Delineation of a Negative Feedback Regulatory Loop That Controls Protein Translation during Endoplasmic Reticulum Stress*

Transient protein synthesis inhibition is an important protective mechanism used by cells during various stress conditions including endoplasmic reticulum (ER) stress. This response centers on the phosphorylation state of eukaryotic initiation factor (eIF)-2 (cid:1) , which is induced by kinases like protein kinase R-like ER kinase (PERK) and GCN2 to suppress translation and is later reversed so translation resumes. GADD34 was recently identified as the factor that activates the type 1 protein serine/threonine phosphatase (PP1), which dephosphorylates eIF-2 (cid:1) during cellular stresses. Our study delineates a negative feedback regulatory loop in which the eIF-2 (cid:1) -controlled inhibition of protein translation leads to GADD34 induction, which promotes translational recovery. We show that activating transcription factor-4 (ATF4), which is paradoxically translated during the eIF-2 (cid:1) -mediated translational block, is required for the transactivation of the GADD34 promoter in response to ER stress and amino acid deprivation. ATF4 directly binds to and trans-activates a conserved ATF site in the GADD34 promoter during ER stress. Examination of ATF4 (cid:2) / (cid:2) MEFs revealed an absence of GADD34 induction, prolonged eIF-2 (cid:1) phosphorylation, delayed protein synthesis recovery, and diminished translational up-regulation of BiP during ER stress. These studies demonstrate


Transient protein synthesis inhibition is an important protective mechanism used by cells during various stress conditions including endoplasmic reticulum (ER)
stress. This response centers on the phosphorylation state of eukaryotic initiation factor (eIF)-2␣, which is induced by kinases like protein kinase R-like ER kinase (PERK) and GCN2 to suppress translation and is later reversed so translation resumes. GADD34 was recently identified as the factor that activates the type 1 protein serine/threonine phosphatase (PP1), which dephosphorylates eIF-2␣ during cellular stresses. Our study delineates a negative feedback regulatory loop in which the eIF-2␣-controlled inhibition of protein translation leads to GADD34 induction, which promotes translational recovery. We show that activating transcription factor-4 (ATF4), which is paradoxically translated during the eIF-2␣-mediated translational block, is required for the transactivation of the GADD34 promoter in response to ER stress and amino acid deprivation. ATF4 directly binds to and trans-activates a conserved ATF site in the GADD34 promoter during ER stress. Examination of ATF4؊/؊ MEFs revealed an absence of GADD34 induction, prolonged eIF-2␣ phosphorylation, delayed protein synthesis recovery, and diminished translational upregulation of BiP during ER stress. These studies demonstrate the essential role of GADD34 in the resumption of protein synthesis, define the pathway for its induction, and reveal that cytoprotective unfolded protein response targets like BiP are sensitive to the eIF-2␣mediated block in translation.
When mammalian cells encounter physiological or chemical stresses that impinge upon the normal folding of proteins in the endoplasmic reticulum (ER), 1 rapid and complex signal transduction cascades, known collectively as the unfolded protein response (UPR), are put into play that are responsible for a number of cytoprotective measures (1)(2)(3)(4). If homeostasis of the ER cannot be re-established, apoptotic pathways are activated to destroy the damaged cell in order to ultimately protect the organism (5). The protective components of the UPR include the transcriptional up-regulation of ER chaperones and folding enzymes. These proteins bind to unfolded proteins that begin accumulating in the ER and prevent their aggregation. In addition, the increased expression of the chaperones allows for the rapid refolding of nascent secretory pathway proteins once homeostasis has been re-established. A second protective response is the transient block in protein synthesis (6). This limits the load of proteins being introduced into a hostile environment where they would be unable to fold. The inhibition of protein synthesis is not limited to ER proteins and is responsible for a third aspect of the response, which is the arrest of cells in the G 1 phase of cell cycle. The inhibition of the synthesis of the short lived cyclin D1 protein results in a rapid depletion of this protein and a concomitant cell cycle arrest (7). Finally, the up-regulation of proteins associated with proteasomal degradation helps to dispose of the accumulated unfolded proteins, further reducing the load on the ER (8,9).
In the past 5 years, many components of the signaling pathways controlling these cytoprotective responses have been described. The ER-localized transmembrane protein, ATF6, encodes a cytosolic transcription factor tethered to an ER luminal "stress sensing" domain (10). ER stress allows ATF6 to travel to the Golgi where it is cleaved by the S1P and S2P proteases, thus liberating the cytosolic transcription factor domain (11,12). Cleaved ATF6 binds to ER stress response elements in the chaperone promoters and up-regulates their transcription. In addition, three transmembrane ER kinases have been identified that also possess luminal "sensing" domains in addition to their cytosolic kinase domains. Two of the kinases, Ire1␣ and ␤ (13,14), are orthologues of the sole ER stress transducer in yeast, Ire1p. Like yeast Ire1p, these proteins possess a unique C-terminal endonuclease domain, which upon activation of the kinase domains targets a specific RNA sequence. Most recently, the XBP-1 transcription factor was shown to be the target of endonuclease activity of the mammalian Ire1 (15)(16)(17). The Ire1-catalyzed removal of 26 nucleotides from XBP-1 mRNA followed by a religation reaction alters the XBP-1 reading frame. The remodeled XBP-1 protein encodes a novel transcription factor with the original DNA binding domain tethered to a new transactivation domain. The targets of spliced XBP-1 are not clear but may be involved in ER-associated degradation (18). The ATF6 transcription factor up-regulates XBP-1 mRNA providing a convergence of the Ire1-ATF6 pathways.
The third ER stress-transducing kinase is pancreatic eIF-2␣ kinase (PEK)/PERK (19,20), which shares homology with the luminal domains of Ire1␣ and -␤ but does not encode an endonuclease domain. The target of the kinase activity of PERK is the eucaryotic translation initiation factor, eIF-2␣. Phosphorylation of eIF-2␣ during ER stress interferes with the assembly of the translation initiation complex, thereby dramatically diminishing translation (21,22). This aspect of the UPR is shared by a number of other cellular stresses, which activate different eIF-2␣ kinases. Interestingly, although most cellular protein synthesis is shut down, translation of the ATF4 transcription factor is specifically induced. ATF4 is constitutively transcribed in non-stressed cells, even though its transcripts cannot be efficiently translated due to the presence of short open reading frames (5Ј-ORF) in its 5Ј-untranslated region (5Ј-UTR) that interfere with initiation at the proper start codon. During ER stress, these short ORFs are no longer utilized which leads to a dramatic increase in ATF4 protein synthesis (23). One target of the ATF4 transcription factor is the CHOP promoter (24,25), which appears to play a role in the induction of apoptosis during ER stress. Unlike the up-regulation of ER chaperones, the inhibition of protein synthesis during the UPR and other cellular stresses is transient. This is accomplished by inhibition of the PERK kinase itself and dephosphorylation of eIF-2␣. Very recently, it was demonstrated that a constitutively expressed double-stranded RNA-activated protein kinase (PKR) inhibitor, p58 IPK , is induced during ER stress and plays an important role in down-regulating PERK phosphorylation later during ER stress (26,27). GADD34 was originally identified in a screen for genes induced during growth arrest and DNA damage (28). Later study reveals that the GADD34 protein directly interacts with the catalytic subunit of type 1 protein serine/threonine phosphatase (PP1) and activates the ability of PP1 to dephosphorylate eIF-2␣, allowing most protein synthesis to resume (29 -31). The transcription induction of GADD34 during ER and other cellular stresses is dependent on the presence and activity of the respective eIF-2␣ kinases (29). The mechanism of GADD34 induction during cellular stress is not presently known.
In this study, we demonstrate that the translational induction of ATF4 in response to eIF-2␣ phosphorylation is immediately upstream of GADD34 induction. ATF4 null MEFs are unable to induce GADD34 mRNA during ER or other cellular stresses, which leads to prolonged eIF-2␣ phosphorylation and diminished protein synthesis recovery. ATF4 is able to bind specifically to and trans-activate a conserved ATF site in the GADD34 promoter in a stress-dependent manner. Through the chromatin immunoprecipitation assay, we further demonstrate that ATF4 directly binds to the GADD34 promoter region encompassing the ATF site in an ER stress-dependent manner. Together, our study delineates an autoregulatory loop in which PERK (or presumably other stress-specific eIF-2␣ kinases) phosphorylates eIF-2␣, in turn inhibits protein synthesis, and allows ATF4 translation. This then leads to GADD34 induction, eIF-2␣ dephosphorylation, and finally the resumption of protein synthesis. Furthermore, our data revealed that BiP up-regulation at the translational level is significantly diminished in ATF4 null cells where protein synthesis inhibition persists, demonstrating that the cytoprotective UPR targets are not immune to the eIF-2␣-dependent translational inhibition and thus underscore the importance of translational recovery early in the response.

EXPERIMENTAL PROCEDURES
Isolation, Genotyping, Culture, and Treatment of WT and ATF4 Null Mouse Embryonic Fibroblasts-Our ATF4 heterozygous mice were generated by Tanaka et al. (32) and were further maintained by Dr. Adrian F. Gombart, Cedars-Sinai Medical Center, who kindly provided our breeding pairs. In keeping with previous reports, homozygous ATF4 null mice are born at significantly less than expected Mendelian ratios and are blind due to the lack of lens development (32,33). Mouse embryos were explanted at day 13.5-14.5 of gestation from ATF4 ϩ/Ϫ animals to obtain embryonic fibroblasts (MEFs) (34). MEFs were propagated in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum, 2 mM glutamine, 0.1 mM nonessential amino acids, 55 M 2-mercaptoethanol, and 10 g/ml gentamicin (Invitrogen).
Both the amniotic membrane of the embryo and the cultured MEFs were genotyped as illustrated before (32) with primer pair "a" and "b" for the WT allele and primer pair "b" and "c" for the mutant allele. a is 5Ј-AAAGGAATGCTCTGGAGTGGAAGACAG; b is 5Ј-AAGCAAAGCT-AAGCCTCCATCTTGTGC; and c is 5Ј-ATCGCCTTCTATCGCCTTCTT-GACGAG. Passage 5-7 MEFs were plated for experiments and were left untreated, treated with Tg (1 M), or cultured in media that lacked leucine for the indicated periods of time (components of the minus leucine media are based on Dulbecco's modified Eagle's medium (Invitrogen 11960) with only leucine left out).
Isolation and Analysis of Cytoplasmic RNA-Cytoplasmic RNA was harvested using the Qiagen RNeasy miniprep kit following the protocol provided by the supplier. 1 g of RNA was used to generate 50 l of cDNA using a poly(dT) primer (1 mM). Semi-quantitative PCR was performed with the fail-safe PCR system from Epicenter using 2 l of cDNA, 0.25 mM of each primer pair, 2.5 units of Taq DNA polymerase, and 33 nM [␣-32 P]dCTP (Amersham Biosciences) for 18 -20 cycles. The PCR products were run on a 4% TBE-acrylamide gel, dried, and exposed to film. For quantification data, each radioactive band was quantified using an Amersham Biosciences PhosphorImager and normalized to the level of ␥-actin transcript measured from the same sample. The linear range was monitored by using serial dilutions (1:5 and 1:25) of the sample that showed the highest radioactive signal in each experimental group. The primer pair for mouse ␥-actin was 5Ј-CCGACGGG-CAGGTGATCAC (upstream) and 5Ј-GAGCAGTTAACTTGAATACA-AGG (downstream) and for mouse BiP (GRP78) was 5Ј-GATTCCAAG-GAACACTGTGGTA (upstream) and 5Ј-CAGTAAACAGCCACTTGGGC (downstream). 20 g of cytoplasmic mRNA from each experimental group was run on a Northern blot as described previously (35). The probe for BiP was an 800-bp piece amplified from mouse cDNA with oligonucleotides mentioned above. The probe for ATF4 was a 600-bp piece amplified from mouse cDNA with oligonucleotides 5Ј-CCAGGGGTTCTGTCTTC-CAC and 5Ј-AGCAAACACAGCAACACAAGCA. The probe for GADD34 was an 825-bp piece amplified from mouse cDNA with oligonucleotides 5Ј-CCCAGACACATGGCCCCG and 5Ј-TTGTCTCAGGTCCTCCTTCC. The probe for GAPDH is a human fragment purchased from Clontech (Palo Alto, CA).
Metabolic labeling and immunoprecipitation were performed as described previously (37). In brief, WT or ATF4Ϫ/Ϫ MEFs were labeled for 30 min with [ 35 S]methionine and -cysteine (Translabel, ICN, Irvine, CA). Cells were then harvested, lysed directly in Nonidet P-40 lysing buffer, and immunoprecipitated with polyclonal rabbit anti-BiP antiserum before electrophoresing on a 10% denaturing SDS gel. 35 S signals were quantified using a Amersham Biosciences PhosphorImager.
Nuclear Extract Purification, Electrophoretic Mobility Shift Assay, and Supershift Assay-WT or ATF4 null MEFs treated with Tg (1 M) for the indicated times were harvested and washed with phosphatebuffered saline twice before nuclear extracts were purified as described (24). The EMSA reaction was performed as described previously (24) with 10 g of nuclear extract, 1 g of dI-dC, 10-fold excess of cold mutant oligonucleotides, and [␥-32 P]ATP-labeled probe (10,000 cpm) or 2 g of supershift antibody when indicated (polyclonal rabbit anti-C/ EBP␤ and anti-ATF4 antisera were purchased from Santa Cruz Biotechnology). The double strand synthetic oligonucleotide probes were annealed and end-labeled with [␥-32 P]ATP as described (24). A 29-bp CHOP C/EBP-ATF WT probe was designed as described previously (24). Similarly, a 29-bp sequence containing Ϫ77 to Ϫ53 bp of the mouse GADD34 promoter (5Ј-GATCGCTCTCCGGTGACGTCAGCACAGCA) was designated mGADD34-ATF. For the mutant oligonucleotides, the GTGACGTCAG sequence of the WT probe containing the ATF site was altered to acttatgtca.
Chromatin Immunoprecipitation (CHIP) Assay-The CHIP assay was performed with the CHIP kit from Upstate Biotechnology, Cell Signaling Solutions (Lake Placid, NY), and according to the protocol provided by Upstate Biotechnology with the following modifications.
Instead of using whole cell lysates, nuclear extracts were harvested from non-stressed and Tg-treated (1 M, 6 h) NIH3T3 cells using method described before (24). The cross-linked chromatin-protein complexes were sonicated to obtain chromatin fragments with an average length of ϳ600 -700 bp. Polyclonal rabbit anti-YY-1 antibody was used in the negative control immunoprecipitation, and a polyclonal rabbit anti-ATF4 antibody was used for ATF4 immunoprecipitation (both from Santa Cruz Biotechnology). Two targeted sequences were amplified by 25 cycles of PCR in the presence of 33 nM [␣-32 P]dCTP. The primer pair used to amplify region A of the mouse GADD34 promoter was 5Ј-GCT-CGGAAATTACGTGAGATCG and 5Ј-GCGCCAACATCGTCCACGCG. The pair for the amplification of region B was 5Ј-GGGCAGCCAGGGC-TATACTG and 5Ј-GGTGTACAAGGAGGTCGGAAG.
Chloramphenicol Acetyltransferase Assay-Sequences encompassing Ϫ721 to ϩ130 bp of the mouse GADD34 promoter was cloned into a CAT reporter vector using PCR-based method as mentioned before (24). The ATF site (TGACGTCA) alone was deleted in the mutant promoter construct with the Stratagene Quickchange TM site-directed mutagenesis kit (Cedar Creek, TX). 1 g of WT or mutant construct was transfected into COS-1 cells, and cell lysates were analyzed for CAT activity as described before (24). Each experimental group was analyzed in triplicate, normalized against protein concentration, and then repeated twice. The error bars shown in Fig. 6D were calculations of standard deviations for each triplicate experiment.

Isolation and Analysis of ATF4 Null Mouse Embryonic Fibroblasts-Recently
, the stress-induced GADD34 protein was shown to bind to the PP1 phosphatase and activate the dephosphorylation of eIF-2␣, thereby triggering the resumption of protein translation during cellular stresses (29 -31). Induction of GADD34 during ER stress and amino acid deprivation stress did not occur in PERKϪ/Ϫ or GCN2Ϫ/Ϫ cells, respectively, which are defective in the initial block in protein synthesis during ER stress or amino acid deprivation (29). This suggests the presence of a feedback regulatory loop, in which the eIF-2␣ kinases block protein synthesis and by an unknown mechanism induce GADD34, which then allows translation to resume. Ron and co-workers (29) suggested that this might occur either via a mechanism similar to the induction of ATF4 where small 5Ј-ORFs that interfere with correct translation initiation are suppressed by the inhibition of protein synthesis thus allowing initiation at the correct site or, conversely, that GADD34 might be a target of the ATF4 transcription factor. To address the role of ATF4 in the activation of GADD34 directly, we isolated WT and ATF4 null mouse embryonic fibroblasts (MEFs) by mating ATF4 ϩ/Ϫ mice. To confirm the lack of ATF4 transcription and translation in these cells, cytoplasmic mRNA and whole cell lysates were harvested from WT and ATF4Ϫ/Ϫ cells that were left untreated or treated with Tg to induce ER stress. The ATF4 transcript was detected using Northern blotting (Fig. 1B), and ATF4 protein expression was detected via Western blotting (Fig. 1A). ATF4 message was detected only in the WT MEFs but not in the ATF4 null MEFs (Fig. 1B). Consistent with published data, ATF4 was constitutively transcribed but not translated in WT cells under non-stressed conditions. ATF4 protein, as detected by Western blotting, was induced in WT MEFs as early as 2 h after Tg treatment but could not be detected even after 8 h of UPR activation in ATF4 null cells (Fig. 1A). When nuclear extracts harvested from either WT or ATF4Ϫ/Ϫ cells were added in gel shift assays, ATF4 binding was detected only with those from WT cells (Fig. 6C). These data confirm the lack of ATF4 transcripts and protein in our null MEFs both before and after ER stress. The targeted disruption of the ATF4 allele was generated by replacing the second exon of ATF4 gene containing the basic leucine zipper domain, which is essential for the dimerization of ATF4 and DNA binding ability, with the neomycin resistance gene (32). Thus, even if a partial ATF4 protein was expressed that was not recognized by our antibody, it would not be functional due to the lack of the leucine zipper domain.
Induction of BiP Transcription Was Intact in ATF4Ϫ/Ϫ Cells-To begin characterizing the ER stress response in the ATF4 null cells, we first analyzed the transcriptional up-regulation of the ER chaperone BiP. BiP transcription is a target of the ATF6 transcription factor (38) and may be partially regulated by Ire1-altered XBP-1 (15,39) and thus represents a read-out of these arms of the UPR. We harvested cytoplasmic mRNA from WT as well as ATF4Ϫ/Ϫ cells that had been treated with Tg for the indicated times, and we performed both Northern blotting and semi-quantitative reverse transcriptase-PCRs to detect the BiP transcript (Fig. 1, B-D). As early as 2 h after Tg treatment, an increase in BiP transcripts could be detected in both cells, with BiP mRNA levels continuing to build throughout the time course (Fig. 1B). Results from three independent reverse transcriptase-PCRs were quantified, normalized to the actin signal, and summarized in Fig. 1D, demonstrating that ATF4 null MEFs responded to Tg treatment by inducing BiP transcripts equally well as their wild type counterparts. This verifies that both WT and ATF4 null cells experienced and responded to ER stress as expected and suggests that at least the ATF6-ER stress response element signaling pathway is intact in the null cells. Furthermore, our data directly demonstrate that ATF4 is not required for the induction of BiP transcription during ER stress.
Prolonged eIF-2␣ Phosphorylation in ATF4Ϫ/Ϫ Cells-To determine whether the PERK activation signal was intact in these cells, we monitored the phosphorylation status of its target protein, eIF-2␣, during the course of ER stress using an antiserum that is specific for the phospho-form. WT and ATF4Ϫ/Ϫ cells were treated with Tg for the indicated periods of time, and protein blots were analyzed ( Fig. 2A). The level of total eIF-2␣ was also measured as a loading control. Phosphorylation of eIF-2␣ peaked from 0.5 to 1.5 h after Tg treatment in both WT and ATF4Ϫ/Ϫ cells, although that in the null cells was higher (Fig. 2B). As expected in the WT MEFs, eIF-2␣ phosphorylation started to decline 2-4 h after Tg treatment until it reached near basal levels. However, the level of eIF-2␣ phosphorylation in ATF4 null MEFs underwent a slower decline and remained comparable with the maximal level in WT cells until 8 h after ER stress. Together, our data demonstrated that even though the UPR-activated eIF-2␣ phosphorylation signal was intact in ATF4 null cells, the eIF-2␣ dephosphorylation signal was diminished. This suggested that ATF4 was required for the dephosphorylation of eIF-2␣ later in the ER stress.
Lack of GADD34 Induction in ATF4Ϫ/Ϫ Cells during Both ER Stress and Amino Acid Deprivation-Recently, Novoa et al. (29,31) demonstrated that GADD34, also known as Myd116 in mouse, promotes the dephosphorylation of eIF-2␣ and is induced by both ER stress and amino acid deprivation. Because eIF-2␣ phosphorylation is prolonged in the absence of ATF4 synthesis, we determined if ATF4 played a role in the regulation of GADD34 transcription. We harvested cytoplasmic mRNA from cells treated with Tg for the indicated periods of time and then performed Northern blot to measure the induction of GADD34 mRNA in WT and ATF4Ϫ/Ϫ cells (Fig. 2C). Similar experiments were also performed using cells deprived of leucine for the indicated times (Fig. 2D). In both cases, the induction of GADD34 transcripts was readily detected in the WT MEFs. However, treatment of the ATF4Ϫ/Ϫ MEFs with either stress-inducing regimen induced GADD34 mRNA to a FIG. 2. ATF4 was required for the dephosphorylation of eIF-2␣ and the induction of GADD34 during cellular stresses. A, cell lysates from WT and ATF4 null MEFs treated with Tg for the indicated times were run on a denaturing SDS gel. Phosphorylated (top panels) and total eIF-2␣ (bottom panels) were detected with respective antibodies. The amount of eIF-2␣ phosphorylation was then quantified, normalized to that of the total eIF-2␣, and summarized in B. C and D, mRNA was isolated from WT or ATF4Ϫ/Ϫ cells that had been treated with Tg (B) or deprived of leucine (C) for the indicated times and run on a Northern blot to detect the levels of GADD34 and GAPDH transcripts. minimal level. Hence, ATF4 is required for the up-regulation of GADD34 during both ER and amino acid deprivation stresses. This presumably explains the persistence of eIF-2␣ phosphorylation in the ATF4 null MEFs during ER stress (Fig. 2B).
The Resumption of Translation after ER Stress Was Delayed and Incomplete in ATF4Ϫ/Ϫ MEFs-To determine the consequences of prolonged eIF-2␣ phosphorylation in the ATF4Ϫ/Ϫ MEFs, we examined its effect on total protein synthesis. Cells were pulse-labeled with [ 35 S]methionine at different time points after Tg treatment as indicated in Fig. 3A. Total cell lysates were directly run on SDS protein gels, which were Coomassie-stained and exposed to film to detect the incorporated 35 S radioactivity. The 35 S signal was then quantified, and the result is summarized in Fig. 3B. In both WT and mutant MEFs, total protein synthesis diminished very quickly (within 1 h) in response to Tg treatment (Fig. 3B). Protein synthesis began to resume in the WT cells after 2 h of treatment, consistent with the kinetics of eIF-2␣ phosphorylation and dephosphorylation detected earlier (Fig. 2B). In the ATF4 null MEFs, protein synthesis diminished immediately after Tg treatment as in WT cells but did not recover with the same kinetics or to the same level as observed with the WT MEFs (Fig. 3B). To examine directly the effect of prolonged eIF-2␣ phosphorylation on a UPR target whose transcriptional up-regulation is not influenced by the absence of ATF4 induction, we measured the synthesis of BiP protein in the two MEFs during ER stress. We immunoprecipitated BiP protein from cells that had been pulse-labeled with 35 S to accurately detect the rate of BiP synthesis at different time points after UPR activation (upper panel of Fig. 3C). An equal amount of each whole cell lysate was run on a separate gel as a loading control (lower panel of Fig.  3C). Although the rates of BiP synthesis before ER stress were comparable in the two MEFs, BiP protein induction during the course of ER stress was almost completely abolished in cells lacking ATF4 expression (Fig. 3D). This block in translation was observed even though the transcriptional up-regulation of BiP in ATF4 null MEFs was similar to that in WT MEFs (Fig.  1, B-D). The synthesis of BiP in WT cells actually decreased 30 min after ER stress, and a significant induction in synthesis did not occur until nearly 4 h after ER stress despite an increase in BiP transcripts at earlier time points, which further demonstrates the sensitivity of BiP synthesis to eIF-2␣ phosphorylation. Thus, the lack of GADD34 induction and the resultant prolonged phosphorylation of eIF-2␣ dramatically impacted on the resumption of protein translation as specifically observed with BiP synthesis.
Identification of a Potential ATF4-binding Site That Is Conserved in the Mouse and Human GADD34 Promoter-The ATF4dependent up-regulation of GADD34 transcription during ER and other cellular stresses could occur either directly via the binding of ATF4 to the GADD34 promoter or indirectly as a consequence of the induction of other ATF4 target genes. Because a full-length GADD34 promoter sequence has not been reported and only ϳ200 bp of human (U83982) and mouse (U83984) GADD34 promoter sequences had been entered in the EST data base, we utilized the published human GADD34

FIG. 3. Persistent eIF-2␣ phosphorylation during the course of ER stress in ATF4 null cells blocked translation recovery and BiP induction.
A and B, WT and ATF4Ϫ/Ϫ MEFs were treated with Tg for the indicated times and then pulse-labeled with [ 35 S]methionine. Lysates were prepared, and a portion of each was run on SDS gels, Coomassie-stained, and exposed to film as shown in A. The total 35 S signal for each lane was quantified and summarized in B with that for the untreated WT MEFs set as 1. C, WT and ATF4Ϫ/Ϫ cells were treated as indicated and labeled with [ 35 S]methionine as in A. The lysates were then immunoprecipitated with a polyclonal antiserum specific for rodent BiP, separated by denaturing SDS-PAGE and detected by autoradiography. An equal amount of total cell lysates for each experimental group was run on a separate SDS gel and Coomassie-stained as loading control. The 35 S signals for the immunoprecipitated BiP protein at different time points after ER stress were then quantified and summarized in D.
cDNA (NM014330) and mouse Myd116 cDNA (NM008654) sequences to search the Ensembl genomic sequence data base at www.ensembl.org. We were able to identify the putative mouse and human GADD34 promoter sequences immediately upstream of the respective exon 1 sequences located on chromosome 7 in mouse and chromosome 19 in human. Both sequences contained the respective ϳ200-bp promoter sequences entered in the EST data base. When the first 1-kb upstream sequences of both promoters were aligned, we observed stretches of homology across the whole sequence (ϳ74%). However, if more stringent parameters were used in the comparison, only about 210 bp of the promoter sequences immediately upstream of the transcription start site can be aligned, revealing a higher homology in this region.
ATF and cAMP-response element-binding protein family members form homodimers or heterodimers to bind to a conserved ATF/cAMP-response element motif, TGACGTCA (40). Only one completely conserved ATF-binding site was identified in the mouse and human sequences. The ATF site is located in a stretch of highly conserved nucleotides and is just 20 bp upstream of the TATA box (Ϫ67 bp to Ϫ60 bp in mouse GADD34 promoter and Ϫ58 bp to Ϫ51 bp in human GADD34 promoter, Fig. 4). There is a stretch of 14 nucleotides just upstream of the GADD34-ATF site that is also completely conserved between the mouse and human GADD34 promoter (underlined in Fig. 4). This is the longest completely conserved sequence between the two GADD34 promoters, suggesting that it might also contain regulatory motif that is important for GADD34 transcriptional regulation during cellular stress conditions. Closer inspection of this motif revealed potential binding sites for both CHOP and NF-B. Further study is required to examine the importance of this conserved motif.
ATF4 Bound to the GADD34 Promoter in an ER Stress-dependent Manner-To examine whether ATF4 directly binds to the GADD34 promoter upon ER stress, we performed CHIP assays using an ATF4 antibody that recognizes epitopes that are exposed while ATF4 binds to its target sequence in gel shift assays (24). The YY-1 antibody that recognizes YY-1 binding to the BiP ER stress response element site (data not shown) was used as a negative control. Cross-linked chromatin-protein complexes prepared from untreated or Tg-treated NIH3T3 cells were fragmented to an average of 600 -700 bp (Fig. 5B). One target sequence designated region A encompasses the ATF site that is immediately upstream of the TATA box and contains the most highly conserved region (210 bp) between the human and mouse GADD34 promoters (Fig. 5A). The second target sequence, region B, was ϳ1.2 kb upstream of the ATF site (Fig.  5A). Activation of the UPR induced ATF4 binding to region A in the GADD34 promoter, whereas binding of ATF4 to the B region was below the level of detection. Antibodies to YY-1 did not immunoprecipitate either region (Fig. 5C). Therefore, our data demonstrate that ATF4 directly binds to the GADD34 promoter in an ER stress-dependent manner and suggests that this binding occurs through a region encompassing the conserved ATF site.
ATF4 Bound to the Conserved GADD34-ATF Site Specifically in an ER Stress-dependent Manner-Next we wished to directly test the binding of ATF4 to the conserved ATF site of GADD34 promoter using in vitro electrophoretic mobility shift assays (EMSA). We synthesized a 29-bp probe encompassing the potential ATF4-binding site (GADD34-ATF) and performed EMSA in the presence of nuclear extracts harvested from WT MEFs treated with Tg for 0, 2, 4, 6, 8, and 14 h (Fig. 6A). Either BSA or an anti-ATF4 antibody was added to each gel shift reaction as indicated. 10-Fold excess of the mutant oligonucleotides with the ATF site mutated was added to each gel shift reaction to exclude binding outside of the ATF site. Complexes that were supershifted with the anti-ATF4 antibody were observed as early as 4 h after Tg treatment (Fig. 6A, lanes 6, 8, 10,  and 12), which is consistent with the kinetics of GADD34 transcript induction (Fig. 2C) and eIF-2␣ dephosphorylation (Fig. 2B) during ER stress in WT MEFs. ATF4 binding became weaker after 8 h of UPR activation (Fig. 6A, lanes 9 and 10) but persisted to the end of our time course of Tg treatment (lanes 11 FIG. 4. Identification of potential ATF4-binding sites that are conserved between human and mouse GADD34 promoters. Promoter sequences for human and mouse GADD34 were obtained from ensemble genomic sequence data base and aligned to identify regions of homology. The transcription initiation site and the TATA box are identified. The conserved ATF/cAMP-response element-binding site (GADD34-ATF) is boxed and indicated. A highly conserved stretch of nucleotides upstream of the GADD34-ATF site contains potential binding sites for CHOP and/or NF-B, which are underlined. and 12). We were unable to observe the original ATF4-binding complex before supershifting with the ATF4 antisera. There was a relatively large diffuse complex (complex A) that formed with the GADD34-ATF probe, which might mask the unshifted ATF4-binding complex. When a control EMSA was performed in parallel using the CHOP C/EBP-ATF composite site as the probe (Fig. 6B, lanes 7-9) to reproduce the ATF4 and C/EBP-␤-binding complexes to the CHOP composite site as we demonstrated previously (24), the diffuse complex (complex A) binding to GADD34-ATF probe migrated at a similar rate as the ATF4-binding complex to the CHOP composite site probe, as pointed out by the arrow in Fig. 6B (lanes 7 and 9). Because the two radioactively labeled probes are at the same length (29 bp), this is consistent with the possibility that the unshifted ATF4-binding complex observed with the GADD34-ATF probe is masked by the diffuse complex (complex A).
To confirm that the binding activity detected by the antisera was specific to ATF4, we performed gel shift assay with the WT GADD34-ATF probe using nuclear extracts from either WT or ATF4 null MEFs (Fig. 6C). Again, ATF4 binding was detected only after ER stress (4 h of Tg treatment) in the WT MEFs (lanes 1-4). Addition of ATF4 antibody to the binding reaction did not change the migration of any complex either before or after ER stress when nuclear extracts from ATF4 null cells were examined (lanes 5-8). Together, our EMSA data demonstrated that an ATF4-binding site exists in the GADD34 promoter region, which is conserved in both the mouse and human promoters and binds to ATF4 specifically in an ER stress-dependent manner.
Finally, to evaluate the role of the ATF site in the context of GADD34 promoter, we cloned both WT GADD34 promoter encompassing 700 bp upstream of the transcription start site or the same region with ATF site mutated. As shown in Fig. 6D, mutation of the ATF site resulted in a significant decrease in the activity of the promoter both before and after ER stress even though the fold inducibility of the promoter is only slightly decreased. This suggests that even though ATF4 plays an important role in transactivating the ATF site, other transcription factors binding to the site under non-stressed conditions may contribute to the basal activity of the GADD34 promoter in this assay. In conclusion, ATF4 directly binds to and transactivates the conserved ATF site of the GADD34 promoter in an ER stress-dependent manner.

DISCUSSION
The UPR-induced translational arrest is essential for the survival of cells that are experiencing ER stress. Animals that are nullizygous for PERK (22) or in which a non-phosphorylated form of eIF-2␣ has been inserted in place of the wild type gene (21) are unable to down-regulate protein synthesis during ER stress and, as a result, experience severe difficulties in handling the accumulation of unfolded proteins which leads to apoptotic cell death. The block in protein synthesis is not unique to ER stress but is also elicited by a number of cellular stresses including amino acid deprivation and heme deficiency (6). Even though each of these stresses activates a unique eIF-2␣ kinase (GCN2 and HRI, respectively) that causes the block in translation, many aspects of the downstream signaling pathways are shared. First, the translational arrests elicited by these stresses are reversible (29,41,42). Recently, Ron and co-workers (29,31) identified GADD34 as a component of the PP1-phosphatase complex that is responsible for dephosphorylating eIF-2␣ and allowing translation to resume. Their study revealed that GADD34 induction is downstream of the eIF-2␣ kinase PERK and GCN2, suggesting the existence of a negative regulatory feedback loop. A second shared aspect of the cellular stress responses is the induction of the ATF4 transcription factor (23), which is largely regulated at the translational level. Many cells possess ATF4 transcripts but do not produce the protein under normal physiological conditions due to the presence of several small ORFs in the 5Ј-untranslated region (5Ј-UTR) of the transcript. Only after the phosphorylation of eIF-2␣ is the correct start site used to translate the ATF4 protein. We showed that ATF4 might contribute to CHOP induction during ER stress (24), and Holbrook and co-workers (25) has demonstrated its involvement in CHOP induction during arsenite poisoning. Most recently, the study by Harding et al. (43) revealed that ATF4 mediates an integrated stress response pathway, which protects the cells from oxidative stress.
Our data showing the absence of GADD34 transcript induction in response to either UPR-inducing agents or amino acid deprivation in ATF4Ϫ/Ϫ cells demonstrated that ATF4 is upstream of and required for GADD34 induction during both stress conditions. Absence of GADD34 induction in ATF4 null cells resulted in persistent eIF-2␣ phosphorylation, delayed translation recovery, and diminished induction of UPR target genes like BiP. Next, combined with the CHIP assay data showing that endogenous ATF4 was able to bind to the GADD34 promoter directly in a stress-dependent manner, we conclude that GADD34 is a direct target of ATF4 during various cellular stresses. Finally, we demonstrated that ATF4 acts through a conserved ATF site, which is important for the activity of the GADD34 promoter during ER stress.
Thus, we have delineated a negative regulatory feedback FIG. 5. CHIP assay demonstrated that ATF4 directly bound to the GADD34 promoter in an ER stress-dependent manner. A, relative positions of the targeting sequences designated regions A and B are indicated. The location of the ATF site within region A is highlighted. B, cross-linked chromatin used for CHIP assay in C were fragmented and run on a DNA-agarose gel and stained with ethidium bromide. C, cross-linked chromatin from untreated or Tg-treated (6 h) NIH3T3 fibroblasts was immunoprecipitated with either anti-ATF4 or anti-YY-1 antibody as negative control. 1% of total input before immunoprecipitation was used as positive control. Immunoprecipitated chromatin fragments were then amplified by PCR methods to detect ATF4 binding to the two regions of GADD34 promoter as described in A.
loop that controls the phosphorylation state of eIF-2␣ during cellular stresses (Fig. 7A). UPR-induced activation of PERK (or other eIF-2␣ kinases activated by their respective stresses) leads to the phosphorylation of eIF-2␣. This in turn inhibits much of the cellular translation and, ironically, allows the translation of ATF4. ATF4 in turn directly trans-activates the GADD34 promoter by binding to the ATF site. GADD34 translation then occurs via a mechanism that is likely to be eIF-2␣independent. GADD34 activates the PP1 phosphatase, which leads to eIF-2␣ dephosphorylation and the resumption of general translation (Fig. 7A). Therefore, phosphorylation of eIF-2␣ sets in place the pathway leading to its dephosphorylation, which ensures that the translation block will be reversed. The fact that GADD34 acts on the phospho-form of eIF-2␣ instead of PERK suggests that it should be able to alleviate the block in protein synthesis for many other cellular stresses that activate eIF-2␣ kinases. The kinetics of this loop or the activation of different downstream components may vary in different cells or under different physiological and pathological conditions. For instance, the phosphorylation of eIF-2␣ in response to doublestranded RNA-activated protein kinase activation by doublestranded RNA appears to be relatively long lived (44), although the mechanism for this difference is not currently understood. In the absence of GADD34-dependent dephosphorylation, as occurred in the ATF4 null cells, continuous cellular stress resulted in prolonged eIF-2␣ phosphorylation and an extended suppression of protein synthesis. During ER stress, the significant loss of BiP protein induction in ATF4Ϫ/Ϫ MEFs due to persistent protein synthesis inhibition suggests that the quick translation recovery is essential for the up-regulation of cytoprotective mechanisms.
A similar feedback loop was hypothesized by the Ron and FIG. 6. ATF4 bound to the conserved GADD-ATF site specifically in an ER stress-dependent manner. A, EMSA was performed using the WT mouse sequence surrounding the GADD34-ATF site as the probe and nuclear extracts harvested from WT MEFs at different time points after Tg treatment. The supershifted ATF4 complex is indicated by an arrow, and the diffuse complex (complex A) that may mask the unshifted ATF4 complex is also indicated. B, gel shift assays were performed with either the WT-mGADD34-ATF probe or the WT hamster CHOP composite site probe as a control. Either non-stressed or ER-stressed (4-h Tg) nuclear extracts were used. The unshifted and the supershifted ATF4 complexes that bind to the WT CHOP composite site probe are indicated. C, a similar gel shift experiment was performed as described in A, whereas nuclear extracts from both WT and ATF4 null MEFs were used. D, either WT GADD34 promoter or one with ATF site mutated was cloned in front of a CAT reporter gene and transfected into COS-1 cells. The cells were then left untreated or treated with Tg for 14 h before cell lysates were harvested for CAT activity.
co-workers (29), along with a second possibility in which GADD34 translation might be regulated via the presence of ORFs in the 5Ј-UTR similar to ATF4. When the 5Ј-UTR of the mouse and human GADD34 transcript were examined, we did locate two small ORFs that are relatively conserved (Fig. 7B). Therefore, it is possible that GADD34 is regulated at both transcriptional and translational levels like ATF4. However, neither the human nor the mouse GADD34 mRNA 5Ј-small ORFs (Fig. 7B) overlaps with the correct translation initiation site as is observed with the human and mouse ATF4 transcripts (23). More importantly, the stop codon of the first 5Ј-ORF (untranslated ORF1) in the mouse GADD34 transcript overlaps with the start codon of the second 5Ј-ORF (untranslated ORF2), thus potentially diminishing the use of the second ORF and, therefore, its effect on interfering with the correct translation start site (45). In addition, unlike ATF4, we see no evidence of GADD34 transcripts present in unstressed MEFs (Fig. 2, C and D).
Interestingly, not all transcripts appear to be equally affected by the phospho-eIF-2␣-mediated block in translation, nor does the resumption of protein synthesis occur equally for all. After 2 h of Tg treatment, ATF4 protein was detected at near maximal level in the wild type MEFs (Fig. 1A), even though eIF-2␣ phosphorylation levels were still elevated ( Fig.  2A) and total protein synthesis was very much decreased (Fig.  3B). In other studies, it appears that the CHOP transcription factor may also be fairly resistant to the eIF-2␣-mediated protein synthesis inhibition, as its translation is detectable almost as early as the ATF4 protein (24). 2 GADD34 may represent a third member of this class of proteins, because its biosynthesis is required to alleviate the translation block. Perhaps because these proteins play critical roles in regulating early components of the stress response, their messages have evolved to escape this translation block. Conversely, synthesis of the BiP chaperone is clearly affected early in the response and in ATF4Ϫ/Ϫ cells. In WT MEFs, increased transcription of the BiP gene was detectable 2 h after ER stress (Figs. 1B and 2). However, BiP translation diminished initially and was just returning to basal rates between 1 and 2 h after activating the UPR (Fig. 3D). Because BiP levels control the activation of UPR signaling molecules (Ire1, PERK, and ATF6) by binding to their luminal domains and keeping them in an inactive form (11,46), it is plausible that an increase in the level of this chaperone too early in the response could shut it down prematurely. An internal ribosome entry site has been identified in the 5Ј-UTR of the BiP mRNA, whose activity does not change after ER stress or amino acid deprivation (44,47). It is unclear whether the internal ribosome entry site-mediated eIF-2␣-independent BiP translation plays an important role in the overall synthesis of the BiP protein. Our data demonstrated that BiP translation is down-regulated just like other genes when eIF-2␣ is phosphorylated during the 1st h of ER stress, and BiP translation does not recover when the protein synthesis inhibition persists even though its message is induced as normal (Fig. 3D). This suggests that the majority of BiP protein is translated via the eIF-2␣-dependent mechanism. Finally, cyclin D1 represents a third class of proteins whose translation is susceptible to eIF-2␣-mediated inhibition and decrease dramatically after stress but do not recover even long after eIF-2␣ is dephosphorylated and general protein synthesis has resumed (7). Because cyclins control entry of the cells into the G 1 phase of cell cycle, it is possible that this protein remains affected to ensure that cells experiencing stress do not propagate. How many proteins are represented in the various classes and how these differences in susceptibility to translation inhibition are achieved are not known.
We conclude that an eIF-2␣-dependent feedback regulatory loop exists to ensure the transient nature of protein synthesis inhibition during the mammalian UPR. It is initiated by ATF4, a product of eIF-2␣ phosphorylation, which directly induces GADD34 transcription, a component of the phosphatase complex that dephosphorylates eIF-2␣, thereby allowing translation to resume. ATF4 null mice cannot induce GADD34, and as a result, show prolonged eIF-2␣ phosphorylation and protein synthesis inhibition.