RANKL from Bone Marrow Adipocytes Regulates Osteoclast Formation and Bone Loss in Mice

Receptor activator of NF-κB ligand (RANKL) is essential for osteoclast formation. The cellular source of RANKL for osteoclastogenesis has not been fully uncovered. Bone marrow (BM) adipocytes derived from bone marrow mesenchymal stromal cells (BMSCs) express RANKL. Here we demonstrated that the AdipoqCre could target bone marrow adipocytes. We crossed the AdipoqCre mice with ranklfl/fl mice to conditionally delete RANKL from BM adipocytes. Conditional deletion of RANKL increased cancellous bone mass in the long bones of growing and adult mice by reducing the formation of trabecular osteoclasts and inhibiting bone resorption but did not affect cortical bone thickness or resorption of calcified cartilage. AdipoqCre; ranklfl/fl mice exhibited resistance to estrogen deficiency and rosiglitazone (ROS) induced trabecular bone loss but showed bone loss induced by unloading. BM adipocytes therefore represent an essential source of RANKL for the formation of trabecula osteoclasts and resorption of cancellous bone during remodeling.


Introduction
Osteoclasts are specialized multinucleated cells derived from the monocytemacrophage hematopoietic lineage and are responsible for resorption of bone matrix (1). During longitudinal bone growth, hypertrophic chondrocytes attract osteoclasts and blood vessels to direct the matrix mineralization (2). Under normal conditions, trabecular bone is resorbed periodically by osteoclasts followed by osteogenesis in the cavities, in a process known as remodeling (3). Excess bone resorption by osteoclasts results in pathological bone loss disorders, including postmenopausal osteoporosis, Paget's disease, and others (4).
Receptor activator of NF-κB ligand (RANKL) is an essential mediator of osteoclast formation during bone resorption (5). In contrast to previous hypothesis, the RANKL from osteoblasts and osteoblast progenitors does not contribute to bone remodeling in adult bone. Instead, matrix-embedded hypertrophic chondrocytes and osteocytes serve as the primary sources of RANKL for mineralized cartilage resorption in the development of growing bone and remodeling of adult bone, respectively (6). Initially RANKL is produced in a membranous form that can be cleaved by proteases to produce soluble RANKL (7). Because most matrix-embedded cells do not come in direct contact with osteoclast progenitors, these cells may control osteoclast formation through the production of soluble RANKL (8). Although it does contribute to osteoclast formation and the resorption of cancellous bone in adult bone, soluble RANKL does not affect cancellous bone mass or structure in growing bone (9). Moreover, bone loss caused by estrogen deficiency in ovariectomized mice is not prevented or reduced by a lack of soluble RANKL (9). These results indicate that the membrane-bound form of RANKL is sufficient for regulating osteoclast formation in a direct contact manner. As RANKL is expressed in various cell types, the cellular source of RANKL for osteoclastogenesis during bone growth and estrogen deficiency remains unclear.
Unlike white or brown adipose tissue, bone marrow (BM) adipocytes derived from BM mesenchymal stromal cells (BMSCs) by lineage tracing have been implicated in osteoclastogenesis and hematopoiesis (10,11). The number of BM adipocytes increases with age as well as under certain pathological conditions (12). BM adipocytes express RANKL, a phenotype not seen in either white or brown adipose tissue (13), and increased BM adipose tissue is associated with increased osteoclast formation (14).
Based on these observations, we hypothesized that BM adipocytes could serve as a source of RANKL for osteoclast formation and pathological bone resorption. We conditionally deleted RANKL in BM adipocytes by crossing the Adipoq Cre mice with rankl fl/fl mice. We found that RANKL derived from BM adipocytes did not affect resorption of calcified cartilage in the growth plate or cortical bone but did contribute to osteoclast formation and resorption of the cancellous bone in the longitudinal bone.
Mice lacking RANKL in BM adipocytes were protected from bone loss after ovariectomy or rosiglitazone administration but not from bone loss induced by unloading. BM adipocytes therefore represent an essential source of RANKL for the trabecular osteoclast formation and cancellous bone resorption in physiological and pathological bone resorption.
In isolated bone marrow cells from Adipoq Cre ; R26-tdTomato flox mice, immunofluorescent staining showed that RANKL colocalized with all tdTomato + cells, which indicated that all BM adipocytes expressed RANKL (Fig. 1C). To verify the specificity of RANKL expression in BM adipose tissue, we examined the rankl gene transcription in epididymal adipose tissue (eWAT), inguinal adipose tissue (iWAT), interscapular brown adipose tissue (iBAT) and BM adipose tissue (BMAT) from wild type C57BJ/6L mice by qPCR as previous reported. (13) The results showed that different from peripheral adipocytes, the BM adipocytes highly expressed rankl (Fig.   1D). Then we isolated BMSCs from Adipoq Cre ; rankl fl/fl mice and induced adipogenic differentiation. During adipogenesis, expression of RANKL in BMSCs decreased markedly in Adipoq Cre ; rankl fl/fl mice but remained unchanged in rankl fl/fl mice ( Fig.   1E-F). These results demonstrate that the Adipoq Cre efficiently deletes RANKL in bone marrow adipocytes and but not in chondrocytes, osteocytes or mesenchymal progenitors.

RANKL knockout in BM adipocytes increased trabecular bone mass
To assess the role of BM adipocyte RANKL in bone mass, we established BM adipocyte RANKL knockout mice by crossing the Adipoq Cre mice with rankl fl/fl mice ( Fig. S3A). Both male and female mice were used in the study. Conclusions are based on data analyses from both sexes, although only data from females are presented here.
For both male and female mice, the body weight and body length were recorded from birth to 8 weeks (Fig. S3B). And Adipoq Cre ; rankl fl/fl mice showed a slightly increase of body weight compared to their control littermates. RANKL deletion in BM adipocytes did not affect RANKL expression in spleen (Fig. S3C) by immunohistochemistry, spleen ( Fig. S3D) and lymph node (Fig. S3E) development by HE staining, or differentiation of T and B cells by flowcytometry ( Fig. S3F-G).
We harvested the femurs of Adipoq Cre ; rankl fl/fl and rankl fl/fl mice at 4, 8 and 16 weeks.
Quantitative computed tomography (μ-QCT) analyses revealed significant increases in trabecular bone mass of Adipoq Cre ; rankl fl/fl mice compared to rankl fl/fl mice as confirmed by increased bone mineral density (BMD), trabecular bone volume (BV/TV), trabecular number (Tb.N) and decreased trabecular spacing (Tb.Sp) ( Fig. 2A-B). No differences in cortical bone thickness or BMD were found between the two groups at 8 weeks ( Fig. 2C-D). Taken together, these results indicate that deleting RANKL from BM adipocytes increase trabecular bone mass.

BM adipocyte RANKL is essential for trabecular bone remodeling
Tartrate-resistant acid phosphatase (TRAP) staining and OCN immunofluorescence staining was performed to assess the roles of BM adipocyte-derived RANKL in trabecular bone remodeling. TRAP staining of trabecular bone revealed a significant decrease in the number of TRAP + cells (osteoclasts) in the bone marrow around the trabecular bones of Adipoq Cre ; rankl fl/fl mice compared to rankl fl/fl mice at 8 weeks ( Fig.   3A a1 vs. a4, B). The number of TRAP + cells under the growth plate ( Fig. 3A a2 vs. a5, C) and periosteal TRAP + cells ( Fig. 3A a3 vs. a6, D) did not show significant difference.
The thickness of growth plate cartilage (Fig. S4A) and tooth eruption ( Fig. S4B) were not significantly different between two groups.
OCN immunofluorescence staining showed fewer osteoblasts around the trabeculae in Adipoq Cre ; rankl fl/fl mice than in rankl fl/fl mice at 8 weeks ( Fig. 3E-F). Calcein staining revealed less trabecular bone formation in Adipoq Cre ; rankl fl/fl mice relative to rankl fl/fl mice at 8 weeks (Fig. 3G). By contrast, the endosteal and periosteal bone formation was not significantly different between two groups ( Fig. S4C-D). The bone histomorphometric parameters of Adipoq Cre ; rankl fl/fl mice and rankl fl/fl mice were showed in Table 1. The above results suggested that RANKL deletion from BM adipocytes impeded trabecular osteoclasts formation and the following osteogenesis and bone formation, namely bone remodeling.
We isolated BMSCs and in vitro Alizarin red staining after osteogenic differentiation showed no significant difference between the two groups ( Fig. S5A). Expression of alkaline phosphatase (ALP) and OCN was comparable in Adipoq Cre ; rankl fl/fl and rankl fl/fl mice ( Fig. S5B-C). In vitro oil red O staining and expression of LPL and PPARγ after adipogenic differentiation of BMSCs revealed no significant difference between Adipoq Cre ; rankl fl/fl and rankl fl/fl mice ( Fig. S5D-F). The above results indicated that RANKL deletion from BM adipocytes did not affect BM osteogenesis and adipogenesis.
Next, we examined serum CTX-1, OCN levels and bone marrow RANKL levels. A recent study reported that the Adipoq Cre targets PDGFR + VCAM-1 + stromal cells, (18) and our previous work showed that RANKL signaling in BMSCs negatively regulates osteoblastogenesis and bone formation. (19) To rule out the effects of RANKL signaling on osteoblast differentiation of BMSCs in vivo, we generated the Adipoq Cre ; rank fl/fl conditional knockout mice. Micro-CT analyses revealed no significant differences between rank knockout in BM adipocytes and rank fl/fl mice at any time point tested ( Fig. S6A-B). Overall, these results indicate that BM adipocytes are an essential source of RANKL for trabecular bone osteoclast formation and subsequent remodeling.

BM adipocyte RANKL did not mediate skeletal mechanical loading
Bone constantly adapts its structure in response to mechanical signals under the control of osteoblastic cells, which senses mechanical loading and regulates osteoclasts formation and trabecular bone remodeling through RANKL (6). To further determine if BM adipocyte derived RANKL mediates skeletal mechanical loading, we carried out unloading test by tail suspension in 8-week old mice for 4 weeks. In each group, micro CT results showed the trabecular bone BMD, BV/TV and Tb.N were significantly decreased and the Tb.Sp was increased ( Fig. 4A-B). The cortical changes were consistent with trabecular bones (Fig. 4C-D). The Adipoq Cre ; rankl fl/fl mice did not prevent bone loss induced by tail suspension. The results support that BM adipocyte RANKL does not mediate skeletal mechanical loading.

BM adipocyte RANKL mediated osteoclastogenesis and bone loss in ovariectomized mice
Increased BM adipose tissue is associated with increased fracture risk in postmenopausal osteoporosis (20). To examine the roles of RANKL from BM adipocytes in bone resorption after estrogen withdrawal, we used a mouse model of ovariectomy (OVX)-induced bone loss. Immunofluorescence staining of adiponectin revealed a significant increase in the number of BM adipocytes following OVX in both Adipoq Cre ; rankl fl/fl and rankl fl/fl mice, with no statistical difference between groups ( Fig.   S7A-B). At 6 weeks post-ovariectomy, trabecular bone mass was significantly decreased in rankl fl/fl mice, a phenotype not seen in Adipoq Cre ; rankl fl/fl mice (Fig. 5A-B). Consistent results were evident via H&E staining (Fig. S7C). TRAP staining showed that in rankl fl/fl mice, after OVX the number of osteoclasts significantly increased while in Adipoq Cre ; rankl fl/fl mice the number did not significantly change ( Fig. 5C-D). The OCN staining for osteoblasts demonstrated consistent results with TRAP staining (Fig. 5E-F). These results indicated that increased bone remodeling after OVX was inhibited in Adipoq Cre ; rankl fl/fl mice compared to their littermates.
ELISA analyses revealed increased serum CTX-1 (Fig. S7D), OCN (Fig. S7E) and bone marrow RANKL (Fig. S7F) levels in rankl fl/fl OVX mice relative to non-OVX controls, with no significant changes in protein abundance in Adipoq Cre ; rankl fl/fl OVX mice. These results demonstrate that RANKL from BM adipocytes is an essential mediator of excess osteoclast formation, bone resorption and increased bone turnover after estrogen withdrawal.

BM adipocyte RANKL contributed to rosiglitazone-induced osteoclastogenesis and bone loss
Rosiglitazone (ROS), a PPARγ activator used to treat type 2 diabetes, is associated with increased BM adipogenesis and fracture risk in both men and women (21,22). The mechanisms underlying these effects are still poorly understood but are likely attributable to increased adipogenesis due to PPARγ activation (23). To explore the roles of BM adipocyte RANKL in this process, we used a model of ROS-induced bone loss. In this model, ROS (10 mg/kg) was administered for 6 weeks. Adiponectin immunofluorescent staining revealed a substantial increase in the number of BM adipocytes after ROS administration in both Adipoq Cre ; rankl fl/fl and rankl fl/fl mice, with no significant differences between the groups (Fig. S8A-B). Trabecular bone mass showed a significant decrease in rankl fl/fl mice but no decrease in Adipoq Cre ; rankl fl/fl mice ( Fig. 6A-B). H&E staining showed consistent results with CT analyses (Fig. S8C).
TRAP staining showed that ROS treatment significantly increased the number of osteoclasts in rankl fl/fl mice but not in Adipoq Cre ; rankl fl/fl mice (Fig. 6C-D). OCN immunofluorescent staining showed that the number of osteoblasts was significantly increased in rankl fl/fl mice after ROS treatment but not in Adipoq Cre ; rankl fl/fl mice ( Fig.   6E-F). Serum CTX-1 (Fig. S8D) and OCN levels were increased in response to ROS in rankl fl/fl mice (Fig. S8E) but not in Adipoq Cre ; rankl fl/fl mice. BM RANKL levels showed no significant differences in Adipoq Cre ; rankl fl/fl mice after treated with ROS (Fig. S8F). These results indicate that increased BM adipocytes are an important source of RANKL for osteoclast formation and bone loss after ROS administration.

Discussion
As an essential factor for osteoclastogenesis, RANKL is widely expressed by a variety of cells types within the bone marrow (24). The cellular source of RANKL for osteoclast formation has not been fully revealed. Mice harboring a conditional knockout of RANKL using Prx1 Cre demonstrate no osteoclast formation, which indicates that mesenchyme-derived cells are an essential source of RANKL in the development of long bones (6). Among the mesenchyme-derived cells, matrix-embedded hypertrophic chondrocytes and osteocytes are the major source of RANKL for osteoclastogenesis, not osteoblasts or their progenitors. Hypertrophic chondrocytes control the resorption of calcified cartilage during bone development, while osteocytes regulate bone remodeling by sensing mechanical changes (6,25). And osteocytes are an essential source of RANKL for cancellous bone remodeling in adult mice but not younger mice (6,24). RANKL is produced in both soluble and membrane-bound forms. Since matrixembedded cells have limited direct contact with osteoclast progenitors and in postmenopausal osteoporosis plasma soluble RANKL is significantly increased (26), it is proposed that the soluble RANKL is a major form to regulate osteoclastogenesis.
Nevertheless, growing mice lacking soluble RANKL showed no deficits in bone mass and equal amounts of bone loss after estrogen deprivation compared to controls (9).
These studies indicate that membrane-bound RANKL is sufficient for bone growth and is responsible for bone loss induced by estrogen deficiency. Therefore, it is mysterious how matrix-embedded cell-derived RANKL regulates osteoclastogenesis and bone remodeling.
Although marrow adipose tissue accounts for almost 70% of the adult bone marrow volume in humans, the functions of BM adipocytes are still very mysterious (27).
Clinically, increased marrow adipose tissue is associated with increased fracture risk in diseases such as postmenopausal osteoporosis, anorexia nervosa and diabetes (20).
Mesenchyme-derived BM adipocytes express RANKL, which is distinctive from other adipose tissues, including both white and brown adipose (9). Several studies have explored the roles of bone marrow adipocytes by using the Adipoq Cre (11,16). Zhou et al. reported that bone marrow adipocytes promoted the regeneration of stem cells and hematopoiesis by secreting SCF (11). The author demonstrated that four weeks after gavaging 6-week-old Adipoq Cre/ER ; R26-tdTomato flox mice with tamoxifen, tdTomato was expressed in 93 ± 5% of perilipin + adipocytes. The tdTomato expressing in the bone marrow besides adipocytes was a subset of LepR + stromal cells: 5.9 ± 3.1% of LepR + cells in Adipoq Cre/ER ; R26-tdTomato flox bone marrow were tdTomato + . They  (9). Our findings showed that the bone marrow soluble RANKL levels were not significantly different between Adipoq Cre ; rankl fl/fl and rankl fl/fl mice, which indicated that BM adipocytes are not an important source of sRANKL production. In some pathological conditions, including PMOP, periodontitis, and inflammatory joint disease, soluble RANKL levels are elevated (33)(34)(35), which indicates that soluble RANKL is functionally involved in these conditions. However, mice lacking soluble RANKL showed a similar amount of bone loss after estrogen withdrawal (9), which implies that soluble RANKL is not essential for osteoclasts formation and bone resorption after estrogen withdrawal. In our study, no bone loss was observed in Adipoq Cre ; rankl fl/fl mice after estrogen withdrawal.
Osteoclast formation and bone resorption were significantly inhibited following the deletion of RANKL in BM adipocytes. Furthermore, the bone turnover rate was significantly lower in Adipoq Cre ; rankl fl/fl mice compared to their littermate controls.
These results suggest that RANKL from BM adipocytes contributes to excess osteoclast formation and bone resorption in pathological bone loss diseases with increased BM adiposity.
ROS is widely clinically used to lower blood glucose levels in diabetic patients via the activation of PPARγ. However, the use of ROS is associated with an increased risk of fractures as well as an increase in BM adipocyte counts due to the activation of PPARγ, as essential mediator of adipocyte differentiation and function (21,22), Although the role of PPARγ in osteoclast differentiation is still a topic of considerable debate (23,(36)(37)(38), our findings showed that administering ROS to Adipoq Cre ; rankl fl/fl mice failed to increase osteoclast formation and bone resorption or induce bone loss, as observed in rankl fl/fl mice. These results indicate that ROS increases osteoclast formation and bone resorption at least in part by increasing the number of BM adipocytes.
Osteocytes sense mechanical loading and secrete RANKL to control osteoclast formation and bone remodeling in adult mice and unloading-induced bone loss (6,25).
Since the Adipoq Cre does not target osteocyte, we further testify whether adipocyte RANKL mediates skeletal mechanical loading. Tail-suspension did induce a significant loss of cancellous bone volume in both Adipoq Cre ; rankl fl/fl and rankl fl/fl mice, indicating adipocyte RANKL did not mediate bone mechanical loading.
Taken together, our study demonstrates that the RANKL from BM adipocytes serves as an essential source of trabecular bone remodeling in both physiological and pathological conditions but does not mediate resorption of calcified cartilage in growth plate or regulate cortical bone remodeling.

In vivo heterotopic bone formation assay
The sorted tdTomato + cells from Adipoq Cre ; R26-tdTomato flox mice and BMSCs were collected and 0.5 × 10 6 cells mixed with 40 mg hydroxyapatite in 100 μl α-MEM and incubated overnight. Then the cells were implanted under the kidney capsules of BALB/c nude mouse for 4 weeks. After retrieval, it was paraformaldehyde-fixed and paraffin-embedded. The slice was stained with OCN, Aggrecan, Perilipin and positive cells were counted.

Micro-CT and X ray
Bone was fixed with 4% paraformaldehyde for 24 h and analyzed by Skyscan 1172 high-resolution micro-CT (Skyscan, Antwerp, Belgium). Scan conditions were set at 8 µm per pixel resolution, 80 kV voltage, and 124 μA current. Using these images, we constructed three-dimensional models and analyzed images by CTAn and CTVol, including bone mineral density (BMD), bone volume/tissue volume (BV/TV), trabecular number (Tb.N), trabecular separation (Tb.Sp), and cortical thickness. And the tooth eruption was evaluated by the X ray.

Histochemistry, immunohistochemistry, and immunofluorescent staining
Femurs were fixed in 4% paraformaldehyde for 24 h and decalcified in 10% EDTA (room temperature) for 2 weeks. After dehydration in an alcohol gradient (50%, 75%, 85%, 95%, or 100%), femurs were embedded in paraffin, cut into 4 µm thick sections Then the sections were incubated with corresponding fluorescence-conjugated secondary antibodies (1:200) and DAPI (1:500) for 1 h at room temperature. Then the samples were scanned on either a Nano Zoomer or 3D Histech system, with exported images analyzed using Case Viewer and NDP view software.

Calcein staining
To detect mineral deposition, we injected mice with alizarin red S (Sigma; 40 mg/kg body weight) 10 days and calcein (Sigma; 20 mg/kg body weight) 3 days before euthanasia. We performed undecalcified bone slicing and calculated the mineral apposition rate (MAR) of trabecular bone and cortical bone.

Statistics
Data are presented as means ± SDs. Independent-samples t tests and one-way ANOVAs were used to assess statistical significance. All analyses were performed using SPSS 21.0 (IBM, Chicago, IL, USA). Statistical significance was set at P < 0.05. Immunofluorescence staining of RANKL and Hoechst in bone marrow cells from