ATF4 is an oxidative stress–inducible, prodeath transcription factor in neurons in vitro and in vivo

Amino acid depletion via the arginine-degrading enzyme arginase is a well-characterized strategy for protecting neurons from oxidative stress–induced cell death (12, 17). Studies in nonneural cells have shown that amino acid deprivation activates cell-survival pathways by activating the bZip transcription factor ATF4 (11). We initiated our studies with the hypothesis that arginase is a multipotent inhibitor of apoptotic neuronal cell death by activating ATF4 and its downstream gene targets. Because endoplasmic reticulum stress by thapsigargin is also associated with increased ATF4 expression (18), we examined the corollary hypothesis that this agent could initiate tolerance to oxidative stress in primary cortical neurons. As predicted, arginase and thapsigargin led to nearly complete protection from neuronal cell death (Fig. 1, A and B) and to accumulation of ATF4 protein (Fig. 1 C). Interestingly, we found that oxidative stress alone was sufficient to induce ATF4 protein levels (Fig. 1 C).

To verify that the protection by arginase and thapsigargin requires ATF4, we cultured cortical neurons from brains of embryonic ATF4^−/− mice (E15.5). Immunohistochemical analysis (Fig. 2 A) using an antibody to microtubule-associated protein 2 (MAP2; which stains neural dendrites) showed that basal viability and morphology of cell bodies and dendrites did not display any differences between ATF4^+/+ and ATF4^−/− neurons. We found that ATF4 mRNA was induced in ATF4^+/+ neurons after oxidative stress and confirmed its absence in ATF4^−/− neurons (Fig. 2 B). We then examined susceptibility to oxidative stress induced by the glutamate analogue homocysteate (HCA) in ATF4^+/+ and ATF4^−/− cortical neurons (Fig. 2, C and D). Contrary to our hypothesis, we found that ATF4^−/− neurons were significantly protected from oxidative stress–induced death. ATF4^−/− neurons demonstrated enhanced resistance, as monitored by MTT reduction (Fig. 2 C) and live/dead staining (Fig. 2 D). Moreover, in ATF4^+/+ neurons, ATF4 induction by treatment with HCA was a result of induction at the translational level (Fig. 2 E). We confirmed the nuclear accumulation of ATF4 in ATF4^+/+ neurons exposed to oxidative stress (Fig. 2 F). Higher nuclear ATF4 activity after oxidative stress was demonstrated by showing higher promoter activity of an ATF4 target gene, tribbles homologue 3 (TRB3), that has been recently associated with cell death. TRB3 mRNA is induced upon HCA treatment in an ATF4-dependent manner (Fig. 2 G). To characterize ATF4 binding on the promoter of TRB3, we generated a 2-kb fragment of the mouse TRB3 promoter (Fig. 2 H). This fragment contains a 33-bp element whose human analogue has been characterized as a DNA binding region of ATF4. Transfection of both ATF4^+/+ and ATF4^−/− neurons with the TRB3WT promoter construct, along with a mutant version lacking the binding element, revealed that this element plays an important but not exclusive role in the induction of the TRB3 promoter after oxidative stress. In the absence of ATF4, the TRB3 promoter activity was not induced in response to oxidative stress.

In fibroblasts, previous studies (11) have established ATF4 as an important prosurvival transcription factor in the context of oxidative stress. Microarray analysis in that study (11) showed that several ATF4-regulated genes in fibroblasts are involved in amino acid import, glutathione biosynthesis, antioxidant defense, and DNA repair. In contrast, in neurons, ATF4 has classically been thought to act mainly as a transcriptional repressor. In this context, ATF4 has been hypothesized to negatively influence survival by repressing expression of genes that are otherwise activated by prosurvival transcriptional activators such as cAMP response element binding protein. To begin to examine whether ATF4 functions primarily as a repressor or activator in neurons and to determine the extent to which these ATF4-regulated genes function similarly in fibroblasts and neurons, we performed a global analysis of gene expression using microarrays in ATF4^+/+ and ATF4^−/− neurons (Fig. 3 A). More specifically, we assessed (a) the physiological genomic effect of ATF4 knockout (comparing ATF4^−/− with ATF4^+/+ neurons), and (b) the effect of ATF4 knockout on the response to oxidative stress (comparing ATF4^−/− with ATF4^+/+ neurons after HCA treatment; Fig. 3 B and Table I). At the chosen statistical threshold (5% false discovery rate), 136 probes were down-regulated in ATF4^−/− versus ATF4^+/+ neurons, compared with 53 that were up-regulated, suggesting a role for ATF4 as a transcriptional activator. Functional gene ontology categories overrepresented in this list include mitochondrion, oxidoreductase activity, and amino acid metabolism. Several of the down-regulated genes have been shown to be positively regulated by ATF4, including the prodeath gene TRB3. Comparison of ATF4^−/− versus ATF4^+/+ neurons after oxidative stress with HCA treatment suggested a fundamental role of ATF4 in modulating oxidative stress–induced gene expression. In fact, 119 probes are dysregulated in response to oxidative stress in ATF4^+/+ neurons, whereas only 3 change in ATF4^−/− neurons. Collectively, these findings suggest that ATF4 is a major upstream regulator of oxidative stress–induced changes in gene expression. Additionally, for most genes in embryonic cortical neurons, ATF4 functions as an activator and not a repressor.

Gene expression array data were obtained from ATF4^−/− fibroblasts that display a higher susceptibility to oxidative stress (11). A significant overlap could be observed between the array data from ATF4^−/− fibroblasts and neurons (Table S1), indicating that ATF4 regulates at least a major part of target genes in a cell type–independent manner. These similarities do not explain the observation that ATF4 is prosurvival in fibroblasts and prodeath in neurons, but they demonstrate the reliability in our analysis. In contrast, a subset of genes, including the prodeath gene TRB3, were ATF4-regulated in neurons but not in fibroblasts. Collectively, these findings provide some understanding of the tissue-specific gene regulation mediated by ATF4 that could account for the divergent phenotypes in dividing fibroblasts versus postmitotic neurons.

Because we used germline knockout animals (19) in this study, we could not exclude the possibility that the decreased susceptibility to oxidative stress was a consequence of a compensatory effect distantly related to ATF4 deficiency. To address this question, we used adenoviral overexpression of a mouse ATF4WT construct. As a control, we used a dominant-negative ATF4 construct harboring a mutation in its DNA binding domain (²⁹²RYRQKKR²⁹⁸ to ²⁹²GYLEAAA²⁹⁸). To confirm the different DNA binding properties of these constructs, we analyzed the 33-bp binding region of the TRB3 promoter by gel-shift (Fig. 4 A) and chromatin immunoprecipitation studies (Fig. 4 B). In fact, only nuclear extracts from cells overexpressing ATF4WT in combination with WT oligonucleotide formed a band specific for ATF4. Consistently, only TRB3 promoter chromatin from cells overexpressing ATF4WT could be PCR amplified using primers flanking the ATF4 binding site. Cotransfection of WT and mutant ATF4 constructs with reporter constructs containing the WT and the mutant binding site confirmed the activating effect ATF4 has on transcription (Fig. 4 C).

To determine whether forced expression of ATF4 in ATF4^−/− neuronal cultures can restore sensitivity to oxidative stress and/or has an effect itself, we overexpressed both constructs along with a GFP control in both ATF4^+/+ and ATF4^−/− cortical neuronal cultures, followed by treatment with HCA (Fig. 5). We confirmed the efficient expression of both ATF4 constructs in cortical neurons by immunohistochemistry (Fig. 5 A) and Western blotting (Fig. 5 B). Specifically, ∼50% of neurons and a small number of glia are infected with the adenoviruses. Infection with ATF4ΔRK was able to protect ATF4^+/+ cortical neuronal cultures from HCA-induced toxicity (Fig. 5, C and D), whereas infection with ATF4WT was capable of rendering ATF4^−/− neurons sensitive to HCA (Fig. 5, E and F). Moreover, infection with ATF4WT itself significantly reduced the viability in both WT and ATF4^−/− neurons. Combination of forced expression of ATF4WT and treatment with HCA resulted in a higher loss of viability than infection with ATF4WT alone, suggesting that at least one more pathway in addition to ATF4 is involved in HCA toxicity. Collectively, these results are consistent with the notion that ATF4 indeed plays a key role in oxidative stress–mediated cell death.

The loss of glutathione is central to the in vitro model of oxidative stress used in this study. This model has been successfully used to describe signaling pathways that prevent cell death downstream of glutathione depletion. However, in fibroblasts, ATF4 has been described as a positive regulator of glutathione metabolism (11). Therefore, we performed glutathione measurements to determine whether ATF4 permits cell death in response to HCA by direct influence of the neuronal glutathione metabolism or by acting downstream of glutathione depletion. In untreated neurons, glutathione concentrations in ATF4^+/+ and ATF4^−/− neurons did not differ (Fig. 6 A). As anticipated, HCA treatment caused a progressive loss of glutathione in ATF4^+/+ neurons. However, ATF4^−/− neurons displayed a markedly slower decline in glutathione concentration. To rule out the possibility that an increased rate of glutathione synthesis is responsible for decreased sensitivity to cystine deprivation–induced glutathione depletion (20), we used buthionine sulfoximine (BSO) to directly inhibit the rate-limiting enzyme of glutathione synthesis, γ-glutamylcysteine synthetase (γ-GCS; Fig. 6 B). BSO resulted in cell death in ATF4^+/+ neurons, whereas ATF4^−/− neurons were essentially resistant (Fig. 6, C and D). As expected from these results, γ-GCS expression did not differ between ATF4^+/+ and ATF4^−/− neurons at the protein level and did not appear in the gene array (Fig. 6 E). Similar to HCA, BSO treatment led to a decline in the concentration of glutathione in ATF4^+/+ neurons, whereas ATF4^−/− neurons displayed a distinctly slower reduction of glutathione (Fig. 6 F). These findings are consistent with the notion that the differences in reduction of glutathione after HCA or BSO is not attributable to changes in synthesis. To determine whether ATF4 directly regulates the neuronal glutathione content, we measured glutathione in neurons with forced expression of ATF4WT, ATF4ΔRK, and GFP (Fig. 6 G). Consistent with the finding that forced expression of ATF4WT can reduce neuronal viability, we found that this viability reduction correlates with a reduction in the neuronal glutathione content. Because cell death attributable to glutathione depletion can be completely blocked by a whole host of classical antioxidants, we combined forced expression of ATF4WT with butylated hydroxyanisol (BHA) treatment (Fig. 6, H and I). BHA is a well characterized antioxidant known to block HCA-induced cell death in cortical neurons (6). In fact, in neurons overexpressing ATF4WT, BHA significantly increased neuronal viability and protected against the additional toxic effect of HCA treatment. Finally, we measured production of ROS by using the redox-sensitive probe 2′,7′-dichlorofluorescin (DCF; Fig. S1). Indeed, ATF4^−/− neurons displayed lower levels of ROS in response to HCA treatment than ATF4^+/+ neurons. These findings are consistent with the notion that ATF4 positively influences ROS levels and leads to a higher consumption of glutathione independent of its synthesis.

Oxidative stress is an established mediator of neuronal loss in cerebral ischemia (21). It is generally accepted that glutathione acts as a major endogenous cerebral antioxidant in the adult brain. In fact, depletion of intracellular antioxidants has been associated with stroke pathology and repletion of antioxidants has reduced neuronal damage (22). Therefore, we hypothesized that ATF4 deficiency might positively influence the antioxidant status in the adult brain and thereby improve the outcome after ischemia-reperfusion injury. To test the hypothesis that ATF4 germline knockout animals have less neuronal loss after stroke, we used a transient middle cerebral artery occlusion (MCAo) model for ischemia-reperfusion injury, as previously described (23).

First, we performed morphological studies to rule out that germline ATF4 deficiency causes a fundamental abnormality in the adult brain that would bias stroke outcome (19, 24–26). Examining sagittal sections of brains from ATF4^+/+ and ATF4^−/−animals (Fig. 7 A, top), we did not detect any significant structural differences in the cerebellum, hippocampus, or cortex. We then assessed the morphology of major cerebral blood vessels of the circle of Willis (Fig. 7 A, middle) (23). Besides a higher degree of tortuosity in the MCA, no major abnormalities were detected in the brains obtained from ATF4^−/− mice. Finally, using immunofluorescent staining for the endothelial-specific marker CD31/PECAM, we did not observe significant differences in the density or morphology of the microvessels (Fig. 7 A, bottom). In addition, important physiological parameters did not differ between ATF4^+/+ and ATF4^−/− animals (Fig. S2 and Table S2).

The procedure used to occlude the MCA (occlusion time = 45 min) led to reproducible infarcts in WT animals involving both the cerebral cortex and the striatum, with sparing of the hippocampus. However, a significantly smaller infarct was observed when the procedure was performed in ATF4^−/− animals (Fig. 7, C and D). The smaller infarct area could be observed in all sections of the brain. Thus, the higher resistance to ischemia-reperfusion injury was not limited to a specific brain area. In fact, the smaller infarct volume was associated with a faster recovery in ATF4^−/− animals, as measured by a simple neurological score (Fig. 7 E). In addition, two simple behavioral tests (Fig. 7, F and G) complemented the morphological data. In consideration of the eye lens malformation regularly observed in ATF4^−/− animals (25), these tests are not dependent on the visual sense. Consistent with the morphological data, ATF4^−/− animals performed significantly better than the ATF4^+/+ control group, suggesting that ATF4 deficiency facilitates recovery after stroke.

Solid evidence supports the hypothesis that oxidative stress is an initiator or propagator of neuronal dysfunction or death in prevalent neurological diseases, including stroke, spinal cord injury, Alzheimer's disease, and Parkinson's disease (3, 27). These studies have led to the somewhat surprising conclusion that oxidants can trigger neuronal death via highly regulated signaling pathways and subsequent activation of prodeath transcription factors, leading to the controlled demise known as apoptosis (28). The current study adds to our understanding of the cellular transcription factors that regulate neuronal viability and function after oxidative stress in primary cortical neurons. Specifically, we show that the transcription factor ATF4 is induced by oxidative stress caused by depletion of the major antioxidant tripeptide, glutathione. We further demonstrate that transgenic deletion of ATF4 renders neurons resistant to neuronal cell death. Prevention of cell death by germline knockout of ATF4 is associated with a preservation of glutathione levels, the primary mediator of death in our oxidative stress model. Consistent with ATF4's role in regulating an early, upstream aspect of the oxidative neuronal death pathway, we found that ATF4 deficiency causes global down-regulation of gene expression and blocks the up-regulation of many genes that are induced by oxidative stress in WT neurons. Accordingly, forced expression of ATF4 was sufficient to promote cell death and loss of glutathione. In addition, we demonstrate that in ATF4^−/− neurons, restoration of ATF4 protein expression reinstated sensitivity to oxidative death. We establish the relevance of protection by ATF4 deficiency by demonstrating that the ATF4 homozygous knockout is capable of protecting the adult mouse brain from stroke-induced injury and disability. Indeed, we show that ATF4^−/− animals recover more easily and maintain proper motor function more efficiently. Although known to be a stress-responsive protein, these results for the first time establish ATF4 as a redox-regulated protein that can function to lower the threshold for oxidative stress–induced death in neurons.

Classically, stress-mediated enhancement of ATF4 levels is known to occur via enhanced efficiency of translation of constant levels of ATF4 mRNA (29, 30). Consistent with this model, we found an up-regulation of ATF4 mRNA translational efficiency after glutathione depletion. However, we also observed an increase in ATF4 mRNA levels after oxidative stress, thereby confirming a recent report describing a role for transcriptional regulation in ATF4 induction by cell stress (31). The distinct increase in ATF4 protein levels in the nucleus of neurons undergoing oxidative stress corresponded to an ATF4-dependent induction of a TRB3 promoter-luciferase reporter. Collectively, these observations argue that oxidative stress–induced changes in ATF4 mRNA levels and translational efficiency lead to increased ATF4 activity in neurons. Because ATF4 overexpression is sufficient to induce death, limiting the amount of ATF4, at least under basal conditions, seems to be necessary for neuronal survival. Mechanisms that control and limit basal ATF4 activity (32–34) appear to be important determinants of neuronal fate. Although this low level of ATF4 appears to be sufficient to regulate a host of genes, its absence does not grossly influence neuronal morphology or basal survival.

Several converging lines of inquiry demonstrate a central role for glutathione metabolism in degeneration and aging in the nervous system. Our findings confirm the primary role that glutathione can play in neuronal death. However, the mechanisms by which a global depletion of glutathione in neurons is transduced into apoptotic events remained elusive. The data presented in this study favor a model in which glutathione depletion causes an ATF4-dependent up-regulation of a coordinated set of genes that orchestrate the timely and irreversible demise of the cell. Part of the orchestrated sequence of events downstream of ATF4 activation includes the increased production of ROS and turnover of glutathione. Although our data clearly establish that ATF4 induction does not influence glutathione synthesis, we cannot determine whether the increased turnover of glutathione is primarily attributable to a change in a glutathione-degrading enzyme with a subsequent increase of ROS, or a secondary consequence of ATF4-mediated ROS production. Increased ROS production would be expected to decrease glutathione levels by enzyme-mediated consumption of the thiol tripeptide. Of note, many mitochondrial genes that could facilitate production of deleterious ROS during respiration are down-regulated in the ATF4^−/− neurons. Future studies will clarify whether ATF4 mediates death primarily by altering glutathione turnover or by dialing up the production of free radicals.

ATF4 was originally described as a transcriptional repressor (35, 36). However, our gene array data suggest that ATF4 acts at least in part as a transcriptional activator. This finding is in line with several more recent publications that describe the activating effect of ATF4 on several target genes, such as heme oxygenase 1 (37, 38), stanniocalcin 2 (39), osteocalcin (40), gadd153/CHOP (41), and TRB3 (42). Some of these target genes have been shown to play an important role in cell death and cell survival in the brain. Pharmacological induction of heme oxygenase 1 was demonstrated to protect neurons from oxidative insults both in vitro and in vivo (43). Indeed, heme oxygenase 1 is induced in WT neurons by oxidative stress; however, it was down-regulated in ATF4^−/− neurons. Likewise, stanniocalcin 2 is up-regulated in response to oxidative stress in WT neurons but was down-regulated in ATF4^−/− neurons. In a recent work (39), stanniocalcin 2 was identified as an oxidative stress–responsive gene with cytoprotective properties. A prodeath ATF4 target gene, TRB3, was strongly dependent on ATF4 and up-regulated in response to oxidative stress. TRB3's role in neuronal apoptosis was described first in a neuronal cell line (44) and has been recently confirmed (42).

The data presented in this work are in support of a strong prodeath role of ATF4. However, previous findings regarding the roles of ATF4 have yielded partly different results. ATF4^−/− fibroblasts are impaired in expressing genes involved in amino acid import, glutathione biosynthesis, and resistance to oxidative stress (11). Therefore, ATF4 was proposed to play a key role in rendering the cell more resistant against the metabolic consequences of endoplasmic reticulum oxidation. Therefore, we initially hypothesized that ATF4 induction by endoplasmic reticulum stress or amino acid depletion is the key event in the protection against oxidative stress mediated by HCA treatment (Fig. 1). The finding that ATF4 activation can be associated with protection from oxidative stress–induced cell death, although it clearly acts as a prodeath transcription factor, points to a complex and context-specific role of ATF4 in the propagation of neuronal cell death. However, both endoplasmic reticulum stress or amino acid depletion initiate several different signaling events in addition to ATF4 induction (2, 45). Which of these signaling events are capable of preventing the deleterious consequences of ATF4 activation requires further investigation. In general, the similarity of data obtained from fibroblasts and from cortical neurons is limited. However, despite the opposite effects on glutathione metabolism and cell survival, a comparison between the array data from fibroblasts with our gene expression data indicates a highly significant overlap. Even though the role of each of the many target genes of ATF4 warrant further investigation, the different metabolic demands of postmitotic neurons and actively dividing fibroblasts might help to explain the distinct role of ATF4 in distinct cell types (46). Additionally, differently expressed dimerization partners of ATF4 might also significantly modify ATF4's function. Indeed, several different cancer cell lines show an elevated level of ATF4 expression (10, 47–50). Moreover, ATF4 is already known to have functions that are limited to a specific cell type such as osteoblasts (40, 51, 52). The precise characterization of these additional signaling events warrants further investigation, and might help to explain why and how these signaling events are capable of preventing the deleterious consequences of ATF4 induction.

We have established that ATF4 also plays an important prodeath role in an in vivo model of stroke. However, the use of conventional ATF4^−/− animals poses a risk of bias on the outcome in a complex paradigm such as stroke (53). Therefore, the results need to be interpreted with caution. We cannot rule out the possibility that ATF4 deficiency in all cell types of the brain and the body has an impact on stroke outcome. However, no overt difference was observed when comparing the brain morphology between WT and ATF4^−/− brains, including the vascular system. In addition, the smaller infarct size in the ATF4^−/− mice could not be attributed to changes in systemic physiological parameters, including body temperature, or changes in cerebral blood flow under basal conditions or after MCAo. However, an issue that was not addressed in this study is whether the ATF4^−/− mice displayed an altered inflammatory response after ischemia that could explain the smaller infarct size. Although a higher propensity to infections or other immune-related alterations has not yet been observed in ATF4^−/− mice, we cannot rule out the possibility that an altered inflammatory response (54) is one reason for the smaller infarct volume in the ATF4^−/− mice. In summary, the strong correlation in the outcome between in vitro and in vivo studies lets us conclude that our results, even when reviewed carefully, suggest an important role for ATF4 in the induction of neuronal cell death. Thus, ATF4 might be an important target for therapeutic intervention against stroke and other neurodegenerative diseases.

All animal procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee of the Weill Medical College of Cornell University. Germline ATF4^−/− mice have been described previously (19). For the culture of mouse embryonic cortical neurons, embryos (15.5 days post conceptionem) were obtained from the mating of ATF4 heterozygous mice.

Primary neuronal cultures were prepared from the cerebral cortices of mouse embryos, as described previously (55), with minor modifications. The mouse hippocampal cell line HT22 was maintained according to standard procedures (56). Transfection of both cortical neurons and HT22 cells was performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Recombinant arginase was provided by D. Ash (Temple University, Philadelphia, PA).

Genomic DNA for PCR was prepared using the DNeasy genomic DNA isolation kit (QIAGEN). The WT alleles were detected using the primer 5′-AGCAAAACAAGACAGCAGCCACTA-3′. The ATF4^−/− alleles (neomycin) were detected using the primer 5′-ATATTGCTGAAGAGCTTGGCGGC-3′. As a common reverse primer, we used 5′-GTTTCTACAGCTTCCTCCTCCACTCTT-3′.

Cell viability was assessed by propidium iodide uptake and retention of calcein (Invitrogen) using an inverted epifluorescence microscope (Axiovert 200M; Carl Zeiss, Inc.). MTT assay was performed according to the manufacturer's instructions (Promega). In brief, absorbance was measured at 570 nm. From each data point, the reference absorbance measured at 690 nm was subtracted. Finally, values were calculated as the percentage of untreated control cells.

The expression plasmids for ATF4WT and dominant-negative ATF4ΔRK (provided by J. Alam, Ochsner Foundation, New Orleans, LA) (37), and the mouse ATF4 5′UTR luciferase construct (D. Ron, New York University, New York, NY) have been previously described. The promoter construct pTRB3 was generated by amplification of genomic mouse DNA using the primers 5′-CTCACTCAGGTGCCTGTAGTGCTCG-3′ and 5′-TCAGCAGAAGCAGCCAGAGGTGTAG-3′, followed by a second round of PCR with the primers 5′-ACACTCGAGAGAGAAACAAATGTGTCATG-3′ and 5′-ACAAAGCTTCTAGAGAGCAAGGAAGAAAG-3′ before cloning into the vector pGL3basic (Promega). The mutant form lacking the 33-bp ATF4 binding site was generated using a site-directed mutagenesis kit (QuikChange; Stratagene). To generate the reporter plasmids with the 33-bp ATF4 binding sequence, the following oligonucleotides were annealed and cloned into pGL3 basic: p33WT, 5′-TCGAGGCAGATTAGCTCAGGTTTACATCAGCCGGGCGGGGATCCA-3′ and 5′-AGCTTGGATCCCCGCCCGGCTGATGTAAACCTGAGCTAATCTGCC-3′; and p33MUT (mutant form), 5′-TCGAGGCAGATTAGCTCAGTCTAAACCTATAGGGGCGGGGATCCA-3′ and 5′-AGCTTGGATCCCCGCCCCTATAGGTTTAGACTGAGCTAATCTGCC-3′. All sequences were verified by automatic DNA sequencing.

Primary neurons and HT22 cells were transfected with the respective reporter construct along with the pTK-Renilla control (Promega). A dual luciferase assay (Promega) was performed using a bioluminometer (MDS Analytical Technologies), according to the manufacturer's instructions.

To generate the adenoviruses ATF4WT and ATF4ΔRK, the expression cassettes of each construct were cloned into the shuttle vector ad5 pVQ-K-NpA. The correct sequence was confirmed by automatic DNA sequencing. Virus generation and amplification were performed by ViraQuest. Infection with adenoviruses was performed at a multiplicity of infection (MOI) of 100.

RNA quantity was assessed with a spectrophotometer (Nanodrop; Thermo Fisher Scientific), and quality was assessed with nanochips (Bioanalyzer; Agilent Technologies). Total RNA was amplified, labeled, and hybridized on arrays (MouseRef-8 Expression BeadChip; Illumina). Data analysis was performed using Bioconductor packages (available at http://www.bioconductor.org; reference 57). Raw data were log2 transformed and normalized using quantile normalization. Analysis of differential expression was performed using a linear model fitting (LIMMA package; reference 58). The obtained p-values were corrected for multiple testing using the false discovery rate method, and a threshold of 0.05 was applied. Microarray data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus under accession no. GSE10470.

Crude nuclear extracts were purified by dialysis using Slide-A-Lyzer MINI Dialysis Units (Thermo Fisher Scientific). The following oligonucleotide probes (Invitrogen) corresponding to the ATF4 binding site of the TRB3 gene were used: TRB3WT, 5′-GATTAGCTCAGGTTTACATCAGCCGGGCGGGGA-3′; and TRB3MUT, 5′-GATTAGCTCAGTCTAAACCTATAGGGGCGGGGA-3′. The oligonucleotides were annealed with complementary DNA and radiolabeled with γ-[³²P]ATP using T4 polynucleotide kinase. After incubation with nuclear extracts, the DNA–protein complexes were resolved in 5% polyacrylamide gels, and the signal was visualized using a PhosphorImager (Fujifilm).

Formaldehyde cross-linking and chromatin immunoprecipitation were performed as described previously (59). The cross-linked chromatin suspension was sonicated using a Sonicator 3000 (Mosonix). Immunoprecipitation was performed with an anti-myc antibody. DNA–protein cross-linking was reversed, followed by an overnight incubation at 65°C. DNA was isolated by phenol/chloroform extraction and subjected to PCR analysis using the primers 5′-GGTCACAGATGGTGCAATCC-3′ and 5′-AACTGAGCAGCTCTCGGAGTC-3′.

Concentrations of reduced glutathione were measured using HPLC (PerkinElmer) equipped with an eight-channel coulometric array detector (ESA, Inc.). Cells were lysed in 5% (wt/vol) metaphosphoric acid and centrifuged at 10,000 g for 10 min to sediment protein. Cell-pellet precipitates were saved for protein determinations. Glutathione concentrations of supernatant fractions were determined by injecting 5-μl aliquots onto an Ultrasphere 5 u, 4.6 × 250 mm, C18 column (Beckman Coulter), and eluting with a mobile phase of 50 mM NaH₂PO₄, 0.05 mM octane sulfonic acid, 1.5% acetonitrile (pH 2.62) at a flow rate of 1 ml/min. Peak areas were analyzed using software from ESA, Inc. Intracellular generation of ROS was determined by DCF. Cortical neurons were seeded in 96-well fluorescence plates. After the treatments indicated in the figures, the cultures were incubated for 45 min with 1.25 μM DCF-H₂. Cells were washed with HBSS and read on a plate reader (Fluorometer; MDS Analytical Technologies) at 485-nm excitation and 530-nm emission wavelengths.

Transient focal cerebral ischemia was induced by MCAo using the intraluminal filament method, as previously described (23). Male mice were anesthetized. A small incision was performed in the skin covering the scalp, and a fiber optic probe was glued to the parietal bone and connected to a laser-Doppler flowmeter (Periflux System 5010; Perimed) for continuous monitoring of cerebral blood flow. For MCAo, a heat-blunted black monofilament surgical suture (6-0) was inserted into the exposed external carotid artery, advanced into the internal carotid artery, and wedged into the circle of Willis to obstruct the origin of the MCA. The filament was left in place for 45 min and withdrawn.

Animal physiology was assessed in ATF4^+/+ and ATF4^−/− mice (n = 3 each). For this purpose, the left femoral artery was cannulated using a polyethylene tube (PE-10; BD Biosciences) to record mean arterial blood pressure. Blood samples were taken for chemical analysis using a hand-held blood analyzer (iSTAT; Abbot Laboratories).

Coronal brain sections were serially cut in a cryostat (Leica) and stained with cresyl violet to identify viable tissue. To correct for the effect of edema, the infarcted area was determined indirectly by subtracting the area of the healthy tissue in the ipsilateral hemisphere from the area of the contralateral hemisphere on each section. Infarction volume was calculated by integration of infarct areas measured in 20 equidistant brain sections that encompassed the whole lesion. Volumes from all sections were summed to calculate total infarct volume.

To assess the morphology of major cerebral blood vessels of the circle of Willis, deeply anesthetized mice were perfused transcardially with a prewarm (37°C) saline solution containing gelatin (20% wt/vol) and India ink (0.25% vol/vol).

Neurological scores were assigned the following values: 0, normal motor function; 1, flexion of torso and contralateral forelimb when the mouse was lifted by the tail; 2, circling to the contralateral side when the mouse was held by the tail on a flat surface, but normal posture at rest; 3, leaning to the contralateral side at rest; and 4, no spontaneous motor activity. The inclined plane test measured the time a mouse managed to hold itself on an inclined glass plate angled at 50° before sliding down. The hanging wire test determined the time it took an animal to cross a distance of 45 cm on a freely hanging narrow metal bar.

Table S1 compares the array data from fibroblasts (11) with the microarray data obtained from neurons. Table S2 summarizes physiological data in ATF4^+/+ and ATF4^−/− mice. Table S3 shows a complete list of all genes differentially expressed after statistical analysis, including p-values and gene ontology information, and a list with the complete gene ontology analysis. Fig. S1 shows measurements of DCF fluorescence in ATF4^+/+ and ATF4^−/− neurons in response to HCA. Fig. S2 shows measurements of cerebral blood flow, body temperature, and blood pressure during MCAo.

The authors wish to thank Jawed Alam for providing us with the expression plasmids for ATF4. The ATF4 5′UTR reporter plasmid was a gift from David Ron. David Ash provided us with recombinant arginase. We thank Brett Langley, Ambreena Siddiq, JoAnn Gensert, Alvaro G. Estevez, Jin Son, Bruce Volpe, and Gary Gibson for sharing various materials and for their invaluable advice and encouragement throughout this study. This work was possible because of the outstanding technical support from Hsin-Hwa Lee, Chu-Hui Peng, Lucretia Batten, and Elizabeth Newman.

This study was supported by the W.M. Burke Foundation, by a grant from the National Institutes of Health to Rajiv R. Ratan (NS40591), and by the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation. Philipp S. Lange is a recipient of a postdoctoral fellowship from the Deutsche Forschungsgemeinschaft (LA 1483/1-1) and the Koeln Fortune Program/Faculty of Medicine, University of Cologne (125/2003). Juan C. Chavez is the recipient of a Scientist Development Grant from the Northeast Affiliate of the American Heart Association (0635556T).

The authors have no conflicting financial interests.