Cripto Is a Noncompetitive Activin Antagonist That Forms Analogous Signaling Complexes with Activin and Nodal*

Cripto plays critical roles during embryogenesis and has been implicated in promoting the growth and spread of tumors. Cripto is required for signaling by certain transforming growth factor-β superfamily members, such as Nodal, but also antagonizes others, such as activin. The opposing effects of Cripto on Nodal and activin signaling seem contradictory, however, because these closely related ligands utilize the same type I (ALK4) and type II (ActRII/IIB) receptors. Here, we have addressed this apparent paradox by demonstrating that Cripto forms analogous receptor complexes with Nodal and activin and functions as a noncompetitive activin antagonist. Our results show that activin-A and Nodal elicit similar maximal signaling responses in the presence of Cripto that are substantially lower than that of activin-A in the absence of Cripto. In addition, we provide biochemical evidence for complexes containing activin-A, Cripto, and both receptor types and show that the assembly of such complexes is competitively inhibited by Nodal. We further demonstrate that Nodal and activin-A share the same binding site on ActRII and that ALK4 has distinct and separable binding sites for activin-A and Cripto. Finally, we show that ALK4 mutants with disrupted activin-A binding retain Cripto binding and prevent the effects of Cripto on both activin-A and Nodal signaling. Together, our data indicate that Cripto facilitates Nodal signaling and inhibits activin signaling by forming receptor complexes with these ligands that are structurally and functionally similar.

The transforming growth factor ␤ (TGF-␤) 3 superfamily contains ϳ40 structurally related ligands that control a wide range of diverse cellular processes, including proliferation, homeostasis, differentiation, and migration (1,2). TGF-␤ ligands exert their biological effects by binding and assembling two types of transmembrane receptors (type I and type II) with intrinsic serine/threonine kinase activities (3,4). There are five type II receptors and seven type I receptors; therefore, these receptors display varying degrees of promiscuity with regard to the ligands they bind (3). Activin type II (ActRII and ActRIIB) and type I (ALK4) receptors are especially promiscuous and form signaling complexes with activins but also several other superfamily members, including GDF1, GDF3, GDF8 (myostatin), and Nodal (5). In cases where multiple TGF-␤ ligands share the same receptors, signaling specificity can be achieved via ligand-specific differences, including variable expression patterns, processing of mature ligands from propeptides, stabilities, receptor affinities, susceptibility to extracellular ligand traps, and modulation by co-receptor proteins (3,4,6).
As an example, activin and Nodal utilize the same signaling receptors, but their mechanisms of ligand-mediated receptor assembly and use of co-receptor proteins differ significantly. On the one hand, activins initiate signaling by first binding Act-RII/IIB with high affinity (7)(8)(9) and then recruiting ALK4, resulting in the formation of active signaling complexes (10,11). By contrast, Nodal lacks intrinsic affinity for ActRII/IIB and ALK4 and requires Cripto (Cripto-1, TDGF1) or a related co-receptor to facilitate its binding to these receptors (12)(13)(14). Cripto is a small, GPI-anchored protein that possesses two modular cysteine-rich domains, an epidermal growth factorlike domain that binds Nodal and a CFC domain that binds ALK4 (13,15,16). Structure/function studies have shown that Nodal assembles type II and type I receptors only in the presence of Cripto. In fact, disruption of Nodal binding to the epidermal growth factor-like domain or ALK4 binding to the CFC domain completely abolishes Cripto-dependent Nodal signaling (12,17,18). Following receptor assembly, propagation of activin and Nodal signals is thought to be essentially the same (i.e. ALK4 is phosphorylated and activated by the constitutively active ActRII/ActRIIB receptor kinase and in turn phosphorylates cytoplasmic Smad2 and Smad3 proteins). Once phosphorylated, Smad2 and Smad3 assemble into complexes together with Smad4 and then migrate into the nucleus, where they interact with transcription factors, co-activators, and corepressors to regulate transcription of target genes (3).
In contrast to its requirement for Nodal signaling, Cripto has been shown to antagonize activin signaling apparently by preventing activin from assembling its receptors (19). For example, we have demonstrated that Cripto forms complexes with activin-A and ActRII/IIB and reduces activin-A cross-linking to ALK4 (20). This observation led us to propose a model in which Cripto inhibits activin-A signaling by competing with ALK4 for access to activin-A-ActRII/IIB complexes. Cripto was also shown to bind activin-B, and in this case it was proposed that Cripto inhibits activin-B signaling by forming nonproductive complexes with activin-B and/or ALK4 (21). Although these studies reported distinct inhibitory complexes, they each presented models in which Cripto competitively inhibits activin signaling by blocking receptor assembly. Notably, these models appear to be incompatible with the well documented role of Cripto as a Nodal co-receptor that facilitates receptor assembly. Specifically, it remains unclear how Cripto can form functional signaling complexes with Nodal while inhibiting activindependent receptor assembly.
In the present study, we provide evidence supporting a new model that explains the seemingly contradictory effects of Cripto on activin and Nodal signaling. We demonstrate that Cripto is a noncompetitive activin antagonist and that activin and Nodal assemble Cripto-containing receptor complexes that are structurally and functionally similar. We also identify ALK4 mutants that bind Cripto and block its effects on activin-A and Nodal signaling. Together, our data suggest that Cripto functions as a molecular switch that causes cellular responses to activin and Nodal to converge via the formation of analogous signaling complexes.

EXPERIMENTAL PROCEDURES
Materials-NuPAGE gels and molecular weight markers were obtained from Invitrogen. Recombinant human activin-A was generated using a stable activin-A-expressing cell line generously provided by Dr. J. Mather (Genentech, Inc., South San Francisco, CA) and was purified by Wolfgang Fischer (Peptide Biology Laboratory, Salk Institute, La Jolla, CA). Recombinant mouse Nodal was purchased from R&D Systems. 125 I-Activin-A was prepared using the chloramine T method as described previously (22). Anti-Myc (9E10) monoclonal antibody and protein G-agarose were purchased from Calbiochem. Polyclonal antibodies directed against ALK4 (23) have been described. A polyclonal anti-Cripto antibody (6900) was produced in rabbits immunized with a peptide from the epidermal growth factor-like domain of Cripto ( 82 CPPSFYGRNCEHD-VRKE 98 ). Anti-FLAG (M2) antibody and FLAG (M2)-agarose were from Sigma. Anti-phospho-Smad2, anti-Smad2/3, and anti-pan-actin antibodies were purchased from Cell Signaling Technologies (Danvers, MA). Horseradish peroxidase-linked anti-rabbit IgG, anti-mouse IgG, 3,3Ј,5,5Ј-tetramethylbenzidine substrate, chemiluminescent substrate (Supersignal TM ), and the BCA protein assay kit were obtained from Pierce. All DNA constructs used in this study were in the pcDNA3.0 expression vector (Invitrogen).
Expression Constructs-The mouse Cripto construct has been described (12) and was a generous gift from Malcolm Whitman (Department of Cell Biology, Harvard Medical School, Boston, MA). The ⌬CFC Cripto deletion mutant was generated using a previously described PCR strategy (24), in which the CFC domain was replaced with an in-frame BamHI site (GGATCC) encoding the amino acids Gly-Ser. The kinasedeleted ALK4 constructs (ALK4-trunc) encode the first 206 amino acids of ALK4 and were generated using standard PCR techniques, as previously described (25). Similar standard PCR strategies were used to make kinase-deleted ActRII constructs (ActRII-trunc) and to introduce epitope tags (25,26). All constructs were confirmed by DNA sequencing.
Transfection and Detection of Cell Surface Protein Expression in HEK 293T Cells-HEK 293T cells were grown in 5% CO 2 to 40 -60% confluence in complete Dulbecco's modified Eagle's medium (with 10% bovine calf serum, penicillin, streptomycin, and L-glutamine) on poly-D-lysine-coated wells or plates. Cells were transfected with indicated DNA constructs using Perfectin (Gene Therapy Systems). For Western blotting, cells were solubilized in radioimmune precipitation buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Nonidet P-40, 0.5% deoxycholate, and 0.1% SDS) containing standard protease inhibitors. SDS-PAGE and electrotransfer to nitrocellulose were carried out using NuPAGE gels and a NOVEX X-cell II apparatus as described previously (26). To detect proteins expressed at the cell surface, intact cells were washed with Hepes dissociation buffer (HDB), blocked with 3% bovine serum albumin/HDB, incubated with anti-FLAG antibody, washed with HDB, and incubated with peroxidase-conjugated anti-mouse IgG. Specific antibody staining was measured using the 3,3Ј,5,5Ј-tetramethylbenzidine peroxidase substrate, as described previously (26).
Luciferase Assays-Luciferase assays were carried out using the A3-luciferase reporter essentially as previously described (20). The A3-luciferase construct contains three copies of the activin response element from the Xenopus laevis Mix.2 promoter linked to a basic TATA box and a luciferase reporter gene. HEK 293T cells were plated on poly-D-lysine-coated 24-well plates at 1 ϫ 10 5 cells/well and transfected (Perfectin) in triplicate ϳ24 h later with 500 ng of DNA/well using 10 -400 ng of DNA constructs (indicated combinations of pcDNA 3.0 vector, Cripto, wild type (WT) or mutant ActRII-trunc, and WT or mutant ALK4-trunc) and 50 ng FAST2 (FoxH1), 25 ng of A3-luciferase, 25 ng of CMV-␤-galactosidase. Cells were treated ϳ24 h following the transfection and then harvested ϳ16 h following treatment. Cells were incubated in solubilization buffer (1% Triton X-100, 25 mM glycylglycine (pH 7.8), 15 mM MgSO 4 , 4 mM EGTA, and 1 mM dithiothreitol) for 30 min on ice, and luciferase reporter activity was measured and normalized relative to CMV-␤-Gal activities.
Smad2 Phosphorylation-HEK 293T cells were plated on 6-well plates at a density of 2 ϫ 10 5 cells/well. 24 h after plating, cells were transfected with 2 g of DNA (1 g of vector and 1 g of Cripto) using Perfectin. 24 h after transfection, cells were serum-starved overnight prior to treatment. Cells were left untreated or treated for 30 min with the indicated doses of activin-A or Nodal. Cells were harvested by adding 150 l of ice-cold radioimmune precipitation buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Nonidet P-40, 0.5% deoxycholate, 0.1% SDS) supplemented with 50 mM ␤-glycerol phosphate, 20 mM NaF, and standard protease inhibitors. Fifty l of 4ϫ SDS-PAGE loading buffer were then added to each sample, and the proteins were separated by SDS-PAGE and blotted to nitrocellulose. Blots were treated with anti-phospho-Smad2 or anti-Smad2/3 antibodies, followed by anti-rabbit IgG antibody conjugated to horseradish peroxidase, and bands were detected using enhanced chemiluminescence.
Covalent Cross-linking-HEK 293T cells were plated on 6-well plates coated with poly-D-lysine at a density of 4 ϫ 10 5 cells/well. Approximately 24 h later, cells were transfected with 2 g of DNA/well (1 g of ActRII-Myc, 0.5 g of ALK4-untagged and 0.5 g of WT-FLAG) using Perfectin and then incubated an additional 48 h before harvesting. Covalent cross-linking was performed by first washing cells in HDB (12.5 mM Hepes, pH 7.4, 140 mM NaCl, 5 mM KCl) and then incubating cells with 125 I-activin-A in binding buffer (HDB containing 0.1% bovine serum albumin, 5 mM MgSO 4 , 1.5 mM CaCl 2 ) at room temperature for 4 h. Cells were washed with HDB and incubated in 0.5 mM disuccinimidyl suberate in HDB for 30 min on ice. Cross-linking reactions were quenched with TBS (50 mM Tris-HCl, pH 7.5, 150 mM NaCl) and then cells were solubilized in lysis buffer (TBS containing 1% Nonidet P-40, 0.5% deoxycholate, 2 mM EDTA) with lysates subjected to immunoprecipitation by using anti-Myc, anti-FLAG, or anti-ALK4 antibodies. Immune complexes were then analyzed by SDS-PAGE and autoradiography. For Nodal competition, cells were pretreated with 100 nM Nodal for 30 min prior to incubating cells with 125 I-activin-A tracer.
Alternatively, P19 embryonal carcinoma cells were plated on 10-cm plates coated with poly-D-lysine at 2 ϫ 10 5 cells/well. Approximately 48 h later, covalent cross-linking was performed by first washing cells in HDB, treating with 10 nM cold activin-A, or leaving cells untreated and then incubating cells with 125 I-activin-A in binding buffer at room temperature for 2 h. Cells were washed with HDB and incubated in 0.5 mM disuccinimidyl suberate in HDB for 30 min on ice. Cross-linking reactions were quenched with TBS, and cells were solubilized in lysis buffer. Lysates were subjected to immunoprecipitation using anti-Cripto or nonspecific rabbit IgG antibodies. Immune complexes were then analyzed by SDS-PAGE and autoradiography.
Co-Immunoprecipitation and Western Blot Analysis-1 ϫ 10 6 HEK 293T cells were plated in 10-cm plates. Approximately 24 h later, cells were transfected with 12 g of DNA/plate (6 g of vector or untagged WT or ⌬CFC Cripto and 6 g of vector or FLAG-tagged WT or mutant ALK4-trunc) using polyethyleneimine, and cells were incubated another 48 h before harvesting. Cells were lysed and scraped on ice in 0.8 ml of cold radioimmune precipitation buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Nonidet P-40, 0.5% deoxycholate, and 0.1% SDS) containing standard protease inhibitors. Cellular lysates were precleared by centrifugation, and 75 l of the total lysates were heated and frozen in 4ϫ SDS-PAGE loading buffer (with dithiothreitol), whereas the remainder of the lysates were incubated overnight at 4°C with FLAG (M2)-agarose. Precipitated complexes were washed three times with cold radioimmune precipitation buffer for 1 h each, eluted from beads, and then subjected to SDS-PAGE and electrotransfer to nitrocellulose, followed by Western blotting using the Cripto antibody.

Cripto Reduces Maximal Activin-A-dependent Luciferase Induction to a Level That Converges with That of Nodal-In
order to compare the roles of Cripto as a co-receptor for Nodal and an inhibitor of activin-A signaling, 293T cells were transfected with the Smad2-responsive luciferase reporter A3-Lux and FAST2 (FoxH1) in the presence of vector or Cripto and then treated with a range of doses of each ligand. Consistent with previous reports that Nodal requires Cripto as an obligatory co-receptor (14), we did not observe Nodal signaling in the absence of Cripto even at the highest dose of 180 nM Nodal (Fig.  1A). By contrast, in the presence of Cripto, Nodal caused a dose-dependent induction of luciferase with an EC 50 of ϳ14 nM and a maximal induction of ϳ20-fold basal (Fig. 1A).
In parallel experiments, we measured the dose-response relationship for activin-A in the absence or presence of Cripto. In the absence of Cripto, maximal doses of activin-A elicited a robust ϳ40-fold induction of luciferase, which was significantly larger than that observed in response to maximal doses of Nodal in the presence of Cripto (Fig. 1, A and B). Activin-A had an EC 50 of ϳ60 pM and was therefore nearly 250 times more potent than Nodal (EC 50 ϳ 14 nM). Consistent with our previous findings (20), activin-A signaling was inhibited by the presence of Cripto, and this inhibition was observed at each dose of activin-A tested (Fig. 1B). However, the potency of activin-A in the presence of Cripto (EC 50 of ϳ41 pM) was essentially the same as in the absence of Cripto (EC 50 of ϳ60 pM), suggesting that activin-A affinity for the type II receptor (ActRII/IIB) was unaffected by Cripto. Notably, despite their disparate potencies, activin-A and Nodal displayed a similar maximal level of luciferase induction (ϳ20-fold basal) in the presence of Cripto (Fig. 1, A and B). Thus, Cripto reduced the maximal signaling capacity of activin-A to a level similar to that of Nodal.
Next, we asked how the level of Cripto expression in these cells affects Nodal and activin-A signaling. 293T cells were transfected with a range of doses of the Cripto expression vector, treated with 30 nM Nodal or 250 pM activin-A, and then resulting luciferase induction was measured. As shown in Fig.  1C, Nodal signaling increased as the level of Cripto DNA was increased, and this effect was mirrored by a corresponding decrease in the activin-A response. Strikingly, as Cripto DNA doses were increased, Nodal and activin-A responses converged at ϳ20-fold basal (Fig. 1C). The Nodal and activin-A responses were indistinguishable following transfection of 300 or 500 ng of Cripto DNA (Fig. 1C), despite the fact that the level of cell surface Cripto expression was substantially higher following transfection with 500 ng of Cripto DNA than it was following transfection of 300 ng (Fig. 1D). Therefore, since maximal inhibition of activin-A signaling by Cripto was partial and achieved at submaximal Cripto levels, we conclude that Cripto is a noncompetitive activin-A antagonist.
Cripto Reduces Activin-A-induced Smad2 Phosphorylation to a Level Similar to That of Nodal-In order to further compare the effects of Cripto on activin-A and Nodal signaling, we measured the relative ability of these ligands to activate Smad2 phosphorylation in the absence or presence of Cripto. In agreement with the luciferase data presented above, we have shown that transfection of Cripto is necessary for Nodal-induced Smad2 phosphorylation (Fig. 2). By contrast, activin-A induced robust Smad2 phosphorylation in these cells in the absence of Cripto. Importantly, and consistent with the luciferase assays, transfec-tion of Cripto into these cells reduced the extent of activin-Ainduced Smad2 phosphorylation to a level resembling that of Nodal (Fig. 2).
In summary, three novel conclusions can be drawn from these data measuring the effects of Cripto on Nodal and activin-A signaling: 1) activin-A has a higher intrinsic efficacy than Nodal; 2) Cripto does not change the potency (EC 50 ) of activin-A but rather acts in a noncompetitive manner to reduce activin-A efficacy; and 3) the efficacies of activin-A and Nodal converge as Cripto expression increases.
Activin-A Assembles Complexes Containing ActRII, ALK4, and Cripto-The similar magnitudes of activin-A and Nodal signaling responses in the presence of Cripto suggested that these ligands form structurally similar complexes containing Cripto and both receptor types. To test this, we transfected 293T cells with ActRII-Myc, ALK4, and/or Cripto-FLAG constructs, as indicated (Fig. 3), labeled them with 125 I-activin-A, and then subjected them to covalent cross-linking. Resulting cell lysates were then subjected to immunoprecipitation using antibodies targeting ActRII-Myc (Fig. 3A), ALK4 (Fig. 3B), or Cripto-FLAG (Fig. 3C). As expected, anti-Myc immunoprecipitation of ActRII led to the isolation of 125 I-activin-A-labeled ActRII (lane 2), ActRII and ALK4 (lane 3), ActRII and Cripto (lane 4), and ActRII, ALK4, and Cripto (lane 5) (Fig. 3A). Although Cripto reduced cross-linking of activin-A to ALK4, it is notable that Cripto does not completely abolish this interaction, raising the possibility that Cripto may exist in a complex with activin-A and both receptor types (Fig. 3A, compare lanes  3 and 5). However, since ActRII was the target in the immuno- Resulting cell lysates were subjected to Western blotting (IB) using phospho-Smad2, Smad2, or Cripto antibodies, as described under "Experimental Procedures."  FEBRUARY 22, 2008 • VOLUME 283 • NUMBER 8 precipitation, we could not rule out the possibility that the labeled bands corresponding to ALK4 and Cripto (Fig. 3A, lane 5) were derived from distinct complexes consisting of activin-A-ActRII-Cripto or activin-A-ActRII-ALK4. This issue was resolved by immunoprecipitation with an anti-ALK4 antibody that resulted in the clear detection of labeled bands corresponding to ActRII, ALK4, and Cripto (Fig. 3B, lane 3). Since ALK4 and Cripto do not bind activin-A in the absence of ActRII (20), this result demonstrates that activin-A, ActRII, ALK4, and Cripto are each indeed present in the same complexes. This finding was further confirmed by our visualization of bands corresponding to 125 I-activin-A-labeled ActRII, ALK4, and Cripto following anti-FLAG immunoprecipitation (Fig. 3C,  lane 3).

Mechanism and Blockade of Cripto
Having shown that activin-A can form complexes with Act-RII, ALK4, and Cripto when they are overexpressed in 293T cells, we attempted to demonstrate the formation of these complexes in cells expressing these proteins at endogenous levels. Mouse embryonal carcinoma P19 cells express relatively high levels of endogenous Cripto, and we therefore subjected these cells to 125 I-activin-A labeling, covalent cross-linking, and immunoprecipitation using an antibody that targets Cripto. To control for nonspecific immunoprecipitation effects of rabbit antisera, we assessed the ability of rabbit IgG to immunoprecipitate similar complexes. As shown in Fig. 3D, 125 I-activin-Alabeled bands corresponding to ActRII, ALK4, and Cripto were visualized following immunoprecipitation with an anti-Cripto antibody but not with nonspecific IgG (Fig. 3D). Furthermore, the appearance of these labeled bands was almost completely abolished when cells were preincubated with 10 nM unlabeled activin-A. Significantly, this is the first evidence that demonstrates that activin-A can form receptor complexes that contain endogenous ActRII, ALK4, and Cripto.
Nodal Competes with Activin-A for the Formation of Cripto-ActRII-ALK4 Complexes-Our results indicate that, like Nodal, activin-A can form complexes that contain both receptor types and Cripto. Given the apparent similarity of Cripto-containing activin and Nodal complexes, we reasoned that Nodal and activin-A might compete with each other to assemble receptor complexes containing Cripto. To test this possibility, 293T cells were transfected with the indicated constructs (Fig.  4A) and then subjected to 125 I-activin-A labeling in the absence or presence of 100 nM Nodal. As shown in Fig. 4A, pretreatment of cells with 100 nM Nodal substantially decreased the intensity of 125 I-activin-A-labeled receptor and Cripto bands, as visualized following Cripto immunoprecipitation, indicating that Nodal competitively disrupted the formation of both activin-A-ActRII-Cripto (Fig. 4A, lanes 3 and 4) and activin-A-ActRII-ALK4-Cripto complexes (Fig. 4A, lanes 5 and 6). Interestingly, Nodal competed with 125 I-activin-A cross-linking to Cripto in the absence of ALK4, further supporting previously published work demonstrating that Nodal can bind Cripto directly (12).
We predicted that activin-A and Nodal would not have additive signaling effects in the presence of Cripto, since they apparently compete to form Cripto-containing receptor complexes that elicit similar signaling responses. To test this directly, we asked whether Nodal could increase the activin-A response in the absence or presence of Cripto. 293T cells were transfected with 400 ng of vector or Cripto and then treated with 100 pM activin-A, 30 nM Nodal, or 100 pM activin-A plus 30 nM Nodal. As shown in Fig. 4B, in the absence of Cripto, activin-A elicits a ϳ27-fold induction of luciferase, and, as predicted, Nodal has no signaling effect either alone or together with activin-A. In cells transfected with 400 ng of the Cripto construct, the activin-A response was inhibited to ϳ18-fold, whereas Nodal caused an ϳ13-fold induction. Importantly, when these cells were treated with activin-A and Nodal together, the signaling response was essentially the same as that observed following treatment with activin-A alone (i.e. Nodal does not increase the activin-A signaling response in the presence of Cripto). These data are consistent with our cross-linking data and suggest that FIGURE 4. Nodal and activin-A compete to form complexes with ActRII, ALK4, and Cripto. A, 293T cells were transfected with the indicated constructs, treated as indicated with 100 nM unlabeled Nodal, and then subjected to 125 I-activin-A cross-linking as described under "Experimental Procedures." Cross-linked complexes were subjected to immunoprecipitation using antibodies targeting Cripto-FLAG, and then immune complexes were resolved via SDS-PAGE and visualized by autoradiography as described under "Experimental Procedures." Corresponding protein levels from total cell extracts were determined by Western blotting. B, 293T cells were transfected with A3-lux, FAST2, and CMV-␤-galactosidase together with 200 ng of empty vector or Cripto DNA as described under "Experimental Procedures." Cells were treated with 100 pM activin-A, 30 nM Nodal, or 100 pM activin-A plus 30 nM Nodal, and resulting normalized luciferase activities are presented as -fold induction.
Nodal and activin-A compete for access to the Cripto-bound fraction of ALK4 to form similar, low efficacy signaling complexes.
Nodal and Activin-A Bind the Same Site on ActRII-Next, we probed the structural similarity of Nodal and activin-A signaling complexes in more detail by testing whether these ligands bind the same ActRII ECD residues. Activin-A binds a hydrophobic surface on the ActRII ECD consisting primarily of Phe 42 , Trp 60 , and Phe 83 , and mutation of any one of these three residues to alanine abolishes binding (26). Although the binding site for Nodal on ActRII has not yet been identified, it might be predicted to differ from that of activin-A, since the ActRIIbinding residues on activin-A are not conserved on Nodal and since Nodal binding to ActRII requires Cripto.
To determine whether Nodal binds the same surface on the ActRII ECD as activin-A, we used transmembrane, kinase-deleted ActRII (ActRII-trunc) constructs that bind cognate ligands, form nonproductive ligand-receptor complexes, and block signaling in a dominant negative manner (28,29). We compared the abilities of WT ActRII-trunc and the ActRIItrunc mutants F42A, W60A, and F83A to block activin-A and Nodal signaling. As a control, we also tested the function of K56A ActRII-trunc, since the K56A mutation did not affect activin-A binding to ActRII (26). 293T cells were transfected with empty vector, Cripto, and various ActRII-trunc constructs, as indicated, and treated with 100 pM activin-A or 30 nM Nodal, and then resulting luciferase induction was measured. Cripto inhibited activin-A signaling and facilitated Nodal signaling, as predicted, such that both ligands signaled at a similar level in its presence (Fig. 5). By contrast, the signaling response of both ligands was nearly abolished when WT ActRII-trunc and Cripto were co-transfected. Importantly, this result indicates that ActRII-trunc acts in a dominant negative manner to block signaling by both activin-A and Nodal in the presence of Cripto. As expected, the K56A ActRII-trunc mutant resembled WT ActRII-trunc in its ability to block activin-A signaling, and it also blocked Nodal signaling. By contrast, the F42A, W60A, and F83A ActRII-trunc constructs were only weakly able to inhibit signaling by both activin-A and Nodal, indicating that in addition to disrupting activin-A binding, each of these ActRII ECD substitutions also prevents Nodal binding (Fig. 5). Therefore, although activin and Nodal have distinct modes of accessing ActRII, these data indicate that both ligands interact with the same hydrophobic surface on ActRII constituted by Phe 42 , Trp 60 , and Phe 83 . This finding supports our model in which both activin-A and Nodal can form ActRII-ALK4-Cripto signaling complexes with similar structures.
Truncated ALK4 Mutants with Disrupted Activin-A Binding Retain the Ability to Bind Cripto-Nodal forms signaling complexes that contain Cripto, ActRII, and ALK4, and in these complexes ALK4 is thought to form distinct contacts with Cripto and Nodal (12). Here we have tested whether ALK4 also has distinct and separable binding sites for Cripto and activin-A. Similar to ActRII-trunc, membrane-anchored, kinase-deleted ALK4 (ALK4-trunc) inhibits activin-A signaling by forming nonproductive complexes with ActRII and activin-A (30). We previously used alanine substitution mutagenesis to identify a cluster of three amino acid residues on the ALK4-ECD that are each required for the dominant negative function of ALK4-trunc (25). These residues, Ile 70 , Leu 75 , and Pro 77 , were subsequently shown to be required for activin-A binding as well as for ALK4 signaling function (25). Since the binding sites on ALK4 for Nodal and Cripto are thought to be distinct, we reasoned that ALK4-trunc mutants with disrupted activin-A binding might retain their ability to bind Cripto.
To test this, we transfected 293T cells with untagged WT Cripto or ⌬CFC Cripto in the absence or presence of FLAGtagged WT ALK4-trunc, L75A ALK4-trunc, P77A ALK4trunc, or I70A ALK4-trunc. Cell lysates were then subjected to anti-FLAG immunoprecipitation followed by Western blot using anti-Cripto or anti-FLAG antibodies. As predicted, WT Cripto co-precipitated with WT ALK4-trunc, whereas ⌬CFC did not (Fig. 6). Importantly, the L75A, P77A, and I70A ALK4trunc proteins each co-precipitated with WT Cripto but not with ⌬CFC, indicating that these ALK4 mutations do not affect Cripto-ALK4 binding. This result demonstrates that ALK4 has distinct and separable binding sites for activin-A and Cripto, and it is consistent with our conclusions drawn from data pre- Corresponding cell lysates were subjected to Western blotting (IB) using Myc or Cripto antibodies as described under "Experimental Procedures." FIGURE 6. Kinase-deleted ALK4 (ALK4-trunc) mutants with disrupted activin-A binding retain Cripto binding. 293T cells were transfected with the indicated Cripto and/or FLAG-tagged ALK4-trunc constructs with resulting cell lysates subjected to anti-FLAG immunoprecipitation (IP) followed by Western blotting using Cripto or FLAG antibodies. Alternatively, total cell extracts were subjected to Western blotting using a Cripto antibody as described under "Experimental Procedures." sented above that activin-A and Nodal can form structurally and functionally similar receptor complexes containing Cripto.

ALK4-trunc Mutants Antagonize the Opposing Effects of Cripto on Activin-A-and Nodal-dependent Luciferase
Induction-Since these ALK4-trunc mutants, deficient in activin binding, retain the ability to bind Cripto, we hypothesized that they might sequester Cripto away from ligand-receptor complexes, thereby reducing Nodal signaling on the one hand while increasing activin-A signaling on the other. To test this possibility, we asked whether these ALK4-trunc mutants could target Cripto function. 293T cells were transfected with empty vector, Cripto, ALK4-trunc, or L75A ALK4-trunc, as indicated. As shown in Fig. 7A, L75A ALK4-trunc had no effect on activin-A signaling in the absence of Cripto but was able to block Cripto antagonism of activin-A signaling in a dose-dependent manner. Conversely, as shown in Fig. 7B, increasing amounts of L75A ALK4-trunc DNA caused a dose-dependent decrease in Cripto-dependent Nodal signaling. However, the highest DNA dose of the L75A ALK4-trunc construct only blocked Nodal signaling by ϳ50%. This may be explained by our observation that Nodal signaling can be facilitated by very low levels of transfected Cripto DNA as well as by the fact that Cripto is overexpressed in this experiment. Therefore, we predict that a high DNA ratio of L75A ALK4-trunc to Cripto may be required to see a complete block of Nodal signaling in this assay. As expected, the I70A and P77A ALK4-trunc mutants possess similar Cripto-blocking capability (data not shown).
In order to ensure that the progressive loss of Cripto function was due to the presence of L75A ALK4-trunc protein and not the result of a decrease in Cripto expression, we measured the levels of these proteins at the cell surface. Fig. 7C shows that the level of cell surface L75A ALK4-trunc expression correlated with the amount of DNA transfected and, importantly, that its expression did not cause the cell surface levels of Cripto to decrease. Finally, we have performed Western blots on total cell lysates from 293T cells transfected with an equal molar ratio of Cripto and L75A ALK4-trunc DNA under conditions that resulted in a blockade of Cripto function. As shown in Fig. 7D, the expression levels of these proteins are unaffected by their co-expression. Therefore, in summary, we have shown that the L75A ALK4-trunc mutant binds Cripto and also prevents Cripto effects on Nodal-and activin-A-induced luciferase expression.
Activin-A and Nodal Form Analogous Receptor Complexes with Cripto via Distinct Assembly Mechanisms-In the absence of Cripto, activin-A forms high potency, high efficacy signaling complexes with ActRII and ALK4. Conversely, Nodal is incapable of signaling in the absence of Cripto (Fig. 8A). The data presented here support a model in which activin-A and Nodal each form similar signaling complexes that contain Cripto 293T cells were transfected with empty vector or Cripto DNA in the absence or presence of wild-type ALK4-trunc or the indicated doses of L75A ALK4-trunc DNA together with A3-lux, FAST2, and CMV-␤-galactosidase constructs as described under "Experimental Procedures." Cells were treated with 250 pM activin-A (A) or 10 nM Nodal (B), and resulting luciferase activities were normalized relative to ␤-galactosidase activities and are presented as percentage of maximal induction. C, cell surface expression of Cripto and L75A ALK4-trunc was measured using an intact cell enzyme-linked immunosorbent assay as described under "Experimental Procedures." D, Cripto and L75A ALK4-trunc constructs were transfected into 293T cells at an equimolar DNA ratio, and then resulting cell lysates were subjected to Western blotting (IB) using Cripto, FLAG, or actin antibodies as described under "Experimental Procedures." (Fig. 8B). According to this model, activin-A first binds its type II receptors with high affinity and then recruits ALK4 and Cripto into high potency, low efficacy complexes. On the other hand, Nodal first binds Cripto before it can assemble ActRII/IIB and ALK4 into low potency, low efficacy complexes. Fig. 8C schematically outlines how the level of Cripto impacts Nodal and activin signaling. In the absence of Cripto (Fig. 8C, a), activin forms only high efficacy signaling complexes, whereas Nodal is unable to signal. As the level of Cripto increases, activin signaling decreases, and Nodal signaling increases until they converge (Fig. 8C, b). At this point, all receptor complexes contain Cripto and therefore have low efficacy. Finally, as shown in Fig. 8D, ALK4-trunc mutants deficient in activin-A binding retain the ability to bind Cripto and block its ability to form signaling complexes with activin-A and Nodal.

DISCUSSION
TGF-␤ ligands are morphogens that regulate gene expression and specify cell fate in a concentration-dependent manner (4,31). However, Smad proteins are the central mediators of TGF-␤ signaling, and therefore the magnitude and duration of the Smad response are critical in determining patterns of gene expression. Importantly, Smad signaling is shaped not only by the concentration of TGF-␤ ligands but also by an array of celland ligand-specific modulators that operate through multiple mechanisms (3,4). Cripto is one such modulator that affects the signaling of several TGF-␤ ligands that activate the Smad2/3 pathway. These include Nodal (13), GDF1 (32), GDF3 (33), activin-A (20), activin-B (21), and TGF-␤1 (34). Interestingly, although Cripto functions as an obligatory co-receptor for Nodal, GDF1, and GDF3, it antagonizes signaling by activin-A, activin-B, and TGF-␤1. Therefore, by altering its level of Cripto expression, a cell can tune its responsiveness to these ligands and vary its level of Smad2/3 signaling over a broad range.
Cripto Reduces Activin Efficacy to a Level Similar to That of Nodal-In the present study, we conducted a detailed comparison of the roles of Cripto as a modulator of activin and Nodal signaling. In contrast to previous reports exclusively postulating that Cripto inhibits activin signaling by forming inactive complexes with activin and/or its receptors, our data support a novel mechanism whereby Cripto participates in functional FIGURE 8. Cripto reduces activin-A efficacy to a level similar to that of Nodal and forms similar signaling complexes with both ligands. A, in the absence of Cripto, activin-A forms high potency, high efficacy signaling complexes with ActRII/IIB and ALK4, whereas Nodal does not bind these receptors or initiate signaling. B, when Cripto is present, activin-A and Nodal each form structurally similar signaling complexes that contain Cripto. Activin-A first binds its type II receptors with high affinity and then recruits ALK4 and Cripto into a high potency, low efficacy complex. Nodal first binds Cripto before it can assemble ActRII/IIB and ALK4 into a low potency, low efficacy complex. C, diagram illustrating the ability of Cripto to modulate activin and Nodal signaling in a dose-dependent manner. In the absence of Cripto, activin signaling is high, whereas Nodal does not signal (a). As Cripto levels increase, activin signaling decreases, and Nodal signaling increases until they converge (b). D, ALK4-trunc mutants deficient in activin-A binding retain the ability to bind Cripto and block its ability to form signaling complexes with activin-A and Nodal.
activin signaling complexes that have reduced signaling capacity. We conclude that Cripto is a noncompetitive activin antagonist, since it reduced activin efficacy without altering activin potency (EC 50 ). Also, its maximal inhibitory effect on activin signaling was partial and attained at submaximal Cripto expression levels. In light of these findings, we propose that activin receptor complexes can exist in two signaling states, a high efficacy state in the absence of Cripto (Fig. 8, A and C, a) and a lower efficacy state in the presence of Cripto (Fig. 8, B and C, b). According to this model, Nodal and other ligands that require Cripto can only signal in the low efficacy state. Activins, on the other hand, signal in the high efficacy state when Cripto is absent and in the low efficacy state when Cripto is present.
This two-state model of activin receptor signaling is supported by our discovery that the maximal signaling responses to activin and Nodal converge as the level of Cripto expression increases. Specifically, Cripto reduced activin-A-dependent luciferase induction and Smad2 phosphorylation to levels indistinguishable from those induced by Nodal. In each of these assays, we found that activin-A and Nodal had similar response maxima in cells expressing high levels of Cripto. However, although Cripto effectively converted activin-A into a partial agonist with efficacy resembling that of Nodal, it did not alter activin-A potency, indicating that Cripto does not affect the affinity of activin-A for ActRII/IIB. In this regard, we found that activin-A was nearly 250 times more potent than Nodal, suggesting that the affinity of activin-A for ActRII/IIB is much higher than that of Nodal for Cripto, although this remains to be tested directly. In summary, we propose that activin and Nodal form functionally similar signaling complexes in the presence of Cripto, despite the fact that these ligands have distinctly different mechanisms of receptor assembly.
Our results demonstrate that activin can elicit a substantially larger signaling response in the absence of Cripto than Nodal can generate in the presence of Cripto. This finding predicts that Nodal will be unable to regulate the expression of certain activin-responsive genes that require high levels of Smad2/3 signaling. On the other hand, our results show that activin and Nodal signaling responses converge in the presence of Cripto and therefore also predict that activin and Nodal will elicit similar biological effects on cells that express Cripto. Notably, these predictions appear to be supported by the demonstration that activin-B and Xenopus Nodal-related proteins (Xnrs) regulate the expression of disparate sets of genes during Xenopus embryogenesis, with activin-B affecting the expression of nearly twice as many genes as Xnrs (35). In addition to cooperating with Xnrs to regulate developmental genes, such as goosecoid, chordin, Xbra, Xnr2, and Derrière (36), Ramis et al. (35) showed that activin-B exclusively regulated the expression of several genes involved in cell cycle control, consistent with its antiproliferative role during gastrulation. In this regard, it is interesting to note that Smad2/3-mediated antiproliferative effects have previously been reported to require high, sustained levels of Smad2/3 signaling (37). Therefore, high efficacy activin signaling may be required to induce growth arrest during Xenopus gastrulation, whereas lower efficacy activin and Xnrs signaling in the presence of Xenopus Cripto proteins may cooperate to control mesodermal patterning. Importantly, these studies appear to be consistent with our observation that the intrinsic efficacy of activin-A is greater than that of Nodal but that these ligands signal similarly in the presence of Cripto. Further studies will be necessary to determine how Cripto expression levels shape gene transcription responses to activin and Nodal ligands.
Activin and Nodal Form Structurally Similar Complexes with ActRII, ALK4, and Cripto-The similar efficacies of activin and Nodal in the presence of Cripto suggested that these ligands assemble Cripto-containing signaling complexes that are structurally similar. In support of this, we provide biochemical evidence for complexes containing activin-A, Cripto, and both receptor types and show that the assembly of such complexes is competitively inhibited by Nodal. We further demonstrate that Nodal and activin-A share the same binding site on ActRII. This result was somewhat surprising, since Nodal, unlike activin, requires Cripto to bind type II receptors (12). Notably, activin residues that mediate type II receptor binding are not conserved in Nodal, and substitution of a 14-amino acid ActRIIbinding region from activin into Nodal resulted in a chimera that could signal in a Cripto-independent manner (38). Despite these differences in their type II receptor binding properties, however, our results clearly demonstrate that activin-A and Nodal bind the same residues on ActRII. We have also shown that ALK4 has distinct and separable binding sites for activin-A and Cripto. This result supports our observation that ALK4 and Cripto are both present in complexes together with activin-A and ActRII and is also consistent with our conclusion that activin-A and Nodal form similar receptor complexes in the presence of Cripto. Together, these results indicate that despite their differing mechanisms of receptor assembly, Nodal and activin form structurally similar complexes with Cripto and activin type I and type II receptors.
Although our data indicate that Cripto is present in activin receptor complexes, they do not address the question of how the presence of Cripto in these signaling complexes reduces activin efficacy. We propose two possible models. First, although we show that Cripto does not preclude ALK4 recruitment into activin signaling complexes as was previously proposed (20), it is possible that ActRII-bound activin has lower affinity for Cripto-bound ALK4 than it has for ALK4 alone. Therefore, by reducing the affinity of activin for ALK4, Cripto could reduce the stability of activin signaling complexes and thereby reduce signaling. Such a mechanism is consistent with our observation here and previously (20) that Cripto reduces activin cross-linking to ALK4. Alternatively, Cripto may act as a wedge that distorts signaling complexes in a way that limits the ability of the ActRII kinase to phosphorylate ALK4 or the ability of ALK4 to phosphorylate Smads or both. Biochemically, these spatial constraints may explain the observed decrease in the efficiency of cross-linking between activin and ALK4 in the presence of Cripto. Such a model postulates that Cripto imposes similar structure/function constraints on signaling complexes assembled by either activin-A or Nodal and is therefore appealing, since it can explain our observation that activin-A and Nodal attain similar signaling maxima in the presence of Cripto. Three-dimensional structures of complexes containing activin or Nodal, their receptors, and Cripto will allow a comparison with similar structures that have been solved in the absence of Cripto (39 -41) and will provide further insight into the structural basis for the effects of Cripto on signaling complexes containing ActRII/IIB and ALK4. According to this revised model, we predict that the three-dimensional structures of activin and Nodal in complex with Cripto and their receptors will be very similar.
ALK4 Mutants Block Cripto Effects on Activin and Nodal Signaling-We have shown here that three ALK4-trunc mutants each lacking activin binding were indistinguishable from WT ALK4-trunc in their ability to bind Cripto. This finding indicates that ALK4 has distinct and separable binding sites for activin and Cripto. We have further demonstrated that these ALK4-trunc mutants inhibit Cripto-dependent effects on activin and Nodal signaling, as illustrated by the L75A ALK4trunc mutant. The effects of this mutant were dose-dependent and apparently stemmed from its ability to bind Cripto and sequester it away from ligand-receptor complexes (Fig. 8D). However, future studies will be required in order to elucidate the precise mechanism by which these ALK4 mutants inhibit specific Cripto functions.
In addition to supporting our model in which Cripto mediates activin and Nodal signaling by forming similar signaling complexes with these ligands, the ability of these ALK4-trunc mutants to block Cripto effects on activin and Nodal signaling suggests that they may have therapeutic value as Cripto inhibitors. For example, Cripto has been identified as a marker of stem cell pluripotency (42), and it has been shown that ES cells lacking Cripto spontaneously differentiate into neurons (43)(44)(45). Therefore, developing a Cripto-blocking reagent, such as L75A ALK4-trunc, may aid efforts aimed at providing cell-based treatments for neurodegenerative disorders. Cripto is also highly expressed in human tumors and is thought to promote tumorigenesis via multiple mechanisms (16), including antagonism of activin (20,21) and TGF-␤ (34) signaling as well as facilitation of Nodal signaling (27). Therefore, these ALK4-trunc mutants may also have utility as cancer therapeutic agents.
In summary, the results presented here provide a molecular mechanism that explains the basis for the opposing effects of Cripto on activin and Nodal signaling. We have shown for the first time that the intrinsic efficacy of activin is higher than that of Nodal, and we have further demonstrated that Cripto forms reduced efficacy complexes with activin and its receptors that structurally and functionally resemble Nodal signaling complexes. Thus, our results reconcile the opposing effects of Cripto on Nodal and activin signaling and suggest a model in which Cripto participates in similar signaling complexes with each ligand.