Genetic analysis of Runx2 function during intramembranous ossification

Bone is generated by two distinct skeletal ossification processes during embryonic and postnatal skeletogenesis. Bones in the calvaria, such as the frontal and parietal bones, ossify through intramembranous ossification, whereas other bones, such as the long bones and vertebrae ossify through endochondral ossification. In the endochondral ossification process, a cartilaginous template is first formed by condensation of mesenchymal stem cells (MSCs), which then differentiate into chondrocytes. Chondrocytes successively proliferate and then undergo hypertrophy and apoptosis through a well-organized mechanism. Upon apoptosis of chondrocytes, capillaries, osteoclasts, bone marrow cells, and osteoblasts begin to invade the cartilage matrix to produce new bones (Kronenberg, 2003). By contrast, in intramembranous ossification, MSCs undergo differentiation into osteoprogenitor cells, and subsequently proliferate and differentiate into osteoblasts that express type I collagen, osteopontin (Opn; Spp1 – Mouse Genome Informatics) and osteocalcin (also known as bone γ carboxyglutamate protein), and synthesize bone matrix (Hall and Miyake, 1992). Osteoblast differentiation occurs at the osteogenic fronts, where progenitor cells are recruited from the surrounding mesenchyme. The calvarial bones grow as sheets between the brain and epidermis, and a suture is responsible for the maintenance of separation between calvarial bones.

Runt-related transcription factor 2 (Runx2) is a cell-specific member of the Runt transcription factor family, which is considered the master gene of osteoblast differentiation (Ducy et al., 1997; Komori et al., 1997). RUNX2 haploinsufficiency causes cleidocranial dysplasia, an autosomal-dominant human bone disease characterized by defective intramembranous ossification with hypoplastic clavicles, patent fontanelles, and sutures (Mundlos et al., 1997; Otto et al., 1997). We recently showed that osteoblast-specific deletion of Runx2 using α1(I)-collagen-Cre (referred to here as Runx2_col(I)^−/−) mice resulted in no overt abnormalities (Takarada et al., 2013). These results indicated that Runx2 is dispensable in α1(I)-collagen-expressing osteoblasts. Hence, the stage of osteoblast differentiation at which Runx2 is necessary remains to be determined. Less is known about the overall differentiation process and the contribution of Runx2 in intramembranous ossification than during endochondral bone formation despite the simpler process of osteoblast differentiation in intramembranous ossification.

Paired related homeobox 1 (Prx1; Prrx1 – Mouse Genome Informatics) is expressed in MSCs in the limb buds and craniofacial mesenchyme (Cserjesi et al., 1992), and the intermediate filament protein nestin, which was first reported as a marker of neuroectoderm progenitors (Lendahl et al., 1990), is also a marker of MSCs (Méndez-Ferrer et al., 2010). Here, we showed that Runx2 is essential for intramembranous ossification through its expression in cells of the Prx1 lineage but not in cells of the nestin lineage. In addition, we also examined the localization of Prx1-GFP- and nestin-GFP-expressing cell populations in calvaria, and showed that Runx2 is heterogeneously expressed in Prx1-GFP⁺ cells, but not nestin-GFP⁺ cells. Functional in vitro analysis indicated that Prx1-GFP⁺Sca1⁺ cells expressed lower levels of Runx2 and were more primitive in terms of multipotency and self-renewal than were Prx1-GFP⁺Sca1⁻ cells. Moreover, we showed that Runx2 is also required for intramembranous ossification through its expression in cells of the osterix (Osx; Sp7 –­ Mouse Genome Informatics) lineage. Our results demonstrate that the essential period of Runx2 function in intramembranous ossification begins at the Prx1⁺Sca1⁺ mesenchymal stem cell stage and ends at the Osx⁺Prx1⁻Sca1⁻ osteoblast precursor stage.

To identify the cell type in which Runx2 must be activated for intramembranous ossification, we employed two types of Cre drivers, Prx1-Cre mice (Logan et al., 2002) and Nestin-Cre mice, which express Cre recombinase under the control of the 2.4-kb Prx1 promoter or the 5.8-kb promoter and 1.8-kb second intron of the Nes gene (Zimmerman et al., 1994), respectively.

Although Prx1-Cre;Runx2^flox/flox (Runx2_prx1^−/−) embryos were alive at embryonic day (E) 18.5, they died at birth because of breathing difficulties. The perinatal lethality of Runx2_prx1^−/− mice was almost similar to that observed in Runx2^−/− mice (Takarada et al., 2013). By contrast, Runx2_nestin^−/− mice were born at the frequency predicted by the Mendelian principles of inheritance (data not shown). Alizarin Red and Alcian Blue staining of skeletal preparations of E18.5 embryos of Runx2_prx1^−/− and Runx2_nestin^−/− mice was performed (Fig. 1A). Runx2_nestin^−/− embryos had no marked abnormalities in skeletal development, compared with wild-type (WT) embryos. By contrast, Runx2_prx1^−/− embryos showed severe mineralization defects in the calvaria, scapula, humerus, radius, ulna, femur, tibia, fibula and sternum. Moreover, the clavicles of these Runx2_prx1^−/− embryos were extremely hypoplastic (Fig. 1B, arrow). It is important to note that, in contrast to the skeletal phenotypes of Runx2_col(II)^−/− mice (Takarada et al., 2013), Runx2_prx1^−/− embryos lacked bones that are formed through either endochondral or intramembranous ossification (Fig. 1B).

Immunohistochemical analyses revealed that expression of the osteogenic markers Osx and Opn, in addition to Runx2, were absent in the parietal bone of Runx2_prx1^−/− mice (Fig. S1A). In addition, adipogenic (fatty acid binding protein 4) and chondrogenic (type II collagen) markers were not detected in parietal bone of either WT or Runx2_prx1^−/− mice (Fig. S1A). Moreover, in vitro studies revealed that Runx2_prx1^−/− mouse-derived calvarial cells, in which the expression of Runx2, but not Prx1, was decreased (Fig. S1B), had less bone-forming potential than those of WT mice (Fig. S1C).

To reveal the phenotypic differences in Runx2 deficiency between cells of the Prx1 lineage and cells of the nestin lineage, we generated Prx1-Cre;Rosa26-tdTomato mice and Nestin-Cre;Rosa26-tdTomato mice, in which cells expressing Prx1 or nestin and their descendants become Tomato⁺, to track cell fates (Fig. 1C,E). In Prx1-Cre;Rosa26-tdTomato mice, Tomato⁺ cells (cells of the Prx1 lineage) were widely distributed in the parietal bone, including intrasutural mesenchyme, osteogenic front, and bony plate surface, and overlapped with Runx2 at the osteogenic front and the bony plate surface in addition to the intrasutural mesenchyme, and with Osx and Opn at the bony plate surface (Fig. 1D; Fig. S2A,B). Almost all Runx2-, Osx- and Opn-positive cells were Tomato positive (cells of the Prx1 lineage) (Fig. S3A), indicating that almost all osteoblasts were derived from the Prx1 lineage, at least in the parietal bone of newborn mice.

In contrast to cells of the Prx1 lineage, in Nestin-Cre;Rosa26-tdTomato mice, Tomato⁺ cells (cells of the nestin lineage) were found in the parietal bone, and were completely separated from Runx2-expressing cells at postnatal day (P) 1 (Fig. 1F), P10 and 12 weeks old (Fig. S4A). Moreover, Runx2 mRNA expression was comparable in WT and Runx2_nestin^−/− mouse-derived calvarial cells (Fig. S4B), and the intensity of immunostaining for Runx2 in parietal bone of adult mice was also comparable in WT and Runx2_nestin^−/− mice (Fig. S4C). Moreover, Tomato⁺ cells (cells of the nestin lineage) did not express Osx or Opn in the parietal bone at P1 (Fig. S2C,D) and P10 (Fig. S4A), suggesting that the Nestin-Cre driver used in this study does not target osteoblasts in calvaria, which supports the lack of a skeletal phenotype at E18.5 and the lack of abnormalities in calvaria of 6-month-old Runx2_nestin^−/− mice (Fig. 1A,B; Fig. S4D).

In contrast to parietal bone, in Prx1-Cre;Rosa26-tdTomato mice, Tomato⁺ cells (cells of the Prx1 lineage) were scattered in the frontal bone (Fig. S5A,B), but some overlapped with Osx and Opn at the surface of the bony plate (Fig. S5C). The fact that the recombination efficiency of the Prx1-Cre driver was reduced in frontal bone, compared with that in parietal bone, might lead to the phenotypic differences in mineralization between parietal and frontal bone in Runx2_prx1^−/− mice (Fig. 1B). Moreover, this could possibly explain why Runx2_prx1^−/− mice had a greater extent of mineralization in frontal bone, compared with that of Runx2^−/− mice. As it has been reported that the entire frontal bone is derived from the cranial neural crest, whereas other skull bones principally originate from the paraxial mesoderm (Chai and Maxson, 2006), Wnt1-Cre, which is a known reporter of neural crest origin (Jiang et al., 2002; Yoshida et al., 2008), could be used to determine whether Runx2 affects frontal bone formation. Taken together, these results revealed that almost all osteoblasts are from the Prx1 lineage, at least in parietal bone of newborn mice, and Runx2 is essential for intramembranous ossification through its expression in cells of the Prx1 lineage, but not in cells of the nestin lineage.

Next, we assessed Runx2 expression in the calvaria of transgenic mice expressing green fluorescent protein (GFP) under control of the 2.4-kb Prx1 promoter (Prx1-GFP⁺) (Kawanami et al., 2009) or the 5.8-kb promoter and 1.8-kb second intron of the Nes gene (nestin-GFP⁺) (Mignone et al., 2004; Ono et al., 2014). Whole-mount and histological analyses revealed the presence of Prx1-GFP⁺ cells in the calvarial suture region (Fig. 2A). Immunostaining showed that Runx2 was highly expressed in Prx1-GFP⁻ cells at the osteogenic front and the bony plate surface (mineralized bone), in addition to Prx1-GFP⁺ cells at the intrasutural mesenchyme in parietal bone (Fig. 2B; Fig. S6A). As shown in Fig. S3B, 91.4% of Prx1-GFP⁺ cells were positive for Runx2. Among Prx1-GFP⁺ cells in the intrasutural mesenchyme, at least two cell types appear to be present: Prx1-GFP⁺ cells expressing Runx2 at higher levels (Runx2^hiPrx1⁺ cells; Fig. S7A, arrows) or at lower levels (Runx2^lowPrx1⁺ cells; Fig. S7A, arrowheads). Moreover, heterogeneous Runx2 expression in Prx1-GFP⁺ cells was defined when a reference nuclear marker, Hoechst33342, was used as the standard (Fig. S7B).

To perform a direct comparison of Prx1-GFP⁺ cells and Prx1-Cre-targeted cells in the parietal bone of newborn mice, we generated triple-transgenic mice carrying Prx1-GFP, Prx1-Cre and a Rosa26-tdTomato reporter (Prx1-GFP;Prx1-Cre;Rosa26-tdTomato mice) (Fig. S8). These mice revealed that all Prx1-GFP⁺ cells overlapped with Prx1-Cre-targeted cells. Moreover, together with the results shown in Fig. S2A,B and Fig. S3A, almost all Osx⁺ and Opn⁺ osteoblasts appear to be Tomato⁺Prx1-GFP⁻ cells. In addition, almost all Prx1-GFP⁺ cells expressed Runx2 (Fig. 2B; Fig. S3B) and all Runx2-expressing cells were Prx1-Cre-targeted cells (Fig. 1D; Fig. S3A). Although further experiments should be performed to reveal the cell types of Tomato⁺Prx1-GFP⁻Osx⁻Opn⁻ cells, these results indicate that Prx1-Cre could target and encompass all populations with the Prx1-GFP allele, and that Prx1⁺ cells could give rise to the calvarial osteoblasts expressing Runx2, Osx and Opn in vivo.

In sharp contrast to Prx1-GFP⁺ cells, nestin-GFP⁺ cells did not express Runx2 protein in the parietal bone of the calvaria (Fig. 2C; Fig. S3B; Fig. S6B) along with markedly lower Runx2 mRNA expression in GFP cells sorted from calvaria of Nestin-GFP mice compared with that of WT mice-derived calvarial cells (Fig. S9A). Together with the normal skeletal phenotype of Runx2_nestin^−/− mice, the present studies imply that nestin⁺ cells might not be required for generation of bone-forming osteoblasts in the calvaria. However, a recent paper by Ono et al. clearly revealed that vasculature-associated cells expressing nestin-GFP encompass early osteoblasts in endochondral ossification (Ono et al., 2014). To delineate the difference between cell types (nestin-GFP⁺ cells and Nestin-Cre-targeted cells) that are differentially targeted in the parietal bone of newborn mice, therefore, we generated triple-transgenic mice carrying Nestin-GFP, Nestin-Cre and a Rosa26-tdTomato reporter (Nestin-GFP;Nestin-Cre;Rosa26-tdTomato mice) (Fig. S9B). These mice revealed that almost all cells targeted by Nestin-Cre were positive for nestin-GFP, whereas nestin-GFP⁺ cells partially overlapped with Nestin-Cre-targeted cells (50.4±10.2%), suggesting that Nestin-Cre preferentially targets a subpopulation of nestin-GFP⁺ cells in the parietal bone of the calvaria (Fig. S9C). These results might indicate problems with one of the Nes promoters, at least in calvarial cells. However, because immunohistochemical analyses revealed that Runx2 was not expressed in both nestin-GFP⁺ cells and Nestin-Cre-targeted cells in the parietal bone of newborn mice, nestin⁺ cells might not be required for generating bone-forming osteoblasts in the calvaria under the experimental conditions (including the mouse line) used in this study.

Because Runx2 appears to be heterogeneously expressed in Prx1-GFP⁺ cells, as shown in Fig. 2B and Fig. S7, and Prx1-Cre can target all populations of cells with the Prx1-GFP allele (Fig. S8), we next classified the populations of Prx1-GFP⁺ cells according to the expression of various MSC surface markers. Cells were isolated from the parietal and frontal bone, including coronal and sagittal suture, and then subjected to flow cytometric analysis. To investigate the size of the fraction of cells coming from the differentiated calvaria and suture region, we used α1(I)-collagen-Cre;Rosa26-tdTomato mice, in which cells expressing α1(I)-collagen and their descendants become Tomato⁺, and Prx1-GFP mice, in which GFP⁺ cells are localized at the intrasutural mesenchyme but not at the osteogenic front and the bony plate surface in the calvaria. Among the total cells isolated from α1(I)-collagen-Cre;Rosa26-tdTomato mice, 10.1% were Tomato⁺ cells (Fig. S10A), whereas among the total cells isolated from Prx1-GFP mice, 11.5% were GFP⁺ cells (Fig. S10B). These results indicate that ∼10% of isolated total cells were from differentiated calvaria and the suture region in our experimental conditions.

In Prx1-GFP⁺ cells of the calvaria, CD29, CD49e and CD51 (Itgb1, Itga5 and Itgav, respectively – Mouse Genome Informatics) were highly expressed, whereas platelet-derived growth factor receptor α (PDGFRα) and CD105, CD90, CD61 and stem cell antigen-1 (Sca1) (Eng, Thy1, Itgb3 and Ly6a, respectively – Mouse Genome Informatics) were significantly expressed to a lesser extent. As expected, the perivascular marker PDGFRβ, the endothelial marker CD31 (Pecam1– Mouse Genome Informatics), and the hematopoietic markers CD45 and Ter119 (Ptprc and Ly76, respectively – Mouse Genome Informatics) were not expressed by Prx1-GFP⁺ cells (Fig. 3A; Fig. S11A). Recent studies revealed that mouse MSCs in the bone marrow are highly concentrated in populations of cells double-positive PDGFRα and Sca1 (PαS) (Morikawa et al., 2009). Flow cytometric analysis revealed that PαS cells comprised a small subset of Prx1-GFP⁺ cells in the calvaria, and 8.4% were Prx1-GFP⁺ cells among PαS cells (Fig. 3A; Fig. S11A).

Because Sca1 alone (10.1±5.3%) was present in a similar percentage of Prx1-GFP⁺ cells as were PαS cells (11.6±5.3%), as shown in Fig. 3A, we used only Sca1 antibody to distinguish cells with MSCs properties among Prx1-GFP⁺ cells. Prx1-GFP⁺Sca1⁺ populations highly expressed CD51, CD61 and CD90, in addition to PDGFRα (97.7±1.9%), indicating that almost all Prx1-GFP⁺Sca1⁺ cells should be Prx1-GFP⁺PαS cells with MSC properties. By contrast, Prx1-GFP⁺Sca1⁻ populations displayed lower expression levels of CD61, CD90, CD105 and PDGFRα (36.7±11.1%) (Fig. 3B; Fig. S11B). These results clearly show that calvarial Prx1⁺Sca1⁺ cells are a more homogeneous population than Prx1⁺Sca1⁻ cells in terms of the expression profiles of conventional MSC surface markers.

We next determined Runx2 expression levels as well as the self-renewal and multi-lineage differentiation potential of Prx1⁺Sca1⁺ and Prx1⁺Sca1⁻ cells. A colony-forming unit-fibroblast assay revealed that the colony-forming capacity of Prx1-GFP⁺Sca1⁺ cells was greater than that of Prx1-GFP⁺Sca1⁻ cells (5.8% versus 0.14%, respectively; Fig. 3C). Furthermore, Prx1-GFP⁺Sca1⁺ cells robustly differentiated into osteoblasts and adipocytes, whereas Prx1-GFP⁺Sca1⁻ cells differentiated into osteoblasts, but not adipocytes (Fig. 3D). Among total calvarial stromal cells, Sca1 expression was clearly inversely proportional to Runx2 expression (Fig. 3E). Moreover, immunohistochemical analyses revealed that Sca1 expression was observed in periosteum and intrasutural mesenchyme (Fig. S12A,B). Prx1-GFP⁺Sca1⁺ cells were observed in a small part of intrasutural mesenchyme and expressed Runx2 at a lower level compared with that in Prx1-GFP⁺Sca1⁻ cells (Fig. S12C). In addition, mRNA expression of Runx2 and MSC markers (Itgb3 and Thy1) was significantly higher (Runx2) and lower (Itgb3 and Thy1) in Prx1-GFP⁺Sca1⁻ cells than in Prx1-GFP⁺Sca1⁺ cells (Fig. 3F; Fig. S13), supporting the notion of heterogeneous Runx2 expression in Prx1-GFP⁺ cells as indicated by the data shown in Fig. S7.

These results suggest that multipotent self-renewing cells were present in the Prx1⁺Sca1⁺ subsets with lower Runx2 expression, rather than in the Prx1⁺Sca1⁻ cells with higher Runx2 expression, in the calvaria (Fig. 3G).

Osx is another transcriptional determinant of osteoblast lineages expressed by osteoblast precursors downstream of Runx2 (Nakashima et al., 2002). Osx⁺ osteoblast precursors differentiate further into mature osteoblasts that produce α1(I)-collagen (Maes et al., 2010). Runx2_prx1^−/− mice were deficient in intramembranous ossification in parietal bone, whereas Runx2_col(I)^−/− mice showed no skeletal abnormalities. In addition, as shown in Fig. 2B and Fig. S6A, Runx2 was strongly expressed in Prx1-GFP⁻ cells at the osteogenic front and the bony plate surface in the calvaria. Moreover, Runx2 mRNA expression was markedly higher in calvarial osteoblasts compared with that in Prx1-GFP⁺ cells (Fig. S9A). We therefore investigated whether Runx2 in cells of the Osx lineage played a role in intramembranous ossification. In parietal bone of Osx-Cre transgenic mice expressing the GFP::Cre fusion protein, almost all Osx-GFP⁺ cells were positioned at the osteogenic front and the surface of mineralized bone (Fig. 4A), in which Prx1-GFP⁺ cells are absent and Runx2 is highly expressed (Fig. 2B; Fig. S7). Moreover, immunohistochemical analyses revealed that Osx was highly expressed in Prx1-GFP⁻ cells at the bony plate surface (mineralized bone) (Fig. 4B), suggesting that Osx is highly expressed by Prx1⁻Runx2⁺ cells in parietal bone. Although Osx-Cre;Runx2^flox/flox (Runx2_osx^−/−) embryos were alive at E18.5, Runx2_osx^−/− mice died at birth, similar to Runx2^−/− mice, Runx2_col(II)^−/− mice (Takarada et al., 2013) and Runx2_prx1^−/− mice. Skeletal analysis of E18.5 Runx2_osx^−/− mice revealed severe mineralization defects in the calvaria, scapula, humerus, radius, ulna, femur, tibia, fibula, sternum and clavicle (Fig. 4C,D). These results clearly revealed that Runx2 is essential for intramembranous ossification through its expression in cells of the Osx lineage.

Our results suggest that osteoblast differentiation in the calvaria begins at the Prx1⁺Sca1⁺ MSC stage with sequential progression to Prx1⁺Sca1⁻ cells, then Osx⁺Prx1⁻Sca1⁻ osteoblast precursors, which eventually form mature α1(I)-collagen⁺ osteoblasts. Furthermore, during the stepwise differentiation of Prx1⁺Sca1⁺ MSCs into mature α1(I)-collagen⁺ osteoblasts, the essential period of Runx2 function in intramembranous ossification probably begins at the Prx1⁺Sca1⁺ MSC stage and ends at the Osx⁺Prx1⁻Sca1⁻ osteoblast precursor stage [before the mature α1(I)-collagen⁺ osteoblasts appear] (Fig. 4E). Although intensive studies have been conducted to identify the functional importance of Runx2 in osteoblast differentiation, to the best of our knowledge, this is the first report showing direct evidence of the cellular origin and essential period of Runx2 function during intramembranous ossification.

Prx1-Cre, Nestin-Cre, Osx-Cre and Rosa26-tdTomato mice were from Jackson laboratory [B6.Cg-Tg(Prrx1-cre)1Cjt/J, B6.Cg-Tg(Nes-cre)1Kln/J, B6.Cg-Tg(Sp7-tTA,tetO-EGFP/cre)1Amc/J and B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J, respectively]. Runx2 conditional knockout mice lacking exon 4 of the Runx2 gene (Runx2^flox/+) were bred in our animal facility. Prx1-GFP mice, Nestin-GFP mice and α1(I)-collagen-Cre mice were provided by Dr Murakami (Case Western Reserve University, OH, USA), Dr Enikolopov (Cold Spring Harbor Laboratory, NY, USA) and Dr Karsenty (Columbia University, NY, USA), respectively. The protocol employed here meets the guideline of the Japanese Society for Pharmacology and was approved by the Committee for Ethical Use of Experimental Animals at Kanazawa University. The numbers of samples and sections used per experiment are stated in the figure legends.

Skeletal double staining with Alcian Blue and Alizarin Red was performed as described previously (Takahata et al., 2012). Sections were stained with Hematoxylin and Eosin (H&E), von Kossa and Alcian Blue stains.

Neonatal mouse calvaria were fixed with 4% paraformaldehyde, followed by immersion in 30% sucrose. Calvaria were then embedded in OCT and cryosectioned at 14 μm (Leica CM 3050), followed by immunostaining with specific antibodies (Table S1). Confocal images were acquired using LSM710 and Zen2009 software (Zeiss).

Neonatal parietal and frontal calvaria (including coronal and sagittal suture) were dissected and washed several times in PBS, followed by treatment with 125 mM EDTA for 10 min (twice) for decalcification. Decalcified calvaria were next treated with 0.2% collagenase (Wako) in PBS at 37°C for 20 min. No cells were collected from calvaria by further collagenase digestion. Calvarial cells were collected by centrifugation (160 g), followed by soaking in water for 5 to 10 s in order to burst the blood cells. Finally, cells were suspended in 2% fetal bovine serum (FBS) in PBS for the staining with fluorescence-labeled antibodies (Table S1). Flow cytometric analysis was carried out using the FACS Verse (BD Biosciences). Data were analyzed by FACS Suite software (BD Biosciences). Cell sorting experiments were performed using an Aria Cell sorter (BD Biosciences).

For the CFU-F assay, 1500 or 3000 sorted cells were plated in 90-mm dishes (Nunc) in DMEM (Wako) supplemented with 20% FBS (HyClone), and incubated for 10 days at 37°C under 5% CO₂. CFU-F colonies were counted after Crystal Violet staining. Osteoblastic differentiation was induced by osteogenic inducer (L-ascorbic acid and β-glycerophosphate), and subsequent alkaline phosphatase staining. Adipogenic differentiation was performed using adipogenic inducers (dexamethasone, pioglitazone, insulin and 3-isobutyl-1-methylxanthine) for the first 2 days, followed by culture with pioglitazone and insulin for a further 14 days, and subsequent Oil Red O staining.

Sorted cells were lysed in Buffer RLT (QIAGEN), and RNA isolation was performed using the RNeasy Mini Kit (QIAGEN). Quantitative real-time PCR was performed using specific primers. The sequences used as primers were as follows: Runx2, 5′-CCTAGTTAGAGTGGTAGCAGAAGC-3′, 5′-ACAGACAACGAAGAAAGTTCCCAC-3′; Itgb3, 5′-CCACACGAGGCGTGAACTC-3′, 5′-CTTCAGGTTACATCGGGGTGA-3′; Thy1, 5′-CCAACCAGCCCTATATCAAGGT-3′, 5′-TGAAGCTCACAAAAGTAGTCGC-3′; Actb, 5′-GGCTCCTAGCACCATGAAGATCAAG-3′, 5′-ATCTGCTGGAAGGTGGACAGTGAG-3′. The relative mRNA expression was determined using the ΔCt method. The gene expression was normalized to Actb.

Primary osteoblasts were prepared from the mouse calvariae at E18.5 by the sequential enzymatic digestion method, as previously described (Takarada et al., 2013). A differentiation-inducing mixture containing 50 μg/ml ascorbic acid and 5 mM β-glycerophosphate was used for the culture of primary osteoblasts. For Alizarin Red and Van Gieson staining, cultured cells were stained with Alizarin Red S (Sigma) or Weigert's iron Hematoxylin (Sigma)/Van Gieson's solution (Wako), respectively.

Quantitative data are expressed as mean±s.e.m. Statistical significance of differences was determined using Student's t-test (Fig. 3F; Fig. S1B; Fig. S4B; Fig. S9A; Fig. S13). Differences were considered significant at P<0.05.

We are very grateful to Drs S. Murakami (Case Western Reserve University, OH, USA) and G. Enikolopov (Cold Spring Harbor Laboratory, NY, USA) for generously providing Prx1-GFP mice and Nestin-GFP mice, respectively. We also thank Dr Gerard Karsenty for critical reading of the manuscript.