Hedgehog signaling is required for cranial neural crest morphogenesis and chondrogenesis at the midline in the zebrafish skull

Cranial neural crest (NC) cells in the vertebrate embryo form most of the cartilage and bones of the skull. The head skeleton has a mediolateral (ML)polarity. Some skeletal elements form singly, at the midline, whereas others develop bilaterally at different distances on either side(De Beer, 1937). Fate-mapping studies of NC in embryos of fish, amphibians, birds and mammals have all shown that streams of NC cells from the hindbrain form most of the pharyngeal skeleton or `viscerocranium', while more anterior NC cells form the braincase or `neurocranium' (Horstadius and Sellman,1946; Chibon,1967; Le Lievre,1978; Couly et al.,1993; Lumsden et al., 1993; Schilling and Kimmel, 1994; Osumi-Yamashita et al., 1994). These NC cells carry intrinsic differences acquired prior to migration that influence their skeletal fates along the anteroposterior (AP) axis(Noden, 1983; Trainor and Krumlauf, 2002), but also require inductive signals from surrounding endoderm and ectoderm (Hall, 1980; Couly et al., 2002; David et al., 2002; Crump et al., 2004a; Crump et al., 2004b; Le Douarin et al., 2004). Many studies have addressed the roles of these signals in AP patterning, but less is known about how skeletal polarity is established along the ML axis.

The anterior neurocranium (ANC) of larval zebrafish consists of paired trabecular rods joined at the midline anteriorly to form the ethmoid plate(Fig. 1A), and these form a striking example of M-L polarity. Genetic studies in zebrafish and mice have revealed that one member of the Hedgehog family of secreted proteins expressed at the midline, Sonic hedgehog (Shh), is required for ANC formation. Hh proteins act as polarizing factors in other contexts. Shh is expressed in the midline prechordal plate mesoderm and floorplate of the CNS adjacent to NC progenitors of the ANC(Strahle et al., 2004). Inactivation of zebrafish shh or other components of its signal transduction pathway causes ML patterning defects in which trabecular cartilages fuse at the midline, which is consistent with shh having a role in polarity (Brand et al.,1996; Schilling,1997; Kimmel et al.,2001). Similarly, loss-of-function mutations in Shh in the mouse cause severe defects in cranial growth and loss of virtually all of the craniofacial bones (Chiang et al.,1996). Humans with mutant SHH exhibit holoprosencephaly(HPE), often accompanied by midline craniofacial defects and palatal clefting(Gorlin et al., 1990; Roessler et al., 1996; Muenke and Beachy, 2000). Recent studies in mice have also shown that NC cells require the essential Hh receptor co-factor Smoothened (Smo) for craniofacial development, suggesting that Shh acts directly on skeletal progenitors(Jeong et al., 2004). Given its well-established role in regulating polarity in the neural tube and limb bud in a concentration-dependent manner(Ingham and McMahon, 2001),Shh is a good candidate for a signal in the NC environment that polarizes skeletal fates along the ML axis. In zebrafish, shh may share this role with its close relative tiggy-winkle hedgehog (twhh),which is expressed in a similar pattern.

Shh is expressed in several craniofacial tissues, including the early embryonic mesendoderm, ventral brain and oral ectoderm, and one or more of these sources could mediate its roles in ANC patterning. Surgical removal of oral ectoderm causes midfacial clefting in chicken embryos, whereas exogenous SHH protein causes midfacial expansion and, in some cases,duplications of midline structures of the ANC such as the nasal bone(Helms et al., 1997; Hu and Helms, 1999; Hu et al., 2003). These and other results have implicated Hh signaling from the facial ectoderm in ML patterning of the face, and in the regulation of NC proliferation and/or survival (Ahlgren et al.,2002). Endodermal expression of Shh is less important in this process, as zebrafish mutants that eliminate endoderm, such as casanova, lack viscerocranial cartilages but still form an ANC(David et al., 2002). Thus, Shh from the oral ectoderm may play a crucial role in ML polarity that is conserved in all vertebrates. However, the precise source of the Hh that controls facial patterning remains unclear.

Here, we present a characterization of NC morphogenesis and skeletal patterning in the sonic you (syu) mutant zebrafish, which disrupts the shh gene, and in embryos treated with the Hh inhibitor cyclopamine (CyA), at different stages of NC development. By constructing a transgenic line in which NC cells fluoresce in the living embryo, we are able for the first time to follow their migratory pathways in detail. We show that two distinct groups of NC form the trabecular and medial ethmoid cartilages of the ANC, contrary to the classical notion that the ethmoid simply forms by trabecular fusion. Both sets of NC cells require Hh signaling, but we argue that they do so at different times and in different regions, and that this helps to explain the spectrum of ML polarity defects observed in fish, mice and humans that are mutant in the Hh pathway. We suggest that Hh signals act both early, to separate skeletogenic NC at the midline, and later, to promote chondrogenesis, and that multiple sources of Shh induce the final skeletal pattern.

Zebrafish embryos were obtained in natural crosses and staged as previously reported (Kimmel et al.,1995). The AB strain was used as wild type. sox10:egfptransgenic fish [Tg(–4.9sox10:egfp)^ba2], which express egfp under the control of the sox10-promoter, were generated by microinjection of a DNA construct containing 4.9 kb of upstream sequence at the one-cell stage (T.J.C. and R.N.K., unpublished). Homozygous sonic you (syu^tq252) and slow muscle omitted (smu^b641) mutants were isolated from their siblings by somite morphology at 20 hours post-fertilization (hpf)(van Eeden et al., 1996;Schuarte et al., 1998; Varga et al.,2001).

Morpholinos targeting zebrafish shh (shh-MO) and tiggy-winkle hedgehog (twhh-MO) were obtained (GeneTools) as reported previously (Nasevicius and Ekker,2000). These were microinjected at the one- to two-cell stage,either alone or in combination. Defects depended on the amount of MO injected;for most experiments, we injected 2.0 ng shh-MO and 0.55 ng twhh-MO (Fig. 1).

For skeletal analysis, larvae were fixed at 4-5 dpf and stained for cartilage with Alcian Blue, after which they were dissected and flat-mounted as described (Javidan and Schilling,2004). In situ hybridization was performed as described previously(Thisse et al., 1993). Probes used in this study were foxd3(Odenthal and Nusslein-Volhard,1998), shh (Krauss et al., 1993), ptc1(Concordet et al., 1996), sox9a (Yan et al.,2002) and sox10(Dutton et al., 2001).

For live imaging of NC, heterozygous sox10:egfp^+/– embryos (homozygous embryos showed variable defects in NC survival), were dechorionated at 12 hpf and mounted in 0.75% agarose on bridged coverslips. To prevent drying, petroleum jelly(Vaseline) was used to form a chamber of embryo medium containing tricaine anaesthetic, and this was sealed with a second coverslip. Time-lapsed imaging was performed using a Zeiss LSM 510 laser confocal microscope at 10 minute intervals. At each time point, approximately 100-μm z-stacks at 5μm intervals were captured. Movies of projected z-stacks, as well as individual z-sections, were configured and individual cells followed manually. To analyze the later distribution of egfp+ cells,embryos were fixed with 4% paraformaldehyde (PFA) for 8 hours at 4°C,washed several times with PBS, and embedded in 1.5% agar after manual yolk removal. Images were captured at 2 μm intervals.

Premigratory and postmigratory NC in sox10:egfp fish were labeled with either the lipophilic fluorescent dye, PKH26 (Sigma), or by intracellular microinjection of TRITC-dextran (Molecular Probes, #1817) into single cells. The distribution of their descendants was determined in the ANC 2-3 days later. Anesthetized embryos were mounted in 3% methylcellulose and a small volume of PKH-solution (20% in dilution solution) was injected through a small cut in the surface ectoderm made with a tungsten needle. On average, this labeled five to ten cells. Injection sites were recorded and the embryos subsequently raised until 80-85 hpf, when they were fixed in PFA for 8-12 hours, dissected and flat-mounted for observation of PKH+ cells.

Intracellular, single-cell labeling with tetramethylrhodamine-dextran(TRITC-dextran; 10,000 MW) was performed by iontophoresis, as previously reported (Schilling and Kimmel,1994). Briefly, dechorionated embryos were mounted in 1.5% agar,agar was removed adjacent to the injection site and TRITC (3% in 0.2 M KCl)was injected iontophoretically; fluorescence was monitored continuously on a Zeiss Axioplan2 fluorescence microscope. Injection sites were recorded within 1 hour after injection, and the embryos were incubated for another 2-3 days to analyze cartilage.

A 10 mM stock solution of cyclopamine (CyA), dissolved in 95% methanol, was diluted in embryo medium to different concentrations between 5-100 μM. Embryos were soaked in CyA solutions at different stages between 0-48 hpf,washed several times in embryo medium, and either fixed at 24-36 hpf for in situ hybridization for ptc1 or raised for another 2-3 days for Alcian staining.

Shh was overexpressed by injecting 0.3 ng of zebrafish shh mRNA into embryos at the 1-2 cell stage. For misexpression experiments at later stages, beads coated in recombinant Shh protein were implanted. CM-AffiGel Blue beads (70-100 μm diameter, Bio-Rad) were incubated in Shh protein(mouse recombinant, 10 μg/ml, R&D Systems) at 4°C for 1 hour. Using a tungsten needle, a small slit was made posterior to the left eye. A protein-coated bead was inserted, and positioned between eye and brain (see Fig. 7C inset). Control beads were coated in 0.1% BSA-PBS. Treated embryos were incubated and fixed for cartilage analysis.

Wild-type donors were injected at the one- to two-cell stage with a mixture of 3% TRITC-dextran and 3% biotinylated-dextran, and cells were transplanted into syu^tq252 mutants (or shh-MO/twhh-MO-injected embryos). Transplants were either placed near the animal pole to target surface ectoderm, or further vegetally to target the ventral brain in these experiments. Donors for ventral brain transplants were injected with small amounts of Taram-A (Tar*; 0.5 ng/embryo),which biases them toward dorsal fates, including floorplate. At 24 hpf, the positions of fluorescent donor cells were recorded and then embryos were raised to 4 dpf for skeletal analysis. Biotin-labeled donor cells were detected with a Vectastain Elite ABC kit (Vectastain) and DAB as the substrate.

Loss of Shh function in the sonic you(syu^tq252) mutant causes variable midline defects and fusions of trabeculae in the ANC (Fig. 1A-C). To address redundant functions of the two closely related Hh-family members shh and twhh in the ANC, we used antisense morpholino oligonucleotides (MOs) to inhibit their translation. Both are expressed at the midline in zebrafish and are partially redundant in patterning of the neural tube (Ekker et al., 1995). Co-injection of the shh-MO and twhh-MO (hereafter referred to as `Hh-deficient') caused midline ANC fusions in a concentration-dependent manner(Fig. 1D-F). Injections of 3 ng/embryo or more led to severe shortening or loss of trabeculae(Fig. 1D), whereas injections of 2 ng/embryo or less caused midline fusions similar to those seen in syu mutants (Fig. 1E). A complete loss of Hh signaling, as seen in the smooth muscle omitted(smu^b641) mutant, eliminates the neurocranium altogether(data not shown) (Chen et al.,2001; Varga et al.,2001).

To determine the stage at which cranial cartilage formation is disrupted in Hh-deficient embryos, we first analyzed chondrogenesis with Alcian staining. In wild types, trabecular cartilage is first detected between 45-48 hpf(Schilling and Kimmel, 1997). Two parallel, single-cell rows of chondrocytes form between the eyes (Fig. 2A), and these fuse in the midline at their anterior ends to form the ethmoidal plate by 56 hpf (Fig. 2B,C). At the same time, the trabeculae elongate rapidly and their posterior ends fuse to the basal plate of the skull(Fig. 2B-D). To examine pre-chondrogenic cells of the ANC in wild type, we analyzed expression of the Sox transcription factor sox9a during NC migration and condensation beneath the brain (Fig. 2I-L). Expression of sox9a mRNA is first detected at 24 hpf (data not shown), and by 30 hpf marks bilateral clusters of cells between the diencephalon and the eyes (Fig. 2I). These elongate by 36 hpf to prefigure the trabeculae(Fig. 2J). Expression of sox9a in ethmoid progenitors is first detected at 42 hpf, and these bridge the anterior ends of the trabecular condensations by 45 hpf(Fig. 2K).

In contrast to wild type, the first chondrocytes detected between the eyes in syu mutants lie further posteriorly, in small clusters near the anterior end of the notochord. These fuse at the midline to form a single rod of cartilage that bifurcates at its posterior end(Fig. 2E-H). sox9aexpression in mutant ANC precursors is delayed compared with wild type(Fig. 2M), but by 36 hpf expression reveals fused condensations that form the ANC, ∼10 hours prior to differentiating as cartilage (Fig. 2N). Thus, the fusions of trabecular cartilages in syuhomozygous larvae result from an earlier NC defect.

Cartilage fusions at the midline in syu mutants could be due to abnormal NC migration or later ectopic fusions of NC streams after their migration beneath the brain. To distinguish between these possibilities, we followed cranial NC morphogenesis in wild-type and Hh-deficient embryos using a newly generated transgenic line in which 4.9 kb of the sox10promoter drives egfp expression in NC. A detailed description of the expression patterns of this transgene in other body regions will be presented elsewhere. Expression in the head begins at 11 hpf and, for the first time,has allowed us to follow whole populations of cranial NC in the living embryo,from their premigratory origins at 14 hpf(Fig. 3A,B), and along their migratory pathways up to 20 hpf (Fig. 3C). Expression also persists in cartilage at 48 hpf(Fig. 3I). Time-lapsed,confocal movies of sox10:egfp expression between 14-20 hpf revealed that NC at midbrain levels migrates anteriorly, between the eyes, into the position of the future ANC (Fig. 3B,C; see Movie 1 in the supplementary material). Tracking of individual cells in these movies revealed a general tendency for NC cells originating in more anterior positions to migrate with more anterior trajectories (Fig. 3D). NC cells that originate at anterior midbrain levels, tend to move along the dorsal edge of the optic vesicle to the anterior tip of the embryo(n=4). By contrast, NC cells that originate further posteriorly, near the midbrain-hindbrain boundary, follow a more ventral trajectory posterior to the eyes by 20 hpf (n=8). Through these movements, NC eventually forms a continuous band of cells that surround the posterior and medial portions of the eyes, the future locations of cartilages of the ANC.

We also examined later stages of NC morphogenesis at the ventral midline in sox10:egfp transgenic embryos. In uninjected controls at 24 hpf, sox10:egfp+ cells from the hindbrain aggregate in the pharyngeal arches, whereas others are widely distributed around the brain and eyes(Fig. 3E). Precursors of the trabeculae at this stage lie beneath the eye, anterior to the stomodeum(arrowhead in Fig. 3E) that delineates the mandibular arch (PA1). These cells are widely separated(∼100 μm) from one another across the midline at 24 hpf, on either side of the brain and pharynx (Fig. 3F,G). sox10:egfp+ cells in this region first aggregate around the mouth at 36 hpf (Fig. 3H), and condense to prefigure trabeculae by 42 hpf (data not shown), six hours prior to cartilage differentiation(Fig. 3I). Once cartilage has differentiated, sox10:egfp expression also includes other cartilages such as the parachordals and the pectoral girdle, which derive from embryonic mesoderm(data not shown).

Next, we investigated NC migration in Hh-deficient embryos using sox10:egfp. This revealed little difference from controls at early stages (n=3, data not shown). At 24 hpf, the head is slightly reduced but the pattern of egfp+ NC cells is similar to controls when viewed laterally (Fig. 3J). A ventral view, however, reveals a dramatic difference. NC cells lying anterior to the mandibular arch fuse across the ventral midline, between the eyes(Fig. 3K,L). Fusion is more pronounced at 36 hpf (Fig. 3M),and leads directly to a midline condensation that forms a single rod of cartilage (Fig. 3N). This midline structure is most likely a fusion of trabeculae, based on the relative positions of the adjacent palatoquadrate and the anterior end of the notochord in Hh-deficient embryos. These defects in Hh-deficient embryos do not appear to result from elevated NC cell death, as determined by acridine orange staining (data not shown). Taken together, these results suggest that Hh signaling is required for the morphogenetic movements that separate NC streams at the ventral midline.

To confirm cartilage identities in syu mutants, we labeled individual premigratory NC cells with fluorescent lineage tracers and tracked their fates in the ANC at 80-85 hpf. In the initial experiments, the lipophilic fluorescent dye, PKH26, was injected extracellularly at different positions in the NC of sox10:egfp transgenics at 13 hpf and followed until they formed cartilage (Fig. 4). This technique typically labeled 5-10 cells. When NC were labeled at the anterior midbrain, just dorsal to the eye(Fig. 4A,K; 200-300 μm from the anterior end), cells were later found in the medial ethmoid, confined to a triangular group of chondrocytes wedged between the anterior ends of the trabeculae (Fig. 4B; n=15). By contrast, cells labeled further posteriorly, by the midbrain-hindbrain boundary (Fig. 4E,K; 300-400 μm) generated progeny confined to the trabecular rods and lateral ethmoid, but not in the medial ethmoid(Fig. 4F; n=15). We confirmed these results by labeling single NC cells intracellularly with iontophoresis of fluorescent dextrans in sox10:egfp transgenics (see Fig. S1 in the supplementary material). Our fate map for the ANC roughly coincides with the anterior limits of sox10 and foxd3expression in NC (Fig. 4I,J). These domains are changing rapidly, as one hour earlier (12 hpf) we could not distinguish the positions of ethmoid and trabecular precursors.

To further investigate the distinct origins of these two skeletal regions,we also made a fate map at 22-24 hpf, near the end of NC migration. Postmigratory NC cells were labeled with PKH26 at three different locations:(1) the anterior end of the head; (2) ventral-anterior to the eye; and (3)ventral-posterior to the eye. When cells at the anterior end were labeled,their progeny formed the median ethmoid plate(Fig. 4C,D; n=8). By contrast, cells ventral-anterior or ventral-posterior to the eye formed lateral ethmoid and trabecular cartilage(Fig. 4G,H; Fig. S2 in the supplementary material; n=15). Similar results were obtained by iontophoretically labeling single NC cells (see Fig. S2 in the supplementary material). These results confirm that NC cells that form the median ethmoid are distinct and follow different migratory pathways, rather than forming as medial trabecular expansions (Fig. 4K).

To confirm the identity of the single midline cartilage in Hh-deficient animals, we performed similar fate mapping studies(Fig. 5A,C). Although posterior cells that normally form the trabeculae and lateral ethmoid contributed to the midline cartilage rod in Hh-deficient animals(Fig. 5D; n=5), more anterior NC cells that normally form the medial ethmoid did not form cartilage(Fig. 5B; n=5). Some labeled, undifferentiated cells were found in the tissue surrounding the cartilage, suggesting that medial ethmoid progenitors were eliminated in these embryos. Thus, the midline cartilage in Hh-deficient animals arises from the same NC population as trabecular precursors in wild type do.

To determine the precise temporal requirements for Hh signaling in the ANC,we blocked Hh signaling using the plant-derived steroidal alkaloid cyclopamine(CyA). CyA directly antagonizes the Hh signal-activation component Smoothened(Cooper et al., 1998; Chen et al., 2002). Embryos were bathed in CyA at different stages between 0-48 hpf and stained for cartilage. Siblings treated in parallel were stained by in situ hybridization for patched 1 (ptc1) at 24-36 hpf as a control for loss of Hh activity in NC (data not shown). Embryos treated with low concentrations (5μM) for a 4-hour period during gastrulation between 4-8 hpf had trabecular fusions similar to syu mutants(Fig. 6B,C; n=30;50%), and, at these concentrations, ptc1 expression was reduced but not eliminated in the NC (data not shown). Higher concentrations, 50-100μM, applied over a similar time period, effectively blocked all ptc1 expression and prevented all cartilage formation anterior to the basicapsular commissure (Fig. 6D). We were surprised at first to find that treatments with these high concentrations for any 4-hour period between 0-24 hpf caused a complete loss of the ANC (Fig. 6E),indicating an absolute requirement for Hh signaling in ANC development during NC migration. However, even early treatments suppress ptc1 expression up to at least 36 hpf, indicating that suppression of Hh signaling by CyA persists long after its removal.

Surprisingly, 50 μM CyA treatment of older embryos (>24 hpf), after NC migration, still disrupted trabecular and ethmoid development. Treatment between 24-30 hpf eliminated most of the ANC, other than a few trabecular chondrocytes, in the location where these cartilages first differentiate in wild types (Fig. 6F; n=30; 90%). Treatment between 30-35 hpf eliminated the medial ethmoid plate, whereas trabecular cartilages were well formed, thereby creating a large palatal cleft (Fig. 6G). Embryos treated between 36-48 hpf showed slight reductions in the ANC(Fig. 6H). No defects were observed with treatments after 48 hpf (Fig. 6I).

Next, we investigated the effects of increasing Hh signaling on ANC formation. We first injected mRNA for zebrafish shh to misexpress it throughout the embryo; this mRNA is only likely to persist through gastrula and neurula stages. The ANC of mRNA-injected wild-type embryos was roughly normal in shape but expanded laterally(Fig. 7B), and in some injected embryos, the medial and lateral ethmoid were separated(Fig. 7B), further suggesting that their precursors differentiate separately. To investigate later roles for Shh, after NC migration, we implanted Shh-coated beads and analyzed skeletal pattern in the ANC. Bead implantation at 20 hpf into the space between eye and brain in wild-type embryos promoted chondrogenesis in the ANC. Both ethmoid and trabeculae expanded laterally compared with BSA-soaked control beads. In some cases, posterior trabecular cartilages were fused to the dorsal(palatoquadrate) cartilage of the mandibular arch(Fig. 7C, compare with Fig. 1A).

Because syu mutants lack a functional Shh protein, we also tested whether Shh-coated beads could rescue syu or shh/twhh-MO injected embryos at different stages of ANC development. Implantation of a BSA-soaked control bead had no effect at any stage(Fig. 7D). By contrast, beads soaked in 10 μg/ml Shh implanted as late as 22 hpf partially restored both head size and ANC patterning in syu(Fig. 7E; n=15). Surprisingly, despite bead placement between the eye and midbrain on one side of the head, rescue often occured bilaterally to nearly a wild-type configuration. Others showed cartilage distortions or ectopic chondrocytes along the medial trabeculae (data not shown; n=10). To determine whether the postmigratory NC distribution is restored by these late Shh applications, we analyzed ANC precursors in MO-injected sox10:egfpembryos after bead implantation. With BSA-beads, egfp+ cells were observed in the midline at 36 hpf (Fig. 7G), showing that the beads themselves did not physically disrupt NC distribution. By contrast, when a Shh-coated bead was implanted at 22 hpf,separation was restored between groups of sox10:egfp+ cells on either side of the midline in syu mutants, although a few cells remained at the midline in some cases (Fig. 7H). Overall head size was not rescued in these experiments(compare Fig. 7G and H with I),suggesting that the role of Hh signaling in midline separation of NC is not simply in growth. These results show that exogenous Hh can rescue ANC pattern in Hh-deficient zebrafish embryos by restoring the cranial NC distribution. They also suggest that the pre-chondrogenic pattern in the ANC is not specified at 22 hpf. By 30 hpf, Shh-coated beads can only partially rescue the syu phenotype (Fig. 7F), and a few hours later they have no effect.

Shh is a secreted protein expressed in several different cranial tissues in the embryo, including the ventral brain, facial ectoderm and pharyngeal endoderm. To address which is required for ANC development, we used mosaic analysis to test the capacity of wild-type cells to rescue ANC development when transplanted into Hh-deficient hosts. Surface ectodermal cells that form epidermis arise from ectoderm near the animal pole of the early gastrula at 6 hpf (Kimmel et al., 1990). Cells were grafted from this location in a wild-type donor embryo co-injected with fluorescent lineage tracers, into the same region in syu mutants and raised to 96 hpf for cartilage staining. These ectodermal grafts expressed shh mRNA (data not shown)and partially rescued trabecular and ethmoid chondrogenesis(Fig. 8A-C; n=3). Rescue only occurred when transplanted cells were located in dorsal oral ectoderm, just above the mouth, that comes to lie beneath the ethmoid(Fig. 8B). To target cells to the ventral forebrain, similar transplants were performed with ectoderm from the dorsal side of the early gastrula near the margin at 6 hpf, where these cells share a common lineage with floorplate cells at posterior hindbrain and spinal levels (Hatta et al., 1991). Dye-labeled cells were grafted from this location in wild type into the same region of syu or shh/twhh-MO injected hosts, and stained for cartilage(Fig. 8D-I). In some cases,these transplants also rescued bilateral trabecular separation(Fig. 8E; n=5).

Shh may act directly on NC or through intermediate signals, but recent tissue-specific removal of Smo function from NC in mice suggests that it acts directly on skeletal progenitors. To distinguish between these hypotheses, we transplanted labeled NC cells from smu mutants, which lack a functional Smo, into the wild-type midbrain NC and followed their fates at 96 hpf (Schilling and Kimmel,1994) (see also Fig. S3 in the supplementary material). These transplanted cells migrated with the host NC but remained undifferentiated and never formed cartilage (0%, n=6). Controls in which wild-type cranial NC cells were transplanted similarly formed cartilage in the ANC at a high frequency (data not shown; 60%, n=25). These results demonstrate that smu is required cell autonomously for skeletogenesis in cranial NC,and suggests that Hh acts directly on NC to induce the ANC.

The anteriormost NC cells in all vertebrate embryos migrate between the eyes to form the anterior neurocranium (ANC) and palate, as well as part of the lower jaw (Stone, 1929; Kontges and Lumsden, 1996). Here, we show that, in zebrafish, the two main cartilages of the larval ANC,the ethmoid plate and trabeculae, arise from distinct precursors(Fig. 9A), and both depend on Hh signaling for their development. In humans, loss-of-function mutations in the SHH gene cause holoprosencephaly (HPE) accompanied by facial anomalies including cyclopia, a single nostril, and cleft lip or palate(Gorlin et al., 1990; Roessler et al., 1996; Nanni et al., 1999). Similarly, disruption of Hh signaling in syu mutant zebrafish or in embryos treated with CyA during gastrulation causes cyclopia, and this is associated with trabecular fusions and loss of the median ethmoid. Genetic screens in the zebrafish have identified many loci required for ML patterning of the larval ANC, some of which encode Hh signaling components(Brand et al., 1996; Schuarte et al., 1998; Karlstrom et al.,1999). We show that fusions result from aberrant NC cell mixing at the ventral midline, and a lack of chondrogenesis in the ANC. Our cell transplantation studies suggest that there are two important Shh sources in this process, the ventral neural tube and the oral ectoderm, consistent with previous studies in chick (Hu et al.,2003). We propose that Hh signaling is required at two separate stages in the ANC: one early that acts to separate cells at the midline, and a second, more prolonged period in which it promotes chondrogenesis(Fig. 9B). We argue that this helps to explain the spectrum of phenotypes that have been described for Hh deficiencies in fish, chick and mouse, as well as the facial defects associated with human HPE.

Cranial NC cells from the midbrain that form the larval ANC do not all follow the same migratory path. The path choice depends on their premigratory locations at anterior or posterior midbrain(Fig. 3). We traced living NC cells in sox10:egfp transgenic embryos and found that cells from the anterior midbrain migrated along the dorsal optic vesicle and optic stalk. By contrast, NC cells from the posterior midbrain migrated behind the optic stalk, many accumulating posterior to the optic vesicle, in a `premandibular'location. These results confirm previous fate maps(Lumsden et al., 1991;Ohsumi-Yamashita et al., 1994; Epperlein et al., 2000) but, until now, the migratory pathways of these cells have only been inferred.

We also show that the AP position of a premigratory NC cell beside the midbrain predicts its fate in the skull(Fig. 4K). By following clones of single cells or small groups labeled with fluorescent lineage tracer dyes,we confirmed that cells migrating along the more anterior pathway above the eye exclusively form median ethmoid cartilage, whereas cells migrating along the more posterior path behind the eye form the trabeculae. Our results corroborate previous fish, amphibian and avian fate maps, showing that the trabeculae cranii derive from mesencephalic NC(Landacre, 1921; Stone, 1929; Langille and Hall, 1988) but reveal an additional, more anterior, ethmoid progenitor population. Thus,rather than simply arising as a medial growth and fusion of the trabeculae(trabeculae communis; Fig. 9A,top panels), as previously assumed (De Beer, 1937; Bertmar,1959), the medial ethmoid and perhaps other midline skeletal elements have a distinct cellular origin(Fig. 9A, bottom panels),possibly under distinct genetic control. In support of this hypothesis, in schmalspur (sur) mutant zebrafish, trabecular defects occur without corresponding changes in the ethmoid plate(Brand et al., 1996), whereas other mutants show defects restricted to the ethmoid(Neuhauss et al., 1996; Piotrowski et al., 1996). More long-term fate mapping studies are required to determine whether nasal and palatal progenitors, which differentiate between 1 and 3 weeks postfertilization in zebrafish (Cubbage and Mabee, 1996), also arise from this anterior NC. These results also have important implications for the interpretation of gene expression patterns in the premigratory NC, which in some cases do not include the most anterior, ethmoid domain.

We also labeled NC cells later, during migration, and followed the contributions of cells along each pathway to cartilage in the ANC. Consistent with our fate maps at earlier stages, we found that only cells at the anterior end near the midline contributed to the medial ethmoid plate. The trabeculae are formed by NC cells that lie more posteriorly, many developing in close association with the mandibular arch. Several recent papers have highlighted similar cells in amphibian and avian embryos, and have argued that they are a part of the mandibular arch, blurring the distinction between maxillary and mandibular (Cerny et al.,2004). Our results indicate that a majority of these cells lie anterior to the stomodeum in zebrafish, just outside of the mandibular arch. In other respects, our fate map is similar to those for the facial prominences in other species (McGonnell et al.,1998).

The variety of skeletal defects caused by Hh-deficiencies remains unexplained. For example, within the spectrum of human patients with Hh-associated HPE, some show midline collapse of facial structures, while others show clefting (Roessler and Muenke,2001; Roessler and Muenke,2003). Our studies suggest that such phenotypic differences reflect stage-dependent requirements for Hh signaling; and separate roles in midline separation and chondrogenesis. We have shown that early NC migration is unaffected in Hh-deficient embryos and that Hh acts later to prevent NC cells from crossing the midline. Shh may induce midline tissue expansion that acts as a physical barrier to movement(Fig. 9B, Step 1). Shh has anti-apoptotic activity (Charrier et al.,2001), and also promotes cell survival and growth in the ventral brain (Britto et al., 2002). Our analysis of sox10:egfp suggests that this is not simply an affect on growth, although we have not formally ruled out this possibility. We did not observe elevated NC apoptosis in Hh-deficient embryos (data not shown),which is consistent with previous reports(Cordero et al., 2004). Alternatively, Shh may directly alter NC cell movements in a concentration-dependent manner, as it has been shown to modify adhesion and NC migration in vitro (Testaz et al.,2001), or may act indirectly to induce a second signal from midline tissue(s), that acts on NC. At least some of these requirements for Hh signaling are direct, as NC cells transplanted from smo mutants into wild-type hosts never form cartilage (see Fig. S3 in the supplementary material), similar to recent findings in mice with a tissue-specific loss of Smo function in NC (Jeong et al.,2004). Future studies are needed to dissect how these early patterning influences of Hh on NC morphogenesis are coupled with effects on growth.

Two lines of evidence point to the fact that there are separate early and late roles for Hh-signaling in skull development(Fig. 9B, Step 2 and Step 3). First, CyA treatment between 24-36 hpf disrupts ANC chondrogenesis, but no longer affects NC morphogenesis at the midline. Similarly, a lack of Hh-signaling in Shh- or Smo-mutant mice causes severe cranial bone loss, and in some cases palatal clefting, presumably when ethmoidal chondrogenesis is disrupted. Late CyA treatment causes similar clefts in chick, suggesting that this late requirement is conserved(Cordero et al., 2004). Our results are also consistent with numerous studies in mammals implicating Hh proteins in skeletal differentiation (reviewed by Karsenty and Wagner, 2002; Kronenberg, 2003; Zelzer and Olsen, 2003). In our model, if there is sufficient signaling to separate NC at the midline, but not for ethmoid induction, clefting may result.

Not all forms of HPE in humans are accompanied by midline facial defects,suggesting that skeletal defects are not simply secondary consequences of forebrain defects. Our mosaic analyses help to explain this, and suggest that ectodermal Hh sources are also crucial for ANC formation. Gastrulation is the only period when CyA treatment in zebrafish causes midfacial fusions, and at these stages our data suggest that the important Hh source is the neural tube. Wild-type ventral neural tube cells locally rescue ANC development when grafted into the forebrain in Hh-deficient embryos, and separate the trabeculae at the midline. However, oral ectodermal grafts can also rescue chondrogenesis in the ANC (Fig. 8). These results are consistent with studies in chick suggesting that Shh from the facial ectoderm promotes maxillary and frontonasal outgrowth(Hu and Helms, 1999; Hu et al., 2003; MacDonald et al., 2004). Ectodermal sources may act together with Shh from the ventral diencephalon and presumptive hypothalamus at these later stages to regulate chondrogenesis.

Taken together, our results suggest that the mechanisms by which Shh controls the coordinated growth and fusion of facial primordia are highly conserved among vertebrates. The ventral neural tube and oral ectoderm at the midline appear to form an organizing center controlling skeletal growth and morphogenesis. These tissues both express shh, and we argue that skeletogenic NC cells must interpret their positions relative to these midline sources of shh and differentiate accordingly. Interestingly,mutations in mice and humans that disrupt squamous epithelia also often cause cleft lip and palate (EEC syndrome), presumably reflecting defects in these epithelial signaling centers (Celli et al.,1999). An interesting direction for the future would be to explore the roles of other growth factors in the cranial ectoderm, and their interactions with Shh in skeletogenesis. As a starting point, mutagenesis screens in zebrafish have uncovered large phenotypic classes of mutants that disrupt the midline, many of which disrupt signaling by the TGFβ family member Nodal (Kimmel et al.,2001). Zebrafish Nodal-related proteins were recently shown to directly regulate shh expression in the zebrafish midline(Muller et al., 2000) and recent evidence in humans has implicated defects in Nodal signaling in HPE(Gripp et al., 2000). With the large mutant collection in zebrafish, we can now begin to study of how such pathways interact during palatal development.

We thank J. Eberhard and C. Kimmel for sharing information prior to publication. We also thank I. Blitz and Schilling Laboratory members for discussions and comments on the manuscript. This work was supported by grants from the March of Dimes (1-FY01-198), NIH (NS-41353, DE-13828) and Pew Scholars Foundation (2615SC) to T.F.S., Wellcome Trust to R.N.K., and an ORS award to T.J.C.