Vimentin tunes cell migration on collagen by controlling β1 integrin activation and clustering

Epithelial–mesenchymal transition (EMT) is an integral process in embryogenesis, wound healing and discrete pathologies such as organ fibrosis and cancer progression. In EMT, epithelial cells undergo phenotypic and functional modifications that accompany their conversion into mesenchymal cells (Acloque et al., 2009; Kaimori et al., 2007; Lamouille et al., 2014; Nieto, 2009; Shibue and Weinberg, 2017; Thiery et al., 2009; Thiery and Sleeman, 2006). Activation of EMT results in reduction of cytokeratin, increased expression of proteins associated with mesenchymal lineages, such as vimentin, fibronectin and type I collagen (Kaimori et al., 2007; Lamouille et al., 2014; Shibue and Weinberg, 2017), and the conversion to a migratory phenotype in which cells increase their expression of β1 integrins (Liu et al., 2015a). These alterations facilitate the migration of mesenchymal cells through dense collagenous matrices, partly by generating long cell extensions that can intercalate between collagen fibrils (Grinnell et al., 2006). Although vimentin is a marker of fibroblast lineages (Venkov et al., 2007), its role in regulating cell migration through collagen is not well understood. Notably, physical reorganization of fibrillar proteins that accompanies the migration of fibroblasts through extracellular matrices is an important collagen remodeling pathway (Feng et al., 2014) and is crucial for the maintenance of organ health in mammals (Cox and Erler, 2011).

Fibroblast migration through fibronectin and collagen matrices requires the formation of adhesions enriched with β1 integrin (Maemura et al., 1995; Zeltz and Gullberg, 2016), which undergoes activation as a result of allosteric modifications that increase binding affinity (Arjonen et al., 2012). Integrin activation may involve bidirectional processes in which tightly regulated binding of intracellular proteins to integrin cytoplasmic tails and matrix ligand binding to extracellular domains contribute to two-way signaling. One of the regulators of ‘inside-out’ integrin activation is talin 1, which binds to β-integrin cytoplasmic tails and anchors actin filaments (Calderwood et al., 2013; Wang et al., 2019). Talin binding alters the orientation of the β integrin transmembrane domain. By separating the α- and β-integrin cytoplasmic subunits, talin induces an extended open conformation of integrin extracellular domains (Kim et al., 2011; Klapholz and Brown, 2017; Shattil et al., 2010). In addition, the association of integrin cytoplasmic tails with a broad repertoire of cytoskeletal and signaling proteins facilitates initiation of integrin clustering, which strengthens cell attachment to matrix ligands and provides physical continuity between extracellular matrix and cytoskeletal polymers such as actin filaments (Baade et al., 2019; Liu et al., 2015b; Puklin-Faucher and Sheetz, 2009). As integrins are activated by ligand binding, during initial attachment to the matrix, cells may adhere quite tightly, even in the absence of ‘inside-out’ integrin activation (Anthis et al., 2009; Tanentzapf and Brown, 2006). High-affinity, talin-binding integrins can form adhesions, but normal cell spreading is only possible with integrins that can recruit the signaling adapter protein paxillin (Pinon et al., 2014).

Talin colocalizes with paxillin near the plasma membrane (Case and Waterman, 2015; Critchley, 2009; Kanchanawong et al., 2010). After initial integrin–extracellular matrix (ECM) engagement and actin reorganization, paxillin contributes to the formation of cell extensions as a result of binding to the Rho GTPases Cdc42 and Rac1 (Hodge and Ridley, 2016; Iden and Collard, 2008; Jaffe and Hall, 2005). One of the downstream effectors of Cdc42 and Rac is the p21-activated kinase 1 (PAK1), which interacts with these small GTPases to regulate the actin assembly required for the formation of cell extensions (Delorme-Walker et al., 2011; Pirruccello et al., 2006; Zhao and Manser, 2012). PAK1 phosphorylates paxillin, which in turn enhances the turnover rate of cell adhesions and regulates leading-edge protrusion and cell migration (Nayal et al., 2006). PAK1 may also regulate the reorganization of vimentin filaments by phosphorylation, a process that modifies filament assembly and maturation, cell adhesion dynamics and cell extension formation (Eriksson et al., 2004; Goto et al., 2002; Li et al., 2006). Currently it remains obscure how vimentin regulates collagen-dependent migration despite its prominent expression in fibroblasts and localization to cell adhesions (Kim et al., 2010; Kreis et al., 2005; Terriac et al., 2017). We considered that vimentin recruits paxillin into focal adhesions, which enables Cdc42-dependent PAK1 activation and, finally, cell extension formation.

Vimentin intermediate filaments are centrally involved in cell extension formation and migration through fibronectin (Ding et al., 2020). In endothelial cells, migration through collagen matrices may depend on the colocalization of vimentin with the α2β1 integrin, a fibrillar collagen receptor (Kreis et al., 2005). Vimentin localizes to cell–matrix adhesions in a structure-dependent manner: small, nascent adhesions contain small vimentin particles whereas mature, large adhesions contain long filaments (Terriac et al., 2017). The abundance of vimentin and its configuration (e.g. squiggles, protofilaments, filaments) regulate cell adhesion strength and migration (Ding et al., 2020; Helfand et al., 2011). Vimentin interacts with adhesion proteins and may aggregate in adhesion complexes to regulate the focal adhesion formation (Dave et al., 2013; Menko et al., 2014; Terriac et al., 2017; Tsuruta and Jones, 2003). Currently, the mechanisms by which vimentin controls cell adhesion, cell extension formation and migration through soft connective tissues are not defined. Here, we examined whether vimentin acts as an adaptor protein that orchestrates the formation, stabilization and turnover of β1 integrin-dependent cell adhesions to collagen in time and space, which in turn leads to cell extension formation and cell migration.

As vimentin intermediate filaments are involved in the regulation of cell shape (Lowery et al., 2015; Mendez et al., 2010), formation and stabilization of extensions required for cell migration on fibronectin substrates (Ding et al., 2020), we examined cell extension formation in NIH-3T3 cells (3T3) and mouse embryonic fibroblasts (mEFs) plated for 4 h on monomeric collagen (100 μg/ml) or fibronectin (10 μg/ml). For reduction of vimentin expression, cells were treated with scrambled siRNA (control) or ON-TARGET VIM siRNA 36 h before re-plating on ECM. In addition, we examined Vim null mEFs obtained from vimentin-knockout mice (Patteson et al., 2019) to assess the effects of long-term vimentin deletion on cell extension formation. Immunoblotting showed greater than 90% reduction of vimentin expression after siRNA knockdown (KD; Fig. 1A). Confocal images of vimentin and actin filaments showed that vimentin expression levels strongly affected cell morphology when cells were cultured on collagen, whereas when plated on fibronectin there was only a modest effect on cell morphology (Fig. 1B). Quantification of cell surface area and circularity, as well as numbers and lengths of cell extensions (defined as processes that extended >10 μm from the cell centroid) indicated that vimentin affects cell morphology on collagen but not fibronectin. In cells with reduced vimentin expression (siRNA-treated or null cells), cell surface area on collagen surfaces was reduced by 45% in 3T3 cells and by more than 30% for mEFs compared with that of control cells (Fig. 1C). Furthermore, in cells with vimentin KD, there was a greater than 60% reduction in the number of cell extensions for cells plated on collagen (Fig. 1D). Similarly, when cells were cultured on collagen, the mean length of extensions was reduced (by 30% for 3T3 cells and 50% for mEFs) as a result of vimentin KD or deletion (Fig. 1E). Finally, analyses of morphology showed that loss of vimentin increased cell circularity by 60–70%, but only in cells that were cultured on collagen (Fig. 1F). These data prompted us to examine the relationship between vimentin expression and the formation of extensions in cells plated on collagen.

As vimentin associates with the α2β1 integrin and is enriched in β1 integrin-containing focal adhesions (Kreis et al., 2005), we assessed whether vimentin affects the expression of collagen adhesion receptors normally expressed in mesenchymal cells, specifically the β1 integrin and DDR2. Besides β3 and β7 integrins, the β1 is also a common integrin subunit in fibronectin receptors (Liao et al., 2002; Pankov and Yamada, 2002). Evaluation of whole-cell lysates showed that vimentin KD increased the level of β1 integrin but with no effect on DDR2 protein expression (Fig. 2A). We examined by flow cytometry the abundance of total membrane-associated and activated β1 integrin at the plasma membrane in non-permeabilized cells. Short-term vimentin KD by siRNA did not affect β1 integrin expression, whereas repeated passaging (n passages>10) of vimentin null mEFs resulted in a slight increase (∼20%) in β1 integrin compared with levels in the control (Fig. 2B). With a separate approach for estimating β1 integrin activation, we analyzed the binding of fibrillar collagen-coated 2 μm beads after 2 h incubation with cells. Flow cytometry showed that loss of vimentin increased collagen-coated bead binding by 40% for 3T3 cells, 170% for vimentin-null mEFs and 68% for vimentin siRNA mEFs (Fig. 2C). When the same analyses were performed with fibronectin beads, there were no substantial differences in β1 integrin activation and binding between wild-type (WT) and vimentin-null cells (Fig. S1A,B). Moreover, the addition of 0.2 mM MnCl₂ to activate integrins did not affect fibronectin-coated bead binding in WT mEFs, whereas in vimentin-null cells, fibronectin-coated beads were bound with the same efficiency as collagen-coated beads (Fig. S1C). Surprisingly, the addition of fibronectin-coated beads to WT and vimentin-null mEFs cultured on fibronectin-coated surfaces resulted in a 2-fold increase in bead binding by vimentin-null mEFs (Fig. S1D). These results were affected by increasing ligand accessibility (examined in a receptor abundance assay, Fig. 2B), which indicated increased capacity for ligand binding by receptors. We also measured β1 integrin activation with a neo-epitope antibody (clone 9EG7; Lenter et al., 1993). For cells plated on collagen, vimentin KD was associated with a 2-fold increase in β1 integrin activation in 3T3 cells (treated with siRNA), a 9-fold increase in vimentin-null mEFs and a 7-fold increase in mEFs treated with siRNA, compared with activity in their respective siRNA control or WT controls (Fig. 2D). The fluorescence intensity of β1 integrin receptors assessed by microscopy (Fig. 2E) was consistent with flow cytometry data.

We also measured adhesion strength using beads coated with fibrillar collagen (1 mg/ml) and a shear force assay (3.5 Pa) (Chong et al., 2007). Quantification of bead binding after repeated washes showed that loss of vimentin increased collagen binding strength of 3T3 cells by more than 64% (Fig. 2F) and of mEFs treated with vimentin siRNA by more than 33% (Fig. 2G). These differences in binding strength disappeared after large numbers of washes (>8). Analysis of vimentin-null mEFs showed consistently stronger collagen binding, which was not dependent on the number of washes. In contrast, KD of vimentin with siRNA reduced the relative binding strength of cells incubated with fibronectin-coated (10 μg/ml) fluorescent beads (23% reduction for 3T3 cells and 49% reduction for mEFs), whereas vimentin deletion in mEFs reduced binding strength of fibronectin-coated beads by 14% (Fig. S1E,F). Notably, exposing cells to increased fibronectin by adding fibronectin-coated beads to cells cultured on fibronectin-coated dishes, resulted in slightly stronger bead binding by vimentin-null mEFs (Fig. S1G).

We assessed whether collagen bead binding was indeed dependent on the β1 integrin with the use of an inhibitory antibody (Lenter et al., 1993). Quantification of bead binding showed that inhibition of the β1 integrin blocked the increase of binding strength caused by vimentin depletion (Fig. 2H). For assessing the specificity of vimentin-mediated activation of the β1 integrin, we examined fibrinogen-coated bead binding, which is dependent on β2 and β3 integrin receptors (Hantgan et al., 2010) but not the β1 integrin. We found that vimentin depletion decreased initial fibrinogen binding strength (Fig. 2I), indicating that vimentin plays a role in the regulation of β1, but not in β3 integrin activation. Furthermore, negative controls that utilized poly-L-lysine-, fibrinogen- and BSA-coated fluorescent beads also showed no differences in binding after vimentin depletion (Fig. S1H). These data support the notion that vimentin plays a role in regulating integrin function activation or inhibition depending on the access of the β1 integrin to collagen or fibronectin, respectively.

As cell binding to certain matrix ligands (e.g. collagen) is associated with β1 integrin activation and clustering (Welf et al., 2012), we examined the spatial distribution of activated β1 integrins. We found that the level of vimentin expression was associated with alteration of activated β1 integrins in focal adhesions (Fig. 3A). In cells with vimentin KD, the size of β1 integrin clusters was reduced by 60% in 3T3 cells and 50% in mEFs cultured on collagen. Re-expression of vimentin in vimentin-null mEFs (Fig. S2A) restored the size of β1 integrin clusters by 42% (Fig. 3B). Contemporaneously, there was a greater than 55% increase in the number of small (∼0.5 μm) β1 integrin clusters after vimentin depletion for all cell types that were cultured on collagen. After vimentin rescue, there were only 11% more clusters compared with the clusters in control cells (Fig. 3C). Furthermore, measurements of the spatial distribution of β1 integrin clusters with respect to the leading edge of the cell showed that loss of vimentin was associated with a 68% reduction of the size of integrin clusters that formed within 1.75 μm of the leading edge of the cell. Moreover, in mEF vimentin-null cells in which vimentin expression was restored (rescue) showed only 17% smaller clusters compared with those in WT cells. We found no differences for β1 integrin clusters that formed at the cell periphery (i.e. <1.75 μm from the leading edge; Fig. 3D). Under identical experimental conditions, cells cultured on fibronectin-coated surfaces showed no differences in the spatial distribution of β1 integrin clusters or cluster size between WT and vimentin-null mEFs (Fig. S2B–D). Further analysis by confocal imaging of cells plated on fibronectin- or collagen-coated surfaces for 3 h and stained with anti-total (non-activation specific) β1 antibody (clone KMI6; Fig. S2E–G) showed marked similarities to data on integrin activation and clustering obtained with the 9EG7 antibody. Finally, we examined the spatial relationship of β1 integrin cluster size with the number of clusters using Pearson correlation (Fig. 3E). These analyses showed that the size of integrin clusters was negatively correlated with numbers of clusters per cell (r<−0.5), which was 3-fold greater for all vimentin-deficient cells: 3T3 siRNA (P<0.04), mEF null (P<0.02) and mEF siRNA (P<0.03). Collectively these results indicate that vimentin affects β1 integrin cluster growth and maturation of focal adhesions. We conclude that vimentin filaments are involved in linking collagen-dependent β1 integrin activation with cluster formation.

The interactions of cell surface integrins with matrix ligands result in integrin clustering and the formation of nascent cell–ECM adhesion structures, such as focal complexes (Geiger et al., 2001). Focal complexes are dynamic protein aggregates that mature into more stable, larger structures (focal adhesions) that form mechanical links between actin filaments and the ECM (Kanchanawong et al., 2010). As vimentin regulates the lifetime, stability and maturation of focal adhesions (Burgstaller et al., 2010), we considered that vimentin may also affect the assembly and function of adhesion-associated molecules in time and space, which in turn may regulate β1 integrin activation. Accordingly, we looked for a link between vimentin expression and quantitative changes in the abundance of proteins in adhesion complexes by quantitative tandem mass spectroscopy. Initial mass spectrometry analyses of collagen bead-associated proteins were used to examine protein abundance ratios for vimentin-null versus WT mEFs. These data showed that after vimentin depletion, there was ∼2.5-fold more talin 1 in collagen adhesions than in those of WT cells (Fig. S3A). Notably, talin regulates β1 integrin activation (Calderwood et al., 2013). Validation of the mass spectrometry data by immunoblotting showed that vimentin depletion increased talin abundance in collagen-bead associated proteins but did not affect whole-cell talin expression levels (Fig. 4A). The total abundance of β1 integrin, as determined by immunoblotting, was higher in vimentin-null mEFs than WT mEFs, which was consistent with flow cytometry analyses (Fig. 2B). However, the same amount of bound integrins was detected in mEF WT and vimentin-null cells for collagen-associated proteins. Finally, we noticed a loss of paxillin in the collagen bead-associated proteins, suggesting that vimentin may affect the recruitment of paxillin to the focal adhesion complex. This loss of paxillin in the focal adhesions was not explained by a global alteration of expression, because loss of vimentin did not change the total expression of talin, α-actinin, paxillin or PAK1 (Fig. 4A,B; Fig. S3B–E).

As previous data suggest that vimentin filaments of cells involved in healing are linked to paxillin-rich focal adhesions (Menko et al., 2014), we examined whether vimentin may directly interact with paxillin during cell migration. Notably, immunoprecipitation data showed an association between vimentin and paxillin (Fig. 4C). Furthermore, we analyzed whether vimentin depletion leads to alterations of the size or number of paxillin-containing focal adhesions. Confocal images (Fig. 4D) showed that after vimentin depletion, the mean area of focal adhesions was reduced by 44% compared with those of WT cells. In addition, the size of focal adhesions was restored by 22% after vimentin re-expression (Fig. 4E). Moreover, our analyses showed that loss of vimentin increased the number of focal adhesions per cell by more than 2-fold (Fig. 4F), which is an indication of less-mature focal adhesions.

One of the most abundant paxillin-associated proteins in focal adhesions is talin1, which is critical for focal adhesion assembly (Dong et al., 2016). Accordingly, we investigated whether talin is involved in the mechanism by which vimentin regulates cell binding to collagen. We examined binding of collagen-coated 2 μm beads in vimentin-WT and -null mEFs after talin deletion and re-expression (Fig. 4G; Fig. S3F). The data showed that after 2 h of incubation with beads, loss of talin reduced bead binding by 68% in WT and vimentin-null mEFs. Talin re-expression resulted in 55% and almost complete restoration of collagen-coated bead binding for WT and vimentin-null cells, respectively, compared with binding by the corresponding control cells (Fig. 4H). Furthermore, high-resolution, sub-diffraction imaging of WT and vimentin-null mEFs (Fig. 4I) indicated that vimentin filaments colocalize with talin (Pearson coefficient, r=0.55; Fig. 4I, upper inset). Small, nonfilamentous vimentin particles were in close proximity with talin, but no colocalization was detected (Fig. 4I, lower inset). Comparisons between cells plated on fibronectin and collagen showed that the colocalization of vimentin and talin was stronger on collagen than on fibronectin (P< 0.02; Fig. 4J), suggesting that vimentin may participate in talin-dependent β1 integrin activation.

Focal adhesions link and transduce bidirectional forces and signals between the actin cytoskeleton and ECM polymers like collagen. The assembly of focal adhesions requires the activity of Rho GTPases, including RhoA and Cdc42 (Hall, 2005), which promote cell extension formation and cell migration through PAK1 (Hodge and Ridley, 2016; Zhao and Manser, 2012). Accordingly, we investigated whether vimentin depletion interferes with cell extension formation through alterations of Cdc42 activation. Cells were plated on collagen for 3 h, lysed and fractionated, and the resulting supernatants were incubated with PDB–sepharose beads to assess the relative abundance of active Cdc42. Loss of vimentin decreased Cdc42 activation by 65–70% for all types of vimentin-deficient cells (Fig. 5A). We assessed whether vimentin depletion affected other Rho proteins by analyzing Rac1 activity. Depletion of vimentin was associated with a 94% reduction of Rac1 activity; the activity was restored to 57% of the control level after vimentin re-expression (Fig. 5B). We also investigated paxillin-dependent Cdc42 activation by transfecting cells with paxillin siRNA and with a GFP–paxillin expression plasmid that was not affected by the paxillin siRNA we used (Fig. S4A). Paxillin KD in WT cells was associated with 97% reduction of Cdc42 activity; this was restored to 50% of control after paxillin re-expression. Analysis of vimentin-null mEFs showed similar (91%) reduction of Cdc42 activity compared with activity in control cells. Paxillin rescue of these vimentin-null mEFs restored Cdc42 activity by 24% (Fig. 5C). Paxillin silencing resulted in substantial differences in cell morphology (Fig. S4C–E) and the abundance of β1 integrin clusters (Fig. S4H,I) in cells plated on collagen but not on fibronectin. These data indicate that vimentin and paxillin have a substantial impact on Cdc42 activation.

As the active form of Cdc42 is necessary for PAK1 activation through autophosphorylation (Rane and Minden, 2014), we examined how loss of vimentin may affect PAK1 activity by assessing PAK1 phosphorylation levels (Fig. 5D). We found that PAK1 phosphorylation after vimentin depletion was indeed reduced (S141 phosphorylation by 35% and T423 phosphorylation by 24%). As an additional control for PAK1-mediated phosphorylation of vimentin, we used the Cdc42 inhibitor ML141 (10 μM), which showed that inhibition of Cdc42 reduced phosphorylation of PAK S141 in WT mEFs by more than 30% (Fig. 5E; Fig. S4B). No significant changes were found in PAK phosphorylation in vimentin-null mEFs. Finally, we analyzed the effect of Cdc42 inhibition on morphology and focal adhesions in vimentin-WT and -null mEFs with ML141. Confocal images (Fig. S4C) and their quantification showed that inhibition of Cdc42 did not mimic all of the morphological changes (Fig. S4D,E) and focal adhesion effects (Fig. S4F–I) that were observed in vimentin-null cells plated on fibronectin. In contrast, collective measures of cell morphology (Fig. S4D,E) and focal adhesions (Fig. S4F–I) showed that ML141 treatment phenocopied the morphological differences observed in WT and vimentin-null mEFs cultured on collagen.

As Cdc42 plays a central role in cell extension formation, we evaluated cell extension formation by immunostaining for the filopodia tip marker myosin-10 (Fig. 5F). These data showed that vimentin depletion was associated with reduced cell extension formation (>56%; Fig. S4J). Wound healing assays in cultured cells treated with scrambled siRNA (control) or with vimentin siRNA and then plated on collagen, showed visually evident differences in the rate of gap closure after vimentin KD. Time-lapse images of in vitro wound closure showed that loss of vimentin delayed wound healing (Fig. 5G). Furthermore, we found no differences in cell migration of vimentin-WT and -null mEFs plated on fibronectin (Fig. S4K). Estimates of migration velocity (Fig. 5H) showed that compared with controls, vimentin KD decreased migration velocity by 36% for 3T3 cells and by more than 35% for mEFs. Collectively these data indicate that vimentin affects the PAK1 activation pathway through inhibition of paxillin-related Cdc42 activation, which in turn contributes to the regulation of cell extension formation and regulation of focal adhesion size in collagen-dependent manner.

During filament assembly, vimentin undergoes several posttranslational modifications (Shi et al., 2016; Snider and Omary, 2014) including phosphorylation, which affects filament assembly and maturation (Eriksson et al., 2004). PAK1 mediates vimentin phosphorylation at S38, S56 and S72 (Ding et al., 2020; Eriksson et al., 2004; Goto et al., 2002; Li et al., 2006) but it is not known whether vimentin phosphorylation regulates β1 integrin-dependent formation of focal adhesions and cell extensions. We first examined vimentin phosphorylation in 3T3 and mEF cells that had been plated for 3 h on collagen-coated surfaces. To evaluate whether inhibition of PAK1-dependent vimentin phosphorylation is involved in β1 integrin activation and cell extension formation, we used IPA3, a PAK-specific inhibitor (Rudolph et al., 2013). Immunoblot with phospho-vimentin antibodies (Fig. 6A) indicated that IPA3 treatment reduced PAK1-dependent vimentin S39, S56 and S71 phosphorylation by 40%, 60% and 20%, respectively, for 3T3 cells. Similarly, there were 45–67% reductions of vimentin phosphorylation in mEFs (Fig. 6B). Immunostaining for vimentin showed that IPA3 treatment caused spatial reorganization of vimentin filaments, which included vimentin aggregation around the nuclei (Fig. 6C). Quantification of extensions showed that IPA3 treatment and inhibition of vimentin phosphorylation was associated with a greater than 70% reduction of the number of extensions (Fig. 6D). Furthermore, the length of extensions in IPA3-treated cells was reduced by 70%, compared with those of vehicle-treated control cells (Fig. 6E). As IPA3 may affect PAK kinases other than PAK1 (Deacon et al., 2008), we examined cells in which mutant PAK1 was expressed. In these experiments, similar results as those obtained with IPA3 were found for vimentin filament reorganization (Fig. 6F) and cell extension length (Fig. 6G) in cells transfected with plasmids carrying PAK1 K299R or PAK1 T423E mutants (in addition to endogenous PAK1 expression; the controls in Fig. 6F-I show expression of WT PAK1 from a plasmid transfection). Cells treated with the plasmid carrying kinase-dead PAK1 K299R showed a greater than 70% reduction in cell extension length. In contrast, in cells transfected with the constitutively active PAK1 T423E, cell extension length was increased (by 15% for 3T3 cells and 37% for mEFs). These results indicate that changes of cell morphology (in this instance, the formation of cell extensions) were related to inhibition of PAK1-dependent vimentin phosphorylation, which in turn reduced vimentin recruitment to focal adhesions. Further, we investigated whether treatment with IPA3 (Fig. S5A) and the various PAK mutants (Fig. 6H) affected the distribution of paxillin in focal adhesions. These data showed that alteration of vimentin assembly and loss of vimentin filaments was associated with paxillin recruitment into focal adhesions. We also examined the distribution of focal adhesions in the cell periphery and cell body. Transfection with a PAK dominant-negative mutant (K299R) reduced the sizes of focal adhesions for WT mEFs and vimentin-null mEFs by 5-fold and 9-fold, respectively. In cells transfected with the constitutively active PAK mutant (T234E), the size of focal adhesions was increased by 27% for WT mEFs and 5-fold for vimentin-null mEFs, compared with control cells (Fig. 6I). Furthermore, inhibition of PAK by IPA3 resulted in similar reductions of focal adhesion areas (Fig. S5B). These data were consistent with the distributions of the β1 integrin, suggesting that PAK1-dependent vimentin phosphorylation is involved in vimentin filament assembly and maturation. Further, we evaluated the impact of PAK-dependent vimentin phosphorylation on the activation of β1 integrin. In cells immunostained for active β1 integrin, confocal microscopy (Fig. 6J) showed that IPA3 inhibited the formation of β1 integrin clusters (Fig. 6K). These results were also consistent with flow cytometry assays showing that inhibition of vimentin phosphorylation blocked β1 integrin activation (Fig. 6L). Furthermore, we analyzed the effect of the PAK1 mutants on β1 integrin activity in vimentin-WT and -null mEFs. These data (Fig. S5C) indicate that inhibition of PAK1 kinase (K299R) caused a 38% and 15% reduction of β1 integrin activity in vimentin-WT and -null mEFs, respectively.

The ability of cells to spread, form extensions, migrate and remodel fibrillar collagen depends on their interactions with underlying matrix molecules (Discher et al., 2005). In addition to actin filaments and actin-binding proteins, vimentin filaments are thought to be centrally involved in cell migration during wound healing (Helfand et al., 2011; Mendez et al., 2010) and cancer invasion (Schoumacher et al., 2010). However, the mechanism of how vimentin regulates the formation of cell extensions in physiological remodeling of mature connective tissue is not well defined. Our central finding is that vimentin affects cell extension formation in a collagen-dependent manner and plays a key role in regulating β1 integrin activation and clustering through the control of talin and paxillin recruitment into focal adhesions. Notably, paxillin regulates Cdc42 activation (Hodge and Ridley, 2016; Iden and Collard, 2008; Jaffe and Hall, 2005), which in turn enables PAK1 activation, vimentin phosphorylation and induction of vimentin filament formation (Ding et al., 2020; Eriksson et al., 2004; Goto et al., 2002; Li et al., 2006). Taken together, these processes contribute to increased cell extension formation. Our data presented here indicate that vimentin in focal adhesions plays a central role as an adaptor protein, which may contribute to mesenchymal cell migration and extracellular matrix remodeling (Pinto et al., 2015; Terriac et al., 2017). These processes, which are associated with EMT, are important for tissue regeneration and repair in wound healing (Stone et al., 2016). Collectively, our results indicate that vimentin affects cell extension formation and cell adhesion by regulating β1 integrin activity and avidity in collagen-dependent manner (Fig. 7).

Vimentin filaments are involved in the determination of cell shape (Lowery et al., 2015; Mendez et al., 2010) and in the development and stabilization of cell extensions during cell migration on fibronectin (Ding et al., 2020). We found that in fibroblasts expressing vimentin, there was rapid formation and elongation of extensions in cells plated on fibronectin and collagen. While the vimentin-depleted cells showed reductions in the number and length of extensions, only the extension length was reduced in cells cultured on fibronectin. Consistent with these data, knockdown of vimentin also strongly affected cell area and circularity on collagen. Whereas cell area provides a general estimate of cell spreading, a reduction of circularity is consistent with the formation of membrane protrusions such as lamellipodia and filopodia (Uynuk-Ool et al., 2017) that are needed for cell migration through the collagen matrix.

Integrins are critically important adhesion receptors that regulate cell attachment to matrix proteins. For binding to collagen, cells use the β1 integrin (Maemura et al., 1995; Zeltz and Gullberg, 2016). The dynamic interaction between integrins and collagen is crucial for the stabilization of membrane protrusions and for the tension application to the ECM and retraction of the cell rear during cell migration; integrin recycling is central to the regulation of this process of adhesion turnover (Caswell et al., 2009). Expression of integrins can dynamically respond to changes in the cell environment, which enables cells to maintain their connection to the ECM (Desgrosellier and Cheresh, 2010; Hynes, 2009). When cells attach to collagen, the high abundance of this matrix protein ensures that it is always available for binding whenever an integrin briefly shifts to a high-affinity conformation (Arjonen et al., 2012). Our data show that vimentin depletion increases the total expression of β1 integrin and the membrane-associated fraction, which is activated in response to collagen binding. In light of these data, we investigated whether the increased β1 integrin affinity caused by vimentin deletion also affects cell adhesion. Flow cytometry and jet-wash evaluations of adhesion strength showed that cell–collagen adhesions were stronger after vimentin deletion, which is consistent with our finding of increased numbers of membrane-associated β1 integrin receptors and increased activation. These data are consistent with previous studies linking integrin activation to reduced receptor recycling rates and increased lysosomal degradation (De Franceschi et al., 2015). Analysis of β1 integrin activation in cells plated on fibrinogen or fibronectin-coated surfaces showed no prominent differences after vimentin deletion.

The ability of cells to sense and respond to collagen depends on spatially and temporally specific alterations of integrin conformation, which in turn regulates activation and aggregation processes (Baade et al., 2019; Liu et al., 2015b; Puklin-Faucher and Sheetz, 2009) that are linked to cluster formation (Welf et al., 2012). Our data show that vimentin also critically regulates β1 integrin receptor clustering on collagen. Despite enhanced integrin affinity and activation after vimentin depletion, integrin clusters remained much smaller than in WT cells. Moreover, we found an inverse relation between the size of β1 integrin clusters and the number of clusters, which is consistent with earlier studies showing that vimentin localizes to nascent adhesions (Terriac et al., 2017) and promotes integrin clustering. These results suggest that vimentin plays a critical role in linking β1 integrin activation to cluster formation in collagen-dependent manner, which is also supported by the vimentin rescue experiments. Previous data has shown that the focal contacts of cultured fibroblasts derived from vimentin-null mice are less regular, less distinct and less mechanically stable than those of WT control cells (Battaglia et al., 2018).

When integrins are activated and bound to matrix ligands, they promote the aggregation of hundreds of adaptors and signaling molecules at their cytoplasmic tails to enable the assembly of a dynamic, macromolecular adhesion complex (Winograd-Katz et al., 2014). The clustering of β1 integrin tail domains leads to the recruitment of talin and paxillin, which in turn creates a platform for proteins that are indirectly linked to integrins, such as PAK and α-actinin (Puklin-Faucher and Sheetz, 2009; Liu et al., 2015a,b; Baade et al., 2019). With mass spectrometry analysis and confirmations by immunoblotting, we assessed the impact of vimentin deletion on proteins in adhesion complexes bound to collagen. Collectively these data indicate that the observed increase of β1 integrin activation in vimentin-null cells is related to higher abundance of talin cell adhesions and talin-dependent, ‘inside-out’ integrin signaling. Previous data has shown that vimentin localizes to cell adhesions at discrete stages of maturation (Terriac et al., 2017). Our findings showed that only filamentous vimentin colocalized with talin, whereas small vimentin particles were in close proximity with talin.

As paxillin colocalizes with talin near the plasma membrane (Dong et al., 2016; Kanchanawong et al., 2010), talin may also interact with vimentin. Indeed, our immunoprecipitation experiments showed association of vimentin and talin in cells migrating on fibrillar collagen. We found that loss of vimentin was associated with decreased recruitment of paxillin into adhesion complexes and that vimentin expression levels were associated with the number and size of adhesion plaques. The colocalization of talin with vimentin filaments suggests that vimentin was incorporated into cell adhesions and may affect its interactions with paxillin. In support of this notion is the previous observation that vimentin expression in epithelial cells increases paxillin turnover in focal adhesions by 4-fold (Mendez et al., 2010). Previous data also indicate that vimentin interactions with focal adhesion proteins regulate the formation of the adhesions (Dave et al., 2013; Menko et al., 2014; Terriac et al., 2017; Tsuruta and Jones, 2003). Taken together, these data indicate that the association of vimentin with paxillin and talin enables the maturation of integrin clusters and the maintenance of collagen adhesion structures that are involved in ‘inside-out’ integrin signaling.

Paxillin is recruited into adhesion structures and, when phosphorylated, paxillin induces Cdc42 activation (Hodge and Ridley, 2016; Iden and Collard, 2008), which in turn drives filopodia formation and cell migration (Chen et al., 2000; Kozma et al., 1995; Nobes and Hall, 1999). We found that vimentin contributes to the activation of Cdc42 and Rac1, and in turn, the formation of cell protrusions. Depletion of vimentin was strongly associated with diminished Cdc42 and Rac1 activity, which correlates with the reduced cell extension formation and cell migration we observed. Notably, Cdc42 is downstream of PAK and enables PAK activation (Rane and Minden, 2014). Our data support the idea that vimentin plays an important role in cell extension formation through control of Cdc42-dependent PAK activation. Moreover, by enhancing recruitment of paxillin into focal adhesions, vimentin may contribute to Cdc42 activation through a paxillin-dependent pathway. Recent data indicate that the paxillin-related focal adhesion protein Hic-5 (also known as TGBB1I1), stabilizes the vimentin network by modulation of Rho GTPases, and its ablation leads to the disassembly of vimentin filaments (Vohnoutka et al., 2019), findings which underpin the involvement of vimentin in adhesion biology.

The formation of cell extensions strongly depends on PAK1-dependent vimentin phosphorylation and filament assembly (Ding et al., 2020). Our findings are consistent with previous data demonstrating that PAK inhibition abrogates vimentin S39, S56 and S72 phosphorylation, as well as vimentin spatial reorganization and cell extension formation (Ding et al., 2020). Furthermore, we found that alterations of vimentin structure are associated with the reduction of β1 integrin cluster size, which is consistent with earlier data linking small vimentin particles with nascent cell adhesions (Terriac et al., 2017).

The formation of integrin clusters is thought to depend on interacting processes that regulate the production, concentration and diffusion of integrin-activating proteins (Welf et al., 2012). Our findings indicate that vimentin affects the organization, structure and function of cell–collagen adhesions by suppressing β1 integrin activation, thereby resulting in well-organized, mature integrin clusters. In contrast, loss of vimentin dissipates local concentrations of integrins, which curtails the growth of integrin clusters and collagen adhesions. Notably, fine tuning of cell adhesion is essential for the formation of cell extensions. In this context, we propose that vimentin filaments are essential for the recruitment of paxillin into adhesion complexes and in the regulation of Cdc42 and PAK1 activation, which are also required for the formation of cell extensions. These processes are ultimately dependent on vimentin phosphorylation mediated by PAK1, which promotes vimentin filament maturation of cell extensions and migration on collagen.

Rat anti-β1 integrin (clone KMI6, ab95623), anti-phospho-vimentin (S72, EP1070Y), goat anti-rabbit IgG H&L (DyLight 488 conjugated, ab150077), rabbit anti-Cdc42 (ab155940), rat anti-α-actinin (EA-53, ab9465), rabbit anti-paxillin (Y113, ab32084), mouse anti-talin1 (ab157808) and anti-PAK1 (phospho S144, ab40795) antibodies were purchased from Abcam (Cambridge, MA). Rabbit anti-phospho-vimentin (S56, 3877S), anti-phospho-vimentin (S39, 13614S), rabbit anti-PAK1 (2602S), rabbit anti-phospho-PAK1 T423 (2601) and rabbit anti-Rac1 (2465S) antibodies were obtained from Cell Signaling Technology (Danvers, MA). Chicken anti-vimentin antibody (N13300-223) was purchased from Novus Biologicals (Littleton, CO). Goat anti-DDR2 antibody (AF2538) was from R&D Systems (Minneapolis, MN). Fibronectin, fibrinogen, ML141 (Cdc42/Rac1 inhibitor), IPA-3 (PAK1 inhibitor), mouse anti-β-actin (A1978), rabbit anti-myosin-10 (APA024223), DAPI was obtained from Sigma-Aldrich (St. Louis, MO). Fluorescent 2 μm beads (09847-5) were from BioTrend (Germany). Alexa Fluor 488-conjugated goat anti-mouse antibody (A11001) Alexa Fluor 568-conjugated goat anti-rabbit antibody (A-11011) and Rhodamine–phalloidin, were from Invitrogen (Carlsbad, CA). Alexa Fluor 647-conjugated AffiniPure F (ab’)₂ fragment donkey anti-chicken IgY (H+L) (703-605-155) was from Jackson ImmunoResearch (West Grove, PA). Rat anti-CD29 (β1 integrin subunit, 9EG7 clone, 553715) was from BD Biosciences (Mississauga, ON). ON-TARGETplus non-target pool (5 nmol), ON-TARGETplus siRNA reagents VIM (5 nmol), ON-TARGETplus siRNA Reagents Pax (5 nmol) and ON-TARGETplus siRNA Tln1 siRNA 3′ UTR CGC and 3′ UTR CAG (custom siRNA, sequences used are available upon request; standard 0.015 µmol regular) were obtained from Horizon (Lafayette, CO). DharmaFECT1 transfection reagent was purchased from Dharmacon (Lafayette, CO). FuGENE HD transfection reagent, BCA protein assay (Pierce), RIPA lysis buffer, IP lysis buffer and Lipofectamine 3000 reagent were obtained from Thermo Fisher Scientific (Mississauga, ON). Type I bovine collagen solution (3 mg/ml) was from Advanced BioMatrix (San Diego, CA). Plasmids containing full-length PAK1 cloned into pCMV6 (plasmid 12209), catalytically inactive mutant PAK1 (K299R; plasmid 12210) or constitutively active PAK1 (T423E; plasmid 12208), GFP–Paxillin (plasmid 15233), GFP–talin1 (plasmid 26724) were obtained from Addgene (Cambridge, MA). pCMV3-mVIM–GFP-spark plasmid was from Sino Biological Inc (Chesterbrook, PA). The FITC anti-rat IgG2a antibody clone MRG2a-83 (407506) was obtained from BioLegend (San Diego, CA)

Mouse NIH-3T3 fibroblasts were obtained from David Calderwood (Yale University, New Haven, CT). Wild-type (mEF WT) and vimentin-null (mEF null) mouse immortalized embryonic fibroblasts, derived from WT and vimentin-knockout mice, respectively, were passaged from a cell line originally isolated by the laboratory of J. Ericsson (Abo Akademi University, Turku, Finland). Vimentin-rescue mEF cells were prepared by transfection of mEF null cells with pCMV3-mVIM–GFP-spark plasmid. Cells were cultured at 37°C in complete DMEM containing 10% Hyclone fetal bovine serum (3T3 cells) or FBS (mEFs) and 10% (v/v) antibiotics. For vimentin siRNA transfection, cells were trypsinized and plated on 100 mm dishes to 60–70% confluence. Cells were treated with Dharmacon ON-TARGET siRNAs plus transfection reagents or with non-targeting control siRNA, following the manufacturer's protocol. To obtain talin1 or paxillin KD cells, mEFs were transfected with respective siRNAs using Lipofectamine 3000. Talin1 or paxillin re-expression was done by co-transfection with GFP–talin1- and GFP–paxillin-containing plasmids, respectively. Cells were grown for 48 h, lysed in RIPA buffer, and whole-cell lysates were collected. Protein concentrations were determined using a BCA protein assay (Pierce). Equal amounts of proteins from each treatment condition were separated on 8% and 10% acrylamide gels and immunoblotted to estimate the efficiency of the KD. For PAK1 mutant experiments, cells were transfected using FuGENE HD DNA transfection reagent according to the manufacturer's instructions.

Cells were plated on collagen-coated 100 mm dishes and lysed with RIPA buffer (25 mM Tris-HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate and 0.1% SDS) supplemented with 1 mM PMSF, 1 mM NaVO₃ and 10 μg/ml protease inhibitor cocktail. Equal amounts of proteins were loaded on 8% or 10% polyacrylamide gels, resolved by SDS–PAGE and transferred to nitrocellulose membranes for 18 h. The membranes were blocked with TBS containing 5% BSA for 1 h followed by overnight incubation with primary antibodies (anti-β-actin was used at a dilution of 1:3000, other primary antibodies were used at a dilution of 1:5000). Membranes were washed and incubated with respective secondary antibodies (1:5000) for 1 h in TBS supplemented with 0.1% Tween-20 at room temperature. All immunoblots were performed in triplicate. Blots were visualized with a Li-Cor Odyssey imager (Lincoln, NB) and quantified densitometrically using ImageJ software (NIH, Bethesda, MD).

In most experiments, cells were cultured on monomeric collagen (0.1 mg/ml), fixed with 1% paraformaldehyde for 15 min, permeabilized in 0.2% Triton X-100 for 10 min, and blocked with 1% BSA. Samples were incubated with appropriate primary antibodies for 1 h at 37°C. Cells were washed with PBS following secondary antibody incubation. In most experiments, cells were stained with DAPI in NP-40 (10 µg/ml) and with Rhodamine–phalloidin for actin filaments. For studies of vimentin, phospho-vimentin (S39, S56 and S72), β1 integrin, paxillin and myosin-10, samples were incubated with appropriate primary antibodies (1:100 in PBS containing 0.2% BSA) for 1 h at 37°C. Cells were washed with PBS and stained with secondary antibody (1:200). Images were obtained with a TCS SP8 confocal microscope (Leica Microsystems, Wetzlar, Germany) equipped with a ×40 oil-immersion objective lens, a Zeiss LSM 800 confocal microscope (Zeiss, Germany) equipped with a ×63 oil-immersion objective lens, and an Olympus SpinSR10 spinning-disk confocal super-resolution microscope. Analysis and quantification of acquired images were done with ImageJ software. The mean cell surface area was determined using the cell outliner plugin (https://imagej.nih.gov/ij/plugins/cell-outliner.html).

Cell extensions were defined as cell protrusions that were >10 μm as measured from the cell centroid to the tip of each extension. Pearson correlation coefficients were determined with JACoP (Just Another Colocalization Plugin; https://imagej.nih.gov/ij/plugins/track/jacop.html) in Fiji (https://imagej.net/Fiji). The spatial distribution of β1integrin clusters with respect to the leading edge was determined using a macro that indicated two separate masks for the cell periphery (1.75 μm from the leading edge) and for the cell body excluding the area of cell periphery. The number and area of β1 integrin clusters were quantified on fluorescence images acquired on the confocal microscope according to the method described by Coelho et al. (2017). Briefly, in Fiji the image background was subtracted using the sliding paraboloid; the local image contrast was enhanced with the CLACHE plugin; the background was minimized with mathematical exponential (EXP); and the brightness and contrast were adjusted automatically followed by auto thresholding. The number and area of β1 integrin clusters were obtained with the Analyze Particles program in Fiji.

For quantification of the total abundance of β1 integrin and the activated fraction of β1 integrin, cells were plated on fibronectin (10 µg/ml)-coated or collagen (1 mg/ml)-coated 100 mm dishes for 3 h (Segal et al., 2001). Cells were washed with ice-cold PBS, scraped immediately with Versene solution and fixed with 1% paraformaldehyde. Cells were washed and incubated with primary antibodies (KMI6 for total β1 integrin and 9EG7 for activated β1 integrin) for 1 h at 37°C. Cells were washed with PBS, stained with FITC-conjugated anti-rat secondary antibodies and analyzed by flow cytometry (Sony SA3800 Spectral Analyzer).

Collagen-coated latex beads (2 μm) were prepared as previously described (Lee et al., 1996). Yellow-green fluorescent beads were incubated with 1 mg/ml acidic solution of bovine type I collagen and neutralized with 1 M NaOH to pH 7.4 to facilitate fibril formation on the bead's surface. Beads were incubated with agitation at 37°C for 1 h. Beads were pelleted, resuspended in PBS, briefly sonicated and counted using a hemocytometer. Cells were counted electronically, and collagen-coated beads were loaded on to the top surfaces of cells at an 8:1 bead:cell ratio for 3 h. To reduce the inclusion of cells bearing loosely or non-specifically bound beads, cells were washed (in PBS), trypsinized and resuspended in DMEM (Segal et al., 2001). BSA (0.01 mg/ml)-, fibrinogen (1 mg/ml)- and poly-L-lysine (0.1%)- coated yellow-green beads (2 μm) were prepared as described above. Samples were analyzed by flow cytometry (Sony SA3800 Spectral Analyzer) to estimate the percentage of cells with bound collagen-coated fluorescent beads.

The relative binding strength of cells to fibrillar collagen or fibrinogen-coated beads was estimated by a shear wash assay (Chong et al., 2007). Experiments were performed as described previously (Coelho et al., 2017) using cells plated on 24-well tissue culture plates at 80–90% confluency. FITC-labeled beads (2 μm) were coated with 1 mg/ml collagen solution as described above. The collagen-coated beads were added to the serum-free medium on the top of cells (8:1 bead:cell ratio) for 2 h. The cells were then subjected to increasing numbers of washes (1–16 washes, shear stress of 3.5 Pa). Cells were fixed with 4% paraformaldehyde for 10 min and stained with DAPI (5 μg/ml). Bound beads and nuclei were visualized by fluorescence microscopy and counted by using the Cell counter plugin in ImageJ (https://imagej.nih.gov/ij/plugins/cell-counter.html).

Cells were lysed in IP lysis buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% NP-40 and 5% glycerol) containing 1 mM PMSF, 1 mM NaVO3, 10 μg/ml leupeptin, and 10 μg/ml aprotinin. Equal amounts of protein from cleared extracts were immunoprecipitated with the Dyna-beads Protein G (Life Technologies) according to the manufacturer's protocol (Invitrogen, Carlsbad, CA, USA) with primary antibodies overnight. Beads were separated using a magnetic stand and washed with an IP lysis buffer. Immunoprecipitated proteins were separated from beads by resuspending in Laemmli sample buffer and boiling for 10 min. The beads were sedimented, and lysates were analyzed by western blotting.

The Cdc42 and Rac1 activation assay was performed with PAK1 binding domain (PBD) bound to glutathione–Sepharose beads. Fibroblasts were plated on collagen-coated 100 mm dishes and incubated for 3 h. Adherent cells were lysed with RIPA buffer. Lysates were clarified by centrifugation, equalized for total volume and protein concentration, and rotated overnight with 30 g of purified PBD-bound beads. The beads were pelleted and washed in PBS, resuspended in Laemmli buffer and boiled. Equal amounts of protein were separated by SDS–PAGE using 15% gels. The Cdc42 activation assay was performed with antibodies against Cdc42, and Rac activation was assayed with antibodies against Rac1.

Cells were cultured to 80–90% confluence on 100 mm tissue culture dishes and then incubated with 1 mg/ml collagen-coated magnetite beads (Sigma-Aldrich). Focal adhesion complexes were isolated from cells as described previously (Wang et al., 2009). In brief, cells were washed with ice-cold PBS to remove unbound beads and scraped into ice-cold cytoskeleton extraction buffer (CKSB; 0.5% Triton X-100, 50 mM NaCl, 300 mM sucrose, 3 mM MgCl₂, 20 μg/ml aprotinin, 1 μg/ml leupeptin, 1 μg/ml pepstatin, 1 mM phenylmethylsulfonyl fluoride and 10 mM PIPES, pH 6.8). The cell–bead suspension was briefly sonicated, and the beads were isolated from the lysate using a magnetic separation stand. The beads were resuspended in fresh ice-cold CKSB and re-isolated magnetically. Then, beads were washed in CSKB, sedimented by centrifugation, resuspended in Laemmli sample buffer and boiled for 10 min to allow the collagen-associated complexes to dissociate from the beads. The beads were separated magnetically, and lysates were analyzed by western blotting.

Vimentin-WT and -null cells were incubated with collagen-coated magnetite beads as described above. Proteins were eluted from beads with 10 mM DTT at 60°C for 1 h. Alkylation was performed using 20 mM iodoacetamide at room temperature for 45 min in the dark. Trypsin (2 μg; Pierce) was added to the samples, which were rotated overnight at 37°C. The eluted proteins were lyophilized in a Speedvac and desalted using a Millipore C18 ziptip. Subsequently, 0.1% formic acid was added to the sample, which was then air-dried with an evaporator. Lyophilized samples were analyzed using a Thermo Scientific Orbitrap Q exactive HFX at the Hospital for Sick Children SPARC BioCentre (Toronto, ON, Canada). Scaffold 4.10 (Proteome Software, Portland, OR, USA) was used for analyzing search results, calculating P-values for each peptide match and matching peptide spectra.

Statistical significance (P<0.05) was determined using the unpaired t-test for two samples or ANOVA for multiple samples. For analysis of the spatial relationship of proteins in cultured cells, Pearson correlation was used (Bolte and Cordelieres, 2006). Analyses were performed with GraphPad Prism 8 software. All experiments were performed at least three independent times in triplicates on separate days. Bar charts show mean±s.e.m.

The authors thank Joao Bronze de Firmino from the imaging facility in the Faculty of Dentistry, University of Toronto for help with developing the Fiji macro and the quantification of β1 integrin distribution. The authors thank the SPARC BioCentre - Molecular Analysis, The Hospital for Sick Children, Toronto, Canada, for assistance with mass spectrometry.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.254359.reviewer-comments.pdf