Twisted gastrulation promotes BMP signaling in zebrafish dorsal-ventral axial patterning

The bone morphogenetic protein (BMP) signaling pathway is essential for many developmental processes (Hogan,1996), one of the earliest of which is to pattern cell fates along the dorsoventral (DV) axis in vertebrates and invertebrates(De Robertis and Sasai, 1996). In Xenopus and zebrafish, BMP signaling is required for the specification of ventral cell fates, whereas its restriction is necessary for proper specification of dorsal tissues(Holley and Ferguson, 1997). Genetic evidence implicating BMP signaling in DV patterning comes from a collection of zebrafish mutants that disrupt factors acting within a BMP signaling pathway (Hammerschmidt and Mullins, 2002). Mutations in genes encoding the ligands Bmp2b(Swirl) (Kishimoto et al.,1997; Nguyen et al.,1998) or Bmp7 (Snailhouse)(Dick et al., 2000; Schmid et al., 2000), the type I BMP receptor Alk8 (Lost-a-fin) (Bauer et al., 2001; Mintzer et al.,2001), or the intracellular effector Smad5 (Somitabun)(Kramer et al., 2002) result in the expansion of dorsal ectodermal and mesendodermal cell fates at the expense of nearly all ventral tissues.

Conversely, zebrafish mutants of the BMP antagonizing genes chordin (chordino) and sizzled (ogon)exhibit a reduction of dorsal and expansion of ventral cell types(Hammerschmidt et al., 1996). Extracellularly, Chordin, Follistatin and Noggin, which are expressed in dorsal domains, antagonize BMP signaling by directly binding and inhibiting BMP ligands, prohibiting activation of their receptors(De Robertis et al., 2000). Sizzled (Ogon) is unique as a BMP antagonist in its ventrally restricted expression pattern and its dependence on Chordin to antagonize BMP signaling(Yabe et al., 2003). Chordin is itself regulated by the Tolloid family of metalloproteases, which degrade Chordin and thus promote BMP signaling(Mullins, 1998). In zebrafish,mutations in tolloid [mini fin; tolloid like 1(tll1) - Zebrafish Information Network] result in the loss of the ventral fin fold, tissue derived from the ventral most region of the gastrula(Connors et al., 1999). This mildly dorsalized phenotype has led to a model in which zebrafish Tolloid(Mini fin) activity restricts Chordin, and thus promotes high levels of BMP signaling, at or near the end of gastrulation(Connors et al., 1999). The combined activity of these extracellular factors ensures proper patterning,and is consistent with a model in which dorsally produced BMP antagonists diffuse laterally to produce a gradient of BMP signaling(Holley and Ferguson, 1997; Thomsen, 1997), which imparts positional information and/or cell fate decisions along the DV axis.

Although the activities of BMP ligands, the extracellular antagonist Chordin and the protease Tolloid are well agreed upon, the function of the secreted factor Twisted gastrulation (Tsg) in DV patterning in vertebrates has been less clear. Tsg displays both BMP promoting and antagonizing activities in gain-of-function and biochemical assays. Overexpression of Tsg at high levels in Xenopus embryos causes ventralization, indicative of high levels of BMP signaling (Oelgeschlager et al., 2000; Ross et al.,2001). When expressed with Tolloid or at high levels relative to Chordin, Tsg can act as an agonist of BMP signaling(Larrain et al., 2001; Ross et al., 2001; Oelgeschlager et al., 2003). Consistent with this, Tsg can release BMP ligands from Chordin cleavage products (Larrain et al.,2001; Oelgeschlager et al.,2000), and Tsg enhances Tolloid-mediated proteolysis of Chordin(Larrain et al., 2001; Scott et al., 2001; Shimmi and O'Connor, 2003; Yu et al., 2000). Conversely,lower levels of overexpressed Tsg antagonize BMP signaling, causing dorsalization (Chang et al.,2001; Ross et al.,2001; Scott et al.,2001). Biochemically, Tsg binds with high affinity directly to BMP ligands and enhances the binding of Chordin to BMPs(Chang et al., 2001; Larrain et al., 2001; Oelgeschlager et al., 2000; Scott et al., 2001),suggesting that Tsg inhibits BMP signaling. One model taking into account these results proposes that Tsg activity depends on the cleavage status of Chordin, and thus on the activity of Tolloid: uncleaved Chordin would elicit strong binding between Tsg and BMPs to antagonize BMP signaling, whereas Tolloid-generated Chordin fragments, which possess residual anti-BMP activity,would be released from BMPs by Tsg to promote BMP signaling(Larrain et al., 2001). This model is based largely upon overexpression experiments, leaving unanswered the question of the endogenous role of Tsg as a pro- or anti-BMP factor in DV patterning of the embryonic axis.

We performed morpholino (MO)-based loss-of-function studies of tsg1 (tsgb - Zebrafish Information Network) in the zebrafish and examined genetic interactions between tsg1 and several BMP signaling component mutants. We found that knockdown of Tsg1 results in a moderately strong dorsalization, consistent with a loss of BMP signaling. We show a genetic interaction between tsg1 and bmp2b(swirl), supporting a role for Tsg1 as a BMP signaling agonist. We demonstrate that mini fin (tolloid) and tsg1cooperate in the promotion of BMP signaling. We also show that Tsg1 knockdown can partially suppress the chordino and sizzled(ogon) ventralized phenotypes, indicating that Tsg1 can act as a pro-BMP ventralizing factor in the absence of Chordin, as well as in the absence of Sizzled (Ogon). We propose that the predominant role of Tsg1 in DV patterning in the zebrafish is to promote BMP signaling, and that this function involves mechanisms that do not rely exclusively on the presence of Chordin or Chordin fragments.

Mutant strains used were mini fin^tm124a,swirl^tc300a, chordino^tt250 and ogon^tm305. Genotyping was performed as described [mini fin^tm124a by Connors et al.(Connors et al., 1999)]; chordino^tt250 by Oelgeschlager et al.(Oelgeschlager et al., 2003); swirl^tc300a by Wagner and Mullins(Wagner and Mullins, 2002),except Taq^α1 restriction enzyme was used to genotype ogon^tm305 using primers rk1-5′ and rk1-3′(Yabe et al., 2003)]. Genotyping was performed on all individual embryos pictured. The results are included in the tables, except Fig. 5C-H and Fig. 6C-F,where homozygous mutant parents were used.

RT-PCR and in situ hybridization indicated that tsg1 transcripts are found in zebrafish one cell stage embryos (not shown). 5′ RACE was performed to obtain the 5′ untranslated sequence of maternal tsg1 mRNA using GeneRacer kit (Invitrogen), according to the manufacturer's instructions. The sequence we obtained (up to the first codon of the open reading frame) was 5′GCATTAGTTGAGTCTGCTCCAGTTTAGTTCTGCTCAGCTCAGAGTTCGGCGGTAAATGCTCTGCAGCAAAACCTCGTGAGTTCAACATCTTCATTCTCAAGGTCCGGTGGGCTCTGTTCGTGTCGTGATG 3′.

Morpholinos were obtained from Gene Tools, LLC. Lyophilized powder was resuspended in water, then diluted into 1×Danieau buffer(Nasevicius and Ekker, 2000)for injection of 1 nl into one-cell stage embryos. The sequence of MO1 was as described by Ross et al. (Ross et al.,2001); MO4, 5′ TAAACTGGAGCAGACTCAACTAATG 3′; MO5,5′ CGCCGAACTCTGAGCTGAGCAGAAC 3′; a four nucleotide mismatch MO to MO1 (mismatch MO1), 5′ CTCATGTTGATGATGAACACCGCAT 3′; a 5 nucleotide mismatch MO to MO5 (mismatch MO5), 5′CCCCCAACTCTCAGCTCAGCACAAC 3′.

mRNA was in vitro transcribed as described(Nguyen et al., 1998) using SP6 mMessage mMachine kits (Ambion) and injected into embryos at the 1- to 2-cell stage. ztsg1 cDNA without its endogenous secretion signal was subcloned by PCR from zTsg1-pT3TS (a gift from M. O'Connor) into a derivative of pCS2 containing the Xenopus Chordin signal peptide and a FLAG epitope downstream of the signal peptide(Oelgeschläger et al.,2000). To create FLAG-tagged ztsg1 with its endogenous secretion signal, the Chordin secretion signal was replaced with ztsg1 sequence using standard molecular biology procedures.

tsg1 RNA was transcribed from these NotI linearized plasmids. tsg2 RNA was transcribed from pCS2-ztsg (a gift from M. Oelgeschläger and E. M. De Robertis) linearized with NotI. All injections were performed on at least three separate occasions. For rescue experiments, MO was injected first, followed by a second injection of mRNA into a random subset of the MO-injected embryos. Embryos injected with MO alone were then compared with those that were co-injected with mRNA.

Wild-type embryos were injected with mRNA encoding FLAG-tagged tsg1. A subset of these embryos were subsequently injected with 32 ng MO1 or mismatch MO1. Batches of five embryos were lysed in 20 μl SDS-PAGE loading buffer (Sambrook et al.,1989), boiled for five minutes, centrifuged for five minutes and the supernatants subject to SDS-PAGE analysis on 12% gels. After transferring to PVDF, membranes were probed with 1:1600 dilution of anti-FLAG antibody(Sigma) followed by 1:3000 dilution of HRP-conjugated sheep anti-mouse antibody (Amersham Biosciences) and detection with ECL plus western blotting detection kit (Amersham Biosciences) according to the manufacturer's instructions.

Labeling with digoxigenin-conjugated nucleotides by terminal deoxynucleotidyl transferase was performed on embryos fixed in 4%paraformaldehyde using ApopTag Detection Kit (Intergen) according to the manufacturer's instructions. After incubation with alkaline phosphatase-conjugated anti-digoxigenin antibody, staining was performed with 330 mg/ml NBT and 170 mg/ml BCIP (Sigma). Whole mount in situ hybridization was performed as described (Nguyen et al.,1998) with probes against bmp4(Chin et al., 1997), pax2.1 (Krauss et al.,1991), krox20 (Oxtoby and Jowett, 1993), myod(Weinberg et al., 1996), eve1 (Joly et al.,1993), gsc (Stachel et al., 1993), foxb1.2 (formerly fkd3)(Odenthal and Nüsslein-Volhard,1998), dlx3 (Akimenko et al., 1994), gata2(Detrich et al., 1995) and chd (Miller-Bertoglio et al.,1997). All images were taken from an MZ12.5 stereomicroscope(Leica) with a ColorSNAP-cf digital camera (Photometrics) and processed using Adobe Photoshop.

To study the function of Tsg1 in early development, we injected an antisense MO oligonucleotide to inhibit Tsg1 translation. This MO, designated MO1, is complementary to the first 25 nucleotides of the tsg1 open reading frame and of identical sequence to a MO used previously(Ross et al., 2001). Injection of 8 ng of MO1 into wild-type embryos resulted in the formation of an edema in the ventral tail vein in 61% of embryos(Fig. 1A,B; Table 1). In these embryos, the tail vein is dilated and blood pools within it. This phenotype is first visible at about 28 hours post fertilization (hpf) as a mild edema, which becomes prominent by 48 hpf (Fig. 1A,B; Table 1). A less severe phenotype was evident in 24% of the embryos, as decreased blood flow through the ventral tail vein (not shown). In all MO-injected embryos, we observed a reduction in anterior tissues at 48 hpf (compare bars in Fig. 1A,B). Surprisingly, the morphants did not display duplications of ventral tail fin tissue, which are typical of ventralized zebrafish mutants (see Figs 5, 6). These defects are similar to those seen in a previous study (Ross et al., 2001), where it was concluded that tsg1 morphants resemble weakly ventralized zebrafish mutants.

To determine if patterning was affected in MO1-injected embryos, we examined gene expression markers of dorsally and ventrally derived tissues. During early somitogenesis stages, we did not observe changes in the expression of pax2.1 or krox20, which mark dorsal ectodermal gastrula derivatives of the midbrain-hindbrain boundary and rhombomeres 3 and 5, respectively (Fig. 1G,K), or the extent of myoD expression to indicate dorsal mesoderm(Fig. 1H,L). Neither did we observe alterations in ventrally derived tissue, as revealed by pax2.1 expression in the pronephric system(Fig. 1H,L) or bmp4ventral tail bud expression (Fig. 1J,N). Additionally, the most anterior expression domains of bmp4 and pax2.1 were not altered (asterisks in Fig. 1I,J,M,N). Based on these results, we conclude that 8 ng of MO1 does not induce significant alterations in DV patterning. Thus, our results are incongruent with those of Ross et al.(Ross et al., 2001), who reported a reduction in dorsal markers and an expansion in ventral markers during somitogenesis stages. We subsequently injected doses of 12, 15, and 16 ng of MO1. Although we saw increased penetrance and severity of the vein phenotype, we did not observe alterations in the expression domains of pax2.1, krox20, myoD, bmp4 or gata2 with these amounts of MO(not shown).

As a change in early patterning could not account for the reduction of anterior structures at 48 hpf, we investigated if the defects were specific or non-specific effects of MO1. To test this, we co-injected in vitro transcribed tsg1 or tsg2 mRNA, which lack sequence complementary to MO1,and assayed for rescue of the defects. We found that either mRNA could rescue the vein edema in a large fraction of embryos(Fig. 1C, Table 1). However, we did not detect rescue of the anterior structures by co-injection of tsg mRNA(Fig. 1C). Examination of live embryos revealed a high degree of cell death that became prominent at 24 hpf,a common nonspecific effect of MOs(Heasman, 2002; Nasevicius and Ekker, 2000). To determine if the decrease in anterior structures could be attributed to cell death, we used a TUNEL-based assay to label the nuclei of dying cells. We observed a substantial increase in cell death in anterior CNS tissues of morphants (Fig. 1D,E), which could not be rescued by RNA co-injection(Fig. 1F). We attribute the reduction in anterior tissues at 48 hpf to nonspecific, MO-induced cell death.

We investigated whether the 8-16 ng tsg1 knockdown may be an incomplete loss of tsg1 function by examining the effects of injecting increasing amounts of MO1. We found that 32 ng of MO1 appeared to dorsalize the embryo moderately to moderately strong, similar to class 3 (C3)and class 4 (C4) dorsalizations (Table 2; Fig. 2K,O, not shown) (Mullins et al., 1996). At no dose did we observe defects consistent with a ventralization. To verify the specificity of MO1, we designed two additional MOs, MO4 and MO5, targeted against non-overlapping sequences in the 5′ untranslated region. We found that injection of 32 ng of MO4 or 25 ng of MO5 induced a moderately strong dorsalization (Fig. 2; Table 2), similar to that of MO1.

We assayed patterning during gastrulation and early somitogenesis to determine if the presumptive dorsalization reflects a change in pattern formation. During mid-gastrula stages, we found that morphants displayed more restricted expression domains of the ventral markers eve1, dlx3, and gata2 (Fig. 2A-C,F-H). Concomitantly, the expression of foxb1.2 and chordin, which are normally dorsally restricted, were expanded into lateral domains(Fig. 2D,E,I,J), consistent with an alteration in DV patterning during the gastrula period when BMP signaling is thought to act. During early somitogenesis, we examined injected embryos for expression of pax2.1, krox20 and myoD, and found that all expression domains of dorsally derived tissues were expanded(Fig. 2L,M,P,Q), indicating a dorsalized phenotype. As controls for the specificity of the dorsalized appearance, we injected up to 60 ng of two different mismatch MOs, which did not cause dorsalization (not shown). To test for the ability of the MOs to block translation, we injected 32 ng of MO1 or its four base mismatch control into embryos that were co-injected with a FLAG-tagged tsg1 mRNA containing the MO1 binding site. We found that MO1 blocked translation of tsg1, whereas the mismatch control did not(Fig. 2R). Furthermore, we could partially rescue the MO-induced dorsalization by co-injection of 2 pg of tsg1 mRNA (Table 2). Taken together, these results demonstrate that a strong knockdown of Tsg1 function by high levels of tsg1 MO induces specific dorsalization phenotypes, strongly suggesting that the in vivo role of Tsg1 in zebrafish is to promote the specification of ventral cell fates.

To examine if Tsg1 promotes ventral cell fates by promoting BMP signaling,we tested whether tsg1 interacts genetically with known BMP pathway component mutants. We injected a sub-dorsalizing amount of tsg1 MO1(16 ng) into embryos from a cross between a swirl (bmp2b)heterozygote and a wild type fish. In contrast to the uninjected,phenotypically wild type swirl (bmp2b) heterozygotes,greater than 95% of the tsg1 MO-injected heterozygotes displayed dorsalized phenotypes at 24 hpf, ranging from weak (class 1, 13%; class 2,36%), moderate (class 3, 28%) to strong (class 4, 21%)(Fig. 3B-D; Table 3). Importantly, 90% of injected wild-type sibling embryos were not dorsalized(Fig. 3A and Table 3). Dorsalization was also evident during somitogenesis by an expansion of dorsally expressed markers in the MO1-injected swirl heterozygotes(Fig. 3E-J). Furthermore, we found that 12 pg of bmp2b mRNA or 20 pg of mRNA encoding the intracellular BMP effector Smad5 could rescue the tsg1 MO5 dorsalization (Table 2). Our ability to rescue Tsg loss-of-function with multiple components of a BMP signaling pathway, as well as the genetic interaction between loss of swirl (bmp2b) and tsg1 function indicates that Tsg1 functions endogenously to promote BMP signaling in dorsoventral patterning.

As discussed above, previous work in Xenopus suggests that the pro-BMP activity of Tsg depends on the ability of Tolloid to cleave Chordin into fragments (Larrain et al.,2001). These fragments contain residual anti-BMP activity, which Tsg can inactivate. To address whether Mini Fin (Mfn) (Tolloid) is required for Tsg1 function, we injected 12.5 ng of MO5 into the progeny of a cross of mfn^tm124a heterozygous adults, which carry a presumptive null mutation in tolloid (Connors et al., 1999). The mfn mutation produces a weakly dorsalized (class 1) phenotype, which is characterized by loss of ventral fin fold tissue (Fig. 4A,B)(Connors et al., 1999). Whereas most (94%, Table 4) tsg1 MO5-injected wild-type or heterozygous siblings were not dorsalized, a large majority of mfn homozygous mutants (76%, Table 4) exhibited a significantly stronger dorsalized phenotype (class 3, Fig. 4C,E) than did uninjected mfn mutants (Fig. 4B,D). These results indicate that Tsg1 ventralizing activity does not depend on the product of the mfn gene or the ability of Mfn to generate Chordin fragments. Additionally, this result uncovers a previously undetected role for Mfn/Tolloid in patterning dorsoanterior tissues, which remain unaltered in mfn mutants(Connors et al., 1999; Mullins et al., 1996), but are expanded by injection of tsg1 MO at amounts insufficient to alter wild-type siblings.

We tested the genetic interaction between Chordin and Tsg1 loss of function. In Xenopus, low levels of Tsg overexpressed simultaneously with Chordin lead to additive dorsalization(Chang et al., 2001; Ross et al., 2001; Scott et al., 2001), whereas higher levels of Tsg can release BMP from Chordin fragments and promote ventralization (Oelgeschlager et al.,2000; Ross et al.,2001). Thus, Tsg might function exclusively through Chordin, and consequently the chordino (chordin) mutation may be epistatic to Tsg loss of function. A zebrafish chordino null mutation causes a moderate ventralized phenotype, including a reduction in dorsoanterior neural and somitic mesodermal tissue, expansion of the ventrally derived blood and duplications in the ventral fin fold(Fig. 5A)(Hammerschmidt et al., 1996). The expected protein product of this mutant chordin allele lacks all its BMP-binding domains (Fisher and Halpern, 1999; Schulte-Merker et al., 1997). We found that injection of 8 ng of MO1 into chordino mutants could suppress duplications of the ventral tail fin fold (Fig. 5B; Table 5), not supporting a mechanism in which chordino is epistatic to tsg1. Additionally, chordino null mutants injected with 25 ng of MO5 displayed more prominent dorsoanterior neural and somitic mesodermal tissue during early somitogenesis stages, compared with uninjected mutant siblings(Fig. 5C-H), indicating a partial suppression of these ventralized defects as well (compare with wild-type embryos; Fig. 2L,M). Thus, the endogenous ventralizing activity of Tsg1 does not rely exclusively on the presence of Chordin or Chordin fragments.

Similarly, we tested whether Tsg depends on Sizzled (Ogon) activity. At 24 hpf, ogon mutants display a mild ventralization, which includes multiple ventral fin folds and expanded blood precursors(Fig. 6A)(Hammerschmidt et al., 1996; Miller-Bertoglio et al., 1999; Solnica-Krezel et al., 1996). When injected with 8 ng MO1, however, ogon mutants displayed a single ventral fin fold, similar to the suppression seen in chordino mutants(Fig. 6B; Table 5). Additionally,injection of 20 ng of MO5 into ogon mutants suppressed the expansion of a ventral marker (eve1, Fig. 6C,E) and the restriction of a dorsal marker (foxb1.2, Fig. 6D,F) during gastrulation. These results indicate that Tsg1 can promote BMP signaling when the BMP antagonistic function of Ogon is deficient. Our observations that tsg1 MO can partially suppress both the chordino and ogon mutant phenotypes is consistent with the dependence of Sizzled(Ogon) dorsalizing activity on the function of Chordin(Yabe et al., 2003).

Surprisingly, like the tsg1 knockdown phenotype, Tsg overexpression also dorsalizes the zebrafish embryo (see also Ross et al., 2001). We injected various amounts of tsg1 mRNA into wild-type embryos and monitored DV patterning by in situ hybridization with pax2.1, krox20and myoD at the eight-somite stage. We found that 100-200 pg of tsg1 mRNA caused a moderate, lateral expansion of all three markers(Fig. 7A-E), and mild to moderate dorsalized phenotypes at 1 day post fertilization (dpf)(Table 6; Fig. 7F-I). About half of the mildly dorsalized embryos (class 1 or class 2, Fig. 7G,H) also displayed small duplications in the fin fold (Fig. 7H inset), similar to overexpression of Xenopus Tsg mRNA in zebrafish (Oelgeschlager et al.,2003) and some weakly dorsalized mutants(Kramer et al., 2002). Higher amounts of tsg1 mRNA caused more severely dorsalized phenotypes (not shown). Lower amounts had negligible effects. At no dose did we observe ventralized phenotypes.

We also injected tsg1 mRNA into ventralized mutant embryos. We intercrossed chordino (dino) heterozygotes, injected their progeny with 75-150 pg tsg1 mRNA, then scored and genotyped the embryos. The majority of homozygotes retained a ventralized phenotype(Table 6) and could be readily distinguished from wild-type and heterozygous siblings. Frequently, this class displayed a partial loss of proximal ventral fin fold tissue, but maintained multiple distal fin duplications (not shown), a phenotype seen in dino^-/-; swirl^+/- mutants and in a small fraction of dino mutants(Wagner and Mullins, 2002; Hammerschmidt and Mullins,2002). In addition, 39% of dino mutants exhibited either a partial or complete suppression of the ventralization or a dorsalized phenotype. Thirty-four percent of homozygous mutants displayed weakly dorsalized phenotypes, indistinguishable from those seen in wild-type or heterozygous injected siblings. Additionally, a small fraction (5%) of dino mutants were rescued to a nearly wild-type phenotype(Table 6), showing that overexpressed Tsg can suppress the absence of Chordin, consistent with overexpressed Tsg antagonizing BMP signaling. Similar results were seen with sizzled (ogon) mutants injected with 75-200 pg of tsg1 mRNA (Table 6),which normally display a weakly ventralized phenotype.

Overall, these experiments show that overexpressed Tsg can antagonize ventral cell fate specification in wild-type and mutant backgrounds. However,our loss-of-function results indicate that endogenous Tsg1 promotes ventral cell fate specification. It is possible that Tsg1 acts as a crucial component of a BMP-promoting complex that cannot form in either the absence of Tsg1 or the presence of excess Tsg (see Discussion).

Here, we demonstrate that a strong knockdown of Tsg1 dorsalizes the zebrafish embryo, indicating that in zebrafish Tsg1 functions predominantly as a ventralizing factor in DV patterning. We demonstrate a genetic interaction between tsg and bmp2b (swirl), consistent with Tsg promoting BMP signaling. Sub-dorsalizing knockdown of tsg1 increases the severity of the tolloid (mini fin) mutant, suggesting that Tsg1 and Mini fin cooperate to pattern anterior tissue, a function not previously attributed to Mini fin (Tolloid). Furthermore, we show that Tsg functions as a ventralizing factor, in part, independently of the dorsalizing factors Chordin and Sizzled (Ogon). In contrast to the strong dorsalizations observed in null bmp2b (swirl), bmp7(snailhouse), MZ-alk8 (lost-a-fin) and MZ-smad5 (somitabun) mutants(Bauer et al., 2001; Kramer et al., 2002; Mintzer et al., 2001; Schmid et al., 2000), we found that the strongest tsg1 knockdown phenotype is a moderately strong dorsalization of the embryo. As we do not know if Tsg1 function is completely knocked down, the moderately strong dorsalization may reflect the complete or partial role of Tsg1 in DV patterning and BMP signal modulation. Nevertheless,our results demonstrate a clear role in the zebrafish for Tsg1 as an agonist in BMP signaling during DV pattern formation.

A previous model of Tsg function suggests that full-length Chordin causes Tsg to act as a BMP antagonist, whereas the presence of Chordin fragments allows Tsg to exhibit pro-BMP behavior(Larrain et al., 2001). Thus,the activity of Tsg may depend on the cleavage status of Chordin. However, the partial suppression of the chordin mutant phenotype by Tsg1 knockdown indicates that endogenous Tsg1 acts as a ventralizing factor in a manner that does not rely entirely on Chordin. These results are consistent with the finding that an overexpressed mutant Tsg that cannot bind BMPs reduces ventral tissues to a greater extent than loss of Chordin alone(Oelgeschlager et al., 2003). Thus, Tsg1 may promote BMP signaling in part by inactivating BMP antagonists in addition to Chordin or Chordin fragments.

However, as tsg1 knockdown only partially suppresses the dino ventralized phenotype and tsg1 is not epistatic to dino, Tsg1 probably also inactivates Chordin in conjunction with Tolloid, as previous studies indicate(Larrain et al., 2001; Oelgeschlager et al., 2000). Moreover, as dino mutants display a range of phenotypes from a mild ventralization that is viable to a moderate ventralization that is lethal(Fisher and Halpern, 1999), it is possible that loss of tsg1 affects the range of dinophenotypes, with the mild phenotype becoming predominant. This could be a direct effect of Tsg on other antagonists, as discussed above, or it could be indirect by unknown mechanism(s) that also modulate the dinophenotype.

Tsg has been reported to enhance Tolloid proteolysis of Chordin(Larrain et al., 2001; Scott et al., 2001; Shimmi and O'Connor, 2003; Yu et al., 2000). We found that a sub-dorsalizing knockdown of Tsg1 in wild type exacerbates the mildly dorsalized phenotype of mini fin (tolloid) mutants(Fig. 5), suggesting a previously unknown role for Mini fin (Tolloid) in patterning rostral tissues in zebrafish (Connors et al.,1999; Mullins et al.,1996). There is evidence that additional Tolloid-related enzymes function during gastrulation (J. Xie and S. Fisher, personal communication),which may normally mask the loss of mini fin (tolloid)during these stages. If Tsg1 increases the rate of proteolysis of multiple Tolloid enzymes, then loss of Tsg could dorsalize the embryo by reducing the ability of multiple Tolloid factors to degrade their targets.

If endogenous Tsg1 promotes ventral cell fates by facilitating BMP signaling, why does Tsg1 overexpression dorsalize the embryo, reflecting a loss of BMP signaling? One possibility is that Tsg functions in a multi-component protein complex, binding both to a BMP ligand, as previously shown (Chang et al., 2001; Larrain et al., 2001; Oelgeschlager et al., 2000; Ross et al., 2001; Scott et al., 2001), and at least one other factor required to promote BMP signaling. Loss of Tsg would disrupt formation of this complex and result in decreased BMP signaling. Excess Tsg would bind independently to both free BMP ligands and the other factor in the complex, again preventing formation of the trimolecular complex and reducing BMP signaling. Thus, both the loss- and gain-of-function phenotypes would cause the same defect(Fig. 8). A similar phenomenon is also observed in the loss- and gain-of-function phenotypes of some Wnt planar cell polarity components and other genes(Gubb et al., 1999; Hiromi et al., 1993; Krasnow and Adler, 1994; Strutt et al., 1997; Tomlinson et al., 1997).

In the mouse, Tsg mutants do not display defects in DV patterning of the embryonic axis (Nosaka et al.,2003; Petryk et al.,2004; Zakin and De Robertis,2004). However, they do exhibit several other defects consistent with possible roles in either promoting or inhibiting BMP signaling. For example, thymocytes from Tsg^-/- animals show increased phospho-Smad1 levels, suggesting an increase in BMP signaling in the absence of Tsg (Nosaka et al., 2003). Tsg^-/- mice also display defective skeletogenesis, similar to the effect of a dominant-negative BMP receptor(Nosaka et al., 2003; Zakin and De Robertis, 2004),indicating a possible defect in BMP signaling. Interestingly, a genetic interaction is also observed in the mouse, as we observe in zebrafish, between a tsg and BMP mutation. Tsg^-/-;Bmp4^+/-animals display holoprosencephaly and craniofacial defects, not seen in either Tsg mutant animals or Bmp4 heterozygotes(Zakin and De Robertis, 2004). These results support a role for Tsg in promoting BMP signaling in head development in the mouse.

In contrast to our study, a recent loss of function study in Xenopus reports that MO-induced knockdown of Tsg causes ventralization, indicating that Tsg predominantly inhibits BMP signaling in Xenopus (Blitz et al.,2003). Tsg knockdown in Xenopus mildly restricts a small fraction of dorsally expressed genes during midgastrulation, with more substantial changes evident during neurula stages. These findings led to the conclusion that in Xenopus Tsg maintains the specification of dorsal cell fates, presumably after BMP signaling has patterned the early DV axis. We find that Tsg1 knockdown in zebrafish causes a moderately strong dorsalization in pattern formation during gastrulation. It is possible that Tsg acts oppositely at different stages and/or activity levels to affect the DV pattern of the embryo or functions in a nonconserved manner in these two organisms.

In Drosophila, Tsg also modulates BMP signaling during DV axial patterning. Tsg functions similarly to the Chordin ortholog Short gastrulation(Sog) in specifying the dorsal-most tissue, the amnioserosa, which requires the highest levels of BMP signaling in the fly embryo(Ashe and Levine, 1999; Francois et al., 1994; Mason et al., 1994; Ross et al., 2001; Zusman and Wieschaus, 1985). In addition to this pro-BMP activity, both Tsg and Sog exhibit anti-BMP activity in dorsolateral regions of the embryo(Ross et al., 2001). Current models suggest that Sog and Tsg bind to and transport BMP ligands toward dorsal regions of the embryo, where they are released from Sog by the activity of Tolloid, thereby generating the highest levels of BMP signal dorsally(Decotto and Ferguson, 2001; Eldar et al., 2002; Shimmi and O'Connor, 2003). In this model, the activity of Tsg relies on the presence of Sog and Tolloid. In vertebrates, there is no evidence for a role for Chordin in promoting gastrula BMP signaling, although it is possible it plays such a role at later stages in tail patterning (Hammerschmidt and Mullins, 2002; Wagner and Mullins, 2002). Thus, all aspects of how Sog/Chordin and Tsg function in DV patterning in vertebrates and invertebrates may not be conserved.

In zebrafish, Tsg1 could act as a BMP antagonist at other stages of development or under particular conditions that we did not detect in our studies. We do not know the nature of the tail vein edema phenotype observed in low level Tsg knockdown embryos (Fig. 1). It may reflect its role as a BMP antagonist, as chordin mutants exhibit a similar edema, although in conjunction with other ventralized defects, which we do not detect in tsg1 morphants. It is likely that the timing, location and/or levels of expression of Tsg,possibly with other factors, are crucial in determining whether Tsg functions to promote or antagonize BMP signaling in different developmental contexts. In zebrafish, the mechanism by which Tsg promotes BMP signaling, and the identity of any additional Tsg-interacting factors, remains unclear. Further work will be required to determine how Tsg acts in relation to other BMP modulating factors, in order to elucidate the mechanism by which Tsg promotes BMP signaling.

We thank M. Oelgeschläger and E. M. De Robertis for the tsg2clone and pCS2 secretion vector, and M. O'Connor for the tsg1 clone. We are grateful to J. Xie and S. Fisher for communicating their work prior to publication. We thank D. Wagner for many helpful discussions; and R. Dosch, T. Gupta, F. Marlow and J. Schumacher for comments on the manuscript and other advice. This work was supported by NIH grant GM56326 and a Developmental Grant Award from the Center for FOP and Related Disorders to M.C.M. and training grants 5 T32 GM07229-28 and 2 T32 HD007516-05 to S.C.L.