Structure of the catalytic fragment of translation initiation factor 2B and identification of a critically important catalytic residue.

Eukaryotic initiation factor (eIF) 2B catalyzes the nucleotide activation of eIF2 to its active GTP-bound state. The exchange activity has been mapped to the C terminus of the eIF2Bepsilon subunit. We have determined the crystal structure of residues 544-704 from yeast eIF2Bepsilon at 2.3-A resolution, and this fragment is an all-helical protein built around the conserved aromatic acidic (AA) boxes also found in eIF4G and eIF5. The eight helices are organized in a manner similar to HEAT repeats. The molecule is highly asymmetric with respect to surface charge and conservation. One area in the N terminus is proposed to be directly involved in catalysis. In agreement with this hypothesis, mutation of glutamate 569 is shown to be lethal. An acidic belt and a second area in the C terminus containing residues from the AA boxes are important for binding to eIF2. Two mutations causing the fatal human genetic disease leukoencephalopathy with vanishing white matter are buried and appear to disrupt the structural integrity of the catalytic domain rather than interfering directly with catalysis or binding of eIF2.

The initiation phase of protein synthesis in eukaryotic cells is a complex series of highly regulated interactions between ribosomal subunits, mRNA, aminoacylated initiator methionyl-tRNA (Met-tRNA i Met ), and eukaryotic translation initiation factors (eIFs). 1 They all function to correctly position Met-tRNA i Met within the 80 S ribosomal P site at the correct AUG initiation codon of every mRNA (1). One of the key regulated initiation factors is eIF2, which delivers Met-tRNA i Met to the 40 S ribosomal subunit as part of an eIF2⅐GTP⅐Met-tRNA i Met ter-nary complex (TC). When TC is bound to both the 40 S ribosomal subunit and the initiator AUG codon of an mRNA, GTPase activating protein (GAP) eIF5 stimulates GTP hydrolysis releasing an eIF2⅐GDP binary complex. eIF2B acts as a nucleotide exchange factor (GEF) and promotes release of GDP from eIF2 and formation of an eIF2⅐GTP complex. Only eIF2⅐GTP can form TC, so by controlling eIF2B function, cells can control TC levels and protein synthesis initiation (1). eIF2B activity is controlled both indirectly by phosphorylation of eIF2, and directly by phosphorylation of eIF2B. Four protein kinases can phosphorylate the eIF2␣ subunit (eIF2␣) at Ser 51 . Each kinase reacts to different cellular stress conditions. GCN2 responds to amino acid starvation (2); PEK/ PERK counters damage caused by unfolded proteins in the endoplasmic reticulum; PKR is activated by double stranded RNA in response to viral infection and HRI is regulated by heme levels in reticulocytes (1). Phosphorylation of eIF2␣ (eIF2(␣P)) reduces the activity of eIF2B by formation of a non-productive eIF2⅐eIF2B complex (3). As eIF2 is more abundant than eIF2B, a small fraction of eIF2(␣P) can have a large effect on eIF2B activity and therefore significantly reduce TC levels. The reduction in TC levels by eIF2(␣P) has opposing effects; overall protein synthesis is lowered, but the translation of stress responsive genes is enhanced.
The activity of mammalian eIF2B can also be controlled directly in response to insulin signaling, which causes glycogen synthase kinase 3 inactivation and thereby contributes to activation of eIF2B (4). This permits increased eIF2B activity and protein synthesis in response to growth-promoting signals. It was recently demonstrated that mutations in eIF2B cause the fatal human genetically inherited brain disorder known as childhood ataxia with central nervous system hypomyelination or vanishing white matter leukoencephalopathy (5,6). eIF2 and eIF2B are proteins with three and five non-identical subunits, respectively. Whereas eIF2␣ appears to be largely required for regulation by phosphorylation, the ␤ and ␥ subunits have a central role in GTPase function. eIF2␥ contains the GDP/GTP binding domain and also binds Met-tRNA i Met (7), whereas eIF2␤ is required to bind both eIF5 and eIF2B (8).
Functions for each of the five subunits of eIF2B have also been assigned from molecular genetic and biochemical studies of the yeast factor. The eIF2B ␣, ␤, and ␦ subunits share extensive sequence similarity and these three subunits form a regulatory subcomplex that mediates the inhibition of eIF2B function in response to eIF2(␣P) (9). This regulatory complex binds eIF2(␣P) with higher affinity than non-phosphorylated eIF2.
In contrast, the ␥ and ⑀ subunits of eIF2B are required for the catalytic function of nucleotide exchange. These subunits share extensive similarity over the entire length of eIF2B␥. eIF2B⑀ subunits from rat, yeast, and Drosophila are capable of nucleotide exchange in vitro (3, 10, 11). We recently demon-* 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.
strated that the C-terminal ϳ200 amino acids of eIF2B⑀ contains the minimal eIF2B catalytic domain (12). This minimal fragment was proposed to contain two functional regions. The C-terminal 115 residues share sequence homology with the C termini of eIF5 and mammalian eIF4G, and has been defined as the W2 domain (two invariant tryptophans), and this domain also includes the shorter eIF5C domain. Within this region there are two AA boxes (rich in aromatic and acidic residues) (8), which are important for mediating protein-protein interactions. Multiple alanine substitutions in the AA boxes of eIF2B⑀ disrupt binding to eIF2␤ and equivalent substitutions in eIF5 have similar effects on eIF2␤ binding (8). The C terminus of mammalian eIF4G is proposed to interact with the protein kinase Mnk1, which phosphorylates eIF4E on serine 209 and has a regulatory role in translation (13).
The second region within this catalytic domain is proposed to function as the catalytic center of the enzyme. This region encompasses residues 518 -583 of the yeast protein and is well conserved in all eIF2B⑀ proteins (12). In addition, mutation of either Thr 552 to Ile (T552I) or Ser 576 to Asn (S576N) directly impairs catalytic function without detectably impairing binding to eIF2 (14). Thus, although the eIF2B⑀ C terminus is only a small fragment of the entire ϳ275-295-kDa eIF2B complex, it contains the major functional regions required for nucleotide exchange and provides a structural model for the related C termini of eIF4G and eIF5. In this study we present the x-ray structure of the eIF2B⑀ catalytic domain to 2.3 Å and genetic evidence that Glu 569 plays a major role in the function of eIF2B.

EXPERIMENTAL PROCEDURES
Protein Preparation, Crystallization, and Structure Determination-DNA encoding residues 524 -712 was amplified by PCR using the pAV1693 plasmid as template (12). The PCR product was cleaved with Eco31I creating overhangs compatible with the NcoI and BamHI cloning sites of the pET24d vector into which it was ligated, resulting in a construct having the internal eIF2B⑀ Met 524 as start methionine. This construct was expressed in Escherichia coli BL21(DE3) Rosetta cells grown in a defined medium containing selenomethionine. The protein was purified to homogeneity by anion exchange chromatography. The columns were equilibrated in buffer A (20 mM Tris-HCl, pH 7.4, 5 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, 0.5 mM EDTA) with 150 mM NaCl and eluted with a gradient to 500 mM NaCl in buffer A. Finally, the fragment was purified by gel filtration using a Superdex 75 (Amersham Biosciences) column equilibrated in buffer A with 100 mM NaCl and concentrated to 5-8 mg/ml prior to crystallization. Crystals were obtained at 5°C by vapor diffusion of drops with equal volumes of protein and reservoir against a solution containing 0.2 M ammonium acetate, 23-25% PEG 2000 monomethyl ether, 100 mM Na citrate, pH 5.75. The crystals belong to the tetragonal space group I422 (a ϭ 106.34 Å, c ϭ 91.71 Å) with one molecule per asymmetric unit. Crystals were flash-frozen in liquid nitrogen directly from the mother liquor supplemented with a few microliters of reservoir solution prior to data collection. Anomalous data collected from a selenomethionine-substituted crystal were collected at 100 K at the EMBL BW7A (Table I). The data were processed and scaled with MOSFLM and SCALA (15), and multiple wavelength anomalous dispersion (MAD) phases were calculated with CNS (16). After density modification a starting model containing 135 residues was build with ARP/wARP (17). This model was completed in O (18), and refined with CNS, using the MAD phases as restraints (Table I). Figures except for Fig. 2 were produced in MOLMOL (19). The coordinates and structure factors have been deposited in the Protein Data Bank (accession code 1PAQ).

RESULTS
Structural Overview-The fragment used for structure determination contained residues Met 524 -Glu 712 of eIF2B⑀. Despite that no degradation was observed during crystallization, no residues beyond Asp 544 -Asp 704 could be traced in either the experimental map calculated from MAD (24) phases based on data from three wavelengths collected from a selenomethionine-substituted crystal ( Fig. 1) or in the final 2F o Ϫ F c map. Hence, the first 20 residues and the last 8 residues are disordered in the crystal. This may explain the R free value of 27.2% (Table I), which is slightly higher than expected. The residues in the N terminus may require residues upstream in eIF2B⑀ or other residues in eIF2B to become ordered, whereas those in the C terminus may become ordered during complex formation with eIF2 (see below). Alternatively, phosphorylation of two conserved serines in the C terminus, as demonstrated for mammalian eIF2B⑀, could have an effect on the structure of this region (4).
The structured fragment of eIF2B⑀ containing residues Asp 544 -Asp 704 (eIF2B⑀C) has overall dimensions of 47 ϫ 43 ϫ 32 Å. The protein is all-helical ( Fig. 2) with ␣-helices I-VIII arranged in pairs with an angle between them of 129 -166°. One such helical hairpin then packs against the next pair with an angle of 35-75°. This results in a left-handed superhelix with an axis approximately perpendicular to the axes of the individual helices (Fig. 3). There are also two 3 10 helices, the first in residues 619 -624 is located right after helix IV, and the second with residues 684 -686 prior to helix VIII. The protein has the overall appearance of a heart (Fig. 3, top) with helices III-VI forming the center, whereas helices I and II and VII-VIII are located on either side. Helices I, III, V, and VII form one surface of the molecule (the A-face), whereas the remaining helices are on the opposite side (the B-face). The three edges of the molecule can be named according to the contributing helices (Fig. 3). The N-edge is made from residues in the helices I and II, whereas helices VII and VIII contribute to the C-edge. Finally, the top edge contains residues from all helices.
The structure can be divided into a structural core containing helices III-VIII with structural homology to other proteins onto which N-terminal helices I and II are attached. This core is organized around the two AA boxes (Figs. 2 and 3), which is shared with eIF5 and mammalian eIF4G. For both eIF2B⑀ and eIF5 the AA boxes are required for interaction with the common binding partner, the N-terminal half of the ␤-subunit from eIF2 (8). In contrast to the exchange activity of eIF2B, eIF5 acts as a GAP by stimulating hydrolysis of GTP bound to eIF2␥ in the preinitiation complex (25). Hence, it is very likely that both eIF5 and eIF2B are in contact with the nucleotide binding pocket of eIF2␥ although no such interaction has been reported.
Structural Homology of eIF2B⑀C to Other Proteins-The structure of eIF2B⑀C is the first structure of a protein containing the AA boxes (see below). A search for homologous proteins with DALI (26) reveals a similar arrangement of helices in several other proteins. The best Z-score, 12.0, is obtained by the 80-kDa subunit of the human nuclear cap binding complex (27), where all eight helices in eIF2B⑀C and eight helices within residues 497-651 in domain 3 of the large subunit of the complex can be superimposed with a root mean square deviation (r.m.s.d.) of 2.0 Å for 122 C ␣ atoms (Fig. 4). The second best match with a Z-score of 7.4 is obtained with a central fragment of initiation factor 4GII (MIF4G), and helices III-VIII of eIF2B⑀C can be matched with the six N-terminal helices of MIF4G with an r.m.s.d. of 2.1 Å over 77 C ␣ atoms (Fig. 4) (28). Importantly, this fragment does not contain the AA boxes, which are found at the C terminus of mammalian eIF4G.
Because eIF2B acts as a GEF and eIF5 as a GAP, structural homology to other GEFs or GAPs is of interest. The only close structural homologues of eIF2B⑀C detected in a DALI search are the Ras GEF (SOS) and GAP (p120 GAP ). Helices III-VIII of eIF2B⑀C superimpose with six helices within residues 605-739 of SOS with a Z-score of 4.7 (78 C ␣ atoms with r.m.s.d. of 3.0 Å). However, the matching fragment of SOS is not part of the catalytic site of SOS, but rather involved in binding of a second Ras molecule, which has been characterized very recently (29). After superimposition of eIF2B⑀C onto the complex between SOS and two molecules of Ras, a loop region of eIF2B⑀C containing non-conserved residues (Gln 603 -Asp 609 ) comes close to both switch I and II of the second Ras molecule. Thus, the Gln 603-609 loop could be implicated in binding to eIF2 and displacing GDP. An important role seems unlikely, however, as sequence is not well conserved among eIF2B⑀ proteins in this region. The first six helices of eIF2B⑀C can also be superimposed on six helices within residues 732-900 of p120 GAP , the GAP of Ras (DALI score 6.1, r.m.s.d. of 2.0 Å for 66 C ␣ atoms), but in contrast to the three other cases, there is a very large insert between the second and the third matching helical hairpin of p120 GAP . Nevertheless, this similarity superimposes the where F c is the calculated structure factor scaled to F o ; R free is identical to R on a subset of test reflections not used in refinement; FOM, figure of merit. The statistics for the Ramachandran plot are residues in most favored plus additionally favored regions/generously allowed regions/disallowed regions. Due to the presence of sclenomethionine, and the use of MAD phases as restraints, both reflections of the Friedel pairs were used for refinement. "GAP finger" Arg 789 (30) on the beginning of helix IV in eIF2B⑀C, but as observed above for the SOS homology, the primary sequence in this area is also not well conserved between species (see below). The corresponding region is also poorly conserved in eIF5, the eIF2 GAP. If this region was important for GAP function in eIF5, it would most likely be better conserved. Hence, the similarity of eIF2B⑀C (and by homology eIF5) to both the Ras GEF and GAP may indicate regions important for protein-protein interactions, but the low sequence conservation would imply the regions do not contain residues directly important for catalysis.
The packing observed for helices III-VIII of eIF2B⑀C is conserved within cap binding complex 80, MIF4G, and the SOS fragment. The resemblance of this helix arrangement in MIF4G to HEAT repeats has previously been described (28). HEAT repeats are often involved in protein-protein or proteinligand interactions (31). The presence of helical repeats resembling HEAT repeats has now been observed in eIF4G and eIF2B, and must also be expected in eIF5 because of the conserved AA box motifs. Finally, eukaryotic elongation factor 3 (the "E" of HEAT) was one of the first proteins that was predicted to contain HEAT repeats (32). Thus it seems that such repeats are widespread among translation factors even though they often cannot be predicted from primary sequence alone, as demonstrated here for eIF2B⑀C.
The Surface Properties of eIF2B⑀C-The surface properties of the fragment with respect to the distribution of charges and exposed conserved residues of putative functional importance are highly asymmetric (Fig. 3, middle and right column, re-spectively). The overall charge of the fragment is quite acidic with an isoelectric point of 4.3, and the surface electrostatic potential is dominated by a very large acidic belt, which starts at the N-edge, continues over the A-face over the top edge of the molecule, and ends at the B-face at the C terminus of helix VIII. A minor positively charged (basic) patch is located on the B-face and top edge around which the acidic belt is "tied." The remaining charges are more or less randomly distributed (Fig. 3). The majority of the residues constituting the acidic belt and basic patch of eIF2B⑀ are conserved. The acidic belt includes residues Asp 564 , Glu 569 , Glu 548 , Glu 583 , Asp 634 , Asp 666 , Glu 670 , Asp 704 , and Asp 671 , which are all highly conserved with respect to charge (the order of the listed residues following the direction of the belt as described above). Three of these residues, Glu 670 , Asp 671 , and Asp 704 , are conserved AA box residues (see below). The basic patch includes Arg 574 , Lys 613 , Lys 623 , and Arg 624 .
As with the electrostatic potential, the exposed conserved residues are primarily found at the B-face, N-edge, and top edge (Fig. 3), and are roughly organized in two patches. The first of these contains residues from helices I and II and associated loops, whereas the second contains residues from helices VII-VIII and their associated loops. In this second area, residues from the AA boxes are dominating and the strictly conserved Tyr 663 , Glu 670 , Trp 699 , and Leu 700 together with the highly conserved Ala 703 form an exposed "handle" at the C-edge (Fig. 3).
The AA Boxes-The AA boxes contain conserved aliphatic, aromatic and acidic residues (Fig. 2). The first box contains 12 conserved residues within amino acids Leu 655 -Trp 676 , which Right, residue conservation mapped on the surface with the following color codes: green, 100% identity (according to the alignment shown in Fig. 2); gold, between 100 and 80% identity; magenta, between 80 and 50% identity; and light gray, below 50% identity. are all located within helices VI and VII (Fig. 5). The second box has eight conserved residues within amino acids Trp 696 -Glu 706 , of which the two last, Glu 705 and Glu 706 , are disordered in our structure. This AA box is located at the C terminus of helix VIII. The first box is important for the structural integrity of the helix III-VIII core. Residues Ala 658 in helix VI and Ile 667 in the following loop contact helix IV and the following 3 10 helix, whereas Leu 655 and Leu 659 in helix VI together with the strictly conserved Trp 676 in helix VII pack with helix V. The indole ring of the Trp 676 also engages in a hydrogen bond with the side chain of Asn 637 , and furthermore, packs with side chains of Met 636 , Lys 675 , and Trp 677 .
The strictly conserved Tyr 663 is at the center of a cluster of five residues from the AA boxes linking helices VI-VIII (Fig. 5). It engages in van der Waal interactions with Glu 670 , Ile 673 , Trp 696 , and Leu 700 , which, except for the tryptophan, are strictly conserved in eIF2B⑀. Furthermore, the hydroxy group of Tyr 663 forms a hydrogen bond with the side chain carboxyl group of Glu 670 . On one side this cluster is flanked by Tyr 674 , which fixes the end of helix VIII through a hydrogen bond from its hydroxy group to the side chain of Asp 704 , the last residue in our structure. To the other side Phe 656 flanks the cluster.
The N-terminal Helices-Although the two N-terminal helices are firmly associated with the six helical core of eIF2B⑀C, residues 544 -576 show significantly higher temperature factors (46.1 Å 2 ) compared with amino acids 577-704 (34.8 Å 2 ). This is not because of very high mobility of a few disordered residues, but rather a general trend for all atoms. Hence, these two helices are by average more mobile than the rest of the structure. This could be of functional importance for the exchange reaction, as these two helices are likely to be directly involved in catalysis, but may also be caused by missing residues from either eIF2B⑀ itself or the other subunits of eIF2B. One indication in favor of their functional importance is that these two helices and the loop to helix III contain three strictly conserved residues. Strictly conserved residues are otherwise only found in the AA boxes. A direct function of residues from the two helices in catalysis is also supported by genetic and biochemical data (12) (see also below). The interface between helices I-II and the rest of the molecule is not extensive in agreement with the elevated temperature factors. Central contacts are formed by the packing of Met 557 and Leu 563 with Arg 595 and Trp 618 from helices III and IV, respectively. Furthermore, there are important polar interactions; Asp 564 forms a salt bridge with Arg 624 , and a water molecule bridges the side chains of Glu 554 and Arg 594 . At the end of helix II Asn 571 and Arg 574 interact with Tyr 581 in helix III.
Mutation of Glutamate 569 Is Critical for in Vivo Function-To identify residues important for catalysis of nucleotide exchange we introduced single alanines in place of selected conserved residues within helices I and II of eIF2B⑀C. As we had previously identified two residues within this region of yeast eIF2B⑀C, Thr 552 and Ser 576 , that when mutated significantly impaired eIF2B function in yeast (14) we suspected that other changes here might significantly reduce eIF2B activity.
We selected conserved residues with side chains implicated in catalysis of nucleotide exchange in other GEFs. Charged residues were selected as these are important for function in many other GEFs. Leucine 938 is important for the exchange reaction in SOS. The RAS-SOS co-crystal structure reveals that it disrupts magnesium binding to the RAS nucleotide binding pocket (36). Thus, two leucine residues were selected. Finally, Asn 578 was chosen as it is universally conserved. Each residue was changed to Ala. The mutations were introduced into the GCD6 gene on a low copy plasmid and were shuffled into a gcd6⌬ strain. Surprisingly, of seven mutations analyzed, six exhibited no obvious growth defect (Fig. 6A). Only one mutation, E569A, had a significant phenotype. It was lethal. This suggested that nucleotide exchange was severely reduced in this strain. The essential function of eIF2B can be overcome in yeast by overexpressing four genes encoding TC factors: the three eIF2 subunits and one of the tRNA i Met genes (12,33). The resulting strain (GP4115) is severely slow growing. We asked whether a plasmid bearing gcd6-E569A or any of our other previously described reduced activity mutants could improve the growth rate of our mutant strain where deletion of gcd6 is rescued by high copy TC. We found that wild type GCD6 and reduced activity mutants F250L, T552I, and S576N fully rescued the slow growth phenotype of this strain (Fig. 6B). The previously described mutation N249K was partially functional in this assay. N249K, like E569A is lethal in an otherwise normal strain, but does retain some eIF2B activity in vitro (14). These results show that growth rate in strain GP4115 is an extremely sensitive in vivo assay for eIF2B function. Remarkably, E569A was unable to rescue growth in our assay and grew as poorly as vector alone. The Gln 500 * mutation was equally unable to rescue growth (Fig. 6B, Q500*). Gln 500 * expresses a protein with a premature nonsense codon that eliminates the eIF2B⑀C region entirely, but is still able to interact with the other four eIF2B subunits to form a non-functional complex (14). Immunoblotting from extracts of these cells (Fig. 6C) confirmed that eIF2B⑀ E569A protein levels were not reduced when compared with the wild type protein. So in summary, an eIF2B complex containing E569A is as defective for eIF2B function in vivo as a complex lacking the eIF2B⑀C region entirely. In contrast other single alanine substitutions to adjacent conserved residues had no major defect in eIF2B function under ideal growth conditions. FIG. 6. Glu 569 is important for catalysis. A, following plasmid shuffling into strain KAY16 (gcd6⌬), serially diluted cultures of yeast strains bearing the indicated mutations were grown on SD medium for 2 days at 30°C. B, growth of the indicated mutants transformed into strain GP4115 (gcd6⌬, high copy TC) on SD medium at 30°C for 3 days. C, immunoblot of cell extracts from the indicated strains from panel B. The subunit-specific antisera on the right was used to probe for the subunit listed on the left of each panel. 10 g protein is loaded lanes 1, 3, and 5; 20 g is loaded in lanes 2, 4, and 6. and Met 640 corresponding to human Trp 628 and Glu 650 , which both cause the brain disease childhood ataxia with central nervous system hypomyelination when mutated to Arg and Lys, respectively, are shown (6). Their function seems to be predominantly structural. The locations of Thr 552 and Ser 576 also are shown (14). When mutated to Ile and Asn, respectively, a slow growth phenotype of mutant yeast is observed in combination with a reduced catalytic activity of eIF2B. Finally, the location of Glu 569 is shown, and mutation of this to alanine is lethal in yeast.

DISCUSSION
The Bipolar Properties of eIF2B⑀C-The catalytic activity of eIF2B has been mapped to residues 518 -712 by genetic analysis, in vitro exchange, and pull-down assays (14). Furthermore, in vitro studies showed that deletion of residues 518 -580 results in loss of exchange activity, whereas binding to eIF2 is preserved (12). These results are in excellent agreement with the structure presented here, because helices I and II are within 518 -580. Hence, they are not required for binding eIF2 but very likely to be involved in catalysis, whereas helices III-VIII are sufficient for eIF2 binding but not for catalysis. One conserved surface patch at helices I and II seems well suited for participation in catalysis as this patch contains Glu 569 (Figs. 3 and 7), and mutation of this residue to alanine eliminates eIF2B function in vivo (Fig. 6). The glutamate is located in the center of the patch at the N-edge/top edge, and flanks a negatively charged depression between helices I and II, which also contains Thr 552 . This threonine is engaged in hydrogen bonding to Glu 548 , hereby fixing this residue. Mutation of Thr 552 and another residue close to this area, Ser 576 , has previously been shown to reduce exchange activity (Table II) (14).
Consistent with our findings that Glu 569 is important for eIF2B catalytic function, structures of complexes between Gproteins and their GEFs have previously emphasized the importance of glutamates or aspartates in the reaction mechanism. In the Arf1-Sec7 complex, the Glu 97 side chain of Sec7 overlaps with the binding site for the Mg 2ϩ and ␥-phosphate. It also forms a salt bridge with the conserved P-loop lysine (34), and mutation of this glutamate reduces exchange activity by orders of magnitude (35). In the Tiam1-Rac1 complex Glu 1047 from Tiam1 interacts extensively with switch I of Rac1, and places an Ile from Rac close to the GDP ribose binding site (36). In the Ras-Sos complex, Glu 942 from SOS forms a hydrogen bond with Ser 17 from the P-loop of Ras, and this prevents binding of both phosphates and Mg 2ϩ to Ras (37). In the EF-Tu⅐EF-Ts complex, Asp 80 from EF-Ts displaces EF-Tu switch II, thereby disrupting the EF-Tu Mg 2ϩ binding site (38). As shown by these examples, the GEF Asp/Glu do not recognize equivalent parts of the nucleotide binding pocket of their target G-protein, but can contribute to exchange by interaction with switch I, switch II, or the P-loop of the G-protein.
In vitro experiments showed that residues 580 -712 are sufficient for binding to eIF2 (12), and this result can now be rationalized, as these residues contain helices III-VIII, the core of the molecule. The eIF2␤ subunit contains three lysine-rich boxes, which mediates binding of eIF2 to both eIF2B and the GAP eIF5 (8,39). These positively charged stretches of eIF2␤ are likely to interact with the acidic belt observed in eIF2B⑀C. The AA box motifs of eIF4G have also been implicated in protein-protein interactions. eIF4G has been shown to bind to the eIF4E kinase, Mnk1. The N-terminal 23 residues of Mnk1 necessary for this interaction also contain a lysine-rich region very similar to the lysine-rich boxes in eIF2␤ (13).
Comparison of eIF2B⑀C with eIF5-The catalytic fragment of eIF2B shares a number of characteristics with the eIF2 GAP, eIF5. The AA boxes are important for interaction of both proteins with the common substrate eIF2␤ (8). Double mutations E346A,E347A and E384A,E385A of rat eIF5, corresponding to yeast eIF2B⑀ Glu 670 -Asp 671 and Glu 705 -Glu 706 (Fig. 2), caused severe defects in eIF5 binding to eIF2␤ (Table II) (40). The hexamutant E345A,E346A,E347A,E384A,E385A,E386A showed strongly decreased binding to eIF2␤ (40). These residues can now be mapped to equivalents in yeast eIF2B⑀C, except for residues Glu 705 and Glu 706 for which we have no electron density. They are located in the acidic belt at the C-edge/top edge of eIF2B⑀C (Figs. 3 and 5). Thus, it is clear that this part of eIF2B⑀C (and eIF5) is critical for binding to eIF2␤. However, one major functional difference between eIF2B⑀C and eIF5 is that Arg 15 essential for GAP activity of eIF5 (41,42) is located at the N terminus of the protein quite distant in the primary structure from C-terminal residues 241-405 in eIF5.
The Exchange Factors of Translation-eIF2␥ shares extensive structural and functional similarity to translation elongation factors eEF1A and prokaryotic EF-Tu. All three are Gproteins that also bind aminoacylated tRNAs and interact with the ribosome. In addition, all three factors require a nucleotide exchange factor: eIF2B, eEF1B, and EF-Ts, respectively. Despite these obvious similarities all three exchange factor catalytic domains do not share structural similarities. eIF2B⑀C is an all-helical protein, whereas the eEF1B catalytic fragment is organized with a central ␤-sheet surrounded by two helices (43) and EF-Ts offers a third structural solution (38). The common properties of the three homologous G-proteins do apparently not impose any restraints on the structure of their exchange factors.
Partial Structural Basis for a Genetic Disease-Childhood ataxia with central nervous system hypomyelination also called leukoencephalopathy with vanishing white matter, is a rare recessive fatal genetic disease. The disease is caused by missense mutations in any of the five eIF2B subunits, and the majority are found in the catalytic eIF2B⑀ subunit (5). Two of these mutations can now be mapped onto the structure of yeast eIF2B⑀C (Fig. 7) (6), whereas the remaining mutations are located elsewhere in eIF2B outside our structure. The diseaselinked mutations of W628R and E650K in human eIF2B⑀ correspond to yeast Trp 618 and Met 640 , respectively. Both are well defined in our structure, and probably have very similar locations in human eIF2B. The tryptophan interacts tightly with Met 557 , as mentioned, and this interaction is vital in the interface between helices I and IV. This interaction should be conserved in human eIF2B⑀, where human Trp 628 probably interacts with human Lys 568 in the same manner as the tryptophan-methionine pair in the yeast subunit. The long aliphatic side chain of the human lysine is equivalent to the aliphatic side chain of the yeast methionine. Preliminary modeling shows that an arginine can roughly fill the space occupied by Trp 618 in the yeast structure. An arginine in this position might be detrimental, as repulsion could occur with the human Lys 568 , but it is also perfectly located for making a hydrogen bond with the backbone between yeast His 561 and Asp 562 . For the second mutation, human E650K, the equivalent yeast Met 640 is located between the Leu 655 from AA box 1 and also in the vicinity of Trp 676 . Modeling indicates that a lysine can be accommodated here, but such a lysine in human eIF2B⑀ might be attracted into a salt bridge with human Asp 651 , and thereby decrease the stability of the hydrophobic core around human Met 663 and Trp 684 . Hence, the yeast structure of eIF2B⑀C shows that both the pathogenic mutations found in the C terminus of human eIF2B⑀ are buried, so these residues are not directly involved in catalysis or in the interface to eIF2 or other parts of eIF2B. This suggests that their mutation cause the phenotype by disturbing the structural integrity of the domain.