Snail induces epithelial cell extrusion by regulating RhoA contractile signalling and cell–matrix adhesion

Transcription factors of the Snail family are important regulators of epithelial morphogenesis, both during development and in post-developmental life (Nieto, 2002). In the archetypal case of Snail (SNAI1), its morphogenetic impact has commonly been ascribed to its ability to promote epithelial-to-mesenchymal transitions (EMTs) (Peinado et al., 2007; Lamouille et al., 2014; Nieto et al., 2016; Dongre and Weinberg, 2018). For example, during gastrulation in sea urchin embryos, Snail is essential for the transformation of epithelial cells into primary mesenchymal cells, leading to their delamination and invasion into the blastocoele (Wu and McClay, 2007). Similarly, Snail has been implicated in tumour invasiveness. EMT typically involves the repression of genes associated with the epithelial phenotype, such as E-cadherin, and the expression of mesenchymal genes, such as those encoding vimentin and N-cadherin (Batlle et al., 2000; Cano et al., 2000). In this model, the morphogenetic impact of Snail reflects cellular transdifferentiation. Downregulation of E-cadherin is thought to allow cells to dissociate through loss of cell–cell junctions, ultimately facilitating their invasion into the stroma.

Snail is also implicated in morphogenetic events in which cellular contractility is a major driving process. During Drosophila gastrulation, formation of the ventral furrow is mediated by contractility in the cortical actomyosin cytoskeleton, regulated by both Snail and another transcription factor, Twist (Martin et al., 2009; Weng and Wieschaus, 2016). Twist induces the expression of Snail in these ventral cells, which elicits the pulsed contractions of medial-apical actomyosin networks that constrict the apical poles of cells and drive formation of the ventral furrow (Martin et al., 2009). Actomyosin-based cortical contractility has also been implicated in basal delamination of cells, both during early mouse development and in Caenorhabditis elegans gastrulation (Pohl et al., 2012). This is coordinated with the assembly of effective cell-to-cell adherens junctions (Roh-Johnson et al. 2012). Indeed, both cortical contractility and E-cadherin are necessary for Snail-induced apical constriction during Drosophila gastrulation (Weng and Wieschaus, 2016). Thus, regulation of cellular contractility provides an alternative pathway for Snail to influence epithelial organization.

In this report, we analyse the ability of Snail to promote the morphogenetic phenomenon of epithelial cell extrusion. This is a ubiquitous event, where minorities of cells are expelled from the epithelium either apically, towards the lumen (apical extrusion) or in a basal direction into the stroma (basal extrusion, also known as delamination). Extrusion is a biomechanical phenomenon that can be induced by diverse cellular changes, ranging from apoptosis to tissue overcrowding. However, our understanding of the mechanisms that drive extrusion remains incomplete; nor do we understand what determines the direction of extrusion, although roles for polarized intracellular contractility (Marshall et al., 2011) and cortical capture of microtubules have been identified (Slattum et al., 2009). Here, we show that expression of a stabilized Snail transgene (Snail^6SA, with six serine residues mutated to alanine) activates cell contractility through a transcriptional network that stimulates the RhoA GTPase. This hypercontractile cortex induces apical extrusion upon mosaic expression when Snail^6SA cells are surrounded by wild-type epithelia. In contrast, when expressed in large groups of cells, Snail^6SA induces sporadic basal delamination, a process that requires both the transcriptional suppression of integrin-based adhesion and RhoA-driven cortical contractility. Therefore, extrusion represents a hitherto unrecognized effect of Snail that is distinct from its canonical role in EMT.

We developed an MCF-7 cell line that inducibly expresses Snail^6SA, a stabilized mutant that is less susceptible to metabolic turnover and predominantly localizes to the nucleus (Zhou et al., 2004). Characteristically, Snail^6SA expression was evident within 6 h of doxycycline induction, increasing to ∼threefold at 48 h (Fig. S1A,B). Strikingly, we found that Snail^6SA cells underwent apical extrusion from the epithelium when they were expressed mosaically within otherwise confluent wild-type monolayers (Fig. 1A,B; Fig. S1C,D). Apical extrusion occurred in small clusters (less than three cells), but not when Snail^6SA was expressed in larger groups. Although apical extrusion is elicited when epithelial cells undergo apoptosis (Michael et al., 2016), extruded Snail^6SA cells showed no evidence of apoptosis, as identified with annexin V labelling (Fig. S1E,F), and cell density around Snail^6SA cells was comparable to control conditions (Fig. S1N,O). Similar apical extrusion was observed when Snail^6SA was expressed in MCF-7 cells that had been CRISPR/Cas9-engineered to express E-cadherin-GFP (MCF-7^E-cad-GFP) (Fig. 1C; Fig. S2A–G, Movie 1) and in Caco-2 cells (Fig. S1G–J). Thus, apical extrusion appeared to be a common epithelial response to the mosaic expression of Snail^6SA.

However, the morphogenetic impact of Snail^6SA depended on the extent of transgene expression within the epithelium. When Snail^6SA was expressed in clusters of more than three cells they were retained in the monolayer and did not undergo apical extrusion (Fig. 1A,B). Indeed, monolayers that uniformly expressed Snail^6SA retained E-cadherin-based cell–cell junctions and overall epithelial integrity (Fig. 1D,E). This is reminiscent of what has been observed when oncogenes, such as Ras and Src, are ubiquitously expressed in epithelial monolayers (Hogan et al., 2009; Kajita et al., 2010). Cell size and cell density in Snail^6SA monolayers were similar to those in control monolayers expressing NLS^mCherry (Ctrl) (Fig. S1K–M). However, closer inspection revealed sporadic numbers of Snail^6SA cells underneath the primary layer of the monolayer (Fig. 1F,G). Furthermore, live imaging of MCF-7^E-cad-GFP-Snail^6SA monolayers showed that some of these cells lost their apical surface to exit the monolayer basally (Movie 2). This suggests that, although Snail^6SA-expressing cells were not apically extruded when they were grown as uniform monolayers, they could undergo sporadic basal delamination.

We next sought to understand the changes in signalling downstream of Snail^6SA that underpin these cell behaviours. For this, we compared the transcriptional profiles of control and Snail^6SA monolayers, as well as apically extruded Snail^6SA cells (ApiEX-Snail^6SA cells). RNA-seq analysis performed on total RNA extracted from three independent monolayer cultures identified 8843 differentially expressed genes in Snail^6SA monolayers (4280 downregulated, 4563 upregulated, FDR<0.05; Table S1). In addition to the expected increase in Snail (P=1.06×10⁻²¹, FDR=5.882478e-18, logFC=6.04), the EMT-inducing transcription factor ZEB1 was also upregulated (P=5.9×10⁻⁹, FDR=8.57E-08, logFC=2.23) (Fig. S3A). The differentially expressed genes were examined for changes in epithelial- and mesenchymal-specific genes using the framework of Tan et al. (2014). Surprisingly, we found no significant differences in the expression of genes associated with EMT in Snail^6SA cells, including those encoding the canonical markers E-cadherin (CDH1) and vimentin (VIM) (Fig. 2A,B; Table S1). Western analysis confirmed that E-cadherin and vimentin protein expression was not altered in Snail^6SA cells, compared with control monolayers (Fig. 2C). Furthermore, there were no significant differences in the expression of genes associated with epithelial polarity (Fig. 2C; Fig. S2H). Overall, this suggests that Snail^6SA does not induce EMT in these cells. In addition to this, RNA-seq analysis identified a total of 984 and 509 differentially expressed genes that were unique to Snail^6SA and ApiEX-Snail^6SA cells (Fig. S2J, Table S2).

To corroborate these findings, we used the EMT signature of Tan et al. (2014) to compare the relative abundance of epithelial and mesenchymal genes between control (NLS^Cherry), Snail^6SA and ApiEX-Snail^6SA cells (Fig. 2D). This analysis revealed that the expression of most of the epithelial and mesenchymal genes was similar across these conditions, implying that Snail^6SA cells show a strong tendency to maintain epithelial identity (Fig. 2D). Furthermore, the epithelial and mesenchymal affinities of all three conditions were similarly based on the EMT score (Tan et al., 2014) (Fig. S2I). Together, these results indicate that Snail^6SA cells maintain epithelial identity whether in monolayers or following apical extrusion. This suggests that EMT is not the cause of apical extrusion of Snail^6SA cells.

What then could cause the extrusion of Snail^6SA cells? Contractile stress at E-cadherin junctions has been implicated in the extrusion of both oncogene-transfected and apoptotic cells (Wu et al., 2014; Rosenblatt et al., 2001; Michael et al., 2016). Accordingly, we interrogated our transcriptomic profiles for changes in potential regulators of the actomyosin cytoskeleton, focusing on monolayers that expressed Snail^6SA. Interestingly, we found that RhoA mRNA levels were moderately elevated in Snail^6SA monolayers (Fig. 3A). This observation was confirmed by quantitative PCR (Fig. S3B) and by immunoblotting Snail^6SA monolayers for total RhoA protein (Fig. 3D,E). This suggested that Snail^6SA might enhance the expression of RhoA. However, RhoA signals to the actomyosin cytoskeleton when it is bound to GTP (Jaffe and Hall, 2005). To test whether the increased expression of RhoA was accompanied by its increased activity in the cells, we isolated GTP-RhoA from Snail^6SA monolayers using Rhotekin-RBD pulldown assays. These showed that overall GTP-RhoA levels were increased in Snail^6SA monolayers compared with control monolayers (Fig. 3B,C). Thus, Snail increased both the expression and activation of RhoA.

RhoA activity is controlled by the balance between guanine nucleotide exchange factors (GEFs), which promote its active GTP-loaded state, and inhibitory GTPase-activating proteins (GAPs) (Jaffe and Hall, 2005). We therefore asked whether transcriptional regulation by Snail^6SA might affect these classes of proteins. For this, we tested whether genes from the curated gene set of the RhoA regulatory pathway were differentially expressed in our cell lines. Analyses revealed that mRNA levels for the RhoA GEFs NET1, DEF6 and ARHGEF28 were increased in Snail^6SA monolayers compared with controls. Conversely, transcript levels of RhoA GAPs DLC1, SRGAP1, ARHGAP5 (p190B RhoGAP) and ARHGAP6, were reduced upon Snail^6SA expression (Fig. 3A). We validated some of these changes by quantitative PCR and western analysis, choosing candidates for which effective antibodies were available. Together, these experiments corroborated that both mRNA and cellular protein levels of the RhoGEF Ect2 were elevated in Snail^6SA monolayers compared with controls (Fig. 3D,G; Fig. S3B). Conversely, mRNA and protein levels of p190B RhoGAP were reduced in Snail^6SA monolayers (Fig. 3D,F; Fig. S3B). Overall, this suggests that transcriptional regulation by Snail^6SA can stimulate RhoA signalling both by upregulating RhoA itself and by altering the GEF:GAP balance to promote its activation.

We then sought to assess where active RhoA was expressed in response to Snail^6SA. For technical ease, we used monolayers induced to uniformly express Snail^6SA in these experiments. Adherens junctions are a prominent site of RhoA signalling in MCF-7 cells (Priya et al., 2015). Consistent with this, levels of active RhoA were increased at the adherens junctions in Snail^6SA monolayers compared with controls (Fig. 4A–C), as identified with AHPH, a location biosensor for GTP-RhoA (Priya et al., 2015; Liang et al., 2017). Furthermore, a FRET-based RhoA sensor also showed increased activity at the adherens junctions of Snail^6SA cells (Fig. 4D,E). As this activity sensor reflects the balance between RhoA activators and inactivators, this result suggests that the network changes seen in our transcriptomic analysis might also affect the expression of GEFs and GAPs at junctions. Indeed, levels of the RhoGEF Ect2 were elevated at junctions (Fig. S3E,F), whereas levels of p190B RhoGAP were repressed, in Snail^6SA cells compared with controls (Fig. S3E,G). This implies that Snail^6SA expression stimulates RhoA signalling at adherens junctions.

Immunofluorescence staining also revealed that cortical levels of F-actin and non-muscle myosin IIA (NMIIA) were increased at the junctions of Snail^6SA cells (Fig. S3E,H,I), without changes in total cellular NMIIA expression (Fig. S3C,D). This was accompanied by increased levels of phosphorylated myosin regulatory light chain (ppMLC), an index of activated myosin II (Fig. S3E,J). Increased actomyosin activity would be predicted to increase tensile forces at junctions. To test this, we first stained for a tension-sensitive epitope of α-catenin (α-18 mAb) (Yonemura et al., 2010). Junctional α-18 mAb staining was increased by expression of Snail^6SA, evidence of increased molecular tension across the cadherin-catenin complex (Fig. 4F,G). This was corroborated by measuring the recoil of E-cadherin-GFP junctions after they were cut by laser ablation. Initial recoil speed was elevated in Snail^6SA cells compared with controls (Fig. 4H,I). However, Snail^6SA did not alter viscous drag at junctions, as inferred from the rate constants (k-values) of relaxation (Fig. S3K). This suggested that the increase in initial recoil velocity principally reflected an increase in junctional tension. Together, these findings imply that Snail expression induces a transcriptional network that upregulates cellular contractility via RhoA. It should be noted that, although these changes were most readily evident at cell–cell junctions, active RhoA was also increased at the apical surfaces of Snail^6SA cells (Fig. 4A–C). This implies that Snail signalling may enhance cortical contractility more globally within the cells.

The transcriptional network that we have identified has the potential to upregulate RhoA signalling in a cell-autonomous fashion. Therefore, we asked whether this also applied when Snail^6SA was mosaically expressed in monolayers. First, we used AHPH to characterize the distribution of GTP-RhoA in Snail^6SA cells that were surrounded by wild-type cells. Compared with control NLS^mCherry-transfected cells, mosaic Snail^6SA cells showed increased cortical levels of AHPH (Fig. 5A–C). This was evident at cell–cell junctions (Fig. 5A,B), associated with increased junctional RhoA and Ect2 RhoGEF, and a concomitant decrease in p190B RhoGAP (Fig. S4A–D). This suggests that Snail^6SA could drive RhoA activation even when neighbouring cells were wild type. There was also increased accumulation of cortical AHPH at the extrajunctional apical domain in these cells (Fig. 5A,C), as observed in Snail^6SA monolayers (Fig. 4A,C). This reinforced the notion that upregulation of RhoA is not confined to cell–cell junctions.

Immunofluorescence further revealed that cortical levels of F-actin and NMIIA were increased at the junctions between Snail^6SA cells and their wild-type neighbours in these mosaic cultures (Fig. 5D–F). This was accompanied by evidence of increased mechanical tension at these heterologous junctions, which showed increased staining with the α-18 mAb (Fig. S4E,F) and increased initial recoil after laser ablation (Fig. 5G,H). Thus, the cell-autonomous upregulation of RhoA by Snail^6SA stimulated junctional actomyosin to enhance contractile tension. Of note, actomyosin was also increased in non-junctional regions of the cortex (Fig. 5D), consistent with the presence of active RhoA at these sites.

Based on these findings, we hypothesized that the increased contractility in Snail^6SA cells might be responsible for their apical extrusion. Consistent with this, depletion of either RhoA or NMIIA in all cells of the monolayers reduced apical extrusion of mosaically expressed Snail^6SA cells (Fig. S4G,H). We then used a myosin regulatory light chain mutant (MRLC^AA) to test specifically whether enhanced contractility in the Snail^6SA cells was necessary for extrusion. MRLC^AA is mutated for the residues (Thr19 and Ser20) whose phosphorylation is necessary to activate NMII. To characterize its effect, we first expressed MRLC^AA in either control MCF-7 cells or monolayers that uniformly expressed Snail^6SA. MRLC^AA reduced junctional tension in both circumstances, as measured by recoil after laser ablation (Fig. S4I–O), suggesting that it could be used as an inhibitor of contractility. Consistent with this, MRLC phosphorylation at the activatory Ser19 site was reduced in cells expressing MRLC^AA (Fig. S4P).

Then, to reduce contractility in Snail^6SA cells that were grown under mosaic conditions, we coexpressed MRLC^AA with Snail^6SA and mixed these double transfectants with a majority of wild-type cells. Strikingly, under these mosaic conditions, we found that MRLC^AA reduced the apical extrusion of Snail^6SA cells by approximately threefold, compared with mosaic Snail^6SA cells where wild-type MRLC was coexpressed (Fig. 5I,J). This implied that the increased contractility of Snail^6SA cells was necessary for their apical extrusion. To test whether contractility alone could trigger apical extrusion, we then mosaically expressed a phosphomimetic MRLC (MRLC^DD), which can enhance contractile tension at apical adherens junctions (Leerberg et al., 2014). However, the mosaic expression of MRLC^DD was not sufficient to cause apical extrusion (Fig. S4Q,R). This suggests that enhanced cortical contractility is necessary, but not sufficient, for apical extrusion of mosaic Snail^6SA cells.

We then sought to understand what caused the sporadic basal delamination observed when Snail^6SA was expressed in large groups of cells. One possibility was asymmetric cell division; however, Snail^6SA monolayers predominantly exhibited planar cell divisions (Fig. S6M,N); nor was basal delamination increased when we expressed MRLC^DD alone (Fig. 6A,B). This suggests that enhanced contractility alone does not account for basal delamination in Snail^6SA monolayers.

As epithelial cells exhibit strong adhesion to the extracellular matrix (ECM) (Mao and Baum, 2015), we hypothesized that Snail^6SA might affect cell–ECM adhesion. To test this, we used the gene sets relating to extracellular matrix reorganization, the basal lamina and cell–matrix adhesion from the Reactome and GO databases to search for differentially expressed genes. Snail^6SA expression was associated with significant downregulation of genes related to cell–ECM adhesion and the ECM (e.g. genes encoding paxillin, zyxin, integrins, collagen and laminin), and upregulation of transcripts of genes influencing matrix remodelling (e.g. encoding matrix metalloproteinases MMP9 and MMP25) (Fig. 6C; Fig. S5A,B). Immunoblotting of pFAK, vinculin and paxillin and immunostaining of paxillin supported defective cell–ECM adhesion in Snail^6SA monolayers (Fig. S6A–F). It was interesting to note that Snail^6SA cells also showed an elevated number of invadopodia per cell (Fig. S6G–I), something associated with loss of cell–ECM adhesion and matrix remodelling in other contexts (Chan et al., 2009).

To test whether loss of cell–ECM adhesion was sufficient to trigger basal delamination, we then depleted paxillin in MCF-7 monolayers using siRNA (Fig. S6J). Basal delamination was moderately increased in paxillin RNAi monolayers (Fig. S6K,L), but mosaic paxillin knockdown did not cause apical extrusion (Fig. S6O,P). This suggested that loss of cell–ECM adhesion might contribute to basal delamination, although delamination rates were lower in paxillin knockdown monolayers than seen in Snail^6SA monolayers (Fig. 1F,G). However, as Snail^6SA cells displayed both increased RhoA signalling and downregulated cell–ECM adhesion, we wondered whether this combination would further enhance basal delamination. To test this, we expressed MRLC^DD in cells depleted of paxillin to simultaneously induce hypercontractility and reduce cell–ECM adhesions. This combination substantially enhanced sporadic delamination in MCF-7 monolayers (Fig. 6D,E) to levels similar to those observed in Snail^6SA monolayers (Fig. 1F,G; Movie 2). To test this further, we dampened contractility in Snail^6SA monolayers by expression of MRLC^AA. Expression of MRLC^AA in Snail^6SA monolayers suppressed delamination to a minimal level, as seen in control conditions (Fig. 6F,G). These findings imply that the combination of enhanced contractility and loss of cell–ECM adhesion is an important driver of basal delamination in Snail^6SA monolayers.

Snail is a major driver of epithelial morphogenesis during development, and its aberrant expression contributes to pathogenetic processes such as malignant invasion and migration (Alberga et al., 1991; Nieto, 2002; Peinado et al., 2007). Our current results identify cell extrusion as an additional element in the morphogenetic armamentarium of Snail, but one whose manifestation crucially depends on the context of its expression. Mosaic expression of Snail^6SA led to apical extrusion of the affected cells, but when expressed in larger groups, Snail^6SA cells were initially retained in the monolayer before undergoing sporadic basal delamination. We found that both forms of extrusion involve the transcriptional activation of RhoA-based cell contractility. Furthermore, it is important to emphasize that whereas apical extrusion was a consistent consequence of mosaic Snail^6SA expression (affecting >80% cells), basal delamination from Snail^6SA monolayers was a more sporadic phenomenon (affecting ∼20% of transgene-expressing cells). This suggests that additional factors might complement the action of contractility to influence the direction of extrusion in these different culture conditions.

Although Snail is best understood as an inducer of EMT (Wu and McClay, 2007; Theveneau and Mayor, 2012), our transcriptomic profiling indicated that Snail^6SA cells largely retained their epithelial identity, both when they were extracted from monolayers and also after they had been extruded. Therefore, EMT did not explain the basis for Snail-induced cell extrusion. Instead, our data identify a key role for cell contractility. Snail has been implicated in promoting contractility during developmental processes, such as apical constriction (Martin et al., 2009; Martin et al., 2010; Weng and Wieschaus, 2016). However, the mechanisms responsible for this latter effect are poorly characterized. Here we found that for the MCF-7 epithelial cell system, enhanced contractility can be explained by modulation of a transcriptional network that ultimately stimulates RhoA signalling. This network exerts its effects by promoting the upregulation of RhoA itself and also by tuning the expression of diverse GEFs and GAPs, with the outcome predicted to skew towards RhoA activation. Consistent with its canonical role in promoting cell contractility, the upregulation of RhoA was accompanied by an enhanced actomyosin cortex that generated mechanical tension.

Importantly, contractility was necessary for both apical and basal extrusion of Snail^6SA cells to occur, as both forms of extrusion were inhibited by coexpression of an inhibitory MRLC^AA transgene. Similarly, contractile stimulation of apoptotic cells is thought to contribute to their apical extrusion (Lubkov and Bar-Sagi, 2014; Michael et al., 2016) and hypercontractility was reported to trigger delamination in embryos (Liu and Jessell, 1998). Polarized patterns of contractility have also been proposed as playing a role in defining the direction in which cells extrude, with apical contractility thought to promote basal delamination and basal contractility promoting apical extrusion (Marshall et al., 2011; Slattum and Rosenblatt, 2014; Gu et al., 2015). However, RhoA and the actomyosin cortex were enhanced in multiple locations within Snail^6SA cells. Although readily evident at adherens junctions, consistent with them being prominent sites of RhoA signalling under control circumstances (Ratheesh et al., 2012; Priya et al., 2015), active RhoA and actomyosin were also enhanced at the apical surfaces of cells. Thus, the transcriptional network elicited by Snail^6SA may enhance cortical contractility quite extensively within the cells, rather than creating polarized patterns that are sufficient to determine the direction of extrusion. This is consistent with our observation that expression of a pro-contractile MRLC^DD transgene was not sufficient to elicit either apical extrusion or basal delamination.

Nevertheless, the notion that additional factors contribute to defining the directionality of Snail-induced extrusion is supported by our finding that the downregulation of cell–ECM interactions collaborates with RhoA-enhanced contractility to induce basal delamination. Thus, our transcriptomic analysis revealed that Snail^6SA expression modulated a suite of genes that are predicted to regulate integrin-based adhesion and enhance turnover of the ECM. Consistent with this, focal adhesions were less apparent in Snail^6SA cells, and paxillin RNAi modestly increased basal delamination. However, it was the combination of paxillin RNAi and MRLC^DD that most strikingly enhanced delamination, thus implicating a crucial role for both cortical contractility and basal adhesion in this process. This implies that Snail^6SA elicits a transcriptional program whereby both enhanced contractility and downregulated cell–ECM adhesion participate in basal delamination. It is interesting to consider whether this combination of changes may be co-opted by oncogenic cells to become invasive (Slattum et al., 2014), despite resistive E-cadherin junctions (Fig. 6H).

In conclusion, we propose that extrusion represents a new way for Snail to regulate epithelial behaviour, independent of its canonical role in EMT. Currently, it is less clear what may collaborate with RhoA to promote apical extrusion when Snail is expressed mosaically, since paxillin RNAi did not induce any degree of apical extrusion. It is interesting to note that MLCK^DD alone did not induce extrusion in our experiments but other pro-contractile signals that can induce apical extrusion also affect F-actin dynamics. These include constitutively active ROCK or MRCK mutants (Lubkov and Bar-Sagi, 2014; Michael et al., 2016; Gagliardi et al., 2018). Therefore, it is possible that Snail alters other modes of cytoskeletal dynamics to help promote apical extrusion. In addition, the balance of forces at the tissue level could contribute to this phenomenon. Mosaic expression of Snail^6SA would be predicted to cause local upregulation of contractility whereas its ubiquitous expression would be expected to lead to an overall increase in tissue tension that was balanced throughout the monolayer. E-cadherin-based adherens junctions possess mechanotransduction pathways that can respond to mechanical stresses (Acharya et al., 2018) and thus carry the potential to respond to local changes induced by mosaic Snail^6SA. We propose that these are important questions for future research.

MCF-7 human breast epithelial adenocarcinoma cells were cultured in DMEM complete growth medium (Gibco) supplemented with 10% FBS (Invitrogen), 100 units/ml penicillin (Gibco) and 100 units/ml streptomycin (Gibco). MCF-7 cells stably expressing NLS^mCherry or mCherry-Snail^6SA were cultured in the same medium as parental MCF-7 cells, supplemented with 60 ng/ml of puromycin. MCF-7 cells stably expressing MRLC^WT-GFP, MRLC^AA-GFP or MRLC^DD-GFP were cultured in the same medium as parental MCF-7 cells. Expression of inducible transgenes was achieved by culturing cells with 3 µg/ml of doxycycline for at least 48 h. Caco-2 human colorectal epithelial carcinoma cells were cultured in RPMI complete growth medium (Gibco) supplemented with 10% FBS (Invitrogen), 100 units/ml penicillin (Gibco), 100 units/ml streptomycin (Gibco), 2 mM L-glutamine (Gibco) and 1% non-essential amino acids (Gibco). HEK-293T cells were cultured in the same medium as parental MCF-7 cells. All cells were cultured at 37°C with 5% CO₂. RNAiMax (Invitrogen) and Lipofectamine 2000 (Invitrogen) were used for transfection of siRNA and plasmids respectively, according to the manufacturer's instructions. All analyses of transfected cells were carried out at 48 h post-transfection, in a confluent monolayer.

NLS^mCherry and mCherry-Snail^6SA in pmCherry-C1 (Clontech) were subcloned into the pTripz Tet-ON lentiviral expression vector (GE-Dharmacon), which had been modified to contain a human PGK promoter to drive the expression of rtTA3, using the AgeI and ClaI restriction sites (Table S3).

MRLC^WT-GFP, MRLC^AA-GFP and MRLC^DD-GFP in pmCherry-C1 or pEGFP-C1 (Clontech) were subcloned into the pLL5.0 lentiviral expression vector using the BsrGI and BamHI restriction sites (Table S3). The pLL5.0 lentiviral vector and packaging constructs pMDLg/pRRE, RSV-Rev and pMD.G were gifts from Dr Jim Bear (UNC Chapel Hill, North Carolina, USA). To generate the above cell lines, pTripz or pLL5.0 vectors were transfected together with the packaging plasmids into HEK-293T cells using Lipofectamine 2000. Medium was replaced 16 h post-transfection and the medium containing virus particles was then collected at 72 h post-transfection. This medium was then used to infect cells supplemented with 8 μg/ml of polybrene. Cells infected with pTripz viral particles were cultured in medium containing 0.5 μg/ml puromycin for 1 week to positively select for transduced cells. The RhoA location biosensor (GFP-AHPH) was a kind gift from M. Glotzer (University of Chicago, USA) and is described previously (Priya et al., 2015; Liang et al., 2017). To generate a stable cell line, MCF-7 cells were transfected with GFP-AHPH using Lipofectamine 2000 and treated with G418 (0.5 mg/ml) for selection of positive cell clones. Cells expressing NLS^mCherry or Snail^6SA were transiently transfected with GFP-AHPH using Lipofectamine 2000 to analyse for active junctional RhoA.

MCF-7 cells stably expressing NLS^mCherry or mCherry-Snail^6SA were treated with 3 μg/ml of doxycycline for 48 h to induce expression of transgenes. Cells were cultured to 95% confluency and total RNA was extracted using a RNA extraction kit (Sigma) according to the manufacturer's instructions. Purified RNA was enriched for mRNA, and library preparation was performed using the TruSeq stranded mRNA kit by the IMB Sequencing Facility (University of Queensland, Institute of Molecular Bioscience). mRNA samples were then sequenced using the NextSeq 75 cycle kit (1×75 bp) on Illumina NextSeq 500.

For qPCR, RNA was extracted from cells expressing NLS^mCherry or mCherry-Snail^6SA as described above. Purified RNA (5 μg) from each sample was used to convert to cDNA using the SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen) according to the manufacturer's instructions. Quantification of cDNA samples were performed in a 96-well plate compatible with the ViiA7 Real-Time PCR System (ThermoFisher Scientific). The transcript level of each gene target was assessed in triplicate with specific primers (Table S3). In each replicate well, 4 μl of cDNA was added to 0.5 μl each of forward and reverse primers (Table S3) and 5 μl of SYBR Green mastermix (Applied Biosystems). The 96-well plate was then briefly vortexed and centrifuged at 112 g for 2 min at 4°C. qPCR of the 96-well plate was performed using the ViiA7 PCR system and the output data further analysed in Excel as outlined (Tables S1 and S2). In brief, the CT (cycle threshold) value indicates the number of PCR cycles required for the fluorescent signal of the target to cross the baseline threshold. Thus, the CT value obtained from the ViiA7 system is inversely proportional to the amount of mRNA target in the sample. The CT mean value was used as a measurement of relative mRNA expression for each target (Table S1). Relative mRNA levels for each target were then normalized to those of GAPDH in each condition. The fold change of normalized mRNA target in each condition was then calculated and normalized to control conditions as outlined (Table S2).

Single-End reads of 35-76 bp in length were quality checked using Fastqc (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were mapped to hg38 using the Subread aligner (Liao et al., 2013). The aligned reads were summarized at the gene level using featureCounts (Liao et al., 2014). Differential gene expression analysis, GO and KEGG pathway enrichment analysis were carried out using the R/Bioconductor package edgeR (Robinson et al., 2010) and limma (Ritchie et al., 2015). The average counts-per-million (cpm) values for the EMT gene signatures (cell lines) given by Tan et al. (2014) were computed across the three biological replicates for NLS^mCherry (Ctrl), Snail^6SA and Snail^6SA (extruded) phenotypes (Foroutan et al., 2018). The gene signatures characteristic of epithelial and mesenchymal phenotypes in all cancer cell lines and tumours were generated by Tan et al. (2014). They have also described a detailed method for scoring cell lines and tumours for EMT phenotypes and have provided the gene sets. The ternary plot was produced using the R package vcd. Specific gene sets for enrichment and visualization were selected based on KEGG pathway membership or through annotation with relevant Gene Ontology terms.

The primary antibodies used for immunofluorescence (IF) and western blotting (WB) were as follows: rabbit polyclonal antibody against α-catenin (Invitrogen: #71-1200; IF 1:50); mouse monoclonal antibody against anillin (Santa Cruz: #sc-271814; IF 1:50, WB 1:500); rabbit polyclonal antibody against ARHGEF28/p190GEF (Life Technologies: #PA558032; WB 1:1000); rabbit polyclonal antibody against ARHGEF31/Ect2 (Merck: #07-1364; IF 1:100, WB 1:1000); mouse monoclonal antibody against ARHGAP5/p190BGAP (BD Bioscience: #611612; IF 1:50, WB 1:1000); rabbit polyclonal antibody against ARHGAP6 (Thermo Fisher Scientific: #PA564170; WB 1:1000); mouse monoclonal antibody against E-cadherin (a gift from P. Wheelock, University of Nebraska, USA; with permission of M. Takeichi; IF 1:50, WB 1:1000); mouse monoclonal antibody against GFP (Rhoche: #11814460001; IF 1:250, WB 1:2000); rabbit polyclonal antibody against GAPDH (Trevigen: #2275; WB 1:10,000); rabbit polyclonal antibody against mCherry (Biovision: #5993; IF 1:250, WB 1:2000); rat monoclonal antibody against mCherry (Thermo Fisher Scientific: #M11217; IF 1:250, WB 1:2000); mouse monoclonal antibody against N-cadherin (Cell Signaling: #14215; WB 1:1000); rabbit polyclonal antibody against myosin IIA (Sigma: #M8064; IF 1:100, WB 1:1000); rabbit polyclonal antibody against myosin IIB (Covance: #PRB-445P; IF 1:100, WB 1:1000); rabbit polyclonal antibody against Par3 (Millipore: #07-330; WB 1:1000); rabbit polyclonal antibody against phospho-myosin light chain 2 (Ser19) (Cell Signaling: #3675; WB 1:1000); mouse monoclonal antibody against RhoA (Santa Cruz: #SC418; IF 1:50, WB 1:500); rabbit monoclonal antibody against Snail (Cell Signaling: #3879; IF 1:100, WB 1:1000); rabbit monoclonal antibody against vimentin (Cell Signaling: #5741; WB 1:1000); rabbit polyclonal antibody against ZO-1 (Invitrogen: #61-7300; IF 1:100, WB 1:1000).

Secondary antibodies used for immunoblotting were anti-rabbit (Bio-Rad: #1706515; 1:4000) or anti-mouse (Bio-Rad: #170516; 1:4000) conjugated with horseradish peroxidase (HRP). Secondary antibodies for immunofluorescence were as follows: anti-rabbit AlexaFluor488 (Invitrogen: #A11034; 1:400), anti-mouse AlexaFluor488 (Invitrogen: #A11029; 1:400), anti-rabbit AlexaFluor546 (Invitrogen: #A11035; 1:400), anti-rat AlexaFluor546 (Invitrogen: #A11081; 1:400), anti-rabbit AlexaFluor647 (Invitrogen: #A21245; 1:400) and anti-mouse AlexaFluor647 (Invitrogen: #A21236; 1:400). Phalloidin conjugated with AlexaFluor647 was used to stain for F-actin (Thermo Fisher Scientific: #A22287; 1:400).

Cells cultured in six-well plates were placed on ice, washed with ice-cold PBS once, and then lysed with either RIPA lysis buffer or SDS sample lysis buffer. Cell lysates were scraped down, resolved on 10% or 12% SDS-PAGE gels and transferred onto a nitrocellulose membrane. Nitrocellulose membranes were then stained with Ponceau S to visually check equal loading of proteins before blocking with 5% milk diluted in Tris-buffered saline with 0.2% Tween-20 for 1 h at room temperature. Specific primary antibodies were then diluted in blocking buffer at the specified dilution as described earlier, added to pre-blocked membranes and left overnight at 4°C. Membranes were washed with TBST three times, 20 min each, at room temperature. Specific secondary antibodies tagged with HRP were diluted in blocking buffer at 1:4000 dilution and incubated with the membrane for 1 h at room temperature. The membrane was then washed three times, 20 min each, at room temperature followed by detection of specific signals by enhanced chemiluminescence (ECL) at room temperature. Luminescence was detected in the dark and captured on highly sensitive X-ray films.

Cells cultured to 95% confluency were fixed with either 4% paraformaldehyde (PFA) in cytoskeletal stabilization buffer for 20 min at 37°C, 100% methanol for 5 min on ice or 10% TCA in PBS for 20 min on ice. For PFA fixation, samples were permeabilized with 0.25% Triton-X-100 for 5 min at room temperature. For TCA fixation, samples were washed three times with 30 mM glycine in PBS, followed by permeabilization with 0.25% Triton-X-100. Fixed samples were blocked for 1 h at room temperature with 5% milk or 3% BSA in PBS, followed by incubation of diluted specific primary antibodies overnight at 4°C in a humid chamber. Samples were then washed with PBS, incubated with diluted specific secondary antibodies for 1 h at room temperature and mounted on glass slides with Prolong Gold containing DAPI (Cell Signaling; #8961). Confocal images were acquired on a Zeiss LSM 710 (63×, 1.4NA Plan Apo objective) using the provided ZEN software. Multiple Z stacks (0.4 μM increment) spanning the height of the sample (cells) were acquired for each region that was imaged.

Quantification of fluorescence intensity at junctions was performed using the line scan function in ImageJ as previously described (Michael et al., 2016). In brief, a line of 20 pixels in width was drawn orthogonal to the cell junction and the pixel intensity obtained using the Plot Profile function. The peak value for each junction was obtained and corrected by subtracting background fluorescence intensity from either end of the measured line. The resulting value is representative of the fluorescence intensity at the measured junction. Junctional fluorescence intensity from a minimum of 20 different junctions was compiled and analysed as one independent experiment per condition. The mean peak fluorescence value was then obtained and calculated from three independent experiments and compared between groups (mean of means). To quantify junctional active RhoA, the fluorescence intensities of junctional and cytoplasmic GFP-AHPH in cells expressing NLS^mCherry or Snail^6SA were measured as described above. GFP-AHPH junctional fluorescence intensity was then corrected for differences in expression levels with the cytoplasmic GFP-AHPH levels for each cell. The average of corrected GFP-AHPH junctional fluorescence values were then compared between groups.

Confocal images of GFP-AHPH in the apical cortical surface area (ROI) were acquired as above. The total amount of GFP-AHPH fluorescence within the ROI was measured by drawing an irregular polygon around the ROI and quantifying using the measure function in ImageJ. The resulting value represents the mean fluorescence intensity (arbitrary unit) of GFP-AHPH in the given ROI surface area. Fluorescence intensities from at least 20 different ROIs were compiled and analysed as one independent experiment per condition. The mean fluorescence value was then obtained and calculated from three independent experiments and compared between groups (mean of means).

Preparation of samples for immunostaining and confocal imaging were performed as above. Cells were stained with tubulin for visualization of microtubules during mitosis division events, and the orientation of cell division was identified by visualizing microtubules spatially in planar (XY) or orthogonal (XZ) view using ImageJ orthogonal views function. Occurrence of symmetric (planar) cell division was identified when astral microtubules are parallel to the epithelium plane, whereas asymmetric (orthogonal) cell division occurs when astral microtubules are perpendicular (90°) to the epithelium plane. The number of cells undergoing either planar or orthogonal orientated mitosis was counted by eye and measured as a percentage of the total number of cells in the confocal image.

Preparation of samples for immunostaining and confocal imaging were performed as above. Cells were stained with E-cadherin for measurement of cell size and with DAPI for cell number counting. Cell size was quantified by drawing an irregular polygon around the ROI within E-cadherin junctions and quantified using the measure function in ImageJ. The resulting value represents the apical cortical surface area of the individual cell (in microns). A minimum of 20 different confocal images for each condition was used, with all cells within each image measured and analysed as one independent experiment per condition. The mean cell size was then obtained and calculated from three independent experiments and compared between groups (mean of means). The total number of cells in confocal images was counted by eye and measured as a ratio of the total area of the confocal image, fixed at 512×512 pixels for all images. The resulting value represents the cell density within the confocal image (cell number per area of confocal image). A minimum of 20 different confocal images were used and analysed as one independent experiment per condition. The mean density of cells was obtained and calculated from three independent experiments and compared between groups (mean of means).

Homogenous populations of MCF-7 cells expressing either NLS^mCherry or Snail^6SA were cultured to 95% confluency and fixed with 4% PFA as described earlier. Phalloidin staining was used to identify F-actin in the invadopodial structures at the basal region of the cell, beneath the nucleus. Confocal images were then acquired on a Zeiss LSM 710 (63×, 1.4NA Plan Apo objective) using the provided ZEN software. Multiple Z stacks (0.4 μM increment) spanning the height of the sample (cells) were acquired for each region that was imaged. Invadopodia-like structures were identified as F-actin puncta accumulation beneath the nucleus. F-actin punctas in Snail^6SA cells were counted and compared with those of control NLS^mCherry cells.

MCF-7 cells expressing either NLS^mCherry or Snail^6SA were cultured for 5 h at low density on glass cover slips to achieve isolated single cells. Cells were then fixed and stained with paxillin for analysis by immunofluorescence. Paxillin punctas around the periphery of cells expressing NLS^mCherry or Snail^6SA (white box in Fig. S6E) were isolated using freehand selections on ImageJ and measured for average fluorescence intensity in the isolated area.

We used the α-18 antibody, a generous gift from the Nagafuchi lab (Nara Medical University, Japan), as a proxy of changes in junctional tension generated by contractile forces at the adherens junction that specifically binds to the ‘exposed’ central domain of the α-catenin protein at cell–cell contacts (Yonemura et al., 2010). Cells were cultured to 95% confluency, fixed with methanol and co-stained with α-18 and α-catenin antibodies at the specified concentration, as described earlier. Junctional fluorescence intensity was imaged by confocal microscopy. The junctional fluorescence intensity of α-18 was calculated as the ratio of α-catenin fluorescence intensity to that of Snail^6SA cells normalized against control NLS^mCherry cells.

The pTriEx-RhoA Biosensor wild-type construct was obtained from addgene, deposited by the laboratory of Klaus Hahn (Pertz et al., 2006) and used for RhoA-FRET experiments. pTriEx-RhoA was co-transfected with either NLS^mCherry or Snail^6SA in MCF-7 cells cultured on glass-bottom dishes. The next day, the medium was replaced by fresh medium containing doxycycline and cells incubated for 48 h to induce expression of NLS^mCherry or Snail^6SA. Live cells were then imaged on a Zeiss LSM 710 microscope (63×, 1.40NA Plan Apo objective). Donor and FRET fluorescence channels were excited using the 458 nm laser and the emission of donor was collected at 470-490 nm; the emission of acceptor molecule during FRET was obtained at 530-590 nm as previously described (Priya et al., 2015; Liang et al., 2017). The images were processed in ImageJ to split the fluorescence channels, generate an image stack of various junctions and create a mask of the apical adherens junction for each image in order to calculate the pixel intensity for various channels (donor, acceptor and FRET). The donor and FRET emission at pixels localized at apical cell–cell junctions were used to calculate the emission ratio of FRET/donor using a custom-made MATLAB script described earlier and available upon request (Ratheesh et al., 2012; Priya et al., 2015).

To assess apical extrusion, cells stably expressing NLS^mCherry or Snail^6SA were counted and mixed with wild-type cells in a 1:100 ratio and plated at approximately 80% confluency to obtain a mosaic population of a single transgenic cell surrounded by wild-type cells. To generate islands or clusters of transgenic cells, NLS^mCherry- or Snail^6SA-expressing cells were mixed with wild-type cells in a 1:50 ratio and plated at 80% confluency. Transgene expression was then induced by treating with 3 μg/ml of doxycycline for 48 h, and the confluent monolayer fixed for analysis by immunofluorescence. DAPI and mCherry were used to identify extruded cells with their nuclei out of the apical plane of the monolayer. E-cadherin was used to identify the apical plane of the monolayer. For live imaging of apical extrusion, MCF-7 cells in which the endogenous E-cadherin was tagged with GFP (using the CRISPR/Cas9 genome editing system) were cultured on glass-bottom dishes and transfected with either NLS^mCherry or Snail^6SA with Lipofectamine 2000 and imaged after 48 h. Time-lapse live cell imaging of the confluent monolayer was performed on a Nikon Ti-E deconvolution microscope (40×, 0.5NA Plan Apo objective), equipped with a 37°C, 5% CO₂ chamber, using the NIS-Elements AR software (Nikon). Images were acquired at 30 min intervals for 24 h.

To assess basal extrusion, cells stably expressing NLS^mCherry or Snail^6SA were treated with 3 μg/ml of doxycycline for 48 h and cultured to approximately 90-100% confluency. E-cadherin was used to identify the apical and baso-lateral regions of the monolayer. DAPI and mCherry immunofluorescence were used to identify basally extruded cells with their nuclei below the basal plane of the monolayer. For live imaging of basal extrusion, MCF-7 cells expressing endogenous E-cadherin tagged with GFP were infected with either NLS^mCherry or Snail^6SA viral particles to generate stable cell lines. Time-lapse live cell imaging of cells stably expressing either E-cad-GFP-NLS^mCherry or E-cad-GFP-Snail^6SA were acquired as described above for live imaging of apical extrusion.

95% confluent cells were treated with 250 nM of etoposide for 5 h and analysed by immunofluorescence as described previously (Michael et al., 2016). Cleaved caspase-3 and DAPI were used to identify apoptotic Caco-2 cells whereas annexin-V labelling and DAPI were used to identify apoptotic MCF-7 cells. Extruded cells were identified as annexin-V or cleaved caspase-3 positive spherical cells with their nuclei out of the apical plane of the monolayer. E-cadherin was used to identify the monolayer.

Pulldown of GTP–RhoA was performed with the RhoA Pulldown Activation Assay Kit (#BK036, Cytoskeleton) as described previously (Priya et al., 2015). Cells expressing NLS^mCherry or Snail^6SA were treated with 3 μg/ml of doxycycline for 48 h and lysed at 95% confluency. The cells were then washed with ice-cold PBS, lysed and processed for the pulldown assay according to the manufacturer's instructions.

The use of a two-photon laser to assess junctional tension has been described previously (Leerberg et al., 2014; Wu et al., 2014; Michael et al., 2016; Liang et al., 2017). Adherens junctions of MCF-7 cells were labelled by tagging endogenous E-cadherin with GFP using the CRISPR/Cas9 genome editing system and were used to identify the apical cell–cell contacts within a monolayer. NLS^mCherry or Snail^6SA were expressed in these genome-edited cells, either stably or transiently for 48 h, within a confluent monolayer. For the measurement of junctional tension in MRLC^AA and MRLC^WT cells, MCF-7 cells were transduced with either MRLC^AA-GFP or MRLC^WT-GFP lentiviral particles to generate stable cell lines. Junctional localization of MRLC^AA and MRLC^WT was used to identify apical cell–cell contacts. All ablation experiments were performed on a Zeiss LSM-510 META confocal microscope (63×, 1.4NA Plan Apo objective) with a 37°C heating stage. Recoil of junctional vertices was recorded by time-lapse imaging at 10 frames with 5 s interval between each frame. Analyses of data were performed on ImageJ as previously described (Michael et al., 2016; Liang et al., 2017). Although recoil measurements are a commonly used index for tension, changes in recoil can be caused by changes in tension and/or changes in the material properties of the junction (such as viscous dissipation or stiffness). Accordingly, we and others such as Fernandez-Gonzalez et al. (2009) have modelled recoil data as a one-dimensional Kelvin–Voigt fibre and used the value of k (elasticity/viscosity coefficient) as proxy for the material properties of the junction (Michael et al., 2016). The observation that k-values do not change suggests that the change in recoil principally reflects changes in tension.

We want to thank our laboratory colleagues, past and present, for their immense support and advice. Optical microscopy was performed at the ACRF/IMB Cancer Biology Imaging Facility, established with the generous support of the Australian Cancer Research Foundation.