Simon A. Hawley,1 Rebecca J. Ford,2 Brennan K. Smith,2 Graeme J. Gowans,1

Sarah J. Mancini,3 Ryan D. Pitt,2 Emily A. Day,2 Ian P. Salt,3 Gregory R. Steinberg,2

and D. Grahame Hardie1

The Na+/Glucose CotransporterInhibitor Canagliflozin Activates AMPKby Inhibiting Mitochondrial Functionand Increasing Cellular AMP LevelsDiabetes 2016;65:2784–2794 | DOI: 10.2337/db16-0058

Canagliflozin, dapagliflozin, and empagliflozin, all recentlyapproved for treatment of type 2 diabetes, were derivedfrom the natural product phlorizin. They reduce hypergly-cemia by inhibiting glucose reuptake by sodium/glucosecotransporter (SGLT) 2 in the kidney, without affectingintestinal glucose uptake by SGLT1. We now report thatcanagliflozin also activates AMPK, an effect also seenwith phloretin (the aglycone breakdown product ofphlorizin), but not to any significant extent with dapagli-flozin, empagliflozin, or phlorizin. AMPK activation oc-curred at canagliflozin concentrations measured inhuman plasma in clinical trials and was caused byinhibition of Complex I of the respiratory chain, leadingto increases in cellular AMP or ADP. Although canagli-flozin also inhibited cellular glucose uptake indepen-dently of SGLT2, this did not account for AMPKactivation. Canagliflozin also inhibited lipid synthesis, aneffect that was absent in AMPK knockout cells and thatrequired phosphorylation of acetyl-CoA carboxylase(ACC) 1 and/or ACC2 at the AMPK sites. Oral adminis-tration of canagliflozin activated AMPK in mouse liver,although not in muscle, adipose tissue, or spleen.Because phosphorylation of ACC by AMPK is known tolower liver lipid content, these data suggest a potentialadditional benefit of canagliflozin therapy compared withother SGLT2 inhibitors.

A recently introduced approach to treatment of type 2diabetes is selective inhibition of sodium/glucose co-transporter (SGLT) 2 (1). SGLT1 and SGLT2 are related

transporters that carry glucose across apical membranesof polarized epithelial cells against concentration gradi-ents, driven by Na+ gradients. SGLT1 is expressed in thesmall intestine and responsible for most glucose uptakeacross the brush border membrane of enterocytes, whereasSGLT2 is expressed in the kidney and responsible for mostglucose readsorption in the convoluted proximal tubules.The first identified SGLT inhibitor was a natural product,phlorizin, which is broken down in the small intestine tophloretin, the aglycone form (Fig. 1). Although phlorizinhad beneficial effects in hyperglycemic animals (2), it in-hibits SGLT1 and SGLT2, causing adverse gastrointestinaleffects (3). This led to development of the synthetic ana-logs canagliflozin (4), dapagliflozin (5), and empagliflozin(6) (Fig. 1), which have 260-, 1,100-, and 2,700-fold se-lectivity for SGLT2 over SGLT1, respectively (6). In meta-analyses of clinical trials in type 2 diabetes, canagliflozin(7), dapagliflozin (8), or empagliflozin (9), as monother-apy or combined with existing therapies, all reduced fast-ing plasma glucose, HbA1c, and body weight. Canagliflozinalso decreased plasma triglycerides (7).

The current front-line therapy for type 2 diabetes ismetformin, a biguanide that lowers plasma glucose pri-marily by reducing hepatic glucose production (10). Met-formin, and the related biguanide phenformin, inhibitComplex I of the respiratory chain (11,12) and activatethe cellular energy sensor AMPK (13,14). Binding to theAMPK-g subunit of AMP and/or ADP, which are elevatedduring cellular energy stress, causes conformational changes

1Division of Cell Signalling and Immunology, School of Life Sciences, University ofDundee, Dundee, Scotland, U.K.2Division of Endocrinology and Metabolism, Department of Medicine, McMasterUniversity, Hamilton, Ontario, Canada3Institute of Cardiovascular and Medical Sciences, College of Medical, Veterinaryand Life Sciences, University of Glasgow, Glasgow, Scotland, U.K.

Corresponding authors: D. Grahame Hardie, d.g.hardie@dundee.ac.uk, andGregory R. Steinberg, gsteinberg@mcmaster.ca.

Received 12 January 2016 and accepted 25 May 2016.

This article contains Supplementary Data online at http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db16-0058/-/DC1.

S.A.H. and R.J.F. contributed equally to this study.

© 2016 by the American Diabetes Association. Readers may use this article aslong as the work is properly cited, the use is educational and not for profit, andthe work is not altered. More information is available at http://diabetesjournals.org/site/license.

2784 Diabetes Volume 65, September 2016

PHARMACOLOGYAND

THERAPEUTIC

S

that activate the kinase via allosteric effects and promotionof net phosphorylation of Thr172 on the AMPK-a subunit(15–18). Metformin and phenformin increase ADP-to-ATPratios and fail to activate AMPK containing a g-subunitmutant that does not bind AMP/ADP (19), confirmingthat their AMPK-activating effects are mediated by in-creases in AMP/ADP. Once activated, AMPK acts to restoreenergy homeostasis by promoting catabolic pathways, in-cluding fatty acid oxidation, while inhibiting anabolic path-ways, including fatty acid synthesis (15,16). Its opposingacute effects on fat synthesis and oxidation are due tophosphorylation of two acetyl-CoA carboxylase (ACC) iso-forms, ACC1 and ACC2. Whether AMPK explains all ther-apeutic benefits of metformin has been controversialbecause its acute effects on hepatic glucose production inmice were reported to be AMPK independent (20,21). How-ever, studies using knock-in mice, in which both ACC iso-forms were replaced by mutants lacking the critical AMPKphosphorylation sites, suggested that the longer-terminsulin-sensitizing effects of metformin are accounted forby phosphorylation and inactivation of ACC1/ACC2 byAMPK (22).

We now report that canagliflozin activates AMPK, inintact cells and in vivo, by a mechanism involving in-hibition of respiratory chain Complex I. Our results raisethe possibility that some therapeutic benefits of canagli-flozin might occur via AMPK activation rather than SGLT2inhibition.

RESEARCH DESIGN AND METHODS

Materials and AntibodiesCanagliflozin, dapagliflozin, and empagliflozin were fromSelleck Chemicals, and phlorizin, phloretin, metformin,phenformin, AICAR, and 2,4-dinitrophenol (DNP) werefrom Sigma-Aldrich. A769662 was synthesized as described(23). Antibodies against phosphorylated (p)Thr172 onAMPK-a (pT172, #2531) were from Cell Signaling Tech-nology. In Fig. 7, antibodies against pACC (#3661) andtotal ACC (#3676) were from Cell Signaling Technology.In other figures, total ACC was detected using streptavidindirectly conjugated to 800 nm fluorophore (Rockland Im-munochemicals), and pACC (14) and total AMPK-a (24)antibodies were as previously described. Anti-GLUT1

(#325510) was from Abcam, and anti-SGLT2 (sc-47402)was from Santa Cruz Biotechnology.

Cell Culture and LysisHEK-293 cells and wild-type (WT) and AMPK knockoutmouse embryo fibroblasts (MEFs) (25) were grown in DMEMwith 25 mmol/L glucose and 10% FBS. Cell lysates wereprepared as described previously (19). For Western blotsshown in Fig. 6C, tissues were homogenized in 5 vols ofHES buffer (20 mmol/L Na HEPES [pH 7.4], 1 mmol/LEDTA, 250 mmol/L sucrose; Roche complete protease in-hibitor cocktail) with a Dounce homogenizer and centri-fuged (7,050g, 20 min, 4°C). The pellet was resuspendedin HES buffer and layered on top of buffer containing1.12 mol/L sucrose before centrifugation in a swing-outrotor (41,500g, 60 min, 4°C). Membranes were collectedfrom the interface of the sucrose layers, diluted in HESbuffer, and centrifuged (150,000g, 60 min, 4°C). The re-sultant plasma membrane–rich pellets were resuspendedin HES buffer (0.2–0.4 mL).

Immunoprecipitate Kinase Assays and Other AnalysesMethods for AMPK assay in immunoprecipitates, SDS-PAGE, Western blotting, and determination of cellularADP-to-ATP ratios and oxygen consumption in HEK-293cells were described previously (19). Lipid synthesis inMEFs was analyzed by starving cells of serum for 3 hand then treating with drug or vehicle in the presenceof [14C]acetate (1 mCi/mL)/0.4 mmol/L Na acetate for3 h. Cells were washed with PBS before extraction to de-termine incorporation of label into total lipid (26). Fattyacid oxidation was measured as etomoxir-sensitive 3H2Oproduction from [3H]palmitate. MEFs were preincubatedwith AMPK activators for 30 min before incubation in[3H]palmitic acid (8 mCi/mL, 110 mmol/L), carnitine(50 mmol/L), fatty acid–free BSA (0.5 mg/mL) in Earle’s-HEPES (116 mmol/L NaCl, 5.3 mmol/L KCl, 0.8 mmol/LMgSO4, 1.8 mmol/L CaCl2, 1 mmol/L NaH2PO4, 20 mmol/LHEPES-NaOH, pH 7.4) in the presence or absence of eto-moxir (50 mmol/L) for 90 min at 37°C. The 3H2O gener-ated was separated and quantified as previously described(27).

Animal ExperimentsAll animal procedures were approved by the McMasterUniversity Animal Ethics Research Board. Male and femalemice (16–20 weeks), WT or ACC1/ACC2 (S79A/S212A)double knock-in (DKI), were housed in a pathogen-free fa-cility under a 12-h light/dark cycle at 23°C, with ad libitumaccess to standard chow and water. Primary hepatocyteswere generated from WT and DKI mice and the followingday were treated for 4 h with canagliflozin or vehicle beforeassessing AMPK and ACC phosphorylation and lipid syn-thesis, as previously described (22,28). For experiments toexamine AMPK and ACC phosphorylation in vivo, canagli-flozin or vehicle (saline solution containing 0.5% carboxy-methyl cellulose, 0.025% Tween-20) was administered byoral gavage (100 mg/kg, 10 mL/g). Mice were anesthetized

Figure 1—Structures of compounds used in this study.

diabetes.diabetesjournals.org Hawley and Associates 2785

and tissues snap frozen in situ as previously described (22).Measurements of the respiratory exchange ratio (RER)were performed in metabolic cages using a protocol (29)in which mice were fasted overnight and refed with chowfor 2 h before being gavaged with canagliflozin or vehicle atthe time of food withdrawal. In separate experiments, micewere treated as above, blood was collected from a nick in thetail, and glucose was measured using a Roche glucometer.

Measurements of Oxygen Uptake/Respirationin Primary Mouse HepatocytesMitochondrial respiration was measured by high-resolutionrespirometry (Oxygraph-2 k; Oroboros Instruments, Inns-bruck, Austria) at 37°C and room air-saturated O2 tension inrespiration buffer (MIRO5) containing EGTA (0.5 mmol/L),MgCl2 (3 mmol/L), K-lactobionate (60 mmol/L), KH2PO4

(10 mmol/L), HEPES (20 mmol/L), sucrose (110 mmol/L),and fatty acid–free BSA (1 g/L). Primary hepatocytes fromWT mice were generated as described above. The follow-ing day they were suspended in 2 mL of respiration buffer,and 800 mL of the suspension was added to the respira-tion chambers. Digitonin (8.1 mmol/L) was added to per-meabilize the cells, and the assay was initiated 5 min later.To measure Complex I–supported respiration, glutamate(5 mmol/L), malate (2 mmol/L), and ADP (2.5 mmol/L)were added, and respiration was allowed to reach steadystate. To measure Complex II–supported respiration, rote-none (1.25 mmol/L), succinate (10 mmol/L), and ADP(2.5 mmol/L) were added, and respiration was allowed toreach steady state. Drugs were then added to respiring cellsat the indicated concentrations.

Measurements of Glucose UptakeCells were incubated in glucose-free Krebs-Ringer phosphatefor 2 h, with AICAR present for the last 1 h or canagliflozin/dapagliflozin for the last 15 min. 2-Deoxyglucose (2DG; 50mmol/L, 1mCi/mL) was added, and cytochalasin B-sensitive2DG uptake was measured over 10 min (30).

Statistical AnalysisSignificance of differences was assessed using the Studentt test, one-way or two-way ANOVA as appropriate, usingthe Sidak test after ANOVA. Values of P , 0.05 wereconsidered statistically significant.

RESULTS

Canagliflozin and Phloretin, but not Dapagliflozin,Empagliflozin, or Phlorizin, Significantly ActivateAMPK in HEK-293 CellsTo assess the effect of canagliflozin on AMPK, we initiallyused HEK-293 cells. A daily 300-mg dose of canagliflozinproduces a peak plasma concentration in humans of ;10mmol/L (31), so we tested concentrations from 1 to30 mmol/L. After incubation for 1 h (Fig. 2A), canagliflo-zin increased AMPK activity at concentrations from 1 to30 mmol/L, with 10 mmol/L giving a similar activation tothat observed with 300 mmol/L A769662, which activatesAMPK by direct binding between the a- and b-subunits(32), or berberine, a mitochondrial inhibitor that activates

AMPK by increasing the cellular AMP-to-ATP ratio (19).AMPK activation by canagliflozin, A769662, and berber-ine was associated with increased phosphorylation of Thr172on AMPK and of the primary AMPK site on ACC (pACC)(Fig. 2A; quantified results shown in Supplementary Fig.1A and B). Figure 2B shows that the effect of 30 mmol/Lcanagliflozin was rapid, reaching a maximum by 20 min.

Figure 2C and D compares the effects of canagliflozin,dapagliflozin, and empagliflozin. Although dapagliflozinand empagliflozin both activated AMPK, this requiredconcentrations .30 mmol/L, and even at 100 mmol/L,their effects were small compared with canagliflozin. Ef-fects on phosphorylation of AMPK and ACC (Fig. 2F–H,quantified results in Supplementary Fig. 1C–F) were con-sistent with this. Because single doses of dapagliflozin(20 mg) or empagliflozin (50 mg) produce peak plasmaconcentrations of only 1–2 mmol/L (33,34), that theseinhibitors would produce significant AMPK activation atnormal therapeutic doses seems unlikely.

We also examined the effects of the natural productphlorizin and its aglycone form, phloretin. Interestingly,phloretin activated AMPK and promoted phosphorylationof AMPK and ACC at concentrations slightly higher thancanagliflozin. However, phlorizin only affected these parame-ters marginally at much higher concentrations (Fig. 2E, I, andJ; quantification of blots in Supplementary Fig. 1G and H).

Canagliflozin Activates AMPK by Inhibiting Complex Iof the Respiratory ChainTo test whether canagliflozin activated AMPK by increas-ing cellular AMP or ADP, we tested its effects in HEK-293cells expressing the WT AMPK-g2 subunit (WT cells) or theAMP/ADP-insensitive R531G mutant (RG cells) (19). Can-agliflozin activated AMPK and promoted its phosphoryla-tion in WT cells but not in RG cells; similar results wereobtained with phloretin (Fig. 3; quantification of blotsin Supplementary Fig. 2). These results suggest thatcanagliflozin and phloretin activate AMPK by increasingcellular AMP or ADP.

Cellular AMP levels are low and difficult to measure incultured cells, so we routinely measure ADP-to-ATP ratiosas a surrogate for AMP-to-ATP ratios (17). Increasing con-centrations of canagliflozin caused increases in the ADP-to-ATP ratio that were significant at 10 and 30 mmol/L, with30 mmol/L canagliflozin producing an effect similar to10 mmol/L phenformin (Fig. 4A). Similar results wereobtained with phloretin (Fig. 4B). The increase in the cel-lular ADP-to-ATP ratio due to canagliflozin was accompa-nied by a reduction of cellular oxygen consumption (Fig.4C). The effect of 30 mmol/L canagliflozin was smaller,although more rapid, than that of 10 mmol/L phenformin.When the uncoupler 2,4-DNP was added 60 min later,there was a large increase in oxygen consumption (repre-senting the maximal respiration rate when mitochondriawere not limited by the supply of ADP) in cells treated withDMSO. However, as canagliflozin concentrations increased,the effects of 2,4-DNP were eliminated, suggesting that,

2786 Canagliflozin Activates AMPK Diabetes Volume 65, September 2016

like phenformin, canagliflozin inhibited the respiratory chainrather than the F1 ATP synthase.

We next used primary mouse hepatocytes permeabilizedwith digitonin and measured the function of respiratorychain Complexes I and II. Figure 4D and E shows that can-agliflozin and dapagliflozin inhibited Complex I–supported

respiration in a concentration-dependent manner, whereasany effect on Complex II was marginal. The inhibitory ef-fect of canagliflozin on Complex I was significantly greaterthan that of dapagliflozin, with estimated half-maximaleffects at 18 6 1 mmol/L for canagliflozin and 40 63 mmol/L for dapagliflozin.

Figure 2—AMPK activation in HEK-293 cells by canagliflozin and other SGLT2 inhibitors. A: HEK-293 cells were treated for 1 h with theindicated concentrations of canagliflozin, A769662, or berberine as positive controls. The upper panel shows kinase assays of immuno-precipitates prepared using anti–pan a antibodies (mean 6 SD, n = 2). The lower panel shows analysis of AMPK and ACC phosphorylationby Western blotting from duplicate dishes of cells. B: Time course of AMPK activation by 30 mmol/L canagliflozin (mean 6 SD, n = 2). C:Comparison of effects of canagliflozin and dapagliflozin at 1-h incubations (mean 6 SEM, n = 4). *P < 0.05 and ****P < 0.0001 indicatesignificance of differences from DMSO controls indicated. D: Comparison of effects of canagliflozin and empagliflozin, as in C. ****P <0.0001. E: Comparison of effects of phloretin and phlorizin, as in C. ****P< 0.0001. Phosphorylation of AMPK and ACC in duplicate wells inresponse to canagliflozin (F ) and dapagliflozin (G), from experiment in C; empagliflozin, from experiment in D (H); and phloretin (I) andphlorizin (J), from experiment in E. See Supplementary Fig. 1 for quantification of these blots.

diabetes.diabetesjournals.org Hawley and Associates 2787

AMPK-Dependent and -Independent Effectsof Canagliflozin on Cellular MetabolismTo examine effects of canagliflozin on metabolic effectsmediated by AMPK, we initially used immortalized MEFs

with a double knockout (DKO) of AMPK-a1 and -a2 orWT controls. Figure 5A shows that AMPK was activatedby concentrations of canagliflozin above 1 mmol/L in WTcells, associated with phosphorylation of AMPK and ACC(blots quantified in Supplementary Fig. 3A and B). Asexpected, AMPK activity was undetectable in DKO cells,and although ACC expression was normal, there was nobasal or canagliflozin-stimulated ACC phosphorylation.We next measured lipid synthesis using [14C]acetate andfatty acid oxidation using [3H]palmitate (Fig. 5B and C).A769662 and phenformin, which activate AMPK by dif-ferent mechanisms, caused inhibition of lipid synthesis inthe WT cells, whereas canagliflozin caused increasing in-hibition at 10 and 30 mmol/L. As expected, the inhibitionby A769662 was abolished in DKO cells, but surprisingly,lipid synthesis was stimulated by both phenformin andcanagliflozin in DKO cells. Fatty acid oxidation (Fig. 5C)was stimulated by AICAR, and this was abolished in theDKO cells, showing that it was AMPK-dependent. How-ever, phenformin and canagliflozin inhibited fatty acidoxidation in the WT and DKO cells. The likely explanationfor this is that these drugs, although activating AMPK,inhibit the respiratory chain and, therefore, prevent reox-idation of NADH and FADH2 generated by fat oxidation,an effect that would be AMPK-independent and that hasbeen previously observed in primary hepatocytes treatedwith metformin (22).

Canagliflozin Inhibits Lipogenesis in HepatocytesThrough a Mechanism Involving Phosphorylationof ACC by AMPKThe inhibition of lipid synthesis by canagliflozin in Fig. 5Bwas associated with phosphorylation of ACC at the AMPKsite(s) (Fig. 5A; results quantified in Supplementary Fig.3B). To test whether the effect of canagliflozin on lipidsynthesis required phosphorylation of ACC1/ACC2, westudied hepatocytes from mice with a DKI of mutationsat the AMPK sites on both ACC isoforms (22). Figure 5Dshows that increasing concentrations of canagliflozincaused progressive inhibition of lipid synthesis in WThepatocytes that was abolished, at least at concentrationsof 10 mmol/L and below, in DKI cells. The effects of500 mmol/L metformin were also abolished in the DKI cells.However, at 30 mmol/L canagliflozin, there was still someinhibition in the DKI cells, although significantly less thanin WT cells. The effects of canagliflozin and metformin onlipid synthesis were accompanied by increased phosphory-lation of AMPK and, in WT cells, of ACC. The pACC andtotal ACC antibodies both detected doublets; the lowerand upper bands are ACC1 (predicted mass 265 kDa) andACC2 (276 kDa) (22). Figure 5E shows that 10–30 mmol/Lcanagliflozin significantly increased pT172/AMPK-a andpACC/ACC (ACC1 + ACC2) in WT mouse hepatocytes.

Canagliflozin Inhibits Glucose Uptake in HEK-293 Cellsand MEFsWe next assessed the effects of the drugs on glucosetransport by measuring 2DG uptake. Canagliflozin inhibited

Figure 3—Comparison of AMPK activation by canagliflozin andphloretin in WT and RG cells. A: WT and RG HEK-293 cells weretreated for 1 h with the indicated concentrations of canagliflozin,and the results of kinase assays in anti-a immunoprecipitates areshown (mean 6 SEM, n = 4 to 6). ****P < 0.0001 indicates signif-icance of differences from vehicle control. B: Analysis of AMPKphosphorylation in duplicate dishes of cells from the same experi-ment as in A. C: Experiment as in A, but using phloretin. ****P <0.0001. D: As in B, but using phloretin. See Supplementary Fig. 2 forquantification of these blots.

2788 Canagliflozin Activates AMPK Diabetes Volume 65, September 2016

uptake by 50–60% in HEK-293 cells and MEFs, whereasthe AMPK activator AICAR had no effect. The results wereidentical in WT MEFs and AMPK DKO MEFs, showingthat this effect of canagliflozin was AMPK-independent.We also assessed the expression of SGLT2 in these celltypes using Western blotting. HEK-293 cells and mouseliver, but not MEFs, expressed a polypeptide that comi-grated with SGLT2 in mouse kidney. HEK-293 cells,MEFs, and mouse liver also appeared to express GLUT1,

although the band did not always comigrate with thebands in control tissue, possibly due to variable glycosyl-ation (Fig. 6C). Attempts to measure expression of otherglucose transporters (GLUT3, SGLT1) by Western blottingwere inconclusive. The effect of canagliflozin to inhibitglucose uptake in HEK-293 cells and MEFs appears to beyet another off-target effect, because dapagliflozin had noeffect on 2DG uptake in either cell type (Fig. 6A and B). Tofurther confirm that the activation of AMPK in HEK-293

Figure 4—Effects of canagliflozin, phloretin, and phenformin on cellular ADP-to-ATP ratios and oxygen uptake. A: Effects of phenforminand increasing concentrations of canagliflozin on cellular ADP-to-ATP ratios. HEK-293 cells were treated with the indicated drug concen-tration for 1 h, and cellular ADP-to-ATP ratios were determined by capillary electrophoresis of perchloric acid extracts (mean6 SEM, n = 4).***P < 0.001 and ****P < 0.0001 indicate significance of differences from vehicle control. B: As in A, but using phloretin in place ofcanagliflozin. C: Effects of phenformin and increasing concentrations of canagliflozin on cellular oxygen uptake. At the point shown by thefirst arrow (“drug”), vehicle (DMSO), phenformin (10 mmol/L), or the indicated concentration of canagliflozin (mmol/L) was added, andoxygen uptake was measured at regular intervals. At the point shown by the second arrow (“DNP”), 2,4-DNP was added, and oxygenuptake was measured for a further 15 min. Results are expressed as percentage of the mean of the first three time points before drugaddition (mean 6 SD; three replicate dishes performed twice, n = 6). After addition of drugs, the effects of 10 mmol/L canagliflozin weresignificantly different from the DMSO control by two-way ANOVA (Dunnett multiple comparison test) at all time points except 67.5 and75 min; effects of 30 mmol/L canagliflozin and 10 mmol/L phenformin were significant at all time points. D and E: Complex I– and ComplexII–supported respiration in permeabilized primary mouse hepatocytes. The cells were permeabilized with digitonin and then provided withADP, glutamate, and malate to stimulate Complex I–supported respiration, or rotenone, succinate, and ADP to simulate Complex II–supportedrespiration. Results are expressed as percent inhibition of Complex I and II by canagliflozin and dapagliflozin (mean 6 SEM, n = 3–5). ****P <0.0001 indicates statistically significant differences between effects of canagliflozin and dapagliflozin by two-way ANOVA.

diabetes.diabetesjournals.org Hawley and Associates 2789

cells was not secondary to its effects on glucose uptake, wecompared the effects of canagliflozin with complete glucoseremoval from the medium (Fig. 6D; quantification of blotsin Supplementary Fig. 3C and D). Although removal of all

glucose caused a modest activation and Thr172 phosphor-ylation of AMPK, activation by 30 mmol/L canagliflozinwas more than threefold larger. Because inhibition of glu-cose uptake by canagliflozin was only partial (Fig. 6A), the

Figure 5—Effects of canagliflozin on AMPK, lipid synthesis, and fatty acid oxidation in intact cells. A: WT and AMPK-a12/2 and AMPK-a22/2

MEFs (AMPK-a1/a2 DKO) were incubated with the indicated concentrations of canagliflozin for 1 h, and AMPK activity was determined (upperpanel) in anti–AMPK-a immunoprecipitates (mean 6 SEM, n = 4). Assays were performed in the DKO cells, although the activity, as expected,was negligible. *P< 0.05 and ****P< 0.0001 indicate significant differences from the vehicle control. The lower panel shows phosphorylation ofAMPK and ACC analyzed in duplicate dishes of cells from the same experiment (see Supplementary Fig. 3 for quantification of these blots). B:Lipid synthesis (incorporation of radioactivity from [14C]acetate into total lipid, i.e., fatty acids and sterols) in WT and AMPK-a1/a2 DKO MEFsincubated in the indicated concentrations of A769662, phenformin, or canagliflozin for 1 h (mean6 SEM, n = 4). **P< 0.01 and ****P< 0.0001indicate significant difference from vehicle controls. C: Fatty acid oxidation (incorporation of radioactivity from [3H]palmitate into 3H2O) in WTand AMPK-a1/a2 DKO MEFs incubated in the indicated concentrations of phenformin, AICAR, or canagliflozin for 1 h. Results are mean 6SEM, n = 4. ****P < 0.0001 indicates significant difference from vehicle controls. D: Lipid synthesis and protein phosphorylation in primarymouse hepatocytes from WT and ACC1/ACC2 DKI mice. The upper panel shows incorporation of radioactivity from [3H]acetate into total lipidmeasured after 4 h (mean 6 SEM, cells from three mice, each performed in triplicate). ****P < 0.0001 indicates significant differences fromcontrol within each genotype. †P< 0.05, ††P< 0.01, and ††††P< 0.0001 indicate significant difference between genotypes at the same drugconcentration. The lower panel shows analysis of protein phosphorylation by Western blotting of a representative example. E: Quantification ofthe increases in phosphorylation in WT hepatocytes of Thr172 on AMPK (left) or Ser79 plus Ser212 on ACC1/ACC2 (right), normalized for theexpression of total AMPK or ACC1/ACC2 and expressed relative to control (mean6 SEM; n = 4 for pT172/AMPK-a, and n = 5 for pACC/ACC).*P < 0.05, **P < 0.01, and ****P < 0.0001 indicate significant differences from vehicle controls.

2790 Canagliflozin Activates AMPK Diabetes Volume 65, September 2016

effect of canagliflozin on AMPK is unlikely to be due to areduced supply of glucose for catabolism.

Canagliflozin Activates AMPK in Mice In VivoTo test whether canagliflozin activated AMPK in vivo, itwas administered to mice (100 mg/kg) by oral gavage, andtissues were collected 3 h later by freeze clamping in situ,which preserves the activation state of AMPK (35). Inliver, Thr172 phosphorylation of AMPK was significantlyincreased by this treatment, as was the phosphorylation ofACC and Raptor at AMPK sites (Fig. 7A and SupplementaryFig. 4). By contrast, significant increases in phosphoryla-tion of AMPK, ACC, and Raptor were not observed inmuscle (tibialis interior), gonadal white adipose tissue, orspleen (Supplementary Fig. 5A–C). We also measured theeffects on the RER of oral administration of canagliflozin atthe time of withdrawing food from previously fasted micethat had been refed for 2 h. Canagliflozin caused a morerapid drop in RER than vehicle in WT mice, indicating amore rapid shift back toward fat rather than carbohydrateoxidation (Fig. 7B). However, this was also observed inACC1/ACC2 DKI mice, showing that the effect was indepen-dent of ACC phosphorylation and, therefore, presumablyof AMPK (Fig. 7C). This reduction in RER by canagliflozin

was likely secondary to the reduction of blood glucose, whichwas similar in WT and ACC DKI mice (Fig. 7D).

DISCUSSION

Recent clinical trials suggest that the SGLT2 inhibitorscanagliflozin, dapagliflozin, and empagliflozin show promisefor reversal of hyperglycemia, either as monotherapy or asadjuncts to existing therapy. Compared with dapagliflozin,canagliflozin also has consistently favorable effects onplasma lipid profiles (7,8,36,37). These differential effectson plasma lipids prompted us to investigate whether cana-gliflozin might have SGLT2-independent effects. Our resultsshow that canagliflozin causes a substantial activation ofAMPK in human and mouse cells at concentrations corre-sponding to the peak plasma concentrations achieved aftertherapeutic doses in humans. By contrast, dapagliflozin andempagliflozin only caused a modest AMPK activation atconcentrations well above their peak plasma concentrations.Thus, activation of AMPK by dapagliflozin or empagliflozinis less likely to be significant in vivo.

Our results demonstrate that AMPK activation is pri-marily due to inhibition of Complex I of the respiratorychain, leading to increases in cellular AMP/ADP that bindto the g-subunit and promote Thr172 phosphorylation.

Figure 6—Canagliflozin inhibits glucose transport in HEK-293 cells and MEFs, but this does not account for AMPK activation. A: Uptake of2DG in HEK-293 cells treated with 10 mmol/L canagliflozin, 10 mmol/L dapagliflozin, or 1 mmol/L AICAR. Results are mean 6 SEM (n = 5).**P < 0.01 indicates significant differences from vehicle controls. B: As in A, but in WT or AMPK-a12/2 and AMPK-a22/2 (AMPK-a1/a2DKO) MEFs. C: Expression of SGLT2 and GLUT1; membrane fractions from MEFs or HEK-293 cells (80 mg protein), mouse kidney (KID) orliver (LIV) (30 mg), mouse brain (Br.) (15 mg), or human placenta (HUM PLA) (7.5 mg) were analyzed by Western blotting and probed using theindicated antibodies. D: Effect of total removal of medium glucose (center) or 30 mmol/L canagliflozin (right) on AMPK activity (top) andphosphorylation of AMPK (pT172) and pACC. Results in the bar chart are mean 6 SEM (n = 3). *P < 0.05, **P < 0.01, and ****P < 0.0001indicate statistical significance of differences.

diabetes.diabetesjournals.org Hawley and Associates 2791

Thus, canagliflozin:

1. increased cellular ADP-to-ATP ratios;2. increased AMPK activation and Thr172 phosphoryla-

tion in cells expressing the WT but not the AMP/ADP-insensitive R531G mutant of AMPK-g2;

3. inhibited oxygen uptake in HEK-293 cells; and4. inhibited oxygen uptake in permeabilized mouse hepato-

cytes provided with substrates that feed into Complex I.

Dapagliflozin also caused a less potent effect on ComplexI, although only at concentrations (10–30 mmol/L) higherthan those observed in human plasma with normal doses.We also found that canagliflozin, but not dapagliflozin,

inhibited 2DG uptake in HEK-293 cells and MEFs in anAMPK-independent manner, indicating that it had addi-tional off-target effects on glucose transport, presumablydue to inhibition of another glucose transporter such asGLUT1. Indeed, previous studies in L6 myotubes have in-dicated that 10 mmol/L canagliflozin can inhibit glucoseuptake by ;50%, an effect that was attributed to GLUT1inhibition (38). However, this is unlikely to account for theAMPK activation observed in our experiments, because evencomplete removal of glucose from the medium had only amodest effect on AMPK activity compared with canagliflozin.

Interestingly, we found that phloretin, the aglyconederivative of phlorizin, also activated AMPK, althoughphlorizin itself was much less effective. Like canagliflozin,

Figure 7—Liver AMPK is activated by oral administration of canagliflozin in mice, but this does not explain changes in RER. A: Phos-phorylation of AMPK, ACC, and Raptor relative to total protein in mouse liver after oral administration of canagliflozin (100 mg/kg) to mice(mean 6 SEM, n = 9–11). *P < 0.05 indicates significant difference by Student t test. B: RER (VCO2/VO2) in 12-h fasted mice that had beenrefed a carbohydrate-rich diet for 2 h between 7:00 and 9:00 A.M. and were then administered vehicle or canagliflozin (30 mg/kg) at the timethat food was withdrawn again. Results are mean 6 SEM (n = 10–12). C: As in B, but experiments performed in ACC1/ACC2 DKI mice.Results are mean 6 SEM (n = 5–7). D: Glucose concentrations in the tail vein during experiments of the type shown in B and C. *P < 0.05and **P < 0.01 indicate statistically significant differences in canagliflozin samples vs. vehicle control. hr, h.

2792 Canagliflozin Activates AMPK Diabetes Volume 65, September 2016

phloretin appeared to act by increasing cellular AMP.Phlorizin (39) and phloretin (40) were both previouslyreported to inhibit the function of isolated mitochondria,but we found that only phloretin is effective in intact cells,perhaps due to greater membrane permeability.

Our studies also show that AMPK activation has theexpected effects on lipid synthesis in intact cells. Thus,three distinct AMPK activators—A769662, phenformin,and canagliflozin—all inhibited the pathway in MEFs, andthese inhibitory effects were abolished in MEFs lackingAMPK, correlating with a complete loss of ACC phosphor-ylation. Surprisingly, phenformin and canagliflozin (butnot A769662) stimulated lipid synthesis in DKO cells.When we measured fatty acid oxidation in the same cells,phenformin and canagliflozin inhibited the pathway in anAMPK-independent manner, which is expected becauseboth compounds inhibit Complex I. The major fates ofcellular acetyl-CoA are oxidation by the tricarboxylicacid cycle or incorporation into lipids. Thus, in the ab-sence of AMPK to inhibit lipid synthesis, acetate fluxmight be diverted into lipid synthesis as oxidation wasinhibited. This provides an explanation for the paradoxi-cal activation of lipid synthesis by phenformin and cana-gliflozin observed in AMPK knockout cells.

Our results in isolated mouse hepatocytes demonstratethat the effects of canagliflozin on lipid synthesis aremediated by phosphorylation of ACC. Thus, a moderateand therapeutically relevant concentration of canagliflo-zin (10 mmol/L) inhibited lipid synthesis in WT hepato-cytes but failed to do so in DKI cells from mice where bothACC isoforms lacked the critical AMPK site. At highercanagliflozin (30 mmol/L), there was some inhibitioneven in DKI hepatocytes, although significantly lessthan in WT cells; this is most likely because both ACCisoforms use ATP as a direct substrate. Once the increasein the cellular ADP-to-ATP ratio becomes substantial, aswith 30 mmol/L canagliflozin (Fig. 4A), decreases in ATPmay limit flux through ACC independently of AMPK. Thisis a revealing demonstration of the physiological role ofAMPK: as ATP falls under situations of energetic stress,the kinase limits the function of energy-consuming path-ways before the ATP concentration has dropped to levelswhere it becomes limiting for the pathway itself. Thisoccurs because AMPK is more sensitive to ATP depletionthan the ATP-consuming enzymes in the pathway.

Our results show that oral administration of canagli-flozin increased Thr172 phosphorylation of AMPK in liverin vivo and phosphorylation of ACC and Raptor, two of itswell-recognized downstream targets. Although we onlytested effects of the drug in normal mice, there is noreason to believe the results would be any different indiabetic models; for example, berberine, which activatesAMPK by the same mechanism as canagliflozin (19), ac-tivates AMPK normally in db/db mice (41). Canagliflozinalso caused a more rapid drop in the RER when admin-istered to fed mice, indicating a more rapid switch to fatversus carbohydrate oxidation. However, this was still

observed in the DKI mice and was, therefore, presumablyindependent of AMPK. It is possible that a reduction inblood glucose caused by canagliflozin causes increasedfat oxidation due to competition between glucose andfat for substrate oxidation (42) and that this obscuresany effect due to ACC phosphorylation by AMPK. Thismight be addressed in future studies using SGLT2 knock-out mice.

Our findings raise the interesting question of whetherdual therapy with canagliflozin and metformin would bemore effective than canagliflozin alone. In a recently re-ported clinical trial of newly diagnosed subjects with type 2diabetes (43), monotherapy with canagliflozin was moreeffective in lowering HbA1c than metformin. Althoughdual therapy was more efficacious than canagliflozin alone,the effects of the two drugs were not additive, as might beexpected if they had distinct mechanisms of action. Bycontrast, in a similar trial using metformin and dapagliflo-zin, monotherapy with dapagliflozin was not more effectivethan metformin, and the effects of dual therapy were closerto being additive (44).

Finally, although the long-term effects of metformin toinhibit hepatic glucose production in mice are AMPK-dependent (22), studies have suggested that its rapid ef-fects are AMPK-independent (20,21). However, the latterauthors agree that the primary effect of metformin is toinhibit the respiratory chain and increase cellular AMP inthe liver, which is then proposed to affect other AMP-sensitive targets such as fructose-1,6-bisphosphatase (20)or adenylate cyclase (21). Because canagliflozin inhibits therespiratory chain and activates AMPK in the liver, it is con-ceivable that these AMP-dependent but AMPK-independenteffects of metformin might also be mimicked by canagliflozin.

Acknowledgments. The authors thank Benoit Viollet (Institut Cochin,Paris) for AMPK knockout MEFs.Funding. S.J.M. and I.P.S. were supported by a Project Grant (PG/13/82/30483)from the British Heart Foundation. The study was supported by the CanadianDiabetes Association (G.R.S.), by the Canadian Institutes of Health Research (G.R.S.),by a Senior Investigator Award from the Wellcome Trust (097726) to D.G.H., and bythe pharmaceutical companies supporting the Division of Signal TransductionTherapy at Dundee (AstraZeneca, Boehringer-Ingelheim, GlaxoSmithKline, MerckKGaA, Janssen Pharmaceuticals, and Pfizer). G.R.S. is a Canada Research Chair inMetabolism and Obesity and the J. Bruce Duncan Chair in Metabolic Diseases.Duality of Interest. No potential conflicts of interest relevant to this articlewere reported.Author Contributions. S.A.H., R.J.F., B.K.S., G.J.G., S.J.M., R.D.P., andE.A.D. established experimental protocols and/or conducted experiments. I.P.S.,G.R.S., and D.G.H. supervised research. G.R.S. and D.G.H. wrote the initialmanuscript. All authors contributed revisions and corrections. G.R.S. and D.G.H.are the guarantors of this work and, as such, had full access to all the data in thestudy and take responsibility for the integrity of the data and the accuracy of thedata analysis.

References1. Rosenwasser RF, Sultan S, Sutton D, Choksi R, Epstein BJ. SGLT-2 in-hibitors and their potential in the treatment of diabetes. Diabetes Metab SyndrObes 2013;6:453–467

diabetes.diabetesjournals.org Hawley and Associates 2793

2. Rossetti L, Smith D, Shulman GI, Papachristou D, DeFronzo RA. Correctionof hyperglycemia with phlorizin normalizes tissue sensitivity to insulin in diabeticrats. J Clin Invest 1987;79:1510–15153. Ehrenkranz JR, Lewis NG, Kahn CR, Roth J. Phlorizin: a review. DiabetesMetab Res Rev 2005;21:31–384. Nomura S, Sakamaki S, Hongu M, et al. Discovery of canagliflozin, a novelC-glucoside with thiophene ring, as sodium-dependent glucose cotransporter2 inhibitor for the treatment of type 2 diabetes mellitus. J Med Chem 2010;53:6355–63605. Meng W, Ellsworth BA, Nirschl AA, et al. Discovery of dapagliflozin: a potent,selective renal sodium-dependent glucose cotransporter 2 (SGLT2) inhibitor forthe treatment of type 2 diabetes. J Med Chem 2008;51:1145–11496. Grempler R, Thomas L, Eckhardt M, et al. Empagliflozin, a novel selectivesodium glucose cotransporter-2 (SGLT-2) inhibitor: characterisation and com-parison with other SGLT-2 inhibitors. Diabetes Obes Metab 2012;14:83–907. Yang XP, Lai D, Zhong XY, Shen HP, Huang YL. Efficacy and safety ofcanagliflozin in subjects with type 2 diabetes: systematic review and meta-analysis. Eur J Clin Pharmacol 2014;70:1149–11588. Sun YN, Zhou Y, Chen X, Che WS, Leung SW. The efficacy of dapagliflozincombined with hypoglycaemic drugs in treating type 2 diabetes mellitus: meta-analysis of randomised controlled trials. BMJ Open 2014;4:e0046199. Liakos A, Karagiannis T, Athanasiadou E, et al. Efficacy and safety ofempagliflozin for type 2 diabetes: a systematic review and meta-analysis.Diabetes Obes Metab 2014;16:984–99310. Natali A, Ferrannini E. Effects of metformin and thiazolidinediones onsuppression of hepatic glucose production and stimulation of glucose uptake intype 2 diabetes: a systematic review. Diabetologia 2006;49:434–44111. Owen MR, Doran E, Halestrap AP. Evidence that metformin exerts its anti-diabetic effects through inhibition of complex 1 of the mitochondrial respiratorychain. Biochem J 2000;348:607–61412. El-Mir MY, Nogueira V, Fontaine E, Avéret N, Rigoulet M, Leverve X.Dimethylbiguanide inhibits cell respiration via an indirect effect targeted onthe respiratory chain complex I. J Biol Chem 2000;275:223–22813. Zhou G, Myers R, Li Y, et al. Role of AMP-activated protein kinase inmechanism of metformin action. J Clin Invest 2001;108:1167–117414. Hawley SA, Boudeau J, Reid JL, et al. Complexes between the LKB1 tumorsuppressor, STRAD alpha/beta and MO25 alpha/beta are upstream kinases in theAMP-activated protein kinase cascade. J Biol 2003;2:2815. Hardie DG. AMPK–sensing energy while talking to other signaling pathways.Cell Metab 2014;20:939–95216. Hardie DG, Schaffer B, Brunet A. AMP-activated protein kinase: an energy-sensing pathway with multiple inputs and outputs. Trends Cell Biol 2016;26:190–20117. Gowans GJ, Hawley SA, Ross FA, Hardie DG. AMP is a true physiologicalregulator of AMP-activated protein kinase by both allosteric activation and en-hancing net phosphorylation. Cell Metab 2013;18:556–56618. Ross FA, Jensen TE, Hardie DG. Differential regulation by AMP and ADP of AMPKcomplexes containing different g subunit isoforms. Biochem J 2016;473:189–19919. Hawley SA, Ross FA, Chevtzoff C, et al. Use of cells expressing gammasubunit variants to identify diverse mechanisms of AMPK activation. Cell Metab2010;11:554–56520. Foretz M, Hébrard S, Leclerc J, et al. Metformin inhibits hepatic gluco-neogenesis in mice independently of the LKB1/AMPK pathway via a decrease inhepatic energy state. J Clin Invest 2010;120:2355–236921. Miller RA, Chu Q, Xie J, Foretz M, Viollet B, Birnbaum MJ. Biguanidessuppress hepatic glucagon signalling by decreasing production of cyclic AMP.Nature 2013;494:256–26022. Fullerton MD, Galic S, Marcinko K, et al. Single phosphorylation sites inAcc1 and Acc2 regulate lipid homeostasis and the insulin-sensitizing effects ofmetformin. Nat Med 2013;19:1649–165423. Göransson O, McBride A, Hawley SA, et al. Mechanism of action ofA-769662, a valuable tool for activation of AMP-activated protein kinase. J BiolChem 2007;282:32549–32560

24. Woods A, Salt I, Scott J, Hardie DG, Carling D. The alpha1 and alpha2isoforms of the AMP-activated protein kinase have similar activities in rat liver butexhibit differences in substrate specificity in vitro. FEBS Lett 1996;397:347–35125. Laderoute KR, Amin K, Calaoagan JM, et al. 59-AMP-activated protein ki-nase (AMPK) is induced by low-oxygen and glucose deprivation conditions foundin solid-tumor microenvironments. Mol Cell Biol 2006;26:5336–534726. Corton JM, Gillespie JG, Hardie DG. Role of the AMP-activated proteinkinase in the cellular stress response. Curr Biol 1994;4:315–32427. Reihill JA, Ewart MA, Salt IP. The role of AMP-activated protein kinase in thefunctional effects of vascular endothelial growth factor-A and -B in human aorticendothelial cells. Vasc Cell 2011;3:928. Davies SP, Carling D, Hardie DG. Tissue distribution of the AMP-activatedprotein kinase, and lack of activation by cyclic-AMP-dependent protein kinase,studied using a specific and sensitive peptide assay. Eur J Biochem 1989;186:123–12829. Hawley SA, Fullerton MD, Ross FA, et al. The ancient drug salicylate directlyactivates AMP-activated protein kinase. Science 2012;336:918–92230. Salt IP, Connell JM, Gould GW. 5-aminoimidazole-4-carboxamide ribonu-cleoside (AICAR) inhibits insulin-stimulated glucose transport in 3T3-L1 adipo-cytes. Diabetes 2000;49:1649–165631. Devineni D, Curtin CR, Polidori D, et al. Pharmacokinetics and pharmaco-dynamics of canagliflozin, a sodium glucose co-transporter 2 inhibitor, in sub-jects with type 2 diabetes mellitus. J Clin Pharmacol 2013;53:601–61032. Xiao B, Sanders MJ, Carmena D, et al. Structural basis of AMPK regulationby small molecule activators. Nat Commun 2013;4:301733. Kasichayanula S, Chang M, Hasegawa M, et al. Pharmacokinetics andpharmacodynamics of dapagliflozin, a novel selective inhibitor of sodium-glucoseco-transporter type 2, in Japanese subjects without and with type 2 diabetesmellitus. Diabetes Obes Metab 2011;13:357–36534. Brand T, Macha S, Mattheus M, Pinnetti S, Woerle HJ. Pharmacokinetics ofempagliflozin, a sodium glucose cotransporter-2 (SGLT-2) inhibitor, coadministeredwith sitagliptin in healthy volunteers. Adv Ther 2012;29:889–89935. Davies SP, Carling D, Munday MR, Hardie DG. Diurnal rhythm of phos-phorylation of rat liver acetyl-CoA carboxylase by the AMP-activated protein ki-nase, demonstrated using freeze-clamping. Effects of high fat diets. Eur JBiochem 1992;203:615–62336. Bailey CJ, Iqbal N, T’joen C, List JF. Dapagliflozin monotherapy in drug-naïve patients with diabetes: a randomized-controlled trial of low-dose range.Diabetes Obes Metab 2012;14:951–95937. Kaku K, Kiyosue A, Inoue S, et al. Efficacy and safety of dapagliflozinmonotherapy in Japanese patients with type 2 diabetes inadequately controlledby diet and exercise. Diabetes Obes Metab 2014;16:1102–111038. Liang Y, Arakawa K, Ueta K, et al. Effect of canagliflozin on renal thresholdfor glucose, glycemia, and body weight in normal and diabetic animal models.PLoS One 2012;7:e3055539. Keller DM, Lotspeich WD. Some effects of phlorizin on the metabolism ofmitochondria. J Biol Chem 1959;234:987–99040. De Jonge PC, Wieringa T, Van Putten JP, Krans HM, Van Dam K. Phloretin -an uncoupler and an inhibitor of mitochondrial oxidative phosphorylation. BiochimBiophys Acta 1983;722:219–22541. Kim WS, Lee YS, Cha SH, et al. Berberine improves lipid dysregulation inobesity by controlling central and peripheral AMPK activity. Am J Physiol En-docrinol Metab 2009;296:E812–E81942. Randle PJ, Garland PB, Hales CN, Newsholme EA. The glucose fatty-acidcycle. Its role in insulin sensitivity and the metabolic disturbances of diabetesmellitus. Lancet 1963;1:785–78943. Rosenstock J, Chuck L, González-Ortiz M, et al. Initial combination therapywith canagliflozin plus metformin versus each component as monotherapy fordrug-naïve type 2 diabetes. Diabetes Care 2016;39:353–36244. Henry RR, Murray AV, Marmolejo MH, Hennicken D, Ptaszynska A, List JF.Dapagliflozin, metformin XR, or both: initial pharmacotherapy for type 2 diabetes,a randomised controlled trial. Int J Clin Pract 2012;66:446–456

2794 Canagliflozin Activates AMPK Diabetes Volume 65, September 2016