Lack of CD47 Impairs Bone Cell Differentiation and Results in an Osteopenic Phenotype in Vivo due to Impaired Signal Regulatory Protein α (SIRPα) Signaling*

Background: CD47 and its receptor SIRPα are suggested to regulate bone metabolism. Results: Lack of CD47 prevents stromal cell SIRPα signaling, which impairs bone marrow stromal cell differentiation, subsequent osteoclast differentiation, and bone homeostasis. Conclusion: CD47 and SIRPα both mediate normal bone cell and bone tissue formation. Significance: CD47/SIRPα may be a future molecular target to modulate bone homeostasis. Here, we investigated whether the cell surface glycoprotein CD47 was required for normal formation of osteoblasts and osteoclasts and to maintain normal bone formation activity in vitro and in vivo. In parathyroid hormone or 1α,25(OH)2-vitamin D3 (D3)-stimulated bone marrow cultures (BMC) from CD47−/− mice, we found a strongly reduced formation of multinuclear tartrate-resistant acid phosphatase (TRAP)+ osteoclasts, associated with reduced expression of osteoclastogenic genes (nfatc1, Oscar, Trap/Acp, ctr, catK, and dc-stamp). The production of M-CSF and RANKL (receptor activator of nuclear factor κβ ligand) was reduced in CD47−/− BMC, as compared with CD47+/+ BMC. The stromal cell phenotype in CD47−/− BMC involved a blunted expression of the osteoblast-associated genes osterix, Alp/Akp1, and α-1-collagen, and reduced mineral deposition, as compared with that in CD47+/+ BMC. CD47 is a ligand for SIRPα (signal regulatory protein α), which showed strongly reduced tyrosine phosphorylation in CD47−/− bone marrow stromal cells. In addition, stromal cells lacking the signaling SIRPα cytoplasmic domain also had a defect in osteogenic differentiation, and both CD47−/− and non-signaling SIRPα mutant stromal cells showed a markedly reduced ability to support osteoclastogenesis in wild-type bone marrow macrophages, demonstrating that CD47-induced SIRPα signaling is critical for stromal cell support of osteoclast formation. In vivo, femoral bones of 18- or 28-week-old CD47−/− mice showed significantly reduced osteoclast and osteoblast numbers and exhibited an osteopenic bone phenotype. In conclusion, lack of CD47 strongly impairs SIRPα-dependent osteoblast differentiation, deteriorate bone formation, and cause reduced formation of osteoclasts.

Osteoblasts are mononucleated cells derived from pluripotent mesenchymal stem cells, which prior to osteoblast commitment also can differentiate into other mesenchymal cell lineages such as bone marrow stromal cells, fibroblasts, chondrocytes, myoblasts, and adipocytes, depending on the activated signaling transcription pathways. The bone morphogenetic protein 2 (BMP-2) is an example of a potent cytokine, which stimulates osteoblast differentiation and increases bone formation, whereas the peroxisome proliferator-activated receptor ␥2 (PPAR␥2) is essential in directing the differentiation of adipocyte lineage cells. Activation of the transcription factor Runx2 (Runtrelated gene 2) is essential for osteoblast differentiation and bone formation. In addition, the transcription factor osterix, supposed to act downstream of Runx2, and other transcription factors also contribute to the control of osteoblastogenesis. Differentiated osteoblasts express matrix proteins (e.g. ␣-1-collagen) and alkaline phosphatase (ALP), a protein associated with mineralization, which also functions as a biochemical marker for osteoblastic differentiation and bone-forming capacity (6,7).
CD47, a cell surface glycoprotein of the Ig superfamily and ubiquitously expressed in the body, was originally discovered as an integrin-associated protein important for regulation of at least ␤1, ␤3, and ␤5 integrin members. In addition, CD47 has been suggested to function as a receptor for thrombospondin (8), and as a ligand for signal regulatory protein ␣ (SIRP␣) (9). SIRP␣ is also an Ig superfamily cell surface glycoprotein shown to be highly expressed in myeloid cells, neurons, endothelial cells, and fibroblasts, but not by T cells or B cells. By binding to CD47 on other cells, SIRP␣ can regulate macrophage-macrophage adhesion, fusion, and formation of giant cells (10,11).
We have previously shown that osteoclast formation is reduced both in vivo and in vitro in the absence of CD47 (12). These findings have been confirmed in vitro both by Uluçkan et al. (13) and Maile et al. (14). However, Maile et al. did not find any reduction in number of osteoclasts in vivo (14). To determine the role of CD47 in bone homeostasis, both in vitro and in vivo, we here investigated whether CD47 is involved in osteoclast and stromal cell/osteoblast differentiation in murine bone marrow cultures. In addition, we studied osteoblast numbers and bone homeostasis in CD47 Ϫ/Ϫ mice.

EXPERIMENTAL PROCEDURES
CD47 Ϫ/Ϫ Mice and SIRP␣ Mutant Mice-Generation of CD47 Ϫ/Ϫ has been described previously (15). Male CD47 Ϫ/Ϫ Balb/c mice, backcrossed to Balb/c (The Jackson Laboratory, Bar Harbor, ME) for 16 or more generations, and their wild-type homozygous littermates were from our own breeding colony. C57BL/6 SIRP␣ mutant mice, lacking most of the SIRP␣ cytoplasmic domain were described previously (16,17). Male SIRP␣ mutant mice or their wild-type littermates, backcrossed for Ͼ10 generations, were from our own breeding colony. Animals were kept in accordance with local guidelines and maintained in a specific pathogen-free barrier facility. The Local Animal Ethics Committee approved all animal procedures.
Bone Marrow Cell Culture (BMC)-Bone marrow cells were isolated from femurs and tibiae of Balb/c mice, 5-9 weeks of age (CD47 ϩ/ϩ and CD47 Ϫ/Ϫ ) and seeded on 48-or 12-well plates (Nunc, Roskilde, Denmark) in ␣-MEM with 10% FCS (Invitrogen), L-glutamine, and antibiotics (Sigma Aldrich), with cell concentration of 10 6 cells/cm 2 . The number of mice used for each experiment differed between 1-4 depending on the size of the experiment. If more than one mouse was used, the bone marrow cells were pooled. After 24 h (day 0) of incubation (37°C, 5% CO 2 ), the medium was replaced and either PTH 10 Ϫ8 M (Bachem, Bubendorf, Switzerland), D3 10 Ϫ8 M (Roche Diagnostics, Mannheim, Germany), or BMP-2 125 ng/ml (R&D Systems, Abdingdon, UK) were added to the cultures. At day 3, medium and test substances were changed. On days 6 -7, the cultures were harvested, and the cells were either lysated for further RNA isolation or fixed with acetone in citrate buffer and subsequently stained for TRAP as described below.
Osteoclast Formation and Resorption on Bone Slices-Bone marrow cells were isolated as described above. Thereafter, the cells were seeded on bone slices in 96-well tissue culturing plates (Nunc, Roskilde, Denmark) in ␣-MEM with 10% FCS (Invitrogen), L-glutamine and antibiotics (Sigma Aldrich), with cell concentration of 10 6 cells/cm 2 . The cells were incubated for 24 h (37°C, 5% CO 2 ), and thereafter, the medium was changed and D3 10 Ϫ8 M (Roche Diagnostics) was added. Medium was changed every third day, and after 10 days of culture, the experiments were harvested and stained for TRAP positivity, osteoclasts were counted in a light microscope, and resorption pits were visualized by an external light source. Moreover, the release of collagen type I fragments into the culture medium during resorption was analyzed by CrossLaps according to the manufacturer's instructions.
Osteoblast Differentiation-Bone marrow cells were isolated and plated as in above described BMC and plated in 24-well tissue culturing plates (Nunc, Roskilde, Denmark). The cells were cultured in ␣-MEM with 10% FCS (Invitrogen), L-glutamine, and antibiotics (Sigma Aldrich). The medium was changed every third day. At day 6, the culture medium was replaced with osteoblastic differentiation medium (␣-MEM with 10% FCS, L-glutamine, antibiotics, 10 mM ␤-glycerol phosphate, and 50 g/ml ascorbic acid). At days 7, 14, and 21, the cells were fixed with acetone in citrate buffer and stained for ALP or von Kossa as described below.
Bone Marrow Stromal Cell Culture-Bone marrow cells were isolated and treated as in BMC except that they were plated in 60 cm 2 culture dishes (Nunc). The cells were cultured in ␣-MEM with 10% FCS (Invitrogen), L-glutamine, antibiotics (Sigma Aldrich). The medium was changed after 3 days. After 7 days, the adhering cells were detached, counted, and plated at a concentration of 2 ϫ 10 4 cells/cm 2 in 24-well and 12-well tissue culturing plates (Nunc). The cells were either lysed for further RNA isolation or stained for ALP or von Kossa at the stated culture time.
Co-culture-Cells used for co-culture experiments were harvested from mice, 5-9 weeks of age, either Balb/c (CD47 ϩ/ϩ and CD47 Ϫ/Ϫ ) or C57BL/6 wild-type or SIRP␣ mutants. Bone marrow cells were isolated and treated as in BMC except that they were plated in 60 cm 2 culture dishes (Nunc). The cells were cultured in ␣-MEM with 10% FCS (Invitrogen), L-glutamine, and antibiotics (Sigma Aldrich). Medium was changed every third day, and after 14 days, the adherent cells were trypsinated, counted, and seeded at 10 5 cells/cm 2 , in 96-well tissue culturing plates (Nunc) and thereafter incubated for 24 h (37°C, 5% CO 2 ). Meanwhile non-adherent bone marrow myeloid precursor cells were isolated by flushing out bone marrow from femurs and tibiae. Following lysis of erythrocytes, the remaining cells were plated in 60 cm 2 tissue culture-treated culture dishes (Nunc) and kept at 37°C for 2 h. Thereafter, the non-adherent cells were collected, counted, and seeded onto the stromal cells at 6 ϫ 10 4 cells/cm 2 . The co-cultures were stimulated with D3 10 Ϫ8 M (Roche Diagnostics, Mannheim, Germany). After 5-7 days, the cultures were harvested, fixed, and stained for TRAP as described below.
TRAP staining was performed by use of the leukocyte acid phosphatase kit (Sigma-Aldrich) according to the manufacturer's instructions. TRAP ϩ cells with three or more nuclei were considered osteoclasts, and the number of multinucleated osteoclasts was counted in a light microscope.
ALP and von Kossa Staining-Osteoblast differentiation was visualized with ALP staining by using Naphtol AS-MX phosphate, disodium salt, and Fast Blue BB (Sigma Aldrich) according to histochemical standard procedure. Mineralization of extracellular matrix was visualized by von Kossa staining with 2% silver nitrate (Fisher Scientific UK) under UV light for 30 min. The result of ALP and von Kossa staining was photographed.
RNA Isolation and First-stranded cDNA Synthesis-Total RNA from adherent bone marrow cell cultures was isolated by using the RNAqueous®-4PCR kit (Ambion, Austin, TX) according to the manufacturer's instructions. By using the high capacity cDNA reverse transcription kit (Foster City, CA), mRNA was reverse-transcribed to cDNA.
ELISA-The amount of synthesized protein of RANKL and M-CSF in culture medium was assessed by measuring the levels of the proteins in BMC using commercially available ELISA kits. RANKL was measured according to manufacturer's instructions (R&D Systems) and M-CSF according to the manufacturer's instructions (AB Frontier, Seoul, Korea).
TRAP Activity Assay-BMC were washed three times in PBS and lysed in Triton X-100 (0.2% in H 2 O). By using p-nitrophe-nol phosphate (1.7 mg/ml) as substrate at pH 4.9 in the presence of tartrate (0.17 M), the TRAP activity was determined. The activity of the enzyme was assessed as the A 405 . The enzyme assay was performed under conditions where the reaction was proportional to amount of enzyme and reaction time.
ALP Activity Assay-Alkaline phosphatase was measured in BMC stimulated with BMP-2 by using a commercially available kit (Sigma Aldrich).
Flow Cytometric Analysis-Bone marrow stromal cells were washed and resuspended in cold PBS. Fifty l of the cells (1 ϫ 10 5 cells) were seeded in a 96-well V-bottomed plate and centrifuged at 1500 rpm for 3 min at 4°C. Thereafter, the cells were resuspended in 30 l staining buffer (PBS ϩ 2% FCS) containing saturating concentrations of Alexa Fluor 488-conjugated anti-SIRP␣ mAb P84, or Alexa Fluor 488-conjugated rat IgG isotype control mAb, for 30 min on ice. Cells were then washed once with 200 l of cold PBS, resuspended in cold PBS/2% FCS/0.05% NaN 3 , and analyzed by flow cytometry (FACS Calibur, Becton Dickinson) and CellQuest software (Beckton Dickinson).
Phenotypic Studies-Histomorphometry and peripheral quantitative computed tomography were used to analyze the bone phenotype of male BalbC CD47 Ϫ/Ϫ mice at the age of 18 and 28 weeks and compared with normal mice at the same age.
Histomorphometry-CD47 ϩ/ϩ or CD47 Ϫ/Ϫ mice were injected with the fluorescent compound calcein (15 mg/kg; Sigma) 10 days and 3 days prior to collection, respectively. At 18 and 28 weeks of age CD47 ϩ/ϩ and CD47 Ϫ/Ϫ were euthanized by cervical dislocation. Both femurs were excised and bisected transversely at the midpoint of the shaft. The distal halves of the right femora were fixed and embedded, undecalcified in methyl-methacrylate resin (Medim-Medizinische Diagnostik, Giessen, Germany), and 5-m sagittal sections were prepared for analysis using Q Win software (Leica Microsystems Pty., Ltd., Sydney, Australia). Sagittal sections were stained for mineralized tissue, and trabecular bone volume, trabecular thickness, and trabecular number were calculated in a sample region extending 4.5 mm proximal to the distal growth plate and encompassing all trabecular bone within the cortical boundaries. Osteoblast OCTOBER 11, 2013 • VOLUME 288 • NUMBER 41 parameters (osteoblast surface, osteoblast number, and osteoid surface) were estimated using sections stained with von Kossa stain and toluidine blue. Only osteoblasts identified adjacent to osteoid surfaces were included in the analysis. Bone formation (mineralizing surface, MS) was estimated using the bone surface (BS) coverage of single-and double-labeled (sLS and dLS) fluorescent bands using the equation MS ϭ ((0.5 ϫ sLS) ϩ dLS) ϫ 100/BS, expressed as a percentage of bone surface. Mineral apposition rate (MAR) was estimated by the distance between the calcein labels divided by the time interval between injections of labels (MAR ϭ interlabel distance divided by 7, expressed in m/day). Bone formation rate (BFR) was calculated after fluorescence microscopy (Leica Microsystems, Heerbrugg, Switzerland) as BFR ϭ MS/bone surface ϫ MAR, expressed in m 2 /m 3 /day. For measurements of osteoclast surface and osteoclast number, sections were stained for TRAP positivity.

Lack of CD47 Impairs Bone Cell Differentiation
Peripheral quantitative computed tomography was used to analyze cortical bone of the tibia, using a Stratec XCT Research SA scanner (Stratec Medizintechnik, Pforzheim, Germany). Scans were conducted using a voxel size of 70 m, scan speed of 5 mm/s and slice width of 1 mm on excised tibia, as described previously (18). Bones were scanned in a single slice 4 mm distal from the proximal margin of the tibia.
Micro-computed Tomography-A Skyscan 1174 scanner and associated analysis software (Skyscan, Aartselaar, Belgium) were used to visualize the three-dimensional bone structure of the distal femoral metaphysis. Following fixation, representative bones were scanned in 70% ethanol. A 0.5-mm aluminum filter was applied to the 50 kV x-ray source, with an exposure time of 3600 ms and sharpening at 40%. Distal femora were scanned at a 6.2 m pixel resolution acquired over an angular range of 180°, with a rotation step of 0.4°. Following reconstruction, 67-m-thick images were created from a point 400 m from the most proximal aspect of the growth plate.
Statistical Analyses-The statistical analyses were performed using one-way analysis of variance (ANOVA) with Levene's homogenicity test, and post hoc Bonferroni's test or, where appropriate, Dunnett's t test. Students t test were used in the histomorphometric analyses (SPSS, version 18). All experiments were performed at least twice with comparable results, and all data are represented as the means Ϯ S.E. Significance levels were set to p Ͻ 0.05 (*), 0.01 (**), or 0.001 (***).

CD47 Is Required for Functional Osteoclast Differentiation in Vitro-
We have previously shown that mice lacking CD47 have reduced numbers of osteoclasts on trabecular bone surfaces in vivo (12). To further elucidate the role of CD47 in osteoclastogenesis within the bone marrow, we studied CD47 Ϫ/Ϫ or CD47 ϩ/ϩ crude BMC, containing both mesenchymal/stromal cells and hematopoietic cells, to mimic the in vivo situation. Addition of PTH (10 Ϫ8 M) or D3 (10 Ϫ8 M) stimulated formation of multinucleated osteoclasts in CD47 ϩ/ϩ BMC (n ϭ 96 Ϯ 15.8 and 142 Ϯ 14.1, respectively) (Fig. 1, A and B). In marked contrast, only a few multinucleated osteoclasts were detected in PTH or D3-stimulated CD47 Ϫ/Ϫ cultures (n ϭ 4 Ϯ 1.1 and n ϭ 8 Ϯ 3.2, respectively) (Fig. 1, A and B). When CD47 Ϫ/Ϫ BMC were cultured on bone slices, we found both fewer osteoclasts (Fig. 1C) and dramatically fewer resorption pits reflected as an 82% reduced release of collagen type I fragments into the culture medium, as compared with that in CD47 ϩ/ϩ cultures (Fig.  1D). Trap/Acp5 mRNA and TRAP protein expressions were substantially increased in CD47 ϩ/ϩ BMC in response to PTH (10 Ϫ8 M) or D3 (10 Ϫ8 M), whereas no (D3), or only a slight (PTH) (p Ͻ 0.001), increase in TRAP protein expression was seen in CD47 Ϫ/Ϫ cultures (Fig. 2, A and B). Quantitative real-time PCR analyses of CD47 ϩ/ϩ BMC revealed that D3 substantially stimulated the mRNA expression of ctr and catK, whereas D3 did not increase ctr expression and only marginally increased catK expression (p Ͻ 0.05) in CD47 Ϫ/Ϫ cultures (Fig. 2, C and D). The mRNA expression of nfatc1, the master transcription factor regulating osteoclastogenesis, and the osteoclast-associated receptor oscar, were both increased in D3-stimulated CD47 ϩ/ϩ cultures but not in CD47 Ϫ/Ϫ cultures (Fig. 3). D3 did not significantly regulate the mRNA expression of the receptors c-fms or SIRP␣, and no significant difference was seen comparing the two genotypes (Fig. 3). However, we found that the mRNA expression of rank was up-regulated in CD47 ϩ/ϩ BMC stimulated with D3 (p Ͻ 0.001), whereas no such response to D3 was seen in CD47 Ϫ/Ϫ BMC (Fig. 3). Additionally, the mRNA expression of dendritic cell-specific transmembrane protein (dc-stamp), a protein of importance for osteoclast fusion was substantially increased in response to D3 in the CD47 ϩ/ϩ cultures (p Ͻ 0.05; Fig. 3), whereas only a minor increase was seen in CD47 Ϫ/Ϫ cultures (p Ͻ 0.05; Fig. 3).
Because production of M-CSF in BMC-cultures is pivotal for osteoclastogenesis due to its capacity to increase survival of osteoclast progenitor cells (3), we next investigated whether this function was impaired in CD47 Ϫ/Ϫ BMC. These experiments revealed that m-csf gene expression was significantly reduced in unstimulated CD47 Ϫ/Ϫ BMC cultures, as compared with that in unstimulated CD47 ϩ/ϩ BMC (p Ͻ 0.05; Fig. 4A). By comparing M-CSF protein levels in culture medium from CD47 ϩ/ϩ or CD47 Ϫ/Ϫ cultures, we confirmed a significantly lower concentration of M-CSF, both in control and D3-stimulated CD47 Ϫ/Ϫ cultures (Fig. 4B). The effect of D3 on M-CSF expression in BMC is to our knowledge unknown, but in our experiments, D3 seems to decrease the M-CSF expression in the normal mice. Analyses of RANKL expression in D3-stimulated CD47 ϩ/ϩ BMC revealed a significant increase in rankl mRNA, as well as RANKL protein in culture medium, whereas no such response to D3 was detected in CD47 Ϫ/Ϫ cultures with the exception of a very slight but significant increase (p Ͻ 0.01) in the protein levels of RANKL in CD47 Ϫ/Ϫ culture medium (Fig. 4, C and D). When stimulating BMC cultures with D3, the osteoprotegerin protein expression decreased in both CD47 Ϫ/Ϫ and CD47 ϩ/ϩ cultures (data not shown), and thus, the RANKL/osteoprotegerin-ratio increased in both genotypes.
CD47 Is Required for Osteoblastogenesis and Bone Formation in Vitro-The strongly reduced production of M-CSF and RANKL in CD47 Ϫ/Ϫ cultures indicated a stromal cell differentiation defect when CD47 was absent. Therefore, we next analyzed osteoblastic differentiation in BMC and found that the major transcription factors for osteoblast differentiation, runx2 and osterix, were increased over time in CD47 ϩ/ϩ cultures (Fig. 5A). However, in CD47 Ϫ/Ϫ cultures, we observed no increase in osterix mRNA expression and only a modest increase in runx2 mRNA expression (Fig. 5A). Moreover, although the expression of Alp/Akp1 and ␣-1-collagen substan-tially increased over time in CD47 ϩ/ϩ BMC, no significant increase of Alp/Akp1 and only a small increase in ␣-1-collagen were seen in CD47 Ϫ/Ϫ cultures (Fig. 5B). Thus, the mRNA expression of Alp/Akp1 and ␣-1-collagen were significantly   . Impaired expression of osteoclast differentiation-associated genes in CD47 ؊/؊ BMC. The expression of nfatc1, oscar, rank, and dc-stamp were all significantly increased when stimulating CD47 ϩ/ϩ BMC with D3 (10 Ϫ8 M). With the exception of dc-stamp, which was weakly but statistically significantly increased, neither of these genes were up-regulated in D3-stimulated CD47 Ϫ/Ϫ BMC. The mRNA expression of c-fms or SIRP␣ were not significantly up-regulated when stimulating BMC from CD47 ϩ/ϩ or CD47 Ϫ/Ϫ with D3 (10 Ϫ8 M). The CD47 ϩ/ϩ control was set to 100%, and data shown are quantitative values normalized to the housekeeping gene ␤-actin. All genes were analyzed in two or more different experiments, and the data shown are the means Ϯ S.E. of four to six samples/group in one representative experiment. ns, not significant. *, p Ͻ 0.05 and ***, p Ͻ 0.001, using ANOVA with Levene's homogenicity test and post hoc Bonferroni's test or, where appropriate, Dunnett's t test. , were lower both in control and D3-stimulated CD47 Ϫ/Ϫ BMC, as compared with that in CD47 ϩ/ϩ BMC. In D3-stimulated CD47 ϩ/ϩ BMC, both rankl mRNA expression (C) and RANKL protein in culture medium (D) were strongly increased, as compared with that in CD47 Ϫ/Ϫ BMC. The unstimulated CD47 ϩ/ϩ control was set to 100%, and data of mRNA analyses are quantitative values normalized to the housekeeping gene ␤-actin. All genes were analyzed in two or more different experiments, and the data shown are the means Ϯ S.E. of four to six samples/group in one representative experiment. ns, not significant. *, p Ͻ 0.05; **, p Ͻ 0.01; and ***, p Ͻ 0.001, using ANOVA with Levene's homogenicity test and post hoc Bonferroni's test or, where appropriate, Dunnett's t test.
lower in CD47 Ϫ/Ϫ BMC when compared with CD47 ϩ/ϩ BMC. Interestingly, at day 3 of culture, we found a slight up-regulation of the adipocyte-specific gene PPAR␥ expression in CD47 Ϫ/Ϫ cultures, as compared with CD47 ϩ/ϩ cultures, whereas no difference in PPAR␥ expression was seen at day six of culture (Fig. 5C).
We further analyzed the stromal cell differentiation capacity in CD47 Ϫ/Ϫ BMC by culturing the cells in osteogenic medium supplemented with ␤-glycerophosphate and ascorbic acid. Under these conditions, CD47 ϩ/ϩ BMCs were positive for the osteoblastic marker ALP at day 14 of culture, when small spots of mineral deposits were also detected (Fig. 5D). In marked contrast, less ALP-staining and no mineral deposits were seen at day 14 in CD47 Ϫ/Ϫ BMC cultures (Fig. 5D). By 21 days of culture, the ALP expression and mineral deposits increased in both CD47 ϩ/ϩ and CD47 Ϫ/Ϫ cultures, but the expression of ALP and formation of mineral were still much lower in CD47 Ϫ/Ϫ cultures (Fig. 5D).
Because BMP-2 is a known stimulator of osteoblastic differentiation (19), we next investigated the effects of exogenous BMP-2 on BMC differentiation toward the osteoblastic lineage. As shown in Fig. 5D, culture of CD47 ϩ/ϩ BMC for 6 days in osteogenic medium alone induced ALP protein expression, as compared with the virtual lack of ALP expression in CD47 Ϫ/Ϫ cultures. Addition of BMP-2 induced a robust increase in ALP protein expression in CD47 ϩ/ϩ BMC, whereas the ALP was maintained at extremely low levels in BMP-2-stimulated CD47 Ϫ/Ϫ cultures (Fig. 6, A and B). Furthermore, there was a marked difference in cellular morphology between the two genotypes. In both unstimulated and BMP-2-stimulated CD47 ϩ/ϩ BMC, the vast majority of the cells showed spreading on the plastic surface. However, cell spreading appeared to be strongly reduced in unstimulated as well as BMP-2-stimulated CD47 Ϫ/Ϫ cultures (Fig. 6B).
Tyrosine phosphorylation of SIRP␣, for which CD47 is a ligand, is important in mediating adhesion-dependent cellular functions (16). In endothelial cells, adhesion-dependent SIRP␣ tyrosine phosphorylation requires CD47 (20). Therefore, we next investigated whether impaired signaling through SIRP␣ could be involved in mediating the reduced osteoblast differentiation in CD47 Ϫ/Ϫ cultures. Flow cytometric analysis of bone marrow stromal cells showed that SIRP␣ was expressed at equal levels by both CD47 ϩ/ϩ and CD47 Ϫ/Ϫ stromal cells (Fig. 6C). When SIRP␣ was immunoprecipitated from stromal cell cultures of CD47 ϩ/ϩ or CD47 Ϫ/Ϫ mice, a strong tyrosine phos- FIGURE 5. Impaired expression of osteoblast-specific genes and osteoblast differentiation in CD47 ؊/؊ BMC. A and B, the mRNA expression of runx2, osterix, Alp/Akp1, and ␣-1-collagen were significantly increased from day 3 to 6 during culture of CD47 ϩ/ϩ BMC. With the exception of runx2, these responses were significantly lower in CD47 Ϫ/Ϫ BMC, as compared with CD47 ϩ/ϩ BMC. C, the expression of the adipocytic gene PPAR␥ was higher in CD47 Ϫ/Ϫ cultures, as compared with that in CD47 ϩ/ϩ cultures on day 3 but not on day 6. Data shown are quantitative values normalized to the housekeeping gene ␤-actin, where the CD47 ϩ/ϩ control day 3 was set to 100%. All genes were analyzed in two or more different experiments, and the data shown are the means Ϯ S.E. of four to six samples/group in one representative experiment. ns, not significant. *, p Ͻ 0.05; **, p Ͻ 0.01; and ***, p Ͻ 0.001, using ANOVA with Levene's homogenicity test and post hoc Bonferroni's test or, where appropriate, Dunnett's t test. D, representative photographs of ALP staining (purple) and von Kossa staining (brown mineral deposits) in BMC from both CD47 ϩ/ϩ and CD47 Ϫ/Ϫ mice after 14 and 21 days of culture in osteogenic medium supplemented with ␤-glycerophosphate and ascorbic acid. OCTOBER 11, 2013 • VOLUME 288 • NUMBER 41 phorylation of SIRP␣ was detected in CD47 ϩ/ϩ cells, whereas the SIRP␣ phosphorylation was virtually blunted in CD47 Ϫ/Ϫ cells (Fig. 6D). Thus, lack of SIRP␣ tyrosine phosphorylation in CD47 Ϫ/Ϫ stromal cells could explain the reduced osteoblastic differentiation in CD47 Ϫ/Ϫ BMC. To further challenge this hypothesis, we investigated differentiation of bone marrow stromal cells from SIRP␣-mutant mice. These mice express normal levels of CD47 as well as the SIRP␣ extracellular domain, but the cytoplasmic SIRP␣ domain is deleted and cannot be phosphorylated upon ligation of the receptor (17). Interestingly, ALP protein expression was strongly reduced in SIRP␣ mutant BMC cultured for 12 days, as compared with that in SIRP␣ wild-type cultures (data not shown). Thus, osteogenic differentiation of bone marrow stromal cells requires CD47-dependent SIRP␣ signaling.

Lack of CD47 Impairs Bone Cell Differentiation
Based on these findings, we next hypothesized that the absence of stromal cell CD47, and subsequent lack of SIRP␣ signaling, could explain the impaired osteoclast formation in D3-stimulated CD47 Ϫ/Ϫ BMC. To address this hypothesis, we conducted D3-stimulated co-culture experiments combining bone marrow stromal cells and bone marrow myeloid precursor cells of either genotype (wild-type, CD47 Ϫ/Ϫ , or SIRP␣ mutant). These experiments revealed that wild-type and CD47 Ϫ/Ϫ myeloid precursor cells generated equal numbers of multinucleated osteoclast on CD47 wild-type bone marrow stromal cells, whereas myeloid precursors of both genotypes generated low numbers of multinucleated osteoclast on CD47 Ϫ/Ϫ stromal cells (Fig. 7A). Similarly, SIRP␣ wild-type and SIRP␣ mutant myeloid precursors gave rise to an equal number of multinucleated cells on SIRP␣ wild-type bone marrow stromal cells, whereas SIRP␣ mutant stromal cells showed a strongly reduced ability to promote multinucleated osteoclast formation in both wild-type and SIRP␣ mutant myeloid precursors (Fig. 7B).
Reduced Osteoclast and Osteoblast Numbers and Impaired Skeletal Parameters, in CD47 Ϫ/Ϫ Mice-To investigate the biological impact of our in vitro observations, we next analyzed osteoclast and osteoblast densities in femoral bones of 18-or 28-week-old male CD47 ϩ/ϩ or CD47 Ϫ/Ϫ mice. We have previously reported a 37% reduction in TRAP ϩ osteoclast surface and a 29% reduction in osteoclast number in the distal femoral metaphysis of 18-week-old CD47 Ϫ/Ϫ mice, as compared with that in age-matched CD47 ϩ/ϩ mice (12). Herein, we further analyzed the osteoclast density in 28-week-old mice and found that CD47 Ϫ/Ϫ mice had a 38% (p Ͻ 0.01) reduction in osteoclast numbers and a 46% (p Ͻ 0.01) reduction in osteoclast surface, as compared with CD47 ϩ/ϩ mice (Fig. 8, A and B). Importantly, osteoblast numbers were also reduced in CD47 Ϫ/Ϫ mice, with a 55 and 47% reduction, respectively, in 18-and 28-week-old CD47 Ϫ/Ϫ mice, as compared with age-matched CD47 ϩ/ϩ mice (p Ͻ 0.001 and p Ͻ 0.01, respectively; Fig. 8, C and D).
To further investigate the effects of reduced osteoclastogenesis and osteoblastogenesis in CD47 Ϫ/Ϫ mice, trabecular FIGURE 6. Blunted osteoblast differentiation in CD47 Ϫ/Ϫ stromal cells is associated with the lack of SIRP␣ tyrosine phosphorylation. A, ALP protein expression was evident following 6 days of CD47 ϩ/ϩ BMC culture in osteogenic medium alone (control) and significantly increased in cultures stimulated for 6 days with BMP-2 (125 ng/ml). CD47 Ϫ/Ϫ BMC expressed remarkably little ALP protein when cultured in osteogenic medium and were unresponsive to BMP-2. B, representative photomicrographs showing purple ALP ϩ cells in CD47 ϩ/ϩ or CD47 Ϫ/Ϫ BMC cultured in osteogenic medium alone (control), or in BMC stimulated with BMP-2 (125 ng/ml) stimulated at 6 days of culture. ALP ϩ cells are stained blue. Data are shown as the means Ϯ S.E. of four to six samples/group in one of two representative experiments. ns, not significant. ***, p Ͻ 0.001, using ANOVA with Levene's homogenicity test and post hoc Bonferroni's test or, where appropriate, Dunnett's t test. C, flow cytometric analysis of cell surface SIRP␣ expression in CD47 ϩ/ϩ or CD47 Ϫ/Ϫ bone marrow stromal cells cultured for 14 days. Cells were labeled with anti-SIRP␣ mAb P84 (gray histogram) or isotype rat IgG1 mAb (open histogram), where both antibodies were conjugated to Alexa Fluor 488. D, adherent bone marrow stromal cells derived from CD47 ϩ/ϩ or CD47 Ϫ/Ϫ mice and cultured for 14 days were lysed, and SIRP␣ was immunoprecipitated from equal amounts of protein. The tyrosine phosphorylation (pY) of SIRP␣ immunoprecipitated from stromal cells of each genotype was determined by Western blot (WB), using anti-phosphotyrosine mAb 4G10. Equal loading of SIRP␣ immunoprecipitates was confirmed by blotting for SIRP␣ using mAb P84. The data are representative of two separate experiments with virtually identical results. microarchitecture was evaluated both in adult and aged mice bone. Histomorphometric analysis of the distal femurs revealed that the trabecular bone volume was significantly reduced by 15% in 18-week-old CD47 Ϫ/Ϫ mice as compared with that in CD47 ϩ/ϩ mice (p Ͻ 0.01; Fig. 9, A and B). In 28-week-old CD47 Ϫ/Ϫ mice, the trabecular bone volume was further reduced to 45% of that in CD47 ϩ/ϩ mice (p Ͻ 0.001; Fig. 9B). These differences in trabecular bone microarchitecture were also confirmed by micro-computed tomography (Fig. 9C). To further investigate this intriguing observation, we next examined bone formation in 18-week-old CD47 ϩ/ϩ or CD47 Ϫ/Ϫ mice using dual calcein labeling in vivo (Fig. 9D). These experiments revealed that the bone formation rate was markedly reduced in 18-week-old CD47 Ϫ/Ϫ mice, being 62% of that in CD47 ϩ/ϩ mice (0.26 Ϯ 0.008 m 2 /m 3 /day versus 0.10 Ϯ 0.030 m 2 /m 3 /day, p Ͻ 0.001; Fig. 9D). This inhibition was consist-ent with a reduction of both osteoblast activity and surface extent, with mineral apposition rate and mineralizing surface reduced by 34 and 42%, respectively (p Ͻ 0.01; Fig. 9D). In 28-week-old CD47 ϩ/ϩ mice, the mineralized surface, mineral apposition rate, and bone formation rate were much lower, as compared with that in 18-week-old CD47 ϩ/ϩ mice (Fig. 9D), and no difference in these parameters were seen between CD47 ϩ/ϩ or CD47 Ϫ/Ϫ mice at 28 weeks of age (Fig. 9D). In line with the reduced mineralization observed in 18-week-old CD47 Ϫ/Ϫ mice, these mice were also found to have a 66% reduction in osteoid surface, as compared with that in agematched CD47 ϩ/ϩ mice (p Ͻ 0.001; Fig. 9D). In 28-week-old CD47 Ϫ/Ϫ mice, osteoid surface was reduced by 43%, as compared with that in CD47 ϩ/ϩ mice (Fig. 9D).
We also investigated the cortical bone compartment in 18-or 28-week-old CD47 ϩ/ϩ and CD47 Ϫ/Ϫ mice, using peripheral FIGURE 7. D3-stimulated bone marrow stromal cells lacking CD47, or expressing mutated non-signaling SIRP␣, are unable to support osteoclastogenesis. A, CD47 ϩ/ϩ D3-stimulated (10 Ϫ8 M) BMS induced similar levels of osteoclast formation in both CD47 ϩ/ϩ and CD47 Ϫ/Ϫ non-adherent BMM. In contrast, D3-stimulated CD47 Ϫ/Ϫ BMS showed strongly reduced capacity to induce osteoclast formation in non-adherent bone marrow myeloid precursor cells of either genotype. B, wild-type D3-stimulated BMS induced osteoclast formation in non-adherent bone marrow myeloid precursor cells from both wild-type and non-signaling SIRP␣ mutant mice. However, osteoclast formation in non-adherent bone marrow myeloid precursor cells of either genotype was strongly reduced when co-cultured with D3-stimulated BMS expressing the non-signaling SIRP␣ protein. Data are shown as the means Ϯ S.E. of four to six samples/ group in one of two representative experiments. ns, not significant. ***, p Ͻ 0.001, using ANOVA with Levene's homogenicity test and post hoc Bonferroni's test or, where appropriate, Dunnett's t test. mut, mutant. FIGURE 8. Reduced osteoclast and osteoblast numbers in bones of CD47 ؊/؊ mice. At 28 weeks of age, the number of osteoclasts covering trabecular bone surface (A) and the total osteoclast surface that covers the trabecular bone (B) were both significantly reduced in CD47 Ϫ/Ϫ mice, as compared with that in CD47 ϩ/ϩ mice. At both 18 (C) and 28 (D) weeks of age, the number of osteoblasts covering the bone surface was significantly lower in CD47 Ϫ/Ϫ mice, as compared with that in CD47 ϩ/ϩ mice. Data are means Ϯ S.E. of 7-10 mice in each age group. **, p Ͻ 0.01 and ***, p Ͻ 0.001, as compared with that in CD47 ϩ/ϩ mice, using Student's t test for unpaired analyses.
quantitative computed tomography analysis. Although this analysis showed a slight decrease in cortical area and thickness in both 18-and 28-week-old CD47 Ϫ/Ϫ mice, the only statistically significant difference was detected in cortical area of 28-week-old CD47 Ϫ/Ϫ mice (Fig. 9D). Note that there were no differences in bone marrow fat content in 28-weekold CD47 ϩ/ϩ and CD47 Ϫ/Ϫ mice (data not shown).

DISCUSSION
The mechanisms regulating bone homeostasis are multiple and still not fully understood. In the present study, we have shown that CD47 is required to activate its receptor SIRP␣ in stromal cells/osteoblasts, which is critically important within the bone marrow for the ability of stromal cells/osteoblasts to promote formation of osteoclasts. Lack of CD47/SIRP␣ signaling in the stromal cell compartment impairs osteoblast differentiation in vivo and in vitro. This functional defect not only impairs osteoid formation and mineralization in vivo but also results in impaired formation of osteoclasts. Together, this suggests an explanation to the osteopenic bone phenotype of CD47 Ϫ/Ϫ mice.
Investigations of osteoclast formation in vitro are based on a number of different experimental systems, which show important variations. One system is based on selection of monocyte/ macrophage progenitors from the bone marrow and differentiation of these cells toward the osteoclastic lineage by addition of M-CSF and RANKL to the culture medium (bone marrow myeloid precursor cell (BMM) cultures) (21). Another system can be used to mimic more physiological conditions in vivo, which is based on the culture of crude bone marrow (BMC cultures), containing both stromal cells (with osteoblast progenitor potential) and hematopoietic progenitor cells (which can differentiate into pre-osteoclasts) (22). In the latter system, which is used in this paper, stimulation with PTH or D3 will induce stromal cells/osteoblasts to produce RANKL, resulting in a natural indirect stimulation of pre-osteoclast differentiation and osteoclast maturation. The importance of stromal cells can be bypassed if M-CSF and RANKL are added to this type of culture.
Our previous data suggest that the physical interaction between CD47 and SIRP␣ is important for osteoclast formation in BMM cultures because functional blocking antibodies against either CD47 or SIRP␣ impaired osteoclast formation in CD47 ϩ/ϩ cultures, and that CD47 Ϫ/Ϫ BMM cultures generated less osteoclasts than CD47 ϩ/ϩ cultures (12). Uluçkan et al. (13) also showed impaired generation of osteoclasts from CD47 Ϫ/Ϫ BMM cells but that the osteoclast formation defect could be rescued by pre-incubation with M-CSF and by increasing the levels of RANKL. We have confirmed the rescued osteoclast formation in CD47 Ϫ/Ϫ BMM cultures by pre-incubating the cells with M-CSF and by adding increased concentrations of FIGURE 9. Reduced bone formation in CD47 ؊/؊ mice. A, representative femoral sections from 18-week-old CD47 ϩ/ϩ and CD47 Ϫ/Ϫ mice stained for mineral with von Kossa. B, at 18 and 28 weeks of age, the trabecular bone volume was lower in CD47 Ϫ/Ϫ mice, as compared with age-matched CD47 ϩ/ϩ mice. C, micro-computed tomography images of left femur from 18-and 28-week-old CD47 ϩ/ϩ and CD47 Ϫ/Ϫ mice. At 18 weeks of age, trabecular bone formation rate, mineral apposition rate, and mineralizing surface were all significantly reduced in femurs from CD47 Ϫ/Ϫ mice, as compared with that in CD47 ϩ/ϩ mice. At 28 weeks of age, the same parameters did not show any significant differences between CD47 ϩ/ϩ and CD47 Ϫ/Ϫ mice. Osteoid surface was significantly lower in CD47 Ϫ/Ϫ mice at both 18 and 28 weeks, as compared with that in CD47 ϩ/ϩ mice. A non-significant (ns) difference was seen in cortical area, when comparing CD47 ϩ/ϩ and CD47 Ϫ/Ϫ mice at 18 weeks of age; however, a 6% statistically significant difference was evident in cortical area at 28 weeks of age. The cortical thickness was not significantly different in the two genotypes at 18 or 28 weeks of age. Data are means Ϯ S.E. of 7-10 mice in each age group. ns, not significant. *, p Ͻ 0.05; **, p Ͻ 0.01; and ***, p Ͻ 0.001, as compared with that in CD47 ϩ/ϩ mice, using Student's t test for unpaired analyses. d, day.
RANKL or a highly efficient truncated form of RANKL (data not shown). This is an interesting observation, suggesting that M-CSF-mediated early differentiation of pre-osteoclasts can rescue the osteoclast phenotype in CD47 Ϫ/Ϫ mice if sufficient RANKL levels are provided.
In the present study, we found that osteoclast formation was virtually abolished in CD47 Ϫ/Ϫ BMC stimulated with PTH or D3 in vitro. In this more complex system, the observed defect cannot only be explained as a pre-osteoclast fusion defect, since gene expression levels of nfatc1, oscar, Trap/Acp5, the ctr, and catK, all important markers of differentiation along the osteoclastic lineage (5,23), were strongly impaired as well, whereas the expression levels of the M-CSF receptor c-fms or that of SIRP␣ were not impaired in CD47 Ϫ/Ϫ BMC. Rather, our data suggest that a key defect in CD47 Ϫ/Ϫ BMC is associated with impaired function and differentiation of stromal cells along the osteoblastic lineage. These cells, when stimulated with PTH or D3, are induced to produce M-CSF and RANKL, which are pivotal to drive osteoclast differentiation and formation in this system and also in vivo (4). It is therefore interesting to note that D3-stimulated M-CSF and RANKL expression in CD47 Ϫ/Ϫ BMC were both strongly reduced at the gene as well as protein levels and that these cultures showed a very low TRAP activity. The importance of this finding is emphasized by data from Maile et al. (14), showing that addition of exogenous M-CSF and RANKL to a crude CD47 Ϫ/Ϫ bone marrow cultures rendered the same TRAP staining levels as in CD47 ϩ/ϩ cultures.
The lack of M-CSF and RANKL expression in CD47 Ϫ/Ϫ BMC pointed toward a phenotypic change in the bone marrow stromal cells when CD47 was absent. To further understand the mechanism behind this, we investigated the hypothesis that the stromal cell defect was due to the absence of CD47-mediated induction of SIRP␣ signaling. We found a strongly decreased tyrosine phosphorylation of SIRP␣ in bone marrow stromal cells when CD47 was absent. Together with the results from the co-culture experiments, showing that bone marrow stromal cells lacking either CD47 or the signaling domain of SIRP␣ could not efficiently support osteoclast formation, this clearly suggests that SIRP␣ signaling in bone marrow stromal cells is necessary for osteoclastogenesis. The fact that osteoclast formation was intact in bone marrow macrophages, lacking either CD47 or the SIRP␣ signaling domain, when cultured on wildtype bone marrow stromal cells, further suggest that the impaired osteoclast formation in CD47 Ϫ/Ϫ BMC cultures was unlikely due to lack of CD47/SIRP␣ interaction in pre-osteoclasts but that the defect was rather isolated to the stromal cells.
In 2011, Maile et al. (14) showed an impaired osteoclast formation in CD47 Ϫ/Ϫ crude bone marrow cells cultured in the presence of M-CSF and RANKL. The authors suggested that the physical interaction between CD47 and SIRP␣ induced SIRP␣ tyrosine phosphorylation, recruitment of SHP-1, and subsequent dephosphorylation of non-muscle cell myosin IIA, which, altogether, would be important to promote fusion and formation of osteoclasts (14). It is hard to interpret these data because the osteoclast formation was driven by addition of exogenous M-CSF and RANKL to the crude bone marrow culture, which based on data from the present study and that of a previous study (13) can compensate for the lack of CD47. The reason why Maile et al. (14) only detected late fusion defects when CD47 was missing in the cultures could be that the early osteoclast precursors proliferated and differentiated in the presence of M-CSF. In marked contrast, data by van Beek et al. (24) in 2009 showed that bone marrow cells from SIRP␣ mutant mice (lacking the signaling SIRP␣ cytoplasmic domain, but expressing the extracellular CD47-binding domain of SIRP␣) generated normal numbers of osteoclasts with the same number of nuclei, as compared with that in wild-type bone marrow cultures, when cultured with exogenous M-CSF and RANKL (24). Notably, van Beek et al. (24) used a truncated form of mouse RANKL with a much lower EC 50 value then the human RANKL that was used by Maile et al. (14), which could be one explanation behind the normal osteoclast formation in SIRP␣ mutant bone marrow cultures.
The stromal cell phenotypic change apparent when CD47 was missing did not only impair osteoclast differentiation but also affected the osteoblast differentiation pathway. At the gene expression level, we found an impaired expression of runx2 and ␣-1-collagen and a virtually blunted expression of osterix and Alp/Akp1, whereas the expression of the adipocyte-associated gene PPAR␥2 was found to be normal. To find out whether the mechanism of action in CD47 Ϫ/Ϫ cells was lack of SIRP␣ signaling, we took advantage of the SIRP␣ mutant mice where the cytoplasmic SIRP␣ domain is deleted and cannot be phosphorylated. Culture of bone marrow stromal cells from these mice showed dramatically decreased alkaline phosphatase activity. These findings support the hypothesis that CD47 has a profound effect on osteoblastic differentiation by inducing activation of SIRP␣ signaling. Further studies of the downstream signaling from SIRP␣ in stromal cells are necessary to understand how the receptor is linked to the master regulator of osteoblast differentiation, RUNX-2.
The strong in vitro phenotype in CD47 Ϫ/Ϫ BMC suggested that there should also be an obvious bone phenotype in these mice in vivo. We previously reported that 18-week-old CD47 Ϫ/Ϫ mice had reduced numbers of osteoclasts in vivo (12), which we in the present study confirmed to also be the case in 28-weeks old CD47 Ϫ/Ϫ mice. By looking at the osteoclast phenotype per se, one would assume that CD47 Ϫ/Ϫ mice would present with an osteopetrotic bone phenotype. However, in agreement with the data from our in vitro BMC experiments, analyses of the skeleton in 18-week-old CD47 Ϫ/Ϫ mice indeed demonstrate an osteopenic bone phenotype including a reduced trabecular bone volume, bone mineral density, bone formation rate, mineral apposition rate, and osteoblast number. Furthermore, aged 28-weekold CD47 Ϫ/Ϫ mice showed a persistent osteopenic bone phenotype with reduced numbers of osteoblasts and osteoclasts. Our bone phenotype data in CD47 Ϫ/Ϫ mice, on the Balb/c background, are in part confirmed by data from C57BL/6 CD47 Ϫ/Ϫ mice where an osteopenic bone phenotype was demonstrated (14). Although lack of CD47 was not associated with a reduced number of osteoclasts in the bones of C57BL/6 mice, which could possibly be explained by strain differences (13,14). Van Beek et al. (24) showed that the SIRP␣ mutant mice had a decreased cortical bone volume, suggested to be due to an increased bone resorbing activity of osteoclasts lacking SIRP␣. OCTOBER 11, 2013 • VOLUME 288 • NUMBER 41

Lack of CD47 Impairs Bone Cell Differentiation
In conclusion, in a physiologically relevant in vitro system where stromal cells/osteoblasts regulate differentiation and formation of osteoclasts, lack of CD47 or the signaling domain of SIRP␣ in stromal cells results in a dramatically impaired osteoclast formation. These findings identify a mechanism where CD47 induces tyrosine phosphorylation of SIRP␣ in bone marrow stromal cells, which appears to be pivotal for normal osteoblast differentiation and ability to stimulate osteoclast formation. The biological importance of this mechanism is also confirmed by the osteopenic bone phenotype in CD47 Ϫ/Ϫ mice.