key: cord-0740390-ck0jpp7e
authors: Horsburgh, Brian C; Kollmus, Heike; Hauser, Hansjörg; Coen, Donald M
title: Translational Recoding Induced by G-Rich mRNA Sequences That Form Unusual Structures
date: 1996-09-20
journal: Cell
DOI: 10.1016/s0092-8674(00)80170-1
sha: 8feebef712cb74f395a89a5b1b6a7f782a8904b2
doc_id: 740390
cord_uid: ck0jpp7e

We investigated a herpesvirus mutant that contains a single base insertion in its thymidine kinase (tk) gene yet expresses low levels of TK via a net +1 translational recoding event. Within this mutant gene, we defined a G-rich signal that is sufficient to induce recoding. Unlike other translational recoding events, downstream RNA structures or termination codons did not stimulate recoding, and paused ribosomes were not detected. Mutational analysis indicated that specific tRNAs or codon–anticodon slippage were unlikely to account for recoding. Rather, recoding efficiency correlated with the G-richness of the signal and its ability to form unusual structures. These findings identify a mechanism of translational recoding with unique features and potential implications for clinical drug resistance and other biological systems.

The different plasmids were transcribed and translated and the extracts assayed for luciferase and ␤-galactosidase. The top line shows the construct with the T3 promoter (oval) driving ␤-galactosidase coding sequences, which are followed by a frameshift insert and luciferase coding sequences. Below are the sequences in the different plasmids tested. Arrows show where the HSV sequences were replaced. To the right of each plasmid are the mean values and standard deviations for the percentage of luciferase activity relative to ␤-galactosidase activity normalized to an in-frame control (BCH1). D.S.S.: downstream structure. Sa, Bg, and Ba represent SalI, BglII, and BamHI sites, respectively (see text for cloning details). Boxes represent termination codons. that the mechanism is unique and may depend on the create T3 LucLac (Figure 1) . A 207 bp SnaBI-SacI bluntended fragment from pTK-4, containing mutant tk se-ability of G-rich sequences to form unusual inter-or intramolecular structures. The results may also have quences sufficient for recoding (Experimental Procedures), was then cloned between the reporter genes, implications for antiviral drug resistance and for expression of other genes.

replacing the HIV slippery sequence (Figure 1 ). This plasmid was called BCH2. Following transcription with T3 RNA polymerase and translation of the synthetic mRNA Results in reticulocyte lysates, luciferase activity from the ␤-galactosidase-luciferase fusion protein can be de-A Reporter Gene System to Analyze Net ϩ1 tected when a recoding event that shifts translation into Recoding In Vitro the ϩ1 frame occurs. The relative enzymatic activities of Previously, we identified a net ϩ1 translational recoding ␤-galactosidase and luciferase reflect the frameshifting event within the tk gene from an ACV-resistant HSV efficiency. To normalize this efficiency, we constructed mutant (Hwang et al., 1994) . Because this recoding event two control plasmids. Reference plasmid, BCH1, was was not very efficient (1%-2%) and was difficult to quancreated by cloning the corresponding wild-type tk setify, we decided to make use of a sensitive and quantitaquences from pTK-wt (Experimental Procedures) betive enzyme reporter system that has been used to intween the reporter genes ( Figure 1 ). This plasmid exvestigate the role of human immunodeficiency virus presses an in-frame ␤-galactosidase-luciferase fusion. (HIV) and human T cell leukemia virus II sequences in Its ratio of luciferase activity to ␤-galactosidase activity frameshifting in transfected cells (Kollmus et al., 1994;  was set at 100%. A negative control plasmid, BCH0, Reil et al., 1993) . To adapt this system for in vitro analywas created. It contained luciferase-coding sequences sis, the reporter genes encoding ␤-galactosidase and in the ϩ1 reading frame relative to ␤-galactosidase luciferase separated by the HIV elements were placed under the control of bacteriophage T3 promoter to but lacked any HSV sequence. This allowed subtraction

In cases in which the corrected values fell below 0.2%, we were unable to detect fusion protein (data not shown).

A 10 Base G-Rich Sequence from the Mutant tk Gene Is Sufficient for Recoding It seemed probable that the recoding event occurred between the first upstream termination codon in the ϩ1 reading frame and the SacI site (Hwang et al., 1994) . Therefore, HSV sequences upstream of this termination codon were deleted, creating pBCH2a. This plasmid, containing 46 nt of HSV sequence, could direct frameshifting at standard levels ( Figure 1) . A further 5Ј deletion of 30 nt yielded plasmid BCH4. The 16 nt of HSV sequence contained within this plasmid was sufficient to promote frameshifting at standard levels. However, a control plasmid, BCH4a, which contained 16 nt of a randomly generated sequence, did not frameshift (Figure 1) . To narrow the minimal signal for frameshifting, we deleted either 3 or 6 nt from the 3Ј or 5Ј end of the 16 nt HSV sequence, creating plasmids BCH5, BCH7, BCH6, and BCH8, respectively ( Figure 1 ). Deletions from the 3Ј end had no effect on frameshift efficiency, whereas 5Ј deletions abolished frameshifting. Therefore, these results indicate that the sequence G(8)AG is suffi- Figure 2 . Translation of Wild-Type and Mutant mRNAs cient to promote net ϩ1 frameshifting at standard levels (A) A representative sample of reticulocyte translation products syn-(1%). However, no recoding was detected when this thesized from synthetic mRNA derived from the mutants described motif was transplanted into a reporter system designed in Figure 1 . The number in each lane corresponds to the described to detect Ϫ1 frameshifting (data not shown). 

In many examples of ribosomal recoding, the presence of a stem-loop or pseudoknot structure acts as a posi-of background "noise" generated within the system (0.05%). For each plasmid, assays were repeated 4 to tive stimulator . A stimulatory stemloop structure derived from HIV sequences (Reil et al., more than 30 times to obtain means and standard deviations. In these assays, BCH2 reproducibly yielded a low 1993) lies 10 nt downstream of the G(8)AG motif in BCH4. However, no such structure is obvious downstream of but significant efficiency (1%) of net ϩ1 frameshifting ( Figure 1 ). This efficiency will be referred to as frame-this motif in its natural context (Hwang et al., 1994) . To test whether the HIV stem-loop affects the HSV recoding shifting at "standard levels."

To confirm that luciferase expression was due to a event, we deleted it in BCHSLϪ. This deletion had no significant effect on luciferase levels produced by net ϩ1 frameshift and not to another translational event, (e.g., internal initiation), translations were carried out in frameshifting (see Figure 1 ). Moreover, insertion of the HIV stem-loop 10 nt downstream of the G(8)AG motif in the presence of 35 S methionine and the products analyzed on SDS-polyacrylamide gels. In each case in the mutant tk gene (pTK-SL ϩ ) did not alter the levels of full-length TK relative to the truncated product (Fig-which we obtained a corrected frameshift value (the mean minus the standard deviation) of greater than or ure 2B). As reviewed in the Introduction, stimulators of eukary-equal to 0.2%, we could observe full-length ␤-galactosidase-luciferase fusion protein. (A representative sample otic frameshifting are thought to act by pausing ribosomes at the recoding site. This model has been borne is shown in Figure 2A ). We therefore consider corrected values of greater than or equal to 0.2% as meaningful.

out in the examples examined thus far (Somogyi et al., from the 44 kDa marker was due to methonine labeling of an endogenous reticulocyte polypeptide]). In contrast, a control plasmid, pPS1a (Somogyi et al., 1993 ; a gift of I. Brierley), which contains a coronavirus pseudoknot, produced a paused product of the expected size, 42 kDa (data not shown). However, as the sensitivity of these assays was limited, we cannot totally discount the possibility that a very low level pause occurs at the HSV recoding site. Nevertheless, in contrast to results obtained in other translational recoding systems, there was no observable detectable pause associated with the HSV recoding event, nor was recoding stimulated by a downstream structure believed to act by pausing ribosomes.

As the sequence defined by deletion analysis was highly G-rich, we wished to assess the importance of monotonous runs of G's (G-strings) in net ϩ1 frameshifting.

Replacing the G(8)AG motif of BCH4 with A(8)AG abolished frameshifting (Figure 1 ), indicating a requirement for G's rather than for any purines. We lengthened the run of G's from 8 in BCH4 to 11 (BCH14) or shortened These results suggest that the recoding event occurs as ribosomes encounter the run of G's. Thus far, however, all attempts to determine the site of recoding by 1993; Tu et al., 1992) . Given the lack of effect of the HIV protein sequencing have failed, primarily due to low stimulator, we asked whether paused ribosomes could frameshift efficiency (B. C. H. and S. Matusfuji, unpubbe detected during recoding in the HSV tk system, using lished data). Therefore, we decided to use site-directed the translational inhibitor edeine (Somogyi et al., 1993) . mutagenesis to address this question. In an initial at-Briefly, ribosomes were allowed to initiate on mRNA in tempt to define the site of frameshifting, we mutated the presence of 35 S methionine. Then, after a short time, the C at position 7 in BCH15 to a T, yielding plasthe drug was added to prevent further initiation. This mid BCH16, i.e., GGG-GGG-CGA-GGC-TGG-G to GGGresulted in a relatively synchronous pulse of ribosomes GGG-TGA-GGC-TGG-G. (Bold characters represent the whose products could be sampled at various times differences; T's indicated in the DNA sequence become throughout the reaction. Interruption of elongation, i.e., U's in the RNA). This mutation resulted in a termination pausing, can result in a detectable "paused" product codon (underlined) in-frame with ␤-galactosidase. This on SDS-polyacrylamide gels.

termination codon did not measurably affect frameshift-To facilitate the analysis, we cloned the EcoRV-SacI ing, since luciferase expression of BCH16 and BCH15 fragment of pTK-4 and pTK-4a (i.e., frameshift-compewere comparable ( Figure 1 ). This indicates that ribotent BCH4 and frameshift-incompetent BCH4a-consomes enter the luciferase reading frame without recogtaining sequences, respectively; Figure 1 ) into the HSV nizing the termination codon. (Also, the termination co-DNA polymerase (pol) open reading frame under the don, which is known to stimulate ϩ1 frameshifting in control of the T3 promoter, forming plasmids pBH19 other systems, did not stimulate the HSV recoding and pBH21, respectively ( Figure 3A ). pBH19 and pBH21 event.) Luciferase expression from pBCH16a, TGAwere linearized with either ScaI, which cuts downstream GGG-GGG-GGA-GGC-TGG-G, which has a TGA codon of protein coding sequences, or with EcoRI. EcoRI cuts inserted in the ␤-galactosidase reading frame immedijust 3Ј to the G-rich sequence; thus, transcription and ately upstream of the G(8) motif, was not detectable translation resulted in a labeled product of the same ( Figure 1 ). The results from BCH 16 and BCH16A, taken size as the potential paused product, approximately 44 together, imply that the frameshift at least initiates be-kDa ( Figure 3B , lane P). ScaI-digested DNA was trantween the termination codons; that is, on the G-string. scribed, the synthetic RNA translated (with edeine added following initiation), and the products electropho-Recoding Is Associated with G-Richness resed through SDS-polyacrylamide gels. Paused prod-Rather Than with Specific tRNAs ucts of the expected size were not detected ( Figure 3B ;

We next asked whether the recoding event is mediated either by specific tRNAs or by a structural element within data not shown [the product below and clearly resolved the mRNA or both. Certain ϩ1 frameshifts are dependent ϩ1 equally well. These mutants included a UAG or UGA termination codon (underlined), which when downupon a specific tRNA occupying the ribosomal P site stream of a recoding site can act as stimulators of pro-and a rare tRNA or termination codon occupying the grammed frameshifting (Matsufuji et al., 1995;  Weiss et ribosomal A site. The hungry codon induces a ribosomal al., 1987). Neither mutant frameshifted ( Figure 1 ). These pause, allowing recoding (Belcourt and Farabaugh, results imply that neither specific glycine tRNAs nor 1990; Farabaugh et al., 1993; Lindsley and Gallant, 1993;  tRNA slippage mechanisms are likely to be responsible Pande et al., 1995; Vimaladithan and Farabaugh, 1994) .

for the recoding event. Rather, the G-richness of the The BCH4 sequence, GGG-GGG-GGA-GGC-TGG-G, mRNA sequence is important. is decoded as GLY-GLY-GLY-GLY-TRP. The likely site of recoding, underlined, is decoded by two tRNA GLY CCC spe-Correlation of the Ability of G-Strings to Form cies, whereas the next two codons, GGA and GGC, are Unusual Structures with Recoding decoded by tRNA GLY UCC and tRNA GLY GCC. None of these is a A plethora of reports indicates that G-rich sequences rare tRNA (Gupta et al., 1980) and consequently should (e.g., telomeres) are able to form unusual structures in not induce a pause to facilitate tRNA/mRNA realignvitro via noncanonical base pairing between G's, e.g., ment. Nevertheless, we wished to investigate the effect Hoogsteen base pairing, to form two-or four-stranded that different glycine tRNA isoacceptors might have on (G-quartet) structures (reviewed by Williamson, 1994). frameshifting. (Glycine tRNAs decode the sequence These structures generally are stabilized by monovalent GGX, where X ϭ G, A, C, or U [Gupta et al., 1980] ). We ions (Williamson, 1994) . We hypothesized that the mutated the third and sixth G of the recoding site (bold G-string affected the local RNA architecture to cause characters above) to A, C, or T (plasmids BCH12, net ϩ1 recoding. BCHH3, and BCHH4, respectively). Analysis of these As DNA G-strings behave similarly to RNA G-strings constructs revealed that BCH12 (GGA-GGA-GGA-GGCin formation of these structures (Smith and Feigon, 1992 ; TGG-G) expressed low but detectable levels of lucifer- Sundquist and Klug, 1989; Zimmerman et al., 1975) , we ase, whereas BCHH3 (GGC-GGC-GGA-GGC-TGG-G) used the DNA oligonucleotides from our mutagenesis and BCHH4 (GGT-GGT-GGA-GGC-TGG-G) did not (Figexperiment to address this hypothesis. First, we subure 1). These data suggest that if specific tRNAs are jected the oligonucleotide used to create BCH4 to circuinvolved in the recoding event, they would have to be lar dichroism experiments. The oligo (2 M) was dis-tRNA GLY CCC decoding GGG or tRNA GLY UCC decoding GGA. Imsolved in either 20 mM Tris (pH 8.0) alone or 100 mM portantly, the results with BCH12 indicate that a KCl or 50 mM NaCl, denatured, and then cooled before tRNA GLY UCC slippage model is unlikely because tRNA does recording wavelengths on a spectrophotometer. Molecnot form a middle base pair with the next ϩ1 codon, ular rotation at specific wavelengths can be measured GAG; the predicted C:A pair has been shown to inhibit because G-quartets are optically active in polarized frameshifting 80-to 240-fold in a ϩ1 frameshift system light. The resulting wavelength scans revealed a minithat utilizes slippage (Curran, 1993) . mum at 240 nm and a maximum at 260 nm in the pres-To address whether G-rich sequences might be inence of K ϩ or Na ϩ ions but not in the absence of salt volved rather than these two specific tRNAs, we placed ( Figure 4 ). This pattern is indicative of parallel quadthe G-strings in different reading frames. We first raplex formation (Balagurumoorthy et al., 1992) . Intermutated the BCH4 sequence, GGG-GGG-GGAestingly, addition of 50 mM NaCl to a pTK-4 RNA transla-GGC-TGG-G, to AGG-GGG-GGG-AGG-CTG-G, creating tion mixture, although lowering overall translation, BCH13 ( Figure 1 ). This mutant, decoded by ARG-GLYresulted in a 2-3-fold increase in TK frameshifting (data GLY-ARG-LEU, frameshifted at standard (1%) levels not shown). (Figure 1 ), indicating that when two Gly GGG codons In a second approach, end-labeled oligonucleotides are bounded by G residues, they direct recoding as well were electrophoresed through native 20% polyacrylas do eight or more G residues comprising three glycine amide gels as described in Experimental Procedures. codons. This finding was underscored and extended by Figure 5A shows that the control oligonucleotide used the construction and expression of BCH18. The seventh to create plasmid BCH4a or oligonucleotides in which nucleotide in the G-string of BCH13 was mutated to the G(8) sequence of BCH4 (Figure 1 ) was changed to an A. While this conserves the purine richness of the either A(8), C(8), or T(8), migrated mainly as homogesequence, it shortens the G-string to 6 nt. BCH18 (AGGneous species of the expected size. However, G-rich GGG-GAG-AGG-CTG-G), decoded by ARG-GLY-GLUoligonucleotides, including those used to create plas-ARG-LEU (only one Gly codon, GGG), expressed lucifermids BCH4, BCH12, BCH13, and BCH14 exhibited three ase at just slightly lower levels than BCH15 and BCH16, different mobilities; faster-and slower-moving species whereas BCH6 (GGG-GGA-GGC-TGG-G), decoded by were observed as well as the species of expected size GLY-GLY-GLY-TRP (one GGG and one GGA codon), did ( Figure 5A ). These results suggest that the G-rich oligonot frameshift (Figure 1 ). However, we would not expect nucleotides fold into compact (fast-migrating) and com-BCH6 to frameshift if the recoding event were a Ϫ2 slip, plex (slow-migrating) structures as a result of G-G base because tRNA GLY CCC is able to slip either ϩ1 or Ϫ2 in mutant pairing. The slower-migrating species ( Figure 5A ) were BCH15, BCH16, and BCH18 but only ϩ1 in mutant BCH6 more responsive to sequence alterations than the faster-(tRNA GLY CCC does not form a middle base pair with the Ϫ2 moving species. Dilution experiments suggested that codon GAG in BCH6; Figure 1 ). Therefore, we conthe faster-moving species were monomeric (data not structed two additional mutants, BCH20 (AGG-GGG-shown). Further, repetition of the experiment in the pres-TGA-GGC-TGG-G) and BCH21 (AGG-GGG-TAG-AGGence of 50 mM NaCl yielded similar results (data not shown). CTG-G), in which tRNA GLY CCC would be able to slip Ϫ2 or Sequence shown at top. The oligonucleotide (2 M) was denatured by boiling for 15 min, then cooled on ice for 30 min before recording wavelengths on an Aviv 62DS spectrophotometer. Wavelength scans were recorded at 1 nm intervals (10 s averaging time), and three scans were averaged. Minima at 240 nm and maxima at 260 nm are characteristic of parallel quadraplexes.

To investigate this further, we compared oligonucleo-recoding event is a ϩ1 frameshift or a Ϫ2 frameshift (B. C. H. and S. Matusfuji, unpublished data). Neverthe-tide complex formation of BCH4 with other G-rich oligonucleotide sequences that directed different levels of less, for the sake of simplicity and brevity, we will assume in the ensuing discussion that recoding occurs frameshifting, i.e., BCHH3 and BCHH4 (background), BCH12 (0.3%), BCH15 (0.6%), and BCH4 (1.0%; Figure  within the G-string as a ϩ1 frameshift. 5B). Labeled oligonucleotides were electrophoresed through polyacrylamide gels as described previously.

The HSV Recoding Event Differs from Previously BCHH3 and BCHH4 sequences did not form the slower-Described Frameshifts: Failure to Detect migrating complex (S; Figure 5B ) or frameshift, whereas Pausing or Stimulation BCH12 sequences both complexed and frameshifted Current models to explain translational frameshifting enweakly and BCH15 sequences both complexed and tail two elements: first, a recoding site at which frameframeshifted at an intermediate level relative to BCH4 shifting occurs; and second, a stimulatory element that sequences (S; Figure 5B ). Thus, there is an excellent increases the frameshift efficiency by pausing ribocorrelation between the efficiency of recoding and the somes at the recoding site ; Faraability of the corresponding oligonucleotides to form baugh, 1996; Gesteland et al., 1992) . Mechanisms that unusual structures.

have been invoked to explain the actual recoding event at the frameshifting site, which both entail a kinetically slow step that is more favorable when ribosomes are Discussion paused, are, first, slippage; and second, specific peptidyl tRNAs, which are thought to facilitate out-of-frame During expression of the tk gene of an HSV drugresistant mutant, recoding occurs to shift translation binding of tRNAs at the A site (reviewed by Farabaugh, 1996) . The efficiency of frameshifting is generally con-into the ϩ1 reading frame (Hwang et al., 1994 ; this study). We have found that a specific G-rich signal embedded sidered to depend on the extent of pausing, and, in particular, weakly functioning recoding sites are most within the tk mRNA corrupts the translational machinery. The ability of sequences to induce recoding correlates affected by the presence of stimulators. We failed to detect paused ribosomes at the HSV tk recoding site well with their ability to form unusual RNA structures. We discuss below in what way this recoding event differs ( Figure 3B ), suggesting that the HSV frameshift does not involve a kinetically slower second step. Given the from previously described translational frameshifts, possible mechanisms to explain the correlation between low efficiency of the tk frameshift, it could be argued that the pause was too short for us to detect. Nevertheless, if RNA structure and recoding, and the potential relevance of our results to herpesviruses and other biological pausing were important for the tk frameshift, one would expect that stimulators would greatly increase frame-systems.

Our mutational analysis is consistent with net ϩ1 re-shift efficiency. Instead, neither a downstream RNA structure nor termination codons increased frameshift coding occurring within the G-string (Figure 1 ; data not shown) but is not definitive. Unfortunately, due to the efficiency. Thus, it appears that the tk frameshift operates via a mechanism that does not entail a kinetically low efficiency of recoding, it has not yet been possible to determine by protein sequencing whether the net ϩ1 slow step that is enhanced by ribosomal pausing. BCHH3 and BCHH4 (the oligonucleotide used to create these mutant was synthesized as a mixture, i.e., tcgaGGRGGRGGAGGCTGGG, where R equals pyrimidine). The right panel shows an autoradiogram of a polyacrylamide gel of the corresponding 32 P-labeled oligonucleotides, run as described in (A). As above, F and S indicate the faster-moving and slowermoving species, respectively.

Moreover, it has been argued that tRNAs should not slip on G-strings because of the predicted strength of Conventional ϩ1 frameshifting utilizing slippage is exemplified by the frameshift in the overlap between the the mRNA-tRNA interactions (Jacks et al., 1988) . Indeed, replacement of A-or U-rich codons with G-or C-rich TYA and TYB genes in the retrotransposon ty1 (Belcourt and Farabaugh, 1990; Farabaugh, 1996) . The frameshift triplets reduced frameshifting in Ϫ1 frameshift systems that utilize slippage (Brierley et al., 1992 ; Jacks et al., requires a 7 nt sequence, CUU AGG C, and occurs by slippage of the P-site tRNA from CUU to UUA. This slip 1988), and replacement of third position U's in the inphase triplet decreased frameshifting in a ϩ1 system, is stimulated by a translational pause induced by the slowly decoded hungry codon AGG in the A site. The suggesting that the weakness of wobble pairs facilitates ϩ1 slippage (Curran, 1993) . One might imagine a sce-frameshift requires no other factors.

If the HSV frameshift were conventional, then its P-site nario whereby the mRNA G-string interacts, for example, with rRNA (see below) in a way that might promote tRNA codons could be GGG or GGA. Given either of these P-site codons, we analyzed several A-site codons that slippage. For example, if the G-string formed non-Watson-Crick base pairs with rRNA, that would weaken could permit ϩ1 slippage while maintaining base pairing at at least two out of three codon-anticodon positions the hydrogen bonds between the O 6 positions of G's in the codon and C's in the anticodon. It would be tempting (UAG [BCH21], UGA [BCH20], GGA [gly; BCH6], GGG [gly; BCH13], and GAG [glu; BCH18]) when placed down-to speculate that the weakening of these bonds could add to the slipperiness of the mRNA-tRNA complex. stream of a single GGG or GGA codon. Of these, UAG and UGA are nonsense codons and are, in general, more However, for this scenario to be viable, we would expect that an A string would also promote net ϩ1 frameshift-slowly decoded than sense codons. (These nonsense codons operated efficiently in the antizyme frameshift ing, since A-U base pairs have even less energy than the weakened G-C base pairs imagined above. This system, which, like the HSV frameshift system, has been characterized in a mammalian system [Matsufuji et al., did not occur (BCH 11; Figure 1) . Moreover, this model would predict that an RNA structure or termination co-1995]). GGA is an intermediately used codon, and GGG and GAG are very commonly used ("well-fed") codons don that enhanced the probability of the second, kinetically slower, slip would increase the efficiency of frame- (Haas et al., 1996) . Thus, if the HSV frameshift were conventional, we would expect higher frameshift effi-shifting, which did not occur. Thus, it is difficult to consider slippage on G-strings as a viable explanation ciencies with A-site codons UAG and UGA, intermediate frameshift efficiencies with GGA codons, and lower for our results. Furthermore, although tRNA GLY UCC will form base pairs at frameshift efficiencies with GGG and GAG codons. This was not observed. Instead, only the well-fed A-site co-positions one and three with the ϩ1 codon, GAG in BCH12, the U:G wobble base pair is relatively weak at dons promoted frameshifting, whereas slowly decoded A-site codons did not (e.g., compare BCH18 and codon position 3 while the clashing middle base pair (A:C) in the slipped codon-anticodon complex should BCH21). Thus, the HSV frameshift is not conventional. result in destabilization. Grosjean and co-workers (1978) demonstrated that centrally positioned C:A or A:C pairs abolished synthetic anticodon-codon complexes. Such mismatches decrease frameshifting more than 80-fold in the ϩ1 E. coli RF2 system (Curran, 1993) , compared with the only 3-fold decrease in the HSV tk frameshift (BCH12, Figure 1) . Thus, the ability of BCH12 to frameshift is very difficult to explain by either conventional slippage or by the unconventional mechanism considered above. Taken together, our results argue that tRNA slippage is highly unlikely to account for the HSV recoding event.

A second mechanism for frameshifting involves specific peptidyl tRNAs, which are thought to facilitate out-offrame binding of tRNAs at the A site (Farabaugh et al., 1993; Matsufuji et al., 1995; Pande et al., 1995; Vimaladithan and Farabaugh, 1994) . In our case, this would implicate tRNA GLY CCC and tRNA GLY UCC as responsible for the frameshift when they occupy the P site. In yeast, GGG codons in the P site can induce frameshifting but only ity that a subpopulation of glycine tRNAs contain 4 nt anticodons and therefore function as ϩ1 frameshift suppressors.

However, as pointed out by Farabaugh et al. (1993) , In summary, the tk recoding event appears to function who considered a similar model to explain the TY3 ϩ1 with the same efficiency with or without a stimulator frameshift, it is more likely that such an excluded base and in the absence of detectable pausing (Figures 1-3) .

would still stack between the paired nucleotides on ei-It does not behave as if it utilizes slippage or peptidylther side. This has been observed in several RNA struc-tRNA-induced misalignment. Thus, the tk recoding tures (Aboul-ela et al., 1995; Kalnik et al., 1989) . Thus, event appears to be truly novel.

in this model, some kind of folding of the mRNA on itself would be required to stabilize a bulge. We have obtained preliminary evidence that the G-string is contained Possible Mechanisms for the tk Recoding Event The efficiency of the tk recoding event correlates with within a structure resistant to RNase T1, consistent with this idea (data not shown). the G-richness of the signal and its ability to form unusual RNA structures. Such structures depend on In our second model (intermolecular), ribosome-mRNA interactions play a role in the occlusion of one the ability of G residues to pair with each other, e.g., via non-Watson-Crick interactions (Williamson, 1994) .

nucleotide. Given the correlation between frameshifting and the ability of the G-strings to form unusual struc-Therefore, we suggest that G-string structure mediates recoding within the ribosome via Hoogsteen or other tures, an appealing possibility would be non-Watson-Crick interactions between the viral mRNA G-string and non-Watson-Crick interactions.

We envision three possible models to account for this. G-rich element within mammalian rRNAs. There already are precedents for rRNA-mRNA interactions affecting In the first model (intramolecular), the G-string in the mRNA forms a structure in the ribosome. Formation of the efficiency of frameshifting in the E. coli release factor 2 and dnaX genes (Larsen et al., 1994; , an intramolecular structure in the ribosome has been proposed to account for the T4 gene 60 ribosomal hop 1990). However, in these cases, the interactions are Watson-Crick and do not involve the recoding signal per se. (Herbst et al., 1994; Weiss et al., 1990) . In the intramolecular model for the tk frameshift, one of the nucleotides A version of this model is cartooned in Figure 6 . Purines within rRNA at the ribosomal A site interact via within the G-string would bulge out so that it does not pair with a tRNA, either in the P site or in the A site.

noncanonical base pairs on the major groove side of a GGG codon within the G-string. Other potential rRNA-Thus, in the ribosome, the two tRNAs bound to the mRNA would form an RNA helix in which an extra nucleo-mRNA interactions are omitted from the cartoon for simplicity. The incoming aminoacyl-tRNA interacts with the tide between the tRNAs is excluded from base pairing. same GGG codon via Watson-Crick base pairs. The translational leakiness may occur more frequently than would otherwise be expected, since genes containing result is a small pseudohelix (tRNA-mRNA and mRNA-rRNA) with a nucleotide bulged out. Supporting evi-G(6) sequences are more common than genes containing G(8) sequences. dence for bulging nucleotides stabilized by non-Watson-Crick base pairs comes from studies on the HIV HSV genes are very G-C-rich and contain higher numbers of guanine repeats than their cellular counterparts. Rev Responsive Element RNA bound to Rev peptide For example, HSV-1 tk genes contain one G(6) and one (Battiste et al., 1994) . In this structure, purine-purine G(7) string, and HSV-2 tk genes contain two G(6) and base pairs form within an internal loop of a helix to create one G(7) string. It has been suggested that the high a quasi-continous helix with the concomitant bulging of mutation frequency in the tk gene of HSV may be a bases (Battiste et al., 1994) .

consequence of these guanine repeats (Kit et al., 1987) , In our third model, a non-Watson-Crick base-paired and, indeed, Hwang and Chen (1995) have obtained structure forms within the ribosome, distorting the riboevidence that frameshift mutations can occur frequently somal A site, thus favoring binding of tRNA to the mRNA in the G(7) string. If this is so, why would a virus retain in the ϩ1 frame. This structure could result from either sequences that accumulate mutations? Perhaps the viintramolecular base pairing or intermolecular base pairrus can tolerate mutations in these sequences because ing between mRNA and rRNA. In this model, no bulge it generates so many wild-type copies per infected cell. in the mRNA is required. However, for this model to fit Nevertheless, one speculation is that the G-string seour data, the distortion of the A site would have to be quences have been retained because they permit the a relatively fast step kinetically; otherwise, frameshifting expression of alternate polypeptides. Regardless, a would be expected to be increased by pausing.

consequence of our results is that the translational ma-Each of these models makes specific predictions that chinery can partly compensate for mutations in the can be tested. Furthermore, any of these models pro-G-string, permitting low level "ribosomal rescue" of the vides a new example for a role of non-Watson-Crick mutations and expression of some of the normal gene base pairs in biology, in this case, translational recoding.

products. In the case of the drug-resistant mutant we have studied, it may be that this low level of TK expression was insufficient to activate ACV effectively but was Implications for Herpesviruses and Other Systems sufficient for pathogenesis in a human, leading to selec-Our results indicate that G-strings are sufficient to intion of the mutant in the infected patient. Given that the duce net ϩ1 recoding in vitro. Furthermore, the degree G(7) string is a mutational hotspot, we predict that other of frameshift efficiency (approximately 1%) from tk se-ACV-resistant mutants associated with human disease quences measured in reticulocyte lysates matches TK will contain the same mutation. Study of other drugactivity quantitated in TK mutant-infected cells using resistant HSV tk mutants may identify different signals new assays developed in our laboratory (S.-H. Chen, that allow relaxation of the constraints involved in read-B. C. H., and D. M. C., unpublished data). Encouragingly, ing frame maintenance. preliminary results indicate that some of our constructs placed under the control of the SV40 promoter recapitu- 

An obvious question is whether recoding events medi-pT3LacLuc was created by cloning the T3 promoter into the HindIII ated by G-strings could be occurring in genes other site of pBgalluc-1 (Reil et al., 1993) . The BCH-1, BCH3-BCH19 plasmids were constructed by cloning synthetic oligonucleotides con-than the mutant HSV tk gene that we have studied. One taining specific HSV sequence into the BglII and SalI sites of possibility is the wild-type tk gene. This gene contains pT3LacLuc. BCH1 and BCH2 were constructed by cloning the the motif G(7)AG, which, like the mutant form G(8)AG, SnaBI-SacI fragment from pTK-wt or pTK-4 into T4 DNA polymerase is sufficient for standard levels of frameshifting in our blunt-ended BglII and SalI sites of pT3LacLuc. (Plasmids TK-wt and reporter system (Figure 1 ). This raises the possibility TK-4 were described previously as pBH15 and pBH13 by Hwang et that the wild-type tk gene normally expresses low levels al. [1994] ). Plasmid BCHSLϪ was created by digesting BCH4 with BglII and BamHI and religating. Oligonucleotides (TCGAG and of a previously undetected polypeptide. This polypep-GATCC) were cloned into SalI-BglII-digested pT3LacLuc to create tide would retain the ATP-binding site of TK but would BCH0. This construct, lacking HSV sequences, has luciferase in the lack the nucleoside binding site and other conserved ϩ1 frame relative to ␤-galactosidase. Plasmids pBH13 and pBH15 residues (Brown et al., 1995) . The question of whether have been previously described (Hwang et al., 1994) . pTK-4a (labothis polypeptide is expressed and, if so, whether it has ratory plasmid BH17) was constructed by digesting pBH13 with SphI and SacI. The intervening 51 nt were replaced with identical any function, is under investigation. oligonucleotide sequences, except that the G(8)AGGCTGGG motif Generally, long G-strings are not found in eukaryotic was changed to GCTCACCATTCGCGAG. Insertion of a duplex of coding sequences, suggesting that there is selection synthetic oligonucleotides with sequences 5Ј GGCCTTCCTACAAG against these motifs. Searches of the database for the GGAAGGCCAGGGAGCT and 5Ј CCCTGGCCTTCCCTTGTAGGAAG motif G(8)AG revealed two occurrences within herpes GCCAGCT into the SacI site of pBH13 created pTK-SLϩ (laboratory genome sequences (types 1 and 2) and one example in plasmid BH13SLϩ). This plasmid contains the HIV stem-loop 10 nt downstream of the G-string. a cellular gene. We hypothesize that low level net ϩ1 Plasmids pBH19 and pBH21, for pausing experiments, were creframeshifts will be detected from other genes containing ated by digesting pBH13 or pBH17 with EcoRV and SacI. The fragthis sequence motif. Our results indicate that shorter ment were T4 DNA polymerase blunt-ended, gel-eluted, and cloned G-strings (e.g., G6), within purine-rich contexts, promote into the EcoRI (T4 DNA polymerase blunt-ended)-EcoRV sites of net ϩ1 recoding, although at lower levels ( Figure 1 ). p911 (Digard et al., 1993) . All plasmid constructs were verified by DNA sequencing.

This implies that biologically relevant G-string-mediated

Benhar, I., and Engleberg, K.H. (1993) . Frameshifting in the expression of E. coli trpR gene occurs by the bypassing of a segment of In vitro transcription reactions were carried out as described previously (Hwang et al., 1994) . Product RNA was recovered by phenol-its coding sequence. Cell 72, 121-130. chloroform extraction and ethanol precipitation in the presence of Brierley, I., Jenner, A.J., and Inglis, S.C. (1992) . Mutational analysis 2 M ammonium acetate. The RNA pellet was dissolved in water of the "slippery sequence" component of a coronavirus ribosomal and checked for integrity by electrophoresis on 1% agarose gels frameshifting signal. J. Mol. Biol. 227, 463-479. containing 0.1% SDS.

Brown, D.G., Visse, R., Sandhu, G., Davies, A., Rizkallah, P.J., Melitz, In ribosomal frameshift assays, serial dilutions of purified RNAs C., Summers, W.C., and Sanderson, M.R. (1995) . Crystal structure were translated in the rabbit reticulocyte lysate system as previously of the thymidine kinase from herpes simplex virus type-1 in complex described (Hwang et al., 1994) . Frameshift efficiencies were calcuwith deoxythymidine and Ganciclovir. Nature Struct. Biol. 2, lated from ␤-galactosidase, and luciferase assays were performed 876-881. exactly as described (LeBowitz et al., 1991; Schenborn and Goiffen, Coen, D.M., and Schaffer, P.A. (1980) . Two distinct loci confer resis-1993) and frameshift efficiencies calculated as described in Results.

tance to acyloguanosine in herpes simplex virus type-1. Proc. Natl. In ribosomal pausing assays, translations were carried out essen-Acad. Sci. USA 77, 2265-2269. tially as described by Somogyi et al. (1993) . The translational inhibitor, edeine (5 M final concentration), was added 5 min after the start Coen, D.M., Kosz-Vnenchak, M., Jacobson, J.G., Leib, D.A., Bogard, of the reaction in order to obtain synchronous initiation. Aliquots (3 C.L., Schaffer, P.A., Tyler, K.L., and Knipe, D.M. (1989) . Thymidine l) were withdrawn at specific intervals, mixed with an equal volume kinase-negative herpes simplex mutant establish latency in mouse of pancreatic RNase A (100 g/ml) in 10 mM EDTA, and incubated trigeminal ganglia but do not reactivate. Proc. Natl. Acad. Sci. USA at room temperature for 10 min. Laemmli's buffer (12.5% glycerol, 86, 4735-4739. 2% bromophenol blue, 25 mM Tris [pH 6.8], 100 mM dithiothreitol, Curran, J.F. (1993) . Analysis of effects of tRNA: message stability and 2% SDS) was added to the samples prior to electrophoresis on frameshift frequency at the Escherichia coli RF2 programmed through SDS-12.5% (wt/vol) polyacryamide gels. The products were frameshift site. Nucl. Acids Res. 21, 1837-1843. analyzed by autoradiography of dried gels.

Digard, P., Bebrin, W.R., Weisshart, K., and Coen, D.M. (1993) . The extreme C-terminus of herpes simplex virus DNA polymerase is Circular Dichroism crucial for functional interaction with processivity factor UL42 and Samples (2 M) were prepared by heating at 95ЊC for 15 min and for viral replication. J. Virol. 67, 398-406. cooled to room temperature, and circular dichroism spectra were Efstatiou, S., Kemp, S., Darby, G., and Minson, A.C. (1989) . The role recorded on an Aviv 62DS spectrophotometer (made available by of herpes simplex virus type-1 thymidine kinase in pathogenesis. J. Professor Stephen Harrison). Wavelength scans were recorded at Gen. Virol. 70, 869-879.

expression of HIV-1 envelope glycoprotein

A mutation in ribosomal protein L9 affects ribosomal hopping during translation of gene 60 from bacteriphage T4

Ribosome gym-USA 91

Hairpin and parallel quartet structures for telomeric sequences

A net ϩ1 frameshift permits synthesis of a

thymidine kinase from a drug-resistant herpes simplex virus mutant. Binding of an HIV Rev peptide to Rev response element RNA induces Proc

An altered spectrum of herpes simplex virus mutations mediated by an antimutator DNA polymerin the yeast retrotransposon Ty: tRNAs induce slippage on a 7 nt minimal site

Ϫ1, ϩ1, ϩ2, ϩ5 and ϩ6 ribosomal frameshifting. Cold Spring Harbor Symp

Signals for ribosomal frameshifting in the Rous sarcoma virus gagpol region

Ribosomal frameshifting from Ϫ2 to ϩ50 nucleotides

Herpes simplex virus thymidine kinase and specific stages of latency in murine trigeminal Williamson

X-ray fiber Conformation of adenosine bulge-containing deoxytridecanucleodiffraction and model-building study of polyguanylic acid and polyitide duplexes in solution

Nucleotide sequence changes in thymidine kinase gene of herpes simplex virus type -2 clones from an isolate of a patient treated with acyclovir

The sequences of and distance between two cis-acting signals determine the efficiency of ribosomal frameshifting in human immunodeficiency virus type I and human T-cell leukaemia virus type II in vivo

Translational accuracy and the fitness of bacteria

rRNA-mRNA base pairing stimulates a programmed Ϫ1 ribosomal frameshift

Simultaneous transient expression assays of the trypanosomid parasite Leishmania using ␤-galactosidase and ␤-glucuronidase as reporter enzymes

On the directional specificity of ribosome frameshifting at a "hungry

Autoregulatory frameshifting in decoding mammalian orthinine decarboxylase antizyme

Pulling the ribosome out of frame by ϩ1 at a programmed frameshift site by cognate binding of aminoacly-tRNA

A heptanucleotide sequence mediates ribosomal frameshifting in mammalian cells

Polyamines regulate the expression of ornithine decarboxylase antizyme in vitro by inducing ribosomal frameshifting

Progressive esophagitis from acyclovir-resistant herpes simplex: clinical roles for DNA polymerase mutant and viral heterogeneity?

Quadraplex structure of Oxytricha telomeric DNA oligonucleotides

Ribosomal pausing during translation of an RNA pseudoknot

Telomeric DNA dimerizes by formation of guanine tetrads between hairpin loops

Ribosomal movement impeded at a pseudoknot required for frameshifting

Special peptidyl-tRNA molecules can promote translational frameshifting without slippage

Slippery runs, shifty stops, backward steps and forward hops

 (1996). Programmed translational frameshifting. Miin the gel as well as the running buffer. DNA samples were in 5 l crobiol. Rev. 60, 103-134. of TE plus salt at the same concentration as the gel and prior to Farabaugh, P.J., Zhao, H., and Vimaladithan, A. (1993). A novel proelectrophoresis were heated to 95ЊC for 20 min, cooled to room grammed frameshift expresses the POL3 gene of retrotransposon temperature, and mixed with 1 l of 30% glycerol-containing marker Ty3 of yeast: frameshifting without tRNA slippage. Cell 74, 93-103. dyes. Gels were dried and visualized by autoradiography.Fyfe, J.A., Keller, P.M., Furman, P., Miller, R.L., and Elion, G.B. (1978). Thymidine kinase from herpes simplex virus phosphorylates the Acknowledgments new antiviral compound, 9-(2-hydroxyethoxymethy)guanine. J. Biol. Chem. 253, 8721-8727. We thank Paul Digard, Martha Kramer, Louane Hann, and Ian Brierley for their comments on the manuscript; Steve Harrison for Gesteland, R.F., Weiss, R.B., and Atkins, J.F. (1992). Recoding: reallowing us to use his circular dichroism equipment; Charles Hwang programmed genetic decoding. Science 257, 1640-1641. and Steven Sacks for collaborating on earlier aspects of this project;Grosjean, H.J., de Henau, S., and Crothers, D.M. (1978). On the and Senya Matsufuji for help on protein sequencing attempts. We physical basis for ambiguity in genetic coding interactions. Proc. also thank the Drug Synthesis and Chemistry Branch of the National Natl. Acad. Sci. USA 75, 610-614. Cancer Institute for the drug edeine. This work was supported by Gupta, R.C., Roe, B.A., and Randerath, K. (1980). Sequence of hu-National Institutes of Health grant AI26126 to D. M. C. and by the man glycine transfer ribonucleic acid (anticodon CCC): determina-German Ministry of Science and Technology. tion by a newly developed thin-layer readout sequencing technique and comparison with other glycine transfer ribonucleic acids. Bio-Received April 16, 1996; revised July 31, 1996. chemistry 19, 1699-1705. Haas, J., Park, E.-C., and Seed, B. (1996. Codon usage limitation