Histone deacetylase is a direct target of valproic acid, a potent anticonvulsant, mood stabilizer, and teratogen.

Valproic acid is widely used to treat epilepsy and bipolar disorder and is also a potent teratogen, but its mechanisms of action in any of these settings are unknown. We report that valproic acid activates Wntdependent gene expression, similar to lithium, the mainstay of therapy for bipolar disorder. Valproic acid, however, acts through a distinct pathway that involves direct inhibition of histone deacetylase (IC(50) for HDAC1 = 0.4 mm). At therapeutic levels, valproic acid mimics the histone deacetylase inhibitor trichostatin A, causing hyperacetylation of histones in cultured cells. Valproic acid, like trichostatin A, also activates transcription from diverse exogenous and endogenous promoters. Furthermore, valproic acid and trichostatin A have remarkably similar teratogenic effects in vertebrate embryos, while non-teratogenic analogues of valproic acid do not inhibit histone deacetylase and do not activate transcription. Based on these observations, we propose that inhibition of histone deacetylase provides a mechanism for valproic acid-induced birth defects and could also explain the efficacy of valproic acid in the treatment of bipolar disorder.

Valproic acid (VPA) 1 is a short-chained fatty acid widely used in humans as an anticonvulsant and as a mood stabilizer (1,2). The effectiveness of VPA as an anticonvulsant was discovered serendipitously when other compounds were dissolved in VPA for administration to animals used in experimental models of epilepsy (1)(2)(3). Since then, VPA has been used to control a variety of seizures, including generalized and partial seizures (1). Several hypotheses have been put forth to explain the anticonvulsant activity of VPA, and, given the efficacy of VPA in diverse forms of epilepsy, it may act through more than one target (1). VPA increases the level of the inhibitory neurotrans-mitter ␥-aminobutyric acid (GABA), with acute administration causing a 15-45% increase in GABA in the brains of rodents (cited in Ref. 1). Because inhibition of GABAergic signaling can cause seizures and potentiation of GABA signaling can prevent seizures, this effect of VPA on GABA levels has been proposed as a mechanism for the anticonvulsant activity of VPA. However, the target(s) of VPA in this setting has not been definitively identified; VPA can stimulate GABA biosynthetic enzymes and inhibit enzymes involved in GABA degradation in vitro, but it is not clear whether these are important in vivo targets of VPA (1,2,4).
VPA is a potent teratogen in humans (5) and is widely studied as a model teratogen in rodents. Although the target of VPA in this setting is unknown, strict structural requirements have been defined for the teratogenic activity of VPA and VPA-related compounds. Thus, potently teratogenic analogues of VPA contain a tetrahedral ␣-carbon bound to a free carboxyl group, a hydrogen, and two alkyl groups (6,7). In contrast, analogues such as valpromide (VPM), in which the carboxyl group is modified to an amide, and 2-methyl-2-propylpentenoic acid (2M2P), in which a methyl group is added to the ␣-carbon, do not cause neural tube defects in mouse embryos. These analogues can still protect against chemically induced seizures in mice (6,7), suggesting that at least some of the clinically observed effects of VPA involve distinct molecular targets.
In the treatment of bipolar disorder, VPA is effective both in acute mania and as a prophylaxis for recurrent mania and depression, similar to lithium (1,2). However, as with lithium, the mechanism of VPA action in bipolar disorder remains unknown. A number of interesting mechanisms have been proposed, but in each case, the direct target of VPA has not been defined. The characteristic delay in response to lithium or VPA has led to the proposal that both drugs act through modulation of gene expression, and this is supported by data from in vitro as well as in vivo systems (8 -12). Furthermore, lithium and VPA can down-regulate expression of protein kinase C isoforms PKC␣ and PKC⑀, induce expression of the anti-apoptotic gene bcl-2, and activate AP-1-dependent transcription (through a direct effect on c-jun activity and by increasing expression of c-Jun (11,12)). Both VPA and lithium also stimulate glutamate release and inositol 1,4,5-trisphosphate accumulation in mouse cerebral cortex slices, although apparently through distinct mechanisms (14). Furthermore, both VPA and lithium have been shown to confer protection from neurotoxic agents (15)(16)(17). In each of these settings, there is a delay in the response, similar to that observed clinically; thus the direct targets of VPA in these settings have not been determined.
Several direct targets of lithium have been identified (reviewed in Ref. 18), including inositol monophosphatase (19,20), a family of related phosphomonoesterases (21), and glycogen synthase kinase-3␤ (GSK-3␤ (22)). GSK-3␤ is a negative regulator of the Wnt signaling pathway, which regulates numerous processes, including axonal remodeling, cellular proliferation, embryonic patterning, and organogenesis (23)(24)(25)(26). Because GSK-3␤ phosphorylates ␤-catenin, leading to its rapid degradation, inhibition of GSK-3␤ by either lithium or Wnt signaling leads to stabilization and accumulation of ␤-catenin protein (27,28); ␤-catenin then translocates to the nucleus where it activates transcription of Wnt-dependent genes by binding to factors of the Tcf/Lef family. Activation of Wnt signaling by lithium has been proposed to explain the similarity between lithium and Wnts in a variety of settings (22), but a role for this pathway in bipolar disorder has not been demonstrated. Although VPA does not inhibit inositol monophosphatase (29,30), it has also been reported to inhibit GSK-3␤-mediated phosphorylation of a peptide derived from the CREB protein in vitro, and exposure of SH-SY5Y cells to VPA can also cause an increase in ␤-catenin protein levels (31), raising the interesting possibility that VPA and lithium both act through inhibition of GSK-3␤. However, VPA has not yet been shown to inhibit GSK-3␤ in vivo nor to activate Wnt-dependent gene expression.
We have further investigated whether VPA activates Wnt signaling and find that VPA can indeed activate Wnt-dependent gene expression, similar to lithium, but through a distinct mechanism that involves direct activation of transcription. We show that VPA potently inhibits histone deacetylase (HDAC), a negative regulator of gene expression in multiple settings, at therapeutically relevant levels. Furthermore, the teratogenicity of VPA in vertebrate embryos is mimicked by the HDAC inhibitor trichostatin A, whereas non-teratogenic analogues of VPA do not inhibit HDAC. These findings lead us to propose that HDAC is an important target of VPA in the pathogenesis of birth defects. HDAC also offers a plausible novel target for VPA action in the treatment of bipolar disorder.

EXPERIMENTAL PROCEDURES
Plasmids-Luciferase constructs containing three wild-type (Lef-OT) or three mutated (Lef-OF) Lef binding sites were gifts from Dr. Bert Vogelstein (Johns Hopkins University, Baltimore, MD) and have been described elsewhere (32). Renilla luciferase plasmids pRL-SV40 and pRL-CMV were purchased from Promega. Plasmids encoding secreted alkaline phosphatase (pSEAP) and enhanced green fluorescence protein were purchased from CLONTECH. Human HDAC1 in pcDNA3.1-myc/His-A (33) was a gift of Dr. T. Kouzarides (The Wellcome/Cancer Research Campaign Institute, Cambridge, UK).
Cell Culture and Transfection-293T and Neuro2A cells were obtained from American Type Culture Center. 293T cells were grown in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum (FBS). Neuro2A cells were grown in Eagle's Basal Medium in 10% FBS. Stable 293T cell lines were prepared by transfecting Lef-OT or Lef-OF together with pcDNA3.1. Cells were selected and maintained in Dulbecco's modified Eagle's medium with 10% FBS and 400 mg/liter G418. For all transfections, cells were plated in 6-well dishes at a density of 2.75 ϫ 10 5 cells per well. Plasmids were transfected as follows: 1 g of Lef-OT and Lef-OF reporter plasmids; 10 ng of pRL-CMV or pRL-SV40; and 0.5 g of pSEAP. Firefly and Renilla luciferase activities for each sample were measured on a Monolight 3010 luminometer (Turner Designs) using the Dual Luciferase assay kit (Promega).
All transfections included 0.5 g of enhanced green fluorescence protein to assess transfection efficiency. In some cases (e.g. overexpression of HDAC1), pSEAP was also transfected, and SEAP activity in culture medium was used as an independent measure of transfection efficiency. Thus, 24 h after transfection, an aliquot of media was removed for SEAP assay, and VPA or LiCl was then added to the cells. After an additional 24 h, cells were harvested for luciferase assay. Valproic acid (sodium salt; Sigma Chemical Co.) and lithium chloride (Sigma) were prepared in sterile water as concentrated stocks and added to the final concentrations as indicated in the figures. Valpromide (kind gift of Katwijk Chemie B.V.) was prepared in Me 2 SO. 2-Methyl-2-propylpentenoic acid and 4-pentenoic acid were purchased from Alfa Aesar.
In Vitro GSK-3␤ and HDAC1 Assays-GSK-3␤ assay was performed as described previously (22), except that MgCl 2 was 1 mM. VPA was added at concentrations indicated in Fig. 3B. For in vitro HDAC assays, myc epitope-tagged HDAC1 was transfected into HeLa cells and immunoprecipitated (36). Immunoprecipitates were washed, resuspended in HD buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 10% glycerol), and stored as frozen aliquots. HDAC1 was then added to a tube containing 40,000 cpm 3 H-labeled acetylated histones (purified from HeLa cells) in 200 l of HD buffer Ϯ VPA, trichostatin A (Sigma), valpromide, or 2-methyl-2-propylpentenoic acid (at concentrations described in Figs. 3, 4, and 7). After rotation for 2 h at 37°C, the reaction was stopped by the addition of 50 l of stop solution (1 M HCl, 0.16 M acetic acid) and released 3 H-labeled acetic acid was extracted and analyzed by scintillation counting. To assay total nuclear HDAC activity, nuclear extracts from HeLa cells (30 g) were used as a source of HDAC activity in place of immunoprecipitated HDAC1, as described (36). HDAC assay was otherwise as described above for HDAC1.
Embryos-Xenopus eggs and embryos were maintained in 0.1ϫ MMR according to standard protocols (37). Stage 8 embryos were incubated in 0.1ϫ MMR containing valproic acid or valpromide (1.0, 2.5, or 5.0 mM) or trichostatin A (25, 50, or 100 nM) for 24 h. Embryos were then transferred to fresh 0.1ϫ MMR and cultured until tadpole stages. VPA stock (2 M) was prepared in water, whereas valpromide (2 M) and TSA (100 M) stocks were prepared in Me 2 SO. Control Me 2 SO-treated embryos developed normally.

VPA Activates Tcf/Lef-dependent Transcription and Synergizes with
Lithium-To test whether VPA can activate Wntdependent gene expression, we generated stable cell lines in human embryonic kidney cells (293T) transfected with firefly luciferase reporters containing either three wild-type Lef binding sites (Lef-OT) or three mutant Lef binding sites (Lef-OF) (32). These two stable cell lines were treated with VPA or lithium chloride (LiCl) for 24 h and then harvested to measure luciferase activity. Cells treated with lithium show a dosedependent increase in Lef-luciferase activity (over 70-fold) for the reporter containing wild-type, but not mutated, Lef sites ( Fig. 1A), as reported for transiently transfected C57MG cells (38). Similarly, VPA also induces Lef-dependent luciferase activity over 20-fold (Fig. 1B). Interestingly, the addition of both drugs to the 293T stable lines resulted in marked synergistic activation of reporter activity (Fig. 1C), with up to 315-fold activation, far exceeding additive effects. This synergy raises the possibility that lithium and VPA act through independent mechanisms in this assay. This could also explain the efficacy of combining lithium and valproate in bipolar disorder patients that are resistant to single drug therapy.
VPA Activates Transcription through Diverse Promoters-To test whether neuronal cells may respond to VPA in a similar manner, Neuro2A cells were transiently transfected with Lef-OT or Lef-OF, together with a control reporter (pRL-SV40) encoding Renilla luciferase driven by the SV-40 promoter, and firefly and Renilla luciferase activities were measured after 24 h. As in 293T cells, VPA activated OT-Lef up to 6-fold in Neuro2A cells (not shown). Surprisingly, VPA also consistently activated the control reporter up to 10-fold (Fig. 1D), with half-maximal activation at 0.8 mM VPA. Transfection efficiency, assessed by frequency of green fluorescence proteinpositive cells or by expression of secreted alkaline phosphatase (SEAP), was similar in each group prior to addition of VPA. Lithium did not stimulate this or other control reporters (not shown).
VPA can activate AP-1-dependent transcription, and an increase in the activity of the SV-40 promoter has been proposed to be due to the presence of AP-1 sites within the SV-40 promoter (39). However, VPA also induces Renilla expression driven by the cytomegalovirus (CMV) promoter (pRL-CMV), which does not contain an AP-1 site (Fig. 4B), suggesting a more general mechanism of activation. VPA has also been reported to activate the Rous sarcoma virus promoter and peroxisomal proliferator-activated receptor-␦-dependent transcription in F9 teratocarcinoma cells (6). The effect of VPA on diverse promoters suggests that VPA acts through a mechanism distinct from lithium and may involve direct activation of transcription.
VPA Increases ␤-Catenin Levels through a Novel Mechanism-Wnt signaling, or exposure to lithium, causes stabilization and accumulation of ␤-catenin protein (26). We therefore examined the effect of VPA on levels of ␤-catenin protein by Western blotting. In Neuro2A cells, both lithium (20 mM) and VPA (2 mM) caused accumulation of ␤-catenin protein, similar to published work on VPA in SY5Y cells (31). However, the rate of ␤-catenin accumulation differed with the two drugs. Lithium caused ␤-catenin accumulation within 30 min of treatment, whereas the effect of VPA was not evident until 10 h after treatment ( Fig. 2A). Although this could reflect differences in the access of VPA and lithium, VPA has been shown to cross the blood-brain barrier within 1 min after intravenous injection and similarly is rapidly taken up by cells in culture (40).
An alternative possibility to explain the delay in ␤-catenin accumulation after VPA exposure is that VPA acts by increasing the expression of ␤-catenin rather than stabilizing the protein. To distinguish between these two possibilities, Neuro2A cells were cultured in the presence of cycloheximide (CHX), an inhibitor of protein synthesis. Agents, such as lithium, that stabilize ␤-catenin should slow its degradation, but no new protein will accumulate. Agents that induce new transcription or translation of ␤-catenin should have no effect on ␤-catenin protein levels in the presence of CHX. Under these conditions, ␤-catenin is rapidly degraded and is almost undetectable after 30 min of CHX treatment (Fig. 2B, lane 2). In the presence of CHX, lithium stabilized existing ␤-catenin protein, slowing the rate of degradation so that ␤-catenin protein was readily detectable at 30 min, 5 h, and 10 h (Fig. 2B, lanes 3, 6,  and 9). Conversely, VPA treatment did not stabilize ␤-catenin; the protein was rapidly degraded within 30 min and was barely detectable at 5 or 10 h (Fig. 2B, lanes 4, 7, and 10), as in cells treated with CHX alone. These observations suggest that VPA acts at the level of transcription or translation.
Northern blot analysis confirmed that VPA increases the level of ␤-catenin mRNA in Neuro2A cells in a dose-and timedependent manner (Fig. 2C), with increased ␤-catenin mRNA detected as early as 4.5 h. These data strongly support that VPA induces ␤-catenin at the level of transcription (or message stability) rather than through post-translational regulation.
VPA can inhibit GSK-3␤-mediated phosphorylation of a CREB peptide in vitro, providing an intriguing potential mechanism for VPA action, but the effect of VPA on GSK-3␤ activity in vivo has not been studied. In vivo inhibition of GSK-3␤ can be followed by examining phosphorylation of the microtubuleassociated protein tau, which is phosphorylated by GSK-3␤ in vivo at specific sites recognized by the PHF-1 antibody (34,41). This phosphorylation is inhibited by lithium in vitro (22) and in vivo (27,28), as well as by other GSK-3␤ inhibitors (18). We therefore examined the levels of tau phosphorylation in mouse Neuro2A cells treated with VPA. VPA from 0.5 to 20 mM did not inhibit tau phosphorylation even after 24 h of exposure (Fig.  3A). Rather, VPA caused a modest increase in the level of tau protein (phosphorylated and unphosphorylated forms). The non-teratogenic analogues of VPA, 2-methyl-2-propylpentenoic acid (2M2P) and 4-pentenoic acid (both at 2 mM) had no effect on levels of tau protein or tau phosphorylation. Furthermore, lithium inhibited tau phosphorylation in the presence of VPA (Fig. 3A), indicating that the tau phosphorylation under these conditions depends on GSK-3␤ activity. VPA also did not inhibit tau phosphorylation in Xenopus oocytes (data not shown).
Because VPA can inhibit phosphorylation of a CREB-derived peptide in vitro (31), we examined whether other GSK-3␤ substrates are also sensitive to VPA. VPA (0.125 to 10.0 mM) did not inhibit GSK-3␤-dependent phosphorylation of GS-2, a peptide derived from glycogen synthase (Fig. 3B). The assay was performed at 1 mM MgCl 2 , but was repeated over a wide range of magnesium concentrations (0.15-10 mM) with similar results. Inhibition of CREB phosphorylation by VPA may reflect substrate-specific inhibition of GSK-3␤, as reported for other GSK-3␤ inhibitors (42). Taken together with the above data, these in vitro and in vivo results suggest that VPA does not act through the same mechanism as lithium in these settings.
VPA Inhibits Histone Deacetylase-The ability of VPA to activate transcription regulated by multiple promoters is reminiscent of molecules that inhibit histone deacetylases (HDACs). HDACs are recruited by a variety of transcription factor corepressor complexes and are believed to repress transcription by reducing the level of acetylation of core histones, thereby altering chromatin structure (43). We therefore assayed HDAC1 activity in the presence of VPA. HDAC1 was overexpressed in HeLa cells and immunoprecipitated, and the release of labeled acetyl groups from 3 H-labeled, acetylated histones was measured (36). VPA inhibits HDAC1 in vitro in a dose-dependent manner, with an IC 50 of 0.4 mM (Fig. 4A), which is within the therapeutic range for VPA therapy in humans. To test whether VPA inhibits HDACs other than HDAC1, we prepared nuclear extracts from HeLa cells, which express multiple HDACs, including HDAC1, 2, 3, 4, and 8 (36,44,45), 2 and assayed HDAC activity as above in the presence of VPA (Fig.  4B). VPA inhibited nuclear HDAC activity, similar to inhibition of isolated HDAC1, with 50% inhibition between 0.5 and 2 mM.
These data show that VPA inhibits HDAC activity in vitro. If VPA also inhibits HDAC in cells, it should cause hyperacetylation of endogenous targets of HDACs, as seen with other HDAC inhibitors. Thus, Neuro2A cells were treated with VPA (0.5-5 mM) or with TSA (300 nM) for 24 h and then histones were isolated (Fig. 5A, lower panel). Histone acetylation was assessed by immunoblotting with an antibody specific to acetylated histone H4. Fig. 5 (upper panel) shows VPA-induced hyperacetylation of H4 is detectable at as low as 0.5 mM VPA. p53 has also been shown to be acetylated and inhibition of HDACs by TSA increases endogenous p53 acetylation (43,46); VPA also caused hyperacetylation of p53 at concentrations as low as 1-2 mM. 3 Because HDACs are important negative regulators of transcription, the increased levels of reporter activity described above may thus be due to inhibition of HDACs. Therefore, we tested whether overexpression of HDAC1 in tissue culture cells could reverse the VPA-induced activation of the Renilla reporter. 293T cells were transfected with CMV-Renilla, with or without HDAC1, and treated with VPA for 24 h. Overexpression of HDAC inhibited CMV-Renilla reporter activity 10-fold showing that this reporter can be regulated by HDAC (Fig. 6). Furthermore, HDAC prevented the transcriptional activation by VPA, suppressing Renilla activity 10-fold in the presence of VPA. These data are consistent with the proposal that HDAC is an in vivo target of VPA.
The HDAC Inhibitor Trichostatin A Mimics VPA Effects on Embryogenesis-VPA is highly teratogenic in humans, causing spina bifida aperta and other neural tube closure defects in 1-2% of offspring of women taking VPA during the first trimester of pregnancy (5). Similarly, VPA causes neural tube defects in rodents, and this has served as a commonly studied model of teratogenesis. However, the molecular target of VPA in vertebrate embryos is still unknown. Because HDAC is inhibited by therapeutically relevant concentrations of VPA, and plays important roles in embryonic development (47)(48)(49), it could be the target of VPA-induced teratogenesis. We have therefore investigated whether the structurally related VPA analogues, valpromide (VPM) and 2M2P, which function as anticonvulsants but are not teratogenic in mice (6,7), are able to inhibit HDAC. Under conditions where VPA and trichostatin A (TSA, a well characterized HDAC inhibitor) potently inhibited HDAC, neither VPM nor 2M2P inhibited HDAC in doses ranging from 0.1 to 5 mM ( Fig. 7A and data not shown). Similarly, TSA potently activated CMV and SV40 reporters, whereas VPM did not (data not shown). These observations are consistent with the failure of other non-teratogenic VPA analogues to activate Rous sarcoma virus reporter genes (6).
Because non-teratogenic VPA analogues do not inhibit HDAC, we then asked whether an established HDAC inhibitor causes defects in embryonic development similar to VPA. Exposure of Xenopus embryos to VPA (5 mM) after the midblastula transition (MBT) had a pronounced effect on development, with marked loss of anterior structures as well as shortening of the anterior-posterior axis in 88% of embryos (n ϭ 82; Fig. 7C), similar to previous work in Xenopus and other amphibians (50), whereas VPM-treated (5 mM) embryos showed apparently normal anterior development (92%, n ϭ 39; Fig. 7D). TSA closely mimicked VPA in a dose-dependent manner: 77% of embryos treated with 100 nM (n ϭ 39) and 21% of embryos treated with 50 nM TSA (n ϭ 39) showed loss of anterior structures and shortening of the A-P axis (Fig. 7E). TSA at 25 nM did not appear to affect anterior development significantly (2%, n ϭ 40), similar to a previous report using 30 nM TSA (51). Similarly, TSA causes developmental defects in mouse embryos (52) that are virtually identical to VPA-induced defects (53), as described below. Exposure of Xenopus embryos to VPA prior to the MBT, when embryonic transcription is relatively silent, had no obvious effect on development (not shown). These later effects of VPA and TSA in Xenopus are reminiscent of the effects of lithium, Xwnt-8, or other activators of Wnt signaling, which also cause anterior truncation when presented after the MBT (54,55). DISCUSSION The mechanisms of action for VPA as an anticonvulsant, mood stabilizer, and teratogen have not been defined. This work shows that VPA is an effective inhibitor of histone deacetylases, with an IC 50 (0.4 mM) well within the therapeutic range of VPA (0.35-0.7 mM in serum), and that VPA, like other HDAC inhibitors, activates transcription from diverse promoters. VPA, like lithium, activates Wnt-dependent gene expression, but unlike lithium, VPA does not inhibit GSK-3␤ in vivo. Rather, we propose that VPA activates Wnt-dependent gene expression through inhibition of HDAC, which in turn leads to both increased expression of ␤-catenin and de-repression of Tcf/Lef (as well as activation of other HDAC-regulated genes). The remarkable similarities in the effects of VPA and TSA in both mouse and Xenopus embryos indicate that inhibition of HDAC may be the mechanism of VPA-induced teratogenicity.
Several hypotheses have been put forth to explain the antiepileptic activity of VPA, and, given the diverse forms of epilepsy that respond to VPA, it may act through more than one target (1). Thus, VPA increases the level of the inhibitory neurotransmitter ␥-aminobutyric acid (GABA), selectively en- hances GABA-mediated inhibition in the cerebral cortex, inhibits AMPA binding to its receptor, and antagonizes voltagedependent sodium channels (1). In vitro, VPA can stimulate glutamic acid decarboxylase, which is involved in GABA biosynthesis, and inhibit GABA transaminase, succinic semialdehyde dehydrogenase, and ␣-ketoglutarate dehydrogenase, enzymes involved in GABA degradation. However, whether these in vitro effects of VPA are sufficient to explain its anticonvulsant activity in vivo remains unclear (1). Because analogues of VPA that do not inhibit HDAC (Fig. 7A) can still protect against chemically induced seizures in rodents (6, 7), inhibition of HDAC may not explain the anticonvulsant activity of VPA, at least in this experimental setting.
Strict structural requirements have been defined for the teratogenic activity of VPA and VPA-related compounds, and these features also appear to be important for inhibition of HDAC (Fig. 7A). Potently teratogenic analogues of VPA contain a tetrahedral ␣-carbon connected to a free carboxyl group, a hydrogen, and two alkyl groups (6,7), features that are also found in the structure of butyrate, a well known inhibitor of HDAC. Thus the non-teratogenic analogues valpromide (VPM), in which the carboxyl group is modified to an amide, and 2-methyl-2-propylpentenoic acid (2M2P), in which a methyl group is added to the ␣-carbon, do not cause neural tube defects in mouse embryos, and neither of these compounds inhibit HDAC (Fig. 5A). In contrast, the HDAC inhibitor TSA causes severe morphological defects in mouse embryos cultured from day 7.5 to 9.5 (52) that are nearly identical to those caused by VPA (53), including anterior neural tube defects, shortening of the anterior-posterior axis, growth retardation, and failure of the embryo to rotate properly. Furthermore, as presented here, Xenopus embryos treated with TSA show marked defects in anterior development, similar to VPA-treated embryos. These data in mice and Xenopus support our proposal that the teratogenic effects of VPA are mediated through HDAC inhibition.
A role for HDAC during embryogenesis has been demonstrated in invertebrates. In Drosophila, strong hypomorphic mutations in Rpd3, a gene homologous to HDAC, cause segmentation defects and embryonic lethality. These segmentation defects appear to arise from altered Even-skipped (eve) repressor activity, resulting in an indirect loss of engrailed expression (47). Additionally, loss of hda-1 in Caenorhabditis elegans causes embryonic lethality at the one-fold stage and rescues endoderm formation in embryos lacking CBP, a histone acetyl transferase (48). The hda-1 gene product is also a component of a complex involved in regulation of vulva development (49). Thus, HDACs play an important role during the embryogenesis of numerous organisms, and interfering with the function of these proteins pharmacologically may mimic the effects of loss-of-function mutations in the developing embryo.
Because VPA and lithium are both used to treat bipolar disorder, it is instructive to consider the similarities and differences between these two drugs (also see Ref. 39). In addition to treating bipolar disorder, VPA and lithium can activate AP-1, induce expression of bcl-2, down-regulate myristoylated alanine-rich C kinase substrate (39), stimulate expression of genes involved in inositol biosynthesis in yeast (56), and inhibit anterior development in Xenopus tadpoles (50,54,55; and this work). Furthermore, both lithium and VPA activate Wnt/Lefdependent gene expression; however, we propose that this effect occurs through inhibition of distinct molecular targets. Although lithium causes stabilization of ␤-catenin protein by inhibiting GSK-3␤, the effects of VPA on ␤-catenin protein levels and Lef reporter activation are likely 2-fold: 1) VPA stimulates transcription of ␤-catenin, suggesting that HDAC inhibits ␤-catenin transcription and 2) VPA stimulates (or de-represses) Tcf/Lef, which has been shown to be repressed by HDAC (57). In support of this, TSA and butyrate have been shown to stimulate Tcf/Lef-dependent transcription in colon carcinoma cell lines through a mechanism that appears to be independent of upstream Wnt signaling (58). Therefore, activation of Tcf/Lef-dependent gene expression could represent a common pathway to explain the therapeutic action of both lithium and VPA in the treatment of bipolar disorder, and this remains an open and intriguing question for future studies.
Histone acetylation has been shown to be an important regulatory mechanism for controlling transcription in ϳ2% of transcribed genes (59). However, a number of non-histone proteins, particularly nuclear proteins, have also been shown to be regulated by acetylation (for review, see Ref. 43)). These factors can be acetylated through the intrinsic acetyltransferase activity of common transcription cofactors such as CBP/p300. In principle, VPA could also inhibit the deacetylation of these non-histone proteins. In support of this, we found that VPA causes hyperacetylation of p53, as well as histone H4. 3 Interestingly, clinical use of VPA in HIV-infected patients causes an increase in the replication of HIV (60). In addition, the HDAC inhibitors TSA and trapoxin induce the transcription of HIV in vivo and in vitro (61). The identification of VPA as an HDAC inhibitor thus offers a plausible explanation for the effect of VPA on HIV levels observed clinically.
Finally, the identification of VPA as an HDAC inhibitor suggests a potential therapeutic role in the treatment of malignant diseases (62). HDAC inhibitors can prevent proliferation and induce differentiation of numerous transformed cell types, including neuroblastoma, erythroleukemia, acute myelogenous leukemia, and carcinomas of the skin, breast, prostate, bladder, lung, colon, and cervix (62)(63)(64)(65)(66)(67). Given the extensive clinical experience with VPA, it may provide a relatively safe, well tested alternative to the use of TSA and trapoxin in the therapy of malignant diseases. Indeed, VPA has been shown to inhibit proliferation and induce differentiation of cell lines derived from human malignant gliomas (13), and it may well find broader clinical use in the treatment of other types of cancer. Whether this occurs through hyperacetylation, and consequent activation of p53, or through regulation of other HDAC targets, is a subject for future study.