E-cadherin-mediated Cell-Cell Attachment Activates Cdc42*

E-cadherin is a transmembrane protein that mediates Ca2+-dependent cell-cell adhesion. Cdc42, a member of the Rho family of small GTPases, participates in cytoskeletal rearrangement and cell cycle progression. Recent evidence reveals that members of the Rho family modulate E-cadherin function. To further examine the role of Cdc42 in E-cadherin-mediated cell-cell adhesion, we developed an assay for active Cdc42 using the GTPase-binding domain of the Wiskott-Aldrich syndrome protein. Initiation of E-cadherin-mediated cell-cell attachment significantly increased in a time-dependent manner the amount of active Cdc42 in MCF-7 epithelial cell lysates. By contrast, Cdc42 activity was not increased under identical conditions in MCF-7 cells incubated with anti-E-cadherin antibodies nor in MDA-MB-231 (E-cadherin negative) epithelial cells. By fusing the Wiskott-Aldrich syndrome protein/GTPase-binding domain to a green fluorescent protein, activation of endogenous Cdc42 by E-cadherin was demonstrated in live cells. These data indicate that E-cadherin activates Cdc42, demonstrating bi-directional interactions between the Rho- and E-cadherin signaling pathways.

teins. Briefly, MBP-WASP-GBD was produced by excising WASP-GBD from the GST plasmid with EcoRI; blunt ends were generated with Klenow. The pMAL-c2X vector (New England Biolabs), which encodes MBP, was digested with PstI and blunt-ended with T4 polymerase. Both the WASP-GBD fragment and pMAL-c2X vector were cut with BamHI. The 214-base pair WASP-GBD fragment was purified from a low melting agarose gel and was inserted into the pMAL-c2X vector. The MBP-WASP-GBD fusion protein was expressed in E. coli and purified over an amylose column.
To generate GFP-WASP-GBD, the WASP-GBD was excised from the GST plasmid with BamHI and EcoRI. The resultant 214-base pair fragment was purified from a low-melting agarose gel and inserted into pEGFP-C1 (CLONTECH) at BglII and EcoRI sites. The sequence of the WASP-GBD in all of the fusion constructs was confirmed by restriction mapping and DNA sequencing.
Assay for the Detection of Activated Cdc42-MCF-7 and MDA-MB-231 human breast epithelial cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% (v/v) fetal bovine serum in a 37°C humidified incubator. L and EL cells were grown in the same medium supplemented with 400 g/ml hygromycin B. Cells were washed three times in phosphate-buffered saline (PBS) (145 mM NaCl, 12 mM Na 2 HPO 4 , and 4 mM NaH 2 PO4, pH 7.2) and quick-frozen in 500 l of lysis buffer (20 mM Hepes, pH 7.4, 150 mM NaCl, 1% (v/v) Nonidet P-40, 20 mM NaF, 20 M GTP, and protease inhibitors). Lysates were thawed, clarified by centrifugation at 15 000 ϫ g for 5 min at 4°C, and precleared by incubating 1.25 mg of lysate for 1 h at 4°C on a rotator with 40 l of glutathione-Sepharose. In selected experiments, 1 mM ATP, 20 M GDP, or 100 M GTP␥S was added to the lysates for 15 min. Protein concentrations were measured and samples analyzed directly as whole cell lysates (ϳ75 g), or equal amounts of protein were incubated with 40 g of GST-WASP-GBD for 2 h at 4°C. GST alone was processed in parallel as a control. Complexes were collected with glutathione-Sepharose, the beads were washed six times with PBS, and proteins were resolved by SDS-PAGE and transferred to PVDF. Blots were probed with anti-Cdc42 antibody followed by horseradish peroxidase-conjugated sheep anti-mouse antibody and developed by ECL.
E-cadherin-mediated cell adhesion was performed essentially as described previously (20). MCF-7 cells were grown to confluence in 60-mm culture dishes. 6 to 8 h later, cells were serum-starved for 16 -18 h and incubated with DMEM containing 4 mM EGTA and 1 mM MgCl 2 for 30 min at 37°C. The EGTA chelates Ca 2ϩ , thereby disrupting E-cadherin homophilic interactions, and the Mg 2ϩ maintains integrin function. The Ca 2ϩ -free medium was removed, and DMEM containing 1.8 mM CaCl 2 was added to induce E-cadherin-mediated cell-cell interactions. In selected experiments, anti-E-cadherin (DECMA-1 clone) antibodies (1:100 dilution), 50 M LY294002, 500 nM wortmannin, or the appropriate vehicle were included in the medium. Cells were lysed at different time intervals after the addition of Ca 2ϩ . In all experiments, the time of addition of Ca 2ϩ was considered 0 min.
Immunoprecipitation-MCF-7 cells were transfected with GFP or GFP-WASP-GBD using FuGENE 6 transfection reagent (Roche Molecular Biochemicals) according to the manufacturer's instructions. The following day, cells were incubated with vehicle or 200 nM bradykinin for 5 min and processed as described above. In selected samples, 100 M GTP␥S was added after lysis. Immunoprecipitation was performed essentially as described previously (21). Briefly, after preclearing, equal amounts of protein were incubated with anti-GFP or anti-myoglobin (an isotype-identical, irrelevant) monoclonal antibody for 3 h at 4°C. Immune complexes were collected for 2 h with 40 l of Gamma Bind G-Sepharose (Amersham Pharmacia Biotech) and washed five times with lysis buffer, and Western blotting was performed as outlined above.
Immunofluorescence Staining and Confocal Microscopy-MCF-7 cells were removed from culture dishes with trypsin, washed twice with PBS, and allowed to attach to Permanox plastic slides overnight at 37°C in DMEM. Cells were serum-starved for 24 h. Following a 30-min incubation with DMEM containing 4 mM EGTA and 1 mM MgCl 2 , cells were incubated with DMEM containing 1.8 mM CaCl 2 . For immunofluorescence staining, cells were washed with PBS and fixed with 3.7% (v/v) formaldehyde for 20 min at 22°C. Permeabilization was performed by adding 0.5% (v/v) Triton X-100 in PBS for 10 min followed by blocking with 2% (w/v) bovine serum albumin in PBS for 1 h at 4°C. After two washes with PBS, cells were incubated for 1 h at 4°C with rhodamine phalloidin (Molecular Probes), or E4.6 anti-E-cadherin monoclonal antibody, followed by fluorescein isothiocyanate-labeled goat anti-mouse antibody. Slides were washed four times with PBS and mounted with Aqua Polymount (Polysciences, Inc.). The specificity of staining was verified by omitting the primary antibody. Confocal laser scanning microscopy was performed with the MRC-1024 Confocal Imaging System (Bio-Rad) as described previously (21).
Live Cell Imaging-For live cell imaging, MCF-7 cells were plated on Lab-Tek II chambered coverglass (Nalge Nunc Int.). The following day, the cells were transiently transfected with GFP or GFP-WASP-GBD. Transfection efficiency was usually 40 -50%. E-cadherin-mediated cell adhesion was performed as described above. Cells were imaged at 1-min intervals for 70 min by time lapse video microscopy using a Zeiss Axiovert S100 microscope. Fluorescence images were collected with the MRC-1024 confocal imaging system and processed as described (21).
Miscellaneous-Protein concentrations were determined with the DC protein assay (Bio-Rad). Densitometry of ECL signals, performed in triplicate, was analyzed with NIH Image. Statistical significance was evaluated by Student's t test, using InStat software (GraphPad Software, Inc.).

RESULTS
Assay of Active Cdc42-The specificity of the WASP-GBD constructs was evaluated with recombinant GTPases in vitro and with endogenous GTPases in cell lysates. Initial characterization was directed toward confirming the specificity of the WASP-GBD for the active (GTP-bound) conformation of the GTPases. A MBP fusion protein of WASP-GBD was incubated with GDP-and GTP␥S-bound forms of GST fusion proteins of small GTPases. The WASP-GBD bound specifically to active GST-Cdc42-GTP␥S (data not shown). Active Rac bound to WASP-GBD but with lower affinity. No specific binding was detected to the active forms of Rho or Ras nor to the inactive (GDP-bound) GTPases (data not shown).
Analysis of endogenous GTPases revealed the presence of constitutively active Cdc42 in MCF-7 cell lysates (Fig. 1A). As anticipated, the addition of GTP␥S dramatically increased the  4) or GFP-WASP-GBD (lanes 1, 3, 5, and 6), followed by incubation with (lanes 4 and 5) or without (lanes 1, 2, 3, and 6) bradykinin (BK) for 5 min. Where indicated, GTP␥S was added after cell lysis. Equal amounts of protein lysate were immunoprecipitated (IP) with anti-GFP (␣GFP) or an isotype-identical irrelevant (␣myo) monoclonal antibody. Immunocomplexes were resolved by SDS-PAGE and transferred to PVDF membrane, and blots were probed with anti-Cdc42 or anti-Rac antibodies. amount of active Cdc42 in the lysate, whereas ATP and GDP had no significant effect (Fig. 1A). GTP␥S did not change the total amount of Cdc42 in the lysate (data not shown). No Cdc42 was detected in samples incubated with GST alone, even in the presence of GTP␥S, verifying that binding was specific for the WASP-GBD. Active Rac bound to the WASP-GBD only in samples spiked with GTP␥S (Fig. 1A). Incubation of MCF-7 cells with bradykinin, a known activator of Cdc42 (22), increased by 3-fold the amount of Cdc42 in GST-WASP-GBD pulldowns (Fig.  1A). By contrast, bradykinin had no effect on the amount of Rac that bound to WASP-GBD. A comparison of the relative amounts of active Rac and active Cdc42 with the total amounts of the GTPases (Fig. 1A) is consistent with WASP preferentially binding Cdc42 over Rac (12,23). Together, these analyses validate the proposition that the WASP-GBD, like the fulllength protein, interacts specifically with the active forms of Cdc42 and, to a substantially lesser extent, Rac.
A construct of the WASP-GBD fused to GFP was created to examine Cdc42 activation in live cells (see below). The validity of the construct was assessed in cells transfected with GFP-WASP-GBD. Immunoprecipitation with anti-GFP antibody revealed that the construct bound Cdc42 (Fig. 1B, lane 3). Activation of Cdc42 with GTP␥S or bradykinin substantially increased the amount of Cdc42 that co-immunoprecipitated with GFP-WASP-GBD (Fig. 1B), demonstrating that the GFP did not interfere with the ability of WASP-GBD to bind active Cdc42. No Cdc42 was detected in anti-GFP immunoprecipitates from samples transfected with GFP alone (Fig. 1B, lanes   2 and 4), indicating that Cdc42 bound specifically to the WASP-GBD. No Rac was detected in the immunoprecipitates, even in the presence of GTP␥S (data not shown).
E-cadherin-mediated Cell-Cell Adhesion Activates Cdc42-There is evidence that cell adhesion regulates the cytoskeleton (24) and that members of the Rho family modulate E-cadherin function (6,25). Therefore, we addressed the important question of whether E-cadherin-mediated adhesion would alter activation of Cdc42. E-cadherin-mediated cell-cell adhesion was abrogated by chelating Ca 2ϩ and then re-initiated by re-introducing Ca 2ϩ (26). This approach induced a time-dependent accumulation of E-cadherin at cell-cell junctions ( Fig. 2A). Adherens junctions were detected in MCF-7 cells 15 min after the addition of Ca 2ϩ , and junction formation was essentially complete by 90 min. This assay was used to explore whether Ecadherin modulates Cdc42 activation. Induction of E-cadherin homophilic adhesion by Ca 2ϩ significantly increased the amount of active Cdc42 in MCF-7 cell lysates (Fig. 2B). Activation of Cdc42 was time-dependent. Increased active Cdc42 was detected at 30 min, peaked at 2.2-fold at 60 min, and returned to basal values by 90 min. Total cellular Cdc42 did not change significantly during this time interval (Fig. 2B). Incubation with EGTA for 30 min altered neither the amount of total or active Cdc42 in the lysates (data not shown).
Initiation of E-cadherin-mediated cell-cell adhesion in MDCK cells resulted in a time-dependent activation of PI3kinase and Akt kinase (20). Akt, a major downstream target of PI3-kinase, is activated by phosphorylation on Ser-473 and Thr-308 by phosphoinositide-dependent protein kinase(s) (27). Engagement of E-cadherin in MCF-7 cells activated endogenous Akt, as revealed by a significant increase in Akt phosphorylation (Fig. 2C). Interestingly, the time course and magnitude (peak increase of 2.0-fold at 60 min) of Akt activation are generally similar to those of Cdc42, except Akt remains increased at 90 min.
E-cadherin-mediated Cell-Cell Adhesion Is Necessary for Cdc42 Activation-Although the above results implicate Ecadherin as the source of Cdc42 activation, other factors could be responsible. Several strategies were adopted to address this question. Analogous experiments were performed in MDA-MB-231 cells, a human breast epithelial cell line that does not contain E-cadherin (21). Incubation of MDA-MB-231 cells with Ca 2ϩ failed to increase active Cdc42 (Fig. 3A). In fact, the amount of active Cdc42 decreased with time. An increased in active Cdc42 was evident in MDA-MB-231 lysates spiked with GTP␥S that were processed in parallel (data not shown). Similarly, the absence of E-cadherin prevented the Ca 2ϩ -dependent cell-cell contact activation of Akt in MDA-MB-231 cells, whereas incubation with epidermal growth factor, a known activator of Akt (28), enhanced Akt phosphorylation (data not shown). A second control was the comparison of L fibroblasts (which lack endogenous E-cadherin) with EL cells, which are L fibroblasts stably transfected with E-cadherin (21). Ca 2ϩ induced a time-dependent activation of Cdc42 in EL cells, whereas, analogous to MDA-MB-231 cells, active Cdc42 decreased in L cells (Fig. 3B). Third, MCF-7 cells were incubated with anti-E-cadherin (DECMA-1) antibodies that block E-cadherin-mediated junction formation (29). As we observed in cells lacking E-cadherin, the antibodies abrogated the E-cadherinmediated increase in active Cdc42 (Fig. 3C). The antibody alone altered neither the amount of total nor active Cdc42. The reason for the decrease in active Cdc42 in the absence of E-  cadherin function is unknown. Nevertheless, together these data strongly support the notion that E-cadherin homophilic association is responsible for the activation of Cdc42.
E-cadherin Induces the Formation of Filopodia-Activation of Cdc42 in fibroblasts leads to the formation of filopodia or microspikes (5). Therefore, we examined the effect of E-cadherin on the actin cytoskeleton. E-cadherin-mediated cell-cell adhesion induced the formation of filopodia in MCF-7 cells (Fig.  4A). Multiple thin actin fibers are present at the cell membrane, both in the intercellular region at areas of cell-cell attachment (arrowheads in Fig. 4A; note also the higher magnification in panels 30b and 60b) and at the free edges of the cell (arrows in Fig. 4A). These filopodia failed to develop when analysis was performed in parallel in the presence of anti-Ecadherin antibody (Fig. 4B). Note that the changes in the actin cystoskeleton generated by E-cadherin are virtually identical to those produced by activation of Cdc42 in MCF-7 cells by bradykinin (Fig. 4C).
Role of PI3-kinase in E-cadherin-mediated Activation of Cdc42-E-cadherin-mediated cell adhesion activates PI3-kinase (20). This observation, coupled with evidence implicating PI3-kinase in the activation of Cdc42 (30), prompted us to evaluate the participation of PI3-kinase in E-cadherin-mediated activation of Cdc42. Inhibition of PI3-kinase with LY294002 (Fig. 5A) or wortmannin (Fig. 5B) abolished the Ca 2ϩ -dependent cell-cell contact activation of Cdc42 in MCF-7 cells. These observations imply the involvement of a PI3-kinasedependent pathway in the activation mechanism.
E-cadherin Induces Activation of Cdc42 in MCF-7 Cells-To visualize active Cdc42 in living cells, the WASP-GBD was fused to GFP and transfected into MCF-7 cells. E-cadherin homophilic interaction was induced with Ca 2ϩ , and the location of GFP-WASP-GBD was studied in real time with laser-scanning confocal microscopy. Both GFP-WASP-GBD and GFP were distributed throughout the cell initially (Fig. 6). Sixty-five minutes after the addition of Ca 2ϩ , a fraction of the GFP-WASP-GBD localized to the plasma membrane (Fig. 6A, arrows). This effect was not seen in isolated cells. Note that the peak of active Cdc42 in cell lysates was detected at 60 min (see Fig. 2B). The translocation required the WASP-GBD, as GFP alone did not accumulate at the plasma membrane (Fig. 6B). Importantly, the presence of anti-E-cadherin (DECMA-1) antibodies abrogated specific localization of GFP-WASP-GBD at the plasma membrane (data not shown).
Injection of the Cdc42-binding domain of WASP blocked actin filament assembly by the Cdc42 guanine nucleotide exchange factor, FGD1 (31). To demonstrate that the low level of GFP-WASP-GBD expressed in MCF-7 cells did not inhibit Cdc42 function, we activated Cdc42 with bradykinin. There was no difference in the development of filopodia among cells transfected with GFP-WASP-GBD (Fig. 6C), GFP (data not shown), or nontransfected cells. DISCUSSION Rho proteins, which control the organization of the actin cystoskeleton in all eukaryotic cells (5), were recently identified as important regulators of cadherin-dependent contacts (13)(14)(15)(16)25). The important question of whether cadherin adhesiveness can trigger the activation of the GTPases (25) has been addressed in this paper.
Using the knowledge that only active Cdc42, and to a lesser extent Rac, binds WASP with high affinity (12, 23), we developed an assay for GTP-bound Cdc42. Fusion proteins of the WASP-GBD bound specifically to active Cdc42, both in vitro and in cell lysates. Factors known to increase the amount of active Cdc42, namely GTP␥S and bradykinin, significantly augmented the amount of Cdc42 bound to the WASP-GBD. Recently, other investigators have used fusion constructs of the GTPase-binding domains of Ras family targets, namely Raf-1, Rhotekin, and p21-activated kinase, to assay activated Ras, Rho, and Rac/Cdc42, respectively (32)(33)(34). These reagents have proved to be valuable in enhancing our understanding of the regulation of small GTPases in cells. However, p21-activated kinase discriminates little between Cdc42 and Rac (35) and cannot be used in live cells. We therefore used WASP-GBD, which has a substantially higher affinity for active endogenous Cdc42 than for active endogenous Rac (see Fig. 1).
Initiation of E-cadherin-mediated cell-cell attachment increased the amount of active Cdc42 in cell lysates. Activation peaked at approximately 60 min, which is consistent with the time course for establishment of adherens junctions in EL cells (36). E-cadherin homophilic interactions were required for activation of Cdc42 as demonstrated by the absence of activation under identical assay conditions in cells lacking E-cadherin. Moreover, preventing the formation of adherens junctions with anti-E-cadherin antibodies (29) abrogated activation of Cdc42. The extent of the increase in active Cdc42 that we observed is similar to that seen for Rac, Cdc42 (34), and Rho (33) in response to other stimuli. The amount of active Cdc42 reflects the entire cellular pool of Cdc42, and it is conceivable, perhaps even likely, that higher concentrations of active Cdc42 may be localized in discrete subcellular pools, particularly those associated with the plasma membrane. A recent study with MDCK cells revealed that E-cadherinmediated attachment enhanced endogenous Akt activity by Ͼ10-fold (20). The magnitude of the increase was greater than and observed slightly earlier than our results, which may be due to differences in the cell lines and assay methodology. We used an antibody specific for the Ser-473-phosphorylated form of Akt, a prerequisite for Akt activation (27). Although accepted as an assay of Akt activity (37), the phospho-specific antibody may underestimate Akt activity compared with direct measurements of kinase activity. This conjecture is supported by the detection of a 4-fold increase in Akt activity by epidermal growth factor under our conditions (data not shown) compared with Ͼ15-fold demonstrated by Pece et al. (20). It is also possible that differences between the cell lines may account for the variability in the extent of Akt activation in response to both E-cadherin and epidermal growth factor.
The GFP-WASP-GBD construct we developed permitted for the first time the visualization of Cdc42 activation in live cells in real time. Beginning approximately 1 h after initiation of E-cadherin attachment, GFP specifically accumulated in discrete puncta at the plasma membrane. Controls revealed that both the WASP-GBD construct and E-cadherin homophilic attachment were necessary to detect this translocation. This observation validates in live cells the previous findings obtained by fractionation that activation induces the translocation of Rho family GTPases to the plasma membrane (38). Because of the diffuse distribution of the GFP construct throughout the cell, it was not possible to establish whether active Cdc42 increased at the region of cell-cell attachment. (Attempts to ascertain this using purified GFP-WASP-GBD protein as an indicator in fixed, permeabilized cells were not successful.) A conceptually analogous strategy, employing GFP fused to the pleckstrin homology domain of selected proteins, has been used to demonstrate that 3Ј-phosphoinositide prod-ucts of PI3-kinase are localized at the plasma membrane (39,40). Although it is not possible to verify unequivocally the target of the GFP-WASP-GBD, our data are consistent with specific identification of active Cdc42 in living cells. WASP is reported to be Cdc42-specific (12,41) and is inhibited in intact cells by dominant negative Cdc42 but not dominant negative Rac or Rho (12). Immunoprecipitation revealed that Cdc42, but not Rac, associated with GFP-WASP-GBD in the cell milieu. The solution structure of Cdc42 in complex with WASP-GBD provides insight into the ability of WASP to distinguish Cdc42 from Rac and Rho (42). Hydrogen bonding between side chains in activated Rac and Rho closes off the pocket contacted by Ile-233 of WASP and would disrupt the hydrophobic contacts to Lys-235 (42).
The mechanism by which E-cadherin activates Cdc42 is unknown. There is no evidence for a direct interaction between Cdc42 and E-cadherin. A potential candidate that may provide a molecular link is IQGAP1, which binds directly to Cdc42 and E-cadherin (16,21,43). IQGAP1 competes with ␣-catenin for binding to E-cadherin/␤-catenin, leading to disruption of cellcell adhesion (43). Based on their observation that active Cdc42 inhibits this effect of IQGAP1, Kaibuchi et al. (16) propose that Cdc42 and IQGAP1 can serve as positive and negative molecular switches of cadherin activity, and they speculate that cell-cell contact would induce the activation of Rho family GTPases. Our results validate this supposition. However, we have no direct evidence that IQGAP1 participates in the Ecadherin-mediated activation of Cdc42, and there are no specific inhibitors nor dominant negative forms of IQGAP1 available to directly test this hypothesis. A second potential intermediary is PI3-kinase, which is activated by E-cadherin (20) and is reported to be both upstream and downstream of Cdc42 (44, 45). We did not detect E-cadherin-induced activation of Cdc42 in the presence of the PI3-kinase inhibitors wort- mannin or LY294002, a structurally distinct inhibitor, implicating the participation of a PI3-kinase-dependent pathway. The process by which PI3-kinase activates Rho family proteins is unknown. The pleckstrin homology domains, which bind phosphatidylinositol lipids and are found in GEFs, have been postulated as the link (46). A third possibility is modulation of Cdc42 cycling. Activation of a GEF or inhibition of a GAP would result in an increase in the amount of active Cdc42 in the cell. The proposed mechanisms are not mutually exclusive, and more than one pathway may be involved. For example, the ␣ isoform of p21-activated kinase-interacting exchange factor, PIX, a GEF for Rac and Cdc42, is activated by PI3-kinase (47).
Regardless of the mechanism, our data indicate that E-cadherin activates Cdc42, yielding further evidence for outside-in signaling by E-cadherin. These results also provide the mechanism by which keratinocytes developed filopodia that were integral to the establishment of E-cadherin-mediated intercellular adhesion (48) and confirm the authors' postulation that Cdc42 was involved. Our findings establish bi-directional communication between Cdc42 and E-cadherin and identify an additional route by which intercellular interactions can influence intracellular signaling pathways.