LKB1 and AMPK maintain epithelial cell polarity under energetic stress

LKB1 is a serine/threonine kinase that is mutated in most cases of Peutz-Jeghers syndrome, which is an autosomal dominant disorder in which patients develop benign hamartomas and a high frequency of malignant tumors (Alessi et al., 2006). Furthermore, LKB1 is also mutated in some sporadic cancers, such as 30% of lung adenocarcinomas, and the expression of the kinase is also down-regulated in a substantial proportion of breast cancers (Sanchez-Cespedes et al., 2002). In both cases, tumors associated with LKB1 mutations usually derive from epithelial tissue. LKB1 is a master kinase that can potentially activate several downstream kinases by phosphorylating a conserved threonine in their activation loops (Lizcano et al., 2004). Two of these kinases have been extensively characterized: PAR-1/microtubule affinity-regulating kinases (MARKs) and AMPK (AMP-activated protein kinase). PAR-1 regulates cell polarity in numerous cell types and organisms (Kemphues et al., 1988; Bohm et al., 1997; Shulman et al., 2000; Benton and St Johnston, 2003; Cohen et al., 2004). AMPK acts as a cellular energy sensor because it is activated by AMP, which accumulates when ATP levels are low (Kahn et al., 2005). AMPK then mediates the cellular response to energetic stress by activating energy-producing activities, while inhibiting energy-consuming ones, such as translation and proliferation. LKB1 regulates both cell polarity and cell growth and division in cell culture and in vivo (Kemphues et al., 1988; Tiainen et al., 1999; Martin and St Johnston, 2003; Baas et al., 2004; Narbonne and Roy, 2006). One hypothesis envisions LKB1 signaling mediated by PAR-1 regulating cell polarity, whereas LKB1 signaling through AMPK could control cell growth and proliferation. However, recent cell culture experiments suggest that AMPK also plays a role in the polarization of MDCK cells by promoting tight junction assembly (Zhang et al., 2006; Zheng and Cantley, 2007). We show that LKB1 and AMPK are required to maintain epithelial polarity and integrity under energy-limiting conditions in Drosophila melanogaster. Therefore, these results provide a potential mechanism to coordinate the regulation of cell polarity and proliferation with energy conditions within a multicellular animal.

AMPK contains three protein subunits, α, β, and γ, which form a heterotrimer. The α subunit (AMPKα) encodes a highly conserved serine/threonine kinase, and the other subunits are regulatory. From a D. melanogaster forward genetic screen for mutants affecting larval neuronal dendrite development (Medina et al., 2006), we identified several lethal mutations in AMPKα. The ethylmethanesulfonate mutants, ampkα¹ and ampkα², contain a single amino acid change (S211L, completely conserved) and a premature stop codon (Q295 STOP), respectively, whereas ampkα³ has a 16-bp deletion creating a stop codon (Y141 STOP; Fig. 1 A). All ampkα mutants, whether homozygous or in trans with a deletion covering the locus, displayed a completely penetrant and nearly identical phenotype, with greatly enlarged plasma membrane domains in dendrites, but not in axonal compartments (Fig. 1 C; unpublished data). In addition, ampkα¹ and ampkα³ could be rescued to viability with either a chromosomal duplication carrying a wild-type ampkα gene, a wild-type AMPKα transgene, or a transgene that is tagged with the red fluorescent protein mCherry (Fig. 1 D; see Materials and methods). The requirement for ampkα is cell autonomous because transgene expression within only neurons rescues the phenotype (Fig. 1 D). Therefore, these mutations represent the first knockouts of the single AMPKα catalytic subunit in the D. melanogaster genome and allow the genetic analysis of AMPK function in vivo.

Although ampkα mutants display a strong phenotype in larval neuronal dendrites, no phenotype was observed in early larval lkb1 mutants (unpublished data), probably because of the large maternal contribution of this protein. To explore the relationship between AMPKα and LKB1 function without the confounding issues caused by the differing maternal contributions of each protein, we chose to examine follicle cells of the D. melanogaster ovary. The follicle cells that surround the oocyte have a typical epithelial architecture with a highly polarized actin cytoskeleton in which the apical surface is marked by dense actin bundles in the apical microvilli, the lateral cortex is covered by a thin actin mesh, and the basal side contains a prominent network of parallel actin stress fibers. This polarized organization of actin typifies many epithelia, including the main mammalian tissue culture model for polarized epithelial cells, MDCK cells (Fievet et al., 2004). We did not observe any actin phenotypes in ampkα³ mutant follicle cells using standard detection procedures (Fig. 2 A). Because AMPK is maximally activated under low cellular energy levels, we also tested the influence of energy stress by strongly reducing the availability of sugar in the D. melanogaster culture medium. Under these conditions, ampkα³ mutant cells display a strong actin phenotype (Fig. 2 A). The density of basal stress fibers is strongly reduced, whereas the amount of apical F-actin increases. This phenotype is highly penetrant under these starvation conditions (98%; n = 49) and is also observed with the two other alleles of ampkα.

Because this phenotype reflects a disruption of the apical–basal polarity of the actin cytoskeleton, we examined other polarity markers within these cells. ampkα mutant clones induced in adult flies fed with high-sugar diets did not show any polarity phenotypes, which is consistent with the absence of an actin phenotype under these conditions (Fig. 2 A). Under energetic starvation conditions, however, ampkα mutant cells show a fully penetrant loss of polarity. Apical markers, such as atypical PKC (aPKC) and Crumbs (Crb) lose their cortical localization completely and appear to be down-regulated, as do the lateral markers Discs large (Dlg) and Coracle (Cora; Fig. 2 A). In contrast, Dystroglycan (Dg), which is normally enriched at the basal cortex, extends into the lateral domain, and occasionally even reaches the apical membrane (Fig. 2 A). This suggests that the phenotype represents an expansion of the basal domain at the expense of the lateral and apical domains.

Although most aspects of apical–basal polarity are completely disrupted in ampkα mutant clones under energetic stress, E-cadherin (ECad) is usually still enriched at the adherens junctions, suggesting that the altered polarity is not a secondary consequence of a loss of intercellular adhesion. The subapical localization of Bazooka (Baz) with cadherin is also maintained in most cases (Fig. 2 B). This indicates that Baz is not in a complex with aPKC in columnar follicle cells, but is instead associated with the adherens junctions, as has recently been described in the D. melanogaster embryo and in neuroepithelial cells of the Zebrafish neural tube (Harris and Peifer, 2005; Afonso and Henrique, 2006).

A considerable proportion of ampkα mutant clones show a more severe phenotype, in which the cells round up and lose their epithelial organization to form multiple layers of cells (Fig. 2 B). In these cases, Baz is now also absent from the cell cortex. Finally, larger mutant clones, particularly at the anterior or the posterior of the egg chamber, show a complete loss of epithelial organization and overproliferate to form small, tumorlike growths (Fig. 2 C).

As one proposed function for AMPK is to sense and maintain cellular ATP levels, the polarity phenotype observed under starvation conditions could be caused by low cellular ATP concentrations. To test this hypothesis, we examined cells that were mutant for tenured (tend). Tend encodes a mitochondrial cytochrome oxidase subunit; therefore, mutants have reduced intracellular ATP concentrations to levels sufficient to maintain cell survival and growth, but not cell division (Mandal et al., 2005). This cell cycle block is believed to require AMPK activation. In agreement with a role for Tend in cell cycle progression, we did not observe tend clones bigger than four to six cells under energetic starvation conditions (Fig. 2 D). In contrast to ampkα mutant cells, however, tend mutant cells showed no polarity defects, ruling out the possibility that the ampkα phenotype is a secondary effect of low ATP levels. We also tested the effect of specific nutrient starvation by feeding flies only glucose, but these conditions did not induce any polarity phenotypes in ampkα mutant cells (Fig. 2 E). Thus, AMPKα is specifically required to maintain epithelial polarity and growth control under conditions of energetic stress.

Because our results indicate that ampkα plays a role in epithelial polarity, we assessed whether the localization of the protein itself is polarized. We also examined LKB1 localization, as it is a potential regulator of AMPK. Transgenic wild-type fusion proteins for both AMPKα and LKB1 rescue lethal null mutants to viability, and should therefore mimic the localizations of the endogenous proteins. LKB1-GFP is mainly found at the apical and lateral cortex of the follicle cells, and is absent from the basal domain (Fig. 3 A). This basal exclusion is surprising, as cortical localization of LKB1 requires its membrane targeting by prenylation of a conserved CAAX motif (Martin and St Johnston, 2003). This suggests that the lipid composition of the basal domain is different from the rest of the plasma membrane and/or that LKB1 posttranslational modifications are asymmetrically controlled. In contrast, mCherry-AMPKα does not show any enrichment or asymmetric localization at the plasma membrane, and it is found distributed throughout the cytoplasm, but absent from the nucleus (Fig. 3 A). The localization of LKB1 suggests that AMPK could be activated specifically at the apical and lateral cortices of the cells. To test this hypothesis, we used an antibody against the LKB1 phosphorylation site of AMPK (phospho-T184). The immunostaining is reduced to background levels in both ampkα and lkb1 mutant clones. This confirms the specificity of the antibody and indicates that LKB1 is the principle AMPK kinase in these cells (Fig. 3 B). In wild-type cells, PhosphoT184-AMPK is found diffusely in the cytoplasm (Fig. 3 B). The effect of AMPK on apical–basal polarity is therefore not related to a polarized distribution of the kinase or its localized activation by LKB1.

Because LKB1 activates AMPK, we wondered if similar phenotypes could be observed in lkb1 mutant cells. lkb1 clones can lead to severe polarity defects in follicle cells in normally fed flies (Martin and St Johnston, 2003). However, these defects are observed only in large clones that are induced in the stem cells that give rise to the follicular epithelium, whereas small lkb1 mutant clones, which are induced after the formation of the epithelium, have no effect on follicle cell polarity or the organization of the actin cytoskeleton (n = 24; Fig. 4 A). This suggests that LKB1 is required for the establishment of epithelial polarity in well-fed flies, but not for its maintenance, as is the case for PAR-1 (Doerflinger et al., 2003). In contrast, under conditions of glucose starvation, small lkb1 clones that were induced after the formation of the follicular epithelium show a fully penetrant polarity phenotype (100%; n = 21). Under these conditions, we observed a loss of the polarized localization of Dlg, aPKC, Crb, and Cora (Fig. 4 A). However, Baz distribution is usually not affected by lkb1 loss of function (unpublished data). Dg extends laterally and occasionally localizes to the apical domain (Fig. 4 A). The actin cytoskeleton is also disturbed, with more F-actin apically and a decreased density of stress fibers on the basal side. Finally, large lkb1 clones lose their epithelial organization completely and overproliferate to form small neoplasms (Fig. 4, B and C). Thus, lkb1 mutant cells exhibit identical phenotypes to ampkα mutant cells under low-energy conditions.

Because lkb1 and ampkα mutant clones lead to very similar polarity defects and LKB1 phosphorylates AMPKα, we wondered if a constitutively active form of AMPKα could rescue the lkb1 phenotype. Therefore, we generated transgenic lines carrying a UAS-AMPKα construct, in which Threonine184 is replaced by an aspartate, which should mimic the activating phosphorylation of this site by LKB1 (Lizcano et al., 2004). The expression of the AMPKα-T184D transgene in lkb1 mutant clones fully rescues their starvation-dependent polarity and overproliferation phenotypes (n = 37), whereas the Gal4 driver alone has no effect (Fig. 5). Furthermore, AMPKα-T184D–expressing mutant clones also have a normal actin cytoskeleton (100%; n = 13; Fig. 5). Thus, the phosphomimetic version of AMPKα completely rescues the lkb1 mutant phenotype under conditions of energetic stress.

The recovery of null mutations in ampkα has allowed the first in vivo analysis of AMPK function in a multicellular organism, which has revealed an unexpected role for the kinase in the maintenance of epithelial polarity, but only under conditions of energetic stress. This implies that at least one of the pathways that normally maintain cell polarity cannot function when cellular energy levels are too low, and that AMPK activation compensates for this defect.

A surprising feature of the ampkα polarity phenotype is that it has opposite effects on the actin cytoskeleton and the cortical polarity cues. In mildly affected clones, basal actin is strongly reduced, with a corresponding increase in the amount of apical actin. In contrast, mutant clones show an expansion of the basal markers into the lateral and apical regions, as well as a loss of lateral and apical markers. Thus, the effects on actin may be independent of other polarity defects, suggesting that AMPK acts though different pathways to regulate actin and cortical polarity in opposite ways.

It is unclear how AMPK regulates the actin cytoskeleton, but it is possible that it acts on only one side of the cell and that the reciprocal changes on the other are caused by a change in the concentration of free G-actin or an actin nucleator, as has been shown for abl mutants during cellularization (Grevengoed et al., 2003). For example, loss of AMPK could increase actin polymerization apically, thereby depleting the pool of free actin that can polymerize basally. Alternatively, ampkα mutants may prevent the formation of basal actin stress fibers, and thus increase the concentration of free actin, which enhances apical actin polymerization.

The cortical polarity defects of ampkα mutant clones also suggest a reciprocal relationship between the basal and apical/lateral membrane domains because the basal domain, marked by Dg, is dramatically expanded, whereas the determinants for the lateral domain (Dlg) and the apical domain (aPKC and Crb) disappear from the cortex. This suggests that there is some form of mutual antagonism between the basal and lateral domains that maintains a sharp boundary between them, as has been described for apical and lateral domains through the inhibitory phosphorylation of Baz (PAR-3) by lateral PAR-1, and of PAR-1 by apical aPKC (Benton and St Johnston, 2003; Suzuki et al., 2004). If this model is correct, AMPK could be required to restrict the extent of the basal domain, with the expansion of this domain in ampkα mutants leading to the exclusion of lateral and apical markers. Indeed, the overexpression of Dg has been found to cause a similar loss of apical and lateral markers to that seen in ampkα clones (Deng et al., 2003). Alternatively, AMPK could be necessary to maintain the localization of the apical and lateral determinants, which in turn prevent the basal domain from extending into these regions.

Mutations in AMPK not only disrupt the polarity of the follicle cell epithelium, but also cause the cells to overproliferate, giving rise to a tumorous phenotype. One possible explanation for this phenotype is that it is caused by the mislocalization and down-regulation of Dlg. Dlg is a member of a class of tumor suppressors in D. melanogaster that also includes Lgl and Scribble, and follicle cell clones mutant for any of these genes overproliferate to form invasive tumors that are similar to those formed by ampkα and lkb1 clones under low-energy conditions (Bilder and Perrimon, 2000; Goode et al., 2005; Hariharan and Bilder, 2006). Furthermore, the tumor suppressor function of these proteins is probably conserved in humans because Scribble restricts proliferation by repressing the G1/S transition, and is a target of the papilloma virus E6 oncoprotein (Nagasaka et al., 2006; Takizawa et al., 2006). This may account for the observation that AMPK is required to trigger the G1/S checkpoint under conditions of energetic stress (Mandal et al., 2005). However, it has also been shown in mammals that AMPK activates TSC2 to repress the insulin–TOR pathway, and thus it functions as a tumor suppressor that inhibits cell growth and division (Inoki et al., 2003, 2005). Loss of this repression might provide an alternative explanation for the overgrowth of ampkα mutant clones.

Although the molecular pathways involved remain to be elucidated, our results demonstrate that ampkα mutant cells lose their polarity under low-energy conditions and overproliferate to give rise to tumorlike growths. The activation of AMPK depends on its phosphorylation by LKB1, and loss of LKB1 produces an identical tumorous phenotype. Thus, the novel functions of AMPK reported in this work may provide a basis for the tumor suppressor function of LKB1.

An ethylmethanesulfonate mutagenesis screen on the X chromosome was performed as previously described (Medina et al., 2006). Early second instar larvae were visually screened for dendritic defects using fluorescent microscopy. The ampkα mutants, lethal at late second instar stages, were mapped to ∼150 kb on the X chromosome using a molecularly defined deficiency (Df[1]Exel6227), an undefined deficiency (Df[1]AD11), and a duplication of the Y chromosome (Dp[1;Y]/Df[1]svr). Predicted coding regions for genes in the region were sequenced using PCR amplicons made from mutant genomic DNA, and one gene (AMPKα; CG3051; NM_057965) was discovered that had mutations in all three alleles.

The wild-type AMPKα transgene was cloned into the pUAST vector (Brand and Perrimon, 1993) as an EcoRI–BglII fragment of an EST, corresponding to an AMPKα-RA transcript (www.flybase.org). The mCherry-AMPKα fusion protein was made using a mCherry construct (provided by R. Tsien, University of California, San Diego, San Diego, CA) at the N terminus fused in-frame to AMPKα into the pUAST vector. The UAS-mCherry-AMPKα transgene rescues viability and fertility when expressed by Ubiquitin-Gal4 in either ampkα¹ or ampkα³ mutants. The phosphomimetic activated form of AMPKα (AMPKα T184D) was made by PCR-based, site-directed mutagenesis converting base C549 to G549. The transgenes were introduced into a w¹¹¹⁸ stock by P element–mediated transformation.

AMPKα alleles were recombined with FRT101 for mitotic recombination. Other mutant stocks used were FRT82B, lkb^4A4-2 and FRT82B, tend. UAS:Cherry-AMPKα and UAS:GFP-LKB1 were expressed in follicle cells using the Cy2-Gal4 driver. Flip-out experiments were performed by crossing UAS:Cherry-AMPKα and UAS:AMPKα-T184D to y, w, hs:Flp; tub-FRT-cc-FRT-Gal4, UAS:GFP and heat-shocking pupae. For rescue experiments, two independent stocks were established and crossed together: w; UAS:AMPKα-T184D/CyO; FRT82B, Ubi:nlsGFP and y, w, hs:flp; da:Gal4, FRT82B,lkb1^4A4-2/TM3,Sb.

Adult flies were placed in vials containing “normal” D. melanogaster food media (5% glucose, 5% yeast extract, 3.5% wheat flour, and agar 0.8%), energetic starvation medium (1% yeast extract, 3.5% wheat flour, and agar 0.8%), or specific nutrient-starvation medium (5% glucose and agar 0.8%). Clones were induced by heat-shocking adult females at 37°C for 2 h on two consecutive days. Females were dissected 2 d after the last heat shock.

Immunofluorescence on ovaries was performed using standard procedures. Primary antibodies were used as follows: rat anti-DECad (1:1,000; Oda et al., 1994); mouse anti-Crb (cq4; 1:50; Developmental Studies Hybridoma Bank); Guinea pig anti-Cora (1:2,000; Fehon et al., 1994); rabbit anti-aPKC (1:500; Sigma-Aldrich); rat anti-Baz (1:500; Wodarz et al., 1999), mouse anti-Dlg (1:50; Developmental Studies Hybridoma Bank); rabbit anti-Dg (1:1,000; Deng et al., 2003); and rabbit anti–phospoT385-AMPK (1:100; Cell Signaling Technology). Actin staining was performed with rhodamine-conjugated phalloidin (Invitrogen). Second instar larvae were dissected in 4% paraformaldehyde, as previously described (Medina et al., 2006). Secondary antibodies coupled with Cy5 (anti–rabbit and anti–guinea pig) or Texas red (anti–mouse and anti–rabbit; Jackson Immuno-Research Laboratories; 1:500) were used. Images of follicle cells were collected on a confocal microscope (Radiance 2000; Bio-Rad Laboratories) with a 40×/1.3 NA objective (Plan Fluor; Nikon) using LaserSharp software. Live images of dendrite morphology were acquired using a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) by using the 488-nm argon line to excite GFP. Larvae were covered in a glycerol solution at 22°C and gently covered with a coverslip (22 × 50 mm; Fisher Scientific) to restrict movement, but not cause bursting of the body wall. Images were taken using a Pan-Neofluar 40×/1.3 NA oil immersion lens with a 2-μm optical slice and LSM Imaging software (Carl Zeiss MicroImaging, Inc.). Images were resized and cropped with Photoshop (Adobe), and imported into Illustrator (Adobe) for labels and arrangement.

We are grateful to those who generously sent stocks and reagents. We thank Pierre Leopold for helpful advice on starvation conditions.

V. Mirouse was supported by the European Molecular Biology Organization, and D. St Johnston was supported by the Wellcome Trust. Grants from the National Institute of Mental Health (MH073155) and the Whitehall Foundation to J.E. Brenman and the National Science Foundation to L. Swick also supported this work.