Identification of caspase 3-mediated cleavage and functional alteration of eukaryotic initiation factor 2alpha in apoptosis.

Induction of apoptosis in a variety of cell types leads to inhibition of protein synthesis. Recently, the cleavage of eukaryotic translation initiation factor 4G (eIF4G) by caspase 3 was described as a possible event contributing to translation inhibition. Here, we report the cleavage of another initiation factor in apoptotic cells, eIF2alpha, that could contribute to regulation of translation during apoptosis. This cleavage event could be completely inhibited by pretreatment of HeLa cells with Z-VAD-fmk. In vitro analysis using purified eIF2 and purified caspases showed cleavage of eIF2alpha by caspase 3, 6, 8, and 10 but not 9. Caspase 3 most efficiently cleaved eIF2alpha and this could be inhibited by addition of Ac-DEVD-CHO in vitro. Comparison of cleavage of phosphorylated versus nonphosphorylated eIF2alpha revealed a modest preference of the caspases for the nonphosphorylated form. When eIF2. 2B complex was used as substrate, only caspase 3 was capable of eIF2alpha cleavage, which was not affected by phosphorylation of the alpha subunit. The eIF2.GDP binary complex was cleaved much less efficiently by caspase 3. Sequence analysis of the cleavage fragment suggested that the cleavage site is located in the C-terminal portion of the protein. Analysis showed that after caspase cleavage, exchange of GDP bound to eIF2 was very rapid and no longer dependent upon eIF2B. Furthermore, in vitro translation experiments indicated that cleavage of eIF2alpha results in functional alteration of the eIF2 complex, which no longer stimulated upstream AUG selection on a mRNA containing a viral internal ribosome entry site and was no longer capable of stimulating overall translation. In conclusion, we describe here the cleavage of a translation initiation factor, eIF2alpha that could contribute to inhibition or alteration of protein synthesis during the late stages of apoptosis.

Initiation of protein synthesis in mammalian cells is a highly regulated process that requires multiple translation initiation factors. The eukaryotic translation initiation factor 2 (eIF2) 1 plays a key role in the initiation of translation. eIF2 is a heterotrimeric protein consisting of three subunits ␣, ␤, and ␥, which forms a so-called ternary complex with GTP and the initiator Met-tRNA. The ternary complex interacts with the 40 S ribosomal subunit thereby forming a 43 S preinitiation complex that is capable of recognizing the proper start codon of the mRNA. It has been shown in yeast that eIF2 itself is involved in this process of start codon selection (1) indicating its importance in translation initiation. Upon joining of the 60 S ribosomal subunit to the 43 S preinitiation complex, the GTP moiety is hydrolyzed and an eIF2⅐GDP complex is released from the ribosome (2). Participation of the eIF2⅐GDP complex in a new round of translation requires the exchange of the GDP moiety for a new GTP molecule, a process carried out by the guanine nucleotide exchange factor, eIF2B. The global rate of protein synthesis in eukaryotes is mainly regulated by the specific phosphorylation of Ser-51 of the eIF2␣ subunit. eIF2␣ is known to be phosphorylated by kinases such as PKR (double-stranded RNA dependent eIF2 kinase) in response to viral infection, and stress conditions (3), HRI (heme-regulated inhibitor) in response to hemin availability and a host of environmental stress conditions (4 -6) and GCN2 from the yeast Saccharomyces cerevisiae in response to amino acid starvation (7). Phosphorylated eIF2 binds eIF2B tightly, forming a poorly dissociable eIF2(␣P)⅐eIF2B complex. Since eIF2B exists in relatively low molar quantities with respect to eIF2 in the cytoplasm in many systems, phosphorylation of as little as 25% of eIF2␣ can be sufficient to sequester virtually all the available eIF2B, thereby blocking the eIF2B catalyzed recycling of the eIF2⅐GDP and subsequently inhibiting protein synthesis completely (8).
It has been well established that protein synthesis is inhibited during apoptosis or programmed cell death but the mechanism of translation inhibition has remained unclear until recently. We and others have recently shown that eukaryotic translation initiation factor 4GI (eIF4GI), which is required for binding of capped mRNA to ribosomal 43 S subunits, was cleaved by caspase 3 during apoptosis (9,10). Similar cleavage of eIF4GI is one of the major causes of translation inhibition during enterovirus infection of cells (11). Likewise, the cleavage of eIF4GI correlates with the observed inhibition of protein synthesis in apoptotic cells (9). Other reports indicate that phosphorylation of eIF2␣ could represent another mechanism by which translation is inhibited during apoptosis. For in-stance, overexpression of PKR in transfected cells resulted in elevated levels of phosphorylated eIF2␣ concomitant with reduced protein synthesis and induction of apoptosis in these cells (12). In addition, it has been reported that PKR is a tumor suppressor protein as it was shown that overexpression of mutant PKR incapable of phosphorylating eIF2␣ resulted in tumor formation (13,14). Similarly, others have described that lowering the level of the eIF2 associated protein, p67, resulted in increased levels of phosphorylated eIF2␣ concomitant with induction of apoptosis in these cells (15). In both cases, the decrease in the overall rate of protein synthesis was ascribed to the phosphorylation of eIF2␣.
Here, we report the cleavage of eIF2␣ by caspase 3 during the late stages of the execution phase of apoptosis in HeLa cells, K562 cells, and Jurkat T cells. The cleavage of the 38-kDa ␣ subunit resulted into a 36-kDa cleavage product as observed by both Coomassie stain and immunoblot analysis. In vitro experiments using purified eIF2 and purified caspases showed that eIF2␣ could be a substrate for caspases 3, 6, 9, and 10 but not caspase 8. Although cleavage by caspase 3 appeared to be most efficient we do not rule out the possibility that other caspases might be responsible for eIF2␣ cleavage in vivo. In vitro experiments using both nonphosphorylated and phosphorylated eIF2␣ as a substrate showed that all caspases had a preference for the nonphosphorylated form of eIF2␣. Interestingly, when eIF2⅐2B complex was used as substrate no preference was observed between nonphosphorylated and phosphorylated eIF2␣. The cleavage site appears to be located in the C-terminal region of the protein as determined by N-terminal amino acid sequencing of the 36-kDa cleavage product. Close examination of this region revealed two putative caspase 3 cleavage sites which would lead to the release of a 14-or 11-amino acid peptide. Interestingly, results indicate that cleavage of eIF2␣ does inactivate the eIF2 complex as determined by in vitro translation assays.

EXPERIMENTAL PROCEDURES
Cells-HeLa S3 cells were grown as described (9). K562 cells were grown at 37°C in RPMI (Irvine Scientific) supplemented with 10% bovine calf serum, 0.5% fetal calf serum (Summit Biotechnology), 100 units of penicillin, and 100 g of streptomycin/ml (Sigma) in a humidified chamber containing 5% CO 2 . For induction of apoptosis, stock solutions of cisplatin, etoposide, cycloheximide (Sigma), or TNF-␣ (Peprotech) were diluted with medium and then incubated with cells at 37°C for the time indicated in each figure.
Preparation of Cell Lysates and Immunoblot Analysis of Fractions-Cell extracts were prepared as described (9). Briefly, HeLa cells were washed with phosphate-buffered saline, resuspended in CHAPS lysis buffer (20 mM Tris pH 7.2, 0.1 M NaCl, 1 mM EDTA, 10 mM dithiothreitol, 0.5% CHAPS, 10% sucrose) and incubated on ice for 30 min. Cell lysates were then centrifuged for 10 min at 10,000 ϫ g at 4°C, supernatants were collected and stored at Ϫ80°C. Cell pellets were resuspended in sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) loading sample buffer, boiled for 10 min, and subjected to SDS-PAGE. Detection of poly(ADP-ribose) polymerase (PARP) cleavage was accomplished by immunoblotting with a mouse monoclonal anti-PARP antibody (Zymed Laboratories Inc.). Detection of eIF4GI cleavage was performed by immunoblot as described previously (9). Ribosomal salt wash fractions from HeLa cells were prepared by standard methods.
Metabolic Labeling of Proteins-Proteins were labeled with [ 35 S]methionine (ICN) as described (9) to determine translation levels in the cells.
Expression and Purification of Caspases-The full-length cDNAs encoding human caspase 3, 6, 9, and the cDNAs encoding caspase 8 (Ser-217 to Asp-479) and caspase 10 (Ile-162 to Ile-479) cloned into pET15b (caspase 8), pET21b (caspases 6, 9, and 10), and pET23b (caspase 3) (all Novagen) were expressed in Escherichia coli BL21(DE3)pLysS. Caspases 3 and 8 were a kind gift of Dr. Claudius Vincenz; caspases 6, 9, and 10 were a kind gift of Dr. Emad Alnemri. The expressed proteins were purified by affinity chromatography on TALON metal affinity resin (CLONTECH) according to the manufacturer's instructions. Working concentrations of active caspases were standardized based on in vitro colorimetric cleavage assays using appropriate substrate peptides coupled to para-nitroanilide (pNA) (9). Caspase activity units are defined as the amount of caspase required to hydrolyze 1 nmol of pNA substrate/min. Various caspases were used at concentrations ranging from 10 to 60 g/ml as determined appropriate by these assays.
eIF2␣ Cleavage Assays-Purified rabbit eIF2 (1 l) was incubated at 37°C with purified recombinant caspase 3 (1 l) in a total volume of 10 l for the time points indicated in each figure. Samples were then analyzed by SDS-PAGE and immunoblotting using a mouse monoclonal antiserum specific for eIF2␣ (20). Caspase 3-treated eIF2 for in vitro translation assays was prepared similarly as described above with the following modification: after incubation at 37°C for 8 h, caspase 3 was inactivated by the addition of the caspase inhibitor Ac-DEVD-CHO (10 M) (Quality Controlled Biochemicals) for 1 h at 37°C and subsequently stored at Ϫ80°C.
Phosphorylation of eIF2-Purified rabbit eIF2 (10 l) or eIF2⅐2B (1 l) was incubated for 1 h at 37°C with HRI (2 l) in buffer containing 10 mM Tris (pH 7.4), 50 mM NaCl, 2 mM MgCl 2 , 0.1 mM ATP (final volume 30 l). Three microliters of phosphorylated eIF2 or eIF2⅐2B was then used in cleavage assays as described above in a final reaction volume of 30 l.
GDP Binding and Guanine Nucleotide Exchange Assay-To saturate eIF2 with GDP for cleavage studies, eIF2 (1 l) was incubated for 10 min at 37°C in the presence of 30 M GDP. To "freeze" the GDP on the eIF2 and stabilize the conformation of eIF2, MgCl 2 was added to a final concentration of 1 mM. For each 20-l aliquot of guanine nucleotide exchange assay mixture, ϳ1.5 pmol of eIF2 was incubated in the presence or absence of 18 units of caspase 3 (Upstate Biotechnology) in buffer containing 20 mM Tris⅐HCl (pH 7.4), 100 mM NaCl, 1 mM dithiothreitol, 0.1 mM EDTA, 0.1% CHAPS, and 10% polyethylene glycol for 3 h at 37°C. [ 3 H]GDP (5 M) was then added and after 10 min of additional incubation, 3 mM Mg(OAc) 2 was added to stabilize the eIF2⅐lsqb] 3 H]GDP complex. After a further 5-min incubation, 100 M unlabeled GDP was added and the amount of eIF2⅐[ 3 H]GDP complex remaining was determined after various times of incubation by filtration of a 20-l aliquot of each assay mixture through nitrocellulose filters as described previously (17).
Isoelectric Focusing-To determine the phosphorylation status of eIF2␣ in vivo, 5 ϫ 10 6 cells from apoptotic or control cultures were taken and resuspended in 200 l of VSIEF buffer (9.5 M urea, 5% CHAPS, 50 mM sodium fluoride, 5% ␤-mercaptoethanol). Samples were then clarified and subjected to isoelectric focusing on vertical isoelectric focusing gels by a slight modification of the method of Maurides et al. (21) as described previously (22).
Plasmid Constructs-T7TS plasmid (gift of P. A. Krieg) contains a multiple cloning site upstream of the Xenopus ␤-globin 3Ј-UTR and a track of 30 adenosyl residues which allows generation of poly(A) mRNAs. The region encoding the 5Ј-untranslated region (5Ј-UTR) of foot and mouth disease virus (FMDV) coupled to the reporter gene chloramphenicol acetyltransferase (CAT) as described (23) was excised from FMDV-CAT with EcoRV and BamHI and subsequently cloned into blunt-ended HindIII-BglII-digested pT7TS plasmid to generate pT7TSFMDVCAT. pT7TSglobinCAT was generated by insertion of a HindIII-SmaI globin-CAT fragment into a HindIII-EcoRV-digested pT7TS plasmid. Plasmid DNAs were linearized with XbaI, purified by phenol/chloroform extraction, and concentrated by alcohol precipitation for use in transcription reactions (see below).
Transcription and Translation-Transcription reactions for FMDV-CAT using T7 RNA polymerase (Promega) were carried out according to manufacturer's recommendations. For capped globin-CAT, transcription was performed using the mMessage mMachine T7 kit from Ambion. In both cases, the resulting RNAs were purified using spin columns (Eppendorf-5 Prime). Translation reactions were performed in rabbit reticulocyte lysate (Promega) according to the manufacturer's recommendations in a 20-l reaction volume. Addition of buffer control, caspase inhibitor Ac-DEVD-CHO, eIF2, caspase 3-treated eIF2 plus Ac-DEVD-CHO were added (2 l) to the translation reaction prior to the addition of [ 35 S]methionine (ICN), and incubated for 10 min at 30°C. After addition of the label, reactions were incubated for another 90 min at 30°C. Samples (2 l) were subjected to SDS-PAGE and analyzed by autoradiography. In addition, translation products were quantified by densitometry (Storm 860 PhosphorImager) or scanning autoradiographs and densitometry using NIH Image. Statistical analyses were performed using Excel.

RESULTS
Cleavage of eIF2␣ during Apoptosis-In order to identify possible target proteins that could contribute to translation inhibition during cell death, apoptosis was induced in multiple cell lines using a variety of apoptosis-inducing agents. We started with a system in which we have previously extensively characterized cleavage of eIF4GI (9) to determine if other factors are also modified. Thus, HeLa cells were treated with 100 M cisplatin and aliquots were harvested over a 24-h time period as indicated (Fig. 1A). Fig. 1A shows the eIF2␣ protein at 0 h with a molecular mass of approximately 38 kDa, at which time cisplatin (100 M) was added to the HeLa cell culture. Subsequent lanes show that cisplatin treatment of HeLa cells led to cleavage of eIF2␣ starting between 6 and 7 h postinduction, neared completion at 24 h and generated a cleavage fragment of about 36 kDa in size. The cleavage of eIF2␣ appears to initiate later than cleavage of the well defined caspase 3 substrate PARP as shown in Fig. 1B and eIF4GI as shown in Fig. 1C. Generation of the characteristic 85-kDa caspase 3-derived PARP cleavage product and eIF4GI cleavage products can be detected as early as 4 h post-induction. We noted that the eIF2␣ cleavage fragment is barely detectable at 24 h postinduction. Two factors are likely to contribute to the poor detection of the eIF2␣ cleavage product: (i) we have noted that the anti-eIF2␣ monoclonal antibody has a lower affinity for the cleavage fragment of eIF2␣ as compared with intact eIF2␣; and (ii) the stability of the cleaved eIF2␣ subunit appears to be lower than the intact protein in vivo. In addition, treatment of HeLa cells with etoposide, cycloheximide, or tumor necrosis factor-␣ (TNF-␣), K562 cells with cisplatin, etoposide, cycloheximide, or MG132, or Jurkat T cells with cisplatin, etoposide, cycloheximide, MG123, or TNF-␣ all resulted in the generation of a similar 36-kDa cleavage product (data not shown), indicating that this type of response can be generated in a wide variety of cell types using different inducers of apoptosis.
The loss of approximately 2 kDa in molecular mass implies that only a small portion of either the N or C terminus of eIF2␣ is removed, which raises the question whether this cleavage event results in inactivation of the eIF2 protein complex and could contribute to inhibition of protein synthesis or may modulate eIF2 activity in another way. Therefore, we further examined whether the cleavage of eIF2␣ correlated with translation inhibition in apoptotic cells. Cisplatin-treated HeLa cells were pulse labeled at the time points indicated for 30 min, subsequently lysed and analyzed by SDS-PAGE. Fig. 1D shows that between 3 and 5 h post-induction the incorporation of [ 35 S]methionine into protein decreased significantly, followed by an abrupt further decline to near zero between 6 and 7 h. Much of this translation decrease is likely due to rapid cleavage of eIF4GI or other factors, however, the latter abrupt decline may correlate with the appearance of eIF2␣ cleavage products.
Cisplatin treatment of HeLa cells produces apoptosis, eIF4GI cleavage, and translation shutoff faster than most other inducer/cell combinations we have tested. Thus, in order to reveal possible correlations between eIF2␣ cleavage and translation shutoff, we explored other apoptotic systems in which eIF4GI cleavage was not so rapid. Fig. 2 summarizes data taken from HeLa cells or K562 erythroblastoid cells treated with etoposide and compares those to data taken from cisplatin-treated HeLa cells. Etoposide treatment of HeLa or K562 cells resulted in slower induction of apoptosis and eIF4GI cleavage, yet the kinetics of cleavage of eIF2␣ were nearly equivalent. In both cases, translation rates declined as cleavage of eIF2␣ proceeded. In particular, 80% of eIF4GI was still intact in K562 cells after 16 h treatment, yet translation rates had declined to 50%. Since studies with poliovirus-infected cells have shown that translation rates can still proceed at near 100% efficiency when up to 80 -90% of eIF4GI has been cleaved (24), this implies that additional events are occurring in apoptosis to inhibit translation. At this same 16-h time point, eIF2␣ was 32% cleaved, suggesting a possible contribution to the translation inhibition phenotype. Interestingly, by 24 h, translation was still occurring in etoposide-treated cells even though 70 -80% of eIF2␣ had been processed. This may reflect the observation (see Fig. 9 below) that eIF2 complex containing cleaved eIF2␣ rapidly exchanges GDP, unlike its unprocessed precursor. Despite testing several combinations of cell types and apoptosis inducers, we have not discovered conditions where eIF2␣ cleavage precedes or occurs independently of eIF4GI cleavage. In summary, these data reveal that cleavage of eIF4GI and eIF2␣ often occur simultaneously, cleavage rates are variable depending upon the apoptosis inducer and cell type used, and that the mechanism of translation inhibition may be multifactorial, involving simultaneous modification of more than one translation factor.
Inhibition of eIF2␣ Cleavage by Z-VAD-fmk-To explore whether caspase activation is required for the cleavage of eIF2␣, we performed in vivo inhibition experiments using the broad spectrum cell-permeable caspase inhibitor Z-VAD-fmk. Cell cultures were preincubated with 75 M Z-VAD-fmk, followed by treatment with apoptosis inducers for 16 h. Analysis of cell lysates by immunoblot (Fig. 3A) show that eIF2␣ cleavage was completely inhibited in HeLa cells pretreated with Z-VAD-fmk regardless of the inducing agent used. The observed inhibition of cleavage was concomitant with a reduction in overall caspase activation as measured by the decrease in release of pNA from the caspase substrate Ac-DEVD-pNA (Fig.  3B). The DEVD-pNA cleavage activity in cell lysates pretreated with Z-VAD-fmk was reduced by more than 70% for each inducer tested. These results suggest that the cleavage of eIF2␣ is caspase-dependent and indicate that caspases could be di- rectly involved in the induction or catalysis of eIF2␣ cleavage.
In Vitro Cleavage of eIF2␣ by Caspases-To identify caspases that could be responsible for the observed eIF2␣ cleavage in vivo, we screened a panel of purified recombinant human caspases for their ability to cleave eIF2␣ in vitro. All purified caspases were pre-standardized for equivalent levels of enzymatic activity before incubation with protein substrates. Fig.  4A shows cleavage of eIF2␣ occurred when HeLa cell ribosomal salt wash (crude initiation factor preparation containing eIF2) was incubated with caspase 3 and 6, and to a lesser extent with caspase 8 and 10, but not with caspase 9. Comparison of these results with in vivo studies shown in Figs. 1 and 3 reveals that the caspases generate an identical eIF2␣ cleavage fragment, further supporting our hypothesis that caspases are indeed responsible for eIF2␣ cleavage. In this experiment caspases 3, 6, or 10 could indirectly induce eIF2␣ cleavage via activation of other proteases present in the HeLa ribosomal salt wash, which subsequently could use eIF2␣ as a substrate. To address this question, we incubated highly purified rabbit eIF2 with purified caspases to see if the observed cleavage in Fig. 4A could be generated directly by caspases. Purified eIF2␣ was cleaved by caspase 3 (Fig. 4B) whereas incubation with caspase 6, 8, and 10 resulted in only minor or undetectable cleavage of eIF2␣. These results might indicate that cleavage of eIF2␣ by caspases 6 and 10 in panel A could have been enhanced by activation of other caspases (e.g. caspase 3) in the lysate. Levels of cleavage obtained with caspase 3 over several experiments were significantly greater than the low, somewhat variable cleavage obtained with other caspases, implying that eIF2␣ is a better substrate for caspase 3 than caspases 6 and 10. However, it is possible that purified eIF2 might not be an appropriate substrate for caspases 6 and 10, which instead may function as a more efficient substrate when it is bound within a higher order translation complex or require other unknown cofactors for efficient cleavage of eIF2␣. Furthermore, in this initial experiment, low amounts (2 units) of caspase units were used to cleave eIF2␣, which also may explain the low level of eIF2␣ cleavage. However, further experiments, in which higher levels (17 units) of caspases were used in cleavage assays, indicated that the amount of cleavage increased greatly only with caspase 3 but not with the other caspases tested (see Fig.  7 below). Overall, these results indicate that eIF2␣ can serve as a substrate for caspases and is cleaved by the effector caspase 3 more efficiently than initiator caspases (caspases 8, 9, and 10).
Direct Cleavage of eIF2␣ by Caspase 3-Based on the results from Fig. 4 we further explored the caspase 3-induced cleavage of eIF2␣ in vitro. First, we performed a time course to analyze the kinetics of the eIF2␣ cleavage reaction. Therefore, purified caspase 3 was incubated with purified eIF2 at 37°C, aliquots were taken at the indicated time points, and analyzed by immunoblot (Fig. 5A). Significant cleavage of eIF2␣ occurred within the first hour followed by complete cleavage of eIF2␣ between 5 and 18 h of incubation. To rule out the possibility that other contaminant proteases in the purified preparations could be responsible for the eIF2␣ cleavage, purified eIF2␣ and purified caspase 3 were incubated in the presence of specific caspase 3 inhibitor Ac-DEVD-CHO (Fig. 5B). Increasing concentrations of the inhibitor coincided with a decrease in the amount of eIF2␣ cleavage product and as little as 1 M of the inhibitor was sufficient to completely block the cleavage of eIF2␣. Overall, these results indicate that caspase 3 is responsible for the observed cleavage of eIF2␣.
Characterization of the Caspase 3 Cleavage Site on eIF2␣-Next we wished to determine the cleavage site on eIF2␣ utilized by caspase 3. Therefore, we performed an in vitro cleavage assay as described for Fig. 5A, transferred the proteins to polyvinylidene difluoride, and subjected the 36-kDa fragment to amino acid sequencing. The sequencing results showed that the N-terminal sequence of the 36-kDa fragment matched 9 of 10 amino acids in the authentic N-terminal amino acid sequence of intact eIF2␣ (Fig. 6A), strongly suggesting that the caspase 3 cleavage site is located in the C-terminal portion of the eIF2␣ protein. Close examination of the amino acid se- quence in the C-terminal portion of the protein revealed the presence of two potential caspase 3 cleavage motifs similar to the preferred sequences DEVD or DXXD that could be targeted for proteolysis (Fig. 6B). Although cleavage at either site would lead to a loss in size of approximately 2 kDa consistent with our findings using in vivo (e.g. Fig. 1) or in vitro (e.g. Fig. 4) experiments, we have not yet been able to confirm which site is actually recognized by caspase 3.
Cleavage of Phosphorylated versus Non-phosphorylated eIF2␣-The biological function of eIF2 is tightly regulated by the phosphorylation of the eIF2␣ subunit. Therefore, we determined whether the phosphorylation status of eIF2␣ influences the ability of caspases to cleave eIF2␣. Purified caspases were incubated with either nonphosphorylated or phosphorylated eIF2␣ (eIF2␣(P)) for 18 h and samples were analyzed by immunoblotting (Fig. 7A). Phosphorylation of the ␣ subunit modestly affected its cleavage, most easily observed with caspase 3 but also faintly visible with caspases 6, 8, and 10 which demonstrated weak cleavage activity in this assay. The ability of caspase 3 to cleave phosphorylated eIF2␣ was confirmed by immunoblotting of samples separated on slab gels by vertical isoelectric focusing (see Fig. 8

below).
A portion of the available eIF2 present in the cell is complexed with eIF2B, the guanine nucleotide exchange factor, which is required to recycle the eIF2⅐GDP complex generated during translation initiation. We were interested in whether caspases could cleave the ␣ subunit in the context of an eIF2⅐2B complex, and if so, would this cleavage be affected by phosphorylation of the eIF2␣. As shown in Fig. 7B, only caspase 3 was able to cleave eIF2␣ in the context of an eIF2⅐2B complex whereas none of the other caspases displayed measurable cleavage activity using this substrate. In addition, the cleavage of eIF2␣ present in the eIF2⅐2B complex was not affected by its phosphorylation (also see Fig. 8C, below).
Cleavage of the eIF2⅐GDP Complex by Caspase 3-eIF2 exists in vivo in at least two alternative conformational states (17,25). When the 60 S ribosomal subunit joins the 48 S initiation complex during the last step of translation initiation, the GTP bound to the eIF2⅐GTP⅐Met-tRNA i ternary complex is hydrolyzed and eIF2 is released as a complex with GDP. In the presence of physiological Mg 2ϩ concentrations, the conformation of the eIF2⅐GDP complex is "locked" in an inactive state from which the bound GDP cannot dissociate at a physiologically relevant rate. The subsequent exchange of GTP for the bound GDP obligatorily requires the interaction of the eIF2⅐GDP complex with eIF2B (17,25). In its GTP-bound state, eIF2 is in its "active" conformation, which can subsequently form a ternary complex with Met-tRNA i . In the absence of Mg 2ϩ in vitro, eIF2 can assume the conformation from which bound GDP freely disassociates in the absence of eIF2B, indicating that Mg 2ϩ is required to freeze the conformation of the eIF2⅐GDP complex.
To determine whether the conformational state of eIF2 affects the ability of eIF2 to serve as a substrate for caspase 3, we incubated purified eIF2 with GDP for 10 min at 30°C followed by addition of magnesium to "lock" the eIF2⅐GDP in its inactive conformation. Surprisingly, when this substrate was tested, cleavage of eIF2␣ by caspase 3 was found to be severely inhibited compared with cleavage of eIF2 (Fig. 8A). This observation could not be ascribed to a possible inhibitory effect of magnesium on caspase 3 activity since we found no such effect on Ac-DEVD-pNA cleavage activity by caspase 3 under similar conditions with the same concentration of magnesium (data not shown). Also, addition of magnesium to eIF2 and eIF2/2B had no significant effect on caspase activity (Fig. 8, panels B and C). Thus, these data support the existence of the previously noted conformational shift in eIF2 caused by the GDP moiety and suggests that the cleavage site for caspase 3 on eIF2␣ is either masked or blocked in this conformation. In light of this result, the modest inhibition of eIF2␣ cleavage observed upon eIF2␣ phosphorylation in Fig. 7A is most likely due to the fact that GDP remains bound to a small fraction of purified eIF2 (26). As such, the Mg 2ϩ added to the kinase reaction would stabilize the binding of this GDP to eIF2 and inhibit its cleavage by caspases.
To determine whether the phosphorylation state of the ␣-subunit of eIF2 affects the ability of caspase 3 to cleave eIF2 containing bound GDP, purified fractions of eIF2 or eIF2⅐2B complex whose ␣-subunit was 40 -50% phosphorylated were preincubated in the presence or absence of GDP (Fig. 8, B and C). The ability of caspase 3 to cleave eIF2␣ was then examined after addition of Mg 2ϩ . Samples were separated on slab gels by vertical isoelectric focusing to separate full-length and cleaved fragments of unphosphorylated and phosphorylated eIF2␣. Immunoblot analysis indicated that the presence of eIF2-bound GDP markedly inhibited the cleavage of both phosphorylated and unphosphorylated eIF2␣ by caspase 3. In contrast, the presence of GDP only marginally inhibited the cleavage of phosphorylated or unphosphorylated eIF2␣ present in the eIF2⅐2B complex.
Effect of Caspase 3-induced Cleavage of eIF2␣ on eIF2 Activity-To initially address the effect of eIF2␣ cleavage on the function of eIF2, we determined the effect of caspase 3 on the ability of eIF2 to bind and exchange GDP. eIF2 was incubated in the presence or absence of caspase 3 for 3 h at 37°C. Samples were then incubated with [ 3 H]GDP for 10 min followed by the addition of 1 mM Mg 2ϩ . Binding of an aliquot of each sample to nitrocellulose filters indicated that cleavage of eIF2␣ by caspase 3 had no effect on the ability of eIF2 to bind GDP (data not shown). Excess unlabeled GDP was then added to the samples and the extent of guanine nucleotide exchange was measured at the times indicated in Fig. 9. SDS-PAGE and immunoblot analysis of aliquots of each sample taken at the end of the incubation verified that the eIF2␣ had been quantitatively cleaved by caspase 3 (data not shown). As previously reported, bound [ 3 H]GDP was chased from eIF2 at slow rate. In contrast, after cleavage of eIF2␣ by caspase 3 over 90% of the eIF2-bound [ 3 H]GDP was chased in 15 min. Thus, exchange of eIF2 bound GDP was no longer dependent upon eIF2B after cleavage of eIF2␣ by caspase 3.
To further address the effect of eIF2␣ cleavage on the activity of the eIF2 complex we took advantage of a well defined characteristic of the 5Ј-UTR of FMDV RNA. FMDV 5Ј-UTR initiates translation via a cap-independent mechanism mediated by a large RNA structure called the internal ribosome entry site (IRES). The 3Ј margin of this structure possesses two initiation codons in tandem which lead to initiation of two forms of the leader protease of the virus (27). The selection of the initiation codon in mRNAs containing the FMDV 5Ј-UTR has been shown to be influenced by the activity of eIF2 and the amount of ternary complex eIF2⅐Met-tRNA⅐GTP (23,28) in a mechanism that may involve direct binding of eIF2 to the viral IRES RNA element (29). In previous studies, it was shown that the addition of eIF2 to in vitro translation lysates programmed with reporter mRNAs containing the FMDV 5Ј-UTR resulted in an increase in initiation from the upstream AUG (23).
Here, we used the same system to test whether caspasecleaved eIF2 could still function in modulating initiation site selection. We used mRNA transcripts containing the CAT gene downstream of either the Xenopus globin 5Ј-UTR (globin-CAT) or the FMDV 5Ј-UTR (FMDV-CAT), which was fused at the second, downstream AUG to CAT. Untreated, intact eIF2, and caspase 3-treated eIF2 were added to in vitro translation assays programmed with these transcripts and their effects on translation were assessed by SDS-PAGE autoradiography (Fig.  10A). As previously reported (23), translation of FMDV-CAT RNA yielded the normal size CAT protein (which comigrated with CAT produced using the globin 5Ј-UTR, Fig. 10, lane h) which initiated from the downstream AUG, and a larger product (preCAT), which initiated from the upstream AUG (lane a). Addition of intact eIF2 resulted in a slight stimulation in overall translation as expected (Fig. 10B) and a significant increase in translation (34%, p Ͻ 0.006) from the upstream AUG which was evidenced by the increase in the ratio of preCAT:CAT (lane f); also in good agreement with previous studies (23). In contrast, addition of caspase 3-treated eIF2 did not stimulate translation in general or specifically increase translation from the upstream AUG (lane e). Immunoblot analysis of caspase 3-treated eIF2 showed that eIF2␣ was completely cleaved before addition to translation lysates (data not shown). This loss of eIF2 function in IRES-AUG selection was completely attributed to the addition of cleaved eIF2 rather than caspase 3 itself since caspase 3 activity (lane d) was completely abolished by the caspase inhibitor Ac-DEVD-CHO (lane c) under the conditions used in this assay. Last, addition of 2-aminopurine, an inhibitor of eIF2␣ kinases (30), which prevents the phosphorylation of eIF2, caused an expected enhancement of the activity of endogenous eIF2 and translation of pre-CAT (lane g) in a similar fashion to addition of intact eIF2 (lane f). Fig. 10B shows averaged translation levels of pre-CAT and CAT generated by the FMDV IRES as well as CAT translated using a globin 5Ј-UTR. These data reveal that addition of eIF2 modestly and reproducibly stimulated translation of CAT from both the FMDV IRES (10.8%, p Ͻ 0.02) and the globin 5Ј-UTR (10.3%, p Ͻ 0.01) when compared with buffer controls. A similar, slightly larger stimulation in translation was observed when 2-aminopurine was added as a positive control. Larger levels of translation stimulation from addition of eIF2 or 2-aminopurine were not anticipated since the translation lysate contained active endogenous eIF2. In contrast, eIF2 which had been cleaved with caspase reproducibly failed to stimulate translation of either mRNA used in this assay. Furthermore, the cleaved eIF2 did not inhibit translation levels, suggesting that this form of eIF2 does not function in a dominant negative manner in this assay. Importantly, however, both cap-dependent (globin-CAT) and cap-independent (FMDV-CAT) translation was strongly inhibited in the presence of active caspase 3 (Fig. 10, A and B), presumably via combined cleavage of eIF4GI, eIF2␣, and potentially other factors in the lysate. This demonstrates that active caspase 3 is sufficient to block translation in reticulocyte lysates. In conclusion, caspase 3 treatment renders the eIF2 complex inactive in stimulation of upstream codon selection on the FMDV IRES and blocks its normal ability to stimulate overall translation initiation in vitro.

DISCUSSION
In this report, we demonstrate that the ␣-subunit of eIF2 is a target for cleavage by caspases upon induction of apoptosis in cultured cells. The cleavage of eIF2␣ was induced in HeLa and K562 cells, but also Jurkat T cells (data not shown) by cisplatin, etoposide, TNF-␣, or MG132 (data not shown). Thus, the cleavage of eIF2␣ does occur generally in apoptotic cells, however, the possible contribution of eIF2␣ cleavage to translation inhibition during apoptosis may vary dependent upon the cell type and apoptotic stimulus utilized. For instance, eIF2␣ cleavage may play a more important role in translation inhibition or modulation in etoposide-treated cells than cisplatin-treated HeLa cells. In fact, we expect that the drastic inhibition of translation in apoptosis results from a multifactorial process involving modification of several initiation factors. To this end, we have also documented cleavage of eIF4GII and poly(A)binding protein in apoptotic cells, also with variable kinetics. 2 Further experiments with cleavage-resistant mutants of eIF4GI and eIF2␣ and other factors will need to be performed to clarify the role and impact of each cleavage event in the process of translation inhibition.
Although eIF2␣ could be cleaved in vitro by several of the caspases tested, caspase 3 appears to be the prime candidate for the observed cleavage in vivo. Compared with caspase 3, cleavage of eIF2␣ by caspase 6, 8, and 10 was inefficient. In addition, only caspase 3 cleaved eIF2␣ that was present in the eIF2⅐2B complex. Since eIF2 is a heterotrimeric complex we tested cleavage of other subunits of eIF2. Although preliminary experiments show possible cleavage of eIF2␤ (but not eIF2␥) in vitro, we do not have evidence that this occurs in vivo.
The data suggest that the eIF2 present in the eIF2⅐2B complex is the preferred target for caspase 3 cleavage in vivo. Caspase 3 efficiently cleaved the ␣-subunit of eIF2 in vitro only when eIF2 was in a conformation from which GDP could readily dissociate (i.e. eIF2 in the absence of Mg 2ϩ ; or eIF2␣ present in the eIF2⅐2B complex). In the absence of Mg 2ϩ , phosphorylation of the eIF2␣ present in eIF2 or the eIF2⅐2B complex had little or no effect on its ability to be cleaved by caspase 3. However, in the presence of physiological Mg 2ϩ concentrations, which locks the conformation of eIF2 complexes containing GDP, cleavage of eIF2␣ present in eIF2⅐GDP and eIF2(␣P)⅐GDP complexes by caspase 3 is severely, if not totally inhibited in vitro. Therefore, the eIF2⅐GDP complex generated at the last step of initiation of translation is not likely to be a physiologically relevant substrate for caspase 3 in vivo. On the other hand, the presence of Mg 2ϩ and GDP had relatively little affect on the ability of caspase 3 to cleave either unphosphorylated or phosphorylated eIF2␣ present in the eIF2⅐2B complex. Thus, while we cannot currently rule out eIF2␣ present in the eIF2⅐GTP⅐Met-tRNA i complex as a potential target for cleavage by caspase 3, our data firmly establishes the eIF2␣ present in the eIF2⅐2B complex as a target for caspase cleavage in vivo.
eIF2 is the second initiation factor we have identified to be cleaved during apoptosis and that potentially contributes to alterations in translation or inhibition of translation in apoptotic cells. eIF2 is a G-protein, which is known to be a critical factor in translation initiation. In its GTP-bound conformation, eIF2 binds Met-tRNA i to the 40 S ribosomal subunit to form a 43 S preinitiation complex and aids in the recognition of the initiation codon (1). In its GDP-bound conformation, eIF2 is inactive. Recycling of eIF2⅐GDP requires its interaction with the guanine nucleotide exchange factor, eIF2B. As discussed in the Introduction, the activity of eIF2 and eIF2B is regulated by phosphorylation of the ␣-subunit of eIF2 by several kinases (e.g. PKR and HRI) in response to certain stimuli which leads to global inhibition of translation. Our data indicate that cleavage of eIF2␣ alters the activity of the eIF2 complex. Cleavage of eIF2␣: (i) generated an eIF2 complex from which GDP can dissociate independent of eIF2B; (ii) eliminated the ability of eIF2 to stimulate translation from upstream AUG codons; and (iii) inhibited the ability of the eIF2 complex to stimulate translation in vitro. Even though our data indicate that eIF2␣ cleavage followed or was coincident with eIF4GI cleavage in cultured cells, it is possible that the alterations in eIF2 activity induced by eIF2␣ cleavage contribute significantly to the inhibition of protein synthesis at later stages in apoptotic cells. Furthermore, it is known that phosphorylation of as little as 25% of all eIF2 is sufficient to cause inhibition of protein synthesis through the ability of phosphorylated eIF2 to sequester eIF2B. Since the eIF2␣ present in the eIF2⅐2B complex has been identified as the likely target for cleavage in vivo, it is possible that cleavage of only a small portion of eIF2 suffices to alter translation.
Based on the results presented in this report, we postulate that caspase 3-induced removal of the C-terminal portion of eIF2␣ could contribute to the regulation of translation in apoptotic cells through three potential mechanisms which may combine to provide a complex phenotype. (i) The cleaved form of eIF2␣ which exchanges GDP may partly counteract PKRdriven translation inhibition after partial caspase activation occurs. (ii) Large levels of cleaved eIF2␣ in apoptotic cells might mimic the effect of eIF2␣ phosphorylation and contribute to a general block in translation. (iii) Cleavage of eIF2␣ by caspase 3 might result in an eIF2 complex similar to that described for yeast, a complex capable of inhibiting protein synthesis, and perhaps, causing a specific up or down-regulation of certain apoptosis-related genes containing ORFs in their 5Ј-UTR. Those yeast studies indicated that removal of the short, highly acidic C-terminal region of eIF2␣ had the same regulatory effect on GCN4 mRNA translation as phosphorylation of eIF2␣ (31). Phosphorylation of eIF2 in yeast stimulates the translation of GCN4 mRNA by slowing the recycling of eIF2⅐GDP by eIF2B. Reduction in the rate of eIF2 recycling stimulates initiation of translation of GCN4 mRNA at the authentic downstream AUG start codon by eliminating the inhibitory effect of upstream ORFs in the 5Ј-UTR of GCN4 mRNA (32).
Could cleavage of eIF2␣ contribute to regulation of apoptosis? Conceivably, cleavage of eIF2␣ could result in translation inhibition that potentially contributes to the induction of apoptosis by suppressing anti-apoptotic genes that normally block apoptosis. Indeed, several reports have shown that treatment of cells with the protein synthesis inhibitor cycloheximide does induce apoptosis by itself, indicating that blocking protein synthesis can lead directly to cell death (33)(34)(35). However, several other publications showed that cycloheximide prevented apoptosis, so the necessity of protein synthesis appears to be dependent on cell type and trigger of apoptosis (36 -38). While extensive phosphorylation of eIF2␣ by an eIF2␣ kinase, such as PKR, could lead to a complete global arrest of translation which induces apoptosis in many cell types, cleavage of eIF2␣ by caspase 3 would yield eIF2 from which GDP can dissociate independent of active eIF2B and potentially relieve global arrest during early stages. Alternatively, cleaved eIF2 might function similar to the yeast GCN4 system (32) to enhance the translation of mRNAs coding for pro-apoptotic proteins that contain upstream ORFs in their 5Ј-UTRs as postulated above, thus contributing to the completion of the apoptotic process.
PKR itself has been shown to induce apoptosis in several systems (39 -41) whereas mutant forms of PKR have been implicated in malignant transformation (13). Srivastava et al. (12) showed that during TNF-induced cell death levels of PKR activity were increased and concomitantly levels of phospho-rylated eIF2␣ were increased as well. Expression of a nonphosphorylated form of eIF2␣, S51A, partially protected against TNF-induced apoptosis, suggesting that PKR is required for and mediates stress-induced apoptosis through eIF2␣ phosphorylation. Other studies showed that increased PKR activity leads to transcriptional and translational up-regulation of Fas and Bax, two pro-apoptotic proteins (42). The mRNAs for both Fas and Bax have upstream ORFs in their 5Ј-UTRs which could account for translational up-regulation through a eIF2␣ phosphorylation-mediated mechanism similar to the regulation of yeast GCN4 mRNA (31,32). Finally, while this article was in review, another group (43) has also shown that eIF2␣ is cleaved in apoptosis, generating results similar to those herein. In that study it was noted that cleaved eIF2␣ was able to repress PKR-mediated suppression of translation, but no mechanistic explanation for that observation was provided (43). It is likely that the enhanced ability of cleaved eIF2 to exchange GDP reported here helps counteract this effect of PKR. Much more investigation is required to unravel these complexities and to fully discern the role of eIF2 cleavage in apoptosis.