Structure-function analysis of the AMPK activator SC4 and identification of a potent pan AMPK activator

The AMP-activated protein kinase (AMPK) αβγ heterotrimer is a primary cellular energy sensor and central regulator of energy homeostasis. Activating skeletal muscle AMPK with small molecule drugs improves glucose uptake and provides an opportunity for new strategies to treat type 2 diabetes and insulin resistance, with recent genetic and pharmacological studies indicating the α2β2γ1 isoform combination as the heterotrimer complex primarily responsible. With the goal of developing α2β2-specific activators, here we perform structure/function analysis of the 2-hydroxybiphenyl group of SC4, an activator with tendency for α2-selectivity that is also capable of potently activating β2 complexes. Substitution of the LHS 2-hydroxyphenyl group with polar-substituted cyclohexene-based probes resulted in two AMPK agonists, MSG010 and MSG011, which did not display α2-selectivity when screened against a panel of AMPK complexes. By radiolabel kinase assay, MSG010 and MSG011 activated α2β2γ1 AMPK with one order of magnitude greater potency than the pan AMPK activator MK-8722. A crystal structure of MSG011 complexed to AMPK α2β1γ1 revealed a similar binding mode to SC4 and the potential importance of an interaction between the SC4 2-hydroxyl group and α2-Lys31 for directing α2-selectivity. MSG011 induced robust AMPK signalling in mouse primary hepatocytes and commonly used cell lines, and in most cases this occurred in the absence of changes in phosphorylation of the kinase activation loop residue α-Thr172, a classical marker of AMP-induced AMPK activity. These findings will guide future design of α2β2-selective AMPK activators, that we hypothesise may avoid off-target complications associated with indiscriminate activation of AMPK throughout the body.


Introduction
AMP-activated protein kinase (AMPK) is a conserved serine/threonine protein kinase that acts as a metabolic fuel sensor and is crucial for maintaining cellular energy homeostasis [1,2]. It is a heterotrimer composed of a catalytic α subunit, an N-terminally myristoylated regulatory β subunit and a nucleotide-sensing γ subunit, each with multiple isoforms (α1, α2, β1, β2, γ1, γ2, γ3) providing the potential to form 12 distinct complexes. The capacity of AMPK to monitor cellular energy status is provided by three exchangeable adenine nucleotide binding sites on the γ-subunit. Increases in AMP/ATP and ADP/ATP adenylate ratios arising from energy stress cause either allosteric activation or increases in phosphorylation on the α-subunit activation loop residue threonine-172 (α-Thr172) by liver kinase B1 (LKB1) or calcium/ calmodulin-dependent kinase kinase 2 (CaMKK2) [3,4]. Under conditions of energy or nutrient stress, AMPK directs metabolism towards ATP-conserving (catabolic) pathways and away from ATP-consuming (anabolic) pathways. It does this by directly phosphorylating and regulating rate-limiting enzymes, and transcription factors regulating their expression, in multiple key biochemical pathways such as fatty acid oxidation and synthesis, mitochondrial biogenesis, skeletal muscle glucose uptake, cholesterol synthesis and autophagy. These pleiotropic effects place AMPK as a promising drug target for the treatment of diseases such as type 2 diabetes mellitus (T2DM), metabolic syndrome, cancer, neurodegeneration and cardiovascular disease [5][6][7][8]. Indeed, some of the pleiotropic effects of the biguanide metformin, the first line treatment for T2DM, have been attributed to AMPK-dependent mechanisms through inhibition of complex 1 in the mitochondrial electron transport chain [9,10]. This leads to impaired ATP production and a consequent increase in AMP/ATP ratio, resulting in canonical nucleotide-dependent activation of AMPK primarily through increased α-Thr172 phosphorylation. Indirect AMPK activation is a hallmark of a large group of natural and synthetic agents that trigger AMPK signalling by inhibiting either mitochondrial function or glycolysis to induce metabolic stress [5].
AMPK isoforms display tissue-specific expression profiles therefore activator isoform selectivity raises opportunities to preferentially target subsets of AMPK complexes. α1, β1 and γ1 isoforms are found in multiple human tissues while α2β2γ1 and α1β2γ1 together make up the majority of AMPK complexes in skeletal muscle [24]. Expression of γ2 is predominantly limited to the heart, whereas γ3 is found almost exclusively complexed to ∼30% of α2β2 complexes in glycolytic skeletal muscle. The importance of activator selectivity to prevent indiscriminate activation throughout the body is demonstrated by the pan AMPK activator MK-8722, which induced robust improvements in key disease hallmarks, such as systemic glucose clearance and improved glycaemic control, through activation of β2AMPK in skeletal muscle of rodent and non-human primate models of T2DM [17]. However, MK-8722 also induced reversible cardiac hypertrophy in both lean and diabetic animal models, attributed to increased ventricular wall area and glycogen content. Although the investigation into off-tissue target effects of MK-8722 was not reported, it can be inferred that MK-8722 induced hypertrophy resulted from activation of AMPK in the heart, since activating mutations in the AMPK γ2 subunit, expressed in the heart, are linked to excessive cardiac glycogen accumulation in Wolff-Parkinson-White syndrome [25][26][27][28].
Recent in vivo studies have shown that MK-8722, PF-739 and 991 induced skeletal muscle glucose uptake independently of α2β2γ3 [29,30]. This important finding, combined with our previous report showing β2-dependent increases in ex vivo glucose uptake in mouse soleus and extensor digitorum longus (EDL) muscles by the imidazopyridine SC4, implicate α2β2γ1 complexes as the physiological target for ADaM site drugs in skeletal muscle [21,29,30]. Furthermore, SC4 activates all six α2 complexes in vitro but demonstrates poor activation potential against α1β2γ1. Our study, supported by computational analysis of the molecular interaction networks involved, demonstrated the importance of the β2-specific residue Asp111 and the SC4 imidazopyridine 4-nitrogen atom for β2 activation by SC4, however we know very little regarding the determinants mediating apparent α2 selectivity [21,31]. Here we describe preliminary SAR analysis of the SC4 phenylphenol and reveal its contribution to the α isoform discriminating properties of this compound. Our findings will aid efforts to develop clinically viable, glucose-controlling drugs through specific activation of α2β2γ1 in skeletal muscle.

Biochemical analysis of pan AMPK activators
We directly compared in vitro potencies of pan activators by performing radiolabel kinase assays with γ1AMPK complexes. Initially, we used GST-tagged γ1AMPK expressed in HEK293T/17 cells and immobilised on glutathione-Sepharose ( Figure 2 and Table 2). Calculated EC 50 values for MSG012 activation were 103.1 nM  with α1β1γ1 and 328.0 nM with α2β2γ1, thus we were unable to demonstrate greater potency for β2 over β1AMPK that was previously described for this compound [32]. EC 50 values for MSG011 ranged from 140.4 nM (α1β1γ1) to 553.4 nM (α2β1γ1). Under these assay conditions, EC 50 values for γ1AMPK activation by MK-8722 (1.2-5.6 mM) and PF-739 (2.0 mM) were 100-1000-fold higher than previously determined by high-throughput, fluorescence-based assays performed in solution (Table 2) [17,18]. These assays used AMPK purified from Sf9 insect cells or E. coli that was pre-treated with CaMKK2. Reduced potency of MK-8722 in our assays was not due to quality of the compound, which we validated by NMR (Supplementary Figure S2) [34], nor was it due to steric hindrance caused by AMPK immobilisation since GST-α1β1γ1 in solution demonstrated similar activation kinetics (Supplementary Figure S3A). However, relative to immobilised GST-α2β1γ1 from HEK293T/17 cells, His-tagged and myristoylated α2β1γ1 AMPK (expressed in E. coli, CaMKK2-treated and assayed in solution) was at least 10-fold more sensitive to activation by both MK-8722 and MSG011 (Supplementary Figure S3B), indicating the GST-affinity tag or CaMKK2 treatment may exert some influence on sensitivity to these activators. MK-8722 at concentrations below 1 mM was previously reported to stimulate AMPK signalling in HepG2 cells and C2C12 myotubes in the absence of increased α-Thr172 phosphorylation [17], which is often used as a marker of AMPK activation by compounds targeting the allosteric ADaM site. MK-8722 above 1 mM elevated α-Thr172 in these cells, although this did not translate to further robust increases in phosphorylation of ACC. To explore this dose-dependent response, we used HEK293T/17 cells transiently expressing WT γ1AMPK or the AMP-insensitive γ1 mutant R299G (Supplementary Figure S4A). Phosphorylation of α-Thr172 was significantly increased in WT γ1-expressing cells by 2.5 mM MK-8722, however the effect was largely abrogated in γ1-R299G-expressing cells. One explanation is that 2.5 mM MK-8722 caused indirect activation of AMPK (α-Thr172 phosphorylation) by inducing a metabolic stress leading to rises in AMP/ATP and ADP/ATP ratios. We examined the effect of MK-8722 incubation on C2C12 mitochondrial function (Supplementary Figure S4B-I). Incubation with 10 mM, but not 1 mM, MK-8722 significantly impaired basal respiration, ATP production, maximal respiration, spare respiratory capacity, non-mitochondrial oxygen consumption and coupling efficiency. Impaired respiration was not due to loss of cell viability was determined by total protein stain (Supplementary Figure S4J).

MSG011 activation of AMPK in cells occurs primarily via an allosteric mechanism
To evaluate the efficacy of MSG011 in cellulo, we incubated mouse primary hepatocytes with 2.5 mM MSG011 for 60 min and tracked changes in phosphorylation of AMPK α-Thr172 (marker of activation) and the AMPK substrate acetyl-CoA carboxylase (ACC) Ser79 (AMPK signalling marker). Both MSG011 and phenformin, a mitochondrial complex I inhibitor that indirectly activates AMPK by dramatically depleting cellular adenylate energy charge (AEC), induced similar increases in phosphorylation of ACC-Ser79 ( Figure 3A and Supplementary Figure S5A). However, unlike phenformin, MSG011 activation of AMPK signalling in hepatocytes occurred independently of significant net phosphorylation of AMPK α-Thr172. MSG011 (2.5 mM) also triggered AMPK signalling (measured by phosphorylation of ACC-Ser79 and the alternate AMPK substrate  analysis was performed by one-way ANOVA with post hoc Dunnett's multiple comparison test. **** P < 0.0001 indicates significant reduction in adenylate energy charge relative to vehicle. Antibodies used are described in Supplementary Table S3. raptor-Ser792) at levels comparable to extreme energy stress in HEK293T and COS7 cell lines, and significantly activated AMPK in commonly used cancer cell lines HeLa and PC3 ( Figure 3B-E and Supplementary Figure S5B-E). In HEK293T and PC3 cells these effects were independent of changes to α-Thr172 phosphorylation. Using LC-MS to directly measure relative cellular levels of AMP, ADP and ATP [16], we found that AEC was unaffected by incubation of HEK293T cells with MSG011 up to 40 mM ( Figure 3F), indicating that activation of AMPK by this compound was not due to elevated cellular ADP or AMP concentrations.
Structural comparison of AMPK/drug complexes reveals binding determinants potentially important for α2 selectivity To structurally characterise the MSG011/AMPK interaction and investigate chemistry for blocking β1 binding we solved the co-crystal structure of activated α2β1γ1 ( phosphorylated on α2-Thr172 and β1-Ser108) in complex with MSG011 at a resolution of 2.95 Å (Supplementary Table S1). As expected, MSG011 was docked in the ADaM site ( Figure 4A and Supplementary Figure S6A) where clear electron density was observed for the drug in both AMPK heterotrimers in the asymmetric unit (Supplementary Figure S6B,C). Both MSG011 molecules were built in similar poses, however, missing electron density around the 2-methylbenzoic acid indicates . This was also a feature of the SC4/AMPK complex structure [21]. Interestingly, α2-Lys29 is in close proximity to this ring and was unable to be completely modelled due to poor electron density around the ε-amino group ( Figure 4B and Supplementary Figure S6A), hence it is plausible that it is participating in both hydrophobic interactions with the 2-methylbenzoic acid of MSG011 through the hydrophobic carbon tail, and hydrogen bonding with phosphorylated β1-Ser108 through its positively charged ε-amino group. There is also flexibility around the acetamido group, which was modelled in different conformations in each heterotrimer (Supplementary Figure S6A-C). This chemical group is not present in previously published ADaM site-directed activators and occupies a novel space at the entrance of the ADaM site, however it does not appear to interact with any AMPK residues that are resolved in this structure ( Figure 4B and Supplementary Figure S6D). MSG011 differs from the α2 selective SC4 only in the LHS terminal ring, with addition of an acetamido and loss of aromaticity and a 2-hydroxyl group. We previously showed that the SC4 2-hydroxyl forms hydrogen bonding with α2-Lys31 [21], suggesting that loss of this interaction confers MSG011 with the ability to activate all AMPK complexes, and conversely that the 2-hydroxyl group is important for α2 selectivity of ADaM site compounds. It is important to note that other ADaM site compounds (SC4, A-769662 and R739) also possess a 2-hydroxyl group in the LHS terminal ring and don't display α2 selectivity, however due to differences in various other components of these drugs the reason for lack of selectivity is uncertain. Given that the majority of ADaM site drugs dock in a similar way, it seems likely that subtle global changes around the ADaM site pocket, both proximal and distal to the binding location, may contribute to isoform specificity and potency.

Discussion
Recent biochemical and genetic data strongly implicate skeletal muscle α2β2γ1 AMPK as the molecular target by which ADaM site drugs stimulate glucose uptake in these tissues. However, detailed structure-function analyses of molecular components underpinning isoform selectivity are lacking as most series reported in the patent literature were solely assayed against α1β1γ1. Here, we show modification of the SC4 LHS phenyl group contributes to α isoform selectivity, although it remains unclear whether the SC4 2-hydroxylphenyl is a negative determinant for α1 activation, or the MSG010 and MSG011 substitutions are positive determinants. Other pan activators MK-8722, PF-739 and I-3-40 do not possess the 2-hydroxy modification of the biphenyl LHS fragment, although they all possess D-mannitol groups in place of the MSG011 2-methylbenzoic acid RHS moiety, which may also contribute to loss of isoform selectivity. Our structural analysis was unable to identify residues with the potential to interact with the MSG011 acetamide group, although it is worth noting that unresolved α and β N-terminal residues are likely in close proximity and may contribute to the ADaM binding pocket in manners specific to α/β isoform combinations. This could also explain why SC4 activates some α1β1 complexes but not α1β2 complexes [21], whereas other contributions, in particular β1-Asn111 and β2-Asp111 are also known to act as key factors in modulating sensitivity of β1and β2-containing AMPK complexes [31]. MSG011 is a pan AMPK activator that effectively stimulated AMPK signalling in a panel of primary, immortalised and cancer cells. In most cases, the extent of AMPK signalling induced by MSG011 was comparable to that induced by severe energy stress. In hepatocytes, HEK293T and PC3 cells, MSG011-induced AMPK signalling was not accompanied by significant increases in phosphorylation of α-Thr172, nor perturbed adenylate nucleotide ratios in HEK293T cells. Thus, MSG011 acts primarily by allosterically enhancing intrinsic AMPK activity rather than by inducing energy stress or protecting phosphorylated α-Thr172 from dephosphorylation, although the latter mechanism may play a minor role in some cell types. Further structure/function analyses of α2β2-selective AMPK activators are warranted to aid the development of novel treatment strategies for major human metabolic diseases. We were surprised by the discrepancy between our calculated EC 50 values for MK-8722 and those previously reported, although some difference in calculated activating potency must be expected given extensive inter-assay variation. Possible reasons are numerous and include the uncharacterised influence of affinity tags used to purify AMPK, substrate composition, assay conditions, kinase detection method and source of recombinant AMPK that influences regulatory post-translational modifications. In terms of the latter, AMPK activation by extended CaMKK2 treatment, common practice in high throughput screening platforms [17,18,35], generates preparations with supraphysiological autophosphorylation of β-Ser108 (>95% vs. <10% basal β-pSer108 stoichiometry from mammalian cells [36,37]). Phosphorylation of β1-Ser108 stabilises the ADaM site and is required for AMPK sensitisation to lower potency activators A-769662, MT47-100, salicylate, long-chain fatty acyl CoA esters and lusianthridin [12,15,16,20,35,37], where the loss of phosphorylation via exchange for Ala reduces activating potency of SC4 by ∼4-fold and 991 by 40-fold [20,21]. To our knowledge the influence of β1-S108 phosphorylation on AMPK activation kinetics by MK-8722 and PF-739 has not been reported, however a similar sensitising effect is likely for all high potency pan activators and consequently assays using highly activated material may be expected to output more potent kinetics than those using more physiologically relevant preparations. It remains largely unknown what influence other abundant phosphorylation sites (e.g. β1-Ser182, α2-Ser345) and modifications (e.g. β-subunit myristoylation, missing in AMPK prepared from E. coli) have on drug sensitisation.
Increasing evidence indicates that ADaM site-targeting activators may induce energy stress at high concentrations, leading to phosphorylation of α-Thr172 by canonical AMP-dependent mechanisms (most recently investigated by Sanders and colleagues [35]) and supporting our previous recommendation that α-pThr172 should not be used in isolation as a marker of AMPK drug bioactivity [37]. Elevated cellular α-pThr172 is often used to demonstrate on-target action of AMPK drugs and is explained mechanistically by a drug-induced conformational change that makes α-pThr172 resistant to dephosphorylation. Here, we found that 2.5 mM MK-8722 induced robust phosphorylation of α-Thr172 in HEK293T/17 cells expressing WT γ1, an effect diminished by ∼55% in cells expressing the AMP-insensitive γ1 mutant R299G. Since A-769662-induced protection of α-pThr172 to phosphatases is retained in the γ1-R299G mutant [38], the most likely explanation in WT-expressing cells is that elevated AMP, arising from mitochondrial toxicity with 2.5 mM MK-8722, accounted for the majority of drug-induced increase in α-Thr172 phosphorylation. However, it is not a simple process to accurately measure the extent of energy stress induced by AMPK drugs; detecting changes in mitochondrial function (e.g. Seahorse) or adenylate nucleotide ratios (e.g. LC-MS) requires elaborate platforms and, in our experience, can produce confounding results since AMPK activators themselves will likely increase the ability of the cell to buffer against loss of mitochondrial ATP production. Relying on the inherent biological property of AMPK to respond to even small fluctuations in AMP/ADP/ATP ratios, a simpler option would be to routinely compare candidate AMPK activators in WT AMPK cell lines with their γ1-R299G stably expressing counterparts to reveal purely allosteric effects on signalling [15].

General experimental details for organic synthesis
Unless stated specifically, all chemicals were purchased from commercial suppliers and used without purification. All reactions were conducted in oven-dried glassware under nitrogen atmosphere. Progress of reactions was tracked by TLC and was performed on silica gel 60 F254 aluminium sheets (0.25 mm, Merck). 1 H (400.13 MHz) and 13 C NMR (100.62 MHz) NMR spectra for each compound were collected from a Bruker Avance III Nanobay spectrometer with a BACS 60 sample changer using deuterated solvents from Cambridge Isotope Laboratories. Chemical shifts (δ, ppm) are reported relative to the solvent peak ( . Coupling constants are reported in hertz (Hz), and the following abbreviations are used to assign the multiplicity of the 1 H NMR signal: s = singlet; bs = broad singlet; d = doublet; t = triplet; q = quartet; quin = quintet; dd = doublet of doublets; m = multiplet. Analytical HPLC was performed on an Agilent 1260 Infinity analytical HPLC coupled with a G1322A degasser, G1312B binary pump, G1367E high-performance autosampler, and G4212B diode array detector. Conditions were as follows: Zorbax Eclipse Plus C18 rapid resolution column (4.6 × 100 mm) with UV detection at 254 and 214 nm, 30°C; the sample was eluted using a gradient system, where solvent A was 0.1% aq. TFA and solvent B was 0.1% TFA in MeCN (5−100% B [9 min], 100% B [1 min]; 0.5 ml/min). High-resolution mass spectra were acquired on an Agilent 6224 TOF LCMS coupled to an Agilent 1290 Infinity LC. All data were acquired and reference mass corrected via a dual-spray electrospray ionisation (ESI) source. Each scan or data point on the total ion chromatogram (TIC) is an average of 13 700 transients, producing a spectrum every second. Mass spectra were created by averaging the scans across each peak and subtracting the background from first 10 sec of the TIC. Acquisition was performed using the Agilent Mass Procedures for the preparation of MSG008 and MSG009 and their intermediates Methyl 5-((6-chloro-5-iodo-1-((2-(trimethylsilyl)ethoxy)methyl)-1H-imidazo [4,5-b] pyridin-2-yl)oxy)-2-methylbenzoate (1) The title compound was prepared using the literature procedure [21], and it was obtained as a colourless oil (783 mg, 1.  [39] (882 mg, 4.10 mmol, 1 equiv), potassium acetate (1.21 g, 12.3 mmol, 3.0 equiv), B 2 pin 2 (1.53 g, 6.03 mmol, 1.5 equiv) and Pd(dppf )Cl 2 ·CH 2 Cl 2 (165 mg, 202 mmol, 5 mol%) were added to a round bottom flask then evacuated and back-filled with nitrogen gas three times. DMF (9.00 ml) was bubbled with nitrogen gas for 10 min before added to the substrates. The reaction mixture was heated at 120°C overnight then concentrated under reduced pressure. The dried residue was dissolved in ethyl acetate

Metabolite extractions
HEK293T cell cultures grown in six-well plates were gently washed in ice cold PBS, lysed with 150 ml of ice cold 0.5 M perchloric acid (Univar) and clarified by centrifugation (16 000g, 3 min, 4°C). An amount of 75 ml of clarified lysate was neutralised with 25 ml of ice cold 2.3 M KHCO 3 (Sigma-Aldrich), incubated on ice for 5 min and then centrifuged (16 000g, 3 min, 4°C) [16]. Supernatants were collected for analysis by liquid chromatography mass spectrometry-mass spectrometry (LC-MS/MS).

Small molecule mass spectrometry
Adenine nucleotide metabolites were measured by LC-MS/MS with modifications to our previously described method [16]. A QTRAP 5500 mass spectrometer (AB Sciex) was operated with the turbo V ion source linked to Prominence UFLC XR LC-20ADXR pumps (Shimadzu), SIL-20AC HT autosampler (Shimadzu) and CTO-20A HPLC column oven (Shimadzu). Both LC and MS instruments were controlled and managed with the Analyst 1.7.1 software (AB Sciex). Nitrogen was provided by a Genius NM3G nitrogen gas generator (PEAK Scientific). The autosampler was set at 4°C and column oven set at 30°C, which housed a 150 mm (length) × 0.5 mm (inner diameter) Hypercarb 3 mm porous graphitic carbon column (Thermo Fisher Scientific). The LC solvent system comprised of 50 mM triethylammonium bicarbonate buffer (TEAB, Sigma-Aldrich) pH 8.5 in pump A, and acetonitrile with 0.5% trifluoroacetic acid (TFA; Sigma-Aldrich) in pump B. A flow rate of 400 ml min −1 was used throughout a gradient program consisting of 0% B (2 min), 0 to 100% B (10 min), 100% B (3 min), 0% B (2 min). Data was analysed with MultiQuant 3.0.2 software (AB Sciex) using area under the LC curve. Calibration curves were determined by linear regression of the peak area ratio of each nucleotide and were required to have a correlation coefficient (R 2 ) of >0.98. The MS and multiple reaction monitoring (MRM) values were optimised by separate infusion of 1 mg ml −1 solution in 50 mM TEAB at a flow rate of 50 mg ml −1 (Supplementary Table S2). All data were acquired in negative ion mode with the spray voltage set to −4500 V, source temperature set to 250°C, ion source gas 1 and 2 set at 30 and 60, respectively, curtain gas set at 20 and collision gas set to high. AEC was calculated from the ratios of [