Mechanical cell competition kills cells via induction of lethal p53 levels

Cell competition is a quality control mechanism that eliminates unfit cells. How cells compete is poorly understood, but it is generally accepted that molecular exchange between cells signals elimination of unfit cells. Here we report an orthogonal mechanism of cell competition, whereby cells compete through mechanical insults. We show that MDCK cells silenced for the polarity gene scribble (scribKD) are hypersensitive to compaction, that interaction with wild-type cells causes their compaction and that crowding is sufficient for scribKD cell elimination. Importantly, we show that elevation of the tumour suppressor p53 is necessary and sufficient for crowding hypersensitivity. Compaction, via activation of Rho-associated kinase (ROCK) and the stress kinase p38, leads to further p53 elevation, causing cell death. Thus, in addition to molecules, cells use mechanical means to compete. Given the involvement of p53, compaction hypersensitivity may be widespread among damaged cells and offers an additional route to eliminate unfit cells.

It has recently been reported that Madin-Darby canine kidney (MDCK) epithelial cells silenced for the polarity gene scribble (scrib KD cells) are eliminated in the presence of wild-type MDCK cells 23 , while they are viable on their own 23 . However, the mechanisms by which scrib KD cells are killed by wild-type cells are largely unknown. We therefore took advantage of this recent observation to investigate the mechanisms of cell competition.
Here we show that scrib KD cells are out-competed by wild-type cells through mechanical insults rather than molecular exchange. We find that scrib KD cells are hypersensitive to compaction and that this is due to elevation of baseline p53 levels, which is both necessary and sufficient to induce hypersensitivity to crowding and confer a mechanical loser status. We further show that on contact with wild-type cells, scrib KD cells become compacted into a high-density arrangement and that compaction is not only required but also sufficient to eliminate scrib KD cells. We also delineate the mechano-transduction cascade that leads to scrib KD cell death. Specifically, we show that scrib KD cells' compaction causes activation of the Rho-associated kinase (ROCK), which in turn activates p38 leading to further p53 elevation and cell death. Overall, this work demonstrates that mechanical forces can be responsible for the elimination of cells during cell competition and that p53 levels play a key role both in instructing the mechanical loser status and in the execution of mechanical cell competition.

Results
Compaction of scrib KD cells induces mechanical competition. It has previously been shown that scrib KD MDCK cells are eliminated when co-cultured with wild-type MDCK cells through cell death and delamination (see ref. 23 and Supplementary Fig. 1a and Supplementary Movie 1, left), while monocultures of scrib KD cells are viable (see ref. 23 and Supplementary Fig. 1b and Supplementary Movie 1, right). To investigate the mechanisms of scrib KD -mediated cell competition, we first asked whether it is mediated by soluble factors, as in other cases of in vitro cell competition 6,24 . Growth rate (doubling time) profiles showed that scrib KD cells in pure cultures divide, albeit at a reduced rate, to reach a steadily maintained number ( Supplementary Fig. 1d), whereas under competing conditions, their numbers collapse following initial growth (Fig. 1a). Interestingly, we found that the growth rate of scrib KD cells is not affected by conditioned medium from competing cultures ( Fig. 1b and Supplementary  Fig. 1c). Similarly, in transwell systems that allow exchange of solutes but prevent cell contact, scrib KD cells grown together with co-cultures of competing (wild-type/scrib KD ) cells grew comparably to scrib KD cells grown with other scrib KD cells ( Fig. 1c and Supplementary Fig. 1c). This indicated that soluble factors are not sufficient to induce cell competition and that cell contact is required. We hypothesized that cell contact enables molecular interactions essential for cell competition, as observed by others 11,12 . However, to our surprise, we found that sustained contact with wild-type cells is not sufficient for elimination of scrib KD cells (Fig. 1d, black arrow and Supplementary Movie 2) and that scrib KD clones are efficiently eliminated only when fully surrounded by wild-type cells (Fig. 1d, white arrow and Supplementary Movie 2). This suggested that a type of exchange other than molecular signalling (which would be enabled by contact) may be needed, and prompted us to look for differences between scrib KD clones that were surrounded and peripheral clones that were simply contacted.
One striking feature of surrounded clones, which is not shared by peripheral clones, is that they reach a dramatically higher cell density than confluent scrib KD pure cultures ( Fig. 1e-g and Supplementary Fig. 1e). scrib KD cells acquire a flattened morphology upon gene silencing 23,25 , which at confluence results in a much lower (B1/3) final density compared with wild-type cells (Fig. 1e-g). However, scrib KD clones surrounded by wild-type cells do not flatten and reach a density that is B4.5fold higher than that of pure scrib KD cultures ( Fig. 1e-g). Furthermore, competing scrib KD cells are taller than when grown in single cultures (Fig. 1h-j). Together, this indicates that as a result of their interaction with wild-type cells, scrib KD cells become more compacted than in their default state and this correlates with and precedes their elimination.
We next asked whether cell compaction plays a role in scrib KD cell elimination, by assessing its effect on scrib KD cells in complete absence of wild-type cells. To that end, we used micropatterns to form scrib KD microcultures of a defined, homogenous density and size (+ ¼ 800 mm, ref. 26). When plated at high density, control cells that had not undergone scribble silencing (without tetracycline, À TET) continued to grow until they reached a maximal density of 61 ± 4 cells per 10,000 mm 2 , which they maintained homeostatically (Fig. 1k,l and Supplementary Movie 3, left), as previously shown 27,28 . Remarkably, scrib KD cultures ( þ TET) instead saw their numbers and densities drop (Fig. 1m,n, dark green and Supplementary Movie 3, right), due to a combination of increased cell death and extrusion, the same two events that lead to the elimination of scrib KD cells during cell competition (Supplementary Movie 1, left and ref. 23). Notably, scrib KD cells seeded at a lower density maintained that initial density (Fig. 1n, light green) indicating that the drop in cell number observed at higher density is not a general response of scrib KD cells to plating. This suggested that scrib KD cells are hypersensitive to compaction. To test this directly, we seeded confluent monolayers of control ( À TET) or scrib KD ( þ TET) cells on stretched polydimethylsiloxane (PDMS) substrates, which we then released to induce cell compression 27 . Control cultures showed no increase in apoptosis upon compression ( Fig. 1o-q). The scrib KD cells, however, displayed a 3-fold increase in apoptosis over their uncompressed baseline cell death, which was already higher than control (Fig. 1p,q). Altogether, these experiments show that scrib KD cells display hypersensitivity to cell density and cannot sustain levels of crowding normally reached by wild-type cells. They further indicate that compaction of scrib KD cells into higher cell densities, like those imposed on them by wild-type cells during competition, is sufficient to induce cell death. This suggests that wild-type cells eliminate scrib KD cells through crowding-induced compaction. We name this new mode of selective elimination of one cell population by another due to differential sensitivity to crowding 'mechanical cell competition', to contrast it with forms of cell competition that rely on molecular exchange.
Corralling promotes but is not needed for scrib KD elimination. As shown in Fig. 1f,g, in subconfluent competing cultures, the density of scrib KD cells surpasses that of surrounding wild-type cells. This indicates that their acquired density does not simply reflect the density of the mixed culture, but is the result of an active process. To understand how scrib KD cells become compacted during competition, we plated cells at low density and observed what happens when wild-type and scrib KD clones first come into contact. Strikingly, we saw that, on contact, both scrib KD and wild-type cells engage in collective cell migration ( Fig. 2a and Supplementary Movie 4). This behaviour was specific to wild-type/scrib KD encounters, as it was not observed upon homotypic encounters of either cell population ( Fig. 2b and Supplementary Movie 5). Interestingly, this migration was highly directional with scrib KD cells always at the migrating front and, conversely, wild-type cells always at the back. We next characterized this migratory behaviour and assessed its contribution to cell compaction and elimination. First, we repeated the above experiment using a fluorescent nuclear label in each population to facilitate cell tracking (Fig. 2c). Kymograph analysis showed rapid activation of collective cell migration at the time of contact between wild-type and scrib KD cells, with both populations moving synchronously (Fig. 2d, top and Fig. 2e). By   contrast, similar kymographs of homotypic collisions did not show significant cell displacement (Fig. 2d, centre and bottom; and Fig. 2e). To further characterize the features of this collective cell migration, we carried out single-cell tracking of both cell populations before and after contact. The analysis of individual cell trajectories shows that single-cell movement is faster and more persistent for both wild-type and scrib KD cells upon heterotypic collision ( Fig. 2f and Supplementary Fig. 2a-d). We   then looked more closely at the dynamic interplay between wildtype and scrib KD cells at the onset of migration. Interestingly, analysis of cell shape at the interface between the two populations revealed that both wild-type and scrib KD cells become elongated at the site of contact (Fig. 2g,h), indicating that they are under anisotropic stress. Moreover, the cells in both populations move in the direction of their short axis (Fig. 2i), suggesting that they could be locally compressed. This could be the result of wild-type cells beginning migration and locally 'piling up' against the scrib KD cells. In agreement with this, particle image velocimetry (PIV) and single-cell tracking revealed that wild-type cells begin migrating towards scrib KD cells on average B2 h before the scrib KD cells start migrating away (Fig. 2j,k and Supplementary Movie 6). Together, the chronology of these events and the local cellular deformations suggest that cells might engage in a behaviour similar to the 'chase and run' migration reported for other cell types 29 . These experiments do not distinguish whether scrib KD cells are pushed by wild-type cells or, conversely, whether scrib KD cells self-compact to avoid closer interaction wild-type cells. However, it is clear that as migration progresses, scrib KD cells become corralled by wild-type cells and are compacted and eliminated (Supplementary Movie 7), indicating that this behaviour may facilitate compaction and outcompetition of scrib KD cells.
To directly test the relevance of directional cell migration in the elimination of scrib KD cells, we next sought to disrupt this behaviour. Adhesion molecules play an important role in collective cell migration 30 and have also been implicated in 'chase and run' among mesenchymal cells 29 . In addition, scribble downregulation has previously been shown to induce intracellular accumulation of E-cadherin 23,31 and we found that it causes an increase of both total and cell surface E-cadherin levels (Supplementary Fig. 2e-g). We therefore reasoned that targeting E-cadherin-mediated adhesion might disrupt directional migration between the wild-type and scrib KD cells. We undertook two separate approaches and found that either blocking E-cadherin function in both populations, through a combination of low calcium and addition of a blocking antibody ( Having established that E-cadherin-mediated adhesion is involved in contact-induced migration between wild-type and scrib KD cells, we next asked whether the upregulation of E-cadherin observed in scrib KD cells (Supplementary Fig. 2e-g) plays a role in this directional cell movement. We therefore generated an inducible double scrib KD E-cadherin knockdown cell line (scrib KD E-cad KD ) and specifically selected clones that displayed partial silencing, enough to bring E-cadherin down to wild-type levels (Fig. 2q, right panel and Supplementary Fig. 2h,i). Indeed, partial downregulation of E-cadherin inhibited contactinduced migration, suggesting that high E-cadherin levels in the scrib KD cells are required for this process ( Fig. 2q and Supplementary Movie 9). As expected, these clones were still outcompeted by wild-type cells ( Supplementary Fig. 2j). In contrast, we found that E-cadherin upregulation alone is not sufficient to cause contact-induced migration, as cells overexpressing E-cadherin at levels comparable to those of scrib KD cells ( Supplementary Fig. 2k) did not engage in directional cell migration with wild-type cells upon contact ( Supplementary  Fig. 2l). Altogether, we conclude that directional cell migration, by enabling corralling and active compaction of scrib KD cells, promotes but is not required for mechanical cell competition, and that E-cadherin is necessary for corralling and active compaction but it does not impact on loser cell status.
p53 is activated in scrib KD cells before cell competition. Key to the outcompetition of scrib KD cells is their hypersensitivity to compaction ( Fig. 1k-q). To identify genes and pathways involved in this behaviour, we carried out transcriptional profiling of scrib KD cells (scrib KD þ TET) and compared it with the transcriptomes of control MDCK cells (scrib KD À TET) and, importantly, of an isolate of scrib KD cells that is resistant to cell competition (scrib RES ). Despite maintaining scribble gene knockdown ( Supplementary Fig. 3a), the scrib RES cells do not display elevated E-cadherin ( Supplementary Fig. 3b), do not engage in contact-induced migration with wild-type cells ( Supplementary Fig. 3c) and are not outcompeted in cell competition assays (Supplementary Fig. 3d and Supplementary Movie 10). Notwithstanding these fundamental differences, we observed that the scrib KD transcriptome is still substantially closer to scrib RES (with 523 differentially expressed genes) than to control (scrib KD À TET) cells (with 1,645 differentially expressed genes; Fig. 3a, left and Supplementary Data 1 and 2). This allowed us to rule out all genes that are differentially expressed in scrib KD versus wild-type cells but are similarly expressed between scrib KD and scrib RES , as these are clearly not sufficient to induce cell competition. Instead, we focused on the small intersection of 306 genes that are differentially expressed between wild-type and scrib KD cells but are also different between scrib KD and scrib RES cells ( Fig. 3a and Supplementary Data 3).
Gene Ontology term enrichment analysis highlighted p53 signalling as the top functionally enriched pathway (Fig. 3a, middle). A number of known p53 target genes were moderately upregulated in scrib KD cells, suggesting p53 activation (Fig. 3a, right). Consistent with this, we found that p21 (a known p53 target 33  p53 is further elevated in scrib KD cells by compaction. Next we asked whether cell competition affects p53 activity. Interestingly, we found that in competing conditions, p53 levels in scrib KD cells increase above their already elevated baseline level, in a way that correlates with the degree of cell compaction (Fig. 3c-e; r ¼ 0.56 by non-parametric Spearman correlation). This suggested that compaction increases p53 levels or, conversely, that higher p53 levels enable compaction. To distinguish between these two possibilities, we looked at how compression alone affects p53, using deformable PDMS substrates as before (Fig. 1o-q). Importantly, we found that compression alone causes an increase in p53 levels (Fig. 3f,g), as seen during competition. This indicates that cell competition induces further p53 activation in the scrib KD cells via compaction-induced mechanical stress.
Given that p53 is upregulated during cell competition, we next sought to ask whether it contributes to this process. Therefore we mutated p53 in scrib KD cells by CRISPR-mediated mutagenesis (scrib KD p53 À / À cells, Supplementary Fig. 4a,b). The scrib KD p53 À / À cells failed to upregulate p21 following ultraviolet irradiation ( Supplementary Fig. 4b) or scribble knockdown ( Supplementary Fig. 4c-e), confirming functional p53 inactivation. Strikingly, we found that p53 inactivation in scrib KD p53 À / À cells was sufficient to partially rescue their low homeostatic cell density (Fig. 3h) and their hypersensitivity to compaction (Fig. 3i). Remarkably, genetic (using scrib KD p53 À / À cells) or chemical (using the inhibitor Pifithrin-a) inhibition of e 0.  p53 in scrib KD cells was also sufficient to prevent their outcompetition ( Fig. 3j and Supplementary Movie 11; Supplementary Fig. 4f). Furthermore, scrib KD p53 À / À cells maintained high E-cadherin levels ( Supplementary Fig. 4g) and still displayed contact-induced migration ( Supplementary Fig. 4h), demonstrating again that corralling is not sufficient for competition. Altogether, we conclude that high baseline p53 activity in scrib KD cells is associated with their loser status and is required for them to acquire a low homeostatic cell density and hypersensitivity to compaction, two key features of the mechanical loser status. Interestingly, though it has long been established that scribble À / À cells are eliminated by cell competition in Drosophila 34,35 , when we tested whether this might happen via mechanical insults in wing imaginal discs we found that, unlike in MDCK cells, scribble À / À wing disc cells did not upregulate E-cadherin ( Supplementary Fig. 5a,b) and their outcompetition was not rescued by p53 inhibition (Supplementary Fig. 5c). It is possible that scribble À / À cells are not eliminated by mechanical cell competition in Drosophila or that the function of Scribble or p53 may not be conserved in this process. Alternatively and perhaps more likely, mechanical cell competition may be redundant with other mechanisms of cell competition that have been described to target scribble À / À cells in that system 36,37 .
ROCK activates p38 leading to p53 elevation and cell death. We next wondered how mechanical stress might lead to p53 activation. A potential candidate was p38 signalling, as it is required for scrib KD cell competition 23 , is known to promote p53 activity [38][39][40] and has also been involved in the response to mechanical stress 41 . Consistent with an involvement of p38, we found that compression alone causes an increase in active phosphorylated (T180/Y182) p38 (ref. 38; P-p38) in scrib KD cells (Fig. 4a,b) and that chemical inhibition of this pathway partially rescues both the homeostatic density of scrib KD cells (Fig. 4c) and their compaction hypersensitivity (Fig. 4d). Moreover, the upregulation of p53 in competing scrib KD cells was reduced by p38 inhibition (Fig. 4e). We conclude that in scrib KD cells, compression induces p53 via activation of p38.
We next asked how compression of scrib KD cells causes p38 activation. We monitored cytoskeletal changes induced by compression and found that both cortical Actin (by phalloidin staining) and active phosphorylated-myosin (P-Myosin) are upregulated in compacted scrib KD cells during competition (Fig. 4f,g). Since the cytoskeletal regulator ROCK 42 is one of the main kinases responsible for Myosin phosphorylation, this suggested that ROCK might be activated. Indeed P-Myosin upregulation was reduced in the presence of a ROCK inhibitor ( Supplementary Fig. 6a,b) and the ROCK target phospho-MYPT1 (ref. 43; p-MYPT1) was also elevated in compacted scrib KD cells (Fig. 4h), indicating that ROCK is activated in compacted scrib KD cells. This was potentially relevant because ROCK has been shown to phosphorylate p38 (ref. 44). Thus, to ask whether ROCK is upstream of p38 activation, we compressed pure cultures of scrib KD cells in the presence or absence of a ROCK inhibitor and looked at P-p38 levels. ROCK inhibition led to a partial reduction of P-p38 levels, thus placing ROCK upstream of p38 signalling (Fig. 4i,j). In addition, ROCK inhibition was sufficient to partially rescue the homeostatic density of scrib KD cells (Fig. 4k) and compression-induced cell death (Fig. 4l). Importantly, inhibition of ROCK was also sufficient to prevent the out-competition of scrib KD cells, with no appreciable cell death observed even though cells were compacted far beyond standard competition densities (Fig. 4m and Supplementary Movie 12). Altogether, these experiments indicate that mechanical cell competition is caused by compaction-induced ROCK activation, which activates p38, leading to p53 elevation and cell death. Interestingly, ROCK has also been implicated in apical extrusion of dying or crowded MDCK cells 45,46 . However, in the case of Scribble competition, ROCK has a different function, as it is involved in p53 activation and cell death. In addition, while in that context ROCK was activated by S1P2 and Piezo signalling 45,46 , inhibition of these pathways had no effect on Scribble cell competition ( Supplementary Fig. 7a,b), suggesting a distinct upstream activation mechanism.
p53 activation turns wild-type cells into mechanical losers. Finally, we decided to address how scribble knockdown earmarks cells as losers. Our data indicated that a loser cell status is not an obligate outcome of scribble silencing, as we could isolate scrib KD cells that are competition resistant (scrib RES cells; Supplementary  Fig. 3d). Instead, our data showed that p53 is necessary for competition and that mild p53 elevation is required for scrib KD cells to acquire key features of the mechanical loser status (Fig. 3h,i). This prompted us to investigate whether p53 might actually be sufficient to induce the mechanical loser status and cell competition. Nutlin-3, a chemical inhibitor of the E3 ubiquitin ligase MDM2 (Mouse Double Minute 2 (ref. 47)), activates p53 in a dosedependent manner. This allowed us to establish conditions to induce mild p53 activation in wild-type MDCK cells. Strikingly, low-level p53 activation was sufficient to induce cell flattening (Fig. 5a) and to lower the homeostatic density of wild-type cells (Fig. 5b). Remarkably, Nutlin-3 treatment was also sufficient to induce compaction hypersensitivity (Fig. 5c). Thus, mild p53 activation is sufficient to induce in wild-type cells all the features that we observed in scrib KD cells that are hallmarks of their mechanical loser status. Therefore, we next asked whether differential p53 levels are sufficient to induce mechanical cell competition. To test this, we generated a p53 knockout cell line (p53 À / À ), which we then co-cultured with wild-type cells. Not surprisingly, simply mixing wild-type and p53 À / À cells was not sufficient to induce competition (Supplementary Fig. 8a). Strikingly however, mild p53 activation in wild-type cells, by Nutlin-3 addition at sublethal doses, caused the elimination of these cells specifically in the co-culture, but not when they were cultured alone (Fig. 5d, Supplementary Fig. 8b and Supplementary Movie 13). Importantly, p53-induced cell competition was indistinguishable from scribble cell competition, as it resulted in loser cell compaction (Fig. 5e) and only caused elimination when cells were compacted by p53 À / À cells (Fig. 5d). Overall, these results demonstrate that mild elevation of p53 is sufficient to phenocopy Scribble cell competition and induce both hypersensitivity to cell crowding and the mechanical loser status in otherwise wild-type cells.
To demonstrate the existence of p53-mediated mechanical cell competition beyond the MDCK experimental paradigm, we turned to primary cultures of epithelial cells, specifically to mouse tracheal epithelial cells (MTECs), and looked for evidence of this process. Under normal conditions, confluent MTEC monolayers show little proliferation or change in cell density (Fig. 5f,g, compare first and second time points, and Supplementary Movie 14, left, before Nutlin-3 addition). Interestingly however, when we induced mild elevation of p53 (by Nutlin-3 treatment) in these cultures, we found that this results in a 26% average reduction in cell density, accompanied by cell extrusion, suggesting acquired hypersensitivity to crowding (Fig. 5f,g, compare before and after Nutlin-3 addition, and Supplementary Movie 14, left, after Nutlin-3 addition). To test whether differential levels of p53 could induce cell competition in MTECs, we mixed Tomatolabelled wild-type MTECs (MTEC Tomato; from Rosa26R-nTomato-nGFP mice 48   ARTICLE from p53 null mice). Before Nutlin-3 addition, the proportion of wild-type (Tomato) MTECs in the monolayer did not change for several days when co-cultured with either p53 À / À cells or more wild-type cells ( Fig. 5f-i, first two time points, and Supplementary Movie 14, before Nutlin-3 addition). However and strikingly, Nutlin-3 addition induced robust cell competition specifically in wild-type/p53 À / À MTEC co-cultures, causing the number of wild-type cells to plummet within 6 days to B17% of their starting number, with pronounced cell death and fragmentation (Fig. 5h,i and Supplementary Movie 14, right). Thus, mild p53 elevation is sufficient to induce crowding hypersensitivity and mechanical cell competition in MTEC cultures.

Discussion
In summary, our work demonstrates that in addition to molecular signals, cells use mechanical means to compete (see model in Fig. 5j,k), a concept that had previously only been speculated about and proposed on theoretical grounds 19,49 . We show that hypersensitivity to crowding provides an additional route to the loser cell status, leading to mechanical cell competition and the elimination of loser cells. One interesting prediction of our findings is that they suggest that competition could take place between genetically identical cells if they are endowed with differential sensitivity to mechanical stress or if they reside in tissues with varying mechanical properties 50,51 , a hypothesis that should be investigated. In addition, given the involvement of p53, a general sensor of cell stress, we suggest that mechanical cell competition may be widespread among the damaged cells. Moreover, future work should also examine the implications of our findings on the behaviour of cancer cells. Since p53 is one of the most commonly mutated genes in cancer 52 , our findings suggest that its loss could enable neoplastic cells (scribble is a tumour-suppressor gene 53,54 ) to evade mechanical cell competition. Identifying the physiological contexts where mechanical cell competition plays a role may help better understand tissue biology and potentially cancer formation.
Note added in proof: Evidence of mechanical competition in Drosophila was reported just prior to acceptance of this work 55 .
Cell culture and plasmids. The cell lines used in this publication are not listed in the database of misidentified cell lines maintained by ICLAC. MDCK cell lines were authenticated in our laboratory by RNA sequencing, which confirmed their canine origin. All cell lines used in this publication have been tested in our laboratory and were found to be negative for mycoplasma infection (EZ PCR Mycoplasma Test Kit, Geneflow).
Wild-type MDCK, MDCK-pTR E-cadherin shRNA (Ecad KD ) 32 and MDCK-pTR scribble shRNA (scrib KD ) cells 23 were a kind gift from Yasuyuki Fujita. Wildtype and E-cad KD MCDK cells were cultured in DMEM (21885; Invitrogen) supplemented with 10% fetal bovine serum (FBS, Invitrogen) in a humidified incubator at 37°C with 5% CO 2 . The scrib KD cells were cultured as described for wild-type cells with the addition of blasticidin 50 mg ml À 1 (Sigma) and G418 800 mg ml À 1 (Invitrogen) to the culture media. To establish MDCK cell lines that stably express a nuclear green fluorescent protein (GFP) or red fluorescent protein (RFP) marker, we used a modified pGIPZ-turboGFP-Puro (Thermo Scientific) plasmid where we replaced turbo-GFP with either GFP-NLS or RFP-NLS. Selection following infection was carried out with puromycin (0.65 mg ml À 1 , Sigma).
Resistant scrib KD MDCK cells (scrib RES ) were generated by culturing MDCK-pTR scribble shRNA cells in 5 mg ml À 1 tetracycline for 20 days. The resulting final population was expanded in the absence of tetracycline and tested for efficient Scribble knockdown by western blotting.
To knock down E-cadherin expression in MDCK scribble shRNA cells (scrib KD E-cad KD ), the recipient cells were stably transfected with a modified version (pSUPERIOR.hygro þ gfp E-cadherin shRNA) of a plasmid provided by Yasuyuki Fujita (pSUPERIOR.neo þ gfp E-cadherin shRNA 32 ). Selection was carried out with Hygromycin B (75 mg ml À 1 , Invitrogen). To overexpress E-cadherin in wildtype MDCK cells, the E-cadherin-GFP cDNA 59 was introduced into a modified pTRIPZ-turboRFP-Puro plasmid (Thermo Scientific) in which turboRFP was replaced by RFP-NLS (pTRIPZ-RFP-NLS-Puro). The cells infected with the pTRIPZ ECAD-GFP-P2A-RFP-NLS construct express doxycycline-inducible E-cadherin-GFP and RFP-NLS as a single transcript. Selection following infection was carried out with puromycin (0.65 mg ml À 1 , Sigma).
Primary mouse tracheal epithelial cells (MTECs) were obtained from 5-month-old animals from 26R-Tomato (Gt(ROSA)26Sor tm1(CAG-tdTomato*,-EGFP*)Ees ) and p53-null (Trp53 tm1tyj ) strains, both of C57BL/6 background, using a protocol adapted from published methods 60 . Tracheas were dissected from the larynx to the bronchial main branches and collected in ice-cold DMEM:F12 (11330-32; Invitrogen) supplemented with a solution of 100 units ml À 1 penicillin and 100 mg ml À 1 streptomycin (Invitrogen). The muscle, vascular tissue and glands were then removed and the trachea cut into three to four rings. Each fragment was washed in phosphate-buffered saline (PBS) and then incubated in Dispase (BD Biosciences) at 7.5 Caseinolytic Units in PBS (total volume 450 ml per trachea) for 25 min at room temperature (RT). Tracheal fragments were then transferred into ice-cold DMEM:F12 and the sheets of epithelial tissue were peeled off. The epithelial sheets and medium were transferred to an ice-cold 1.5 ml tube, and pelleted twice at 500 g for 3 min with a PBS wash in between. The pellets were resuspended in 0.05% TE (Invitrogen) supplemented with 5 mM EDTA for 30 min at 37°C. Five hundred microlitres of DMEM:F12 supplemented with 5% FBS was added to stop the reaction. The cells were pelleted (500 g, 3 min), resuspended in MTEC/Plus media and plated on collagen-coated (50 mg ml À 1 rat tail collagen I (BD Biosciences)/0.02 M acetic acid) 24-well tissue culture inserts (BD Falcon) in MTEC/Plus media at approximately 5 Â 10 4 cells per insert. The MTEC cells were cultured in MTEC/Plus media consisting of: DMEM:F12 basal media supplemented with a solution of 100 units ml À 1 penicillin and 100 mg ml À 1 streptomycin, 10 mg ml À 1 insulin (Invitrogen), 5.5 mg ml À 1 transferrin (Invitrogen), 6.7 mg ml À 1 selenium (Invitrogen), 0.1 mg ml À 1 cholera toxin (Sigma), 25 ng ml À 1 epidermal growth factor (R&D Systems), 30 mg ml À 1 bovine pituitary extract (Invitrogen), 5% FBS, 15 mM HEPES and 0.01 mM freshly added retinoic acid (Sigma) in a humidified incubator at 37°C with 5% CO 2 .
Conditioned media and transwell assays. For competition-conditioned medium experiments, media were conditioned for 48 h by cultures of GFP-labelled scrib KD and wild-type cells with (competition condition) or without (mock condition) tetracycline. Recipient GFP-labelled scrib KD cells were pre-treated with tetracycline for 24 h and 10,000 cells per well were seeded into 24-well plates with 1 ml of conditioned medium. For transwell assays, GFP-labelled scrib KD cells (10,000 per well) were seeded at the bottom of a 24-well plate. Co-cultures at a ratio of 1:10 (GFP-labelled scrib KD : wild-type or GFP-labelled scrib KD : scrib KD ) were plated onto a transwell insert (3.0 mm pore size-polyester membrane, Corning) at 1,500 cells per insert). All scrib KD and GFP-labelled scrib KD cells were pre-treated for 24 h with tetracycline before plating.
Fences system. Where applicable, immunofluorescence of cell competition was carried out in a 24-well plate using 'fences' (Aix Scientifics, http://www.aix-scientifics.co.uk/en/fences.html). The system allows two different populations of cells to be seeded on either side of a silicone barrier, thus allowing experimental and control conditions to be cultured, treated, imaged and (where applicable) stained within the same well/coverslip.
Cell competition and contact-induced migration assays. Cell competition assays on MDCK cells were carried out in 24-well plate fences. Control cultures were plated in the centre of the fence (1,000 cells per fence) at a ratio of 1:10 GFPlabelled scrib KD : scrib KD . Competition cultures were seeded on the outside of the barrier (8,000 cells per fence) at a ratio of 1:10, scrib KD : wild-type cells. The fences were removed approximately 5 h after plating and the culture medium was replaced with or without addition of tetracycline. Twenty hours later, the culture medium was replaced with phenol red-free DMEM ( þ 10% FBS and 1% L-glutamine, Invitrogen) with or without tetracycline. Where specified, the chemical inhibitors were also added at this point, except for the ROCK, Piezo and S1P2 inhibitors, which were added after an additional 24 h. Live imaging was started 2-4 h after the final media change and was continued for at least 50 h with regular media changes every 2-3 days.
For cell migration assays, the cells were seeded as for competition assays, except that they were plated at a lower density (1,000 cells in the centre of the fence, 2,000-5,000 cells in the outer chamber) and imaged more frequently (every 10 min to every 2 h). Alternatively, where specified, cells for competition or migration assays were plated on gridded tissue culture plates (m-Dish 35 mm Grid-500, Ibidi) at 6,250 cells per plate. This allowed us to find specific clones of cells for immunofluorescence analysis after live imaging had finished. Co-cultures, unless otherwise specified, were plated at a ratio of 1:10 (scrib KD : wild-type).
The MTEC cells were plated in collagen-coated tissue culture inserts and allowed to grow for approximately 2 weeks until they reached homeostatic density before commencing experiments. In control cultures, wild-type (unlabelled) cells were plated with wild-type Rosa26R-Tomato (nuclear red) cells (MTEC Tomato) 48 cells at a 2:1 ratio. In competition cultures, unlabelled p53-null cells (MTEC p53 À / À ) and wild-type Rosa26R-Tomato cells (MTEC Tomato) were plated at a 2:1 ratio. Nutlin-3 was added at 17.5 mM on day 3 of live imaging. The medium was changed every 2 days.
Homeostatic cell density assays. These assays were carried out in 96-well plates. GFP-labelled scrib KD and scrib KD p53 À / À cells (8,000-12,000 cells per plate) were pre-treated with tetracycline for 12 h before plating where applicable, in the presence or absence of inhibitors. Where applicable, Nutlin-3 was added 24 h after plating. The whole well was then imaged and the individual images were stitched together for further analysis to obtain data on total cell number and density. The plates were imaged every 6 h and cell culture medium and drug treatments were changed every 2 days.
Micropattern assays. For assays in micropatterns, GFP-labelled scrib KD cells with or without tetracycline pre-treatment (72 h) were plated onto circular adhesive patterns (+ 800 mm) within a PEG cell-repellent surface 26,61 . The cells were flushed with culture medium 5 h after plating and the medium was replaced with phenol red-free DMEM ( þ 10% FBS and 1% L-glutamine, Invitrogen) with or without tetracycline. Cell culture medium and drug treatments were changed every 2 days.
PDMS-based cell compression assays. The GFP-labelled scrib KD cells were plated onto a stretched flexible silicone substrate (Gel pak PF-60-X4, 150 mm thickness, Teltek), held in a custom-made chamber (GREM; http://www.jove.com/ video/51193/stretching-micropatterned-cells-on-a-pdms-membrane). Before plating, the clamped membranes were coated with 25 mg ml À 1 fibronectin/PBS (Sigma) for 1 h at 37°C. The membranes were stretched precisely by 2 cm, which provided a 57% stretch over the resting length (unless otherwise specified). A PDMS rectangular chamber, with two compartments (6.6 Â 13 mm each) was attached to the membrane with Baysilone paste (GE Bayer). Two densities (low and high) of tetracycline pre-treated or Nutlin-3 pre-treated (48 h) GFP-labelled scrib KD cells were plated, one in each compartment. High-density cells were plated between 75,000 and 120,000 cells per compartment, forming a confluent monolayer; low-density cells were plated at 25,000-35,000 cells per compartment. The cells were allowed to adhere for 24 h and then the membrane was released to induce compression. The cells were fixed in 4% PFA/PBS (3 h after release for p53 staining, 1.5 h for P-p38 staining and 5 h for cleaved Caspase-3 staining) and processed for immunofluorescence. As per design, low-and high-density cells were stained and imaged from the same stretcher avoiding sample-to-sample variability. The p38 inhibitor was added one day before plating; the ROCK inhibitor was added 1 h before releasing the membrane.

E-cadherin-blocking experiments.
To block E-cadherin-dependent junctions, 20 h after the addition of tetracycline, the cells were incubated in 10 mM EDTA/ PBS for 5 min, followed by a 20 min incubation in calcium-free DMEM (Invitrogen). The cells were then cultured in calcium-free DMEM ( þ 10% FBS and 1% L-glutamine, Invitrogen), supplemented with an anti-E-cadherin-blocking antibody (1:200; ref. 57). Live-image analysis was started 2-4 h after the final media change and continued for at least 50 h.
Immunofluorescence. For immunofluorescence, the cells were cultured on glass coverslips or on gridded dishes (m-Dish 35 mm Grid-500, Ibidi). The cells were fixed for 10 min in 4% PFA/PBS, quenched for 10 min in 50 mM NH 4 Cl/PBS and then permeabilized for 10 min with 0.1% Triton X-100/PBS. The cells were blocked in 2% BSA, 2% FBS/PBS for 30 min. Primary and secondary antibodies were diluted in blocking solution diluted 1:1 in PBS. The primary antibodies were incubated for a minimum of 1 h at RT, followed by washes in PBS; secondary antibodies were incubated for a minimum of 30 min at RT followed by washes in PBS. Coverslips were mounted with FluorSave (Millipore). For immunostaining against phosphorylated proteins, fixing solution was supplemented with PhosSTOP (1 tablet per 10 ml, Sigma), all PBS solutions were substituted with TBS, and blocking solution was substituted with 5% BSA/TBS. For surface immunostaining, the cells were washed in ice-cold phenol red-free DMEM (Gibco) and incubated with primary antibody diluted in ice-cold phenol red-free DMEM at 4°C for 45 min. The cells were then washed in ice-cold PBS before fixation at RT in 4% PFA/PBS for 10 min. Secondary antibody staining was then carried out as outlined previously.
Western blotting. The cells were lysed in 1% SDS/PBS and 10-20 mg of protein was separated on a 4-12% gradient gel (Invitrogen) and transferred onto PVDF membrane for standard ECL blots (Anachem) or Immobilon FL PVDF for LI-COR blots (Millipore, 0.45 mm pore size). The membranes were blocked with 5% Marvel/ 0.05% Tween-20/PBS (PBST) for 1 h at RT, incubated in primary antibodies (in 2% Marvel/PBST) overnight at 4°C, washed in PBST and incubated in secondary HRP-conjugated antibodies for ECL blots or Infrared fluorescent dyes (IRDye-800CW and IRDye-680RD) for LI-COR blots, diluted in 2% Marvel/PBST for 1 h at RT. After washes in PBST, the membranes were developed using standard ECL (GE Healthcare) or scanned with a LI-COR Odyssey CLx near-infrared fluorescence imaging system. Quantifications for LI-COR blots were carried out using Image studio lite (http://www.licor.com/bio/products/software/image_studio_lite/ ?gclid=COWG_Yze98MCFYvpwgodfR8Adg), using Actin to normalize the samples for loading. For original full blots, see Supplementary Fig. 9.
RNA-seq and differential expression analysis. RNA-seq libraries were prepared with the TrueSeq RNA sample preparation V2 kit (Illumina) according to the manufacturer's instructions, and sequenced on an Illumina HiSeq 2000 instrument in single-read mode at 36 or 40 base length. The resulting fastq files were filtered for low-quality reads (oQ20) and low-quality bases were trimmed from the ends of the reads (oQ20).
Genome-based RNA-seq mapping was carried out using Canis lupus familiaris 3.1 (NCBI/Dog Sequencing Consortium) as a reference genome. Transcript sequences were assigned to genome using BLAT 62 . The resulting mappings were filtered by a mismatch threshold (2%), as well as requiring 90% of the transcript to match the genome and all exons to match a single chromosome. This resulted in 21,571 transcripts mapping to the genome. This mapping was used as a junction file for Tophat 2 (ref. 63), which was used to map the RNA-seq reads to the genome. To provide gene names, transcript sequences were downloaded from the NCBI RefSeq database in March 2013 (24,538 sequences). Orthologues were found against the Mus musculus proteome (downloaded in January 2013-NCBI RefSeq) using Inparanoid 64 . The sequences were further annotated using InterProScan 65 to provide both InterPro Domains and Panther ontology terms 66 . For differential expression, read counts were generated by quantifying overlaps with transcript locations. These were then used to generate RPKMs. Comparisons were made between pairs of conditions, each with at least four replicates. For a transcript to be included, counts per million had to be above 10 for all samples in at least one condition and within 2-fold between replicates. Differentially expressed transcripts were then called using EdgeR 67 . Hits were selected applying the following thresholds: P40.05 log FC (fold change) 40.5. Gene Ontology terms overrepresented among these lists were found using David Bioinformatics Resources 68 , in particular, KEGG pathway analysis.
Imaging and image analysis. Quantifications shown in the figures are from a single representative experiment of a minimum of three independent repeats per experiment, unless otherwise specified in the figure legend.
Fixed samples were imaged with a Leica SP5 or SP8 confocal microscope. For live imaging, the cells (kept at 37°C and 5% CO 2 ) were imaged using a Nikon BioStation CT with a Â 10 air objective with imaging frequency between every 10 min and every 6 h (as indicated in the movie time stamps), with media changes every 2-3 days. For each live imaging experiment, at least five fields were imaged by time lapse and analysed. Movie sequences or individual field time points that lost focus during the experiment were discarded from further analysis, as this precluded accurate cell counting.
For Fig. 1a-c and Supplementary Fig. 1d, quantifications of cell number over time were carried out using the open-source image analysis software Cell Profiler (http://www.cellprofiler.org/), using the nuclear GFP signal to segment cells.
To measure cell density in competition assays, the number of nuclei was manually counted using DAPI and/or nuclear GFP and divided by surface area, as calculated in Fiji (http://fiji.sc/Fiji; Fig. 1g and 2n, Fig. 5e, Supplementary Fig. 1e).
For Fig. 1k-n, cell number over time was quantified with the open-source image analysis software Icy 69 , using Icy Protocols. The nuclei were detected using the wavelet-based spot detection plugins. Areas with cysts, where nuclei were not visible due to overexposure or because they were out of focus, and areas without cells were excluded from both cell counting and from area measurements on a frame by frame basis. For Fig. 3h, Fig. 4c,k and Fig. 5b, entire wells of a 96-well plate were imaged by tiling. Individual images were then stitched and processed as above.
Cell height (mm, Fig. 1h-j) was measured from apical to basal membranes using the open image analysis software Fiji.
Cell death in compression assays was quantified as the number of activated caspase-3 positive death events (Fig. 1o-q, Fig. 3i, Fig. 4d,l and Fig. 5c) divided by total number of DAPI-positive nuclei.
Kymographs were generated by first registering the hyperstack containing the different channels (bright-field, RFP and GFP) to remove any global motion or drift over time of the cells (Fig. 2d). The position of the centroid for each nucleus at each time point was extracted and the RFP and GFP channels were then projected on the same mean direction of motion by re-slicing the stacks and then projecting the maximum intensity of the centroid. For the control kymographs (GFP-labelled scrib KD : scrib KD ; RFP-labelled WT: WT), this was done only using the bright-field channel and one fluorescent channel.
The directionality index of migrating cells (D ¼ Euclidean distance/total distance, Supplementary Fig. 2b,d) was calculated after manual tracking of individual cell nuclei (440 cells and 45 movies per condition) with the aid of ImageJ (http://imagej.nih.gov/ij/). Data analysis was performed with Matlab (MathWorks Inc.).
Analysis of cell shape and migratory features (Fig. 2g-k). The aspect ratio, which is defined as the ratio of the long axis to the short axis of cells, was measured from three independent movies by manually fitting the best fit ellipses to single cells. To plot the distribution of angles between the cell's long axis and the cell's direction of motion, we determined for each cell the orientation of the cell's long axis and calculated its angular deviation with respect to the cell's direction of motion (data were binned at 40°interval), by manual tracking. To measure the displacement of wild-type and scrib KD cells, we carried out PIV analysis, using a custom algorithm based on the MatPIV software package for MATLAB, and measured the displacement of individual cells within the first row of contact. Trajectories across different movies were adjusted so that the first time point where both populations formed a broad interface and moved concertedly was set at t ¼ 5 h. Distance is the projected distance along the axis that joins both cell populations.
In Figs 3c-e and 4e, local density was measured for each cell by taking the sum of the Gaussian-weighted distances to all other cells within 50 mm using a Gaussian function with s ¼ 20.
For Fig. 5f-I, quantifications of cell number over time were carried out in ImageJ (http://imagej.nih.gov/ij/), using the nuclear Tomato signal to segment cells.
In Supplementary Fig. 2h-i, E-cadherin staining intensity was measured using Fiji. The individual cells were manually selected on an average z stack projection and their integrated density values were recorded.
Statistical analysis. No statistical methods were used to predetermine sample size. Every experimental condition and treatment was carried out alongside a complete control set of experiments or no treatment control. The sample size was chosen to see a statistical difference between data sets. In the few instances where no difference was observed, sample size was at least as big as in conditions that had shown a difference. The experiments were not randomized and there was no blinding during experiments or analysis, as samples were marked. We carried out a minimum of independent three repeats for each experiment, unless otherwise specified in the figure legend.
The t-Test was used in Supplementary Fig. 1e and Supplementary Fig. 2b,d, where data were normally distributed and with equal variance. The Wicoxon ranksum test was used in Supplementary Fig. 3f. Non-parametric Spearman correlation was used in Fig. 3e. Otherwise, the non-parametric KS test was used for all the statistical tests, removing the requirement for normally distributed data and equal variance. Throughout: *Po0.05, **Po0.005, ***Po0.0005.