Bone Morphogenetic Proteins Signal through the Transforming Growth Factor-β Type III Receptor*

The bone morphogenetic protein (BMP) family, the largest subfamily of the structurally conserved transforming growth factor-β (TGF-β) superfamily of growth factors, are multifunctional regulators of development, proliferation, and differentiation. The TGF-β type III receptor (TβRIII or betaglycan) is an abundant cell surface proteoglycan that has been well characterized as a TGF-β and inhibin receptor. Here we demonstrate that TβRIII functions as a BMP cell surface receptor. TβRIII directly and specifically binds to multiple members of the BMP subfamily, including BMP-2, BMP-4, BMP-7, and GDF-5, with similar kinetics and ligand binding domains as previously identified for TGF-β. TβRIII also enhances ligand binding to the BMP type I receptors, whereas short hairpin RNA-mediated silencing of endogenous TβRIII attenuates BMP-mediated Smad1 phosphorylation. Using a biologically relevant model for TβRIII function, we demonstrate that BMP-2 specifically stimulates TβRIII-mediated epithelial to mesenchymal cell transformation. The ability of TβRIII to serve as a cell surface receptor and mediate BMP, inhibin, and TGF-β signaling suggests a broader role for TβRIII in orchestrating TGF-β superfamily signaling.

kinase-1 (ALK1) (6), ALK2, ALK3, or ALK6 (7). The serine/ threonine kinase activity of the type I receptor is activated upon recruitment and phosphorylation by a type II receptor, either the BMP type II receptor (BMPRII), or one of the activin type II receptors (ActRII or ActRIIB) (8). Upon activation the type I receptor phosphorylates the intracellular effector proteins, Smad1/5/8 transcription factors, which complex with the common Smad, Smad4, and enter the nucleus to induce BMP-mediated target gene transcription (1). Whereas most BMPs are able to elicit distinct cellular effects, the mechanism by which a limited number of cell surface receptors mediate these divergent effects remains to be established.
Co-receptors are important components of many signaling pathways (9). The TGF-␤ type III receptor (T␤RIII or betaglycan), endoglin (10), and members of the repulsive guidance molecule family, DRAGON, RGMa, and hemojuvelin (11)(12)(13), have been characterized as TGF-␤ superfamily co-receptors. T␤RIII is an abundantly and ubiquitously expressed cell surface receptor that enhances binding of all three isoforms of TGF-␤ to the TGF-␤ signaling receptor complex (14), and is required for high affinity cell surface binding of TGF-␤2. T␤RIII also binds inhibin, another TGF-␤ superfamily member (15). In addition to directly regulating ligand availability, T␤RIII also alters the subcellular localization of the signaling receptor complex through interactions with the PDZ domain containing protein, GIPC (16), and ␤-arrestin2 (17). The demonstration that T␤RIII is required for TGF-␤2-stimulated epithelial to mesenchymal transformation (EMT) in vitro (18) and the embryonic lethality of the T␤RIII knock-out mouse (19,20) has fostered consideration of a unique and non-redundant role for T␤RIII that is independent of ligand presentation to the kinase receptor complexes.
Several observations suggest that T␤RIII may serve as a cell surface receptor for BMP. First, BMP shares structural similarity with ligands known to bind T␤RIII (21). Second, T␤RIII shares extracellular domain homology with endoglin (22,23), which binds BMP-2 and BMP-7 in the presence of their respective type II receptors (24). Finally, T␤RIII is a heparan sulfate proteoglycan (25,26) and these glycosaminoglycan modifications have been shown to mediate basic fibroblast growth factor binding to T␤RIII (27). As BMP has a strong affinity for heparan sulfate (28), these modifications may confer the ability of T␤RIII to bind BMP as well. Here we investigate whether T␤RIII functions as a cell surface receptor for BMP.
Adenoviruses for EMT assays were generated using the pAdEasy system (30). All concentrated viruses were titered by performing serial dilutions of the concentrated virus and counting the number of GFP-expressing 293 cells after 18 -24 h.
Adenoviruses containing sequences for human T␤RIII and non-targeting control short hairpin RNA were generated by Dharmacon and inserted into a vector co-expressing the DS-Red fluorophore using the Adeno-X TM Expression System (Clontech). Specificity for T␤RIII has previously been demonstrated (31,32). Recombinant human BMP-2, BMP-4, BMP-7, GDF-5, and soluble T␤RIII were purchased from R&D Systems.
BMP receptors were detected with an anti-hemagglutinin antibody (Roche). A polyclonal antibody against the cytoplasmic tail of T␤RIII was generated by our laboratory and characterized previously (33). The ␤-actin antibody was purchased from (Sigma). Both anti-mouse and anti-rabbit antibodies were purchased from Amersham Biosciences.
Iodination of BMP Family Members-125 I-BMP were generated using the chloramine T method as previously described (34). Ten micrograms of carrier-free recombinant human BMP-2, BMP-7, BMP-4, and GDF-5 were used for each labeling. 125 I-TGF-␤1 was purchased from Amersham Biosciences.
Cross-linking and Immunoprecipitation of Receptors-Binding assays were performed as previously described (35). COS-7 cells (150,000 cells/well in 6-well plates) were transiently transfected with 2 g (or otherwise indicated) of plasmid DNA using FuGENE 6 (Roche) 18 h after plating. Transiently transfected cells were incubated for 3 h at 4°C with 10 nM BMP (150 pM TGF-␤1), unless indicated otherwise. Endogenous T␤RIII studies were incubated overnight at 4°C. Competition studies were carried out similarly, except that the indicated concentrations of cold BMP-7 (ϫ0.1, 1, 10, and 100) were added alongside the hot ligand (2 nM) for 3 h at 4°C. Cell surface complexes were cross-linked with disuccinimidyl suberate and quenched with 1 M glycine. The cells were then lysed with RIPA buffer containing protease inhibitors and immunoprecipitated at 4°C. Before immunoprecipitation lysate was removed for control gel anal-ysis. The immunoprecipitated proteins were resolved using SDS-PAGE. These gels were subsequently dried and exposed to an audioradiograph.
Surface Plasmon Resonance-BMP-2 (1600 response units) was immobilized on a CM5 sensor chip using amine coupling (sodium acetate pH 4.5). Soluble T␤RIII (R & D Systems) was diluted in running buffer (10 mM Hepes, pH 7, 0.15 M NaCl, 3 mM EDTA, and 0.005% Surfactant P20) and flowed at concentrations ranging from 7.81 to 250 nM for 5 min at a flow rate of 50 l/min. TGF-␤1 (900 response units) was immobilized in sodium acetate, pH 5.5. In the TGF-␤ studies, sT␤RIII (concentrations ranging from 8.75 to 560 nM) was flowed at a rate of 50 l/min for 3 min. The resulting sensograms were then fit using nonlinear least squares analysis and numerical integration of differential rate equations and the fits were then analyzed by considering the distribution of the residuals.
Smad1 Phosphorylation-PC-3 cells were plated at 125,000 cells/well and infected with adenovirus (50 multiplicity of infection) containing either non-targeting control short hairpin RNA (shRNA) or shRNA directed against human T␤RIII 24 h after plating. The cells were then incubated for 96 h, serumstarved for 5 h, and treated with the indicated concentrations of rhBMP-2 for 10 min followed by direct lysis. Smad1 phosphorylation was assayed by Western blot with phospho-Smad1 antibody (Cell Signaling), with total Smad1 antibody as a loading control (Cell Signaling).
Viral Injections and Collagen Gel Assays-Injections and assays were performed as previously described by Desgrosellier et al. (36) with the exception of the addition of vehicle (bovine serum albumin/HCl), 200 pM TGF-␤2, or 5 nM BMP-2, BMP-4, BMP-7, or GDF-5 12 h after placement of the explant on collagen. Each GFP-expressing cell was scored as epithelial, activated, or transformed as described (36). For the total number of explants and cells counted, refer to supplemental Tables 3 and 4.

T␤RIII Is a Cell Surface Receptor for BMP-2-
To determine whether T␤RIII functions as a BMP receptor, we expressed T␤RIII, along with the BMP receptors, ALK3, ALK6, or BMPRII, in COS-7 cells, which express low endogenous levels of these cell surface receptors and assessed BMP-2 binding by chemically cross-linking 125 I-BMP-2 to binding partners on the cell surface. As expected, 125 I-BMP-2 bound to ALK3 and ALK6, but not to BMPRII, which cannot bind ligand on its own (Fig. 1A). 125 I-BMP-2 was also detected bound to T␤RIII in the presence of ALK3, ALK6, and BMPRII (Fig. 1A, lanes 3, 5, and 7) suggesting that T␤RIII binds BMP-2. T␤RIII expression also modestly increased BMP-2 binding to ALK3 and ALK6, but did not confer ligand binding to BMPRII (Fig. 1A).
BMPRII binds BMP ligands only when in complex with either ALK3 or ALK6 (8). To determine whether T␤RIII affects the ability of BMPRII to bind BMP in the presence of ALK3 or ALK6, we expressed T␤RIII with these traditional BMP signaling complexes. As expected, BMPRII could be detected bound to 125 I-BMP-2 when co-expressed with ALK3 or ALK6 (Fig. 1B,  lanes 1 and 3), and T␤RIII did not significantly alter BMP-2 binding to BMPRII (Fig. 1B, lanes 2 and 4).
To determine whether BMP receptor expression was necessary for the ability of T␤RIII to bind BMP-2, we expressed T␤RIII alone. In the absence of ALK3 and ALK6, 125 I-BMP-2 formed a cross-linked complex with fully processed T␤RIII in a dose-dependent fashion (Fig. 1C, left) establishing that BMP-2 is able to bind T␤RIII independent of other ligand binding receptors.
BMPs bind heparan sulfate with high affinity (28,37) and T␤RIII is a heparan sulfate and chondroitin sulfate proteoglycan (38). These glycosaminoglycan modifications are important for basic fibroblast growth factor binding to T␤RIII (27), but not for TGF-␤ (38) or inhibin binding (39). To determine whether these glycosaminoglycan modifications were important for BMP binding to T␤RIII, we used a mutant of T␤RIII in which the serines (Ser-535 and Ser-546) necessary for glycosaminoglycan chain attachment are converted to alanines preventing this modification (T␤RIII⌬gag) (38). In these studies, the core protein of T␤RIII⌬gag was affinity labeled with 125 I-BMP-2 in a dose-dependent fashion (Fig. 1C, right) indicating that the heparan sulfate modifications were not necessary for BMP-2 binding to T␤RIII.
T␤RIII exists in two forms, a membrane bound form and a soluble form, sT␤RIII, derived from ectodomain shedding (38). sT␤RIII consists of the extracellular domain of T␤RIII and is able to bind TGF-␤, sequester ligand from the cell surface receptors, and antagonize TGF-␤ signaling (38,40). To determine whether sT␤RIII is able to bind BMP-2, we exposed recombinant, purified sT␤RIII to 125 I-BMP-2. As with membrane-bound T␤RIII, sT␤RIII was affinity labeled with 125 I-BMP-2 in a dose-dependent fashion (Fig. 1D, lanes 2-4), with a BMP-2 binding pattern similar to that of the well characterized T␤RIII ligand, TGF-␤1 (Fig. 1D, lane 1). These data demonstrate that sT␤RIII is able to bind BMP-2 and confirm that the binding of BMP-2 to the extracellular domain of T␤RIII is direct.
Kinetics and Affinity of BMP Binding to T␤RIII-To characterize the interaction between T␤RIII and BMP-2 we used surface plasmon resonance (also known as BIAcore), a sensitive method to measure protein-protein interactions (41,42). BIAcore has been used to define the interactions of multiple TGF-␤ superfamily ligands with their receptors (6,(43)(44)(45). Both BMP-2 and TGF-␤1 were immobilized to a dextran sensor chip and purified sT␤RIII was the analyte. Upon mathematically fitting the response curves, the model that best fit the binding of T␤RIII to BMP-2 was the bivalent analyte (or avidity) model (Fig. 2 and supplemental  Table S1). This model also provided the best fit for TGF-␤1 binding to T␤RIII, based on previous BIAcore studies (44) and confirmed here (supplemental Fig. 1 . BMP-2 binds to the extracellular domain of T␤RIII. COS-7 cells expressing hemagglutinin-tagged ALK3, ALK6, or BMPRII (A); ALK3 or ALK6 and BMPRII (B) with or without wild-type (wt) human T␤RIII were exposed to 10 nM 125 I-BMP-2, cross-linked, immunoprecipitated, and separated by SDS-PAGE and detected by phosphorimaging. C, COS-7 cells expressing wt T␤RIII or T␤RIII with serine to alanine mutations at glycosaminoglycan attachment sites (T␤RIII⌬gag) were exposed to increasing doses of 125 I-BMP-2 (2, 5, 10, 25, and 50 nM, respectively) (top panel). Total cellular T␤RIII and ␤-actin are shown as expression and loading controls (bottom panels). D, soluble T␤RIII (sT␤RIII) was exposed to 125 I-TGF-␤1 (150 pM) as a positive control and increasing doses of 125 I-BMP-2 (2, 5, and 10 nM) followed by chemical cross-linking and immunoprecipitation with an antibody against the extracellular domain of T␤RIII. The proteins were then resolved by SDS-PAGE. The data are representative of three independent experiments.
both TGF-␤1 and BMP-2 (supplemental Table S2). Using the bivalent model, we established kinetic and thermodynamic constants for BMP-2 interacting with the ligand binding site on sT␤RIII, with data from three independent experiments establishing a dissociation constant of 10 M (Ϯ3.66) for BMP-2 and 5 M (Ϯ1.71) for TGF-␤1. These results are comparable with previously published reports for T␤RIII and TGF-␤1 (44).
These studies indicate that the affinity of T␤RIII for BMP-2 is on the same order of magnitude as the affinity of T␤RIII for TGF-␤1. Multiple Members of the BMP Subfamily Bind to T␤RIII-To determine whether T␤RIII could bind other BMP subfamily members, we radiolabeled representative members of the three distinct BMP subfamilies, BMP-4, BMP-7, and GDF-5 ( Fig. 3A) (1). 125 I-BMP-2, 125 I-BMP-4, 125 I-BMP-7, and 125 I-GDF-5 each formed a cross-linked complex with both the fully processed form of T␤RIII, along with T␤RIII⌬gag (Fig. 3B), suggesting that a broad range of BMP family members can bind to the core protein of T␤RIII. Intriguingly, there were subtle differences in the binding patterns of certain BMP subfamily members to T␤RIII, particularly to the core protein.
To define the specificity of the interaction of BMP with T␤RIII, we performed competition experiments with iodinated ligand in the presence of increasing concentrations of unlabeled ligand. Consistent with a previous report (46), we were unable to specifically compete off BMP-2 binding with excess cold ligand (data not shown). The nonspecific binding of 125 I-BMP-2 is likely due to the presence of basic residues at the amino terminus of BMP-2, as previously reported (46). Accordingly, we investigated specificity using BMP-7, which also binds T␤RIII, and for which these amino-terminal basic residues are not present. Here, 100-fold excess cold ligand successfully competed off 125 I-BMP-7 from wild-type T␤RIII (Fig. 3C) with ϳ40% nonspecific binding remaining ( Fig. 3C and supplemental Fig. S2). Unlabeled BMP-7 also competed with 125 I-BMP-7 for binding to T␤RIII⌬gag (Fig. 3D). Taken together, these data support the ability of T␤RIII to specifically bind a broad range of BMP subfamily members.
To establish the physiological relevance of BMP binding to T␤RIII, we investigated whether BMP family members bound to endogenous T␤RIII. For these studies, we used NIH3T3 cells, which abundantly express T␤RIII. Both 125 I-BMP-2 and 125 I-BMP-7 bound to high molecular weight complexes corresponding to the fully processed endogenous form of T␤RIII, which were specifically immunoprecipitated by a T␤RIII antibody, and not by preimmune serum, in a pattern similar to that of 125 I-TGF-␤1 (Fig. 3E). These data confirm that both BMP-2 and BMP-7 are able to bind to endogenous T␤RIII.
BMP Binds to Both T␤RIII Ligand Binding Domains-Our BIAcore data suggested two ligand binding sites for BMP-2 and TGF-␤1 on the core protein of T␤RIII. Consistent with this, previous studies have established two TGF-␤ binding regions on T␤RIII, with one in the membrane-distal half (Binding Region 1) and one in the membrane-proximal half (Binding Region 2) (Fig. 4A) (29,47). In contrast to TGF-␤, inhibin binds selectively to Binding Region 2 (39). To further investigate BMP binding to T␤RIII, we defined the regions of T␤RIII that mediate BMP binding. We expressed extracellular domain deletions of T␤RIII, including those lacking either Binding Regions 1 or 2 and then assessed their ability to bind BMP-2 and BMP-7. Both 125 I-BMP-2 (Fig. 4B) and 125 I-BMP-7 (Fig. 4C) exhibited a binding pattern identical to that of 125 I-TGF-␤1 (Fig. 4D). BMP-2 and BMP-7 bound T␤RIII mutants with either Binding Region 2 (Fig. 4, B and C, lanes 4 and 5) or Binding Region 1 (Fig. 4, B and C, lane 6) deleted. In contrast, when portions of both of these regions are deleted, no binding occurred for either BMP-2 or BMP-7 (Fig. 4, B and C, lane 3), providing further support for specific binding of both BMP-2 and BMP-7 to the other T␤RIII constructs. In addition, like TGF-␤1, BMP-2 and BMP-7 appear to preferentially bind Binding Region 1 (Fig. 4, B and C,  compare lanes 5 and 6), which is the region most similar to endoglin. These data establish that both BMP-2 and BMP-7 can bind to either of the two ligand binding motifs of T␤RIII, similar to TGF-␤1, validating the bivalent model for BMP binding to T␤RIII.
T␤RIII Enhances Ligand Binding to ALK-3 and ALK-6-As a co-receptor, T␤RIII has an established role in presenting ligand, leading to enhanced TGF-␤ binding to T␤RII and increasing TGF-␤ signaling (35), while also enhancing inhibin binding to ActRII to facilitate inhibin-mediated antagonism of activin signaling (15). We observed a slight increase in BMP-2 binding to ALK3 and ALK6 in the presence of T␤RIII (Fig. 1A).
To determine whether T␤RIII alters BMP binding to the BMP signaling receptors, the effect of increasing T␤RIII expression FIGURE 2. The bivalent analyte model best fits the kinetics for BMP-2 binding to T␤RIII using surface plasmon resonance (BIAcore). A, global fitting analysis was carried out on the response units of T␤RIII (the analyte) binding to rhBMP-2 (1600 response units) immobilized using amine coupling to a CM5 dextran sensor chip after subtracting out bovine serum albumin binding as background. T␤RIII was flowed over BMP-2 at a rate of 50 ml/min for 5 min at concentrations ranging from 7.81 to 250 nM. The data were then fit to kinetic models. Represented is the best fit of the data, the bivalent analyte (also known as the avidity model). This data are representative of three independent experiments. B, graphical representation of the residuals of the data fit to the bivalent analyte model. on the binding of either 125 I-BMP-2 or 125 I-BMP-7 to ALK3 and ALK6 was examined. T␤RIII significantly increased binding of 125 I-BMP-2 to ALK-3 in an expression-dependent manner (Fig. 5A), with a maximal 2-fold increase (Fig. 5C). Expression of T␤RIII also significantly enhanced 125 I-BMP-2 binding to ALK6 ϳ2-fold (Fig. 5B) and 125 I-BMP-7 (Fig. 5D) binding to ALK6 about 3-fold (Fig. 5E). These increases in binding were due to increasing T␤RIII expression and not due to altered ALK3 or ALK6 levels. T␤RIII expression did not enhance BMP-4 binding to ALK6 (supplemental Fig. 3) nor BMP-2 binding to BMPRII in the absence of either ALK3 or ALK6 (Fig. 1A, lanes 6 and 7), suggesting that T␤RIII does not function to confer BMP ligand binding to BMP receptors unable to bind BMP subfamily members independently. Taken together, these data suggest that one function for T␤RIII in binding BMP subfamily members is to enhance BMP binding to their respective ligand binding receptors, without altering ligand binding specificity.
Loss of Endogenous T␤RIII Expression Reduces BMP Responsiveness-T␤RIII-mediated ligand presentation to the signaling receptors increases signaling by the respective ligand, as has been demonstrated for TGF-␤ (35) and inhibin (15). To assess whether T␤RIIImediated BMP presentation to BMP receptors regulated BMP signaling, we used shRNA to decrease endogenous T␤RIII expression and assessed effects on BMP-mediated Smad1 phosphorylation. As a model system we used the human prostate cancer cell line, PC-3, which express moderate levels of endogenous T␤RIII (data not shown), express BMPRII, ALK3, and ALK6, and are BMP responsive, including BMP-induced Smad1 phosphorylation (48). PC-3 cells were infected with non-targeting control shRNA and human T␤RIII shRNA, which concurrently expresses DS-Red fluorophore. Fluorescent images demonstrated similar infection efficiency and phase-contrast images demonstrated similar viability (data not shown). shRNA to T␤RIII consistently decreased endogenous T␤RIII expression by greater than 50%, whereas the non-targeting control shRNA had no effect on endogenous T␤RIII expression, as previously reported (31, 32) (data not shown). This loss of T␤RIII expression was associated with a significant decrease in BMP-2-mediated Smad1 phosphorylation, particularly at low BMP-2 concentrations, where phospho-Smad1 induction was approximately 50% of that in non-tar- A, evolutionary tree diagram generated by the MacVector program from NCBI sequence alignment of known T␤RIII ligands and the BMP members used in this study. B, COS-7 cells expressing wild type (wt) T␤RIII or T␤RIII⌬gag were exposed to 125 I-BMP-2, 125 I-BMP-4, 125 I-BMP-7, 125 I-GDF-5, or 125 I-TGF-␤1 as indicated and chemically cross-linked followed by immunoprecipitation with an antibody to the cytoplasmic tail of T␤RIII. Total cellular T␤RIII is shown as an expression control (bottom panel). *, indicates a nonspecific band. C and D, COS-7 cells expressing wild type T␤RIII (C) or T␤RIII⌬gag (D) were simultaneously exposed to 2 nM 125 I-BMP-7 in the presence of increasing amounts of cold BMP-7 (0.2, 2, 20, and 200 nM) as indicated, followed by chemical cross-linking and immunoprecipitation. E, NIH3T3 cells were exposed to 125 I-BMP-2, 125 I-BMP-7, and 125 I-TGF-␤1. All lysates were immunoprecipitated with either preimmune serum or a T␤RIII antibody, separated by SDS-PAGE, and detected by phosphorimaging. The data are representative of three independent experiments.
geting control cells (Fig. 6, A and B). These data suggest that T␤RIII serves to present BMP-2 to either ALK3 or ALK6 and increase cellular sensitivity to BMP-2.

BMP-2-specific Interaction with T␤RIII Yields a Functional
Response-To establish further support for a role for T␤RIII in BMP signaling, we used a well characterized model of T␤RIII signaling that scores cell invasion into a three-dimensional collagen matrix to determine whether BMP-2 binding to T␤RIII initiated signaling (49). During embryonic development, a subset of endothelial cells in the atrioventricular (AV) cushion of the heart undergo EMT as an early step in valve formation. This process has been well studied using explants of the AV cushion that have been placed onto collagen gels. Endothelial cells from this AV cushion form a compact epithelial sheet composed of rounded, tightly packed cells on the surface of the collagen, whereas transformed cells are identified morphologically and functionally as those that elongate and invade into the collagen matrix. Endothelial cells in these AV cushions express T␤RIII, as well as T␤RII and T␤RI, along with the BMP receptors, BMPRII and ALK3, and undergo EMT in response to exogenous TGF-␤ (18,50). In contrast to AV cushion endothelial cells, ventricular endothelial cells lack T␤RIII and do not undergo EMT, even in response to an excess of TGF-␤. However, expression of T␤RIII in these ventricular endothelial cells results in TGF-␤2-induced EMT, demonstrating a unique requirement for T␤RIII and suggesting a non-canonical role for T␤RIII in addition to ligand presentation (18).
This in vitro assay system, where EMT is dependent on the presence of T␤RIII, is currently the only described assay for T␤RIII signaling (18). To determine whether BMP-2 is also involved in T␤RIII-mediated transformation, chick ventricular endothelial cells expressing either GFP or T␤RIII and GFP were incubated with TGF-␤2 or BMP-2. Neither ligand alone induced transformation of control infected cells (Fig.  7E). However, expression of T␤RIII conferred BMP-2-and TGF-␤2-induced EMT (Fig. 7E) as measured by a 2-fold increase in the percentage of transformed cells (cells elongating and invading the collagen gel (Fig. 7, C and D)), and a concomitant decrease in the percentage of epithelial cells (cells rounded and remaining on the surface of the gel (Fig.  7, A and B)). Importantly, BMP-2 (5 nM) induced transformation to a similar extent as 200 pM TGF-␤2, and both BMP-2-and TGF-␤2-induced transformation were entirely dependent on T␤RIII expression. These data demonstrate that BMP-2 requires T␤RIII to mediate this biological response, consistent with T␤RIII and BMP-2 functioning as a receptor-ligand pair. Because other BMP family members also bind to T␤RIII, we investigated the ability of these BMP members (BMP-4, BMP-7, and GDF-5) to induce EMT in this model system. Surprisingly, only BMP-2 induced EMT, suggesting a specific functional role for BMP-2 binding to T␤RIII in mediating EMT during heart development (Fig. 7F).

DISCUSSION
Here we demonstrate, for the first time, that T␤RIII is able to bind BMP, another class of ligands within the TGF-␤ superfamily. Importantly, T␤RIII is able to bind a broad range of BMP ligands, including BMP-2, BMP-4, BMP-7, and GDF-5. We demonstrate that this binding is specific, through both competition studies and studies demonstrating that disruption of both ligand binding domains prevents BMP binding to T␤RIII. Functionally, T␤RIII serves as a BMP co-receptor by increasing BMP binding to the BMP signaling receptors, ALK3 and ALK6, and increasing BMP signaling, and as a BMP receptor in a bio-logically relevant system. Whereas glycosylphosphatidylinositol-linked co-receptors have recently been identified for BMP (DRAGON, RGMa, and hemojuvelin) these receptors each have limited tissue distribution (12). In contrast, T␤RIII is a ubiquitously expressed transmembrane receptor (51). T␤RIII has previously been identified as a co-receptor for both TGF-␤ and inhibin based on its ability to bind these ligands and enhance their cellular effects. Data presented here suggest that T␤RIII functions similarly for the BMP family of ligands. Given the broad tissue distribution of T␤RIII and its ability to bind all major classes of ligands in the TGF-␤ superfamily, T␤RIII is poised to act as a major orchestrator of TGF-␤ superfamily signaling. As members of the TGF-␤ superfamily are able to antagonize each other, including BMP-7 antagonism of TGF-␤-induced fibrogenesis in mesengial cells (52) and distinct inhibin antagonism of activin and BMP signaling (53), T␤RIII may be the common component for these TGF-␤ superfamily ligands mediating this antagonism. Consistent with this hypothesis, we have demonstrated that BMP-2 can compete with TGF-␤1 for binding to T␤RIII (data not shown). Current studies are delineating the hierarchy of ligand binding to T␤RIII to further define the role of T␤RIII as a moderator of TGF-␤ superfamily signaling.
Here we demonstrate that BMP-2 binding to T␤RIII elicits a functional response in ventricular endothelium. These data are consistent with a suggested role for BMP-2 in normal cardiac development. A recent report demonstrated that conditional ablation of BMP-2 in heart muscle, where it is expressed during development, results in failure of the T␤RIII expressing endothelial cells in the adjacent valve forming region of the heart to undergo EMT (54,55). Furthermore, we show that T␤RIII has enhanced functional interaction with ALK3 in the presence of BMP-2. Conditional ablation of ALK3 from the endothelium also results in a failure of EMT in the valve forming regions (50). Taken together, these data suggest that BMP-2 produced by the myocardium may act directly through T␤RIII on endothelial cells in the heart to stimulate EMT and valvulogenesis. . T␤RIII presents BMP to the BMP type I receptors. COS-7 cells expressing either ALK3 (A) or ALK6 (B and D), in the absence or presence of increasing amounts of T␤RIII (0, 0.5, 1.0, 1.5, and 2.0 g, respectively), were exposed to 2 nM radiolabeled ligand (BMP-2 (A and B) or BMP-7 (D), as indicated) followed by chemical crosslinking and immunoprecipitation of the BMP receptor using an hemagglutinin antibody. The proteins were separated by SDS-PAGE and detected by phosphorimaging. ␤-Actin is shown as a loading control (bottom panel). Images are representative of three independent experiments. Graphical representation of the average change in signal intensity of BMP-2 binding to ALK3 and ALK6 (C) or BMP-7 binding to ALK6 (E) from three independent experiments. Densitometry using ImageJ software was used to determine the signal intensity. The densitometry of each band was normalized to the signal intensity of the ligand bound to the respective BMP receptor without T␤RIII. Two-tailed Student's t test was used to determine statistical significance in comparison to the respective BMP receptor without T␤RIII; *, p Ͻ 0.05; **, p Ͻ 0.005; ***, p Ͻ 0.0005.
The differences noted between BMP ligand binding to T␤RIII and the biological effect of these BMP ligands may be explained by the relative expression level of the ligands and the identity of the receptors recruited into the signaling complex. The related TGF-␤ superfamily co-receptor, endoglin, participates in differentially activating ALK1 and ALK5 in endothelial cells, with low TGF-␤ concentrations activating ALK1 and stimulating angiogenesis, whereas higher concentrations activate ALK5 to inhibit angiogenesis (56 -58). The present data demonstrate that BMP-2 can cause functional recruitment of T␤RIII to ALK3 consistent with a role for ALK3 in stimulating endothelial cell transformation. In contrast, T␤RIII leads to enhanced BMP-7 binding to ALK6, which is not expressed in the endothelial cell model used here (50,59). Thus, these differences in receptor expression may account for the differences in functional response between BMP-2 and the other ligands (BMP-4, BMP-7, and GDF-5) and supports a model where specific ligands may initiate the assembly of particular receptor complexes to generate diversity in biological effect.
Members of the BMP family activate cellular responses through the formation of an active signaling receptor complex containing the type I and type II receptors (8). BMP binding data presented here suggest that T␤RIII does not alter the formation of this active complex. Future studies will be aimed at characterizing the effect of T␤RIII on the formation, stability, and activity of the signaling complex and determine whether T␤RIII alters ligand binding to the other BMP type II receptors, including ActRII and ActRIIB.
A number of BMP family members, including BMP-2, BMP-4, and BMP-7, have been associated with heart development. Here we demonstrate that T␤RIII is uniquely essential for BMP-2-induced EMT in the developing heart, as T␤RIII expression, did not confer BMP-4-or BMP-7-induced EMT. Our data are consistent with recently published data indicating that BMP-4 is dispensable for EMT in the heart (60) and that BMP-7 antagonizes TGF-␤1-induced EMT, whereas having no effect on EMT by itself (61). Thus, EMT is not a physiologically relevant assay for BMP-4 and BMP-7. These data further support the hypothesis that T␤RIII does not confer BMP function, but facilitates their function. Further investigation will be required to define the effect of T␤RIII on other BMP family ligands in physiologically relevant assays.
Binding of BMP to T␤RIII occurs through two ligand binding domains on the core protein of T␤RIII, similar to TGF-␤, and FIGURE 6. Loss of T␤RIII decreases BMP-sensitivity. PC-3 cells were infected with adenovirus expressing either a non-targeting control shRNA (NTC) or shRNA directed against human T␤RIII (shRIII). A, PC-3 cells were then treated with the indicated concentrations of BMP-2 (0, 0.5, 1.0, 2.0, and 5.0 nM, respectively) for 10 min and phospho-Smad1 and total Smad1 were detected by Western blot. Image is representative of two independent experiments. C, densitometry was performed on Western blot images using ImageJ. Graphed are raw densitometry units from two independent experiments for both shNTC and shRIII. Error bars represent mean Ϯ S.E. Two-tailed Student's t test was used to determine statistical significance; *, p Ͻ 0.05. FIGURE 7. T␤RIII is essential for BMP-2 and TGF-␤2-mediated transformation in chick heart ventricular explants. A-D, chick ventricular explants infected with adenovirus expressing GFP. Photomicrographs of representative explants to illustrate the phenotypes of the infected cells. A and B, brightfield and fluorescent images with the plane of focus at the surface of the collagen pad. A sheet of rounded, adjacent cells scored as epithelial cells (asterisks) is seen adjacent to the cardiac muscle. Cells scored as activated are no longer rounded and have separated from adjacent cells but remain on the surface of the collagen pad (arrowheads). C and D, brightfield and fluorescent images with the plane of focus within the collagen pad. Epithelial cells are found on the surface (asterisk). Elongated cells in the gel are scored as transformed (arrowheads). E, chick ventricular explants infected with adenovirus expressing either GFP or T␤RIII and GFP were incubated with either vehicle (bovine serum albumin/HCl), 200 pM TGF-␤2, or 5 nM BMP-2 as indicated. F, T␤RIII-dependent EMT is specific to BMP-2. Infected explants were incubated with 5 nM BMP-2, BMP-4, BMP-7, and GDF-5 as indicated. After incubation for 36 h, the explants were fixed and GFP-expressing cells were scored as epithelial, activated, or transformed. The data are a graphical representation of the average percentage of the total GFP cells scored as epithelial, activated, or transformed from three independent experiments (refer to supplemental Tables S3 and S4 for actual counts). Two-tailed Student's t test was used to determine statistical significance.
does not require the heparan sulfate and chondroitin sulfate modifications to the extracellular domain of T␤RIII. Whereas our data indicate that the glycosaminoglycan chains are not necessary for BMP binding, they may still contribute to enhance or alter ligand binding. The mechanisms regulating post-translational processing of T␤RIII are not well understood and likely to be cell type-specific. 3 The effect of differential post-translational processing on altering interactions of TGF-␤ superfamily members with T␤RIII warrants additional investigation.
In addition to glycosaminoglycan modifications, the extracellular domain of T␤RIII is proteolytically cleaved from the membrane and shed into the extracellular environment. Here we demonstrate that recombinant, purified sT␤RIII can bind BMP, indicating that the interaction is direct and that anchorage to the cell membrane and proximity of other BMP binding components are not necessary for ligand binding. BMP is secreted into the extracellular environment as an active ligand and the bioavailability of BMP is tightly controlled by a number of soluble BMP antagonists that bind ligand and sequester BMP from the signaling receptors, including Noggin, Chordin, Follistatin, and gremlin (4,7). The ability of sT␤RIII to bind BMP-2 suggests that sT␤RIII may be an additional mechanism by which the bioavailability of BMPs is regulated.
Expression of both BMP and T␤RIII are essential during embryonic development. Here we demonstrate that T␤RIII is important for BMP-induced EMT in ventricular cells of the chick heart. Therefore, defects in development of the T␤RIII knock-out mice may not only be due to alterations in TGF-␤ signaling, but also alterations in BMP signaling. In addition to the role of BMP in development, alterations in the BMP signaling pathway have been linked to a number of hereditary human diseases, including primary pulmonary hypertension, juvenile polyposis syndrome, ovarian dysgenesis 2, and A2 brachydactyly (62,63). BMP signaling also has an emerging role in regulating cancer biology with effects on glioblastoma (64), along with breast (65,66), ovarian (67), pancreatic (68), colon (69), and prostate cancer cells (70). The expression of T␤RIII is lost in a number of these same cancer types, including breast (33), ovarian (71), pancreatic (32), and prostate cancer (72,73). Here we demonstrate that loss of T␤RIII expression results in a decrease in cellular sensitivity to BMP, as assayed by Smad1 phosphorylation. Further defining the contribution of T␤RIII to BMP signaling will aid in establishing the mechanism by which T␤RIII functions during tumorigenesis, and whether alterations in T␤RIII expression or function are linked to other human diseases.