Primary structure requirements for Xenopus nodal-related 3 and a comparison with regions required by Xenopus nodal-related 2.

Transforming growth factor-beta superfamily members play important roles in the early development of animals. Activin and the Xenopus nodal related proteins 1, 2, and 4 induce muscle actin from Xenopus ectodermal explants, whereas the bone morphogenetic proteins 4 and 7 induce ectoderm to differentiate as epidermis. Bone morphogenetic proteins are antagonized by soluble binding proteins such as noggin and chordin, which leads to expression of neural cell adhesion molecule in animal caps. The transforming growth factor-beta superfamily member Xenopus nodal-related 3 also induces the neural cell adhesion molecule through inhibition of bone morphogenetic proteins. Therefore, whereas Xenopus nodal-related 2 and 3 share a high amount of sequence homology, they lead to very different cell fates. This study investigates the functional domains that distinguish the activities of these two factors. It was found that mutually exclusive regions of nodal-related 2 and 3 were required for activity. The central region of the mature domain is required for nodal-related 2 to induce muscle actin, whereas the N- and C-terminal ends of the mature domain are required for nodal-related 3 to induce neural cell adhesion molecule. These results help to define the minimal domains required for the unique activities of these factors.

The transforming growth factor-␤ (TGF-␤) 1 superfamily is composed of a myriad of related secreted proteins that are important regulators of development and physiology in both vertebrates and invertebrates (1). Most TGF-␤s are synthesized as proproteins that are biologically inactive until proteolytically processed at R-X-(K/R)-R and R-X-X-R consensus sequences by subtilisin-like proprotein convertases (2)(3)(4). Active TGF-␤ proteins consist of two 12-16-kDa peptides that show varying ability to function as homo-and heterodimers. Superfamily members have seven highly conserved cysteines in the C-terminal mature domain that form intra-and interchain disulfide bonds. Within the monomer, disulfide bond pairs are formed between the first and fifth, second and sixth, and third and seventh cysteines. The fourth conserved cysteine makes an interchain bond between dimer subunits. Another conserved feature of the superfamily is a glycine residue between the second and third cysteines, within the consensus sequence CXGXC. It is thought that steric hindrance would require a glycine (Fig. 1, indicated by an asterisk) in this position for proper folding, because two disulfide bonds form a closed ring on either side of this residue, which prevents its substitution (5,6).
Three-dimensional structural studies of TGF-␤s 1, 2, and 3, as well as bone morphogenetic proteins (BMPs) 2 and 7 predict a conserved "cysteine knot" structure that has been described as a "left hand" (Fig. 1) (5)(6)(7)(8)(9)(10). The "heel" of the hand structure is an ␣-helix formed by the amino acids between the third and fourth cysteines. Extending out from this are two long loops consisting of ␤-sheets that form "fingers." The N-terminal amino acid region of the mature domain extending from the cleavage site to the first conserved cysteine represents the "thumb" (also referred to as the proknot sequence, (10)). Members of the TGF-␤ subfamily (TGF-␤s 1-5) and activin have two additional conserved cysteines in this region, which form a disulfide bond anchoring a short ␣-helix to the first ␤-sheet of finger 1. Because most other superfamily members share only the seven conserved cysteines, they lack this additional disulfide bond. In BMP2 and BMP7, the thumb region is disordered and cannot be resolved in electron density maps, so the structure of this region in these superfamily members is unknown.
Members of the nodal subfamily of the TGF-␤s play important roles in vertebrate mesoderm induction and patterning (11)(12)(13). Whereas the mouse and chick appear to have single members of the nodal gene family, duplications have led to at least four family members in Xenopus (Xnr1-4) and two in zebrafish (squint and cyclops) (14 -16). Comparison of predicted amino acid sequences indicates that Xnr1, Xnr2, and Xnr3 are more closely related to each other than to Xnr4. Most notably, Xnr1, Xnr2, and Xnr3 share the unique feature of having the sequence CXXC between the fourth and fifth conserved cysteines. This sequence is also found in chick nodal and zebrafish squint. Xnr4, mouse nodal, and zebrafish cyclops, as well as the majority of other TGF-␤ superfamily members, have the sequence CC for these two cysteines. Xenopus nodal-related 3 (Xnr3) is unique among the nodal subfamily in having several primary structure features that diverge from the TGF-␤ superfamily consensus (17). First, Xnr3 is missing the last of the seven conserved cysteines. Second, whereas all other superfamily members have a glycine located between the second and third cysteines, Xnr3 has a serine in this position. Perhaps this substitution is allowed because Xnr3 lacks one of the two disulfide bonds that constrains this residue to a glycine in other superfamily members. Together these observations suggest that Xnr3 does not form the characteristic knot structure (Fig. 1).
Xnr3 also has biological activity in developing Xenopus embryos that differ markedly from the other nodal-related factors.
In Xenopus animal cap induction assays, Xnr1, 2, and 4 induce the mesodermal markers brachyury (when assayed at early gastrula stage) and muscle actin (MA) at tailbud stage (11,18). All three also have the ability to rescue mesoderm in VegTdepleted embryos (19). The related factors in the mouse and zebrafish (nodal, squint, and cyclops) induce mesodermal markers in the Xenopus assay (11,16,20). However, Xnr3 blocks the activities of the TGF-␤ family members BMP4 and activin, does not induce brachyury or MA, and instead induces the neural cell adhesion molecule (NCAM). Based on these and other findings, we have speculated that Xnr3 may function as a BMP4 and activin receptor antagonist (21). Although there are several possible mechanisms by which Xnr3 could inhibit these TGF-␤ superfamily members, observations that Xnr3 can inhibit soluble activin protein excludes a mechanism by which dysfunctional dimers were formed. Also, the fact that Xnr3 does not inhibit a constitutively active receptor supports the receptor antagonist model (21). In addition to Xnr3, several other TGF-␤s have been postulated to be antagonists, including lefty, inhibin, and antivin (22)(23)(24)(25). A comparison of the primary structure of these putative antagonists does not readily suggest common features to account for their activities. Thus, it is possible that the putative TGF-␤ receptor antagonists evolved independently and may use different strategies for binding, but not activating, receptors.
The nodal-related factors in Xenopus present a unique model system for studying structural features that are responsible for divergent activities of closely related TGF-␤ superfamily members. Among the nodal-related genes, Xnr3 is most closely related to Xnr2. Using a Xenopus animal cap assay and Northern blotting, the NCAM-inducing activity of Xnr3 and the MAinducing activity of Xnr2 are easily distinguished. We have used this assay to characterize a number of chimeras and mutations of these two factors to determine which regions are required for specifying their divergent activities. The results show that the regions needed for NCAM and MA induction are different and that these regions are different from those found to be important for TGF-␤ activity.
For making the C terminus of Xnr2 resemble Xnr3 in chimera 8, a pair of complementary oligonucleotides was used to fill in between BamHI and Eco57I sites, 5Ј-TTGTGGATGAGTGTGGATTCAAGGAC-ATGTAAG-3Ј and 5Ј-GATCCTTACATGTCCTTGAATCCACACTCATC-CACAATC-3Ј.
For making chimera 11 (the mutation changing amino acids ENA to FKP), a three part ligation was done using the following: 1) a ϳ570-base pair fragment produced using polymerase chain reaction from construct 9 with the T7 promoter primer and the mutagenic oligonucleotide 5Ј-TATCCTTTAAATGAAACCGAGAATGCAACGAACCATGCC-3Ј; 2) a pair of complementary oligonucleotides synthesized to fill in between DraI and NsiI sites, 5Ј-TATAGGTGTGAGGGAGCCTGTCCTATTCCT-TT-3Ј and 5Ј-AAAGGAATAGGACAGGCTCCCTCACACCTATAGGCA-3Ј; and 3) a construct 9 vector linearized with NsiI and ApaI.
The presence of the desired mutations was confirmed by dideoxy sequencing using Sequenase (US Biochem) or Big Dye Terminator cycle sequencing (Applied Biosystems). In cases where a chimera was made by ligating fragments, the regions where two pieces were joined within the coding region of the gene was also sequenced to verify the ligation had maintained the frame (Table I).
RNA Synthesis and Injection-Capped RNAs were synthesized from linearized plasmid templates using the mMessage mMachine (Ambion). Transcripts were checked by formaldehyde-containing agarose gel electrophoresis. Xnr3 template was linearized as described previously (17). The Xnr2 template consisted of a 1.4-kilobase cDNA insert (clone UDL3) in pBluescript SK-(Stratagene) (11). The Xnr2 plasmid was FIG. 1. A, alignment of the mature domains from the five TGF-␤ superfamily members that have been crystallized (TGF-␤s 1, 2, and 3 and BMPs 2 and 7), Xnr2, and Xnr3. Sequences that comprise specific features of the known structures are indicated, and the seven highly conserved cysteines are numbered. B, ribbon diagram of BMP7 illustrating the features used to describe the structure as a left hand. The numbers correspond to the numbered cysteines in A above and the intrachain disulfide bonds are shown. Cysteine four joins the dimer subunits, and the conserved glycine is indicated by an asterisk. It should be noted that the thumb domain is only two residues long because the structure for the N-terminal residues was not determined (14). linearized by XhoI digestion and RNA-transcribed with T3 RNA polymerase. Injection of 1 ng of RNA was at one cell stage 1, and animal cap explants were isolated at stages 8 -9.
RNA Extraction and Analysis-Total RNA was isolated from animal cap explants and embryos using Trizol Reagent (Life Technologies, Inc.). For Northern analysis, 20 caps were used/sample and electrophoresed on formaldehyde-containing agarose gels (29). Gels were transferred to Hybond-N nylon membrane (Amersham Pharmacia Biotech) by capillary action overnight and hybridized with QuikHyb (Stratagene). Random-primed 32 P-labeled probes were prepared using the Prime-a-gene system (Promega) with isolated fragments from NCAM (30), elongation factor 1 ␣ (31), and MA (also referred to as cardiac actin) (32).
Western Blotting-Xenopus embryos were injected with 2.5 ng of RNA, and 20 animal caps were isolated and grown to stages 20 -25. Animal caps were homogenized in reducing Laemmli sample buffer (33) and run on a 12% polyacrylamide gel at 175 V for an hour. Gels were transferred to Hybond-ECL nitrocellulose (Amersham Pharmacia Biotech) for 1 h at 40 V and blocked for 90 min in TBS with 3% dried milk (filtered through Whatman 4 paper) and 0.1% Tween (TBSM-T). Anti-Xnr3 antibody was diluted 1:2500 and incubated overnight. The blots were then washed 4 times for 10 min in TBSM-T, incubated with goat anti-rabbit horseradish peroxidase secondary diluted 1:5000 in TBSM-T for 40 min, and washed 6 times for 10 min in TBS-T with a final 10 min wash in TBS without detergent. Detection was done with SuperSignal (Pierce).
Antibodies-The Xnr3 antibody was made in rabbit against the synthetic peptides HVSTVPPKPIEEIKPEC and CHVSTVPPKPIEEIKPE, which correspond to the amino acids at the N terminus of the putative mature domain (the thumb domain). These peptides were linked to keyhole limpet hemocyanin and were used to immunize two rabbits. Serum from both rabbits was tested for binding of antibody to antigen using an alkaline phophatase-linked secondary and detection with pnitrophenyl phosphate. Antibody was purified from serum of the rabbit producing the higher titer using the antigenic peptides coupled to agarose (Pierce SulfoLink Kit).

Xnr3 Requires N-terminal Domain of Predicted Mature
Protein-Epitope tagging is a commonly used technique in the study of the structure and function of TGF-␤ superfamily members. Epitope tags have been successfully inserted into the thumb regions of many TGF-␤ superfamily members, including dorsalin, Xnr1, Vg1, and activin (34,35). 2 However, construction of similarly modified versions of Xnr3 yielded surprising results that provide clues to the structural requirements for Xnr3 activity. The addition of tags (Myc or hemagglutinin) into an equivalent position of the Xnr3 thumb blocked NCAMinducing activity (Fig. 2), even though Western blotting indicated that the epitope-tagged Xnr3 proteins were synthesized (data not shown). In contrast to other TGF-␤ superfamily members, this domain in Xnr3 is sensitive to alteration and therefore appears to play a role in protein function.
A second protein modification that is commonly used in the study of the TGF-␤s is the exchange of prodomains between two family members. This is done because various expression and in vivo assay systems process TGF-␤ superfamily members with varying efficiencies (36,37). These fusion proteins are usually made to take advantage of the optimized cleavage site of the heterologous prodomain. Prodomains can be successfully swapped between distantly related family members to yield active protein (35)(36)(37)(38). Fusion proteins have been successfully made joining the proregion of activin to the Xnr2 mature domain (38). However, the primary sequence of the mature domain of Xnr3 has a number of unusual features, and the prodomain plays a role in protein secretion, processing, and stability (39,40). Therefore, we sought to determine if there was a strict requirement of the Xnr3 prodomain for biological activity. A chimera was made with the activin prodomain and cleavage site-linked to the Xnr3 mature domain. This activin⅐Xnr3 fusion protein had no biological activity when assayed in the Xenopus animal cap induction assay for NCAM (data not shown). To test a prodomain from a more closely related superfamily member, the Xnr2 prodomain was fused to the Xnr3 mature domain. Because the results with epitope tagging suggested that some feature of the thumb region was important for activity, the fusion of the pro-and mature domains was made 18 amino acids N-terminal of the Xnr3 putative cleavage site. As with the activin fusion protein, no NCAM activity of the fusion protein could be detected (Fig. 2). Although negative data are by their nature not conclusive, our observations suggest that the ability to swap prodomains both within the nodal family and between more distantly related superfamily members does not extend to Xnr3. These and previous experiments with epitope tags indicate that certain features within the Xnr3 2 W. Smith, unpublished observations.  prodomain are required for biological activity. The most likely processing site for Xnr3 is the sequence RRLRR (amino acids 270 -274), although another less likely site is found in the flanking region (residues 179 -182). The processing site at residues 270 -274 was mutated to RKLSS (cmXnr3) to assess the importance of the site for activity. We hoped to use this strategy to make a dominant negative cleavage mutant Xnr3, as had been done successfully with TGF-␤1, activin, BMPs, and Xnr2 (41)(42)(43)(44)(45). Surprisingly, cmXnr3 is still functional as a neural inducer (Fig. 2). Western blotting of extracts from mRNA-injected animal caps using a polyclonal antibody directed at a mature domain peptide showed that the cleavage site mutation dramatically reduced the amount of processed prodomain, with the subsequent appearance of the unprocessed proprotein (Fig. 3). Even though it is possible that some processing could occur at amino acids 179 -183, it would be unlikely. Therefore, the unprocessed protein or a protein containing a significant amount of prodomain has biological activity.
We further examined the importance of the Xnr3 thumb region, including the putative processing site and surrounding residues. A chimera was made in which a 22-amino acid sequence was deleted (residues 270 -291), extending from the 5 residues in the putative cleavage site toward the C terminus to include an additional 17 amino acids. This deletion mutant was therefore lacking the region immediately N-terminal of the location where epitope tags were inserted and has seven amino acids before the first conserved cysteine. This chimera had no apparent activity when injected into whole Xenopus embryos (data not shown) or when assayed in explanted animal caps for NCAM induction (Fig. 2). Together these results point to structural requirements around the cleavage site for Xnr3 activity that are not found in Xnr2 or nodal.
Chimeras of Xnr3 and Xnr2 Reveal Different Domains for Activity-To further characterize regions of Xnr3 required for its unique NCAM-inducing activity, we made chimeric proteins fusing regions of Xnr3 to the closely related factor Xnr2, which does not have direct NCAM-inducing activity (46). First, we divided the mature region of Xnr2 and Xnr3 approximately into thirds by making a set of six chimeras (Fig. 4A). Because prior experiments indicated that Xnr3 likely required different regions for activity than TGF-␤s, this approach was used to narrow down the possible sequences that might be involved in the activity of Xnr3. The N-terminal one-third includes the thumb and "finger 1," the center region the ␣-helix heel, and the C-terminal portion contains "finger 2". To make the fusion chimeras, we added a number of restriction sites, most of which resulted in no change in amino acid sequence. However, a BanII site was added to Xnr3 at nucleotide 1127 to correspond with a BanII site found in Xnr2, which resulted in changing methionine 371 to a serine. This substitution alone resulted in no apparent change in Xnr3 activity (data not shown). The activity of the chimeras was examined by injecting in vitro transcribed mRNA into the animal pole of Xenopus embryos at the one-cell stage. For each chimera, 20 animal caps were dissected at late blastula stage and grown until stage 25. RNA isolated from the animal caps was analyzed by Northern blotting for presence of NCAM, MA, and elongation factor 1 ␣ transcript, which was used as a loading control (Fig. 4B).
Chimera 1 contained the prodomain and first third of the mature domain of Xnr2 fused to the C-terminal two-thirds of Xnr3. It was not an active MA inducer and had very little to no NCAM-inducing activity (Fig. 4). Chimera 2 is the reverse of chimera 1 and contains the prodomain and first 51 residues of Xnr3 with the C-terminal region of Xnr2. This chimera was consistently active in MA induction (n ϭ 6) indicating that the C-terminal 74 residues are sufficient for Xnr2-like MA-inducing activity. This chimera was judged to be negative for NCAM induction because it had little, or no, activity five of six times tested, and the low level of NCAM transcript most likely resulted from secondary induction (46).
In the complementary chimeras 3 and 4, domain swaps between Xnr2 and Xnr3 were restricted to the C-terminal onethird. All but the last 30 residues of chimera 3 were from Xnr2, FIG. 3. Xnr3 cleavage mutant (cmXnr3) is proteolytically processed at lower levels than wild-type. Extracts from Xenopus animal caps injected with mRNA encoding either wild-type or cmXnr3 were analyzed by Western blotting using an antibody against a synthetic peptide from the Xnr3 mature domain. Both the proprotein (uncleaved) and the processed, mature protein are indicated. The same result was obtained in each of three independent experiments.

FIG. 4. Inductive activities of Xnr3 and Xnr2 chimeras in Xenopus animal caps.
A, chimeras 1-6 divide the mature domain into approximately three equal parts. The data from a number of experiments are tabulated in the two columns on the right. The number in brackets following the plus or minus sign shows how many independent experiments resulted in that outcome. A plus sign indicates that the chimera had NCAM-or MA-inducing activity comparable to wild-type Xnr3 or Xnr2. A minus sign signifies no signal for NCAM or MA transcripts. The symbol ϩ/Ϫ is used in cases where activity was slightly above background, but much weaker than the activity of the controls. B, representative Northern blots. Each Northern blot is shown with four controls: whole embryo, uninjected animal caps, Xnr2 RNA-injected animal caps, and Xnr3 RNA-injected animal caps. Experimental lanes that are grouped together are from the same experiment and were assayed together. In some cases it was necessary to cut out intervening lanes that were not relevant to the figure. The MA probe cross-reacts to a lesser extent with cytoskeletal actin; the MA-specific signal is the lower band. EF1␣, elongation factor 1 ␣. and chimera 4 had Xnr3 sequence except for 28 amino acids at the C terminus. Neither of these chimeras had NCAM-or MA-inducing activity. The low level of MA seen in one case with chimera 4 was perhaps because of contamination. Thus, whereas the C-terminal 74 residues of Xnr2 were able to confer MA-inducing activity to the fusion protein, a smaller segment consisting of only the C-terminal 28 residues was inactive. When the middle third of the mature domain of Xnr2 was replaced with the corresponding domain of Xnr3 (chimera 5; Fig. 4A), MA-inducing activity was also lost and there was no detectable NCAM-inducing activity. However, a reciprocal fusion protein that replaced the middle third of Xnr3 with Xnr2 (chimera 6; Fig. 4A) retained near wild-type Xnr3 NCAMinducing activity in the absence of MA transcript. In summary, these experiments showed that Xnr2 and Xnr3 have different structural requirements for activity. In Xnr3, the prodomain and all or part of the first and last thirds of the mature protein are necessary for NCAM induction, whereas Xnr2 requires the C-terminal two-thirds of the mature domain to act as an MA inducer. Fig. 4 pointed to large domains required for activity. In further experiments we were able to narrow these regions to smaller domains. Because the results presented in Fig. 1 indicated that the residues surrounding the cleavage site of Xnr3 were required for activity, we examined this area in greater detail. To allow for domain swaps in the proximity of the thumb domain, an AvaI restriction site was added to Xnr2 in a location equivalent to one found in Xnr3 (Fig. 5). The addition of this restriction site required changing amino acids 303 and 304 of Xnr2 from TL to PG, which by itself did not alter Xnr2 activity (data not shown). We made chimera 7 to test if the region closest to the cleavage site was required, as suggested by the results summarized in Fig. 2. This chimera resembled chimera 6, except that a larger segment of Xnr2 extending past the first conserved cysteine replaced the middle region of Xnr3, and the resulting fusion protein had 22 amino acids of Xnr3 between the cleavage site and the Xnr3/Xnr2 junction (Fig. 5A). This chimera was still active for NCAM induction at wild-type levels (Fig. 5B).

Xnr3 Requires 22 Amino Acids at the N Terminus and Four Residues at the C Terminus of the Mature Domain for NCAMinducing Activity-The domain swaps detailed in
The results of experiments with chimeras 6 and 7 point to regions at the C-and N-terminal regions of the mature domain of Xnr3 that are important for activity. We made an additional chimera ( Fig. 5A; chimera 8) to examine the C-terminal region of Xnr3 more closely. The C-terminal four amino acids of Xnr3 were hypothesized to be important for activity because this region was most divergent between Xnr3 and the consensus sequence for TGF-␤ superfamily members. Chimera 8 is identical to chimera 7, except that chimera 8 has only the four C-terminal amino acids of Xnr3. This fusion protein combining the Xnr3 proregion, N-terminal 22 amino acids of the Xnr3 mature domain, and four C-terminal residues had NCAM-inducing activity comparable to that of wild-type Xnr3 (Fig. 5, A  and B).
Residues Important for Proper Folding of other TGF-␤ Family Members Appear Unimportant for Xnr3-As detailed in the Introduction, Xnr3 has several features that differ from the consensus for TGF-␤ superfamily members, including the lack of the seventh conserved cysteine and the substitution of a serine for a glycine residue between the second and third conserved cysteines. Mutation of this serine back to glycine did not alter the biological activity of Xnr3, even in chimeras in which a seventh cysteine was added (data not shown). We used chimera 9 to test whether the substitution of the conserved glycine residue by serine in Xnr3 was responsible for the lack of activity of chimera 4, which contains the center region of Xnr3. Chimera 9 was identical to chimera 4 except that the serine to glycine substitution had been made (Fig. 5A). This chimera yielded an unexpected result. The fusion protein consistently induced high levels of NCAM transcript (Fig. 5B). This result, and the fact that Xnr3 can tolerate changes to positions known to be crucial in other TGF-␤ superfamily members (47), suggest fundamental structural differences.
The Center Third of the Xnr2 Mature Domain Is Sufficient for Muscle Actin Induction-Whereas the presence or absence of the seventh cysteine did not alter the activity of wild-type Xnr3, the importance of this alteration to the activity became evident in the context of the Xnr2/Xnr3 chimeras. In chimera 6, which had strong NCAM-and no MA-inducing activity, only the middle third of the mature domain of Xnr3 was substituted with the Xnr2 sequence (Fig. 4). However, when the seventh cysteine was added back (chimera 10; Fig. 6A), the chimera acquired strong MA-inducing activity in three and weak activity in one of six independent trials (Fig. 6B). Despite the variability in the induction of MA transcript, chimera 10 induced NCAM above background levels in all six trials (Fig. 6). We speculate that chimera 10 may possess both NCAM-and MAinducing activity, and the relative induction of these two differentiated states may result in the variability seen for MA induction (see below and "Discussion"). This result identified a 46-amino acid stretch that is able to confer MA-inducing activity to Xnr3, provided that all of the cysteines are present.
Additional mutations were made to determine if the sequences responsible for conferring MA-inducing activity could be narrowed further. Within this 46-amino acid segment, 28 amino acids are identical between Xnr2 and Xnr3. To find out which of the remaining 18 remaining amino acids were critical, we hypothesized that necessary residues would be ones conserved among Xnr1, 2, and 4, which all induce MA, but not with Xnr3. The amino acid sequence FKP found in this region of Xnr2 (residues 346 -348) would appear to be a likely candidate (Fig. 6C). The identical sequence is found in Xnr1, whereas the corresponding sequence in Xnr4 is VKP, representing one conservative change. However, in Xnr3, the analogous amino acids FIG. 5. A, Xnr2/Xnr3 chimeras indicate that 22 amino acids at the N terminus and four residues at the C terminus of the mature domain are sufficient for NCAM-inducing activity (chimera 8). B, representative Northern blots. Samples were processed and scored as in Fig. 3. EF1␣, elongation factor 1 ␣.
are ENA, which substitutes a charged residue in the first position, and does not include a proline. To test the importance of these residues, chimera 11 was made to substitute the Xnr3 sequence at this region into chimera 10 (Fig. 6A). Although chimera 10 induced MA to levels comparable to Xnr2 half of the times tested, chimera 11 never induced MA to this extent. No MA transcript was seen in two cases, and a third independent experiment induced a reduced level of MA relative to Xnr2, although the signal was still above "ϩ/Ϫ" levels ( Fig. 6B).
A Chimera That Induces Both Neural and Mesoderm Reveals That These Are Two Distinct, Separable Activities-Chimeras 6 and 10 are identical except for the presence of the seventh conserved cysteine in chimera 10 (Fig. 6). Whereas chimera 6 was a strong NCAM inducer only, chimera 10 strongly induced both NCAM and MA markers. In Xenopus animal cap assays, dorsal mesoderm inducers will often secondarily induce neural tissue. The degree of secondary neural induction is variable because of the amount of uncommitted animal cap ectoderm remaining to be induced by factors derived from the dorsal mesoderm. Whereas secondary neural induction may explain the presence of both neural and dorsal mesoderm markers in some of the assays with chimera 10, in other assays no MA was detected. It is thus possible that chimera 10 possessed both direct mesoderm (MA) and neural (NCAM) inducing activities simultaneously. To investigate this possibility further, a Myc tag was added back in the region known to eliminate NCAMinducing activity from Xnr3 (Fig. 2) but which does not interfere with the MA-inducing activity of Xnr2. The resulting fusion protein, chimera 12, had strong MA-inducing activity in two of three trials and greatly reduced, or eliminated, NCAMinducing activity (Fig. 7).

DISCUSSION
Requirements for Xnr3 NCAM-inducing Activity-Experimental manipulation of Xnr3 primary structure helped to reveal regions important for activity. Specifically, proregion fusions, epitope tags, and site-directed mutations indicated that the thumb domain (48) was sensitive to amino acid additions or deletions. Three-dimensional structure predictions of BMP7 indicate that the thumb domain has minimal interactions with the remainder of the folded mature domain (5). It is likely that the thumb domain of Xnr3 is similarly unstructured. Therefore, Xnr3 would appear to be unique among related family members in having such strict structural requirements for this domain. This domain may be involved in a receptor interaction or it could play an indirect role in the overall folding of the protein, in which case disruption of this domain would affect the structure of another region.
Although many other TGF-␤s, including nodal and Xnr2, are able to function when produced as fusion proteins with heterologous proregions, this does not appear to be the case for Xnr3. Xnr3 fusions failed to produce active protein when the junction was in the thumb domain, which is done commonly for other TGF-␤s, or even N-terminal of the putative cleavage site. Surprisingly, even fusion of Xnr3 to the prodomain of the closely related Xnr2 failed to make an active protein. In contrast, the Xnr3 prodomain can be substituted into Xnr2 (e.g. chimera 2, Fig. 4). Previous studies on nodal have shown that prodomains play a crucial role in determining stability of the mature domain (39). Therefore, one possibility is that the strict requirements for Xnr3 prodomain relate to protein stability and that the presumed structural anomalies of Xnr3 can not be stabilized by other prodomains. Mutations to the putative cleavage site appear to indicate that processing is not required for Xnr3 activity, even though residues surrounding this region are necessary for activity. Because there are other distal potential cleavage sites, it is possible that Xnr3 may still be processed to a lesser extent. Mutations of the putative cleavage site in Xnr2 result in a protein with dominant negative activity, whereas similar mutations to Xnr1 and Xnr4 do not diminish activity (44). It was speculated that the residual activities of the mutant Xnr1 and Xnr4 were because of processing at alternative sites.
The results of the initial domain swaps between Xnr2 and Xnr3 (Fig. 4) showed that the prodomain, first, and last onethird of the Xnr3 mature domain were required for activity. Additional chimeras revealed that NCAM-inducing activity re- FIG. 6. A, the center third of the Xnr3 mature region is sufficient for MA-inducing activity provided a seventh cysteine is added (compare chimeras 6 and 10). B, representative Northern blots. Samples were processed and scored as in Fig. 3. C, alignment of the 46 residues comprising the center one-third of all four Xenopus nodal-related genes. The three amino acids in bold are those selected for mutagenesis in Xnr2 (in chimera 11). EF1␣, elongation factor 1 ␣. quired the thumb domain but not the region between the first and second cysteines. Residues at the C terminus are required as well, as indicated by the first series of chimeras. However, just the most C-terminal four amino acids appear to be sufficient for NCAM induction in combination with the Xnr3 proand thumb domains (chimera 8). Structural predictions of Xnr3 based on BMP7 predict that the C-terminal region is in close proximity to the thumb domain, and that it is possible these two areas interact (49 -51).
The activity of chimera 9 is particularly hard to explain. Chimera 9 differs from chimera 4 only by the substitution of a glycine for a nonconsensus serine residue found between the second and third cysteine residues of Xnr3 (Fig. 5A). Whereas chimera 4 had virtually no MA-or NCAM-inducing activity, chimera 9 had strong NCAM-inducing activity. Thus with chimera 9, NCAM-inducing activity was present even though the chimera lacked an Xnr3 sequence at the C terminus. It is possible that changing serine to glycine allows disulfide bonds to form between the second and sixth and the third and seventh cysteines, as in most TGF-␤s, which could mimic the folding of Xnr3 in the absence of the N-terminal 4 amino acids.
Requirements for Xnr2 MA-inducing Activity-Unlike Xnr3, Xnr2 has the conserved features of a typical TGF-␤ superfamily member, including all seven conserved cysteines and the sterically required glycine between cysteines two and three. Tertiary modeling based on the BMP7 and TGF-␤2 structures predict that Xnr2 has a very similar structure (49 -51). In chimera 2, the entire prodomain and first third of the mature domain of Xnr2 was substituted with the corresponding region of Xnr3. This chimera retained MA-inducing activity similar to that of wild-type Xnr2. On the other hand, chimera 1, which contained the reciprocal swaps between Xnr2 and Xnr3, had very little or no activity. The results from chimera 10 showed that the requirements for MA-inducing activity could be reduced even further to the center one-third of the mature domain provided that a C-terminal seventh cysteine was added. Without the seventh cysteine (chimera 6), the fusion protein had strong NCAM-inducing activity but no MA-inducing activity. Having identified the central third of the Xnr2 mature domain as containing essential features for MA-inducing activity, we compared the sequences of the four Xenopus nodal-like genes within this region to find differences that might account for the divergent activities of Xnr3 versus Xnr1, -2, and -4. Near the center of this region we identified the 3-amino acid sequence (F/V)KP in Xnr1, -2, and -4, which was substituted as ENA in Xnr3. Site-directed mutations to change these three amino acids in Xnr2 to the corresponding sequence in Xnr3 resulted in a fusion protein with greatly reduced MA-inducing activity but that retained strong NCAM-inducing activity. If the alignment of Xnr2 with the three-dimensional structures of BMP7 and TGF-␤ is used to model Xnr2, these three amino acids would be just N-terminal of the ␣-helix forming the heel of the handshaped fold (49 -51).
The domain that we have identified as being essential for MA-inducing activity by Xnr2 differs from regions known to be required for the activity of TGF-␤1 and TGF-␤2. Similar domain swapping studies have identified residues that are responsible for the divergent activities in TGF-␤1 and TGF-␤2 (52,53). It was shown that exchanging residues 92-98 of the mature domain was sufficient to change the activity of TGF-␤1 to resemble that of TGF-␤2 in a LS513 cell growth assay. In addition, this protein no longer bound to the TGF-␤1 receptor, T␤RII, which TGF-␤1 recognizes, but TGF-␤2 does not. Significantly, this part of the protein forms an extended surface loop forming the end of finger 2 and therefore is likely to be involved in receptor interactions (52). To test if similarly positioned residues played a role in distinguishing Xnr2 and Xnr3 activities, the sequences of Xnr2, Xnr3, TGF-␤-1, and TGF-␤-2 were first compared to determine which residues in the Xnrs were analogous to positions 92-98 of TGF-␤. In Xnr2, the corresponding amino acid sequence is EDGEVVL, whereas in Xnr3 it is ENEDFIL. We were confident in the alignment of these sequences because all four proteins share a tyrosine immediately preceding this sequence, and it is followed in 12 residues by the sixth conserved cysteine. Because the first and last amino acids are already shared between the two proteins, mutagenesis was done to change residues DGEVV in Xnr2 to NEDFI. Surprisingly, this mutant Xnr2 protein retained wildtype MA-inducing activity (data not shown). Whereas it is possible that the residues in Xnr2 analogous to those identified in TGF-␤1 and TGF-␤2 lie slightly more N-or C-terminal to those mutated, our Xnr2/Xnr3 chimera results suggest that very different regions of the nodal and TGF-␤s may be required for activity. Recent results consistent with this possibility have suggested that signaling by Xnr2 might be atypical for a TGF-␤ superfamily member. It has been found that a mutant Xnr2 with the fourth cysteine changed to a serine retains MA-inducing activity (44). This cysteine forms the interchain bond in the ligand dimer in other TGF-␤ superfamily members (5)(6)(7)(8)(9)(10). In activin, mutation of this residue results in protein with only 2% of wild-type biological activity (47).
Both Direct Neural and Mesoderm Inducing Activities may Co-exist in One Molecule-Chimera 10 had properties consistent with both Xnr2 and Xnr3 activities. Although it induced MA in three of six independent assays, it had strong NCAMinducing activity every time it was tested. Alterations to the structure of chimera 10 appear to independently disrupt one type of activity or the other. When the seventh cysteine was absent (chimera 6), making the protein more Xnr3-like, MAinducing activity was lost, but NCAM-inducing activity remained strong. Likewise, as discussed above, if the amino acid sequence FKP in the central domain of Xnr2 was mutated to ENA as in Xnr3 (chimera 11), MA inducing but not NCAMinducing activity was reduced. Finally, if an epitope tag was added in a position known to disrupt Xnr3 activity, MA-inducing activity was retained, but NCAM-inducing activity was greatly reduced or absent (chimera 12). Fig. 8 summarizes this work and shows the regions necessary for NCAM-or MA-inducing activity. The shaded areas required for activity are mutually exclusive. We show that a protein containing both domains has NCAM-and MA-inducing functions (chimera 10). We speculate that chimera 10 may be able to bind both as an agonist to the putative nodal receptor and as an antagonist to the BMP4 receptor. As suggested by chimera 11, mutation of critical residues in this middle portion of the protein may interfere with binding to the hypothetical nodal receptor and therefore prevent MA induction but without reducing NCAM induction. Conversely, the addition of an epitope tag in the thumb domain would block the inhibitory binding to the BMP receptor but not the activation of a nodal receptor.
The nodal-related proteins are a distinct subfamily of the TGF-␤s. There is a single nodal protein in mouse, whereas duplications have led to multiple nodal-like genes in zebrafish and Xenopus, which have two and four nodal-related genes, respectively. Xenopus has the most diverged family member, Xnr3. The structure and function of Xnr3 is unique among the nodal relateds, and a similar protein has yet to be found in any other animal. Perhaps Xnr3 was able to evolve its unusual characteristics because Xenopus has several redundant copies of the nodal gene. Xnr2 has an activity similar to the other nodals. However, this work has shown that the regions necessary for Xnr2 activity are different than those required by TGF-␤ itself and that the nodals may have evolved a different signaling strategy from other members of the superfamily.