AMP-activated protein kinase selectively inhibited by the type II inhibitor SBI-0206965

Inhibition of the metabolic regulator AMP-activated protein kinase (AMPK) is increasingly being investigated for its therapeutic potential in diseases where AMPK hyperactivity results in poor prognoses, as in established cancers and neurodegeneration. However, AMPK-inhibitory tool compounds are largely limited to compound C, which has a poor selectivity profile. Here we identify the pyrimidine derivative SBI-0206965 as a direct AMPK inhibitor. SBI-0206965 inhibits AMPK with 40-fold greater potency and markedly lower kinase promiscuity than compound C and inhibits cellular AMPK signaling. Biochemical characterization reveals that SBI-0206965 is a mixed-type inhibitor. A co-crystal structure of the AMPK kinase domain/SBI-0206965 complex shows that the drug occupies a pocket that partially overlaps the ATP active site in a type IIb inhibitor manner. SBI-0206965 has utility as a tool compound for investigating physiological roles for AMPK and provides fresh impetus to small-molecule AMPK inhibitor therapeutic development.


Inhibition of the metabolic regulator AMP-activated protein kinase (AMPK) is increasingly being investigated for its therapeutic potential in diseases where AMPK hyperactivity results
in poor prognoses, as in established cancers and neurodegeneration. However, AMPK-inhibitory tool compounds are largely limited to compound C, which has a poor selectivity profile. Here we identify the pyrimidine derivative SBI-0206965 as a direct AMPK inhibitor. SBI-0206965 inhibits AMPK with 40-fold greater potency and markedly lower kinase promiscuity than compound C and inhibits cellular AMPK signaling. Biochemical characterization reveals that SBI-0206965 is a mixedtype inhibitor. A co-crystal structure of the AMPK kinase domain/SBI-0206965 complex shows that the drug occupies a pocket that partially overlaps the ATP active site in a type IIb inhibitor manner. SBI-0206965 has utility as a tool compound for investigating physiological roles for AMPK and provides fresh impetus to small-molecule AMPK inhibitor therapeutic development.
The ability to maintain energy homeostasis during acute or chronic periods of nutrient shortfall is an essential characteristic of all living organisms. A direct molecular link between nutrient supply and energy demand is provided by AMP-activated protein kinase (AMPK), 5 a key regulator of cellular and whole-body metabolism (1,2). AMPK is an evolutionarily conserved serine/threonine kinase that senses, and is activated by, low adenylate charge (elevated AMP/ATP and ADP/ATP ratios) resulting from energetically demanding processes such as muscle contraction or reduced energy supply caused by hypoxia or nutrient deprivation. AMPK redirects cellular metabolism by down-regulating numerous ATP-consuming anabolic processes (e.g. protein, cholesterol, and fatty acid synthesis) and up-regulating similarly diverse ATP-producing catabolic processes (e.g. fat oxidation, glycolysis, and autophagy) to restore energy balance. It does this by direct phosphorylation, either modulating the activities of rate-limiting enzymes in multiple metabolic processes or regulating transcriptional activities of the factors governing their expression. For example, AMPK phosphorylation of cytosolic acetyl-CoA carboxylase 1 (ACC1) and mitochondrial-associated ACC2, inhibits de novo lipogenesis and promotes fat oxidation, respectively. AMPK signaling has also been associated with a range of nonmetabolic regulatory roles (e.g. circadian rhythm, mitochondrial fission, and appetite control).
The metabolic dimensions associated with major human diseases, such as type 2 diabetes, cancer, and inflammatory disorders, have encouraged efforts to develop small-molecule AMPK activators. Patented examples now number in the hundreds. One of the first pharmacological AMPK activators discovered was 5-aminoimidazole-4-carboxamide ribonucleoside (AICAR) (12). AICAR is taken up by cells via the adenosine transport system (13) and converted to the monophosphorylated derivative ZMP, which functions as an AMP-mimetic. Other AMPK agonists can be broadly classified as direct activators (those that bind to drug sites located either between the kinase domain and ␤-CBM (e.g. A-769662, salicylate, 991, and PF-937 (14 -17)) or within the ␥ subunit (e.g. C2 (18,19)) or as indirect activators (those that commonly induce energy imbalance through mitochondrial toxicity, including metformin, xenobiotics, and other natural products).
Pharmacological AMPK inhibition provides potential strategies to treat obesity (appetite suppression), type 2 diabetes (enhanced insulin secretion), and stroke (neuroprotection) (11), and AMPK hyperactivity has also been linked to pathogenesis of neurodegeneration (20). The role of AMPK in cancer is complex; whereas initial studies demonstrated a tumor-suppressive role, AMPK signaling also contributes to the metabolic adaptations associated with tumor growth (e.g. increased glycolytic flux (the Warburg effect) and maintenance of ATP and NADPH) and promotes anchorage-independent proliferation (21)(22)(23). AMPK also promotes autophagic processes, via phosphorylation of ULK1, to maintain homeostasis in the neoplastic cell (24). Thus, AMPK is considered pro-tumorigenic under certain circumstances, underpinning the attraction of AMPK inhibition as a strategy for cancer treatments.
Current availability of small-molecule AMPK inhibitors, either for clinical application or as research tools to delineate AMPK's physiological roles, is extremely limited. By far the most widely applied AMPK inhibitor, the pyrazolopyrimidine derivative compound C (dorsomorphin), was originally selected from a high-throughput screen and used to confirm AMPK-dependent effects of AICAR and metformin in cultured hepato-cytes (25). Compound C is an ATP-competitive inhibitor and binds to the highly conserved active site of AMPK (26). However, in vitro screening has shown that compound C is promiscuous, inhibiting multiple kinases with similar or greater potency than AMPK (27). Numerous studies have since described off-target or AMPK-independent cellular effects, including inhibition of bone morphogenetic protein type I receptors ALK2, ALK3, and ALK6 (28), hypoxia-induced HIF-1 activation (29), preadipocyte proliferation (30), and macrophage chemotaxis (31). Compound C also blocks AICAR cellular uptake through competition for adenosine transporter binding sites, which largely accounts for its suppressive effects on AICAR-mediated AMPK activation (13). In light of these undesirable properties, conclusions drawn from using compound C as an AMPK inhibitor are now viewed with extreme caution, and recommendations are to avoid its application altogether (27). Other characterized direct antagonists include MT47-100, a low-potency, ␤2-AMPK allosteric inhibitor that intriguingly also activates ␤1-complexes (32), and SU6656, that paradoxically stimulates net cellular AMPK signaling by promoting phosphorylation of Thr-172 by LKB1 (33).
Here, we report the discovery of the small-molecule SBI-0206965 as a direct, type IIb AMPK inhibitor that demonstrated improved potency and kinase selectivity relative to compound C. A crystal structure of the AMPK ␣2 kinase domain/SBI-0206965 complex revealed a binding pocket that partially overlapped the ATP site; however, SBI-0206965 displayed mixed-competitive kinetics. SBI-0206965 inhibited AMPK signaling in a variety of cell types. Our study also highlights limitations pertaining to SBI-0206965 when used in conjunction with AICAR.
To assess kinase selectivity in terms of activity inhibition, we profiled SBI-0206965 (0.25 M) against a diverse panel of 50

High-potency AMPK inhibitor
High-potency AMPK inhibitor kinases (Fig. S1A). Compound C (2.5 M) was screened in parallel for direct comparison (Fig. S1B). AMPK in this panel was assayed at 20 M ATP, hence the need to profile inhibitors below the IC 50 concentrations determined in our assays at 200 M ATP (Fig. 1, C and D). Of the kinases in the panel, AMPK was inhibited to the greatest extent by both SBI-0206965 (22 Ϯ 2% residual activity) and compound C (4 Ϯ 1%); however, SBI-0206965 displayed a preferable selectivity profile, inhibiting activity of only five other kinases by Ͼ50% (compared with 23 kinases with compound C) and seven other kinases by Ͼ30% (compared with 31 kinases with compound C) ( Fig. 2 and Fig.  S1). SBI-0206965 inhibited LKB1 and CaMKK2 activities by 4 and 17%, respectively (Fig. 2).

SBI-0206965 is a mixed-type AMPK inhibitor
We investigated the inhibitory mechanism of SBI-0206965 by performing dose-response assays at different ATP concentrations using ␣1␤1␥1 AMPK. SBI-0206965 IC 50 S2C). This is in contrast to the apparent ␣K i (also commonly termed K i Ј) of 892.3 nM (Fig. S2D), which indicates that SBI-0206965 has a lower affinity for AMPK when ATP is bound. Equally, ATP has a lower affinity for AMPK when SBI- In the context of AMPK regulation, these kinetic properties of SBI-0206965 are consistent with mixed-type inhibition.
To confirm the binding mode of SBI-0206965, we solved a 2.9 Å resolution crystal structure of the inhibitor complexed to ␣2 kinase domain (residues 6 -278), in which the activation loop Thr-172 was mutated to the phosphomimetic Asp (Table S1). A similar crystallization construct was used previously to visualize the compound C-binding pocket (26) (PDB entry 3AQV) and adopted a structure essentially identical to that of the Thr-172-phosphorylated WT kinase domain in the heterotrimeric complex (16,19). SBI-0206965 was found to occupy a pocket located between the kinase N-and C-lobes and hinge region, which partially overlaps the compound C-binding site ( Fig. 4A and Fig. S3A). Specifically, the trimethoxyphenyl and pyrimidinyl rings of SBI-0206965 lie approximately within the same plane as, and overlap substantially with, the phenyl and pyrazolo[1,5-a]pyrimidinyl rings, respectively, of the compound C core. The positioning of the electronegative bromine atom of SBI-0206965, probably contributing to increased potency relative to compound C, was confirmed by anomalous scattering at 13.6 keV (Fig. 4B) that revealed a large single peak for bromine (Table S1). The bromine moiety was found to occupy a large cavity bordered by residues Val-30, Lys-45, Ile-77, Met-93, Ala-156, and Asn-162.
We could not unambiguously place the N-methylbenzamide group of SBI-0206965 within the electron density; therefore, we

High-potency AMPK inhibitor
modeled two distinct conformations with equal occupancy (Fig. 4C). In one position (conformation A), the aromatic ring is rotated almost 80°with respect to the compound plane, and the N-methyl group extends into the space occupied by the catalytic loop, Mg 2ϩ -chelating residue Asn-144, and the ATP ␣-phosphate in structures of active kinases (35). In the alternate position (conformation B), the benzamide ring is rotated 180°r elative to conformation A, and the N-methyl group is directed toward P-loop residues Leu-22 and Gly-23 (Fig. 4C). In both conformations, SBI-0206965 makes two electrostatic contacts with the main chain of Val-96 in the kinase hinge region; otherwise, drug binding is mainly stabilized by hydrophobic contacts with hinge (Tyr-95), N-lobe (Leu-22, Ala-43, and the gate-keeper residue Met-93), and C-lobe (Ile-77, Gly-99, Glu-100, Leu-146, and Ala-156) residues.
Comparisons with either apo-(PDB entry 2YZA), compound C-complexed (26), or activated (PDB entry 4ZHX) (19) kinase domain structures reveal the ␣2/SBI-0206965 complex contains many of the hallmarks of an unproductive kinase. In both compound C-and SBI-0206965-bound structures, the C-␣-helix adopts a "swung-out" position, ensuring that the glutamatelysine salt bridge, required for efficient phosphoryl transfer, is unformed (Fig. S3A). These residues (Lys-45 and Glu-64) are instead interdigitated by the activation loop residue Ser-161, hydrogen bonding with Glu-64. Glu-64 forms a further hydrogen bond with the backbone of Leu-160. Compared with the

High-potency AMPK inhibitor
compound C-bound structure, the DFG motif (Asp 157 -Phe-Gly 159 ) in the activation loop of our SBI-0206965-bound structure is displaced away from the ATP active site by a maximum of 2.9 Å (Phe-158 ␣C), presumably as a consequence of the nonplanar structure of SBI-0206965 relative to compound C. However, in common with the ␣2/compound C structure, this motif adopts a unique conformation that is neither "DFG-in," in which Asp-157 is appropriately positioned to chelate Mg 2ϩ ions required for coordination of ATP phosphate groups, nor classical "DFG-out," in which the motif flips by 180°such that the Phe side chain now occludes the ATP pocket and Asp-157 is no longer able to coordinate Mg-ATP for catalysis (Fig. S3B). Instead, the inhibited ␣2 DFG motif adopts an intermediate conformation; Asp-157 is sterically hindered from adopting the "DFG-in" conformation by the compound N-methylbenzamide and is oriented toward the N-lobe (Fig. S3B), with Phe-158 vacating the regulatory R-spine to displace His-137 in the kinase HRD motif (Fig. 4D). A consequence of this displacement is rearrangement of the HRD backbone and positioning of the Arg-138 side chain to prevent coordination with the phosphate group of pThr-172.

SBI-0206965 inhibits cellular AMPK signaling
We explored the efficacy of SBI-0206965 to inhibit AMPK signaling in a range of cell lines. In HEK293 cells, glucose starvation triggered a 1.9-fold increase in pThr-172, indicative of

High-potency AMPK inhibitor
energy stress, and this was accompanied by a 3.8-fold increase in pACC (Fig. 5A). The addition of SBI-0206965 to glucose-free medium significantly suppressed pACC at concentrations of Ն5 M, probably due to direct AMPK inhibition because SBI-0206965 did not induce reductions in pThr-172, reflecting suppressed LKB1 activity (Fig. 5A). SBI-0206965 concentrations of Ն5 M also blocked AMPK signaling in HEK293 cells treated with AICAR (Fig. 5B); however, this was probably due to inhibition of AICAR cellular uptake, reminiscent of compound C, because SBI-0206965 Ն 5 M also resulted in significantly reduced intracellular ZMP accumulation (Fig. S4A). SBI-0206965 Յ 30 M did not significantly affect ratios of adenine nucleotides in HEK293 cells, indicative of mitochondrial toxicity; however, we note a trend toward reduced adenylate energy charge at 30 M (Fig. S4B). 5 M SBI-0206965 also significantly suppressed pACC in the neuronal cell line SH-SY5Y, in which CaMKK2-mediated AMPK signaling was triggered by the addition of the calcium ionophore ionomycin (Fig. 5C).

Discussion
SBI-0206965 was originally identified as an ATP-competitive inhibitor of the autophagy initiator kinase ULK1, with the ability to suppress cellular ULK signaling and ULK1-mediated survival of lung cancer and glioblastoma cells when coupled with nutrient stress (34). In this initial report, AMPK signaling in MEFs (evidenced as ACC phosphorylation) was unchanged following incubation with 50 M SBI-0206965, despite increased Thr-172 phosphorylation, which may have arisen from fluctuations in adenine nucleotide ratios at high concentrations (Fig.  S4B). SBI-0206965 has subsequently been shown to suppress non-small-cell lung cancer cell growth (36). We now show that SBI-0206965 can be repurposed as a potent and kinase-selective inhibitor of the metabolic coordinator AMPK, providing a useful alternative to the almost singularly used, yet promiscuous, compound C. Because AMPK is a major positive regulator of autophagy through direct phosphorylation of ULK1 (37) and other autophagy-associated proteins (e.g. VSP34 and Beclin-1), simultaneous inhibition of both AMPK and ULK1 signaling supports the application of SBI-0206965 as a highly effective suppressor of prosurvival autophagic responses in tumor cells. We previously used SBI-0206965 as an agent to confirm ULK1-mediated phosphorylation of the drug-sensitizing AMPK ␤1-subunit residue Ser-108, until recently regarded as an autophosphorylation site (8). Our demonstration now that SBI-0206965 also inhibits AMPK does not detract from our previous conclusions, because Ser-108 autophosphorylation is a cis event (10), and in our study, kinase-inactive AMPK enzyme was used as the cellular substrate.
Our biochemical characterization of SBI-0206965 as an AMPK antagonist highlights several limitations of use that

High-potency AMPK inhibitor
demand consideration. Both Egan et al. (34) and our study here indicate a highly selective kinase profile for SBI-0206965; however, it must be noted that our combined, activity-based analyses cover only 57 kinases besides AMPK ␣1 and ␣2, representing just 11% of the human kinome. In light of our data implicating AMPK as being able to accommodate mixed-type inhibitors, we consider that activity, rather than ATP-competitive binding, is a more definitive metric of inhibitor selectivity. Besides ULK1 and -2, JAK3 and Src are flagged in both analyses as possible kinase targets for SBI-0206965, and we cannot exclude the possibility that others exist. We strongly advise against using SBI-0206965 in combination with AICAR when investigating cellular AMPK signaling. We also note that cellular incubations at high concentrations may induce Thr-172 phosphorylation through fluctuations in AMP/ATP and ADP/ ATP ratios, rather than via stimulation of LKB1-mediated phosphorylation as with SU6656, which has the potential to confound interpretation of results.
Structural characteristics used to classify small-molecule kinase inhibitors label SBI-0206965 as a type IIb AMPK inhibitor (38,39). Foremost, the SBI-0206965-binding pocket overlaps the ATP site in the ␣2 kinase domain, with the DFG motif positioned in a nonclassical "out" conformation. The Asp-157 side chain is not "in" (i.e. unable to coordinate phosphate-stabilizing Mg 2ϩ ions), and Asn-144/Phe-158 (7.7 Å) and Glu-64/ Phe-158 (10.1 Å) C␣ atomic distances are consistent with a "DFG-out" classification (40). Additionally, the SBI-0206965 molecule does not extend into the back cleft of the ␣2 kinase domain, and the R-spine is distorted upon binding. Type IIb inhibitors bind to sites that incompletely overlap the ATP active site and are usually regarded as ATP-competitive. However, members of this inhibitor class have also been described as "ATP noncompetitive" (41,42) and "indirectly ATP-competitive" (43), perhaps because comprehensive enzyme kinetic profiling is rarely reported alongside structural studies.
We have confirmed that SBI-0206965 partially shares its binding site on AMPK kinase domain with ATP but under our assay conditions is not ATP-competitive, instead displaying a mixed-type inhibition profile (Fig. 3D). Mixed-type inhibition, by definition, is a conceptual mixture of competitive (increased K m substrate) and uncompetitive inhibition (reduced V max ), in which binding of the inhibitor affects substrate binding, and vice versa (44). From a structural perspective, mixed-type inhibition would appear incompatible with overlapping inhibitor/ substrate binding sites. However, numerous studies have reported mixed-type inhibition by agents that have either been demonstrated to bind at the enzyme's active site or most likely do (i.e. transition state inhibitor analogues) (45)(46)(47)(48)(49)(50). Specifically, a type II, active site binding inhibitor of insulin-like growth factor 1 receptor tyrosine kinase activity has been reported to display mixed-type kinetics with respect to ATP (45), whereas the tyrphostin inhibitor AG1296 displayed either competitive or mixed-type characteristics, depending on the activation state of its target kinase in platelet-derived growth factor receptor (46). Additionally, crystal structures of adenylate cyclase complexed to P-site inhibitors (adenine nucleosides/adenine nucleoside 3Ј-phosphates) show active site binding; these inhibitors display either noncompetitive (a special from of mixed inhibition) or uncompetitive kinetic properties (47). Active site binding by mixed inhibitors has also been inferred by direct methods (51,52) or from close structural relationship to the substrate (53,54). Type II kinase inhibitors such as SBI-0206965 probably stabilize the "DFG-out" conformation, an arrangement less favorable for ATP binding (55). Thus, the binding sites occupied by SBI-0206965 or ATP represent distinct conformational arrangements, each reducing the binding affinity of the competing ligand.
SBI-0206965 increased K m(ATP) , indicating a preference for binding to the free enzyme rather than the enzyme-substrate complex. This is consistent with the location of the inhibitorbinding site resolved in our structure and our observation that SBI-0206965 cannot displace AMPK from ATP-agarose (Fig.  3C). This indicates that SBI-0206965 can only bind to an actively cycling AMPK, as the binding site must first become available following release of the ADP product from the active site. Furthermore, for the inhibitor-bound AMPK complexes, [ATP] in the activity assay may be below K s ATP, precluding an increase of IC 50 at the same rate as increase of [ATP], as dictated by the Cheng-Prusoff equation (56). This probably also explains why SBI-0206965 is very effective at inhibiting AMPK in cells; type II inhibitors typically display high cellular potency, whereas millimolar [ATP] often prevents type I inhibitors (e.g. compound C) from maintaining the potency observed in cellfree activity assays (43).
In summary, we show here that SBI-0206965 displays preferable characteristics, relative to compound C, as an AMPK inhibitor in vitro. Further studies are required to reveal its efficacy in vivo; however, biochemical and structural data provided in this study suggest that SBI-0206965 is a promising lead for the development of a new class of AMPK inhibitors with therapeutic potential.

Kinase inhibitor profiling
Inhibitor profiling (Express screen) was performed by the International Centre for Kinase Profiling, Medical Research Council Protein Phosphorylation and Ubiquitylation Unit (MRC-PPU), University of Dundee, UK.
Crystallization-pET21b plasmid for expression of ␣2 kinase domain (residues 6 -278) incorporating T172D mutation, with a PreScission protease cleavable N-terminal His 6 tag, was synthesized by General Biosystems. Constructs were sequenceverified, and expressed constructs were mass-verified by TOF MS.

Protein crystallization
His-␣2(6 -278) (T174D) was generated as described above for His-␣2(6 -279), except the CaMKK2 incubation step was omitted. His tag was removed with GST-tagged PreScission protease treatment (overnight, 4°C) after the buffer exchange step. PreScission protease was removed using GSH-Sepharose and ␣2(6 -279) (T174D) repurified by SEC. Concentrated ␣2(6 -279) (T174D) (4 mg/ml) was incubated with 0.5 mM SBI-0206965 on ice for 30 min and centrifuged at 10,000 rpm for 3 min before setting crystallization experiments. Protein was mixed equally 1:1 at room temperature with a reservoir solution containing 9 -14% ethanol, 5 mM MgCl 2 , 10 mM tris(2-carboxyethyl)phosphine, and 3-7 mM MnCl 2 . Diffraction quality crystals were obtained through streak seeding with a cat's whisker. Crystals appeared after 2-3 days and reached full size after 1-2 weeks. Crystals were then incubated in a cryoprotectant-containing reservoir solution with an addition of 25% ethylene glycol. Data were collected on both MX1 and MX2 beamlines at the Australian Synchrotron (Melbourne, Australia). Data were processed and integrated using XDS (57) and scaled using AIM-LESS from the CCP4 suite (58). The structure was solved by molecular replacement using Phaser from the CCP4 suite (59) and 3AQV as the search model. Iterative rounds of model building and refinement were performed using Coot (60) and Buster (https://www.globalphasing.com/buster/) 6 (61), respectively. Data for the bromine anomalous map was collected on the MX2 beamline at the Australian Synchrotron near the bromine edge at 13.6 KeV, calculated on the X-ray anomalous scattering website (http://skuld.bmsc.washington.edu/scatter/ AS_form.html). 6 An anomalous map was generated by Phenix.refine (62). SBI-0206965 molecular coordinates and restraints were generated using the PRODRG web server (63). Structural validation was performed using Molprobity (64). Omit maps were generated using Buster, and figures were created using PyMOL.

LKB1 phosphorylation assays
200 ng of unphosphorylated His-␣2␤1␥1 was incubated with 20 ng of LKB1/STRAD␣/MO25 in the presence of buffer A supplemented with 2 mM MgCl 2 , 200 M ATP, and compounds (final 1% DMSO) or 1% DMSO vehicle, as indicated, at 32°C. Reactions were terminated by the addition of SDS sample buffer and immunoblotted for pThr-172.

Immunoblotting
Samples were separated by SDS-PAGE on a 12% gel (7% for ACC), previously enriched using streptavidin-Sepharose (10), and transferred to Immobilon-FL polyvinylidene difluoride membrane (EMD Millipore). Membrane was blocked in PBS ϩ 0.1% Tween 20 (PBST) with 2% nonfat milk for 30 min at 22°C and then incubated for either 1 h or overnight with primary antibodies (dilutions in PBST). After washes with PBST, membranes were incubated with anti-rabbit or anti-mouse IgG secondary antibodies, fluorescently labeled with IR680 or IR800 dye, for 1 h. Immunoreactive bands were visualized on an Odyssey IR imaging system with densitometry analyses determined using ImageStudioLite software (LI-COR Biosciences).

ATP-agarose immobilization
20 g of Thr-172-phosphorylated His-␣2(6 -279) kinase domain was incubated with ATP/compounds for 30 min at room temperature, before the addition of 10 l of ATP-agarose suspended in 200 l of buffer B (50 mM HEPES, pH 7.4, 100 mM NaCl, 10% glycerol, 5 mM MgCl 2 ). Beads were incubated on a rotating wheel at 4°C for 2 h, before washing three times with 500 l of buffer A supplemented with 0.1% Tween 20. Beads were resuspended in 15 l of SDS sample buffer, boiled, and immunoblotted for AMPK ␣.

Cell culture and incubations
HEK293 and SH-SY5Y mammalian cells were cultured in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal bovine serum and 1% penicillin/streptomycin antibiotics, at 37°C with 5% CO 2 . Cells were incubated with fresh DMEM for 3 h, before simultaneous incubation with glucose-free DMEM and or/compounds as indicated. Glucose-starved cells were washed 4 h post-treatment with ice-cold PBS and harvested by rapid lysis using ice-cold lysis buffer (50 mM Tris⅐HCl, pH 7.4, 150 mM NaCl, 50 mM NaF, 1 mM sodium pyrophosphate, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 1% Triton X-100, and cOmplete protease inhibitor mixture). AICAR/ionomycintreated cells were harvested similarly, except this was performed 1 h post-treatment. Lysates were clarified by centrifugation (14,000 rpm; 3 min; 4°C) and flash-frozen in liquid N 2 until processed.

Nucleotide measurements
Adenine nucleotides and adenylate energy charge were measured by LC-MS from HEK293 perchlorate extracts, as described previously (32).

ZMP measurements
All ZMP measurements were acquired using LC-MS from perchlorate extracts, using a method similar to that described previously for nucleotides. Briefly, LC conditions were optimized for a 150-mm (length) and 0.5-mm (inner diameter) Hypercarb column (3 m; Thermo Fisher Scientific Australia Pty. Ltd.). The LC solvent system was (A) 25 mM triethylammonium bicarbonate buffer at pH 7.8 and (B) acetonitrile with 0.1% TFA. ZMP was eluted at a flow rate of 500 l/min in a gradient program consisting of 100% A (5 min), 0 -25% B (10 min), 50 -80% B (5 min), and 100% B (5 min). Data were analyzed with Multiquant 2.0.2, utilizing the area under the LC chromatogram for the corresponding ZMP peak. Calibration curves were obtained by linear regression of the peak area ratio of a ZMP standard. The MS conditions and MRM values for ZMP were optimized by separate infusion of 1 g/ml solution in 25 mM triethylammonium bicarbonate buffer at a flow rate of 50 l/min.

Statistical analysis
The data are presented as mean values Ϯ S.E. of at least three independent experiments unless stated. Statistical analyses were performed using GraphPad Prism version 7.0c. Ordinary one-way analysis of variance (ANOVA) with Dunnett's multiple-comparison test was used for all comparisons unless stated.