ActRII Extracellular Domain Complex Provides New Insights into the Cooperative Nature of Receptor Assembly

Molecular Cell, Vol. 11, 605–617, March, 2003, Copyright 2003 by Cell Press

The BMP7/ActRII Extracellular Domain Complex Provides New Insights into the Cooperative Nature of Receptor Assembly Jason Greenwald,1 Jay Groppe,1 Peter Gray,2 Ezra Wiater,2 Witek Kwiatkowski,1 Wylie Vale,2 and Senyon Choe1,* 1 Structural Biology Laboratory and 2 Clayton Foundation Laboratories for Peptide Biology The Salk Institute La Jolla, California 92037

Summary Activins and bone morphogenetic proteins (BMPs) elicit diverse biological responses by signaling through two pairs of structurally related type I and type II receptors. Here we report the crystal structure of BMP7 in complex with the extracellular domain (ECD) of the activin type II receptor. Our structure produces a compelling four-receptor model, revealing that the types I and II receptor ECDs make no direct contacts. Nevertheless, we find that truncated receptors lacking their cytoplasmic domain retain the ability to cooperatively assemble in the cell membrane. Also, the affinity of BMP7 for its low-affinity type I receptor ECD increases 5-fold in the presence of its type II receptor ECD. Taken together, our results provide a view of the ligand-mediated cooperative assembly of BMP and activin receptors that does not rely on receptor-receptor contacts. Introduction A large number of secreted proteins interact with transmembrane receptor kinases to transmit intercellular signals. Structural rearrangements of these receptors are key to the initiation of an intracellular signal cascade (Hunter, 2000; Ullrich and Schlessinger, 1990). Receptor tyrosine kinases (RTKs) are a well-characterized class of receptors for which ligand binding induces oligomerization in the membrane allowing crossphosphorylation of the intracellular kinase domains. Ligands such as human growth hormone (de Vos et al., 1992), erythropoietin (Syed et al., 1998), and fibroblast growth factor (Pellegrini et al., 2000) act as a centerpiece around which the receptors cluster, whereas ligands such as transforming growth factor-␣ (Garrett et al., 2002) and epidermal growth factor (Ogiso et al., 2002) induce conformational changes in their receptors that influence the direct oligomerization of these receptors exclusively through receptor-receptor interactions. Regardless of the means by which they are brought together, the assembly of RTKs is a pivotal mechanistic step in transmembrane signaling. In contrast to RTKs, the processes in the membrane that lead to assembly and signal propagation of receptor serine/threonine kinases (RSKs) are less well understood. This latter class is mainly comprised of re* Correspondence: [email protected]

ceptors for transforming growth factor-␤ (TGF-␤) superfamily ligands. The TGF-␤ superfamily, including the activins and bone morphogenetic proteins (BMPs), consists of a large number of structurally related polypeptide growth factors that regulate a broad array of cellular processes including cell proliferation, differentiation, adhesion, and death (Massague´, 1998). Both the ligands of this superfamily and their receptors have highly conserved structural motifs. The ligands are disulfide-linked dimers that belong to the cystine-knot growth factor family (McDonald and Hendrickson, 1993) and interact with two distinct types of receptors, termed type I and type II. Characteristic structural features of the receptors are a threefinger toxin fold (Greenwald et al., 1999) in the ligand binding extracellular domain (ECD), a single transmembrane spanning domain, and an intracellular serine kinase domain. The distinction between the two types of receptors is based on sequence conservation in their kinase domain and the presence in type I receptors of the Gly-Ser-rich (GS) juxtamembrane activation domain. Although the two types of receptors are structurally related, they serve different roles in signaling. After ligand is bound concurrently to both of the receptor types, the constitutively active kinase domain of the type II receptor phosphorylates the type I receptor on multiple sites in its GS activation domain (Wrana et al., 1994). The activated type I receptor then initiates the downstream signaling process by phosphorylating Smad proteins in the cytoplasm (Massague´ and Wotton, 2000). The 2-fold symmetric nature of the ligand extends to the formation of the signaling complex so that there are four receptors (two type I and two type II) binding to one ligand dimer. The practical significance of the 2-fold symmetry in TGF-␤ signaling is not known; however, a kinase-defective type I receptor mutant can functionally complement an activation-defective type I receptor mutant (WeisGarcia and Massague´, 1996), indicative of cooperative interaction between cytoplasmic domains of receptors. Studies of the receptor signaling mechanism of TGF-␤ and activin have demonstrated that the type II receptor is a high-affinity receptor and that its presence is required for the type I receptor to bind to the ligand (Attisano et al., 1996; Lebrun and Vale, 1997; Wrana et al., 1994). This cooperativity is less pronounced and reversed with BMP2 and 4 which were shown to have moderately higher affinity for their type I than their type II receptor (Knaus and Sebald, 2001; Koenig et al., 1994). However, the homologous ligand, BMP7 (60% identical to BMP2) retains the TGF-␤/activin-like preference for its type II receptor. These differences reflect the diversity of the more than 30 ligands in the superfamily. Furthermore, the relatively small number of receptors (seven type I and five type II) necessitates that the receptors have multiple specificities. The activin type II receptor (ActRII) has a particularly broad specificity, bridging subfamilies by binding to both activin and BMP ligands. Although it is a high-affinity receptor for activin and BMP7, ActRII, like BMPRII, is the lower-affinity receptor for BMP2. A broad specificity also exists in the type I

Molecular Cell 606

receptor, ALK2, transmitting signals from both BMP7 (Macias-Silva et al., 1998) and the distantly related Mu¨llerian inhibiting substance (MIS) (Clarke et al., 2001; Visser et al., 2001). Currently, the molecular mechanisms that underlie cooperative receptor assembly and the origin of multiple specificity and variable affinity in the TGF-␤ superfamily are unknown. The present study focuses on the functional and structural aspects of receptor assembly, with an emphasis on activin and BMP specificity. BMPs play a central role in many developmental processes including dorsoventral patterning and organogenesis, in addition to having multiple, diverse functions in mature organisms (Hogan, 1996; Kingsley, 1994; Lim et al., 2000). The temporal and spatial activity of BMPs is regulated during development at the level of expression and also by antagonists that bind to BMPs and prevent them from signaling (Zimmerman et al., 1996) while facilitating their diffusion to allow for proper gradient formation (Eldar et al., 2002). BMP7 is expressed in many tissues during embryogenesis and is essential for eye and kidney development and maintenance (Dudley et al., 1995; Luo et al., 1995). Since BMP7 can also induce new bone formation (Sampath et al., 1992), it is an attractive therapeutic agent for bone regeneration in adults (Pecina et al., 2001). Given the biological importance of the activin and BMP signaling pathways, there has been great interest in elucidating the structures of the ligand-receptor complexes. The structure of ActRII-ECD (Greenwald et al., 1999) allowed us to predict the ligand binding surface on its concave face based on the conservation of exposed hydrophobic residues. Subsequently, this interface was validated by mutational analysis (Gray et al., 2000). In addition, the type II interface on BMP2 has been mapped to the convex “knuckle” epitope (Kirsch et al., 2000b). Thus, it was a surprise to find that the binding surfaces in the TGF-␤3/TGF-␤RII-ECD crystal structure were different on both ligand and receptor (Hart et al., 2002). In this paper, we describe the crystal structure of ActRIIECD, in complex with BMP7. The structure is consistent with the predicted interfaces and is distinct from that of the TGF-␤ complex with its type II receptor. Because BMP2 and BMP7 are functionally and structurally very similar, our complex could be confidently combined with the published structure of BMP2 in complex with the BMP type Ia receptor ECD, BMPRIa-ECD (Kirsch et al., 2000a). The resulting model of the entire six-chain assembly (four receptor ECDs, one BMP dimer) indicates that there is no direct physical interaction between any of the receptor ECDs. Interestingly, truncated type I and type II receptors for BMP7 that lack their entire intracellular domain were found to crosslink to the ligand in a cooperative manner, indicating that the intracellular domains, although critically involved in signal transduction, are dispensable for complex formation with the ligand. We also find that the isolated ECDs show a moderate increase in affinity for their ligand in the presence of their high-affinity counterpart. Hence, the observed cooperative binding of type I and type II receptors may be a result of interactions between transmembrane domains of the two types of receptors and/or ligand conformational changes induced upon complex formation that lead to an increased affinity for the individual receptors. Furthermore, the restricted two-dimensional diffusion in

the membrane and the multivalent nature of the complex should increase its stability by coupling the individual ligand-receptor interactions. Results The BMP7/ActRII-ECD Complex The complex of BMP7 with ActRII-ECD was crystallized and the structure determined by multiple isomorphous replacement with anomalous scattering (MIRAS) at a resolution of 3.3 A˚ (Table 1). In parallel, free BMP7 was crystallized and its structure independently refined to 2.0 A˚ resolution. The complex consists of a single BMP7 dimer and two molecules of ActRII-ECD. The structural model includes residues 6–100 of ActRII-ECD (out of 1–102) and residues 36–139 of BMP7 (out of 1–139). The N-terminal 35 amino acids of BMP7 remain disordered, as in the unbound ligand (Griffith et al., 1996). Present in the model are two N-acetyl glucosamines on ActRIIECD and one branched mannose chain on BMP7. BMP7 contains a cystine-knot fold and has a single disulfide bond that covalently connects the two chains of the dimer (gold and rust in Figure 1). Within each monomer, two pairs of anti-parallel ␤ strands, forming first a short and then a long finger, stretch outward like wings from the cystine-knot core of the dimer. The characteristic curvature of the wing-like extensions creates concave and convex surfaces on the butterflyshaped ligand. The long ␣-helix (␣3) from one monomer packs against the other monomer in the concave surface of the dimer. As a representative three-finger toxin fold, ActRII-ECD (green in Figure 1) has a disulfide-bonded core as well, from which three pairs of anti-parallel ␤ strands reach outward like fingers from a palm. They are also curved, resulting in convex and concave faces on the molecule. In the complex, the concave face of ActRII-ECD and the convex face of the BMP7 wing, both exposed hydrophobic surfaces, form the binding interface. The hydrophobic interface, with a buried surface area of 676 A˚2 per subunit, is mainly composed of side chain interactions (Figures 2C and 2D), providing a structural scaffold on which the binding specificity can be encoded by amino acid identities. ECDs of the Type I and Type II Receptors Make No Contact The structure of the BMP7/ActRII-ECD complex was combined with the previously reported structure of the BMP2/BMPRIa-ECD complex (Kirsch et al., 2000a) to create a model for the heterohexameric BMP7/ActRII/ BMPRIa complex (Figures 3A and 3B). BMPRIa was placed in the model by overlaying BMP2 and BMP7 from the two binary complexes using all 103 C␣ atoms of BMP7 (rms fit of 0.683 A˚), resulting in a ternary complex model with excellent surface complementarity. In the model there is no direct interaction between the individual receptor ECDs. This is remarkable considering the paradigm of activin and TGF-␤ signaling in which the complex of the ligand and the type II receptor recruits the type I receptor (Attisano et al., 1996; Lebrun and Vale, 1997; Wrana et al., 1994). The C termini of the four receptors (yellow in Figure 3) all extend from the same side of the complex yet are quite distant from one an-

Assembly of Activin/BMP Receptors 607

Table 1. Crystallographic Data, Phasing, and Refinement of BMP7 and the BMP7/ActRII-ECD Complex

Beamline Wavelength (A˚) Number of observations Number of unique reflections Resolution range (A˚) Average I/␴I Completeness (%) Anomalous completeness Rsym (%) Riso (%) Phasing Number of sites Phasing power (centric/acentric) Isomorphous Anomalous Mean FOM (centric/acentric) Refinement Resolution range (A˚) Rcryst (%) Rfree (%)b Average B factor (A˚2 ) Rms deviation Bonds (A˚) Angles (deg) Number of atoms Protein/Sugar Water Ramachandran plot non-gly, -pro, -terminal residues in: most favored regions additional allowed regions generously allowed regions a b

BMP7 Native

BMP7/ActRII Native

BMP7/ActRII K2PtCl4

BMP7/ActRII K2PtCN4

SSRL 9-1 0.98 67,831 15,770 100–2 (2.08–2)a 17.5 93.3 (99.6)

SSRL 11-1 1.03 31,531 8,320 100–3.3 (3.42–3.3) 21.3 96.7 (98.6)

5.4 (33.3)

5.5 (33.2)

SSRL 7-1 1.08 12,438 4,394 100–3.69 (3.82–3.69) 15.2 69.6 (24.6) 54.3 5.8 (39.9) 43.0

ALS 5.0.1 1.00 50,683 5,685 100–3.69 (3.83–3.69) 14.0 94.0 (91.5) 80.4 8.8 (39.1) 18.2

2

6

1.21/1.50 –/0.36

0.67/0.74 –/0.56

0.60/0.45 40–2 22.8 24.4 42.3

100–3.3 23.8 27.9 71.2

0.03 2.462

0.032 2.978

849 81

1,695

81 (88%) 10 (10.9%) 1 (1.1%)

138 (78.45%) 35 (19.9%) 3 (1.7%)

Numbers in parentheses correspond to the highest resolution shell. Calculated from 5% of data not used in refinement.

other (II-II, 83 A˚, and I-I, 66 A˚). Since the model lacks ⵑ16 residues in ActRII and ⵑ9 in BMPRIa between the structurally defined C termini and the membrane, the intertransmembrane segment distances are difficult to predict. Nevertheless, with the given constraints, homomeric interactions of the transmembrane domains seem unlikely. The C termini of ActRII-ECD and BMPRIa-ECD in the four-receptor model are 27 A˚ apart, leaving open the possibility of heteromeric interactions between a type I and type II transmembrane segment. TGF-␤ and BMP/Activin Type II Receptor Interfaces Are Distinct The structure of the BMP7/ActRII-ECD complex raises several key issues regarding the universality of the TGF-␤ signaling mechanism. First, the interface observed in the BMP7/ActRII-ECD complex is completely distinct and nonoverlapping with that of the TGF-␤3/ TGF-␤RII-ECD complex (Hart et al., 2002). Although there is an unusually large conformational rearrangement of the TGF-␤3 dimer in the complex, the distinct interface of TGF-␤3/TGF-␤RII is consistent with mutation data on both TGF-␤ and TGF-␤RII (Hart et al., 2002; Qian et al., 1996). The sequence alignment in Figure 2 highlights the difference in the type II interfaces with green or pink on the interface residues. Second, the TGF-␤RII-ECD is bound close enough to the type I receptor interface (based on the BMPRIa site) that the receptor ECDs could

make direct contact, unlike the BMP and activin receptors. Thus, the juxtaposed surfaces of TGF-␤3 and TGF␤RII would provide a new interface for cooperative binding of the type I receptor. The relative position of ActRII versus TGF-␤RII is depicted in Figure 3C in which each of the type II receptor complexes is overlaid with one half of the BMP2/BMPRIa complex. The large structural difference between BMP2 and TGF-␤3 in the complexes (Figure 3C) necessitated that the alignment of these two ligands be performed using only 14 residues as reported in the TGF-␤3/TGF-␤RII-ECD structure (Hart et al., 2002). Therefore, the relative orientation between the type I and type II receptors in the TGF-␤3 model is variable and dependent on the alignment procedure. Activin and BMP Share a Conserved Type II Receptor Interface The residues of ActRII, BMP2, and activin predicted by mutagenesis to be important for ligand/type II receptor interactions are all at the interface of the BMP7/ActRIIECD complex (Figures 1C and 2C). Mutagenesis studies of BMP2 (Kirsch et al., 2000b) identified five residues where substitutions resulted in weaker binding to ActRII and BMPRII (Ser113, Leu115, and Leu125 to Ala; Ala58 and His63 to Asp; BMP7 numbering as in Figure 2). All of these residues make direct contacts to the receptor and are buried in the interface (pink in Figure 2C). We previously identified three residues (Phe42, Trp60,

Molecular Cell 608

Assembly of Activin/BMP Receptors 609

Phe83) that form a hydrophobic patch on the surface of ActRII whose individual substitutions with alanine abrogated activin signaling (Gray et al., 2000). These three residues directly contact the hydrophobic residues in BMP7 described above, emphasizing the importance of the hydrophobic nature of the interface. ActRII is a dual specificity receptor, transducing signals from both BMP and activin ligands (Yamashita et al., 1995), whereas BMPRII only binds BMPs. To demonstrate that ActRII binds activins and BMPs in the same manner, we used dominant-negative ActRII constructs truncated at residue 152 (ActRIItrunc) to probe six interfacial residues. ActRIItrunc, which lacks a cytoplasmic domain, was previously shown to act in a dominant-negative fashion with respect to activin signaling by forming nonproductive complexes with the activin type I receptor, ALK4 (Tsuchida et al., 1995). We find that the kinase-deficient ActRIItrunc suppressed BMP signaling in a similar manner (Figure 4B). Furthermore, all six substitutions introduced to the ligand binding surface of the ActRIItrunc caused comparable effects on BMP and activin responses, reducing ActRIItrunc’s ability to suppress signaling. In particular, substitutions at the core of the binding site were the most severe (Figure 4, constructs 6–8).

Conserved Interface Residues and Binding Specificity Our results with ActRIItrunc and those for activin, BMP2, and ActRII referenced above indicate that activin and BMP7 bind ActRII through a common interface. Therefore, the sole specificity of BMPRII for BMP can be viewed as a subset of the dual specificity of ActRII. Punt, a Drosophila type II receptor for the BMP2 ortholog Dpp (Letsou et al., 1995), binds to human activin with high affinity (Childs et al., 1993) and appears to be another dual specificity receptor. The fact that it transduces BMP and activin-like signals (Brummel et al., 1999) and the recent identification of an activin-like ligand in Drosophila (Kutty et al., 1998) provide further evidence that Punt is an ortholog of ActRII rather than BMPRII. In addition, mature Drosophila activin is 62% similar to human, while the set of type II interface residues is 78% similar (Lys102 is identical). The positive charge of Lys102 in activin (Leu125 in BMP7) has been shown to be important for binding to ActRII (Wuytens et al., 1999). This residue in BMP7 is buried in the interface and contacts Trp60 on ActRII. Modeling a lysine at position 125 in BMP7 places its ⑀-amino group within hydrogenbonding distance of the ␥-carboxylate of Glu29 in ActRII and allows for a hydrophobic packing of the aliphatic arm of the lysine side chain against Trp60. Since Glu29 is conserved in all known activin binding type II receptors (ActRIIa, ActRIIb, and Punt) but absent in all others,

glutamate at this position may be a specificity determinant by interacting with Lys102 of activin. Given the similarities between the receptors, the simplest explanation for their different specificities would be that BMPRII either lacks a critical activin binding component or contains a feature that diminishes activin binding. Figure 2D demonstrates that the conservation of receptor residues at the BMP7/ActRII interface is correlated with the conservation of the ligand residues that they contact. Hence, the pairs of nonconserved positions appear to have coevolved as specificity determinants. We probed two nonconserved interface residues on ActRII, replacing them with their counterparts in BMPRII, to see if they were responsible for the lack of activin binding in BMPRII. L61S or V81Y substitutions on ActRIItrunc indeed diminished its dominant-negative effect not only for activin signals but also for BMP7 (Figure 4). Other differences between the two receptors may account for their specificity, including a three-residue insertion in BMPRII at the interfacial A loop (blue in Figures 1 and 2) which could interfere with activin binding. As well, the charged mobile loop (M loop, Figures 1 and 2), which is conserved only among activin binding type II receptors (including Punt), may provide an activin-specific interaction that is lacking in BMPRII. Even though the M loop is near the interface in the BMP7/ActRII-ECD complex, it remains highly mobile as in unbound ActRII and is not part of the binding interface. The most striking feature of the BMP7/ActRII interface is that it consists almost exclusively of side chain-side chain interactions (83% of contacts; Figure 2D) and hydrophobic interactions (72% of contacts). Here, a hydrophobic contact is defined as two residues whose nonpolar atoms come within 4 A˚ of each other. Hydrophobic surfaces are often the primary source of the binding energy in protein-protein complexes (Clackson and Wells, 1995), and the large percentage of hydrophobic contacts in the BMP7/ActRII interface can support its high affinity despite the moderate amount of buried surface area (676 A˚2 per molecule). BMPRIa has a much larger binding surface with BMP2 (1130 A˚2 ), but only 27% of the contacts are between two hydrophobic groups, so the difference in area of their buried hydrophobic surfaces is not as significant. The TGF-␤3/ TGF-␤RII interface (515 A˚2 ) is also less hydrophobic in nature with 37% of the contacts between two nonpolar groups. In contrast to the diverse percentages of hydrophobic contacts, the buried nonpolar surface areas (carbon and sulfur) are similar: 481 A˚2 buried in BMP7/ ActRII, 683 A˚2 in BMP2/BMPRIa, and 306 A˚2 in TGF␤3/TGF-␤RII. Interestingly, the structurally similar ActRII and BMPRIa bind through their homologous concave surfaces. In fact, the positions in BMPRIa that correspond to the five core residues in the ActRII interface

Figure 1. The Structure of the BMP7/ActRII-ECD Complex (A and B) Ribbon diagrams of the BMP7/ActRII-ECD complex (A) with the 2-fold symmetry axis vertical and the membrane facing side at the bottom and (B) the view from above (BMP7, gold and rust; ActRII-ECD, green; cystine sulfurs, yellow space-filling). (C) Stereo view of the interface between BMP7 and ActRII in an orientation close to (A). The residues within 4 A˚ of the binding partner as well as Glu29 are displayed as balls and sticks. Glu29 and those residues whose mutations are known to affect binding (pink, Figure 2) are labeled. In (A)–(C) significant conformational changes are highlighted (dark blue), and in (C) they are overlaid with the unbound conformations (light blue). This figure was made using MOLSCRIPT (Kraulis, 1991).

Molecular Cell 610

Figure 2. Sequence Alignments of TGF-␤ Superfamily Ligands and Type II Receptor ECDs (A and B) Residue numbers above the sequences refer to ActRII and BMP7 and secondary structure elements refer to the complex. Residues that come within 4 A˚ of the binding partner at the type II interfaces are boxed and shaded green or pink, and those at the type I interface (BMP2) are boxed. The individual residues of activin, BMP2, and ActRII that have been shown by mutagenesis to be important for type II receptor binding are shaded in pink. The cysteines are shaded yellow and the conserved cysteines of the folds are boxed. The regions of conformational change (A and B loops) are lettered in blue. The alignments were performed by CLUSTALW (Thompson et al., 1994) using profiles from a structural alignment caclulated by STAMP (Russell and Barton, 1992). The gray boxes indicate the regions defined by the STAMP alignment procedure. The cysteine score in the Gonnet matrix was modified to ensure correct alignment of the conserved cysteines. (A) and (B) were made using ALSCRIPT (Barton, 1993). (C) A space-filling model of the complex dissociated by a 180⬚ rotation of ActRII from the right half of the complex in Figure 1A. The colors are as in (A) and (B) with the BMP2 mutations mapped onto BMP7. (D) Catalog of BMP7-ActRII contacts showing the types of interactions and the conservation of each residue. The residue colors are as in (C)

Assembly of Activin/BMP Receptors 611

Figure 3. The Model of the BMP7/ActRII/BMPRIa Six-Chain Signaling Complex BMPRIa (purple) was placed in the complex by aligning the BMP2/BMPRIa structure (Kirsch et al., 2000a) with BMP7. (A) Side view as in Figure 1A is shown as a solvent accessible surface. The horizontal line represents the plane of the membrane. (B) Bottom view (opposite from Figure 1B). Sugars are in black. The C termini are marked with yellow dots and the horizontal distances between them as projected onto the plane of the membrane are 83 A˚ for type II-type II, 66 A˚ for type I-type I, and 27 A˚ and 68 A˚ for type Itype II. (A) and (B) were prepared with DINO (Philippsen, 2001). (C) Stereo view of ActRII and TGF-␤RII bound to their respective ligands overlaid with the BMP2/BMPRIa structure. The BMP7/ActRII complex was aligned as in (A) using the entire ligand. In order to overlay the TGF-␤RII binding site (tip of finger 2 on TGF-␤3) with BMP2, only 14 residues (86–92 and 98–104 of BMP2) were used for the alignment (Hart et al., 2002). The color scheme is BMP2, white; BMP7, gold; ActRII, green; TGF-␤3, blue; TGF-␤RII, red; BMPRIa, purple. The receptors and their C termini are labeled.

(pink in Figure 2) are central to the BMP2/BMPRIa interface. Truncated Types I and II Receptors Bind Cooperatively Expanding on the observation that truncated type II receptors lacking kinase domains act in a dominant-negative manner by forming signaling-impaired complexes with type I receptors, we tested whether or not the intracellular domain of either receptor is required for complex formation. By constructing type I and type II receptors

that are truncated close to the transmembrane segments and tagged with different antibody epitopes, we could test for crosslinking of 125I-labeled ligands to cells expressing combinations of truncated and full-length receptors. Figures 5A and 5B show that 125I-activin crosslinked to FLAG-ALK4trunc upon coexpression with myc-ActRIItrunc but not when expressed alone. Likewise, 125I-BMP7 crosslinked to its type I receptor, ALK2 or myc-ALK2trunc only upon coexpression with FLAGActRII or FLAG-ActRIItrunc (Figures 5C and 5D). Lanes 2 and 3 of Figures 5A (activin) and 5C (BMP7) demonstrate

and their conservation (BMP7 with activin and ActRII with BMPRII) is shown as black bars for identical residues, gray for conservative changes, and white for nonconservative changes. The interaction type is colored yellow for nonpolar-nonpolar, blue for h-bonded, and white for nonpolarpolar. *, the conserved hydrophobic core residues of the interface; –, main chain-side chain contacts; all others are side chain-side chain.

Molecular Cell 612

Figure 4. Mutation of Selected ActRII-ECD Residues Inhibits the Ability of Kinase-Deleted ActRII to Function as a Dominant-Negative HepG2 cells were transfected with ActRIItrunc constructs: (1) wildtype, (2) K56A, (3) L61S, (4) V81A, (5) V81Y, (6) F83A, (7) F42A, (8) W60A, or (9) empty vector. Cells were subsequently, treated with vehicle or (A) 10 nM activin A or (B) 10 nM BMP7 and resulting ligand-dependent fold induction of luciferase activity is shown.

that ActRII is crosslinked to either ligand when expressed alone, independent of its cytoplasmic domain. Lanes 4 and 5 of Figures 5B (activin) and 5D (BMP7) are blank, demonstrating the inability of type I constructs to crosslink to their ligand when expressed alone. However, lanes 6–9 of Figures 5B and 5D show that the type I receptors (full-length or truncated) invariably bind and crosslink to the ligand when coexpressed with a type II receptor (full-length or truncated). The presence of both receptor types in lanes 6 and 7 of Figures 5C and 5D indicates that the BMP7 complex that is formed in the membrane is sufficiently stable to remain intact during immunoprecipitation (IP). Although FLAG-ActRIItrunc shows up as a very faint band in the anti-FLAG IP when cotransfected with ALK2 (Figure 5C, lane 8), it crosslinks efficiently to BMP7 as seen in the anti-ALK2 IP (Figure 5D, lane 8). In this case, the Lys residue in the FLAG epitope may be involved in a crosslinking event which would block its IP while the noncrosslinked FLAG-ActRIItrunc efficiently pulls down a crosslinked ALK2. Conformational Changes and Cooperative Interactions in Receptor ECDs In light of the structural model, the cooperative crosslinking with truncated receptors suggests that there may be some ligand-mediated effects on the extracellular side of the membrane. Although we cannot be certain

of their origins, there are two large structural changes in the BMP7/ActRII complex that may provide sources of allostery for a cooperative interaction. Comparison of the free and bound conformations of BMP7 and ActRII-ECD shows that a loop in ActRII-ECD near the binding interface switches conformations producing displacements of up to 19 A˚ (A loop, blue in Figures 1 and 2). This motion results from slight rearrangements in the interfacial residues that propagate into the A loop. A second intriguing movement occurs in BMP7 at the type I interface, far from its interface with ActRII. The loop preceding helix ␣3 of BMP7 moves into a conformation resembling that of the loop in the BMP2/BMPRIa complex (B loop, blue in Figures 1 and 2). Comparison with the B loop conformation in free BMP7 suggests that the new conformation is required to accommodate BMPRIa binding, although we cannot eliminate the possibility that crystal packing influences this loop movement. To further probe the roles of the ECDs in the formation of a heterotetrameric receptor complex, we measured the interactions between the isolated ECDs of ActRII, ALK2, and Thickveins (Tkv) with BMP2, BMP7, and activin A by surface plasmon resonance (BIAcore). In these experiments, one protein was immobilized on a flexible dextran matrix (the surface) and the other (the analyte) flowed over the surface while interaction between the two proteins was measured as a mass increase on the surface. Similar to published BIAcore studies with TGF-␤ proteins (De Crescenzo et al., 2001; Hatta et al., 2000; Kirsch et al., 2000b; Knaus and Sebald, 2001), we find that immobilized ActRII-ECD displays a 150- to 1500fold higher affinity for its ligands than does ActRII-ECD for the immobilized ligands (Table 2). Two immobilized receptor molecules may bind to a single ligand, thus increasing the apparent affinity through a proximity effect. If the ligand is immobilized, two receptor molecules bind independently, at the lower affinity of a single receptor. Likewise, immobilized Tkv, the ECD of the Drosophila type I BMP receptor, has a higher affinity for BMP2 (3.6 nM versus 200 nM). As expected, the invertebrate receptor Tkv-ECD is comparable to BMPRIa-ECD in its affinity for BMP2 (Kirsch et al., 2000b; Penton et al., 1994). Although the immobilized receptors have higher affinities, they are not suited for measuring cooperative enhancements because of the experimental need for either two distinct surfaces (immobilized highaffinity receptor) or very high concentrations of the highaffinity receptor (immobilized low-affinity receptor). Using the immobilized ligands, we could then measure on a single surface the affinity of one receptor in the presence of a near-saturating concentration of another receptor in order to detect any cooperative increase in affinity. Tkv-ECD induced a 2-fold increase in the affinity of ActRII-ECD for BMP2 (5.5 to 2.3 ␮M). Similarly, ActRIIECD caused a 5-fold increase in the affinity of ALK2ECD for BMP7 (143 to 28 ␮M). The magnitude of the cooperativity may be underestimated because, due to experimental limitations, we used concentrations of the high-affinity receptors that did not saturate the ligand surface. The significance of the relatively small magnitude is validated by the fact that the affinity of ActRIIECD for activin or BMP7 is unchanged with the addition of the nonspecific (for activin) or lower-affinity (for BMP7) receptor Tkv (Table 2).

Assembly of Activin/BMP Receptors 613

Figure 5. Activin and BMP7 Crosslinking to Cell Surface Receptors 293T cells transfected with the indicated receptor constructs were treated with 125I-labeled activin or BMP7, washed, and then crosslinked with DSS. The cells were solubilized, split into two pools, and precipitated with either a type I or type II receptor-specific antibody. (A) Crosslinking to 125I-activin and IP with anti-myc (ActRII). (B) Crosslinking with 125I-activin and IP with anti-FLAG (ALK4). (C) Crosslinking to 125I-BMP7 and IP with anti-FLAG (ActRII). (D) Crosslinking to 125I-BMP7 and IP with anti-myc (ALK2). The full-length ALK2 has no epitope tag so lanes 4, 6, and 8 were precipitated with ALK2-specific antibodies.

Discussion A general assumption has been that the TGF-␤ ligands all bind to their receptors and signal through them in a similar manner; however, we show that the BMP7/ActRII interface is distinct from that of TGF-␤/TGF-␤RII. This disparity in type II receptor binding modes is surprising in light of the conserved structural framework and many functional parallels in the TGF-␤ superfamily. We have shown that ActRII binds to BMP7 and activin with the same interface and that it shares the same binding site

on BMPs with BMPRII (Kirsch et al., 2000b). A sequence alignment of the type II receptors shows that many interface residues in ActRII are conserved in BMPRII (Figure 2B) but not in TGF-␤RII or MISRII. Interestingly, TGF␤3/TGF-␤RII interface residues are not well conserved in either MIS or MISRII, suggesting that a third unique binding interface may exist for the MIS/MISRII complex. Cytoplasmic biochemical events leading to the activation of the type I receptor by the type II receptor have been well studied (Wrana et al., 1994); however, the structural framework through which ligand binding ori-

Table 2. Ligand and Receptor Affinity Data from BIAcore Analysis Interaction Ligand/Receptor

Immobilized Receptor kon [1/M*s]/koff [1/s]

Activin/ActRII BMP7/ActRII BMP2/ActRII BMP7/ALK2

5.9 ⫻ 105 /3.5 ⫻10⫺5 5.6 ⫻ 105 /6.9 ⫻ 10⫺4 7.2 ⫻ 105 /2.7 ⫻ 10⫺2 n.d.

KD [nM] 0.059 1.2 38

Immobilized Ligand KD [␮M] 0.045 1.7 5.4 143

0.048 (6.6 ␮M Tkv) 1.5 (6.6 ␮M Tkv) 2.3 (6.6 ␮M Tkv) 28.1 (6.8 ␮M ActRII)

The immobilized receptor data were fit to a kinetic model from which KD is calculated as koff/kon. The immobilized ligand data were best fit by equilibrium dose-response, yielding the calculated KD. The rightmost column with parentheses represents measurements in the presence of a second receptor ECD at a fixed high concentration given in parentheses. n.d. indicates that there was no significant signal up to 1 ␮M of the analyte.

Molecular Cell 614

ents the receptors, allowing for transphosphorylation is not clear. The model of the four-receptor signaling complex (Figure 3) shows that ActRII-ECD binds to the ligand further away from the membrane than does BMPRIa-ECD. Because the C-terminal linkage to the transmembrane segment is seven residues longer on the ActRII ECD than on BMPRIa-ECD, this relative orientation seems feasible. However, a splice variant of ActRIIb (Attisano et al., 1992) shortens this linkage by eight residues to only eight in length. Thus, to span the ⬎25 A˚ distance from the ECD to the base of the ligand, the ligand might be drawn into the membrane or ActRIIb pulled out. The observation that complexes constitutively signal upon the removal of either receptor ECD (Zhu and Sizeland, 1999) supports the possibility that a relative shift of the receptors in or out of the membrane plays a role in the signaling mechanism. As there are no direct contacts between the receptor ECDs in the BMP/activin model, the cooperativity that we observe with the isolated ECDs could be an allosteric effect. Conformational changes such as the movement of the B loop preceding ␣3 in BMP7 could contribute to the increased affinities. However, the magnitude of ligand-mediated cooperativity measured with the isolated ECDs seems insufficient to account for the extent of cooperative binding seen with activin and TGF-␤ (Attisano et al., 1996; Lebrun and Vale, 1997; Wrana et al., 1994). Additional interactions between receptors could clearly increase the affinity of ligand-receptor interactions. In the TGF-␤ superfamily, ligand-independent receptor homo-oligomers (Chen and Derynck, 1994; Gilboa et al., 1998, 2000) as well as ligand-independent hetero-oligomers of type I and type II activin and BMP receptors exist in the membrane (Attisano et al., 1996; Gilboa et al., 2000; Lebrun and Vale, 1997). As demonstrated by the higher affinities measured with receptors immobilized on the BIAcore surface where two spatially restrained (or dimeric) receptors bind to a single ligand (Table 2), the oligomerization state of the receptors can profoundly affect the affinity of the complex. The absence of direct contact between receptor ECDs raises the question of whether interactions between transmembrane or cytoplasmic domains could contribute to cooperative ligand binding. As with most single transmembrane receptors, the transmembrane regions of 23–26 amino acids are just long enough to span the membrane if oriented approximately perpendicular to it. Therefore, the distances between the C termini of the type I (66 A˚) and type II (83 A˚) ECD in the BMP2 and BMP7 complexes make homomeric contacts between transmembrane segments appear unlikely (Figure 3B). The type I and type II C termini in the model come within 27 A˚ of each other which may be close enough for their transmembrane segments to interact. Although the kinase domains must interact for transphosphorylation to occur, it remains to be determined what role these domains play in ligand-receptor assembly (Huse et al., 1999). We show that the receptor intracellular domains are not required for the initial formation of a stable complex. The stability of the cooperatively formed complexes lacking these domains is evidenced by the observation that the BMP7/ALK2/ActRII complex remains intact through solubilization and immunoprecipitation (Figure 5).

Protein-protein interactions are important factors to consider for the cooperative nature of receptor assembly; however, it is important to note that the limited twodimensional diffusion of receptors in the membrane also impacts the thermodynamics of ligand binding. Even if the receptors are not directly touching, they are indeed coupled through their spatial restriction in the membrane. The initial binding of ligand by a higher-affinity receptor has the effect of increasing the local concentration of correctly oriented ligand at the membrane, thus increasing the affinity of subsequent receptor binding events. Hence, the affinity of a ligand for its four-receptor complex is a function of all of the individual affinities, both high and low. Our findings that no direct interactions are made between activin/BMP type I and type II receptor ECDs and that the intracellular domains are dispensable for cooperative receptor assembly yield new insights that should lead to a better understanding of the molecular mechanism of signal transduction in the TGF-␤ superfamily. Experimental Procedures Protein Expression, Purification, and Crystallization The ECDs of Thickveins (45–144) and ALK2 (1–103) were expressed in E. coli as thioredoxin fusion proteins using a modification of published procedures (Hatta et al., 2000; Kirsch et al., 2000c). Mouse ActRII-ECD (residues 1–102) was purified from a P. pastoris expression system as described (Greenwald et al., 1998). Recombinant human activin A was generated using a stable activin-expressing cell line generously provided by Dr. J. Mather (Genentech, Inc.). Recombinant human BMP7 expressed in CHO cells (Sampath et al., 1992) was provided by Curis (Boston, MA), and recombinant human BMP2 was obtained from Genetics Institute (Cambridge, MA). Lyophilized BMP7 was dissolved in water at 4 mg/ml and then mixed 3:1 with 200 mM Tris-HCl, 2.8 M NaCl, 7.2% CHAPS (pH 7.4) yielding a 3 mg/ml solution in 50 mM Tris-HCl, 0.7 M NaCl, and 1.8% CHAPS. This solution was diluted 10-fold with a solution that had 2.2 molar equivalents of ActRII-ECD in 20 mM Tris-HCl, 0.7 M NaCl (pH 7.4). The complex was concentrated 20-fold in a Vivaspin 20 concentrator with a 5 kDa cut-off (Vivascience) and then purified on a Superdex 200 gel filtration column (Pharmacia) equilibrated in 20 mM TrisHCl, 0.7 M NaCl (pH 7.4). The eluted complex was concentrated to 8 mg/ml and crystallized from hanging drops at 23⬚C in 1 M Na acetate with 100 mM imidazole (pH 7). The crystals are in the spacegroup P6222 with a ⫽ b ⫽ 141.6 A˚ and c ⫽ 91.9 A˚. BMP7 crystals grew in 20% MPD, 100 mM Na acetate (pH 4.5), in the previously reported crystal form (Griffith et al., 1996) of P3221 with a ⫽ b ⫽ 101.3 A˚ and c ⫽ 41.8 A˚. Data Collection, Phasing, and Refinement BMP7 crystals were flash cooled in liquid nitrogen directly from the crystallization drop. The BMP7 coordinates (1 bmp) were used as a starting model for rebuilding and refinement (Griffith et al., 1996). BMP7/ActRII-ECD crystals were equilibrated in cryoprotectant for at least 12 hr before freezing. Because complex crystals could not be transferred directly into cryoprotectant without a significant loss of diffraction, the cryoprotectant (mother liquor containing 30% glycerol) was added directly to the crystallization drop and then equilibrated over a well of cryoprotectant. After this treatment crystals were transferred into fresh cryoprotectant for heavy atom soaking or flash frozen in liquid nitrogen. In this way the PtCl4 derivative was prepared by soaking a preequilibrated crystal in 1 mM PtCl4 for 16 hr. The PtCN4 derivative was prepared by the addition of cryoprotectant and 100 mM PtCN4 to a crystal in its original crystallization drop and then flash cooled in liquid nitrogen after 16 hr. All data were integrated and scaled with HKL2000 (Otwinowski, 1993). MIRAS phasing to 3.7 A˚ with MLPHARE (CCP4, 1994), and subsequent solvent flattening and histogram matching with DM (CCP4, 1994) yielded an electron density map in which the two models

Assembly of Activin/BMP Receptors 615

could be placed. There is one BMP7 and one ActRII-ECD molecule per asymmetric unit leaving a solvent content of 75%. Manual rebuilding of much of the model using O (Jones et al., 1991) was required before refinement could proceed. Refinement in Refmac5 (CCP4, 1994) with two TLS groups defined as the two independent protein chains yielded a model with good bond geometry with 78% of the non-Gly/Pro residues lying in the most favored region and none in the disallowed regions of the Ramachandran plot.

cooperativity. The association and dissociation times for the kinetic data were adjusted according to the affinity of the interaction and varied from 20 min for BMP2/ActRII equilibrium measurements to 180 min for activin/ActRII kinetic measurements. The kinetic data from at least four concentrations were globally fit to a 1:1 Langmuir binding model.

Truncated Receptor Constructs Truncated receptors for ALK2, ALK4, and ActRII were all constructed in pcDNA3 with an N-terminal epitope tag on ActRII (FLAG or myc) and ALK4 (FLAG) and five tandem C-terminal myc tags on ALK2. FLAG-ActRIItrunc and myc-ActRIItrunc both have stop codons after Val152 (numbering without signal peptide), FLAG-ALK4trunc has a stop codon after Thr183, and myc-ALK2trunc is truncated and fused to the myc epitopes after Leu126. The truncations leave 0, 10, and 57 residues in the intracellular domains of ALK2, ActRII, and ALK4, respectively.

We thank T. Hunter for critical comments on the manuscript and Curis, Inc. (S. Jones) and Genetics Institute (A. Celeste) for the supply of BMP7 and BMP2. We are grateful to E. Komives and A. Baerga-Ortiz for assistance with the BIAcore data collection and T. Roosild and the staff at SSRL and ALS for help with data collection. The ALS is supported by the Director, Office of Science, Office of Basic Energy Sciences, Materials Sciences Division, of the U.S. DOE. The SSRL Structural Molecular Biology Program is supported by the DOE, Office of Biological and Environmental Research, and by the NIH, National Center for Research Resources, Biomedical Technology Program, and the National Institute of General Medical Sciences. This work was supported by grants from the NCI and NIH.

Luciferase Assays Human HepG2 cells were seeded in 24-well plates at 175,000 cells per well and transfected 24 hr later with Gene Porter 2 (Gene Therapy Systems) according to the manufacturer’s instructions. Wells were transfected in triplicate using 1 ␮g DNA per well at a ratio of 800 ng ActRII construct:100 ng CMV-␤-galactosidase:100 ng luciferase reporter (3TP-Lux [Ca´rcamo et al., 1994] for activin-A-treated cells, BRE-luc [Hata et al., 2000] for BMP7-treated cells, each generously provided by Joan Massague´). The cells were treated with activin-A or BMP7 for ⵑ16 hr, harvested in solubilization buffer (1% Triton X-100, 25 mM glycylglycine [pH 7.8], 15 mM MgSO4, 4 mM EGTA, and 1 mM dithiothreitol), and then luciferase reporter activity was measured and normalized relative to ␤-galactosidase activities. Crosslinking 293T cells were seeded in poly-D-lysine coated 6-well plates at 400,000 cells per well and transfected 24 hr later with Perfectin (Gene Therapy Systems) according to the manufacturer’s instructions. Cells were transfected using 2 ␮g DNA per well at a type II receptor:type I receptor ratio of 2:1. Approximately 48 hr following transfection, media was aspirated from each well, and 1 ml DMEM containing 0.1% BSA and either 2 ⫻ 106 cpm per ml 125I-Activin-A or 5 ⫻ 106 cpm per ml 125I-BMP-7 were added to each well. Cells were incubated at 25⬚C for 4 hr and washed with 1 ml dissociation buffer (HDB; 12.5 mM HEPES, 140 mM NaCl, and 5 mM KCl [pH 7.4]) and treated with 1 ml ice cold HDB containing 0.5 mM disuccinimidyl suberate (DSS). Cells were incubated on ice for 30 min, aspirated, rinsed with 1 ml HDB, and then lysed in 1 ml solubilization buffer (TBS containing 1% NP-40, 0.5% deoxycholate, and 2 mM EDTA) on ice for 1 hr. Solubilized material was removed from each well and divided equally between two tubes containing antibody for IP of type II receptors or type I receptors and then incubated for 16 hr at 4⬚C. Immune complexes were precipitated by adding protein G (or for rabbit antibody, protein A) agarose beads and incubating for an additional 2 hr at 4⬚C. Beads were washed with 1 ml solubilization buffer, samples were heated in sample buffer, and proteins were resolved via SDS-PAGE and visualized via autoradiography. Surface Plasmon Resonance (BIAcore) Analysis The experiments were carried out on a BIAcore 3000 system, and the data were processed with the BIA evaluation software version 2.2.4 (BIAcore). Proteins were immobilized by primary amine coupling on a CM5 sensor chip. Sensor chip 1 had BMP2, BMP7, and activin A immobilized in flow cells 2, 3, and 4. Sensor chip 2 had ALK2, ActRII, and Tkv immobilized in flow cells 2, 3, and 4. Flow cells 1 were treated with the immobilization reagents as a background to be subtracted from the other flow cells. Immobilizations were done in 10 mM Na acetate (pH 5.1) with 40–80 ␮g/ml of ligand for 2 min on chip 1 and 250–400 ␮g/ml of receptor ECD for 14 min on chip 2. All measurements were done with a flow rate of 5 ␮l/min in 20 mM Tris-HCl, 500 mM NaCl, 0.005% surfactant P20 (pH 7.4), supplemented with 0.36% CHAPS detergent for chip 2. The equilibrium binding experiments were fit to a 1:1 association model without

Acknowledgments

Received: October 30, 2002 Revised: January 30, 2003 References Attisano, L., Wrana, J.L., Cheifetz, S., and Massague´, J. (1992). Novel activin receptors: distinct genes and alternative mRNA splicing generate a repertoire of serine/threonine kinase receptors. Cell 68, 97–108. Attisano, L., Wrana, J.L., Montalvo, E., and Massague´, J. (1996). Activation of signalling by the activin receptor complex. Mol. Cell. Biol. 16, 1066–1073. Barton, G.J. (1993). ALSCRIPT: a tool to format multiple sequence alignments. Protein Eng. 6, 37–40. Brummel, T., Abdollah, S., Haerry, T.E., Shimell, M.J., Merriam, J., Raftery, L., Wrana, J.L., and O’Connor, M.B. (1999). The Drosophila activin receptor baboon signals through dSmad2 and controls cell proliferation but not patterning during larval development. Genes Dev. 13, 98–111. Ca´rcamo, J., Weis, F.M., Ventura, F., Wieser, R., Wrana, J.L., Attisano, L., and Massague´, J. (1994). Type I receptors specify growthinhibitory and transcriptional responses to transforming growth factor beta and activin. Mol. Cell. Biol. 14, 3810–3821. CCP4 (Collaborative Computational Project 4) (1994). The CCP suite: programs for protein crystallography. Acta Crystallogr. D 50, 760–776. Chen, R.H., and Derynck, R. (1994). Homomeric interactions between type II transforming growth factor-␤ receptors. J. Biol. Chem. 269, 22868–22874. Childs, S.R., Wrana, J.L., Arora, K., Attisano, L., O’Connor, M.B., and Massague´, J. (1993). Identification of a Drosophila activin receptor. Proc. Natl. Acad. Sci. USA 90, 9475–9479. Clackson, T., and Wells, J.A. (1995). A hot spot of binding energy in a hormone-receptor interface. Science 267, 383–386. Clarke, T.R., Hoshiya, Y., Yi, S.E., Liu, X., Lyons, K.M., and Donahoe, P.K. (2001). Mullerian inhibiting substance signaling uses a bone morphogenetic protein (BMP)-like pathway mediated by ALK2 and induces SMAD6 expression. Mol. Endocrinol. 15, 946–959. De Crescenzo, G., Grothe, S., Zwaagstra, J., Tsang, M., and O’Connor-McCourt, M.D. (2001). Real-time monitoring of the interactions of transforming growth factor-␤ (TGF-␤) isoforms with latency-associated protein and the ectodomains of the TGF-␤ type II and III receptors reveals different kinetic models and stoichiometries of binding. J. Biol. Chem. 276, 29632–29643. de Vos, A.M., Ultsch, M., and Kossiakoff, A.A. (1992). Human growth hormone and extracellular domain of its receptor: crystal structure of the complex. Science 255, 306–312. Dudley, A.T., Lyons, K.M., and Robertson, E.J. (1995). A requirement

Molecular Cell 616

for bone morphogenetic protein-7 during development of the mammalian kidney and eye. Genes Dev. 9, 2795–2807.

al. (1994). Characterization and cloning of a receptor for BMP-2 and BMP-4 from NIH 3T3 cells. Mol. Cell. Biol. 14, 5961–5974.

Eldar, A., Dorfman, R., Weiss, D., Ashe, H., Shilo, B.Z., and Barkai, N. (2002). Robustness of the BMP morphogen gradient in Drosophila embryonic patterning. Nature 419, 304–308.

Kraulis, P.J. (1991). MOLSCRIPT: a program to produce both detailed and schematic plots of protein structures. J. Appl. Crystallogr. 24, 946–950.

Garrett, T.P., McKern, N.M., Lou, M., Elleman, T.C., Adams, T.E., Lovrecz, G.O., Zhu, H.J., Walker, F., Frenkel, M.J., Hoyne, P.A., et al. (2002). Crystal structure of a truncated epidermal growth factor receptor extracellular domain bound to transforming growth factor ␣. Cell 110, 763–773.

Kutty, G., Kutty, R.K., Samuel, W., Duncan, T., Jaworski, C., and Wiggert, B. (1998). Identification of a new member of transforming growth factor-beta superfamily in Drosophila: the first invertebrate activin gene. Biochem. Biophys. Res. Commun. 246, 644–649.

Gilboa, L., Wells, R.G., Lodish, H.F., and Henis, Y.I. (1998). Oligomeric structure of type I and type II transforming growth factor ␤ receptors: homodimers form in the ER and persist at the plasma membrane. J. Cell Biol. 140, 767–777.

Lebrun, J.J., and Vale, W.W. (1997). Activin and inhibin have antagonistic effects on ligand-dependent heteromerization of the type I and type II activin receptors and human erythroid differentiation. Mol. Cell. Biol. 17, 1682–1691.

Gilboa, L., Nohe, A., Geissendorfer, T., Sebald, W., Henis, Y.I., and Knaus, P. (2000). Bone morphogenetic protein receptor complexes on the surface of live cells: a new oligomerization mode for serine/ threonine kinase receptors. Mol. Biol. Cell 11, 1023–1035.

Letsou, A., Arora, K., Wrana, J.L., Simin, K., Twombly, V., Jamal, J., Staehling-Hampton, K., Hoffmann, F.M., Gelbart, W.M., Massague´, J., et al. (1995). Drosophila Dpp signaling is mediated by the punt gene product: a dual ligand-binding type II receptor of the TGF ␤ receptor family. Cell 80, 899–908.

Gray, P.C., Greenwald, J., Blount, A.L., Kunitake, K.S., Donaldson, C.J., Choe, S., and Vale, W. (2000). Identification of a binding site on the type II activin receptor for activin and inhibin. J. Biol. Chem. 275, 3206–3212.

Lim, D.A., Tramontin, A.D., Trevejo, J.M., Herrera, D.G., Garcia-Verdugo, J.M., and Alvarez-Buylla, A. (2000). Noggin antagonizes BMP signaling to create a niche for adult neurogenesis. Neuron 28, 713–726.

Greenwald, J., Le, V., Corrigan, A., Fischer, W., Komives, E., Vale, W., and Choe, S. (1998). Characterization of the extracellular ligandbinding domain of the type II activin receptor. Biochemistry 37, 16711–16718. Greenwald, J., Fischer, W.H., Vale, W.W., and Choe, S. (1999). Threefinger toxin fold for the extracellular ligand-binding domain of the type II activin receptor serine kinase. Nat. Struct. Biol. 6, 18–22. Griffith, D.L., Keck, P.C., Sampath, T.K., Rueger, D.C., and Carlson, W.D. (1996). Three-dimensional structure of recombinant human osteogenic protein 1: structural paradigm for the transforming growth factor ␤ superfamily. Proc. Natl. Acad. Sci. USA 93, 878–883. Hart, P.J., Deep, S., Taylor, A.B., Shu, Z., Hinck, C.S., and Hinck, A.P. (2002). Crystal structure of the human T␤R2 ectodomain-TGF␤3 complex. Nat. Struct. Biol. 9, 203–208. Hata, A., Seoane, J., Lagna, G., Montalvo, E., Hemmati-Brivanlou, A., and Massague´, J. (2000). OAZ uses distinct DNA- and proteinbinding zinc fingers in separate BMP-Smad and Olf signaling pathways. Cell 100, 229–240. Hatta, T., Konishi, H., Katoh, E., Natsume, T., Ueno, N., Kobayashi, Y., and Yamazaki, T. (2000). Identification of the ligand-binding site of the BMP type IA receptor for BMP-4. Biopolymers 55, 399–406. Hogan, B.L. (1996). Bone morphogenetic proteins: multifunctional regulators of vertebrate development. Genes Dev. 10, 1580–1594.

Luo, G., Hofmann, C., Bronckers, A.L., Sohocki, M., Bradley, A., and Karsenty, G. (1995). BMP-7 is an inducer of nephrogenesis, and is also required for eye development and skeletal patterning. Genes Dev. 9, 2808–2820. Macias-Silva, M., Hoodless, P.A., Tang, S.J., Buchwald, M., and Wrana, J.L. (1998). Specific activation of Smad1 signaling pathways by the BMP7 type I receptor, ALK2. J. Biol. Chem. 273, 25628–25636. Massague´, J. (1998). TGF-beta signal transduction. Annu. Rev. Biochem. 67, 753–791. Massague´, J., and Wotton, D. (2000). Transcriptional control by the TGF-beta/Smad signaling system. EMBO J. 19, 1745–1754. McDonald, N.Q., and Hendrickson, W.A. (1993). A structural superfamily of growth factors containing a cystine knot motif. Cell 73, 421–424. Ogiso, H., Ishitani, R., Nureki, O., Fukai, S., Yamanaka, M., Kim, J.H., Saito, K., Sakamoto, A., Inoue, M., Shirouzu, M., and Yokoyama, S. (2002). Crystal structure of the complex of human epidermal growth factor and receptor extracellular domains. Cell 110, 775–787. Otwinowski, Z. (1993). Data collection and processing. In Proceedings of the CCP4 Study Weekend, L. Sawyer, N. Isaacs, and S. Burley, eds. (Warrington, UK: Daresbury Laboratory), pp. 56–62.

Hunter, T. (2000). Signaling—2000 and beyond. Cell 100, 113–127.

Pecina, M., Giltaij, L.R., and Vukicevic, S. (2001). Orthopaedic applications of osteogenic protein-1 (BMP-7). Int. Orthop. 25, 203–208.

Huse, M., Chen, Y.G., Massague´, J., and Kuriyan, J. (1999). Crystal structure of the cytoplasmic domain of the type I TGF ␤ receptor in complex with FKBP12. Cell 96, 425–436.

Pellegrini, L., Burke, D.F., von Delft, F., Mulloy, B., and Blundell, T.L. (2000). Crystal structure of fibroblast growth factor receptor ectodomain bound to ligand and heparin. Nature 407, 1029–1034.

Jones, T.A., Zou, J.Y., Cowan, S.W., and Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Crystallogr. A 47, 100–119.

Penton, A., Chen, Y., Staehling-Hampton, K., Wrana, J.L., Attisano, L., Szidonya, J., Cassill, J.A., Massague´, J., and Hoffmann, F.M. (1994). Identification of two bone morphogenetic protein type I receptors in Drosophila and evidence that Brk25D is a decapentaplegic receptor. Cell 78, 239–250.

Kingsley, D.M. (1994). The TGF-␤ superfamily: new members, new receptors, and new genetic tests of function in different organisms. Genes Dev. 8, 133–146. Kirsch, T., Sebald, W., and Dreyer, M.K. (2000a). Crystal structure of the BMP-2-BRIA ectodomain complex. Nat. Struct. Biol. 7, 492–496.

Philippsen, A. (2001). DINO: Visualizing Structural Biology (http:// www.dino3d.org).

Kirsch, T., Nickel, J., and Sebald, W. (2000b). BMP-2 antagonists emerge from alterations in the low-affinity binding epitope for receptor BMPR-II. EMBO J. 19, 3314–3324.

Qian, S.W., Burmester, J.K., Tsang, M.L., Weatherbee, J.A., Hinck, A.P., Ohlsen, D.J., Sporn, M.B., and Roberts, A.B. (1996). Binding affinity of transforming growth factor-␤ for its type II receptor is determined by the C-terminal region of the molecule. J. Biol. Chem. 271, 30656–30662.

Kirsch, T., Nickel, J., and Sebald, W. (2000c). Isolation of recombinant BMP receptor IA ectodomain and its 2:1 complex with BMP-2. FEBS Lett. 468, 215–219.

Russell, R.B., and Barton, G.J. (1992). Multiple protein sequence alignment from tertiary structure comparison: assignment of global and residue confidence levels. Proteins 14, 309–323.

Knaus, P., and Sebald, W. (2001). Cooperativity of binding epitopes and receptor chains in the BMP/TGF␤ superfamily. Biol. Chem. 382, 1189–1195.

Sampath, T.K., Maliakal, J.C., Hauschka, P.V., Jones, W.K., Sasak, H., Tucker, R.F., White, K.H., Coughlin, J.E., Tucker, M.M., Pang, R.H., et al. (1992). Recombinant human osteogenic protein-1 (hOP-1) induces new bone formation in vivo with a specific activity comparable with natural bovine osteogenic protein and stimulates osteoblast

Koenig, B.B., Cook, J.S., Wolsing, D.H., Ting, J., Tiesman, J.P., Correa, P.E., Olson, C.A., Pecquet, A.L., Ventura, F., Grant, R.A., et

Assembly of Activin/BMP Receptors 617

proliferation and differentiation in vitro. J. Biol. Chem. 267, 20352– 20362. Syed, R.S., Reid, S.W., Li, C., Cheetham, J.C., Aoki, K.H., Liu, B., Zhan, H., Osslund, T.D., Chirino, A.J., Zhang, J., et al. (1998). Efficiency of signalling through cytokine receptors depends critically on receptor orientation. Nature 395, 511–516. Thompson, J.D., Higgins, D.G., and Gibson, T.J. (1994). CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids Res. 22, 4673–4680. Tsuchida, K., Vaughan, J.M., Wiater, E., Gaddy-Kurten, D., and Vale, W.W. (1995). Inactivation of activin-dependent transcription by kinase-deficient activin receptors. Endocrinology 136, 5493–5503. Ullrich, A., and Schlessinger, J. (1990). Signal transduction by receptors with tyrosine kinase activity. Cell 61, 203–212. Visser, J.A., Olaso, R., Verhoef-Post, M., Kramer, P., Themmen, A.P., and Ingraham, H.A. (2001). The serine/threonine transmembrane receptor ALK2 mediates Mullerian inhibiting substance signaling. Mol. Endocrinol. 15, 936–945. Weis-Garcia, F., and Massague´, J. (1996). Complementation between kinase-defective and activation-defective TGF-beta receptors reveals a novel form of receptor cooperativity essential for signaling. EMBO J. 15, 276–289. Wrana, J.L., Attisano, L., Wieser, R., Ventura, F., and Massague´, J. (1994). Mechanism of activation of the TGF-␤ receptor. Nature 370, 341–347. Wuytens, G., Verschueren, K., de Winter, J.P., Gajendran, N., Beek, L., Devos, K., Bosman, F., de Waele, P., Andries, M., van den Eijndenvan Raaij, A.J., et al. (1999). Identification of two amino acids in activin A that are important for biological activity and binding to the activin type II receptors. J. Biol. Chem. 274, 9821–9827. Yamashita, H., ten Dijke, P., Huylebroeck, D., Sampath, T.K., Andries, M., Smith, J.C., Heldin, C.H., and Miyazono, K. (1995). Osteogenic protein-1 binds to activin type II receptors and induces certain activin-like effects. J. Cell Biol. 130, 217–226. Zhu, H.J., and Sizeland, A.M. (1999). Extracellular domain of the transforming growth factor-beta receptor negatively regulates ligand-independent receptor activation. J. Biol. Chem. 274, 29220– 29227. Zimmerman, L.B., De Jesus-Escobar, J.M., and Harland, R.M. (1996). The Spemann organizer signal noggin binds and inactivates bone morphogenetic protein 4. Cell 86, 599–606. Accession Numbers The coordinates for free BMP7 (1LXI) and the BMP7/ActRII-ECD (1LX5) complex reported here have been deposited in the Protein Data Bank.