Register for Mechanobiology 2016!

α2β1 integrin controls association of Rac with the membrane and triggers quiescence of endothelial cells
Laurence Cailleteau, Soline Estrach, Raphael Thyss, Laurent Boyer, Anne Doye, Barbara Domange, Nils Johnsson, Eric Rubinstein, Claude Boucheix, Teni Ebrahimian, Jean-Sebastien Silvestre, Emmanuel Lemichez, Guerrino Meneguzzi, Amel Mettouchi


Integrin receptors and their extracellular matrix ligands provide cues to cell proliferation, survival, differentiation and migration. Here, we show that α2β1 integrin, when ligated to the basement membrane component laminin-1, triggers a proliferation arrest in primary endothelial cells. Indeed, in the presence of strong growth signals supplied by growth factors and fibronectin, α2β1 engagement alters assembly of mature focal adhesions by α5β1 and leads to impairment of downstream signaling and cell-cycle arrest in the G1 phase. Although the capacity of α5β1 to signal for GTP loading of Rac is preserved, the joint engagement of α2β1 interferes with membrane anchorage of Rac. Adapting the ‘split-ubiquitin’ sensor to screen for membrane-proximal α2 integrin partners, we identified the CD9 tetraspanin and further establish its requirement for destabilization of focal adhesions, control of Rac subcellular localization and growth arrest induced by α2β1 integrin. Altogether, our data establish that α2β1 integrin controls endothelial cell commitment towards quiescence by triggering a CD9-dependent dominant signaling.


Interactions between endothelial cells and the extracellular matrix have a central role in the regulation of angiogenesis processes. Endothelial cells can contact either basement membrane or provisional matrix proteins, and adopt a resting or motile and proliferative behavior, respectively (for reviews, see Adams and Alitalo, 2007; Avraamides et al., 2008). Endothelial cells sense these environmental cues through integrin-adhesion receptors that signal to determine cell behavior. The large repertoire of integrins allows transmission of a broad spectrum of intracellular signals. Binding of specific extracellular matrix ligands to cell-surface integrins, the associated conformational activation, and intracellular signaling are relatively well-studied processes (Humphries et al., 2006; Luo et al., 2007). By contrast, much less is known about mechanisms that lead to the coordination of signaling routes originating from the concomitant engagement of diverse integrins.

Several reports show the crucial role of α5β1 and α2β1 integrins, respective receptors of fibronectin and laminin or collagen, in angiogenesis. Genetic deletion of α5 integrin in mice leads to embryonic death owing to extensive vascular and mesodermal defects (Francis et al., 2002). Consistently, α5 integrin antagonists inhibit angiogenesis induced by tumors or growth factors in adult mice (Bhaskar et al., 2007; Kim et al., 2000). Furthermore, α5β1 integrin was shown to influence endothelial cell survival in vivo (Kim et al., 2002) and induces endothelial cell proliferation (Mettouchi et al., 2001). Recent studies on α2-integrin-knockout mice have established that although they present no developmental angiogenesis defects, the adult mice display strong pathological angiogenesis deregulations, because tumor and wound angiogenesis are enhanced (Grenache et al., 2007; Zweers et al., 2007). Moreover, the angiostatic effect of endorepellin is mediated by its binding to α2β1 integrin (Woodall et al., 2008). How extracellular matrix cues are integrated by these two integrins in endothelial cells to coordinate their behavior and adapt their response remains to be addressed.

Progress has been made by showing that regulation of Rac-GTPase activity is a signaling hub in endothelial-cell growth controlled by α5β1 integrin, upon adhesion to fibronectin. Importantly, activation of Rac allows accumulation of cyclin D1 in the presence of growth factors, which determines cell-cycle progression (Mettouchi et al., 2001). Endothelial cells are one of the few cell types using α2β1 integrin as a receptor for laminin (Elices and Hemler, 1989; Kirchhofer et al., 1990; Languino et al., 1989). Strikingly, we previously observed that α2β1 integrin does not activate Rac or induce cell proliferation upon contact with laminin-1 (Mettouchi et al., 2001). In the present study, we found that when α2β1 integrin is ligated to laminin-1, it interferes with the capacity of α5β1 integrin to generate mature focal adhesions and propagate downstream signals for cell proliferation. This results in a strong block of cell proliferation and the maintenance of endothelial cells in a quiescent state. We show that among the laminin-binding integrins expressed in endothelial cells, α2β1 integrin is the predominant member that acts in this capacity. We investigated membrane-proximal events involved in the dominant signaling mechanism of α2β1 integrin using a new interaction-screening ‘split-ubiquitin’ technique. This allowed us to identify an interaction between the integrin α2 chain and the CD9 tetraspanin. We show that the α2 integrin and CD9 signaling axis restricts proper α5β1-integrin-mediated focal-adhesion assembly and downstream Rac activation. We thus show in this study that α2β1 integrin conveys a dominant signal to lock the cell cycle by impairing Rac-GTPase membrane anchoring, thereby defining this integrin receptor as an important player in endothelial cell fate.


Engagement of laminin-1 and integrin triggers a dominant signal leading to endothelial cell quiescence

Laminin (Lm) and fibronectin (Fn) are two components of the endothelial cell microenvironment, which in vitro lead either to quiescence or proliferation of HUVEC endothelial cells, respectively (Mettouchi et al., 2001). To study how endothelial cells integrate antagonist signals, we took advantage of a culture system to investigate cell behavior upon plating on a composite matrix made of Fn and Lm-1, bound by α5β1 and α2β1 integrins, respectively. Although α2β1 integrin functions as a cell-surface receptor for collagen on some cells such as platelets, it functions as both a laminin and collagen receptor on a more restricted subset of cells, including endothelial cells (Elices and Hemler, 1989; Kirchhofer et al., 1990; Languino et al., 1989). Adhesion of endothelial cells to Lm-1 is robustly inhibited by α2-integrin-blocking antibodies, whereas blocking antibodies directed against α3 or α6 subunits had only marginal effects (supplementary material Fig. S1). This shows that α2β1 integrin is the main receptor for Lm-1 in HUVECs.

Cells adhere to the extracellular matrix through major structures of close apposition of the cell plasma membrane to the substrate, encompassing integrins, talin, vinculin, paxillin, tensin and α-actinin, which are hierarchically organized with signaling molecules and actin in clusters called focal adhesions (Berrier and Yamada, 2007). Cells plated on Fn displayed a spread phenotype, harboring mature focal adhesions (Fig. 1A, arrows) as determined by their content of vinculin, paxillin and phospho-tyrosine (Tyr-P), and actin stress fibers. By contrast, endothelial cells plated on Lm-1 showed reduced spreading and accumulation of cortical actin. In these cells, vinculin, paxillin and Tyr-P signals were distributed to the cell periphery in adhesive structures resembling focal complexes (Fig. 1A, arrowheads). Endothelial cells plated on a composite matrix containing Fn and Lm-1 (3:1) displayed intermediate phenotypes, showing a low content of actin stress fibers, compared with Fn-plated cells, with a cortical actin staining similar to that observed on Lm-1-plated cells. The cells also displayed smaller focal adhesions and presented peripheral focal complexes (Fig. 1A). The presence of a similar amount of Fn in the composite matrix was verified (supplementary material Fig. S1). Upon examination of integrin localization (Fig. 1B), we observed that α2 integrin accumulated at the cell periphery and never clustered in focal adhesions, regardless of the nature of the extracellular matrix. Integrin α5 concentrated in focal adhesions when cells were in contact with Fn. By contrast, α5 integrin displayed a partially dispersed membrane localization, together with perinuclear accumulation, in cells plated on Lm-1. Interestingly, we observed that cells plated on the composite matrix (Fn:Lm, 3:1) showed an intermediate distribution pattern for the α5 integrin. These observations suggested an interfering effect of Lm-α2β1 on Fn-induced adhesion structures, thus raising the possibility that α2 integrins might exert a dominant effect on α5 integrin functions.

Fig. 1.

Dominant-interfering effect of laminin-1 on the assembly of focal adhesions and actin stress fibers mediated by fibronectin. (A) G0-synchronized HUVECs were detached and replated in complete SFM for 2 hours on fibronectin (Fn), Laminin-1 (Lm) or a mixed matrix composed of both proteins in the ratio 3:1 (Fn:Lm). Cells were stained with antibodies against vinculin, paxillin, phosphorylated tyrosine residues (Tyr-P; labeled as p-Tyr) or with phalloidin-FITC and observed with a fluorescence microscope to visualize the extent of focal-adhesion formation and actin-cytoskeleton fasciculation. (B) Immunolocalization of endogenous α5β1 and α2β1 integrins in the same conditions. Scale bars: 20 μm.

To further test our hypothesis, we investigated whether the signaling function of α5β1 integrin was also affected upon α2β1 integrin co-engagement. Endothelial cells were cultured on a composite matrix made of different ratios of Lm and Fn and assayed for their proliferation state. We observed that addition of Lm-1 to a Fn matrix diminishes cell proliferation in a dose-dependent manner (Fig. 2A). This decrease in proliferation was associated with a defect in accumulation of cyclin D1, which is a restrictive event required for the G1 to S-phase transition (Fig. 2A, inset). Analysis of the DNA content confirmed that addition of Lm-1 blocked cells in the G1 phase of the cell cycle (Fig. 2B). Addition of Lm-1 also reduced FAK autophosphorylation (Fig. 2C), a proximal signal triggered by α5β1 integrin ligation. To demonstrate the involvement of α2β1 integrin in the dominant effect of Lm-1 on α5β1 integrin functions, we performed selective RNA interference knockdown experiments. There was no effect of Lm-1 on focal adhesions initiated by α5β1 in cells knocked down for α2 integrin (Fig. 2D; supplementary material Fig. S2A). We verified that α2β1 integrin silencing had no effect on expression of α5 integrin (supplementary material Fig. S2A). Moreover, knock down of α2β1 integrin in endothelial cells rescued their proliferation capacity when plated on the Fn:Lm compostite matrix, indicating that the cell-cycle restriction imposed by Lm-1 involves α2β1 integrin ligation (Fig. 2E). These results rule out nonspecific effects, such as steric obstruction of Fn–α5β1-integrin recognition. Together, our data thus suggest that α2β1 integrin is a dominant switch that triggers endothelial cell quiescence.

Fig. 2.

The dominant-interfering effect of laminin-1 is mediated by α2β1 integrin and leads to the inhibition of α5β1 integrin signaling and cell-cycle arrest. (A) G0-synchronized HUVECs were plated in complete SFM for 24 hours on fibronectin (Fn) alone or in combination with Laminin-1 (Lm) at different ratio, Fn amounts being constant. The graph shows the percentage (mean ± s.d.) of BrdU incorporation (n=3, P<0.005). The defect in BrdU incorporation induced by Lm is associated with a defect in cyclin D1 production at 24 hours, as shown by western blot (representative of three independent experiments). (B) G0-synchronized HUVECs were plated in complete SFM on fibronectin (Fn), Laminin-1 (Lm) or a mixed matrix in the ratio 3:1 (Fn:Lm) for 24 hours. DNA content was analyzed by flow cytometry upon propidium iodide staining. (C) G0-synchronized HUVECs were detached (S) and replated in complete SFM on fibronectin (Fn), Laminin-1 (Lm) or a mixed matrix in the ratio 3:1 (Fn:Lm) following a time course. FAK activity was measured by immunoblot using a specific antibody against the phosphorylated tyrosine auto-activation residue of FAK (Y-397). (D) RNA interference to knock down the α2 subunit of integrin in HUVEC cells demonstrates that the dominant effect of laminin-1 is mediated by ligation of α2β1 integrin. Immunolabeling was performed after adhesion for 2 hour on a matrix composed of Fn and Lm (3:1). Representative of three independent experiments. Scale bar: 20 μm. (E) RNA interference to knock down the α2 subunit of integrin allows for HUVEC proliferation recovery on the Fn:Lm matrix (mean ± s.d., n=5, P<0.004).

α2β1 integrin is specifically involved in the dominant induction of endothelial cell quiescence

To assess whether other Lm receptors contribute to the dominant effect on α5β1 integrin function in endothelial cells, we compared the capacity of cells plated on the Fn:Lm composite matrix to form mature focal adhesions upon selective knockdown of α3β1 and α6β1 integrins. Whereas knockdown of α2β1 integrin allowed massive focal adhesion formation on a Fn:Lm matrix, knockdown of α3 or α6 integrin had only minor effects, as shown by immunolabeling for paxillin (Fig. 3A; supplementary material Fig. S2B).

In a complementary approach, we assessed whether other α2β1 integrin ligands could trigger this dominant effect on α5β1 integrin focal-adhesion clustering. The α2β1 integrin binds to the filamentous type I collagen and also serves in cultured endothelial cells as a receptor for the basement membrane protein preparation Matrigel and for a human laminin preparation (Hum-Lm) (supplementary material Fig. S1). This commercial laminin preparation was shown to contain laminin-2, laminin-8 and laminin-10 (Wondimu et al., 2006). We labeled focal adhesions of endothelial cells plated on a mixture of Fn and either collagen-I, Hum-Lm or Matrigel™. We found that addition of Matrigel™ or Hum-Lm to Fn led to dispersal of focal-adhesion structure in a similar way to Lm-1, whereas collagen-I was unable to trigger this effect (Fig. 3B). The capacity of each isolated protein to nucleate focal adhesions is shown in supplementary material Fig. S3. These data suggest that all ligands of α2β1 integrins are not equivalent in their ability to trigger dispersion of α5β1-integrin-initiated focal adhesions. We could show that α2β1 integrin is not the sole collagen-I receptor in our cells because adhesion blocking on this matrix was maximal in the presence of a combination of blocking antibodies against α2β1 and α1β1 integrins (supplementary material Fig. S1). Interestingly, α1β1 integrin is a unique collagen receptor that is involved in cell proliferation (Pozzi et al., 1998), and one could hypothesize that α2β1 integrin signaling is dominant on α5β1 integrin but not on other proliferative integrins such as α1β1 integrin. Alternatively, ligand-specificity studies have highlighted key mechanistic differences between the binding of α2 integrin to collagen and laminin (Dickeson et al., 1998). These could result in distinct conformational activation of the receptor and different downstream signaling. In line with this hypothesis, recent studies describe the existence of several conformations of the α2 integrin, linked to different activation states, which determine the extent of cell spreading (Van de Walle et al., 2005).

Identification of binding partners of α2β1 integrin using the split-ubiquitin screen

To determine the molecular mechanism of α2β1 integrin dominant signaling, we performed an original split-ubiquitin screen to isolate partner proteins. The transmembrane region and cytoplasmic tail of α2 integrin were fused to the C-terminal half of ubiquitin (Cub), followed by the RUra3p reporter protein, to generate a bait referred to as α2-CRU. Upon interaction of the bait with a protein encoded from a cDNA fused to the N-terminal half of ubiquitin (Nub), reassembly of the ubiquitin halves into a native-like ubiquitin occur, RUra3p reporter is cleaved and degraded by the N-end-rule pathway (Fig. 4A). Ura3p is a yeast enzyme that catalyzes the decarboxylation of orotidine-5′-phosphate for the synthesis of uracil and also converts 5-fluoro-orotic acid (5-FOA) into the toxic compound 5-fluorouracil. Yeast that express interacting Nub and Cub fusion proteins are therefore uracil auxotrophic and 5-FOA resistant (Johnsson and Varshavsky, 1994; Wittke et al., 1999). Correct insertion of our bait in the yeast plasma membrane was rated by interaction with yeast proteins that reside in diverse subcellular compartments, fused to Nub. The localization of the bait and prey in the same compartment is enough to allow reassociation of the native ubiquitin motif. The strong 5-FOA resistance of yeast expressing α2-CRU together with fusion proteins known to be expressed in the plasma membrane (Nub-Sso1p), in the late Golgi and the plasma membrane (Nub-Snc1p), in the early Golgi (Nub-Sed5p) and in the endoplasmic reticulum (Nub-Sec62p) confirmed that α2-CRU is properly integrated into the plasma membrane and is distributed over the secretory pathway (Fig. 4B). We transformed into the α2-CRU-expressing yeast a human endothelial cDNA library fused to Nub(A), a point mutant of Nub (I13 to A), which displays lower affinity to Cub. The cleavage of the reporter therefore becomes strictly dependent on the interaction between the two fusion proteins and will not arise upon simple colocalization. Twenty clones harboring a coding sequence in-frame with Nub were confirmed positive. Two independent clones encoding a partial and full-length sequence of the transmembrane protein CD9 were identified (Fig. 4C). This shows that the split-ubiquitin screen is a powerful approach to probe for integrin-interacting proteins.

Fig. 3.

Effect of alternative α2β1 ligands and laminin-1 receptors on formation of focal adhesions in endothelial cells. (A) RNA interference to knock down the α2, α3 and α6 subunits of integrins in HUVEC cells demonstrates that the α2β1 integrin is the main receptor involved in the focal adhesion dispersion observed in the presence of laminin-1. Paxillin immunolabeling was performed after adhesion for 2 hours on a matrix composed of Fn and Lm (3:1). Results are representative of three independent experiments. (B) G0-synchronized HUVECs were plated in complete SFM for 2 hours on fibronectin (Fn), or a mixture of Fn and collagen-I (1.5:1), matrigel (3:1) or human laminin (Hum-Lm, 3:1). The cells were stained with antibodies against α5β1 integrin, vinculin and paxillin, and observed using a fluorescence microscope to visualize the formation of focal adhesions. Scale bars: 20 μm.

Cd9-knockout mice display abnormal adult angiogenesis

Several studies have reported a connection between the CD9 tetraspanin and integrins in diverse differentiation processes (Kurita-Taniguchi et al., 2001; Schwander et al., 2003; Tachibana and Hemler, 1999; Tanio et al., 1999). Thus, we further focused on the potential involvement of the α2β1-integrin–CD9 tetraspanin complex in regulating proliferation arrest of endothelial cells and angiogenesis.

Mice lacking CD9 are born healthy and grow normally, but suffer from reduced fertility in females because of a failure in sperm-egg fusion (Le Naour et al., 2000; Miyado et al., 2000). Many integrin-knockout mice display no developmental angiogenic defects, which is in contrast to their important role in adult angiogenesis. Therefore, we monitored angiogenesis in Cd9−/− adult mice using the subcutaneous matrigel plug assay, with bFGF as an inducer of vessel growth. Our data revealed that Cd9−/− animals failed to develop a new vasculature in the matrigel plug (Fig. 4D-F). Whereas wild-type (WT) mice developed numerous perfused vessels that stained for the endothelial specific glycoprotein endomucin, Cd9-null mice showed poor vascular infiltration of the implant. These results unravel an unsuspected role of the CD9 tetraspanin in modulating adult angiogenesis.

CD9 interacts with α2 integrin in endothelial cells and selectively co-localizes with α5 integrin in the presence of laminin

We further investigated the interaction and localization of endogenous CD9 and α2 integrins in endothelial cells. Using the technique of sequential immunoprecipitation from surface-labeled cells, we detected the α2 integrin in the CD9 complexes that were present at the cell surface of stably growing endothelial cells (Fig. 5A). This confirms the interaction between α2 integrin and CD9 that we discovered using the split-ubiquitin assay. We further addressed the pertinence of this biochemical interaction between CD9 and α2 integrins at the cellular level by confocal microscopy. The staining profiles CD9 and α2 integrin were similar, with both proteins present at the cell periphery in structures resembling focal complexes, and in dot-like arrangements throughout the cell plasma membrane (Fig. 5B). We observed that CD9 colocalized with α2 integrin on each extracellular matrix, the colocalization being maximal on the Lm-1 matrix. This indicates that a portion of CD9 can localize with α2 integrin in its resting state, and that ligation of the integrin is accompanied with a relocation of the tetraspanin in adhesion structures containing this integrin (Fig. 5B, top panel, overlay). When cells were in contact with Fn, CD9 did not colocalize with α5 integrin. This indicates that the localization of these proteins is exclusive when integrin α5 is bound to its ligand. Interestingly, colocalization of CD9 and α5 integrin was observed if cells are plated on Lm or Fn:Lm composite matrix (Fig. 5B, bottom panel, overlay), suggesting elaborate regulation of the formation of these complexes.

Fig. 4.

A new interaction screen allows isolation of the CD9 tetraspanin as an α2β1 integrin partner involved in angiogenesis. (A) The split-ubiquitin screen principle. Interaction between the bait and the prey allows for the re-formation of a ubiquitin motif that directs the URA3 reporter protein to proteasomal degradation. Yeast that do not encode interacting proteins are counter-selected on 5-FOA-containing medium because this compound is transformed by URA3 into the toxic 5-fluoro-uracil product, leading to death of the colony. This technique therefore allows for positive selection of yeast clones bearing proteins presenting affinity for the transmembrane protein of interest. In our case, the bait is composed of the transmembrane and cytosolic portion of the human α2 integrin. (B) Characterization of the yeast strain encoding the α2 integrin bait (referred to as α2-CRU), that confers to the parental JD53 strain histidine (his) and uracil (ura) prototrophy. The insertion of α2-CRU in the plasma membrane was verified through its capacity to interact with yeast proteins fused to the high intrinsic affinity Nub(I) region and resident in vacuoles, the endoplasmic reticulum (ER), early or late Golgi or the plasma membrane. The growth capacity of the clones, summarized in the table, was assessed on 1.5 mg/ml 5-FOA. (C) Isolation of two independent clones conferring the capacity to grow on 5-FOA-containing medium, corresponding to the CD9 tetraspanin. Control clones encoding the Tom22 protein or Nub(A) alone do not survive increasing doses of 5-FOA. (D-F) The in vivo angiogenesis response in adult Cd9-knockout mice was monitored using the subcutaneous matrigel plug assay. (D) Histological analysis of implant sections after 14 days, Masson's trichrome staining (scale bar: 100 μm). (E) The angiogenic response in WT (n=11) and Cd9−/− (n=8) mice was scored as detailed in the Materials and Methods. (F) Immunolabeling of the endothelium-specific protein endomucin in sections of the matrigel implants confirms defective vascularization of the plug in Cd9-null mice (scale bar: 50 μm).

Altogether, we show that CD9 interacts with α2β1 at the surface of endothelial cells. Engagement of α2β1 with Lm triggers enhanced CD9 colocalization with this integrin, and is associated with a new ability of CD9 to colocalize with α5β1.

Involvement of CD9 in α2-integrin-induced destabilization of focal adhesions and cell-cycle arrest

We next evaluated the role of CD9 in the destabilization of focal adhesion and interference with signaling of α5β1 integrin. Following overexpression of CD9 in cells plated on Fn, vinculin and paxillin redistributed from focal adhesions to focal complexes and accumulated in the cytoplasm (Fig. 6A; supplementary material Fig. S2C). This pattern resembled that observed in cells adhering on Lm. Interestingly, expression of CD9 also resulted in a relocation of α5β1 integrin from focal adhesions to small dot-like membrane structures and perinuclear vesicles, reminiscent of α5β1 integrin localization in cells on Lm-1 (Fig. 6A). CD9 expression did not affect localization of α2β1 integrin (data not shown). We then tested the effect of CD9 on cell-cycle progression by measuring incorporation of BrdU in WT or CD9-overexpressing endothelial cells. Consistent with inhibition of focal-adhesion assembly, CD9 overexpression strongly inhibited cell-cycle progression of cells plated on Fn (Fig. 6B). Thus, CD9 overexpression in cells plated on Fn produces similar effects as addition of Lm-1 to WT cells.

Next, we investigated whether cellular depletion of CD9 impaired the specific effects of Lm and α2β1 integrin on endothelial cell adhesion and proliferation. We examined whether depletion of CD9 using shRNA knockdown could relieve the block in cell proliferation constrained by the interaction between Lm and α2β1 integrin. As shown in Fig. 6C, a reduction of CD9 levels was associated with a partial rescue of the α2-integrin-induced cell-cycle block. Depletion of CD9 in cells plated on a composite matrix also led to partial recovery of focal-adhesion nucleation, as visualized by clusters of vinculin and paxillin that contained little α5β1 integrin (Fig. 6D; supplementary material Fig. S2D). In conclusion, CD9 has a key role in the α2β1-integrin-induced hindrance of focal-adhesion assembly and endothelial cell proliferation.

Fig. 5.

CD9 and α2β1 integrin selectively associate in endothelial cells. (A) After cell-surface biotinylation, HUVECs were lysed and CD9 and α2 integrin were immunoprecipitated and processed for SDS-PAGE and western blot analysis. Hybridization of the blots with streptavidin-HRP shows the presence of proteins at the surface, and their correct extraction for the analysis. Subsequently, the CD9 immunocomplex was dissociated and the α2 integrin was immunoprecipitated from the complex. Streptavidin-HRP blotting reveals the presence of the α2 integrin in complex with CD9 at the cell surface. (B) Confocal microscopy analysis of CD9, α2 and α5 integrins in HUVECs plated in complete SFM for 2 hours on fibronectin (Fn), laminin (Lm) or a mixed matrix composed of both proteins (Fn:Lm, 3:1). Images represent single optical z sections, the overlay shows zones of colocalization as a yellow signal. Scale bars: 20 μm.

α2β1 integrin and CD9 regulate membrane association of Rac

Rac activity is required for cyclin D1 expression and cell-cycle progression (Coleman et al., 2004; Mettouchi et al., 2001). Activation of Rac necessitates its GTP loading and membrane association. Several studies have highlighted the requirement of membrane stabilization of microdomains enriched in cholesterol and caveolin, called CEMM or RAFT, by integrins for the proper activation of Rac. Such domains enable Rac anchoring in this particular lipid environment (del Pozo et al., 2004; Del Pozo et al., 2002). We measured association of Rac to membrane upon adhesion of endothelial cells to the different extracellular matrices (Fig. 7A). Although ligation of α5β1 integrin to Fn was associated with Rac activation and translocation to membrane (Fig. 7A,B,C), the presence of Lm did not allow association of Rac with the membrane, as shown by absence of the GTPase from the membrane fraction when endothelial cells contact a matrix composed of both Fn and Lm (Fig. 7A). Nevertheless, GTP-Rac could be detected in cells contacting a matrix composed of both Fn and Lm (Fig. 7B,C). Therefore, these results suggest that Lm triggers an inhibitory signal on GTP-Rac membrane anchoring. Using RNAi to mediate knockdown of α2β1 integrin in endothelial cells, we could restore proper Rac membrane localization when cells adhered on the Fn:Lm (3:1) composite matrix (Fig. 7D; supplementary material Fig. S5A). This indicates that α2β1 integrin restricts Rac membrane localization. In a similar approach, knock down of the α2 integrin downstream effector CD9 also led to recovery of Rac proteins in the membrane fraction (Fig. 7E; supplementary material Fig. S5B). Using a Fn:Lm composite matrix at 3:1 and 2:1 ratios, we compared the effects of knockdown of CD9 and α2 integrin on Rac membrane association. We found that cells in which α2 integrin was knocked down recovered Rac membrane localization even if the dose of Lm was raised (Fn:Lm, 2:1). By contrast, knock down of CD9 allowed membrane association of Rac on Fn:Lm at (3:1) ratio but was inefficient in the presence of higher concentrations of Lm (Fig. 7F). This reflects either the need to completely abrogate CD9 expression in cells to block α2β1 integrin signaling efficiently, or the contribution of other effectors downstream of α2 integrin.

Although integrin-mediated adhesion to Fn is known to promote and stabilize CEMM-RAFT microdomains, tetraspanins organize specific detergent-resistant membrane microdomains referred to as tetraspanin-enriched microdomains (TEMs) that are distinct from rafts (Charrin et al., 2009; Claas et al., 2001; Hemler, 2005; Yanez-Mo et al., 2009). The localization of tetraspanins in these TEMs is important to form the tetraspanin web and elicit specific downstream functions (Charrin et al., 2003; Yang et al., 2002). Therefore, we investigated, using sucrose gradient fractionation experiments, the segregation pattern of the CD9 tetraspanin in the membrane when endothelial cells contact a particular extracellular matrix (Fig. 7G). CD9 was present both in light fractions, characterized by the presence of flotillin, and in heavy fractions of the membrane when cells contacted Fn alone. By contrast, the presence of Lm triggered preferential concentration of CD9 in the light fractions. These data suggest that α2β1 integrin ligation induces a change in membrane compartmentalization of CD9 that could reflect changes in binding partners and/or organization of the tetraspanin web at the plasma membrane, and might be linked to interference on signaling of α5β1 integrin to Rac.


Cells in vivo are exposed to multicomponent matrices of varying composition. Angiogenesis causes resting endothelial cells to loosen contact with basement membrane so that they can migrate and proliferate through a provisional matrix. This change of substratum requires engagement of a different array of integrins. It is important to decipher how integrin cross-talk could coordinate the cell response in the early or late phase of the angiogenesis process, situations where Lm and Fn binding integrins can be simultaneously ligated. Here, we demonstrate that the α2β1 integrin dominantly interferes with the capacity of α5β1 integrin to assemble mature focal adhesions and to signal, thereby blocking cell proliferation. Using the split-ubiquitin assay, a new screening technique adapted to transmembrane proteins, combined with functional analysis, we demonstrate the major role of CD9 tetraspanin in mediating α2β1-integrin-dominant signaling in endothelial cells. Integrin α2β1 and CD9 restrict Rac-GTP membrane localization and provoke cell-cycle arrest.

Fig. 6.

CD9 acts as a downstream target of the α2β1 integrin for cell-cycle control and focal-adhesion formation in endothelial cells. Effect of CD9 overexpression (A,B) or depletion upon RNA interference (C,D) in HUVECs. (A) HUVECs transiently transfected with pcDNA3-EGFP (Ctrl) or pCDNA3-CD9-EGFP were plated on fibronectin in complete SFM for 2 hours. The extent of focal-adhesion clustering was determined by immunofluorescence on fixed cells stained for paxillin, vinculin and α5 integrin. (B) The effect of CD9 overexpression on cell-cycle progression was assessed by BrdU incorporation on HUVECs plated for 24 hours in complete SFM on increasing doses of fibronectin. The graph shows the percentage (means ± s.d.) of transfected cells that incorporated BrdU, compared with mock cells (value set to 100%). (C,D) HUVECs transiently co-transfected with pDsRed2-mito and pSuper-CD9 were plated on a matrix composed of fibronectin and laminin-1 (3:1) in complete SFM. (C) Effect of CD9 depletion on HUVEC cell-cycle progression measured by BrdU incorporation for 24 hours in the DsRed2-positive population. Graph represents the inhibition of BrdU incorporation by Lm in mock (pSup-Ctrl) compared with CD9 sh-RNA (pSup-CD9) transfected cells (means ± s.d.; n=4). (D) The extent of focal-adhesion assembly was visualized by confocal immunofluorescence showing paxillin, vinculin and α5 integrin pattern in CD9-depleted cells (arrow) compared with surrounding non-transfected counterparts. For a better visualization, the DsRed2 signal (in red), indicating transfected cells, is only shown on the CD9 panel. Scale bar: 20 μm.

Dominant mode of action for α2β1 integrin in endothelial cells

Our study reveals a dominant-interfering action of α2β1 integrin on α5β1 integrin in endothelial cells to control proliferation. During wounding or cancer development, endothelial cells need to adopt a proliferative state to generate a new vessel branch, and to return to rest when the process terminates. The relief of α2β1 integrin engagement via basement-membrane degradation that occurs early in angiogenesis would gradually allow endothelial cells to transduce α5β1 integrin and growth-factor-receptor proliferation signals provided by the provisional matrix. Reports that investigated the role of α2β1 integrin in vivo established its crucial role in the regulation of tumor and wound angiogenesis, because animals in which α2 integrin is knocked down display enhanced angiogenesis (Grenache et al., 2007; Zweers et al., 2007). Mice lacking α2β1 integrin present increased production of MMPs upon wounding, which might contribute to remodeling of the extracellular matrix and stimulation of angiogenesis (Grenache et al., 2007). The increased tumor angiogenesis has been linked to a compensatory mechanism involving VEGFR1 overexpression in endothelial cells in which α2 integrin is knocked out (Zhang et al., 2008). Nevertheless, a relief of dominance exerted by α2 integrin on other integrins might also cause increased angiogenesis. Determination of the respective contribution of these pathways in the animal phenotype awaits analysis of α5β1 integrin signaling in these mice.

We analyzed angiogenesis in CD9-knockout animals using the matrigel plug assay. Surprisingly, we observed a defect in plug vascularization in response to bFGF. Although numerous vessels could be observed around the matrigel implant in WT and CD9-deficient animals (supplementary material Fig. S4B), vascularization of the plug only occurred in WT mice. Interestingly, when we analyzed the capacity of CD9-depleted HUVEC cells to form in vitro capillary-like structures in 3D matrigel, a process involving cell migration, we similarly observed a defect in network formation (supplementary material Fig. S4A). Thus, the phenotype of CD9-knockout mice could reflect a combination of cellular deregulation, including the inability of Cd9−/− endothelial cells to properly migrate and invade the implant.

Signaling downstream of α2β1 integrin

Integrin signaling can be either common to all integrins or heterodimer specific. Signaling events specific for α2β1 integrin have not been extensively explored, yet several studies converge on the activation of p38 kinase. In fibroblasts, activation of p38 by α2 integrin leads to collagen gene transcription and cell migration on type I collagen (Ivaska et al., 1999; Klekotka et al., 2001). In endothelial BAE cells exposed to shear stress, ligation of α2β1 integrin mediates p38 activation and subsequent NF-κB inhibition (Orr et al., 2005). In our model, we did not detect any p38 activation upon plating HUVECs on Lm-1. Moreover, treatment of cells with the specific p38 inhibitor SB203580 had no effect on the disruption of focal adhesions mediated by α2 integrin (data not shown). Activation of PKA is known to inhibit proliferation, disrupt the actin cytoskeleton and induce a stellate morphology in many cell types. In particular, PKA is activated in suspended cells, and repressed upon binding to fibronectin. This allows phosphorylation of proteins of the focal adhesion cytoplasmic plaque, activation of PAK, MAPK and cell-cycle progression (Howe and Juliano, 2000). Inhibition of PKA by engagement of α5β1 integrin is important during angiogenesis for endothelial cell migration and survival (Kim et al., 2002). Therefore, a role of PKA activation in the dominant function of α2β1 integrin was considered. Treatment of cells with the pharmacological PKA inhibitor H89 did not restore the formation of mature focal adhesions or the organization of actin stress fibers on a Fn:Lm matrix (data not shown). The lack of signaling candidates explaining the dominant effect of α2β1 integrin on endothelial cell proliferation prompted us to search for membrane-proximal α2-integrin-interacting partners that could mediate this signal. We have successfully adapted the split-ubiquitin sensor approach to screen a mammalian cDNA library against the transmembrane and cytosolic region of the integrin. This was done considering the crucial contribution of the transmembrane region to differential signaling in the integrin α-chain (Wary et al., 1998). This new approach enabled us to find an association between the transmembrane cytoplasmic region of α2 integrin and the tetraspanin CD9. The association of various tetraspanins with integrins is documented, modulating post ligand-binding events such as adhesion strengthening or receptor recycling (Hemler, 2005; Liu et al., 2007). Although interaction of CD9 with α3 or α6 integrins has been observed in several cell types (Berditchevski, 2001), an association with α2 integrin was only reported in vascular smooth muscle cells (Scherberich et al., 1998). Our results in yeast suggest the direct interaction of these two molecules. It is not clear to date whether any other tetraspanin, in addition to CD151, can contact directly integrins. Concerning the CD151 interaction with α3 integrin, the interaction interface has been mapped to the C-terminal region of the tetraspanin large extracellular loop (EC2) and the extracellular domain of α3 integrin. The two CD9 clones we isolated correspond to a full-length molecule and a portion encompassing the last five amino acids of the EC2 domain and the last transmembrane and cytosolic domains (residues 188-228). Both molecules interact in yeast with a portion of α2 integrin that does not contain the extracellular region. This indicates the occurrence of a newly identified type of interaction that is independent of the integrin ectodomain. Yet, we do not know whether in endothelial cells, this interface is the only region involved or whether it acts in concert with alternative binding sites within CD9 and α2 integrin. Further extensive mapping and biochemical interaction studies will be needed for a better understanding of this process.

Fig. 7.

α2β1 integrin and CD9 control Rac-GTPase membrane association in endothelial cells. (A) Immunoblots showing the distribution of Rac proteins between membrane (m) and cytosolic (c) fractions from HUVECs placed in suspension (S) or plated in complete SFM for 45 minutes on fibronectin (Fn), laminin-1 (Lm) or a mixed matrix (Fn:Lm, 3:1). The quality of membrane and cytosolic fractions is verified by the exclusive presence of transferrin receptor and Rho-GDI, respectively. (B) Time course of Rac activation upon plating on Fn or a mixed matrix (Fn:Lm, 3:1), measured by GST-PBD affinity purification. (C) Level of Rac-GTP present in HUVECs plated for 40 minutes in complete SFM on the different matrices (means ± s.d.; n=4). (D) RNAi mediated knockdown of α2 integrin in endothelial cells restores association of Rac with membrane (m) upon adhesion on the mixed matrix (Fn:Lm, 2:1). (E) RNAi-mediated knockdown of CD9 in endothelial cells restores association of Rac with membrane (m) upon adhesion on the mixed matrix (Fn:Lm) at the 3:1 ratio but not at the 2:1 ratio. (F) Quantification of Rac membrane association (mean % ± s.d.) in three independent experiments to knock down α2 integrin and CD9 performed on Fn:Lm mixed matrix at 3:1 and 2:1 ratios. The amount of Rac associated with membrane was set to 100% in control cells plated on Fn. This comparative analysis indicates that α2 depletion is more potent, because it allows localization of Rac to membrane in presence of high amounts of Lm. (G) Distribution of the CD9 tetraspanin and control proteins in isopycnic sucrose gradients. HUVECs were plated for 45 minutes on the indicated matrices in complete SFM and lysed in 1% Brij 58. One-sixth of the lysate was then centrifuged in a discontinuous sucrose density gradient and eight fractions were collected from the top. Each fraction was resolved by non-reducing SDS-PAGE, and immunoblotted for CD9, flotillin and transferrin receptor (Tf-R). The graph quantifies the portion of CD9 present in the flotillin-positive fractions.

Regulation of Rac-GTPase localization by α2β1 integrin and CD9

We provide evidence in the present study that α2β1 integrin regulates Rac activity by controlling its membrane targeting. This novel mode of regulation involves CD9 mobilization. Rac cycles between a GTP-bound active state and a GDP-bound inactive state, and this process is regulated by GTPase-activating proteins (GAPs) and guanine nucleotide exchange factors (GEFs) (Schmidt and Hall, 2002). Importantly, GTPases need to translocate from the cytosol to the membrane to complete their activation cycle, because their coupling to downstream effectors is precluded when they are dissociated from membranes. In this study, we demonstrate that ligation of α2β1 integrin impairs GTP-Rac membrane localization in endothelial cells. RNA interference of α2 integrin or CD9 rescued proper Rac subcellular localization in cells contacting the Fn:Lm matrix.

Several mechanisms might account for the modulation of Rac association to membrane by α2β1-integrin–CD9 signaling. GEFs can be involved in the association of Rac with membranes because they are often membrane bound and interact with GTPases at the plasma membrane. The localization of β-Pix in focal adhesions was involved in membrane translocation of Rac and ruffle formation in spread cells (ten Klooster et al., 2006). However, our data argue against a crucial role for GEFs in the control of Rac membrane localization upon α2β1 integrin and CD9 stimulation, because the interfering effect is clearly separable from GTP loading. Another possibility comes from the fact that binding of Rac to membranes requires prenylation and electrostatic interaction of its C-terminal polybasic region with the negatively charged inner leaflet of plasma membrane. A change in surface charge was correlated with the metabolism of phospho-inositides and changes in cytosolic Ca2+ during phagocytosis. This leads to dissociation of Rac from the plasma membrane at the site of sealing phagosomes (Ueyama et al., 2005; Yeung et al., 2006). The clustering of α2β1 integrin and CD9 in the presence of the relevant ligand could signal to trigger similar changes in the charge of the endothelial membrane surface, leading to the block of the membrane association of GTP-Rac or its fast removal from the membrane. A third mechanism ensues from work by Del Pozo and colleagues, which shows that adhesion-dependent control of Rac activity lies in the capacity of integrins to stabilize raft CEMM domains at the membrane upon adhesion on fibronectin. Rafts are detergent-insoluble membrane microdomains containing lipids in a liquid-ordered phase, in apposition to the liquid-crystalline fluid phase of the rest of the membrane. These domains also contain GPI-anchored proteins, acylated proteins and occasionally caveolin. GTP-Rac has high affinity for rafts and anchors in the membrane at these sites (del Pozo et al., 2004). When cells are placed in suspension, rafts are endocytosed and the membrane is thereby cleared of potential Rac binding sites (del Pozo et al., 2004; Del Pozo et al., 2002). Our observations could be in favor of a balance between formation of cholesterol- or caveolin-enriched and tetraspanin-enriched membrane domains, depending on which β1 integrin heterodimer is ligated by the surrounding extracellular matrix. Interestingly, TEM domains are known to be distinct in composition from rafts (Charrin et al., 2009; Hemler, 2005; Yanez-Mo et al., 2009). We could hypothesize that stimulated α2β1 integrins, together with the CD9 tetraspanin, organize alternative detergent-resistant membrane microdomains. Each domain could have a different impact on the composition of the local membrane and the association of Rac. Future extensive investigations in this direction will help to establish whether α2β1 integrin and CD9 modify the plasma-membrane environment to regulate adhesion-dependent signaling and cell-cycle control.

Materials and Methods

Cell culture and transfection

Primary human vein endothelial cells (HUVECs) were cultured as described (Mettouchi et al., 2001). All experiments were carried out with G0-synchronized cells in a serum-free human endothelial SFM medium (Invitrogen) containing 20 ng/ml bFGF, 10 ng/ml EGF, 1 μg/ml heparin, 5 μg/ml insulin, 5 μg/ml transferrin, 5 μg/ml selenium and 4.7 μg/ml linoleic acid (ITS+1, Sigma), referred to as ‘complete SFM’. Cells were electroporated at 300 V, 450 μF with 20 μg plasmids pCDNA3-EGFP, pCDNA3-CD9-EGFP and pSuper-CD9 (GAAGGACGTACTCGAAACC). In the latter case, 4 μg pDsRed2-mito (Clontech) was co-transfected as a marker. For RNAi experiments, the ON-TARGETplus SMARTpool siRNA L-004566, L-004571, L-007214 (Dharmacon) respectively targeting α2, α3 and α6 integrin subunits, RNAi against Cd9 (GAGCAUCUUCGAGCAAGAA) or luciferase as a control (CGUACGCGGAAUACUUCGA) were used at 100 nM using Lipofectamine-2000 transfection reagent (Invitrogen) or the magnetofection technology (OZ Biosciences) following the manufacturer's instructions. Human plasma fibronectin, Matrigel, rat collagen I and mouse EHS-laminin-1 were purchased from BD-Bioscience or Upstate; human laminin was from AbD Serotec. All proteins were used at 5 μg/cm2 unless otherwise specified. In mixed matrices, Fn was kept constant at 5 μg/cm2. Ratios indicate the relative amount of each molecule, based on their molecular mass.

Cell-biology techniques

Immunofluorescence was performed done on paraformaldehyde-fixed cells, permeabilized in 0.5% Triton X-100. FITC-phalloidin 1 μg/ml (Sigma) was used to visualize actin. The adhesive structures were stained with antibodies against Paxillin (BD-Transduction Laboratories), phospho-tyrosine (P-Tyr-100, Cell Signaling), vinculin (h-vin1, Sigma). The distribution of α2β1 and α5β1 integrins and CD9 was examined using AB1936, AB1928 (Chemicon) and ML13 (Pharmingen) antibodies. BrdU incorporation was performed on G0-synchronized cells plated for 24 hours; 10 μM BrdU was added in the last 5 hours and was immunodetected using the cell proliferation kit (Boehringer). Fluorescence analysis was performed on an Axiophot (Zeiss) or TCS-SP Confocal microscope (Leica). Image treatment (contrast adjustment, median filter) was performed with GIMP-2 software (GNU Image Manipulation Program). DNA content was determined upon staining cells with 40 μg/ml propidium iodide in 0.1% NP-40 and analysis on a BD FACScan™ flow cytometer.

Biochemical techniques

SDS-PAGE analysis was performed upon cell lysis in Laemmli buffer or Brij 58 buffer (1% Brij58, 10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM MgCl2, 1 mM CaCl2, 10 mM NaF, 1 mM Na3VO4, 1 mM AEBSF, 10 μg/ml aprotinin, leupeptin, pepstatin A). Immunoblots were done with the following primary antibodies and revealed by chemiluminescence (Millipore): Cyclin D1 (DCS-6, Neomarkers), FAK (BD-Transduction Laboratories), phosphorylated FAK (Y397-Biosource), tubulin (TUB2.1, Sigma), GAPDH (FL335, Santa-Cruz), CD9 (TS9, produced by E.R.), Rac (BD-Transduction Laboratories), α2 integrin (AB1936, Chemicon), α5 integrin (AB1928, Chemicon), Flotillin (BD-Transduction Laboratories), Transferrin receptor (H68.4, Zymed), Rho-GDI (A20, Santa-Cruz) and HA tag (HA-11, Babco).

Surface-association analysis was performed upon biotinylation of HUVEC surface proteins at 4°C for 30 minutes in 10 mM HEPES pH 7.3; 150 mM NaCl, 0.2 mM CaCl2, 0.2 mM MgCl2 containing 0.5 mg/ml Ezlink-Sulfo-NHS-LC-Biotin (Pierce). Cells were lysed in 1% Brij 58 buffer and processed for immunoprecipitation using antibodies against CD9 (TS9), α2 integrin (AK-7, Diaclone) or control IgG (DAKO). The CD9 complex was dissociated in 1% Triton X-100, 0.2% SDS for 2 hours at 4°C, and the presence of associated α2 integrin was assessed by subsequent immunoprecipitation, SDS-PAGE and immunoblotting using streptavidin-HRP (Pharmacia). Equilibrium density-gradient centrifugation was performed upon lysis for 30 minutes at 4°C of 7×106 cells in 0.7 ml of 1% Brij58 buffer. 120 μl was mixed with 240 μl OptiPrep™ (Sigma) to yield 40% sucrose. Then, 300 μl of lysate-sucrose mixture was sequentially overlaid with 600 μl of 30% sucrose in the same buffer without detergent and with 300 μl of detergent-free buffer. The tubes were centrifuged at 100,000 g for 2 hours at 4°C (S55S rotor, Sorvall-RC-M120GX). Fractions of 150 μl were collected from the top of the tube and resolved on SDS-PAGE. All steps were done at 4°C.

Recruitment of Rac to cellular membranes was measured on 4×106 cells adhered for 45 minutes. Cells were scraped in 5 ml cold PBS, pelleted and homogenized in 250 μl cold SI buffer (250 mM sucrose, 3 mM imidazole, pH 7.4, 1 mM PMSF). Cells were lysed by passing 40 times through a 25 G needle (U-100 Insulin, Terumo). Nuclei were removed by centrifugation for 10 minutes at 10,000 g at 4°C. Protein concentration of the post-nuclear supernatants (PNS) was normalized. PNSs were centrifuged for 1 hour at 100,000 g, 4°C. Supernatants correspond to the cytosolic fraction. Pellets, homogenized in an equal volume of SI buffer, correspond to membranes. 30 μl of each fraction was resolved by SDS-PAGE and blotted for Rac and Rho-GDI and transferrin receptor to control fractionation. When RNAi depletion was performed, HA-Rac (pKH3-HA-Rac) was co-transfected and fractions were blotted for HA. Rac-GTP affinity-purification experiments were performed as described (Mettouchi et al., 2001).

Yeast experiments

Yeast strain JD53 (MATα his3200 leu2-3,112 lys2-801 trp163 ura-52) was transformed using the classical lithium-acetate-PEG method with the pRS-313-α2-CRU vector to generate the JD53-α2CRU strain. The bait vector was derived from pRS-313-STE14-CRU (Wittke et al., 1999) by replacing the insert with the transmembrane and cytosolic portion of human α2 integrin (residues 1125-1181). JD53-α2CRU was transformed with a cDNA library constructed from HUVECs in proliferation, quiescent and sprouting states from 2.5 hours and 24 hour cultures and cloned as 3′ fusion to the Nub(A) region (Creator-SMART cDNA library construction kit, Clontech). Library screening was done with an efficiency of 106 c.f.u./ml in the presence of 1 mg/ml 5-FOA (Euromedex). Positive clones were confirmed upon independent transformation in the JD53-α2CRU strain, and culture drops were spotted upon serial 1:10 dilutions starting with OD 0.2 on increasing doses of 5-FOA as indicated in Fig. 4. Negative controls were the vector coding for Nub(A) alone or in fusion with the yeast Tom22 mitochondrial membrane protein, facing the cytosol.

In vivo angiogenesis

500 μl Matrigel (10 mg/ml, BD-Bioscience) containing 1 μg/ml bFGF (R&D Systems) was injected subcutaneously in the back of 8-week-old C57BL/6 Cd9−/− females and WT littermates. After incubation for 14 days, mice were sacrificed and the implants were paraffin embedded to proceed with Masson Trichrome staining of 8 μm sections. In parallel, sections were processed to detect endothelial cells by immunohistochemistry using the anti-endomucin antibody (7C7, gift from Dietmar Vestweber, Max-Planck-Institute of Molecular Biomedicine, Münster, Germany). Angiogenic scores were determined as follows: score 1, no cell colonization or slight peripheric coat around the plug; score 2, thickened cell coat with marked structural infiltration in the plug; score 3, deep and massive infiltration with erythrocyte presence. Scores were assessed blindly and verified by two manipulators. Animals were maintained and handled according to the regulations of the European Union and the French Department of Health and Agriculture.

Endothelial cell culture in 3D matrigel

3D cultures in matrigel were performed with 5×104 cells per well in eight-chamber Lab-Tek plates (Nunc) containing 5 mg/ml matrigel in complete SFM. The process was followed for 24 hours at 37°C. Selection of cells depleted of CD9 was performed using pan mouse IgG-specific magnetic beads (Dynal). Briefly, negative sorting was realized upon 4°C incubation of the cells with the ML13 antibody against CD9 followed by anti-mouse IgG-coated beads (four beads per cell). The supernatant (CD9-negative cells) was collected after retaining the CD9-positive cells on a magnet. Levels of CD9 depletion were controlled in each individual experiment. Control cells were incubated with non-immune antibody and subjected to the same treatment.

FACS analysis of surface-protein expression

Cells were detached and incubated with antibodies against α2 (AK-7), α3 (ASC-1), α6 (GOH3) integrins or against CD9 (ML13), followed by secondary FITC-conjugated antibody. Cells were then fixed in 1% formaldehyde and mean fluorescence intensities were determined on a BD FACSCalibur flow cytometer.

ELISA detection of deposited fibronectin

Plates were coated with a fixed amount of fibronectin, either alone or in combination of increasing doses of Lm-1, yielding different Fn:Lm molecular ratios. The amount of Fn effectively coated on the tissue-culture surface was quantified by ELISA using anti-human fibronectin polyclonal antibody (kind gift from Gertraud Orend, INSERM U682, Strasbourg, France) and secondary HRP-coupled antibody. After washing, wells were incubated in a solution of 0-phenylenediaminedihydrochloride substrate (Sigma) for 5 minutes in the dark. Color yields were determined at 490 nm in an ELISA reader.


We wish to thank R. Agami, A. Scherberich, H. Prats, G. Orend and D. Vestweber for generous gifts of material; F. Prodon of the C3M Cell Imaging Facility, M. Billard, A. Duvivier, A. Loubat, S. Paquier-Isnard, J. Pouysségur and L. Gagnoux for contributing to the completion of the project and members of the laboratory for critical reading of the manuscript. This work was supported by institutional funding from INSERM, CNRS to A.M., grants from the Association pour la Recherche sur le Cancer (ARC 4478 & 4906), the Agence Nationale de la Recherche (ANR-A05135AS, ANR 0378 & ANR-RPV07055ASA), fellowships from INSERM to S.E. and Ligue Nationale contre le Cancer to R.T.


  • Accepted April 7, 2010.


View Abstract