Mapping of Functional Domains in Eukaryotic Protein Synthesis Initiation Factor 4G (eIF4G) with Picornaviral Proteases IMPLICATIONS FOR CAP-DEPENDENT AND CAP-INDEPENDENT TRANSLATIONAL INITIATION*

Cap-dependent binding of mRNA to the 40 S ribosomal subunit during translational initiation requires the as- sociation of eukaryotic initiation factor 4G (eIF4G; formerly eIF-4γ and p220) with other initiation factors, notably eIF4E, eIF4A, and eIF3. Infection of cells by picornaviruses results in proteolytic cleavage of eIF4G and generation of a cap-independent translational state. Rhinovirus 2A protease and foot-and-mouth-disease virus L protease were used to analyze the association of eIF4G with eIF4A, eIF4E, and eIF3. Both proteases bi- sect eIF4G into N- and C-terminal fragments termed cp N and cp c . cp N was shown to contain the eIF4E-binding site, as judged by retention on m 7 GTP-Sepharose, whereas cp c was bound to eIF3 and eIF4A, based on ultracentrifugal co-sedimentation. Further proteolysis of cp N by L protease produced an 18-kDa polypeptide termed cp N2 which retained eIF4E binding activity and corresponded to amino acid residues 319-479 of rabbit eIF4G. Further proteolysis of cp c

GTP-Sepharose, whereas cp c was bound to eIF3 and eIF4A, based on ultracentrifugal co-sedimentation. Further proteolysis of cp N by L protease produced an 18-kDa polypeptide termed cp N2 which retained eIF4E binding activity and corresponded to amino acid residues 319-479 of rabbit eIF4G. Further proteolysis of cp c yielded several smaller fragments. cp C2 (~887-1402) contained the eIF4A binding site, whereas cp C3 (~480-886) contained the eIF3 binding site. These results suggest that cleavage by picornaviral proteases at residues 479-486 separates eIF4G into two domains, one required for recruiting capped mRNAs and one for attaching mRNA to the ribosome and directing helicase activity. Only the latter would appear to be necessary for internal initiation of picornaviral RNAs.
Translation of eukaryotic cellular mRNA into protein is a complex process involving nearly 200 RNA and protein components interacting in a regulated fashion to ensure timely expression of genetic information (1). Viruses often alter the host cell translational machinery to allow more efficient expression of virally encoded proteins. One of the most dramatic examples of this occurs upon picornaviral infection. The picornaviridae are small, icosohedral, positive-stranded RNA viruses of considerable clinical and veterinary importance, containing such members as poliovirus, rhinovirus, coxsackievirus, encephalomyocarditis virus, and foot-and-mouth-disease virus (FMDV) 1 (2,3). Infection of mammalian cells by most members of this family (all genera except cardioviruses and hepatitis A virus) impairs the ability of the host cell to translate capped mRNAs. Translation of the uncapped viral RNA still proceeds, however, by a cap-independent mechanism whereby ribosomes bind internally at specific sites on the viral RNAs (4,5). Thus, picornaviral infection results in the conversion of the predominant mode of translation initiation from cap-dependent to capindependent.
The host cell shutoff by rhinoviruses, enteroviruses, and aphthoviruses is thought to be mediated, at least in part, by the virally induced cleavage of a protein synthesis initiation factor, eIF4G 2 (6)(7)(8). The loss of the ability to translate capped host mRNAs correlates with the conversion of eIF4G, detected immunologically, from a cluster of bands migrating on SDS-PAGE at 200-220 kDa to one of 100-130 kDa (6,9). This event is mediated by virally encoded proteases (10,11). The cleavage products of eIF4G detected in vivo are electrophoretically identical to those generated in vitro using highly purified recombinant forms of the viral proteases (12). Although early evidence suggested that viral proteases played only an indirect role in eIF4G proteolysis (13,14), recent results support direct cleavage of eIF4G by virally encoded proteases (15)(16)(17). However, the observation that eIF4G cleavage is not sufficient for total host cell shutoff in vivo (18,19) suggests that additional events may play a role in the inhibition of cap-dependent translation. eIF4G functions during initiation of translation through association with other members of the eIF4 group of translation factors (reviewed in Refs. [20][21][22]. This group includes eIF4A, a 46-kDa bi-directional ATP-dependent helicase; eIF4B, a 70-kDa RNA-binding phosphoprotein which enhances eIF4A activity; eIF4E, a 25-kDa cap-binding phosphoprotein; and eIF4G itself, a 154-kDa phosphoprotein which can be isolated in complexes with eIF3, eIF4A, eIF4B, and eIF4E (23)(24)(25). The best characterized of these complexes, eIF4F, consists of eIF4A, eIF4E, and eIF4G (23). These polypeptides collectively recognize the m 7 GTP-containing cap, unwind mRNA secondary structure, and facilitate binding of the 40 S ribosomal subunit.
The mechanism by which cleavage of eIF4G by picornaviral proteases contributes to an inhibition of cap-dependent translation is not understood, but some functions associated with eIF4F appear to be inactivated. Following poliovirus infection, cross-linking of eIF4 polypeptides to the cap structure is altered (9,26,27). The pattern of cap-binding protein complexes recovered from infected cells is distinctly different from that of uninfected cells, suggesting that infection disrupts macromolecular complexes important for cap-dependent initiation (28,29). Since no alteration in other eIF3 or eIF4 polypeptides is apparent as a result of infection (6,30,31), cleavage of eIF4G is thought to be responsible for the observed changes. This is supported by the fact that protein complexes containing intact eIF4G restore cap-dependent translation in extracts of infected cells (23,32,33). Also, extracts of uninfected cells, when treated with purified viral proteases, lose their ability to translate capped mRNA, but addition of eIF4G-containing complexes restores this activity (8,16).
Not only does cleavage of eIF4G result in inactivation of some functions important for cap recognition, but there is also evidence that eIF4G may be important for cap-independent translation of viral RNAs. Addition of the eIF4F complex stimulates in vitro translation from picornaviral internal ribosome entry site sequences (34)(35)(36). In contrast to cap-dependent initiation, cleavage of eIF4G by picornaviral proteases does not abrogate this stimulation but rather enhances it (16,37,38). This suggests that eIF4G cleavage products play a direct role in cap-independent translation of viral RNAs (16,37). Picornaviral proteases which cleave eIF4G fall into two separate classes. The 2A proteases of rhino-and enteroviruses are small thiol proteases with structural similarities to chymotrypsin and α-lytic protease (39)(40)(41). The primary cleavage site in rabbit eIF4G is Arg 486 -Gly 487 (15), although a secondary site of unknown location has been suggested for the 2A protease of HRV2 (12). The L protease of FMDV more closely resembles papain in structure (42). It initially cleaves eIF4G at Gly 479 -Arg 480 and subsequently at multiple sites as yet unidentified (12).
To understand better the role of eIF4G in cap-dependent and -independent initiation, we have examined the effect of proteolysis on the association of eIF4G with other initiation factors using recombinant viral proteases. We provide evidence that proteolysis of eIF4G at the primary site separates functional domains. We also report the localization of a second L protease cleavage site in eIF4G which further defines the binding region for eIF4E. These results suggest a model for the role of eIF4G cleavage products in translation of picornaviral RNA.

EXPERIMENTAL PROCEDURES
Materials -The C4 column for reverse phase separation of eIF4G cleavage products was obtained from Vydak (Hesperia, CA). HPLCgrade acetonitrile was obtained from Fisher. m 7 GTP-Sepharose was purchased from Pharmacia Biotech Inc. m 7 GTP and antipain were obtained from Sigma. Peptide 9, consisting of the sequence KEFEKD-KDDDWFEC, was synthesized by Bio-Synthesis, Inc. (Dallas, TX). This sequence corresponds to amino acid residues 653-666 of human eIF4G (43) except that residue 664 was changed from Glu to Val to enhance the yield of peptide synthesis, and the C-terminal Cys was added to aid in coupling to a solid support. HPLC-grade trifluoroacetic acid, maleimide-activated forms of keyhole limpet hemocyanin and bovine serum albumin, and Freund's complete and incomplete adjuvant were obtained from Pierce. An Affi-Gel HZ immunoaffinity kit was purchased from Bio-Rad. Elastatinal was obtained from Calbiochem. Horseradish peroxidase-conjugated goat anti-rabbit IgG was obtained from Vector Laboratories (Burlingame, CA). All other chemicals were of reagent grade.
Antisera-The production of rabbit anti-eIF4G 1230-1248 (raised against a synthetic peptide corresponding to amino acid residues 1230-1248 of human eIF4G), anti-eIF4G 327-342 , and anti-eIF4G 403-416 antisera was described elsewhere (43). A mouse monoclonal antibody against eIF4A was a generous gift from Dr. Hans Trachsel, Bern, Switzerland. A goat anti-eIF3 antibody was kindly donated by Dr. William Merrick, Case Western Reserve University. Peptide 9 was coupled to maleimide-activated keyhole limpet hemocyanin using the protocol supplied by the manufacturer and used to raise anti-eIF4G 653-666 antibodies in rabbits (500 jug/immunization; injections bimonthly for 3 months).
Purification of Proteins -Rabbit reticulocyte lysate was prepared as described (44). Preparation of ribosomal salt wash and high salt postribosomal supernatant from rabbit reticulocytes, and purification of eIF4F and eIF3:eIF4F complexes from ribosomal salt wash, were as described previously (25,45). Recombinant HRV2 2A and FMDV L proteases were expressed in bacteria and purified as described previously (12,16).
Reverse Phase Column Fractionation of eIF4G Cleavage Products -For the generation of cp N2 for N-terminal sequence analysis, high salt postribosomal supernatant (70 ml) from rabbit reticulocytes, dialyzed against buffer A 100 (20 mM MOPS, pH 7.6, 10% glycerol, 0.1 mM EDTA, 0.25 mM dithiothreitol, 100 mM KC1) was incubated with L protease (0.5 µg/ml) for 20 min at 30 °C and then applied to a 2-ml m 7 GTP-Sepharose column equilibrated in buffer A 100 at 4 °C. The column was washed with 20 ml of buffer A 100 and 10 ml of buffer A 100 containing 100 µM GTP. Bound material was eluted in buffer A 100 containing 70 µM m 7 GTP. The eluate (3 ml) was further incubated with L protease (3 µg/ml) at 30 °C for 20 min to complete the conversion of cp N to cp N1 and cp N2 . The reaction mixture was cooled on ice, brought to 0.1% in trifluoroacetic acid, and fractionated by HPLC using a C4 column (0.45 x 15 cm) equilibrated in buffer B (0.1% trifluoroacetic acid). The column was developed at 1 ml/min with 5 ml of buffer B followed by a 1-ml step to 35% buffer C (0.1% trifluoroacetic acid in 95% acetonitrile) and a linear 39-ml gradient to 60% buffer C. Fractions of 1 ml were collected throughout and those shown by SDS-PAGE and immunoblotting to contain cp N1 and cp N2 were pooled and reapplied to the C4 column. The column was developed with 5 ml of buffer B followed by a 1-ml step to 30% buffer C and a linear 40-ml gradient to 45% buffer C. Fractions of 0.5 ml were collected throughout and those containing cp N2 were pooled and subjected to automated Edman degradation with an Applied Biosystems model 477A sequenator at the Louisiana State University Core Research Laboratories, New Orleans.
Generation of eIF4G Cleavage Products and m 7 GTP-Sepharose Column Fractionation-For analysis of the time course of digestion, eIF4F (30 µg/ml), either in the presence or absence of eIF3 (70 µg/ml), was incubated at 30 °C with L protease (12 µg/ml) in buffer A 140 (buffer A 100 except the KC1 concentration was 140 mM). Aliquots (30 µl) were removed at various times and subjected to SDS-PAGE and immunoblotting. For analysis of the eIF4G-eIF4E interaction by m 7 GTP-Sepharose chromatography, eIF4F (27 fig/ml) was incubated in buffer A 150 in the presence of either 2A (8 µg/ml) or L protease (12 µg/ml) at 30 °C for 60 min. The reactions were stopped with elastatinal (500 µM) or antipain (1 mM), respectively, and placed on ice. Samples were then loaded onto a 100-µl m 7 GTP-Sepharose column equilibrated in buffer A 100 , the column washed with 300 µl of buffer A 100 , and the bound material eluted with buffer A 100 containing 70 µM m 7 GTP. Ultracentrifugal Sedimentation -Reactions (30 µl) containing the eIF3 eIF4F complex (560 µg/ml) were incubated at 30 °C in buffer A 80 in the absence or presence of either 2A protease (8.3 µg/ml) or L protease (95 µg/ml) for either 50 or 5 min, respectively. Reactions were stopped with protease inhibitors as described above, placed on ice, layered onto 4.4-ml 15-30% sucrose gradients prepared in buffer A 80 and centrifuged in an SW60.1 rotor at 45,000 rpm for 16 h.
Immunologic Procedures -For immunoblotting, unstained proteins were transferred to polyvinylidine difluoride membranes in a BioRad Mini Trans-Blot cell using the manufacturer's recommendations. Detection of immunoreactive species was performed as described previously (43) except incubations were performed at 22 °C. Anti-eIF4G 653-666 antiserum was used at a 1:1000 dilution. Anti-eIF4G 1230-1248 resin was prepared using the Affi-Gel HZ immunoaffinity kit. The antibody was purified on protein A-Sepharose, and 1.5 mg were coupled through the carbohydrate moiety by a hydrazide linkage to 2 ml of agarose beads. For immunoadsorption, various combinations of eIF4A and eIF4F were incubated with FMDV L protease at 30 °C in buffer D 50 (buffer A 50 , containing 0.1% Tween 20). Reactions were stopped by the addition of 1 mM antipain. Aliquots (50 µl) were added to 450 µl of buffer D 50 containing 0.5% non-fat dry milk and 100 µl of a 25% slurry of anti-eIF4G 1230-1248 resin in buffer D 50 containing 0.5% non-fat dry milk. The mixture was incubated for 2 h at 4 °C with continuous mixing and then the resin was sedimented by centrifiigation at 1000 x g for 30 s, washed four times with cold buffer D 50 , and proteins eluted with 50 µl of 1.25 M NaCl, 0.25 M MgCl 2 .
Electrophoresis-SDS-PAGE was performed for 90 min at 100 V using a Hoefer minigel electrophoresis system (46). Protein bands were visualized by staining with either Coomassie Blue or silver (47) or immunoblotting.

RESULTS
Cleavage of eIF4G by L Protease -A previous study indicated that, in addition to the primary cleavage of rabbit eIF4G by L protease at Gly 479 -Arg 480 , there were cleavages at several other sites (12). To characterize these secondary cleavages further, we combined purified recombinant L protease with rabbit reticulocyte eIF4F and performed a time course of digestion ( Fig.  1). Proteolytic fragments were visualized by staining with silver (A) or immunoblotting with antibodies against different regions of eIF4G (B and C). Antibodies against peptides corresponding to amino acid residues 327-342, 403-416, and 1230-1248 were developed in an earlier study (43). However, the central portion of the eIF4G sequence was not represented by these antibody reagents. We therefore developed an additional antibody against residues 653-666. As observed previously (12), L protease initially makes a single cleavage, separating the N-terminal one-third of eIF4G (here designated cp N ) from the C-terminal two-thirds (designated cp c ; see Fig. 2). eIF4G migrates on SDS-PAGE as a series of heterogeneous bands with aberrantly slow mobility (Fig. 1, lane 1); the slowest species migrates at 220 kDa (6,25,48), in contrast to the 154 kDa calculated from the amino acid sequence (43). The cause of the heterogeneity and aberrant mobility is not known, but the region of eIF4G responsible for this behavior appears to be located in cp N (12,15). Like intact eIF4G, cp N migrated abnormally slowly as a heterogeneous cluster of bands, with an apparent molecular mass of 100-130 kDa, despite its true size Further digestion with L protease caused the cp N fragments to disappear and a heterogeneous cluster of fragments, designated cp N1 , to appear (A, lanes 4-7). All members of the cp N1 cluster arose simultaneously and therefore are likely to have resulted from cleavage of cp N at a single site. The heterogeneity of cp N1 is similar to that of cp N and eIF4G itself, suggesting that the source of heterogeneity in eIF4G is localized to the cp N1 region. Immunoblotting with an anti-eIF4G 327-342 antibody detected a band with estimated molecular mass of 28 kDa, designated cp N2 , which appeared simultaneously with the degradation of cp N (B, lanes 2-4). With further digestion, some of cp N2 was converted to smaller species termed cp N3 (B, lanes [4][5][6][7]. Immunoblotting with anti-eIF4G 327-342 antibodies (B) and anti-eIF4G 403-416 antibodies (data not shown) indicated that cp N2 and cp N3 contained amino acid residues 327-416, whereas cp N1 did not. This permitted the orientation of cp N1 , cp N2 , and cp N3 as shown in Fig. 2.
Further cleavage of the 103-kDa cp c yielded a number of smaller fragments (cp C1 -cp C5 ). The relative abundance and kinetics of appearance of the fragments together with immunoblotting with anti-eIF4G 653-666 (Fig. 1C) and anti-eIF4G 1230-1248 (12) antibodies is consistent with the following order of events. Initial cleavage of cp C by L protease can occur at one of two nearby sites (see Fig. 2). This yields two overlapping polypeptides, cp C1 (59 kDa) and cp C2 (55 kDa), both of which can be seen in A, lanes 4-7, and both of which react with anti-eIF4G 1230-1248 antibodies (12). Secondary cleavage of cp C1 generates cp C2 (note gradual loss of cp C1 and appearance of cp C2 in Fig. LA). When cp c is initially cleaved to cp C1 , the N-terminal half is represented by cp C4 (40 kDa), which reacts with anti-eIF4G 653-666 antibodies (C, lanes 3-7, and Fig. 2). When cp c is initially cleaved to cp C2 , the N-terminal half is represented by cp C3 (45 kDa), which also reacts with anti-eIF4G 653-666 antibodies (C, lanes 3-7). With longer times of incubation, cp C3 and cp C4 disappear simultaneous with the appearance of cp C5 (35 kDa; C, lanes 4-7). Neither cp C3 , cp C4 , nor cp C5 were recognized by anti-eIF4G 1230-1248 (data not shown). These results indicate that cp ci and cp C2 are derived from the C-terminal portion of cp c and that cp C3 , cp C4 , and cp C5 are derived from the N terminus (Fig. 2). Based on the SDS-PAGE-derived molecular masses, cp C3 and cp C2 approximately correspond to amino acid residues 480-886 and 887-1402 of rabbit eIF4G, respectively. eIF3 is a multisubunit initiation factor that associates with eIF4F (29). Previously we showed that eIF3 does not affect cleavage of eIF4G by 2A proteases from HRV2 and coxsackievirus serotype B4 (15). However, Wyckoff et al. (49,50) have presented evidence that eIF3 is required for cleavage of eIF4G by poliovirus 2A protease. To determine if eIF3 affected either the rate or sites of cleavage of eIF4G by FMDV L protease, we repeated the time course in the presence of this factor at a concentration equimolar to that of eIF4F (Fig. 1, lanes 8-14). No cleavage of eIF4E or the various subunits of eIF3 was observed. The rate of appearance of the eIF4G cleavage products was unchanged except for a slight heterogeneity of the cp C4 band; eIF3 caused a reduction in cp C4 and the appearance of a new band ~1 kDa smaller in size (cf. lanes 4-7 with lanes  11-14 in C). This suggests that eIF3 may block the accessibility of a protease site or expose a new one in this region of eIF4G. The N-terminal Region of eIF4G Contains the eIF4E-binding Site -As noted above, the eIF4F complex consists of eIF4A, eIF4E, and eIF4G. To determine which portions of eIF4G bind to eIF4E, we digested purified rabbit eIF4G with HRV2 protease 2A and subjected the reaction products to chromatography on m 7 GTP-Sepharose (Fig. 3). Analysis of column fractions by SDS-PAGE followed by either Coomassie Blue staining (A) or immunoblotting with an anti-eIF4G 327-342 antibody (B) or a monoclonal anti-eIF4A antibody (C) indicated that neither cp c nor eIF4A were retained on the column, but rather were found in the flow-through (FT) fractions (A and C, lanes 1-3). However, the cp N bands were retained on the column and co-eluted with eIF4E (A and B, lanes 6 and 7). This suggests that the cp N region of the protein contains the eIF4E-binding site.
A similar analysis of eIF4F treated with L protease was performed (Fig. 4). In this case, proteins in column fractions were resolved by SDS-PAGE and visualized by either silver staining (A) or immunoblotting with anti-eIF4G 1230-1248 (B) or anti-eIF4G 327-342 antibodies (C). (The fact that cp N stains much more strongly than cp c with silver has been documented previously (15), but the reason is not known; it may be due to groups of charged amino acid residues observed in the cp N portion of eIF4G (43) or posttranslational modifications which introduce hydrophylic or charged groups.) cp c , fragments of cp c , and eIF4A were not retained on the column but were found in the flow-through fractions (A and B, lanes 1-3). cp N1 was also detected in the flow-through (A, lanes 1-3). Most of the cp N2 , however, was retained on the column and co-eluted with eIF4E (A and C, lanes 6-8), indicating that the eIF4E-binding site is contained within this region of eIF4G. The fact that some of the cp N and cp N2 are found in the flow-through as well (lanes 1-3) may indicate dissociation from eIF4E during chromatography.
In order to define more precisely the region of eIF4G that binds to eIF4E, we determined the L protease cleavage site at the cp N1 /cp N2 junction. Fragments of eIF4G produced by a brief digestion with L protease were subjected to reverse phase HPLC on a C4 column (Fig. 5). The peaks in fractions 40-50 contained cp N2 and cp N3 and were well resolved from the peak in fractions 51-56, which contained cp N1 . Although cp N2 and cp N3 eluted in the same general region, it was possible to resolve the majority of the cp N2 (fractions 44-47) from the cp N3 (fractions [40][41][42][43]. cp N2 was subjected to N-terminal sequence analysis by automated Edman degradation (Table I). Alignment of the resultant sequence with the rabbit polypeptide sequence determined by cDNA cloning indicated that L protease cleaves at Lys 318 -Arg 319 , which corresponds to Lys 317 -Arg 318 in human eIF4G. This indicates that the eIF4E-binding site lies between amino acid residues 319 and 479.
Knowledge of this site allows a comparison of the true molecular masses of cp N1 and cp N2 to the SDS-PAGE-derived masses. The calculated molecular mass for cp N1 based on this cleavage site is 32.5 kDa, but the observed mobility on SDS-PAGE for cp N1 bands is 80-100 kDa. Also, cp N1 is heterogeneous, whereas cp N2 is not (Fig. 5B). cp N2 migrates as a 28-kDa polypeptide despite its true molecular mass of 18 kDa. This indicates that cp N1 is responsible for most of the heterogeneity and aberrant mobility of eIF4G.
Knowledge of the cp N1 /cp N2 junction also allows one to make conclusions about further cleavage products of cp N2 . The size difference between cp N2 and cp N3 is 2-4 kDa, but both are recognized by the anti-eIF4G 327-342 antibodies ( Figs. 1 and 5). This epitope is located only nine amino acid residues (1.2 kDa) from the N terminus of cp N2 . The fact that cp N3 is recognized by anti-eIF4G 327-342 antibodies suggests that it has the same N terminus as cp N2 and hence that the cleavage converting cp N2 to cp N3 is at its C terminus (see Fig. 2).
The C-terminal Region of eIF4G Contains the eIF4A-and eIF3-binding Sites -In intact eIF4F, the eIF4A subunit is retained on m 7 GTP-Sepharose even though only eIF4E has affinity for m 7 GTP (23)(24)(25)48). This has been interpreted to mean that both eIF4E and eIF4A have affinity for eIF4G, although another study has indicated that eIF4E and eIF4A can bind to each other directly (51). The results in Figs. 3 and 4 demonstrate that eIF4A is no longer retained on m 7 GTP-Sepharose after treatment with 2A or L protease. This could mean either that eIF4A has been released as a free protein by proteolytic digestion of eIF4G or that it is bound to a portion of eIF4G which is not retained on m 7 GTP-Sepharose. To distinguish between these possibilities, we took advantage of the fact that eIF3, an eight-subunit initiation factor with molecular mass of over 500 kDa (20), can be isolated in a complex with eIF4F and purified by m 7 GTP-Sepharose chromatography (45). Proteolytic fragments of eIF4G that remain associated with eIF3 should co-sediment with this factor in sucrose gradients, whereas nonassociated components should remain near the top of the gradient.
The eIF3·eIF4F complex was isolated, digested with either 2A or L protease, and the products subjected to sedimentation analysis (Fig. 6). The sedimentation of eIF3 was indicated by immunoreactivity of the 110-kDa eIF3η subunit. In the absence of protease, eIF4G and eIF3η sedimented at approximately 18 S with the eIF3·eIF4F complex (A, lanes 6-9). The majority of the eIF4A (70%) co-sedimented with this complex as well, although some apparently dissociated during sedimentation. Cleavage of eIF4G disrupted the eIF3·eIF4F complex. When eIF4G was cleaved by 2A protease, the two fragments of eIF4G did not co-sediment. Rather, all of the cp N sedimented near the top of the gradient (B, lanes 1-2), whereas all of the cp C and most (75%) of the eIF4A co-sedimented with eIF3 (lanes [6][7][8][9]. This indicates that cp C and eIF4A remain associated with eIF3 after 2A cleavage of eIF4G, but cp N does not. Limited digestion of eIF4G by L protease disrupted its interaction with both eIF4A and eIF3 (C). Whereas all remaining intact cp C and some eIF4A still co-sedimented with eIF3 (lanes [5][6][7][8], most of the eIF4A (80%), as well as all of the cp C1 and cp C2 , shifted to the top of the gradient (lanes 1-4). The proportion of eIF4A at the top of the gradient was approximately equal to the proportion of cp C cleaved. Longer digestion with L protease resulted in complete cleavage of cp C and all of the eIF4A shifting to the top of the gradient (data not shown). The cp C3 fragment co-sedimented with eIF3 (lanes 6 and 7), whereas cp C4 and cp C5 were released and sedimented near the top of the gradient (data not shown). These results suggest that the eIF3-binding site is located within cp C3 .
Since eIF4A co-sedimented with cp C , the eIF4A binding site is likely to be contained within this region. The fact that eIF4A shifts to the top of the gradient when cp C is further digested with L protease (C) could mean that either eIF4A is free or is bound to a fragment of eIF4G which has shifted to the top of the gradient, e.g. cp C1 or cp C2 . To determine whether eIF4A has affinity for cp C1 or cp C2 , we covalently attached purified anti-eIF4G 1230-1248 antibody to agarose beads and used this immunoaffinity resin to test for physical association between eIF4A and fragments of eIF4G. L protease was incubated with either eIF4A alone (Fig. 7, lane 1) or eIF4A mixed with eIF4F (lanes  2 and 3), the reactions subjected to immunoadsorption, and the bound fractions analyzed (lanes [4][5][6]. (eIF4A was added to the eIF4F preparation because eIF4F prepared by the method of Lamphear and Panniers (25) is substoichiometric for eIF4A.) The immobilized antibody bound cp C , cp C1 , and cp C2 as expected (A, cf. lanes 2 and 3 with lanes 5 and 6); the strong band  6. Ultracentrifugal fractionation of the protease-treated eIF3·eIF4F complex. The eIF3·eIF4F complex was incubated in the absence (A) or presence of the 2A (B) or L (C) proteases, layered onto 15-30% sucrose gradients, and centrifuged as described under "Experimental Procedures." Gradients were fractionated and aliquots subjected to SDS-PAGE on 8.5% gels. 3n, immunoblot probed with anti-eIF3 antibodies; 4A, immunoblot probed with anti-eIF4A antibodies; 4G, cp c , cp c1 , and cp C2 , immunoblot probed with anti-eIF4G 1230-1248 antibodies; cp N , immunoblot probed with anti-eIF4G 327-342 antibodies; cp C3 , immunoblot probed with anti-eIF4G 653-666 antibodies. Lane S represents the starting material which was layered onto each gradient. Fraction 1 corresponds to the top of the gradient. below cp C2 migrating at 50 kDa is the highly immunoreactive IgG heavy chain, which probably results from insufficient removal of the immunoaffinity resin from eluted samples). eIF4A alone was not bound to the immunoaffinity resin (B, lane 4) unless eIF4G fragments were present (B, lanes 5 and 6). In a similar experiment in which the digestion with L protease was continued until all of the cp c1 was converted to cp C2 , eIF4A was also retained on the immunoaffinity resin (data not shown), indicating that the ~4 kDa which is removed from cp c1 by the secondary L protease cleavage does not contain the eIF4A binding site.
As a control, the anti-eIF4G 653-666 antibody was used to test whether cp C3 , cp C4 , or cp c5 were bound to the immunoaffinity resin. If this were the case, e.g. that cp C3 remained bound to cp C2 after proteolytic cleavage of cp c , it would not be possible to distinguish whether eIF4A were bound to cp C2 or cp C3 . The experiment indicated that cp c3 , cp C4 , or cp C5 , though detectable in the starting material (Fig. 7C, lanes 2 and 3), was not retained by the resin (lane 5). (Contamination by the IgG heavy   FIG. 7. Immunoadsorption of L protease-treated eIF4F with immobilized anti-eIF4G 1230-1248 antibodies. eIF4A (2 µg, lanes 1 and 4) or eIF4A plus eIF4F (2 µg each, lanes 2, 3, 5, and 6) were incubated with L protease in 50 µ1 of buffer D 50 . Aliquots (5 µl) were removed for gel analysis (lanes 1-3) and the remainder subjected to immunoadsorption on an anti-eIF4G 1230-1248 resin as described under "Experimental Procedures." The eluate (bound fraction) was then analyzed by SDS-PAGE on 8.5% gels (lanes 4-6). A, 12 µg of protease were used and incubation was for either 30 min (lanes 1, 2, 4, and 5) or 90 min (lanes 3 and 6). The immunoblot was probed with anti-eIF4G 1230-1248 antibodies. The band migrating at ~50 kDa in lanes 5 and 6 is rabbit IgG heavy chain, which contaminates some of the eluates. B, same as A except the immunoblot was probed with anti-eIF4A antibodies. In this case the IgG heavy chain does not react, since the secondary antibody used was goat anti-mouse. C, same as A except incubation with protease was for 30 min, 3 µg of protease were used in lanes 1, 2, 4, and 5, and 6 µg were used in lanes 3 and 6. Also, the immunoblot was probed with anti-eIF4G 653-666 antibodies. chain did not permit analysis of cp C3 in lane 6, but cp C4 or cp C5 were clearly absent.) cp c , on the other hand, which contains the epitope for both anti-eIF4G 1230-1248 and anti-eIF4G 653-666 antibodies, was detected in the bound fractions (Fig. 7C, lanes 5  and 6). Thus, the immobilized anti-eIF4G 1230-1248 antibody selectively adsorbs cp c1 and cp C2 , but not cp C3 , cp C4 , or cp C5 . In a similar experiment using anti-eIF4G 327-342 antibodies, it was determined that cp N2 did not bind to the immunoaffinity resin (data not shown). In summary, these results indicate that there is a binding site for eIF4A in the cp C2 fragment of eIF4G.

DISCUSSION
Previous studies have established that eIF4G can form complexes with initiation factor polypeptides whose activities are associated with each of the events necessary for recruitment of host cell mRNA for translation. These events are cap recognition, a property of eIF4E (reviewed in Ref. 52); ATP-dependent unwinding of mRNA secondary structure, which is a property of eIF4A alone but one which is greatly enhanced by the pres-ence of eIF4B and eIF4G (53,54); and ribosome binding, which requires eIF3 (55)(56)(57). Purification of a high salt ribosomal wash by gel filtration (23) or m 7 GTP-Sepharose chroniatography (25) leads to co-elution of eIF4A, eIF4B, eIF4E, eIF4G, and eIF3 in a large macromolecular complex. The present study indicates that treatment with proteases which have high specificity for eIF4G separates these major functional activities into two complexes, one containing the cap recognition function and one containing the helicase and ribosome binding functions (Figs. 3, 4, 6, and 7). In vivo, the 2A and L proteases appear to carry out only the cleavage event which generates cp c and cp N (12). Therefore, the activities of these fragments are most relevant to viral replication. However, further proteolysis by L protease in vitro, again specific for eIF4G, results in the separation of the ribosome binding function (eIF3) and the helicase function (eIF4A; Fig. 6). This suggests that these three functional activities of initiation factors involved in mRNA recruitment (cap binding, RNA helicase, and ribosome binding) are brought together principally through an interaction with eIF4G rather than with each other. cp N is the region of eIF4G responsible for binding eIF4E, representing amino acid residues 1-486 of rabbit eIF4G in the case of 2A protease cleavage (1-479 in the case of L protease). Finer mapping with L protease narrows this region to amino acids 319-479. The criterion in either case is retention on m 7 GTP-Sepharose. Previously it was reported that poliovirusinduced cleavage products of eIF4G, as detected by a monoclonal antibody, were retained on cap-analog columns, but the location of the epitope was unknown (27)(28)(29). With the availability of antibodies to defined regions of eIF4G, it is now possible to conclude that the monoclonal antibody used earlier recognizes the same cleavage products as antipeptide antibodies against amino acid residues 327-416 (43). The validity of m 7 GTP-Sepharose retention as a basis for assigning the eIF4Ebinding region to cp N is reinforced by the observation that cp N in the absence of eIF4E is not retained on m 7 GTP-Sepharose (28). cp C , representing amino acids 480-1402, is the region of eIF4G which binds to eIF4A and eIF3, as demonstrated by co-sedimentation (Fig. 6) and immunoadsorption (Fig. 7). cp C3 contains the eIF3-binding site, based primarily on co-sedimentation, but also supported by the observation that eIF3 alters cleavage within cp C3 (Fig. 1C). cp C2 contains the eIF4A binding site, since both eIF4A and cp C2 are released from eIF3 by L protease and co-sediment (Fig. 6C). Also, eIF4A specifically binds to the anti-eIF4G 1230-1248 affinity resin, but only in the presence of cp C2 or larger fragments containing cp C2 (Fig. 7B).
Putting these two sets of data together suggests a domain model for eIF4G wherein the cap recognition function is in the N-terminal one-third of the molecule and the unwinding and ribosome binding functions are in the C-terminal two-thirds (Fig. 8). The region separating the N-and C-terminal domains may be a flexible hinge or loop which is more exposed to proteases. In support of this prediction, analysis of secondary structure motifs suggests that this region has a high probability of β-turns (58). Furthermore, a synthetic peptide based on a different portion of the eIF4G sequence is cleaved by HRV2 protease 2A (17), but this site is not cleaved in intact eIF4G (15), suggesting that it is in a region of the molecule which is not as accessible as the putative hinge region. Finally, fragments of eIF4G similar in size to cp N are observed upon raising Ca 2+ levels in extracts of uninfected cells, suggesting that the junction between cp N and cp c is more susceptible to normal intracellular proteases (59).
Is such a domain model compatible with current ideas about cap-dependent initiation? Two models for the mRNA-binding Only one of the current models for cap-dependent initiation, the stepwise assembly model, is depicted (see text). eIF4E binds to the mRNA cap, whereas eIF4G binds to eIF3 on the 43 S initiation complex. The protein-protein interaction between eIF4E and eIF4G brings the mRNA to the ribosome and the unwinding machinery, represented by eIF4A. The wavy portion of mRNA indicates secondary structure. The ability of eIF4G to bind eIF4A, eIF3, and eIF4E is critical for mediating cap-dependent joining of mRNA to the ribosome and positioning of the RNA helicase at the 5'-end of the mRNA. B, cap-independent initiation. During replication of rhino-, entero-, and aphthoviruses, cleavage of eIF4G separates the eIF4Ebinding domain (cp N ) from the eIF3-and eIF4A-binding domain (cp C ). This produces an altered 43 S initiation complex (*), disrupting the ability of eIF4G to mediate cap-dependent recruitment of RNA to the ribosome but not altering the binding of cp C to the ribosome or helicase activity required for internal initiation. step of initiation have been proposed, here referred to as the stepwise assembly model and the preformed complex model. Recruitment of capped mRNAs requires the coupling of cap recognition, unwinding, and scanning. The models differ primarily in which of these functions occur on the ribosome. The division of eIF4G into functional domains, however, is compatible with either model. The stepwise assembly model states that eIF4E first recognizes the cap as a free polypeptide, i.e. not in the eIF4F complex, whereas eIF4G is present in the 43 S initiation complex prior to mRNA recruitment, bound to eIF3 (Ref. 60; see Fig. 8). eIF4A, although largely found in the postribosomal supernatant, may also be bound to eIF4G and hence also be present in the 43 S initiation complex. The joining of eIF4E to eIF4G brings mRNA to the 40 S ribosomal subunit and to the unwinding machinery. In this model, the cp C region would serve to anchor eIF4G to the ribosome via eIF3 and provide a binding site for eIF4A (and possibly eIF4B as well), whereas the cp N domain would provide a flexible arm to receive the incoming eIF4E:mRNA complex (Fig. 8A). Cleavage of eIF4G by viral proteases would permit eIF4E to bind the cap and the eIF4E mRNA complex to bind cp N (Fig. 8B, upper portion), but these events would occur apart from the unwinding machinery and the 40 S ribosomal subunit. Hence, no recruitment of capped mRNA to the ribosome would result. The preformed complex model states that eIF4F exists prior to formation of any complexes between mRNA, initiation factors, or ribosomes (3,21,22,61). This eIF4F complex recognizes the cap of free mRNA and unwinds its secondary structure. When a sufficiently long stretch of secondary structure is unwound, the 40 S ribosomal subunit binds and begins scanning. A domain structure for eIF4G as proposed here would suggest that eIF4E, bound to the cp N2 region of eIF4G, first recognizes the cap and directs the mRNA to the unwinding machinery which is bound to the cp C region. Cleavage of eIF4G by viral proteases would yield separate complexes capable of cap recognition and unwinding, but since unwinding would not be directed to the 5'-end of the mRNA, ribosome binding to single-stranded RNA would be frequently unproductive.
Is such a domain model for eIF4G compatible with current ideas about cap-independent initiation? Translation of picornaviral RNA requires eIF4A, and inactive variants of eIF4A have a dominant negative effect on translation of both capped and picornaviral mRNAs (36). This inhibition can be overcome by the addition of normal eIF4A, but the addition of eIF4F is six times more effective, underscoring the importance of eIF4G for unwinding. Binding of picornaviral RNA to the ribosome requires an internal ribosome entry site sequence (reviewed in Ref. 62). As eIF3 is involved in mRNA binding to ribosomes (see above), it must be presumed that translation of picornaviral RNA requires eIF3. In summary, translation of picornaviral RNA requires factors involved in unwinding (eIF4A, eIF4B) and ribosome binding (eIF3) but not cap recognition (eIF4E). Cleavage of eIF4G by picornaviral proteases separates the functions required for cap-independent translation (those attached to cp C ) from those which are not (those attached to cp N ; see Fig. 8B). It is interesting in this light that the primary cleavage sites in eIF4G for 2A protease (15) and L protease (12) are only seven amino acid residues apart, despite the fact that the proteases are structurally distinct and that the amino acid sequences surrounding the cleavage sites bear little similarity. Thus, there has been evolutionary conservation among the entero-, rhino-, and aphthoviruses of function (separation of cp N from cp C ) rather than recognition of a specific amino acid sequence.
cp C is thus likely to contain the functions required for capindependent initiation. A binding site for eIF3 in cp C permits attachment to the ribosome. Consistent with this, we have observed that cp C sediments with ribosomes in a micrococcal nuclease-treated rabbit reticulocyte lysate (data not shown). A binding site for eIF4A in cp C suggests that unwinding functions can take place in the absence of cp N . The unwinding activity eIF4A is greatly enhanced when it is present in the eIF4F complex (53,54); based on the location of the eIF4A binding site, it is likely that this stimulatory activity of eIF4G resides in cp C . Also, there is a proposed RNA-binding region in eIF4G which is within cp C (63), and this region may be important for recruitment of mRNAs to the ribosome or stabilization of mRNA-ribosome interactions during translation of cellular and viral mRNAs. The proposed requirement of cp C for internal initiation may explain a paradoxical observation: picornavirus infection "inactivates" eIF4F (6, 64), yet eIF4F has been shown to stimulate internal cap-independent initiation from viral sequences (34,35,65). The proposed domain model would state that the added eIF4F is supplying functions contained in the cp C region which are needed for cap-independent translation; eIF4F is "inactivated" by 2A or L protease only with respect to cap-dependent translation. Based on this consideration, one would predict that cp C alone would stimulate cap-independent initiation through its eIF3, eIF4A, and possibly RNA binding properties.