IMPACT, a Protein Preferentially Expressed in the Mouse Brain, Binds GCN1 and Inhibits GCN2 Activation*

Translational control directed by the eukaryotic translation initiation factor 2 (cid:1) -subunit (eIF2 (cid:1) ) kinase GCN2 is important for coordinating gene expression programs in response to nutritional deprivation. The GCN2 stress response, conserved from yeast to mammals, is critical for resistance to nutritional deficiencies and for the control of feeding behaviors in rodents. The mouse protein IMPACT has sequence similarities to the yeast YIH1 protein, an inhibitor of GCN2. YIH1 com-petes with GCN2 for binding to a positive regulator, GCN1. Here, we present evidence that IMPACT is the functional counterpart of YIH1. Overexpression of IMPACT in yeast lowered both basal and amino acid star-vation-induced levels of phosphorylated eIF2 (cid:1) , as described for YIH1 (31). Overexpression of IMPACT in mouse embryonic fibroblasts inhibited

The control of protein synthesis plays an important role in diverse physiological conditions as part of homeostatic mechanisms and long-term memory formation and in pathological conditions such as diabetes, brain ischemia, epilepsy, and other neurodegenerative disorders (1)(2)(3)(4)(5)(6)(7)(8). At the cellular level, many signaling networks that affect the rate of protein synthesis involve the phosphorylation of Ser 51 in the eukaryotic translation initiation factor 2 ␣-subunit (eIF2␣) 1 by a family of protein kinases that are each activated by different cellular stress conditions. The heterotrimeric factor eIF2 is responsible for binding the initiator methionyl-tRNA i Met in a GTP-dependent mode and delivering it to the 40 S ribosomal subunit. When the initiator AUG codon is encountered, eIF2 is released in a GDP-bound form, with the subsequent formation of the 80 S elongating ribosome. The exchange of GDP for GTP on eIF2 (to allow for further rounds of initiation) is catalyzed by the guanine nucleotide exchange factor eIF2B. Phosphorylation of eIF2␣ can have a profound inhibitory effect on overall protein synthesis because phosphorylated eIF2␣ (eIF2␣(P)) is a competitive inhibitor of eIF2B, which is limiting in cells (9). Concomitant with this global translation inhibition, eIF2␣ phosphorylation can lead to preferential translation of mRNAs encoding stressrelated proteins. Thus, eIF2␣ phosphorylation can regulate both general and specific translation.
There are four known eIF2␣ kinases in mammals (reviewed in Ref. 10): GCN2, activated by amino acid starvation through the binding of uncharged tRNA to its regulatory region; PEK/ PERK, an endoplasmic reticulum transmembrane protein activated by endoplasmic reticulum stress; PKR, activated by double-stranded RNA produced during viral infection; and HRI (heme-regulated inhibitor kinase), present mainly in reticulocytes and activated by heme deprivation.
GCN2, the sole eIF2␣ kinase present in the yeast Saccharomyces cerevisiae (11), is found in mice in three isoforms differing only in their N-terminal sequences (12). The most abundant and ubiquitously expressed isoform contains all of the features of the yeast counterpart, including an N-terminal domain, which is required in vivo for activation of the kinase domain through its interaction with the activator GCN1 (13,14); a pseudo-kinase domain followed by the kinase region; a region with similarity to histidyl-tRNA synthetases, implicated in the recognition of uncharged tRNA, which then signals for activation of the kinase domain; and a C-terminal domain, involved in the interaction of GCN2 with the ribosomes (9,15,16). GCN1 is required for activation of GCN2, and it is thought that GCN1 acts as a chaperone to transport uncharged tRNAs that enter the A site of ribosomes to the tRNA-binding domain of GCN2 (13,17). GCN1 works in concert with GCN20 to form a complex with GCN2 on the ribosome, required for activation of the kinase (13,16,17).
In yeast, GCN2 is required for growth under amino acid starvation conditions. Its activation by high levels of uncharged tRNAs that accumulate under these conditions leads to eIF2␣ phosphorylation and thus to the translation of GCN4, a transcriptional activator of hundreds of genes involved in amino acid biosynthesis (18,19). Mammalian GCN2 has been shown to be required for adaptation to amino acid deprivation in mice and is activated under conditions of low availability of amino acids (20,21). Although there is no GCN4 ortholog in mammalian cells, the levels of a related transcriptional activator, ATF4, are induced by eIF2␣ phosphorylation through a mechanism of translation reinitiation similar to that described for yeast GCN4 (22,23). ATF4 (also known as CREB-2) enhances expression of additional basic leucine zipper transcriptional regulators (including CHOP/ GADD153 and ATF3) that together contribute to expression of a large number of genes involved in metabolism, redox chemistry, and apoptosis (21,24,25).
Mice lacking GCN2 are viable; however, the GCN2-deficient animals display aberrant translation in the liver, enhanced skeletal muscle loss, and increased morbidity in response to amino acid deprivation (26). Recently, GCN2 has been shown to be directly involved in the feeding behavior of mice. Phosphorylation of eIF2␣ in the anterior piriform cortex is observed immediately following intake of diets poor in essential amino acids (27). Interestingly, contrary to wild-type animals, which tend to avoid meals lacking even one of the essential amino acids, GCN2 Ϫ/Ϫ mice are deficient in this aversive behavior (28,29).
The GCN2 N-terminal domain has a sequence motif called the GI domain, which is also present in the N-terminal half of the yeast protein YIH1 and its mammalian ortholog, IMPACT (14). IMPACT was originally identified in a screen for imprinted genes in mice. Both YIH1 and IMPACT contain, in the C-terminal half, a conserved sequence (Ancient domain) found also in bacterial proteins (14,30).
Because of the similarity between the GI domains of YIH1 and GCN2, it has been proposed that YIH1 acts as an inhibitor of GCN2 activation mediated by GCN1 through competition with GCN2 for GCN1 binding. Indeed, recent data in yeast have demonstrated that the binding of GCN2 to GCN1 can be reduced by overexpression of YIH1 and that this leads to reduced eIF2␣ phosphorylation, indicative of impaired GCN2 activation (31). The in vivo evidence in yeast clearly indicate that YIH1 inhibits GCN2 activation. However, no condition was found in which this inhibitory action would be physiologically relevant to yeast cells because a deletion of YIH1 has no apparent phenotype. Because YIH1 was also found to bind to G-actin, it has been proposed that localized action of YIH1 in yeast cells might regulate the activity of GCN2 in regions where protein synthesis must be maintained at high levels, such as near the growing bud (31).
Given the relevance of GCN2 in mammalian metabolism and behavior and the involvement of eIF2␣ phosphorylation in several pathological conditions, we decided to study the function of IMPACT. We show here that IMPACT is the mammalian functional counterpart of YIH1. IMPACT binds to GCN1 and acts as an inhibitor of mouse GCN2. We also demonstrate that IMPACT is preferentially expressed in the brain in mice and is especially abundant in the hypothalamus. The levels of IM-PACT correlated inversely with the basal levels of eIF2␣(P) in all tissues examined. Our results strongly suggest that IM-PACT acts as an inhibitor of GCN2 in the mammalian brain. These findings have profound physiological implications in the control of phosphorylation of eIF2␣ and consequently in the expression of ATF4 in different brain areas and specific neuronal cells.

EXPERIMENTAL PROCEDURES
Yeast Methods-Standard yeast methods were employed (32). Yeast strain H1511 (MAT␣, ura3-52, trp1-63, leu2-3, leu2-11, GAL2 ϩ ) (33) was grown in synthetic complete medium lacking amino acids to select for plasmids and supplemented with 2% glucose or 10% galactose as carbon source. The plasmid encoding YIH1 fused to glutathione Stransferase (GST) for expression in yeast under the control of the galactose-inducible GAL1 promoter has been described previously (31). The plasmid encoding IMPACT fused to GST under the control of the GAL1 promoter was constructed by introducing BglII and HindIII sites by PCR into the cloned sequence of IMPACT present in plasmid pBE435 (see below) and cloning into the BamHI-HindIII sites of vector pES128-9-1 as described previously (13). The response of yeast strains to amino acid starvation was studied either by scoring for growth on solid medium lacking histidine and containing 3-aminotriazole (3-AT) at the concentration indicated or by growing cells for 4 h in liquid culture containing 30 mM 3-AT and measuring the levels of eIF2␣ phosphorylation by immunoblot analysis as described previously (31).
Animals-Male Swiss albino mice (20 -30 g) were decapitated, and brains and other tissues were removed as quickly as possible, washed with phosphate-buffered saline (PBS), and immediately processed.
IMPACT and mGCN1 Cloning and Protein Purification-A 1-kb DNA fragment comprising the complete open reading frame of IMPACT was obtained by RT-PCR from whole brain RNA using the primers described above. The IMPACT cDNA was inserted as an EcoRI-NotI fragment into the pET28a plasmid (Novagen), generating plasmid pBE435. The N-terminal His-tagged recombinant protein was expressed in E. coli Rosetta (DE3) cells grown in LB medium with 100 g/ml kanamycin and 100 g/ml chloramphenicol after induction with 0.1 mM isopropyl ␤-D-thiogalactopyranoside at 23°C overnight. Purification of the protein was performed using nickel-nitrilotriacetic acid (Ni-NTA; Qiagen Inc.) essentially as recommended by the manufacturer. Briefly, the cells were harvested, resuspended in lysis buffer (10% sucrose, 0.2 M NaCl, and 50 mM Tris-HCl (pH 7.5)), frozen, incubated with lysozyme (1 mg/ml) for 30 min on ice, and briefly sonicated. The cell lysates were centrifuged, and the supernatant was applied to Ni-NTA equilibrated with binding buffer (500 mM NaCl, 20 mM Tris-HCl (pH 8.0), and 5 mM imidazole). The column was washed with washing buffer (500 mM NaCl, 20 mM Tris-HCl (pH 8.0), and 20 mM imidazole). The His-tagged protein was eluted with elution buffer (500 mM NaCl, 20 mM Tris-HCl (pH 8.0), and 1 M imidazole) and dialyzed against 20 mM Tris-HCl (pH 7.5) and 10 mM 2-mercaptoethanol. The mGCN1 sequence (GenBank TM gi:51827542) encoding residues 2204 -2651 was obtained by RT-PCR and cloned as a 1.3-kb BamHI-XhoI fragment into plasmids pET28a and pGEX6p3. Expression was carried out in E. coli BL21(DE3) cells for the His-tagged protein and in DH5␣ cells for the GST fusion. Extracts were prepared as described above. The insoluble recombinant proteins present in the bacterial extract pellet were solubilized by 8 M urea. The His 6 -mGCN1 protein used for immunization was purified on Ni-NTA in the presence of urea as recommended by the manufacturer, followed by preparative SDS-PAGE and elution from the gel. The GST-mGCN1 protein was purified from the urea-solubilized pellet by preparative SDS-PAGE, followed by elution of the protein from the polyacrylamide gel slice.
Mouse Embryonic Fibroblast (MEF) Cell Transfection and Amino Acid Starvation Conditions-For overexpression of IMPACT in MEF cells, an EcoRI-NotI fragment encoding IMPACT was isolated from plasmid pBE435 and placed under the control of the cytomegalovirus promoter in plasmid pCI-neo (Promega), which contains the SV40 origin of replication, originating plasmid pBE514. MEF cells immortalized by the SV40 large T antigen were cultured in Dulbecco's modified Eagle's medium (BioWhittaker, Inc.) supplemented with 1 mM nonessential amino acids, 100 units/ml penicillin, 100 g/ml streptomycin, and 10% fetal bovine serum and transfected using Lipofectamine (Invitrogen) as described previously (25). The non-transfected and transfected MEF cells were subjected to amino acid starvation by culturing in Dulbecco's modified Eagle's medium without leucine (BioWhittaker, Inc.) for the indicated number of hours. Lysates were prepared as described, and equal amounts of proteins were analyzed by immunoblotting using antibodies specific to eIF2␣(P), CHOP, ATF4, or ␤-actin as described (25). Additionally, wild-type GCN2 ϩ/ϩ and GCN2 Ϫ/Ϫ MEF cells were subjected to leucine starvation, and lysates were analyzed by immunoblotting.
Preparation of Monospecific Antibodies against IMPACT and mGCN1-Polyclonal antibodies were produced by immunizing rabbits with purified recombinant His 6 -IMPACT and His 6 -mGCN1 proteins. Monospecific antibodies were obtained by incubating the immune sera with His 6 -IMPACT or GST-mGCN1 immobilized on nitrocellulose membrane. One milligram of purified protein was submitted to preparative SDS-PAGE and transferred to a Hybond-C membrane. A strip of the membrane containing the protein, as visualized by Ponceau staining, was blocked with 5% nonfat milk in PBS, followed by incubation with antiserum at a 1:10 dilution in PBS for 3 h. After washing with PBS, bound antibodies were eluted with 0.1 M glycine (pH 2.5) and 1 mM EGTA for 10 min. The pH was immediately neutralized by adding an equal volume of 0.1 M Trizma (Tris base).
Immunoblot Analysis-For immunoblot analysis of IMPACT expression and eIF2␣ phosphorylation, Laemmli sample buffer was added to the samples; and after boiling for 3 min, the proteins were separated 12% SDS-polyacrylamide gel and transferred to nitrocellulose membrane (Hybond-C Extra, Amersham Biosciences) at 1 A for 1 h using the buffer conditions described previously (34). The membrane was blocked with 5% nonfat milk in PBS (for anti-IMPACT antibodies) or in Trisbuffered saline (for anti-eIF2␣(P) antibodies; BIOSOURCE) and 0.1% Tween 20 for 1 h at room temperature, followed by overnight incubation at 4°C with anti-IMPACT antibodies (1:500 dilution in PBS, 5% nonfat milk, and 0.1% Tween 20) or anti-eIF2␣(P) antibodies (1:1000 dilution in Tris-buffered saline, 1% bovine serum albumin, and 0.1% Tween 20). After three washings with 0.1% Tween 20 in PBS or Tris-buffered saline, bound antibodies were detected with horseradish peroxidaseconjugated protein A (Amersham Biosciences) diluted 1:4000 in PBS or with horseradish peroxidase-conjugated goat anti-rabbit IgG (Santa Cruz Biotechnology, Inc.) diluted 1:2000 for anti-IMPACT and anti-eIF2␣(P) immunoblotting, respectively. After incubation for 1 h at room temperature and washing with PBS or Tris-buffered saline, the bound antibodies were detected using the ECL chemiluminescence system (Amersham Biosciences). After stripping the membranes by incubation with 100 mM 2-mercaptoethanol, 2% SDS, and 62.5 mM Tris-HCl (pH 6.7) at 50°C for 30 min, the same membrane was incubated with anti-actin antibodies (1:300 dilution; Sigma) or anti-eIF2␣ antibodies (1:1000 dilution; BIOSOURCE) and processed as described above. eIF2␣(P) levels in yeast whole cell extracts were determined as described (31). The conditions for immunoblotting of MEFs for the detection of ATF4, CHOP, eIF2␣(P), and ␤-actin were essentially as described previously (25).
Pull-down Assays-GST pull-down assays were performed as described previously (13). GST-yeast GCN1 (yGCN1)-(2052-2428) and GST-yGCN1-(2052-2428)(R2259A) fusion proteins and GST were purified from E. coli cells carrying plasmids pES123-B1, pES164-2A, and pGEX2T, respectively (31). Briefly, E. coli DH5␣ cells carrying the specified plasmid were grown in LB medium containing ampicillin (100 g/ml) to 0.8 A 600 nm , and the induction of recombinant protein expres-sion was carried out by incubation with 0.1 mM isopropyl ␤-D-thiogalactopyranoside at 30°C for 4 h. Cells were collected and resuspended in PBS, and the extract was prepared as described above. The GST fusion proteins were purified from the soluble fraction of the cell extract on glutathione-Sepharose (Amersham Biosciences) as recommended by the manufacturer. The purified proteins were dialyzed against 20 mM Tris-HCl (pH 8.0) and 1 mM EDTA. For the pull-down assays, the purified proteins (20 g) were immobilized on 20 l of glutathione-Sepharose beads and incubated with 500 g of brain extract prepared as described above in a total volume of 200 l in 30 mM Tris-HCl (pH 7.5), 50 mM KCl, 10% glycerol, 0.1 mM phenylmethylsulfonyl fluoride, and 5 mM ␤-mercaptoethanol for 2 h at 4°C. After washing with PBS, the beads were resuspended in 20 l of 4ϫ Laemmli sample buffer and boiled for 3 min, and proteins were separated on 12% SDS-polyacrylamide gel. After transfer to a Hybond-C membrane, the proteins were stained with Ponceau S. Western blotting using anti-IMPACT antibodies was performed as described above.
Co-immunoprecipitation of mGCN1 and IMPACT-Brain extracts (5 mg) were precleared by incubation with 20 l of protein A-agarose beads and 1 l of preimmune serum in the buffer described above for the preparation of extracts from mouse tissues. The supernatant was then incubated overnight at 4°C with 20 l of protein A-agarose beadbound anti-mGCN1 antibodies, an irrelevant IgG, or buffer only. The beads were washed three times with the same buffer, and the bound material was resolved by SDS-PAGE, followed by immunoblotting using anti-IMPACT or anti-mGCN1 antibodies.
Immunohistochemistry for IMPACT-Animals were deeply anesthetized with a thionembutal overdose (100 mg/kg) and perfused through the heart with 50 ml of saline, followed by 300 ml of 4% paraformaldehyde at 4°C. The brains were removed and cryoprotected in 30% sucrose in PBS for 24 h. Coronal sections (40 m thick) were collected, washed with PBS for 30 min, and incubated in blocking buffer (PBS containing 0.3% Triton X-100 and 0.5% normal goat serum (Vector Laboratories)) for 30 min, followed by incubation with anti-IMPACT monospecific antibodies (diluted 1:100 in blocking buffer) at room temperature for 24 h. The sections were then washed with PBS for 30 min, incubated with goat anti-rabbit biotinylated IgG (1:200 dilution) for 2 h, washed again with PBS for 30 min, and incubated with avidin-biotin complex (Elite ABC kit, Vector Laboratories) for 90 min. The bound antibodies were detected with a nickel-intensified diaminobenzidine tetrahydrochloride reaction.

Overexpression of IMPACT in Yeast Cells Causes a Gcn Ϫ
Phenotype-To determine whether IMPACT is functionally related to YIH1, we overexpressed this protein in yeast cells. Overexpression of YIH1 imparts a Gcn Ϫ phenotype to yeast cells (31), which can be identified by the inability to grow in the presence of 3-AT, an inhibitor of the HIS3 enzyme necessary for the biosynthesis of histidine. To grow in the presence of 3-AT, cells have to synthesize GCN4, a transcriptional activator of amino acid biosynthetic genes and salvaging of nutrients. GCN4 is under translational regulation and is synthesized under conditions that lower the amounts of active ternary complex, such as when eIF2␣ is phosphorylated. Thus, cells unable to phosphorylate eIF2␣ will not translate GCN4 and therefore will not overcome the inhibitory action of 3-AT, visible by impaired growth. To assess the effect of IMPACT on the ability of cells to grow in 3-AT, IMPACT was expressed in yeast as a fusion with GST under the control of the galactose-inducible promoter. Cells overexpressing GST-YIH1, also under the control of the galactose-inducible promoter, were used as a control. Cells overexpressing IMPACT displayed a Gcn Ϫ phenotype, i.e. impaired growth in the 3-AT-containing medium, as did cells overexpressing YIH1 (Fig. 1A). The levels of both proteins were similar, as shown by immunoblotting performed in extracts prepared from galactose-induced cultures using antibodies directed against GST (Fig. 1B).
The Gcn Ϫ Phenotype of IMPACT Overexpression in Yeast Is Due to Inhibition of eIF2␣ Phosphorylation-The Gcn Ϫ phenotype is an indication of a defect in eIF2␣ phosphorylation and GCN4 translational derepression, such as in cells with a deletion of the eIF2␣ kinase GCN2. We then investigated the levels of eIF2␣ phosphorylation in cells overexpressing IMPACT. There is a basal level of eIF2␣(P) even under optimal growth conditions, which is then significantly elevated when cells are grown under amino acid starvation conditions or elicited by the addition of 3-AT. As shown in Fig. 2, the basal levels of eIF2␣ phosphorylation were decreased in cells overexpressing IM-PACT (compare lanes 2 and 3 with lane 4), similar to the effect observed for YIH1 overexpression (compare lane 1 with lanes 2 and 3). IMPACT overexpression also hindered activation of GCN2 under starvation conditions (lanes 6 and 7 versus lanes 8 -10) to the same extent as did YIH1 overexpression, lowering eIF2␣ phosphorylation levels to approximately half of the eIF2␣(P) levels found upon GST overexpression. These data thus indicate that IMPACT inhibits GCN2 activation in yeast, probably by interacting with and sequestering yGCN1.
Overexpression of IMPACT in MEFs Inhibits GCN2 Activation upon Leucine Starvation-GCN2 is required for induced eIF2␣ phosphorylation and enhanced expression of ATF4 and its target gene CHOP in response to amino acid limitation in mammalian cells (21,25). To determine whether IMPACT inhibits mammalian GCN2, MEF cells were transfected with a plasmid expressing IMPACT under the control of the cytomegalovirus promoter. There were high levels of IMPACT in the transfected MEF cells as judged by immunoblot analysis compared with the cells carrying only the vector or the non-transfected cells (Fig. 3B). Activation of GCN2 was determined by phosphorylation of eIF2␣ upon incubation of the cells in medium lacking leucine. The non-transfected cells showed a significant increase in eIF2␣(P) that was the result of activation of GCN2, as illustrated by the observation that eIF2␣(P) levels were diminished in GCN2 Ϫ/Ϫ cells that were subjected to leucine starvation (Fig. 3A). Cells carrying the vector alone showed an increase in the basal levels of eIF2␣(P) in the presence of leucine compared with the non-transfected cells, suggesting perhaps that the transfection procedure and/or high replication rate of the plasmid activates an eIF2␣ kinase; however, activation of GCN2 upon leucine starvation was clearly evident in these cells. Although displaying the same amount of basal eIF2␣(P) as the vector-containing cells under non-starvation conditions, MEF cells overexpressing IMPACT clearly did not show the same increase in eIF2␣(P) levels upon leucine depletion (Fig. 3B). The signal for eIF2␣(P) remained constant throughout the 6 h of incubation without leucine. Given that eIF2␣(P) signals for the increased expression of ATF4 and CHOP, we also investigated whether overexpression of IM-PACT affects the levels of these two proteins. Consistent with an earlier report (25), translational induction of ATF4 expression contributed to early expression of this transcriptional regulator, within 1 h of leucine limitation in the wild-type or vector-transfected MEF cells. The levels of the ATF4 target gene CHOP were also induced in these cells within 3 h of leucine starvation. Notably, overexpression of IMPACT resulted in minimal production of ATF4 and CHOP upon amino acid starvation compared with the cells transfected with the vector (Fig. 3B). These results clearly show that IMPACT inhibits activation of the mammalian GCN2 stress pathway in response to nutrient deprivation.
IMPACT Binds Yeast and Mammalian GCN1-To determine whether IMPACT inhibits GCN2 by binding to GCN1, as does its yeast ortholog, we first addressed whether IMPACT could bind yGCN1. It has been shown that residues 2052-2428 of yGCN1 are sufficient for interaction with both yeast GCN2 and YIH1 (13). The region comprised by residues 2052-2428 of the yeast protein shows 34% identity and 61% similarity to the equivalent region in mGCN1. The mutation R2259A in this region has been shown to abolish the interaction of yGCN1 with both yGCN2 and YIH1 when present in the complete protein in vivo and in a GST-yGCN1-(2052-2428) fusion protein in vitro (13,31). Arg 2259 is identical and neighboring sequences are identical or highly similar in all sequenced orthologs, as shown in the alignment of Fig. 4A. We then used the purified GST-yGCN1-(2052-2428) fusion protein in pull-down experiments with extracts prepared from mouse brain (see below). As shown in Fig. 4B, brain IMPACT was immobilized on glutathione beads through interaction with GST-yGCN1-(2052-2428). Because only a small fraction of IMPACT present in the brain extract associated with the immobilized GST-yGCN1 protein (ϳ0.2-0.5%), we used also the yGCN1 mutant in which residue 2259 was altered from arginine to alanine (GST-yGCN1-(2052-2428)(R2259A)) to address the specificity of binding. This mutant protein was unable to bind to IMPACT FIG. 2. IMPACT overexpression in yeast inhibits GCN2. Shown are the results from immunoblot analysis of whole cell extracts prepared from the yeast strains described in the legend to Fig. 1 grown to exponential phase in medium containing galactose in the absence or presence of 30 mM 3-AT using antibodies directed against total eIF2␣ and eIF2␣(P). All lanes contained 10 g of total protein, except for the GST-expressing cells subjected to starvation conditions, for which 5, 10, and 20 g were used in the immunoblot analysis. (Fig. 4B). The small amount of binding observed for the wildtype GST-yGCN1-(2052-2428) protein could be due to the use of a fragment of a heterologous protein that may bind with low affinity to the endogenous IMPACT protein present in the extracts. Alternatively, it is possible that the native IMPACT protein is present in a large complex that may not be stable enough to be retained by the GST-yGCN1 beads.
To show that IMPACT interacts with mGCN1, we raised antibodies against part of the mGCN1 protein (residues 2204 -2651, containing the putative GCN2/IMPACT-interacting region) for use in a co-immunoprecipitation assay. The complete sequence of mGCN1 cDNA indicates a protein of 2806 residues, with a predicted mass of 307 kDa. These antibodies recognized a protein of ϳ280 kDa and, to a lesser degree, a slower migrating protein; the latter was also recognized by the preimmune serum. The smaller than predicted mass was also observed for the native yGCN1 protein. As shown in Fig. 4C, IMPACT was immunoprecipitated along with mGCN1 from brain extracts using the purified antibodies. mGCN1 and IMPACT were not detected in the control, in which unrelated purified IgG was used in the same amount as the purified anti-mGCN1 antibodies. Taken together, these results provide strong evidence that IMPACT binds specifically to GCN1.
A very small percentage of IMPACT was found associated with GCN1 in brain extracts as detected by this co-immunoprecipitation assay. This result is not surprising given that a more extensive interaction would probably block the amino acid starvation response mediated by GCN2. Interestingly, in the yeast model, the in vivo interaction between YIH1 and yGCN1 can be detected only when YIH1 is overexpressed (31). Thus, it is possible that the IMPACT-GCN1 complexes occur exclusively in a small population of neuronal cells that express high levels of IMPACT.
IMPACT Is Preferentially Expressed in the Hypothalamus-It has been previously shown through in situ and Northern hybridizations in mice that the IMPACT mRNA is preferentially expressed in the brain (30). Using highly specific antibodies directed against IMPACT raised in this study, we analyzed in detail the abundance of this protein in mouse tissues. Immunoblots of extracts obtained from several organs showed that the IMPACT pro-

FIG. 3. Inhibition of GCN2 activation by IMPACT in mammalian cells.
A, GCN2 ϩ/ϩ and GCN2 Ϫ/Ϫ MEF cells were subjected to leucine starvation conditions for the indicated number of hours (3 and 6 h) or to no stress (0 h), and immunoblot analyses were carried out to measure activation of the GCN2 stress pathway. B, non-transfected GCN2 ϩ/ϩ MEF cells or GCN2 ϩ/ϩ MEF cells transfected with the pCI-neo plasmid vector alone or with the plasmid expressing IM-PACT were grown in medium lacking leucine for the indicated number of hours (1, 3, and 6 h) or with no stress (0 h). Equal amounts of protein lysates were analyzed by immunoblotting using antibodies against IMPACT, CHOP, ATF4 eIF2␣(P), or ␤-actin.

FIG. 4. IMPACT binds yeast and mammalian GCN1.
A, conservation of GCN1 sequences. Shown is the alignment of the region of GCN1 around Arg 2259 (numbering relative to the yeast protein) (arrow) from Mus musculus (Mm; GenBank TM gi:51827542), Homo sapiens (Hs; gi:41149891), S. cerevisiae (Sc; gi:477122), Neurospora crassa (Nc; gi:32410355), and Arabidopsis thaliana (At; gi:5042415), with identical residues indicated in reverse boldface and conserved residues boxed. B, GST pull-down. Wild-type yGCN1-(2052-2428) (R2259) and mutant yGCN1-(2052-2428)(R2259A) (A2259) fused to GST and GST alone (20 g each) purified from E. coli were incubated with 20 l of glutathione-Sepharose beads and with 500 g of whole extracts from mouse brains. The proteins associated with the beads (100% of the bound material; pellet lanes) were subjected to SDS-PAGE and detected by immunoblotting with monospecific antibodies raised against IMPACT (upper panel) and by Ponceau staining (lower panel). The input lanes correspond to one-fiftieth of the extract used in the pull-down assay. C, co-immunoprecipitation. Protein A-agarose beads alone (Ϫ) or coupled to monospecific antibodies raised against mGCN1 or to an irrelevant IgG were incubated with total brain extract (5 mg). Half of the material bound to the beads was subjected to 12% SDS-PAGE, followed by immunoblotting with anti-IMPACT antibodies, and the other half was subjected to 6% SDS-PAGE, followed by immunoblotting with anti-GCN1 antibodies. The lanes containing the supernatants from the immunoprecipitations contained 50 and 100 g of total protein in the anti-IMPACT and anti-GCN1 immunoblots, respectively. tein was highly expressed in the brain (Fig. 5A), correlating with the previous data on the mRNA abundance. To determine IMPACT expression in different brain areas, immunohistochemistry was performed using mouse brain slices. As a control, we used nonspecific IgG purified from preimmune serum in the same concentration as the purified antibodies directed against IMPACT. As shown in Fig. 6, IMPACT was found to be expressed at high levels in some scattered neurons in several brain areas, as shown here for the hippocampus, where a few neurons showed strong labeling, and in the cortex, where layer II of the piriform cortex showed more intense labeling. Expression in other neuronal cells in these regions was comparatively very low. On the other hand, IMPACT was found to be highly expressed in the majority of neurons in several hypothalamic regions. Strong labeling was evident around the third ventricular region, including the paraventricular, dorsomedial, and posterior hypothalamic nuclei; the preoptic area; and the suprachiasmatic nuclei (SCN), among others. These results are in agreement with data obtained from the Mouse GNF Gene Expression Database (available at expression.gnf.org/cgi-bin/index.cgi), where the hypothalamus had 10 times more IMPACT mRNA compared with other organs and ϳ3 times more compared with other brain areas (35). These results were further confirmed and quantitated by Western blotting of extracts prepared from the cortex, hippocampus, and hypothalamus (Fig. 5B). The abundance of IM-PACT in the hypothalamus relative to the other brain parts was clearly evident.
The Hypothalamus Displays the Lowest Basal Levels of eIF2␣ Phosphorylation-Because our results showed that IMPACT overexpression inhibited mouse GCN2, we hypothesized that the high levels of IMPACT in the hypothalamus may inhibit endogenous GCN2, resulting in low basal levels of eIF2␣ phosphorylation in this brain region compared with other areas with low IMPACT expression. We then determined the ratio of eIF2␣(P) to total eIF2␣ by immunoblotting of extracts from the cortex, hippocampus, and hypothalamus. As shown in Fig. 7A, the basal levels of eIF2␣(P) were much lower in the hypothalamus, where high amounts of IMPACT were found, than in the hippocampus and cortex.
It was possible that the low eIF2␣(P) levels in the hypothalamus could be instead due to lower levels of GCN2 or GCN1. In the mouse transcriptome microarray data base (35), GCN2 was found to be equally expressed in all of the brain areas analyzed here. However, the microarray data did not differentiate among the three isoforms of GCN2 present in mice, all of them known to be expressed in the brain (12). The ␤-isoform has complete homology to other GCN2 homologs, including the region at the N terminus that may interact with GCN1. The ␣-isoform lacks the N-terminal 280 amino acid residues and is therefore not a target for GCN1 binding. The ␥-isoform starts at an amino acid corresponding to position 86 of ␤GCN2 and carries six additional residues in the N terminus. This isoform contains part of the GI domain. To investigate the presence of the ␤GCN2 isoform in the different brain areas, we performed RT-PCR using oligonucleotides specific for this isoform. Furthermore, GCN1 mRNA was not reported to be present in the transcriptome microarray data base (35), and we therefore carried out RT-PCR analysis for the GCN1 transcript in the different brain areas. The results indicate that both ␤GCN2 and GCN1 are expressed in the hypothalamus, cortex, and hippocampus (Fig. 7B). Thus, the different basal levels of eIF2␣(P) in these areas might be related to the levels of IM-PACT and thus to the differential basal activation of GCN2.
To provide another comparative analysis, we also quantitated the levels of eIF2␣(P) in the heart, where there is little IMPACT protein. Basal eIF2␣(P) levels were found to be very high in the heart compared with those in the hypothalamus (Fig. 7C), therefore providing further support for an inverse relationship between the abundance of IMPACT and eIF2␣ phosphorylation.

DISCUSSION
It has been recently shown that YIH1 regulates activation of the eIF2␣ kinase GCN2 in yeast through its interaction with GCN1 (31). In this study, we have provided in vivo evidence that IMPACT, the mammalian ortholog of YIH1, inhibits both yeast and mouse GCN2 by its ability to bind to GCN1. We suggest that IMPACT is a negative regulator of GCN2 in mammals.
As a prediction of this hypothesis, high levels of IMPACT should lead to low intrinsic activation of GCN2 in mouse tissues and therefore to lower basal levels of eIF2␣ phosphorylation, as observed in the yeast model. We were able to show that this correlation occurred in all of the mouse tissues tested. In the heart, where almost no IMPACT could be detected relative to the brain, high levels of eIF2␣(P) were found. In the hypothalamus, where IMPACT showed the highest expression, eIF2␣(P) levels were very low. Other eIF2␣ kinases or the activity of eIF2␣-specific phosphatases may also participate in establishing the basal levels of eIF2␣(P) in the different brain areas and organs studied. However, it is reasonable to assume that differences in eIF2␣ kinase activity among organs or cells FIG. 5. Preferential expression of IMPACT in the brain. Shown are immunoblots of extracts from the indicated organs (A) and brain areas (B) using anti-IMPACT antibodies (upper panels). ␤-Actin was used to normalize the amounts of total protein added to each lane by incubating the same filter with anti-␤-actin antibodies after stripping the first antibodies (lower panels). In B, different amounts of total protein were loaded (30, 15, 10, and 5 g, as indicated by the triangles above the lanes). The ratios of IMPACT to actin, plotted on the graph, were calculated using the values obtained from different total protein loads running in parallel, which showed a linear range of signal for both IMPACT and actin, obtained from at least three independent immunoblot analyses identical to the one shown in B, and normalized to the ratio obtained for the hypothalamus. Error bars indicate S.D. values. of a normal animal should rely heavily on GCN2 and PEK/ PERK, both of which are intrinsically related to metabolic regulation. Low glucose activates PEK/PERK and possibly GCN2, in analogy to yeast (36) and from the phenotypes of knockout animals suggesting an overlap of PEK/PERK and GCN2 in glucose sensing, whereas amino acid deprivation ac- FIG. 6. Differential expression of IMPACT in neuronal cells. Shown are the results from immunohistochemistry of the dentate gyrus area of the hippocampus (A and B), the piriform cortex (D and E), and the hypothalamus (G and H) using anti-IMPACT antibodies and from control immunohistochemistry of the hippocampus (C), piriform cortex (F), and hypothalamus (I) performed with identical amounts of nonspecific IgG isolated from preimmune serum by affinity purification on protein A-Sepharose. Quantification of these antibodies relative to anti-IMPACT antibodies was performed by dot blot using horseradish peroxidase-conjugated protein A, followed by detection with ECL. A, C, D, F, G, and I are shown in the same scale, with the scale bar in A corresponding to 300 m. B, E, and H are at a higher magnification, with the scale bar in B corresponding to 100 m. gcl, granule cell layer.
FIG. 7. IMPACT levels correlate inversely with eIF2␣(P) levels. Varying amounts of total extracts (30, 15, 10, and 5 g, as indicated by the triangles above the lanes) from the indicated brain areas (A) or organs (C) were analyzed for the levels of eIF2␣(P) by immunoblotting using antibodies specific to eIF2␣(P) (upper panels), and after stripping, the same filters were probed with antibodies against total eIF2␣ (lower panels). The ratios of eIF2␣(P) to eIF2␣ shown in the graphs were calculated by assigning a value of 1 to the ratio of the signal intensity of eIF2␣(P) to the signal intensity of eIF2␣ obtained for the hypothalamus. The signal intensity values were obtained within a linear range determined from the different amounts of total protein loads. The results represent data obtained from at least three independent immunoblots of different pools of extracts. Error bars indicate S.D. values. RT-PCR products were obtained from total RNA isolated from the cortex (lanes 1), hippocampus (lanes 2), and hypothalamus (lanes 3) using oligonucleotides specific for ␤-actin, ␤GCN2, and GCN1 (B). The first lane contained 100-bp ladder DNA size standards. tivates GCN2. Both mechanisms are necessary for maintaining homeostasis and should be constantly monitored. Thus, GCN2 activity can be considered as an important contributor to the levels of phosphorylated eIF2␣ in mammals under normal physiological conditions. The results shown here strongly suggest that, under amino acid starvation conditions, IMPACT has an important role in controlling the levels of eIF2␣(P) through GCN2 inhibition, which may be more relevant in specific populations of neuronal cells.
The hypothalamus is critically involved in the maintenance of homeostasis, such as the control of body temperature and the balance of fluids and energy, and it is constantly adjusting the organism's metabolism and behavior to its immediate needs. It is interesting to speculate that, due to the constant signaling required from neurons in the hypothalamus, protein synthesis must be maintained at constant high levels even under conditions in which GCN2 would be activated in other cell types, such as amino acid starvation. Thus, the mechanism of inhibition of GCN2 activation represented by overexpression of IMPACT may be important for the function of neurons in this area. Along these lines, it is possible that an inhibitor of PEK/PERK may also be overexpressed in this brain region. The SCN of the hypothalamus show the highest expression of IMPACT as determined from microarrays (35). We have not quantitated IMPACT or eIF2␣(P) in the SCN as a separate region in immunoblots, but immunohistochemistry analyses suggested elevated IMPACT levels in the SCN (data not shown). The SCN are involved in circadian rhythm determination and maintenance. The possibility that the SCN neurons have an even more stringent control over eIF2␣ phosphorylation is intriguing.
Recent findings concerning activation of GCN2 in the anterior piriform cortex upon feeding a low amino acid diet are not discrepant relative to our results showing high levels of IM-PACT in the piriform cortex. Upon close inspection of the anterior piriform cortex, IMPACT was found mainly in interneurons in layer II (data not shown), whereas in animals subjected to a low amino acid diet, only a few pyramidal neurons in the anterior piriform cortex show eIF2␣(P) labeling (27). Certainly, co-localization studies will be highly relevant.
YIH1 has been shown to bind to G-actin in yeast. It is not clear what role the interaction between YIH1 and actin plays in yeast cells. However, low actin levels lead to impairment of GCN2 activation, suggesting that the resulting larger pool of free YIH1 would interact with GCN1, preventing activation of GCN2 (31). We were not able to detect binding of ␤-actin to purified His 6 -IMPACT or GST-IMPACT added to brain extracts (data not shown). It is possible that the actin-binding site in this protein might be hidden in the conformation of the recombinant protein. Thus, the issue of IMPACT binding to actin in mammals should be further analyzed.
To the best of our knowledge, this work is the first to show that eIF2␣ is differentially phosphorylated in mammalian brain areas under normal conditions. The intrinsic levels of eIF2␣(P) may be important for determining expression of ATF4, a protein that is proposed to play pivotal roles in neurons. In addition to its role in the build up of a stress recovery program, ATF4 has been shown recently to negatively regulate long-term potentiation: transgenic mice expressing an inhibitor of ATF4 activity have enhanced hippocampus-based spatial memory and long-term potentiation, and ATF4 is related to the Aplysia CREB-2 memory suppressor gene (37). Expression of IMPACT in the hippocampus may be relevant for the attenuation of expression of ATF4. However, our observation that IMPACT was overexpressed exclusively in some interneurons in this region is surprising. Despite the wealth of information on the functional role of ATF4/CREB-2 in the biochemical pathways associated with learning and memory, there is no information regarding the distribution of ATF4 over different populations of hippocampal neurons. Thus, further understanding of the role played by IMPACT in regulation of the basal and amino acid starvation-induced levels of eIF2␣(P) (and consequently, in the levels of ATF4) in specific neuronal cells of the hippocampus may add new insights to the molecular mechanisms involved in long-term potentiation.