Inactivating negative regulators of cortical branched actin enhances persistence of single cell migration

Cell migration is a critical physiological process during embryo morphogenesis, especially during gastrulation, and in some cell types of the adult, such as immune cells patrolling the organism. Most adult cells, however, do not migrate, but can nonetheless be induced to migrate in pathological conditions, for example, during cancer progression. Untransformed cells classically use the mesenchymal mode of cell migration, which relies on the formation of adherent membrane protrusions called lamellipodia. Lamellipodia are powered by polymerization of branched actin by the Arp2/3 complex (Ridley, 2011; Wu et al., 2012). Several families of nucleation-promoting factors (NPFs) activate the Arp2/3 complex at different subcellular locations. At the plasma membrane of mammalian cells, the most important families of NPFs are the WAVE family, comprising three paralogous proteins, WAVE1, WAVE2 and WAVE3, and the N-WASP family, comprising WASP and N-WASP (Molinie and Gautreau, 2018). For cell migration, WAVE family proteins play a central role in lamellipodium formation, with a variable contribution of N-WASP family proteins depending on the exact cell system (Bieling and Rottner, 2023).

WAVE family proteins are embedded into a multiprotein complex, which is activated by the small GTPase Rac1 at the leading edge of mammalian cells (Ding et al., 2022; Rottner et al., 2021). The control of branched actin polymerization by Rac1 is central in many cell systems and has been reported to control, for example, cell migration of endodermal cells in the zebrafish embryo during gastrulation (Woo et al., 2012; Giger and David, 2017). Moreover, branched actin is thought to feedback on Rac1 activation and thus to sustain polymerization of branched actin where it was previously polymerized (Castro-Castro et al., 2011; Krause and Gautreau, 2014). Thereby the Rac1–WAVE–Arp2/3 pathway controls persistence of cell migration.

Many proteins inhibit the Rac1–WAVE–Arp2/3 pathway at different levels. Arp2/3 activation can be blocked by several Arp2/3 inhibitory proteins (Chánez-Paredes et al., 2019; Zhao et al., 2020). Arpin occupies one of the two NPF-binding sites of Arp2/3 and thus prevents WAVE from activating it (Fregoso et al., 2022). Therefore, arpin depletion by knockdown (KD) or knockout (KO) promotes migration persistence (Dang et al., 2013; Simanov et al., 2021). Branched actin networks can also be debranched by coronins and glia maturation factor (GMF) family proteins that recognize Arp2/3 at the branched junction (Molinie and Gautreau, 2018). The WAVE regulatory complex is inhibited by the Nance–Horan syndrome family protein NHSL1, which interacts with the WAVE complex and can even replace WAVE in its complex (Law et al., 2021; Wang et al., 2023). Specific GTPase-activating proteins inhibit the small GTPase Rac1 to restrict lamellipodial protrusions and cell migration (Parrini et al., 2011; Yamazaki et al., 2013). The ability of GTP-bound Rac1 to activate the WAVE regulatory complex is restricted by CYFIP-related Rac interactor (CYRI) proteins, which compete with WAVE complexes for Rac1 binding (Fort et al., 2018; Yelland et al., 2021).

The combinatorial complexity in the assembly of WAVE and Arp2/3 complexes provides additional means to regulate membrane protrusions and cell migration. The paralogous subunits ARPC1B and ARPC5L assemble Arp2/3 complexes that are more active than those assembled with ARPC1A and ARPC5 (Abella et al., 2016). We have previously shown that depletion of ARPC1A allows the assembly of more ARPC1B-containing complexes, because common subunits are no longer distributed between the two paralogous proteins, and that ARPC1B-containing Arp2/3 complexes promote cortical branched actin (CBA) and migration persistence (Molinie et al., 2019). Similarly, we have shown that the CYFIP2-containing WAVE complexes are less readily activated by Rac1 than CYFIP1-containing WAVE complexes and that CYFIP2-depleted cells display increased lamellipodial protrusions and migration persistence (Polesskaya et al., 2022 preprint). Because of this balance of paralogous subunits, ARPC1A and CYFIP2 exhibit an apparent inhibitory activity on membrane protrusion and migration persistence, even though these subunits belong to complexes that promote branched actin polymerization. CBA polymerized by the Rac1–WAVE–Arp2/3 pathway does not only control cell migration, but also the decision to enter a new cell cycle. In single cells, migration persistence was found to inversely correlate with the duration of the G1 phase (Molinie et al., 2019).

In epithelial cells, cell–cell junctions also depend on the Rac1–WAVE–Arp2/3 pathway for their assembly and maintenance (Li et al., 2020; Verma et al., 2012). When cells reach a high density, they suppress migration and proliferation (Puliafito et al., 2012; Streichan et al., 2014). Upon wounding of the monolayer, cells resume migration and proliferation to heal the wound in a coordinated manner that is regulated in time and space (Palamidessi et al., 2019; Poujade et al., 2007). Because of this coordination between cells, wound healing is a more complex process than single cell migration. Epithelial cells can also detach from each other through an epithelial-to-mesenchymal transition (EMT), which relies on changes of gene expression (Thiery et al., 2009). There appear to be multiple partial EMT states that determine migration modes (Nieto et al., 2016). Vimentin (VIM) is an intermediate filament protein that is a marker of EMT. It is initially widely expressed in the embryo and becomes restricted to mesenchymal cells (Paulin et al., 2022). Vimentin is involved in cell migration and appears to be critical for wound healing in vivo (Eckes et al., 1998, 2000).

Thus, an important question is how negative regulators of WAVE-dependent polymerization of branched actin maintain the migration of epithelial cells under control. Here, we show that enhancing polymerization of branched actin at the cell cortex by removing as many as three negative regulators promoted single cell migration, but not collective migration of mammary epithelial cells. Cells adapted to these long-term perturbations by altering gene expression, in particular, by downregulating vimentin expression. Vimentin downregulation enhanced migration persistence of single epithelial cells.

We recently characterized the role of the Rac1–WAVE–Arp2/3 pathway in sustaining migration in a direction chosen at random by single MCF10A cells (Molinie et al., 2019; Polesskaya et al., 2022 preprint). The human MCF10A cell line is derived from a fibrosis of the mammary breast (Soule et al., 1990). Cells are diploid and not transformed, as they do not form tumors when grafted into immunocompromised mice (Worsham et al., 2005). These epithelial cells are quite plastic, as single cells detach from epithelial islets in vitro and adhere again to other cells when they meet. Arpin, ARPC1A and CYFIP2 play negative roles towards the polymerization of CBA at different levels of the Rac1–WAVE–Arp2/3 pathway (Fig. 1A) and were inactivated in this study.

To generate KO clones for each of these CBA negative regulators, we created insertions/deletions (indels) in the open reading frame (ORF) by transfecting synthetic gRNAs with the purified Cas9 protein into MCF10A cell line. In this protocol with no antibiotic selection, screening of about a hundred clones by western blotting allowed the isolation of at least two independent KO clones for each of these genes (Molinie et al., 2019; Polesskaya et al., 2022 preprint). We analyzed the genomic DNA of these clones to characterize the indels. Genomic DNA encompassing the Cas9-mediated cut was amplified and the PCR product was sequenced. The PCR sequence could be directly read if the two alleles were the same. This was the case for arpin KO clone #1. If the PCR sequence could not be read, it indicated that the two alleles generated different sequences that overlap. In this case, the PCR product was cloned and individual plasmids were then sequenced to unambiguously determine the sequence of the two alleles. In all clones, we were able to identify the indels that accounted for the two KO alleles due to frameshift (Fig. S1). For each of these CBA negative regulators, the two KO clones had increased migration persistence, although to a different extent (Fig. S2). This systematic increase of migration persistence was not associated with consistent variations of cell speed or mean square displacement (Fig. S2), as we previously reported for MCF10A cells when the Rac1–WAVE–Arp2/3 pathway was perturbed (Molinie et al., 2019; Polesskaya et al., 2022 preprint).

To explore the potential synergistic role of regulatory proteins, we combined their KOs. Our goal was to see whether we could isolate a ‘super-migrator’ cell line, as our protocol of gene inactivation, which does not require antibiotic selection, allows us to perform the whole procedure again in a previously obtained KO cell line. For each CBA negative regulator, we further edited the KO clone that displayed the most increased migration persistence. We managed to obtain the three possible double KOs: arpin+ARPC1A KO, arpin+CYFIP2 KO and ARPC1A+CYFIP2 KO (Fig. 1B). To our surprise, none of the double KOs migrated more persistently than single KOs (Fig. 1C,D; Fig. S3). We thus attempted to combine the three KOs and this time, we obtained a super-migrator cell line that migrated more persistently than any single or double KO of ARPIN, CYFIP2 and ARPC1A (Fig. 1C,D; Movie 1).

To characterize the triple KO cells we had isolated, we first examined lamellipodia in fixed cells by immunofluorescence using phalloidin staining and an antibody targeting cortactin, a branched actin marker. MCF10A cells do not form prominent lamellipodia (Molinie et al., 2019). Staining of lamellipodia in triple KO cells was not different, with little enrichment of cortactin (Fig. 2A). We thus decided to examine lamellipodial dynamics. Kymographs extracted from phase-contrast imaging revealed that the rate of membrane protrusion, but not retraction, was significantly enhanced in triple KO cells compared with that in parental cells (Fig. 2B,C). Upon transfection of cells with a plasmid expressing mCherry–actin and imaging by total internal reflection fluorescence structured illumination microscopy (TIRF-SIM), we were able to image actin dynamics in lamellipodia (Fig. 2D; Movie 2). As expected, we found that the actin polymerization rate at the lamellipodium edge, i.e. the sum of retrograde flow with membrane protrusion, was significantly enhanced in triple KO cells compared with that in parental cells (Fig. 2E).

A difficulty associated with this approach is that we could not systematically study two independent clones for combined KOs, even though we had observed significant differences between single KO clones. This rule of keeping two independent clones for each genotype would mean four clones for each of the three double KOs and eight clones for the triple KO. Moreover, if a clone adapts to its genotype, then the route taken to sequentially introduce mutations might potentially also affect the phenotype. In other words, the sequence in which compound KOs were obtained might also influence the phenotype. Testing all possible routes would double the number of double KO clones and multiply the number of triple KO clones by six. To be fully rigorous, the total number of clones to compare with parental cells would amount to (4×2)+(4×2)+(8×6)=64 clones. This large number of clones prompted us to compare the super-migrator cell line we had obtained to cells with transient depletion of the three CBA negative regulators.

We transiently transfected MCF10A cells with pools of siRNAs targeting each of the three CBA negative regulators (Fig. 3A). Upon siRNA-mediated depletion, double KD cells had the same increased migration persistence as that of single KD cells (Fig. 3B; Fig. S4). Only the triple KD had increased persistence compared with that of all single or double KDs. The phenotype of the triple KD was thus most similar to the triple KO clone we had isolated. A difference between the two approaches is that triple KD cells did not exhibit the slight increase of cell speed, previously seen in triple KO cells (Fig. 3C). Together, these experiments showed that MCF10A cells greatly increased migration persistence upon downregulation of three, but not two, CBA negative regulators, whether this downregulation was performed by KO or KD.

To evaluate the physiological relevance of these observations, we turned to zebrafish embryos. We have previously characterized the migration of endodermal cells, which internalize at the beginning of gastrulation and then disperse over the yolk surface as single cells through random walks (Pézeron et al., 2008). This cell-autonomous migration process is governed by the Rac1–Arp2/3 pathway (Giger and David, 2017; Woo et al., 2012). Using morpholinos, we knocked down arpin, ARPC1A and CYFIP2 in endodermal cells and transplanted some cells into wild-type (WT) receiver zebrafish embryos (Fig. 4A; Fig. S5). Transplanted cells were tracked by the expression of a histone 2B–mCherry fusion protein (Fig. 4B; Movie 3).

When CBA negative regulators were tested in isolation, KD of CYFIP2 or arpin significantly increased migration persistence of endodermal cells (Fig. 4C), as we previously reported in the collective migration of prechordal plate cells (Dang et al., 2013; Polesskaya et al., 2022 preprint). Double KDs of arpin and CYFIP2 led to an even stronger increase in persistence. ARPC1A KD had no effect on the migration of endodermal cells. Consistently, KDs of ARPC1A and CYFIP2 together led to a phenotype similar to that of a simple CYFIP2 KD. However, KDs of ARPC1A and arpin increased cell persistence compared with the single KD of arpin, as if ARPC1A potentiated the effect of arpin. The triple depletion of arpin, ARPC1A and CYFIP2 rendered cells significantly more persistent than all single or double depletions (Fig. 4C; Fig. S5). Triple KD cells were also faster than other cells (Fig. 4D). In conclusion, results observed in the zebrafish embryo largely mirrored those obtained on MCF10A cells in culture, suggesting that KDs of the three CBA negative regulators turn cells into super-migrators in vivo as well as in cultured cells.

We then asked whether promoting actin assembly and turning on migration persistence would be associated with defects. We thus decided to challenge our MCF10A KO clones into a morphogenetic assay, in which single mammary cells proliferate and develop acini at the surface of Matrigel (Debnath et al., 2003). We have previously reported that arpin KO cells formed normal acini, albeit bigger than those of parental cells due to increased proliferation (Molinie et al., 2019). This was also the case upon KO of ARPC1A or CYFIP2 (Fig. 5A,B). Acinus sizes reached by double KOs were not different from those of single KOs, but the biggest acini from triple KO reached a size greater than that of acini differentiated from single and double KOs. Parametric statistics could not be applied because the size distribution of acini differentiated by the triple KO became scattered, with coexistence of small and large acini. Most large acini from the triple KO developed a lumen, more frequently so than the acini differentiated from parental cells (Fig. 5C,D). Most small acini of the triple KO did not develop a lumen, in line with delayed morphogenesis. Some of these small acini, however, displayed an irregular shape, together with heterogeneous laminin deposition (Fig. 5C). In conclusion, excessive activation of the Rac1–WAVE–Arp2/3 pathway did not only increase cell proliferation, as previously reported (Molinie et al., 2019), but also increased the heterogeneity of the cell population.

We then sought to analyze the triple KO for collective cell migration in a wound healing assay to examine how cells would coordinate with each other during cell migration, as Rac1-dependent polymerization of branched actin is critical both for lamellipodial protrusions at the front edge and for the maintenance of adherens junctions (Fenteany et al., 2000; Verma et al., 2004, 2012). The triple KO line that was super-migrating at the single cell level did not improve wound healing, and even slightly decreased it, if anything (Fig. 6A,B). The migrating front of the triple KO, however, appeared different from that of parental cells (Movie 4). MCF10A cells close the wound with a front that homogeneously progresses, unlike MDCK cells that organize multicellular ‘fingers’ pulled by leader cells (Poujade et al., 2007). The triple KO MCF10A cells appeared more similar to MDCK cells than did parental MCF10A cells, with protruding fingers at the front. To quantify this effect, we measured the distance covered by the front over time. Overall, the mean value was quite similar between parental and triple KO cells; however, the variance was higher in triple KO than in parental cells owing to the fingers at the leading front (Fig. 6C).

We then decided to examine whether KO of CBA negative regulators would affect gene expression. We first analyzed all the genes of the Rac1–WAVE–Arp2/3 pathway in all KO cell lines that we had isolated for potential compensatory changes in gene expression. We found that, in line with nonsense mRNA-mediated decay due to premature stop codons (Popp and Maquat, 2016), the levels of CYFIP2, ARPIN and ARPC1A mRNAs were greatly decreased when they contained indels in KO cell lines (Fig. S6). Downregulation of CYFIP2 and ARPC1A was not compensated by transcriptional upregulation of the paralogous genes CYFIP1 and ARPC1B. Expression of genes encoding subunits of WAVE and Arp2/3 complexes was overall not significantly altered in KO cell lines, nor was the expression of genes encoding Rac small GTPases (Fig. S6, Table S1).

Because of the differential phenotype of the triple KO between single cell and collective migration, we also measured the expression of nine EMT genes: VIM, CDH1, CDH2, SNAI1, SNAI2, TWIST1, ZO1 (or TJP1), ZEB1 and ZEB2. Among EMT-related genes, we found that the expression of vimentin (VIM) and the transcription regulator ZEB2 was strongly downregulated in all KOs, including in single KOs (Fig. 6D; Fig. S6). ZEB2 controls expression of the vimentin gene in MCF10A cells (Bindels et al., 2006). Downregulation of vimentin gene expression translated into strongly decreased protein levels in all KO clones, but this effect was not observed upon transient KD of CBA inhibitors (Fig. 6E). These results on selected genes showed that CBA can have an impact on gene expression. We thus performed RNA sequencing (RNAseq) on all KO cell lines to study the global impact of CBA inhibitors (Table S2). 787 differentially expressed genes with an adjusted P-value (P_adj) less than 0.01 were identified between super-migrator triple KO and parental cells. Gene Ontology (GO) analysis revealed that most differentially expressed genes belonged to the category of genes involved in ‘cellular component organization or biogenesis’, encoding ‘cytosolic proteins’ involved in ‘protein binding’ (Table S3). The most strongly upregulated gene in triple KO cells was indeed BEX3, which encodes a signaling adaptor protein fulfilling these criteria. The most strongly downregulated gene was GPAT2, which encodes a mitochondrial glycerol 3-phosphate acyltransferase. Importantly, these two genes were respectively upregulated or downregulated in all single KO cells, indicating that their expression was sensitive to the amount of CBA (Fig. S6).

When we examined vimentin expression and organization by immunofluorescence (Fig. 6F), we found that vimentin was organized in a dense network in the majority of parental MCF10A cells. In contrast, single and double KOs had a decreased percentage of cells displaying such a network of vimentin filaments, and the triple KO even more so (Fig. 6G). In the wound healing assay, there was no absolute association between the leader cell phenotype and vimentin expression (Fig. 6H). However, actively migrating leader cells in fingers were more likely to display a vimentin network than follower cells (Fig. 6I).

We wondered whether vimentin downregulation in KO cell lines was promoting or opposing the increased migration persistence of these lines. We first examined the role of vimentin in parental MCF10A cells using two independent siRNAs (Fig. 7A). Vimentin depletion dramatically increased migration persistence of single cells (Fig. 7B; Fig. S7; Movie 5). Overexpression of untagged vimentin, however, did not yield any phenotype on persistence or other migration parameters of single cells (Fig. S7). In collective migration, vimentin-depleted MCF10A cells were less efficient at migrating than control cells, as they took longer to close the wound (Fig. 7C). Similar to triple KO cells, single vimentin-depleted cells displayed an increased rate of membrane protrusion, but not retraction (Fig. 7D). Unlike triple KO cells, however, membrane protrusions of single vimentin-depleted cells covered a wider distance, but lasted shorter than those of parental cells. Therefore, the profound downregulation of vimentin in triple KO cells might contribute to their exceptional migration persistence. To confirm this point, we isolated stable vimentin-expressing cells from the triple KO super-migrating clone (Fig. 7E). As expected, these clones displayed a higher proportion of cells with vimentin networks than that for the triple KO population (Fig. 7F). Most importantly, migration persistence of these cells was efficiently rescued, albeit not completely (Fig. 7G; Fig. S8). These results suggest that the transcriptional downregulation of vimentin in super-migrating cells contributes to their migration persistence (Fig. 8).

In this work, we combined the inactivation of three genes that antagonize migration persistence in MCF10A cells. In both KD and KO, there was no additional effect when two genes were inactivated compared with that of a single inactivation, as if the system was buffered against an overly dramatically enhanced persistence. Surprisingly, the addition of the third inactivation revealed dramatically enhanced persistence in the generated cells, which we referred to as super-migrators, suggesting that the third inactivation crossed a threshold. This enhanced persistence of cell migration was presumably due to the enhanced actin polymerization rate we observed at membrane protrusions of triple KO cells, together with their enhanced protrusion rate.

In zebrafish embryos, when we examined endodermal cells during gastrulation, the phenotype was roughly the same, as the triple inactivation generated cells that migrated more than any single or double KD. In the zebrafish system, however, an additional complexity was that the single ARPC1A KD did not generate a phenotype on its own, but amplified the persistence when combined with other KDs. Even if our super-migrators were very persistent, they were not yet similar to fish keratocytes, which are the most persistent vertebrate cells. To obtain the persistence and the characteristic morphology of fish keratocytes, it is perhaps required to combine our triple KO with inactivation of other negative regulatory genes, such as the genes encoding the Nance–Horan syndrome protein NHSL1 (Law et al., 2021) and the Rac1 competitor CYRI (Fort et al., 2018).

The super-migrator phenotype of MCF10A cells was obtained with both triple KD and triple KO, but the mechanisms were different, because we identified downregulation of vimentin with long-term inactivation in triple KO, but not in triple KD cells. Expression analyses showed that downregulation of the genes encoding GPAT2, ZEB2 and vimentin was observed in all single Kos, indicating that it is a likely cell response to enhanced branched actin at the cortex. ZEB2 is a transcription factor of the zinc finger family, which was shown to control vimentin expression in MCF10A cells (Bindels et al., 2006). MCF10A cells are plastic epithelial cells, where cells in islets coexist with single cells, in regular culture conditions containing EGF. EGF induces vimentin expression in MCF10A cells (Bindels et al., 2006). We found that vimentin downregulation was, to a large extent, responsible for the persistence of triple KO, even though this mechanism was not at play in the super-migrators obtained upon triple KD. In line with our observation, vimentin intermediate filaments were previously reported to antagonize the actin retrograde flow from membrane protrusions (Costigliola et al., 2017) and disassembly of vimentin filaments was required to induce membrane protrusions by Rac1 signaling (Helfand et al., 2011).

The downregulation of the EMT genes ZEB2 and vimentin in super-migrators indicates that cells respond to enhanced CBA by becoming more epithelial-like and less mesenchymal-like. The role of vimentin in fibroblasts is very distinct from the role we uncover here in epithelial cells. Vimentin KO mouse embryonic fibroblasts were found to be less motile in vitro and in vivo upon wound healing (Eckes et al., 1998, 2000). Unlike MCF10A cells, rat fibroblasts were less directionally persistent when vimentin was knocked out (Vakhrusheva et al., 2019). These opposite roles of vimentin in the persistence of fibroblasts and epithelial cells are striking. Along this line, it is interesting to note that the persistent fish keratocytes are also single cells dissociated from an epithelial monolayer (Rapanan et al., 2014), as if the transcriptional program of epithelial cells renders cells more persistent than fibroblasts when these epithelial cells happen to be isolated.

Super-migrator cells appeared less fit in collective behaviors than did parental cells. Wound healing was less smooth with the apparition of leader cells at the front edge. Acinus morphogenesis was also affected. The triple KO cells differentiated into large acinus structures as expected, given the role of CBA in controlling cell cycle progression, but also into small acinus structures, which were abnormal in shape and extracellular matrix deposition. Importantly, despite their increased migration persistence as single cells, there was no sign of invasiveness, in line with the fact that these cells were driven towards a more epithelial program. Heterogeneity of the triple KO population was most evident at the level of vimentin expression and organization. Populations of cancer cells and aging cells are usually more heterogenous than populations of normal cells (Caspersson et al., 1963; Mahmoudi et al., 2019). Our work thus reveals that overactivation of the signaling pathway polymerizing CBA alters gene expression programs in a variable manner in the different cells of the population and renders epithelial cells less able to coordinate in collective behaviors.

The MCF10A cell line was from the collection of breast cell lines organised by Thierry Dubois (Institut Curie, Paris), where they were authenticated. MCF10A cells were cultured in Dulbecco's modified Eagle medium (DMEM)/F12 medium (Gibco) supplemented with 5% horse serum (Sigma-Aldrich), 100 ng/ml cholera toxin (Sigma-Aldrich), 20 ng/ml epidermal growth factor (Sigma-Aldrich), 0.01 mg/ml insulin (Sigma-Aldrich), 500 ng/ml hydrocortisone (Sigma-Aldrich) and 100 U/ml penicillin/streptomycin (Gibco). The MCF10A cell line and its derivatives were routinely tested for mycoplasma and found to be negative.

Human vimentin ORF (GenBank KU178388.1) was amplified from a vector provided by Dr Alexander Minin (Institute of Protein Research, Russian Academy of Sciences). The PCR product was cloned into the home-made MXS CAG Blue SV40pA PGK Puro bGHpA vector between the FseI and AscI restriction sites and sequenced (Eurofins Genomics, Ebersberg, Germany) to ensure that no mutations appeared after amplification. MCF10A cells were transfected using Lipofectamine 3000 (Invitrogen). Two days after transfection, 1 µg/ml puromycin was added. Single clones were isolated by cloning rings, expanded and characterized.

For triple KD, cells were transfected using Lipofectamine RNAiMAX (Invitrogen) with 3 nM ARPC1A esiRNA (EHU105471, Sigma-Aldrich) (Molinie et al., 2019), 20 nM arpin siRNA (5′-GUGGAUGUAUCUCGGCACA-3′, onTarget Plus, Dharmacon) (Dang et al., 2013; Molinie et al., 2019; Simanov et al., 2021) and/or CYFIP2 siRNA (J-021477-05-0002, onTarget Plus, Dharmacon) (Polesskaya et al., 2022 preprint). Specific depletion was controlled with an equivalent amount of non-targeting siRNA (D-001810-01-05, onTarget Plus, Dharmacon) or GFP esiRNA (EHUEGFP, Sigma). siRNA-induced depletion of vimentin was obtained with 20 nM of siRNAs from Sigma-Aldrich (#1, 5′-GUCUUGACCUUGAACGCAAdTdT-3′, and #2, 5′-GGUUGAUACCCACUCAAAAdTdT-3ʹ) (Maier et al., 2015) and controlled with siCtrl (5′-AAUUCUCCGAACGUGUCACGUUU-3′) (Fokin et al., 2021). Cells were harvested or imaged after 2 or 3 days.

KO cell lines were generated using the CRISPR/Cas9 system. The following gRNAs were used: negative control, 5′-AAAUGUGAGAUCAGAGUAAU-3′; ARPIN, 5′- GAGAACUGAUCGAUGUAUCU-3′; ARPC1A, 5′-UAAGAAGAACGGGAGCCAGU-3′; and CYFIP2, 5′-CAUUUGUCACGGGCAUUGCA-3′. Cells were transfected with the CRISPR RNA (crRNA):trans-acting CRISPR RNA (tracrRNA) duplex and the purified Cas9 protein by Lipofectamine CRISPRMAX Cas9 Transfection Reagent (all reagents were from Thermo Fisher Scientific). Cells then were subjected to dilution at 0.8 cells/well in 96-well plates. Clones were screened by western blotting. Targeted loci were amplified with the following primers and sequenced: ARPINfor, 5′-CCTGACAAGGTTCCTCCTGG-3′; ARPINrev, 5′-TGCTGCTCAACACAGCCTTA-3′; ARPC1Afor, 5′-ATTGACAGTTGTACGTGTCTCTG-3′; ARPC1Arev, 5′-AAAGGAAGAGTGCCTGATTTGGA-3′; CYFIP2for, 5′-GTTTCCACAGAGAGCTTGCG-3′; and CYFIP2rev 5′-GGAGCTCAAGAAAGTGAGTAGTG-3′. In the case of overlapping signals, PCR products were cloned (Zero Blunt PCR Cloning Kit, Thermo Fisher Scientific) and independent plasmids were sequenced to identify the two alleles.

Embryos were obtained by natural spawning of Tg(-1.8gsc:GFP)ml1 adult fishes (Doitsidou et al., 2002). All animal studies were performed in accordance with the guidelines issued by the Ministère de l'Education Nationale, de l'Enseignement Supérieur et de la Recherche and were approved by the Direction Départementale des Services Vétérinaires de l'Essonne and the Ethical Committee N°59.

A translation-blocking morpholino (Gene Tools, Philomath, OR, USA) against ARPC1A (5ʹ-ATCTTCAAAGAATTTGCACCTCTGC-3ʹ) was designed for this study, whereas morpholinos targeting CYFIP2 (5ʹ-CGACACAGGTTCACTCACAAAACAG-3ʹ) or arpin (5ʹ-GTTGTCATAAATACGACTCATCTTC-3ʹ) were previously used and validated by rescue experiments in Dang et al., 2013; Polesskaya et al., 2022 preprint. To perform ARPC1A rescue experiments, morpholino-insensitive mRNAs were synthesized from a pCS2+ plasmid containing the zebrafish coding sequence (Twist Bioscience), as the morpholino targets the 5′ untranslated region.

Cells were forced to adopt an endodermal identity through the expression of the activated form of the Nodal receptor acvr1ba (acvr1ba*) (Peyriéras et al., 1998; Giger and David, 2017). Their nuclei were labeled through the expression of mCherry-tagged histone 2B from microinjected mRNA. All mRNAs were synthesized from pCS2+ plasmids with the mMessage mMachine SP6 kit (Thermo Fisher Scientific).

Donor embryos were injected at the eight-cell stage with 0.2 nl of a solution containing acvr1ba* (0.6 ng/µl; synthesized from a pCS2+ plasmid, as stated above), histone 2B–mCherry mRNA (30 ng/μl) and morpholinos against ARPC1A (0.2 mM), arpin (0.2 mM) or CYFIP2 (0.2 mM), alone or in combination. Small groups of GFP expressing endodermal cells were transplanted at the shield stage (6 h post fertilization) to the animal pole of an untreated host (Giger and David, 2017; Boutillon et al., 2022). Embryos were then cultured in embryo medium (Westerfield, 2007) with 10 U/ml penicillin and 10 μg/ml streptomycin. Transplanted embryos were mounted in 0.2% agarose in embryo medium and imaged between the shield stage and 85% epiboly (6–9 h post fertilization) under an inverted TCS SP8 confocal microscope (Leica) equipped with environmental chamber (Life Imaging Services) at 28°C, using a HCX PL Fluotar 10×/0.3 objective (Leica). Visualization of three-dimensional movies and nuclei tracking were done using Imaris (Bitplane). Cell migration parameters were extracted using custom codes in MATLAB (MathWorks) (Boutillon et al., 2022; https://github.com/danio368/Fokin_JCS_2024.git) and autocorrelation was computed using published Excel macros (Gorelik and Gautreau, 2014).

For western blots, the following antibodies were used: anti-CYFIP2 (Sigma-Aldrich, SAB2701081, 1:1000), anti-tubulin clone DM1A (Sigma-Aldrich, T9026, 1:2000), anti-ARPC1A (Sigma-Aldrich, HPA004334, 1:200), anti-Arpin (home-made antibody, Dang et al., 2013; 1:500), anti-p62/DCTN4 clone H-4 (Santa Cruz Biotechnology, sc-55603, 1:200) and anti-vimentin, clone V9 (Santa Cruz Biotechnology, sc-6260, 1:200). For immunofluorescence, anti-laminin V, clone D4B5 (Merck, MAB19562, 1:100), anti-cortactin, clone 4F11 (Merck, 05-180-I-100UL, 1:200) and anti-vimentin, clone V9 (sc-6260, Santa Cruz Biotechnology, 1:100). Secondary goat anti-mouse and anti-rabbit antibodies conjugated with Alexa Fluor 488, 555 and 647 used for immunofluorescence at 1:300 dilution were from Life Technologies (A21236, A21424, A11029, A11034, A21429, A21429). Secondary goat anti-mouse and anti-rabbit antibodies conjugated with alkaline phosphatase used for western blots (1:3000) were from Promega (S3721, S3731).

Cells were lysed in 50 mM HEPES, pH 7.7, 150 mM NaCl, 1 mM CaCl₂, 1% NP40, 0.5% sodium deoxycholate and 0.1% SDS supplemented with a protease inhibitor cocktail (Roche). Lysates were spun at 4°C and 20,000 g and supernatants were mixed with LDS sample buffer (Thermo Fisher Scientific) and 2.5% β-mercaptoethanol. SDS-PAGE was performed using NuPAGE 4–12% Bis-Tris gels (Thermo Fisher Scientific). After transfer, nitrocellulose membranes were blocked in 5% skimmed milk, incubated with primary antibodies, then with secondary antibodies conjugated with alkaline phosphatase and developed with NBT/BCIP as substrates (Promega). Uncropped western blots are displayed in Fig. S9.

For quantitative real-time PCR (qRT-PCR), total RNA was extracted from cell lines using the NucleoSpin RNA Plus Kit (Macherey–Nagel). Specific mRNAs were quantified from the cycle number (Ct value) at which the increase in the fluorescence signal started to be detected by the laser detector of the QuantStudio 7 Flex Real-Time PCR System (Thermo Fisher Scientific) as previously described (Bieche et al., 2001). Gene expression levels were normalized to TBP expression levels (NM_003194) used as an endogenous RNA control and to gene expression in the control condition, the MCF10A parental cell line, to calculate the fold change. Nucleotide sequences of the primers used are detailed in Table S4.

For RNAseq, the library was prepared using 150 ng of total RNA and QuantSeq 3′ mRNA-Seq reverse (REV) Library Prep Kit (Lexogen, Vienna, Austria). The pool was sequenced on a NovaSeq 6000 SP 2×75 bp flow cell (Illumina, San Diego, CA, USA). The raw sequencing data were deposited in the Gene Expression Omnibus (GEO) database under the accession number GSE244924. The BlueBee Genomics Platform (Lexogen, Vienna, Austria) was used to generate normalized counts for each differential expression analysis. To identify genes for which the expression was significantly misregulated, a fold change in expression of >1.5 for each KO cell line compared to the expression level in parental MCF10A cells was considered. To restrict the search to genes that were misregulated with a very high level of significance, the P_adj-value threshold was set to the least P_adj-value we found for a knocked-out gene. Analysis was carried out in R using the BiomaRt (Durinck et al., 2009) and ComplexHeatmap (Gu et al., 2016) packages. GO analysis of differentially expressed genes was carried out using the g:Profiler website (https://biit.cs.ut.ee/gprofiler/gost). Table S3 reports molecular functions (GOMF), biological processes (GOBP) and cellular components (GOCC) of differentially expressed genes, as well as their mapping in pathways derived from Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome (REAC) and WikiPathways (WP) databases.

Videos of individual cell migration in two dimensions were acquired on an inverted Axio Observer microscope (Zeiss) equipped with a Pecon Zeiss incubator XL multi S1 RED LS (heating unit XL S, temperature module, CO₂ module, heating insert PS and CO₂ cover), a definite focus module and an ORCA-Flash4.0 V3 digital CMOS camera. Pictures were taken every 5 or 10 min for 24 h using the Plan-Apochromat 20×/0.80 or Plan-Apochromat 10×/0.40 air objectives. For wound healing, wounds were produced in a cell monolayer either by a pipette tip or by removing inserts from 35 mm μ-dishes (ibidi). Cells were imaged every 20 min during 24 h. For fast acquisition of membrane dynamics, pictures were taken every 2 s for 5 min using the 63×/1.40 oil immersion objective.

For TIRF-SIM imaging, cells transfected with pmCherry-beta-actin (a gift from Klemens Rottner, Zoologisches Institut, Technische Universität Braunschweig, Germany) were plated onto glass-bottomed dishes (ibidi, 81158) that were coated with fibronectin. Images were acquired at 2 s intervals during 2 min using three phase-shifted angles, each with three fringe patterns, on a DeltaVision OMX SR (GE Healthcare) microscope. High-resolution images were reconstructed in softWoRx (AppliedPrecision).

MCF10A cells were seeded on top of polymerized Matrigel (CB-40230C, Corning) in Millicell EZ SLIDE 8-well glass chamber slides (PEZGS0816, Millipore) in a medium containing 4 ng/ml EGF (4 ng/ml; Preprotech, AF-100-15) and 1% serum and supplemented with 2% Matrigel. During the next 3 weeks, the medium was regularly changed. Acini were then fixed in 2% PFA in PBS permeabilized with 0.5% Triton X-100, rinsed with PBS/glycine (130 mM NaCl, 7 mM Na₂HPO₄, 3.5 mM NaH₂PO₄, 100 mM glycine), blocked in IF buffer (130 mM NaCl, 7 mM Na₂HPO₄, 3.5 mM NaH₂PO₄, 0.1% BSA, 0.2% Triton X-100 and 0.05% Tween-20) containing 10% fetal bovine serum (FBS; Thermo Fisher Scientific, 11573397) first and then with IF buffer containing 10% FBS and 20 µg/ml goat anti-rabbit Fc fragment (111-005-046, Jackson ImmunoResearch). Acini were incubated with the primary antibodies in the secondary block solution washed with IF buffer and then incubated with secondary antibody in IF buffer containing 10% FBS. Then acini were incubated with DAPI, rinsed with IF buffer, and mounted with Mount Liquid Antifade (Abberior) and sealed with nail polish. Overview images of acini were taken on an Olympus CKX53 microscope, equipped with a DP22 camera (Olympus) and DP2-SAL firmware (Olympus). Acinus size was then calculated in FIJI by manually contouring acini.

Cells were seeded on glass coverslips coated with 20 µg/ml bovine fibronectin (Sigma-Aldrich) for 1 h at 37°C. Cells were fixed in PBS containing 3.2% PFA, then quenched with 50 mM NH₄Cl, permeabilised with 0.5% Triton X-100, blocked in 2% BSA and incubated with antibodies (1–5 µg/ml for the primary antibody, 5 µg/ml for the secondary antibody) or with 1:3000 diluted SiR-actin (Tebu-Bio). Nuclei were counter stained with DAPI (Life Technologies). Images were acquired using an Axio Observer microscope (Zeiss). Images of acini were obtained on a SP8ST-WS confocal microscope equipped with a HC PL APO 63×/1.40 oil immersion objective, a white light laser and HyD and PMT detectors.

Image analysis was performed in ImageJ or FIJI software. In the single cell migration assay, only cells that had been freely migrating for 7 h or more were taken into account. Cell trajectories were acquired with the ImageJ Manual tracking plugin. Random migration of single cells and migration persistence, based on the angular shift between frames, was analyzed as previously described (Dang et al., 2013) using the DiPer program (Gorelik and Gautreau, 2014). To assess the speed of wound closure, wound areas were manually drawn and measured at different time points. Variance of migration (mean±s.d.) in wound healing was calculated by measuring the distance moved by the leading edge perpendicular to the initial wound edge, for each pixel along the wound. To this end, time-lapse images were first thresholded, segmented in FIJI and analyzed using custom-made MATLAB scripts.

To analyse membrane dynamics, kymographs were obtained along a line that was perpendicular to the cell edge. Protrusion and retraction rates corresponding to protruding or retracting lamellipodia were extracted from the kymographs as tangents of angles. The protrusion length was measured manually. The duration of protrusion was defined as the total duration of observation (5 min) divided by the number of events. To analyse TIRF-SIM data, kymographs were generated in FIJI using manually drawn lines that followed the direction of actin retrograde flow. Protrusion/retraction speed and rearward flow were obtained as a tangent of the angle made with the time axis in kymographs. The actin assembly rate is the sum of protrusion speed and rearward flow.

where A is the autocorrelation, t is the time interval, A_min is the plateau and τ is the time constant of decay. The plateau value A_min was set to zero for the cell lines in vitro, as they did not display overall directional movement. Time constants τ, reflecting directional persistence, were then compared using one-way ANOVA on non-linear mixed-effect models to take into account the resampling of the same statistical unit. A minimum of 20 cells were analysed per condition to achieve sufficient statistical power. No data points were excluded from the analysis.

Other statistical analyses were carried out with GraphPad Prism software (v7.00) and Microsoft Excel 2016. When not stated otherwise, one-way ANOVA and Kruskal–Wallis test were used. The Shapiro–Wilk normality test was applied to examine whether the data fit a normal distribution. If data satisfied the normality criterion, one-way ANOVA was followed by a post hoc Tukey's multiple comparison test. If not, a non-parametric Kruskal–Wallis test was followed by a post hoc Dunn's multiple comparison test. Pairwise comparisons were assessed by either a two-tailed unpaired t-test if data followed a normal distribution or by the Mann–Whitney test otherwise. Four levels of significance were distinguished: *P<0.05; **P<0.01; ***P<0.001; ****P<0.0001.

We thank Dmitry Guschin and Pierre Mahou for excellent technical assistance, and Alexander Minin for providing vimentin cDNA. We also thank the Polytechnique Bioimaging Facility for confocal microscopy partly supported by Région Ile-de-France (interDIM) and Agence Nationale de la Recherche (ANR-11-EQPX-0029 Morphoscope2, ANR-10-INBS-04 France BioImaging).

This article has an associated First Person interview with the first author of the paper.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261332.reviewer-comments.pdf