Epithelial cell-cell contacts regulate SRF-mediated transcription via Rac-actin-MAL signalling

Epithelial monolayers line many body compartments and are characterised by strong adhesions between neighbouring cells to restrict transcellular passage of macromolecules and solutes. These cell-cell contacts include tight junctions and adherens junctions, which are functionally but dynamically connected to the circumferential actin belt (Drees et al., 2005; Gumbiner, 2005; Halbleib and Nelson, 2006; Matter and Balda, 2003; Yamada et al., 2005). Transmembrane proteins within these apical junctional complexes link adjacent cells together, while they are tethered to cytoplasmic plaque proteins that function as adapters and regulators. A key component of the adherens junctions is E-cadherin, which forms homophilic trans-interactions at sites of cell-cell contacts in the presence of Ca²⁺ (Gumbiner, 2005; Halbleib and Nelson, 2006). These are vital for initiating and maintaining epithelial architecture in vitro and in vivo (Gumbiner et al., 1988; Larue et al., 1994; Perl et al., 1998).

Formation of epithelial junctions involves Rho family GTPases and a dynamic and contractile actin cytoskeleton (Braga et al., 1997; Braga and Yap, 2005; Jou and Nelson, 1998; Nakagawa et al., 2001; Noren et al., 2001; Sahai and Marshall, 2002; Vasioukhin et al., 2000). However, during embryonic development, tissue remodelling and tumour progression, disintegration of epithelial junctions is a key event, allowing epithelial-mesenchymal transition (EMT). In vitro, junction disassembly can be induced by depletion of extracellular Ca²⁺ (Klingelhofer et al., 2002). Internalisation and breakdown of the apical junctional complexes involves clathrin-mediated endocytosis and actomyosin contractility (de Rooij et al., 2005; Ivanov et al., 2004). This process is initiated and controlled during physiological and pathological conditions by cytokines, toxins and growth factors such as TGFβ, and leads to developmental or pathological EMT (Grunert et al., 2003; Janda et al., 2006; Masszi et al., 2004). Indeed, Ca²⁺ depletion activates the promoter of the mesenchymal marker smooth muscle actin, which is potentiated by TGFβ (Fan et al., 2007; Masszi et al., 2004). The disassembly of epithelial junctions can also be caused by constitutively active or dominant-negative variants of Rho and Rac GTPases (Lozano et al., 2003).

Rho-mediated changes in actin dynamics regulate transcription in serum-stimulated fibroblasts (reviewed by Posern and Treisman, 2006). A key regulatory step is the release of the transcriptional co-activator MAL/MKL1/MRTF-A from monomeric G-actin (Miralles et al., 2003). Upon RhoA-induced dissociation from G-actin and nuclear accumulation, MAL activates serum response factor (SRF)-dependent target genes (Mack et al., 2001; Miralles et al., 2003; Vartiainen et al., 2007). Both MAL and SRF are widely expressed and form complexes with promoter elements termed CArG-boxes (Posern and Treisman, 2006). Serum-stimulated SRF activity in fibroblasts is blocked by inhibition of RhoA and ectopic expression of either non-polymerisable actin mutants such as actin R62D or dominant-negative MAL constructs (Hill et al., 1995; Miralles et al., 2003; Posern et al., 2002). Vice versa, Rho family GTPases, LIMK, Dia1 and F-actin stabilising actin mutants such as actin G15S activate SRF (Copeland and Treisman, 2002; Hill et al., 1995; Posern et al., 2004; Sotiropoulos et al., 1999). In mice, SRF and the MAL paralogue MRTF-B are essential, whereas MAL/MRTF-A-depleted mice show only minor defects in lactation, probably owing to partial functional redundancy (Arsenian et al., 1998; Li et al., 2005; Li et al., 2006; Oh et al., 2005; Sun et al., 2006).

Owing to their important role, epithelial junctions have been studied extensively, but how they influence gene expression is less clear. Using the largely serum-insensitive MDCK and EpRas cell lines as model systems, we show here that the dissociation of epithelial junctions activates SRF via actin dynamics and MAL. Time-course experiments, concentration dependence and the use of E-cadherin-deficient cell lines suggest that disruption of epithelial junctions triggers the transcriptional activation following Ca²⁺ withdrawal. The small GTPase Rac is critically involved in SRF activation and target gene induction in epithelial cells, whereas ROCK and actomyosin contraction are not sufficient.

Disassembly of cell-cell adhesions is a key process during the disintegration of epithelial sheets and epithelial-mesenchymal transition, but little is known about how this influences transcriptional regulation. To test this, we transiently transfected MDCK cells with a luciferase reporter plasmid for the transcription factor SRF. This reporter is activated by the G-actin-MAL pathway, but not by Ras-MAPK-TCF signalling (Hill et al., 1995; Miralles et al., 2003). The transfected MDCK cells were allowed to form typical cell-cell junctions in confluent epithelial monolayers. We then exchanged the culture medium with medium containing physiological (1.8 mM) or reduced amounts of Ca²⁺. We found a significant induction of SRF upon Ca²⁺ withdrawal (Fig. 1A), as measured by luciferase activity. This effect was concentration dependent, with a threshold of about 0.04 mM Ca²⁺. By contrast, confluent MDCK cells do not respond significantly to serum (Fig. 1C; P=0.2 in Student's unpaired t-test). Moreover, the induction of the reporter by low Ca²⁺was strictly dependent on SRF: a mutated reporter lacking functional SRF-binding sites failed to show any induction (Fig. 1C), although its luciferase activities were reduced 20-fold in both conditions (data not shown).

Withdrawal of extracellular Ca²⁺ results in the breakdown of cell-cell junctions including tight junctions and the E-cadherin-containing adherens junctions (Capaldo and Macara, 2007; Klingelhofer et al., 2002; Nakagawa et al., 2001). Monitoring E-cadherin localisation by immunofluorescence demonstrated a loss of E-cadherin staining at cell-cell junctions when Ca²⁺ concentration was reduced (Fig. 1B). Similarly, the disruption of epithelial cell-cell adhesion and the subsequent cell rounding became apparent in phase-contrast microscopy. The threshold Ca²⁺ concentration of 0.04 mM strictly correlated for all effects, including F-actin rearrangement (Fig. 3C), suggesting that dissociation of epithelial cell-cell contacts is linked to SRF activation. EGTA treatment to deplete Ca²⁺ ions from confluent MDCK cells led to comparable results, but we observed strong toxicity upon extended treatment, maybe owing to perturbation of the intracellular level of divalent cations (data not shown). By contrast, medium exchange to low Ca²⁺ did not elicit any apparent harm to MDCK cells for several days (not shown).

To follow the time course of SRF activation, we engineered a reporter construct harbouring a destabilised EGFP controlled by three SRF-binding sites (Fig. 2A). MDCK cells were stably transfected and EGFP expression and cell shape were analysed by time lapse microscopy. EGFP was observable 3 hours after medium exchange to low Ca²⁺ and maintained for around 7 hours before it declined, consistent with a transient SRF activation (Fig. 2B and supplementary material Movie 1). Induction was preceded by dissociation of cell-cell adhesions, which was completed within 15 minutes (Fig. 2B and supplementary material Movie 2). Medium exchange alone, however, did not induce EGFP expression (Fig. 2B, mock). Quantitation of the fluorescence intensity showed that EGFP expression peaked at 8 hours (Fig. 2C), consistent with the luciferase reporter activation. Several independently isolated clones showed similar results, with variations mainly in their background and signal-to-noise ratio (data not shown). Together, these data indicate that SRF-mediated transcription is transiently activated by dissociation of cell-cell adhesions in epithelial cells.

To ensure proper basolateral polarisation, we repeated the luciferase reporter assay with cells seeded on Transwell filters (Grunert et al., 2003). Again, SRF was activated following Ca²⁺ withdrawal (Fig. 1C). This demonstrates that full epithelial polarisation on microporous membranes neither negatively nor positively influences the required signalling, and that the junctions that form on uncoated tissue culture dishes are sufficient for regulation. Thus, the likely reason for the observed effects of Ca²⁺ withdrawal on SRF activity is the dissociation of E-cadherin dependent apical junctions. To further test this, cells lacking E-cadherin expression were subjected to Ca²⁺-reduced medium. Neither MDA-MB 435S cells, a mammary carcinoma cell line, nor AGS cells, a gastric adenocarcinoma cell line, responded to Ca²⁺ withdrawal (Fig. 1D). There was no indication that basal SRF activity is increased in these cell lines, which could disguise a further activation (not shown). Whereas MDA-MB 435S did not show proper epithelial morphology, AGS cells still formed epithelial sheet-like cell clusters, although both lines lacked any detectable E-cadherin expression (Fig. 1D, inset). However, the principal signalling pathways from Rho family GTPases to SRF are intact, as shown by reporter induction with activated Rac1 (Fig. 1D) and RhoA (not shown). This demonstrates that intact apical junctions function as the Ca²⁺ sensor responsible for SRF regulation in epithelial cells.

In fibroblasts, serum stimulation of SRF is blocked by increased level of cellular G-actin, because the SRF co-activator MAL/MRTF binds to G-actin and is thereby inhibited (Miralles et al., 2003; Posern et al., 2002). We thus asked whether elevation of the cellular G-actin by ectopic expression of actin or latrunculin B treatment interferes with signalling to SRF in epithelial cells. Transfection of wild-type β-actin and the non-polymerisable mutant actin R62D efficiently reduced SRF activation by Ca²⁺ withdrawal (Fig. 3A). Likewise, latrunculin B blocked SRF reporter activity. By contrast, the F-actin stabilising mutant actin G15S activated SRF to similar levels both at physiological and reduced Ca²⁺ concentrations (Fig. 4A; data not shown), consistent with its effect in fibroblasts (Posern et al., 2004). Expression of either actin wild type, R62D or G15S did not change the E-cadherin localisation at normal or low Ca²⁺ levels, excluding the possibility that the effects of the actin mutants on SRF were caused by altering formation or disruption of epithelial junctions in transfected cells (Fig. 3B). The F-actin cytoskeleton, visualised by phalloidin staining, showed a prominent reorganisation when subjected to low Ca²⁺ (Fig. 3C). Whereas diffuse F-actin staining disappeared, thick bundles of cortical actin formed in the separating cells when the Ca²⁺ level dropped below the threshold of 0.04 mM. This suggests that changes in actin dynamics, particular in the G-actin level, transduce a signal to SRF in dissociating epithelial cells.

To investigate the role of the SRF co-activator MAL/MRTF-A, we transfected full-length MAL and a constitutively active form, MAL ΔN, which lacks the repressive RPEL motifs responsible for G-actin binding (Miralles et al., 2003). Full length MAL (f.l.) activated SRF, which was further elevated by Ca²⁺ withdrawal, whereas MAL ΔN was sufficient to activate SRF fully even in the presence of Ca²⁺ (Fig. 4A. By contrast, two different dominant-negative forms of MAL, MAL ΔNΔB1 and MAL ΔNΔC, both slightly but significantly reduced the activity of SRF upon Ca²⁺ withdrawal (Fig. 4A).

The crucial step in SRF regulation by MAL is the dissociation of the inhibitory G-actin:MAL complex (Vartiainen et al., 2007). Using co-immunoprecipitation, we analysed the actin:MAL complex in MDCK cells upon epithelial disintegration. Ca²⁺ withdrawal resulted in the rapid and transient dissociation of MAL from actin (Fig. 4B). By contrast, latrunculin B stabilised the complex, whereas cytochalasin D, an activator of MAL and SRF, resulted in actin:MAL dissociation. Moreover, the upward shift probably indicates the phosphorylation of MAL, which is associated with SRF activation, similar to the previous observations in fibroblasts (Miralles et al., 2003). Together, the results demonstrate that SRF regulation in dissociating epithelial cells occurs via G-actin and MAL/MRTF co-activators.

Rho family GTPases are involved in both adherens junction formation and SRF regulation (Braga and Yap, 2005; Hill et al., 1995; Nakagawa et al., 2001). We reasoned that Rho family GTPases are particularly and rapidly sensitive to shear stress. We therefore analysed the GTP loading of RhoA, Rac1 and Cdc42 using Rhotekin-RBD and Pak-CRIB pull-down assays, respectively, in MDCK cells after medium exchange to Ca²⁺-reduced or normal medium. Rac1 and RhoA, but not Cdc42, were rapidly activated upon Ca²⁺ reduction (Fig. 5A). Pairwise quantitation showed that the activation of Rac1 and RhoA declined below background level after 2 hours (Fig. 5A). Averaged over three independent experiments, peak activation of Rac at 3 minutes was 3.9±0.9-fold (for Rho it was 3.2±1.0-fold; data are given ± s.e.m.). This transient activation of Rac1 and RhoA indicates a role for Rho GTPases not only in formation of epithelial cell-cell contacts but also in junction disruption.

To determine whether RhoA and Rac1 are able to activate SRF in MDCK cells, cells were transfected with constitutively active versions of either Rac1 or RhoA. Both were sufficient to activate SRF-mediated transcription, as determined by the luciferase reporter (Fig. 5E).

To test directly whether SRF activation upon Ca²⁺ withdrawal depends on Rho signalling, we took advantage of the cell-permeable exoenzyme C3 from Clostridium botulinum (Tat-C3), a specific inhibitor of RhoA, and two isoforms of Toxin B from C. difficile, the Rho/Rac/Cdc42-inactivating TcdB and the Rac/R-Ras-inactivating TcdBF (Huelsenbeck et al., 2007; Sahai and Marshall, 2002). Ca²⁺ was removed from confluent epithelial MDCK cells pretreated with either inhibitor and cells were analysed for SRF activity. Induction of SRF by low Ca²⁺ was found to be responsive to concentration-dependent inhibition by TcdB and TcdBF, but not by Tat-C3, suggesting that Rac1 rather than RhoA is required (Fig. 5B). Inhibition of either RhoA by Tat-C3 or TcdB or Rac1 by TcdBF was confirmed using the respective effector pull-down assays (Fig. 5C); importantly, Tat-C3 was fully functional in the polarised epithelial cells and completely blocked GTP-loading of RhoA following Ca²⁺ withdrawal [from 3.7±1.1-fold to 0.99±0.15-fold (n=3)]. Treatment with TcdBF, TcdB or Tat-C3 interfered neither with the preformed epithelial junctions (Fig. 5D) nor with their disruption (data not shown) during the course of the experiment. Prolonged treatment of MDCK cells with the Rho family inhibitors, however, resulted in reticulated dissociation of the epithelial sheets, demonstrating that some activity of Rac1 and RhoA is necessary for the maintenance of proper cell-cell adhesion (data not shown).

In sharp contrast, SRF activation in NIH 3T3 fibroblasts was concentration dependently inhibited by Tat-C3 (Fig. 5F), showing that RhoA is essential for serum-stimulation of fibroblasts as reported (Hill et al., 1995). In epithelial cells, however, Rac1 rather than RhoA transmits the signal from dissociating epithelial junctions to SRF.

We next extended our study of SRF activation to more physiological conditions under which junctions are disassembled or rearranged, independently of the extracellular Ca²⁺ level. During HGF-induced scattering of MDCK cells, a small but significant induction of SRF activity correlated with the disappearance of cell-cell contacts (Fig. 6A,B). Additionally, we used EpRas cells, which undergo epithelial-mesenchymal transition in vitro and in vivo following TGFβ treatment (Grunert et al., 2003). In contrast to the dog kidney MDCK cells, for which it has been questioned whether they have a typical zonula adherens (Miyake et al., 2006), EpRas cells are derived from the murine mammary gland and thus also allow monitoring of endogenous gene expression. First, following Ca²⁺ withdrawal, SRF was activated more than fivefold both in EpRas and the parental EpH4 cells (not shown), comparable with our observations in MDCK. Moreover, treatment of EpRas with TGFβ for 3 days, which results in disassembly of the epithelial junctions and actin reorganisation also induced SRF activity (Fig. 6C,D). This suggests that SRF activation is a general response upon dissociation of epithelial cell junctions.

To test the induction of transcription of endogenous SRF target genes, we used quantitative real-time RT-PCR. Vinculin (Vcl) and smooth muscle α-actin (Acta2) are known SRF target genes which are activated via the actin-MAL, but not by the MAPK-TCF pathway (Du et al., 2004; Gineitis and Treisman, 2001). Both Vcl and Acta2 mRNAs were upregulated after 3 hours in low Ca²⁺ (Fig. 6E). The induction of Acta2, A marker for epithelial-mesenchymal transition, was diminished when Rac1 was blocked by either TcdBF or TcdB pretreatment (Fig. 6F). By contrast, inhibition of RhoA with Tat-C3 did not significantly reduce Acta2 induction (Fig. 6F). Consistent with the previous luciferase assays performed in MDCK cells, this result demonstrates the critical role of Rac1 (but not RhoA) in the pathway analysed.

Previous studies have shown that actomyosin contractility is required for internalisation and disassembly of the apical junctional complexes, as well as cell scattering (de Rooij et al., 2005; Ivanov et al., 2004). We therefore tested the effect of Blebbistatin, an inhibitor of non-muscle myosin II ATPase activity. Blebbistatin pretreatment inhibited the activation of SRF upon Ca²⁺ withdrawal (Fig. 7A). Similarly, inhibition of ROCK by Y-27632 abrogated SRF activation (Fig. 7A), in agreement with a recent study (Fan et al., 2007). To exclude side effects, we tested the activation of Rac1 and RhoA, which was maintained in Blebbistatin-treated cells (Fig. 7B). The internalisation of E-cadherin from the cell periphery was reduced (Fig. 7C), consistent with previous studies (Ivanov et al., 2004). We also observed, however, disorganised F-actin staining, and a loss of cytoskeletal reorganisation and cortical bundling following Ca²⁺ withdrawal (Fig. 7C). This suggests that actomyosin contractility is involved in SRF activation, probably by perturbing cytoskeletal F-actin structures.

We therefore examined whether actomyosin contractility plays a direct role in SRF-mediated transcription. In cells co-transfected with activated ROCK kinase, an upstream activator of myosin light chain phosphorylation, no elevated SRF activity was observed (Fig. 7D). In addition, inhibiting myosin light chain phosphatase by calyculin A also failed to activate SRF, despite both stimuli resulted in thick contractile F-actin bundles in phalloidin-stained cells (Fig. 7D,E). This demonstrates that actomyosin contractility is not sufficient for SRF activation and may not directly transmit signals from dissociating epithelial junctions to MAL.

We demonstrate that the disruption of cell-cell contacts in confluent epithelial monolayers activates specific gene expression in the nucleus. Three components of the signalling pathway, the GTPase Rac, G-actin and MAL/MRTF, are both required and sufficient for activation of the transcription factor SRF in epithelial cells (see Fig. 8). By titrating the extracellular Ca²⁺ level below a threshold of 0.04 mM, we show that dissociation of E-cadherin-dependent epithelial junctions precedes the induction of transcription in a SRF reporter cell line. This SRF activation correlates with changes in the actin cytoskeleton and dissociation of the G-actin:MAL complex. Ectopic expression of non-polymerisable mutant actins that bind to and inhibit MAL (Posern et al., 2002) block SRF activation. Conversely, F-actin stabilising mutant actins (Posern et al., 2004) activate SRF even in the presence of Ca²⁺, demonstrating an inhibitory role for monomeric actin in transcriptional activation of disintegrating epithelial sheets.

Both adherens junctions and tight junctions are dynamically connected to the actin cytoskeleton, and disrupting them would expectedly interfere with F-actin structures. However, we demonstrate here that the small GTPase Rac is critically involved. The relationship between junctions and Rho family GTPases has been studied extensively, but mainly during formation of junctions (Braga et al., 1997; Noren et al., 2001) or HGF-induced scattering (Lynch et al., 2006; Zondag et al., 2000). We provide evidence that both Rac1 and RhoA are quickly and transiently GTP loaded when epithelial sheets are subjected to 0.02 mM Ca²⁺. By contrast, treatment with EGTA for 30 minutes has previously been shown to inhibit Rac localisation and activation, whereas RhoA remained unchanged under such conditions (Balzac et al., 2005; Nakagawa et al., 2001). This apparent discrepancy is probably explained either by junction-independent effects of EGTA, consistent with our observed toxicity upon extended treatment, or by the transient nature of the Rac and Rho activation, which previously might have been missed.

Activated forms of both RhoA and Rac are sufficient to activate SRF. However, only Rac proved to be required for activation of SRF-mediated transcription. We used clostridial cytotoxins to investigate the involvement of GTPases. Tat-C3 is a cell-permeable RhoA-specific ADP-ribosyltransferase (Sahai and Marshall, 2002). TcdB inhibits RhoA and Rac/Cdc42 by glucosylating Thr-37 and Thr-35 within the effector region, respectively (Huelsenbeck et al., 2007). Finally, TcdBF harbours the same receptor-binding and internalisation domain as TcdB, but has an altered glucosyltransferase specificity, targeting selectively Rac and R-Ras (Chaves-Olarte et al., 2003; Huelsenbeck et al., 2007). Comparing the effects of these toxins allows us to address the specific requirements for GTPases, unlike the expression of dominant-negative constructs that already interfere with the formation of junctions when reseeded to form confluent epithelial sheets (data not shown) (Braga et al., 1997; Jou and Nelson, 1998). Moreover, overexpression of RacN17 or RhoN19 may not necessarily result in the specificity previously anticipated (Wells et al., 2004). With the concentrations and times used here, the toxins efficiently blocked their specific target GTPase, while neither perturbing existing junctions (Fig. 5) nor junction disassembly following Ca²⁺ withdrawal (data not shown).

The specific requirement for Rac, but not for Rho, during SRF activation by dissociating epithelial junctions is contrasted by the situation in serum-stimulated fibroblasts. In NIH3T3 cells, the RhoA-inactivating C3 exoenzyme efficiently blocks SRF (Fig. 5F) (Hill et al., 1995). The effect of TcdBF is difficult to assess in fibroblasts, because these cells quickly round up and detach upon treatment; however, in detached cells, no concentration-dependent reduction of serum-induced SRF activity was observed (data not shown). In addition, confluent epithelial cells are hardly stimulatable by serum (Fig. 1C). These findings indicate that epithelial disintegration and serum stimulation use distinct upstream pathways, which converge at the level of G-actin to activate MAL/SRF dependent transcription.

Our studies with the toxins also exclude a role for R-Ras following Ca²⁺ withdrawal, as TcdB, which does not block R-Ras, inhibits SRF. A close relative to R-Ras, Rap1, has also been implicated in the regulation of cell-cell contacts of epithelial cells (Kooistra et al., 2007). Neither overexpression of RalGDS-RBD, a R-Ras- and Rap1-specific inhibitor, nor the Rap1-specific exchange factor Epac (with deleted regulatory cAMP-binding domain) affected SRF activity, suggesting that Rap1 is not directly involved (data not shown).

What is the function of RhoA activation following dissociation of transcellular junctions, if it is not involved in SRF activation? The Rho-ROCK signalling axis has been shown to disrupt junctions, and endocytosis and disassembly of the apical junctional complexes following Ca²⁺ withdrawal, as well as scattering of MDCK cells, depends on actomyosin contractility (de Rooij et al., 2005; Ivanov et al., 2004; Sahai and Marshall, 2002). Thus, RhoA may affect the internalisation and degradation of the junctional protein complexes via ROCK and regulation of non-muscle myosin II activity. Consistent with this, junctional proteins fail to internalise following Ca²⁺ withdrawal when pretreated with Blebbistatin (Fig. 7C) (Ivanov et al., 2004). This does not contradict a Rac-dependent signal from the junctions to MAL: the homophilic interactions are disrupted and Rac is still activated by Ca²⁺ withdrawal (Fig. 7B), but internalisation and actin remodelling is blocked (Fig. 7C). The low level of basal RhoA activity which is maintained during our Tat-C3 treatment appears to be sufficient to permit the Ca²⁺-induced actin remodelling and internalisation of junctional complexes (not shown), despite the complete block of induced GTP-loading (Fig. 5C).

The activation of SRF-mediated transcription and the loss of E-cadherin staining at sites of cell-cell contacts both exhibit a comparable concentration dependency on Ca²⁺, suggesting a causative connection between the two effects. It is important to note that the Ca²⁺ threshold reported here is considerably lower than that required for junction formation, however. This demonstrates that data from such reverse experiments are not directly comparable. Our experimental setup depends on the ability of the medium to deplete Ca²⁺ from existing cell junctions in a confluent monolayer, which probably explains the low concentration reported here and in previous studies (Capaldo and Macara, 2007; Klingelhofer et al., 2002; Pokutta et al., 1994).

Some mammary breast carcinoma and gastric adenocarcinoma cell lines lack the expression of E-cadherin. These cells do not only fail to form adherens junctions, but also lack proper tight junctions and desmosomes. We found that these E-cadherin deficient cells do not show any SRF activation in response to low Ca²⁺, suggesting a role of either E-cadherin or affected transmembrane proteins, including TJ components, in the cellular machinery sensing cell-cell dissociation. Future work will identify the molecular nature of the sensor and will characterise the link to Rac-dependent SRF-mediated transcription.

The ROCK inhibitor Y-27632 and the myosin ATPase inhibitor Blebbistatin blocked induction of SRF. While this manuscript was in preparation, another study showed that the smooth muscle α-actin promoter, which is regulated by MAL and SRF, is induced by disruption of intercellular contacts (Fan et al., 2007), consistent with our findings. In addition, they also showed that Y-27632 and Blebbistatin suppresses the smooth muscle α-actin promoter (Fan et al., 2007). At first sight, this indicates a role for actomyosin contractility. We showed, however, that Blebbistatin treatment leads to disassembly of the cellular F-actin and abrogates the cytoskeletal changes elicited by Ca²⁺ withdrawal (Fig. 7C). Consistent with this, loss of force has been shown to trigger the breakdown of F-actin and adhesion sites (Giannone and Sheetz, 2006). Thus, we speculate that a basal level of ROCK activity, myosin light chain phosphorylation and actomyosin contractility (Fan et al., 2007) is required for maintaining the integrity and treadmilling of the actin cytoskeleton, which is a prerequisite for signal transmission from Rac to MAL (Fig. 8). However, we cannot exclude more general effects of Blebbistatin on gene expression.

A direct role for actomyosin in SRF regulation is also inconsistent with our finding that inducing contractility with calyculin A or activated ROCK is not sufficient to induce SRF in epithelial cells (Fig. 7). Moreover, ROCK also fails to induce SRF in fibroblasts (Sahai et al., 1998). By contrast, activated mouse Dia1, which does not facilitate bundling or contraction but Rho- and ROCK-independent actin polymerisation, activates SRF in epithelial cells even in the presence of Ca²⁺ (data not shown). Thus, changing the treadmilling cycle of actin, and thereby the G-actin population in the cell, provides a critical signal, whereas the ROCK-actomyosin signalling axis is not sufficient for SRF regulation in epithelial cells.

The disintegration of epithelial junctions is a key event during epithelial-mesenchymal transition (EMT), which occurs during development and cancer metastasis. We demonstrate here the induction of the MAL-dependent SRF gene expression program during epithelial disintegration. The two known MAL-dependent SRF targets we tested, the genes encoding smooth muscle α-actin and vinculin, were upregulated in mouse mammary epithelial cells. The induction of vinculin could potentially result in stabilisation of cell-cell and cell-matrix adhesions (Ziegler et al., 2006). By contrast, smooth muscle actin is considered a mesenchymal marker, also in the EpRas EMT model system we used (Grunert et al., 2003). This raises exciting questions about the functional outcome of the dissociation-induced MAL and SRF activation, potentially pointing towards a positive feed-forward loop during EMT. Indeed, a very recent report showed that MRTFs are crucial mediators of EMT induced by TGFβ (Morita et al., 2007). The precise nature of these regulatory circuits remains to be elucidated.

Details are available upon request. (–)-Blebbistatin, latrunculin B and calyculin A were purchased from Calbiochem. SRF luciferase reporter and actin mutants have been described previously (Posern et al., 2004; Posern et al., 2002). The control reporter lacking functional SRF-binding sites was derived from p3DA-Luc, with CCATATTAGG mutated to CCCAATCGGG (Hill and Treisman, 1995). Deletions in murine MAL were as follows: ΔN, 1-171; ΔB1, 316-341; ΔC, 563-1021; numbering based on pEF-MAL-HA (f.l.) (Miralles et al., 2003). Other plasmids were for Rac V12 (Hill et al., 1995), myc-ROCKΔ4 and RhoQ63L in pcDNA3 (gifts from Reinhard Faessler). The degradable EGFP expression plasmid p3DA-d2EGFP was generated by replacing the CMV promoter of pd2EGFP-N1 (BD Clontech) with three SRF-binding sites in front of a Xenopus actin TATA-box (Mohun et al., 1987). TcdB and TcdBF were purified as described from Clostridium difficile strains VPI 10463 and 1470, respectively (Huelsenbeck et al., 2007). The Tat-C3 expression plasmid was a gift from Eric Sahai (Cancer Research UK, London), and the fusion protein was expressed and purified from E. coli BL21 (Sahai and Olson, 2006).

Cells were grown in DMEM (Gibco) supplemented with 10% foetal calf serum (FCS, Gibco), except for EpRas cells, which require 4% FCS. For cells grown in confluent monolayers, the medium was refreshed every 24 hours. To disrupt Ca²⁺-dependent cell-cell contacts cells were washed once and then cultured in low calcium medium (calcium-free DMEM (Gibco) with 0.02 mM Ca²⁺ and FCS). For this medium, FCS was depleted of divalent cations using the Chelex 100 resin (Biorad) as described (Brennan et al., 1975). To generate MDCK cells stably expressing the SRF-EGFP reporter, transfected cells were selected in 600 μg/ml G418 and further maintained at 300 μg/ml.

Transfections were carried out using Lipofectamine (Invitrogen) for MDCK, NIH3T3, AGS and EpRas cells, and TransFast (Promega) for MDA-MB-435S cells, according to the manufacturers' protocols. 1.5×10⁶ cells per 10-cm diameter dish were transfected with 1.2 μg p3DA-Luc, 1 μg pRL-TK, together with the indicated plasmids in a total of 5 μg vector. 18 hours after transfection cells were reseeded to form confluent monolayers (600,000 per 1-cm diameter or 175,000 per transwell filter). 24 hours after reseeding, the medium was exchanged and 7 hours later cells were lysed. Cells were pretreated with Tat-C3 for 15 hours, TcdBF or TcdB for 4 hours, Blebbistatin for 2.5 hours, latrunculin B or calyculin A for 30 minutes. NIH3T3 cells were transfected and serum stimulated as described (Posern et al., 2002). Firefly luciferase activity was normalized to either protein content or pRL-TK luciferase activity, as indicated. Figures show mean ± s.e.m. of at least three independent experiments. For each immunoprecipitation, 1×10⁷ MDCK cells were electroporated with 20 μg of DNA in a GenePulser Xcell with CE and PC module (BioRad) using the square wave protocol (voltage, 250 V; pulse length, 10 ms; number of pulses, 3; pulse interval, 0.1 seconds) in a 4 mm GenePulser cuvette (Biorad). Cells were seeded in a 6-cm diameter dish. 36 hours later, medium was exchanged and cells were lysed in RIPA and immunoprecipitated in 1% Triton-X buffer (Miralles et al., 2003) for 2 hours, using M2 anti-flag agarose beads (Sigma-Aldrich).

For immunofluorescence microscopy, cells were fixed with 4% paraformaldehyde, permeabilized in 0.2% Triton X-100 and blocked with 10% FCS, 1% gelatine, 0.05% Triton X-100 in PBS. Staining condition were as follows: E-cadherin (DECMA-1, Sigma-Aldrich), 1:1000; M2 anti-Flag (Sigma-Aldrich), 1:500; rhodamin phalloidin (Molecular Probes), 1:40; anti-Myc (9E10, CR-UK), 1:100; Alexa Fluor 488 goat anti-mouse IgG (H+L) (Molecular Probes), 1:1000; TRITC anti-rabbit (Dako Cytomation), 1:40. Micrographs were taken using a Zeiss Axioplan 2 with MetaVue software (Molecular Devices), or a LEICA TCS SP2 AOBS confocal laser-scanning microscope. Live cell imaging was carried out on a Zeiss Axiovert microscope with MetaMorph software. The induction profile showing the time course of EGFP expression was calculated by subtracting the arbitrary fluorescence intensity at each time point in the mock experiment from the equivalent arbitrary fluorescence intensity upon Ca²⁺ withdrawal.

For quantitative RT-PCR, murine epithelial EpRas cells were seeded to form a confluent monolayer (550,000 per 1-cm diameter). 36 hours later the medium was exchanged, and after 3 hours RNA was isolated. Cells were pretreated with 1 μM Tat-C3 (15 hours), 0.25 μg/ml TcdBF (4 hours) or 0.3 ng/ml TcdB (4 hours). The inhibitor treatment was maintained throughout the assay. RNA preparation (Qiagen) and first-strand cDNA synthesis (ABgene) were carried out according to the manufacturers' protocol. For cDNA synthesis, 1 μg of RNA and anchored oligo-dT primers were used. For cDNA quantitation, 1/40th of the RT reaction was mixed with gene-specific primers (0.5 μM), MgCl₂ (3 mM) and LightCycler FastStart DNA Master SYBR Green I mix (1.5 μl; Roche) to a total volume of 15.5 μl. The primers used are: hprt, tcagtcaacgggggacataaa (forward), ggggctgtactgcttaaccag (reverse); acta2, tgacgctgaagtatccgataga (forward), gtacgtccagaggcatagagg (reverse); and vinc, ggccggaccaacatcagtg (forward), atgtaccagccagatttgacg (reverse). PCR was carried out on a LightCycler instrument (Roche) according to the manufacturer's instructions. Calculations were made using the ΔΔC_t method (Winer et al., 1999).

Preparation of GST-Rhotekin-RBD and GST-PAK-CRIB fusion protein was carried out using E. coli BL21 DE3 Rosetta. Induction of expression by 1 mM IPTG at 30°C for 90 minutes was followed by bacterial lysis in 50 mM Tris (pH 7.5), 1% Triton X-100, 150 mM NaCl, 5 mM MgCl₂ and 1 mM DTT. The lysate was sonicated on ice (6×10 seconds) and the 25,000 g supernatant was snap-frozen in aliquots. Optimised amounts of lysate were used for coupling to glutathione sepharose beads (GE Healthcare) directly prior to GTPase pull-down assays (Ren and Schwartz, 2000). 1×10⁷ MDCK cells were seeded in 10 cm dishes to form a confluent monolayer and cultured for 36 h. After medium exchange, cells were lysed in Rho lyses buffer [50 mM Tris (pH 7.5), 1% Triton X-100, 500 mM NaCl, 10 mM MgCl_2, 0.5% sodium desoxycholate, 0.1% SDS, protease inhibitors]. The cleared lysates were incubated with immobilised GST-Rhotekin-RBD at 4°C for 35 minutes and afterwards washed three times in Rho wash buffer [50 mM Tris (pH 7.5), 1% Triton X-100, 500 mM NaCl, 10 mM MgCl₂, protease inhibitors). The Rac1/Cdc42-GTP pull-down assay was performed accordingly, except for 150 mM NaCl in lysis and wash buffer. As controls, cell lysates were incubated either with uncoupled beads alone or preincubated with GTPγS (1 mM for 10 minutes) prior to precipitation. Bound proteins were detected by western blotting using RhoA (Santa Cruz), Rac1 or Cdc42 (Transduction Laboratories) antibodies. Densitometric analysis of western blots was carried out with AIDA software (Raytest) and normalised to total lysates or ratios.

This work was made possible by generous support from Axel Ullrich at the MPI of Biochemistry. We thank Richard Treisman (Cancer Research UK, London) for various constructs, Sina Bartfeld (MPI of Infection Biology, Berlin) for AGS cells, Thomas Wirth (University Ulm) for EpRas cells, Eric Sahai (Cancer Research UK, London) for Tat-C3, and Reinhard Faessler (MPIB, Martinsried) for RhoA, Dia and ROCK. Laura Leitner, Carolin Schächterle and Monika Rex-Haffner helped establishing techniques and reagents in the laboratory. We also thank Michael Sixt for critically reading of the manuscript. Funding was obtained from the Wilhelm Sander-Stiftung to G.P.