Distinct roles for Hedgehog and canonical Wnt signaling in specification,differentiation and maintenance of osteoblast progenitors

The mammalian skeleton forms from three distinct cell lineages(Nakashima and de Crombrugghe,2003). Facial bones and the cranium are derived from the neural crest. The base of the skull, the parietal bones and the axial elements of the ribs and vertebrae are derived from paraxial mesoderm. The sternum and long bones are formed from lateral plate mesoderm. Although several cell lineages contribute to skeletal structures, each gives rise to a common bone matrix-secreting cell type, the osteoblast. The requirement of this cell type for the production of bone is clearly demonstrated by the targeted mutation of Runx2 (Komori et al.,1997; Ducy et al.,1997; Otto et al.,1997) and osterix1 (Osx1; Sp7 - Mouse Genome Informatics) (Nakashima et al.,2002), transcriptional regulators expressed in osteoblast progenitors where both mutants lack all bone.

Bone can be produced by two distinct mechanisms, direct differentiation of osteoblasts from mesenchymal progenitors (intramembranous ossification, e.g. skull and face) or by formation of bone on a cartilage scaffold (endochondral ossification, e.g. remainder of the skeleton). Endochondral ossification, the principle focus of this study, is initiated by the condensation of multipotent mesenchymal progenitor cells into structures that anticipate skeletal elements of the adult (reviewed by Kronenberg,2003). Chondrocytes are the first cell type to form, starting out as immature proliferative cells that express type 2 collagen (Col2a1)that subsequently mature into postmitotic type 10 collagen(Col10a1)-expressing hypertrophic chondrocytes. Osteoblast progenitors can first be identified in the inner layer of perichondrial cells that lie immediately adjacent to the zone of hypertrophic chondrocytes, the periosteum, where the first bone matrix is deposited. Death of hypertrophic chondrocytes and vascular invasion result in the formation of a new area of mineralization, the primary spongiosa, within the shaft of the long bones.

Several lines of evidence implicate Hedgehog (Hh) and canonical Wnt signaling in the regulation of endochondral ossification (reviewed by Kronenberg, 2003; Karsenty, 2003). Indian hedgehog (Ihh) is expressed by prehypertrophic chondrocytes and plays an essential role in coordinating the growth and differentiation of chondrocytes, both directly and through the control of other factors, notably parathyroid hormone-related peptide (Pthrp; Pthlh - Mouse Genome Informatics). In addition, Ihh appears to act directly on perichondrial mesenchyme to initiate an osteogenic program in osteoblast progenitors; in the absence of an Ihh input, these cells adopt an alternate chondrogenic fate(Long et al., 2004). The failure of activation of Runx2, a crucial early determinant of the osteoblast lineage, indicates that Hh signaling acts to initiate an osteogenic program. Whether Hh signaling is required at later stages of the osteogenic program has not been addressed.

Initial in vivo evidence for canonical Wnt activity in osteogenesis came from studies of human and mouse mutations in low-density lipoprotein receptor-related protein 5 (Lrp5) which encodes a co-receptor(together with the frizzled family of multi-pass membrane proteins) for Wnt ligands. Human genetic analysis identified mutations in LRP5 where bone mass was increased (activating mutations)(Boyden et al., 2002), or decreased, in osteoporosis pseudoglioma syndrome (OPPG; null mutations)(Gong et al., 2001). These observations have been supported by parallel mouse studies(Kato et al., 2002). Together,these findings pointed to a role for a Wnt-mediated process in regulating bone mass. More recently, manipulation of the canonical Wnt pathway by regulating the activity of β-catenin, which together with members of the Lef/Tcf family forms a transcriptional effector complex for this arm of the Wnt signaling pathway, has shed light on this mechanism. Canonical Wnt signaling in osteoblasts appears to regulate the production of osteoprotegerin (OPG), a factor that acts on the other key cell type of bone metabolism, the osteoclast, inhibiting osteoclast-mediated bone resorption(Glass, 2nd et al., 2005).

In addition to Wnt function in the maintenance of bone homeostasis, several recent developmental studies have suggested that Wnts play a role in the specification of osteoblasts. Specifically, Hill et al.(Hill et al., 2005) used a Prx1-cre line to conditionally inactivate β-catenin function,where Prx1-cre is active in the developing head and limb mesenchyme at E9.0 prior to skeletogenesis (Hill et al., 2005). This resulted in a dramatic reduction in the size of long bone skeletal elements and in an overall failure to develop bone. Characterization of osteoblast development in this model demonstrates that Runx2 expression was detectable; however, these mutants failed to express Osx1, a Runx2-dependent transcriptional regulator that, like Runx2, is essential for all bone development. In a second study, Day et al. (Day et al., 2005) used Dermo1-cre and Col2α1-cre mouse lines to remove β-catenin function in mesenchymal condensations prior to chondrocyte and osteoblast development (a later stage than Prx1-cre),and in cartilaginous condensates prior to the specification of osteoblasts,respectively. In both models, an overall reduction in the size of skeletal elements was observed that was accompanied by an apparent arrest of osteoblast development at a Runx2⁺, Osx1⁺ precursor stage. By contrast, a third study using the same Dermo1-cre andβ-catenin conditional mouse lines reported an earlier arrest at a Runx2⁺, Osx1^- stage(Hu et al., 2005). Thus,although there is some disagreement as to the phenotypical outcomes, all studies indicate that β-catenin activity within skeletal elements is required for formation of mature osteoblasts. Furthermore, the failure to complete an osteogenic program was associated with the appearance of ectopic chondrocytes, suggesting a potential link between canonical Wnt signaling and inhibition of a chondrogenic pathway within osteoblast progenitors(Hill et al., 2005; Day et al., 2005).

As these models remove β-catenin function broadly within the skeletal anlage, not specifically in the osteoblast lineage, the issue of whether canonical Wnt/β-catenin activity acts directly within the osteoblast lineage to promote an osteogenic program remains to be resolved. Here, we have used a novel Osx1-GFP::Cre mouse strain to investigate the direct roles of hedgehog and canonical Wnt signaling in early Runx2⁺,Osx1⁺ osteoblast precursors and their derivatives. These data demonstrate an essential role for canonical Wnt signaling, but not for Hh signaling, in the progression of osteoblast precursors to mature,matrix-secreting osteoblasts. Interestingly, cell fate analysis demonstrates that removal of β-catenin activity in Runx2⁺,Osx1⁺ osteoblast precursors gives rise to ectopic chondrocytes, suggesting, along with earlier data(Hill et al., 2005; Day et al., 2005), an extended role for canonical Wnt signaling in the suppression of an alternate chondrocytic fate within osteoblast precursors. In contrast to loss-of-function studies, enhanced β-catenin activity rapidly accelerates this program leading to a dramatic expansion of osteoblast precursors and the premature synthesis of a mineralized bone matrix in the long bones. However,differentiation to a terminal osteocalcin⁺(Oc⁺; Bglpa1 - Mouse Genome Informatics)osteoblast is blocked by stabilization of β-catenin. Thus, canonical Wnt/β-catenin signaling plays crucial roles at specific stages of the osteogenic program.

Col2α1-cre3(Long et al., 2001),β -catenin^c/c(Brault et al., 2001), Catnblox^ex3/lox^ex3(Harada et al., 1999),β -catenin^+/n(Haegel et al., 1995), Smo^c/c, Smo^+/n(Long et al., 2001) and Sox2-cre (Hayashi et al.,2002) mouse lines have been described previously. The Wnt7b conditional allele will be described elsewhere.

The Osx1-GFP::Cre mouse line was generated by pronuclear injection of a bacterial artificial chromosome (BAC) containing the Osx1 gene targeted at exon 1, using standard BAC recombination methods(Lee et al., 2001). Specifically, primers Osx5 (CTC TCC CTT CAC CCT CTC CCA CTG GCT CCT CGG TTC TCT CCA TCT GCC TGA CTC CTT GGG ACC CGG TCC CCA GCT CGA GGA GAA TTC GCT GTC TGC GAG G) and Osx9 (TAG GCA TGG ATT AGG ACC AGG AAG ATT GTA GCT GGT TTC TTA AGG AAG GGA ACA GTT ACC TCA AGC AGA GCT ATT CCA GAA GTA GTG AGG) containing 80- and 70-nucleotide Osx1-specific homology arms(underlined), respectively, were used to PCR amplify the transgenic tTA regulated GFP::Cre construct from the plasmid pTGCK (M. T. Valerius and A.P.M., unpublished), a derivative of plasmids triTAUBi(Kistner et al., 1996), pBS592(Le et al., 1999) and pIGCN21(Lee et al., 2001), using high-fidelity platinum pfx polymerase, following the manufacturer's directions (Invitrogen).

The PCR product was gel purified prior to electroporation into EL250 bacteria containing the BAC clone RP23-399N14 (Children's Hospital Oakland Research Institute, Oakland, CA). Recombinants were selected on bacterial plates containing kanamycin and PCR screened for homologous recombination events. The kanamycin selection cassette was removed from correctly targeted BAC clones by L-arabinose induction of flpe recombinase prior to being grown up and nucleobond purified (BD Biosciences). Osx1-GFP::Cre mice were genotyped using primers Osx10 (CTC TTC ATG AGG AGG ACC CT) and TGCK-3′(GCC AGG CAG GTG CCT GGA CAT). Activity of Osx1-GFP::Cre was assessed by mating the Osx1-GFP::Cre mouse to R26RlacZ reporter mice;embryos were collected and lacZ stained as described previously(Soriano, 1999). Alternatively, limbs were removed, fixed, cryo-embedded, sectioned and then stained for lacZ activity.

In our studies, we failed to observe any significant differences between skeletons of Osx1-GFP::Cre and β-catenin^c/nembryos and those of wild-type litter mates. Data are shown whereβ-catenin removal is compared with β-catenin^c/nindividuals to control for reduced levels of β-catenin activity where only one allele is active. Skeletons were stained as described previously(Long et al., 2001). Embryonic limbs were dissected and fixed in sodium phosphate-buffered 4%paraformaldehyde over night at 4°C. Limbs were then washed in phosphate-buffered saline and either transferred to and stored in 70% ethanol prior to being paraffin-wax processed, embedded and sectioned, or cryo-protected in 30% sucrose prior to being frozen mounted in OCT and sectioned. In situ hybridization with ³⁵S-labeled probes and BrdU analysis of cell proliferation were carried out as described previously(Long et al., 2001). Hematoxylin/eosin, von Kossa and Safranin O staining were performed using standard histological methods. TUNEL analysis was performed using the ApopTag in situ Apoptosis Detection Kit (Chemicon), as described in the manufacturer's instructions.

Earlier reports (Hu et al.,2005; Day et al.,2005; Hill et al.,2005) where β-catenin activity was removed prior to or after cartilage condensation have come to somewhat different conclusions about the requirement for canonical Wnt/β-catenin signaling in the initial specification of osteoblast progenitors. In a similar approach, we generated Col2α1-cre3;β-catenin^c/cconditional knockout embryos in which β-catenin function was specifically removed from the mesenchyme-derived cartilaginous condensates that give rise to both the cartilage and the bone of the skeleton(Long et al., 2001). Hence Col2α1-cre3 is active both in osteoblasts and in chondrocyte lineages.

Although the size and structure of the mutant growth plate was comparable between β-catenin^c/c orβ -catenin^c/+ littermates, and vascular invasion and seeding of the marrow cavity was evident from the presence of red blood cells within the forming marrow cavity, no histologically identifiable bone matrix was observed in the mutant long bones (Fig. 1).

These observations were corroborated by molecular analysis. Col1α1 and Runx2, markers of early osteoblast progenitor cells, were expressed in mutant tibia at E18.5, albeit at lower levels than in β-catenin^c/c littermates. Specifically, Col1α1 was expressed by cells within a wedge-like structure of mesenchyme invading into the presumptive marrow cavity. Furthermore, Osx1, a marker for osteoblast precursor cells whose function is downstream of Runx2(Nakashima et al., 2002), was expressed in the perichondrial region where osteoblast progenitors first arise, coincidental with the appearance of Col1α1-expressing cells in the invading mesenchymal wedge. By contrast, Oc, a definitive marker gene of terminally differentiated osteoblasts, was absent from most long bones; the occasional Oc⁺ cell observed most likely reflected mosaicism in the activity of the cre transgene(Fig. 1; data not shown). These results are in general agreement with recent studies(Hu et al., 2005; Day et al., 2005; Hill et al., 2005). However,our observation that, despite robust Osx1 expression, no mineralized bone matrix is formed suggests that canonical Wnt signaling may be required at a later stage downstream of Osx1 in the osteogenic program.

Although these experiments are consistent with an intrinsic requirement for canonical Wnt/β-catenin signaling within the developing osteoblast lineage, we cannot rule out an alternative role for canonical Wnt/β-catenin signaling in the cartilage component that might indirectly regulate the osteoblast pathway. Indeed, cartilage development is perturbed in this and other similar models in which early growth and hypertrophic differentiation are disrupted (Hu et al.,2005; Day et al.,2005; Hill et al.,2005) (data not shown), and Wnt signaling is itself implicated in the chondrogenic program (Day et al.,2005). Furthermore, hypertrophic chondrocyte-derived factors(notably Ihh) play a crucial role in endochondral osteoblast development(reviewed by Kronenberg,2003). To address signaling directly in the osteoblast lineage, we generated a BAC transgenic mouse line in which expression of a Tet-off regulatable GFP::Cre fusion protein is placed under the transcriptional regulation of the Osx1 promoter. The Tet-off cassette provides an additional level of potential temporal control of GFP::Cre activity within the osteoblast lineage that has not been examined in this study. The data herein and our unpublished data (S.J.R. and A.P.M.) indicate that this is an effective strategy for the dual regulation of gene activity. However,unrelated studies in which this regulatory system has been introduced by gene targeting into several loci suggest that this strategy can lead to dominant phenotypes. These are not always observed, as in this strategy, suggesting a gene-, cell type- and/or context-dependent mechanism, the nature of which is unclear.

As an initial step to characterize expression and activity of the Osx1-GFP::Cre transgene, transgenic mice were mated to Rosa26lacZ reporter (R26R) mice(Soriano, 1999). lacZexpression, indicative of GFP::Cre activity, was observed in all boney elements (endochondral and membranous) consistent with endogenous Osx1 expression (Nakashima et al., 2002) (Fig. 2B). Bone-specific activity was evident when GFP::Cre fluorescence and lacZ expression were examined in tibial sections(Fig. 2C,D). GFP::Cre and lacZ expression were observed in the inner bone-forming perichondrium, adjacent to hypertrophic chondrocytes, and sporadically in hypertrophic chondrocytes. Similarly, analysis at E18.5(Fig. 2F-H) and postnatal day 10 (Fig. 2I-K) demonstrated that expression of lacZ and GFP::Cre fluorescence were largely restricted to the periosteum and primary spongiosa. Thus, Osx1-GFP::Cre is active specifically within the osteoblast lineage and expression is maintained throughout embryonic and early postnatal development. The fact that GFP::Cre activity is largely absent from chondrocytes in which Osx1 normally shows weak expression(Nakashima et al., 2002)suggests either that low-level GFP::Cre activity is inefficient in initiating recombination, or, alternatively, that chondrocyte-specific regulatory elements are absent in the BAC transgene.

To determine whether β-catenin is required downstream of Osx1in the osteoblast lineage, Osx1-GFP::Cre;β-catenin^c/n conditional knockout embryos were collected at E14.5, E16.5 and E18.5, and their gross skeletal organization analyzed (Fig. 3). At E14.5, Osx1-GFP::Cre;β-catenin^c/n mutant skeletons were comparable in size to wild-type embryos. At E16.5, no Alizarin red staining was observed despite clear zones of hypertrophy in mutant limbs. By E18.5, mineralization of the endochondral skeletal elements was visible;however, as in the Col2α1-cre3 removals, the remaining mineralization was associated with hypertrophic chondrocytes. Importantly, in contrast to the Col2α1-cre3 removal of β-catenin activity, which is specific to the endochondral skeleton, Osx1-GFP::Cre;β-catenin^c/n embryos lacked the membranous bone of cranial ossification centers. Hence, there is a complete loss of bone deposition that is reminiscent of the loss of osteoblast determinants, such as Runx2(Komori et al., 1997; Ducy et al., 1997; Otto et al., 1997) and Osx1 (Nakashima et al.,2002).

Histological analysis of long bones at E14.5 showed that the size and organization of the developing tibia was comparable to wild type with respect to chondrocyte differentiation and proliferation, as we had expected from the demonstrated specificity of the Osx1-GFP::Cre transgene (data not shown). Analysis of tibial elements at E16.5 and E18.5 indicated that the size and structure of the growth plate was comparable to wild type. The only detectable mineralized matrix was associated with hypertrophic chondrocytes and, in addition to the normal growth plate, hypertrophic chondrocytes ectopically lined the periosteal region(Fig. 4 and data not shown). Vascular invasion is critically linked with osteogenesis; however, the adjacent forming marrow cavity was well vascularized, and Mmp9, Mmp13and Vegf, which are associated with vascular invasion were expressed normally (Fig. 4). Together,these observations support a primary role for β-catenin downstream of Osx1 in osteoblast specification.

To characterize osteoblast development, we examined expression of the osteoblast cell state marker genes Col1α1, Runx2, Osx1and Oc (Fig. 4). Expression of Col1α1, Runx2 and Osx1 was observed in the periosteal region and in a mesenchymal wedge invading the marrow cavity in E18.5 Osx1-GFP::Cre;β-catenin^c/n mutant tibia(Fig. 4). Thus, as expected Osx1⁺ osteoblasts were present. However, no Oc⁺ osteoblasts were detected indicating a failure of osteoblast progression to terminal Oc⁺ osteoblasts in mutant embryos. No apparent difference was observed in either cell death or cell proliferation between wild-type and mutant tibia (data not shown).

Several signaling systems are known to play important roles during osteoblast development (reviewed by Kronenberg, 2003; Karsenty, 2003). Ihhnull mice fail to differentiate osteoblasts, but arrest in this pathway appears to occur at an initiating step: the transition of an unspecified cell to a pre-osteoblast (St-Jacques et al.,1999; Long et al.,2004). As expected, the Ihh expression domain of prehypertrophic chondrocytes was present, consistent withβ -catenin^c/n embryos, and active Ihh signaling was evident from the upregulation of Ptch1 (the Hh receptor and primary target of Hh signaling) in the periosteal region where osteoblast specification initiates (Fig. 4).

Dickkopf homolog 1 (Dkk1), a negative regulator of Wnt signaling,and T-cell factor 1 (Tcf1; Tcf7 - Mouse Genome Informatics),a transcriptional mediator of Wnt signaling, are transcriptional targets of canonical Wnt signaling (Roose et al.,1999; Cong et al.,2003). Within the developing long bone, these genes were expressed within the osteoblast-forming region in the perichondrium and the primary spongiosa of β-catenin^c/c embryos, but were absent from Osx1-GFP::Cre;β-catenin^c/n mutants(Fig. 4), consistent with regulation by a β-catenin-mediated Wnt signaling pathway. In addition,bone morphogenic protein 3 (Bmp3) expression, which is normally detected throughout the periosteum and primary spongiosa, was absent in the mutant, suggesting that Bmp3 signaling lies downstream of canonical Wnt signaling (Fig. 4). Thus, Osx1-GFP::Cre removal of the β-catenin gene leaves Hh signaling intact in the bone forming region but terminates canonical Wnt signaling and,with this, progressive osteoblast development.

Further insight into the relationship between Hh and Wnt function comes from the osteoblast-specific removal of Hh signaling in Osx1⁺ osteoblast precursors and their descendents. As discussed earlier, Ihh is required for the specification of early Runx2⁺ osteoprogenitors, prior to the requirement for canonical Wnt signaling. However, whether this represents a transient signal input or the onset of a more extended role for Ihh signaling in osteoblast development has not been addressed. We generated Osx1-GFP::Cre;Smo^c/n embryos to remove smoothened (Smo)activity (and, consequently, all responsiveness to Hh signaling) after specification of a Runx2⁺;Osx1⁺ osteoblast progenitor. Analysis of Osx1-GFP::Cre;Smo^c/n embryos at E18.5 by whole-mount skeletal preparation, histological analysis and in situ hybridization with chondrogenic and osteogenic marker genes demonstrated a normal program of chondrocyte and osteoblast development(Fig. 5; data not shown). Thus,Hh signaling does not appear to play an essential role in the terminal differentiation of osteoblasts beyond an Osx1⁺ cell state.

Previous studies have indicated that when osteoblast differentiation is arrested prior to Osx1 expression the loss of a terminally differentiated osteoblast is accompanied by the appearance of ectopic chondrocytes (Nakashima et al.,2002; Long et al.,2004; Hill et al.,2005; Day et al.,2005). Consistent with this view, analysis of the expression of collagen 2α1 (Col2α1) and collagen 10α1(Col10α1), markers of proliferating and postmitotic hypertrophic chondrocytes, highlight ectopic chondrocytes lining the periosteal region and an invading mesenchymal wedge adjacent to Col1α1 (a marker of early osteoblast progenitors), Runx2 and Osx1 expressing cells (Figs 1, 4). Thus, the presence of Runx2 and Osx1 within an osteoblast precursor does not appear to be sufficient to maintain an osteogenic program in the absence ofβ-catenin; osteoblast precursors convert to a chondrocyte fate uponβ-catenin removal. This conclusion was further supported by performing double-labeled fluorescent in situ hybridization on wild-type and mutant tibia, comparing the expression of Col2α1 and Col10α1 with the activity of the Osx1-GFP::Cre transgene. Furthermore, Osx1-GFP::Cre-mediated activation of the R26R allele in Osx1-GFP::Cre;β-catenin^c/n;R26R mice demonstrates that the ectopic chondrocytes arise from Osx1-GFP::Cre-expressing osteoblast precursors (data not shown).

We next addressed whether activation of the canonical Wnt/β-catenin signaling pathway in osteoblast precursor cells influenced osteoblast differentiation. To address this, we mated Osx1-GFP::Cre mice to those carrying a conditional β-catenin allele (Δex3;designated Catnb^{lox(ex3)/lox(ex3)})(Harada et al., 1999). Cre-mediated excision of exon 3 encoded N-terminal regulatory sequences produces a stabilized form of β-catenin, cell autonomously upregulating canonical Wnt signaling within the osteoblast lineage.

Osx1-GFP::Cre;Catnb^lox(ex3)/+ mutants died at birth. Skeletal preparations at E14.5, E16.5 and P0 revealed that mutant embryos,overall, had shorter limbs in comparison with Catnb^lox(ex3)/+ wild-type littermates(Fig. 6). Alizarin red staining of the mineralized bone matrix was first evident at E16.5 in whole-mount skeletons. At this time, mutants appeared to have an intense and broader ossification center in the long bones in comparison with their wild-type counterparts, although ossification in the skull bones was delayed. By P0, a thick bony matrix characterized all long bones in the mutants, and bone formation was now visible in several cranial regions.

Histological analysis of the long bones at E14.5 and E16.5 showed, an abnormal wedge-shaped growth plate with very few identifiable hypertrophic chondrocytes when compared with wild-type tibias (Figs 7, 8). Analysis of histological sections at E13.5 (data not shown) and E14.5(Fig. 7) using von Kossa and Alizarin red staining techniques revealed extensive premature mineralization of the mutant tibia at this time; no mineralization was detected in the wild type, even at E14.5 (Fig. 7). At E16.5, the primary spongiosa was readily identifiable in the wild-type tibia but no primary spongiosa-like matrix was observed in the mutant. Rather,the matrix resembled the dense matrix normally restricted to the cortical bone-forming region that generates the bone collar(Fig. 8).

To characterize the observed premature ossification of the mutants, we examined expression of the osteoblast differentiation markers Col1α1, Osx1 and Oc. While Col1α1 was expressed in the periosteal region of the wild type at E14.5, and the primary spongiosa at E16.5, expression of Col1α1 was observed throughout the periosteal region and entire central region of the tibia in the mutants(Fig. 7; data not shown). A similar dramatic expansion was also observed in the expression of Osx1, suggesting that the stabilization of β-catenin in osteoblast precursor cells resulted in the promotion of osteoblast development. Although the synthesis of bone matrix was activated prematurely at E13.5, we failed to observe Oc⁺ terminal osteoblasts prior to the normal onset of Oc expression at E16.5 (data not shown). Furthermore, although Oc⁺ osteoblasts were clearly identifiable in the wild-type tibia at E16.5, Oc⁺osteoblasts were rare in the mutant; when present, these cells were exclusively restricted to the periosteal region and expressed low levels of Oc relative to their wild-type counterparts(Fig. 8). The expansion of the osteoblast lineage was accompanied by a 3-fold increase in proliferation in osteoblast-forming regions along the length of the periosteum in mutants at E14.5 (Fig. 7). Together, these observations suggest that the stabilization of β-catenin results in a marked increase in proliferation of an Osx1⁺ precursor population and an accelerated progression of an osteoblast program to mature bone-secreting osteoblasts. Although these cells actively synthesize a bone matrix, they do not progress to a terminal state, characterized by high Oc⁺ expression. Thus, the cessation of canonical Wnt/β-catenin signaling may normally accompany this progression in vivo. As most Oc⁺ cells are normally observed in the marrow cavity, Oc⁺ expression may reflect an absence of local canonical Wnt signaling in this region.

Recent evidence indicates that canonical Wnt/β-catenin signaling positively regulates bone matrix formation by suppressing osteoclast development (Glass et al.,2005). Thus, a failure of osteoclast development, and,consequently, a loss of bone matrix turnover, could explain the premature formation of a bone matrix following osteoblast-specific stabilization ofβ-catenin. However, when wild-type limbs were examined, it was evident that Trap-positive osteoclasts were not present at E14.5 (data not shown). Thus, the accelerated bone matrix phenotype precedes any role for osteoclasts in bone remodeling. However, when osteoclasts were normally detected in the wild type at E16.5, they were completely absent from the mutant (data not shown). Consistent with the findings of Glass et al.(Glass et al., 2005), that the osteoclast inhibitor osteoprotegerin (Opg; Tnfrsf11b - Mouse Genome Informatics) is regulated by canonical Wnt signaling, the failure of osteoclast formation correlated with a dramatic upregulation of Opgupon osteoblast-specific stabilization of β-catenin relative to wild-type osteoblasts at E14.5 (Fig. 7;data not shown).

Analysis of Dkk1 and Tcf1 expression (Wnt pathway modulators and targets) indicated that active Wnt signaling was clearly upregulated in the osteoblast population in response to the activation ofβ-catenin (data not shown), which is consistent with the activation of canonical Wnt signaling and direct regulation of Dkk1 and Tcf1 transcription by β-catenin. Despite over-mineralization of the tibia, vascular invasion was evident, and mutants expressed Mmp9and Mmp13 (data not shown). In addition, the population of Col10α1- expressing hypertrophic chondrocytes was clearly reduced (Fig. 7; data not shown). TUNEL analysis revealed that mutants displayed a significant increase in cell death at the interface of hypertrophic chondrocytes and the mineralized matrix, suggesting that TUNEL-positive cells were likely to represent hypertrophic chondrocytes (data not shown). Finally, we examined the expression of indicators of Ihh signaling. These appeared to be largely unaltered, suggesting that the observed increase in bone matrix does not result from an overt change in the Hh-regulation of osteoblast precursors(data not shown).

The nature of the putative Wnt ligand that would mediate Wnt action in the osteoblast lineage is unclear. A recent report has highlighted Wnt7bas a candidate (Hu et al.,2005). Wnt7b expression was reported in the osteoblast-forming region of the long bone; furthermore, expression was dependent upon Ihh signaling, and was upregulated in response to Hh signaling in Hh-stimulated osteoblast specification of C3H10T1/2 mesenchymal progenitor cells (Hu et al., 2005). However, we recently generated embryos in which all Wnt7b function was removed and we have not observed any defects in skeletal development(S.J.R. and A.P.M., unpublished).

We have used genetic approaches to modify Hh and Wnt signaling in the developing osteoblast lineage of the mouse. These data point to sequential and genetically separable roles for these signaling pathways in the progressive specification of osteoblasts within the endochondral skeleton(Fig. 9).

Osteoblast progenitors can first be identified within the inner perichondrium adjacent to, and coincident with, the first appearance of hypertrophic chondrocytes. This tight linkage reflects a crucial role for Ihh signaling (St-Jacques et al.,1999; Chung et al.,2001). Ihh is produced by pre-hypertrophic chondrocytes and appears to act directly on perichondrially located osteoblast progenitors to specify the osteoblast precursors(St-Jacques et al., 1999; Long et al., 2004). To date,all markers of this program in the endochondral-, but not the membranous bone-, derived skeleton are dependent on an initial Ihh input. Furthermore, Hh activates osteoblast development in a variety of mesenchymal and skeletogenic cell types in vitro (Nakamura et al.,1997; van der Horst et al.,2003; Long et al.,2004). In the absence of signaling, perichondrial osteoblast progenitors in the perichondrium adopt a chondrocyte fate, as evidenced by the layer of immature chondrocytes that surround the hypertrophic chondrocytes in Ihh mutants (St-Jacques et al.,1999), and by the ectopic chondrogenesis exhibited by perichondrially localized Smo mutant cells in chimeric mice(Long et al., 2004). Although this suggests that a potential cell, frequently termed an osteochondroprogenitor (OCP), resides within the perichondrium, a conclusion supported by clonal analysis in vitro(Nakase et al., 1993), there is currently no evidence that these cells ordinarily give rise to chondrocytes. Thus, during normal development, Ihh signaling appears to act as a switch within a specific population of inner perichondrial mesenchyme to initiate a program of bone formation. Failure to activate this switch results in cells adopting an alternative chondrocyte pathway of development(Fig. 9). Given the crucial role for Ihh signaling in regulating the temporal and spatial program of early osteoblast commitment, what role does Ihh play beyond this stage? Our results do not support an on-going role for Ihh signaling in progression along the osteoblast pathway of differentiation(Fig. 5). When Smoactivity is removed in Osx1⁺ osteoblast precursors, normal bone secreting Oc^High osteoblasts are generated, and the endochondral skeleton at birth is indistinguishable from wild type. Whether this is also true in the adult is currently under investigation. Clearly, Ihh continues to be expressed postnatally within prehypertrophic chondrocytes, and the upregulation of Ihh signaling at sites of fracture repair implicates Ihh in skeletal homeostasis, repair and regeneration (Vortkamp et al.,1998; van der Eerden et al.,2000; Le et al.,2001).

In contrast to Ihh, the manipulation of β-catenin activity provides evidence against an essential role for canonical Wnt signaling in early osteoblast specification. When β-catenin activity was removed in skeletal progenitors with Prx1-cre, Runx2⁺ prior to specification of the skeletal anlage in the limbs, osteoblast precursors were specified, but these failed to progress to Osx1⁺ precursors and no bone matrix was laid down (Hill et al.,2005) (Fig. 9). Although the precise kinetics of the inactivation of the β-catenin conditional allele and turnover of existing β-catenin protein in this model are not clear, it seems reasonable to conclude from the early expression of the Cre driver that β-catenin is not essential for the specification of a Runx2⁺ osteoblast precursor. Furthermore, the results are consistent with a role for canonical Wnt/β-catenin signaling in the transition to an Osx1⁺ osteoblast precursor. Other studies using later active Cre driver mice that, like Prx1-cre, are active in both the chondrocyte and osteoblast compartments of the endochondral skeleton give conflicting results. Although all reports fail to observe a role forβ-catenin in the initial specification of Runx2⁺osteoblast precursors, using an identical Dermo1-cre transgenic line to remove β-catenin from pre-cartilaginous endochondral skeletal condensates, Hu et al. (Hu et al.,2005) observed an arrest of osteoblast specification at a Runx2⁺, Osx1^- stage, whereas Day et al.(Day et al., 2005) reported a progression, albeit at reduced levels, to a Runx2⁺,Osx1⁺ precursor and arrest at this stage, suggesting a continuing role for β-catenin in osteoblast differentiation beyond the Osx1⁺ osteoblast precursor.

This interpretation is also supported by Col2α1-cre-mediated removal(Day et al., 2005) (this study; Fig. 9). However, the occasional appearance of Oc^High osteoblasts(Day et al., 2005) (this study)and the broad activity of these Cre strains within both osteoblast and chondrocyte lineages prevents a rigorous assessment of the specific direct roles for canonical Wnt/β-catenin signaling in osteoblast differentiation.

Our data, in which β-catenin activity is removed specifically from Runx2⁺, Osx1⁺ osteoblasts, provides compelling evidence that β-catenin is essential within the osteoblast lineage for the specification of an Osx1⁺ osteoblast to a bone-secreting osteoblast (Fig. 9). Furthermore, the demonstration that the stabilization ofβ-catenin, and, consequently, the activation of canonical Wnt signaling,within osteoblast precursors expands this population and accelerates the progression to a bone matrix-secreting osteoblast indicates that canonical Wnt/β-catenin signaling may regulate both the proliferation and maturation of the osteoblast precursor pool. Interestingly, continued activation of canonical Wnt/β-catenin signaling arrests osteoblasts at an Oc^Low stage, suggesting that progression to an Oc^High state may require the downregulation of this signaling input in the primary spongiosa where the majority of Oc^High cells are located(Fig. 9).

In addition to a proposed cell autonomous role for β-catenin in promoting a bone-secreting osteoblast pathway, recent studies suggest that canonical Wnt/β-catenin signaling may play a non-cell autonomous role in suppressing the development of bone-matrix-degrading osteoclasts through the production of OPG (Glass et al.,2005). Our data lend further support to this view. We observed a dramatic upregulation of Opg upon stabilization of β-catenin in Osx1⁺ osteoblast precursors. However, as the normal appearance of osteoclasts occurs after the onset of bone matrix secretion in this model, the premature bone matrix deposition observed at E14.5 precedes any role for osteoclast suppression in this early phenotype. By contrast,OPG-mediated inhibition of osteoclast development may contribute to the massive expansion of a cortical bone-like matrix in E16.5 long bones of Osx1-GFP::Cre;Catnb^lox(ex3)/+ embryos.

Interestingly, in earlier reports, in which Cre removed β-catenin broadly in skeletal structures, ectopic chondrocytes were observed in perichondrial regions where cortical bone first arises(Hill et al., 2005; Day et al., 2005). Here, we show that the osteoblast-specific removal of β-catenin results in osteoblast precursors in which Cre was active, adopting a chondrocyte-like fate. These ectopic chondrocytes appear to undergo a transition from early Col2α1 immature chondrocytes to Col10α1 hypertrophic chondrocytes, as in the normal growth zone. Thus, β-catenin acts both to promote an osteoblast program and to block an alternative program of chondrogenesis within osteoblast precursors: an Ihh-mediated activity at an earlier stage in the osteoblast lineage.

How do Hh and Wnt signals act at the molecular level to sequentially regulate osteoblast differentiation? Hh signaling has been shown to modify cellular responsiveness to other signals, most notably several members of the Bmp family (Murtaugh et al.,1999). Bmps are implicated in the specification of both chondrocytes and osteoblasts (reviewed by Hoffmann and Gross, 2001), as well as in the subsequent modification of the osteogenic program, where some Bmps promote bone formation [such as Bmp2, Bmp7, Bmp6 and Bmp9(Peng et al., 2003)], although Bmp3 acts as a negative regulator of bone formation(Daluiski et al., 2001). In vivo, Bmp signaling is essential for the normal induction of endochondral anlage, and, both in vivo and in vitro, Bmps can promote an osteoblast program of mesenchymal cell differentiation. One possible role for Ihh signaling is to alter the responsiveness of perichondrial osteoblast progenitors to a Bmp input, modifying the response from chondrocyte inducing to osteoblast inducing. Consistent with this model, Bmp-mediated induction of an osteoblast phenotype in vitro requires prior Ihh signaling(Long et al., 2004), and has been shown to function, at least in part, by induction of a Wnt autocrine loop(Rawadi et al., 2003). In addition, Bmp2 signaling has been shown to upregulate the expression of Sox9 (Healy et al.,1999; Zehentner et al.,1999), a key early determinate of chondrocytes, whereas Sox9 is downregulated in the perichondrial region where osteoblast differentiation initiates (Yamashiro et al., 2004). Whether Sox9 downregulation is Ihh dependent is unclear but, given the ectopic chondrocyte phenotype adopted by perichondrially localized Smo mutant cells, it is reasonable to presume that this, as with normal chondrogenesis, requires high levels of Sox9 activity.

The transcriptional activation domain of Sox9 has also been reported to directly interact with specific armadillo repeats in β-catenin. On the basis of these interactions, it has been proposed that Sox9 inhibitsβ-catenin activity, promoting β-catenin degradation(Akiyama et al., 2004), such that overexpression of Sox9 generates a similar phenotype to loss-of-function of β-catenin. Thus, the balance of Sox9 andβ-catenin may regulate alternate programs of chondrocyte and osteoblast development, respectively. When Sox9 levels are high and β-catenin levels are low, a chondrocyte program may be favored. By contrast, high levels ofβ-catenin and low levels of Sox9 may act on appropriately specified progenitors to promote an osteoblast fate. Interestingly, Prx1-cre-mediated stabilization of β-catenin in skeletal progenitors leads to a dramatic loss of the endochondral skeleton, suggesting that the timing of action of β-catenin (downstream of Ihh in perichondrial osteoblast precursors) may be crucial to its normal osteoblast role (Hill et al., 2005).

Canonical Wnt ligands have been shown to stimulate Runx2expression and Runx2 is itself essential for osteoblast development(Gaur et al., 2005). However,in our study, and in other studies discussed earlier, β-catenin is not essential for the initial activation of Runx2. These findings do not exclude the possibility that the Wnt input may normally regulate either the level or duration of Runx2 expression in the osteoblast lineage. By contrast, Prx1-cre removal of β-catenin suggests that Osx1expression is β-catenin dependent. Analysis of the expression of each gene in osteoblast precursors in Runx2 and Osx1 mutants indicates that Osx1 activation lies downstream of Runx2(Nakashima et al., 2002); thus Runx2, although required in vivo for Osx1 activation, is not sufficient in the absence of β-catenin. Furthermore, Bmp2 activates Osx1 in Runx2 mutant cells(Lee et al., 2003); thus, the molecular hierarchies and interactions underlying Osx1 activation are uncertain, although its crucial role in the specification of all osteoblasts has been clearly demonstrated (Nakashima et al., 2002).

After the initial appearance of a bone matrix, a subset of osteoblasts activates Oc; Oc is considered a late marker of terminal osteoblasts. Oc is itself a direct target of Runx2 regulation, where Runx2 binds a cis-regulatory region within the Oc promoter(Paredes et al., 2004). Our data indicate that when β-catenin levels remain high in the osteoblast,secretion of a bone matrix is promoted, but osteoblasts express only low levels of Oc (Oc^Low, Fig. 9). Thus, a loss of a canonical Wnt input appears to accompany the progression to an Oc^High state. Interestingly, Lef1, aβ-catenin-binding partner and target of canonical Wnt/β-catenin signaling in the osteoblast lineage (Hu et al., 2005), has been shown to inhibit Runx2-mediated activation of the Oc promoter; Lef1- and Runx2-binding sites lie adjacent to each other in the relevant cis-regulatory region within Oc(Kahler and Westendorf, 2003). Thus, the direct integration of distinct regulatory inputs on the Ocpromoter may explain the observed block in Oc activation in our study.

Finally, what is the identity of the postulated Wnt signal(s) mediating the proposed canonical Wnt signaling? Several Wnts have been reported to be expressed in the developing skeletal anlage(Hu et al., 2005). Of particular interest is Wnt7b. Wnt7b is expressed in the bone-forming region downstream of Ihh, providing a potential link between the Ihh and Wnt pathways (Hu et al.,2005). However, although Wnt7b may play a redundant role with another factor, our data indicate that Wnt7b is not essential for osteoblast development. Thus, as with other regions of the embryo where studies of β-catenin action have made a strong case for a canonical Wnt input, most notably in stem cell maintenance in the mammalian gut (reviewed by Pinto and Clevers, 2005), the regulatory ligand(s) remains to be identified.

We would like to thank Dr M. Todd Valerius for providing the pTGCK vector,Drs Tom Carroll, Jing Yu and Jay Rajagopal for Wnt7b skeletal analysis, and Dr Ernestina Schipani for providing in situ probes and advice and support for histological analysis. S.J.R. was supported by fellowships from the National Health and Medical Research Council of Australia (#301299)and the Arthritis Foundation (#401683). Work in A.P.M.'s laboratory was supported by a program grant from the NIH (NIH-NIDDK PO1 #DK56246).