key: cord-0862486-oir3rlb7
authors: Brierley, Ian; Jenner, Alison J.; Inglis, Stephen C.
title: Mutational analysis of the “slippery-sequence” component of a coronavirus ribosomal frameshifting signal
date: 1992-09-20
journal: Journal of Molecular Biology
DOI: 10.1016/0022-2836(92)90901-u
sha: bdbfc53f7815dfd0af23044962ffff1f3ad4f296
doc_id: 862486
cord_uid: oir3rlb7

Abstract The ribosomal frameshift signal in the genomic RNA of the coronavirus IBV is composed of two elements, a heptanucleotide “slippery-sequence” and a downstream RNA pseudoknot. We have investigated the kinds of slippery sequence that can function at the IBV frameshift site by analysing the frameshifting properties of a series of slippery-sequence mutants. We firstly confirmed that the site of frameshifting in IBV was at the heptanucleotide stretch UUUAAAC, and then used our knowledge of the pseudoknot structure and a suitable reporter gene to prepare an expression construct that allowed both the magnitude and direction of ribosomal frameshifting to be determined for candidate slippery sequences. Our results show that in almost all of the sequences tested, frameshifting is strictly into the −1 reading frame. Monotonous runs of nucleotides, however, gave detectable levels of a −2 +1 frameshift product, and U stretches in particular gave significant levels (2% to 21%). Preliminary evidence suggests that the RNA pseudoknot may play a role in influencing frameshift direction. The spectrum of slip-sequences tested in this analysis included all those known or suspected to be utilized in vivo. Our results indicate that triplets of A, C, G and U are functional when decoded in the ribosomal P-site following slippage ( XXX YYYN) although C triplets were the least effective. In the A-site (XXY YYY N), triplets of C and G were non-functional. The identity of the nucleotide at position 7 of the slippery sequence (XXXYYY N ) was found to be a critical determinant of frameshift efficiency and we show that a hierarchy of frameshifting exists for A-site codons. These observations lead us to suggest that ribosomal frameshifting at a particular site is determined, at least in part, by the strength of the interaction of normal cellular tRNAs with the A-site codon and does not necessarily involve specialized “shifty” tRNAs.

contains a, signal in the genomic RNA which briefly overlapping open reading frames (ORFs) la and lb. with lb in the -1 frame with respect to la t Author t,o whom a11 correspondence should be (Boursnell et al., 1987) . The frameshift signal is located at the overlan region and its conseauence is oouytes. some 30",, 11t the ribosomes are induced to change reading frame (Brie&g rt a/., 1987 (Brie&g rt a/., , 1989 . Mutagenesis experimclnts have revealed that the necessary information for efficient fra,meshift.ing is contained wit,hin an 86 nucleotide stretch located at the overlap region (Brierley et al., 1989) . The frameshift signal is composed of two elements. a heptanucleotide sequence WCAXAC. the site where t)he ribosome is thought. to change reading frame (,Jac*ks rt CI. (.. 19886) positioned at a defined distance upstream from an RNA pseudoknot structure in t,he mRNA (see Pleij & Bosch. 1989) . Both elements are WSWt,ial for efficient frameshifting. However, in the c+ase of the pseudoknot structure, no primary sequencxe determinant appears to be involved in the frameshift process; as long as the overall structure is maintained, frameshifting is highly efficient (Brie&y rt &., 1991) . The IKV frameshift signal belongs t,o a class of efficient -I frameshift sites first described for the vertebrate ret'rovirus Rous sarcoma virus (KRV: Jacks $ Varmus. 198.5) . In recent Fears, examples of t,his kind of frameshift. expression strategy have been predicted of documented to occur in many other systems (for a review see Atkins et al., 1990) . Tn RSV, frameshifting links the overlapping gag and pal coding regions enabling production of the gag-pal polyprotein from which reverse t'ranscriptase is derived. An alignment of the gag-pal, gag-pro or pro/p01 overlap sequences of many ret,roviruses or related eukaryotir elements known or suspected to utilize frameshift'ing highlighted a conserved heptanucleot,ide motif containing two homopolymeric triplets of the order XXXYYYN (where X = A, G or IT, Y = A or I' and N is A, (1 or CJ: Jacks et al., 19886) . Amino acid sequencing of the transframe protein has revealed that this sequence motif (termed the "slippery" sequence) is the actual site of the rihosomal frameshift in mouse mammary tumour virus (MMTV, gag-pro junction; Hizi et nl.. 1987) , human immunodefieiency virus type 1 (HTV-1: Ja,cks rt nf., 1988a), RSV (<Jacks et al., 1988b) , and is likely to be the slip-site in all the other systems including THV. sequence analysis has suggested that RNA pseudoknots are present downstream from the slip-sites in a large number of cases (Brierley rf al., 1989 (Brierley rf al., : t'en T>am et al., 1990 . Thus slippery sequences and RNA pseudoknots (or other secondary structures) appear to be common elements of this class of -1 frameshift event. To account for the conserved slip-site motif in retroviruses, Jacks et al. (19886) proposed a simultaneous slippage model of frameshifting (see Fig. 1 ) in which t'wo adjacent ribosomebound tRNAs decoding the slip sequence in the zero reading frame (i.e. X XXY YYZ) slip back simultaneously by one nucleotide such that bot,h tRSAs are in the -1 phase (XXX YYY) and are basepaired to the mRNA in two out of t,hree anticodon positions. Following t.ranslocation. translation proceeds in the -1 frame. Support for this model comes from studies of the slippery-sequences of RSV [Jacks et al., 19886) , TBV (Brierley et al., 1989 ) and I----P j A GGUAC Peptldyl transfer 3 nt translocation I codondacoded Figure 1 . The simultaneous slip model of frameshifting. as proposed by Jacks ~, t al. (198%) . In this example. tRNAL'" carrying the nascent peptide (jagged line) and tRNA*"" are shown bound to the IRV la frame eodons WA and AAC in the ribosomal P(peptidyi)-and A(aminoacyl)-sites.

In about 30"/ of' cases. both tRNAs slip back by I nucleotide in the 5' direction, such that they now pair with the lb frame codons UUU and AAA and form base-pairs (dots) only with the first 2 nucleotides of each codon. Following peptidyl transfer and translocation, the next lb frame codon. CGG, is brought into the &site, where it is decoded by tRS,4*'g. In this example, which is based on that presented by Jacks at al. (1988b) , the tRNA anticodon sequences are shown as standard Watson-Crick base-pairs. The actual anticodon sequences are described and discussed in the text.

the yeast double-stranded RNA virus, L-A (Ijinman at al.: 1991) , where it has been found that mutations that reduce the homopolymeric nature of the slip sequence greatly reduce frameshifting, presumably by reducing the post-slippage mRNA-tRNA complementarity. In those systems in which t,he amino acid sequence of the transframe prot'ein has been determined, it is known that the frameshift occurs within the slippery sequence at the second codon of the tandem slippery pair, i.e. at that codon decoded in the ribosomal A-site (XXXYYYN. zero __-) frame, A-site codon underlined). Amongst natural retroviral slip sequences (* Jacks et al., 19883) and in probably all cases of this class of frameshift signal found to date in viruses (ten Dam et al., 1990) , only three A-site codons (AAC, CUA and UUU) are represented, and it has been suggested that only certain "shifty" tRNAs are able to slip at, the A-site (Jacks et al., 1988b) . The potential involvement of specialized tRNAs in the frameshift process was further implicated by the observation that the presence of a G nucleotide at position 7 of the slipsequence (XXXYYYG) greatly reduces frame-shifting in eukaryotic systems (Inglis et al., 1990; Dinman et al., 1991) but not in prokaryotic systems (Weiss et al., 1989; Tsuchihashi, 1991) .

Here, we describe a detailed analysis of the kind of slippery sequences that can function at t,he IB1 frameshift site. Our approach was to replace the TBV slip-sequence with candidate sequences and then to test the ability of synthetic mRNAs containing the variant sites to direct frameshifting in a cell-free translation system. We hoped to be able to determine the frameshift efficiency of a large number of sites. particularly position 7 variants, since this would allow the slipperiness of a range of A-site-decoding tRNAs to be investigated. We firstly confirmed that the site of frameshifting in IBV was at the UUUAAAC sequence, and then used our knowledge of the pseudoknot structure (Brierley et al., 1991) and a suitable report,er gene t:o design an expression construct in which all the terminat,ion codons within the pseudoknot were removed such that frameshifting into any reading frame could be monitored simultaneously. Thus. for each candidate slip sequence, both the magnitude and, to some extent, the direction of frameshifting could be det,ermined. Our results show that in almost all of the sequences tested, frameshifting was strictly into the -I reading frame. However, detectable levels of -S/+ 1 frameshifting (0~5%) were seen with monotonous runs of nucleotides (A,N, C,N, GsS) or when a U triplet was present in the A-site. Significant levels of -2/ + 1 frameshifting were observed with monotonous U tracks (U,N; 2% to S'+/b) and eight consecutive U bases gave very high levels (21%). These observations support the widely held view that runs of U bases in mRNAs are particularly slippy. With respect to the kind of sequences that could functionally replace the IBV slip sequence, we found that, triplets of A, C: G and U were functional when placed in that region of the slip-sequence decoded in the ribosomal P-site following slippage (XXXYYYZ; -1 frame P-site codon underlined), but C-triplets were the least slippy. Only triplets of A and L' were functional in the ribosomal A-site (XXXmZ; -1 frame A-site codon underlined), in agreement with related studies of RSV (Jacks et al., 19888) and L-A frameshifting (Dinman et al., 1991) . Surprisingly, a number of mutations created in the first slippery codon (XXXYYYK, decoded in the P-site) were found to display high levels of frameshifting, despite the apparent reduction in the homopolymeric nature of the slip sequence. This suggests that in certain cases only minimal postslippage mRNA-tRNA pairing is required for efficient frameshifting. The identity of the nucleotide at position 7 of the slip-sequence was found to be a critical determinant of frameshift efficiency and we show here that a hierarchy of frameshifting is seen for the various pre-slippage A-site codons. For the series, XXXAM, the hierarchy for K was C > A N U >>G; for XXXUm, the hierarchy was U > A + C > G. These observations lead us to suggest that frameshifting at a particular site is determined, at least in part, by the strength of interaction of normal tRNAs with the A-site codon and does not necessarily involve specialized "shifty" t)RNAs.

(a) Site-specific mutagenrsia Site-directed mutagenesis was carried out. hy a procedure based on that described by Kunkel (1985; Brierley el nl., 1989) . 411 the plasmids used in t,his study contain the intergenir region of the filamentous bacteriophage fl (Dotto et al., 1981) enabling single-stranded plasmid DNA to be generated following infection of plasmid-carrying bacteria with bacteriophage R408 (R~sel et al., 1986) . I7racil-containing.

single-stranded DXA4 substrates for mutagenesis were prepared by R408-superinfection of plasmid-carrying Escherichia coli RZ1032 (*ells (dutung-; Kunkel. 1985) . Oligonucleotides for mutagenesis were synthesized using an Applied Riosystems 3814 DSA synthesizer and the mutagenesis react,ions performed as before (Brierley et al., 1989) . Mutants were identified by dideoxy sequencing (Sanger et al., 1977) (Hames. 1981 region. To confirm that ribosomes frameshift at the UUUAAAC sequence, two mutant derivatives of pFS8 were prepared.

In pFS8.5, the la frame (zero frame) glycine codon immediately downstream from the UUUAAAC sequence was converted by muta-genesis to a stop codon (UGA). In pFS8.19, the same mutation was introduced but in addition the leucine codon just upstream from the UUUAAAC stretch (in the -1 or lb frame) was changed to a termination codon (UAA). In pFS8.19, the upstream and downstream

ORFs of the reporter gene overlap solely by the seven nucleotides that constitute the putative slip-site UUUAAAC. For ribosomal frameshifting to be observed in this mutant, ribosomes would have to slip within the heptanucleotide sequence. As can be seen in Figure  2 (b), both pFS8.5 and pFS8.19 show frameshifting at wild-type levels, demonstrating that frameshifting does indeed occur at' this position.

(b) A frameshift cassette to monitor all reading f Tame8

We wished to design a construct in which frameshifting into any reading frame could be monitored simultaneously, since this was not possible with pFS8. As the wild-type IBV pseudoknot contains a number of stop codons, we firstly designed an artificial "minimal" pseudoknot based on our knowledge of t,his RNA structure (Brierley et al., 19911 , in which loop 2 was reduced from 32 to only 8 nucleotides in length, and all the termination codons were changed to sense codons (see Fig. 3 ). The changes were carefully chosen so as to form a pseudoknot whose structure and predicted stability bore close resemblance to the wild-type pseudoknot. As can be seen in Figure 3 , the minimal pseudoknot differs from the wild-type in a number of ways; loop 2 is 24 nucleotides shorter, the G-A mismatched pair in stem 1 of the wild-type pseudoknot is replaced by a G.U base-pair, the G nucleotide of loop 1 of the wild-type structure is replaced by a C nucleotide and, finally, the minimal pseudoknot has ,no stop codons. There is no difference in the lengths of the two stems in each structure. Previous work (Brierley et al., 1991) has shown that mutations that change the nucleotide sequence of the pseudoknot without greatly influencing the overall structure do not inhibit the frameshift process. Thus, reducing the length of loop 2 to eight nucleotides and changing the sequence of loop 1 ought not to influence frameshifting. However, by replacing the mismatched G-A pair in stem 1 by G.U, one may expect, a moderate increase in the stability of stem 1 and hence a small increase in the frameshift efficiency (Brierlev et aZ., 1991; see later). Having designed the minimal pseudoknot, we then searched available reporter genes for regions in which significant lengths of ORF were present in all three frames. A suitable candidate found was the influenza PB2 gene that contained such a region just downstream from a unique BgZII site beginning at position 456 (Young et al., 1983) . Our strategy wa,s to clone the minimal pseudoknot into the BgZII site such that, the upstream portion of the PB2 gene was in-frame with the IBV la coding region. As the downstream PB2 information was open (to a greater or lesser extent) in all three reading frames, we with a mismatched pair (G7-A24) in stem 1. The singlestranded connecting loops contain 2 (loop 1) and 32 (loop 2) nucleotides, respectively. The termination codon of the la ORF (UGA) is boxed, as is the slippery sequence (UCUAAAC). The nucleotides in loop 2 that, were retained in the minimal pseudoknot are in a larger, bold font. In the minimal pseudoknot, which contains no termination codons, the stems are of the same length and base-pair composition as the wild-type pseudoknot, except that the G7-A24, mismatched pair is replaced by a IJ7.G24 pair. Three other differences are present; the loop 1 G nucleotide is changed to a C; loop 2 contains only 8 of the 32 nucleotides present in the wild%type structure, and the glycine codon immediately downstream from the slippery sequence is replaced by a termination codon (in the la reading frame). could monitor frameshifting into any frame simultaneously by examining the size of the frameshift products on SDS/polyacrylamide gels. The strategy for the construction of this plasmid is shown in Figure 4 and detailed in Materials and Methods. An oligonucleotide (90mer) was synthesized that contained sequences complementary to the minimal pseudoknot, the slip-sequence and a bacteriophage T7 RNA polymerase promoter. A shorter oligonucleotide was then annealed that contained the reverse complement of the T7 promoter sequence and the single-stranded regions filled by primer extension using the large fragment of DNA polymerase I. The blunt-ended molecules were selfligated and then digested with BglII. The resulting BgZII pseudoknot "cassette" was cloned into the BglII site of plasmid pING12 at the aforementioned site in the PB2 gene. The construct was then modified in two ways. Firstly, an inconvenient termination codon (UAA) within one of the PB2 frames was changed to serine (UCA), such that the particular frame in question (zero frame) was extended by some 60 amino acid codons in order to ease assignment of translation products to reading frame. Secondly, a PvuII restriction site was introduced at the 3' end of the inserted cassette. This was carried out to remove a termination codon that was inad-vertently missed during design of the long oligonucleotide, and to provide a suitable restriction site 46X I. Brie&y et al. Figure 4 . The construction of plasmid pFScass5 has been described in detail in the text. Briefly. an oligonucleotide containing sequences complementary to a T7 promoter and the IBV minimal pseudoknot (oligol) was annealed to an oligonucleotide containing the T7 promoter sequence (oligo2) and the single-stranded regions filled with the Klenow fragment of DNA polymerase I. Following self-ligation, a BglII frameshift-signal-containing DNA fragment was isolated and ligated into a reporter gene, the influenza PI32 gene, to create pFScass5. The diagram of pFScass5 shows the minimal frameshift signal (in white) flanked by the influenza PB2 reporter gene (shaded). Linearization of the plasmid with BamHl and in. vitro transcription using SP6 RNA polymerase yields an mRNA (2.4 kb) that. when translated in rabbit reticulocyte lysates, is predicted to produce a 19 kDa product corresponding to ribosomes that terminate at the new la termination codon (UGA) located immediately downstream from the CUUAAAC sequence (stippled), and a 22 kDa ( -1) frameshift product. Any ribosomes that enter the -2/ + 1 reading frame would produce an 85 kDa product, and any zero frame ribosomes, a 28 kDa product. Plasmids pFScass6 and 7 differ from pFScass5 only by the deletion (pFScass6) or insertion (pFScass7) of a single C nucleotide just downstream from the inserted frameshift cassette.

for the generation of short, pseudoknot-containing T7 transcripts for structural studies (not addressed here). The resulting construct, pFScass5 is shown in Figure 4 . any ribosomes that enter the -2 (or + 1) phase would give an 85 kDa product, and a 28 kDa product would be seen for any ribosomes that, do not frameshift yet in some way suppress the termination codon and continue translation in the zero reading frame.

(c) The IBV frameshift is strictly into the -I phase Shown in Figure  5 are the translation products from pFScass5 and two related constructs, pFScass6 and 7, which differ from pFScass5 only by The 2 stacked stems of the pseudoknot are drawn horizontally and are shaded. A number of mutants created in pFScass5 are depicted. In 5.39, the T7 promoter was deleted but the la frame was maintained.

A number of spacing mutants (pSM) are shown in which 1 or 2 nucleotides were deleted or inserted from the spacer sequence. These mutants were prepared in the background of either pFScass6 or 7 such that frameshifted ribosomes produce a 22 kDa product in each case (see the text). A mutation at the base of stem 1, and known to destabilize the wild-type RNA pseudoknot (Brierley et al., 1991) was also prepared (5.15, CCCC to GGGG). The gel shows the reticulocyte lysate translation products synthesized in response to mRNAs derived from BarnHI-digested pFScass5 or mutant templates. Polypeptides were labelled with [35S]methionine. separated on an SDS/17.5O& polyacrylamide gel and detected by autoradiography.

the deletion (pFScass6) or addition (pFScass7) of a C nucleotide downstream from the cassette. Insertion or deletion of the single nucleotide ought to reassign the translation products to a different class of frameshift. Thus in pFScass6, the -1 frame is 28 kDa; the -2( + l), 22 kDa and the zero frame 85 kDa. In pFScass7, the -1 frame is 85 kDa, the -2( + 1) frame 28 kDa and the zero frame, 22 kDa. As can be seen from the translations the constructs produced the expected -1 frameshift products; 22 kDa in pFScass5,28 kDa in pFScass6 and 85 kDa in pFScass7. Thus, all the predicted open reading frames were present and correct. Estimates of the frameshift efficiency of the constructs indicate that some 40% of ribosomes enter the -1 phase at the frameshift site. This is some 10% higher than that seen in the wild-type pseudoknot construct (pFS8) and probably reflects, at least in part, the increased stability of the stem 1 helix in the pFScass constructs. (Using the base-stacking rules of Turner et al. (1988) it can be estimated that introducing the G*U base-pair in stem 1 increases the helix stability by 38 kcal mol-' (1 cal = 4.184 J), from -12.4 kcal mol-' (stem 1 of pFS8) to -16.2 kcal mol-' (stem 1 of pFScass5).) In the pFScass5 translation, the absence of significant quantities of 85 kDa product suggests strongly that frameshifting is strictly into the -1 reading frame. Although a +2 shift would place ribosomes in the same phase, it is very unlikely that this occurs, since the ribosome-bound tRNAs could form only one P-site mRNA-tRNA base-pair and no A-site pairs following a + 2 slip. In addition, we have shown previously that the primary sequence of the triplet downstream from the slip sequence is unimportant in the frameshift process (Brierley et al., 1989) . Furthermore, a + 2 slip is not compatible with the results of amino acid sequencing of retroviral trans-frame proteins (see Introduction).

In order to check that the frameshift observed in pFScass5 was dependent upon the downstream pseudoknot, we made a destabilizing change at the base of stem 1 in pFScass5.15 (see Fig. 5 transcriptions. However, when the T7 promoter was deleted in construct pFScass 5.39 (Fig. 5) , the minor products were still present. It is therefore more likely that these proteins arise as a result of low-level aberrant translation events in the in V&O system. Although it is likely that some minor products may co-migrate with the authentic stopped and frameshift products on the gels, the comparative abundance of the minor bands is low, and hence can introduce only a small error into the calculated frameshift efficiencies.

demonstrated that the precise distance between the slip-sequence and the pseudoknot (6 nucleotides) had to be maintained; increasing or decreasing this distance by three nucleotides greatly reduced or abolished frameshifting (Brierley et al., 1989) . We wished, therefore, to confirm that the six nucleotide spacing between the slip sequence and the minimal pseudoknot structure was optimal. Shown in Figure  5 are the translations of four spacing mutants in which this distance was decreased or increased by one or two nucleotides. In these mutants, the size of the -1 frameshift product was 22 kD in each case, since the mutations were introduced into the background of either pFScass6 (-2 deletion, pSM5, f 1 insertion pSM3) or pFScass7 (-1 deletion, pSM2; +2 insertion, pSM8) in order to assign the -1 frameshift to this size class and ease visualization and quantification of the frameshift efficiencies. The results show that the optimal spacing distance is indeed six nucleotides;

shortening or lengthening the spacing by one or two nucleotides steadily decreased the frameshift efficiency. Spacing distances of five and seven nucleotides gave reasonable frameshifts (297 o/o and 21.7 y. respectively), but with spacing distances of four or eight nucleotides, the efficiency was considerably reduced (6:& and 13.2 "/o respectively).

RNA pseudoknots have been proposed to play a role in the suppression of termination codons in some animal virus systems (ten Dam et aZ., 1990) and are known to be important in translational readthrough of the termination codon that separates the gag and pal coding regions of the retrovirus Moloney murine leukaemia virus (MuLV: Honigman et al., 1991; Wills et al., 1991) . We were interested to see if the IBV pseudoknot could also promote termination codon suppression.

The pFScass6 construct allows such an event to be monitored, since any ribosomes that fail to frameshift and to terminate at the UGA codon downstream from the slip-sequence continue in the zero reading frame and produce an 85 kDa protein.

However, although a distinct 85 kDa product was produced upon translation of mRNA derived from pFScass6 (see Fig. 5 ), the abundance of this product was very low (0.50/,, or less). Thus: it seems that at least at the stop codonpseudoknot ' spacing distance in pFScass6 (3 nucleotides), the TBV pseudoknot cannot suppress signilicantly a UGA termination codon.

(d) Slip-site requirements for eficient frameshifting

Having characterized the pFScass5 construct, we set out to examine the kinds of slippery sequence that could specify efficient frameshifting in the context of the IBV pseudoknot. We concentrated our efforts on the analysis of homopolymeric runs, either fully monotonous (e.g. A,N) or matched pairs of triplets (e.g. A,U,N), since it was known that point mutations that reduce the homopolymeric nature of the slip site greatly diminish or abolish the frameshift (Jacks et al., 1988a,b; Brierley et al.. 1989; Dinman et al., 1991) . Shown in Figure 6 are the results of translations of a large number of slip site variants.

These were chosen to include all those considered t'o be utilized in viva (ten Dam et al., 1990) . Each particular slip site is represented by four constructs, since the identity of the seventh nucleotide of the candidate sequence was varied in each case. The frameshift efficiencies measured for each construct are detailed in Table 1 . Although a large number of constructs were studied, several common features emerge from the analysis. . Slip-sequence requirements for frameshifting. In this Figure, the frameshift properties of a large number of slip-sequence mutants are shown. In most of these, the wild-type IRV slip-sequence UUUAAAC was replaced by a candidate sequence. for example AAAAAAA.

For each class of sequence (e.g. AAAAAAN), N represents either A. C. G or I! at the 7th position of the slip-sequence. The gels show the reticulocyte lysate translation products synthrsized in response to mRNAs derived from BumHI-digested pFScass5 or mutant derivatives. Polypeptides were labelled and analysed as described in the legend to Fig. 5 . The bottom of each track is labelled with a number to identify each mutant,. 1991); G3A,C (ten Dam, Brierley.

Inglis & Pleij, unpublished results) ). An exception to this is the slip sequence of HIV-I, which comprises a homopoly merit U stretch (U,A) and promotes higher levels of frameshifting (5%; Jacks et al., 1988a; Wilson rf al., 1988) . Examination of the level of frameshifting seen when these particular slip sequences are placed upstream from the IBV pseudoknot (Table  1 ) reveals that frameshifting is substantially increased. For the sites that in isolation show very low levels of frameshifting, the pseudoknot amplifies the slip enormously, resulting in efficiencies between lS(>,) and 40%.

Even the "naturally slippery" C,S sequence is stimulated some five-to tenfold. These observations indicate that the pseudoknot can act as enhancer of frameshifting, and this is consistent with models for frameshifting in which the function of the pseudoknot is to impede the progress of ribosomes (Jacks et al., 1988a,b) . It is possible that the pseudoknot can arrest ribosomes in the act of decoding the slip site; perhaps increasing the decoding time such that ribosome-bound tRNAs can re-align on the mRNA with improved efficiency. As can be seen in Figure 6 , however, a considerable number of sites tested do not show high levels of frameshifting.

This suggests that the pseudoknot can amplify the signal only when the slip sequence contains inherent potential for slippage.

In the simultaneous slippage model for -1 retroviral frameshifting (Jacks et al., 1988) , the two ribosome-bound tRNAs paired in the zero frame slip back simultaneously by one nucleotide and each retain two (or more) out of three mRNA-tRNA Table 1 The -I frameshift eficiencies in the analysis shouln in Figure 6 Construct base-pair contacts (see Fig. 1 ). However, as is clear from the results shown in Figure 6 and experiments with the slip sites of RSV (Jacks et al., 19886) and yeast L-A (Dinman et al., 1991) , simply having the potential for the formation of four out of six postslippage pairs is not necessarily commensurate with an efficient slippery sequence. If one considers only the sites that conform to the XXXYYYN motif, there are, from our analysis (see Table I ), two main classes of slip sequence that are non-functional.

In the first class, a C or G triplet is present in the part of the slip-sequence that is decoded in the ribosomal A-site following slippage (XXXCCCN or XXXmN; -1 frame A-site codon underlined). In the second class, the seventh nucleotide of the motif is a G (XXXYYYG).

Thus the only A-site codons that can function in frameshifting in the reticulocyte lysate system are AAA, AAC, AAC and UUA, UUC, UUU. The requirements for specific nucleotides in that part of the slip sequence decoded in the ribosomal P-site were less stringent,. with all four nucleotide triplets functioning to some extent (XXXYYYN; -1 frame P-site codon underlined). However, C-triplets gave, in general. very low frameshift levels, perhaps reflecting the fact that no P-site C-triplets have been described as yet, in naturally occurring slip sequences (ten Dam et al., 1990) .

Examination of the candidate slip sites shown in Figure 6 reveals that within each group the ident,ity of the seventh nucleotide greatly influences the frameshift.

If this nucleotide is a G (A-site codon AAC: or IJUG)? frameshifting is greatly reduced, as was seen in t'he yeast L-A system (Dinman et al.. 1991) . As the presence of the IBV pseudoknot downstream from the slip site amplifies the frameshift level of inherently slippery sequences to easily quantifiable levels, it has been possible to determine a hierarchy of frameshifting with respect' to the identity of the seventh nucleotide. When the A-site codon is AAN. this hierarchy is C>A k IJ>>G; when the codon is UIJN, the hierarchy is U> A -C>G. The likeliest, explanation for these observations is that the tRNAs that decode the A-site codons vary in their ability to frameshift. It is not, known whether this is a result of the presence of unusual "shifty" tRNAs in the ret,iculocyte lysate, or whether the effect is simply due to differing strengths of interaction of the anticodons of normal tRNAs with the A-site codons. This is considered in the Discussion.

(iii) I'nusua,l P-&e combinations Although most naturally occurring slip-sequences contain two homopolymeric triplets, there are three examples (to date) that do not conform to this organization: the slip sequence at the MMTV pro-pal overlap (GGALJUUA; Moore et al., 1987; Jacks et al., 1987) , the related sequence of the red clover necrotic mosaic virus (RCNMV) RNA-l ~27.~01 overlap (GGAUUUIJ; Xiong & Lommel, 1989) and that of the coronavirus-like equine arteritis virus (EAV) lalb overlap (GUUAAAC;

den Boon et al., 1991) . It is possible to accommodate the MMTV pro-p01 and 1. Brierley et al. RCNMV p27-pol slip-sequences within the simultaneous-slip model by invoking the formation of a post-slippage G.IJ mRNA-tRNA base-pair at the non-wobble position 2 in the P-site following a simultaneous-slip (see Table 2A ). Indeed, when we changed the IBV slip-sequence to GGAUUUA (MMTV) or the related sequence GGAAAAC, some 15% and 22% of t,he ribosomes frameshifted at those sites respectively, supporting the idea that a GU pair can be tolerated at this position (see Fig. 7 and Table 2 ). The EAV slip sequence (GUUAAAC), however, is more problematical, in that the simultaneous slip model would predict the formation of an A.U pair at position 2 and a non-Watson-Crick G-A mismatch at the non-wobble position 1 in the P-site. Nevertheless, this slip sequence has been shown to specify a 20% frameshift in the context of an RNA pseudoknot in vivo (den Boon et al, 1991) , suggesting that either the G.A mismatched pair can contribute to post-slippage tRNA-mRNA stability or that, under certain circumstances, stable postslippage pairing is not required. When we tested this slip sequence in the context of the IBV pseudoknot, frameshifting occurred at high efficiency (35% :  Table 2A , Fig. 7) . We further investigated which nucleotide could be tolerated at position 1 of the slip sequence by completing a series of mutants in which the IBV sequence was modified at this position (NCUAAAC) and then prepared a series of other slip sequences in which the first position nucleotide was varied against a background of different first slippery codons (N&AAC, NMUUC, NmAAC, NmtJUC; zero frame, P-site codon underlined). The sites were chosen such that the NAAUUUC NGGAAAC NUUAAAC NUUUUUC ACGUACGUACGUACGUACGU Figure 7 . Unusual P-site combinations. In this Figure, the frameshift properties of a number of slip-sequence mutants are shown. In most of these, the wild-type IBV slip-sequence was replaced by a candidate sequence, e.g. GGAUUUA.

For each class of sequence, N represents either A, C, G or U at the 1st position of the slip-sequence. The gel shows the reticulocyte lysate translation products synthesized in response to mRNAs derived from BarnHI-digested pFScass5 or mutant derivatives. Polypeptides were labelled and analysed as described in the legend to Fig. 5 . class of tRNA decoding t,he slip-sequence at, t hcb P-site varied in each case, since the observation of highly efficient frameshifting at the EA\' slip sequence raised t'he possibi1it.y t,hat' the ability to slip eficiently in the P-site, yet mispair at position I may be related to the t,ype of tRNA in the P-site.

The in vitro translations of the position I mutants are shown in Figure 7 . Tn Table 2B . the frameshift efficiency measured for each construct. the relative efficiency expressed as a proportion of the eficiency of t,he parental construct and the likely post,-slippage mRNA-tRNA contacts formed are shown. Tt is clear from the results that. in most cases. a mutation in position 1 greatly reduces the frameshift, eficiency and this supports the view that,, in general. stable post-slippage tRh'A-mRNA pairing is important in the frameshift process. However. a number of the position 1 mutants retained the ability to specify efficient frameshifting, namely GAAC'ULTC (60°io of the parental level), ACC 'AAAC' (34:i, , ) , GLXJA,4A (' (84'&,) and C:UUI:CUC (51%). The overall eficiency of frameshifting within each group did depend to some extent on thr Psite tRNA: I 'I'A and I'l.:l:-decoding tRSAs were, in general, more slippy than AAA and GC:A-decoding tRh'As. However. it was conspicuous that wit,hin each group the most efficient mutant sites were those in which a G nucleotide was present at position I : whether forming a C+rT or a (>.A pair post-slippage.

The explanation for this observation is at. present unclear. ln a further experiment. we tested the sequence GUCAAA('. which ought to retain. in terms of standard Watson-(Trick base-pairing, following slippage. a sole UC base-pair at position 2 in the P-sit'e. This construct (see Fig. 7 and Table  %A ) displayed a frameshift, eficiency of 14!,,, despite the apparent paucity of pairing in the P-site. Frameshifting can t,hus occur at high efficiency under certain circumstances. with minimal basepairing in t,he P-sitrl.

(iv) Frameshift dirrction In the majorit'y of slip-sequence variants t'est,ed in this study, frameshift'ing was strictly into the -1 reading frame. However, for a number of sites. an 85 kDa product was seen, and this corresponds to ribosomes shifting into the -2 reading frame (or t,he + 1 frame; we cannot, at present, distinguish between the 2 possibilities). In all the cases in which t,his product appeared, the slip sequence contained a I' triplet in the A site (i.e. S ',U,N) or was fully monotonous (S,). The abundance of the -2/ + 1 frameshift product was, in general, extremely low (efficiency less than 0.50/). However, when monotonous r stretches (C,N) were tested, significant levels (2 9,, to 5(!6) of -2/ + 1 frameshifting were observed. Indeed in the U7 construct (pFRcasG.lO), which was a stretch of eight U bases (since the c&on immediately downstream from the slip sequence was the zero-frame termination codon, CGA), frameshifting was remarkably efficient, with some 39:& of ribosomes entering the -1 phase, and and the presence (PK+) or absence (F'ti-) of a pseudoknot in bhese constructs is indicated above earh track.

In the PK-constructs, the pseudoknot was destabilized by converting 4 C nucleotides at the base of stem I of the pseudoknot to C: residues (as in construct 51.5. see Fig. 5 ). The polypeptides produced by each mutant, were labelled and analysed as described in the legend to Fig. Ti. a further 21 '!o, the -2/ + 1 frame. We wished to test whether -2/ + I frameshifting was sensitive to the presence of the RNA pseudoknot and so made two further constructs in which the pseudoknot structure downstream from the li,C and (1, constructs was destabilized by making a complementary change in stem 1 (to create pFScass5.83 (U&J, PK ) and pFScass5.14 (Us, PK-. see Fig. X ): the complementary change in stem 1 was identical with t'hat made in pFScass5.15. see Fig. 5 ).

Removal of the pseudoknot in the C,C construct greatly reduced the level of both -I (from 27(yo to 2%) and -2/+ 1 (from 2.2(+, to less than 0.5(&) frameshifting: raising t#hr possibility that both classes of frameshift are dependent upon the RNA pseudoknotJ. However, removal of the pseudoknot, in the U, c*onstruct had a much less dramatic effect on frameshifting; with the -1 level reduced by about half (to 21.7?/,) and the -2/ + 1 level only marginally (from 21 (!/o to 200/b). Frameshifting on t,hr U, stretch therefore appears to be relatively indeperldent of downstream RNA structure. Runs of IT bases in coding sequences have been implicated in frameshift events in other systems, including the yeast mitochondrial oxi 1 gene (Fox 8 Weiss-Krummer. 1980 ), bacteriophage 1'7 gene IO (Dunn $ St'udier, 1983 ) and t'he trpE gene of' Salmonella tyiphimurium (Atkins et al., 1983) . The high levels of frameshifting associated with the 'I:, sequence suggests that when ribosomes encounter this stretch t'he ability to monitor reading frame is lost, since t,hey are distributed into each of the three reading frames at high efficiency (39oh, -1; Sl%, -2/+ 1; 40 %, zero frame). Furthermore, even when the downstream pseudoknot was destabilized, reading frame maintenance was still aberrant (22 %, -1; 20%, -2/+ 1; 58%, zero frame). Why the two sequences respond in different ways to the removal of the pseudoknot is not clear, but possible explanations are given in the Discussion. Frameshifting at the gag-pol overlap in HIV-l occurs at a UsA sequence and evidence has accumulated that, unlike most frameshift sites, no downstream RNA structures appear t,o be involved in frameshifting in this system (Wilson et al., 1988; Madhani et al., 1988) . The possibility that a -S/+ 1 frameshift product could be produced from this sequence in &vo has been raised (Kingsman et aZ., 1990) . Our analysis of the U,C construct described here suggests that such a frameshift product would be of low abundance.

(a) The A-site slippery codon: tRNAs and sibosomal framesh@ing

In the reticulocyte lysate system, and in the context of the IBV pseudoknot, only six codons are functional at the A-site, namely AAN (where N = A, C or U) and UUN (where N = A, C or C). Triplets of G or C and codons ending in G gave very low levels of frameshifting at this position. It is clear from this study that the IBV pseudoknot can act as an enhancer for the frameshift process, and appears to be considerably more effective than the pseudoknots predicted to form downstream from the RSV (Jacks et al., 19886) and L-A (Dinman et al., 1991) slip sites. When the slip-sequences of RSV (A,UsA) and L-A (G,U,A) were placed upstream from the IBV pseudoknot, frameshifting at these sites was at a four-and ninefold greater level, respectively, t.han that seen in their natural context)s. The enhanced frameshifting afforded by the IBV pseudoknot allowed the determination of a hierarchy of frameshift efficiency of A-site codons. ln order to explain this hierarchy, we propose that slip sequences are decoded by the normal pool of cellular tRNAs and the ability to slip at a particular codon is determined by the st'rength of the codonanticodon interaction in the ribosomal A-site before tRNA slippage. We suggest that if this interaction is relatively weak, then slippage is more likely to occur. The ability to form two or more post-slippage mRNA-tRNA contacts is taken as a prerequisite for frameshifting at the A-site. The strength of the interaction between mRNA and tRNA is likely to he influenced considerably by the kind of base-pair that forms between the 3' base of the codon and the 5' base of the anticodon (position 34) at the wobble position (Crick, 1966 ; for a review, see Bjiirk et aZ., 1987) . On this basis, we have surveyed the nature of the anticodons of the tRNAs predicted to be decoding the second slippery codon. We limited the analysis to tRNAs of non-transformed mammalian cells, excluding mitochondrial tRNAs and form our arguments on the basis of only those tRNAs whose sequence and modification pattern have been determined. As a source of information, we used the tRNA sequence compilations assembled by Sprinzl et al. (1989) and the EMBL and NEWEMBL databases.

For the AAN hierarchy, AAA and AAG are decoded by tRNALy"; AAC and AAU by tRNA*"". In mammalian cells, two lysine-acceptor tRNAs are present, tRNALyS U9UU (anticodon 5' U9UU 3'; U9 = 5-methoxycarbonylmethyl-2-thiouridine) and tRNALY" CUU (Raba et aZ., 1979) . On first inspection, one may expect tRNALyS U'UU to recognize both the AAA and the AAG lysine codons, and tRNALyS CUU to recognize only AAG. However. the t J9 modification influences the codon preference of tRNALyS U9UU such that it preferentially recognizes the AAA codon with little aiIinity for the AAG codon (Lustig et al., 1981) . Thus, it, is likely that the AAA and AAG codons are decoded by a separate tRNA. As the AAGCUU codon-anticodon pairing is likely to be more stable than the AAA.UUU pairing, then, according to our hypothesis, AAG should frameshift poorly, since it forms strong pre-slip contacts with the tRNA. This is supported by the lack of frameshifting seen in slip-sequences ending with AAG. The AAA codon, however, would form a weaker pre-slip contact and, as is seen, may be expected to frameshift at a higher level. In the case of tRNA*"", which decodes AAC and AAU, only one tRN*4 species has been detected to date: tRNA*"" (antieodon 5' QUU 3'; Q = queuosine: ('hen & Roe. 1978 ('hen & Roe. , 1980 Roe et al., 1979) . The Q modification appears to influence codon choice in such a way that tRNA*"" QUU recognizes both AAC and AAU. with a slight preference for AAU (Meier et al., 1985) .

If the wobble base-pairs between Q-C and Q-U are less strong than a GC pair, then one may expect both AAC and AAU to be slippery codons. As tRNA*"" &IX has a slight preference for AAU, one would expect AAC to be more slippery, and this correlates with the experimental observation. Thus, for the AAN series, there is some support for the observed hierarchy C > U-A>>G, based on the strength of the pre-slip codon-anticodon pairing.

For the UUN series, UUA and UUG are decoded by tRNALe"; UUC and UUU by tRNAPh'. Of the mammalian leucine tRNAs described to date, there are two species that are likely to decode UUA and UI'(:; namely, tRNALe" (IAG; I = inosine : Pirtle et al., 1980 ) and tRNAL '" (CAG: Sprinzl & Gauss, 1983) . The presence of inosine at the wobble posit'ion allows the formation of 1.A. IC and I.lJ but not 1.G base-pairs. Thus, tRNAL'" IAG recognizes the UI'A but not the UUG codon. The isoacceptor tRNALe" CAG will recognize only the CUG codon and may be expected to form a strong CG wobble pair and not to frameshift efficiently. The 1.A wobble pair formed upon decoding I'UA by tRNALe" IAG would be weaker than t,he CG pair and thus one would expect UUA t,o frameshift at a higher efficiency than UUG, as is seen. In the case of t,RNAPh'. which decodes UIJC and UI'I:. only one tRNA species has been detected to date, tRNAPh" (G3AA; G3 = 2'-o-methylguanosine: Roe et al., 1975; Keith & Dirheimer, 1978; Lin et aE., 1980) . Unfortunately, the effects on codon choice of this modification (if any) have not been studied, so it is not possible to comment on the likely affinit,y of this tRNA for the UUC and UUU codons. If the frameshift hypothesis is correct, however, one would expect this tRNA to recognize both UUC and C'I'U with a slight preference for UUC. However, the G3C and G3.U pairs would be predicted to be less strong than an unmodified GC pair, since both UUC and UIJIJ codons allow efficient frameshifting. 5O'jb. 'l'suchihashi (1991) has proposed an explanation for this observation hased on the knowledge that only one tRNALyS species has been reported in E. c'ol'i; tRNALy" 5' 1J81JIJ 3' (IJ8 = 5-methylaminomethyI-X-thiouridine: (Jhakraburtty et al., 1975) .

Tsuchihashi (1991) suggests that as the IJ8 modificaat,ion appears to weaken the interaction bet.ween this tRNA and the AAG codon (Yokoyama rt al., 1985) , then the recognition of AAQ may be poor and the tRNALyS able to slip at high efficiency. In E. coli. therefore.

AAG. in contrast to the situation in eukaryotes, is a very slippery codon. and this mav well be a consequence of having only a single tRNALys species. In eukaryotes, t'he presence of t,he tRNALyS CUlJ isoacceptor may well prevent frameshifting at the AAG codon in the context of "ret'rovirus-type" frameshift signal. Hatfield and colleagues have proposed an alternative explanation for the limited number of codons known to be functional in frameshifting, in which frameshifting is mediated by a subset' of hypomodified isoacceptor tRNAs (Hatfield et aZ.. 1990 ). The tRNAs t'hat would decode the t'hree A-site codons found in nat,urally occurrin are tRNA*"" (codon AAC), r+h;li~ti;~ncc~ tRNA tRNAL'" (GUA). These tRNAs are characterized by the fact that tRNA*"" contains the highly modified Q base in the 5' position of its anticodon. tRNAPh' the highly modified Wyebutosine (Y 1) or Wyebutoxosine (Y2) base (Sprinzl et al. 1989 ) just to the 3' side of the anticodon, and t,RNAL'" lacks a highly modified base in its anticodon loop. The hypothesis suggests that hypomodified varianm of tRNAAs" and t'RNAPh' (and the already "hypo"-modified tRNAL'") would be able to slip at frameshift sites, since the lack of a highly modified base would create more space in and around the frameshift site, which may facilitate frameshifting by allowing greater flexibility of movement of the anticodon (HatfieId et al., 1989; Hatfield & Oroszlan. 1990 ). Experimental support for the idea that hypomodified variants of tRNA*"" and tRNAPh" could be involved in the frameshift process comes from an analysis of the modification status of these tRNAs in cells infected with various retroviruses (Hatfield et al., 1989 Almost all of the mutants tested in the Psite analysis described here were concerned with position 1 of the slip-sequence, since we were intrigued by the observation that the slip-sequence of EAV (GUUAAAC) was functional in viwo (den Boon et al.. 1991) . In our assay, this sequence stimulated frameshifting at a level of 35%. only slightly less efficient than the wild-type IBV sequence (CUUAAAC). This raised the possibility that the predicted postslippage base-pairs, G-A at position 1 and U.A at position 2, may be sufficiently stable to allow frame shifting.

An alternative possibility was t.hat the leucine tRNA decoding UUA had some unique property that facilitated slippage without the formation of strong post-slip contacts with the mRNA. To test t.his, we prepared a number of position 1 variants of a range of slip-sequences and tested frameshifting.

It was found that in most cases, and in agreement with the necessity for retention of the homopolymeric nature of the slip-sequence, frameshifting was greatly reduced. In four of the mutants, however, significant frameshifting was observed (from 3596 to 135% of the parental level). Three of these had G at position I (GUUAAAC, GUUUUUC, GAAIXUC) and it was clear that in all the mutants tested, the presence of G at the first position conferred the highest frameshift level within each class. The fourth functional position 1 mutant was AUUAAAC.

We have been unable to arrive at' a satisfactory explanation for these observations. One possibility is that during a frameshift, unusual basepairs can form and are stable at position 1. Current models for the frameshift process suggest that the RNA structures downstream from the slip-site may function to stall ribosomes over the slippery sequence codons; this pause could directly influence the fidelity of mRNA decoding, and promote the realignment of ribosome-bound tRNAs into the -1 reading frame (Jacks et al., 1988a,b) . Thus, the postslip G.G, C.A and A.A pairs that may form in the functional position 1 mutants (see Table 2 ) could be allowed under the circumstances of a paused ribosome. Surprisingly, the sequence GUCAAAC was also found to stimulate reasonable levels of frameshifting (140/6), despite being expected t'o form only a G-A mismatch at position 1 and a LJC pair at po&ion 2 post,-slippage (see Table BA ). Thus, certain sequencaes do not appear to adhere strongly to the two-out-of-three post-slip contact,s code of the simultaneous slip model. Clearly, more informa.tion is needed before the molecular basis of this P-site misreading is understood, particularly wit,h respecxt to the functionality of G-nucleotides at position 1.

In our analysis of the direction of frameshifting, we have found that at the majority of slip sequences, frameshifting was strict.ly into the -1 reading frame. Significant quantities of -2/+ 1 product, however, were seen when fully homopolymerit U runs (I:,N) were tested (2 to 5%) and a US stretch gave high levels (21%). An important question concerning the frameshift process is whether RNA pseudoknots can influence frameshift direction. Tn the simplest case, RNA pseudoknots may function solely to pause ribosomes; such ribosomes can then frameshift in a direction determined by the nature of the slippery sequence. In the majority of frameshift sites, the likeliest slip is a -1 slip, since this produces the most stable post-slippage mRNA-tRNA base-pairing. In the case of homopolymeric runs, identical post-slip contacts can form following -1 or -2 slips (e.g. in the case of the U,C sequence) or even -1, -2 or + 1 slips (e.g. in the case of the UB sequence). If the pseudoknot plays a role in determining the direction of frameshifting, however, then in the presence of a pseudoknot one would expect a particular class of frameshift to predominate. We investigated this possibility by destabilizing the pseudoknot structure in constructs containing a U,C or U, slip-sequence and tested for frameshifting (see Fig. 8 ). In the U& construct, this resulted in a reduction of both -I and -2/+ 1 frameshifting. In the 1;s construct. however, removal of the pseudoknot had little effect on -2,' + 1 frameshifting, but' reduced -1 frameshifting by about half. These apparently inconsistent effects can be explained if the pseudoknot is able to specifically enhance 5'.wards slippage (minus direction). Tn the C,C construct, the 85 kDa -2/+ 1 product probably arises from a -2 slip rather than a + 1 frameshift, since the phenylalanine tRX'A decoding the A-site codon can form identical post-slip contacts with the mRNA following either a -1 or -2 slip, but not from a + 1 slip. If the pseudoknot does enhance frameshifting in t,he -5' direction, t,hen the reduction in both -1 and -2/( + 1) frameshifting observed in the U,C PK-construct. is perhaps expecbed. In the case of the I:, construct, destabilization of the pseudoknot once again caused a reduction in -1 frameshifting but. in this case, little effect on -2/ + 1 frameshifting was seen. This may be related to the likelihood that both forward and backward slips can occur at the U, sequence: perhaps in the absence of a pseudoknot. + 1 frameshifts occur. whereas in the presence of the structurr, frameshifting may occur mainly into the -2 (and -1) reading frame. As we are unable to dist#inguish between -2 and + 1 shifts in this assay, these int)erpretat)ions must be viewed as speculative, but, the results lend some support to t.he idea that RNA pseudoknots can influence the direction of frameshifting.

In summary, our investigation int,o the magnitude and directionality of ribosomal frameshifting described here offers an explanat,ion for thr observation that many natura,l slip-sequences employ imperfect. tandem slippery codons (e.g. IHV, l'l'l:A.4AC; RSV AAAUI'UA; HIV-1 : ITlXJl'lJUA).

Firstly, for those sequences ending in .4Ax: a pezect second slippery codon (AAA) would frameshift less well: hence all natural sit.es of this class utilize the AAC codon (ten Dam et nl., 1990) . Indeed. recent work with the MMTV frameshift signal has shown that AAC is the optimal codon (Chamorro et al., 1992) . Secondly, in thcl case of those sequences ending with I'IZ. although perfecting t.he triplet t,o lXX: may increase t,he frameshift efficiency, it, may also start t.o generate undesirable -2; + 1 frameshifts. ki'r thank the Agricultural and Food ResearcLh Council and the Medical Research Council for supporting this investigation.

We are grateful to Xi&y Rolley for her assiat.ance in the early part of this work. Many thanks to Dolph Hatfield for communicating informabion before publication.

The nucleotide sequence of the first externally suppressible -I frameshift mutant and of some nearby leaky frameshift mutants

Ribosome gymnast,ics-degree of difficulty 9.3, style 10.0. CelI

Transfer RNA modification

(lompletion of the sequence of the genome of the c*oronavirus avian infectious bronchitis virus

An eflicient ribosomal frame-shifting signal in the polymerase-encoding region of the coronavirus IBV

haracterisation of an efficient coronavirus ribosomal frameshifting signal: requirement for an RSA pseudoknot

MutatIonal analysis of the RNA pseudoknot. component of a coronavirus ribosomal frameshifting signal

An RPiA pseudoknot and an optimal heptatneric shift site are required for highly efficient ribosomal frameshifting on a retroriral messenger RNA

The nucleotide sequem:e of rat. liver t.R~~4As". Hiochew. Biophys. Rrs

Structural comparison of human, bovine, rat and Walker 256 carcinosarcoma asparaginyl-tRli.4

Codon-anticodon pairing. The wobble hypothesis

Eyuine arterit,is virus is not a togavirus but belongs t.o the csoronaviruslike superfamily

pEMBL: a new-family of single-stranded plasmids

Complex formation between influenza virus polymerase proteins expressed in Xenopus oocyt,es. l'irology

A -I ribosomal frntneshift in a douhle stranded RNA virus of yeast forms a gag-p01 fusion protein

Functional analysis of bacteriophage fl intergrnic region. l-iroloyy

Complete nuc:leotide sequence of bacteriophage T7 DNA and the lorations of Tl genetic elements

An introduction to polyacrylamide gel electrophoresis

The wwhere, &at and hou: of ribosomal fratneshifting in retroviral prot,ein synthesis

Chromatogra,phic analysis of t)he aminoacyl-tRxAs which are required for translation of codons at and around the ribosomal frameshift, sites of HIV. HTLV-1 and BLV. I'iroloc~?/, 173. 73674%. Hat,field, I)., Smith. I)

hara.cterisation of mouse mammary tumour virus :/ay-pol gene products and t'he ribosomal frameshift by protein sequencing. t'roc

Wolf. I)

ris acting RNA sequences control thr gag-pal transla.tion readthrough in murine Ieukaemia virus

Thr ril)osomal frameshift signal of infectious bronchitis virus. In f'ost-trn?lsaription,al flontrol of &nf~ Exprrssion

Two t+ic~ient rihosomal frameshifting evc,nts are rrcluirtrd for synthesis of mouse mammary tumor virus ycry-related polyprotrins

haracterisation of ribosomal frameshifting in HIV-1 !lay-f,ol expression. Maturr (Lo&on)

Signals for rihosomal frameshifting in the Rous sarcoma virus gag-pal region. ('P/I. 55

The primary structure of rabbit. calf and bovine liver tRsAPh

HIV ~01 expression via a. ribosomal frameshifi. Tn f'o.st-tmnsoiptional

sate ASI series H: (Tell Riology

Func~t~ional messenger RNAs are produced by SP6 in vitro transcription of c*lonetl v1)X.h. A'~ucl

Rapid and efficient, site-specifics tnutagenesis without phenotypic selection

The nuoleotidr sequence of two bovine lens phenylalanine tRKAs

Complete nucleotide sequence of a milk-transmitted mouse mammary tumor virus: Two frameshift suppression events required for translation of gag and pal

The nucleotide sequence of a major species of tRNA from bovine liver

RNA pseudoknots: Structure, detection and prediction

E. coli ribosomes rephase on retroviral frameshift signals at, rates ranging from 2 to 50 percent

Evidence that a downstream pseudoknot, is required for translational readt,hrough of the Moloney murine leukaemia virus gag stop codon

Nucleotide sequence of three isoaccepting lysine tRNAs from rabbit liver and SV40-t,ransformed mouse fibroblasts

HIV expression strategies: ribosomal frame-shifting is directed by a short sequence in both mammalian and yeast systems

Sequence studies on tRNAPh' from human placenta: ('omparisons with known sequences of tRNAPh" from ot,her normal mammalian tissues

The complete nucaleotide sequence and genome organisation of red clover necrotic mosaic virus RNA-I

Improved Ml3 phage cloning vectors and host strains: Nucleotide sequences of the M13mp18 and pCiC19 vectors

Comparison of rat liver and Walker 256 carcinosarcoma tRNAs

An improved filamentous helper phage for generating singlestranded plasmid DNA

DNA sequencing with chain-terminating inhibitors

Molecular mechanism of codon recognition by tRNA species with modified uridine in thr first position of the anticodon

Cloning and expression of influenza virus genes

Lustig, F.. Elias, P., Axberg, T., Samuelsson, T., Sprinzl, M. & Gauss, D. H. (1983) . Compilation of transfer Tittawella, 1. & Lagerkvist, U. (1981