Endoglin controls cell migration and composition of focal adhesions: function of the cytosolic domain.

Mutations in the human endoglin gene result in hereditary hemorrhagic telangiectasia type 1, a vascular disorder characterized by multisystemic vascular dysplasia, arteriovenous malformations, and focal dilatation of postcapillary venules. Previous studies have implicated endoglin in the inhibition of cell migration in vivo and in vitro. In the course of studies to address the relationship of the conserved cytosolic domain to endoglin function, we identified zyxin, a LIM domain protein that is concentrated at focal adhesions, as an interactor with endoglin in human umbilical vein vascular endothelial cells. This interaction is localized within the 47-amino acid carboxyl-terminal cytosolic domain of endoglin, and maps within zyxin residues 326-572. The endoglin-zyxin interaction was found to be largely mediated by the third LIM domain of zyxin, and is specific for endoglin because the homologous cytosolic domain of the transforming growth factor-beta type III receptor, betaglycan, fails to interact with zyxin. Expression of endoglin is associated with reduction of zyxin, as well as its interacting proteins p130(cas) and CrkII, from a focal adhesion protein fraction, and this reduction is correlated with inhibition of cell migration. We also show that endoglin-dependent: (i) inhibition of cell migration, (ii) reduction of focal adhesion-associated p130(cas)/CrkII protein levels, (iii) tyrosine phosphorylation of p130(cas), and (iv) focal adhesion-associated endoglin levels are mediated by the cytosolic domain of endoglin. These results suggest a novel mechanism of endoglin function involving its interaction with LIM domain-containing proteins, and associated adapter proteins, affecting sites of focal adhesion.

Endoglin is a 190-kDa homodimeric transmembrane glycoprotein composed of 95-kDa disulfide-linked subunits (1). Mutations in the gene encoding endoglin have been linked to the human disease: hereditary hemorrhagic telangiectasia type 1 (HHT1) 1 (2,3), an autosomal dominant inherited vascular dis-order characterized by localized vascular dysplasia and a tendency toward arteriovenous malformations. Arteriovenous malformations occur in ϳ20% of patients and are associated with life-threatening complications including stroke and brain abscess. Pathological features found in HHT telangiectases include focal dilatation of postcapillary venules and a prominence of actin stress fibers in pericytes (4 -6). Mice lacking endoglin die from defective angiogenesis characterized by failure of vascular smooth muscle investment of embryonic blood vessels, suggesting a defect in vascular smooth muscle cell development (7). The molecular mechanism underlying the angiogenic phenotype in the endoglin null mouse remains unclear.
In vivo data points to a role for endoglin in the vascular response to injury. Findings from this laboratory showed that endoglin was expressed in human aortic smooth muscle cells in atherosclerotic plaques but was absent from normal smooth muscle (8). Ma et al. (9) found that expression of endoglin in normal porcine coronary arteries was restricted to endothelial cells and adventitial fibroblasts, and a minority of medial smooth muscle cells. However, following angioplasty, endoglin was present on adventitial myofibroblasts and medial smooth muscle cells of porcine coronary arteries. These workers also found that antisense oligonucleotides to endoglin decreased its expression and antagonized the TGF-␤-mediated inhibition of smooth muscle cell migration in vitro, suggesting that endoglin may be required for modulation of smooth muscle cell migration by TGF-␤. In addition, overexpression of endoglin in fibroblasts (10) and prostatic cells (11) leads to an inhibition of cellular migration. These studies implicate endoglin in the control of cell migration within the context of the injured vessel wall and in embryonic development, although no molecular mechanism has been proposed to link endoglin expression to cell migration.
The two known TGF-␤ type III receptors, endoglin and betaglycan, are auxiliary components of the TGF-␤ receptor system that associate with the types I and II signaling receptors and modulate the cellular responses to TGF-␤ (12)(13)(14). Only limited information regarding the relationship of the structure of endoglin to its function is available. The cytosolic domains (CDs) of betaglycan and endoglin comprise 46 and 48 amino acid residues, respectively, are very similar bearing 71% homology at the protein level (12), and are highly conserved between mammalian species. Whereas the CDs of endoglin and betaglycan are both serine-and threonine-rich, and to similar extents (SϩT, 40%), the distributions of serine and threonine in the two CDs are quite distinct. The serine residues in the endoglin CD are more centrally clustered, whereas serine clustering in the betaglycan CD is more COOH-terminal. Despite these differences, potential similarities also exist. It has been suggested that the carboxyl-terminal three amino acids of endoglin (15) and betaglycan (16) resemble sequences that may bind class 1 PDZ domains. To date, proteins, other than the TGF-␤-receptors, that interact with endoglin and which could provide insight into the role of endoglin in vascular disease, vessel remodeling, and angiogenesis, have not been described. This paper is the first to examine the interaction of the LIMdomain protein zyxin, with endoglin, and describes functional consequences of endoglin expression including coordinate changes in the composition of sites of focal adhesion and cellular migration.

EXPERIMENTAL PROCEDURES
Plasmids and Retroviral Constructs-The CD of endoglin was amplified with primers designed to encompass amino acids 612-658 of human endoglin: 5Ј-AAGAATTCTACATCTACTCGCACACCCGT and 5Ј-CGGGATCCTATGCCATG-CTGCTGGTGGA, and cloned into pAS-2-1 (Clontech). The CD of endoglin was subcloned into pGEX4T3 (Amersham Biosciences). The CD of betaglycan was amplified with primers designed to encompass amino acids 805-849 of human betaglycan: 5Ј-GCTCGAGCTACATCTATTCTCACACA and 5Ј-CTAGGGCC-GTGCTGCTGCTGGA, and cloned into pGEX4T3. Zyxin constructs were amplified with the primers listed below and cloned into pMALC2 (New England Biolabs), engineered to contain an HA epitope between maltose-binding protein and the zyxin fragments. The zyxin forward primer (5Ј-GGAATTCCCATGGCGGCCCCCCGC) was designed to amplify beginning at amino acid 328 of the human zyxin cDNA. The zyxin reverse primers were as follows: LIM1, to amplify to amino acid 438: 5Ј-TCAAGTGTAACAGCCCTCGCA; LIM2, to amplify to amino acid 499: 5Ј-TCACTGCTTGTGGTAGTCGGG; and LIM3, to amplify to amino acid 572: 5Ј-TCAGGTCTGGGCTCTAGC. Human endoglin cDNA from pcEXVEndoL (17) was subcloned into pcDNA 3.1 (Invitrogen), the Tet-responsive pBI-EGFP vector (Clontech), and the retroviral vector pWZL (18). pWZL is based on the pBABE retrovirus (19), which uses an internal ribosomal entry site to drive hygromycin resistance. The -⌬C mutant of endoglin, lacking the cytosolic domain, and the -⌬SMA deletion mutant, lacking the 3 carboxyl-terminal amino acid residues, were constructed by creation of termination codons immediately after amino acid residues 611 or 655, respectively, by site-directed mutagenesis (QuikChange TM , Stratagene), and subsequently cloned into the pWZL retrovirus. Human zyxin cDNA was obtained in Bluescript from ATCC (clone number 1016104), and was subcloned into pcDNA 3.1, pEGFP-C1 (Clontech) engineered to contain an HA epitope upstream of zyxin. All constructs were confirmed by DNA sequence analysis.
Yeast Two-hybrid Construction and Screening-The Matchmaker Two-hybrid system 2 (Clontech) was used to isolate proteins that physically interact with the cytosolic domain of endoglin. pAS2-END-CD was transfected into yeast and used to screen a human lung cDNA library constructed in the GAL-4 AD vector, pACT2 (Clontech). Potential protein interactions between the DNA-BD/endoglin and the AD/ lung library proteins were identified by cotransfecting both constructs into yeast (CG-1945) followed by determination of their ability to grow on medium lacking histidine. Interactions in His ϩ clones were confirmed by assaying for ␤-galactosidase activity. The identities of the potential proteins that interact with the endoglin CD were determined by DNA sequence analysis.
GST Binding Analysis-Zyxin in pcDNA3.1 was transcribed and translated in vitro in the reticulocyte lysate programmed with T7 polymerase (Promega) in the presence of [ 35 S]methionine according to the manufacturer's instructions. For binding assays, GST-CD fusion proteins or GST alone (1.5-5 g) were incubated with radiolabeled zyxin in 200 l of GST binding buffer (10 mM Tris-HCl, pH 7.5, 100 g/ml bovine serum albumin (BSA), 150 mM NaCl, 1 mM ZnCl 2 , and 1 mM MgCl 2 ). After washing with binding buffer, proteins were eluted in SDS gel loading buffer (0.2% SDS, 20% glycerol, 0.2% bromphenol blue, and 100 mM Tris, pH 6.8) and separated by SDS-PAGE. Radiolabeled proteins were visualized using a phosphorimager (Typhoon 8600, Amersham Biosciences) following exposure of the fixed dried gel to phosphorscreens for 1 h. Bands corresponding to zyxin were quantified using ImageQuant software version 5.2 (Amersham Biosciences). For GST pull down reactions from cells, GST-protein-containing bacterial lysates were diluted in GST binding buffer (1ϫ phosphate-buffered saline (PBS), 2.5 mM MgCl 2 , 2 mM dithiothreitol) and added to human embryonic kidney 293 (HEK 293) cell lysates obtained as described below. Proteins were electrotransferred to polyvinylidene fluoride membranes (Millipore Corp.) and immunoblotted using the indicated antibodies.
Construction of Tetracycline-inducible Cells-GM7372 cells were used to prepare stable Tet-responsive cell lines expressing the VP16Tet activator protein. Cells were transfected with the pTet-On vector (Clontech) followed by selection with G418 according to the manufacturer's recommendations. Isolated tetracycline-resistant clones were expanded and tested for inducibility by transient transfection with the pTREluciferase vector followed by assay for luciferase activity in the presence or absence of doxycycline (Sigma, 1 g/ml). Doxycycline-responsive GM7372 clones were used for subsequent transfection with endoglincontaining Tet-responder plasmids. GM7372 Tet-responder cells transfected with the pBI-endoglin construct were selected using hygromycin-B. Clones were expanded and tested for Tet-dependent induction of endoglin by Western blotting. Doxycycline-inducible endoglin expressing clones (GM7372-EL) were further subjected to Western blotting to estimate endoglin levels at maximal induction (1 g/ml doxycycline) and the doxycycline dose dependence of induction (data not shown).
Construction of Retroviral-transduced Cell Lines-Subconfluent BOSC23 cells (20) were transiently transfected with either control pWZL, or pWZL bearing -FL, -⌬C, or -⌬SMA endoglin constructs. The transfection medium was changed 16 -24 h later, and virus was harvested 48 h post-transfection, filtered through 0.4-m low protein binding filters (Millipore Inc.), and used immediately or stored at Ϫ70°C. C3H10T1/2 cells were plated at 50% confluency in 6-well culture dishes, and the following day the medium was removed and replaced with 1.5 ml of DMEM containing 10% FBS, 0.5-1.0 ml of the retrovirus preparation, and 8 g/ml Polybrene (Sigma). Following incubation with virus for 24 h, the medium was removed and replaced with DMEM containing 10% FBS and 400 g/ml hygromycin-B. Selection and expansion over the course of 10 -14 days yielded a pooled polyclonal cell population. Immunofluorescence analysis was used to confirm the efficiency of retroviral gene expression, which was found to be greater than 95% in all cases (data not shown).
Antibodies, Western Blot, and Co-immunoprecipitation Analysis-Anti-GST antibody was obtained from Amersham Biosciences. Anti-HA, anti-vinculin, anti-zyxin (goat, C19, and rabbit, H-200), anti-EGFP, and anti-p130 cas antibodies were obtained from Santa Cruz Inc. Anti-endoglin P4A4 monoclonal antibody was obtained from the Developmental Studies Hybridoma Bank (NICHD, National Institutes of Health). Anti-endoglin monoclonal clone 35, and anti-CrkII antibodies were obtained from Transduction Laboratories. Anti-focal adhesion kinase (FAK) antibody was obtained from Sigma. Anti-phosphotyrosine peroxidase-labeled antibody (RC20:PO) was obtained from BD Biosciences. Primary antibodies were used at the dilutions recommended by the manufacturer in Tris-buffered saline containing 1% Tween 20 (TBST) and 5% nonfat dry milk. The phosphotyrosine antibody conjugate RC20:PO was used in TBST containing 5% BSA. Horseradish peroxidase-labeled antibodies were visualized by ECL chemiluminescence (Amersham Biosciences).
For Western blot analysis, cells were harvested by trypsinization, washed three times with PBS, and lysed in extraction buffer (20 mM Tris-HCl, pH 7.5, 0.15 M NaCl, 1% Nonidet P-40) containing protease inhibitor mixture (Roche Diagnostics) according to the manufacturer's instructions. For co-immunoprecipitation studies cell culture plates were washed three times with ice-cold PBS and lysates were prepared using a modified radioimmunoprecipitation analysis (RIPA) buffer (150 mM NaCl, 50 mM Tris-HCl, pH 7.5, 500 M EDTA, 100 M EGTA, 1.0% Triton X-100, and 1% sodium deoxycholate) containing protease and phosphatase inhibitors (Phosphatase Inhibitor Mixture Set II, Calbiochem).
Analysis of p130 cas tyrosine phosphorylation was conducted essentially as previously described (23) with the following modifications. Cells were starved for 24 h in serum-free DMEM containing the soluble form of the TGF-␤ type II (TGF-␤R:Fc) receptor (Ref. 24; 100 ng/ml, R & D Systems). Following three washes with serum-free DMEM, cells were either harvested (untreated) or stimulated with 2 ng/ml TGF-␤1 (R & D Systems) for 10 min. Culture plates were washed three times with ice-cold PBS, and lysates were prepared using RIPA buffer (150 mM NaCl, 50 mM Tris-HCl, pH 7.5, 500 M EDTA, 100 M EGTA, 0.1% SDS, 1% Triton-X100, and 1% sodium deoxycholate) containing protease and phosphatase inhibitor mixtures.
Lysates were centrifuged at 14,000 ϫ g for 5 min, and aliquots of cleared cell lysates containing the same amount of protein were incubated with the indicated antibodies for 3 h at 4°C, followed by incubation with Protein A/G Plus-Sepharose (Santa Cruz) for 1 h at 4°C. The immune complexes were washed three times with lysis buffer, eluted by boiling for 5 min in SDS gel loading buffer, and analyzed by SDS-PAGE. Proteins were electrotransferred to polyvinylidene fluoride membranes (Millipore Corp.) and immunoblotted using the indicated antibodies. Quantitation of bound peroxidase-labeled anti-rabbit antibody and peroxidase-labeled anti-phosphotyrosine antibody, RC20:PO, was accomplished by phosphorimager analysis of ECL chemiluminescence.
Immunofluorescence and Confocal Microscopy-Cells were seeded on 25-mm fibronectin-coated glass coverslips (BioCoat, BD Biosciences) and grown to 50 -70% confluency. Cells were transfected with EGFPzyxin, and either empty vector or pcDNA containing -FL or -⌬C endoglin. Following an additional incubation for 48 h, cells were washed twice with 1ϫ PBS and fixed with 4% paraformaldehyde for 8 min. Coverslips were washed with PBS and incubated in blocking solution (5% BSA, 0.1% Triton X-100, 0.1% Tween 20 in PBS) for 2 h at room temperature. Primary anti-human endoglin antibody (SN6h, Dako), diluted 1:200 in blocking solution, was added to slides and then incubated for 1 h at room temperature. Cells were washed twice with 1ϫ PBS, and antimouse Cy3 conjugate, diluted 1:1000 in blocking solution, and incubated with the cells for 30 min. After a final 1ϫ PBS wash, coverslips were mounted on glass slides using Aquamount (Lerner Laboratories) and observed using the LTCS-SP confocal system with an inverted DMIRBE microscope using a ϫ100 objective (Leica).
Affinity Purification of Focal Adhesion-associated Proteins-Tosylactivated magnetic microspheres (4.5 m diameter; Dynal) were coated with an RGD-containing peptide (Biomol) as recommended by the manufacturer. Cells were plated at high density (10 6 cells per 10-cm plastic culture dishes). Twenty-four hours later, cells were dispersed with trypsin-EDTA, washed twice in 1% BSA/DMEM, placed in polypropylene tubes (Costar), suspended (1 ϫ 10 6 cells/ml) in BSA/DMEM containing magnetic RGD-labeled microspheres (2 ϫ 10 7 /ml), and rotated for 30 min at 37°C. For experiments involving cytokine-dependent effects on focal adhesion proteins, TGF-␤1 and activin A (R&D Systems) were used at 2 ng/ml. Microspheres and bound cells were isolated by magnetic collection and suspended in ice-cold cytoskeleton extraction buffer (CSK-EB, 50 mM NaCl, 300 mM sucrose, 3 mM MgCl 2 , and 10 mM PIPES, pH 6.8) in 2-ml Eppendorf tubes (25). All subsequent procedures were carried out on ice. The magnetic bead pellet was transferred to freshly prepared complete CSK-EB (CSK-EB containing 0.5% Triton X-100 and protease and phosphatase mixtures), sonicated with 5 pulses (VWR Scientific, output setting, 1.3; duty cycle, 30) from a Branson sonifier. The microspheres were magnetically collected and washed 5 times with complete CSK-EB. Proteins in the bead complexes and total cell lysates (TCL) were dissolved in gel loading buffer, separated by SDS-PAGE, and transferred to polyvinylidene difluoride membranes using an electroblotting system (Bio-Rad).
Migration Assays-Cells were harvested by light trypsinization and washed twice in serum-free DMEM containing 200 g/ml BSA and counted using a hemacytometer (AO Scientific Instruments). Final volumes were adjusted so that all cell preparations were at final concentrations of 10 6 cells/ml. Cells (40,000) were then added to the upper chamber of a 48-well microchemotaxis chamber (Neuro Probe) and allowed to migrate from 200 g/ml BSA (upper chambers) to either 200 g/ml BSA as a migration negative control, or DMEM containing FBS at the indicated concentrations. Following incubation for 5 h at 37°C and at 5% CO 2 , migrating cells were fixed in 50% aqueous methanol for 5 min and then stained with PBS containing 5 g/ml propidium iodide (Calbiochem). Migrating cells were quantitated by phosphorimager scanning of the cell-associated propidium iodide fluorescence. Scanning was conducted using a 532-nm excitation laser. Fluorescence was detected at 580 nm using a 30-nm band pass filter. Each measurement is the average of at least 6 microchemotaxis chamber wells.

Zyxin Interacts with the Cytosolic Domain of Endoglin-In
attempts to elucidate the function of endoglin by identifying proteins that interact with it, we screened a human lung cDNA library using the endoglin cytosolic domain as bait. Approximately 2 ϫ 10 5 yeast transformants were screened, and His ϩ ␤-galactosidase ϩ clones were identified for further study. DNA sequence analysis identified the COOH-terminal region of the LIM domain-containing phosphoprotein, zyxin, corresponding to amino acid residues 326 -572, as a potential interactor.
In vitro affinity binding was used to characterize the specificity of the endoglin-zyxin interaction. Full-length zyxin in pcDNA3.1 was used to produce zyxin mRNA and 35 S-labeled zyxin protein using the reticulocyte-coupled transcription-translation (TNT) system. Crude bacterial lysates containing the GST-endoglin CD or GST-betaglycan CD fusion proteins, or GST only, were incubated with the TNT lysate. A clear dose response of 35 S-labeled zyxin protein bound to the GST affinity resin in terms of GSTendoglin CD added (Fig. 1A, histogram bars 2-4). The histogram bars corresponding to no added GST (bar 1), GST-betaglycan CD (bars 5-7), or empty GST vector (bars 8 -10) produced only background binding. The inset panel in Fig. 1A shows the result of immunoblot analysis of constant volumes (5 l) of GST proteins using anti-GST antibody to confirm equivalence of the protein levels corresponding to GST, endoglin-GST, and betaglycan-GST used in the binding reactions. These results indicate that the interaction of zyxin with the endoglin CD is markedly greater than with the betaglycan CD.
GST pull down assay of transfected full-length zyxin (GFP-HA-zyxin) in pcDNA3.1 was conducted in HEK 293 cells using the GST-endoglin CD and betaglycan CD fusion proteins. HEK 293 cells were chosen for this experiment because they express negligible levels of endogenous zyxin by Western blot analysis (data not shown). Western blot analysis of the binding reactions confirms binding of zyxin by the GST-endoglin CD construct (Fig. 1B, lane 1). Interactions between zyxin and GSTbetaglycan CD (lane 2) or GST alone (Lane 3) were not observed under these conditions. These results confirm the specificity of the interaction for the endoglin CD as compared with that of betaglycan. HUVEC were next used as a source of endogenous endoglin (26) and zyxin (27) for co-immunoprecipitation. Endoglin co-immunoprecipitated with zyxin, as shown in the upper panel of Fig. 1C, using an anti-zyxin antibody (immunoprecipitation, anti-zyxin; Western blot, anti-endoglin) thus confirming that the endoglin-zyxin interaction occurs between endogenous proteins in a primary endothelial cell.
To further investigate the requirement of the cytosolic domain of endoglin for the interaction of endoglin with zyxin in intact cells, retrovirally infected C3H10T1/2 cells bearing fulllength (-FL) endoglin and a carboxyl-terminal deletion mutant lacking the entire cytosolic domain (Ϫ⌬C) were constructed. In addition, we constructed a truncated endoglin mutant lacking the carboxyl-terminal three amino acids because these residues have been proposed to constitute a PDZ liganding motif in betaglycan (-⌬SMA (15,16)). Fig. 1D (upper panel) shows coimmunoprecipitation of endogenous zyxin in -FL and -⌬SMA containing cells. The interaction with -FL endoglin, evident in lane 1, is weaker that in cells expressing the -⌬SMA mutant (lane 2), suggesting that the truncated endoglin CD may possess higher affinity for zyxin. Total zyxin levels (middle panel, Fig. 1D) did not change appreciably, indicating that the difference in binding between -FL and -⌬SMA is not likely to be attributable to a change in levels or stability of zyxin. -⌬C endoglin-infected cells did not permit co-immunoprecipitation of zyxin, although -⌬C expression levels were comparable with -FL and -⌬SMA (lower panel, Fig. 1D), confirming the requirement for the cytosolic domain, but not the carboxyl-terminal SMA sequence, for zyxin binding.
Wild-type, but Not Cytosolic Domain-deleted Endoglin Colocalizes with Zyxin in Vitro-GM7372 endothelial cells were next transfected with EGFP-zyxin, and empty vector control, or constructs containing -FL or -⌬C endoglin. Following fixation and staining with anti-endoglin antibody, cells were imaged by scanning confocal microscopy. Cells cotransfected with EGFP (Fig. 2, panels A1-C1) showed diffuse cytoplasmic and nuclear staining with no evidence of labeling at sites of focal adhesion. In contrast, cells transfected with EGFP-zyxin (Fig. 2, panels A2-C2) show staining in distinct sites of focal adhesion (arrows, panels A2 and B2). This pattern of EGFP-zyxin fluorescence is consistent with previous reports of zyxin subcellular localization (28, 29) within sites of focal adhesion. Immunofluorescent staining for endoglin (panels A3-C3) produced no detectable signal in the control pcDNA vector transfectant (panel A3). In contrast, cells transfected with -⌬C endoglin (B1-4) exhibit a strong signal for endoglin (B3) but show only spurious colocalization as indicated in the merged image shown in Fig. 2, panel B4. Cells transfected with -FL endoglin (C1-4) show comparable immunofluorescence arising from transfected endoglin (C3) compared with that observed for -⌬C endoglin (B3). In contrast to the -⌬C endoglin-expressing cells, however, distinct colocalization can be observed in panel C3 and the merged image shown in panel C4 (C2-C4, arrows). The strongest regions of colocalization appear to be in ruffle-like structures along edges of the cell. Well defined sites of focal adhesion, as seen in panels A2 and B2, are less evident in the endoglin-EGFP-zyxin co-transfectants shown in C2, suggesting that zyxin undergoes redistribution from focal adhesion sites to ruffle-like structures.
Zyxin LIM Domains Interact with the Endoglin Cytosolic Domain-To further refine the structural basis of the interaction between endoglin and the zyxin-(326 -572) fragment, we examined the contributions of the LIM subdomains of zyxin to endoglin binding. HA-tagged deletion constructs included the nuclear export signal and extended through the LIM1, LIM2, or LIM3 domains of zyxin, as shown schematically in Fig. 3A. As shown in Fig. 3B, affinity binding of the GST-endoglin CD was significantly reduced upon loss of the zyxin LIM3 domain. The lowermost panel of Fig. 3B shows that removal of both LIM2 and LIM3 domains eliminates detectable interaction with zyxin. The righthand panel of Fig. 3B provides confirmation of the equivalence of expression of the various zyxin LIM domain deletion mutants in total cell extracts. These data indicate that the LIM1 domain is not sufficient to bind endoglin. Additionally, endoglin binding and specificity with respect to betaglycan CD binding resides primarily in the zyxin LIM2-3 domain with a larger contribution by the LIM3 domain.
Endoglin Expression Inhibits Cell Migration and Alters the Composition of Sites of Focal Adhesion-Because zyxin has been shown to localize to sites of focal adhesion (28 -30) and to undergo subcellular redistribution during cell migration (31), we examined the coordinate effects of endoglin expression on the composition of sites of focal adhesion and cell migration. RGD-labeled magnetic microspheres provided an approach to biochemical characterization of sites of focal adhesion (25) and were used to induce and isolate RGD binding, integrin-rich sites of focal adhesion from viable cells. Doxycycline-inducible GM7372-EL cells were grown to confluence and subsequently incubated in the absence or presence of doxycycline for 24 h, harvested by brief trypsinization, and incubated with either RGD-or control RGE-labeled magnetic microspheres. Western blot analysis, shown in the upper panel of Fig. 4A, demonstrates that vinculin, a focal adhesion-associated protein (32)(33)(34), is equivalently isolated by RGD microspheres (lanes 1 and  3) irrespective of doxycycline treatment, and therefore endoglin expression. Control RGE-labeled magnetic microspheres (lanes 2 and 4) show little evidence of bound vinculin, confirming that the RGD-labeled magnetic microspheres are highly selective for RGD-binding focal adhesion sites. Analysis of the bound material for ␤-actin, as compared with total cellular ␤-actin, shown in the middle and lower panels of Fig. 4A, respectively, indicates that there is negligible cytosolic or microfilament contamination of this fraction. These results confirm that RGDlabeled magnetic microspheres provide a cellular fraction from doxycycline-treated or untreated cells that is enriched in focal adhesion-associated protein, as previously reported (25).
To determine the effect of endoglin expression on zyxin sub- labeled microspheres at the 0 -10 ng/ml levels of doxycycline indicating that, like vinculin, expression of endoglin has little effect on FAK localization to the focal adhesion protein fraction. The slight decrease in microsphere-bound FAK at the highest level of endoglin induction (as shown in Fig. 4, panel B) suggests that, under these conditions, there may be a decrease in the level of focal adhesion-associated protein captured. The bottom panel of Fig. 4B indicates that total cell lysate zyxin levels do not appreciably change as a result of induction of endoglin synthesis with doxycycline. This result confirms, as discussed above (Fig. 1C), that total zyxin levels are constant and show no evidence of down-regulation or accelerated protein turnover, and provides biochemical evidence that endoglin expression causes dose-dependent reduction of zyxin from newly formed sites of focal adhesion.
Zyxin interacts with p130 cas , a protein that is also associated with focal adhesions in a complex with CrkII (28). Therefore, to investigate the possible downstream functional consequences of endoglin-dependent subcellular localization of zyxin, we determined whether endoglin induction altered the localization of endogenous p130 cas . For this experiment, GM7372-EL cells were transfected with EGFP-HA-zyxin as a positive control for endoglin-dependent subcellular localization of zyxin and in one set of cells, endoglin expression was induced with doxycycline. As shown in the upper inset of Fig. 4C, EGFP-HA-zyxin was reduced in the focal adhesion-associated fraction in doxycycline-treated cells, consistent with the findings presented in Fig. 4B. Western blot analysis for p130 cas , shown in the second panel of Fig. 4C, also shows evidence of loss from the focal adhesion fraction. Western blot analysis of the total cell lysate, using anti-HA and anti-p130 cas antibodies, shown in the third and fourth panels of Fig. 4C, respectively, indicates that total EGFP-zyxin and p130 cas levels did not change appreciably upon doxycycline treatment. Western blot analysis of the RGDbound fraction, using anti-FAK antibody shown in the lower panel of Fig. 4C, confirms equivalent levels of protein in this fraction. These results indicate that expression of endoglin mediates coordinate reduction of both zyxin and endogenous p130 cas in the RGD-bound protein fraction.
Finally, GM7372-EL cells treated identically to those used for Fig. 4C were analyzed for their ability to migrate in a microchemotaxis chamber using serum as the stimulus, in the presence or absence of endoglin induction by doxycycline. As seen in Fig. 4D, GM7372-EL cells induced to express endoglin FIG. 4. Effect of endoglin expression on focal adhesion-associated proteins and cellular migration. A, affinity magnetic enrichment of focal adhesion proteins. RGD, or negative control RGE peptide-labeled magnetic microspheres were used to purify RGD-binding focal adhesion-associated protein fractions from doxycycline (Dox)-treated and untreated GM7372-EL cells. Vinculin is used as a marker for focal adhesion-associated protein and ␤-actin for purity of this fraction. B, GM7372-EL cells were transiently transfected with human zyxin in pcDNA3.1. Different levels of endoglin expression were induced 24 h later with the indicated levels of doxycycline. Following an additional 24 h, focal adhesion-associated protein was isolated using RGD-labeled magnetic microspheres. C, GM7372-EL cells were transiently transfected with EGFP-HA-zyxin. Endoglin expression was induced 24 h later with the indicated levels of doxycycline. After an additional 24 h, focal adhesionassociated protein was isolated using RGD-labeled magnetic microspheres. RGD microspheres (RGD Beads) and TCL were analyzed for transfected EGFP-HA-zyxin and endogenous p130 cas using the indicated antibodies. Anti-FAK antibody was used to confirm the equivalence of this protein bound to microspheres. D, cellular migration of GM7372-EL doxycycline-inducible endoglin cell lines. Cells were washed twice in 200 g/ml BSA in serum-free DMEM and allowed to migrate for 5 h from 200 g/ml BSA (upper chambers) to either 200 g/ml BSA (negative control), or 10% FBS (n ϭ 12 wells for each point). Migrating cells were fixed, and stained with propidium iodide. Migrating response, expressed as the relative fluorescence of the propidium iodide was measured at 580 nm, is plotted in the y axis. Error bars represent the mean Ϯ S.E. WB, Western blot. exhibit a reduced migration response to serum as the stimulus. These results indicate an association between endoglin expression, loss of p130 cas from newly formed sites of focal adhesion, and reduction in cellular migration.

Endoglin-dependent Inhibition of Cell Migration and Alteration of Focal Adhesion Composition Requires the Endoglin
Cytosolic Domain-Next, we examined whether changes in focal adhesion composition and cell migration required an intact endoglin CD. Stable -FL and -⌬C endoglin-expressing C3H10T1/2 cells were examined for their ability to migrate as described above. Fig. 5A shows that the absolute level of migration was found to be dependent on the level of serum used as the stimulus, -FL endoglin expression consistently inhibited migration of C3H10T1/2 cells, relative to control cells. In contrast, the -⌬C endoglin mutant had no effect on cell migration using either 1 or 3% serum as the stimulus (Fig. 5A). Western blotting of lysates from cells used for migration analysis (Fig.  5B) indicates that the levels of -FL and -⌬C endoglin were approximately equivalent. These results demonstrate that the endoglin cytosolic domain is necessary for inhibition of cell migration.
C3H10T1/2 cells expressing -FL and -⌬C endoglin were next used to assess the levels of p130 cas and CrkII collected on RGD-labeled microspheres. The top panel of Fig. 5C shows that, consistent with the data shown in Fig. 4, p130 cas is reduced in the -FL endoglin focal adhesion-associated protein fraction as compared with the levels of p130 cas seen in the total cell lysate. In contrast, the -⌬C-endoglin-derived focal adhesion protein fraction shows no reduction in p130 cas , indicating a requirement for the endoglin cytosolic domain for reduction of p130 cas in the focal adhesion-associated protein pool. Similarly, the p130 cas -interacting protein CrkII is reduced or absent from the -FL endoglin-, but not in the -⌬C-endoglin-derived focal adhesion-associated protein fraction, as compared with the total cell lysate levels of CrkII (Fig. 5C, middle two panels). The lower two panels of Fig. 5C demonstrate equivalence of protein loading for both the RGD-labeled microsphere, and total cellular protein, using anti-FAK and anti-endoglin antibodies for these fractions, respectively. These data show that inhibition of C3H10T1/2 cell migration, and reduction of the levels of p130 cas and CrkII in the focal adhesion-associated protein fraction are both correlated with endoglin expression, and are endoglin CD-dependent.
Because tyrosine phosphorylation of p130 cas is mediated by TGF-␤1 (23), we next investigated whether endoglin, a modulator of TGF-␤ signaling, affected tyrosine phosphorylation of p130 cas . To do this, C3H10T1/2 cells were first serum-starved in the presence of recombinant soluble TGF-␤R:Fc receptor to reduce endogenous levels of TGF-␤ ligands (24). As shown in Fig. 6A, Western blot analysis of control, -FL, or -⌬C endoglinexpressing C3H10T1/2 cells for p130 cas -associated tyrosine phosphorylation indicated that p130 cas tyrosine phosphorylation was reduced in -FL, but not control or -⌬C endoglin expressing cells (Fig. 6A, upper panel). Total 130 cas was essentially unchanged (Fig. 6A, lower panel) indicating that there is no appreciable degradation of p130 cas under these conditions. Tyrosine phosphorylation of p130 cas was weakly induced by TGF-␤1 treatment, and this induction appeared to be substan-

FIG. 5. The endoglin cytosolic domain is required for endoglin-dependent inhibition of migration.
A, microchemotaxis chamber assay of C3H10T1/2 endoglin transduced cell lines. C3H10T1/2 cell lines bearing control vector pWZL, -FL, or -⌬C endoglin were examined for their ability to migrate using the modified microchemotaxis chamber. Cells were allowed to migrate for 5 h from 200 g/ml BSA (upper chambers) to either 1 or 3% FBS (n ϭ 6, bars 1-3 and 4 -6, respectively) in the lower chambers of the migration apparatus. Bar 7 represents the average for the BSA-containing lower chambers (n ϭ 12) used as a negative control for migration. Migrating cells were fixed and stained with propidium iodide. The migrating response was determined as above and plotted on the y axis. Error bars represent mean Ϯ S.E. B, Western blot of cells used in A using anti-endoglin antibody (␣-END) and anti-␤-actin antibody (␣-␤ Actin) as a gel loading control. C, identically treated cells were harvested and used for isolation of focal adhesion-associated protein. The magnetic RGD-labeled microsphere-bound (RGD), and TCL fractions were subjected to Western blot (WB) analysis using anti-CrkII, anti-p130 cas , anti-FAK, and anti-endoglin antibodies. tially unaffected by endoglin expression (Fig. 6B). These data indicate that endoglin expression reduces the overall level of p130 cas tyrosine phosphorylation in a cytosolic domain-dependent fashion, but that this effect does not appear to be substantially mediated by TGF-␤1.

Deletion of the Endoglin Cytosolic Domain Increases Endoglin Concentration in Newly Formed Focal Adhesions-
The focal adhesion-associated protein fraction from C3H10T1/2 cells were next examined for binding of wild-type and mutant endoglin polypeptides (depicted schematically in Fig. 7A). These experiments indicate that -⌬C endoglin, shown in the upper panel of Fig. 7B, appears in the RGD-microsphere bound fraction, whereas -FL and -⌬SMA endoglin are much reduced in this fraction. Larger scale (2 ϫ 10 6 cells per lane) preparations of RGD-bound focal adhesion proteins did reveal -FL endoglin as a constituent of the RGD-bound fraction, albeit at much lower levels than seen for the RGD-bound -⌬C endoglin (Fig.  7C). Treatment of cells during RGD-microsphere binding with either TGF-␤1 or activin A produced no significant differences in the levels of either -FL or -⌬C RGD microsphere-bound endoglin (Fig. 7C). DISCUSSION In the present study we present data addressing functional roles for the endoglin CD. We show, for the first time, that endoglin-dependent inhibition of cell migration in vitro requires the endoglin CD. Second, we provide data indicating that the endoglin CD interacts with zyxin and mediates endoglin-depend-ent changes in the levels of p130 cas /CrkII in an RGD-bound, focal adhesion-associated protein fraction. Finally, we show that the endoglin CD governs the degree of association of endoglin with the focal adhesion-associated protein fraction. These data support a hypothesis whereby endoglin expression initiates a process that is mediated by a protein-protein interaction involving the endoglin CD and the LIM domains of zyxin, or possibly other related LIM-domain proteins. In this hypothesis, endoglin expression causes alteration of the levels of the interacting adaptor proteins, p130 cas and CrkII, within newly formed sites of cellular focal adhesion, and the overall level of p130 cas tyrosine phosphorylation. We propose that these changes contribute to the negative control of cell migration.
To date no cytosolic domain interactors for endoglin have been documented, although interactors with the endoglin-homologous cytosolic domain of betaglycan have been described (35). In the context of the TGF-␤ type III receptors, the interaction of zyxin with endoglin possesses potential functional specificity because zyxin does not interact with the betaglycan CD. This observation provides insights into the structural determinants of this interaction because endoglin and betaglycan share strong structural homology in their cytosolic domains (12). Conversely, we show that a region of structural similarity between endoglin and betaglycan, the putative carboxyl-terminal PDZ-domain binding motif (15), is not required for, and its absence may actually enhance, the interaction of endoglin with zyxin. Indeed, while the interaction with zyxin does not require the endoglin carboxyl-terminal SMA sequence, interactors with betaglycan do require the carboxyl-terminal PDZ-liganding motif (35). This result could be because of abrogation of binding of as yet unrecognized endoglin interactor proteins, or to changes in the phosphorylation of the cytosolic domain in the -⌬SMA construct.
The region of greatest dissimilarity between the endoglin and betaglycan CDs corresponds to the endoglin CD sequence RSPSKREPVVAVAAPASSESSSTN. Because this region of endoglin is rich in serine residues, it may be a site of regulation of the endoglin-zyxin interaction by serine phosphorylation (36). Endoglin CD phosphorylation is mediated by the TGF-␤ receptors (37) although other protein serine/threonine kinases have not been ruled out. Because bacterially expressed CDs used in this study are not phosphorylated, the interaction between zyxin and the endoglin CD can occur in the absence of phosphorylation, although the possibility of phosphorylation-dependent regulation of the affinity of the interaction with zyxin is not excluded. It will be important to determine the consequences of phosphorylation on the endoglin-zyxin interaction.
The interaction between endoglin and zyxin was initially suspected to be weak, or transient, because endoglin and zyxin do not appear to be fully colocalized as judged by confocal microscopy (Fig. 2, A2-C2). However, demonstration of co-immunoprecipitation in HUVECS using a stringent RIPA-based buffer suggests that the interacting proteins may populate structures that are not easily dissociated or isolated.
The region of zyxin found to interact with endoglin involves the second and third LIM domains of zyxin. Other proteins that have been shown to interact within the LIM region of zyxin include CRP2 (38,39), h-warts/LATS1 (40), and p130 cas (28). CRP2 binds to the LIM1 domain of zyxin (38,39), and h-warts/ LATS1 and p130 cas bind to LIM domains 1 and 2 (28,40). However, interactions have not yet been attributed to the zyxin LIM3 domain. It is therefore interesting to speculate that endoglin expression may alter the subcellular localization of multiple proteins, in addition to p130 cas and CrkII, via assembly of endoglin and zyxin into multiprotein complexes.
Both zyxin and zyxin-related protein (ZRP-1) interact with FIG. 6. Expression of -FL, but not -⌬C endoglin suppresses p130 cas tyrosine phosphorylation levels. A, C3H10T1/2 cells, grown as above, were serum-starved for 48 h and subsequently stimulated with 2 ng/ml TGF-␤1 for 10 min. Total cell lysates were prepared and used for immunoprecipitation of p130 cas followed by Western blot analysis using anti-phosphotyrosine pY20-peroxidase antibody (upper inset) and anti-p130 antibody (lower inset). B, phosphorimager analysis of Western blots (WB) of three separate experiments. Separate experiments were normalized for chemiluminescent detection of total p130 cas . The y axis is the ratio of the phosphorimager chemiluminescence of phosphotyrosine (pY) divided by the total p130 cas values (pY/ p130(Cas)). Error bars represent the mean Ϯ S.E. p130 cas (28). The present work describing endoglin-dependent loss of p130 cas from sites of focal adhesion provides evidence for a subcellular redistribution function for zyxin-p130 cas interaction. The ZRP-1/p130 cas interaction was mapped to the ZRP-1 LIM domains but appeared to depend on LIM domains 1-2 (28), whereas the endoglin-zyxin interaction appears dependent to a similar degree on LIM domains 2-3 raising the possibility that these interactions could form a multiprotein complex containing zyxin, p130 cas /CrkII, and endoglin. It is important to note that, whereas our data implicate zyxin in the mechanism by which endoglin expression inhibits cellular migration, the sole involvement of zyxin seems unlikely because the zyxin gene family is highly conserved, especially in the LIM domains, and appears redundant, as evidenced by the lack of phenotype in the zyxin-null mouse (41). Evidence that endoglin may interact with other LIM domain proteins will be discussed elsewhere, 2 but the observation that p130 cas interacts with either zyxin or ZRP-1 suggests the involvement of multiple related LIM domain proteins in endoglin function.
The adaptor protein, p130 cas , interacts with other focal adhesion-associated proteins including CrkII. Anchorage-dependent phosphorylation of p130 cas leads to its coupling to the adaptor protein CrkII and promotion of cell migration (42). These observations, taken together with dose-dependent and structure-dependent consequences of endoglin expression suggest a role for the p130 cas -CrkII complex in endoglin-dependent inhibition of cell migration. Based on these data, we suggest a mechanism whereby endoglin, in conjunction with zyxin or related proteins, causes reduction of the levels of p130 cas /CrkII in the nascent focal adhesion, negatively mediating cell migration. This mechanism does not appear to involve protein degradation because, in no case could we detect endoglin-dependent reduction in the levels of zyxin, CrkII, or p130 cas in the total cellular lysates. We were able, however, to detect reduction of tyrosine phosphorylation of p130 cas that was dependent on the presence of endoglin bearing an intact cytosolic domain. Thus, endoglin expression may result in alteration of tyrosine phosphorylation of the p130 cas -CrkII complex, which may occur in response to TGF-␤-related ligands (23,43), and suggests that endoglin may participate in cross-regulation between TGF-␤dependent serine/threonine and tyrosine kinase signaling pathways. Although TGF-␤1 did not strongly influence the overall endoglin-dependent change in p130 cas tyrosine phosphorylation, additional work will be required to determine whether reduction of p130 cas tyrosine phosphorylation results from endoglin-dependent effects on the TGF-␤ receptors or altered subcellular localization of zyxin, p130 cas -CrkII, or other constituents of the focal adhesion complex.
Finally, -FL endoglin was found to be associated with the focal adhesion-associated protein pool at a very low level relative to -⌬C endoglin. This association was greatly enhanced by removal of the cytosolic domain of endoglin (Fig. 7C), although removal of the carboxyl-terminal SMA tripeptide had no enhancing effect (Fig. 7B). Thus the endoglin CD, exclusive of the putative COOH-terminal PDZ-liganding motif, appears to mediate the association of endoglin with the focal adhesion-associated protein pool. These data may reflect significantly different rates of recruitment of endoglin into the nascent focal adhesion, mediated by cytosolic domain residues exclusive of the terminal potential PDZ-liganding motif. Alternatively, endoglin may initially localize to sites of focal adhesion via interactions within its extracellular or transmembrane domains, and the cytosolic domain mediates a high dissociation rate from the focal complex. Interestingly, the association of neither -FL nor -⌬C endoglin was altered by the presence of TGF-␤1 or activin A. Because endoglin phosphorylation is reduced by TGF-␤1 (36), this result suggests that endoglin CD phosphorylation is either unaffected by TGF-␤1 or activin A under these conditions, or that CD phosphorylation does not strongly affect the association of endoglin with focal adhesions. More work will be required to determine the exact role of endoglin CD post-translational modification on these processes.
Human endoglin possesses an RGD tripeptide in its extracellular domain that may participate in integrin-based interactions (1,44), although the lack of conservation of this tripeptide in mouse, pig, and rat endoglin suggests that, if functional, this may represent a recent adaptation. The extracellular domain of endoglin also contains a zona pellucida (45) domain that is thought to participate in protein-protein interactions within the extracellular space (46), although no role for this motif has been established in the case of endoglin. It is interesting to note, however, that the Drosophila protein, piopio (Pio), was recently recognized as an apically secreted extracellular matrix protein that has an important role in the regulation of epithelial tube diameter and branching. Pio, like endoglin (45), possesses an extracellular zona pellucida domain and a COOH-terminal sequence whose closest mammalian homologue appears to be the endoglin CD (47). Other genes that mimic Pio-mutant phenotypes in Drosophila include steamer duck (48), also known as PINCH, an evolutionarily conserved LIM domain protein that is postulated to act as part of an integrin-dependent signaling complex that colocalizes to sites of actin filament anchorage in both muscle and wing epithelial cells. Thus, there may be a precedent for a set of interactions involving Pio, PINCH, and sites of cell adhesion, which may be related to the roles postulated here for endoglin and zyxin, respectively. HHT1 is characterized by localized vascular dysplasia with pathological features that include focal dilation of postcapillary venules and a prominence of actin stress fibers in pericytes. It will be interesting to determine whether the formation of actin stress fibers can be perturbed by the presence or absence of ligands that bind within the zyxin LIM domains. We may speculate that stoichiometric reduction of endoglin protein levels by half, as is observed in HHT1, could contribute to cumulative alteration of the detailed composition of proteins located within the focal adhesion. Such alterations could impact cell structure, cell adhesion, and mobility, potentially affecting vascular remodeling and repair, and contributing to the clinical picture of HHT1.