Protein Phosphatase PP5 Controls Bone Mass and the Negative Effects of Rosiglitazone on Bone through Reciprocal Regulation of PPARγ (Peroxisome Proliferator-activated Receptor γ) and RUNX2 (Runt-related Transcription Factor 2)*

Peroxisome proliferator-activated receptor γ (PPARγ) and runt-related transcription factor 2 (RUNX2) are key regulators of mesenchymal stem cell (MSC) differentiation toward adipocytes and osteoblasts, respectively. Post-translational modifications of these factors determine their activities. Dephosphorylation of PPARγ at Ser-112 is required for its adipocytic activity, whereas phosphorylation of RUNX2 at serine 319 (Ser-319) promotes its osteoblastic activity. Here we show that protein phosphatase 5 (PP5) reciprocally regulates each receptor by targeting each serine. Mice deficient in PP5 phosphatase have increased osteoblast numbers and high bone formation, which results in high bone mass in the appendicular and axial skeleton. This is associated with a substantial decrease in lipid-containing marrow adipocytes. Indeed, in the absence of PP5 the MSC lineage allocation is skewed toward osteoblasts and away from lipid accumulating adipocytes, although an increase in beige adipocyte gene expression is observed. In the presence of rosiglitazone, PP5 translocates to the nucleus, binds to PPARγ and RUNX2, and dephosphorylates both factors, resulting in activation of PPARγ adipocytic and suppression of RUNX2 osteoblastic activities. Moreover, shRNA knockdown of PP5 results in cells refractory to rosiglitazone treatment. Lastly, mice deficient in PP5 are resistant to the negative effects of rosiglitazone on bone, which in wild type animals causes a 50% decrease in trabecular bone mass. In conclusion, PP5 is a unique phosphatase reciprocally regulating PPARγ and RUNX2 activities in marrow MSC.

MSC 2 allocation toward adipocyte or osteoblast lineages relies on the balance between the activity of two transcription factors, PPAR␥ and RUNX2. PPAR␥ also regulates insulin sensitivity. The insulin sensitizing and pro-adipocytic activities of PPAR␥ are determined at the level of posttranscriptional modifications (PTMs), which include phosphorylation, acetylation, and sumoylation (1). Similarly, osteoblastic activity of RUNX2 is regulated by PTMs, among which serine phosphorylation is the most recognized (2). Dephosphorylation of PPAR␥ at Ser-112 determines its pro-adipocytic activity, whereas phosphorylation of RUNX2 at Ser-301 and Ser-319 promotes its proosteoblastic activity (1)(2)(3). We recently reported that Ser-112 in PPAR␥ and Ser-301/Ser-319 in RUNX2 are phosphorylated by the same MAPK kinases, leading to activation of RUNX2 and inhibition of PPAR␥ activities (4). The reciprocity in PPAR␥ and RUNX2 activities determined by their phosphorylation status suggests that a common mechanism regulates their activities at the level of serine dephosphorylation.
PPAR␥ is a target of thiazolidinedione (TZD) class of antidiabetic drugs with insulin sensitizing and pro-adipocytic activities. TZDs, including rosiglitazone and pioglitazone, unbalance bone remodeling, which results in bone loss and fractures (5)(6)(7)(8). The pathological effects of TZDs on bone are explained by decreased bone formation due to reduction in osteoblast number accompanied by an increase in marrow adipocyte number and increased bone resorption (6,7). PPAR␥ preferentially stimulates differentiation of MSC to the adipocyte lineage by direct transcriptional regulation of adipocytic genes via binding to PPAR response elements (PPREs) at gene promoters. However, the mechanism by which osteoblast differentiation and gene expression are suppressed upon TZD treatment is less defined. An analysis of marrow MSC transcriptome in response to rosiglitazone showed robust changes in gene expression as soon as 2 h after treatment (9). These changes included up-regulation of adipocytic genes possessing multiple PPREs at promoters. However, down-regulated osteoblastic genes lacked PPREs in their promoters, suggesting that a PPAR␥-independent mechanism may be involved in this pro-This article has been withdrawn by the authors following concerns raised by the Journal. Specifically, the image for total ERK in cess (9). We recently reported that prolonged treatment with rosiglitazone decreases MAPK activity and is associated with a decrease in RUNX2 phosphorylation at Ser-319 and transcriptional activity (4). However, the delayed TZD effect to decrease MAPK activity does not explain the rapid suppression by rosiglitazone of osteoblast-specific genes otherwise positively regulated by RUNX2 (9). We recently identified protein phosphatase 5 (PP5) as responsible for dephosphorylation of Ser-112 and activation of PPAR␥ pro-adipocytic activity (10). PP5 is a unique member of the family of serine/threonine phosphoprotein phosphatases in that it contains a tetratricopeptide repeat (TPR) domain. Activation of PP5 requires binding of its TPR domain to the protein complex undergoing dephosphorylation (11). We reasoned that PP5 reciprocally regulates PPAR␥ and RUNX2 activities by binding to and dephosphorylating Ser-112 pPPAR␥ and Ser-301,Ser-319 pRUNX2 and that this regulates MSCs fate toward adipocytic and away from osteoblastic lineage. Here we show that PP5 regulates PPAR␥ and RUNX2 activities through selective dephosphorylation, is a major contributor to TZD-induced bone loss, and promotes the MSC shift toward adipogenesis and away from osteoblastogenesis.

PP5 Reciprocally Regulates PPAR␥ and RUNX2 Activities-
The ability of PP5 protein phosphatase to dephosphorylate  pPPAR␥ and Ser-319 pRUNX2 was examined in Cos7 cells ectopically expressing either PPAR␥ (intact or mutated at Ser-112) or RUNX2 (intact or mutated at Ser-301 and Ser-319) proteins. To ensure phosphorylation of Ser-112 and Ser-319 and assess PP5-specific activity toward these PTMs, Cos7 cells were co-transfected with either a construct expressing constitutively active MEK1 (Meksp or SP) and/or a construct expressing PP5. As shown in Fig. 1A pRUNX2 increased in the presence of MEK1 but decreased in the presence of PP5, and MEK1-induced phosphorylation of Ser-319 was reduced in the presence of PP5 (Fig.  1B). As expected, substitution of serine 319 by alanine in RUNX2 (SA mutant) led to loss of reactivity with anti-Ser-319 pRUNX2 antibody. MEK1 is a threonine/tyrosine kinase that phosphorylates and activates MAPK via the extracellular signal-regulated kinases (ERK) pathway. Thus, the levels pERK serve as a control of MEK1 activity. As shown in Fig. 1, A and B, transfection of MEK1 increased pERK levels regardless of the presence of PP5 or functional phosphorylation sites in PPAR␥ or RUNX2. This indicates that MEK1 activity is not affected by PP5 and that Ser-112 and Ser-319 are direct targets for PP5 phosphatase activity.
Next, we tested how Ser-112 and Ser-319 PTMs correlate with transcriptional activities of both factors. Increased phosphorylation at Ser-112 by MEK1 resulted in decreased PPREmediated transcriptional activity of PPAR␥ (Fig. 1C). Overexpression of PP5 activated PPAR␥; however, in the presence of MEK1 this activity returns to control level. The basal activity of the construct carrying the phosphorylation site mutation, S112A (SA), elevated PPAR␥ transcriptional activity, whereas the activity of a construct carrying S112E mutation (SE), which mimics phosphorylation of this site, significantly decreased PPAR␥ basal transcriptional activity. Neither SA nor SE constructs responded to PP5 with increased transcriptional activity (Fig. 1C). In conclusion, phosphorylation of Ser-112 inhibits PPAR␥ PPRE-specific transcriptional activity, a response that is reversed by PP5.
The effect of PP5 on transcriptional activity of RUNX2 is opposite to the effect on PPAR␥ (Fig. 1D). MEK-dependent phosphorylation at Ser-319 was associated with increased RUNX2 transcriptional activity, which was reduced to 50% in the presence of PP5. Serine to alanine mutations of the Ser-301 and Ser-319 sites also reduced MAPK-dependent transcriptional activity by 50%, and this activity was not reduced further by PP5. Because PP5 reduced MAPK-dependent Ser-319 phosphorylation to basal levels, we conclude that other non-PP5sensitive sites also participate in MAPK regulation of RUNX2. As expected, the basal transcriptional activity of combined phosphomimetic mutant S301E/S319E (SE) was high and was not affected by any of the tested conditions.
The reciprocal regulation of PPAR␥ and RUNX2 activities was also observed in conditions of PP5 partial deficiency. Lentiviral expression of the shRNA to PP5 in bone marrow-derived bipotential U-33/␥2 cells, which reduced PP5 protein levels by 40%, resulted in a 40 -50% increase in the levels of Ser-112 pPPAR␥ and Ser-319 pRUNX2 proteins ( Fig. 2A). These correlated with decreased expression of PPAR␥-regulated gene markers and with either increased expression of RUNX2-induced markers or with increased suppression of Axin2, which is negatively regulated by RUNX2 (Fig. 2B) (12).

PP5 Regulates PPAR␥ and RUNX2 Activities
axial and appendicular skeletons. Animals at the age of 5 months were used in all experiments because their skeleton is mature, and adipocyte accumulation in the marrow cavity, representing PPAR␥ transcriptional activity regulated by PP5, is detectable. The trabecular bone fractions in male proximal tibiae ( Fig. 3A) and L4 vertebrae (Fig. 3C) are 70 and 40% higher, respectively, than wild type (WT) controls. Trabecular bone in tibiae of PP5KO mice have more trabeculae and a high degree of structural complexity, measured as connectivity density, whereas all trabecular parameters were increased in vertebrae, yielding an overall favorable microarchitecture (Table 1). Fat volume in the marrow of proximal tibiae is decreased substantially in PP5KO mice (Fig. 3B). PP5 deficiency increased cortical bone diameter and bone area but had no effect on cortical thickness (Table 2). It does, however, increase tissue mineralized density (TMD) in tibia midshaft, indicating a higher degree of mineralization. The increased mineralization of the mid-shaft is associated with increased strength, as indicated by increased polar moment of inertia, and resistance to bending across the maximal (Imax/Cmax) and minimal centroidedge (Imin/Cmin) ( Table 2). The increased bone mass in PP5KO mice correlates with increased osteoblast activity measured as a daily mineral apposition rate and bone formation rate (BFR) in tibia bone doublelabeled with calcein (Fig. 3, D and E). This correlated with increased osteoblast numbers (Fig. 3E). Interestingly, osteoclast number was also elevated; however, the ratio of osteoblast to osteoclast was higher in PP5KO than in WT mice, suggesting higher bone formation over bone resorption (Fig. 3E).
The gene expression profile of osteoblasts freshly isolated from the endosteal surface of femora confirmed their increased activity in the absence of PP5. The expression of all three osteoblast-specific transcription factors (RUNX2, DLX5, and Osterix) and Wnt10b ligand were increased (Fig. 3F). Conse- To analyze an effect of Ser-112 phosphorylation on PPAR␥ activity, Cos7 cells were transfected with expression vectors for non-mutated (WT) or S112Amutated (SA) PPAR␥ and either empty vector LacZ (C) or constitutively active MEK1 (SP) or PP5 or both (SPϩPP5). Similarly, to analyze an effect of Ser-301/Ser-319 phosphorylation on RUNX2 activity, Cos7 cells were transfected with expression vectors for WT or S301A/S319A-mutated (SA) RUNX2, and either empty vector LacZ (C) or constitutively active MEK1 (SP) or PP5 or both (SPϩPP5). Whole cell extracts were harvested 48 h post transfection and analyzed with Western blots. A, PP5 effect on PPAR␥ phosphorylation. Western blots were analyzed using antibodies specific to either phosphorylated Ser-112 of PPAR␥ ( Ser-112 pPPAR␥) or against total PPAR␥ (T-PPAR␥), or against phosphorylated Thr-202/Tyr-204 of ERK (P-ERK), or against total ERK (T-ERK). Graphs represent densitometry analysis of Western blots (n ϭ 3-6). Signals from bands representing phosphorylated proteins were normalized to signals from total proteins. B, PP5 effect on RUNX2 phosphorylation. Blots were analyzed using either antibodies specific to phosphorylated Ser-319 of RUNX2 ( Ser-319 pRUNX2) or antibodies specific to total RUNX2 (T-RUNX2) or against phosphorylated Thr-202/Tyr-204 of ERK (P-ERK),or against total ERK (T-ERK). Graphs represent densitometry analysis as in A (n ϭ 3-6). C, analysis of transcriptional activity of PPAR␥. The same samples of Cos7 cells as in A expressing either C, SP, PP5, or SPϩPP5 were co-transfected with PPRE-luc reporter plasmid and either expression vectors for WT or PPAR␥ phosphorylation mutants S112A (SA) and S112E (SE). PPAR␥ activity was measured after 48 h. Firefly luciferase activities were normalized to Renilla pyriformis luciferase activities. D, analysis of transcriptional activity of RUNX2. Samples of Cos7 cells as in B expressing either C, SP, PP5, or SPϩPP5 were co-transfected with 6OSE2-luc reporter plasmid and either expression vectors for WT or RUNX2 phosphorylation mutants S301A/S319A (SA), and S301E/S319E (SE) and analyzed as in C. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. quently, the expression of genes regulated by these transcription factors and controlling osteoblast maturation and bone mineralization, such as matrix extracellular phosphoglycoprotein (Mepe), phosphate regulating endopeptidase homolog (Phex), osteopontin (Spp1), and bone sialoprotein (Ibsp) were consistently up-regulated (Fig. 3F).
Osteoblast differentiation and bone formation are regulated by osteocytes, which produce sclerostin and DKK1, both proteins acting as inhibitors of Wnt signaling (13). The pattern of gene expression for these proteins in osteocytes freshly isolated from femur cortical bone of PP5KO mice was consistent with increased support for bone formation. Although the expression of Sost gene encoding sclerostin was not changed, the expression of Dkk1 was decreased by 4-fold (Fig. 3G). Similar to the osteoblast fraction, Wnt10b expression in osteocytes increased substantially, consistent with an elevation in Wnt signaling activity. Indeed, expression of Axin2, cyclin D (CycD), and connexin 43 (Cx43) is up-regulated consistent with Wnt pathway activation. Surprisingly, the expression of osteoprotegerin (Opg), which is known to be positively regulated by Wnt signaling, is decreased in osteocytes derived from PP5KO mice (Fig.  3G). Taken together, these results suggest increased support for bone formation by osteocytes deficient in PP5. Of note, expression of the marker of osteocyte maturation, Dmp1, did not differ between PP5KO and WT mice, indicating that osteocyte function, but not differentiation, is altered in PP5KO mice (Fig. 3G).
PP5 Deficiency Skews MSC Fate-To determine whether PP5 regulates lineage commitment of MSCs, the number of mesenchymal cells with a potential to form colonies differentiating toward osteoblast or adipocyte phenotypes were analyzed in ex vivo MSC cultures using a colony-forming unit assay (cfu). As shown in Fig. 4A, the number of MSCs with the ability to form fibroblast-like colonies was not different between PP5KO and WT animals. However, PP5 deficiency resulted in a lower potential to form adipocytic (CFU-AD) and a higher potential to form osteoblastic (CFU-OB) colonies (Fig. 4A). The role of PP5 in determining MSC differentiation potential was reflected by the pattern of osteoblast and adipocyte gene markers expressed. Thus, in basal conditions MSCs deficient in PP5 are characterized by higher expression of osteoblast-specific gene markers (Fig. 4B) and growth factors (Fig. 4C) and lower expression of white adipocyte-specific gene markers (CD36 and Lpl) (Fig. 4D) as compared with WT marrow MSCs. Interestingly, PP5 deficiency increased expression of brown adipocyte markers (Ucp1 and Prdm16) in these cells (Fig. 4D).
We previously reported that brown-like or beige adipocytes in the bone marrow support osteoblast differentiation and bone formation by producing bone anabolic factors (14). To determine the role of PP5 in "beiging" of adipocytes, we knocked down PP5 in AD2 pre-adipocytic cells using lentivirus carrying specific shRNA. Similarly as in U-33/␥2 cells ( Fig. 2A), silencing of PP5 in AD2 cells resulted in increased phosphorylation of both RUNX2 and PPAR␥ (Fig. 4E). Moreover, a decrease in PP5 protein levels (Fig. 4E) resulted in increased expression of beige adipocyte markers (Prdm16, Tbx15, HoxC2, and Dio2), decreased expression of white adipocyte marker (Tcf21), and increased the expression of bone anabolic Igfbp2 and Wnt10b (Fig. 4F). To confirm that adipocytic cells deficient in PP5 produce activities that support osteoblast differentiation, we cocultured bipotential U33/␥2 cells with conditioned media from AD2 PP5KD cell cultures and measured alkaline phosphatase activity, a bona fide marker of osteoblastic activity. As shown in Fig. 4G, conditioned media collected from AD2 PP5KD donor cells increased alkaline phosphatase activity in U33/␥2 recipient cells. Thus, the regulatory effect of PP5 on PPAR␥ activity may also include regulation of white and beige adipocyte phenotypes and suggests a contribution of PP5-deficient beige adipocytes to the high bone mass of PP5KO mice.
PP5 Phosphatase Binds and Regulates PPAR␥ and Runx2 Activities in Response to the PPAR␥ Agonist Rosiglitazone-Because rosiglitazone induces adipocyte and suppresses osteoblast phenotypes in marrow MSCs, we examined the role of PP5 in this process. As shown in Fig. 5, A and B, rosiglitazone treatment causes simultaneous PP5 recruitment to the PPAR␥ and RUNX2 complexes. Interestingly, regardless of their activation FIGURE 2. PP5 knockdown increased PPAR␥ and RUNX2 phosphorylation leading to increased PPAR␥ and reduced RUNX2-specific transcriptional activity. A. to achieve stable knockdown of PP5 protein expression, U-33/␥2 cells were infected with lentiviral constructs encoding either scrambled (Sc) or PP5-specific (KD) shRNA. Whole cell extracts were analyzed by Western blot with antibodies specific to Ser(P)-112 of PPAR␥, or Ser(P)-319 of RUNX2 and antibodies against total PPAR␥ and RUNX2. HSP90 protein levels served as a loading control. Graphs represent densitometry analysis of Western blots (n ϭ 6). Signals from bands representing phosphorylated proteins were normalized to signals from total proteins. B, the effect of PP5 silencing in U-33/␥2 cells on expression of gene markers controlled by either PPAR␥ or RUNX2 transcriptional activities (n ϭ 6). Graphs represent -fold change in expression of analyzed markers in KD as compared with scrambled conditions. All transcripts expression was normalized to 18S RNA. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001.

TABLE 2 Cortical parameters of midshaft tibia analyzed with mCT
T.Ar, total area; B.Ar, bone area; M.Ar, marrow area; Ct.Th., cortical thickness; pMOI, polar moment of inertia; Imax/Cmax, resistance to bending measured across the bone by the distance along the maximal centroid-edge; Imin/Cmin, resistance to bending measured across the bone by the distance along the minimal centroid-edge; TMD, tissue mineralized density. n ϭ 5 mice per group.

PP5 Regulates PPAR␥ and RUNX2 Activities
status, both factors are already in complexes with HSP90 chaperone, which facilitates PP5 binding and activation via its TPR domain (11). Complex formation with either PPAR␥ or RUNX2 is associated with a partial redistribution of PP5 protein from the perinuclear space to the nucleus where it co-localizes with both transcription factors (Fig. 5, C and D) and correlates with their dephosphorylation (Fig. 5E). Silencing of PP5 using lentiviral expression of PP5-specific shRNA blocks rosiglitazoneinduced dephosphorylation of PPAR␥ Ser-112 and RUNX2 Ser-319 (Fig. 5E). Consequently, PP5KD makes PPAR␥ resis-tant to rosiglitazone activation and RUNX2 resistant to rosiglitazone inhibition at their respective promoters (Fig. 5F) and at endogenous genes (Fig. 5G). PP5 Protects from Rosiglitazone-induced Bone Loss-The above results suggested that PP5KO mice may be resistant to the negative effects of rosiglitazone on bone. To test this, WT and PP5KO mice received rosiglitazone at the dose of 11 mg/day/kg of body weight for 8 weeks. Bone mineral density (BMD) measured at the beginning and the end of the experiment showed that PP5KO mice are refractory to the decrease in  NOVEMBER 18, 2016 • VOLUME 291 • NUMBER 47 BMD caused by rosiglitazone observed in WT mice (Fig. 6A). Consistent with our previous results, rosiglitazone increased body weight in WT animals, but the weight of PP5KO mice remained the same as control animals (Fig. 6B). Analysis of trabecular bone in proximal tibia confirmed bone loss in WT mice receiving rosiglitazone. Bone mass was decreased by 50% due to a decrease in trabeculae number (Fig. 6C). None of these changes were observed in the bone of PP5KO mice. Both tra-becular bone mass and structure in tibia (Fig. 6C) and vertebra (Fig. 6D) were preserved in PP5KO animals receiving rosiglitazone. Moreover, rosiglitazone had a very modest effect (1.8-fold increase) on fat accumulation in the marrow cavity of PP5KO, whereas such accumulation was very robust (40-fold increase) in the bone of WT animals (Fig. 6C). In contrast to WT, rosiglitazone did not affect the number of osteoblasts and osteoclasts and did not affect BFR in PP5KO mice (Fig. 6E). Consistent with E, lentiviral knockdown (KD) of PP5 in preadipocytic AD2 cells. Whole cell extracts were analyzed by Western blot with antibodies specific to Ser(P)-112 of PPAR␥ or Ser(P)-319 of RUNX2 and antibodies against total PPAR␥ and RUNX2. HSP90 protein levels served as a loading control. Graphs represent densitometry analysis of Western blots (n ϭ 6). Signals from bands representing phosphorylated proteins were normalized to signals from total proteins. Sc, scrambled. F, the effect of PP5 KD in AD2 cells on expression of gene markers for beige adipocytes and bone anabolic factors. G, the effect of conditioned media collected from AD2 cells (donor cells) infected with lentivirus carrying either scramble (Sc) or PP5-specific (KD) shRNA on alkaline phosphatase (ALP) activity in recipient U33/␥2 cells. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001 versus WT mice or Sc transfected cells.

PP5 Regulates PPAR␥ and RUNX2 Activities
these measurements, osteoblasts isolated from the endosteal surface of femora of rosiglitazone-treated PP5KO mice maintained high expression of all three osteoblast-specific transcription factors (Runx2, Dlx5, and Osx), whereas the expression of Dlx5 and Osx was decreased in rosiglitazone-treated WT mice (Fig. 5F). Interestingly, osteocytes isolated from rosiglitazone-treated WT mice had increased expression of Sost, Dkk1, and Rankl, suggesting an osteocyte contribution to the negative effect on bone formation and the increase in bone resorption observed in WT animals (Fig. 6G). This osteocyte response to rosiglitazone was not observed in PP5KO mice (Fig. 6G). These results indicate that PP5 mediates rosiglitazone-induced bone loss.

Discussion
Phosphorylation is the most common PTM regulating protein activities. We previously identified ERK/MAP kinases as responsible for reciprocal regulation of PPAR␥ and RUNX2 activities at the level of serine phosphorylation (2)(3)(4). Using this mechanism, MAPK may control the ratio of osteogenesis to marrow adipogenesis in a variety of physiological conditions, including skeletal loading, where Ser-319 pRUNX2 is increased (15,16). However, little is known about mechanisms opposing bone anabolic signals and stimulating marrow adipogenesis, such as the response to the TZD, rosiglitazone. Here we show for the first time that PP5 phosphatase is a unique regulator of both PPAR␥ and RUNX2 activities associated with serine dephosphorylation. PP5-mediated dephosphorylation of PPAR␥ at Ser-112 and RUNX2 at Ser-319 stimulates adipogenesis and suppresses osteoblastogenesis from marrow MSC. Thus, PP5 phosphatase provides a missing link for understanding mechanisms controlling MSC differentiation to adipocytes at the expense of osteoblasts. The role of PP5 in regulation of MSC lineage commitment is reflected in the high bone mass of mice deficient in this phosphatase and their protection from the negative effect of rosiglitazone. Thus, this single phosphatase is a major contributor to the negative effect of rosiglitazone on bone mass and osteoblast differentiation. Mice deficient in PP5 have a high number of active endosteal osteoblasts, which are resistant to rosiglitazone. Similarly, primary MSCs derived from PP5KO mice are neutral to the anti-osteoblastic and pro-adipocytic effects of rosiglitazone. This provides definitive evidence that PP5 is a unique phosphatase reciprocally regulating RUNX2 and PPAR␥ activities.
PP5 belongs to the family of tetratricopeptide repeat proteins, which can bind HSP90 chaperone in order to modulate nuclear receptor activity (11). We previously demonstrated that PP5 phosphatase mediates cellular lipid metabolism through reciprocal control of glucocorticoid receptor-␣ (GR␣) and PPAR␥ (10). Here, we demonstrate that a similar reciprocal mechanism between PPAR␥ and RUNX2 regulates MSC differentiation to osteoblasts and adipocytes. Moreover, we showed that, like PPAR␥, RUNX2 forms a complex with HSP90 and, upon activation of adipogenesis, PP5 binds to this complex and dephosphorylates RUNX2 at Ser-319. This finding demonstrates that PP5 can modulate the activity of transcription factors other than nuclear receptors and that the mechanism by which PP5 regulates client activity is common and involves formation of a complex with HSP90.
Although PP5 overexpression was able to reduce MAPKdependent phosphorylation of both PPAR␥ (Ser-112) and RUNX2 (Ser-319) close to unstimulated levels, there were notable differences in its effects on transcriptional activity. Thus, PP5 totally reversed MAPK-dependent suppression of PPAR␥ transcriptional activity but only partially inhibited MAPK stimulation of RUNX2. The implication of this finding is that MAPK phosphorylates and activates RUNX2 using sites in addition to those that are dephosphorylated by PP5.
As previously described (17,18) and as confirmed here in marrow MSCs, PP5 of unstimulated cells is found predominantly in the cytoplasm with increased concentration in the perinuclear region. Surprisingly, rosiglitazone treatment of cells caused at least a partial redistribution of PP5 to the nucleus where it co-localized with PPAR␥ and RUNX2. As determined by co-immunoprecipitation, rosiglitazone also caused recruitment of PP5 to the Hsp90-containing PPAR␥ and RUNX2 complexes. As described below, rosiglitazone-induced recruitment of PP5 to PPAR␥ is not without precedent. However, the binding of PP5 to RUNX2 in response to rosiglitazone is surprising and suggests similar protein-protein interaction as with PPAR␥.
Although it remains to be clarified what precise mechanisms cause translocation of PP5 to the nucleus and induce its binding to PPAR␥ and RUNX2, the phenomenon is highly similar to the hormone-induced recruitment of another TPR-containing chaperone, FKBP52, to nuclear receptors. In the latter case, binding of cognate hormone leads to interaction of FKBP52 with both the glucocorticoid and mineralocorticoid receptors and movement of the receptors from cytoplasm to nucleus with involvement of dynein-dynactin motor complexes (19 -21).
With respect to the physiological role of PP5, it is important to note that rosiglitazone and other TZD agonists are not endogenous activators of PPAR␥. That role is performed by various polyunsaturated fatty acids, which also activate PP5 (22)(23)(24). We, therefore, propose that the major function of PP5 in the regulation of osteogenesis is to act as a fatty acid controlled rheostat maintaining the balance between PPAR␥ and RUNX2 activities.
PP5 activity is not restricted to regulation of MSC lineage commitment. We found that mice deficient in PP5 also have higher osteoclast numbers, which remained unchanged upon rosiglitazone treatment. Although this finding is not a focus of this report, in separate studies we recently demonstrated that PPAR␥ regulation of osteoclast differentiation involves phosphorylation of Ser-273, the same serine that is also responsible for insulin sensitization (25). Thus, the role of PP5 phosphatase in the maintenance of bone homeostasis may go beyond regulation of adipocyte and osteoblast differentiation and may include hematopoietic cell differentiation and regulation of energy metabolism. These aspects of PP5 activity are currently under investigation.
PP5 deficiency also affects osteocyte support of bone remodeling by changing the gene expression profile to favor bone formation and activation of Wnt signaling. It also changes the osteocyte response to rosiglitazone. Thus, PP5 deficiency decreases expression of Dkk1, a Wnt pathway inhibitor, which is predominantly produced in osteocytes and, as we showed recently, is under transcriptional control of PPAR␥ (25). Consistently, Wnt pathway activity is increased, which is reflected in increased expression of Axin2, CycD, and Cx43 in osteocytes, and increased expression of Wnt10b in both osteocytes and osteoblasts deficient in PP5. However, a decreased expression of Opg in osteocytes is unexpected and suggests that in addition to the Wnt, pathway Opg expression may be regulated by other mechanisms. Indeed, as we recently reported that Opg expression can be down-regulated by mechanism requiring dephosphorylation of Ser-273 in PPAR␥ protein (25). Although not shown here, the Ser-273 is dephosphorylated in PP5KD cells.
Importantly, although in WT animals rosiglitazone up-regulates expression of both Dkk1 and Sost, PP5-deficient osteocytes are refractory to rosiglitazone. Similarly, the expression of Rankl, which is increased by rosiglitazone, remains unchanged in the absence of PP5 phosphatase. It remains to be established whether expression of these genes in osteocytes is under PP5 control via regulation of PPAR␥ and RUNX2 activities.
Another interesting feature of PP5 is its apparent ability to modulate PPAR␥ activity controlling white versus beige adipocyte differentiation. MSCs deficient in PP5 have increased expression of genes associated with a beige adipocyte phenotype and increased expression of cytokines with pro-osteoblastic activity. We showed previously that marrow adipocytes having beige characteristics produce factors that are anabolic for bone, including WNT10b and IGFBP2 (14). Thus, increased support for bone formation from marrow cells with beige adipocyte characteristics may contribute to the increased bone mass of mice deficient in PP5.
In conclusion, PP5 is the first identified phosphatase reciprocally regulating PPAR␥ and RUNX2 activities. PP5 unifies activities of these transcription factors at the level of MSC lineage commitment and bone response to rosiglitazone.

Experimental Procedures
Animals PP5 Ϫ/Ϫ (PP5KO) mice were described previously (26). The colony of PP5KO and WT mice were maintained at the University of Toledo Health Science Campus. All animals were housed in a 12-h dark-light cycle and had free access to standard chow (Harlan Teklad 2016; Haslett, MI). The animal treatment and care protocols conformed to National Institutes of Health Guidelines and were performed using a University of Toledo Health Science Campus Institutional Animal Care and Utilization Committee protocol.
All presented experiments were performed on 5-month-old skeletally mature males, the age when marrow adipocyte accumulation is detectable. There are no differences in skeletal characteristics between males and females of PP5KO mice (not shown). To test the skeletal effect of rosiglitazone administration, WT and PP5KO mice were divided into two groups (n ϭ 3-6 per group) and fed for 8 weeks with either non-supplemented chow or chow supplemented with rosiglitazone at the dose 20 mg/kg/day. During the experiment, food and water intake per cage were monitored, and the average intake of rosiglitazone per mouse was calculated at the end of experiment. There were no differences between groups within each mouse strain in daily food intake and water intake. Calculated dose of effective drug intake in WT mice was 19.1 mg/kg/day and in PP5KO mice was 20.2 mg/kg/day.

Bone Analysis
Microcomputed tomography (mCT) of the tibiae and L4 vertebrae was performed using the CT-35 system (Scanco Medical AG, Bassersdorf, Switzerland), as previously described (27). Briefly, scans were performed at 70 peak kilovoltage (kVp) energy and 113-A intensity settings and using 7-m voxel. Images of trabecular bone were segmented at a 289 threshold value using the per mille scale. The analysis of bone microstructure conformed to recommended guidelines (28).
For lipid evaluation, decalcified bone specimens were stained for 1 h in solution containing 2% osmium tetroxide prepared in 0.1 M sodium cacodylate buffer, pH 7.4, according to the protocol (27). Staining was carried out in an exhaust hood and away from light due to osmium tetroxide toxicity and light sensitivity. Images of lipid depositions were acquired at 70-kVP and 113-A settings and 12-m nominal resolution. Image segmentation was done under global threshold conditions by applying a gray scale threshold of 480 -1000 using the per mille scale with the three-dimensional noise filter set to sigma 1.2 and support 2.0. Lipid volumes were calculated directly from individual voxel volumes in three-dimensional reconstructions. Global BMD was measured by dual-energy X-ray absorptiometry (DXA) using a PIXI-mus Small Animal Densitometer (manufactured by LUNAR, Madison, WI).

Bone Histomorphometry
To permit static and dynamic bone histomorphometry, mice were intraperitoneally injected with 20 mg/kg calcein (Sigma) 8 and 2 days before sacrifice, and undecalcified tibiae were fixed in 70% ethanol and embedded in methyl methacrylate and sectioned. Pictures were taken by a Zeiss Axiovert 40 CFL Fluorescence Microscope (Carl Zeiss Microscopy, Thornwood, NY) equipped with a Micropublisher 3.3 Megapixel Cooled CCD Color Digital Camera (QImaging, Surrey, BC, Canada). Analysis of BFR was confined to the secondary spongiosa of proximal tibia and was performed using the Nikon NIS-Elements BR3.1 system. The measurements were collected with a 40ϫ objective (numerical aperture, 0.5) at room temperature from 6 representative fields per bone sample. The terminology and units used were those recommended by the Histomorphometry Nomenclature Committee of the American Society for Bone and Mineral Research (29).

Immunofluorescent Imaging
U-33/␥2 cells were seeded on laminin-coated coverslips in 6-well plates at 100,000 -200,000 cells/well in DMEM contain-ing dialyzed serum (catalog #SH30079.3, HyClone, Logan, UT). Before fixing, cells were treated with or without rosiglitazone for 1 h. Cells were fixed in 3% paraformaldehyde for 15 min and blocked at room temperature for 1 h in PBST (phosphate-buffered saline ϩ 0.25% Triton X) containing 1% BSA. Incubation with primary antibody was done overnight at 4°C. After 3 washes in PBST, membranes were incubated with Alexafluor anti-rabbit (Alexafluor 488; green) and anti-mouse (Alexafluor 594; red) secondary antibodies (Abcam PLC) at 1:4000 dilution in PBST for 1 h at room temperature. Coverslips were then mounted on labeled glass slide using Fluoromount-G and DAPI (Sigma) for nuclear staining. Fluorescent images of the fixed cells were obtained using a Leica TCS SP5 laser scanning confocal microscope (Leica Microsystems, Bannockburn, IL) equipped with conventional solid state and a Ti-sapphire tunable multiphoton laser (Coherent, Santa Clara, CA). Images were acquired in the XYZ plane in 1-m steps with a 63ϫ oil objective (numerical aperture, 1.40) at room temperature. Images were acquired with the LAS AF software in sequential scan mode. Images are two-dimensional projections of the of image stack as labeled. The figures show representative cells from each condition. At least 50 -100 cells from each condition were inspected.

Cell Culture Experiments
Murine marrow-derived cell lines representing bipotential (U-33/␥2 cells) and pre-adipocytic (AD2 cells) phenotypes have been previously described (30). The knockdown of PP5 was achieved by viral infection of U-33/␥2 cells with a specific short hairpin RNA (shRNA) for PP5 (CACGAGACAGACAACAT-GAACCAGATCTA). A control cell line was made using a scramble shRNA that was not specific to any known RNA sequence. Three hours after the infection cells were washed with fresh medium and allowed to recover; 48 h after the infection, cells were washed, followed by selection with puromycin (5 ng/liter) for 7 days with medium changes every 24 h. To confirm knockdown of PP5, Western blot analysis was performed on the cell lines. Primary bone marrow cultures were established from femur marrow aspirates and differentiated as described (6).

Extraction of Osteoblast-and Osteocyte-enriched Fractions
Cell fractions enriched in osteoblasts or osteocytes were isolated by sequential collagenase digestion of femora bone according to previously described protocol (35).

Gene Expression Analysis Using Quantitative Real-time RT-PCR Analysis
One g of total RNA was converted to cDNA using the Verso cDNA synthesis kit (Thermo Fisher Scientific, Waltham, MA). PCR amplification of the cDNA was performed by quantitative real-time PCR using TrueAmp SYBR Green qPCR SuperMix (Smart Bioscience, Maumee, OH) and processed with StepOne Plus System (Applied Biosystems, Carlsbad, CA). The thermocycling protocol consisted of 10 min at 95°C, 40 cycles of 15 s at 95°C, 30 s at 60°C, and 20 s at 72°C and finished with a melting curve ranging from 60 to 95°C to allow distinction of specific products. Relative gene expression was measured by the comparative CT method using 18S RNA levels for normalization. Primers were designed using OligoPerfect Designer (Thermo Fisher Scientific).

Statistical Analysis
Data are presented as the means Ϯ S.D. and were analyzed using statistical software package of SigmaPlot (version 13.0). p values Ͻ0.05 was considered statistically significant.