key: cord-0008801-7obfhwey
authors: Ryabova, Lyubov A; Pooggin, Mikhail M; Hohn, Thomas
title: Viral strategies of translation initiation: Ribosomal shunt and reinitiation
date: 2004-01-07
journal: Prog Nucleic Acid Res Mol Biol
DOI: 10.1016/s0079-6603(02)72066-7
sha: 8d884680783a659485086fa05094495051e00f98
doc_id: 8801
cord_uid: 7obfhwey

Due to the compactness of their genomes, viruses are well suited to the study of basic expression mechanisms, including details of transcription, RNA processing, transport, and translation. In fact, most basic principles of these processes were first described in viral systems. Furthermore, viruses seem not to respect basic rules, and cases of “abnormal” expression strategies are quiet common, although such strategies are usually also finally observed in rare cases of cellular gene expression. Concerning translation, viruses most often violate Kozak's original rule that eukaryotic translation starts from a capped monocistronic mRNA and involves linear scanning to find the first suitable start codon. Thus, many viral cases have been described where translation is initiated from noncapped RNA, using an internal ribosome entry site. This review centers on other viral translation strategies, namely shunting and virus-controlled reinitiation as first described in plant pararetroviruses (Caulimoviridae). In shunting, major parts of a complex leader are bypassed and not melted by scanning ribosomes. In the Caulimoviridae, this process is coupled to reinitiation after translation of a small open reading frame; in other cases, it is possibly initiated upon pausing of the scanning ribosome. Most of the Caulimoviridae produce polycistronic mRNAs. Two basic mechanisms are used for their translation. Alternative translation of the downstream open reading frames in the bacilliform Caulimoviridae occurs by a leaky scanning mechanism, and reinitiation of polycistronic translation in many of the icosahedral Caulimoviridae is enabled by the action of a viral transactivator. Both of these processes are discussed here in detail and compared to related processes in other viruses and cells.

Eukaryotic chromatin seems to be very uneconomically used. Large regions are devoid of genes, large distances usually separate individual genes, and large and multiple introns separate the true coding regions of a gene. The situation is fundamentally different with virus genomes. These have evolved to optimize the available genomic space to allow a high density of coding information and c/s-acting control elements such as promoters, polyadenylators, and replicators (1) . Noncoding regions are minimized and DNA sequences are often multiply used, that is, several overlapping reading phases of viral DNA are used to encode proteins, and coding regions also contain c/s-elements. Often, polyproteins are produced that are later cleaved to individual functional proteins, which reduces the numbers of promoters and polyadenylators on the DNA as well as on 3'-and 5'-untranslated regions on the RNA. Apparently, this justifies the cost of the extra proteinase required for polyprotein processing. Furthermore, virus RNAs are often multiply used, either through alternative splicing or by various mechanisms of polycistronic translation. In addition some viral proteins are multifunctional.

The class of reverse-transcribing elements (2) , that is, retroviruses, pararetroviruses, and retrotransposons, although related in type and number of genes, have evolved a variety of different strategies of gene economy (Fig. 1) . Hepadnaviruses have the highest degree of gene and signal overlap: The envelope-coding region is completely included in a different reading phase of the pol gene. Hepadnaviruses also make use of multiple transcription initiation from a single promoter. This leads to two types of capsid protein and three types of envelope protein, distinguished by different N-termini. Complex retroviruses, such as human immunodeficiency virus-1 (HIV-1), are specialists in alternative splicing. This leads to six different mRNAs originating from a single original VIRAL 3 and 192. The processing sites of the PVCV polyprotein are not known. HBV Pol expression does not occur by frameshifting; conflicting results exist regarding whether it is translated via leaky seanning/reinitiation or internal ribosome entry as well.

Mechanisms of translation initiation have been intensively studied and grouped into several categories that differ in the pathway by which ribosomal subunits reach the initiator AUG codon (reviewed by Jackson in Ref. 6 ).

The majority of eukaryotic mRNAs are monocistronie. Their 5'-leader sequences are capped, short, and unstructured, and they direct initiation from the AUG codon nearest to the capped 5~-end via a cap-dependent ribosomescanning mechanism (7, 8) . The 40S ribosomal subunit, together with associated VIRAL STRATEGIES OF TRANSLATION INITIATION   5 initiation factors, engages the mRNA at the capped 5'-end and migrates linearly until it encounters the first AUG codon. This occurs in several steps: the m 7 GpppX cap at the 5'-end of most mRNAs is recognized by subunit 4E of eukaryotic initiation factor (elF) 4F, which contains in addition elF4G, elF4A (an ATP-dependent helicase), and ATP. Ribosomes separate into 40S and 60S subunits before binding mRNA. The 40S subunit binds elF3 and the elF2-GTPinitiation methionyl transfer RNA (Met-tRNAi) ternary complex (for review, see Refs. 9, 10) . elF5 has been implicated in bridging elF3 and elF2 (11) . The complex requires also elF1 and elF1A to begin scanning to search for the correct AUG start codon (12) . At the AUG codon, which is recognized by base pairing with the anticodon in Met-tRNAi, the 40S subunit stops, a 60S subunit joins to form an 80S ribosome in the presence of factor elF5B (13) , and elongation begins.

The mechanism of cap-dependent scanning has been elaborated in mammalian and yeast cell systems, but most of its features are also valid for plants (14, 15) . Scanning can be strongly affected by c/s-acting elements and transacting factors: (1) strong secondary structures (below -50 kcal/mole) completely block scanning, causing stalling of 40S ribosomes at the 5'-side of the hairpin (16) (17) (18) (19) ; (2) multiple AUG start sites upstream of the main ORF reduce efficiency of downstream initiation (20, 21) ; and (3) RNA-binding proteins can specifically block scanning or 40S loading on the capped 5'-end (2~, 23).

About 10% ofeukaryotic mRNAs (particularly those encoding growth-related factors, tumor suppressors, transcription factors, and protooncogenes) and many viral RNAs contain long, often structured leaders with one or multiple small ORFs upstream of the main coding region, which would interfere with the normal scanning process. To account for this, alternative modes of translation initiation have been proposed that allow the presence of inhibitory elements in the leader to be overcome: (1) "leaky scanning" (8) , (2) internal initiation (24) (25) (26) , (3) reinitiation after small ORF translation (see reviews by Hinnebusch, Ref. 27, and Morris and Geballe, Ref. 28) , and (4) nonlinear ribosomal scanning or ribosomal shunting (29) (30) (31) (32) (33) (34) .

Leaky scanning allows the bypass of the upstream AUG codon in at least two cases. A start codon located close (< 10 nucleotides, nt) to the cap site (35) (36) (37) or an AUG in a suboptimal nucleotide context for initiation (20) (or a non-AUG start codon in an optimal context) is poorly recognized by ribosomes. The latter effect can be suppreSsed by a structural element located 14 nt downstream of the start codon (38) . Thus, in some cases, upstream ORFs might be bypassed by leaky scanning, obviating the need for reinitiation at the downstream AUG start site. 6 RYABOVA ET AL.

Internal initiation involves direct binding of the 40S ribosome to a region (internal ribosome entry site, IRES) at, or upstream of, the authentic AUG codon, and is mainly used by viruses which have developed a strategy to shut down cap-dependent initiation in a host cell (for review see Refs. 39, 40, 10) .

This mechanism does not require a cap structure on the mRNA and thus does not depend on elF4E and the N-terminal part of eIF4G containing the elF4Ebinding domain (41) . Therefore, in contrast to cap-dependent initiation, internal initiation is not impaired by cleavage of elF4G by viral proteases. The bestcharacterized IRESs are from picornaviruses and hepatitis C virus (HCV), classical swine fever virus (CSFV), and bovine viral diarrhea virus (BVDV) (42, 43) .

After translating an upstream ORF, terminating ribosomes might remain associated with the mRNA, continue scanning, and initiate at a downstream start codon with an efficiency depending both on mRNA c/s-elements and transacting factors (28) . Reinitiation requires the recruitment of initiation factors and, in general, can occur in eukaryotic RNAs when the upstream ORF is short (2--30 codons), with reinitiation frequency increasing with the distance between the short ORF (sORF) and the "main" ORF (44, 27, 45, 46) . The requirement for scanning 40S ribosomes to recruit initiation factors, including the ternary complex elF2-GTP-Met-tRNAi, is the limiting step for reinitiation of translation (27) . The decrease of reinitiation frequency with the increase of the length of the sORF suggests that, during the short translation event, some initiation factors remain associated with the translational machinery after termination of translation and can be used for reinitiation (44, 19) .

Scanning posttermination ribosomes might be affected by the mRNA structure surrounding the stop codon of the sORF (47) , the structure of the sORFencoded peptide (48, 49) , and, in some special cases, by degradation of the mRNA (50, 51) .

After translating an upstream ORF longer than about 30 codons, the terminating ribosome is believed to dissociate from the mRNA. However, there are a few reports suggesting reinitiation at AUG codons located a short distance upstream of the long ORF termination codon in artificial RNA constructs in mammalian cells (52, 53) .

Ribosome shunting, first described in CaMV, is a nonlinear scanning mechanism in which initially scanning ribosomes are transferred directly from a 5'-donor site to a 3~-acceptor site without linear scanning of an intervening region (30) . Shunting 

The plant viral family Caulimoviridae and its animal counterpart Hepadnaviridae, including the well-known human hepatitis B virus (HBV), have been proposed to be classified into a suborder Pararetrovirinae of the order Retrovirales of reverse-transcribing elements (2) . The members of Pararetrovirinae-also known as pararetroviruses---contain DNA in mature virions, in contrast to the RNA-containing retroviral virions. It is worth mentioning that virions of spumaviruses--a separate genus of Retrovirinae (or retroviruses)----can also contain DNA (54) . This, together with other striking similarities to pararetroviruses (reviewed in Refs. 55, 56) , places them between the pararetro-and orthoretrovirus suborders of the Retrovirales. The DNA of pararetroviruses is targeted to the nucleus of infected cells, where host RNA polymerase II transcribes a terminally redundant pregenomic RNA, which is eventually reverse-transcribed by the viral reverse transcriptase complex to produce the DNA genome. Retroviruses and pararetroviruses use their (pre)genomic RNA and its spliced versions as polycistronic mRNAs that can be translated into more than one protein. Strategies of polycistronic translation include: (1) frameshifting (in most retroviruses; reviewed in Refs. 3, 57) , (2) leaky scanning [e.g., RTBV (5); HIV-1 (58)], and (3) reinitiation activated by a viral transactivator [discovered first in CaMV (59); see below]. In the controversial case of the HBV Pol ORF (the second cistron following the overlapping Core ORF), translation is believed to occur by leaky scanning combined with reinitiation after translation of a short ORF within the Core ORF (60--62) (see Fig. 1 ). However, alternative initiation mechanisms such as the internal ribosome entry suggested in earlier studies (63) and ribosome shunting have not been convincingly ruled out. Moreover, in duck hepatitis B virus, which represents a distinct genus of Hepadnaviridae, a mechanism other than leaky scanning, with some, but not all features of internal, initiation, has been implicated in Pol translation (64) .

To regulate the multiple usage of pregenomic RNA for replication as well as translation, pararetrovirnses have adapted to versatile properties of the eukaryotic translation machinery. In HBV, passage of the translating 80S ribosome (but not the scanning 40S ribosomal subunit) through the RNA packaging signal (e) disrupts its secondary structure, thereby preventing encapsidation/reverse transcription of the pre-Core RNA (65) (Fig. 2) . On the other hand, the reduced ability of the 40S ribosome to scan through such secondary structures (18) accounts for efficient functioning of the ~ signal for packaging of the shorter Core RNA, which lacks the 5' portion with the AUG start codon of the pre-Core ORF (65) (see Fig. 2 ). The e structure-mediated retardation of the scanning ribosome heading toward the Core start codon might allow the formation of the e-Pol complex and initiation of reverse transcription on a subpopulation of Core RNAs. Moreover, possible collision of the Pol-nascent DNA complex with the scanning ribosome may help to relocate this complex to the 3'-end of Core RNA in the poorly understood process of template switching that is required for a complete cycle of reverse transcription (66) (67) (68) . Interestingly, despite its negative effect on scanning, and therefore on initiation of polycistronic translation of the Core RNA, the E structure plays a positive role in pre-Core RNA translation, because it ensures the recognition of the upstream pre-Core AUG, which is in a suboptimal context (69), in accordance with Kozak's rule (38) . This should ensure complete exclusion ofpre-Core RNA from the packaging process.

Plant pararetroviruses such as CaMV and RTBV exploit the shunting property of the scanning ribosome (31, 70) (see below in greater detail) to allow the bypass of a large region of the pregenomic RNA leader containing a putative packaging signal (for CaMV, see Fig. 2 ). We speculate that regulation of shunting efficiency may affect conformations of the intervening structure and thereby modulate the interaction of the viral coat protein and the putative packaging signal. In the case of CaMV, this specific RNA--coat protein interaction has been demonstrated by using the yeast three-hybrid system (71) . The sorting of genomic RNA for translation and replication has been studied in retroviruses. In some studies, the existence of two separate pools of RNA, one for translation, the other for packaging, was proposed (e.g., 73). However, in many complex retroviruses, such as ]entiviruses (e.g., HIV-1) and alfaretroviruses (e.g., Rous sarcoma virus, RSV), RNA for packaging seems to be sequestered away from the pool of translated RNAs (see, e.g., Ref. 74). The case of RSV illustrates the complexity of possible interplay between the host translation machinery and viral c/s-and trans-acting factors involved in RNA packaging. The 379-nt-long leader of RSV RNA preceding the Gag ORF contains three phylogenetically conserved sORFs and extensive secondary structure (Fig. 2) . The 5'-proximal sORF, sORF 1, is the major ribosome-binding site (75, 76) . sORF 3 is also translated, and the efficiency of this translation event is thought to regulate a conformational switch in the structure of the RNA packaging signal, qJ (77) (78) (79) .

The conflicting reinitiation/leaky scanning models of sORF 3 and Gag ORF translation deduced from those early studies have been recently revisited by the group of J.-L. Darlix (80) , who identified a bipartite IRES driving the two initiation events. A new model proposes that the translating ribosomes stall at sORF 1, most likely due to the stable downstream structure and/or a particular feature of the encoded peptide (MAGPLIP), which resembles the mammalian S-AdoMet decarboxylase regulatory peptide (MAGDIS), which blocks proper termination (28, 81) thereby preventing reinitiation at downstream start codons.

On the other hand, the Gag AUG start codon is recognized by ribosomes entering internally at the 3' IRES. Thus, both events would preserve the intervening structure containing the o2 signal. In addition, ribosomes can enter at a somewhat weaker IRES, roughly mapped to the intervening region upstream of the sORF 3 AUG start codon. This latter ribosome binding followed by sORF 3 translation might interfere with Gag binding to the o2 signal. The same study (80) also proposes a mechanism for translation initiation on the RSV spliced v-Src mRNA. The splicing event creates a new sORF of 30 nucleotides, where the Gag ATG is in-frame with a stop codon located behind the splice junction (see Fig. 1 ). Presumably, IRES-driven recognition of the Gag AUG would lead to translation of this sORF followed by reinitiation at the v-Src AUG. In fact, the 66-nt region between the sORF and the v-Src ORF favored efficient reinitiation when placed after a large cistron on an artificial bicistronic mRNA (80) , thus supporting the reinitiation model. IRES is located just downstream of the Gag ORF start codon (83) and the preceding leader sequence containing the o2 signal does not seem to exhibit any IRES activity (87) (Fig. 2 ). Scanning ribosomes migrating through the o2 signal may interfere with the packaging process, unless they are retarded by the leader secondary structure as in the case of HBV Core RNA described above. Notably, the structural configuration of the RSV RNA leader with a 5'-proximal sORF terminating in front of an extended stem-loop structure resembles the shunt configuration found in plant pararetroviruses (72) . The experimental evidence does not necessarily exclude the possible occurrence of shunt-dependent initiation events on RSV RNA as suggested in earlier work by Darlix and coworkers (75) .

A tandem array of Gag (or Core, in the case of HBV) and Pol genes represents the core of reverse-transcribing elements that provides the structural and enzymatic functions essential for replication. The expression of Gag and Pol, and VIRAL STRATEGIES OF TRANSLATION INITIATION   11 their ratio, are tightly regulated at different stages of replication. Translational control plays a major role in this regulation. In most retroviruses, Pol is expressed from genomie RNA as a Gag-Pol fusion protein (e.g., 88) , which is thought to be coassembled into the virion together with the much more abundant Gag protein; and eventually processed to release the mature reverse transcriptase. The correct ratio of Gag protein to Gag-Pol protein is critical for viral infectivity (e.g., 20:1 for HIV-1; 89). The mechanisms driving Gag-Pol translation in retroviruses are (1) programmed ribosome "-1" frameshifting (reviewed in Refs. 90, 91) and (2) stop-codon readthrough [in type C retroviruses, e.g., UAG readthrough in murine leukemia virus (92) ; reviewed in Ref. 93 ]. In both cases, 5-10% of the translating ribosomes produce the Gag-Pol fusion protein, and the others terminate at the Gag stop codon. A structural stem-loop or pseudoknot element downstream of the "slippery" heptamer in the case of frameshifting, or the suppressed stop codon in the case of readthrough, is believed to pause 80S translating ribosomes to increase the efficiency of these events about three orders of magnitude above background. Interestingly, transient pausing of translating (or scanning) ribosomes at various structural elements in mRNA seems to be a common feature of unusual translation mechanisms such as frameshifting, readthrough, hopping (see Ref. 94 and references therein), and shunting (see below).

In hepadnaviruses and many plant pararetroviruses, the Gag and Pol genes are uncoupled and Pol is translated from its own AUG start codon (95) (96) (97) . However, different translation strategies are used to initiate Pol translation. Whereas hepadnaviruses seem to rely on leaky scanning or internal entry abilities of the 40S ribosome (as discussed above), plant caulimoviruses, including CaMV and two other genera of Caulimoviridae, code for a transactivator protein (transactivator/viroplasmin, TAV) that allows polycistronic translation of viral RNAs by a reinitiation mechanism (see below). In CaMV, the Gag and Pol genes (ORFs IV and V, respectively) are located internally both on the pregenomic RNA and its spliced derivatives. Their expression in plant protoplasts requires the presence of TAV (98) . ORF IV (Gag) is, most likely, translated as a second cistron from one of the three spliced RNAs (99), although its translation from pregenomic RNA is not excluded. The mechanism of Pol translation has not been investigated in sufficient detail to draw any definitive conclusion (97; also M. Schultze, unpublished PhD thesis, University of Basel, Basel, Switzerland). Theoretically, ORF V (Pol) could be translated in the presence of TAV from the ORF IV-expressing spliced RNA as the third cistron, or from pregenomic RNA as the sixth cistron. Alternatively, a monocistronic RNA species (either subgenomic or spliced RNA) for ORF V might exist. Indeed, there have been occasional, unconfirmed reports of a 22S RNA in CaMV-infected plants (e.g., 100). In this regard, an additional link has been established between plant caulimoviruses and animal spumaviruses (see above); in human foamy virus, Gag and Pol translational events are also uncoupled (101) and Pol is expressed from a separate, spliced RNA (102) .

In contrast to the three genera of Caulimoviridae coding for TAV (CaMV-like, soybean chlorotic mottle virus-like, and, possibly, cassava vein mosaic virus-like viruses; see Ref. 103) , the three other genera of plant pararetroviruses (RTBVlike, badna-viruses, and PVCV-like viruses) do not possess any homology to TAV consensus sequences. Strikingly, their genome organization further differs in that they encode Gag and Pol within a single polyprotein, thus predicting a 1:1 ratio upon proteolytic processing. The paradox of the apparent surplus of Pol protein that might be expressed by these viruses remains to be understood.

Interestingly, the Schizosaccharomyces pombe retrotransposon Tfl also contains a single ORF for Gag and Pol without any obvious means for overexpressing Gag protein (104).

The RNA leaders of CaMV and other members of the Caulimoviridae contain all three types of elements inhibitory for scanning: (1) a low-energy elongated hairpin (105), (2) several sORFs (72) , and (3) a putative packaging signal (72) that, in the case of CaMV, interacts with the viral coat protein (71) . These would make translation initiation at a downstream ORF difficult (Figs. 2 and 3A). However, translation downstream of the leader of CaMV occurs with reasonable efficiency in plant protoplasts (106, 30) , in several different in vitro systems [wheat germ extract (107), reticulocyte lysate (81) , and yeast extract (108)], and in transgenic plants (109) . This cannot be explained by internal ribosome entry, since initiation at the AUG of ORF VII downstream of the CaMV leader is strongly cap-dependent (107, 110) .

Originally, it was found that certain parts of the leader support downstream translation, whereas others are inhibitory (30) . This analysis suggested a mechanism termed the "ribosomal shunt" by which ribosomes bypass the inhibitory leader elements during normal cap-dependent scanning (30, 31) . Antisense oligonucleotides strongly inhibited translation if directed against 3'-and 5'-proximal parts of the leader, but not if directed against the central region The ribosomal shunt was reconstructed by providing the leader in two RNA molecules transcribed from separate plasmids (31) . The first, spanning residues 1-300, contains the shunt "take-off site" and the second contains the "shunt 14 RYABOVA ET AL.

landing site" (see below) and a reporter ORF. If these two RNA molecules anneal, the elongated hairpin could form, reconstituting the shunt structure. That the shunt can indeed occur from these two separate molecules was demonstrated in plant protoplasts, albeit with lowered efficiency (31).

l. "TAKE-OFF" AND "LANDING" SITES Extensive analysis of the effects of insertions of a strong stem interfering with scanning ribosomes, as well as start codon insertions, revealed shunt takeoff and landing sites flanking the bypassed region of the leader. It is assumed that formation of the leader hairpin structure promotes shunting by bringing the shunt landing site upstream of the first main ORF into close spatial proximitywith a shunt take-off site downstream of sORF A, the first 5~-proximal sORF (110) .

Indeed, the CaMV shunt landing site has been mapped on the 3t-proximal part of the elongated hairpin to between positions 548 and 562 (31) . In RTBV, the shunt landing site has been precisely mapped to just downstream of the leader hairpin (70) . CaMV sORF A turns out to be a key element promoting shunting in CaMV and other pararetroviruses. The first indication of the importance of sORF A for shunting came from experiments where replacement of the sORF A AUG start codon with an UUG codon nearly abolished translation downstream of the CaMV leader (31) . Extensive mutational analysis of sORFs within the 35S RNA leader, separately and in different combinations, revealed that only mutations in sORF A frequently reverted on passage of the respective CaMV mutants in planta (see Table 1 in Ref. 111 ). In addition, mutations leading to the disruption of the elongated hairpin also reverted. Moreover, a striking correlation has been found between the efficiency of ribosomal shunt and viral infectivity; point mutations in sORF A which reduced the level of shunt-dependent expression also reduced infectivity of the virus in turnip plants (112) . Mutational analysis of sORF A confirmed its important role in ribosomal shunt in different in vitro systems (81, 108) and in plant protoplasts (113) .

Comparison of sORF A and several artificial sORFs of different length revealed that the optimal length of the 5~-proximal sORF for efficient shunting was between 2 and 10 codons; longer sORFs significantly reduced shunt-mediated translation (108, 110, 113) . The peptide sequence of sORF A in general was not important for shunting or infectivity of the virus (112, 114) . A notable exception is that a one-codon (start-stop) ORF does not promote shunting at all (113) . Note, however, that in special cases the sORF of optimal length promotes shunting with only a low efficiency (81) .

Another critical parameter was found to be the spacing between the stop codon of sORF A and the base of the leader hairpin. For optimal shunting " this distance should be about 5-10 nucleotides (110, 113) . Moving the position of sORF A closer to the hairpin base would allow a corresponding number of additional stem base pairs to melt during translation of sORF A, and, as a consequence, the shunt landing site is shifted to a position further upstream (81) . Shifting of sORF A further away from the hairpin base precludes melting of the hairpin; the landing site is moved to position further downstream, and the shunting efficiency is strongly diminished. Thus, the position of the shunt landing site depends on the take-off site, which is itself determined by the end of the 5'-proximal sORE

The stem structure downstream of the 5'-proximal sORF supports nearoptimal shunting; a reduction in the number of base pairs reduced its efficiency (110, 115) . In vitro analysis of the strength of the CaMV elongated hairpin, in particular its most stable lower part (so-called stem section 1), suggested the importance of base pairings rather than primary sequence of stem section 1. Physical disruption of stem section 1 by mutations in the left arm of the hairpin structure resulted in a significant reduction of shunt expression, but this could be restored by compensatory second-site mutations that restored secondary structure (110, 113, 115) . Moreover, an artificial strong stem positioned downstream of sORF A was well able to replace the CaMV hairpin, and an artificial shunt structure could be assembled from the Kozak stem mentioned above with an artificial sORF upstream (108, 116) . It was therefore proposed that the combination of sORF followed by a stem could be a universal signal promoting the bypass of internal leader region. In line with this, the sORF/stem combination, but little of the primary sequence, is conserved in most of the Caulimoviridae (72).

Shunting ribosomes seem to initiate with low fidelity, since they can initiate at non-AUG start codons if such codons are positioned within or near the shunt landing site. After some scanning in the 3~-direction, fidelity improves significantly and non-AUG codons are no longer well recognized (81) . These results indicate that, upon landing, shunting ribosomes might be deficient in factors responsible for correct AUG start codon recognition. The situation is similar to that in the case of IRES-mediated initiation, where enhanced recognition of non-AUG codons placed at the entry site has been reported (117) .

In the case of RTBV, ORF I lacks a proper AUG initiation codon and shuntdependent translation is initiated at an AUU codon (70) . Shunting ribosomes most likely land directly on this non-AUG codon, apparently without scanning. An AUG start codon introduced at the position of the AUU codon gives a six-to sevenfold increase in the level ofORF I expression, whereas an AUG placed 9 nt upstream of the authentic start site is not recognized. The efficiency of ORF I start codon recognition is low, and effective leaky scanning through the AUU codon as well as through further downstream start codons in suboptimal context 16 RYABOVA ET AL. allows polycistronic translation of three of the four large ORFs on the RTBV RNA (5) (see Fig. 1 ).

Translation downstream of the CaMV leader could occur either via reinitiation by shunting ribosomes or via an IRES induced after 5~-proximal sORF translation. To distinguish between these two possibilities, sORF A was replaced by an artificial, six-amino-acid-long sORF, MAGDIS, which is known to be inhibitory for downstream start eodon recognition (49, 118) . This sORF caused stalling of the translating ribosome near the sORF stop codon (81), thus preventing scanning toward the 3'-end and downstream reinitiation in a bicistronic construct. The stalling event strongly represses expression of a downstream reporter gene in wheat germ extract, reticulocyte lysate, and plant protoplasts, in both linear and shunt-mediating constructs, suggesting that shunting involves a reinitiation process (81) . The inhibitory effect of the MAGDIS sORF demonstrates that the 5'-proximal sORF must be translated before shunting occurs. Mutant sORFs MAGDI or MAGRIS, which did not cause stalling, allowed efficient progression toward the 3~-end. Thus, CaMV-type shunting can be considered as modified reinitiation.

Further support for this hypothesis comes from the fact that the CaMVencoded reinitiation factor TAV significantly enhances shunt-dependent expression in plant protoplasts (31, 113) .

The host protein requirement for shunting is not clear. However, the efficiency of shunting varies significantly between protoplasts originating from host and nonhost plants (106) . On the other hand, shunting functions well in reticulocyte lysate and yeast, indicating that plant-specific factors are not required.

For the CaMV-type shunt mechanism, the 5'-proximal sORF translation event leads to the bypass of the structural element downstream. This could indicate that the short translation event might selectively remove some of the canonical initiation factors associated with the 48S initiation complex that may be inhibitory for ribosomal shunt, such as the initiation factor elF4F-associated elF4A helicase, and elF4B, which could melt the secondary structure of the stem. However, other factors might still be associated with the ribosome to promote shunting, for example, eIF3. The current shunt model (Fig. 3A) can be summarized as follows: (1) Ribosomes start cap-dependent linear scanning from the cap until they reach a 5'-proximal sORF start codon located just upstream of a strong structural element. (2) The sORF is translated and terminated, rendering 40S ribosomes shunt-and reinitiation-competent. (3) The ribosomes bypass the structured region and are able to reiuitiate at an AUG or a non-AUG start codon immediately downstream or resume scanning and reiuitiate at an AUG start codon located further downstream (81, 113) .

INITIATION 17

After translation of sORF A, a fraction of ribosomes shunt, but others can continue to migrate linearly toward the center of the CaMV leader (113, 119) . Here, they can recognize some of the central sORFs, reinitiate according to a distance-dependent mechanism, then dissociate from the mRNA after termination of translation ( Fig. 3A; 119) . In reticulocyte lysate, the fraction of shunting ribosomes greatly exceeds the fraction of linearly migrating ribosomes (119) .

The arrangement of an sORF followed by a stem structure occurs twice in the CaMV leader; if the main shunt is impaired or ribosomes manage to scan through stem section 1, the combination of sORF B and stem section 2 can support shunting in vivo and in vitro (119) (Fig. 3A) . The function of the second shunt might just be to contribute to a tighter protection of the central leader structure, which is thought to be required for packaging. This alternative shunt pathway, although not essential, may increase viral fitness in planta.

sORF-dependent shunting has not been reported for viruses other than Caulimoviridae. The molecular details of shunting or of related phenomena proposed for other viruses remain obscure. However, three intensively studied cases of shunting might shed some light on general mechanism of ribosomal shunting.

The adenovirus tripartite leader is a 200-nt-long noncoding region, which facilitates translation of viral mRNAs late in infection (Fig. 3B) . The adenovirus late mRNAs transcribed from the major late promoter all contain the tripartite leader. Structural analysis of the leader predicted that the 5~-end is unstructured, whereas the 3~-part includes several moderately stable stem-loops (120) (Fig. 3B) . As was shown for Caulimoviridae, translation initiation starts at the 5'-end of the tripartite leader, which does not function as an IRES (121) . In uninfected cells, initiation downstream of the tripartite leader occurs by both scanning and shunting mechanisms with about similar efficiencies (33), again resembling the situation in CaMV, where a significant proportion of scanning ribosomes migrates toward the central part of the leader (113, 119) .

The shunting phenomenon in adenovirus has been demonstrated mainly by insertion of strong hairpin structures (-80 kcal/mole) or AUG codons within the leader, either close to its 5'-end or in the middle part, where both types of insertion would affect the level of translation at the downstream ORF if this region is scanned (33) . These insertions are inhibitory in the unstructured leader regions comprising the first 80 nt and about last 35 nt and, accordingly, these regions are believed to be scanned linearly. In contrast, insertions into the structured middle part of the leader are not inhibitory and therefore are considered to be potentially shunted. However, it was found recently that this structured part contains elements essential for efficient shunting. These regions are complementary to the 3'-end of 18S ribosomal RNA (34) (see Fig. 3B ). The first of these is located within the first 80 nucleotides of the tripartite leader, while the other two are located in the central structured domain. A high level of ribosomal shunting was found to be dependent on the presence of any two of these three regions. Some kind of redundancy exists, since the removal of each of these regions separately had little influence on shunting and scanning efficiency. However, the combination of the second and the third complementarities are functionally dominant, that is, their combined removal significantly impaired shunting. Thus, the shunt take-off and landing sites are apparently degenerate and might be affected by the particular complementarity region present.

The functional role of the complementarity regions in ribosomal shunt remains to be understood. Despite lack of direct evidence for prokaryotic Shine-Dalgarno-type interactions in eukaryotes, a growing number of examples are accumulating where interactions between 5'-noncoding mRNA and 18S RNA regions either promote internal initiation (122) or facilitate the binding of 40S ribosomes to RNA leaders (S. Zhanibekova, R. Akbergenov, and B. Iskakov, personal communication). Similar sequences are found at uORF4 of the yeast GCN4 leader (47) and close to the shunt landing site in CaMV (98) . These complementarity regions might induce direct binding of ribosomes or rebinding of dissociated scanning ribosomes, or they might cause stalling of scanning ribosomes migrating within these regions. In this regard, an additional link might be established between adenovirus and CaMV shunt: In both cases shunting might be induced by stalling, mediated by the complementarity region in the case of adenovirus and the stem structure in the case of CaMV.

Consistent with this idea, replacement of sORF A by one of the complementarity regions from the adenovirus leader promotes shunting in the context of the CaMV leader with an efficiency similar to that of the wild-type leader (L. Ryabova and T. Hohn, unpublished) . The use of 18S RNA complementarities as shunt elements might be universal, since similar complementarities in the case of another adenovirus RNA, the IVa2 mRNA, which is synthesized during the intermediate phase of viral infection, promote bypass of internal leader regions during late stages of viral infection (34) .

It is interesting that the basal shunting activity identified in the absence of all three complementarity elements is found to be approximately 5% of the shunting efficiency of the original tripartite leader and considered to represent a basic shunt, the mechanism of which is unknown (34) . The tripartite leader confers the ability to eliminate or gradually reduce the normal requirement for eIF4E or eIF4F (121, 123) . Late (123) . Significantly, ribosomal shunt mechanisms in adenovirus also function exclusively in conditions when elF4F is inactivated, during late adenovirus infection or heat shock conditions (33) .

Like the CaMV shunt, which is activated by TAV (112) , adenovirus shunt can also be influenced by a viral protein, IA-100K, which is involved in the preferential translation of adenovirus late mRNAs (124) . Its mechanism of action in the shunt process and whether this is specific for adenovirus late RNA translation remain to be clarified. Two molecular models have been suggested to explain the phenomenon of ribosome shunt in adenovirus (125) (126) . One of these, the polycistronic P/C mRNA, contains two overlapping ORFs and is known to initiate protein synthesis at multiple start codons (127) . The order of these initiation start sites from the capped 5'-end is ACG sl of the C' ORF, AUG 1°4 of the P, V, and W ORFs (moderate initiation context), AUG n4 of the C ORF, AUGs ls3/2°1 of the Y1/Y20RFs (these two AUGs are in the weakest initiation context), and AUG 1523 of the X ORF (Fig. 3C ). Three proteins (P, V, and W) contain the same N-terminal 317 amino acids (aa), the X protein represents approximately the C-terminal 95 aa of the 568-aa-long P protein, and C, C', Y1, and Y2 (215, 204, 183, and 175 aa, respectively) represent a nested set of four C proteins with a common C-terminus (32, 128, 129) . Translation of the P/C mRNA is cap-dependent. Accordingly, its function is inhibited by cap analogs, insertion of a 5'-proximal stem-loop, and poliovirus infection (130) . ORFs C', P, and C are translated by linear leaky scanning, while the Y1/Y2, and presumably X, start sites are accessed by shunting.

No specific shunt "take-off" site could be identified by deletion mutagenesis. Moreover, artificial unstructured leaders fused to the start site of C' ORF mediated efficient shunting, leading to normal levels of Y1/Y2 expression (130) .

Both Y1 and Y2 expression yields are not affected by mutation of these AUGs to ACGs, suggesting that the fidelity of start codon recognition is strongly reduced. This is reminiscent of the situation found in CaMV and RTBV, although in the latter cases AUGs in the landing site are much more efficiently recognized than non-AUGs. AUGs inserted in two positions in front of the Y1 and Y2 start sites were not recognized as start sites and thus had little effect on the yield of Y1/Y2 proteins, suggesting that shunting ribosomes are directed precisely to the non-AUGs, much as in the case of CaMV and RTBV. Thus, the shunt landing site might be located between the two start sites of the Y1 and Y20RFs.

It can be speculated that shunting on Sendai virus RNA might also be induced by stalling of the scanning ribosome, which may be caused by the possible combination of unknown leader c/s-elements and downstream initiation events at the C', P, and C initiation codons. Pausing of the ribosome during an initiation event (131), further induced, for example, by secondary structure, might create an obstacle for the following ribosome; such pausing then would induce shunting of this ribosome past the stalled first ribosome, allowing the former to initiate further downstream.

Another candidate ORE the AUG of which is reached by discontinuous ribosome scanning, appeared to be ORF X located at the 3'-end of the P/C mRNA (29) (see Fig. 3C ).

Papillomaviruses are small, double-stranded DNA (dsDNA) viruses that replicate extrachromosomally in the nuclei of infected cells (reviewed in Ref. i32). Primary transcripts of early and late virus RNAs overlap and are processed by complex splicing and polyadenylation events. Human papillomavirus RNAs, like those of the Canlimoviridae, appear to be polyeistronie, that is, encode more than one functional protein (133) (134) (135) . However, post-transcriptional gene regulation in this case is poorly understood and requires further investigation.

Two pathways of initiation of translation have been suggested for expression of the polycistronic papillomavirus RNA containing three long ORFs, E6, E7, and E 1, reading from the 5'-end: "extreme" leaky scanning for expression of E7 and ribosomal shunt for E1 (136, 137) . Despite the very tight arrangement of ORFs E6, ET, and E1 (only 6-bp intergenie space) and 5'-dependent expression of all these ORFs, translational coupling via readthrough or reinitiation is apparently not used. Using the established criterion of examining the effect of insertion of stable stem-loops and upstream strong start eodons, both linear scanning to reach E7 [without initiating at any of the 13 AUGs preceding the E70RF (137)] and nonlinear scanning (ribosomal shunt) for E1 initiation (136) have been demonstrated. In the latter ease, the shunt "take-off" site is presumably located within the first half of ORF E6, and shunt "landing" would occur somewhere just upstream of the ORF E1 AUG. This indicates that shunting is VIRAL STRATEGIES OF TRANSLATION INITIATION 21 not a randomly organized process in papillomavirus, although the specific set of conditions, such as secondary structure or protein requirements, that can trigger the bypass mechanism are not known.

Hepatitis C virus mRNAs under control of the HCV IRES were inefficiently translated compared with capped and polyadenylated mRNAs. Addition of a cap and a polyA tail on the HCV mRNAs revealed that these structures interacted with the HCV IRES in a synergistic manner to load ribosomes onto the HCV mRNAs, thereby strongly enhancing translation. The positive effect of the cap and the polyA tail on initiation of translation at the initiator AUG embedded in the HCV IRES might have been the result of a discontinuous scanning, or shunting, mechanism where ribosomes are translocated from the cap site to the IRES (138) .

Internal initiation in poliovirus might also include a shunting step to bypass the intervening region between the upstream AUG of the IRES used for direct ribosome binding and the downstream AUG used as the translation start site (139) .

A combination of internal ribosome entry, leaky scanning, and/or shunting was also proposed for the expression of ORF 3b from polycistronic porcine transmissible gastroenteritis coronavirus (TGEV) mRNA 3 (140).

The type of shunting mechanism employed by adenovirus might also be used by the cellular mRNAs encoding hsp70 and c-fos (34) ; the leader sequence of both these RNAs contains a single element of complementarity to 18S rRNA and promotes initiation downstream via shunting and scanning mechanisms (34).

One unconfirmed report suggested an "internal ribosome repositioning" mechanism to explain differential expression of Mycl and Myc2 isoforms in the human c-myc 5'-UTR (141) . However, alternative mechanisms such as internal initiation have not been ruled out (142, 143) .

After translating an upstream ORF of more than about 30 codons, the terminating ribosome is believed to dissociate from the eukaryotic mRNA in a scanning-and reinitiation-incompetent state. Although we have some knowledge of the first steps of the termination process in eukaryotes and the factors associated with it, our understanding of termination is far from complete. The ribosomal release factor (RRM), which triggers 50S or 70S ribosome release from the mRNA in prokaryotes (144), is not found in eukaryotes. Thus, the 22 RYABOVA ET AL.

fate of the eukaryotic ribosome after translation termination remains unclear. However, following sORF translation, some ribosomes can be assumed to remain associated with the mRNA, continue scanning, and reinitiate. Very little is known about factor requirements for reinitiation of translation. The GCN4 case clarified the role ofeIF2 in recognition of the second start site for the reinitiating 40S subnnit (27) . These experiments also suggest that eIF2 is not essential for the 40S scanning process in yeast, eIF3 would apparently also be required to promote binding of the ternary complex to the 40S ribosome. The role of eIF3 and other canonical initiation factors in reinitiation remains to be clarified.

Several groups have reported the existence of potentially bicistronic gene structures in eukaryotes, which could require reinitiation steps for their expression. Phylogenetic analysis of the SNRPN (small nuclear ribonucleoprotein N, StuN) mRNA reveals a highly conserved coding sequence SNURF (SNRPN upstream reading frame, 71 amino acids in length) upstream of the SNRPN ORF; both SNURF and SNRPN are produced from a bicistronic transcript in normal human and mouse tissues (145) . In mammals, active LINE 1 elements of retroviral origin produce two proteins from a single transcript (146) (147) (148) . Several bicistronic messages have been reported in Drosophila melanogaster, at the stoned locus (149), at the Adh locus (150) , and from the mei-gl8 gene (151) .

Interestingly, the overlapping stop-start codon UGAUG, found in embryonic exons as a result of alternative splicing during development, converts the monocistronic adult-type message encoding glutamie acid decarboxylase (GAD) into a bicistronic one coding for a 25-kDa leader peptide and a 44-kDa enzymatically active truncated GAD (152) . It remains to be seen if any of these mRNAs use internal initiation, reinitiation, or leaky scanning. Also, for each case of a potentially polycistronic RNA, careful analysis of additional promoter regions, splice sites, and degradation mechanisms is required (153) .

The best-studied examples of polycistronic mRNAs are found in the plant viral family Caulimoviridae, which contain up to seven long ORFs within a single RNA. However, the strategy employed to effect polycistronic translation differs in RTBV-like viruses, which use a leaky scanning mechanism, and CaMV-like viruses, which use a reinitiation mechanism activated by a viral protein.

The pregenomie RNA of RTBV and other bacilliform Caulimoviridae contains four or three main ORFs, respectively, and serves for translation of at least three proteins, the products of ORFs I, II, and III (5) . In RTBV, the splicing event between the first sORF and ORF IV provides a mRNA for the production of the ORF IV protein (154) (see Fig. 1 ). The first ORF recognized after shunting in RTBV is ORF I, which begins with the non-AUG start codon AUU (70) .

According to the Kozak scanning model, this non-AUG eodon could be easily VIRAL STRATEGIES OF TRANSLATION INITIATION 23 bypassed by linear migrating ribosomes or shunting ribosomes landing upstream ofORF I. A remarkable property of ORFs I and II is the complete lack of AUGs in any reading phase to impede scanning ribosomes and preclude downstream initiation. The start codon of ORF II is in an unfavorable sequence context and about half of the scanning ribosomes ignore it and initiate at the start codon of ORF III, which is in a strong initiation context (5) . Thus, ribosomes loaded at the capped 5'-end of the RTBV RNA first shunt the leader sequence and then scan over a distance of about 900 nt upstream of ORF III, confirming the reported high processivity of scanning ribosomes in vitro (155) . Mutation of the ORF I AUU codon to create a strong initiation codon leads to a drastic decrease in expression of ORFs II and III in vivo and in vitro (5) .

The pregenomic RNA of CaMV and its internally spliced derivatives (99) serve as polycistronic mRNAs for a number of viral proteins. There are several lines of evidence that these polycistronic RNAs are indeed used for viral protein production in CaMV-infected plants. First, only one subgenomic RNA, the 19S RNA encoding the transactivator/viroplasmin protein (TAV), has been identified. The other ORFs are translated from the 35S RNA or its spliced versions. Second, ORFs I-V are tightly arranged and often overlap by a single base, suggesting that their translation might be linked. Indeed, some mutations within ORFs VII and II are polar, typical for polycistronic prokaryotic mRNAs, meaning that they affect translation of the following ORF (156, 157) .

Translation of the 35S RNA is 5'-end-dependent and the first translatable ORF is sORF A within the leader. Thus, translation of major ORFs on the CaMV RNA requires reinitiation. Reinitiation on the polycistronic 35S RNA and its spliced versions is activated in the presence of TAV (59, 98, 99) , which is expressed early in infection. Transient expression of derivatives of the CaMV genome, each containing the upstream sequence of a specific ORF fused to the CAT or GUS reporter genes, was shown in plant protoplasts (98) . A reporter gene fusion to the start codon of ORFs VII and II was expressed at 15% of the level of the monocistronic control, but at 100% if a second plasmid expressing the TAV ORF was provided; fusions to the start codons of ORFs I, III, IV, and V were expressed only in the presence of TAV (100% for ORF I, 33% for ORF III, 50% for ORF IV, and 3% for ORF V). TAV-mediated transactivation of expression of all the major ORFs on the polycistronic pregenomic RNA has been also reported for figwort mosaic virus (FMV; [158] [159] [160] and peanut chlorotic streak virus (161) , and TAV of CaMV and FMV can reciprocally activate polycistronic translation in these viruses (158) .

RYABOVA ET AL.

The use of partially or completely artificial constructs has confirmed the role of TAV in general activation of polycistronic translation (19, 45) . When bicistronic constructs, for example, consisting of ORFs VII and CAT, or GUS and CAT, were analyzed in plant protoplasts, the level of expression of the second ORF was always under TAV control, whereas expression of the first ORF was not affected by TAV. Specific c/s sequence signals are not required for transactivation of second ORF translation, since TAV can activate reinitiation after translation of any first ORF in an artificial bicistronic RNA (19, 45) . Notably, the presence of a sORF (optimally around 30 codons) upstream of both long ORFs strongly enhances the process, activating expression of the first ORF 2-fold and the second 8-to 10-fold. A long overlap of the major ORFs (130 nt) inhibits transactivation, whereas a short overlap (17 nt) is permissible. In caulimoviruses a short overlap between two ORFs, via an AUGA, is very common, although expression of the downstream ORF is TAV-dependent as well (19) . A stem structure at the cap site inhibits expression of both reporters, whereas a stem between the two ORFs inhibits only expression of the second (19) .

Similar results were obtained in yeast when the 35S RNA promoter was used to direct bicistronic RNA transcription (162) and in transgenic plants (163) .

Transgenic plants expressing TAV exhibit chlorosis and other virus-like symptoms (164) (165) (166) (167) (168) . TAV expression in these plants also correlates with changes in plant morphology and development, suggesting that TAV might transactivate translation from as-yet-unknown complex mRNAs encoding regulatory proteins involved in development.

TAV is mainly expressed from the 19S subgenomic RNA, but it is also able to transactivate its own expression from the 35S RNA (169, 170) . TAV is very abundant in the cytoplasm of infected cells and forms a dense matrix, the so-called inclusion bodies or viroplasm. In early phases of infection, these are surrounded by polyribosomes (171) (172) (173) . These "tethered" ribosomes could direct CaMV translation products into the inclusion bodies, where all proteins expressed from CaMV RNAs are found (174, 175) , including heterologous nonviral proteins encoded by transgenic CaMV RNA (176) .

TAV has many functions in the life cycle of the virus. In addition to polycistronic translation, it controls virus assembly and replication and determines the host range (see Ref. 177 for review). Virion assembly might be guided by the interaction of TAV with coat protein (178) . Moreover, it appears that TAV is involved in the stabilization of other viral gene products (179) . polypeptide. In addition to the defect in transactivation, MAV has so far only been found to be active in Nicotiana plumbaginifolia protoplasts, in contrast to full-length TAV, which is active in protoplasts of a number of dicot and monocot plants (59, 158) . The C-terminal part ofTAV (CTAV; Fig. 4 ) can efficiently inhibit the transactivation activity of the entire protein, suggesting the CTAV is able to sequester host factors that are essential for transactivation activity (180) .

Notably, the N-terminal region located upstream of the MAV domain was found also to affect TAV-mediated transactivation activity (181) . TAV transactivates polycistronie translation by association with host translational machinery, namely polysomes and associated proteins. Indeed, TAV was found in polysomes isolated from CaMV-infected turnip plants, and it cosediments with polysomes isolated from healthy turnip plants (182) . This interaction is apparently mediated by interaction with key components of the host translational machinery, namely the 60S large ribosomal subunit via at least two ribosomal proteins, L24 (182) and L18 (183) , and with initiation factor eIF3 via its subunit g (182) . The L18 ribosomal protein interacts with the MAV domain of 26 RYABOVA ET AL.

NTAV, while both the central segment of elF3g, including the Zn-finger motif, and the N-terminal half of ribosomal protein L24, compete for the same binding site within one of the RNA-binding domains (RBa) of CTAV (Fig. 4) .

It is remarkable that both ribosomal subunits and elF3 can be found in a complex with TAV in vitro, but TAV association with 40S is indirect, that is, mediated by elF3. In contrast, elF3 association with 60S is mediated by TAV. Thus, the existence of two ternary complexes, TAV/elF3/40S and elF3/TAV/60S, might play an essential role in the transactivation process. Significant accumulation of elF3 in polysomes isolated from CaMV-infected plants correlates well with TAV accumulation, whereas only traces of elF3 were found in polysomes isolated from healthy turnip plants (182) .

Transient overexpression of elF3g has a strong negative effect on TAVmediated polycistronic translation in plant protoplasts (182) , which correlates well with the reported in vitro interaction between TAV and complete eIF3, suggesting that a TAV/eIF3 complex is active in transactivation. Archaebacterial homologs of L24 and L18 are located on the interface and the external surface of the 60S ribosome, respectively (184, 185) . In eukaryotes, the location of L24 in the main factor-binding site on the interface of the 60S ribosome has also been reported (186; R. Beckmann, personal communication).

This suggests that, in the eukaryotic 60S ribosome, these proteins are located too far apart to interact with the same molecule of TAV. Thus, the 60S ribosome is presumably capable of binding two TAV molecules simultaneously on its external and internal surfaces. These observations can explain why transient overexpression of L24 has a strong positive effect on TAV-mediated polycistronic translation in plant protoplasts (182) . Interaction of TAV with the main factor-binding site on the 60S ribosome via L24 might indeed lead to inhibition ofprotein synthesis during the late phase of viral infection, promoting a switch to viral assembly. On the other hand, L24 might have an extraribosomal function and a separate protein might be involved in 60S subunit turnover, as shown previously for yeast L24 (187) .

The ribosomal protein L18, due to its location on the external surface of the 60S ribosome, might interact with other partners such as TAV without affecting 80S ribosome formation.

The accumulation of TAV as the main matrix protein in viral inclusion bodies together with the coaccumulation of proteins derived from the polycistronic CaMV RNA suggests that TAV-enhanced translation occurs on the inclusion body surface. The TAV molecules that form the inclusion bodies might present a perfect surface for ribosome and eIF3 recruitment; the concomitant increase in local concentration of these factors might contribute to the increased reinitiation rate observed in the presence of TAV. RNA-binding activities of TAV, mediated by its ssRNA-and DNA-binding domains within its C-terminal part (180) , and the dsRNA-binding domain within the MAV region (188) might further enhance local TAV concentration in polysomes (Fig. 4 ).

Step 2

Step 3 $ FIG. 5. Model of TAV function during translation of the polycistronic RNA. Steps 1-3 are described in the text. elF3, its subunit g, elF1 (1), elF2 (2), eIF5 (5), tRNA, L24 (24), L18 (18) , and TAV are indicated. Recycling of elF2 is shown by a dotted arrow.

Despite the abundance of data on TAV molecular interactions with the host translational machinery, we do not yet understand the exact mechanism of TAV-mediated reinitiation of translation. However, the accumulated data allow us to offer the following three-step working model (Fig. 5) .

Step 1. Primary TAV acquisition by elF3. Several observations suggest that TAV is preferentially sequestered by the translational machinery during sORF translation. (1) Ribosomes often translate an sORF and then reinitiate efficiently downstream of it (44, 28) . This suggests that some of the eIFs are still associated with the translational machinery after sORF translation, but not after translation of larger ORFs. (2) Polycistronic translation under TAV control is enhanced by an sORF (or CaMV ORF VII) positioned in the leader (19, 45) . interacts with eIF3 as well as with eIF3 bound to the 40S ribosome (182) .

(4) elF3 might be associated with the 40S ribosome without interfering with the 80S ribosome formation (189) . We assume that after sORF translation, the 40S ribosome including certain eIFs remains loosely attached to the mRNA and can resume scanning. We further speculate that eIF3 is one of the remaining factors, whereas elF2/GDP, and perhaps also elF4F and eIF4B, are preferentially released, elF2/GTP/Met-tRNAi can then be reacquired after resumption of scanning via the bridging function of elF5 interacting with elF3c (190) and TAV could be acquired through its association with eIF3 via its subunit g, perhaps even docking at the same position as other elF3-interacting proteins. Yeast elF3g interacts with elF4B (191) , and we have confirmed that this interaction also occurs in plants (H.-S. Park, T. Hohn, and L. A. Ryabova, unpublished) ; TAV could thus be sequestered by elF3g upon release of elF4B.

Step 2. Preservation of reinitiation-competent ribosomes during translation of the first large ORF. During translation of a longer ORF in the absence of TAV, eIF3 will be released during the translation elongation step, leaving ribosomes incompetent for reinitiation. In the presence of TAV, however, the interaction of elF3 with the translating ribosome might be stabilized. We do not know in this case whether eIF3 holds its original position or is moved to a new location within the translation machinery. An exciting possibility is that elF3 is delivered to the 60S ribosome via the L18 ribosomal protein, which is located at the 60S external surface. We were able to show the existence of both TAV/elF3/40S and eIF3/TAV/60S complexes in vitro. This raises the possibility that elF3 is transferred to the 60S ribosome apparently via the L18/TAV interaction. We do not think that the alternative L24/TAV interaction is involved at this step, since L24 and elF3g compete for the same binding site on TAV and L24 is located at the 60S/40S interface.

Step 3. Reinitiation at the second large ORF. We postulate that during the translation termination process, the binding site on the 40S ribosome for elF3 becomes available again and elF3 could be transferred back to this site, leading to reacquisition of elF2/GTP/Met-tRNAi and scanning to find the second ORF start codon.

The translation process is the most important step in transforming genetic into functional information. The interplay of initiation factors and regulatory regions on the leaders and trailers of RNA is the most important parameter controlling translation, both quantitatively and qualitatively. Although the basic translation initiation factors and some details of the translation process are known, control of translation initiation varies from case to case and is controlled VIRAL STRATEGIES OF TRANSLATION INITIATION 29 by environmental factors, much as is transcription. Scientists try to find rules that make the behavior of biological entities and processes predictable. However, also in biology, rules are there to be broken and often the exceptions lead to deeper understanding of the rule. Viruses are among the best model organisms for studying these parameters due to the rapid evolution of their translation control mechanisms and the high density of regulatory sequences located on their genome. Members of the Caulimoviridae are excellent examples of such viral rule-breakers, circumventing linear scanning by shunting and allowing reinitiation by interactions of a viral protein with the translational machinery. The future will tell whether similar mechanisms occur also during translation of cellular messenger RNAs.

The Strategies of plant ~rus gene expression: models of economy

Classification of reverse transcribing elements: a discussion document

Retroviruses

Petunia vein clearing virus (PVCV): a potential retrovirus in petunia

Rice tungro bacilliform virus open reading frames II and III are translated from polycistronic pregenomie RNA by leaky scanning

Comparative view of initiation site selection mechanisms

The scanning model for translation an update

Initiation of translation in prokaryotes and eukaryotes

Pathway and mechanisms ofinitiation ofprotein synthesis

Molecular mechanisms of translation initiation in eukaryotes

Cloning and characterization of the p42 subunit of mammalian translation initiation factor 3 (eIF3): demonstration that elF3 interacts with elF5 in mammalian cells

Eukaryotic ribosomes require initiation factors 1 and 1A to locate initiation codons

The joining of ribosomal subunits in eukaryotes requires elF5B

The plant translational apparatus

Translation in plants rules and exceptions

Internal binding of eukaryotic ribosomes on poliovirus RNA: Translation in HeLa cell extracts

Influence of mRNA secondary structure on initiation by eukaryotic ribosomes

Circumstances and mechanisms of inhibition of translation by secondary structure in eukaryotic mRNAs

Translation of a polycistronic mRNA in presence of the cauliflower mosaic virus transactivator protein

Context effects and inefficient initiation at non-AUG eodons in eukaryotic cell-free translation systems

Regulation of translation in eukaryotic systems

Proteins binding to 5' untranslated region sites: a general mechanism for translational regulation of mRNAs in human and yeast cells

Ribosomal pausing and scanning arrest as mechanisms of translational regulation from cap distal iron responsive elements

Internal initiation of translation of eukaryotic mRNA directed by a sequence derived from poliovirus RNA

Initiation of encephalomyocarditis virus RNA translation: the authentic initiation site is not selected by a scanning mechanism

Dualinitiation sites of protein synthesis on foot and mouth disease virus RNA are selected following internal entry and scanning of ribosomes in vivo

Translational regulation of yeast GCN4. A window on factors that control initiator-trua binding to the ribosome

Upstream open reading frames as regulators of mRNA translation

Scanning independent ribosomal initiation of the Sendai virus X protein

Positive and negative control of translation by the leader of cauliflower mosaic virus pregenomic 35S RNA

Nonlinear ribosome migration on cauliflower mosaic virus 35S RNA

Scanning independent ribosomal initiation of the Sendai virus Y proteins in vitro and in vivo

Selective translation initiation by ribosome jumping in adenovirus infected and heat-shocked cells

Translationbyribosome shuntingonadenovirus andhsp70mRNAs facilitated by complementarity to 18S rRNA

Translation initiation at a downstream AUG occurs with increased efficiency when the upstream AUG is located very close to the 5' cap

A short leader sequence impairs the fidelity of initiation by eukaryotic ribosomes

Cell-specific translational regulation of S adenosylmethionine decarboxylase mRNA. Influence of the structure of the 5' transcript leader on regulation by the upstream open reading frame

Downstream secondary structure facilitates recognition of initiator codons by eukaryotic ribosomes

Internal initiation of translation in eukaryotes: the picornavirns paradigm and beyond

Internal ribosome initiation of translation and the control of cell death

Functional dissection of eukaryotic initiation factor 4F: the 4A subunit and the central domain of the 4G subunit are sufficient to mediate internal entry of 43S preinitiation complexes

Picornavirns RNA translation: roles for cellular proteins

Internal ribosome entry site within hepatitis C virus RNA

Effects of intereistronic length on the efficiency of reinitiation by eukaryotic ribosomes

Role of an upstream open reading frame in the translation of polycistronic mRNA in plant cells

Efficiency of reinitiation of translation on human immunodeficiency virus type 1 mRNAs is determined by the length of the upstream open reading frame and by intercistronic distance

Sequences that surround the stop codons of upstream open reading frames in GCN4 mRNA determine their distinct functions in translation control

Ribosome regulation by the nascent peptide

Cell specific translational regulation of S-adenosylmethionine decarboxylase mRNA. Dependence on translation and coding capacity of the cis-acting upstream open reading frame

Making sense of nonsense in yeast

Post termination ribosome interactions with the 5 IUTR modulate yeast mRNA stability

Effect of upstream reading frames on translation efficiency in simian virus 40 reeombinants

Termination reinitiation occurs in the translation of mammalian cell mRNAs

Evidence that the human foamy virus genome is DNA

Foamy viruses are unconventional retroviruses

Foamy viruses: between retroviruses and pararetroviruses

Translational frameshifting: implicatinos for the mechanism of translational frame maintenance

Mechanism of translation of monocistronic and multicistronic human immunodeficiency virus type 1 mRNAs

Posttranscriptional transactivation in cauliflower mosaic virus

Evidence for the involvement of a ribosomal leaky scanning mechanism in the translation of the hapatitis B virus pol-gene from the viral pregenome RNA

Translation of the hepatitis B virus P gene by ribosomal scanning as an alternative to internal initiation

Translational regulation of hepatitis B virus polymerase gene by termination-reinitiation of an upstream minicistron in a length dependent manner

Internal entry of ribosomes and ribosomal scanning involved in HBV P gene expression

Mechanism of translation of the hepadnaviral polymerase (P) gene

Translational inactivation of RNA function: Diserimination against a subset of genomic transcripts during HBV nucleocapsid assembly

Expression of functional hepatitis B virus polymerase in yeast reveals it to be the sole viral protein required for correct initiation of reverse transcription

Formation of a functional hepatitis B virus replication initiation complex involves a major structural alteration in the RNA template

Effects of genomic length on translocation of hepatitis B virus polymerase-hnked oligomer

The encapsidation signal of hepatitis B virus facilitates preC AUG recognition resulting in inefficient translation of the downstream genes

Position-dependent ATT initiation during plant pararetrovirus Rice Tungro Bacilliform Virus translation

Interaction of the cauliflower mosaic virus coat protein with the pregenomic RNA leader

A short open reading frame terminating in front of a stable hairpin is the conserved feature in pregenomic RNA leaders of plant pararetroviruses

Metabolism of viral RNAin murine leukemia virus infected cells; evidence for differential stability of viral message and virion precursor RNA

Comparison of viral genomic RNA sorting mechanisms in human immunodeficiency virus type 1 (HIV-1), HW-2, and Moloney murine leukemia virus

Structure function relationship of Rous sarcoma virus leader RNA

Role of the open reading frames of Rous sarcoma virus leader RNA in translation and genome packaging

Effects of the open reading frames in the Rous sarcoma virus leader RNA on translation

The first and third uORFs in RSV leader RNA are efficiently translated: implications for translational regulation and viral RNA packaging

Autogenous regulation or RNA translation and packaging by Rous sarcoma virus Pr76gag

Rous sarcoma virus translation revisited: characterization of an internal ribosome entry segment in the 5' leader of the genomic RNA

Ribosome shunting in the cauliflower mosaic virus 35S RNA leader is a special case of reinitiation of translation functioning in plant and animal systems

An internal ribosome entry segment promotes translation of the simian immunodefieiency virus genomie RNA

The human immunodeflciency virus type 1 gag gene encodes an internal ribosome entry site

An internal ribosomal entry mechanism promotes translation of murine leukemia virus gag polyprotein precursors

An internal ribosomal entry signal in the rat VI_3O region of the Harvey murine sarcoma virus leader and its use in dicistronic retroviral vectors

Characterization of an internal ribosomal entry segment within the 5' leader of avian reticuloendotheliosis virus type A RNA and development of novel MLV-REV-based retroviral vectors

The human immunodeficiency virus type 1-5' packaging signal structure affects translation but does not function as an internal ribosome entry site structure

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

Maintenance of the Gag/Gag-Pol ratio is important for human immunodeficiency virus type 1 RNA dimerization and viral infectivity

Programmed translational frameshifting

Recoding: dynamic reprogramming of translation

Translation of MuLV and MSV RNAs in nuclease treated reticuloeyte extracts: enhancement of the gag-pol polypeptide with yeast suppressor tRNA

Readthrough suppression in the mammalian type C retroviruses and what it has taught us

Drop off during ribosome hopping

Biosynthesis of the reverse transcriptase of hepatitis B viruses involves de novo translational initiation not ribosomal frameshifting

Synthesis and encapsidation of duck hepatitis B virus reverse transcriptase do not require formation of core polymerase fusion proteins

The reverse transcriptase gene of CaMV is translated separately from the capsid gene

Expression from polycistronic cauliflower mosaic virus pregenomic RNA

Splicing of Cauliflower Mosaic Virus 35S RNA is essential for viral Infectivity

Detection of a subgenomic mRNA for gene V, the putative reverse transcriptase gene of cauliflower mosaic virus

Foamy virus reverse transcriptase is expressed independently from the Gag protein

Human foamy virus replication: a pathway distinct from that of retroviruses and hepadnaviruses

Cassava Vein mosaic virus (CsVMV), type species for a new genus of plant double stranded DNA viruses?

Novel gene expression mechanism in a fission yeast retroelement: Tfl proteins are derived from a single primary translation-product

Alternative structures of the cauliflower mosaic virus 35 S RNA leader: implications for viral expression and replication

Differential inhibition of downstream gene expression by the CaMV 35S RNA leader

Plant ribosome shunting in vitro

Termination and peptide release at the upstream open reading frame are required for downstream translation on synthetic shuntcompetent mRNA leaders

Nonlinear ribosome migration on cauliflower mosaic virus 35S RNA in transgenic tobacco plants

Ribosome shunting in cauliflower mosaic virus. Identification of an essential and sufficient structural element

Forced evolution reveals the importance of short open reading frame A and secondary structure in the cauliflower mosaic virus 35S RNA leader

Ribosome shunt is essential for infectivity of cauliflower mosaic virus

Role of a short open reading frame in ribosome shunt on the cauliflower mosaic virus RNA leader

Shunting is a translation strategy used by plant pararetroviruses (Caulimo~ridae)

Regulation of CaMV 35 S RNA translation is mediated by a stable hairpin in the leader

A stable hairpin preceded by a short open reading frame promotes nonlinear ribosome migration on a synthetic mRNA leader

Unique features of internal initiation of hepatitis C virus RNA translation

Cell-specific translation of S-adenosylmethionine decarboxylase mRNA. Regulation by the 5' transcript leader

Continuous and discontinuous ribosome scanning on the cauliflower mosaic virus 35 S RNA leader is controlled by short open reading frames

Secondary structure analysis of adenovirus tripartite leader

Translation by the adenovirus tripartite leader: elements which determine independence from cap-binding protein complex

Identification of two short internal ribosome entry sites selected from libraries of random oligonucleotides

A late adenovirus factor induces elF 4E dephosphorylation and inhibition of cell protein synthesis

Adenovirus-speciflc translation by displacement ofkinase Mnkl from cap initiation complex elF4F

Adenovirus inhibition of cellular protein synthesis and preferential translation of viral mRNAs

Replication of paramyxoviruses

8endai virus contains overlapping genes expressed from a single mRNA

ACG, the initiator codon for a Sendai virus protein

Translation initiation potential of the 5' proximal AUGs of the polycistronic P/C mRNA of Sendal virus. A multipurpose vector for site specific mutagenesis

Sendai virus Y proteins are initiated by a ribosomal shunt

Ribosome pausing and stacking during translation of a eukaryotic mRNA

Human Papillomaviruses and Cervical Cancer

E2 of cottontail rabbit papiilomavirus is a nuclear phosphoprotein translated from an mRNA encoding multiple open reading frames

Human papiilomavirus type 18 E6*, E6, and E7 protein synthesis in cell free translation systems and comparison of E6 and E7 in vitro translation products to proteins immunoprecipitated from human epithelial ceils

Translation of the human papiUomavirus type 16 E7 oncoprotein from bicistronic mRNA is independent of splicing events within the E6 open reading frame

Human papillomavirus type 18 E1 protein is translated from polycistronic mRNA by a discontinuous scanning mechanism

Leaky scanning is the predominant mechanism for translation of human papillomavirus type 16 E7 oncoprotein from E6/E7 bicistronic mRNA

Cap and polyA tail enhance translation initiation at the hepatitis C virus internal ribosome entry site by a discontinuous scanning, or shunting, mechanism

Construction ofregulatable picornavirus IRESes as a test of current models of the mechanism of internal translation initiation

Downstream ribosomal entry for translation of coronavirus TGEV gene 3b

Differential expression of Myel and Myc2 isoforms in cells transformed by eIF4E: evidence for internal ribosome entry site

Alternative translation of the proto oncogene c-myc by an internal ribosome entry site

Translation of the human c-mye P0 tricistronic mRNA involves two independent internal ribosome entry sites

Evidence for in vivo ribosome recycling, the fourth step in protein biosynthesis

An imprinted, mammalian bicistronic transcript encodes two independent proteins

Translation of the rat LINE bicistronie RNAs in vitro involves ribosomal reinitiation instead of frameshifting

Translation of the human LINE-1 element

The full length transcript of the I factor, a LINE element of Drosophila melanogaster, is a potential bicistronic RNA messenger

The stoned locus of Drosophila melanogaster produces a dieistronic transcript and encodes two distinct polypeptides

The Adh-related gene of Drosophila melanogaster is expressed as a functional dicistronic messenger RNA: multigenic transcription in higher organisms

Two genes required for meiotic recombination in Drosophila are expressed from a dicistronic message

Distinct protein forms are produced from alternatively spliced bicistronic glutamic acid decarboxylase mRNAs during development

New ways of initiating translation in eukaryotes?

Splicing in a plant pararetrovirns

Primer extension analysis of eukaryotic ribosome mRNA complexes

Initiation of translation of the cauliflower mosaic virus genome from a polycistronic mRNA: evidence from deletion mutagenesis

The molecular buiology of cauliflower mosaic virus and its application as plant vector

Gene VI of flgwort mosaic virus (caulimo virus group) functions in posttranscriptional expression of genes on the full length RNA transcript

Gene VI of figwort mosaic virus activates expression of internal cistrons of the full-length polycistronic transcript

The fun-length transcript of a caulimovirns is a polycistronic mRNA whose genes are transactivated by the product of gene VI

Gene expression regulated by gene VI of caulimovirus: transactivation of downstream genes of transcripts by gene VI of peanut chlorotic streak virus in transgenic tobacco

Expression of a plant viral polycistronic mRNA in yeast, Saccharomyces cerivisiae, mediated by a plant virus translation transactivator

Cauliflower mosaic virus gene VI controls translation from dicistronic expression units in transgenic arabidopsis plants

Diseases symptoms in transgenic tobacco induced by integrated gene VI of cauliflower mosaic virus

Cauliflower mosaic virus gene VI produces a symptomatic phenotype in transgenie tobacco plants

A disease syndrome associated with expression of gene VI of caulimovirus may be a non host reaction

Cauliflower mosaic virus gene VI causes growth suppression, development of necrotic spots and expression of defence-related genes in transgenic tobacco plants

Arabidopsis thaliana expressing the Cauliflower Mosaic Virus ORF VI transgene has a late flowering phenotype

Regulation of canlimovirus gene expression and the involvement of cis-acting elements on both viral transcripts

Transcription from the CaMV 19S promoter and autocatalysis of translation from CaMV RNA

Infection of turnip brassica-rapa oultivar perviribis protoplasts with cauliflower mosaic virus

Fine structure of zinnial leaf tissues infected with dahlia mosaic virus

Ultrastructure ofcarnation etched ring virus-infected Saponaria vaccaria and Dianthus caryophyllus

A second cauliflower mosaic virus gene product influences the structure of the viral inclusion body

Protein encoded by ORFI of cauliflower mosaic virus is part of the viral inclusion body

The expression, localization, and effect of a human interferon in plants

Rararetroviruses and retroviruses: a comparative review of viral structure and gene expression strategies

Interaction between Cauliflower mosaic virus inclusion body protein and capsid protein implications forviral assembly

Requirement of cauliflower mosaic virus open reading frame VI product for viral gene expression and multiplication in turnip protoplasts

Molecular dissection of the cauliflower mosaic virus translational transactivator

Mutational analysis of cauliflower mosaic virus gene VI: changes in host range, symptoms, and discovery of transactivation-positive, noninfectious mutants

A plant viral "reinitiation" factor interacts with the host translational machinery

The cauliflower mosaic virus translational transactivator interacts with the 693 ribosomal subunit protein L 18 of Arabidopsis thaliana

Primary structures of five ribosomal proteins from the archaebacterium. Halobacterium marismortui and their structural relationship to eubacterial and eukaryotic ribosomal proteins

The complete atomic structure of the large ribosomal subunit at 2.4 A resolution

Localization of ribosomal proteins on the surface of mammalian 60S ribosomal subunits by means of immobilized enzymes. Correlation with chemical crosslinking data

Ribosomal protein 1_30 is dispensable in the yeast Saccharomyces cerevisiae

A common 40 amino acid motif in eukaryotic RNases H-1 and caulimovirus ORF VI proteins binds to duplex RNAs

Eukaryotic initiation factor 3 does not prevent association through physical blockage of the ribosomal subunit-subunit interface

A multifactor complex of eukaryotic initiation factors, eIFl, elF2, eIF3, elF5, and initiator tRNA(Met) is an important translation initiation intermediate in vivo

A 110-kilodalton subtmit of translation initiation factor eIF3 and an associated 135-kilodalton protein are encoded by the Saccharomyces cerevisiae TIF32 and TIF31 genes

Hepatitis B virus replication