Identification of a Functional Binding Site for Activin on the Type I Receptor ALK4*

Activins, like other members of the transforming growth factor- (cid:1) (TGF- (cid:1) ) superfamily, initiate signaling by assembling a complex of two types of transmembrane serine/threonine receptor kinases classified as type II (ActRII or ActRIIB) and type I (ALK4). A kinase-deleted version of ALK4 can form an inactive complex with activin and ActRII/IIB and thereby acts in a dominant negative manner to block activin signaling. Using the complex structure of bone morphogenetic protein-2 bound to its type I receptor (ALK3) as a guide, we intro-duced extracellular domain mutations in the context of the truncated ALK4 (ALK4-trunc) construct and as-sessed the ability of the mutants to inhibit activin function. We have identified five hydrophobic amino acid residues on the ALK4 extracellular domain (Leu 40 , Ile 70 , Val 73 , Leu 75 , and Pro 77 ) that, when mutated to alanine, have substantial effects on ALK4-trunc dominant negative activity. In addition, eleven mutants partially affected activin binding to ALK4. Together, these residues likely constitute the binding surface for activin on ALK4. Cross-linking studies measuring binding of 125 I-activin-A to the ALK4-trunc mutants in the presence of ActRII implicated the same residues. Our results indicate that there is only a partial overlap of the binding sites on ALK4 and ALK3 for activin-A

Activins, like other members of the transforming growth factor-␤ (TGF-␤) superfamily, initiate signaling by assembling a complex of two types of transmembrane serine/threonine receptor kinases classified as type II (ActRII or ActRIIB) and type I (ALK4). A kinase-deleted version of ALK4 can form an inactive complex with activin and ActRII/IIB and thereby acts in a dominant negative manner to block activin signaling. Using the complex structure of bone morphogenetic protein-2 bound to its type I receptor (ALK3) as a guide, we introduced extracellular domain mutations in the context of the truncated ALK4 (ALK4-trunc) construct and assessed the ability of the mutants to inhibit activin function. We have identified five hydrophobic amino acid residues on the ALK4 extracellular domain (Leu 40 , Ile 70 , Val 73 , Leu 75 , and Pro 77 ) that, when mutated to alanine, have substantial effects on ALK4-trunc dominant negative activity. In addition, eleven mutants partially affected activin binding to ALK4. Together, these residues likely constitute the binding surface for activin on ALK4. Cross-linking studies measuring binding of 125 Iactivin-A to the ALK4-trunc mutants in the presence of ActRII implicated the same residues. Our results indicate that there is only a partial overlap of the binding sites on ALK4 and ALK3 for activin-A and bone morphogenetic protein-2, respectively. In addition three of the residues required for activin binding to ALK4 are conserved on the type I TGF-␤ receptor ALK5, suggesting the corresponding region on ALK5 may be important for TGF-␤ binding.
The signaling events initiated by activin require binding of two types of transmembrane serine/threonine receptor kinases classified as type II (ActRII or ActRIIB) and type I (ALK4). Both receptors are transmembrane proteins with ligand binding activity in the extracellular domain and serine/threonine kinase activity in the intracellular domain (15). The activin type II receptors are the primary ligand-binding proteins and can bind ligand with high affinity in the absence of type I receptors (18). The type I receptor, however, is unable to bind ligand in the absence of the type II receptors (19,20). In the receptor complex, the constitutively active type II receptor kinase phosphorylates ALK4 in the regulatory GS domain, a glycine-and serine-rich segment on the membrane-proximal side of the kinase domain, and this phosphorylation leads to activation of ALK4 (21). Once activated, ALK4 binds and then phosphorylates cytoplasmic Smad proteins, which form part of the post-receptor signal transduction system (22).
Recently obtained crystal structure data have greatly advanced our understanding of how members of the TGF-␤ superfamily interact with their receptors. The complex structure of BMP-7 dimer bound to the ActRII-ECD shows that the ActRII-ECD makes contact with only one of the dimer subunits (23). The binding interface revealed by the x-ray structure agrees with the binding affinities available for mutants of ActRII (24), activin-A (25,26), and BMP-2 (27). The structure of BMP-2 in complex with BMP receptor IA (ALK3, BR1A) was also solved recently (28), revealing a type I receptor ECD fold similar to that of the ActRII-ECD. In the structure of the complex, ALK3 binds to the finger-helix groove of the BMP-2 dimer in such a way that each ALK3-ECD molecule is in contact with both BMP-2 monomers. The ligand binding interface on ALK3 for BMP-2 is characterized by a groove on the concave surface of the ECD and by residues in the short ␣ helix.
Based on the crystal structure of the BMP2⅐ALK3-ECD complex and homology modeling, we have subjected the ALK4-ECD to alanine scanning mutagenesis with the goal of identifying the amino acid residues required for activin binding. We predicted that by individually mutating ALK4-ECD residues to alanine we would be able to identify amino acids that are important for activin binding while minimizing the structural changes caused by mutation. Using this approach, we have identified five hydrophobic amino acid residues (Leu 40 , Ile 70 , Val 73 , Leu 75 , and Pro 77 ) that, when individually mutated to alanine, had substantial effects on activin binding to ALK4. In addition, eleven other mutants moderately or weakly disrupted activin binding. Homology modeling suggests that these residues interact with each other to form a contiguous surface on the concave face of ALK4 that is a likely activin binding interface.

EXPERIMENTAL PROCEDURES
Materials-NuPAGE gels and molecular weight markers were obtained from Invitrogen. Recombinant human activin-A was generated using a stable activin-expressing cell line generously provided by Dr. J. Mather (Genentech, Inc., South San Francisco, CA) and were purified by Wolfgang Fischer (Peptide Biology Laboratory, Salk Institute, La Jolla, CA). 125 I-Activin-A was prepared using the chloramine T method as described previously (29). Anti-FLAG (M2) antibody was from Sigma. Horseradish peroxidase-linked anti-mouse IgG, 3,3Ј,5,5Ј-tetramethylbenzidine substrate, chemiluminescent substrate (Supersignal TM ), and the BCA protein assay kit were obtained from Pierce. The hALK4 (20) constructs used in this study were in the pcDNA3 expression vector (Invitrogen).
Mutagenesis of ALK4 -A kinase-deleted ALK4 construct (ALK4trunc) that encodes the first 206 amino acids of ALK4 was generated using standard PCR techniques. To incorporate the amino-terminal FLAG tag (following glycine 28) and to generate mutations in the ECD of ALK4-trunc or full-length ALK4, we utilized an overlapping PCR strategy (24). Primers were constructed to incorporate a 5Ј-HindIII site and a 3Ј-EcoRI site, and PCR products were gel-purified and digested with both enzymes and then subcloned into HindIII/EcoRI-digested pcDNA3 vector to yield mutant receptor constructs. For each construct, the mutated amino-terminal ECD region was confirmed by DNA sequencing.
Transfection and Detection of Cell Surface Expression of ALK4 in Intact HEK293T Cells-HEK293T cells were grown in 5% CO 2 to 40 -60% confluence on poly-D-lysine-coated 6-well plates in complete Dulbecco's modified Eagle's medium (with 10% bovine calf serum, penicillin, streptomycin, and L-glutamine). Cells were transfected with wildtype ALK4, ALK4-trunc, or ALK4-trunc ECD mutants using Perfectin (Gene Therapy Systems). For co-transfection of ActRII and ALK4, a 2:1 ratio of their respective cDNAs was used. For Western blotting, cells were solubilized in 200 l of 1% Triton X-100, and protein concentrations were determined using the BCA method according to the manufacturer's instructions. SDS-PAGE and electrotransfer to nitrocellulose were carried out using NuPAGE gels and a NOVEX X-cell II apparatus as described previously (24). To detect FLAG-tagged ALK4 constructs expressed at the cell surface, intact cells were fixed in paraformaldehyde, incubated with anti-FLAG antibody, washed, and treated with peroxidase-conjugated anti-mouse Ig. Specific antibody staining was measured using 3,3Ј,5,5Ј-tetramethyl-benzidine peroxidase substrate as described previously (24).
Covalent Cross-linking-Covalent cross-linking was performed by incubating transfected HEK293T cells (4 ϫ 10 6 ) with 125 I-activin-A (10 6 cpm/well) for 2 h at room temperature in binding buffer (24) with gentle rocking. Cells were washed, resuspended in 1 ml of Hepes buffer (HDB; 12.5 mM Hepes, pH 7.4, 140 mM NaCl, and 5 mM KCl), brought to 0.5 mM disuccinimidyl suberate, and incubated for 30 min on ice. Cross-linking reactions were quenched with TBS (50 mM Tris-HCl, pH 7.5, 150 mM NaCl), cells were solubilized in 1% Triton X-100 lysis buffer for 30 min on ice, and the resultant supernatant was subjected to immunoprecipitation using anti-FLAG antibody. Immune complexes were analyzed by SDS-PAGE and autoradiography.
Computer Modeling of ALK4-ECD-Computer modeling and molecular mechanics employed Discover/Insight II (Accelrys, San Diego, CA) using the cff91 force field hosted on a Silicon Graphics Octane work station running IRIX64, version 6.5. Construction of an interactive computer graphical model of ALK4 employed homology modeling based on the x-ray crystallographic structure of ALK3 reported by Kirsch et al. (28) in their structure of the BMP-2⅐ALK3 complex (Protein Data Bank entry 1ES7). Based on the primary sequence alignment of ALK3 and ALK4, 55 residues were identified for single-residue substitution; eight deletions of one residue or more and a single insertion of five residues were similarly identified. Single residue substitutions were performed with retention of side chain orientation where possible; in the event of severe overlap, manual rotation was conducted to minimize steric hindrance. Local perturbations caused by deletion were allowed to anneal by restrained minimization wherein the backbone atoms of the nascent ALK4 model were tethered in place with the exception that the residues on either side of a deletion were allowed to move. A simple ␤ turn structure was imposed on the five-residue insertion spanning ALK4 residues Lys 72 -Val 76 . This insertion attempted to maximize the solventaccessible area of Lys 72 and Glu 74 and to exploit a naturally arising salt bridge involving Asp 56 and Lys 72 . Final overlap of the experimentally observed ALK3 and theoretically derived ALK4 structures was 1.25 Å root mean square over the common backbone.

Selection of Amino Acid Residues on ALK4 for Mutagenesis-
A truncated version of ALK4 in which the cytoplasmic kinase domain is deleted forms an inactive complex with activin and ActRII and therefore acts in a dominant negative manner to block activin-A signaling (32). We showed that ALK4-trunc blocks activin-A signaling when transfected into HEK293T cells (see Fig. 4), and we proceeded to introduce mutations within the ECD of the ALK4-trunc construct with the goal of identifying residues that block activin binding and therefore inhibit its dominant negative function. Kirsch et al. (28) identified a groove on ALK3 that constitutes the BMP-2 binding interface (Fig. 1), and we have individually mutated each of the residues on ALK4 corresponding to these groove- forming residues. In addition, we have mutated ALK4-ECD residues in the broader region implicated by the BMP-2⅐ALK3 complex structure (Fig. 1). We predicted that by individually mutating ALK4-trunc residues to alanine we would be able to identify amino acids that are important for activin binding while minimizing the structural changes caused by mutation.
Expression of ALK4-trunc in HEK293T Cells-To facilitate detection of ALK4-trunc expressed in HEK293T cells, we used PCR to introduce a FLAG epitope tag (DYKDDDDK) at the extreme amino terminus immediately following the putative signal peptide (after Gly 28 ). To demonstrate that the FLAGtagged ALK4-trunc constructs were expressed in 293T cells, we initially performed Western blot experiments. Fig. 2A shows that wild-type and selected mutant ALK4-trunc proteins are expressed at comparable levels in 293T cells. An antibody directed against the FLAG epitope tag recognizes an ϳ26-kDa protein (expected size 22.6 kDa) expressed in cells transfected with ALK4-trunc constructs but not vector alone ( Fig. 2A). To demonstrate that the FLAG-tagged ALK4-trunc constructs were expressed at the cell surface we used an intact cell enzyme-linked immunosorbent assay (24) (Fig. 2B). Together, these results demonstrate that the FLAG epitope is present on the ALK4-trunc constructs, that the wild-type and mutant receptors are expressed in this system at similar levels, and that the receptors are expressed at the cell surface.
Covalent Cross-linking of Activin to ActRII and Wild-type or Mutant ALK4-trunc Constructs-To identify ALK4-trunc mutants with altered activin binding, we co-expressed the mutants in HEK293T cells together with ActRII, and performed affinity labeling experiments. Immunoprecipitation of crosslinked complexes with anti-FLAG antibodies directed against ALK4-FLAG led to the identification of a number of ALK4 residues important for activin-A binding (Fig. 3). In cells transfected with ActRII-Myc and wild-type ALK4-trunc, the predominant species of ϳ40 kDa represented ALK4-trunc cross-linked to a single activin subunit (the ϳ55-kDa species presumably represented ALK4-trunc cross-linked to the activin dimer). The ActRII⅐activin cross-linked complex (ϳ80 kDa) was ineffectively immunoprecipitated using anti-FLAG antibodies consistent with previous studies (23) that have shown the activin⅐ActRII⅐ALK4-trunc complex to be unstable following solubilization. Of the 26 ALK4-trunc mutants tested, five had significant effects on activin binding whereas several others had intermediate effects. Mutants L40A and V73A formed cross-linked complexes with 125 I-activin-A more efficiently than wild-type ALK4-trunc (Fig. 3). In contrast, mutants I70A, L75A, and P77A cross-linked poorly to 125 I-activin-A (Fig. 3). Mutants P71A, K72A, V76A, G79A, F82A, R91A, and T93A FIG. 2. ALK4 mutants transfected into HEK293T cells are expressed at the cell surface. A, solubilized extracts prepared from cells transfected with empty vector, ALK4-trunc, or ALK4-trunc with selected ECD mutations were normalized with respect to protein levels and then subjected to Western blot analysis using an anti-FLAG antibody as described under "Experimental Procedures." B, bar graph illustrating the cell surface expression of wild-type (black bar) and mutant (gray bars) FLAG-tagged ALK4-trunc constructs as detected by anti-FLAG antibody in intact HEK293T cells as described under "Experimental Procedures." Expression was determined relative to vectortransfected cells (white bar) .   FIG. 3. Covalent cross-linking of 125 I-activin-A to ActRII and wild-type or mutant ALK4-trunc constructs. HEK293T cells were transfected with empty vector (pcDNA3), ActRII-myc, ActRII-myc ϩ ALK4-FLAG, ActRII-myc ϩ ALK4-trunc, or ActRII-myc ϩ ALK4-trunc mutants and cross-linked to 125 I-activin-A as described under "Experimental Procedures." Cross-linked complexes isolated by immunoprecipitation using an anti-FLAG antibody were resolved by SDS-PAGE and visualized by autoradiography.
showed some decrease in binding to 125 I-activin-A compared with wild-type ALK4-trunc (Fig. 3). These results were supported by immunoprecipitation using antibodies directed against ActRII-Myc (data not shown). The Myc antibody effectively immunoprecipitated activin⅐ActRII complexes but was less efficient in isolating activin bound to ALK4-trunc. Nevertheless, the same pattern of 125 I-activin-A cross-linking to the ALK4-trunc mutants was evident.
Mutation of Amino Acids Required for Activin Binding Affects the Dominant Negative Activity of ALK4-trunc in HEK293T Cells-We tested the relative abilities of wild-type and mutant ALK4-trunc constructs to block the activin-A induction of the TGF-␤/activin responsive luciferase reporter construct A3-Lux in HEK293T cells. As shown in Fig. 4, activin-A induced A3-Lux reporter activity when this plasmid was cotransfected with FAST2 (31) into HEK293T cells. Activin-induced A3-Lux activation was decreased ϳ65% when wild-type ALK4-trunc was co-transfected (Fig. 4). However, consistent with the cross-linking data, receptors with alanine substitutions at any one of the three residues identified to be essential for activin binding (Ile 70 , Leu 75 , or Pro 77 ) were not able to exhibit dominant negative activity (Fig. 4). In addition, the ALK4-trunc mutants with increased activin-A binding (L40A and V73A; see Fig. 3) were more effective dominant negative inhibitors than wild-type ALK4-trunc (Fig. 4), despite equal expression at the cell surface (Fig. 2B). Mutant L40A decreased activin-induced A3-Lux activation by ϳ80% relative to vectortransfected cells whereas mutant V73A decreased activation by ϳ90% relative to vector. The behavior of most of the other ALK4-trunc mutants in the functional assay (Fig. 4) corresponded to their ability to cross-link activin (Fig. 3). Mutants P71A, V76A, G79A, F82A, R91A, and T93A, which showed decreased binding to 125 I-activin-A compared with wild-type ALK4-trunc (Fig. 3), also displayed impaired dominant negative activity (suppressing the activin-induced luciferase response by 35-50%; see Fig. 4). Mutants S55A, A78G, and S87A, although displaying activin binding comparable with wild-type ALK4-trunc (Fig. 3), were less effective dominant negative inhibitors (Fig. 4). All other mutants suppressed activin-induced luciferase activity to the same extent as wild-type ALK4trunc. Together, these results provide further evidence in intact cells demonstrating the importance of the Leu 40 -Ile 70 -Val 73 -Leu 75 -Pro 77 hydrophobic residues in mediating activin/ ALK4 binding and activin signaling.

FIG. 4. Mutation of ALK4-ECD residues required for activin-A binding blocks the dominant negative activity of ALK4-trunc.
HEK293T cells were transfected in triplicate with A3-Lux, cytomegalovirus-␤-galactosidase, FAST2, and empty vector (pcDNA3), wild-type ALK4-trunc, or the indicated ALK4-trunc mutants. Cells were incubated in the presence or absence of 1 nM activin-A and assayed for luciferase activity as described under "Experimental Procedures." -Fold induction of activin-A-treated cells relative to untreated cells is shown with error bars indicating S.D. This experiment was repeated four times with similar results, and a representative experiment is shown.

FIG. 5. Activin-A-and activin-B-induced luciferase activity requires the same ALK4-ECD residues.
HEK293T cells were transfected in triplicate with A3-Lux, cytomegalovirus-␤-galactosidase, FAST2, and empty vector (white bar), wild-type ALK4-trunc (black bar), or the indicated ALK4-trunc mutants (gray bars). Cells were incubated in the presence or absence of 1 nM activin-A or 1 nM activin-B and assayed for luciferase activity as described under "Experimental Procedures." -Fold induction of activin-A-and activin-B-treated cells relative to untreated cells is shown with error bars indicating S.D. This experiment was repeated three times with similar results, and a representative experiment is shown.

ALK4 Binding Determinants Are Conserved for Activin-A and
Activin-B-We tested whether wild-type and selected mutant ALK4-trunc constructs could also block activin-B induced luciferase activity. Fig. 5B shows that activin-B-induced A3-Lux activity was suppressed ϳ55% upon overexpression of wild-type ALK4-trunc (compared with ϳ75% suppression of activin-A induced A3-Lux activity; see Fig. 5A). ALK4-trunc constructs with alanine substitutions at Ile 70 , Leu 75 , or Pro 77 were poor dominant negative inhibitors of both activin-A-and activin-B-induced luciferase activity (Fig. 5, A and B). In addition, the V73A mutant was a more effective dominant negative inhibitor than wild-type ALK4 for both activin-A and activin-B. Mutations at residues Ser 38 and Phe 82 had little effect on ALK4 dominant negative activity for either activin isoform (Fig. 5). Overall, these results indicate that activin-A and activin-B interact with the same residues on ALK4.

Mutation of Amino Acids Required for Activin Binding Affect the Ability of ALK4 to Mediate Activin Signaling in Mink Lung
Epithelial Cells-Mv1Lu mink lung epithelial cells have low levels of activin type I and type II receptors and very limited responsiveness to activin (19,20). However, these cells show strong transcriptional and antiproliferative responses to activin when transfected with appropriate activin receptors (19,20) and therefore provide an ideal system to compare the effects of wild-type and mutant ALK4. Mv1Lu cells co-transfected with the TGF-␤/activin responsive luciferase reporter construct 3TP-Lux and empty vector responded poorly to treatment with 1 nM activin-A, showing a 3-fold induction in luciferase activity (Fig. 6). In contrast, cells transfected with wildtype ALK4 or the M53A mutant (which had no effect on ALK4trunc function in the dominant negative assay) mediated a 12-fold induction in luciferase activity. Alanine mutations at residues Ile 70 and Leu 75 were shown previously to be essential for activin binding and resulted in full-length ALK4 mutants that were unable to mediate activin signaling (Fig. 6).
Structure-Function Relationship Analysis of ALK4 -The three-dimensional atomic structure of ALK4 has yet to be determined. Therefore, we constructed a homology model of the ALK4-ECD (Fig. 7B) based on the ALK3-ECD structure (Fig.  7A) (28). Despite relatively low sequence homology (ϳ30% in the structurally defined ECD regions) of these type I receptors, the ten cysteine residues can be aligned and are thought to confer the same general fold (three-finger toxin fold) in ALK4 as observed in the ALK3 structure. Our modeling was also instructed by the ALK5-ECD model Guimond et al. (33) produced based on the ALK3-ECD structure. ALK4 and ALK5 have higher sequence homology (ϳ38% in the structurally defined ECD regions) than that shared between ALK4 and ALK3 (Fig. 1). In addition, there are fewer non-homologous regions (i.e. gaps in sequence alignments) observed between ALK4 and ALK5 compared with ALK4 and ALK3. Overall, the modeled ALK4-ECD structure preserves the open left hand topology of the ALK3-ECD structure, with a concave face and a convex face stemming from the curvature of the central ␤-sheet (Fig. 7, A  and B). However, the pre-helix extension in the ALK4-ECD model significantly alters the conformation of this region with respect to the ALK3-ECD structure (Fig. 7C). In addition, predicted groove-forming residues of ALK4, including Met 53 , Ser 55 , and Thr 93 , are only partially exposed on the concave surface of the receptor suggesting a possible distortion and masking of the groove in the ALK4-ECD compared with the ALK3-ECD (Fig. 7B).
Mapping of the ALK4 residues mutated in this study onto the modeled structure of the ALK4-ECD is shown in Fig. 7B. Our mutagenesis studies have identified five hydrophobic amino acid residues (Leu 40 , Ile 70 , Val 73 , Leu 75 , and Pro 77 ) that are each important for activin binding and that, together with contributions from several other residues, likely provide an interface for ligand binding and responsiveness. Based on the ALK3 structure (28), residues Leu 40 and Ile 70 are predicted to be groove-forming residues on ALK4 (Fig. 7). Residues Val 73 , Leu 75 , and Pro 77 , however, are in the region of ALK4 that has low homology with ALK3 (Fig. 1). Modeling predicts that this region of ALK4 (Pro 71 -Ala 78 ; see Fig. 7) would form a largely hydrophobic loop prior to the short ␣ helix. Fig. 7 shows that residues Val 73 , Leu 75 , and Pro 77 of this loop are in close proximity to residues Leu 40 and Ile 70 , and this hydrophobic patch likely mediates activin binding. Other residues that were mutated in this study are also indicated on the surface of the model of the ALK4-ECD. Residues Val 76 , Ala 78 , Lys 80 , Phe 82 , Tyr 83 , Arg 91 , and Thr 93 , which when mutated partially affected activin signaling, surround the hydrophobic patch. In contrast, residues that when mutated had no effect on ALK4 function (e.g. Ser 38 , Met 53 , Lys 72 , Pro 81 , Glu 88 , Asp 89 , and Leu 90 ) are, in general, distant to the putative binding surface.

DISCUSSION
The signaling events initiated by activin require binding of two types of transmembrane serine/threonine receptor kinases classified as type II (ActRII or ActRIIB) and type I (ALK4). Truncations or frameshift mutations within the kinase domain of ALK4 disrupt the antiproliferative effects of activin in pituitary adenomas (34) and pancreatic carcinomas (35) suggesting that ALK4 acts as a tumor suppressor. In addition, ALK4 has important roles in development as highlighted by the phenotype observed in ALK4 Ϫ/Ϫ embryos, in which the epiblast and the extraembryonic ectoderm are disorganized resulting in disruption and developmental arrest of the egg cylinder before gastrulation (36). Structural mapping of functional sites of ALK4 could, therefore, provide insights into the mechanisms underlying activin and related ligands signaling via ALK4 and activin's function in normal and disease physiology.
In recent years, crystal structure data have greatly advanced our understanding of how members of the TGF-␤ superfamily interact with their receptors. The structure of BMP-2 in complex with BMP receptor IA (ALK3) revealed a type I receptor fold similar to that of the ActRII-ECD (28,37). This structure shows that each ALK3-ECD molecule makes contacts with both subunits of the BMP-2 dimer. The ligand binding interface on ALK3 is characterized by a groove on the concave surface similar to the ligand binding surface on ActRII (28,37). The groove is formed by residues His 43 and Pro 45 on one side and Lys 79 , Met 78 , and Gln 86 on the other side; the floor of the groove is largely hydrophobic because of the side chains of Ile 99 , Phe 60 , and Met 78 , and the groove ends in a partly polar and partly hydrophobic hollow surrounded by side chains of Ile 62 , Ile 99 , and Arg 97 . The groove is filled by residues from the pre-helix loop of BMP-2. Outside the groove, Phe 85 of ALK3 fits with knob-into-hole packing into a hydrophobic pocket on BMP-2 and was proposed to be a key determinant for BMP-2 binding (28).
Based primarily on information from the crystal structure of the BMP2⅐ALK3-ECD complex, we subjected the ALK4-ECD to alanine scanning mutagenesis in an effort to identify the amino acid residues required for activin binding and function. We selected 26 residues on the ALK4-ECD that we individually mutated to alanine in the context of the kinase-deleted receptor. These mutants were expressed at similar levels as determined by Western blot experiments and were targeted to the cell surface based on detection of their respective amino-terminal FLAG epitopes on intact cells.
Our results indicate that there is only a partial overlap of the binding sites on ALK4 and ALK3 for activin-A and BMP-2, respectively. Of the predicted groove-forming residues on ALK4 (Ser 38 , Leu 40 , Met 53 , Ser 55 , Ile 70 , Pro 71 , Tyr 83 , Ser 86 , Arg 91 , and Thr 93 ) only mutation of residues Leu 40 and Ile 70 had pronounced effects on ALK4 function. Surprisingly, Phe 82 , which corresponds to Phe 85 on ALK3, and which was predicted to be a key feature generally required for type I receptor-ligand interactions (28), is not absolutely required for activin-A binding to ALK4. Of greater importance are residues (Val 73 , Leu 75 , and Pro 77 ) in the region preceding the short ␣ helix (Lys 80 -Leu 85 ) where there are five residues in ALK4 that do not have corresponding residues on the ALK3-ECD. Modeling predicts that this stretch of amino acids would form a largely hydrophobic loop prior to the short ␣ helix. This loop, together with contributions from residues Leu 40 and Ile 70 , likely forms a functional activin binding surface. Guimond et al. (33) in their recent study of the ALK5-ECD discussed the probable functional importance of non-homologous regions of type I receptors. Here we show that a region of low structural similarity between ALK4 and the BMP type I receptors (ALK2, ALK3, and ALK6) determines ligand binding.
Interestingly, of the seven ALKs, only ALK4 and ALK5 (type I TGF␤ receptor) contain the pre-helix loop extension. In addition, three of the ALK4 residues we have shown to be required for activin binding (Ile 70 , Leu 75 , and Pro 77 ; the ILP motif) are conserved in ALK5 (Ile 72 , Leu 77 , and Pro 79 ). Although limited mutagenesis of the ALK5-ECD has been performed (33), residues Ile 72 , Leu 77 , and Pro 79 were not targeted. Based on structural homology between activin and TGF-␤ and their respective type I receptor ECDs, we predict that Ile 72 , Leu 77 , and Pro 79 on the ALK5-ECD are likely to play an important role in TGF-␤ binding. The additional pre-helix residues present in ALK4 and FIG. 7. Model of the ALK4-ECD. A, the molecular surface of ALK3 (Protein Data Bank entry 1ES7) is color-coded; groove-forming residues are marked in orange whereas additional residues that come within 4.5 Å of BMP-2 are highlighted in pink. The horizontal bar indicates the plasma membrane. Residues constituting the putative activin binding surface are indicated on an ALK4-ECD space-filling model (B). Alanine substitutions of residues colored yellow had no effect on activin-A binding to the ALK4-ECD. Alanine substitutions of residues colored blue had either a minor (light blue) or major (dark blue) effect on activin binding. Residues colored green had enhanced ligand binding when mutated to alanine. C, stereo view of the ALK4-ECD model (black) superimposed on the ALK3-ECD (blue). The pre-helix extension of ALK4, the ␣ helix of ALK3, and the amino and carboxyl termini of the ECDs are indicated.
ALK5 may contribute to the inability of activin and TGF-␤ to bind to their type I receptors in the absence of their type II receptors (19,20). BMP type I receptors lack this pre-helix loop extension and can directly associate with their respective ligands (38).
Because activin does not bind ALK5, and TGF-␤ does not bind ALK4, residues on the ALK4-ECD other than the conserved ILP motif presumably determine ligand specificity. Mutation of residue Val 73 (Ile 75 on ALK5) to alanine resulted in a 2-fold increase in activin binding to ALK4 in the presence of ActRII, relative to wild-type ALK4. Moreover, in functional assays, mutant V73A was more efficient than wild-type ALK4 in mediating activin signaling. The mechanism by which mutant V73A mediates increased activin activity is unclear. As described previously (19), the affinity of activin for ActRII was not increased significantly by co-expression with wild-type ALK4, and it also appeared to be unaffected by the V73A mutant (data not shown). In addition, the V73A mutant does not bind ligand in the absence of ActRII (data not shown). Val 73 is in close proximity to the ILP motif and, based on the BMP-2⅐ALK3 structure, likely contacts residues in the wrist epitope of the activin dimer following activin binding to ActRII. It is possible that the conservative valine to alanine mutation at position 73 increases the affinity of ALK4 for the activin⅐ActRII complex. Other residues, including Leu 40 (Thr 42 on ALK5), are probably involved in determining ligand specificity; however, the identification of an activating mutation at residue 73 further highlights the importance of the pre-helix loop extension of ALK4 for activin binding.
Three of the hydrophobic residues required for activin binding and signaling (the ILP motif) are conserved between ALK4 and ALK5 and may mediate the similar manner in which activin and TGF-␤ interact with their type I receptors. The pre-helix extension that is integral to the activin binding site on ALK4 is absent in ALK2, ALK3, and ALK6 and may explain the differences in type I receptor binding observed between activins and BMPs. Future studies could address these questions of ligand specificity and mechanisms of type I receptor binding by generating ALK4/ALK3 (or ALK5) chimeras. Changing important ALK4 residues individually or in combination to the corresponding residues on ALK3 may yield chimeric ALK4 receptors that could bind activin and/or BMP-2 in the absence of type II receptors. The identification of residues on ALK4 that restrict activin binding in the absence of type II receptors could provide insight into the structural basis for the type I receptor recruitment model that describes activin signaling.
The importance of characterizing the ALK4 binding interface is further highlighted by the roles this receptor plays in development. Recent evidence indicates that signaling by other TGF-␤ ligands, including nodal, GDF-1, and Vg1, is mediated by ActRII/IIB and ALK4 (39,41). Nodal, GDF1, and Vg1 have been implicated in early embryonic patterning events (39,40). However, unlike activin, these other TGF-␤ ligands require additional co-receptors such as the EGF-CFC protein Cripto to generate signals (42). Current evidence suggests that Cripto, or other EGF-CFC proteins, bind to ALK4 and form a complex with nodal-related ligands and ActRII/IIB (40). It will be interesting to determine whether the Cripto binding site on ALK4 is distinct from the activin binding site. If the binding sites do not overlap, then specific activin and nodal agonists or antagonists targeting ALK4 could be generated. Such compounds could be used in the regulation of a wide array of diverse biological processes including hormone release, cell proliferation, differentiation, and pattern formation during embryogenesis (16,43).