Identification of Activating Transcription Factor 4 (ATF4) as an Nrf2-interacting Protein FOR HEME OXYGENASE-1 GENE REGULATION*

Nrf2 regulates expression of genes encoding enzymes with antioxidant ( e.g . heme oxygenase-1 (HO-1)) or xe-nobiotic detoxification ( e.g . NAD(P)H:quinone oxidoreductase, glutathione S -transferase) functions via the stress- or antioxidant-response elements (StRE/ ARE). Nrf2 heterodimerizes with small Maf proteins, but the role of such dimers in gene induction is contro-versial, and other partners may exist. By using the yeast two-hybrid assay, we identified activating transcription factor (ATF) 4 as a potential Nrf2-interacting protein. Association between Nrf2 and ATF4 in mammalian cells was confirmed by co-immunoprecipitation and mammalian two-hybrid assays. Furthermore, Nrf2 z ATF4 dimers bound to an StRE sequence from the ho-1 gene. CdCl 2 , a potent inducer of HO-1, increased expression of ATF4 in mouse hepatoma cells, and detectable induction

Overproduction of oxygen free radicals, attenuation of antioxidant systems, or both, commonly in response to extracellular stimuli, disturbs the cellular redox status and leads to oxidative stress. Such conditions typically elicit an adaptive response aimed at reversing this imbalance and maintaining redox homeostasis. In part, this adaptive response includes the activation of specific signaling pathways and, ultimately, the coordinate induction of a select set of genes that encode proteins with distinct activities that individually and collectively manifest antioxidant and cytoprotective functions. Central to this induction process are redox-sensitive transcription factors, such as nuclear factor-B (NF-B) 1 and activator protein-1, arguably the two most prominent regulators of this cellular response mechanism (reviewed in Refs. 1 and 2).
Recent studies from several laboratories (3)(4)(5)(6)(7)(8) have implicated another transcription regulator, Nrf 2, with a potentially significant role in the adaptive response to oxidative stress. Nrf 2 belongs to the CNC-bZIP subfamily of basic region/ leucine zipper (bZIP) transcription factors. CNC-bZIP proteins are distinguished from other bZIP subfamilies, including those composed of Jun, Fos, ATF/CREB, or Maf factors, in that they also contain a Cap'n'Collar (CNC) structural motif homologous to a region within the Drosophila homoeotic selector protein encoded by the cap'n'collar gene (9). bZIP proteins function as obligate dimers; for example, individual Jun-Jun or Jun-Fos dimers are commonly and collectively referred to as activator protein-1 transcription factors. Sequences necessary for both dimerization and DNA binding reside within the bipartite bZIP domain.
Limited but consistent observations (6,8) suggest that under normal conditions, and as is the case for NF-B factors, Nrf 2 exists in an inactive, cytoplasm-localized state, in part or fully as a consequence of binding to the cytoskeleton-associated protein Keap1. After exposure of cells to electrophiles or oxidative stress-generating agents, the cytoplasmic retention mechanism is inactivated, and Nrf 2 is transported to the nucleus by an as yet uncharacterized mechanism(s) but one that may involve protein kinase C-mediated phosphorylation of Nrf 2 (8). Within the nucleus, Nrf 2 activates transcription of a select set of target genes by binding to distinct but very similar DNA elements, individually or alternatively referred to as the NF-E2-binding site (10), the Maf recognition element (MARE, 11), the stress-response element (12), or the antioxidant-response element (13). Many of the Nrf 2 target genes (3-5, 7, 8) encode proteins that play a central role in the adaptive response to oxidative stress. Among others, these include heme oxygenase-1 (HO-1), an enzyme that catalyzes the rate-limiting reaction in heme degradation, a catabolic pathway that leads to the production of bilirubin, a potent antioxidant; NAD(P)H:quinone oxidoreductase (NQO), which catalyzes two-electron reduction of quinones, preventing the participation of such compounds in redox cycling and oxidative stress; ␥-glutamylcysteine synthase, which catalyzes the rate-limiting reaction in glutathione biosynthesis; and glutathione S-transferase, which conjugates hydrophobic electrophiles and reactive oxygen species with glutathione.
Nrf 2, like other CNC/bZIP proteins and Fos family members, belongs to a sub-class of bZIP factors with leucine zipper motifs incapable of self-dimerization. Consequently, sequencespecific DNA binding and subsequent induction of target gene transcription requires association of Nrf 2 with other transcription factors. In accordance with the paradigm established by NF-E2 (10), the first CNC-bZIP containing mammalian transcription factor isolated, the most prominent dimerization partners of Nrf 2 are the small Maf proteins, MafF, MafG and MafK (also referred to as p18 (14)). The precise function of such Nrf 2⅐Maf dimers, however, is controversial, as they have been proposed to function as both positive (5) and negative regulators (15) of ARE-dependent gene transcription. Jun-Nrf 2 complexes have also been implicated as positive effectors of ARE-dependent genes (16).
Given our incomplete understanding of Nrf 2 function, the propensity of bZIP proteins to form inter-and intra-family dimers (17,18), and of transcription factors in general to form complexes that tend to provide both diversity to, and discrimination of, genetic responses to extracellular stimuli, we reasoned that additional Nrf 2-containing complexes exist intracellularly and that such complexes would likely regulate Nrf 2 target gene expression. Accordingly, we have used the yeast two-hybrid screening procedure to identify proteins that associate with Nrf 2. Herein, we report the identification of ATF4 as a Nrf 2-interacting protein and explore the potential role of ATF4 in the regulation of one Nrf 2 target gene, ho-1.

Materials
Tissue culture media were from Life Technologies, Inc., and fetal bovine serum was obtained from Mediatech. Restriction endonucleases and other DNA-modifying enzymes were purchased from either Life Technologies, Inc., or New England Biolabs. Oligonucleotides were synthesized by IDT, Inc. Radiolabeled nucleotides were obtained from PerkinElmer Life Sciences. Reagents for luciferase assays were purchased from Sigma. Anti-mouse Nrf 2 was kindly provided by Dr. M. Yamamoto. Antibodies against other transcription factors, including anti-human Nrf 2, and HO-1 were obtained from Santa Cruz Biotechnology and StressGen Biotechnologies Corp., respectively. All other chemicals were reagent grade. Mammalian Expression Plasmids and Dominant-negative Mutants-Expression plasmids encoding mouse Nrf 2 (pEF/Nrf 2), p18 (pEF/p18), and the mutant p18 (pEF/p18M) (19) were kindly provided by Dr. Stuart Orkin. Mouse and rat ATF4 cDNAs were cloned into pEF/myc/mito and pcDNA3.1/myc-his (Invitrogen), respectively, to generate pEF/mATF4 and pCMV/rATF4. A dominant mutant of mouse ATF4 was generated by overlap extension using PCR resulting in a protein with 6 amino acid substitutions within the DNA-binding domain ( 292 RYRQKKR 298 to 292 GYLEAAA 298 ). The amplification product was cloned into pEF/myc/mito to generate the plasmid (pEF/mATF4M). The dominant mutant of Jun D (pCMV/JunDM) was constructed by cloning the 591-base pair BssHII/BssHII (blunt-ended) fragment of the mouse Jun D cDNA into the vector pCMV-Tag2B (Stratagene). This manipulation deletes amino acid residues 1-169 resulting in a transactivation domain-deficient mutant of Jun D similar to one described earlier (20). Dominant mutants of c-Jun and Nrf 2 have been described previously (4).
Yeast and Mammalian Two-hybrid (Y2H and M2H) Constructs-The "bait" plasmid, pDBLeu-Nrf 2, for Y2H was constructed in the following manner. The mouse Nrf 2 cDNA sequence encoding amino acid residues 393-581 (numbering as in Ref. 21) was amplified by PCR using the primer pair Nrf 2-1, 5Ј-CACGCGTCGACTATGCGTGAATCCCAATG-3Ј, and Nrf 2-2, 5Ј-TCCTCCGGATATCAGTTTTTCTTTGTAT-3Ј. The amplified product was digested with SalI and EcoRV restriction endonucleases (recognition sites underlined) and cloned between the SalI and StuI sites of the pDBLeu vector (Life Technologies, Inc.) in-frame with the Gal4 DNA-binding domain (Gdbd). The integrity of the mouse Nrf 2 cDNA and production of the fusion protein was confirmed by DNA sequencing and Western blotting, respectively. The mammalian Gdbd vector, pEG, was constructed by cloning the Gdbd (residues 1-147) into pEF/myc/mito. Mouse Nrf 2 sequences (aa residues 13-581 or 314 -581) were subsequently cloned downstream of, and in-frame with, the Gdbd to generate pEG/Nrf 2 plasmids. The "activation domain" vector (pAD) was constructed by cloning an 870-base pair BglII/HindIII (blunt-ended) fragment of mouse Nrf 2 (aa 13-302) into pCMV-Tag2B. Full-length ATF3, p18 and rat ATF4, sequences were subsequently cloned into pAD in-frame with the Nrf 2 sequence.
Reporter Gene Plasmids-pFRluc, containing 5 tandem copies of the Gal4-binding site, was obtained from Stratagene. The construction of pE1-luc, containing the mouse ho-1 gene distal enhancer 1, and pStREluc, containing three copies of the mouse ho-1 StRE3, has been described previously (4,22). Plasmid pCMV/␤-gal, encoding the Escherichia coli ␤-galactosidase gene, was kindly provided by Dr. Ping Wei.

Yeast Two-hybrid (Y2H) Screening
Screening was carried out using the Y2H system from Life Technologies, Inc. Briefly, plasmid pDBLeu-Nrf 2 was introduced into yeast strain MaV203 (MAT␣, leu2-3, 112, trp1-901, his3⌬200, ade2-101, gal4⌬, gal80⌬, SPAL10::URA3, GAL1::lacZ, HIS3 UAS GAL1 ::HIS3@ LYS2, can1 R , cyh2 R ), and transformants were selected and purified on medium lacking leucine. Subsequently, DNA representing rat brain or liver cDNAs, cloned into the Gal4 activation domain vector pPC86, was transformed into MaV203/pDBLeu-Nrf 2 strain. Transformants were selected on medium containing 25 mM 3-amino-1,2,4-triazole but lacking tryptophan, leucine, uracil, and histidine. Positive colonies were assayed for activation of the lacZ reporter gene. Plasmids isolated from the positive colonies were rescued in E. coli HD10B and re-assayed for interaction activity by transformation into the MaV203/pDBLeu-Nrf 2 strain and growth on selection media. The 5Ј end of the rat ATF4 cDNA was isolated by PCR amplification (5Ј-rapid amplification from cDNA ends) from a rat brain Marathon-Ready cDNA mixture (CLONTECH) according to the manufacturer's recommendation using two different gene-specific primers ATF4-1 (5Ј-TAGGACTCAGGGCTCATACAGAT-GCCA-3Ј) and ATF4-2 (5Ј-TTGAAGTGCTTGGCCACCTCCAGATAG-3Ј) and the adaptor primer provided in the kit. Both amplification products were purified and cloned into the pT-Adv vector (CLONTECH Laboratories Inc). Automated DNA sequence analysis was carried out by the Howard hughes Medical Institute Biopolymer/W. M. Keck Foundation Biotechnology Resource Laboratory at Yale University. Identical 5Ј sequences were obtained from clones derived from both amplification products.

Mammalian Cell Culture, Transfection, and Enzyme Assays
COS-7 (African green monkey kidney), Hepa (mouse hepatoma), and MCF-7 (human mammary epithelial) cells were cultured in Dulbecco's modified Eagle's medium, whereas HeLa (human cervical carcinoma) cells were cultured in Eagle's minimal essential medium. All media were supplemented with 0.45% glucose, 10% fetal bovine serum, 50 g/ml gentamicin sulfate, and 10 ng/ml insulin (MCF-7 only). Transient transfection of luciferase constructs was carried out by the calcium phosphate precipitation technique as described previously (23) or with Fugene 6 transfection reagent (Roche Molecular Biochemicals) according to the manufacturer's recommendation. Additional details are provided in the figure legends. Transfection efficiency was monitored by co-transfection with pCMV/␤-gal. Preparation of cell extract and measurement of luciferase activity were carried out as described previously (24). ␤-Galactosidase activity was measured using the Galacto-Light (Tropix, Inc.) chemiluminescent assay kit according to the manufacturer's protocol.

Recombinant Protein and Electrophoretic Mobility Shift Assays (EMSA)
Full-length (p18, Fos B, and ATF3) or nearly full-length (mATF4, residues 31-349) coding regions were cloned downstream of, and in-frame with, the hexa-histidine tag in the T7 RNA polymerase-based prokaryotic expression vector series pET30a-c (Novagen Inc.). Recombinant proteins were purified from inclusion bodies by nickel affinity chromatography as per the manufacturer's protocol or according to the protocol of Holzinger et al. (25). Nrf 2M protein (residues 393-581), containing the DNA binding and leucine zipper dimerization domains, was synthesized by coupled in vitro transcription and translation reaction as described previously (26). EMSA was carried out as described previously (23) using a double-stranded oligonucleotide containing the sequence 5Ј-TTTTCTGCTGAGTCAAGGTCCG-3Ј (core StRE underlined) as probe. Five l of Nrf 2M synthesis product and/or 100 ng of recombinant protein were used in EMSA reactions.

Co-immunoprecipitation and Western Blot Analysis
COS-7 cells were transfected 24 h after plating (1 ϫ 10 6 /100-mm plate) with a total of 5 g of either empty vector (pEF/myc/mito), pEF/mATF4, pEF/Nrf 2, or a combination of these plasmids using Lipofectin transfection reagent (Life Technologies, Inc.) as specified by the manufacturer. Cells were harvested 36 h after transfection and resuspended in 200 l of lysis buffer (10 mM Tris-HCl (pH 7.5) containing 0.5% (v/v) Nonidet P-40, 150 mM NaCl, and 1 mM EDTA). Cell lysates were cleared by centrifugation, and immunoprecipitation was carried out with 100 g of cell lysate using protein G-agarose beads as described (27). Immune complexes were eluted from the beads with 2ϫ SDS-PAGE sample buffer and subjected to denaturing polyacrylamide gel electrophoresis. Western blotting was carried out as described previously (4). All antibodies were used at dilutions recommended by the respective suppliers.

Identification of ATF4 as an Nrf 2-interacting Protein-To
identify proteins that interact with Nrf 2, a cDNA fragment encoding the C-terminal portion of mouse Nrf 2 was amplified by PCR and cloned in-frame and downstream of the Gdbd. The resulting fusion protein was used as "bait" in a yeast twohybrid screening assay as described under "Experimental Procedures." From a total of 2 ϫ 10 6 yeast transformants, harboring either rat brain or liver cDNA/Gal4 activation domain fusions, seven independent clones (3 liver and 4 brain) that encoded Nrf 2-interacting polypeptides were identified after a series of selection methods.
The nucleotide sequences of the inserts within the positive clones were determined, and the results indicated that five of the cDNAs were derived from the same mRNA. A sequence similarity search using "blastn" revealed significant similarities to sequences encoding mouse ATF4 and human CREB2 (ATF4) (28,29) suggesting that these clones encoded the rat FIG. 1. Comparison of rat, mouse, and human ATF4 protein sequences. The amino acid sequence alignment was generated using the ClustalW algorithm (44). Residues conserved in all sequences are marked with asterisks. Colons and periods indicate conservative and semi-conservative substitutions, respectively. The bZIP domain (basic region and leucine zipper I) and the second zipper (zipper II) are indicated. The leucine (or corresponding) residues within the heptad repeat of the zippers are marked with filled circles above. Mouse ATF4 accession number is M94087; human ATF4 accession number is XP_010004 FIG. 2. Co-immunoprecipitation of Nrf 2 by anti-ATF4 antibody. COS-7 cells were transfected with empty vector or the indicated transcription factor expression plasmid. Cell lysates (100 g of protein) were subjected to immunoprecipitation (IP) with anti-ATF4 (A) or anti-Nrf 2 (N) antibodies (Ab) in the presence or absence of the corresponding blocking peptide (BP). Cell lysates (25 g) or immune complexes (total sample from a single precipitation) were size-fractionated on a denaturing 10% polyacrylamide gel; proteins were transferred to nitrocellulose, and the Nrf 2 protein (marked by arrow) was detected by Western blotting. homolog of ATF4. Since none of the isolates contained the initiation codon, the full-length ATF4 cDNA was obtained by PCR amplification from a rat brain cDNA library, cloned, and subjected to DNA sequence analysis.
The deduced amino acid sequence of rat ATF4 is aligned with those of mouse and human ATF4 in Fig. 1. Between these three species, ATF4 exhibits 84.4% sequence conservation and the rat protein displays 94.8 and 87.2% sequence identity to mouse and human ATF4 s, respectively. As expected, the highest degree of conservation is observed within the basic region (DNA binding) and the adjacent leucine zipper (dimerization) domains. The second "leucine zipper" region, to which a function has yet to be assigned, exhibits greater divergence although the repeating leucine (or corresponding) residues at every 7th position are completely conserved in the three proteins.
Interaction between Nrf 2 and ATF4 in Mammalian Cells-Association between Nrf 2 and ATF4 in mammalian cells was confirmed by co-immunoprecipitation experiments and mammalian two-hybrid assays. For the former, expression plasmids encoding ATF4 or Nrf 2 were transfected, individually or in combination, into COS-7 cells; the cells were lysed, and the lysates subjected to immunoprecipitation with anti-ATF4 or anti-Nrf 2 antibodies in the presence or absence of the corresponding blocking peptide. Immunoprecipitates were subsequently analyzed by Western blotting. As shown in Fig. 2 5 and 7).
The results from co-immunoprecipitation experiments were corroborated by mammalian two-hybrid assays. In these experiments, nearly full-length mouse Nrf 2 (aa 13-581; Fig. 3A) or the C-terminal portion of Nrf 2 (aa 314 -581; Fig. 3B) was fused to the Gdbd, and these fusions served as interaction targets. Sequences encoding test proteins were fused in-frame to an N-terminal region of Nrf 2 (amino acids 13-302) that contains a potent transcription activation domain (AD). Gdbd-Nrf 2-(13-581) strongly trans-activated a luciferase reporter gene under the control of Gal4-binding sites, pFRluc. Co-expression of Nrf 2 AD further increased luciferase activity by 4 -5-fold suggesting self-interaction between Nrf 2 proteins. AD fusions containing full-length rat ATF4 or mouse ATF3 exhibited even greater trans-activation capabilities, ϳ35-fold above control and 7-fold above AD alone. The MafK (p18) fusion served as a positive control and exhibited the highest interaction activity. Gdbd-Nrf 2-(314 -581) contains the DNA interaction and dimerization (i.e. bZIP) domains but is transcriptionally inactive. Co-expression of AD did not stimulate luciferase activity suggesting that Nrf 2 self-interaction is limited to the N-terminal portion of Nrf 2. ATF3, ATF4, and p18 interacted with Gdbd-Nrf 2-(314 -581) with the following rank order: p18 Ͼ Ͼ ATF4 Ͼ Ͼ ATF3. Presumably these interactions reflect dimerization between leucine zipper domains. Relative to p18, ATF4 exhibits greater association with Gdbd-Nrf 2-(13-581) than with Gdbd-Nrf 2 (314 -581) (ϳ40% versus 5%), even though the latter is produced at higher levels intracellularly (data not shown).
ATF4, in Association with Nrf 2, Binds to the Stress-responsive Element (StRE)-To understand the consequence of ATF4/ Nrf 2 interaction on gene regulation, we initially examined the binding of ATF4 and Nrf 2, individually or as a mixture, to the StRE, a cis-acting element known to regulate inducer-mediated ho-1 gene activation in an Nrf 2-dependent manner (4,22,26). As expected, Nrf 2 did not bind to the StRE (Fig. 4). ATF4 alone also did not bind to the StRE but, in the presence of Nrf 2, exhibited significant binding. Presumably, this DNA-protein interaction reflects the activity of ATF4⅐Nrf 2 dimers. p18, which can form homodimers and highly stable heterodimers with Nrf 2 was used as a positive control for protein dimerization and DNA binding. p18 homodimers, which are known to interact with MARE and NF-E2 binding sites, also bound to the StRE, but the strongest binding was observed with p18⅐Nrf 2 FIG. 3. Interaction between ATF4 and Nrf 2 in mammalian two-hybrid assays. HeLa cells were plated (1 ϫ 10 5 /well of 6-well plates) 48 h prior to transfection by CaPO 4 -DNA co-precipitation. Cells in each well were transfected for 6 h with a DNA mixture consisting of 3 g of pFRluc, 1 g of pCMV/␤Gal, 2 g of the indicated Gdbd plasmid, and 2 g of an empty vector (EV) or the indicated activation domain (AD) plasmid. Cells were cultured for an additional 48 h and then lysed. Eight and 4% of the cell extracts were used to measure luciferase and ␤-galactosidase activities, respectively. ␤-Galactosidase-normalized luciferase activities are presented. A, luciferase activity obtained with the empty vector (EV), an average of 10,100 light units, was arbitrarily assigned a value of 1. Such standardization was not possible in B because no activity was detected in the absence of plasmids encoding interacting proteins. Each data bar represents the average Ϯ S.E. from 3 to 6 independent experiments.
FIG. 4. The Nrf 2⅐ATF4 heterodimer binds to the StRE. Protein synthesis and EMSA reactions were carried out as described under "Experimental Procedures." EMSA gels were exposed to x-ray film for 16 h. Lanes designated "ϪNrf 2" contained in vitro transcription/translation products from reactions directed by the empty expression vector pGEM2. Nonspecific complexes resulting from the transcription/translation extracts are marked with asterisks. 5. ATF4 enhances and p18 represses Nrf 2-mediated trans-activation of pE1-luc. Hepa cells were plated (5 ϫ 10 5 /well of 6-well plates) 24 h prior to transfection by CaPO 4 -DNA co-precipitation. Cells in each well were transfected for 6 h with a DNA mixture consisting of 2 g of pE1-luc, 1 g of pCMV-␤-Gal, 1 g of pEF/Nrf 2 or pEF/myc/mito, and the indicated amount of pCMV/rATF4 or pEF/ p18. Total DNA was equalized with an appropriate empty expression vector. Cells were cultured for an additional 48 h and then lysed. Eight and 4% of the cell extracts were used to measure luciferase and ␤-galactosidase activities, respectively. Luciferase activity was normalized to ␤-galactosidase activity in the same extract and is presented as a percentage of activity in cells transfected without pCMV/rATF4 or pEF/p18. Each data point represents the average Ϯ S.E. from three independent experiments. Average fold trans-activation by Nrf 2 (in the absence of ATF or p18) was 53.4. heterodimers. The relative affinities (and/or stability) of ATF4⅐Nrf 2 dimers and p18⅐Nrf 2 dimers for the StRE correlated with the relative affinities of protein-protein interactions observed in the mammalian two-hybrid assays. FosB, which cannot form homodimers and would not be expected to dimerize with Nrf 2, served as a negative control and did not exhibit specific binding in the absence or presence of Nrf 2. Unlike ATF4, recombinant ATF3, presumably ATF3 homodimers, bound weakly to the StRE, but the binding was decreased in the presence of Nrf 2.

FIG.
ATF4 Enhances, whereas p18 Inhibits, Nrf 2-mediated Trans-activation of the ho-1 Enhancer-Binding of ATF4⅐Nrf 2 dimers to the StRE suggested a role for ATF4 in ho-1 gene regulation. This potential function was investigated further by examining the ability of ATF4 to trans-activate the ho-1 enhancer, E1, in the reporter construct pE1-luc. In Hepa cells, co-transfection of an ATF4 expression plasmid, up to the maximum level tested, decreased basal pE1-luc expression by 20 -25% (Fig. 5). ATF4, however, had a synergistic effect on Nrf 2dependent pE1-luc expression, increasing luciferase activity up to 2-fold. In contrast, co-expression of p18 dramatically inhibited Nrf 2-mediated trans-activation of E1. This inhibition may, at least in part, be attributed to p18 homodimers as overexpression of p18 alone also inhibited basal pE1-luc expression.
Cadmium Stimulates ATF4 Expression-Previous studies from our laboratory (4,26) and other laboratories (7) have demonstrated the requirement for Nrf 2 in inducer-dependent ho-1 gene regulation. To determine the role, if any, of ATF4 in this process, we first examined the effect of cadmium, a known HO-1 stimulant, on ATF4 expression as earlier reports had suggested that ATF4 is a stress-response protein (30,31). As shown in Fig. 6, treatment of Hepa cells with 100 M CdCl 2 increased ATF4 levels in a time-dependent manner to greater than 10-fold above basal values. Of the other transcription factors tested, only expression of ATF3 and c-Jun, both documented stress-responsive proteins (32,33), was enhanced by cadmium. Interestingly, the temporal profile of c-Jun and ATF3 induction matched that of HO-1 accumulation with the earliest detectable enhancement observed at 2 h post-treatment. Increased expression of ATF4, on the other hand, was detected within 30 min after treatment of cells with CdCl 2 , even prior to the detectable accumulation of ho-1 mRNA by this agent in Hepa cells (34).
A Dominant-negative Mutant of ATF4 Inhibits Basal and Cadmium-stimulated E1 Activity in Hepa Cells-To establish further the role of ATF4 in ho-1 gene regulation, we generated a dominant-negative mutant of ATF4 and examined its effect on pE1-luc expression. Overexpression of the mutant ATF4 inhibited both basal and cadmium-stimulated luciferase activity in a dose-dependent manner (Fig. 7). This effect was qualitatively and quantitatively similar to that observed with an Nrf 2 dominant mutant. Interestingly, an analogous mutant of p18 inhibited cadmium-induced but not basal activity. Mutants of c-Jun or Jun D either enhanced or had no effect on pE1-luc expression.
The ATF4 Dominant-negative Mutant Does Not Inhibit Cadmium-stimulated pE1-luc Expression in MCF-7 Cells-We have recently reported that cadmium is a potent activator of the ho-1 gene in MCF-7 human mammary epithelial cells, stimulating ho-1 mRNA accumulation by 300 -400-fold, and that Nrf 2 is an important regulator of this response (26). Consistent with our previous finding, the Nrf 2 mutant significantly inhibited pE1-luc expression in the presence or absence of CdCl 2 (Fig. 8). The p18 mutant also inhibited both basal and cadmium-stimulated activities to similar levels. The ATF4 mutant, however, inhibited only basal luciferase activity, reveal-ing cell-dependent differences in the mechanism of ho-1 gene activation by cadmium and the role of ATF4 in this process.

DISCUSSION
In this study we have identified ATF4 as an Nrf 2-interacting protein and have provided evidence implicating ATF4 in basal and cadmium-induced regulation of the ho-1 gene, the latter presumably in cooperation with Nrf 2. Whereas association between Nrf 2 and ATF/CREB family members has not been previously documented, such an interaction is not necessarily unexpected as both classes of factors belong to the bZIP superfamily. In this regard it is noteworthy that among bZIP protein, ATF4 exhibits relatively promiscuous interaction activity. For instance, ATF4 forms heterodimers with c-Jun, c-Fos, and Fra-1 proteins under conditions where ATF2 and ATF3 heterodimerize only with c-Jun and ATF1 does not form any detectable heterodimers at all (17). In addition, ATF4 also forms heterodimers with CCAAT/enhancer-binding proteins (35), a relatively more distant subfamily (based on comparison of DNA-binding site sequences) than the Jun or Fos subfamilies within the bZIP superfamily.
Although not directly tested, it is not unreasonable to assume that Nrf 2 and ATF4 form classical bZIP dimers via their FIG. 6. Cadmium induces ATF3, ATF4, c-Jun, and HO-1 expression in Hepa cells. Approximately 5 ϫ 10 5 cells were plated in each well of a 6-well plate. Cells were cultured in complete medium for 48 h and, subsequently, in serum-free medium for 24 h. Cells were exposed to 100 M CdCl 2 for the indicated time (hours). Western blot analyses were carried out as described under "Experimental Procedures" using antibodies directed against the indicated proteins. leucine zipper structures. Indeed, in all likelihood, it is this type of interaction that will most effectively elicit the proper alignment of the basic regions necessary for sequence-specific DNA binding as demonstrated herein. This conclusion is supported by the fact that the C-terminal portion of Nrf 2 containing the bZIP domain was used in the original Y2H screening, and the observations that ATF4 exhibits positive interaction with a truncated Nrf 2 in M2H assays (Fig. 3) and in EMSA (Fig. 4). Based on the difference in the apparent relative affinities of ATF4 for nearly full-length Nrf 2 versus the truncated protein (see Fig. 3) and additional preliminary observations, we cannot rule out the possibility that ATF4 interacts with additional domains within the Nrf 2 polypeptide. In this regard, it is interesting that ATF3 interacts very poorly with the truncated Nrf 2 but is as effective as ATF4 in binding to the larger Nrf 2 protein. At present, the precise location of any of the interaction sites (within Nrf 2 or ATF4) is unknown. The functional significance of any non-leucine zipper associations is also not obvious. However, we note that ATF4 is known to associate with non-bZIP factors, including the Tax protein of the human T-cell leukemia virus type 1 (36,37) and the ␥-aminobutyric acid type B receptor (38). Interestingly, in both situations the interacting site within ATF4 was localized to the bZIP domain and, in the case of Tax, the association leads to enhanced trans-activation capacity for the Tax protein, similar to that observed here for Nrf 2.
The ho-1 gene is activated by a variety of stress-associated agents including the substrate heme, heavy metals, tumor promoters, UV irradiation, and inflammatory cytokines. Induction of the mouse gene by most stimuli is mediated by two 5Ј, distal enhancer regions, E1 and E2, each containing multiple StREs. The StREs are sufficient and necessary for inducer-dependent gene activation (reviewed in Ref. 12). Our recent analyses (4,26) have implicated Nrf 2 in the mechanism of ho-1 gene activation by several agents and in particular by cadmium. For instance, stable expression of a dominant-negative mutant of Nrf 2, but not of c-Jun, diminishes cadmium-induced ho-1 mRNA accumulation by 75-90% in L929 fibroblasts and MCF-7 cells. Similarly, in MCF-7 cells, overexpression of the Nrf 2 mutant inhibits induction of an E1-regulated luciferase reporter gene by cadmium, and mutants of E1 that are not trans-activated by Nrf 2 are also unresponsive to cadmium.
Identification of ATF4 as an Nrf 2-interacting protein led us to speculate that ATF4, possibly in cooperation with Nrf 2, regulates ho-1 gene expression. The transfection experiments demonstrating the inhibitory effects of the ATF4 dominant- negative mutant on basal and cadmium-induced E1 activity in Hepa cells provides support for this idea. It is important to point out that transfection studies of this nature, by themselves, do not conclusively demonstrate the role of a specific transcription factor in gene regulation. This limitation arises because of the tendency of bZIP proteins to dimerize with multiple partners. Consequently, inhibition of gene activation observed with a specific dominant mutant can be attributed not only to the corresponding endogenous protein (if capable of homodimerization) but also to any of its dimerization partners, one or more of which may be the actual effector protein(s).
Clearly, under such circumstances, it is important to obtain corroborative data for the role of a given factor in gene regulation. For ATF4, such correlative evidence includes the observations that ATF4⅐Nrf 2 dimers bind to the StRE, that ATF4 expression is enhanced by cadmium, and that ATF4 has a synergistic effect on Nrf 2 trans-activation of E1. The latter finding, in particular, distinguishes ATF4 from p18, which exhibits an inhibitory effect similar to that of MafG on Nrf 2 trans-activation of the ARE (5,15). In the case of Nrf 2, a role in ho-1 gene regulation has been further substantiated with the use of nrf 2-targeted mice (39) and cells (7). A similar analysis for ATF4 and p18 would of course be very informative.
Our contention that ATF4 in part regulates ho-1 gene expression is consistent with emerging data that indicate ATF4 is a stress-response protein and, consequently, would function as a regulator of the adaptive response to such stress. For example, anoxia, which stimulates HO-1 synthesis, strongly increases the expression and DNA binding activity of ATF4 in fibroblasts (30). Arsenite, another HO-1 inducer, also stimulates ATF4 DNA binding activity in pheochromocytoma PC12 cells (31). In addition, ATF4 levels are enhanced in endothelial (40,41) and Jurkat (42) cells in response to homocysteine and the calcium ionophore A23187, respectively, two agents known to cause endoplasmic reticulum stress. Finally, ATF4 expression is increased in cell lines resistant to various DNA-targeting drugs (43).
The difference between Hepa and MCF-7 cells with respect to the ability of the ATF4-dominant mutant to modulate cadmium-dependent E1 activity is puzzling but certainly points to cell-specific differences in the induction mechanism. Perhaps, intracellularly, productive or transcriptionally competent interaction between ATF4 and Nrf 2 requires an additional cofactor(s) which may be expressed in a cell-specific manner. This cofactor concept is somewhat similar to one postulated by Venugopal and Jaiswal (16) who have proposed that formation of Nrf 2-Jun complexes is dependent on one or more presently uncharacterized cytoplasmic proteins. One consequence of this hypothesis is that it apparently precludes a role for Nrf 2-ATF4 complexes in cadmium-dependent ho-1 gene activation in MCF-7 cells and, based on our previous studies noted above, requires the use of other Nrf 2-containing complexes. Given the tendency for bZIP proteins to form multiple and distinct associations, this requirement is not necessarily insurmountable. Studies are under way to characterize further the role of ATF4, particularly with respect to cell and inducer specificity, in the regulation of ho-1 and other Nrf 2 target genes.