Vimentin filaments regulate integrin–ligand interactions by binding to the cytoplasmic tail of integrin β3

Precise regulations of cell–matrix interactions and cell adhesion play pivotal roles in many biological processes, such as cell growth, differentiation, formation of blood vessels, axon guidance, wound healing, immune responses, hemostasis and other important functions (Gumbiner, 1996; Hynes, 2002). Integrins, heterodimeric transmembrane proteins composed of α and β subunits, are major cell adhesion receptors that control cell adhesion (Hynes, 2002). Owing to the importance of integrins in these diverse biological functions and their potential as a target for treatment of many different human diseases caused by improper cell adhesion (Goodman and Picard, 2012), the regulatory mechanism of integrin-mediated cell adhesion has been studied extensively (Kim et al., 2011; Shattil et al., 2010). These efforts have resulted in identification of many integrin regulators that bind to integrin cytoplasmic tails, such as talin (Calderwood et al., 1999; Tadokoro et al., 2003), filamin (Lad et al., 2008), Gα13 (Gong et al., 2010), kindlin (Harburger et al., 2009; Moser et al., 2009) and others (Kim et al., 2011; Ye et al., 2014).

Talin, one of the best studied integrin activators, consists of an N-terminal ∼50-kDa globular head and a ∼220-kDa C-terminal rod domain. The talin head domain (THD) can be divided into four subdomains, F0, F1, F2 and F3 (Elliott et al., 2010), and the F3 subdomain contains the integrin-binding site (Calderwood et al., 2002). The head domain has a sixfold higher binding affinity than full-length talin for the integrin β tail, suggesting that the integrin-binding site in talin is masked in full-length talin (Yan et al., 2001). In agreement with this, overexpression of the THD or talin F2F3 domain (talin F23) activates integrin, whereas overexpression of full-length talin has little effect on integrin activation (Han et al., 2006). Thus, the integrin-binding site in talin might be unmasked by intracellular signaling during integrin activation (Calderwood, 2004). It has been suggested that the small G protein Rap1 (which has two isoforms, Rap1a and Rap1b) is involved in this unmasking process (Han et al., 2006). Upon activation by protein kinase C, Rap1 is localized to the plasma membrane and recruits its effector molecule, Rap1-GTP-interacting adaptor molecule (RIAM, also known as APBB1IP) (Han et al., 2006; Wynne et al., 2012), which binds to talin to reveal the integrin-binding site (Chang et al., 2014; Yang et al., 2014).

Some studies have found that other small G proteins are also involved in the integrin activation process (Kinbara et al., 2003). For example, a constitutively active form of R-Ras enhances integrin activation (Zhang et al., 1996), presumably through the activation of phosphatidylinositol-3 kinase (Berrier et al., 2000) and/or the integrin-binding protein filamin (Gawecka et al., 2010). In contrast, a constitutively active form of H-Ras is known to suppress integrin activation by activating Raf and its downstream kinases ERK1 and ERK2 (ERK1/2, also known as MAPK3 and MAPK1, respectively) (Hughes et al., 1997). These Ras signaling pathways might activate an integrin regulator, other than talin, that can bind to integrins and regulate their function, and regulator binding can be a promising therapeutic target for treatment of integrin-mediated diseases. However, unlike the Rap1-mediated integrin activation signaling cascade, the detailed molecular mechanisms of how Ras signaling pathways activate integrins are still unknown. Thus, we chose to study the H-Ras signaling pathway and investigated how H-Ras is involved in integrin suppression with the aim to find a new integrin regulator.

In the effort to find such an integrin regulator, we identified vimentin in its filament form as an integrin regulator. Involvement of vimentin in integrin-mediated cell adhesion and migration has been demonstrated in many studies (Satelli and Li, 2011). During the epithelial-to-mesenchymal transition (EMT), when benign epithelial tumor cells turn into highly invasive malignant tumor cells, vimentin expression is induced and even used as a molecular marker for EMT (Thiery, 2002). In addition, vimentin-deficient mice have severe defects in the wound healing process (Eckes et al., 2000). The cellular localization of vimentin filaments, which are connected to focal adhesions, presumably through vimentin–plectin–integrin-β3 interaction, also indicates a possible role of vimentin in integrin regulation (Bhattacharya et al., 2009; Eckes et al., 1998; Tsuruta and Jones, 2003). Moreover, knocking down the expression of vimentin significantly reduces cell adhesiveness under conditions of flow (Bhattacharya et al., 2009). Despite these reports suggesting a role of vimentin in integrin function, the exact molecular mechanism as to how vimentin regulates cell adhesion and migration has remained unclear. In this study, we suggest that vimentin filaments underneath the plasma membrane can provide integrin attachment sites by direct interaction with integrin tails. These interactions can result in integrin clustering and enhanced integrin-mediated cell adhesion. Thus, our study helps explain how vimentin can control integrin-mediated cellular processes, such as adhesion, migration and invasion. We also suggest that disrupting the integrin–vimentin interaction could be a promising therapeutic strategy to inhibit human diseases caused by unwanted adhesion.

We first investigated the relationship between the H-Ras signaling and the well-known talin-mediated integrin activation process to gather information on the unidentified effector of H-Ras signaling. Because unmasking of the integrin-binding site in talin is known to be required for talin-induced integrin activation (Yang et al., 2014), the H-Ras signaling pathway might exert its effects by regulating a signaling pathway that promotes such an unmasking process. To test this idea, we transfected talin F23 into integrin-αIIbβ3-expressing Chinese hamster ovary cells (denoted CHO/αIIbβ3), together with a constitutively active form of H-Ras, H-Ras(G12V). We stained the cells with PAC1, an antibody specific to active integrin αIIbβ3. Although talin F23 can bypass the unmasking process for integrin activation, expression of H-Ras(G12V) still inhibited the talin-F23-induced integrin activation (Fig. 1A,B) suggesting that H-Ras signaling is not be involved in the unmasking process. Next, we examined whether H-Ras signaling could directly regulate the affinity of talin for the integrin β tail, for example, through modification of talin. To this end, we used a CHO cell line stably expressing Raf fused to the estradiol receptor (denoted CHO/RafER), in which Raf–ERK signaling can be activated by 4-hydroxytamoxifen (4-OHT) to turn on integrin suppression (Hughes et al., 1997). The THD and talin F23 expressed in CHO/RafER cells were pulled down by beads bound to the integrin β1 tail protein, but the pulldown efficiency of wild-type talin was reduced compared to that of THD or talin F23 (Fig. 1C). These data support the hypothesis that the integrin-binding site is masked in wild-type talin. Importantly, activation of Raf by 4-OHT treatment did not change the interaction between the integrin β tail and talin proteins (Fig. 1C). Thus, H-Ras signaling might not regulate the affinity of talin for integrin.

As another possible suppressive mechanism, we hypothesized that H-Ras might recruit a competitive inhibitor for talin binding to integrin. If this hypothesis is correct, an excess amount of talin F23 would prevent the binding of the putative competitive inhibitor and overcome the suppression caused by H-Ras signaling. We found that increasing amounts of talin F23, co-expressed with a fixed amount of H-Ras(G12V), in CHO/αIIbβ3 cells, did not overcome this suppression (Fig. 1D). In contrast, the suppressive effect of an artificial competitive inhibitor, a talin F23(L325R) mutant that binds but does not activate integrin (Wegener et al., 2007), was overcome by overexpression of wild-type talin F23 (Fig. 1D). From these observations, we conclude that generation of a competitive inhibitor for the talin–integrin interaction might not be the way by which H-Ras suppresses integrin affinity. Instead, the inability of excess talin to overcome the H-Ras-mediated suppressive effect suggests that H-Ras can modulate the function of integrin in a way that does not change the talin–integrin interaction.

The integrin β tail is the primary site for the regulation of integrin affinity by known regulators (Kim et al., 2011; Shattil et al., 2010). To determine whether the integrin β tail is required for H-Ras signaling to suppress integrin activation, we constructed a truncation mutant that lacked the β tail (integrin αIIbβ3Δ717). Using this mutant, we tested whether the integrin β tail was involved in H-Ras-mediated suppression of integrin activation. We found that PAC1 binding to αIIbβ3Δ717 was unchanged by H-Ras(G12V), whereas PAC1 binding to wild-type integrin was increased by talin F23 and decreased by H-Ras(G12V), as expected (Fig. S1). These results suggest that the integrin β tail is required for the suppressive effect of H-Ras signaling on integrin activation.

Results described above suggest that H-Ras might (1) recruit an integrin inhibitor that binds to the integrin β tail, or (2) induce dissociation of an integrin activator from the tail. With this information, we decided to perform a proteomic screen to identify the putative regulator, using the β3 tail protein as bait. In the screening, we tried to identify any protein that interacted with integrin differently in the absence or presence of H-Ras signaling. We found that 4-OHT treatment on CHO/RafER cells inhibited the binding of a 55-kDa protein to β3 tail (Fig. 2A), suggesting that the protein was a possible H-Ras effector. Further analysis with two-dimensional electrophoresis similarly showed that a 55-kDa spot with a pI of ∼5.0 was substantially reduced in the 4-OHT-treated condition (Fig. 2B). Using mass spectroscopy, we identified the 55-kDa protein as vimentin. Western blot analysis using an anti-vimentin antibody revealed that vimentin expressed in CHO/RafER cells was co-precipitated with beads bound to integrin β tail protein and that 4-OHT treatment reduced this binding (Fig. 2C), confirming the results from the proteomics study. The degree of vimentin binding seemed to represent different levels of vimentin in the cell extracts, because the amount of vimentin extracted from the 4-OHT-treated cells was also reduced (Fig. 2C). Indeed, the amount of detergent-soluble vimentin was substantially reduced when H-Ras(G12V) was co-transfected (Fig. 2D), presumably due to reorganization of vimentin filaments. In immunocytochemistry experiments, expression of H-Ras(G12V) resulted in the disappearance of vimentin intermediate filaments in most of the cytoplasmic areas and induced their dense localization around the nuclei (Fig. 2E, asterisks; Fig. 2F, red filled circles). In contrast, vimentin expression under standard conditions was observed to form conventional intermediate filaments all over the cytoplasm, including in the peri-membrane regions (Fig. 2E,F, red empty circles). From these results, we hypothesized that vimentin filaments localized near the plasma membrane might positively regulate integrin–ligand interactions, and one of roles of the H-Ras signaling is to induce the translocation of vimentin to the peri-nuclear region, where no vimentin–integrin interaction takes place.

With the hypothesis, we focused on the possible activating effect of vimentin filaments in integrin–ligand interactions. We observed that forced expression of vimentin in the presence of THD caused an increase in PAC1 binding (Fig. 3A). Therefore, we concluded that vimentin does have a role in enhancing integrin–ligand interactions. In addition, THD-induced integrin activation was significantly reduced in CHO cells stably expressing vimentin short hairpin RNA (shRNA), when compared to control-shRNA-expressing cells (Fig. 3B). We also confirmed that the surface expression level of integrin αIIbβ3 was not significantly altered by the transfection of vimentin and/or talin or by the knockdown of vimentin (Fig. 3A,B, bar graphs). In contrast to the reduced PAC1 binding observed in cells stably expressing vimentin shRNA, however, we wish to note that slight knockdown of vimentin by transient transfection of vimentin small interfering RNA (siRNA) sometimes induced an increase in PAC1 binding instead (data not shown), suggesting a complicated and context-dependent role of vimentin in integrin regulation (see Discussion). Because vimentin expression alone without the THD did not increase PAC1 binding (Fig. 3A, red line), we suspected that the role of vimentin might be to induce integrin clustering leading to increased avidity for ligand binding, rather than to change the affinity state of integrin. PAC1 binding represents the combined effect of integrin affinity and integrin clustering because PAC1 is an immunoglobulin M that forms a pentamer, providing a total of ten epitope-binding sites (Shattil et al., 1985). Thus, to specifically measure integrin affinity, we used a monomeric ligand, the tenth repeat of type III fibronectin (FN10), for which binding is dependent only on integrin affinity (Ye et al., 2013). Interestingly, we found that vimentin did not enhance THD-induced binding of FN10 (Fig. 3C, top panel). However, vimentin enhanced the binding of the multimeric ligand cross-linked FN10 (Fig. 3C, bottom panel). The fact that only the binding of a multimeric ligand, but not a monomeric ligand, to the integrin was affected by vimentin expression suggests that vimentin can increase integrin–ligand interactions by inducing the clustering of activated integrins. Consistent with this, the perimeters of the integrin αIIbβ3 clusters stained in suspended CHO/αIIbβ3 cells were significantly increased by co-expression of the THD and vimentin, when compared to cells transfected with empty vector or THD only (Fig. 3D).

Vimentin contains a non-structured head domain (denoted H) at the N-terminus, two coiled-coil domains (often denoted as C1 and C2) in the middle and a C-terminal tail domain (denoted T) (Fig. 4A) (Chernyatina et al., 2012). We examined which domain is responsible for integrin binding. We divided vimentin into N-terminal and C-terminal halves at the linker region between C1 and C2, and the binding of these constructs (expressed in CHO cells) to purified maltose binding protein (MBP)-fused integrin β3 tail was tested. We found that the vimentin head to C1 region bound to the β3 tail protein, but the C2 to tail region did not (Fig. 4B). In a subsequent experiment, we found that a purified glutathione-S-transferase (GST)-fused vimentin head domain bound directly to a purified β3 tail specifically in a concentration-dependent manner (Fig. 4C), showing that vimentin can interact directly with the integrin tail through its head domain. Deletion of 13 amino acids at the C-terminal end of β3 tail (Δ750, Fig. 4D) completely blocked the interaction, suggesting that this region is involved in the interaction. In case of the vimentin head domain, deletions after 46th amino acid (Δ46 and Δ71) did not block the interaction, but a deletion after 21st amino acid (Δ21) did (Fig. 4E). Thus, vimentin amino acids 21–45 seem to be crucial for the binding.

We thought that the observed vimentin-induced change in integrin avidity might be due to the ability of vimentin to make filamentous structures and provide many integrin-binding sites (from the vimentin head domain) under the plasma membrane. If that hypothesis is true, overexpression of a vimentin head domain that is incapable of making filamentous structures might compete with endogenous vimentin filaments for integrin binding, thus abolishing the effect on avidity. In agreement with this idea, the vimentin head domain protein inhibited the interaction between endogenous vimentin and purified β3 tail in a concentration-dependent manner (Fig. 4F). In contrast, the vimentin head domain protein did not compete with other integrin regulators, such as filamin A and talin (Fig. 4F). In addition, THD-induced β3 transmembrane domain tilting was not inhibited by the vimentin head domain in a purified system (Fig. S2), again confirming no competition between talin and vimentin. Interestingly, in agreement with the hypothesis above, expression of the vimentin head domain largely inhibited THD-induced PAC1 binding whereas surface expression of the integrin was not altered (Fig. 5A). Moreover, we observed that expression of the vimentin head domain caused detachment of vimentin filaments from the plasma membrane regions and induced vimentin aggregation near the nuclei, which is very similar to the effect of H-Ras(G12V) (Fig. 5B, asterisk). These results confirm the role of vimentin in regulating integrin–ligand interactions. Moreover, these results also suggest that integrin–ligand interactions are enhanced by the binding of vimentin filaments, not vimentin itself, to integrin.

Because the integrin-binding site was mapped to the vimentin head, we assumed that the vimentin head and C1 domain protein should inhibit THD-induced integrin adhesiveness just like the vimentin head, and that the C2 and tail domain should have no effect. Very surprisingly, we found that both constructs enhanced the THD-induced PAC1 binding (Fig. 5C), although this observation was less pronounced for the head and C1 domain. Integrin surface expression was not altered by those constructs (Fig. 5C, bar graph). We examined whether the head and C1 domain, or C2 and tail domain had an intrinsic ability to increase integrin–ligand interactions. Similar to full-length vimentin, neither fragment induced PAC1 binding when expressed without THD (not shown). However, we found that these proteins destroyed the filamentous vimentin structures and induced punctuated vimentin staining patterns throughout the cells (Fig. 5D, arrows). This pattern is similar to that formed by aggregates of the vimentin filament precursor observed at the lamellipodia of migrating fibroblasts (Goldman et al., 1996; Helfand et al., 2011; Strelkov et al., 2002) or at cell adhesion sites during cell spreading (Correia et al., 1999). In agreement with our observation, short peptide either from the N- or C-terminal region of vimentin (R1 and R2 in Fig. 4A) induces fragmentation of vimentin filaments by inhibiting vimentin assembly (Goldman et al., 1996; Helfand et al., 2011; Strelkov et al., 2002). Theoretically, once vimentin filaments are converted into a large number of smaller pieces, the vimentin fragments can be released from the filamentous structure and move more freely to the plasma membrane than vimentin in the filamentous form. In addition, the filaments would contain enough vimentin molecules to mediate integrin clustering. Thus, our interpretation of these interesting phenomena is that reorganization of vimentin filaments into smaller fragments upon expression of the head and C1, or C2 and tail domain might be responsible for the enhanced integrin adhesiveness. If we consider both the positive (by inducing fragmentation of vimentin filaments) and negative (by inhibiting integrin–vimentin filament interactions) effects, the reason why the head and C1 domain induces a weaker integrin-ligand interaction than the C2 and tail domain (Fig. 5C) can also be explained.

Next, we investigated the role of the vimentin–integrin interaction on cell migration and metastasis. 4T1 cells are derived from a highly invasive mouse breast cancer cell line from the BALB/c strain (Aslakson and Miller, 1992) and expresses integrin β3 (Fig. S3A). These cells bind to fibrinogen, a ligand of β3 integrin (integrin αVβ3) expressed in non-platelet cells. This binding increased in the presence of THD and vimentin (Fig. S3B,C), similar to integrin αIIbβ3 (Fig. 3A). To see the consequence of integrin–vimentin filament interaction loss, we infected 4T1 cells with lentivirus encoding the vimentin head. The vimentin head in the lentiviral construct contains a downstream internal ribosome entry site (IRES) followed by the gene encoding GFP. Thus, we sorted GFP-positive cells and pooled them to generate a population of vimentin-head-expressing cells (Fig. 6A). As a control, we also collected GFP-negative cells (Fig. 6A). Although vimentin expression in both GFP-positive and -negative cells was rarely detected in normal culture conditions, treatment with transforming growth factor (TGF)-β, the signal that induces EMT (Liu et al., 2012), significantly induced vimentin expression (Fig. 6B). In our migration assay, both of these cell populations, GFP-negative and -positive cells, showed a relatively low migration rate in the absence of TGF-β (Fig. 6C). When treated with TGF-β, there was a tenfold increase in the migration rate of GFP-negative cells (Fig. 6C,D). However, in the case of GFP-positive cells, the TGF-β-induced migration was largely suppressed (Fig. 6C,D). Consistent with this result, BALB/c mice that were intravenously injected with GFP-positive 4T1 cells also exhibited a reduction in nodule formation on the lung (Fig. 6E,F). Because the effect of vimentin head must be dependent on the presence of vimentin filaments, we randomly selected metastatic nodules and examined vimentin expression in each nodule. Interestingly, induction of vimentin expression was observed in all the nodules tested [eight nodules from GFP-positive cells (Fig. 6G) and eight nodules from GFP-negative cells (data not shown)], suggesting a crucial role for vimentin in cancer metastasis. From these results, we suggest that one of the roles of vimentin during metastasis is to regulate integrin adhesiveness and that inhibition of vimentin-filament-dependent integrin regulation might reduce cell migration and metastasis. However, we do not rule out the possibility that expression of vimentin head affects other functions of vimentin involved in cell migration, rather than inhibiting the integrin-β3–vimentin-filament interaction.

In this report, we first identified vimentin filaments as new regulators for integrin function. Our study suggests that vimentin filaments underneath the plasma membrane have many vimentin heads exposed along the filaments, and thus can provide many integrin-binding sites for integrin clustering, which in turn can enhance integrin–ligand interaction (Fig. S4). Because integrins localize to the plasma membrane, this interaction is only possible near the plasma membrane. Thus, we suggest that translocation of vimentin to the peri-nuclear region, for example by H-Ras signaling, might have a negative effect on the integrin–ligand interaction. However, owing to the experimental difficulty of testing the suppressive effect of H-Ras with vimentin filaments kept intact under the plasma membrane, we were unable to determine whether the suppressive effect of the H-Ras is solely dependent on the rearrangement of vimentin filaments. Therefore, at the moment, the initial question of how H-Ras suppresses integrin function has not been fully established. Second, our study provides a possible explanation of how organization of vimentin filaments is associated with cell adhesion. Not only long vimentin filaments, but also smaller ones containing a high enough number of vimentin subunits, would induce changes in integrin clustering. Therefore, regulation of the local assembly or disassembly of vimentin filaments near the cell surface could be a way of controlling integrin-mediated adhesiveness at the region involved. Reducing vimentin expression level to a certain degree might reduce the number of the filaments near cell surface (negative effect) or induce disassembly of vimentin filaments leaving smaller filaments capable of inducing integrin clustering (positive effect). Because of these possible dual effects, we think transient knockdown of vimentin (description on Fig. 3B) might cause different results depending on the degree of knockdown efficiency and the knockdown-induced status of vimentin filaments in each experiment. Finally, our results showed that the vimentin head impairs cell migration and metastasis (Fig. 6), likely by inhibiting integrin–vimentin interaction. Unlike after the knockdown of vimentin, overexpression of the vimentin head would detach the integrin from vimentin filaments regardless of the status of the filaments, which can consistently reduce the clustering effect of vimentin. Thus, we suggest that targeting the integrin–vimentin interaction could be a potential therapeutic approach to treat cell-adhesion-mediated diseases.

It is well known that the expression of vimentin correlates with an invasive cellular phenotype, and forced expression of vimentin can convert epithelial cells into invasive mesenchymal cells (Mendez et al., 2010). Our study suggests that induction of vimentin expression during EMT can increase the overall adhesiveness of tumor cells toward the extracellular matrix (ECM). The increased cell–ECM interaction might become dominant over cell–cell interactions, which would cause individual cell migration or invasion. In addition to vimentin expression, the dynamic regulation of vimentin structure seems to be important in cell adhesion. For example, during cell adhesion, many smaller fragments of vimentin filaments are observed (Correia et al., 1999). In addition, the small fragments, which are similar to those observed in this study (Fig. 5D), have been suggested to modulate the formation of lamellipodia (Helfand et al., 2011). According to our study, vimentin filaments in mesenchymal cells might work as a reservoir for integrin regulators; once the filaments are broken into many smaller fragments in the cellular region where increased cell–ECM interaction is required, the fragments are able to modulate cell attachment at the site by increasing integrin avidity (Fig. S4). Therefore, our study can explain the role of expression and reorganization of vimentin filaments in adhesiveness.

Owing to their crucial roles in extracellular matrix binding, integrins have been considered as potential therapeutic targets to reduce unwanted cell adhesion and migration. Indeed, many anti-integrin drugs have been investigated for curing autoimmune diseases, cancer, thrombosis and so on (Goodman and Picard, 2012). However, to the best of our knowledge, there are only four integrin antagonists currently used in the clinic: three integrin αIIbβ3 antagonists, blocking platelet adhesion for the treatment of acute thrombosis, and one antagonist targeting the α4 integrin, blocking leukocyte adhesion for the treatment of autoimmune diseases. The mechanism of action of these and other integrin-targeting drugs under investigation in clinical trials is direct blockage of integrin–ligand interaction by targeting the ligand-binding sites in integrins through small molecules or monoclonal antibodies (Goodman and Picard, 2012). Given the importance of integrins in normal physiology, complete blockage of integrin function might disrupt not only the newly formed cell adhesion but also the pre-existing cell adhesion, thus causing severe adverse effects such as the detachment of adherent cells, which might account for the failure of anti-integrin drugs targeting those cells. Even in floating cells, bleeding problems can occur with the use of αIIbβ3 antagonists (Quinn et al., 2003) and viral diseases can occur with the use of the α4 integrin antagonist (Langer-Gould et al., 2005). In addition, several reports have demonstrated that drugs targeting integrin–ligand binding sites can stimulate integrin signaling when in low concentration, rather than block it (Reynolds et al., 2009), presumably because the antagonizing drugs act as a ligand (agonism). Because of the mechanism-based toxicity and agonism, many attempts have been made to identify and target integrin regulators instead. These regulators might serve as an alternative way to control the function of integrins indirectly, whereby the mode of action would only target newly formed adhesions and should not have the associated adverse effects. In this regard, we propose that the integrin–vimentin interaction could be a potent target to overcome the current huddle in anti-integrin therapy. Based on our study, the inhibition of integrin–vimentin interaction significantly reduced the migration as well as metastasis of cancer cells (Fig. 6). Considering that the role of vimentin we found in our study is to induce integrin clustering and thus to enhance integrin–ligand interaction, blocking the interaction would not perturb pre-existing cell adhesion but will inhibit only new adhesion formation. At the same time, the mechanism of action will not provoke integrin signaling. Therefore, we believe that our study not only explains the role of vimentin in cell adhesion, but also provides a novel therapeutic target for treatment of cell-adhesion-mediated diseases, such as metastasis.

CHO cells and 4T1 cells were purchased from the Korean Collection for Type Culture and the American Type Culture Collection, respectively. CHO/RafER and CHO/αIIbβ3 cells were kind gifts from Mark Ginsberg (University of California San Diego, CA), and they were authenticated by 4-hydroxytamoxifen (4-OHT) responsiveness and D57 binding, respectively, during their use. HA-tagged talin constructs (wild type, head domain and F2F3 domain), HA-tagged H-Ras(G12V), integrin αIIb and integrin β3 constructs (provided by Mark Ginsberg) were described previously (Han et al., 2006). An untagged version of H-Ras(G12V) was kindly provided by Young Do Yoo (Korea University, Seoul, Republic of Korea). The GFP-tagged THD was generated by inserting a PCR-amplified THD into pEGFP-C1 (Clontech). Vimentin cDNA was purchased from the Korea Human Gene Bank and used in PCR to generate a HA-tagged vimentin head domain (amino acid residues 2–95), a head and C1 domain (H-C1, amino acid residues 2–254), and a C2 and tail domain (C2-T, amino acid residues 255–466) construct in pcDNA3, as well as a GST-fused vimentin head domain in pGEX4T-1 (GE Healthcare). To generate the His-pMAL/β3 tail construct for purification of the maltose binding protein (MBP)–β3-tail fusion protein, the integrin β3 tail region was amplified by PCR and cloned into BamHI and XhoI sites of His-pMAL vector (provided by Hyun Kyu Song, Korea University, Seoul, Republic of Korea). Short hairpin RNA (shRNA) against vimentin was generated by ligation of annealed oligonucleotide containing the target sequence (5′-AATACCAAGACCTGCTCAATC-3′) into a lentivirus vector, pLKO.1 (Addgene). Its scrambled sequence (5′-GCAATGCATGCCATACTAACA-3′) was used to generate the control shRNA construct. These lentivirus vectors were transfected into Lenti-X 293T cell line (Clonetech) to generate lentivirus particles as previously described (Kim et al., 2012). Complex-specific anti-integrin-αIIbβ3 antibody (D57, 10 µg/ml for flow cytometry), activation-specific anti-αIIbβ3 antibody (PAC1, mouse ascetic fluid, 1:400 for flow cytometry) and anti-integrin-β3-tail antiserum (Rb8275, rabbit serum, 1:1000 for western blotting) were provided by Mark Ginsberg. Anti-vimentin (Sigma-Aldrich, V6389, 1:1000 for western blotting, 1:100 for cytochemistry), anti-HA (Santa Cruz Biotechnology, sc-805, 1:1000 for western blot, 1:100 for cytochemistry), anti-β-actin (Santa Cruz Biotechnology, sc-130656, 1:1000 dilution for western blotting), anti-GAPDH (Santa Cruz Biotechnology, sc-20357, 1:500 for western blotting), anti-GFP (Santa Cruz Biotechnology, sc-9996, 1:200 for western blotting) and anti-GST (GE Healthcare, RPN1236, 1:1000 for western blotting) antibodies were obtained commercially.

To prepare monovalent ligand for integrin αIIbβ3, the tenth repeat of fibronectin type III (FN10) was used as previously described (Ye et al., 2013). Briefly, pGEX-6P-1 vector containing FN10 was introduced into E. coli strain BL21-DE3, and protein was expressed by adding IPTG to the culture. Expressed GST–FN10 was purified using glutathione–Sepharose-4B (GS4B, GE Healthcare) according to the manufacturer's instructions. The elution buffer was exchanged to PreScission protease cleavage buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1 mM DTT, pH 7.5) using Zebra™ spin desalting columns (Thermo Scientific), and the purified proteins were incubated with PreScission protease (GE Healthcare) at 4°C overnight to remove GST. The free GST was removed by incubating the mixture with GS4B for 6 h at 4°C. The flow-through was loaded to the desalting column equilibrated with PBS for the biotinlyation reaction using EZ-Link NHS-Biotin (Thermo Scientific). To prepare the multimeric ligand, purified GST–FN10 was first biotinylated and then incubated with a fourfold molar excess of 1,11-bis(maleimido)triethylene glycol [(BM(PEG)3, Thermo Scientific] at room temperature for 1 h to allow cross-linking events to occur. The reaction was treated with 20 mM DTT and incubated for 15 min at room temperature to quench all unreacted reagent, followed by a desalting step. MBP–β3-tail protein was purified from BL21-DE3 cells transformed with His-pMAL/β3 tail using amylose resin (New England BioLabs) according to the manufacturer's guide. Similarly, GST-tagged vimentin head protein was purified from BL21-DE3 cells transformed with pGEX4T-1/Vim-H using GS4B.

CHO cells transfected with various vimentin constructs were lysed with a lysis buffer [1% Triton X-100, 150 mM NaCl, 50 mM HEPES, pH 7.4, protease inhibitor cocktail (Roche)]. Cell lysates were incubated for 20 min at 4°C with agitation, homogenized by passing through a 26-gauge needle three times and then clarified by centrifugation at 17,000 g for 20 min. The cell lysates (or GST proteins as indicated) were added to purified MBP-fused integrin β3 tail (or GST-tagged vimentin head protein) and incubated for 2–16 h at 4°C with agitation. The mixtures were further incubated with amylose resin (or GS4B) for 2 h. Beads were washed three times and the bound proteins were eluted by boiling the beads in SDS-PAGE sample buffer. Eluted proteins were analyzed by western blotting using anti-vimentin antibody or anti-HA antibody.

Cells were transfected with GFP–THD (or pEGFP-C1 empty vector) and vimentin cDNAs (or empty vector) at a 1:8 ratio, as indicated in each experiment. At 24 h post transfection, cells were detached and ∼5×10⁵ cells were incubated with 4 µg/ml PAC1 or 100 µg/ml biotinylated monomeric or oligomeric FN10. Specific binding was calculated as the MFI–MFI₀ [where MFI is mean fluorescence intensity of binding of each ligand and MFI₀ is mean fluorescence intensity of the binding in the presence of 10 mM EDTA (for PAC1 binding) or 100 µM integrilin, an integrin-αIIbβ3-specific inhibitor (for monomeric or cross-linked FN10 binding)]. Because monomeric and cross-linked FN10 might bind to endogenous integrins in CHO/αIIbβ3 cells, only the integrin-inhibitable binding was considered as specific FN10 binding to integrin αIIbβ3 (Ye et al., 2013). The reaction was incubated for 30 min at room temperature. For monomeric or cross-linked FN10-binding assays, cells were fixed with 3.7% formaldehyde for 10 min after the incubation. After washing, cells were further incubated with fluorophore-conjugated secondary reagents, allophycocyanin-conjugated anti-IgM or streptavidin. The stained cells were analyzed using a FACSCalibur (BD Biosciences). Dot plots were generated by WinMDI (Scripps institute). MATLAB R2014a (MathWorks) was used to calculate the geometric means of PAC1 binding at different levels of integrin expression and to generate dot plots with the geometric means indicated as red dots, as previously described (Kim et al., 2009).

CHO cells on gelatin-coated cover glasses were fixed with 3.7% formaldehyde and then permeabilized with phosphate-buffered saline (PBS) containing 0.1% Triton X-100 (PBS-t). After treatment with a blocking solution (10% goat serum and 0.5% gelatin in PBS-t), cells were stained with primary antibodies (anti-vimentin antibody and/or anti-HA antibody) and the appropriate fluorophore-conjugated secondary antibodies. After the cover glasses were mounted using Fluorescence Mounting Medium (Dako) containing Hoechst 33342 (Invitrogen), cells were observed under a fluorescence microscope (Ti-E, Nikon) equipped with a 100× (1.4 NA) plan-apochromat objective lens. More than ten microscopic fields were selected based on the expression of transfection marker, and the fluorescence images were collected by using a charge-coupled device camera (DS-Qi2, Nikon) and deconvoluted by using NIS-Elements AR (Nikon).

Upper chambers of a Transwell^® culture plate with 8-μm pores (Corning) were incubated with 10 µg/ml fibrinogen in PBS overnight. A 4T1 cell suspension (5×10⁴ cells/ml) was placed in the chamber, and cells were allowed to migrate through membrane pores during an incubation at 37°C in 5% CO₂ for 20 h. Next, cells were fixed and stained with Cell Staining Solution (Cell Biolabs). After washing with PBS, images of migrated cells on the opposite side of the membrane were captured with an inverted microscope (Motic AE2000). The number of migrated cells in each field was counted and the average number of triplicate samples was displayed as a bar graph with s.e.m.

7-week-old female BALB/c mice were purchased from Orient Bio, Korea. Approximately 10⁶ GFP-negative or -positive 4T1 cells in 100 µl PBS were injected into the tail veins of the mice (n=5 per each condition). After 2 weeks, all mice were killed with cervical dislocation and isolated lungs were frozen at −80°C before taking pictures with a digital camera. All the nodules displayed in the pictures were counted to generate the scatter plots. Animal studies were approved by the Institutional Animal Care and Use Committee of Korea University.

We thank Mark Ginsberg (University of California San Diego, CA) for his valuable input on this study.