A novel regulatory mechanism for Fgf18 signaling involving cysteine-rich FGF receptor (Cfr) and delta-like protein (Dlk)

The fibroblast growth factor (FGF) family consists of 22 members in mice and humans, and has pleiotropic roles. Gene targeting studies have clearly demonstrated that some FGFs are indispensable to the development of many organs and tissues, including the limbs, lungs, liver, brain, testes, hair follicles and skeleton (Eswarakumar et al., 2005). FGFs transduce intracellular signals via FGF receptors (FGFRs) with a tyrosine kinase. As there are four genes encoding FGFRs in mice and humans, and Fgfr1, Fgfr2 and Fgfr3 each have two splice variants, the FGFR family consists of seven members. Each FGFR binds a subset of FGFs and transduces signals by forming an active receptor complex (Eswarakumar et al., 2005). Besides FGFRs, there are several FGF-binding molecules that also contribute to FGF signaling. Heparan sulfate proteoglycans (HSPGs) are necessary for the dimerization and activation of FGFRs by FGFs (Spivak-Kroizman et al., 1994; Yayon et al., 1991). Klothos are transmembrane proteins that modify the specificity of FGFRs by forming complexes with them and are indispensable for signaling by the FGFs with endocrine functions: i.e. Fgf19 (Fgf15 — Mouse Genome Informatics), Fgf21 and Fgf23 (Kurosu et al., 2007; Kurosu et al., 2006; Suzuki et al., 2008b; Urakawa et al., 2006). FGFR-like 1 (Fgfrl1), which shares sequence homology with FGFRs in its extracellular domain, but lacks an intracellular tyrosine kinase, is also involved in FGF signaling (Wiedemann and Trueb, 2000). Fgfrl1 seems to be a decoy receptor that negatively regulates FGF signaling (Trueb et al., 2003), and has indispensable roles in the development of the diaphragm (Baertschi et al., 2007). Thus, the modulation of FGF signaling by FGF-binding molecules plays important roles in normal developmental processes.

Cysteine-rich FGF receptor (Cfr; Glg1 — Mouse Genome Informatics) is another non-FGFR FGF-binding molecule, which was identified as a transmembrane molecule with affinity for Fgf1 and Fgf2 (Burrus and Olwin, 1989), and subsequently shown to bind Fgf3 and Fgf4 (Burrus et al., 1992; Kohl et al., 2000). Although Cfr binds FGFs via its large extracellular domain consisting of 16 repeats of an unique motif known as the Cfr repeat, it has no sequence homology with FGFRs and its intracellular domain consists of a short peptide of 13 amino acids without a kinase (Zhou et al., 1997). Thus, Cfr is a unique FGF-binding protein and may play a role in FGF signaling like the other FGF-binding proteins described above; however, its functions remain to be elucidated.

In this study, we identified Cfr as a delta-like protein (Dlk)-binding molecule. Dlk, also known as preadipocyte factor 1 (Pref-1), is a transmembrane protein with six epidermal growth factor (EGF) repeats in its extracellular domain. Dlk has sequence homology with Delta, a ligand for Notch, but lacks the Delta/Serrate/LAG-2 (DSL)-motif required for the activation of Notch. Dlk is abundantly expressed in various embryonic tissues (Smas and Sul, 1993) and we previously identified Dlk as a cell surface marker for hepatoblasts, embryonic hepatic progenitor cells (Tanimizu et al., 2003). Dlk-deficient mice show several developmental abnormalities such as perinatal death, growth retardation, skeletal abnormalities, increased amounts of adipose tissue and abnormal B cell development (Moon et al., 2002; Raghunandan et al., 2008), indicating that Dlk plays a fundamental role in development. Having identified Cfr as a Dlk-binding protein, we were interested in the functions of Cfr and the relation between Cfr and Dlk. To reveal the functions of Cfr, we generated Cfr-deficient mice and found that some of them die shortly after birth and show growth retardation, tail distortion and cleft palate. Among all the Fgf-deficient mice published, Fgf18-deficient mice were most similar to Cfr-deficient mice in terms of these phenotypes (Liu et al., 2002; Ohbayashi et al., 2002). Fgf18 activates intracellular signaling mainly via Fgfr3c (Zhang et al., 2006), and it is well established that Fgf18-Fgfr3c signaling plays a central role in skeletal development (Haque et al., 2007). Fgf18 inhibits the proliferation of chondrocytes (Liu et al., 2002; Ohbayashi et al., 2002), and several Fgfr3 mutations in humans are known to be a cause of dwarfism (Horton et al., 2007). In this paper, we show genetic and physical interaction between Cfr and Fgf18, and a positive regulatory role of Cfr in Fgf18 signaling. Moreover, we also noticed that the phenotypes of Cfr-deficient mice are similar to those of Dlk-transgenic mice (Lee et al., 2003) and found that Dlk inhibits the physical interaction between Cfr and Fgf18. Taken together, these results imply that Dlk interferes with the positive regulatory role of Cfr in Fgf18 signaling by abrogating the binding between Cfr and Fgf18. Thus, our study reveals a novel regulatory mechanism for Fgf18 signaling involving Cfr and Dlk.

Cfr-deficient mice were generated by using a gene-trapped 129Ola embryonic stem (ES) cell line (see Results). The cell line used for this research project, BayGenomics clone KST005 (catalog number 000716-UCD), was obtained from the Mutant Mouse Regional Resource Center (MMRRC), a NCRR-NIH funded strain repository, and was donated to the MMRRC by the NIH and NHLBI supported BayGenomics consortium. The mouse model will be made available (BRC No. RBRC03897) from the RIKEN BioResource Center (RIKEN BRC), which is participating in the National Bio-Resource Project of the MEXT, Japan, please contact them at animal@brc.riken.jp for inquiries, or search the catalog at http://www.brc.riken.go.jp/lab/animal/en/. The details of Fgf18 targeting were described previously (Ohbayashi et al., 2002). The Cfr-mutant and Fgf18-mutant mice used in this study had been back crossed with C57BL/6 wild-type mice at least eight and ten times, respectively. All experimental procedures in this study were approved by the institutional animal care and use committee of the University of Tokyo.

The antibodies used were as follows: anti-Actin (sc-1616) purchased from Santa Cruz Biotechnology, rat monoclonal antibody against Dlk raised in our laboratory (Suzuki et al., 2008a); anti-Cfr rabbit serum raised against the N-terminal domain of Cfr without a signal sequence (28-256 amino acid residues), which was expressed in Escherichia coli, and anti-Fgf18 rabbit serum raised against full-length Fgf18 without a signal sequence, which was fused to a GST tag and expressed in E. coli.

The extracellular domain of Dlk (1-303 amino acid residues) fused to the human IgG Fc region (Dlk-Fc) was expressed in COS7 cells and purified from the culture supernatant with a Hi Trap column (GE Healthcare), and then biotinylated with a biotinylation module (GE Healthcare) for flow cytometric analysis. First, we constructed a cDNA library of HPPL with the FastTrack 2.0 mRNA Isolation Kit (Invitrogen) and SuperScript Choice System (Invitrogen) in the pMXs retroviral vector. The plasmid was then converted to retrovirus using PLAT-E cells. Ba/F3 cells were infected with the virus library and Ba/F3 cells that bound Dlk-Fc were enriched by IMag (Becton Dickinson) cell sorting using biotinylated Dlk-Fc protein. Genomic PCR of the sorted cells revealed that some of them had the full-length Cfr sequence in their genomes (see Results).

We utilized a Ba/F3 cell line, which expresses a fusion protein of the extracellular domain of Fgfr3c and the intracellular domain of Fgfr1 described previously (Ornitz et al., 1996). For simplicity, we refer to this fusion protein as Fgfr3c in this report.

We constructed a pMXs-IRES-GFP vector encoding the full-length Cfr and produced retroviruses from the vector using PLAT-E cells as described previously (Miyaoka et al., 2006). Ba/F3 cells were infected with the viruses and cells expressing GFP were sorted by FACSVantage (Becton Dickinson).

For proliferation assay, Ba/F3 cells were deprived of cytokines for at least 6 hours, and then 2×10³ Ba/F3 cells were cultured in 100 μl RPMI-1640 (SIGMA) containing 10% fetal bovine serum (EQUITECH-BIO), 2 mM L-glutamine, 50 μg/ml gentamicin (Wako), and the indicated reagents per 96 well for 2 or 3 days. We assessed the extent of proliferation by adding 10 μl WST-1 reagent (Roche) per 96 well, incubating at 37°C for 2 hours, and monitoring absorbance at 450 nm subtracted by basal absorbance at 650 nm. We set eight wells per one condition in all the experiments in this report.

For cDNA synthesis, total RNA was isolated with TRIzol Reagent (Invitrogen), and then reverse transcribed with a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). For analysis, SYBR Premix Ex Taq (Takara) and Light Cycler (Roche) were used.

For immunohistochemistry, embryos were dissected and fixed with methanol/DMSO (4:1) at 4°C overnight, and then incubated in methanol/DMSO/H₂O₂ (4:1:1) at room temperature for 8 hours. Embryos were rehydrated and incubated with primary antibodies at 4°C overnight. They were then incubated with anti-rat IgG-HRP at 4°C overnight. Signal was visualized with 0.06% DAB/0.06% NiCl₂ in PBS with 0.2% BSA, 0.5% Triton X-100 and 0.03% H₂O₂.

For immunofluorescent staining, embryos were dissected and fixed with Zamboni's fixative at 4°C overnight, and then gradually substituted to 20% sucrose in PBS. The samples were sectioned and incubated with 5% skimmed milk in PBS for 2 hours at room temperature. They were then incubated with first antibodies at 4°C overnight, and with second antibody conjugated with Alexa Fluor 488 or 555 (Invitrogen) for 2 hours at room temperature. The sections were embedded with Gel/Mount (Cosmo Bio) containing Hoechst 33342.

For examining β-gal activity, embryos were dissected and fixed with 4% paraformaldehyde in PBS at room temperature for 10 minutes. After being washed with PBS, they were incubated in 1 mg/ml X-Gal, 5 mM ferricyanide, 5 mM ferrocyanide, 2 mM MgCl₂, 0.02% Nonidet-P 40 and 40 mM HEPES in PBS at 37°C for appropriate periods.

The detailed methods for western blot analysis and northern blot analysis were described previously (Miyaoka et al., 2006).

Mice were dissected, eviscerated, and fixed with 95% ethanol overnight. They were then incubated in acetone overnight. After a brief rinse with water, the samples were incubated in 20% acetate, 75% ethanol and 0.15 mg/ml Alcian Blue 8GX overnight. Then, the samples were washed with 70% ethanol for 8 hours, and cleared with 1% KOH overnight. For counterstaining, 0.05 mg/ml Alizarin Red in 1% KOH was used, and the samples were cleared in 1% KOH in 20% glycerol. All steps were done at room temperature.

The extracellular domain of Cfr (1-869 amino acid residues) fused tandemly to a His tag and a FLAG tag at its C-terminus (Cfr-EC) was expressed in COS7 cells and purified by using His Trap column (GE Healthcare). Protein G Sepharose (GE Healthcare) was used for precipitation of Fc-fused proteins. Cfr-EC was immunoprecipitated with anti-Cfr serum and protein G Sepharose, or anti-FLAG M2 agarose (SIGMA). These beads were incubated with culture supernatants or the lysis buffer (Miyaoka et al., 2006) containing their target proteins at 4°C for 4 hours, and then washed with the lysis buffer four times.

First, we attempted to identify molecules that bind to Dlk by using an expression screening method. To find cells expressing Dlk-binding proteins on their surface, the extracelluar domain of Dlk was fused to the Fc domain of human IgG (Dlk-Fc) and flow cytometry was used to find cells that bound Dlk-Fc. We found that Dlk-Fc strongly bound to HPPL, a hepatocyte progenitor cell line (Tanimizu et al., 2004) (Fig. 1A). To clone a cDNA encoding the Dlk-binding protein, a retroviral cDNA expression library was constructed from HPPL and introduced into Ba/F3 cells. Magnetic cell sorting was used to enrich Ba/F3 cells that bound Dlk-Fc. After three rounds of enrichment, we obtained several Ba/F3 clones that bound Dlk-Fc. Oligonucleotide primers specific for the retrovirus sequence were used to amplify the sequences introduced into the Ba/F3 genomes by the retrovirus. Sequencing of the amplified DNA fragments revealed that four of the eight clones contained in their genomes a full-length Cfr cDNA sequence (Fig. 1B). Other amplified DNA fragments seemed to be non-specific, because they were all truncated, not full-length fragments, and some of them were inserted in the reverse orientation to the retrovirus promoter (data not shown). As clones 1, 3 and 4, and 6 in Fig. 1B had different non-specific fragments inserted in the genomes, these clones were independently generated and acquired the affinity for Dlk-Fc by the Cfr insertion. We confirmed binding between Cfr and Dlk by immunoprecipitation. The extracellular domain of Cfr (Cfr-EC) was expressed in COS7 cells together with either Dlk-Fc or the extracellular domain of CD4 fused to the Fc domain of human IgG (CD4-Fc), and the Fc-fusion proteins were immunoprecipitated from the culture supernatants. As shown in Fig. 1C, Cfr-EC was co-precipitated with Dlk-Fc but not CD4-Fc. Then, we purified Cfr-EC, Dlk-Fc and CD4-Fc, and found that the purified Cfr-EC was also co-precipitated with the purified Dlk-Fc (Fig. 1D,E). These results indicate that Cfr directly interacts with Dlk. However, as the signal of the co-precipitated purified Cfr was relatively weak compared with that from the crude culture supernatants (Fig. 1C,E), it might be possible that the interaction is enhanced by other factors.

As there has been no report on the physiological role of Cfr, we generated Cfr-deficient mice by utilizing a gene-trapped ES cell line. Because this ES cell line has a β-geo cassette with a splice acceptor inserted in the first intron of the Cfr locus, normal splicing between the first and second exons was interrupted, resulting in the production of a protein in which the peptide from the first exon is fused with β-geo (Fig. 2A). The ES cells were introduced into C57BL/6 blastocysts to generate chimeric mice, and a mouse with ES-cell-derived cells in its germline was obtained. First, we confirmed the site of the β-geo cassette in the first intron of the Cfr locus by inverse PCR (data not shown), and located it, in the correct orientation, about 10 kbp 3′ downstream from the end of the first exon (Fig. 2A). The insertion was further confirmed to be correct by PCR with primers indicated in Fig. 2A (Fig. 2B). To confirm the abrogation of Cfr protein expression, lysates from embryonic day 11.5 (E11.5) whole embryos of each genotype were subjected to a western blot analysis with antiserum raised against the N-terminal domain of Cfr. An embryo homozygous for the gene-trapped allele showed a protein larger than the wild-type Cfr, the molecular weight of which corresponds to that of a fusion protein comprising the peptide from the first exon of Cfr and β-geo. Both the intact Cfr and the fusion protein were detected in a heterozygous embryo (Fig. 2B). Because no intact Cfr protein was detected in the embryo homozygous for the gene-trapped allele, and the first exon of Cfr contains only its signal peptide with a few additional amino acid residues (117 amino acid residues), it is highly likely that the gene-trap results in a null mutation. Therefore, we refer to the gene-trapped allele as ‘−’ hereafter. Mating between Cfr^+/− mice resulted in the generation of embryos with the normal Mendelian segregation pattern until E18.5; however, about 90% of Cfr^−/− mice died within 2 days after birth and there were only a few Cfr^−/− mice alive at 3 weeks of age (Table 1). Adult Cfr^−/− mice had no obvious abnormality in their appearance, but their size and weight were significantly smaller than those of wild-type mice (data not shown). As a preliminary histological analysis of two Cfr^−/− adult male survivors showed no obvious pathological abnormality in the brain, heart, lung, liver, spleen, kidney, stomach, guts, thymus, skin, skeletal muscle, eye or testis (data not shown), the cause of lethality is currently unknown. Because we identified Cfr as a Dlk-binding molecule, we considered the possibility that the lack of Cfr might affect the expression of Dlk. However, neither the mRNA nor protein level of Dlk was altered in Cfr-deficient mice (see Fig. S1A-D in the supplementary material). Because the inserted β-geo cassette expresses β-gal activity via the promoter of the inserted gene, the β-gal activity in Cfr^+/− embryos represents the Cfr promoter activity. At E7.5, Cfr was expressed only in the epiblast, and thereafter ubiquitously at all stages examined, E8.5-14.5 (see Fig. S2A in the supplementary material and data not shown). Northern blot analysis using total RNA from various adult tissues revealed that Cfr was expressed in all the tissues examined, although levels were variable (see Fig. S2B in the supplementary material). Thus, Cfr is expressed broadly in the body from early embryonic stages to adulthood.

At E18.5, Cfr^−/− embryos were smaller in size and weighed less than Cfr^+/+ and Cfr^+/− littermates (Fig. 3A,B). We also noticed that tails of all Cfr^−/− E18.5 embryos and neonates were distorted, suggesting an abnormal skeletal development. In fact, Alcian Blue/Alizarin Red staining of bones and cartilages revealed that tail skeletons of Cfr^−/− neonates were distorted, whereas the other skeletons were almost normal (Fig. 3C). The distortion seemed to be caused by an irregular arrangement of the tail cartilages (Fig. 3D). In addition, about 19% (7/37) of Cfr^−/− neonates had a bloated abdomen filled with air (Fig. 3E). As cleft palate in mice often results in an accumulation of air in the stomach and a bloated abdomen, we investigated the palates of Cfr^−/− neonates, and found that about 30% (11/37) of the mice exhibited cleft palate (Fig. 3F). All of the Cfr^−/− mice with a bloated abdomen had a cleft palate. We investigated craniofacial skeletal elements and found incomplete fusion of maxillary shelves in Cfr^−/− neonates with cleft palate (Fig. 3G). The other skeletal elements were apparently normal in Cfr^−/− mice (Fig. 3C,G and data not shown). Cfr^+/− mice showed no obvious abnormality. These results indicate that Cfr is essential for normal development.

Although Cfr is ubiquitously expressed in embryos, as shown in Fig. S2 in the supplementary material, the skeletal abnormality of Cfr^−/− mice prompted us to investigate more detailed expression patterns of Cfr in embryonic skeletons. We examined expression patterns of Cfr as well as Dlk in limbs, vertebrae and maxillae of E14.5 wild-type embryos by immunofluorescent staining and found that Cfr and Dlk were specifically expressed in prehypertrophic chondrocytes, and reserve and proliferating chondrocytes, respectively, in humeri (Fig. 4). The same expression patterns were observed in other limb skeletons (data not shown). Cfr and Dlk were also expressed in vertebrae and maxillae, although the specificity of cell-types or stages was unclear (see Fig. S3A,B in the supplementary material). Thus, Cfr and Dlk are expressed in developing skeletons, contributing to skeletal development.

Because Cfr is an FGF-binding molecule, we considered that the observed phenotypes of Cfr-deficient mice were caused by deregulation of some of the FGF signaling pathways. To address this possibility, we compared Cfr-deficient mice with all the Fgf-deficient mice reported previously, and noticed that the phenotypes of Cfr^−/− mice, such as perinatal death, growth retardation, tail distortion and cleft palate, are markedly similar to those of Fgf18-deficient mice. Therefore, we searched for a functional link between Cfr and Fgf18. First, we investigated whether there is any genetic interaction between Cfr and Fgf18 by mating Cfr^+/− mice with Fgf18^+/− mice. Among the offspring, wild-type, Cfr^+/− and Fgf18^+/− mice had no obvious phenotype. Cfr^+/−;Fgf18^+/− double heterozygous (D-HT) mice were viable and fertile (data not shown). They were comparable in body weight to other genotypes and had a normal tail morphology at birth (see Fig. S4A,B in the supplementary material). However, D-HT mice were significantly smaller than wild-type mice at 3 months after birth, whereas neither Cfr^+/− nor Fgf18^+/− mice showed such a phenotype (Fig. 5A). The phenotype was more evident in D-HT male mice, as they were already smaller than wild-type male mice at 1 month after birth (Fig. 5A). Moreover, the tails of D-HT mice were distorted by 7 days after birth like those of Cfr-deficient mice and Fgf18-deficient mice, but neither Cfr^+/− nor Fgf18^+/− mice showed tail distortion (Fig. 5B,C). The tail distortion in D-HT mice was visible from postnatal day 3 and retained even in adulthood. About 83% (19/23) of adult D-HT mice showed the malformation. These phenotypes clearly indicate genetic interaction between Cfr and Fgf18, suggesting functional cooperation.

We further analyzed the genetic interaction between Cfr and Fgf18 by crossing D-HT mice. As was the case for the offspring from mating of Cfr^+/− mice (Table 1), some Cfr^−/− neonates from mating of D-HT mice died after birth. Mortality among Cfr^−/−;Fgf18^+/− mice was similar to that among Cfr^−/− mice, indicating that the lack of one Fgf18 allele did not affect the viability of Cfr^−/− mice (data not shown). Likewise, skeletons of Cfr^−/− and Cfr^−/−;Fgf18^+/− neonates showed no significant difference (Fig. 5D). Consistent with the finding that the complete loss of Fgf18 resulted in perinatal death and skeletal abnormalities (Liu et al., 2002; Ohbayashi et al., 2002), Cfr^+/−;Fgf18^−/−, Cfr^−/−;Fgf18^−/− as well as Fgf18^−/− mice died at birth. Fgf18^−/− neonates showed slightly distorted ribs and the additional loss of one Cfr allele (Cfr^+/−;Fgf18^−/−) enhanced the distortion of the ribs and caused curvature of the spine. In Cfr^−/−;Fgf18^−/− double knockout neonates, the distortion of the ribs and curvature of the spine were much more severe than in Cfr^+/−;Fgf18^−/− mice, although the number of ribs was normal and there was no fusion of the ribs (Fig. 5D). These results clearly indicate that Cfr and Fgf18 play indispensable and overlapping roles in skeletal development.

We examined the possibility that the genetic interaction of Cfr with Fgf18 resulted from their physical interaction. To address this possibility, we prepared recombinant Fgf18 protein with a GST tag (GST-Fgf18), and used the GST tag alone as a control (Fig. 7A). We also prepared culture supernatant of COS7 cells transfected with an expression vector for Cfr-EC. The purified GST-Fgf18 or GST tag was added to the supernatant, and Cfr-EC was immunoprecipitated. Western blot analysis revealed that GST-Fgf18 was co-immunoprecipitated with Cfr-EC, whereas the GST tag was not (Fig. 7B). These results clearly indicate that Fgf18 binds to Cfr and suggest that Cfr positively regulates Fgf18 signaling through this physical interaction.

To further confirm this possibility, we utilized a Ba/F3 cell line expressing Fgfr3c, which proliferates depending on Fgf18 (Ornitz et al., 1996). We chose Fgfr3c because it has the highest affinity for Fgf18 (Zhang et al., 2006) and is expressed in prehypertrophic chondrocytes in developing limb skeletons, where Cfr is also expressed (Ohbayashi et al., 2002) (Fig. 4). We infected these cells with viruses containing Cfr cDNA or without cDNA as a control, and assessed their responses to Fgf18 by WST-1 assay. Expression of Cfr enhanced the responses of Ba/F3 cells with Fgfr3c to Fgf18 (Fig. 6A), indicating that Cfr positively regulates Fgf18-Fgfr3c signaling. By contrast, Ba/F3 cells expressing Cfr alone failed to respond to Fgf18, whereas they proliferated in response to IL-3 (Fig. 6B). Based on these results, we concluded that Cfr enhances Fgf18 signaling via Fgfr3c but does not directly activate the intracellular signaling. As it is well known that heparin or HSPG is required for FGF signaling via FGFRs (Spivak-Kroizman et al., 1994; Yayon et al., 1991), we examined the effect of heparin on the Cfr function. We monitored responses of Ba/F3 cells expressing both Fgfr3c and Cfr to Fgf18 in the presence or absence of heparin and found that they failed to respond to Fgf18 at all in the absence of heparin, demonstrating that heparin is required for the function of Cfr (Fig. 6C).

The results described above demonstrate genetic, physical and functional interaction between Cfr and Fgf18. However, it still remained unknown how Dlk is involved in the interaction between Cfr and Fgf18.

As Cfr physically interacts with both Fgf18 and Dlk, we examined whether the binding of Dlk to Cfr affects the binding between Cfr and Fgf18. For this purpose, Cfr-EC was expressed in COS7 cells together with Dlk-Fc or CD4-Fc as a control. GST-Fgf18 fusion protein was then added to the culture supernatants and Cfr-EC was immunoprecipitated. Western blot analysis of the immunoprecipitates revealed that co-immunoprecipitation of Fgf18 with Cfr-EC was severely blocked in the presence of Dlk-Fc, but not CD4-Fc (Fig. 7C). These results indicate that Dlk interferes with the binding of Fgf18 to Cfr, resulting in interruption of the positive regulatory role of Cfr in Fgf18 signaling.

Cfr is a unique FGF-binding protein that binds FGFs via its long extracellular domain, but has a very short intracellular peptide of 13 amino acid residues without any known motifs for signaling. Therefore, it seems unlikely that Cfr directly activates the intracellular signaling pathways, which is in fact supported by the results using Ba/F3 cells expressing Cfr (Fig. 6B). Although Cfr seems to have no paralog in either mice or humans, it is evolutionally conserved and present even in Caenorhabditis elegans (data not shown), suggesting that Cfr is a fundamental component of the FGF signaling pathway. However, its physiological role remained unknown.

In this study we generated a Cfr-deficient mouse line and found that these mice exhibit perinatal lethality, growth retardation, distorted tails and cleft palates (Fig. 3). Another Cfr-deficient mouse line generated independently from this study shows the same phenotypes (H. Okae and Y. Iwakura, personal communication), supporting our results. Interestingly, among various Fgf-deficient mice reported so far, Fgf18-deficient mice exhibit phenotypes very similar to those of Cfr-deficient mice, such as perinatal death, growth retardation, tail distortion and cleft palate. Given this striking similarity, we considered that there is a functional link between Cfr and Fgf18. To address the genetic interaction of Cfr with Fgf18, double heterozygous mice were generated by mating Cfr^+/− mice with Fgf18^+/− mice. Although neither Cfr^+/− nor Fgf18^+/− mice showed any phenotype, the double heterozygous mice exhibited tail distortion after birth, suggesting genetic interaction (Fig. 5B,C). The phenotype is also strikingly similar to that of Fgfr3-deficient mice, which also exhibit tail distortion after birth (Deng et al., 1996). Because it has been well established that Fgf18 transmits its signal via Fgfr3c (Davidson et al., 2005), these observations further support the idea that Cfr is positively involved in Fgf18 signaling. Furthermore, co-immunoprecipitation assays clearly demonstrated that Cfr binds Fgf18 (Fig. 7B), and Ba/F3 cells expressing Fgfr3c with or without Cfr provided direct evidence for the positive regulatory role of Cfr in the Fgf18-Fgfr3c signaling pathway (Fig. 6A). This is the first report on the physiological functions of Cfr and the first evidence for the existence of a novel regulatory component of the Fgf18 signaling pathway. Fgfr3c is known to be expressed in proliferating and prehypertrophic chondrocytes to control their proliferation. Co-expression of Cfr with Fgfr3c in prehypertrophic chondrocytes supports the idea that Cfr plays a role in Fgfr3c signaling in those cells (Fig. 4). Because it has been well established that deregulation of the Fgf18-Fgfr3c signaling pathway leads to achondroplasia (Horton et al., 2007), it would be intriguing to investigate whether Cfr is also involved in human skeletal diseases.

Because Cfr has been shown to bind Fgf1, Fgf2, Fgf3 and Fgf4 (Burrus et al., 1992; Kohl et al., 2000), it may play a role in their signaling as well. In addition, the deletion of one Cfr allele in Fgf18-deficient mice augmented the skeletal abnormality of Fgf18-deficient mice and the deletion of both Cfr alleles in Fgf18-deficient mice resulted in a much more severe phenotype (Fig. 5D), suggesting that Cfr is involved in not only Fgf18 signaling but also other signaling pathways. However, as Fgf4-deficient mice die very early in their embryonic development (Feldman et al., 1995), and their phenotype is totally different from that of Cfr-deficient mice, Fgf4 is unlikely to be a physiological ligand for Cfr. By contrast, Fgf3-deficient mice show growth retardation and distorted tails, and some of them die perinatally similarly to Cfr-deficient mice, although their phenotypes are different from those of Cfr-deficient mice, e.g. inner ear defects in Fgf3-deficient mice (Mansour et al., 1993). Therefore, it is possible that Cfr contributes to Fgf3 signaling. Because Fgf1-deficient and Fgf2-deficient mice show rather mild phenotypes or no phenotype (Dono et al., 1998; Miller et al., 2000), it is hard to speculate about any possible involvement of Cfr in their signaling based on phenotypes. While there is currently no information as to the binding of Cfr to other FGFs, Cfr may contribute to signaling by some of them. In particular it would be interesting to investigate whether Cfr contributes to signaling by the FGFs that play a role in skeletal development, such as Fgf9 (Hung et al., 2007).

We identified Cfr as a Dlk-binding molecule (Fig. 1). Dlk is well known as preadipocyte factor 1 (Pref-1), an inhibitor of adipogenesis (Smas and Sul, 1993). In addition to preadipocytes, Dlk has also been reported to affect in vitro differentiation of various cell types such as osteoblasts (Abdallah et al., 2004), hematopoietic progenitors (Moore et al., 1997), thymocytes (Kaneta et al., 2000) and B-cell progenitors (Bauer et al., 1998). However, the mechanism by which Dlk functions still remains largely unknown. To reveal its functions, attempts were made to identify Dlk-binding molecules with the yeast two-hybrid system, resulting in several candidates such as growth arrest specific gene 1 (Baladron et al., 2002; Nueda et al., 2008). Because the extracellular domain of Dlk alone exerts its biological activities (Mei et al., 2002; Ohno et al., 2001; Smas et al., 1997), we used the extracellular domain of Dlk to search for its binding partners displayed on the cell surface and identified Cfr.

Because Dlk binds to Cfr, we investigated whether Dlk has any effect on the binding between Cfr and Fgf18 and revealed that Dlk blocks the binding (Fig. 7C). If the physical interaction of Cfr with Fgf18 is necessary for Cfr to positively regulate Fgf18 signaling, Dlk is likely to abrogate the positive role of Cfr in the signaling. We noticed that the phenotypes of Cfr-deficient mice are similar to those of Dlk-transgenic mice: e.g. both mice show growth retardation throughout their life and tail distortion (Lee et al., 2003). As loss of Cfr and forced expression of Dlk resulted in similar phenotypes, it is likely that Cfr and Dlk function in opposite ways. This is consistent with the result that Dlk inhibits the physical interaction between Cfr and Fgf18. Based on these results, we propose a model for the roles of Cfr and Dlk in Fgf18 signaling (Fig. 7D): i.e. in the absence of Dlk, binding of Cfr to Fgf18 positively regulates the signaling, and Dlk blocks the interaction between Cfr and Fgf18 by binding to Cfr, interrupting the positive regulatory role of Cfr in Fgf18 signaling. If this model is correct, the levels and timing of the expression of Cfr and Dlk would modulate Fgf18 signaling, fulfilling normal developmental processes. Recently, it was reported that Dlk regulates mesenchymal cell fate, such as chondrocyte, osteoblast and adipocyte lineages, through Sox9 (Wang and Sul, 2009). It is well known that Fgf18 also regulates chondrocyte and osteoblast development. Therefore, it is tempting to speculate that Fgf18 signaling is involved in the mechanism of cell-fate determination.

Recent studies have revealed that the FGF signaling via FGFRs is finely tuned by FGF-binding molecules other than FGFRs, such as HSPGs, Klothos and Fgfrl1 (Kurosu et al., 2007; Kurosu et al., 2006; Spivak-Kroizman et al., 1994; Suzuki et al., 2008b; Trueb et al., 2003; Urakawa et al., 2006). These molecules significantly contribute to normal developmental processes mediated by the FGF signaling pathways. This study has revealed, for the first time, in vivo roles of another FGF-binding molecule, Cfr, and unexpected links among Dlk, Cfr and FGF signaling, and provides evidence for the presence of a novel regulatory mechanism of the FGF signaling pathway.

We thank Drs T. Itoh, N. Tanimizu and E. Saijou, and Mr S. Saito for helpful discussions and technical assistance. We thank Ms K. Suzuki for preparing anti-Dlk antibody. We are also grateful to Drs A. Nakano, H. Okae, Y. Iwakura (the University of Tokyo), N. Mizushima, Y. Ogawa (Tokyo Medical and Dental University), J. Kanno, S. Kitajima, Y. Takahashi (National Institute of Health Science), N. Ohbayashi (Tohoku University) and T. Okubo (Okazaki Institute for Integrative Biosciences) for helpful advice and support. We thank Drs D. M. Ornitz (Washington University) and N. Itoh (Kyoto University) for providing us with cDNAs of FGFRs and Fgf18, respectively. Y.M. is a recipient of a JSPS Research Fellowship for Young Scientists.