FoxO1 Protein Cooperates with ATF4 Protein in Osteoblasts to Control Glucose Homeostasis*

Background: The skeleton regulates glucose metabolism and energy expenditure. Results: Two transcription factors interact to regulate the activity of an osteoblast-secreted hormone favoring energy metabolism. Conclusion: The skeleton utilizes an intricate transcriptional machinery to maintain energy homeostasis. Significance: Transcription factor-mediated regulation of energy metabolism by the skeleton has potential applications in diseases of abnormal glucose metabolism. The Forkhead transcription factor FoxO1 inhibits through its expression in osteoblasts β-cell proliferation, insulin secretion, and sensitivity. At least part of the FoxO1 metabolic functions result from its ability to suppress the activity of osteocalcin, an osteoblast-derived hormone favoring glucose metabolism and energy expenditure. In searching for mechanisms mediating the metabolic actions of FoxO1, we focused on ATF4, because this transcription factor also affects glucose metabolism through its expression in osteoblasts. We show here that FoxO1 co-localizes with ATF4 in the osteoblast nucleus, and physically interacts with and promotes the transcriptional activity of ATF4. Genetic experiments demonstrate that FoxO1 and ATF4 cooperate to increase glucose levels and decrease glucose tolerance. These effects result from a synergistic effect of the two transcription factors to suppress the activity of osteocalcin through up-regulating expression of the phosphatase catalyzing osteocalcin inactivation. As a result, insulin production by β-cells and insulin signaling in the muscle, liver and white adipose tissue are compromised and fat weight increases by the FoxO1/ATF4 interaction. Taken together these observations demonstrate that FoxO1 and ATF4 cooperate in osteoblasts to regulate glucose homeostasis.

FoxO1, one of the four FoxO isoforms of the Forkhead family of transcription factors, is highly expressed in insulin-responsive tissues, including pancreas, liver, skeletal muscle, and adipose tissue. In all these tissues FoxO1 orchestrates the transcriptional cascades regulating glucose metabolism, in part by being a major target of insulin signaling. In most cells insulin signaling favors FoxO1 phosphorylation; this results in FoxO1 nuclear exclusion, thereby preventing its transcriptional activity. In its most recently discovered mode of action in the control of energy metabolism, FoxO1 was shown to act as a transcriptional link between the skeleton and the pancreas as well as insulin target tissues by regulating the novel endocrine function of the skeleton in energy homeostasis (1)(2)(3)(4)(5). Indeed, through its expression in osteoblasts FoxO1 decreases ␤-cell proliferation and function, resulting in a decrease in insulin secretion (1). It also suppresses insulin sensitivity in insulin-target tissues such as adipose tissue, the liver, and the muscle. These effects compromise glucose metabolism and increase blood glucose levels. This function of FoxO1 is due to its ability to promote carboxylation and inactivation of osteocalcin, an osteoblast-secreted hormone that favors, insulin secretion, and sensitivity and energy expenditure (1). Adding another level of complexity to this function in a feedback mode of regulation, FoxO1 is also a target of insulin signaling in osteoblasts (3). Insulin suppresses the activity of osteoblastic FoxO1, thus, promoting osteocalcin bioactivity.
In the context of whole body physiology it is remarkable that the exact same transcriptional mediator of insulin actions in all peripheral insulin-sensitive target organs also regulates the metabolic activity of osteocalcin and its insulin-up-regulating as well as insulin-sensitizing functions. This property establishes FoxO1 as a common unifying link of insulin signaling among all glucose-regulating organs. At the same time it raises the question of how such a ubiquitously expressed transcription factor could fulfill in osteoblasts a function that it does not fulfill in other cell types; that is, to affect the glucose-regulating function of other organs. To address this question we searched for osteoblast-specific or osteoblast-enriched transcription factors that could be effectors or co-regulators of FoxO1 signaling in its metabolic functions in osteoblasts.
ATF4 is a transcription factor that accumulates predominantly in osteoblasts and acts through them to affect glucose metabolism and insulin sensitivity (6). Analysis of Atf4 Ϫ/Ϫ mice showed that these animals had a metabolic phenotype similar to that of mice lacking FoxO1 in osteoblasts and characterized by enhanced insulin secretion and insulin sensitivity in the liver, fat, and muscle. Thus, we examined whether FoxO1-mediated regulation of glucose homeostasis occurs through its interaction with ATF4. Here we show that FoxO1 engages in a functional complex with ATF4 in osteoblasts. In this complex the two transcription factors synergize to regulate glucose metabolism, insulin production, and insulin sensitivity.

EXPERIMENTAL PROCEDURES
Mice-All the protocols and experiments were conducted according to the guidelines of the Institute of Comparative Medicine, Columbia University. Generation of FoxO1 fl/fl , ␣1(I)Collagen-Cre (␣ 1 (I)Col-Cre), and Atf4 ϩ/Ϫ mice has previously been reported (7)(8)(9)(10). Mice with osteoblast specific deletion of FoxO1 (FoxO1 osb Ϫ/Ϫ ) were generated by crossing FoxO1 fl/fl mice with transgenic mice expressing Cre under the control of the osteoblast-specific collagen type 1A1 promoter (␣1(I) Collagen-Cre) as previously described (1). Genotyping was performed at weaning stage by PCR analysis of genomic DNA. In each experiment the mice used were of the same genetic background as they were all littermates. In all experiments data presented were obtained from male animals.
Histological Analysis of Pancreatic Islets, White Adipose Tissue, and Liver Sections-Histological analysis was performed as previously described (1). Briefly, fat and pancreata were collected, fixed overnight in 10% neutral formalin solution, embedded in paraffin, sectioned at 4 m, and stained with hematoxylin and eosin. Pancreatic sections were immunostained for ␤ cells using guinea pig anti-swine insulin polyclonal antibody (Dako). ␤-Cell proliferation was assessed using an antibody recognizing Ki67 antigen, the prototypic cell cyclerelated nuclear protein expressed by proliferating cells in all phases of the active cell cycle. ␤-Cell area represents the surface positive for insulin immunostaining divided by the total pancreatic surface. ␤-Cell mass was calculated as ␤-cell area multiplied by pancreatic weight. Livers were cryoembedded, sectioned at 5 m, and stained with Oil red O (Crystalgen).
Cell Cultures-Primary osteoblasts were prepared from calvaria of 5-day-old pups as previously described (2) and were cultured in fresh ␣MEM and 10% FBS. The OB-6 bone marrowderived osteoblastic cell line has been described (11) and was cultured under the same conditions as calvaria.
Transient Transfections and Luciferase Assays-Cos-7 cells were seeded in 48-well plates at a density of 10 4 cells/well. One day after plating, cells were transfected with Lipofectamine Plus (Invitrogen) according to the manufacturer's protocol. We carried out co-transfections of the FoxO1 or ATF4 expression plasmids with either a FoxO-reporter construct or an Osteocalcin-reporter construct or an ESP-reporter construct (50 ng). The Esp-luc, which bears mutations in the FoxO1 and ATF4 site, was generated by site-directed mutagenesis using the QuikChange site-directed mutagenesis kit (Stratagene) according to the manufacturer's instructions. Nucleotide substitutions at positions 3, 4, and 5 of these elements are indicated in italic: Foxo1, 5Ј-TGGGGTT-3Ј (WT: TGTTTTT) and ATF4, 5Ј-ACGGAA-3Ј (WT: ACATCA). The total amount of DNA was adjusted to 270 ng/well with pCMV5 or pcDNA control vectors. Transfection was stopped by adding 20% FBS. Lucifer-ase assays were carried out using the Dual Luciferase Reporter Assay System (Promega), and luciferase activity was quantified using Fluostar Omega. pRL-CMV Renilla luciferase control vector (20 ng) (Promega) was cotransfected as an internal standard to normalize for transfection efficiency. Normalized luciferase activity is presented as -fold induction over the empty vector control (EV, 3 considered 1). All experiments were repeated at least twice.
Western Blotting and Immunoprecipitation-Osteoblasts or bone extracts (50 g) from wild type and FoxO1 osb Ϫ/Ϫ or Atf4 Ϫ/Ϫ mice were analyzed on a SDS-polyacrylamide gel, transferred to a PVDF membrane, and immunoblotted with pFoxO1 (Cell Signaling), FoxO1, or ATF4 primary antibodies (Santa Cruz Biotechnology, Inc.). For immunoprecipitation, 100 g of bone/cell lysates protein were incubated with 2 g of specific antibodies and 20 l of protein A/G-agarose beads (Santa Cruz) overnight at 4°C on a rotating device. To exclude DNA-mediated effects, co-immunoprecipitation was performed in the presence of ethidium bromide (50 g/ml). Alternatively, cell extracts were treated with benzonase (10 units/ reaction) for 30 min. Image J software was used for gel band quantitative densitometric analysis.
Protein Interaction Analysis-GST-FoxO1 (aa 1-300, 290 -655) constructs and His-ATF4 (aa 1-151, 110 -121, 186 -349) constructs have been described (9,12,13). Full-length GST-FoxO1 construct was obtained from Addgene. The GST and His fusion proteins were expressed in Escherichia coli strain BL21pLys by isopropyl thiogalactose induction. Extracts of the GST fusion protein-transformed cells were coupled with glutathione-Sepharose beads. After SDS-PAGE and Coomassie Brilliant Blue staining for monitoring coupling efficiency, GST fusion protein-bound beads were incubated with purified His fusion proteins overnight, washed extensively, and eluted by boiling in SDS-PAGE loading buffer. Bound proteins were visualized by Western blot using anti-His antibody.
Immunocytochemistry-Primary osteoblasts were grown on 8-well chamber slides and treated with either vehicle or 100 M H 2 O 2 for an hour. After a treatment period cells were fixed with ice-cold methanol for 5 min at Ϫ20°C. After washing with 1ϫ PBS, cells were blocked with 10% normal goat serum in PBS for 1 h and then incubated overnight with indicated primary antibodies (1:200 dilution in 1% goat serum in PBS) in a humidified chamber at 4°C. Samples were washed with 1ϫ PBS and incubated with CY3-conjugated (red) and CY2-conjugated (green) secondary antibodies (Jackson ImmunoResearch) for 1 h at room temperature to visualize FoxO1 and ATF4, respectively. Cells were counterstained with DAPI to show the nuclear morphology. Images were acquired with a Nikon 80i Eclipse Microscope using a Retiga digital camera.
Real-time Quantitative PCR Analysis-DNase I (Invitrogen)treated RNA was reverse-transcribed at 42°C with SuperScript II (Invitrogen). The expression of all the genes was measured by real-time quantitative PCR with the SYBR Green master mix using ␤-actin as endogenous control with 1 cycle at 95°C for 10 min followed by 40 cycles at 95°C for 30 s and 60°C for 1 min. The primer sequences are given in supplemental Table 1.
Electrophoretic Mobility Shift Assay (EMSA)-30-Mer complementary oligonucleotides spanning the ATF4 and the FOXO1 binding site of the murine Esp promoter and first intron or mutated ATF4 site and the FOXO1 site were used to perform gel shift assays. The wild type sequence of the oligonucleotides used for ATF4 site was 5Ј-AGCATCCTGCCAACA-TCACCAAGAACCGGT-3Ј and for FOXO1 site was 5Ј-CAT-TCCCACGCATGTTTTTCTCACCCGTTC-3Ј, The mutated sequence (shown in lowercase) for ATF4 site was 5Ј-AGCAT-CCTGCCAACggaACCAGAACCGGT-3Ј and for FOXO1 site was 5Ј-CATTCCCACGCATGgggTTCTCACCCGTTC-3Ј. Overlapping oligonucleotide strands were heat-denatured and annealed overnight. 50 ng of duplex oligonucleotides was 5Ј-end labeled with [␥-32 P]ATP and T 4 polynucleotide kinase (Promega) and purified using a Micro Bio-Spin P-30 column (Bio-Rad). The eluted probe was used for EMSA. Aliquots of the nuclear preparations from murine primary calvaria cells (5 g of protein) were incubated for 20 min at 27°C with 2 g of poly(dI-dC) with or without unlabeled specific or nonspecific DNA competitor or ATF4 and FOXO1 antibodies (Santa Cruz) in binding buffer followed by the addition of the labeled oligonucleotide probe and incubation for 30 min at 27°C. The samples were separated by electrophoresis on a 6% nondenaturing polyacrylamide gel and analyzed by using Typhoon Phosphor-Imager (Molecular Dynamics).
Metabolic Studies-An intraperitoneal glucose tolerance test was performed by administering 2 g of glucose per kg of body weight intraperitoneally after an overnight fast. Blood glucose was monitored using blood glucose strips (Diabetes Association) and the Accu-Check glucometer. For the insulin tolerance test, mice were fasted from 4 to 6 h and injected intraperitoneally with insulin (0.5 units/kg of body weight), and blood glucose levels were measured at the indicated times. Insulin tolerance test data are presented as percentage of initial blood glucose concentration. Body composition was measured with NMR (Bruker Optics).
Physiological Assays-Sera were collected by heart puncture from mice in the fed state. Blood was kept on ice for 15 min before centrifugation for 15 min at maximum speed. Insulin (Mercodia) was measured by enzyme-linked immunosorbent assay (ELISA). Serum osteocalcin was measured by an immunoradiometric assay (Immutopics) according to the manufacturer's instructions. A triple ELISA system was used for quantification of mouse total, carboxylated, and uncarboxylated osteocalcin as described previously (14). This method measures the exact serum concentration of GLU, GLA13, and total osteocalcin and distinguishes between completely decarboxylated osteocalcin (GLU-OCN) and the osteocalcin decarboxylated on Glu-13, a residue that is critical for in vivo activation of osteocalcin.
Statistical Analyses-Results are given as the means Ϯ S.E. Statistical analyses were performed using unpaired two-tailed Student's t or a one way analysis of variance (Student-Newman-Keuls) for more than two groups.

RESULTS
An interaction between FoxO1 and ATF4 was demonstrated by a co-immunoprecipitation experiment showing that the two transcription factors physically associate in osteoblasts as well as in bone (Fig. 1, A and B). This interaction was DNA-independent as demonstrated by the lack of any effect of ethidium bromide or benzonase on the FoxO1/ATF4 complex (Fig. 1C). We also defined the binding interface between FoxO1 and ATF4 by using deletion mutants of each protein. For FoxO1 we used GST-tagged full-length or two GST-tagged fragments that carry either the N-terminal domain, aa 1-300 containing the Forkhead domain, or the C-terminal domain, aa 290 -655 containing the transactivation domain of FoxO1 (Kitamura et al. These results establish that FoxO1 and ATF4 physically interact through their transactivation and DNA binding domains, respectively. Immunohistochemical analysis in primary osteoblasts confirmed that FoxO1 and ATF4 co-localize in the nucleus (Fig.  1F). However, both proteins can also be found in the cytoplasm as they are known to shuttle between the two subcellular compartments. In the absence of any stimuli from growth factors or stress-related stimulus, the two transcription factors are predominantly located in the cytoplasm. Growth factor or stress signals stimulate their translocation to the nucleus, where they actively operate to initiate transcriptional events that promote multiple cell functions. Indeed, stressing the cells by treatment with H 2 O 2 induced nuclear accumulation and colocalization of FoxO1 and ATF4 in the nucleus (Fig. 1G).
We reasoned that if FoxO1 and ATF4 participate in a functional complex, their interaction would occur at either the transcriptional or the protein activation level. Therefore, we first examined whether one transcription factor affects the expression of the other. Atf4 expression and protein levels were not altered in FoxO1-deficient osteoblasts (Fig. 1, H and J). Similarly, ATF4 inactivation had no effect on gene expression or protein levels of FoxO1 in osteoblasts (Fig. 1, I and Ks), indicating that transcriptional events were not involved in the interaction between FoxO1 and ATF4. In contrast, either transcription factor could transactivate the other. ATF4 stimulated FoxO1 activity as measured on a FoxO1 reporter construct carrying FoxO binding sites from the IGF-1 promoter (Fig. 1L). Because this reporter construct does not contain an ATF4 binding site, the ability of ATF4 to enhance FoxO1 activity on this promoter suggests that ATF4 enhances FoxO1 activity without directly binding to its promoter. In support of this observation, phosphorylation of FoxO1 at Ser-256 was increased in the bones of Atf4 Ϫ/Ϫ mice (Fig. 1K). Because FoxO1 phosphorylation at Ser-258 indicates FoxO1 inactivation, this result suggests that FoxO1 activity is decreased in Atf4 Ϫ/Ϫ bones. Simi-larly, forced expression of FoxO1 along with ATF4 also enhanced the transactivating ability of ATF4 as measured by OG2-Luc, a reporter construct carrying 147 bp of the osteocalcin promoter fused to the luciferase gene. This construct contains one ATF4 binding site (9) (Fig. 1M).
Improved Glucose Metabolism in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ Mice-Having established that the two transcription factors are involved in a functional interaction, we asked whether ATF4 is involved in the glucose-regulating action of FoxO1 through osteoblasts. As a first approach to answering this question we compared the metabolic phenotype of mice lacking either FoxO1 or ATF4 in osteoblasts (1,6). Both deletions decreased glucose levels, improved glucose tolerance and insulin sensitivity, and were associated with increased insulin secretion and high energy expenditure, indicating that the two transcription factors may synergize on osteoblasts to regulate energy metabolism. We tested this hypothesis genetically by using compound mutant mice lacking a single allele of FoxO1 in osteoblasts and ATF4 (FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ ). First we searched for an effect on blood glucose levels. FoxO1 or ATF4 deletion led to low blood glucose levels at the random fed state, whereas FoxO1 or ATF4 happloinsufficiency had a minimal or no effect on blood glucose ( Fig. 2A). Happloinsufficiency for both transcription factors resulted in a low glucose phenotype similar to that of knock-out animals. Similar to glucose levels, the ability to metabolize glucose was improved as seen by the improvement of glucose tolerance in FoxO1 osb ϩ/Ϫ ; Atf4 ϩ/Ϫ mice versus FoxO1 osb ϩ/Ϫ or Atf4 ϩ/Ϫ and wild type control animals (Fig. 2B).
The lower glucose levels of ATF4 and FoxO1 double happloinsufficient mice was associated with an increase in serum insulin levels to an extent similar as that of FoxO1 osb Ϫ/Ϫ and Atf4 Ϫ/Ϫ animals (Fig. 2C). Consistent with the increase in insulin levels, islet size, islet numbers, ␤-cell area, and ␤-cell mass were increased in Atf4 ϩ/Ϫ ;FoxO1 osb ϩ/Ϫ mice (Fig. 2, D-G). The increase in insulin levels in the serum of FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mutant mice was due to an increase in ␤-cell proliferation that was equal to the increase seen in mice lacking FoxO1 in osteoblasts or lacking ATF4 (Fig. 2, H-I).
FoxO1 and ATF4 Function Synergistically to Control Insulin Sensitivity-Low glucose levels and the improvement in glucose tolerance in the FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice was not only due to increased insulin production. Indeed, in addition to hyperinsulinemia, insulin sensitivity, assessed by an insulin tolerance test, was increased with double FoxO1 and Atf4 happloinsufficiency as compared with FoxO1 osb ϩ/Ϫ or Atf4 ϩ/Ϫ mice or wild type littermates (Fig. 3A). Consistent with higher insulin levels and greater insulin sensitivity, the expression of the insulin target gene Pgc1␣ was increased in the muscle of FoxO1 osb ϩ/Ϫ ; Atf4 ϩ/Ϫ mice as compared with FoxO1 osb ϩ/Ϫ , Atf4 ϩ/Ϫ , and wild type animals (Fig. 3B). The expression of two Pgc1␣ target genes, Nrf1 and Mcad, was also increased (Fig. 3, C and D).
To uncover the mechanism leading to increased insulin sensitivity in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice, we examined the expression profile of various insulin-regulating adipokines. Expression levels of the insulin-sensitizing hormone adiponectin (15) were increased by FoxO1 and Atf4 haploinsufficiency in osteoblasts (Fig. 3E). In contrast, expression of the insulin-sensitizing hormone leptin (16) or expression of resistin, an adipokine mediating insulin resistance (17), was not affected in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice (Fig. 3, F and G). These observations were consistent with the lack of any effect of FoxO1 or Atf4 deletion in leptin and resistin expression in FoxO1 osb Ϫ/Ϫ and Atf4 Ϫ/Ϫ mice. Confirming the regulation of adiponectin by the FoxO1/ATF4 interaction in osteoblasts and the increase in insulin sensitivity, expression of the adiponectin targets acyl-CoA oxidase, peroxisome proliferator-activated receptor-␣ (Ppar␣), and uncoupling protein 2 (Ucp2) was increased in the muscle of FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice (Fig. 3, H-J). The up-regulation of insulin-sensitive genes Pgc1␣ and Mcad and the adiponectin targets Ppar␣ and Ucp2 in the muscle may indicate an increase in energy expenditure in FoxO1 osbϩ/Ϫ ;Atf4 ϩ/Ϫ mice. All four genes are involved in mitochondrial biogenesis, and increased mitochondrial respiration increases fatty acid oxidation, which leads to a decrease in lipogenesis, increase in insulin sensitivity, and improved glucose transport. More specifically, PPAR␣ is a key regulator of fatty

FoxO1 Interacts with ATF4
MARCH 16, 2012 • VOLUME 287 • NUMBER 12 acid oxidation in skeletal muscle. Mcad is involved in the first step of the mitochondrial ␤-oxidation of fatty acids. Deficiency in MCAD is the most common inborn error observed in the processing of mitochondrial ␤-oxidation of fatty acids, and it is one of the most common inherited disorders of metabolism. The synergistic effect of FoxO1 and ATF4 in the expression of these genes suggests that the two transcription factors interact to regulate mitochondrial activity and metabolism in the muscle.
In parallel with the improved glucose metabolism, gonadal fat was decreased in mice happloinsufficient for FoxO1 and Atf4, as compared with single heterozygous FoxO1 osb ϩ/Ϫ or Atf4 ϩ/Ϫ mice or wild type animals (Fig. 4A). The decrease in gonadal fat was similar to that seen in FoxO1 osb Ϫ/Ϫ or Atf4 Ϫ/Ϫ mice. Total fat content was decreased (Fig. 4B), whereas lean mass and body weight were not affected in FoxO1 osb ϩ/Ϫ ; Atf4 ϩ/Ϫ mice (Fig. 4, C and D). However, whereas Atf4 Ϫ/Ϫ mice show a decrease in fat content, lean mass, and body weight, FoxO1 osb Ϫ/Ϫ mice showed no changes in any of the three parameters. Therefore, regulation of total fat and muscle mass appears to be independent of a synergistic interaction between the two transcription factors and due to ATF4 deletion. It is FIGURE 2. Improved glucose metabolism and increased insulin production in FoxO1 osb ؉/؊ ;Atf4 ؉/؊ mice. A, blood glucose at random feeding is shown; n ϭ 8. B, glucose tolerance test; n ϭ 8 mice/group. C, shown are serum insulin levels at random feeding is shown; n ϭ 8. Insulin staining (D) and increased islet number (E), ␤-cell area (F), and ␤-cell mass (G) Ki67 immunostaining (H) shows larger islets and increased ␤-cell proliferation in the pancreas of FoxO1 osb ϩ/Ϫ ; Atf4 ϩ/Ϫ/Ϫ mice (I). n ϭ 5 mice/group. In all panels bars indicate means Ϯ S.E. *, p Ͻ 0.05 versus WT; #, p Ͻ 0.05 versus FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ group. All mice were 2 months of age.

FoxO1 Interacts with ATF4
possible that these additional properties of ATF4 are due to the fact that it was deleted in all cell types rather than just osteoblasts as was the case with FoxO1. Alternatively, ATF4 may regulate distinct protein(s) in osteoblasts with additional metabolic effects. Indeed, ablation of osteoblasts in a mouse model, has indicated that in addition to osteocalcin, other osteoblastsecreted proteins may confer the glucose-regulating properties of the skeleton (5). The decrease in gonadal fat weight demonstrated in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice occurred despite their hyperinsulinemia and improved insulin sensitivity, suggesting a defect in adipogenesis that would be independent of insulin regulation. Indeed, adipocyte numbers were decreased by 23% in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice (Fig. 4E). In contrast, adipocyte size was increased despite the decrease in perigonadal fat weight further, suggesting that adipocyte differentiation may be compromised in these mice (Fig. 4F).
Consistent with the decrease in gonadal fat weight, the expression of the adipogenic gene C/EBP␣ and two lipolytic genes perillipin and triglyceride lipase (Tgl), whose expression is inhibited by insulin, was decreased in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice as compared with FoxO1 osb ϩ/Ϫ , Atf4 ϩ/Ϫ , and wild type mice (Fig. 4, G-I). Expression of lipoprotein lipase (Lpl) was unaffected (Fig. 4J). These molecular changes indicated that whereas both adipogenesis and lipolysis may be regulated by a FoxO1/ATF4 interaction in osteoblasts, lipogenesis and fatty acid uptake are probably not affected.
Finally, we looked for a potential involvement of the FoxO1 and ATF4 synergism in the control of insulin signaling in another major glucose-regulating organ, the liver. Suggesting a role for the FoxO1/ATF4 interaction, expression of the insulin target FoxA2, which regulates lipogenesis and ketogenesis during fasting, was increased in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice as
FoxO1/ATF4 Interaction Suppresses Osteocalcin Activity-To identify the molecular target that mediates the effects of osteoblast-expressed FoxO1 and ATF4 to the pancreas and the other insulin sensitive target tissues, we focused on osteocalcin and examined whether the two transcription factors co-operatively affect osteocalcin expression or activity. Osteocalcin expression in the bone and serum levels were not altered in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice as compared with single heterozygous animals or wild type littermates (Fig. 5, A and B). As expected, expression and serum levels of osteocalcin were increased in FoxO1 osb Ϫ/Ϫ but decreased in Atf4 Ϫ/Ϫ mice (Fig. 5,  A and B). Consistent with these observations, in cotransfection experiments performed in Cos-7 cells, a FoxO1 expression vector decreased the activity of a reporter construct containing a 2.9-kb fragment of the osteocalcin promoter and first intron fused to the luciferase gene (Fig. 5C). On the other hand, ATF4 stimulated the activity of the osteocalcin reporter construct (Fig. 5C). The osteocalcin reporter contains the previously identified Ϫ1270 bp (TGTTTTG), -1074 bp (TGTTTT), and ϩ250 bp (TGTTTGC) FoxO1 binding sites as well as an ATF4 binding site at Ϫ75 bp (1,9). However, transcriptional repression by FoxO1 is rarely mediated through direct binding to any of the known FoxO1 binding sites (18). Thus, we examined whether FoxO1 represses osteocalcin transcription through inhibiting Runx2. We found that whereas Runx2 expression was not altered in the bones of FoxO1 osb Ϫ/Ϫ mice (Fig. 5D), the ability of Runx2 to activate osteocalcin transcription was inhibited by FoxO1 (Fig. 5E). The Runx2 binding site (Ϫ146 bp) is present in the reporter construct used.
Because osteocalcin expression was not affected in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice, we examined potential changes in its activity. Osteocalcin exists in a carboxylated or partially uncarboxylated (undercarboxylated) form. In its under/uncarboxylated form, osteocalcin acts as a hormone to favor alto-gether ␤-cell proliferation, insulin secretion, insulin sensitivity, and energy expenditure (2). We examined whether FoxO1 and ATF4 regulate osteocalcin activity by measuring its degree of carboxylation in the serum. We found that 33% of osteocalcin present in the serum of FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice was undercarboxylated (Fig. 5F), as measured by decarboxylation on Glu-13, a residue that has been shown to be critical for in vivo activation of osteocalcin (14). In contrast, only 23% of undercarboxylated osteocalcin was present in the serum of FoxO1 osb ϩ/Ϫ or Atf4 ϩ/Ϫ or wild type mice. Undercarboxylated osteocalcin was increased in FoxO1 osb Ϫ/Ϫ and Atf4 Ϫ/Ϫ mice to levels similar to that of FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice. Osteocalcin bioactivity can be regulated in a bimodal mechanism of action. In the first mechanism osteocalcin activity is negatively regulated by another gene expressed in osteoblasts, Esp. Protein-tyrosine phosphatase (OST-PTP), the product of Esp, decreases osteocalcin bioactivity by favoring its carboxylation (2). In the second mechanism bone resorption induces a change in the pH that spontaneously decarboxylates and activates osteocalcin (3). Respective to the latter mechanism, we did not detect any increases in osteoclast function in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice that would indicate an increase in bone resorption (Fig. 5G). However, expression of Esp was reduced in the bone of FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ mice as compared with heterozygous FoxO1 osb ϩ/Ϫ or Atf4 ϩ/Ϫ and to wild type mice (Fig. 5H). The mechanism of the stimulatory effect of FoxO1 and ATF4 on Esp expression was explored further using Cos-7 cells. Transfection of either FoxO1 or ATF4 stimulated the activity of Esp as measured using an Esp reporter construct that carries 722 bp of the promoter region and 1095 bp of the first intron and exon of the gene (region Ϫ722 to ϩ1095) (Fig.  5I). This region of Esp contains a FoxO1 binding site at 947 bp of the first intron of the gene and one ATF4 binding site present at position Ϫ340 bp (1,9). A combination of different amounts of FoxO1 and ATF4 showed a synergistic effect of the two transcription factors in up-regulating Esp-luc activity. Collectively, these observations suggested that down-regulation of Esp expression in FoxO1 osb ϩ/Ϫ ;Atf4 ϩ/Ϫ as compared with FoxO1 osb ϩ/Ϫ or Atf4 ϩ/Ϫ mice could account for the decrease in osteocalcin carboxylation and the metabolic phenotype of the FoxO1;Atf4 mutant mice.
The synergistic mode of action of FoxO1 and ATF4 on Esp expression was further examined. Site directed mutagenesis was performed on each of the FoxO1 or ATF4 binding sites

FoxO1 Interacts with ATF4
present in the Esp promoter. Mutation of the ATF4 binding site (ACATCA) present at Ϫ340 bp of the Esp promoter abolished binding of ATF4 to Esp (Fig. 5J). Similarly, mutation of the TGTTTTT binding site of FoxO1 at 947 bp of the first intron of the Esp gene abolished binding of FoxO1 to Esp (Fig. 5K). We found that mutation of either the ATF4 or FoxO1 binding site abolished activation of Esp by each transcription factor (Fig.  5L). In addition, combination of FoxO1 and ATF4 also failed to induce activation of Esp when either the ATF4 or FoxO1 binding site was mutated. Thus, synergistic activation of Esp by ATF4 and FoxO1 requires both factors to be bound simultaneously to DNA. Collectively, these observations indicate that FoxO1 and ATF4 interact in osteoblasts to control glucose homeostasis by regulating the carboxylation of osteocalcin through regulating Esp expression.

DISCUSSION
In this study we have shown that two transcription factors, FoxO1 and ATF4, interact in osteoblasts to control their endocrine properties as regulators of glucose homeostasis. Osteoblast-expressed FoxO1 and ATF4 cooperate to increase blood glucose levels, trigger glucose intolerance and insulin insensitivity, and hinder insulin signaling in insulin-sensitive target tissues such as the muscle, liver, and white adipose tissue. Pancreatic function is also affected as the FoxO1/ATF4 interaction in osteoblasts leads to suppression of ␤-cell proliferation with a subsequent decrease in insulin production.
The hormonal mediator of the energy homeostatic properties of the skeleton is a protein specifically expressed by osteoblasts, osteocalcin (2). Glucose metabolism, insulin sensitivity, and energy expenditure are all favored by osteocalcin as shown by genetic models of gain and lack of osteocalcin function. The metabolic activity of osteocalcin is determined by its degree of carboxylation. Osteocalcin is carboxylated in three glutamic residues in a vitamin K-dependent post-translational modification that confers to it high affinity to minerals present in bone. However, it is the undercarboxylated form of osteocalcin (the one in which Glu-13 is not carboxylated) that is metabolically active on ␤-cells (14). Osteocalcin carboxylation is under the control of insulin signaling in osteoblasts, which acting through its receptor suppresses the expression of the anti-osteoclastogenic cytokine osteoprotegerin (Opg) and as a result stimulates bone resorption (3). In turn, the acidic environment of the resorptive lacunae promotes osteocalcin decarboxylation. Esp, protein-tyrosine phosphatase that is also expressed in osteoblasts, acts as a substrate of the insulin receptor. Esp antagonizes insulin signaling in osteoblast and by doing so promotes carboxylation and, therefore, osteocalcin inactivation. Confirming the genetic studies, administration of uncarboxylated osteocalcin to mice lowers blood glucose levels by increasing insulin production and insulin sensitivity and favors white adipose tissue metabolism by enhancing energy expenditure (19). Our studies show that FoxO1 and ATF4 potently control the glucose-regulating properties of osteoblasts precisely by controlling osteocalcin activity, as their cooperative action suppresses osteocalcin carboxylation. This effect is in line with and a consequence of their ability to concomitantly stimulate expression of Esp by directly binding to the Esp promoter.
A calculation of the actual amounts of undercarboyxlated osteocalcin in the different groups suggests that whereas wild type, FoxO1 osb ϩ/Ϫ , and Atf4 ϩ/Ϫ mice have the same amount (30 ng/ml), active osteocalcin levels are increased in Atf4 ϩ/Ϫ ; FoxO1 osb ϩ/Ϫ mice (42 ng/ml), very high in the FoxO1 osb Ϫ/Ϫ mice (90 ng/ml) but low in the AtF4 Ϫ/Ϫ mice (23 ng/ml). Based on this information, the levels of undercarboxylated osteocalcin in these mice do not completely explain the improved metabolic phenotype of FoxO1 osb ϩ/Ϫ and Atf4 Ϫ/Ϫ animals. At the present time we can foresee two explanations for this apparent inconsistency. The first one is that the ratio of undercarboxylated to total osteocalcin is a more reliable marker of osteocalcin activity than the total level of undercarboxylated osteocalcin. This would imply that either carboxylated and undercarboxylated osteocalcin may have opposite effects on metabolism or that total osteocalcin levels itself may affect glucose homeostasis. In support of this hypothesis are studies suggesting that total and carboxylated osteocalcin can both influence glucose homeostasis in humans (20,21). Alternatively, it is possible that the metabolic phenotype of the Atf4 and FoxO1 osb null animals is not solely due to a change in osteocalcin levels but is the result of additional actions of these transcription factors in osteoblasts. In support of the latter contention, recent evidence suggests that there are osteocalcin-independent influences of osteoblasts on energy metabolism (5).
The cooperative mode of interaction with another transcription factor as a means of regulating target protein activity is not common for the spectrum of FoxO1 functions. Indeed, most frequently FoxO1 has been shown to regulate the expression of transcription factors that are required for cell differentiation and function. For example, in the adipose tissue FoxO1 suppresses the expression of a master adipogenic transcription factor, PPAR␥, through direct binding to the promoter. However, these effects are transcriptional and involve direct regulation of the expression of FoxO1 target transcription factors. Unlike them, the interaction of FoxO1 with ATF4 occurs at the protein level, is independent of changes in gene expression, and results in regulation of the activity of common transcriptional targets shared by both transcriptional factors. A similar mode of FoxO1 interactions with other transcription factors has been reported in both the pancreas and white adipose tissue. In the latter, FoxO1 physically interacts with C/EBP␣ to promote expression of adiponectin (22) but suppresses PPAR␥ activity by competitively inhibiting the formation of a PPAR␥-RXR functional complex (23). In the pancreas FoxO1 negatively regulates expression of Pdx1, which regulates ␤-cell development and mass by competing with the transcription factor FoxA2 for binding to the Pdx1 promoter (24). This is one of the main effects that mediates the suppressive effect of pancreatic FoxO1 on ␤-cell differentiation (25). A similar functional and physical interaction of FoxO1 with Notch1 is described in myoblasts, which leads to corepressor clearance from the Notch effector Csl, leading to stabilization of a functional FoxO1-Notch1 complex (12).
We reveal a mechanism by which two transcription factors, the broadly expressed FoxO1 and the osteoblast-enriched ATF4, interact in osteoblasts to regulate glucose metabolism and insulin sensitivity. In view of FoxO1 biology, this mecha-nism of action may find a counterpart in the regulation of glucose metabolism and insulin signaling in other glucose-regulating organs where other transcription factors are involved in cell fate determination and function, such as ␤-cell development, myocyte formation, and hepatic cell function. Respective to skeletal biology, we show that FoxO1 and ATF4 regulate energy metabolism at least in part in an osteocalcin-dependent manner. The fact that many recent clinical studies have suggested that osteocalcin is a marker of glucose tolerance (2, 4, 26 -29) attributes to the FoxO1/ATF4 interaction a biological significance with potential applications in diseases of abnormal glucose metabolism.