The role of tolloid/mini fin in dorsoventral pattern formation of the zebrafish embryo

Bone morphogenetic proteins (Bmps) and their signaling pathways are essential in the development of numerous tissues in vertebrates (reviewed in Hogan, 1996). One of the earliest developmental processes involving Bmp signaling is the establishment of the dorsoventral axis of the embryo. Overexpression experiments in the frog and fish implicate a Bmp signaling pathway in ventral cell fate specification (reviewed in Mullins, 1999; Thomsen, 1997). Mutants of bmp2b in the zebrafish (swirl mutants) demonstrate a requirement for Bmp2b in the specification of nearly all ventral cell fates (Kishimoto et al., 1997; Nguyen et al., 1998b). Repression of Bmp signaling is necessary for dorsal mesodermal and neural development and is mediated through the action of Bmp antagonists, molecules secreted from dorsal tissue that can bind and prevent Bmp ligands from activating their receptors (reviewed in Graff, 1997; Thomsen, 1997). Zebrafish mutants of chordin (chordino mutants; Fisher et al., 1997; Schulte-Merker et al., 1997), one of the Bmp antagonists, exhibit a moderate ventralized phenotype, consistent with models of chordin (chd) function.

Ectopic expression and loss-of-function studies suggest that cell types derived from different dorsoventral regions of the gastrula are specified by a gradient of Bmp activity, with high levels specifying ventral and low levels lateral cell fates (Dosch et al., 1997; Jones and Smith, 1998; Knecht and Harland, 1997; Neave et al., 1997; Nguyen et al., 1998b; Wilson et al., 1997). The Bmp activity gradient is not reflected in gene expression levels of Bmp2 or Bmp4, since these genes are expressed uniformly in ventral regions or throughout the embryo (Chin et al., 1997; Nikaido et al., 1997, reviewed in Thomsen, 1997). The gradient of Bmp activity is thought to be generated by the diffusion of the Bmp antagonists, Chd, Noggin and Follistatin, from the dorsal organizer of the embryo towards ventral regions (Dosch et al., 1997; Jones and Smith, 1998; Knecht and Harland, 1997; Wilson et al., 1997). The presumptive gradients of Bmp antagonists would thus establish a gradient of Bmp activity in an opposite direction to that of the Bmp antagonists. In dorsal regions there would be little or no Bmp activity, in lateral regions low activity, and in ventral regions the highest activity.

A conserved TGF-β signaling pathway establishes the dorsoventral axis of the Drosophila embryo (although designation of the axes is inverted between vertebrates and invertebrates; reviewed in De Robertis and Sasai, 1996; Mullins, 1998). In the fly, the decapentaplegic (dpp) and short gastrulation (sog) genes are orthologues to bmp2/4 and chd, respectively. dpp specifies dorsal cell fates, while sog, expressed in ventral regions, diffuses dorsally and can repress dpp signaling. A further modulator of Dpp activity is the astacin family metalloprotease, Tolloid (Tld). Mutants of tld display phenotypes similar to moderate or weak dpp alleles (Ferguson and Anderson, 1992b). Increasing the level of dpp can suppress tld loss-of-function defects, suggesting that Tld acts to enhance Dpp activity (Ferguson and Anderson, 1992a,b).

Genes related to Drosophila tld have been isolated in zebrafish (tld, Blader et al., 1997), Xenopus (Xolloid, Goodman et al., 1998; Piccolo et al., 1997, and Xtld, Lin et al., 1997; Maéno et al., 1993), mouse (mTld/Bmp1, Fukagawa et al., 1994; Takahara et al., 1994, and mTll, Takahara et al., 1996), and human (mTld, Takahara et al., 1994). Overexpression of Xolloid/tld in frog and fish embryos causes a moderate ventralization of the embryo (Blader et al., 1997; Goodman et al., 1998; Piccolo et al., 1997), a phenotype resembling that of chd mutants in the zebrafish (Fisher et al., 1997; Hammerschmidt et al., 1996a). In epistasis tests, Xolloid/tld overexpression blocks a dorsalization induced by chd, but did not affect a dorsalization caused by noggin or follistatin (Blader et al., 1997; Marqués et al., 1997; Piccolo et al., 1997). These results suggest that Xolloid/tld acts upstream of chd to specifically inhibit its activity. Biochemical evidence in the fly, frog and fish demonstrates that Tld can proteolytically cleave Chd/Sog protein (Blader et al., 1997; Marqués et al., 1997; Piccolo et al., 1997). Additional data indicate that the released Bmp ligands can bind and activate their receptors (Piccolo et al., 1997).

In the fly, Tld plays a fundamental role in modifying the activity of Sog to generate a Bmp signaling gradient while, in the vertebrate embryo, the role played by Tld in dorsoventral patterning has not been established. Null mutants of the mouse Bmp1/mTld gene display defects in ventral body wall closure, likely due to improper Procollagen processing in the amnion (Suzuki et al., 1996). No abnormality in dorsoventral axis formation was found in these mutants; however, the mTll gene may play a role in this process in the mouse.

Here, we report on the genetic analysis of tolloid function in dorsoventral patterning of the zebrafish embryo. Overexpression of tld mRNA in embryos of six different zebrafish dorsalized mutants rescued the phenotype of only one of these mutants, mini fin (mfn). Linkage analysis, molecular cloning and DNA sequence analysis of the tld gene from several mutant mfn alleles, demonstrate that mfn is a mutation in the tld gene. We found that mfn is required to establish the most ventral cell types of the tail: the ventral fin, somites and vasculature. The loss of Mini fin/Tolloid activity is first manifested by a reduction in ventrally restricted gene expression at the end of gastrulation. Furthermore, some mfn mutants exhibit an aberrant expansion of chd gene expression. We propose that mfn/tld gene activity is required at the end of gastrulation and within the tail bud, when the dorsal and ventralmost marginal cells of the embryo lie in close proximity and Chd may diffuse to the ventralmost vegetal regions. Tld/Mfn would act to inhibit Chd function at these stages, thus generating high Bmp activity levels in ventral vegetal regions of the embryo, which then specify ventral tail cell fates.

In situ hybridizations were performed essentially as described by Schulte-Merker et al. (1992). The following probes were used: bmp4 (Chin et al., 1997), dlx3 (Akimenko et al., 1994), eve1(Joly et al., 1993), flk1 (Thompson et al., 1998), msxB (Akimenko et al., 1995), msxD (Akimenko et al., 1995), myoD (Weinberg et al., 1996), tld (Blader et al., 1997) and chd (Miller-Bertoglio et al., 1997). For double in situ hybridization, embryos were incubated with both fluorescein- and digoxigenin-labeled antisense probes. The fluorescein-labeled probe was detected with anti-fluorescein antibodies and visualized with 5-bromo-6-chloro-3-indolyl phosphate and tetrazolium red. After antibody removal (0.1 M glycine, 0.1% Tween-20), embryos were incubated with anti-digoxigenin antibodies and processed as described above. Embryos were scanned with a Progres 3012 digital camera (Kontron Elektroniks) and the images processed with Adobe software.

mfn mutants were identified by their mutant phenotype at bud and later stages, or at bud and earlier stages by 25% percent of the embryos exhibiting a different staining pattern. When less than 25% showed an altered pattern, embryos were genotyped by PCR using a restriction fragment length polymorphism (RFLP) created by the mutation. In these cases, pictures were taken to document the in situ expression in individual embryos, then embryonic DNA was made as described below. PCR was performed using primers flanking the mutation (tm124a allele: S12 TGTGTCAAGCATAAAGACTGG, S13 TCTTCTCAATGAAAGTCACGC), the PCR product was cut with MspI, and run on a 2% agarose gel. The mfn^tm124a mutation causes the loss of an MspI restriction site.

Mutant embryos from different mfn alleles were placed into categories (A-E) at 2 days of development, when the ventral tail fin is well defined and variations in phenotype are the most pronounced.

Synthetic tld mRNA was made from pCS2:Ztld-3′MT (Blader et al., 1997) and injected into the yolk of 1- to 4-cell-stage embryos as described in Westerfield (1995). Mutants of the following alleles were injected: swirl^dta72, somitabun^dtc24, snailhouse^ty68a, piggytail^ty40a and lost-a-fin^tm110b (all described in Mullins et al., 1996). Embryos from crosses between transheterozygous adult mfn fish (mfn^ty130a/mfn^tf215a) were injected, so that all progeny were mutant and could not be mistaken for wild-type siblings. A two-fold range of tld mRNA (80-160 pg) was injected.

Crosses between Tübingen (TÜ) strain fish carrying either the mfn^tm124a, mfn^tb241c or the mfn^ty130a mutation and polymorphic WIK or AB strain fish were used to generate map crosses. The tm124a allele was originally reported to be an allele of the piggytail gene (Mullins et al., 1996). Additional complementation analyses revealed that tm124a is an allele of the mfn gene. The mapping procedure and the WIK line are described in Rauch et al. (1997) and Knapik et al. (1996). Pools of about 25 F₂mfn mutant embryos and wild-type siblings were collected separately and stored in methanol at −20°C. We found that embryos stored in methanol yielded higher amounts of genomic DNA. Embryonic DNA and PCR was performed as described by Gates et al. (1999), with the following PCR conditions: 94°C for 1 minute, 5 cycles of 94°C for 30 seconds, 54°C for 2 minutes and 73°C for 1 minute followed by 35 cycles of 94°C for 30 seconds, 55°C for 30 seconds and 73°C for 1 minute. Linkage was found to markers on LG1: Z5508, 15 recombinants/156 meioses and Z1705 (see text) in a mfn^tm124a map cross; Z1463, 1 recombinant/86 meioses in a mfn^tb241c map cross.

To estimate the map position of the tld gene, PCR was carried out on 94 zebrafish/mouse somatic cell hybrid (Ekker et al., 1996) and 93 radiation hybrid cell lines (Hukriede et al., in preparation). The following oligonucleotides were used: GCATGGACAGTTTG-GCTGC and CGCTGACATAAACAATGGGC, within the 3′ untranslated region (UTR; GenBank accession #AF027596). PCR conditions were as previously described (Chevrette et al., 1997).

Single F₂ mutant and wild-type embryos from a mfn^ty130a mapping cross line were tested for segregation of the mfn mutation to an RFLP in the tld 3′UTR. The 3′UTR was amplified by PCR using primers Tld3′a, TTGAGAGACTGAAGACAGGG and Tld3′b, CAGAAAG-TCAGTCAACAGCG, and the product cut with MspI.

About 50 mutant embryos were collected at 24 hours postfertilization (hpf) from crosses between heterozygotes of five mfn alleles (ty130a, tm124a, tf215a, tn217b, tb241c; Mullins et al., 1996). Control wild-type embryos from the TÜ strain (the strain in which the mutations were induced) were also collected. Total RNA was extracted with TRIzol Reagent (Gibco/BRL) according to the manufacturer’s instructions. First-strand cDNA was synthesized from total RNA with the Superscript Preamplification System for First Strand Synthesis (Life Technologies). Two to three independent PCR reactions were performed using the primer pairs Tld5′a/Tld5′b, Tld1/Tld2 and Tld3/Tld4 on wild-type and mutant embryo cDNA. The PCR was performed as follows: 94°C for 3 minutes; 40 cycles of 57°C for 1 minute, 70°C for 3 minutes, 94°C for 1 minute; 57°C for 1 minute and 70°C for 20 minutes.

Primers used:

Tld5′a AGGGAAATGGGCACGTTTGG Tld5′b AGCAGCAGATATAAGGAACCC Tld1ACACACATCTGGAGGTCTAGG Tld2ACACACACTCCTTAGATGGG Tld3AAGTTCTGGTCTCTACAGACAG Tld4TCAGAGAGCGCAAGACACC PCR products were subcloned using the TA-PCR cloning kit (Invitrogen) or the pGEM-T Easy Vector System (Promega) according to the manufacturer’s instructions. Clones from independent PCR reactions were sequenced using primers spaced the length of the gene. Sequences were analyzed using MacMolly software.

Our previous rescue results using overexpression of bmp ligand genes (Nguyen et al., 1998b) implicate two genes with dorsalized mutant phenotypes, somitabun and snailhouse, to act within a Bmp signaling pathway to establish ventral cell fates. Tld acts upstream of the Bmp ligands to cleave Chd and release Bmps. To examine whether snailhouse and somitabun act upstream or downstream of tld, we injected wild-type tld mRNA into 1- to 4-cell-stage mutant embryos and assayed for rescue of the mutant phenotype at 24-30 hpf. Overexpression of tld could not rescue somitabun or snailhouse mutant embryos (Table 1), suggesting that these genes act downstream of, or parallel to, tld in a Bmp signaling pathway. As expected, tld overexpression did not rescue swirl/bmp2b mutants (Table 1), since swirl/bmp2b is predicted to function downstream of tld.

We investigated whether three additional dorsalized mutants, piggytail, lost-a-fin and mfn (Mullins et al., 1996), could be rescued by overexpression of tld mRNA. We observed no rescue of piggytail or lost-a-fin mutant embryos (Table 1), indicating that these two genes do not act upstream of, or as, a tld metalloprotease. In contrast, tld overexpression rescued a majority of mfn mutant embryos to a wild-type or slightly ventralized phenotype (Table 1; Fig. 1), suggesting that the mfn gene functions upstream of, or encodes, a tld metalloprotease.

To determine if mfn could correspond to the tld gene, we examined the chromosomal locations of mfn and tld. We mapped the mfn mutation to a chromosomal position using simple sequence length polymorphic (SSLP) markers located throughout the genome (Knapik et al., 1996, 1998). Marker Z1705 on linkage group 1 (LG1) appeared linked to mfn in pooled F₂ mutant DNA (data not shown). Linkage analysis in individual F₂ mutant embryos placed mfn approximately 3 cM from marker Z1705 (5 recombinants in 156 meioses, Fig. 2A,B). Examination of markers distal and proximal to Z1705 suggests that the mfn gene lies between Z1705 and Z1463, 1 to 3 cM from Z1463 (see Materials and Methods).

To establish the chromosomal position of the tld gene, we used a collection of zebrafish/mouse somatic cell hybrid (Ekker et al., 1996) and radiation hybrid (RH) panels (N. A. Hukriede, L. Joly, M. Tsang, J. Miles, P. Tellis, J. A. Epstein, W. B. Barbazuk, F. N. Li, B. Paw, J. H. Postlethwait, T. J. Hudson, L. I. Zon, J. D. McPherson, M. Chevrette, I. B. Dawid, S. L. Johnson and M. Ekker, personal communication). Of 94 somatic cell hybrid lines tested, six were positive for the tld gene, five of which have one or more known markers on LG1 (Fig. 2B). Examination of LG1 markers suggests that tld is found in an interval that separates markers Z1463 and Z3705. Linkage to microsatellite markers from this chromosomal region using the RH panel indicated a strong linkage between tld and Z1463 (LOD=20; 4.0 cR). Based on a preliminary estimate of the average breakpoint frequency of this RH panel (Hukriede et al., personal communication), tld would be within a Mbp from Z1463. To directly determine the proximity of tld to the mfn mutation, we examined linkage between an RFLP in the tld gene and the mfn mutation. We found no recombinants between the RFLP and the mfn mutation in 160 meioses (Fig. 2C), confirming that the mfn mutation is very closely linked to the tld gene, likely within 1 cM. Thus tld is an excellent candidate for the gene mutated in mfn.

To determine if mfn is a mutation in the tld gene, we cloned and sequenced tld cDNA clones from five of the nine mfn alleles. The zebrafish tld gene is predicted to encode a 1023 amino acid secreted protein constituting an amino-terminal signal sequence and pro-domain, followed by a metalloprotease domain, five CUB domains and two EGF repeats positioned between the CUB domains (Fig. 2D; Blader et al., 1997). The CUB domains and the EGF repeats are both thought to mediate protein-protein interactions (reviewed in Davis, 1990). Here we report on the identification of mutations in the five mfn alleles examined.

Nonsense mutations were identified in three mfn alleles. In the mfn^tm124a allele, a point mutation changes amino acid 181 from a glycine residue to a stop codon (Fig. 2D). This mutation occurs in the amino-terminal region of the metalloprotease domain, resulting in a predicted truncated protein missing both the functional metalloprotease and the protein-protein interaction domains. We expect this mutation is a strong loss-of-function allele, very likely a null mutation.

Stop codons were also found within the first and second CUB domains in mfn^tb241cand mfn^tf215a, respectively (Fig. 2D). Although the active site of the tld protein, the metalloprotease domain, is present in the predicted mutant proteins, the putative protein-protein interaction domains are mostly absent in these two alleles. Therefore, these proteins may not recognize and cleave their target, Chd.

We identified missense mutations within the mfn^tn217b and mfn^ty130a alleles (Fig. 2D). The mfn^tn217b mutation alters a leucine residue to proline at position 429 in the first CUB domain. Proline residues are known to disrupt certain secondary structures of proteins. This mutation occurs in a region proposed to form a β-strand (Bork and Beckmann, 1993) and may alter the Tld protein conformation. The mfn^ty130a allele changes a highly conserved cysteine in the metalloprotease domain to a tryptophan residue. In the Astacin protease, this cysteine forms an intramolecular disulfide bridge with a cysteine located 20 amino acids carboxy-terminal to it (Stöcker et al., 1993; Bode et al., 1992). Since these cysteine residues are invariant within the astacin protease family, it is likely that the disulfide bridge is present in all family members (Stöcker et al., 1993). The mfn^ty130a mutation would disrupt this disulfide bond, thus altering the Tld protein conformation.

We have further characterized the mfn mutant phenotype to ascertain the function of tld/mfn in the zebrafish embryo. The recessive mfn phenotype is the weakest (class 1) of five dorsalized phenotypes identified in a large-scale mutant screen (Mullins et al., 1996). The phenotype is first morphologically discernible in some mfn mutant embryos at the bud stage (end of gastrulation) by a slight thickening of cells in the tail bud or a slightly ovoid shape of the embryo (Fig. 3A,B). By the 3-somite stage, all mutant embryos exhibit an abnormally shaped tail bud (Fig. 3C,D), which is reduced in length (arrows) and protrudes away from the yolk (asterisk). This phenotype contrasts with the ventralized mutant phenotypes of chordino and mercedes embryos, which display a broader and flattened tail bud (Fisher et al., 1997; Hammerschmidt et al., 1996a). During later stages of somitogenesis, the tail bud of mfn mutant embryos continues to protrude abnormally away from the yolk (data not shown, Mullins et al., 1996).

At 24 hpf, mfn mutants display a range of phenotypes, which we divided into five categories (A-E) based on the degree to which ventral tail tissues are reduced. We established these categories to investigate the relative strengths of the different mfn alleles. The strongest mfn phenotype, category E, exhibits a complete loss of the ventral tail fin, frequently associated with reductions in caudal tail vasculature and ventral somitic mesoderm (Fig. 3H). Category D mutant embryos display a loss of more than half of the ventral tail fin (Fig. 3F), while in category C mutants half or less than half of the fin is absent (Fig. 1A). Category C and D mutant embryos occasionally exhibit a bifurcation of the tail (Fig. 3G). Category B mutants display an obvious notch in the ventral tail fin, while category A mutants exhibit a subtle reduction or small split in posterior ventral tail fin tissue (data not shown).

We used the criteria above to categorize mutant embryos of the nine mfn alleles and compare the strengths of their mutant phenotypes. We found that mutants of most alleles, including the presumptive null allele, mfn^tm124a, exhibit all categories (A-E) of phenotypes but to different degrees (Fig. 4). Our results suggest that there may be stronger (mfn^ty130a andmfn^tm124a) and weaker (mfn^tt203 and mfn^tf211a) mfn alleles. However, since a complete range of phenotypes is observed in a putative null allele, the significance of the different phenotypic strengths for the various alleles is difficult to ascertain. The range of defects for a particular allele could reflect different degrees of compensatory regulation in mutant embryos. The strength of the phenotype could also be influenced by differences in genetic background. Further analysis of the strength of the mfn/tld mutations will require a biochemical determination of residual Tld activity. Interestingly, mutants of the Tld substrate, Chordin, also display a range of ventralized phenotypes at 24 hpf (Hammerschmidt et al., 1996a). Together with our analysis of mfn, this suggests some compensatory regulation occurring in modulating Bmp signaling levels.

Fish homozygous mutant for mfn are semiviable. Some mutant embryos survive to adulthood and appear nearly wild type (Fig. 3I) or are shorter in length and lack all or part of their tail and anal fins (Fig. 3J). In crosses with mfn homozygous females, no maternal-effect phenotype was detected.

To investigate whether the ventral somitic mesoderm, vasculature and tail fin defects of mfn mutant embryos are due to a reduction of these tissues rather than their abnormal morphology, we examined several genes specifically expressed in these tissues. The msxD gene is expressed in the prospective median fin fold tissue by the 6-somite stage, specifically labeling cells at the caudal tip of the embryo (Akimenko et al., 1995). In 8-somite-stage mfn mutants, msxD expression was reduced or absent (Fig. 5A,B). At the 14-somite stage, the msxB and dlx3 genes are also expressed in cells of the presumptive median fin fold (Akimenko et al., 1994, 1995). In mfn mutants, a gap in expression is observed caudal to the tail bud (Fig. 5C,D, data not shown). At 24 hpf, the expression patterns of these genes correspond precisely to the phenotypes observed in live mutant embryos (compare Fig. 3F-H with Fig. 5F-H, respectively). Thus, the reduction in the ventral tail fin at 24 hpf in mfn mutant embryos is apparent as early as the 8-somite stage by a loss of prospective fin fold tissue.

Many mfn mutants fail to establish normal blood circulation. To investigate whether a loss of vasculature is associated with this defect, we examined the expression of flk1, a marker for vascular endothelial cells (Fouquet et al., 1997; Liao et al., 1997; Thompson et al., 1998). The degree to which the flk1 expression pattern was altered typically corresponded to the severity of the mfn mutant phenotype. At 24 hpf, weak mfn mutant embryos displayed normal flk1 expression (data not shown). An aberrant flk1 expression pattern was observed in some mfn mutants (Fig. 5J), which may reflect an abnormal morphology and/or reduction in flk1 expression. Other mfn mutant embryos displayed a reduction of flk1 expression in caudal regions of the tail (Fig. 5K), indicating that the loss of vasculature observed in some mfn mutants is due to a reduction in the number of endothelial precursor cells.

We analyzed myoD expression (Weinberg et al., 1996) at 24 and 32 hpf to examine tail somite formation in mfn mutant embryos. Weak and most of the moderate strength mfn mutant embryos (category A-D) display a normal myoD expression pattern (data not shown). Moderate mfn mutant embryos (category C, D) with a bifurcated tail tip frequently contain somitic tissue in both forks (Fig. 5N). Stronger mfn mutant embryos (category E with kinked tails) usually lack or exhibit a reduction in ventral somitic tissue posterior to the bent notochord (Fig. 5M and data not shown). The partial loss of three ventral tissue types of the tail, somitic mesoderm, blood vessel endothelial cells and fin tissue suggests that mfn does not act in the specification of a particular cell type, but rather functions in establishing the pattern of ventral tail tissues.

Vasculature and ventral tail fin tissue have been fate mapped to ventral regions of the blastula or gastrula embryo (Lee et al., 1994; Woo and Fraser, 1998). Since these tissues are reduced or absent in mfn mutant embryos, we investigated whether these losses are reflected at earlier stages in development in the altered expression of markers of ventral regions. We examined the expression of eve1 (Joly et al., 1993) and bmp4 (Chin et al., 1997; Nikaido et al., 1997), which are expressed in ventral (or posterior) tissue from mid-gastrulation through somitogenesis stages. Analysis of eve1 and bmp4 expression in mfn mutant gastrula (80-90% epiboly) did not reveal a clear reduction of ventral cell fates. We detected a smaller domain of eve1 and bmp4 expression in a small fraction of embryos (<10%) in some broods. We genotyped these embryos and found that embryos with the most severely reduced expression correlated to mfn homozygous mutant embryos, but we also found a small fraction of wild-type embryos with slightly reduced domains (data not shown).

By the bud stage, however, many mfn mutant embryos display a reduction in eve1 expression (Fig. 6A,B) and the posterior (or ventral) bmp4 expression domain (Fig. 6C,D) within the tail bud. These markers, which continue to be expressed in the extending tail bud during somitogenesis stages, remain reduced in all mfn mutant embryos (data not shown). Thus, mfn/tld may act during late gastrulation and is clearly functioning at the end of gastrulation. Moreover, these data indicate that tld specifies tissues of ventral character within the tail bud.

We examined the expression pattern of tld from early gastrulation to 24 hpf to investigate how tld expression may account for its role in ventral or posterior cell fate specification. From the shield to bud stages, tld is expressed at a low level with a spotty appearance in superficial cells in both dorsal and ventral animal pole regions (Fig. 6E,F and data not shown). At 55% epiboly, a clearing is observed in a dorsal quadrant, except for expression close to the edge of the margin (Fig. 6E). At 70% epiboly, tld is expressed uniformly in ventral regions in cells below the superficial clusters and strongly around the margin (Fig. 6G), similar to earlier stages (Fig. 6E). At 80-90% epiboly, the ventral domain no longer extends to the margin, but instead lies in the animal half of the embryo (Fig. 6H). As previously reported (Blader et al., 1997), at the bud stage tld expression is observed transiently along the edge of the neural plate and the marginal tld expression domain now appears to lie within the tail bud (Fig. 6I), where it continues to be strongly expressed during somitogenesis stages (Fig. 6J,K). By 20 somites, tld is expressed in prospective vascular tissue (Fig. 6J), similarly to flk1. Thus, the expression of tld within the margin late in gastrulation and within the tail bud is consistent with a role for tld in the establishment of ventral tail cell fates.

Bmp2 and Bmp4 gene expression is under the control of an autoregulatory feedback mechanism (Jones et al., 1992; Kim et al., 1998; Kishimoto et al., 1997; Metz et al., 1998; Nguyen et al., 1998b), while chordin expression is negatively regulated by Bmp signaling (Miller-Bertoglio et al., 1997; Schulte-Merker et al., 1997). To examine whether tld gene expression is regulated similarly to the Bmp ligand genes, we examined tld expression in chordino and swirl mutant embryos. In chordino mutants at bud stage and throughout somitogenesis stages, the tail bud expression domain of tld is broadened relative to wild type (Fig. 6K,L, and data not shown). In swirl/bmp2b dorsalized mutant embryos, tld expression is reduced at bud and later stages (Fig. 6M, data not shown). Thus, tld gene expression at these stages is regulated positively by bmp2b/swirl and negatively by chd.

We investigated the spatial relationships of the chd, tld and bmp4 expression domains at the bud stage, when we know mfn/tld functions, to examine possible functional relationships between these genes. Double in situ hybridizations of chd, tld and chd, bmp4 at the bud stage in wild-type embryos shows that tld and bmp4 expression is excluded from the more anteriorly located chd expression domain (Fig. 6N,O). Moreover, tld expression lies adjacent to the chd expression domain, while a prominent gap is found between the bmp4 and chd expression domains. The bmp4 gene expression domain extends further laterally and posteriorly than does tld (data not shown) and tld expression lies closer to the anterior side of the tail bud than does bmp4. Thus tld mRNA is in the correct location for Tld to play a role in modulating presumptive Bmp4 activity levels through cleavage of Chd.

We examined the involvement of tld/mfn in shaping the chd and bmp4 expression domains. At the bud stage, chd expression is expanded posteriorly and laterally in some mfn mutants (Fig. 6O,P), while tld and bmp4 expression domains are reduced in all mutant embryos (Fig. 6O,P and C,D, respectively). At the 7-somite stage, an expansion of chd is also observed in some, but not all, mfn mutants, which likely reflects the strength of the mfn mutant phenotype. It is possible that the range of mfn mutant phenotypes is caused, in part, by the degree to which chd gene expression is expanded. Altogether, these results are consistent with Tld positively modulating Bmp signaling levels, which is then reflected in the positive regulation of bmp4 and tld gene expression and negative regulation of chd expression.

We have shown that mini fin encodes the tolloid gene, based on three lines of evidence. First, overexpression of tld mRNA specifically rescues mfn mutants to a wild-type phenotype. Second, tld and mfn are very closely linked in the zebrafish genome. Lastly, three stop codon and two missense mutations were found in the five mfn alleles analyzed, verifying that mfn encodes tld.

The mfn phenotype is characterized by a partial to complete loss of the ventral tail fin, and a reduction in caudal tail vein and somitic mesoderm in strong mutant embryos. The reduction in three different ventral tail tissues suggests that mfn is not involved in the differentiation or specification of a particular cell type, but rather may act in the establishment of the general pattern of the tail. Vasculature and ventral tail fin tissues have been fate mapped to ventral regions of blastula (Lee et al., 1994) and mid-gastrula (Woo and Fraser, 1998) embryos, respectively, suggesting that ventral cell fate specification is affected in mfn mutant embryos. A reduction in the expression of two markers of ventral tissue at the end of gastrulation in mfn mutants supports such a role for mfn/tld. Additional results and observations, as discussed below, add strength to the hypothesis that mfn/tld normally is involved in the specification of ventral cell fates at the end of gastrulation in the zebrafish embryo, likely by increasing Bmp activity.

Phenotypes identical to those of mfn mutants are observed in other dorsalized mutant embryos, including dominant phenotypes of swirl/bmp2b (Mullins et al., 1996). Characterization of homozygous swirl mutant gastrula shows that swirl is involved in the specification of nearly all ventral cell fates (Nguyen et al., 1998b). The presumed reduction in Bmp2b activity in swirl heterozygotes produces a partially penetrant phenotype, suggesting that Bmp2b activity is at a threshold level for normal development in these mutant embryos. Interestingly, this partial loss of bmp2b gene activity results in a weak dorsalization identical to the mfn/tld phenotype described here. Since Tld can enhance Bmp activity through the cleavage and inactivation of Chd (Blader et al., 1997; Marqués et al., 1997; Piccolo et al., 1997), loss-of-function mutations in tld are expected to decrease Bmp activity. The similarity in the dominant loss-of-function swirl/bmp2b phenotype to the mfn/tld mutant phenotype strongly supports a role for mfn/tld in increasing Bmp activity in ventroposterior cells, which then specifies ventral tail cell fates.

In frog and fish embryos, bmp2 and bmp4 gene expression is under the control of an autoregulatory mechanism (Jones et al., 1992; Kim et al., 1998; Kishimoto et al., 1997; Metz et al., 1998; Nguyen et al., 1998b). Chd can negatively regulate bmp gene expression (Hammerschmidt et al., 1996b; Kim et al., 1998; Metz et al., 1998), presumably by inhibiting Bmp activity and interrupting the autoregulatory feedback loop. In a reciprocal manner, Bmp activity can negatively regulate chd gene expression (Miller-Bertoglio et al., 1997; Schulte-Merker et al., 1997). Consistent with a role for Tld in modulating Bmp activity, we observe a reduction in the domain of bmp4 gene expression and an expansion of chd gene expression in mfn/tld mutants. Tld likely regulates bmp gene expression by cleaving Chd, thus increasing Bmp activity, which would then induce additional bmp transcription through the autoregulatory loop. The expression of tld may also be part of this feedback regulatory loop, since tld gene expression is expanded in chordino mutants and reduced in swirl mutants.

Our analysis of the mfn/tld mutant phenotype indicates that Mfn/Tld plays little or no role during early gastrulation in the zebrafish, although it is expressed at this time. We discuss two alternative hypotheses for this lack of mfn/tld function. The first hypothesis is that a second tld-related gene in zebrafish exists and acts redundantly to mfn/tld. It is likely that additional tld-related genes are present in the zebrafish, since two tld-related genes have been identified in the mouse, human and frog. In Xenopus, for example, Xolloid (Xld) and one long splice variant from the Bmp1 locus, called Xtolloid (XTld), have been reported (Goodman et al., 1998; Lin et al., 1997). Overexpression of a dominant-negative form of Xolloid (DN-Xld) moderately dorsalizes anterior tissue of the Xenopus embryo (Piccolo et al., 1997), suggesting an early function for Xld during gastrulation (Piccolo et al., 1997). Xld, however, is not detectable by whole-mount in situ hybridization during gastrulation, arguing against it functioning at these stages (Goodman et al., 1998). At least one of the transcripts from the Bmp1 locus is expressed during gastrulation (Goodman et al., 1998; Lin et al., 1997) and could be the activity inhibited by the DN-Xld. Overexpression of a DN-tld in the zebrafish causes a low frequency of predominantly weak dorsalization phenotypes (Blader et al., 1997), inconsistent with mfn/tld or another tld gene playing a significant role early in gastrulation. However, it is possible that the zebrafish DN-tld is not as effective as the frog DN-Xld, which contains a different mutation. The mechanisms by which the DN-Tld/Xld act are not understood, leaving open the possibility that Bmp activity is reduced by the DN-Xld, independent of normal Tld function.

Further experimental analysis is required to establish whether Tolloid inhibits Chordin activity during early gastrulation.

An alternative hypothesis for the lack of detectable mfn/tld function early in gastrulation is that neither Mfn/Tld nor another Tld-related protein inhibits Chordin activity at these stages. It is possible that the large circumference of the zebrafish embryo is sufficient to generate the necessary Bmp activity gradient through the presumed diffusion of dorsally expressed Chordin and other Bmp antagonists into ventral regions. A corollary to this hypothesis could explain the more severe defects observed in the fly tld mutant compared to the zebrafish mfn mutant. Null mutant tld embryos in the fly are lethal and exhibit a moderate ventralization, affecting the dorsal ∼25% of the fate map (Arora and Nüsslein-Volhard, 1992), compared to the weak semiviable dorsalization of the mfn mutant. The smaller circumference of the fly embryo, ∼630 μm (Ashburner, 1989) compared to 2200 μm in the zebrafish (Kimmel et al., 1995), may necessitate a more prominent role for the fly Tld in repressing Sog function in dorsal regions. Sog appears to diffuse at least 12 to 15 cell diameters (Biehs et al., 1996), or up to 120 μm, from its ventral expression domain and may diffuse to the most dorsal point of the fly embryo, ∼135 μm. A function of the fly Tld, which is expressed in the dorsal 40% of the embryo, may be to restrict Sog repressive action in the dorsalmost regions in order to generate the necessary Bmp activity gradient. In zebrafish, this gradient may be established during early gastrulation solely by diffusion of Chd, and possibly other Bmp antagonists, independent of Tld function.

The first consistent alteration in mfn/tld mutant embryos is apparent at the time of tail bud formation at the end of gastrulation (the bud stage), indicating that tld/mfn functions at least by this stage. Since tld is strongly expressed within the posterior tail bud, we hypothesize that tld/mfn functions within the tail bud itself in the specification of ventral tail cell fates. Similarly, Xld may also play a role in ventral cell fate specification within the Xenopus tail bud, since its expression is specifically localized to prospective tail cell types at neurula and later stages (Goodman et al., 1998; Piccolo et al., 1997).

Previous studies based on analysis of cell movements and gene expression domains within the tail bud suggest that tail formation results from a modified form of gastrulation in the tail bud (Catala et al., 1995; Gont et al., 1993; Kanki and Ho, 1997; Knezevic et al., 1998; Pasteels, 1943; Tucker and Slack, 1995). Anterior and posterior domains of the tail bud express genes that are dorsally and ventrally restricted in gastrulation, respectively (Gont et al., 1993). Fate map data also support a restriction of dorsally derived gastrula cell fates to the anterior tail bud (Kanki and Ho, 1997). Moreover, cell transplantation experiments between anterior and posterior tail bud regions indicate that cells of the tail bud are not determined at the bud stage (Kanki and Ho, 1997; J. Kanki, personal communication). Hence, the juxtaposed expression domains of chd to bmp4 and tld within the tail bud suggests that these genes may function in patterning prospective tail tissues from the bud stage onwards.

We propose that Tld/Mfn function is crucial during late gastrulation and tail bud stages to repress Chd activity. At these stages, the ventralmost vegetal cells of the late gastrula and tail bud lie in close proximity to the dorsal-expressing Chordin cells. It is conceivable that at these stages Chordin could diffuse, if unimpeded, to the ventralmost regions and inhibit Bmp signaling (Fig. 7). We hypothesize that the high expression of tld in the margin of the late gastrula and within the tail bud acts to limit Chd activity, thereby increasing Bmp activity, which then specifies ventral tail cell fates. When functional Tld/Mfn is reduced in the embryo, higher levels of Chd are expected to cause a reduction in Bmp activity. This would then affect the transcriptional autoregulatory loop of bmp and chd gene expression, further decreasing Bmp activity and expanding the domain of Chd antagonism. Consequently, the Bmp activity levels required for ventral tail cell fate specification are not attained, resulting in the loss of ventral tail tissues.

We cannot yet account for two potential modes of spatial regulation of Tld activity. First, Tld is processed from its zymogen form to an active protease (Marqués et al., 1997; Piccolo et al., 1997). In the fly embryo, only a small fraction of Tld is processed by a yet unknown mechanism (Marqués et al., 1997). Likewise, in vertebrate embryos, it is not known how Tld is processed and if it is regulated in some manner. Second, in tissue culture experiments, the fly Tld cleaves Sog much more efficiently when bound to Dpp (or other BMPs; Marqués et al., 1997; Nguyen et al., 1998a), while Xld and the zebrafish Tld can act on Chd alone (Blader et al., 1997; Piccolo et al., 1997) or in complex with BMPs (Piccolo et al., 1997). The kinetics of Tld cleaving Chd versus a Chd-Bmp complex and the endogenous requirements for cleavage have not been established. Hence, we make no assumption as to where in the margin or tail bud Tld cleaves Chd.

Graded Bmp activity appears to be established through the action of specific Bmp antagonists. It is clear that additional molecules, such as Tld, are also critical in modulating Bmp activity levels. The precise role played by Tld in the zebrafish and fly in regulating the Bmp activity gradient remains unclear. Its primary function may be to limit Chd/Sog activity in regions where the entire bmp expression domain lies near to the chd/sog expression domain, and Chd/Sog may diffuse across most or all of the bmp expression domain to inhibit most Bmp signaling (e.g. in the the tail bud of the zebrafish embryo and the dorsoventral axis of the fly embryo). Alternatively, it may be required to shape or refine the gradient in a particular manner to generate the readout necessary to specify different cell fates in particular domains. A further role of Tld may be in the generation of Sog/Chd cleavage products, which could play additional roles in development as suggested by Marqués et al. (1997). These and other questions remain to be resolved.

We thank Patrick Blader and Uwe Strähle for generously providing tolloid injection and in situ constructs, and Mark Fishman and Ela Knapik for kindly providing SSLP marker information prior to publication. We also thank Alvin Chin, Valarie Miller-Bertoglio, Marnie Halpern, Marie-Andrée Akimenko, Eric Weinberg, Jean-Stefan Joly, Didier Stainier and Len Zon for in situ probes, Dan Kessler, Steve DiNardo, Michael Granato, Dan Wagner, Bettina Schmid, Vu Nguyen and Keith Mintzer for critical review of the manuscript, and Andrea Payne and Chapell Miller for care of the fish. We thank Bettina Schmid for generation of homozygous mfn fish and Vu Nguyen for some expression data in swirl mutants. Special thanks to Lucille Joly for help with mapping, Neil Hukriede for assistance with RH data analysis, and John Kanki for sharing unpublished observations. This work was supported by a grant from the NIH to M. C. M. (R01-GM56326), funding from the Medical Research Council of Canada to M. E., and training grants (T32 HD07305 and T32 HD07516 01) to S. A. C.