An essential role for Fgfs in endodermal pouch formation influences later craniofacial skeletal patterning

The cartilages and bones that form the skeleton of the face and, in mammals, the middle ear, are derived from a specialized population of ectomesenchyme, the cranial neural crest(Le Douarin, 1982; Weston et al., 2004). Cranial neural crest cells (NCC) originate adjacent to neural ectoderm and migrate in three streams (mandibular, hyoid and branchial) to form seven pharyngeal arches. Segmentation of NCC into distinct streams is coupled to the segmentation of the hindbrain into rhombomeres (R1-7)(Kontges and Lumsden, 1996). NCC that contribute to the formation of the mandibular arch delaminate adjacent to posterior midbrain-R2 and do not express Hox genes, whereas NCC of the hyoid and branchial arches originate next to R4 and R6-R7, respectively,and are Hox-positive (Schilling and Kimmel, 1994; Trainor and Krumlauf, 2001).

Fibroblast growth factors (Fgfs) are a family of extracellular signaling molecules that have been implicated in diverse facets of vertebrate craniofacial development. Fgf8 from the oral surface ectoderm induces patterns of gene expression in adjacent mandibular mesenchyme, subdividing the mandibular arch into rostral odontogenic and caudal skeletogenic fields(Tucker et al., 1999) and controlling the position of the jaw joint(Wilson and Tucker, 2004). In addition to functions in mandibular arch development, Fgfs have roles in the development of cartilages derived from the hyoid and branchial arches and in the formation of pharyngeal pouches. Pouches are outpocketings of the pharyngeal endoderm that interdigitate with the crest-derived pharyngeal arches. Fgf8^neo/– mice, which are hypomorphic for Fgf8, display a range of craniofacial abnormalities that include reductions in cartilages and bones derived from all pharyngeal arches and disorganized endodermal pouches (Abu-Issa et al., 2002). Likewise, in the zebrafish acerebellar(ace) mutant, a strong loss-of-function mutation of fgf8(hereafter referred to as fgf8^–), hyoid cartilage is reduced and pouches are misshaped (Draper et al., 2001; Reifers et al.,1998; Roehl and Nusslein-Volhard, 2001).

Increasing evidence suggests that signals from the pharyngeal endoderm pattern the bones and cartilages of the pharyngeal arches. Analysis of casanova (cas) mutant zebrafish, which make no endoderm(Alexander et al., 1999),suggest that endoderm is required for the development of all pharyngeal cartilages (David et al.,2002). In tbx1 (van gogh) mutant zebrafish,pharyngeal pouches are largely absent and cartilages are misshaped and fused with those of adjacent arches (Piotrowski and Nusslein-Volhard, 2000), suggesting that pouches contribute to the segmentation of NCC into distinct arches. In addition, transplantation experiments in chick show that foregut endoderm is both necessary and sufficient to induce the shape and orientation of pharyngeal skeletal elements(Couly et al., 2002). One role of pharyngeal endoderm may be to locally promote the survival of skeletogenic NCC (Crump et al., 2004). Fgf3 has been shown to be required in pouch endoderm for the survival of skeletogenic NCC of the hyoid and branchial arches(David et al., 2002; Nissen et al., 2003), and a similar NCC survival-promoting role for endodermal Fgf8 has been proposed but not proven in zebrafish (Walshe and Mason,2003a). Clearly, understanding how pharyngeal endoderm develops is critical for understanding the later development of the pharyngeal skeleton.

In this study, we investigate an earlier role for Fgfs in the formation of pharyngeal pouches. Whereas in fgf8^– animals pharyngeal pouches are variably misshaped, we find that reducing both Fgf8 and Fgf3, by injecting fgf8^– animals with an fgf3 morpholino (fgf3-MO), leads to a complete failure of pouch formation. We use time-lapse microscopy to show that pharyngeal pouches form by the directed lateral migration of periodic clusters of endodermal cells. In fgf8^–; fgf3-MO animals,pharyngeal endodermal cells are present but their lateral migration is disorganized and discrete pouches fail to form. We use the Fgf receptor-inhibiting drug SU5402 to show that Fgf signaling is required during early somite stages for first pouch formation. At these stages, fgf8is expressed in the head in lateral mesoderm(Reifers et al., 2000) and in midbrain-hindbrain boundary (MHB)-R2 and R4 domains of the hindbrain(Maves et al., 2002; Reifers et al., 1998). Starting at 4-somites (11.3 hours post-fertilization), fgf3expression overlaps fgf8 expression in neural MHB and R4 domains(Maves et al., 2002; Walshe et al., 2002), and mesodermal and neural Fgf expression domains are in close proximity to developing pharyngeal endoderm. At later stages, fgf3 and, to a lesser extent, fgf8 are expressed in the pharyngeal pouches(David et al., 2002; Walshe and Mason, 2003a). However, we use mosaic analysis to show that Fgfs are required in mesodermal and neural domains, and not in the pharyngeal endoderm, to rescue pharyngeal arch structure in fgf8^–; fgf3-MO animals. Thus, we find an essential, Fgf-dependent function of the brain and head mesoderm in controlling segmentation of the pharyngeal endoderm into pouches.

In addition to their requirement in pouch formation, we find that Fgf8 and Fgf3 have redundant, essential functions in pharyngeal cartilage development. In fgf8^–; fgf3-MO animals, hyoid and branchial cartilages are largely absent and mandibular cartilages are reduced. By imaging pouch structure early and cartilage structure later in individual sides of animals in which Fgf signaling has been manipulated, we find that altered pouches correlate with later rearrangements of the cartilage pattern. This analysis suggests that pharyngeal pouch structure is a critical determinant of the pharyngeal cartilage pattern. We present a model in which an earlier function of Fgfs in pouch formation, in addition to their well-documented role as pouch-secreted survival factors later in development,contributes to the diversity of craniofacial phenotypes seen in fgf8mutants.

Zebrafish (Danio rerio) were raised and staged as previously described (Kimmel et al.,1995; Westerfield,1995). Time (hpf) refers to hours post-fertilization at 28.5°C. The wild-type line used was AB. Homozygous acerebellar^ti282a (ace) mutant embryos were scored by their loss of the cerebellum or loss of midbrain pax2aexpression (Brand et al., 1996; Reifers et al., 1998). fli1-GFP albino transgenic fish are the same as TG(fli1:EGFP)^y1; alb^b4(Lawson and Weinstein, 2002)and H2A.F/Z:GFP transgenic fish are as described(Pauls et al., 2001).

The fgf8 MOs E2I2 and E3I3(Draper et al., 2001) were used at 0.5 mg/ml each. To generate fgf8^–; fgf3-MO embryos, we pressure-injected fgf8^–embryos at the one-cell stage with 5 nl of the fgf3B (1.0 mg/ml) +fgf3C (0.25 mg/ml) MO combination. As previously described(Maves et al., 2002), this fgf3 MO dose gave highly reproducible fgf8^–; fgf3-MO phenotypes, scored as either the lack of ears or the lack of R5 krox20 staining.

Alcian Green staining was performed as described(Miller et al., 2003). For flat-mount dissections, Alcian-stained animals were digested for 1 hour in 8%trypsin at 37°C and transferred to 100% glycerol. Cartilages were dissected free from surrounding tissues with fine stainless-steel insect pins and photographed using a Zeiss Axiophot 2 microscope. Image background was cleaned up with Adobe Photoshop. For immunocytochemistry, embryos were prepared as described (Maves et al.,2002). Antibodies were used at the following dilutions: rabbit anti-GFP, 1:1000 (Molecular Probes); Zn8, 1:400(Fashena and Westerfield,1999; Trevarrow et al.,1990), goat anti-rabbit Alexa Fluor 488 and anti-mouse Alexa Fluor 568, both 1:300 (Molecular Probes).

The following cDNA probes were used: dlx2(Akimenko et al., 1994); krox20 (Oxtoby and Jowett,1993); pax2a (Krauss et al., 1991); axial(Odenthal and Nusslein-Volhard,1998); nkx2.7 (Lee et al., 1996); pea3(Brown et al., 1998). Probe syntheses and whole-mount in-situ hybridizations were performed as previously described (Hauptmann and Gerster,1994; Jowett and Lettice,1994).

fli1-GFP embryos were manually dechorionated and incubated in 40μl of EM with 0.4 mM SU5402 (Calbiochem) in agar wells. SU5402 was diluted from a 40 mM stock in DMSO. After 1- or 4-hour incubations, embryos were washed vigorously in EM. For 4-hour incubation experiments, sibling controls were fixed and processed for in-situ hybridizations with pea3.

Transplant techniques were as described(Maves et al., 2002). For endoderm transplants, donor embryos were injected at the 1-cell stage with an`Alexa568' mixture of 2% Alexa Fluor 568 dextran and 3% lysine-fixable biotin dextran (10,000 M_r, Molecular Probes) along with activated Taram-A receptor (TAR^*) RNA prepared according to David et al.(David et al., 2002). At 40%epiboly (ca. 4 hpf) donor TAR^* tissue was moved to the margins of fgf8^–; fgf3-MO; fli1-GFP host embryos. For neural and mesodermal transplants, donor embryos were injected at the 1-cell stage with Alexa568. For neural transplants, donor tissue was taken from the animal cap at shield stage (ca. 6 hpf) and moved to a position approximately two germ ring widths from the margin and 70° from dorsal in fgf8^–; fgf3-MO; fli1-GFP hosts(Maves et al., 2002). For mesodermal transplants, donor tissue was taken from the margin at 50% epiboly(ca. 5 hpf) and moved to the margins of fgf8^–; fgf3-MO; fli1-GFP hosts(Kimmel et al., 1990). All hosts were screened using a fluorescence stereomicroscope at 34 hpf, and only hosts with substantial, tissue-specific contributions to the pharyngeal endoderm, hindbrain or cranial mesoderm were used for subsequent analysis. In addition, the mesodermal transplant technique produced six embryos with contributions to both the hindbrain and cranial mesoderm. In order to control for variability in the effectiveness of the fgf3-MO, only fgf8^–; fgf3-MO; fli1-GFP hosts in which the ear was missing in at least one side were used for the analysis. In control transplants, mutant siblings in which donor tissue did not contribute to head tissues, we never observed the presence of an ear on only one side. At 34 hpf, confocal images of select host embryos were analyzed for pharyngeal arch structure. Rescue of arch structure was scored as complete if the first pouch and mandibular and hyoid arches were indistinguishable from wild type,and rescue was scored as partial for all other embryos with arch structure subjectively more organized than in fgf8^–; fgf3-MO; fli1-GFP controls.

For the endoderm movies (see Movies 3-6 in supplementary material),pharyngeal endoderm was labeled by transplanting donor tissue injected with a mixture of Alexa568 and TAR^* into GFP hosts. Embryos were selected for large fractions of labeled donor endoderm and bright GFP fluorescence using a Leica MZ FLIII fluorescence stereomicroscope. After manual dechorionation and anesthetization with buffered ethyl-m-aminobenzoate methane sulfanate (MESAB)(Westerfield, 1995), embryos were transferred to 0.2% agarose in embryo media (EM) with 10 mM HEPES and MESAB and then mounted onto a drop of 3% methylcellulose on a rectangular coverslip with three superglued #1 square coverslips on each side. A ring of vacuum grease was added around the embryo to make an airtight seal upon addition of the top coverslip. A heated stage kept the embryos at 28.5°C. Approximately 80 μm Z-stacks at 2 μm intervals were captured every 6 minutes using a Zeiss LSM5 Pascal confocal fluorescence microscope. At each time point, Z-stacks were projected with maximum intensity onto a single plane. Time-lapse recordings were further processed with Adobe Premiere. For single time point confocal sections, embryos were mounted without vacuum grease.

As previously reported (Roehl and Nusslein-Volhard, 2001), fgf8^– zebrafish had incompletely penetrant and expressive defects in the formation of pharyngeal pouches and cartilages (Fig. 1B,B′,F and especially Fig. 2I,J). Since the fgf8^– phenotype is relatively mild, we wondered if other Fgfs were partially redundant with Fgf8 in patterning pharyngeal arches. Fgf3 was a good candidate, as it has been shown to act redundantly with Fgf8 to pattern the posterior hindbrain, forebrain and ear(Maroon et al., 2002; Maves et al., 2002; Phillips et al., 2001; Walshe et al., 2002; Walshe and Mason, 2003b). In addition, Fgf3 has been shown to play a role in zebrafish pharyngeal arch development (David et al.,2002; Nissen et al.,2003). Although fgf3-MO animals had largely normal pouches (Fig. 1C,C′)(David et al., 2002), we found that in fgf8^–; fgf3-MO animals no pouches were made (Fig. 1D,D′). In addition, we found that Fgf8 and Fgf3 acted redundantly to promote pharyngeal cartilage development. In fgf3-MO animals, the ceratobranchial (CB) cartilages, which derive from the branchial arches located posterior to the hyoid arch, were largely absent(David et al., 2002; Nissen et al., 2003). However,whereas hyoid cartilages of fgf3-MO and fgf8^– animals were relatively mildly affected at 4 days (Fig. 1F,G), nearly all of the hyoid and CB cartilages in fgf8^–; fgf3-MO animals were absent (Fig. 1H). Although reduced in size, mandibular cartilages were patterned correctly in fgf8^–; fgf3-MO animals (inset to Fig. 1H).

Since fgf8^–; fgf3-MO animals have both pouch and cartilage defects, we wondered whether defects in pouch development could be responsible for later cartilage losses. In order to study correlations between pouch and cartilage defects in individual animals, we took advantage of the fact that partially reducing Fgf function with an fgf8 morpholino (fgf8-MO), or genetically with the fgf8 mutation, causes variably penetrant and expressive phenotypes. We imaged pharyngeal arch structure in live fgf8-MO; fli1-GFP and fgf8^–; fli1-GFP embryos at 28 hpf and subsequently raised individuals to 4 days to examine cartilage. The fli1-GFP transgene(Lawson and Weinstein, 2002)labels cranial NCC shortly after ventrolateral migration and perdures as cells differentiate into the pharyngeal cartilage elements. fli1-GFP also marks the developing vasculature but is not expressed in the pharyngeal mesoderm or endoderm (except for early, transient expression in the second pouch; see below). At pharyngula stages, the endodermal pouches are evident as non-fli1-GFP-expressing regions separating the fli1-GFP-expressing NCC of the pharyngeal arches (black in Fig. 2A). In individual sides of fgf8-MO; fli1-GFP and fgf8^–; fli1-GFP animals we found a correlation between early arch structure and alterations of the hyoid cartilage pattern(Fig. 2C-J and Table 1). In all sides with abnormal first pouch morphology we observed hyoid cartilage alterations later. In some cases, the first pouch was `deformed'(Fig. 2C), invading hyoid NCC territory, and this arch phenotype was most often linked to a complete loss of the dorsal hyomandibular cartilage element(Fig. 2D). In other cases, the first pouch was `shifted' to a more posterior position(Fig. 2E), and this `shift' was correlated with changes in the shape and position of dorsal hyoid cartilage(Fig. 2F). In the most striking example of pouch–cartilage shape correlations, a small ectopic pouch, in addition to the normal first pouch, formed in the middle of the hyoid NCC territory (Fig. 2G), and in these same sides an ectopic cartilage element developed later in the hyoid arch (Fig. 2H). To confirm that the non-GFP-expressing regions observed in live animals probably corresponded to misshapen and ectopic pouches, and not another non-GFP-expressing tissue such as mesoderm, we fixed and stained fgf8^– animals displaying similar GFP phenotypes with the endoderm-labeling Zn8 antibody(Trevarrow et al., 1990). In numerous examples, non-GFP expressing regions in fgf8^– animals, of similar shapes and positions to those observed in the live analysis, were found to be Zn8-positive (red in Fig. 1B and Fig. 2I; and data not shown). Lastly, whereas the majority of fgf8^– animals with normal arch morphology early had no defects in the hyoid cartilage pattern later, we observed graded reductions of the hyomandibular cartilage element in some sides with no pouch defects (Table 1). Thus, Fgf8 is likely to have other functions in cartilage development, in addition to its role in controlling pouch development. Nonetheless, we conclude that, in contrast to simple cartilage losses, the alterations in hyoid cartilage shape and position observed in a subset of animals reduced for Fgf8 are most tightly correlated with early changes in pouch morphology.

After NCC migration the third, most posterior, NCC mass segments into the five branchial, or gill-bearing, arches from which the five CB cartilages subsequently develop (Fig. 3A). It has been previously shown that the segmentation of NCC into distinct arches requires the segmentation of the pharyngeal endoderm into pouches(Piotrowski and Nusslein-Volhard,2000). In animals reduced for Fgf8, we observed a range of CB cartilage phenotypes that suggests defects in the segmentation of NCC into distinct branchial arches. These phenotypes include reductions in the number of CB cartilages (Fig. 3B) and fusions of cartilages of adjacent segments(Fig. 3C); incompletely formed CB cartilages were also observed (Fig. 3C). In addition, as reported for fgf8^–animals (Roehl and Nusslein-Volhard,2001), more posterior pharyngeal pouches were reduced and disorganized in fgf8-MO; fli1-GFP animals(Fig. 2C,E,G). We investigated whether the CB cartilage defects seen in fgf8-MO animals might be secondary to posterior pouch formation defects that result in reduced branchial NCC segmentation.

In order to understand the cellular basis of NCC segmentation, we made time-lapse recordings of pharyngeal arch development in wild-type (see Movie 1 in supplementary material) and fgf8-MO (see Movie 2 in supplementary material) animals. Time-lapse recordings were made of wild-type animals expressing both the pan-nuclear H2A.F/Z:GFP and the NCC-expressing fli1-GFP (12-38 hpf; Movie 1). In wild-type animals, NCC migrated in three streams to ventrolateral regions, where they contributed to the formation of the pharyngeal arches. Starting at 12 hpf (5-somites),H2A.F/Z:GFP-labeled NCC were seen migrating in two streams (mandibular and hyoid) anterior to, and one stream (branchial) posterior to, the developing otic vesicle. By around 16 hpf, the NCC finished their migration and began to express fli1-GFP as they condensed to form the arch masses. Shortly after the initiation of fli1-GFP expression in NCC, the first branching of the branchial mass occurred as the third pouch was formed(Fig. 3D). Over the next 20 hours, the fourth, fifth and sixth pouches formed in an anterior-posterior(AP) wave of development, and by 38 hpf the branchial mass had been subdivided into the five segments from which the CB cartilages would arise(Fig. 3E-G).

By contrast, in the fgf8-MO; fli1-GFP example shown (see Movie 2 in supplementary material), one fewer branchial segment formed. We found that in this animal the reduced number of segments was due to the failure of the fourth pouch to develop. While the third pouch (i.e. the first pouch to subdivide the branchial mass) formed normally(Fig. 3H), the fourth pouch initiated outgrowth yet failed to fully subdivide the branchial mass into a new segment (Fig. 3I). By 38 hpf the fifth and sixth pouches had fully formed, but the fourth pouch had retracted, and what in wild-type animals would have been the second and third posterior branchial segments had fused together, resulting in one less branchial segment (Fig. 3J,K). Consistent with this animal forming one less branchial segment, we found that one less CB cartilage developed (data not shown). These results suggest that at least some, and possibly all, of the CB cartilage defects seen in fgf8^– animals are the result of a failure of posterior endodermal pouches to form properly and segment branchial NCC into discrete arches.

In order to determine when Fgfs act to control pouch development, we inhibited Fgf signaling at different times of development using the Fgf receptor antagonist SU5402. As extended (24-hour) treatment of embryos with SU5402 causes widespread death of NCC(David et al., 2002), we performed shorter treatments of SU5402 in order to dissociate requirements for Fgf signaling in pharyngeal pouch formation from those in NCC survival. After addition of SU5402 for 1- to 4-hour periods, followed by a washout, embryos were scored for pouch defects at 34 hpf. In the case of 4-hour treatments, we examined effectiveness of inhibition of Fgf signaling by expression of the Ets factor, pea3, in treated siblings. pea3 is a downstream target of Fgf signaling (Roehl and Nusslein-Volhard, 2001), and we observed partial-to-complete inhibition of pea3 expression during the treatment and gradual recovery after washout (data not shown, summarized in Fig. 4 legend). In embryos treated with SU5402 from 10 to 14 hpf, we found that the first pouch was specifically lost in 39% of embryos (Fig. 4B) and misshapen in another 26% of embryos (data not shown). In addition, when animals with first pouch defects were raised to 4 days, we observed specific losses of hyoid cartilage(Fig. 4E). By contrast, SU5402 treatments from 6 to 10 hpf and 14 to 18 hpf had lesser effects on pouch development (Fig. 4G). We found a similar temporal requirement for Fgf signaling using 1-hour SU5402 treatments (Fig. 4C,H). The highest penetrance of first pouch shape defects was seen when 1-hour SU5402 treatments began between 9 and 13 hpf. Interestingly, later (>20 hpf)treatments with SU5402 produced variable defects in the development of more posterior pouches but not the first pouch(Fig. 4F), consistent with more posterior pouches forming later in an AP temporal wave of development. In summary, we find that Fgf signaling has a peak requirement in first pouch development from 10 hpf (tailbud) to 14 hpf (10-somites).

As Fgf8 and Fgf3 are expressed in neural and mesodermal tissues at early somite stages, and then in the pharyngeal endoderm at later stages, we investigated which Fgf sources are required for pouch formation. Using transplantation techniques (see Materials and methods), we introduced wild-type tissues at pre-shield stages (<6 hpf) into fgf8^–; fgf3-MO; fli1-GFP embryos and then assayed for rescue of first pouch development based on GFP-labeled arch structure. First, we found that wild-type endoderm was not able to rescue first pouch structure (Fig. 5B-B′′) compared with control contralateral sides that did not receive transplants (Fig. 5A-A′′). Wild-type mesoderm(Fig. 5C-C′′) or neural tissue (Fig. 5D-D′′) alone was able to only partially rescue pouch and arch structure in less than half of fgf8^–;fgf3-MO; fli1-GFP hosts (Fig. 5F). By contrast, transplantation of wild-type neural and mesodermal tissues together completely rescued pouch structure in 3/6 embryos,and partial rescue was seen in another 2/6 embryos(Fig. 5E-E′′). As neural tube transplants alone were sufficient to rescue the ear and cerebellar defects of fgf8^–; fgf3-MO embryos(Fig. 5D′′′),yet failed to completely rescue pouch structure(Fig. 5D′), we conclude that the requirements for both neural and mesodermal tissues is not simply a function of restoring neural structure in fgf8^–;fgf3-MO embryos. Thus, Fgf8 and Fgf3 are required additively in both the neural tube and mesoderm, but not the endoderm, to promote morphogenesis of the pharyngeal endoderm into pouches.

In order to test whether the lack of pouches is due to a general reduction in pharyngeal endoderm, we examined early markers of pharyngeal endoderm in fgf8^–; fgf3-MO animals. In wild-type embryos, the first pouch began to form around 16 hpf (see below). At 18 hpf, nkx2.7 and axial normally were expressed in lateral pharyngeal endoderm, in particular the first pouch and regions where the second and more posterior pouches would form(Fig. 6A,E). In fgf3-MO animals, nkx2.7 and axial expression was similar to that seen in wild-type animals(Fig. 6C,G), whereas in fgf8^– animals nkx2.7 and axialwere present but first pouch staining was variably absent(Fig. 6B,F). Strikingly, in fgf8^–; fgf3-MO animals, these markers revealed that a significant amount of pharyngeal endoderm was present, yet there was no clear evidence of pouches(Fig. 6D,H). In addition, as assayed by axial staining, we saw no differences in the amount of pharyngeal endoderm at an earlier stage (10 hpf) between fgf8^–; fgf3-MO and wild-type animals (data not shown). These results suggest that Fgf8 and Fgf3 act to promote the segmentation of pharyngeal endoderm into pouches and not endoderm generation.

Similarly, in order to understand the losses of the NCC-derived cartilages in fgf8^–; fgf3-MO animals, we examined the development of NCC in doubly reduced animals. In 18 hpf wild-type embryos, dlx2 expression marked three postmigratory NCC masses: the mandibular, hyoid and branchial primordia(Fig. 6I). In fgf3-MO and fgf8^– embryos, the three dlx2-positive masses resembled those in wild-type animals(Fig. 6J,K). By contrast, in fgf8^–; fgf3-MO embryos, only two dlx2-positive masses were present(Fig. 6L). Based on their positions with respect to the R3 domain of the hindbrain, we interpreted these two masses as a disorganized mandibular mass and a single fused mass incorporating hyoid and branchial NCC. As hyoid NCC are generated adjacent to R4 and branchial NCC develop adjacent to R6-R7 domains(Schilling and Kimmel, 1994; Trainor and Krumlauf, 2001),the NCC fusions are probably due to the absence of intervening R5-R6 territory in fgf8^–; fgf3-MO embryos(Maves et al., 2002; Walshe et al., 2002). We observed similar dlx2 phenotypes at 12 hpf (data not shown). By 33 hpf, dlx2 labeled the mandibular and hyoid arches and four branchial segments in wild-type animals (Fig. 6M). In fgf3-MO and fgf8^–animals, dlx2 expression was only mildly reduced compared with wild-type controls (Fig. 6N,O). However, dlx2 expression in fgf8^–; fgf3-MO animals revealed that by 33 hpf most NCC were absent except for a reduced mandibular population (Fig. 6P). We observed similar NCC losses in fgf8^–; fgf3-MO animals based on the NCC expression of the fli1-GFP transgene(Fig. 1D). Moreover, the selective disappearance of hyoid and branchial NCC between 18 hpf and 33 hpf was consistent with the later specific losses of the hyoid and branchial cartilages in fgf8^–; fgf3-MO animals. In conclusion, we found requirements for Fgf8 and Fgf3 in both the early organization (12-18 hpf) and the later survival (33 hpf) of NCC.

Understanding the role of Fgfs in pouch formation requires a detailed knowledge of the cell behaviors underlying pouch development in wild-type animals. Surprisingly, little is known about the mechanism of pouch formation in any species. In order to investigate the morphogenesis of pouch endoderm directly, we made time-lapse recordings of Alexa568-labeled developing endoderm in H2A.F/Z:GFP; fli1-GFP zebrafish (10-30 hpf: see Movies 3,4 in supplementary material). As labeled endoderm was generated by a combination of TAR^* induction and transplantation techniques (see Materials and methods for details), a large fraction, but not all, of the endodermal cells can be seen in the recordings. H2A.F/Z:GFP labels the nucleus of every cell (Pauls et al.,2001) and helps to identify landmarks, whereas fli1-GFP labels postmigratory NCC. At 10 hpf of wild-type development, Alexa568-labeled pharyngeal endodermal cells were scattered along the surface of the yolk(Fig. 7A,A′), a distribution that closely resembled endodermal axial expression at this time (Reiter et al.,2001). Over the next 6 hours, endodermal cells underwent a medial migration and became increasingly packed together near the midline(Fig. 7B,B′ show a 14 hpf intermediate stage). Shortly after medial migration, endodermal cells that would give rise to the first pouch began to extend thin cytoplasmic processes and migrate back out laterally (18 hpf: Fig. 7C,C′ and more clearly in Movie 4). At this time, the first postmigratory NCC began to turn on fli1-GFP. During the next few hours, endodermal cells continued to migrate laterally in a directed fashion, and by 22 hpf the first pouch was nearly fully formed (Fig. 7D,D′). In addition, clusters of endodermal cells situated periodically along the AP axis migrated laterally to form progressively more posterior pouches in an AP wave of development. By 30 hpf in this recording,the positions of the first three pouches were clearly seen relative to the fli1-GFP-labeled NCC of the arches(Fig. 7E,E′). In three wild-type recordings, the lateral migration of endodermal cells was observed to underlie the formation of all labeled pouches. We conclude that the directed lateral migration of periodically spaced endodermal cells is the mechanism that generates the pharyngeal pouches.

In order to understand the cellular basis for the lack of pouches in animals reduced for Fgf8 and Fgf3, we made time-lapse recordings of pharyngeal endoderm development in fgf8^–; fgf3-MO;H2A.F/Z:GFP animals (10-30 hpf: see Movies 5, 6 in supplementary material). For technical reasons, we visualized pharyngeal endoderm by transplanting labeled, TAR^*-induced endoderm from wild-type donors into fgf8^–; fgf3-MO; H2A.F/Z:GFP hosts (see Materials and methods). However, as we knew that wild-type endoderm failed to make pouches in an fgf8^–; fgf3-MO background, we inferred that the endoderm defects we describe here would be the same as in non-mosaic fgf8^–; fgf3-MO animals. At 10 hpf, we saw a similar distribution of pharyngeal endodermal cells over the surface of the yolk as in wild-type animals(Fig. 7F,F′), consistent with our earlier finding that fgf8^–; fgf3-MO animals had no defect in 10 hpf endodermal axialexpression. In addition, the medial migration and subsequent compaction of endodermal cells near the midline was largely normal in fgf8^–; fgf3-MO animals(Fig. 7G-H′). However, by 18 hpf we saw defects in the lateral migration of endodermal cells(Fig. 7H,H′). Although endodermal cells extended cytoplasmic processes in fgf8^–; fgf3-MO animals, these processes were not always directed laterally and often retracted (see Movie 6 in supplementary material). By 22 hpf, endodermal cells had failed to migrate laterally and discrete pouches were not seen(Fig. 7I,I′). Moreover,putative pouch endodermal cells did not align into regularly spaced arrays and by 30 hpf had formed a single, unsegmented clump of cells laterally(Fig. 7J,J′). We conclude that Fgf8 and Fgf3 are not required for the initial generation or medial migration of pharyngeal endodermal cells. Instead, Fgfs have later functions in the directed lateral migration and regular spacing of pharyngeal endodermal cells along the AP axis, processes critical for the formation of discrete pouches.

We have demonstrated an essential role for Fgf signaling in the formation of pharyngeal pouches. Whereas fgf8^– animals had variable defects in pouch formation, animals reduced for both Fgf8 and Fgf3 lacked all pharyngeal pouches. In addition, transient inhibition of Fgf signaling with the Fgf receptor antagonist SU5402 caused similar defects in pouch formation. This essential function of Fgfs in pouch formation is probably conserved among vertebrates, as mice hypomorphic for Fgf8 or Fgf receptor 1 (FgfR1) have similar pouch defects to those in fgf8^– zebrafish(Abu-Issa et al., 2002; Trokovic et al., 2003).

Previous to this study, little was known about the cellular behaviors underlying pouch formation. Pharyngeal pouches arise as lateral branches of the foregut endoderm. Branching morphogenesis is a common theme in organogenesis, and various cellular mechanisms, such as clefting and cell migration, have been implicated in branch generation(Chuang and McMahon, 2003; Ghabrial et al., 2003; Sakai et al., 2003). By directly imaging pouch formation in developing zebrafish embryos, we showed that cell migration is the mechanism that drives the lateral branching of the pharyngeal endoderm into pouches. After coalescence of pharyngeal endoderm near the ventral midline, subsets of endodermal cells migrated laterally at periodic intervals along the AP axis to form pouches. As cells could be seen extending cytoplasmic processes laterally during migration, we propose that chemotactic or substrate cues in the local environment guide pouch endodermal cells to lateral positions.

Several lines of evidence indicate that the lack of pouches in fgf8^–; fgf3-MO animals is probably due to a defect in the later morphogenesis of pouch endoderm. Based on the expression of endodermal markers such as axial and nkx2.7, pharyngeal endoderm was present in fgf8^–; fgf3-MO animals, although we do not know if mediolateral patterning of the endoderm was completely normal. Whereas axial and nkx2.7 expression suggest that pouch formation was defective at early stages in fgf8^–; fgf3-MO animals, a lack of pouch formation was clearly seen later using the Zn8 antibody or transplant techniques to label endoderm. A role for Fgfs in the morphogenesis of pouch endoderm was most evident in time-lapse recordings of endodermal development in fgf8^–; fgf3-MO embryos. Whereas the generation, medial migration and coalescence of endodermal cells were largely normal, we saw defects in both the later lateral migration and AP positioning of pharyngeal endodermal cells in fgf8^–; fgf3-MO embryos. The migration of endodermal cells was delayed, and the thin cytoplasmic processes characteristic of migrating cells were disorganized in fgf8^–; fgf3-MO embryos. Thus, in the absence of Fgfs, putative pouch endodermal cells had the ability to migrate but could not orient themselves along the mediolateral axis. In addition, whereas in wild-type embryos pouch endodermal cells migrated laterally at periodic AP positions, in fgf8^–; fgf3-MO embryos endodermal cells remained a continuous mass occupying the anterior pharyngeal region. We propose that Fgf signaling may regulate both the migration and AP positioning of pouch endodermal cells, and future experiments are needed to elucidate the extent to which these processes are interrelated.

Our transplantation experiments demonstrated that Fgf8 and Fgf3 are required additively in the neural keel and head mesoderm to rescue first pouch formation in fgf8^–; fgf3-MO embryos. Consistent with this, inhibition of Fgf signaling from tailbud (10 hpf) to 10-somites (14 hpf), stages at which Fgf8 and Fgf3 are expressed in the neural keel and lateral mesoderm, blocked first pouch formation. An attractive model is that signals from the neural keel help to pattern both the pharyngeal endoderm and premigratory NCC (Fig. 8A). Such a strategy would link the two sources of pharyngeal segmentation, segmentation of NCC into distinct streams and segmentation of the endoderm into pouches, to the earlier segmentation of the hindbrain. Intriguingly, mandibular NCC originate from ectomesenchyme adjacent to the MHB and are Hox-negative, and the first pharyngeal pouch develops in the vicinity of MHB-R2 and is also Hox-negative (Hunt et al., 1991; Miller et al.,2004). By contrast, the second pouch and hyoid NCC develop adjacent to R4 and are both Hox-positive. Whereas development of the first two pouches connects to segmental expression of Fgf in the hindbrain, it is less clear how development of the more posterior pouches would be regulated. All pouches were lost in animals lacking both Fgf8 and Fgf3, and transient inhibition of Fgf signaling, at times later than those used to inhibit first pouch formation, disrupted the formation of more posterior pouches. However,pouches three through six develop at stages when Fgfs are no longer expressed in the hindbrain and head mesoderm. Further analysis will be required to determine how similar the functions of Fgfs in posterior pouch formation are to those described here for first pouch formation.

Although we showed that Fgfs are essential for pouch morphogenesis, our results do not distinguish between direct and indirect functions of Fgfs in pouch outgrowth. For example, Fgfs from the lateral mesoderm may act directly as chemoattractants for the lateral migration of pharyngeal endodermal cells. In Drosophila, the branching of the trachea also requires Fgf signaling (Klambt et al.,1992; Sutherland et al.,1996), and tracheal cells have been shown to migrate toward ectopic sources of Fgf (Sato and Kornberg,2002). Alternatively, Fgfs may influence pouch formation indirectly by regulating patterning of the hindbrain and lateral mesoderm. By 16 hpf, pharyngeal endodermal cells in close proximity ventrally to the neural keel and medially to lateral mesoderm begin to migrate to form the pouches(Fig. 8B). However, we could block first pouch formation by inhibiting Fgf signaling from 10-14 hpf,although we cannot rule out that there is a delay between addition of the drug and effective inhibition of Fgf signaling. In one model, the role of Fgfs would be to establish segmental signals in the hindbrain that control the later lateral migration of endodermal cells at periodic positions along the AP axis. Consistent with this, animals reduced for Fgf8 and Fgf3 have hindbrain defects that include losses of MHB and segments R1, R5 and R6, and reductions in size of additional rhombomeres (Walshe et al., 2002; Maves et al.,2002). If pouch endodermal cells are responding to segmental cues in the hindbrain for their migration, the reduced size and segmentation of the hindbrain may explain why endodermal cells did not migrate at discrete AP positions and ultimately formed compressed clumps in fgf8^–; fgf3-MO animals(Fig. 8C,D). In addition, as fgf8^– but not fgf3-MO zebrafish have defects in MHB structure (Reifers et al.,1998), a lack of early Fgf8-dependent MHB signals might explain why fgf8^– but not fgf3-MO embryos had variable first pouch defects. Similarly, Fgf8 has been shown to control both the gene expression profile and morphogenesis of the lateral head mesoderm(Reifers et al., 2000). Thus,Fgfs might promote pouch formation indirectly by controlling the positioning and expression of pouch guidance factors in the hindbrain and lateral mesoderm. Future experiments that address where Fgf signaling is required, for example by manipulating Fgf receptor function in the endoderm and other tissues, should help to clarify how directly Fgfs act to control pouch formation.

In fgf8^–; fgf3-MO animals, no pouches formed and little or no cartilage was made from the Hox-expressing NCC of the hyoid and branchial arches. Similarly, transient inhibition of Fgf signaling during early somite stages led to correlated losses of both the first pouch and dorsal hyoid cartilage. However, in fgf8^–; fgf3-MO animals, mandibular cartilages, which are derived from NCC that do not express Hox genes, were less affected. By contrast, casmutant zebrafish lack all endoderm and are missing cartilages derived from all pharyngeal arches (David et al.,2002), implying that Fgf-independent endodermal signals pattern cartilages of the mandibular arch. As pharyngeal endoderm was present but not segmented into pouches in fgf8^–; fgf3-MO animals, we conclude that the outpocketing of the pharyngeal endoderm to form pouches is a critical event that allows endoderm to induce cartilage in Hox-positive NCC.

By analyzing individual sides of fgf8^– animals,we found a correlation between early changes in pouch structure and later rearrangements of the cartilage pattern. In our studies of integrinα5 mutant zebrafish, we found that the first pouch promotes the local compaction and survival of NCC that give rise to specific regions of dorsal hyoid cartilage(Crump et al., 2004). In fgf8 mutant sides in which the first pouch was shifted in position or an ectopic pouch formed in presumptive hyoid NCC territory, we observed similar positional shifts of dorsal hyoid cartilage or ectopic cartilage elements. These correlations are consistent with the abnormal first pouch promoting hyoid cartilage development in abnormal locations. In other cases a deformed first pouch invaded NCC territory that, based on our previous fate mapping (Crump et al., 2004),normally forms dorsal hyoid cartilage, and this deformed pouch was correlated with a subsequent loss of dorsal hyoid cartilage. In addition, defects in the formation of more posterior pouches were correlated with losses and fusions of the ceratobranchial cartilages. Thus, whereas early pouch defects are largely predictive of later cartilage alterations, the precise interpretation of the resultant cartilage defects in fgf8^– animals is complicated by the fact that Fgf8 probably has multiple functions in pharyngeal cartilage development.

Our finding that pharyngeal pouches were essential sources of patterning information for the cartilages of the Hox-expressing hyoid and branchial arches is consistent with work in chicken showing that different types of foregut endoderm interact with Hox-expressing versus non-Hox-expressing NCC to specify cartilage pattern. In these studies, involving the transplantation and ablation of endoderm domains at stages prior to pouch morphogenesis, anterior endoderm can respecify cartilage pattern when transplanted adjacent to Hox-negative, but not Hox-positive, NCC(Couly et al., 2002). However,more posterior endoderm can respecify cartilages derived from Hox-positive NCC(Ruhin et al., 2003). Based on our analysis of Fgf function in zebrafish, we predict that the posterior endoderm domains that induce cartilage from Hox-positive NCC in chicken will include endodermal regions that form pharyngeal pouches during later embryogenesis.

Lastly, what are the pouch-derived factors that promote cartilage development? Recent evidence suggests that Fgfs themselves are expressed later in the pouches and promote the survival of skeletogenic NCC. Studies in zebrafish have shown that Fgf3 is required in the pharyngeal endoderm for the survival of hyoid and branchial NCC (David et al., 2002; Nissen et al.,2003). As Fgf8 is also expressed, albeit less strongly, in the pouches, it has been proposed that Fgf8 may act redundantly with Fgf3 as an endoderm-derived NCC survival factor(Walshe and Mason, 2003a). Interestingly, we did observe graded reductions of dorsal hyoid cartilage in some fgf8^– animals that had normal first pouches,consistent with Fgf8 also having a role in the later survival of NCC. Moreover, the lack of hyoid and branchial cartilages in fgf8^–; fgf3-MO animals is most consistent with a survival defect of postmigratory Hox-positive NCC. Based on dlx2 expression, hyoid and branchial NCC are present early but disappear later in fgf8^–; fgf3-MO animals. However, as pouches do not form in fgf8^–; fgf3-MO animals, the NCC survival defects are probably due in part to there being no pouches to secrete survival factors such as Fgf8 and Fgf3. As has been observed in tooth and lung development(Chuang and McMahon, 2003; Jernvall and Thesleff, 2000),it is becoming apparent that Fgfs also have multiple, temporally distinct functions during pharyngeal ontogeny. We propose that, in addition to a later function as pouch-derived survival factors, an essential early function of Fgfs in endodermal pouch morphogenesis may help explain the diversity of craniofacial phenotypes seen in fgf8^- animals.

We thank John Dowd and the UO Fish Facility for abundant help with raising fish; Jose Campos-Ortega for providing the H2A.F/Z:GFP line before publication; Craig T. Miller, Le Trinh, Nick Osborne and Tom Schilling for helpful discussions, especially about endoderm; and Johann Eberhart for comments on the manuscript. J.G.C. is an O'Donnell Fellow of the Life Sciences Research Foundation. L.M. was supported by a fellowship from the Damon Runyon-Walter Winchell Cancer Research Fund. Research is funded by NIH grants DE13834 and HD22486.