Early Hedgehog signaling from neural to oral epithelium organizes anterior craniofacial development

Hedgehog (Hh) signaling is crucial for the development of the anterior face and skull. Loss of function of a key Hh molecule, Sonic hedgehog (Shh), causes a wide variety of craniofacial defects, and it is clear that Shh is involved in multiple steps of craniofacial development(Cordero et al., 2004). However, the existence of these multiple signaling events means that the precise roles of Hh signaling in individual steps remain largely unresolved.

Evidence that Hh plays a role in early craniofacial patterning comes from two different kinds of analyses. First, whereas complete loss of either Shh or the Hh co-receptor, Smoothened (Smo), causes nearly complete cartilage loss(Brand et al., 1996; Chen et al., 2001; Chiang et al., 1996; Hu and Helms, 1999; Roessler et al., 1996), less severe disruptions of the Hh pathway produce defects more specific to the anterior neurocranium (Brand et al.,1996). Second, targeted inactivation of Smo in the mouse,as well as functional studies in the chicken have shown that Hh signaling is required for the patterning and outgrowth of the precursors of the craniofacial skeleton (Abzhanov and Tabin,2004; Hu et al.,2003; Jeong et al.,2004; Washington et al.,2005). These analyses suggest a model in which the precursor domain for the anterior craniofacial skeletal is especially sensitive to Hh disruption. Therefore, we have detailed the role that Hh plays in patterning the specific precursor domain of the anterior neurocranium.

Here, we show that zebrafish Hh pathway mutants lose a specific domain of neural crest-derived skeletal precursors, which normally condense on the superior surface, or roof, of the stomodeum. By fate mapping and time-lapse analyses, we find that the neural crest cell precursors for the anterior neurocranium and pterygoid process, in the upper jaw, migrate to condense onto the stomodeal roof. Via genetic mosaic analyses, we next show that the ventral presumptive brain is the crucial source of Hh and that stomodeal ectoderm, not the neural crest cells, is the crucial recipient of the Hh signal. Moreover,cyclopamine treatment shows that Hh acts at the end of gastrulation,substantially before crest cells migrate into the pharyngeal arches. Therefore, we conclude that Hh has a specific early role in establishing the local epithelial environment that promotes the development of the anterior craniofacial skeletal elements.

Embryo care has been described previously(Westerfield, 1993). fli1:GFP transgenic embryos express GFP in cartilage-producing cells as well as vasculature (Lawson and Weinstein, 2002). Heterozygous carriers of the hypomorphic shh allele syu^tq252 and the smoothenednull allele smu^b577 were maintained in both AB and fli1:GFP backgrounds and will be referred to as shh^- and smo^- throughout the text for clarity. gsc:GFP transgenic zebrafish were generated by inserting a PCR generated 3 kb 5′ upstream promoter region of the gsc gene,primers GC64 (5′ AGGACTACCAAGCTTACTTTCTAACAATGCAAAGC 3′) and GC66(5′ TAGGACGAGGGATCCCATCACAAGCGAAAAGATGTG 3′), into pG1 (gift of Chi-Bin Chien) and injecting early one-cell embryos with 160 ng/μl linearized plasmid. Cyclopamine treatments(Hirsinger et al., 2004) were performed on fli1:GFP transgenic embryos. Shh and Twhh morpholinos are described elsewhere (Nasevicius and Ekker, 2000).

colIIa (Yan et al.,1995), fgf8 (Reifers et al., 1998), pitx2(Essner et al., 2000), shh (Krauss et al.,1993) and sox9a(Cresko et al., 2003) probes were used for in situ hybridization(Westerfield, 1993). For cartilage and bone staining, 4 day postfertilization (dpf) zebrafish embryos were stained with Alcian Blue and flat mounted(Kimmel et al., 1998). Time-lapse recordings, 10-15 minutes/frame, and confocal analysis of fli1:GFP and smo^-;fli1:GFP embryos were performed according to established protocols (Crump et al., 2004a).

Fate-map analyses with Kaede (Ando et al., 2002) and microelectroporation(Crump et al., 2004b) were performed in albino;fli1:GFP or albino;fli1:GFP; h2a:GFP transgenic embryos to allow for specific visualization of neural crest cells and chondrocytes. Kaede mRNA was injected at the one-cell stage. Green to red photoconversion of Kaede in a three- to five-cell diameter column was achieved using a DAPI filter set and a 30 μm pinhole. Embryos were imaged using a Zeiss Pascal confocal microscope and LSM software immediately post-labeling and again at 4.5 to 5 dpf. The original and skeletal locations of labeled cells were compiled into a fate map using the eye, the ventral limit of the crest population and the curvature of the yolk as landmarks at 12 hpf and the eye, stomodeum and first pharyngeal pouch as landmarks at 24 hpf. Statistical analysis was performed using JMP software(SAS Institute, Cary, NC).

Genetic mosaics were generated as described elsewhere(Crump et al., 2004a; Maves et al., 2002). Transplants placed just to the future ventral side of the animal pole contribute to stomodeal ectoderm (Fig. 7C). Transplanting gsc:GFP-positive cells into the gsc:GFP-positive region of host embryos at 30% epiboly targeted axial mesoderm. An anterior craniofacial `skeletal index' was used to quantify the level of rescue in shh + twhh MO-injected embryos: 0, no anterior neurocranium or pterygoid process; 1, anterior neurocranium: a very short midline bar, no pterygoid process; 2, anterior neurocranium: a longer midline bar extending between the eyes, pterygoid process present; 3, anterior neurocranium with bilateral trabeculae and slightly truncated ethmoid plate,pterygoid process present; 4, wild type.

Brand et al. described a phenotypic series of zebrafish `midline' mutants in which cartilages of the anterior region of the skull base, or neurocranium,are increasingly reduced or lost (Brand et al., 1996; Kimmel et al.,2001). Subsequent identification of the mutated genes, e.g. gli1 and gli2, showed that all of them are in the Hh pathway(Karlstrom et al., 1999; Wolff et al., 2004). We observe that hypomorphic shh mutant larvae display the entire phenotypic series reported by Brand et al. In the most severe hypomorphs, the anterior region of the neurocranium, including the midline ethmoid plate and the bilateral pair of trabeculae, is entirely absent. The more posterior neurocranium is only relatively mildly reduced(Fig. 1A,B). The boundary between the anterior deleted and posterior nondeleted regions is where the trabeculae meet the basal plate (`polar cartilage' region)(Cubbage and Mabee, 1996; Schilling and Kimmel, 1997). Although not reported previously, we also discovered that a specific anterior region of the palatoquadrate cartilage, the pterygoid process, was also missing in severely affected shh hypomorphs(Fig. 1B). The pterygoid process articulates with the ethmoid, thus connecting the jaw skeleton to the neurocranium and acting as the upper jaw cartilage(DeBeer, 1937). Additionally,the posterior portions of the palatoquadrates fuse into a single element bridging the midline. Whereas partial loss of Hh signaling predominantly affects anterior craniofacial cartilages, more complete reduction of the pathway, in severe smo loss-of-function mutants(Chen et al., 2001), deletes nearly all cartilage (Fig. 1C).

From these findings, we hypothesized that Hh signaling plays at least two roles. One function, revealed in the hypomorphic shh mutant and examined in detail here, is to pattern the formation of the anterior craniofacial skeleton specifically. A second function, revealed in the smo mutant, would direct differentiation of nearly all cartilages later (Long et al., 2003; Long et al., 2001). This hypothesis predicts that smo mutants would display early specific defects, even though specific cartilage loss is not observed later due to the generalized cartilage deficiency. To look for early patterning defects in smo mutants, we examined the expression of two of the earliest chondrogenic markers, sox9a (Fig. 2) and colIIa (data not shown). Both markers prefigure the wild-type neurocranium at 2 dpf. However, anterior neurocranial expression is completely missing in smo mutants, while posterior neurocranial expression remains. Hence, as predicted, we see an early smo^- patterning defect that correlates with the shh^- skeletal defect, even though these defects are obscured in older smo mutants owing to the generalized cartilage loss.

To understand the early role of Hh signaling in development of the anterior-most head skeleton we need a detailed knowledge of how this region develops. Pharyngeal cartilages originate from neural crest cells in zebrafish as in other vertebrates (Köntges and Lumsden, 1996; Langille and Hall, 1988; Sadaghiani and Thiebaud, 1987; Schilling and Kimmel, 1994; Tan and Morriss-Kay, 1986). We now show that the zebrafish anterior neurocranium is also neural crest derived (see also Wada et al., 2005). We used photoconversion of Kaede protein to fate map small clusters of cells in the premigratory anterior cranial neural crest, resulting in labeling in descendant chondrocytes of the trabeculae and ethmoid cartilages(Fig. 3). We also confirmed(Schilling and Kimmel, 1994)that the palatoquadrate and Meckel's cartilage derive from this population. However, labeling never included the polar cartilages or basal plate,consistent with the proposed mesodermal origin of these more posterior cartilages (Couly et al., 1993; Huang et al., 1997).

The fates of specific skeletal elements might be nonrandomly distributed within the premigratory crest, in spite of overlaps being present. The more anterior crest cells preferentially contribute to the anterior neurocranium(Fig. 3A,B), and the more posterior cells contribute to the first arch pharyngeal skeleton (Meckel's cartilage and posterior palatoquadrate; Fig. 3E,F). Notably, the pterygoid process of the palatoquadrate maps largely in between these two regions (Fig. 3C,D), in fact more frequently overlapping with neurocranial fates than with the posterior palatoquadrate (Fig. 3G,H).

We examined fates of the more anterior region at higher resolution by marking single crest cells (occasionally neighboring cell pairs) by microelectroporation before (Fig. 4A) or after (Fig. 4E) migration to the first pharyngeal arch. This method shows the precursors for the pterygoid process and anterior neurocranial cartilages, and the dermal parasphenoid bone to be largely localized to a small region along the side of the neural keel and just posterior to the rudiment of the eye(Fig. 4, data not shown). The premigratory domain is largely surrounded by crest cells that contribute, in part, to non-skeletal tissues such as the sclera of the eye.

This study also reveals that after migration, the neural crest-derived cells generating the skeletal fates described above all populate the first pharyngeal arch. Here, those cells forming specifically the pterygoid process and anterior neurocranium reside within a compact condensation closely associated with invaginated oral, or stomodeal ectoderm. Furthermore, the cells in question occupy just the part of the condensation that is present on the roof of the stomodeal epithelium (Fig. 4E-H, the extent of the domain is demarcated by the broken line in Fig. 4F). In addition, we labeled maxillary osteocytes from precursors within this domain in two cases(data not shown). Fate mapping of postmigratory posterior palatoquadrate and Meckel's cartilage precursors within the first arch is described elsewhere(J.G.C., M.E.S., J.K.E. and C.B.K., unpublished). In summary, we show that the cartilages that have an early and sensitive requirement for Hh signaling also have a distinctive origin in the neural crest, both before and after crest cells migrate to the first pharyngeal arch.

What is the effect of reduced Hh signaling on early development of the crest-derived anterior craniofacial precursors? We observed no changes in crest cell migration between wild types and smo mutants, either by examining expression of a crest cell marker, sox9b (12 and 18 hpf,data not shown) or by Kaede labeling of premigratory neural crest cells (see Fig. S1 in the supplementary material). However, we found that postmigratory smo^- cells in the first arch fail to undergo condensation on the roof of the stomodeum. To show this defect directly, we made confocal time-lapse recordings and followed cells expressing the fli1:GFP transgene, which labels crest cells beginning shortly after their migration and throughout skeletogenesis. At 20 hpf, postmigratory neural crest cells distribute in a broad anteroposterior swath in the first arch of wild-type and smo^- embryos alike(Fig. 5A,E). Similarly, in both wild type and smo^-, the inpocketed stomodeal ectoderm upon which the crest cells normally condense becomes evident at 21 hpf as an unlabeled (dark) indentation of the GFP-expressing cell mesenchyme(Fig. 5B,F, arrowhead). Beginning at 22 hpf, the wild-type crest cells form a prominent condensation,coating both the stomodeal roof and floor, i.e. the superior and inferior surfaces (Fig. 5C,D, see Movie 1 in the supplementary material). By contrast, in smo mutant embryos,neural crest cells condense only poorly or not at all on the stomodeal roof,whereas cells condense seemingly normally on the floor of the stomodeum(Fig. 5G,H, see Movie 2 in the supplementary material). Thus, the selective requirement for smo⁺ in the condensation of crest cells on the stomodeal roof, but not floor, corresponds exactly to the selective loss of the anterior neurocranium and pterygoid process cartilages that we fate mapped to this superior region of the condensation. By 30 hpf, we could detect few, if any,crest cells condensed on the roof of the stomodeum in smo^-embryos (see below, Fig. 7B). Time lapse analyses of smo^- embryos from 22-48 hpf show that the mesenchymal cells that would normally condense tightly on the roof of the stomodeum instead remained loosely associated with one another and eventually migrated immediately posterior to the eye (see Movie 3 in the supplementary material). In addition, we detect no cell death in these neural crest cells even as late as 60 hpf (Acridine Orange labeling, data not shown),suggesting that Hh signaling is necessary for the condensation but not survival of this subpopulation of first arch neural crest cells.

The association of the crest-derived mesenchyme with the superior surface of the stomodeal ectoderm, intimate in the wild type and missing in the smo mutant, prompted us to look, using marker expression, for defects in the smo^- stomodeum that could indicate a missing crucial epithelial-mesenchymal interaction(Crump et al., 2004b; Hall, 1992). pitx2(expressed broadly in wild type, Fig. 6A), fgf8 (lateral stomodeum, Fig. 6C) and shh(medial stomodeum, Fig. 6E) are all strikingly downregulated in the smo^- stomodeum(Fig. 6B,D,F). Some shh-expressing cells, probably endodermal but possibly stomodeal, are still present in the mutant near the midline(Fig. 6F, see Fig. S2 in the supplementary material). The stomodeum is not simply missing in smomutants; we can detect it by morphology with DIC optics and it expresses epithelial markers (see Fig. S3 in the supplementary material, data not shown). Moreover, we detected no cell death by Acridine Orange staining, loss of proliferation by anti-phosphohistone H3 or alterations of the early gastrula fate map of the stomodeum in smo^- embryos (data not shown).

Smo-dependent gene expression points to the stomodeal ectoderm as a potential direct target of Hh signaling, in which case the disruption of condensation of the crest-derived mesenchyme in smo mutants could be indirect. However, the opposite interpretation is also possible: work in mouse has shown that reception of Hh signaling is required by neural crest cells for craniofacial development (Jeong et al.,2004), and neural crest cells have been shown to regulate the timing of gene expression in avian facial ectoderm(Schneider and Helms, 2003). To determine which tissue needs to receive Hh signaling to promote crest cell condensation, we used mosaic analysis with wild-type and smo^- embryos. In wild-type hosts that received transplants of smo^- crest cell precursors, smo^-crest cells condensed apparently normally on the stomodeal roof(Fig. 7A,C). This rescue suggests that Smo functions nonautonomously in crest condensation. However, we rarely observed smo^- cells in cartilages at 4 dpf in these mosaics (data not shown), consistent with a later autonomous Smo-dependency in crest for cartilage differentiation (as also reported recently by Wada et al., 2005). In the reciprocal experiment, wild-type crest when transplanted into smo^- hosts failed to condense on the stomodeal roof, just as in smo mutants (Fig. 7B,D), and again suggesting that the Smo requirement for condensation is crest nonautonomous. Furthermore, time-lapse analyses of mosaics directly show that wild-type crest cells in the smo^- environment enter the region superior to the stomodeum (Fig. 7E) but they fail to condense (Fig. 7F-H,see Movie 4 in the supplementary material), just as described for the smo mutants themselves (Fig. 5).

These experiments show that Smo function is not directly required in neural crest cells for condensation. Hence, we used mosaic analysis to examine the other possibility, that the stomodeum itself is the target of Hh signaling. By transplanting the rudiment of the stomodeum, we generated mosaic embryos in which wild-type donor cells contributed to the stomodeum in smo^- hosts. In these mosaics, the smo^-crest cells condensed on the transplanted wild-type stomodeum normally, and also correctly expressed the chondrogenic marker sox9a(Fig. 8C,D). Rescue of both the condensation and sox9a expression was unilateral in the mosaics,corresponding to the side that received the wild-type stomodeal transplant(Fig. 8D,E). Sometimes the transplants included non-stomodeal facial ectoderm. However, stomodeal ectoderm was the only tissue that always correlated with rescue, suggesting that the response to Hh signaling is required directly within the stomodeal precursors. A robust condensation on the stomodeal roof persists for at least 1 day in the mosaics (through at least 48 hpf; see Fig. S4 in the supplementary material). However, no cartilage develops, which is consistent,as argued above, with a later direct requirement for Smo function in the crest-derived skeletogenic cells. Again, the result indicates the occurrence of a later Hh-signaling event required for skeletogenesis, one separate from the early signaling for condensation.

Our mosaic analyses provide direct evidence that the reception of Hh signaling is required in the stomodeal ectoderm for the condensation of crest cells upon its roof. When does signaling to the stomodeum occur and what is the signaling source? To address the first part of this question, we used timed treatments with cyclopamine to disrupt the Hh pathway at specific stages and assayed the effects. We found that 4 hours of cyclopamine exposure, if initiated at 8-12 hpf or earlier, completely eliminated both stomodeal pitx2 expression and crest cell condensation on the stomodeal roof(Fig. 9A,C). Later treatments,even if longer in duration, were less effective by these assays (e.g. an 8 hour treatment from 14-22 hpf; Fig. 9A,D). Consistent with previous findings(Hirsinger et al., 2004), we found ptc1 levels to be highly disrupted 1 hour after cyclopamine treatment begins (data not shown). Therefore, the crucial time for Hh signaling to correctly specify marker expression in the stomodeum is about 10 hpf, shortly after gastrulation. Continued Hh signaling after 10 hpf is also likely to be important (Fig. 9A). Regardless of timing, all cyclopamine treatments resulted in severe pharyngeal cartilage loss that included first arch palatoquadrate and Meckel's cartilages (data not shown). Additionally, all cyclopamine treated embryos lacked the anterior neurocranium(Fig. 9E,F). However, as crest cells condense on the stomodeal roof in later treated embryos, the subsequent loss of the anterior neurocranium is probably due to the cell-autonomous requirement for Hh signaling as described for the mosaic analyses. Indeed, ptc1, a Hh signaling readout gene, is not expressed by neural crest cells at 14 hpf, but can be detected in neural crest cells adjacent to the eye at 22 hpf (data not shown).

At 10 hpf, the crucial time when the stomodeum must receive Hh signaling,the only Hh sources present in the head rudiment are Shh and its co-ortholog Twhh. Specifically, both are expressed in the midline by the ventral neural keel and the prechordal plate (Ekker et al., 1995; Krauss et al.,1993). Here, we use Hh-deficient mosaics to test which tissue is the crucial source of the Hh signal. To create animals deficient for both Shh and Twhh, we co-injected two morpholinos that individually block translation of shh or twhh (shh-MO; twhh-MO). The double morpholino treatment consistently resulted in the loss of crest cell condensation on the stomodeal roof (Fig. 10A) and a resulting loss of the anterior craniofacial skeleton,similar to the most severe shh hypomorphic mutants(Fig. 10; compare D and E with Fig. 1A and B, see legend for quantification). The stomodeal expression of pitx2, shh and fgf8 in shh-MO; twhh-MO embryos was greatly diminished (data not shown), although less consistently than in smomutants. Transplantation of wild-type presumptive ventral midbrain into shh-MO; twhh-MO hosts was able to significantly rescue both condensation and skeletal phenotypes (Fig. 10B,F), and the level of rescue positively correlated with the number of donor cells populating the ventral brain (data not shown). By contrast, even in instances when donor cells extensively populated the prechordal plate and notochord, the embryos still developed virtually no anterior craniofacial skeleton (Fig. 10C,G). Therefore, only the ventral brain is a significant Hh source for promoting condensation of crest cells on the stomodeal roof.

At the end of gastrulation, Hh secreted from the ventral brain primordium signals to ectodermal precursors of the stomodeum(Fig. 11A). Hh signaling specifies a relay, the ability of the stomodeum to interact with cranial neural crest cells that arrive in the vicinity some 8 to 10 hours later(Fig. 11B). The relaying signal from the stomodeum clearly does not involve Hh, probably occurs at close range, and is required for the condensation of neural crest cells upon the stomodeal roof. These condensed cells will generate the anterior craniofacial skeleton - the pterygoid process of the palatoquadrate in the upper jaw and the anterior neurocranium.

Furthermore, it is clear that for proper craniofacial development multiple reciprocal signaling interactions must continue after this early Hh-mediated signal and its relay, that we characterize in this paper. The recent work of Wada et al. (Wada et al.,2005), supported by our own study, clarifies that in zebrafish there occurs another, later signaling interaction that redeploys Hh, probably now secreted by oral ectoderm and acting directly on the neural crest. The second Hh signal requires the first (Fig. 6). Furthermore, specifying the expression of Shh in the stomodeum appears to require signaling to the ectoderm from the neural crest(Marcucio et al., 2005).

The hypothesis just described, particularly the understanding that Hh functions in two separate interactions, can resolve seemingly contradictory findings in zebrafish, chicken and mouse. Defects in the distribution of neural crest cells have been demonstrated in Shh-null mice(Washington et al., 2005),suggesting that the early role of Hh in organizing the condensation of crest cells on the stomodeal roof may be conserved between zebrafish and mice. In mouse mosaics, Smo mutant crest cells can form facial prominences in a wild-type host (Jeong et al.,2004), just as we observe smo mutant crest cells can develop condensations in zebrafish. Furthermore, in both species the mosaics reveal a crest-autonomous requirement for Smo in the formation of skeletal elements that stem from these prominences or condensations. In the chicken, early disruption of Hh signaling causes loss of expression of Shh and Fgf in facial ectoderm(Marcucio et al., 2005),matching our findings in the zebrafish, even though Marcucio et al. provide an alternative interpretation to ours, namely that the neural crest relays a Shh signal to the facial epithelium. Although our hypothesis fully explains their findings, their model could not work for zebrafish because it requires an early smo function in the neural crest, and we show this function is dispensable. We propose that the same early signaling cascade underlying anterior craniofacial development is widely shared among vertebrates.

How does the stomodeum organize the condensation and later development of neural crest cells? Facial ectoderm has been shown to be important for the proper outgrowth of anterior craniofacial skeletal elements(Hu et al., 2003; MacDonald et al., 2004), and here we present data that the oral ectoderm is essential for the initial condensation of neural crest cells that contribute to these skeletal elements. The same kinds of ectomesenchymal-epithelial interactions are crucial for skeletal development in the second arch as well as the first(Crump et al., 2004a; Crump et al., 2004b). In both arches, crest cells condense on an epithelium and later chondrogenesis is strictly correlated with, and presumably dependent upon, condensation. However, in the second arch, the crucial epithelium revealed in our studies is the first pharyngeal pouch, derived from endoderm. Furthermore, we find that Hox gene expression appears to specify the ability of second arch crest cells to interact with the pouch endoderm (J.G.C., M.E.S., J.K.E. and C.B.K.,unpublished). First arch crest cells lack Hox gene expression and this difference could, in part, underlie whether crest cells interact with endoderm as in the second arch, or with stomodeal ectoderm as in the first arch.

The transcription factor pitx2 is an excellent candidate for a gene responding to Hh that regulates the role the stomodeum plays in crest cell condensation. Pitx2 mutations in mouse and human are known to cause craniofacial defects (Lu et al.,1999; Saadi et al.,2003; Saadi et al.,2001), but it is unclear if these defects are due to the loss of Pitx2 specifically in the stomodeum. Earlier requirements for pitx2 in zebrafish (Essner et al., 2000) have obscured potential anterior craniofacial defects in loss-of-function analyses (J.K.E., unpublished). Conditional expression of dominant-negative Pitx2 (Saadi et al., 2003; Saadi et al.,2001) within stomodeal ectoderm may be useful for determining the role of Pitx2 in craniofacial development in the future.

Secreted signaling factors might be involved in the stomodeal-crest cell interaction. Along with pitx2, both fgf8 and shhare expressed by the zebrafish stomodeum(Miller et al., 2004; Miller et al., 2000) and are lost in smo mutants (this study). We observed little or no overlap of shh and fgf8 in the wild type, and this apparent restriction may be functionally relevant because interfaces of shh and fgf expression are known to direct craniofacial outgrowth in avian embryos (Abzhanov and Tabin,2004; Hu et al.,2003; MacDonald et al.,2004). We only detect stomodeal shh and fgf8expression after the time of crest cell condensation, arguing that their role is indeed in promoting outgrowth, rather than signaling the crest cells to condense. Although a recent study in zebrafish has reported that the stomodeum can partially rescue midline ethmoid plate defects in more mildly affected Hh mutants and shh-MO animals (Wada et al., 2005), we find no ability of the stomodeum to rescue the condensation and development of anterior craniofacial precursors in more severely affected shh-MO; twhh-MO embryos (data not shown).

We show that normally first arch crest cells closely adhere to both the roof and floor of the stomodeal epithelium. Our time-lapse analyses with mosaics show that wild-type crest cells in the smo mutant environment contact but fail to condense on the roof of the stomodeum, whereas crest cells condense normally on the mutant stomodeal floor. A clear inference is that the stomodeum provides differential adhesive substrates on the roof and floor and that Hh signaling selectively patterns the roof. Cell-ECM and cell-cell adhesion, implicated in condensing craniofacial neural crest cells(McKeown et al., 2005), may provide a mechanism for Hh mediated stabilization of anterior craniofacial neural crest cells. It will be of great interest to discover if any adhesion molecules are differentially expressed on stomodeal roof versus floor and if the expression of adhesion molecules on the roof is Hh dependent.

We greatly appreciate comments on the manuscript provided by R. Marcucio,C. T. Miller and T. Piotrowski. We also to thank W. A. Cresko for his statistical expertise. We are deeply indebted to J. Dowd, J. Wofford and Bonnie Ullmann for their expert fish care and technical help. We thank T. F. Schilling, N. Wada and Y. Javidan for sharing unpublished data. This work was supported by NIH grants HD22486 and DE13834 to C.B.K.; NIH Ruth L. Kirschstein NRSA HD043584 to J.K.E.; and an O'Donnell Fellowship of the Life Sciences Research Foundation to J.G.C.