Initial anteroposterior pattern of the zebrafish central nervous system is determined by differential competence of the epiblast

Early in development, the vertebrate central nervous system (CNS) is regionalized along the anteroposterior (A-P) axis into forebrain, midbrain, hindbrain and spinal cord. Based upon analyses using amphibian embryos, it has been proposed that the A-P neural pattern is induced by the combined action of two signals produced by the organizer-derived dorsal mesoderm (two-step model; Nieuwkoop, 1950; Toivonen and Saxén, 1968). The first signal (activator) initiates neural development, inducing neural tissue of an anterior type and the second signal (transformer) converts the neural tissue induced by the first signal into progressively more posterior types of neural tissue (hindbrain and spinal cord) with increasing concentration. A number of candidates for the first signal have been identified recently, namely Noggin, Follistatin and Chordin (Lamb et al., 1993; Hemmati-Brivanlou et al., 1994; Sasai et al., 1994). Each of these proteins is expressed in the dorsal mesoderm during gastrulation, as appropriate for neural-inducing molecules, and each induces the formation of neuroectoderm with an anterior character. As for the second signal, retinoic acid (Conlon, 1995; Blumberg et al., 1997), FGFs (Kengaku and Okamoto, 1993, 1995; Lamb and Harland, 1995; Cox and Hemmati-Brivanlou, 1995) and Wnts (McGrew et al., 1995) have been suggested as candidate molecules. In combination or alone, these factors on Xenopus gastrula ectoderm generate a wide range of neural tissues along the A-P axis (For review, see Doniach, 1995), supporting the notion that the nature of inducers determines the type of the neural tissues. This notion, however, can not fully explain the results obtained in classical experiments of organizer transplantation: whenever a secondary axis is induced, its A-P polarity is the same as that of the primary axis. Thus, its anterior end faces the host animal pole, suggesting that A-P positional values are present throughout the embryo including the ventral region. These observations have led us to examine the origin of A-P polarity in vertebrate CNS using the zebrafish embryo.

During embryogenesis in the zebrafish, the first morphological indicator of axial asymmetry appears at the beginning of gastrulation, with a thickening called the ‘embryonic shield’ on the dorsal side of the marginal zone. The shield is believed to be equivalent to the amphibian ‘organizer’ or amniote ‘node’. This is supported by expression in the shield of genes also expressed in the Xenopus organizer and mouse or chicken node, such as goosecoid (gsc; Blum et al., 1992; Cho et al., 1991; Izpisua-Belmonte et al., 1993; Stachel et al., 1993), lim1 (Taira et al., 1993; Barnes et al., 1994; Toyama et al., 1995) and chordin (Sasai et al., 1994; Schulte-Merker et al., 1997) and by the ability of the shield to induce secondary axis when transplanted in the ventral region (Oppenheimer, 1936a; Shih and Fraser, 1996). Recently, Sagerström et al. (1996) analyzed shield function in vitro using an explant assay: the shield preferentially induced anterior neural markers when conjugated to animal caps, while the shield itself formed an axis with A-P neural pattern. However, the mechanisms that regulate zebrafish A-P axis formation in vivo are still unclear. For example, how much A-P information does the shield impart to the epiblast? Recent grafting experiments (Woo and Fraser, 1997) demonstrate that there is a signal from the germ ring that can impart anteroposterior pattern, but leave the nature of the signal unresolved. Does such an interaction impart a differential competence to the epiblast?

To investigate whether or not the response of zebrafish epiblast to inducing tissue depends on the position of the responding tissues along the A-P axis, we transplanted early organizer tissues into the ventral region of zebrafish embryos and related induced structures and gene expression patterns to the location of the grafts in the host. By performing this experiment, we were able to determine whether the epiblast cells had A-P information or whether such information could be imposed by the organizer. We present data showing that initial A-P patterning of zebrafish CNS depends on differential competence within the epiblast and that FGF-mediated signal produced by the non-axial marginal mesoderm is involved in the specification of the epiblast.

Zebrafish (Danio rerio) embryos were obtained from natural crosses of wild-type fish of the Oregon AB background. Collected embryos were maintained at 28.5°C and sorted into 1/3 Ringer (39 mM NaCl, 0.97 mM KCl, 1.8 mM CaCl₂, 1.7 mM HEPES at pH 7.2) and staged according to hours postfertilization at 28.5°C and morphological criteria (Kimmel et al., 1995).

For Hensen’s node (HN) transplantation, 50-100 μm fragments of chicken HN (stage 4), which included the most potent region, the medial sector of both the epiblast and mesenchyme (Storey et al., 1992), were manually removed and inserted into the ventral region of the zebrafish host during shield stage using a sharpened tungsten needle.

For transplantation of shield or ventral marginal tissues, fertilized eggs at the 2-to 8-cell stage were injected with a mixture of rhodamine- and biotin-dextran (Molecular Probes) (Miyagawa et al., 1996) into the yolk through a glass micropipette. The injected dye spreads through intercellular cytoplasmic connections to all cells of the blastoderm. When the embryonic shield became visible (6 h), shield or ventral marginal tissues were cut out with small scissors and then transplanted through a glass micropipette into the ventral region of a shield-stage host or animal pole of a sphere-stage host, respectively. Visualization of the labelled donor cells was conducted exactly as described previously (Westerfield, 1993; Miyagawa et al., 1996).

For transfection and transplantation of COS7 cells, we essentially followed the protocol described by Tonegawa et al. (1997). The plasmids that we used were pCDM8 (In Vitrogen) containing Xenopus noggin and chordin or lacZ cDNAs. The cell aggregate of approximately 50 μm diameter was used for transplantation. Transplantation of the yolk cell or blastomeres at the blastula stage was performed as described (Mizuno et al., 1996; Miyagawa et al., 1996). Grafted hosts were allowed to develop at 28.5°C for an appropriate period in 1/3 Ringer.

In some grafted zebrafish embryos, the graft was found dorsal in the vicinity of the original host axis and did not demonstrate its characteristic inductive activity. Therefore, we analyzed embryos only in which the graft remained in lateral or ventral regions.

Ventralized zebrafish embryos were obtained by the removal of the vegetal yolk hemisphere as described (Mizuno et al., 1997) with some modifications. The vegetal yolk mass (about half of the total yolk mass) was squeezed out through a small hole made at the vegetal end by a glass needle. The operation was carried out between 5 and 20 minutes after fertilization, which produced gsc-negative embryos at a high frequency (more than 80%; T. Mizuno et al., unpublished data).

Capped sense RNAs were synthesized using the MEGAscript large-scale transcription kit (Ambion) from the plasmid containing full-length Xenopus dominant negative FGF receptor (XFD) or HAVnot cDNA (Griffin et al., 1995). HAVnot RNA used as a control encodes nonfunctional FGF-R in which a mutation is introduced in the ligand-binding domain. The mRNAs synthesized were purified by gel filtration (NICK Column, Pharmacia Biotech) and diluted to 0.4-0.6 μg/μl with distilled water and injected into 1-cell-stage embryos. Injected embryos were cultured in 1/3 Ringer until use.

Whole-mount in situ hybridization was carried out as previously described (Takeda et al., 1994) with some modifications. Briefly, embryos fixed with 4% paraformaldehyde/PBS were dehydrated with methanol and rehydrated with PBS/0.1% Tween 20. They were then transferred to hybridization mixture (5× SSC, 50 μg/ml of heparin, 100 μg/ml of calf thymus DNA, 10 μg/ml of tRNA, 50% formamide, 0.1% Tween 20) and incubated for at least 1 hour. The hybridization mixture was then replaced with the mixture containing 100 ng/ml of DIG-labelled RNA probe (and fluorescein-labelled one for double in situ hybridization) and the embryos were incubated at 57°C overnight. After a rinse with 5× SSC, they were washed in 2× SSC/50% formamide at 65°C for 30 minutes, 2× SSC for 10 minutes and soaked in RNase buffer (0.5 M NaCl, 10 mM Tris-HCl at pH 8.0) for 10 minutes. Following the treatment with 20 μg/ml of RNase A at 37°C for 1 hour, a series of washings (2× SSC for 10 minutes; 2× SSC/50% formamide at 65°C for 30 minutes; 0.2× SSC at 55°C for 15 minutes) was performed. Embryos were rinsed with MAB(150 mM NaCl, 100 mM maleic acid at pH 7.5)/0.1% Tween 20 and soaked in blocking solution (2% FCS, 0.2% Tween 20, 0.2% TritonX-100 in MAB) at 4°C for 1 hour. The embryos were then incubated with alkaline-phosphatase (AP)-conjugated anti-DIG Fab fragments diluted 1:8000 in blocking solution at 4°C overnight. After washing three times with blocking solution for 30 minutes, the embryos were rinsed three times with AP reaction buffer (50 mM MgCl₂, 100 mM NaCl, 0.1% Tween 20, 1 mM levamisole, 100 mM Tris-HCl at pH 9.0) for 10 minutes. Detection was performed with BM-purple (Boehringer Mannheim). After stopping the coloring reaction, we washed the specimens three times with AP reaction buffer excluding levamisole and stored then in 4% paraformaldehyde/PBS.

Double in situ hybridization was performed using ELF-97 mRNA In Situ Hybridization Kit (Molecular Probes) as described in Jowett and Yan (1996).

Following in situ hybridization, some embryos were immediately stained with anti-Ntl antibody (Schulte-Merker et al., 1992). The samples were washed three times with MAB for 10 minutes and twice with MABD (1% BSA, 1% DMSO in MAB) for 30 minutes. After blocking with MABDN (2% FCS in MABD) for 30 minutes, they were incubated with rabbit anti-Ntl polyclonal antibody diluted 1:1000 in MABDN at 4°C overnight. Following a series of washings with MABD (three times for 5 minutes, 4 times for 30 minutes), the samples were blocked with MABDN for 30 minutes and then incubated with biotin-conjugated goat anti-rabbit IgG diluted 1:500 in MABDN at 4°C overnight. A series of washings was carried out as follows: MABD 5 times for 5 minutes, 4 times for 30 minutes, rinse with MAB and PBS, PBSDT (1% DMSO, 0.1% TritonX-100 in PBS) three times for 5 minutes. The signals were detected with ABC staining Kit according to the manufacturer’s instructions (Vector Laboratory Inc).

For histological analysis, the specimens were embedded in paraffin or Technovit 8100 (HERAEUS KULZER, Wehrheim) and cut at 8-10 μm.

In addition to the fish organizer (embryonic shield) from a shield-stage embryo, we also used chicken Hensen’s node (HN; stage 4, head organizer) and mammalian COS7 cells transfected with Xenopus noggin and chordin cDNA (Noggin/Chordin COS7; Lamb et al., 1993; Sasai et al., 1994) as a source of inducing tissue in the present study (Fig. 1A). In previous heterospecific organizer experiments between fish and salamander (Oppenheimer, 1936b), between mouse and Xenopus (Blum et al., 1992), between chick and Xenopus (Kintner and Dodd, 1991), and between chick and fish (Hatta and Takahashi, 1996), axial structures and/or neural-specific markers were ectopically induced, suggesting that signalling molecules from the organizer is conserved during vertebrate evolution.

As expected, we found that all inducing tissues examined were able to induce secondary axes when transplanted at midblastula to early gastrula stage into the ventral side of the fish embryo (Fig. 1B,C,F). However, they behaved differently in the secondary axes. Grafted embryonic shield contributed to the axial mesoderm and the ventral part of the neural tube (Fig. 1D,E), while grafted chicken HN and Noggin/Chordin COS7 showed no sign of self-differentiation presumably due to a different cellular environment, but were present in a cell mass within the neuroepithelium (Figs 1B, 2E) and under the neural tube (Fig. 1H) respectively. The secondary axes induced by HN or Noggin/Chordin COS7 tended to show cyclopean (one-eyed head), probably due to the lack of axial mesoderm: neither gsc (a marker for early organizer and prechordal plate) nor no tail (ntl; zebrafish orthologue of mouse Brachyury; a marker for the marginal mesoderm and notochord; Schulte-Merker et al., 1992) was detected in the secondary axis (Hatta and Takahashi, 1996; data not shown). Histological sections show that the neural tube, especially in the region anterior to the HN or Noggin/Chordin COS7 cell mass, did not exhibit dorsoventral (D-V) polarity (Fig. 1G). Thus, we had expected that, by use of chicken HN or Noggin/Chordin COS7, we could examine direct influences of the transplants, excluding interactions between axial mesoderm and the epiblast.

We examined gene expression patterns around the grafted tissues at the gastrula stage. We mainly focused on the anterior CNS and used the otx2 gene as a marker for midbrain and forebrain. The expression of zebrafish otx2 appears abruptly at high levels in a triangular patch at the animal pole of the early gastrula (about 6.5 hours postfertilization (h)), a dorsal region that contains cells fated to become midbrain and forebrain (Li et al., 1994; Mori et al., 1994; Fig. 6A). We also used the hoxa-1 as a marker for the posterior neural tissues (hindbrain and spinal cord) (Alexandre et al., 1996; Fig. 6A). The anterior border of hoxa-1 expression corresponds to the rhombomere 3/4 boundary. These genes are the earliest among known genes showing region-specific expression in CNS.

We first examined the expression pattern of otx2 in host embryos with HN grafts (stage 4; head organizer; Storey et al., 1992). When host embryos received HN grafts at the early gastrula stage (shield, 6 h), a neural plate-like structure (thickened epiblast) was formed around the grafted HN (Fig.2A,E) at the late gastrula stage (bud stage, 10 h), and ectopic expression of otx2 was detected around the grafts (30/37; 30 positive out of 37 successful operations in which the grafts are located in the ventroanimal region). No transcripts of gsc were detected in the secondary axis (0/10) induced by grafted HN, indicating that probably no axial mesoderm was formed. Thus, it was concluded that ectopic expression of otx2 was directly induced by the grafted HN.

Ectopic expression of otx2 was frequently induced around the anterior edge of the grafted HN. However, the induced expression domain always had a sharp posterior boundary which was at the same level (latitude) as the endogenous otx2 expression domain on the dorsal side (Fig. 2B-D), suggesting that only the anterior epiblast is competent to express otx2 in response to the grafted HN.

To examine the onset of competence to respond to organizer activity, we transplanted HN earlier before the formation of the endogenous shield. The transplantation was done at late blastula stage (sphere to dome stage, 4 h) and examined at mid gastrula stage (70-75%-epiboly, 8 h). The otx2 expression domain in the ventral region was enlarged as compared with that in later transplantation, and sometimes fused with the endogenous one at the animal pole, forming a circumferential expression domain around the pole (18/22; Fig. 2F). Again, the ectopic expression domain shared the same posterior boundary as the host domain. In most experiments, the inducing tissue was grafted halfway between the blastoderm margin and the animal pole (Fig. 1A). When HNs of the same stage were grafted near the blastoderm margin, the grafts were located in the host trunk region and no ectopic expression of otx2 was observed (21/21; Fig. 2G). These data suggested that the otx2-competent region is established by early gastrulation.

One possible explanation as to why HN is unable to specify A-P positional values within the ectoderm is that chicken cells are unable to contribute to any axial tissues, such as notochord and prechordal plate, that are known to have region-specific inducing properties (Hemmati-Brivanlou et al., 1990; Ang et al., 1994). We therefore examined whether endogenous organizer tissue (embryonic shield) could impart A-P positional information to the epiblast. In spite of the presence of the axial mesoderm derived from the transplanted shield (visualized by biotin-peroxidase staining), the posterior boundary of ectopic expression of otx2 was always located at the same latitude as that of the host dorsal expression (26/30; Fig. 2H-K). By contrast, the donor cells, which were distributed in the host trunk region by posterior transplantation (transplantation near the blastoderm margin), never induced otx2 expression (25/25; Fig. 2L) but induced posterior type CNS, which was shown by ectopic expression of krox-20 (a marker gene of rhombomere 3 and 5; Oxtoby and Jowett, 1993) in later stages (Fig. 2N). otx2 was induced only when the inducing tissues were present near the animal pole (compare Fig. 2M with Fig. 2N).

Our results suggest that the primary role of organizer tissues transplanted to the ventral side of the embryo is to convert non-neural to neural ectoderm. To investigate this issue further, instead of transplanting organizer tissue, which consists of different subpopulations of cells (Kimmel et al., 1990, zebrafish; Shih and Fraser, 1995, zebrafish; Storey et al., 1995, chick), we used COS7 cells secreting Noggin and Chordin, both potent neural-inducing molecules (Lamb et al., 1993; Sasai et al., 1994). In transplanted embryos, neither gsc nor ntl was ectopically induced in the hosts, indicating that no axial mesoderm was formed around the grafts (data not shown). The Noggin/Chordin COS7 exactly mimicked the organizer transplants. In response to the neural inducers, only the epiblast near the animal pole expressed otx2 while the epiblast near the blastoderm margin expressed hoxa-1. Again, each expression boundary on the ventral side was shared by that of the host domain on the dorsal side, indicating that both anterior and posterior information are present in the epiblast (Fig. 2O-R).

Taken together, the transplantation of organizer tissues demonstrated that a differential competence of the epiblast existed in entire embryo along the A-P axis.

The above results did not rule out the possibility that the endogenous early organizer specifies the A-P pattern in the epiblast and that our transplants are simply unable to override this initial specification. To test this possibility, we transplanted Noggin/Chordin COS7 into ventralized zebrafish embryos that lacked a shield and all axial derivatives (Mizuno et al., 1997; Fig. 3A), and determined whether the induced neural tissue had A-P polarity. We confirmed that, in ventralized embryos, ntl was normally expressed in the blastoderm margin but that neither gsc (a marker for early organizer) nor otx2 expression was detected (Fig. 3B-E). As shown in Fig. 3F,G, otx2 was induced by the graft only in the anterior region of the ventralized embryos and the expression domain had a clear posterior boundary (9/9), indicating that neural tissue induced in the absence of early organizer possessed an A-P pattern.

We then asked which tissues and factors are involved in the specification of the epiblast. We first focused on the yolk cell which is known to be a source of mesoderm-inducing signals in fish (Mizuno et al., 1996). When transplanted in the animal pole region at the blastula stage (Fig. 4A), the yolk cell suppressed otx2 (34/34) and induced hoxa-1 (31/32) in the surrounding cells (Fig. 4B,C). Histological examination revealed ectopic hoxa-1 expression in the epiblast close to the mesoderm, which was induced by the grafted yolk cell (data not shown). These results suggest that the yolk cell and/or mesoderm posteriorize the epiblast during normal development.

Given the potential role for fibroblast growth factors (FGFs) in mesoderm induction (Kimelman and Kirschner, 1987), we blocked the mesoderm formation by the injection of mRNA encoding dominant negative Xenopus FGF receptor (XFD) at the 1-cell stage. As reported (Amaya et al., 1991; Griffin et al., 1995), inhibition of FGF-receptor signalling by XFD suppressed the mesoderm formation, leading to complete loss of trunk and tail. Furthermore, we frequently observed that, in XFD-injected embryos, the posterior boundary of endogenous otx2 expression domain shifted posteriorly near the margin (15/21), while expression of hoxa-1 was greatly reduced (8/8) (Fig. 4D-G). In the injected embryos, the effect was sometimes limited probably due to an uneven distribution of injected RNA (Griffin et al., 1995). However, this kind of sample revealed a close correlation between a loss of the marginal mesoderm (indicated by a loss of Ntl staining) and a posterior shift in the otx2-expression domain (Fig. 4E), suggesting that the mesoderm in the margin is a primary tissue involved in the A-P specification of the epiblast. The transplantation experiments (Fig. 4H) confirmed that the otx2-competent region on the ventral side also expanded near the margin in the XFD-injected embryos: ectopic expression domain did not have a clear posterior boundary, extending along the transplanted inducing tissue (Fig. 4I,J).

To directly examine the role of marginal mesoderm, the ventral marginal cells from early gastrula (shield stage, 6 h) were transplanted in the animal-pole region (future otx2-expression region) of the blastula embryo (sphere, 4 h) and examined for gene expression at the gastrula stage (Fig. 4K). As shown in Fig. 4L,M, the marginal cells suppressed otx2 expression (7/7) and induced hoxa-1 (11/16) in the surrounding host cells, implying that the cells in the margin possess posteriorizing activity.

Since XFD was overexpressed in whole embryos, RNA injection experiments did not rule out the possibility that FGF-R signalling is directly involved in restricting the expression of otx2 in the anterior region. To test this possibility, we examined whether cells expressing XFD were still able to respond to A-P positional information by transplanting XFD-injected blastomeres into the normal host blastula (Fig. 5A). XFD-injected cells, when incorporated into the mesoderm region, did not express ntl (Fig. 5B,C). By contrast, XFD-injected cells strictly followed the normal otx2-expression pattern in the host (15/15; Fig. 5D-G): they expressed otx2 when located within the host expression domain while they never expressed outside the region, implying that FGF-R signalling is not directly involved in defining the otx2-boundary. The presence of XFD mRNAs in transplanted cells was confirmed by double in situ hybridization with XFD and otx2 probes (Fig. 5F,G).

In this paper, we present several lines of evidence that initial A-P pattern within neural tissues is determined by differential competence of the epiblast. Involvement of the marginal mesoderm in A-P specification of CNS has been shown by Woo and Fraser (1997). We further show that the signals from the marginal mesoderm impart a differential competence to the epiblast.

Although we use three different inducing tissues, embryonic shield, chicken Hensen’s node and mammalian COS7-secreting organizer factors, we obtained essentially the same results. They induced neural tissue but the A-P identity of the induced neural tissue is dependent upon the position of the inducing tissues within the epiblast. Thus, otx2, an anterior neural marker, was only ever induced in anterior regions of the embryo, irrespective of the position of the grafts, and hoxa-1, a posterior neural marker, was induced only in the posterior regions. Furthermore, the boundary of each ectopic expression domain on the ventral side was always at an equivalent latitude to that of the endogenous expression of the dorsal side of the embryo. These results demonstrated that the competent region for each neural marker exists in a circumferential ring in the epiblast along the A-P axis.

It is worth emphasizing that differential competence to neural inducers is not intrinsic to the epiblast but that its establishment depends on the signals from the marginal mesoderm. The competence can be easily changed by transplantation or suppression of the marginal mesoderm. Thus, the differential competence of the epiblast is detected only in vivo in the presence of the marginal mesoderm. Furthermore, we do not know exactly when differential competence is established in the epiblast. However, our early transplantation experiments (Fig. 4L,M) suggest that the marginal mesoderm influences the epiblast at least by the onset of gastrulation.

Our finding, differential competence in the epiblast along the A-P axis, seems somewhat contradictory to a widely accepted model for neural induction in other vertebrates. In the model, regional specification of the CNS depends on the nature of the inducers but not on differential competence of the ectoderm; for example, the early organizer or anterior axial mesoderm induces head structures, while the late organizer or posterior mesoderm induces the spinal cord (for example, Spemann, 1938). However, it is worth noting that recent experiments supporting this model have been carried out in vitro or without examining the influence of the host tissue (Hemmati-Brivanlou, 1990, Xenopus; Storey et al., 1992, chick; Ang and Rossant, 1993, mouse). Furthermore, we observed that, as development proceeded, the otx2 expression domain in the secondary axis was posteriorly displaced with respect to that of the host (see Figs 1B, 2M), probably due to a difference in cell growth; the induced axis is usually smaller than that of the host. Thus, an examination at later stages like classical experiments would give the impression that the expression boundary of a secondary axis bears little relationship to that of a host axis.

Our data do not exclude the notion that the nature of inducing tissues determines the type of CNS. Sagerström et al. (1996) have shown that explanted zebrafish shield becomes into an axis with precise A-P neural pattern, suggesting that the shield itself has also A-P information. Probably, together with differential competence of the epiblast, the signals from the axial mesoderm are required for regional specification of CNS, especially in later development (Shimamura and Rubenstein, 1997).

In previous experiments by Hatta and Takahashi (1996) and Shih and Fraser (1996), secondary axes were frequently induced in zebrafish embryos by the organizer transplantation. However, those secondary axes lacked anterior head structures, which could be interpreted as the results of posterior transplantation: they grafted organizer tissues into the ventral marginal zone of zebrafish embryos. Thus, unlike in amphibians, the organizer must be grafted anteriorly near the animal pole to obtain a complete secondary axis in zebrafish.

The experiments with ventralized zebrafish embryos have shown that the A-P pattern in the epiblast is not imposed by early organizer. Since ultraviolet (UV) irradiation does not work well in fish embryos (Strähle and Jesuthasan, 1993), ventralized zebrafish embryos were obtained by the removal of the vegetal yolk hemisphere soon after fertilization. As in goldfish (Mizuno et al., 1997), zebrafish embryos from which the vegetal yolk mass had been removed showed a typical ventralized phenotype: no dorsal structures such as embryonic shield, notochord and neural tube were formed, leading to a bilaterally symmetrical morphology (T. Mizuno et al., unpublished data). In situ hybridization analysis confirmed that neither gsc nor otx2 was expressed in the embryos while ntl was normally expressed in the margin, indicating that mesoderm induction proceeds normally without D-V specification (Fig. 3). This phenotype is similar to that of Xenopus ventralized embryos generated by UV irradiation (Cook and Smith, 1987). Surprisingly, otx2 was induced by Noggin/Chordin COS7 with a clear A-P pattern in the ventralized embryo, indicating that differential competence in the epiblast appears independent of D-V axis formation.

Our transplantation experiments show that the yolk cell that is known as a mesoderm inducer (Mizuno et al., 1996) up-regulated the posterior neural marker and suppressed the anterior marker in the surrounding cells. Since the mesodermal cells in the margin, when transplanted in the animal-pole region, mimicked the effect of the yolk cell on A-P neural patterning, it is reasonable to conclude that the marginal mesoderm induced by the yolk cell posteriorizes the epiblast during normal development. This is consistent with recent studies showing the posteriorizing activity of the non-axial mesoderm in zebrafish and chicken embryos (Sagerström et al., 1996; Bang et al., 1997; Woo and Fraser, 1997). Especially, Woo and Fraser (1997) have shown in zebrafish that the marginal cells posteriorized forebrain progenitors when grafted nearby, resulting in an ectopic hindbrain-like structure. In addition to transplantation experiments, we show here a close correlation between the loss of the marginal mesoderm and a posterior shift in the otx2-expression domain in XFD-injected embryos, confirming that the marginal mesoderm is a source of posteriorizing signals. Since A-P information also exists on the ventral side where no neural induction takes place, it is likely that the posteriorizing signals have no neural-inducing activity but they affect the competence of the epiblast.

There is accumulating evidence in Xenopus that FGFs or FGF-R signalling function in specifying posterior identity in the developing neural tube (Kengaku and Okamoto, 1993 and 1995; Lamb and Harland, 1995; Cox and Hemmati-Brivanlou, 1995; Doniach, 1995). The present study showed that, in XFD-injected zebrafish embryos, the posterior boundary of both endogenous and ectopic otx2-expression domains shifted near the margin. These results demonstrate that all the epiblast remains competent to express otx2 if the FGF-R signalling is blocked. However, the requirement of FGF-R signalling seems indirect because XFD-injected cells, transplanted in the posterior region of the host, still responded to A-P positional information. Woo and Fraser (1997) also demonstrated that FGF beads, implanted in the animal pole of zebrafish gastrula, caused severe perturbation of the host’s forebrain but did not induce the posterior neural marker, suggesting that FGFs may act in concert with other factors to be responsible for the posteriorizing activity. Consistent with these results in zebrafish, the conclusions of recent transgenic and in vitro tissue recombination experiments using Xenopus indicate that FGF-R signalling may not be solely responsible for the neural posteriorizing activity (Kroll and Amaya, 1996; Bang et al., 1997).

Retinoic acid (RA) is an attractive candidate (Conlon, 1995; Blumberg et al., 1997). However, RA may not be an endogenous posteriorizing factor in zebrafish early embryos, because it has been reported that, although the amount of expression was increased, the anterior boundary of hoxa-1 expression remained unchanged in RA-treated zebrafish gastrula (Alexandre et al., 1996). Therefore, the nature of the endogenous posteriorizing signal(s) remains unclear.

From our present data, we present a model for neural induction and A-P specification in zebrafish (Fig. 6B, C), which is similar to that of Woo and Scott (1997). There exists two distinct signals, posteriorizing signals involved in A-P specification of the epiblast, and organizer factors (Schulte-Merker et al., 1997) involved in neural induction. This model is different from the classical amphibian two-step model (Nieuwkoop, 1950; Toivonen and Saxén, 1968) in that posteriorizing signals (may be equivalent to transformers) are secreted by non-axial mesoderm and that the two signals are independently regulated by the yolk cell which is responsible for induction and dorsal specification of the mesoderm (Mizuno et al., 1997).

Before and/or during neural induction, the epiblast near the blastoderm margin, including on the ventral side, loses the competence to express anterior neural markers under the influence of posteriorizing signals produced by the marginal mesoderm. When the epiblast on the dorsal side is exposed to neural inducers produced by the organizer region, the posteriorized epiblast near the margin expresses posterior neural markers, while the anterior epiblast (probably the equivalent of the amphibian animal cap), which escapes from posteriorization, expresses anterior neural markers. Thus, A-P positional values are present in the entire embryo as differential competence of the epiblast, which imposes an initial A-P pattern on zebrafish CNS. Probably due to this A-P information present in the entire epiblast, the secondary axis induced on the ventral side always forms in the same direction as the primary axis, irrespective of the orientation of the grafted organizer.

We would like to thank Drs M. Mishina and H. Mori for providing zebrafish otx2 cDNA, Dr S. Shulte-Marker for ntl cDNA and anti-Ntl antibody, Dr T. Jowett for zebrafish krox-20 cDNA, Dr N. Holder for hoxa-1 cDNA, Dr E. Amaya for XFD cDNA and Dr Y. Sasai for Xenopus noggin and chordin cDNAs. This work was supported in part by grants-in-aids from the Ministry of Education, Science and Culture of Japan, by CREST (Core Research for Evolutional Science and Technology) of Japan Science and Technology Corporation (JST), by Bio-Design Project by the Ministry of Agriculture, Forestry and Fisheries, Japan, and by research funds of the Asahi Glass Foundation and Naito Foundation.