Direct Binding of Translation Initiation Factor eIF2γ-G Domain to Its GTPase-activating and GDP-GTP Exchange Factors eIF5 and eIF2Bϵ*

The GTP-binding eukaryotic translation initiation factor eIF2 delivers initiator methionyl-tRNA to the 40 S ribosomal subunit. The factor eIF5 stimulates hydrolysis of GTP by eIF2 upon AUG codon recognition, whereas the factor eIF2B promotes guanine nucleotide exchange on eIF2 to recycle the factor for additional rounds of translation initiation. The GTP-binding (G) domain resides in the γ subunit of the heterotrimeric eIF2; however, only eIF2β, and not eIF2γ, has been reported to directly bind to eIF5 or eIF2B. Using proteins expressed in yeast or recombinant systems we show that full-length yeast eIF2γ, as well as its isolated G domain, binds directly to eIF5 and the ϵ subunit of eIF2B, and we map the interaction sites to the catalytically important regions of these factors. Consistently, an internal deletion of residues 50-100 of yeast eIF5 impairs the interaction with recombinant eIF2γ-G domain and abolishes the ability of eIF5 to stimulate eIF2 GTPase activity in translation initiation complexes in vitro. Thus, rather than allosterically regulating eIF2γ-G domain function via eIF2β, our data support a model in which the GTPase-activating factor eIF5 and the guanine-nucleotide exchange factor eIF2B modulate eIF2 function through direct interactions with the eIF2γ-G domain.

The GTP-binding eukaryotic translation initiation factor eIF2 delivers initiator methionyl-tRNA to the 40 S ribosomal subunit. The factor eIF5 stimulates hydrolysis of GTP by eIF2 upon AUG codon recognition, whereas the factor eIF2B promotes guanine nucleotide exchange on eIF2 to recycle the factor for additional rounds of translation initiation. The GTP-binding (G) domain resides in the ␥ subunit of the heterotrimeric eIF2; however, only eIF2␤, and not eIF2␥, has been reported to directly bind to eIF5 or eIF2B. Using proteins expressed in yeast or recombinant systems we show that full-length yeast eIF2␥, as well as its isolated G domain, binds directly to eIF5 and the ⑀ subunit of eIF2B, and we map the interaction sites to the catalytically important regions of these factors. Consistently, an internal deletion of residues 50 -100 of yeast eIF5 impairs the interaction with recombinant eIF2␥-G domain and abolishes the ability of eIF5 to stimulate eIF2 GTPase activity in translation initiation complexes in vitro. Thus, rather than allosterically regulating eIF2␥-G domain function via eIF2␤, our data support a model in which the GTPase-activating factor eIF5 and the guanine-nucleotide exchange factor eIF2B modulate eIF2 function through direct interactions with the eIF2␥-G domain.
The initiation of protein synthesis in eukaryotic cells requires the coordinated activity of at least 10 eukaryotic initiation factors (eIFs), 2 with several of the factors composed of multiple polypeptide chains. Interactions among the initiation factors and the ribosome promote initiator methionyl-tRNA (Met-tRNA i Met ) and mRNA binding to the 40 S ribosomal subunit and subsequent joining of the large 60 S ribosomal subunit to form the translationally competent 80 S ribosome. The factor eIF2 is responsible for delivering Met-tRNA i Met to the 40 S subunit in the first step of translation initiation (reviewed in Refs. 1 and 2). The eIF2 is composed of three polypeptide chains: the core eIF2␥ subunit (GCD11 in yeast), to which the eIF2␣ (SUI2) and eIF2␤ (SUI3) subunits bind. The eIF2␥ subunit contains a consensus GTP-binding (G) domain at its N terminus, and structural studies on the eIF2␥ homolog from archaea revealed a three-domain structure that closely resembles the structure of the translation elongation factor EF-Tu (EF1A) (3,4). The structural similarity between eIF2␥ and EF-Tu is consistent with the common function of the two proteins to bind aminoacyl-tRNA in a GTP-dependent manner and to deliver the aminoacyl-tRNA to the ribosome. Mutational analyses of eIF2␥ support a similar mode of aminoacyl-tRNA binding by eIF2␥ and EF-Tu, wherein the 3Ј (aminoacyl) end of the tRNA binds to domain II (3,5,6). Interestingly, the C-terminal end of eIF2␣ binds to an adjacent conserved surface on domain II (3,7). The N terminus of the eIF2␤ subunit contains three lysine-rich segments (K-boxes) that are absent from the corresponding archaeal protein and do not appear to be important for binding to eIF2␥ (7,8). Moreover, the binding site for eIF2␤ on eIF2␥ has not been resolved.
Following binding of the eIF2 ternary complex (eIF2⅐GTP⅐Met-tRNA i Met ) to the 40 S subunit and association of other initiation factors, including eIF1, eIF1A, eIF3, and eIF5 (TIF5 in yeast), the so-called 43 S preinitiation complex binds to an mRNA near the 5Ј cap structure in a reaction facilitated by the eIF4 family of factors (reviewed in Ref. 1). The 43 S complex then scans the mRNA in a 5Ј to 3Ј direction in search of an AUG start codon. Base pairing between the 5Ј-CAU-3Ј anticodon of the Met-tRNA i Met in the 43 S complex with an AUG codon is thought to halt the scanning ribosome and trigger GTP hydrolysis by eIF2. This GTP hydrolysis reaction is stimulated by the putative GTPase-activating protein (GAP) eIF5 (9,10). Following GTP hydrolysis, eIF2 and other initiation factors are released from the 40 S ribosome (11,12), which then joins with the 60 S subunit in a reaction catalyzed by a second GTPase eIF5B (13). The eIF2 is released from the ribosome bound to GDP, and, like a number of G proteins, eIF2 has a higher affinity for GDP than for GTP (ϳ100-fold difference for yeast eIF2). The nonenzymatic exchange of GTP for GDP on eIF2 occurs slowly, and the guanine nucleotide exchange factor (GEF) eIF2B enhances the rate of this reaction and recycles eIF2 for use in subsequent rounds of translation initiation (2). Phosphorylation of the eIF2␣ subunit on Ser 51 by one of the stress-responsive eIF2␣ kinases (GCN2, PERK, PKR, or HRI) converts eIF2 from a substrate to an inhibitor of eIF2B and thereby blocks cellular protein synthesis (reviewed in Ref. 14).
The eIF2B is composed of five subunits that can be both physically and functionally separated into two subcomplexes (reviewed in Ref. 2). The regulatory subcomplex, composed of the ␣ (GCN3 in yeast), ␤ (GCD7), and ␦ (GCD2) subunits of eIF2B, binds to eIF2␣ in a manner stimulated by Ser 51 phosphorylation (15). The catalytic subcomplex, composed of the ␥ (GCD1) and ⑀ (GCD6) subunits of eIF2B, binds to eIF2 and catalyzes guanine nucleotide exchange in vitro (16). Further in vitro studies mapped the eIF2B GEF activity to the C-terminal ϳ200 residues of eIF2B⑀ (17), which was found to fold into an eight-stranded ␣-helical bundle resembling a HEAT repeat (18). Interestingly, the eIF2B⑀ HEAT domain contains two regions of conserved aromatic and acidic amino acids referred to as AA-boxes (or the W2 domain) that are also found in eIF5 (see Fig. 1) (19,20). These AA-boxes in both eIF2B⑀ and eIF5 have been shown to directly interact with the K-boxes in the N-terminal half of eIF2␤ ( Fig. 1) (20 -22). Multiple alanine substitutions in the AA-boxes of eIF2B⑀ and eIF5 disrupt binding to eIF2␤ (20). Consistently, these AA-box mutations in eIF5 and the corresponding K-box mutations in eIF2␤ impair initiation complex assembly (23)(24)(25).
GST or various GST-eIF5 fusions bound to glutathione-Sepharose beads were suspended in 500 l of binding buffer (20 mM HEPES, pH eIF5 and eIF2B⑀ Bind eIF2␥-G Domain 7.4, 150 mM KCl, 25 mM MgCl 2 , 5 mM NaF, 1 mM EDTA, 2 mM ␤-ME, 1% Triton X-100, 1% skimmed milk, and complete protease inhibitor mixture), incubated at 4°C for 1 h on a nutator to block nonspecific binding sites, and then mixed with the full TNT lysates containing 35 Slabeled full-length eIF2␥ or eIF2␥-G domain and incubated for 3 h at 4°C. Alternatively, the binding reactions contained 35 S-labeled FLAGtagged eIF2␥-G domain that was purified from rabbit reticulocyte lysates using anti-FLAG M2 agarose (Sigma) followed by elution with FLAG-peptide (100 g/ml) according to the manufacturer's instructions. Following binding, the beads were washed four times with 1 ml of binding buffer without skimmed milk, and bound proteins were eluted in 2ϫ SDS loading buffer, boiled for 5 min, and separated by 4 -20% SDS-PAGE. Gels were stained with GelCode (Pierce) followed by autoradiography.
GTPase Assays-Translation initiation factors eIF1, eIF1A, and eIF2, Met-tRNA i Met and 40 S ribosomal subunits were purified as described previously (29). GST-eIF5 fusion proteins were purified as described above with the following additional steps. The glutathione-Sepharose beads were washed with 5 bed volumes of cleavage buffer (50 mM Tris-Cl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1 mM dithiothreitol) and then suspended in 1 bed volume of cleavage buffer. Following the addition of 20 units of PreScission protease (Amersham Biosciences), the mixture was incubated overnight at 4°C on a nutator. The cleaved GST, uncleaved GST-eIF5, and protease were pelleted by centrifugation at 1000 ϫ g for 5 min at 4°C, and the liberated eIF5 in the supernatant was dialyzed overnight at 4°C in storage buffer (20 mM Tris-Cl, pH 7.4, 150 mM NaCl, 1 mM dithiothreitol, and 10% glycerol). The model AUGcontaining mRNA (30) was obtained from Dharmacon, Inc. (Lafayette, CO). The GTPase reactions and the calculations of the rate of GTP hydrolysis with the different versions of eIF5 were performed as described by Algire et al. (30), except eIF2 ternary complexes were formed in the presence of [␥-33 P]GTP (10 mCi/ml, 3000 Ci/mmol), and 33 P i release was monitored by thin layer chromatography on polyethyleneimine-cellulose in 0.8 M lithium acetate and quantified using a PhosphorImager.

RESULTS
Direct Interaction of eIF5 and eIF2B⑀ with eIF2␥-To study the interaction of eIF2␥ with other translation initiation factors, we generated a fusion construct to overexpress in yeast cells GST fused to full-length (FL) eIF2␥. Expression of the GST fusion protein was under the control of a galactose-inducible promoter. Following induction of GST or GST-eIF2␥-FL expression, WCEs were prepared, and the GST or GST-eIF2␥ fusion protein was immobilized on glutathione beads. Products of the pull-down reactions were separated by SDS-PAGE and analyzed by immunoblot analyses using antibodies specific for various translation factors. As shown in Fig. 2 (top), both GST and GST-eIF2␥ were expressed and pelleted with the glutathione beads. The FL GST-eIF2␥ fusion protein readily interacted with the ␣ and ␤ subunits of eIF2 (Fig.  2, lane 11), consistent with the idea that eIF2␥ is the keystone of the eIF2 complex (3). In addition, GST-eIF2␥ (FL) interacted with eIF5 and eIF2B⑀ (Fig. 2, lane 11) but not with the factor eIF3 nor with the small ribosomal subunit (RPS2p is a constituent of the yeast 40 S ribosomal subunit).
Previous studies revealed that three lysine-rich segments (K-boxes) located in the N-terminal half of eIF2␤ mediate a direct interaction with a conserved sequence element consisting of two segments rich in acidic and aromatic amino acids (AA-boxes) and found near the C terminus of eIF5 and eIF2B⑀ (20 -22). Thus, the ability of eIF2␥-FL to pull down eIF5 and eIF2B⑀ could reflect an interaction bridged by eIF2␤ rather than a direct interaction of eIF2␥ with these two effectors of eIF2 function. It was previously shown that substitution of alanine for all seven lysine residues in the first and second K-boxes of eIF2␤ (the K12 mutation) impaired the binding of eIF2␤ to both eIF5 and eIF2B⑀ (20). In order to test if eIF5 and eIF2B⑀ bind directly to eIF2␥ or whether this interaction is bridged by eIF2␤, we examined the binding of eIF5 and eIF2B⑀ to eIF2␥ in strains expressing the eIF2␤-K12 mutant. However, prior to this experiment, we first tested whether the eIF2␤-K12 mutation impaired the binding of eIF5 and eIF2B⑀ to eIF2␤ under the conditions of our pull-down assays. As shown in Fig. 3A (lane 2), eIF2␥, eIF5, and eIF2B⑀ were readily co-precipitated with FLAG-tagged wild-type eIF2␤ in FLAG pull-down assays. In contrast, only eIF2␥ was co-precipitated with the FLAG-tagged eIF2␤-K12 mutant (Fig. 3A, lane 4). Thus, the eIF2␤-K12 mutation did not affect eIF2 (␣ ϩ ␤ ϩ ␥) complex formation, but it did impair the interaction between eIF2␤ and both eIF5 and eIF2B⑀ under the conditions of our experiments.
Next, we examined, in the same whole cell extracts, the impact of the eIF2␤-K12 mutation on the ability of the GST-eIF2␥-FL fusion protein to pull down other factors. As expected, the GST-eIF2␥-FL fusion protein readily interacted with eIF2␣ and eIF2␤-K12 (Fig. 3B, lane 12). Interestingly, GST-eIF2␥-FL also bound to eIF5 and eIF2B⑀ in the eIF2␤-K12 mutant strain (Fig. 3B, lane 12). Thus, disruption of the K-boxes in eIF2␤ did not impair the interaction between FL eIF2␥ and eIF5 or eIF2B⑀, consistent with the idea that this interaction is direct and not bridged by eIF2␤. Providing further support for this idea, and as will be discussed below, the interaction between eIF2␥ and eIF5 was likewise not impaired by mutations in the AA-boxes located near the C terminus of eIF5 that disrupt the interaction between eIF5 and eIF2␤.
Despite the inability of isolated eIF2␥ domains to bind eIF2␣ or eIF2␤, the isolated eIF2␥-G domain (GST-G) as well as the GST-G ϩ D2 fusion protein were able to pull down both eIF5 and eIF2B⑀ (Fig.  2, lanes 3 and 9). Consistent with the inability of the GST-G and GST-G ϩ D2 fusion proteins to interact with eIF2␤, the binding of eIF5 and eIF2B⑀ to the GST-G and GST-G ϩ D2 fusions was unaffected by the eIF2␤-K12 mutation (Fig. 3B, lanes 4 and 10). These results support the idea that the eIF2␥-G domain directly binds to both eIF5 and eIF2B⑀, consistent with their functions to promote GTP hydrolysis and GTP-GDP exchange on eIF2. It is noteworthy that the various GST-eIF2␥ fusion proteins did not pull down the 40 S ribosomal subunit (RPS2p) ( Fig. 2A, odd-numbered lanes), indicating that the interactions detected in the pull-down assays are direct and not mediated by mutual binding of the translation factors to the same 40 S subunit. Quantification of the results revealed that GST-eIF2␥-FL and GST-G pulled down 1-9% of the eIF5 in the WCEs. This low percentage binding is consistent with the incorporation of eIF5 into the multifactor complex with eIF2 and eIF3 and with the lack of phenotype associated with overexpression of the various GST-eIF2␥ fusions in yeast cells (data not shown). Finally, as shown in Fig. 2 (top panel), the GST-G fusion was expressed at higher levels than the GST-eIF2␥-FL fusion protein (compare lanes 3 and 4  with lanes 11 and 12). Since the GST-eIF2␥-FL fusion is predicted to interact with eIF5 both directly through the G domain and indirectly via eIF2␤, the equivalent eIF5 pull-down by GST-eIF2␥-FL and GST-G may be due to the higher expression of the GST-G fusion protein.
The N-terminal Domain of eIF5 Specifically Binds to the eIF2␥-G Domain-In addition to promoting GTP hydrolysis by eIF2 in ribosomal complexes bound to an AUG codon, eIF5 plays an important role earlier in the translation initiation pathway by enhancing the binding of the eIF2 ternary complex to the 40 S ribosomal subunit (23)(24)(25). This latter function of eIF5 is mediated by the AA-boxes at the C terminus of eIF5, which enable eIF5 to bridge an interaction between eIF2 (via the K-boxes in eIF2␤) and eIF3 (23). We confirmed the importance of the eIF5 AA-boxes by examining the ability of wild-type and mutant forms of eIF5 to bind to eIF2 and eIF3 under the conditions used for our GST pull-down experiments. Immunoprecipitation of FLAG-tagged eIF5 from crude yeast extracts co-precipitated eIF3 and eIF2. In contrast, substitution of alanine for seven conserved residues in the second AAbox of eIF5 (eIF5-7A mutant) abolished the interaction with eIF3a and eIF2␤ (data not shown).
Despite disrupting the eIF5-eIF2␤ interaction, the eIF5-7A mutation did not impair the binding of eIF5 to eIF2␥. The GST-eIF2␥ fusion proteins were expressed in the isogenic strains expressing FLAG-tagged eIF5 or eIF5-7A, and pull-down assays were used to assess eIF2␥ binding to other translation factors. As shown in Fig. 4A, the GST-FL eIF2␥ fusion readily bound eIF5, eIF2B⑀, and eIF2␣ (lane 12), whereas the GST-G and GST-G ϩ D2 fusions bound both eIF5 and eIF2B⑀ (lanes 4 and 10). Importantly, all of these interactions were not impaired by the eIF5-7A mutation (Fig. 4B, lanes 4, 10, and 12). Interestingly, the GST-G fusion pulled down a greater percentage of eIF5-7A than of eIF5 (lane 4 in Fig. 4, A and B). To account for this enhanced interaction, we propose that the eIF5-7A mutation, which weakens the binding to eIF3 and eIF2␤, may liberate eIF5 from the multifactor complex. Accordingly, this free eIF5-7A is then available to readily interact with the GST-G fusion protein, resulting in the increased binding in the pull-down assay. We conclude that the AA-boxes in eIF5 that mediate a direct interaction with eIF2␤ are not critical for the binding of eIF5 to eIF2␥. Thus, these results provide further support for the idea that the eIF2␥-G domain directly binds to both eIF5 and eIF2B⑀.
Since the GST pull-down assays employed in these studies rely on the co-expression of the GST fusion and the interacting protein in the same yeast cell, it is possible that other yeast proteins may bridge the eIF2␥-eIF5 interaction. To test whether eIF5 can bind to eIF2␥ in vitro, we expressed in bacteria and purified GST-eIF5 fusion proteins consisting of eIF5 residues 1-405 (FL), residues 1-279 (NTD), or residues 280 -FIGURE 3. The interaction of eIF2␥ with eIF5 and eIF2B⑀ is independent of the K-boxes in eIF2␤. As described in Fig. 2, GST or various GST-eIF2␥ fusion proteins were overexpressed in yeast strains expressing FLAG-tagged eIF2␤ (H2888) or FLAG-tagged eIF2␤-K12, in which the seven lysine residues in the first two K-boxes of eIF2␤ are substituted by alanines (H2892). A, K-box mutations disrupt the binding of eIF2␤ to eIF5 and eIF2B⑀. WCEs from strains H2888 and H2892 overexpressing GST were incubated with anti-FLAG M2-agarose, and proteins bound to the resin were analyzed by immunoblot analysis using anti-FLAG antibodies to detect eIF2␤ and specific antisera, as indicated and described under "Experimental Procedures," to detect other translational components. Lanes 2 and 4, the entire pellet (P) from the precipitation reactions; lanes 1 and 3, 10% of the input (In) amounts of the WCEs. Results of two independent co-immunoprecipitation reactions are summarized in the table on the right. B, K-box mutations in eIF2␤ do not impair the binding of eIF2␥ to eIF5 and eIF2B⑀. WCEs from strain H2892 expressing the indicated GST or GST-eIF2␥ fusion proteins were mixed with glutathione-Sepharose beads, and the interacting proteins were analyzed by immunoblot analysis as described in the legend to Fig. 2. Even-numbered lanes, the entire pellet (P) from the pull-down reactions; odd-numbered lanes, 10% of the input (In) amounts of the WCEs. Results of three independent pull-down reactions are summarized in the table on the right. eIF5 and eIF2B⑀ Bind eIF2␥-G Domain 405 (CTD). The GST fusion proteins were added to rabbit reticulocyte lysates programmed to express either eIF2␥-FL (Fig. 5A) or the isolated eIF2␥-G domain (Fig. 5B). Purified GST-eIF5-FL and GST-eIF5-NTD, but not GST alone, readily interacted with FL eIF2␥ (Fig. 5A, lanes 3 and  4 versus lane 2). Moreover, GST-eIF5-FL and GST-eIF5-NTD bound the isolated eIF2␥-G domain (Fig. 5B, lanes 4 and 8). Quantification of the binding, taking into account the molar amounts of GST fusion proteins recovered in the pellets, revealed that ϳ3% of the total input eIF2␥-FL or eIF2␥-G was bound by the GST-eIF5-FL and GST-eIF5-NTD fusion proteins. In contrast, the GST-eIF5-CTD fusion displayed a very weak ability to pull down the eIF2␥-G domain (Fig. 5B, lane 3; Ͻ1% binding). These results suggest that eIF2␥ directly binds to eIF5; however, it is possible that a component in the reticulocyte lysates is bridging the interaction between these two factors. To test this possibility, we expressed FLAG-tagged eIF2␥-G domain in reticulocyte lysates and then purified the protein using anti-FLAG resin. The purified eIF2␥-G domain bound to purified recombinant GST-eIF5-FL but not to GST (Fig. 5C, lane 3 versus lane 2). Again, GST-eIF5-FL bound ϳ3% of the input eIF2␥-G domain. Thus, we conclude that eIF2␥ directly binds to eIF5 and that the primary determinants for this binding are located in the NTD of eIF5.
Consistent with the results of these in vitro pull-down assays, previous studies implicated the NTD of eIF5 in regulating eIF2␥ activity. It was proposed that Arg 15 in the eIF5 NTD functions as an "arginine finger" to catalytically stimulate the hydrolysis of GTP by eIF2␥ (9,10). Consistent with this possibility, substitution of alanine, lysine, or methionine for Arg 15 in mammalian eIF5 significantly impaired GTP hydrolysis by eIF2 in 48 S initiation complexes (9,10). To further map the eIF2␥-G domain binding site on eIF5, we expressed in bacteria and purified GST-eIF5 fusion proteins lacking the N-terminal residues 1-50 (GST-eIF5⌬1-50), lacking residues 50 -100 (GST-eIF5⌬50 -100), or lacking residues 75-100 (GST-eIF5⌬75-100). Whereas GST-eIF5⌬1-50 and GST-eIF5⌬75-100 readily interacted with the eIF2␥-G domain expressed in reticulocyte lysates, pulling down 6 and 4%, respectively, of the input (Fig. 5B, lanes 5 and 7), the GST-eIF5⌬50 -100 fusion protein failed to interact with the eIF2␥-G domain above the background levels observed with GST (Fig. 5B, compare lanes 2 and 6; Ͻ1% binding). These results indicate that the primary binding determinants for the eIF2␥-G domain are present in the NTD of eIF5 and map between residues 50 and 100.
To test the ability of the various eIF5 deletion mutants to stimulate GTP hydrolysis by eIF2, the eIF5 portion of the GST-eIF5 fusion proteins was liberated by protease cleavage. Following purification, wildtype or mutant forms of eIF5 were mixed with 40 S ribosomal subunits, a model mRNA, the factors eIF1 and eIF1A, and preformed eIF2⅐[␥-33 P]GTP⅐Met-tRNA i Met ternary complexes. Measurements of the rate of GTP hydrolysis in the presence of saturating amounts of wild-type eIF5 revealed an observed rate constant of 0.12 s Ϫ1 (Fig. 5D), which is comparable with what has been reported previously (30). Interestingly, the eIF5 NTD (residues 1-279) was sufficient for stimulating GTP hydrolysis (k obs ϭ 0.11 s Ϫ1 ), whereas the CTD (k obs ϭ 0.01 s Ϫ1 ) lacked activity (Fig. 5D). Consistent with its defective binding to the eIF2␥-G domain, eIF5⌬50 -100 failed to stimulate eIF2 GTPase activity (k obs ϭ 0.01 s Ϫ1 ; Fig. 5D). In contrast, despite their abilities to bind to the eIF2␥-G domain, both eIF5⌬1-50 (0.03 s Ϫ1 ) and eIF5⌬75-100 (0.02 s Ϫ1 ) failed to stimulate GTP hydrolysis by eIF2 (Fig. 5D). Interestingly, the results obtained with the eIF5⌬1-50 mutant resemble the abolished GAP function and partially impaired eIF2␥ binding activity associated with mutating the putative arginine-finger (R15A) in full-length eIF5 (data not shown). Taken together, these results reveal that the eIF5 NTD is both necessary and sufficient for binding to eIF2␥ and stimulating eIF2 GTPase activity. Moreover, deletion of eIF5 residues 50 -100, which leaves Arg 15 intact, impaired both eIF2␥ binding and GAP activities, supporting the idea that the direct binding of eIF5 to eIF2␥ identified in this paper is important for stimulating eIF2 GTPase activity.
Like the "arginine finger" mutations in eIF5, the ssu2-1 mutation in yeast eIF5 impaired the stimulation of eIF2 GTP hydrolysis activity (24). Originally identified as a suppressor of the temperaturesensitive phenotype of the sui1-17 mutation in yeast eIF1, the ssu2-1 mutation was proposed to restore translational fidelity by lowering the rate of GTP hydrolysis by eIF2 at non-AUG start sites (24). It is noteworthy that the ssu2-1 mutation substitutes Ser for Gly 62 (G62S) in the region of eIF5 (residues 50 -100) that we showed was important for binding to the eIF2␥-G domain. To test whether the eIF5-G62S mutation affected the binding of eIF5 to eIF2␥, we expressed the various GST-eIF2␥ fusion proteins in an ssu2-1 mutant yeast strain, which only expresses the mutant form of eIF5. As shown in  . The binding of eIF2␥ to eIF5 is independent of the AA-boxes required for the interaction of eIF5 with eIF2␤ and eIF3. As described in the legend to Fig. 2, GST or various GST-eIF2␥ fusion proteins were overexpressed in yeast strains expressing FLAGtagged eIF5 (H2894), FLAG-tagged eIF5-7A in which seven conserved acidic and aromatic residues in the second AA-box in eIF5 were substituted by alanines (H2895), or untagged eIF5-G62S (F708). A and B, 7A mutation in eIF5 does not impair binding to eIF2␥. WCEs from strain H2894 (A) and H2895 (B) expressing the indicated GST or GST-eIF2␥ fusion proteins were mixed with glutathione-Sepharose beads, and the interacting proteins were analyzed by immunoblot analysis as described in the legend to Fig. 2 (except eIF5 was detected using anti-FLAG antibodies). Even-numbered lanes, the entire pellet (P) from the pull-down reactions; odd-numbered lanes, 10% of the input (In) amounts of the WCEs. C, the ssu2-1 mutation in eIF5 does not impair the interaction with eIF2␥. WCEs from strain F708 expressing eIF5-G62S (ssu2-1 mutant protein defective in promoting eIF2 GTPase activity) and the indicated GST or GST-eIF2␥ fusion proteins were mixed with glutathione-Sepharose beads. The binding of eIF5 (anti-TIF5) and eIF2B⑀ (anti-GCD6) to the GST fusion proteins was analyzed by immunoblot analysis as indicated. Even-numbered lanes, the entire pellet (P) from the pull-down reactions; odd-numbered lanes, 10% of the input (In) amounts of the WCEs. Results are representative of two independent experiments. eIF5 and eIF2B⑀ Bind eIF2␥-G Domain MAY 5, 2006 • VOLUME 281 • NUMBER 18 ing eIF5 to eIF2␥, and they likewise suggest that the Gly 62 residue, like Arg 15 , may play a more critical role in the catalytic activation of GTP hydrolysis by eIF2 on the ribosome.
The eIF2␥-G Domain Directly Binds the Minimal Catalytic Fragment of eIF2B⑀-Previous studies mapped the guanine nucleotide exchange catalytic site to the C terminus of eIF2B⑀ (residues 518 -712) (17,18,31). Moreover, eIF2B⑀ residues 524 -712 were found to fold into a stable structure consisting of eight ␣ helices and resembling a HEAT repeat (18). Interestingly, this catalytic fragment of eIF2B⑀ contains the AA-boxes that interact with eIF2␤. Assuming the interaction we detected between eIF2B⑀ and eIF2␥ is relevant to catalysis of guanine nucleotide exchange on eIF2␥, we predicted that this small catalytic fragment of eIF2B⑀ should bind to eIF2␥. To test this hypothesis, we expressed GST or GST-G in a yeast strain also overexpressing a FLAG-tagged C-terminal fragment of eIF2B⑀ (residues 524 -712). As shown in Fig. 6, GST-G (lane 4) but not GST alone (lane 2) was able to pull-down FLAG-eIF2B⑀ 524 -712 , consistent with the idea that the catalytic center of eIF2B directly contacts the eIF2␥-G domain to catalyze guanine nucleotide exchange.

DISCUSSION
Previous studies established the direct interaction between the N-terminal K-boxes in eIF2␤ and the C-terminal regions of eIF2B⑀ and eIF5 containing the AA-boxes (20 -22). It was proposed that these interactions facilitate an additional, and perhaps transient, interaction between eIF5 or eIF2B⑀ and eIF2␥; however, this latter interaction was never observed. Here we showed that eIF5 and eIF2B⑀ directly interact with eIF2␥. When expressed in yeast, a GST-eIF2␥-FL fusion protein was able to pull-down both eIF5 and eIF2B⑀ as well as the ␣ and ␤ subunits of eIF2 (Fig. 2). The ability of the GST-eIF2␥-FL fusion to pull down eIF5 and eIF2B⑀ was unaffected by eIF2␤-K12 and eIF5-7A mutations that impair direct interactions between eIF2␤ and eIF2B⑀ or eIF5 (Figs.  3 and 4). Thus, the interaction between eIF2␥-FL and eIF5 and eIF2B⑀ is direct and not bridged by eIF2␤. Further supporting this idea, GST fusion proteins consisting of the eIF2␥-G or G ϩ D2 domains were able to pull down the native eIF5 and eIF2B⑀ in yeast cells but not eIF2␣ or eIF2␤ (Figs. 2-4). Supporting the significance of these interactions, the eIF2␥-G domain-binding site was mapped to the C-terminal catalytic HEAT domain of eIF2B⑀ (Fig. 6). Finally, using recombinant proteins,

eIF5 and eIF2B⑀ Bind eIF2␥-G Domain
we confirmed the direct interaction between eIF5 and the eIF2␥-G domain and we mapped the eIF2␥-G domain-binding site to the N-terminal portion of eIF5 (Fig. 5) that was previously implicated in GAP activity. Thus, our data support a model in which eIF5 and eIF2B⑀ directly bind to the eIF2␥-G domain, consistent with their functions to promote GTP hydrolysis and GDP-GTP exchange on eIF2.
Our results help clarify the role of the eIF2␤ interaction with eIF5 and eIF2B⑀. Previously, it was proposed that the direct interaction between the N-terminal K-boxes in eIF2␤ and the C-terminal AA-boxes of eIF5 facilitated recruitment of eIF5 to the 40 S subunit to trigger GTP hydrolysis by eIF2 following start codon recognition (20 -22). Alternatively, since the slow growth phenotype of a yeast strain expressing the eIF5-7A mutant was partially suppressed by overexpression of eIF2 (20), the eIF5-eIF2␤ interaction may facilitate (together with an eIF3a-eIF2 interaction) eIF2 binding to the ribosome (prebound with eIF5 and eIF3) (20,25,32). The finding that both the GAP (eIF5) and GEF (eIF2B⑀) for eIF2 bound to the eIF2␤ subunit was initially surprising, given the presence of the G domain in the eIF2␥ subunit. Thus, two models could be proposed. 1) Binding of eIF5 or eIF2B⑀ allosterically activates eIF2␤ to promote GTP hydrolysis and/or GTP-GDP exchange on eIF2␥, or 2) the N terminus of eIF2␤ serves as a high affinity binding (docking) site for eIF5 and eIF2B⑀ and recruits these latter proteins to eIF2␥, where they directly promote GTP hydrolysis or nucleotide exchange.
The allosteric model, in which the eIF2␤ subunit activates a GAP and GEF activity inherent in eIF2␥, is consistent with the lack of eIF5, eIF2B⑀, and K-boxes in eIF2␤ in archaea. Whereas homologs of eIF2␣, eIF2␤, and eIF2␥ were readily identified in archaea and shown to form an eIF2 complex (7,33), functional homologs of eIF5 and eIF2B have not been identified. Accordingly, the K-boxes in eIF2␤ evolved along with the requirement for an external factor to promote GTP hydrolysis and GDP-GTP exchange. It is notable that all experiments examining eIF2 GTPase activity as well as nucleotide exchange have utilized eIF2 holocomplexes rather than the isolated eIF2␥ subunit (6,9,10,16,17,31). Whereas this may reflect protein solubility problems with free eIF2␥, it also is consistent with the idea that eIF2␤ activates GAP and GEF activities in eIF2␥.
Our results, revealing the direct binding of eIF5 and eIF2B⑀ to the G domain of eIF2␥, support the second hypothesis that eIF2␤ simply serves as a high affinity docking site for eIF5 and eIF2B⑀, which in turn directly regulate eIF2␥ function. Whereas our data do not rule out a regulatory function for eIF2␤, the fact that both eIF5 and eIF2B⑀ bind to eIF2␥ supports the idea that these factors directly modulate eIF2␥ function. Since the binding of eIF5 or eIF2B⑀ to eIF2␥ was only detected when one of the proteins was overexpressed, these direct interactions are apparently weak. Accordingly, we propose that the interaction of the C-terminal regions of eIF5 and eIF2B⑀ with eIF2␤ serves to recruit these factors to the eIF2 complex (Fig. 1, black arrows). Following their recruitment, the effector regions in the N-terminal half of eIF5 and the C-terminal segment of eIF2B⑀ then interact with the eIF2␥-G domain to promote GTP hydrolysis or nucleotide exchange (Fig. 1, gray arrows). The idea that eIF2B⑀ directly catalyzes guanine-nucleotide exchange on eIF2␥ is consistent with the identification of critical "exchange" residues in eIF2B⑀ (17,18,31). Similarly, a direct and catalytically important interaction between eIF5 and eIF2␥ is supported by the identification of a putative "arginine-finger" in eIF5 (9,10). In analogy to other GAPs (34), this conserved Arg residue in eIF5 is predicted to insert into the GTP-binding site of eIF2␥ and help catalyze GTP hydrolysis. However, our results indicate that the putative arginine finger (Arg 15 ) is not essential for eIF5 binding to eIF2␥ (Fig. 5B, lane 3; GST-⌬1-50-eIF5 binds eIF2␥-G) (data not shown). Moreover, whereas our results in Fig. 5D reveal that the N terminus of eIF5 is both necessary and sufficient to promote eIF2 GTPase activity, additional experiments are required to prove that Arg 15 in eIF5 functions as a direct catalytic residue in the GTP hydrolysis reaction by eIF2␥.
Finally, our results reveal unique properties for the segments of eIF5 and eIF2B⑀ containing the conserved AA-boxes. The x-ray structure of the C-terminal fragment of eIF2B⑀, which we show here binds to the eIF2␥-G domain, revealed a bundle of eight ␣-helices arranged in a so-called HEAT repeat previously observed in the nuclear cap-binding protein CBP80 and the translation factor eIF4G (18). The first AA-box in the eIF2B⑀ structure is located near the core of the HEAT domain and is predicted to be important for the structural integrity of the fold. Likewise, the second AA-box in eIF2B⑀ is located in helix VIII and may help pack this helix to the core of the structure. It is likely that the C-terminal segment of eIF5, which, like eIF2B⑀, contains two AA-box motifs, folds into a similar HEAT domain-like structure. Presumably, the HEAT domains in eIF2B⑀ and eIF5 interact with the N-terminal segment of eIF2␤ (containing the K-boxes) in a structurally similar manner. However, whereas the HEAT domain of eIF2B⑀ binds to both eIF2␤ and eIF2␥, the HEAT domain of eIF5 binds only to eIF2␤, whereas the N-terminal portion of eIF5 binds to eIF2␥. Presumably, a common surface on the HEAT domains of eIF5 and eIF2B⑀ provides a docking site for the N terminus of eIF2␤, whereas a distinct and unique surface on the eIF2B⑀ HEAT domain binds to eIF2␥ and promotes guanine-nucleotide exchange. Thus, whereas the HEAT domain is commonly thought of as a protein-docking site, the HEAT domain in eIF2B⑀ has, in addition, acquired a catalytic function.