Involvement of the Aphthovirus RNA Region Located between the Two Functional AUGs in Start Codon Selection

Transcription

1 Virology 255, (1999) Article ID viro , available online at on Involvement of the Aphthovirus RNA Region Located between the Two Functional AUGs in Start Codon Selection Sonia López de Quinto and Encarnación Martínez-Salas 1 Centro de Biología Molecular Severo Ochoa, Consejo Superior de Investigaciones Científicas- Universidad Autónoma de Madrid, Cantoblanco Madrid, Spain Received September 10, 1998; returned to author for revision November 2, 1998; accepted January 7, 1999 Initiation of translation in picornavirus RNAs occurs internally, mediated by an element termed internal ribosome entry site (IRES). In the aphthovirus RNA, the IRES element directs translation initiation at two in-frame AUGs separated by 84 nucleotides. We have found that bicistronic constructs that contained the IRES element followed by the fragment including the aphthovirus start codons in front of the second gene mimicked the translation initiation pattern of viral RNA observed in infected cells. In those constructs, the frequency of initiation at the first AUG was increased by a sequence context that resembled the favorable consensus for cap-dependent translation, although initiation at the second site was always preferred. In addition, we have found that initiation at the second start codon was not diminished under conditions in which the first initiation codon was blocked by antisense oligonucleotide interference. Interestingly, mutations that positioned the second AUG out-of-frame with the first AUG did not interfere with the frequency of initiation at the second one. On the contrary, IRES-dependent translation initiation in bicistronic constructs lacking the sequences present between functional AUGs in the viral RNA was sensitive to the presence of out-of-frame initiator codons and hairpins in the spacer region. This remarkable difference in start codon recognition was due to the nucleotide composition of the RNA that separated the IRES from the initiator codon. Thus our results indicate that the region located in the aphthovirus RNA between functional AUGs is involved in start codon recognition, strongly favoring selection of the second start AUG as the main initiator codon Academic Press INTRODUCTION 1 To whom reprint requests should be addressed. Fax: emartinez@cbm.uam.es. The 5 noncoding region of picornavirus RNAs contains an element that directs internal initiation of translation, known as the internal ribosome entry site (IRES), whose structure is essential to the determination of activity (reviewed in Hellen and Wimmer, 1995; Jackson and Kaminski, 1995). According to their secondary structure, the picornavirus IRES are classified into three groups: type I includes enteroviruses and rhinoviruses; type II, cardioviruses and aphthoviruses; and type III, hepatoviruses. The conservation of primary sequence within the IRES elements of the Picornaviridae family is poor with the exception of a polypyrimidine tract (PolyY) (Kühn et al., 1990; Pilipenko et al., 1992) close to the 3 border of the IRES. In type I, the stretch that separates the PolyY from the authentic initiator codon is larger than that in the other types, and it has been considered as an spacer. In contrast, in type II the sequences required for IRES activity extend to the vicinity of the functional start codon (Hellen and Wimmer, 1995; Jackson and Kaminski, 1995). The nucleotide (nt) stretch located between the PolyY and the functional start codon accepts many substitutions in all the cases. For example, in foot-and-mouth disease virus (FMDV), the sole member of the aphthovirus group, this stretch contains the highest concentration of genetic heterogeneity within the IRES (Escarmís et al., 1992). Initiation of translation can occur at more than one start codon in some picornaviruses. The best example of two functional start codons is FMDV, which contains two in-frame AUGs separated by 84 nt (Beck et al., 1983). Both start codons are functional, and thus two forms of the L protein (Lab and Lb) are found during infection (Sangar et al., 1987). The amino acid sequence of the N-terminus in Lab is highly variable, suggesting that this part of the L protein is not essential, which is in agreement with the finding that Lb and Lab do not differ in proteolytic activity and specificity (Medina et al., 1993). Interestingly, the presence of the AUG that corresponds to the second initiation codon, but not the first one, is essential for viral replication (Cao et al., 1995). Two functional start codons have been also reported in other picornavirus containing type II and III IRES elements (Kaminski et al., 1994; Kong and Roos, 1991; Tesar et al., 1992). Recently, it has been shown that IRES-driven translation in modified hepatitis A virus cdnas can initiate efficiently at two start codons separated by nt (Zhang and Kaplan, 1998). Studies of the effect of size and sequence changes of /99 $30.00 Copyright 1999 by Academic Press All rights of reproduction in any form reserved. 324

2 FMDV IRES-DRIVEN TRANSLATION 325 the spacer region located between the 3 border of the IRES and the functional start codon have allowed the definition of the parameters that influence start codon selection. Thus in poliovirus, in which the functional AUG is located at position 743 although the entry site is in front of a cryptic AUG, revertants of defective mutants originated by insertion or linker scanner mutagenesis either restored the correct spacing or created a new AUG at the right distance (Pilipenko et al., 1992; Haller and Semler, 1992). A model in which a starting window defines the site where the ribosome starts translation or a scanning process depending on the presence of a start codon with a good sequence context was proposed for Theiler s murine encephalomyelitis virus (Pilipenko et al., 1994). The 11th AUG of encephalomyocarditis virus (EMCV) is the functional one, and the actual entry site is located very close to this start codon (Jackson and Kaminski, 1995). Deletions between the IRES element and the authentic AUG result in the use of the 12th AUG. In contrast, initiation at the 10th AUG occurs only if an IRES element is absent (Kaminski et al., 1994). The sequence context of the 11th AUG codon in EMCV modulates its use efficiency, requiring an A at 3 position (Davies and Kaufman, 1992). Furthermore, a poor sequence context at AUG 11th results in increased initiation at the next AUG. Conserved sequences adjacent to the initiation codon are also important for hepatitis C virus (HCV) IRES function, probably due to their role in RNA-ribosome complex stabilization (Pestova et al., 1998). However, initiation of translation does not seem to involve scanning in HCV IRES-promoted translation (Rijnbrand et al., 1996, 1997). Concerning the FMDV IRES, which has been reported to be one of the most efficient elements (Borman et al., 1997; Ramesh et al., 1996), the parameters influencing recognition frequency of initiator codons are still under study. Translation initiation of FMDV RNA is directed by an IRES element that is located several hundred of nt downstream from the 5 end of the RNA, spanning from nt 1 to 465, relative to the A of the first functional AUG in the viral RNA (Kühn et al., 1990). This highly structured cis-acting element efficiently promotes cap-independent internal initiation of translation in bicistronic constructs (Belsham and Brangwyn, 1990; Martínez-Salas et al., 1993), provided that essential structural motifs, e.g., the double-stranded structure at the base of domain 3, or the GNRA motif at a distal loop of this domain, are not perturbed (López de Quinto and Martínez-Salas, 1997; Martínez-Salas et al., 1996). We have recently shown that bicistronic constructs containing hairpins and/or out-of-frame AUGs in the spacer region that separated the 3 border of the FMDV IRES from the initiator codon reduce translation efficiency of the second cistron (López de Quinto and Martínez-Salas, 1998). With the aim of determining whether this conclusion also applied to bicistronic RNAs that resembled the viral RNA in their translation initiation zone, we studied the parameters influencing start codon recognition in constructs that contain the IRES and initiator codons positioned as in the FMDV RNA. We show here that bicistronic constructs with the IRES region extended to the second AUG mimic the pattern of translation initiation found during FMDV infections, providing a useful tool to study start codon recognition. Using these constructs, we observed that the frequency of initiation at the first functional codon depends on its sequence context, but that initiation at the second start codon is always preferred. In addition, we show that initiation at the second start codon is not diminished under conditions where the first start codon is blocked by antisense interference or in frameshift mutants that positioned the second AUG out-of-frame with the first one. RESULTS Bicistronic constructs with the IRES region extended to the second AUG mimic the pattern of translation initiation found in FMDV infections Translation of the FMDV polyprotein starts at two inframe AUG codons separated by 84 nt and is directed by an IRES element that spans from nt 1 to 465, relative to the A of the first initiator AUG. The relative use of each AUG is dependent on the viral strain. We have observed that certain viral isolates, such as VR100, showed an increased frequency of initiation at the first AUG, relative to the parental strain, C-S8c1 (Fig. 1A). Thus although 90% of the C-S8c1 RNA molecules were translated to produce the short form of the L protein (Lb) rather than the large one (Lab), the situation in VR100 was close to 50%. The sequences of C-S8c1 and VR100 RNA differ by 1% of their nt composition, with substitutions distributed all along the genome (Díez et al., 1990). In this regard, differences in mobility of Lb from C-S8c1 and VR100 could be attributed to amino acid changes found in the Lb protein (Díez, 1990). With the aim of understanding the parameters affecting start codon selection in those RNA variants, we analyzed the pattern of IRES-dependent translation initiation in bicistronic RNAs of the form CAT-IRES-luciferase, which included the 84 nt of the viral RNA located between the IRES region and the second functional AUG. For this purpose, we prepared the bicistronic constructs p lucc and p lucr, in which truncated forms of the luciferase gene could be initiated at any of the two FMDV start codons directed by their respective IRES (Fig. 1B). After the transfection of BHK-21 cells, the mobility of the two polypeptides of 149 and 121 amino acids was precisely detected. To correct for transfection efficiency, CAT protein was determined in all the extracts. Derivatives of the parental C-S8c1 viral RNA p lucc translated its RNA predominantly from the second start codon, whereas derivatives of the VR100 virus p lucr used both AUGs

3 326 LÓPEZ de QUINTO AND MARTÍNEZ-SALAS FIG. 1. Pattern of translation initiation in different FMDV isolates. (A) Autoradiogram of a 15% SDS acrylamide gel loaded with immunoprecipitated 35 S-methionine-labeled L proteins synthesized during the course of an infection of BHK-21 cells with the FMDV strains indicated at the top of each lane. (B) Diagram of the construction of the plasmid p lucc. CAT stands for chloramphenicol-acetyl transferase, the first cistron of the bicistronic RNA. The striped rectangle represents the 467 nt that conform the IRES, whereas the black rectangle corresponds to the first 84 nt of the FMDV polyprotein including the two start codons, named 1st and 2nd. The number of residues of the polypeptide encoded by the second unit of the bicistronic RNA, depending on the translation initiation codon used, is indicated. (C) Autoradiogram of a 15% SDS-polyacrylamide gel loaded with immunoprecipitated extracts of S-methionine-labeled BHK-21 cells, infected with vtf7-3 vaccinia virus and transfected with the plasmids indicated at the top of each lane. No DNA correspond to the mock-transfected cells. Arrows indicate the position of polypeptides initiated at the first or the second functional start codon, as well as the CAT protein. (Fig. 1C). Thus translation initiation in these bicistronic constructs mimicked the pattern observed with the viral RNAs during the course of an infection (see Fig. 1A). The frequency of initiation at the first FMDV functional codon depends on its sequence context, but initiation at the second start codon is always preferred Sequence analysis of p lucc and p lucr constructs, as well as their respective parental viral RNAs, revealed four substitutions in the 551-nt region that contains the IRES and the functional AUGs (Fig. 1B). Defining the A of the first functional AUG as position 1, two of these substitutions are present at positions 376 and 15. To determine whether the mutation present in the upstream region of the IRES that was previously shown to affect IRES efficiency (Martínez-Salas et al., 1993) was participating in the shift in start codon use observed in the VR100 RNA, we exchanged fragments of the plasmids containing the mutation affecting 376 position. Because construct p lucc/ 376R behaved as p lucc and construct p lucr/ 376C behaved as p lucr (Fig. 1C), we concluded that this position is not involved in the selection of translation start codon. In addition to the substitution of A to G at position 15 located in the 3 border of the PolyY, sequence changes of A to G at position 4 and of U to C at position 23 were present in p lucc and p lucr, respectively (Fig. 2A). To determine the contribution of the nt present at position 4 to start codon selection, we generated the substitutions indicated in Fig. 2A and analyzed their effect on AUG recognition in vivo, in combination with the two substitutions previously found at position 15. In each case, CAT translation efficiency was used as an internal control of each transfection assay. Comparison of the pattern of initiation in the group of mutants that changed the sequence at position 4 while maintaining that of positions 15 and 23 found in p lucc and p lucr revealed that in vivo, there was an increase in the frequency of initiation at the first start codon when a G was present at 4 (Fig. 2B). These data were fully consistent with the pattern of initiation observed in p lucc and p lucr constructs (Fig. 1C), carrying the sequence of the viral isolates. In addition, we observed that the presence of auatposition 4 exerted a strong negative effect, reducing significantly the frequency of initiation in vivo at the first initiator AUG, whereas A and C residues at this position had an intermediate effect. Finally, to assess the contribution of the residue present at position 23 to the observed shift in start codon recognition, we substituted the U present in p lucc to G and the C residue present in p lucr to U, G, or A. The results obtained were consistent with a main effect of the residue present at 4, and therefore, they have been presented together (Fig. 2B). Densitometric analysis of the data from at least three

4 FMDV IRES-DRIVEN TRANSLATION 327 FIG. 2. Effect of sequence context of the first start codon on the frequency of IRES-dependent translation initiation. (A) Summary of the nt differences in constructs p lucc and p lucr close to the first AUG and the changes introduced by site-directed mutagenesis at positions 4 and 23. (B) Autoradiogram of a 16.5% SDS-polyacrylamide gel loaded with immunoprecipitated extracts of S-methionine-labeled BHK-21 cells transfected with plasmids producing RNAs with the nt indicated at the top at positions 15, 4, and 23. Arrows indicate the position of the truncated luciferase forms initiated at the first and second functional AUGs, in addition to the CAT protein. (C) Relative frequency of initiation at the first IRES-dependent start codon. For each mutant, the percentage of the intensity of the polypeptide initiated at the first AUG has been made relative to the sum of the intensity of polypeptides initiated at the first plus the second initiator AUG, which was defined as 100%. The average from at least three independent experiments are plotted in decreasing order of initiation efficiency at the first site. Error bars correspond to the S.E.M. Sequence present in p lucc and p lucr constructs at positions 15, 4, and 23 has been indicated with C and R capital letters, respectively. (D) Average frequency of initiation at the second IRES-dependent site. Results from at least three independent experiments were normalized to the intensity of the polypeptide initiated at the second functional AUG in the construct p lucc, included in all the experiments, which was defined as 100%. The reference RNA contains the nt A at position 15,Aat 4, and U at 23, as outlined in a rectangle. For consistency, mutants have been ordered as in C. independent experiments reinforced our conclusions. When the frequency of initiation at the first functional AUG was calculated relative to the total initiation events at both AUGs, mutant sequences readily grouped around the nt present at 4, showing a decreasing gradient when the G was changed to A, C, and U (Fig. 2C). Therefore, the nt present at 4 position was a key factor in determining recognition of the first functional start codon as an initiation site. In addition, the results of the densitometric analysis suggested that the nt at position 15 may modulate the effect of 4. In mutants having agoranaatposition 4, in most of the cases agat 15 had a more positive influence on the frequency of initiation at the first start codon than an A at the same position. On the other hand, data corresponding to the group of mutants that contained the same residue at 15 and 4 while changing the sequence at 23 indicated that the residue present at the latter position may contribute to some extent to the recognition of the first AUG. Thus constructs with a G at 4, which displayed the highest frequency of initiation of the first codon, contained acoragat 23. However, a U at this position was correlated with a decrease in the recognition frequency of the first AUG (Fig. 2C). From these data, we also concluded that initiation at the second functional AUG was always preferred. Even in the best favorable context for the first start codon, its frequency of initiation never reached the 50% value of the total of first plus second AUGs (Fig. 2C). It is interesting to note that the average frequency of initiation at the second start AUG, normalized to the value observed for p lucc in each experiment, varied from 74% to 115%, at

5 328 LÓPEZ de QUINTO AND MARTÍNEZ-SALAS the most (Fig. 2D), indicating that none of the changes observed in the frequency of initiation at the first AUG interfered strongly with the recognition of the second start codon. Inhibition of initiation at the first FMDV functional codon does not interfere with initiation at the second functional AUG The results shown above raised the question of whether recognition of the second AUG was somehow dependent on the accessibility of the first start codon to the translation machinery. Antisense oligodeoxynucleotides (ODNs) form stable hybrid molecules that block biological processes such as translation (Le Tinévez et al., 1998; Wakita and Wands, 1994). We used this approach to examine their effect on translation initiation driven by the FMDV IRES element at both functional AUG codons. Thus in vitro transcribed bicistronic RNAs encoding the truncated forms of luciferase that could be initiated at any of the two FMDV start codons were used to program in vitro translation reactions in reticulocyte lysates. The use of this in vitro system has the advantage that we can control the ratio of antisense ODN to RNA in the reaction. A representative example of the results obtained using a 1:20 molar ratio of RNA to ODN is shown in Fig. 3. As observed in the lanes that do not have any ODN added, the pattern of initiation at each AUG in both types of transcripts reproduced the results observed in vivo (compare with Figs. 1 and 2). Statistical analysis of the results from several independent experiments showed that the addition of an antisense ODN complementary to the first initiator AUG (ATG1) decreased the frequency of initiation at this start codon in both constructs, p lucc and p lucr (Fig. 3). The addition of an antisense ODN complementary to the second start site (ATG2) decreased the frequency of initiation at the second initiator AUG, in agreement with previous results (Gutiérrez et al., 1993, 1994). Antisense block of the second initiator AUG also reduced the intensity of the polypeptide band initiated at the first AUG (Fig. 3), suggesting interference with progression of the polypeptide chain initiated at the first start AUG. In contrast to the constant inhibitory effect of ODNs ATG1 and ATG2, the modifications observed in the pattern of initiation with a control ODN that contained a randomized sequence of the second start AUG (ATG-R) varied up and down 1.5- to 2-fold. In addition, an ODN complementary to the PolyY was not inhibitory. To correlate the observed inhibitory effect with annealing of ODNs to the RNA used for translation, identification of the hybrid position in the target RNA was required. To this end, 32 P-labeled transcripts annealed to each of the ODNs used in the in vitro translation assay were treated with commercial RNase H. Samples were then run in denaturing gels to compare patterns of mobility, FIG. 3. Effect of antisense oligonucleotides targeted to each initiator AUG on the frequency of initiation. Autoradiogram of a 16.5% SDSpolyacrylamide gel loaded with immunoprecipitated products of 10 l of the in vitro translation reactions programmed with the RNA transcripts indicated at the top, which have been annealed to a 20-fold molar excess of the ODN indicated on each lane. ATG1 and ATG2 ODNs are complementary to the first and second initiation codons, respectively; ATG-R contains a randomized sequence of ATG2; PolyY is complementary to the PolyY of the FMDV IRES. The bottom panel shows the intensity of the CAT polypeptide present in 5 l of the corresponding crude extracts. Arrows indicate the position of the relevant polypeptides. The quantitative analysis of the relative frequency of initiation at each AUG modified by the presence of the ODNs added to the in vitro translation reaction (mean S.E.M.) is shown in the bottom. Data from at least three independent experiments, quantified in a PhosphorImager, were normalized to the intensity of the CAT polypeptide of each lane and made relative to the intensity of the polypeptide bands initiated at the first or second AUG in the absence of any ODN added to the reaction. using a sequence ladder to identify the length of cleavage products. As shown in Fig. 4A, ATG1, ATG2, and PolyY ODNs formed hybrid molecules that constituted substrates for RNase H cleavage at positions consistent with the formation of the predicted antisense-sense hybrid. No differential pattern was observed for the ATG-R ODN between RNase H and RNase H, even in overexposed films. The observation that antisense block of the first initiator AUG did not diminish the frequency of initiation at the second functional AUG (Fig. 3) suggested that recognition of the second start codon was independent of the accessibility of the first AUG to the translation machinery. To test the possibility that the initiation observed at the second AUG was due to 5 end-dependent initia-

6 FMDV IRES-DRIVEN TRANSLATION 329 moved two or three positions. However, no differential pattern was observed in the case of ATG-R and PolyY ODNs. Furthermore, undetectable amounts of the labeled transcript is substrate of other RNases activities, as deduced from the intensity of the larger bands of the transcript. FIG. 4. (A) Mapping of RNA-ODN hybrid. Representative example of a 6% acrylamide denaturing gel loaded with an aliquot corresponding to 7000 cpm/lane, showing an RNase H cleavage analysis of the indicated ODNs annealed to 32 P-labeled transcript, prepared from p lucr plasmid. Arrows depict the position of the preferred cleavage sites, as indicated by a DNA sequence run in parallel. (B) Transcript stability in in vitro translation assays. Pattern of 32 P-labeled transcript annealed to ODNs, incubated for the time (min) indicated at the top. A 6% acrylamide denaturing gel gel was loaded with an aliquot corresponding to 28,000 cpm/lane. tion caused by an RNase H cleavage of RNA-ATG1 hybrid molecules, we determined the stability of the transcripts in the mixture of reticulocytes. Thus RNA was incubated in the presence or absence of the corresponding ODN for 0 and 15 min with the reticulocyte mixture, under the same conditions used for in vitro translation. After extraction of the RNA, the samples were loaded onto denaturing acrylamide gels to analyze the pattern of mobility (Fig. 4B). As observed in the lane with no ODN added, after 15 min of incubation at 30 C, transcript molecules retained the mobility observed in the samples incubated for 0 min. Quantification of the results indicated that only 3 4% of the total RNA was processed by an endogenous RNase H activity in the case of ATG1 and ATG2 (2.9% and 4.3%, respectively). Relative to the pattern obtained with commercial RNase H, preferred cleavage position in the reticulocyte RNase H activity is Initiation at the second functional AUG in vivo is independent of the second AUG being out-of-frame with the first one Due to the strong bias to initiate at the second AUG, we tested whether positioning of both AUGs in-frame was a requisite to initiate translation correctly. To this end, we assayed constructs p lucc- and p lucr-, which contained a deletion of 1 nt at residues 22 or 23 respectively, positioning the second AUG out-offrame with the first one (Fig. 5A). As a result of these frameshift mutations, two stop codons are now present in-frame with the first functional AUG, at codons 11 and 17. Transfection of these constructs in BHK-21 cells, in parallel to their respective parental DNAs, indicated that the frequency of initiation at the second AUG was not decreased in RNAs that contained the second AUG outof-frame with the first one (Fig. 5B). Furthermore, analysis of data from three independent experiments showed that initiation at the second start codon was 129% in constructs lucc- and 118% in lucr-, relative to their respective parental constructs (Fig. 5C). Therefore, these results indicated that recognition of the second AUG as the main initiator codon was independent of whether the second functional AUG is in-frame with the first one. Interestingly, the viral RNA sequence between AUGs adopts a stem-loop conformation with an energy of 12.6 kcal/mol according to computer folding predictions, but translation initiation at the second AUG in p lucc and p lucr was not interfered. The overall secondary structure is maintained in p lucc- (Fig. 6) and p lucr- constructs, although the energy of these structures is slightly modified. Similarly, the substitutions introduced in positions 4 and 23 by site-directed mutagenesis do not modify this structure to a significant extent (data not shown) and led to energy increments between 1.6 to 0.1 kcal/mol. Results described above are in contrast to those observed in bicistronic constructs lacking the viral RNA region present between both functional start codons. Transcription from T7 or Tk promoter in pbic plasmid yields a bicistronic RNA of the form CAT-IRES-luciferase (Martínez-Salas et al., 1993). In this vector, the authentic initiator codon of the luciferase gene is preceded by 35 nt of its own untranslated region (Fig. 7A). Derivatives of pbic containing out-of-frame AUGs and/or stable hairpins between the 3 end of the FMDV IRES and the luciferase initiator codon showed a strong reduction in the luciferase translation efficiency (López de Quinto and

7 330 LÓPEZ de QUINTO AND MARTÍNEZ-SALAS FIG. 5. Effect of frameshift mutations on the initiation at the second functional AUG. (A) Nucleotide sequence of mutants bearing a deletion of 1 nt (open arrow) at positions 22 and 23 in p lucc and p lucr constructs, respectively, that resulted in the presence of UAA codons (underlined) at positions 31 and 49 in both cases. The out-of-frame start codons (lowercase) as well as all initiator AUGs and the PolyY in front of the second AUG are underlined. The parental constructs have been included for completeness. (B) Autoradiogram of a representative example of a 16.5% SDS-polyacrylamide gel loaded with immunoprecipitated extracts of S-methionine-labeled BHK-21 cells transfected with plasmid indicated at the top of each lane. (C) Average frequency of initiation at the second AUG. Results from three independent experiments were normalized to the intensity of the polypeptide initiated at the second functional AUG in the parental constructs p lucc, and p lucr, which were defined as 100% on each case. Martínez-Salas, 1998). To confirm these striking differences, we performed in vitro translation assays programmed with RNAs derived from these two different types of bicistronic constructs. As expected from the results obtained previously in vivo, luciferase translation from bicistronic transcripts prepared from construct pbic-1n, which contains one AUG out-of-frame with the luciferase initiator codon (Fig. 7A), was strongly diminished relative to the RNA derived from the plasmid pbic (Fig. 7B), leading to a reduction of luciferase expression of 92%. However, relative to their control constructs, translation initiation at the second AUG in p lucc- and p lucr- was barely affected by the presence of the out-of-frame AUG (Fig. 7C). Statistical analysis of the results from four independent experiments revealed that initiation at the second IRES-dependent AUG was 71% in p lucc- and 59% in p lucr-, relative to their respective in-frame parental constructs. Cap-dependent translation in these RNAs was monitored by the intensity of the CAT polypeptide, which was used to normalize the reticulocyte lysate activity between assays. The pattern of IRES-driven translation initiation observed in vitro is akin of the situation found in vivo (compare with Fig. 5, and see López de Quinto and Martínez-Salas, 1998). The presence of a stable hairpin with internal energy of 71.6 kcal/mol (construct pbic-6x, Fig. 7A) was highly deleterious for correct initiation at the luciferase start codon in vitro, leading to a reduction of luciferase expression of 95% (Fig. 7B), in agreement with the observations found in vivo (López de Quinto and Martínez- Salas, 1998). Taken together, these results allowed us to conclude that the viral RNA region present between the two functional AUGs in the aphthovirus RNA is involved in start codon recognition, strongly favoring selection of the second AUG. DISCUSSION Distinct FMDV viral isolates initiate translation with different frequencies at each of the two in-frame functional start codons separated by 84 nt. We show here that this pattern of initiation was mimicked by artificial RNAs containing the IRES region fused to the viral region present between functional start codons in front of the second cistron. Thus RNA produced from the construct

8 FMDV IRES-DRIVEN TRANSLATION 331 FIG. 6. Secondary structure of the FMDV RNA stretch including the functional AUGs. Predictions of secondary structure were obtained with the M-Fold program and the Squiggles output of the University of Wisconsin package. Both functional AUGs are boxed. An arrow depicts the nt deleted in p lucc-. p lucr initiated at the first initiator AUG more frequently than that transcribed from p lucc, as did their parental viral RNAs. Therefore, these bicistronic constructs constitute a suitable tool to study the parameters that influence start codon selection in the context of the viral RNA in the absence of infection. In this report, we have shown that the modified frequency of initiation at the first start codon was a direct effect of sequence context because in vitro generated substitutions that incorporated a G residue at the position 4 always initiated at the first site more frequently than those having A, C, or U. However, in constructs that contained the first AUG out-of-frame with the second functional AUG, the latter is as efficiently recognized as in constructs with both AUGs in frame. These observations are consistent with the fact that IRES-dependent initiation at the second FMDV functional AUG was preferred to the first start codon, independently of the context of the first initiator AUG. Different models for AUG selection were compatible with the results shown in this report. The finding of a clear effect of 4 position on start codon recognition is reminiscent of the scanning model proposed for capdependent translation (Grunert and Jackson, 1994; Kozak, 1987, 1997). In contrast, the strong preference for initiation at the second functional AUG even in the presence of frameshift mutations was an argument against it. This observation is particularly relevant in the case of the p lucr- construct because the first AUG in p lucr is used close to 50%. Scanning of the ribosome through the RNA after entry was proposed on the basis that two additional AUGs inserted in-frame with the natural FMDV start codons in the presence of the IRES induced a reduction in the expression of the downstream gene (Belsham, 1992). However, also in this case, there was a bias toward the utilization of the second natural initiator AUG of FMDV that was now the fourth start codon. Antisense ODNs complementary to either the first or the second start AUG efficiently blocked translation initiation from their respective target codons. According to the scanning model, the prediction is that preventing translation from the first codon should affect initiation

9 332 LÓPEZ de QUINTO AND MARTÍNEZ-SALAS FIG. 7. Comparison of the effect of out-of-frame mutations in constructs that do or do not contain the FMDV region found between functional start codons in in vitro translation assays. (A) Sequence of the spacer region between the 3 end of the IRES and the start codon of the luciferase gene in pbic. Relative to pbic, pbic-1n and pbic-6x contain one out-of-frame AUG (underlined) provided by insertion of one NcoI linker and a stable hairpin provided by the insertion of six XbaI linkers (brackets), respectively (López de Quinto and Martínez-Salas, 1998). (B) Autoradiogram of a 12% SDS-polyacrylamide gel loaded with 5 l ofthein vitro translation reactions programmed with the RNA transcripts indicated at the top. The bottom panel shows the intensity of the CAT polypeptide present in the same crude extracts. (C) Autoradiogram of a 16.5% SDS-polyacrylamide gel loaded with immunoprecipitated products of 10 l ofthein vitro translation reactions programmed with the RNA transcripts indicated at the top. The bottom panel shows the intensity of the CAT polypeptide present in 5 l of the corresponding crude extracts. Arrows indicate the position of the relevant polypeptides. from the second one. On the other hand, if direct recognition of the initiator codon is occurring, translation from the second start codon is expected to be independent of the presence of a block at the first site. Remarkably, we have observed that antisense interference of the first site did not decrease the frequency of initiation at the second one, suggesting that initiation at the second AUG is independent of the recognition of the first AUG. The limited amount of RNase H activity exhibited by the reticulocytes mixture used in the translation assay is not sufficient to explain the observed use of AUG2 during ODN ATG1 inhibition of translation. Fewer than 5% of the transcript molecules are cleaved in RNA-ATG1 or RNA- ATG2 hybrid molecules. Furthermore, trimming of the RNA-PolyY ODN hybrid molecule may lead to recognition of a cryptic AUG present at 8 position (Figs. 6 and 7A). This would lead to a decrease in initiation at the first functional AUG, a result that was not observed (Fig. 3). At present, we do not know why the PolyY ODN is not inhibitory. Several possibilities, including lack of hybrid formation or displacement by protein binding, could account for this effect. From experiments shown in Fig 4A, it is evident that this ODN hybridizes with its target at the appropriate position. In agreement with the results obtained using the PolyY ODN, there are reported examples of ODNs that are not inhibitory despite their ability to hybridize with HCV IRES transcripts (Wakita and Wands, 1994). In fact, only 6% of ODNs used in the literature are efficient inhibitors (Tu et al., 1998). Binding affinity has been correlated with biological activity of antisense ODNs (Lima et al., 1997). Conversely, potent ODNs targeted to the translation initiation codon do not promote RNase H-dependent degradation (Tu et al., 1998). Altogether, our results are compatible either with a leaky scanning process or with the existence of two alternative entry sites for each FMDV start codon. Under regular circumstances, the first AUG is mostly bypassed, and only a favorable context increases its use. The use of the second start codon when the first site is inhibited by an antisense ODN as well as the lack of reduction of initiation at the second AUG in the frameshift mutants, p lucc- and p lucr-, favors the existence of a second entry site. In this regard, substitutions of the PolyY CTTTTCCT (Fig. 5A) present 19 nt in front of the second AUG to purines led to the expression of short polypeptides originated from incorrect initiation sites (Cao et al., 1995). Compared with other IRES-dependent translation initiation, including those of EMCV and HCV (Kaminski et al., 1994; Pestova et al., 1998; Rijnbrand et al., 1996) that

10 FMDV IRES-DRIVEN TRANSLATION 333 share some of the FMDV functional characteristics, the FMDV initiation sites seems to have peculiar properties. Although the FMDV IRES extends up to the first initiation site (Kühn et al., 1990), this one is not normally used unless its sequence context is made favorable, as shown here. Furthermore, the fact that initiation at the second functional AUG in vivo is independent of whether the second AUG is out-of-frame with the first one, together with the result that inhibition of initiation at the first start codon in vitro did not interfere with initiation at the second functional AUG, suggests that the second AUG is the main IRES-dependent initiator codon. In agreement with our results, FMDV cdna mutants that had the first AUG mutated were viable but not those that contained mutations to nonstarter codons in the second functional AUG (Cao et al., 1995). Using bicistronic constructs devoid of the 84- nt region present downstream of the aphthovirus IRES, we have recently shown that IRES-dependent translation initiation was sensitive to the presence of out-of-frame AUGs or hairpin structures in close proximity to the 3 end of the FMDV IRES in vivo (López de Quinto and Martínez-Salas, 1998) and in vitro (present report). On the contrary, as shown here, in constructs that contained the FMDV 84 nt region, neither the presence of out-of-frame AUGs nor the secondary structure of this viral region reduced the ability of the translation machinery to recognize the second AUG as the main IRES-dependent initiator codon. Comparison of the sequences present in these two types of constructs (compare pbic and p lucc in Figs. 5A and 7A) reveals that the PolyY present in p lucc is absent in pbic. Although the distance of the initiator AUG is 35 nt in one case and 1 and 84 in the other, it has been shown that the distance per se is not essential because a spacer of 95 nt in the derivative pbic-d (López de Quinto and Martínez-Salas, 1998) was tolerated for IRES-dependent translation initiation. As shown by the mutants described in Fig. 2, a short distance to the IRES border (1 nt) is not a limitation to see initiation at the first AUG but, rather, an effect of consensus sequence around the first initiator codon. Furthermore, a stop codon in-frame with the first AUG in p lucr- or p lucc- in front of the second functional FMDV AUG did not interfere with initiation at the second AUG. Conversely, in the construct pbic-1n (Fig. 7A), a stop codon in-frame with the first initiator AUG (the additional out-of-frame AUG) just in front of the authentic luciferase initiator codon did not abolish inhibition of translation. This observation also applied to constructs pbic-2n, pbic-3n, and pbic-4n (López de Quinto and Martínez-Salas, 1998). Therefore, what is important to direct initiation at the second FMDV AUG is the 84-nt region, not the presence of a stop codon in-frame with the first start codon in p lucc- or p lucr-. The comparative results obtained with these constructs indicated that the viral RNA segment present between the FMDV functional AUGs is not simply an spacer in respect to IRES-driven translation initiation. In this regard, the first indication of a direct involvement of these sequences in the selection of start codon came from the observation that their presence was required to reproduce the pattern of translation initiation found in the viral RNA. Therefore, on the basis of the different behavior shown by bicistronic constructs that did or did not contain the viral sequences present between functional AUGs, we conclude that the 84-nt region located downstream of the aphthovirus IRES contains sequences actively involved in the recognition of the second AUG as the authentic FMDV initiator codon. Plasmid constructions TABLE 1 Oligonucleotide Sequences Primer Nucleotide sequence (5 3 ) a Orientation ATG2 GTTTTTGTCAACTTCCATTTTTCCTGC Antisense NR4 b CACGAGCTCAGCAGGTTTCC Sense EcoRVluc CCACCTGATATCCTTTGTATTTAA Antisense ATGluc GGAAAAATGGAAGGTTGACAAAAAC Sense ClaIluc GGGTGTTTGTAACAATATCGATTCC Antisense DEG4,23 GNATACAACTGACTGTTTTANCGC Antisense ATG1 CAGTCAGTTGTATTCATAGGGTC Antisense PolyY TGTAAAGGAAAGGGTGCCGAC Antisense a N indicates A, C, G, or T. Oligonucleotides were synthesized by Isogen Bioscience bv (Amsterdam, The Netherlands). b Primer NR4 was described previously (Martínez-Salas et al., 1996). MATERIALS AND METHODS On transcription, the constructs used in this study produce bicistronic RNAs of the form CAT-IRES-luciferase, in which translation initiation of the CAT gene is cap dependent, whereas initiation of luciferase is IRES dependent. The plasmids p lucc and p lucr, which encode a truncated form of the luciferase protein, were prepared in two steps. First, the IRES region of FMDV was extended to the second translation start codon of the viral RNA and fused to the luciferase start codon, using a PCR procedure (Higuchi et al., 1988). Briefly, C-S8c1 and VR100 RNAs (Díez et al., 1990) were subjected to reverse transcription followed by PCR amplification as described (Martínez-Salas et al., 1993), using the primers ATG2 and NR4 (see Table 1 for all primers). In parallel, most of the luciferase coding region from plasmid pbic (Martínez-Salas et al., 1993) was PCR amplified with primers EcoRVluc and ATGluc. About 0.2 pmol of each PCR product was mixed and annealed by their complementary ends present in ATG2 and ATGluc primers. Then, a second PCR round was performed using NR4 and EcoRVluc primers. The reaction products, digested with SacI and EcoRV, were ligated to the large

11 334 LÓPEZ de QUINTO AND MARTÍNEZ-SALAS fragment of plasmid pbic digested with the same enzymes, yielding plasmids pfmdlucc and pfmdlucr. Second, the 1292-bp XbaI-EcoRV luciferase fragment from pfmdlucc and pfmdlucr was deleted by consecutive digestion with EcoRV and HindIII (yielding two fragments of 1431 and 5048 bp) followed by XbaI digestion of the 1.4-kb fragment. The resulting HindIII-XbaI fragment of 192 bp was ligated via its HindIII end to the corresponding 5-kb EcoRV HindIII fragment. One-end ligation products were then blunt-ended and religated. Plasmids p lucc/ 376R and p lucr/ 376C were prepared by exchanging the 610-bp NcoI-HindIII fragments of p lucc and p lucr, respectively. Before expression analysis, the sequence of the entire length of each region under study was obtained using Sequenase or Thermosequenase (Amersham Life Sciences). The region around the first AUG in the bicistronic constructs p lucc and p lucr was subjected to sitespecific mutagenesis by PCR as described (Martínez- Salas et al., 1996). The sequences of external primers NR4 and ClaIluc and mutagenic oligonucleotide DEG4,23 are given in Table 1. The products of the second PCR were digested with HindIII and ClaI, purified by agarose gel electrophoresis, and ligated to the large fragment of p lucc and p lucr, similarly digested, to produce the constructs with the sequence described in Fig. 2. The sequence of mutants p lucc- and p lucr-, which contain a deletion of 1 residue at positions 22 and 23, respectively, is shown in Fig. 5A. Infection and transient expression assays BHK-21 cells were infected with FMDV and 35 S-methionine radiolabeled as described (Martínez-Salas and Domingo, 1995). Immunoprecipitation of the L polypeptides was carried out with a rabbit polyclonal anti-l sera, kindly provided by Dr. E. Beck. Immunoprecipitated complexes were separated in 15% SDS-acrylamide gels run on Tris-glycine buffer. BHK-21 monolayers 80 90% confluent were infected with the vaccinia virus recombinant vtf7-3 (Fuerst et al., 1986) when transcription from the T7 promoter was desired at 1 h before transfection with the relevant plasmid. Liposome-mediated transfection (Rose et al., 1991) was carried out as described (López de Quinto and Martínez- Salas, 1997). Soluble extracts were prepared by lysing the cells in 0.5% Nonidet P-40, 120 mm NaCl, and 50 mm Tris-HCl, ph 7.8, followed by centrifugation at 12,000 rpm for 5 min in a microfuge. When required, transfected BHK-21 cells were radiolabeled 20 h after transfection, during a3hpulse, with 100 Ci of [ 35 S]methionine (1190 Ci/mmol TRAN 35 S-label; ICN Biochemicals). Cells were kept for 1 h in methionine-free medium before labeling. A rabbit polyclonal anti-luciferase serum (Cortex) was used to immunoprecipitate the truncated forms of luciferase basically as described (Martínez-Salas and Domingo, 1995). Crude extracts from transfected cells were diluted 2-fold in RIPA buffer (50 mm Tris-HCl, ph 7.5, 150 mm NaCl, 1% Triton X-100, 0.1% SDS, 1% NaDOC) and allowed to bind for 4 h at 4 C to complexes of protein A Sepharose CL-4B with sera, preformed during a minimumof4hat4 C. Four washes with RIPA and 1 wash with TBS (20 mm Tris-HCl, ph 7.5, 140 mm NaCl) were carried out before the addition of disruption buffer (50 mm Tris-HCl, ph 6.8, 12% glycerol, 0.1% bromophenol blue, 2% -mercaptoethanol, and 4% SDS). Immunoprecipitated complexes were separated by electrophoresis in 16.5% SDS-polyacrylamide gels run on Tricine buffer (Shagger and von Jagow, 1987). When indicated, CAT protein was determined by immunoprecipitation with an anti-cat antibody (5 Prime 3 Prime). Translation products were quantified using a Molecular Dynamics densitometer or an IP Eraser BAS, Fujifilm, PhosphorImager. At least three independent experiments, which always included the control p lucc to normalize the data within experiments, were used for statistical analysis. CAT protein intensity as well as CAT activity, determined as described (Martínez-Salas et al., 1993), was parallel to the initiation at the second FMDV AUGinp lucc and its derivatives. In vitro translation and antisense block of initiation codons Before in vitro transcription with T7 RNA polymerase (New England Biolabs) to produce bicistronic transcripts encoding CAT and the truncated or full-length forms of luciferase, all plasmids (p lucc and pbic derivatives) were linearized with HpaI. In vitro translation of nuclease-treated reticulocyte lysates (Promega) were programmed with 0.5 g of the desired RNA, previously heated at 70 C during 5 min, and translated during 1 h at 30 C, in 25 l of 50% reticulocyte lysate in the presence of 25 Ci of [ 35 S]methionine. When indicated, the oligonucleotides ATG1, ATG2, and ATG-R (kindly provided by Dr. F. Sobrino) and PolyY (sequences given in Table 1) were mixed with the template RNA at 1:20 molar ratio before heating at 70 C. Truncated forms of luciferase were immunoprecipitated to eliminate endogenous translation backgrounds, whereas CAT and the fulllength form of luciferase were directly detected in the translation reaction. Translation products were treated with 50 g/ml RNase A, mixed with disruption buffer, and electrophoresed through 16.5% SDS-acrylamide gels run on Tricine buffer. To quantify the data, the intensity of polypeptide bands initiated at the first or the second start codons, determined as indicated above, were made relative to the intensity of the CAT polypeptide obtained in the same extract. RNA ODN hybrid mapping. The 32 P-labeled transcripts were prepared in vitro using T7 polymerase and [ - 32 P]CTP (Amersham Life Sci-

12 FMDV IRES-DRIVEN TRANSLATION 335 ences) from ClaI-linearized p lucc or p lucr plasmids to produce transcripts of 2012 nt. After denaturation for 2 min at 90 C, 14,000 cpm corresponding to 25 ng of 32 P-labeled transcripts was annealed to each of the ODNs at a molar ratio of 1:20 (as in the translation assay) in 20 mm Tris-HCl, ph 7.5, 100 mm KCl, 10 mm MgCl 2,0.1 mm DTT, and 5% (w/v) sucrose, and subsequently incubated with 0.3 unit of RNase H (GIBCO BRL) during 15 min in a final volume of 20 l. After phenol extraction, the resulting products were resolved in 6% acrylamide denaturing gels, run in parallel to a known sequence. The predicted cleavage site of RNA-ODN hybrid will be around positions (ATG2), (ATG1), and (PolyY) from the 3 end of the labeled transcript. Transcript stability The 32 P-labeled RNA (250 ng, 140,000 cpm), in the presence or absence of the indicated ODN at a molar ratio of 1:20, was incubated with the reticulocytes mixture under the same conditions used for in vitro translation. After an incubation at 30 C for 0 and 15 min, samples were diluted in 10 mm Tris, ph 7.5, 1 mm EDTA, and 0.1% SDS, and the RNA was extracted using 5 volumes of guanidine thiocyanate-phenol using Tripure reagent (Boehringer-Mannheim). Aliquots of extracted RNA were loaded in 6% denaturing acrylamide gels in parallel to a DNA sequence. ACKNOWLEDGMENTS We are grateful to E. Beck for providing the anti-l sera, to E. Domingo in whose laboratory the immunoprecipitation of the L polypeptides from infected cells was performed, and to F. Sobrino for the gift of ATG-R ODN. We thank C. Gutiérrez and J. P. García-Ruiz for their continuous support and encouragement. We also thank C. Gutiérrez for helpful suggestions on the manuscript. This work was supported by grants PM from DGES and 08.2/0024/97 from CAM and by an Institutional grant from Fundación Ramón Areces. REFERENCES Beck, E., Forss, S., Strebel, K., Cattaneo R., and Feil, G. (1983). Structure of the FMDV translation initiation site and of the structural proteins. Nucleic Acid Res. 11, Belsham, G. J. (1992). Dual initiation sites of protein synthesis on foot-and-mouth disease virus RNA are selected following internal entry and scanning of ribosomes in vivo. EMBO J. 11, Belsham, G. J., and Brangwyn, J. K. (1990). A region of the 5 noncoding region of foot-and-mouth disease virus RNA directs efficient internal initiation of protein synthesis within cells: Involvement with the role of L protease in translational control. J. Virol. 64, Borman, A. M., Le Mercier, P., Girard, M., and Kean, K. M. (1997). Comparison of picornaviral IRES-driven internal initiation of translation in cultured cells of different origins. Nucleic Acid Res. 25, Cao, X., Bergman, I. E., Füllkrug, R., and Beck, E. (1995). Functional analysis of the two alternative translation initiation sites of foot-andmouth disease virus. J. Virol. 69, Davies, M. V., and Kaufman, R. J. (1992). The sequence context of the initiation codon in the encephalomyocarditis virus leader modulates efficiency of internal translation initiation. J. Virol. 66, Díez, J. (1990). PhD dissertation. Universidad Autónoma de Madrid. Díez, J., Dávila, M., Escarmís, C., Mateu, M. G., Domínguez, J., Pérez, J. J., Giralt, E., Melero, J. A., and Domingo, E. (1990). Unique amino acid substitutions in the capsid proteins of foot-and-mouth disease virus from a persistent infection in cell culture. J. Virol. 64, Escarmís, C., Toja, M., Medina, M., and Domingo, E. (1992). Modifications of the 5 untranslated region of foot-and-mouth disease virus after prolonged persistence in cell culture. Virus Res. 26, Fuerst, T. R., Niles, E. G., Studier, F. W., and Moss, B. (1986). Eukaryotic transient expression system based on recombinant vaccinia virus that synthesizes bacteriophage T7 RNA polymerase. Proc. Natl. Acad. Sci. USA 83, Grunert, S., and Jackson, R. J. (1994). The immediate downstream codon strongly influences the efficiency of utilization of eukaryotic translation initiation codons. EMBO J. 15, Gutiérrez, A., Martínez-Salas, E., Pintado, B., and Sobrino, F. (1994). Specific inhibition of aphthovirus infection by RNAs transcribed from both 5 and 3 noncoding regions. J. Virol. 68, Gutiérrez, A., Rodríguez, A., Pintado, B., and Sobrino, F. (1993). Transient inhibition of foot-and-mouth disease virus infection of BHK-21 cells by antisense oligonucleotides directed against the second functional initiator AUG. Antiviral Res. 22, Haller, A. A., and Semler, B. L. (1992). Linker scanning mutagenesis of the internal ribosome entry site of poliovirus RNA. J. Virol. 66, Hellen, C. V. T., and Wimmer, E. (1995). Translation of encephalomyocarditis virus RNA by internal ribosome entry. Curr. Top. Microbiol. Immunol. 203, Higuchi, R., Krummel, B., and Saiki, R. K. (1988). A general method of in vitro preparation and specific mutagenesis of DNA fragments: Study of protein and DNA interactions. Nucleic Acid Res. 16, Jackson, R. J., and Kaminski, A. (1995). Internal initiation of translation in eukaryotes: The picornavirus paradigm and beyond. RNA 1, Kaminski, A., Belsham, G. J., and Jackson, R. J. (1994). Translation of encephalomyocarditis virus RNA: Parameters influencing the selection of the internal initiation site. EMBO J. 13, Kong, W.-P., and Roos, R. P. (1991). Alternative translation initiation site in the DA strain of Theiler s murine encephalomyelitis virus. J. Virol. 65, Kozak, M. (1987). An analysis of 5 -noncoding sequences upstream from 699 vertebrate messenger RNAs. Nucleic Acid Res. 15, Kozak, M. (1997). Recognition of AUG and alternative initiator codons is augmented by G in position 4 but is not generally affected by the nucleotides in positions 5 and 6. EMBO J. 16, Kühn, R., Luz, N., and Beck, E. (1990). Functional analysis of the internal translation initiation site of foot-and-mouth disease virus. J. Virol. 64, Le Tinévez, R., Mishra, R. K., and Toulmé, J. J. (1998). Selective inhibition of cell-free translation by oligonucleotides targeted to an mrna hairpin structure. Nucleic Acids Res. 26, Lima, W. F., Brown-Driver, V., Fox, M., Hanecak, R., and Bruice, T. W (1997). Combinatorial screening and rational optimization for hybridization to folded hepatitis C virus RNA of oligonucleotides with biological antisense activity. J. Biol. Chem. 272, López de Quinto, S., and Martínez-Salas, E. (1997). Conserved structural motifs located in distal loops of the aphthovirus internal ribosome entry site domain 3 are required for internal initiation of translation. J. Virol. 71, López de Quinto, S., and Martínez-Salas, E. (1998). Parameters influencing translational efficiency in aphthovirus IRES-based bicistronic expression vectors. Gene 217, Martínez-Salas, E., and Domingo, E. (1995). Effect of expression of the aphthovirus protease 3C on viral infection and gene expression. Virology 212, Martínez-Salas, E., Regalado, M. P., and Domingo, E. (1996). Identifica-