Local light-activation of the Src oncoprotein in an epithelial monolayer promotes collective extrusion

Transformed isolated cells are usually extruded from normal epithelia and subsequently eliminated. However, multicellular tumors outcompete healthy cells, highlighting the importance of collective effects. Here, we investigate this situation in vitro by controlling in space and time the activity of the Src oncoprotein within a normal Madin–Darby Canine Kidney (MDCK) epithelial cell monolayer. Using an optogenetics approach with cells expressing a synthetic light-sensitive version of Src (optoSrc), we reversibly trigger the oncogenic activity by exposing monolayers to well-defined light patterns. We show that small populations of activated optoSrc cells embedded in the non-transformed monolayer collectively extrude as a tridimensional aggregate and remain alive, while the surrounding normal cells migrate towards the exposed area. This phenomenon requires an interface between normal and transformed cells and is partially reversible. Traction forces show that Src-activated cells either actively extrude or are pushed out by the surrounding cells in a non-autonomous way. To study the onset of carcinogenesis, the authors use an optogenetics approach to selectively activate with light the Src oncoprotein in cells of an epithelium with a good control in time and space. The transformed cells collectively extrude as a tridimensional aggregate from the cell sheet while remaining alive and in parallel cells of the monolayer converge toward the locus of extrusion.

D uring the initial stage of cancer, cell transformation resulting from the activation of oncoproteins and/or inactivation of tumor suppressor proteins, occurs in isolated cells 1 . Interactions between transformed cells and their normal neighbors are known to be key regulators not only at the early stage of carcinogenesis, but also later during the tumor progression 2 . Several in vitro experiments have looked at the competition between normal and transformed cells by studying sparse transformed cells embedded in a monolayer of normal cells [3][4][5][6][7][8] . In these experiments, normal cells are co-cultured with cells expressing cancer-promoting mutations in an inducible manner, such as a temperature shift or an addition of antibiotics. When cells overexpressing an oncogene (H-RasV12 3 or v-Src 4 ) or the tumor suppressor-knockdown Scribble 6 are plated with normal Madin-Darby Canine Kidney (MDCK) cells, the isolated transformed cells are either extruded from the monolayer (for Ras or Src cells) or become apoptotic (for Scribble cells). The fate of transformed cells in healthy tissues has been studied in vivo, where genetic tools enable the creation of mosaic tissues of transformed cells. In Drosophila, Vidal et al. showed that overactivated Src cells were basally extruded and died by apoptosis when they were surrounded by wild-type cells 9 . Similarly, Kajita et al. showed that v-Src expressing cells are apically extruded in zebrafish 4 . Such extrusions are also observed in Drosophila and mammals when cells with different growth rates compete 10 . In this case, cell fitness drives the elimination of the slower-growing cells. These studies must be put in parallel with others highlighting the suppressive influence of the epithelium on premalignant cells to ensure that suboptimal cells do not accumulate during development 11 .
Despite these advances, the above-mentioned experiments did not control in space or time the transformed population pattern. Powerful optogenetics tools have been used to control gene or protein expression at the single cell level 12,13 and even at the scale of a group of cells 14,15 . Here, we take advantage of this strategy to locally tune the oncogenic activationleading to transformation in a mammalian cell culture. We focus on Src, which is a membrane-bound tyrosine kinase whose proto-oncogene activity is tightly regulated in normal cells [16][17][18][19] . Src is central in many cellular pathways that are involved in cellular growth, cell-cell and cell-matrix adhesion, motility, and invasion 20 . Increased levels and activity of Src have been observed in several types of human cancers, and Src has been proposed to induce tumorigenesis through cancer-related pathways [21][22][23] by promoting both migration and invasion 24 . We have chosen to study competition between normal and Src-transformed cells based on (i) Src potential oncogenic activity, (ii) its invasive properties and (iii) its essential role in single cell extrusion 4 .
In the present paper, we question whether collective effects would emerge after a group of contacting cells embedded in a normal monolayer has been transformed. We develop an innovative set-up using optogenetics to control, in time and space, the pattern of oncogenic Src cells, in order to study the competition between a well-defined population of Src cells and their normal neighbors in a monolayer. We show that this local transformation results in the collective extrusion of the group of Src-activated cells from the monolayer. The extruded cells remain alive and appear as a tridimensional aggregate displaying some, but not all, hallmarks of the epithelial-to-mesenchymal transition (EMT). This tissue deformation requires an interface between normal and Src cells and is partially reversible when the light activation is stopped.

Results
The MDCK optoSrc cells. The light-sensitive cryptochrome 2 (CRY2) has been extensively used for optogenetics applications through blue light-dependent binding to its partner, the truncated cryptochrome-interacting basic-helix-loop-helix (CIBN) 25,26 . We used a stable MDCK cell line engineered to express a synthetic light-sensitive version of the oncoprotein Src ("optoSrc") based on the CRY2-CIBN light-gated heterodimerizer system. The optoSrc is a cytosolic potentially active Src mutant that is deprived of membrane anchoring 27 , this anchoring is required to support the transforming activity of this oncogene. In darkness, the optoSrc is thus inactive in the cytosol. Blue light induces CRY2-CIBN dimerization resulting in the relocalization at the membrane of optoSrc (Fig. 1a, Supplementary Fig. 1)−where it can phosphorylate its physiological substrates and express its oncogenic activity. When blue light is switched off, the CRY2-CIBN dimer splits up: the optoSrc diffuses back to the cytoplasm with a characteristic time of a few minutes 25,26,28 . To limit the photo-toxicity, MDCK optoSrc cells were exposed to short pulses of blue light which results in efficient activation of the optoSrc. In the following, "blue light" always refers to such pulses of light.
Collective behaviors of activated MDCK optoSrc cells. We first checked that MDCK optoSrc cells, when not exposed to blue light, developed normally into an epithelial monolayer, and had the same phenotype as normal MDCK cells. Blue light did not significantly change the proliferation rate of MDCK optoSrc cells (Supplementary Table 1). This is consistent with the observation that elevated Src is not associated with enhanced proliferation for other cell lines 29,30 . We also studied the collective migration of MDCK optoSrc cells with a barrier assay (Supplementary Movie 1). In the absence of blue light, these cells migrated collectively as MDCK wild-type with migration fingers preceded by leader cells as previously reported 31,32 . In contrast, if illuminated with blue light, the MDCK optoSrc monolayer rapidly, within minutes, displayed the characteristics of a migrating monolayer of MDCK cells expressing the activated form of Src 33 . The front edge of the monolayer did not display leader cells and migrated 60% faster than a normal MDCK monolayer. Cells at the front edge exhibited membrane ruffles, consistent with the role of Src in adhesion remodeling: Src phosphorylation typically induces actin ruffle formation while disrupting focal adhesions 34 . We also observed the detachment of individual cells at the front edge of the monolayer. All of these characteristics disappeared when blue-light illumination was stopped.
Effect of blue light on the single-cell extrusion rate. Single-cell extrusion rate in a homogeneous monolayer of identical cells, whether MDCK normal or optoSrc, was not significantly impacted by a global exposition to blue light: the amount of blue light used for photo-activation did not foster single cell extrusions from homogeneous monolayers ( Supplementary Fig. 2). In contrast, when we illuminated with blue light a monolayer of normal MDCK cells containing isolated optoSrc cells, the activated optoSrc cells were expelled 15 times more frequently than without illumination ( Supplementary Fig. 2). These results are consistent with those previously reported for isolated transformed cells in a normal MDCK monolayer 3,4 . They also show that single-cell extrusion is caused by the heterogeneity of the cell monolayer.
Local activation of Src results in a collective extrusion. To study the competition between a group of Src cells and their surrounding normal cells, we projected a well-defined pattern of blue light on a homogeneous confluent monolayer of MDCK optoSrc cells (Fig. 1b). We used a Digital Mirror Device (DMD) interfaced with an inverted microscope to project the blue light pattern, which directly defines the cell transformation pattern. Our standard conditions of illumination correspond to a circular pattern of diameter 70 μm (Fig. 1c), with a 200-ms pulse of light at 470 nm, every 5 min (Methods). Upon such exposure, cells from the illuminated area extruded simultaneously from the monolayer (Fig. 1d, Supplementary Movie 2). Importantly, these extruded cells remained cohesive and formed an aggregate that grew in size, fed by newly extruded cells as time went on. This phenomenon of "collective extrusion" was observed in 95% of the experiments for which the initial cell density was comprised between 1700 and 3000 cells per mm 2 , and started to occur after 21±10 h (standard deviation (SD), n = 112) of illumination in our standard conditions (Fig. 1e).
Contrary to single extruded cells, which readily detach from the monolayer, the 3D cluster made of extruded cells remained well attached to the monolayer, despite the hydrodynamic flows generated by several washing steps. Confocal microscopy confirmed that the extruded aggregate was budding on top of the monolayer at the location of the blue-light stimulation, or in its close neighborhood (Fig. 1f). Hoechst labeling of the nuclei did not reveal condensed or fragmented chromatin (Fig. 1g), indicating that collectively extruded cells are not apoptotic, contrary to individually extruded cells. The height of the aggregate was 40±10 μm (standard error of the mean (SEM), n = 12) after 60 h of blue-light stimulation, corresponding to 210±30 extruded cells (SEM, n = 6). Given that there were initially about ten cells in the exposed area, and that the density of non-exposed cells was only multiplied by a factor 3 within the 60 h of the experiment, we conclude that the growth of the aggregate is mostly due to the arrival of new cells and not to proliferation. Tracking of the nuclei during the collective extrusion confirmed that cells extrude simultaneously (Supplementary Movie 3).
Such a collective extrusion was not observed in presence of Src inhibitors (Fig. 1e, Supplementary Fig. 3) or for a homogeneous monolayer of normal MDCK cells illuminated in the same conditions. Moreover, when cell-cell junctions were disrupted with egtazic acid (EGTA), the aggregate could not form. Since the membrane recruitment of Src is essential for its transforming activity 35 , we also experimented with an optoSrc system deleted of its CIBN component: the recruitment of optoSrc to the membrane was thus not allowed (Supplementary Fig. 1). Upon light-activation, we did not observe collective extrusion for this construct (Fig. 1e), indicating that Src must be over-activated at the membrane to trigger such a collective extrusion. This observation is consistent with the role of Src in the regulation of cell-cell adhesion: in tissue culture, activation of Src is correlated with internalization of cadherin complexes 36 . In that sense, the collective extrusion could result from a localized decrease of cell-cell adhesions for the group of activated cells embedded in the monolayer.
Control, in time and space, of the collective extrusion. Our setup allows a good control of the transformed cells in space and time. The DMD allows us to modify the geometry of blue-light pattern projected, while maintaining a constant local light intensity projected on the monolayer. In parallel, we can control the amount of optoSrc at the membrane with the frequency of the blue-light pulses: the less frequent the light pulses, the less time the kinase remains at the membrane and can phosphorylate its substrates 25,28 .
The area of the blue-light pattern affected the underneath area of the aggregate but not its appearance time (Fig. 2a, b): we find that the underlying area is about 20% larger than the region of illumination. In contrast, the appearance time T ap of the budding structure, but not its size, was function of the frequency of the pulses: T ap is delayed as Δt increases (Fig. 2c). This reveals that the level of Src at the cells membrane directly affects the kinetics of the collective extrusion: the higher the level of Src activation, the faster the formation of the budding structure. The appearance time as a function of the period Δt of blue light was empirically fitted with a saturation curve T ap ¼ T max 1 À ke À Δt τ . We found T max = 47 ± 3 h, k = 0.9 ± 0.1, and τ = 9 ± 3 min. This characteristic time τ is larger than the previously reported CIBN/CRY2 dissociation time of~3 min 25,28 . We conclude that τ is a time constant of the integrated behavior of the cells, a cumulative effect in time of the over-activation of Src, and possibly, its downstream effectors. The plateau of T ap (Δt) at long times logically coincides with a decrease of the probability to observe collective extrusion events (Fig. 2d). The value of T max around 50 h suggests that if collective extrusion has not occurred during the first 50 h of bluelight stimulation, it is unlikely to happen at all. Hence, the frequency of the blue light is a key parameter controlling the kinetics of the appearance of the aggregate.
Since we control in time and space the collective extrusion process by tuning the area of illumination and the frequency of stimulation, our optogenetics-based system is thus a powerful in vitro tool to trigger, on-demand, the formation of a 3D aggregate of transformed cells in a normal epithelial monolayer.
Molecular markers of the collective extrusion. Tissue crowding has been shown to favor cell delamination in epithelial tissues [37][38][39][40] . To investigate the role of crowding in the collective extrusion phenomenon, we measured the probability P bud of observing the formation of a 3D aggregate during the first 60 h of stimulation, as a function of the initial cell density (Supplementary Fig. 4). For initial densities d i corresponding to confluent MDCK optoSrc cells monolayer (1700 cells per mm² ≤ d i ≤ 3000 cells per mm²), P bud was high (95%, n = 112). If the monolayer of MDCK optoSrc cells was initially below confluence (700 cells per mm² ≤ d i ≤ 1700 cells per mm²), P bud was reduced by half, suggesting that a minimal amount of surrounding cells is needed. Surprisingly, when density was too high at the beginning of blue-light stimulation (d i ≥ 3000 cells per mm²), the probability of budding was also reduced (P bud ∼0.7).
Since P bud does not correlate directly with cell density, we looked for molecular changes between MDCK optoSrc cells that were stimulated or not. F-actin strongly accumulates in exposed cells, as revealed by live-imaging of SiR-actin during collective extrusion (Fig. 3a, b and Supplementary Fig. 5). This actin recruitment suggests that a contractile mechanism is at play in the collective extrusion process. However, the distribution of myosin was diffuse in the budding structure ( Supplementary Fig. 5), ruling out the presence of a supracellular actomyosin cable that would drive the extrusion. Logically, treatment with calyculin A (resp. Y-27632) which increases (resp. decreases) cell contractility accelerated (resp. delayed) the average appearance time (Fig. 3c). In parallel, we performed immunostaining of monolayers of MDCK optoSrc cells either fully exposed or not exposed. Upon blue-light illumination, E-cadherin loses its localization at the membrane and accumulates in the cytoplasm, in agreement with previous observations showing that elevated expression of active Src disrupts E-cadherin localization 41 . Besides, vimentin was enriched in the stimulated cells ( Supplementary Fig. 6). Similar trends were observed in collectively extruded aggregates (Fig. 3d, e). Down-regulation of E-cadherin and expression of vimentin have already been observed in other cell lines for which c-Src is elevated 42 . Furthermore, during single Src-transformed cell extrusion, vimentin has been shown to accumulate at the interface between normal and transformed cells 4 . A relocalization of E-cadherin from the plasma membrane to the cytoplasm and an increase of vimentin expression are both hallmarks of the EMT. However, this transition from an epithelial phenotype to a mesenchymal phenotype appears to be only partial, not least because the level of E-cadherin at junctions remains high enough to ensure a good adhesion between extruded cells.
Role of the surrounding non-activated cells. While cells in the exposed pattern collectively extruded, the surrounding cells converged towards the region of stimulation (Fig. 4a, Supplementary Movie 4). We measured that a 70 μm-diameter region of illumination induced a convergent flow of non-exposed cells in a 150 μm-wide disk around the collective extrusion (Fig. 4b, c). The presence of this convergent cell flow during illumination suggests that the local activation triggers a long-range response mediated by cell-cell interactions and these migrating cells feed the extruded cluster. Turning off the illumination once the 3D structure was formed reversed the flow of cells (from the extrusion site outwards) and the cells of the extruded cluster reintegrated the monolayer (Fig. 4a, Supplementary Movie 4). This observation not only demonstrated that the process of collective extrusion was partially reversible, but also confirmed that extruded cells were still alive.
To test if collective extrusion results from interaction between the two cell types, we shed blue light on optoSrc cells confined on a circular adhesive pattern 43,44 that matched the area of light stimulation. When the whole confined MDCK optoSrc monolayer was fully illuminated with blue light, collective extrusion events were observed in only 10% of the cases (Fig. 4d, n = 10 samples from 2 different experiments) instead of 95% for non-confined monolayers, for an initial cell density comprised between 1700 and 3000 cells per mm 2 in both cases. This confirms our hypothesis that collective extrusion requires the presence of wild-type cells surrounding the populations of Src cells.
Traction forces point away from the activated Src cluster. Regarding the mechanical contribution involved in collective extrusion, we consider three possible scenarios compatible with the observed convergent cell flow (Fig. 5a-c). In the first scenario, the activated optoSrc cells collectively contract and "pull" the surrounding monolayer inwards (Fig. 5a). The second scenario is the opposite situation where the surrounding normal cells actively "push" on the group of activated optoSrc cells (Fig. 5b). In the last scenario, the group of activated optoSrc cells is actively escaping from the monolayer (Fig. 5c), which prompts the surrounding monolayer to migrate inwards to cover the surface released by this process 45 . Some of these scenarios can be distinguished by measuring the traction forces exerted by the cells on their substrate during the collective extrusion 46 . These forces were measured by Traction Force Microscopy (TFM) 47,48 . For these experiments, the MDCK optoSrc monolayer was grown on a fibronectin-coated polyacrylamide gel of Young modulus 10 kPa loaded with fluorescent beads. Note that the TFM technique requires a soft substrate, while we generally carried out experiments on glass. Although collective extrusion is slowed down on a fibronectin-coated polyacrylamide gel of~10 kPa, the final outcome is globally unchanged. From the analysis of the beads' displacements, we concluded that the traction forces were highest at the interface between the two populations, and directed outwards (Fig. 5d-f). Two hypotheses are compatible with such a force map. In the first one, the collective extrusion would be a non-autonomous process in which the surrounding cells actively migrate towards the exposed area and effectively push the Src cells out of the monolayer (Fig. 5b). The second scenario would correspond to an active extrusion of the Src cells that would release new area for the surrounding normal cells to migrate on (Fig. 5c). We note that the flow and traction force patterns measured during collective extrusion are similar to those measured during single cell extrusion 49 , but at a supra-cellular scale.

Discussion
Optogenetics enabled us to activate the oncoprotein Src in mammalian tissue culture with a good control in space and time. We created mosaic tissues by over-activating Src in well-defined subpopulations in an epithelial monolayer, giving rise to two interacting cell populations. We find that groups of activated optoSrc cells collectively extrude from the monolayer giving rise to 3D aggregates. These spheroid-like structures remained strongly attached to the monolayer (resisting hydrodynamic flows) and were not observed in presence of EGTA, which disrupts cell-cell junctions. Altogether, these observations point to a new phenomenon of "collective extrusion". Collective extrusion requires a surrounding monolayer and the recruitment of Src at the membrane.
We note that the cells in the aggregate differ in the number of pulses they received. The cells that have been more exposed to blue light are more likely to be the first to extrude. They are located at the top of the aggregate and exhibit a high expression of vimentin. The cells that entered last into the region of illumination do not immediately extrude, and they still have sufficient level of E-cadherin at the membrane to ensure strong attachment of the aggregate to the monolayer.
Previous work has evidenced the extrusion of individual Srcexpressing cells dispersed in a normal monolayer 4 : these single-cell extrusions share common features with the collective extrusion we report here. First, in both single and collective extrusion, cells are extruded apically (towards the lumen). Second, it occurs in an apoptosis-independent manner. Third, it happens only when transformed cells are surrounded by normal cells. However, there are two major differences between collective and single extrusion processes. First, the collective extrusion is reversible. Second, contrary to single-cell extrusion, where cells are weakly attached to the monolayer (and often dissociate from it), the collectively extruded cells remain alive and form a cohesive spheroid strongly attached to the underlying substrate. We hypothesize that the collective extrusion protects the extruded cells as a group and prevents cell death. c Effect of drugs modifying contractility on the probability of budding P bud and the appearance time T ap . Calyculin A (1 nM) decreases T ap (p = 0.021, Wilcoxon rank sum test) and increases P bud , while Y27632 (10 μM) increases T ap (p = 0.013, Wilcoxon rank sum test) and decreases P bud . Horizontal black lines correspond to the mean values. The error bar of P bud is the uncertainty as described in the Methods (n Cal = 28, n DMSO = 30, n Y27632 = 55 and n water = 24, all coming from at least 3 different experiments). d, e A monolayer of Madin-Darby Canine Kidney (MDCK) optoSrc cells was subjected to the standard conditions of illumination for 60 h and fixed with paraformaldehyde. d Immunostaining for E-cadherin (orange) in a monolayer displaying a budding aggregate reveals that E-cadherin is depleted from the membrane for Src-activated cells (black circle) compared to non-illuminated cells (dashed white square). Cell nuclei were stained with Hoechst (purple). e Reconstructed side view of a confocal acquisition after an immunostaining for vimentin (orange) in a monolayer of MDCK optoSrc cells displaying a budding aggregate. Vimentin is strongly expressed in the cells at the top of the budding structure, compared to cells in the monolayer. Cell nuclei were stained with Hoechst (cyan). Scale bars: 50 μm Cell density has been identified as the main parameter controlling single cell extrusion in various studies in epithelia 39,49,50 . In our case, at high cell density, we measured a decrease of P bud in parallel with the apparition of clusters of optoSrc-CRY2 in the cytoplasm (Supplementary Fig. 4). To account for this surprising result, we propose that cell crowding favors the homooligodimerization of CRY2 to the detriment of the CRY2/CIBN dimerization 51 . This would limit the recruitment and over-activation of optoSrc at the membrane. At very low density, blue light stimulation of optoSrc cells did not result in extrusion but in a completely different outcome: an islet of a few optoSrc cells exposed to blue light quickly (within a couple of hours) spreads out and breaks apart like mesenchymal cells ( Supplementary  Fig. 7), which is consistent with the fact that over-activation of Src induces an EMT 23  supports the idea that the extruded cells are likely undergoing an EMT-like process, albeit maintaining a cohesive structure. In this context, it would be of particular interest to confirm this analysis by studying other EMT markers, such as snail or beta-catenin to get a complete characterization of the Src-activated cells.
Collective extrusion requires a frontier between two populations which have different Src levels at the membrane. The tissue deformation therefore results from the interactions between these two populations. Recent studies have demonstrated the importance of mechanical forces in the competition that takes place between normal and transformed epithelial cells 37,53,54 . We showed that the traction forces are pointing outwards at the interface between the two cell populations, while the surrounding normal cells converge towards the aggregate and eventually get incorporated in it. A non-autonomous extrusion process for which surrounding normal cells actively push the Src cells is The non-exposed monolayer is actively pushing on the group of activated optoSrc cells. In such non-autonomous extrusion, we expect the forces generated on the substrate to be directed outwards. c The group of optoSrc cells is actively escaping the monolayer, which closes the resulting gap by migrating inwards. The forces generated are also expected to be directed outwards. compatible with such a force map. It is interesting to note that pathological cyst formation in Drosophila imaginal discs has also been proposed to be a non-autonomous process 55 . However, in that case, the tissue deformation is triggered by the interface contractility between two distinct cell populations. Very recently, Messal et al. analyzed how clusters of Krastransformed cells trigger morphological changes in tissue architecture of pancreatic ducts 56 . Our results are consistent with their observation of endophytic lesions made of transformed cells growing inwards for ducts of large diameter. Although tumor formation is a complex multifactorial process, we show here that a group of contacting Src-activated cells is enough to create a tumor-like aggregate. Since Src is known to be involved in tumor cell dissemination 57 , the ability to create clusters of Src-activated cells on demand using an epithelial cell line is a valuable tool to investigate the role of Src activity in tumor dissemination.
More generally, our results show how tridimensional cellular aggregates can develop by collective extrusion from an epithelial monolayer, based on the simultaneous transformation of a group of neighboring cells, which are extruded but thrive as a collective. Such photo-inducible models may also allow for the study of biological processes that involve partial and reversible EMT, as it has been observed in circulating tumor cell clusters (micro metastases)−in which partial EMT of cancer cells has been associated with increased metastatic properties 58,59 .

Methods
Cell culture. The optoSrc construct refers to the OptoSrc + CIBN-GFP-Caax described in another of our publication 27 . Madin-Darby Canine Kidney (MDCK) and MDCK optoSrc cells were both cultured in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 1% Penicillin-Streptomycin and 10% (vol/ vol) fetal bovine serum (all from Gibco) at 37°C, 5% vol CO 2 and 95% relative humidity. Cells were seeded in 6-well glass bottom plates (IBL, Austria) at roughly 0.5 million cells per well, and left to incubate overnight. For time-lapse acquisition, DMEM was replaced with FluoroBrite DMEM (Gibco) -to reduce medium autofluorescence -similarly supplemented with Penicillin-Streptomycin and fetal bovine serum, but not with GlutaMAX.
Population doubling time and single-cell extrusion rate. Cells were either not exposed or exposed to blue light using our standard conditions of illumination in complete DMEM. For the population doubling time T, cells were manually counted at six different time-points between t = 0 (seeding time) and t = 60 h over 5 fields of view for each conditions (1 movie). T was estimated by fitting the number of cells as a function of time t with the equation n 0 2 t=T where n 0 was the initial number of cells (Supplementary Table 1). For homogeneous monolayers (made of one cell type), extrusions were counted in a 150 × 150 μm 2 region for six samples (2 movies) over 40 ± 3 h after the confluence. For mixed monolayers, extrusion rate was defined as the proportion of isolated cells that have been extruded in the span of 60 h (Supplementary Fig. 2).
Microscopy. Time-lapse acquisitions were performed on an inverted microscope (Zeiss Axiovert 200 M) equipped with a temperature, CO 2 and humidity regulation (Life Imaging Services, Switzerland), a motorized stage for multipositioning (Marzhauser, Germany) and a CCD camera (CoolSnap HQ2 from Photometrics, USA or Retiga R6 from QImaging, Canada). A 40×/0,75NA/Ph2 objective (Olympus, Japan) was used, and images were taken every 15 min for 60 h. Phase and fluorescence light sources were, respectively, a pE-100 and a pE-300-white (CoolLed, UK). Image acquisition was computer controlled with Metamorph (Molecular Devices, USA). For confocal observation, the monolayers were observed with a LSM880 NLO with a 40×/1.3 OIL/DIC II PL APO VIS-IR or a 63×/1.4 OIL/ DIC II PL APO objectives (Zeiss, Germany). For live imaging of F-actin, we used SiR-Actin (Spirochrome) at 100 nM supplemented with verapamil (10 μM). For live imaging of nuclei, we used SiR-DNA (Spirochrome) at 2 μM.
Blue-light activation. Blue-light patterned illumination was made using a Mosaic2 (Andor, UK) coupled to a X-Cite XLED (Lumen Dynamics, BDX LED module 450-495 nm) using the GFP filters (excitation filter BP 450-490 nm and dichroic mirror FT 510) and controlled with the "Mosaic Targeted Illumination" Metamorph plugin. Unless otherwise mentioned, the illuminated region was a disk of diameter 70 μm and the intermittent illumination consisted of a 200-ms pulse of blue light every Δt = 5 min. For these standard conditions of illumination, the light power was measured to be 20 μW after the objective, with a PM30 optical powermeter (Thorlabs, US). Given that the characteristic dissociation time of the complex CRY2/CIBN is~3 min 25,28 , setting Δt = 5 min ensures a moderate level of CRY2 at the membrane. The photo-activation parameters (region size, frequency and pulse duration) were controlled using custom Metamorph journals. Three pulse durations were tested (50,200 and 400 ms): since the duration of the pulse did not obviously change the apparition time of the collective extrusion (Supplementary Fig. 8), we choose to set it to 200 ms, the lowest value that enables a correct GFP imaging.
Drugs. PP2 (Sigma), Src inhibitor #5 (Biaffin GmbH) and calyculin A (Life technologies) were dissolved in dimethyl sulfoxide (DMSO) and added to the medium to reach a final concentration of 10 μM, 10 nM, and 1 nM, respectively. The control wells were supplemented with equivalent volumes of DMSO. Y-27632 (Sigma) and egtazic acid (EGTA, Euromedex) were diluted in distilled water and used at a final concentration of 10 μM and 2 mM, respectively. The drugs were added when medium was changed, between 1 and 4 h preceding the beginning of the photoactivation, and left for the duration of the experiment.
Preparation of the adhesive patterns. The detailed protocol for adhesive patterns can be found in others of our publications 43,44 . Briefly, plasma-cleaned glass slides were first PEGylated with a robust surface treatment. Domains were defined by photolithography directly on the cell-repellent coating in such a way that it remained protected by the photoresist everywhere except at the desired location of the adhesive domains. Using the photoresist as a mask, the polyethylene glycol (PEG) coating was etched on the photoresist-free areas with a plasma cleaner, revealing the underlying glass. The resist was then dissolved away leading to clean glass domains surrounded by PEG-coated glass.
Image analysis. The appearance time T ap of the bud was evaluated by eye on the phase contrast images, with an uncertainty of ± 2 h, as the moment in which the first budding cell starts to emerge from the monolayer, in conjunction with an apparent convergence of the monolayer towards the activated region. Single extruded cells that did not subsequently belong to the 3D structure were not considered. The initial density was determined by manually counting the number of cells in the field of view for the first frame of the acquisition. Image processing was performed with custom ImageJ macros 60 and Matlab (MathWorks, US) scripts. The number of cells in the bud were counted on 3D confocal images, by automatic segmentation of cell nuclei using the surface tool in Imaris (Bitplane, UK). Tracking of nuclei labelled with SiR-DNA was done manually with the cell counter Image J plugin.
Immunostaining. Cells were fixed for 10 min in 4% paraformaldehyde (Euromedex) and permeabilized for 4 min using 0.1% Triton X-100 in phosphatebuffered saline (PBS) with 0.1% bovine serum albumin (BSA, all reactants from Sigma). The cells were then blocked in a solution of 2.5% normal goat serum (NGS, Santa Cruz) in PBS-BSA for 1 h, incubated with the selected primary antibody (references below) for 1 h and washed with PBS. They were then incubated with the secondary antibody (references below) for 30 min at room temperature, washed with PBS, Hoechst-labeled (NucBlue Live ready Probes, ThermoFisher) and mounted on glass slides using ProLong Gold (Life technologies). Primary antibodies: E-cadherin (BD Biosciences, 610181, mouse at a concentration of 1:100); vimentin (Dako M0725 clone V9, mouse at a concentration of 1:300); p-MLC (Cell signaling 3674, rabbit at a concentration of 1:100). Secondary antibodies: goat antimouse A633 (Molecular Probes); goat anti-rabbit (Sigma, SAB4600141) both at a concentration of 1:1000.
Particle image velocimetry (PIV). PIV was used to analyze the velocity fields in the cell monolayers 61,62 using the MatPIV toolbox for Matlab (Mathworks, US). The window size was set to 64 pixels (~23 μm), with an overlap of 0.25. Sliding average over 10 h was performed.
Traction force microscopy (TFM). The protocol for TFM was adapted from Tse et al. 63 . Briefly, coverslips were plasma-cleaned for 10 min, treated with 3aminopropyltrimethoxysilane (2% vol/vol in isopropanol, Sigma) and glutaraldehyde (0.5% vol/vol in ddH 2 O, Sigma), then dried with compressed air. These coverslips are referred to as "activated". In parallel, microscope glass slides were incubated with fibronectin bovine protein (Gibco) in phosphate-buffered saline (PBS) at 25 μg per mL for 30 min at room temperature, then rinsed with PBS and left to air dry. We then made a gel of Young's modulus 10 kPa by mixing 40% acrylamide solution, 2% bis-acrylamide solution (Bio-Rad) and 1% (vol/vol) of fluorescent beads (FluoSpheres 0.2 μm dark red fluorescent (660/680), Life technologies). Ammonium persulfate (1% vol/vol, Bio-Rad) and TEMED (1‰ vol/vol, Bio-Rad) were added to this mix to start the polymerization of the acrylamide gel. Then 30 μL of the solution was applied on the fibronectin-coated glass slide immediately after drying, and an "activated" coverslip was placed on top. During the polymerization, the gel attaches covalently to the activated surface on one side, and attaches to the fibronectin proteins on the other side. This step, inspired by the deep-UV patterning technique 64 , allows to coat the surface of the gel with fibronectin. When the polymerization was complete, the sandwiched gel was immersed in PBS, and the fibronectin-coated glass slide was carefully detached from the gel. The gel was then thoroughly rinsed with PBS and incubated in culture medium for 45 min at 37°C, before the cells were seeded on it, and left to adhere overnight. The coverslip bearing the gel was then placed under the microscope in a POCmini-2 cell cultivation system (Pecon GmbH). At the end of the experiment, the cells were carefully detached from the gel using trypsin (TrypLE, Gibco), to take the reference image of the beads at rest. To compute the traction forces, we used the TFM Fiji plugin available online 65 .
Probability of observing a collective extrusion. The probability of observing a collective extrusion was defined as P bud ¼ k n , where k is the number of occurrences of the collective extrusion observed over n experiments. We consider the binomial distribution with parameters n and p ¼ P bud 100 . The confidence interval for p was estimated using ffiffiffiffiffiffiffiffiffiffi ffi p 1Àp ð Þ n q .
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.