Discoidin domain receptor 1 controls linear invadosome formation via a Cdc42–Tuba pathway

Type I collagen fibrils are present in tumors, where they were long considered to be a simple physical and structural barrier to inhibit tumor progression and metastasis. However, type I collagen is overexpressed in a large number of cancers, and, paradoxically, a high expression is correlated with an increased risk of metastasis, for instance in breast and lung cancers (Ramaswamy et al., 2003; Gilkes et al., 2013). Collagen overexpression is not the only factor involved in cancer progression. Indeed, the size, diameter, morphology, and cross-linking of type I collagen fibrils have an impact on tumor cell proliferation and metastatic growth (Levental et al., 2009; Cox et al., 2013). Moreover, type I collagen fibrils promote the activity of matrix metalloproteases (MMPs; Ruangpanit et al., 2001).

We previously discovered that type I collagen fibrils are powerful and physiological inducers of invadosomes, which are F-actin–rich structures able to degrade the ECM (Juin et al., 2012). The term invadosomes refers to podosomes in normal cells as well as to invadopodia in cancer cells. Both are matrix-degrading structures allowing matrix remodeling and cell invasion due to the activity of MMPs such as MMP2, MMP9, and MT1-MMP (Hoshino et al., 2013). Invadosomes in some cancers correlate with their ability to metastasize (Eckert et al., 2011). Moreover, invadosomes were recently involved in tumor cell extravasation and demonstrated to be a therapeutic target for metastasis (Leong et al., 2014). Invadosome formation, organization, and activation are controlled by RhoGTPases such as RhoA, Rac1, and Cdc42 (Moreau et al., 2003; Di Martino et al., 2014) and also by Src kinases (Tarone et al., 1985; Linder et al., 2000; Hauck et al., 2002). The invadosome basic module corresponds to a central F-actin core composed of actin-binding proteins like neuronal Wiskott–Aldrich Syndrome protein (N-WASP), the Arp2/3 complex, and cortactin, which is associated with scaffold proteins such as Tks5 (Destaing et al., 2011; Linder et al., 2011; Murphy and Courtneidge, 2011). This actin core may be surrounded by a ring of regulating proteins like integrins, vinculin, and talin. Invadosomes are found as individual items, aggregates, or organized into “rosettes” according to cellular models and context. They are constitutive in various cancer cells and in osteoclasts, but in most cell types they are absent in basal conditions, although inducible by various stimuli including cytokines (PDGF, VEGF, and TGF-β) or various compounds (phorbol esters, cytotoxic necrotizing factor 1, and sodium fluoride; Albiges-Rizo et al., 2009). Our recent data showed that type I collagen fibrils induce invadosome formation in most cell types tested, such as endothelial cells and fibroblasts. Moreover, type I collagen fibrils promoted a linear reorganization of invadopodia in cancer cell lines, which was associated with an increase in ECM-degrading activity. Invadosomes induced or reorganized by collagen I aligned along the collagen fibers, and we thus called them linear invadosomes. Two studies have confirmed the induction of linear invadosomes upon cell contact with collagen fibrils (Monteiro et al., 2013; Schachtner et al., 2013). Interestingly, although β1 integrin family members are the major receptors for type I collagen (Leitinger, 2011) and are associated with classical invadosomes in many cell types, we found that they were not necessary for linear invadosome formation (Juin et al., 2012), raising the question about the ECM receptor involved.

Discoidin domain receptors (DDRs) are a ubiquitously expressed family of receptors known to interact with collagens, in particular fibrillar collagens I–III (Shrivastava et al., 1997; Vogel et al., 1997). DDRs only bind collagens in their native physiological triple-helical conformation and do not recognize denatured collagens such as gelatin (Konitsiotis et al., 2008). The DDR receptor family belongs to the large group of receptor tyrosine kinases (RTKs) and is composed of two members, DDR1 and DDR2. Ligand interaction with DDRs promotes tyrosine autophosphorylation as with classical RTKs, although with very slow and persistent kinetics (Vogel et al., 1997). The DDRs are considered to be collagen sensors and act on tissue homeostasis, as well as on many cellular processes, including cell proliferation and differentiation, cell adhesion, cell migration, and invasion (Leitinger, 2014). These latter properties clearly connect them with cancer. Indeed, several recent studies show that the DDRs are often up-regulated in various cancers (for review see Valiathan et al., 2012). Notably, DDR1 was found overexpressed in lung and breast cancers (Barker et al., 1995; Ford et al., 2007), where a high expression level was correlated with a poor prognosis and metastasis formation (Yang et al., 2010; Valencia et al., 2012; Miao et al., 2013).

Because both DDR1 and collagen I are overexpressed in cancers and associated with metastasis development, and as type I collagen fibrils promote linear invadosome formation, we hypothesized that DDR1 could be the collagen I receptor involved in the formation of linear invadosomes and subsequent cellular invasion.

For this study, we selected breast cancer and lung cancer cell lines with high levels of DDR1 expression. We found that MDA-MB-231 and A549 cells, derived from human breast and lung cancers, respectively, express DDR1 (see Fig. 2 B and Fig. S2 A). We first analyzed the formation of invadopodia in these cells. As shown by dual F-actin/cortactin immunostaining on fluorescent gelatin, A549 cells do not form constitutive invadopodia, whereas MDA-MB-231 cells do (Fig. 1 A). Consequently, only MDA-MB-231 cells degrade gelatin in the in situ zymography assay. However, when seeded on collagen I fibrils, both cell types were able to form linear invadosomes (Fig. 1, B–D). These dynamic structures formed along collagen fibrils are composed of F-actin, cortactin, and Tks5, which are classical markers for invadosomes (Fig. 1, B and C; and Video 1). These results confirm our previous data demonstrating that type I collagen fibrils reorganized invadopodia from MDA-MB-231 cells into linear invadosomes (Juin et al., 2012). In addition, we show that type I collagen fibrils strongly induced linear invadosomes in cancer cells that do not exhibit constitutive invadopodia (Fig. 1 D). In MDA-MB-231 cells, the invadosome reorganization was also associated with an increase in the percentage of cells exhibiting these structures (Fig. 1 D) and was correlated with an increase in the global degradation activity of cells (Fig. S1 A). Altogether, we show that cancer cells expressing DDR1 can form linear invadosomes when plated on type I collagen fibrils and that contact with type I collagen fibrils increases the ability of the cells to degrade the ECM.

To investigate whether DDR1 played a role in linear invadosome formation, we analyzed DDR1 subcellular localization when cells were plated onto type I collagen fibrils. In order to do this, we transfected or infected MDA-MB-231 cells with either a DDR1-Flag construct or a DDR1-GFP lentiviral construct. We found that tagged DDR1 colocalized with linear invadosomes and type I collagen fibrils in MDA-MB-231 cells (Fig. 2 A). This result was confirmed with endogenous DDR1 when using an anti-DDR1 antibody in MDA-MB-231 and A549 cells (Fig. S1, B and C).

To determine DDR1 involvement in linear invadosome formation, we used an RNA interference strategy. Two to three distinct siRNAs were used to deplete DDR1 in both cell types (Fig. 2, B and C; and Fig. S2, A–C), and linear invadosomes were quantified upon plating on type I collagen fibrils. We found that depletion of DDR1 promoted a significant decrease in the percentage of cells able to form linear invadosomes in both cell types (Fig. 2, C and D; and Fig. S2, B and C). It also strongly decreased the number of linear invadosomes per cell (Fig. 2 E), altogether highlighting a major role of DDR1 in linear invadosome formation. To confirm these data, we performed a rescue experiment. We found that lentiviral-mediated expression of DDR1-GFP restored linear invadosome formation in cells transfected with a DDR1 siRNA, which is associated with a colocalization between Tks5 and DDR1 (Fig. S2, D and E). We have previously shown that collagen I–induced linear invadosomes were able to degrade not only gelatin but also collagen I fibrils themselves (Juin et al., 2012). Using second harmonic generation (SHG) microscopy that allows collagen fibril visualization without any staining, we thus quantified the consequences of DDR1 depletion on collagen fibril degradation (Gailhouste et al., 2010). As expected, the decrease of linear invadosome formation was correlated with a decrease in the cell capacity to degrade type I collagen fibrils (Fig. 2 F). Altogether, these results demonstrate the critical role of DDR1 in the formation and activity of type I collagen-induced invadosomes.

These results raised the question about a potential role of DDR1 in the formation and function of classical invadosomes. We thus silenced DDR1 in MDA-MB-231 and Huh6 cells, which both exhibit constitutive invadopodia. Interestingly, whereas we were not able to localize DDR1 at invadopodia, we found that decreasing DDR1 expression using two different siRNAs altered invadosome formation and decreased cell degradation capacity in MDA-MB-231 and Huh6 cells (Fig. S3).

As DDR1 is a tyrosine kinase receptor, we investigated the involvement of DDR1 kinase activity in linear invadosome formation and degradation function. To this end, we used nilotinib, developed as a Bcr-Abl kinase inhibitor but later shown to inhibit DDR1 kinase activity highly efficiently (Day et al., 2008). We first confirmed using immunoprecipitation that type I collagen promoted DDR1 tyrosine phosphorylation and that nilotinib almost completely abrogated it (Fig. 3 A). We found however that nilotinib treatment did not affect linear invadosome formation (Fig. 3, B–D). Indeed, type I collagen stimulation was still able to reorganize F-actin along fibrils, and Tks5 remained associated with the structures (Fig. 3 B). As assessed by the quantification of the SHG (Fig. 3 E) or of the cleaved collagen antibody signal (Fig. S4), we also found that nilotinib treatment did not affect linear invadosome degradation activity.

Moreover, we used three independent monoclonal antibodies that block DDR1 autophosphorylation without interfering with collagen binding (Carafoli et al., 2012), and we obtained the same results (Fig. 3 F), demonstrating that linear invadosome formation and activity are indeed independent of DDR1 kinase activity. In the next part of this study, we thus aimed at understanding which signaling pathway is responsible for the role of DDR1 in linear invadosome formation.

c-Src is well known as a key molecule implicated in the formation and activity of classical invadosomes (Tarone et al., 1985). Src inhibition or depletion is sufficient to abolish classical invadosome formation. In addition, c-Src has been shown to be required in DDR signaling for full phosphorylation after ligand binding (Dejmek et al., 2003; Yang et al., 2005). This prompted us to examine c-Src involvement in DDR1-induced linear invadosome formation. As expected, the c-Src inhibitor PP2 abolished invadopodia formation and degradation activity in MDA-MB-231 cells plated on gelatin (Fig. 4 A). Surprisingly, when MDA-MB-231 cells were seeded on type I collagen, or on a mixed matrix composed of gelatin associated with type I collagen fibrils, PP2 treatment had no impact on linear invadosome formation, on gelatin degradation (Fig. 4, B and C), or on type I collagen fibril degradation (Fig. 4 C). This latter finding is supported by the fact that the cleaved collagen signal was not modified upon PP2 treatment when compared with the control condition (Fig. S4).

Moreover, we demonstrated that DDR1 depletion did not modify c-Src phosphorylation whether cells were plated on gelatin or collagen I (Fig. 4 D). The lack of involvement of c-Src was confirmed using SYF cells, which do not express either Src, Yes, or Fyn, three members of the Src kinase family; and, as control, SYF-Src cells, which are the same cells that stably express c-Src (Fig. 4 E). We first confirmed that both cell lines express DDR1 (Fig. 4 F). We found that control and SYF cells do not exhibit invadopodia on gelatin (Fig. 4 G). However, when plated on a mixed matrix, SYF cells had the same potential as control cells to form linear invadosomes, and these invadosomes were fully active at degrading the matrix (Fig. 4 H). All these results show that c-Src is not involved in the formation or in the degradation activity of type I collagen–induced linear invadosomes.

It is well established that RhoGTPases, principally RhoA, Rac1, and Cdc42, control actin cytoskeleton remodeling and invadosome formation (Linder et al., 2011). Using siRNAs targeting these three proteins, we investigated their respective involvement in linear invadosome formation. We used two distinct siRNAs per GTPase and first checked their efficiency by specifically depleting their corresponding targets (Fig. 5 A). We then measured their impact on linear invadosome formation. We found that only Cdc42 depletion had an impact on linear invadosome formation (Fig. 5, B and C). We then expressed constitutively active and inactive forms of Cdc42 in MDA-MB-231 cells seeded on type I collagen fibrils. We found that the constitutively active form of Cdc42, GFP-V12Cdc42, colocalized with linear invadosomes (Fig. 6 A), unlike the Cdc42 dominant-negative form, GFP-N17Cdc42 (Fig. 6 B). Moreover, we found that expression of GFP-V12Cdc42 enhanced the ability of MDA-MB-231 cells to form linear invadosomes, whereas expression of GFP-N17Cdc42 had the opposite effect (Fig. 6 C). It has been shown that type I collagen fibrils can promote Cdc42 activation (Sato et al., 2003). We thus analyzed the activity level of Cdc42 in cells plated on type I collagen fibrils with or without depletion of DDR1. We first confirmed that type I collagen significantly promoted Cdc42 activation, and found that this effect of type I collagen was abolished in cells with DDR1 depletion (Fig. 6 D). This result was strengthened by the colocalization of DDR1 with the active form of Cdc42 (Fig. 6 E). In addition, using the Raichu Cdc42 biosensor (Itoh et al., 2002) on living cells seeded on type I collagen, we showed a signal corresponding to activated Cdc42 along collagen fibrils (69 hits with a high FRET ratio along collagen fibrils out of 79 cells observed; Fig. 6 F). All these data show that Cdc42 is involved in relaying the collagen I signal through DDR1 for the formation of linear invadosomes.

To go further concerning the link between DDR1 and Cdc42, we searched for a GEF involved in Cdc42 activation upon type I collagen fibril induction. For this purpose, we performed an RNAi screen targeting 14 Cdc42-specific GEFs (Table S1) on cell ability to form linear invadosomes (Cook et al., 2014). Our screen revealed that the depletion of the GEF Tuba impacts on linear invadosome formation. Tuba is a Cdc42-specific GEF but also acts as a scaffold protein to link dynamin with actin regulatory proteins such as N-WASP. To confirm this result, we used two distinct siRNAs to deplete Tuba expression in MDA-MB-231 and A549 cells (Fig. 7, A and B). We demonstrated that Tuba depletion induces a decrease in cell ability to form linear invadosomes in both cell types (Fig. 7, A–C). In addition, we observed a colocalization between DDR1 and Tuba in linear invadosomes of DDR1-GFP–expressing cells that supports a link between these two molecules (Fig. 7 D). Thus, this is the first demonstration of the involvement of Tuba in invadosome formation. To address Tuba participation in classical invadopodia, we analyzed Tuba localization in MDA-MB-231 cells seeded on gelatin. Interestingly, Tuba did not colocalize with classical invadosomes while it was present on linear invadosomes (Fig. 7 E). These data suggest that DDR1 can recruit Tuba that can specifically activate Cdc42 to induce linear invadosome formation.

DDR1 is known to be involved in cancer cell invasion and metastasis induction (Valiathan et al., 2012). Because type I collagen fibrils are part of the tumor microenvironment, we studied whether DDR1 was involved in the invasion of a 3D collagen gel by linear invadosome-bearing tumor cells. We first demonstrated that MDA-MB-231 cells were able to form linear invadosomes in a 3D collagen gel (Fig. 8 A). We further used an invasion assay (Lopez et al., 2005) consisting of a type I collagen gel polymerized into Boyden chambers. Gels were polymerized at a 1 mg/ml concentration of type I collagen at 37°C for 1 h. In this condition, cells need proteolysis to invade the gel according to the study of Wolf et al. (2013). Cells were seeded on top of the collagen gel and fixed after 1 h and 3 d. We confirmed that DDR1 depletion remained constant over the studied time frame (Fig. 8 B). Using quantitative confocal z-stack analysis (Fig. 8 C), we found that DDR1 depletion blocked the cell’s ability to invade the collagen gel. In the control condition, approximately half of the cells were able to invade the gel (Fig. 8 D). In contrast, DDR1 down-regulation abolished cell invasion. Confocal analysis showed that after 3 d, most of the DDR1-silenced cells remained stacked at the gel surface and did not enter into the gel (Fig. 8 E). To control that cells used a proteolysis-dependent mode of migration to invade the collagen gel, we used an MMP inhibitor, GM6001. We found that GM6001 totally blocked the cell’s ability to penetrate the gel, demonstrating the MMP involvement in this process (Fig. S5, A and B). Consequently, GM6001 treatment abolished the cleaved collagen signal observed in the control condition (Fig. S5 C). As PP2 and nilotinib treatments did not inhibit type I collagen degradation in 2D (Fig. S4), we also checked for cell invasion in 3D in these conditions. PP2 treatment did not have an impact on cell invasion, whereas nilotinib treatment only induced a slight decrease in the cell capacity to invade the collagen gel (Fig. S5, A and B). These data demonstrate the crucial involvement of DDR1 in type I collagen matrix invasion, which we found to be MMP dependent and Src independent.

This study has revealed the link between the collagen receptor DDR1 associated with the development of metastasis, and invadosomes, which are protrusive F-actin structures used by tumor cells to degrade the ECM and promote invasion. Herein, we confirmed our previous findings showing the importance of type I collagen fibrils as powerful invadosome inducers (Juin et al., 2012) and extend them to cancer cells. Most cell types are able to form linear invadosomes, including endothelial cells, fibroblasts, cancer cells, Src-transformed cells (Juin et al., 2012), or, as shown by another group, megakaryocytes (Schachtner et al., 2013). We demonstrated that the simple contact of cancer cells with type I collagen fibrils can promote formation of linear invadosomes and consequently activate their capacity to degrade the ECM. In the case of cancer cells constitutively exhibiting invadopodia, type I collagen fibrils induced their reorganization into linear invadosomes, increased the percentage of cells presenting linear invadosomes as compared with classical invadopodia, and also increased the capacity of the cells to degrade the ECM. Owing to their capacity to localize the degradation machinery along fibrils, linear invadosomes could be implicated in the proteolytic breakdown of the ECM, which favors invasive migration, either in physiological conditions such as angiogenesis, or in pathological situations such as cancer. Thus, there is a great interest in developing our understanding of the molecular pathways required for the formation and function of linear invadosomes.

Although integrins are required for the formation and activity of classical invadosomes (Destaing et al., 2010; Beaty et al., 2013), we found that they are not necessary for linear invadosome formation (Juin et al., 2012), thus raising the question of the identity of the collagen receptor responsible for their formation. Four major classes of vertebrate transmembrane receptors are known to interact directly with the native collagen triple helix: collagen-binding β1 integrins, DDRs, glycoprotein VI (GPVI), and leukocyte-associated immunoglobulin-like receptor-1 (LAIR-1; Hidalgo-Carcedo et al., 2011; Leitinger, 2011). Our own data eliminated β1 integrins (Juin et al., 2012). Because GPVI is present only on platelets and LAIR-1 on leukocytes, we turned our attention to the DDRs, which are ubiquitously expressed.

We found that DDR1 colocalized within linear invadosomes, and, using a RNAi approach, found that it was involved in their formation and in their ability to degrade type I collagen fibrils. These results establish DDR1 as the collagen I receptor required for linear invadosome formation (Fig. 9) and raise the question of DDR1’s role in classical invadosomes. We showed that DDR1 depletion decreased classical invadosome formation and activity. Importantly, the effect of DDR1 on classical invadosomes was observed on gelatin, which is not known as a DDR1 substrate. Accordingly, we were unable to colocalize DDR1 with classical invadosomes. However, these data suggest an involvement of DDR1 in classical invadopodia. Because DDR1 controls integrin activation (Xu et al., 2012), it could modulate invadosome formation and activation directly or indirectly. Moreover, DDR1 and integrins present common signaling pathways, which suggests a cooperative action to form classical invadosomes (Shintani et al., 2008).

Because DDR1 is an RTK, we investigated the role of its kinase activity in invadosome formation. We demonstrated, using nilotinib as an inhibitor as well as blocking antibodies, that DDR1 kinase activity is not necessary for the formation of linear invadosomes. This is in agreement with our previous observations. Indeed, we found that linear invadosomes already appear within 10 min of cell seeding on type I collagen, a kinetic that is not compatible with DDR autophosphorylation, a slow process requiring >30 min before being detectable (Shrivastava et al., 1997; Vogel et al., 1997; Juin et al., 2012). Interestingly, it was previously shown that the role of DDR1 in promoting collective migration was also independent of its kinase activity (Hidalgo-Carcedo et al., 2011). In our hands, the lack of requirement for c-Src activity, demonstrated with both a pharmacological antagonist and the use of SYF cells, is also in line with these findings because it was shown that c-Src was necessary for the full DDR phosphorylation (Dejmek et al., 2003; Yang et al., 2005). In addition, although c-Src has been historically associated with invadosomes, our data show that collagen I–induced invadosomes are independent of c-Src activity. Thus, linear invadosomes are the first described c-Src kinase–independent invadosomes.

How then does DDR1 signal for invadosome formation? As the RhoGTPases RhoA, Rac1, and Cdc42 have been largely involved in invadosome formation and organization, we investigated their role in DDR1-dependent linear invadosome formation. We clearly showed that only Cdc42 is involved. This is supported by the drastic effect of Cdc42 silencing, the blockage of collagen I–induced Cdc42 activation in cells transfected with DDR1 siRNAs, and the localization of the active form of Cdc42 (GFP-V12Cdc42 protein and Cdc42 biosensor) at linear invadosomes. Conversely, the GFP-N17Cdc42 dominant-negative form decreased the cell’s ability to form linear invadosomes on collagen and did not colocalize with linear invadosomes. Most invadosome models are controlled and regulated by several Rho-GTPases. Our results are thus important, as only very few models have been described so far in which only Cdc42 and not RhoA or Rac1 are implicated in invadosome formation. This specificity of linear invadosomes, together with their restricted molecular composition, which we reported previously (Juin et al., 2012), supports the idea that collagen-induced invadosomes correspond to a minimal form of invadosomes (Di Martino et al., 2014). Intriguingly, another study showed that overexpression of tagged forms of DDR1 in MDCK cells decreased Cdc42 activation by collagen (Yeh et al., 2009). The reason for this discrepancy is unclear, but it may be relevant to explain why MDCK cells are unable to form linear invadosomes and to degrade the ECM upon collagen stimulation (unpublished data).

Altogether, our data demonstrated that collagen I–induced invadosomes rely on a DDR1, Cdc42-dependent pathway (Fig. 9). We further identified Tuba as the major Cdc42GEF involved in linear invadosome formation. Several other Cdc42GEFs such as FGD1 or Vav1 were shown to be involved in invadopodia formation (Ayala et al., 2009; Razidlo et al., 2014), but this is the first demonstration of the involvement of Tuba. Moreover, Tuba colocalized with DDR1 into linear invadosomes but not with classical invadosomes, which allowed us to identify Tuba in addition to DDR1 as a specific marker of linear invadosomes. Tuba is a 177-kD protein containing four SH3 domains in its N terminus: a central GEF domain, followed by a BAR domain and two SH3 domains in the C terminus (Cestra et al., 2005; Xu et al., 2012). The involvement of Tuba is in line with the involvement of N-WASP in linear invadosomes, as Tuba was shown to be involved in N-WASP–dependent cytoskeletal rearrangements (Salazar et al., 2003; Kovacs et al., 2006). Though we describe here the involvement of Tuba in MDA-MB-231 and in A549 cells, it is now clear that each cancer cell expresses its own pattern of GEFs among the 70 RhoGEFs in the human genome, (Cook et al., 2014; Razidlo et al., 2014), which suggests that other GEFs may also be involved in linear invadosome formation according to the cell type. In addition, the link between DDR1 and Tuba is probably not direct, as we were not able to show an interaction between both proteins. Indeed, other DDR1-interacting proteins could also be involved. For instance, on one hand several studies have identified DDR1 partners previously linked to the invadosome machinery (Murphy and Courtneidge, 2011), like PYK2 (Shintani et al., 2008), Nck2 (Koo et al., 2006), and PI3K (Dejmek et al., 2003), whereas others found DDR1-interacting partners that bind DDR1 regardless of its phosphorylation status, like Syk (Dejmek et al., 2005), E-cadherin (Hidalgo-Carcedo et al., 2011), and the Par3/Par6 cell polarity proteins (Hidalgo-Carcedo et al., 2011). Further studies will tell if these proteins are involved in the role of DDR1 on linear invadosome formation and activation.

In our study, we confirmed the role of DDR1 in cell invasion. However, interestingly, we established that inhibition of Src with PP2 or of DDR1-kinase activity with nilotinib did not affect drastically type I collagen degradation and cell invasion. This differs from the findings from other studies showing an impact of PP2 treatment on cell invasion (Angers-Loustau et al., 2004). This point could be explained by the conditions used to perform these assays such as the use of a non–type I collagen matrix, the collagen source, or the absence of serum starvation. It is clear that the kinase activity of the receptor is crucial for DDR1 signaling that promotes, for example, cell proliferation. But we demonstrated here that cells can sense type I fibrils and start a degradation process independent of its kinase activity and in the absence of a requirement for c-Src.

Thus, upon contact with type I fibers, DDR1 is able to recruit the actin machinery associated with a strong matrix degradation activity. Although the proteolytic mechanisms used by linear invadosomes are still being investigated, a recent study has shown that the Scar homologue (WASH) and the exocyst complex are involved in delivering MT1-MMP–positive late endosomes focally to linear invadosomes (Monteiro et al., 2013). It is well known that MMPs, such as pro-MMP2, can be activated by the culture of cells on fibrillar collagen I (Azzam and Thompson, 1992; Ruangpanit et al., 2001) in an MT1-MMP–dependent manner (Takino et al., 2004). Thus, we propose that DDR1 is the sensor used by tumor cells to interact with fibrillar collagen I, leading to the organization of invadosomes that concentrate the proteolytic machinery of the cells to facilitate invasiveness. Because of its capacities to stimulate cell invasion and its overexpression in different cancers, DDR1 should be a good target for the prevention of metastasis.

Nilotinib, anti-Tks5 (rabbit, M-300), anti-DDR1 (rabbit, C-20), anti-GAPDH (mouse, FL-335), anti-myc (mouse, 9E10), and anti-RhoA (mouse, 26C4) antibodies were purchased from Santa Cruz Biotechnology, Inc. Mouse anti-DDR1 monoclonal antibodies (1F7, 1F10, 3E3, and 5D5) were provided by B. Leitinger (National Heart and Lung Institute, Imperial College London, London, England, UK) and produced as described previously (Carafoli et al., 2012). Mouse anti–β-actin (clone AC-15), anti-tubulin (T6074), and anti-Flag (clone M2) antibodies were purchased from Sigma-Aldrich. We also used anti-DDR1 (rabbit, D1G6) from Cell Signaling Technology. Anti-cortactin (mouse, p80/85), anti-Rac1 (mouse, 23A8), anti-Src CT (rabbit, clone NL19), anti–phospho-Src (mouse, Tyr416), and anti-phosphotyrosine (mouse, 4G10) antibodies and GM6001 were purchased from EMD Millipore. Anti-Cdc42 (mouse, clone 44) antibody was purchased from BD. PP2 was from Abcam. Anti–collagen type I cleavage site antibody (rabbit, Col1 3/4 short C) was purchased from immunoGlobe. Rabbit polyclonal anti-Tuba antibody was provided by P. De Camilli (Yale University, New Haven, CT; Salazar et al., 2003; Cestra et al., 2005). Secondary antibodies FluoProbes 488, 547H, and 647H anti–rabbit and anti–mouse antibodies were purchased from Interchim. F-actin was stained with Phalloidin-FluoProbes 647, 547H, 488, or 405 (Interchim). Hoechst 34580 (Invitrogen) was used to stain nuclei. To visualize the collagen I network, we labeled 0.4 mg/ml fibrillar collagen I with 10 µg/ml Alexa Fluor 546 or 647 carboxylic acid succinimidyl ester (Invitrogen).

pDDR1-Flag construct was provided by M. Bendeck (Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Ontario, Canada). Human DDR1 full length fused to flag sequence was cloned in pcDNA3.1/Zeo (−) using EcoRI–BamHI cloning sites, placed under the control of a cytomegalovirus (CMV) promoter. pEGFP-Cdc42 WT, pEGFP-V12Cdc42, myc-V12-Cdc42, and pEGFP-N17Cdc42 have been described previously (Moreau et al., 2003). GFP-Cdc42–tagged constructs were cloned in pEGFP-C1 using BglII–KpnI cloning sites and placed under the control of a CMV promoter. pLVX-EF1α-DDR1-acGFP was constructed from pcDNA 3.1/zeo–DDR1–myc (provided by G.D. Longmore, ICCE Institute, Washington University School of Medicine, St. Louis, MO; Zhang et al., 2013) by subcloning DDR1 full length under the control of an EF1α promoter into pLVX-EF1α–acGFP-N1 (Takara Bio Inc.) using AfeI and BamHI restriction sites. pRaichu-Cdc42 was provided by M. Matsuda (Graduate School of Biostudies, Kyoto University, Kyoto, Japan; Aoki and Matsuda, 2009). pRaichu-Cdc42 was derived from the pCAGGS eukaryotic expression vector and encoded a chimeric protein, Raichu-Cdc42. pLifeact-mRuby lentiviral vector was obtained by subcloning Lifeact-mRuby from pmRFPRuby-Lifeact, provided by R. Wedlich-Soeldner (Max Planck Institute of Biochemistry, Martinsried, Germany; Riedl et al., 2010), using BglII and SpeI restriction sites in pRRLsin-MND-MCS-WPRE lentiviral plasmid, placed under the control of an MND promoter. pTks5-GFP was provided by S.A. Courtneidge (Sanford-Burnham Medical Research Institute, La Jolla, CA). Human Tks5 was cloned in pcDNA3.1/Zeo (−) using XhoI–Kpn I cloning sites, placed under the control of a CMV promoter.

MDA-MB-231 cells (human breast cancer cell line) were from American Type Culture Collection and were maintained in L-15 medium and Glutamax-I (Invitrogen) supplemented with 10% fetal calf serum and 100 U/ml penicillin–streptomycin (Invitrogen). A549 (human lung adenocarcinoma cell line) were from Sigma-Aldrich, provided by F. Delom (INSERM, Bordeaux, France), and SYF and SYF c-Src fibroblasts were from A. Wiedmann (INRA Val de Loire, Tours, France; Klinghoffer et al., 1999). Huh6 cells (human hepatoblastoma cell line) were provided by C. Perret (Cochin Institute, Paris, France). A549 and SYF cell lines were cultured in Dulbecco’s modified Eagle’s medium with 4.5 g/liter glucose Glutamax-I (Invitrogen) supplemented with 10% fetal calf serum (Pan Biotech GmbH) and 100 U/ml penicillin–streptomycin (Invitrogen). The Huh6 cell line was cultured in Dulbecco’s modified Eagle’s medium with 1 g/liter glucose Glutamax-I (Invitrogen) supplemented as with A549 cells.

SiRNA oligonucleotides (100 nM) were transfected with Lipofectamine RNAiMax (Invitrogen) according to the manufacturer’s instructions. The siRNA sequences for human DDR1 were as follows. DDR1 #1, 5′-GAAUGUCGCUUCCGGCGUGUU-3′; DDR1 #2, 5′-GAGCGUCUGUCUGCGGGUAUU-3′ according to published sequences (Hidalgo-Carcedo et al., 2011). DDR1 #3 (SI05130706; QIAGEN) targets the 3′ UTR of DDR1 mRNA. The antisense strand siRNA was targeted against GTPases using the 21-nucleotide sequences 5′-AAGAAGTCAGCATTTCTGTC-3′ for hRhoA #1, 5′-AAGTTCTTAATTTGCTTTTCC-3′ for hRac1 #1, and 5′-AAGATAACTCACCACTGTCCA-3′ for hCdc42 #1 according to published sequences (Deroanne et al., 2003); 5′-AAGGAGATTGGTGCTGTAAAA-3′ for hRac1 #2 as previously published (Grise et al., 2012); and 5′-AGGTGGATGGAAAGCAGGTA-3′ for hRhoA #2 and 5′-GAGATGACCCCTCTACTATTG-3′ for hCdc42 #2. A control siRNA targeted against luciferase (CT) 5′-CGTACGCGGAATACTTCGA-3′ was purchased from Eurofins MWG Operon, Inc. siRNAs used for the GEF screen were purchased from QIAGEN and are referenced in Table S1. The second siRNA sequence for human Tuba was as follows: Tuba #2, 5′-GAGCUUGAGGGAACAUACAAGAUUU-3′, as previously published (Rajabian et al., 2009). For transient transfection of MDA-MB-231 cells, the Amaxa Nucleofector kit V (Amaxa Inc.), JET PRIME (PolyPlus; Ozyme), or Lipofectamine 2000 (Invitrogen) was used according to the manufacturer’s instructions. 5 µg of DNA was added per well of a six-well plate. Cells were allowed to grow 24 h after transfection before use.

For the rescue experiment, cells were transfected with siRNA DDR1 #3 to silence endogenous DDR1 as described in the same paragraph. 2 d after, siRNA DDR1-expressing cells were infected with lentivirus particles expressing DDR1-GFP at a multiplicity of infection of 2.5 and selected using puromycin antibiotic at a concentration of 1 µg/ml. The rescue was observed comparing the proportion of cells able to form linear invadosomes in cells receiving siRNA control to siRNA DDR1 only and siRNA DDR1 + DDR1-GFP. To generate the MDA-MB-231 cell line stably expressing Lifeact-mRuby, cells were transduced at a multiplicity of infection of 10.

To detect GTP-active bound Cdc42, MDA-MB-231 transfected with CT or DDR1 #1 siRNA were cultured for 2 h on type I collagen fibrils or overnight on plastic. 50 µg of protein was subjected using the G-LISA Cdc42 Activation Assay Biochem kit (Cytoskeleton, Inc.) according to the manufacturer’s instructions.

Coverslips were coated with Oregon green gelatin (Invitrogen), fixed with 0.5% glutaraldehyde (Electron Microscopy Sciences), and washed three times with PBS (Invitrogen). Cells were seeded on coated coverslips and incubated overnight before fixation and staining.

Cells were fixed with 4% paraformaldehyde, pH 7.2, for 10 min, permeabilized with 0.2% Triton X-100 for 10 min, and incubated with various antibodies. Cells were imaged with an SP5 confocal microscope (Leica) using a 63×/NA 1.4 Plan Neofluor objective lens. To prevent contamination between fluorochromes, each channel was imaged sequentially using the multitrack recording module before merging.

Collagen polymerization and linear invadosome quantifications were made as described previously (Juin et al., 2012). In brief, collagen was diluted at 0.5 mg/ml in DPBS 1×, then polymerized for 4 h at 37°C before cell seeding. Cells were seeded for 4 h on collagen before fixation. Confocal images of isolated cells were obtained using an SP5 confocal microscope (Leica) using a 63×/NA 1.4 Plan NeoFluor objective lens. Cell surface area was measured upon phalloidin staining, and Tks5 staining was used as a marker for linear invadosomes. We used a custom macro (macros 1–3, available as supplemental files) with ImageJ software (W. Rasband, National Institutes of Health) that allowed measurement of all required parameters of linear invadosomes: number, size (using the Feret diameter, the longest distance between any two points), and area (A.U.). Collagen degradation using the anti–collagen type I cleavage site antibody (rabbit, Col1 3/4 short C) was done using a custom macro (see supplemental files) with ImageJ.

The SHG imaging system consists of a confocal TCS SP2 scanning head (Leica) mounted on an inverted microscope (DMIRE2; Leica) and equipped with a MAITAI femtosecond laser (Spectra Physics). A 10× dry objective lens (NA 0.4; Leica) was used for applying an 820-nm excitation to the sample. The SHG signal was collected in the forward direction using the condenser (S1, NA = 0.9–1.4; Leica), and the two-photon-excited fluorescence (TPEF) was epi-collected in the backward direction. IRSP 715 band-pass and 410-nm infrared (IR) filters (10-nm full width at half-maximum [FWHM]) were placed before the photomultiplier tube.

All image analysis was performed with the ImageJ software, using a custom macro (see supplemental files). For collagen quantification, SHG images were thresholded and the mean pixel numbers corresponding to collagen were converted to square micrometers, multiplying by a factor of 8.583 and taking into account the point spread function of the objective. For cell counts, the TPEF images were thresholded and watersheded before performing the “Analyze Particles” function.

The invasion assay was adapted from Lopez et al. (2005). 1 mg/ml type I rat tail collagen (BD) was used. Costar Transwell inserts (8-µm pore; Corning) and gels were allowed to polymerize at 37°C for 1 h. Collagen gel matrices were then hydrated with DMEM (Life Technologies) supplemented with 50% fetal bovine serum (Biomedia) for 4 h. Cells were washed twice in serum-free medium, trypsinized, counted, placed in the upper chamber of the Transwell insert, and allowed to invade for the indicated time points. After invasion, the Transwell inserts were removed from the plate and the quantity of invading cells into the gel matrix was determined by F-actin staining.

Cells were lysed in radio-immunoprecipitation assay buffer (25 mM Tris HCl, pH 7.5, 150 mM NaCl, 1% IGEPAL, 1% sodium deoxycholate, and 0.1% SDS), sonicated, incubated at 95°C for 5 min, and loaded onto a 10% or 12% SDS-PAGE gel. Proteins were blotted onto a nitrocellulose membrane (Sigma-Aldrich), blocked with 5% bovine serum albumin, and probed with primary antibody overnight. Membranes were then washed and incubated with the corresponding secondary antibody, and signals were acquired and quantified with the Odyssey system (LI-COR Biosciences).

Raichu-Cdc42 biosensor was used for FRET imaging to measure Cdc42 activity (Aoki and Matsuda, 2009). MDA-MB-231 cells transfected with the biosensor were plated in µ-Dish 35 mm, high glass bottom plates (Ibidi) coated with type I collagen. 2 h later, living cells were imaged using an inverted microscope (TE Eclipse; Nikon) equipped with a motorized heated and CO₂-regulated incubator. Images were taken using a Nikon 100× Plan-Apochromat VC 1.4 oil objective lens and captured with an EM charge-coupled device (CCD) camera (C9100-13, ImagEM; Hamamatsu Photonics) controlled by the MetaMorph 7.0 software. A ratio image of YPF/CFP was created to represent FRET efficiency, which correlated with the activities of the G proteins. Pseudocolored ratio images were generated from images of CFP and FRET channels, as described previously (Hodgson et al., 2006). For quantification, the frequency of colocalization between active probe and collagen I fibrils was observed and presented as the number of cells presenting active probe on collagen I fibrils per number of cells observed (two independent experiments).

MDA-MB-231 cells stably expressing Lifeact-mRuby were transfected with pTks5-GFP construction. The next day after transfection, cells were plated on 14-mm glass-bottom dishes, No. 1.5 thickness (MatTek) before being coated with 633 fluorescent collagen type I. Cells were imaged 2 h after seeding, with or without PP2 in DMEM 4.5 g/liter glucose, Hepes, no phenol red, and 10% fetal calf serum at 37°C without CO₂. A picture was taken every 4 min for 1 h with a confocal microscope (SP5; Leica).

For colocalization quantification, we used Co-localization Finder Version 1.2 from C. Laummonerie and J. Mutterer (Institut de Biologie Moleculaire des Plantes, Strasbourg, France) on ImageJ version 1.48. Results were presented as the mean of 10 fields quantified.

Data were reported as the mean ± SEM of at least three experiments. Statistical significance (P < 0.05 or less) was determined using a paired t test or analysis of variance (ANOVA) as appropriate and performed with GraphPad Prism software (GraphPad Software).

Fig. S1 demonstrates the impact of type I collagen fibrils on cell degradation activity and confirms the DDR1 localization with linear invadosomes. Fig. S2 reveals DDR1 involvement in linear invadosome formation in A549 cells and shows a rescue experiment on cells depleted for DDR1 and infected with DDR1-GFP. Fig. S3 describes the DDR1 role in classical invadosome formation and activity. Figs. S4 and S5 demonstrate that the role of DDR1 in collagen degradation and cell invasion is MMP dependent but Src independent. Table S1 contains the list of siRNAs tested to screen for GEFs involved in linear invadosome formation. Video 1 shows the dynamics of Tks5-GFP and F-actin on MDA-MB-231 cells seeded on labeled type I collagen fibrils. ImageJ macros 1 and 2 were used to determine the number and the size of linear invadosomes per cell and collagen degradation. ImageJ macro3 was used to quantify SHG signal.

We are grateful to Dr. A. Wiedmann and Dr. C. Perret for the SYF and Huh6 cell lines, respectively. We thank Drs. S. Courtneidge, P. De Camilli, M. Bendeck, R. Wedlich-Soeldner, G. Longmore, and M. Matsuda for constructs. We thank the Bordeaux Imaging Center for help with fluorescence quantification (BIC) and P. Chavrier and A. Blangy for helpful discussions.

A. Juin is supported by a predoctoral fellowship from the Ministère de l’Enseignement Supérieur et de la Recherche. J. Di Martino is supported by a PhD fellowship from Institut National de la Santé et de la Recherche Médicale/Région Aquitaine. L. Paysan is supported by a PhD fellowship from Région Aquitaine. E. Henriet is supported by a PhD from the Ministère de l’Enseignement Supérieur et de la Recherche. This work was supported by grants from ANR-13-JJC-JSV1-0005, l’AFEF (Association Française pour l’Etude du Foie), La Ligue Nationale contre le Cancer, and Association pour la Recherche sur le Cancer. V. Moreau and J. Rosenbaum are supported by funding from Equipe Labellisée Ligue Nationale contre le Cancer 2011 and Grant Institut National du Cancer - Direction Générale de l’Offre de Soins - Institut National de la Santé et de la Recherche Médicale 6046.

The authors declare no competing financial interests.