Structural basis of allosteric and synergistic activation of AMPK by furan-2-phosphonic derivative C2 binding

The metabolic stress-sensing enzyme AMP-activated protein kinase (AMPK) is responsible for regulating metabolism in response to energy supply and demand. Drugs that activate AMPK may be useful in the treatment of metabolic diseases including type 2 diabetes. We have determined the crystal structure of AMPK in complex with its activator 5-(5-hydroxyl-isoxazol-3-yl)-furan-2-phosphonic acid (C2), revealing two C2-binding sites in the γ-subunit distinct from nucleotide sites. C2 acts synergistically with the drug A769662 to activate AMPK α1-containing complexes independent of upstream kinases. Our results show that dual drug therapies could be effective AMPK-targeting strategies to treat metabolic diseases.

A MP-activated protein kinase (AMPK) is a metabolic stress-sensing kinase responsible for regulating metabolism in response to energy supply and demand. During metabolic stress the cellular AMP/ATP ratio increases leading to activation of AMPK, which in turn switches off energyconsuming anabolic pathways and switches on catabolic pathways to restore ATP levels. The AMPK abg heterotrimer comprises a catalytic a-subunit associated with b and g regulatory subunits (Fig. 1a). Adenine nucleotides (ATP, ADP and AMP) bind interchangeably to the g subunit where three of the four cystathionine b-synthase (CBS) tandem-repeat sequences provide nucleotide-binding sites 1 . Allosteric activation of AMPK by AMP appears to involve all three sites but site 3 may be the most important 2,3 . AMP binding initiates AMPK signalling by promoting phosphorylation of AMPK on a-Thr172 by upstream kinases LKB1 and Ca 2 þ /calmodulin-dependent protein kinase 2 (CaMKK2). Once phosphorylated on a-Thr172 AMPK is further activated by AMP, and AMP sustains AMPK signalling by inhibiting dephosphorylation of a-pThr172 (ref. 4). AMP stimulation of aThr172 phosphorylation is dependent on N-terminal myristoylation of the b-subunit, since we demonstrated that AMP stimulation by both upstream kinases is lost following b-G2A mutation 2 . ATP on the other hand opposes allosteric activation by AMP as well as promoting the dephosphorylation of a-Thr172 (ref. 5). The a-regulatory subunit-interacting motif 2 (aRIM2; Fig. 1a) 6 is required for sensing the adenine nucleotide bound state of the g-subunit and transducing this signal to the a-catalytic domain, resulting in either stimulation (AMP) or inhibition (ATP) of AMPK activity 1,6-8 .
There has been keen interest in developing AMPK-activating drugs for potential therapeutic use in treating metabolic diseases including type 2 diabetes, obesity and cardiovascular disease. A number of small-molecule activators have been identified 9 and two of these, A769662 (Abbott Laboratories) and 991 (Merck Sharp and Dohme Corporation and Metabasis Therapeutics), were shown in milestone structures to bind to a site formed between the small lobe of the a-subunit kinase domain and the b-subunit carbohydrate-binding module (CBM) 8,10 termed the allosteric drug and metabolite-binding site (ADaM) 11 . Salicylate, the active metabolite of aspirin, is also thought to bind to this site 12 . ADaM site stabilization is enhanced by phosphorylation of the b-subunit residue Ser108 (refs 13,14). We have shown that the need for b-pSer108 can be bypassed and that dephosphorylated AMPK (or dephosphorylation mimic a1-T172Ab1-S108Ag1 mutant) can be activated synergistically by A769662 and AMP whereas there is negligible activity with either activator alone 13 .
Another potent small-molecule allosteric activator of AMPK, 5-(5-hydroxyl-isoxazol-3-yl)-furan-2-phosphonic acid (compound 2, C2, Metabasis Therapeutics), was identified by screening an AMP mimetic library 15 . It is thought that C2 activates both AMPK a1 and a2 isoforms in cell-free assays by binding to the g-subunit. Although C2 is impermeable to cells, its corresponding di-iso-propyl phosphoester prodrug (compound 13, C13) preferentially stimulates AMPK a1 in mammalian cells 16 . AMP and C2 likely share a common AMPK-activating mechanism as their allosteric effects are not additive, and sensitivity to both is lost in the g2-R531G AMP-insensitive mutant 16 . The purpose of our study was to identify the C2-binding site(s) on AMPK and investigate the relationship between activation of AMPK by A769662 and C2.
We have determined the crystal structure of the C2: AMPK complex, revealing two C2-binding sites in the g-subunit distinct from nucleotide sites. C2 acts synergistically with the drug A769662 to activate AMPK a1-containing complexes independent of upstream kinases. Our results raise the possibility that dual drug therapies could be effective AMPK-targeting strategies to treat metabolic diseases.

Results
Allosteric activation of AMPK. We found that C2 activates both AMPK g1 and g2 complexes (a1b1g1 and a1b2g1; a1b1g2 and a1b2g2) to a similar degree as AMP, whereas neither C2 nor AMP-activated g3-containing complexes (a1b1g3 and a1b2g3) (Fig. 1b). Lack of AMP activation of g3-containing complexes has also been reported recently 17 but not in all studies 18,19 .
Since ATP concentration was previously shown to alter AMP dependence 5 , we investigated C2 activation of AMPK a1 at low (20 mM) and physiological (2 mM) ATP concentrations. Comparison of dose-response curves revealed that the half-maximal concentrations for C2 activation were similar at  Fig. 1c,d), implying that C2 and ATP binding is non-competitive. In contrast, previous studies have shown that the concentration of AMP required for halfmaximal activation increases with higher concentrations of ATP, which is consistent with AMP/ATP competition at the g-subunit allosteric sites 5 . We also observed that C2 activation becomes strongly co-operative at high ATP, as shown by a shift in the Hill coefficient from 1.4 ± 0.3 (20 mM ATP) to 2.3 ± 0.4 (2 mM ATP) (Fig. 1c,d). In contrast, AMP does not exhibit co-operative binding at high ATP concentrations (Hill coefficient 1.0 ± 0.2; Supplementary Fig. 1 and Supplementary Table 1). We are uncertain of the structural basis for the effect of ATP on C2 co-operative binding as well as the kinetic basis for ATP not influencing AMP activation in a co-operative manner. The g-subunit resembles a classic multimeric-enzyme with its 4 CBS tandem repeats. One interpretation is that ATP binding to the g-subunit influences C2 co-operativity but we have been unable to obtain crystals with both C2 and ATP bound to the g-subunit thus far. In the event ATP cannot bind to g in the presence of C2 we cannot formally eliminate the possibility that ATP binding at the kinase active site (a-subunit) rather than the g-subunit may induce C2 co-operative binding.
Novel C2-binding site in the c-subunit. To gain further insight into the mechanism of C2-mediated AMPK activation we solved the X-ray crystal structure of full-length a2b1g1 isoform cocrystallized with C2 and AMP to 2.99 Å resolution ( Table 2). The structure revealed two molecules of C2, bound within the solventaccessible core of the g-subunit. The phosphate groups of both C2 molecules overlap the phosphate binding sites of AMP in sites 1 and 4, while the 5-(5-hydroxyl-isoxazol-3-yl)-furan moieties occupy novel binding sites independent from the AMP-binding sites ( Fig. 3a-c; Supplementary Fig. 3a). The asymmetric unit contained two AMPK heterotrimers. One heterotrimer contained AMP bound at each of CBS sites 3 and 4 ( Fig. 3a, white) whereas the other heterotrimer was devoid of AMP and fortuitously contained two molecules of C2 instead (Fig. 3a-c). The two distinct liganded complexes allowed us to comprehensively probe the differences between AMP and C2 binding to AMPK. Our AMP-bound heterotrimer closely resembles the previously published AMP-bound a2b1g1 structure (PDB 4CFE) 8 . The two molecules superpose with a r.m.s. deviation of 0.55 Å over 884 Ca atoms, with key AMP-interacting residues having the same conformation. In contrast, the C2-bound molecule has an r.m.s. deviation of 0.75 Å (over 864 Ca atoms) when superimposed with 4CFE 8 . The largest structural difference between the C2 and AMP-bound heterotrimer is in the b-CBM and the N-lobe of the kinase domain, where a shift of up to 4.0 Å occurs (Fig. 3a). This shift reflects the formation of a canonical kinase salt bridge, formed between conserved residues a-Lys45 and a-Glu64, in the C2-bound heterotrimer. This important interaction is regarded as an active kinase conformation signature 20 and is conspicuously absent in previous a2b1g1 structures 8 . The C2-bound g1-subunit conformation closely mimics that of the AMP-bound complex. Despite the considerable chemical differences and different binding orientations between C2 and AMP the side chains of the g1-subunit residues are remarkably similar in conformation, with the notable exceptions being g-Arg70 and g-His298 ( Supplementary Fig. 4a,b). A region that does distinguish between the two heterotrimers is the nucleotide sensing a-subunit RIM2 motif 1,6,8 . Strong electron density was observed for the a-RIM2 in the AMP-bound heterotrimer, but only very weak electron density was observed for a-RIM2 in the corresponding C2-bound heterotrimer and could therefore not   Fig. 3b). This is a C2 specific effect as the previously published AMPK structure 4CFE (crystallized in the same space group (P21) and solved to an equivalent resolution of 3.02 Å) revealed strong electron density for both AMP-bound heterotrimers in the asymmetric unit 8 . Importantly, these data are consistent with the a2-subunit RIM2 being incompletely engaged with the g-subunit when C2 is bound.
The two C2 molecules occupy the interface between the CBS-binding sites 1, 3 and 4, with the phosphate group of C2 occupying a similar position to that of AMP-bound phosphate group in sites 1 and 4 (Fig. 3b). For this reason we have named specific C2-binding sites g-pSite-1 and g-pSite-4, referring to the equivalent AMP phosphate group position (Fig. 3b,c).
In both sites the C2 molecule makes six protein-mediated hydrogen bond contacts, while C2 in g-pSite-4 makes an additional p-stacking interaction with g-His298 (Fig. 3c). All of the residues that contact C2 are shared with AMP, yet the binding constant for C2 is 4100-fold higher than that for AMP 15,16 . The basis of C2's increased affinity is not obvious to us.
Mutational analysis of the C2-binding site. To validate the C2-binding sites, we expressed three AMPK g1 mutants (T89E, H151E and H298E) in COS7 mammalian cells and measured their sensitivity to C2 or AMP. We initially found that all the g1 mutants displayed increased but variable Thr172 phosphorylation and basal activities compared with the wild-type (WT)  g1-containing AMPK complexes. We therefore co-expressed the mutants with the phospho-mimetic AMPK a1(T172D), which is more amenable to studying the effect of these g mutations on allosteric activation ( Supplementary Fig. 2b) 5,21 . C2 and AMP activated the AMPK a1(T172D) b1g1 enzyme to a similar extent (E 8-and 10-fold, respectively; Fig. 4a). The g-T89E and g-H298E mutants had increased basal activities (9.5±0.2 and 11.4 ± 0.2 nmol min À 1 mg À 1 lysate for g-T89E and g-H298E, respectively) in the absence of C2 and AMP compared with g1 WT (1.4 ± 0.1 nmol min À 1 mg À 1 lysate), and were insensitive to further activation by either C2 or AMP (Fig. 4a). It is possible that substituting for a negatively charged glutamate residue at the T89 and H298 positions mimics the effect of the phosphate groups of AMP or C2, resulting in constitutive activation. This would also explain the increased Thr172 phosphorylation observed with these mutants. The gH151E mutation completely abolished AMPK activation by C2 (Fig. 4a,c), while only partially reducing AMP-dependence and sensitivity compared to the wild-type control (Fig. 4a,b).

C2-induced conformational rearrangements of the c-subunit.
A superposition of the two AMPK heterotrimers reveals two major side-chain conformational changes between AMP and C2 binding. Firstly, the imidazole ring of the aforementioned g-His298 has flipped 180°and rotated at the Cb490°(B 6.0 Å) to form a p-stacking interaction with C2 in g-pSite-4 while simultaneously blocking site 3, mimicking ATP-bound AMPK (PDB 4EAK) 3 (Supplementary Fig. 4a). The conformational difference of the g-His298 side chain is critical for AMP or C2 binding. Secondly, the position of C2 at g-pSite-1 causes the sidechain guanidine group of g-Arg70 to shift approximately 4 Å away from the AMP-bound conformation. In this new conformation, g-Arg70 would clash with critical AMP-sensing residue a-Glu368 from the aRIM2 (Supplementary Fig. 4b).
The clash may cause a-Glu368 to lose hydrogen bonds to site 3 AMP-interacting residues g-Arg70 and g-Lys170, that tether a-Glu368 to the g-subunit 1,6,8 . In turn, this causes the aRIM2 disengagement. The loss of RIM2 association with the g-subunit leads to the loss of protection from a-Thr172 dephosphorylation and a concomitant reduction in Vmax with C2 as shown by Hunter et al. 16 . In support of this proposal, previous studies have shown that a-Glu368 mutation results in a comparable reduction in Vmax with AMP stimulation and a-Thr172 dephosphorylation protection [6][7][8] .
AMPK isoform specificity of C2 allosteric activation. Secondary structure elements of aRIM2 are crucial for the isoform specificity of C2. Structural comparison of a1 and a2 isoform aRIM2/ g-subunit interactions reveal the a-helical a1RIM2 (PDB 4CFH) 1 forms a strong interaction (three hydrogen bonds and six saltbridges) 22 with the g-subunit in a groove between g-helix a10 and g-helix a4, formed upon AMP or C2 binding ( Supplementary  Fig. 3b,c). In contrast, a2 isoforms have a loop motif (PDB 4ZHX and 4CFE) 8 that forms a single hydrogen bond and four salt-bridges 22 with the g-subunit (Supplementary Fig. 3b,c). We propose the weaker interacting a2 isoform aRIM2 is incompletely engaged upon C2 binding relative to the a1 isoform, resulting in the a-isoform specificity for allosteric and synergistic activation and protection from a-Thr172 dephosphorylation. We solved the X-ray crystal structure of a heterotrimeric a2/a1 RIM chimaera, a2(1-347)/a1(349-401)/a2(397-end) b1g1 (which has previously been shown to completely restore maximal allosteric activation and C2-mediated protection of a-Thr172 dephosphorylation 16 ), co-crystallized with C2 to 2.99 Å (pdb: 5EZV; Table 2). Unlike the aforementioned crystal structure (4ZHX), a2/a1 RIM swap chimaera was crystallized in the absence of AMP, revealing two C2-bound AMPK heterotrimers in the asymmetric unit. Both of the C2-bound heterotrimers (5EZV) closely resembled the previous C2-bound heterotrimer (molecule 1 4ZHX), with an r.m.s deviation of 0.520 over 869 Ca atoms (molecule 1 5EZV). The C2 molecules occupy the same binding space and g-subunit residue interactions as the AMP/C2-bound structures (4ZHX). Importantly the structure reveals density for seven a1 RIM 2 residues (Fig. 5), including the main-chain of the nucleotide sensing a-E364 (a2 E368), further supporting our proposal that the a1 RIM2 interacts more strongly with the g-subunit.
Interestingly there was weak density for the carboxyl group of the E364 side-chain, possibly indicating a transient interaction with the g-subunit residue R70.

Discussion
In this study we showed that two topographically disparate drug sites on AMPK, one located at the classic a-kinase domain/ b-CBM interface (ADaM site) and the other located within the solvent-accessible core of the g-subunit (g-pSite-1/g-pSite-4), can be exploited to achieve synergistic activation of unphosphorylated a1-AMPK independent of AMP and upstream phosphorylation events. This provides a major advance to previous demonstrations of AMPK synergistic activation that have been dependent on AMP. Importantly, identification of C2-binding sites on the g-subunit represents an entirely unexplored dimension for future rational drug design.
To evaluate the impact of g-mutation on C2/AMP activity (Fig. 4): FLAG-a1(T172D)b1g1 (wild-type and indicated mutants) AMPK complexes were immobilized on anti-FLAG M2 affinity Agarose gel ( Supplementary Fig. 2b) before SAMS activity assay. The expression levels of the g-mutants were assessed by immunoblot. Membrane was probed overnight at 4°C with a FLAG primary antibody (1:2,000 dilution in PBS-T) labelled with the fluorescent dye IR680 (LI-COR Biosciences) and probed for 1 h at room temp with HA primary antibodies (1:1,000 dilution in PBS-T) fluorescently labelled with IR800 and analysed as above.
AMPK activity assay. AMPK activity assay was conducted as described previously 13 . Briefly, AMPK complexes were purified as above and washed three times with a 40:1 v/v ratio of assay buffer/resin (assay buffer: 50 mM Hepes pH 7.4,