The smad5 mutation somitabun blocks Bmp2b signaling during early dorsoventral patterning of the zebrafish embryo

In amphibia, early D-V patterning of the embryo is a multi-step process governed by maternally and zygotically supplied signaling proteins. Under maternal control, a coarse D-V pattern is set up in the developing mesoderm, consisting of ventral mesoderm spanning most of the marginal region of the blastula, and dorsal mesoderm in a rather small dorsal marginal domain. This initial pattern is subsequently refined by zygotic ventralizing and dorsalizing signals, leading to the specification of intermediate fates. Several signaling proteins have been identified which might be involved in this refinement process: the Bone Morphogenetic Proteins 2 and/or 4 (Bmp2/4), members of the TGFβ superfamily of growth factors, which have ventralizing activities, and their antagonists Chordin, Noggin and Follistatin (for reviews, see Thomsen, 1997; Heasman, 1997). Recent evidence indicates that Bmp2/4 function as instructive morphogens which determine positional identities along the entire D-V axis in a dose-dependent fashion (Dosch et al., 1997), while Chordin, Noggin and Follistatin attenuate this ventralizing activity on the dorsal side of the embryo by physical interaction with Bmp proteins (Piccolo et al., 1996; Zimmerman and Harland, 1996), and thereby establish the putative morphogenetic D-V gradient of Bmp2/4 activity.

Genetic evidence revealing the requirement of Bmp2b and its antagonist Chordin during D-V patterning has recently been obtained in the zebrafish. In large-scale mutant screens at least eight complementation groups have been identified defining eight genes required for dorsoventral development (Mullins et al., 1996; Hammerschmidt et al., 1996a). The phenotype of the strongest of the dorsalized mutants, swirl, is caused by null mutations in bmp2b, one of the two bmp2 genes known in zebrafish (Martínez-Barberá et al., 1997; Kishimoto et al., 1997; Nguyen et al., 1998), while the ventralization of dino mutants is due to a null mutation in the zebrafish chordin gene, therefore designated chordino (Schulte-Merker et al., 1997).

In the fruitfly Drosophila melanogaster, embryonic D-V patterning is also regulated by structural homologues of Chordin and Bmp2/4, called Short gastrulation (Sog) and Decapentaplegic (Dpp) respectively. Screens for dominant maternal enhancers of dpp identified two genes, mad (mothers against decapentaplegic) and medea, which enhance the D-V phenotype of dpp mutants (Raftery et al., 1995). mad is also required for later organogenic processes acting downstream of the Dpp receptors to mediate Dpp signaling in target cells (Newfeld et al., 1997; Wiersdorff et al., 1996).

Similar functions as transducers of TGFβ signaling were demonstrated for the Mad-related Sma proteins in the nematode C. elegans and Smad proteins in vertebrates (for reviews, see Kretschmar and Massagué, 1998; Attisano and Wrana, 1998). In vertebrates, two Smad proteins, Smad1 and Smad5, show particularly high homology to Drosophila Mad. Together with Smad8, they are involved in the mediation of ventralizing signaling by Bmp2/4, the vertebrate homologues of Dpp (Graff et al., 1996; Thomsen, 1996; Liu et al., 1996; Hoodless et al., 1996; Suzuki et al., 1997), while the more distantly related Smad2 and Smad3 proteins mediate signaling by other TGFβ proteins. Smad4 is identical to the tumor suppressor DPC4 (Hahn et al., 1996) and shows high homology to the Drosophila Medea protein (Hudson et al., 1998; Wisotzkey et al., 1998; Das et al., 1998). It appears to be a common mediator of all TGFβ superfamily members, interacting with both Smad1- and Smad2-type proteins (Lagna et al., 1996; Zhang et al., 1997). Biochemical evidence indicates that Smad1 and Smad2 proteins form homo-trimers which are recruited and phosphorylated at their C terminus by ligand-activated Bmp- and TGFβ-receptors, respectively. After phosphorylation, these ‘receptor-activated’ Smad trimers associate with Smad4 trimers, translocate to the nucleus and participate in transcriptional complexes (for reviews, see Kretschmar and Massagué, 1998; Attisano and Wrana, 1998). While the ‘receptor-activated’ Smads are all positive transducers of TGFβ signaling, two other, more distantly related Smad proteins, Smad6 and Smad7, antagonize TGFβ signaling (reviewed by Whitman, 1997).

The requirement of both Smad2 and Smad4 during early development and mesoderm formation has been recently demonstrated using gene targeting in the mouse (Sirard et al., 1998; Waldrip et al., 1998; Nomura and Li, 1998). Here, we describe a zebrafish mutant in smad5, a transducer of Bmp2/4 signaling which is active during early steps of embryonic D-V patterning.

Polymerase chain reaction was carried out using a zebrafish gastrula cDNA library (kind gift from David Grunwald) as template and previously reported degenerated primers (Graff et al., 1996) to amplify a 180 bp zebrafish smad5 cDNA fragment which was used to screen the gastrula cDNA library at high stringency. For further experiments we used a 2.6 kb clone with 365 bp of 5′ UTR, 1395 bp of coding region, giving rise to a predicted protein of 465 aa, and approx. 850 bp of 3′ UTR, including a polyadenylation site and a polyA tail. An in-frame TAA 23 triplets upstream of the start codon indicates that the coding region is complete (sequence not shown, accession number AF127920).

smad5 was mapped in F₂ offspring of a Tue X WIK reference cross as described by Rauch et al. (1997), using a smad5 RFLP (restriction fragment length polymorphism) identified between the P₀ parents of the reference cross.

The sbn mutation was mapped via its dominant zygotic effect which was fully penetrant, causing a mild dorsalization. A male sbn carrier of the Tue background was mated to a WIK wild-type female and heterozygous F₁ males were crossed to TL wild-type females (Haffter et al., 1996). The SSLP markers linked to sbn were indistinguishable in TL and WIK, but different in the Tue (sbn) line. Thus, recombinants between the SSLP markers and the sbn mutation were scored among the phenotypically wild-type F₂ siblings as heterozygous for the SSLP markers. In addition, a direct linkage analysis of smad5 and sbn was carried out using a tetra-PCR approach as described below.

RT-PCR with total RNA isolated from the offspring of two sbn^tc24 carriers was performed to amplify three partially overlapping fragments spanning the complete smad5 open reading frame. The resulting fragments were cloned using a TA cloning kit (Invitrogen); 14 independent clones of each fragment were sequenced, and 7/14 of the clones with the 3′ fragment showed a single missense mutation (ACA→ATA).

DNA from single embryos or amputated tail fins of adult fish was prepared according to the method of Westerfield (1994). Genotyping was performed by allele specific tetra-PCR (Ye et al., 1992), identifying wild-type, heterozygous or homozygous mutant genotypes in a single PCR reaction with four primers. Two reverse primers containing the nucleotide of the wild-type or mutant cDNA sequence at their 3′ end were combined with outer primers to give rise to wild-type and mutant-specific DNA fragments of different sizes.

For sbn^tc24smad5, we used: 94°C (3 minutes), 10 cycles of 94°C (30 seconds), 65°C (45 seconds), 72°C (1 minute), 25 cycles of 94°C (30 seconds), 48°C (45 seconds), 72°C (1 minute), followed by 72°C (10 minutes). Outer sense primer (GCAACTACCATCATGGCTTTC), outer antisense primer (AAAACAAAGGCTTCCTCCCAGG), wild-type sense primer (ACAGACAAGATGTGAC), mutant antisense primer (AGCAGGGGGTGCTTA). For the bmp2b swr^ta72 mutation we used: 94°C (3 minutes), 10 cycles of 94°C (30 seconds), 57.5°C (45 seconds), 72°C (1 minute), 20 cycles of 94°C (30 seconds), 48°C (45 seconds), 72°C (1 minute), followed by 72°C (10 minutes). Outer sense primer (GCTATCATGCTTTCTACTG), outer antisense primer (GTTTTGTTCATTCAACATAAAT), wild-type sense primer (TGT-GGGTGCCGATGA), mutant antisense primer (TTGGGGA-GATTGTTCC).

Total RNA from staged embryos (Westerfield, 1994) was isolated using Trizol LS reagent (Gibco/BRL). Northern blot analysis was carried out (Bauer et al., 1998) using randomly labeled smad5 or smad1 cDNA as probe.

For the transcription construct p64TS-smad5, the zebrafish wild-type smad5 coding region containing an upstream Kozak sequence (ACC) was amplified via PCR and cloned into pSP64TS (Krieg and Melton, 1984). To generate p64TS-smad5(sbn), a 325 bp BglII/SpeI fragment of smad5 cDNA from sbn^tc24 mutant embryos which carries the C→T substitution was used to replace the corresponding fragment of p64TS-WTsmad5. pCS2-smad1(sbn) and pCS2-smad2(sbn) encoding zebrafish Smad1 and zebrafish Smad2 with the sbn-specific Thr→Ile mutation were generated via a PCR-based site-directed mutagenesis strategy essentially as described by Ho et al. (1989). To generate pCS2-hsmad4, a HindIII (blunt-ended)-EcoRI fragment of human SMAD4 (Zhang et al., 1996) was cloned into StuI-EcoRI digested pCS2+ (Rupp et al., 1994).

For the expression constructs pXeX-smad5, pCSKA-smad5, pXeX-smad5(sbn) and pCSKA-smad5(sbn), the coding region of wild-type and mutant smad5 was amplified and cloned into the BamHI site of pXeX (Johnson and Krieg, 1994) or pCSKA (Harland and Misher, 1988).

For synthesis of capped mRNA, all p64TS constructs were linearized with SmaI and all pCS2 constructs with NotI, followed by in vitro transcription with SP6 RNA polymerase using the Ambion Message Machine kit. Injections of mRNA and plasmid DNA were carried out as described by Hammerschmidt et al. (1998).

For Xenopus experiments, synthetic wild-type smad5, wild-type smad1, smad5(sbn) and/or smad1(sbn) RNA was injected cell-by-cell into all four animal blastomeres at the 8-cell stage. At stage 9, animal caps were explanted and cultured until sibling embryos reached stage 11.5. Intact caps were harvested and processed for RT-PCR analysis as previously described (Bouwmeester et al., 1996).

The following zebrafish strains were used: as wild type, TL (Haffter et al., 1996) and Ekkwill (Knapik et al., 1998), and for mutant analyses, the bmp2b allele swirl (swr^ta72) and the smad5 allele somitabun (sbn^tc24; Mullins et al., 1996). Cell transplantations were carried out essentially as described by Ho and Kane (1990). Whole-mount in situ hybridization with digoxigenin-labeled RNA probes and photography were according to Hammerschmidt et al. (1996a).

The zebrafish smad5 cDNA

A full length zebrafish cDNA, designated zebrafish smad5, was cloned. Its deduced amino acid sequence shows 91.6 % and 91.4% identity to human and mouse Smad5, 87.9%, 87.4% and 87.0% identity to human, mouse and Xenopus Smad1, respectively, and 77.5% identity to rat Smad8, indicating that it is the Smad5 orthologue.

To investigate whether the phenotype of any of the isolated zebrafish mutants was caused by mutations in the smad5 gene, the smad5 cDNA and mutations causing D-V phenotypes were mapped relative to SSLP markers. Mapping with a StyI RFLP in 75 F₂ individuals of the Tue x WIK reference cross positioned zebrafish smad5 into linkage group 14 (Knapik et al., 1998), 15.9 cM from the SSLP marker z4592 (LOD 10.2), 15.7 cM from z4291 (LOD 10.4) and 2.3 cM from z3290 (LOD 16.0). Of the five investigated dorsalizing mutations (swirl, somitabun, snailhouse, lost-a-fin, minifin; Mullins et al., 1996) one, somitabun (sbn^tc24), mapped to the same region, 5.2 cM from z4592 (LOD 44.6) and 4.4 cM from z4291(LOD 50.3). The genetic distances of sbn to the investigated SSLP markers are shorter than for smad5. This is expected since the sbn mapping strategy only measures male recombination rates, while mixed male and female recombination was used to map smad5. The sex-averaged zebrafish map of SSLP markers is reduced in size compared to the zebrafish RAPD map, solely based on female meiosis (Knapik et al., 1998; Johnson et al., 1996).

For a direct linkage analysis between the sbn mutation and the smad5 gene, we designed a tetra-PCR assay that allowed us to distinguish between wild-type and sbn smad5, taking advantage of the change of a single nucleotide in the smad5 coding region of sbn mutants (Fig. 1B; see below). Upon genotyping, 50 of 50 embryos showing the zygotic dominant sbn phenotype were smad5 heterozygotes, while 50 of 50 wild-type sibling embryos contained only the smad5 wild-type allele. In addition, in genotyping experiments of adult males and females, 52 of 52 sbn carrier fish were heterozygous for the smad5 mutation, while 50 of 50 non-carrier silbings were smad5 wild type. Thus, the smad5 gene and the mutation causing the zygotic dominant effect of sbn are <0.5 cM apart (no recombination in 202 meioses).

To investigate the molecular nature of the sbn^tc24 mutation, smad5 cDNA was amplified from sbn mutant embryos via RT-PCR. Sequencing revealed a single C→T change at nucleotide position 1286 of the smad5 coding region which leads to a Thr→Ile change at amino acid position 429 in the L3 loop region of the Smad5 MH2 domain. This threonine is conserved in all currently known Smad proteins except the inhibitory Drosophila protein Dad (Tsuneizumi et al., 1997) and Smad4, a common transducer of signaling by members of the TGFβ superfamily (Hahn et al., 1996) (Fig. 1A). The L3 loop has been shown to be involved in the interaction of ‘receptor-activated’ Smad proteins with their respective receptors and with Smad4 (Shi et al., 1997; Lo et al., 1998).

To investigate the strength of the sbn^tc24 allele of zebrafish smad5, misexpression experiments in Xenopus animal cap explants were carried out, comparing the ventral mesoderm-inducing potentials of wild-type and sbn^tc24 Smad5. Animal caps injected with wild-type smad5 mRNA contained significant amounts of Xhox3 mRNA (a ventral marker) and Xbra RNA (a pan-mesodermal marker). But no such transcripts could be detected by RT-PCR after injection of sbn^tc24smad5 RNA, so its protein has retained very low or no ventralizing activity (Fig. 1C). Co-expression experiments to ask whether the smad5 mutation also accounts for the dominant negative effect of sbn (see below) revealed that the ventralizing effect of wild-type smad5 RNA was inhibited by about 80% by equal amounts of co-injected sbn^tc24smad5 RNA, and 90% by twofold amounts (Fig. 1D). To study whether sbn^tc24 Smad5 might also interfere with other related Smad proteins, sbn^tc24smad5 mRNA was coinjected with zebrafish wild-type smad1 RNA. In addition, smad1(sbn) mRNA, bearing the Thr→Ile mutation found in sbn^tc24, was co-injected with wild-type smad5. In both cases, the antimorphic version also inhibited the wild-type paralogue, but more weakly than its own wild-type version (Fig. 1E,F).

The temporal expression patterns of zebrafish smad1 and smad5 in wild-type embryos have been investigated by developmental northern blot analysis. The mRNA profile indicates that smad5 is expressed both maternally and zygotically. Levels of maternally supplied smad5 mRNA drop to about 50% of their intial value during cleavage and early blastula stages, rise again at the sphere stage after the onset of zygotic gene expression at midblastula stages (Kane and Kimmel, 1993), and decline thereafter, with a strong reduction during early gastrula stages (between shield stage and 80% epiboly, Fig. 2A). smad1 shows a different temporal expression profile with no mRNA detectable prior to midgastrulation at cleavage, blastula or early gastrula stages (Fig. 2A). The same temporal smad1 expression pattern was revealed via developmental RT-PCR analysis (not shown). Studies of the spatial expression pattern of smad5 by whole-mount in situ hybridizations revealed a uniform and ubiquitous distribution of both maternal and zygotic smad5 transcripts at cleavage (Fig. 2B), blastula (Fig. 2C) and gastrula stages (Fig. 2D,F).

Consistent with the maternal and zygotic expression of smad5, the somitabun mutation sbn^tc24 has both a dominant maternal and a dominant zygotic effect, which lead to dorsalizations of different degrees.

Heterozygous sbn females crossed to either wild-type or heterozygous males give rise to 100% strongly dorsalized embryos (C4 phenotypes; Mullins et al., 1996, see Table 1 for further explanation of the various phenotypes) characterized by a body axis wound up in a snailshell-like fashion at day one of development (Fig. 3A,B). The strength of dorsalization among the different offspring of two sbn^tc24 heterozygous parents was indistinguishable, whether the embryos were homozygous mutant (Fig. 3G), heterozygous (Fig. 3F), or wild-type (Fig. 3E). We were able to rescue sbn^tc24 homozygous mutant embryos and raise them to adulthood (see below), and thus could also generate offspring from homozygous mutant mothers (Fig. 3H). These showed the same degree of dorsalization as embryos from heterozygous mothers, indicating that the dominant maternal effect of the sbn mutation is not enhanced by removing the remaining maternal and zygotic contributions of wild-type smad5 mRNA.

sbn^tc24 also displays a dominant zygotic effect as seen in 50% of the offspring of a sbn heterozygous male and a wild-type female. This dominant zygotic phenotype is viable and is characterized by loss of the ventral tail fin at day 1 of development (C1; Fig. 3C).

Rescue and phenocopy of the somitabun phenotype by wild-type and mutant smad5

To further confirm that the dorsalization of sbn^tc24 mutant embryos is caused by the mutation found in the smad5 gene, smad5 RNA injections were carried out to rescue the mutant phenotype. In addition, mRNA injections of the antimorphic smad5 (sbn) allele into wild-type embryos were carried out to phenocopy the sbn mutant phenotype.

The dorsalized phenotype of most embryos from a cross of two sbn heterozygous parents was rescued in a dose-dependent manner upon injection of wild-type smad5 mRNA, whereas injection into dorsalized swr(bmp2b) mutants had no effect (Table 1). A high proportion of the smad5-injected sbn embryos reached wild-type condition and could be raised to adulthood (Fig. 4A,B; Table 1). For 82 such adult mutants genotyped via tetra-PCR of genomic tail fin DNA, 21/55 males and 13/27 females were wild type (41%), 28/55 males and 8/27 females were smad5 heterozygous, and 6/55 males and 6/27 females were homozygous mutant (14%) (25% expected wild-type and 25% homozygous mutant fish if rescue efficiency is independent of zygotic genotype). smad5 homozygous mutant adults of both sexes were fertile, and males gave 100% C1-phenotype embryos when mated to wild-type females.

The sbn mutant phenotype could also be rescued, with a weaker frequency, by injecting the closely related zebrafish smad1 RNA (Table 1); injecting zebrafish smad2 (A. D. and M. H., unpublished) or lacZ mRNA had no effect (Table 1). sbn mutants were also rescued by injecting human smad4 RNA (Fig. 4E,F; Table 1), the putative partner of all ‘receptor-activated’ Smad proteins.

To phenocopy the sbn phenotype, smad5(sbn) mRNA bearing the mutation found in sbn^tc24 was injected into wild-type embryos. While 25 pg wild-type smad5 mRNA had no effect, 25 pg mutant smad5(sbn) mRNA led to a significant dorsalization up to C4 strength (Fig. 4C,D; Table 1). When the Thr→Ile mutation found in sbn^tc24 was introduced into other zebrafish smad cDNAs, it converted Smad1 into a dorsalizing agent of similar strength to Smad5(sbn), whereas mutated smad2(sbn) RNA injected into wild-type embryos had no effect (Table 1).

The weakness of the dominant zygotic effect of sbn^tc24 compared to its dominant maternal effect suggests that the smad5 gene product acts at very early stages in development. To appraise the time window of Smad5 action, we carried out both sbn rescue and phenocopy experiments, injecting plasmid DNAs that drive smad5 expression under the control of two promoters with different temporal activation profiles: the Xenopus EF1α promoter, which is activated right after midblastula transition when zygotic gene expression starts, and the cytoskeletal actin (CSKA) promoter, which is strongly activated during gastrulation but very weakly expressed at earlier stages (Hammerschmidt et al., 1998). Only when gene expression was under the control of the EF1α promoter, did injection of wild-type smad5 DNA into sbn embryos lead to a rescue, and injection of mutant smad5 DNA into wild-type embryos to a phenocopy of the sbn phenotype. No effect was observed upon injection of the corresponding CSKA constructs (Table 1). These data suggest that smad5 acts after midblastula transition and before gastrulation.

Smad5 has been previously described as a putative mediator of ventralizing signaling by Bmps. In the zebrafish, the bmp2b null mutations swr^ta72 and swr^tc300 lead to dorsalized embryos (Kishimoto et al., 1997; Nguyen et al., 1998) very similar to sbn mutant embryos, indicating that bmp2b and smad5 are both involved in ventral development. In contrast to sbn, the effect of swr^ta72 is purely zygotic, suggesting that Bmp2b has no maternal function (Kishimoto et al., 1997). To investigate whether smad5 and bmp2b interact genetically, sbn,swr double mutant embryos were generated. While crosses between sbn heterozygous males and wild-type females led to 50% weakly dorsalized (C1) and 50% wild-type embryos, and crosses between wild-type males and swr heterozygous females to 100% wild-type embryos, a significantly stronger dorsalization was found in a quarter of the offspring of sbn heterozygous males and swr heterozygous females (25% C3, 23% C1, 52% wild type; n=588; 7 crosses; compare Mullins et al., 1996). Genotyping of 20 individuals of each class identified the C3 embryos as sbn,swr double heterozygotes (Fig. 3D), while the C1 embryos were sbn heterozygous and wild type for swr. This indicates that the dominant zygotic effect of the smad5 mutation sbn^tc24 is substantially enhanced by the loss of one functional bmp2b allele. The same appears to be true for the dominant maternal effect of sbn. Crosses between sbn heterozygous females and swr heterozygous males yielded 50.5% more strongly dorsalized embryos (C5; see Table 1), and 49.5% C4 embryos (n=327; 2 crosses), compared to 100% C4 embryos in crosses between sbn heterozygous females and wild-type males. Genotyping revealed that 10 of 10 tested C5 embryos were swr heterozygotes, while 10 of 10 C4 embryos were swr wild type.

Analysis of swr mutant embryos has revealed that bmp2b is required for the maintenance of its own expression (Kishimoto et al., 1997). To investigate the role of smad5 on bmp2b expression, we compared the bmp2b expression pattern in sbn mutant, swr(bmp2b) mutant, and wild-type embryos.

At the sphere stage, approx. 1 hour after the onset of zygotic transcription, the bmp2b expression patterns of both sbn and swr mutants were indistinguishable from that of wild-type embryos (Fig. 5A-C). At the onset of gastrulation (shield stage), however, both mutants display a dramatic reduction in bmp2b transcripts (Fig. 5D-F). Thus, the initial induction of bmp2b expression is not affected by the maternal effect of the sbn mutation, but the later, Bmp2b-dependent maintenance of bmp2b expression appears to be impaired.

The comparison of the bmp2b expression patterns in sbn and swr (bmp2b) mutant embryos suggests that the sbn mutation blocks the aforementioned positive Bmp2b autoregulation. To test this idea more directly, the effects of exogenous Smad5 and Bmp2/4 on bmp2b expression in sbn mutant embryos were analyzed. smad5 (Fig. 5H; compare with 5D), but not Xbmp4 mRNA (Fig. 5I), injected into sbn mutant embryos led to a rescue of bmp2b mRNA levels at early gastrula stages, while increased bmp2b mRNA levels were observed after Xbmp4 RNA injection in wild-type embryos. This suggests that the sbn mutation acts downstream of Bmp2b, blocking the mediation of Bmp2/4 signaling during late blastula and early gastrula stages.

In contrast to this unresponsiveness in early bmp2b expression, bmp2/4-injected sbn mutant embryos displayed a striking response in their morphology, first apparent at the end of gastrulation. Depending on the amounts of injected zebrafish bmp2b or bmp4 or Xenopus bmp4 mRNA, the strong dorsalization of sbn mutants could be significantly normalized (C4 to C1), or even converted to a ventralization (C4 to V3, Table 1). Since this effect was also obtained with embryos deriving from two sbn homozygous mutant parents, it can be ruled out that the rescue by Bmp2/4 was mediated by residual maternal or zygotic wild-type Smad5 (Fig. 6A,B; Table 1).

To determine when during development exogenous Bmp2/4 can override the effect of the sbn mutation, we expressed human bmp4 under the control of the cytoskeletal actin promoter in sbn mutant embryos. In contrast to the corresponding smad5 transgene, expression of the human bmp4 gene under this promoter led to a striking rescue and even ventralization of sbn mutant embryos (Fig. 6C,D; Table 1), suggesting that the sbn-independent response of the zebrafish embryo to exogenous Bmp2/4 occurs during gastrula stages.

In the bmp2/4 RNA injection experiments described above, the actual in vivo concentrations of Bmp2/4 protein are unknown. To investigate whether sbn cells can respond to Bmp2/4 under physiological conditions, cell transplantation experiments were carried out. Labeled cells from either wild-type embryos or embryos deriving from a cross of two sbn heterozygous or two sbn homozygous mutant parents were transplanted into wild-type embryos. Homochronic transplantations were carried out at the sphere stage, when sbn mutant embryos still show normal marker gene expression, or at the shield stage, shortly after the onset of gastrulation when sbn mutants display a dramatic loss in bmp2b mRNA levels. In all cases, donor cells gave rise to blood (Fig. 6E,F) and ventral tail fin (Fig. 6G,H), tissues completely absent in sbn mutant embryos. The frequencies of ventral tissue contribution were similar, independently of whether donor cells were wild-type or sbn mutant (see Table 2), and whether they were transplanted at late blastula or early gastrula stages. Furthermore, individual host embryos with blood cells deriving from both donors were obtained after simultaneous transplantation of differently labeled wild-type and sbn mutant cells (not shown). This indicates that in a wild-type environment, sbn mutant cells can form the ventral-most derivatives, thereby showing the maximal possible response to endogenous ventralizing signals. Thus, in contrast to the early mediation of Bmp2/4 signaling, the final specification of ventral cell fates can occur in the absence of functional Smad5.

Many recent publications deal with the function of Smad proteins as mediators of TGFβ signaling. Direct genetic evidence for this function was provided for Drosophila Mad, which mediates Decapentaplegic (Dpp) signaling during midgut and eye development (Newfeld et al., 1997; Wiersdorff et al., 1996). In addition to these later zygotic functions, mad is also required for early D-V patterning of the fly embryo, and mutants lacking both maternally and zygotically contributed mad gene products exhibit the same ventralized phenotype as dpp mutant embryos (Das et al., 1998).

Here, we show that in the zebrafish one of the homologues of mad, smad5, is involved in the mediation of Bmp2/4 signaling during early phases of embryonic D-V pattern formation. Like Drosophila mad, zebrafish smad5 mRNA is both maternally and zygotically supplied. The antimorphic smad5 allele somitabun (sbn^tc24) leads, when maternally supplied, to a strong dorsalization, similar to the phenotype caused by loss of the ventralizing signal Bmp2b (Kishimoto et al., 1997; Nguyen et al., 1998), indicating a role of smad5 in early ventral development. Smad5 may also be involved in later developmental processes, but its later function is not completely indispensable, as suggested by rescue experiments of smad5 mutant embryos with exogenous wild-type smad5 RNA, which is usually degraded by early somitogenesis stages (Hammerschmidt et al., 1998). While uninjected embryos from two sbn heterozygous parents display dorsalizations of equal strengths, irrespective of their zygotic genotype, rescued homozygous mutant adults were obtained with a lower than expected frequency, indicating a lower survival rate compared to their sbn heterozygous and wild-type siblings.

The alterations in Smad5 activity caused by the somitabun mutation have been investigated in Xenopus animal cap explants, measuring the ventral mesoderm-inducing properties of different Smad5 and Smad1 alleles – although such assays may not reflect the in vivo function of Smad1/5 proteins, as the initial induction of ventral mesoderm is unaffected in sbn mutant zebrafish embryos. In contrast to the wild-type version, sbn^tc24 Smad5 showed no or a very weak ventralizing activity. In addition, co-expression experiments in Xenopus animal caps and comparison of sbn homozygous and heterozygous mutant zebrafish embryos suggest that sbn^tc24 Smad5 acts like an antimorph that inactivates most of wild-type Smad5 protein when present in equimolar ratios. Together, these results suggest that embryos from sbn mutants lack all or at least most of Smad5 activity.

However, this loss of wild-type Smad5 activity is not necessarily the basis for the sbn mutant phenotype, since sbn mutant Smad5 protein could also inhibit other Smad proteins, as shown in the animal cap assay for smad5 and smad1. The observed dorsalization of sbn mutant embryos could equally well result from a general inhibition of Smad1, Smad5 and other as yet unidentified additional members of the Smad1/5 family. It is thus currently impossible to state to what extent smad5 itself is actually required for early dorsoventral patterning, although preliminary data suggest that it might be essential: another mutation, captain hook (cpt^m169; Solnica-Krezel et al., 1996), appears to interfere specifically with zygotic smad5 expression, and leads to a mild dorsalization similar to that caused by the zygotic effect of sbn (A. D. and M. H., unpublished results).

There are several hints as to how the point mutation found in sbn^tc24 Smad5 may cause its dominant negative effect molecularly. According to current models, Smad5 proteins normally exist as homo-trimers which upon activation by Bmp receptors bind to Smad4 trimers to form transcriptionally active regulators. The mutated threonine residue T(429) of Smad5 is located at a solvent-exposed position within a region of the MH2 domain, the L3 loop. This loop is involved in the interaction of receptor-regulated Smad proteins with the kinase domain of the TGFβ receptors and with Smad4 (Shi et al., 1997; Lo et al., 1998). According to in vitro studies carried out with Smad2, replacement of the threonine corresponding to T(429) in Smad5 by lysine or alanine causes a dramatic reduction in the affinity to Smad4 and TGFβ receptors, while the homo-trimer formation is unaffected (Lo et al., 1998). Similar properties have been proposed for the naturally occurring inhibitory Smad6 protein, which can bind to Smad1, but not to Smad2 or Smad4 (Hata et al., 1998). In its L3 loop, Smad6 protein differs only in a few amino acid residues from the sequence conserved in the receptor-regulated Smad protein, including a T(431)C change of two amino acids downstream of T(429). Apparently, the T(429)I change in sbn^tc24 Smad5 yields an inhibitory version of Smad5 protein, trapping wild-type Smad5 protein in signaling-incompetent trimers whose interaction with either the Bmp2/4 receptors or the Smad4 trimers is strongly impaired. Our observation that the maternal effect of the sbn mutation could be rescued by injecting high amounts of human smad4 mRNA indicates that interaction with Smad4, rather than activation by the Bmp receptors is affected by the sbn mutation.

Co-expression experiments in Xenopus animal cap explants with a dominant negative version of the Bmp receptor have indicated that Smad1 and Smad5 act downstream of Bmp signaling. In this study, we analyzed the relationship between bmp2b and smad5 by genetic tests. The enhancement of the dominant effect of the sbn mutation by the loss of one bmp2b wild-type allele in sbn,swr double heterozygotes indicates that smad5 and bmp2b interact during ventral specification of the early zebrafish embryo. Their epistatic relationship cannot be addressed in such double mutant analyses because mutations in both genes lead to a similar phenotype. Epistasis analyses are further complicated by the existence of an autoregulatory positive feedback loop of Bmp2b on its own expression (Kishimoto et al., 1997) which, as shown in this work, depends on Smad5. Thus, during the maintenance phase of bmp2b expression, Bmp2b appears to act both upstream and downstream of Smad5. Accordingly, exogenous Bmp2/4 failed to rescue the early defects of smad5 mutants on bmp2b expression, and exogenous Smad5 failed to rescue the bmp2b mutant phenotype.

In contrast to its maintenance, the initiation of bmp2b expression in smad5 mutant blastula embryos occurs normally, indicating that smad5 plays no role in its induction, and possibly not prior to Bmp2b action at all. Accordingly, exogenous smad5 expressed after midblastula transition can compensate for the loss of endogenous Smad5 activity. Together, these data support a model according to which smad5, despite its maternal expression and the maternal effect of the sbn mutation, has no maternal function per se. Rather, smad5 gene products appear to be present so early to ensure that cells are competent to process Bmp2b signaling when zygotic bmp2b expression starts.

These findings define three distinct phases of D-V patterning: in the first phase, an initial coarse D-V pattern is set up, when the future dorsal mesoderm, the equivalent of the amphibian Spemann organizer, is induced in a small dorsal marginal domain characterized by the expression of chordino and opposed by the expression of bmp2b in the rest of the embryo. This pattern is most likely set up under maternal control, and is independent of Bmp2b and Smad5, although in Xenopus, Bmp2/4 and Smad1/5 are dicussed as maternal components involved in the induction of ventral mesoderm (Graff et al., 1996).

In the second phase, which is under zygotic control and dependent on Bmp2b and Smad5, this initial D-V pattern is refined, leading to the transformation of the broad and uniform expression of bmp2b to a graded pattern with ventral-to-dorsal progressively dropping bmp2b mRNA levels. This transformation is governed by the antagonizing action of Bmp2b and its mediator Smad5 on one side and the Bmp2b-antagonist, Chordino, on the other side. In swirl(bmp2b) and somitabun(smad5) mutant embryos, bmp2b mRNA levels drop throughout the entire embryo, indicating that bmp2b and smad5 are involved in the maintenance of bmp2b expression, while in chordino mutant embryos, the initial broad bmp2b expression pattern is maintained (Hammerschmidt et al., 1996b), indicating that Chordino signaling from the dorsal side is required for the clearing of bmp2b mRNA in dorsolateral regions of the late blastula embryo. Judging from in situ hybridization patterns in wild-type and mutant embryos, the establishment of this Bmp2b gradient, which is thought to have morphogenetic character defining positional values and cell fates along the D-V axis, appears completed by the onset of gastrulation.

Several observations point to a function of Bmp2/4 beyond this time point, defining a third phase of D-V patterning. In Xenopus, the morphogenetic action of Bmp4 is thought to occur at early gastrula stages (Jones et al., 1996). There are indications that the same might be true in zebrafish. Here, embryos lacking functional Bmp2b (Hammerschmidt et al., 1996b) or Smad5 (this work) can be rescued by injecting a DNA construct with the expression of Bmp4 under the control of the cytoskeletal actin promoter, which is active during gastrulation (Hammerschmidt et al., 1998). Furthermore, cells from shield stage smad5-deficient embryos transplanted into wild-type embryos can be rescued to form the ventral-most derivatives like blood or ventral tail fin, tissues normally missing in mutant embryos (this work).

This suggests that Bmp2/4 function during D-V pattern formation is twofold. First, during blastula stages Bmp2b serves to ensure the maintenance of its own expression, a prerequisite for establishing the morphogenetic Bmp2/4 gradient by the antagonistic action of Bmp2b and Chordino. The morphogenetic Bmp2/4 action itself, to determine cell fates along the D-V axis in a dose-dependent fashion, seems to take place later, during early gastrula stages, after the gradient has been set up.

The smad5 mutation sbn^tc24 described here nicely dissects these two phases of zygotic Bmp2/4 action, as it affects just the early phase, when the Bmp2/4 gradient is set up, whereas the later mediation of Bmp2/4 signaling to induce final cell fate specifications can occur normally.

This dispensability of Smad5 function at later stages might be caused by the presence of other mediators of Bmp2/4 signaling with redundant functions, such as e.g. Smad1, which starts to be made during early gastrula stages. However, we find that Smad1 can also be inhibited by Smad5(sbn), so this redundant transducer of Bmp2/4 signaling could be another Smad protein, or even a completely unrelated transcription factor. In any case our results indicate that there must be an additional, Smad5(sbn)-independent transcription factor downstream of Bmp2/4 specifically involved in the third phase of D-V patterning, the interpretation of the putative morphogenetic Bmp2/4 gradient.

We are very grateful to Claudia F. B. de Olveira and Beate Fischer for their help during the mapping of smad5, Heike Maifeld for her help in genotyping sbn via tetra-PCR, and Hermann Bauer for confirming the sbn genotyping via single strand conformation analysis. We thank Drs Jon Graff and Douglas Melton for degenerated primers to amplify vertebrate smad genes, and for the plasmids pSP64T-tBr and pSP64T-Xbmp4, Drs Ying Zhang and Rik Derynck for human SMAD4, Drs Kathryn Helde and David Grunwald for the zebrafish gastrula cDNA library, and particularly Dr Chris Wright for cesium-chloride gradient-purified pCSKA-BMP4 plasmid. We are very indebted to Dr Kris Vleminckx for introducing us to Xenopus laevis, and to Drs Patrick Blader, Ference Muller and Uwe Strähle for communicating unpublished results. A. D. was supported by a predoctoral long-term fellowship from the Boehringer Ingelheim Fonds, Stuttgart.