Leaky Scanning Is the Predominant Mechanism for Translation of Human Papillomavirus Type 16 E7 Oncoprotein from E6/E7 Bicistronic mRNA

Abstract

Human papillomaviruses (HPV) are unique in that they generate mRNAs that apparently can express multiple proteins from tandemly arranged open reading frames. The mechanisms by which this is achieved are uncertain and are at odds with the basic predictions of the scanning model for translation initiation. We investigated the unorthodox mechanism by which the E6 and E7 oncoproteins from human papillomavirus type 16 (HPV-16) can be translated from a single, bicistronic mRNA. The short E6 5′ untranslated region (UTR) was shown to promote translation as efficiently as a UTR from Xenopus β-globin. Insertion of a secondary structural element into the UTR inhibited both E6 and E7 expression, suggesting that E7 expression depends on ribosomal scanning from the 5′ end of the mRNA. E7 translation was found to be cap dependent, but E6 was more dependent on capping and eIF4F activity than E7. Insertion of secondary structural elements at various points in the region upstream of E7 profoundly inhibited translation, indicating that scanning was probably continuous. Insertion of the E6 region between Renilla and firefly luciferase genes revealed little or no internal ribosomal entry site activity. However when E6 was located at the 5′ end of the mRNA, it permitted over 100-fold-higher levels of downstream cistron translation than did the Renilla open reading frame. Internal AUGs in the E6 region with strong or intermediate Kozak sequence contexts were unable to inhibit E7 translation, but initiation at the E7 AUG was efficient and accurate. These data support a model in which E7 translation is facilitated by an extreme degree of leaky scanning, requiring the negotiation of 13 upstream AUGs. Ribosomal initiation complexes which fail to initiate at the E6 start codon can scan through to the E7 AUG without initiating translation, but competence to initiate is achieved once the E7 AUG is reached. These findings suggest that the E6 region of HPV-16 comprises features that sponsor both translation of the E6 protein and enhancement of translation at a downstream site.

The majority of eukaryotic mRNAs are monocistronic, that is, they encode only a single functional open reading frame (ORF). A few viral and cellular mRNAs have extended 5′ untranslated leader sequences (5′UTR) which can contain numerous small ORFs. These types of mRNA present a problem for an understanding of the translational machinery because, according to the scanning model for translational initiation (29, 31, 32), ribosomal subunits contact the 5′ end of the mRNA (assisted by the cap structure and its binding proteins) and then scan in a 5′ to 3′ direction until they encounter a suitable start codon, at which point they initiate translation. Upon termination of translation, ribosomes are not immediately competent to reinitiate translation. Numerous examples have been reported in which mRNA translation does not conform to the basic predictions of the scanning model because the leader sequences are long and contain AUG codons upstream of the primary ORF (reviewed in references 17, 23, and 39). Several different mechanisms have been described by which the translation complexes are able to negotiate leader sequences containing upstream AUGs (uAUGs). If upstream start codons are in a sequence context which is a poor match for the Kozak consensus A/GCCATGG (6, 26, 30), then a proportion of scanning complexes may fail to initiate at the AUG and continue scanning to the next AUG. This is known as leaky scanning. Alternatively, initiation may occur at the AUG of an upstream ORF (uORF), but following termination a proportion of the ribosomal 40S subunits remain attached to the template and resume scanning, gradually regaining competence to reinitiate translation at a downstream site. This mechanism is known as termination-reinitiation. Ribosomes may also bypass the 5′ end of the mRNA altogether and initiate from internal entry sites. Internal ribosomal entry sites (IRES) have been demonstrated in picornaviruses, hepatitis C virus, and several cellular genes (23). Another mechanism, known as ribosomal shunting, has been observed in cauliflower mosaic virus, adenovirus, and Sendai virus, in which ribosomal initiation complexes first contact the 5′ end of the mRNA, scan for a short distance, and then translocate to a remote position without scanning through the intervening sequences (11, 16, 35, 56, 57, 73).

In most circumstances, the presence of a uORF inhibits initiation at downstream AUGs, and it often appears that the sole function of the uORF is to regulate expression of the primary ORF of the mRNA (17). Human papillomaviruses (HPVs), on the other hand, are unique in that many of their mRNAs appear to be truly multifunctional, that is, they code for more than one functional protein through independent, tandemly arranged ORFs (2, 5, 22, 51, 69). The mechanisms by which these mRNAs are translated and the implications of this arrangement for posttranscriptional gene regulation are as yet poorly understood.

HPVs are small, double-stranded DNA viruses which infect cutaneous or mucosal epithelia. Some types of HPV infect genital mucosal epithelia, giving rise to genital warts or cervical intraepithelial neoplasia (CIN). While both conditions present significant medical challenges, CIN is important because it is a precursor lesion to invasive carcinoma of the cervix. HPV type 16 (HPV-16) and related types are classed as high-risk viruses because of their association with high-grade CIN and cervical carcinoma (reviewed in reference 70). Transcriptional analysis of HPV infections from isolated lesions, cell lines, and organotypic raft cultures has revealed that most HPV mRNAs are bicistronic or polycistronic (1). This has generated difficulties with understanding the HPV life cycle because it is not possible to infer that a specific viral protein is expressed by simply observing the abundance of mRNAs which contain the ORF. With the two major oncoproteins, E6 and E7, the implications of polycistronic transcripts are more far reaching. High-risk HPVs use a single promoter to drive expression of mRNAs containing both E6 and E7, while in low-risk viruses each of the two genes has its own promoter (65). In a study of the high-risk HPV-16, we showed previously that E7 synthesis is attenuated by the requirement for translation from E6/E7 bicistronic mRNAs generated by the P₉₇ promoter (69). E7 synthesis was restricted whether or not the transcript had undergone a differential splicing event in the E6 ORF (the *I splice; see Fig. 1), and we contended that the sole function of this splice is to restrict E6 expression in balance with E7. We suggested that the bicistronic E6/E7 mRNA arrangement provides a motor for the evolution of the high-risk virus types, both in the acquisition of strong transforming activities of the E6 and E7 proteins to overcome their restricted expression and in their high propensity to deregulate in a coordinate, prooncogenic manner.

FIG. 1.

(A) Structures of plasmids made to investigate the efficiency of the endogenous E6 5′UTR and to determine whether scanning from the 5′ end of the mRNA was required for both E6 and E7 translation. All inserts were under the control of the T7 promoter and terminator. Details of individual constructs are given in the text. (B) 5′UTR leader sequences from the constructs shown in A. The E6 start codon is shown in bold, and endogenous HPV-16 sequences are underlined. Transcription start points are marked by an arrow. HPV-16 refers to the endogenous leader sequence which results from transcription from the P₉₇ promoter in HPV-16. (C) Capped RNAs from the constructs indicated were synthesized in vitro and then translated in RRL in the presence of [³⁵S]cysteine, and labeled proteins were visualized by SDS-PAGE and autoradiography. (D) The relative levels of E6 expression from the various constructs were quantitated by PhosphorImager analysis and expressed relative to the E6 levels found in pNL67. Data are from 10 replicate experiments using two different RNA preparations, and error bars indicate standard errors. (E) Capped RNAs from the constructs indicated were translated as above and then subjected to RIPA using an E7 antibody, followed by SDS-PAGE and autoradiography. Sizes are shown in kilodaltons. (F) PhosphorImager quantitations of E7 expression from the indicated constructs were expressed relative to the E7 levels found in pNL67.

Despite the posttranscriptional restrictions in their expression, both E6 and E7 oncoproteins are produced in sufficient quantities to permit virus replication and, in vitro, host cell immortalization. This suggests that, in addition to inhibiting E7 translation, the E6 region may function to facilitate E7 translation from its downstream position. In this study we have analyzed the mechanism by which E7 can be translated from bicistronic RNAs encoding E6 and E7 ORFs. We concluded that the majority of ribosomes which translate E7 contact first the 5′ end of the mRNA and scan linearly through the E6 region without initiating translation at any of the 13 uAUGs preceding the E7 ORF. Once the E7 start codon is encountered, however, initiation is efficient and accurate. These results suggest that HPVs use an extreme form of leaky scanning in order to translate polycistronic mRNAs.

MATERIALS AND METHODS

Construction of recombinant clones.

The numbering system for HPV-16 is specified in reference 61. HPV-16 sequences were cloned from the archetypal Heidelberg isolate (a gift from H. zur Hausen) except for clones containing the *I splice, which were obtained from SiHa cells by reverse transcription (RT)-PCR as described previously (69).

pHET6 contains HPV-16 sequences from nucleotides (nt) 97 to 654 cloned into the transcription vector pET-3 (53). pHET7 contains HPV-16 sequences from nt 553 to 879 in pET-3. pNL67 was created from a pBluescriptII vector containing the E6/E7 region (pKE67) by insertion of an oligonucleotide linker between a KpnI vector site and the HPV-16 EcoO109I site at nt 112. The linker sequence was (top) 5′CgatatcTGCAATGTTTCAG and (bottom) 5′GTCCTGAAACATTGCAgatatcGGTAC, where the E6 ATG codon is shown in boldface. The linker contained an EcoRV site (shown in lowercase) which was used to recover the HPV E6/E7 fragment (nt 97 to 879) as an EcoRV-BamHI fragment. This fragment was inserted into transcription vector pET-7 (a gift from F. W. Studier) (53) between the StuI and BamHI sites. The T7 promoter initiates transcription at the site shown in Fig. 1B (53). pNL*I was derived from an RT-PCR clone containing HPV-16 nt 97 to 875 with the *I intron (nt 226 to 409) removed. The strategy was otherwise the same as for pNL67.

To produce pβ67, plasmid pET-7 was first modified by insertion of oligonucleotides comprising the Xenopus β-globin 5′UTR. The oligonucleotides were (top) 5′GAATACAAAGCTTGCTTGTTCTTTTTGCAGAAGCTCAGAATAAACGCTCAACTTTGGCAG and (bottom) 5′GATCCTGCCAAAGTTGAGCGTTTATTCTGAGCTTCTGCAAAAAGAACAAGCAAGCTTGTATTC and were inserted between the StuI and BamHI sites of pET-7 to produce pET7β. A BamHI site was engineered immediately preceding HPV-16 nt 97 in pKE67, and the E6/E7 region (nt 97 to 879) was recovered as a BamHI fragment and ligated into the BamHI site of pET7β. pβ*I was produced by similarly cloning the *I RT-PCR fragment into pET7β.

For pβ2^o67, pET-7 was modified by insertion of an oligonucleotide linker containing the Xenopus β-globin leader sequence and secondary structural element between the StuI and BamHI sites to produce pET7β2^o. The inserted sequence was 5′ATCGAATACAAGCTTGCTTGTTCTTTTTGCAGAGGTACCGGTACCGGTACCGGTACCGGTACCGGTACCGGTACCAGCTCAGAATAAACGCTCAACTTTGGCAGGATCC. A BamHI fragment containing the E6/E7 region (nt 97 to 879) was then inserted into the BamHI site of pET7β2^o. pβ2^o*I was produced by cloning the *I RT-PCR fragment into the BamHI site of pET7β2^o.

To produce pNL67cat, pβ67cat, and pβ2^o67cat, the chloramphenicol acetyltransferase (CAT) gene from pSV2cat was amplified by PCR using a forward primer that changed the CAT start codon to comprise an NsiI site, as found at the endogenous E7 start codon. The 3′ primer introduced a PstI site, and the CAT PCR fragment was used to replace the E7 ORF from the NsiI (nt 566) to PstI (nt 879) sites in pKE67. The resulting plasmid was used to derive pNL67cat, pβ67cat, and pβ2^o67cat using the same strategies as described above for pNL67, pβ67, and pβ2^o67, respectively. Plasmids pNL*Icat, pβ*Icat, and pβ2^o*Icat were produced using a similar strategy except that the CAT PCR fragment was used to replace the E7 ORF in the context of the *I spliced RT-PCR clone prior to cloning into the pET-7-derived transcription vectors.

pRV2^o67 was constructed by insertion of a KpnI-EcoO109I oligonucleotide containing a secondary structure with a ΔG of −62.2 kcal/mol between the KpnI (vector) and EcoO109I (nt 112) sites of pKE67. The oligonucleotide linker contained an EcoRV site 5′ to the secondary structural element. The E6/E7 region (nt 97 to 879) plus secondary structural element were excised as an EcoRV-BamHI fragment and inserted into pET-7 between the StuI and BamHI sites. The resulting sequence at the 5′ end of E6 is shown in Fig. 5B.

FIG. 5.

Use of secondary structural elements to inhibit scanning through the E6 region. (A) Diagram showing the constructs used and the locations of the secondary structures. (B) Details showing the sequence of some of the secondary structures and their locations relative to the E6 and E7 ORFs. Transcription start sites are indicated by arrows. Endogenous HPV sequences are underlined. ter, termination codon. (C) Representative PAGE gels of E6 and E7 expression of the various RNAs, translated in RRL. Sizes are shown in kilodaltons. (D) Quantitative PhosphorImager analysis of E6 (top) and E7 (bottom) levels produced by translation of each RNA in RRL. Levels are normalized relative to NL67 RNAs. The data are taken from nine independent experiments using two different RNA preparations, and bars represent standard errors. (E) The plasmids indicated were transfected into HeLa cells and activated by infection with vTF7-3. Extracts were made at 24 h posttransfection, and E7 protein was detected by Western blotting and ECL.

For pEco2^o67, the vector pKE67 was first modified by the insertion of the 5′ linker as described for pNL67. This intermediate was further modified by insertion of a secondary structure oligonucleotide (ΔG = −62.2 kcal/mol) into the EcoO109I (nt 112) site (see Fig. 5C), followed by recovery of the E6/E7 region as an EcoRV-BamHI fragment and cloning into pET-7.

To produce pBgl2^o67, PCR mutagenesis was first used to create a BglII site at nt 154 (AGAGCT changed to AGATCT) in pKE67. A secondary structural oligonucleotide (ΔG = −76.8 kcal/mol) was then inserted into the BglII site. The vector was further modified and cloned into pET-7 as described for pNL67. A similar strategy was used to produce pE6-3′2^o67, where a BglII site was created at nt 552 (AGCTGT changed to AGATCT) followed by insertion of a secondary structural oligonucleotide (ΔG = −62.2 kcal/mol) and cloning into pET-7. Note that the secondary structural element in pE6-3′2^o67 shifts the E6 termination codon two positions upstream. pBgl2^o67_TGA was cloned in the same way as pBgl2^o67 except the oligonucleotide was modified slightly to introduce a frameshift mutation, resulting in termination at nt 175. The structure of the oligonucleotide is shown in Fig. 5B. pNde2^o67 contained a secondary structure oligonucleotide with a ΔG of −62.2 kcal/mol inserted at the nt 280 NdeI site.

The core of the Kozak consensus sequence for start codons is A/GCCATGG (6, 26, 30). For the purposes of this paper, AUG codons were considered to have a strong Kozak context if there was an A at position −3 or G's at both −3 and +1. Codons were considered to have an intermediate context if they contained a G at −3 but no G at +1. All other contexts were considered to be weak. pΔ1, pΔ2, pΔ3, and pΔ123 were created by PCR mutagenesis of pKE67. In pΔ1, the initiation codon at position 148 was modified from TTATGC to TTGTGC, where the altered nucleotide is in italics, the initiation codon is in boldface, and the modified codon in the E6 ORF is underlined. In pΔ2, the start codon at nt 190 was changed from GAATGT to GAGTGT. For pΔ3, the mutation GATGGG to GCTGGG was introduced at nt 270. In pΔ123, all three mutations were combined. Note that pΔ1 and pΔ2 do not introduce amino acid changes into the E6 ORF, whereas pΔ3 results in an Asp-to-Ala substitution. The modified E6/E7 regions were recovered and cloned into pET-7 as described above for pNL67.

The dual luciferase vector pGL3R2 and the c-myc IRES vector pGL3utr were gifts from A. E. Willis (71). To produce pGLE6SD, mutations were first introduced pKE67 at nt 223 and 227 to eliminate the *I splice donor sites present in E6. The alteration was ACGTGAGGTAT to ACGCGAGCTAT, where the *I splice donor is shown in bold and the modified nucleotides are shown in italics. This mutation has been shown previously to prevent *I splicing (60). The region nt 97 to nt 658 was recovered by PCR and cloned between the EcoRI and NcoI sites of pGL3R2, as shown in Fig. 6A. The 5′ junction was gaattcAACTGCAATGTTT, where the EcoRI cloning site is shown in lowercase and the E6 AUG is in boldface. The 3′ junction was TCA GCT CCC ATG GCC, where the firefly luciferase (F-luc) initiation codon is shown in boldface and the NcoI cloning site is in italics. This junction fused Ser31 of E7 (underlined) in-frame to the F-luc ATG via an Ala-Pro linker. pΔRL67 was derived from pGLE6SD by deletion of the EcoRV-EcoRI Renilla luciferase (R-luc) fragment. Monocistronic F-luc and E7-F-luc fusion vectors, designated pΔRL and pΔRLΔ6, respectively, were produced in order to control for a potential impairment of firefly luciferase activity resulting from the fusion between the 31 N-terminal amino acids from E7 and F-luc. pΔRL was produced from pGL3R2 by deletion of the EcoRV-EcoRI R-luc fragment. pΔRLΔ6 was derived from pGLE6SD by deletion of an EcoRV-PvuII (nt 553) fragment comprising R-luc and E6. The relative activities of pΔRL and pΔRLΔ6 were used to establish a correction factor for use with E7-F-luc fusion vectors.

FIG. 6.

Dual luciferase vectors were used to screen the E6 region for IRES activity and for level of translational readthrough permitted. (A) Structures of the constructs used. R-luc, Renilla luciferase; F-luc, firefly luciferase; E7/F-luc, E7-firefly luciferase fusion ORF. The names of the constructs are indicated on the left. pGL3R2 and pGL3utr were described before (71). (B) The plasmids were transfected into HeLa cells, and R-luc and F-luc activities were determined 48 h later. Luciferase activities were normalized relative to the F-luc level from pGL3R2, which was assigned a value of 1. Data are drawn from four independent experiments, and error bars indicate standard errors.

All constructs used in this study were verified by DNA sequencing using an ABI Prism 373A automated sequencer. Constructs were propagated in Escherichia coli XL1-Blue except for those containing secondary structures, which were maintained in E. coli SURE (both strains from Stratagene).

In vitro transcription and translation.

RNA was synthesized in vitro using a Promega Ribo-Max kit according to the manufacturer's instructions. To produce capped RNA, m⁷G(5′)ppp(5′)G was added to 30 mM, and the GTP concentration was limited to 0.75 mM for 30 min at 37°C, and then the GTP concentration was increased to 7.5 mM for a further 60 min of incubation. The reaction was then treated with DNase, the RNA was purified using Qiagen RNAeasy columns, and the yield was determined spectrophotometrically. Purified RNA (25 μg/ml) was translated in 25-μl reactions in the presence of [³⁵S]cysteine (11 μCi per reaction) in 67% rabbit reticulocyte lysate (Promega Flexi-Lysate) for 60 min at 30°C. Reactions contained 70 mM added KCl unless otherwise stated and 2.0 mM Mg²⁺ (final concentration). Reactions were found to initiate accurately and were insensitive to moderate changes in RNA concentration under these conditions (data not shown). Upon termination of the reaction, phenylmethylsulfonyl fluoride (PMSF) and aprotinin were added to concentrations of 0.5 mM and 1 μg/ml, respectively. For direct visualization of labeled material, 0.5 μl of translation reaction mixture was mixed with polyacrylamide gel electrophoresis (PAGE) loading buffer and resolved by 20% PAGE, followed by autoradiography or PhosphorImager (Storm; Molecular Dynamics) analysis. For immunoprecipitation, 2 μl of translation reaction was diluted to 50 μl with radioimmunoprecipitation assay (RIPA) buffer (1% NP40, 0.5% deoxycholate, 0.1% sodium dodecyl sulfate [SDS], 50 mM Tris-HCl, 150 mM NaCl, 0.5 mM PMSF and 1 μg of aprotinin per ml). Antibody was added to 1:50 dilution followed by incubation at 4°C overnight. Protein A Sepharose beads (20 μl) were added, and incubation was continued for 1 h at 4°C. After extensive washing with RIPA buffer, the beads were boiled in 20 μl of PAGE loading buffer and resolved by 20% PAGE. Antibodies used for immunoprecipitation were E7 N-terminal antipeptide 2-18 polyclonal rabbit antibody 145-3R (67); RPαE7 rabbit polyclonal raised against an E. coli-expressed MS2-E7 fusion protein (68). Hu+ and Hu− were human sera previously characterized as positive and negative, respectively, for E7 antibodies (68). A10-60 guinea pig polyclonal E7 antipeptide 17-36 antibody (67) was also used.

Vaccinia virus infection and transfection assay.

HeLa cells (3 × 10⁵) in a 25-cm² flask were infected with 10 PFU of vaccinia virus vTF7-3 (15) (a gift from B. Moss) per cell for 60 min. Cells were then transfected with 5 μg of plasmid expression vector using 30 μl of SuperFect reagent (Qiagen) according to the manufacturer's protocol. At 24 h postinfection, cells were lysed in 150 μl of PAGE loading buffer and disrupted by passing through a Qiashredder column (Qiagen); 20 μl of lysate was resolved by SDS–15%-PAGE, followed by Western blotting to Immobilon P membrane. E7 protein was detected using mouse monoclonal antibody ED17 (Santa Cruz). The blots were developed using 1:2,000 biotinylated anti-mouse IgG (DAKO) then 1:2,000 streptavidin-horseradish peroxidase (DAKO) and revealed by enhanced chemiluminescence (ECL) (Pharmacia-Amersham).

Transfection and luciferase assay.

For transfection of dual luciferase vectors, six-well plates were inoculated with HeLa cells at 10⁵ cells/well. The following day they were transfected with 2 μg of DNA using 10 μl of SuperFect reagent (Qiagen). Cells were harvested at 48 h posttransfection, and equal amounts of protein were analyzed for luciferase activities using a Promega “Stop and Glo” dual luciferase kit. Results were normalized by assigning the F-luc activity from pGL3R2 a value of 1 and expressing all other readings relative to this value.

RNA structural modeling.

The secondary structural model for the HPV-16 mRNA sequence from nt 97 to nt 660 was produced according to the method of Zucker (25, 75) using the GCG MFOLD software. Because reliable predictions cannot be made from a sequence of this length, the sequence was broken up into segments comprising 100 nt, each succeeding segment overlapping the previous by 50 nt. Up to 20 alternative structures for each window were sought, and structures that were represented most frequently were identified. The boundaries of any stem-loop structures were identified, and each stem-loop was modeled again in isolation. Once the features of consistently predicted structures had been identified, aligned sequences of HPV-31 and HPV-18 (14) were examined for the presence of similar structures. Phylogenetically conserved base pairings were then used to constrain the model.

RESULTS

The short E6 5′UTR is fully efficient in supporting initiation at the E6 AUG.

In our previous analysis of HPV-16 E6/E7 translation, we had shown that the E7 ORF is not translated by ribosomes which have reinitiated after translating E6. We showed that ribosomes which bypass the E6 start codon are capable of proceeding efficiently to translate E7 (69). The ability of ribosomal scanning complexes to leaky-scan through the E6 AUG may be facilitated by the proximity of the E6 start codon (at nt 104) to the 5′ end of the mRNA (at nt 97). We wished to investigate whether the short E6 5′UTR allowed readthrough of the E6 start codon by leaky scanning and whether such readthrough would facilitate E7 expression from an E6/E7 bicistronic transcript. A T7 polymerase-driven expression vector was constructed in which the natural leader sequence is closely mimicked (pNL67 in Fig. 1A and B). This construction necessitated the addition of a single G residue onto the E6 5′UTR, since this is required for T7 promoter functionality (53), and an A>T transversion at position 98 to create an EcoRV restriction site. For comparison, a construct was produced in which the highly efficient 5′UTR from Xenopus β-globin mRNA (13) was cloned upstream of the E6 5′UTR (pβ67 in Fig. 1A and B). In order to study the effect of the E6 5′UTR in the context of E6*I spliced transcripts, equivalent constructs were produced in which the *I intron had been removed (pNL*I and pβ*I in Fig. 1A). As controls, monocistronic constructs were used which produce only E6 (pHET6) or only E7 (pHET7).

RNA was produced from each of these constructs by in vitro transcription, purified, and quantitated, and equivalent amounts were used to program rabbit reticulocyte lysate (RRL) translation reactions containing [³⁵S]cysteine. Samples of lysate were resolved by SDS-PAGE, and the E6 proteins were visualized by autoradiography. A representative gel is shown in Fig. 1C and the cumulative results are presented in Fig. 1D. The amount of E6 protein produced from pNL67 was equal to or slightly greater than the E6 levels produced by pβ67 (Fig. 1C, lanes 3 and 4). This indicated that the E6 5′UTR, as mimicked in pNL67, was fully efficient for loading ribosomal scanning complexes onto the E6 AUG. The E6*I protein was not detected in these experiments, supporting previous observations that this protein may have limited stability (51).

The E7 protein could not be visualized well by direct autoradiography in these experiments because its expression level is restricted when it is translated from bicistronic mRNA (69) and because an endogenous protein-tRNA complex in RRL interferes with migration. Accordingly, E7 protein synthesis was visualized by RIPA using an antipeptide antiserum (67). Analysis of E7 synthesis (Fig. 1E and F) showed that addition of the efficient β-globin leader sequence did not compromise E7 synthesis, as would have been expected if this sequence had prevented the leaky scanning of the E6 AUG. In fact, E7 synthesis was slightly decreased with the β-globin leader, but no more so than the E6 synthesis. The E7 experiments also allowed the examination of the effect of splicing. In support of our previous findings, splicing of the *I intron had little or no effect in facilitating E7 synthesis when either the NL or β-globin leader sequences were present (Fig. 1E and F).

The E6 5′UTR is scanned by ribosomes that translate the E7 ORF.

If the ribosomal initiation complexes which engage the E7 AUG scan through the E6 AUG without initiation, it would be expected that blocks to scanning located upstream of E6 would also inhibit E7 expression. The constructs pβ2^o67 and pβ2^o*I were produced, containing an element capable of forming a stable hairpin structure (ΔG = −65.4 kcal/mol) inserted into the β-globin leader sequence upstream of E6 (Fig. 1A and B). A similar element has been shown previously to present an effective block to scanning ribosomal initiation complexes (47). Insertion of this element into the E6/E7 RNA resulted in a profound inhibition of E6 synthesis, as expected (Fig. 1D and C, lane 5). Insertion of the secondary structural element was also found to inhibit E7 synthesis from both spliced and unspliced RNAs (Fig. 1F and E, lanes 14 and 17). While this inhibition did not appear to be as profound as the inhibition of E6, this may be due to the more sensitive assay used for E7 detection. We concluded that E7 was translated predominantly by ribosomal initiation complexes which scanned through the region upstream of the E6 AUG. Taken together with our previously published evidence that E7 is not translated by termination-reinitiation (69), these observations suggest that E7 is translated by ribosomes that scan in from the 5′ end of the mRNA and bypass the E6 start codon by a mechanism that does not require proximity of the E6 AUG to the 5′ end.

The E6 ORF allows a high level of readthrough to E7.

In the preceding experiments, it was not possible to assess the abundance of E7 proteins relative to E6 because different methods were required for their detection. A new set of constructs was produced in order to obtain an estimate of the levels of E6 and E7 synthesis relative to each other and thereby facilitate an analysis of the degree of readthrough permitted by the E6 region. These constructs were analogous to the series described above except that the E7 AUG was fused to the CAT gene, taking care to retain the initiation codon in the endogenous E7 context (see Materials and Methods). The increased size of the E7-CAT protein alleviated the problem of PAGE migration interference by endogenous RRL protein-tRNA complex. Translation of the E7-CAT protein was visible by direct autoradiography of the same gels as E6. This allowed, after correction for the numbers of cysteine residues present in each protein, direct determination of the relative levels of E6 and E7-CAT translation (Fig. 2). Using pNL67cat, the construct which mimics the natural E6 5′UTR, E7-CAT translation ran at a consistent 25 to 35% of E6 translation, suggesting that the E6 ORF permits a high degree of readthrough. E7 translation initiation appears therefore to be remarkably efficient, considering there are some 13 AUG codons upstream of the E7 start codon. This relative rate of initiation was maintained when the E6 5′UTR was substituted for the highly efficient β-globin leader sequence in pβ67cat (Fig. 2A, lanes 2 and 3), and splicing in pNL*Icat and pβ*Icat had little apparent effect on initiation at the E7 AUG (E7CAT in Fig. 2A, compare lanes 2 with 5 and 3 with 6). Insertion of the secondary structure into the β-globin leader inhibited both E6 and E7-CAT expression (Fig. 2, pβ2^o67cat and pβ2^o*Icat), confirming the observation that initiation at the E7 AUG depended on scanning through the UTR.

FIG. 2.

Comparison of E6 and E7-CAT translation levels. The constructs used were the same as in Fig. 1 except the E7 ORF was replaced by the CAT gene. (A) Capped RNAs were produced and translated in RRL in the presence of [³⁵S]cysteine and then visualized by SDS-PAGE and autoradiography. Note that the intensity of the E7-CAT band is underrepresented because it contains 5 cysteine residues, whereas E6 contains 14 cysteines. Sizes are shown in kilodaltons. (B) The relative levels of E6 and E7-CAT expression from the indicated constructs were determined by PhosphorImager analysis, corrected for the relative abundances of cysteine residues in the E6 and E7-CAT proteins, and expressed relative to the E6 levels found in pNL67cat. Data were drawn from 10 independent experiments using two RNA preparations. Error bars indicate standard errors.

Translation of E7 from bicistronic RNA is cap dependent.

We next investigated whether E7 translation relied upon a cap structure at the 5′ end of the bicistronic mRNA. Capped and uncapped bicistronic mRNAs were synthesized from pNL67 and then used to program RRL reactions. To some of the translation reactions, free m⁷G(5′)ppp(5′)G cap analogue was added in order to inhibit eIF4F binding to the mRNA (Fig. 3). E6 and E7 protein synthesis was detected by PAGE, RIPA, and autoradiography as above. For E6, translation of capped and uncapped transcripts resulted in an 8- to 10-fold increase in E6 synthesis (Fig. 3A). Addition of free cap analogue strongly inhibited E6 synthesis from capped but not uncapped pNL67 RNAs, as expected. Synthesis of E7 from pNL67 bicistronic transcripts was also potentiated by capping, although the magnitude of this effect was less than had been observed with E6 (Fig. 3B). This indicated that synthesis of both E6 and E7 from bicistronic transcripts is dependent on an m7G(5′)ppp(5′)G cap structure on the 5′ end of the mRNA, even though E7 has a reduced cap dependency relative to E6. This observation supported the idea that the majority of ribosomes that ultimately translate E7 first contact the 5′ end of the mRNA and are assisted by eIF4F complexes.

FIG. 3.

Cap dependency of E6 and E7 translation from bicistronic transcripts. pNL67 was used to transcribe RNA in vitro in the presence (Capped) or absence (Uncapped) of m⁷G(5′)ppp(5′)G. Both types of RNA were used to program RRL in either the presence or absence of 1 mM added soluble m⁷G(5′)ppp(5′)G. (A) [³⁵S]cysteine-labeled E6 protein was visualized by SDS-PAGE and autoradiography. (B) Labeled E7 protein was visualized by RIPA, SDS-PAGE, and autoradiography. The levels of E6 or E7 produced were expressed relative to the amounts expressed by capped transcripts translated in the absence of added soluble cap analogue. Data are from five independent experiments, and standard errors are indicated.

Heat shock and potassium chloride concentrations have differential effects on E6 and E7 synthesis.

It has been shown previously that heat shock rapidly inhibits translation due to inactivation of translational initiation factors eIF4F and eIF2. In RRL, the predominant effect of heat shock is the prevention of cap recognition by eIF4F by inactivation of its eIF4E subunit (34, 46). Moreover, with adenovirus late mRNAs, heat shock has been shown to reveal a ribosomal shunt mechanism sponsored by the tripartite leader (73). Therefore, the effects of heat shock on E6 and E7 translation were investigated, initially using pNL67-derived RNAs. Heat shock lysates were incubated at 42°C for 30 min prior to programming with RNA. As shown in Fig. 4A and B, heat shock strongly inhibited E6 expression and, to a lesser degree, E7 expression. In order to confirm this apparent difference between E6 and E7, the experiments were repeated using pNL67cat RNAs, which allowed accurate quantitation of the relative levels of E6 and E7 translation. Heat shock inhibited both E6 and E7-CAT expression (not shown), but the effect on E7-CAT expression was less pronounced than on E6 expression, so that the E7/E6 ratio increased to near unity in heat shock (Fig. 4C). These observations supported the view that E7 translation is dependent on cap recognition by eIF4F and further suggested that E7 translation has a reduced dependency on canonical translational initiation factors compared to E6.

FIG. 4.

Effects of heat shock and KCl concentration on E6 and E7 synthesis from bicistronic transcripts. Capped RNAs from pNL67 were translated in RRL as before or in RRL that had been incubated at 42°C for 30 min prior to programming. E6 (A) and E7 (B) levels were determined as previously and expressed relative to the levels in non-heat-shocked samples. (C) The experiment was conducted similarly except pNL67cat was used and the data were expressed as the ratio of E7-CAT to E6 protein synthesis. (D) pNL67-derived RNAs were translated at various concentrations of added KCl. For E6 and E7, the data were expressed relative to the level of the respective protein observed at 70 mM KCl. Data are based on five or more replicates of each experiment, and standard errors are indicated.

Cap-dependent translation typically exhibits a higher optimum KCl concentration than translation from uncapped RNAs. For capped RNA, the optimum would be expected in the region of 70 to 90 mM (24). We determined the optimum potassium concentrations for E6 and E7 translation, using pNL67 RNA as a template (Fig. 4D). For E6, the potassium optimum was estimated as 80 mM, whereas E7 translation occurred efficiently over a much broader range of KCl concentrations, with an optimum of approximately 55 mM. This resilience of E7 translation to low KCl concentrations may reflect a reduced dependency on cap binding by eIF4F.

Most ribosomes that initiate at the E7 AUG scan first through the E6 region.

We examined whether initiation at the E7 AUG was dependent upon scanning thorough the E6 region by inserting stable secondary structures upstream of the E6 AUG and within the E6 ORF itself, using pNL67 a base construct. The series of constructs is shown in Fig. 5A and B. Secondary structures of this stability are expected to block ribosomal scanning but not translating ribosomes (27, 37). RNAs were isolated and translated as before. Insertion of a secondary structure upstream of the endogenous 5′UTR in pRV2^o67 resulted in complete abolition of E6 protein synthesis (Fig. 5C, lane 4, and Fig. 5D) as anticipated. Similarly, insertion of a secondary structure 10 nt downstream of the E6 AUG in pEco2^o67 also abolished E6 synthesis (Fig. 5C, lane 5, and Fig. 5D), probably because of the proximity of the secondary structure to the start codon (27). Secondary structures inserted in the body of the E6 ORF (pBgl2^o67 and pNde2^o67) and at its 3′ end (pE63′2^o67) resulted in production of E6 protein at levels 30 to 40% of wild type (Fig. 5C, lanes 7 to 9, and Fig. 5D). This was perhaps caused by interference of E6 translation by scanning complexes that are approaching the E7 ORF and queue upstream of the scanning block (see below) (27).

E7 protein synthesis was inhibited by approximately 80% by each of the secondary structural elements inserted either upstream of E6 or within the E6 ORF (Fig. 5C and D). This means that most of the ribosomes which translate E7 arrive at the E7 AUG by scanning through the RNA from the 5′ end. It was noted, however, that placement of the secondary structure close to the E7 AUG in pE63′2^o67 was consistently more inhibitory than placements further upstream. This suggests that a small minority of ribosomes may scan only the region immediately 5′ to the E7 AUG. Alternatively, this could mean that scanning complexes increase their competence to initiate as they approach the E7 start codon and are, in this state, more susceptible to inhibition by scanning blocks. One of the constructs, pNde2^o67, was reproducibly more permissive for E7 expression than others (mean expression level of 9 experiments, 32 ± 3 compared with 19 ± 5 for pBgl2^o67). Such behavior could be indicative of a discontinuous scanning mechanism, but this would comprise only a minor component of total E7 synthesis. We favor an alternative explanation that the pNde2^o67 insertion site occurs within a region of endogenous secondary structure and is therefore reduced in effectiveness (see below).

In our previous study we provided evidence that E7 is not translated frequently by termination-reinitiation of ribosomes that have translated E6 (69). In order to investigate this point further, one of the secondary structures was modified so that it produced a frameshift and termination event in the E6 ORF (Fig. 5A and B, pBgl2^o67_TGA). This increased the space between the (truncated) E6 cistron and the E7 AUG. Such a configuration might be expected to rescue E7 synthesis if termination-reinitiation were operating but continue to inhibit E7 synthesis if scanning through the E6 region were the predominant mechanism involved (28, 38). As shown in Fig. 5C (lane 18) and 5D, placement of the secondary structure in this configuration still inhibited E7 synthesis substantially and was only marginally increased relative to the other scanning block RNAs. While features other than the distance between cistrons may affect reinitiation propensities, these observations are consistent with the interpretation that scanning through the E6 region, rather than termination-reinitiation, is the predominant mechanism for E7 synthesis.

We investigated whether RNAs containing the scanning blocks behaved similarly in cells as in RRL. The T7-driven constructs were transfected into HeLa cells followed by activation of the T7 promoters by superinfection with the recombinant vaccinia virus vTF7-3, which expresses T7 RNA polymerase. E7 protein expression was detected at 24 h posttransfection by Western blotting. As shown in Fig. 5E, the secondary structural elements were strongly inhibitory to E7 expression. This indicated that scanning through the E6 region was the major mechanism for ribosomal initiation at the E7 AUG in the cellular translation system, as in the RRL.

The E6 region does not encode an efficient IRES.

If ribosomal initiation at the E7 AUG were sponsored by an internal ribosomal initiation site as opposed to a leaky scanning mechanism, then initiation should occur independently of the presence of upstream cistrons. The E6 ORF was examined for its ability to act as an IRES for E7 translation by cloning the E6 region between the Renilla luciferase (R-luc) and firefly luciferase (F-luc) genes in vector pGL3R2 (71), producing pGLE6SD (Fig. 6A). This vector contained the entire E6 ORF and leader sequence from the transcription initiation site at nt 97 under control of the simian virus 40 (SV40) early promoter. In addition, this construct contained splice site mutations in the E6 ORF to prevent *I intron splicing. Because previous reports have implicated downstream coding sequences as integral parts of some IRES elements (50), the 3′ end of the HPV insert in pGLE6SD included the first 93 nt of the E7 ORF fused in-frame to the F-luc gene. Monocistronic control constructs containing F-luc or the E7/F-luc fusion protein were produced and used to correct for any loss of luciferase activity resulting from the inclusion of the E7 sequences in the fusion protein. As a positive control for IRES activity, we used pGL3utr, which contains the c-myc IRES inserted between the R-luc and F-luc genes (71). These constructs were used to transfect HeLa cells, and the levels of Renilla and firefly luciferase were determined at 48 h posttransfection. As shown in Fig. 6B, F-luc expression was minimal from the bicistronic vector pGL3R2, as expected. Addition of the c-myc UTR resulted in a greater than 40-fold increase in expression from the downstream F-luc cistron, in line with previously published observations (71). However, when the E6/E7 region was substituted between the R-luc and F-luc genes (pGLE6SD), a less than fourfold increase in firefly luciferase activity was observed, revealing little or no IRES activity.

We used the dual luciferase system to investigate the degree of readthrough permitted when the E6 ORF was positioned at the 5′ end of the mRNA (construct pΔRL67 in Fig. 6A), compared with the levels obtained when R-luc was positioned upstream of the reporter gene (in pGL3R2). As shown in Fig. 6B, placement of the E6 ORF upstream of the reporter allowed a greater than 100-fold-higher level of readthrough than was permitted by the R-luc ORF. Reference to the monocistronic controls allowed us to estimate that 1 in 4 ribosomes contacting the 5′ end of the pΔRL67 mRNA initiated at the E7 AUG, whereas in the R-luc-F-luc mRNA, only 1 in 500 ribosomes initiated at the AUG of the downstream cistron. The results from these transfection experiments agreed well with the estimates made using the RRL system. We concluded that although the E6 region of HPV-16 does not comprise an efficient IRES, it does allow a remarkable degree of scanning through to the E7 AUG when placed at the 5′ end of the mRNA.

Out-of-frame AUGs located within the E6 region are not recognized by scanning ribosomal initiation complexes.

We had shown previously that two minicistrons located immediately upstream of the E7 AUG had no inhibitory effect on E7 synthesis from a *I spliced bicistronic template (68). An analysis of the start codons in the E6 region showed that there are 12 cistrons internal to the E6 ORF, all but one of which are out of frame with respect to E6 (Fig. 7A). Three of these had AUGs which we categorized as having intermediate or strong matches with the Kozak consensus for efficient translational initiation. If these codons were recognized efficiently by scanning ribosomal complexes which had missed the E6 start codon, then the expected result would be a suppression of E7 synthesis. We investigated whether these AUGs were recognized and thus presented an impediment to E7 synthesis by deleting the start codons of these three ORFs, both singly and in combination. As shown in Fig. 7B and C, removal of these internal start codons had minimal effect on E7 synthesis. We concluded that the ribosomal complexes which scan beyond the E6 AUG are not competent to initiate at the three internal AUGs despite their rather good Kozak consensus matches.

FIG. 7.

Effect of ORFs within E6. (A) Diagram showing the locations of reading frames located 5′ to E7. The sequence contexts surrounding the start codons of reading frames with intermediate (I) or strong (S) Kozak consensus sequence matches are shown on the right. Mutants refers to constructs in which the indicated start codons were eliminated. (B) RNA was produced from the constructs indicated and translated in RRL. E7 protein levels were determined by RIPA. Data from 10 independent experiments were expressed relative to the levels from the monocistronic pHet7. Standard errors are indicated. (C) Representative autoradiogram from the experiments in B.

Recognition of the E7 AUG is accurate and nonleaky.

In a previous study using E7 baculovirus expression vectors, we had shown that when the E7 AUG is expressed using a synthetic leader with a poor Kozak consensus context, leaky scanning of the E7 AUG occurs. The result is a secondary initiation event at an in-frame internal methionine at position 595, producing an N-terminally truncated E7 protein (E7Δ) which can be detected by some but not all E7 antisera (68) (Fig. 8, left panel). Substitution of the endogenous context of the E7 AUG abolished synthesis of the truncated E7 product (Fig. 8, bE7Pvu). The experiments described above using RRL employed the antipeptide antibody 145-3R, which detects only the full-length E7 protein. Using several antisera that detect both full-length and E7Δ proteins, we investigated whether the lack of recognition of start codons in the E6 region was also reflected in a poor recognition of the E7 AUG. A monocistronic construct in which E7 was in its natural context (pHet7) permitted very little leaky scanning, as evidenced by only very low abundance of the E7Δ product. With the bicistronic construct pNL67, no E7Δ product was detected (Fig. 8, right panel), even on prolonged exposure of the autoradiograms (not shown). We concluded that although start codons upstream of the E7 AUG were not recognized well, ribosomal subunits which scan as far as the E7 AUG initiate accurately and do not continue their leaky scanning behavior. This observation suggests that scanning complexes may gain additional competence to initiate as they approach the E7 AUG.

FIG. 8.

Accuracy of initiation at the E7 AUG. The left panel shows E7 expression from baculoviruses in which the E7 AUG is in a weak (bE7) or strong (bE7Pvu) Kozak context. E7 protein was detected by RIPA using a seropositive human serum (Hu+), controlled with a seronegative human serum (Hu−). E7Δ indicates the N-terminally truncated product of an initiation at nt 595. The right panel shows translation in RRL of E7 RNAs from the monocistronic pHet7 and the bicistronic pNL67 constructs. E7 was detected by RIPA using antisera which are capable of detecting the E7Δ product (RPαE7, Hu+, and A10-60). The presence of an E6 band in some pNL67 tracks is due to nonspecific adsorption of E6 protein to protein A-Sepharose and to E6 seropositivity of the Hu+ antiserum. Lane M, size markers. Sizes are shown in kilodaltons.

Secondary structural model of the E6 region of HPV-16.

The secondary structure of the HPV-16 sequence from nt 97 to 660 was modeled using the method of Zuker (25, 75). Windows of 100 nt, overlapping by 50 nt, were first modeled, and the model was refined by examining each of the emergent secondary structures in isolation. The model was further refined by using phylogenetic alignment with HPV-31 and HPV-18. A graphic representation of the model is presented in Fig. 9, with the following salient features. No consistent structure could be predicted for the first approximately 160 nt, suggesting a lack of pronounced secondary structure. The region is also AT rich, consistent with a predicted low degree of secondary structure. Four stem-loops labeled 1 to 4 were consistently predicted in different plots for HPV-16 and were conserved in both HPV-31 and HPV-18. Accordingly, the model shown was constrained to form the conserved stem-loop structures 1 to 3 and to leave the first 160 nt unpaired. Other features which were consistently predicted for HPV-16 were further stem-loop structures, labeled 180, 410, and 480. A “dog leg” structure, comprising a long stem with an unpaired loop from nt 549 to 553, was present in all predicted structures for HPV-16 encompassing this region, but it was not conserved phylogenetically. This structure also included the E7 AUG at nt 562. The sites at which the secondary structures shown in Fig. 5 were inserted are indicated in the model. It can be seen that all of the inserted secondary structures with the exception of Nde2^o occurred at a position of low predicted endogenous secondary structure. The Nde2^o oligonucleotide was inserted into the descending arm of conserved stem-loop 1. This may interfere with the folding of the Nde2^o structure and could account for its slightly decreased effectiveness as a scanning block.

FIG. 9.

Model for the secondary structure of HPV-16 E6/E7 mRNA from nt 97 to 660 created using GCG Mfold software. The model was developed by scanning 100-nt windows overlapping by 50 nt, followed by phylogenetic alignment with HPV-18 and HPV-31. Loops 1 to 4 were phylogenetically conserved; loops 180, 410, and 480 and the “dog leg” were consistently predicted for HPV-16. Conserved base pairings that were used to constrain loops 1 to 3 are indicated by bold lines flanking the helix. No consistent structure was predicted for the first 160 nt, so the model was constrained to prevent the first 160 nt from pairing. The positions of the secondary structural insertions, AUG deletions, *I splice sites, and E6 and E7 start codons are indicated.

DISCUSSION

The HPV-16 genome is approximately 8 kb long, comprising eight ORFs, all encoded on the same DNA strand. The promoter locations for HPV-16 have not yet been completely defined (4, 18, 19, 65, 66), but it is clear from studies of a variety of papillomavirus types that there is not a 1:1 correspondence between promoters and ORFs (1). Therefore, HPVs must rely on posttranscriptional mechanisms to regulate expression of the various ORFs (1, 21, 59).

The structures of mRNAs from several HPV types have been defined using a variety of mapping techniques, and in all cases it appears that RNAs are produced that encode more than one ORF (7–9, 12, 20, 22, 40–44, 52, 54, 55, 62, 63). In several instances, for example E7, E2, and E5 in HPV-16, no RNAs have been defined in which the ORF is not preceded by an upstream coding ORF. Numerous investigators have noted that these polycistronic papillomavirus RNAs appear to be capable of producing more than one protein concurrently (2, 5, 22, 49, 51, 72), but the mechanisms underlying this have remained obscure. Termination-reinitiation (65, 72), leaky scanning (69), and shunting (49) have all been suggested previously as potential mechanisms.

In the high-risk genital HPV types, all known transcripts encoding E7 contain the E6 or E6* ORF upstream of the E7 start codon. In our previous study, we argued against termination-reinitiation by ribosomes that had translated the E6 ORF or its spliced variant E6*I as a mechanism by which ribosomes gain access to the E7 AUG. One of the main predictions of the termination-reinitiation hypothesis is that the *I splice would act to increase the efficiency of E7 synthesis (60, 65) because the E6 ORF is shortened and the intercistronic space has been increased (28, 38). The termination-reinitiation model is attractive, therefore, in that it provides a rationale for existence of the *I splice. The results in the present study (Fig. 1E and F) reconfirm that the *I splice has minimal effect on the efficiency of E7 synthesis in vitro and support the view that termination reinitiation does not play a major role in E7 translation. This point was further supported by experiments with the construct pBgl2^o67_TGA (Fig. 5A and B) in which a prematurely terminated E6 ORF was engineered to contain a region of secondary structure. If the predominant route to the E7 AUG is to scan rather than translate through the E6 region, then the inhibitory effects of the secondary structure ought to be revealed, as was observed in the experiments (Fig. 5D).

If the *I splice does not function to facilitate E7 expression, then what is its function? It has been reported that in transfected cells the C-terminally truncated product from the *I ORF can derepress p53 function by antagonizing the action of full-length E6 (48). However, we and others have been unable to detect the truncated E6* products immunologically even in systems where they are grossly overexpressed, suggesting that they may have limited stability (51, 67, 69). It remains uncertain whether these proteins have a physiologically relevant role. It is obvious that the *I splice prevents the bicistronic mRNA from synthesizing full-length E6 protein. Maintaining an appropriate balance between E6 and E7 expression may be a crucial function for the virus, and given the constraints of the bicistronic transcription unit, the *I splice may be the only way in which this can be achieved.

In our previous study we also showed that translation of the E6 (or E6*I) ORF is inhibitory to E7 synthesis (69). Evidence for this was that deletion of the E6 AUG potentiated E7 translation while forcing efficient translation of E6 by embedding its start codon in an encephalomyocarditis virus leader sequence resulted in reduced E7 translation. This could indicate that E7 synthesis requires leaky scanning of the E6 AUG, even though the Kozak consensus match of the E6 AUG is of intermediate strength. Alternatively, it could be that translation through the E6 ORF disrupts an IRES or other sequence element required for E7 translation.

Although the presence of the E6 ORF upstream of E7 causes an attenuation of E7 synthesis, one of the more interesting findings of the current study is how efficient E7 synthesis actually is despite the presence of a functional, coding ORF upstream. The E6 region of HPV-16 is extremely permissive for readthrough to the E7 AUG, allowing the E7 protein to be synthesized at a rate of approximately 25 to 35% of the E6 level. This permissivity was demonstrated both in RRL and in transfected HeLa cells. By comparison, an unrelated cistron (Renilla luciferase) resulted in a readthrough of only 0.2% to a downstream cistron (Fig. 6). The E6 region must have properties which facilitate initiation at downstream sites. The model that is most consistent with the results presented here is one involving an extreme leaky scanning mechanism. Ribosomal initiation complexes bind first to the 5′ end of the mRNA and then scan through the E6 region without initiating translation at the 13 AUG codons they encounter until E7 is reached. Once reached, initiation at the E7 AUG is efficient and accurate.

The high levels of readthrough to E7 were shown to be dependent upon interactions with the 5′ end of the mRNA. Evidence for this was that synthesis of E7 was potentiated by capping of the transcript and was inhibited by soluble cap analogues. Heat shock, which in RRL depletes factors necessary for binding the cap structure (eIF4F) (34, 45), inhibited E7 translation. Moreover, insertion of stable secondary structures in close proximity to the 5′ ends of either the β-globin 5′UTR or the endogenous leader sequence resulted in inhibition of E7 expression as well as that of E6. This showed that the majority of ribosomes which translate E7 interact first with the cap structure at the 5′ end of the mRNA and scan at least a short distance along the RNA. These observations are also consistent with the inability of the E6 region to demonstrate significant IRES activity in the dual luciferase assay (Fig. 6), where high readthrough to E7 depended on the E6 ORF being closest to the 5′ end of the RNA.

Ribosomal shunting, or discontinuous scanning, occurs when ribosomal initiation complexes bind first to the 5′ end of an mRNA, scan for a short distance, and then translocate to a site further downstream where they can resume scanning or initiate translation (16, 35, 73). In operational terms, shunting can be differentiated from scanning by the ability of shunting ribosomes to bypass an inserted secondary structural element which would inhibit scanning ribosomes (23). We inserted stable secondary structures at a number of locations within the E6 5′UTR and the E6 ORF. At each location, the secondary structures profoundly inhibited the level of E7 translation. This indicated that the majority of ribosomes which translate E7 scan continuously through the E6 region without initiating translation.

Some of the observations made in this study do not fit well with predictions drawn from a simple version of the leaky-scanning hypothesis. First, one would expect that E6 synthesis must be compromised in order to permit scanning ribosomes to pass through to E7. Previous studies have shown that when 5′UTR regions are short, they can promote leaky scanning of the first AUG (33, 64). The structure of the E6 primary transcript has a very short (7 nt) leader sequence. This setting, along with the intermediate-strength Kozak consensus context, would seem to suggest that the E6 AUG is susceptible to leaky scanning. In HPV-18 and HPV-31, the E6 AUG is also found in close proximity to the 5′ end of the respective mRNAs (20, 41, 58), leading some authors to assume that E6 cannot be translated from such structures (49). However, when a proven-efficient, unstructured leader sequence from the β-globin gene (13) was added the HPV-16 E6 5′UTR, we observed little change in the efficiency of E6 AUG utilization in RRL. It was not possible to test directly the behavior of the endogenous E6 5′UTR in cellular transfections because the low activity of the HPV-16 P₉₇ promoter in such assays necessitated the use of a heterologous (SV40 early) promoter producing a fusion mRNA. Nevertheless, in HeLa cells, the E6 region was found to promote high-level readthrough to E7 even though it was preceded by an SV40 leader sequence from the vector. It therefore appears that the E6 5′UTR is fully capable of promoting initiation at the E6 AUG and that, if E7 synthesis does indeed depend on leaky scanning of the E6 AUG, then it depends on features of the E6 start codon other than proximity to the 5′ end of the mRNA.

A second set of observations which is inconsistent with a simple leaky-scanning hypothesis concerns the role of upstream AUG codons. If ribosomal scanning complexes manage to miss the E6 AUG, there are still 12 more AUGs to encounter before the E7 AUG. In this study we examined the three of these which have start codons in Kozak contexts that we classified as intermediate or strong. Our data suggest that these cistrons are usually ignored, probably by ribosomes scanning past them. We reached a similar conclusion regarding the two cistrons (MC1 and MC2) located immediately upstream of E7 (69). This suggests that leaky scanning is promoted throughout the length of the region upstream of the E7 AUG. In the closely related HPV-31, there are eight AUG codons upstream of E7, of which three are in strong Kozak contexts and three are in intermediate contexts. These observations suggest that the extreme degree of readthrough permitted by the region upstream of E7 is not simply a function of the immediate sequence context of the upstream AUGs. Apparently neutral uAUGs have been observed previously: an optimal-context AUG codon in the cytomegalovirus pp150 gene 5′UTR has no apparent effect on downstream translation regardless of the intercistronic spacing. The mechanism for bypass in this case has not been determined (3).

Although E7 translation demonstrated a 5′ end, cap, and eIF4F dependency, the magnitude of dependency was always lower for E7 than for E6. In some experiments, such as heat shock using pNL67cat, the E7/E6 ratio increased to almost unity. A possible explanation for this comes from examination of the secondary structural model shown in Fig. 9. This model predicted that the first 160 nt of the bicistronic mRNA are relatively unstructured. Previous studies have shown that a long, unstructured region at the 5′ end of an mRNA can result in a reduced dependency on cap recognition by eIF4F (10, 36, 74). Most probably, 40S ribosomal subunits cannot attach to a 5′ end efficiently when extensive secondary structure is present adjacent to the loading site. In this case the attachment depends on the import of eIF4F-associated helicase activities to unwind the structure. However, when little or no secondary structure is present in proximity to the 5′ end, 40S subunits can attach without extensive reliance on eIF4F-associated helicases. It is possible, therefore, that E7 is less reliant on 5′-end binding by eIF4F because its mRNA has, in effect, a long leader sequence that has a relatively low level of secondary structure surrounding the 5′ end.

We also noted that secondary structures placed very close to the E7 AUG (in pE63′2^o67) were two- to threefold more inhibitory than similar structures placed further upstream. Moreover, when secondary structures were placed upstream of both E6 and E7 cistrons (in pβ2^o67 and pRV2^o67), E6 always appeared to be more strongly inhibited than E7. This observation could indicate that a minority of the ribosomes which translate E7 undergo an internal initiation event, scanning only the region immediately upstream of E7. In agreement with this, Fig. 6 shows that the IRES trap vector pGLE6SD produced levels of E7-F-luc protein at levels three- to fourfold over background. Such a secondary mechanism might have an enhanced role during the natural life cycle of the virus or when cap-dependent translation becomes severely restricted. An alternative explanation may be that ribosomal scanning subunits undergo a qualitative change as they approach the E7 AUG, which results in an increased competence to initiate translation coincident with an increased susceptibility to scanning blocks. This would be consistent with the inability of scanning subunits to recognize start codons as they transit the E6 ORF.

In summary, it appears that, with respect to E7 initiation, the E6 region has properties of both a translational inhibitor and a translational enhancer. Translational inhibition, which is a necessary consequence of the requirement to express E6 from the same transcript, is compensated for by an extraordinarily high degree of leaky scanning permitted by the E6 region.