Glycogen Synthase Kinase-3β Inhibits the Xenobiotic and Antioxidant Cell Response by Direct Phosphorylation and Nuclear Exclusion of the Transcription Factor Nrf2*

The transcription factor Nrf2 (nuclear factor E2-related factor 2) regulates the expression of antioxidant phase II genes and contributes to preserve redox homeostasis and cell viability in response to oxidant insults. Nrf2 should be coordinated with the canonical cell survival pathway represented by phosphatidylinositol 3-kinase (PI3K) and the Ser/Thr kinase Akt but so far the mechanistic connections remain undefined. Here we identify glycogen synthase kinase-3β (GSK-3β), which is inhibited by Akt-mediated phosphorylation, as the link between both processes. Using heme oxygenase-1 (HO-1) as a model phase II gene, we found that both PI3K and Akt increased mRNA and protein levels of this enzyme. Pharmacological inhibitors (LiCl and PDZD-8) and genetic variants of GSK-3β (constitutively active and dominant negative mutants) indicated that PI3K/Akt activates and GSK-3β inhibits the antioxidant response elements of the ho1 promoter and pointed Nrf2 as directly involved in this process. Indeed, GSK-3β phosphorylated Nrf2 in vitro and in vivo. Immunocytochemistry and subcellular fractionation analyses demonstrated that the effect of GSK-3β-mediated phosphorylation of Nrf2 is to exclude this transcription factor from the nucleus. Nrf2 up-regulated the expression of HO-1, glutathione peroxidase, glutathione S-transferase A1, NAD(P)H: quinone oxidoreductase and glutamate-cysteine ligase and protected against hydrogen peroxide-induced glutathione depletion and cell death, whereas co-expression of active GSK-3β attenuated both phase II gene expression and oxidant protection. These results contribute to clarify the cross-talk between the survival signal elicited by PI3K/Akt and the antioxidant phase II cell response, and introduce GSK-3β as the key mediator of this regulation mechanism.

in the oxidation of cellular macromolecules and leads to cell death. The detrimental accumulation of ROS plays an important role in multiple pathologies, including neurodegenerative disorders, cancer, atherosclerosis, diabetes, and aging (1)(2)(3)(4). To maintain redox homeostasis, aerobic cells have developed an antioxidant armamentarium that includes a group of antixenobiotic genes termed phase II detoxification genes (5,6). These genes include NAD(P)H:quinone oxidoreductase 1, glutathione S-transferases, glutamate-cysteine ligase, glutathione peroxidases, and heme oxygenase-1 (HO-1). The enzyme HO-1 degrades heme to release equimolar amounts of free iron, carbon monoxide, and biliverdin, the latter being converted to bilirubin by the ubiquitous enzyme biliverdin reductase (7). Induction of a moderate intracellular heme catabolism through HO-1 represents an adaptive and protective response to oxidative injury because bilirubin is a very potent antioxidant (8,9) and because low levels of CO are cytoprotective (10).
Transcriptional up-regulation of phase II genes, including HO-1, is critically dependent on a common cis-acting enhancer sequence, termed antioxidant response element (ARE) (11,12). Numerous studies have provided clear evidence that the transcription factor nuclear factor-E2-related factor 2 (Nrf2) is a crucial molecule in the regulation of basal and induced expression of phase II genes through regulation of AREs (13,14). Nrf2 is a member of the basic leucine zipper transcription factor family featuring a cap'n'collar motif (15,16). It forms heterodimers with members of the Jun (17) and small Maf (18) proteins to bind AREs and transactivate ARE-containing genes. AREs are also subjected to negative regulation by heterodimers made with small Mafs, such as MafK, and the heme-binding protein Bach1 (19,20).
Considering that both Nrf2 and Bach1 form heterodimers with small Maf proteins in the nucleus to activate or repress AREs, respectively, it is clear that Nrf2 levels must be tightly regulated (21). Under conditions where high levels of the phase II gene products are not required, Nrf2 interacts with the BTB-Kelch protein Keap1, and this association promotes Nrf2 ubiquitination by the cullin-3-ROK1 complex and subsequent degradation by the proteasome (22)(23)(24)(25)(26)(27)(28). Therefore, under normal conditions, Nrf2 protein levels are barely detectable. On the other hand, oxidant insults promote the dissociation of the Nrf2-Keap1 complex, resulting in stabilization of the Nrf2 protein, translocation to the nucleus, and up-regulation of phase II genes (29). However, the molecular mechanisms leading to the control of nuclear Nrf2 accumulation remain unexplored.
Recent reports have characterized import and export signals in Nrf2 and have suggested that this factor might by subjected to regulation at the level of cellular location (30,31). Phosphorylation is a common post-translational modification implicated in the regulation of numerous transcription factors (32)(33)(34) but the possible role of Nrf2 phosphorylation in the regulation of its subcellular distribution remains unknown.
Glycogen synthase kinase-3␤ (GSK-3␤) is a Ser/Thr kinase involved in a variety of metabolic processes that include glycogen metabolism, Wnt signaling and sensitization to apoptosis (35). Regarding oxidative stress, GSK-3␤ intervenes in oxidative-stress-mediated apoptosis through caspase-3 by releasing cytochrome c from mitochondria (36). In fact, it has been shown that inhibition of GSK-3␤ activity by either overexpression of the GSK-3␤-binding protein, FRAT-1, or the use of a kinase-dead dominant negative mutant of GSK-3␤, or pharmacological inhibitors such as lithium, results in a decreased susceptibility of cortical neurons to trophic withdrawal-induced cell death (37). Therefore, GSK-3␤ has emerged as a new regulator of cell death but the mechanism whereby GSK-3␤ sensitizes cells to external insults such as oxidant injury is unknown.
GSK-3␤ is tightly regulated by the survival pathway represented by phosphatidylinositol-3 kinase (PI3K) and its downstream effector the Ser/Thr kinase Akt. This pathway is activated in response to growth factors and neurotrophins but also by external oxidants such as H 2 O 2 (38,39), ␤-amyloid (40), etc. PI3K and Akt inhibit apoptosis by mechanisms that may involve in part the inactivation of GSK-3␤ (41,37) because GSK-3␤ activity is negatively regulated by Akt-mediated phosphorylation at Ser-9 in the pseudosubstrate domain (41). At this time, the possible connection between PI3K/Akt/GSK-3␤ and the regulation of the Nrf2-induced antioxidant response remains to be established.
In this study, we identify Nrf2 as a substrate of GSK-3␤ and we show that this kinase governs its subcellular location. Our observations provide a mechanistic interpretation for the proapoptotic effects of GSK-3␤ through inhibition of Nrf2 function and contribute to elucidate the regulation of this transcription factor.

EXPERIMENTAL PROCEDURES
Cell Culture and Reagents-Human embryonic kidney (HEK) 293T cells were grown in DulbeccoЈs modified EagleЈs medium supplemented with 10% fetal bovine serum and 80 g/ml gentamycin. Transient transfection of these cells was performed with calcium phosphate, using the reagents from Sigma-Aldrich. GSK-3␤ inhibitor TDZD-8 was purchased from Calbiochem; Hemin, 4HT, and H 2 O 2 were from Sigma-Aldrich, and lithium chloride was purchased from Merck.
Preparation of Nuclear and Cytosolic Extracts-Cytosolic and nuclear fractions were prepared as previously described (58). Briefly, 5 million cells were washed three times with cold PBS and harvested by centrifugation at 1100 rpm for 10 min. The cell pellet was resuspended carefully in three pellet volumes of cold buffer A (20 mM HEPES, pH 7.0, 0.15 mM EDTA, 0.015 mM EGTA, 10 mM KCl, 1 mM phenylmethylsulfonyl fluoride, 20 mM NaF, 1 mM sodium pyrophosphate, 1 mM sodium orthovanadate, 1% Nonidet P-40, 1 g/ml leupeptin). Then, the homogenate was centrifuged at 500 ϫ g for 5 min. The supernatant corresponding to the cytosolic fraction was resolved in SDS-PAGE and immunoblotted with anti-V5 or anti-PDI antibodies. The nuclear pellet was resuspended in five pellet volumes of cold buffer B (10 mM HEPES, pH 8.0, 0.1 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 20 mM NaF, 1 mM sodium pyrophosphate, 1 mM sodium orthovanadate, 25% glycerol, 0.1 M NaCl, 1 g/ml leupeptin). After centrifugation in the same conditions indicated above, the nuclei were resuspended in two pellet volumes of hypertonic cold buffer C (10 mM HEPES, pH 8.0, 0.1 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 20 mM NaF, 1 mM sodium pyrophosphate, 1 mM sodium orthovanadate, 25% glycerol, 0.4 M NaCl, 1 g/ml leupeptin) and were incubated for 30 min at 4°C in a rotating wheel. Nuclear debris was removed by centrifugation at 900 ϫ g for 20 min at 4°C. The supernatant was resolved in SDS-PAGE and immunoblotted with anti-V5 and anti-Sp1 antibodies.
In Vitro Kinase Assays-In vitro phosphorylation was performed in two different ways: (i) using bacterially expressed His-tagged Nrf2, isolated with the ProBond TM purification system (Invitrogen), and (ii) using immunocomplexes with V5-tagged Nrf2 or AU5-tagged c-Jun, that were immunoprecipitated with anti-V5-or anti-AU5-specific antibodies, respectively, from HEK293T cells transiently transfected with pcDNA3.1 Nrf2⌬ETGE-V5 or pCEFL AU5-c-Jun. In both cases, the kinase assays were performed using 5 ng of active recombinant GSK-3␤ (Upstate Biotechnology) per reaction; briefly, substrate was incubated with the kinase and 5 Ci of [␥ 32 P]ATP in 25 l of reaction buffer (10 mM MgCl 2 , 100 M ATP in 40 mM MOPS, pH 7.0, and 1 mM EDTA) for 20 min at 30°C with continuous shaking. Kinase reactions were resolved in SDS-PAGE, transferred to immobilon-P membranes, and exposed to autoradiography or immunoblotted.
Immunoprecipitation-Cells were washed once with cold PBS and harvested by centrifugation at 1100 rpm for 10 min. The cell pellet was resuspended in three pellet volumes of cold lysis buffer (340 mM NaCl, 20 mM Tris-HCl, pH 7.5, 1 mM phenylmethylsulfonyl fluoride, 20 mM NaF, 1 mM sodium pyrophosphate, 1 mM sodium orthovanadate, 1% Nonidet P-40, 10% glycerol, 1 g/ml leupeptin). Lysates were incubated for 20 min at 4°C in a rotating wheel. Then, they were precleared by centrifugation at 13,000 rpm for 10 min. To prevent the NaCl interference in association between the protein and its specific antibody, NaCl concentration was diluted to 200 mM. 5 l of the corresponding antibody were added per lysate, and after incubation for 3 h at 4°C in a rotating wheel, gamma-bind Sepharose-protein G was added (Amersham Biosciences), and then incubated for one more hour at 4°C. The complexes were harvested by centrifugation, washed five times with lysis buffer, resolved in SDS-PAGE, and immunoblotted.
Immunocytochemistry-HEK293T cells were seeded in 24-well plates (75,000 cells per well), on poly-D-Lys covered slides, cultured for 16 h and transfected with calcium phosphate. After 24 h from transfection, cells were, in some cases, treated with lithium chloride 5 mM (for 6 h). Then, cells were washed with cold PBS and fixed by incubation for 15 min at room temperature with 4% paraformaldehyde. After three 5-min washes with PBS, cells were permeabilized with 0.25% Nonidet P-40 for 10 min. The slides were incubated with primary antibodies for 1 h and 30 min at 37°C in a humidified chamber. Then, cells were washed three times with PBS and incubated with secondary antibodies for 45 min under the same conditions. To visualize the nuclei, cells were stained with DAPI. The fluorescence images were captured using appropriate filters in a Leica DMIRE2TCS SP2 confocal microscope (Nussloch, Germany). The lasers used were Ar 488 nm for green fluorescence, Ar/HeNe 543 nm for red fluorescence and finally ArUV 351 nm 364 nm for UV fluorescence. Primary antibodies used in immunocytochemistry were mouse anti-V5 (Invitrogen); mouse anti-FLAG (Sigma Aldrich); mouse anti-AU5 (Covance), and rabbit anti-HA (Abcam). Secondary antibodies were Alexa Fluor 488-conjugated goat anti-mouse IgG and Alexa Fluor 546-conjugated goat anti-rabbit IgG (Molecular Probes).
Semiquantitative RT-PCR-Total cellular RNA was extracted with TRIzol reagent (Invitrogen). Equal amounts (1 g) of RNA from each treatment were reverse-transcribed (75 min, 42°C), using 5 units of avian myeloblastosis virus reverse transcriptase (Promega, Madison, WI) in the presence of 20 units of RNAsin (Promega). Semiquantitative PCR amplification of cDNA was performed in 25 l of PCR buffer, containing 10 mM Tris-HCl, pH 9.0, 50 mM KCl, 5 mM MgCl 2 , 0.1% Triton X-100, 0.6 units of TaqDNA polymerase (Promega) and 15 pmol of each synthetic gene-specific primer for selected human phase II genes as well as human GSK-3␤ and murine Nrf2. To ensure that equal amounts of cDNA were added to the PCR, we amplified the ␤-actin housekeeping gene. After an initial denaturation step for 3 min at 94°C, amplification of each cDNA was performed at the optimal number of cycles within linear range, using a thermal profile of 30 s at 94°C (denaturalization), 30 at 58°C (annealing), and 30 s at 72°C (elongation). Table 1 shows the forward and reverse primers used for each gene analyzed, the size of the PCR fragment and the number of PCR cycles for amplification in the linear range. The amplified PCR products were resolved by 5% acrylamide/bisacrylamide gel electrophoresis and stained with ethidium bromide.
GSH Levels-The levels of reduced glutathione (GSH) were determined using the GSH-sensitive probe monochlorobimane (Molecular Probes). 60,000 cells were seeded on 24-well plates and 16 h later, they were transfected as indicated. 200 M l-buthionine-(S,R)-sulfoximine (BSO) was added to the medium and incubated for 24 h. During the last 6 h, cells were treated with 0.5 mM H 2 O 2 as indicated in Fig. 8A. 100 M monochlorobimane was added during the last hour. Plates were analyzed immediately using a Fluoroscan fluorometer (Labsystems Oy, Helsinki, Finland) (20-ms integration time, 355-nm excitation filter, and 485-nm emission filter). Basal autofluorescence was determined from cells that were not incubated with the probe and was subtracted to all samples. The background incorporation of the probe to -SH protein groups was subtracted from each sample with its corresponding BSOtreated GSH-depleted controls.
LDH Levels-The levels of extracellular LDH were determined using the cytotoxicity detection kit of LDH (Roche Applied Science). 60,000 cells were seeded on 24-well plates and 16-h later, they were transfected as indicated. After 24 h, cells were treated with 0.5 mM H 2 O 2 for 6 h as showed in Fig. 8B. Then, 100 l per well of supernatant were carefully removed and transferred into corresponding wells of a 96-well microplate for determination of extracellular LDH levels. Then, the assay medium was removed from the adherent cells, and replaced with 1 ml of Dulbecco's modified Eagle's medium with 1% Triton X-100. Again, 100 l of supernatant from the lysed cells were transferred to a 96-well microplate for determination of intracellular LDH levels. To determine LDH activity 100 l of reaction mixture were added to each sample and incubated for 30 min at room temperature. Optical density was measured using a microplate reader (ELISA), at 490 nm and 600 nm. The fraction of extracellular LDH was represented as fold of increase over control untreated cells.
Image Analysis, Quantification, and Statistics-Different band intensities corresponding to ethidium bromide detection of DNA samples or Western blot detection of protein samples were quantified using the Scion Image program. Student's t test was used to assess differences between groups. A p value Ͻ 0.001 was considered significant. Unless indicated, all experiments were performed at least three times with sim- 5Ј a Linear range was achieved with the indicated number of cycles using a 20-ng cDNA template and the thermal profile described under "Experimental Procedures." b NQO1, NAD(P)H:quinone oxidoreductase.
ilar results. The values in the graphs correspond to the mean of at least three samples. Error bars indicate S.D.

RESULTS
First, we set up the conditions to investigate phase II gene induction by the PI3K/Akt survival pathway using HO-1 as a model. HEK293T cells were transiently transfected with expression vectors for active, membrane-targeted versions of PI3K (PI3K-CAAX) and Akt1 (myr-Akt1), with an efficiency of transfection close to 90% (data not shown). HO-1 mRNA and protein levels were analyzed after 24 h from transfection. As shown in Fig. 1A, PI3K-CAAX-and myr-Akt1-transfected cells exhibited increased levels of Thr 308 -phosphorylated, active Akt. Interestingly, Akt activation correlated with increased HO-1 mRNA and protein levels determined by immunoblot and semiquantitative RT- PCR (Fig. 1, A and B). Next, we used an expression vector for a conditionally active chimera of Akt, myr-Akt1-HA-ER*. In this protein, myristoylated HA-tagged Akt1 was fused at its C-terminal end with a mutated version of the hormone binding domain from the estrogen receptor (42). This fusion protein is expressed as an inactive form and becomes activated in the presence of 4-hydroxytamoxifen (4HT). HEK293T cells were transiently transfected with expression vectors for myr-Akt1-HA-ER * and HA-tagged GSK-3␤. Then cells were maintained in serum-free medium for 16 h and finally stimulated with 1 M 4HT. Fig. 1C shows the activity of the endogenous, 59 kDa Akt and the ectopically expressed, 110 kDa fusion protein as determined with an activation-specific anti-(phospho-Thr 308 )-Akt antibody. Under nonstimulated conditions, we observed a low level of endogenous and myr-Akt1-HA-ER* activation. However, addition of 4HT resulted in a timedependent phosphorylation/activation of this chimera, but not the endogenous Akt. Activation of myr-Akt1-HA-ER* by 4HT was evident after 2 h and was maximal between 4 and 8 h. Interestingly, the 4HTinduced activation of this chimeric kinase narrowly correlated with the inactivation of both endogenous GSK-3␤ and ectopically expressed HA-GSK-3␤, as determined with an anti-(phospho-Ser 9 )-GSK-3␤ specific antibody that identifies GSK-3␤ in its Akt-phosphorylated inactive state. Finally, addition of 4HT produced a time-dependent increase in HO-1 protein levels in the myr-Akt1-HA-ER* cells and did not have a significant effect on control vector-transfected cells (data not shown). The increase in HO-1 protein levels was also evident after 4 h in the presence of 4HT and remained elevated for at least 8 h. By contrast, HO-2 protein levels were not altered by Akt, in agreement with the notion that this HO isoenzyme is not regulated at the level of gene expression (7). Considering that the activation of Akt and the increase in HO-1 protein levels correlated with the inactivation of GSK-3␤, we focused our study on the putative effect of GSK-3␤ as an inhibitor of HO-1 expression.
We analyzed if GSK-3␤ inhibitors had an effect on HO-1 protein levels. We used the most well-established GSK-3␤ inhibitor, lithium (IC 50 2 mM), and one of the recently commercialized GSK-3␤ specific inhibitors, the thiadiazolidinone TDZD-8 (IC 50 2 M). As shown in Fig.  2A, the standard inducer of this pathway, hemin, produced a strong increase in HO-1 protein levels, but, interestingly, both GSK-3␤ inhibitors produced a modest but reproducible increase in HO-1 protein levels at concentrations that are compatible with GSK-3␤ inhibition.
Because HO-1 is regulated primarily at the level of transcription, we explored further the effect of the GSK-3␤ inhibitors on the reg- ulation of HO-1 gene expression by using a luciferase reporter construct that contains 15 kb of the mouse ho1 promoter. As shown in Fig. 2B, PI3K-CAAX induced a 5-fold increase in luciferase activity compared with vector transfected cells. Of note, lithium and TDZD-8 induced by themselves a 5-8 increase in promoter activity confirming that these inhibitors increase HO-1 protein levels through up-regulation of ho1 gene expression. Moreover, addition of either GSK-3␤ inhibitor to cells transfected with PI3K-CAAX resulted in a significant increase in promoter activation, suggesting that both PI3K and either lithium or TDZD-8 cooperate to inactivate GSK-3␤ and to induce HO-1 expression.
To identify the putative response element that is targeted by GSK-3␤ in the 15-kb promoter fragment, we used genetic inhibitors of three candidate transcription factors that might be regulated by PI3K and Akt: Nrf2, heat shock factor-1 (HSF-1) and nuclear factor-B (NF-B). Nrf2 activity was blocked by overexpression of the bZIP dimerization domain of Nrf2 (⌬Nrf2(DN)) that competes with endogenous Nrf2 for heterodimerization. NF-B activity was blocked by overexpression of a degradation resistant mutant of IB␣ (IB␣(S32A/S36A)) that sequesters NF-B in the cytosol. HSF-1 activity was inhibited with a HSF-1 dominant negative mutant (43). As shown in Fig. 2C, the promoter activation by lithium was insensitive to dominant negative IB␣(S32A/ S36A) and HSF-1(DN), but was fully blocked by ⌬Nrf2 (DN). Therefore, these results suggest that GSK-3␤ might be repressing a signaling protein involved in regulation of AREs.
To confirm that indeed GSK-3␤ is inhibiting HO-1 expression at the level of the AREs, we used genetic variants of GSK-3␤ and a luciferase reporter containing 3 tandem sequences of the mouse ho1 ARE. As shown in Fig. 2D, PI3K-CAAX specifically activated this promoter element. The mutant GSK-3␤-⌬9, which lacks the first nine N-terminal residues including Ser 9 and, therefore, is insensitive to PI3K/Akt phosphorylation/inhibition, fully prevented the PI3K-induced activation of the AREs. Finally, a dominant negative GSK-3␤ mutant, carrying a Lys to Arg substitution in the ATP acceptor site (GSK-3␤(K85R)), partially activated the AREs by itself, just as the GSK-3␤ inhibitors did, and cooperated with PI3K-CAAX to further activate the AREs. Hence, pharmacological and genetic approaches allowed the identification of AREs as targets for positive regulation by PI3K/Akt and negative regulation by GSK-3␤.
We explored the possibility that Nrf2 might be a direct target of GSK-3␤. We could hardly detect endogenous or ectopically expressed wild-type Nrf2 in the absence of an oxidant stimulus (data not shown). Therefore, to analyze the effect of GSK-3␤ on this factor in the absence of other variables, we used a Nrf2 mutant lacking the ETGE sequence for binding to Keap1 (see "Discussion"). As a rule, GSK-3␤ requires the cooperation of another kinase, a priming kinase, that phosphorylates the substrate in an S/T residue located at ϩ4 position. To circumvent

GSK-3␤ Phosphorylates and Excludes Nrf2 from the Nucleus
this problem, we used as a substrate source V5-tagged Nrf2⌬ETGE, immunoprecipitated from HEK293T-transfected cells because this Nrf2 should be already phosphorylated by whatever is the priming kinase. As a control, we immunoprecipitated under the same conditions AU5-tagged c-Jun, that is a well-reported transcription factor substrate of GSK-3␤. In vitro kinase assays were performed with commercially available GSK-3␤. As shown in Fig. 3A, in the absence of GSK-3␤, we could not detect a significant incorporation of 32 P in c-Jun or in Nrf2 indicating that the immunocomplex alone does not contain any kinase capable of phosphorylating these transcription factors. However, when we added purified GSK-3␤ to the immunocomplexes, we found a significant incorporation of 32 P in both c-Jun and Nrf2. Moreover, phosphorylation of both substrates was reduced when lithium (5 mM) was included in the kinase reaction. To further explore the possibility that GSK-3␤ might phosphorylate Nrf2 in vitro, we used bacterially expressed His-tagged mouse Nrf2 as a substrate. Despite the lack of phosphate-primed substrate, again we observed two radioactive bands of 47 and 110 kDa corresponding to autophosphorylated GSK-3␤ and recombinant Nrf2-His respectively. These results indicate that GSK-3␤ phosphorylates Nrf2 in vitro. To determine if this kinase also phosphorylates Nrf2 in vivo, we transiently cotransfected HEK293T cells with expression vectors for constitutively active GSK-3␤-⌬9 and either Nrf2⌬ETGE-V5 or AU5-c-Jun as a positive control. After 24 h, cell lysates were immunoblotted with anti-V5 and anti-AU5 antibodies to detect the protein bands corresponding to these transcription factors and with anti-phospho-Ser antibodies. As shown in Fig. 3C, both anti-AU5 and anti-V5 antibodies showed several retarded bands, consistent with phosphorylation. For c-Jun, cotransfection with active GSK-3␤ resulted in substantial enrichment of a slowly migrating form of AU5c-Jun of about 80 kDa. This band was also recognized with the anti-pSer antibody, indicating that indeed it is phosphorylated in Ser. But more interestingly, the anti-V5 antibody evidenced several slowly migrating forms of Nrf2-V5 in the GSK-3␤-⌬9 cotransfected cells that were also identified by the anti-pSer antibody. Taken together these results indicate that Nrf2 is phosphorylated at Ser residues by GSK-3␤.
Next, we analyzed how GSK-3␤-phosphorylation might affect Nrf2 function. First, we analyzed if phosphorylated Nrf2 is more sensitive to degradation and we found that lithium did not alter the abundance of endogenous Nrf2 (data not shown). Then, we explored the possibility that GSK-3␤ might be altering the subcellular location of Nrf2. HEK293T cells were cotransfected with GSK-3␤-⌬9 and either Nrf2-⌬ETGE-V5 or FLAG-Bach1, V5-Keap1, or AU5-c-Jun as controls. As shown in Fig. 4, in the absence of active GSK-3␤-⌬9, both Bach1 and Keap1 exhibited a mostly cytoplasmic location (in agreement with Refs. 44 and 45) whereas the transcription factors Nrf2 and c-Jun were located at the nucleus (in agreement with Refs. 46 and 44). Interestingly, when co-expressed with GSK-3␤-⌬9, most Nrf2 was found in the cytoplasm, and the distribution of Bach1, Keap1, and c-Jun remained unaltered.
To further confirm that Nrf2 redistribution toward the cytoplasm was due to GSK-3␤, we analyzed the Nrf2 localization in the presence of GSK-3␤ inhibitors. As shown in Fig. 5A, active myr-Akt1-HA preserved the nuclear location of Nrf2 but the Akt-insensitive GSK-3␤ mutant, GSK-3␤-⌬9, excluded Nrf2 from the nucleus. On the other hand, lithium, that is capable of inhibiting the ⌬9 mutant through com- . GSK-3␤ regulates the subcellular distribution of Nrf2, but not that of Bach1, Keap1, or c-Jun. HEK293T cells were cotransfected with active HA-GSK-3␤-⌬9 and either FLAG-Bach1 or V5-Keap1 or AU5-c-Jun or Nrf2⌬ETGE-V5 as indicated. 24 h from transfection, cells were analyzed by immunocytochemistry. GSK-3␤ was detected with rabbit anti-HA antibody and developed with a secondary antibody conjugated with red Alexa-546 (left column), whereas the other antigens were detected with the corresponding mouse anti-FLAG, anti-V5, or anti-AU5 and developed with a secondary antibody conjugated with green Alexa-488 (middle column). DAPI nuclear staining, merged with the green fluorescence is shown in the right column. For quantification, subcellular distribution was classified into three categories: cytoplasmic-dominant accumulation (white sectors), nuclear-dominant accumulation (black sectors), and similar distributions between cytoplasm and nucleus (gray sectors). Results correspond to samples of 500 cells. Only cells expressing the two ectopically expressed proteins were counted. The arrowheads in the Nrf2/GSK-3␤-⌬9 panels point a cell that was transfected only with Nrf2⌬ETGE-V5 and serves as an internal control. MAY 26, 2006 • VOLUME 281 • NUMBER 21 petition with Mg 2ϩ , partially reversed the distribution of Nrf2 toward the nucleus. In additional experiments, we analyzed the nuclear/cytoplasmic distribution of Nrf2 in subcellular protein fractions. As shown in Fig. 5B, in HEK293T cells transfected with Nrf2⌬ETGE-V5 alone, this protein was detected only in the nuclear fraction but cotransfection with GSK-3␤-⌬9 shifted Nrf2 to the cytosolic fraction. Moreover, lithium (5 mM) partially reversed the effect of GSK-3␤ and about 40% of the Nrf2 protein was driven back to the nucleus. Taken together, these results indicate that GSK-3␤ shifts the subcellular distribution of Nrf2 toward the cytoplasmic compartment.

GSK-3␤ Phosphorylates and Excludes Nrf2 from the Nucleus
These results suggest that GSK-3␤ might interfere with the transactivating function of Nrf2. To analyze this possibility, HEK293T cells were transiently cotransfected with the luciferase reporter construct carrying 3 tandem sequences of mouse ho1 AREs (ARE-LUC), and a fixed amount of either empty vector or GSK-3␤-⌬9 (0.6 g/well). In addition, cells were also cotransfected with increasing amounts of the Nrf2⌬ETGE-V5 expression vector as indicated in Fig. 6. Nrf2 alone induced a dose-dependent activation of the ARE-LUC reporter that was maximal with 0.1 g of Nrf2 plasmid/well. By contrast, GSK-3␤-⌬9 highly reduced ARE induction by Nrf2 in all cotransfections. These results indicated that, indeed, GSK-3␤ interferes with Nrf2 transactivating activity and provided the conditions to further analyze the functional relevance of this interference.
Then, we extended the finding that GSK-3␤ blocks Nrf2-induced HO-1 expression to other phase II genes. HEK293T cells were cotransfected with 0.7 g of Nrf2⌬ETGE or 35 g of GSK-3␤-⌬9 expression vectors in p100 plates as indicated in Fig. 7. The total amount of transfected DNA was equalized to 40 g with empty vector. After 48 h, the levels of the indicated phase II enzymes were analyzed by semiquantitative RT-PCR. As shown in Fig. 7, HO-2, which is not an inducible phase II enzyme, was not affected by overexpression of either Nrf2 or GSK-3␤. By contrast, the phase II genes coding for HO-1, GSTA1, GCL-M, Gpx-4, and NQO1 were significantly up-regulated by Nrf2 overexpression to different degrees ranging between 2-and 8-fold. Interestingly, and in agreement with Fig. 6, GSK-3␤ attenuated the transcriptional activity to Nrf2 on these genes. These results indicate that the negative regulation of Nrf2 function by GSK-3␤ described here for HO-1 as a model has a wider effect on all phase II genes analyzed.
Considering that phase II genes protect cells from oxidative aggressions, we analyzed if GSK-3␤ might reduce the antioxidant and cytoprotective function of Nrf2 against hydrogen peroxide. First we measured the intracellular levels of reduced glutathione, the major non-protein-reducing agent within the cell, with the fluorescent probe monochlorobimane (47). Unspecific basal incorporation of the probe was determined in cells treated for 24 h with BSO, an irreversible inhibitor of glutamate cysteine ligase, which is the rate-limiting enzyme in glutathione biosynthesis (48). HEK293T cells were transfected with expression vectors for Nrf2⌬ETGE-V5 or GSK-3␤-⌬9 and with empty vector under similar conditions as those shown in Fig. 7. Then, cells were maintained in serum-free medium for 16 h and finally treated with 0.5 mM H 2 O 2 for 6 h. During the last hour of incubation, the medium was supplemented with 100 M monochlorobimane (49). The amount of fluorescent GSHmonochlorobimane adduct that is generated by the enzymatic action of intracellular glutathione S-transferase was measured fluorometrically. As shown in Fig. 8A, both control-and Nrf2-transfected cells exhibited similar GSH levels in the absence of H 2 O 2 as expected from the GSHinduced negative feedback regulation of GSH synthesis (48) (see "Discussion"). Interestingly, the GSH pool of control vector-transfected cells was significantly reduced by a 6-h incubation with H 2 O 2 whereas the pool of Nrf2-transfected cell remained unaltered. GSK-3␤-transfected cells exhibited partially lower GSH levels than control cells. Moreover, in cells cotransfected with both GSK-3␤ and Nrf2, GSH levels were not restored when cells were challenged with H 2 O 2 . These results indicate that GSK-3␤ prevents the antioxidant function of Nrf2 by interfering at least in part with GSH metabolism.
Finally, we determined whether the GSK-3␤-driven reduction in Nrf2 function might sensitize cells to induction of cell death by H 2 O 2 . Loss of plasma membrane integrity, a hallmark of cell death, was quantified as the release to the culture medium of cytosolic LDH (50).

DISCUSSION
In this study we report the negative regulation of the transcription factor Nrf2 by GSK-3␤, a kinase that sensitizes cells for cell death. The regulation involves phosphorylation and nuclear exclusion of Nrf2, therefore preventing binding and activation of the AREs located in the phase II gene promoters.
Most of this study was done using HO-1 as a prototypic phase II gene. Of note, while the levels of this isoenzyme were up-regulated in response to Akt activation or to overexpression of Nrf2, the levels of the HO-2 isoenzyme, which is not transcriptionally regulated (7), remained unchanged. Moreover, the results with HO-1 were confirmed in essence for four additional phase II antioxidant genes, GSTA1, Gpx4, GCL-M, and NQO1.
Considering that phase II genes protect cells from oxidant attack, in recent years we have searched for a mechanistic connection between regulation of these genes and the canonical survival pathway represented by PI3K and its downstream effector Akt. In fact, we have reported the up-regulation of the phase II antioxidant enzyme heme oxigenase-1 (HO-1) in response to nerve growth factor (NGF) (51) and to the phytochemical carnosol (52) by a mechanism that requires PI3K activation. However, phase II gene promoters, including HO-1s, contain a large number of putative regulatory sequences and therefore, the mechanistic connection at the molecular level has remained elusive. Interestingly, a large part of the NGF effect on HO-1 transcriptional induction was sensitive to the protein synthesis inhibitor cycloheximide (51), indicating that NGF induction requires a short half-life protein.
This evidence prompted us to analyze the role of Nrf2, a transcription factor that targets phase II genes at the antioxidant response element and has a half-life of just a few minutes (25).
First, we analyzed if Nrf2 and other regulators of the pathway, such as MafG, MafK, Keap1, and Bach1 could be targeted by direct Akt-driven phosphorylation. Only Keap1 contains a weakly putative site for Akt phosphorylation at Ser 53 ( 46 QHGNRTFSYTLEDHT 60 ) that is conserved in several species. However, in in vitro kinase assays with recombinant Keap1 and purified Akt1 we found a very low stoichiometry of phosphate incorporation compared with other well-established Akt substrates such as Bad (data not shown). Therefore, other mechanisms had to be involved in Nrf2 regulation by PI3K/Akt.
Bearing in mind that active GSK-3␤ makes cells more sensitive to oxidative stress and apoptosis, and that GSK-3␤ is inhibited by Akt through phosphorylation at Ser 9 , we analyzed if GSK-3␤ might be regulating negatively this antioxidant pathway. Indeed, pharmacological inhibitors of GSK-3␤ slightly increased HO-1 protein levels. Moreover, these inhibitors and genetic mutants of GSK-3␤ cooperated in the regulation of AREs by PI3K, further suggesting that PI3K activates and GSK-3␤ inhibits AREs.
Then, we searched for specific components of the cell signaling pathway leading to AREs activation that might by regulated by GSK-3␤. Interestingly, mouse Nrf2 has several candidate sites that conform the consensus sequence for phosphorylation by GSK-3␤ ((S/T)XXX(S/T)). We found that, indeed, GSK-3␤ phosphorylates eukaryotic and bacterially expressed Nrf2 in vitro. Moreover, cotransfection experiments with V5-tagged Nrf2 demonstrated that this transcription factor, like c-Jun, is a target of GSK-3␤ in vivo. The identification of the GSK-3␤phosphorylated residues is complicated by the fact that mouse Nrf2 has up to 10 putative GSK-3␤ phosphorylation sites, many of them conserved in other species, and that this kinase usually acts in concert with another kinase that primes the substrate via phosphorylation at position ϩ4. We are currently identifying the specific residues that are phosphorylated by GSK-3␤ and the nature of the priming kinase.
To determine how GSK-3␤ is regulating the function of Nrf2, we first analyzed the stability of this protein. We could not detect a significant difference in Nrf2 half-life in the absence or presence of active GSK-3␤ (data not shown). Therefore, we focused our study on a recently reported mechanism of regulation of Nrf2 activity, which is based on subcellular localization. We found that GSK-3␤ alters the distribution of Nrf2 to keep it in the cytosol. Nrf2 has two putative nuclear localization signals (NLS) and one nuclear export signal (NES) (31), (45). Therefore, GSK-3␤ might be retaining Nrf2 in the cytoplasm or might be increasing the rate of export. In fact, both mechanisms of regulation have been proposed for other transcription factors. For instance, cytoplasmic retention has been reported for the forkhead transcription factor FOXO-3a (53). Upon Akt phosphorylation, 14-3-3 proteins bind phospho-FOXO3a and sequester this transcription factor in the cytosol. However, our preliminary data and the analogy with other transcription factors such as NF-AT (34) and cyclin D1 (54) suggest that GSK-3␤ may be increasing the rate of nuclear export.
In previous studies, we and others have observed that pharmacological inducers of Nrf2 activate HO-1 expression and provide protection against oxidant insults. Interestingly, however, when we inhibited HO-1 activity we still detected a significant level of oxidant protection by the Nrf2 inducers (52). Those observations suggested that other enzymes are also participating in oxidant protection and, accordingly, here we have observed the up-regulation of other cytoprotective phase II genes, namely Gpx4, GSTA1, GCL-M, and NQO1, which constitute several fronts of defense against ROS to preserve redox homeostasis. One of these genes, GCL-M constitutes the modulatory subunit of glutamatecysteine ligase, the rate-limiting enzyme involved in GSH synthesis. It was surprising, at first glance, that up-regulation of this enzyme did not produce a clear increase in basal GSH levels. This is probably due to the well-documented fact that GSH exerts negative feedback regulation over its own synthesis by at least two mechanisms: (i) competition for the binding site for glutamate in the catalytic subunit of glutamatecysteine ligase (GCL-C), and (ii) prevention of dimmer formation between GCL-C and GCL-M by keeping critical cysteine SH groups of these subunits in its reduced state (48). However, in the presence of H 2 O 2 , the GSH pool was substantially reduced in the control vector transfected cells but not in the Nrf2 transfected cells, therefore unmasking the higher capacity of these cells to safeguard a reducing environment. On the contrary, GSK-3␤ attenuated the restoration of GSH levels even in the Nrf2 co-transfected cells, consistently with its inhibitory role on Nrf2 function. The attenuation of Nrf2 antioxidant capacity by GSK-3␤ resulted in reduced cytoprotection because cells cotransfected with Nrf2 and GSK-3␤ exhibited higher mortality in the presence of H 2 O 2 than cells cotransfected with Nrf2 and empty vector.
In recent years, it has been shown that there is a loss in oxidative stress tolerance with aging that is linked to a parallel reduction in Akt activity (55) and to an increase in GSK-3␤ activity (56). Interestingly, the decline in Nrf2 function also sensitizes cells to oxidative stress during aging (4). Therefore, it is tempting to speculate that the molecular mechanisms described here for Nrf2 regulation may explain the progressive increase in cells susceptibility to oxidants that occurs during aging.