The crystal structure of the glutathione S-transferase-like domain of elongation factor 1Bgamma from Saccharomyces cerevisiae.

The crystal structure of the N-terminal 219 residues (domain 1) of the conserved eukaryotic translation elongation factor 1Bgamma (eEF1Bgamma), encoded by the TEF3 gene in Saccharomyces cerevisiae, has been determined at 3.0 A resolution by the single wavelength anomalous dispersion technique. The structure is overall very similar to the glutathione S-transferase proteins and contains a pocket with architecture highly homologous to what is observed in glutathione S-transferase enzymes. The TEF3-encoded form of eEF1Bgamma has no obvious catalytic residue. However, the second form of eEF1Bgamma encoded by the TEF4 gene contains serine 11, which may act catalytically. Based on the x-ray structure and gel filtration studies, we suggest that the yeast eEF1 complex is organized as an [eEF1A.eEF1Balpha.eEF1Bgamma]2 complex. A 23-residue sequence in the middle of eEF1Bgamma is essential for the stable dimerization of eEF1Bgamma and the quaternary structure of the eEF1 complex.

The protein biosynthesis process is divided into initiation, elongation, and termination. During initiation Met-tRNA i Met is bound to the ribosomal P-site and base-paired with an initiator AUG codon on the mRNA. In the elongation cycle, the aminoacylated tRNA is brought to the ribosomal A-site by eukaryotic elongation factor (eEF) 1 1A (eEF1A), a 50-kDa G-protein. The ribosome acts as a GTPase activator for eEF1A in the presence of a correct codon-anticodon match between the aminoacylated tRNA and the A-site codon of mRNA. eEF1A hydrolyzes its bound GTP, and eEF1A⅐GDP leaves the ribosome (reviewed in Ref. 1). The yeast elongation factor 1 complex (eEF1) consists of eEF1A and elongation factor 1B (eEF1B), which is the guanine nucleotide exchange factor for eEF1A. In all eukaryotes eEF1B contains at least two subunits, ␣ and ␥. In metazoans a third subunit, eEF1B␤, which shares high sequence similarity to eEF1B␣, is present (2,3). eEF1B␣ and eEF1B␤ are catalytic subunits of the exchange factor, and eEF1B␣ is essential for viability in Saccharomyces cerevisiae (4). The interaction between eEF1B␥ and eEF1B␣ involves the N-terminal portion of both proteins (5). The eEF1B␥ subunit in Artemia salina enhances the activity of the eEF1B␣ subunit by 100% when added in a 1:1 molar ratio in vitro (6).
There are unique aspects of eEF1B␥ function other than association with eEF1B␣. In A. salina 5% of the eEF1B␣⅐ eEF1B␥ complex in the cell is associated with membranes, and eEF1B␥ can associate with tubulin (6). Studies in human fibroblasts indicated that the eEF1 complex is predominantly associated with the endoplasmic reticulum, possibly anchored via eEF1B␥ (7). Association between eEF1B␥ and mRNA has also been reported (8). eEF1B␥ in S. cerevisiae was identified through a screen for calcium-dependent membrane-binding proteins (9). From the protein sequence a gene was identified named CAM1 (calcium and membrane binding), and its disruption does not affect the viability of the cell. The same gene was later identified as acting as a dosage extragenic suppressor of a cold-sensitive mutant, and named drs2 (deficient for ribosomal subunits), which is deficient in the assembly of 40 S ribosomal subunits (10). The gene identified was named TEF3 and the protein it encodes Tef3p. A second isoform of eEF1B␥ encoded by the TEF4 gene in S. cerevisiae has also been identified (11). The sequence identity between Tef3p and Tef4p is only 64.5%, with the highest conservation in the C terminus. It is further believed that protein synthesis is under the control of the cell cycle during meiosis and mitosis. eEF1B␥ from Xenopus laevis and Carassius auratus was found to be a substrate both in vivo and in vitro for the cell division control M-phase promoting factor (12,13). Taken together these results may indicate additional functions for this protein. Very recently Tef3p has been identified in a complex binding to the msrA promoter suggesting a function in regulation of expression of methionine sulfoxide reductase (14).
The N-terminal domain of eEF1B␥ has sequence motifs characteristic of the Theta class glutathione transferases (GST) and, in accordance with this, was suggested to form homodimers and be enzymatically active (15). Recently, GST activity was observed toward the model substrate 1-chloro-2,4dinitrobenzene (CDNB) with the recombinant eEF1B␥ subunit from Oryza sativa expressed in Escherichia coli and for the full native eEF1B complex (16). Likewise, the silk worm (Bombyx mori) eEF1B␥ has been observed to bind to GSH-Sepharose (17).
Soluble GSTs (EC 2.5.1.18) are proteins involved in the cellular three-phase metabolism of exogenous and endogenous * 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 xenobiotics (18,19). The GSTs belong to the phase II system of proteins and catalyze the conjugation of the nucleophilic sulfhydryl group of GSH to a number of electrophilic compounds. The GSTs function by decreasing the pK a of GSH, thereby allowing its deprotonation and the formation of a more reactive thiolate anion. In the Alpha, Mu, Pi, and Sigma GST classes, this anion is maintained in the active site through an interaction with a tyrosine residue. In contrast, Theta and Theta-like GSTs utilize a conserved serine residue as the primary catalytic active site residue (20). Two GST enzymes, Gtt1p and Gtt2p, have been identified and characterized in S. cerevisiae (21). Recombinant protein expressed in E. coli exhibited GST activity toward CDNB.
Limited proteolysis demonstrates that yeast and human eEF1B␥ consist of two structural domains connected by a flexible linker (this article and see Ref. 22). We here present the first crystallographic structure of the eEF1B␥ N-terminal domain 1 (residues 1-219) from the yeast S. cerevisiae determined by the single anomalous dispersion (SAD) technique at 3.0 Å resolution. Consistent with predictions of a GSH-binding motif (15), the structure of domain 1 is very similar to those of GST enzymes. Importantly, structural details of the catalytic site are conserved in the yeast eEF1B␥ domain 1. Both domain 1 and the slightly longer domain 1Ј (residues 1-242) were not able to bind to a GSH matrix, nor did they show any activity toward the GST model substrate CDNB. A model for the eEF1 complex in S. cerevisiae is suggested based on this structure and the previously reported structure of eEF1A in complex with the catalytically active fragment of eEF1B␣ (23).

EXPERIMENTAL PROCEDURES
Expression of Full-length eEF1B␥-Plasmid pTKB532 containing the full-length S. cerevisiae TEF3 gene with an N-terminal histidine tag and inserted into a pET9d plasmid was transformed into E. coli BL21 cells. The cells were grown in LB media containing 50 g/ml kanamycin at 37°C until an A 600 value of 0.6 -0.8, and protein was expressed with 0.25 mM isopropyl-␤-D-thiogalactopyranoside for 3-4 h. The cells were harvested by centrifugation. The following purification steps were performed at 0 -6°C. The cells were resuspended in lysis buffer (250 mM KCl, 40 mM Tris-HCl, pH 7.6, 5 mM MgCl 2 , 0.5 mM EDTA-NaOH, pH 7.5, 5 mM ␤-mercaptoethanol, and 0.1 mM phenylmethylsulfonyl fluoride (PMSF)) and sonicated. The lysate was centrifuged at 50,000 rpm for 75 min, and the pH of the supernatant was adjusted to at least 7.6 and loaded on a 5-ml charged HiTrap chelating column (Amersham Biosciences) equilibrated in lysis buffer. The column was washed and protein eluted with lysis buffer containing 50 and 120 mM imidazole HCl, pH 7.8, respectively. The eluted protein was concentrated in a YM-30 Centricon (Millipore) and loaded on a 120-ml (2 ϫ 60 cm) Superdex 200 Prep Grade gel filtration column (Amersham Biosciences) equilibrated in 250 mM KCl, 20 mM Hepes-KOH, pH 7.2, 5 mM DTT, and 1 mM PMSF.
Limited Proteolysis of Full-length eEF1B␥-Full-length eEF1B␥ was mixed with trypsin (Sigma) in a 640:1 weight ratio and kept on ice for 16 h. The reaction was stopped by adding 1 mM PMSF, and the result was analyzed by SDS-PAGE. The salt concentration was lowered to ϳ100 mM by mixing with 20 mM Hepes, pH 7.2. Ion exchange chromatography on a 1-ml Mono-S column (Amersham Biosciences) led to the isolation of two distinct domains of eEF1B␥. The column was equilibrated in buffer A (100 mM KCl, 20 mM Hepes-KOH, pH 7.2, and 0.5 mM DTT) and eluted with a gradient from buffer A to buffer B (700 mM KCl, 20 mM Hepes-KOH, pH 7.2, and 0.5 mM DTT). In agreement with the predicted high pI of 9.1 for residues 1-219, the N-terminal basic fragment bound to the column, although the acidic C-terminal fragment was in the flow-through.
Subcloning-Plasmid pTKB176 containing the S. cerevisiae TEF3 gene encoding eEF1B␥ was used as a template for PCR amplification of the eEF1B␥-(1-219) fragment by using primers 5Ј-CATGCCATGGCA-CATCACCATCACCATCACTCTCAAGGTACTTTATATGCT-3Ј and 5Ј-GAAGATCTTTATTATTGAGGGGGACTCAATGG-3Ј. The primers introduce a His 6 tag and an NcoI restriction site at the AUG and a BglII restriction site 3Ј of the new stop codon. Boldface letters indicate the restriction sites, and the His 6 tag is underlined. The PCR product was digested with NcoI and BglII and cloned into plasmid pET11d digested with NcoI and BamHI producing plasmid pTKB588. Based on the observations made with the eEF1B␥-(1-219) construct, a second construct containing residues 1-242 of eEF1B␥ was made with the same 5Ј primer and the 3Ј primer 5Ј-GAAGATCTTTATTATGGCTTGGCTTC-CTCCTT-3Ј, producing plasmid pTKB611.
Expression and Purification of SeMet-substituted Protein-Plasmids were transformed into electrocompetent BL21(DE3)B834 E. coli cells. The SeMet media contained 1 mg/liter vitamins (riboflavin, niacinamide, pyridoxine monohydrochloride, and thiamine), 40 mg/liter of all amino acids except methionine, 40 mg/liter of seleno-L-methionine, 25 mg/liter FeSO 4 , 4 g/liter glucose, 2 mM MgSO 4 , 2 g/liter NH 4 Cl, 6 g/liter KH 2 PO 4 , 25.6 g/liter Na 2 HPO 4 ⅐7H 2 O, and 100 g/ml ampicillin. All amino acids, vitamins, and glucose were filtered through a 0.45-m filter (Sartorius). The rest of the ingredients were autoclaved prior to mixing (24). An overnight bacterial culture was used to inoculate SeMet media, and after 24 h of growth at 37°C, an A 600 of ϳ1.0 was achieved. Protein expression was induced with isopropyl-␤-D-thiogalactopyranoside at a final concentration of 0.5 mM for 6 h, and cells were harvested by centrifugation. All purification steps were performed at 0 -6°C. The cells were resuspended in lysis buffer (250 mM KCl, 50 mM Tris-HCl, pH 7.6, 5 mM MgCl 2 , 5 mM ␤-mercaptoethanol, and 0.1 mM PMSF) and sonicated. The rest of the protocol is identical to that of the full-length protein, except that gel filtration was omitted.
GST Activity Assay-The proteins to be used in the assay were dialyzed overnight against PBS buffer (140 mM NaCl, 2.7 KCl, 10 mM Na 2 HPO 4 , and 1.8 mM KH 2 PO 4 , pH 7.3). The absorbance was recorded at 1-min intervals for 5 min on an LKB Biochrom Ultrospec II spectrophotometer at a wavelength of 340 nm. The reaction mixture consisted of a total volume of 1 ml containing 100 mM KH 2 PO 4 , pH 6.5, or 100 mM Tris-HCl, pH 7.6, 1 mM of CDNB, and GSH and a given amount of protein. As a reference, the absorbance at 340 nm was measured using the reaction mixture and a volume of phosphate-buffered saline equal to the volume of protein sample used in the other sample mixtures. The activity of the proteins were compared with the activity of GST from Schistosoma japonicum (Sj26) encoded in a pGEX-1 vector (25).
Data Collection and Processing-Crystals were transferred to stabilization buffer (2.48 M (NH 4 ) 2 SO 4 , 2 mM reduced GSH, 3 mM DTT, 1 mM NaN 3 , 0.1 mM EDTA-NaOH, pH 7.5, 25 mM Mes-NaOH buffer, pH 6, and 25 mM Hepes-NaOH buffer, pH 7) and stored for 2 days. The crystals were then transferred to cryobuffer (stabilization buffer ϩ 20% glycerol) and frozen immediately in a stream of nitrogen gas at 100 K. A three-wavelength MAD data set was collected at the selenium K-edge on the beamline ID29 ESRF, France. The data were processed and reduced with MOSFLM and SCALA (27). The selenium sites were located with SHELXD (28), and SAD phases to 3 Å resolution were calculated with CNS (29) by using the data collected at the x-ray wavelength corresponding to the peak of the x-ray fluorescence spectrum. After density modification an initial model was constructed with O (30). An initial refinement with a model containing residues 2-210 in CNS (29) resulted in the drop of R free from 40 to 26%. Iterative cycles of manual model building and refinement with the experimental phases as restraints resulted in a structure with all 219 amino acids (Table I).
Due to the presence of selenomethionine, and the use of MAD phases as restraints, anomalous pairs were used for refinement. The quality of the structure was inspected during rebuilding with OOPS2 (31) and finally with PROCHECK (32). The secondary structure was assigned with DSSP (33). All figures were made with PyMOL (34).

Expression of eEF1B␥
Fragments-In order to study the organization of the eEF1 complex in yeast, we initially expressed full-length TEF3-encoded eEF1B␥ in E. coli. However, this protein showed undesirable properties for structural studies such as a strong tendency to aggregate at physiological salt conditions (results not shown). Based on limited trypsin digestion and sequence alignment, we constructed two deletion mutants. The smallest consisted of residues 1-219, domain 1, which has clear sequence homology to GST proteins (15). Based on sequence alignment with the human GST T2-2 sequence, we also expressed residues 1-242, domain 1Ј. Both recombinant fragments of eEF1B␥ were highly soluble (data not shown).
In gel filtration assays domain 1 eluted at a position similar to carbonic anhydrase (29 kDa) indicating a monomer, whereas domain 1Ј eluted at a position similar to bovine serum albumin (66 kDa) indicating a dimer (Fig. 1). These results suggest that residues 220 -242 are required for the stable dimerization of Tef3p eEF1B␥ in S. cerevisiae. Proteolytic studies of the fulllength S. cerevisiae Tef3p showed that the protein could be cleaved into two stable fragments of 26 and 21 kDa, respectively, as assessed by SDS-PAGE. The C-terminal domain 2, residues 237-415, elutes as a monomer from a gel filtration column (Fig. 1). Domain 2 does not interact with eEF1B␣ and can be separated from the N-terminal fragment by ion exchange (data not shown). Both domain 1 and 1Ј were able to form stable complexes with eEF1B␣ (data not shown). Although the amino acid sequence suggests homology to the GSH-binding motif of GST proteins, no such enzymatic activity was observed with domain 1 or 1Ј toward the GST marker substrate CDNB, whereas activity was demonstrated for a recombinant GST (data not shown).
The Structure of Domain 1-Although large single crystals of domain 1Ј were obtained, they diffracted only weakly to ϳ4.5 Å. Crystals of selenomethionated domain 1 diffracted better, and a three wavelength data set was collected at 3 Å resolution. The density map obtained from SAD phases was superior to that obtained from MAD phases, most likely due to radiation damage during data collection at the remote and inflection wavelengths. Initially the crystals were believed to be tetragonal, and 100 degrees of data were collected for each wavelength. Radiation damage was observed already within the peak wavelength collected first, and only the first 50 degrees of data were used, but due to the high symmetry this was sufficient to obtain a redundancy of 10.7 (Table I). The asymmetric unit contains one molecule of domain 1 and a sulfate ion and has a solvent content of 63%. The experimental map obtained after density modification was easily interpretable over almost the full length of the protein (Fig. 2). The mean residue real space correlation calculated with O between the refined model and the experimental map is 0.83. For comparison the correlation is 0.91 between the model and the final 2F o Ϫ F c map, but this map inevitably suffers to some degree from model bias, so the correlation to this map is likely to be artificially high. Visually, there is rather little difference between the two maps indicating a very high quality of the input SAD phases. These phases were used as MLHL restraints in all cycles of refinement. Although the same test sets were used in all refinement steps, there is an unusual small difference between R and R free (Table  I). However, it has been demonstrated that there is a strong correlation between R free and the phase error, and with good experimental phases the MLHL target is superior for refinement (35). Our structure fully confirms the power of using good experimental phases as restraints with respect to obtaining low values of R free .
The refined crystal structure of S. cerevisiae eEF1B␥ domain 1 contains all 219 amino acids and a sulfate ion (Fig. 3A). The  structure shows striking similarity to GST proteins, and one of the crystallographic dimers is organized as a known GST dimer (Fig. 3). Like GST enzymes the monomer consists of two subdomains. Subdomain 1 N , residues 1-74, has a central fourstranded ␤-sheet flanked on one side by the parallel helices ␣1 and ␣3 and on the other solvent-exposed side by the ␣2 helix. A linker of 15 residues (75-89), of which residues 77-83 forms a helix, connects subdomain 1 N to subdomain 1 C containing residues 90 -219. As in GST structures we named the helix in the linker ␣3B. This helix formation is not seen in the human Theta class GST T2-2 structure, the Arabidopsis thaliana GST Theta class structure, nor the Australian sheep blowfly Lucilia cuprina GST structure (20,36,37). However, a short 3 10 helix is found in the linker connecting subdomains 1 N and 1 C in the structures of maize GST 1-3 and GST 1-4 from the mosquito Anopheles dirus (38,39).
Subdomain 1 C contains five helices: ␣4, ␣5, ␣6, ␣6B,and ␣7 (Figs. 3 and 4). As in the human GST T2-2 structure, helix ␣4 is irregular and is more accurately described as three helices ␣4A, ␣4B, and ␣4C. The three helices of eEF1B␥, however, are almost coaxial. Helix ␣5 runs antiparallel with almost the full length of ␣4 and is slightly bent. Helix ␣6 contains at its start the N-capping box ((S/T)XXD) and a hydrophobic staple motif, both of which are highly conserved among GSTs and appear to greatly stabilize their fold (40). This helix runs parallel to helix ␣4 and bends away from the subdomain 1 N of the protein at its C-terminal end. Compared with the human GST T2-2 structure, there is an additional 3 10 helix between helices ␣6 and ␣7, termed ␣6B. A similar helix is seen in the A. thaliana GST Theta class structure (37). Subdomain 1 C has a hydrophobic core created by an intricate ring stacking of the side chains of Although present in the crystallization solution and in the cryobuffer, no electron density for GSH was observed in the electron density maps. A large spherical piece of electron density surrounded by Arg-11, Arg-13, and Arg-171 was observed at the N terminus of helix ␣1, which cannot be attributed to protein. Given the fact that the protein was crystallized in ammonium sulfate, the basic environment, and that the human GST T2-2 and the maleylacetoacetate isomerase/glutathione transferase , MAAI/GST Z1-1 structures have a sulfate ion in a similar position (36,41), this density was modeled as a sulfate ion (Fig. 5).  3. A, the eEF1B␥ domain 1 monomer viewed from the dimer interface side. The N-terminal subdomain 1 N containing the sulfate ion is shown on the left and subdomain 1 C on the right. The N and C termini are labeled along with the secondary structural elements and the sulfate ion. The last 20 amino acids constituting the C terminus of subdomain 1 C have been colored orange for clarity. B, the eEF1B␥ GST-like dimer viewed along the crystallographic 2-fold axis. Monomer A is shown on the left in cyan with subdomain 1 N at the bottom containing the sulfate ion and subdomain 1 C at the top, and monomer B is shown on the right in yellow. C, the human GST T2-2 monomer with 1-menaphthyl-GSH and a sulfate ion shown in ball and stick bound in the active site (36)). Helices ␣8B and ␣9, shown in orange, completely encapsulate the active site.
The eEF1B␥ Dimer-Although in solution domain 1 does not dimerize, the crystallization conditions induce dimer formation around a crystallographic 2-fold axis (Fig. 3B). The A and B monomers interact primarily by contacts between ␤4 and ␣3 from the subdomain 1 N of one subunit and ␣4 from the subdomain 1 C of the adjacent subunit. This interaction decreases the solvent-accessible surface by 1237 Å 2 per monomer. The interactions between the monomers in the dimer are both hydrophobic and polar. The side chains of Leu-60 A (␤4), Ala-65 A (␣3), Tyr-68 A (␣3), and Tyr-69 A (␣3) form the hydrophobic core of the dimerization interactions with Leu-90 B , Gln-93 B , Ala-94 B , and Ile-97 B , all from ␣4A. The methylene groups of Glu-62 A , Met-64 A (␣3), Lys-72 A (␣3), and Arg-98 B (␣4A) all contribute to the dimerization through their side chain methylene groups at the periphery of this hydrophobic core. The polar and charged residues Tyr-68, Lys-72, Asp-87, and Gln-93 and their symmet-rical equivalents from the other monomer all point toward the 2-fold symmetry axis of the dimer, creating a polar/charged pocket in the dimer interface. At the 3 Å resolution of the structure, water-mediated contacts cannot be observed. The hydrophobic "lock-and-key" motif described for mammalian GST structures was not present in eEF1B␥ Tef3p (42). This involves a hydrophobic key, usually Phe or Tyr, from the loop preceding ␤3 and a lock composed of residues from ␣4 and ␣5 of the other monomer. The hydrophobic key in this structure is represented by Leu-46, but the side chain of this residue is too short and is not near a putative lock motif. The bacterial (Proteus mirabilis) Beta and squid Sigma class GSTs both have a polar rather than a hydrophobic dimeric interface and also lack the lock and key motif (43,44). In contrast to the classical V-shaped dimer interface observed in other GST structures, as well as the two Theta-like structures, a more close-packed FIG. 4. A, amino acid sequence alignment between Tef3p eEF1B␥ domain 1 and hGST T2-2. The sequences have been aligned based on their structure. The secondary structure elements of the two proteins are indicated above and below the sequence, respectively. The eEF1B␥ sequence has been colored based on an alignment between 20 eEF1B␥ sequences made with ClustalW. Similar residues have a light gray background; residues identical in 50% or more of the sequences have a dark gray background, and strictly conserved residues have a black background. The active residue in hGST T2-2 has a black background, and three other residues important for the binding of GSH are boxed. The similarity between the two sequences is indicated below by : (similar) and * (identical). B, partial sequences of eEF1B␥ from different species with the putative eEF1B␤ recognition loop. Same coloring as in A.
interface is observed in eEF1B␥. Finally, the stacking of symmetry related arginine guanidinium groups in the dimer interface of Alpha, Mu, Pi, and Sigma GST classes (42, 44 -46) is not observed in the structure presented here. This is consistent with the observations in the L. cuprina GST structure (20). The only Arg residue near the dimeric interface, Arg-98 from helix ␣4A, points away from the 2-fold axis of the dimer and toward the solvent. With respect to the interface and the lack of stacked residues, eEF1B␥ is similar to the bacterial GST Beta class protein (43).
Domain 1Ј of eEF1B␥ has a lower level of identity (27.9%) compared with domain 2 (57.9%) when comparing the two S. cerevisiae forms with those from human, X. laevis, and A. salina (11). An alignment of the GST homology regions from 20 eEF1B␥ sequences resulted in the identification of 12 conserved residues (Fig. 4A). Two of the conserved residues, cis-Pro-50 and Glu-62 from S. cerevisiae eEF1B␥, are conserved in the GST proteins and are involved in the positioning and binding of glutathione in the active site (47). Three other identical residues, Thr-151, Arg-190, and Thr-194 from subdomain 1 C , form an accessible surface patch opposite to the dimer interface (Fig. 6). The largest region of conserved residues between the aligned eEF1B␥ sequences are found around the hydrophobic staple motif located between ␣5 and the start of ␣6 and are involved in the proper folding of this motif (40).
A consensus pattern derived for the Theta class GSTs showed that the residues unique to this class mainly cluster on the hydrophilic surface and flanking loops of helix ␣2 (47). This region contains the largest difference between GST classes, so it possibly plays a class-specific role in the reaction mechanism. The largest discrepancies between the eEF1B␥ sequences are seen in the region around the C terminus of helices ␣3 and ␣3B in the linker region between subdomains 1 N and 1 C . A short sequence between these two domains is common in mammalian GSTs. Another notable fact is the lack of sequence conservation at helix ␣4B, which in our structure forms a 3 10 helix. Surprisingly, there are no strictly conserved residues at the eEF1B␥ dimer interface. This feature has also been suggested to be a Theta class GST characteristic (48). The dimerization region contains only partly conserved residues, such as Ala-94 from ␣4A and Ala-65 from ␣3.
The Putative Active Site-The backbone density for the region homologous to the active site of GSTs is continuous in the initial SAD electron density map, but some of the side chain densities are weak. The side chain of a potential catalytic residue, Tyr-7, is oriented away from the active site, and no rotamer could bring it near the putative active site. Residues Arg-11 and Arg-13 have weak side chain density, which suggest high mobility. The side chain of Arg-13 is within hydrogen bonding distance of the backbone carbonyl group between Val-49 and cis-Pro-50 (Fig. 5). This carbonyl group is coordinating the amino group of the sulfhydryl moiety of active sitebound GSH in GSTs. The backbone density for Val-49 and Pro-50 is good, and only a cis-Pro could be modeled at this position in agreement with a cis-Pro in all other GST structures. The side chain of Lys-48, conserved as Lys-53 in the hGST T2-2 structure, has weak side chain density, indicating high mobility. In the presence of a ligand molecule such as GSH, it might recognize the side chain carbonyl group of the glutamyl moiety, as in the hGST T2-2 structure, or the glycinecarboxylate group of GSH. The density for Glu-62 was also of good quality and could easily be modeled. Residue Glu-62 lies in a generously allowed region of the Ramachandran plot. This residue is involved in the binding of GSH in GST proteins and lies in a similar position in the Ramachandran plots of all GSTs. Asp-106 from monomer B appears to be able to coordinate GSH in the active site of monomer A. This is as far as we know in accordance with all other GST structures except human Omega class, squid Sigma class, and the two Theta-like GSTs from A. thaliana and L. cuprina (20,37,44,49).
In the hGST T2-2 structure a sulfate ion is bound in a tetrahedral fashion to Gln-12, Arg-107, Trp-115, and Arg-239 and a water molecule. In our structure it is coordinated to the backbone amide of the Ile-12 peptide bond and the guanidinium group of Arg-171 (Fig. 5). The positive dipole moment of helix ␣1 also contributes to the affinity for the ion at this position. Due to the low resolution of the structure, no water molecules could be modeled, but water molecules are likely to be involved in the coordination of this ion.
The hydrophobic binding pocket, or H-site, is the binding site for the secondary substrate in GSTs, i.e. the cellular toxic compound. Due to the coaxial nature of helices ␣4A, ␣4B, and ␣4C in our structure, the residues corresponding to helix ␣4C in the hGST T2-2 structure that are responsible for forming part of the H-site are not near a putative H-site in the isolated domain 1, and they are not conserved. The loop between ␤2 and ␣2 responsible for forming the other part of the H-site in hGST T2-2 is too short and positioned too far from a possible H-site in eEF1B␥ domain 1. Crystal contacts around the loop region preceding helix ␣2 may be responsible for dislocating this region.

DISCUSSION
The Quaternary Structure of eEF1-The characterization of an eEF1 complex in S. cerevisiae was first performed by Saha and Chakraburtty (2). They determined the ratio of eEF1A⅐ eEF1B␥⅐eEF1B␣ subunits to be 2:1:1, and estimated the molecular mass of the complex to be ϳ200 kDa. Based on the crystal structure and the gel filtration studies presented here together with the structure of the eEF1A⅐eEF1B␣ complex (23), a model for eEF1 in yeast can be proposed (Fig. 7). This results in a 2:2:2 stoichiometric composition of the three yeast components eEF1A, eEF1B␣, and eEF1B␥ for the eEF1 complex, yielding a theoretical molecular mass of 240 kDa in agreement with the organization of the eEF1B␤-deficient eEF1 complex II from Artemia (50). In the hGST T2-2 dimer, the two C-terminal helices are oriented in an anti-parallel fashion resulting in the two C-terminal residues from each monomer being almost in hydrogen bonding distance. If eEF1B␥ in its full-length form adopts a similar secondary structure in this region, the stabilization of an eEF1B␥ dimer could be due to a coiled-coil structure in this region possibly formed by some of the residues 220 -242. The size of the N-terminal fragment originally isolated after limited trypsinolysis, and which formed a dimer in a gel filtration assay, seem to have an intermediate size between recombinant domain 1 and domain 1Ј when analyzed by SDS-PAGE. It is therefore likely that not all of residues 220 -242 are required for the stable dimerization.
Several other models for eEF1 have been proposed (50 -54). The one that is most consistent with the model presented here is the A. salina model in which [eEF1A⅐eEF1B␣⅐ eEF1B␥⅐eEF1B␤⅐eEF1A] 2 , with a molecular mass of 408 kDa, constitutes the eEF1 complex (50). To elaborate on this model, we can say that the GST-like domain 1Ј of eEF1B␥ mediates the dimerization of the complex and that the C-terminal part of eEF1B␣ interacts with the N terminus of eEF1A (23). eEF1B␤ has been suggested to mediate the dimerization in two mammalian models (51,54). The only other model in which eEF1B␥ is explicitly suggested to dimerize is in the rabbit eEF1 model (53), in which eEF1B␣ also is suggested to dimerize. The Xenopus model, in which eEF1B␥ is suggested to form a trimer that dimerizes (52), seems less likely based on the structure presented here.
The conserved patch on the solvent-exposed side of eEF1B␥ may be part of an interaction area with eEF1B␣ ( Figs. 6 and 7). An interaction site for eEF1B␣ close to the putative active site of eEF1B␥ (see below) may help facilitate communication between the exchange activity and the putative GST-like activity of eEF1B␥. However, until the proper substrate for the latter has been identified, in vitro experiments demonstrating such a linkage are not feasible. Based on sequence alignment, it appears that eEF1B␥ from many species have a longer loop region between ␤2 and ␣2 (Fig. 4B), and this appears to correlate with the presence of eEF1B␤. Hence, this loop could be important for the interaction between eEF1B␥ and eEF1B␤ in metazoans. This loop is well separated from the conserved surface patch, which might be involved in interaction with eEF1B␣. Alternatively, it could form part of the supposedly active site in eEF1B␥, possibly adding flexibility to the region and allowing induced fit upon substrate binding and catalysis. An induced fit mechanism has been suggested for the maize GST I enzyme (38), which was suggested to belong to the Theta class (55). This solvent-exposed loop region is involved in the formation of the H-site in GST proteins, so this region may also define specificity for the putative substrate. The consensus sequence derived from sequences containing the long loop is 38 FXXGX(T/ S)N(K/R)(T/S) 46 (Fig. 4B). The boldface letters indicate conserved residues; X indicates any residue, and the numbers correspond to the human sequence. The animal sequences seem to be slightly longer in this region as compared with plant sequences, but these seem to have slightly longer sequence around ␣3B. One exception is eEF1B␥ from the protozoans Trypanosoma cruzi and Leishmania infantum, which have a Cys residue instead of the conserved Phe residue at position 38.
Is eEF1B␥ Catalytically Active?-The overall structure of the Tef3p dimer and especially the conserved cis-Pro-50 and Glu-62 all suggest that eEF1B␥ in yeast is catalytically active as a GST protein as demonstrated previously for recombinant rice eEF1B␥ (16), despite our own failure to demonstrate this with the fragments of yeast eEF1B␥. The evolutionary maintenance of such a specific structure for no reason seems very unlikely. Because the active site of eEF1B␥ Tef3p has a high degree of FIG. 6. Surface representation of conservation as described in Fig. 4 mapped on the eEF1B␥ domain 1 monomer. Residues less than 50% identical are colored red, and blue indicates 100% identity. Residues with between 50 and 100% identity are colored gray. A, eEF1B␥ viewed from the dimer interface in the same orientation as Fig. 3A. B, view from the solvent-exposed face. The three 100% conserved residues Thr-151, Arg-190, and Thr-194 form a patch on the solvent-exposed side of the N terminus of eEF1B␥. homology to other GST active sites, and although no catalytic residue could be identified, a novel GST activity mechanism with a very specific secondary substrate should not be ruled out. Arginine is a conserved active site residue in class Alpha GSTs except in a chicken liver GST. The structural equivalent in eEF1B␥ Tef3p is Arg-13, which is also conserved in the Tef4p form of yeast eEF1B␥. The conformation of Arg-13 in eEF1B␥ Tef3p could explain why neither of the two Tef3p constructs were able to bind to a GSH affinity matrix because it overlaps with the putative GSH-binding site. Mutational studies of the equivalent Arg-15 from the human Alpha class GST, hGST A1-1, showed that alteration of this amino acid reduced the catalytic activity of the enzyme (56). A conformation of eEF1B␥ Arg-13 similar to the one in hGST A1-1 could not only facilitate the binding of GSH in the active site but could also at the same time bring the charged ␦-guanido group of Arg-13 closer to the sulfate ion, thereby stabilizing it further. The involvement of an Arg residue in the active site no longer seems to be ␣-classspecific. The squid Sigma class GST also has an Arg residue at a similar position (44). Other residues conserved between the hGST A1-1 and Tef3p are Arg-13 and Arg-11, respectively. Their function seems far from similar, however. Whereas the A1-1 Arg-13 points away from the active site, Arg-11 from eEF1B␥ Tef3p points into the putative active site. In Tef4p this residue is a Ser, which may be involved in the catalytic activity of Tef4p eEF1B␥ based on alignment with the catalytic Ser residue in human GST T2-2. The human Pi class GST hGST P1-1 also has two Arg residues, Arg-11 and Arg-13 (46). The side chain of Arg-11 points toward the glutamyl-carboxylate moiety of GSH and its position thereby differs from both the equivalent positions in hGST A1-1 and Tef3p. Chemical modification studies of Arg-13 have confirmed its involvement in binding of GSH in the active site (57).
Most of the eEF1B␥ proteins have a tyrosine or threonine residue in or near the putative active site in eEF1B␥. A Tyr at position 9 in the Tef3p can position its hydroxyl group in a location very similar to that of the Ser-11 hydroxyl group of the hGST T2-2 structure, suggesting that this residue functions as the active site in the majority of the eEF1B␥s. However, yeast Tef4p, Candida albicans, and Schizosaccharomyces pombe eEF1B␥ have a Ser residue in this position. Interestingly, rice (O. sativa) eEF1B␥, which showed GST activity toward the GST model substrate CDNB, does not seem to possess a potential catalytic residue near the putative active site (16).
The existence of additional genes in S. cerevisiae coding for proteins with GST activity has been excluded previously (21). Tef3p and Tef4p, and possibly the related open reading frame YGR201C, similar to the N terminus of eEF1B␥ (58), may in this context be thought of as GST-like proteins that may catalyze highly specific reactions, perhaps also at specific locations in the cell. The TEF3 and TEF4 genes are not essential for growth, but in contrast to extra copies of the TEF3 gene, extra copies of the TEF4 gene were not able to suppress the coldsensitive growth of the drs2 ribosome assembly mutant (11). These observations indicate that Tef3p and Tef4p have different functions in the cell.
The crystal contacts in the region around the end of ␤2 and the loop preceding helix ␣2 may be responsible for the absence of secondary structure in the C-terminal 20 residues and perhaps a dislocated ␣2 helix. These two regions are highly involved in the formation of the H-and G-sites, respectively, in GST structures. If residues 200 -242 in the full-length eEF1B␥ adopt a conformation similar to that of hGST T2-2, Phe-210 in Tef3p might form part of the hydrophobic binding site. In the long random coil of the protein constituting the C terminus of our structure, there is a proline-rich area with Pro-214, -217, and -218. This region may be functionally homologous to the Pro-rich linker region Pro-226, -228, and -230, which spans the top of the active site cavity connecting helix ␣8B and ␣9 in the human GST T2-2 Theta class protein.
The GST-like domain of eEF1B␥ is of interest as related to the in vivo effects of the loss of eEF1B␥ in yeast. Deletion of either TEF3 or TEF4 results in increased resistance to oxidative stress. 2 This effect is additive at least partially through the catalytic activity of the eEF1B complex. Thus, this domain may play a novel role in regard to the activity of the eEF1B complex and the response to oxidative stress.