Crystal Structure of the N-terminal Segment of Human Eukaryotic Translation Initiation Factor 2α*

Eukaryotic translation initiation factor 2α (eIF2α) is a member of the eIF2 heterotrimeric complex that binds and delivers Met-tRNA i Met to the 40 S ribosomal subunit in a GTP-dependent manner. Phosphorylation/dephosphorylation of eIF2α at Ser-51 is the major regulator of protein synthesis in eukaryotic cells. Here, we report the first structural analysis on eIF2, the three-dimensional structure of a 22-kDa N-terminal portion of human eIF2α by x-ray diffraction at 1.9 Å resolution. This structure contains two major domains. The N terminus is a β-barrel with five antiparallel β-strands in an oligonucleotide binding domain (OB domain) fold. The phosphorylation site (Ser-51) is on the loop connecting β3 and β4 in the OB domain. A helical domain follows the OB domain, and the first helix has extensive interactions, including a disulfide bridge, to fix its orientation with respect to the OB domain. The two domains meet along a negatively charged groove with highly conserved residues, indicating a likely site for protein-protein interaction.

Eukaryotic translation initiation factor 2 (eIF2) 1 is a GTPbinding protein that plays a central role in initiating translation. It binds charged initiator tRNA (Met-tRNA i Met ) in a GTPdependent manner to form the ternary complex, eIF2⅐GTP⅐Met-tRNA i Met . This ternary complex binds to the 40 S ribosomal subunit, and additional initiation factors (including eIF3 and eIF1A) join to form the 43 S preinitiation complex and assist in recognizing the start codon (reviewed in Refs. [1][2][3][4][5]. Recognition of the translational site is accompanied by eIF5mediated hydrolysis of eIF2-bound GTP, which releases an eIF2⅐GDP binary complex along with several other initiation factors. For another round of initiation, the GDP in the eIF2 binary complex must be exchanged for GTP in a reaction catalyzed by the multimeric protein factor eIF2B (previously called guanine nucleotide exchange factor) (6). The 40 S⅐mRNA⅐Met-tRNA i Met complex joins the 60 S ribosomal subunit to form an 80 S initiation complex that can enter the elongation phase of protein synthesis.
In eukaryotes, eIF2 is a heterotrimer composed of ␣ (36 kDa)-, ␤ (38 kDa)-, and ␥ (52 kDa)-subunits, which appear to remain associated throughout the initiation cycle. Cross-linking and genetic studies have suggested that both ␤and ␥-subunits are implicated in guanine nucleotide and Met-tRNA i Met binding (7)(8)(9). In addition, the ␤-subunit was shown to interact specifically with eIF5 during GTP hydrolysis and also to bind mRNA (10 -12). The ␥-subunit participates in the recognition of the start site for protein synthesis (13).
The phosphorylation/dephosphorylation of a conserved serine (Ser-51) in the ␣-subunit is the major regulator of protein synthesis in eukaryotes (reviewed in Ref. 2). Phosphorylation of Ser-51 shuts off protein synthesis (14). Three protein kinases, HRI (heme-regulated inhibitor), PKR (RNA-dependent protein kinase), and GCN2, have been identified that specifically phosphorylate Ser-51 in response to a variety of cellular stresses including viral infection, heat shock, heavy metals, and deprivation of amino acids or serum.
The phosphorylated eIF2⅐GDP complex released during protein synthesis initiation binds eIF2B with much higher affinity than does eIF2⅐GTP, and binding to eIF2B does not allow the exchange GDP for GTP that is necessary for further rounds of initiation (reviewed in . A recent paper by Nika and co-authors (18) has shown that unphosphorylated eIF2␣ also meditates nucleotide exchange. The rate of eIF2B-catalyzed nucleotide exchange increases in the absence of the ␣-subunit compared with wild type, which suggests that nucleotide exchange requires direct interaction between eIF2␣ and eIF2B and that phosphorylation strengthens this interaction. There are no structural data on any of the eIF2 monomers to provide a structural context for these important regulatory interactions.
Here we report the structure of human eIF2␣ determined by x-ray crystallography using multiple anomalous dispersion (MAD) on a selenomethionine (SeMet)-labeled baculovirus-expressed eIF2␣ at 1.9 Å resolution. The crystallization of eIF2␣ was achieved only by using limited proteolysis techniques, and the final structure comprises residues 3-182, roughly the Nterminal two-thirds of full-length human eIF2␣.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Baculovirus expressing human eIF2␣ was kindly provided by Dr. Jane-Jane Chen (MIT). Human eIF2␣ was overexpressed in Sf9 insect cells grown in Cyto-SF9 medium (Kemp Biotechnologies, Frederick, MD). A cell pellet from 1 liter of culture was lysed in 100 ml of 20 mM Tris-HCl (pH 8.0), 150 mM NaCl, 5 mM EDTA, 5 mM dithiothreitol, and 1 ml of protease inhibitor mixture set III * This work was supported by Grants CA59021 from the National Institutes of Health (to J. C.). Portions of this work were carried out at the Cornell High Energy Synchrotron Source (CHESS), which is supported by the National Science Foundation under award DMR-9311772, using the resources of the Macromolecular Diffraction Facility at CHESS (MacCHESS), which is supported by Grant RR-01646 from the National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The  1 The abbreviations used are: eIF2␣, eukaryotic translation initiation (Calbiochem). The lysate was centrifuged at 48,000 ϫ g for 1 h, and the supernatant was applied to a Q-Sepharose (Amersham Biosciences) column equilibrated with 20 mM Tris-HCl (pH 8.0) and 150 mM NaCl.
The column was washed with the same buffer and eluted with a linear gradient from 150 to 500 mM NaCl. The fractions containing eIF2␣ were combined and dialyzed against 25 mM potassium phosphate (pH 7.0), centrifuged, and applied to a hydroxylapatite (Bio-Rad, CHT type 1) column equilibrated with the same buffer. The column was washed with the same buffer, and the protein was eluted with a linear gradient from 25 to 300 mM potassium phosphate. The fractions with eIF2␣ were combined and dialyzed against 20 mM Hepes (pH 7.4) plus 100 mM NaCl. The dialyzed sample was applied to an SP-Sepharose (Amersham Biosciences) column, washed with the same buffer, and eluted with a linear gradient from 100 to 400 mM NaCl in 20 mM Hepes (pH 7.4). The selenomethionine-labeled eIF2␣ was prepared by Kemp Biotechnologies as described (19) and purified under the same conditions. Protein sample was dialyzed against 10 mM Hepes, pH 7.5, 100 mM NaCl and concentrated up to 10 mg/ml based on its calculated extinction coefficient (20).
Crystallization and Data Collection-Extensive crystallization experiments with freshly prepared eIF2␣ were performed without success. Hints of crystallization appeared 6 months after the drops were set up, which suggested partial proteolysis. Although the sample looked very pure by SDS-gel electrophoresis and appeared as a single band, traces of one or more proteases could certainly have been present. Pursuing this idea, eIF2␣ was concentrated to 30 mg/ml to speed up proteolysis and incubated at 4°C for approximately 1 month.
These aged samples crystallized readily at 22°C by using the sitting drop vapor diffusion method (21). The crystallization of selenomethionine-containing eIF2␣ was performed using a reservoir solution containing 10 -16% polyethylene glycol 4000 in 100 mM sodium acetate (crystallization conditions for the original eIF2␣ sample) plus the addition of 0.2 M zinc chloride. The drops contained equal amounts (2.0 l) of reservoir and protein solution. Rectangular crystals appear after 3 days.
In retrospect, the addition of zinc chloride, which is essential for crystallization, undoubtedly facilitates crystallization by connecting two interacting molecules through coordination with the solvent-exposed His-10.
A MAD data set was collected with flash-frozen crystals at Cornell High Energy Synchrotron Source (CHESS F2 beamline). Data were reduced and scaled using software programs MOSFLM and SCALA, respectively (22). SeMet eIF2␣ crystals diffract up to 1.9 Å resolution and belong to space group P222 1 with unit cell dimensions a ϭ 37.28 Å, b ϭ 44.20 Å, c ϭ 121.42 Å. Data collection statistics are given in Table  I. Assuming a molecular mass of ϳ21.4 kDa (corresponding to the first 182 residues) and 1 molecule in the asymmetric unit, the solvent content is 47.0% by volume, with a calculated V M value of 2.3 Å 3 /Da (23).
Structure Determination and Refinement-The crystal structure of eIF2␣ was solved with the MAD phasing procedure (24). The positions of the two selenium sites were determined using SOLVE 2 (25) and refined using SHARP (26). Density modification to 2.1 Å using SOLO-MON (27) produced an electron density map in which most of the protein residues could be identified unambiguously with the program O ( Fig. 1) (28). Native data from 30 to 1.9 Å were used for structure refinement. Successive rounds of rebuilding, interspersed with torsion angle simulated annealing (CNS (29)) and positional and individual B-factor refinement (REFMAC5 (30)) were carried out to generate a model with R factor and R free values of 19.80 and 23.26%, respectively (see Table I). The model contains 180 amino acid residues, 4 zinc ions, and 96 solvent sites treated as water oxygens. The average B-factor for the entire model is 36.6 Å 2 . According to PROCHECK (31), 92.2% of the residues lie on the most favored regions, 7.1% on additional allowed regions, and 0.6% on generously allowed regions. No residues were found in not allowed regions.
Relatively few baculovirus-expressed proteins have had sufficiently high SeMet incorporation to allow successful MAD phasing (19). It is worth mentioning that the N-terminal piece found in the crystal structure contains three methionine residues, but only SeMet-20 and Se-Met-44 positions were used to solve the structure. Although SeMet-20 and SeMet-44 are located in rigid parts of the secondary structure, SeMet-29 is part of a more mobile ␤-turn. Even after a complete refinement of the structure, there is no clear electron density that could account for the side chain of SeMet-29.

RESULTS AND DISCUSSION
The structure of an ϳ21.4-kDa N-terminal fragment of human eIF2␣ was determined using x-ray crystallography at 1.9 Å resolution. The crystal structure contains residues 3-182 with a disordered loop from residues 51 to 68 (Fig. 2). The final model, with approximate dimensions of 57 ϫ 36 ϫ 35 Å, is divided into two major domains: the OB domain and the helical domain.
OB Domain-The first 87 N-terminal residues have an oligonucleotide-binding fold (OB fold) (32), with a five-stranded antiparallel ␤-barrel arranged with a Greek key topology capped by a turn of 3 10 ␣-helix located between ␤3 and ␤4 (Fig. 2).
A ␤-hairpin connects ␤1 and ␤2; ␤2 and ␤3 are linked by a four-residue loop and ␤4 and ␤5 by a three-residue loop. The loop connecting ␤3 and ␤4 is longer by 17 residues, and it was not completely modeled due to a lack of interpretable electron density. The visible part, residues 48 -50, begins with a turn of 3 10 helix that is stabilized through a hydrogen bond between the carbonyl group of Leu-46 and nitrogen of Glu-49. Serine 51 2 Found on the Web at www.solve.lanl.gov. comes just after this 3 10 helix (Fig. 2). Residues 63 and 64 at the end of the loop are also visible with a hydrogen bond between the carbonyl group of Arg-63 and the main chain nitrogen of Arg-66. ␤1 contains a ␤-bulge at residue Val-23, and a lefthanded Gly-65 ( and have the same absolute values as the right-handed Gly but with opposite signs) is present at the beginning of ␤4; both are standard features of the OB fold.
The eIF2␣ N-terminal domain is the latest example of the large oligonucleotide/oligosaccharide-binding fold. The structural classification of the protein data base (SCOP (34)) counts at present 61 different structures containing an OB fold, divided into seven superfamilies. Using the DALI server software (35), the OB domain of human eIF2␣ was identified as a member of the nucleic acid-binding superfamily. Examples of members of this family are: S1 RNA-binding domain from the Escherichia coli polynucleotide phosphorylase (33), cold shock proteins A and B (36,37), domain II of eukaryotic translation initiation factor 5A (38,39), translational initiation factor IF1 (40), and the N-terminal domain of aspartyl-tRNA synthetase (41), among others.
Although eIF2␣ shows very little, if any, sequence homology with members of this family, their structures can easily be superimposed, and the root mean square deviation along the coordinates of all C ␣ atoms is around 2.0 Å. The main differences are restricted to loop regions, especially the length of the loop connecting ␤3 and ␤4 containing the phosphorylation site. Among the eIF2␣ sequences, the ␤3 to ␤4 residues are highly conserved (Fig. 3); in human eIF2␣, this region contains a turn of a 3 10 helix (Fig. 2) as observed in the NMR structure of E. coli polynucleotide phosphorylase (33) and the crystal structure of translation initiation factor 5A from Pyrobaculum aerophilum (38). Other family members have different structures in this connecting region. Translational initiation factor IF1 (40) and the N terminus of aspartyl-tRNA synthetase (41) contain an ␣-helix, and cold shock proteins A and B have a long loop lacking a defined secondary structure (36,37).
The site proposed as the RNA binding site is found in nearly the same position in all members of the nucleic acid-binding superfamily; this is the ␤-barrel region, where three loops connecting ␤1 and ␤2, ␤3 and ␤4, and ␤4 and ␤5 come together (Fig. 2). In most cases RNA interaction seems to involve solvent-exposed aromatic residues and positively charged residues on the surface of the protein. The only structure available for this group of proteins bound to nucleic acids is the complex of the N-terminal fragment of aspartyl-tRNA synthetase with the anti-codon loop of its cognate tRNA (41) where there is a -stacking interaction between one of the bases and a conserved Phe residue. In eIF2␣ the corresponding region contains two tyrosine residues, Tyr-32 and Tyr-81, clustered with nonpolar residues Met-44, Leu-46, Ile-82, and Gly-43 and two residues, Glu-42 and Asp-83, with negatively charged side chains (Fig. 2).
The OB domain of human eIF2␣ does not have any of the clustered positive charges that are observed for the other members of this family. This structural feature is consistent with the observation that, unlike other family members in which biochemical and genetic experiments and limited structural data support the idea of direct interaction with nucleic acids (36 -37, 40 -41), there is little or no evidence that eIF2␣ binds RNA (18). Most cross-linking and genetic experiments suggest that ␤and ␥-subunits of eIF2 are the subunits involved in initiator tRNA and ribosomal RNA binding (8,42).
Helical Domain-The helical domain comprises residues 88 -182. In contrast to the highly conserved architecture of the OB domain, the helical domain adopts a previously unreported fold consisting of a 28-residue-long ␣-helix (␣1), a series of small ␣-helices (␣2, ␣3, ␣4, and ␣6), and one 3 10 helix folded into a very compact domain (Fig. 2). The OB and helical domains are linked by a disulfide bridge between Cys-69 in ␤4 and Cys-97 in ␣1 (Fig. 2). Alignment of eIF2␣ sequences suggests that this disulfide bridge is only possible in mammalian eIF2␣ (Fig. 3). The last residue in our model, Arg-182, is solvent-exposed, and some uninterpretable electron density can be observed around this region, suggesting that proteolytic attack may not be specific.
Groove between the OB and Helical Domains-A negatively charged cavity is formed where the OB and helical domains come together (Fig. 4). The residues defining this groove come Interestingly, all of the residues found in this region are highly conserved among the species (Fig. 4). The ␤-subunit of eukaryotic translation initiation factor 2 (eIF2␤ (12)) contains a polylysine motif. The negatively charged groove in eIF2␣ might serve as the direct interaction site for the polylysine repeat.
Phosphorylation Site-eIF2␣ regulates protein synthesis through phosphorylation/dephosphorylation of Ser-51. Phosphorylation of eIF2␣ converts eIF2 from a substrate into a competitive inhibitor for the guanine exchange factor eIF2B. The eIF2 binding experiments demonstrate that eIF2B binds to phosphorylated eIF2 with a higher affinity than to unphosphorylated eIF2.
It has been suggested that eIF2␣ is required for the interaction between eIF2 and eIF2B (8). The high affinity interaction between eIF2␣ and the regulatory domain composed of ␣and ␦-subunits of eIF2B is required for the proper binding of eIF2 to the catalytic domain of eIF2B, allowing normal rates of nucleotide exchange (43). The phosphorylation site of eIF2␣ lies on the long loop connecting ␤3 and ␤4 of the OB domain, and it is close to the nominal RNA-binding site (Fig. 2). The lack of electron density suggests that this region is very mobile. Not surprisingly, the NMR structure of the translational initiation factor IF1 from E. coli shows considerable flexibility for the corresponding loop (40).
The simplest way for eIF2B to detect and respond to phosphorylation at Ser-51 would be for eIF2B to make direct contact with eIF2 in the Ser-51-containing loop region. An interesting feature of the highly conserved loop sequence is the predominance of positively charged residues (Arg-52, Arg-53, Arg-54, Arg-56, Lys-60, Arg-63) following Ser-51 (Fig. 3). This proximity and conservation, combined with the recent results that eIF2␣ and the regulatory domain eIF2B interact independently of phosphorylation, suggest a model in which the positively charged residues participate in an interaction that is enhanced by phosphorylation.
In summary, the present study provides the first structural analysis of any member of the eIF2 heterotrimeric complex and provides a framework for further genetic, biochemical, and structural analyses of protein-protein interactions involving eIF2␣.
FIG. 3. Sequence alignment of eIF2␣ proteins. Similar residues are highlighted in cyan, and the phosphorylation site is yellow. The conserved positive charges found near the phosphorylation site are highlighted in pink. The first three sequences are mammalian, and the last four are yeast. Alignment was performed using MULTALIN (45) and picture using ALSCRIPT (46).

FIG. 4. A view of the highly conserved pocket between ␣1 of the helical domain (upper right) and the OB domain (lower left).
The molecular surface was calculated with SPOCK (47) and colored according to the sequence conservation. Completely conserved residues are indicated in red, those more than 50% conserved are in orange, and those less than 50% conserved are in blue. The completely conserved groove is negatively charged.