Bmp and Fgf signaling are essential for liver specification in zebrafish

Data generated using the mouse embryo indicate that the mammalian liver develops from the ventral foregut endoderm in response to inductive signals such as fibroblast growth factors (Fgfs) from the cardiac mesoderm and bone morphogenetic proteins (Bmps) from the septum transversum mesenchyme (STM)(Jung et al., 1999; Rossi et al., 2001; Calmont et al., 2006) (reviewed by Zaret, 2002; Zhao and Duncan, 2005). An in vitro mouse embryonic tissue explant system was used to show that Fgfs can replace the cardiac mesoderm to induce hepatic gene expression(Jung et al., 1999), that inhibitors of Fgf signaling block hepatic specification(Jung et al., 1999; Calmont et al., 2006), and that a Bmp inhibitor, noggin, blocks hepatic specification by the STM(Rossi et al., 2001). The crucial role of Fgf and Bmp signaling in hepatic specification was also revealed in chick embryos using chick tissue explant cultures and ex vivo cell/bead implantation experiments (Zhang et al., 2004). In tissue explants, FGF2 could substitute for the cardiac mesoderm to induce hepatic markers such as HHEX and albumin in the anterior lateral endoderm, and noggin blocked the expression of these markers (Zhang et al., 2004). Implantation of noggin-expressing cells into the anterior lateral regions of chick embryos inhibited HHEX expression in the anterior lateral endoderm, and that of BMP2-soaked beads induced ectopic HHEXexpression in the endoderm just posterior or lateral to the endogenous HHEX expression domain (Zhang et al., 2004).

While Foxa and Gata genes have been shown by genetic analysis to regulate the competence of foregut endodermal cells to respond to hepatic inductive signals (Bossard and Zaret,1998; Cirillo et al.,2002; Lee et al.,2005a), no hepatic inducer has been validated genetically. Mice carrying mutations in individual components of the Fgf or Bmp signaling pathways either do not survive to the point of hepatic specification(Yamaguchi et al., 1992; Deng et al., 1994) or exhibit no defects in hepatic specification, possibly due to compensation by other components of the pathway (Weinstein et al., 1998; Miller et al.,2000).

In addition to their role in hepatic specification, Fgf and Bmp signaling regulate the morphogenetic outgrowth of the hepatic endoderm. Data from the tissue explant system mentioned above indicated that FGF8 and BMP4 also contribute to the morphogenetic outgrowth of the hepatic endoderm(Jung et al., 1999; Rossi et al., 2001), and analysis of Bmp4 mutant mouse embryos provided genetic evidence for the role of Bmp4 in liver bud formation(Rossi et al., 2001).

In zebrafish, hepatoblast specification is thought to occur at approximately 22 hours post-fertilization (hpf) as marked by the localized endodermal expression of hhex and prox1(Ober et al., 2006), two transcription factor genes also expressed in mouse(Burke and Oliver, 2002) and chick (Zhang et al., 2004)hepatoblasts. Recently, the analysis of prometheus (prt; wnt2bb - ZFIN), a zebrafish mutant that exhibits profound defects in hepatoblast specification, was reported(Ober et al., 2006). The prt gene encodes a novel Wnt2b homolog, indicating that canonical Wnt signaling is essential for hepatoblast specification. However, no zebrafish mutants that implicate Fgf or Bmp signaling in hepatoblast specification have been reported thus far.

Both Fgf and Bmp signaling play crucial roles in gastrulation and embryonic patterning (Amaya et al., 1991; Kishimoto et al., 1997);therefore, it is essential to block Fgf and Bmp signaling after gastrulation in order to examine later developmental events such as hepatoblast specification. In addition, the functional redundancy that stems from the existence of multiple Fgf and Bmp ligands and receptors may complicate the identification of mutants defective in hepatoblast specification. Here, we have overcome these technical and biological challenges by using transgenic zebrafish lines that overexpress dominant-negative Bmpr1a(Pyati et al., 2005) or fgfr1 (Lee et al.,2005b) under the regulation of the heat shock cognate 70-kd protein (hsp70) promoter. The dnBMPR and dnFgfr1 proteins can completely block most, and possibly all, Bmp(Graff et al., 1994) and Fgf(Amaya et al., 1991) signaling,respectively, and hsp70 allows one to temporally control transgene expression (Halloran et al.,2000). We show that blocking Bmp or Fgf signaling resulted in profound defects in hepatoblast specification. Surprisingly, hepatocyte differentiation occurred in embryos temporarily lacking Bmp signaling,although with a substantial delay.

Embryos and adult fish were raised and maintained under standard laboratory conditions (Westerfield,2000). We used the following mutant and transgenic lines: laf^tm110b (Bauer et al., 2001; Mintzer et al.,2001), Tg(hsp70l:dnBmpr-GFP)(Pyati et al., 2006), Tg(hsp70l:dnfgfr1-EGFP)(Lee et al., 2005b), Tg(lfabp:dsRed) (Her et al., 2003), Tg(hsp70:alk8) and Tg(hsp70:bmp2b) ^fr13(Chocron et al., 2007).

Embryos were heat shocked at various stages by transferring them into a pre-warmed plate containing egg water on a heat block. Tg(hsp70l:dnBmpr-GFP) embryos were heat shocked for 25 minutes at 40°C; Tg(hsp70l:dnfgfr1-EGFP) embryos for 20 minutes at 37°C. After heat shock, the plate containing the embryos was transferred into a 28°C incubator, and embryos were harvested between 30 and 40 hpf. As GFP expression is maintained in these embryos for at least 1 day after heat shock, hemizygous embryos expressing GFP were easily sorted. Heat shock at 40°C (but not 37°C) substantially delays development of wild-type and hemizygous embryos, making it difficult to estimate the exact stage of embryos based on incubation time. Thus, we assigned stage based on gut morphology of wild-type embryos revealed by foxa3 expression(Field et al., 2003). To overexpress Bmp2b while blocking Fgf signaling, embryos were heat shocked for 20 minutes at 37°C. Hemizygous Tg(hsp70:bmp2b)^fr13 embryos were easily distinguished based on the yolk extension defect, and further tested for presence of the hsp70:bmp2b transgene by PCR using genomic DNA after in situ hybridization, using the following primers:5′-CATGTGGACTGCCTATGTTCATC-3′ and 5′-GAGAGCGCGGACCACGGCGAC-3′.

Whole-mount in situ hybridizations were performed as previously described(Alexander et al., 1998), using the following probes: hhex (Ho et al., 1999), prox1(Glasgow and Tomarev, 1998), gata4/5/6 (Reiter et al.,1999), foxa3(Odenthal and Nusslein-Volhard,1998), cp (Korzh et al., 2001) and pdx1(Milewski et al., 1998).

Wild-type embryos were injected at the one- or two-cell stage with 10 ng gata4 MO and/or 2.5 ng gata6 MO(Holtzinger and Evans, 2005),and they were assayed between 37 and 38 hpf.

Among the transcription factor genes expressed in the liver, hhexand prox1 are the earliest markers of liver development in zebrafish(Ober et al., 2003; Wallace and Pack, 2003; Ober et al., 2006). Their expression in the zebrafish liver is initiated at approximately 22 hpf and maintained thereafter (Ober et al.,2006). To determine whether Bmp signaling is essential for hepatoblast specification in vivo, we utilized Tg(hsp70l:dnBmpr-GFP) fish(Pyati et al., 2005). Embryos obtained from outcrossing a hemizygous Tg(hsp70l:dnBmpr-GFP)fish were heat shocked at 18 hpf, a stage before the onset of hhexand prox1 expression in the liver primordium. We found that the earliest liver markers, hhex and prox1, were barely expressed in the liver region of the heat-shocked hemizygous embryos(Fig. 1B,C,E,F, arrows),whereas hhex expression in the pancreatic islet and prox1expression in the interrenal primordium appeared unaffected(Fig. 1B,C,E,F, arrowheads). These data indicate that Bmp signaling occurring after 18 hpf is essential for hepatoblast specification. Hepatocyte differentiation, as assessed by ceruloplasmin (cp) expression, was also absent in the heat-shocked hemizygous embryos (Fig. 1Q). In order to visualize the entire endoderm, we examined the expression of the endodermal marker foxa3(Odenthal and Nusslein-Volhard,1998; Field et al.,2003). Leftward bending of the gut(Horne-Badovinac et al., 2003)was often defective in the heat-shocked hemizygous embryos(Fig. 1N,O; bracket). These data suggest that Bmp signaling is also required for the morphogenesis of the gut but not for the maintenance of endodermal gene (e.g. foxa3)expression. As gata4 and gata6 are expressed in the liver primordium (see Fig. S1 in the supplementary material), are essential for hepatocyte differentiation in zebrafish(Holtzinger and Evans, 2005)and have been associated with Bmp signaling(Rossi et al., 2001; Zaret, 2002; Zhao and Duncan, 2005), we examined their expression in the heat-shocked hemizygous embryos. Both gata4 and gata6 were strongly expressed in the liver region and weakly expressed in the intestinal endoderm in wild-type embryos(Fig. 1G,J, arrows and brackets). In the heat-shocked hemizygous embryos, the expression of gata4 and gata6 in the liver region was greatly reduced but not completely abolished, whereas the expression of these genes in the intestinal endoderm appeared relatively less affected(Fig. 1H,I,K,L, arrows versus brackets). These data suggest that Bmp signaling may regulate gata4and gata6 expression in the liver primordium.

As our data indicated that Bmp signaling was essential for hepatoblast specification in zebrafish and a mutation in alk8 (also known as laf, acvr1 - ZFIN), a type I Bmp receptor, was identified in a forward genetic screen for genes involved in endodermal organ morphogenesis(E.A.O., H. Verkade, H. A. Field, P. D. Si Dong, P. Aanstad, T. Sakaguchi, M. Bagnet, C. Munson, W.-S. Chung, C.H.S., S. Curado, R. Anderson, J. Frantsve,D. Beis, T. Bartman and D.Y.R.S., unpublished), we next examined liver development in alk8 mutant embryos(Bauer et al., 2001; Mintzer et al., 2001). The expression of hhex and prox1 appeared greatly reduced in the mutants at 26 and 34 hpf compared with their wild-type siblings(Fig. 2A-D, arrows; data not shown), but the reduction appeared less severe than that seen in embryos in which Bmp signaling was blocked at 18 hpf(Fig. 2B,D versus Fig. 1C,F, arrows), suggesting that other Bmp receptors or maternally deposited Alk8 might compensate for the loss of zygotic Alk8 function in hepatoblast specification. The expression of gata4, gata6 and cp in alk8 mutant embryos appeared similar to that in embryos in which Bmp signaling was blocked at 18 hpf(Fig. 2F,H,L), further suggesting that Alk8 is part of the Bmp signaling machinery regulating hepatoblast specification. However, the leftward bending of the gut appeared unaffected in alk8 mutant embryos(Fig. 2J, bracket), suggesting that other Bmp receptors or maternally deposited Alk8 function might compensate for the loss of zygotic Alk8 function in this morphogenetic process.

To delineate the time-window during which Bmp signaling is required for hepatoblast specification, as well as investigate whether Bmp signaling is required for the maintenance of the specified liver progenitors, we also blocked Bmp signaling at 22, 24, 26 and 30 hpf. When Bmp signaling was blocked at 22 hpf, hhex and prox1 expression in the liver region was greatly reduced (Fig. 3B,C,H,I,arrows). However, this expression was slightly higher than that seen in embryos heat shocked at 18 hpf (Fig. 3B,H versus Fig. 1B,E). In addition, the percentage of embryos that showed severely reduced expression of hhex and prox1 in the liver region was lower than that seen in embryos heat shocked at 18 hpf[Fig. 3C (33%) versus Fig. 1C (56%); Fig. 3I (37%) versus Fig. 1F (50%)]. Furthermore,the expression of gata6 and cp in the 22 hpf heat-shocked hemizygous embryos was slightly higher than that in the embryos heat shocked at 18 hpf (Fig. 3N; see Fig. S2G,H in the supplementary material; Table 1). When Bmp signaling was blocked at 24 hpf, hhex and prox1 expression in the liver region was reduced but clearly present in all embryos (data not shown). Altogether, these data suggest that Bmp signaling between 18 and 22 hpf is crucial for hepatoblast specification. When Bmp signaling was blocked at 26 hpf, hhex and prox1 were clearly expressed in the liver region, although their expression was reduced compared with wild-type siblings (Fig. 3D-F,J-L, arrows); gata4 and cp but not gata6 expression was also reduced in the heat-shocked hemizygous embryos (Fig. 3Q,R; see Fig. S2D,E,J,K in the supplementary material; Table 1). However, when Bmp signaling was blocked at 30 hpf, gata4 and cp expression appeared unaffected (data not shown). Taken together, these data suggest that Bmp signaling continues to regulate liver development after its initial role in hepatoblast specification.

We next examined the timing of Alk8-mediated Bmp signaling in liver development by rescuing the liver defects of alk8 mutant embryos with an hsp70:alk8 transgene. When embryos from alk8^+/- females crossed with homozygous Tg(hsp70:alk8);alk8^+/- males were heat shocked at 10 hpf, the Tg(hsp70:alk8) mutant embryos appeared unaffected morphologically, whereas the mutant siblings that were not heat shocked exhibited cardiac defects (data not shown). As Bmp signaling between 18 and 22 hpf is crucial for hepatoblast specification, we also heat shocked embryos at 18 hpf. The cardiac defects were not rescued in these heat-shocked alk8 mutant embryos, facilitating their identification. We found that in alk8 mutant embryos heat shocked at 18 hpf, the expression of hhex, prox and cp in the liver region appeared unaffected (Fig. 4C,I,O,arrows) compared with wild-type siblings(Fig. 4A,G,M, arrows). To delineate the time-window when the liver defects could be rescued in alk8 mutant embryos, we heat shocked the embryos at later stages (26 and 34 hpf). Surprisingly, the expression of hhex, prox and cp could still be rescued when the embryos were heat shocked at 34 hpf (Fig. 4F,L,R, arrows). These data suggest that the alk8 mutant endodermal cells remain competent to respond to Bmp signaling for at least 16 hours after Bmp signaling is first required for hepatoblast specification.

Gata transcription factors have been reported to play an essential role in rendering endodermal cells competent to become hepatocytes by unfolding chromatin (Bossard and Zaret,1998; Cirillo et al.,2002). Among Gata transcription factor genes in zebrafish, gata4 and gata6, but not gata5, are expressed in the hepatic endoderm (see Fig. S1 in the supplementary material; data not shown). When both genes were knocked down by morpholino oligonucleotides(MOs), hepatocyte differentiation, as assessed by transferrin(transferrin-a - ZFIN) expression at 3 days post-fertilization (dpf),was completely abolished (Holtzinger and Evans, 2005). In embryos in which Bmp signaling was blocked at 18 hpf, gata4 and gata6 expression was greatly reduced in the liver region (Fig. 1H,I,K,L,arrows). Thus, we hypothesized that the hepatoblast specification defect in embryos lacking Bmp signaling might result from the reduction of gata4 and gata6 expression, causing endodermal cells to lose competence to become hepatoblasts. To test this hypothesis, we injected gata4 and gata6 MOs, and examined the expression of the hepatoblast specification markers, hhex and prox1, and a differentiation marker, cp, at 37-38 hpf. Injections of gata4 or gata6 MO fully blocked hepatocyte differentiation in 63 and 71% of the embryos, respectively; injections of both MOs fully blocked hepatocyte differentiation in 95% of the embryos(Fig. 5L-N). Injections of gata4, gata6 or both MOs did not fully block hepatoblast specification (Fig. 5B-E,G-J),but the double MO injections significantly reduced hhex expression in 50% and prox1 expression in 69% of the embryos(Fig. 5E,J). The less severe defects in the double morphants than in embryos lacking Bmp signaling could be due to variation in the efficacy of the MO-based depletion. Altogether, these data suggest that the hepatoblast specification defect in embryos lacking Bmp signaling may at least partially result from the reduction of gata4and gata6 expression.

To determine whether Fgf signaling is essential for hepatoblast specification in vivo, we utilized Tg(hsp70l:dnfgfr1-EGFP)fish (Lee et al., 2005b). Embryos obtained from outcrossing a hemizygous Tg(hsp70l:dnfgfr1-EGFP) fish were heat shocked at 18 hpf, a stage before the onset of hhex and prox1 expression in the liver primordium. We found that hhex and prox1 expression in the liver region was almost absent in the heat-shocked hemizygous embryos,whereas hhex expression in the pancreatic islet and prox1expression in the interrenal primordium appeared only weakly affected(Fig. 6B,C,E, arrows versus arrowheads), suggesting that Fgf signaling after 18 hpf is essential for hepatoblast specification. Hepatocyte differentiation, as assessed by cp expression, was also absent in the heat-shocked hemizygous embryos(Fig. 6P). The leftward bending of the gut did not occur in the heat-shocked hemizygous embryos(Fig. 6M,N; brackets),suggesting that Fgf signaling is also required for the morphogenesis of the gut. The expression of gata4 and gata6 in the liver region was greatly reduced in the heat-shocked hemizygous embryos, whereas their expression in the intestinal endoderm appeared relatively less affected(Fig. 6G,H,J,K, arrows versus brackets), suggesting that Fgf signaling may also regulate gata4 and gata6 expression in the liver primordium. In general, the liver defects in embryos in which Fgf signaling was blocked at 18 hpf were similar to those in embryos in which Bmp signaling was blocked at the same stage.

We next examined the time-window during which Fgf signaling is required for hepatoblast specification by blocking Fgf signaling at later stages. When Fgf signaling was blocked at 22 hpf, hhex and prox1 expression in the liver region was greatly reduced(Fig. 7B,C,H, arrows). However,this expression was slightly higher than that seen in embryos heat shocked at 18 hpf (Fig. 7B,H versus Fig. 6B,E), and the percentage of embryos that barely expressed hhex in the liver region was lower than that in embryos heat shocked at 18 hpf[Fig. 7C (44%) versus Fig. 6C (67%)]. In addition,the expression of gata4, gata6 and cp in the 22 hpf heat-shocked hemizygous embryos was slightly higher than that in the embryos heat shocked at 18 hpf (Fig. 7M,N; see Fig. S3B,C,H in the supplementary material; Table 1). Altogether, these data suggest that Fgf signaling between 18 and 22 hpf plays an essential role in hepatoblast specification.

As hhex and prox1 expression in the liver primordium is initiated approximately at 22 hpf (Ober et al., 2006), we blocked Fgf signaling at 26 hpf to determine whether Fgf signaling is required for the maintenance of the specified liver progenitors. hhex and prox1 were clearly expressed in the liver region in the hemizygous embryos heat shocked at 26 hpf, although their expression was reduced compared with wild-type siblings(Fig. 7D-F,I-K, arrows); gata4, gata6 and cp expression were also reduced(Fig. 7P,Q; see Fig. S3E,F,J,K in the supplementary material; Table 1). However, when Fgf signaling was blocked at 30 hpf, gata4 and cp expression appeared unaffected (data not shown). These data suggest that Fgf signaling continues to regulate liver development after its initial role in hepatoblast specification.

Our data indicate that both Bmp and Fgf signaling are essential for hepatoblast specification in zebrafish; therefore, we next wanted to investigate the relationship between these signaling pathways in this process. As Bmp signaling is essential for early embryonic patterning, we utilized a transgenic line, Tg(hsp70:bmp2b)^fr13, to enhance Bmp signaling after gastrulation. Embryos obtained from crossing a hemizygous Tg(hsp70:bmp2b)^fr13 fish with a hemizygous Tg(hsp70l:dnfgfr1-EGFP) fish were heat shocked at 18 hpf. Surprisingly, we found that hhex and prox1expression in the liver region mostly recovered in the heat-shocked double hemizygous embryos (Fig. 8E,F,J,K, arrows), whereas their expression was greatly reduced in the heat-shocked single hemizygous embryos lacking Fgf signaling(Fig. 8C,D,I, arrows). Furthermore, cp expression in the liver region mostly recovered in a majority of the double hemizygous embryos compared with the single hemizygous embryos lacking Fgf signaling (Fig. 8P,Q versus N,O). These data indicate that overexpressing bmp2b can partially compensate for the loss of Fgf signaling during hepatoblast specification.

The fact that the liver defects in alk8 mutant embryos can be rescued by overexpressing wild-type alk8 as late as 34 hpf suggests that in the absence of hepatic inductive signals endodermal cells maintain competence to give rise to hepatocytes. We further investigated this hypothesis by using Tg(hsp70l:dnBmpr-GFP) fish. Embryos obtained from outcrossing a hemizygous Tg(hsp70l:dnBmpr-GFP)fish were heat shocked at 18 hpf and harvested at 2, 3, 4, 5 and 6 dpf. The GFP signal of the heat-shocked hemizygous embryos became undetectable under a fluorescence microscope about 36 hours after the heat shock. The hepatocyte differentiation marker, cp, was barely expressed at 2 dpf in embryos transiently lacking Bmp signaling (Fig. 9B), but was clearly expressed at 3 dpf in 20% of the embryos(Fig. 9E). The percentage of embryos exhibiting distinct, hepatocyte cp expression increased to 60% at 4 dpf and 100% at 5 dpf (Fig. 9H and data not shown). Note the similar size of the yolk between wild-type and heat-shocked hemizygous embryos at 2, 3 and 4 dpf(Fig. 9A-H) as well as the similar size of the pectoral fins at 4 dpf(Fig. 9G,H, arrows), indicating that the delayed cp expression in the heat-shocked hemizygous embryos was not caused by developmental delay. We also examined the expression of another hepatocyte differentiation marker, fatty acid binding protein 1a,liver (fabp1a), using Tg(fabp1a:dsRed) fish(Her et al., 2003). DsRed expression, indicating fabp1a expression, in wild-type siblings was clearly detected starting at 3 dpf, while it was absent at this stage in embryos in which Bmp signaling had been blocked at 18 hpf(Fig. 9I, red circles versus squares). However, DsRed expression in embryos in which Bmp signaling had been blocked at 18 hpf was detectable in 33% of the embryos at 5 dpf, and in 90% of the embryos at 6 dpf (Fig. 9Jand data not shown). Thus, blocking Bmp signaling leads to an apparent lack of hepatoblast specification, as assessed by hhex and prox1expression, in about 50% of the embryos at 29-32 hpf(Fig. 1C,F, arrows). However,by 6 dpf, 90% of the embryos showed a clear presence of hepatocytes as assessed by DsRed expression, indicating that endodermal cells maintain competence to give rise to hepatocytes for longer than anticipated.

In this study, we first examined the role of Bmp and Fgf signaling in early liver development by utilizing Tg(hsp70l:dnBmpr-GFP) and Tg(hsp70l:dnfgfr1-EGFP) fish. By blocking Bmp or Fgf signaling at various stages, we generated in vivo evidence that Bmp and Fgf signaling are essential for hepatoblast specification. The essential role of Bmp signaling in early liver development is further supported by the observation that alk8 mutant embryos exhibit a severe reduction in the expression of several liver markers. Furthermore, our data revealed that foregut endodermal cells maintain competence to give rise to hepatocytes in the absence of Bmp signaling.

It has been suggested that Bmp signaling plays multiple roles in early liver development (Rossi et al.,2001). Initially, Bmp signaling appears to induce or maintain the expression of Gata4, a competence factor gene, in the mouse foregut endoderm, suggesting that it regulates hepatic competence. Subsequently, Bmp signaling appears to induce hepatic specification. Third, Bmp signaling is required for liver bud formation. The evidence supporting the first and second roles of Bmp signaling came from in vitro tissue explant experiments, and the evidence supporting its third role came from the analysis of Bmp4mutant mouse embryos combined with in vitro tissue explant experiments(Rossi et al., 2001). In our study, we provide genetic evidence for the role of Bmp signaling in hepatoblast specification. By blocking Bmp signaling at 18 and 22 hpf and by examining alk8 mutant embryos, we reveal the essential role of Bmp signaling in hepatoblast specification; by blocking at 26 and 30 hpf, after hepatoblast specification has occurred, we show that Bmp signaling is not essential for the maintenance of specified liver progenitor cells but continues to regulate liver development.

It has been proposed that Fgf signaling plays at least two roles in early liver development based on in vitro tissue explant experiments(Jung et al., 1999). FGF1 and FGF2 appear to be sufficient to induce hepatic specification; FGF8 appears to regulate the outgrowth of the hepatic endoderm. However, no liver defect has been described in Fgf1;Fgf2 double mutant mice(Miller et al., 2000), and there are no known mouse mutations affecting Fgf signaling that block hepatic specification or the outgrowth of the hepatic endoderm. Because the mouse genome appears to encode 22 Fgfs and four Fgf receptor tyrosine kinases(Itoh and Ornitz, 2004), it is likely that functional redundancy of these molecules will prevent one from finding severe liver defects in single or double mutant mice. A recent study in mouse (Calmont et al., 2006)showed that overexpressing Spry2, which inhibits signaling from receptor tyrosine kinases, including Fgf receptors, in nascent hepatic cells decreases their survival. However, culturing mouse embryos in the presence of Fgf inhibitors did not appear to affect endodermal cell survival. Here, we provide in vivo evidence for the role of Fgf signaling in hepatoblast specification. By blocking Fgf signaling at 18 and 22 hpf, we reveal the essential role of Fgf signaling in hepatoblast specification; by blocking at 26 and 30 hpf, after hepatoblast specification has occurred, we show that Fgf signaling is not essential for the maintenance of specified liver progenitors but continues to regulate liver development.

While data in mouse support the argument that the Bmp and Fgf signals involved in hepatic specification are expressed in the STM(Rossi et al., 2001) and cardiac mesoderm (Jung et al.,1999), respectively, data in chick show expression of BMP2 and several Fgfs (FGF1, 2, 8 and 12) not in the mesodermal tissues surrounding the hepatogenic endoderm but in the hepatogenic endoderm itself (Zhang et al.,2004). Our data do not provide information about the spatial expression of the inducers; thus, we have to consider Bmp and Fgf genes,expressed in the endoderm and/or surrounding mesoderm at the time of hepatoblast specification. Although the expression pattern of all Bmps (at least seven) and Fgfs (at least 22) has not been investigated, a number of Bmps (bmp2a, 2b, 5 and 6) and Fgfs (fgf8, 10, 17band 24) appear to be expressed in the mesoderm and/or endoderm of the liver-forming region around the time of hepatoblast specification (ZFIN expression database). Furthermore, genetic data have suggested that Gdf6a(Radar), a member of the TGF-β superfamily, is a ligand of Alk8(Sidi et al., 2003) and gdf6a is also expressed in the endoderm at the time of hepatoblast specification (ZFIN expression database); thus, not only Bmps but other members of the TGF-β superfamily have to be considered for their role in hepatoblast specification.

The fact that Wnt, Bmp and Fgf signaling are essential for hepatoblast specification in zebrafish raises the question of whether they function in the same or parallel pathways. Data from tissue explant experiments showing that exogenous addition of FGF2 was not able to induce hepatic gene expression in the presence of the Bmp inhibitor noggin(Jung et al., 1999; Rossi et al., 2001), indicate that Fgf signaling does not function downstream of Bmp signaling. Our studies showing that Bmp2b can partially compensate for the loss of Fgf signaling also suggest that Fgf signaling does not function downstream of Bmp signaling. However, in order to define the epistatic relationship between these two signaling pathways, it will be necessary to identify the Bmp and Fgf ligands required for hepatoblast specification and to analyze their expression in embryos lacking Fgf and Bmp signaling, respectively.

It will also be important to address whether Wnt signaling functions in parallel with, downstream of, or upstream of Bmp and Fgf signaling. Interestingly, another zebrafish homolog of Wnt2b, wnt2ba, which appears to be essential for pectoral fin development, appears to regulate the expression of Bmp and Fgf genes (Ng et al., 2002; Mercader et al.,2006). Initially, tbx5 expression in the lateral plate mesoderm appears to be induced by wnt2ba-mediated signaling and appears to induce the expression of fgf24 and other Fgf genes. Fgf signaling in turn appears to induce prdm1 expression, which subsequently appears to induce fgf10 and bmp2b expression in the lateral plate mesoderm. These data indicate that Wnt signaling is upstream of both Bmp and Fgf signaling in pectoral fin development(Ng et al., 2002; Mercader et al., 2006). wnt2bb is essential for liver specification(Ober et al., 2006), and we tested whether it lies downstream of Bmp or Fgf signaling in this process. We found that wnt2bb expression in the lateral plate mesoderm in embryos in which Bmp or Fgf signaling was blocked at 18 hpf appeared unaffected (data not shown), suggesting that Wnt signaling does not lie downstream of Bmp or Fgf signaling in hepatoblast specification.

In the absence of hepatic inductive signals, endodermal cells have at least three choices: adopt a different fate such as pancreas, maintain an uncommitted, competent state, or die. Data from in vitro culture experiments indicated that endodermal cells adopt a pancreatic fate in the absence of Bmp(Rossi et al., 2001) or Fgf signaling (Deutsch et al.,2001); those from in vivo mosaic analyses suggested that endodermal cells die in the absence of Fgf signaling(Calmont et al., 2006). However, we observed that expression of pdx1, used as a pancreatic marker in the in vitro culture experiments, was not expanded into the liver-forming region in embryos lacking Fgf signaling (data not shown) and that hepatocyte differentiation occurred in alk8 mutant embryos overexpressing wild-type alk8 at 34 hpf(Fig. 4R) as well as in embryos transiently lacking Bmp signaling (Fig. 9). Our data suggest that endodermal cells maintain an uncommitted, competent state in the absence of inductive signals.

Because there is weak expression of hhex and prox1 in alk8 mutant embryos (Fig. 2B,D, arrows), it is not clear whether the recovery of hepatocyte differentiation in those embryos results from the proliferation of a few specified hepatoblasts or from de novo hepatoblast specification, or a combination of both. Half the embryos temporarily lacking Bmp signaling showed a complete absence of hhex and prox1 expression in their liver-forming region at 30-32 hpf (Fig. 1C,F). Yet hepatocyte differentiation eventually occurred in a great majority of those embryos, suggesting delayed hepatocyte specification. Altogether, these data indicate that endodermal cells in the liver-forming region remain competent to differentiate into hepatocytes, an observation that may help explain why some wnt2bb mutants eventually form a liver(Ober et al., 2006).

We thank Ujwal Pyati and David Kimelman for the Tg(hsp70l:dnBmpr-GFP) line; Todd Evans for gata4and gata6 morpholinos; Ana Ayala and Steve Waldron for excellent fish care; Chantilly Munson for discussions and critical readings of the manuscript; and other Stainier lab members for technical help and discussion. D.S. was supported by an NIH institutional NRSA training grant (NIH 5T32HL007544), C.H.S. by an NIH postdoctoral fellowship (DK068891) and E.A.O. by the UCSF Liver Center through an NIH pilot feasibility grant. This work was supported in part by grants from the NIH (NIDDK) and the Packard Foundation to D.Y.R.S.