adenine dinucleotides) have been shown to interact with γ-sites [11,29]. NADH suppresses AMP allosteric activation of purified enzyme, and NADPH was shown by hydrogen–deuterium exchange mass spectrometry (HDX-MS) to bind to γ-site 3, however, the effects of NADH/NADPH interactions on AMPK regulation in vivo remain unclear.
The furan-2-phosphonic acid derivative C2 is a potent allosteric activator of α1-AMPK complexes, with structural studies subsequently revealing C2 binds in tandem to novel sites within the γ-subunit solvent accessible core. These sites are distinct from, but partially overlap, the adenylate nucleotide sites and involve many common binding determinants [18,30], in line with biochemical analysis demonstrating non-additive allosteric effects between AMP and C2. In particular, the two phosphate groups of bound C2 molecules occupy a similar position to that of ribose phosphates bound at sites 1 and 4.
Debate surrounds which of AMP or ADP, if either, is the primary AMPK-activating nucleotide [4,31]; the problem being exacerbated by technical difficulties associated with accurately measuring [AMP] from cell or tissue lysates. Many studies measure [AMP] indirectly from [ADP] and [ATP] measurements (generally with an HPLC-based method) using the assumption that the adenylate kinase reaction (which interconverts 2ADP ↔ ATP + AMP) is at equilibrium . Under this premise, it is simple to show that [AMP] fluctuates as the square of [ADP], indicating AMP provides the more sensitive read-out of cellular energy stress . However, this oversimplification discounts efficient clearance mechanisms, driven by AMP deaminase and 5′-nucleotidase, to rapidly convert intracellular AMP to IMP and inosine. The relative and temporal contributions made to AMPK activation and signalling by AMP and ADP remains an important and open question in the field, one which is likely to be addressed with the application of ultrasensitive liquid chromatography–mass spectrometry (LC–MS) techniques to relatively measure [AMP], [ADP] and [ATP] simultaneously from a single biological sample [3,32]; initial results from LC–MS-based observations indicate that, in fact, absolute [AMP] fluctuates to an almost similar extent as [ADP] in cultured cells subjected to a cellular energy stress [3,32]. While an extensive study has shown that AMP is more potent than ADP in triggering certain AMPK regulatory mechanisms (see below) , [ADP] is consistently 5- to 10-fold higher than [AMP]. Combined with our previous arguments  we reiterate the view that ADP is a major positive regulator of AMPK activation.
Increased intracellular AMP:ATP and ADP:ATP ratios, arising from physiological (exercise, nutrient stress) or pathological processes (hypoxia) trigger exchange of Mg-ATP for AMP/ADP at γ-sites 1 and/or 3, regulating AMPK signalling via a tripartite mechanism, each of which is antagonized by ATP (Figure 1). First, AMP/ADP binding stimulates phosphorylation of Thr172 (pThr172) in the α-subunit kinase activation loop, by either of two upstream kinases, liver kinase B1 (LKB1) and possibly Ca2+/calmodulin-dependent protein kinase kinase-2 (CaMKK2) [2,12]. The sensitivity of CaMKK2-mediated Thr172 phosphorylation to AMP/ADP has been challenged in two studies [1,16]; an explanation for this discrepancy is currently unclear but could possibly be attributed to the CaMKK2 isoform (of which there are 7 known) used in the studies or differences in the assay conditions. Nevertheless, this step critically requires myristoylation of the β-subunit, as recombinant AMPK purified from Escherichia coli (which does not have endogenous N-myristoylation capacity) or non-myristoylated β-G2A mutants extracted from mammalian cells are insensitive to AMP/ADP stimulation of Thr172 phosphorylation [2,12,33]. The current model depicts AMP/ADP binding de-repressing molecular stabilization induced by ATP, resulting in the ejection of the myristoyl group from an intracellular binding pocket, mapped by HDX-MS to the kinase C-lobe in the ATP-bound complex . This is akin to the myristoyl switch mechanism employed by other proteins, e.g. recoverin, HIV-1 Gag and ADP-ribosylation factor 1 . While details of the precise mechanism are still to be resolved, AMP/ADP binding presumably increases accessibility of the activation loop, as well as promoting myristoyl group-dependent targeting of AMPK to lysosomal membranes (possibly via association with AXIN scaffolds), the major cellular site of Thr172 phosphorylation due to co-localisation with LKB1 [35,36]. However, AMP-independent mechanisms may also be in play to promote lysosomal/endosomal AMPK targeting . Second, AMP/ADP sustains AMPK activity due to suppression of pThr172 dephosphorylation by phosphatases, a mechanism involving γ-site 3 ‘sampling’ by the α regulatory motif RIM2 (in particular, α residues Glu364 and Arg365) [11,37]. Structural and biochemical data suggest a hypothetical model in which AMP binding to site 3 results in recruitment of α-RIM2 to the γ surface, leading to a conformational change that buries the α activation loop in an interface between kinase domain and regulatory core. In this arrangement, phosphatase access to pThr172 is presumably limited, possibly mediated by a partially unresolved β loop (residues 179–194) in proximity to this interface since the removal of β residues 1–187 abolishes dephosphorylation protection by AMP . Contradictory evidence to this model is provided by a structure of α2β1γ1 complexed with C2, in which the activation loop adopts a similar conformation to that seen with AMP and is also buried against the regulatory core, even though C2 is not effective in protecting pThr172 from dephosphorylation in α2 AMPK complexes [18,30]. Hence, it is clear that we do not yet have a full understanding of the molecular mechanism by which AMP and ADP suppress pThr172 dephosphorylation. Thirdly, phosphorylation of Thr172 sensitizes AMPK to ≤10-fold allosteric stimulation (depending on subunit composition) by AMP but not ADP [11,38]. Allosteric activation is also mediated through RIM2 detection of AMP at γ-site 3, but is not dependent on the β-CBM, with structural evidence suggesting the α-RIM2/γ interaction results in sequestration of the α-AID, which maintains the kinase domain in an ‘open’, inactive state under energy replete conditions, away from its interactions with both small and large lobes of the kinase domain, to de-repress AMPK activity . The net effect of these three regulatory mechanisms is 100- to 1000-fold increase in AMPK activity, although in vivo it is likely that only a small fraction of the entire AMPK cellular pool is active at any one time point.
The ADaM site was first structurally defined following co-crystallization of AMPK α2β1γ1 complexed to the high-affinity activator 991 , and since confirmed as the allosteric binding site for a range of second-generation AMPK drugs (PF-739, complexed to α1β1γ1 ; SC4 complexed to α2β1γ1 and α2β2γ1 ), the aspirin metabolic product salicylate  and most likely the allosteric antagonist MT47-100 . The site is a predominantly hydrophobic 15 Å long channel, formed at the interface between β-CBM and the small lobe of the α-subunit kinase domain which, with respect to β1 complexes, is stabilised by phosphorylation of the β1 residue Ser108 through electrostatic interactions between phosphate and basic α residues Lys31 and Lys33 . Ser108 phosphorylation is either critical for activation by low-affinity ADaM site agonists (e.g. salicylate, A-769662) or further enhances potency of high-affinity agonists (e.g. 991, SC4) [19,20,32,41,42]. Other important binding determinants have been characterized as β1-Arg83, which forms a π-stacking interaction with the benzimidazole rings of PF-739 and 991 or the imidazopyridine ring of SC4, and β2-D111, which confers SC4 sensitivity to β2-AMPK, abundantly expressed in skeletal muscle and responsible for triggering glucose clearance in response to exercise [19–21]. In recent major breakthroughs, small-molecule activation of skeletal muscle AMPK has been validated as a strategy to treat T2D [21,43]; these AMPK activators fall within a broader class of molecules we have termed importagogs (substances that induce or augment uptake of another substance into cells or tissues .
Similar to AMP, ADaM site agonists allosterically activate AMPK (e.g. SC4: ∼2.5-fold maximal activation with α1 complexes, ∼5-fold with α2 complexes) and protect α-pThr172 from dephosphorylation (Figure 1). How ADaM site occupancy leads to AMPK activation is presently unclear; structural and mutational evidence suggests activating compounds recruit a β-subunit helical motif, C-terminal to the CBM termed the C-interacting helix, to the C-helix in the kinase domain small lobe important for integrity of the ATP active site, thereby inducing a ‘closed’, activated form of the enzyme [17,20,44]. This mechanism of activation is reminiscent of that employed by cyclin-dependent kinases following interaction with cyclin regulators.
AMPK activation independent of α-Thr172 phosphorylation
Recent biochemical and cellular evidence indicates that, in contrast with the previously held tenet of AMPK activation, significant enzyme activity can be achieved in the complete absence of activation loop phosphorylation (Figure 1) [32,41]. This has profound consequences for AMPK research since Thr172 phosphorylation has for many years been used as the standard metric for assessing AMPK activity in vivo or in cells. In 2014, we discovered, rather surprisingly, that AMPK α1β1γ1 expressed in E. coli can be significantly activated by the ADaM site agonist A-769662 without prior treatment with LKB1 or CaMKK2, an effect that was lost by pre-incubation and dephosphorylation with λ-phosphatase. The sensitizing phosphorylation site was identified as β1-Ser108, autophosphorylation of which likely accumulated during expression of AMPK as a result of low intrinsic activity and lack of pSer108 phosphatases in E. coli. We now know that β1-Ser108 is autophosphorylated by a cis event that requires prior AMPK activation through Thr172 phosphorylation. Nonetheless, Ser108 phosphorylation persists in cells for a longer duration than the rapidly dephosphorylated Thr172 following removal of the activating stimulus, presumably because of reduced phosphatase pressure . This raises an intriguing scenario in which distinct AMPK complexes remain sensitive to ADaM site ligands (drugs/endogenous metabolite(s)) long after they lose the ability to respond to AMP allosteric activation (loss of Thr172 phosphorylation).
A further layer of allosteric regulation emerged from an even more surprising observation; that AMPK devoid of both Thr172 and Ser108 phosphorylation could be activated >1000-fold when co-incubated with AMP and A-769662, but not by either alone . We subsequently demonstrated allosteric synergism following co-incubation with A-769662 and C2 (AMP mimetic) or AMP and SC4 [18,19], however, the effect did not extend to ADP/A-769662 co-incubation, was specific for α1β1 AMPK complexes and was dependent on the integrity of all three functional γ nucleotide sites. These observations using purified AMPK preparations, with which phosphosites can be strictly controlled, indicate a highly co-ordinated intramolecular regulatory mechanism: we hypothesize that AMP binding at γ-sites leads to stabilization of the ADaM site located >70 Å away, with drug binding presumably resulting in stabilization of the activation loop in the open conformation to permit ATP and substrate binding. Recent crystal structures of unphosphorylated AMPK in complex with AMP and ADaM site agonists, in which the activation loop adopts the active-like conformation despite lack of Thr172 phosphorylation, lends support to this mechanistic model [22,45]. Although synergistic activation has the potential to trigger substantial AMPK activity, particularly since the majority of the cellular AMPK pool is retained in the unphosphorylated state during periods of low energy stress, the physiological relevance of Thr172-independent signalling, and the contribution it makes to total AMPK signalling if any, is unclear. We previously demonstrated that Thr172-independent AMPK signalling in cells only expressing an α2(T172A) mutant following co-incubation with A-769662 and phenformin (an AMP-inducing agent) , however, a report demonstrating absolute dependence of AMPK signalling on Thr172 phosphorylation in cells lacking LKB1 and CaMKK2, in response to 991 and 2-deoxyglucose (elevates AMP/ATP through inhibition of glycolysis), provides conflicting evidence . Regardless, the mechanism provides potential combinatorial treatment strategies for disease states in which AMPK activation is beneficial, but coincides with loss of upstream kinase activity, for example, genetic deletion of LKB1 in a variety of cancer models such as non-small cell lung carcinoma , or down-regulation of CaMKK2 associated with schizophrenia and bipolar disorder .
ULK1 ligand switching
Current models of AMPK allosteric regulation depict scenarios in which sensitivity to discrete sites is mediated by dynamic phosphorylation events; Thr172 phosphorylation is required for significant AMPK enzyme activity following AMP allosteric activation, whereas sensitivity to ADaM site agonists is facilitated or enhanced by Ser108 phosphorylation [32,41]. We recently made the observation that catalytically inactive AMPK complexes are phosphorylated at Ser108 in response to glucose starvation, A-769662 or phenformin . Given Ser108 is cis-autophosphorylated, this strongly indicated that Ser108 was also a substrate for trans modification, and we subsequently identified the autophagy initiator ULK1 as a cellular upstream kinase for Ser108. ULK1 activity is triggered by AMPK- and mTOR-dependent processes in response to nutrient stress, leading to elevated autophagic flux through phosphorylation and recruitment of proteins involved in autophagosome formation and maturation . Thus, ULK1 stabilizes the ADaM site and sensitizes β1-complexes to A-769662 and in a previous study, ULK1 signalling was also found to suppress AMPK activity by an unknown mechanism that resulted in decreased Thr172 phosphorylation . Combined, the dual effects of ULK1 signalling on AMPK regulation can be predicted to reduce sensitivity to AMP (lowered Thr172 phosphorylation) whilst simultaneously sensitizing AMPK to an endogenous ADaM site metabolite, in a ligand switching mechanism as described for other allosterically regulated proteins, e.g. nuclear receptors and β3 integrins [50,51].
Importance of the field: AMPK is a critical regulator of cellular and whole-body energy balance that has been proposed to play a role in major diseases including obesity, T2D, inflammation-associated disorders and cancer. Accordingly, AMPK has emerged as a promising drug target for both the prevention and treatment of these diseases. In this regard, understanding the molecular mechanisms that govern AMPK allosteric regulation is crucial not only for advancing our fundamental knowledge of this important metabolic regulator but also to leverage the development of selective, small-molecule activators as potential therapeutics.
Summary of current thinking: Thr172 phosphorylation has long been regarded as the major mechanism by which AMPK is activated and was thought to be an absolute requirement to initiate downstream signalling. On the other hand, direct allosteric regulation was considered a minor component of the overall activation mechanism, however, we now know that AMPK can be activated entirely allosterically and independently of Thr172 phosphorylation by simultaneously occupying the ADaM and γ-sites with A-769662 and AMP (or C2), respectively [18,41]. Whether there are physiological conditions where AMPK is activated purely allosterically remains unknown, however, it provides proof of concept that combinatorial therapies could be exploited to activate AMPK, particularly in disease states where upstream kinase signalling is impaired.
Comment on future directions: An outstanding question in the AMPK field is whether drugs that target the allosteric ADaM site mimic a natural ligand that stimulates AMPK activity. Although such a ligand has yet to be identified, there is compelling evidence to indicate it may exist. A structural commonality of ADaM site agonists is a phenylphenol moiety that resides within the hydrophobic channel bordered by α- and β-subunits ; it is, therefore, tempting to speculate that a natural AMPK ligand also possesses a long hydrophobic moiety. Two studies have shown that AMPK can be allosterically activated by coenzyme-A (CoA) conjugates of fatty acid analogues [52,53], whose acyl chains could be envisaged to bind in the large hydrophobic cavity in the ADaM site . Indeed, a CoA-conjugate of the fatty acid analogue bempedoic acid selectively activates AMPK heterotrimers containing the β1-subunit isoform, analogous to the effects of the ADaM site activator A-769662 (Figure 1). However, other binding determinants accommodating the CoA group must be involved since the non-esterified bempedoic acid, which bears a striking resemblance to palmitic acid, does not activate β1 AMPK . Little is known about the mechanisms by which these fatty acyl-CoA conjugates activate AMPK, and is an unexplored area that represents an important missing chapter in our understanding of AMPK regulation.
allosteric drug and metabolite
AMP-activated protein kinase
Ca2+/calmodulin-dependent protein kinase kinase-2 (CaMKK2)
liver kinase B1
type 2 diabetes
A.L.d.S.A.M. was supported in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001. J.W.S. and J.S.O. are supported by National Health and Medical Research Council project grants [1138102 and 1145265, respectively], St. Vincent's Institute of Medical Research and in part by the Victorian Government's Operational Infrastructure Support Program.
The Authors declare that there are no competing interests associated with the manuscript.