SOX9 is a major negative regulator of cartilage vascularization, bone marrow formation and endochondral ossification

Replacement of cartilage by bony tissue in the foetal growth plate of long bones, ribs and vertebrae, in a process summarized as endochondral ossification, occurs in a series of distinct chondrocyte differentiation steps, which are tightly regulated in a concerted action by a variety of growth and differentiation factors (Ballock and O'Keefe, 2003; Lefebvre and Smits, 2005; Provot and Schipani, 2005). It begins with growth and differentiation of small hyaline chondrocytes in the epiphysis to rapidly proliferating chondrocytes, followed by alignment of these cells into vertical columns of flattened, lens-shaped chondrocytes. Differentiation of proliferating cells into prehypertrophic and hypertrophic chondrocytes is marked by an up to ten-fold increase in cell volume and development of a granular cell surface with numerous microvilli, which release matrix vesicles required for cartilage mineralization. This step is associated with substantial matrix remodelling: the hyaline cartilage matrix, comprising aggrecan, type II, VI and XI collagen made by resting and proliferating chondrocytes, is substituted by a calcifiying matrix deposited by hypertrophic chondrocytes that produce type X collagen and alkaline phosphatase (Ballock and O'Keefe, 2003; Olsen et al., 2000; Ortega et al., 2004). Further maturation of chondrocytes in the lower hypertrophic zone to terminally differentiated chondrocytes is marked by upregulation of osteopontin (Franzen et al., 1989), Vegfa (Gerber et al., 1999), and Mmp13 (Johansson et al., 1997). Vascular sprouts invade the hypertrophic zone in the diaphysis from the perichondrium, and hypertrophic cartilage is resorbed by osteoclasts. In the growth plate, hypertrophic chondrocytes partially die by apoptosis or survive for some time as posthypertrophic chondrocytes, encapsulated in primary endochondral bone trabeculae (Gebhard et al., 2008).

Differentiation of proliferating chondrocytes to prehypertrophic and hypertrophic chondrocytes is regulated in a complex, synergistic manner by growth factors of the fibroblast growth factor (FGF) and bone morphogenetic protein (BMP) families, by growth hormones including insulin-like growth factors (IGFs), thyroxin and parathyroid hormone-related protein (PTHrP), by Indian hedgehog, which induces PTHrP (Ballock and O'Keefe, 2003; Goldring et al., 2006; Kronenberg, 2006; Lanske et al., 1996; Provot and Schipani, 2005), and by WNT factors (Day and Yang, 2008; Hartmann, 2007). RUNX2, the major transcription factor required for bone formation (Ducy et al., 1997; Komori et al., 1997), plays a key role in endochondral ossification as it promotes not only expression of type X collagen and maturation to hypertrophic chondrocytes (Inada et al., 1999; Kim et al., 1999; Stricker et al., 2002; Takeda et al., 2001), but also induces Vegfa in hypertrophic chondrocytes (Zelzer et al., 2001), which is required for capillary invasion into hypertrophic cartilage (Carlevaro et al., 2000; Gerber et al., 1999; Maes et al., 2004; Zelzer et al., 2002). Furthermore, RUNX2 induces expression of the matrix metalloproteinase Mmp13 in hypertrophic chondrocytes (Hess et al., 2001; Jimenez et al., 1999; Porte et al., 1999; Selvamurugan et al., 2004; Wang et al., 2004), which loosens up the matrix of hypertrophic cartilage in order to allow invasion of bone marrow sprouts (Inada et al., 2004; Johansson et al., 1997; Selvamurugan et al., 2004).

Whereas RUNX2 controls several genes involved in late stages of chondrocyte maturation (Takeda et al., 2001), SOX9 is, together with SOX5 and SOX6, a key regulator for early stages of chondrogenesis in the limb bud and somite mesenchyme, for chondrocyte proliferation and for expression of cartilage matrix genes (de Crombrugghe et al., 2001; Lefebvre and de Crombrugghe, 1998; Lefebvre and Smits, 2005). SOX9 is a transcription factor of the SRY family, regulating sex determination, chondrocyte differentiation and numerous other developmental events. In cartilage development, Sox9 is expressed in all chondroprogenitor cells; it is essential for the formation of cartilage blastema of the limb mesenchyme, for proliferation and differentiation of chondrocytes in the foetal growth plate and for regulation of cartilage-specific genes including Col2a1, Col9a1, Col11a1, aggrecan and others. In the foetal and juvenile growth plates, Sox9 is expressed in resting and proliferating chondrocytes, with a maximum of expression in prehypertropic chondrocytes, but disappears completely from the hypertrophic zone (Zhao et al., 1997) (see also this paper).

The complete absence of Sox9 from hypertrophic chondrocytes suggested to us that Sox9 downregulation might be required to allow the onset of subsequent events of cartilage-bone transition such as induction of angiogenesis, cartilage resorption and formation of bone marrow and endochondral bone trabeculae. In order to test this hypothesis, we generated transgenic mouse lines misexpressing Sox9 in hypertrophic chondrocytes under a strong Col10a1 promoter in the context of a BAC-Col10a1 promoter. Recently, we have shown that expression of transgenes such as lacZ (Gebhard et al., 2007) or Cre (Gebhard et al., 2008) under the control of a BAC-Col10a1 promoter in transgenic mice is efficient and highly specific for hypertrophic chondrocytes. Here, we show that Sox9 misexpression in the hypertrophic zone of the growth plate under this promoter severely suppresses bone marrow formation, cartilage resorption and endochondral ossification in transgenic mice, resulting in reduced bone length. By in situ hybridization and real-time PCR analysis, we demonstrate substantial downregulation of Vegfa, Mmp13 and osteopontin expression in the transgenic growth plate as a result of Sox9 overexpression, indicating that Sox9 overexpression in hypertrophic chondrocytes inhibits their terminal differentiation. Most importantly, we show that SOX9 has the capacity to downregulate Vegfa gene transcription by direct interaction with regulatory SRY elements in the Vegfa gene. These data indicates that SOX9 is a major inhibitor of cartilage vascularization.

The BAC targeting vector pCol10a1-Sox9-Neo was prepared by replacing the lacZ cassette of the placH-Col10A1-Neo vector described previously (Gebhard et al., 2007) with the complete Sox9 cDNA sequence, including a poly A site of bovine growth hormone. Homologous recombination of the pCol10-Sox9-Neo vector into the BAC clone RP23-192A7 [BACPAC Resources Center, Children's Hospital Oakland Research Institute (CHORI), Oakland CA, USA] bearing the complete mouse Col10a1 gene was performed in E. coli according to Lee et al. (Lee et al., 2001) as described previously (Gebhard et al., 2007). BAC-Sox9 clones were tested by PCR (for primer sequences, see Table S1 in the supplementary material) and pulse field gel electrophoresis as described previously (Gebhard et al., 2007). From two BAC-Col10a1-Sox9-Neo clones, DNA was prepared, linearized with PISceI enzyme (NEBiolabs), purified by molecular sieve chromatography (Gebhard et al., 2007) and used for generating transgenic mice. Five Sox9 transgenic founders were obtained that tested positive for the BAC-Col10a1-Sox9 transgene by PCR using primers P1 and PSox9Rev (P2, Fig. 1B; for primers see Table S1 in the supplementary material). Transgene copy number was analyzed by real-time PCR for genomic Col10a1 using primers located in exon 3 and intron 2 as described previously Gebhard, 2008) (see Table S2 in the supplementary material).

For morphological analysis, foetal and postnatal skeletons were freed from adherent tissue, fixed in 95% ethanol and stained for cartilage with Alcian Blue, clarified in KOH and counterstained for bone with Alizarin Red as described previously (Bi et al., 2001). For immunohistochemistry, tissues were fixed in 4% paraformaldehyde at 4°C for 24 hours, decalcified in 0.5 M EDTA from embryonic day 18.5 (E18.5) onwards, dehydrated and embedded in paraffin. Immunohistochemical analysis of nuclear SOX9 with rabbit anti-SOX9 antibody (kindly provided by Dr V. Lefebvre, Cleveland Hospital, OH, USA), followed by peroxidase-labelled anti-rabbit IgG was performed as described previously (Hattori et al., 2008). Type I collagen was stained on decalcified paraffin sections with rabbit anti-rat type I collagen (Abcam ab 21286), but using biotinylated goat anti-rabbit Biotin (Amersham) as secondary antibody, followed by Steptavidin-Phosphatase and Fast Red as a colour detection system. Osteoclast activity was detected by staining for tartrate-resistant acid phosphatase (TRAP) using a leukocyte acid phosphatase kit (Sigma-Aldrich, Poole, UK). For detection of PECAM (CD31), frozen sections were fixed in methanol, pretreated with testicular hyaluronidase (Sigma) and stained with a rat anti-mouse anti-CD31 (Pharmingen), followed by counterstaining with Cy3-labelled anti-Rat IgG.

For in situ hybridization and immunohistochemistry, mouse tissues were fixed in 4% paraformaldehyde at 4°C for 24 hours, decalcified in 0.5 M EDTA from E18.5 onwards, dehydrated and embedded in paraffin. In situ hybridization with digoxigenin-labelled mouse antisense RNA probes using anti-digoxigenin-labelled alkaline phosphatase and BM Purple as a detection system was performed as described previously (Schmidl et al., 2006). RNA probes for mouse Col2a1 and Col10a1 were prepared from the 3′ coding region (Schmidl et al., 2006). The specificity of the probe for mouse Sox9 is described by Zhao (Zhao et al., 1997). Riboprobes for Ihh (Bitgood and McMahon, 1995) and Mmp13 (Yamagiwa et al., 1999) were kindly provided by Dr Vortkamp (University of Essen, Germany); the Runx2 probe was kindly provided by Dr Mundlos (Stricker et al., 2002); and Mmp9 (Reponen et al., 1995) was provided by Dr P. Angel (DKFZ, Heidelberg, Germany). A 387 bp riboprobe hybridizing to exons 1-3 of mouse Vegfa (base #1021-1408, gene #NM 00125250.5) recognizing all Vegfa splice variants and a 677 bp fragment specific for m-chondromodulin were generated by PCR amplification (for primers, see Table S1 in the supplementary material).

Epiphyseal and hypertrophic cartilage was dissected under the binocular from 5-day-old (P5) Sox9-transgenic and wild-type long bones and digested with trypsin and collagenase as described (Surmann-Schmitt et al., 2008). Chondrocytes were cultured in DMEM/F12 containing 5% FCS for 1 day and total RNA was harvested using an RNeasy Kit (Qiagen). Reverse transcription (RT) was performed with 0.5 μg total RNA and the resulting cDNA was amplified in triplicate using the SYBR-Green PCR Assay (TOYOBO SYBR Green PCR Master Mix; TOYOBO, Osaka, Japan or Absolute QPCR SYBR Green Fluorescein, Thermo Scientific), and products were detected with the Light Cycler System (Roche, Basel, Switzerland) or the I-cycler CFX (BioRad). PCR reactions were incubated for 15 minutes at 95°C, followed by 50 amplification cycles of 30 seconds annealing at 60°C, 40 seconds extension at 72°C and 30 seconds denaturation at 95°C. GAPDH, cylophilin A and actin mRNA were used to standardize the total amount of cDNA (for primers, see Table S1 in the supplementary material).

A human VEGFA reporter gene construct covering −1960/+379 bp of the human VEGFA gene (Tischer et al., 1991) was kindly provided by Dr K. Lyons (UCLA, Los Angeles, USA) (Nishida et al., 2009). Various truncated fragments of this reporter gene were prepared by PCR using sequence-specific oligodeoxinucleotide primers (Fig. 6A; see also Table S1 in the supplementary material). COS7 cells were co-transfected with 1 μg of reporter constructs, 100 ng of β-galactosidase expression vector and either 1 μg of Sox9 expression vector or a mock vector; luciferase and β-galactosidase were measured as described previously (Hattori et al., 2006).

³²P-dCTP labelled oligonucleotide probes (33 bp) complementary to the two SRY boxes at −230 bp of the VEGFA gene and corresponding mutant SRY probes were prepared by PCR. Gel shift assays with recombinant Sox9 (Hattori et al., 2008) and antibodies (Hattori et al., 2006) were carried out with 1 ng of poly (dG-dC) and 20 μg of bovine serum albumin (BSA).

Small interference Sox9 RNA were electroporated according to manufacturer's instructions (Amaxa) into mouse primary rib chondrocytes, prepared from newborn mice as described previously (Hattori et al., 2008). The cells were collected 48 hours after electroporation and mRNA or total protein were extracted. The effects of Sox9 siRNA on Sox9 and Vegfa mRNA levels were analyzed by real-time PCR, standardized to GAPDH; changes in SOX9 protein levels were assessed by western blotting, standardized to actin.

Rib chondrocytes prepared from newborn mice were transfected with a vector expressing full-length Sox9 linked to a HaloTag (Promega). DNA was isolated after fragmentation by ultrasonication, and a chromatin immunoprecipitation assay (ChIP) was performed using a resin binding covalently to the HaloTag. DNA fragments precipitated from Sox9-transfected and untransfected cells, as well as total genomic DNA before precipitation, were used as templates for PCR amplification, using primers specific for the Vegfa repressor region and β-actin gene as a negative control, as indicated above. For ChIP of endogenous Sox9, chromatin from primary mouse chondrocytes isolated from the proliferating/resting zone or hypertrophic zone of newborn mouse rib cartilage was prepared and fragmented as described above, whereas ChIP for PCR amplification was done with rabbit anti-SOX9 or control IgG.

MicroCT images of mouse tibiae were acquired on a laboratory cone-beam microCT scanner developed at the Institute of Medical Physics, University of Erlangen-Nuremberg, Germany, for ultra-high resolution imaging (ForBild scanner). It uses a μ-Focus x-ray tube (Hamamatsu) and a 2D cooled CCD detector array (1024×1024 elements, 19 μm pitch; Photometrics, USA) with a dynamic range of 16 bit. For the actual project, the following acquisition parameters were used: voltage, 40 kV, 500 projections; matrix, 1024×1024; voxel size in the reconstructed image, 10 μm, isotropic. The data were processed and analyzed in Amira (Mercury) and MagNan (BioCom). Grey-value images were clipped in the range of 2000 to 10,000 and the isosurfaces were generated by a marching cube algorithm at a threshold of 3000.

For misexpression of Sox9 in hypertrophic chondrocytes, a pCol10a1-Sox9-FRTNeoFRT targeting vector containing full-length Sox9 cDNA was inserted into exon 2 of a BAC-Col10a1 clone by homologous recombination in E. coli as described previously (Gebhard et al., 2007) (Fig. 1A). BAC clones with correct insertions were tested by PCR and restriction mapping, purified by gel chromatography and analyzed by pulse field gel electrophoresis (Gebhard et al., 2008). After injection of linearized and purified BAC DNA into pronuclei of fertilized oocytes, five Sox9 transgenic lines were obtained and verified by PCR (Fig. 1B). Quantitative PCR of genomic Col10a1 revealed that the transgenic founders contained between one and four copies of the BAC-Col10a1-Sox9-Neo transgene (see Table S2 in the supplementary material). All founders, except founder #4779, bearing four transgenic copies were fertile over several generations.

Preparation of skeletal elements revealed that the long bones of Sox9 transgenic newborns were severely deficient of bone marrow (Fig. 1Ca). In all three investigated BAC-Col10a1-Sox9 transgenic lines, phenotypic alterations and gene expression patterns were similar or identical, thus excluding that the observed phenotype could be an artefact caused by random insertion of the BAC transgenes into the genome. Bone marrow formation started to resume a few days after birth (Fig. 1Cb,c). FACS analysis of bone marrow cells revealed a relatively lower content of granulocytes but a higher content of lymphocytes in 3-week-old transgenic versus wild-type animals (data not shown).

A morphological analysis of the transgenic offspring revealed that misexpression of Sox9 in hypertrophic chondrocytes significantly impaired postnatal skeletal growth and bone length. Already at P4, long bones of transgenic animals were significantly shorter that those of their wild-type littermates (Fig. 1C-E). Between P4 and P21, the average bone length of transgenic animals was about 20% reduced as compared with wild-type littermates (Fig. 1D-G). Similarly, the length of the ribs, thorax circumference and entire body length were reduced in transgenic animals, levelling off at about 10% in 10-week-old transgenic animals (data not shown). Furthermore, ribs of transgenic animals frequently showed an odd-shaped curvature (Fig. 1F).

Histological analysis of long bones and ribs by Alcian Blue staining demonstrated that bone marrow invasion and resorption of hypertrophic cartilage starting in the diaphysis of the humerus between E14.5 (data not shown) and E15.5 (Fig. 2Ba,Ca) was retarded in transgenic bones as compared with wild-type bones. At E16.5, the bone marrow space of a transgenic tibia was only half of that of a wild-type littermate (Fig. 2Aa,b). Accordingly, staining for the endothelial marker CD31/PECAM and TRAP staining for osteoclasts demonstrated a delay of vascular invasion from the perichondrium and onset of cartilage resorption in transgenic cartilage models by one day at E15.5 compared with wild-type littermates (Fig. 2Ba-c,Ca,b). The lag in vascularization and resorption became more prominent during subsequent development (Fig. 2Bd,e,Cc,d), leaving long cones of non-resorbed hypertrophic cartilage in the diaphysis of transgenic bones with little space for bone marrow (Fig. 2Ac,d). TRAP staining revealed the appearance of osteoclasts in the diaphysis of both transgenic and wild-type femurs at E15.5 and 18.5 (Fig. 2C), but fewer osteoclasts were observed in transgenic bones owing to reduced bone marrow volume.

In the non-resorbed hypertrophic cartilage tissue, chondrocytes partially lost the columnar arrangement (Fig. 2Ac,Cc,De; see also Fig. S1A in the supplementary material) and became diverse in morphology and gene expression pattern (Fig. 3; Fig. 4). The length of non-resorbed hypertrophic cartilage and occupation of bone marrow space reached a maximum at about P8 (Fig. 2Ae,f). Starting at P8 to P18, depending on the joint, secondary ossification of epiphyseal cartilage was also delayed, as shown in Fig. 2 for a P18 tibia head (Fig. 2Ag,h).

Although in wild-type long bones, spicules of endochondral bone are retained in the diaphysis when growth plates move toward the epiphyses (Fig. 2Ad,f,Df; see also Fig. S1 in the supplementary material), no endochondral bone trabeculae were seen in the Sox9 transgenic bones (Fig. 2Ac,e,De; see also Figs S1, S5 in the supplementary material). Cortical bone was somewhat more disorganized and slightly more thickened in transgenic animals than in wild-type littermates at E19 through P2 (Fig. 2Da-d), but the difference disappeared around P8 (Fig. 2De-h; Fig. 7). Mineralization of cartilage and bone was, however, not impaired by Sox9 misexpression (see Fig. S1 in the supplementary material).

TUNEL staining revealed a strong increase in apoptotic cells in the area of non-resorbed cartilage of transgenic animals in comparison to the low rate of apoptotic cells in the wild-type growth plate (see Fig. S2A in the supplementary material). Conversely, PCNA staining also revealed a significant number of mitotic cells in the non-resorbed cartilage cone, indicating heterogeneity in the cell fate of Sox9-overexpressing hypertrophic chondrocytes (see Fig. S2B in the supplementary material).

In situ hybridization of Sox9 transgenic offspring between E15.5 and P8 confirmed high levels of Sox9 misexpression under the BAC-Col10a1 promoter in hypertrophic chondrocytes, coinciding with the pattern of Col10a1 mRNA as shown in Fig. 3 for E16.5 and E18.5 (Fig. 3Aa,c,Ba,c) and Fig. S2 for newborns (see Fig. S2 in the supplementary material). By contrast, Sox9 mRNA was absent from the hypertrophic zone of wild-type littermates (Fig. 3Ab,Bb; see also Fig. S2 in the supplementary material). In accordance with high Sox9 mRNA expression, high levels of nuclear staining for Sox9 protein were observed in hypertrophic chondrocytes of Sox9 transgenic growth plates (Fig. 4I). In the resting, proliferative and prehypertrophic zones of growth plate cartilage, the distribution of endogenous Sox9 mRNA and protein was similar in transgenic and wild-type littermates (Fig. 3Aa,b,Ba,b), with a maximum in the prehypertrophic chondrocytes. Sox9, as well as Col10a1, continued to be expressed in the non-resorbed hypertrophic chondrocytes (Fig. 3Aa), with decreasing levels towards the tip of the non-resorbed cartilage cone (Fig. 3Ac,Bc; see also Fig. S3 in the supplementary material).

In order to understand the molecular mechanism by which Sox9 misexpression in hypertrophic chondrocytes inhibited vascular invasion, cartilage resorption and formation of bone marrow and trabecular bone, the expression pattern of major genes and factors regulating these events was investigated. In situ hybridization analysis with a mouse Vegfa-specific probe detecting all Vegfa splice variants revealed the absence of Vegfa mRNA from all transgenic hypertrophic chondrocytes, shown in Fig. 3 for a transgenic E16.5 (Fig. 3Ae) and E18.5 tibia (Fig. 3Be), thus explaining the delay in vascular invasion. In wild-type littermates, significant Vegfa expression was seen in lower hypertrophic chondrocytes (Fig. 3Af,Bf). Suppression of Vegfa expression in Sox9 transgenic hypertrophic cartilage was confirmed by quantitative real-time PCR analysis of chondrocyte mRNA, prepared after microdissection and separation of epiphyseal from hypertrophic cartilage of P5 wild-type and transgenic mice (Fig. 5).

In both wild-type and transgenic bones, Vegfa mRNA signals appeared at E18.5 in the hypoxic environment of the proliferating zone and in the presumptive zone of the secondary ossification centre in the epiphysis (Fig. 3Be,f). Sox9-induced impairment of cartilage vascularization was not a result of enhanced expression of chondromodulin, another major anti-angiogenic factor of hyaline cartilage (Shukunami and Hiraki, 2001). In both wild-type and transgenic bones, chondromodulin was expressed in the resting and proliferating zone of the growth plate but was absent from the non-resorbed cartilage of transgenic animals and from wild-type hypertrophic chondrocytes (Fig. 4G,H).

The lack of cartilage resorption suggested impaired activity of Mmp13 in hypertrophic chondrocytes and/or reduced osteoclast activities in the bone marrow. In situ hybridization analysis (Fig. 3; Fig. 4), as well as quantitative PCR analysis (Fig. 5), revealed that Mmp13 was strongly suppressed in transgenic hypertrophic chondrocytes but not in osteoclasts and other bone marrow cells (Fig. 3Ag,Bg; Fig. 4A), whereas it was pronounced in the lower hypertrophic zone of wild-type bones (Fig. 3Ah,Bh; Fig. 4B). Real-time PCR analysis also showed reduced levels of RANKL expression in transgenic hypertrophic chondrocytes (Fig. 5), indicating reduced activation of osteoclasts, which is consistent with the reduced density of osteoclasts seen by TRAP staining in Sox9 transgenic bone marrow (Fig. 2Cc). MMP9, which is expressed mostly by osteoclasts, monocytes and other bone marrow cells, was seen both in wild-type and transgenic bones at the cartilage-bone marrow interface (Fig. 4C,D). The total number of cells expressing Mmp9 was, however, still lower in transgenic cartilage owing to the reduced amount of bone marrow, consistent with lower Mmp9 mRNA levels in transgenic hypertrophic cartilage (Fig. 5).

The lack of Vegfa and Mmp13 expression in the transgenic hypertrophic cartilage indicated that by overexpression of Sox9 under the Col10a1 promoter, terminal differentiation of hypertrophic chondrocytes in the non-resorbed cartilage cone was impaired. This notion was confirmed by analyzing the expression of osteopontin, another marker of terminal differentiated hypertrophic chondrocytes. Osteopontin was absent from transgenic hypertrophic chondrocytes (Fig. 3Ai,Bi) but unaffected in periosteal and trabecular bone. Furthermore, foci of Ihh (Fig. 4E), a marker of prehypertrophic and early hypertrophic chondrocytes (Vortkamp et al., 1996), were seen in the transgenic non-resorbed cartilage (Fig. 4F), indicating that Sox9 overexpression retained chondrocytes in an early hypertrophic state.

As Vegfa, Mmp13 and RANKL are downstream targets of RUNX2 (Jimenez et al., 1999; Hess et al., 2001; Mori et al., 2006; Usui et al., 2008), the Runx2 expression pattern was investigated by in situ hybridization analysis of transgenic and wild-type cartilage. At E16.5, Runx2 expression appeared to be extended in the upper part of the transgenic and wild-type tibia (Fig. 3Ak,Bk) as compared with the wild-type prehypertrophic zone (Fig. 3Al,Bl) but was reduced in the lower hypertrophic zone of transgenic cartilage cone (Fig. 3Ak). In the normal growth plate, Runx2 is most strongly expressed in the prehypertrophic zone (Fig. 3Al,Bl), indicating that Sox9 overexpression causes extension of the prehypertrophic zone towards the diaphysis (Fig. 3Ak,Bk). A quantitative PCR analysis of P5 transgenic and wild-type chondrocytes confirmed that, during further development, relative Runx2 expression levels also declined in transgenic hypertrophic cartilage in comparison with wild-type levels (Fig. 5). The absence of Runx2 in the lower zone of the transgenic cartilage cone might also contribute to the suppression of Vegfa, Mmp13 and RANKL expression in that zone.

We tested the possibility that the lack of Vegfa expression in transgenic hypertrophic cartilage is caused not only by the deficiency of RUNX2, but also by direct transcriptional control by Sox9. DNA sequence analysis of the human VEGFA gene indicated several putative Sox9-binding SRY sites in the first exon, that are also conserved in the mouse and rat Vegfa genes (Fig. 6A). The functionality of these elements was investigated by luciferase reporter gene assays with various VEGFA reporter genes (Fig. 6A). Co-transfection of COS7 cells with Sox9 suppressed the activity of the 1/230 VEGFA reporter gene containing a fragment with two SRY boxes in a dose-dependent manner (Fig. 6Ba,b). The suppressive effect of SOX9 was strongly impaired when a mutation was introduced into the SRY site (Fig. 6Bc). The activity of reporter genes containing 1/379 and −1960/379 VEGFA sequences was lower than that of the 1/230 fragment, suggesting the presence of further silencing elements in the VEGFA promoter, although the longer reporter genes were also inhibited by SOX9 (Fig. 6Bb).

Binding of SOX9 to the regulatory SRY sites in the VEGFA exon 1 was confirmed by electromobility shift assays. SOX9 caused a gelshift of a 33 bp wild-type VEGFA-oligonucleotide spanning both SRY sites at +230 bp and a supershift was observed in the presence of anti-SOX9 (Fig. 6C, lane 5). The gelshift was completely inhibited by a 400-fold molar excess of unlabelled wild-type SRY oligonucleotide, but not by the mutated oligonucleotide (Fig. 6C, lanes 9 and 10).

To confirm the Vegfa-suppressing effect of SOX9 in vivo, mouse primary chondrocytes were treated with Sox9 siRNA and the effect on endogenous Vegfa mRNA levels was measured by real-time PCR. Sox9 siRNA stimulated Vegfa mRNA levels (Fig. 6Da) and it suppressed both endogenous Sox9 mRNA and protein levels (Fig. 6Db,c). Col10a1 expression was not significantly changed (data not shown), indicating that after Sox9 siRNA treatment, the differentiation state of the chondrocytes was not altered.

Evidence for direct binding of Sox9 to the SRY site in the Vegfa promoter in vivo was provided by chromatin immunoprecipitation (ChIP) of fragmented DNA from mouse primary chondrocytes after transfection with HaloTag-labelled Sox9, using primers specific for the Vegfa SRY region (Fig. 6Ea). Also, binding of endogenous SOX9 to the Vegfa SRY site in primary mouse rib chondrocytes prepared from the resting/proliferating zone was seen after ChIP with anti-SOX9, whereas no signal was obtained with primary hypertrophic chondrocytes (Fig. 6Eb; for characterization of hypertrophic and resting mouse rib chondrocytes, see Fig. S5 in the supplementary material). This indicates that the lack of Vegfa expression observed in Sox9 transgenic cartilage might be due to direct suppression by SOX9 binding to the regulatory SRY element in the Vegfa gene, and also possibly indirectly owing to Runx2 downregulation by SOX9.

Interestingly, in postnatal development, the remaining cone of hypertrophic cartilage of Sox9 transgenic animals started to get resorbed from the periphery through the cortical bone shaft rather than from the diaphysis, as occurring in normal endochondral ossification (Fig. 2Ae). Irregular bulges of mesenchymal tissue were protruding from the periosteal bone collar and resorbed the hypertrophic cartilage in a manner similar to the secondary ossification process of the epiphysis (see Fig. S4 in the supplementary material). The invading cells contained osteoclasts, as seen by TRAP staining (see Fig. S4 in the supplementary material), as well as osteoblasts, which deposited osteoids on the resorbed cartilage surface, as shown by in situ hybridization for Col1a1 expression and antibody staining for type I collagen (see Fig. S4 in the supplementary material).

A massive invasion of the bone shaft of Sox9 transgenic animals by bone marrow sprouts was confirmed by microCT analysis of 3-week-old transgenic bones, showing numerous invasion channels protruding the cortical bone shaft, which were not seen in wild-type long bones (Fig. 7A,B). MicroCT images of vertical and horizontal section planes demonstrated a reduction in the diameter of Sox9 transgenic bones and confirmed a substantial deficiency of spongy bone trabeculae in transgenic bones in comparison with wild-type animals (Fig. 7).

Hyaline cartilage is — besides vitreous body and cornea — one of the few avascular tissues in the body. The only cartilaginous tissue that becomes vascularized in a controlled physiological process is the hypertrophic cartilage in growth plates of long bones, ribs and vertebrae. Most strikingly, in foetal cartilage models, this is also the only cartilaginous zone that does not express any Sox9 (Lefebvre et al., 1997; Lefebvre and Smits, 2005; Ng et al., 1997), suggesting to us that Sox9 downregulation in the hypertrophic zone might be a necessary step to allow vascular invasion and endochondral ossification. Overexpression of Sox9 in hypertrophic chondrocytes under a BAC-Col10a1 promoter in transgenic mice confirmed this hypothesis. Vascular invasion of hypertrophic cartilage and bone marrow formation were severely impaired; consequently, cartilage resorption and endochondral bone formation in the foetal and newborn skeleton were substantially retarded and bone growth was reduced. As a result of impaired cartilage resorption, cones of non-resorbed hypertrophic cartilage accumulated in the diaphyses of Sox9 transgenic mice. In situ hybridization and immunohistochemical analysis, as well as microCT imaging, revealed a severe deficiency in trabecular bone formation in Sox9 transgenic animals, whereas cortical bone formation was less affected. Interestingly, 2-3 weeks after birth, transgenic animals developed a rescue mechanism to resorb cartilage and restore bone marrow formation by enhanced invasion of bone marrow sprouts through the cortical bone collar. Thus, transgenic mice grew to adulthood and developed a functional skeleton, although with shorter ribs and extremities than their wild-type littermates.

In searching for the molecular mechanism of this SOX9-induced inhibition of vascularization, we analyzed the expression pattern and protein distribution of several genes involved in endochondral ossification in the developing skeleton of transgenic mice and their wild-type littermates. The striking delay in bone marrow formation in Sox9 transgenic bones suggested a deficiency in VEGFA levels. In cartilage models of long bones, VEGFA is synthesized by hypertrophic chondrocytes (Carlevaro et al., 2000; Gerber et al., 1999) and induces invasion of capillaries, beginning in the diaphysis between days E14 and E15 of embryonic development (Zelzer et al., 2004). Our in situ hybridization analysis demonstrated Vegfa expression in wild-type hypertrophic chondrocytes at E15.5 but the complete absence of Vegfa from hypertrophic cartilage of Sox9 transgenic animals. Accordingly, staining for endothelial cells with anti-CD31 and for osteoclasts with TRAP revealed a strongly delayed invasion of endothelial cells and osteoclasts into the diaphysis at E15.5 in Sox9 misexpressing cartilage models.

Besides Vegfa, the expression of Mmp13 and osteopontin, two further markers of terminal differentiation of hypertrophic chondrocytes, was completely inhibited in the growth cartilage of Sox9 transgenic mice. Previously it was shown that SOX9 is required to prevent conversion of proliferating chondrocytes into hypertrophic chondrocytes (Akiyama et al., 2002). Here we demonstrate that forced expression of Sox9 in hypertrophic chondrocytes under the BAC-Col10a1 promoter is possible, but prevents their differentiation into terminal hypertrophic chondrocytes.

Occasional foci of Ihh expression, as well as continued expression of Col2a1 and Runx2 in the non-resorbed cartilage cone of Sox9 transgenic bones, indicated that some cells could have been arrested or even redifferentiated into the prehypertrophic stage (Vortkamp et al., 1996), although the entire level of Ihh expression in the transgenic growth plate was lower than in the wild-type growth plate. At the same time, numerous dividing cells were observed in the non-resorbed cartilage cone of Sox9 transgenic bones, which is consistent with the ability of SOX9 to stimulate chondrocyte proliferation (Akiyama et al., 2002).

Vegfa expression in hypertrophic chondrocytes is upregulated by Runx2 (Maes et al., 2004; Zelzer et al., 2001) and by HIF1α (Schipani et al., 2001). Because Runx2 expression was downregulated in Sox9 transgenic animals, which is consistent with in vivo and in vitro findings (Akiyama et al., 2002; Zhou et al., 2006), a possible reason for the downregulation of Vegfa in transgenic hypertrophic chondrocytes might therefore be the suppression of Runx2 by SOX9 in the lower hypertrophic cartilage cone. Similarly, suppression of Mmp13 and RANKL, both downstream targets of RUNX2 (Hess et al., 2001; Mori et al., 2006), might be a result of Runx2 downregulation by misexpressed Sox9. In the non-resorbed hypertrophic cartilage cone, however, we still observed residual Runx2 mRNA signals as well as HIF1α protein (our unpublished observation), whereas Vegfa mRNA was completely absent until E18.5, thus explaining the substantial delay of cartilage vascularization after Sox9 overexpression in the hypertrophic zone. Our data indicate that the lack of Vegfa expression seen in transgenic cartilage is not only owing to reduced Runx2 expression, but also a result of direct suppression of Vegfa gene transcription by SOX9. In fact, several putative SOX9-binding SRY elements are located in the promoter and first exon of the mammalian VEGFA gene. Our reporter gene studies focusing on a SRY site located at 230 bp downstream of the transcription start site of the VEGFA gene supported the notion that SOX9 directly suppresses VEGFA reporter gene activity. Direct binding of SOX9 to this SRY site was confirmed by electromobility shift assays in vitro and by ChIP with anti-SOX9 in chondrocytes in situ. Accordingly, siRNA-mediated knockdown of Sox9 mRNA in mouse chondrocytes stimulated Vegfa expression.

This is in apparent conflict with a recent study by Eshkar-Oren (Eshkar-Oren et al., 2009), who provided evidence based on in vivo and in vitro experiments that SOX9 is involved in VEGFA upregulation in condensing limb mesenchyme. However, as VEGFA is regulated by a number of tissue-specific, positive and negative regulatory transcription factors including HIF1, RUNX2 and others, and in light of further putative SOX9 binding sites in the VEGFA promoter, it is possible that SOX9 might enhance or suppress VEGFA expression depending on the cell type, the developmental stage and the context of different co-regulatory factors.

The suppression of Vegfa expression in hypertrophic cartilage of the Sox9 transgenic mice might account for the inhibition of cartilage vascularization in Sox9 transgenic animals, but not for the substantial suppression of cartilage resorption. In a cartilage-specific Vegfa-null mouse (Zelzer et al., 2004), capillary invasion was severely impaired and associated with reduced mineralization and a high rate of cell death. Cartilage resorption was also retarded but not to the same extent as in the Sox9-overexpressing mice. The strong inhibition of cartilage resorption observed in Sox9-overexpressing mice is apparently owing to a deficiency of Mmp13 and Mmp9 and reduced number and activity of osteoclasts. Mmp13-deficient mice also show delayed vascular invasion, a prolonged hypertrophic zone and altered structure of endochondral bone trabeculae (Stickens et al., 2004). Similarly, deletion of MMP9 from cartilage resulted in a delay of cartilage resorption (Vu et al., 1998), whereas inactivation of both Mmp13 and Mmp9 genes caused substantial expansion of the hypertrophic zone, reduced bone marrow cavity, delayed recruitment of osteoblasts and drastically shortened bones (Stickens et al., 2004). However, in these mice, Vegfa expression in hypertrophic cartilage was even enhanced and angiogenesis was normal, therefore the chondrocyte morphology in the growth plate was similar to that of a wild-type bone with a normal cartilage-bone marrow interface and a clear separation between the proliferating and hypertrophic zones.

The substantial delay in cartilage resorption in the Sox9 transgenic growth plate raised the question concerning the fate of the hypertrophic chondrocytes in BAC-Col10a1-Sox transgenic mice with respect to mitotic activity, cell death, gene expression and metabolic activity. The results of the TUNEL staining indicated high, but variable, rates of apoptosis in different joints; this variability is consistent with a study showing substantial differences in the pace and pattern of growth plate turnover in different joints (Wilsman et al., 2008). The enhanced apoptosis seen in transgenic hypertrophic cartilage might be caused by the deficiency of VEGFA, which is not only an angiogenic factor, but also an important growth factor for chondrocyte proliferation in the growth plate (Zelzer et al., 2004). Thus, specific deletion of Vegfa in cartilage caused enhanced chondrocyte apoptosis (Zelzer et al., 2004).

In conclusion, the profound impact of Sox9 overexpression in hypertrophic chondrocytes on the suppression of terminal differentiation by inhibiting Vegfa, Mmp13, RANKL and Opn expression indicates a so far unprecedented role of SOX9 as a major negative regulator of cartilage vascularization, cartilage resorption and formation of trabecular bone in the growth plate. This supports the hypothesis that downregulation of Sox9 in the hypertrophic zone of the growth plate is a necessary event to allow these processes to occur. Our data show that the anti-angiogenic affect of SOX9 is owing to direct transcriptional suppression of Vegfa by SOX9, whereas further effects of Sox9 overexpression might include direct or indirect suppression of Runx2, Mmp13 and RANKL. Furthermore, the overexpression experiment revealed a remarkable ability of the vertebrate skeleton to develop rescue mechanisms to restore bone formation in response to genetic manipulations.

We acknowledge the professional help by Dr D. Mielenz, Division of Molecular Immunology in the FACS analysis of bone marrow cells and by Dr K. Knaup and the IZKF Junior Research Group, both Nikolaus-Fiebiger Center, in HIF1α staining. We thank Prof. M. Takigawa, Okayama University Graduate School, for generous support of this work and Hiroshi Ikegawa and Ayako Ogo for technical assistance. The work was financially supported by the Deutsche Forschungsgemeinschaft (Ma 534-23-1), by a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (#19592145) to T.H. and by the IZKF of the University Hospital Erlangen (Core Unit Z2, A.H.).