VEGF-A stimulates podosome-mediated collagen-IV proteolysis in microvascular endothelial cells

Vascular endothelial cells line the entire circulatory system from the heart to the smallest capillaries and play a key role in the maintenance and remodelling of the vascular system. The physiology of these cells depends heavily on cytoskeleton rearrangements that are controlled by soluble factors, matrix proteins, cell–cell interactions and mechanical forces. Actin-based cytoskeletal structures called podosomes have been described in endothelial cells (Billottet et al., 2008) and the elucidation of their precise role in endothelial biology thus represents a major challenge. So far, most of our knowledge on podosomes comes from studies performed in macrophages, osteoclasts or immature dendritic cells where they are most commonly found. They are described as dynamic cell–matrix contact structures that combine several key abilities, including adhesion, matrix degradation and mechanosensing (Linder and Kopp, 2005; Linder and Wiesner, 2015). Although the triggering signal for podosome initiation is not known yet, Src family kinases have a pivotal role in their formation (Murphy and Courtneidge, 2011). Structurally, podosomes comprise a core of F-actin associated with actin regulatory proteins such as Arp2/3, WASp or N-WASp (also known as WAS and WASL, respectively), cortactin, dynamin, gelsolin and cofilin, which contribute to their dynamic behaviour, and the scaffolding protein and Src substrate Tks5 (also known as SH3PXD2A), which constitutes a reliable podosome marker. Around the central core, integrins, adaptors and signalling proteins, including FAK (also known as PTK2), vinculin, talin and paxillin, form an adhesive ring domain. Recent studies have shown that acto-myosin filaments radiating from the dense F-actin cores interconnect podosomes and enable their collective behaviour (Proag et al., 2015; van den Dries et al., 2013).

In contrast to cells of the myelomonocytic lineage, where podosomes develop evenly in the first steps of integrin-mediated cell spreading, podosome formation is an inducible phenomenon for endothelial cells. Most studies have been performed with human umbilical vein endothelial cells (HUVECs) (Osiak et al., 2005) exposed to phorbol esters (Guegan et al., 2008; Seano et al., 2014; Tatin et al., 2006) or with aortic endothelial cells exposed to TGFβ (Curado et al., 2014; Varon et al., 2006). In these situations, podosomes arise in clusters or ring-shaped superstructures that are often called rosettes. However, recent studies have shown that matrix proteins are also involved in podosome induction, architecture and arrangement. In the absence of any specific cytokine (or pharmacological) stimulation, a collagen-I (Col-I) substrate elicits the formation of atypical linear podosomes aligning along collagen fibrils (Juin et al., 2013), whereas in a confined environment, fibronectin induces the formation of either isolated entities or podosome rosettes (Spuul et al., 2015). Such diversity points to a multiplicity of regulatory mechanisms for podosome formation in endothelial cells and raises the question of whether these podosomes are functionally equivalent.

Podosome-forming cells share the common feature of traveling across tissues, suggesting that podosome lytic enzymes overcome extracellular matrix (ECM) barriers. Among them, the transmembrane matrix metalloprotease MT1-MMP (also known as MMP14) is found at podosomes in all types of cells forming these structures and plays a prominent role (Linder, 2007). However, in vitro, many studies are limited to the use of the traditional fluorescent gelatin invasion assay that highlights the matrix-degrading capabilities of podosomes, showing only the gelatinolytic activity on the denatured Col-I matrix. Thus, it remains unclear whether podosomes are used by cells to travel across all types of ECM encountered during their transtissular migration or if there are cell type and matrix protein specificities for their induction and function.

In the context of angiogenesis, degradation of the ECM scaffold enables the vasculature to be reshaped (Davis and Senger, 2005). Endothelial cells within existing blood vessels breach the basement membrane [whose main components are collagen-IV (Col-IV) and laminin], which forms a sleeve around the capillary, to escape from the parent vessel and migrate within the interstitial matrix (mainly composed of Col-I and fibronectin), which is found between cells. Neovessel formation is completely abroagated in MT1-MMP-null mice (Chun et al., 2004). To explore the role of podosomes during the angiogenesis invasive step, we chose to use microvascular cells (human microvascular endothelial cells, HMVECs), which form the sprouting capillaries, and examine their response to the canonical angiogenic factor VEGF-A. We also used various matrices to address their substrate specificities and characterised podosome proteolytic capacities under each condition. We found that a small fraction of the HMVEC population assembled podosomes upon adhesion regardless of the matrix. When cells were exposed to VEGF-A, podosome formation was stimulated but this response occurred only on basement membrane proteins. Finally, we found that Col-IV, but not laminin, stimulated the formation of matrix-degrading podosomes upon VEGF-A stimulation. Collectively, these results show that podosomes display cell type and matrix protein specificities for their induction and function.

We used primary HMVECs for these studies as physiological angiogenesis involves endothelial cells from capillaries. HMVECs are large cells with prominent stress fibres (Fig. 1A). At the lowest confocal plan, we found F-actin colocalised with cortactin, Tks5 and other podosomal proteins (Linder and Aepfelbacher, 2003) (Fig. S1) in punctate structures forming circular clusters or rings. These were identified as genuine podosome rosettes by performing the gelatin degradation assay (Fig. 1B) and by the detection of MT1-MMP and phosphorylated Src (active Src) (Tatin et al., 2006; Varon et al., 2006) markers (Fig. 1C). We also visualised the typical polygonal arrangement of vinculin around F-actin and cortactin cores (Walde et al., 2014), and F-actin interconnecting cables by stimulated emission depletion (STED) microscopy (Labernadie et al., 2010; Luxenburg et al., 2007; van den Dries et al., 2013) (Fig. 1D). In these unstimulated cultures, the proportion of podosome-forming cells never exceeded 5.5% of the total cell population.

We next examined whether the number of cells presenting podosome rosettes was altered upon angiogenic stimulation. In HUVECs, VEGF stimulates podosome formation with a peak response observed 5 min after the addition of the cytokine and a return to baseline values 1 h later (Osiak et al., 2005). A prompt dose-dependent increase in podosome-forming cells also occurred in HMVECs exposed to soluble VEGF-A, but this response (20 min) reached its maximum at 24 h (Fig. 1E–G). In the gelatin assay, matrix degradation became detectable after 1 h and steadily increased for the first 3 h (Fig. 1H–J). However, after that point, the degradation index of VEGF-A-treated cells increased dramatically and both the number and size, as well as the intensity of degraded areas (loss of matrix fluorescence) rose significantly (Fig. 1K,L). The pattern of the degraded areas in the fluorescent matrix suggested several possibilities in terms of cell motility: a reduced motility of VEGF-treated cells that would stay still and degrade larger holes in the matrix before crawling away, an increased motility of these cells, leaving behind more numerous holes over a wide area, or a combination of both. We therefore first analysed cells individually, then classified them into two groups, matrix-degrading and non-degrading cells, that we followed over a period of three consecutive hours in the second phase of the 24-h response to VEGF-A (i.e. at least 3 h after adhesion onset, Fig. 1H,I). Regarding matrix-degrading cells, we found that VEGF-A-treated cells had a higher velocity and higher displacement as compared to untreated cells. For non-degrading cells, velocity and displacement were not significantly different between the two groups (Fig. S2). We conclude that the matrix degradation pattern observed for VEGF-treated cells results from both a higher degradation potential (Fig. 1L) and a multiplication of degradation areas associated with their higher motility.

Because VEGF-A stimulates cell proliferation in sparse cultures, podosome induction could be somehow linked to the mitotic state. We therefore compared podosome formation and matrix degradation in sparse and confluent cultures. In addition, this exploration had the potential to provide some insights into the in vivo situation where endothelial cells form a continuous lining in the microvessel. By itself, confluence favoured podosome rosette formation but this response was insensitive to VEGF-neutralising antibodies, thus ruling out an autocrine effect (Fig. 2A,B). Collectively, these results indicate that VEGF-A is a regulator of podosome formation and function for HMVECs. We distinguished two classes of podosomes, constitutive podosomes with low degrading capacities whose formation was stimulated shortly after VEGF-A addition, and induced podosomes that had acquired a higher degrading potential upon prolonged VEGF-A exposure.

As none of the tested markers (Fig. S1) discriminated between the two types of podosome rosettes, we examined the cell subpopulations. The sialomucin CD34 is a surface adhesion glyocoprotein that has been shown to be expressed in a small subset of cultured endothelial cells (Siemerink et al., 2012). No direct correlation could be established between these cells and podosome-forming cells (Fig. 2C). However, confluence failed to stimulate podosome formation in either untreated or VEGF-A-treated CD34-positive cells (Fig. 2C), suggesting that this signal was already part of their phenotype. In unstimulated cultures, CD34-positive cells were morphologically similar to CD34-negative cells (Fig. 2D). In contrast, VEGF-A-stimulated CD34-positive cells were polarised and presented filopodia (Fig. 2D).

MT1-MMP and MMP2 localise at podosomes in HUVECs (Guegan et al., 2008; Osiak et al., 2005; Tatin et al., 2006). Besides its important contribution to ECM cleavage, MT1-MMP also acts as a membrane-anchored activator, processing inactive pro-MMP2 into active MMP2 at the cell surface. We therefore analysed the expression, localisation and activity of MT1-MMP after 24 h of VEGF-A stimulation in HMVECs. VEGF-A treatment did not alter MT1-MMP expression levels in cells seeded on either uncoated or gelatin-coated substrata (Fig. 3A,B). As MT1-MMP is known to be actively transported to the cell surface of podosome sites (Wiesner et al., 2010), we also examined MT1-MMP cell surface exposure in control and stimulated cells seeded on gelatin. The analysis of biotinylated cell surface proteins revealed greater exposure of MT1-MMP at the surface of VEGF-A-stimulated HMVECs than in untreated cells (Fig. 3C). Again, similar results were obtained when cells were seeded on uncoated substrata (Fig. 3D). MT1-MMP staining of non-permeabilised cells was brighter at podosome rosettes in VEGF-A-treated cells than in controls (Fig. 3E–G). Taken together, these results indicate that VEGF-A-induces a qualitative change in podosome rosettes over a 24-h period and that increased MT1-MMP cell surface exposure discriminates between VEGF-A-induced podosomes and constitutively formed podosomes.

MT1-MMP regulation depends on multiple intracellular trafficking events (Linder and Wiesner, 2015). In HUVECs, p190RhoGAP-B (also known as ARHGAP5) regulates MT1-MMP cell surface presentation (Guegan et al., 2008) but its regulation by VEGF-A has not been explored. We observed more recruitment of p190RhoGAP-B to VEGF-A-induced podosomes than constitutive podosomes in HMVECs, but no change for p190RhoGAP-A (also known as ARHGAP35) or ARAP3, a third RhoGAP component (Guegan et al., 2008; Yu et al., 2013) (Fig. 4A). The cellular expression levels of the three proteins remained unaltered upon VEGF-A stimulation (Fig. 4B). Silencing p190RhoGAP-B had no effect on constitutive podosomes but it selectively suppressed the VEGF-A-induced podosome rosettes and associated gelatinolytic capacities (Fig. 4C–E). These results indicate that p190RhoGAP-B could also be involved in podosome induction. To get some insights into how p190Rho-GAP-B regulates the podosome increase in response to VEGF-A, we focused our analysis on its function. Because p190RhoGAP-B exhibits its catalytic activity preferentially towards RhoA, we evaluated RhoA activity at podosomes using a conformation-specific monoclonal antibody that recognises RhoA-GTP but not RhoA-GDP. VEGF-A stimulation was associated with an increase of RhoA-GTP at podosomes and depletion of p190RhoGAP-B prevented this response (Fig. 4F). Taken together, these data indicate that podosome induction in response to VEGF-A stimulation is associated with increased MT1-MMP exposure at podosome sites through a pathway involving p190RhoGAP-B, which contributes to the spatial regulation of RhoA activity at podosomes.

We next examined the contribution of genuine ECM proteins to VEGF-A-induced podosome formation. In vitro, HMVECs produced Col-IV, laminin, Col-I and fibronectin, and VEGF-A stimulated the expression of Col-IV (Fig. 5A). When cells were seeded on these proteins, a small fraction of the HMVEC population assembled podosomes upon adhesion regardless of the matrix (Fig. S3). We observed both the short-term and long-term VEGF-A podosome responses on the basement membrane proteins Col-IV, laminin and Matrigel (a mixture of ECM components containing predominantly laminin and to a lesser extent Col-IV), but no induction on fibronectin or Col-I (Fig. 5B). Accordingly, matrix degradation was stimulated on Col-IV but not on fibronectin or Col-I (Fig. 5C–E). As expected, Col-I also elicited the formation of collagenolytic linear podosomes along fibrils (Juin et al., 2012) but VEGF-A did not increase the number of cells presenting them (Fig. 5F,G). In agreement with previous findings in an ex vivo set-up (Rottiers et al., 2009), HMVEC podosome rosettes were unable to degrade laminin. Laminin–FITC remained a suitable substrate for other types of podosomes, such as those formed by MDA-MB-231 cells (Fig. S4A). Thus, Col-IV was found to be the only matrix protein that stimulates basement membrane degradation upon treatment of cells with VEGF-A.

We then verified MT1-MMP exposure at podosome sites upon VEGF-A stimulation in cells plated on Col-IV versus laminin substrata. On Col-IV, a quantitative analysis of MT1-MMP cell surface exposure showed that MT1-MMP staining of non-permeabilised cells was consistently more intense with VEGF-A-treated cells than with untreated cells. In contrast, no differences were found when cells were seeded on laminin or Matrigel (Fig. 5H,I), indicating that MT1-MMP was not recruited to podosome rosettes elicited on these matrices. Podosome rosettes stained for α2β1 integrin in cells seeded on Col-IV but α6β1 integrin was absent from the podosome areas when cells were seeded on laminin (Fig. S4B).

Collectively, these results indicate that basement membrane proteins are conducive for podosome formation in HMVECs exposed to VEGF-A and that podosomes mediate Col-IV but not laminin degradation. In contrast, VEGF-A does not promote podosome formation in HMVECs seeded on either fibronectin or Col-I.

Given that p190RhoGAP-B expression appeared to be essential for podosome induction, we examined its relocalisation upon VEGF-A stimulation when cells were seeded on the various ECM proteins. Increased recruitment of p190RhoGAP-B at podosomes was observed only when cells were seeded on Col-IV matrices (Fig. 6A). Because laminin did not elicit such a response, p190RhoGAP-B relocalisation did not appear to be the signalling component for podosome induction on this matrix protein but rather for the podosome-degrading function. We also explored Src regulation as Src both interacts with adhesion molecules and is a crucial downstream effector of VEGF receptors (Claesson-Welsh and Welsh, 2013). By analysing the autophosphorylation of Tyr418 within the kinase activation loop as a marker of Src activation, we found that Col-IV, but none of the other matrices, enhanced Src activity in HMVECs exposed to VEGF-A (Fig. 6B). Combined, these results show that Col-IV synergises with VEGF-A to regulate podosome induction and degrading function.

Most of our knowledge on podosomes comes from studies performed in cells of the monocytic lineage, that is, single cells which form these cytoskeletal structures spontaneously. However, podosomes have now been described in many other cell types where they have the fundamental difference of being inducible but not constitutive. Podosome architecture and arrangement also vary depending on the cell type being considered (Juin et al., 2012,, 2013; Spuul et al., 2015). This raises the question as to what extent our knowledge on monocytic cell podosomes can be extrapolated to those found in other cell lineages. In this study, we characterised podosomes formed by microvascular endothelial cells, which form the capillary endothelium in vivo. We identified two types of podosomes, constitutive podosomes that form in adherent endothelial cells in the absence of specific stimulation and induced podosomes that arise in response to VEGF-A. In both situations, endothelial cell podosomes were organised in rosettes and the basic bipartite podosome architecture remained similar to that described for individual podosomes in monocytic cells. In endothelial cells, we also visualised the actin cables interconnecting neighbouring podosomes, which are probably instrumental in generating the characteristic rosette pattern.

In HMVECs, podosome formation appeared not to be restricted to the CD34⁺ subpopulation (Siemerink et al., 2012). Rather, we found that confluence positively regulated podosome formation independently of VEGF-A, suggesting that either some junctional components or cell–cell communication and signalling pathways participate in this response. A link between junctional proteins and podosomes has been made because the actin-binding zonula occludens protein 1 (ZO-1) has been found at podosomes in smooth muscle cells (Kremerskothen et al., 2010). However, the involvement of Notch signalling, which links the fate decisions of one cell to those of its neighbours during angiogenesis, appear a more likely mechanism (Phng and Gerhardt, 2009). Here, we show that restoring cell–cell junctions and direct cell communication in vitro promotes podosome formation in a context where endothelial cells are activated by the in vitro culture conditions. Therefore, the relevance of these findings needs to be addressed in vivo using established angiogenesis models (Norrby, 2006).

We previously characterised the occurrence of podosomes in endothelial cells derived from liver sinusoids (LSECs) (Juin et al., 2013). These specialised capillaries exhibit morphological features that differentiate them from classical capillaries, such as the lack of a continuous basal membrane (Le Bail et al., 1990), which facilitates their specific functions in the liver. A large fraction of LSECs (>60%) in culture form constitutive podosomes and they do not organise in rosettes. In addition, VEGF-A does not detectably affect podosome formation in these cells. Thus, the constitutive podosomes in LSECs differ substantially from those found in HMVECs.

The podosome response to VEGF-A reached its maximum after 24 h, in contrast to HUVECs where the peak was observed at 5 min and the level returned to baseline within 1 h (Osiak et al., 2005). VEGF-A-induced and constitutive podosomes shared similar components but exhibited marked differences in terms of gelatinolytic activity. Such changes correlated with increased MT1-MMP cell surface exposure and increased recruitment of p190RhoGAP-B at podosomes. In addition, p190RhoGAP-B was also required for podosome induction in response to VEGF-A, suggesting that induced podosomes are assembled de novo. Although it is possible that p190RhoGAP-B regulates these two events independently, at this stage, we cannot exclude that MT1-MMP regulation impacts on podosome induction in response to VEGF-A. p190RhoGAP-B is known to be primarily regulated by subcellular localisation (Matheson et al., 2006) and to regulate the spatial regulation of RhoA activity (Ponik et al., 2013). In VEGF-A-stimulated HMVECs, we show that the local activation of RhoA at podosomes was dependent on p190RhoGAP-B. Thus, whereas constitutive podosomes are primarily regulated by cell confluence, VEGF-A-induced podosomes are controlled by a pathway involving p190RhoGAP-B.

As for cells of the myelomonocytic lineage, most substrata seem to be permissive to the formation of podosomes in HMVECs. However, some ECM proteins were found to be more favourable than others. Vitronectin is the preferential matrix for podosome formation in osteoclasts (Fuller et al., 2010) whereas fibrinogen is the best for macrophages (Labernadie et al., 2010). We have now found that laminin and Col-IV are the preferential substrates for podosome induction in HMVECs. These results contrast with those obtained with HUVECs, where laminin was shown to impair podosome formation (Seano et al., 2014). Our findings show the instrumental roles played by matrix proteins in podosome formation in various cell types and again highlight fundamental differences between endothelial cells originating from different vascular beds.

Our results also reveal a role for matrix proteins in the response to VEGF-A, which eventually dictate podosome function. In cells seeded on Col-IV, VEGF-A induced the formation of podosomes capable of degrading the underlying matrix, whereas, in cells seeded on laminin, VEGF-A induced inactive podosomes proteolytically. VEGF-A activates Src and this can occur through several pathways (Caron et al., 2009; Meyer et al., 2008; Sun et al., 2012). On Col-IV, VEGF-A stimulation increased Src autophosphorylation, p190RhoGAP-B relocalisation and MT1-MMP cell surface exposure at podosome sites whereas these events did not occur on laminin, thus suggesting differences in VEGF-A signalling on these two matrices. The molecular mechanisms of these responses remain to be established but likely involve signalling crosstalk between VEGF-A receptor and integrins (Plow et al., 2014).

The data presented in Table 1 summarise our findings. The results obtained on uncoated matrix were similar to those obtained on Col-IV, consistent with our finding that cultured HMVECs produce large amounts of Col-IV. Gelatin (denatured Col-I) gave similar results to those with globular Col-IV. Thus, in HMVECs, Col-IV is both a master regulator of podosome formation and a potent inducer of Col-IV degradation. We conclude that VEGF-A-induced podosomes in HMVECs subsume specific functions in relation to the basal membrane. They disrupt the Col-IV bonds to loosen the laminin–Col-IV scaffold (Rowe and Weiss, 2008), which subsequently leads to the dissociation of the laminin meshwork (Yurchenco et al., 2004). In these cells, laminin-induced podosomes might mediate other podosomal function such as cell adhesion (Veillat et al., 2015). Interstitial matrix proteins did not promote podosome formation even in the presence of VEGF-A, suggesting that, during sprouting angiogenesis, podosomes are not involved in the migration that follows the extrication of the activated endothelial cell from the parental vessel (Davis and Senger, 2005). We hypothesise that by promoting podosome formation, VEGF-A enables endothelial cells to overcome the basement membrane barrier to allow sprouting from the existing vasculature as well as for re-entering the network by facilitating anastomosis of vessel segments cleared of basement membrane coverage.

Human pulmonary microvascular endothelial cells (HMVECs, Lonza, tested free of contamination) were maintained in complete endothelial cell growth medium (EGM-MV; Promocell, a culture medium that does not contain VEGF) containing antibiotics at 37°C in a 5% CO₂ humidified atmosphere and used between passages two and seven. Cells were trypsinised and seeded in complete medium at 1×10⁵ cells per plate in 3.5-cm dishes for western blot experiments or at 1.5×10⁴ cells per glass coverslip in four-well plates for immunofluorescence imaging. For stimulation, human recombinant VEGF-A (used at 25 ng/ml in all experiments, except for the dose-response experiment) was obtained from Promocell. Neutralising VEGF antibodies (MAB293) were obtained from R&D Systems. Podosome formation assays were performed by double staining for F-actin and cortactin.

Matrix protein coatings were performed on glass coverslips as follows: gelatin (Sigma), diluted to 0.1% in PBS at 37°C for 30 min; plasma-derived fibronectin (Sigma), diluted to 20 µg/ml in PBS at 37°C for 90 min; Matrigel (growth factor-reduced Matrigel™ basement membrane matrix, BD Biosciences) diluted with complete medium to 1 mg/ml at 37°C for 30 min; Col-IV (Sigma) diluted to 10 µg/ml in PBS at 4°C overnight; Col-I (Sigma), diluted to 0.2 mg/ml in PBS at 37°C for 30 min; and laminin (placental laminin containing mainly vascular LN-511, Sigma) diluted to 20 µg/ml in PBS at room temperature for 60 min. After washing with PBS, coverslips were incubated for at least 20 min with culture medium before adding the cells.

For the gelatin degradation assay, glass coverslips were first coated with Oregon-Green-488-conjugated gelatin diluted to 0.1% in PBS at room temperature for 30 min, washed with PBS and fixed with 0.5% glutaraldehyde (EMS) for 15 min. After washing with PBS, coverslips were incubated in 5 mg/ml sodium borohydride for 15 min under constant agitation, sterilised in 70% ethanol, then washed three times in PBS and finally incubated with culture medium before adding the cells. For assays on fluorescent matrices, cells were seeded on either fibronectin–TRITC (Cytoskeleton), Col-IV–FITC or Col-I–FITC (Invitrogen), laminin–FITC (a kind gift from Jean-Philippe Bouchara, Groupe d'Etude des Interactions Hôte-Pathogène, Université d'Angers, France), as indicated above and, after adhesion, treated or not with VEGF-A for the time indicated, then fixed and processed for immunofluorescence staining. Quantification of degradation areas on fluorescence-labelled matrices was performed for at least 10 fields (×10 objective lens) for each coverslip. The areas of degradation were quantified by using ImageJ software. Degraded areas were thresholded and measured by the Analyze Particles function of ImageJ software. The loss of matrix-associated fluorescence (total degraded areas) is presented as the intensity of matrix degradation. The total degradation area (expressed in µm²) was then normalised for the number of cells (degradation index). Control values were arbitrarily taken as 100%.

Cells (on a 6-cm dish) were washed twice with ice-cold PBS and 0.5 mg/ml EZ-Link™ Sulfo-NHS-Biotin (Thermo Scientific) was added for 30 min (under constant shaking at 4°C). Cells were washed twice with quenching solution (100 mM Glycine in PBS) followed by quenching for 15 min (under constant shaking at 4°C). Cells were washed twice with PBS and lysed with 300 µl immunoprecipitation buffer [20 mM Tris-HCl pH 7.6, 150 mM NaCl, 1% Triton X-100, 1 mM EDTA pH 8, 1 mM EGTA pH 8, 2.5 mM pyrophosphate, 1 mM β-glycerophosphate, protease inhibitor cocktail solution added at 1:25 dilution and phosphoSTOP phosphatase inhibitor solution added at 1:10 dilution (Roche)]. Biotin-labelled material was captured by streptavidin–Sepharose beads (GE Healthcare). Biotinylated cell surface fractions and intracellular fractions were separated and analysed by immunoblotting. Bound proteins were detected by western blot analysis.

siRNA transfection into HMVECs was performed by transfection with double-stranded short interfering RNA (siRNA, 50 nM) using the Nucleofector technology. The 21-bp duplex Smartpools of siRNA used for p190RhoGAP-A (sc-44077), p190RhoGAP-B (sc-97682), ARAP3 (sc-61988) were obtained from Santa Cruz. Control siRNA (All-star negative control, #1027281) was purchased from Qiagen. Cells were stimulated with VEGF-A 24 h after transfection as indicated above and analysed.

Antibodies against cortactin (clone 4F11), P-tyrosine (clone 4G10, Platinum, #05-1050), fascin, αVβ3 (clone LM609, MAB1976) and integrin α6 (clone GoH3, MAB135) were obtained from Millipore. Antibodies against N-WASp (clone 30D10, #4848) were from Cell Signaling Technologies. Antibodies against pMLC (Ab2480), MT1-MMP (Ab51074), Nck1 (Ab14588), laminins (Ab30320), DynII (Ab3457) and Col-I (Ab6308) were from Abcam. Antibodies against Grb2 (LB-B4079) were from Lifespan Biosciences. Antibodies against Tks5 (sc-30122), CD44 (sc-7946), IQGAP (sc-376383) were obtained from Santa Cruz Biotechnology. Antibodies against P-Y418-Src were from MBL (#AT-7135). Antibodies against Fgd1 (#HPA000911), myosin IIA (#M8064), vinculin (clone hVIN-1, #V9131), ARAP3 (#HPA042887), tubulin (clone DM1A, #T9026) and fibronectin (#F3648) were purchased from Sigma. Cleaved Col-I antibodies (targeting the cleavage site, G775-I776) were obtained from ImmunoGlobe^®. Antibodies against CD34 (#M-6175) were obtained from Dako. Antibodies against Dbnl were purchased from Proteintech (#13015-1-AP). Monoclonal antibodies against p190RhoGAP-A (#610150) and p190RhoGAP-B (#611613) were obtained from BD Biosciences. Conformation-specific antibodies for RhoA-GTP- were from NewEast Biosciences (#26904). Laboratory-made antibodies against integrin α2 (clone JBS2) were a gift from Linda Pilarski (Cross Cancer Institute, Canada), those against Arp2/3 were kindly given by Matthew Welch (University of California, Berkeley, CA) and those against WIP were kindly given by Ines Anton (CNB-CSIC, Madrid, Spain). For secondary detection, species-specific Alexa-Fluor-488- or Alexa- Fluor-647-labelled secondary antibodies (Jackson Laboratories) were used.

Cells were washed in ice-cold PBS and cell lysates were prepared by adding 100 µl of Laemmli buffer on cell monolayers. Samples were boiled for 5 min. Cell lysates (20 µg of proteins) were subjected to SDS-PAGE and proteins were transferred from gels onto immobilon polyvinylidene difluoride membranes (GE Healthcare). Membranes were saturated with 5% bovine serum albumin in Tris-buffered saline containing 0.1% Tween 20 for 1 h and incubated with appropriate primary antibodies (used at a 1:500 dilution except those against tubulin and β-actin that were used at a 1:10,000 dilution). Overnight at 4°C. Then, primary antibodies were revealed using horseradish-peroxidase-coupled anti-mouse-IgG or anti-rabbit-IgG (Jackson laboratories) secondary antibodies for 1 h. Finally, horseradish peroxidase activity was revealed using a chemiluminescence kit (GE-healthcare) according to the manufacturer's instructions. The amounts of proteins detected by western blotting were determined by scanning the autoradiograph (densitometry), followed by processing of the data with ImageJ. All values were normalised against tubulin or β-actin, which was used as a loading control.

Subconfluent cells (unless otherwise indicated) were fixed for 30 min at room temperature in 4% (w/v) paraformaldehyde in PBS and permeabilised with 0.1% Triton X-100 (Sigma) and incubated with primary antibodies (1:100, except cortactin and vinculin, 1:300). F-actin was visualised with Alexa-Fluor-546- or -488-conjugated phalloidin and, when indicated, the nuclei were stained using Hoechst 33342 (Molecular Probes) together with secondary antibodies (1:100). For MT1-MMP cell surface exposure, the permeabilisation step was omitted and cells were incubated for 1 h with MT1-MMP antibodies after several PBS washes. Another 30 min step with 4% PFA was performed to fix the antibodies before permeabilising with 0.1% Triton X-100 and immunofluorescent staining as above. The coverslips were washed in water and mounted on microscope slides with ProLong Gold Antifade containing (or not containing) 4′,6-diamidino-2-phenylindole (DAPI; Life Technologies). For STED, phalloidin-conjugated Alexa-ATTO 647N (Invitrogen) for F-actin detection or ATTO647N-labelled secondary antibodies (Invitrogen) for vinculin or cortactin detection were used (at a 1:100 dilution).

Confocal fluorescent images were acquired with a Zeiss LSM 510 Meta inverted laser scanning fluorescence microscope equipped with acquisition software (LSM 510 acquisition software; Zeiss) and a ×63 NA 1.4 oil immersion objective. Triple- or quadruple-colour imaging for DAPI or Hoechst 33342, Alexa-Fluor-546–phalloidin, and Alexa-Fluor-488-labelled and/or Alexa-Fluor-647-labelled secondary antibodies was performed using selective laser excitation at 350, 543, 488 and/or 647 nm, respectively. Each channel was imaged sequentially using the multitrack recording module before merging. Cells showing podosome rosettes were quantified by scoring at least 300 cells for each coverslip. Quantification of proteins at podosome rosettes was computed using the ImageJ Analyze Particles function by taking the mask of the rosette according to the cortactin staining and applying the mask to the image corresponding to the protein analysed. Integrated density was used as readout. To analyse the number of podosomes presenting MTI-MMP, F-actin- and cortactin-positive podosome dots were scored when positive for MT1-MMP. STED super-resolution images were acquired with a Leica DMI6000 inverted TCS SP5 AOBS microscope combined with the regular confocal mode for the companion markers.

To assess cell motility, HMVECs were seeded at a density of 15×10³cells per well on Oregon-Green-488–gelatin-coated glass-bottom 24-well plates (Greiner). Hoescht 33342 (10 ng/ml) was added to label the nuclei and the cells were either left untreated or treated with 25 ng/ml of VEGF-A. Live-cell imaging was performed with a Eclipe Ti Nikon videomicroscope coupled with NIS analysis software, based on a 10× NA 0.30 objective lens (Nikon), a Hamamatsu Digital CCD C10600-10B camera, and an environmental chamber to maintain cells at 37°C in a 5% CO₂ humidified atmosphere (Life Imaging Services). Videos were acquired every 15 min for a period of 3 h in bright-field, green and blue filters (Oregon-Green-488–gelatin and Hoescht 33342, respectively). Mean cell velocity and cell displacement were calculated with the ImageJ plugin TrackMate. The software was used to automatically track nuclei and characterise their trajectories amongst stacks of images.

Each experiment was performed at least three times and quantification values represent the mean±s.d. of three independent experiments if not stated otherwise. Statistical analysis was performed with GraphPad Prism 6 (GraphPad Software, Inc., San Diego, CA). Significance was determined by using a Student's t-test or one-way ANOVA (Bonferroni multiple comparison test between selected pairs), and P<0.05 was considered statistically significant.

We thank Prof. Andreas Bikfalvi for sharing some of the reagents used in the study and Laure-Anne Cussac for help with quantification experiments. We thank Dr J. P. Bouchara (Groupe d'Etude des Interactions Hôte-Pathogène, EA 3142, Université d'Angers, France) for providing laminin–FITC, Dr Linda Pilarski (University of Alberta, Canada) for providing anti-integrin-α2 antibodies (JBS2) and Ines Antón (CNB-CSIC), Madrid, Spain) for providing WIP antibodies. We also thank Ray Cooke for editing of the manuscript prior to submission.