Identification of Two Amino Acids in Activin A That Are Important for Biological Activity and Binding to the Activin Type II Receptors*

Activins are members of the transforming growth factor-β family of growth and differentiation factors. In this paper, we report the results of a structure-function analysis of activin A. The primary targets for directed mutagenesis were charged, individual amino acids located in accessible domains of the protein, concentrating on those that differ from transforming growth factor-β2, the x-ray crystal structure of which is known. Based on the activities of the recombinant activin mutants in two bioassays, 4 out of 39 mutant proteins (D27K, K102A, K102E, and K102R) produced in a vaccinia virus system were selected for further investigation. After production in insect cells and purification of these four mutants to homogeneity, they were studied in bioassays and in cross-linking experiments involving transfected receptor combinations. Mutant D27K has a 2-fold higher specific bio-activity and binding affinity to an ActRIIA/ALK-4 activin receptor complex than wild type activin, whereas mutant K102E had no detectable biological activity and did not bind to any of the activin receptors. Mutant K102R and wild type activin bound to all the activin receptor combinations tested and were equipotent in bioassays. Our results with the Lys-102 mutants indicate that the positive charge of amino acid 102 is important for biological activity and type II receptor binding of activins.

Activins are members of the transforming growth factor-␤ family of growth and differentiation factors. In this paper, we report the results of a structure-function analysis of activin A. The primary targets for directed mutagenesis were charged, individual amino acids located in accessible domains of the protein, concentrating on those that differ from transforming growth factor-␤2, the x-ray crystal structure of which is known. Based on the activities of the recombinant activin mutants in two bioassays, 4 out of 39 mutant proteins (D27K, K102A, K102E, and K102R) produced in a vaccinia virus system were selected for further investigation. After production in insect cells and purification of these four mutants to homogeneity, they were studied in bioassays and in cross-linking experiments involving transfected receptor combinations. Mutant D27K has a 2-fold higher specific bio-activity and binding affinity to an ActRIIA/ALK-4 activin receptor complex than wild type activin, whereas mutant K102E had no detectable biological activity and did not bind to any of the activin receptors. Mutant K102R and wild type activin bound to all the activin receptor combinations tested and were equipotent in bioassays. Our results with the Lys-102 mutants indicate that the positive charge of amino acid 102 is important for biological activity and type II receptor binding of activins.
The TGF-␤ 1 family consists of a large group of structurally related, but functionally diverse polypeptides that control the growth and differentiation of many cell types in vitro and in vivo (1)(2)(3)(4). TGF-␤s, activins, and bone morphogenetic proteins (BMPs) exert their biological effects through binding to two types of serine/threonine kinase receptors, termed type I (Ϯ 53 kDa) and type II (Ϯ 70 kDa) receptors (1,5,6). Type I and II receptors can form high affinity receptor complexes at the cell surface and this is necessary for signal transduction (7)(8)(9)(10)(11). Overexpressed type II receptors can bind ligand in the absence of type I receptor with moderate affinity, while it is generally accepted that type I receptors require type II receptors to bind ligand in the high affinity receptor complex. The type II receptor phosphorylates the type I receptor after ligand binding, and the latter propagates the signal to downstream effectors, the Smad proteins (see Ref. 12).
TGF-␤ members are biologically active as dimers. Like other members of the TGF-␤ family, the activins are synthesized as large precursor proteins consisting of a signal peptide, a glycosylated prodomain and a mature domain. The maturation of activin requires intracellular cleavage by protein convertases, such as furin, at the basic cleavage site which separates the mature chain from the prodomain (13,14). Removal of the prodomains from the precursor dimer is necessary for biological activity of the mature 25-kDa dimer, since unprocessed high molecular weight forms of activin A display no biological activity (13,15,16).
Thus far, TGF-␤2, TGF-␤3, and BMP-7 (also called osteogenic protein-1 (OP-1)) have been crystallized, and the threedimensional structures of the mature, dimeric molecules have been elucidated (17)(18)(19). These proteins share a common threedimensional polypeptide folding pattern, although their amino acid sequence identity is limited to 36% (BMP-7 compared with TGF-␤2). Hence, it is likely that this structure is the prototype for the whole family and might be extrapolated to activins as well. The common fold of the monomer is defined by seven cysteines that are conserved throughout the family. Six of these form intrachain disulfide bonds and make up the cystine knot, while the seventh cysteine forms an interchain disulfide bond that stabilizes the dimer (17)(18)(19). By analogy with a left hand, the monomeric structure consists of the N-terminal thumb region, two antiparallel pairs of ␤-strands that build up four fingers, two loops that connect the fingers (loop 1 connects finger 1 and 2; loop 2 connects finger 3 and 4), and a long ␣-helix at the heel of the hand (see Fig. 1). This prototype structure defines four solvent-accessible, flexible and divergent regions, which may contain putative receptor-binding sites, i.e. the N terminus, loop 1, loop 2, and the C-terminal end of the long ␣-helix (17)(18)(19).
Amino acids important for biological activity have been defined by limited structure-function analysis of TGF-␤ members and by molecular characterization of naturally occurring mutations that cause drastic phenotypes in different organisms. Mutation analysis revealed that the nine cysteines, including the seven conserved cysteines in the family, of mature activin A are essential for either the biosynthesis or the (full) biological activity of activin A (20), and that a phenylalanine to isoleucine substitution at position 21 of activin B creates a dominantinterfering protein (16). In mature TGF-␤1, the C-terminal portion (amino acids 83-112) has recently been defined as necessary for high affinity binding to the TGF-␤ receptor type II (T␤RII) (21). In addition, frameshift and/or point mutations (i.e. replacement of the first conserved cysteine by tyrosine) in hCDMP-1 (human cartilage-derived morphogenetic protein), and growth and differentiation factor (GDF)-8, have been identified in human chondrodysplasia and double muscling in cattle, respectively (22)(23)(24).
However, no specific receptor binding determinants are known for any TGF-␤ member. Detailed mutagenesis studies of TGF-␤ family members would provide insight into how such mutations affect their biological activities, and this may facilitate the development of therapeutic agents that can be used in TGF-␤-related diseases. In order to identify amino acids important for receptor binding and biological activity, we started structure-function analysis of activin A by introducing single amino acid substitutions in the mature domain, in regions that are thought to be involved in receptor interaction (18). In this way, we identified two amino acids in activin A which are important for its biological activity and its interaction with the type II receptor: Asp-27 and Lys-102, located in loop 1 and 2, respectively.

EXPERIMENTAL PROCEDURES
Mutagenesis-Oligonucleotide-directed mutagenesis was performed using plasmid PTZ18R, which contains a mouse activin A cDNA cloned in the sense orientation with respect to the T7 promoter, and the Muta-Gene phagemid in vitro mutagenesis kit (Bio-Rad). Mutations were introduced using the single-stranded sense oligonucleotides listed in Table I. Mutant TGF-␤2L1 was constructed using two complementary oligonucleotides representing the DNA sequence of loop 1 of TGF-␤2: 5Ј-C-GATTTCAAGAGAGATCTAGGGTGGAAATGGATACACGAACCCT-3Ј and 5Ј-CCGGAGGGTTCGTGTATCCATTTCCACCCTAGATCTCTCTT-GAAAT-3Ј This sequence was cloned into an activin mutant construct in which a ClaI site and a MroI site had been introduced upstream and downstream, respectively, of the loop 1 (Val-18 to Pro-32) sequence.
Mutant activin V18I/S19D/N26I/D27G was generated by partial double annealing via the nine 3Ј-nucleotides of 5Ј-GAAACAGTTCTTTATC-GATTTCAAGGACATTGGCTGGATTGGCTGG-3Ј. This generated two extra mutations (N26I and D27G) in addition to those generated by oligonucleotide V18I/S19D. All mutations, listed in Fig. 1B, were confirmed by DNA sequencing. Vaccinia Virus T7 Expression System-This expression system has been described previously (13), but it was applied in a slightly modified manner. Subconfluent HeLa or PK15 cells (10-cm 2 dishes) were infected with a recombinant vaccinia virus expressing phage T7 RNA polymerase (multiplicity of infection: 5) for 1 h at 24°C. These cells were then transfected with T7 promoter-containing plasmids encoding the wild type and mutant activins A, using DOTAP (Boehringer Mannheim). For the heterodimerization assay, cells were cotransfected with T7 plasmids encoding zebrafish activin B (a gift from F. Rosa, U368 INSERM, Ecole Normale Supérieure, Paris, France). Cells were incubated with this DNA/DOTAP mix for 6 h at 37°C in an atmosphere containing 5% CO 2 . Cells were then washed with methionine-free minimum essential medium before starvation in this medium for 1 h at 37°C. The cells were next pulse-labeled for 1 h by addition of 1 ml of the same medium containing 50 Ci of [ 35 S]methionine and [ 35 S]cysteine (ICN). The cells were chased by addition of 1 ml of Dulbecco's modified Eagle's medium supplemented with 20 g/ml bovine serum albumin and a 10-fold higher concentration of cold methionine than is normally present in this medium. After 14 h at 37°C, the medium was collected and centrifuged, and the supernatant was frozen. Samples were prepared for electrophoresis after precipitation of the proteins with trichloroacetic acid, as described previously (13).
FSH Assay in Rat Pituitary Cells-The follicle-stimulating hormone (FSH) assay was performed as described (25). Briefly, primary rat pituitary cells were cultured for 2 days in serum-free medium (as specified in Ref. 25) containing dilutions of the activin A mutant proteins. The medium was then collected, and the FSH concentration was determined by radioimmunoassay (RIA). Each mutant activin was

5Ј-CATGATTGTGGAGGAGTGTAAGTGCTCCTGAATTCGCCAGGTCCC-3Ј
added to three wells, and the RIA for FSH was performed in duplicate using the FSH-RIA kit (NIDDKD, National Institutes of Health, Rockville, MD) according to Denef et al. (26).
Mesoderm Induction Assays in Xenopus-Xenopus embryos were obtained by in vitro fertilization (27). They were maintained in 10% Normal Amphibian Medium (28) and staged according to Nieuwkoop and Faber (29). Animal pole regions were dissected from mid-blastula (stage 8) embryos (30) and cultured in 75% Normal Amphibian Medium containing 0.1% (w/v) bovine serum albumin and wild type or mutant activin (2.5 ng/ml). A preliminary assessment of mesoderm induction was based on the elongation of the animal caps. Animal pole regions were then frozen on dry ice, and expression of the mesoderm-specific gene Brachyury (Xbra) (31) was assayed by RNase protection analysis as described by Jones et al. (32).
Radiolabeling of Activins and Follistatin-Wild type and mutant activins, and follistatin were iodinated using a modified chloramine-T method (33). Two g of protein (in 10 l of 30% acetonitrile, 0.1% trifluoroacetic acid) were diluted with 10 l of 600 mM sodium phosphate (pH 7.5) and 5 l of Na 125 I (0.25 mCi; Amersham Pharmacia Biotech) and 5 l of phosphate-buffered saline (137 mM NaCl, 2.7 mM KCl, 6.5 mM Na 2 HPO 4 and 1.5 mM KH 2 PO 4 ). To initiate the radioiodination, 10 l of chloramine-T (100 g/ml in 50 mM sodium phosphate (pH 7.5); Sigma) was added. After 2 min, the iodination was stopped by addition of 20 l of 50 mM N-acetyl-L-tyrosine (Sigma), 200 l of 60 mM sodium iodide, and 200 l of 10 M ultrapure urea (Life Technologies). Subsequently, the reaction mixture was passed over a Sephadex G-25 column (Amersham Pharmacia Biotech), which was equilibrated and eluted with phosphate-buffered saline containing 0.1% (w/v) hemoglobin (Sigma). Peak fractions, with specific activities of 30 -100 Ci/g of protein, were routinely obtained, pooled, and stored at Ϫ80°C.
Activin/Follistatin Cross-link-Radioiodinated follistatin288 (FS288) was cross-linked to cold wild type and mutant activins using bis-sulfosuccinimidyl suberate (BS 3 ; Pierce) (33). Approximately 2 ng of iodinated FS288 (5 l) was incubated with 500 l of activin-containing conditioned medium prepared as described above. After 2 h of incubation at 4°C on a rotary shaker, 125 l of 5 mM BS 3 in HEPES-buffered saline (150 mM NaCl and 20 mM HEPES; Life Technologies, Inc.) was added and the reaction was incubated for 1 h at 4°C. Activin/follistatin complexes were purified using wheat germ agglutinin-agarose (Sigma) beads (33). They were separated by SDS-PAGE under reducing conditions and visualized by autoradiography.
Receptor Binding Studies-PK15 cells (28-cm 2 dishes) were transfected with different combinations of activin receptors using the vaccinia virus-T7 system as described above. On the second day, the cells were washed with ice-cold binding medium (HEPES-buffered Dulbecco's modified Eagle's medium (pH 7.5) containing 0.2% (w/v) bovine serum albumin) for 10 min. Cells were incubated with 150 pM labeled activin A in 1.5 ml of binding medium for 2 h at 4°C. For competition studies, cells were incubated with a constant amount (150 pM) of 125 Ilabeled wild type activin and (simultaneously added) different amounts of cold wild type or mutant activins. Then, iodinated activin was removed by gently and repeatedly washing the cells with ice-cold HEPESbuffered saline containing 0.9 mM CaCl 2 . Activin was cross-linked by incubation in 1.5 ml of HEPES-buffered saline containing 1 mM BS 3 for 30 min at 4°C. The reaction was then quenched for 5 min at 4°C by addition of 150 l of 10ϫ quench solution (10ϫ: 10 mM Tris (pH 7.5), 2 mM EDTA, and 200 mM glycine). The cells were scraped from the plates in 1 ml of detachment buffer (10 mM Tris (pH 7.4), 1 mM EDTA, 10% (v/v) glycerol, 0.5 g of aprotinin/ml, 0.5 g of leupeptin/ml, and 0.3 mM phenylmethylsulfonyl fluoride), and collected by centrifugation (5 min at 4°C). The pellet was then dissolved in 50 l of solubilization buffer (10 mM Tris (pH 7.4), 1 mM EDTA, 125 mM NaCl, 1% (v/v) Triton X-100, 0.5 g of aprotinin/ml, 0.5 g of leupeptin/ml, and 0.3 mM phenylmethylsulfonyl fluoride), followed by incubation for 40 min on ice. Proteins were separated by SDS-PAGE and visualized by autoradiography.
Large Scale Production and Purification of Activins-The large scale production of wild type and mutant activins was performed using a baculovirus expression system. The mutant cDNA was inserted in the baculotransfer vector pVL1393 under transcriptional control of the baculoviral polyhedrin promotor. Recombinant baculovirus was generated by homologous recombination in Spodoptera frugiperda cells (Sf9) cotransfected with the recombinant transfer construct and Baculogold TM virus AcNPV DNA (PharMingen). Recombinant virus was plaque-purified and amplified to high titer stock for production. Activin was purified from conditioned medium of recombinant baculovirus infected Sf9 cells, harvested 72 h after infection.
Purification of wild type and mutant activins was performed by use of an optimized four-step purification protocol in which the conditioned medium is diafiltrated and concentrated in the presence of 6 M urea and loaded onto an anion exchange column (Fractogel-(EMD)-TMAE, Merck). The flow-through is then loaded on a Protein Pack Sulfonyl (Millipore) cation exchange column. The 150 mM NaCl fraction is then adjusted to 10% acetonitrile, 0.1% trifluoroacetic acid (v/v) and separated by RPC-4 (Fractogel-butyl) reversed phase chromatography. Mutant activins are recovered in the 30 -34% acetonitrile fraction (34) and further purified on a RPC-8 (Brownlee octyl) column run as a polishing step. Quantification of these pure activins was obtained through amino acid composition analysis.

RESULTS
Mutagenesis-Some mutations in structurally important regions of TGF-␤ family members have been reported to lead to improper biosynthesis of these ligands (15,16,20,35,36). In our study, this was also the case when severe changes in activin A were introduced. For example, the substitution of loop 1 of activin A by the equivalent region of TGF-␤2 led to undetectable protein expression levels in the vaccinia virus-T7 system, both in the secreted and in the intracellular fraction (data not shown). It is likely that intracellular degradation occurred, as has been suggested for most cysteine mutants of activin A and TGF-␤1 (20,36). To avoid synthesis and intracellular trafficking problems due to structural changes, we anticipated that the majority of mutant activins used in this study should be generated by single amino acid substitution.
Four solvent-accessible regions can be deduced from the three-dimensional structure of TGF-␤2 and BMP-7 (17)(18)(19): the N terminus, loop 1, loop 2, and the C-terminal segment of the long ␣-helix (18; see also Fig. 1). These regions are the most flexible structures in the dimer, and their sequences are divergent throughout the family, which marks them as good candidates for receptor interaction. Most of the mutations introduced are single alanine substitutions at charged residues in these domains (Fig. 1A). A large panel of 39 activin A mutants was constructed by oligonucleotide-directed mutagenesis (Fig.  1B).
Synthesis and Secretion of the Mutant Polypeptides-Synthesis of this large panel of activin polypeptides was first analyzed in HeLa cells using the T7 vaccinia virus-based expression system. As suggested previously, these cells have sufficient levels of endogenous furin to support correct and efficient processing of the activin A precursor (13). Synthesized proteins were visualized by metabolic labeling followed by SDS-PAGE. We assessed both the maturation of activin A mutants to a 25-kDa dimer as well as their capacity to heterodimerize with zebrafish activin B; a secreted activin AB heterodimer can be resolved in SDS-PAGE because activin A homodimers have a slower migration than activin B homodimers. Nearly all activin A mutant polypeptides were processed like the wild type precursor and they heterodimerized efficiently (and predominantly) with activin B, as observed previously (Ref. 37 and data not shown; only activin dimers of D27K, K102A, K102E and K102R, respectively, are shown in Fig. 2). This indicates that the overall structure of the precursor polypeptides and their intracellular folding and dimerization in the rough endoplasmic reticulum are not altered. As well as analyzing their ability to dimerize, the ability of the mutants to bind follistatin (FS), an antagonistic binding protein of activin, was tested by crosslinking. All mutant activins tested (including K102A and K102E), formed complexes with FS288 like wild type activin (shown for K102A and K102E in Fig. 3). This again suggests that their overall structure is not dramatically altered, if at all.
A modified version of the vaccinia virus-T7 expression system was used to produce all mutant activins (see "Experimental Procedures"). De novo synthesized proteins were pulselabeled early in infection and then chased with excess cold amino acids. Assuming that their production rates do not dif-fer, this allowed us to normalize the different mutant activin concentrations in the conditioned medium relative to wild type by detecting and quantifying the signal of the 25-kDa dimer using PhosphorImager analysis. The concentration of activins (Ϯ 30 ng/ml) was high enough in the conditioned medium for use in different sensitive bioassays.

Biological Activities of vv-T7 Produced Mutant Activins-
The crude conditioned media described above were used for a preliminary characterization of the bio-activities of the different mutant activins. The first bioassay was based on the stimulatory effect of activin on the production of FSH by pituitary cells (38). Wild type activin A stimulated FSH production in a dose-dependent manner with a maximum stimulation of 2-2.5fold compared with non-stimulated cells at concentrations of 2.5-5 ng of activin/ml. Purified activin A exhibited a halfmaximal stimulation (ED 50 ) of FSH production at a concentration of 0.4 ng/ml in our modified FSH assay (25), which is more sensitive than previously used assays (38). Most of the unpurified mutant activin preparations behaved like wild type activin A in this assay (data not shown). However, mutant K102E was not significantly active, while K102A consistently displayed a lower activity than wild type activin A. Loss of bioactivity of K102E was restored to wild type levels when this lysine (Lys-102) was replaced with another positively charged residue (mutant K102R). Mutant activin A bearing a D27K substitution appeared to be more active in this assay (data not shown, but see below).
Activin causes animal cap explants of early Xenopus embryos to undergo a rapid and dramatic morphogenetic response (39), allowing a provisional assessment of the mesoderm-inducing activities of different activin mutants. Mesoderm-inducing activities of the crude vv-T7 produced mutant activins were first assessed by observing the elongation of Xenopus animal caps cultured in activin-containing media. The animal cap assay is very sensitive, since it has an ED 50 (50% of the animal caps show elongation) of 0.2 ng of activin/ml. Nearly all activin variants induced weak to strong elongation (like wild type activin) of animal caps, but mutant K102E showed no elongation (data not shown).
Based on these two biological assays, we selected four activin mutants for further analysis. Three mutants bear amino acid substitutions at Lys-102 (Ala, Glu, and Arg), and the positive charge at this position is apparently critical for biological activity (see also below). The D27K mutant was also selected because it displayed higher specific activity.
Biological Activities of Purified Mutant Activins-In order to analyze the selected mutant proteins in more detail, they were produced in large amounts using the baculovirus expression system. Four baculovirus recombinants were generated and used to infect insect cells (Spodoptera frugiperda (Sf9) cells). The secreted 25-kDa dimer was purified from 1.5-3 liters of conditioned medium. The mutant proteins were purified to homogeneity in four steps using a modification of a previously published purification protocol for activin A (34). This yielded pure activins, as judged by SDS-PAGE followed by silver stain- ing (data not shown). Quantification of these pure activins was obtained through amino acid analysis.
In order to confirm the data obtained with conditioned media, the mesoderm-inducing activities of the purified mutant activins were studied using the animal cap assay. At a concentration of 2.5 ng/ml, all mutant activins and wild type activin A caused clear elongation of the animal caps, except for mutant K102E (data not shown). The mesoderm-inducing activities of the activins were confirmed by studying expression of Xenopus Brachyury (Xbra), which is induced in an immediate-early fashion in amphibian embryos by activin. Both D27K and K102R induced expression of Xbra to levels similar to those induced by wild type ligand, whereas virtually no Xbra expression was detected in animal caps incubated with K102E activin (Fig. 4). Mutant K102A was less potent than wild type activin A in this assay.
The FSH release assay was also repeated with purified activins. Different dilutions of the mutant proteins were tested in order to generate a dose-response curve for each mutant activin. Mutant D27K stimulated FSH levels to 240% of the unstimulated control (100% level), whereas maximum stimulation with wild type activin A was 200% (Fig. 5). This indicates that the D27K mutant has a higher intrinsic biological activity in this assay, a conclusion consistent with the observation that the onset of stimulation occurred with a lower dose of D27K compared with wild type. The ED 50 values (i.e. the concentration of protein that results in half of the maximum stimulation) are 0.4 ng/ml for wild type versus 0.2 ng/ml for D27K, suggesting that D27K has a higher affinity for the receptor complex.
The K102A and K102E mutants were less efficient at FSH stimulation. Their maximal stimulation was 160% of the unstimulated control and this required very high concentrations of ligand, which are known for wild type activin to result in nonspecific effects in the assay (including luteinizing hormone stimulation; Ref. 13). Also, the onset of stimulation occurred at higher doses for K102A and K102E compared with wild type activin A. Overall, K102A and K102E were about 50-fold less potent than wild type activin in this assay. However, the K102R mutant had FSH stimulatory capacities comparable with wild type activin.
These data confirm those obtained with the unpurified mutant activins produced in HeLa cells with the vaccinia virus-T7 system. Mutants K102A and K102E had little or no biological activity, while mutant K102R was as active as wild type activin. The improved agonistic properties of mutant D27K became obvious in the FSH assay. Mutant D27K is active at a 4-fold lower concentration compared with wild type and this mutant activin also generated a higher level of FSH stimulation than wild type.
Receptor Binding of Purified Mutant Activins-In order to test the binding of mutant activins to type I and type II receptors, affinity cross-linking experiments were performed on cells overexpressing different (mouse) activin receptor combinations. These receptors (IIA, IIB, ALK-2, and ALK-4) were expressed in kidney (PK15) cells using the vaccinia virus-T7 expression system (40). Radiolabeled wild type activin, and mutants D27K and K102R, all bound to the activin receptors tested (type II receptors and type II-I receptor combinations, respectively), whereas no binding could be detected with the K102E mutant (Fig. 6). K102A binds to ActRIIA/ALK-4 and ActRIIB/ALK-4, but interacts only very weakly with type II/ ALK-2 combinations. The fact that activins bind better to complexes containing ALK-4 than those containing ALK-2 suggests that ALK-4 is the type I receptor that responds to activin in vivo, a conclusion consistent with previous observations on the binding of activin to primary pituitary cells (25). The lack of detectable cross-linked complexes with the K102E mutant was not due to the method of cross-linking itself. BS 3 uses lysine residues to cross-link, and Lys-102 in activin A is not essential for cross-linking by BS 3 , since mutant K102R could still be cross-linked to receptor complexes in a manner similar to wild type.
In order to compare the binding affinities of the different mutant proteins with that of wild type activin, competition cross-linking experiments were performed on PK15 cells transfected with ActRIIA and ALK-4. These transfected cells were affinity-labeled using a constant amount of 125 I-activin A (150 pM) in the presence of increasing concentrations (5-, 10-, and 20-fold excess) of cold mutant or wild type activin. Wild type activin A and mutant D27K competed for binding to the Act-RIIA/ALK-4 receptor complex efficiently, whereas mutants K102A and K102E did not (Fig. 7). Interestingly, K102R competed efficiently for binding to ALK-4 in the ActRIIA/ALK-4 receptor combination, but very poorly for binding to ActRIIA alone (data not shown).
Quantification (using a PhosphorImager) indicates that D27K (K D 350 pM) has a 2-fold higher affinity than wild type FIG. 4. Analysis of Xenopus Brachyury (Xbra) expression induced by different activins. Animal caps were cultured in 2.5 ng/ml mutant or wild type activin, and frozen for analysis at the equivalent of stage 11. RNA was isolated and hybridized with radioactive probes specific for Xbra and ornithine decarboxylase (ODC) as a loading control.

FIG. 5. FSH release from pituitary cells by wild type activin A and mutants.
Rat pituitary cells were seeded in 24-well plates, and the medium was changed to fresh medium containing purified wild type or mutant activins. The cells were incubated for 2 days, and FSH released into the medium was measured by RIA. Results are represented as the mean values with standard deviations. Shown is one representative experiment with triplicate incubations (with each dilution) of the cells. q, wild type activin; E, D27K; ϫ, K102E; f, K102A; Ⅺ, K102R. activin (K D 600 pM, a figure confirmed by Scatchard analysis; data not shown) for binding to ActRIIA/ALK-4 receptor complexes. The affinity of K102A for this receptor complex is lower than that of wild type activin; although this mutant is able to bind to ActRIIA/ALK-4 (Fig. 6), it is not able to compete with wild type activin, at least at the concentrations of competitor tested. The 2-fold stronger binding of D27K for the ActRIIA/ ALK-4 receptor complex is consistent with the results from the FSH assay, where mutant D27K had a 2-fold lower ED 50 than wild type activin. The relative binding affinities of mutants K102A, K102E, and K102R for the ActRIIA-ALK-4 receptor complex are also in agreement with their activities in the FSH assay. DISCUSSION In the present study, we have identified two individual amino acids in activin A that are important for biological activity as assessed by their ability to stimulate FSH release by gonadotropic pituitary cells and to induce mesoderm in Xenopus animal cap assays: residue K102, located in loop 2 of each subunit of the dimer, and D27, in loop1. Substitution of the positively-charged amino acid (K102) with a neutral (A) or negatively charged (E) residue greatly reduces activin function, whereas mutant K102R has no effect on activin bio-activity, suggesting that a positive charge at position 102 is crucial for activity in these assays. Substitution of D27 with K results in a mutant protein with a 2-fold higher specific activity than wild type activin. This study adds important new results to previously obtained data concerning the structure and function of activins, which have demonstrated that phenylalanine 21 of zebrafish activin B and 2 cysteine residues (Cys-4 and Cys-12 in the mature protein) of human activin A are important for biological activity (16,20). However, the precise level at which the phenylalanine 21 mutant affects the biological activity of activin has not been determined.
Affinity cross-linking experiments indicate that Lys-102 is crucial for interaction with the type II receptor and, as predicted by the current model of receptor activation, also for binding to a type II (A/B)-type I (ALK-2/4) receptor complex while mutant D27K can be cross-linked to the ActRIIA/ALK-4 receptor complex with a 2-fold higher efficiency than wild type activin. Since D27K displays a higher biological activity, we do not believe that this more efficient cross-linking occurs because of the introduction of an additional lysine, but that it rather reflects a higher binding affinity of D27K for the receptor combination tested here. The latter could be the result of a higher rate of association and/or a lower rate of dissociation.
Lys-102 is positioned in a region (loop 2) of the ligand previously shown to be important for high affinity binding of TGF-␤1 to T␤RII (21). Other approaches, using antagonistic peptides that block binding of TGF-␤ to its receptors, have defined the W/RXXD motif of the N-terminal segment of the long ␣-helix of TGF-␤s as a primary determinant for receptor binding (41). However, the W/RXXD motif does not seem to be involved specifically in type II receptor binding, as a peptide containing this motif also blocks binding of TGF-␤ to the high molecular weight type III and type V receptors (42,43). Moreover, such peptide studies usually need high concentrations of peptides to interfere with the function of the wild type molecule, which can lead to nonspecific effects, as reported previously (44).
All mutant activins, including K102A and K102E, can be cross-linked to the activin-binding and inhibitory protein follistatin (FS288). Many conclusions can be drawn from this observation. First of all, together with the fact that all activin A variants can form mature homodimers and heterodimers with activin B, this indicates that their overall three-dimensional structure is not dramatically altered. However, at this stage it is not clear whether Lys-102 or Asp-27 introduces local structural changes in the receptor binding pocket of the ligand or whether these mutations are directly involved in interaction with the receptor. This is difficult to assess, since conformationspecific monoclonal antibodies for ligands of the TGF-␤ family are not available. In addition, it is significant that Lys-102 and Asp-27 are located in the most flexible and solvent-accessible loop regions of the ligand, which favor the hypothesis that they interact directly with the receptors. Strikingly, Lys-102 and Asp-27 are conserved in BMP-7 and GDF-5, which have been Binding of iodinated wild type activin to transfected PK15 cells and competition with cold wild type or mutant activins. ActRIIA was cotransfected with ALK-4 using the vaccinia virus-T7 system. The cells were affinity-labeled using 125 I-wild type activin in the presence of different concentrations of cold competitor (wild type or mutant activin A), followed by cross-linking with BS 3 . Samples were analyzed by SDS-PAGE (8% gels) followed by autoradiography (data not shown) and analysis using a PhosphorImager. Competition of 125 I-wild type binding by cold wild type or mutant activins was quantitated by the amounts of radioactivity in the ActRII complex (II) using a PhosphorImager. Experiments were repeated using different concentrations of cold activins (both wild type and mutant), and representative data are shown. Ϫ, no competitor added.
shown to bind to and signal through ActRIIA or ActRIIB containing receptor complexes (25,45). In contrast, TGF-␤s have, respectively, a lysine (Lys) and a glutamic acid (Glu) residue at these positions, and TGF-␤s do not bind ActRIIA or IIB. Both these observations with BMP-7 and GDF-5 (25,45) supported the notion that these amino acids (Asp-27 and Lys-102) are important for (type II) receptor recognition.
Second, since K102E does not bind to the activin type IIA and IIB receptors, this suggests that the receptor binding determinant of activin is (at least in part) distinct from the follistatin binding determinant. Consistent with this idea, we note that BMPs can also bind follistatin and that follistatin can form a trimeric complex with BMP and its receptor (46). A peptide approach has defined two contact sites in activin that are necessary for interaction with follistatin (47). These sites encompass amino acids 15-29 and 99 -116 in activin A. Although Lys-102 is localized in one of these regions, it is not necessary for binding to follistatin, suggesting that follistatin may act by masking this amino acid and thus prevents activin from binding to the activin type II receptors, as reported previously (33).
An ideal antagonistic ligand would be able to interfere with wild type ligand function by binding to its receptor(s) without activating the signal transduction cascade. In the TGF-␤ family, such an antagonist would bind with a normal or even higher affinity to a type II receptor, but not at all to a type I receptor. Although we performed an extensive mutagenesis study, such an antagonistic activin variant was not found. Possibly, a more drastic change is needed to interfere with binding to type I receptors or no strict separate binding determinants exist for binding to type II and type I receptors. A dominant-negative ligand, distinct from an antagonist, might interfere with wild type function in two ways: either by altering the affinity of the mutant/wild type heterodimer for its receptor(s) or by interfering with the processing of the wild type ligand. Co-translation of the wild type and dominant-negative ligands would thereby deplete the endogenous pool of activin. Such dominant-negative variants of activin B, BMP-7 and BMP-4, have been described, and in these the consensus cleavage site for the protein convertase is modified into a noncleavable sequence (16,35). In addition, certain cysteine mutants of different TGF-␤ members have been identified as dominant negative; however, such mutations may result in nonspecific inhibition of ligand secretion (20,23). The K102E mutant is likely to act as a dominant negative construct of the first kind, because homodimers and heterodimers with activin B are still secreted but, at least in the case of the K102E homodimer, cannot interact with the type II receptor. Future studies, for example using RNA injection experiments in Xenopus embryos, can investigate this question in more detail. Future work should also try to extrapolate our data to other ligands of the TGF-␤ family. In this way, our work will contribute to the design of agonistic, dominant negative, and antagonistic variants of TGF-␤ members. These variants might help in the development of new therapeutic agents, e.g. for use in bone repair, wound healing, fibrosis, immune modulation, and acute kidney insufficiency.