INTRODUCTION

Top Abstract Introduction Procedures Results Discussion References

Initiation of nearly all eukaryotic mRNAs proceeds by a cap-dependent mechanism whereby the AUG nearest the 5'-end serves as the initiation codon (1). Yet other modes of initiation codon selection are used in special cases, e.g. leaky scanning, termination-reinitiation, ribosome shunting, and internal initiation (2). In the latter case, ribosomes are directed to internal AUGs by an internal ribosome entry site (IRES).¹ Internal initiation has been demonstrated by both in vitro and in vivo experimentation for picornaviruses (3), certain other viruses (4-6), and a growing number of cellular mRNAs (7-12).

Picornaviral IRESs have been the most characterized to date (3). The minimal size of picornaviral IRESs is ~450 nt, with further deletion into this sequence dramatically decreasing IRES activity. These IRESs have been divided into three groups: (i) entero- and rhinoviruses, (ii) cardio- and aphthoviruses, and (iii) hepatoviruses. Within each group, there is strong conservation of IRES secondary structure, somewhat less conservation of primary structure, but little conservation between groups apart from the existence of a polypyrimidine tract (PPT) located ~25 nt from the 3'-end of the IRES and a 3'-terminal AUG. The precise sequence of the PPT is important for IRES function in entero- and rhinoviruses but not cardioviruses (2). Several host cellular proteins bind to, and in some cases stimulate translation from, picornaviral IRESs, including La (13, 14), the polypyrimidine tract-binding protein (15-17) and eIF4G (18).

The mechanisms of initiation codon selection in picornaviral IRESs are likewise divided into three models. For cardioviruses such as encephalomyocarditis virus, and perhaps hepatoviruses as well, the 40 S ribosomal subunit is thought to enter at the 3'-terminal AUG of the IRES and use it as the translation initiation codon (the "landing" model). For entero- and rhinoviruses, the ribosomal subunit appears to enter at the same AUG but then scans for the next downstream AUG (the "scanning" model). The aphthoviruses, e.g. foot and mouth disease virus, may combine landing and scanning, as translation initiates at both the 3'-terminal AUG of the IRES and also at the next downstream AUG.

Another determinant of IRES utilization is the host protein synthesis machinery. Under normal conditions, most mRNAs utilize the cap-dependent translation pathway, but in cells infected with entero-, rhino-, and aphthoviruses, this pathway is shut down and viral mRNAs use instead a cap-independent mechanism (3). The switch from cap-dependent to cap-independent initiation is mediated by the proteolytic cleavage of eIF4G (19-21). eIF4G functions as a linking protein which joins, by virtue of its binding sites for eIF4E, eIF3, eIF4A, and poly(A)-binding protein, the various factors involved in mRNA recruitment to the 40 S ribosomal subunit (22-24). The action of the 2A proteases of entero- and rhinoviruses (25, 26) or the L protease of foot and mouth disease virus (27) releases the N-terminal portion of eIF4G, bound to the cap-binding protein eIF4E, from the initiation complex but leaves the C-terminal portion of eIF4G bound to eIF3 and eIF4A in the initiation complex (22). Such modified initiation complexes can apparently participate in internal initiation but not cap-dependent initiation, although the mechanism is not clear (18, 28, 29).

In comparison with picornaviral IRESs, relatively little is known about cellular (non-viral) IRESs. The 5'-UTRs of BiP (7), Ultrabithorax (12), and Antennapedia (8) mRNAs are, respectively, 220, 968, and 252 nt, but the locations and sizes of the IRESs within them are not known. In the case of the 318-nt 5'-UTR of fibroblast growth factor 2 mRNA, the IRES resides in a 165-nt segment (9). In the 1022-nt 5'-UTR of platelet-derived growth factor 2 mRNA, the IRES is present in a 395-nt segment (11). None of these cellular mRNAs contains a picornavirus-like PPT nor is there significant homology among them or with picornaviral IRESs.

The 5'-UTR of human eIF4G mRNA (30) is unusually long (368 nt), compared with typical 5'-UTRs of cellular mRNAs (31), and contains four upstream open reading frames, suggesting that its translation would be extremely inefficient if the scanning mechanism were used. Previously we demonstrated, using reporter constructs transfected into K562 cells, that the first 357 nt of the 5'-UTR of eIF4G mRNA has IRES activity (10). In the present study we have further characterized the IRES of eIF4G with respect to size, sequence requirements, and mechanism of initiation codon selection. Knowledge of this mechanism may shed light on how internal initiation occurs in other RNAs, both viral and cellular. Because of the unique role of eIF4G in protein synthesis initiation, this may also give insight into such questions as the mechanism by which intracellular levels of eIF4G are maintained and the balance between cap-dependent and cap-independent initiation.

	EXPERIMENTAL PROCEDURES

Top Abstract Introduction Procedures Results Discussion References

Materials-- Luciferin, restriction endonucleases, the DNA Cycle Sequencing System (fmol^®), enzymes used for in vitro transcription, and S1 nuclease were purchased from Promega (Madison, WI). RNase T1 was obtained from Life Technologies, Inc. (Gaithersburg, MD). RNase V1 was purchased from Pharmacia Biotech Inc. (Piscataway, NJ). Acetyl-coenzyme A was obtained from Sigma. Radioisotopes were purchased from ICN (Costa Mesa, CA). Silica gel thin layer chromatography plates (LK5F, 150 Å) were obtained from Whatman (Clifton, NJ). Pfu polymerase was purchased from Stratagene (La Jolla, CA).

Plasmids-- Deletions in the DNA corresponding to nt 1-357 of human eIF4G mRNA (30) were made from both 5'- and 3'-ends by PCR using Pfu polymerase. The PCR products used to make the first three 5'-deletions contained HindIII sites at the 3' but not 5' termini. It was therefore necessary to incorporate a 5'-HindIII site by cloning into pBluescript KS (Stratagene, La Jolla, CA) at the EcoRV site, which is adjacent to a HindIII site, and then screening for an insert with the same orientation as the T3 promoter. The fragments were excised with HindIII and inserted into pGL2/CAT/LUC (10) at the unique HindIII site between coding regions for CAT and luciferase. (Henceforth, plasmids names are abbreviated, e.g. pGL2/CAT/LUC is abbreviated C/L, pGL2/CAT/4G/LUC is abbreviated C/4G/L, etc.) PCR products for the rest of the plasmids shown in Fig. 1A and the 3'-deletion plasmids shown in Fig. 2A contained HindIII sites at each end. These were digested with HindIII and inserted into the HindIII site of C/L. The plasmid C/4G+/L contained DNA corresponding to the full-length 5'-UTR and the first three codons of eIF4G mRNA (nt 1-377), fused upstream and in-frame with the luciferase gene. This plasmid was made by a three-step PCR strategy. In the first step, pHFC5 (30) was used as template. The product contained sequences corresponding to nt 1-377 of eIF4G mRNA followed by nt 1-17 of the luciferase coding region. In the second step, pGL2/LUC (10) was used as template. The product contained sequences corresponding to nt 358-377 of eIF4G mRNA followed by nt 1-55 of the luciferase coding region. In the third step, gel-purified PCR products from the first two steps were mixed and used as templates for PCR. The product, which had a HindIII site at its 5'-end and a NarI site at its 3'-end, was digested with both enzymes and inserted into C/L at the same sites. Plasmids containing mutations in the 3'-region of the 5'-UTR (Figs. 3A, 4A, 5A, and 6A) were made by the same strategy but with C/4G+/L as template. Plasmid C/4G+B322-363/LA (Fig. 6A) was made by PCR with C/4G+B/LA as template. The resulting product, which had a 5' HindIII site and 3' BclI site, was digested with the same enzymes and inserted into C/4G+B/LA at the same sites. The plasmid used for in vitro transcription, pKS4GMCS, was made in the following way. The PCR product corresponding to nt 1-357 of human eIF4G mRNA (10) was inserted into pKS (Stratagene, La Jolla, CA) at the HindIII site. The polylinker upstream of the insert was removed by digesting with XbaI and SacI and re-ligating the plasmid. The structures of plasmids were confirmed by restriction digestion and sequencing (32) using double-stranded DNA as template.

Transient Expression-- K562 cells were transfected by electroporation using a Gene Pulser from Bio-Rad (Hercules, CA) as described previously (10). Electroporation was performed in triplicate for each plasmid and utilized 30 µg of DNA unless otherwise specified. Luciferase activity was measured (33) using a Monolight 2010 Luminometer from Analytical Luminescence Laboratory (San Diego, CA). CAT activity (10) was used as an internal control to normalize luciferase activity in each sample. Corrected luciferase activity values were averaged and standard deviations determined.

Northern Blot Analysis-- Total RNA was isolated from transfected K562 cells (34) and poly(A)-containing mRNA was purified (35). Poly(A)-containing mRNA from each sample (~1 µg) was used for electrophoresis and Northern blot analysis (36). The hybridization probe was synthesized as described previously (10) and consisted of RNA complementary to nt 47-547 of luciferase mRNA (where the initiating AUG is nt 1).

In Vitro Transcription and Labeling of eIF4G RNAs-- A 5'-end labeled form of the eIF4G 5'-UTR (nt 1-357) was produced by in vitro transcription of HindIII-linearized pKS4GMCS in the presence of 10 µCi/µl [-³²P]GTP (37). The nucleotide concentrations used in 100-µl reactions were 0.5 mM for ATP, CTP, and UTP but 0.25 mM for GTP.

Enzymatic Probing-- The ³²P-end labeled RNA was subjected to partial digestion with S1 and V1 nuclease as described previously (38, 39). RNA markers were synthesized by limited alkaline hydrolysis and RNase T1 nuclease digestion of the labeled RNA.

	RESULTS

Top Abstract Introduction Procedures Results Discussion References

5'-Deletion Analysis of the eIF4G 5'-UTR-- The initial characterization of IRES activity in eIF4G mRNA employed a portion of the 5'-UTR representing 357 nt of the total 368 nt (10). However, it was not clear how much of this sequence was required for IRES activity. We therefore prepared a series of plasmids expressing forms of the 5'-UTR that were deleted from either 5'- or 3'-ends. To serve as a guide in choosing segments to be deleted, we examined the distribution of single- and double-stranded regions in the 5'-UTR. Subjecting the 357-nt sequence to a folding algorithm (M-fold, GCG Sequence Analysis Package) produced a series of structures, all having free energies of folding between 105 and 110 kcal/mol. Many of the predicted stems and loops were common among the structures. To test the validity of these various structures, a 5'-end labeled transcript representing nt 1-357 was subjected to partial enzymatic digestion with single- and double-strand-specific nucleases (S1 and V1; see "Experimental Procedures"). In the theoretical structure exhibiting the best agreement with experimental results, 75% of the digestion sites were in the predicted single- and double-stranded regions (data not shown). These results indicate that additional experimental information will be required to arrive at a precise secondary and tertiary structure of the 5'-UTR. Nonetheless, the regions of double- and single-stranded structure were used as a guide for making deletions in the 5'-UTR. Each successive 5'-deletion (Fig. 1A) corresponded to the removal of an additional region of double-stranded RNA (based on nuclease sensitivity).

View larger version (44K): [in this window] [in a new window]   Fig. 1.   5'-Deletion analysis of the human eIF4G IRES. A, construction of vectors. DNA segments corresponding to nt 1-357 of human eIF4G mRNA or the indicated 5'-deletions were inserted between the CAT and luciferase coding regions of plasmid pGL2/CAT/LUC (abbreviated C/L) as described under "Experimental Procedures." Asterisks represent ATG triplets and the hatched box, the polypyrimidine tract. B, expression of luciferase activity. K562 cells were transfected by electroporation with the plasmids shown in A, and CAT and luciferase activities were measured as described under "Experimental Procedures." Relative luciferase activity was calculated by normalizing luciferase activity with CAT activity. The bars represent the average ± S.D. of three independent transfections. C, four of the nine plasmids in A were used to test the dose-response of luciferase expression to input DNA. Relative luciferase activity was measured as in B and then normalized to the 10-µg point.

3'-Deletion Analysis of the eIF4G 5'-UTR-- The 3'-boundary of the IRES was similarly defined by deletion analysis (Fig. 2A). Removal of 58 nt from the 3'-end completely abolished IRES activity (Fig. 2B, C/4G300-357/L versus C/4G/L). However, an RNA with only 38 nt deleted, transcribed from plasmid C/4G322-357/L, restored wild-type levels of luciferase. This indicates that the sequence between nt 300 and 321 is critical for IRES activity. More than half of this sequence consists of a 12-nt PPT.

View larger version (22K): [in this window] [in a new window]   Fig. 2.   3'-Deletion analysis of the eIF4G IRES. A, construction of vectors. DNA segments corresponding to nt 1-357 of human eIF4G mRNA or the indicated 3'-deletions were inserted between the CAT and luciferase coding regions of plasmid C/L. B, relative luciferase activity was measured as in Fig. 1B.

View larger version (22K): [in this window] [in a new window]   Fig. 3.   The effect of sequence variations in the PPT on the IRES activity. A, construction of vectors. DNA segments corresponding to the entire 5'-UTR (nt 1-368) plus the first three codons (nt 369-377) of human eIF4G mRNA, or the indicated sequence variants in the PPT (nt 304-315), were inserted between the CAT and luciferase coding regions of plasmid C/L. B, relative luciferase activity was measured as in Fig. 1B.

The Role of the PPT-- Since deletion of the region containing the PPT abolished IRES activity (Fig. 2), it was of interest to determine whether a PPT per se was needed or merely a spacer between the IRES and the initiation codon. We therefore substituted a polypurine stretch consisting of six A and six G residues for the 12-nt PPT (Fig. 3A, C/4G+Pu/L). This polypurine tract completely abolished IRES activity (Fig. 3B, C/4G+Pu/L versus C/4G+/L), indicating the need for a PPT. To determine whether a specific sequence in the PPT was required, we inverted the PPT sequence (Fig. 3A, C/4G+yP/L). This reduced IRES activity by 50%, but the modified 5'-UTR was still 50-fold more effective in directing luciferase expression than the negative control (Fig. 3B, C/4G+yP/L versus C/L). The plasmid expressing this inverted PPT contains a central sequence of CTTTC that is the same as in the original plasmid (Fig. 3A). To rule out the possibility that this precise sequence was important, we made a different construct in which the original six T residues and six C residues were rearranged into a TC repeat sequence (Fig. 3A, C/4G+TC/L). Expression from C/4G+TC/L was further decreased but still 30-fold higher than expression from C/L (Fig. 3B). These results suggest that a PPT is needed, but that the wild-type sequence is not critical for IRES activity.

Initiation Codon Used in the Bicistronic mRNA-- The extended construct C/4G+/L (Fig. 3A) expresses an RNA containing two in-frame AUG codons immediately upstream of the luciferase coding region, either one of which could, in principle, initiate translation of luciferase. The first of these, at nt 369 of the eIF4G mRNA (30), is presumed to be the in vivo initiation codon for eIF4G, although this has not heretofore been proven. This supposition is based on its location at the beginning of a 4191-nt open reading frame in eIF4G mRNA, but it is also possible that a non-AUG codon near nt 369 initiates translation. The next downstream AUG in the RNA product of C/4G+/L is the original initiation codon for luciferase itself. For a correct interpretation of mutations in the IRES, it was necessary to determine which initiation codon was being used in transfected K562 cells. We therefore mutated the ATG corresponding to the original luciferase initiation codon to AAG (Fig. 4A, C/4G+/LA). Expression of luciferase was not significantly altered (Fig. 4B), suggesting that the AUG at nt 369 was used. To confirm this, we mutated the ATG corresponding to this AUG to AAG in plasmid C/4G+/LA to produce C/4G+A/LA). This abolished luciferase expression altogether (Fig. 4B), proving that the AUG at nt 369 was the initiation codon. It was also possible that the immediate sequence context of the AUG at nt 369 played a role in its utilization for initiation. To test this, we mutated the AUG at 369 to AAG in plasmid C/4G+A/L. However, luciferase expression was at the same level as with C/4G+/L (Fig. 4B). In addition to establishing that the AUG at nt 369 is used, these results demonstrate that the presence of three extra amino acid residues at the N terminus of luciferase does not significantly affect expression of enzymatic activity.

View larger version (27K): [in this window] [in a new window]   Fig. 4.   Determination of the translation initiation codon used for luciferase in the bicistronic mRNA. A, construction of vectors. DNA segments corresponding to nt 1-377 of human eIF4G mRNA or the indicated mutations (lowercase) of ATG to AAG codons were inserted between CAT and luciferase coding regions of plasmid C/L. The first ATG (underlined) in plasmid C/4G+/L corresponds to the putative initiation codon for eIF4G mRNA at nt 369. The second ATG corresponds to the original initiation codon of luciferase mRNA. B, relative luciferase activity was measured as in Fig. 1B.

Mechanism of Initiation Codon Selection-- In picornaviruses, there are three types of mechanisms determining initiation codon selection: landing, scanning, and mixture of both landing and scanning (see Introduction). To determine the mechanism of initiation codon selection for the eIF4G IRES, we produced three plasmids from C/4G+/LA containing three different out-of-frame ATGs, all of which were in an identical translation initiation context (41), a context that was in fact better than the context at nt 369 (Fig. 5A). When the out-of-frame AUG was introduced upstream of the PPT (Fig. 5A, C/4G+298A/LA), it did not significantly affect luciferase expression (Fig. 5B, C/4G+/LA versus C/4G+298A/LA). This suggests that if ribosome scanning is involved in initiation codon selection, it must start somewhere downstream of nt 298. When out-of-frame AUGs were introduced at either nt 319 (C/4G+319A/LA), which is immediately downstream of the PPT, or nt 340 (C/4G+340A/LA), which is 25 nt downstream of the PPT, activity was reduced by approximately 80% (Fig. 5B). This suggests that these out-of-frame AUGs provide alternative initiation sites for scanning 40 S ribosomal subunits.

View larger version (29K): [in this window] [in a new window]   Fig. 5.   The effects of out-of-frame AUGs on luciferase expression initiated from the AUG at nt 369. A, construction of vectors. DNA segments corresponding to the entire 5'-UTR plus first three codons of human eIF4G mRNA, or the indicated mutations in this sequence, were inserted between CAT and luciferase coding regions of plasmid C/L in which the original initiation codon of luciferase was mutated to AAG. Mutations (lowercase) introduced out-of-frame AUG codons (underlined) in the transcribed RNA at nt 298, 319, or 340. PPT signifies the polypyrimidine tract. The grouping of nucleotide residues into triplets corresponds to the reading frame of luciferase. B, relative luciferase activity was measured as in Fig. 1B.

Changing the Length of the Sequence between the PPT and the Initiation Codon-- In cardioviruses, the ~25-nt sequence between the PPT and the initiation codon serves as a spacer, and the length of this spacer affects IRES activity (2, 3). The length of the spacer is critical in the landing model but not the scanning model. Shortening the spacer shifts initiation to the next downstream AUG. We investigated the role of a putative spacer in the eIF4G IRES by deleting sequences corresponding to nt 322-363 of eIF4G mRNA. To remove this sequence from plasmid C/4G+/LA, it was necessary to introduce a BclI site at nt 363-368 by changing the A at nt 363 to T. The resultant plasmid, C/4G+B/LA, was used to construct plasmid C/4G+B322-363/LA, in which the sequences from nt 322 to 363 were removed (Fig. 6A). This deletion shortened the spacer between the PPT and the initiator AUG from 53 to 11 nt. However, the expression of luciferase was not decreased significantly (Fig. 6B, C/4G+B/LA versus C/4G+B322-363/LA), arguing against the need for a spacer of a particular length. If initiation had occurred at the next downstream AUG in luciferase mRNA, an inactive enzyme would have been produced (Fig. 4, C/4G+A/LA).

View larger version (20K): [in this window] [in a new window]   Fig. 6.   The effect of a "spacer" between the PPT and initiation codon. A, construction of vectors. A DNA segment corresponding to the entire 5'-UTR plus first three codons of human eIF4G mRNA, but with an A  T mutation at nt 363 to produce a BclI site (B), was inserted between CAT and luciferase coding regions of plasmid C/L, in which the original initiation codon of luciferase was mutated to AAG, to make C/4G+B/LA. The same sequence lacking nt 322-363 was also inserted to make C/4G+B322-363/LA. B, relative luciferase activity was measured as in Fig. 1B.

	DISCUSSION

Top Abstract Introduction Procedures Results Discussion References

Progressive deletion from the 5'-end of the 5'-UTR of eIF4G mRNA produces both stimulatory and inhibitory effects (Fig. 1). Without the knowledge of the precise structure of the 5'-UTR, it is impossible to draw firm conclusions from these changes in IRES activity, but based on picornaviral IRESs (2), these may reflect loss of secondary structure elements that confer positive or negative effects, refolding of the 5'-UTR into more or less active conformations, or loss of protein-binding sites. The deletion of nt 1-57 or 1-78 correlates, respectively, with removal of the first two or three upstream open reading frames of the 5'-UTR. Yet it is unlikely that the upstream open reading frames per se produce this inhibitory effect, since mutation of the upstream AUGs to AAGs has no detectable effect on IRES activity (10). A surprising increase of IRES activity also arose after the deletion of nt 231-256. In hepatitis A (42-44) and hepatitis C (45) viruses, it has also been shown that deletions or mutations in the 5'-UTR can increase IRES activity.

In contrast to the variable effects of 5' deletion, there is a total loss of IRES activity upon 3'-deletion of 58 nt (Fig. 2). This suggests that the PPT from nt 304-315 is critical for IRES activity, since replacement with an equal number of purine residues abolished IRES activity (Fig. 3). Functional PPTs have not been described in the other non-viral, cellular IRES discovered to date (7-9, 11, 12), but they are a conserved element in picornaviral IRESs (2). In the case of poliovirus, the 5'-proximal UUUC of the PPT is strictly required for IRES activity (46-48). In encephalomyocarditis virus, by contrast, the PPT can be replaced with purine residues with only partial loss of IRES activity (49, 50). Thus, this feature of the eIF4G IRES sets it apart from all other IRESs characterized to date: unlike other cellular IRESs, the eIF4G IRES requires a PPT for activity; unlike poliovirus, the eIF4G PPT has no obligatory sequence requirement; unlike encephalomyocarditis virus, it cannot be replaced by purines.

What is the minimum amount of RNA sequence needed for full IRES activity? The positive effect on IRES activity caused by the deletion of nt 1-57 appears to balance the negative effect of deleting nt 58-160, with the result that a 196-nt RNA consisting of only nt 161-357 (Fig. 1B, C/4G1-160/L) has IRES activity equal to the complete 5'-UTR. Since sequences beyond nt 321 have neither positive nor negative effects on IRES activity (Figs. 2 and 6), an RNA of only 161 nt, from nt 161 to 321, should have full IRES activity. (This construct has not been made.) Another fragment of the eIF4G IRES with the same activity as the complete 5'-UTR is the 101-nt stretch from nt 257-357 (C/4G1-256/L). An even shorter RNA of only 81 nt, transcribed from plasmid C/4G1-276/L, still has detectable IRES activity (55-fold higher than the negative control C/L). With such a short stretch of RNA, however, one cannot rule out the possibility that some of the luciferase translation is due to termination-reinitiation. Picornaviral IRESs, by contrast, are much larger (~450 nt) (2, 3). Of the viral IRESs other than those of picornaviruses, the smallest IRES is that of hepatitis C virus at 200-300 nt (4, 51). Non-viral IRESs appear to be smaller than viral IRESs, with that of fibroblast growth factor 2 mRNA at ~165 nt (9) and that of BiP mRNA at ~220 nt. The reason may be that the cellular IRESs are presumably used only for translation while viral IRESs have other functions as well, e.g. replication (52).

The experiments presented here give some insight into the mechanism of initiation codon selection in eIF4G mRNA. First, mutational analysis indicates that the AUG at nt 369 is the initiation codon (Fig. 4). The fact that multiple eIF4G bands are observed on SDS-polyacrylamide gel electrophoresis (30) has given rise to speculation that eIF4G is initiated at several different translation initiation sites, but this is now less likely. Second, the data in Fig. 5 suggest that the 40 S ribosomal subunit lands between nt 298 and 319. The upstream limit is based on the finding that an out-of-frame AUG at nt 298 had no effect on luciferase expression. The dramatic decrease in luciferase expression caused by introduction of out-of-frame AUGs at nt 319 and 340 suggests that the downstream limit is at nt 319. This also suggests that ribosomal subunit scanning is involved, because initiation codons can be recognized by 40 S subunits regardless of whether they are located at nt 319, 340, or 369. Further support for the scanning model comes from experimental results that are not consistent with a landing model. In the landing model, used by encephalomyocarditis virus, the spacing between the PPT and the initiator AUG is critical for expression (2). Yet in the case of the eIF4G IRES, similar expression is observed when the spacing is 12 nt (Fig. 6, C/4G+B322-363/LA), 35 nt (Fig. 2, C/4G 322-357/L), 55 nt (Fig. 4, C/4G+/L), 62 nt (Fig. 4, C/4G+A/L), and 70 nt (Fig. 2, C/4G/L). With respect to models of initiation codon selection, the eIF4G IRES is more similar to entero- and rhinovirus IRESs (53-55) than cardiovirus IRESs (49) because scanning seems to occur and because a 20-25 nt spacer is not needed. The failure of out-of-frame AUGs downstream of the PPT to completely eliminate luciferase expression suggests leaky scanning.

Despite these similarities of the eIF4G IRES to those of picornaviruses, it must be recognized that the cellular IRESs may operate in a fundamentally different manner than picornaviral IRESs. This is because cellular IRES-containing mRNAs contain two regulatory elements, the cap and poly(A) tail, that are not present in picornaviral RNAs. These elements have strong, positive roles on translation initiation (56, 57). It is conceivable that one or both of these elements play a role in translation of cellular IRES-containing mRNAs. The cap, for instance, may be a factor in the autoregulation of eIF4G protein levels. In such a scenario, if the cellular capacity to carry out cap-dependent translation were high, the cap of eIF4G mRNA would compete with the IRES for binding of the 40 S ribosomal subunit, but this would not lead to eIF4G synthesis. If the capacity to carry out cap-dependent initiation were low due to low eIF4G levels, the cap would not compete with the IRES for 40 S subunit binding and eIF4G mRNA would be translated. A similar scenario could be envisioned concerning poly(A)-mediated translation initiation.