Conditional inactivation of FGF receptor 2 reveals an essential role for FGF signaling in the regulation of osteoblast function and bone growth

Mutations in fibroblast growth factor receptor (FGFR) 2 are responsible for several clinically distinct craniosynostosis syndromes in humans(Naski and Ornitz, 1998; Ornitz and Marie, 2002; Wilkie, 1997). These syndromes share a common feature; premature fusion of at least one of the cranial sutures. However, craniosynostosis syndromes also have distinct facial features and some have characteristic limb and joint abnormalities and central nervous system dysfunction.

Crouzon syndrome affects suture development and is not associated with limb abnormalities, whereas Pfeiffer syndrome and Jackson-Weiss syndrome are characterized by both craniosynostosis and broad medially displaced toes(Jabs et al., 1994; Rutland et al., 1995). In contrast, Apert syndrome patients develop coronal synostosis, severe bony and soft tissue syndactyly with associated joint fusion, and often have mental retardation (Cohen, 2000). The unique features of craniosynostosis syndromes probably reflect differences in the mechanism by which FGFR activity is altered by specific missense mutations(Wilkie, 1997).

Four FGFR tyrosine kinases bind with varying affinity and specificity to a family of 22 FGF ligands (Ornitz and Itoh,2001). Ligand binding specificity is regulated by specific sequences in the extracellular region and by the alternative splicing of exons encoding the carboxyl-terminal half of immunoglobulin (Ig) domain III. Alternative splicing of Fgfr2 is tissue specific, resulting in epithelial variants (b splice forms) and mesenchymal variants (c splice forms)(Miki et al., 1992; Naski and Ornitz, 1998; Orr-Urtreger et al., 1993). Ligand binding studies demonstrate that mesenchymally expressed ligands such as FGF7 and 10, activate FGFR2b, whereas FGF2, 4, 6, 8 and 9 activate FGFR2c(Ornitz and Itoh, 2001; Ornitz et al., 1996).

The majority of missense mutations in FGFR2 result in some ligand-independent dimerization, phosphorylation and constitutive receptor signaling. Interestingly, a single missense mutation which causes Apert syndrome (P252R or S253W in FGFR2) abolishes its ability to discriminate between epithelially and mesenchymally expressed FGF ligands, providing an explanation for the increased severity of Apert syndrome compared to other craniosynostosis syndromes resulting from mutations in FGFR2(Anderson et al., 1998; Yu et al., 2000; Yu and Ornitz, 2001).

Gain-of-function mutations in FGFR3 result in three related dwarfing chondrodysplasia syndromes; hypochondroplasia, achondroplasia (ACH) and thanatophoric dysplasia (TD) (Naski and Ornitz, 1998; Ornitz and Marie, 2002). Fgfr3 is expressed in the cartilage of the developing embryo, prior to formation of ossification centers. In the epiphyseal growth plate, Fgfr3 is expressed in proliferating and prehypertrophic chondrocytes. Mouse models for ACH and TD demonstrate that activation of FGFR3 inhibits chondrocyte proliferation and differentiation(Chen et al., 1999; Chen et al., 2001; Iwata et al., 2001; Li et al., 1999; Naski et al., 1998; Segev et al., 2000; Wang et al., 1999). In contrast, mice lacking Fgfr3 exhibit skeletal overgrowth(Colvin et al., 1996). Together, these data establish Fgfr3 as a negative regulator of endochondral bone growth.

Mice homozygous for null alleles of Fgfr2 die at embryonic day 10.5 (E10.5) with multiple defects in organogenesis, including the absence of limb buds (Xu et al., 1998). Specific deletion of the b exon of Fgfr2 results in a neonatal lethal phenotype, also lacking limb buds (De Moerlooze et al., 2000). This phenotype is similar to that of mice lacking Fgf10 (Min et al.,1998; Sekine et al.,1999). At E10.5, skeletal development has not progressed beyond the condensation stage. However, Fgfr2 is expressed at high levels in condensed mesenchyme that will give rise to cartilage and bone and later in the perichondrial and periosteal tissues that give rise to osteoblasts(Ornitz and Marie, 2002). To evaluate the function of Fgfr2 in skeletal development, we used the conditional gene silencing approach to specifically disrupt Fgfr2signaling in the chondrocyte and osteoblast lineages. We show that Fgfr2 is not required for osteoblast differentiation but is essential for osteoblast proliferation and for the maintenance of osteoblast anabolic function.

The Dermo1 targeting vector(Fig. 1A) was constructed by inserting a 5′ 7 kb EcoRI-NotI genomic fragment and a 3′ 1.5 kb BspEI-HindIII genomic fragment, which flanks exon 1 of Dermo1, into EcoRI-EcoRV, and EcoRV-HindIII sites of pBS-KS, respectively. A crecDNA sequence (Lewandoski et al.,1997b) was cloned into a SalI-KpnI site at the 3′ end of the 5′ arm to generate pBS-7-cre. A FRT2-neo selectable marker, was constructed by inserting a PGK/neo cassette(Tybulewicz et al., 1991) into the SnaBI site of pFRT2 (Dymecki,1996). The FRT-neo fragment was then excised with EcoRI and cloned into an EcoRI site at the 5′ end of the 3′ arm to generate pBS-1.5-neo. The targeting construct was assembled by cloning the 5′ 7-cre fragment into a 5′ NotI site in pBS-1.5-neo. Following linearization, RW4 ES cells were electroporated with the targeting vector and selected with G418. Antibiotic resistant clones were screened by DNA blot analysis of XhoI-digested and EcoRV-digested genomic DNA using external 5′ and 3′ probes, respectively(Fig. 1A). Primer D1(5′-AACTTCCTCTCCCGGAGACC) and D2 (5′-CCGGTTATTCAACTTGCACC), which were localized in the intron sequence 5′ of exon 1 of Dermo1and in 5′ end of the cre sequence, respectively, were used for PCR genotyping of Dermo1-cre (370 bp product)(Fig. 1A, Fig. 2D).

The 34-bp loxP sequence was introduced into Fgfr2 genomic sequence with two complementary oligonucleotides, loxP1 5′-CTAGAGATATCATAACTTCGTATAGCATACATTATACGAAGTTATTand loxP2 5′-CTAGAATAACTTCGTATAATGTATGCTATACGAAGTTATGATATCT,which were annealed to form a loxP linker with overhanging XbaI ends(underlined). An EcoRV site (bold) was placed at the 5′ end of the loxP sequence for further cloning and DNA blot genotyping. A 5.7 kb EcoRI-HindIII genomic fragment, which contained exon 7(IIIa), exon 8 (IIIb) and exon 9 (IIIc) of Fgfr2, was used to generate the 5′-half of the targeting vector by replacing a BamHI site in intron 7 with a SpeI site and inserting the loxP linker. A 4.4 kb HindIII genomic fragment, which contained exon 10 (encoding the transmembrane (TM) domain), was used to generate the 3′-half of the targeting vector by inserting the loxP linker into a XbaI site in intron 10 and the FRT-neo cassette into the EcoRV site of the loxP linker. RW4 ES cells were transfected and screened by DNA blot analysis of EcoRV/HindIII-digested and EcoRV-digested DNA using 5′ and 3′-probes, respectively(Fig. 2A).

To generate the Fgfr2^Δ allele, Fgfr2^+/flox heterozygous mice were mated withβ -actin-cre mice (Lewandoski et al., 1997a) and F₁ animals were screened for the presence of Fgfr2^flox and β-actin-crealleles. The double heterozygous mice were then mated to wild-type mice to obtain a germline-transmitted Fgfr2^Δ allele.

Primer F1 (5′-ATAGGAGCAACAGGCGG), F2 (5′- TGCAAGAGGCGACCAGTCAG)and F3 (5′- CATAGCACAGGCCAGGTTG) were used for PCR genotyping of Fgfr2 alleles. F1 and F2 produced a 142 bp and a 207 bp fragment from wild type and Fgfr2^flox alleles, respectively, and F1 and F3 produced a 471 bp fragment from the Fgfr2^Δ allele(Fig. 2D).

Skeletons were prepared as described previously(Colvin et al., 1996). Radiography studies were carried out using a Specimen Radiography System(Faxitron X-ray Corporation, IL) X-ray source. Bone mineral density was determined using dual energy X-ray absorptiometry (DEXA) with a PIXImus Mouse Densitometer (GE Lunar Medical Systems).

For whole-mount staining, E11.5 embryos were fixed in 4% paraformaldehyde in PBS for 60 minutes, washed in PBS and stained in β-gal staining buffer(5 mM K₃Fe(CN)₆, 5 mM K₄FE(CN)₆·3H₂O, 1 mM MgCl₂,0.01% sodium desoxycholate, 0.009% NP40, 0.002% X-gal) for 8 hours, at 4°C. After post-fixing in 4% paraformaldehyde, the limb buds were cut off,embedded in paraffin wax and sectioned (4 μm). E15.5 embryos, immediately after dissection were embedded in OCT at -20°C and sectioned (8 μm). Frozen sections were dried at room temperature for 20 minutes, fixed in 4%paraformaldehyde for 10 minutes and stained for 2 hours at 37°C. At postnatal ages, the bones of the mice were dissected, fixed in 4%paraformaldehyde in PBS, decalcified in 14% EDTA for 10 days at 4°C and embedded in OCT at -20°C. 8 μm frozen sections were stained for 2 hours at 37°C. All sections were counter stained with Nuclear Fast Red.

Tissues were fixed in 4% paraformaldehyde in PBS, decalcified in 14% EDTA and embedded in paraffin wax. Sections were stained with Hematoxylin and Eosin(H&E), von Kossa staining (for minerized bone tissues) or tartrate resistant alkaline phosphatase (TRAP) staining (for osteoclast activity). Computer imaging using AxioVision 3.0 software (Zeiss) was used for histomorphometric analysis. The bone volume and tissue volume were measured on the trabecular bone region of the proximal tibia, including both the primary and secondary spongiosa. The length of proliferating and hypertrophic chondrocyte zones and the length of the metaphysis was measured along the midline of the proximal tibia growth plate.

Osteoblasts and osteoclasts in the primary spongiosa were counted in a region extending 100 μm from the chondro-osseous junction and including the entire width of the metaphysis. The total areas counted were measured using AxioVision 3.0 image software (Zeiss).

The mineral apposition rate (MAR) was determined by calcein double labeling. Mice were injected with calcein (20 mg/kg) 13 days and 4 days prior to sacrifice. Following dissection, long bones were fixed in 70% ethanol and embedded in polymethyl methacrylate resin for sectioning. Coronal sections of comparable anatomic position were examined by fluorecence microscopy. The mean distance between the calcein double labels was measured using AxioVision 3.0 image software (Zeiss).

Anti-BrdU immunohistochemistry was carried out as previously described(Naski et al., 1998). Proliferating osteoblasts were counted as described above for total osteoblasts. The percentage of BrdU-positive nuclei versus total nuclei was calculated as the proliferation index.

Total RNA was prepared from E10.5 whole embryos or from legs of E16.5 embryos using the RNeasy kit (Qiagen) according to the manufacturer's instructions. At E10.5, the amniotic membrane was removed and Fgfr2^Δ/Δembryos were identified by lack of limb buds. At E16.5, legs were dissected from the whole embryos and the skin were carefully removed. First strand cDNA synthesis and subsequent PCR amplification were carried out as described previously (McEwen and Ornitz,1997). The forward primer 5′-CAAAGGCAACTACACCTGCC and the reverse primer 5′-CAGCCATGACTACTTGCCCG used in RT-PCR analysis were located in exon 6 and 12 of Fgfr2, respectively. qRT-PCR was performed on an Applied Biosystems Gene Amp 5700 Sequence Detection system using Sybr Green fluorescent dye binding to PCR products(Bustin, 2000). Osteocalcin gene expression was quantified as previously described(Willis et al., 2002).

In situ hybridization was performed as described previously(Liu et al., 2002). The Fgfr2 IIIcTM domain probe, which is 348 bp long and includes exon 9(IIIc) and 10 (TM) sequences, was generated by PCR amplification with forward primer 5′-CCGCCGGTGTTAACACCAC and reverse primer 5′-TGTTACCTGTCTCCGCAG and cloned into the pGEM-T EASY vector (Promaga). Plasmids used for generating ³³P-labeled riboprobes were: Fgfr1 (Peters et al.,1992), Fgfr2-TK (De Moerlooze et al., 2000), collagen type I(Metsaranta et al., 1991); osteocalcin (provided by K. Nakashima); osteopontin(provided by K. Lee) and Cbfa1 (Runx2; provided by K. Nakashima).

Dermo1 is a basic helix-loop-helix transcription factor that is highly expressed in mesodermal tissues during embryogenesis. During skeletal development, Dermo1 is expressed in condensed mesenchyme and later in perichondrial and periosteal cells surrounding cartilage(Li et al., 1995). Because the transcriptional regulatory elements of Dermo1 have not been characterized, homologous recombination was used to replace the first exon of Dermo1 with cre recombinase(Fig. 1A). One hundred and thirty ES cell clones were screened and two homologous recombinants were identified (Fig. 1B). Both clones were injected into blastocysts and one germline chimera was obtained(Fig. 1C). Mice heterozygous for the targeted allele (Dermo1^cre/+) were phenotypically normal.

To test in vivo function and tissue specificity of Dermo1-CRE, Dermo1^cre/+ mice were mated to Rosa26 reporter mice (R26R) in which the original β-galactosidase (β-gal)insertion into the Rosa26 locus was disrupted by insertion of a loxP-neo cassette (Soriano,1999). The lacZ gene can be reactivated through CRE-mediated excision of the loxP-neo sequence(Soriano, 1999). In double heterozygous (Dermo1^cre/+; R26R/+) embryos or miceβ-gal activity was detected at the surface of the embryo as early as E9.5 and in mesodermal tissues such as branchial arches and somites(Fig. 1D). Little β-gal activity was detected in neural and ectodermal tissues(Fig. 1D,E). During endochondral ossification, β-gal activity was first detected in condensed mesenchyme from which both chondrocytes and osteoblasts are derived(Fig. 1F). Later in developmentβ-gal activity was detected in chondrocytes in growth plate cartilage and in osteoblasts in the perichondrium, periosteum and endosteum(Fig. 1G,I,J,L). Occasionally a clone of chondrocytes was observed that failed to express β-gal activity(Fig. 1G,I). Interestingly,bone marrow cells and osteoclasts were negative for β-gal activity(Fig. 1L). During intramembranous ossification in developing sutures, β-gal activity was present at the osteogenic fronts and in surrounding mesenchymal tissues (data not shown).

In situ hybridization studies were used to compare the pattern of Fgfr2 expression with that of Dermo1-CRE activity. Consistent with published data (Delezoide et al., 1998; Orr-Urtreger et al., 1991; Peters et al.,1992), Fgfr2 was expressed throughout all phases of skeletal developmental. Initially, Fgfr2 was expressed in condensing mesenchyme (Delezoide et al.,1998; Orr-Urtreger et al.,1991; Peters et al.,1992). Later in development, Fgfr2 was predominantly localized to perichondrial and periosteal tissue and weakly to endosteal tissue and trabecular bone (Fig. 1H,K). Fgfr2 was also present in periarticular chondrocytes but was not detected in growth plate chondrocytes(Fig. 3D,E). In developing sutures, Fgfr2 expression was found in osteoblasts in the osteogenic fronts (data not shown). The prominent expression of Fgfr2 in osteoblast lineages suggested an essential function for FGFR2 during skeletal development. Because β-gal activity patterns indicated that osteoblasts were lineage descendants of cells that expressed Dermo1-cre, the function of FGFR2 during osteoblast development and bone formation can be studied by a Dermo1-CRE-mediated conditional gene silencing approach.

To generate a floxed (flanked by loxP sites) allele of Fgfr2 that could be inactivated by CRE-mediated recombination, a conditional targeting vector was constructed (Fig. 2A). A 5′ loxP site was inserted in the intron between exon 7 (IIIa) and 8 (IIIb), and a 3′ loxP site was inserted in the intron downstream of exon 10 (TM). An FRT-neo cassette was cloned just upstream of the 3′ loxP site. This design flanked Fgfr2 sequences encoding the alternatively spliced Ig domain IIIb, IIIc and TM domain encoding exons with loxP sites. Of 343 ES cell clones screened for homologous recombination,28 3′ end homologous recombinants were identified(Fig. 2B). On screening for the presence of the distal 5′ loxP site, 4 of 28 clones were correctly targeted and also contained the 5′ loxP site(Fig. 2C). These 4 clones were injected into blastocysts and two transmitted the targeted allele into the germline.

Mice heterozygous for the Fgfr2-flox allele(Fgfr2^+/flox) were bred to β-actin-cretransgenic mice (Lewandoski et al.,1997a) to create a germline null allele of Fgfr2(Fgfr2^Δ) (Fig. 2A). The consequence of different allelic combinations of wild type Fgfr2(Fgfr2⁺), Fgfr2^flox and Fgfr2^Δ alleles showed that all allelic combinations(Fgfr2^+/flox, Fgfr2^flox/flox, Fgfr2^+/Δ or Fgfr2^flox/Δ) were phenotypically wild type, except for the Fgfr2^Δ/Δ embryos, which died between E10 and E11. Fgfr2^Δ/Δ embryos developed no limb buds, and failed to form a functional placenta, a probable cause of early embryonic lethality (data not shown). These phenotypes were consistent with that of other Fgfr2 null mice(Xu et al., 1998), suggesting that the two inserted loxP sites used for conditional targeting of Fgfr2 are functional and that the Fgfr2^Δallele is a null allele.

The level of Fgfr2 transcription in E10.5 wild type and Fgfr2^Δ/Δembryos was evaluated by semi-quantitative RT-PCR using primers in exons flanking the three targeted exons (Fig. 3H). Similar amounts of mRNA were transcribed from both wild type Fgfr2 and Fgfr2^Δ alleles, suggesting that deletion of exons 8-10 did not affect mRNA transcription, processing, or stability. However, proteins translated from the Fgfr2^Δ transcript, if any, would be non-functional because of the lack of exons required for ligand binding (IIIb or IIIc) and membrane insertion (TM domain).

To conditionally inactivate Fgfr2 in the developing skeleton, Fgfr2^+/Δ; Dermo1^cre/+ double heterozygous mice were mated with Fgfr2^flox/flox homozygous mice(Fig. 2D, Fig. 3A). Fgfr2conditional knockout (Fgfr2^cko) mice(Fgfr2^flox/Δ;Dermo-1^cre/+) were born alive with the expected Mendelian frequency (∼25%). The efficiency of Dermo1-CRE-mediated recombination was assessed by detecting Fgfr2 expression in developing long bones of control and Fgfr2^cko mice by in situ hybridization (Fig. 3B-E)and by RT-PCR (Fig. 3H). An in situ hybridization probe, derived from exons 9 and 10 (encoding IIIc and TM domain, respectively), only detected intact Fgfr2 in normal control but not Fgfr2^cko femurs of E16.5 and P10 mice(Fig. 3B,C, and data not shown). In contrast, a probe derived from the tyrosine kinase domains of Fgfr2 detected both intact and deleted Fgfr2. The probe showed similar expression patterns in both Fgfr2^cko and control femurs of E16.5 and P10 mice (Fig. 3D,E, and data not shown). RT-PCR analysis also revealed that intact Fgfr2 transcripts were undetectable in skeletal tissues dissected from Fgfr2^cko mice(Fig. 3H). Together, these data demonstrated that the Fgfr2^flox allele was effectively targeted by Dermo1-CRE-mediated recombination in developing skeletal tissues. In most experiments an Fgfr2^Δ allele was incorporated to increase the efficiency of CRE-mediated recombination. However, Fgfr2^flox/flox, Dermo1^cre/+ mice appeared phenotypically identical to Fgfr2^flox/Δ;Dermo-1^cre/+ mice.

FGFR1 is another member of the FGF receptor family that closely resembles FGFR2 in both structure and function. Fgfr1 could be partially redundant with Fgfr2 and could be upregulated in mice lacking a functional Fgfr2 allele. Fgfr1 expression was therefore examined. The results showed that Fgfr1 expression clearly overlapped with that of Fgfr2 in developing bone. However, no difference in Fgfr1 expression between Fgfr2^cko and control mice was observed (Fig. 3F,G).

Fgfr2^cko embryos and newborn pups were phenotypically similar to littermate controls. However, during early postnatal development, Fgfr2^cko pups exhibited severe growth retardation and by four weeks of age, Fgfr2^cko mice were 40-50% smaller than controls (Fig. 3I,J). After a characteristic plateau in growth between 14 and 22 days, Fgfr2^cko mice showed a relatively normal growth curve. However, adult Fgfr2^cko mice remained 30-40% smaller than controls. Fgfr2^cko mice were fertile and had a similar life span to that of controls (>1year).

All Fgfr2^cko mice exhibited a dwarfism phenotype. At P15 and P21, the femur length of Fgfr2^cko mice was 87% and 79% of control (Table 1),respectively. Skeletal preparations showed a shortened axial and appendicular skeleton (in 21 of 23 skeletons examined) as well as a domed-shaped skull(Fig. 4A,B). Other skeletal abnormalities were also found in Fgfr2^cko mice with variable severity. The midline sutures (sagittal and metopic) remained patent and the occipital arch was open at the dorsal midline (in 5 of 7 mice examined) (data not shown). Vertebrae also showed abnormalities, including a non-ossified gap in the dorsal midline of both cervical and thoracic vertebrae and absence of the spinous processes (in 7 of 7 mice examined)(Fig. 4C,D). In Fgfr2^cko mice, several tarsal joints failed to develop (13 of 13 mice examined) (Fig. 4E-H). The origin of the joint fusion appeared to be a failure of cavitation of the cartilaginous anlage prior to ossification of these bones(Fig. 4E,F).

In addition to the decreased size of the appendicular skeleton, bone mineral density (BMD) was reduced in all Fgfr2^cko mice examined. Radiographic analysis showed that Fgfr2^cko mice had decreased BMD compared with age-matched control mice(Fig. 4I,J). Quantitative assessment of the femur and lumbar vertebra by DEXA showed that beginning at 3 weeks of age Fgfr2^cko mice had significantly decreased BMD. Bone density of Fgfr2^cko mice increased with age, but always remained lower than that of age-matched control mice even in 58 week-old mice (Fig. 4K).

Because the osteoblast is the only cell type that engages in bone formation and was one of the cell types effectively targeted by Dermo1-CRE,reduced bone density in Fgfr2^cko mice was likely the result of decreased bone formation as a consequence of osteoblast malfunction. Previous in vitro studies suggested that FGFR signaling affects osteoblast differentiation by regulating osteoblast-specific gene expression(Chikazu et al., 2001; Newberry et al., 1996; Newberry et al., 1997; Zhou et al., 2000). Reduced bone formation in Fgfr2^cko mice could result from down-regulation of genes required for osteoblast differentiation. The expression of several osteoblast marker genes were examined by in situ hybridization and quantitative RT-PCR (qRT-PCR). At E16.5, although functional Fgfr2 transcripts were no longer present, skeletal development was not dramatically affected. Skeletal preparations at this age show no evidence of delayed ossification (data not shown) and long bones were of normal size with a well developed and ossified bone collar(Fig. 5A,D). The expression of the genes Cbfa1, osteopontin and osteocalcin showed a similar pattern and intensity in Fgfr2^cko and control mice(Fig. 5B,C,E,F and data not shown). However, during early postnatal development, histological analysis showed dramatic differences between Fgfr2^cko mice and normal controls in osteogenic regions (see below). At P7, expression of Cbfa1, collagen type I, osteopontin and osteocalcin in Fgfr2^cko mice showed decreased signal intensity and area,when compared with that of controls (Fig. 5G-L). This corresponded with reduced total osteoblast number in osteogenic regions of Fgfr2^cko mice (see below). qRT-PCR showed a modest decrease in osteocalcin expression in Fgfr2^cko mice (80% of control, P<0.05). osteopontin expression was not significantly different in Fgfr2^cko mice and control mice. These data suggest that loss of Fgfr2 does not block osteoblast differentiation.

At embryonic stages, the major osteogenic region is located in the perichondrium and periosteum, which form the bone collar. Vascular invasion of hypertrophic chondrocytes from the central bone collar allows formation of the metaphysis adjacent to the hypertrophic chondrocyte zone(Karsenty and Wagner, 2002; Ornitz and Marie, 2002). The metaphysis contains a large number of active osteoblasts, arranged along calcified spicules of cartilage (Caplan and Pechak, 1987). These cells progressively lay down bone matrix and longitudinally extend trabecular bone. Histologically, osteoblasts that are actively engaged in depositing bone matrix protein appear as plump cuboidal cells with a homogenous basophilic cytoplasm. At the periphery of the metaphysis, multiple longitudinally extended trabeculae merge with perichondrial tissue to form new diaphyseal cortical bone.

In Fgfr2^cko mice, the metaphysis contained significantly less trabecular bone and in some instances was nearly devoid of trabecular bone (Fig. 6A,B, Table 2 and data not shown). Morphometric analysis showed that trabecular zone length and width was reduced, resulting in a 40% decrease in metaphyseal area(Table 2). Because osteoblasts were distributed throughout the entire metaphysis, the total osteoblast number was reduced. However, osteoblast cell density was not significantly different in the primary spongiosa in Fgfr2^cko mice(Table 3). Significantly, in Fgfr2^cko mice, metaphyseal osteoblasts were disorganized along the trabecular surface and showed an atrophic morphology with a granular-appearing cytoplasm (Fig. 6C,D).

Perichondrial and periosteal tissue was also severely affected in Fgfr2^cko mice. The perichondrium (contiguous with the periosteum) lies adjacent to the growth plate. This multi-layered fibrous collagenous structure consists of a mineralized bone collar, an osteoprogenitor cell layer and an outermost fibroblast-like cell layer(Fig. 6E). Perichondrial/periosteal thickness in Fgfr2^cko mice was reduced by approximately 50% in regions adjacent to the growth plate and the metaphysis (Fig. 6F). Additionally, the osteoprogenitor cell layer was more compact and contained fewer and atrophic-appearing cells.

The periosteal surface in the mid-diaphysis is another osteogenic zone in developing long bone (Caplan and Pechak,1987). In control mice, active osteoblasts line the periosteal surface of cortical bone. In Fgfr2^cko mice, the number of active osteoblasts residing in the diaphyseal periosteum was decreased(Fig. 6G,H). Additionally, in control mice the mid-diaphysis of the femur contained many lacunae, which were occupied by osteoblasts. However, in Fgfr2^cko mice, the mid-diaphyseal region of the femur contained only a few lacunae. The lacunae that were present were more compact and contained fewer atrophic-appearing osteoblasts (Fig. 6H).

The ability of osteoblast to produce bone matrix was assessed by measuring MAR by calcein double labeling. During the third and forth postnatal week the MAR on Fgfr2^cko mice was too low to measure while control mice had a MAR value of 2.11±0.03 μm/day on the diaphyseal endosteum and 2.43±0.07 μm/day on the diaphyseal periosteum of the femur(Fig. 7A,B).

During skeletal development, the osteoblast lineage undergoes continued proliferation to accommodate rapid bone growth during early postnatal development. The decreased total number of osteoblasts in osteogenic regions of Fgfr2^cko mice suggested that osteoblast proliferation could be dependent on FGFR2 signaling. BrdU labeling studies identified zones of proliferation in the perichrondrium and metaphysis, as well as in the mid diaphyseal periosteum. These osteogenic regions contained fewer proliferating osteoblasts in Fgfr2^cko mice compared with controls(Fig. 7C-F, Table 3).

Another possible reason for decreased cell number is increased cell death due to loss of Fgfr2. However, total numbers of apoptotic cells,analyzed by TUNEL labeling and caspase 3 immunohistochemistry, were similar in femur sections of Fgfr2^cko and control mice at both embryonic and postnatal stages (data not shown). These results suggested that Fgfr2 was essential for proliferation but not survival of osteoblasts.

Longitudinal bone growth is tightly coupled to proliferation and differentiation of chondrocytes in the growth plate. When either process is disrupted, longitudinal bone growth is affected. Since Fgfr2^cko mice also showed a dwarfism phenotype, the growth plate of Fgfr2^cko mice was examined. At all developmental stages examined, the growth plate of Fgfr2^cko mice showed an overall intact histomorphologic architecture(Fig. 6A,B). In Fgfr2^cko mice the length of the proliferation zone was similar to that of controls (Table 2) and the BrdU labeling index was not significantly changed(Fig. 7C,D). However, the length of the hypertrophic zone was greatly reduced during early postnatal development (Table 2). The length of the hypertrophic zone is controlled by the rate of proliferating chondrocyte differentiation and by osteoclast (chondroclast)-mediated degradation at the chondro-osseous junction(Gerber et al., 1999; Turner et al., 1994).

The Indian hedgehog (IHH) signaling pathway is an important regulator of chondrocyte proliferation and differentiation(Karsenty and Wagner, 2002; Ornitz and Marie, 2002). The IHH receptor, Patched, is expressed by proliferating chondrocytes(St-Jacques et al., 1999) and its level of expression is a measure of the strength of the IHH signal(Chen and Struhl, 1996; Ingham, 1998). No significant difference was observed in Patched expression in Fgfr2^cko and control mice (data not shown), indicating that IHH activation in the growth plate was intact during skeletal development. These observations suggest that the primary cause of dwarfism and the reduced size of the hypertrophic zone are unlikely to be caused by defective chondrocyte differentiation.

Osteoclasts were examined by staining for TRAP activity. During embryogenesis (E16.5), no significant difference was found between control and Fgfr2^cko mice (data not shown). However during early postnatal development, although the distribution pattern of TRAP-positive cells was similar in Fgfr2^cko and control mice, the relative number of TRAP-positive cells in the chondro-osseous junction increased by 40% (n=6 mice, P<0.001) in Fgfr2^cko mice (Fig. 7G,H). Interestingly, osteoclasts were also more mature (larger)than in control mice.

The CRE-LoxP system is an efficient method for tissue-specific gene modification during development. The success of this strategy depends on tissue-specific expression of CRE in the anlage of a targeted tissue at a time prior to the expression of the targeted gene. In this study a Dermo1-cre knockin mouse was used to probe the function of Fgfr2 during skeletal development. Dermo1-CRE activity was present in condensed mesenchyme which gives rise to both the chondrocyte and osteoblast lineage. The specificity of Dermo1-CRE-mediated recombination was confirmed by lineage tracing analysis using the R26R mouse (Soriano,1999). This analysis identified both osteoblasts and chondrocytes as lineage descendants of cells that express Dermo1-cre. The efficiency of CRE-mediated β-gal activation in these lineages approached 100 percent. In situ hybridization confirmed the efficiency of Dermo1-CRE-mediated gene deletion by demonstrating the absence of the floxed Fgfr2 exons in skeletal tissue.

Fgfr2 is one of the earliest genes expressed in the mesenchymal condensation (Delezoide et al.,1998; Orr-Urtreger et al.,1991; Peters et al.,1992); probably before the very first expression of Dermo1-cre (Li et al.,1995). It is not known how long it takes for complete CRE-mediated gene inactivation after cre is first expressed and translated. It is therefore possible that FGFR2 protein is synthesized and retained in condensing mesenchymal cells, even after the Fgfr2 gene is disrupted. The precise timing of Fgfr2 inactivation could therefore lead to heterogeneity in the onset and possible severity of the skeletal phenotype. Because of this potential delay in Fgfr2 inactivation, the complete function of FGFR2 in the mesenchymal condensation has not been clearly addressed by this study. However, based on the comparative expression patterns of Dermo1-cre and Fgfr2, we conclude that the skeletal phenotype in Fgfr2^cko mice is a direct cell autonomous consequence of loss of FGFR2 function in the osteoblast, following the formation of the mesenchymal condensation. Interestingly, introduction of a frame shift mutation into the c exon of Fgfr2 resulted in a viable mouse with a similar degree of skeletal dwarfism and similar defects in skeletal morphology (Eswarakumar et al.,2002). However, in Fgfr2IIIc^-/- mice a delay in ossification was observed whereas in Fgfr2^cko mice there was no delay in ossification. This may reflect differences in timing of Fgfr2 inactivation.

Recent studies have established the transcription factor Cbfa1 as a determinant of osteoblast differentiation. Cbfa1 is the earliest and most specific gene expressed during osteoblast differentiation. In Cbfa1-deficient mice, the osteoblast differentiation process is completely blocked (Komori et al.,1997; Otto et al.,1997). CBFA1 controls the expression of genes that encode various bone matrix proteins. The similar expression of Cbfa1 and other osteoblast-specific genes in both Fgfr2^cko and control mice indicated that Fgfr2 is not necessary for the early stages of osteoblast differentiation. However, osteoblast differentiation is normally accompanied by profound morphological and functional changes. Osteoprogenitor cells, which appear morphologically like fibroblasts, are relatively small and have a compact cytoplasm. Although osteoprogenitor cells produce bone matrix proteins, their main function is to maintain the progenitor pool through proliferation. In contrast, mature osteoblasts are plump cuboidal cells with ample cytoplasm. Electron microscopy studies reveal that the cytoplasm of mature osteoblasts contains large amounts of rough endoplasmic reticulum and an extensive Golgi apparatus, consistent with their function in synthesizing large amounts of bone matrix proteins. The presence of atrophic cytoplasm in osteoblasts of Fgfr2^cko mice and the dramatically decreased MAR strongly indicates that their protein synthesis capacity was reduced. Fgfr2 was expressed by osteoprogenitor cells in perichondrial and periosteal tissues (Fig. 1H,K) and by mature osteoblasts in the metaphysis, and in diaphyseal lacunae (Fig. 1K). Notably, osteoblasts in diaphyseal lacunae were post mitotic(Fig. 7E). The Fgfr2expression pattern and the osteogenic phenotype in Fgfr2^cko mice suggests that FGFR2 signaling regulates proliferation of osteoprogenitor cells and the anabolic function of mature osteoblasts. Consistent with this observation, the effect of implanting FGF2 beads over the coronal suture in mice suggested that the primary function of FGFR2 is to regulate osteoblast proliferation in suture mesenchyme(Iseki et al., 1999).

It has been shown that FGFR1 signaling can directly regulate Cbfa1expression in osteoblasts (Zhou et al.,2000). In this study, loss of Fgfr2 did not affect Cbfa1 expression, suggesting that FGFR2 signaling may function in a Cbfa1-independent manner. Whether Cbfa1 is directly involved in osteoblast proliferation is not clear. Interestingly, low-density lipoprotein receptor-related protein (Lrp)-5-deficient mice have been shown to have a Cbfa1-independent decrease in osteoblast proliferation and bone formation (Kato et al., 2002). Since LRP5 mediates WNT signaling, these data,together with this study, suggest that proliferation of the osteoblast lineage is regulated by multiple signaling pathways that do not require CBFA1 as a downstream effector. Moreover, the morphological characteristics of mature osteoblasts are similar to other cell-types engaged in the synthesis and secretion of large amounts of proteins. This also suggests non-lineage-specific regulation of metabolic function because, unlike the complete lack of bone formation in Cbfa1-deficient mice, the diminished bone formation in Fgfr2^cko mice suggests that FGFR2 signaling is complementary to that of CBFA1 during skeletal development. Indeed, we have identified a transcriptional activation domain in CBFA1 that is markedly enhanced in response to FGFR2 activation (D.A.T. and D.M.O.,unpublished).

Skeletal dwarfism is often associated with decreased chondrocyte proliferation. In ACH, the hypertrophic zone is shortened as a result of reduced chondrocyte proliferation and differentiation(Naski et al., 1998; Ornitz and Marie, 2002). In contrast, Fgfr2^cko mice have decreased bone length without apparent defects in chondrocyte proliferation. However, similar to Fgfr2IIIc^-/- mice(Eswarakumar et al., 2002), Fgfr2^cko mice also exhibit a shortened hypertrophic zone. Interestingly, the decreased length of the hypertrophic zone in Fgfr2^cko mice was accompanied by an increase in the number of mature osteoclasts in the chondro-osseous junction. Increased osteoclast activity could account for the decrease in hypertrophic zone length by increasing the rate of removal of calcified hypertrophic chondrocyte matrix. Furthermore, in contrast to ACH, in which bone formation in the diaphysis is not affected and diaphyseal cortical bone has normal thickness(Rimoin et al., 1970), Fgfr2^cko mice showed a dramatic reduction in diaphyseal thickness. These data suggest that the pathogenesis of the dwarfism in Fgfr2^cko mice is significantly different from that of ACH and that metaphyseal osteogenesis, which is severely affected in Fgfr2^cko mice, is an important regulator of longitudinal bone growth.

Fgfr2^cko mice exhibit decreased bone density throughout their life (examined up to 58 weeks). The finding that FGFR2 is required by the osteoblast to regulate mineral deposition suggests that FGF signaling may be important for the maintenance of bone density.

We thank Ling Li, Heather Walker and Craig Smith for their excellent technical help, X. Hua for microinjection, the Washington University Siteman Cancer Center Embryonic Stem Cell core facility for ES cell culture and the MBP histology core facility. This work was supported by NIH grants HD39952,CA60673, DK52446 (D.A.T.) and DK52574 (microinjection) and a generous contribution from the Virginia Friedhofer Charitable Trust.