Wnt Protein Signaling Reduces Nuclear Acetyl-CoA Levels to Suppress Gene Expression during Osteoblast Differentiation*

Developmental signals in metazoans play critical roles in inducing cell differentiation from multipotent progenitors. The existing paradigm posits that the signals operate directly through their downstream transcription factors to activate expression of cell type-specific genes, which are the hallmark of cell identity. We have investigated the mechanism through which Wnt signaling induces osteoblast differentiation in an osteoblast-adipocyte bipotent progenitor cell line. Unexpectedly, Wnt3a acutely suppresses the expression of a large number of genes while inducing osteoblast differentiation. The suppressed genes include Pparg and Cebpa, which encode adipocyte-specifying transcription factors and suppression of which is sufficient to induce osteoblast differentiation. The large scale gene suppression induced by Wnt3a corresponds to a global decrease in histone acetylation, an epigenetic modification that is associated with gene activation. Mechanistically, Wnt3a does not alter histone acetyltransferase or deacetylase activities but, rather, decreases the level of acetyl-CoA in the nucleus. The Wnt-induced decrease in histone acetylation is independent of β-catenin signaling but, rather, correlates with suppression of glucose metabolism in the tricarboxylic acid cycle. Functionally, preventing histone deacetylation by increasing nucleocytoplasmic acetyl-CoA levels impairs Wnt3a-induced osteoblast differentiation. Thus, Wnt signaling induces osteoblast differentiation in part through histone deacetylation and epigenetic suppression of an alternative cell fate.

Developmental signals in metazoans play critical roles in inducing cell differentiation from multipotent progenitors. The existing paradigm posits that the signals operate directly through their downstream transcription factors to activate expression of cell type-specific genes, which are the hallmark of cell identity. We have investigated the mechanism through which Wnt signaling induces osteoblast differentiation in an osteoblast-adipocyte bipotent progenitor cell line. Unexpectedly, Wnt3a acutely suppresses the expression of a large number of genes while inducing osteoblast differentiation. The suppressed genes include Pparg and Cebpa, which encode adipocyte-specifying transcription factors and suppression of which is sufficient to induce osteoblast differentiation. The large scale gene suppression induced by Wnt3a corresponds to a global decrease in histone acetylation, an epigenetic modification that is associated with gene activation. Mechanistically, Wnt3a does not alter histone acetyltransferase or deacetylase activities but, rather, decreases the level of acetyl-CoA in the nucleus. The Wnt-induced decrease in histone acetylation is independent of ␤-catenin signaling but, rather, correlates with suppression of glucose metabolism in the tricarboxylic acid cycle. Functionally, preventing histone deacetylation by increasing nucleocytoplasmic acetyl-CoA levels impairs Wnt3a-induced osteoblast differentiation. Thus, Wnt signaling induces osteoblast differentiation in part through histone deacetylation and epigenetic suppression of an alternative cell fate.
A major challenge in modern biology is to understand the mechanism of cellular differentiation in metazoans. Although sharing the same genes, the diverse cell types of a given organism each exhibit a unique profile of gene expression. Gene expression in turn is regulated by the local chromatin milieu at each gene locus. At the molecular level, this milieu reflects epigenetic modifications of both DNA (e.g. methylation) and the associated proteins such as histones (1). Among histone modifications, acetylation and methylation have been most extensively studied. Whereas histone acetylation is often associated with active gene expression, the roles of histone methylation are more complex depending on the position of the lysine residue being modified (2,3). Thus, epigenetic modifications of the chromatin are at the core of cell differentiation. Although it is known that a small number of developmental signals are chiefly responsible for cell differentiation, it is not well understood how these signals confer the epigenetic modifications necessary for cell type-specific gene expression. Studies to date have focused mostly on the direct downstream transcription factors that engage specific genomic loci and recruit chromatin-modifying enzymes. However, it is becoming increasingly clear that various metabolic intermediates, such as NAD ϩ , acetyl-CoA, and ␣-ketoglutarate, can also influence the epigenetic modifications of the chromatin milieu (4 -6). Whether or not developmental signals may exert epigenetic regulation through intermediate metabolites has not been explored.
Wnt signaling is a critical regulator of cell differentiation in metazoans both during embryogenesis and in postnatal tissue homeostasis (1, 2. Wnt proteins are secreted glycoproteins that activate multiple intracellular signaling cascades. In the most well characterized pathway, Wnt stabilizes ␤-catenin, which enters the nucleus and interacts with the Lef/Tcf family of transcription factors to activate downstream target genes. In addition, Wnt proteins can also activate the Rho family of small GTPases, the Ca 2ϩ pathway, PKC␦, mTORC1, and mTORC2 (7)(8)(9)(10)(11)(12)(13). Through the various effectors, Wnt signaling has been recently shown to regulate cellular metabolism in both normal and cancerous cells (14 -17). Wnt signaling has been shown to regulate chromatin modifications through ␤-catenin, which forms complexes with a number of chromatin-modifying enzymes including histone acetyltransferases (HAT), 3 TIP60, and CBP/p300, the H3K4 histone methyltransferase MLL1⅐ MLL2 complex, and the H3K79 histone methyltransferase Dot1-containing complex (18 -21). However, potential epigenetic regulation by Wnt through ␤-catenin-independent mechanisms is largely unexplored. Wnt signaling has emerged as an important mechanism regulating bone formation in mammals (22). In the mouse embryo, deletion of ␤-catenin or both Lrp5 and Lrp6 in the skeletogenic progenitors abolishes osteoblast differentiation, indicating that Wnt signaling through ␤-catenin is critical for embryonic osteoblastogenesis (23)(24)(25)(26)(27). In postnatal life, loss-and gain-offunction mutations in LRP5 cause low and high bone mass syn-dromes, respectively, in humans (28 -30). Moreover, deficiency in SOST, a secreted inhibitor that prevents the binding of WNT to LRP5 or LRP6, results in high bone mass in human patients (31,32). In the mouse, deletion of Lrp5 causes osteopenia (33,34), whereas loss of SOST increases bone mass (35). The mechanism through which Wnt signaling stimulates osteoblast differentiation, however, remains incompletely understood.
Here we present evidence that Wnt signaling induces osteoblast differentiation in part through rapid epigenetic suppression of genes responsible for alternative cell fates. Specifically, Wnt3a acutely reduces nuclear acetyl-CoA, the necessary substrate for histone acetyltransferases, resulting in a global decrease in histone acetylation. These data reveal a novel mechanism by which Wnt signaling suppresses alternative cell fates through modulation of acetyl-CoA levels in the cell.

Experimental Procedures
Cell Culture-ST2 cells were cultured in ␣-MEM (Gibco) supplemented with 10% FBS (Invitrogen), 5 mM glucose, and 2 mM glutamine. All experiments were carried out at a seeding density of 13,000 cells/cm 2 . Wnt treatments were initiated 24 h after plating by replacing the medium with ␣-MEM supplemented with 50 ng/ml recombinant Wnt3a (R&D Systems) versus vehicle control (0.1% CHAPS in PBS) or by replacing the medium with ␣-MEM diluted 1:1 with Wnt3a-or L-cell (L)conditioned medium (CM). In the Dkk1 experiment the cells were pretreated with 500 ng/ml Dkk1 for 1 h before treatment with Wnt3a and Dkk1 for 6 h. BMP2 treatments were initiated by replacing the medium with ␣-MEM supplemented with either 300 ng/ml recombinant BMP2 (R&D Systems) or vehicle control (0.1% HCl). In indicated experiments growth media was supplemented with 10 ng/ml trichostatin A or 20 mM sodium acetate. For Wnt3a-or BMP2-induced mineralization, cells were treated with Wnt3a or BMP2 with or without the other reagents as indicated for 72 h and then switched to mineralization medium (␣-MEM supplemented with 50 mg/ml ascorbic acid (Sigma) and 10 mM ␤-glycerophosphate (Sigma)) for 6 days with a change of medium every 48 h. Mineralization was assayed by von Kossa staining.
shRNA or siRNA Knockdowns-Lentiviral vectors expressing shRNA against ␤-catenin and Pparg were obtained from the RNAi core at Washington University School of Medicine. The lentiviral vector pLKOpuro was modified to express shRNAs targeting ␤-catenin, Pparg, or either LacZ or RFP as negative controls ( Table 1). The shRNA-expressing lentiviral vector was co-transfected with plasmids pMD2.g and psPax2 into 293 cells. Viral containing media were collected and filtered. ST2 cells were plated at 13,000 cells/cm 2 and infected for 12 h and then recovered for 36 h in regular media. ATP citrate lyase (Acly) was knocked down with the siRNA On-Targetplus SMART pool (Dharmacon) as follows. ST2 cells were plated at 4200 cells/cm 2 and transfected with 25 nM SMARTpool #104112 and 0.25 l of DUO transfection reagent (Dharmacon) for 48 h.
Histone Deacetylase (HDAC) and HAT Activity Assays-HDAC activity was measured in 50 g of nuclear extract with the Fluor-de-lys HDAC fluorometric activity assay kit (Enzo Life Sciences, BML-AK500). HAT activity was measured in 50 g of nuclear lysate by a colorimetric HAT assay (Abcam, ab65352). Each assay was performed in technical triplicate from three independent experiments.
RNA Isolation and qPCR-Total RNA was isolated from cultured cells using the RNAeasy kit with on-column DNase treatment (Qiagen). Reverse transcription was performed using 100 ng of total RNA with the iScript cDNA synthesis kit (Bio-Rad). Reactions were set up in technical and biological triplicates in a 96-well format on an ABI StepOne Plus using SYBR green chemistry (SsoAdvanced, Bio-Rad). The PCR conditions were 95°C for 3 min followed by 40 cycles of 95°C for 10s and 60°C for 30s. Gene expression was normalized to 18S rRNA, and relative expression was calculated using the 2 Ϫ(⌬⌬Ct) method. Primers were used at 0.1 M, and their sequences are listed in Table 2. PCR efficiency was optimized, and melting curve analyses of products were performed to ensure reaction specificity.
RNA-Seq-Poly-A RNA was isolated from 20 g of total RNA using Oligo-dT beads. mRNA was fragmented and reversetranscribed into double-stranded cDNA. cDNA was bluntended followed by the addition of A base to 3Ј end and the ligation of sequencing adapters to the ends. The fragments then underwent PCR amplification for 12 cycles. The resulting libraries were sequenced using the Illumina HiSeq-2000 as single reads extending 42 bases (GTAC, Washington University). The raw data were demultiplexed and aligned to the reference genome (mm9) using TopHat. Both Cufflinks and Partek Genomic Suite were used independently to assemble transcripts and analyze expression. Gene Ontology analysis was performed using the GORILLA platform.
ChIP-seq-Chromatin and protein complexes from 1.5 ϫ 10 7 ST2 cells were cross-linked for 10 min in 1% formaldehyde and flash-frozen. Chromatin was sonicated to an average size of 200 -450 bp using a Sonics Vibracell sonicator (model Vcx 500). Chromatin complexes were immunoprecipitated using antibodies that recognized the histone modifications H3K9ac (Millipore 17-658) and H3K4me3 (Millipore CS200580). ChIP DNA was purified with the PCR purification kit (Qiagen), blunt-ended, and ligated to the sequencing adapters. The fragments were size-selected to 200 -500 base pairs and underwent PCR amplification for 15 cycles. The resulting libraries were sequenced using the Illumina HiSeq-2000 as single reads extending 42 bases. The raw data were demultiplexed and aligned to the reference genome (mm9) using Bowtie. MACS was used for peak calling and generation of Wiggle plots. Transcription start site enrichment, meta-gene analysis, and heatmaps were generated using Cistrome. Western Blotting-ST2 cells were scraped in lysis buffer containing 50 mM Tris (pH 7.4), 15 mM NaCl, 0.5% Nonidet P-40, and a protease inhibitor mix (Roche Applied Science #04693124001). Protein concentration was estimated by the BCA method (Pierce). Protein (20 g) was resolved on 12% polyacrylamide gel and subjected to immunoblot analysis using the respective antibodies. Prestained dual color protein standards were used as molecular weight markers (Bio-Rad). After protein transfer, the blots were typically cut into multiple strips according to the molecular weight markers so that proteins of different sizes could be detected simultaneously from the same blot. Antibodies against Acly (Cell Signaling #4332, 1:1000 dilution) and ␤-actin (Cell Signaling #4970, 1:1000 dilution) were used to detect the respective protein levels. The immunoblots were blocked for 1 h at room temperature in 5% BSA (TBS, 0.1% Tween) followed by an overnight incubation at 4°C in their respective diluted primary antibody solutions. Membranes were then washed three times using TBS/Tween 0.1% and further incubated with the secondary antibody, HRP goat antirabbit (Invitrogen, catalogue no. G21234) in 5% BSA (TBS/ Tween 0.1%) for 1 h at room temperature (1:5000). All blots were developed using either the Immun-Star WesternC Chemiluminescence kit or the Clarity ECL substrate (Bio-Rad). Each experiment was repeated for a minimum of three times with three independently prepared protein samples.
Acetyl-CoA Extraction-For the extraction of acetyl-CoA, 12-15 ϫ 10 7 cells were trypsinized, washed, and incubated in hypotonic lysis buffer for 30 min, and the nuclei were pelleted at 3000 ϫ g for 5 min at 4°C. Nuclei were resuspended in 10% trichloroacetic acid (Sigma). 100 ng of propionyl-CoA (Sigma) standard was added, and the nuclei were sonicated at 20% amplitude for 30 s and centrifuged at 15,000 ϫ g for 5 min at 4°C to precipitate the protein pellet. The nuclear lysate was bound to an HLB Oasis 3cc column, washed, and eluted in 600 l of methanol.
Acetyl-CoA Quantification by Mass Spectrometry-HPLC mass spectrometric analysis of acetyl-CoA was conducted on a Thermo Finnigan (San Jose, CA) TSQ Vantage triple-stage quadrupole instrument with Thermo Accela UPLC operated with Xcalibur software. Lipid extracts with propyl-CoA internal standard were injected with a Thermo Accela autosampler and separated with a Thermo BETASIL C18 reversed-phase column (100 ϫ 2.1 mm; 3-m particle size) coupled to the electrospray ionization source, where the skimmer was set at ground potential, the electrospray needle was set at 4.0 kV, and temperature of the heated capillary was 300°C. Tandem mass spectrometry was operated in the negative-ion mode in which precursor-ion scan of m/z 134.1 with a collision energy of 21 V was employed. The tandem quadrupole mass spectra were acquired in the profile mode in the mass range of 390 -420 Da with a scan rate of 1 scan/s for detection of the acyl-CoA as the [M-2H] 2Ϫ ions. The mass resolution of the instrument was tuned to 0.6 Da at halfpeak height.
HPLC separation was carried out with a step gradient in which mobile phase A was 5% acetonitrile in water with 0.1% NH 4 OH and mobile phase B was 0.1% NH 4 OH in 95% acetoni- AGGCGTTTATCCCCCGATTC CGTAACCAAACCCAGCCAGG trile/water.
Step gradient elution was performed with an initial condition of 100% A for 4 min and switched to mobile phase B for an additional 4 min. The column was then re-equilibrated at 100% B for 10 min before subsequent injections. The flow rate was 50 l/min at ambient temperature. Nuclear Citrate Measurement-Nuclear citrate concentration was measured using the citrate fluorometric assay kit (Sigma, MAK057). For the extraction of citrate, nuclei from ST2 cells treated with L-or Wnt3a-conditioned media were resuspended in 0.1 ml of citrate assay buffer. Insoluble material was removed by centrifugation at 15,000 ϫ g for 10 min at 4°C. Nuclear supernatants were concentrated on 10-kDa spin columns (Abcam ab93349) at 10,000 ϫ g for 20 min. 25 l of nuclear extract was used for each technical replicate.
Citrate concentration was extrapolated from a standard curve and normalized to cell number. Each assay was performed in technical duplicate from six independent experiments.
CO 2 Trap Experiments-The procedure is as described previously (37). Briefly, ST2 cells were seeded at 4 ϫ 10 5 /flask and cultured in media containing 2 l of 0.1 mCi/ml 14 C-labeled glucose (total radioactivity 444,000 dpm) with or without Wnt3a (100 ng/ml) in the sealed CO 2 trap system for 24 h before CO 2 was harvested for the measurement of radioactivity.  consumption, an aliquot of the medium was taken for measurements before the injection of H 2 SO 4 .

Results
Wnt3a Acutely Suppresses Gene Expression-To investigate how Wnt signaling induces osteoblast differentiation, we applied Wnt3a to ST2 cells, a progenitor cell that possesses both osteogenic and adipogenic potential but undergoes osteoblast differentiation in response to Wnt3a (13,38,39). Wnt3ainduced osteoblast differentiation was confirmed by the timedependent up-regulation of marker genes such as Alpl, Ibsp, Bglap, and Sp7 (Fig. 1A). Interestingly, Sp7 and Bglap were initially suppressed by Wnt3a before being up-regulated at the later time points. To gain insights about the early transcriptional response to Wnt3a, we profiled the transcriptome by RNA-seq in these cells after 6 and 24 h of treatment. For comparison, we conducted similar experiments with BMP2, another potent osteogenic signal for ST2 cells. Unexpectedly, Wnt3a suppressed the mRNA levels of a large number of genes as early as 6 h of treatment (144 genes with Ͼ2-fold decrease), and the number of suppressed genes surpassed those induced at 24 h (487 versus 329 genes, with a Ͼ2-fold change) (Fig. 1B). In contrast, BMP2 altered the expression of much fewer genes at 6 h, and the number of genes suppressed was considerably lower than those induced at both 6 and 24 h (Fig. 1B). The prominent feature of gene suppression by Wnt3a was unexpected, as gene activation through ␤-catenin is presumed to be the dominant mode of Wnt3a signaling. We performed gene ontology (GO) analysis of genes suppressed at least 2-fold by Wnt3a using GOrilla (40). This analysis revealed that the suppressed transcripts were significantly enriched for genes involved in "positive regulation of cell differentiation" (GO: 0045597, p ϭ 7.22E Ϫ10 ) and "positive regulation of transcription, DNA-dependent" (GO:0045893, p ϭ 1.87E Ϫ06 ). These genetic cadres include transcription factors known to regulate the differentiation of adipocytes (e.g. Pparg, Cebpa, and Id2) and chondrocytes (e.g. Sox5 and Sox9) (see Fig. 1C for the RNAseq alignment for the representative gene Pparg). With quantitative PCR, we confirmed that these adipogenic or chondrogenic transcription factors were all suppressed upon 6 h of Wnt3a treatment (Fig. 1D). Like recombinant Wnt3a, virally expressed Wnt10b or Wnt7b, both known to induce osteoblast differentiation, caused similar suppression of those genes (data not shown). Others have also observed the suppressive effect of Wnt10b on Pparg and Cebpa and further demonstrated that knockdown of Pparg or Cebpa induced spontaneous osteoblast differentiation in ST2 cells (38). We confirmed that Pparg knockdown (by ϳ50%) alone was sufficient to induce the expression of osteoblast markers Alpl and Ibsp, albeit less efficiently than Wnt3a (Fig. 1E). In contrast to Wnt, BMP2 did not suppress Pparg or Cebpa nor did it impair adipocyte differentiation in ST2 cells (data not shown). Thus, unlike BMP, Wnt appears to induce osteoblast differentiation in part through suppression of adipogenic transcription factors and the alternative cell fate.
Wnt3a Acutely Reduces Histone Acetylation-To investigate the mechanism of gene suppression by Wnt3a, we first evaluated if ␤-catenin-mediated transcription was required. Expression of a dnTCF4 completely abrogated the induction of ␤-catenin target genes like Lef1 by Wnt3a but had no effect on the suppression of Pparg or Cebpa at either 6 or 96 h of Wnt treatment (Fig. 2, A and B). Similarly, knockdown of ␤-catenin had no effect on Wnt3a-dependent suppression (Fig. 2C). These results are consistent with the fact that Wnt7b suppressed gene expression even though it did not activate ␤catenin-dependent gene transcription in ST2 cells (13). Thus, gene suppression in response to Wnt3a is unlikely to be secondary to gene activation mediated by the ␤-catenin⅐Tcf complex.
Because histone acetylation and methylation are associated with gene transcription, we next evaluated the effect of Wnt3a on such modifications of bulk histones. Remarkably, after 6 h of treatment, Wnt3a reduced the acetylation of histone H3, a modification associated with active gene transcription, and the effect lasted for at least 24 h (Fig. 3, A and B). In contrast, BMP2 did not suppress histone acetylation (Fig. 3B). The suppression of histone acetylation by Wnt3a likely required Lrp5/6 signal- ing as Dkk1, which is known to interfere with the Wnt-Lrp5/6 interaction, notably diminished the effect (Fig. 3C). A survey of several histone methylation marks indicated that Wnt3a did not alter the levels of H3K9me3, H3K27me3, H3K4me3, or H3K36me3, although it increased H3K79me3, likely due to the methyltransferase activity of the ␤-catenin⅐Dot complex induced by Wnt signaling as previously reported (19,21). Thus, Wnt signaling acutely reduces histone acetylation, and this event may account for the rapid suppression of gene expression.
We next sought to determine whether the decrease in bulk histone acetylation was reflected by changes on the chromatin. For this, we performed chromatin Immunoprecipitation followed by deep sequencing (ChIP-Seq) using antibodies specific for acetylated H3K9 (H3K9ac) or tri-methylated H3K4 (H3K4me3) to localize these modifications to particular loci on the chromatin. These experiments yielded 4565 genomic loci with decreased H3K9ac compared with only 1461 with increased H3K9ac after 6 h of Wnt3a stimulation. When examined individually, each autosome contained many more loci with decreased than increased H3K9ac (Fig. 3D). In contrast, Wnt3a only minimally affected H3K4me3, with 162 genomic loci showing an increase compared with 141 loci with a decrease. More importantly, all promoter regions (defined as Ϯ3 kb flanking the transcription start site) in the genome on average exhibited a decrease in H3K9ac but a slight increase in H3K4me3 (Fig. 3E). When the promoter regions of all suppressed versus induced genes were analyzed separately, we saw that Wnt3a caused a decrease in H3K9ac in the former but an increase in H3K4me3 in the later (Fig. 3, F and G). Finally, 355 of the 4655 loci with decreased H3K9ac were located within 3 kb of a transcription start site and showed significant correlation with gene suppression (p ϭ 2.25 ϫ 10 Ϫ23 , hypergeometric test), as exemplified by Pparg and Sox5 (Fig. 3, H and I). Conversely, Wnt3a-induced genes significantly overlapped with genes exhibiting increased H3K9ac in their promoters (p ϭ 0.00067, hypergeometric test). There was no overlap between genes with decreased H3K9ac in the promoter and genes induced or vice versa. Thus, Wnt3a induces a genome-wide reduction in histone acetylation that correlates with broad gene suppression.
Wnt3a Decreases Nuclear Acetyl-CoA and Citrate Levels-To determine if decreased histone acetylation was required for gene suppression by Wnt signaling, we used trichostatin A (TSA) to inhibit both class I and II HDACs. TSA greatly increased histone acetylation in ST2 cells regardless of Wnt3a (Fig. 4A). Importantly, preventing histone deacetylation with TSA completely abolished Wnt3a-induced suppression of Pparg and Cebpa as well as osteoblast differentiation assayed by von Kossa staining (Fig. 4, B and C).
We next investigated how Wnt3a decreases histone acetylation. We found no change in either HDAC or HAT activities in assays with nuclear extracts after 6 h of Wnt3a treatment (Fig.  4D). Therefore, we hypothesized that Wnt3a suppressed histone acetylation by decreasing available acetyl-CoA, the substrate for HAT. We directly measured acetyl-CoA levels with mass spectrometry in nuclear extracts prepared from ST2 cells. These experiments revealed that Wnt3a reduced nuclear acetyl-CoA levels by 25% after 6 h of treatment (Fig. 4E). Thus, Wnt3a suppresses histone acetylation and gene expression likely by reducing nuclear acetyl-CoA levels.
Because citrate is the precursor for acetyl-CoA, we next examined the potential effect of Wnt3a on citrate levels in the nucleus. For this, we measured citrate levels in nuclear extracts prepared from ST2 cells treated with Wnt3a-conditioned media (Wnt3a-CM) or the control media. Wnt3a-CM reduced nuclear citrate levels by 24% after 6 h of treatment (Fig. 4F).
Thus, Wnt3a appears to reduce the nuclear acetyl-CoA concentration by suppressing citrate levels. Changes in Acetyl-CoA Production Affect Histone Acetylation and Gene Expression-Our findings so far predict that changing nucleocytoplasmic acetyl-CoA levels would affect Wnt3a-induced histone deacetylation and gene suppression. To test this prediction, we treated ST2 cells with supraphysiological levels of acetate, which is converted to acetyl-CoA by ACECS1 and expected to elevate nucleocytoplasmic acetyl-CoA levels ( Fig.  5A) (5). Acetate by itself slightly increased H3K9ac and completely reversed the effect of Wnt3a on histone acetylation (Fig.  5B). Importantly, acetate not only blunted the suppression of Pparg and Cebpa by Wnt3a but also impaired Wnt3a-induced osteoblast differentiation as assayed by both gene expression and von Kossa staining (Fig. 5, C-E). We next used siRNA to knock down Acly, the principle enzyme responsible for gener-ating nucleocytoplasmic acetyl-CoA from citrate in mammalian cells (Fig. 5A) (5). Acly knockdown, similar to Wnt3a, markedly decreased H3K9ac levels and suppressed Pparg and Cebpa expression, but Wnt3a and Acly knockdown did not have an additive effect on gene suppression (Fig. 5, F and G). These results further support the conclusion that Wnt3a suppresses histone acetylation by reducing acetyl-CoA levels.
Wnt3a Suppresses Glucose Entry to TCA Cycle to Reduce Histone Acetylation-Glucose oxidation in the TCA cycle is a primary source of citrate for acetyl-CoA production, and we have previously shown that Wnt signaling stimulates glucose consumption and lactate production from glucose (5,41). To examine directly the effect of Wnt on glucose oxidation through the TCA cycle, we used a CO 2 trapping method to FIGURE 5. Changes in acetyl-CoA production affect histone acetylation and gene expression. A, schematic for acetyl-CoA production from citrate or acetate. Acly converts citrate into acetyl-CoA for histone acetylation. Acetate is converted into acetyl-CoA independent of citrate. B, effects of sodium acetate on H3K9ac in ST2 cells treated with Wnt3a-or L-cell CM for 6 h. Propionic acid was used as the negative control. H3K9ac was normalized to total H3. Data are -fold change Ϯ S.D. from three independent experiments. C-E, effects of sodium acetate on Wnt3a-induced gene suppression at 6 h (C), osteoblast differentiation at 72 h (D), or matrix mineralization (E). F and G, effects of Acly knockdown (siAcly) on histone H3K9ac (F) or gene expression with or without Wnt3a treatment (G). siNTC, non-targeting control siRNA. Coomassie staining shows equal loading of acid-extracted histones in F. *, p Ͻ 0.05, n ϭ 3. Error bars: S.D. capture radioactive 14 CO 2 produced from glucose labeled with 14 C at specific positions. In this method, [3,4-14 C 2 ]glucose produces 14 CO 2 upon entry into the TCA cycle via conversion of pyruvate into acetyl-CoA and 14 CO 2 (Fig. 6A). On the other hand, [6-14 C]glucose produces 14 CO 2 during the conversion of citrate to ␣-ketoglutarate after completion of the first round of the TCA cycle (Fig. 6B). Despite a greater amount of total glucose consumption, the cells treated with Wnt3a released significantly less 14 CO 2 from [3,4-14 C 2 ]glucose than the control (Fig.  6C). Upon normalized to glucose consumption, the percentage of glucose entering the TCA cycle was reduced by nearly 50% by Wnt3a over the control (Fig. 6D). On the other hand, the amount of glucose that completes the TCA cycle, as measured by 14 CO 2 radioactivity released from [6-14 C]glucose, was not affected by Wnt3a (Fig. 6, E and F). Thus, Wnt signaling reduces the amount of glucose entering the TCA cycle, and this may be responsible for the decrease of citrate and acetyl-CoA levels in the nucleus.
Our data predict that reducing glucose entry into the TCA cycle should mimic the effect of Wnt3a on histone acetylation. As a proof of principle, withdrawal of glucose from the culture media for 12 h markedly decreased histone acetylation in ST2 cells (Fig. 7A). We next tested whether preventing glucose metabolism into the TCA cycle was sufficient to decrease histone acetylation in the absence of Wnt signaling. Specifically, we overexpressed Pdk1, which phosphorylates and inhibits pyruvate dehydrogenase, thus preventing pyruvate oxidation and entry into the TCA cycle (Fig. 7B). Pdk1 overexpression not only reduced histone acetylation by 47% but also suppressed gene expression of Pparg and Cebpa in ST2 cells (Fig. 7, C and  D). Thus, Wnt signaling suppresses histone acetylation likely through reprogramming of glucose metabolism.

Discussion
We have provided evidence that Wnt signaling induces osteoblast differentiation in part through epigenetic suppression of gene expression. The data support the mechanism that Wnt3a acutely reprograms glucose metabolism to reduce the nuclear level of citrate and acetyl-CoA, resulting in a decrease in histone acetylation independent of ␤-catenin. The current study expands our understanding about how Wnt regulates gene expression.
The study provides a mechanistic link between reduced TCA metabolism of glucose and the early phase of osteoblast differentiation in response to Wnt. Our finding is reminiscent of a recent report that decreased glycolysis reduces acetyl-CoA and histone acetylation during early differentiation of embryonic stem cells (42). Thus, a glycolytic switch appears to trigger cellular differentiation in multiple settings.
Our study reveals that Wnt signaling induces osteoblast differentiation in part through epigenetic suppression of transcription factors specifying the alternative fate of adipocyte. In contrast, BMP2 does not utilize the same mechanism but rather induces the expression of Sp7, an osteogenic transcription factor. Thus, in ST2 cells, Wnt3a and BMP2 induce osteoblast differentiation through apparently different mechanisms. It should be noted that ␤-catenin is also necessary for Wnt3ainduced osteoblast differentiation in ST2 cells as knockdown of ␤-catenin essentially abolished the induction of Alpl. Thus, WNT signaling induces osteoblast differentiation both by gene activation via ␤-catenin and by gene suppression through acetyl-CoA-mediated epigenetic regulation.
Our findings may have broad implications for understanding epigenetic regulations of gene expression. The prevailing view holds that epigenetic changes require de novo binding of transcription factors to specific DNA sequences, which subsequently recruit chromatin-modifying enzymes. Our results, however, indicate that developmental signals can act through intermediate metabolites to alter chromatin modifications and gene expression. The acute response in the level of histone acetylation likely reflects the dynamic nature of acetylation versus deacetylation on the genomic loci already occupied by the respective enzymes. Although our in vitro assays did not detect any change in total Hdac or Hat activity in nuclear extracts in response to Wnt3a, the results do not exclude the possibility that the enzymatic activity might be altered in vivo. Indeed, a recent study has linked decreased intracellular pH with global histone deacetylation (43). Considering our previous finding that Wnt signaling stimulates lactate production from glucose, it is conceivable that a lower intracellular pH could result and also contribute to histone deacetylation (41). Although previous studies have shown that chromatin acetylation decreased with differentiation of embryonic stem cells, they did not demonstrate a causal relationship between deacetylation and differentiation (44). Interestingly, Acly knockdown induces differentiation in C2C12 myoblasts but does not affect differentiation of 3T3L1 preadipocytes even though it prevents lipid accumulation, indicating that modulation of acetyl-CoA levels likely has a context-dependent role during differentiation (5,45). Finally, the present study raises the possibility that other substrates or co-factors for chromatin-modifying enzymes may also respond to physiological signals in a biologically meaningful way to affect gene expression.
Author Contributions-C. M. K., E. E., and J. C. conducted most of theexperiments.F.-F H.andJ. T.performedmassspectrometricmeasurements of acetyl-CoA. F. L. conceived then idea for the project. C. M. K. and F. L. analyzed the data and wrote the paper.