Endoglin Is an Accessory Protein That Interacts with the Signaling Receptor Complex of Multiple Members of the Transforming Growth Factor-β Superfamily*

Endoglin (CD105) is a transmembrane glycoprotein that binds transforming growth factor (TGF)-β1 and -β3, and coprecipitates with the Ser/Thr kinase signaling receptor complex by affinity labeling of endothelial and leukemic cells. The present study shows that in addition to TGF-β1 and -β3, endoglin interacts with activin-A, bone morphogenetic protein (BMP)-7, and BMP-2 but requires coexpression of the respective ligand binding kinase receptor for this association. Endoglin cannot bind ligands on its own and does not alter binding to the kinase receptors. It binds TGF-β1 and -β3 by associating with the TGF-β type II receptor and interacts with activin-A and BMP-7 via activin type II receptors, ActRII and ActRIIB, regardless of which type I receptor partner is coexpressed. However, endoglin binds BMP-2 by interacting with the ligand binding type I receptors, ALK3 and ALK6. The formation of heteromeric signaling complexes was not altered by the presence of endoglin, although it was coprecipitated with these complexes. Endoglin did not interact with BMP-7 through complexes containing the BMP type II receptor, demonstrating specificity of its action. Our data suggest that endoglin is an accessory protein of multiple kinase receptor complexes of the TGF-β superfamily.

Endoglin is the target gene for the dominantly inherited vascular disorder hereditary hemorrhagic telangiectasia type 1 (HHT1) (36). HHT is characterized by frequent nose bleeds, mucocutaneous telangiectases, and the development of arteriovenous malformations predominantly in lung, brain, and the gastrointestinal tract that lead to recurrent hemorrhage and shunting (37). We have recently shown that mutant forms of endoglin are degraded intracellularly and that HHT1 is associated with reduced levels of surface endoglin on endothelial cells and activated monocytes (38).
Endoglin was shown to bind TGF-␤1 and -␤3 with high affinity, but not -␤2 (39), suggesting it mimics the isoform specificity of T␤RII. Endoglin was coprecipitated with T␤RII and a type I receptor in endothelial and leukemic cells (38, 40 -42). Human umbilical vein endothelial cells (HUVEC) were shown to have a single high affinity binding site representative of T␤RII complexes (39). Furthermore, when overexpressed in U937 monocytic cells, endoglin did not alter binding (42). These previous studies suggested that endoglin alone may not bind TGF-␤. We tested this using a COS1 transfection system and now demonstrate that endoglin requires the coexpression of T␤RII to bind TGF-␤1 and -␤3. In addition, it binds activin-A, BMP-2, and BMP-7 only in the presence of their respective ligand binding receptors. We also demonstrate that endoglin associates with heteromeric signaling receptor complexes of multiple members of the TGF-␤ superfamily.

EXPERIMENTAL PROCEDURES
Cell Culture and Transfections-Endothelial cells were derived from HUVEC of newborns by previously published procedures and maintained as described (28). NCTC2071 fibroblasts were cultured as published (43). COS1 cells were maintained and transiently transfected with expression constructs using the DEAE-dextran-chloroquine method as reported (44,45). Assays were performed 2 days after transfection.
Antibodies-P3D1 and P4A4 hybridoma to human endoglin were provided by E. A. Wayner (Seattle, WA) and were described and characterized extensively (50) . Murine IgG1 (Coulter Electronics) was used as an isotype control for these two monoclonal antibodies (mAb). For immunoprecipitation of T␤RII, the polyclonal antisera (pAb) C16, which was raised in rabbits by immunization with a synthetic peptide corresponding to amino acids 550 -565 of the highly conserved carboxyl terminus of human type II TGF-␤ receptor, was used (Santa Cruz Biotechnology Inc.). For immunoprecipitation of HA-and FLAG-tagged TGF-␤ superfamily receptors, monoclonal antibodies 12CA5 (Boehringer Mannheim) and M2 (IBI, Eastman Kodak), respectively, were used.
For all other ligands tested, HUVEC or transfected COS1 were incubated with 250 pM 125 I-TGF-␤3, 800 pM 125 I-activin-A, 2 nM 125 I-BMP-2, or 1-2 nM 125 I-BMP-7 for 3-4 h, cross-linked with DSS, and analyzed as described above using Triton X-100 in the lysis solution. Conditions for immunoprecipitation and analysis were also the same as above; however, some assays included the mAb P4A4 (1.6 g) to endoglin, anti-HA mAb 12CA5 (1.5 g), or anti-FLAG mAb M2 (3 g) to tagged receptors.
Metabolic Labeling and Western Blot Analysis-Endoglin and T␤RII expression in transfected COS1 cells were quantitated by metabolic labeling in Fig. 1 (38). In other experiments, endoglin protein levels in transfected COS1 cells were moni-FIG. 1. Endoglin requires coexpression of T␤RII for binding to TGF-␤1. A, COS1 transiently transfected with empty vector (pCMV5), endoglin (END) and/or T␤RII were affinity-labeled with 200 pM 125 I-TGF-␤1, cross-linked with DSS, and solubilized in Triton X-100. Aliquots of total lysates each containing 20 g of total protein were fractionated on 4 -12% SDS-PAGE followed by autoradiography. Aliquots of lysates were immunoprecipitated with mAb P3D1 (␣END), pAb C16 (␣T␤RII), and control IgG1 as indicated. Arrows indicate the position of monomeric endoglin (END) and T␤RII (RII) separated under reducing conditions (R), and of endoglin dimers, oligomers (OLIGO) and T␤RII when fractionated non-reduced (NR). B, aliquots of COS1 cells transfected in A were metabolically labeled with [ 35 S]Met and solubilized in Triton X-100. Lysates containing equivalent cpm were immunoprecipitated with mAb P3D1 (lanes 1-4) and pAb C16 (lanes 5 and 6) and fractionated by SDS-PAGE using reducing conditions. The left bracket indicates monomeric glycosylated endoglin (END) and precursor in lysates of cells expressing endoglin (lanes 2 and 4). Fully glycosylated RII and precursor bands are indicated by the right bracket (lanes 5 and 6). tored by Western blot analysis. Aliquots of total lysates from affinitylabeled transfected cells were separated by SDS-PAGE (4 -12% gradient, non-reducing conditions) and assayed by immunoblotting using mAb P4A4 (1.6 g/ml) to endoglin as described previously (38). For determination of BMPRII/FL protein levels, aliquots of total cell lysates were separated by SDS-PAGE and assayed by immunoblotting with mAb M2 to FLAG as described (45). Immunoblots were visualized using the enhanced chemiluminescence detection kit (ECL ® ; Amersham Pharmacia Biotech) using the protocols provided. Multiple exposures using Hyper-film (Amersham Pharmacia Biotech) were obtained.
Cell Surface Biotinylation-Equivalent numbers of HUVEC or transiently transfected COS-1 cells at subconfluence (ϳ90%) were surfacelabeled with biotin in the absence of added ligand as reported (38). Cells were lysed and immunoprecipitated as described under "Binding and Affinity Labeling" except lysates were precleared for at least 1 h with Protein A-Sepharose ® CL-4B (Amersham Pharmacia Biotech). Eluates were fractionated on 4 -12% SDS-PAGE gels, transferred to polyvinylidene difluoride nylon membranes, and blocked as described for Western blots. Membranes were probed with streptavidin-horseradish peroxidase (400-fold dilution in Tris-buffered saline with Tween 20; Amersham Pharmacia Biotech) for 20 min, and biotinylated proteins were detected using ECL (Amersham Pharmacia Biotech); multiple exposures were obtained.

RESULTS
Endoglin Requires the Coexpression of T␤RII to Bind TGF-␤1-Previous studies have shown that endoglin interacts with the TGF-␤ binding complex. To investigate the nature of this complex, we transiently expressed endoglin in COS1 cells in the presence or absence of T␤RII. Cell monolayers were affinity-labeled with 125 I-TGF-␤1, chemically cross-linked using DSS, and cell lysates fractionated by SDS-PAGE either directly or after immunoprecipitation using antibodies to endoglin or T␤RII (Fig. 1A). Compared with the vector control, no affinitylabeled proteins were detectable when endoglin was expressed alone (compare lanes 4 -6 with lanes 1-3), whereas T␤RII expressed alone bound TGF-␤1 strongly (lanes 7-9). Control experiments confirmed that endoglin was expressed at levels comparable to that of T␤RII (Fig. 1B). We previously showed that endoglin is processed by COS1 cells and is expressed on the cell surface in the absence of T␤RII (38). In contrast, when endoglin was coexpressed with T␤RII, affinity-labeled endoglin migrating as a monomer of 105-120 kDa was detected (Fig. 1A, lane 11). Since endoglin and T␤RII migrate close to each other on reducing SDS-PAGE gels, affinity-labeled proteins were also analyzed on non-reducing gels (lanes 14 -17). Under these conditions, products characteristic of dimers (180 kDa) and oligomers (Ͼ200 kDa) of endoglin were detected (39) and could be immunoprecipitated using anti-endoglin (Fig. 1A, lane 15). Immunoprecipitation with anti-T␤RII revealed that endoglin coprecipitated with T␤RII (85-95 kDa) (Fig. 1A, lane 16). Thus we conclude that endoglin is unable to bind TGF-␤1 when expressed alone.
We also analyzed the effect of chemical cross-linking on 125 I-TGF-␤1 binding to endoglin (Table I). TGF-␤1 binding to endoglin alone was very low in the presence of the cross-linker DSS, and may represent association of endoglin with the low levels of endogenous TGF-␤ receptors present on COS1 cells (see Fig. 1A, lanes 1-3). However, in the absence of DSS, no binding of TGF-␤1 to endoglin was observed (Table I). When T␤RII was coexpressed with endoglin, strong binding of 125 I-TGF-␤1 to endoglin was observed in DSS-treated cells, while binding was significantly less in non-cross-linked samples. In contrast to endoglin, TGF-␤1 binding to T␤RII showed little dependence on the chemical cross-linker. These data suggest that the binding of TGF-␤1 to endoglin is unstable in the absence of a cross-linker.
Interaction of Endoglin with TGF-␤1⅐T␤RII on HUVEC Is Preserved in Digitonin-We next examined the binding of 125 I-TGF-␤1 to HUVEC (Table II), which express both T␤RII and endoglin. In Triton X-100, cpm eluted from anti-T␤RII immunoprecipitates showed little dependence on the presence of cross-linker whereas, those eluted from anti-endoglin immunoprecipitates were dependent on cross-linker. Since detergents such as Triton X-100 can disrupt membrane protein complexes, we also solubilized cells with milder detergents like CHAPS TABLE I Effect of chemical cross-linking on 125 I-TGF-␤1 binding to endoglin and T␤RII transiently expressed in COS1 cells COS1 cells were transiently transfected with the indicated gene constructs endoglin (END) and T␤RII in the pCMV5 vector. Two days after transfection, cells were incubated with 200 pM 125 I-TGF-␤1, treated with or without DSS, solubilized with Triton X-100. T␤RII and endoglin were immunoprecipitated with pAb C16 (␣T␤RII) and mAb P3D1 (␣END), respectively, from aliquots of total lysates containing equivalent total protein content. Immunoprecipitates were eluted from Protein A-or G-Sepharose in 1% SDS (Ͼ95%) and counted in a ␥ counter. Counts/min (cpm) above background (IgG) for immunoprecipitates are reported. 125  solubilization of membrane proteins HUVEC were incubated with 200 pM 125 I-TGF-␤1, treated with or without DSS, and solubilized with Triton X-100, CHAPS, or digitonin plus protease inhibitors. T␤RII and endoglin were immunoprecipitated with pAb C16 (␣T␤RII) and mAb P3D1 (␣END), respectively, from aliquots of total lysates containing equivalent total protein content. Immunoprecipitates were eluted from Protein A-or G-Sepharose in 1% SDS (Ͼ95%) and counted in a ␥ counter. Counts/min (cpm) above background (IgG) for immunoprecipitates are reported. and digitonin known to better preserve some protein/protein interactions (55). Of the three detergents tested, CHAPS was most effective in preserving TGF-␤1 interaction with T␤RII. In CHAPS, we also recovered 12% of the radiolabeled TGF-␤1 in the anti-endoglin immunoprecipitates in the absence of the cross-linker versus 4% recovery in Triton X-100 (Table II). In digitonin, however, we recovered 36% of the radiolabeled TGF-␤1 in anti-endoglin immunoprecipitates relative to crosslinked samples, which was comparable to 46% in the anti-T␤RII immunoprecipitates, despite an overall reduced efficiency of lysis. These data suggest that TGF-␤1 interaction with endoglin can be preserved in mild detergents. Immunoprecipitation of T␤RII⅐endoglin complexes using pAb C16 directed to the COOH terminus of T␤RII consistently yielded efficient coprecipitation of endoglin (see Fig. 1). In contrast, anti-endoglin coprecipitated little T␤RII that required overexposure of autorads for visualization. Since these experiments were routinely performed in Triton X-100, we determined whether CHAPS or digitonin might preserve endoglin⅐T␤RII complexes in anti-endoglin immunoprecipitates. HUVEC were affinity-labeled with 125 I-TGF-␤1 and lysed in Triton X-100, CHAPS, or digitonin, and the lysates subjected to immunoprecipitation with antibodies directed against T␤RII or endoglin (Fig. 2). The profile of endogenous receptor complexes immunoprecipitated from Triton X-100-solubilized HUVEC (Fig. 2, lanes 4 and 5) was similar to that previously reported (38,39) and was comparable to that observed in COS1 cells coexpressing T␤RII and endoglin (see Fig.  1A). Similar results were obtained in CHAPS-solubilized cells (lanes 6 and 7). However, in the presence of digitonin, we observed efficient coprecipitation of affinity-labeled T␤RII with the anti-endoglin (lanes 9 and 15). Furthermore, the relative intensities and profile of digitonin solubilized receptor complexes were similar in the anti-endoglin and the anti-T␤RII immunoprecipitates (compare lane 8 with lane 9, and lane 14 with lane 15). Thus digitonin preserved endoglin association with both T␤RII and TGF-␤1. We have shown that detergents that disrupt T␤RII association with endoglin in anti-endoglin immunoprecipitates also disrupt coprecipitation of TGF-␤ with endoglin in the absence of cross-linker. Furthermore digitonin, which preserved association of T␤RII with endoglin, also led to coprecipitation of TGF-␤1 with endoglin in the absence of crosslinker. Thus, the ability of endoglin to maintain interactions with TGF-␤1 is dependent on its association with T␤RII. Based on these results, we propose that endoglin is not itself a TGF-␤ receptor, but rather is cross-linked with the ligand through association with the TGF-␤ type II receptor.
Endoglin Interacts with Heteromeric Receptor Complexes Containing ALK5 and Does Not Disrupt Their Formation-ALK5 (T␤RI) preferentially interacts with ligand-bound T␤RII to generate a TGF-␤ receptor signaling complex. ALK5 is unable to bind TGF-␤ on its own, but does so when coexpressed with T␤RII (9). Having established that endoglin recognizes ligand-bound type II receptors, we investigated whether it could modulate binding and subsequent formation of heteromeric complexes between ALK5 and T␤RII (Fig. 4). No binding is observed when endoglin and ALK5 are coexpressed (Fig. 4,  lanes 1 and 2), but cotransfection of T␤RII leads to binding of TGF-␤1 to both ALK5 and endoglin (lanes 3-5). Anti-endoglin immunoprecipitated ALK5 and T␤RII with endoglin (lanes 6 and 7). The anti-HA immunoprecipitates showed that the heteromeric complex between ALK5 and T␤RII is not affected by endoglin expression (lanes 8 and 9). Together, these data show that endoglin can interact with kinase receptor complexes containing ALK5, but does not enhance overall binding, nor modulate the association of receptor I with receptor II.
Endoglin Binds Activin-A or BMP-7 When Coexpressed with ActRII or ActRIIB 2 -As endoglin was cross-linked to TGF-␤1 and -␤3 through its association with T␤RII, we tested whether it could interact with other type II receptors of the TGF-␤ superfamily. We first examined activin-A, which binds to two related type II receptors, ActRII and ActRIIB, and signals through a mechanism similar to that defined for TGF-␤ receptors (44). COS1 cells were transiently transfected with endoglin alone or together with ActRII or ActRIIB 2 (a ligand binding functional isoform of ActRIIB; Ref. 16) and were then affinitylabeled using 125 I-activin-A (Fig. 5A). When transfected alone, endoglin did not bind activin-A, despite efficient expression of the protein (Fig. 5A, lanes 1 and 2). However, in COS1 cells coexpressing type II receptors, anti-endoglin immunoprecipitated 125 I-activin-A cross-linked to endoglin dimers and oligomers (lanes 12 and 13), and coprecipitated activin type II receptors (lanes 10 -13). Thus endoglin can form complexes with ActRII or ActRIIB 2 bound to activin-A. As these receptors can also bind BMP-7 (27), we determined whether endoglin might also associate with activin type II receptors bound to BMP-7 (Fig. 5B). Endoglin alone did not bind BMP-7, but did so in the presence of coexpressed ActRII and to a lesser degree with ActRIIB 2 (bottom panel). Reduced efficiency of cross-linking of ligands to ActRIIB versus ActRII has been reported (16).
Endoglin Does Not Interact with BMP-7⅐BMPRII Receptor  Fig. 1. ALK5 was tagged at the carboxyl terminus with HA. All samples were fractionated reduced. Shown are total lysates (lanes 1-5) and eluates from immunoprecipitates with mAb P3D1 (␣END) and with mAb 12CA5 (␣HA) in lanes 6 and 7 and lanes 8 and 9, respectively. Arrows indicate the positions of the affinity-labeled products, endoglin (END), T␤RII (RII), and the type I receptors (RI).

FIG. 5. Endoglin binds activin-A or BMP-7 when coexpressed with activin type II receptors.
A, COS1 cells were transiently transfected with various combinations of pCMV5 empty vector, pCMV5-END, and HA-tagged pCMV5-ActRII or pCMV5-ActRIIB 2 , affinity-labeled with 800 pM 125 I-activin-A, and analyzed as in Fig. 1. Total lysates that were fractionated reduced (R) are shown in lanes 1-6; affinitylabeled type II receptors (RII) are indicated by left bracket. Endoglin expression was analyzed by Western blot of an aliquot of this total lysate as in Fig. 3B (lower panel, lanes 1-6). Eluates from immunoprecipitates with mAb P3D1 (␣END) were fractionated reduced in lanes 7-11. Right brackets indicate the positions of monomeric affinity-labeled endoglin (END) and type II receptor (RII). Corresponding eluates from lanes 10 and 11 were fractionated non-reduced (NR) in lanes 12 and 13, respectively, and lane 14 represents the negative control for these conditions. Arrows indicate the position of endoglin dimers and oligomers (OLIGO) and ActRII affinity-labeled with activin-A. B, COS1 cells were transiently transfected with various combinations of pCMV5 empty vector, pCMV5-END, and/or pCMV5-ActRII/HA, pCMV5-Ac-tRIIB 2 /HA, or Flag-tagged BMPRII as indicated, affinity-labeled with 1 nM 125 I-BMP-7, and analyzed as in Fig. 1. Total lysates fractionated under reducing conditions (top panel) reveal affinity-labeled type II receptors (RII). Endoglin expression was analyzed by Western blotting of an aliquot of these total lysates as in Fig. 3B. BMPRII/FL was also analyzed by Western blotting of total lysates using mAb M2 (␣FLAG). Eluates from immunoprecipitates with mAb P3D1 (␣END) were fractionated non-reduced in lower panel. Arrows indicate the position of endoglin (END) dimers and oligomers (OLIGO) affinity-labeled with BMP-7.
Complexes-In these studies, BMP-7 was unable to interact with BMPRII alone despite efficient expression of this receptor in COS1 cells (Fig. 5B). Coexpression of endoglin did not alter this binding. Since binding of BMP-7 to BMPRII was shown previously to be dependent on the coexpression of the type I receptors ALK6 or ALK2 (24), we investigated the ability of endoglin to interact with these complexes. Fig. 6 (A and B) demonstrates binding of BMP-7 to ALK6⅐BMPRII or ALK2⅐BMPRII complexes and is similar to that observed with ActRII or ActRIIB 2 (upper panels). However, endoglin could not associate with the BMPRII complexes, while it could with the ActRII and ActRIIB 2 complexes (middle panels). Indeed, under reducing conditions anti-endoglin immunoprecipitates revealed coprecipitating affinity-labeled proteins corresponding to the type II receptors, ActRII or ActRIIB 2 together with the type I receptor, ALK2 (Fig. 6B, lanes 9 -12), but it did not alter the formation of any of these complexes as seen in the anti-HA immunoprecipitates (lower panels). Thus, endoglin is not found associated with BMP-7⅐BMRPRII complexes containing ALK6 or ALK2.
Endoglin Associates with BMP-2 When Coexpressed with the Type I Receptors ALK3 and ALK6 -Unlike TGF-␤, activin, and BMP-7, BMP-2 initiates signaling by first interacting with the type I receptors ALK3 or ALK6 and then recruits type II receptors into a signaling complex. Since endoglin interaction with ligands appears to require expression of the ligand-binding component of the heteromeric Ser/Thr kinase receptor complex, we determined whether endoglin might interact with BMP-2 in the presence of type I receptors. Endoglin alone was unable to bind BMP-2, but did so upon coexpression with ALK3 or ALK6 (anti-endoglin (␣END) panels, lanes 1 and 3 in Fig. 7,  A and B). Interestingly, when endoglin was coexpressed with either ALK3 or ALK6 and BMPRII, we observed a substantial decrease in the association of endoglin with BMP-2 (Fig. 7A, compare lane 5 with lane 3; Fig. 7B, compare lane 6 with lane 3, anti-endoglin (␣END) panels). This occurred despite efficient formation of BMP-2 binding receptor complexes in these cells as seen in total lysates and anti-Flag and anti-HA immunoprecipitates, and may reflect the inability of endoglin to associate with BMPRII, as noted above.
We also tested for the association of endoglin with BMP-2 in the presence of ActRII and ActRIIB 2 . Consistent with previous observations on the Drosophila type II receptor, punt (56), we were unable to observe any binding of BMP-2 to ActRII (Fig.  7A, lane 8) or ActRIIB (data not shown) when these receptors were expressed alone. However, in the case of cells coexpressing ALK3 and either ActRII or ActRIIB 2 , the association of endoglin with BMP-2 was comparable to that observed with ALK3 alone (Fig. 7A, lanes 7 and 10 compared with lane 3). Similar results were obtained in the case of cells coexpressing ALK6 and ActRII or ActRIIB 2 (Fig. 7B, lanes 8 and 10 compared with lane 3), although coexpression of ActRIIB 2 yielded lower overall levels of BMP-2 binding (seen in total lysates; FIG. 6. The type I receptors ALK2 or ALK6 neither modulate nor induce endoglin interactions with BMP-7⅐ActRII, RIIB, or BMPRII complexes. A, COS1 cells were transiently transfected with pCMV5-ALK6/HA, with or without pCMV5-END, and/or ActRII/His, ActRIIB 2 , BMPRII/FL as indicated, affinity-labeled with 1 nM 125 I-BMP-7 and analyzed as in Fig. 1. Total lysates fractionated under reducing conditions are shown in top panel, where affinity-labeled type II receptors (RII) and ALK6 (RI) are indicated by arrows. Endoglin expression was analyzed by Western blot of an aliquot of these total lysates as in Fig. 3B, shown just below top panel. Eluates from immunoprecipitates with mAb P3D1 (␣END) were fractionated non-reduced in middle panel. Arrows indicate endoglin (END) dimers and oligomers (OLIGO) affinity-labeled with BMP-7. Immunoprecipitation analysis of ALK6/HA (␣HA) or BMPRII/FL (␣FLAG) is shown in the lower panel, and serves as the control for binding to and formation of heteromeric complexes. Affinity-labeled type II receptors (RII) co-precipitating with ALK6 (RI) are indicated with arrows. B, COS1 cells were transiently transfected with pCMV5-ALK2/HA, with or without pCMV5-END, and/or ActRII/His, ActRIIB 2 , or BMPRII/FL as indicated and affinitylabeled with 2 nM 125 I-BMP-7. All samples were processed as in A. In addition, anti-endoglin immunoprecipitates were fractionated reduced (R) showing coprecipitation of endoglin (END) with RII and ALK2 (RI) (right panel, lanes 9 -12). lanes 7-10); this may account for the reduced level of BMP-2 bound to endoglin observed in these transfectants. We also confirmed that endoglin could associate stably with BMP-2 receptor complexes containing the cotransfected ALK3 or ALK6 as they were coprecipitated with anti-endoglin as seen under reducing conditions (Fig. 7, A and B, lanes 11-13).
Endoglin Associates with the Ligand Binding Receptors-Since our data suggest endoglin interacts with ligand binding receptors, we next tested whether endoglin could directly bind these receptors in the absence of exogenously added ligand (Fig. 8). In COS1 cells that were transfected with type II receptors, with or without endoglin, we analyzed receptor interactions by surface biotinylation, solubilization in digitonin, and specific immunoprecipitation. We found T␤RII and ActRII coprecipitated with anti-endoglin as seen under both non-reducing and reducing conditions (Fig. 8A, lanes 3-6). This was not observed in BMPRII-expressing cells (lanes 7 and 8) nor in the controls (lanes 1 and 2). When the same transfectants were analyzed with anti-T␤RII (lanes 9 -12), anti-HA (lanes 13-16), or anti-FLAG (lanes [17][18][19][20], endoglin coprecipitated with T␤RII, ActRII, but not BMPRII, respectively. Endoglin also interacted with ActRIIB 2 (data not shown) in the absence of added ligand. In a similar series of experiments, we found that the BMP type I receptors, ALK2, ALK3, and ALK6, did not coprecipitate with endoglin (Fig. 8B, lanes 3-16). Furthermore, in COS1 cells expressing endoglin alone, we found no evidence for interaction with endogenous receptors (lane 2). These data demonstrate that endoglin interacts specifically with the ligand binding type II receptors T␤RII, ActRII, and ActRIIB 2 .
However, association with type I receptors may require ligand and/or chemical cross-linkers for detection by coimmunoprecipitation. Furthermore, these latter interactions are not likely dependent on the expression of endogenous type II receptors, since the interaction of endoglin with the low level of any endogenous type II receptors was undetectable both by affinity labeling and cell surface labeling.
We also investigated the interaction between endogenous endoglin and T␤RII, which are both expressed in endothelial cells and NCTC2071 fibroblasts (43). For this, the cells were surface-labeled by biotinylation and solubilized in digitonin prior to immunoprecipitation (Fig. 8C). We observed that under non-reducing and reducing conditions, endoglin coprecipitated with T␤RII in the absence of added ligand (compare lanes 2, 5,  8, and 11 with lanes 1, 4, 7, and 10). However, as with the affinity labeling experiments (Fig. 2), this interaction was disrupted in Triton lysates (Fig. 8C, lanes 13-18). Together, these data suggest that endoglin associates with the type II receptors in the absence of ligand.

DISCUSSION
The present studies show that endoglin can interact with TGF-␤1, ␤3, activin-A, BMP-7, and BMP-2, but requires the coexpression of the respective ligand-binding kinase receptor partner for binding and specificity. For TGF-␤1, its association with endoglin is better demonstrated with the use of weak detergents that do not disrupt the interaction of endoglin with T␤RII. These results strongly suggest that endoglin binds TGF-␤ secondarily to its association with T␤RII already bound FIG. 7. Endoglin binds BMP-2 when ALK3 or ALK6 are coexpressed. A, COS1 cells were transiently transfected with pCMV5-ALK3/HA, with or without pCMV5-END, and/or ActRII/His, Ac-tRIIB 2 or BMPRII/FL as indicated and affinity-labeled with 2 nM 125 I-BMP-2. Analysis is the same as in Fig. 6B. B, COS1 cells were transiently transfected with pCMV5-ALK6/HA, with or without pCMV5-END, and/or ActRII/His, Ac-tRIIB 2 , or BMPRII/FL as indicated and affinity-labeled with 2 nM 125 I-BMP-2. Analysis is the same as in Fig. 6B. to ligand. This would explain why the specificity and affinity of endoglin for TGF-␤ isoforms mimics that of T␤RII. It is likely that this is true for all ligands that interact with endoglin.
Endoglin was first defined as a component of the TGF-␤ receptor system when betaglycan was sequenced and found to be similar to endoglin in particular, in the cytoplasmic tail where these two proteins are 71% identical (39). Betaglycan is a proteoglycan (Ͼ200 kDa) also called a type III receptor that is required for presenting TGF-␤2 to the kinase receptor complex T␤RII⅐ALK5, and promoting signaling by this isoform (57). Betaglycan also binds other isoforms, ␤1 and ␤3, on its own and potentiates binding to cells ultimately enhancing the response of cells to these ligands. Betaglycan acts as a dual modulator of ligand access to the signaling receptors, as it can be released from the cell membrane and binds ligand in soluble form; thus, it is clearly defined as a receptor (58). No known signaling domains have been identified in its structure; however, like endoglin, its short cytoplasmic tail is highly conserved among species (43,59). Endoglin has often been compared with betaglycan and postulated to function in a similar fashion by affecting the binding of TGF-␤1 and TGF-␤3 to the signaling receptors (35,39,40,60,61), and it has also been described as an auxiliary receptor (42). Our results clearly demonstrate that endoglin does not function like the type III receptor betaglycan, as it needs coexpression of a ligand binding receptor to interact with ligand. Endoglin cannot bind ligand on its own; it does not alter overall binding to the kinase receptors, but mimics the specificity of the ligand binding receptor it interacts with. Indeed, previous studies have shown that overexpression of full-length functional endoglin in U937 does not alter the binding affinity of receptor complexes (42), and we have obtained similar results when endoglin is expressed in 3T3 fibroblasts. 2 As we show endoglin is not a true receptor, we define it is an accessory protein that interacts with the ligand binding receptors of multiple members of the TGF-␤ superfamily.
We demonstrate that endoglin not only interacts with ligand binding receptors, but also associates with multiple heteromeric receptor complexes (summarized in Fig. 9A). For TGF-␤1, ␤3, activin, and BMP-7, the type II receptors bind ligand and recruit the type I receptor partners into a high affinity complex. Endoglin can interact with the ligand binding type II receptors, T␤RII, ActRII and ActRIIB 2 , regardless of which type I receptor partner is coexpressed, and associates with these type II receptors in the absence of exogenous ligand. Endoglin does not disrupt the formation of the signaling receptor complexes, and can be coprecipitated with these complexes. It could not, however, interact with BMP-7⅐BMPRII complexes demonstrating specificity of endoglin action, nor did it interact with BMPRII in the absence of ligand. However, the type I receptors ALK3 and ALK6 bind BMP-2 and recruit the type II receptor partners into a high affinity complex. Endoglin also interacts with these ligand binding type I receptors and their respective type II receptors (Fig. 9A). However, the association of endoglin with ALK-3 and ALK-6 was only detectable in the presence of ligand and a chemical cross-linker, which suggests these associations might be more transient. Interestingly, we observed a reduction in BMP-2 binding to endoglin when BM-PRII was coexpressed, yet the ligand binding to BMPRII was not altered. Since endoglin probably binds ligands secondary to its associations with the ligand binding receptors, these data suggest that BMPRII may compete with endoglin for ALK-3 or ALK-6 and that endoglin may be excluded from these complexes. Thus, endoglin is an accessory protein that interacts with multiple heteromeric receptor complexes containing TGF-␤s, activins, and BMPs.
We have shown that endoglin interacts with specific ligand binding receptors, and we postulate it is recruited into active receptor complexes in this way (see model in Fig. 9B). What is the significance of these findings? There are several levels of receptor function that could be affected by the presence of endoglin. It could modulate the activity of the receptor kinase complex. However, we have found that endoglin does not alter the transphosphorylation of ALK5 by T␤RII. 3 The receptor complexes that endoglin interacts with contain the signal transducing receptors ALK2, 3, 5, or 6. Since endoglin interacts with activin type II receptors, it is likely to be found in heteromeric complexes containing ALK4 (ActRIB) (Fig. 9). As multiple signal transducing type I receptors interact with endoglin, it may function to modulate receptor activation of downstream events, such as Smad signaling (Fig. 9). For instance, endoglin might regulate BMP signaling through Smads 1, 5, and 8 or TGF-␤/activin signaling through Smads 2 and 3. Although 3 N. P. Barbara and J. L. Wrana, unpublished data.
FIG. 8. In the absence of exogenous ligand, endoglin associates with T␤RII and ActRII but not with BMPRII or the BMP type I receptors. A, COS1 cells were transiently transfected with pCMV5-T␤RII, ActRII/HA, BMPRII/FL, with or without pCMV5-END as indicated and surface-labeled with NHS-LC-biotin, solubilized with 1% digitonin, immunoprecipitated with mAbs P3D1 (␣END), 12CA5 (␣HA), M2 (␣FLAG), and pAb C16 (␣T␤RII), and eluates were fractionated under non-reducing (top panels) and reducing conditions (bottom panels) as in Fig. 2. Gels were transferred to polyvinylidene difluoride nylon membranes, probed with streptavidin-horseradish peroxidase, and detected by ECL. Brackets indicate the positions of biotinylated surface-expressed endoglin (END) dimers, oligomers (OLIGO), and monomeric type II receptors (RII) (non-reduced). Arrows indicate biotinylated monomeric endoglin and the type II receptors (reduced). B, COS1 cells were transiently transfected with pCMV5-ALK2/HA, ALK3/HA, ALK6/HA, with or without pCMV5-END, and analyzed as in A. C, HU-VEC and NCTC2071 fibroblasts were surface-labeled with NHS-LC-biotin, solubilized with 1% digitonin (left) or 1% Triton (right), and analyzed as in Fig. 2. Arrows indicate the positions of biotinylated endoglin dimers, oligomers, and monomeric T␤RII (lanes 1-6; non-reducing conditions) and monomeric endoglin and T␤RII (lanes 7-18; reducing conditions). endoglin expression has no effect on the induction of the plasminogen activator inhibitor promoter by TGF-␤ or activin-A in mink lung epithelial cells, 3 a more comprehensive look at downstream events may be warranted. Furthermore, endoglin could recruit other proteins or a novel Smad into the signaling complex, thereby inducing a specific nuclear response (Fig. 9).
Previous studies have shown that endoglin plays a role in the TGF-␤ pathway, as overexpression of endoglin modulates some but not all TGF-␤1 responses in U937 monocytes (42). Endoglin has also been implicated in the regulation of trophoblast differentiation, a process stimulated by activin and inhibited by TGF-␤1 and -␤3 (62, 63). As we now show that endoglin can interact with activin type II receptors, it might be functioning in activin as well as TGF-␤ receptor complexes during placental development. Furthermore, activin and TGF-␤ exert multiple effects on many cell types including erythrocytes, endothelial cells, stromal fibroblasts, and mesenchymal cells where endoglin is expressed (1, 64).
A major role for endoglin in the vasculature was inferred by the finding that it is mutated in HHT1 (36). It is currently unclear what molecular mechanisms underlie HHT pathology; however, our studies now implicate pathways involving activins and BMPs, as well as TGF-␤s. Both TGF-␤ and activin are known to inhibit the proliferation of endothelial cells in culture (65,66). TGF-␤ is directly implicated in vascular development and thought to control interaction between endothelial cells and smooth muscle cells (67,68). BMPs may also be involved in these processes. BMP-2 and -7 can act on vascular smooth muscle cells to inhibit their proliferation without stimulating extracellular matrix synthesis, whereas activin-A has a growth-stimulatory effect (69). The BMP-like factor GDF-5, which binds ALK6, induces angiogenesis while BMP-2 does not (70). This is not surprising, as vascular endothelial cells do not have specific binding sites for BMP-2 (71). Together, these studies suggest a role for BMPs in regulating the maintenance and/or formation of vasculature involving both endothelial and smooth muscle cells where their specific response to these ligands depends on the receptors they express. The downstream effectors responsible for mediating these responses have yet to be identified. In this context, it is interesting to note that ALK1 is the target gene for HHT2 (72). Recently, ALK1 was shown to signal BMP-like responses, yet its ligand in endothelial cells is unclear (73). Furthermore, a class of inhibitory Smads were recently shown to be expressed at high levels in endothelial cells during laminar shear stress (74 -77). As stress to blood vessels has been implicated in the development of arteriovenous malformations (78), vascular Smads might also play a role in the pathology of HHT. These findings, together with the demonstration that endoglin is an accessory protein interacting with multiple receptor kinase complexes, support the notion that HHT could involve altered responses of the vasculature in pathways additional to TGF-␤. Elucidating the mechanisms of how endoglin functions as an accessory molecule in the TGF-␤ superfamily is critical to understanding the molecular mechanisms underlying the development of HHT and biological processes where endoglin is expressed.
FIG. 9. Model of endoglin interaction with multiple kinase receptor complexes of the TGF-␤ superfamily. A, summary of data demonstrating the ligands that endoglin interacts with, the ligand binding receptors and the respective partners present in the heteromeric complexes. B, the type II (RII) and type I receptors (RI) are related Ser/Thr kinase receptors, where RII is constitutively phosphorylated (P) and transphosphorylates RI in the Gly/Ser-rich domain (open circle) upon formation of a ligand containing high affinity complex. For TGF-␤1, ␤3, activin, and BMP-7, the type II receptors bind ligand, whereas, for BMP-2, the type I receptors ALK3 and ALK6 bind ligand. Endoglin interacts with these receptors and becomes a component of the heteromeric receptor complexes. This interaction is specific, as endoglin does not interact with BMP-7⅐BMPRII. Endoglin does not alter the association of RI with RII and thus interacts with a heteromeric complex that can initiate downstream events. We postulate that endoglin might modulate signals, by acting on known Smads and altering transcriptional responses or by functioning in alternate pathway(s) leading to specific transcriptional responses.