BMP9-regulated angiogenic signaling plays an important role in the osteogenic differentiation of mesenchymal progenitor cells

Mesenchymal stromal progenitor cells (MSCs) are multipotent progenitors which can undergo self-renewal and differentiate into multi-lineages, including osteogenic, chondrogenic, and adipogenic lineages (Deng et al., 2008; Pittenger et al., 1999; Prockop, 1997). Although MSCs have been isolated from numerous tissues, one of the major sources in adults is the bone marrow stromal cells. Osteogenic differentiation is a sequential cascade that recapitulates most, if not all, of the molecular events occurring during embryonic skeletal development (Olsen et al., 2000). Bone morphogenetic proteins (BMPs) play an important role during development (Deng et al., 2008; Luu et al., 2007; Shi and Massagué, 2003) and have been shown to regulate stem cell proliferation and osteogenic differentiation (Varga and Wrana, 2005; Zhang and Li, 2005). BMPs belong to the TGFβ superfamily; and there are at least 14 BMPs in humans and rodents (Deng et al., 2008; Hogan, 1996; Luu et al., 2007; Shi and Massagué, 2003).

We have found that BMP9 is one of the most potent BMPs among the 14 types of BMPs in inducing osteogenic differentiation of MSCs both in vitro and in vivo by regulating several important downstream targets during BMP9-induced osteoblast differentiation (Cheng et al., 2003; Kang et al., 2004; Luo et al., 2004; Luther et al., 2011; Luu et al., 2007; Peng et al., 2003; Peng et al., 2004; Sharff et al., 2009; Tang et al., 2009). BMP9 (also known as growth differentiation factor 2, or GDF-2) was identified in the developing mouse liver (Song et al., 1995). As one of the least studied BMPs, BMP9 has been shown to play some roles in inducing and maintaining the cholinergic phenotype of embryonic basal forebrain cholinergic neurons (López-Coviella et al., 2000), inhibiting hepatic glucose production and inducing the expression of key enzymes of lipid metabolism (Chen et al., 2003), and stimulating hepcidin 1 expression (Truksa et al., 2006).

Osteogenesis usually involves two distinct pathways (Olsen et al., 2000): intramembranous bone formation for the flat bones of the skull and endochondral bone formation for long bones. Endochondral bone formation occurs in close spatial and temporal association and proximity to capillary invasion, so that angiogenesis and osteogenesis must be tightly coupled (Olsen et al., 2000; Wan et al., 2010; Wang et al., 2007). Conflicting results have implicated BMP9 as either an angiogenesis inducer in endothelial cells (Castonguay et al., 2011; Cunha et al., 2010; Mitchell et al., 2010; Park et al., 2012; Scharpfenecker et al., 2007; Suzuki et al., 2010; Yao et al., 2012) or as a potent anti-angiogenic factor (David et al., 2008). Although it is well recognized that osteogenic and angiogenic pathways are well coordinated during bone formation (Wan et al., 2010), it is unclear how these processes are linked in MSCs stimulated by osteogenic factors, such as BMPs.

Here, we investigate whether hypoxia-inducible factor 1α (HIF1α)-mediated angiogenic signaling plays any role in BMP9-regulated osteogenic differentiation of MSCs. HIF1α is a well established regulator of angiogenic cascade, which usually regulates many development processes (Majmundar et al., 2010; Wan et al., 2010). We find that BMP9 directly induces HIF1α expression in MSCs through Smad1/5/8 signaling. Exogenous expression of HIF1α potentiates BMP9-induced osteogenic differentiation of MSCs both in vitro and in vivo, while siRNA-mediated silencing of HIF1α or HIF1α inhibitor CAY10585 profoundly blunts BMP9-induced osteogenic signaling in MSCs. HIF1α expression regulated by cobalt-induced hypoxia recapitulates the synergistic effect between HIF1α and BMP9 in osteogenic differentiation. Mechanistically HIF1α is shown to exert its synergistic effect with BMP9 by inducing both angiogenic signaling (e.g. VEGF) and osteogenic signaling in MSCs. Taken together, our findings should not only expand our understanding of the molecular basis behind BMP9-regulated osteoblastic lineage-specific differentiation, but also provide an opportunity to harness BMP9-induced synergy between osteogenic and angiogenic signaling pathways in regenerative medicine.

Our previous studies have revealed that BMP9 is one of the most osteogenic factors for inducing osteoblastic differentiation in MSCs (Cheng et al., 2003; Deng et al., 2008; Kang et al., 2004; Luther et al., 2011; Luu et al., 2007). However, the molecular mechanisms underlying BMP9 functions in MSCs remain to be fully elucidated. Here, we investigated if HIF1α plays any role in BMP9 osteogenic signaling in MSCs. By using semi-quantitative PCR (sqPCR) analysis, we found that HIF1α was upregulated in MSC line C3H10T1/2 cells at 48 hours after BMP9 transduction (Fig. 1A). Similar results were obtained by western blotting analysis using anti-HIF1α antibody (Fig. 1B). These findings suggest that HIF1α may function as a downstream target of BMP9 signaling.

We next tested if HIF1α is a direct target of BMP9 signaling by analyzing if its genomic promoter can interact with BMP-specific Smad1/5/8. We conducted ChIP analysis using Smad1/5/8 antibody or isotype IgG to pull down genomic DNA from MSCs transduced with BMP9 or GFP. HIF1α promoter-specific primers were shown to yield expected products in BMP9 stimulation- and Smad1/5/8 pulldown-dependent fashion (Fig. 1C). These results suggest that HIF1α expression may be regulated by BMP9 through BMP-specific R-Smad1/5/8.

If HIF1α is an important target of BMP9-mediated osteogenic signaling, we hypothesized that exogenous expression of HIF1α would enhance BMP9-induced osteogenic differentiation of MSCs. We found that endogenous HIF1α expression was almost undetectable in MSCs when cultured under low fetal calf serum conditions (Fig. 2Aa). To effectively introduce exogenous HIF1α in MSCs, we constructed a recombinant adenovirus expressing human HIF1α (AdR-HIFα) and demonstrated that AdR-HIF1α mediated a robust transgene expression in MSCs (Fig. 2Ab), which were effectively transduced by adenoviral vectors (Fig. 2Ac).

While exogenous HIF1α expression alone did not exert any significant effect on osteogenic early marker alkaline phosphatase (ALP) activity, HIF1α was shown to exhibit a profound synergistic effect on BMP9-induced ALP activity in C3H10T1/2 MSCs (Fig. 2Ba). Quantitatively, HIF1α-mediated synergistic effect on ALP activity in BMP9-transduced C3H10T1/2 cells was increased by 79%, 80% and 175% on days 5, 7, and 9, respectively (Fig. 2Bb). A similar synergistic effect was also found in iMEFs as HIF1α was shown to increase ALP activity in BMP9-transduced iMEFs by 20%, 64%, and 165% on days 3, 5, and 7, respectively (Fig. 2Bc). Moreover, we analyzed the effect of HIF1α on late osteogenic markers osteopotin (OPN) and osteocalcin (OCN), and found that exogenous HIF1α expression augmented BMP9-induced expression of OPN and OCN (Fig. 2Ca,Cb). We also demonstrated that HIF1α enhanced matrix mineralization in BMP9-transduced MSCs (Fig. 2Cc). Taking these results together, HIF1α is shown to potentiate BMP9-induced osteoblastic commitment and terminal differentiation of MSCs in vitro.

To further confirm if HIF1α is a crucial mediator of BMP9 osteogenic signaling, we constructed a recombinant adenovirus that expresses a pool of four siRNAs targeting mouse HIF1α coding region using our recently established pSOS system (Luo et al., 2007b), resulting in AdR-simHIF1α, which was shown to effectively silence HIF1α expression in MSCs at as early as 48 hours post infection (Fig. 3A). BMP9-induced ALP activity was significantly decreased in iMEFs by AdR-simHIF1α to 9%, 8%, and 29% of the BMP9 control's on days 3, 5, and 7, respectively (Fig. 3Ba). Qualitatively histochemical staining revealed similar results and the expression of simHIF1α effectively reduced BMP9-induced ALP activity in MSCs (Fig. 3Bb). Furthermore, simHIF1α expression almost completely blunted BMP9-induced matrix mineralization in MSCs as illustrated by Alizarin Red S staining (Fig. 3C). These results strongly suggest that HIF1α may play a crucial role in BMP9-induced osteogenic differentiation of MSCs.

While the above in vitro studies established that HIF1α may play an important role in BMP9-mediated osteogenic signaling, it was imperative to demonstrate if HIF1α played such a role in vivo. Using our previously established stem implantation assay (Chen et al., 2010; Huang et al., 2012a; Huang et al., 2012b; Luo et al., 2010; Zhang et al., 2010), we injected the iMEFs transduced with BMP9 and HIF1α, simHIF1α, or RFP subcutaneously into the flanks of athymic nude mice for 4 weeks. The iMEFs transduced with RFP, HIF1α or simHIF1α alone failed to form any detectable masses (data not shown). We found that HIF1α significantly augmented BMP9-induced bony mass formation whereas simHIF1α inhibited BMP9-induced bone formation (Fig. 4Aa). The gross size differences were further confirmed by iso-surface three-dimensional (3D) analysis of the μCT imaging data (Fig. 4Ab). The hit map analysis of average mineral density revealed that exogenous HIF1α expression increased the average mineral density of the bone masses formed by BMP9-transduced MSCs, and conversely, silencing HIF1α expression significantly decreased the average mineral density (Fig. 4Ac).

The retrieved samples were further subjected to histologic analysis and other special staining. Both H&E staining and trichrome staining revealed that HIF1α significantly enhanced BMP9-induced bone formation and mineralization and that silencing HIF1α inhibited BMP9-induced osteogenesis (Fig. 4Ba,Bb). However, changes in HIF1a expression levels did not affect BMP9's effect on chondrogenesis (Fig. 4Bc). Quantitative analyses indicate that co-expression of BMP9 and HIF1α significantly increased the average thickness of trabeculae and the percentage of trabecular area over total area, whereas knocking down HIF1α expression exhibited an inhibitory effect (Fig. 4Ca,Cb). These in vivo findings are supported by the in vitro studies. Collectively, our results thus far strongly indicate that HIF1α is a crucial mediator of BMP9 osteogenic signaling and that exogenous HIF1α expression augments BMP9-induced osteogenic differentiation of MSCs and produces more mature bone.

We also studied the functional importance of HIF1α in BMP9 signaling by using CAY10585, a novel small molecule inhibitor of HIF1α accumulation and gene transcriptional activity. We found that BMP9 and/or HIF1α-induced ALP activity in C3H10T1/2 cells was inhibited by CAY10585 in a dose-dependent fashion (Fig. 5Aa). The inhibitory effect of CAY10585 on ALP activity was also confirmed by ALP histochemical staining (Fig. 5Ab). Furthermore, CAY10585 was shown to effectively inhibit BMP9 and HIF1a-induced late stage osteogenic differentiation, as demonstrated by Alizarin Red mineralization staining (Fig. 5B). Thus, these results confirm that HIF1α plays a crucial role in BMP9-induced osteogenic differentiation.

HIF1α is frequently induced in hypoxia condition. Cobalt chloride has been widely used a hypoxia inducer. We tested if BMP9-induced osteogenic differentiation would be affected under hypoxia condition. HIF1α was induced in the MSCs treated with cobalt chloride, which could be effectively silenced by AdR-simHIF1α (Fig. 6A). In fact, cobalt chloride treatment in MSCs synergized with BMP9 in ALP induction in a dosage and time-dependent manner (Fig. 6B). The synergistic effect between cobalt and BMP9 on ALP activity (Fig. 6Ca) and matrix mineralization (Fig. 6Bb) could be effectively reduced by siRNAs targeting HIF1α. These results suggest that HIF1α may play a crucial role in mediating the synergistic effect of cobalt-induced hypoxia with BMP9 in MSCs.

HIF1α is an established angiogenic regulator; but its role in MSCs is not well defined. We found that HIF1α in MSCs was able to induce known target genes, such as VEGF and β-catenin (Fig. 7Aa), and late osteogenic marker OPN (to a lesser extent, OCN) (Fig. 7Ab). Expression of these genes was further enhanced by BMP9 in MSCs. We further examined the expression of angiogenic marker von Willebrand factor (vWF) in the ectopic bone masses retrieved from the studies described in Fig. 4, and found that co-expression of BMP9 and HIF1α increased von Willebrand factor expression, which was inhibited by simHIF1α (Fig. 7B). Taking our in vitro and in vivo findings together, our results strongly indicate that BMP9-regulated HIF1α expression not only plays an important role in transducing angiogenic signaling but also functions as a crucial mediator of osteogenic signaling in BMP9-stimulated MSCs (Fig. 7C).

HIF1α is a well established regulator of angiogenic cascade, which usually coordinates with many developmental processes, including skeletal development (Majmundar et al., 2010; Wan et al., 2010). However, it is not clear if HIF1 exerts any effect on BMP9-regulated osteogenic differentiation. In this report, we investigate if HIF1α-mediated angiogenic signaling plays any role in BMP9-regulated osteogenic differentiation of MSCs. We have found that BMP9 upregulates HIF1α expression in MSCs through Smad1/5/8 signaling. Exogenous expression of HIF1α potentiates BMP9-induced osteogenic differentiation of MSCs while silencing HIF1α or HIF1α inhibitor CAY10585 profoundly blocks BMP9-induced osteogenic signaling in MSCs. Furthermore, HIF1α expression regulated by cobalt-induced hypoxia recapitulates the synergistic effect between HIF1α and BMP9 in osteogenic differentiation. Mechanistically HIF1α is shown to exert its synergistic effect with BMP9 by inducing both angiogenic signaling (e.g. VEGF) and osteogenic signaling in MSCs. Thus, our results strongly indicate that HIF1α-induced angiogenic signaling plays a crucial role in BMP9-initiated osteogenic differentiation of MSCs.

Hypoxia-inducible factors (HIFs) regulate target genes that mediate adaptive responses, such as angiogenesis, to reduced oxygen availability (Wan et al., 2010). HIF complex has 1 of 3 α subunits (HIF1α, HIF2α, or HIF3α) bound to the aryl hydrocarbon receptor nuclear translocator (ARNT, also known as HIFβ). The level of HIF1α and HIF2α proteins is regulated by ongoing ubiquitination and proteasomal degradation following enzymatic prolyl hydroxylation on an oxygen-dependent degradation domain (ODD) (Wan et al., 2010). Under normoxic conditions HIF1α undergoes prolyl hydroxylation and is ligated by the E3 ubiquitin ligase von Hippel-Lindau protein (pVHL), and then processed for degradation by the proteosome. Prolyl hydroxylases require oxygen, iron, and 2-oxoglutarate as cofactors (Wan et al., 2010). Under hypoxia, HIF1α prolyl hydroxylation is inhibited, and dimerizes with the ARNT HIF-1, which in turn complexes with coactivator p300 and regulates downstream targets (Wan et al., 2010).

It has been reported that angiogenesis and osteogenesis are well coordinated processes during bone development (Wan et al., 2010; Wang et al., 2007). Osteogenesis usually involves two distinct pathways (Olsen et al., 2000). For intramembranous bone formation, which gives rise to the flat bones of the skull, MSCs differentiate directly into osteoblasts. However, during endochondral bone formation bones are formed through the formation of a chondrocyte anlage, onto which osteoblasts differentiate and deposit bone. Endochondral bone formation occurs in close spatial and temporal association and proximity to capillary invasion, so that angiogenesis and osteogenesis must be tightly coupled (Olsen et al., 2000, Wan et al., 2010).

During osteogenesis MSCs elicit angiogenic signals to recruit new blood vessels into bone. Overexpression of HIF1 in osteoblasts through disruption of the pVHL increases angiogenesis and osteogenesis, which involve the action of VEGF on the endothelial cells (Wan et al., 2010). In fact, mice overexpressing HIF1α in osteoblasts through selective deletion of the Vhl expressed high levels of VEGF and developed extremely dense, heavily vascularized long bones (Wang et al., 2007). Conversely, mice lacking HIF1α in osteoblasts had significantly thinner and less vascularized long bones were than those of controls (Wang et al., 2007). Thus, the activation of HIF1α pathway in developing bone may increase bone modeling events through cell-nonautonomous mechanisms to coordinate the timing, direction, and degree of new blood vessel formation in bone (Wan et al., 2010). Furthermore, HIF1α pathway has been shown as a crucial mediator of neoangiogenesis for skeletal regeneration and bone healing (Wan et al., 2008).

Although it is well recognized that osteogenic and angiogenic pathways are well coordinated during bone formation, it is unclear how these processes are linked in MSCs stimulated osteogenic factors, such as BMPs. We have demonstrated that BMP9 directly upregulates HIF1α expression in MSCs, which in turn induces both osteogenic factors and angiogenic factor VEGF. Thus, potent osteogenic factors, such as BMP9, may induce a tightly-regulated convergence of osteogenic and angiogenic signaling in MSCs, and subsequently lead to efficient bone formation.

HEK-293 and C3H10T1/2 cells were from ATCC (Manassas, VA). The iMEFs were immortalized mouse embryonic fibroblasts as previously described (Huang et al., 2012a). The cell lines were maintained in the conditions as described (Cheng et al., 2003; Luo et al., 2008; Peng et al., 2003; Tang et al., 2009). HIF1α inhibitor CAY10585 was purchased from Cayman Chemical (Ann Arbor, MI), and was prepared in DMSO (as 50 µM stock solution, kept at −80°C). Unless indicated otherwise, all chemicals were purchased from Sigma-Aldrich (St Louis, MO) or Fisher Scientific (Pittsburgh, PA).

Recombinant adenoviruses were generated using AdEasy technology as described (Cheng et al., 2003; He et al., 1998; Kang et al., 2009; Kang et al., 2004; Luo et al., 2007a). The coding regions of human BMP9 and HIF1α were PCR amplified and cloned into an adenoviral shuttle vector and subsequently used to generate recombinant adenoviruses in HEK-293 cells. The siRNA target sites against mouse HIF1α coding region were cloned into the pSES adenoviral shuttle vector (Luo et al., 2007b) to generate recombinant adenoviruses. The resulting adenoviruses were designated as AdBMP9, AdR-HIF1α, or AdR-simHIF1α. AdBMP9 also expresses GFP, whereas AdR-HIF1α and AdR-simHIF1α express RFP as a marker for monitoring infection efficiency. Analogous adenovirus expressing only monomeric RFP (AdRFP) or GFP (AdGFP) were used as controls (He et al., 1998; Luo et al., 2007a; Luo et al., 2004; Peng et al., 2004; Sharff et al., 2009; Si et al., 2006; Tang et al., 2009).

Total RNA was isolated using TRIZOL Reagents (Invitrogen) and used to generate cDNA templates by RT reaction with hexamer and M-MuLV Reverse Transcriptase (New England Biolabs, Ipswich, MA). The first strand cDNA products were further diluted five- to tenfold and used as PCR templates. Semi-quantitative RT-PCR (sqPCR) was carried out as described (Huang et al., 2009; Rastegar et al., 2010; Su et al., 2011; Zhang et al., 2010; Zhu et al., 2009). PCR primers (supplementary material Table S1) were designed by using the Primer3 program to amplify the genes of interest (∼150–180 bp). A touchdown cycling program was as follows: 94°C for 2 minutes for 1 cycle; 92°C for 20 seconds, 68°C for 30 seconds, and 72°C for 12 cycles decreasing 1°C per cycle; and then at 92°C for 20 seconds, 57°C for 30 seconds, and 72°C for 20 seconds for 20–25 cycles, depending on the abundance of a given gene. PCR products were resolved on 1.5% agarose gels. All samples were normalized by the expression level of GAPDH.

Subconfluent C3H10T1/2 cells were infected with AdGFP or AdBMP9. At 30 hours after infection, cells were cross-linked and subjected to ChIP analysis as previously described (Sharff et al., 2009; Si et al., 2006; Tang et al., 2009). Smad1/5/8 antibody (Santa Cruz Biotechnology) or control IgG was used to pull down the protein-DNA complexes. The presence of HIF1α promoter sequence was detected by using two pairs of primers corresponding to mouse HIF1α promoter region.

ALP activity was assessed by a modified Great Escape SEAP Chemiluminescence assay (BD Clontech, Mountain View, CA) and/or histochemical staining assay (using a mixture of 0.1 mg/ml napthol AS-MX phosphate and 0.6 mg/ml Fast Blue BB salt) as described (Chen et al., 2010; Cheng et al., 2003; Kang et al., 2009; Kang et al., 2004; Luo et al., 2004; Luo et al., 2008; Peng et al., 2004; Sharff et al., 2009; Tang et al., 2009; Zhang et al., 2010). For the chemiluminescence assays, each assay condition was performed in triplicate. The results were repeated in at least three independent experiments.

C3H10T1/2 cells and iMEFs were seeded in 24-well cell culture plates and infected with adenoviruses and/or treated with inhibitors. The cells were cultured in the presence of ascorbic acid (50 µg/mL) and β-glycerophosphate (10 mM) for 10–14 days. Mineralized matrix nodules were stained for calcium precipitation by means of Alizarin Red S staining as described previously (Cheng et al., 2003; Kang et al., 2009; Kang et al., 2004; Luo et al., 2004; Luo et al., 2008; Peng et al., 2004; Sharff et al., 2009; Tang et al., 2009).

Western blotting was carried out as previously described (Huang et al., 2009; Rastegar et al., 2010; Su et al., 2011; Zhang et al., 2010; Zhu et al., 2009). Briefly, cells were collected in Lysis Buffer. Cleared total cell lysate was denatured by boiling and resolved by 10% SDS-PAGE. After electrophoretic separation, proteins were transferred to an Immobilon-P membrane. Membrane was blocked with SuperBlock Blocking Buffer, and probed with anti-HIF1α or anti-β-actin (Santa Cruz Biotechnology, Santa Cruz, CA), followed by incubation with a secondary antibody conjugated with horseradish peroxidase. The proteins of interest were detected by using SuperSignal West Pico Chemiluminescent Substrate kit.

The cultured cells were fixed with 10% formalin, washed with PBS, and permeabilized with 1% NP-40. The fixed cells were blocked and incubated with an anti-osteocalcin, or osteopontin antibody (Santa Cruz Biotechnology). After being washed, cells were incubated with biotin-labeled secondary antibody for 30 minutes, followed by incubating cells with streptavidin-HRP conjugate for 20 minutes at room temperature. For von Willebrand factor expression, immunohistochemical staining on paraffin-embedded tissues was also carried out with a von Willebrand factor (vWF) antibody (DaKO, Carpinteria, CA). The presence of the expected protein was visualized by DAB staining and examined under a microscope. Stains with control IgG, were used as negative controls.

The iMEFs were infected with AdBMP9/AdRFP, AdBMP9/AdR-HIF1α, or AdBMP9/AdR-simHIF1α. At 16 hours post infection, cells were harvested, and resuspended in PBS for subcutaneous injection (5×10⁶ cells/injection) into the flanks of athymic nude (nu/nu) mice (5 animals/group, 4-6 weeks old, female, Harlan Sprague-Dawley). At 4 weeks post implantation, animals were sacrificed. Implantation sites were retrieved for μCT analysis, histologic evaluation, and other stains. All specimens were imaged using the μCT component of the GE Triumph (GE Healthcare, Piscataway, NJ, USA) trimodality preclinical imaging system. All image data analysis was performed using Amira 5.3 (Visage Imaging, San Diego, CA); and 3D volumetric data and bone mean density hit maps were obtained as previously described (Chen et al., 2010; Luo et al., 2010; Zhang et al., 2010).

Retrieved tissues were fixed, decalcified in 10% formalin and embedded in paraffin. Serial sections of the embedded specimens were stained with hematoxylin and eosin (H&E). Masson Trichrome and Alcian Blue stains were carried out as previously described (Chen et al., 2010; Kang et al., 2009; Kang et al., 2004; Luo et al., 2010; Luo et al., 2004; Luo et al., 2008; Sharff et al., 2009; Tang et al., 2009; Zhang et al., 2010).

All quantitative experiments were performed in triplicate and/or repeated three times. Data were expressed as mean ± s.d. Statistical significances between vehicle treatment versus drug-treatment were determined by one-way analysis of variance and the Student's t-test. A value of P<0.05 was considered statistically significant.

The authors wish to thank Chad Hanley of the Department of Radiology at The University of Chicago for his assistance and advice on μCT scanning and imaging analysis. The authors declare no conflict of interest.