Osteoblast-specific expression of Fra-2/AP-1 controls adiponectin and osteocalcin expression and affects metabolism

Bone-forming osteoblasts and mesenchyme-derived adipocytes share common progenitor cells. Adipokines, such as leptin, resistin and adiponectin (Adipoq) are primarily produced by adipocytes and are important regulators of metabolism (Kawai et al., 2009; Shetty et al., 2009). Bone has recently emerged as an endocrine organ regulating glucose metabolism through osteoblast-specific secretion of osteocalcin (Ocn) (Karsenty and Ferron, 2012). Osteoblast-specific expression of the transcription factors FoxO1 and ATF4 regulate glucose and insulin metabolism, energy expenditure and fertility through transcriptional regulation of Esp (osteotesticular protein tyrosine phosphatase) (Rached et al., 2010; Yang and Karsenty, 2004; Yang et al., 2004; Yoshizawa et al., 2009), thus decreasing Ocn decarboxylation (Ferron et al., 2008; Ferron et al., 2010; Oury et al., 2011; Yoshizawa et al., 2009). ATF4 is a member of the activator protein (AP)-1 transcription factor consisting of a variety of dimers composed of members of the Fos, Jun and ATF proteins. In mice loss- and gain-of-function mutations of Fos and ATF members, and to a lesser extent of Jun members, demonstrate that these factors are important for osteoblast differentiation and function (Wagner and Eferl, 2005). Whereas Fra-1 (Fos-related antigen 1, Fosl1) and ΔFosB (the short isoform of FosB) promote osteoblast differentiation and activity (Eferl et al., 2004; Kveiborg et al., 2002; Lefterova and Lazar, 2009; Zhao et al., 2008), Fra-2 (Fos-related antigen 2, Fosl2) is important in normal physiology and disease, such as in pulmonary fibrosis (Eferl et al., 2008). Fra-2 also regulates chondrocyte differentiation (Karreth et al., 2004) and bone remodeling by controlling osteoclast size and activity (Bozec et al., 2008). Furthermore, Fra-2 controls bone formation by transcriptional regulation of the collagen isoform Col1a2 and Ocn expression in mouse and human osteoblasts (Bozec et al., 2010). However, the effects of Fra-2 expression beyond bone, for example, on metabolism, have not yet been described.

Here, we show that osteoblast-specific Fra-2 expression regulates Adipoq and Ocn levels, thereby affecting systemic glucose and insulin metabolism. Importantly, Fra-2 expression in osteoblasts represses Adipoq at the transcriptional level. These data provide a new link between the endocrine function of the skeleton and systemic metabolism regulated by AP-1.

We generated osteoblast-specific Fra-2 mutant mice to investigate the role of Fra-2 in osteoblasts on bone homeostasis and metabolism. Mice with the Osx-tTA-Cre allele were crossed to mice carrying two different Fra-2 alleles to either delete Fra-2 (Fosl2^Δob) or express Fra-2 (Fosl2^ob-tet) in osteoblasts (Rodda and McMahon, 2006). Both mutant mice were maintained without doxycycline to allow Fra-2 deletion or transgene expression during development. Bones of Fosl2^Δob mutants expressed lower levels of Fra-2 mRNA and protein, whereas Fosl2^ob-tet mice expressed higher levels of Fra-2 (supplementary material Fig. S1a–f). No change in Fra-2 mRNA or protein expression was detected in other organs, such as pancreas, muscle, liver or inguinal fat pad in either Fosl2^Δob or Fosl2^ob-tet mice (supplementary material Fig. S1a–f). Furthermore, immunohistochemical stainings of Fra-2 in long bones showed increased Fra-2 expression in osteoblasts and chondrocytes in Fosl2^ob-tet mice, whereas in Fosl2^Δob mice, Fra-2 staining was hardly detectable (supplementary material Fig. S1c,d). This indicates that Fra-2 expression or deletion is restricted to mesenchymal cells of the bone compartment.

Histomorphometric analyses of bones from two-month-old Fosl2^Δob mice revealed a decrease in bone volume, trabecular thickness and number, whereas an increase in these parameters was observed in Fosl2^ob-tet mice (Fig. 1A,B). Fosl2^Δob mice displayed a decrease in osteoblast surface, whereas osteoclast surface was increased, suggesting disturbed bone homeostasis (Fig. 1A). In contrast, Fosl2^ob-tet mice displayed increased osteoblast surface and decreased osteoclast surface (Fig. 1B). Adipocyte number and size were similar in control and Fosl2^Δob or Fosl2^ob-tet tibia (supplementary material Fig. S2a,b). Bone formation assessed by calcein labeling was decreased in Fosl2^Δob mice and increased in Fosl2^ob-tet animals, whereas, no statistical difference in resorption defined by serum CTX could be detected in the two mutant strains (Fig. 1A,B). Furthermore, expression of type 1 collagens and Ocn were decreased in Fosl2^Δob long bones and increased in Fosl2^ob-tet long bones (Fig. 1C,D). Circulating levels of Ocn were decreased in Fosl2^Δob mice and increased in Fosl2^ob-tet mice (Fig. 1E). These data indicate that changes in the bone-forming activity of osteoblasts are responsible for the low and high bone mass phenotype observed in Fra-2 mutant mice. In vitro differentiation experiments were next performed to analyze whether the bone phenotype was cell autonomous. Osteoclast differentiation and osteoclast marker expression, such as Acp5 (also known as TRAP) and cathepsin K, were not altered (supplementary material Fig. S3a–c). Fosl2^Δob primary osteoblasts produced fewer mineralized nodules, when cultured in osteogenic conditions, whereas increased nodule formation was observed in Fosl2^ob-tet cultures (supplementary material Fig. S4a,c). In addition, collagen content, and Col1a1, Col1a2 and Ocn mRNA levels were decreased in Fosl2^Δob cultures and increased in Fosl2^ob-tet cultures (supplementary material Fig. S4a–d). To determine whether adipocytes, which share common precursors with osteoblasts were altered, mesenchymal progenitors isolated from Fosl2^ob-tet and Fosl2^Δob calvariae were differentiated into adipocytes. Oil Red O staining and expression analyses of peroxisome proliferator-activated receptor γ (PPARγ) and CCAAT/enhancer-binding protein α (CEBPα), early adipocyte markers, revealed a comparable adipogenic potential of Fosl2^Δob and Fosl2^ob-tet cells when compared to control cells, whereas these markers were hardly detectable in osteoblast cultures (supplementary material Fig. S5a–c). These data demonstrate that genetic manipulation of Fra-2 in osteoblasts specifically affects the bone-forming activity of the osteoblast lineage.

Fra-2 was recently shown to control leptin expression in adipocytes (Wrann et al., 2012). We therefore measured the levels of serum adipokines and Ocn isoforms in Fra-2 mutant mice. No changes were observed for leptin and resistin (supplementary material Fig. S6a,b), whereas using the hydroxyapatite (HA) assay followed by enzyme-linked immunosorbent assay (ELISA), total, carboxylated and undercarboxylated Ocn levels were decreased in Fosl2^Δob sera and increased in Fosl2^ob-tet sera (supplementary material Fig. S6c,d). Interestingly, total Adipoq levels were significantly increased in Fosl2^Δob sera and decreased in Fosl2^ob-tet sera (Fig. 2A,B). Moreover, the high molecular weight (HMW) Adipoq, which is the active form (Tilg and Moschen, 2006) as well as the low (LMW) and middle molecular weight (MMW) forms of Adipoq with less defined functions (Takeda et al., 2002) were similarly altered (Fig. 2A,B). Surprisingly, Adipoq expression was unchanged in its main production site, the adipose tissue of Fosl2^ob-tet or Fosl2^Δob mice (supplementary material Fig. S6e). Interestingly, we found that Adipoq mRNA levels were increased in long bones of Fosl2^Δob and decreased in Fosl2^ob-tet mice, suggesting that Adipoq can be produced in bone cells (Fig. 2C). To determine which cells can express Adipoq in long bones, Adipoq production and expression were measured in Fosl2^ob-tet and Fosl2^Δob primary cells cultured in osteogenic or adipogenic conditions. Despite a lower level of Adipoq mRNA and protein in osteoblast cultures compared to adipocyte cultures, Adipoq production and expression by mineralized osteoblasts was increased in the absence of Fra-2 and decreased in Fra-2 overexpressing cells, whereas no difference was observed in cells cultured in adipogenic conditions (Fig. 2D; supplementary material Fig. S6f,g). These findings were confirmed by western blot analyses (Fig. 2E,F).

Moreover, sections from transgenic reporter mice (Adipoq-cre; ROSA26-lox-stop-lox-RFP) were stained for red fluorescent protein (RFP) and Ocn (Eguchi et al., 2011). Positive staining for RFP was observed in Adipoq-cre reporter fat cells (Fig. 2G). Interestingly, osteoblasts, osteocytes and adipocytes were positively stained in Adipoq-cre reporter long bone sections (Fig. 2H). Importantly, the RFP staining was localized to areas positive for Ocn staining on subsequent sections (Fig. 2H). Moreover, co-immunofluorescence analyses showed co-expression of RFP and Ocn in long bones (Fig. 2H). These results indicate that bone is a potential source of Adipoq and that Adipoq expression and secretion by osteoblasts is Fra-2 dependent.

To determine whether Fra-2 controls Adipoq transcription in osteoblasts, a 5 kb Adipoq promoter fragment, which includes a putative non-consensus binding region for Fra-2 at position −3.8 kb, was analyzed. Using chromatin immunoprecipitation (ChIP), with anti-Fra-2 antibodies on Fosl2 control and knockout primary osteoblasts, Fra-2 binding at this site was observed (Fig. 3A). To further characterize the specificity and potential dimerizing partners of Fra-2 on the Adipoq promoter, ChIP was performed for IgG and other AP-1 transcription factors. Fosl2-knockout osteoblasts were used as a control for Fra-2 antibody specificity and to identify the dimerization partner of Fra-2. No binding of IgG, ATF-4 and Fra-1 to the Adipoq promoter could be detected and Fra-2 did not bind a region around −1100 (Fig. 3A). However, FosB, c-Fos and JunB could bind the Adipoq promoter in the presence or absence of Fra-2 (Fig. 3A). JunD and c-Jun clearly bound the Adipoq promoter in wild-type cells, whereas decreased binding of c-Jun and no binding of JunD was detected in the absence of Fra-2, suggesting that in osteoblasts c-Jun and JunD probably bind to this fragment as a heterodimer with Fra-2 (Fig. 3A). Additional ChIP assays were performed with antibodies directed against methylated histone H3 to analyze chromatin modifications on the Adipoq promoter (Fig. 3B). ChIP and quantitative PCR (qPCR) analyses of the Adipoq promoter revealed high levels of active (H3K4me3) and decreased repressive (H3K27me3) methylation marks in Fra-2-deficient cells. This indicates that the Adipoq promoter was more active in the absence of Fra-2 (Fig. 3B). ChIP on ChIP Fra-2 and c-Jun, and Fra-2 and JunD confirmed that c-Jun and JunD might be the dimerizing partner of Fra-2 on the Adipoq promoter because in the absence of Fra-2, neither c-Jun nor JunD antibodies were able to pull down the Adipoq promoter chromatin fragment (Fig. 3C). In addition, Adipoq promoter activity assessed by luciferase assay was decreased by the addition of Fra-2, c-Jun, JunD or a combination of these three AP-1 members, confirming that these factors are repressing Adipoq expression (Fig. 3D,E). These data demonstrate that Fra-2 can repress Adipoq transcription in osteoblasts.

Because Ocn and Adipoq are important regulators of metabolism, we next analyzed the metabolic parameters of osteoblast-specific Fra-2 mutant mice. Six-week-old mice were placed on a normal diet (ND) or on a high fat diet (HFD) for six weeks. Fosl2^Δob mice were heavier under both ND and HFD conditions with an increased gonadal fat pad weight (Fig. 4A), whereas Fosl2^ob-tet mice were leaner with decreased weight and decreased fat pad weight under both conditions (Fig. 4B). Insulin and glucose levels were next analyzed. Although mild difference in glucose and insulin levels could be detected in Fosl2^Δob mice (Fig. 4C,D), Fosl2^ob-tet mice had lower serum glucose levels and lower circulating insulin compared to controls under all tested conditions (Fig. 4E,F). In addition, glucose-stimulated insulin secretion tests (GSIS) indicated that the relative insulin response to glucose stimulation was higher in Fosl2^ob-tet mice (supplementary material Fig. S7a). Insulin content was next measured in Fosl2^ob-tet bones and pancreata. Bone insulin content was not altered, whereas insulin content was decreased in Fosl2^ob-tet pancreata, suggesting a defect in insulin synthesis (supplementary material Fig. S7b). To further analyze glucose and insulin metabolism, glucose (GTT) and insulin (ITT) tolerance tests were employed. A decrease in glucose tolerance and insulin sensitivity was observed in Fosl2^Δob mice compared to littermate controls in ND and HFD conditions (Fig. 4G; supplementary material Fig. S7c). In contrast, Fosl2^ob-tet mice showed a remarkable increase in glucose tolerance and insulin sensitivity in both conditions (Fig. 4H; supplementary material Fig. S7d). Moreover, an insulin-induced increase of phosphorylated AKT (pAKT) was attenuated in Fosl2^Δob, but increased in Fosl2^ob-tet fat, liver and muscle (supplementary material Fig. S7e–g). These data demonstrate that Fra-2 expression in osteoblasts altered insulin and glucose responses, probably through affecting Ocn and Adipoq levels.

Finally, intravenous injections of adenoviruses expressing Adipoq (Ad-Adipoq) partially rescued the metabolic phenotype in Fosl2^ob-tet mice. The levels of circulating Adipoq were increased and similar in Fosl2^ob-tet and littermate control mice under a ND, when compared to PBS- or Ad-Empty-injected mice (supplementary material Fig. S8a). In addition, the levels of Ocn remained higher in Fosl2^ob-tet mice under all conditions (supplementary material Fig. S8b). This suggests that Ad-Adipoq injection can only modify the Adipoq levels within the physiological range without affecting Ocn secretion, allowing us to discriminate between the role of Adipoq and Ocn in our model. The weight of Fosl2^ob-tet mice after Ad-Adipoq injection remained lower than the weight of control mice, although the decrease did not reach statistical significance under a HFD (supplementary material Fig. S8c). Fat pad weight in ND was similar to controls after Ad-Adipoq injection (supplementary material Fig. S8d), whereas a lower fat pad weight was still observed under a HFD (supplementary material Fig. S8d). This suggests that Ocn is probably responsible for the remaining body and fat pad weight alterations observed in Fosl2^ob-tet mice. Importantly, glucose and insulin levels were comparable between mutant and control mice after Ad-Adipoq treatment in ND (supplementary material Fig. S8e,f). Under HFD, the levels of insulin remained lower in Fosl2^ob-tet mice, whereas glucose levels were restored (supplementary material Fig. S8d,e,f). GTT and ITT assays revealed a complete rescue of glucose and insulin handling in Ad-Adipoq-injected mice under ND (supplementary material Fig. S8g,h). In HFD after Ad-Adipoq injection, despite a slight increased insulin sensitivity persisting in Fosl2^ob-tet mice, Ad-Adipoq-injected mice had a similar response to glucose (data not shown). These data support our notion that osteoblast-specific Fra-2 expression can regulate glucose metabolism and that Adipoq produced by osteoblasts contributes to the defects in glucose and insulin handling.

In this report, we have shown a new endocrine role for osteoblast-specific Fra-2 expression, which regulates the transcription of Adipoq and Ocn leading to altered bone physiology and metabolic responses (Fig. 4I). Ocn (Bozec et al., 2010) and Adipoq were found to be transcriptionally regulated by Fra-2. The expression of Fra-2 in osteoblasts induced a leaner phenotype, lower glucose and insulin levels, an improved glucose response and insulin sensitivity in Fosl2^ob-tet mice.

Fos proteins have multiple functions in normal bone physiology, regulating bone mass through cell autonomous control of osteoblast differentiation (Bozec et al., 2010; Eferl et al., 2004; Grigoriadis et al., 1993). Distinct target genes are regulated by the Fos-related proteins Fra-1 and Fra-2, which control osteoblast differentiation and activity, rather than osteoblast numbers (Bozec et al., 2010; Eferl et al., 2004). Fra-2 transcriptionally regulates leptin, which can control energy homeostasis in mesenchymal cells (Wrann et al., 2012). Surprisingly, leptin was not affected in Fra-2 mutant mice, whereas the levels of the adipokine Adipoq were found to be altered in Fra-2 mutant sera. Adipoq is known to enhance the action of insulin by several mechanisms, including suppression of gluconeogenesis, regulation of fatty acid metabolism and modulation of Ca²⁺ signaling in skeletal muscle (Kadowaki et al., 2006). Adipoq self-associates into larger structures e.g. homotrimers (LMW), hexamers (MMW) or dodecamers (HMW), with the latter described as the bioactive form (Engl et al., 2007). In sera of Fosl2^ob-tet mice, a decrease in LMW and MMW, and HMW Adipoq is associated with a decrease in insulin levels, and increased insulin sensitivity. Adipoq has been described as an adipokine produced mainly by adipocytes in fat tissue; surprisingly, its expression was unchanged in the fat pad of osteoblast-specific Fra-2 mutant mice. Importantly, Adipoq was expressed in osteoblasts and osteocytes, confirmed by Adipoq reporter mice (Eguchi et al., 2011). These findings support previous reports indicating that Adipoq and its receptors are expressed in bone-forming cells in humans (Berner et al., 2004) and in mice (Shinoda et al., 2006) as well as our microarray data on osteoblast cultures (M.A., unpublished data). We found that Adipoq expression is controlled by Fra-2 in a cell-autonomous manner in osteoblasts; Fra-2 directly represses the transcription of Adipoq with JunD or c-Jun as dimer partners. The levels of Adipoq in the Fra-2 mutant mice are inversely correlated with the osteogenic potential of these cells in vitro and with bone mass in vivo. Studies in vitro and in vivo have yielded contradictory results regarding the effects of Adipoq on bone cell function. Adipoq has been shown to either stimulate or suppress osteogenesis in vitro (Lee et al., 2009; Shinoda et al., 2006; Williams et al., 2009) and the results from in vivo studies using Adipoq knockout or overexpressing mice are also conflicting (Ealey et al., 2008; Oshima et al., 2005; Shinoda et al., 2006; Williams et al., 2009). This indicates that Adipoq could additionally contribute to the bone phenotype by suppressing bone forming activity in osteoblasts in an autocrine fashion.

Our data suggest that Adipoq is a molecule linking bone physiology to systemic metabolism in an endocrine fashion, much like Ocn. Further experiments would be necessary to define its precise action given that the Fra-2 mutant mice described in this study show deregulation of at least two ‘bone hormones’, Adipoq and Ocn. Both forms of Ocn – the bioactive undercarboxylated Glu13 and the carboxylated Gla13-Ocn were induced by Fra-2. The Glu-Ocn controls insulin secretion and sensitivity (Ferron et al., 2010; Yoshizawa et al., 2009). Although Glu-Ocn has also been reported to induce Adipoq secretion by white adipocytes (Ferron et al., 2008; Lee and Karsenty, 2008; Lee et al., 2007), reduced circulating Adipoq was observed in Fosl2^ob-tet mice despite increased Ocn. This suggests that Adipoq expression might exert its action independently of Ocn in Fra-2 mutant osteoblasts. Moreover, administration of exogenous Adipoq increased the circulating Adipoq to levels comparable to controls. Interestingly, under all conditions Ocn levels remained high in Fosl2^ob-tet mice. Because the restoration of Adipoq levels through Ad-Adipoq in Fosl2^ob-tet mice was sufficient to rescue glucose and insulin responses under a ND, we conclude that decreased Adipoq is partially responsible for the metabolic phenotype. The absence of complete rescue of insulin responses when Fosl2^ob-tet mice were challenged with HFD might be due to the low Ad-Adipoq dose and/or to the timeline of the treatment.

Fra-2 has not yet been associated with a metabolic phenotype in humans, although it is possible that polymorphisms in Fra-2 exist. Fra-2 target genes, such as leptin, Adipoq and Ocn are associated with effects on glucose metabolism in mice and humans. Furthermore, increased Ocn levels in mice were associated with improved glucose handling and protection against the development of type 2 diabetes following diet-induced obesity (Ferron et al., 2012). The protection from diet-induced metabolic effects in Fosl2^ob-tet mice challenged with a HFD allowed us to speculate that Fra-2 in osteoblasts protects against the adverse effects of obesity. These data add new evidence to the notion that osteoblasts are endocrine cells and that the skeleton is an endocrine organ. Therefore, modulating the activity of AP-1 transcription factors in osteoblasts might provide means for targeted therapies against metabolic diseases.

Osteoblast-specific Fra-2 deficient mice (Fosl2^Δob) were generated by crossing Fosl2 flox/flox mice (Eferl et al., 2007) with mice carrying the Osx-tTA-Cre allele (Rodda and McMahon, 2006). The tetracycline (tet)-switchable Fosl2 allele was introduced 3′ of the col1a1 locus using a recombinase-mediated single-copy transgene integration strategy in embryonic stem cells (ESCs) (Beard et al., 2006). Southern blot analyses confirmed correct recombination in ESCs and correct germ-line transmission. Doxycycline-switchable expression of the FLAG-tagged mouse Fosl2 cDNA sequence and the dsRed reporter were confirmed in ESCs and in mice with a tetracycline activator expressed from the Rosa26 locus. Crossing mice with the tet-switchable Fosl2 allele with Osx-tTA-Cre mice generated the osteoblast-specific Fosl2^ob-tet mice. In the absence of doxycycline, these mice express Fra-2. Males were used for the metabolic studies. All mice were maintained on a mixed C57Bl6/129 background and littermates were used as controls.

All mice were bred and maintained without doxycycline to ensure Cre and Fosl2 expression before birth and in adult. Genotyping was performed by PCR analyses of genomic DNA from tail biopsies. The generation of Fosl2^−/− mice has been described previously (Bozec et al., 2008; Bozec et al., 2010; Eferl et al., 2007). Mutant mice expressing Td Tomato, a reporter for red fluorescent protein (RFP) under the control of the Adipoq promoter were obtained by crossing the lox-STOP-lox Td Tomato mice with Adipoq-Cre mice (Eguchi et al., 2011). For HFD experiments, six-week-old mice were challenged with a high fat diet (D12331 from Research Diet NC) for six weeks. All mouse experiments were performed in accordance with local and institutional regulations.

For the GTT assay, glucose (2 g/kg of body weight) was injected intraperitoneally (i.p.) after 6 hours of fasting, and blood glucose was monitored using blood glucose strips and the accu-check glucometer (Roche) at indicated times. For the ITT assay, insulin (0.55 U/kg of body weight) was injected i.p. after 3 hours of fasting and blood glucose levels were measured at indicated time.

Mice were fasted overnight and injected i.p. with vehicle or insulin (1 U/kg of body weight). At 15 minutes after injection, mice were killed and proteins were isolated from fat, liver and muscle. Western blotting for pAKT (Cell Signaling) and AKT (Cell Signaling) was performed.

Blood was collected by heart puncture. Serum levels of insulin (Mercodia), leptin (R&D Systems), resistin (R&D system), total Adipoq and HMW Adipoq (Alpco and Abcam) were quantified by ELISA. Protein from pancreas and bone were isolated, and insulin and protein levels were quantified using insulin ELISA (Mercodia) and the Bradford kit (PIERCE), respectively.

Sera from mice were added to hydroxyapatite (HA) to achieve a final concentration of 25 mg/ml. After 60 minutes HA beads were pelleted by centrifugation and HA-unbound osteocalcin was used. The osteocalcin level in supernatant and initial samples were measured by ELISA (Demetitec diagnostic).

Bones were fixed in 10% neutral formalin and decalcified. Bone were embedded in paraffin and sectioned at 5 µm. Immunohistochemical staining for Fra-2 (Abcam), RFP (molecular probes), Adipoq (Abcam and Santa Cruz Biotechnology), TRAP (Sigma) and Ocn (Enzo laboratories) were performed on deparaffinized sections using the LSAB or Envision detection kit (Dako). For immunofluorescence on paraffin sections, secondary antibodies used were conjugated to Texas Red (The Jackson Laboratory) and counterstainings were performed with DAPI. Double immunofluorescence was performed on paraffin sections using RFP and Ocn antibodies and secondary antibodies conjugated to Texas Red or FITC (Abcam) in a simulatenous protocol (Abcam). For histomorphometrical analysis, undecalcified lumbar spines were embedded in methylmethacrylate and 5-µm sections were cut along the sagittal plane. Sections were stained with Toluidine Blue, and modified von Kossa/van Gieson. Quantitative histomorphometry was performed on Toluidine-Blue-stained sections according to standard protocols (Parfitt et al., 1987) using the Osteomeasure histomorphometry system (Osteometrix). Experiments were performed in a blinded fashion.

Calvariae were sequentially digested for 30 minutes in modified Eagle's medium type α (α-MEM) containing 0.1% collagenase and 0.2% dispase. Cells isolated in fractions 2–3 were combined as an osteoblastic cell population, expanded for 2 days in α-MEM with 10% fetal calf serum (FCS) and plated at a density of 5×10⁵ cells/well. Medium was supplemented with 10 mM β-glycerophosphate and 50 µg/ml ascorbic acid for osteoblast cultures or with 1 mM dexamethasone (Sigma-Aldrich), and 10 mg/ml insulin (Sigma-Aldrich) for adipocyte cultures. After 2 weeks of culture, RNAs were extracted from osteoblast or adipocyte cultures. To control cell differentiation, Alizarin Red (Sigma-Aldrich) and Oil Red O (Sigma-Aldrich) staining were performed for osteoblast and adipocyte cultures.

Bone marrow cells were isolated from six-week-old mice, put into suspension overnight and cultured in 24 wells (5×10⁵ cells/well) for 2–7 days with macrophage colony-stimulating factor (20 ng/ml; R&D) and Rankl (5 ng/ml; R&D); TRAP straining (Sigma-Aldrich) was then performed.

Total RNA was isolated with the TRIzol protocol (Invitrogen). For bone samples whole long-bones, including marrow, was extracted. cDNA synthesis was performed using 2 µg RNA with the Ready-To-GoTM You-Prime It First-Strand-Beads and random primers (Amersham Biosciences) according to the manufacturer's protocols.

qPCR reactions were performed using SYBR Green (Molecular Probes) on MasterCycler ep Realplex (Eppendorf). Primers for genes analyses used for real-time PCR are available upon request. The comparative CT method was used to quantify the amplified fragments. RNA expression levels were normalized to those of actin. Results are shown as mean± s.d. of gene expression relative to actin.

Whole long bone, pancreas, fat pad and culture cells extracts were prepared with cell RIPA lysis buffer and western blots were performed with the antibodies against the following proteins: Fra-2 (Invitrogen), Flag (Sigma), Adipoq (Abcam and Santa-Cruz) and for normalizing, actin (Sigma).

ChIP was performed according to standard protocols with antibodies against Fra-2, Fra-1, FosB, c-Fos, JunB, JunD, c-Jun, ATF4 (Santa Cruz Biotechnology), K4 (Upstate) and K27 (kind gift from Thomas Jenuwein). The specific binding of Fra-2 antibody was verified in parallel using a corresponding pair of wt and ko osteoblasts in 3 independent experiments.

The promoter–reporter vectors of adiponectin promoter were generated using pGL4.23[luc2/minP] vector following the kit instructions (Promega). 6×10⁴ HEK293 cells/well were plated in 24-well dishes. 1.5 µg of the luciferase reporter construct, 0.2 µg of the Renilla internal control (pHRG-tk; Promega) and 0.2–1 µg of each AP-1 expression vector were co-transfected in triplicate using Lipofectamine (Invitrogen). cJun–Fra-2 forced dimer constructs were generated as previously described (Bakiri et al., 2002). Luciferase activity was quantified using the Dual Luciferase kit (Promega).

All experiments were repeated at least three times and performed in triplicate. Statistical analysis was performed using the two-tailed non-parametric Student's t-test. The significance levels are: *P<0.05, **P <0.01 and ***P<0.001. Data are shown as means±s.d.

We are very grateful to Jean-Pierre David, Rama Khokha, Michele Petrucelli, Francesco Real, Romeo Ricci, Mercedes Rincon, Özge Uluckan and Juan Guinea-Viniegra for critically reading the manuscript and Lionel Gresh for help in generating the Fosl2^ob-tet transgenic line. We also thank Thomas Jenuwein for providing the methyl histone antibodies.