Context-specific requirements for Fgfr1 signaling through Frs2 and Frs3 during mouse development

Members of the receptor tyrosine kinase (RTK) superfamily elicit similar cellular, transcriptional and biochemical responses when stimulated in cultured cells (Fambrough et al.,1999; Schlessinger,2000), yet direct specific, non-redundant cellular responses in developing organisms. It has been proposed that qualitative differences in signal transduction underlie RTK functional specialization. Such differences could be in ligand sensitivity, effector pathway use or the strength or duration of signaling. This model has gained support from in vivo domain swap experiments, which have demonstrated biologically significant differences in RTK signal transduction (Klinghoffer et al., 2001; Madhani,2001; Hamilton et al.,2003). Analyses of mice harboring signaling domain mutations have also shown that effector pathways are differentially required within the RTK superfamily, and that many signaling events observed in biochemical studies are dispensable or required only for a subset of the in vivo functions of receptors (Partanen et al.,1998; Tallquist et al.,2000; Maina et al.,2001; Klinghoffer et al.,2002; Tallquist et al.,2003).

The mammalian fibroblast growth factor (Fgf) signaling network consists of four high affinity RTKs and at least 22 ligands(Ornitz, 2000; Böttcher and Niehrs,2005). Mouse knockout studies have identified numerous roles of Fgfs, and demonstrated that all essential embryonic roles are mediated by two receptors, Fgfr1 and Fgfr2(Deng et al., 1994; Yamaguchi et al., 1994; Deng et al., 1996; Arman et al., 1998; Yu et al., 2000; Böttcher and Niehrs,2005). We have focused our studies on Fgfr1, which is required for postimplantation growth, mesodermal migration and patterning during gastrulation, and for somitogenesis. Fgfr1^nullmouse embryos also exhibit posterior truncations and neural tube closure defects (Deng et al., 1994; Yamaguchi et al., 1994). Studies of hypomorphic and conditional alleles have identified additional roles of this receptor in node regression, neural stem cells, olfactory bulb development, and patterning of the anteroposterior axis, central nervous system and pharyngeal arches (Partanen et al., 1998; Tropepe et al.,1999; Xu et al.,1999; Hébert et al.,2003; Trokovic et al.,2003a; Trokovic et al.,2003b).

Though many roles of Fgfr1 are known, the mechanisms by which this receptor signals in vivo have not yet been elucidated. Active Fgfrs can directly engage relatively few proteins, namely Crk, Grb14, Shb, PLCγ and Frs2,3(hereafter referred to collectively as `Frs')(Mohammadi et al., 1991; Wang et al., 1996; Kouhara et al., 1997; Larrson et al., 1999; Reilly et al., 2000; Cross et al., 2002). Biochemical studies have implicated Frs adaptors as the key mediators of Fgfr signal transduction. These proteins interact constitutively with the intracellular juxtamembrane regions of Fgfr1 and Fgfr2, unlike signaling proteins downstream of other RTKs that are recruited after receptor activation. Fgfr activation leads to phosphorylation of Frs tyrosine residues to which Grb2 and SHP2 are subsequently recruited, initiating PI3K and MAPK signaling (Wang et al., 1996; Kouhara et al., 1997; Xu et al., 1998; Ong et al., 2000; Hadari et al., 2001). This adaptor-mediated mechanism may distinguish the kinetics, amplitude or subcellular localization of Fgfr signal transduction from other RTK signaling events.

Frs adaptors are conserved among vertebrates, and expression patterns suggest that the two mammalian isoforms perform non-redundant functions in vivo (McDougall et al., 2001; Gotoh et al., 2004). Loss-of-function studies have not been reported for Frs3, but Frs2^-/- mouse embryos die ∼E7.5 with defects in extra-embryonic development (Hadari et al., 2001; Gotoh et al.,2005). Mouse chimera and Xenopus knockdown studies have further implicated Frs2 in mesoderm development and convergent extension,respectively (Akagi et al.,2002; Gotoh et al.,2005). These studies are informative with respect to Frs2 roles,but they are difficult to interpret in terms of specific growth factor pathways because of the promiscuity of Frs adaptors. In addition to Fgfrs, Frs can interact with active neurotrophin receptors and are phosphorylated downstream of other RTKs (Rabin et al.,1993; Ong et al.,2000). Furthermore, both adaptors associate strongly with the cell cycle regulatory protein p13^suc1, and may thus play a role in cell cycle regulation or progression (Rabin et al., 1993; Ong et al.,1996).

To assess the developmental requirements for Frs-mediated signaling downstream of Fgfr1, we have generated mice lacking the Frs-binding site on this receptor (Fgfr1^ΔFrs). We found that Frs-mediated signaling is dispensable for Fgfr1 functions during gastrulation and somitogenesis, but is required for Fgfr1 roles in neurulation, tail bud and pharyngeal arch development. Furthermore, we found that primary embryonic cells do not require this signaling interaction to elicit strong and sustained MAPK responses to Fgf, and that MAPK activation in vivo is not grossly affected by the Fgfr1^ΔFrsmutation. These results indicate that Frs adaptors are not the exclusive effectors of Fgfr1 signal transduction in vivo.

Targeting vectors were generated in pPGKneoF2L2DTA, a plasmid containing a DTA selection cassette and a neomycin cassette flanked by nested loxP and FRT sites. For knock-in (KI) vectors (Fgfr1^wtKI,Fgfr1^ΔFrs), the short arm of homology, an XbaI/HindIII fragment spanning introns 6-7 of Fgfr1, was inserted upstream of the 5′ loxP site in the same orientation as the neomycin cassette. Intron 7-partial cDNA cassettes were generated through two rounds of PCR SOEing (splicing by overhang extension,oligos are given in Table S1 in the supplementary material)(Pogulis et al., 2000). In the first SOEing reaction, nucleotides 1276-1356 (amino acids 407-433) were deleted from a mouse Fgfr1 cDNA construct to generate an Fgfr1^ΔFrscDNA. In the second round, partial Fgfr1^wt, Fgfr1^ΔFrs cDNAs (exons 8-17) were spliced to the 3′ end of intron 7, preserving the endogenous intron 7/exon 8 junction. All PCR products incorporated into targeting vectors were generated using Pfu (Stratagene) and were verified by sequencing. Intron 7-partial cDNA cassettes were digested with Tth111I and ligated to the long arm of homology, a ∼4.1 kb Tth111I/PstI genomic fragment. Resultant fragments were cloned between the 3′ loxP site and DTA(Fig. 1B).

For the Fgfr1^floxex4 vector, the short arm of homology,exon 4 fragment, and part of the 3′ long arm of homology were PCR amplified from wild type 129Sv genomic DNA (oligos are given in Table S1 in the supplementary material). The exon 4 fragment was cloned between loxP sites, upstream of the FRT-flanked neomycin cassette in pPGKneoF2L2DTA. The short arm was inserted upstream of the 5′ loxP site; the intron 5 PCR fragment was ligated to a genomic HindIII/EcoRV fragment to complete the long arm, which was inserted between the second loxP site and DTA(Fig. 1C).

AK7 ES cells were electroporated with linearized targeting vectors, G418 selected and PCR screened for integration at the Fgfr1 locus (data not shown). Targeting was verified by Southern analysis with the following probes: Fgfr1^KI-5′ external(NcoI/XmaI), internal (SpeI/NciI),3′ external (PstI/HindIII); Fgfr1^Δex4-5′ external (PCR product from 129Sv genomic DNA, oligos are given in Table S1 in the supplementary material); internal (NotI/NcoI).

Correctly targeted ES cells were injected into wild type C57BL/6J blastocysts to generate chimeras. The neomycin cassette caused early embryonic lethality of KI homozygotes and was excised through breeding to Meox2-Cre or FlpeR mice (Farley et al.,2000; Tallquist and Soriano,2000). Complete excision of neomycin (Fgfr1^KI,Fgfr1^floxex4) and exon 4 (Fgfr1^floxex4)was verified by Southern analysis (data not shown). Tail biopsies were routinely PCR genotyped (oligos are given in Table S1 in the supplementary material). In KI homozygotes, embryonic RT-PCR confirmed that targeting did not disrupt alternative splicing of the extracellular domain or splicing of upstream exons into the cDNA (data not shown).

Whole-mount in situ hybridization was performed using standard procedures.

Embryonic RNA was prepared using Trizol (GibcoBRL) and reverse transcribed as described previously (Chen and Soriano,2003). Diluted cDNA pools were amplified using the following primer pairs: FGFR1.52/FGFR1.36, FGFR2ex11.51/FGFR2ex13.31,Timm_cDNA+/Timm_cDNA- (see Table S1 in the supplementary material). Triplicate PCR reactions were cycled within a linear range of amplification and products were detected and quantified using Sybr Gold (Molecular Probes) and NIH ImageQuant software. Fgfr1 and Fgfr2 RT-PCR products were normalized to Timm (mitochondrial membrane transport protein)levels.

Whole-mount phospho-MAPK IHC was performed as described previously(Corson et al., 2003), with minor modifications.

For thin section immunohistochemistry, embryos were fixed overnight (4%PFA), embedded in paraffin wax and sectioned. After rehydration, slides were pretreated in (3% H₂O₂, 10% methanol, 1×PBS). Antigen retrieval was performed by microwaving slides in 10 mM citrate buffer(pH 6.0) (phosphohistone H3, Upstate Biotechnology #06-570) or 10 mM EDTA (pH 8.0) (pMAPK, Cell Signaling #9101S). Samples were blocked in 5% serum/PBS and incubated with antibody overnight. Immunocomplexes were detected using biotinylated secondary antibodies with ABC Elite and DAB substrate kits (Vectastain).

Freshly isolated embryos were stained in Nile Blue (0.003% in DMEM, 0.5%FBS) for 30-60 minutes (37°C) and washed in PBS. For TUNEL analysis,embryos were fixed, embedded and sectioned as for immunohistochemistry. Rehydrated sections were incubated in (0.3% Triton-X-100, 1x PBS), digested with proteinase K, and incubated in TdT reaction mix for 1 hour, 37°C [30 mM Tris pH 7.5, 140 mM cacodylate, 1 mM CoCl_2, 10 μM dig-dUTP,0.3 U/μL TdT (Roche)]. Reactions were stopped in PBS and TUNEL-positive nuclei were visualized by IHC with anti-Dig-AP Fab (Roche) and NBT/BCIP detection.

Skeletons were prepared and stained with Alcian Blue and Alizarin Red using a standard protocol.

E12.5-14.5 embryos were trypsinized (5′, 37°C), gently disaggregated with a Pasteur pipette, and plated on gelatin in DMEM, 15% FBS. To generate immortalized lines, p1 MECs were infected with an E6LTTNLoxP retrovirus transducing the SV40 Large T antigen(Berghella et al., 1999). Infected cells were selected with 500 μg/ml G418 and resistant colonies were pooled.

MECs were seeded at 1.5-1.8×10⁴ cells/cm²,serum starved (24-36 hours, 0.5% FBS) and stimulated with heparin and Fgf(aFgf and bFgf gave similar results; Research Diagnostics). Cells were then washed with cold PBS and lysed in HNTG (20 mM HEPES pH 7.9, 150 mM NaCl, 1%Triton X-100, 10% glycerol, 1.5 mM MgCl₂, 1 mM EGTA, 1 μg/ml aprotinin, 1 mM PMSF, 10 mM NaPP_i, 0.2 mM activated Na₂VO₃, 50 mM NaF). IPs were performed overnight in HNTG; immunocomplexes were pulled down using protein A- or G-PLUS agarose(Santa Cruz) and washed in HNTG or (500 mM NaCl, 10 mM Tris pH 7.5, 2 mM EDTA,1% NP40). SDS-PAGE and western blotting were performed according to standard protocols, with phospho-specific and -nonspecific antibody blots blocked in 1%BSA and 3% milk, respectively.

Antibodies: pMAPK (Cell Signaling); Frs2 (Santa Cruz H-91); Fgfr2 (mouse monoclonal, Research Diagnostics); RasGAP (70.3; gift of Andrius Kazlauskas; Valius et al., 1993);phosphotyrosine (clone 4G10, Upstate Biotechnology); pAkt (Cell Signaling);Crk (BD Transduction Labs); and actin (clone AC-15, Sigma).

We used a partial cDNA knock-in approach to generate mice lacking the Frs binding site on Fgfr1 without disrupting alternative splicing of exons encoding the extracellular domain (Fig. 1) (Werner et al.,1992; Yan et al.,1992; Johnson and Williams,1993). This deletion mutation has previously been shown to abrogate Fgfr1 binding to both Frs2 and Frs3 without disrupting other receptor functions and interactions (Xu et al.,1998). As a control, we used the same approach to generate wild-type knock-in mice (Fgfr1^wtKI; the untargeted allele is designated Fgfr1⁺).

Following two or more backcross generations to the C57BL/6J background, we recovered nearly the expected ratio of Fgfr1^wtKI/wtKIanimals at E16.5, and a number of homozygotes survived to adulthood with normal fertility (Table 1). We therefore selected this background (>75% contribution from C57BL/6J) for our studies. On this genetic background, Fgfr1^wtKI/wtKIembryos are often smaller than littermates and exhibit digit 3/4 fusions (data not shown). Importantly, however, Fgfr1^ΔFrs/ΔFrs phenotypes discussed below are either absent or notably less penetrant/severe in Fgfr1^wtKI/wtKI embryos (see Fig. S1 in the supplementary material). Phenotypic differences between the knock in lines were not due to differences in Fgfr1 expression: Fgfr1 mRNA levels are similar in untargeted, Fgfr1^wtKI/wtKI and Fgfr1^ΔFrs/ΔFrs embryos on the high percentage C57 background (Fig. 1E).

Because the Fgfr1^wtKI/wtKI phenotype is background dependent (data not shown), we revisited the null phenotype on the (>75%)C57BL/6J background. We generated mice harboring a null allele(Fgfr1^Δex4, Fig. 1A,C) and backcrossed them more than two generations to C57BL/6J. Compared with published Fgfr1^null embryos, which were analyzed on different mixed backgrounds,Fgfr1^Δex4/Δex4embryos on the (>75%) C57BL/6J background survive slightly later in embryogenesis (E11.5 vs. E7.5-9.5). However, they are always developmentally arrested prior to E9.5 and exhibit phenotypes similar to published Fgfr1^null embryos, including a developmental delay,mesodermal migration and patterning defects, craniorachischisis and posterior truncations (Fig. 2; Fig. 3B)(Deng et al., 1994; Yamaguchi et al., 1994). Fgfr1^Δex4/Δex4 embryos also have enlarged hearts with disrupted looping (data not shown).

Fgfr1^ΔFrs/+ animals are viable and indistinguishable from littermate controls, but Fgfr1^ΔFrs/ΔFrs mice were never recovered postnatally (Table 1). Timed matings indicated that Fgfr1^ΔFrs/ΔFrs embryos survive much later in embryogenesis than do Fgfr1^Δex4/Δex4 embryos, with a drop in viability only after E15.5 (Fig. 2A).

Fgfr1 roles in early mesodermal development were rescued in∼80% of Fgfr1^ΔFrs/ΔFrsembryos. During gastrulation, Fgfr1 is required for mesodermal migration through the primitive streak. In Fgfr1^Δex4/Δex4 (and published null) embryos, this migration is impaired and cells accumulate in the streak. Consequently, there is an expansion of axial mesoderm and reduction of paraxial mesoderm, which remains disorganized and fails to form somites (Fig. 2B,C)(Deng et al., 1994; Yamaguchi et al., 1994; Ciruna et al., 1997; Ciruna and Rossant, 2001). We examined the development of axial and paraxial mesoderm in Fgfr1^ΔFrs/ΔFrs embryos by Shh and Meox1 in situ hybridizations, respectively. Although Fgfr1^ΔFrs/ΔFrs embryos are developmentally delayed by ∼1 day, most mutants exhibit normal expression of both mesodermal markers following gastrulation(Fig. 2D-G). This indicates that mesodermal migration is rescued in these embryos. Furthermore, whereas Fgfr1^Δex4/Δex4 embryos never form somites, Fgfr1^ΔFrs/ΔFrsembryos undergo normal somitogenesis, as evidenced by the segmented pattern of Meox1 staining.

Thirty to 50% of Fgfr1^ΔFrs/ΔFrs embryos isolated E10.5-11.5 exhibited a defect in spinal neural tube closure(Fig. 3A,C). This was never observed in Fgfr1^wtKI/wtKI embryos, and differs from the neurulation phenotype observed in Fgfr1^Δex4/Δex4 embryos in which the neural tube remains open along the entire rostrocaudal axis(craniorachischisis; Fig. 3B). The spinal neural tube (caudal to approximately somite 8) closes by the elevation and fusion of neural folds in a progressive rostral-to-caudal manner(Shum and Copp, 1996; Copp et al., 2003). In some Fgfr1^ΔFrs/ΔFrs embryos, the spinal neural tube remained completely open at E10.5, which could indicate a mere delay in the closure process (Fig. 3C, right). In others, however, the spinal neural tube was closed at discrete points but open in the intervening regions(Fig. 3C, left). Later in development, some Fgfr1^ΔFrs/ΔFrs embryos completed neural tube closure and others exhibited varying degrees of spina bifida (Fig. 3D-F, data not shown). In one embryo (E17.5), spina bifida was accompanied by loss of spinal cord marginal layer tissue and caudal meningomyelocele (extrusion of the spinal cord and meninges from the vertebral column; data not shown).

Histological analysis revealed altered morphology of the canal through mutant neural tubes in the spinal region. This was observed in mutant embryos that completed closure, as well as in those that retained regions of open neural tube (Fig. 3E,F). During neurulation, spinal neural folds bend towards the midline through the formation of neuroepithelial hingepoints; this morphogenetic process occurs by different mechanisms at different axial levels. In the upper- and mid-spinal regions, the neuroepithelium bends at one medial hingepoint and the lateral edges fold towards the dorsal midline, generating a slit-shaped neural canal(e.g. Fig. 3D). In the lower spinal region, two dorsolateral hingepoints (DLHP) form in addition to the medial hingepoint, resulting in a diamond-shaped canal(Shum and Copp, 1996). In Fgfr1^ΔFrs/ΔFrsembryos, the canal is diamond shaped at all spinal levels(Fig. 3E,F), suggesting that this second morphogenetic mechanism is used to close the entire spinal neural tube. Cranial neuroepithelial folding appeared normal in all Fgfr1^ΔFrs/ΔFrs embryos (data not shown).

Normally, notochordal Shh inhibits DLHP formation at mid- and upper spinal levels, while signals from the dorsal ectoderm are thought to play an inductive role (Copp et al.,2003). In Fgfr1^ΔFrs/ΔFrsembryos, the notochord appears normal, extends through the spinal region, and expresses Shh (Fig. 2E, Fig. 3D-F). Although ectodermal-inducing signals have not yet been identified, it has been proposed that DLHP formation involves localized proliferation or apoptosis in the dorsolateral neuroepithelium (Copp et al.,2003). We analyzed proliferation and apoptosis in Fgfr1^ΔFrs/ΔFrs neural tubes and found that although the mitotic index is not significantly altered in these embryos, they do exhibit an increase in cell death within the spinal neural tube (Fig. 3G). However,as apoptosis was not specifically localized to dorsolateral regions (data not shown), this probably reflects a role of Fgfr1-Frs signaling separate from its role in neurulation.

Neural tube closure defects are often associated with posterior truncations, and it has been postulated that the curvature of the body axis generated during tail bud outgrowth provides tension in the neural folds that facilitates closure (Copp et al.,1982; Copp, 1985; Gofflot et al., 1997; Peeters et al., 1998; Finnell et al., 2003). Over 80% of Fgfr1^ΔFrs/ΔFrsembryos exhibit posterior truncations of varying severity(Fig. 4A-C, see Fig. S1A in the supplementary material). Truncations persist through embryogenesis and in severe cases are accompanied by a truncation of the notochord at the level of the hindlimb and failure to bifurcate caudal limb fields and organs(Fig. 4D,E, data not shown). In previous Fgfr1 reports, posterior truncations were associated with ectopic posterior neuroectoderm and suggested to be secondary to the mesodermal defects (Ciruna et al.,1997; Ciruna and Rossant,2001). However, in Fgfr1^ΔFrs/ΔFrs embryos, we observed truncations in the absence of these other phenotypes(Fig. 2, data not shown),indicating that Fgfr1-Frs signaling directly impacts tail bud development.

All tissue layers of the tail bud form during secondary gastrulation from remnants of the primitive streak and node. In most Fgfr1^ΔFrs/ΔFrs embryos, Shh in situ hybridization and Hematoxylin and Eosin staining of embryo sections indicated full extension of the notochord, and hence, correct positioning of tail bud progenitors, at the onset of secondary gastrulation(Fig. 2E, data not shown). We investigated whether the Frs site deletion directly affected patterning or outgrowth of the tail bud. Mutant analysis and expression studies have identified genes expressed and required in the tail bud during secondary gastrulation (Roelink and Nusse,1991; Takada et al.,1994; Greco et al.,1996; Gofflot et al.,1997; Copp et al.,2003). Except in the most severely truncated embryos, tail bud gene expression (T, Wnt3a, Fgf8, Hoxb1) and proliferation appeared normal in Fgfr1^ΔFrs/ΔFrstail buds (Fig. 4F-I, data not shown). However, TUNEL analysis revealed an increase in cell death throughout caudal tissues of Fgfr1^ΔFrs/ΔFrs embryos(Fig. 4I). Cell death may hinder cell migrations or reduce the number of tail bud progenitors during secondary gastrulation.

The second PA is hypoplastic in Fgfr1^ΔFrs/ΔFrs embryos by E10.5, and histological analysis revealed a dramatic reduction in the number of mesenchymal cells populating this arch. This population normally includes both mesoderm and neural crest cells. Unlike cells within wild-type PAs,mesenchymal cells within mutant PA2 are neither proliferating nor closely apposed to the surrounding epithelia (Fig. 5A-C). We also observed a loss of Fgf8 expression in PA2,which has been shown to promote survival of PA mesenchyme(Fig. 5D,E)(Frank et al., 2002; Macatee et al., 2003). Although TUNEL analysis on sections did not reveal ectopic apoptosis of cells within Fgfr1^ΔFrs/ΔFrs arches(data not shown), the loss of Fgf8 signals may contribute to the lack of PA2 mesenchymal proliferation or affect migration of neural crest cells into the arches of mutant embryos. In a previous study of Fgfr1 hypomorphs,PA2 was found to be hypoplastic because of a neural crest migration defect(Trokovic et al., 2003a). Likewise, we observed impaired neural crest cell migration into Fgfr1^ΔFrs/ΔFrs PAs by Sox10 in situ hybridization (Fig. 5F-H) (Kuhlbrodt et al.,1998). Nile Blue staining revealed ectopic cell death along the path of neural crest migration (prior to PA entry; Fig. 5I-L).

At later stages, we noted some recovery of PA2 size and did not observe prominent craniofacial defects that would be expected if PA neural crest development failed entirely. We therefore analyzed the formation of PA crest derivatives to determine whether neural crest migration recovered later in development. Analysis of craniofacial skeletal elements at E15.5 indicated that PA2 NCC derivatives are selectively affected in Fgfr1^ΔFrs/ΔFrs embryos. We did not observe major hypoplasia or malformation of the jaws (PA1-derived;data not shown) or PA1 derivatives in the middle ear(Fig. 6A-D). By contrast,cartilages derived from PA2, including the stapes and styloid process of the middle ear and the lesser horns of the hyoid, were missing or hypoplastic in mutant embryos (Fig. 6).

Phenotypic data demonstrated that Fgfr1^ΔFrs is expressed and functional in vivo: we observed neither recapitulation of the null phenotype, as would be expected if the mutant receptor were non-functional,nor gain-of-function phenotypes indicative of ligand-independent activation. However, phenotypic data also indicated that Fgfr1 transduces Frs-independent signals in some developmental contexts. To investigate the nature of these signals, we derived mouse embryonic cells (MECs) from wild-type and mutant embryos, and used them in biochemical studies.

We first used immortalized MECs to verify that the mutant receptor signaled as expected based on previous reports. In these cells, we observed rapid and robust phosphorylation of Frs2 in wild type, but not Fgfr1^ΔFrs/ΔFrs, cells stimulated with Fgf (see Fig. S2A, part i in the supplementary material). We also observed an Fgf-stimulated decrease in Frs2 protein level in wild-type but not mutant cells, concomitant with Frs2 phosphorylation (see Fig. 2A, part ii in the supplementary material). In MAPK response assays, mutant immortalized MECs exhibited reduced sensitivity to Fgf and drove weaker signaling responses as assessed by phosphorylation of MAPK (see Fig. S2B in the supplementary material). These data recapitulate findings of previous studies performed in immortalized Frs2^-/- MECs(Hadari et al., 2001).

Cellular transformation may affect signal transduction and thus confound results of biochemical studies. Therefore, to approximate the in vivo situation, we re-examined Fgf responses in low passage (p2) non-immortalized MECs. Surprisingly, Fgfr1^ΔFrs/ΔFrs and wild-type p2 MECs responded with indistinguishable MAPK activation profiles, both in terms of response duration and Fgf dose sensitivity(Fig. 7A). We did not observe a notable PI3K pathway response, assayed by Akt phosphorylation, in cells of either genotype (Fig. 7A). In addition, neither Crk nor PLCγ were tyrosine phosphorylated (i.e. activated) in p2 MECs following Fgf stimulation(Fig. 7B, data not shown). Thus, these pathways are probably not mediating the observed MAPK response in Fgfr1^ΔFrs/ΔFrs cells. Furthermore, compared with wild-type cells, the Frs2 phosphorylation response was severely diminished in mutant cells despite comparable levels of total Frs2 (Fig. 7C). We expect that residual Frs activation in Fgfr1^ΔFrs/ΔFrs cells occurs indirectly, as it has previously been shown by yeast two hybrid that the amino acid 407-433 deletion completely abolishes the interaction of the Fgfr1 cytoplasmic domain with Frs2 (Xu et al.,1998).

We next investigated whether signaling through Fgfr2 could compensate for the lack of Fgfr1-Frs signaling by activating Frs2 (low level) and MAPK in Fgfr1^ΔFrs/ΔFrs MECs. Fgfr2 expression levels are not altered in Fgfr1^ΔFrs/ΔFrs (p2) MECs or embryos, as might be expected if this receptor was upregulated to compensate for reduced Fgfr1 signaling (Fig. 7D,E). Furthermore, we analyzed the activation state (tyrosine phosphorylation) of Fgfr2 in p2 MECs and found that although both wild-type and mutant MECs had basal levels of tyrosine phosphorylated Fgfr2, the amount of active Fgfr2 was not increased in cells of either genotype in response to Fgf. Surprisingly, the basal (-Fgf) level of activated Fgfr2 was elevated in Fgfr1^ΔFrs/ΔFrs MECs, and these cells responded to Fgf with a decrease in Fgfr2 phosphorylation(Fig. 7D). These data are not consistent with a role of Fgfr2 in compensatory activation of Frs and MAPK,and instead suggest that Fgfr1-Frs signaling transregulates Fgfr2 activity.

The MAPK signaling responses in p2 MECs did not directly translate into Fgf responsiveness in a proliferation assay: only two out of four mutant MEC cultures proliferated in response to Fgf, whereas all four responded with comparable MAPK and PI3K responses (Fig. 7A,F, data not shown). The growth curves did, however, reflect the relative phenotypic severity of embryos from which mutant cells were derived(data not shown). Variation in mutant MEC proliferative responses is probably due to differences in cell physiology or developmental staging between severely and mildly affected embryos. Nonetheless, the ability of two mutant lines to proliferate in response to Fgf indicates that Fgfr1-Frs signaling is not essential in directing this cellular response.

Previous reports have implicated the MAPK pathway as an effector of Fgf signaling in vivo (Corson et al.,2003). Our primary MEC data support this model and suggest that the Fgfr1-Frs signaling is not required for Fgf-induced MAPK activation. To determine the physiological relevance of the MEC results, we examined MAPK activation in Fgfr1-dependent developmental contexts by whole-mount immunohistochemistry. First, we analyzed stage-matched embryos at E8.5-9.5,when the caudal region of the embryo is still undergoing gastrulation and more rostral axial levels are undergoing somitogenesis and neurulation. We found that phospho-MAPK staining in gastrulating mesoderm and somites was not altered in mutant embryos (Fig. 8A-F). In addition, we observed a thin line of phospho-MAPK staining along the medial edges of the neural folds, where neuroepithelial fusion takes place. This domain of MAPK activation was present in both mutant and control embryos (arrows, Fig. 8C,F), despite the disruption of neural tube closure in Fgfr1^ΔFrs/ΔFrs embryos. Just after neurulation, when we observed an increase in cell death in mutant neural tubes, MAPK is not activated in wild-type or mutant neural tubes (data not shown). Thus, the cell survival role of Fgfr1-Frs signaling in the neural tube is probably facilitated by a different downstream pathway.

Later, during early PA2 development, phospho-MAPK immunostained dispersed cells throughout the pharyngeal arch region in both mutant and control embryos(Fig. 8G,J). As development progressed, phospho-MAPK became concentrated in control embryos at the rostral edge of PA1 and the caudal edge of PA2(Fig. 8H). In mutant embryos at this stage, although PA1 immunostaining appeared normal, intensification of phospho-MAPK staining was not observed in caudal PA2(Fig. 8K). This domain corresponds to the Fgf8 expression domain that is lost in mutants; however,the difference in MAPK activation observed in whole mount may be secondary to the failure of NCCs to populate this arch. Premigratory NCCs and NCCs within PA3 stained strongly with the phosphoMAPK antibody(Fig. 8I,L, data not shown).

Together, these results demonstrate that MAPK activity is not globally disrupted in Fgfr1-dependent contexts in Fgfr1^ΔFrs/ΔFrsembryos. Thus,either Fgfr1 does not contribute significantly to cellular phosphoMAPK levels in these contexts, or Fgfr1 signaling to MAPK is not affected by the Frs site deletion.

Our results demonstrate that Fgfr1-Frs signaling is essential during embryogenesis but is required only in a subset of Fgfr1-dependent contexts. Abrogation of the Fgfr1-Frs interaction does not disrupt Fgfr1 functions during gastrulation or somitogenesis, but does result in late embryonic lethality and defects in the development of the tail bud, neural tube, and pharyngeal arches. The specific array of phenotypes observed in Fgfr1^ΔFrs/ΔFrs embryos is distinct from that observed in Fgfr1^wtKI homo- or hemizygotes, previously described hypomorphs with reduced Fgfr1 levels, or embryos harboring an Fgfr1^PLCγ binding site mutation (Table 2, see Fig. S1 in the supplementary material)(Partanen et al., 1998; Xu et al., 1999; Trokovic et al., 2003a). This suggests that Fgfr1^ΔFrs/ΔFrsphenotypes are not general consequences of reduced Fgfr1 activity, but are indeed due to disrupted signaling through the Frs interaction site. Interestingly, activating mutations in Fgfr1 that cause human craniosynostosis syndromes also have context-specific effects restricted mainly to craniofacial and limb development(Passos-Bueno et al., 1999). Thus, developmental contexts differ in their requirements for specific signaling pathways downstream of Fgfr1, as well as for regulation of Fgfr1 activity.

Because Fgfr1^ΔFrs/ΔFrsembryos exhibited rescue of Fgfr1^null mesodermal phenotypes and early lethality, analysis of these signaling mutants enabled us to clarify later roles of Fgfr1 in neural tube and tail bud development. We thus identified a novel role of Fgfr1 in spinal neurulation. In contrast to craniorachischisis, which in Fgfr1^Δex4/Δex4 embryos is probably secondary to mesodermal defects, Fgfr1^ΔFrs/ΔFrs neural tube closure defects affect only the spinal region and are not accompanied by notochord or paraxial mesoderm defects. Our histological data demonstrate that Fgfr1-Frs signaling impacts the morphogenetic folding of the spinal neuroepithelium. Frs2 was previously implicated in convergent extension movements (Akagi et al., 2002),disruption of which can alter the size, shape and signaling properties of the notochord and neural plate, and thus result in neural tube closure defects. However, whereas previously described convergent extension mutants lack medial hinge points and exhibit notochord defects, Fgfr1^ΔFrs/ΔFrsembryos form ectopic DHLPs and do not exhibit defects in mesoderm or medial bending,(Copp et al., 2003). Our results are therefore more consistent with a role of Fgfr1-Frs signaling in modulating neuroepithelial responsiveness to DLHP-inducing or -repressing signals.

The neural tube closure defect in Fgfr1^ΔFrs/ΔFrs embryos may be exacerbated by the accompanying defect in tail bud development. Previously,posterior truncations in Fgfr1^null animals were postulated to be secondary to mesoderm defects. However, the high penetrance of posterior truncations, together with the low incidence of mesodermal defects in Fgfr1^ΔFrs/ΔFrs embryos suggest that Fgfr1 instead plays a direct, Frs-dependent role in tail bud development. The Frs site deletion did not affect patterning or proliferation in the tail bud during secondary gastrulation, but resulted in ectopic apoptosis throughout the caudal region of mutant embryos. Cell death could contribute to posterior truncations by depleting tail bud progenitors or hindering cell migration during secondary gastrulation.

In Fgfr1^ΔFrs/ΔFrsembryos, Sox10-positive neural crest cells are hindered in their migration into the first and second pharyngeal arches, though at later stages,craniofacial defects in these embryos are restricted predominantly to PA2 NCC derivatives. This implies that the Sox10-positive population migrating towards PA1 recovers or is not essential for the development of some PA1 NCC derivatives. Specific disruption of PA2 and/or PA2 NCC development in Fgfr1^ΔFrs/ΔFrs embryos contrasts with general Fgfr1 hypomorph phenotypes that affect NCC derivatives of both PA1 and PA2 (Trokovic et al., 2003a). Fgfr1-Frs signaling may be primarily required for PA2 patterning, or may participate in reciprocal signaling between the epithelium and NCCs that facilitates PA2 entry or survival of NCCs. In Fgfr1^ΔFrs/ΔFrs embryos, an epithelial Fgf8 expression domain is lost in PA2, and the NCC migration defect is accompanied by an increase in cell death along the migration pathway. Preliminary results of conditional mutagenesis experiments indicate that ectopic cell death in the pharyngeal region reflects a NCC-nonautonomous effect of the Frs site deletion (R.V.H. and P.S.,unpublished).

It has long been presumed that Fgfrs signal primarily through the MAPK pathway, and previous biochemical studies implicated Frs adaptors as important mediators of MAPK activation downstream of Fgfr1. However, we have demonstrated that the Fgfr1-Frs interaction is dispensable for early developmental roles of Fgfr1 and for activation of the MAPK pathway,both in primary embryonic cells and in vivo. Wild-type and Fgfr1^ΔFrs/ΔFrs primary cell responses to Fgf stimulation were indistinguishable in terms of both dose sensitivity and signaling kinetics, and MAPK phosphorylation appeared normal in several Fgfr1-dependent contexts in Fgfr1^ΔFrs/ΔFrs embryos. Furthermore, proliferative responses of non-immortalized Fgfr1^ΔFrs/ΔFrs cells suggest that this pathway is not essential in driving Fgf-induced proliferation,although cells from embryos with more severe phenotypes were growth impaired. The discrepancy between our biochemical data and those published previously probably reflects cell type-specific signaling mechanisms, effects of Frs2 disruption on receptors other than Fgfr1, or effects of cellular transformation on the ability of other pathways to signal downstream of Fgfr1.

Our work, as well as previous studies of Fgfr1-PLCγ signaling,demonstrates that individual signaling events downstream of Fgfr1 are required in a context-specific manner during development. In contexts where Fgfr1 functions are not compromised by the Frs site deletion, other pathways may drive cellular responses either independently of or additively with Frs adaptors. Aside from Frs2 and Frs3, only PLCγ, Crk, Grb14 and Sef have been shown to interact directly with activated Fgfr1, although Src, STAT and Shc are also activated downstream of Fgfr1 in some cell types(Mohammadi et al., 1991; Klint and Claesson-Welsh,1999; Larrson et al.,1999; Reilly et al.,2000; Tsang et al.,2002; Kovalenko et al.,2003). Sensitivity to the Frs site deletion could reflect the availability of alternate signaling pathways or the level of overall Fgfr1 signal required in a given context, rather than selective use of the Frs pathway. Thus, Fgfr1-Frs signaling may be used more broadly during development than is revealed by Fgfr1^ΔFrs/ΔFrsphenotypes.

Fgfr1 and Fgfr2 can form heterodimers in vitro(Bellot et al., 1991). If such heterodimers were to form in vivo, Fgfr1^ΔFrs signals could get shunted to the Frs pathway downstream of the Fgfr2 subunit in Fgfr1^ΔFrs/ΔFrs embryos. This predicts that heterodimers would rescue Fgfr1^ΔFrs function in contexts co-expressing Fgfr1 and Fgfr2, and that Fgfr2 would be activated by Fgf stimulation in primary cells. Our data do not support this model: Fgfr1^ΔFrs/ΔFrsembryos exhibit defects in the tail bud and pharyngeal arches, where Fgfr1 and Fgfr2 are co-expressed, but not during gastrulation or somitogenesis when the receptors have distinct expression patterns(Orr-Urtreger et al., 1991; Yamaguchi et al., 1992; Walshe and Mason, 2000). Furthermore, we did not observe compensatory upregulation of Fgfr2mRNA in vivo, or ligand-dependent activation of Fgfr2 in primary cells. Instead, basal levels of active Fgfr2 are notably elevated in Fgfr1^ΔFrs/ΔFrs cells,suggesting that Fgfr1-Frs signaling transregulates Fgfr2. This may coordinate Fgfr responses and/or serve a homeostatic role in vivo. Deregulation of Fgfr2 activity may contribute in a gain-of-function manner to Fgfr1^ΔFrs/ΔFrs phenotypes. Resolution of this issue will require further biochemical analysis and crosses between Fgfr1^ΔFrs and Fgfr2 mutant mouse lines.

We thank Janet Rossant for providing Fgfr1 genomic clones; Mitch Goldfarb for advice on the Frs site deletion; Philip Corrin, Jason Frazier and Marc Grenley for excellent technical assistance; and Raj Kapur, Henk Roelink and our laboratory colleagues for critical reading of the manuscript. This work was supported by an NSF predoctoral fellowship and the CMB Training Grant (NIH#GM07270) to R.V.H., and by grants HD 24875 and HD 25326 to P.S.