The PDZ domains of zonula occludens-1 induce an epithelial to mesenchymal transition of Madin-Darby canine kidney I cells. Evidence for a role of beta-catenin/Tcf/Lef signaling.

The integrity of cell-cell contacts such as adherens junctions (AJ) and tight junctions (TJ) is essential for the function of epithelia. During carcinogenesis, the increased motility and invasiveness of tumor cells reflect the loss of characteristic epithelial features, including cell adhesion. While beta-catenin, a component of AJ, plays a well characterized dual role in cell adhesion and signal transduction leading to epithelial cell transformation, little is known about possible roles of tight junction components in signaling processes. Here we show that mutants of the TJ protein zonula occludens protein-1 (ZO-1), which encode the PDZ domains (ZO-1 PDZ) but no longer localize at the plasma membrane, induce a dramatic epithelial to mesenchymal transition (EMT) of Madin-Darby canine kidney I (MDCKI) cells. The observed EMT of these MDCK-PDZ cells is characterized by a repression of epithelial marker genes, a restricted differentiation potential and a significantly induced tumorigenicity. Intriguingly, the beta-catenin signaling pathway is activated in the cells expressing the ZO-1 PDZ protein. Ectopic expression of the adenomatous polyposis coli tumor suppressor gene, known to down-regulate activated beta-catenin signaling, reverts the transformed fibroblastoid phenotype of MDCK-PDZ cells. Thus, cytoplasmic localization of the ZO-1 PDZ domains induces an EMT in MDCKI cells, most likely by modulating beta-catenin signaling.

In addition to mediating cell-cell adhesion, TJ regulate the paracellular diffusion across epithelial monolayers and the maintenance of the asymmetric distribution of proteins and lipids to the apical and basolateral plasma membrane domains of epithelial cells (1)(2)(3). The tight junction protein zonula occludens protein 1 (ZO-1) is part of a multi-protein complex and binds directly to the integral TJ proteins occludin and to members of the claudin family (4), thereby linking the TJ to the cytoskeleton via a direct or indirect interaction with actin (5). ZO-1 belongs to the membrane-associated guanylate kinase (MAGUK) protein family and contains three PDZ domains (6), an Src homology 3 (SH3) domain, a guanylate kinase (GUK) homology domain, and a proline rich C-terminal region (see Fig. 1). Since occludin lacks PDZ-binding motifs, binding between occludin and ZO-1 probably does not involve the PDZ domains. The function of the GUK domain, which lacks kinase activity in the MAGUK proteins analyzed so far, is not known but has been suggested to be important for binding of ZO-1 to occludin (5). ZO-2 and ZO-3, two additional members of the MAGUK protein family present in TJ, show extensive homology to each other and to ZO-1. ZO-3 interacts with ZO-1 and the cytoplasmic C-terminal tail of occludin, but does not bind ZO-2 (7). ZO-2 binds directly to ZO-1 and occludin. Actin cosedimentation studies showed that ZO-2, ZO-3, and occludin all interact directly with F-actin in vitro and colocalize with actin aggregates at cell boarders in cytocholasin D-treated MDCK cells. The suggested model at the moment is that two independent complexes comprising ZO-1-ZO-2 and ZO-1-ZO-3 exist (rather than a three-member complex, ZO-1-ZO-2-ZO- 3), and that these complexes link the tight junction to the actin cytoskeleton (8). Several other proteins have been described to interact with ZO-1, but the domains involved in binding or the physiological relevance of the interactions are, in most cases, unknown.
Carcinogenesis is a multistep process well characterized in human colon cancer. Mutations in the adenomatous polyposis coli (APC) gene are thought to initiate the process, leading to aberrant crypt foci that develop into areas of benign epithelial hyperplasia or dysplasia and adenomas. Progression of these areas to carcinomas in situ and malignant tumors depends on further changes in the transformed cells such as mutations in p21 Ras and p53 and the gradual loss of a number of characteristic features of differentiated epithelial cells. This process, also known as epithelial to mesenchymal transition (EMT), includes the disruption of apical-basolateral polarity, the disassembly of AJ and TJ, and the ability of the cells to degrade the basement membrane, to migrate, and to form metastases at distant sites.
A reduced intercellular adhesion is a requisite for the higher motility and invasiveness of tumor cells and the expression or integrity of several components of AJ (i.e. ␣-catenin, ␤-catenin, ␥-catenin/plakoglobin, E-cadherin) is altered or lost in different types of carcinoma (9). ␤-Catenin plays a dual role as a structural component of AJ (10) and as a signaling molecule in the Wnt signaling pathway (11). In the absence of Wnt glycoproteins, the Ser/Thr-specific glycogen synthase kinase 3␤ phosphorylates ␤-catenin, APC, and axin/conductin (12,13), which are present as a multi-protein complex in the cytosol. Phosphorylated ␤-catenin is rapidly ubiquitinated and degraded by the proteosomal pathway (14). Binding of Wnt glycoproteins to the Frizzled family of receptors results in the inactivation of glycogen synthase kinase 3␤ and thereby to an enhanced stability of ␤-catenin. Stabilized ␤-catenin can translocate into the nucleus where, in association with members of the Tcf/Lef transcription factor family, it regulates gene expression (15,16), probably by recruiting the basal transcription machinery to promoter regions of Wnt target genes such as cyclin D1 (17,18). Oncogenic transformation of mammalian cells is closely linked to the signaling function of ␤-catenin (19). Intestinal cells carrying mutations in APC that activate the ␤-catenin/Tcf/Lef signaling pathway develop into adenomas and adenocarcinomas (20). In addition, human colorectal neoplasms expressing a wild-type APC often show mutations in ␤-catenin that activate its signaling capacity (9). Furthermore, mice expressing a dominant allele of the ␤-catenin gene develop adenomatous intestinal polyps and nascent microadenomas, providing further evidence that activated ␤-catenin signaling contributes to cancer development (21).
A few observations suggested that TJ components, in addition to their structural role, may also be involved in signaling events. ZO-1 is related to the Drosophila discs-large tumor suppressor (Dlg-A), a component of septate junctions in Drosophila implicated in signaling during mitosis. Dlg proteins with mutations in the PDZ and SH3 domains cause neoplastic overgrowth of larval imaginal disc epithelial cells (22). The Drosophila orthologue of ZO-1, tamou, has been implicated in regulating the expression of extramacrochaetae (23), the fly orthologue of the inhibitor of differentiation protein. ZO-1 itself has been found in the nucleus of migrating epithelial cells at the edge of wounded monolayers or in epithelial cells induced to migrate by HGF (24).
To explore possible additional functions of ZO-1 besides its role as a structural component of TJ, we expressed progressive C-terminal deletion mutants of the protein in epithelial MD-CKI cells. Surprisingly, mutants encoding only the N terminus including the PDZ domains no longer localized at the plasma membrane and induced a dramatic loss of the epithelial phenotype of MDCKI cells. This EMT included changes in the differentiation potential and tumorigenicity of the cells, together with a repression of epithelial (e.g. E-cadherin) and an induction of mesenchymal (e.g. fibronectin) marker genes. Interestingly, ␤-catenin/Tcf/Lef signaling was activated in MCDK-PDZ cells, indicating an involvement of ␤-catenin/Tcf/ Lef signaling in the induction of the observed EMT. Thus, our results show that the cytosolic localization of the PDZ domains of ZO-1 leads to the transformation of MDCKI cells, most likely through a direct or indirect modulation of the ␤-catenin/Tcf/Lef signaling pathway.

MATERIALS AND METHODS
Plasmids-Human ZO-1 cDNA (GenBank accession no. L14837) was kindly provided by J. Anderson. The following deletion mutants (see Fig. 1) were created by the PCR technique using Pwo polymerase (Roche): ZO-1-PDZ (amino acids 1-568), ZO-1-PS (amino acids 1-759), ZO-1-PSG (amino acids 1-798), and wild-type ZO-1. A FLAG epitope tag (5Ј-gattacaaagacgatgacgataaa-3Ј) was introduced into each 3Ј oligonucleotide to generate a ZO-1 fusion protein with the FLAG tag at the C terminus. The different PCR products were cloned into the pCRblunt vector (Invitrogen), cut out with EcoRV, and cloned into the eukaryotic retroviral expression vector pLNCX (CLONTECH), cut with HpaI. Fulllength human APC cDNA (GenBank accession no. M74088) was kindly provided by B. Vogelstein, and a myc tag (5Ј-gaacaaaaactcatctcagaagaggatctgaat-3Ј) was introduced at the 3Ј end by PCR. The neomycin resistance gene in the retroviral pLNCX vector was replaced by a hygromycin-thymidine kinase fusion cDNA (kindly provided by C. Karreman) to yield the pLHygTkCX vector, into which the myc-tagged APC cDNA was cloned. The plasmid coding for the GST-E-cadherin cytoplasmic fusion protein was kindly provided by A. Ullrich and described elsewhere in detail (25).
Cell Culture-MDCKI cells were cultured in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal calf serum (FCS) (Life Technologies, Inc.), 100 units/ml penicillin and streptomycin, and 2 mM glutamine. MDCKI cells were plated at a density of 1 ϫ 10 4 cells/cm 2 on 10-cm plates (Nunc) the day before transfection using the calcium phosphate technique as described (26). Briefly, 16 g of plasmid DNA (Nucleobond purified) were mixed with 40 l of 2.5 M CaCl 2 to a final volume of 400 l. An equal volume of 2ϫ BBS (50 mM BES (Sigma), 280 mM NaCl, 1.5 mM Na 2 HPO 4 , pH 6.5) was added, and after 10 min at room temperature the mixture was added to the cells. The cells were incubated overnight at 3%CO 2 , washed twice with PBS, and incubated at 5% CO 2 . For stable transfections, 750 g/ml G418 or 350 g/ml hygromycin, respectively (Life Technologies, Inc.), were added 24 h after transfection. Single cell clones were established using either limited dilution or cloning rings.
Transepithelial electrical resistance (TER) was measured 3 days after 5 ϫ 10 4 cells/cm 2 had been plated on Transwell filters (Costar). TER was determined by applying an AC square wave current of 620 mA at 12.5 Hz across a cell monolayer plated on a 6.5-mm diameter Transwell filter. The voltage deflection was measured with a pair of Ag/ AgCl voltage sensors (EVOM, World Precision Instruments). TER values were calculated by subtracting the blank values from the filter and the medium, and were normalized to the area of the monolayer (filter).
Indirect Immunofluorescence-5 ϫ 10 4 cells/cm 2 were grown on coverslips to confluence, washed twice with PBS, and fixed with 3% freshly prepared paraformaldehyde for 25 min at room temperature. The cells were washed and permeabilized with 0.5% Triton X-100 10 min at room temperature. Unspecific binding was blocked with 10% goat serum in PBS for 1 h at room temperature, and primary antibodies (M2 monoclonal ␣-FLAG antibody; Eastman Kodak Co.) or fluorescein isothiocyanate-labeled phalloidin (Sigma; diluted 1:100 in 10% goat serum) were added for 2 h at room temperature. After washing the cells several times with PBS, bound primary antibodies were detected with fluorochrome-coupled isotype-specific secondary antibodies (Alexa). Coverslips were mounted (16.7% Mowiol, 33% glycerol in 120 mM Tris-HCl, pH 8.5) and viewed with a conventional fluorescence microscope (Leica).
Cell Culture in Collagen Gels-Cells were trypsinized, and 2 ϫ 10 4 cells were added to 800 g of collagen type I (Promocell) in 10ϫ DMEM to yield a final volume of 2 ml in 1ϫ DMEM, 10% FCS and plated in 24-well plates (Costar). The plates were incubated 10 min at room temperature and 45 min at 37°C before 1 ml of DMEM including 10% FCS was added. Where indicated, HGF (40 ng/ml, kindly provided by W. Birchmeier) was added to the medium and incubation was continued until a ductlike morphology became visible. Cells were fixed with 3% freshly prepared paraformaldehyde, permeabilized with 0.5% Triton X-100, and stained with 0.2% Carmin-Hemalaun in H 2 O overnight.
Tumor Formation in Nude Mice-Cells were resuspended at a cell density of 2 ϫ 10 6 cells in 50 l of DMEM and injected subcutaneously into the flank region of Swiss nude mice (IGR Villejuif, Paris, France) using five animals per cell line. Tumor formation was monitored by measuring the width (W) and length (L) of the tumors with W Ͻ L. The tumor volume was calculated according to the formula (W 2 ϫ L ϫ /6).
Reporter Gene Transcription Assays-Cells were plated at a density of 1 ϫ 10 4 cells/cm 2 in six-well plates (Costar) the day before transfec-tion. Transfection was performed using the calcium-phosphate technique as described above with 4 g of luciferase reporter constructs containing either multimerized wild-type (TOP-FLASH) or mutant (FOP-FLASH) Tcf/Lef binding sites (kindly provided by H. Clevers). As a control for transfection efficiency, 1 g of ␤-galactosidase construct under the control of the simian virus 40 promoter was included in each transfection. Cells were washed 24 h after transfection, and extracts were prepared in 400 l of reporter lysis buffer (Promega). Luciferase and ␤-galactosidase activity were assayed according to the manufacturer's protocol using the luciferase assay kit from Promega. The relative luciferase units corresponding to the enzymatic luciferase activities obtained for the TOP or FOP reporter gene transcription, respectively, were normalized to the relative ␤-galactosidase activity. To allow easier comparison of the transcriptional activities, the background transcriptional activity represented by the FOP values was subtracted from the TOP values. Each transfection was done in triplicate, and the luciferase and ␤-galactosidase activities of each sample were measured in triplicate. The assay was performed in three independent experiments.
Cellular Extracts-To analyze the expression of proteins or to determine the free pool of ␤-catenin, cells were washed with ice-cold PBS and scraped from the plate in ice-cold lysis buffer (20 mM imidazole-HCl, pH 6.8, 100 mM KCl, 2 mM MgCl 2 20 mM EDTA, 300 mM sucrose, 0.1 mM sodium orthovanadate, 1 mM NaF, 0.2% Triton X-100, and freshly added protease inhibitors (25 g/ml aprotinin, 25 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride; Sigma). The lysates were spun at 15,000 rpm in a table-top centrifuge and the supernatant (Triton X-100-soluble fraction) was snap-frozen in liquid nitrogen and stored at Ϫ80°C. The pellet (Triton X-100-insoluble fraction) was re-extracted in RIPA buffer (10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.1% SDS, 1% Triton X-100, 10% glycerol) for 30 min at 4°C. The extract was spun at 15,000 rpm, and the supernatant was also snap-frozen in liquid nitrogen. Protein concentration was determined according to Bradford.
To analyze the distribution of the different ZO-1 proteins to membrane and cytosol fractions, cells were washed twice with ice-cold PBS and scraped into homogenization buffer (25 mM Tris-HCl, pH 7.4, 2 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 10 mM ␤-mercaptoethanol, 10% glycerol, 1ϫ protease inhibitor mixture (Roche Molecular Biochemicals)). After incubating for 10 min on ice, the cells were homogenized in a Dounce homogenizer and insoluble material was pelleted in a low speed centrifugation (5 min, 500 ϫ g). The supernatant was centrifuged at 100,000 ϫ g for 90 min at 4°C to obtain a membrane pellet and a cytosol fraction. The membrane pellet was extracted with homogenization buffer containing 1% Triton X-100 for 30 min at 4°C, and insoluble material was pelleted for 30 min at 4°C and 100,000 ϫ g. The cytosol and the Triton X-100-soluble membrane fraction were analyzed by Western blot to detect the different ZO-1 proteins.
Affinity Precipitation-Equal amounts of cell lysates were precleared with glutathione-Sepharose for 30 min at 4°C and incubated with 5 g of purified GST-E-cadherin cytoplasmic fusion protein or a 3-fold molar excess of GST immobilized on glutathione-Sepharose (Amersham Pharmacia Biotech). The resulting complexes were washed three times with 20 mM HEPES, pH 7.5, 150 mM NaCl, 10 mM pyrophosphate, 10 mM NaF, 0.2 mM ammonium molybdate, 10% glycerol, 0.1% Triton X-100, 2 mM sodium orthovanadate; bound complexes were separated by SDS-PAGE and transferred onto nitrocellulose membranes, and ␤-catenin was visualized by immunodetection.

Generation of MDCKI Cells Expressing Epitope-tagged ZO-1
Constructs-ZO-1 wild-type and deletion mutants encoding the PDZ domains (PDZ), the PDZ and the SH3 domains (PS) or the PDZ, SH3 and GUK domains (PSG) were constructed, each carrying a C-terminal FLAG epitope tag (Fig. 1). The different cDNAs were subcloned into the pLNCX expression vector and transfected into MDCKI epithelial cells. These cells possess characteristics similar to the principal cells of the collecting systems in the kidney and polarize under appropriate conditions. Compared with the MDCKII cell line, MDCKI cells show a significantly increased TER (28,29), 2 indicating the formation of well established TJ. After transfection, G418-resistant cells were selected, pooled, and used for further analysis. Immunofluorescence experiments using M2 anti-Flag antibodies showed that the cells in a given pooled population of G418resistant cells homogeneously expressed the different ZO-1 proteins (data not shown, see below).
ZO-1 Mutants That No Longer Localize at the Plasma Membrane Induce an EMT-To characterize the subcellular distribution of the different ZO-1 proteins, cells were grown to confluence and the localization of the tagged proteins was detected by indirect immunofluorescence (Fig. 2a). In MDCK-ZO-1 cells, the transfected FLAG-tagged wild-type ZO-1 localized at the plasma membrane to regions of cell-cell contact (Fig 2a, panel  A), probably the TJ since these cells displayed an increased TER (see below). Similarly, in MDCKI cells transfected with the PSG construct, the expressed protein was present at regions of cell-cell contact (panel C). In contrast, however, deletion mutants lacking the GUK domain no longer localized at the plasma membrane in transfected MDCK-PS cells (data not shown) or MDCK-PDZ cells (panel B).
The apparent cytosolic localization of the ZO-1 PDZ construct observed by immunofluorescence (Fig. 2a, panel B) was confirmed biochemically. MDCK-PDZ and MDCK-PSG cells were homogenized in the absence of detergent; cytosol and membrane fractions were prepared as described under "Materials and Methods" and analyzed by Western blot. As shown in Fig. 2b, the transfected ZO-1 PSG protein was almost exclu-2 M. Reichert and W. Hunziker, unpublished observations. sively detected in the membrane fraction. In contrast, the ZO-1 PDZ protein was mostly recovered in the cytosolic fraction. This experiment thus confirms the immunofluorescence data and shows that the ZO-1 PDZ protein no longer localizes at the plasma membrane. These results are consistent with previous data (5) indicating that the GUK domain may be critical for the localization of ZO-1 at the plasma membrane, presumably by binding to occludin.
Neither MDCKI cells expressing the wild-type ZO-1 nor cells transfected with the empty pLNCX vector (MDCK-pLNCX) showed apparent changes in morphology when compared with the parental cell line, and the morphology of MDCK-PSG cells was also not altered. Surprisingly, however, the expression of proteins encoding the PDZ domains but lacking the GUK domain led to a dramatic change in the morphology of MDCK-PDZ and MDCK-PS cells (panel B and data not shown). The cells lost their epithelial phenotype and instead displayed a fibroblast-like morphology with long lamellipodia. Alterations in the morphology of MDCK-PDZ cells correlated with changes in the organization of the cytoskeleton. The typical cortical actin ring of polarized epithelial cells was observed in control MDCK-pLNCX cells (Fig. 2a, panel F). In contrast, MDCK-PDZ cells displayed actin stress fibers normally absent from polar- were plated on Transwell filters and incubated for 3 days. TER was measured, and values were calculated by subtracting the blank values from a filter with medium alone and normalized to the area of the monolayer (filter). Each experiment was done in triplicate. b, MDCK-pLNCX (panels A and C) or MDCK-PDZ (panels B and D) cells were plated in 24 well plates in 2 ml of complete medium containing 800 g of collagen I. Where indicated (panels C and D), HGF (40 ng/ml) was added to the medium and incubation was continued until differentiation morphology became visible. Cells were fixed with 3% paraformaldehyde, permeabilized with 0.5% Triton X-100, and stained with Carmin-Hemalaun. The formation of cysts and budlike structures (arrows) as well as ducts (arrowheads) was only observed in control cells.

TER above the background of empty filters was observed for MDCK-PDZ cells, indicating that these cells no longer formed monolayers with intact TJ.
Further characterization of the growth characteristics of MDCK-PDZ cells revealed that these cells had lost the inhibition of proliferation upon cell-cell contact, resulting in cell multilayering (data not shown).
When grown organotypically in collagen type I gels, MDCK cells form cysts that, in the presence of HGF, differentiate into branching ducts (30). While control MDCK-pLNCX cells still formed cysts in collagen type I gels (Fig. 3b, panel A) and, in the presence of HGF, ducts (panel C), MDCK-PDZ cells did not establish cell aggregates but grew as single cells (panel B). In the presence of HGF (panel D), MDCK-PDZ cells did not change their morphology, indicating that these cells had lost their potential to differentiate.
MDCK-PDZ Cells Are Tumorigenic in Nude Mice-To determine whether MDCK-PDZ cells were tumorigenic, an equal number of control cells (MCDK, MDCK-pLNCX, and MDCK-ZO-1) or MDCK-PDZ cells were injected subcutaneously into the hind flank of nude mice. Tumor formation at the site of injection was observed over a 10-week period after which the final tumor size was determined. Neither injection of MDCK-pLNCX cells (Fig. 4) nor parental MDCKI or MDCK-ZO-1 cells (data not shown) resulted in tumor growth. In contrast, however, injected MDCK-PDZ cells led to the formation of a significant tumor mass at the site of injection. Despite the variation in tumor volume from one animal to the other, which may reflect the injection of pooled polyclonal cell populations, tumors derived from MDCK-PDZ cells were in all but one case at least more than 4 times the volume of the control cell mass.

The Expression of Epithelial Markers Is Repressed in MDCK-PDZ Cells, Whereas the Expression of Fibroblastoid Markers Is
Induced-Next, we determined if the loss of the epithelial phenotype induced by the PDZ domains correlated with changes in protein levels or gene expression. Depending on the availability of antibodies or sequence information for the canine proteins or cDNAs, respectively, we analyzed protein and/or mRNA levels in control and MDCK-PDZ cells. To distinguish between soluble or membrane-associated proteins and proteins tightly bound to the cytoskeleton or present in the nucleus, cells were first extracted with 0.2% Triton X-100 (Triton X-100-soluble fraction) and the TX-100-insoluble fraction was then resolubilized in RIPA buffer (Triton X-100-insoluble fraction). The lysates were subjected to Western blot analysis.
Changes in mRNA levels were analyzed by semi-quantitative RT-PCR.
As shown in Fig. 5a, the amount of typical epithelial marker proteins like cytokeratins or E-cadherin was significantly reduced in MDCK-PDZ cells. Semi-quantitative RT-PCR furthermore showed decreased mRNA levels for E-cadherin, occludin, and endogenous ZO-1 in MDCK-PDZ cells (Fig. 5b), indicating that the loss of at least some epithelial marker proteins reflected changes in gene expression and/or mRNA stability. Interestingly, while the endogenous ZO-1 (␣ϩ) transcript was reduced in MDCK-PDZ cells, a new transcript, identified by sequencing the PCR product as the alternatively spliced ZO-1 (␣Ϫ), was present. The ZO-1 (␣Ϫ) splice form is often found in cells that show an increased plasticity in cell-cell contacts and junction stability (31,32).
In addition, the amounts of ␣, ␤and ␥-catenin/plakoglobin protein were reduced in MDCK-PDZ cells (Fig. 5a). In particular, the amount of the different catenins present in the Triton X-100-soluble fraction was reduced. Catenin mRNA levels were either not significantly altered or slightly elevated in MDCK-PDZ cells (data not shown), indicating that the lower protein levels in MDCK-PDZ cells were not due to reduced mRNA levels. Additionally, the amount of APC protein in the different fractions was reduced in MDCK-PDZ cells and a redistribution of APC from a detergent-soluble into a more detergent-resistant fraction was observed (Fig. 5a). Immunofluorescence analysis revealed that APC was present in clusters at the end of lamellipodia (data not shown), possibly correlating with the localization of APC observed during migration processes (33).

FIG. 5. Expression of epithelial and mesenchymal marker proteins in MDCK-PDZ cells.
a, cell extracts containing Triton X-100soluble and -insoluble material were prepared as described under "Materials and Methods" and analyzed by Western blot using the indicated antibodies. b, 0.5 g of total RNA was analyzed by semi-quantitative RT-PCR using specific primer pairs for the indicated genes and the amplified PCR fragments were separated on 0.5-2% agarose gels. c, MDCK-pLNCX and MDCK-PSG cells were lysed in RIPA buffer and the lysates analyzed by Western blot using antibodies against E-cadherin.
The endogenous ZO-1 protein is mainly detected in the Tritoninsoluble fraction in MDCK-pLNCX and MDCK ZO-1 cells, and immunofluorescence analysis revealed that ZO-1 is localized at the plasma membrane (data not shown). Consistent with the RT-PCR analysis, the amount of endogenous ZO-1 in MDCK-PDZ cells is strongly reduced (Fig. 5a). Concomitant with the repression of epithelial markers, mRNA levels of the mesenchymal marker genes fibronectin and vimentin were increased in MDCK-PDZ cells (Fig. 5b).
Importantly, and in contrast to MDCK-PDZ cells, E-cadherin was still expressed in MDCK-ZO-1 and MDCK-PSG cells (Fig.  5, a and c), which retained the polarized epithelial phenotype, excluding that the expression of exogenous ZO-1 per se alters the expression of the analyzed marker genes. These results thus further confirm the inverse correlation between plasma membrane association of the PDZ domains and the induction of an EMT.
In conclusion, the profound changes in MDCK-PDZ cells were paralleled by the repression of epithelial and the induction of mesenchymal marker genes, respectively.
␤-Catenin/Tcf/Lef Signaling Is Constitutively Activated in MDCK-PDZ Cells-It has been suggested that ␤-catenin/Tcf/ Lef signaling may negatively regulate E-cadherin and cytokeratin expression due to the presence of Tcf/Lef binding sites within their promoter regions (34,35). This prompted us to investigate whether this signaling pathway was activated in MDCK-PDZ cells. Free cytosolic ␤-catenin can act as a regulator of transcription after translocation to the nucleus and interaction with members of the Tcf/Lef family of transcription factors (36,37). To determine whether the free pool of ␤-catenin was elevated in MDCK-PDZ cells, the amount of ␤-catenin in cell lysates from control and MDCK-PDZ cells that is able to bind to a GST fusion protein comprising the cytoplasmic tail of E-cadherin was determined (25,38). As shown in Fig. 6a, slightly less free momomeric ␤-catenin from MDCK-PDZ cell lysates bound to the immobilized E-cadherin tail fusion protein as compared with control cells, indicating that the cytosolic free pool of ␤-catenin was not increased in MDCK-PDZ cells. This observation is consistent with the lower amounts of ␤-catenin found in the Triton-soluble pool of cell lysates of MDCK-PDZ cells (Fig. 5a).
Since recent reports have shown that an increased transcriptional activity of ␤-catenin/Tcf/Lef does not necessarily correlate with the detection of an increased free pool of ␤-catenin (39, 40), we directly analyzed ␤-catenin/Tcf/Lef signaling in MDCK-PDZ cells using the TOP-FOP Tcf/Lef luciferase reporter constructs, a well established assay to measure transcriptional activation due to activated ␤-catenin signaling (20). Briefly, MDCK-pLNCX, MDCK-ZO-1, MDCK-PSG, and MDCK-PDZ cells were transfected with plasmids encoding multimerized wild-type (TOP) or mutant (FOP) LEF binding sites followed by a luciferase reporter gene (36). As a control for transfection efficiency, an SV40-driven ␤-galactosidase cDNA was cotransfected. Relative luciferase activities were calculated as described under "Materials and Methods." As shown in Fig. 6b, the relative transcriptional activity of the ␤-catenin/Lef complex was 6 times higher in MDCK-PDZ cells as compared with vector transfected cells. Thus, despite the lack of an increased free pool of ␤-catenin, ␤-catenin/Tcf/Lef signaling is constitutively activated in MDCK-PDZ cells.
Ectopic Expression of APC Reverts the ZO-1 PDZ-induced EMT and Abolishes ␤-Catenin/Tcf/Lef Signaling-When the loss of the epithelial phenotype induced by the expression of mislocalized ZO-1 PDZ domains involves an activated ␤-catenin/Tcf/Lef signaling, transfection of MDCK-PDZ cells with APC, a known negative regulator of ␤-catenin/Tcf/Lef signaling (41,42), may affect the phenotype of MDCK-PDZ cells. We therefore stably transfected MDCK-PDZ cells with a cDNA encoding the human APC cDNA or, as a control, with the empty vector, pLHygTKCX.
While cells transfected with the empty vector retained their fibroblastoid morphology (Fig. 7a, panel A), cells transfected with the APC cDNA reverted to an epithelial phenotype (panels B and C). Several individual clones derived from cells transfected with APC were isolated and analyzed by RT-PCR to confirm the expression of the human APC cDNA. The phenotypic reversion to the epithelial morphology was found to correlate with the expression level of APC (Fig. 7b). Western blot analysis confirmed that the reverted phenotype did not result from a loss of the expression of the ZO-1 PDZ protein (Fig. 7c). Furthermore, the phenotypic conversion of MDCK-PDZ cells ectopically expressing APC was paralleled by the re-expression of E-cadherin (Fig. 7c) and the repression of the transcriptional activity of the ␤-catenin/Lef complex (Fig. 7d). FIG. 6. Free pool of ␤-catenin and activation of ␤-catenin/Tcf/ Lef signaling in MDCK-PDZ cells. a, a GST fusion protein comprising the cytosolic tail of E-cadherin was purified from Escherichia coli (25). Equal amounts of precleared cellular lysates were incubated with 5 g of purified GST-E-cadherin cytoplasmic fusion protein or as a control a 3-fold molar excess of GST immobilized on glutathione-Sepharose. The resulting complexes were washed, and bound ␤-catenin was visualized by Western blotting. The blots were also probed with an anti-E-cadherin antibody to confirm that similar amounts of immobilized GST-E-cadherin tail fusion protein were used in the precipitations. Similar results were obtained in three independent experiments. b, MDCK-pLNCX, MDCK-ZO-1, MDCK-PSG, or MDCK-PDZ cells were transiently cotransfected with TOP or FOP reporter constructs and a plasmid encoding the ␤-galactosidase cDNA under the control of the simian virus 40 promoter as described under "Materials and Methods." Extracts were prepared and assayed 24 h after transfection for luciferase and ␤-galactosidase activity. The luciferase activities obtained from the TOP or FOP reporter constructs were normalized to ␤-galactosidase activity, and FOP values were subtracted from the TOP values. Each transfection was done in triplicate, and the luciferase and ␤-galactosidase activities of each sample were measured in triplicate. Three independent transfections were performed, and a representative experiment is shown.
Thus, the ectopic expression of APC reverted the EMT of MDCK-PDZ cells and led to a repression of the ␤-catenin/Lef transcriptional activity, consistent with a direct or indirect effect of the ZO-1-PDZ protein on the ␤-catenin/Tcf/Lef signaling pathway. DISCUSSION We show that expression of ZO-1 mutants encoding the N terminus comprising the PDZ domains that localize to the cytosol induces a dramatic EMT in MDCKI cells. Localization of ZO-1 at the plasma membrane required the presence of the GUK domain, and no EMT was induced by ZO-1 mutants that were properly targeted to the plasma membrane. Uncloned polyclonal cell populations were used for the experiments to exclude effects due to clonal selection. However, similar phenotypes were observed for cell clones obtained by limited dilution (data not shown). Characterization of ZO-1 PDZ protein expression levels in individual clones indicated that the observed effect was not due to the expression of large amounts of the PDZ domains, since similar effects where observed in clones expressing the PDZ domains at levels barely detectable by Western blot (data not shown), thus arguing for a physiological relevant observation rather than an expression artifact. Furthermore, the loss of the epithelial phenotype strictly correlated with the expression of ZO-1 mutants encoding the PDZ domains that failed to associate with the plasma membrane (i.e. ZO-1-PDZ and ZO-1-PS) and was never observed in cells transfected with the other ZO-1 constructs or in vector-transfected control cells.
The EMT induced by cytosolic ZO-1 PDZ proteins was characterized by a fibroblastoid morphology of the MDCK-PDZ cells with changes in the organization of the actin cytoskeleton, the loss of contact inhibition of proliferation and differentiation potential, as well as an increased tumorigenicity in nude mice. These phenotypic alterations correlated with a reduced expression level of proteins characteristic for epithelial cells (i.e. Ecadherin and cytokeratins). Ectopic expression of APC in MDCK-PDZ cells reverted the transformed phenotype and led to the re-expression of E-cadherin and cell-cell contact formation. Down-regulation of E-cadherin expression correlates with increased tumor invasion, metastasis, and poor clinical prognosis (43). Furthermore, inhibition of E-cadherin function with interfering antibodies alters the morphology of MDCK cells and enables them to invade both collagen gels and embryonic chicken heart tissue (43,44), consistent with the importance of E-cadherin in maintaining the epithelial phenotype. However, ectopic expression of full length E-cadherin was not sufficient to revert the fibroblastoid phenotype of MDCK-PDZ cells (data not shown). Since nonepithelial cadherins like N-cadherin can promote motility and invasiveness of epithelial cells regardless of the presence of E-cadherin (45), it will be interesting to determine if MDCKI cells express nonepithelial cadherins or if their expression is induced in MDCK-PDZ cells.
MDCK-PDZ cells also showed an increase in the mRNA levels of fibronectin and vimentin. Vimentin is a type III intermediate filament protein normally expressed in cells of mesenchymal origin. Using an in vitro wound healing model, it has been shown that vimentin mRNA and protein expression were exclusively induced in cells at the wound's edge that were actively migrating toward the center of the lesion. The vimentin protein disappeared when the cells became stationary after the wound closure (46). This correlates with the observed migrating phenotype of MDCK-PDZ cells. The changes at the protein level reflected, at least for some proteins, corresponding changes in mRNA levels, indicating that the ZO-1 PDZ protein had an effect on gene expression and/or mRNA stability.
Intriguingly, ␤-catenin/Tcf/Lef signaling was constitutively activated in MDCK-PDZ cells. ␤-Catenin/Tcf/Lef has been implicated in signaling events leading to EMT in vivo during embryonic development (47) and in vitro in epithelial cells (48). We observed an inverse correlation between ␤-catenin/Tcf/Lef transcriptional activity and the expression of E-cadherin and cytokeratins, consistent with a negative regulation of transcription by ␤-catenin signaling via Tcf/Lef binding sites present in the promoter regions of the E-cadherin and cytokeratin genes (34,35,49).
Free cytosolic ␤-catenin, upon translocation into the nucleus and interaction with Tcf/LEF family members, can activate the transcription of specific target genes. Despite a significant increase in the transcriptional activation of the TOP reporter construct by ␤-catenin, the free pool of ␤-catenin was not elevated in MDCK-PDZ cells. Transcriptional activation via ␤-catenin/Tcf/Lef without an increase in the free pool of ␤-catenin is not unprecedented and has been described for the integrin-linked kinase-mediated transformation (39). Furthermore, nuclear localization of ␤-catenin and Tcf/Lef may not be sufficient to activate gene expression, implicating additional cell type-specific regulatory factors (40). An example of an additional regulatory mechanism may be the interaction of the Drosophila orthologue of ␤-catenin, Armadillo, with Teashirt, a transcription factor involved in the establishment of trunk segment identity and whose nuclear localization depends on Wnt signaling (50).
Our results demonstrating a contribution of ␤-catenin/Tcf/ Lef signaling in the EMT induced by the ZO-1 PDZ protein are in line with several studies implicating activated ␤-catenin/Tcf/ Lef signaling in epithelial cell transformation (16,51,52). Modest overexpression of ␤-catenin in MDCK cells leads to cellular transformation by affecting contact inhibition of proliferation, anchorage-independent growth, anoikis, and radiation-induced cell cycle arrest (51). The ectopic expression of a constitutively active N-terminal deletion mutant of ␤-catenin in MDCKII cells results in a dispersed fibroblastoid morphology (48). Furthermore, an activated mutant of ␤-catenin induces dysplasia and adenoma in transgenic mice (52,53). Loss of APC thereby activating ␤-catenin signaling also leads to dysplasia and adenoma formation (53,54). Interestingly, during HGF-induced migration of MDCKII cells, an increase in tyrosine phosphorylation of ␤-catenin correlates with the translocation of ZO-1 to the cytosol and HGF inhibits the reassembly of ZO-1 at the plasma membrane (55).
The mechanism by which the ZO-1 PDZ protein activates ␤-catenin/Tcf/Lef signaling and induces the observed EMT is currently under investigation, and several possibilities can be considered. The constitutive cytosolic localization of the ZO-1 mutant could interfere with the proper negative regulation of a signaling function of ZO-1 at the plasma membrane, similar to the negative regulation of ␤-catenin/Tcf/Lef signaling by cadherins (56 -59). Alternatively, the ZO-1 PDZ protein could interact with molecules of the Wnt signaling pathway or with components of adherens junctions (i.e. ␣-catenin; Ref. 60), thereby altering the equilibrium between free and sequestered ␤-catenin or interfering with the signaling pathway. Interestingly, the hDlg protein, which shows homology to ZO-1, interacts with APC (61). While it is not known whether ZO-1 binds APC, such an interaction would be consistent with our observation that the ectopic expression of APC, known to negatively regulate activated ␤-catenin/Tcf/Lef signaling (41), reverted the EMT in MDCK-PDZ cells. Yet another possible link could involve the interaction between ZO-1 and AF-6 (62, 63), which in turn can recruit the de-ubitquitinating enzyme FAM to sites of cell-cell adhesion (64). Since FAM also interacts with ␤-catenin and E-cadherin in vitro, FAM may normally stabilize cellcell contacts by de-ubiquitinating proteins at these sites. Interestingly, p21Ras interferes with the interaction between ZO-1 and AF-6 (65), thereby contributing to the perturbation of cell-cell adhesion in p21Ras-transformed cells. Finally, because ZO-1 was shown to translocate to the nucleus of migrating epithelial cells (24), the ZO1-PDZ protein itself could be engaged in an active signaling process. Since, in contrast to in vitro model systems, mutations in APC or ␤-catenin that activate ␤-catenin/Tcf/Lef signaling in vivo normally lead to less dramatic changes of the epithelial phenotype (i.e. induction of hyperplasia or dysplasia and adenomas; Refs. 53 and 54), the observed profound effects raise the possibility that the ZO-1 PDZ protein itself may play a more direct role in mediating the observed EMT.
In conclusion, our results show that expression of the ZO1-PDZ protein in MDCKI cells leads to a fibroblastoid, transformed phenotype of these cells in vitro and an increased tumorigenicity in vivo, paralleled by the activation of ␤-catenin/ Tcf/Lef transcriptional activity. Thus, a dysregulation of ZO-1 localization or function may contribute to an EMT and possibly to the development of tumors. Indeed, in breast carcinomas, the absence of ZO-1 at the plasma membrane correlates with tumor formation (66). The novel finding that ␤-catenin/Tcf/Lef signaling can be modulated by a member of the TJ, together with the established regulation by components of AJ and integrins, indicates that this signaling pathway plays a general role in integrating signals generated in response to cell-cell and cellmatrix interactions.