RESEARCH ARTICLE

AMP-activated protein kinase promotes epithelial-mesenchymaltransition in cancer cells through Twist1 upregulationMeera Saxena*,§, Sai A. Balaji§, Neha Deshpande§, Santhalakshmi Ranganathan, Divya Mohan Pillai, SravanthKumar Hindupur‡ and Annapoorni Rangarajan¶

ABSTRACTThe developmental programme of epithelial-mesenchymal transition(EMT), involving loss of epithelial and acquisition of mesenchymalproperties, plays an important role in the invasion-metastasiscascade of cancer cells. In the present study, we show thatactivation of AMP-activated protein kinase (AMPK) using A769662led to a concomitant induction of EMT in multiple cancer cell types, asobserved by enhanced expression of mesenchymal markers,decrease in epithelial markers, and increase in migration andinvasion. In contrast, inhibition or depletion of AMPK led to areversal of EMT. Importantly, AMPK activity was found to benecessary for the induction of EMT by physiological cues such ashypoxia and TGFβ treatment. Furthermore, AMPK activationincreased the expression and nuclear localization of Twist1, anEMT transcription factor. Depletion of Twist1 impaired AMPK-induced EMT phenotypes, suggesting that AMPK might mediate itseffects on EMT, at least in part, through Twist1 upregulation. Inhibitionor depletion of AMPK also attenuated metastasis. Thus, our dataunderscore a central role for AMPK in the induction of EMT and inmetastasis, suggesting that strategies targeting AMPK might providenovel approaches to curb cancer spread.

KEY WORDS: AMPK signaling, EMT, Twist

INTRODUCTIONMetastasis, the spread of tumor cells from a primary site to a distantorgan, is accountable for more than 90% of cancer-related mortality(Chaffer and Weinberg, 2011). This process involves local invasionof cells from a primary tumor into the surrounding tissue,intravasation into the nearby microvasculature, migration throughthe blood and lymphatic vessels, extravasation at a distant organ site,and their proliferation to form a secondary macroscopic tumor(Chaffer and Weinberg, 2011). It is believed that the evolutionarilyconserved developmental program of epithelial-mesenchymaltransition (EMT) bestows cancer cells with properties that enablethem to initiate the invasion-metastasis cascade (Yang et al., 2006).

Indeed, EMT is considered to be an important contributor in theinvasion and metastasis of several types of cancer (Thiery et al.,2009).

During EMT, epithelial cells shed their intercellular contacts,cell-extracellular matrix interactions and apico-basal polarity, andattain more mesenchymal properties such as increased expression ofmesenchymal markers, motility and invasive capabilities (Polyakand Weinberg, 2009). EMT can be stimulated by a plethora of cuessuch as hypoxia, signaling through TGFβ, receptor tyrosine kinases,Notch and Wnt (Polyak and Weinberg, 2009). In response to thesecues, several transcription factors such as Twist1, Snail (also knownas SNAI1), Slug (also known as SNAI2), FOXC2, Zeb1 and Zeb2(Polyak and Weinberg, 2009) are upregulated, which serve asimportant downstream mediators of EMT. Of these, Twist1 isconsidered as a master regulator of EMT (Yang et al., 2004). Severalmechanisms of regulation of EMT have been elucidated in recentyears (Polyak and Weinberg, 2009). However, a discreet hierarchyor a central orchestrator linking these various mechanisms in theEMT interactome has yet to be elaborated.

AMP-activated protein kinase (AMPK) is an evolutionarilyconserved stress-sensing kinase that becomes activated under avariety of stresses, including hypoxia, ischemia and nutrientdeprivation, leading to a change in the ATP:AMP ratio (Steinbergand Kemp, 2009). It is a heterotrimeric serine/threonine proteinkinase consisting of a catalytic α subunit (α1 or α2), a scaffolding βsubunit (β1 or β2) and a nucleotide-binding γ subunit (γ1, γ2 or γ3)(Hardie et al., 2012). Besides its pertinent role in the regulation ofenergy homeostasis, AMPK has been implicated in severalphysiological processes including cell growth (Xiang et al.,2004), cell division (Jones et al., 2005), maintenance of epithelialcell polarity (Zhang et al., 2006) and autophagy (Liang et al., 2007).Apart from these, AMPK activity has also been associated with cellmigration in normal physiology (Nakano et al., 2010). For instance,AMPK activation was shown to be important for migration ofhuman umbilical vein endothelial cells (HUVECs) (Nagata et al.,2003) and transendothelial lymphocyte migration (Martinelli et al.,2009). In another study, inhibition of AMPKwas shown to suppressTGFβ-mediated apoptosis and EMT of normal murine hepatocytes(Wang et al., 2010). A few reports have also associated AMPKsignaling with cancer cell migration and invasion (Chen et al., 2011;Chiu et al., 2009; Kim et al., 2011). Another recent study showedthat indirect activation of AMPK by mitochondrial dysfunctioninduced EMT in lung cancer cells (He et al., 2016). In contrast,some studies have reported an inhibitory role for AMPK inEMT of breast, prostate and lung cancer cells, mainly using5-aminoimidazole-4-carboxamide ribonucleotide (AICAR) andmetformin as pharmacological activators of AMPK (Banerjeeet al., 2016; Chou et al., 2014; Cufí et al., 2010; Han et al., 2015; Linet al., 2015; Qu et al., 2014). Thus, there are conflicting reportsabout the role of AMPK signaling in EMT and cancer metastasis,Received 11 July 2017; Accepted 20 June 2018

Department of Molecular Reproduction, Development and Genetics, IndianInstitute of Science, Bangalore 560012, Karnataka, India.*Present address: Department of Biomedicine, University of Basel, Mattenstrasse28, 4058 Basel, Switzerland. ‡Present address: Biozentrum, University of Basel,Klingelbergstrasse 50/70, 4056 Basel, Switzerland.§These authors contributed equally to this work

¶Author for correspondence (anu@iisc.ac.in)

S.A.B., 0000-0002-6539-9660; N.D., 0000-0002-3740-2324; S.R., 0000-0001-5531-2173; S.K.H., 0000-0002-8078-4006; A.R., 0000-0003-0972-2184

This is an Open Access article distributed under the terms of the Creative Commons AttributionLicense (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use,distribution and reproduction in any medium provided that the original work is properly attributed.

1

© 2018. Published by The Company of Biologists Ltd | Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

suggesting that AMPK activation might have cell-type- and context-specific effects. This might, in turn, depend on the upstreamactivators of AMPK used, some of which could additionally haveAMPK-independent effects. This incongruity needs to be resolvedbefore AMPK-targeted therapeutics are widely used for cancertreatment.In light of these conflicting reports, we investigated the role of

AMPK in regulating EMT in multiple cancer types. In the presentstudy, we show that AMPK activation with a direct, allostericactivator A769662 suffices to induce a functional EMT in multiplecancer cells, whereas its inhibition or depletion leads to reversal ofEMT. Importantly, we show that AMPK is required for EMTinduction by upstream stimuli relevant to cancer progression, suchas hypoxia and TGFβ. Furthermore, we show that AMPK mediatesits effects, at least in part, through upregulation of Twist1, a majorregulator of EMT, and increasing its nuclear localization.

RESULTSTo explore a possible role for AMPK in EMT, we initiated our studyby investigating the effects of AMPK activation and inhibition onthe expression of markers associated with EMT in a variety ofcancer cell lines including breast cancer cell lines MCF7, T47D,BT474 and MDA-MB-231, melanoma cell line MDA-MB-435Sand lung adenocarcinoma cell line A549. First, we determined theendogenous expression of epithelial marker E-cadherin (E-cad) andmesenchymal markers vimentin (Vim) and N-cadherin (N-cad) inthese cell lines (Fig. S1A). The cancer cell lines BT474, MCF7 andT47D exhibiting an E-cad+ and Vim− phenotype representepithelial cell types. A549 exhibited an E-cadlow and Vimlow

phenotype, reminiscent of a partially epithelial-mesenchymal-transitioned cell line, whereas MDA-MB-231 and MDA-MB-435S cells exhibited an E-cad− and Vimhigh phenotype, representingmesenchymal cell types. The pattern of expression of these markersmatched that reported in the literature (Lombaerts et al., 2006).To begin our study, we investigated whether AMPK activation

would promote the induction of EMT in epithelial-type cancer celllines. Most previous studies have used AICAR or metformin foractivating AMPK; however, these have additional AMPK-independent effects as well (Foretz et al., 2010; Kalender et al.,2010). To overcome this problem, we used A769662 (Cool et al.,2006), which directly activates the heterotrimeric form of AMPKboth allosterically and by inhibiting its dephosphorylation (Sanderset al., 2007). We confirmed treatment with A769662 by performingimmunoblot analyses to detect the levels of phosphorylated acetyl-CoA carboxylase (ACC), a bona fide substrate of AMPK (Witterset al., 1991). An increase in the phosphorylation of ACC isindicative of AMPK activation (Fig. S1B). Treatment of T47D andMCF7 cells with A769662 led to an increase in the transcript levelsof mesenchymal markers N-cad and Vim and EMT transcriptionfactors Snai1, Slug and Zeb1 (Fig. 1A). Moreover, since decrease inthe protein levels of epithelial markers and disruption of adherensjunctions are hallmarks of EMT (Huang et al., 2012), we alsoinvestigated the effects of AMPK activation on the protein levels ofthe epithelial markers epithelial cadherin (E-cad; also known asCDH1) and zona occludens protein 1 (ZO-1, also known as TJP1).Immunocytochemical analysis revealed a significant reduction in E-cad levels upon AMPK activation for 48 h in these cell lines(Fig. 1B). Reduced expression of E-cad and ZO-1 was also detectedusing an immunoblotting approach (Fig. S1B). Similar results wereobtained with AMPK activation in BT474 cells (Fig. S1C,D,E).However, in spite of the changes at transcript levels (Fig. 1A andFig. S1C), AMPK activation for 48 h did not lead to the detection of

the mesenchymal markers Vim and N-cad at protein levels in theseepithelial cell types (data not shown). Thus, this set of data revealedthat AMPK activation in the epithelial cancer cells suffices to bringabout a partial EMT involving the downmodulation of epithelialmarkers.

Next, we investigated the effects of AMPK activation andinhibition on cancer cells that had already undergone EMT tovarious extents, such as A549, MDA-MB-231 andMDA-MB-435Scells. Activation of AMPK in the partially epithelial-mesenchymal-transitioned A549 cells, bearing an E-cadlow and Vimlow phenotype,led to an increase in the mRNA levels of mesenchymal markers andEMT transcription factors (Fig. 1C). Immunofluorescence analysisalso revealed a reduction in E-cad and increase in Vim proteinexpression (Fig. 1D). Reduction in the expression of the epithelialmarkers E-cad and ZO-1, and increase in the expression of themesenchymal markers Vim and N-cad, was also detected usingimmunoblotting (Fig. S1F). Thus, AMPK activation furtherpromoted the EMT phenotype of A549 cells. In contrast,inhibition of AMPK using 10 μM Compound C, a potent,reversible, and ATP-competitive inhibitor of AMPK (Zhou et al.,2001), led to a reduction in Vim and N-cad expression at bothmRNA and protein levels, and an increase in E-cad and ZO-1protein levels (Fig. 1C,D and Fig. S1F).

In MDA-MB-231 cells bearing an E-cad− and Vimhigh

phenotype, AMPK activation led to a further increase in themRNA expression of mesenchymal markers Vim and N-cad andEMT transcription factors Snai1, Slug and Zeb1, whereas AMPKinhibition led to a reduction in their transcript levels (Fig. 1E). Atprotein levels, we observed a significant increase in Vim expressionupon AMPK activation and a moderate decrease upon AMPKinhibition (Fig. 1F and Fig. S1G). As reported previously (Sommerset al., 1991), we failed to detect the expression of E-cad in MDA-MB-231 cells (Fig. S1A), and AMPK inhibition also did not lead toits re-expression in these cells (data not shown). However, AMPKactivation led to a decrease in the expression of ZO-1, whereas itsinhibition led to increased ZO-1 in MDA-MB-231 cells (Fig. S1G).Similar results were obtained with another mesenchymal cell line,MDA-MB-435S (Fig. S1H).

Because pharmacological inhibition of AMPK (with CompoundC) might have off-target effects (Liu et al., 2014), we additionallyundertook RNAi-mediated depletion of AMPK. To do so, we usedBT474 and A549 cells stably expressing AMPKα2 shRNA underconstitutive and inducible promoters, respectively, each showing∼50% and ∼70% knockdown of AMPK (Fig. S1I and J). Depletionof AMPKα2 in the epithelial BT474 cells led to a further increase inthe expression of epithelial markers E-cad and ZO-1 (Fig. S1I),although we failed to see any expression of Vim (data not shown). Inthe partially epithelial-mesenchymal-transitioned A549 cells,AMPK knockdown led to an increase in the expression of E-cadand ZO-1 while reducing the levels of Vim (Fig. S1J; clone #1).Immunofluorescence analysis further revealed an increase in E-cadexpression and a decrease in Vim expression upon AMPK depletionin these cells (Fig. S1K). Similar results were obtained in A549 cellswhere AMPKwas stably knocked down using another shRNA oligosequence (Fig. S1L; clone #4). Similarly, mesenchymal MDA-MB-231 cells that stably express AMPKα2 shRNA (Hindupur et al.,2014) and showed ∼80% knockdown (Fig. S1M) revealed areduction in the levels of Vim and N-cad (Fig. S1M,N). Theseresults were validated further using an additional, inducible shRNAtargeting AMPKα2 (Fig. S1O; clone #4) in MDA-MB-231 cells.Similar results were obtained upon AMPK knockdown in yetanother mesenchymal cell type, MDA-MB-435S (Fig. S1P). Thus,

2

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

Fig. 1. AMPKactivitymodulatesEMTmarkerexpression. (A-F)Cancercell linesT47D,MCF7(A,B),A549(C,D)andMDA-MB-231(E,F)werecultured in thepresenceof 100 μM AMPK activator (A769662), 10 μM AMPK inhibitor (Compound C) or DMSO (vehicle control) as specified for 48 h. Thereafter, cells were harvested andsubjected toqRT-PCRanalysis for thespecified transcripts.Graphs inpanelsA,CandErepresent foldchange ingeneexpressionnormalized to β2M;errorbars represents.e.m., n=3. Parallel dishes were fixed using 4% paraformaldehyde to undertake immunocytochemistry analysis for the specified proteins (B,D,F). Photomicrographsshow representative fluorescent images taken using anOlympus 1X71microscope at 20×magnification and processed using ImageJ software. E-cad is stained in red;Vim isstained ingreen; the nucleus isstained inblue (Hoechst 33342).Graphs inpanelsB,DandF represent relative fluorescence intensitymeasurementsnormalizedto number of nuclei. Error bars represent s.e.m., n=3. Scale bars: 20 μm. *P<0.05; **P<0.01; ***P<0.001; ns, not significant. A7, A769662; CC, Compound C.

3

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

these data revealed that in cancer cells that have undergone EMT tovarying extent, AMPK activation further promoted, whereas itsinactivation or depletion reversed the morphological changesassociated with EMT, suggesting that AMPK plays a key role inregulating the process of EMT.Attainment of motility and invasive capabilities are hallmark

features of a functional EMT (Polyak and Weinberg, 2009). Toassess whether AMPK can modulate the motility and migration ofcancer cells, we undertook a scratch assay with BT474, A549,MDA-MB-231 and MDA-MB-435S cells treated with eitherAMPK activator A769662 or its inhibitor Compound C. Inkeeping with a failure to induce mesenchymal marker expressionin epithelial cell lines, AMPK activation failed to alter migration ofBT474 cells (Fig. 2A and Fig. S2A). In the partially epithelial-mesenchymal-transitioned A549 cells, A769662 treatment led to amarginal increase in migration whereas Compound C treatmentresulted in a significant decrease in the motility (Fig. 2B andFig. S2B). Interestingly, mesenchymal MDA-MB-231 and MDA-MB-435S cells showed a significant increase in migration whentreated with A769662 and a significant inhibition of motility withCompound C treatment (Fig. 2C, Fig. S2C, Fig. 2D and Fig. S2D).To further confirm these data, we additionally undertook yet anothercell migration assay with MDA-MB-435S cells involvingquantitative real-time impedance measurement using an ECIS(electric cell-substrate impedance sensing)-based technique (Honget al., 2011a). In corroboration with the scratch assay results, wefound that although A769662 treatment led to a marked increase inmigration of MDA-MB-435S cells (as revealed by an increase inimpedance), Compound C treatment led to a decrease in migration(as revealed by a decrease in impedance) compared with the controlcells (Fig. S2E). Further, to confirm the results with Compound C,we employed RNAi-mediated knockdown of AMPKα2. Consistentwith pharmacological inhibition of AMPK, A549, MDA-MB-231and MDA-MB-435S cells stably expressing AMPKα2 shRNA ortransfected with siRNA targeting AMPKα2 also showed decreasedmigration compared with scrambled shRNA control cells (Fig. 2E,Fig. S2F, Fig. 2F, Fig. S2G, Fig. 2G and Fig. S2H). Knockdown ofAMPK failed to change the migration in BT474 cells (data notshown). The effect of AMPK inhibition/knockdown on themigratory properties of mesenchymal cells (Fig. 2F, Fig. S2G,Fig. 2G and Fig. S2H) was more dramatic compared with thepartially epithelial-mesenchymal-transitioned A549 cells (Fig. 2Eand Fig. S2F).To determine the effects of AMPK activation and inhibition on

the invasive potential of cancer cells, we performed in vitro Boydenchamber invasion assays. Since mesenchymal cells showedmaximal effect on AMPK modulation, we undertook invasionassays with these cell types. Compared with vehicle-treated controlcells, we found that whereas A769662 treatment caused asignificant increase in the invasiveness of MDA-MB-231 andMDA-MB-435S cells, Compound C treatment decreased theirinvasive potential (Fig. 2H,I). Furthermore, knockdown ofAMPKα2 in MDA-MB-231 and MDA-MB-435S cells also led toa marked decrease in their invasive potential compared with controlcells (Fig. 2J, Fig. S2I and Fig. 2K). These results collectivelydemonstrated that whereas AMPK activation increased themigratory and invasive capability of cancer cells, thereby bringingabout a functional EMT, its inhibition or knockdown reversed EMT,suggesting that AMPK plays a critical role in the morphogeneticprocesses involved with EMT.Next, we investigated the role of AMPK in the context of

physiological stimuli such as hypoxia, a known trigger of EMT

(Thiery et al., 2009; Yang et al., 2008). Intriguingly, hypoxia is alsoa well-known trigger for AMPK activation (Mungai et al., 2011).Based on our data above, we hypothesized that hypoxia-inducedEMT might be mediated by AMPK activation. To test this, weexposed epithelial BT474 cells stably expressing either scrambledor AMPKα2 shRNA to the hypoxia mimetic CoCl2. Increase in thelevels of Hif1α on treatment with CoCl2 revealed the generation of ahypoxic environment (Fig. 3A). Consistent with a previous study(Mungai et al., 2011), treatment with CoCl2 also triggered anincrease in AMPK activity, as revealed by elevated pACC levels(Fig. 3A). Compared with untreated control cells, a decrease in theexpression of E-cad and ZO-1 on CoCl2 treatment revealed theinduction of EMT in scrambled-shRNA-expressing BT474 cells.Interestingly, hypoxia-induced EMT was inhibited upon AMPKknockdown in these cells (Fig. 3A). Immunoblot analysis revealedsimilar results for partial epithelial-mesenchymal-transitioned A549cells stably expressing AMPKα2 shRNA and incubated underhypoxic conditions (3% oxygen) in a tri-gas incubator (Fig. 3B;clone #4). Similarly, in mesenchymal MDA-MB-435S cells, anincrease in the expression of Vim and N-cad in 3% oxygen revealedthe induction of EMT under hypoxia, which was inhibited in thepresence of the AMPK inhibitor Compound C (Fig. 3C). Similarresults were obtained upon treatment with CoCl2 in MDA-MB-231cells stably expressing AMPKα2 shRNA (Fig. S3A) or in thepresence of Compound C (Fig. S3B). These data reveal that AMPKmight be required for hypoxia-induced EMT.

Besides hypoxia, another potent inducer of EMT is TGFβ(Polyak and Weinberg, 2009; Yang and Weinberg, 2008). It haspreviously been shown that treatment of murine hepatocytes andAML12 cells with TGFβ leads to activation of AMPK signaling(Wang et al., 2010). Thus, we speculated that AMPK activity mightalso be required for TGFβ-induced EMT of cancer cells. To explorethis, MDA-MB-435S cells were treated with TGFβ for 48 h. Anincrease in the protein levels of pSmad2 (Fig. 3D) confirmed theactivation of TGFβ signaling in these cells. In keeping with the dataof Wang et al., we also saw an increase in AMPK activity uponTGFβ activation, as revealed by an increase in the levels of pACC(Fig. 3D). Furthermore, increased levels of Vim and N-cad revealedthe induction of EMT (Fig. 3D). Intriguingly, TGFβ-induced EMTwas significantly inhibited in the presence of the AMPK inhibitorCompound C (Fig. 3D). Similar results were obtained upon TGFβtreatment in the presence of Compound C in A549 (Fig. S3C) andMDA-MB-231 cells (Fig. S3D). To further corroborate the dataobtained with pharmacological inhibition of AMPK, we addressedthe role of AMPK in TGFβ-induced EMT by subjecting MDA-MB-231 cells stably expressing either scrambled or AMPKα2 shRNA toTGFβ treatment for 48 h. An increase in the levels of Vim and N-cadrevealed the induction of EMT marker expression. Depletion ofAMPK prevented TGFβ-mediated induction of EMT markers atprotein levels (Fig. S3E). Together, these data reveal that AMPKmight be required for TGFβ-mediated induction of EMT.

Ras signaling has also been shown to promote EMT (Mukherjee,2010) and increase AMPK activity (Rios et al., 2013). To testwhether AMPK is also important for Ras-induced EMT, weinhibited AMPK in HMLER breast cancer cells (Elenbaas et al.,2001) using Compound C. A reduction in the levels of pAMPK andpACC revealed the effects of Compound C (Fig. S3F). We observedthat inhibition of AMPK led to a decrease in the expression ofmesenchymal markers such as Vim and N-cad while increasing theexpression of the epithelial marker E-cad in HMLER cells(Fig. S3F). Thus, collectively these results indicated a critical rolefor AMPK in the induction of EMT by various upstream stimuli.

4

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

Because our study highlighted the role of AMPK in EMTinduction, and EMT is considered to be a prerequisite formetastasis (Polyak and Weinberg, 2009), we next investigatedthe requirement of AMPK function for metastasis in vivo. Todo so, we subjected MDA-MB-231 cells stably expressing eitherscrambled- or AMPKα2-shRNA to the experimental tail veinmetastasis assay. While several microscopic metastatic colonies

were detected in the lungs of mice injected with scrambledshRNA cells, no metastatic growth was detected in miceinjected with AMPKα2 knockdown cells (Fig. 4A, Fig. S4A).Similar results were obtained with MDA-MB-435S cells stablyexpressing scrambled or AMPKα2 shRNA (Fig. 4B, Fig. S4B).An MTT assay revealed no significant change in cellproliferation when AMPK was depleted (Fig. S4C), together

Fig. 2. See next page for legend.

5

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

suggesting that AMPK functions might be required for themetastasis of these cancer cells.Next, we investigated the mechanisms by which AMPK might

regulate EMT. Induction of EMT is associated with the upregulationof several transcription factors such as Twist1, Snail, Slug and Zeb1(Yang and Weinberg, 2008). Of these, Twist1 has been consideredto be a major regulator of EMT, invasion and metastasis (Wanget al., 2016; Yang et al., 2004). Therefore, we analyzed theexpression of Twist1 under conditions of AMPK activation. Wenoticed an AMPK-mediated increase in Twist1 transcripts in avariety of cell lines tested, including MCF7, BT474, A549, MDA-MB-231 and MDA-MB-435S cells (Fig. 5A). Notably, theepithelial-mesenchymal-transitioned cells showed a greaterincrease in Twist1 transcript levels upon AMPK activationcompared with the non-invasive cells (Fig. 5A). A western blotanalysis revealed elevated Twist1 protein levels in the presence ofA769662 (Fig. 5B and Fig. S5A). An immunocytochemicalanalysis further confirmed elevated Twist1 levels (Fig. 5C).To validate whether Twist1 regulation is through AMPK

activation, we next investigated the effects of AMPK inhibitionon Twist1 expression in the epithelial-mesenchymal-transitionedcancer cells lines A549, MDA-MB-231 and MDA-MB-435S.Indeed, in these cell lines, inhibition of AMPK with Compound Cled to a significant decrease in the levels of Twist1 transcripts(Fig. 5D). Consistent with this, MDA-MB-231 and MDA-MB-435S cells stably expressing AMPKα2 shRNA also showed lessexpression of Twist1 compared with scrambled cells (Fig. 5E).

Importantly, inhibition of AMPK prevented the induction of Twist1expression by CoCl2 or TGFβ treatment (Fig. 5F,G). Takentogether, these data suggested that AMPK might mediate itseffects on EMT through upregulation of Twist1.

To gauge the role of Twist1 downstream of AMPK in the processof EMT, we investigated the effects of Twist1 depletion in AMPK-mediated EMT response of cancer cells. For this, we generatedMDA-MB-231 cells stably expressing Twist1 shRNA (Mani et al.,2008). These cells showed ∼90% reduction of Twist1 both attranscript level (Fig. 5H) and protein level (Fig. S5B,C) comparedwith control shGFP cells. Consistent with the literature (Mani et al.,2008), Twist1 knockdown led to a reduction in the expression ofEMT markers such as Vim, N-cad, fibronectin 1 (FN1; also knownas CIG), Snail and Slug (Fig. 5I). As shown previously, AMPKactivation with A769662 led to increased EMTmarker expression inthe control shGFP cells; however, AMPK activation failed to do soin the Twist1 knockdown cells (Fig. 5I). Similar results wereobtained in MDA-MB-435S cells expressing Twist1 siRNA oligos(Fig. S5D,E), indicating that AMPK might mediate its effects onEMT, at least in part, through Twist1.

Furthermore, to determine whether AMPK requires Twist1 formediating functional changes associated with EMT, we investigatedthe effects of AMPK activation in the migration and invasion ofTwist1 knockdown cells. As seen before, AMPK activation led to anincrease in the migration and invasive potential of control shGFPcells (Fig. 5J,K and Fig. S5F); however, AMPK activation inshTwist1 cells failed to do so (Fig. 5J,K and Fig. S5F).

Because Twist1 is a transcription factor and its nuclearlocalization is critical for its function, we investigated nuclearlocalization of Twist1 under AMPK-activated conditions. As seenin the immunofluorescence images (Fig. 5C), in addition toincreased levels of Twist1, we also noted increased nuclearlocalization of Twist1 in the AMPK-activated condition. BecauseAMPK activation led to a significant increase in Twist1 in MDA-MB-231 cells (Fig. 5A and Fig. S5A), we performed subcellularfractionation experiments with the same cells. Cell fractionationexperiments also showed elevated nuclear Twist1 upon AMPKactivation (Fig. 6A). Furthermore, immunofluorescence imaging ofMDA-MB-231 and MCF7 cells transiently transfected with a GFP-tagged Twist1 construct also showed increased nuclear localizationof Twist1 in the A769662-treated condition (Fig. 6B), suggestingAMPK-mediated post-translational modifications of Twist1 leadingto its nuclear localization. Because phosphorylation by mitogen-activated protein kinases (MAPKs), extracellular signal-regulatedkinases (ERKs) and Jun N-terminal kinases (JNKs) has been shownto increase serine phosphorylation and stability of Twist1 (Honget al., 2011b), we investigated the effect of AMPK activation onTwist1 stability. In cycloheximide chase experiments, inhibition ofAMPK led to a decrease in Twist1 stability (Fig. 6C). Moreover, weobserved increased Twist1 stability in the presence of AMPKactivator A769662 (data not shown). Taken together, our datarevealed that AMPK activation leads to an increase in theexpression, stability and nuclear localization of Twist1 protein,which ultimately could promote EMT.

In conclusion, our study uncovers a central role for AMPK in thepositive regulation of EMT in cancer cells, mediated, at least in part,through Twist1 upregulation.

DISCUSSIONEMT is suggested to be an initial and important step in the invasion-metastasis cascade (Thiery et al., 2009). Therefore, an increasedunderstanding of the molecular players involved in this process is

Fig. 2. AMPK induces a functional EMT. (A) Scratch assay was performedwith BT474 cells after treatment with 100 µM AMPK activator (A769662; A7) orDMSO (vehicle control). The graph represents time kinetics of woundconfluence percentage, calculated by IncuCyte ZOOM software. (B) Scratchassay was performed with A549 cells after treatment with 100 µM AMPKactivator (A769662), 10 µM AMPK inhibitor (Compound C) or DMSO (vehiclecontrol). The graph represents time kinetics of wound confluence percentage,calculated by IncuCyte ZOOMsoftware. Error bars represent s.e.m., n=2. (C,D)Scratch assay was performed with MDA-MB-231 (C) and MDA-MB-435S (D)cells after treatment with 100 µM AMPK activator (A769662), 10 µM AMPKinhibitor (Compound C) or DMSO (vehicle control). Graphs represent timekinetics of wound confluence percentage, calculated by IncuCyte ZOOMsoftware. Error bars represent s.e.m., n=3 and 2, respectively. (E) Scratchassay was performed with A549 cells stably expressing inducible shRNAagainst AMPKα2 (clone #4) cultured with and without doxycycline. The graphrepresents time kinetics of wound confluence percentage, calculated byIncuCyte ZOOM software. Error bars represent s.e.m., n=2. (F) Scratch assaywas performed with MDA-MB-231 cells stably expressing scrambled shRNA orAMPKα2 shRNA. The graph represents time kinetics of wound confluencepercentage, calculated by IncuCyte ZOOM software. Error bars represents.e.m., n=2. (G) Scratch assay was performed with MDA-MB-435S cellstransfected with control siRNA or AMPKα2 siRNA. Graphs represent thedistance migrated between the cell edges by 48 h, calculated by ImageJsoftware. Error bars represent s.e.m., n=3. (H,I) Boyden chamber invasionassay was performed for 24 h with MDA-MB-231 (H) and MDA-MB-435S cells(I) treated with 100 µM AMPK activator (A769662), 10 µM AMPK inhibitor(Compound C) or DMSO (vehicle control). Graphs represent number of cellsinvaded per field (9-10 fields were counted per experiment). Error barsrepresent s.e.m., n=3. (J,K) Boyden chamber invasion assay was performed for24 h with MDA-MB-231 cells stably expressing scrambled shRNA or AMPKα2shRNA; n=2 (J) or MDA-MB-435S cells transfected with control siRNA orAMPKα2 siRNA; n=3 (K). Graphs represent number of cells invaded per field(9-10 fields were counted per experiment). Error bars represent s.e.m. For allscratch/migration assay experiments above, cells were pretreatedwith 10 µg/mlMitomycinC for 2 h before the scratch/woundwasmade. For all migration assayexperiments performed using IncuCyte ZOOM, 10× magnification (wide mode)was used. *P<0.05; **P<0.01; ***P<0.001. A7, A769662; CC, Compound C;Ctrl, control; Scr, scrambled; sh, shRNA.

6

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

Fig. 3. AMPK is necessary for EMT induction bydifferent upstream stimuli. (A) BT474 cells stablyexpressing scrambled shRNA or AMPKα2 shRNAwere cultured in the presence of 150 µM CoCl2 for48 h. Thereafter, cells were harvested andimmunoblot analysis was undertaken. The graphrepresents densitometric quantification of thespecified proteins normalized to α-tubulin; errorbars represent s.e.m.; n=2. (B) A549 cells stablyexpressing inducible shRNA against AMPKα2(clone #4) were cultured with and withoutdoxycycline in ambient oxygen (20% O2) or inhypoxic (3% O2) conditions in a tri-gas incubatorfor 48 h. Thereafter, cells were harvested andsubjected to immunoblot analysis. The graphrepresents densitometric quantification of thespecified proteins normalized to α-tubulin; errorbars represent s.e.m.; n=2. (C) MDA-MB-435Scells were cultured in ambient oxygen (20% O2) orin hypoxic (3% O2) conditions in a tri-gas incubatorin the presence of 10 µM AMPK inhibitor(Compound C) for 24 h. Thereafter, cells wereharvested and subjected to immunoblot analysis.The graph represents densitometric quantificationof the specified proteins normalized to α-tubulin;error bars represent s.e.m.; n=4. (D) MDA-MB-435S cells were cultured with 5 ng/ml TGFβ in thepresence or absence of 10 µM AMPK inhibitor(Compound C) for 24 h. Thereafter, immunoblotanalysis was undertaken for the specified proteins.The graph represents densitometric quantificationof the specified proteins normalized to α-tubulin;error bars represent s.e.m.; n=4. *P<0.05;**P<0.01; ***P<0.001. α-Tub, α-tubulin; CC,Compound C; Doxy, doxycycline; ns, notsignificant; Scr, scrambled; sh, shRNA.

7

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

likely to lead to a better understanding of cancer metastasis andidentify newer targets for cancer treatment. Although severalsignaling pathways and physiological conditions have beenidentified as activators of EMT, and a plethora of transcriptionfactors have been identified as downstream effectors of EMT, it hasremained unclear whether these various inducers of EMT functionindependently, work in conjunction, or converge on a commondownstream mediator. Our study identifies a central role for AMPKin mediating an EMT downstream of several physiological cues.Even though the LKB1-AMPK axis was initially associated withtumor-suppressive roles (Huang et al., 2008), recent studies havedemonstrated pro-tumorigenic functions for AMPK, includinginhibition of apoptosis, promoting anchorage-independent growthand, more recently, cancer cell migration (Hindupur et al., 2014;Jeon and Hay, 2015; Kim et al., 2011; Ng et al., 2012; Wang et al.,2010). In our study, we provide evidence for a direct role of AMPKin the induction of EMT in cancer cells, as well as in themaintenance of the EMT phenotype. We further show thatactivation of AMPK suffices to bring about a functional EMTcharacterized by increased migration and invasion potential.Moreover, we show that AMPK activation is mandatory for EMTinduction mediated by upstream physiological cues such as hypoxiaand TGFβ. Additionally, we show that AMPK mediates its effectson EMT via upregulation of TWIST1 gene expression and itsincreased nuclear localization.AMPK activation with A769662 affected the expression of

epithelial markers E-cad and ZO-1 to different extents in differentcell lines promoting an EMT. In the epithelial breast cancer cellsMCF7, T47D and BT474, AMPK activation decreased theexpression of epithelial markers E-cad and ZO-1 and caused

dissolution of adherens junctions from the cell membrane (the firsthallmark of an active EMT), thus pushing the cells towards initiatingan EMT. Furthermore, even though AMPK activation upregulatedthe transcript levels of mesenchymal markers (Vim and N-cad) inthese cells, their protein levels are not changed (data not shown),and consistent with this, their migratory properties also did notchange. Hence, these cells seem to only ‘partially’ undergo an EMT.In metastable A549 cells, which already seem to be in a partial EMTstate as evident by E-cadlow and Vimlow phenotype, AMPKactivation significantly decreased the protein expression of E-cadand ZO-1 and upregulated mesenchymal markers (Fig. S6). Thus,metastable A549 cells, which seem to be ‘primed’ for EMT, arehence pushed further into EMT in response to AMPK activation.Mesenchymal MDA-MB-231 cells do not express E-cad whileexpressing ZO-1 at a low basal level. AMPK activation was able tofurther downregulate ZO-1 expression in these mesenchymal cells,and significantly increased the expression of mesenchymal markers(Fig. S6) and their migratory and invasive potential. Hence, itappears that although AMPK activation is able to downregulate theexpression of epithelial markers across all states of EMT indifferent cell types, additional cues might be required to alsoincrease the protein expression of mesenchymal markers inepithelial cells (MCF7, T47D and BT474) to bring about acomplete, functional EMT. In contrast, primed partial epithelial-mesenchymal-transitioned cells (A549) and mesenchymal cells(MDA-MB-231 and MDA-MB-435S) seem to already possessthe requisite machinery to bring about a complete, functionalEMT in response to AMPK activation.

On the other hand, AMPK inhibition or depletion increased theepithelial characteristics of the cells while decreasing mesenchymal

Fig. 4. AMPK knockdown impairs metastasis of breast cancer cells. (A,B) MDA-MB-231 and MDA-MB-435S cells stably expressing scrambled shRNAor AMPKα2 shRNA were injected through the tail vein of nude mice and scored for lung metastasis. Photomicrographs show histologic sectioning ofrepresentative lung metastases originating from MDA-MB-231 (A) and MDA-MB-435S (B) cells (magnification ×20; hematoxylin and eosin staining). Scale bars:30 μm. Graphs represent incidence of lung metastases found in mice in experiments A and B; n=5 mice for each group. Scr, scrambled; sh, shRNA.

8

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

properties. In the epithelial BT474 cells, AMPK knockdown led to afurther increase in epithelial markers (Fig. S6). In the partiallyepithelial-mesenchymal-transitioned A549 cells, AMPK inhibitionor knockdown led to an increase in epithelial markers E-cad and

ZO-1 in addition to reduction in mesenchymal markers Vim and N-cad. AMPK inhibition or depletion in the mesenchymal MDA-MB-231 and MDA-MB-435S cell lines decreased the expression ofmesenchymal markers Vim and N-cad (Fig. S6); however, it failed

Fig. 5. See next page for legend.

9

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

to bring back E-cad expression (data not shown). It is known that theE-cad promoter is hypermethylated in MDA-MB-435S cells,completely silencing E-cad expression, whereas it is partiallymethylated in MDA-MB-231 cells, leading to a very low expressionof E-cad that cannot be detected by western blotting (Lombaertset al., 2006). It is possible that AMPK signaling may not be able toalleviate such hypermethylation-induced gene silencing. These datasuggested that AMPK is required to maintain the EMT phenotype,whereas its inhibition leads to reversal of EMT, also known asmesenchymal-to-epithelial transition (MET) (Nieto et al., 2016).Our data revealed that activation of AMPK using A769662

promotes EMT. In contrast, metformin, also an AMPK activator,has been shown to inhibit the EMT phenotype of breast cancercells (Qu et al., 2014). Another study showed that activation ofAMPK by OSU-53, a new allosteric activator of AMPK, reversedEMT (Chou et al., 2014). Metformin was shown to inhibit EMTby AMPK-mediated abrogation of Erk activity, leading todownregulation of Snail and Slug in lung and breast cancer

cells (Banerjee et al., 2016). AMPK signaling has been shown toinhibit EMT in other cancer types such as melanomas, lung andprostate cancer (Chou et al., 2014; Kim et al., 2012; Lin et al.,2015). These discrepancies could be due to the extent of AMPKactivation by different pharmacological agents, or due to theirnon-specific activities. Metformin, a biguanide used as a first-lineanti-diabetic drug, is an indirect modulator of AMPK. It inhibitscomplex I of the mitochondrial respiratory chain, causing anincrease in intracellular AMP levels, thereby activating AMPK(Owen et al., 2000). However, AMPK-independent roles ofmetformin have been reported (Foretz et al., 2010; Kalender et al.,2010) and its use as a bona fide AMPK activator needs furtherinvestigation. A769662 and AICAR, both analogs of AMP, binddirectly to the auto-inhibitory domain of AMPK and do not causea change in the ATP:ADP ratio or alter mitochondrial function toexert their action. However, AICAR monophosphate (ZMP), theactive form of AICAR within cells, is a natural intermediate of thepurine nucleotide synthesis pathway and is metabolized byAICAR transformylase (Kim et al., 2016). Therefore, the turnoverof the active form of AICAR within cells is dependent on theirproliferation rate. Thus, these modulators of AMPK mightactivate it to different extents. Our results with A769662showed that AMPK is a positive regulator of EMT in multiplecancer cell lines. We have additionally corroborated our resultswith genetic approaches using either siRNA- or inducible-shRNA-mediated knockdown.

Most importantly, our data revealed that TGFβ, hypoxia, andoncogenic Ras-mediated EMT are significantly inhibited bydepletion or inhibition of AMPK. This suggested that activationof AMPK could be a key central event in the EMT program. Recentstudies have identified critical functions of AMPK underpathological/physiological stresses associated with tumorprogression, including nutrient deprivation, hypoxia, and matrixdetachment (Hindupur et al., 2014; Steinberg and Kemp, 2009).Our identification of a stress-response kinase as a necessarycomponent of the EMT program additionally supports the notionthat the phenomenon of EMT might be a stress response that aidscancer cell dissemination and metastasis. Furthermore, given thatthe earlier studies have linked the attainment of an EMT phenotypewith the acquisition of stemness properties (Mani et al., 2008), it isplausible that AMPK activity may determine the differentiationstatus of epithelial cancers.

EMT is largely effected by a set of transcription factors whoseexpression and function are under tight control. Twist1, consideredas a master regulator of EMT and metastasis (Yang et al., 2004), isexpressed in a variety of cancers (Ansieau et al., 2010). Severaltranscriptional regulators of Twist1 have been identified, includingHIF1α, interleukin-6 (IL-6), NF-κB and the Smad-dependent TGFβpathway (Cheng et al., 2008; Pham et al., 2007; Tan et al., 2012).Our study showed that Twist1 gene expression is transcriptionallyregulated by AMPK. Indeed, AMPK has been shown to regulategene expression by phosphorylating various transcription factors(Cantó and Auwerx, 2010), or via its effects on chromatin structureby modulating histone deacetylase activity or histones (Bungardet al., 2010; McGee et al., 2008). However, the precise mechanismof Twist1 transcriptional regulation by AMPK remains to beelucidated.

Additionally, our study revealed that AMPK activation led toincreased stability and nuclear localization of Twist1. Such amulti-step regulation of Twist1 by AMPK suggests that the cellularevents downstream of the AMPK-Twist1 axis play a critical role inorchestrating the EMT program. Twist1 is also known to be

Fig. 5. AMPK regulates Twist1 expression. (A) Various cancer cell lines(MCF7, BT474, A549, MDA-MB-231 and MDA-MB-435S) were cultured in thepresence of 100 µM AMPK activator (A769662) or DMSO (vehicle control) for48 h. Thereafter, cells were harvested for RNA isolation and subjected toqPCR analysis for Twist1 gene expression. Graph represents fold changes(normalized to housekeeping gene β2M) of Twist1 gene expression in thespecified cell lines; error bars represent s.e.m., n=3. (B) MDA-MB-435S cellswere cultured in the presence of 100 μM AMPK activator (A769662) or DMSO(vehicle control) for 48 h. Thereafter, cells were harvested and subjected toimmunoblot analysis for Twist1 and α-tubulin. The graph representsdensitometric quantification of the Twist1 protein normalized to α-tubulin; errorbars represent s.e.m., n=3. (C) MDA-MB-231 cells were cultured in thepresence of 100 µM AMPK activator (A769662) or DMSO (vehicle control) for24 h. Thereafter, the dishes were subjected to immunocytochemistry analysisfor Twist1 protein. Photomicrographs show representative fluorescent imagestaken at 20× magnification using an Olympus 1X71 microscope; Hoechst33342 was used for nuclear staining, and was pseudo-colored green forrepresentation. Twist1 is stained in red. Scale bars: 10 μm. The graphrepresents cytoplasmic and nuclear fluorescence intensity (FI) measurementsper cell, n=30 cells quantified in each of the three biological replicates; the errorbar represents s.e.m. (D) qPCR analysis for Twist1 gene expression in A549,MDA-MB-231 and MDA-MB-435S cells cultured in the presence of 10 μMAMPK inhibitor (Compound C) or DMSO (vehicle control) for 48 h. The graphrepresents fold change (normalized to β2 M); error bars represent s.e.m., n=3.(E) qPCR analysis for Twist1 gene expression in MDA-MB-231 and MDA-MB-435S cells stably expressing scrambled shRNA or AMPKα2 shRNA. Thegraph represents fold change (normalized to β2M); error bars represent s.e.m.,n=3. (F,G) MDA-MB-231 and BT474 cells were cultured in the presence of10 µM AMPK inhibitor (Compound C) with 150 µM CoCl2 (F) or TGFβ (G) for48 h. Thereafter, cells were harvested and quantified for Twist1 mRNA levelsby qPCR. The graphs represent fold change (normalized to β2M); error barsrepresent s.e.m., n=3. (H) Quantification of Twist1 mRNA levels by qPCR inMDA-MB-231 cells stably expressing GFP shRNA or Twist1 shRNA. Thegraph represents fold change of Twist1 gene expression (normalized to β2M);error bars represent s.e.m., n=3. (I) MDA-MB-231 cells stably expressing GFPshRNA or Twist1 shRNA were cultured in the presence of 100 µM AMPKactivator (A769662) or DMSO (vehicle control) for 48 h. Thereafter, the cellswere harvested and qPCR analysis was undertaken for the specifiedtranscripts. The graph represents relative gene expression of the specifiedgenes (normalized to β2M expression); error bars represent s.e.m., n=4.(J,K) MDA-MB-231 cells stably expressing GFP shRNA or Twist1 shRNAwerecultured in the presence of 100 µM AMPK activator (A769662) or DMSO(vehicle control). (J) The graph represents the distance migrated between thecell edges in a scratch assay in 48 h, calculated by Image J software; error barsrepresent s.e.m., n=3. (K) The specified cells treated similarly were alsosubjected to Boyden chamber invasion assay. The graph represents the totalnumber of cells invaded per field; error bars represent s.e.m., n=4. *P<0.05;**P<0.01; ***P<0.001. α-Tub, α-tubulin; A7, A769662; CC, Compound C; Scr,scrambled; sh, shRNA.

10

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

phosphorylated by kinases such as Akt and MAPK, which affectits anti-apoptotic function and stability (Hong et al., 2011b;Vichalkovski et al., 2010). Although Twist1 possesses an AMPKconsensus motif (data not shown), AMPK-mediated phosphorylation

of Twist1 is yet to be specified. Interestingly, we also observed asignificant increase or decrease in levels of other EMTtranscription factors such as Snail, Slug and Zeb1 on AMPKactivation or inhibition, respectively. Furthermore, like Twist1,

Fig. 6. See next page for legend.

11

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

Snail and Slug also possess an AMPK consensus motif in theirprotein sequence (data not shown), suggesting that AMPK mightadditionally regulate these EMT transcription factors byphosphorylating them. Hence, although in the present study wehave investigated the AMPK-Twist1 axis, we speculate thatAMPK might also be able to regulate EMT mediated by othertranscription factors.The eventual outgrowth of metastatic colonies at a secondary

organ site is the culmination of a series of steps referred to as theinvasion-metastasis cascade. EMT is proposed to be one of the earlysteps that helps initiate this cascade, enabling the epithelial cells atthe primary tumor site to undergo morphogenetic changes andassume a mesenchymal morphology that enables the cancer cells tomigrate and invade following detachment from the extracellularmatrix. In fact, previous studies have shown that AMPK is activatedon detachment from the matrix (Hindupur et al., 2014; Jeon andHay, 2012), supporting our finding that AMPK activation leads toEMT. After traversing through blood/lymphatic vessels, theepithelial-mesenchymal-transitioned cancer cells extravasate at asecondary site, revert their morphology to an epithelial phenotype,re-attach to the matrix, and then undergo a clonal outgrowth calledmicro-/macro-metastasis. This process is referred to as MET (Nietoet al., 2016). Indeed, we have previously shown that re-attachmentto the matrix downmodulates AMPK activity (Sundararaman et al.,2016), suggesting that at later stages of metastasis, AMPK might beinactivated, causing MET and thus helping the formation of a newtumor growth at a secondary site. This is consistent with ourobservation that AMPK inhibition leads to a reduction in EMTmarkers. Thus, matrix-adhesion state could regulate AMPK activity,which in turn could regulate EMT and MET reversibly, thusenabling metastasis (Fig. 6D).Taken together, our study elicits an important role for AMPK in

tumor progression by regulating EMT, migration and invasion ofbreast cancer cells. The requirement of AMPK downstream ofseveral cues that are known activators of EMT in cancer cellssuggests a possible central role for AMPK in orchestrating the EMT

program. Targeting AMPK might thus serve as a novel therapeuticstrategy to control cancer metastasis. Because AMPK plays a majorrole in maintaining energy homeostasis under metabolically stressedconditions, its inhibition for a therapeutic window with minimalsystemic effects might be exploited for cancer treatment in the futurewith the development of specific AMPK inhibitors for clinical use.

MATERIALS AND METHODSCell culture and reagentsMDA-MB-231, MCF7, T47D (breast epithelial cancer cell lines), MDA-MB-435S (melanoma) and A549 (lung cancer) cells were cultured inDMEM (Sigma-Aldrich, St. Louis, MO) supplemented with 10% FBS,penicillin (1 kU/ml) and streptomycin (0.1 mg/ml). BT474 (breastcancer) cells were cultured in RPMI (Sigma-Aldrich) supplementedwith 10% FBS, penicillin (1 kU/ml) and streptomycin (0.1 mg/ml).HMLER cells (provided by Prof. Robert Weinberg, Department ofBiology, Massachusetts Institute of Technology, Cambridge, MA) werecultured in DME-F12 medium supplemented with hEGF (10 ng/ml),hydrocortisone (0.5 μg/ml) and insulin (10 μg/ml).

Pharmacological chemicals used in this study include A769662 (100 µM;purchased from University of Dundee, Dundee, UK), 6-[4-(2-piperidin-1-ylethoxy-phenyl)]-3-pyridin-4-yl-pyrrazolo [1,5-a]-pyrimidine (CompoundC) (10 µM; Calbiochem, Gibbstown, NJ) and cycloheximide (100 μg/ml;Sigma-Aldrich). Absolute dimethyl sulfoxide (DMSO) (Calbiochem) wasused as a vehicle for A769662 and Compound C. Recombinant humanTGFβ (R&D Systems, Minneapolis, MN) was used at a concentration of5 ng/ml. For hypoxia-based experiments, cells were either cultured in 3%oxygen using a tri-gas incubator (Thermo Fisher, Waltham, MA) or treatedwith 150 μM cobalt chloride.

Antibodies and immunoblottingWhole cell lysates for western blotting were prepared with lysis buffercontaining 1% NP-40 detergent, 0.5% sodium deoxycholate, 0.1% SDS,50 mM sodium fluoride, 1 mM sodium orthovanadate, 10 mM sodiumpyrophosphate (Sigma-Aldrich) and protease inhibitors (Roche, Mannheim,Germany). Protein concentration was estimated with Bradford reagent andan equal amount of protein (25 μg) was resolved by SDS-PAGE using Bio-Rad apparatus, transferred to a PVDF membrane (Millipore, Billerica, MA)and probed with appropriate antibodies. For all western blots involvingphospho-protein analysis, the levels of respective total proteins were probedand shown to be equal. α-Tubulin (Calbiochem) served as a loading control.HRP-coupled secondary antibodies were obtained from The JacksonLaboratory, and immunoblots were visualized using PICO reagent(Pierce, Waltham, MA). Primary antibodies used were: phospho-AMPK(Thr172), total AMPKα2, phospho-ACC, total ACC (recognizes bothisoforms 1 and 2), phospho-Smad2, total Smad2 and vimentin (CellSignaling Technology, Beverly, MA); N-cadherin and E-cadherin(Epitomics, Burlingame, CA); Twist1 (Sigma-Aldrich); and HIF1α(Upstate, Billerica, MA). Multiple panels for a given experiment weredeveloped from the same blot by re-probing. Alternatively, the same lysateswere loaded as technical replicates; for each technical run, α-tubulin wasprobed as a loading control. All primary antibodies were used at 1:1000dilution in 5% BSA, whereas HRP-conjugated secondary antibodies wereused at 1:3000 dilution.

Quantitative densitometric analysis of the EMT-related immunoblots wasundertaken using ImageJ software. Protein expression was normalized to α-tubulin; the normalized expression of control was set to 1, and an increase ordecrease in expression under various conditions was calculated relative tothe control.

Nuclear-cytoplasmic fractionationFor fractionation experiments MDA-MB-435S cells were grown in 90-mmdishes until they reached 80% confluence. The cells were then treated withA769662 for 2 h and lysed in mild lysis buffer [0.5%NP-40, 10 mM sodiumpyrophosphate, 1 mM sodium orthovanadate, and 1× protease inhibitor(Roche) diluted in PBS]. The supernatant obtained after spinning the lysateserved as the cytoplasmic fraction. The pellet obtained was gently washed

Fig. 6. AMPK mediates EMT via increased Twist1 stability and its nucleartranslocation. (A) MDA-MB-231 cells were cultured in the presence of 100 μMAMPK activator (A769662) or DMSO (vehicle control) for 48 h. Thereafter, cellswere harvested, cytoplasmic and nuclear proteins were fractionated andsubjected to immunoblot analysis for Twist1, pAMPKα, lamin and α-tubulin, n=3.The graph represents densitometric quantification of nuclear and cytoplasmicTwist1 levels; error bars represent s.e.m., n=4. *P<0.05. (B) MCF7 and MDA-MB-231 cells were transiently transfected with Twist1 tagged to GFP andsubsequently cultured in the presence of 100 µM AMPK activator (A769662) orDMSO (vehicle control) for 24 h. Thereafter, the cells were fixed and localizationof GFP fluorescence was analyzed. Photomicrographs show representativefluorescent images taken at 20× magnification using an Olympus 1X71microscope; Hoechst 33342 was used for nuclear staining, and was pseudo-colored red for representation. Twist1-GFP is in green. Scale bars: 10 μm. Thegraphs represent nuclear and cytoplasmic localization of Twist1-GFP, n=30 cellsquantified in each of the three biological replicates; error bars represent s.e.m.*P<0.05; **P<0.01. (C) MDA-MB-435S cells were cultured in the presence of10 μMAMPK inhibitor (Compound C) or DMSO (vehicle control) for 6 h followedby inhibition of protein translation by cycloheximide treatment for 0, 2, 4 and 6 h.Cells were harvested and subjected to immunoblot analysis for Twist1 and α-tubulin. The graph represents densitometric quantification of Twist1 levelsnormalized to α-tubulin. (D) Model: Stresses faced by tumor cells in the primarytumor environment, such as hypoxia, nutrient deprivation and matrixdetachment, activate AMPK, which in turn induces EMT, allowing tumor cells toinitiate the invasion-metastasis cascade. Re-attachment at distant sites causesinhibition of AMPK signaling, enabling reversal of EMT and establishment ofsecondary tumors. α-Tub, α-tubulin; A7, A769662; CC, Compound C; CHX,cycloheximide.

12

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

twice with 1× PBS, subjected to a freeze-thaw cycle at −80°C and thenresuspended in western lysis buffer cocktail. The supernatant obtained afterspinning the lysate served as the nuclear fraction. The cytoplasmic and thenuclear fractions were electrophoresed on an SDS-PAGE gel andsubsequently subjected to immunoblotting.

TransfectionAll plasmid transfections were performed using Lipofectamine 2000(Invitrogen, Carlsbad, CA) as per the manufacturer’s instructions. TwistshRNA was provided by Prof. Sendurai Mani (MD Anderson CancerCenter, Houston, TX).

RNAi experimentsAMPKα2 knockdown stable cells were generated by transfecting BT474,MDA-MB-435S and MDA-MB-231 (Hindupur et al., 2014) cells with apool of four shRNA constructs (HuSH-29 shRNA vectors; Origene,Rockville, MD) targeting AMPKα2. Stable cells were generated usingpuromycin selection followed by sorting for RFP expression (encoded bythe vector) and were expanded and frozen for future use. Knockdown wasconfirmed by immunoblotting. Simultaneously, all of the cell types weretransfected with scrambled shRNA and stable cells were generated as above.MDA-MB-231 and BT474 cells with a stable knockdown of AMPKα2 werevalidated and described previously (Hindupur et al., 2014).

Inducible AMPKα2 knockdown cells were generated by transfecting thepTRIPZ vector expressing shRNA against AMPKα2 under the Tet-Onpromoter [Seq no. 1: V2THS_57638 (Clone #1); Seq no. 4: V2THS_375319(Clone #4); Dharmacon, Pittsburgh, PA] in A549 and MDA-MB-231 cells.Stable transfectants were selected using puromycin, followed by sorting forRFP (encoded by the vector) after induction with doxycycline (5 µg/ml) for48 h. Knockdown was confirmed using immunoblotting. For furtherexperiments, uninduced stable cells were used as control.

For siRNA transfections, cells were seeded in 60-mm dishes at 50%confluence and were transfected with non-targeting control siRNA (Sigma-Aldrich), Twist1 siRNA (Sigma-Aldrich) or AMPKα2 siRNA (CellSignaling Technology) using Oligofectamine (Invitrogen) as per themanufacturer’s instructions.

RNA extraction and RT-PCRTotal RNA was isolated using TRI reagent (Sigma-Aldrich). Reversetranscription of mRNA was carried out using a Gene-Amp RNA PCRcDNA synthesis kit (Applied Biosystems, Carlsbad, CA). Specific primers(Table S1) were designed using Primer 3.0 software. Hypoxanthine-guaninephosphoribosyltransferase (HPRT) or β2 microglobulin (β2M) were used asnormalizing control.

Wound-healing scratch assayMigration assay was performed using an ImageLock 96-well plate in theIncuCyte ZOOM live-cell analysis system (Essen Bioscience), where 2×104

of control and knockdown of MDA-MB-231, BT474 and A549 cells wereseeded into each well and treated with 10 µg/ml of Mitomycin C(Calbiochem) for 2 h. A scratch wound was made using WoundMaker™(Essen Bioscience). Cells were then washed with 1× PBS and incubatedwith culture medium without serum and imaged every 1 h. The extent ofwound healing was determined as wound confluence percentage usingIncuCyte software (Essen Bioscience). Each individual experiment hadthree to five technical replicates.

Independently, an equal number of MDA-MB-435S cells were seeded in60-mm dishes. siRNA transfection was performed for 48 h. Cells weretreated with 10 μg/ml Mitomycin C (Calbiochem) for 2 h to arrestproliferation, following which, two wounds were made using a P-200pipette tip. Photomicrographs were taken at 0 and 48 h of wound generation.The distance migrated was quantified using ProgRes capture software andplotted as a difference of distance migrated between 0 and 48 h.

ECIS wound-healing assayReal-time quantitative wound-healing assays were performed using theElectric Cell-substrate Impedance Sensing (ECIS; model 1600R; AppliedBioPhysics, Troy, NY) and the 8W10E ECIS arrays. MDA-MB-435S cells

were treated with A769662 or Compound C for 48 h followed bytreatment with Mitomycin C for 2 h to arrest proliferation, trypsinized anda total of 2×105 live cells seeded in three wells of the 8W10E ECIS array.One well with only medium (no cells) served as a negative control toprovide the baseline changes in impedance. Cells were allowed to attachfor 24 h, after which a wound was made using an elevated AC voltagepulse with a frequency of 40 kHz, and duration of 30 s. Dead cells werewashed away by replacing the medium. Migratory response over the next24 h was measured in real time by recording the recovery of electricalimpedance.

Matrigel invasion assayMDA-MB-231 and MDA-MB-435S cells treated with A769662 orCompound C for 48 h were trypsinized and a total of 50,000 live cellswere seeded in the BD BioCoat™Matrigel™ invasion chambers. DMSOserved as the vehicle control. After 24 h, the non-invaded cells wereremoved, and the invaded cells were fixed with 4% paraformaldehyde andstained with Crystal Violet Blue. Photomicrographs of the invasionchambers were taken at 10× magnification. The number of invaded cellswas counted from these images using ImageJ software and plotted as agraph. MDA-MB-231 (2×104 cells) and MDA-MB-435S (5×104) cellsstably expressing control shRNA or AMPKα2 shRNA were similarlysubjected to Matrigel invasion assays.

Metastasis assayA total of 1×106 cells (control or AMPKα2 knockdown cells) in 50 µl ofDMEM were injected into the lateral vein of nude mice. The mice wereeuthanized after 9 weeks of injection. The lungs were dissected out, fixed in4% paraformaldehyde overnight and embedded in paraffin. Hematoxylinand eosin staining was undertaken on multiple lung sections and scored by apathologist for the detection of metastatic foci. Animal experiments wereperformed in compliance with CPCSEA guidelines.

ImmunofluorescenceVarious cancer cell lines or stable cells generated were treated with A769662(100 µM), Compound C (10 µM) or vehicle control DMSO for 24 or 48 h asmentioned earlier and fixed with 4% paraformaldehyde at room temperaturefor 2 h. The cells were permeabilized with PBS containing Triton-X 100,blocked with 1% BSA and probed with antibodies (1:200 dilution) at 4°Covernight. The cells were probed with secondary antibody tagged with FITCor Cy3 (1:200 dilution) for 2 h at room temperature. The cells were thenwashed, mounted and visualized under an epifluorescence microscope(Olympus IX71). Further image processing and quantification wasperformed using ImageJ software. For relative fluorescence intensity (RFI)quantification, fluorescence intensity was first normalized to the number ofnuclei; the normalized intensity of control was set to 1, and the change inintensity under various conditions was plotted relative to the control.

Proliferation assayMDA-MB-231 (5×103), BT474 (8×103) and A549 (5×103) cells expressingscrambled shRNA or AMPKα2 shRNAwere seeded into a 96-well plate tomeasure their proliferation rates. Cells were seeded in triplicates for eachtime point (day 0 to day 4). At each time point, 20 µl of MTT reagent (5 mg/ml) was added to each well of the plate and incubated for 4 h. The incubationmedium was then removed and 100 µl of DMSO was added to each well todissolve the formazan crystals, then absorbance was measured using a platereader at 570 nm. A proliferation rate graph was plotted using GraphPadPrism software.

Determination of protein stabilityFor Twist1 protein stability experiments following AMPK inhibition,MCF7cells were seeded in eight 60-mm dishes at 30-40% confluence, andtransfected with 1 μg of pcDNA3-Flag Twist1 each from a commonDNA-transfection reagent mix. After 24 h of seeding, four disheswere treated with Compound C for 6 h, following which, 100 μg/mlcycloheximide was added to all of the dishes. Compound-C-treated anduntreated dishes were harvested at 0, 2, 4 and 6 h of cycloheximide treatmentfor immunoblotting.

13

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

StatisticsAll statistical analysis was performed using ratio t-test one-way ANOVAand two-way ANOVA with GraphPad Prism 5.0 software. All data werepresented as means±s.e.m. of three independent experiments unlessotherwise stated. P<0.05 was considered to be statistically significant.

AcknowledgementsWe thank Prof. Robert Weinberg for HMLER breast epithelial cell lines. We thankProf. Sendurai A. Mani for the Twist-shRNA construct. We acknowledgeGopalkrishnashetty Sreenivasmurthy Sravan and Shubham Pandey for helping withwestern blot experiments.

Competing interestsThe authors declare no competing or financial interests.

Author contributionsConceptualization: M.S., S.A.B., N.D., S.R., D.M.P., S.K.H., A.R.; Methodology:M.S., S.A.B., N.D., S.R., D.M.P., S.K.H., A.R.; Validation: M.S., S.A.B., N.D., S.R.;Investigation: M.S., A.R.; Writing - original draft: M.S., S.A.B., D.M.P., A.R.; Writing -review & editing: S.A.B., N.D., S.R., A.R.; Supervision: A.R.; Project administration:A.R.; Funding acquisition: A.R.

FundingM.S. was a recipient of a CSIRSenior Research Fellowship; S.A.B. was a recipient ofa Prime Minister’s Research Fellowship. This work was supported by a WellcomeTrust DBT India Alliance Fellowship [grant number 500112-Z-09-Z] awarded to A.R.We also acknowledge funding from the University Grants Commission [grant number6-4/2016(IC)], Indian Institute of Science (IISc), the DBT-IISc partnershipprogramme, and support fromDepartment of Biotechnology , Ministry of Science andTechnology, Government of India, to the Department of Molecular Reproduction,Development and Genetics. Deposited in PMC for immediate release.

Supplementary informationSupplementary information available online athttp://jcs.biologists.org/lookup/doi/10.1242/jcs.208314.supplemental

ReferencesAnsieau, S., Morel, A.-P., Hinkal, G., Bastid, J. and Puisieux, A. (2010).TWISTing an embryonic transcription factor into an oncoprotein. Oncogene 29,3173-3184.

Banerjee, P., Surendran, H., Chowdhury, D. R., Prabhakar, K. andPal, R. (2016).Metformin mediated reversal of epithelial to mesenchymal transition is triggeredby epigenetic changes in E-cadherin promoter. J. Mol. Med. 94, 1397-1409.

Bungard, D., Fuerth, B. J., Zeng, P.-Y., Faubert, B., Maas, N. L., Viollet, B.,Carling, D., Thompson, C. B., Jones, R. G. and Berger, S. L. (2010). Signalingkinase AMPK activates stress-promoted transcription via histone H2Bphosphorylation. Science 329, 1201-1205.

Canto, C. andAuwerx, J. (2010). AMP-activated protein kinase and its downstreamtranscriptional pathways. Cell. Mol. Life Sci. 67, 3407-3423.

Chaffer, C. L. andWeinberg, R. A. (2011). A perspective on cancer cell metastasis.Science 331, 1559-1564.

Chen, M.-B., Wu, X.-Y., Gu, J.-H., Guo, Q.-T., Shen, W.-X. and Lu, P.-H. (2011).Activation of AMP-activated protein kinase contributes to doxorubicin-induced celldeath and apoptosis in cultured myocardial H9c2 cells. Cell Biochem. Biophys.60, 311-322.

Cheng, G. Z., Zhang, W. Z., Sun, M., Wang, Q., Coppola, D., Mansour, M., Xu,L. M., Costanzo, C., Cheng, J. Q. and Wang, L.-H. (2008). Twist istranscriptionally induced by activation of STAT3 and mediates STAT3oncogenic function. J. Biol. Chem. 283, 14665-14673.

Chiu, Y.-C., Shieh, D.-C., Tong, K.-M., Chen, C.-P., Huang, K.-C., Chen, P.-C.,Fong, Y.-C., Hsu, H.-C. and Tang, C.-H. (2009). Involvement of AdipoR receptorin adiponectin-induced motility and alpha2beta1 integrin upregulation in humanchondrosarcoma cells. Carcinogenesis 30, 1651-1659.

Chou, C.-C., Lee, K.-H., Lai, I.-L., Wang, D., Mo, X., Kulp, S. K., Shapiro, C. L. andChen, C.-S. (2014). AMPK reverses the mesenchymal phenotype of cancer cellsby targeting the Akt-MDM2-Foxo3a signaling axis. Cancer Res. 74, 4783-4795.

Cool, B., Zinker, B., Chiou, W., Kifle, L., Cao, N., Perham, M., Dickinson, R.,Adler, A., Gagne, G., Iyengar, R. et al. (2006). Identification and characterizationof a small molecule AMPK activator that treats key components of type 2 diabetesand the metabolic syndrome. Cell Metab. 3, 403-416.

Cufı, S., Vazquez-Martin, A., Oliveras-Ferraros, C., Martin-Castillo, B., Joven, J.and Menendez, J. A. (2010). Metformin against TGFbeta-induced epithelial-to-mesenchymal transition (EMT): from cancer stem cells to aging-associatedfibrosis. Cell Cycle 9, 4461-4468.

Elenbaas, B., Spirio, L., Koerner, F., Fleming, M. D., Zimonjic, D. B., Donaher,J. L., Popescu, N. C., Hahn, W. C. and Weinberg, R. A. (2001). Human breast

cancer cells generated by oncogenic transformation of primary mammaryepithelial cells. Genes Dev. 15, 50-65.

Foretz, M., Hebrard, S., Leclerc, J., Zarrinpashneh, E., Soty, M., Mithieux, G.,Sakamoto, K., Andreelli, F. and Viollet, B. (2010). Metformin inhibits hepaticgluconeogenesis in mice independently of the LKB1/AMPK pathway via adecrease in hepatic energy state. J. Clin. Invest. 120, 2355-2369.

Han, B., Cui, H., Kang, L., Zhang, X., Jin, Z., Lu, L. and Fan, Z. (2015). Metformininhibits thyroid cancer cell growth, migration, and EMT through the mTORpathway. Tumor Biol. 36, 6295.

Hardie, D. G., Ross, F. A. and Hawley, S. A. (2012). AMPK: a nutrient and energysensor that maintains energy homeostasis. Nat. Rev. Mol. Cell Biol. 13, 251-262.

He, K., Guo, X., Liu, Y., Li, J., Hu, Y., Wang, D. and Song, J. (2016). TUFMdownregulation induces epithelial-mesenchymal transition and invasion in lungcancer cells via amechanism involving AMPK-GSK3beta signaling.Cell. Mol. LifeSci. 73, 2105-2121.

Hindupur, S. K., Balaji, S. A., Saxena, M., Pandey, S., Sravan, G. S., Heda, N.,Kumar, M. V., Mukherjee, G., Dey, D. and Rangarajan, A. (2014). Identificationof a novel AMPK-PEA15 axis in the anoikis-resistant growth of mammary cells.Breast Cancer Res. 16, 420.

Hong, J., Kandasamy, K., Marimuthu, M., Choi, C. S. and Kim, S. (2011a).Electrical cell-substrate impedance sensing as a non-invasive tool for cancer cellstudy. Analyst 136, 237-245.

Hong, J., Zhou, J., Fu, J., He, T., Qin, J., Wang, L., Liao, L. and Xu, J. (2011b).Phosphorylation of serine 68 of Twist1 by MAPKs stabilizes Twist1 protein andpromotes breast cancer cell invasiveness. Cancer Res. 71, 3980-3990.

Huang, X., Wullschleger, S., Shpiro, N., McGuire, V. A., Sakamoto, K., Woods,Y. L., McBurnie, W., Fleming, S. and Alessi, D. R. (2008). Important role of theLKB1-AMPK pathway in suppressing tumorigenesis in PTEN-deficient mice.Biochem. J. 412, 211-221.

Huang, R. Y.-J., Guilford, P. and Thiery, J. P. (2012). Early events in cell adhesionand polarity during epithelial-mesenchymal transition. J. Cell Sci. 125, 4417-4422.

Jeon, S.-M. and Hay, N. (2012). The dark face of AMPK as an essential tumorpromoter. Cell. Logistics 2, 197-202.

Jeon, S.-M. and Hay, N. (2015). The double-edged sword of AMPK signaling incancer and its therapeutic implications. Arch. Pharm. Res. 38, 346-357.

Jones, R. G., Plas, D. R., Kubek, S., Buzzai, M., Mu, J., Xu, Y., Birnbaum, M. J.and Thompson, C. B. (2005). AMP-activated protein kinase induces a p53-dependent metabolic checkpoint. Mol. Cell 18, 283-293.

Kalender, A., Selvaraj, A., Kim, S. Y., Gulati, P., Brûle, S., Viollet, B., Kemp,B. E., Bardeesy, N., Dennis, P., Schlager, J. J. et al. (2010). Metformin,independent of AMPK, inhibits mTORC1 in a rag GTPase-dependent manner.Cell Metab. 11, 390-401.

Kim, E.-K., Park, J.-M., Lim, S., Choi, J. W., Kim, H. S., Seok, H., Seo, J. K., Oh,K., Lee, D.-S., Kim, K. T. et al. (2011). Activation of AMP-activated protein kinaseis essential for lysophosphatidic acid-induced cell migration in ovarian cancercells. J. Biol. Chem. 286, 24036-24045.

Kim, H.-S., Kim, M.-J., Kim, E. J., Yang, Y., Lee, M.-S. and Lim, J.-S. (2012).Berberine-induced AMPK activation inhibits the metastatic potential of melanomacells via reduction of ERK activity and COX-2 protein expression. Biochem.Pharmacol. 83, 385-394.

Kim, J., Yang, G., Kim, Y., Kim, J. and Ha, J. (2016). AMPK activators:mechanisms of action and physiological activities. Exp. Mol. Med. 48, e224.

Liang, J., Shao, S. H., Xu, Z.-X., Hennessy, B., Ding, Z., Larrea, M., Kondo, S.,Dumont, D. J., Gutterman, J. U., Walker, C. L. et al. (2007). The energy sensingLKB1-AMPK pathway regulates p27(kip1) phosphorylationmediating the decisionto enter autophagy or apoptosis. Nat. Cell Biol. 9, 218-224.

Lin, H., Li, N., He, H., Ying, Y., Sunkara, S., Luo, L., Lv, N., Huang, D. and Luo, Z.(2015). AMPK inhibits the stimulatory effects of TGF-beta on Smad2/3 activity, cellmigration, and epithelial-to-mesenchymal transition. Mol. Pharmacol. 88,1062-1071.

Liu, X., Chhipa, R. R., Nakano, I. and Dasgupta, B. (2014). The AMPK inhibitorcompound C is a potent AMPK-independent antiglioma agent.Mol. Cancer Ther.13, 596-605.

Lombaerts, M., van Wezel, T., Philippo, K., Dierssen, J. W. F., Zimmerman,R. M. E., Oosting, J., van Eijk, R., Eilers, P. H., van de Water, B., Cornelisse,C. J. et al. (2006). E-cadherin transcriptional downregulation by promotermethylation but not mutation is related to epithelial-to-mesenchymal transition inbreast cancer cell lines. Br. J. Cancer 94, 661-671.

Mani, S. A., Guo, W., Liao, M.-J., Eaton, E. N., Ayyanan, A., Zhou, A. Y., Brooks,M., Reinhard, F., Zhang, C. C., Shipitsin, M. et al. (2008). The epithelial-mesenchymal transition generates cells with properties of stem cells. Cell 133,704-715.

Martinelli, R., Gegg, M., Longbottom, R., Adamson, P., Turowski, P. andGreenwood, J. (2009). ICAM-1-mediated endothelial nitric oxide synthaseactivation via calcium and AMP-activated protein kinase is required fortransendothelial lymphocyte migration. Mol. Biol. Cell 20, 995-1005.

McGee, S. L., van Denderen, B. J. W., Howlett, K. F., Mollica, J., Schertzer, J. D.,Kemp, B. E. and Hargreaves, M. (2008). AMP-activated protein kinase regulatesGLUT4 transcription by phosphorylating histone deacetylase 5. Diabetes 57,860-867.

14

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience

Mukherjee, S. (2010). The Emperor of All Maladies: A Biography of Cancer.New York, NY: Scribner.

Mungai, P. T., Waypa, G. B., Jairaman, A., Prakriya, M., Dokic, D., Ball, M. K. andSchumacker, P. T. (2011). Hypoxia triggers AMPK activation through reactiveoxygen species-mediated activation of calcium release-activated calciumchannels. Mol. Cell. Biol. 31, 3531-3545.

Nagata, D., Mogi, M. and Walsh, K. (2003). AMP-activated protein kinase (AMPK)signaling in endothelial cells is essential for angiogenesis in response to hypoxicstress. J. Biol. Chem. 278, 31000-31006.

Nakano, A., Kato, H., Watanabe, T., Min, K.-D., Yamazaki, S., Asano, Y.,Seguchi, O., Higo, S., Shintani, Y., Asanuma, H. et al. (2010). AMPK controlsthe speed of microtubule polymerization and directional cell migration throughCLIP-170 phosphorylation. Nat. Cell Biol. 12, 583-590.

Ng, T. L., Leprivier, G., Robertson, M. D., Chow, C., Martin, M. J., Laderoute,K. R., Davicioni, E., Triche, T. J. and Sorensen, P. H. B. (2012). The AMPKstress response pathway mediates anoikis resistance through inhibition of mTORand suppression of protein synthesis. Cell Death Differ. 19, 501-510.

Nieto, M. A., Huang, R. Y.-J., Jackson, R. A. and Thiery, J. P. (2016). Emt: 2016.Cell 166, 21-45.

Owen, M. R., Doran, E. and Halestrap, A. P. (2000). Evidence that metforminexerts its anti-diabetic effects through inhibition of complex 1 of the mitochondrialrespiratory chain. Biochem. J. 348, 607-614.

Pham, C. G., Bubici, C., Zazzeroni, F., Knabb, J. R., Papa, S., Kuntzen, C. andFranzoso, G. (2007). Upregulation of Twist-1 by NF-kappaB blocks cytotoxicityinduced by chemotherapeutic drugs. Mol. Cell. Biol. 27, 3920-3935.

Polyak, K. and Weinberg, R. A. (2009). Transitions between epithelial andmesenchymal states: acquisition of malignant and stem cell traits. Nat. Rev.Cancer 9, 265-273.

Qu, C., Zhang, W., Zheng, G., Zhang, Z., Yin, J. and He, Z. (2014). Metforminreverses multidrug resistance and epithelial-mesenchymal transition (EMT) viaactivating AMP-activated protein kinase (AMPK) in human breast cancer cells.Mol. Cell. Biochem. 386, 63-71.

Rios, M., Foretz, M., Viollet, B., Prieto, A., Fraga, M., Costoya, J. A. and Senaris,R. (2013). AMPK activation by oncogenesis is required to maintain cancer cellproliferation in astrocytic tumors. Cancer Res. 73, 2628-2638.

Sanders, M. J., Ali, Z. S., Hegarty, B. D., Heath, R., Snowden, M. A. and Carling,D. (2007). Defining the mechanism of activation of AMP-activated protein kinaseby the small molecule A-769662, a member of the thienopyridone family. J. Biol.Chem. 282, 32539-32548.

Sommers, C. L., Thompson, E. W., Torri, J. A., Kemler, R., Gelmann, E. P. andByers, S. W. (1991). Cell adhesion molecule uvomorulin expression in humanbreast cancer cell lines: relationship to morphology and invasive capacities. CellGrowth Differ. 2, 365-372.

Steinberg, G. R. and Kemp, B. E. (2009). AMPK in health and disease. Physiol.Rev. 89, 1025-1078.

Sundararaman, A., Amirtham, U. and Rangarajan, A. (2016). Calcium-oxidantsignaling network regulates AMP-activated Protein Kinase (AMPK) activationupon matrix deprivation. J. Biol. Chem. 291, 14410-14429.

Tan, E.-J., Thuault, S., Caja, L., Carletti, T., Heldin, C.-H. and Moustakas, A.(2012). Regulation of transcription factor Twist expression by the DNAarchitectural protein high mobility group A2 during epithelial-to-mesenchymaltransition. J. Biol. Chem. 287, 7134-7145.

Thiery, J. P., Acloque, H., Huang, R. Y. J. and Nieto, M. A. (2009). Epithelial-mesenchymal transitions in development and disease. Cell 139, 871-890.

Vichalkovski, A., Gresko, E., Hess, D., Restuccia, D. F. and Hemmings, B. A.(2010). PKB/AKT phosphorylation of the transcription factor Twist-1 at Ser42inhibits p53 activity in response to DNA damage. Oncogene 29, 3554-3565.

Wang, X., Pan, X. and Song, J. (2010). AMP-activated protein kinase is required forinduction of apoptosis and epithelial-to-mesenchymal transition. Cell. Signal. 22,1790-1797.

Wang, Y., Liu, J., Ying, X., Lin, P. C. and Zhou, B. P. (2016). Twist-mediatedepithelial-mesenchymal transition promotes breast tumor cell invasion viainhibition of Hippo pathway. Sci. Rep. 6, 24606.

Witters, L. A., Nordlund, A.-C. and Marshall, L. (1991). Regulation of intracellularacetyl-CoA carboxylase by ATP depletors mimics the action of the 5′-AMP-activated protein kinase. Biochem. Biophys. Res. Commun. 181, 1486-1492.

Xiang, X., Saha, A. K., Wen, R., Ruderman, N. B. and Luo, Z. (2004). AMP-activated protein kinase activators can inhibit the growth of prostate cancer cellsby multiple mechanisms. Biochem. Biophys. Res. Commun. 321, 161-167.

Yang, J. and Weinberg, R. A. (2008). Epithelial-mesenchymal transition: at thecrossroads of development and tumor metastasis. Dev. Cell 14, 818-829.

Yang, J., Mani, S. A., Donaher, J. L., Ramaswamy, S., Itzykson, R. A., Come, C.,Savagner, P., Gitelman, I., Richardson, A. andWeinberg, R. A. (2004). Twist, amaster regulator of morphogenesis, plays an essential role in tumor metastasis.Cell 117, 927-939.

Yang, J., Mani, S. A. and Weinberg, R. A. (2006). Exploring a new twist on tumormetastasis. Cancer Res. 66, 4549-4552.

Yang, M.-H., Wu, M.-Z., Chiou, S.-H., Chen, P.-M., Chang, S.-Y., Liu, C.-J., Teng,S.-C. and Wu, K.-J. (2008). Direct regulation of TWIST by HIF-1alpha promotesmetastasis. Nat. Cell Biol. 10, 295-305.

Zhang, H., Singh, R. R., Talukder, A. H. and Kumar, R. (2006). Metastatic tumorantigen 3 is a direct corepressor of theWnt4 pathway.Genes Dev. 20, 2943-2948.

Zhou, G., Myers, R., Li, Y., Chen, Y., Shen, X., Fenyk-Melody, J., Wu, M., Ventre,J., Doebber, T., Fujii, N. et al. (2001). Role of AMP-activated protein kinase inmechanism of metformin action. J. Clin. Invest. 108, 1167-1174.

15

RESEARCH ARTICLE Journal of Cell Science (2018) 131, jcs208314. doi:10.1242/jcs.208314

Journal

ofCe

llScience