key: cord-0008743-hy8zihx6
authors: Hatfield, Dolph L.; Levin, Judith G.; Rein, Alan; Oroszlan, Stephen
title: Translational Suppression in Retroviral GENE Expression
date: 2008-02-28
journal: Adv Virus Res
DOI: 10.1016/s0065-3527(08)60037-8
sha: 253d52a717629685ceadeab5430aedaaaee99566
doc_id: 8743
cord_uid: hy8zihx6

This chapter summarizes the present state of knowledge concerning translational suppression in retroviruses. Other viruses, using similar mechanisms, are mentioned only briefly and tangentially. Retroviruses are a unique class of viruses that have been found in all classes of vertebrates but not in other organisms. Perhaps, their most distinctive properties are the flow of information from RNA to DNA early in the infectious process, and the subsequent integration of the viral DNA into the chromosomal DNA of the host cell. Retroviruses are the causative agents of acquired immunodeficiency syndrome (AIDS) and of a variety of neoplastic diseases in man and domestic animals. Elements with striking similarities to retroviruses, termed retrotransposons, occur in yeast and many other eukaryotes; elements sharing some characteristics with retroviruses have also recently been observed in prokaryotes. Because of the apparent relationship between retroviruses and retrotransposons, this chapter discusses of retrotransposons as well as retroviruses. Though all retroviruses utilize translational suppression in pol-protein synthesis, different groups of retroviruses use two completely distinct types of translational suppression. One of these is in-frame or readthrough suppression and the other is ribosomal frameshifting.

Retroviruses are a unique class of viruses that have been found in all classes of vertebrates, but not in other organisms. Perhaps their most distinctive properties are the flow of information from RNA to DNA early in the infectious process, and the subsequent integration of the viral DNA into the chromosomal DNA of the host cell. Retroviruses are the causative agents of acquired immunodeficiency syndrome (AIDS) It is interesting to note that elements with striking similarities to retroviruses, termed retrotransposons, occur in yeast and many other eukaryotes; elements sharing some characteristics with retroviruses have also recently been observed in prokaryotes (reviewed in Garfinkel, 1991) . Because of the apparent relationship between retroviruses and retrotransposons, we will consider retrotransposons as well as retroviruses in this review.

In addition to structural proteins, all retroviruses encode at least three enzymes: a protease (PR), which processes the internal proteins of the virion during virus maturation; reverse transcriptase (RT), which copies the genomic RNA of the virion into DNA when the particle infects a cell; and integrase (IN) , which catalyzes the insertion of the viral DNA into the chromosomal DNA of the host cell (Dickson et al., 1984) . Because these enzymes all function in the free virus particle or in the early stages of infection, they must be incorporated into the virion during virus assembly. However, because they act catalytically, they are needed in much lower amounts than the structural proteins of the virion.

Retroviruses have evolved a remarkable mechanism for expression of the genomic sequences encoding these enzymes. This mechanism, translational suppression, appears to fulfill simultaneously both of the requirements noted above, because it results in a relatively low level of expression of the enzymes and provides a way for the enzymes to be incorporated into the nascent virus particle.

It is now clear that the genomic RNA of the virus is the mRNA for the internal structural proteins of the virus (termed the Gag proteins) and the viral enzymes. As indicated in Fig. 1 , the gag coding sequences are found a t the 5' end of this mRNA. The enzyme-coding region (generally referred to as the pol gene) is immediately 3' of the termination codon a t FIG. 1. Expression of retroviral pol and pro genes from a single gag-pro-pol translational unit by in-frame readthrough, single frameshift, and double frameshift, as illustrated for the respective viruses. The gag, pro, and pol open reading frames are shown together with the following symbols for termination codons: 0 UAG; A, UAA, and V, UGA. The boldface horizontal bars represent the primary translational products, which are processed into smaller functional units as indicated. The protein nomenclature used is that of Leis et al. (1988) ; MA, matrix; CA, capsid; NC, nucleocapsid; PR, protease; RT, reverse transcriptase; and IN, integrase. is, In-frame suppression; and fs, frameshift. Locations of transframe proteins (TF) are also shown. The numbers indicate the approximate molecular weights of the proteins. Arrows indicate the site of translation initiation. in-frame readthrough: the end of the gag gene. Most of the ribosomes engaged in Gag protein synthesis terminate peptide chain elongation in response to this termination codon (as would be expected), resulting in the synthesis of the Gag structural polyprotein. However, a minority of these ribosomes engage in translational suppression: that is, they continue peptide synthesis beyond the termination codon, generating a large Gag-Pol fusion protein. Because the Gag polyprotein precursor normally performs the self-assembly processes responsible for virus assembly, it seems very likely that the Gag moiety of the Gag-Pol fusion protein participates in this self-assembly along with the authentic Gag polyprotein, so that the Gag-Pol fusion protein is incorporated into the virus particle. Thus the use of translational suppression in pol gene expression simultaneously modulates the relative level of Pol protein synthesis and provides for the inclusion of the Pol proteins in the virion. Though all retroviruses utilize translational suppression in Pol protein synthesis, different groups of retroviruses use two completely distinct types of translational suppression. One of these is in-frame or readthrough suppression, and the other is ribosomal frameshifting.

In the viruses using readthrough suppression, the gag and pol coding sequences are in the same reading frame and are separated by a single UAG termination codon. A minority of the ribosomes engaged in Gag protein synthesis insert a n amino acid in response to this UAG triplet, rather than terminating synthesis and releasing the product, and then continue beyond it to translate the pol sequences.

In contrast, in the viruses using ribosomal frameshifting, the gag and pol coding sequences are out of frame with respect to each other, with the pol sequences placed in the -1 position relative to the gag sequences. At some point near, but prior to the termination codon signaling the end of the gag gene, a minority of the ribosomes engaged in Gag protein synthesis translate a codon and insert the corresponding amino acid, but advance only two, rather than three, bases. Thus, this subpopulation of ribosomes shifts from the gag reading frame to the pol reading frame: this transition allows it to bypass thegag termination codon and synthesize a Gag-Pol fusion protein. (As will be discussed below, some yeast retrotransposons exhibit ribosomal frameshifting in the + 1, rather than the -1, direction.)

The distinction between these two mechanisms should be emphasized. Ribosomal frameshifting may occur a t a considerable distance on the mRNA from the gag termination codon, and eliminating this termination codon by mutation does not prevent frameshifting (Jacks et al., 1988a) . Thus, frameshifting is quite independent of the presence of the termination codon. In contrast, readthrough suppression represents unusual behavior of ribosomes at a termination codon.

Some retroviruses, including mouse mammary tumor virus (MMTV), the mammalian type D viruses, and members of the human T cell leukemia virus (HTLV) group, actually use ribosomal frameshifting twice, rather than once, in the synthesis of the pol gene product: once between the gag and protease-coding sequences, and again between the protease gene (pro) and that for RT and IN (see Fig. 1 ). (Another deviation from the general schemes presented above should also be noted: in the avian type C and the foamy retroviruses, the PR is encoded on the 5' side of the gag termination codon, so that it is encoded within the gag region rather than the pol region of the genome.)

The present review summarizes our present state of knowledge concerning translational suppression in retroviruses. Other viruses using similar mechanisms are mentioned only briefly and tangentially. For a description of the historical development of our understanding of this subject (as regards retroviruses), readers are referred to an excellent review by Jacks (1990). 11. READTHROUGH SUPPRESSION As noted above, a single inframe UAG termination codon separates the gag and pol genes in the mammalian type C retroviruses (Shinnick et al., 1981; Tamura, 1983; Herr, 1984; Etzerodt et al., 1984; Kato et al., 1987; Weaver et al., 1990) . The synthesis of a large Gag-Pol fusion protein in murine leukemia virus (MuLVbinfected cells, at a molar ratio of approximately 1:20 with that of the Gag polyprotein, was originally observed by Jamjoom et al. (19771, who suggested the existence of a translational control mechanism governing the synthesis of this product. As discussed in a recent review by Jacks (1990) , an obvious alternative was the presence of a distinct gag-pol mRNA from which the termination codon in the viral genome had been removed by splicing.

Virtually all of the studies on the mechanism of Gag-Pol synthesis in mammalian type C retroviruses have been performed with MuLVs. One important experimental approach that has been used in analyzing the synthesis of the fusion protein has been in uitro translation (Jackson and Hunt, 1983) . Early experiments (Kerr et al., 1976; Murphy et al., 1978) showed that translation reactions programmed with virion RNA were capable of synthesizing the Gag-Pol precursor as well as the Gag polyprotein. Indeed, the ratio of Gag to Gag-Pol produced in these in uitro systems appeared to be comparable to that observed in the infected cell. Thus gag-pol mRNA is present in the virus particle. Because virion RNA is apparently a single, homogeneous species, the gag-pol mRNA is evidently indistinguishable from the genomic RNA of the virus in its approximate size and composition. However, these observations could not exclude the possibility that virions contained a second RNA species, distinct from the genomic RNA by virtue of the fact that the termination codon a t the end of gag had been removed by splicing.

In a significant extension of these studies on in uitro translation of virion RNA, it was found that the relative level of synthesis of the Gag-Pol fusion protein could be increased by the addition of purified yeast amber suppressor tRNA (Philipson et al., 1978; Murphy et al., 1980) . Although no sequence data were available at the time these experiments were performed, this finding strongly suggested that a translational suppression mechanism was responsible for the synthesis of the Gag-Pol fusion protein, and that a UAG codon was present between the gag and pol regions of the viral genome.

More recently, these experiments have been refined by using mRNA synthesized in uitro from an infectious clone of proviral DNA, rather than RNA isolated from virions, to direct the synthesis of both Gag and Gag-Pol polyproteins in rabbit reticulocyte lysates (Feng et al., 1989a) . In general, results obtained with this system reflected the earlier observations with viral RNA quite closely. The synthesis of Gag-Pol product in response to this synthetic, presumably completely homogeneous mRNA is obviously strong support for the idea that the two proteins are synthesized from the same template as a result of translational suppression.

Remarkably, the mechanism of synthesis of the Gag-Pol fusion protein was in large part clarified by a single, seminal observation. In 1985, Yoshinaka et al. reported the isolation and N-terminal amino acid sequence of the PR of Moloney murine leukemia virus (Mo-MuLV) (Yoshinaka et al., 1985a) . Because the PR gene is at the 5' end of the pol coding region (Levin et al., 1984; Crawford and Goff, 1985; Katoh et al., 19851, it was expected that the protein would begin with an amino acid sequence encoded entirely 3' of the gag termination codon. However, when the actual N-terminal sequence was compared with the MuLV proviral DNA sequence (Shinnick et al., 19811, it was found that the first four residues represented the last four codons of the gag gene. They were followed by a glutamine residue, and then by the amino acids encoded at the 5' end of the pol gene. It thus appears that the PR is formed by cleavage of the Gag-Pol fusion protein at a site four residues before the last gag-encoded amino acid. Similar results were also obtained by sequencing the PR of feline leukemia virus (FeLV) (Yoshinaka et al., 1985b) . The protein sequences showed clearly that (1) PR is synthesized by translation of a mRNA containing the UAG termination codon present in the viral genome, rather than by a spliced mRNA lacking this codon, and (2) the UAG termination codon is suppressed in uiuo by a glutamine tRNA.

1. Comparison of Cellular and Viral Termination Signals: Evidence for "Suppression Signal" in Viral mRNA

The fact that the UAG codon at the gag-pol junction of mammalian type C retroviral RNAs is suppressed a t a significant level implies that normal cells of higher eukaryotes contain tRNAs capable of inserting an amino acid in response to this termination codon. However, the UAG termination codons found a t the ends of many coding sequences in the cellular genome (Kohli and Grosjean, 1981; Brown et al., 1990) are not suppressed significantly (Capone et al., 1986; Sedivy et al., 1987; Martin et al., 1989) . How can we explain this striking difference between translation of cellular and viral mRNAs? One simple explanation for the efficiency of termination at normal cellular termination sites would be that they are actually multiple, tandem termination codons. However, survey of a number of eukaryotic coding sequences showed that this is not the case (Kohli and Grosjean, 1981; Brown et al., 1990) . Another possibility is that normal termination codons are associated with signals favoring termination, in essence "protecting" them from the cellular tRNAs capable of suppression. Finally, the viral mRNA may contain signals promoting the suppression event. As discussed by Valle and Morch (1988) and below, it seems likely that both of these latter hypotheses are correct.

The nature of the difference between cellular and viral mRNAs was approached by the construction and analysis of nonsense mutants (mutants containing termination codons at internal positions) in reporter genes, including chloramphenicol acetyltransferase (Capone et aZ., 1986; Martin et al., 1989) and poliovirus (Sedivy et al., 1987) . These termination codons are thus at sites that, unlike the retroviral gag-pol junction, are not designed for efficient suppression, but are also not the location of natural termination codons. When these mutant genes were expressed in normal mammalian cells, it was found that they are not suppressed to a detectable extent (Capone et al., 1986; Sedivy et al., 1987; Martin et al., 1989) . This observation suggested that suppression during translation of the viral mRNA occurs because this RNA contains positive signals favoring suppression.

The possibility that type C retroviral mRNA contains signals promoting suppression was tested directly in a series of experiments by Panganiban (1988) . He isolated a restriction fragment of the MuLV genome containing 37 codons from the 3' end ofgag, the UAG codon a t the gag-pol junction, and 62 codons from the 5' end of pol. This fragment, corresponding to the region of MuLV mRNA near the gag-pol junction, was inserted into a construct at the 5' end of the lac2 gene. When this construct was transfected into mammalian cells, a significant level of readthrough of the UAG codon was observed; indeed, the level of suppression (about 10%) was quantitatively comparable to that which actually takes place during translation of the viral genome in uiuo. A control construct, containing an inframe UAG codon but lacking the viral sequences, showed no detectable suppression. In a somewhat analogous experiment, Honigman et al. (1991) changed a CAG codon within the Mo-MuLVgag gene (nt 1623 -1625 Shinnick et al., 1981) to UAG, and observed no suppression of this UAG codon in an in uitro translation system.

These results provided direct evidence that the viral mRNA contains signals promoting suppression a t the gag-pol junction, and also implied that these signals do not extend beyond the limits of the restriction fragment used in Panganiban's experiments (Panganiban, 1988) . In addition, because the constructs did not encode any viral proteins, and because essentially identical results were obtained in both virusinfected and uninfected cells (Panganiban, 19881 , the data argue that the viral signals that result in suppression at the gag-pol junction are completely cis-acting.

On the other hand, a number of studies suggested that natural termination codons are, in fact, found in contexts that are unfavorable for suppression. Thus, when nonsense suppressor tRNAs (mutant tRNAs whose anticodons pair with termination codons, but which can be acylated with amino acids and function in translation) were introduced into higher eukaryotic cells together with the nonsense mutants of chloramphenicol acetyltransferase and poliovirus discussed above, the suppressor tRNAs could be shown to suppress these termination codons (Capone et al., 1986; Sedivy et al., 1987) . However, the presence of these tRNAs had a surprisingly small effect on the pattern of cellular protein synthesis observed in two-dimensional electrophoresis (Bienz et al., 1981) . There was also very little effect on cell growth (Sedivy et al., 1987) . This observation, that the suppressor tRNAs did not detectably suppress many of the "natural" termination codons at the ends of cellular genes, strongly suggested that the latter codons are in contexts protecting them from suppression (or promoting efficient termination).

At present, there is little information on the nature of the signals discussed above. In prokaryotes, it has been shown that nonsense mutants of l a d (Miller and Albertini, 1983; Bossi, 1983) or other genes (Engelberg-Kulka, 1981 ) are more efficiently suppressed if the nucleotide immediately following the termination codon is a purine. Studies of this type have not, to our knowledge, been performed in eukaryotes.

When the sequence around natural termination codons in eukaryotic genes was analyzed, a very strong bias was found for purines at the position immediately 3' of the termination codon (Kohli and Grosjean, 1981; Brown et al., 1990) . This bias was even more striking in genes expressed a t a high level (Brown et al., 1990) . These observations might suggest that a purine at this position is a n important element of the hypothetical signal promoting efficient termination at natural termination sites. However, the suppressible termination codon at the gagpol junction of all known mammalian type C retroviruses is also followed by a G residue! Clearly, the viral signal promoting suppression must extend beyond this position.

One approach that might point t o signals favoring suppression is to compare sequences of different viruses that use readthrough suppression, to determine whether conserved sequences occur near the suppressible termination codon. Figure 2 presents the sequences of the 20 codons on either side of the gag termination codon of Mo-MuLV (Shinnick et al., 19811, AKRMuLV (Herr, 19841 , spleen necrosis virus (SNV) (Weaver et al., 19901 , and baboon endogenous virus (BaEV) (Kato et al., 1987) . (The latter viruses are much more distantly related to the two MuLVs than the two MuLVs are to each other.) Since all of these sequences were obtained from infectious molecular clones, they all represent portions of mRNAs which successfully engage in readthrough suppression.

Inspection of Fig. 2 shows that there is very limited sequence conservation [8 out of 60 nucleotides (nt), or 1381 on the 5' side of the termination codon. However, there are several striking features on the 3' side. These include a GG pair immediately beyond the UAG codon; a GU CAG GG sequence in the second, third, and fourth pol codons; a run of (Herr, 19841 . spleen necrosis virus (SNV1 (Weaver et al., 19901 , and baboon endogenous virus IBaEV) (Kato et al., 1987) . The nucleotide sequence 60 bases 5' of the UAG termination codon and 60 bases 3' ofthe UAG codon in Mo-MuLV, AKR MuLV, SNV. and BaEV is shown. Nucleotide positions are indicated for Mo-MuLV (Shinnick et nl., 1981,. Nucleotides that are identical in all four viruses are denoted by an asterisk. The UAG codon present in each sequence is boxed. six pyrimidines in the seventh, eighth, and ninth pol codons (all C residues except for one U in spleen necrosis virus); and a run of six G residues, followed by CA, in the eighteenth, nineteenth, and twentieth pol codons. There are also a number of conserved bases between the run of pyrimidines and that of the Gs. In all, nearly 60% of the bases in this 60-nt stretch are identical in the four viruses. The degree of conservation observed on the 3' side of the gag termination codon is strongly suggestive of a possible role for these sequences in suppression.

One obvious possibility is that the signal for suppression is contained in secondary structures in the viral RNA, rather than in specific sequences; such structures clearly play a role in many instances of ribosomal frameshifting (see below). One candidate structure is a potential stem-loop in MuLV depicted in Fig. 3 . To investigate this possibility, Jones et al. (1989) made point mutations in sequences surrounding the Mo-MuLV gug-pol junction that would destroy the stem and measured the effect on viral infectivity. Changes that would allow base pairing in the stem (Fig. 31 , e.g., C2220 to U and G2252 to A, or C2220 to U alone, led to the production of infectious virions, whereas mutations that would destabilize the secondary structure, e.g., A 2223 to C and G2252 to A, or G2252 to A alone, did not. On the basis of these observations,

Flc. 3. Potential RNA secondary structure in Mo-MuLV RNA a t the gag-pol junc- Jones et al. ( 1989) suggested that a region of secondary structure near the UAG codon must be preserved; however, their study did not test the effect of destabilizing mutations upstream of the UAG codon (e.g., A2223 to C alone) nor did they measure suppression directly. In a related series of experiments carried out in uitro, Honigman et al. (1991) introduced destabilizing mutations into residues in the stem of the putative stem-loop structure ( Fig. 3 ) a t positions 5' (nt 2222-2226, GACCC to AAUAU) and 3' (nt 2246-2250, GGGUC to UCAUG) of the UAG codon. The upstream mutation had no effect on suppression in the in uitro system; in contrast, the downstream mutation prevented readthrough. These results led Honigman et al. (1991) to conclude that a secondary structure involving the UAG codon and nearby nucleotides at the MuLV gug-pol junction is unlikely to be important in suppression and in addition emphasized the role of the downstream sequences. It is interesting that stem-loop structures similar to the one shown in Fig. 3 probabiy do not exist in the viral mRNAs of other mammalian type C retroviruses that undergo readthrough suppression (Panganiban, 1988; ten Dam et al., 1990) . Even more importantly, it has been pointed out (Panganiban, 1988 ) that a stem-loop structure containing the UAG codon is unlikely, because the structure would have to be disrupted before ribosome movement and misreading of the UAG codon could occur.

Computer analysis of the Mo-MuLV RNA sequence (ten Dam et al., 1990 ) also raised the possibility that the gag-pol mRNA contains a pseudoknot structure (pseudoknots are stem-loop structures in which the bases in the loop are paired with bases downstream of the stem [Pleij et al., 1985; Schimmel, 1989 ) see Section 111). This structure might involve a long stretch of six C residues (nt 2256 -2261 Shinnick et al., 1981) beginning 19 nt downstream of the UAG codon, which could interact with a run of six G residues (nt 2289-2294) to form the second base-paired region of the pseudoknot. These runs of C and G residues are fairly well conserved in other mammalian type C retrovirus genomes (ten Dam et al., 1990) ( Fig. 2 ). In one study it was found that mutation of nucleotides in the first stem of a putative pseudoknot structure inhibited readthrough in uitro, but compensatory mutations did not restore activity (Honigman et al., 1991) .

Though particular structures have not yet been identified, it is clear that mutations in downstream sequences have an inhibitory effect on readthrough. In addition, it is noteworthy that mutation of a conserved sequence (Honigman et al., 1991) (Fig. 2 ) immediately 3' of the Mo-MuLV UAG termination codon, GGAG (nt 2238-2241) to ACGC, completely abolished in uitro synthesis of a Gag-Pol fusion protein (Honigman et al., 19911. In summary, the mutational data, as well as the sequence conservation (Fig. 21 , are all consistent with the possibility that, as in prokaryotes (Engelberg-Kulka, 1981; Miller and Albertini, 1983; Bossi, 1983) , the suppression signal is contained within the downstream sequences.

The exact number of nucleotides required for readthrough in the MuLV system is under investigation. Based on a n analysis similar to that shown in Fig. 2 , Feng et al. (1990a) designed a miniconstruct of Mo-MuLV mRNA containing the last two codons of gag, the UAG termination codon at thegag-pol junction, and the first 19 codons ofpol, and could show that the UAG codon was suppressed in rabbit reticulocyte lysates (Feng et al., 1990a) . This result suggested that a limited region of viral mRNA contains all the sequences needed for suppression and provided additional evidence that these sequences are largely, if not entirely, downstream sequences.

The mutational analysis discussed above indicates that at least part of the signal that governs readthrough suppression is contained within the primary sequence of the viral mRNA. An important question to consider is whether this signal is specific for the UAG codon or whether other termination codons can be substituted within the nucleotide context required for suppression. Feng et al. (1989b) used oligonucleotidedirected mutagenesis to change the UAG codon at the Mo-MuLV gagpol junction to UAA or UGA. Both UAA and UGA were suppressed with the same efficiency as UAG in rabbit reticulocyte lysates (Feng et al., 1989a,b) . In the case of UAA, however, the system had to be supplemented with additional tRNA; rabbit liver tRNA or tRNA from uninfected or MuLV-infected NIH/3T3 cells were equally effective (Feng et al., 198915) . This observation suggested that the tRNA that suppresses UAA is not abundant in the usual calf liver tRNA-supplemented rabbit reticulocyte lysate (Jackson and Hunt, 1983; Feng et al., 1989a) and that the UAA suppressor tRNA is not unique to mouse cells. In uiuo experiments carried out by transfecting intact viral genomes with UAA or UGA instead of UAG into Chinese hamster ovary (CHO) cells (conditions that do not permit selection of revertants to wild type) led to production of infectious virions with approximately the same titer as wild type (Feng et al., 1989b) . Similarly, the viral capsid (CAI protein and the Pol proteins, RT and IN, were present in equivalent amounts in virions derived from mutant and wild-type genomes (Feng et al., 1989b3 . Indeed, the Gag and Gag-Pol precursor proteins were synthesized to the same extent in the cells transfected with UAG-, UAA-, or UGA-containing viral genomes (A. Rein, unpublished observations 1989) . From these results, Feng et al. (1989b) concluded that (1) the signal(s) for UAG suppression are effective with UAA and UGA; (2) UAA is not an absolute termination codon in higher eukaryotes, as had been previously thought (Geller and Rich, 1980; Valle and Morch, 1988) ; and (3) mammalian cells and cell extracts contain tRNAs capable of suppressing UGA and UAA termination codons that appear in a retroviral context. Jones et al. (1989) also reported that mutant Mo-MuLV viral genomes with alternate stop codons can give rise to infectious virus particles.

The discovery that all three termination codons are suppressible in the MuLV system raised the possibility of identifying previously unknown suppressor tRNAs. The approach used was to translate a miniconstruct mRNA that has a short leader sequence containing an AUG codon, followed by two codons from the 3' end of gag, a termination or sense codon, 19 codons from the 5' end ofpol, and the binding domain of protein A. The N-terminal amino acid sequence of the product was then determined by the Edman degradation technique (Feng et al., 1990a) . The predicted amino acid sequence of the first 20 amino acids of the fusion product is shown in Fig. 4 ; position 8 represents the residue a t the Gag-Pol junction. Comparison of the Edman degradation patterns obtained using mRNAs with UAG and CAG (a glutamine code word) showed that the relative amounts of radioactive glutamine incorporated a t the readthrough site and a t two positions in Pol (Fig. 4) were the same for both products. This result demonstrated that UAG is translated predominantly, if not exclusively, as glutamine and pointed to the parallel between suppression in uitro and in uiuo, where the UAG termination codon is known to be read as glutamine (Yoshinaka et al., 1985a) . In addition, sequence analysis of the UAA fusion protein showed that UAA, like CAG and UAG, directs the sole incorporation of glutamine at the Gag-Pol junction (Feng et al., 1990a) . This finding represented the first (and to date only) identification of an amino acid inserted in response to UAA in a higher eukaryote. Surprisingly, in the case of UGA, three amino acids, arginine, cysteine, and tryptophan, were inserted a t the GAG-Pol junction (Feng et al., 1990a) . It is of interest that early codon recognition studies (Marshall et al., 1967; Caskey et al., 1968; Hatfield, 1972) indicated that arginine, cysteine, and tryptophan tRNAs can respond to UGA in the ribosomal binding assay of Nirenberg and Leder (1964) . However, misreading of UGA as cysteine and arginine during protein synthesis had not been previously described for higher eukaryotes.

The rabbit reticulocyte lysate system used for this study (Feng et al., 1989a) presumably contains other known mammalian suppressor tRNAs, i.e., two leucine UAG suppressors (Valle et al., 1987) and a UGA suppressor that is acylated with serine (Hatfield, 1972; Hatfield et al., 1982a) and ultimately converted in uiuo to selenocysteyl-tRNA . Despite the presumed presence of these suppressor tRNAs, leucine and serine were not inserted a t the Gag-Pol junction and were detected only a t the predicted positions ( Fig. 4) in the fusion products (Feng et al., 1990a) . This strict specificity exhibited by the MuLV suppression system raises the intriguing possibility that nucleotide context not only affects the efficiency of suppression, but also determines which tRNAs will function in suppression.

Whereas cis-acting viral sequences clearly play a significant role in MuLV readthrough suppression, nonviral trans-acting factors are also of major importance. These factors presumably include (1) the single mammalian release factor that mediates termination in response to all three termination codons (Konecki et al., 1977) ; (2) normal cellular tRNAs that act as suppressor tRNAs by misreading termination codons; and (3) other factors, such as ribosomal proteins or RNA sequences, which may have functional significance, but have not yet been identified in mammalian systems. The molecular mechanisms underlying the interaction between these factors are poorly understood. Presumably, there is a competition between suppressor tRNA and a release factor that determines whether suppression can take place. In addition, it has been proposed that suppression may be promoted by base pairing between one or two nucleotides immediately downstream of the termination codon in the message and the corresponding bases 5' of the anticodon in the tRNA (Engelberg-Kulka, 1981; Panganiban, 1988) .

Although it has been known for some time that a glutamine tRNA suppresses the UAG codon at the MuLV and FeLV gag-pol junctions (Yoshinaka et al., 1985a,b) , the glutamine isoacceptor that mediates this suppression has not been identified. Kuchino et al. (1987) sequenced two glutamine tRNAs from mouse liver: (1) a major species with the anticodon CUG and (2) a minor species (only 1-296 ofthe major species) having the anticodon U,UG. The sequences of these glutamine tRNAs were the same except for the 5' position of the anticodon and the nucleotides a t positions 4 and 68 in the acceptor stem; both tRNAs occurred in hypo-and hypermodified forms. Both forms of the minor tRNA species were able to weakly suppress the tobacco mosaic virus (TMV) UAG codon in an in vitro suppression assay, but neither form of the major species had any suppressor activity (Kuchino et al., 1987) . Interestingly, misreading of a UAG termination codon by the minor glutamine isoacceptor would involve unusual codon-anticodon recognition by mispairing at both the first and third positions of the codon. Because the MuLV and TMV systems are not identical, it is not clear that the minor glutamine tRNA functions as the suppressor in MuLV infection as proposed (Kuchino et al., 1987) . Thus, in contrast t o the situation with MuLV (Feng et al., 1990a1 , the tRNA specificity for i n uitro suppression of the TMV UAG codon is less stringent and several tRNAs exhibit suppressor activity, including glutamine (Kuchino et al., 1987) and leucine (Valle et al., 1987) as well as the hypomodified tyrosine tRNA, which is the TMV suppressor i n uiuo (Bienz and Kubli, 1981; Beier et al., 1984a) .

Several groups have considered the question of whether virus infection affects the suppressor tRNA population. Kuchino et al. (1987) reported that infection with Mo-MuLV increased the amount of the minor glutamine tRNA. Other investigators have obtained different results. Feng et al. (1989a) found that infection with Mo-MuLV did not change the chromatographic profile of glutamine isoacceptors or the level of glutamine acceptor activity. Moreover, it could be shown that suppression of the Mo-MuLV UAG termination codon i n uitro was stimulated to the same extent by tRNA isolated from MuLV-infected or uninfected NIH13T3 cells (Feng et al., 1989a) . Similarly, as noted above, Panganiban (1988) observed that i n uiuo suppression of the UAG codon a t the gag-pol junction occurred with the same efficiency in Mo-MuLV-infected and uninfected mouse cells. Taken together, these findings led Feng et al. (1989a) to conclude that the glutamine suppressor tRNA occurs normally within the tRNA population of uninfected cells and is not altered or induced in response to virus infection. In addition, the observation that all three termination codons can be suppressed with the same efficiency in vitro and i n vivo (Feng et al., 1989b) and the fact that several distinct suppressor tRNAs can function within the MuLV context (Feng et al., 1990a ; see below) are difficult to reconcile with a requirement for viral induction of suppressor tRNA.

The observation that a glutamine residue was inserted i n uitro in response to a UAA termination codon at the MuLV gag-pol junction (Feng et al., 1990a) clearly indicated that a glutamine tRNA mediates suppression of UAA in mammalian cells. As in the case of UAG, it is not known which isoacceptor functions as the UAA suppressor in the MuLV system. Suppression of both UAA and UAG termination codons by glutamine tRNAs has a precedent in yeast. The yeast glutamine tRNA, which can suppress UAA, normally decodes CAA (Pure et al., 1985) , whereas a different isoacceptor, which normally recognizes CAG, suppresses UAG (Weiss and Friedberg 1986; Lin et al., 1986; A. . Whether UAA and UAG are suppressed by two distinct glutamine tRNAs in mammalian cells as they are in yeast is not known. In this connection, it may be relevant that additional tRNA must be added to mammalian extracts for efficient suppression of UAA, but not of UAG (Feng et al., 1989b (Feng et al., , 1990a .

Because the UGA termination codon at the gag-pol junction was decoded as three amino acids, arginine, cysteine, and tryptophan, a number of different tRNAs must mediate UGA suppression. Tryptophan tRNA involvement in UGA suppression has already been observed in normal mammalian cells and bacteria. Geller and Rich (1980) proposed that mammalian tryptophan tRNA can function as a UGA suppressor based on their finding that partially purified tryptophan tRNA from reticulocyte lysates stimulates in uitro suppression of a UGA termination codon in P-hemoglobin mRNA. In bacteria, wild-type tryptophan tRNA and a mutant tryptophan suppressor tRNA with a G-to-A change at position 24 (Hirsh, 1971 ) decode the UGG tryptophan codon and UGA in vitro (Hirsh and Gold, 1971 ) and in uiuo (Raftery et al., 1984) . Interestingly, Buckingham and Kurland (1977) found that the suppressor tRNA also decodes the UGU cysteine codon with low efficiency in uitro. Because mammalian tryptophan tRNA, like its bacterial counterpart (Hirsh, 19711 , is expected to have a CCA anticodon, interaction with UGA may require C-A mispairing at the third position of the codon. Similarly, insertion of cysteine (UGU and UGC codons) in response to UGA would be expected to involve mispairing at the third position of the codon. Although arginine has six codons and several isoacceptors (Hatfield, 1972) , the most likely candidate for suppressor activity in this case would appear to be a CGA-decoding tRNA, which could suppress UGA by G-U mispairing a t the first position of the codon, in analogy to the interactions of glutamine tRNAs with UAA and UAG in yeast (Pure et al., 1985; Weiss and Friedberg, 1986; Lin et al., 1986; W. A. Weiss et al., 1987) and possibly in MuLV. The possibility that as of yet unidentified specialized suppressor tRNA(s) are involved in readthrough suppression at the MuLV gug-pol junction should also be considered. The subject of suppressor tRNAs in readthrough suppression in higher eukaryotes has also been reviewed by Valle (1989 ), Hatfield et al. (1990a , and Valle and Haenni (1991) .

The work cited in this section shows that mammalian cells contain either four or five distinct species (depending on whether the same glutamine tRNA is used in UAG and UAA suppression) that can suppress termination codons at the MuLVgag-pol junction. As of yet none of these tRNAs has been definitively identified or characterized, but this will be important for future studies on the mechanism of readthrough suppression. For example, mutational analysis of tRNA structure as well as mRNA context should provide insights into the nature of the interactions between cis-and trans-acting factors.

Though the present discussion has focused on readthrough suppression in retroviruses, it should be noted that this mechanism is also used by other viruses to modulate the level of synthesis of fusion proteins. Thus, in several alphaviruses, including Sindbis virus, a single UGA codon separates two open reading frames (Strauss et al., 1983 (Strauss et al., , 1984 (Strauss et al., , 1988 . It has been shown that this UGA codon is suppressed in uiuo (Li and Rice, 1989) . A number of plant viruses, including TMV (Pelham, 1978; Goelet et al., 1982) , carnation mottle virus (Guilley et al., 1985) , and beet necrotic yellow vein virus (Bouzoubaa et al, 19861 , and use readthrough of a UAG codon. An elegant analysis by Skuzeski et al. (1991) has shown that in the case of TMV, the signal responsible for readthrough suppression is confined to the two codons immediately 3' of the termination codon. It is intriguing to note that many plant viruses exhibiting readthrough suppression have a nearly identical sequence in this position, whereas others, such as carnation mottle virus, have a different sequence (Valle, 1989) .

In both TMV (Ishikawa et al., 1986) and Sindbis virus (Li and Rice, 19891, as in MuLV (Feng et al., 1989b1 , readthrough occurs with each of the three possible termination codons. Despite this similarity in the different viral systems, sequence comparison shows no obvious homology in the sequences surrounding the suppressible termination codon (Feng et al., 1990b) .

As noted in Section I, ribosomal frameshifting alters the reading frame of mRNA during translation, resulting in the expression of a single protein from two or more overlapping genes. Ribosomal frameshifting may operate in one of two directions, altering the reading frame in either the 5' or 3' direction. This phenomenon is well known in bacteria and has been reviewed elsewhere (Dayhuff et al., 1986; Craigen and Caskey, 1987; R. B. Weiss et al., 1987; R. Weiss et al., 1988; Hughes et al., 1989; Atkins et al., 1990; Murgola, 1990) . In eukaryotes, a shift in the reading frame in the 3' direction has been described thus far only in yeast, whereas that in the 5' direction has been described in yeast, plants, and animals. For example, the retrovirus-like retrotransposon, Ty, and the double-stranded RNA viruslike particle, L-A, both contain two large overlapping reading frames that are aligned differently in yeast (for review see Wickner, 1989 ). In Ty, the different reading frames are aligned by a frameshift of one nucleotide in the 3' (or + l ) direction Clare et al., 19881 , whereas in L-A they are aligned by a frameshift of one nucleotide in the 5' (or -1) direction (Icho and Wickner, 1989; Dinman et al., 1991) . In higher eukaryotes, ribosomal frameshifting occurs or is suspected of occurring in the -1 direction in a number of mammalian and avian retroviruses, in the avian infektious bronchitis virus (IBV) (Brierley et al., 1987, 19891 , in certain pladt viruses (Miller et al., 1988; Xiong and Lommel, 19891 , in transposable elements in Drosophila, and in the mouse intracisternal A-particle (mouse IAP) (for reviews see Jacks, 1990; Hatfield et al., 1990a,b; Hatfield and Oroszlan, 1990) . In fact, the gag and pol genes of most vertebrate retroviruses occur in different reading frames and ribosomal frameshifting in the -1 direction is required to align the overlapping frames. Interestingly, as noted above, some of these retroviruses require two frameshift events, one between gag-pro and one between pro-pol, to express the Gag-Pro-Pol fusion protein (see Fig. 1 ).

The means of unequivocally demonstrating ribosomal frameshifting is to sequence the transframe protein (i.e., the protein that spans the overlapping reading frame) through the frameshift site and compare the resulting peptide to the corresponding RNA (template) sequence (Hizi et al., 1987; Jacks et al., 1988a,b; Weiss et al., 1989; Nam et al., 1992) . The fact that viral genes may be tandem, lie in different reading frames, and appear to be overlapping does not necessarily mean that they are expressed by ribosomal frameshifting, even though the gene organization may be analogous to that of other genetic systems utilizing frameshifting, and, for that matter, even though such genes are expressed as a fusion protein. For example, the cauliflower mosaic virus capsid protein and RT genes are tandem and lie in different reading frames, but RT is expressed separately from the capsid protein (see Schultze et al., 1990; Wurch et al., 1991, and references therein) . In addition, in the hepatitis B virus, the X and C genes are expressed as a fusion protein (where these genes may occur in different reading frames), but recent evidence suggests that this fusion protein is not synthesized by ribosomal frameshifting (see Lo et al., 1990, and references therein) .

In the present review, we have included those genetic systems in which ribosomal frameshifting has unequivocally been shown to occur, or is suspected of occurring based on the presence of a n established frameshift signal within the overlapping region (see Table I ). We examine ribosomal frameshifting, in both the -1 and + 1 directions, in detail below. For comparison, several other, nonretroviral systems will also be briefly considered.

Frameshifting in the -1 direction in eukaryotes was first demonstrated by Jacks and Varmus (19851, who reported that both the Gag protein and the Gag-Pol fusion protein of Rous sarcoma virus (RSV) could be synthesized in rabbit reticulocyte lysates programmed with a single species of RNA encoding the RSV gag gene and an adjacent portion of the downstream pol gene. The fact that both polypeptides were formed from a single species of RNA in approximately the same ratios as found in uiuo provided strong evidence that ribosomal frameshifting accounted for the alignment of the different reading frames in RSV. Ribosomal frameshifting was unequivocally demonstrated when the in uiuo-made transframe protein spanning the gag-pro overlap of MMTV was sequenced and found to contain amino acid residues that matched the corresponding nucleotide template, except for a shift by one nucleotide in the -1 direction (Hizi et al., 1987) . A detailed examination of the frameshift site, of information encoded in viral RNA for frameshifting, of possible models for frameshifting, of unique features of the frameshift site, and of the possible role of tRNA in frameshifting are presented below. (Hizi etal., 1987; Jacks et al., 1987; Moore et al., 1987) ; BLV (Rice etal., 1985; Sagata etal., 1985) ; HTLV-1 (Seiki et al., 1983; Hiramatsu et al., 1987; Inoue et al., 1986) ; STLV-1 (Inoue et al., 1986) ; HTLV-2 (Mador et al., 1989; Shimotohno et al., 1985) ; EIAV (Stephens et al., 1986; Kawakami et al., 1987) ; MHV (Lee et al., 1991) ; IBV (Brierley et al., 1987) ; BEV (Snijder et al., 1990) ; SRV-1 (Power et al., 1986) ; SRV-2 (Thayer et al., 1987) ; MPMV (Sonigo et al., 1986) ; Visna (Sonigoet al., 1985) ; mouse IAP (Mietz et al., 1987) ; BYDV (Miller et al., 1988) ; 17.6 (a transposable element in Drosophila) (Saigo et al., 1984) ; RCNMV (Xiong and Lommel, 1989) ; HIV-1 (Jacks et al., 1988b; Ratner et al., 1985; Wain-Hobson et al., 1985; Sanchez-Pescador et al., 1985) ; HIV-2 ; SIV (Franchini et al., 1987; Chakrabarti et al., 1987) ; gypsy (a transposable element in Drosophila) (Marlor et al., 1986) ; RSV (Hughes et al., 1989; Jacks et al., 1988a; Schwartz et al., 1983) ; L-A (a double-stranded RNA viruslike particle in yeast) (Dinman et al., 1991) .

Underlined bases designate heptanucleotide sequences within the overlaps that are associated or are suspected of being associated with frameshifting (see text and Jacks et al., 1988a; Jacks, 1990; Hatfield et al., 1990a,b) . The bold letter at the 3' end of the frameshift signal designates that the precise site of the frameshift has been established by sequencing the transframe peptide of one member of the subclass (see text and footnote c).

Frameshift signals are placed into classes on the basis of the consensus sequence (i.e., AAAC, UUUU, and UUUA) and into subclasses on the basis of the upstream triplet such that members of subclasses have identical heptanucleotide signals. The fact that the transframe peptide has been sequenced from one member of the subclass, establishing the precise site ofthe frameshift (see text), demonstrates that the precise site for each member is known.

The boundaries of the overlapping region (also called the overlap or frameshift window) are set by the termination codon in the 0 frame (e.g., the termination codon a t the end ofgag) and the nearest upstream termination codon in the -1 frame. The frameshift must occur of course within the overlapping region. The size of the overlap can vary from seven nucleotides [as observed in the frameshift window of red clover necrotic mosaic virus (RCNMV)] (Xiong and Lommel, 1989 ; see below) to several hundred nucleotides in length (as observed in the pro-pol frameshift window of HTLV-2 (Shimotohno et al., 1985) . The site of the frameshift may occur anywhere within the overlap from the extreme 3' end to the 5' end.

A search of the overlapping regions within the gag-pol genes of retroviruses and the equivalent regions in other viruses and retroelements of higher organisms, including those in Drosophila and in the mouse IAP, each of which requires -1 frameshifting for alignment of different reading frames, reveals the occurrence of one of three common consensus sequences. As shown in Table I , the sequences are A AAC, U UUA, or U UUU, where AAC, UUA, and UUU decode asparagine, leucine, and phenylalanine, respectively, in the 0 frame (see also Jacks et al., 1988a) . It is of interest to note that IBV (Brierley et al., 1987) and two plant viruses, barley yellow dwarf virus (BYDV) (Miller et al., 1988) and RCNMV (Xiong and Lommel, 19891 , each contain one of the common sequences within the respective overlap shared by two different reading frames. As will be discussed in a later section, the signal for frameshifting that encompasses the common consensus sequence is actually a heptanucleotide, as shown in Table I. The frameshift site is the 3' base a t the end of the consensus sequence. This was most clearly demonstrated by sequencing the transframe peptide generated i n uitro from the RSV sequence, UUAUA (Jacks et al., 1988a1 , where UUA is a leucine codon that is read inthe gag frame and is part of the u UUA consensus sequence, and the pro-pol sequence of HTLV-1, AACCA (Nam et al., 19911, where AAC is an asparagine codon that is readin the pro frame and is part of the A AAC consensus sequence (see Table I ). As shown in Fig. 5 , leucine and isoleucine were generated from the UUAUA RSV sequence, and thus the UUA codon was read as leucine in thegag (or 0 ) frame and the AUA codon was read as isoleucine in the pol (or -1) frame. Likewise, &paragine and proline were generated from the HTLV-1 pro-pol AACCA sequence and thus AAC was decoded as asparagine in the pro frame, and -CCA was decoded as proline in the pol frame. These studies show Fic. 5. Nucleotide and amino acid sequences of the transframe proteins a t the frameshift site in selected retroviruses. Underlined nucleotides designate the frameshift signals. The amino acid sequences above the nucleotide sequences are translated from the 0 reading frame and those below are translated from the ~ 1 reading frame. The arrows point to the frameshift site; u is the site ofthe mutation in Rous sarcoma virus IRSV) (see text).

unequivocally that the site of the frameshift is the 3' end of the consensus sequence. It should be noted that frameshifting has not yet been shown to occur a t the 3' end of the U UUU common sequence in any naturally occurring (wild-type) overlap. However, it seems likely that the 3' U in U UlJU is a frameshift site, because changing the RSV consensus sequence, U UUL, to U U U F promotes frameshifting and incorporates phenylalanine into the transframe peptide in uitro ( Jacks et al., 1988a) .

The transframe protein had been sequenced from other retroviruses in studies performed earlier than those described above. For sequencing, the MMTV Gag-Pro transframe protein was purified from virus (Hizi et al., 1987) and the human immunodeficiency virus (HIV) peptide was synthesized in vitro from the gag-pol construct (Jacks et al., 1988b) . In MMTV, the sequence AAA --AAC UUG UAA occurs at the 3' end of gag where A AAC is the consensus sequence (Table I and Fig. 5 ). The transframe protein generated from this sequence contained lysine and asparagine (decoded by AAA and AAC, respectively, in the 0 frame), followed by leucine (decoded by either UUG in the 0 frame or -CUU in the -1 frame) and then valine (decoded by GUA in the -1 frame) (Hizi et al., 1987) , but due to overlapping Leu codons in the 0 and -1 frames, the precise site of the frameshift could not be determined. However, by analogy to the results described above with HTLV-1, it seems likely that the frameshift occurred at the 3' end of the AAC codon within the MMTV gag-pro overlap as shown in Fig. 5 .

In sequencing the transframe peptide generated in vitro from HIV-1, Jacks et al. (198813) observed that the shift occurred at a leucine residue corresponding to the Leu codon, UUA, which is part of the consensus sequence U UUA (Table I and Fig. 5) . However, both leucine and phenylalanine were present a t the position of the frameshift (in a ratio of 791, making assignment of the precise site uncertain. It is of interest to note that Weiss et al. (1989) translated the HIV frameshift signal in Escherichia coli cells, sequenced the resulting transframe peptide, and observed both leucine and phenylalanine at the frameshift site in a ratio of about 3:l. A similar analysis of the transframe peptide generated from the MMTV gag-pro frameshift signal, A AAA AAC, in E . coli yielded asparagine and lysine a t the frameshift site in a ratio of about 7:3 (Weiss et al., 1989) . These investigators proposed that a major shift occurred at the normal frameshift site Le., on the UUA codon in HIV and on the AAC codon in MMTV), whereas a minor shift occurred a t the immediate upstream codon (i.e., on the UUU codon in HIV-1 and on the AAA codon in MMTV) (see Fig. 5 ). Kingsman et al. (1990) have also considered the possibility that the HIV-1 U UUU UUA frameshift sequence is quite slippery (Wilson et al., 19881 , such that a minor shift that occurs at the upstream UUU codon would account for the occurrence of two amino acids at the frameshift site. Other possibilities that may account for the occurrence of two amino acids at the HIV frameshift site as observed by Jacks et al. (1988b) also warrant consideration. For example, Jacks et al. (1988b) proposed that a portion of the Leu-tRNA that is decoded by UUA a t the ribosomal A-site in the 0 frame may come off the ribosome after the slip to the -1 reading frame. This --event would expose the U U U codon, which then would be decoded by Phe-tRNA. Another possibility is that the frameshift site, which most certainly has unique features that make it slippery, may be more favorable to misreading such that Phe-tRNA misreads the leucine codon in HIV-1 and Lys-tRNA misreads the asparagine codon in MMTV prior to the frameshift (see further discussion on misreading within the frameshift signal in Section III,A,5).

The studies discussed above were carried out in heterologous systems (Jacks et al., 1988b; Weiss et al., 1989) . However, it is important to know whether two amino acids occur at the frameshift site in the naturally occurring transframe proteins (i.e., in the Gag-Pol fusion protein of HIV-1 and the Gag-Pro fusion protein of MMTV) synthesized under normal physiological conditions in virus-producing cells. A transframe peptide derived from the naturally occurring HIV Gag-Pol fusion protein was sequenced (eight amino acids in length) and was found to contain Phe-Leu-Arg (L. Henderson, personal communication 19911, where Phe-Leu is read in the gag frame (decoded by U U U U U A ) and Arg is read in the pol frame (decoded by AGG, where A is the siteof the frameshift) (see Fig. 5 ). However, this study did not fully exclude the possibility that a second transframe peptide with the sequence Phe-Phe-Arg may also exist in HIV-1. Similarly, amino acid sequencing of the naturally occurring transframe protein of MMTV did not indicate microheterogeneity (Hizi et aZ., 19871 , but the possibility cannot be excluded here either. Therefore, additional studies will be required to establish the very important point of whether amino acid sequence heterogeneity exists at the frameshift site in the naturally occurring transframe proteins or peptides.

As discussed above, the sequence of various transframe peptides showed that the frameshift occurred at the 3' end of the consensus sequences, as shown in Table I . Although these studies pinpoint the exact site of the frameshift, they do not determine how the different reading frames are aligned, i.e., whether alignment occurs by overlapping reading (where the base at the 3' end of the consensus sequence would be read twice) or by doublet decoding (where only two of three bases of the 0 frame codon within the consensus sequence would be read). However, an experiment by Jacks et al. (1988a) in which a single base change at the 3' end of the frameshift signal results in two new amino acids in the transframe peptide demonstrates that the alignment occurs by overlapping reading. Jacks et al. (1988a) changed the RSV U U A U A sequence, which codes for leucine in the 0 frame and isoleu-cine in the -1 frame, to UUUUA, where U is the altered base a t the 3' end of the frameshift signal (see Fig. 5 ,. The resulting transframe peptide generated from this sequence contained phenylalanine, which was decoded by UUU, and leucine, which was decoded by UUA (see Fig.  5 ). The base at the 3' end of the frameshift signal is therefore read twice, once in the 0 frame and once in the -1 frame; thus, the alignment of the different reading frames occurs by overlapping reading as originally proposed by Hizi et al. (1987;  see also Hatfield and Oroszlan, 1990) .

Two different kinds of information have been identified in viral RNA that have a role in signaling the frameshift event: (1) as noted above, a heptanucleotide sequence that encompasses the frameshift site on its 3' end and the immediate six upstream bases (Table I ) ( Jacks et al., 1988a; Wilson et al., 1988; Brierley et al., 1989; Dinman et al., 1991; Nam et al., 19921, and (2) RNA secondary structure, which occurs just downstream of the heptanucleotide sequence (Jacks et al., 1987 (Jacks et al., , 1988a Brierley et al., 1989; Weiss et al., 1989; Dinman etal., 1991) . Both types of information, which have been termed cis-acting sequences (for review see Jacks, 19901 , are further examined below.

a. Information at the Frameshift Site. The heptanucleotide sequence that signals the frameshift event (see Table I ) was identified largely by site-directed mutagenesis studies within and/or around the frameshift region (Jacks et al., 1988a,b; Nam et al., 1988; Wilson et al., 1988; Brierley et al., 1989; Dinman et al., 1991; Nam et al., 1992) . In RSV, alteration of the -~ U UU bases within the consensus --U UUA sequence (where UUA is read in the 0 frame and A is the site of the frameshift) inhibitedibosomal frameshifting most severely ( Jacks et al., 1988a) . Alteration of any of the three bases immediately upstream of the RSV consensus sequence (i.e., A AA within A AAU --UUA) reduced the level of frameshifting, but only moderately compared to that observed with the --U UU bases.

A similar observation was made in the double-stranded yeast virus, L-A, in which changes in the first three bases of the heptanucleotide frameshift signal (i.e., in G GG of the G GGU UUA frameshift signal) had a smaller inhibitory effect on frameshifting than those occurring in the downstream U UU sequence (Dinman et al., 1991) . Alteration of the first triplet to any identical three bases (i.e., G GG to C CC, A AA, or U UU) within the L-A frameshift signal maintained efficient frameshifting. The changes to pyrimidines in the first triplet resulted in higher levels of frameshifting than the corresponding purines, with U UU giving the highest level. The latter observations suggest that the homopolymeric U sequences may provide an extremely slippery signal (see also Weiss et al., 1989; Kingsman et al., 1990; and Section III,A,3,b) .

Frameshift signals in the HTLV-1 gag-pro and pro-pol overlaps have also been identified by mutagenesis studies. Alteration of the gag-pro A AAA AAC sequence to A AUA UUC inhibited frameshifting (Nam et al., 19881 , as did alteration of the pro-pol U UUA AAC sequence to U UUA AGC (Nam et al., 1991) . Deletion of the U triplet in the pro-pol frameshift signal also inhibited frameshifting, but changing the sequence immediately downstream of the AAC codon from CAGAA to UGCAG did not affect the frameshift event (Nam et al., 1991) .

Mutation of the 3' terminal A in the sequence U UUU UUA, where -A is at the frameshift site, to-any of the other three basesdid not inhibit the level of frameshifting in RSV or HIV (Jacks et al., 1988a; Wilson et al., 1988) . In L-A, changing the 3' terminal A in UUA to U or C likewise did not inhibit frameshifting, but changing this s i c to G did reduce the level of frameshifting by 5to 10-fold (Dinman et al., 1991) . The frameshift event was enhanced slightly by changing this A to U (note that this change results in a new consensus sequence, U UUU) in all three viruses examined ( Jacks et al., 1988a; Wilson et al., 1988; Weiss et al., 1989; Dinman et al., 1991) . The fact that altering the U UUA sequence in L-A to A AAC maintained wild-type frameshift levels demonstrates that the same set of consensus signals that have been observed to be associated with -1 frameshifting in higher eukaryotes (i.e., U UUA, U UUU, and A AAC) also can work efficiently in yeast (Dinman et al., 1991) .

In contrast to the results described above in which mutants at the 3' end of the homopolymeric U sequence had only moderate effects on the level of ribosomal frameshifting in mammalian protein synthesis, changing the base at the 3' end of the homopolymeric A frameshift signal had far more pronounced effects (Chamorro et al., 1992) . Alteration of the C in the MMTV gag-pro frameshift (A AAA AAC) signal to -U reduced the level of frameshifting severalfold, whereas changing this base to A or G was even more inhibitory, with the A AAA AAG sequence exhibiting the most severe inhibition. In contrast, changing the 3' terminal C in the MMTV gag-pro (homopolymeric A) frameshift signal to A or G increased, and did not decrease, the level of frameshifting in E . coli (Weiss et al., 1989) . The level of frameshifting increased about 1.5 times with a 3' terminal A and about 30 times with a 3' terminal -G.

In mammalian cells, AAC and AAU are decoded by the same isoacceptor and in E. coli the same tRNA decodes AAA and AAG. Presumably, in each case the two codons are normally translated with roughly equal efficiencies by the corresponding cognate isoacceptors. It is therefore of considerable interest that the levels of frameshifting are quite different with AAC compared to AAU in mammalian cells, as are the levels in E. coli with AAG compared with AAA. The implications of these observations are further considered in Section 111, A,5.

One major point to be emphasized in the present discussion is that the frameshift signal that encompasses the frameshift site is a heptanucleotide sequence, which was identified largely by mutagenesis studies (Jacks et al.. 1988a,b; Wilson et al., 1988; Brierley et al., 1989; Dinman et al., 1991) . This conclusion is further supported by the observation that efficient frameshifting occurs when the heptanucleotide frameshift signal in IBV is altered such that it is flanked on both sides by a termination codon [i.e., when the codon immediately upstream in the -1 frame and that immediately downstream in the 0 frame of the heptanucleotide signal are stop codons (S. Inglis, personal communication 1989) l. Furthermore, the frameshift window in RCNMV is only seven nucleotides in length (Xiong and Lommel, 1989) .

b. Information Downstream of the Frameshift Site. Not all of the information necessary for frameshifting is located within the heptanucleotide sequence shown in Table I . Initially, Jacks et al. (1987) demonstrated that the ability of the gag-pro and pro-pol overlaps of MMTV to promote frameshifting was lost when these regions were inserted into another genetic context. Thus, these two overlaps (which contain the heptanucleotide frameshift signal) are not sufficient in themselves to carry out frameshifting. Jacks et al. (1987) also suggested that potential stem-loop structures that occur just downstream of the overlaps in MMTV may provide the additional information required for the frameshift event (see above). The presence of a potentially stable RNA secondary structure just downstream of homopolymeric A sequences in the gag-pro overlaps of bovine leukemia virus (BLV) and HTLV-1 was first noted by Rice et al. (1985) . Potential stem-loop structures occur just downstream of the frameshift sites or suspected sites in each overlapping region sequenced to date from higher eukaryotes (actually within 9 to 10 bases of the frameshift site in all cases; see Jacks, 1990; Kingsman et al., 1990) . Interestingly, in a survey of several hundred bases on either side of the frameshift site, Le et al. (1989) found that the most thermodynamically stable secondary structure was always a stem-loop located within several bases downstream of the frameshift site. Although the stem-loop structures vary considerably in size of both the stem and the loop, the type of configuration that can be generated, in base composition, and in stability, the striking conservation of their position suggests that they may have an important role in frameshifting. Several examples of possible stem-loop structures are shown in Fig. 6 .

Efficient frameshifting has been shown to be dependent on the presence of a stem-loop structure that occurs just downstream of the frameshift site in RSV (Jacks et al., 1988a) , in IBV (Brierley et al., 1989) , in L-A (Dinman eta1.,1991) , and in thepro-pol overlap ofHTLV-1 (Nam et aZ., 1991) . Jacks etal. (1988a) used deletion-substitution and site-specific mutations to show a direct relationship between the presence of a stem-loop structure and the efficiency of frameshifting in RSV. Deletion of bases within the stem-loop structure virtually abolished frameshifting. Disruption of base pairings within the stem by generating specific stem-destabilizing mutations resulted in a decrease in frameshifting, whereas restoring these base pairings by generating specific stem-restabilizing mutations rescued frameshifting ( Jacks et al., 1988a) .

Site-directed alteration of a number of specific bases further downstream of the stem-loop structure in IBV inhibited frameshifting (Brierley et al., 1989) . The latter bases are complementary to those in the loop of the stem-loop, which raises the possibility that many of the downstream bases interact with the loop in IBV, resulting in a tertiary structure known as a pseudoknot. The role of the stem-loop structure and pseudoknot in IBV frameshifting has been examined in further detail (Brierley et al., 1991) . For efficient frameshifting in IBV, the two pseudoknot stems must be in close proximity to each other and must be essentially intact. However, small changes in the loops of the pseudoknot did not affect frameshifting. These investigators also observed that the pseudoknot could not be replaced by a simple stem-loop structure of similar overall size and composition; thus the pseudoknot conformation is a requirement for frameshifting.

In the double-stranded RNA viruslike particle, L-A, evidence has been presented that the stem-loop structure that is immediately downstream of the frameshift site exists as part of a pseudoknot, and the entire pseudoknot structure is required for efficient frameshifting (Dinman et al., 1991) . Furthermore, the stem-loop structures that occur immediately downstream of the frameshift signals or proposed frameshift signals in a number of retroviruses (ten Dam et d., 19901, as well as in murine coronavirus gene 1 (Lee et al., 19911 , are capable of forming a pseudoknot. In MMTV, the presence of a pseudoknot downstream of the gag-pro frameshift signal is required for optimal frame- shifting (Chamorro et al., 1992) . The occurrence and role of pseudoknots in retroviruses and in other RNA structures have been reviewed elsewhere (Schimmel, 1989; Wyatt et at., 1989; Pleij, 1990; ten Dam et al., 1990) . It is of interest t o note that the position of the stem-loop structure relative to the frameshift site is critical for efficient frameshifting, as has been carefully demonstrated for IBV (Brierley et al., 1989) . Altering the distance between the stem-loop structure and the frameshift site by as few as three bases in either direction inhibits frameshifting in IBV. Thus, it appears that the ribosomal frameshift event, at least in some retroviruses, requires a carefully positioned downstream stemloop structure that may, as in the case of IBV (Brierley et al., 1989) , exist as part of a pseudoknot.

The role of the downstream RNA secondary and/or tertiary structures may be to impede translation a t the frameshift site long enough for the shift to occur (Rice et al., 1985; Jacks et al., 1987; Brierley et al., 1989; Weiss et al., 1989; Atkins et al., 1990; Jacks, 1990; Kingsman et al., 1990 ). As noted above, the stability of the stem-loop structures, their sizes, and the type of configurations that can be generated from them vary considerably. It remains to be determined how the variation in the stem-loop structure influences the efficiency of frameshifting.

The initial experiments involving the role of the potential stem-loop structure in the frameshift event in HIV suggested that this structure was not required for efficient frameshifting. For example, Madhani et al. (1988) tested a variety of mutations encompassing the potential stem-loop downstream of the HIV frameshift site, and these mutations (with one exception) had no effect on frameshifting. Wilson et al. (1988) focused on the ability of a short oligonucleotide encoding the heptanucleotide U UUU UUA frameshift signal in HIV to carry out frameshifting. They observed that this sequence worked as efficiently in vitro (in rabbit reticulocyte lysates) as in vivo (in yeast cells) whether or not the downstream stem-loop structure was present. However, it should be noted that these experiments were not carried out under normal physiological conditions for HIV gag-pol expression. Recent in vivo studies (carried out in mammalian cells) provide evidence that the downstream stem-loop structure in HIV is required for optimal frameshifting (H. Varmus, personal communication 1991) , demonstrating that this structure indeed has an important role for efficient synthesis of the HIV Gag-Pol fusion protein.

It should also be noted that the experiments which show that HIV frameshifts equally well with and without a downstream stem-loop structure (Jacks et al., 1988b; Wilson et al., 1988) suggest that this frameshift signal is more slippery than those examined in other genetic systems in which the role of the downstream stem-loop has been shown (under similar assay conditions) to be required for efficient frameshifting (Jacks et al., 1987; Jacks et al., 1988a; Brierley et al., 1989) . These observations led Kingsman et at. (1990) to speculate that the homopolymeric A and U frameshift sequences (i.e., A AAA AA and U UUU UU) may be more slippery than other frameshift signals, and therefore that their requirements for a downstream stem-loop structure to aid the frameshift event may be less stringent. However, in eukaryotes the requirements for efficient frameshifting with a perfect homopolymeric A (i.e., all of the bases are As) sequence apparently are different from that of a perfect homopolymeric U sequence. Mutation of the 3' terminal C in the A AAA AAC frameshift signal to A (or to G) severely inhibits frameshifting (Chamorro et al., 1992) . In addition, Dinman et al. (1991) have observed that frameshifting in yeast was more efficient when the L-A frameshift signal contained homopolymeric pyrimidine sequences than when it contained homopolymeric purine sequences. Interestingly, in the study by Dinman et al. (19911, the most efficient frameshift signal contained six tandem U bases.

Translation of the MMTV gag-pro and HIV frameshift signals with and without the corresponding downstream stem-loop structure in E. coli cells shows that the presence of this structure has only a slight to moderate effect on enhancing frameshifting in this heterologous system (Weiss et al., 1989) . The transframe peptide generated from the MMTV gag-pro frameshift signal (A AAA AAC) in E. coli with and without the downstream stem-loop structure was sequenced (Weiss et al., 1989) . Interestingly, asparagine occurred predominantly a t the frameshift site with the intact stem (in an approximate ratio of 3:l with lysine), whereas in the absence of the stem-loop, lysine occurred predominantly a t the frameshift site (in an approximate ratio of 2:l with asparagine). One interpretation of these results is that the stem specifically enhances frameshifting on the AAC codon, with a minor frameshift occurring on the upstream AAA codon, whereas in the absence of the stem, the major shift occurs on the first slippery codon, AAA, with a minor shift on the downstream AAC codon (Weiss et al., 1989) . This observation is further considered below (see Section III,A,5).

During the frameshift event the aminoacyl-tRNA, which is located a t the ribosomal A-site, and the peptidyl-tRNA, which is located a t the ribosomal P-site, are translocated by one nucleotide in the 5' direction. It is not entirely clear how frameshifting is accomplished, but Jacks et al. (1988a) have proposed that it occurs by simultaneous slippage of both the aminoacyl-tRNA and the peptidyl-tRNA by one nucleotide in the -1 direction, resulting in both tRNAs decoding a new set of codons (see Fig. 7A ). Following slippage, the ribosome is prepared to read the -1 frame; normal transfer of the peptidyl-tRNA to the aminoacyl-tRNA and its translocation to the P-site bring the first codon in the -1 frame to the A-site. Then, normal decoding of the A-site and transfer of the nascent peptide to the incoming aminoacyl-tRNA consummate reading in the -1 frame (Fig. 7A) .

Site-directed mutagenesis studies that show that all seven bases within the heptanucleotide frameshift signal are essential to efficient frameshifting (Jacks et al., 1988a; Wilson et al., 1988; Brierley et al., 1989; Dinman et al., 1991) and sequence analyses of the transframe peptide (Hizi et al., 1987; Jacks et al., 1988a,b; Weiss et al., 1989; Nam et al., 1991 1 that show that the shift to the -1 reading frame occurs a t the ribosomal A-site ( Jacks, 1990) provide support for the simultaneousslippage model. These studies strongly suggest that the six bases within the 0 frame of the signal span both the ribosomal A-and P-sites and the six bases within the -1 frame (following the slippage) also span the A-and P-sites. After the slip to the -1 frame has taken place, the aminoacyl-tRNA and peptidyl-tRNA decode a new set of codons; the base sequence within the frameshift signal is such that a minimal amount of mismatching occurs (see Fig. 8 and Section III,A,5). Weiss et al. (1989) have proposed a somewhat different simultaneousslippage model for frameshifting based on their observations on translation of the retroviral homopolymeric A and U frameshift signals in E . coli. The major difference between the Jacks and Varmus model (for review see Jacks, 1990 ) and that of Weiss et al. is that the latter takes into account the probable presence of three sites on the ribosome, the aminoacyl-tRNA (A), the peptidyl (PI, and the exit (El sites, and the possibility that the shift occurs after peptide bond formation (see Fig.  7B ). The Jacks and Varmus model proposes that the frameshift occurs before peptide bond formation while the ribosome is stationary. Weiss et al. (1989) suggested that the downstream stem-loop structure may exert its influence during translocation of the 3' codon (within the frameshift signal) from the A-to P-site and the slip may occur while the tRNAs exist, albeit transiently, as hybrids in the E-IPand P-/A-sites immediately after peptide bond formation (see Fig. 7B and the legend for details).

-Simultaneous-slippage models for ribosomal frameshifting in the -1 direction. (A) The simultaneous-slippage model (Jacks et al., 1988a; Jacks, 1990 ) (see text for details). (B) The simultaneous-slippage model of Weiss et al. (1989) . This model the occurrence of three ribosomal frameshift sites (A, P, and E) and suggests that the shift occurs after peptide bond formation Interestingly, the tRNA may exist (albeit transiently) as a hybrid occupying simultaneously the EIP-and P-/A-sites. This model that the growing polypeptide remains stationary at the P-site (see Weiss et al., 1989, for details and additional references) . Table I ). The signals are arranged into four classes (columns A-D) depending on the codon-anticodon interaction after the frameshift event as follows: Shift from the 0 to the -1 frame results in misreading, or reading two out of three bases in both the ribosomal A-and P-sites (A), just the A-site (B), orjust the P-site (C); in column D, a shift to the -1 frame results in reading only one base in the P-site and two bases in the A-site by the standard Watson-Crick base pairings. Squiggly lines signify the nascent polypeptides attached to tRNAs in the P-sites; AA represents the amino acid attached to tRNA in the A-site, and the dashed lines represent mismatching in codon-anticodon interactions between standard Watson-Crick base pairs.

Unique Features of the Frameshift Site Jacks et al. (1988a) noted that only A AAC, U UUU, and U UUA occur within the frameshift signals of retroviruses and other genetic elements in eukaryotes (see Table I ), in which AAC, UUU, and UUA are decoded at the ribosomal A-site in the 0 frame. The observation that only three codons occurred at the ribosomal A-site and the finding that altering the consensus sequence within the frameshift signal to A AAA or G GGG inhibits frameshifting led these investigators to propose the existence of specialized "shifty" tRNAs that promote frameshifting (Jacks et al., 1988a) . These tRNAs and the corresponding codons are tRNAASn (AAC), tRNAPhe (UUU), and tRNALe" (UUA). These tRNAs are characterized by the fact that tRNAAsn contains the highly modified queuine (Q) base in the 5' position of its anticodon, tRNA contains the highly modified wyebutoxine (Wye) base in the 3' position next to its anticodon (see Hatfield et al., 1990b , and references therein), and tRNAL"" lacks a highly modified base in its anticodon loop (Valle et al., 1987) .

The status of the tRNAs utilized in and around the frameshift signals in cells infected with HIV-1, BLV, HTLV-1, and simian retrovirus-1 (SRV-1) has been examined (Hatfield et al., 1989, and unpublished observations) . Interestingly, most of the Phe-tRNA from HIV-and SRV-1-infected cells lacked the highly modified Wye base in its anticodon loop, and most of the Asn-tRNA from BLV-, HTLV-I-, and SRV-1infected cells lacked the highly modified Q base in its anticodon loop. Thus, a correlation exists between the occurrence of hypomodified Asn-tRNA and Phe-tRNA in retrovirus-infected cells and their utilization in translating codons within the respective frameshift signals (see Table I ).

The appearance of hypomodified isoacceptors in retrovirus-infected cells, most certainly, is not a virally encoded phenomenon. Thus, a question may be raised as to the cause of hypomodification of specific isoacceptors in the host tRNA population after viral infection. This question has been addressed previously by Katze and collaborators (19831, who considered a number of possibilities to explain a deprivation of Q base in tRNA in tumor cells. I t should be noted that Q base is obtained in the diet of mammals and is inserted in tRNA by an enzyme designated as queuine tRNA-ribosyltransferase (tRNA-GRT). These investigators suggested that a deficiency of Q base in tumor cell tRNA occurs in part because the requirements for this base exceed the dietary intake. This may be due to an increase in tRNA turnover and growth rate, to an inefficient salvage pathway, andlor to the possible occurrence of inhibitors of tRNA-GRT. Of these possibilities, it is tempting to speculate that in retrovirus-infected cells, the metabolism of the host may be altered in response to viral infection such that the host produces a new metabolite or more of a given metabolite, which acts as an inhibitor of an enzyme involved in production of the hypermodified base within tRNA. For example, 7-methylguanine and pteridine occur in mammalian cells and these metabolites are inhibitors of tRNA-GRT (see Katze et al., 1983; French et al., 1991, and references therein) .

What is the possible role of hypomodified isoacceptors in ribosomal frameshifting? Clearly, the lack of a hypermodified base in the anticodon loop of tRNA would create more space in and around the frameshift site; in turn, this might facilitate frameshifting by allowing greater flexibility of movement of the anticodon Hatfield and Oroszlan, 1990) . It is of interest to note that the presence of modified bases within the anticodon loop of tRNA restricts wobble, whereas their absence expands the decoding potential (Randerath et al., 1979; Bienz and Kubli, 1981; Beier et al., 1984a,b; Bjork et al., 1987 Bjork et al., , 1989 Wilson and Roe, 1989; Claesson et al., 1990) . More specifically, among these studies it has been shown that some tRNAs lacking a modified base in their anticodon promote frameshifting (Bjork et al., 1989) , whereas others promote misreading (Randerath et al., 1979; Bienz and Kubli, 1981; Beier et al., 1984a,b; Bjork et al., 1987; Wilson and Roe, 1989; Claesson et al., 1990) . In addition, with respect to tRNAs normally containing a Q or Wye base, the coding properties of tRNAs lacking Q base (Bienz and Kubli, 1981; Beier et al., 1984a,b; Bjork et al., 1987; Meier et al., 1985) and Wye base (Smith and Hatfield, 1986 ) differ from those of the corresponding fully modified isoacceptors. In light of these studies, it is tempting to speculate that the "shifty" tRNAs that promote frameshifting are hypomodified isoacceptors.

In the simultaneous-slippage model of frameshifting, the tRNAs involved must have a dual function. First, they must promote frameshifting, and then after the shift to the -1 frame, they must misread or read only two out of three bases of the new set of codons (see Fig. 8 ). Interestingly, the frameshift signals are designed to minimize misreading after the frameshift event. Within the heptanucleotide frameshift signals involved in a shift to the -1 reading frame (see Table I ), the bases in the first two positions of the downstream codon k e . , UU, AA, or GG) are identical to the base in the 3' position of the upstream codon (i.e., U, A, or G, respectively). The shift to the new reading frame, therefore, maintains similar codon-anticodon interactions provided the tRNAs in the ribosomal A-and P-sites misread the base in the 3' position of the -1 frame codon or read only two out of three bases. Codon-anticodon complexes within the various frameshift signals in retroviruses sequenced to date are summarized in Fig. 8 . The only exceptions to the presence of identical bases in the first two positions of the downstream codon and the terminal position of the upstream codon in the frameshift signals shown in Table I are in that of RCNMV and the pro-pol signal of MMTV. These two heptanucleotide signals contain an Asp codon, GAU, and a shift to the -1 reading frame results in Asp tRNA decoding a glycine codon, GGA. The shift onto the GGA codon requires that mismatching takes place between the first and second positions of the Asp tRNA anticodon and the middle and third positions of the corresponding codon as shown in Fig. 8 D. The frameshift site manifests several unique features. For example, only three naturally occurring codons (UUA, UUU, and AAC) have been found to occupy this site (Jacks et al., 1988a; Jacks, 1990) . Moreover, the site and the heptanucleotide sequence that encompasses the frameshift site constitute a slippery region such that the reading frame of the corresponding mRNA may be altered (Jacks et al., 1988a; Wilson et al., 1988; Weiss et at., 1989; Jacks, 1990; Kingsman et al., 1990 ). An additional feature of the frameshift site, in contrast to what occurs in normal translation, is that the same isoacceptor may decode one cognate codon more efficiently than another at this site. That is, tRNAPhe decodes UUU slightly more efficiently than UUC (Jacks et al., 1988a; Wilson et al., 1988; Weiss et al., 1989; Dinman et al., 1991) and tRNAAsn decodes AAC several times more efficiently than AAU (Chamorro et al., 1992) . InE. coli the same tRNALys decodes AAG a t the mutant MMTV gag-pol frameshift site about 20 times more efficiently than AAA (Weis et al., 1989) . Clearly, the frameshift site involves a form of misreading because the same isoacceptor presumably decodes these cognate codons with similar efficiencies at other mRNA sites. Meier et al. (1985) have shown that in the absence of Q base (G is in the wobble position of the anticodon), tRNAHi" shows a strong preference for CAC codons, whereas tRNAHIs with Q base shows a slight preference for CAU codons. This study provides a model for Q-deficient tRNAs preferentially reading XAC codons, as is found in decoding the asparagine AACIAAU codon set a t the MMTV gag-pro wild-type and mutant frameshift sites (see above). Thus, it is possible that the frameshift site has a specific requirement for a hypomodified isoacceptor that can preferentially decode one of its cognate codons. In addition, because the frameshift site invokes misreading (at least among cognate tRNA codons), then in light of the observation that two amino acids may occur at the HIVgag-pol (Jacks et al., 1988b; Weiss et al., 1989) and MMTV gag-pro (Weiss et al., 1989) frameshift sites, a question can be raised whether the occurrence of two amino acids is caused by (1) frameshifting a t separate sites (as suggested in Weiss et al., 1989) or (2) a high level of misreading at a single frameshift site as a result of using heterologous systems (see also Section III,B,2).

The only examples observed thus far in eukaryotes of frameshifting in the +1 direction are in the Ty elements that occur in the yeast, Saccharomyces cereuisiae. The Ty elements are a family of retrotransposons that are about 5.5 kilobases in length and are flanked by direct repeats of 330-340 bp designated delta sequences in Tyl and Ty2 and sigma sequences in Ty3. The Ty elements replicate through a DNA intermediate in a fashion similar to retrovirus replication and Ty occurs within viruslike particles designated Ty-VLPs. The Ty proteins of the VLP coat are encoded within TYA, which is analogous to the retroviral gag gene: the Ty PR, RT, and IN are encoded within TYB, which is analogous to the retroviral pol gene (for reviews see Wickner, 1989; Garfinkel, 1991) . T Y A and TYB overlap each other by 38-44 bp and the TYB reading frame is offset from TYA by one base in the 3' direction. TYB is expressed as a fusion protein with the TYA gene product, and expression of the fusion protein requires ribosomal frameshifting in the + 1 direction (Clare and Farabaugh, 1985; Mellor et al., 1985; Wilson et al., 1986; Clare et al., 1988; Belcourt and Farabaugh, 1990;  S. Sandmeyer, personal communication 1991 1. Thus, expression of TYA-TYB is like that observed in retroviruses with the exception that circumventing the termination codon at the end of the TYA (gag) gene is effected by ribosomal frameshifting in the + 1 instead of the -1 direction.

Deletion and site-directed mutagenesis studies of bases within a 14-oligonucleotide sequence, which was previously shown to promote frameshifting in Tyl (Clare et al., 19881 , identified seven bases that are responsible for the frameshift event . The seven bases are CUU AGG C; these codons are in the 0 reading frame. The studies demonstrated that all the information necessary for altering the reading frame in the + 1 direction in Tyl is present in the seven-nucleotide sequence that constitutes the frameshift signal. Ty3 also frameshifts in the +1 direction and the signal for this event has been shown to exist somewhere within a 21-bp region of the 38-bp overlap; but, interestingly, this 21-bp region does not contain the sevenbase signal used by Tyl (S. Sandmeyer, personal communication 1991).

Belcourt and Farabaugh (1990) prepared a construct with the frameshift signal for Tyl 15 nucleotides (five codons) downstream of an initiation codon. The sequence of the transframe peptide generated from this construct showed that Leu-Gly, but not Leu-Arg, was decoded by the CUU _AGG U frameshift signal, where A denotes the site of the frameshift and Leu-Gly are decoded by CUUAand GGU, respectively. Interestingly, the tetramer CUUA contains overlapping leucine codons in the 0 (CUU) and + 1 (UU& reading frames , and yeast cells contain a leucine tRNA capable of decoding all six leucine codons (Weissenbach et al., 1977) . Thus, the frameshift in Tyl involves a slippage from one leucine codon in the 0 reading frame to an overlapping leucine codon in the + 1 reading frame ; see also below) and the site of the frameshift is the 3' base of the downstream overlapping leucine codon.

The heptanucleotide frameshift signal in Tyl (CUU AGG C) has two unusual features. First, as noted above, it contains overlapping leucine codons in the 0 (CUU) and +1 (UUA) reading frames. Second, the arginine codon AGG within the frameshift signal is normally decoded by a tRNA, tRNA&&, which is present in low amounts in the host cell (lkemura, 1982) . Belcourt and Farabaugh (1990) demonstrated that increasing the intracellular levels of tRNA&& in yeast decreases the level of frameshifting in Tyl, providing strong evidence that the absence of this tRNA from the ribosomal A-site enhances the frameshift event. These investigators also determined that slippage from the first to the overlapping, downstream leucine codon is essential to the frameshift event and presumably occurs with peptidyl-tRNAL'" . They introduced a leucine tRNA gene with anticodon AAG into the host, whose gene product could decode CUU but not UUA codons. Transfer RNAk& would be expected to compete with the "shifty" tRNA, tRNAPzG, which is capable of reading all six leucine codons (Weissenbach et al., 1977) . The level of frameshifting in Ty was severely inhibited by tRNAki%, demonstrating that slippage from CUU to UUA is essential for frameshifting in Ty. These observations led Belcourt and Farabaugh (1990) to propose a "peptidyl-tRNA slippage" model for frameshifting in the + 1 direction in Ty. In this model, the relatively low abundance of tRNAcAifir results in a pause in translation a t an AGG codon. If this codon partially overlaps an upstream slippery CUUA sequence, then in the absence of an occupied A-site, the peptidyl-tRNA&Y; (which is on the CUU codon in the P-site) slips one base foward to decode UUA. The slippage event establishes the + l reading frame and normal translation then proceeds ).

As noted above, Ty3 does not have a sequence within a 21-bp region that promotes frameshifting comparable to the Tyl seven-nucleotide frameshift signal 6 . Sandmeyer, personal communication 1991 1. Candidates for the frameshift signal within the 21-bp region involve a n alanine GCG codon and an arginine CGA codon, which are used infrequently in yeast (and hence their cognate isoacceptors are present in low abundance) (see Ikemura, 1982) . Either one of these codons may serve a similar function as AGG in the Tyl frameshift signal. The codon immediately upstream of GCG is AAG ( a lysine codon) and that of CGA is AAC (an asparagine codon), and a change to the + l reading frame would mean that either tRNhL& slips to misread GGA or tRNA(& slips to misread ACC. With respect to the latter tRNA it is of interest to note that yeast tRNA lacks the highly modified Q base in the anticodon; as noted in Section 111,A,5, tRNAA"" without Q base is proposed as the shifty isoacceptor in some of the -1 frameshift events. Ultimately, however, the site of the frameshift in Ty3 will have to be identified by mutagenesis studies and by sequencing the transframe protein.

The present review has focused on our present understanding of translational suppression in retroviral gene expression. The existence of this phenomenon obviously raises a number of important questions:

What are the precise mechanisms of readthrough suppression and ribosomal frameshifting? What are the signals in the retroviral RNA that induce these unusual behaviors in the cellular translational machinery? Why have different retroviruses evolved completely distinct mechanisms that apparently accomplish the same ends? Finally, to what extent is either of these modes of translational suppression used as a regulatory mechanism in the synthesis of host cell proteins? As should be clear from the foregoing discussion, these questions in general remain to be answered; retroviruses appear to offer an invaluable tool for the analysis of translational mechanisms in higher eukaryotic cells.

Remarkably, there is currently no evidence that these suppression mechanisms are used by the host. It thus seems possible that an understanding of these phenomena will suggest approaches, including the use of antisense RNA or new types of antiviral drugs, which could help combat the induction of disease by these viruses.

Since this review was prepared, mutational analysis of artificial constructs including the Mo-MuLVgag-pol junction has ( 1) shown that the 57 pol nucleotides immediately 3' of the gag termination codon are necessary and sufficient for suppression, and ( 2 ) provided strong evidence that the two base-paired stems in the proposed pseudoknot struc-ture in this region (ten Dam et al., 1990) are crucial for suppression (Wills et al., 1991; Feng et al., 1992) .

New Craigen

Molecular Biology ofTumor Viruses

New Aspects of Positive-Strand RNA Viruses

Proc. Natl. Acad

The Retroviruses

I n

Nucleic Jacks, T. 11990)

Proc. Natl. Acad. Sci

Recent Results Cancer

Retroviral Proteases

Control ofHIV Gene Expression

I n "Transfer RNAs and Other Soluble RNAs

Transfer RNA: Structure, Properties and Recognition

Virology Tamura

I n "Translation in Eukaryotes

Cold Spring Harbor Weiss

Proc. Natl

Proc. Natl. Acad. Sci

The authors thank Ian Brierley, Mario Chamorro, David Garfinkel, Ray Gesteland, Anne-Lise Haenni, Lou Henderson. Alik Honigman, Stephen Inglis, Amos Panet, C. W. A. Pleij, Suzanne Sandmeyer, James Skuzeski, Edwin ten Dam, Harold Varmus, and Reed Wickner for communicating data prior to publication. We thank Cheri Rhoderick and Carol Shawver for preparation of the manuscript, and Richard Frederickson for assistance with graphics.Research sponsored in part by the National Cancer Institute, DHHS, under contract No. N01-CO-74101 with ABL. The contents of this publication do not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.