Cytoprotection by pre-emptive conditional phosphorylation of translation initiation factor 2

Stress associated with eIF2α phosphorylation protects HT22 cells against oxidative glutamate toxicity

Extracellular glutamate depletes intracellular cysteine and glutathione, leading to oxidative glutamate toxicity (Murphy et al, 1989), a process believed to play a role in neuronal cell death mediated by extracellular glutamate (Schubert and Piasecki, 2001). HT22 cells are an immortal mouse hippocampal cell line that is resistant to glutamatergic excitotoxicity but susceptible to oxidative glutamate toxicity and represents a useful model system for studying cellular responses to oxidative stress (Morimoto and Koshland, 1990; Maher and Davis, 1996; Tan et al, 2001). To test the hypothesis that eIF2α phosphorylation and activation of the ISR promotes resistance to oxidative glutamate toxicity, we first measured the ability of manipulations known to promote eIF2α phosphorylation to protect HT22 cells from the lethal effects of glutamate. Pretreatment of HT22 cells with tunicamycin or thapsigargin, ER stress-inducing agents that cause eIF2α phosphorylation and activate the ISR (Harding et al, 2003), was protective against subsequent challenge with glutamate (Figure 1A).

The stress paradigm used, 1 h treatment with the ER stress-inducing drug followed by 5 h recovery before exposure to glutamate, was sufficient to induce both eIF2α phosphorylation and subsequent expression of the ISR targets ATF4, GADD34 and CHOP (Figure 1B). The survival benefit was dose-dependent and the protective effect of the different stressors correlated with their ability to promote eIF2α phosphorylation (data not shown). The brief interval between the preconditioning ER stress and application of glutamate favors a true survival advantage afforded by the preceding stress and discounts selection for stress-resistant cells and their proliferation at the expense of a susceptible sub-population as the basis for the increased survival of the pretreated cells. We conclude that stressful manipulations that increase the level of phosphorylated eIF2α promote resistance to glutamate toxicity in HT22 cells.

To further examine the correlation between eIF2α phosphorylation and resistance to glutamate, we measured the impact of blocking eIF2α phosphorylation in HT22 cells. To this end, we overexpressed an activated form of GADD34, the regulatory subunit of an eIF2α-specific holophosphatase complex, in HT22 cells (Novoa et al, 2001). As predicted, HT22 cells that stably overexpressed an active fragment of GADD34 had no detectable phosphorylated eIF2α under basal conditions or when stressed (Figure 2A). Inhibition of eIF2α phosphorylation markedly compromised cell survival in regular media. However, the defect was rescued by adding the reducing compound β-mercaptoethanol (βME) to the media (Figure 2B and C). These observations indicate a role for eIF2α phosphorylation in the basal resistance of HT22 cells to oxidative stress and are consistent with the phenotype of mouse embryonic fibroblasts (MEFs) with a compromised ISR (Harding et al, 2003).

Cells cultured in media containing anti-oxidants such as βME are resistant to cell death induced by 5 mM glutamate (Miyamoto et al, 1989; Davis and Maher, 1994) (and data not shown). However, we found that susceptibility to glutamate toxicity could be restored if cells previously growing in the media containing βME were switched to media lacking βME during the exposure to glutamate. About 20% of parental HT22 cells, previously cultured in βME, were dead after 24 h exposure to 5 mM glutamate in media lacking βME (Figure 2D). This compared with ∼65% death following a similar exposure in cells that had not been cultured previously in βME -containing media (Figures 1A and 4C). In this paradigm of switch from βME (+) to βME (−), GADD34-overexpressing HT22 cells were hypersensitive to cell death induced by glutamate treatment. Only 70% of such cells survived 24 h exposure in 2 mM glutamate, as compared to 100% survival of the inactive GADD34^ΔC-overexpressing control cells or cells transduced with empty vector. Following exposure to 5 mM glutamate, there was 36% survival of GADD34-overexpressing HT22 cells compared with approximately 80% survival in the control cell lines (Figure 2D). Withdrawal of the βME for the period of time used for glutamate exposure did not compromise survival of the GADD34-expressing HT22 or the control cells (data not shown). This experiment indicates that blocking eIF2α phosphorylation sensitizes HT22 cells to oxidative glutamate toxicity.

A ligand-activated eIF2α kinase that promotes conditional translational inhibition and downstream target gene expression

The experiments described above establish a correlation between eIF2α phosphorylation and resistance to oxidative glutamate toxicity. However, many signaling pathways are activated in stressed cells and it was therefore impossible to attribute the protective effect of the preconditioning preceding stress specifically to eIF2α phosphorylation and activation of the ISR. To uncouple eIF2α phosphorylation from stress, we took advantage of the fact that activation of PERK is initiated by oligomerization, which leads to trans-autophosphorylation and increased ability to phosphorylate its substrate, eIF2α. Normally, this oligomerization event is driven by the stress-sensing ER lumenal domain of PERK; however PERK activation can be subordinated to any heterologous oligomerization domain (Bertolotti et al, 2000; Liu et al, 2000). Therefore, we fused the eIF2α kinase effector domain of PERK to a polypeptide containing two modified FK506 binding domains (Fv2E) and expressed the fusion protein, Fv2E-PERK, as a soluble cytoplasmic protein in HT22 cells (Figure 3A). The modified FKBP domain (Fv2E) is a high-affinity receptor for a cell-permeant synthetic small organic molecule, AP20187, that can be engaged simultaneously by two modified Fv2E domains (Spencer et al, 1993). The bivalent ligand thereby serves as a dimerizer/oligomerizer, and when added to cells can induce clustering of Fv2E-containing fusion proteins.

Because it lacks an ER lumenal domain, Fv2E-PERK does not respond to ER stress but allows for drug-induced activation of the eIF2α kinase domain of PERK without subjecting the cells to ER stress. Addition of AP20187 to Fv2E-PERK-expressing HT22 cells resulted in rapid activation of the kinase as reflected in a shift to a slower mobility trans-autophosphorylated form (Figure 3B). Fv2E-PERK remains phosphorylated for the duration of exposure of cells to AP20187. Furthermore, phosphorylation of endogenous eIF2α approaches saturation within 15 min and expression of the ISR target genes, GADD34 and CHOP, was observed within 2 h of drug treatment. AP20187 has no impact on eIF2α phosphorylation or target gene expression in parental HT22 cells lacking Fv2E-PERK (Figure 3B), and Fv2E-PERK does not require endogenous PERK for its activity nor does it activate endogenous PERK (see below, Figure 6B). As expected, treatment of Fv2E-PERK-expressing cells with AP20187 also results in translational inhibition demonstrated by reduced ³⁵S-methionine incorporation into newly synthesized protein. The magnitude of the reduction is similar to those observed in ER stressed cells (Figure 3C).

The Fv2E-PERK transgenic cells allowed us to survey the gene expression program activated by an eIF2α kinase under conditions in which parallel stress-induced signaling pathways were not activated. To facilitate comparison between the profiles of genes expressed in AP20187-treated Fv2E-PERK cells and a previously performed survey of the ISR (Harding et al, 2003), we carried out the expression microarray analysis in Fv2E-PERK containing MEFs (see below, Figure 6). The mRNA level of 506 genes was increased two-fold or more at 4 or 8 h following AP20187 treatment, compared with the untreated Fv2E-PERK positive cells (Figure 3E and Table I).

This AP20187-dependent response overlapped with the response to the ER stress-inducing drug tunicamycin. Furthermore, the overlap between the two responses substantially coincided with the portion of the ER stress response that is dependent on eIF2α phosphorylation, as it is compromised in Perk−/− cells (Figure 3E, Group A, Table I). This subset of genes encodes enzymes involved in amino acid metabolism, thiol metabolism and sufficiency, and resistance to oxidative and other cellular stresses (Table I): all recognized target genes of the ISR. Remarkably, many genes encoding ER chaperones and other classical targets of the ER stress response such as genes involved in protein degradation were also activated by Fv2E-PERK. The latter observations are consistent with a prominent role of eIF2α phosphorylation in controlling such classic unfolded protein response genes, which was revealed previously by analysis of cells with mutations affecting the ISR (Scheuner et al, 2001; Harding et al, 2003) (Table I).

To determine directly the role of eIF2α phosphorylation in activated gene expression by Fv2E-PERK, we expressed the chimeric protein in fibroblasts derived from mouse embryos homozygous for a serine 51 to alanine mutation of endogenous eIF2α gene (eIF2α^A/A, the wild type being eIF2α^S/S). This mutation abolishes the phosphorylation site used by eIF2α kinases and deregulates eIF2 activity (Scheuner et al, 2001). Despite expressing high levels of active Fv2E-PERK, AP20187-treated eIF2α^A/A cells have no phosphorylated eIF2α and are unable to activate the downstream targets ATF4 and GADD34 (Figure 3D). Expression microarrays showed that the cluster of genes induced by both Fv2E-PERK activation and by ER stress in wild-type (eIF2α^S/S) cells (Group A in Figure 3E) failed to undergo induction in the Fv2E-PERK-expressing mutant eIF2α^A/A cells, thus establishing an essential role for eIF2α phosphorylation in the activation of the ISR by Fv2E-PERK.

A smaller subset of genes that were induced by ER stress but not by Fv2E-PERK activation was also identified (Figure 3E, Group B); these presumably represent targets of parallel pathways active in the ER stress response (Harding et al, 2002; Kaufman, 2002). A third category of genes was prominently induced by activation of Fv2E-PERK in wild-type eIF2α^S/S cells, but less so by ER stress and not at all in mutant eIF2α^A/A cells. Among these were genes transiently induced by Fv2E-PERK (higher at 4 h than at 8 h), a subset that may be enriched in short-lived mRNAs stabilized by translational repression, which peaks early in the AP20187-treated wild-type eIF2α^S/S cells but is later reversed as GADD34 is induced. This mechanism of gene activation is likely to be less important in tunicamycin-treated cells, because of the gradual onset of eIF2α phosphorylation and more modest translational repression effected by that drug (Harding et al, 2000b). Such differences may account for the larger number of genes activated by Fv2E-PERK than by tunicamycin (data not shown and supplementary tables).

AP20187-induced Fv2E-PERK activation protects cells against oxidative glutamate toxicity, peroxynitrite donor and ER stress

Treatment of Fv2E-PERK-expressing HT22 cells with AP20187 for as little as 15 min was sufficient to induce eIF2α phosphorylation (Figures 3B and 4A). Subsequent activation of downstream gene expression was enhanced after the drug was withdrawn, as translation rates were recovering (Figure 4A and B). Thus, a short pulse of AP20187 followed by withdrawal of the drug mimics the kinetic profile of a response of cells to stress (Brostrom and Brostrom, 1998; Novoa et al, 2003). Therefore, cells were exposed to AP20187 for 15 min, the dimerizer was removed and, following a variable recover period of 5 or 12 h, the cells were exposed to glutamate. AP20187 pretreatment protected Fv2E-PERK-expressing HT22 cells from subsequent exposure to glutamate, but did not affect the survival of parental HT22 cells (Figure 4C and D). The survival advantage of pretreatment correlated with the dose of dimerizer used and was still apparent 12 h after withdrawal of the dimerizer, a point at which translation had recovered to nearly normal levels (Figure 4B).

Oxidative glutamate toxicity is attributed to inhibited import of cystine and consequent glutathione depletion, resulting in accumulation of endogenous peroxides (Murphy et al, 1989). To determine if eIF2α phosphorylation affects the development of oxidative stress or acts downstream of it, we measured levels of endogenous peroxides induced by glutamate using dichlorofluorescin (DCF), a cell-permeant molecule that is oxidized by endogenous peroxides and other oxidants to the fluorescent dichlorofluoresceine. AP20187 had no measurable effect on DCF fluorescence in parental HT22 cells (Figure 5A and B) whereas glutamate treatment resulted in a profound increase in DCF fluorescence as detected by FACscan (Figure 5C and D). DCF fluorescence levels fell only when the cells died, an event reflected by permeability to propidium iodide (Figure 5 A–D).

AP20187 pretreatment of Fv2E-expressing cells did not affect DCF fluorescence in cells that were not exposed to glutamate (Figure 5E and F), but markedly reduced DCF fluorescence in the cells exposed to glutamate (Figure 5G and H). Treatment with AP20187 caused some death among the Fv2E-PERK-expressing cells (compare the upper left quadrants in Figure 5E and F), as predicted by the pro-apoptotic effects of another eIF2α kinase PKR (Srivastava et al, 1998). However, in cells exposed to glutamate, AP20187 dramatically reduced the fraction of propidium iodide positive (dead) cells (compare the upper left quadrants in Figure 5G and H). AP20187 pretreatment also protects against glutamate-induced glutathione depletion (Figure 5I). The latter observation links the protective effect of the eIF2α phosphorylation induced by Fv2E-PERK to the previously established role of oxidative glutamate toxicity in glutathione depletion (Tan et al, 1998a). These experiments show that activation of an eIF2α kinase prevents oxidative stress and cell death in HT22 cells exposed to glutamate.

To measure the effect of Fv2E-PERK activation on the sensitivity of cells to other stressful conditions, we introduced the transgene into immortalized mouse embryo fibroblasts, which are sensitive to both ER stress and nitrosative stress. In these cells too, AP20187 treatment activated Fv2E-PERK, phosphorylated eIF2α and activated downstream genes. This is apparent in wild-type cells and even more so in Perk−/− cells which, lacking the endogenous eIF2α kinase activity, were able to tolerate higher levels of Fv2E-PERK (Figure 6A and B). AP20187 protected the Fv2E-PERK transgenic cells with a wild-type eIF2α genotype (eIF2α^S/S) against the lethal effects of the ER stress-inducing agent tunicamycin and the peroxinitrite donor SIN-1 (Figure 6C and D). No protection was afforded by AP20187 to the Fv2E-PERK transgenic cells with the serine 51 to alanine mutation in eIF2α (eIF2α^A/A).

Cell culture and generation of stable cell lines

The HT22 subclone of HT-4 cells were a kind gift of Pamela Maher, and were maintained at approximately 50% confluence in regular media consisting of DMEM (Life Science Technologies), 10% fetal calf serum (FCS, Atlanta Biologicals), 1 × pen/strep and 2 mM L-glutamine (Life Science Technologies) as described (Maher and Davis, 1996). Wildtype, Perk−/− and eIF2α^A/A MEFs immortalized with SV40 large T antigen were grown in the same medium (Harding et al, 2000b; Scheuner et al, 2001).

Cells expressing active GADD34 were produced by transducing HT22 cells with pBabe, a retrovirus encoding the active A1 fragment of hamster GADD34 (Novoa et al, 2001). GADD34^ΔC-overexpressing cells were made by transducing HT22 cells with retrovirus encoding the inactive GADD34 mutant lacking the C-terminal 121 amino acids, which contain the domain required for binding the catalytic subunit of the phosphatase. The transduced HT22 cells were selected for stable expression of the GADD34 fragments with 4 μg/ml puromycin (Calbiochem) and maintained in regular media supplemented with 55 μM βME and 1 × MEM non-essential amino acids (NEAA) from Life Technologies. The level of expression of the active A1 fragment of GADD34 in pools of transduced cells used in these experiments was similar to that observed previously and is estimated to be several fold higher than the endogenous levels in stressed cells (Novoa et al, 2001).

The use of FK506 binding domains to create drug-activated fusion proteins has previously been described (Spencer et al, 1993). A modified version of the FK506 binding domain has recently been engineered to bind specifically to a synthetic dimerization-inducing ligand, AP20187, which is incapable of interacting with endogenous FK506 binding proteins (www.ariad.com/regulationkits). The Fv2E-PERK construct was made using the ARIAD Regulated Homodimerization Kit. Briefly, the portion of the mouse Perk cDNA encoding the kinase domain kinase domain (amino acids 537–1114) was fused to two tandem modified FK506 binding domains (Fv2E) from the pC₄M-Fv2E plasmid. The 5′ myristoylation signal encoded in pC₄M-Fv2E was removed and the Fv2E-PERK transgene was shuttled into a retroviral vector, pBabe (Morgenstern and Land, 1990), and used to produce replication-defective retrovirus as previously described (Landau and Littman, 1992). HT22 cells, Wildtype MEFs, Perk−/− MEFs and eIF2α^A/A MEFs were transduced with pBabe retrovirus encoding Fv2E-PERK, plated to clonal dilution, and selected for stable expression of the transgene by the addition of 4 and 2.5 μg/ml puromycin, respectively. Levels of Fv2E-PERK were consistently higher in pools of transduced Perk−/− MEFs and eIF2α^A/A MEFs than in the wild-type MEFs and the mutant cells stably express the transgene at high levels, whereas the wild-type cells tend to lose expression over time. These observations may reflect low levels of basal activity of the chimeric kinase (leak), which is well tolerated by cells lacking endogenous kinase activity (Perk−/−) and in cells with a mutant target (eIF2α^A/A) but is poorly tolerated by the wild-type cells. Several cell lines expressing equal levels of Fv2E-PERK were established from the clonal populations and these were used in the functional studies shown in Figures 3 and 6. All the assays described in this manuscript were performed in at least two different Fv2E-PERK cell lines and in the parental cells. The data shown in the figures are from representative cell lines.

Survival assays

Cells were plated at a density of 2.5–5.0 × 10³ cells/96-well plate or 1.5–3.0 × 10⁴ cells/24-well plate. After pretreatment with AP20187 (ARIAD) or mock pretreatment with an equal volume of the carrier ethanol, HT22 cells were washed once in regular media and then maintained in regular media for the duration of recovery. Following the recovery period, the media were replaced with fresh media containing L-glutamic acid (Sigma) for 24 h. MEFs were mock pretreated or pretreated for 18 h with AP20187 and then exposed to 3-morpholinosydnonimine (SIN-1, Sigma) for 6 h. The cells were switched to media containing 0.5 mg/ml MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyl tetrazolium bromide, Sigma) for 4 h and the reaction was stopped by the addition of an equal volume of stop buffer (50% isopropanol, 10% SDS, 10 μM HCl). The formazan crystals were allowed to dissolve during a 16 h incubation at 37°C and the optical density was read at 550 nm. Experiments were performed in duplicate or triplicate and repeated twice.

For the colony outgrowth assay, HT22 cells were plated at a density of 1 × 10³ cells/6-well plate and maintained for 6 days in non-supplemented media or media supplemented with 55 μM βME. MEFs were plated at a density of 5 × 10³ cells/6 well plate and pre-treated with 1 nM AP20187 for 7 h, at which time tunicamycin was added for an additional 16 h (0.25 μg/ml for wild type, eIF2α^S/S cells and 0.025 μg/ml for the mutant eIF2α^A/A cells). The cells were allowed to recover and resume growth and were fixed with 4% formaldehyde 6 days later prior to staining with crystal violet. The data shown are representative wells from experiments repeated four times.

Array analysis

Two independently isolated Fv2E-PERK-expressing Perk−/− transformed MEF cell pools of each treatment group (0, 4, or 8 h with 2 nM AP20187) were used to isolate total RNA using the guanidine thiocyanate–acid–phenol extraction method. The use of Perk−/−, Fv2E-PERK(+) cells was favored because of the higher level of Fv2E-PERK expression. The comparison group consisted of mutant eIF2α^A/A cells expressing comparable levels of Fv2E-PERK (Figure 3D). Fluorescent labeled RNA probes for each of the samples were prepared and hybridized to Affymetrix mouse U74Av2 high-density oligonucleotide arrays as previously described (Lockhart et al, 1996). Primary image analysis of the arrays was performed using the Genechip 3.2 software package (Affymetrix, Santa Clara, CA). The raw data from the hybridization experiments were analyzed by GeneSpring™. The raw signal strength from each gene was normalized to the mean signal strength of all genes from the same chip to obtain the normalized signal strength. Then, to allow visualization of all data on the same scale for subsequent analysis, the normalized signal strength of each gene was divided by the median signal strength for that gene among all samples to obtain the normalized expression level. The data were compared to the results obtained from previously published data from wild type, Perk−/− or Atf4−/− untreated or tunicamycin-treated samples (Harding et al, 2003). Genes that were assigned a ‘present' signal flag (Genechip 3.2 software) in 8 h AP20187-treated Fv2E-PERK samples or tunicamycin-treated wild-type cells and that had a raw signal strength of greater than 50 in tunicamycin-treated wild-type cells were further analyzed. In all, 653 genes that were induced two-fold by either 4 or 8 h of AP20187 treatment of Fv2E-PERK cells or by 6 h of tunicamycin treatment in wild-type fibroblasts were identified; 506 of these were induced at least two-fold by AP20187 treatment in Fv2E-PERK cells and the remainder were induced by at least two-fold in tunicamycin-treated wild-type cells and to varying degrees in Fv2E-PERK cells (see supplementary tables). The complete data set has been submitted to the NCBI GEO database Accession GDSH05

Immunoblotting and translation assays

For analysis of Fv2E-PERK, eIF2α, ATF4, GADD34 and CHOP by immunoblotting 6 × 10⁵ cells were plated onto 10 cm dishes 16 h prior to treatment. Cell extracts were prepared using harvest buffer (10 mM HEPES pH 8.0, 50 mM NaCl, 0.5 M sucrose, 0.1 M EDTA, 0.5% Triton X-100) containing both protease inhibitors (1 mM dithiothreitol (DTT), 2 μg/ml pepstatin, 4 μg/ml aprotinin, 100 μM PMSF) and phosphatase inhibitors (10 mM tetrasodium pyrophosphate, 100 mM NaF, 17.5 mM β-glycerophosphate). The low-speed supernatant (1000 g) containing cytoplasmic proteins was collected and nuclear extracts were made by vortexing the nuclei for 15 min at 4°C in buffer containing 20 mM HEPES pH 7.9, 400 mM NaCl, 1 mM EDTA, 1 mM EGTA, 0.1% IGEPAL CA-630, plus protease inhibitors. Equal amounts of protein were loaded per lane in SDS–PAGE gels (30–50 μg of cytoplasmic extract, 20 μg of nuclear extract). Immunoprecipitation procedures for detecting endogenous PERK and immunoblotting procedures for PERK, eIF2α, ATF4, GADD34 and CHOP have previously been described (Bertolotti et al, 2000; Harding et al, 2000a, 2000b; Novoa et al, 2001).

To measure new protein synthesis, cells were labeled for 10 min with 50 μCi/ml ³⁵S-Translabel (ICN) in pulse media (90% DMEM without methionine and cysteine, 10% regular DMEM, 5% dialyzed FCS, 1 × pen/strep, 2 mM L-glutamine). The dishes were flooded with ice-cold PBS containing 1 mM cycloheximide, 0.6 mg/ml L-methionine and 0.96 mg/ml L-cysteine, washed twice in PBS, and harvested in 1% Triton X-100-containing buffer as described previously (Harding et al, 2000b). In all, 15 μg of protein/lane was loaded onto a 9% gel for detection of ³⁵S-methionine incorporation by autoradiography and 25 μg of protein/lane was loaded for detection of Fv2E-PERK, eIF2α and GADD34 by immunoblot.

Assays for endogenous peroxides and glutathione

Parental HT22 and Fv2E-PERK-expressing cells were plated at 1 × 10⁵ cells/60 mm dish and pretreated for 1 h with 1 nM AP20187 and allowed to recover in regular media for 5 h prior to the addition of 5 mM glutamate for 10 h. The media were removed to collect floating cells and pooled with cells trypsinized from the dishes. The cells were pelleted and resuspended in fresh media containing 10 μM 2′,7′-dichloro-dihydrofluoresceine diacetate (DCF, Molecular Probes) and incubated for 15 min at 37°C. Cells were then pelleted, washed once in ice-cold PBS–2% FCS, and resuspended in PBS–2% FCS containing 1 μg/ml propidium iodide (PI, Roche). After a 5-min incubation on ice, the cells were analyzed for DCF fluorescence (FL-1, green channel) and PI positivity (FL-3, red channel) by a Becton Dickson FACScan. In all, 10 000 ungated cells were counted.

Glutathione levels were measured using the modified DTNB-GSSG reductase recycling assay (Griffith, 1980; Anderson, 1985).