key: cord-0928857-1m9s5bus
authors: Brierley, Ian; Rolley, Nicola J.; Jenner, Alison J.; Inglis, Stephen C.
title: Mutational analysis of the RNA pseudoknot component of a coronavirus ribosomal frameshifting signal
date: 1991-08-20
journal: Journal of Molecular Biology
DOI: 10.1016/0022-2836(91)90361-9
sha: 21d2a4fc141bfe400dda7eda59f19c34480365a3
doc_id: 928857
cord_uid: 1m9s5bus

Abstract The genomic RNA of the coronavirus IBV contains an efficient ribosomal frameshift signal at the junction of the overlapping 1a and 1b open reading frames. The signal is comprised of two elements, a heptanucleotide “slip-site” and a downstream tertiary RNA structure in the form of an RNA pseudoknot. We have investigated the structure of the pseudoknot and its contribution to the frameshift process by analysing the frameshifting properties of a series of pseudoknot mutants. Our results show that the pseudoknot structure closely resembles that which can be predicted from current building rules, although base-pair formation at the region where the two pseudoknot stems are thought to stack co-axially is not a pre-requisite for efficient frameshifting. The stems, however, must be in close proximity to generate a functional structure. In general, the removal of a single base-pair contact in either stem is sufficient to reduce or abolish frameshifting. No primary sequence determinants in the stems or loops appear to be involved in the frameshift process; as long as the overall structure is maintained, frameshifting is highly efficient. Thus, small insertions into the pseudoknot loops and a deletion in loop 2 that reduced its length to the predicted functional minimum did not influence frameshifting. However, a large insertion (467 nucleotides) into loop 2 abolished frameshifting. A simple stem-loop structure with a base-paired stem of the same length and nucleotide composition as the stacked stems of the pseudoknot could not functionally replace the pseudoknot, suggesting that some particular conformational feature of the pseudoknot determines its ability to promote frameshifting.

Viruses employ a wide variety of strategies to co-ordinate and control gene expression. Over the last few years it has been recognized that a number of viruses of higher eukaryotic organisms, particularly the retroviruses, utilize ribosomal frameshifting to control expression of their replicases (for a review, see Craigen & Caskey, 1987; ten Dam et al., 1990) . Ribosomal frameshifting is a directed change in translational reading frame which allows the production of a single protein from two (or more) overlapping genes, and was first described for the vertebrate retroviruses, Rous sarcoma virus (RSVf) (Jacks & Varmus, 1985) and mouse mammary tumour virus (MMTV) (Moore et al., 1987; Jacks et al., 1987) . Retroviral frameshifting appears to be a mechanism for regulating the expression of the viral RNA-dependent DNA polymerase; one termination codon in RSV and two in MMTV are suppressed by -1 ribosomal frameshifts to generate the Gag/Pal (RSV) and Gag/Pro/PO1 (MMTV) polyproteins from which the viral polymerases are subsequently $ 

Although there is little experimental evidence, it employed by the avian coronavirus infectious bron-is highly likely that pseudoknots are more generally chitis virus (IBV). The 5' end of the TBV genomic involved in the process of eukaryotic ribosomal RNA contains two briefly overlapping open reading frameshifting, since sequences capable of forming frames (ORF) la and lb (formerly Fl and F2), with such structures can be found at the frameshift sites lb in the -1 reading frame with respect to la of many retroviruses (Brierley et al., 1989 ; ten Dam (Boursnell et al., 1987) . We have shown that the lb et al., 1990 ) and a number of other viruses suspected ORF is expressed as a fusion with the upstream la to use frameshifting (ten Dam et al., 1990) . Indeed, ORF following a highly efficient (30%) ribosomal a pseudoknot structure has been implicated in the frameshift event that takes place within the overlap frameshift signal of RSV (Jacks et al., 1998) . region (Brierley et al., 1987) . More recently, we Pseudoknots were first identified as structural investigated the precise sequence components of the elements at the 3' end of certain plant viral RNAs frameshift signal and defined an 86 nucleotide (Rietveld et aE., 1982 (Rietveld et aE., , 1983 (Rietveld et aE., , 1984 Joshi et al., 1983 ; stretch that is in itself sufficient to direct frame- van Belkum et al., 1985) . It has been suggested that shifting in a heterologous genetic context (Brierley the formation of such structures in RNA may not be et al., 1989) . The signal is composed of two distinct uncommon (Pleij et al., 1985) and examples have elements; a heptanucleotide "slippery" sequence, been found in viral and messenger RNAs, in ribo-UUUAAAC, the probable site of the ribosomal slip somal RNAs and potentially in the catalytic sites of (Jacks et aE., 1988 ) and a tertiary RNA structure in some ribozymes (for reviews, see Pleij & Bosch, the form of an RNA pseudoknot downstream from 1989; Pleij, 1990) . Almost, all pseudoknots identified this sequence. We were able to establish the pre-to date, including the IBV pseudoknot, are of the sence of this tertiary RNA structure by creating, hairpin-loop type (ten Dam et al., 1990) , which form using site-directed mutagenesis, a number of when nucleotides in the single-stranded loop of a complementary and compensatory base changes hairpin-loop base-pair with a complementary within the predicted RNA helices of the pseudoknot sequence elsewhere in the RNA chain (see Fig. l(a) ). and then testing the ability of synthetic RNA tran-

The resulting configuration contains two basescripts containing the mutant sequences to promote paired stems, Sl and 82, which are thought to stack frameshifting in a cell-free translation system. These coaxially to form a quasi-continuous, extended experiments indicated that the pseudoknot was double-helix. The stacked helices are assumed to essential for high efficiency frameshifting and, in adopt the conformation of the A-type REA helix addition, had to be positioned at a precise distance and are connected by single-stranded loops I,1 and downstream from the slippery sequence for frame-L2, which span the major and minor grooves of the shifting to occur. helix, respectively. Stacking of the two stems in a . Shown in Figure  l (b) is the proposed structure of the IBV pseudoknot, based on nucleotide sequence analysis and our initial mutagenesis data (Brierley et al.. 1989 ) and conforming to current building principles (Pleij et aE., 1985; Pleij & Bosch, 1989) . In this model, the stems formed by base-pairing between the PKl . PK3 and PK2. PK4 sequences have been stacked coaxially to generate a 16 base-pair quasi-continuous helix with one mismatched pair (G7-A24) in Sl. The connecting loops Ll and L2 contain two and 32 nucleotides, respectively.

The mechanism by which the pseudoknot promotes frameshifting

is not yet clear, but it has been suggested that the elongating ribosome encounters and is required to unwind the pseudoknot whilst translating the slippery sequence codons, and that' this interaction may slow or stall the passage of the ribosome along the mRNA, promoting a (-1) frameshift' at the slippery sequence (Jacks et al., 1988; Brierley et al., 1989) . A clearer understanding of how such events could occur, however, has been hindered by uncertainties over the precaise structure of the pseudoknot. In addition, the potential contributions of individual elements of the pseudoknot structure and, indeed, of particular nucleotides within the structure to the frameshift process remain poorly characterized. We therefore set out to investigate the structure of the TKV pseudoknot by site-directed mutagenesis on the premise that nucleotide changes predicted to destabilize the structure should be inhibitory to frameshifting.

Our results largely confirm the model shown in Figure l (b) and suggest that frameshifting in thr IBV system is not dependent on any primary sequence determinants within the pseudoknot; as long as t,he overall structure is maintained, frameshifting occurs at high efficiency.

We further show that) the pseudoknot cannot be replaced in the frameshift signal by a simple hair-pin. Thus, the contribution of this tertiary structure is not simply due to it,s energetic stability, but rather to its unusual (Lonformation.

Site-sprcific mutations within the IBV frameshift region were prepared using a procedure based on the method described by Kunkel (1985) (Brierley et al., 1989) . Plasmid pFS8 (or mutant derivatives, see below) contains the intergenic region of the filamentous bacteriophage fl (Dotto et aI.; 1981) enabling single-stranded pFS8 DNA to be generated following infection of plasmid-carrying bacteria with bacteriophage R408 (Russel et al., 1986) . I'racil-containing.

single-stranded DXA substrates for mutagenesis were prepared by R408-superinfection of plasmid-carrying Escherichia coli RZ1032 cells (dutuny-: Kunkel, 1985) . Mutagenic oligonucleotides were svnthesized using an Applied Riosystems 381A DNA synthesizer and the mutagenesis reactions performed as before (Brierley et al.. 1989) . Mutants were identified by dideoxy sequencing (Sanger et al., 1977) of single-stranded DNA templates rescued from E. coli ,JMlOl (Yanisch-Perron et al.. 1985) .

(b) ('onstruction of plasmids Plasmid pFS8 (see the text) was constructed from plasmid pFS7 (Brierley et aE., 1989) by introducing an oligonucleotide (sequence 5' AATTAATACGACTCACTA-TAGGGAGA 3') containing the bacteriophage T7 RNA polymerase promoter just downstream from the bacteriophage SP6 RNA polymerase promoter of pFS7, bv sitedirected mutagenesis. Plasmid pFS8.47, a derivative of pFS8 was constructed as follows. Firstly, a unique XhoI restriction endonuclease cut-site was introduced into loop 2 of the IBV pseudoknot (see the text) at position 12,406 in the IBV genome (Boursnell et al., 1987) by insertion mutagenesis to create plasmid pFS8.1. Plasmid pVBZ+ (Brierley et al.. 1987) Plasmids for transcription were prepared hy t,he alkaline lysis mini-preparation method (Hirnboim & Daly, 1979) and linearized by digestion with SmaI. Digests were extracted once with a mixture of phenol and chloroform (1 : 1, by vol.) and the aqueous phase passed through a Sephadex G-50 spin column (Maniatis et al., 1982) equilibrated with water. Linearized template was concentrated by precipitation with ethanol and transcribed as described (Brierley et al., 1987) . except that T7 RPU'A polymerase (Gihco-Bethesda Research Laboratories) replaced SP6 RNA polymerase and the concentration of each ribonucleotide in the reaction was doubled (to I mM each of ATP, CTP. I:TP and 0%)5 mM-GTP). Purified mRNAs were translated in rabbit reticulocyt,e lysates as described (Brierley et al., 1987) and translation products analysed on SDS/ 10 y0 (w/v) polyacrylamide gels according to standard procedures (Hames, 1981) . The relative abundance of non-frameshifted or frameshifted products on the gels was estimated by scanning densitometry and adjusted to take into account the differential methionine content of the products. Linearization of the plasmid with SmaI and in vitro transcription using T7 RNA polymerase yields an mRNA (2.8 kb)

which, when translated in rabbit reticulocyte lysates, produces a 45,909 Da product corresponding to ribosomes that terminate at the la termination codon within the la/lb overlap region, and a 95,990 Da (-1) frameshift product corresponding to a PBl (5')-la-lb-PBl-(3') fusion protein. (b) Reticulocyte lysate translation products synthesized in response to mRNA derived from SmaI-digested pFS8. The RNA was translated and products were labelled with [35S]methionine as described in Materials and Methods. Polypeptides were separated on an SDS/lOO/o polyacrylamide gel and detected by autoradiography.

frameshift process was to create defined mutations within the structure and to assess the effect of these changes on the efficiency of frameshifting in a cellfree translation system. The frameshift assay shown in Figure 2 was almost exactly as described (Brierley et al., 1989) , except that synthetic mRNAs were generated using bacteriophage T7 RNA polymerase. Plasmid pFS8 has a 230 bp cDNA region derived from the IBV la/lb overlap region (and containing the essential 86 nucleotides) cloned within a reporter gene (PBl of influenza virus A/PR8/34) which, in turn, is flanked by the 5' and 3' non-coding regions of the Xenopus fi-globin gene (Krieg & Melton, 1984) downstream from a T7 promoter. Linearization of the plasmid with SmaI followed by transcription results in the production of a capped and polyadenylated 2.8 kb mRNA designed such that on translation in rabbit reticulocyte lysates, ribosomes which terminate at the la ORF stop codon produce a 45,000 Da product (the "stopped" product) and those that frameshift, a 95,000 Da product (Brierley et al., 1989) . In our analysis of the pseudoknot, we selected four particular features for study; the stems, the nucleotides where the stems are thought to stack, the G7-A24 mismatched pair and the loops (see Fig. l (b) and Fig. 3 ). We first examined the effect on frameshifting of disrupting and reforming the basepaired regions within Sl and 52 predicted by our model (Fig. l(b) ) and the results of these changes are shown in Figure 3 (a). For each region studied, our strategy was to change the nucleotides of each strand of the predicted base-paired regions to their complementary nucleotides in separate constructs, and then to create double-mutant, pseudo-wild-type constructs in which both changes are made and so should be compensatory. We also introduced a number of additional point mutations into the stems, and these are shown in Figure 4 . In the main, the results of the stem analysis support strongly the idea that the overall stability of each of the pseudoknot stems is related to its ability to promote frameshifting. As can be seen in Figure 3 , all the complementary changes predicted to destabilize the stems reduced or abolished frameshifting.

In the double-mutant, pseudo-wild-type constructs, in which the stems are predicted to be restabilized, frameshifting was restored to high levels (15 to 30%) in all cases. Consistent with the importance of stem-stability to the frameshift process is the observation that the efficiency of frameshifting was less dramatically reduced (to about 15%) in mIcEID Em 6.62 6.57 6.61 less; (+) 10 to 20%; (+ +) wild-type (25 to 30%). The shaded nucleotides in the stems represent the 2 blocks of nucleotides that were tested by this method and shown to be base-paired in a previous analysis (Brierley et aE., 1989) . 95 l kDa* 45 kDa* constructs where the substitutions were at the very ends of the Sl helix (pFS8.32: 856, Fig. 3(a) ). These changes would be expected to destabilize the structure only slightly and this is supported by estimations of the expected change in free energy upon formation of the mutant structures compared to that of the wild-type structure using the base-pair stacking rules of Turner et al. (1988) . As it is not possible to apply these rules to pseudoknots as a whole (since the contribution of the loops and the stacking of 52 upon Sl to the free energy of formation of the pseudoknot is not known), we considered only the Sl stem-loop structure in isolation. The changes in free energy calculated were -12.4 kcal mol-' for wild-type Sl, -11.4 kcal mol-' for Sl in pFS8.32 (Gl-G30 at the base of Sl) and -196 kcal mol-' for Sl in pFS8.56 (Cl-C30 at the base of Sl). More central changes, which are expected to be highly destabilizing, greatly reduced frameshifting. This was particularly apparent when G. C base-pairs were changed (pFS8.11: 8.15, 8.31, %33 and 8.52 in Sl; pFS8.50: 857, 8.61 and &63 in 52) where the frameshift efficiency was 2% or less. Changes in A. U base-pairs were also inhibitory, although in the examples studied, less dramatic in Sl (pFS8.49, 10%) than in S2 (pFS8.53, lo/). Calculations of the predicted stability of Sl in the more central Sl mutants using the Turner rules support the hypothesis that in these mutants, Sl is considerably less stable. The changes in free energy calculated were -8 kcal mol-' (8.11 and 8*15), -5.8 kcal mol-' (8.31 and 8.33) and -8.4 kcal mol-' (8.49).

A point mutation analysis of the mismatched G7-A24 pair in Sl (Fig. 4) tity of nucleotides at this position of the pseudoknot is not critical, In a construct in which the G7-A24 mismatched pair was changed to unpaired G7-G24 (pFS8.21), frameshifting was unaffected. In other constructs where base-pairing was promoted, frameshifting was at slightly greater than the wild-type level (35%) (pFS8.30: U7. A24; pFS8.38: G7. C34). Thus, forming a canonical Watson-Crick base-pair actually improves the frameshift process and this seems once again to be related to an increase in the overall stability of Sl. Calculations of the predicted stability of Sl in these mutants using the Turner rules support this increase in stability (pFS8.30: -164 kcal mall'; pFS8.38: -18.3 kcal mall').

The final feature studied in the stem analysis concerned the nucleotides at the junction where Sl and S2 are thought to stack coaxially. Such an arrangement of stems is strongly suggested from models of the pseudoknots in plant viral tRNA-like structures, and is one of the central features of the pseudoknot building principle (Pleij et al., 1985) . The results of a mutational analysis of the "stacking region" of the IBV pseudoknot are shown in Figure  5 . We expected that the introduction of a mismatched nucleotide pair at the top of Sl or bottom of S2 would destabilize the pseudoknot and inhibit frameshifting, since the stacking of 52 upon Sl in such mutants would be energetically less favourable. Thus, when we changed G20 to C20 (pFS8.10) such that the predicted Ull .G20 basepair at the top of Sl would be replaced by unpaired Ull-C20, we were surprised to find that frameshifting was still efficient (15%). This result could be explained by suggesting that in this mutant, Ul 1 may be displaced into loop 1 and replaced by G62 from loop 2 such that a new G62. C20 base-pair could form at the junction between Sl and 52 ( Fig. 5(a) ). In order to test this possibility, G62 was changed to C62 in a pFS8.10 background (to create pFS8.34) such that we could be confident that only mismatched base-pairs were present at' the top of Sl (either Ull-C20, or C62-C20 if Ull is displaced). Once again, this mutant displayed efficient frameshifting (15%). A similar etliciency was seen in pFS8.39, in which Ull was replaced by Gil such that an unmatched Gll-G20 pair was present at' the top of Sl (Fig. 5(b) ). Thus, a standard base-pair at t'he top of Sl is not absolutely required. It appears, however, that base-pairing at this location can contribute to frameshifting, albeit to a limited extent. In pFS8.40, we replaced Ull by Cl1 such that a standard Cl 1 *G20 pair would be present; in this construct the frameshift efficiency was, if anything, slightly better (30 to 35%) than that seen with the wild-type structure. A second piece of evidence comes from the analysis of a double mutant, pFS8.14. In pFS8.13, a Ull to All transversion at the top of Sl had little effect upon frameshifting, but in combination with an adjacent GlO to Cl0 change (pFS8.14), frameshifting was abolished. As the GlO to Cl0 change in isolation produces a mutant (pFS8.15) in which a frameshift product can still be detected, it seems that the Ul 1. G20 base-pair does provide some stabilization ( Fig. 5(b) ).

The observation of a wild-type frameshift in pFS8.13 (All and G20 at the top of Sl), however, indicates that the level of frameshifting seen in mutants created at, the top of Sl may, perhaps, not be determined simply by the presence or absence of a base-pair at this position. This possibility is supported by our analysis of the contribution of the C63. G19 frameshift process (Fig. 5(c) ). In pFS8.46, a point correlation of frameshift efficiency with base-pair mutation that created unpaired G63-G19 at this formation at the stacking region is not fully underposition reduced the frameshift efficiency by half to stood, but may be related to the possibility that the about, 15'&, a level similar to that seen with the structure of the RNA in this region of the pseudoequivalent change at the top of Sl (Gil-G20, knot is unusual. The n.m.r. analysis of a short pFS8.39). In a mutant in which the corresponding synthetic RNA pseudoknot performed by Puglisi et C63-Cl9 mismatch was created (pF8.54), frameal. (1990) revealed that although the two pseushifting was unaffected (30%), yet surprisingly, in a doknot stems did, indeed, stack, the A-form pseudo-wild-type double mutant (G63.Cl9, geometry of the RNA helix was distorted at the pFS8.59), frameshifting was once again reduced (to junction of the loops and the stacked stems. Thus, 15%). This was an unexpected observation, since in the biological effect of point mutations which the analysis of the Ull and G20 pair at the top of influence the particular bases at this region (parti-Sl, wild-t,ype frameshifting was seen with all the cularly, in the case of the IBV pseudoknot, Ull and constructs in which the bases at this position were C63) probably cannot be interpreted simply on the paired and raised the possibility that there may be a basis of the ability to form base-pairs. Clearly, more specific requirement for the C63 nucleotide at the detailed information on the three-dimensional strucbottom of 52. However, in an additional mutant ture of the stacking region of the IBV pseudoknot is (pFS8.67) in which C63 was changed to A63 to needed before the effects of changes in this region create unpaired A63-G19, frameshifting was at the can be fully interpreted. Nevertheless, the finding wild-type level. The apparent inconsistencies in the that preventing base-pair format,ion in the stacking region had only a limited effect on the frameshift process raised the possibility that direct stacking of the two stems was not an essential requirement of the process, To investigate this, we sought to separate Sl and 52 by inserting three nucleotides between G20 of Sl and G19 of 52 (pFS8.60: AAA insertion) such that stacking could only occur if this insertion was looped out of the helix (Fig. 5(d) ). In order to rule out the possibility that the two nucleotides of loop 1 were insufficient to span the increased distance between the two helices, the insertion was also introduced into a variant construct (pFS8.23) in which a three nucleotide insertion (AAA insertion) had previously been made in Ll (and shown to be functional: see Fig. 6 , pseudoknot loop analysis) to create pFS8.66. As can be seen in Figure 5 AAA insertion between Sl and 52 reduced the efficiency of frameshifting in both pFS8.66 (wild-type background, 1%) and pFS8.66 (pFS8.23 background, 5%) supporting the view that "intact" individual stems are not sufficient for high efficiency frameshifting, and that Sl and S2 need be in close proximity.

On the basis of the known co-ordinates for synthetic RNA double helices (Arnott et al., 1972) and from model building studies of pseudoknots (summarized in Pleij et al., 1985; Dumas et al., 1987) it can be estimated that loop 1 and loop 2 need to span approximately 11 A and 40 A, respectively (1 A = 61 nm), in order to bridge the deep and shallow grooves of the RNA helix, respectively, and connect Sl and S2. Assuming that a single nucleotide can span at most a distance of about 8 A (Saenger, 1984) , loop 1 and loop 2 would have to have a minimum length of two and six nucleotides, respectively, in order to connect the helices without distorting the idealized pseudoknot structure. If frameshifting requires only the formation of the correct tertiary structure and does not depend on the presence of particular nucleotides within the loops, altering the sequence and length of the loops within the constraints of the required length ought not to affect frameshifting. This was indeed the case (Fig. 6) . Either of the nucleotides proposed to comprise loop 1 could be changed without affecting frameshifting (pFS8.7: G12 to C12; pFS8.17: Al3 to U13) and an insertion of an extra three nucleotides in the loop was tolerated (pFS8.23: AAA insertion). As the correct reading frames have to be maintained in the pFS8 construct in order to monitor the frameshift, only insertions and deletions in multiples of three nucleotides were possible. Thus, we were not able to delete proposed Ll nucleotides. Deletion analysis was, however, possible with loop 2, and the results obtained were in good agreement with the predicted minimal length of this loop (6 nucleotides). In pFS8.2, all but eight nucleotides of L2 were deleted without effect, but in pFS8.18, in which a further three nucleotides were deleted (leaving just 5 nucleotides to span the helix), frameshifting was greatly inhibited.

We interpret this observation in terms either of a distortion or an abolition of the pseudoknot structure in this mutant. As was the case of loop 1, the identity of loop 2 nucleotides was unimportant in the frameshift process. When seven of the eight nucleotides of L2 in pFS8.2 were changed to their complements (in construct pFS8.28), frameshifting occurred with wild-type efficiency. (The 8th nucleotide, G62, the last nucleotide in L2 was changed to C62 without dramatic effect in mutant pFS8.34, described earlier.) In principle, there are no upper limits for the length of L2 (or Ll) (Pleij et al., 1985) and we have been able to insert six nucleotides corresponding to a unique XhoI restriction site into L2 without effect (pFS8.1). However, when a 467 bp DNA fragment derived from the influenza PB2 gene (see Materials and Methods) was inserted in-frame into this XhoI site, frameshifting was greatly reduced (pFS8.47; see Fig. 6 ). The reason for this is not clear, but may be a consequence of competition for the PK2 sequence between the authentic PK4 sequence and sequences within the inserted PB2 segment (see Discussion, below).

The mechanism by which the pseudoknot promotes ribosomal slippage is not known, but one possibility is that the translating ribosome slows or stalls as it reaches this structure and that this can promote slippage at the adjacent slippery site. Consistent with this idea is the observation that the pseudoknot must be correctly positioned with respect to the slip site; insertion or deletion of three nucleotides in the six nucleotide intervening sequence severely inhibits frameshifting (Brierley et al., 1989) . The way in which the structure might slow the ribosome is uncertain, but the effect may be simply dependent on the stability of the pseudoknot and the overall energy required to unwind it. If this were the case, it might be expected that a simple stem-loop, comprising an equivalent set of base-pairs, would promote frameshifting as well, if not better, than the pseudoknot if encountered in the same genetic context. To test this, the pseudoknot in pFS8 was deleted and a simple stem-loop structure containing a base-paired stem, of the same length and base-pair composition as the stacked stems of the pseudoknot was introduced at a position six nucleotides downstream from the slippery sequence (pFS8.26). This construct did not display efficient frameshifting, however (0.5 %, Fig. 7) , suggesting that the pseudoknot has some particular structural feature which is required for the effect. It might be argued that, in this case, the failure to detect high levels of a frameshifted product was due not to a reduction in slippage, but to a general inability of ribosomes to translate through the hairpin structure. However, when the upstream and downstream PBl ORFs were placed in the same reading frame by introduction of a single nucleotide (C) just upstream from the slippery sequence in a control construct (pFS8.271, only the full-length translation product was observed. That the stemloop structure actually forms in pFS8.26 is supported by the observation that in an additional construct (pFS8.65), in which the predicted stemloop was destabilized by a complementary change in the stem, we were not able to detect a frameshift product. Thus, the simple stem-loop cannot functionally replace the pseudoknot..

Much of the available information regarding the structure of RNA pseudoknots has been derived from direct chemical and enzymatic cleavage analysis of pseudoknot-containing RNA fragments, in combination with phylogenetic sequence comparisons (Pleij, 1990) and from model building studies (Dumas et al., 1987) . Recently, the threedimensional conformation of a short (24 nucleotides), pseudoknotted, synthetic oligoribonucleotide has been determined by n.m.r. and the structure obtained was in good agreement with that which had been predicted from previous building principles; the pseudoknot contained two helical stem regions that were stacked to form a continuous helix. A powerful and complementary approach that can be employed in the determination of RNA structure is to generate variants within the predicted structure by site-directed mutagenesis, and then to test the formation of the structure by a functional assay. However, such assays for pseudoknot formation have only recently been described and are few in number. Tang & Draper (1989) were able to investigate the base-pairing interactions formed in a pseudoknot located at the 5' end of the E. eoli a-mRNA by taking advantage of the ability of the a-mRNA-encoded S4 protein to bind to the pseudoknot during the process of autoregulation of a-mRNA expression. By mutating specific nucleotides in the predicted stems of the pseudoknot, and measuring the affinity of binding of S4 to these variants, the base-pairs predicted by classical structure mapping methods (Deckman et al., 19871 were confirmed and additional interactions discovered. Similarly, Mans et al. (1990) have studied the tRNA-like structure of turnip yellow mosaic virus RNA by creating mutations within the structure, and testing for the ability of E. coli RNase P to cleave the pseudoknot. Here, we have used this approach to characterize the struc-tural organization of a naturally occurring pseudoknot in the genomic RNA of a coronavirus. By creating a comprehensive series of changes within the pseudoknot region by mutagenesis and testing for the ability of these mutants to promote ribosomal frameshifting in a cell-free translation system, we have been able to investigate the structure of the IBV pseudoknot, and to compare it with that proposed from the currently available building rules. Moreover. the analysis has important implicat'ions for the mechanism of ribosomal frameshifting.

(a) The structure of the IBV pseudoknot

The results of our mutagenic analysis clearly demonstrate that the RNA downstream from the slippery sequence folds into a pseudoknot of the hairpin-loop type. The model proposed in Figure  l (b) is largely supported by this analysis; point mutations and complementary and compensatory base changes within the stem regions confirm, in the main, the base-pairs predicted. We have interpreted the mutagenesis results in terms of the expected stability of the two pseudoknot stems in mutant constructs on the basis that neither of the constituent, hairpins alone is capable of stimulating high efficiency frameshifting. The point mutational analysis has shown that, with certain exceptions, a single mutation in either stem is sufficient to destabilize the structure and reduce frameshifting. Although Sl is some four base-pairs longer than S2, its st,ability is undoubtedly reduced by the presence of the central G-A mismatched pair, since frameshifting is improved if this pair is replaced by a Watson-Crick pair. A similar situation is observed in a related coronavirus, mouse hepatitis virus (MHV). The frameshift region of MHV contains a pseudoknot and there is considerable sequence covariance within the predicted base-paired regions of the MHV and IBV pseudoknots (Bredenbeek et al., 1990) . In MHV, a G-A mismatched pair is present in S2, and in this case, the helix is four basepairs longer than 82 of IBV. It appears, therefore, that each stem must possess a certain minimum stability in order that the pseudoknot be functional. Tn the complementary and compensatory basechange analysis (Fig. 3) it was observed that in a number oi double-mutant, pseudo-wild-type constructs, t)he frameshift' efficiency was not fully restored to the wild-type level. The reason for this is uncertain.

since in at least one of the cases (pFS8.12. Fig. 3 ) the stability of Sl in the pseudowild-type construct (-12.9kcalmol~') , as predicted by the Turner et al. (1988) rules, is similar to that predicted for the wild-type Sl hairpin loop ( -12.4 kcal mol ') . Clearly, more information on pseudoknot, thermodynamics and, indeed, the threedimensional structure of pseudoknots is needed before these subtle effects on frameshift efficiency can be fully understood.

When the G.(: base-pairs at the start of the Sl helix (Gl . ('30) were mutated to unpaired nucleot'ides (Fig. 3) . we noted a small reduction in frame-shift efficiency and have argued that as the changes are at the end of the helix, they may be expected to have a less dramatic effect on the overall stability of the structure. We cannot rule out the possibility, however, that the observed reduction in frameshifting was not a result of a destabilization of Sl below a critical threshold, but rather a result, of increasing the effective slippery sequencepseudoknot spacing distance by one nucleotide.

Another feature of the pseudoknot where simple interpretations based on expected stability are not possible is the stacking region. where base-pair formation does not appear to be required for efficient frameshifting.

In a recent analysis of the stability requirements for RNA pseudoknot formation, Wyatt et al. (1990) were able to demonstrate that a synthetic pseudoknotted oligonucleotide (26 nucleotides) was only marginally more stable t'han either of t'he constituent hairpins (1.5 to 2 kcal mol-' at 37°C). Moreover. the increase in enthalpy observed upon forming the pseudoknot' was less than the predicted gain; this was probably a consequence of both distortions in stacking between the two stems and to positive enthalpic contributions of the loop regions. A number of the mutations introduced into the stacking region of the IBV pseudoknot gave a frameshift profile that was difficult to interpret. in simple terms of base-pair formation; this may well be related to a structural distortion in this region. The availability of a large collection of pseudoknot mutants creat,ed in this study should allow an investigation into the energetics of pseudoknot formation by biochemical and biophysical methods.

Our analysis of the pseudoknot loops has revealed that, as was the case with the pseudoknot stems, no specific nucleotides are involved in t,he fra,meshift process. Each loop ca,n tolerate a small insertion without influencing frameshift'ing, and theoretical considerations for the minimal predicted lengths of the loops (Pleij et nE., 1985) appeared to be substantiated in the case of loop 2. Although there is no theoretical upper limit for loop length (Pleij et al.. 1985) we found that, at least in the one example studied (pFS8.47), loop 2 could not t,olerate a large (467 bp) insert. Pseudoknots wit,h large loop sizes have been proposed in the structure of RNase P (James et al., 1988) and in 16 S rRNA. where most of the rRNA secondary structure is cbontainrd within loop 2 of a pseudoknot (Moazed Br Noller. 1987 ). In addition. a long-range pseudoknot has been proposed in a structural model for mitochondrial group I introns (Davies et 01.. 1982) . In the case of pFS8.47, we have speculated that) pseudoknot formation may be prevented by competition between the authentic PK4 sequence and alternative stretches within the inserted information.

A second possibility concerns the rate of translation of L2 in this mutant. Having melted Sl. the elongating ribosome would take considerably longer to translate L2 in pFS8.47 than in the wild-t'ype sit,uation and this may impede refolding of the pseudoknot to such an extent that t>he next ribosome translating the mRNA may not encounter a pseudoknot and would therefore not frameshift. It maybe significant that in the pseudoknots predicted to form at ribosomal frameshift sites in other systems, the longest loop length is only 69 nucleotides, and in the vast majority of cases, the loops are considerably shorter (ten Dam et al., 1990) .

(b) Implications for ribosomal frameshifting

The mutational analysis so far indicates that the formation of the pseudoknot is critical for high efficiency frameshifting and has helped to provide a detailed picture of the structure involved. A summary of the changes which have been made is shown in Figure 8 . The "non-essential" nucleotides of loop 2 (defined by mutant pFS8.2) are not included in the Figure, hence the 44 nucleotides depicted represent our estimate of the minimal number required for the formation of the IBV pseudoknot. As each of the nucleotides has been changed to an alternative nucleotide in one or more of the various mutant constructs, the particular nucleotide sequence of the pseudoknot per se is not important in the frameshift process.

Nevertheless, efficient frameshifting depends upon the formation of the correct structure 3' *A +U.G+ +A +G* C+ +A*U+ *A +C.G+ Loop2 +c *$A-Uv Stem1 +G *U-A+ *U +G * C+ *G-C+ *u *G-C+ 5' Figure 8 . Summary of changes in the pseudoknot region. The "non-essential" nucleotides of loop 2 (see the text) are not included in the Figure; the 44 nucleotides depicted represent our estimate of the minimal number required for the formation of the IBV pseudoknot. *, indicates change not inhibitory; +, change giving intermediate phenotype; -+, inhibitory change but can be compensated by complementary change elsewhere. downstream from the slippery site; the pseudoknot could not be functionally replaced by a stem-loop structure of a similar or greater predicted stability. It remains unclear how the pseudoknot actually causes ribosomal slippage. The most likely explanation is ribosomal pausing (Jacks et al., 1988) , and we are investigating possible mechanisms within the framework of two models. In the first model, pausing occurs as a result of a direction interaction of a ribosomal protein(s) or additional component(s) of the translation apparatus with the pseudoknot. The control of mRNA translation through specific recognition by proteins of pseudoknotted RNAs has been documented in a number of prokaryotic systems (McPheeters et al., 1988; Tang 6 Draper, 1989; Philippe et al., 1990) , although in these cases, the proteins concerned bind to, or induce the formation of, pseudoknots in the 5' non-coding regions of mRNAs and are involved in autoregulation. In this investigation we have been unable to demonstrate any requirement for specific pseudoknot nucleotide sequences in the frameshift process that could not be interpreted in terms of the pseudoknot structure. The inability of the simple stem-loop structure (pFS8.26) to direct efficient frameshifting, however, supports the view that some unique feature of the pseudoknot plays a role in the process and the possibility that sequence-independent recognition of, for example, a pseudoknot loop(s) by a protein factor cannot be discounted. Autogenous regulation of gene 32 protein expression in bacteriophage T4 involves an interaction between the protein and the loop(s) of a pseudoknot in the 5' non-coding region of the gene 32 mRNA (McPheeters et al., 1988) . In the case of the IBV pseudoknot, eight nucleotides would have to suffice for such an interaction with loop 2, since 24 nucleotides of this loop could be deleted without effect (pFS8.2).

An alternative and attractive possibility for a pausing mechanism is that RNA pseudoknots may present an unusually resistant structure to a ribosome-associated RNA helicase activity functioning to promote local unwinding of mRNA during elongation. The lack of frameshifting in the simple stem-loop construct is significant in that this structure can be predicted to be much more stable than the pseudoknot, and so a simple energetic barrier to translation is not sufficient for the effect. This hypothesis is based upon the observation of Wyatt et al. (1990) that small pseudoknots are only marginally more stable energetically than either of their constituent hairpins, and likely to be less stable than a hairpin-loop with a comparable number of basepairs.

It should be possible to test these hypotheses through in vitro analysis of the interaction of short, pseudoknotted RNAs and components of the translation apparatus. This work was supported by an AFRC link grant, LRG 171, awarded to S.C.I. We are grateful to Cornelis Pleij, Edwin ten Dam and Jan van Duin for stimulating discussion.

Optimised parameters for RNA double-helices

A rapid alkaline extraction procedure for screening recombinant plasmid DNA

The primary structure and expression of the second open reading frame of the polymerase gene of the coronavirus MHV-A59; mechanism

An efficient ribosomal frame-shifting signal in the polymerase-encoding region of the coronavirus TBV

Characterisation of an efficient coronavirus ribosomal frameshifting signal: requirement for an RNA pseudoknot

Tomley, F. M. & Binns, M. M

ompletion of the sequence of the genome of the coronavirus avian infectious bronchitis virus

Translational frameshifting: where will it stop?

Making ends meet: a model for RNA splicing in fungal mitochondria

S4+mRNA translation regulation complex 1. Thermodynamicss of formation

Functional analysis of bacteriophage fl intergenic region

3-D graphics modelling of the tRNA-like 3' end of turnip vellow mosaic virus RNA: structural and functional implications

An introduction to polyacrylamide gel electrophoresis

Expression of the Rous sarcoma virus pal gene by ribosomal frameshifting

Two efficient ribosomal frameshifting events are required for synthesis of mouse mammary tumor virus gag-related polyproteins

Signals for ribosomal frameshifting in the Rous sarcoma virus gag-pol region

The secondary structure of ribonuclease P RNA, the catalytic element of a ribonucleoprotein enzyme

tRNA-like structures of plant viral RNAs: conformational requirements for adenylation and aminoacetylation

Functional messenger RNAs are produced by SP6 in vitro transcription of cloned cDNAs

Rapid and efficient site-specific mutagenesis without phenotypic selection

Molecular Cloning: A Laboratory LWanual

Interaction of RNase P from Escherichia eoli with pseudoknotted structures in viral RNAs

Autogenous regulatory site on the bacteriophage T4 gene 32 messenger RNA

Interaction of antibiotics with functional sites in 16 S ribosomal RXA

Complete nucleotide sequence of a milktransmitted mouse mammary tumor virus: two frameshift suppression events required for translation of gag and pal

Target site of Escherichia coli ribosomal protein 815 on its messenger RNA

Pseudoknots: a new motif in the RNA game

RNA pseudoknots: structure, detection and prediction

A new principle of RNA folding based on pseudoknotting

Conformation of an RNA pseudoknot

The tRNA-like structure at the 3' terminus of turnip yellow mosaic virus RNA. Differences and similarities with canonical tRNA

Three dimensional models of the tRNA-like 3' termini of some plant viral RNAs

The three dimensional folding of the tRN,4-like structure of tobacco mosaic virus RNA. A new building principle applied twice

An improved filamentous helper phage for generating singlestranded plasmid DNA

Principles of ,V'ucleic Acids Structures

DNA sequencing with chain-terminating inhibitors. f'roc

rnusual mRNA pseudoknot structure is recognised by a protein translational repressor

RNA pseudoknots: translational frameshifting and readthrough on viral RNAs

167-192. pseudoknots; stability and loop size requirements

Improved Ml3 phage cloning vectors and host strains: nucleotide sequences of the Ml3 mp18 and pUC19 vectors

Cloning and 204 nucleotides long 3' non-coding region of tobacco expression of influenza virus genes. In The Origin of mosaic virus RNA

RNA pp