Mammalian diaphanous-related formin Dia1 controls the organization of E-cadherin-mediated cell-cell junctions

It is well established that both the formation and maintenance of cell-cell junctions in epithelial cells depend on the actin cytoskeleton. Formation of cadherin-mediated cell-cell junctions (adherens junctions) is accompanied by profound changes in the actin cytoskeleton and accumulation of polymerized actin at the contact area (Bershadsky, 2004; Mege et al., 2006). Inhibition of actin assembly by treatment with pharmacological reagents such as latrunculin or cytochalasin prevents the formation and strengthening of cadherin-mediated adhesions in addition to disrupting those already in existence (Angres et al., 1996; Chu et al., 2004; Ivanov et al., 2005; Vasioukhin et al., 2000). However, the molecular mechanisms underlying the crosstalk between cadherin signaling and the actin cytoskeleton are not fully understood. In particular, the specific role of proteins regulating actin polymerization in the formation and maintenance of the adhesion contacts has yet to be elucidated.

The basic unit of the adherens junction is a cadherin-catenin complex, which consists of transmembrane adhesion receptors known as classic cadherins. These interact in a homophilic manner with the cadherins of neighboring cells. The cytoplasmic domains of classic cadherins interact directly with proteins of the armadillo family: p120 catenin by means of the juxtamembrane domain, and β-catenin or plakoglobin by means of the C-terminal domain. Both β-catenin and plakoglobin can bind to α-catenin, which is another essential element of the cadherin-catenin complex (for reviews, see Gumbiner, 2005; Perez-Moreno and Fuchs, 2006; Wheelock and Johnson, 2003). Monomeric α-catenin interacts with β-catenin or plakoglobin, whereas homodimers of α-catenin can bind to actin filaments and regulate their polymerization and crosslinking (Weis and Nelson, 2006). p120 catenin is another protein functioning at the interface between cadherin and actin, and this regulates Rho GTPases (Reynolds, 2007) and interacts with the actin-binding protein cortactin (Boguslavsky et al., 2007).

Regulators of actin nucleation appear to be crucial for the control of cadherin-mediated cell-cell adhesion. For example, the Arp2/3 complex, which promotes the formation of a branched actin filament network (Higgs and Pollard, 2001; Nicholson-Dykstra et al., 2005), localizes to adherens junctions and physically associates with the cadherin-catenin complex (Kovacs et al., 2002; Verma et al., 2004). Moreover, activators of Arp2/3 function, WAVE (Yamazaki et al., 2006), N-WASP (Ivanov et al., 2005) and cortactin (El Sayegh et al., 2004; Helwani et al., 2004), demonstrate a similar localization. Knockdowns of these components induce weakening of adherens junctions, exemplified by the reduced density of cadherin and its associated proteins.

Another group of molecules controlling actin polymerization comprises proteins of the formin family, which are characterized by the presence of highly conserved formin-homology 1 and 2 (FH1 and FH2) domains and are involved in nucleating the polymerization of linear actin filaments and promoting their elongation (Faix and Grosse, 2006; Goode and Eck, 2007; Higgs, 2005). Formin-1, which was historically the first representative of this group of proteins, is now known to be a partner for α-catenin (Kobielak et al., 2004). Formin-1 localizes to cadherin-mediated cell-cell junctions in an α-catenin-dependent manner; its association with the cadherin-catenin complex is necessary for the formation and maintenance of adherens junctions in keratinocytes (Kobielak et al., 2004).

The potential role in cadherin-mediated adhesion of Diaphanous-related formins is of particular interest as their activity is controlled by small G-proteins of the Rho family (Faix and Grosse, 2006; Higgs, 2005; Olson, 2003). In several cell types, Rho, Rac and Cdc42 are essential for, and activated by, cadherin-mediated cell-cell adhesion (Braga, 2002). Although the roles of Cdc42 and, especially, Rac can at least partially be related to the regulation of Arp2/3-mediated actin polymerization (Kraemer et al., 2007; Yamazaki et al., 2006), the mechanism underlying the apparent Rho dependence of adherens junctions (Braga et al., 1997; Sahai and Marshall, 2002) is still poorly understood. Diaphanous-related formin 1 (Dia1) is one of the Rho targets; RhoA binding triggers the transition of Dia1 from a closed, inactive conformation to an open, active conformation (Li and Higgs, 2003; Otomo et al., 2005; Rose et al., 2005). It has been reported that the disruption of adherens junctions induced by inhibition of Rho could be rescued by overexpression of a constitutively active form of Dia1 (Sahai and Marshall, 2002). This finding suggests that Dia1 participates in the Rho-dependent regulation of adherens junctions.

Here, an RNAi-mediated silencing approach was used to demonstrate that Dia1 is required for the integrity of adherens junctions. Moreover, we show that, upon activation by RhoA, Dia1 localizes to adherens junctions, which leads to their dramatic strengthening and enlargement. Our data therefore show that the formin protein Dia1 is an indispensable component and a potent regulator of cell-cell junctions in epithelial cells.

To investigate the role of Dia1 in cell-cell adhesions, we selected the MCF7 human breast cancer epithelial cell line. These cells preserve normal epithelial E-cadherin-mediated adherens junctions enriched in α- and β-catenins, actin and actin-associated proteins, including vasodilator-stimulated phosphoprotein (VASP), as well as apical tight junctions (van Deurs et al., 1987). Immunofluorescence staining with an antibody against Dia1 revealed a diffuse distribution of the endogenous protein over the cytoplasm, with some weak enrichment at the cell-cell contacts (Fig. 1A-C).

To analyze the role of Dia1 in the formation of cell-cell junctions, we transiently transfected cells with a pSuper vector (Brummelkamp et al., 2002) bearing short hairpin RNA sequences (shRNA) specific for the silencing of expression of human Dia1 (hDia1; si-hDia1 vector). To mark the transfected cells, we co-transfected the shRNA-encoding plasmid with the pGFP-C1 plasmid. Control cells were transfected with pGFP-C1 together with an empty pSuper vector. Expression of si-hDia1 led to a significant decrease of hDia1 in MCF7 cells, as revealed by western blot analysis of GFP-labelled cells (Fig. 1D) and immunofluorescence staining with an antibody against Dia1 (Fig. 1E,F). Densitometric measurement of the levels of Dia1 showed that less than 10% of the endogenous Dia1 remained in the cells expressing si-hDia1.

Comparison of E-cadherin staining in control and hDia1-knockdown MCF7 cells showed a dramatic decrease in E-cadherin localization at the shared cell borders, especially when both neighboring cells were depleted of Dia1 (compare Fig. 1H with K). In >80% of such cell pairs, the cell-cell junctions marked by E-cadherin staining were abnormal, that is either were not distinguishable at all or were discontinuous and less intensely stained than junctions in control cells (Fig. 1K,M). Intercellular contacts between Dia1-deficient and neighboring control cells preserved normal E-cadherin-staining in less than 60% of the cases (Fig. 1M). Among control cells (both non-transfected, or transfected with empty pSuper plasmid), more than 80% of junctions preserved a normal (uniformly bright) E-cadherin staining pattern (Fig. 1H,M). Differences in actin staining along contacting cell borders were, however, less pronounced (Fig. 1I,L).

To confirm that the phenotype observed in si-hDia1 cells is due to the suppression of the target protein, we performed rescue experiments (Fig. 2A-G). For such experiments, we co-transfected human MCF7 cells with the si-hDia1 vector and with a vector encoding GFP-fused mouse Dia1 (mDia1), the sequence of which contained three nucleotide mismatches compared with the human Dia1-shRNA sequence (Fig. 2A). The human shRNA sequence that knocked-down Dia1 in human cells did not affect expression of GFP-mDia1. Indeed, western blot analysis of the cell lysates with the non-species-specific antibody against Dia1 revealed that, although the amount of endogenous hDia1 decreased in the cells expressing si-hDia1, the amount of GFP-mDia1 did not change (Fig. 2B). Although transfection with the si-hDia1 vector alone abolished E-cadherin localization to cell-cell junctions, as in our previous experiments (Fig. 2C,D), the cells transfected with si-hDia1 and therefore lacking endogenous Dia1, but expressing GFP-mDia1, showed prominent localization of E-cadherin to cell-cell junctions (Fig. 2E,F). Scoring of the intact E-cadherin-positive contacts (Fig. 2G) confirmed that expression of GFP-mDia1 is sufficient to restore the junctional localization of E-cadherin in the cells transfected with the si-hDia1 vector, showing that loss of E-cadherin localization was indeed the result of downregulation of hDia1.

To assess quantitatively the changes in the junctional proteins upon Dia1 knockdown, we first compared total E-cadherin and β-catenin amounts in control and knockdown cells by western blotting. In these experiments, MCF7 cells stably transfected with the pSuper-retro-shLacZ (control) and the pSuper-retro-shDia1 vectors were analyzed (Fig. 3A,B). Densitometric analysis of the western blots revealed that, in spite of a >90% reduction in the hDia1 expression, the amounts of E-cadherin and β-catenin (normalized against the levels of α-tubulin in the corresponding samples) did not decrease (Fig. 3B).

To quantify the levels of E-cadherin and actin localized to cell-cell junctions, we scanned the fluorescence intensity along segments crossing the junctions in a perpendicular direction (small squares in Fig. 1H,K). In control cells, such scanning revealed narrow maxima of fluorescence (`peaks') corresponding to the junctions (Fig. 3C). For the analysis of the junctions in Dia1-knockdown cells, we chose only those fluorescence profiles on which the fluorescence peaks were distinguishable (Fig, 3D). Comparison of the fluorescence intensities corresponding to the peak values revealed that, even in these partially preserved junctions between Dia1-knockdown cells, the fluorescence intensity of E-cadherin was reduced on average by ∼50% in comparison with that of the control (Fig. 3E, black bars). At the same time, there were no significant differences between the average levels of cytoplasmic E-cadherin fluorescence between control and Dia1-knockdown cells (data not shown). A similar analysis, measuring the phalloidin fluorescence corresponding to peak values at the cell-cell interface, revealed a slight, albeit statistically significant, decrease of the junctional actin in Dia1-knockdown cells (Fig. 3E, gray bars). It should be noted that these measurements quantify junctional actin (Zhang et al., 2005), but not the actin of peripheral bundles sometimes found adjacent to the junctions.

To investigate how Dia1 knockdown affects the localization of other major components of adherens junctions, we performed immunofluorescence experiments with antibodies against α-catenin (Figs 4 and 5) and β-catenin (Fig. 4). We also assessed the localization of the actin-binding protein VASP, known to be an important modulator of actin organization at adherens junctions (Scott et al., 2006; Vasioukhin et al., 2000) (Fig. 5). As intercellular junctions are sites of high protein tyrosine kinase and phosphatase activity (Sallee et al., 2006; Volberg et al., 1991), we also probed the distribution of tyrosine-phosphorylated junctional components using an antibody detecting phosphotyrosine (Fig. 5). The stainings revealed that, as with E-cadherin localization, the localizations of all the aforementioned components were strongly diminished at the cell-cell interface of Dia1-knockdown cells in comparison with the junctions between control and knockout cells or between two control cells.

MCF7 cells can form tight junctions (van Deurs et al., 1987) and, as shown in supplementary material Fig. S1A-C, express and localize to the cell-cell junctions the protein cingulin, a bona fide component of tight junctions (Citi et al., 1988). Downregulation of mDia1 led to disappearance of the junctional cingulin staining from some, but not all, junctions between affected cells. In particular, some junctions lacking E-cadherin demonstrated well-preserved cingulin localization (supplementary material Fig. S1D-F). In some cases, however, the junctions between Dia1-knockdown cells demonstrated staining for neither E-cadherin nor cingulin (supplementary material Fig. S2A).

To reveal differences in the de novo formation of junctions between control and Dia1-deficient cells, the junctions were disrupted by Ca²⁺ depletion and then allowed to recover at the normal Ca²⁺ concentration (supplementary material Fig. S2). Non-transfected cells, or cells transfected with GFP and empty pSuper vector, restored the junctions almost completely in 30-60 minutes following the Ca²⁺ switch, as revealed by stainings for E-cadherin, actin and cingulin (supplementary material Fig. S2A,B). Dia1-knockdown cells were apparently more sensitive to Ca²⁺ depletion than control cells (supplementary material Fig. S2A). After the Ca²⁺ switch, these cells restored their spread morphology and showed F-actin accumulation at the cell-cell interfaces. Nevertheless, only a small fraction of the junctions between GFP-labelled si-hDia1-expressing cells demonstrated proper E-cadherin or cingulin localization even 2 hours after the Ca²⁺ switch (supplementary material Fig. S2A,C). Thus, Dia1 is required for the formation of junctional complexes typical of normal epithelial cells.

As Dia1 is required for the integrity of the adherens junctions, we investigated whether it localized to these structures. As mentioned above, staining of MCF7 cells with an antibody specific for Dia1 showed a diffuse distribution in the cytoplasm, with some weak enrichment at regions of cell-cell junctions (Fig. 1A-C). Transfection of MCF7 cells with a construct encoding full-length mouse Dia1 protein (GFP-mDia1) showed that full-length mDia1 was diffusely distributed in the whole cytoplasm and did not demonstrate any apparent enrichment at the cell-cell junctions (Fig. 6A). Neither the actin cytoskeleton nor the adherens junctions of the cells were affected by the overexpression of full-length mDia1, as revealed by staining for actin and β-catenin (Fig. 6B,C). However, when full-length mDia1 was coexpressed with a constitutively active RhoA protein (RhoA-V14), a significant enrichment of mDia1 was observed at the junctions between affected cells (Fig. 6D). As active RhoA triggers a transition of Dia1 from an inactive `closed' to an active `open' conformation (Li and Higgs, 2003; Otomo et al., 2005; Rose et al., 2005), this result demonstrates that the activation of the mDia1 molecule strongly enhances its affinity for the junctions.

To assess quantitatively the effect of RhoA-mediated activation on localization of GFP-mDia1, we measured the GFP fluorescence profiles along segments crossing the cell-cell interfaces visualized by E-cadherin or β-catenin immunofluorescence. We calculated a GFP-mDia1 localization index as the difference between fluorescence intensities at the cell-cell interface and in the adjacent cytoplasm, normalized to the fluorescence intensity value obtained for the cytoplasm (Fig. 6J). The difference between the distributions of GFP-mDia1 in control and RhoA-V14-expressing cells was dramatic. The junctional fluorescence in cells transfected with GFP-mDia1 alone was lower than the fluorescence in the cytoplasm, whereas, in the presence of active Rho, the fluorescence intensity of GFP-mDia1 at the junctions was significantly higher than in the cytoplasm (Fig. 6J). Interestingly, this effect was observed when both of the cells in contact were affected. At the interface between cells expressing GFP-mDia1 plus RhoA-V14 and control, non-transfected cells, the former produced numerous filopodial projections with GFP-mDia1 localized to their tips (Fig. 6D inset), but enrichment of junctional GFP-mDia1 was not observed.

As could be predicted from previous studies (Tominaga et al., 2000; Watanabe et al., 1999; Watanabe et al., 1997), the coexpression of GFP-mDia1 and RhoA-V14 triggered the formation of numerous bundles of actin filaments (Fig. 6E). Furthermore, the level of junctional F-actin increased at the contacts between such cells, in comparison with that of non-transfected cells or cells expressing only GFP-mDia1 [Fig. 6E (arrow) and 6K, gray bars]. Remarkably, coexpression of GFP-mDia1 and RhoA-V14 not only increased the junctional levels of GFP-mDia1 and F-actin but also strongly promoted accumulation of β-catenin at the cell-cell junctions (Fig. 6F). This result again was most apparent when both neighboring cells expressed GFP-mDia1 and RhoA-V14. β-Catenin staining at the interface between such cells was much stronger than at the contacts between cells expressing GFP-mDia1 only, or between two non-transfected cells [Fig. 6F (arrow) and 6K, black bars]. The junctional levels of E-cadherin (Fig. 7A,B) and α-catenin (not shown) were also dramatically augmented. Examination of the cells expressing GFP-mDia1 plus RhoA-V14 by confocal microscopy (Fig. 7A-E) revealed that the average height of the E-cadherin-positive junctions between such cells at the X-Z plane [Fig. 7C,D (arrows)] increased approximately twofold in comparison with the height of junctions between control cells (Fig. 7C-E). Thus, coexpression of exogenous mDia1 and constitutively active RhoA increases the width of the `adhesion belt' between the affected cells.

Not only exogenous GFP-mDia1 but also endogenous Dia1 translocates to the junctions upon expression of constitutively active RhoA-V14. In the RhoA-V14-transfected culture, cells expressing active Rho formed prominent actin bundles and demonstrated spectacular localization of endogenous hDia1 to the contact between affected cells (Fig. 6G,H). This was accompanied by some enhancement of cell-cell junctions, as revealed by staining for β-catenin (Fig. 6I and supplementary material Fig. S3C). The effect on β-catenin recruitment was, however, less prominent in cells transfected with RhoA-V14 alone than in the cells coexpressing RhoA-V14 with GFP-mDia1 (compare Fig. 6F with 6I and supplementary material Fig. S3C).

It is worth noting that expression of RhoA-V14 did not result in any enhancement of cell-cell junctions in the cells with downregulated endogenous hDia1 (supplementary material Fig. S3D-F). Assessment of the β-catenin-positive contacts between cells coexpressing GFP-RhoA-V14 and si-hDia1 revealed that the percentage of normal junctions between these cells was as low as that between cells expressing si-hDia1 only (supplementary material Fig. S3G). These results, therefore, show that Dia1 is a Rho target that is required for the stimulatory effect of Rho on the cell-cell adherens junctions.

Dia1 activation by itself was not sufficient for the localization of this protein to adherens junctions as truncated constitutively active forms of mDia1 (Fig. 8A) (Watanabe et al., 1999), mDia1-ΔN1, mDia1-ΔN2 and mDia1-ΔN3 did not localize to adherens junctions (Fig. 8B; and data not shown) and did not increase the levels of junctional actin and β-catenin (Fig. 8C,D). To analyze further the molecular requirements for the junctional localization of Dia1, we prepared mDia1 that lacked the C-terminal DAD domain but had an intact N-terminal region (Fig. 8A). Owing to the absence of the DAD domain, this mutant (CFP-mDia1-ΔDAD) maintains its open conformation with a freely accessible Rho-binding domain (Otomo et al., 2005; Rose et al., 2005). Its overexpression should therefore lead to Rho sequestration and produce a phenotype resembling that induced by inhibition of endogenous Rho. Indeed, when the level of CFP-mDia1-ΔDAD expression was high, the transfected cells had an altered morphology and the integrity of the cell-cell junctions was often disrupted in the same way as in cells transfected with dominant-negative Rho (data not shown). However, at moderate expression levels, when intact adherens junctions were still preserved, CFP-mDia1-ΔDAD clearly localized to cell-cell junctions (Fig. 8E).

Thus, the N-terminal region of mDia1 (residues 1-263) seems to be involved in the localization of Dia1 to the junctions. Surprisingly, coexpression of mDia1-ΔN3 together with constitutively active RhoA was sufficient to produce some mDia1-ΔN3 localization to the cell-cell junctions (Fig. 8F); however this was less prominent than the localization of CFP-mDia1-ΔDAD under the same conditions (Fig. 8G).

Taken together, these results suggest that not only activation but also the junctional localization of Dia1 is required for the augmentation of adherens junctions. Our data are consistent with the idea that the N-terminal region of Dia1 is important for the correct localization of the activated Dia1 to cell-cell junctions. In addition, we found evidence for an additional, N-terminus-independent, mechanism of Dia1 junctional localization that apparently depends on RhoA.

In view of the accumulating data on the involvement of microtubules in the regulation of cell-cell junctions (Ivanov et al., 2006; Mary et al., 2002; Stehbens et al., 2006; Waterman-Storer et al., 2000), we tested whether microtubules are important for the effect of Dia1 on adherens junctions. We added nocodazole to the cells co-transfected with RhoA-V14 and GFP-mDia1 four hours after transfection, so that expression of these proteins occurred in conditions of microtubule disruption. In control cells, with intact microtubules, expression of GFP-mDia1 and RhoA-V14 led to the localization of GFP-mDia1 to the cell-cell junctions and a significant enhancement of adhesion protein recruitment to these structures (Fig. 6D-F and Fig. 9A-C). Incubation with nocodazole was sufficient to disrupt the microtubule network of affected cells almost completely (compare Fig. 9A with D). In cells lacking microtubules, the localization of GFP-mDia1 to adherens junctions was less pronounced than in non-treated cells (compare Fig. 9B with E) as a significant fraction of GFP-mDia1 was often diffusely distributed across the cytoplasm (Fig. 9E). Nevertheless, even in these conditions, expression of GFP-mDia1 together with RhoA-V14 still significantly enhanced the junctional localization of β-catenin (Fig. 9F) as well as other junctional components (data not shown). Thus, mDia1 activated by Rho also strengthens adherens junctions in the absence of microtubules.

The activity of myosin II has been reported to be involved in the formation and maintenance of adherens junctions (Conti et al., 2004; Miyake et al., 2006; Shewan et al., 2005; Yamada and Nelson, 2007). To test the role of myosin II in the Dia1-Rho-induced enhancement of adherens junctions, we treated cells co-transfected with RhoA-V14 and GFP-mDia1 with blebbistatin, a specific inhibitor of nonmuscle myosin II (Straight et al., 2003). Blebbistatin-treated cells experienced the complete disappearance of actin stress fibers (compare Fig. 6E with Fig. 9G). At the same time, a highly exaggerated formation of actin-rich filopodia was observed at the interface between neighboring, contacting cells (Fig. 9G). GFP-mDia1 was mainly localized to these filopodia (Fig. 9H) and was sometimes concentrated in spots at the filopodia tips. Immunofluorescence visualization of E-cadherin revealed its enrichment in these filopodia-like structures at the cell-cell interface (Fig. 9I). Thus, inhibition of myosin II does not interfere with Dia1-induced filopodial activity, but instead blocks the strengthening of adherens junctions in the cells expressing Dia1 together with active RhoA.

We found that downregulation of Dia1 in human MCF7 breast carcinoma cells caused adherens junctions to disintegrate. In Dia1-knockdown cells, the levels of the proteins of the cadherin-catenin complex (E-cadherin, α-catenin, β-catenin, etc.) were significantly reduced at the adherens junctions. This effect was especially prominent when both contacting cells lacked Dia1. A normal partner can partially compensate for the deficiency of a knockdown cell; and a similar effect can be seen also when control and E-cadherin-knockdown cells contact each other (Capaldo and Macara, 2007). Intercellular junctions between Dia1-knockdown cells are more sensitive to Ca²⁺ depletion than those between control cells; moreover, a Ca²⁺ switch does not induce formation of proper cell-cell junctions between Dia1-knockdown cells, as revealed by staining with antibodies against E-cadherin and against a marker of tight junctions, cingulin. Before Ca²⁺ depletion, however, some junctions between Dia1-knockdown cells demonstrate a normal localization of cingulin, even when E-cadherin is completely absent. Combined, these results suggest that Dia1 knockdown interferes with the formation and maintenance of E-cadherin-mediated adherens junctions, whereas its effect on the localization of a tight junction protein is secondary.

Dia1 depletion is not accompanied by any decrease in the total amount of E-cadherin or β-catenin. Moreover, Dia1 knockdown in MCF7 cells did not produce severe changes in the overall organization of actin, and this is in agreement with several previous studies analyzing the effects of Dia1 knockdown in other cell types (Hotulainen and Lappalainen, 2006; Yamana et al., 2006). Careful quantification revealed, however, a small but statistically significant decrease of the junctional F-actin levels at the contacts between Dia1-knockdown cells; and the level of the actin-binding protein VASP decreased dramatically. Thus, Dia1 downregulation, most probably, results in a local effect in junctional actin organization rather than a systemic alteration of the whole actin cytoskeleton.

This conclusion is consistent with our observations that active RhoA induces massive translocation of both exogenous GFP-Dia1 and endogenous hDia1 to adherens junctions in MCF7 cells. The mechanism of Dia1 junctional localization, however, seems to be complex. We demonstrated that Dia1 activation by itself is not sufficient for its junctional localization as constitutively active Dia1 variants lacking the N-terminal domain (ΔN1, ΔN2 and ΔN3) fail to localize, whereas Dia1-ΔDAD, which preserves the N-terminal domain, does localize properly. These results are consistent with a recent study (Seth et al., 2006) showing that the N-terminal part of mDia1 is required for its membrane localization.

Interestingly, Dia1-ΔN3 does localize to the junctions if co-transfected with constitutively active RhoA, even though such localization was less prominent than the localization of constructs preserving the N-terminus. Thus, Dia1 localization to the adherens junctions seems to be driven by both N-terminus-dependent and N-terminus-independent interactions. The RhoA-induced localization of the Dia1-ΔN3 variant suggests that, upon activation by RhoA, certain protein(s) might bind Dia1 in an N-terminus-independent manner and so deliver it to the junctions. We hypothesize that the general membrane localization can be driven by an N-terminus-dependent mechanism, as suggested by Seth et al. (Seth et al., 2006), whereas an additional Rho-dependent mechanism drives the specific junctional localization. RhoA itself was reported to be enriched at cell-cell junctions in some experimental situations (Takaishi et al., 1995).

Whatever the mechanism of Dia1 junctional localization, the fact that such localization significantly enhances the junctions is most interesting. Indeed, recruitment of high levels of F-actin and bona fide adherens junction proteins into junctions between cells expressing GFP-mDia1 together with Rho-V14 even causes an increase in the width of the adhesion belt, as revealed by confocal microscopy. It is worth noting that the general reorganization of the actin cytoskeleton observed in these cells (development of numerous parallel actin filament bundles) is not sufficient for the enhancement of adherens junctions by itself. Augmented formation of actin bundles can be observed upon expression of constitutively active Dia1 lacking the N-terminal domain, and therefore nonlocalized to the junctions, or even by expression of active RhoA-V14 in the condition of Dia1 knockdown. In both cases, the formation of prominent actin bundles is not accompanied by enhancement of adherens junctions. Thus, the effect of Dia1 activation on adherens junctions (similar to the effect of Dia1 knockdown) is not a consequence of gross changes in actin polymerization, rather a result of local actin reorganization at the junctional areas.

Although the best-documented activities of Dia1 are the enhancement of actin nucleation and elongation (Kovar et al., 2006; Li and Higgs, 2003; Moseley et al., 2004; Romero et al., 2004), several in vivo studies suggest that Dia1 also affects microtubule stability and co-alignment (Goulimari et al., 2005; Ishizaki et al., 2001; Palazzo et al., 2001). We checked whether the stimulatory effect of Dia1 on adherens junctions depended on microtubules. We have shown that prolonged incubation with a microtubule-disrupting drug did indeed reduce (but did not entirely abolish) localization of Rho-activated mDia1 to the adherens junctions; the mechanism of this interesting effect certainly warrants further investigation. At the same time, microtubule disruption did not prevent augmentation of adherens junctions upon overexpression of mDia1 and Rho-V14. Therefore, these results do not exclude the idea that microtubules play some role in the delivery of Dia1 to the adherens junctions, but they do show that Dia1 does not affect adherens junctions by means of an involvement with microtubules.

However, the formation of Dia1-driven mature adherens junctions clearly requires the activity of myosin II. Although evidence for the involvement of myosin in the regulation of adherens junctions was found previously (Conti et al., 2004; Grosheva et al., 2006; Miyake et al., 2006; Shewan et al., 2005; Yamada and Nelson, 2007), the effect of myosin II inhibition we observed in our system (production of filopodia in the cell-cell junction area, instead of junction enhancement) was dramatic. It is possible that two pathways downstream of RhoA cooperate to induce junction maturation: one inducing actin polymerization by means of Dia1 and another inducing activation of myosin II by means of Rho kinase.

Several mechanisms based on Dia1-mediated local actin polymerization could potentially be important for the formation and maintenance of cadherin-mediated adhesions. The apparent stimulation of formation of filopodia by active mDia1, which resembles the well-documented effects of its homolog Dia2 (Pellegrin and Mellor, 2005; Peng et al., 2003; Schirenbeck et al., 2005; Wallar et al., 2006), might play a role at the early stages of adhesion, when contacts made through filopodia-like projections are important (Vasioukhin et al., 2000). It is worth noting that the formation of Dia1-induced filopodia occurred mainly at the interface between contacting cells, whereas few filopodia were directed to the substratum. The mechanism of such selectivity deserves further investigation.

Our experiments with inhibition of the activity of myosin II allowed the dissection of the two stages of Dia1-mediated adherens junction strengthening: the formation of filopodia, which is preserved or even exaggerated in conditions of myosin II inhibition, and the formation of mature linear junctions, for which cooperation with myosin II is required. The predicted ability of formins and, in particular, Dia1 to enhance actin polymerization upon application of mechanical force (Kozlov and Bershadsky, 2004) could play a role in the processes of myosin-II-driven force-induced assembly of cadherin adhesions.

At the later stages of assembly, a submembranous sheet of Dia1-nucleated actin filaments could increase adhesion by reducing the lateral mobility of cadherin receptors in the plane of the membrane; this mechanism does not require a direct link between cadherin receptors and the actin filaments (Baumgartner et al., 2003; Iino et al., 2001; Sako et al., 1998). Another interesting possibility is the actin-polymerization-dependent control of turnover of adhesion receptors. For example, modulation of the organization of submembranous actin might protect cadherin from endocytosis, and this could be a very important pathway of adhesion regulation (Ivanov et al., 2004; Le et al., 1999; Troyanovsky et al., 2006).

Finally, future studies are required to verify and elucidate the proposed Dia1-dependent mechanisms, as well as investigate the relationship between Dia1 and other factors affecting actin polymerization in the regulation of the assembly of cell-cell adhesions. Even though the Arp2/3 complex (Kovacs et al., 2002; Verma et al., 2004), WAVE (Yamazaki et al., 2006), N-WASP (Ivanov et al., 2005), cortactin (El Sayegh et al., 2004; Helwani et al., 2004), VASP (Scott et al., 2006; Vasioukhin et al., 2000), formin-1 (Kobielak et al., 2004) and Dia1 (this study) are all involved in the regulation of adherens junctions, the roles of these proteins probably do not overlap entirely. For example, the prominent and unique phenotypes observed upon Dia1 knockdown and Dia1 activation are indicative of the specific and nonredundant functions of this protein in the control of cell-cell adhesion. Dia1-dependent regulation represents a novel facet in the crosstalk between the functions of the cytoskeleton and cadherins.

MCF7 cells were cultured in Dulbecco's modified Eagle's medium supplemented with antibiotics, 2 mM glutamine (Gibco, Rhenium, Jerusalem, Israel) and 10% heat-inactivated fetal calf serum (Biological Industries, Kibbutz Beit Haemek, Israel) at 37°C in a 5% CO₂ humidified atmosphere. Cells were replated using standard trypsin-EDTA solution. Nocodazole was obtained from Sigma-Aldrich (St Louis, MO), and blebbistatin [(±)-1-phenyl-1,2,3,4-tetrahydro-4-hydroxypyrrolo[2,3-b]-7-methylquinolin-4-one] from Calbiochem (San Diego, CA). Both reagents were added to the growth medium from stock solutions in DMSO, so that final the concentration of DMSO in solution did not exceed 0.06%. Calcium-depletion experiments were performed in calcium-magnesium-free Hanks balanced salt solution (Sigma-Aldrich) containing 4 mM EGTA. After 15 minutes, the EGTA-containing HBSS was replaced with calcium-containing DMEM, and reconstitution of cell-cell contacts was monitored for up to 2 hours.

The pSuper Dia1-shRNA (called here si-hDia1), the pSuper-retro-Dia1-shRNA and the pSuper-retro-LacZ vectors were kindly provided to us by Y. Ziv and Y. Shiloh (Sackler School of Medicine, Tel Aviv University, Israel). Briefly, a small interfering oligonucleotide specific for human Dia1 and corresponding to its sequence from base 707 to 725 (GCATGAGATCATTCGCTGC) was synthesized, annealed and cloned in the pSuper plasmids (Brummelkamp et al., 2002). The sequence was then verified by DNA sequencing. The plasmids encoding GFP-fused full-length mDia1 (GFP-mDia1), as well as its constitutively active truncated mutants mDia1-ΔN1, mDia1-ΔN2 and mDia1-ΔN3 described previously (Watanabe et al., 1999), were kindly provided by S. Narumiya (Faculty of Medicine, Kyoto University, Japan). The CFP-ΔDAD truncated mutant lacking the DAD domain (residues 1-1140) (Fig. 7A) was derived from the GFP-mDia1 construct by PCR and ligated into SalI/BglII in the vector pECFP-C1 (BD Biosciences).

Transient transfections were carried out in six-well plates by DreamFect reagent (Oz Biosciences, Parc scientifique de Luminy, Marseille CEDEX 9, France) according to the manufacturer's instructions. The ratio between si-hDia1 plasmid and pGFP-C1 was 3:1. Four hours later, the cells were replated onto 13 mm round coverslips or glass-bottomed culture dishes (MaTek, Ashland, MA) pre-coated with 10 μg/ml fibronectin (tissue-culture grade, Sigma-Aldrich).

To assess the hDia1 expression, cells transfected with pGFP-C1 and empty pSuper plasmids (control) or pGFP-C1 and pSuper-siDia1 plasmids were collected by FACS sorting 48-72 hours following transfection and lysed. A mouse monoclonal antibody (mAb) against p140mDia1 and a mouse mAb against tubulin (Sigma-Aldrich) were used for western blotting.

An MCF7 si-hDia1 stable cell line was obtained by transfection with the pSuper-retro-shDia1 plasmid (the Dia1-shRNA sequence was the same as for the transient transfection). pSuper-retro-LacZ plasmid, bearing an irrelevant shRNA against the LacZ operon, was used as a control. Selection was conducted in the presence of 0.4 μg/ml puromycin (Sigma-Aldrich) for 10 days. Cell lysates were prepared and checked by western blot, which showed an ∼90% reduction of mDia1 protein. Tubulin was used as a loading control. Quantification of western blot signals was performed using the ImageJ software.

Cells processed for immunostaining were fixed and permeabilized 48-72 hours after transfection by incubation for 3 minutes in 3.7% paraformaldehyde (PFA) with 0.5% Triton X-100, followed by PBS washing and incubation in 3.7% PFA at room temperature for an additional 30 minutes. For microtubule immunostaining, fixation with a solution containing 3% paraformaldehyde, 0.05% glutaraldehyde and 0.25% Triton X-100 was used. Cells were immunostained by sequential incubation with the primary and secondary antibodies, for one hour each. After washing with PBS, cells were mounted in Elvanol (Mowiol 4-88, Hoechst, Frankfurt, Germany).

The following primary antibodies were used for staining: mouse mAb against p140mDia1 (BD Biosciences, Heidelberg, Germany); mouse mAb against human E-cadherin (BD Biosciences); rabbit polyclonal anti-β- and α-catenins and monoclonal anti-α-tubulin, clone DM1A (Sigma-Aldrich); mouse mAb against VASP kindly provided by Jurgen Wehland (Helmholtz Centre for Infection Research, Braunschweig, Germany); mouse mAb against phosphotyrosine (clone 4G10, Upstate Biotechnology); and a rabbit polyclonal antibody against cingulin, a gift of Sandra Citi (Dept of Molecular Biology, University of Geneva, Switzerland). The secondary antibodies used were: goat anti-mouse or anti-rabbit, conjugated to Cy3, Cy5 or Alexa 488 fluorochromes (Jackson Laboratories, West Grove, PA). F-Actin was visualized using a coumarin-phalloidin probe (Small et al., 1988) obtained from Sigma-Aldrich.

Images of fixed cells were recorded using the DeltaVision System (Applied Precision, Issaquah, WA) that included an Olympus IX71 inverted microscope equipped with a CoolSnap HQ camera (Photometrics, Tucson, AZ), a 100 W mercury lamp and excitation and emission filter wheels. Images were acquired with an Olympus plan ApoN 60× 1.42 N/A objective.

Laser-scanning confocal microscopy (LSCM) was performed using a Radiant 2000 laser scanning system operated in conjunction with an Eclipse TLE300 inverted microscope (Nikon Corporation, Tokyo, Japan) equipped with a 63× oil-immersion objective, running under Bio-Rad software (Bio-Rad Laboratories, Hercules, CA). Serial XY optical sections were recorded and XZ and YZ profiles were produced using the Bio-Rad software package. Contrast adjustment of raw images was performed in Adobe Photoshop 7 or ImageJ. Panels were assembled in Adobe Photoshop 7.

The integrity of the adherens junctions was quantified by estimation of the fraction of cell-cell contacts preserving intact, non-disturbed E-cadherin-positive junctions. A contact between two cells was considered preserved when the total length of a continuous E-cadherin-positive line was at least 50% of the distance between the corresponding vertices of the cell. ImageJ software was used for the length measurements. Quantitative assessment of adherens junction integrity in Dia1-knockdown cells was performed in three individual experiments by measuring 30 pairs of cells for each case considered.

Levels of E-cadherin, β-catenin, actin and GFP fluorescence intensity at cell-cell junctions were quantified using the line-scan tool in MetaMorph software (Molecular Devices Corporation, Sunnyvale, CA). Intensity measurements were performed along segments oriented perpendicular to the contact, as described by Verma and colleagues (Verma et al., 2004). Intensity profiles (Fig. 3C,D) were calculated by averaging the segments covering a square measuring 20×20 pixels. Between 2 to 5 of such squares per contact were measured in order to cover the entire contact length. The peak pixel intensity values were taken as the measure of fluorescence intensity at the contact (see Results); for the quantification of GFP localization at the junction, we introduced a localization index (I). To calculate this, we subtracted the highest value of cytoplasmic intensity (C) from the peak value of intensity at the junction (J) and divided this difference by the cytoplasmic intensity C, such that: I=(J–C)/C. About 30 contacts were analyzed per experiment, and each experiment was repeated three times.

We are grateful to Y. Shiloh and Y. Ziv (Sackler School of Medicine, Tel Aviv University, Israel) and to S. Narumiya (Faculty of Medicine, Kyoto University, Kyoto, Japan) for providing plasmids, and to Sandra Citi (Department of Molecular Biology, University of Geneva, Switzerland) for the antibody to cingulin. We thank Miri Lev Ran for technical assistance and Mike Grant and Lisa Rogers for critically reading the manuscript. L.C. was supported by a S. Lombroso Fellowship for Cancer Research. C.B. was supported by the Wellcome Centre core award 077100. This study was funded in part by grants from the Israel Sciences Foundation, Minerva Foundation and Pasteur-Weizmann Council to A.D.B. A.D.B. holds the Joseph Moss Professorial Chair in Biomedical Research.