Department of Biochemistry and Molecular Biology, Louisiana State University Medical Center, 1501 Kings Highway, Shreveport, Louisiana, LA 71130-3932, USA

Correspondence to: Arrigo De Benedetti, Department of Biochemistry and Molecular Biology, Louisiana State University Medical Center, 1501 Kings Highway, Shreveport, Louisiana, LA 71130-3932, USA

eIF4E is essential for translation initiation, but its overexpression causes malignant transformation. Recent work demonstrated that eIF4E/F participates in exposing and locating alternate translation start codons during scanning. Translation initiation of several important protooncogenes and growth-regulators, such as Myc and FGF-2, can start at CUG start codon(s) upstream of the normal open reading frame (ORF). The resulting amino-terminal extension alters the properties of these proteins and their intracellular distribution. In cells overexpressing eIF4E, c-myc is overexpressed and particularly the larger, CUG-initiated form (Myc1). Recent reports suggest that synthesis of Myc2, the normally expressed AUG-initiated form, is mediated by an IRES. To determine what role eIF4E might play in c-myc expression, the c-myc 5' untranslated region (UTR) was fused in-frame to CAT reporters, and several more derivative constructs were made. In vitro translation experiments (with and without eIF4E/F); expression in CHO cells transformed with eIF4E; and deletion/mutation analysis demonstrated that Myc1 is translated by a scanning mechanism, while Myc2 is translated by Internal Ribosome Repositioning. Moreover, the existence of a true IRES in the 5'UTR was contradicted by its failure to direct translation of a circular transcript, in contrast to hsp70. The c-myc 5'UTR also failed to engage in translation in the absence of functional eIF4F, after cleavage of the eIF4G component with CVB4 protease-2A. The Internal Repositioning Element (IRPE) in c-myc 5'UTR was delimited to nucleotides (nt) 394 - 440 from the P1 transcription start site.

translation initiation factor 4E; c-myc expression; internal ribosome repositioning; oncogenic transformation; alternate start-codon selection

Introduction

C-myc is a well characterized proto-oncogene, held to be a transcriptional regulator with dual repressor and activator functions, and it plays fundamental roles in cell proliferation, differentiation and apoptosis (Gandori and Eisenman, 1997). Mutations of the gene and alterations in its expression are implicated in several malignancies, including Acute Myelogenous Luekemia and pre-leukemia (Reddy and Sulcova, 1997). Translocations involving c-myc are implicated in Burkitt's Lymphoma (Dalla-Favera et al., 1982; Taub et al., 1982; Ar-Rushdi et al., 1983) and altered translation of c-myc may be involved in the genesis of Bloom's Syndrome (West et al., 1995).

Transcription of the human and mouse c-myc can start from four promoters, designated P0, P1, P2 and P3, of which P1 and P2 are most commonly used (Bentley and Groudine, 1986). Both P1 and P2 transcripts can encode two products from alternate initiation at two in-frame start codons, CUG and AUG, resulting in two proteins designated Myc1 and Myc2 (Hann et al., 1988). In a small number of mRNAs encoding important regulatory proteins, translation can initiate at upstream non-AUG codons, typically in addition to a downstream and in-frame AUG. The resulting N-terminally extended product(s) have altered function and cellular distribution. CUG codon initiation has been identified in c-myc (Hann et al., 1998), FGF-2 (Florkiewicz and Sommer, 1989), L-myc (Dosaka-Akita et al., 1991), lyl-1, (Mellentin et al., 1989), pim-1 (Saris et al., 1991), and int-2 (Acland et al., 1990). Myc1 and Myc2 are nuclear phosphoproteins with apparent molecular weights of 70 and 64 kDa, respectively, even though they differ by only 14 amino acids at the N-terminus. The two proteins reportedly differ in transcriptional activity (Hann et al., 1994). Both Myc1 and Myc2 can transactivate the E-box Myc Site (EMS), but Myc1 can also bind to the CCAAT-enhancer (C/EBP) motif. Overexpression of Myc1, in the absence of Myc2, was reported to inhibit cell growth (Hann et al., 1994), whereas the normally predominant Myc2 is responsible for growth and proliferation. Depending on the specific conditions, differential synthesis of Myc1 and Myc2 could activate or inactivate different subsets of genes.

The 5'UTR of human c-myc mRNA is inhibitory for translation in most systems, including Xenopus oocytes, but not in COS cells or HeLa extracts (Parkin et al., 1988; Fraser et al., 1996). Definitive conclusions from these studies were hampered by the short half-life of Myc and its autoregulatory suppression when overexpressed (Pause and Sonenberg, 1993; De Benedetti et al., 1994). Most investigators have thus resorted to CAT fusions in order to study the translational regulation of c-myc in vivo. The sequences between the P1 and the P2 promoter have been found to be most inhibitory to translation in Xenopus oocytes (Lazarus, 1992; Fraser et al., 1996). Therefore, we have focused our investigation primarily on P1-initiated transcripts. Recent work with the human c-myc 5'UTR also suggests the existence of an IRES (Internal Ribosome Entry Site), as assayed with a bicistronic reporters (Nanbru et al., 1997; Stonely et al., 1998). These studies, however, do not rule out other possibilities and do not address the problem of translation initiation at the upstream CUG codon (Myc1). Since cells that overexpress the translation initiation factor eIF4E are able to relieve the translational repression imposed by highly structured 5'UTRs, we set to study the effect of elevated eIF4E on translation of c-myc and its impact on Myc1 : Myc2 ratios.

Results

Synthesis of Myc1 and Myc2 proteins is increased in CHO-4E cells

Overexpression of eIF4E in CHO or CREF cells leads to elevated Myc synthesis (De Benedetti et al., 1994) attributable to increased translation initiation, based on alterations in the polysome distribution of c-myc mRNA (data not shown). The level of Myc2 protein was increased 3 - 5-fold in CHO-4E over CHO-BK (control cells). Myc1 was not detectable in CHO-BK extracts, but was clearly expressed in CHO-4E, again supporting a translational effect by eIF4E which, as a component of the helicase machinery operating during scanning may be involved in the process of recognition of the CUG start codon. We thus set out to study this mechanism of alternate initiation, and its dependence on eIF4E, with CAT reporters using in vitro and in vivo systems.

Addition of eIF4F or eIF4E relieves the translational inhibition of c-myc 5'UTR

We made several constructs containing the 5'UTR of human c-myc fused upstream of a CAT reporter. This construct (E1CAT) was transcribed in vitro (with and without caps) to test whether exon 1 was inhibitory to translation of CAT in reticulocyte lysate (RRL). Indeed, E1CAT was translated 20-fold less well than the BSCAT control transcript on a mole basis (Figure 1a, lane 4). The addition of 3 or 32 pmol of eIF4F caused a strong increase in translation of E1CAT. After normalization for the different amounts of input mRNA, translation of E1CAT supplemented with 32 pmol of eIF4F was almost as efficient as that of BSCAT. This indicates that the level of eIF4F required to overcome exon 1 inhibition is at least tenfold over the endogenous level in RRL. No amount of eIF4F had any effect on the translation of BSCAT since this mRNA does not possess an inhibitory 5'UTR and the endogenous eIF4F is sufficient to drive its translation at the maximal rate (for that amount of input transcript). Since eIF4E is the limiting component of the eIF4F complex, we also determined that recombinant eIF4E alone could stimulate E1CAT translation in RRL (Figure 1b). From this and several other experiments we concluded that the full-length 5'UTR of c-myc is highly inhibitory for translation in RRL, and that this inhibition can be relieved by added eIF4E/F. As such, the translation of c-myc should be considered cap- and eIF4E-dependent.

Translation initiation at c-myc-CUG is remarkably efficient in RRL

In E1CAT, the CUG start site in c-myc exon 1 is out of frame with the AUG of CAT, hence, no product initiated from the c-myc CUG can be detected. To mimic the Myc1 and Myc2 situation, we made the E1CATsh1 and the E1CATsh4 reporter constructs placing the CUG and AUG in-frame (see diagram in Figure 1d). Translation from the alternate CUG site gave a slower-mobility product in addition to the normal band from the CAT - AUG (Figure 1c, lane 2). This supports the finding that Myc1 is indeed generated from initiation at the CUG site (Hann et al., 1988) and that this mechanism of initiation is retained in a heterologous construct. Translation from the CUG was approximately 25% of that from CAT - AUG, which is remarkably efficient (CUG initiation is normally only a few per cent as efficient). The addition of 32 - 64 pmol of eIF4F increased the translation at CUG to about 40% that of CAT - AUG (data not shown). In the CATsh4 construct, the CUG is also in-frame with the CAT - AUG, but the sh4 linker adds a potential stem with a G of -73 kcal/mol. The CUG and the AUG initiated forms of this construct were synthesized in RRL, albeit poorly (Figure 1c, lane 4) and the ratio of the two products was qualitatively similar to that from E1CATsh1. If translation from CAT - AUG occurred by a simple scanning (or leaky-scanning) mechanism, interpolation of a stable hairpin upstream of the AUG should have resulted in a reduction of initiation at CAT - AUG and in altered CUG/AUG ratio, but this was not the case. Later on, we also show that changing the CUG to an AUG codon has no effect on the pattern of translation: clearly against a scanning model of initiation at the CAT - AUG.

Initiation at the CUG is enhanced by eIF4E overexpression, while interpolation of secondary structure downstream of CUG does not affect CAT - AUG initiation in vivo

The E1CATsh1 and the E1CATsh4 constructs were subcloned into the mammalian expression vector pREP11 to generate p11sh1 and p11sh4, and these episomal constructs were stably transfected into CHO-BK and CHO-4E cells. Cell extracts were resolved by SDS - PAGE gels and probed with an anti-CAT antibody. In CHO-BK, the CAT - AUG form was expressed efficiently (Figure 1e), but CAT-CUG was weakly made, or slightly better with the p11sh4 construct (lane 2). This is probably due to arrest of any residual scanning immediately past the CUG codon, which increases the likelihood of its recognition as a potential start site (Kozak, 1990). In CHO-4E cells the AUG-initiated form was notably decreased, but the CUG initiated form was expressed efficiently (Figure 1e, lanes 3, 4). These results suggest a scanning, eIF4E-dependent mechanism for initiation at CUG, but the mechanism for initiation at the CAT - AUG remained unclear. The fact that translation of CAT - AUG was efficient in CHO-BK would indicate that it may be internally-initiated and that the inhibitory effect by the c-myc 5'UTR seen in vitro may be largely bypassed in vivo. Initiation at CAT - AUG by leaky scanning was considered improbable, as this should have occurred more readily in CHO-4E, given the excess of cap-dependent helicase in these cells. Overall, it seemed that translation of CAT-AUG was initiated internally, possibly via an IRES. From these experiments it also became clear that the element(s) that mediates the internal initiation cannot be between the CUG and the AUG start sites and it must be upstream of CUG. For one thing, translation at the internal AUG was replicated in a heterologous construct (CAT) in which only half the sequences between CUG and AUG are from c-myc. Secondly, the efficiency of initiation at CAT - AUG was little affected by the sh4 hairpin, particularly in CHO-BK cells. If the element facilitating internal translation at CAT - AUG was located between CUG and AUG, interpolation of a strong hairpin should have decreased its functionality.

Translation of circular E1CAT and hsp70 transcripts: translation at CAT - AUG is not mediated by an IRES

The central postulate of the scanning model for translation of eukaryotic mRNA is that pre-initiation complexes assemble at the 5'-end and migrate in the 3' direction until the first AUG in good context is located (Kozak, 1978). A corollary to this is that a free 5'-end must be available for initiation to occur. Indeed, early experiments by Kozak ruled out translation of circular transcripts in RRL, although this occurs in prokaryotes. This dramatic distinction between eukaryotic and prokaryotic protein synthesis has remained mostly unchallenged until recently an efficient method to generate circularized mRNAs was developed (Chen and Sarnow, 1995). The method utilizes a deoxy-oligonucleotide complementary to the beginning and the end of a transcript to bring the 5' and 3' ends in juxtaposition; then, the addition of T4 DNA ligase efficiently seals the gap (see diagram in Figure 2a). By this method, a circularized transcript containing the IRES of EMCV was efficiently translated in RRL (Chen and Sarnow, 1995). We now show that translation of a circularized human hsp70 transcript, a mRNA with little translational dependence on eIF4E/F (Kevil et al., 1995; Joshi-Barve et al., 1992), also occurs readily in RRL. The transcription/ligation reactions of hsp70 are shown in Figure 2b(left panel). The purified linear (~) and circular () transcripts were used to program RRL and showed equivalent translations (Figure 2b, right panel). If initiation occurred randomly from internal sites on the circular transcript, no full-length products would be expected. Although some premature termination is observed in these translations, there is no gross difference between linear and circular mRNAs, indicating the translation of the entire ORF occurs precisely. The possibility of nicking of the circular template upstream of the ORF to generate a linear mRNA is very unlikely, given the similar efficiency of the two translations. However, we investigated this possibility by boiling and extensively nicking the circular transcript (symbolized by [white circle with intersecting plus sign}). In this case, no convincing translation products were obtained even after a long film exposure (Figure 2b, lane 4), indicating that an intact circular transcript is responsible for the synthesis of hsp70 in lane 3. Thus, translation of circularized hsp70 defines an authentic IRES-mediated translation for this mRNA, by a more stringent criterion than that typically imposed with bicistronic constructs, since ribosomal access from the 5'-end is physically precluded in circular transcripts.

When the same strategy was tested on translation of a circularized E1CAT transcript, the result was negative. Only a very faint band, attributable to a trace contamination of gel-purified linear transcript, can be seen in lane 2 of Figure 2c (right panel). Thus, the putative IRES contained in the 5'UTR of c-myc (Nanbru et al., 1997) failed to direct ribosomal entry and/or positioning at the CAT - AUG, indicating that its translation requires a free 5'-end for ribosome loading. The experiments in Figure 2 argue against the existence of an IRES in the c-myc 5'UTR, or at least that it is clearly different from those of EMCV or hsp70. Therefore while translation at CAT - AUG is probably internally initiated, it is not likely to involve internal entry of ribosomes.

Translation driven by the c-myc 5'UTR is abolished upon cleavage of eIF4G

Another feature of cap-independent IRES-driven translation, is its resistance to cleavage of the eIF4G component of eIF4F (Lamphear and Rhoads, 1996 and refs therein); a strategy exploited by picornaviruses to selectively shut-off host protein synthesis. We thus tested whether the putative c-myc IRES has translational capacity that is resistant to eIF4G cleavage by coxsackievirus 2A protease. We tested the ability of E1CATsh1 to be translated in 2A protease-treated RRL (Figure 3). For control, we also tested the translation of hsp70 mRNA, and a bicistronic construct pXLJ-EMC. Translations of hsp70 and of the IRES-driven NS' protein in pXLJ-EMC were virtually unaffected by cleavage of eIF4G (Figure 3, lanes 7 and 5). This was expected for two mRNAs that are translated by IRES-mediated mechanism. In contrast, translation of E1CATsh1 was suppressed in 2A-treated RRL (Figure 3, lane 3). Note that the translation inhibition of E1CATsh1 upon eIF4G cleavage was even greater than that observed for cyclin B, the cap-dependent ORF in pXLJ-EMC (lane 5). It is then clear that translation from the c-myc 5'UTR requires intact eIF4G. While it is not known if all IRES-containing mRNAs can be translated when eIF4G is cleaved, and indeed there may be some exceptions (Borman and Kean, 1997), this generally argues against an IRES-mediated model of translation initiation.

Mutation and deletion analysis indicates a scanning mechanism for CUG-initiation and ribosome repositioning for AUG initiation

We generated a series of deletions in the 5'UTR of E1CAT to identify the role of cis-acting sequences, and we also mutated the CUG to an AUG. The resulting constructs were transcribed and translated in RRL, and the salient results are as follows. Changing the CUG to AUG in the E1CATsh1 construct did not significantly alter the pattern of translation (compare lane 1 with lane 4 in Figure 4a). In particular, translation of the downstream CAT - AUG was not decreased, indicating again that it is not recognized via scanning. In the context of the sh4 construct, containing the stable hairpin, the relative translation was slightly better for the upstream AUG (Figure 4, lane 2). This could indicate that, in the context of an interpolated hairpin, the presence of an upstream AUG helps capture all of the scanning ribosomes, but it has no effect on those that have already assembled internally at the CAT - AUG.

Deletion of 283 or 380 nucleotides (lanes 5 and 6) from the P1 promoter, in the normal E1CATsh1 construct, resulted in a twofold stimulation of translation of both CUG and AUG products over the full-length transcript (lane 4), after normalization to a non-specific reference band (Figure 4). Deletion of 394 nucleotides showed a fourfold stimulation of both products (lane 7). Further deletions of 436 or 441 nucleotides caused a 25% reduction in overall translation respect to the full-length E1CATsh1 and a relative increase in the AUG over the CUG product (lanes 8 and 9). The ratio of CUG to AUG products remained at 1 : 1 in all deletions except for the 436 and 441 where the ratio was 2 : 3. Only after deletion of about 440 - 500 nt, in addition to mutating the CUG to AUG, was translation from the CAT - AUG completely eliminated (lane 10). These results indicate that starting with the 436 and 441 constructs, the mechanism of translation becomes a more canonical leaky scanning process, and that sequences between 394 and 441 likely encompass the IRPE.

Placement of a stable hairpin at the 5'-end inhibits translation of both CAT products

Results from the deletion series are consistent with the hypothesis that initiation at both CUG and AUG begins by scanning through an upstream inhibitory region in c-myc 5'UTR, as previously suggested by other studies (Lazarus, 1992). This argues against the classic model of internal initiation, which should be insensitive to inhibitory sequences upstream of the IRES. However it was possible that these upstream sequences could compete for some folding pattern required by the IRES. In order to assess this possibility, we tested the effect of upstream secondary structure in the deletion construct -394, which contains the necessary functional elements and is translated more efficiently. We then placed a strong hairpin in front of this 5'UTR instead of the normal inhibitory region. The stability of the hairpin (G» -85 Kcal/mol) is such that it effectively arrested scanning-mediated initiation, as tested on BSCAT (data not shown). The predicted structure is so stable that it cannot appreciably anneal to anything but itself, eliminating the possibility of affecting the folding pattern of the putative IRES. The synthesis of both CAT - CUG and CAT - AUG products from this construct (-394-hpn) was severely reduced (Figure 4a, compare lanes 7 and 11) arguing against an IRES model of translation. Taken together, these results are best explained by a model of internal ribosome repositioning on the c-myc 5'UTR (see model in Figure 4c). In this model, cap-mediated ribosome recruitment and initial scanning is subsequently diverted by an IRPE, whereby the ribosome is repositioned (shunted) to the downstream AUG (of either CAT or Myc2).

Discussion

The expression of c-myc is subject to complex regulation at multiple levels. This complexity underlies the importance for the appropriate balance of c-myc products to cell survival and function. Clearly, translational regulation affecting the ratio of Myc1 and Myc2 (Hann et al., 1994) may alter the spectrum of responsive genes. In some multiple myeloma lines, c-myc is known to be aberrantly regulated at the translation level (Paulin et al., 1996). The ability of eIF4E to enhance the synthesis of both Myc1 and Myc2, but more drastically Myc1, implies cap-dependent ribosomal entry from the 5'UTR. The ability of recombinant eIF4E to relieve the translation inhibition of E1CAT reporters (both in- and out-of-frame) in RRL confirmed cap-dependent entry from the 5'UTR (note that in RRL we have control over most components of the translation apparatus). The recent reports of IRES activity in the c-myc 5'UTR in the context of bicistronic constructs in vivo, do not constitute proof for existence of an IRES in the natural context. The fact that the translation of c-myc is highly dependent on eIF4E is also inferred by its activation upon stimulation of the FRAP/mTOR pathway, which leads to the dissociation of the inhibitory proteins, 4E-BPs, from eIF4E (Mendez et al., 1997). This was recently confirmed by a group who initially postulated the existence of the IRES in c-myc (West et al., 1998).

Our work provides evidence for an internal ribosome repositioning element (IRPE), or shunting, as being operative. Clearly, isolated IRPEs could stimulate internal initiation in bicistronic constructs, given the similar function that IRES and IRPE must perform. One common way to establish IRES activity is to place a stable hairpin structure at the 5'-end of the first mRNA in a bicistronic construct. Translation of the upstream mRNA will then be inhibited due to arrest of the scanning machinery, but not that of the downstream IRES-driven mRNA. The stem - loop structure must be stable enough to block translation by scanning: a G near -80 kcal/mol to completely knock out initiation (Koromilas et al., 1992). A hairpin structure was used to probe for IRES activity in the c-myc 5'UTR in the context of a bicistronic construct (Nanbru et al., 1997), but the hairpin had a G of only -60 kcal/mol. In contrast with the aforementioned result, when a hairpin with a G of -85 kcal/mol was used with the 394 construct, which contains the functional element and displays the strongest translation, a strong reduction of both CAT products was observed. Moreover, when the c-myc 5'UTR was put through the more rigorous test of translation initiation from a circular transcript (see Jackson, 1996 for a review), which clearly lacks a free 5'-end, no IRES activity was found.

Internal ribosome repositioning (shunting) was first described for the Cauliflower Mosaic Virus (CaMV) 35s RNA (Fütterer et al., 1993). Similarly, in Adenovirus mRNAs the tripartite leader was found to provide an IRPE (shunting, jumping) function (Yueh and Schneider, 1996). Perhaps, the most critical distinction between IRES and IRPE activity is that both scanning and repositioning must coexist in mRNAs containing an IRPE. In some cases, the scanning machinery could eventually exclude repositioning, due to melting of the IRPE. This is what happens in the presence of excess eIF4F (helicase) complex, associated with the scanning ribosome, along the c-myc 5'UTR. The results from experiments in vivo also show that expression from the CAT - AUG is efficient in CHO-BK cells, but initiation at CUG was negligible with the p11sh1 construct, although the p11sh4 construct allows some CUG initiation. In general, initiation at CUG is bypassed altogether by the IRPE in normal CHO cells. In CHO-4E cells, however, initiation at AUG is reduced and that at c-myc CUG is concomitantly increased (in both sh1 and sh4 constructs). Overall, this suggests that initiation at the AUG occurs by internal translation initiation, not to be confused with internal entry. Initiation at CUG is probably by normal scanning aided by excess eIF4E/F helicase. It should be noted that the sequence surrounding the CUG codon is in perfect context, A at the minus three position and G at the plus one position (Kozak, 1978). Changing the CUG to AUG, in the context of the full 5'UTR, did not increase its recognition in RRL. Perhaps, this is also due to the presence of a natural hairpin a short distance downstream from the CUG (Figure 1d), that may dramatically improve its recognition (Kozak, 1990). This turned to our advantage, since the highly efficient initiation at the CUG by scanning made it very improbable that initiation at the downstream AUG could occur by leaky scanning. We should also mention that introduction of a UAG-stop two codons after the CUG abolishes Myc1 synthesis, but has no effect on synthesis of Myc2 (Hann et al., 1988). This is inconsistent with leaky scanning, which would have predicted the release of myc mRNA by at least a fraction of the ribosomes. Our results with the E1CAT reporter gave a similar result, since a competing initiation at the CUG start site should result in an out-of-frame peptide that terminates after 71 residues. This translational derailment, nonetheless, had no obvious effect on the rate of initiation at CAT - AUG, indicating independence between the two modes of initiation at CUG and AUG sites. The deletion analysis also corroborates the hypothesis for a scanning/IRPE model. Both AUG and CUG products are translated more efficiently when the 5'UTR is progressively shortened up to nt 436, at which point both the CUG and AUG products diminish. This indicates that initiation at both CUG and AUG begins by scanning through an inhibitory upstream region. In fact, when a hairpin was placed in front of the best deletion construct (-394) the inhibition of translation was restored, which argues against the idea of an IRES-driven mechanism. Results from the deletions tentatively delimit the IRPE to 394 - 440 nt. The pattern of translation of 436 and particularly 441 was different from that of the smaller deletions, suggesting that translation of these transcripts has reverted to a regular scanning process. However, only with the 500 and the CUG mutated to an AUG, was the CAT - AUG product completely eliminated.

In conclusion, we propose a model for translation initiation in c-myc where a ribosome enters the mRNA from the 5'UTR and scans until it reaches the IRPE (Figure 4c). The ribosome then pauses and will either resume scanning through the IRPE structure (with the help of excess eIF4E/F), or it is repositioned at the internal AUG. In cells that overexpress eIF4E the scanning machinery may prevail, allowing for increased initiation at the CUG codon and decreasing initiation at the canonical AUG. In normal cells, initiation at the AUG (Myc2) is prevalent, although different physiological conditions may alter the ratio of Myc1 and Myc2 in response to growth or differentiation signals. This mechanism of translational regulation may be shared by other regulated mRNAs with similar architecture, like FGF-2, PDGF2, and VEGF. We have evidence that this is probably the case for FGF-2 (Kevil et al., 1995; Nathan et al., 1997). It is possible that initiation via IRPEs may be quite common. There is no firm evidence to conclude that scanning is a linear, continuous process, as opposed to a discontinuous 5' to 3' migration and alignment of initiation complexes with structural elements along the mRNA 5'UTR. Indeed, a very similar mechanism, with very similar experimental evidence, was recently proposed for the translation of the Y Sendai virus proteins (Latorre et al., 1998), although this mechanism has never before been shown for the translation of a cellular mRNA, like c-myc.

Materials and methods

Plasmids and vectors

E1CAT was constructed by inserting an XhoI/HindIII fragment containing exon I of c-myc (from P1) from pGEM-Myc1 upstream of the CAT ORF in a BSCAT construct made by subcloning CAT into Bluescript KS-(Stratagene). To place the CUG codon of exon I in-frame with the CAT - AUG, the E1CATsh1 construct was created by inserting a HindIII/SmaI linker in the HindIII site of the E1CAT. A stable hairpin structure (-73 Kcal/mol) in E1CATsh4 was obtained by inserting four HindIII/SmaI adapters in tandem into the HindIII site of E1CAT. The p11sh1 and p11sh4 episomal vectors were created by subcloning the E1CATsh1 and E1CATsh4 reporters, cut with BamHI and XhoI, into the Bg/II-XhoI sites of pREP11. The full-length human hsp70 cDNA was subcloned in the HindIII-BamHI sites of Bluescript KS^-. Plasmid pXLJ - EMC is a bicistronic construct composed of the cyclin B2 protein and the influenza NS' ORF separated by the EMC - IRES.

Cell culture, transfection, and selection

CHO-BK and CHO-4E cells, which contain a G418 selectable marker, were previously described (De Benedetti et al., 1994). The BKV-based episomal vector confers uniform expression to mass-cultured cells. CHO-BK and CHO-4E were transfected with either p11sh1 or p11sh4 using Lipofectin. PREP11 (derived from pREP10, Invitrogen) is also episomal, thereby circumventing the need for clonal selection, and carries a Hyg^R cassette. The cells were grown in DMEM plus 10% FBS and selected with 200 g/ml G418 and 400 g/ml Hygromycin.

Mutagenesis and deletion constructs

Mutation of the upstream CUG to an AUG was performed by PCR. REVERSE+ and MUTCA2 primers were used to amplify the 5'UTR of c-myc and introduce the C to A change into the E1CATsh1 and E1CATsh4 plasmids. The mutated fragment was denatured and used as a `mega-primer' with the SK2T7 primer to make the full-length E1CATsh1/A or E1CATsh4/A, and obtaining two in-frame AUGs. The 5' primer includes a T3 RNA polymerase promoter sequence so that PCR products could be used directly for in vitro transcription reactions. The HMCA1 and SK2T7 primers were used to create a PCR fragment with mutated CUG to AUG and a full deletion of the c-myc 5'UTR. This fragment was subcloned into BamHI/HindIII sites of Bluescript using the engineered HindIII site in HMCA1 and the BamHI site at the 3' end of the PCR product. The primers are: SK2T7-5' ACT CAC TAT AGG GCG AAT TGG 3', REVERSE+-5' AAC AGC TAT GAC CAT GAT TAC GCC 3', HMCA1-5' AGC AAG CTT AGC TGC TTA GAC GAT GGA TTT TTT TCG GGT 3', MUTCA2-5'ACC CGA AAA AAA TCC ATC GTC TAA GCA GCT 3'. Details for PCR conditions will be provided upon request. All constructs were confirmed by sequencing with either Sequenase (UBI) or the fmol Sequencing Kit (Promega). The general deletion series was constructed using the Erase-a-Base system by Promega. Clones were screened analytically by PvuII digestion and further by sequencing. The -394-hpn construct was generated by inserting a synthetic hairpin (with a calculated G»-85 Kcal/mol) at the 5'-end. To introduce the hairpin, -394 was cleaved with ApaI (between the T3 promoter and the remainder of the c-myc 5'UTR). The linker 5'-GAAGCT TC GAAAGAGAGG-AAGAAAGGGAAGAGGAGAGAAGAGAGGCC-3' was ligated to both sides of the ApaI cleavage site, destroying it in the process. The ends of the linker (GAAGCTT) are self-complementary and also contain a HindIII site to simplify the screening. After the addition of dNTPs and Klenow polymerase to extend the ends and T4 DNA ligase to close the nicks, a synthetic hairpin containing 78 bp was generated. The same strategy was used to generate a control construct: BSCAT-hpn.

In vitro transcriptions and translations

In vitro transcription reactions were performed according to the manufacturer (Promega). Constructs linearized at the 3' of the inserts with either BamHI or XbaI, or PCR-generated templates, were used directly in transcription reactions with T3 RNA polymerase (Promega). Transcripts were analysed on agarose gels and their concentrations were estimated by ethidium promide staining. Typically, equal amounts (»100 ng/ml) were used to program rabbit reticulocyte lysate (RRL, Promega), except for BSCAT, which was used at 3-5 ng/ml. In all the reactions, the amount of added transcripts were not saturating for translation. Translation reactions were incubated at 30°C for 1 h, and the products, labeled with ³⁵S-methionine (TranSlabel, ICN), were separated on a 15% SDS - PAGE gel. The RRL was diluted 1 : 2 final in reactions. The signal was visualized by enhanced fluorography with DMSO - PPO.

Isolation and purification of eIF4F from RRL and recombinant eIF4E from E. coli; cleavage of eIF4G with 2A protease

The eIF4E and eIF4F were purified by m⁷GTP-Sepharose chromatography. Purity was verified by SDS - PAGE and silver staining. The eIF4F was prepared from RRL (Clemens, 1987) as described in (De Benedetti et al., 1991). Recombinant eIF4E was purified from extracts of the DL21 (DE3)pLys-S overproducing strain of E. coli. Cleavage of endogenous eIF4G in RRL with coxsackievirus (serotype B4) 2A protease, and translation in this cell-free system was described by Lamphear and Rhoads (1996). These conditions were previously shown to result in greater than 80% cleavage of eIF4G and inhibition of globin mRNA translation.

Acknowledgements

We are indebted to Nissim Hay (University of Chicago) for the human c-myc cDNA; Richard Jackson (University of Cambridge) for the pXLJ - EMC bicistronic construct; Rick Morimoto (Northwestern University) for the human hsp70 cDNA; Bhavesh Joshi (Marine Biotechnology Institute, Baltimore) for the pREP11 construct; Warren Bryce (Abbot Laboratories, North Chicago) for the rabbit c-Myc antiserum; Barry Lamphear (LSUMC) for the CVB4 protease 2A. This work was supported by NSF grant MCB9513756 and NIH grant CA69148-01A1.

Acland P, Dixon M, Peters G and Dickson C. (1990). Nature 343, 662-665. MEDLINE

Ar-Rushdi A, Nishakura K, Erickson J, Watt R, Rovera G and Croce C. (1983). Science 222, 390-393. MEDLINE

Bentley DL and Groudine M. (1986). Nature 321, 702-706. MEDLINE

Borman AW and Kean KM. (1997). Virology 237, 129-136. MEDLINE

Chen CY and Sarnow P. (1995). Science 268, 415-417. MEDLINE

Clemens MJ. (1987). In: Transcription and Translation: A practical approach, Rickwood D and Hanes BD (eds). IRL Press Ltd. Oxford, pp. 263-276.

Dalla-Favera R, Baregni M, Erikson J, Patterson D, Gallo RC and Croce C. (1982). Proc. Natl. Acad. Sci. USA 79, 7824-7827. MEDLINE

De Benedetti A, Joshi-Barve S, Rinker-Schaeffer CW and Rhoads RE. (1991). Mol. Cell. Biol. 11, 5435-5445. MEDLINE

De Benedetti A, Joshi-Barve S, Graff JR and Zimmer SG. (1994). Mol. Cell. Diff. 2, 347-371.

Dosaka-Akita H, Rosenberg RK, Minna JD and Birrer MJ. (1991). Oncogene 6, 371-378. MEDLINE

Florkiewicz RZ and Sommer A. (1989). Proc. Natl. Acad. Sci. USA 86, 3978-3981. MEDLINE

Fraser SD, Wilkes-Johnston J and Browder LW. (1996). Oncogene 12, 1223-1230. MEDLINE

Fütterer J, Kiss-László Z and Hohn T. (1993). TIBS 22, 177-181.

Gandori C and Eisenman RN. (1997). TIBS 22, 177-181. MEDLINE

Hann SR, Dixit M, Sears RC and Sealy L. (1994). Gene Dev. 8, 2441-2452.

Hann SR, King MW, Betley DL, Anderson CW and Eisenman RN. (1988). Cell 52, 185-195. MEDLINE

Jackson RJ. (1996). In: Translational Control, Hershey JWB, Mathews MB and Sonenberg N, Rickwood D and Hanes BD (eds). Cold Spring Harbor: Cold Spring Harbor Laboratory Press, pp. 71-112.

Joshi-Barve S, De Benedetti A and Rhoads RE. (1992). J. Biol. Chem. 267, 21038-21043. MEDLINE

Kevil C, Carter P Hu B and De Benedetti A. (1995). Oncogene 11, 2339-2348. MEDLINE

Koromilas AE, Lazaris-Karatzas A and Sonenberg N. (1992). EMBO J. 11, 4153-4158. MEDLINE

Kozak M. (1978). Cell 15, 1109-1123. MEDLINE

Kozak M. (1990). Proc. Natl. Acad. Sci. USA 87, 8301-8305. MEDLINE

Lamphear BJ and Rhoads RE. (1996). Biochemistry 35, 15726-15733. MEDLINE

Latorre P, Kolakofsky D and Curran J. (1998). Mol. Cell. Biol. 18, 5021-5031. MEDLINE

Lazarus P. (1992). Oncogene 7, 1037-1041. MEDLINE

Mellentin JD, Smith SD and Cleary ML. (1989). Cell 58, 77-83. MEDLINE

Mendez R, Kollmorgen G, White MF and Rhoads RE. (1997). Mol. Cell. Biol. 17, 5184-5192. MEDLINE

Nanbru C, Lafon I, Audiger S, Gensac M-C, Vagner S, Huez G and Prats A-C. (1997). J. Biol. Chem. 272, 32061-32066. MEDLINE

Nathan C-A, Carter P, Liu L, Li BD, Abreo F, Tudor A, Zimmer SG and De Benedetti A. (1997). Oncogene 15, 1087-1094. MEDLINE

Parkin N, Darveau A, Nicoloson R and Sonnenberg N. (1988). Mol. Cell Biol. 8, 2875-2883. MEDLINE

Paulin FEM, West MJ, Sullivan NF and Willis AE. (1996). Oncogene 13, 505-513. MEDLINE

Pause A and Sonnenberg N. (1993). Curr. Opin. Structural Biol. 3, 953-959.

Penn LJZ, Brooks, MW, Laufer, EM and Land H. (1990). EMBO. J. 9, 1113-1121. MEDLINE

Reddy, KS and Sulcova V. (1997). Cancer Genet. Cytogenet. 95, 206-209. MEDLINE

Rhoads, RE. (1993). J. Biol. Chem. 266, 3017-3020.

Saris CM, Domen J and Berns A. (1991). EMBO J. 10, 655-664. MEDLINE

Stonely M, Paulin FEM, Le Quesne JPC, Chappell SA and Willis AE. (1998). Oncogene 16, 423-428. MEDLINE

Taub R, Kirsh I, Morton C, Lenoir G, Swan D, Tronick S, Aaronson S and Leder P. (1982). Proc. Natl. Acad. Sci. USA 79, 7837-7841. MEDLINE

West M, Sullivan NF and Willis AE. (1995). Oncogene 11, 2515-2524. MEDLINE

West MJ, Stoneley M and Willis AE. (1998). Oncogene 17, 769-780. MEDLINE

Yueh A and Schneider RJ. (1996). Gen. Dev. 10, 1557-1567.

Zuker M. (1989). In: Methods in Enzymology, Dahlberg and Abelson (eds). Academic Press, New York., pp. 262-288.

Figure 1 Effect of the c-myc 5'UTR and eIF4E/F on translation of CAT reporters in RRL and cells overexpressing E1F4E. (a) Translations in RRL of capped CAT (lanes 1 - 3), or E1CAT transcripts (lanes 4 - 6) with the indicated amounts of either rabbit e1F4F, or recombinant e1F4E (b) are shown. (c) Translations of E1CAT, E1CATsh1 and E1CATsh4 transcripts, which show initiation at the CUG and AUG start sites as indicated by arrows. Translation of E1CATsh4 was consistently weaker in vitro (about 2 - 3-fold, and was run on a different gel. (d) Diagram depicting the structure of E1CAT, E1CATsh1, and the E1CATsh4 constructs in either Bluescript or subcloned in pREP11. (e) Western blot of CHO-BK and CHO-4E cell extracts containing either the E1CATsh1 or the E1CATsh4 reporters, after subcloning into the pREP11 vector. Total protein (20 g) extracted from the indicated cells was loaded on SDS - PAGE gels. After transfer to Immobilon-P (Millipore), the blot was probed for 1 h with anti-CAT antibody (5prime-3prime Inc.) At a dilution of 1 : 500. After washing three times in PBS plus 0.05% Triton X100, anti-rabbit peroxidase conjugated secondary antibody (Vector Labs) was used at a 1 : 1000 dilution for 1 h, and washed as before. The blot was visualized by enhanced chemiluminescence

Figure 2 (a) Diagram of the strategy to generate circular mRNA templates. For circularization, the hsp70 and E1CAT constructs were linearized with XbaI and transcribed with T3 polymerase. Annealing of the ends with the bridging oligo (5'-AGCTTTTGTTCCCTAGAACTAGTG-3') which is complementary to the transcription initiation region of T3 and to the BamHI - XbaI in the MCS, and the ligation conditions were described in (Chen and Sarnow, 1995). The very 5' and 3'-ends of the two transcripts are identical). The ligation mix was heated for 1 min at 90°C to melt-out the oligo, then chilled on ice, and the RNA was precipitated with 5 M LiCl. The transcripts were fractionated on a 2% agarose slab gel in TBE with 5 M urea to resolve the linear and circular forms. The linear and circular products were gel-purified and extracted with RNaid (Bio101), and used to program RRL. (b) Transcription/ligation and translation of hsp70 in RRL. Lane 1 in Figure 2b (left panel) shows a sample of in vitro transcribed hsp70 (symbolized by ~, for linear). An aliquot was annealed to the bridge-oligo and ligated with T4 DNA ligase (lane 3, symbolized by , for circularized). Typically 30 - 75% of the transcript becomes circularized, and migrates aberrantly (lane 3); note that conditions favor intra-molecular reactions (circularization) to inter-molecular (dimer formation), and dimers migrate significantly more slowly than the circular product. A control, without bridge-oligo, is shown in lane 2. Gel-purified linear and circular transcripts were translated in RRL (right panel b). Lane 1 is no mRNA (-); lane 2 is the linear transcript (~); lane 3 is the circularized transcript (); lane 4 was programmed with a circular transcript that was nicked by boiling (symbolized by [white circle with intersecting plus sign}). (Lane 4 was from 2 weeks autoradigraphy, instead of 2 days for lanes 1 - 3). (c) Transcription/ligation and translation of E1CAT in RRL. Lane 1 in (left panel) shows a sample of in vitro transcribed E1CAT. Lane 2 is an aliquot of the ligation. An aliquot of gel-purified circular template is shown in lane 3. As a control for circularization, an aliquot of the circular template was boiled for 5 min to nick the circle randomly and regenerate the linear form (lane 4). These transcripts were used to program RRL (c, right panel). Lane 1 is no mRNA (-). Lane 2 shows translation of the gel-purified, circularized transcript (). Lane 3 is the gel-purified, linear transcript (~). Lane 4 is straight E1CAT transcript, not gel-purified. (Note that the translation in lane 3 was cleaner, albeit slightly weaker, than that in lane 4). Lane 5 shows translation of BSCAT (1/30th the amount of E1CAT transcript in lane 4)

Figure 3 Translations in RRL treated with CVB4 2A-protease. Translation of the transcripts, indicated at top, was carried in parallel reactions in RRL treated or not with 2A-protease, as indicated by a (+). The bicistronic construct pXLJ-EMC, that expresses cyclin-B of X. laevis as the upstream cap-dependent reporter, and the influenza NS' protein as the second reporter driven by the EMCV IRES, was used as general control for these translations. The position of the CAT - CUG and CAT - AUG products (lane 2) is indicated by a (); the positions of cyclin B () and that of NS' (*) are indicated in lane 4

Figure 4 Effect of deletions and mutations on translation of CUG- and AUG-initiated CAT in RRL. (a) Translation of the various transcripts from the deletion series, or CUG to AUG mutation in RRL. The constructs sh1/A and sh4/A (lanes 1 and 2) are derived form E1CATsh1 and E1CATsh4, respectively, by site-directed mutagenesis of the CUG to AUG. Lanes 4 - 10 are translations of E1CATsh1 and progressive deletions of the 5'UTR. Lane 10 is a 500 nt deletion of the 5'UTR and the CUG - AUG mutation. The position of the CAT bands and that of a non-specific trnaslation product, that was used to normalize the results, is indicated on the left. (b) Diagram of the deletions made in the 5'UTR of the E1CATsh1. (c) Model of internal repositioning, depicting a scanning ribosome being repositioned to the internal CAT-AUG by the IRPE. The model was generated by the energy minimization analysis using Zuker's m-fold-based software (Zuker, 1989)