Stimulation of Histone Deacetylase Activity by Metabolites of Intermediary Metabolism*

Background: Little is known about potential regulation of non-sirtuin HDACs by cellular metabolites. Results: HDAC activity is stimulated by conjugated CoA derivatives and NADPH and is inhibited by free CoA. Conclusion: Cellular metabolites required for anabolism directly stimulate HDAC activity. Significance: Cellular HDAC activity may be modulated in response to the metabolic state of a cell. Histone deacetylases (HDACs) function in a wide range of molecular processes, including gene expression, and are of significant interest as therapeutic targets. Although their native complexes, subcellular localization, and recruitment mechanisms to chromatin have been extensively studied, much less is known about whether the enzymatic activity of non-sirtuin HDACs can be regulated by natural metabolites. Here, we show that several coenzyme A (CoA) derivatives, such as acetyl-CoA, butyryl-CoA, HMG-CoA, and malonyl-CoA, as well as NADPH but not NADP+, NADH, or NAD+, act as allosteric activators of recombinant HDAC1 and HDAC2 in vitro following a mixed activation kinetic. In contrast, free CoA, like unconjugated butyrate, inhibits HDAC activity in vitro. Analysis of a large number of engineered HDAC1 mutants suggests that the HDAC activity can potentially be decoupled from “activatability” by the CoA derivatives. In vivo, pharmacological inhibition of glucose-6-phosphate dehydrogenase (G6PD) to decrease NADPH levels led to significant increases in global levels of histone H3 and H4 acetylation. The similarity in structures of the identified metabolites and the exquisite selectivity of NADPH over NADP+, NADH, and NAD+ as an HDAC activator reveal a previously unrecognized biochemical feature of the HDAC proteins with important consequences for regulation of histone acetylation as well as the development of more specific and potent HDAC inhibitors.

HDACs 4 catalyze the removal of acetyl groups from lysine residues of a wide array of substrate proteins in addition to histones, including transcription factors and other nuclear and cytoplasmic proteins. Based on phylogenetic sequence analysis, HDACs are classified into class I (HDAC1-3 and -8), class II (HDAC4 -7 and -9), class III, also known as sirtuins (SIRT1-7), and class IV (HDAC11) (1,2). HDACs are ubiquitously expressed and play roles in regulation of cell growth, differentiation, and death (1,3). They show deregulated expression in pathological states such as cancer and have therefore garnered significant interest as therapeutic targets (4). There are currently more than 80 clinical trials underway to determine the therapeutic efficacies of several HDAC inhibitors that have shown anti-neoplastic functions in cell culture and animal studies. Two HDAC inhibitors, Vorinostat and Romidepsin, have been approved for use in treatment of cutaneous T-cell lymphoma (5,6). Other HDAC inhibitors have been used successfully as mood stabilizers and anti-epileptics and have shown promise in treatment of diverse neurological diseases such as Huntington and Alzheimer disease, amyotrophic lateral sclerosis, and spinal muscular atrophy (7,8). Therefore, it is critical to fully understand the molecular mechanisms that regulate the activities of HDACs for the development of more effective HDAC inhibitors and more informed use of the current drugs in the clinic.
HDACs are generally found in multiprotein complexes, the functions of which can be regulated through several mechanisms. These complexes can be recruited to specific genomic loci by DNA-binding proteins, thereby bringing HDACs to close proximity of their histone substrates (9 -11). HDAC-containing complexes are localized in both the nucleus and cytoplasm but can shuttle between the two subcellular compartments, affecting their accessibility to relevant substrates (12,13). Modulating the stability of interactions between HDACs and other members of the complexes in which they reside can also affect the overall activity of HDAC complexes (14,15).
Other regulatory mechanisms may affect the HDAC enzymatic activity more directly. Acetylation of HDAC1 protein reduces its enzymatic activity both in vivo and in vitro (16). Sumoylation of HDAC1 increases its enzymatic activity, and phosphorylation of HDAC1 stimulates both its activity and complex formation (17,18). The catalytic activity of class III HDACs (Sirtuins) depends on the presence of the oxidized form of nicotinamide adenosine dinucleotide (NAD ϩ ) (19 -21). Because the availability of NAD ϩ is linked to cellular metabolism, the metabolic state of the cell could be a direct regulator of sirtuins (22). Such direct regulation by a metabolic cofactor may transmit information on the cellular energy state to the chromosome, influencing nuclear functions such as gene expression and DNA replication.
However, little is known about direct regulation of non-sirtuin HDAC activity by metabolic intermediates. A first example of this kind of regulation is the binding and inhibition of HDAC1 and -2 by the endogenous lipid mediator sphingosine 1-phosphate (23). Because HDACs release free acetate anions from chromatin, we asked whether intermediates of nitrogen and carbon metabolism that generate or consume two-carbon units as acetate directly regulate HDAC activity.
Here, we show that coenzyme A (CoA) derivatives, such as acetyl-CoA, butyryl-CoA, and malonyl-CoA, as well as NADPH stimulate the activity of class I HDACs on histones, whereas free CoA inhibits HDAC activity in vitro. Detailed kinetic analysis (24) showed that CoA derivatives and NADPH cause mixed activation of recombinant HDAC2, affecting both the affinity of the enzyme for the substrate as well as its catalytic activity. Inhibition by free CoA is substrate-independent and is characterized by decreased reaction velocity while at the same time increasing the affinity of the enzyme for histones. We further show that HDAC activation by NADPH may be relevant in vivo as inhibition of NADPH production increases global histone acetylation. Many of the identified metabolites are products of catabolic pathways of glucose and amino acids that together with NADPH, a source of reducing power, are required for anabolic reactions that fuel cell growth and replication. Our data therefore suggest that cellular HDAC activity may be tightly linked to cellular biosynthetic capacity. Furthermore, identification of natural activators and inhibitors of HDACs that contain a nucleotide-like moiety may enhance our understanding of structure-function relationship and better inform the design and development of HDAC inhibitors.

EXPERIMENTAL PROCEDURES
Experiments in Figs. 1-3 were performed with recombinant HDACs obtained from US Biological. All metabolic compounds were purchased from Sigma as sodium or lithium salts.
Generation of Recombinant WT HDAC1 and HDAC1 Mutants-An EcoRI-HDAC1-His 6 -SalI fragment was generated by PCR using a commercially available cDNA for HDAC1 (OriGene). Recombinant HDAC baculovirus was generated using the Bac-to-Bac expression system (Invitrogen), following the manufacturer's instructions. HDAC1 mutants were created by site-directed mutagenesis on the pFastbac-HDAC1 plasmid using QuikChange Site-directed mutagenesis kit (Agilent).
Purification of Recombinant HDAC Protein-Sf9 cell pellets were washed with ice-cold PBS and resuspended in hypotonic buffer (20 mM HEPES-KOH, pH 8.0, 10 mM NaCl, 10% glycerol, 2.5 mM ␤-mercaptoethanol, 1ϫ protease inhibitor mixture (Roche Applied Science)). The cells were incubated on ice for 30 min and then disrupted using a Dounce homogenizer. Nuclei were spun down at 7000 ϫ g for 10 min, and the cytoplasmic supernatant (containing the majority of the recombinant HDAC protein) was transferred to a fresh tube and re-centrifuged at 22,000 ϫ g for 30 min to remove debris. The recombinant protein was purified using the TALON Metal Affinity Resin (Clontech) following the manufacturer's instructions and dialyzed in HDAC storage buffer (20 mM HEPES, 150 mM KCl, 30% glycerol, and 2.5 mM ␤-mercaptoethanol).
Isolation of 3 H-Labeled Histones from HeLa Cells-HeLa cells were grown exponentially in suspension to a density of 6 ϫ 10 5 cells/ml. Histones were labeled and extracted as described previously (25). Concentration of histones was determined by BCA protein assay (Pierce), and their molarity was calculated using the average molecular weight of histones H2A, H2B, H3, and H4.
HDAC Assay-Recombinant HDACs or nuclear extracts were incubated with 3 H-labeled histones and metabolic intermediates (quantities as indicated in figures) for 30 min at 37°C in Buffer D (13.3 mM HEPES-KOH, pH 7.9, 7 mM MgCl 2 , 13.3% glycerol, 66.7 mM KCl, and 66.7 M EDTA). The reaction was stopped with 0.2 volume of 5ϫ Stop Solution (0.6 M acetic acid, 3.6 M HCl), and the reaction product ( 3 H-labeled acetate) was extracted with 2 volumes of ethyl acetate. Extracted 3 H-labeled acetate was quantified using a liquid scintillation counter (TriCarb2800TR, PerkinElmer Life Sciences). HDAC assays with fluorescent substrate were performed using the Fluor de Lys TM -Green HDAC assay kit (Enzo Life Sciences).
Histone Extraction for Western-Cell cultures were grown in DMEM plus serum to 80% confluency, and 5 ϫ 10 6 cells were washed twice with ice-cold PBS and resuspended in 750 l of Buffer A-H (10 mM HEPES-KOH, pH 7.9, 1.5 mM MgCl 2 , 10 mM KCl, 0.34 M sucrose, 10% glycerol, and freshly added 1 mM DTT and 1ϫ protease inhibitors (Roche Applied Science)). After addition of 0.1% Triton, cells were lysed for 8 min on ice. Nuclei were pelleted at 1500 rcf for 5 min at 4°C and washed with 1 ml of Buffer A-H for 1 min at 4°C. Nuclei were pelleted at 1500 rcf for 2.5 min at 4°C and resuspended in 5 pellet volumes of extraction buffer (432 mN H 2 SO 4 , 28 mM ␤-mercaptoethanol, 10% glycerol). Histones were extracted for 15 min on ice. After centrifugation at 16,000 rcf for 10 min, histones were precipitated from the supernatant by addition of 20% TCA followed by centrifugation at 16,000 rcf for 10 min at 4°C. The pellet was washed once with 1 ml of 70% EtOH, dried, and resuspended in water.
Immunoprecipitation-Cells were grown in DMEM to ϳ90% confluency, washed twice with ice-cold PBS, and lysed in 100 l of lysis buffer (20 mM HEPES-KOH, pH 7.9, 100 mM KCl, 1 mM EDTA, 0.5% Nonidet P-40, 1ϫ protease inhibitor mixture (Roche Applied Science) and 1 mM DTT) per 10 6 cells for 30 min on ice and centrifuged at 16,000 rcf for 10 min at 4°C. 1 ml of clarified whole cell extract was incubated with 2 g of ␣HDAC1, ␣HDAC2 (Bethyl Laboratories A300-713A and A300-705A, respectively), or rabbit IgG (Millipore) for 1 h at room temperature. 70 l of Dynabeads protein A (Invitrogen) were washed with lysis buffer and used for precipitation for 1 h at 4°C. Immunoprecipitated protein was washed three times for 10 min at room temperature with 1 ml of lysis buffer and one time with HDAC assay buffer, and HDAC activity was assayed.
Cell Fractionation-Cell fractionation was performed as described previously (26).
Data Analysis-EC 50 values for Fig. 2, A and B, were determined with SigmaPlot, using the nonlinear regression analysis for a four-parameter logistic curve. Curve fitting and determination of K m , V max , and Hill values for Fig. 2, C-J, were performed with the enzyme kinetics module using the "Enzyme Activator-Nonessential" option and Padgraph Prism (Sigma-Plot). Parameters for Fig. 3A were determined by fitting experimental data to a function describing a double-sigmoid curve (Equation 1), as described previously (27), Using the curve-fitting tool of MATLAB and assuming that F min ϭ 0, the data best fitted Equation 2, where x indicates log10-transformed CoA concentrations; y indicates fold activation, and the denominators of the fractions in parentheses are the log10-transformed EC 50 values of F1 and F2. Curve fitting and determination of K m , V max , and Hill values for Fig. 3, B-F, were performed with the enzyme kinetics module using the "Single Substrate-Single Inhibitor" function and Padgraph Prism (SigmaPlot), and the curves with the highest R 2 were chosen as best fit.

RESULTS
In Vitro HDAC Activity Is Regulated Directly by Specific Cellular Metabolites-Because histone acetylation requires acetyl-CoA and deacetylation releases acetate anions, we asked if metabolites of pathways that generate or consume two-carbon units directly regulate the activities of class I HDACs. To assay HDAC activity, 25-50 nM recombinant HDAC1 or HDAC2 were used in absence or presence of 1 mM of each metabolite shown in Fig. 1A using 50 M 3 H-labeled HeLa histones as substrate. As expected, addition of 1 mM sodium butyrate, a known inhibitor of HDACs, inhibited the activities of HDAC1 and HDAC2 (Fig. 1A). We found that the in vitro activities of HDAC1 and HDAC2 increased by ϳ1.5-fold with addition of acetyl-, acetoacetyl-, and succinyl-CoA, by ϳ2-fold with butyryl-, isobutyryl-, glutaryl-, HMG-, and malonyl-CoA, and by 2-3-fold with methylmalonyl-, crotonyl-, and ␤-methylcrotonyl-CoA relative to no compound control (Fig. 1A). However, the long chain fatty acid derivative palmitoyl-CoA inhibited HDAC1 and HDAC2 activities (Fig. 1A). All activating metabolites shared the presence of the CoA module. However, free CoA reduced the activities of HDAC1 and HDAC2 by 63 and 52%, respectively. Succinate had no effect on the in vitro HDAC activity. Interestingly, although sodium butyrate and free CoA inhibited HDAC1 and more so HDAC2 activities, butyryl-CoA increased the activities of the same HDACs by 2-fold (Fig. 1A).
The chemical structure of coenzyme A includes an adenosine nucleotide phosphorylated on the ribose moiety. Considering that a phosphorylated ribose module is present also in the adenosine nucleotide of NADP(H), we tested if NADP(H) could similarly affect HDAC activity. We found that 1 mM NADPH, but not NADP ϩ , stimulated class I HDAC activity ϳ2-fold (Fig.  1B). Neither NAD ϩ nor NADH showed any significant effect when assayed at the same concentrations ( Fig. 1B), indicating that the phosphorylated ribose moiety may be required for the NADPH ability to activate HDACs. Taken together with the above, we conclude that a variety of natural CoA derivatives that function as biosynthetic metabolites as well as NADPH, the major reducing power for biosynthesis, are able to stimulate the in vitro activities of recombinant HDACs.
CoA Derivatives and NADPH Enhance HDAC Activity in Nuclear Extracts-To test if the CoA derivatives, NADPH and free CoA, affect HDAC activity in their native complexes, we determined the effects of the metabolites on HDAC activity in nuclear extracts from four different cancer cell lines. As shown in Fig. 1C, HDAC activities of nuclear extracts from two breast cancer (MCF7 and MDA-MB-231) and two prostate cancer (LNCaP and PC3) cell lines were increased by ϳ1.5-2.2-fold upon addition of 500 M of the indicated CoA derivatives or NADPH. Addition of the same amount of free CoA slightly reduced HDAC activity of the nuclear extracts. We therefore conclude that the CoA derivatives or NADPH enhance activities of HDACs in their natural molecular environment.
CoA Derivatives and NADPH Enhance the Activity of HDAC1-and HDAC2-containing Complexes-Next, we immunoprecipitated HDAC1 or HDAC2 from whole cell lysates of different cancer cell lines and assayed their response to the CoA compounds. Addition of crotonyl-CoA increased activities of HDAC1 and -2 complexes immunoprecipitated from the MDA-MB-231 cells by up to 2.5-fold, whereas free CoA caused significant inhibition (Fig. 1D). Addition of NADPH also increased HDAC activity of HDAC1 and -2 complexes from three different cell lines from 1.6-to 2.6-fold (Fig. 1E). Altogether, our data indicate that CoA derivatives and NADPH can stimulate class I HDAC activity in vitro and in vivo.
Stimulation of HDAC Activity by Metabolites Depends on Histone Substrates-Because allosteric activation could depend on the type of substrate used (28), we wished to determine whether activation of HDAC1 and HDAC2 by CoA derivatives or NADPH depends on histones as substrate. Thus, we used Fluor de Lys instead of histones as substrate in HDAC activity assays. Fluor de Lys is a small synthetic substrate for histone deacetylases, which is converted to a fluorophore upon deacetylation. Addition of 500 or 1000 M malonyl-CoA, HMG-CoA, or NADPH had essentially no effect on HDAC1 and HDAC2 activities when Fluor de Lys was used as substrate (Fig. 1F). However, free CoA and trichostatin A inhibited deacetylation of Fluor de Lys by these deacetylases (Fig. 1F). These data indicate that activation of HDAC1 or HDAC2 by CoA derivatives or NADPH is substrate-dependent, although their inhibition by free CoA is not.
Half-maximal Effective Concentration (EC 50 ) of Activators Is the Micromolar Range-To determine at which range of concentration NADPH and acetyl-CoA function as metabolic activators of HDACs, we assayed the extent of deacetylation of 50 M histones by 25 nM HDAC2 in the presence of increasing concentrations of each activator, ranging from 50 to 2500 M. The response curve for NADPH showed increased activation up to 2500 M, yielding an ϳ2.3-fold activation with an EC 50 ϭ 541.9 M (Fig. 2A). Similar to NADPH, acetyl-CoA increased activation up to ϳ2-fold at 2500 M, yielding an EC 50 ϭ 602.9 M (Fig. 2B). These data indicate that NADPH is a somewhat stronger activator of HDAC2 compared with acetyl-CoA in vitro.
Activation by NADPH and Acetyl-CoA Is Due to V-type Activation Counteracted by K-type Inhibition-To understand the nature of HDAC2 activation by NADPH and acetyl-CoA, we used our in vitro assay to analyze the enzymatic kinetics in the presence or absence of these compounds. 25 nM of recombinant HDAC2 was incubated with increasing concentrations of labeled histones, ranging from 0.5 to 50 M, and in the presence of 0, 250, 500, 1000, and 2000 M of each activator. The resulting data were fitted using the SigmaPlot enzymatic module (nonessential activation) and expressed as micromolars of histones deacetylated per min (see under "Experimental Procedures"). Fig. 2, C and D, shows the Michaelis-Menten graphs for HDAC2 activation by NADPH and acetyl-CoA, respectively.  (Fig. 2D). The slight difference of V max and K m values in these two experiments is likely due to different batches of commercially acquired HDAC2 protein. Activation by acetyl-CoA is slightly lower than that by NADPH, having a K a(AcCoA) of 251.75 M.
To determine how V max and K m change at each concentration of activators, we determined the parameters W and Q. W is the ratio of V max in the presence of saturating effector to V max in the absence of effector (V max(inf) /V max(0) ). Q is the opposite ratio of the corresponding K m values (K m(0) /K m(inf) ) ( Table 1) (24). The product WQ is Ͼ1 for enzyme activation, which is the case for both NADPH and acetyl-CoA (Table 1). In the case of NADPH, W Ͼ1 but Q Ͻ1, indicating that the enzyme velocity is increased, but affinity for substrate is decreased (Fig. 2E). In the case of acetyl-CoA, the curves indicate a more complex pattern. Both V max and K m values increase with increasing effector concentrations up to 500 M, when a plateau is reached for V max but K m decreases. This indicates that up to 500 M of the effec-tor increases the reaction velocity more than it causes a decrease in affinity for the substrate, after which further activation is mainly achieved by regaining affinity for the substrate (Fig. 2F). Interestingly, at very low substrate concentrations of 0.5 M histones, the presence of NADPH slightly inhibits HDAC activity, indicating that under these conditions K-type  inhibition may prevail (Fig. 2G). We did not observe inhibition by acetyl-CoA at the lowest substrate concentrations tested (Fig. 2H). Activation by NADPH or Acetyl-CoA Does Not Affect Cooperativity-To determine whether NADPH or acetyl-CoA enhances enzyme activity by causing cooperative binding of the HDACs to histones, we used the kinetic data from Fig. 2, C and D, to calculate the Hill values (29) using Graphpad Prism. As shown in Fig. 2, I and J, and Table 2, Hill values at all effector concentrations are close to 1, indicating that enzyme-substrate binding is not cooperative at any effector concentration.
Kinetics of Inhibition of HDAC Activity by Free CoA-We next investigated the nature of HDAC2 inhibition by free CoA in a similar manner as for the activators above. Using the same curve fitting function of SigmaPlot as for Fig. 2, A and B, did not yield any statistically significant result. We noticed that the experimental data were best fitted with a double sigmoidal curve. The resulting curve is the sum of two sigmoid functions, where the V max of function 1 (F1) equals V min of function 2 (F2) (Fig. 3A and under "Experimental Procedures") (R 2 ϭ 0.9992). At low concentrations of CoA (Ͻ200 M), the data are best described with a curve determined by F2 with an IC 50(2) of 24.17 M. Concentrations between 100 and 200 M CoA correspond to a plateau of the double sigmoidal curve. Concentrations Ͼ200 M yielded a second inhibitory response, best fitted with F1 with an IC 50(1) of 1526.5 M. Although the latter concentration is unlikely to be reached within the cellular environment, the IC 50 (2) is well in the range of well known HDAC inhibitors such as sodium butyrate, valproate, or phenyl butyrate with IC 50 values ranging from 27 to 106 M for HDAC1 or HDAC2 (30).
This dual nature of the dose response may reflect different modes of inhibition at different concentrations of the inhibitor. We therefore explored a wider range of CoA concentrations  M as compared with 250 -2000 M in Fig. 2, C and  D) to generate the Michaelis-Menten plots. An attempt to fit all the data into a single graph did not result in statistically significant modeling. However, when data up to 75 M CoA were separated from those above 500 M, curves were fitted using the Michaelis-Menten equation (R 2 Ͼ95%) and kinetic parameters could be determined (Fig. 3, B and C, and Table 1).
Plotting logV max and logK m values against increasing amounts of free CoA shows that both V max and K m values diminish, giving a WQ value of 0.846 (Fig. 3D and Table 1). Inhibition is therefore a result of V-type inhibition (W ϭ 0.093; Table 1) countered by Q-type activation (Q ϭ 9.07, Table 1), indicating strong reduction of reaction velocity, yet increasing the affinity of the enzyme for the substrate. Interestingly, at lower CoA concentrations, K m values diminish more rapidly than V max ; this relation inverts at higher concentrations of free CoA (Ͼ500 M), which possibly reflects the reason for the double sigmoidal shape of the curve in Fig. 3A. We did not observe changes for binding cooperativity as Hill values were found to be close to 1 (Fig. 3, E and F, and Table 2).
Structural Basis of HDAC Activation by Metabolites-The structures of NADPH and CoA contain a phosphorylated adenine nucleotide. Thus, we searched for possible structural motifs in either HDAC1 or HDAC2 that could accommodate a phosphorylated adenine nucleotide. HDAC1 and HDAC2 are almost entirely made up of an HDAC domain flanked by relatively short N-and C-terminal unstructured regions. HDAC domains consist of an array of ␤-sheets sandwiched between two rows of ␣-helices that resemble a Rossmann-fold (31). This structural arrangement is typical of proteins that interact with nucleotides. We therefore searched for chemical similarities as well as sequence and structural alignments between HDAC1 and -2 and proteins that are known to bind either CoA derivatives or nicotinamide adenosine dinucleotides such as cytochrome P450 reductase or NADPH oxidases (see Table 3). We surmised that regions of similarity may include candidate residues that contribute to adenine nucleotide binding. We then mutated such residues in HDAC1 to either change their charge or bulkiness or to the corresponding sequence of class II HDACs to potentially maintain HDAC activity but affect "activatability." In total, we generated 17 mutant HDAC1 proteins with single or multiple point mutations (see Table 3 for a description of and rational for construction of the mutants).
Recombinant wild type (WT) HDAC1 and the various mutants were generated using the Sf9 insect cells (Fig. 4A) and tested for activity and activatability by either ␤-methylcrotonyl-CoA or NADPH, as these compounds had the strongest effect on the WT enzyme. All mutations reduced HDAC activity when compared with the corresponding WT with six mutants retaining 30 -80% of WT HDAC activity and 13 mutants less than 25%. Six mutants had essentially no catalytic activity and were considered catalytically dead (Fig. 4B). Interestingly, although the remaining mutants affected the overall activity, most maintained their activatability by either NADPH or ␤-methylcrotonyl-CoA (Fig. 4C). Only one mutant, G300I, showed slightly decreased activatability, albeit not reaching statistical significance relative to WT (1.8 versus 2.2-fold, p value ϭ 0.09; Fig. 4C). Although the G300I mutant has low basal HDAC activity, other mutants such as G138I showed even lower basal activity, yet retained their activatability (Fig. 4, B and C). Fig. 4D shows the crystal structure of HDAC2 (32) with the Gly-300 residue highlighted in blue. This glycine is located deep within the catalytic pocket, close to the catalytic residues highlighted in Fig. 4D in red. The G300I mutation replaces the small glycine with a bulky isoleucine, which may reduce the size of the catalytic pocket. This position is unlikely to be the site of interaction with the activator, yet G300I may hinder the effects of the activator on the enzyme-substrate interaction.
A second mutation, D181N, changed the charge on the surface of the protein and, similar to G300I, caused a drastic reduction of the basal HDAC activity. Yet, this mutation made the mutant consistently more responsive to the activator compared with WT HDAC1 (3.1-fold versus 2.2-fold, respectively, p value ϭ 0.01; Fig. 4, B and C). The corresponding residue in HDAC8, Asp-183, has been shown to take part in the catalytic charge-relay system (33). This raises the possibility that the  activators may render catalysis less dependent on the negative charge of this aspartate. Although we were unable to identify a mutation that would retain the basal activity of the enzyme but abolish its activatability, our analyses of the HDAC mutants show that the enzyme can potentially be modified in a manner that alters its response to the activators. Further analyses are required to define the region(s) of the HDAC proteins that interact with the identified activators.
Inhibition of Glucose-6-phosphate Dehydrogenase (G6PD) Increases Histone Acetylation in Vivo-Our next aim was to determine whether regulation of HDACs by metabolites is relevant in an in vivo environment. NADPH is known to be present in the nucleus, where HDAC1 and -2 are active. G6PD, the rate-limiting enzyme of the pentose phosphate pathway, is a main producer of NADPH. We surmised that because inhibition of this protein decreases cellular NADPH levels (34), this may lead to decreased HDAC activity and increased levels of histone acetylation. We chose two breast cancer cell lines, MDA-MB-231 and MDA-MB-453, that express G6PD to different extents and should therefore differ in their response to G6PD inhibition. Total cell extracts from each line were separated into cytoplasmic), nucleoplasmic, and chromatin fractions and analyzed by Western blotting (Fig. 5A). As expected, tubulin was detected exclusively in the cytoplasm, whereas an antibody specific for histone H3 lysine 18 acetylation (H3K18ac) stained chromatin only (Fig. 5A). G6PD was expressed at significantly higher levels in MDA-MB-453 cells. Interestingly, a minor fraction of G6PD was detected in both nucleoplasm and chromatin fractions of the MDA-MB-453 cells (Fig. 5A), indicating that some G6PD may be associated with chromatin in the nucleus of MDA-MB-453 cells.
We incubated both MDA-MB-231 and MDA-MB-453 cells with 5-dehydroepiandrosterone (DHEA; 100 M), a known inhibitor of G6PD (35), or solvent only (DMSO) for 12 h, extracted histones, and analyzed their acetylation levels by Western blotting. We examined three acetylation sites in each FIGURE 4. D181N and G300I mutations in HDAC1 alter its response to activating metabolites. A, Coomassie staining of recombinant HDAC1 WT and mutant proteins. B, HDAC activity of recombinant mutant proteins was determined using saturating amounts of 3 H-labeled histones. The graph represents the relative HDAC activity of mutant proteins compared with WT levels. C, recombinant WT and mutant HDAC1 proteins were incubated with saturating amounts of 3 H-labeled histones in the absence or presence of 2 mM NADPH, and HDAC activity was assessed. The graph represents fold change in HDAC activity compared with basal histone deacetylation. Error bars indicate the standard deviation of three independent experiments. A two-tailed Student's t test was used to calculate p values (**, p Ͻ 0.01; ***, p Ͻ 0.001). D, image of a fragment of the crystal structure of HDAC2 (32) in which the location of the corresponding residues of Asp-181 and Gly-300 are indicated in green and dark blue, respectively. Catalytic residues (His-145, His-146, Asp-181, and His-183) are indicated in red.
of histones H4 and H3. As shown in Fig. 5B, DHEA-treated MDA-MB-231 cells did not show increased histone acetylation compared with DMSO-treated cells. In contrast, MDA-MB-453 cells, which had high levels of G6PD, showed increased histone acetylation at all sites analyzed with the strongest effects on H4K8ac and H4K12ac. We conclude that inhibition of G6PD by DHEA causes increased global histone acetylation.

DISCUSSION
We have identified several natural metabolites that function as allosteric activators of HDAC activity in vitro. These metabolites all share an adenosine phosphate moiety and include several CoA derivatives and NADPH. Some of the CoA derivatives are intermediates of pathways that break down carbohydrates or amino acids. For instance, glutaryl-CoA, crotonyl-CoA, and HMG-CoA are generated in the pathway of lysine degradation, whereas ␤-methylcrotonyl-CoA and succinyl-CoA are degradation products of leucine and methionine, respectively. Acetyl-CoA is a central molecule in the intermediary metabolism and can be generated from degradation of glucose, pyruvate, amino acids, as well as ␤-oxidation of fatty acids. Other identified CoA derivatives are involved in anabolic reactions such as malonyl-CoA and HMG-CoA, which are the main precursor for fatty acid and cholesterol syntheses, respectively. NADPH, the main reducing agent for biosynthetic reactions, also increased HDAC activity in vitro. Interestingly, ATP was also shown to activate class I HDACs, albeit possibly indirectly through ATP-dependent molecular chaperons that co-precipitated with the HDAC complexes (15) The structure of the HDAC domain contains a Rossmann-like fold, a motif that is found commonly in proteins such as dehydrogenases that bind nucleotides. Thus, the HDAC structure may have evolved to respond to one or more metabolites with nucleotide containing moieties.
The allosteric stimulation of HDAC activity by the CoA derivatives and NADPH followed a mixed activation kinetics in which both K m and V max values were increased. Following a method of analysis described by Reinhart (24) established that activation was a product of V-type activation counteracted by Q-type inhibition. So the compounds increased the overall HDAC activity by increasing the reaction velocity despite some reduction in the affinity of the enzyme for the substrate. The reasons for the decrease in the K m value at the highest concentration of acetyl-CoA are unclear but could be due to contamination by free CoA, which lowers the K m value of HDACs for histones or nonspecific effects on HDAC-histone interactions.
The metabolites failed to stimulate HDAC activity when Fluor de Lys, a synthetic small peptidic substrate, was used instead of histones in the HDAC reaction, suggesting that the activating metabolites must affect the substrate-enzyme interaction(s) in the regions of substrate other than the acetylated residue itself. Although the level of activation was at most 2-3fold, histone acetylation is rapidly turned over in vivo, and therefore, small changes in the balance between acetylation and deacetylation activities can potentially lead to large differences in steady state histone acetylation levels.
Unlike the thioester-conjugated CoA derivatives, the unconjugated free CoA molecule inhibited HDAC activity on both histones and Fluor de Lys. The inhibition by free CoA is the V-type inhibition with reduced enzyme velocity but increased affinity of the enzyme for the substrate. Overall, the similar extent of decreased V max and K m values closely resembles an uncompetitive inhibitory kinetics. This occurs when V max and K m values decrease at the same rate. Uncompetitive inhibition is rare in enzymology and is based on the principle that the inhibitor binds preferentially to the enzyme-substrate complex and not to the free enzyme. Inhibition of G6PD by DHEA is, in fact, an example of uncompetitive inhibition (36). The rarity of uncompetitive inhibition may be due to potential accumulation of a pathway's intermediates to toxic levels, as the inhibition cannot be overcome by increased substrate concentrations. Thus, in cases in which competitive inhibition is negligible, uncompetitive inhibition can have large effects on enzymatic activity (37).
Interestingly, although free CoA and butyrate, a well known HDAC inhibitor, inhibited HDAC activity, butyryl-CoA stimulated HDAC activities of HDAC1 and -2. This finding suggests that the ability of butyrate to inhibit HDACs may stem from structural mimicry of natural HDAC activators.
Our kinetic analyses indicate that IC 50 values for either inhibitors or activators is in the micromolar range. This is comparable with the intracellular levels reported for metabolites such as free CoA or acetyl-CoA, which can reach cytoplasmic concentrations of 100 and 60 M, respectively. It is therefore possible that intracellular concentrations of these metabolites can reach local concentrations in the effective range. Such a scenario has been reported for the in vitro activity of the Sir2 histone deacetylase, which requires NAD ϩ at up to 1 mM (19).
Our limited mutational analysis of HDAC1 did not identify a mutant that retained WT levels of activity, even when the mutations were based on the corresponding residues in HDAC1 orthologues. This may not be surprising, as most of the enzyme consists of the HDAC domain itself. However, two mutations, one within the catalytic pocket and one on the surface of HDAC1, did alter its response to the activator. Additional studies, including perhaps the structure of HDAC1 with an activator, could reveal the residues that are critical for binding of activators by HDACs.
Inhibition of G6PD has been shown to decrease levels of NADPH (34). As predicted by our in vitro findings, pharmacological inhibition of G6PD increased histone acetylation in MDA-MB-453 cells, presumably by decreasing NADPH levels and thus, reducing HDAC activity. Like NADPH, many of the metabolites identified in this study are needed for syntheses of macromolecules needed for growth. Whether all or some of these compounds, or other structurally related molecules, are true activators of HDACs in vivo needs further investigation, but their presence may signal availability of biosynthetic capacity to the HDACs. The alteration of HDAC activity may in turn result in changes in gene expression patterns that promote cellular growth. Whatever the case may be, modulation of in vitro HDAC activity by anabolic metabolites with a nucleotide moiety could aid the design and development of more effective HDAC inhibitors.