Characterization of the two eIF4A-binding sites on human eIF4G-1.

Eukaryotic translation initiation factor 4G-1 (eIF4G) plays a critical role in the recruitment of mRNA to the 43 S preinitiation complex. eIF4G has two binding sites for the RNA helicase eIF4A, one in the central domain and one in the COOH-terminal domain. Recombinant eIF4G fragments that contained each of these sites separately bound eIF4A with a 1:1 stoichiometry, but fragments containing both sites bound eIF4A with a 1:2 stoichiometry. eIF3 did not interfere with eIF4A binding to the central site. Interestingly, at the same concentration of free eIF4A, more eIF4A was bound to an eIF4G fragment containing both eIF4A sites than the sum of binding to fragments containing the single sites, indicating cooperative binding. Binding of eIF4A to an immobilized fragment of eIF4G containing the COOH-terminal site was competed by a soluble eIF4G fragment containing the central site, indicating that a single eIF4A molecule cannot bind simultaneously to both sites. The association rate constant, dissociation rate constant, and dissociation equilibrium constant for each site were determined by surface plasmon resonance and found to be, respectively, 1.2 x 10(5) m(-1) s(-1), 2.1 x 10(-3) s(-1), and 17 nm for the central site and 5.1 x 10(3) m(-1) s(-1), 1.7 x 10(-3) s(-1), and 330 nm for the COOH-terminal site.

The initiation of translation of most eukaryotic mRNAs involves the sequential recruitment of Met-tRNA i , mRNA, and the 60 S ribosomal subunit to the 40 S ribosomal subunit, catalyzed by the various groups of initiation factors (1). One of the most highly regulated steps is the recruitment of mRNA, which requires recognition of the 5Ј-terminal m 7 GTP-containing cap and 3Ј-terminal poly(A) tract, unwinding of 5Ј-terminal secondary structure, and binding to the 43 S initiation complex. These steps are mediated by members of the eIF4 1 group of initiation factors (eIF4A, eIF4B, eIF4E, and eIF4G) as well as poly(A)-binding protein.
eIF4A is the prototypical member of the DEXD/H-box protein family of nucleic acid helicases (2,3). It functions as an ATPdependent, bi-directional RNA helicase and RNA-dependent ATPase (4 -7). The ␥-phosphate on the bound nucleotide has been shown to mediate changes in eIF4A conformation and RNA affinity. ATP binding and hydrolysis produce conformational changes in eIF4A that alter the RNA-protein interactions (8,9,3). A current model for protein synthesis initiation envisions eIF4A in the role of unwinding mRNA secondary structure in the 5Ј-untranslated region to allow the 40 S ribosomal subunit to bind the mRNA and/or scan it for the first AUG. The observation that dominant negative variants of eIF4A inhibit translation is consistent with such a role (10). Interestingly, the RNA helicase activity of eIF4A is ϳ20 times greater when bound to eIF4G than as a free protein (6,11).
Initial studies indicated that the NH 2 -terminal one-third of mammalian eIF4G binds poly(A)-binding protein (13) and eIF4E (20,21), the central one-third binds eIF3 (20), and the COOH-terminal one-third binds eIF4A (20) and Mnk1 (23,24). These portions of eIF4G correspond with cleavage products of picornaviral proteases (cp N (aa 1-634), cp C3 (aa 635-1041), and cp C2 (aa 1042-1560), respectively; Ref. 20), suggesting that they represent individual structural and functional domains of the eIF4G molecule (see Fig. 1). The central domain has high amino acid sequence homology to all eIF4G proteins, whereas the COOH-terminal domain is poorly conserved (25). Additional studies revealed a second eIF4A-binding site in the central domain (22). The central domain of eIF4G is sufficient to catalyze cap-independent binding of ribosomes to RNA (17) and cap-independent but 5Ј-end-dependent translation in vitro (26) and in vivo (27). The function of the COOH-terminal domain is less clear. Variant forms of eIF4G containing both of the intact eIF4A-binding sites are more active in formation of the 48 S initiation complex and stimulation of cap-dependent translation than forms containing only one site (28). This has led to the proposal that interaction of eIF4A with the central domain is necessary for translation, whereas interaction of eIF4A with the COOH-terminal domain plays a modulatory role (28).
A complex of initiation factors termed eIF4F can be isolated by treatment of the 100,000 ϫ g ribosomal pellet with 0.5 M KCl (29,30). The eIF4F complex from mammalian cells contains eIF4A, eIF4E, and eIF4G (29,30), but the complex from wheat germ (31,32), yeast (33,34), and Drosophila (35) contains only eIF4E and eIF4G. Even in mammalian eIF4F preparations, the amount of eIF4A varies, and passage of eIF4F over a phosphocellulose column completely removes eIF4A (11,36). The stoichiometries of the mammalian eIF4F subunits have not been reported. The discovery of a second eIF4A-binding site in eIF4G (22) raises the question of whether two eIF4A molecules can bind simultaneously or whether a single eIF4A binds to both sites, as suggested recently (28). Similarly, the affinities of eIF4A to the two sites in mammalian eIF4G have not been reported. Since binding of eIF4A to eIF4G serves not only to localize eIF4A in the vicinity of the ribosome and mRNA but also enhances its intrinsic helicase activity, we undertook an analysis of the two eIF4A-binding sites in human eIF4G.
Amino Acid and Nucleotide Nomenclature-Several cDNAs for eIF4G have been described that differ in the predicted length of the NH 2 -terminal polypeptide sequences (12,13,37). The aa and nt numbers used in the current work correspond to the larger form of eIF4G that begins with MNTPSQ (13). In this numbering system, the primary entero-and rhinoviral 2A protease cleavage site (38) is located between Arg-641 and Gly-642 (nt 1823-1828) (see Fig. 1). Fragments of eIF4G expressed from recombinant plasmids are named using inclusive aa numbers. Thus, S-eIF4G(877-1078) would contain the eIF4G sequence from aa 877 to 1078 fused at the NH 2 terminus to thioredoxin, a His 6 sequence, and the S-peptide of RNase A (which increases the molecular mass by ϳ20 kDa) (see Fig. 1). The plasmid expressing this protein would be called pTS4G(877-1078).
In the case of S-eIF4G(613-1078), it was necessary to perform an additional purification step on heparin-Sepharose CL-6B. The eluate from nickel nitrilotriacetic acid-agarose was diluted with an equal volume of buffer B 0 (20 mM Tris-HCl, 2 mM EDTA, 5 mM ␤-mercaptoethanol, 0.1% (v/v) Tween 20, and 5% (v/v) glycerol, pH 7.5) to reduce the salt concentration and then applied to a heparin-Sepharose CL-6B column equilibrated with the same buffer except containing 100 mM KCl (buffer B 100 ). The protein was eluted with buffer B 400 . eIF4G(642-1560) (used as a standard) and eIF4G(642-1078) were produced by incubation of S-eIF4G(613-1560) and S-eIF4G(613-1078), respectively, with recombinant Coxsackievirus 2A protease at 50 g/ml for 1 h at 4°C (41). In the case of eIF4G(642-1560), the resultant COOH-terminal portion was further purified from the S-peptide-tagged NH 2 -terminal portion by adsorption of the latter to S-protein-agarose.
eIF3, eIF4A, and eIF4F Purification-Purification and 14 C-labeling of eIF4A by reductive methylation was performed as described previously (42). The particular batch of [ 14 C]eIF4A used in these experiments was labeled to a specific activity of 187 cpm/pmol. eIF3, eIF4A, and eIF4F were purified from the ribosomal high salt wash of rabbit reticulocyte lysate by m 7 GTP-Sepharose and Mono Q chromatography (43). The eIF4A peak was rechromatographed on Mono Q with a shallower salt gradient. The eIF3 peak from the initial Mono Q chromatography was further purified by gel filtration on a SW300 column (Waters, Milford, MA) in buffer C 150 plus 5% (v/v) glycerol.
Protein Binding Assays on S-protein-agarose-Binding of S-eIF4G fragments with eIF3 and eIF4A was performed using S-protein-agarose. After 40 min of preincubation of S-eIF4G fragments with eIF3 or eIF4A on ice, proteins were mixed with at least a 10-fold molar excess of S-protein-agarose and incubated for 2 h in buffer C 150 containing 1% milk proteins at 4°C. The resin was washed 4 times with 200-l aliquots of buffer C 150 . Proteins were eluted in SDS-electrophoresis buffer and analyzed by SDS-PAGE (44), with detection by Coomassie Blue staining, Western blotting, or autoradiography.
Western Blotting-For immunoblotting, proteins were transferred after SDS-PAGE to an Immobilon-P membrane (Millipore, Bedford, MA) using a Mini Trans-Blot cell (Bio-Rad). The membrane was incubated with a 1:1000 dilution of the primary antibodies in 5% milk proteins in buffer E (20 mM Tris-HCl, 150 mM NaCl, and 0.05% Tween 20, pH 7.5) for 1 h at room temperature, washed three times for 10 min with buffer E, and incubated with secondary anti-mouse or anti-rabbit antibodies conjugated with alkaline phosphatase (Vector Laboratories, Inc., Burlingame, CA) at a dilution of 1:2000 in 5% milk proteins in buffer E for 1 h at room temperature. Blots were developed with the nitro blue tetrazolium and 5-bromo-4-chloro-3-indolyl-phosphate color development substrate (Promega, Madison, WI).
Quantitation of Binding Data-Quantitation of eIF4G fragments and eIF4A separated by SDS-PAGE was performed using a ScanMaker III laser densitometer (Microtek) and ImageQuant software, Version 3.3 (Molecular Dynamics). Experimental data were compared with standard curves consisting of purified S-eIF4G fragments and eIF4A for which the concentration has been determined with the Bio-Rad protein assay kit. Curve fitting was performed using SigmaPlot software, Version 4.01 (SPSS, Inc.). In cases of eIF4A binding with S-eIF4G(613-1078) or S-eIF4G(1078 -1560), the data were fit with an equation describing the Langmuir isotherm, where BR is the binding ratio, i.e. the molar ratio of bound eIF4A to the eIF4G fragment, n is number of eIF4A-binding sites on the eIF4G fragment, [eIF4A] f is the concentration of eIF4A not bound to the resin, and K d is the dissociation equilibrium constant for the eIF4A⅐eIF4G complex.
[eIF4A] f was calculated as the difference between total eIF4A and bound eIF4A. A nonlinear least squares fit was performed in which n and K d were allowed to vary. An equation that takes into account two dissimilar binding sites (45) was used in titrations involving S-eIF4G(613-1560), where n 1 and K d1 represent the number of binding sites and dissociation equilibrium constant for the first binding site, respectively, and n 2 and K d2 represent the number of binding sites and dissociation equilibrium constant for the second.
SPR Analysis of S-eIF4G⅐eIF4A Interactions-SPR was carried out using a BIAcore 2000 instrument (Biacore, Inc., Piscataway, NJ). eIF4A was immobilized on a research grade CM5 sensor chip using the amino coupling kit supplied by the manufacturer in 10 mM sodium acetate, pH 4.5. The surface density of immobilized eIF4A was 200 -300 RU. One RU corresponds to an immobilized protein density of 1 pg/mm 2 (46). The portion of the sensor chip in the first flow cell, used as a control, was subjected to activation and blocking in the same way as the eIF4Acontaining cells but without added protein. The signals generated in the control flow cell were subtracted from the experimental signals to correct for refractive index changes and nonspecific binding.
All kinetic experiments were carried out in buffer D at 25°C and a flow rate of 50 l/min for S-eIF4G(613-1078) and S-eIF4G(1078 -1560) and 20 l/min for S-eIF4G(975-1078). At least five different concentrations of each S-eIF4G fragment were injected for each experiment. The first injection contained buffer without the S-eIF4G fragment. Between injections, the surface was regenerated with buffer F (20 mM HEPES-KOH, 500 mM KCl, 3 mM EDTA, 0.1% Tween 20, and 2 mM ␤-mercaptoethanol, pH 7.5) at a flow rate of 70 l/min and contact time of 2 min, followed by buffer D for 1 min.
Kinetic and equilibrium constants were calculated using the curvefitting facility of the BiaEvaluation software, Version 3 (Biacore, Inc.). Binding data were globally fit to the 1:1 Langmuir binding model (AϩB p AB). For this model, the function that describes the response (R t ) as a function of time (t) during the injection phase is, where R eq is the response at equilibrium, C is the molar concentration of S-eIF4G fragment, t 0 is time at the start of injection, and k a and k d are association and dissociation rate constants, respectively. For the post-injection phase, the response is, where R 0 and t 0 are the values of R t and t, respectively, when sample is replaced with buffer, and k d is the dissociation rate constant. Values for the statistical closeness of fit, 2 , were always below 1, indicating that the simple 1:1 model of interaction correctly described the experimental data.

RESULTS
Definition of eIF4A-binding Sites on eIF4G-Two eIF4A binding domains have been described for mammalian eIF4G, one in the central domain (22) and the other in the COOHterminal domain (20) (Fig. 1). Evidence has recently been presented that the central eIF4A-binding site is located between aa 672 and 970 (28). To study these two binding sites sepa-rately, we prepared various recombinant portions of eIF4G fused to an NH 2 -terminal tag containing the S-peptide of RNase A (Fig. 1). [ 14 C]eIF4A was incubated with each of these S-tagged eIF4G fragments, and the resultant complex was captured on S-protein-agarose. Material bound to the resin was eluted, subjected to SDS-PAGE, and analyzed for the eIF4G fragment by Coomassie Blue staining ( Fig. 2A) and for eIF4A by autoradiography (Fig. 2B). S-eIF4G(613-1078) and S-eIF4G(1078 -1560) were observed to bind eIF4A (lanes 1 and 2), but the other S-eIF4G fragments did not (lanes 3-5).
Since both eIF4A and eIF4G are capable of binding RNA, there was a possibility that the observed binding was due to an RNA linker. Therefore, the experiment was repeated after treatment of the proteins with micrococcal nuclease to remove any E. coli RNA that may have copurified with the recombinant proteins. The results were not altered from those seen in Fig. 2 (data not shown), indicating that the binding occurs in an RNA-independent manner. eIF4A and eIF3 Do Not Compete for Binding to the Central Domain of eIF4G-The central domain binds eIF4A as well as eIF3 (20,22). To test whether this domain can bind eIF4A and eIF3 simultaneously, we performed a competition experiment. Reaction mixtures contained S-eIF4G(613-1078), a 5-fold molar ratio of [ 14 C]eIF4A to S-eIF4G(613-1078), and a 2-, 5-, or 10-fold molar ratio of eIF3 to S-eIF4G(613-1078) (Fig. 3, lanes  2, 3, and 4, respectively). After incubation, the S-eIF4G(613-1078) and any bound proteins were captured with S-proteinagarose. The results indicated that eIF3 was retained on the S-protein-agarose (Fig. 3A, lanes 2-4). Furthermore, there was no decrease in eIF4A binding when increasing amounts of eIF3 were added, whether measured by Coomassie Blue staining (Fig. 3A, lanes 2-4) or autoradiography (Fig. 3B, lanes 2-4). There was no nonspecific binding of eIF4A or eIF3 to the resin in the absence of the eIF4G fragment (Fig. 3, lane 8).
eIF4G(613-1078) and eIF4G(1078 -1560) Each Bind eIF4A with a Stoichiometry of 1:1 but with Different Affinities-To estimate the stoichiometry and equilibrium binding constants of eIF4A with S-tagged eIF4G fragments, we carried out titration experiments using S-protein-agarose. S-eIF4G(613-1078) was incubated at a concentration of 0.3 M with either no eIF4A or with a 0.2-5-fold molar ratio of eIF4A. After adsorption to S-protein-agarose, the material bound to the resin was eluted and analyzed by SDS-PAGE, Coomassie Blue staining, and autoradiography (Fig. 4A). One reaction mixture containing eIF4A but no eIF4G fragment indicated that nonspecific binding of eIF4A was negligible (lane 12). The binding data were quantitated using standard curves, and a nonlinear least squares fit of the experimental data was performed using Equation 1 (Fig. 4B). The results indicated that the number of eIF4A-binding sites on S-eIF4G(613-1078), n, was 1.1 and that the equilibrium dissociation constant, K d , was 96 nM (Fig. 4B). A linear transform of the data indicates the presence of a single binding site (Fig. 4B, inset).
A similar experiment was performed with S-eIF4G(1078 -1560). The S-eIF4G(1078 -1560) was incubated at a concentration of 0.6 M with a 0.5-10-fold molar ratio of eIF4A to S-eIF4G(1078 -1560) (Fig. 4C). The data were quantitated as described above. The nonlinear least-squares fit yielded n ϭ 1.1 and K d ϭ 660 nM, indicating that eIF4A binds S-eIF4G(1078 -1560) with a stoichiometry of 1:1 (Fig. 4D). Scatchard analysis confirmed the existence of a single binding site on S-eIF4G(1078 -1560) (Fig. 4D, inset). These and several other similar experiments indicated that, within experimental error, the stoichiometry of eIF4A binding to both the central and COOH-terminal sites was 1:1.

Stoichiometry of Binding of eIF4A with eIF4G(613-1560)-
The forgoing result indicated that one molecule of eIF4A could bind to either the central or COOH-terminal sites of eIF4G. The question then arose as to whether binding to these sites is mutually exclusive. We therefore titrated an eIF4G fragment containing both the central and COOH-terminal sites [S-eIF4G(613-1560)] using the same methodology. S-eIF4G(613-1560) was incubated at a concentration of 0.075 M with eIF4A at a 1-16-fold molar ratio (Fig. 4E). Material bound to the resin was analyzed by SDS-PAGE.
Unique values of n and K d for each site cannot be determined from empirical saturation data when there are two dissimilar binding sites on the same molecule (45). We therefore computed a theoretical curve from Equation 2 using the n and K d values for S-eIF4G(613-1078) (1.1 and 96 nM, respectively) and S-eIF4G(1078 -1560) (1.1 and 660 nM, respectively) that were determined experimentally in Figs. 4, B and D. The theoretical curve was similar to the experimental data, but several points fell above the line (see below). In addition, the maximum eIF4A binding to S-eIF4G(613-1560) approached a stoichiometry of 2:1 (Fig. 4F). Binding of eIF4A to the S-eIF4G(543-1560) fragment was also performed. In two separate experiments, the binding of eIF4A to that fragment also approached a 2:1 stoichiometry (data not shown).
As expected, plotting the theoretical curve calculated from Equation 2 in the manner used to produce a linear transform of the data of Figs. 4, B and D, resulted in a curved line (Fig. 4F,  inset). The fact that several of the points fall above the theoretical curve means that the affinity of eIF4A for an eIF4G fragment containing both sites is higher than predicted by Equation. 2, which merely sums the individual contributions of the two separate sites. This suggests that there is cooperativity

eIF4G-1 Interactions with eIF4A
between sites, i.e. that the binding of eIF4A to one site increases the affinity to the other. Further support for cooperative binding is provided in the following experiment.
The Presence of Two eIF4A-binding Sites Enhances the Affinity of eIF4A for eIF4G-Cooperativity between sites would predict that an eIF4G fragment containing both sites would bind more eIF4A than the sum of binding to fragments containing each separate site. To test this, we individually incubated S-eIF4G(613-1078) (central site), S-eIF4G(1078 -1560) (COOH-terminal site), and S-eIF4G(613-1560) (both sites), each at 0.1 M, with varying concentrations of eIF4A. The eIF4G fragments and bound eIF4A were then captured with S-protein-agarose and analyzed by Western blotting (Fig. 5). eIF4A incubated with a protein consisting of all elements at the NH 2 terminus of the S-eIF4G fusion proteins (thioredoxin, His 6 -tag, and S-tag) but no eIF4G sequences was not retained on the resin (data not shown). At 0.39 M eIF4A, the BR of eIF4A to S-eIF4G(1078 -1560) was only 0.02, whereas the BR to S-eIF4G(613-1078) was 0.65, which is expected from the higher affinity of the latter. The predicted BR for S-eIF4G(613-1560), if there is no cooperativity between sites, is the sum of these, or 0.67. The observed BR, however, was 1.35, representing a 2-fold enhancement. At 0.65 M eIF4A, the cooperative effect was not as high, which is to be expected since the sites were more nearly saturated. The sum of the BR values for the individual sites was 0.68, but the observed BR for S-eIF4G(613-1560) was 1.16, a 1.7-fold enhancement. Finally, at the highest eIF4A concentration, 1.3 M, where both eIF4Abinding sites were nearly saturated, the enhancement was only 1.2-fold. Thus, binding of eIF4A to an eIF4G fragment that contains both eIF4A-binding sites is more than additive. These observations support the idea of positive cooperativity between the two sites.
Kinetic Measurements of eIF4A-eIF4G Interactions Using SPR-The assessment of binding properties by retention of eIF4A⅐S-eIF4G complexes on S-protein-agarose represents a nonequilibrium situation, because the agarose beads must be washed several times with buffer not containing eIF4A to reduce nonspecific binding. It is suitable for determination of stoichiometries because a theoretical maximal binding is approached with increasing eIF4A concentration. However, the technique produces a systematic overestimate of K d values (underestimate of binding affinities) because the average eIF4A concentration during binding and washing is less than the concentration during binding alone. We therefore turned to SPR, a technique that is not subject to this potential source of error, for the determination of binding constants.
eIF4A was immobilized on a sensor chip by the amino-coupling procedure, and eIF4G fragments were passed over it. Binding of S-eIF4G(613-1078) was measured over a range of concentrations from 3 to 100 nM (Fig. 6A). Binding of S-eIF4G(1078 -1560) was measured over 10 to 1000 nM (Fig.  6B). As a control, S-eIF4G(975-1078), which carries the same tags as two other eIF4G fragments but does not contain an eIF4A-binding site (Fig. 2), was passed over the immobilized eIF4A at concentrations ranging from 3 to 300 nM (Fig. 6C).  (lanes 1Ϫ11, respectively). The S-eIF4G(613-1078) and bound eIF4A were captured with a 10-fold molar excess of S-protein-agarose, subjected to SDS-PAGE on an 8% gel, and visualized with Coomassie Blue and autoradiography. B, the amounts of S-eIF4G(613-1078) and eIF4A in A were quantitated by autoradiography and by scanning stained bands and comparing the signals to standard curves of the two purified proteins electrophoresed on the same gel. BR is the molar ratio of eIF4A to S-eIF4G(613-1078), and [eIF4A] f is the total eIF4A concentration minus the complexed eIF4A. The curve is a least squares fitting of Equation 1 to the data, in which n and K d are allowed to vary. The curve shown corresponds to n ϭ 1.
The best fit to the experimental data for binding of S-eIF4G(613-1078) and S-eIF4G(1078 -1560) to eIF4A was provided with the simple 1:1 binding model (thin lines in Figs. 6, A and B; see "Experimental Procedures"). The low values of the average residuals (deviation of actual from theoretical) and 2 (Fig. 6D) indicate that the 1:1 binding model provides a good fit to the data. The rate constants obtained for association (k a ) and dissociation (k d ) are shown in Fig. 6D. An apparent dissociation constant, K d , can be calculated from k d /k a (Fig. 6D). However, it should be noted that there may be hidden intermediates in the binding of eIF4A to eIF4G, and the rate constants obtained only reflect the slowest step in this process. The results show that S-eIF4G(613-1078) associates with eIF4A about 24-fold faster than does S-eIF4G(1078 -1560). S-eIF4G(613-1078) also dissociates from eIF4A 1.2-fold faster than does S-eIF4G(1078 -1560). As a result, S-eIF4G(613-1078) has an affinity for eIF4A that is 19-fold higher than S-eIF4G(1078 -1560). K d values obtained by SPR are considerably lower than those obtained by S-protein-agarose for the same protein-protein interactions, (17 versus 96 nM for S-eIF4G(613-1078) and 330 versus 660 nM for S-eIF4G(1078 -1560)). This is consistent with the fact that the S-proteinagarose protocol represents a nonequilibrium condition and is expected to underestimate binding affinities.
eIF4G (613-1078) and eIF4G(1078 -1560) Compete for Binding to eIF4A-The presence of two eIF4A-binding sites on eIF4G evokes two possible models regarding the domain or surface of eIF4A that binds each site. In Model 1, suggested previously (28), a single eIF4A molecule can bind simultaneously to both sites in eIF4G. This would occur if two different surfaces of eIF4A were bound to the two sites. This model is not contradicted by the finding of a 2:1 stoichiometry with S-eIF4G(613-1560) (Fig. 4F), since at sufficiently high eIF4A concentrations, a second eIF4A molecule could displace the first, thus converting an eIF4A⅐eIF4G complex, with two contacts between eIF4G and eIF4A, to a (eIF4A) 2 ⅐eIF4G complex, with a single contact between eIF4G and each of the eIF4A molecules. In Model 2, the same surface of eIF4A binds to either site on eIF4G. In this model, binding of eIF4A to one site on eIF4G prevents its simultaneous binding to the other site. An alternative version of Model 2 that would predict the same results is that the two surfaces on eIF4A overlap each other.
To distinguish between these two models, we performed a competition experiment. One eIF4G fragment containing the COOH-terminal eIF4A-binding site and an S-tag, S-eIF4G(1078 -1560), was incubated with [ 14 C]eIF4A in the presence of increasing amounts of an eIF4G fragment containing the central eIF4A-binding site but no S-tag, eIF4G(642-1078) (the latter was produced by cleavage of S-eIF4G(613-1078) with Coxsackievirus 2A protease. Then the S-tagged eIF4G fragment was captured with S-protein-agarose (Fig. 7).
Two outcomes are predicted if Model 1 is correct. First, the untagged eIF4G fragment would be retained on S-protein-agarose in a "sandwich" [S-protein-agarose⅐S-eIF4G(1078 -1560)⅐eIF4A⅐eIF4G(642-1078)]. In Model 2, no untagged eIF4G fragment should be bound to the resin. The second prediction is that addition of the untagged eIF4G fragment should not decrease the amount of eIF4A bound to the S-protein-agarose if Model 1 is correct. In fact, based on the observation of positive cooperativity between the two eIF4A-binding sites (Figs. 4F and 5), one would expect enhanced binding of eIF4A to the S-protein-agarose. Model 2, on the other hand, predicts that the untagged eIF4G fragment in solution would compete with the S-tagged eIF4G fragment for binding with eIF4A, resulting in reduction of eIF4A bound to the resin.
As a control, one tube contained S-eIF4G(1078 -1560), [ 14 C]eIF4A, untagged eIF4G(642-1078), and eIF3 (Fig. 7, lane  6). If S-eIF4G(1078 -1560) and eIF4G(642-1078) were to make a sandwich via eIF4A, it would be possible to detect eIF3 bound to the S-protein-agarose, since eIF3 binds to eIF4G(642-1078) (20). However, there was no detectable eIF3 on the gel (Fig. 7A,  lane 6). Another set of controls contained untagged eIF4G(642-1078) and [ 14 C]eIF4A but no S-eIF4G(1078 -1560) (lanes 7-9). The amount of eIF4G(642-1078) bound to the resin was low (0.03-0.46 pmol; Fig. 7B). Most significantly, there was no difference in the amount of eIF4G(642-1078) bound to the resin in the presence or absence of S-eIF4G(1078 -1560) (Fig. 7B,  lanes 3-5 versus 7-9), indicating that the binding of eIF4G(642-1078) to the resin was nonspecific. All these finding are consistent with Model 2 but not Model 1, indicating that the two eIF4A-binding sites on eIF4G compete with each other for binding to eIF4A and fail to form a sandwich with eIF4A. DISCUSSION In our earlier study on the domain structure of eIF4G, we presented evidence that eIF3 was bound to the central domain, whereas eIF4A was bound to the COOH-terminal domain (22). A subsequent study showed that eIF4A was also bound to the central domain (28). This raised the question of the relationship between the eIF3-and eIF4A-binding sites in the central domain. A previous study proposed that these two sites overlapped (28). This suggestion was based on the observation that mutations affecting eIF4A binding to the central region of eIF4G also caused loss of eIF3 binding (22,28). Here we show that eIF3 and eIF4A do not compete for binding in the central domain, which suggests that the eIF4A-and eIF3-binding sites are distinct.
The stoichiometries of mammalian eIF4F subunits have not been previously determined. Furthermore, the relative amount of eIF4A in any given preparation of rabbit eIF4F varies. 3 eIF4A is lost from the mammalian eIF4F complex by phosphocellulose chromatography (36,11), Mono Q chromatography (43), and ultracentrifugation of eIF4F on sucrose gradients (20). For these reasons, merely measuring the relative amount of eIF4F subunits in purified preparations would not be expected to yield meaningful results. The approach we have taken here is to measure the direct interaction of purified eIF4A and recombinant eIF4G fragments. Titration of different S-tagged eIF4G fragments with eIF4A demonstrated that eIF4G fragments containing each of the individual eIF4A-binding sites bind eIF4A with a 1:1 stoichiometry, but fragments containing both sites bind eIF4A with a 1:2 stoichiometry.
We showed that the affinities of the two eIF4A-binding sites are different using two different methodologies, static binding on S-protein-agarose and kinetic binding using SPR. Both methods indicated that the central site has a higher affinity for eIF4A than the COOH-terminal binding site. The two methods did not agree quantitatively, however, with apparent K d values by SPR being lower than those by S-protein-agarose. The SPR results are expected to be more accurate, since SPR measures binding and dissociation instantaneously, whereas the S-protein-agarose method measures only the protein bound after several washes to remove nonspecifically bound protein. Given the propensity of eIF4F to lose eIF4A during purification, it is not surprising that the apparent affinities as determined by SPR are higher. The fact that the best fit of the SPR data was provided by the 1:1 kinetic binding model provides further evidence for a 1:1 stoichiometry for each site.
The basis for the difference in affinities between the central and COOH-terminal eIF4A-binding sites is revealed by the kinetic results. Turnover of eIF4A occupancy of the central site is faster than that of the COOH-terminal site. eIF4A associates with eIF4G(613-1078) about 24-fold faster than with eIF4G(1078 -1560) (k a ϭ 1.21 ϫ 10 5 versus 5.1 ϫ 10 3 M Ϫ1 s Ϫ1 ), but it also dissociates from eIF4G(613-1078) 1.2-fold faster than from eIF4G(1078 -1560) (k d ϭ 2.1 ϫ 10 Ϫ3 versus 1.7 ϫ 10 Ϫ3 s Ϫ1 ). This more rapid turnover of eIF4A bound to the central site may have mechanistic consequences.
Our data suggest that the binding of two molecules of eIF4A to eIF4G occurs cooperatively. The amount of eIF4A bound to an eIF4G fragment containing both sites was greater than the sum of eIF4A bound to the individual sites, the effect being less pronounced as both sites became saturated (Fig. 5) eIF4G-1 Interactions with eIF4A recent observation that alteration of either one of the eIF4Abinding sites by site-directed mutagenesis results in a form of eIF4G that fails to bind to eIF4A, despite the presence of an intact second site (28). The proposed cooperativity between sites may also explain why eIF4F complexes purified from plant (31,47,48), yeast (33), and Drosophila (35) do not contain eIF4A, despite the fact that direct binding of eIF4A to eIF4G can be demonstrated in wheat germ (49,50) and yeast (51,52) and that the affinity of yeast eIF4G for eIF4A (K d Х 30 nM; Ref. 52) is comparable with that of the central site of human eIF4G (K d ϭ 17 nM; Fig. 6). These plant and yeast eIF4Gs have high homology to the central domain of human eIF4G but low homology to the COOH-terminal domain. Consequently, there may be no enhancement of eIF4A binding by the COOH-terminal domain in the nonmammalian eIF4Gs.
Recently the central domain of eIF4G was postulated to work as a "ribosome recruitment core," in concert with other factors, based on its ability to form 48 S pre-initiation complex and drive initiation of translation (17,27,28). The COOH-terminal domain was proposed to play a modulatory role in this process, since an eIF4G fragment containing both eIF4A-binding sites is more active than one in which the COOH-terminal site has been inactivated by site-directed mutagenesis (28). Similarly, UV-cross-linking of RNA to eIF4A is stimulated more by an eIF4G fragment containing both binding sites than by either one separately (17). Our observation that binding of eIF4A to eIF4G is cooperative could explain these results by ensuring a higher affinity interaction between eIF4G and eIF4A.
Two models of interaction of mammalian eIF4G with eIF4A have been proposed (28). One model would require that eIF4A have two different surfaces for binding to eIF4G, allowing eIF4A to be sandwiched between the central and COOH-terminal binding sites. It was suggested that a resultant change in eIF4G conformation makes it more active in translation. The other model proposes that the central domain of eIF4G is sterically hidden from free eIF4A by the COOH-terminal domain. In this model, eIF4A first binds to the COOH terminus and is later transferred to the central region (28). Both models propose a 1:1 stoichiometry of binding eIF4A and eIF4G. We show here that eIF4G fragments containing the individual eIF4A-binding sites compete with each other and fail to form a sandwich on eIF4A. This indicates that the determinants on eIF4A that are recognized by these two different binding sites are either the same or overlapping, which is incompatible with the first model. Titration experiments showed a 1:2 stoichiometry of binding of eIF4G to eIF4A, which is incompatible with the second model. Also, the kinetic experiments indicate that eIF4A binds faster to the central site than to the COOHterminal site, contrary to the second model. Thus, our data allow us to propose a new model whereby eIF4G binds two molecules of eIF4A simultaneously. Furthermore, binding of these two molecules of eIF4A to eIF4G appears to occurs cooperatively. The implications of such a model on the concerted action of paired eIF4A molecules acting to accomplish the processive unwinding of mRNA remain to be explored.