Transforming Growth Factor β Suppresses Osteoblast Differentiation via the Vimentin Activating Transcription Factor 4 (ATF4) Axis*

Background: Transforming growth factor β (TGFβ) inhibits osteocalcin (Ocn) transcription and osteoblast differentiation. Results: The inhibition of Ocn expression and osteoblast differentiation by TGFβ is blunted upon lack of Atf4 or vimentin knockdown. Conclusion: ATF4 and vimentin are novel downstream targets of TGFβ in osteoblasts. Significance: Understanding mechanisms by which transcription factors are regulated is crucial for developing effective anabolic drugs for bone. ATF4 is an osteoblast-enriched transcription factor of the leucine zipper family. We recently identified that vimentin, a leucine zipper-containing intermediate filament protein, suppresses ATF4-dependent osteocalcin (Ocn) transcription and osteoblast differentiation. Here we show that TGFβ inhibits ATF4-dependent activation of Ocn by up-regulation of vimentin expression. Osteoblasts lacking Atf4 (Atf4−/−) were less sensitive than wild-type (WT) cells to the inhibition by TGFβ on alkaline phosphatase activity, Ocn transcription and mineralization. Importantly, the anabolic effect of a monoclonal antibody neutralizing active TGFβ ligands on bone in WT mice was blunted in Atf4−/− mice. These data establish that ATF4 is required for TGFβ-related suppression of Ocn transcription and osteoblast differentiation in vitro and in vivo. Interestingly, TGFβ did not directly regulate the expression of ATF4; instead, it enhanced the expression of vimentin, a negative regulator of ATF4, at the post-transcriptional level. Accordingly, knockdown of endogenous vimentin in 2T3 osteoblasts abolished the inhibition of Ocn transcription by TGFβ, confirming an indirect mechanism by which TGFβ acts through vimentin to suppress ATF4-dependent Ocn activation. Furthermore, inhibition of PI3K/Akt/mTOR signaling, but not canonical Smad signaling, downstream of TGFβ, blocked TGFβ-induced synthesis of vimentin, and inhibited ATF4-dependent Ocn transcription in osteoblasts. Thus, our study identifies that TGFβ stimulates vimentin production via PI3K-Akt-mTOR signaling, which leads to suppression of ATF4-dependent Ocn transcription and osteoblast differentiation.

Transforming growth factor ␤ (TGF␤) 2 regulates many biological processes including patterning during development, cell proliferation, differentiation, apoptosis, and other physiological and pathological conditions such as wound healing, fibrosis, and cancer growth and metastasis. In the mammalian skeleton, TGF␤ is one of the most abundant cytokines stored in bone matrix; and once activated from its latent form, it promotes osteoblast proliferation and mesenchymal stem cell recruitment to active bone remodeling sites (1)(2)(3)(4)(5), while inhibits osteoblast differentiation (6 -8). The canonical Smad-dependent signaling has been identified as a mediator of TGF␤ in skeletal cells as well as in many other cell types. In response to ligand binding, TGF␤ receptors phosphorylate Smad2 and/or Smad3, which in turn bind to Smad4 to induce translocation into the nucleus (9), where the Smad complex either binds directly to DNA or indirectly to other transcription factors to regulate gene transcription (10).
It has become clear now that multiple Smad-independent or noncanonical signaling pathways are activated in response to TGF␤ ligands in various cell types. These include: MAP kinase (MAPK) pathways in intestine, lung epithelial cells, and breast cancer cells; Rho-like GTPase signaling pathways during epithelial to mesenchymal transition (EMT) in epithelial cells and primary keratinocytes; and mammalian targets of rapamycin (mTOR) through phosphatidylinositol 3-kinase (PI3K) and Akt pathways during EMT in fibroblasts (11)(12)(13)(14)(15)(16)(17)(18). Although each of these signaling pathways has been shown to play an important role in skeletal biology, whether TGF␤ activates non-canonical signaling pathways in osteoblasts is unknown.
Activating transcription factor (ATF4) is a leucine zippercontaining transcription factor belonging to the CREB family and was originally identified as an osteoblast-specific transcription factor required for osteocalcin (Ocn) transcription and osteoblast differentiation (19). Ocn mRNA is exclusively expressed in differentiated osteoblasts hence it is often used as a marker gene of mature osteoblasts (20). Together with Runx2, the first reported osteoblast-specific transcription factor, ATF4 activates Ocn transcription in vitro and in vivo through direct binding to its cognate osteoblast-specific element 1 (OSE1) of the Ocn promoter (19,21,22). In osteoblasts, TGF␤ targets Runx2 via a canonical Smad signaling pathway to achieve its inhibition of both Runx2 and Ocn transcription, thereby suppressing osteoblast differentiation (6). However, it is possible that TGF␤ targets other effectors at the transcriptional level to inhibit Ocn transcription in osteoblasts.
We have recently identified that in osteoblasts vimentin binds directly to ATF4 through its first leucine-zipper domain, which prevents ATF4 from binding to its cognate DNA OSE1 on the Ocn promoter, leading to inhibition of ATF4-dependent Ocn transcription and osteoblast differentiation (23). Vimentin is a member of the intermediate filament protein family and the most widely accepted molecular marker of mesenchymal cells. Moreover, its mRNA is often up-regulated in response to TGF␤ during EMT and cancer progression (24,25). Consistent with its inhibitory role during osteoblast differentiation, vimentin expression is down-regulated when osteoblasts progress toward a fully differentiated stage (23). Since this suggested that one component of vimentin regulation involves the regulatory control of its expression, we searched for the extracellular ligands that govern its expression. In doing so, we noticed that TGF␤ stimulated vimentin mRNA in C2C12 myoblastic cells (26), a cell type of mesenchymal origin that can differentiate into chondrocytes and osteoblasts (27). Given that TGF␤ and vimentin both negatively regulate Ocn transcription and osteoblast differentiation, we hypothesized that TGF␤ targets vimentin and ATF4 to suppress Ocn transcription and osteoblast differentiation.
Here we present evidence that TGF␤ requires endogenous ATF4 to inhibit Ocn transactivation in primary osteoblasts and osteoblastic cell lines. With the delivery of a monoclonal anti-TGF␤ antibody to mice, we show that ATF4 is also required for TGF␤ to increase bone mass. Employing a series of molecular and biochemical approaches, we demonstrate that TGF␤ directly up-regulates vimentin production at post-transcriptional level, via PI3K-Akt-mTOR signaling, but not Smad signaling, to achieve its inhibition of Ocn transcription. Therefore, our study identifies two novel effectors, vimentin and ATF4, that act downstream of TGF␤ in the regulation of osteoblast differentiation via PI3K-Akt-mTOR signaling.

EXPERIMENTAL PROCEDURES
Materials-Tissue culture media and fetal bovine serum were purchased from Invitrogen. Anti-vimentin antibodies were from Santa Cruz Biotechnology and Biovision for V9 antirat vimentin and anti-mouse vimentin (#3634), respectively. Antibodies for ATF4 (C20) and Sp1 (PEP2) were from Santa Cruz Biotechnology, HA tag was from Abcam (ab9110); Flag tag was from Sigma (M2); and phospho-Smad2/3 was from Cell Signaling. Recombinant human (rh) TGF␤1 is from R&D sys-tems. All chemicals were from Sigma unless indicated otherwise.
Northern Blot Hybridization-Total RNA from indicated sources was isolated using TRIzol (Invitrogen) according to the manufacturer's protocols. Total RNA (5 g) was resolved in 1% agarose gel and transferred onto nylon membranes. After crosslinking with UV light, the membranes were hybridized in 6ϫ SSC buffer at 60°C overnight with the following probes: partial cDNA of mouse vimentin from 792 to 1218, mouse Atf4 covering 287 nucleotides of 5Ј-untranslated region and 180 nucleotides of coding region, full-length mouse Ocn gene 2, and mouse Gapdh cDNAs.
Transient DNA Transfection and Luciferase Assay-COS1 cells were seeded at a density of 2.5 ϫ 10 5 /well in 6-well culture dishes for 20 h and then transfected with 2 g/well of Flag-ATF4 or HA-vimentin expression plasmid. For reporter assays, cells were plated in 24-wells at a density of 2.5 ϫ 10 4 /well. p6ϫOSE1-Luc (0.25 g/well) together with 0.05 g/well of ␤-galactosidase (␤-gal) were transfected into COS1 cells using Lipofectamine (Invitrogen). Cells were lysed 24 h post DNA transfection, and the luciferase and ␤-gal activity was measured using cell lysate. Fold activation or inhibition of luciferase activity was calculated by normalization of luciferase activity with ␤-gal activity. For vimentin knockdown, 2T3 osteoblastic cells were plated in 6-well culture dishes at 2.5 ϫ 10 5 /well and then transiently cotransfected with psiRNA (1 g/well) empty vector or psiRNA-Vim with indicated reporter constructs (1 g/well).
Primary Calvarial Osteoblast Isolation and Osteoblast Differentiation Assay-Calvariae were collected from neonatal (P3) pups, pressed on Science Brand Kimwipes to remove blood and surrounding tissues, cleaned with 1ϫ PBS, and then subjected to series of collagenase digestions as described previously (28). Cells released from the first 2 digestions were discarded to enrich the numbers of osteoblastic cells. Cells released from the third digestion were plated in ␣MEM containing 10% FBS until confluence, which was defined as passage 0.
For differentiation assays, primary cells from passage 1 or osteoblastic cell lines were grown in 24-well plates in ␣MEM supplemented with 5 mM ␤-glycerophosphate and 100 g/ml ascorbic acid for either 2 or 12 days for alkaline phosphatase staining or von Kossa staining, respectively, as previously described (29).
2G7 Treatment-Ten pairs of 10-week-old WT and Atf4 Ϫ/Ϫ littermates were treated with anti-TGF␤ monoclonal antibody 2G7 or control IgG at 10 mg/kg. The intraperitoneal injections of each antibody were administered 3 times per week for 4 weeks (30).
Micro-computed Tomography (CT) Analysis-Mouse femurs were collected and fixed overnight in 4% PFA (pH 7.4) and then 70% ethanol. Trabecular bone of the metaphysis was evaluated using an ex vivo CT imaging system (Scanco CT40; Scanco Medical, Bassersdorf, Switzerland). Each femur was fit into a specimen tube and aligned with the scanning axis. Tomographic images were acquired at 55 kV and 145 mA with an isotropic voxel size of 12 m, and at an integration time of 250 ms with 500 projections collected per 180°C rotation. For 100 slices proximal to the growth plate, contours were fit to the inner layer of the cortical bone to define the total tissue volume (TV). Applying a Gaussian noise filter with a variance of 0.8 and support of 1 as well as a threshold of 300 mg of hydroxyapatite (HA) per cm 3 to distinguish mineralized from non-mineralized tissue for each femur scan, we quantified the volume of trabecular bone tissue (BV) per TV (BV/TV) as well as volumetric density (in mgHA/cm3) of the bone-mineralized tissue (BMD) (31).
Western Blot Analysis-Total cell lysates from indicated cell types were isolated with radioimmunoprecipitation assay (RIPA) buffer containing 50 mM Tris-Cl, pH 7.6, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Nonidet P-40, and protease inhibitors (Complete ULTRA Protease Inhibitor Mixture Tablets, Roche Applied Science). Nuclear extracts were isolated using high-salt nuclear extraction buffer on hypotonic buffer-swelled cells as described (32). Total cellular proteins or nuclear extracts (50 g per lane) were loaded onto 10% SDS-PAGE and Western blots were performed following standard protocols. Antibody concentrations were 0.5-1 g/ml.
Statistics-Data are presented as mean Ϯ S.D. Statistical analyses were performed using Student's t test. All in vitro experiments were repeated at least three times.

ATF4 Is Required for TGF␤ to Inhibit Osteoblast
Differentiation-To investigate whether ATF4 was downstream of TGF␤, we first established stable ROS17/2.8 cell lines carrying a reporter luciferase gene driven by 6 repeats of wild type (WT) ATF4 binding element OSE1, p6xOSE1-Luc. As controls, we also made two other stable cell lines carrying a luciferase gene driven by either 6 repeats of a mutant ATF4 binding element, p6xmOSE1-Luc to which ATF4 fails to bind (21), or 3 repeats of the activating protein 1 (AP1) binding element, p3xAP1-Luc. Treatment of these reporter cell lines with rhTGF␤1 at various concentrations, ranging from 0.2 to 2 ng/ml, inhibited dose-dependently luciferase activity driven by 6xOSE1, but not by 6xmOSE1 (Fig. 1, A and B), indicating that the inhibitory effect of TGF␤ on ATF4-dependent Ocn promoter activity relies on ATF4 binding to its cognate DNA. Furthermore, TGF␤ inhibition is specific to the Ocn promoter and not a general phenomenon of leucine zipper-containing transcription factors since TGF␤ failed to reduce luciferase activity in reporter cells driven by AP1 DNA binding element repeats (Fig. 1C).
To determine whether ATF4 was necessary for TGF␤ to suppress endogenous Ocn expression and osteoblast differentiation, we utilized the Atf4-deficient (Atf4 Ϫ/Ϫ ) mouse model (33). Freshly isolated primary calvarial osteoblasts (passage 1) from wild type (WT) and Atf4 Ϫ/Ϫ pups were treated with rhTGF␤1 (0.5 ng/ml) under osteogenic induction media for 10 days, and total RNA was collected for qRT-PCR analyses. Our data showed that TGF␤ inhibited endogenous Ocn expression by 86% in WT calvarial osteoblasts but only by 43% in Atf4 Ϫ/Ϫ calvarial osteoblasts (Fig. 1D). Consistent with this observation, TGF␤ reduced alkaline phosphatase (ALP) activity in WT control cells by 46% but had no such effect in Atf4 Ϫ/Ϫ calvarial cells (Fig. 1E). Lastly, low doses of TGF␤ (0.1 and 0.2 ng/ml) decreased the number of mineralized nodule formation by 90% in WT but only 20% in Atf4 Ϫ/Ϫ calvarial osteoblasts (Fig. 1F). Collectively, these data indicate that lack of endogenous ATF4 attenuates the potency of TGF␤ to inhibit endogenous Ocn expression, ALP activity and differentiation or mineralization of osteoblasts in vitro.
To address whether ATF4 was also required for TGF␤ in vivo and because treatment of WT mice with 1D11 an anti-TGF␤ monoclonal antibody blocking all three isoforms of TGF␤ (34), increased trabecular bone volume fraction in mice (30), we treated WT and Atf4 Ϫ/Ϫ mice with 2G7, an monoclonal antibody with similar reactivity to the 1D11 (35). WT control and Atf4 Ϫ/Ϫ mice were injected intraperitoneally with a TGF␤ monoclonal antibody 2G7 for 4 weeks. Consistent with the results reported previously using 1D11 (30), CT analysis revealed that 2G7 treatment increased bone volume fraction in WT mice by more than 30%, however, it could not improve the low bone mass in Atf4 Ϫ/Ϫ mice (Fig. 1, G and H). Thus, these results confirmed that ATF4 is an important molecule downstream of TGF␤ in vivo.
TGF␤ Stimulates Vimentin Post-transcriptionally in Osteoblasts-To determine whether TGF␤ suppressed ATF4-dependent activation of Ocn and osteoblast differentiation (Fig. 1) via a direct effect on ATF4, we then analyzed whether TGF␤ inhibited ATF4 expression. Primary calvarial osteoblasts (passage 1) were treated for 10 days with rhTGF␤1 at concentrations of 0.2, 1, or 2 ng/ml under osteogenic culture conditions, and then total RNA and protein were analyzed. As expected, endogenous Ocn expression decreased whereas ATF4 expression slightly increased, in response to TGF␤ in a dose-depen-dent manner (Fig. 2, A and B). These results suggested that the suppression of ATF4 transcriptional activity by TGF␤ is not due to a decrease in the level of ATF4 but may be via an indirect mechanism.
One of the candidates was vimentin, because we recently found that it acted as a suppressor of ATF4 in osteoblasts (23) and others have shown that its expression was stimulated by TGF␤ (26). To address whether vimentin was the mediator linking TGF␤ signaling and ATF4, we first treated ROS 17/2.8 cells with TGF␤ and found it could induce vimentin protein expression dose-dependently. Surprisingly, however, TGF␤ did not increase vimentin mRNA expression at all of the concentrations tested (Fig. 2, C and D). These data indicated that in osteoblasts TGF␤ regulates vimentin expression at the protein level and not at the mRNA level, which was different from what had been observed in C2C12 cells previously (26). To further verify our findings, we also treated a panel of osteoblasts along with C2C12 cells (as a positive control) with low dose rhTGF␤1 (0.2 ng/ml), which significantly suppressed WT primary osteo-blast differentiation in vitro (Fig. 1F). The results indicated that TGF␤ increased vimentin protein level but not its mRNA level in all tested cells, including mouse primary bone marrow stromal cells (BMSC), calvarial osteoblasts (POB), the 2T3 mouse preosteoblastic cells, and the MC3T3-E1 mouse committed osteoblasts. However and consistent with previous report (26), TGF␤ in the C2C12 cells increased both protein and mRNA vimentin levels (Fig. 2, E and F). Thus, these data demonstrated that TGF␤ up-regulates vimentin expression at the post-transcriptional level in osteoblasts.
Vimentin Is Required for TGF␤ to Inhibit Ocn Expression-To understand whether vimentin played an essential role in mediating the inhibitory effect of TGF␤ on ATF4 transcriptional activity and Ocn transcription, we knocked down endogenous vimentin expression in 2T3 mouse osteoblastic cells by transfection with siRNA-Vim (23) and the p6xOSE1-Luc as a readout for ATF4 transcriptional activity, and then tested their response to TGF␤ by reporter assays. 2T3 mouse cells instead of primary osteoblasts were used to ensure the transfection effi-FIGURE 1. ATF4 is required for TGF␤ to inhibit osteocalcin (Ocn) transcription. A-C, TGF␤ inhibits ATF4-dependent reporter activity. Luciferase activity is decreased by treatment of rhTGF␤1 at indicated concentrations in stable ROS17/2.8 reporter cells. p6xOSE1-Luc, luciferase gene driven by 6 repeats of ATF4-binding elements reporter (A); p6xmOSE1-Luc, 6 repeats of mutant ATF4-binding element driving luciferase reporter (B); p3xAP1-Luc, 3 repeats of unrelated leucine-zipper protein AP1-binding site-driven reporter (C). D, inhibition of endogenous Ocn expression by TGF␤ is attenuated in Atf4 Ϫ/Ϫ calvarial osteoblasts. Quantitative RT-PCR (qRT-PCR) analysis of total RNA isolated from WT and Atf4 Ϫ/Ϫ calvalrial osteoblasts treated with vehicle (Ϫ) or rhTGF␤1 (ϩ, 0.5 ng/ml) in osteogenic medium for 10 days. Note that TGF␤ decreased endogenous Ocn mRNA level in WT calvarial osteoblasts, which was attenuated (48% inhibition) in Atf4 Ϫ/Ϫ mutant osteoblasts. E, TGF␤ reduced alkaline phosphatase (ALP) activity in WT but not in Atf4 Ϫ/Ϫ osteoblasts. ALP assay of WT and Atf4 Ϫ/Ϫ calvarail osteoblasts treated with vehicle (Ϫ) or rhTGF␤ (ϩ, 0.5 ng/ml) in osteogenic medium for 2 days. F, TGF␤ inhibits mineralized nodule formation in WT but not in Atf4 Ϫ/Ϫ osteoblasts. von Kossa staining of calvarial osteoblasts treated with rhTGF␤1 (0.5 ng/ml) in osteogenic medium for 10 days. Note that TGF␤ reduced the number of mineralized nodules (black colonies) dramatically (Ͼ90%) in the WT cultures but only slightly in Atf4 Ϫ/Ϫ osteoblasts (Ͻ20%). G, ATF4 is required for the anabolic effect of anti-TGF␤ antibody in vivo. CT analysis of trabecular bones of WT and Atf4 Ϫ/Ϫ femurs treated with control IgG antibody or anti-TGF␤ monoclonal antibody (2G7, ␣TGF␤) neutralizing three forms of TGF␤ ligand for 4 weeks. H, quantification of data shown in G. Note that 2G7 treatment increased trabecular bone volume (BV) versus total tissue volume (BV/TV) in WT femurs by 30% but failed to rescue the low BV/TV in Atf4 Ϫ/Ϫ femur. n ϭ 6. ciency. We also transfected 2T3 cells with siRNA empty vector to serve as negative control cells. Fig. 3A (lanes 1 and 2), rhTGF␤1 (0.5 ng/ml) decreased luciferase activity by 36%, which was expected and consistent with what we observed previously in ROS17/2.8 reporter cells (Fig. 1A). Luciferase activity increased 2.7-fold in 2T3 cells transfected with siRNA-Vim, due to the knockdown of endogenous vimentin that removed its suppression of ATF4 transcriptional activity (23). When TGF␤ was added to these cells, ATF4 reporter activity remained 2.5fold higher in cells containing siRNA-Vim than in cells containing siRNA control vector (Fig. 3A, lanes 3 and 4). The knockdown of endogenous vimentin was confirmed by the absence of vimentin mRNA in siRNA-Vim transfected cells, which could not be rescued by rhTGF␤1. Consistent with our previous finding (23) and transfection results shown in Fig. 3A, a low level of endogenous Ocn expression in 2T3 cells transfected with siRNA control vector was inhibited by rhTGF␤1 (0.5 ng/ml) treatment; whereas this inhibition was blunted in cells transfected with siRNA-Vim (Fig. 3B). qRT-PCR data indicated that TGF␤1 inhibited endogenous Ocn expression by 60% in cells transfected with empty siRNA vector. As expected, vimentin knockdown in 2T3 preosteoblasts by siRNA-Vim increased endogenous Ocn expression by 4-fold, which was not affected by TGF␤1 (Fig. 3C). Therefore, these data validated that endogenous vimentin is required for TGF␤ to suppress ATF4-dependent Ocn transcription in vitro.
To determine whether vimentin protein level also fluctuated in vivo in response to TGF␤ administration, we analyzed vimentin content in the long bones of 2G7 or control antibodytreated mice. Western blot showed that 2G7 treatment decreased the abundance of endogenous vimentin in long bone lysates compared with IgG control antibody treatment while ATF4 abundance remained constant in both control and 2G7 treated long bone samples (Fig. 3D). Thus, these results strongly suggested that TGF␤ suppresses Ocn transcription and osteoblast differentiation via up-regulation of vimentin expression.
An increased protein level can be attributed to increased synthesis and/or decreased degradation. To determine which of these two possibilities contributed to the TGF␤-induced accumulation of vimentin protein in osteoblasts, we tested the effects of cycloheximide, a protein synthesis inhibitor, MG115, a proteasome inhibitor, on vimentin expression. We included Flag-ATF4 as a positive control, because it accumulates in many non-osteoblastic cell lines upon MG115 treatment (36). Analysis of nuclear extracts of COS1 cells transfected with HAvimentin or Flag-ATF4 expression vector showed that treat-  OCTOBER 19, 2012 • VOLUME 287 • NUMBER 43 ment of MG115 (10 g/ml) did not alter the abundance of HAvimentin but did, as expected, increase Flag-ATF4 considerably (Fig. 3E). To address whether endogenous vimentin in osteoblasts behaved similarly to HA-vimentin in COS1 cells, we cotreated ROS17/2.8 cells with rhTGF␤1 (0.5 ng/ml), MG115 (10 g/ml), or two lysosomal inhibitors chloroquine and ammonium chloride. As can be seen in Fig. 3F, the TGF␤-induced increase in vimentin expression was not affected by MG115, chloroquine (100 M), or ammonium chloride (50 mM), but strongly decreased to basal level by cycloheximide (15 g/ml). Collectively, these data indicated that protein neosynthesis is required for TGF␤ to stimulate vimentin expression in osteoblasts.
Western blot demonstrated that both wortmannin and rapamycin (1 and 5 nM) drastically reduced vimentin protein expression induced by TGF␤, which correlated with a strong decrease in Akt and S6K phosphorylation. SB505124 (0.1 and 0.2 nM) effectively decreased TGF␤-induced Smad2 phosphorylation but not TGF␤-induced vimentin expression (Fig. 2,  B-D). As expected, wortmannin and rapamycin (1 and 5 nM) did not inhibit the TGF␤-induced Smad2 phosphorylation (Fig.  4, B--E), demonstrating that these inhibitors do not have an off-target effect at these concentrations. Moreover, overexpression of a dominant negative form of Akt, Akt-AA (38), blocked the increase in vimentin expression induced by TGF␤, which also correlated with an inhibition of S6K phosphorylation induced by TGF␤ (Fig. 4E). Collectively, these data indicated that activation of PI3K-Akt-mTOR signaling, but not the canonical Smad signaling, is required for the stimulation of vimentin protein expression by TGF␤.
To further elucidate the functional relevance of this noncanonical signaling pathway in controlling vimentin expression on ATF4-dependent activation of Ocn expression, we treated ROS17/2.8 reporter cells containing p6xOSE1-Luc with various kinase inhibitors. Consistent with their effects in abolishing the effect of TGF␤ on vimentin production, wortmannin and rapamycin both blunted the inhibition of luciferase activity by TGF␤, whereas SB505124 did not affect it (Fig. 4F). Taken together, these results strengthened the finding that PI3K-Akt-mTOR signaling, but not Smad signaling, acts downstream of TGF␤ to mediate the inhibition of Ocn transcription.

DISCUSSION
In this study, we show that TGF␤ up-regulates post-transcriptionally vimentin expression in osteoblasts through a PI3K-Akt-mTOR non-canonical TGF␤ pathway to inhibit osteoblast differentiation. Thus, this study defines vimentin and ATF4 as downstream mediators of TGF␤, providing two additional modulation points of this ancient and important pathway.

ATF4 and Vimentin as Downstream Targets of TGF␤ in
Osteoblasts-The finding of the monoclonal TGF␤-neutralizing antibody 1D11 to promote mechanical strength in young adult mice (30) motivated us to investigate the mechanism(s) whereby anti-TGF␤ treatment promotes bone mass and fracture resistance. Here, we provided evidence that the inhibition of Ocn expression, ALP activity, and mineralized nodule formation by TGF␤ was attenuated by 50% in Atf4 Ϫ/Ϫ osteoblasts compared with the WT osteoblasts. Moreover, the anabolic effect of 2G7 in WT mice (a 30% increase in trabecular bone volume fraction) is completely blunted in Atf4 Ϫ/Ϫ mice (Fig. 1). These data allow us to conclude that ATF4 is required for transmitting TGF␤ signals in osteoblasts in both cultured primary cells and in vivo. Interestingly, TGF␤ did not decrease ATF4 expression but increased vimentin protein abundance in osteoblasts and bones in vivo (Figs. 2 and 3), which in turn led to the suppression of ATF4 transcriptional activity. Furthermore, silencing endogenous vimentin by siRNA-Vim attenuated the effects of TGF␤ (Fig. 3). Therefore, we conclude that both ATF4 and vimentin are downstream targets of TGF␤. Further investigations are necessary to confirm the in vivo roles that vimentin play in osteoblasts.
Studies using cell and transgenic mouse models have shown that vimentin promotes cell growth in Ras-transformed cells (41) but inhibits cell differentiation (42). Loss of vimentin (Vim Ϫ/Ϫ ) in mice results in failures of vascular adaptation, which leads to pathological conditions such as reduced renal mass (43), glial cellmalformation (44), impaired wound healing (45), decreased arterial resistance to sheer stress (46), and disturbed leukocyte homing to lymph nodes (47). It will be of interest to investigate whether all or part of these defects are attributable to the essential role that vimentin plays in transmitting the signals of TGF␤, given the versatile and essential roles that TGF␤ plays in many physiological and pathological conditions, including the ones affected in Vim Ϫ/Ϫ mice. Furthermore, vimentin is often strongly expressed in undifferentiated mesenchymal cells. Overexpressing vimentin in mesenchymes under its own promoter inhibited cell differentiation (48,49). Thus, it is plausible to view TGF␤ as one of the local growth factors that is required to maintain the "stemness" of mesenchymal stem cells in bone through its up-regulation of vimentin in bone. Thirdly, as an intermediate filament (IF) protein, vimentin also undergoes spatial reorganization in a variety of cell types, in response to stimulation with physiological signals, including TGF␤ (data not shown). It will be important to understand the molecular basis whereby vimentin, a molecule that mostly resides in cytosol, travels into the nucleus to modulate the activity of tissue-specific transcription factors.
Analogous to the signaling axis of TGF␤-vimentin-ATF4 is that of TGF␤-Smad3-Runx2, which regulates Ocn transcription and osteoblast differentiation. In the latter pathway, upon TGF␤ ligand stimulation, Smad3 is first phosphorylated by TGF␤ receptors and then it binds to Smad4 and translocates into the nucleus, where the complex of Smad3/4 interacts with Runx2 (6), a master regulator of osteoblast differentiation (50 -53). Subsequent studies demonstrated that the Smad3/4 complex also recruits histone deacetylases (HDAC) 4 and -5 to the Runx2-binding site of Ocn promoter to suppress Ocn transcription and osteoblast differentiation (54). It is currently unknown whether the vimentin-ATF4 complex can recruit additional repressors, such as HDAC family members, to the OSE1 binding site. More importantly, the relative contribution and potential interactions of TGF␤-vimentin-ATF4 and TGF␤-Smad3-Runx2 pathways remain to be studied.
The Role of PI3K-Akt-mTOR Signaling in Osteoblasts-It has previously been reported that specific inhibition of PI3K signaling with wortmannin stimulates human mesenchymal stem cells to differentiate into osteoblasts (55). Similarly, rapamycin also promotes human embryonic stem cells to differentiate into osteoblasts (56). A positive role of these kinase inhibitors of the PI3K-Akt-mTOR pathway in the regulation of osteoblast differentiation is substantiated by the evidence that both wortmannin and rapamycin blunted the inhibitory effect of TGF␤ on ATF4-dependent Ocn transcription in ROS17/2.8 cells containing the p6xOSE1-Luc reporter (Fig. 4). At odds with these results, a low concentration (0.1 nM) of rapamycin has been shown to inhibit Ocn expression and osteoblast differentiation by suppressing the expression of Runx2, in primary mouse bone marrow stromal cells and MC3T3-E1 mouse osteoblastic cells (57). It is currently unknown whether these conflicting observations are a reflection of difference in cell lines, i.e. human versus mouse cell lines, or in drug concentrations. Furthermore, it needs to be further evaluated the in vivo relevance of the newly discovered TGF␤-PI3K-Akt-mTOR-vimentin-ATF4-Ocn axis. Regardless, our current findings represent a novel paradigm involving vimentin and ATF4 as downstream effectors of TGF␤ signaling in the regulation of osteoblast differentiation.
This study does not exclude factors other than TGF␤ acting as physiological upstream regulators of vimentin expression in osteoblasts. Indeed, parathyroid hormone (PTH), an important bone anabolic agent when used intermittently in vivo, has been reported to suppress the de novo biosynthesis of vimentin in human osteoblastic cells (58). Interestingly, PTH has also been shown to increase the expression and activity of ATF4 in osteoblasts (59). Thus, it is possible that PTH down-regulates vimentin expression in differentiating osteoblasts, which may counteract the action of TGF␤ and lead to stimulation of ATF4's transcriptional activity and osteoblast differentiation. From this perspective, our study provides the first step toward to understanding the complexity of signaling cascades that control the anabolic action of TGF␤ and PTH in bone. It is important to further determine whether vimentin is a convergent point that integrates both TGF␤ and PTH signals to finely tune the differentiation process of osteoblasts.
TGF␤ as a potent osteotropic factor has been extensively studied (60) and our understanding of how it regulates bone development and remodeling has continued to evolve (4,5,30,(61)(62)(63). It is encouraging that in vivo inhibition of TGF␤ activity can stimulate bone formation (30). However, evaluating the long-term effects of TGF␤ neutralizing antibodies on the skeleton will be important, since TGF␤ promotes osteoblast proliferation and migration (63)(64)(65), two functions that are also required for bone formation. In addition, better understanding the molecular signaling pathways downstream of TGF␤ should allow one to target selective molecules in osteoblast, thereby limiting off-target effects of general inhibition of TGF␤ activity and the possible development of an immune response caused by the use of an antibody-based bone anabolic strategy.