Bmp2 in osteoblasts of periosteum and trabecular bone links bone formation to vascularization and mesenchymal stem cells

The bone morphogenic protein (BMP) signaling pathway is important for developmental and postnatal physiological processes (Bonilla-Claudio et al., 2012; Dutko and Mullins, 2011; Sieber et al., 2009; Watabe and Miyazono, 2009; Yoon et al., 2005; Zhao et al., 2002). BMP signaling pathways are common for BMP ligands and involve three different type I BMP receptors and two different type II receptors plus negative regulators of BMP signaling, such as noggin and gremlin (Dutko and Mullins, 2011). Selected polymorphisms in the BMP2 gene are associated with increased fracture risk and osteoporosis (Styrkarsdottir et al., 2003). Deletion of other BMP family members Bmp7 and Bmp4 suggests that they play minor roles in bone and cartilage development in the embryo or postnatally, whereas Bmp2 is critical for chondrocyte function and endochondral bone development (Bandyopadhyay et al., 2006; Shu et al., 2011; Tsuji et al., 2008). Removal of Bmp2 in mouse mesenchymal precursor cells in 10 d.p.c. embryos shows a minor role for Bmp2 during endochondral bone formation, but adult mice have small bones that show an increased fracture risk and failure to heal (Tsuji et al., 2006).

Vascularization and angiogenesis are linked to cartilage and bone formation and to fracture healing (Araldi and Schipani, 2010; Jacobsen et al., 2008; Maes et al., 2012; Schipani et al., 2009). During aging or glucocorticoid treatment, bone angiogenesis is reduced and brittleness in bone is increased (Weinstein, 2010; Weinstein et al., 2010). VegfA overexpression in bone marrow leads to formation of new bone and associated blood vessels (Maes et al., 2010a). BMP signaling in osteoblasts increases VegfA through Osterix and Runx2/Hif1a (Deckers et al., 2002; Kwon et al., 2011; Maes et al., 2012; Zhang et al., 2009). Recently, Liu and colleagues (Liu et al., 2012) demonstrated that VegfA is required in the progression of osteoblast precursors (Osterix⁺) to mature osteoblasts.

Mesenchymal stem cells (MSCs) from both the periosteum and bone marrow, are progenitors of osteoblasts (Bianco et al., 2010; Schipani and Kronenberg, 2008) and express CD146 and α-SMA. MSCs are associated with microvascular structures in bone (Crisan et al., 2008; Grcevic et al., 2012; Sacchetti et al., 2007). α-SMA⁺ cells isolated by lineage tracing studies are a class of MSCs that differentiate into osteoblasts (Grcevic et al., 2012). Osterix⁺ preosteoblasts, which represent a non-proliferating transient state, move into bone marrow in association with vascular structures during bone development (Maes et al., 2010b; Park et al., 2012).

Bmp2 expression is activated in MSCs (smooth muscle cells α-SMA⁺) on new induced blood vessels during bone formation and angiogenesis that is activated during distraction osteogenesis (Matsubara et al., 2012). There is first, rapid new blood vessel and associated candidate MSC formation, followed by new bone formation at the distraction site. From the above observations, we hypothesize that the Bmp2 gene has multiple roles at different stages of osteogenesis, ranging from a requirement for terminal differentiation to driving vascularization and formation of new MSCs.

In these studies we: (1) characterized the bone phenotype in a Bmp2-cKO^ob model in both the periosteum and in the bone marrow, including histomorphometry; (2) mechanically tested the long bones to evaluate bone quality; (3) performed in situ hybridization and immunocytochemistry of the key genes and proteins involved in osteoblast and vascular biology; (4) created 3D vascular maps of the vascular bed in both the periosteum and bone marrow; (5) performed in vitro tests for cell autonomy of MSC number and their capacity to differentiate into osteoblasts; and (6) performed in vitro testing of Bmp2 function by deletion of Bmp2 in enriched MSCs.

Fig. 1A and supplementary material Fig. S1C show that Bmp2 expression was reduced by 85–90% in the osteoblasts at 5 days, 1 month and 4 months in periosteum and trabecular bones (Yang et al., 2012). Bmp2 protein levels were also greatly reduced in osteoblasts (Fig. 1B). The body weight of the Bmp2-cKO^ob mouse changed little as assayed to 6 months (supplementary material Fig. S1D, top panel). Humerus length showed a 5% decrease at 2 weeks with no change between control and Bmp2-cKO^ob at 2–4 months (supplementary material Fig. S1D, bottom panel). Alizarin Red and Alcian Blue staining of newborns showed the skeleton was normal in the Bmp2-cKO^ob animals, with a slight decrease in the overall size of the vertebrae compared with heterozygotes or WT controls (supplementary material Fig. S2). The 3.6Col1a1-Cre is active in embryonic tissues that express type 1 collagen (E15–E19) as we and others have observed with Rosa26–loxP–Stop–loxP­–lacZ assays, and this suggests that the Bmp2 gene is minimally required in embryonic Col1a1-expressing tissues for proper development. Another possibility is that maternal Bmp2 crosses the placenta to rescue potential embryo stage bone phenotypes, which would minimize the bone phenotype. Maternal Bmp2 can play a role in embryonic phenotypes, as we have shown (Singh et al., 2008).

Two weeks after birth, Bmp2-cKO^ob mice had thinner bones and smaller unfused vertebrae with a 5% decrease in length compared with controls (WT or Hets) [supplementary material Fig. S1E (bottom right) and Fig. S2 (bottom panels)]. With time, the bone osteopenic-like phenotype, with thinner bones in the Bmp2-cKO^ob became progressively advanced, as assayed by digital radiography up to 1 year of age (Fig. 1C). We focused on males for statistical evaluation, and phenotypes were similar in females and males by X-ray analysis.

As analyzed by dynamic bone formation rate, both mineralizing surface (MS/BS = active surface area) and the mineral apposition rate (MAR, rate of new mineral deposition) were reduced by 20% in both the bone marrow trabeculae (Fig. 1D, top) and in the periosteum (Fig. 1D, bottom), leading to a 40% overall decrease in bone formation rate (BFR). BFR in the periosteum of femurs also showed a 40% decrease in 2-month-old Bmp2-cKO^ob compared with controls (data not shown).

μCT analysis of vertebrae and long bones at 4 months (n = 5 per group) shows vertebrae are smaller by 35% in the Bmp2-cKO^ob with an unfused dorsal region (Fig. 2A). We observed an increase in rod-like quality of trabeculae of the Bmp2-cKO^ob. A 40% reduction in bone volume to total volume (BV/TV) was found in the trabecular region, as well as a 20% decrease in trabecular thickness. The tibia, femur and humerus showed a 30–40% decrease in bone volume compared with controls, but similar cortical thickness (data not shown).

Apparent bone mineral density (aBMD) from 2 weeks to 8 months of age demonstrated a progressive decrease in aBMD with age, especially in the spine (20% decrease in L4–L6 region) at 25 weeks and a 14–16% decrease in tibia and femur of Bmp2-cKO^ob (Fig. 2B). At 6 and 8 months, a similar 20% decrease in the spine aBMD was noted.

By three-point bending of femurs and resulting stress–strain or force–displacement curves, we demonstrated a 40% decrease in stiffness and 50% decrease in work to fracture (W_f) (Fig. 2C) in the Bmp2-cKO^ob bones. Overall toughness of the bones was also decreased 35% in Bmp2-cKO^ob bones (Fig. 2C, top right). Post-yield toughness in Bmp2-cKO^ob bones was decreased by 75% (Fig. 2C, bottom right), demonstrating that the bones are much more brittle.

There were many flat and immature osteoblasts and fewer mature cuboidal polarized osteoblasts in Bmp2-cKO^ob compared with control littermates, as assayed from 5 days to 6 months (Fig. 3A and supplementary material Fig. S3A). By 8 weeks, there is an 80% decrease in the region of mature osteoblasts covering the bone surface and a decreased number of mature osteoblasts per bone surface, OB.S/BS and N.OB/BS (Fig. 3A, bottom left and supplementary material Fig. S3A), respectively. We observed no significant change in the osteoclasts (TRAP+ multinucleated cells, red stain) covering bone surfaces (OC.S/BS) or the number of osteoclasts per unit of bone surface (N.OC/BS) after 2 weeks of age. At 2 weeks, we noted a 10–15% decrease in OC.S/BS and N.OC/BS in the Bmp2-cKO^ob (Fig. 3A, bottom right and supplementary material Fig. S3A, right), respectively.

Osterix, Collagen Type 1a1 (Col1a1) and Osteocalcin expression is reduced by 80% in femurs of 5-day-old Bmp2-cKO^ob (Fig. 3B). Decreased expression of Col1a1 and Osterix mRNA was first noted as early as 5 days, and the difference in expression between control and Bmp2-cKO^ob progressed with age (supplementary material Fig. S3B). There was a 70% decrease in number of osteoblasts positive for phosphorylated Smad1, Smad5 or Smad8 (p-Smad1/5/8) in bones from Bmp2-cKO^ob at 5 days in alveolar bone and at 2 weeks in tibia (supplementary material Fig. S4). In 6-week-old mice, we noted with an anti-phospho-Smad1/5/8 antibody nuclear staining (purple) in osteoblasts of control mice and little cytoplasmic or nuclear localization in the Bmp2-cKO^ob (supplementary material Fig. S5, top panels). Also, 20% of the vascular associated cells around small blood vessels (<30 µm), had a robust nuclear (purple to pink) phospho-Smad1/5/8 signal. Little staining of phospho-Smad1/5/8 was observed in the Bmp2-cKO^ob blood vessels (supplementary material Fig. S5, middle and bottom panels).

Bones of the Bmp2-cKO^ob are less red and show reduced blood vessels as assayed with Col IV immunohistochemistry (Fig. 4A). There are reduced levels of CD31 in blood vessels in the Bmp2-cKO^ob (supplementary material Fig. S6). We performed vascular perfusion (n = 3, control and Bmp2-cKO^ob, 7 months) and μCT 3D mapping of the vascular structure in the metaphysis, diaphysis and periosteum of long bones. We used the BaSO₄ perfusion method as described in the Materials and Methods (David et al., 2009). We observed a 50% decrease in blood vessels in the diaphysis bone region and a 75% decrease in blood vessel volume per trabecular volume in the Bmp2-cKO^ob (Fig. 4B). In the periosteum, we observed a 75% decrease in blood vessel volume to total volume, a 60% decrease in blood vessel volume to bone surface in the Bmp2-cKO^ob and a 75% decrease in blood vessel number per bone surface (Fig. 4C). Blood vessels greater than 20 µm were not affected severely in the Bmp2-cKO^ob. However, in the metaphysis, with the many small capillaries and sinusoidal regions in close association with the bone (less than 50 µm away from bone and less than 15 µm in diameter), we observed a marked 90% decrease in these small-bone-associated microvessels in the Bmp2-cKO^ob (Fig. 4D, right side). A 3D demonstration of the reduced small blood vessels in the Bmp2-cKO^ob and close association of blood vessels with the trabecular bone is shown in supplementary material Movie 1. VegfA mRNA expression was decreased in Bmp2-cKO^ob mice by 60–80% (Fig. 5A). VegfA protein levels in the trabecular region were reduced by ∼70% (Fig. 5B).

α-SMA protein (immunoreactivity) was reduced by 60–70% in the Bmp2-cKO^ob tibia trabecular region and in the cambial layer of the periosteum (Fig. 6A). CD146 or MCAM was reduced by ∼60% in the Bmp2-cKO^ob vascular-associated cells next to the endothelial cells (Fig. 6B, top). Reduced numbers of CD146 cells were observed in the periosteum of 5-day-old and 1-month-old Bmp2-cKO^ob compared with controls (data not shown).

Using α-SMA-Cherry reporter mice, a 70% reduction in α-SMA⁺ cells per unit of marrow volume was observed in the trabecular region. Confocal analysis on the vascular wall demonstrated a reduced number of α-SMA⁺–Cherry cells per unit length of blood vessel (Fig. 6B, bottom). Angiopoietin 1 (Ang-1) levels were reduced by 63% in the bone marrow of tibia Bmp2-cKO^ob compared with controls, as shown in Fig. 6C. Ang-1 is reported to be important in stable interactions between MSCs and endothelial cells (Carmeliet and Jain, 2011).

There is a 90% reduction in the capacity of clonal bone marrow stromal cells (BMSCs) of Bmp2-cKO^ob cells, in vitro, to form mineralized matrix (CFU-OB) compared with cultures from heterozygous or wild-type animals, as measured using the methods described in the Materials and Methods (Chen et al., 2007) (Fig. 6D, right). A 70% reduction in the number of CFU-F clones in the Bmp2-cKO^ob BMSC cultures, an in vitro measure of MSCs, was also noted in the Bmp2-cKO^ob (Fig. 6D, left). There was a significant 30% reduction of CFU-F in the Control-Het, suggesting a threshold effect of Bmp2 on MSC clones. rBmp2 (40 ng/ml) added at initial plating to the CFU assay had no effect on the number of CFU-F clones (data not shown).

Using extracellular matrix, BMSCs maintain their stem cell properties and can be greatly expanded from any BMSC culture. BMSCs were grown for 10 days on XC-marrowECM (StemBioSystems LLC, San Antonio, TX). The cells were released with collagenase type 2 and then replated at high density. Over time, we tested the cell response with 100 ng/ml of VegfA. BMSCs grown on the XC-marrowECM were 88% α-SMA positive (α-SMA⁺) when analysed by FACS (Fig. 7A). VegfA induces Bmp2 mRNA expression at 4, 24 and 48 hours (Fig. 7A, right). The VegfA induction of Bmp2 mRNA was validated by qRT-PCR using a TaqMan assay and FAM-labeled primers (data not shown).

Supplementary material Fig. S7A shows phase-contrast images (top) and the VonKossa/VanGieson mineral-collagen stain images (bottom) for 7 and 10 day cultures. The α-SMA⁺ BMSCs differentiated into well-organized mineral structures with a characteristic sponge-like or honey-comb appearance of the mineral structures as observed in primary calvarial osteoblasts and in 2T3 osteoblast models (Bonewald et al., 2003; Ghosh-Choudhury et al., 1996). Control cultures form these mineral structures and rBmp2 accelerates osteoblast differentiation and increased minerals, as quantified in supplementary material Fig. S7A. Northern gene expression analysis at day 10 shows high levels of Osterix, Osteocalcin, Col1a1 and Bmp2 expression with increased levels in the rBmp2-treated cultures (supplementary material Fig. S7B).

α-SMA⁺ BMSCs were then cultured and expanded from Bmp2^fx/fx mice. These cells are 80–90% α-SMA⁺ as determined by FACS, as described above. The α-SMA⁺ cells were treated with either Adenovirus-5 with CMV–GFP (Ad5–GFP) or Adenovirus-5 with CMV–Cre (Ad5–Cre). After 2 days, the cells are collected and replated at high density to be confluent the next day or two. The cultures were changed to differentiation medium (0 day), plus or minus 100 ng/ml rBmp2. Cultures were collected at 1 and 8 days, for gene expression analysis, and at 8 days for evaluation of mineral quality and quantity. As shown in Fig. 7B, by dark-field and phase-contrast microscopy, and VonKossa/VanGieson stain, the control cultures (Bmp2^flx/flx; Ad5-GFP) formed well-organized mineral structures as defined above, and rBmp2 accelerated this process. However, in α-SMA⁺ BMSCs in which the Bmp2 gene is deleted with Ad5–Cre (Bmp2^flx/flx; Ad5-Cre), few mineral structures formed, and importantly, rBmp2 at 100 ng/ml could not rescue this defect in differentiation. Mineral structures were quantified as shown in Fig. 7B. Gene expression analysis of α-SMA⁺ BMSC cultures, with and without the endogenous Bmp2 gene, was carried out at day 1 and day 8 (Fig. 7C). Expression of the Bmp2 confirms early reduced Bmp2 levels by day 1, with almost complete loss of Bmp2 expression at day 8 (Fig. 7C, top left). Osterix expression at day 1 in Bmp2^flx/flx; Ad5-GFP cells was induced fourfold by rBmp2, whereas Cre-treated cells showed little response to rBmp2 (Fig. 7C). Similarly at day 8, Osterix expression continued to increase and was further stimulated by rBmp2 in Bmp2^flx/flx; Ad5-GFP cells, whereas Bmp2^flx/flx; Ad5-Cre cells had reduced Osterix levels in the basal state and showed little response to rBmp2. Basal Col1a1 expression was reduced in Bmp2^flx/flx; Ad5-Cre cells compared with Bmp2^flx/flx; Ad5-GFP cells at 1 and 8 days. In Bmp2^flx/flx; Ad5-Cre cells at 8 days, there was no rBmp2 increase in Col1a1 expression compared with the Bmp2^flx/flx; Ad5-GFP cells (Fig. 7C). Similar results were seen with Osteocalcin and Dmp1 gene expression patterns, with Bmp2^flx/flx; Ad5-Cre cells having lost most basal expression as well as displaying a much reduced rBmp2 response. VegfA expression changed little between Bmp2^flx/flx; Ad5-GFP and Bmp2^flx/flx; Ad5-Cre cells at day 1, whether or not they were treated with rBmp2. However, at day 8 there was a threefold increase in VegfA expression in the more mature mineralizing cultures. VegfA expression was slightly higher in control Bmp2^flx/flx; Ad5-GFP cells at 8 days than in Bmp2-treated Bmp2^flx/flx; Ad5-GFP cultures because VegfA levels in the more mature osteoblast–osteocyte cultures were decreasing compared with the less differentiated Bmp2^flx/flx; Ad5-GFP 8 day control cultures (data not shown). However, this late stage (8-day) increase in VegfA expression in Bmp2^flx/flx; Ad5-GFP cells was suppressed in the Bmp2^flx/flx; Ad5-Cre cells, suggesting that differentiation state is a critical component of VegfA levels, and the endogenous Bmp2 gene is required for this increased VegfA expression. In support of these in vitro studies, Bmp2 protein immunoreactivity was frequently associated with the small microvascular walls in the metaphysis region in wild-type mice. However, in the Bmp2-cKO^ob mice, vascular-associated Bmp2 protein was greatly reduced (supplementary material Fig. S7C). This suggests that Bmp2 produced by mature osteoblasts induces Bmp2 expression in the vascular associated MSCs in a paracrine manner because the 3.6Col1a1–Cre is not active in vascular-associated candidate MSCs (supplementary material Fig. S1B).

The 3.6Col1a1-cre;Bmp2-cKO^ob model deletes Bmp2 in early osteoblasts and demonstrates a failure of osteoblasts to differentiate to mineralizing state. This was reflected by reduced expression of Osterix, Col1a1 and Osteocalcin and by reduced BMP canonical signaling in vivo. The bones were much more brittle, as assayed by mechanical testing. We discovered a link between osteoblast terminal differentiation and control of the vascular bed and associated MSCs that is controlled by the osteoblast Bmp2 gene. We note a massive decrease in micro-capillaries in bone marrow, decreased CD31, CD146, Angiopoietin 1 and α-SMA levels in vivo in Bmp2-cKO^ob and a decrease in both MSC number and capacity to differentiate with CFU-F and CFU-Ob assays. VegfA levels are reduced in osteoblasts of Bmp2-cKO^ob animals and provide a mechanistic link between osteoblasts and angiogenesis and associated MSCs controlled by the Bmp2 gene. VegfA can activate endogenous Bmp2 production in vascular associated α-SMA⁺ BMSCs in vitro. The proposed mechanism is through activation of the Akt–β-catenin pathway (Maes et al., 2010a) because β-catenin can directly activate transcription of Bmp2 through a TCF–β-catenin response region in the Bmp2 promoter and the Akt pathway has been shown to activate Bmp2 transcription (Ghosh-Choudhury et al., 2002; Zhang et al., 2013).

Recent findings (Liu et al., 2012) support the idea that VegfA can also drive commitment of osteoblast precursors to mature osteoblasts, and suggest that VegfA acts in part by unknown intracrine mechanisms. VegfA overexpression in bone marrow leads to new bone formation as well as massive angiogenesis (Maes et al., 2010b). We suggest that there is a synergistic interaction of VegfA^ob and Bmp2^ob coming from mature osteoblasts that is driving new blood vessel formation and Bmp2^ob is activating Osterix, Col1a1, VegfA and Bmp2 in MSCs. Osterix is known to directly activate many osteoblast differentiation genes, such as Col1a1 and VegfA (Tang et al., 2012), and rBmp2 is known to auto-regulate endogenous Bmp2 expression (Ghosh-Choudhury et al., 1996; Ghosh-Choudhury et al., 2001; Zhao et al., 2002). VegfA plays an intracrine role in driving the MSCs towards the osteoblast lineage (Liu et al., 2012).

A model is presented in Fig. 8 in which osteoblast Bmp2 (Bmp2^ob) induces VegfA from osteoblasts (VegfA^ob) stimulates new vascularization and microvessels (capillaries) with associated MSCs. Angiopoietin 1 plays a key role in integration and stabilization of the MSCs with the vascular–endothelial structures (Carmeliet and Jain, 2011). VegfA^ob and Bmp2^ob from osteoblasts, in turn, activate endogenous Bmp2 expression in these vascular-associated MSCs (Bmp2^msc). The endogenous Bmp2^msc once activated, is then required for progression of the MSCs to osteoblasts and appears to function in some unknown intracrine manner that might allow endogenous Bmp2 to accumulate locally on the plasma membrane, possibly by unconventional secretion mechanisms (USMs) (Nickel, 2003; Askarinam et al., 2013). Bmp2 from osteoblasts could also drive endothelial cells to undergo an endothelial mesenchymal transformation (EndoMT), as happens in pathological heterotopic ossification (HO), and these progenitors cells could then contribute to skeletal ossification as well (Medici et al., 2010; Medici and Olsen, 2012; Lounev et al., 2009).

Our data also suggest that greatly reduced collagen production in the bone matrix in the absence of Bmp2 in osteoblasts leads to bones of poor quality and increased brittleness (Burr, 2002). The combined effects of failure of late stage osteoblast differentiation, leading to poor bone quality, is linked to defects in formation of microvessels closely associated with new bone formation and the home of at least one class of MSC required for formation of bone.

The link between Bmp2 and angiogenesis is, first, supported by elevated generalized BMP signaling in osteoblasts to increase VegfA production and increase bone and blood vessels (Zhang et al., 2009). Second, during distraction osteogenesis, Bmp2 expression is induced in the vascular-associated cells (candidate MSCs) and possibly in the endothelial cells as well (Matsubara et al., 2012). Third, antibody blockage of VegfR1 and VegfR2 leads to much reduced numbers of blood vessels, reduced new bone formation and reduced Bmp2 expression in the distraction osteogenesis model (Jacobsen et al., 2008). Fourth, conditional removal of VegfA in osteoblast precursors (Osterix⁺) leads to greatly reduced bone formation (Liu et al., 2012). Fifth, mechanical-load-induced new bone in the mouse ulna causes an increase in Bmp2 and VegfA expression in vascular associated cells in the periosteum within 1 day (Martinez et al., 2010). Emerging evidence suggests that at least one location for the MSC niche is on the microvessels, and these perivascular-associated MSCs express high levels of α-SMA and CD146 (Sacchetti et al., 2007). Mouse α-SMA⁺ vascular-associated cells show properties of MSCs by lineage tracing studies (Grcevic et al., 2012). These α-SMA⁺ cells, isolated from fat tissue, when placed in an irradiated mouse, can reconstitute the bone marrow and bone and differentiate into osteoblasts (Kalajzic et al., 2008). Our data, at least in vitro, show that α-SMA⁺ BMSC cultures require the endogenous Bmp2 gene to progress to differentiated osteoblasts.

The periosteum of bone controls periosteal bone expansion and is critical for determining bone size and is also an important target for mechanical loading (Allen et al., 2004; Bandyopadhyay et al., 2008). Fatigue loading of bone can induce Bmp2 expression in vascular-associated cells in periosteum (Wohl et al., 2009). Bmp2, specifically in periosteal progenitors, plays a key role in bone healing (Wang et al., 2011). The major phenotype in Bmp2-cKO^ob and in the Prx1–Cre Bmp2 model (Tsuji et al., 2006) is long bones that are all much thinner in cross-sectional diameter, supporting a major role in periosteal function. α-SMA⁺ cells in the periosteum are greatly reduced in the Bmp2-cKO^ob model.

In summary, we demonstrate that the Bmp2 gene in osteoblasts is required in terminal differentiation and proper mineral-matrix formation, in both periosteum and trabecular bone. Ablation of Bmp2 leads to poor quality bone. We also provide evidence that terminal differentiation, driven by Bmp2 in osteoblasts, is linked to angiogenesis and MSCs by both VegfA level and endogenous Bmp2 protein in MSCs. The phenotype of the Bmp2-cKO^ob model in several aspects mimics what happens to bone during aging in mice and humans, such as reduced vascularization and associated MSCs and decreased osteoblast number and differentiation.

All animal experiments were performed according to approved guidelines. Details of production of the Bmp2^fx model are given in supplementary material Fig. S1A. A 5.0 kb KpnI-AccIII 5′ fragment and a 4.5 kb AccIII-SalI 3′ fragment were used. A polII-neo cassette flanked by FRT sites and loxP site, a splicing acceptor site that splices a human alkaline phosphatase expression cassette (Moon et al., 2000) in-frame with coding exon 2, was introduced into an AccIII site in the 3′ untranslated region of exon 3. A 5′ loxP site marked by a PstI site was placed at XhoI site in intron 2. Mice heterozygous for Bmp2 floxed allele with a polII-neo cassette (fn/+, fx with neo cassette) were bred with C57BL/6 Flipper mice containing a β-actin promoter driving flip recombinase (Jackson Labs) to remove the neo cassette (fx/+). The fn/+ (fn is the Bmp2 floxed allele with the neo cassette before removal by the flip recombinase, and both exon 3 and neo cassette are removed at the same time by Meox2-Cre) mice were also bred with Meox2-Cre mice (Jackson Labs) to delete exon 3 and neo cassette and generate recombined null allele. Bmp2 Rec/+ is the same as Bmp2^fx/+ after recombination with Cre from the Meox2 promoter.

Genotypes were determined by Southern blots with 5′ and 3′ external probes (supplementary material Fig. S1A, bottom left). For PCR analyses, primers A and B were used to amplify fragments from wild type (262 bp) and Bmp2 floxed alleles (362 bp for both fn and fx). Primers A and C were used to detect null allele (367 bp) generated by Zhang and Bradley (Zhang and Bradley, 1996). Primers D and F were used to detect flip-dependent deletion of the neo cassette (450 bp for fx, 194 bp for wild type with primers D and E), and primers A and F were used to detect Cre-dependent deletion of the floxed region (450 bp for Rec = Bmp2 floxed allele after Cre recombination) (supplementary material Fig. S1A, bottom right). Primers: A (5′-AGCATGAACCCTCATGTGTTGC-3′); B (5′-CTCGTATCCATCTCATCCTCATTCG-3′); C (5′-GAGACTAGTGAGACGTGCTACT-3′); D (5′-AGGGTTTCAGGTCAGTTTCCG-3′); E (5′-TCCGAAGGTAAGTGTGCTTGG-3′); F (5′-GATGATGAGGTTCTTGGCGG-3′); G (5′-AATGGTTTGTCCTACGCATCG-3′).

Bmp2 null (B2KO is the Bmp2 allele with complete exon 3 deletion globally) heterozygotes were crossed with themselves and wild-type, B2KO/+ and B2KO/KO embryos were evaluated around 8–9 d.p.c. Intercrosses between mice heterozygous for B2Rec (Bmp2 floxed allele recombined after Cre event)/+ were also set up to obtain wild-type, B2Rec/+ (Bmp2^fx-recombined/+ = Bmp2 Hets) and B2Rec/Rec (Bmp2^fx-recombined/Bmp2^fx-recombined = Bmp2 null) embryos, using the MeoxCre mice that delete in the blastocyst. The recombined fx allele (B2Rec) was confirmed with primers A+F. B2KO (Bmp2 allele with permanent global removal of Bmp2 gene)/+ mice for Bmp2 were crossed with B2Rec/+ mice to obtain B2Rec/KO embryos and to compare their morphology with B2KO/KO (standard global Bmp2 KO embryo) and B2Rec/B2Rec (Bmp2-KO^Meox2-Cre) embryos. Data are shown in supplementary material Fig. S1A, far right. These results demonstrate that our Bmp2^fx allele, without the neo cassette, is totally functional since deletion of both Bmp2^fx alleles in homozygous state (Bmp2-KO^Meox2-Cre) at the blastocyst stage gives the same phenotype as the global Bmp2 KO animals, with failure to form a heart or amnionic structures (Zhang and Bradley, 1996).

Only the 3.6Col1a1-Cre allele from the male germline was used for mouse crosses to assure specificity and selectivity (Liu et al., 2004). However, the 3.6Col1a1-Cre is active in a variety of cells expressing type 1 collagen in the embryo but at much lower levels than after birth, when bone formation and Col1a1 expression greatly increases. Deletion of the Bmp2 gene in these embryonic tissues apparently has a minor bone phenotype, as shown in supplementary material Fig. S2. Another possibility is that during embryogenesis, maternal-derived Bmp2 ‘rescues’ these potential phenotypes, as we have shown in global partial Bmp2 KO (when we have the neo cassette in the Bmp2 gene = Bmp2^fn/Bmp2^fn); however, there is rescue of the Bmp2 hypomorphic embryonic defects if the mother has one WT allele (Singh et al., 2008).

The bone-specific expression postnatally with the 3.6Col1a1-Cre model was validated by crossing with Rosa26-loxP-stop-loxP-td-Tomato allele in which the tdTomato reporter (orange-red) expresses when a Cre event occurs [B6.Cg-Gt(Rosa)26Sortm9(CAG-tdTomato)Hze/J]. As shown in supplementary material Fig. S1B, we observed the 3.6Col1a1-Cre (orange-red Tomato+ cells) activity predominantly expressed in the bone-associated osteoblasts at 1 month of age. There are a few dendritic cells in the diaphyseal marrow region that resemble preosteoblasts, as previously described in the 3.6Col1a1–GFP model (Kalajzic et al., 2008). We observed strong Cre activity in several layers of the periosteum and associated tendons (supplementary material Fig. S1B, right). No Cre activity was detected in the vast majority of the bone marrow nor in the muscle or growth plate chondrocytes and importantly, not in or around the blood vessel sinusoidal regions of the metaphysis.

We then crossed Bmp2 floxed allele with 3.6Col1a1-Cre mice to generate the 3.6Col1a1-Cre;Bmp2fx/Bmp2fx or 3.6Col1a1-Cre;Bmp2fx/Bmp2KO, referred to as Bmp2-cKO^ob, to selectively delete the Bmp2 gene in preosteoblasts.

Mice were sacrificed for Alizarin Red and Alcian Blue staining to evaluate ossified tissue and cartilage respectively, at birth and 2 weeks of age, as described (McLeod, 1980).

Bone tissues were collected from Bmp2-cKO^ob and their littermate controls at the regions of vertebrae, femur, tibia and mandible. Bones from the left side were prepared for radiology study, and the right bones were prepared for bone histological evaluation. Bone samples were fixed in RNase-free 4% paraformaldehyde, demineralized in RNase-free 15% EDTA for 6 weeks at 4°C and embedded in paraffin. 8 µm sections were then prepared for Hematoxylin and Eosin (HE) with Orange E, tartrate-resistant acid phosphatase (TRAP) staining, immunohistochemistry and in situ hybridization. Bone samples were fixed in 70% ethanol for 3 days followed by plastic resin embedding without demineralization and sections cut at 8 µm. Bone histomorphometric analysis was carried out on the HE- and TRAP-stained samples using Bioquant (www.bioquant.com). At least two different litters at each age, and their appropriate littermates, were evaluated for most parameters, with multiple sections and photographs of each section for each test, as described in detail in each figure legend.

Male mice aged 2 weeks, 1, 2, 4, 6 and 8 months and 1 year, from over 200 litters, were evaluated by X-ray. Bone samples were processed and stored in 70% ethanol for high resolution X-rays or μCT of individual bones. Digital X-ray images were captured using Faxitron radiographic inspection unit (Model 8050-020, Field Emission Corporation). Bone mineral density (BMD) was evaluated using GE Lunar PIXImus II densitometer. Quantification μCT analysis was carried out for vertebrae and long bones, using a μCT40 Scanco Medical, Zurich, Switzerland (10 µm isometric voxel resolution at 200 msecond exposure, 2000 views and 5 frames per view).

Fluorochrome labeling agents Calcein Green (5 mg/kg) and Alizarin Red (20 mg/kg) were injected peritoneally with a 7-day interval in 1-month-old and 4-month-old male mice. The mice were sacrificed and tissues were collected 3 days after the second injection, followed by the preparation procedure for plastic sectioning as described above. Utilizing Bioquant, MS/BS and MAR were measured and the BFR was then calculated in both the periosteum and trabecular bones using both vertebrae and femurs.

Each femur was placed on lower support points of a three point bending fixture with anterior side down. The span (L) between the lower supports was 8 mm. Next, the hydrated bone was loaded-to-failure at 3.0 mm/minute (DynaMight 8841, Instron, Canton, OH) to generate a force versus displacement (P–d) curve. Mechanical and structural properties, and implied material properties have been determined and defined (Nyman et al., 2011).

Digoxigenin-labeled complementary RNA probes were transcribed from linearized plasmids that encode Bmp2 exon 3, Col1a1, Osterix, Osteocalcin, Dmp1, VegfA and CD146 mouse cDNAs. In situ hybridization was performed following the method previously described (Gluhak-Heinrich et al., 2008; Yang et al., 2012). Probe concentrations were determined to be in the linear range by dot-blot analysis. Numerical estimation was obtained by ImageJ quantification of the blue digoxigenin–antibody complex (Gluhak-Heinrich et al., 2010; Yang et al., 2012). Two independent littermates at a given age were used with 2–3 sections, with multiple photographs, evaluated from each animal and each probe.

Immunohistochemistry was performed as previously described (Gluhak-Heinrich et al., 2010; Yang et al., 2012). The red reactive substrate for alkaline phosphatase on the secondary anti-rabbit IgG antibody was quantified by ImageJ with the appropriate no primary antibody controls as described in detail (Yang et al., 2012; Gluhak-Heinrich et al., 2010). Rabbit anti-mouse Osterix Ab22552 (1∶50), Rabbit anti-mouse Collagen-IV Ab6586 (1∶500), Rabbit anti-mouse Vegf Ab46154 (1∶200), Rabbit anti-mouse CD146 Ab75769 (1∶200) and Bmp2 antibody Ab14933 were obtained from Abcam Limited (Boston, MA). The phospho-Smad1/5/8 antibody was from Cell Signaling Technology (#9511). The relative capillary density was estimated by the number of capillaries multiplied by the average length of capillaries in the defined area, with the ColIV data. The α-SMA antibody Ab5694 (1∶200) and the CD31 antibody Ab28364 (1∶50) were also obtained from Abcam.

BMSCs were collected from the femurs and tibias of 3- to 4-month-old wild-type or Bmp2^fx/fx mice, followed by centrifugation at 1000 r.p.m. at room temperature and then resuspended in Hanks' Balanced salt solution. Cell debris was removed by filtration. Eagle's MEM with Hanks' salts, penicillin (100 U/l), streptomycin (0.1 mg/l) and L-glutamine (PSG) (2 mM) supplemented with 15% FBS was used for BMSC cultures. The BMSCs were expanded on a special extracellular matrix (XC-marrowECM, StemBioSystems) for 10 days, which maintains mesenchymal stem cell properties. Loose and non-adherent cells were removed and cells were then washed twice with PBS and removed with collagenase (Worthington, Type 2, CLS2, 340 U/mg) digestions as described (Chen et al., 2007). From the long bones (femur and tibia) of five mice, we generated over 40–50 million α-SMA⁺ BMSC cells, WT or Bmp2^fx/fx. The cells were plated in six-well plates on plastic at 500,000 cells per well and tested for VegfA regulation of Bmp2 expression in 0.05% FCS.

α-SMA⁺ levels were determined using mouse monoclonal IgG antibody to α-SMA-conjugated to FITC (Sigma, Clone 1A4). Appropriate mouse IgG–FITC was used as a control. Sample-test cells for FACS analysis were first permeabilized with 0.01% Triton X-100 on ice for 10 minutes before adding antibodies. FACS analysis was carried out with the BD Aria FACS analysis instrument (BD Biosciences). The BMSC α-SMA⁺ cells (>80% α-SMA⁺) were placed in 0.05% FCS for 24 hours and then VegfA was added at 100 ng/ml for 4, 24 and 48 hours and Bmp2 expression levels were determined by northern analysis and qRT-PCR using 18S ribosomal as control and TaqMan assay, Applied Biosystem FAM label primers (Rn 18S, Mm03928990 and mBmp2, Mm01340178). Long-term cultures of α-SMA⁺ BMSCs were carried out in 15% FCS αMEM, 50 µM ascorbic acid and 5 mM beta-glycerol-phosphate (differentiation medium), plus and minus rBmp2 at 100 ng/ml. The cultures were stopped at 7 and 10 days to evaluate mineral-matrix production using VonKossa and VanGieson stains (Ghosh-Choudhury et al., 1996), phase-contrast microscopy for mineral and collagen matrix and subsequently RNA extraction and amplification of polyA RNA to ds cDNA and then cRNA, for northern analysis. Von Kossa staining was evaluated and quantified from low magnification (40×) photographs taken under constant setting brightfield illumination. Six independent areas per well, in duplicate wells per test-time were quantified for ‘black’ regions using a constant threshold (dependent on illumination setting of photographs) setting using ImageJ software.

In some experiments, BMSC preparations with the XC-ECM were prepared with bone marrow from Bmp2^fx/fx animals. The cells were again tested for α-SMA by FACS and were over 85% positive. These α-SMA⁺ BMSC^bmp2fx/fx cells were then set up at 50% confluence in 10 cm plates in 2% FCS α-MEM and infected with either Ad5-CMV-GFP or Ad5-CMV-Cre at 10,000 particles per cell (Virus Development Laboratory, Baylor College of Medicine, Houston, TX). After 2 days, over 90% of the GFP-infected cells were GFP⁺. The cells were trypsinized and replated on six-well rat-tail-collagen-coated (1 to 10 dilution of BD Biosciences rat tail collagen in water, for 30 minutes, followed by air drying) plates at 700,000 cells per well and allowed to reach confluence, changed to differentiation medium, plus or minus rBmp2 at 100 ng/ml. Medium was changed every 2–3 days. Samples were stopped at 8 days (most dynamic range for differentiation) for evaluation of mineral structures by VonKossa, VanGieson, phase and dark-field microscopy and stopped at day 1 and day 8 for RNA extraction, amplified polyA RNA to ds cDNA and cRNA, northern analysis using Bmp2 exon 3 probe and full-length coding cDNAs for Osterix, Col1a1, Osteocalcin, Dmp1 and VegfA probes.

The freshly isolated BMSCs were expanded and prepared for CFU assays at clonal densities, as previously described (Chen et al., 2007), using 4-month-old control WT, control Het and Bmp2-cKO^ob mice. After 2 weeks of culture, CFU-F colonies were evaluated using Crystal Violet staining. CFU-OB colonies were evaluated by VonKossa staining after osteogenic induction with the addition of 50 µM ascorbic-acid-2-phosphate, 10 mM β-glycerophosphate and 100 mM dexamethasone. CFU colony assays with or without 40 ng/ml rBmp2 was also performed.

An α-SMA1–Cherry reporter transgenic model was used to monitor vascular associated MSCs in vivo. The transgene consisted of a 2 kb α-SMA (Acta1 gene) promoter region linked to the Cherry reporter cDNA (Grcevic et al., 2012). The bone tissues from 1- and 2-month old α-SMA-Cherry;Bmp2cKO^ob and their littermates, α-SMA-Cherry;Bmp2Het or α-SMA-Cherry;WT, were collected and fixed with 4% paraformaldehyde at 4°C for 1–2 days, followed by demineralization in 15% EDTA (changed every day) for 7–10 days and then placed in 30% sucrose until ready to embed. Tissues were embedded in OCT-Tissue-Tek #4583 (Electron Microscopy Sciences, Hatfield, PA) and sectioned at 8 µm thickness, using the Kawamoto tape system for preservation of bone tissue morphology. The α-SMA⁺ population in vivo was then observed after DAPI staining on the tissue-tape using a confocal Olympus FV 1000 microscope. The visualized images were analyzed using a FV10-ASW 2.1 Viewer (Olympus) and reconstruction of stacks was created with the AutoVisualize component of AutoQuant (Media Cybernetics, Bethesda, MD).

3D vascularization of periosteal and bone marrow regions was carried out using the BaSO₄ method that allows detailed visualization of the microvascular structures in bone marrow and on the periosteum at 2–5 µm resolution (David et al., 2009). Vascularization was assessed with a threshold of 200, whereas bone was at a threshold of 600 by μCT. The vascular maps were independently isolated for the metaphysis region, a volume just below the growth plate that includes the trabecular bone and a volume in the diaphysis region, near the middle of the bone shaft of the femurs and tibias that were analyzed. The periosteal vascular region was carried out in a similar manner, near the metaphysis region. The diameter of the blood vessels were then binned into different size classes and plotted as total blood vessels within a given diameter range.

Student's t-tests and ANOVA were performed to compare data between Bmp2-cKO^ob and their littermate wild-type and heterozygous mice using GraphPad Prism software and Microsoft Excel. P<0.05 was considered as statistical significance. For immunocytochemistry and in situ hybridization, at least two independent litters were evaluated for each probe or antibody with two to three sections, with multiple photographs per section from a given animal. All error bars in the figures represent the s.d.

We thank NIEHS for allowing S. E. Harris to work on producing the Bmp2 floxed mouse model with help of members of Dr Yuji Mishina's lab. This study is dedicated to Gregory R. Mundy.