key: cord-0010669-yefxrj30
authors: Yelverton, Elizabeth; Lindsley, Dale; Yamauchi, Phil; Gallant, Jonathan A.
title: The function of a ribosomal frameshifting signal from human immunodeficiency virus‐1 in Escherichia coli
date: 2006-10-27
journal: Mol Microbiol
DOI: 10.1111/j.1365-2958.1994.tb00310.x
sha: 325e50e217cb21ebe9e8b3899f1e3d5f1d8f7650
doc_id: 10669
cord_uid: yefxrj30

A 15‐17 nucleotide sequence from the gag‐pol ribosome frameshift site of HIV‐1 directs analogous ribosomal frameshifting in Escherichia coli. Limitation for leucine, which is encoded precisely at the frameshift site, dramatically increased the frequency of leftward frameshifting. Limitation for phenylaianine or arginine, which are encoded just before and just after the frameshift, did not significantly affect frameshifting. Protein sequence analysis demonstrated the occurrence of two closeiy related frameshift mechanisms. In the first, ribosomes appear to bind leucyl‐tRNA at the frameshift site and then slip leftward. This is the 'simultaneous slippage’mechanism. In the second, ribosomes appear to slip before binding amlnoacyl‐tRNA, and then bind phenylaianyl‐tRNA, which is encoded in the left‐shifted reading frame. This mechanism is identicai to the‘overlapping reading’we have demonstrated at other bacterial frameshift sites. The HIV‐1 sequence is prone to frame‐shifting by both mechanisms in E. coli.

Ribosomes normaiiy maintain a constant reading frame from AUG to the finish, but they are capabie of slipping into an alternative reading frame at an average frequency of the order of 10 " (Atkins etai, 1972; J. A. Gallant etai, unpubiished) . In certain special cases, much higher frequencies of ribosome frameshifting occur. These cases include production of polypeptide release factor 2 of Escherichia coli, which depends upon a rightward frameshift within the coding sequence (Craigen et at.. 1985; Craigen and Caskey, 1987; Weiss et ai, 1987; Curran and Yarus, 1988) ; translation of the reverse transcriptase of the yeast Ty element, which also depends upon a rightward frameshift (Clare ef ai, 1988) ; and translation of the RNA of severai retroviruses, which express gag-pol and gag-pro-pol polyproteins by means of leftward frameshifts (reviewed by Hatfield and Oroszlan, 1990; Cattaneo, 1989) .

Ribosomal frameshifting in both rightward and leftward directions has also been shown to occur at certain 'hungry' codons whose cognate aminoacyi-tRNAs are in short supply (Gallant and Foley, 1980; Weiss and Gailant, 1983; 1986; Gallant et ai, 1985; Kurland and Gallant, 1986) . Not all hungry codons are equally prone to shift: in a survey of 21 frameshift mutations of the rllB gene of phage T4, Weiss and Gallant (1986) found that oniy a minority were phenotypicaily suppressible when challenged by limitation for any of several aminoacyl-tRNAs.

The context njies governing ribosome frameshifting at hungry sites are under investigation, and have been defined in a few cases (Weiss et al., 1988; Gallant and Lindsiey, 1992; Peter et ai. 1992; Koior ef a/., 1993; Lindsiey and Gallant, 1993) . So far these sequences do not resembie any of the naturally occurring shifty sites summarized in the first paragraph above, in order to find out whether these two categories of ribosome frameshifting are mechanisticaliy reiated, we have tested the susceptibility of a well-studied retroviral frameshift site to manipulation by aminoacyl-tRNA limitation in E. coli We have directed our analysis to the shifty site at the gag-pol junction of HIV-1 both because of its clinical interest, and because certain features render it convenient for analysis.

In some viral systems, baroque secondary structures in the mRNA downstream of the frameshift site are required to augment frameshifting levels (Jacks et ai, 1988b; Brierley et ai, 1989) . In the case of HIV-1, however, although a stem-loop structure might exist downstream of the frameshift site (Jacks et ai, 1988a) , direct modification or elimination of the stem-loop sequence has little effect on the rate of frameshifting (Madhani et ai, 1988; Weiss etai. 1989) . Moreover, Wilson etal. (1988) demonstrated that a short (21 nucleotide) sequence of HIV-1 without the stem-loop was sufficient to direct a high level of frameshifting in heterologous in vitro systems.

The site of ribosomal frameshifting at the siippery sequence U-UUU-UUA has been directly established by amino acid sequencing of frameshifted proteins (Jacks et ai., 1988b) , and the participation of certain aminoacyl-tRNAs has been clearly implicated by mutagenesis of the monotonous tract of uridines (Jacks et ai, 1988b; Wiison et ai, 1988) . Our purpose was to discover whether the ribosomal frameshifting directed by a very short sequence in HiV-1 could be reproduced by E coli ribosomes in vivo, and, if so, whether we could alter the rate of frameshifting by regimens that change the relative abundance of key aminoacyl tRNAs encoded at or near the frameshift site. Weiss etal. (1989) have also reported that a 52 nucleotide fragment from HIV-1 is sufficient to direct ribosomal frameshifting in an E. coli system. In this report we present evidence that a much shorter 15-17 nucleotide sequence derived from HIV-1 is sufficient to direct the same ribosomal frameshift event in E. coli as in eukaryotes. We also show that in E. coli the rate of ribosomal frameshifting on that sequence can be increased by limitation for leucine, the amino acid encoded at the frameshift site. Protein sequence analysis of the product indicates the occurrence of two siightiy different mechanisms of shifting.

The strategy behind the construction of our assay system for ribosomai frameshifting may be understood with reference to Fig. 1 . When eukaryotic ribosomes decode the HIV mRNA sequence . . . UUUUUUAGGG . . ., shown as nucleotides 1-10 in Fig. 1A , the adenine at position 7 appears to be read twice: first, as the third position of a leucine codon (UUA) and then again as the first position in the overlapping arginine codon (AGG Ratner et al. (1985) . In a heterologous mammalian in w/ro translation system, most of the frameshift product has the amino acid sequence . . . Asn-Phe-Leu-Arg (Jacks et al., 1988b) , where Leu is encoded by positions 5-7 and Arg is encoded by positions 7-9. Some mutations that result in increased or decreased expression of frameshift products in a heterologous test system are shown above and below the nucleotide sequence, respectively (Wilson ef a/., 1988) . 'N' signifies a mutation to any non-U base. Double underlines at positions -10 and +16 mark the boundaries of a fragment that directs the synthesis of a frameshift protein product in a heterologous yeast system (Wilson et at.. 1988) . The singly underlined G at position -6 is the 5' boundary of a fragment that directs the synthesis of a frameshift protein product in a mammalian iri vitro system (Jacks e( al., 1988b) . B. and C. A portion of &ie mRNA sequences expressed from tecZframeshift alleles HIV13, HIV13-A3, HIV201, and HIV201-U7 are depicted. Numbers above the nucleotide sequence correspond to analogous positions of the HIV-1 gag-po/junction. Doubly underlined nucieotides mark the twundaries of sequence thai is consen/ed with respect to HIV-1. Synthesis of p-gaiactosidase from the alleles requires a leftward frameshift. Amino acids are shown for the mature protein, afler in vivo cleavage of the initiating W-terminal fomiyl-met residue. Constructs pHIV13 and pHIV201. and their variants pHIV13-A3 and pHIV201-U7, are described in Fig. 1 . (The sequence of the critical heptanucleotide at the frameshift site is shown in parenthesis after each construct's designation.) All constructs were transformed into a derivative of CP79 {relA2 thr' leu his ~ arg~ thi~) carrying a complete deletion of lacZ. Methods of cultivation, and enzyme and protein assay were as described previously Peter etal., 1992) . Cells were grown into exponential phase in M63-glucose medium supplemenled with all required amino acids plus He and Val. The lac promoter was induced (2mM IPTG and 2.5mM cAMP) for about one doubling. Data are reported ± standard error of the mean, with the number of replicate induced cultures in parentheses. These values include all the unstarved control cultures from the various starvation experiments.

to take place in about 10% of ribosomal transits (Jacks etai. 1988b) . in HiV-1, the outcome of the leftward ribosomal frameshift is the successful production of the gag-pol fusion protein, in the assay system we have devised, the outcome of an analogous ieftward frameshift by E coli ribosomes wiii be the successful production of the enzyme p-galactosidase from genetically frameshifted alleles of the /acZ gene. We have previousiy used an assay system of similar design to demonstrate that lysyl-tRNA starvation can amplify ribosomai frameshifting in either direction at iysine codons, given certain context ruies (Gailant and Peter etai. 1992; Lindsiey and Gailant, 1993 ).

Aileles to be tested were constructed by the ligatlon of paired complementary oligonucleotides into the H/ndlll-BamH\ site of pBWHOO, as described in Galiant and Lindsiey (1992) . Figure 1 shows the sequence of the translated strand from the region of our constructs that reproduces the gag-pol frameshift signal from HIV-1. The /acZ frameshift alieles carried on plasmids pi^iV13 and pHIV201 are constructed so that a shift to the left by one base, as in the expression of the gag-po/fusion of HIV, is required to generate active enzyme. The two constructs both carry a short sequence identical to the region around the frameshift site in the gag-po/overlap of HiV-1 for 15 nucleotides in pHIV13 and 17 nucleotides in pHIV201 (see Fig. 1 ); they differ slightly from one another severai bases downstream of the frameshift site. Host ceils carrying either of these piasmids produce active enzyme at about 1% of the efficiency of cells carrying a control lacZ* plasmid (Table 1) . This basal value is close to the value (1.8%) observed by Weiss et al. (1989) for frameshifting on a much longer HIV-derived sequence in a similar p-galactosidase reporter. It is also much higher than the frequency of leftward frameshifting (0.03-0.2%) we observed previously at sequences unrelated to HIV (Gailant and Lindsiey, 1992) . The presence of the HIV sequence in our reporter thus leads to an unusually high frequency of leftward frameshifting.

Modification of the critical heptanucieotide sequence from U UUU UUA to U UAU UUA in plasmid pHIV13-A3 decreased frameshifting about fivefold, while modification of the heptanucleotide to U UUU UUU in pHIV201-U7 increased frameshifting by two-to threefold (Tabie 1). These genetic resuits are analagous to earlier findings in other reporter systems (Jaci<s et ai. 1988b; Wilson et ai. 1988; Weiss et ai, 1989) and suggest that the heptanucleotide is the predominant site of leftward frameshifting in our reporter system as well.

Chemical evidence that this is indeed the case was obtained from protein sequence analysis. The protein encoded by pHIV201 was purified and subjected to automated Edman sequence analysis, as previously described Peter et al., 1992) . The results ( Fig. 2 ) are readily interpreted with reference to the nucleotide sequence and coding potential shown in Fig. 1 . For the first 10 cycles, the amino acid sequence corresponds to normal transiation in the (0) reading frame: T-M-i-T-P-S-S-N-F-L. After this position, the signal (R-E-G-l-P) clearly corresponds to the leftward reading frame. Thus, a leftward shift occurred after incorporation of the leucine at position 10, exactly the frameshift cfiaracteristic of HIV translation in eukaryotic systems (Jacks et ai, 1988b) . Similar resuits were obtained with plasmid pHiV13 (data not shown). Thus, our short cloned sequence reproduces the specificity of the HIV frameshift signal when translated by E. coli ribosomes.

We asked whether the HIV sequence's rather high level of shiftiness could be modulated by stalling the ribosome at hungry codons within the frameshift region. Phe and ieu are the last two amino acids encoded in the initial reading frame before the shift, and Arg is the first encoded in the new reading frame after the shift. We imposed iimitation for phe-tHNA by inhibiting the synthetase with the analogue phenylaianine-hydroxamate (PHX). Graded limitations for arginine or ieucine were imposed by growing auxotrophic host cells on a range of concentrations of slowiy utiiized methyl esters of these amino acids (Buston and Bishop, 1955) . Tables 2 and 3 present the results of ali these limitation regimens.

Considering first Arg and Phe (Table 2) , we observe that both limitations substantiaily reduced the differential rate of enzyme synthesis expressed from both the frameshift construct and the lacZ* control. The reduction was about the same in both genotypes, indicating that neither arginine limitation nor phe-tRNA limitation had a specific effect on the frequency of frameshifting within the HIV sequence. In order to ensure leucine sufficiency, the host was a ieucine prototroph isogenic with our normal host, and it was grown nevertheless in the presence of 100 jig ml" ^ leucine. The /ac promoter was induced for one doubling and two litres of cells were harvested. Beta-galactosidase was isolated as described previously Peter ef ai. 1992 ) and subjected to automated Edman sequencing. The raw data for picomole amounts of PTH amino acids in each cycle were corrected for background and lag using Applied Biosystems software, as described . The figure presents the sum of three independent sequencing runs, one performed at Illinois and two at Riverside. In every cyle, the most abundant PTHamino acid released predominates over others by at least a factor of two, affording a clear signal of the primary sequence. Secondary signals reflect a variety of artefacts. One is minor peaks which anticipate the next cycle (for example, F in cycle 8, R in cycle 10, E in cycle 11, and G in cycle 12}, presumably owing to a minority of the protein that is one amino acid shorter on the /V-terminus than the bulk of the material. Exponential cells carrying plasmid pHIV201 or the control lacZ' plasmid pBWIlOO were subjected to limilation for phenylalanyt-WNA by addition of phenylalanine-i^ydTOxamate (PHX) al the indicated concentrations; and for arginine by centrifuging and washing the cells, and resuspending them in growth medium containing the indicated concentrations of arginine-methyl-ester (AME), with a control culture containing arginine at our standard concentration of 100 Jig ml V The /ac promoter was induced as in Table 1 , and cells harvested for assay after about one doubling. Cultures are listed in order of decreasing growth rate. Growth rate varies directly with the concentration of AME, which serves as a source of required arginine, and inversely with the concentration of PHX, which inhibits phenylaianine activation.

in the case of leucine limitation, a different outcome was suggested in preliminary screening by a crude, qualitative Petri plate assay (Kuriand and Gallant, 1986) . Quantitative data on log-phase cells are shown in Table 3 . In the lacZ*ĉ ontroi, leucine limitation progressiveiy restricted enzyme synthesis, reducing it very severeiy at iow growth rates, in the HIV construct, on the other hand, there was no such restriction, but ratlner a smali stimuiation of enzyme synthesis.

if this effect is caused by frameshifting at the ieucine codon within the HIV sequence, then it should be eliminated by eliminating that ieucine codon. As an additional control, therefore, we tested 201-U7, a construct that is identicai to 201 except for the conversion of the leucine UUA codon to UUU (see Fig. 1 ). In this construct, ieucine limitation did not stimulate enzyme synthesis as in 201, but ratfier reduced it, as it did in the wiid type (Table 4) . The reduction was a good deal less than in the wiid type, suggesting that some frameshifting at various positions still occurs in 201-U7, so as to offset partly the effect of reduced lacZ expression. However, the difference between tine response of 201 and that of 201-U7 demonstrates that the leucine codon eliminated in the latter must be a major determinant of frameshifting induced by leucine limitation.

Decisive identification of the frameshift site during leucine iimitation was obtained from protein sequence anaiysis, as described earlier. The resuits are shown in Table 3 . Relative growth rates and differential rates of (3-galactosidase synthesis during leucine limitation. Exponential cells carrying plasmid pHIV201 were spun, washed, and resuspended in medium supplemented with the indicated concentrations of leucine-methyl-ester (LME). These were grown in parallel with a control aliquot retumed to growth in leucine at lOOtigml''. The entries show growth rates and differential rates of enzyme synthesis relative to the control leucine culture, expressed as mean values i^ standard error (number replicate experiments). The mean control differential rates in these experiments were 2.3 ± 0.16 (6} for pHIV201 and 191 ± 44 (4) forpBWIIOO. Fig. 2 and Fig. 3 . In order to normalize to the same number of picamoles, all the values for the leucine-starved protein were multiplied by the ratio (unstarved/ starved) of the sums of the predominant peaks (N in cycle 8, F in 9, R in 11, and E in 12). The ratios of leucine to phenylaianine in cycle 10 of the two sets of data were as follows. For the unstarved case, the ratios in three runs were 2.55, 1.9, and 3.5, yielding an average of 2,65 with a standard error of 0.46. For the ieucine-starved case, the ratios in four runs were 0.52, 0.58, 0.68, and 0.51, yielding an average of 0.57 with a standard error ot 0.039. The difference is highly significant (P less than 0.005) by the la test (Crow et ai.. 1969) . The nucleotide sequence in the region of interest is shown below the cycle bar plot, with the expected amino acids read in the initial (0) reading frame shown above the nucleotide sequence, and the expected amino acids read in the leftward reading frame shown below. Solid bars = HIV-201 unstarved (+Leu); open bars = HIV--201 stan/ed (+LME) normalized to unstarved values. Fig. 3 . The protein sequence Is very similar to that of the protein from unstarved cells (Fig. 2) , with one significant exception. For the first nine cycles, the initial T-M-i-T-P-S-S-N-F sequence of the noimal (0) reading frame is clear, as It was in the protein from unstarved ceils. From cycle 11 onwards the signature of the leftward reading frame, R-E-G-i-P-G, is equaily clear. The one difference between the starved and unstarved proteins is at cycle 10, the position of the leucine codon. The predominant amino acid in the starved protein is F (phenyialanine) rather than L at this position.

Three considerations argue that carryover of F from cycle 9, the preceding position, does not contribute significantly to the peak of F in cycle 10 of the leucine-starved protein. First, the data in every cycle have been corrected for carryover by means of the Applied Biosystems lagcorrection aigorithm (Huni^apiller, 1986) . The effectiveness (3) Exponential cells carrying plasmid pHIV201-U7 were treated as in Table 3 . The mean control differential rate was 4.7 ± 0.20 (3). of this correction can be seen in the very low levels of carryover throughout the cycle bar plots of Figs 2 and 3, and of our earlier publications (Peter et ai, 1992; Gallant and Lindsiey, 1992) . Second, amino acids differ in their susceptibility to carryover from one cycle to the next (proline, for example, being notoriously bad), and it is carryover of F that we need to assess in considering these data. In fact. Figs 6 and 7 of Gallant and Lindsiey (1992) present exactly the needed controi: F was the amino acid present in cycle 7 of both sequences, and cycle 8 corresponded to a 'hungry' codon, just as in the present case. In those two sets of data, which represent the average of four independent sequencing runs in each case, carryover of F from cycle 7 to the next cycle averaged 8.3% (standard error = 2.8%), far less than the relative ievel of F found in cycle 10 of the present cycle bar plots.

Finally, the starved and unstarved proteins are virtually identical at every position except cycle 10, in which the difference in regard to phenylaianine and leucine was replicated in several independent sequencing runs on protein of each type. A summary of this comparison of the starved and unstarved proteins in and around cycle 10, and the statistics for this cycle, are presented In Fig. 4 .

Phenylaianine corresponds to the UUU triplet overlapping the hungry ieucine codon in the leftward reading frame. We conclude that most ribosomes which produced P-galactosidase slipped ieftward when stalled at the leucine codon before aminoacyl-tRNA binding, and then bound phe-tRNA cognate to the overlapping phenylaianine codon. However, cycle 10 contains a significant secondary peak of leucine as well, suggesting that a minority of ribosomes bound leucyl-tRNA and then shifted left, just as in unstarved cells.

Our wild type lacZ* controi shows a marked decrease in p-galactosidase synthesis during limitation for each of tiie amino acids tested. This response is typical of relA ~ cells, and is primarily caused by a dependence of /acZtranscription on ppGpp during amino acid limitation (Primakoff and Artz, 1979; Yangefa/., 1979; Primakoff, 1981; Foieyefa/., 1982) , compounded by errors in transiation at some hungry codons (reviewed by Gallant, 1979; Cashe! and Rudd, 1987; Parker, 1989) . Expression of the /acZ gene carrying the HIV frameshift signal should be subject to the same relA~ defects and is indeed reduced to the same extent as wild-type during limitation for arginine or for phenyialanyl-tRNA (Table 2) .

During (eucine limitation, in contrast, enzyme synthesis is increased rather than decreased in cells carrying the frameshift allele pHIV201 (Table 3 ). In 201-U7, which is identical to 201 except for the absence of the leucine codon at the frameshift site, leucine limitation provokes a decrease in enzyme synthesis (Table 4) , as it does in the iacZ* control. The fact that enzyme synthesis increases with leucine limitation in 201 means that an increase in frameshifting more than compensates for the sharp decrease in lacZ expression found in tine control experiments with the wild type and with 201-U7. The increased frameshifting attributable to this one leucine codon can be estimated as the ratio of the effect of ieucine limitation on 201 to that on 201-U7. in the former case, enzyme synthesis was two times greater in 3)igmi"' of leucinemethyl-ester than in leucine, whereas in the tter case it was three times less. By this estimate, therefore, the frequency of frameshifting at the critical leucine codon was increased about sixfold at 3|igmr^ of the leucine analogue, a limitation regimen which reduced growth about sevenfold.

The protein sequence of the frameshifted material (Fig. 3) demonstrates that the ribosomes read the sequence in a normal fashion for the first nine codons, up to the hungry codon calling for leucine at position 10. At this position, the predominant amino acid was F (phenylaianine) rather than L (leucine), implying that the ribosomes slipped leftward to decode the TTT triplet for phenyialanine, which overiaps the hungry codon from the left. This overlapping mode of reading is entirely analogous to the case of leftward frameshifting at a different hungry codon we have described earlier .

However, there is also a significant secondary peak of leucine at position 10, amounting to one third of the total signal (Fig. 3) . This is the amino acid encoded in (he initial or (0) reading frame. The presence of leucine here strongly suggests that a minority of ribosomes bound leucyl-tRNA and then slipped leftward, just as a majority of ribosomes did under conditions of (eucine sufficiency.

Thus, the HIV sequence is subject to starvationpromoted frameshifting on E. coli ribosomes, and by two seemingly different mechanisms. One is overlapping reading at the hungry codon. like the cases we have analysed elsewhere (Weiss and Gallant, 1986; Gailant and Linsdiey, 1992) . The second is zero-frame reading and subsequent siippage-inserting leucine at position 10 in this caseanaiagous to other studies of the translation of HIV (Jacks etai, 1988b; Weiss etal.. 1989) or rous sarcoma virus (RSV) (Jacks etai. 1988a) in heteroiogous systems.

In fact, a close reading of the data of Jacks e/a/. (1988b) reveais that the HiV sequence also displayed both modes of frameshifting in the eukaryotic cell-free translation system they analysed. The predominant amino acid at the frameshift site In their study was leucine. but there was aiso about 20-25% phenyialanine (see Fig. 2 of Jacks et ai, 198Bb) . indicative of overiapping reading. Our results in unstarved cells (our Fig. 2) indicate about 30% phenyialanine.

These two modes of frameshifting share fundamentai similarities, most strikingly in regard to their sequence requirements in the four positions to the left of the frameshift site. In the case of the mechanism of (0) frame tRNA binding followed by slippage, mutation studies in this region strongiy suggest that the frameshift is facilitated by a leftward slip of peptidyl-tRNA in the P-site (Jacks et al.. 1988a.b; Wilson ef al., 1988; Weiss et ai, 1989) . In the case of tiie overlapping reading mechanism, our mutation studies suggest a similar involvement of a peptidyl shiiX {KO\OT et ai. 1993) . Thus, in both mechanisms the leftward slippage at the ribosome's A-site is facilitated if the adjacent peptidyl-tRNA finds complementary pairing with a left-shifted triplet at the P-site.

The distinction between the two meciianisms seems, at first thought, to reside in the order in which aminoacyl-tRNA binding and slippage of the message occur. However, this distinction may be more apparent than real. It could be that the aminoacyl-tRNA that reads tiie overlapping codon first enters the A-site in the (0) frame by way of a mismatch. In the present case, this would mean phe-tRNA pairing initialiy with the UUA leucine codon, an interaction which demands one mismatch in the third position. This non-cognate interaction would of course be favoured by a shortage of leucyl-tRNA, particularly in our relA~ host cells which suffer increased recognition errors at hungry codons (Gallant, 1979; Cashel and Rudd, 1987; Parker. 1989 ).

Subsequently, leftward slippage of the phe-tRNA in the A-site to the overiapping UUU triplet would provide complementarity at all three positions, and thus improve stabiiity. in this way, an initial recognition error in the (0) reading frame could be resolved by way of a reading frame error, an example of what has elsewhere been termed 'error coupiing' (Kuriand and Gaiiant. 1986 ).

The poi gene product expressed from intact HiV in human host cells has yet to be characterized at the amino acid sequence level. Consequently, it is unknown whether the amino acid at the frameshift junction is ieucine or phenylaianine (or a mixed population), and it is therefore unknown which mode(s) of frameshifting operates under natural conditions. The overlapping reading mechanism is worth bearing in mind because of the obvious possibilities it presents for physiological controi.

In prokaryotic systems, frameshifting by overlapping reading is a natural outcome of imbalances in the aminoacyl-tRNA pools (Atkins et ai. 1979; Gallant and Foley. 1980; Weiss and Gallant, 1983; 1986; Kuriand and Gallant, 1986; Bruce et ai. 1986; Weiss and Gallant. 1986 ). Our present resuits demonstrate that a translational frameshift signal from HIV-1 can be reguiated in this fashion in response to changing ieveis of leucyl-tRNA. Similar regulation might be a simpie but effective metliod for establishing physioiogical control of the expression of any frameshifted gene. Many such genes function to replicate and spread viruses, transposable elements, and retroviruses (reviewed by Hatfield and Oroszlan, 1990; Cattaneo. 1989) . In these cases, ribosome frameshifting during aminoacyl-tRNA imbalance might represent a mechanism for iinking element movement to the state of the celiuiar economy.

The degree of aminoacylation of tRNAs provides a sensitive difference signal of ceilular activity. Anabolism generates the amino acids which charge the tRNAs, while translation drains these amino acids into growing proteins. Hence, the tRNA charging levei reflects the balance of these two flows, in bacterial celis, tRNA aminoacylation levels operate both specific attenuation control circuits (Landick and Yanofsky, 1987) and the highly pleiotropic stringent controi system (Gallant, 1979; Cashel and Rudd. 1987) . and may regulate the expression of individual translation factor genes (Grunberg-Manago, 1987) .

This speculation might provide a rationale for the various different shifty sequences found in the gag-pol aj\6 gagpro-pol overiaps of different retroviruses, and the comparable regions of different coronaviruses. Perhaps each unique shifty sequence has been evoiutionariiy optimized to respond to a tRNA deficiency that is the most sensitive difference signal for its particular host ceils. Likewise, evoiutionary pressure to maintain a response to a cell-specific aminoacyl-tRNA signal might also explain why retroviruses maintain non-optimal translational frameshift sequences which must in some cases be augmented by baroque mRNA secondary structures downstream (Weiss et ai. 1989) .

Finally, we might note that our data do not distinguish between a simultaneous slippage mechanism at the leucine codon (as proposed by Jacks and Varmus) and an overlapping reading mechanism at the next codon. If ribosomes in unstarved cells paused at the GGG codon following the leucine codon, then arg-tRNA could enter by overlapping reading and shift the reading frame leftward (see Fig. 4 ). The amino acid sequence in this case would be exactly that ascribed to simultaneous slippage, and which we have demonstrated in unstan/ed E. coli cells. We have no independent reason to expect ribosome stalling at the GGG codon in growing celis, except that it is a rare codon and caiis for a rare tRNA. Experiments designed to investigate this possibiiity are under way.

Frameshift alleles were constructed by modification of the plasmid vector pBWIIOO. pBWIlOO, a kind of gift of Bob Weiss, is a modified pBR322 in which nucleotides 1-1417 were replaced by nucleotides 1-4625 (NIH GenBank co-ordinates) of the E coli lac operon. The lac operon fragment has been modified by the deletion of nucleotides 641-1070, the removal of the natural EcoRI site within the /acZ gene, and by the addition of the pUC9 polylinker (Vieira and Messing, 1982) between codons 4 and 5. The resultant pBWIlOO is Lad Z*Y"A , and confers resistance to ampicillin. In typical constructions, about 1 jig of pBWIlOO was cut to completion with restriction endonucleases H/ndlll and SamHI (Bethesda Research Labs) at their single sites within the pUC9 polylinker. About lOOng of the prepared vector was used without purification in a 20 nl ligation containing a 10-fold molar excess of paired complementary oligonucleotide primers, kindly prepared by Yim Foon Lee (Howard Hughes Institute, University of Washington) on an Applied Biosystems 380B synthesizer. Ligations were carried out at room temperature for 3-12 h. then transformed into a lac~ strain for screening on Xgal indicator plates. Plasmid DNA from (leaky) lac candidates was examined by restriction analysis and by DNA sequencing of the salient region of the lacZ gene. Sequences of the oligonucleotide pairs were: HIV13/HIV13-A3, 5'-AGCTCTAATT^TTTAGG-GA-3; and 5'-GATCTCCCTAAAIAATTAG-3'; H1V201, 5-AGCTCTAAI 1 I I I IAGGGAAGG-3' and 5-GATCCCTTCC-CTAAAAAATTAG-3'; HIV201-U7, 5'-AGCTCTAAl 1 I I I I I-GGGAAGG-3' and 5'-GATCCCTTCCCAAAAAAATTAG-3'.

Bacterial strains and culture conditions were as described previously (Weiss etai., 1988) , except that minimal glucose-M63 medium was supplemented only with isoleucine and valine in addition to the strains' growth requirements. The lacZ plasmids were kept under selection through the presence of 1mgml ^ carbenicillin. tn most experiments. the host cells also carried pPYIOII, a compatible plasmid (spectinomycin resistant, streptomycin resistant) carrying a complete lacl'^ gene. Selection for this plasmid was maintained through the presence of 50|igmr'' spectinomycin in the overnight cultures used to inoculate our experimental cultures, but spectinomycin was not present during induction of lacZ.

Cell-free extracts were made by resuspending cells at a final protein concentration of 1 -2 mg ml" \ and sonicating for 20 s at ice-water temperature in a Braunsonic 1510 instrument; cell debris was centriiuged out at 26000 x g in a Sorvaii refrigerated centrifuge. Beta-galactosidase activity was measured by incubating up to 50 MI of extract with 250 iil ONPG (0.8mgml ' in 50 mM sodium phosphate buffer. pH 7.0) at room temperature, and monitoring the accumulation of o-nitrophenol at A420 using a continuously recording Gilford 2400-2 spectrophotomer. One enzyme unit is defined as the amount of enzyme required to produce a change in absorbance of one per minute. Protein concentrations were determined by the method of Lowry et al. (1951) using egg albumin as a protein standard. Differential rates in enzyme units per mg of protein were computed for the period of Induction (approximately one doubling in 2mM IPTG, 2.5 mM cAMP) with or without growth limitation (as indicated).

Purification of 0-galactosidase by immunoprecipitation with specific antiserum was as described in Gallant and Lindsiey (1992) . N-terminal amino acid analysis of the purified protein was performed by automated Edman degradation on Applied Biosystems 477 A pulsed-liquid phase sequencer and 470 A gas phase sequencer at University of Illinois Biotechnology Center and University of California, Riverside. Data were analysed using Applied Biosystems software for background and lag correction.

Low activity of p-galactosidase in frameshift mutants of Escherichia coii

Normal tRNAs promote ribosomal frameshifting

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

tRNA anticodon replacement experiments show that ribosomal frameshifting can be caused by doublet decoding

Synthetic a-amino-phydroxycaproic acids

The Stringent Response

How 'hidden' reading frames are expressed

Efficient translational frameshifting occurs within a conserved sequence of the overlap between two genes of a yeast Tyl transposon

Translational frameshifting: where will it stop?

Bacterial peptide chain release factors: conserved primary structure and possible frameshift regulation of release factor 2

Use of tRNA suppressors to probe regulation of Escherichia coli release factor 2

Mechanism of the rel defect in beta-galactosidase synthesis

Stringent Control in E. coli

On the causes and prevention of mistranslation

Leftward ribosome frameshifting at a hungry codon

Some puzzle of translational accuracy

Regulation of the expression of aminoacyl-tRNA synthetases and translation factors

The where, what, and how of ribosomal frameshifting in retroviral protein synthesis

Automated amino acid sequence assignment: Development of a fully automated protein sequencer using edman degradation

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

Characterization of ribosomal frameshifting in HIV-1 gag-poi expression

On the role of the P-site in leftward ribosome frameshifting at a hungry codon

The secret life of the ribosome

Transcription Attenuation

On the directional specificity of ribosome frameshifting at a hungry codon

Protein measurement with the Folin phenol reagent

Signals for expression of HIV pol gene by ribosomal frameshifting. In The Control of Human Retroviral Gene Expression

Errors and alternatives in reading the universal genetic code

Context rules of rightward overiapping reading

In vivo role of the relA* gene in the regulation of the /ac operon

Positive control of lac operon expression in vitro by guanosine 5'-diphosphate 3'-diphosphate

Complete nucleotide sequence of the AIDS virus, HTLV-III

The pUC plasmids, a M13mp7 derived system for insertion mutagenesis and sequencing with synthetic universal primers

Mechanism of ribosome frameshifting during translation of the genetic code

Frameshifting suppression in aminoacyl-tRNA limited cells

Slippery runs, shifty stops, backward steps, and forward hops: -2, -1, +1, +2, -i-5. and +6 ribosomal frameshifting

On the mechanism of ribosomal frameshifting at hungry codons

E. coli ribosomes re-phase on retroviral frameshift signals at rates ranging from 2 to 50 percent

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

Differential sensitivity of gene expression in vitroio inhibitation of DNA gyrase

This work was supported by Grant GM13626 and from Grant GM07735 from the National Institutes of Health.