Induction and differentiation of the zebrafish heart requires fibroblast growth factor 8 (fgf8/acerebellar)

The heart is the first organ to form and function during vertebrate embryogenesis. The fish heart can be viewed as a prototypical vertebrate heart, since the process of heart formation occurs in a similar way in all vertebrates (DeHaan, 1965; DeRuiter et al., 1992; Stainier and Fishman, 1992; Stainier et al., 1993; Fishman and Chien, 1997). In zebrafish, cardiogenic precursors involute shortly after the onset of gastrulation and migrate medially as part of the lateral plate mesoderm. The myocardial precursor cells form two tubular primordia on each side of the midline at early somitogenesis. These primitive tubes then fuse by the 20-somite stage and enclose the endocardial precursor cells to give rise to the definitive heart tube consisting of an inner, endocardial layer and an outer, myocardial layer (Stainier and Fishman, 1992; Stainier et al., 1993; Fishman and Chien, 1997). Around 24 hours post-fertilization (hpf), the heart starts beating and circulation begins. The definitive heart tube is subsequently divided into different chambers, with the atrium and the ventricle being the most prominent heart structures. The atrium can be distinguished from the ventricle by differential expression of myosin heavy chains already at 24 hpf, and a morphological distinction becomes apparent soon thereafter (Stainier and Fishman, 1992; Stainier et al., 1993).

Although heart development has been well described morphologically, the molecular events underlying this process are only beginning to be understood (for reviews see Lyons, 1996; Olson and Srivastava, 1996; Fishman and Chien, 1997; Mohun and Sparrow, 1997). Several families of transcription factors have been implicated in heart development. A key family are the vertebrate Nkx2.5 genes, which are homologues of the Drosophila gene tinman. In Drosophila, tinman is expressed in the heart (or dorsal vessel) and visceral mesoderm, and is required for heart formation (Azpiazu and Frasch, 1993; Bodmer, 1993; Harvey, 1996). In vertebrates, Nkx2.5 is the earliest marker for heart precursors in the lateral plate mesoderm (Lints et al., 1993; Komuro and Izumo, 1993; Chen and Fishman, 1996). Overexpression of nkx2.5 in Xenopus or zebrafish causes a general increase in heart size (Chen and Fishman, 1996; Cleaver et al., 1996). Targeted disruption of Nkx2.5 in the mouse does not prevent formation of the heart tube, but cardiac looping is absent and the expression of downstream cardiac transcription factors is disturbed (Lyons et al., 1995; Tanaka et al., 1999). At least two more Nk2 class homeobox genes, nkx2.7 and Nkx2.8, are expressed early in the developing heart, with nkx2.7 being expressed even before nkx2.5 in the zebrafish (Lee et al., 1996; Brand et al., 1997).

A direct target gene regulated by tinman is the myocyte enhancer binding factor-2 (D-mef2) of the MADS box-containing proteins, and loss of D-MEF2 function prevents formation of cardiac muscle (Lilly et al., 1994; Bour et al., 1995; Lilly et al., 1995; Olson et al., 1995; Gajeweski et al., 1997). Likewise, the four mef2 genes in vertebrates are expressed in precardiac mesoderm and mouse mutants for MEF2C exhibit heart abnormalities (Edmondson et al., 1994; Molkentin et al., 1996; Ticho et al., 1996; Lin et al., 1997). In addition, several members of the GATA family of zinc finger domain transcription factors are involved in cardiac development. GATA4 is expressed in the presumptive heart mesoderm and overexpression of GATA4 in mouse embryonic stem cells enhances cardiogenesis (Kelley et al., 1993; Heikinheimo et al., 1994; Laverriere et al., 1994; Jiang and Evans, 1996; Grepin et al., 1997). In contrast, the failure of the bilateral heart primordia to migrate medially and to fuse in GATA4 knock-out mice is thought to be a secondary consequence of a defective endoderm (Kuo et al., 1997; Molkentin et al., 1997; Narita et al., 1997a,b). Finally, basic region helix-loop-helix (bHLH) domain proteins play important roles at later stages of cardiac morphogenesis, e.g. in chamber-specific gene expression (Srivastava et al., 1995). Although localized expression of transcription factors is important for heart development, little is known about the signaling molecules that control this expression. Endoderm, ectoderm and Spemann’s organizer have been implicated as sources for cardiac mesoderm inducing signals (Lyons, 1996; Fishman and Chien, 1997; Mohun and Sparrow, 1997). Misexpression of BMPs in anterior mesoderm of chick embryos suggested that BMPs may induce expression of the cardiac transcription factors Nkx2.5 and GATA4, and BMPs are expressed close to the heart primordium (Schultheiss et al., 1997; Andrée et al., 1998). In Drosophila, tinman is indeed regulated by the BMP homologue decapentaplegic (dpp) (Frasch, 1995). In addition, tissue culture studies have demonstrated cardiogenic effects for Activin-A and FGF2, and treatment of non-precardiac mesodermal explants from early chick embryos with a combination of FGF4 and BMP2 is able to induce cardiogenesis in vitro (Sugi and Lough, 1995; Lough et al., 1996).

Here we show that the secreted signaling factor Fgf8 is required in vivo for heart development of the zebrafish. Although expression of several Fgfs has been described in the developing heart (Parlow et al., 1991; Spirito et al., 1991; Engelmann et al., 1993; Mason et al., 1994; Crossley and Martin, 1995; Zhu et al., 1996; Hartung et al., 1997; Miyake et al., 1998), studying the function of Fgfs in development of the vertebrate heart has been difficult, often due to early lethality of the mutants or possible functional redundancy. Analysis of the available loss-of-function mutations has not revealed a specific function for Fgf signaling in the developing heart so far (Feldman et al., 1995; Dono et al., 1998; Meyers et al., 1998). Similarly, 3 out of 4 vertebrate Fgf receptors are expressed during heart development, but their inactivation in mouse embryos has not been informative with respect to cardiac development (Orr-Urtreger et al., 1991; Yamaguchi et al., 1991; Peters et al., 1992; Deng et al., 1994; Arman et al., 1995; Sugi et al., 1995; Thisse et al., 1995; Colvin et al., 1996; Weinstein et al., 1998).

In contrast, functional studies in Drosophila do suggest a role for Fgf signaling in cardiac development (Beiman et al., 1996; Gisselbrecht et al., 1996; Michelson et al., 1998).

The isolation of many mutants affecting zebrafish heart development in the systematic screens for embryonic zebrafish mutants opens up the possibility to study the genetic control of vertebrate heart development in great detail (Haffter et al., 1996). The zebrafish mutant acerebellar (ace) was orginally classified as a brain mutant (Brand et al., 1996) in which the fgf8 gene is inactivated (Reifers et al., 1998). Here we show that analysis of this mutant unexpectedly reveals a pivotal role for Fgf8 in myocardial induction. We examine in detail the expression and function of Fgf8 in the developing zebrafish heart, and propose that Fgf8 functions together with Bmps in induction of heart-specific gene expression upstream of nkx2.5 and gata4.

Zebrafish were raised and kept under standard laboratory conditions at about 27°C (Westerfield, 1994; Brand and Granato, 1999). Mutant carriers were identified by random intercrosses. To obtain embryos showing the mutant phenotype, two heterozygous carriers for a mutation were crossed to one another. Typically, the eggs were spawned synchronously at dawn of the next morning, and embryos were collected, sorted, observed and fixed at different times of development at 28.5°C. In addition, morphological features were used to determine the stage of the embryos, as described by Kimmel et al. (1995). In some cases, 0.2 mM phenylthiourea (PTU) was added to prevent melanization. Isolation and characterization of acerebellar is described by Brand et al. (1996) and Reifers et al. (1998).

Microangiography was essentially done as described by Weinstein et al. (1995). Briefly, embryos were injected with 0.01 mm diameter fluorescent latex beads (Molecular Probes). Bead suspension was diluted 1:1 with 2% BSA in deionized distilled water, sonicated and subjected to centrifugation for 2 minutes in an Eppendorf microcentrifuge. Dechorionated embryos were anaesthetized with tricaine as described and placed on an injection platform (Westerfield, 1994). A large bolus of bead suspension was injected into the sinus venosus. The fluorescent beads were uniformly distributed throughout the vasculature of the embryo within minutes. Specimens were either photographed on a Zeiss Axioskop microscope or scanned using the confocal microscope (Leica TCS4D) as rapidly as possible. Images were assembled using Adobe Photoshop.

In situ hybridizations (ISH) were done as described by Reifers et al. (1998), and histology is described by Kuwada et al. (1990). Immunohistochemistry is described by Stainier and Gilbert (1990). MF20, anti-myosin heavy chain, was obtained from the Developmental Studies Hybridoma Bank. MF20 stains all cardiac chambers. S46 was a kind gift from Frank Stockdale, Stanford University. S46 specifically reacts with the atrium and the sinus venosus. Double-labeled embryos were stained with monoclonal antibodies S46 (IgG1) and MF20 (IgG2b) followed by fluoresceinated goat anti-mouse IgG1 and rhodaminated goat anti-mouse IgG2b. Heart length was measured on photographs of individually mounted wild-type and acerebellar embryos, heart length is given as percentage of wild-type length.

fgf8, subcloned into pCS2+ (Rupp et al., 1994), and XFD (Amaya et al., 1991) were linearized and transcribed using the SP6 message machine kit (Ambion). The amount of RNA injected was estimated from the concentration and volume of a sphere of RNA injected into oil at the same pressure settings. Typically, about 25 pg of fgf8 RNA or XFD RNA were injected; RNA was dissolved in 0.25 M KCl with 0.2% phenol red and backloaded into borosilicate capillaries prepared on a Sutter puller. During injection, RNA was deposited into the cytoplasm of 2-cell stage embryos; in embryos after the first cleavage, the RNA usually stays in the progeny of the injected blastomere, as judged from the often unilateral distribution of control lacZ RNA, as detected with anti-β-gal antibody (Promega, 1:500) after ISH (Dornseifer et al., 1997). The embryos were fixed at the 6- to 8-somite stage prior to ISH and antibody staining.

Heparin acrylic beads (Sigma) were washed in ethanol for 30 minutes, then washed twice in PBS for 5 minutes and finally coated with recombinant mouse FGF8 protein (R&D Systems; 0.25 mg/ml). Protein-coated beads were stored for several weeks at 4°C without detectable loss of activity. FGF8 or PBS control beads were implanted in the lateral plate mesoderm of 5- to 7-somite stage acerebellar mutant embryos, the embryos were fixed at the 10- to 12-somite stage prior to ISH. For pharmacological inhibition of Fgfr activity, wild-type embryos were treated with SU5402 (Calbiochem; Mohammadi et al., 1997) at 8 μm in embryo medium at 28.5°C in the dark for the indicated time periods.

Wild-type embryos examined at 36 and 48 hpf (n=107) have successfully established normal circulation, whereas ace embryos fail to do so, with 23% having no circulation, and 77% of the acerebellar mutants (n=159) having variably reduced circulation. To examine whether the defect in the circulatory system of acerebellar mutant embryos is due to a malformation of the heart itself or the blood vessel system, we analysed vessel formation by microangiography, i.e. visualization of circulation by injecting fluorescent latex beads into the sinus venosus, and by staining for blood vessel marker gene expression. Microangiography at 48 hpf revealed an overall normal organization of the blood vessels in those acerebellar embryos that established circulation (Fig. 1A-D). Likewise, the expression pattern of flk-1, a receptor tyrosine kinase that is a marker for blood vessel endothelial cells (Fouquet et al., 1997; Liao et al., 1997) appears largely normal in all acerebellar mutants, except for an absence of vessels in the dorsal brain (Fig. 1E,F), which is most likely a secondary effect due to the brain defects in mutant embryos (Reifers et al., 1998, Picker et al., 1999). Taken together, these results suggest that blood vessel formation in acerebellar mutants is overall normal; in contrast heart development is severely abnormal.

The zebrafish heart is composed of four subdivisions: the sinus venosus, the posterior atrium, the anterior ventricle and the bulbus arteriosus (Stainier and Fishman, 1992). The prominent atrium can be readily distinguished from the adjacent ventricle in living wild-type embryos (Fig. 1G,H). Heartbeat frequency is approximately normal in the mutants at 28 hpf (106±7 beats/minute, n=5) versus the wild-type (111±6 beats/min, n=5), but the heart of acerebellar mutant embryos is severely dysmorphic (Fig. 1I,J). Histological sections show that subdivision into an inner endocardial and outer myocardial layer occurs normally (Fig. 1K,L). In contrast, immunohistochemical analysis with the myosin heavy chain antibodies MF20 and the ventricular-specific antibody S46 and measurements of heart length shows that the heart is overall shorter, and that in particular the ventricular part is reduced (Fig. 1M-P).

Since the heart is affected in acerebellar mutant embryos and since fgf8 expression during zebrafish heart development has not been analyzed previously, we examined the expression pattern of fgf8 in the developing heart in detail, relative to several markers for cardiac precursors by double in situ hybridisation. Already at the blastula stage, fgf8 is expressed all around the margin of the embryo (Fürthauer et al., 1997; Reifers et al., 1998), which includes the fate map position of cardiac precursors ventrolaterally (Warga and Kimmel, 1990; Stainier and Fishman, 1992). The expression of fgf8 at the mid-hindbrain boundary at tailbud stage (Fig. 2G-I) is located close to the posterior part of the expression domain of nkx2.7, which includes cardiac precursors in zebrafish (Lee et al., 1996). From the 3-somite stage onwards fgf8-positive cells can be observed in the lateral plate mesoderm in the area of the incipient heart field (Fig. 2A,B). At the 4-somite stage fgf8 is expressed in the lateral plate mesoderm adjacent to the posterior part of the gata4 expression domain (Fig. 2J-L), which includes the heart precursors (Kelley et al., 1993; Jiang and Evans, 1996; Serbedzija et al., 1998). Mesodermal fgf8 expression partially overlaps with the expression domain of nkx2.5 at the 7-somite stage (Fig. 2M-O), for which it has been shown that the anterior cells contribute to the heart (Serbedzija et al., 1998). At the 8-somite stage a subset of the fgf8 expression lateral to the embryonic axis includes the myocardial precursors (Fig. 2C,D), which arrive in this region at this timepoint in development (Stainier and Fishman, 1992). The fgf8 expression overlaps partially with the expression domains of gata4 and almost completely with nkx2.5 at the 11-somite stage (Fig. 2P-U). At the 19-somite stage the bilateral myocardial primordia meet caudal to the mid-hindbrain boundary to form a mesodermal ring-like structure which expresses fgf8 (Fig. 2E). At 36 hpf atrium and ventricle can be morphologically distinguished, and fgf8 expression is stronger in the ventricle (Fig. 2F); expression persists at least until 48 hpf, the latest stage we examined. These data suggest that Fgf8 could function during the earliest stages of cardiac precursor development, and in the differentiated heart.

Since fgf8 is expressed early on in close association with the heart precursors and acerebellar embryos fail to develop proper hearts, we analysed the expression of cardiac marker genes in acerebellar mutant embryos (Table 1 and Fig. 3). nkx2.7, gata4 and nkx2.5 are affected from the onset of expression. Importantly, gata6-positive cells are present in mutant embryos, demonstrating that the lateral plate mesoderm cells that are normally destined to become cardiogenic precursors are present (Fig. 3I,J). In addition, we could not detect an increase in apoptotic cell death in the lateral plate mesoderm of acerebellar embryos (data not shown). Expression of nkx2.7 is strongly reduced as early as the 1-somite stage in acerebellar embryos (Fig. 3A,B), but recovers quickly. Expression of the zinc-finger transcription factor gata4 is also clearly reduced in mutant embryos at the 3- to 6-somite stage (Fig. 3C,D). The most severe effect was detected for nkx2.5 expression, which was strongly reduced or absent between the 3-somite and 15-somite stages in acerebellar embryos (Fig. 3E,F). At the 11-somite stage the fgf8 expression domain partially overlaps the nkx2.5 domain and reaches further posterior at least to the level of rhombomere 5. In acerebellar mutant embryos the anterior part of nkx2.5 expression is severely reduced and the expression in the posterior subdomain is completely absent (Fig. 3G,H). At late somitogenesis stages, all heart markers tested (nkx2.5, nkx2.7, gata4, gata6 and bmp4) recover in the intensity of their expression in the remaining heart precursors of acerebellar mutant embryos (Table 1). However, a size-reduction of the mutant hearts remains detectable with these markers even at later stages (Fig. 3K,L and not shown). We conclude that fgf8 expression partially overlaps with nkx2.5 in heart precursor cells and that Fgf8 is required for expression of nkx2.5. Interestingly, the requirement may be stronger in the posterior part of the nkx2.5 domain, since this expression is completely lost in acerebellar mutants (Fig. 3G,H). This early defect in mutant embryos may account for the subsequent loss of heart structures.

The requirement of acerebellar/fgf8 during the earliest stages of cardiac gene expression raised the possibility that Fgf8 is an inducer of heart development. To further examine this issue, we injected synthetic RNAs encoding fgf8 or a dominant negative Fgf receptor into wild-type and acerebellar embryos. In addition to the described dorsalizing effect (Fürthauer et al., 1997; Reifers et al., 1998), fgf8 injection can restore gata4, but not nkx2.5 expression in mutant embryos at the 6-somite stage (Fig. 4A,B). 14 out of 45 (31%) acerebellar mutant embryos show strong gata4 expression after fgf8 injection on the injected side. Surprisingly, none of the injected mutant embryos re-expressed nkx2.5 in response to fgf8 overexpression by RNA injection (n=26; but see below); lacZ control injections gave no effect (not shown).

While gata4 expression is restored in mutant embryos, neither nkx2.5 nor gata4 expression was induced outside their normal expression domain following fgf8 RNA injection into wild-type or ace mutant embryos (Fig. 4A and not shown). This suggests that competence to respond to Fgf8 signaling is restricted to these subregions of lateral plate mesoderm including the prospective heart primordium.

We suspected that the failure to rescue nkx2.5 expression in the mutant embryos by RNA injections could be due to the dorsalizing influence of fgf8 overexpression during gastrula stages (Fürthauer et al., 1997; Reifers et al., 1998). To examine this possibility further and to test whether Fgf8 signaling could occur directly within the heart field during postgastrulation stages, we implanted Fgf8 protein-coated beads at the 5- to 6-somite stage into the lateral plate mesoderm of wild-type and acerebellar mutant embryos. 8 out of 10 operated mutant embryos strongly reexpressed nkx2.5 on the bead side; implantation of PBS beads (n=6) gave no effect (Fig. 4E,F). Interestingly, implantation of Fgf8-soaked beads into wild-type embryos also results in a slight posterior expansion of the endogenous nkx2.5 domain (Fig. 4G). These data indicate that gata4 expression is dependent on Fgf8 signaling, while nkx2.5 expression requires in addition to Fgf8 another factor, which may have been delocalized during gastrulation by the dorsalizing effect of fgf8 RNA injections (see Discussion). Taken together, these experiments show that Fgf8 signaling is required but not sufficient for gata4 and nkx2.5 expression during early somitogenesis stages.

To further study the requirement of Fgf signaling for the expression of nkx2.5 and gata4, we analysed the expression in wild-type embryos where Fgf signaling has been blocked either pharmacologically or by injecting RNA encoding the dominant negative Fgf receptor XFD (Amaya et al., 1991). 32 out of 73 (44%) injected wild-type embryos show a strong reduction or even absence of gata4 expression. In the case of nkx2.5, 67% (35 out of 52) of the XFD-injected embryos have altered gene expression (Fig. 4C,D). Interestingly, the described effects could only be observed if the misexpressed RNA was located in mesodermal cells (data not shown), suggesting that Fgf signal reception is required only within the mesodermal layer. While these results are consistent with a dependence of nkx2.5 and gata4 expression on Fgf signaling in the heart field, they could also be explained by an earlier requirement for Fgf signaling in dorsoventral patterning during gastrulation. To test whether Fgfs are specifically required after gastrulation during early somitogenesis for heart development, we treated wild-type embryos with SU5402, a potent inhibitor of Fgfr1 function (Mohammadi et al., 1997). Since SU5402 blocks Fgfr1 activity by binding to a region identical in all four Fgfrs (Johnson and Williams, 1993), it probably blocks all Fgf signals, including Fgf8. Inhibitor treatment during early somitogenesis results in a phenocopy of the acerebellar heart phenotype (not shown), including failure to initiate nkx2.5 expression (Fig. 4I-K). This confirms that Fgf signaling is required specifically during early somitogenesis for initiation of heart development. As in acerebellar homozygous embryos (see above, Fig. 3F and Table 1), nkx2.5 expression recovers after transient treatment in early somitogenesis (Fig. 4J,M). Continuous inhibitor treatment, however, prevents this recovery (Fig. 4K,N), suggesting that recovery is due to other Fgfs that act during later stages of heart development.

fgf8 itself is later expressed specifically in ventricular tissue (Fig. 2F), and acerebellar mutants show strongly impaired ventricle development (Fig. 1O,P). The ventricle defect could either be a consequence of abnormal formation of the primordium, or could represent an independent later function of Fgf8. To distinguish these possibilities, Fgf signaling was inhibited by SU5402 treatment after initial primordium formation, from late somitogenesis onwards. This treatment results in absence of ventricular tissue and apparent enlargement of atrial tissue, indicating that Fgf signaling continues to be required specifically for ventricular development (Fig. 4O,P).

Our analysis shows that fgf8 is expressed in and required for the development of cardiogenic precursors of the zebrafish heart. Homozygous acerebellar/fgf8 mutants fail to initiate proper gene expression of the cardiac transcription factors nkx2.5 and gata4, resulting in a severely malformed heart. Protein bead implantations and receptor inhibition show that Fgf8 can act directly on the myocardial primordium during postgastrulation stages. We therefore propose that Fgf8 functions in induction of the earliest cardiac gene expression (Fig. 4H).

In addition to its role in cardiac precursor development (see below), fgf8 is more specifically required in the ventricle of the zebrafish heart at later stages, since this is the most strongly affected structure in acerebellar embryos. This finding is consistent with the predominant expression of fgf8 in the ventricle. In addition to fgf8, fgf1, fgf7 and fgf12 are reported to be expressed in a chamber-specific manner. Additional fgfs are expressed during heart development (fgf2, 4, 13 and 16) (Parlow et al., 1991; Spirito et al., 1991; Engelmann et al., 1993; Mason et al., 1994; Zhu et al., 1996; Hartung et al., 1997; Miyake et al., 1998). Although data concerning their function in heart development are lacking, the presence of these additional Fgfs may account for the variability of the acerebellar heart phenotype, since the other Fgfs might partially compensate for the lack of Fgf8. The stronger effect on nkx2.5 expression in myocardial precursors of inhibitor-treated embryos compared to the expression in acerebellar mutants, and the loss of ventricular tissue after late inhibitor treatment are consistent with this possibility.

Our analysis of the vascular system in acerebellar mutants shows that the defect in circulation is largely, if not exclusively, due to the severely dysmorphic heart. This is consistent with the fact that fgf8 is expressed during several stages of heart development. fgf8 expression is closely associated with some or all heart precursor cells from blastula stage onwards and is later on expressed in the ventricle of the heart proper. Since Fgf8 is likely to be a secreted molecule (Baird, 1994; Fernig and Gallagher, 1994; Fürthauer et al., 1997; Reifers et al., 1998), it may very well function in signaling processes in the heart mesoderm. Consistent with this possibility, the Fgf receptors fgfr1, fgfr2 and fgfr4 are expressed during vertebrate heart development (Orr-Urtreger et al., 1991; Peters et al., 1992; Sugi et al., 1995; Thisse et al., 1995). Functional inactivation of the four mouse Fgf receptors causes either no heart phenotype or results in early embryonic lethality preventing analysis of heart development (Yamaguchi et al., 1991; Deng et al., 1994; Arman et al., 1995; Colvin et al., 1996; Weinstein et al., 1998). Our study predicts that Fgfrs probably also function in heart development, possibly in a redundant fashion. In vitro binding studies of Fgf8 protein to Fgfrs suggest Fgfr3 and Fgfr4 as the most likely candidates (MacArthur et al., 1995; Ornitz et al., 1996; Blunt et al., 1997).

Our results demonstrate that Fgf8 is necessary for initiation of cardiac gene expression, since the onset and early expression of gata4, nkx2.5 and, less severely, nkx2.7 in the heart precursor field are affected in acerebellar mutant embryos. acerebellar embryos also display minor defects in mesoderm during gastrulation (Reifers et al., 1998). These defects are however unlikely to account for the heart phenotype described here, since fgf8 is expressed in and can function independently in the developing heart precursors. The results of our Fgf8 bead implantations and of the inhibitor treatment both argue that Fgf8 functions in the heart primordium independently of its expression in general endomesodermal precursors during gastrulation (Reifers et al., 1998). Fgf8 function is required for the expression of cardiac genes for different periods of time: from its onset at the tailbud/1 somite stage, nkx2.7 is very transiently affected, whereas gata4 and nkx2.5 are strongly affected until mid- and late somitogenesis, respectively. After this early requirement, the expression recovers at least in the remaining heart cells of acerebellar mutants. This recovery most likely reflects functional redundancy with other Fgf family members at later stages in cardiac mesoderm. Recent studies in frog and mouse demonstrated that the tinman-related Nkx genes in vertebrates are essential for heart development (Grow and Krieg, 1998; Tanaka et al., 1999). In addition, it has been shown in vitro that GATA4 enhances cardiogenesis (Grepin et al., 1997). Since GATA4 may cooperate with Nkx2.5 in activating downstream cardiac genes, and since GATA4 can act as a transcriptional activator of cardiac Nkx2.5 expression (Durocher et al., 1997; Durocher and Nemer, 1998; Searcy et al., 1998; Lien et al., 1999), it appears that these two transcription factors work within a transcriptional network to drive cardiac development. Since both factors are affected very early in acerebellar mutant embryos, we suggest that the failure of early myocardial nkx2.5 and gata4 expression may largely account for the severe malformation of the mutant hearts at later stages.

gata6, another cardiac transcription factor of the GATA family (Kelley et al., 1993; Heikinheimo et al., 1994; Laverriere et al., 1994; Jiang and Evans, 1996), is not affected in acerebellar mutants. gata6 expression may therefore occur independently of Fgf signaling. The presence of gata6-positive cells also demonstrates that the failure to activate proper expression of nkx2.7, nkx2.5 and gata4 is not due to the absence of the cardiac precursors, a possibility that was raised by Fgf involvement in cell migration processes (Bodmer and Venkatesh, 1998; Sun et al., 1999).

As the loss-of-function situation in acerebellar mutants clearly demonstrates, gata4 expression is initially dependent on Fgf8 function. In addition, fgf8 is sufficient to restore gata4 expression in the endogenous domain. This suggests that Gata4 may be a target for Fgf8 signaling. Interestingly, however, fgf8 RNA injections cannot restore nkx2.5 expression, while a bead applied after gastrulation as a localized source for Fgf8 protein is able to do so. One possible explanation is that an additional factor is needed for proper nkx2.5 expression, which might have become delocalized or suppressed by the dorsalizing effect that fgf8 RNA overexpression has on the zebrafish gastrula (Fürthauer et al., 1997; Reifers et al., 1998). A candidate for this additional signal is BMP2, since BMP2-coated beads or BMP2-producing cells can induce Nkx2.5 expression in chicken embryos (Schultheiss et al., 1997; Andrée et al., 1998). Previous exposure to BMP2 or another BMP signal in gastrulation could therefore be necessary for responsiveness of nkx2.5 to Fgf8 induction. Our observation that nkx2.5 expression is not restored by fgf8 RNA injections into acerebellar mutants is consistent with this possibility, since fgf8 overexpression during gastrulation suppresses bmp2 expression (Fürthauer et al., 1997), thereby eliminating the required second signal at early stages. In Fgf8 bead implanted embryos, bmp2 expression in gastrulation is not affected due to the later stage of implantation, hence allowing rescue of nkx2.5 expression (Fig. 4E).

Several lines of evidence suggest a role for BMP2 in cardiogenesis. (i) BMP2 is expressed close to precardiac mesoderm and later on in the heart (Lyons et al., 1989, 1990; Schultheiss et al., 1997; Nikaido et al., 1997; Andrée et al., 1998). BMP2 mutant embryos show cardiac defects, and a subset of the BMP2 mutant embryos does not express Nkx2.5. BMP2 may therefore be directly required for cardiogenesis, although the alternative possibility that the heart abnormalities are secondary to altered patterning of the gastrula cannot yet be ruled out (Hogan, 1996; Zhang and Bradley, 1996; Kishimoto et al., 1997). (ii) The Nkx2.5 heart enhancer contains partial consensus sequences for the Smad transducers of BMP signaling. Although this element also contains GATA binding sites that are required for expression (Searcy et al., 1998; Lien et al., 1999), they cannot be sufficient since Fgf8-driven gata4 expression does not restore nkx2.5 expression in acerebellar embryos. (iii) The general idea that Nkx2.5 expression is regulated by the BMP pathway is also suggested by Drosophila studies (Frasch, 1995). (iv) In vertebrates, Lough et al. (1996) demonstrated in vitro that the combined action of BMP2 and FGF4, but neither factor alone, promotes cardiogenesis in non-precardiac mesodermal explants. Taken together, these data and our observations suggest that Bmp2 may be the second signal needed and that Fgf8 cooperates with Bmp signals to initiate cardiogenesis (Fig. 4H). A similar, but antagonistic cooperation between FGF and BMP signals has been suggested to occur in tooth development (Neubüser et al., 1997).

Interestingly, Fgf8 may be differentially required for the expression of nkx2.5. The caudal part of the nkx2.5 expression appears to be more dependent on functional Fgf8 than the rostral part, since this caudal subdomain is always affected more strongly in acerebellar mutants. These cells contribute to the otic vesicle (Serbedzija et al., 1998), which also shows malformations in acerebellar mutants (Brand et al., 1996; S. Léger and M. B., unpublished). The anterior notochord has been shown to produce an inhibitory effect on the caudal-most nkx2.5 expression (Goldstein and Fishman, 1998), and the possible differential requirement for Fgf8 in this domain suggests that Fgf8 may counteract this inhibitory signal from the notochord in wild-type embryos. This idea is further supported by our Fgf8 bead implantations into wild-type embryos, which result in a caudal extension of nkx2.5 expression. The same expansion can also be observed in ntl mutants lacking a normal notochord, or after notochord ablation in wild-type embryos (Goldstein and Fishman, 1998). Thus, in this circumstance an Fgf signal, probably Fgf8, is even sufficient to activate nkx2.5 expression in a longitudinal domain of the lateral plate mesoderm.

The involvement of Bmp and Fgf signals in cardiac induction may provide an explanation for the ability of the heart precursor field to regulate itself after laser ablation (Serbedzija et al., 1998), since non-ablated cells in the lateral plate mesoderm, which usually would not give rise to heart tissue, may now be driven into cardiac fate by the inductive signals they receive. Since fgf8 continues to be expressed in the heart, it may well perform, apart from its role in induction of early gene expression in heart precursors, additional functions in development of the heart, e.g. in polarization or control of proliferation, as has been proposed for other tissues (Picker et al., 1999). Human fgf8 mutations are not yet known, but given our findings it is conceivable that heart disease would be among the symptoms.

We thank members of the Brand and Stainier labs for discussion, Wieland B. Huttner, Jeremy Garwood and Deborah Yelon for comments on the manuscript, Heike Böhli for excellent technical assistance, Alexander Picker for help with confocal microscopy, Bernadette Fouquet for the microangiography protocol, Frank Stockdale for the gift of S46 antibody, and Shantha Shanmugalingam for help with the inhibitor protocol. This work was supported by grants from the Deutsche Forschungsgemeinschaft (SFB317) and the Förderprogramm Neurobiologie Baden-Württemberg (to M. B.), a HHMI predoctoral fellowship to E. C. W., and grants from the American Heart Association, the NIH and the Packard foundation (to D. Y. R. S.).