Integrin (cid:1) 6 (cid:2) 4-erbB2 Complex Inhibits Haptotaxis by Up-regulating E-cadherin Cell-Cell Junctions in Keratinocytes*

Keratinocyte integrins (cid:1) 6 (cid:2) 4 and (cid:1) 3 (cid:2) 1 bind laminin-5, a component of basement membranes. We previously demonstrated that in keratinocytes, haptotactic migration on laminin-5 was stimulated by anti- (cid:2) 1 integrin-activating antibody TS2/16, whereas antibodies to (cid:1) 6 and (cid:2) 4, respectively, blocked TS2/16-induced, (cid:1) 3 (cid:2) 1-de-pendent migration. Moreover, (cid:1) 6 (cid:2) 4-associated haptotaxis inhibition was linked to a phosphatidylinositol 3-kinase (PI3K) pathway and required erbB2 activation. erbB2, the ligand-less member of the epidermal growth factor receptor family, was shown to form a complex with the hemidesmosomal integrin (cid:1) 6 (cid:2) 4. Here, we demonstrate that (cid:1) 6 (cid:2) 4 inhibitory effects on haptotaxis are abolished by an anti-E-cadherin antibody, which inter-feres with cell-cell adhesion. Furthermore, antibodies to (cid:1) 6 and (cid:2) 4 stimulated adhesion to an E-cadherin-Fc recombinant protein. In addition, anti- (cid:1) 6/ (cid:2) 4 antibodies increased colony size in plated cells, stimulated cell-cell aggregation, and up-regulated E-cadherin localization to cell-cell contacts. These effects were abolished when erbB2 or PI3K were blocked. These results indicate that stimulation of (cid:1) 6 (cid:2) 4 increases E-cadherin-mediated cell-cell adhesion and that this mechanism depends on erbB2 activation. The molecule that links (cid:1) 6 (cid:2) 4 with E-cadherin may be the small GTPase cdc42, an effector of PI3K, because dominant-negative cdc42 abolished the inhibitory effect of anti- (cid:1) 6/ (cid:2) 4 antibodies and increased basal migration, whereas constitutively active cdc42 prevented the TS2/16-induced increase in haptotaxis. These findings suggest a model whereby (cid:1) 6 (cid:2) 4 can aug-ment cell-cell adhesion and slow down haptotaxis over laminin-5 and point to the (cid:1) 6 (cid:2) 4-erbB2 heterodimer as an important signaling complex for the formation of cohesive keratinocyte layers.

Epithelia are cohesive sheets of cells that form a protective barrier between the body interior and either the outside envi-ronment or organ lumen. To build such tissues, cells must be incorporated into a collective and the whole cell population must stably adhere to the underlying substrate, the basement membrane (BM). 1 In general, cell-BM and cell-cell adhesion are mediated by distinct receptor systems, among which integrins and cadherins, respectively, play a decisive role (1).
Integrins are ␣␤ heterodimers with large extracellular domains and short, but functionally essential cytoplasmic domains (2). Interactions between integrins and the BM not only determine structural features of a tissue but also transmit signals about the surrounding microenvironment to epithelial cells, which in turn determine cellular behavior. BM-binding integrins include ␣3␤1 and ␣6␤4, which are both receptors for the extracellular matrix (ECM) component laminin-5 (Ln-5) (3) but are recruited to different cell adhesion structures. Integrin ␣3␤1 is localized to focal contacts (4) and links Ln-5 to the actin cytoskeleton, mediating cell spreading and migration (5,6), whereas ␣6␤4 is a component of hemidesmosomes (HDs), connecting Ln-5 to the keratin filament network (5). HDs are specialized anchoring complexes that allow static adhesion of epithelial cells to the BM and thus guarantee high mechanical stability (7). In the absence of integrin ␤4, stratified epithelia, i.e. consisting of several layers like the epidermis of the skin (8), detach from the BM, leading to the severe blistering disease epidermolysis bullosa (7). Therefore, ␣6␤4 may play an essential role during the integration of cells into epithelial sheets.
Binding to Ln-5 via ␣3␤1 in focal contacts or ␣6␤4 in HDs transmits distinct molecular signals, which control keratinocyte haptotactic (ECM-directed) migration differently (6). This may be the reason why Ln-5 was found to support both static adhesion and cell migration (9 -11). Clustering of ␣3␤1 by Ln-5 stimulates focal adhesion kinase (FAK) (6), which is involved in the turnover of focal adhesions and modifications of the cytoskeleton, leading to an increase in cell motility (12). Furthermore, extracellular signal-regulated kinases (ERKs) 1 and 2 are stimulated (6), which are mitogen-activated protein kinases found to up-regulate cell migration (13). Thus, ␣3␤1 is the integrin that acts in response to the role of Ln-5 as a migratory substrate. In contrast, stimulation of ␣6␤4 inhibits ␣3␤1-controlled cell motility via an increase in PI3K activity (6). This result correlates with the finding that Ln-5 may slow down keratinocyte migration, probably due to the formation of cell anchoring hemidesmosomal complexes (10,11). However, it is in contradiction to the observation that ␣6␤4 activation leads to a PI3K-dependent increase in Matrigel invasion of breast and colon cancer cells (14). This discrepancy may be due to the different migratory substrates tested (Ln-5 versus Matrigel), the level of ␤4 expression (endogenous protein versus overexpressed protein), the mechanism of migration (haptotaxis versus chemotaxis or invasion), or the different cell lines analyzed (keratinocytes versus breast and colon cells). To trigger PI3K activation that leads to a decrease in haptotaxis, ␣6␤4 uses erbB2, a member of the epidermal growth factor/erbB receptor family (15), as an adapter protein to interact with downstream signaling pathways (6,16). We and others have demonstrated specific complex formation between ␣6␤4 and erbB2 (6,16), as well as ␣6␤4-dependent activation of erbB2 kinase activity (6). Cooperation between integrins and growth factor receptors is a common theme and may be a prerequisite for ECM-dependent control of biological processes such as cell motility during wound repair, inflammation, or organogenesis (17).
Besides cell-BM association, cell-cell contact formation is essential for the assembly of cohesive epithelial sheets and is mediated by several types of junctions, including tight junctions, desmosomes, and adherens junctions (AJs). Recent evidence has exposed a key role for AJs in the regulation of intercellular adhesion as formation of tight junctions and desmosomes depends on the preceding assembly of AJs (18,19). AJs are composed of the transmembrane protein E-cadherin and associated cytoplasmic proteins, like ␤or ␥-catenin and p120-catenin (p120). The ␤and ␥-catenins interact with ␣-catenin, which links the cadherin-catenin complex to the actin cytoskeleton, whereas p120 has regulatory functions (20). Studies in primary keratinocytes underscore the importance of actin in initiating cell-cell adhesion, because cells send out filopodia, which penetrate and project into neighboring cells. Clustered at the filopodia tips are AJs, which zip opposing cell surfaces together and seal them into continuous sheets (21). Formation of AJs is regulated by several different mechanisms, including phosphorylation of components of the complex or alterations in the interaction with actin (20). Furthermore, the small GTPases rac1, rhoA and cdc42, which are known regulators of the actin cytoskeleton and various cell adhesion events, have been shown to play an active role in the regulation of AJs (18,22).
In this study, we report that clustering ␣6␤4 can stimulate E-cadherin-controlled cell-cell adhesion in an erbB2-dependent manner and that this event blocks ␣3␤1-mediated keratinocyte haptotaxis. Thus, ␣6␤4 is an integrin that triggers both cell-BM and cell-cell adhesion by binding to the same ECM component Ln-5. These characteristics make ␣6␤4 the ideal regulator of epithelial sheet formation, controlling important processes such as wound healing and cancer invasion. Our results further point to a unique role for erbB2 as signaling adaptor in the regulation of ␣6␤4-controlled cellular behavior.
tive cdc42(V12) fused to a HA tag were gifts from Dr. K. Mizuno (Tohoku University, Sendai, Japan). Constructs encoding dominantnegative PI3K regulatory subunit p85⌬iSH2-N and dominant-negative erbB2 (HER2VEK753A) have been described previously (6). All cDNAs were subcloned into the retroviral vector pLNCX (Clontech, Palo Alto, CA). Virus production in PT67 packaging cells (Clontech) and HaCaT infection were performed as described in the manufacturer's protocol. A retroviral vector encoding enhanced green fluorescent protein was used to assess infection efficiency, which was at least 95% in each experiment. Stable HaCaT clones were selected with 0.7 mg/ml G418 (Geneticin, Invitrogen).
Migration and Adhesion Assays-Transwell migration assays were performed as described (6). Filters were coated with 0.25 g/ml (HaCaT and A431) or 0.4 g/ml (NHEK) Ln-5 in PBS. Cells (HaCaT and NHEK: 1.2 ϫ 10 5 cells/100 l/filter; A431: 6 ϫ 10 4 cells/100 l/filter) in migration medium, MM (culture medium without fetal calf serum or Single-Quots) were preincubated with mAbs, reagents, or vehicle for 30 min at room temperature before plating. Each experiment was done at least three times, and results are expressed as mean Ϯ S.D. of relative cell migration with non-stimulated cells set as 1.
Adhesion assays were performed as described in Higgins et al. (24) with minor modifications. Microtiter 96-well plates were coated with 5 ng/well (HaCaT), 7 ng/well (A431), or 10 ng/well (NHEK) E-cadherin-Fc protein in TBS buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM CaCl 2 ) overnight at 4°C. Then, wells were blocked with 5% dry milk in TBS for 2 h at 37°C. Cells were detached and washed twice in adhesion buffer (20 mM Hepes, 137 mM NaCl, 3 mM KCl, 1 mM CaCl 2 , 1 mM MgCl 2 , 0.1% bovine serum albumin). Next, cells (7.5 ϫ 10 4 HaCaT/ NHEK cells/100 l/well; 5 ϫ 10 4 A431 cells/100 l/well) were preincubated with mAbs for 30 min at room temperature before plating in wells, which were washed twice with adhesion buffer after blocking. for 20 min at 37°C, HaCaT cells were plated on coverslips coated with anti-integrin mAbs, as described in Fig. 1. After 5 h at 37°C, cells were subjected to actin staining, using rhodamine-phalloidin. Inhibitors, but not control, prevent the ␣6␤4-initiated colony organization (compare with Fig. 1). One representative experiment out of three is shown. B, quantification of the assay shown in A. Single cells and doublets were counted in 12 microscopic fields per slide. Results are expressed as mean Ϯ S.E. (n ϭ 12). Black bars represent cells treated with vehicle (Me 2 SO, control), white and gray bars represent cells preincubated with erbB2 inhibitor and PI3K inhibitor, respectively. One representative experiment out of three is shown (bar, 150 m). * Significantly different values, p Ͻ 0.05, Student's t test. ** Significantly different value, p Ͻ 0.1, Student's t test. g/ml) for HaCaT, 35% Percoll for A431, or 40% Percoll for NHEK containing 9 mg/ml NaCl) were added to each well. Adherent cells were fixed for 15 min with 50 l/well of 25% glutaraldehyde, washed with PBS and stained with crystal violet (0.5% in 20% MeOH) for 10 min. Excess dye was washed off with water, cells were solubilized in 1% SDS, and absorbance was measured at 595 nm. Bars represent mean percentage of control adhesion Ϯ S.D. (n ϭ 3) with non-treated control cells set at 100%. All values have had background subtracted, which represents cell adhesion to wells blocked with milk. Aggregation Assays-Microtiter 96-well plates were coated overnight at 4°C with 5% dry milk in PBS. HaCaT cells (1.2 ϫ 10 5 cells/100 l/well) in MM were preincubated with AG825, AG1478, or vehicle (Me 2 SO) for 15 min at room temperature, followed by mAbs (TS2/16, 40 g/ml; GoH3, 30 g/ml; AA3, 50 g/ml) for an additional 15 min. Cells were seeded and immediately processed to evaluate the number of isolated cells at time 0, or they were allowed to aggregate at 37°C for 2 h. Then, 150 l of PBS was added, and the cells were passed three times through a 1000-l pipette tip and then transferred to a 48-well plate containing 250 l of 6% paraformaldehyde in PBS. Photographs of six fields in each well were taken and cells were counted. Aggregation was quantified using the index (N 0 Ϫ N t )/N 0 . N 0 is the number of isolated cells at time 0 and N t the number of isolated cells after 2 h. Each experiments was done at least three times.
Immunoprecipitations-Cell culture dishes (6 cm) were coated with 1 g of Ln-5/dish in PBS at 4°C overnight and then prewarmed to 37°C for 1 h. HaCaT cells (2 ϫ 10 6 ) in MM were preincubated with mAbs for 30 min at 37°C in suspension before plating onto ligand-coated dishes.
(If cells were co-stimulated with TS2/16 and a second antibody, the latter was added 10 min before TS2/16.) After 5 h at 37°C, attached cells were rinsed in PBS and lysed for 1 h on ice in 0.5 ml of lysis buffer containing 40 mM Tris, pH 7.5, 150 mM NaCl, 0.5% Triton X-100, 6 mM EDTA, 100 mM NaF, 1 mM sodium orthovanadate, 10 mM sodium pyrophosphate, 1 mM phenylmethylsulfonyl fluoride, 10 g/ml leupeptin, and 1 tablet/50 ml of complete protease inhibitor mixture (Roche Applied Science). Lysates were centrifuged at maximum speed for 10 min, and mAbs were added to the supernatant for 3 h at 4°C. Then, protein-mAb complexes were collected with Protein-G-Sepharose (Amersham Biosciences) for 1 h at 4°C. Complexes were washed three times with ice-cold lysis buffer before boiling in SDS-PAGE loading buffer. Proteins were separated on SDS-PAGE gels and blotted, and blots were incubated with the primary antibodies diluted 1:2000 in 5% milk in PBS containing 0.1% Tween 20 overnight at 4°C. Each sample was divided in two and analyzed for ␣-catenin and ␤-catenin content or for E-cadherin and ␣-, ␤-, ␥-, or p120 catenin content, using the ECLplus system (Amersham Biosciences) and a STORM 860 fluorometer. Production of GST-PBD and GTPase Assays-Production of GST-PBD protein and GTPase assays were performed as described in del Pozo et al. (25) with minor modifications. Detached cells were pretreated with mAbs as described in immunoprecipitation experiments and were seeded in Ln-5-coated dishes. After 5 h at 37°C, cells were chilled on ice, washed with ice-cold PBS, and lysed in buffer containing 50 mM Tris-HCl, pH 7.4, 100 mM NaCl, 10 mM MgCl 2 , 0.5% Triton X-100, 10% glycerol, 1 mM phenylmethylsulfonyl fluoride, 1 g/ml aprotinin, and 1 g/ml leupeptin. Lysates were then incubated with 20 g of FIG. 3. Antibodies to both ␣6␤4 and ␣3␤1 stimulate cell-cell aggregation. A, HaCaT cells were stimulated with antiintegrin mAbs as described in Fig. 1 and were allowed to aggregate for 2 h at 37°C. Then, cells were processed as described, photographs of six microscopic fields per well were taken and cells were counted. Results are expressed as mean Ϯ S.D. (n ϭ 6). One representative experiment out of four is shown. B, cells were pretreated with vehicle (Me 2 SO, control), erbB1-specific inhibitor AG1478 (250 nM) or erbB2-specific inhibitor AG825 (100 nM) before stimulation with anti-integrin mAbs and subjection to aggregation assays. One representative experiment out of three is shown. C, mock-transfected (control) cells or cells expressing dominant-negative erbB2 (HER2VEK753A) or dominant-negative PI3K regulatory subunit p85⌬iSH2-N were subjected to aggregation assays. One representative experiment out of three is shown. D, immunoblotting of equal amounts of total cell lysates with anti-erbB2 and anti-p85 antibody, respectively, to confirm expression of infected cDNAs in C.
GST-PBD-glutathione beads (Amersham Biosciences) for 30 min at 4°C. Bound protein was washed three times with lysis buffer and eluted with SDS sample buffer. Bound cdc42 or rac1 was detected by Western blotting using an anti-cdc42 or anti-rac1 mAb (BD Transduction Laboratories). Total lysates were also analyzed for the presence of cdc42 or rac1 for normalization.

Stimulation of ␣6␤4 Establishes Cell Morphology and Cell-
Cell Contact Formation Similar to Ln-5-Given the role of BM components in establishing an epithelial phenotype, e.g. the formation of cohesive sheets, cells should form better organized, tighter monolayers when plated on Ln-5 than on other, non-BM, ECM components. To test this, we plated HaCaT keratinocytes on Ln-5 and on fibronectin and stained the actin cytoskeleton (Fig. 1A). On Ln-5, cells were neatly positioned next to each other, forming a cobblestone-like monolayer with well developed cell-cell contacts. Actin was generally organized as a cortical ring. On fibronectin, cell boundaries were less well pronounced, and in many instances cells overlapped each other. Actin formed a robust network of stress fibers. Thus, Ln-5 and fibronectin induce distinct cell morphologies in HaCaT.
Ligation of ␣6␤4 Induces the Formation of Larger Cell Clusters Than Ligation of ␣3␤1, and This Process Needs Active erbB2 and PI3K-Looking at larger numbers of cells plated on anti-integrin mAbs, it became apparent that anti-␣6 or anti-␤4 mAbs, alone or together with anti-␤1 mAb, induced cells to form clusters, which we term colonies, that contained more cells than colonies on anti-␤1 mAb only ( Fig. 2A). In an attempt to quantify this phenomenon, we counted single cells and doublets per microscopic field, assuming that a decrease in the number of single cells and doublets is the result of an increase in colony size, if the total number of plated cells is constant. Such analysis revealed that, when plated on anti-␤1 mAb, 54.33 Ϯ 3.38 cells existed as single cells or doublets, whereas the remaining cells formed larger colonies. This number decreased to 29.00 Ϯ 2.33 and 18.33 Ϯ 1.37 single cells or doublets on anti-␣6 mAb and anti-␤4 mAb, respectively, and was even lower when plated simultaneously on mAbs to ␣6 and ␤1 (15.00 Ϯ 1.38) or mAbs to ␤4 and ␤1 (12.56 Ϯ 0.84) (Fig. 2B). In summary, ␣6␤4 is the major Ln-5-binding integrin that triggers cell-cell adhesion.
Preincubating cells with inhibitors of erbB2 or PI3K had no effect on the number of single cells and doublets on the anti-␤1 mAb (Fig. 2, A and B). However, these inhibitors reduced the size and number of colonies on anti-␣6 or anti-␤4 mAbs, whether or not anti-␤1 mAb was present, as judged by an increase in the number of single cells and doublets compared with each Me 2 SO control (Fig. 2, A and B). These data indicate that clustering of ␣6␤4 triggers an increase in cell-cell adhesion, which results in the formation of larger colonies and a decrease in the number of single cells or doublets. This event depends on active erbB2 and PI3K, because inhibition of these kinases increases the number of single cells and doublets, an indication of a decrease in cell-cell adhesion.

Ligation of ␣6␤4 Triggers Stronger Cell-Cell Aggregation
Than Ligation of ␣3␤1, and This Process Depends on Active erbB2 and PI3K-To further test a potential role of ␣6␤4 in cell-cell adhesion, we performed aggregation assays (Fig. 3). Anti-␣6 and anti-␤4 mAbs, alone or together with anti-␣3 mAb, induced HaCaT aggregation that was stronger than aggregation in the presence of anti-␣3 mAb alone (Fig. 3A). Similar results were found when, instead of anti-␣3 mAb, anti-␤1 mAb was used (data not shown). Integrin ␣6␤4-dependent stimulation needed active erbB2 but not erbB1, because specific inhibition of the former but not the latter receptor tyrosine kinase abolished the stimulatory effect of anti-␣6 and anti-␤4 mAbs on aggregation (Fig. 3B). Furthermore, overexpression of the dominant-negative erbB2 variant HER2VEK753A or the dominant-negative regulatory PI3K domain p85⌬iSH2-N (6) blocked the increase in cell-cell aggregation triggered by ␣6␤4 ligation but had no effect on ␣3␤1-induced aggregation (Fig. 3C). These results suggest that activation of both Ln-5-binding integrins ␣3␤1 and ␣6␤4 can stimulate aggregation of cells, however, with ␣6␤4 showing a more than 2-fold higher potential. Thus, similar to increased cell cluster formation on anti-␣6␤4 mAbs (Fig. 2), ␣6␤4-stimulated cell-cell aggregation in suspension depends on active erbB2 and PI3K.
Activation of Both ␣6␤4 and ␣3␤1 Stimulates E-cadherin Homophilic Interactions-Because cell-cell adhesion in epithelial cells is initially controlled by AJs (18), we next analyzed whether ␣6␤4 clustering can stimulate E-cadherin homophilic interactions. To this end, we tested primary normal human

FIG. 5. Clustering of ␣6␤4 integrin stimulates localization of E-cadherin at cell-cell contact sites and this effect is blocked by either erbB2 or PI3K inhibitors.
HaCaT cells were cultured in fetal calf serum-containing, complete medium until subconfluent (A) or were pretreated with vehicle or inhibitors, as described in Fig. 2, before plating on coverslips coated with Ln-5 (B) or mAbs (C) (anti-␣3, P1B5, 20 g/ml; anti-␤1, TS2/16, 40 g/ml; anti-␣6, GoH3, 30 g/ml; anti-␤4, AA3, 50 g/ml). After incubation for 5 h at 37°C, E-cadherin and actin were stained, using DECMA-1-AlexaFluor488 (green) and rhodamine-phalloidin (red), respectively. In cells on anti-integrin mAbs, E-cadherin localization at cell-cell contact sites was more pronounced on anti-␣6␤4 mAbs than on anti-␣3␤1 mAbs, indicating that the Ln-5 receptor responsible for that same effect on Ln-5 substrate is ␣6␤4, rather than ␣3␤1. The effect of ␣6␤4 but not ␣3␤1 clustering was blocked by erbB2 and PI3K inhibitors (bar, 50 m). epidermal keratinocytes (NHEKs) (Fig. 4A), HaCaT (Fig. 4B), and the epidermoid carcinoma cell line A431 (Fig. 4C) in adhesion assays on a recombinant E-cadherin-Fc protein substrate. In this assay, non-treated cells showed basal adhesion activity (set at 100%), which was entirely E-cadherin-mediated and not due to binding via Fc receptor or ECM secreted by these cells, because (i) adhesion was completely prevented in the presence of blocking anti-E-cadherin mAb DECMA-1, (ii) cells did not bind to wells coated with control IgG (data not shown), and (iii) cells did not adhere to wells coated with milk only (data not shown). When cells were stimulated with anti-integrin mAbs, they showed an increase in adhesion with mAbs to ␣6 and ␤4 and also with the anti-␤1 mAb and the anti-␣3 mAb A3X8. Anti-␣3 mAb P1B5, which inhibits ␣3␤1-Ln-5 interaction, and anti-␣5 mAb KH72, which blocks adhesion to fibronectin, had no stimulatory but an inhibitory effect on basal E-cadherin adhesion activity, indicating mAb specificity. Thus, in this experimental setup, clustering of both Ln-5-binding integrins ␣3␤1 and ␣6␤4 can increase E-cadherin homophilic interactions in the three cell types tested. An involvement of integrin ␣3␤1 in mediating cell-cell adhesion has been demonstrated before (27)(28)(29). However, these are the first results indicating that AJs play a role in this process.

Anti-␣6␤4 mAbs Stimulate E-cadherin Localization at Cell-Cell Contact Sites in an erbB2/PI3K-dependent Manner-To
confirm the stimulatory effect of ␣3␤1 and ␣6␤4 clustering on AJs formation, we then analyzed the localization of E-cadherin in HaCaT cells plated on Ln-5 or on anti-integrin mAbs, using the anti-E-cadherin mAb DECMA-1 conjugated with Alexa Fluor488 dye. Subconfluent cells grown for several days in complete culture medium were used as staining control. In such cells, E-cadherin was typically localized at cell-cell contact sites (Fig. 5A). A similar pattern was seen to be developing in cells cultured for 5 h on Ln-5 (Fig. 5B). When plated on antiintegrin mAbs, E-cadherin staining at the cell membrane was rather weak on anti-␣3 mAb P1B5, whereas increased staining was observed on anti-␤1 mAb (Fig. 5C). These results were consistent with the previous finding that TS2/16 but not P1B5 can stimulate E-cadherin homophilic interactions (Fig. 4). However, E-cadherin localization at cell-cell contact sites was considerably increased on mAbs to ␣6 and ␤4, alone or in combination with anti-␤1 mAb, suggesting that activation of ␣6␤4 can trigger more pronounced AJs formation than clustering of ␤1 integrins. Thus, the stimulatory effect of Ln-5 on cell-cell adhesion may be mediated by ␣6␤4 rather than ␣3␤1. Notably, inhibition of erbB2 and PI3K reduced E-cadherin staining at cell-cell contact sites in cells seeded on mAbs to ␣6 and ␤4, whereas it had no blocking effect in cells plated on anti-␣3 or anti-␤1 mAbs. These data underscore the importance of erbB2 as signaling adapter in ␣6␤4-dependent pathways. To test whether the observed mechanism is also valid in primary cells and is not specific for an immortal cell line only, we repeated the described experiments, using NHEK cells. E-cadherin staining in these cells was overall much more intense than in HaCaT and A431 (data not shown), matching the observation that immortalized and transformed cells tend to lose E-cadherin expression (30,31). Like HaCaT, NHEKs formed a cobblestone-like monolayer on Ln-5 (Fig. 6A). Furthermore, cell-cell contact formation was stronger on mAbs to ␣6 and ␤4 than on Ln-5 or on anti-␣3 or anti-␤1 mAbs. Anti-␤1 mAb did not down-regulate the effect of the anti-␣6/anti-␤4 mAbs (Fig. 6B). Preincubation with inhibitors of erbB2 or PI-3K blocked the stimulatory effect of anti-␣6␤4 mAb clustering but showed no influence on cell-cell adhesion on mAbs to ␣3 or ␤1. These data further support the concept that, in keratinocytes, ␣6␤4 clustering can stimulate cell-cell adhesion and that erbB2 is an adaptor molecule necessary to trigger this cellular behavior.
Stimulation of ␣6␤4 Promotes AJ Formation in an erbB2/ PI3K-dependent Manner-To further characterize the increase in cell-cell adhesion triggered by ␣6␤4 stimulation, we analyzed formation of AJs by co-immunoprecipitation experiments. HaCaT cells were preincubated with Me 2 SO (control) or PI3K inhibitor and were plated on Ln-5 in the absence or presence of anti-integrin mAbs. Then, E-cadherin was immunoprecipitated and the amount of co-precipitated ␤-, ␥-, and p120 catenin, respectively, was assessed by Western blotting. Complex formation between E-cadherin and ␤-catenin or ␥-catenin was only weakly up-regulated and was most prominent when cells were co-stimulated with anti-␣6 and anti-␤1 mAbs (Fig. 7, A   FIG. 7. Integrin ␣6␤4 clustering induces only minor changes in ␤and ␥-catenin interaction with E-cadherin. HaCaT cells were preincubated with Me 2 SO (control) or PI3K inhibitor LY294002 (20 M) for 10 min at 37°C. Then, anti-integrin mAbs were added as described in Fig. 1. After 30 min at 37°C, cells were seeded in Ln-5 coated dishes (1 g/dish), and were incubated at 37°C for 5 h. Then, cells were lysed and lysates subjected to immunoprecipitations. Equal amounts of precipitated E-cadherin were analyzed by Western blotting, using antibodies to E-cadherin, ␤-catenin (A) or ␥-catenin (B). Ratios of precipitated ␤-catenin to E-cadherin (A) and ␥-catenin to E-cadherin (B) is quantified in the bar graph. Mean Ϯ S.E. (n ϭ 6) of relative ␤-catenin or ␥-catenin pull-down and one representative experiment are shown. and B). Importantly, the increase in E-cadherin-catenin interaction was inhibited when PI3K was blocked. There was no detectable change in complex formation between E-cadherin and p120 (data not shown). The rather low effect of ␣6␤4 ligation on association between ␤and ␥-catenin with E-cadherin suggested different molecules as targets to regulate ␣6␤4dependent AJ formation. Kaibuchi et al. (32) reported that cell-cell adhesion can be controlled on the level of ␣-catenin/␤catenin interaction without changing the stoichiometry of the ␤-catenin-E-cadherin complex. This may be the regulatory mechanism active in HaCaT as stimulation with mAbs to ␣6 or ␤4, alone or in combination with anti-␤1 mAb, increased the amount of ␣-catenin associated with ␤-catenin (Fig. 8A). The observed up-regulation of ␣-catenin-␤-catenin complex formation may be an indication of increased cell-cell adhesion. Anti-␤1 mAb alone had only a weak effect. Again, the stimulatory potential of the anti-␣6 and anti-␤4 mAbs was blocked when erbB2 or PI3K were inhibited. Importantly, ␣6␤4 ligation also up-regulated ␣-catenin-E-cadherin complex formation in a PI3Kdependent manner (Fig. 8B). Based on these results we conclude that ␣6␤4 clustering on Ln-5 triggers erbB2 and PI3K activation, followed by increased association between ␣-catenin and the ␤-catenin-E-cadherin complex. This increase may be an indication of stronger cell-cell adhesion, because it can result in enhanced association of the cadherin-catenin complex with the actin cytoskeleton.
Integrin ␣6␤4-controlled Inhibition of ␣3␤1-Mediated Haptotactic Migration on Ln-5 Is Dependent on E-cadherin-We previously demonstrated that keratinocyte haptotactic migration on Ln-5 is mediated by integrin ␣3␤1 and that this motility is down-regulated by ␣6␤4. Inhibition of migration by ␣6␤4 was dependent on erbB2 activation, followed by PI3K stimulation (6). To test whether an increase in E-cadherin-controlled cellcell adhesion was the reason for the ␣6␤4-triggered decrease in haptotaxis, we performed Transwell migration assays in the presence of the blocking anti-E-cadherin mAb DECMA-1 (Fig.  9). Under control conditions, NHEK, HaCaT, and A431 cells showed basal migration activity on Ln-5, which was further increased in the presence of the stimulatory anti-␤1 mAb. As expected (6), this migration was inhibited by mAb clustering of ␣6␤4 (Fig. 9). When cells were preincubated with DECMA-1, basal migration was not affected and the cells responded to the anti-␤-1 mAb with an increase in motility. However, the blocking effect of the anti-␣6 and anti-␤4 mAbs on anti-␤1 mAbinduced motility was abolished (Fig. 9). These findings corroborate our conclusion that integrin ␣6␤4, which is known to FIG. 8. Integrin ␣6␤4 clustering upregulates AJ formation on the level of ␣-catenin/E-cadherin interaction and in an erbB2/PI3K-dependent manner. HaCaT cells were preincubated with Me 2 SO (control), erbB2 inhibitor AG825 (100 nM) or PI3K inhibitor LY294002 (20 M) for 10 min at 37°C. Then, anti-integrin mAbs were added (anti-␤1, TS2/16, 40 g/ml; anti-␤4, AA3, 50 g/ml, anti-␣6, GoH3, 30 g/ml) as described in Fig. 1. After 30 min at 37°C, cells were seeded in Ln-5 coated dishes (1 g/dish), and were incubated at 37°C for 5 h. Then, cells were lysed and lysates subjected to immunoprecipitations. A, equal amounts of precipitated ␤-catenin were analyzed by Western blotting, using antibodies to ␣-catenin or ␤-catenin. Ratio of precipitated ␣-catenin to ␤-catenin is quantified in the bar graph. Mean Ϯ S.E. (n ϭ 7) of relative ␣-catenin pulldown and one representative experiment are depicted. B, equal amounts of precipitated E-cadherin were analyzed by Western blotting, using antibodies to E-cadherin or ␣-catenin. Ratio of precipitated ␣-catenin to E-cadherin is quantified in the bar graph. Mean Ϯ S.E. (n ϭ 4) of relative ␣-catenin pull-down and one representative experiment are shown. stimulate strong cell-ECM adhesion, also signals up-regulation of E-cadherin-mediated cell-cell adhesion in keratinocytes, resulting in down-regulation of cell migration on Ln-5 and incorporation of cells into a cohesive, quiescent sheet.
The Small GTPase cdc42 Is Involved in the ␣6␤4-controlled Inhibition of Haptotactic Migration on Ln-5-AJ formation can be controlled by the small GTPases cdc42 and rac1 (18,32). Therefore, we analyzed cdc42 and rac1 activity, using the GST-PBD pull-down assay to differentiate active from non-active forms of these enzymes. HaCaT cells were preincubated with or without PI3K inhibitor and anti-integrin mAbs and were plated on Ln-5. Active cdc42 and rac1 were precipitated with the GST-PBD fusion protein and were quantified by Western blotting. Stimulation with anti-␣6 mAb, but not with anti-␤1 mAb, triggered an increase in cdc42 activity, which was blocked in the presence of the PI3K inhibitor (Fig. 10A). In contrast, rac1 did not show any ␣6␤4/PI3K-specific differences in activity (data not shown). These results suggest that clustering of ␣6␤4 leads to activation of cdc42 downstream of PI3K stimulation. An increase in cdc42 activity might be the reason for increased cell-cell adhesion and the resulting decrease in cell motility. To confirm that cdc42 is involved in the ␣6␤4-contolled down-regulation of ␣3␤1mediated haptotaxis on Ln-5, dominant-negative or constitutively active cdc42 variants were transiently overexpressed in HaCaT cells, using a retroviral system, and transfectants were analyzed in Transwell migration assays. Mock transfected cells responded to the different anti-integrin mAbs like wild type cells (Fig. 10B). However, cells overexpressing constitutively active cdc42(V12) did not respond to the stimulatory effect of the anti-␤1 mAb. Thus, high cdc42 activity prevents ␣3␤1-mediated HaCaT migration on Ln-5. In contrast, cells overexpressing dominant-negative cdc42(N17) showed an increase in basal migration activity, which was as high as anti-␤1 mAb-induced motility, and was not blocked by the anti-␣6 mAb. This finding indicates that ␣6␤4 can block ␣3␤1-controlled migration only if functional cdc42 is present. Next, stable transfectants were plated on the different anti-integrin mAbs and the number of single cells and doublets was assessed. When constitutively active cdc42(V12) was overexpressed, the number of single cells and doublets was slightly lower than in control cells (Fig. 10C). However, overexpression of dominant-negative cdc42(N17) led to an increase in the number of single cells and doublets when compared with each Me 2 SO control, indicating that active cdc42 is necessary to trigger the ␣6␤4-controlled increase in cell-cell adhesion (Fig. 10C). Taken together, our results demonstrate that activation of cdc42 downstream of the ␣6␤4-erbB2 complex and PI3K increases cell-cell adhesion, resulting in down-regulation of ␣3␤1-mediated haptotaxis on Ln-5.

DISCUSSION
This study provides evidence that integrin ␣6␤4 can upregulate E-cadherin-mediated cell-cell adhesion, which in turn inhibits ␣3␤1-controlled keratinocyte haptotaxis on Ln-5. This is a new strategy by ␣6␤4 to down-regulate cell motility that complements its ability to firmly anchor epithelial cells to the underlying BM by HDs. Together, these two mechanisms may allow the stable integration of cells into multicellular tissues, because it is necessary to build epithelial sheets (8). This phenomenon may be valid in keratinocytes in general, because we made the same observation in primary keratinocytes, HaCaT spontaneously immortalized keratinocytes, and in A431 epidermoid carcinoma cells.
Inhibition of ␣3␤1-controlled migration on Ln-5 is dependent on ␣6␤4-mediated activation of erbB2 and PI3K (6). Therefore, we expected that the ␣6␤4-induced increase in cell-cell adhesion, which inhibited motility, would also be mediated by an increase in erbB2 and PI3K activity. Our data support this concept, because erbB2 and PI3K inhibitors blocked the ␣6␤4-triggered formation of larger cell colonies and aggregates, the increased E-cadherin staining at cell-cell boundaries, and the formation of AJs. These findings further underscore the significance of erbB2 as an adaptor molecule in ␣6␤4-downstream signaling. Importantly, no soluble erbB ligand was necessary, but ␣6␤4 clustering was sufficient to activate erbB2 (6). Cooperation between ␣6␤4 and erbB2 may be a hallmark of quiescent cells to efficiently promote adhesion-dependent regulation of epithelial growth and survival versus cell migration during wound healing. PI3K may be involved in both inhibition and stimulation of cell motility (33,34). In Madin-Darby canine kidney cells, migration on laminin-1 was inhibited due to a PI3K-dependent increase in AJ formation. However, on collagens, motility was up-regulated and this effect was again PI3K-driven (34). Distinction between these substratedependent responses seemed to occur downstream of PI3K, because they were dependent on the intracellular localization of the FIG. 9. Anti-E-cadherin antibodies reverse the inhibitory effect of antibody-mediated ␣6␤4 clustering on ␣3␤1-dependent Ln-5 haptotactic migration. Cell migration assays were performed with NHEK (A), HaCaT (B) or with A431 (C) cells in Transwell chambers coated with Ln-5. Cells in serum-free culture medium were incubated in suspension at room temperature for 10 min with DECMA-1 (50 g/ml). Control cells were without DECMA-1. Then, mAbs to ␣6 and ␤4 (GoH3, 30 g/ml; AA3, 40 g/ml) were added, and 10 min later anti-␤1 mAb (TS2/16, 40 g/ml) was added. After another 10 min incubation, aliquots were seeded and the chambers were incubated at 37°C. All mAbs were present in both chambers. Five hours later, cells were fixed and stained. Migration was quantified by counting cells migrated through filters (eight microscopic fields on each of two filters for each condition). Results are expressed as mean Ϯ S.D. (n ϭ 3) of relative cell migration with non-stimulated cells set at 1.
small GTPase rac1 (34). We also analyzed cdc42 and rac1 as possible PI3K effectors and detected an increase in cdc42 but not rac1 activity upon ␣6␤4 stimulation, which was blocked in the presence of the PI3K inhibitor. Our data with dominant-negative and constitutively active cdc42 further demonstrated that cdc42 activation resulted in increased cell-cell adhesion and subsequent down-regulation of ␣3␤1-mediated motility on Ln-5.
Surprisingly, anti-␣3␤1 mAbs also stimulated aggregation and adhesion to the E-cadherin-Fc fusion protein. Previous studies suggested that ␤1 integrins can mediate cell-cell adhesion via homophilic or heterophilic integrin-integrin interactions involving ␣2␤1, ␣3␤1, and ␣4␤1 (27)(28)(29)35), whereas our data indicate the involvement of AJs. The role of ␤1 integrins in cell-cell adhesion was questioned, because a blocking anti-␤1 mAb did not interfere with intercellular junction organization (36). Studies in multicellular aggregates plated on Ln-5 demonstrated that activation of ␣3␤1 can lead to cell dissociation and spreading (37). In contrast, experiments with melanoma explants support a role of ␤1 integrin in cell-cell adhesion, which is not in contradiction to a migratory phenotype but is even a prerequisite, if cells need to move collectively, as blocking ␤1 integrin initiated a transition from sheet migration to single-cell movement (38). The initial phase of Transwell migration may be predominantly a mimic for single-cell migration, because cells are introduced into the assay as a monocellular suspension. Later on, cells may start to form ␣3␤1-controlled AJs, but they remain mobile, because sheet migration may now dominate. In contrast to ␣3␤1, ␣6␤4 is involved in blocking rather than promoting migration, as previously shown in primary human keratinocytes (10,39), HaCaT cells (6) and in ␤4-deficient PA-JEB keratinocytes, which were rescued by introducing a ␤4 construct (40). Importantly, Geuijen and Sonnenberg (40) showed that ␣6␤4-triggered motility inhibition was not simply due to increased adhesion to Ln-5 but was a result of increased stability of the interaction between ␣6␤4 and Ln-5. Thus, ␣6␤4-triggered simultaneous up-regulation of both cell-ECM and cell-cell adhesion may result in an adhesive, non-migratory phenotype and may explain the observation that Ln-5 can guide keratinocytes to switch from a migratory-to a dormant, integrated epithelial phenotype (11,41). Concerted substrate-and intercellular adhesion regulation is a prerequisite for tissue rearrangement during morphogenesis or remodeling (42).
A function of ␣6␤4 as motility inhibitor is in apparent contradiction to published literature, which proposes a pro-migratory potential of this integrin. Indeed, overexpression of ␤4 in breast cancer cells normally lacking ␣6␤4 led to increased ECM-independent chemotaxis and invasion and a functionblocking anti-␣6␤4 mAb inhibited migration of colon carcinoma cells on laminin-1 (14). Epidermal growth factor-induced migration was the result of HDs down-regulation (43) and was even actively promoted by ␣6␤4 due to rac1 stimulation (44). Increased ␣6␤4 expression has been detected in squamous carcinoma of lung, head, and neck (45,46), and de novo synthesis FIG. 10. Stimulation of AJ formation by integrin ␣6␤4 requires cdc42 activity. A, HaCaT cells were preincubated with Me 2 SO (control), PI3K inhibitor LY294002 (20 M) and anti-integrin mAbs as described in Fig. 7. Total cell lysates were prepared and cdc42 activity assays were performed using a GST-PAK-CD fusion protein that selectively binds GTP-bound cdc42. Precipitated, active cdc42 was analyzed by Western blotting and levels compared with total levels of endogenous cdc42. Ratio of active to total cdc42 is quantified in the bar graph as mean Ϯ S.D. (n ϭ 3) of relative cdc42 activity. B, HaCaT cells infected with retrovirus encoding constitutively-active cdc42(V12) or dominant-negative cdc42-(N17) or mock transfectants were subjected to haptotactic migration assays on Ln-5 as described in Fig. 9. Data are mean values Ϯ S.D. of three independent experiments. Non-stimulated cells were set as 1. Inset panels on the right show expression of infected cDNAs (indicated underneath the panels) confirmed by Western blotting of equal amounts of total cell lysates with an anti-HA antibody. Arrow indicates cdc42-HA variants cdc42(V12) and cdc42(N17). (*) Nonspecific background signal visible in all samples analyzed. C, transfectants expressing constitutively-active cdc42(V12) or dominant-negative cdc42(N17) and control cells were plated on coverslips coated with anti-integrin mAbs. After eight hours at 37°C, the number of single cells and doublets was assessed as described in Fig. 2. Results are expressed as mean Ϯ S.E. (n ϭ 12) of relative number of single cells and doublets with cells plated on anti-␤1 mAb set at 1. One representative experiment out of three is shown. of ␣6␤4 in thyroid cells or suprabasal keratinocytes promoted tumorigenesis (47)(48)(49)(50), indicating a role of ␣6␤4 in cancer formation and dissemination. In contrast, overexpression of ␣6␤4 in ␤4 null keratinocytes, breast, and urinary bladder cancer cells led to down-regulation of cell motility (40,51,52). Furthermore, cancer specimen analyses showed loss or down-regulation of ␣6␤4 and/or Ln-5 in breast and prostate cancer tissue (51,(53)(54)(55)(56)(57)(58) and an inverse correlation between ␤4 levels and dissemination potential in gastric cancer (59). These results document an involvement of ␣6␤4 in the blockade of cell motility and transformation, supporting a normal cell phenotype. These data indicate that ␣6␤4 may both inhibit and support cell motility, depending on the type of motility (haptotaxis versus chemotaxis versus invasion), the microenvironment, and the cellular context. Because ␣6␤4 can associate with a multitude of proteins, e.g. erbB1 (43) and erbB2 (6,16), hepatocyte growth factor receptor Met (60), insulin receptor substrates 1 and 2 (14), or macrophage-stimulating protein receptor RON (61), which themselves are implicated in cell motility and cancer formation and progression (62)(63)(64)(65), cell-type-specific differences in protein levels of these receptors and their ligands may have profound effects on ␣6␤4 performance. In addition, ␣6␤4 function may be dramatically influenced by differences in the AJ complex composition. Therefore, in cells expressing low levels or no E-cadherin, e.g. in certain tumor cells (30, 31), ␣6␤4 may not be able to down-regulate cell motility.
Based on the data presented here and our published results (6), we suggest a working model for a signal transduction pathway activated by ␣6␤4 clustering that stimulates cell-cell adhesion and results in inhibition of ␣3␤1-controlled haptotaxis on Ln-5. We envision that this mechanism slows down cells and eventually leads to static adhesion on Ln-5. This model allows ␣6␤4 to control keratinocyte motility by two strategies: (i) by AJ-controlled cell-cell adhesion and/or followed by (ii) HD-mediated cell-ECM interaction. Such a tightly controlled mechanism may be of particular importance during the formation and maintenance of epithelial sheets, as cells simultaneously need to slow down and increase attachment to neighboring cells, e.g. when a wound is closed. Alternatively, downregulation of E-cadherin-mediated cell-cell adhesion is a prerequisite to become mobile in processes such as initiation of wound healing, embryonic development, and cancer invasion and metastasis (22). Our finding that the ␣6␤4-erbB2 complex can manipulate E-cadherin function identified these proteins as new targets in modulating cell aggregation and demonstrates that ␣6␤4 does more than just anchor cells to the BM.