Matrix mechanics regulates epithelial defence against cancer by tuning dynamic localization of filamin

Matrix stiffness regulates the extrusion of HRas^V12-transformed cells in EDAC

First, to examine whether matrix stiffness could really alter the outcome of EDAC, we used a well-established mammalian model of EDAC^9,13,32 (Fig. 1a, b) and performed EDAC experiments on ECM of varying stiffness that satisfied the range of physiological and pathological elasticity of healthy and fibrotic epithelial matrix^30,33 (Fig. 1c). The EDAC model involved normal epithelial cells competing against and eliminating transformed cells expressing a constitutively active HRas protein (HRas^V12) (Fig. 1a, b). A tetracycline-inducible promoter controlled HRas^V12 expression, which enabled us to initiate the competition process when intended. We first mixed normal or wild-type epithelial cells (MDCK-WT) and cells with tetracycline-inducible GFP-tagged HRas^V12 stably integrated into the genome (MDCK-GFP-HRas^V12) in 40:1 ratio (Supplementary Fig. 1a) and cultured a mosaic monolayer of these populations in the absence of tetracycline. Subsequently, the addition of a stable tetracycline-derivative, doxycycline, in the medium triggered HRas^V12 expression, which became apparent at 30 min post-induction. HRas^V12-transformed cells started rounding up after 3 h, and most of them extruded within eight-to-ten hours (Supplementary Fig. 1b, Supplementary Video 1, Fig. 1b). We performed this experiment on collagen I-coated hydrogel substrates of six individual discrete stiffness values, having an elastic modulus of 1.2, 4, 11, 23, 35, or 90 kPa^30,33 (Fig. 1c). For each stiffness, we counted the fraction of HRas^V12-expressing colonies that extruded at 6 h post-induction (Fig. 1c) and 4 h post-induction (Supplementary Fig. 1c) and observed that this fraction decreased drastically on substrates stiffer than 11 kPa (Fig. 1c). Relevantly, this result does not depend on the ECM protein coating since we also observed a similar inhibitory effect of ECM stiffness on EDAC when we performed the experiments on laminin and basement membrane extract (BME)-coated substrates (Supplementary Fig. 1d). We thus classified 1.2, 4, and 11 kPa substrates as ‘soft’, mimicking healthy epithelium, and 23, 35, and 90 kPa substrates as ‘stiff’ mimicking fibrotic epithelium (Fig. 1c). Interestingly, previous studies had recorded an increase in ECM stiffness from 0.1 to 5 kPa in healthy epithelium to 25–100 kPa in fibrotic epithelium^34,35, which justified our classification and the stiffness transition value (>11 kPa) that we found. We further found that mutant cell extrusion in absence of doxycycline was significantly low on soft substrates (Supplementary Fig. 1e) when compared with doxycycline-induced extrusion, confirming the active extrusion of HRas^V12-expressing cells on soft substrate. We next wondered what happened to the transformed cells that did not extrude on the stiff substrate. These cells remained in the monolayer and eventually showed long basal protrusions and prominent basal actin fibres (Fig. 1d). These features were absent in normal cells. On observing these cells up to 60 h post-induction, we noticed that HRas^V12-transformed cells remained in the monolayer, started dividing, and the colony size expanded (Supplementary Fig. 1f, Supplementary Video 2). To further check if the effect of ECM stiffness on EDAC is not cell-line specific, we performed cell competition experiments between normal and HRas^V12-expressing cells in two other epithelial cell lines, namely Eph4 and Caco-2, by growing them either on 4 kPa (soft) or on 90 kPa (stiff) ECM. In both cell lines, we observed a significantly less fraction of extruded colonies of HRas^V12-expressing cells on 90 kPa ECM than on 4 kPa ECM (Supplementary Fig. 1g), consolidating the generality of our observations with MDCK cells. Collectively, these results demonstrate that ECM stiffness has a decisive effect on the efficacy of EDAC-associated cell competition, where stiff ECM inhibits the elimination of transformed cells.

a Fluorescence images of GFP-HRas^V12 expressing colony extrusion on soft (4 kPa) and stiff (90 kPa) substrates in XY-plane; followed by an illustration depicting the visual metric employed to quantify extrusion. Rounded-up cells expressing GFP-HRas^V12 (as seen on soft ECM) are taken as extruded. Non-extruded GFP-HRas^V12 cells remain in-plane with other cells, as evident on stiff ECM. (Bottom panels) The yellow-dotted lines visually guide the epithelium surface in XZ-plane. GFP-HRas^V12 cells extruded over this surface on soft ECM (top) whereas they remained within this surface on stiff ECM (bottom). White arrowheads indicate GFP-HRas^V12 cells. b Diagram representing different phases of extrusion of transformed cells and stiffness-dependent outcome of EDAC. c Scatter bar plot depicting the fraction of GFP-HRas^V12 expressing colonies extruded over substrates of varying stiffness at 6 hpi. Distinct decrease in extrusion of transformed cells observed with increase in substrate stiffness. The number of colonies counted is indicated inside each bar. Data are mean ± s.e.m. collected over three independent biological replicates. Statistical significance was assessed using Mann–Whitney t-test (two-tailed). p = 2.8490e−08. d Cytoskeletal morphology of non-extruded colonies over stiff ECM at 24 hpi. White arrowheads indicate basal actin fibres associated with HRas^V12– cells on stiff ECM (90 kPa), stained with AlexaFluor647-Phalloidin. Inset: Magnified view of the yellow-boxed region with actin fibres pointed out by white arrowheads. Scale bars = 20 μm (a, XY-view), 10 μm (a, XZ-view; d).

Differential localization of filamin on soft versus stiff matrix determines EDAC efficacy

We next looked for the molecular mechanism by which stiff ECM inhibited EDAC-induced cell extrusion. Extrusion of transformed cells requires remodelling of the actin cytoskeleton in the normal cells that directly interface with the former^19,21,22,23. Since ECM stiffness alters the cellular localization of many force-sensitive cytoskeleton-related proteins^36,37,38, we initially checked whether any actin-binding or actin-crosslinking proteins showed a localization difference on soft versus stiff ECM in normal cells (Supplementary Fig. 2). An actin filament cross-linking protein, FilaminA (FLNA, referred to as filamin hereafter), emerged as the most promising candidate (Fig. 2a), considering its stiffness-sensitive perinuclear localization in normal cells (Fig. 2a, b, Supplementary Fig. 3a). On soft ECM, filamin localized to cytoplasm and cell–cell interface, while on stiff ECM, a significant fraction of filamin molecules localized to perinuclear region (Fig. 2a, b, Supplementary Fig. 3a). Actin counter-staining and subsequent confocal microscopy (Supplementary Fig. 3b) and ultrastructure expansion microscopy (U-ExM)³⁹ (Fig. 2c, Supplementary Fig. 3c) revealed that perinuclear filamin molecules co-localized with perinuclear actin cytoskeleton on the stiff substrate. While previous work had reported filamin accumulation at normal cell-transformed cell interface and depletion of filamin in normal cells abrogated EDAC¹⁹, this stiffness-sensitive perinuclear localization is a unique finding. Relevantly, on stiff ECM, filamin showed depleted interfacial localization and enhanced perinuclear localization than on soft substrate (Fig. 2b). We next asked whether this filamin localization pattern could be relevant to EDAC. In the normal cells interfacing with HRas^V12-expressing cells, filamin showed increased interfacial fraction (Fig. 2b), indicating that during competition, filamin relocates to cell–cell interface¹⁹. We, therefore, hypothesize that perinuclear filamin on stiff ECM perhaps sequesters this interfacial pool, making a large fraction of filamin unavailable for EDAC. To test this hypothesis, we stably over-expressed filamin in normal cells to compensate for the loss of interfacial fraction on stiff ECM (Fig. 2d, left). The filamin over-expressing cells showed clear interfacial and perinuclear filamin pools, at the same time, on stiff ECM (Fig. 2d, left, Supplementary Fig. 3d). We then quantified the fraction of extruded HRas^V12-expressing colonies during the competition between filamin-overexpressing normal cells and HRas^V12-expressing cells (Fig. 2d, right). Filamin overexpression indeed rescued the extrusion of transformed colonies on stiff ECM, rendering EDAC insensitive to substrate stiffness (Fig. 2d). Finally, we validated that the differential localization of filamin on soft versus stiff ECM is not matrix coating-specific by exploring filamin localization in MDCK cells, grown on soft and stiff ECM coated with either laminin or BME (Supplementary Fig. 3e). We also validated that the same is not cell line-specific by reproducing the stiffness-dependent filamin localization in Eph4 and Caco-2 cells (Supplementary Fig. 3f). These results, together, showed that differential localization of filamin on soft versus stiff ECM plays a critical role in EDAC.

a Immunostaining image of mosaic monolayer of MDCK-WT:MDCK-GFP-HRas^V12 cultured on soft (top panels) and stiff (bottom panels) ECM. From left to right: GFP-HRas^V12, filamin in the mosaic monolayer and magnified view of the boxed regions. Cyan arrowheads indicate interfacial filamin enrichment on soft ECM, and yellow arrowheads indicate perinuclear filamin localization on stiff ECM. Scale bars = 10 μm. b Box-and-whiskers plot depicting the fraction of filamin mean fluorescence intensity at perinuclear and interfacial regions, per cell, on soft (4 kPa) and stiff (90 kPa) ECM. Statistical significance was assessed using an unpaired Student t-test with Welch’s correction (two-tailed). p = 6.35542e−09. For boxplots, centre line denotes median, box displays the interquartile range, whiskers indicate range not including outliers (1.5× interquartile range). c Post expansion images of LifeAct-GFP MDCK cells stained for endogenous filamin and cultured on either 4 or 90 kPa PAA gels. Filamin localization differences are indicated by yellow arrowhead (perinuclear) and cyan (interfacial) in the insets. Scale bar = 2 µm; Inset = 1 µm. Fluorescence intensity line scan plots for the red line marked in the filamin channel. d (Left) Fluorescence images of a mosaic monolayer of filamin-over expressing MDCK cells and MDCK-GFP-HRas^V12 co-cultured on a stiff substrate with a magnified view of the boxed region. Yellow and cyan arrowheads indicate perinuclear and interfacial filamin accumulation respectively. (Right) Scatter bar plot depicting the fraction of GFP-HRas^V12 expressing colonies extruded over substrates of varying stiffness. For each stiffness, left bars are for mock MDCK-WT:MDCK-GFP-HRas^V12 and right bars are for MDCK-mApple-FLNA:MDCK-GFP-HRas^V12 mosaic populations. Stable over-expression of filamin in surrounding cells rescued extrusion of transformed populations on the stiff substrate. The number of colonies counted is indicated inside each bar. Data are mean ± s.e.m. collected over three independent biological replicates. Statistical significance was assessed using unpaired Student’s t-test with Welch’s correction (two-tailed). Significance value (when p < 0.0001) p = 9.27612e−09. Source data (b, d) are provided as a Source Data file. e Photoconversion of mEos2-filamin to study filamin localization dynamics. mEos2-filamin cell interfacing with an HRas^V12 cell was stimulated and tracked for 180 s, as depicted in the schematic. Snapshots of mEos2-filamin dynamics indicate dynamic changes in localization. Kymographs of interfacial (i, green box) or perinuclear (ii, white box) regions show enrichment of filamin at the interface (soft ECM, top panel) or perinuclear region (stiff ECM, bottom panel). Scale bars = 5 μm.

We then performed direct experiments to study the dynamics of filamin localization during EDAC on soft versus stiff ECM and to elucidate the effect of perinuclear sequestration of filamin on stiff ECM during EDAC (Fig. 2e, Supplementary Videos 3 and 4). To this end, we expressed moderate levels of filamin tagged with a green-to-red photoconvertible fluorescent protein, mEos2, in normal cells and selected those mEos2-filamin expressing cells that interfaced with at least one HRas^V12-expressing cell. We then photoconverted a population of mEos2-tagged filamin molecules from green to red, at a point nearly halfway between the cell–cell interface and the cell nucleus (Fig. 2e). We subsequently studied where those photo-converted filamin molecules localized during EDAC. On repeated cycles of photoconversion, filamin molecules invariably moved to the cell–cell interface on soft ECM (Fig. 2e, top panels, Supplementary Video 3). On stiff ECM, however, photoconverted filamin predominantly moved to the perinuclear region (Fig. 2e, bottom panels, Supplementary Video 4). Moreover, a separate set of photobleaching experiments in mApple-filamin expressing cells showed that on stiff ECM, filamin had two dynamically different populations—interfacial and perinuclear (Supplementary Fig. 3g, h), in terms of the speed of recovery after photobleaching. On stiff ECM, the perinuclear population was more dynamic and recovered faster than the interfacial population (Supplementary Fig. 3g, h). A comparison of the filamin population dynamics between soft and stiff ECM interestingly revealed that the perinuclear pool on stiff ECM matched very closely to that of the interfacial population on soft ECM (Supplementary Fig. 3h). This indicated that from a dynamics perspective, the perinuclear population of filamin on stiff ECM behaved more similarly to the interfacial population of filamin on soft ECM than to the interfacial population on stiff ECM (Supplementary Fig. 3h). Importantly, photobleaching experiments did not reveal any dynamic differentiation of filamin localization on soft ECM where a prominent perinuclear population was anyway missing. Both fixed-cell and dynamic experiments provided converging evidence proving that the perinuclear cytoskeleton acted like a sink on stiff ECM, by reducing the fraction of filamin molecules available for EDAC at the interface between normal and transformed cells. Therefore, on stiff ECM, the interaction between normal and transformed cells fails to initiate the extrusion of transformed cells.

Cdc42 and perinuclear cytoskeleton determine differential filamin localization

We next asked what molecular signalling pathways decided the differential filamin localization on soft and stiff ECM. To this end, small RhoGTPase Cdc42 is one of the strongest filamin-binding proteins with a very high interaction score of 0.979 in STRING protein interaction database (https://string-db.org/). Given that RhoGTPases, in general, play an important role in mechanotransduction, we speculated whether Cdc42 might have different activation on soft versus stiff ECM. Transfecting the normal cells with a förster resonance energy transfer (FRET)-based Cdc42 activity sensor⁴⁰ indeed revealed stiffness-dependent differences in Cdc42 activity (Fig. 3a). Cdc42 activity at the cell–cell interface was broader and stronger (Fig. 3b) on soft ECM than on stiff ECM. As an alternative representation for Cdc42 activity, staining for a Cdc42-activating guanine nucleotide exchange factor (GEF), Tuba⁴¹, also indicated higher interfacial Cdc42 activation (Fig. 3c). Localization of Tuba to the cell–cell interface was clearly more prominent on soft ECM than on stiff ECM (Fig. 3c). We then asked whether the interfacial localization of filamin on soft ECM might be a consequence of interfacial activation of Cdc42. To this end, we treated the normal cells cultured on soft ECM with a Cdc42-activity inhibitor, ML141, and studied the localization of filamin upon this inhibitor treatment (Fig. 3d). This experiment revealed that upon ML141 treatment, the interfacial localization of filamin vanished while a faint perinuclear filamin ring appeared even on soft ECM (Fig. 3d, e. Supplementary Fig. 4a). Interestingly, we could also induce interfacial localization on stiff ECM by a reverse manipulation, where we expressed a constitutively active Cdc42^Q61L in some cells (Fig. 3f, Supplementary Fig. 4b). As compared to surrounding non-transfected cells, Cdc42^Q61L-expressing cells showed enhanced interfacial localization of filamin, on both soft and stiff ECM. Also, the stiff ECM-specific perinuclear filamin ring disappeared in Cdc42^Q61L-expressing cells (Fig. 3f, left). In contrast, Cdc42^Q61L itself showed prominent interfacial as well as perinuclear localization on stiff ECM (Fig. 3f, middle). Hence, taken together, these experiments proved that Cdc42 activation drives the interfacial localization of filamin, especially on soft ECM. However, given that filamin and Cdc42^Q61L did not co-localize at the perinuclear region (Fig. 3f, right), they indicated that Cdc42 might not be directly responsible for the perinuclear localization of filamin on stiff ECM. While these experiments depicted cell-autonomous interaction between Cdc42 and filamin, cell competition is a non-cell-autonomous event. Considering that filamin accumulated at the interface between normal and transformed cells during competition, we asked whether Tuba might also show similar accumulation. Tuba membrane localization was indeed promoted at the interface between normal and HRas^V12-transformed cells, where filamin accumulated (Supplementary Fig. 4c).

a Raichu-Cdc42 FRET biosensor expressed in MDCK-WT cells cultured on soft (top) and stiff (bottom) ECM. b (Top) Line scan of grey values from the FRET channel (representative white double-arrowed line in (a) from multiple cells cultured on both soft and stiff ECM indicate higher FRET values at cell–cell interface on soft ECM. Bold line plot depicts the mean grey values with the shaded region accounting for mean ± s.e.m of grey values from n = 10 (4 kPa) and 8 (90 kPa) cells. (Bottom) Box-and-whiskers plot of mean FRET index from soft and stiff ECM shows a significant reduction in cells cultured on stiff ECM, indicative of higher Cdc42 activity on soft ECM. Statistical significance was assessed using unpaired Student’s t-test with Welch’s correction (two-tailed). n = 14 (4 kPa) and 18 (90 kPa). c Immunofluorescence images of MDCK-WT cells co-stained for filamin and Tuba (Cdc42-GEF). Yellow arrowhead indicates enrichment of Cdc42-GEF at the interfacial region on soft ECM (top) and at the perinuclear region on stiff ECM (bottom). d Immunofluorescence images of MDCK-WT cells cultured on soft ECM treated with Cdc42 inhibitor: control DMSO (top) or ML141 (bottom) and immunostained for filamin. Green arrowhead indicates enrichment of filamin at the perinuclear region on soft ECM (bottom) post-treatment with ML141. Red arrowheads indicate the direction of line scans, starting from near the nucleus and extending to periphery e. (Top) Line scan of normalized grey values of filamin along red lines from multiple cells in (d) shows perinuclear filamin enrichment peaks with ML141 treatment (orange line). The bold line plot depicts the mean grey values with the shaded region accounting for mean ± s.e.m of grey values from n = 5 cells. (Bottom) Box-and-whiskers plot of the fraction of filamin localization in cells cultured on soft ECM and treated with ML141 (orange) or without treatment (Ctrl, blue). n = 13 (Ctrl) and 18 (ML141 treated) cells. f MDCK-WT cells transfected with constitutively active Cdc42 (Cdc42^Q61L) and immunostained for filamin. Yellow arrowheads indicate increased interfacial enrichment of filamin on stiff ECM. Red arrowheads indicate enriched areas of constitutively active Cdc42. Scale bars = 10 μm. For boxplots, the centre line denotes the median, the box displays the interquartile range, whiskers indicate the range not including outliers (1.5× interquartile range). Source data (b, e) are provided as a Source Data file.

We then asked what recruits filamin to the perinuclear cytoskeleton on stiff ECM. In non-epithelial cells, a refilin family protein, FAM101B or refilinB, stabilizes perinuclear actin networks by associating with filamin⁴² (Fig. 4a, Supplementary Fig. 4d). Thus, FAM101B seemed to be a likely candidate that recruited filamin to perinuclear cytoskeleton on stiff ECM. As preliminary evidence, we indeed found that on soft ECM, FAM101B distributed all over the cytoplasm (Supplementary Fig. 4e). However, on stiff ECM, it co-localized with filamin predominantly in the perinuclear region (Supplementary Fig. 4e). To test whether filamin–FAM101B interaction might be responsible for the perinuclear localization of filamin, we generated mutant filamin, FLNA-[19-22] or dnFLNA, that carried only four filamin repeats. dnFLNA had been known to have a dominant-negative effect on the interaction between endogenous filamin and FAM101B⁴². Stably expressing dnFLNA in normal cells indeed decreased the perinuclear localization of endogenous filamin and increased its interfacial pool (Fig. 4b, f). We moreover generated a dominant-negative FAM101B (dnFAM101B) mutant⁴² that lacked one of the filamin-binding domains, BD2. Stable expression of this mutant also decreased the perinuclear localization of filamin and increased its interfacial pool (Fig. 4c, f). Together, these experiments indicated that filamin–FAM101B interaction plays a crucial role in the localization of filamin to the perinuclear actin cytoskeleton. Interestingly, the perinuclear actin cytoskeleton is connected to the nuclear lamina via the LINC (linker of nucleoskeleton and cytoskeleton) complex. This linkage enables direct transmission of extracellular cues such as matrix mechanics to the nuclear force-sensing machinery^43,44,45 (Fig. 4a). In fact, using a FRET-based molecular tension sensor module⁴⁶, inserted in the middle of a LINC complex protein, Nesprin1, we measured lower FRET efficiency (Supplementary Fig. 4f). This result implied higher LINC complex tension on stiff ECM than on soft ECM (Supplementary Fig. 4f). We, therefore, asked whether cytoskeleton–nucleoskeleton mechanical linkage could additionally be a critical factor in the stiffness-sensitive perinuclear localization of filamin. To this end, we first delinked the perinuclear cytoskeleton from nuclear lamina by disrupting the LINC complex with a dominant-negative Nesprin1 (dnNesprin1 or Nesprin1-KASH) lacking the cytoskeleton binding domain⁴⁴. In another set of experiments, we disrupted the nucleoskeleton with a LaminB1 mutant (dnLaminB1or XLaminB1∆2+)⁴⁷ that disassembled the nuclear lamina. Consequently, normal cells stably expressing dnNesprin1 or dnLaminB1 (Fig. 4d, e) showed decreased perinuclear localization of endogenous filamin (Fig. 4f), indicating that in addition to and perhaps upstream of FAM101B, the perinuclear cytoskeleton–nucleoskeleton mechanical linkage was indeed a critical factor for the perinuclear localization of filamin.

a Schematic representation for filamin structure, depicting its interactions with FAM101B and F-actin. The filamin-FAM101B-actin localization at perinuclear space is enabled by the complex’s interaction with LINC complex, which responds to extracellular force cues. b–e Immunofluorescence images of mApple-dnFLNA-MDCK (b), MDCK-mApple-dnFAM101B (c), mApple-dnNesprin1-MDCK (d) and mApple-dnLaminB1-MDCK (e) stained for filamin. Red arrowheads indicate interfacial filamin localization on stiff ECM. Scale bars = 10 μm. f Box-and-whiskers plot depicting the fraction of filamin mean fluorescence intensity at perinuclear and interfacial regions, per cell, on stiff (90 kPa) ECM; n = 56 (mock), 43 (dnFLNA), 26 (dnFAM101B), 30 (dnNesprin1) and 24 (dnLaminB1) cells over three independent experiments. Statistical significance was assessed using an unpaired Student’s t-test with Welch’s correction (two-tailed). Significance values: p = 2.56898e−16 (mock-dnFLNA), 1.13087e−09 (mock-dnFAM101B), 6.35474e−09 mock-dnNesprin1), 6.62199e−10 (mock-dnLaminB1). For boxplots, the centre line denotes the median, the box displays the interquartile range, whiskers indicate the range not including outliers (1.5× interquartile range). Source data (b–e) are provided as a Source Data file.

Taken altogether, these experiments involving manipulation of Cdc42 activity vis-a-vis inhibition of FAM101B–filamin interaction or disruption of LINC complex provided the molecular basis of the interfacial and perinuclear localization of filamin. They further implied that there could be a delicate competition between filamin–Cdc42 interaction and filamin–perinuclear cytoskeleton interaction (Supplementary Fig. 4g). Filamin–Cdc42 interaction drives filamin molecules towards the cell–cell interface and is stronger on soft ECM than on stiff ECM (Supplementary Fig. 4g). In contrast, filamin–perinuclear cytoskeleton interaction drives them towards the perinuclear region and is stronger on stiff ECM than on soft ECM (Supplementary Fig. 4g).

Rescuing EDAC on stiff ECM

Having identified the molecular signalling pathway that favours filamin localization to perinuclear cytoskeleton on stiff ECM, we finally wondered whether perturbing this localization would restore EDAC on stiff ECM. To this end, we generated modified normal cells stably expressing the mutants that abolished perinuclear localization of filamin and increased interfacial filamin on stiff ECM (Fig. 4b–e, Supplementary Fig. 5a), including dnFLNA, dnFAM101B, dnNesprin1 or dnLaminB1. We created three clones per mutant to eliminate any clone-specific bias (Clone A: Fig. 4b–e; Clones B and C: Supplementary Fig. 5f–i). We then tested whether these cells with increased interfacial filamin could extrude the transformed cells on stiff ECM when the former surrounded the latter (Fig. 5a). Under this experimental condition, we indeed observed significantly increased extrusion of transformed on stiff ECM (Fig. 5b–e, Supplementary Figs. 5j–m). For example, stable dnFLNA expression in normal cells surrounding the HRas^V12-expressing colonies indeed rescued the extrusion of HRas^V12-transformed colonies on stiff ECM (Fig. 5b, Supplementary Fig. 5j). Similarly, normal cells stably expressing dnFAM101B could extrude HRas^V12-expressing colonies with equal efficacy, irrespective of substrate stiffness (Fig. 5c, Supplementary Fig. 5k). Finally, stably expressing either dnNesprin1 or dnLaminB1 in normal cells also rescued the extrusion of transformed cells on stiff substrate (Fig. 5d, e; Supplementary Fig. 5l, m, respectively). Hence, all four mutants that reduced the perinuclear localization of filamin (Fig. 4b–e), either by disrupting filamin–FAM101B interaction (dnFLNA and dnFAM101B) or by disrupting the nuclear mechanotransduction (dnNesprin1 and dnLaminB1), also made cell competition more or less insensitive to ECM stiffness (Fig. 5b–e). Collectively, these experiments established the perinuclear localization of filamin on stiff ECM to be a clear molecular cause behind the failure of EDAC on stiff ECM and suggested possible therapeutic targets in future. Consolidating our experimental results, we discovered and established matrix mechanics as a crucial micro-environmental factor that decided the success or failure of EDAC and also elucidated the underlying molecular mechanism by which matrix mechanics regulated EDAC, thus integrating processes occurring across several length-scales (Fig. 5f).

a An illustration showing the experimental design for rescuing EDAC on stiff ECM and the effect of different mosaic populations on extrusion of transformed cells on stiff ECM. b–e Scatter bar plots depicting the fraction of GFP-HRas^V12 expressing colonies extruded over substrates of varying stiffness. For each stiffness, left bars are for mock MDCK-WT: MDCK-HRas^V12GFP and right bars are for mApple-dnFLNA-MDCK:MDCK-GFP-HRas^V12 (b), MDCK-mApple-dnFAM101B:MDCK-GFP-HRas^V12 (c), mApple-dnNesprin1-MDCK:MDCK-GFP-HRas^V12 (d) or mApple-dnLaminB1-MDCK:MDCK-GFP-HRas^V12 (e) mosaic populations. Stable expression of dnFLNA, dnFAM101B, dnNesprin1 or dnLaminB1 in surrounding cells rescued extrusion of transformed populations on the stiff substrate. The number of colonies counted is indicated inside each bar. Data are mean ± s.e.m. collected over three independent biological replicates. Statistical significance was assessed using Unpaired t-test with Welch’s correction (two-tailed). Significance value (when p < 0.0001) p = 9.27612e−09. Source data (b–e) are provided as a Source Data file. f Summary of the integrative mechanism that connects the processes occurring at different length scales. We elucidate matrix mechanics, a microenvironmental factor (a), influences the molecular dynamics of filamin (b and c) to determine the outcome of EDAC (d).