Structural variation and functional importance of a D-loop–T-loop interaction in valine-accepting tRNA-like structures of plant viral RNAs

Abstract

Valine-accepting tRNA-like structures (TLSs) are found at the 3′ ends of the genomic RNAs of most plant viruses belonging to the genera Tymovirus, Furovirus, Pomovirus and Pecluvirus, and of one Tobamovirus species. Sequence alignment of these TLSs suggests the existence of a tertiary D-loop–T-loop interaction consisting of 2 bp, analogous to those in the elbow region of canonical tRNAs. The conserved G₁₈·Ψ₅₅ pair of regular tRNAs is found to covary in these TLSs between G·U (possibly also modified to G·Ψ) and A·G. We have mutated the relevant bases in turnip yellow mosaic virus (TYMV) and examined the mutants for symptom development on Chinese cabbage plants and for accumulation of genetic reversions. Development of symptoms is shown to rely on the presence of either A·G or G·U in the original mutants or in revertants. This finding supports the existence and functional importance of this tertiary interaction. The fact that only G·U and A·G are accepted at this position appears to result from steric and energetic limitations related to the highly compact nature of the elbow region. We discuss the implications of these findings for the various possible functions of the valine-accepting TLS.

INTRODUCTION

The genomic RNAs of various genera of positive strand RNA plant viruses carry a structure at their 3′ end that resembles tRNA both structurally and functionally (1–3). While these structures are heavily conserved within each genus, the degree of structural similarity to tRNA can vary dramatically between genera. On the other hand, similar tRNA-like structures (TLSs) are found in virus genera that otherwise show little relationship. The tRNA-like feature may thus have evolved independently in different virus groups and subsequently spread horizontally by RNA recombination (2).

Various functions have been ascribed to these TLSs, including a telomere-like function that ensures the integrity of the conserved 3′-terminal sequence CCA, a role in the specific binding of the viral RdRp complex, a function as the main recognition site for the viral coat protein and a role in the functional switch of the RNA from a template for translation to one for replication. The experimental data presently available suggest that a single TLS may perform multiple functions, but that the relative importance for these various functions depends on the species (2,3).

While virtually all TLSs studied so far can be aminoacylated in vitro and are indeed aminoacylated in vivo, the degree to which they appear to mimic tRNA to the human eye varies. The structures of the histidine- (Tobamovirus) and tyrosine-accepting (Bromovirus, Cucumovirus and Hordei virus) TLSs contain several additional stems and loops as compared with tRNA, while the equivalents of the canonical D- and T-loops appear severely degenerate, if present at all. Much closer in overall structure to normal tRNAs are the 3′ ends of the valine-accepting RNAs, which belong to the genera Tymovirus, Furovirus, Pomovirus and Pecluvirus. The main peculiarity of all these TLSs is the fact that the acceptor stem is constructed as a pseudoknot, such that the remainder of the genomic RNA is not attached opposite the 3′-terminal CCA sequence, but between the D-stem and the anticodon stem (Fig. 1).

Figure 1.

Secondary structure model for the TLS of TYMV. For colors see the legend to Figure 2. Potential tertiary interactions as discussed in the text are indicated with grey bars.

The fact that the degree of overall similarity of a TLS to tRNA appears to be coupled to the nature of the attached amino acid may suggest a direct relation to the properties of the plant aminoacyl-tRNA synthetases that perform the aminoacylation. Sunn-hemp mosaic virus (SHMV), a virus belonging to the genus Tobamovirus, appears to have picked up a TLS of the Tymovirus type. Indeed, its RNA is not histidylated like the other Tobamoviruses, but valylated (4). This example shows that the detailed nature of the TLS is not imposed by the RdRp, which is still of the Tobamovirus type.

To understand better the role of the TLS and the nature of its molecular interactions during the life cycle of the virus, it is interesting to examine in closer detail the nature of the tRNA mimicry. The fact that the structure of cytoplasmic tRNAs is conserved to an extreme degree whereas TLSs do not mimic all the details may reflect that only a subset of the functions and interactions of normal tRNAs are employed by the viral RNA. By defining which elements of the structure are mimicked and which are not, and relating this to what is known about their roles in the interactions of tRNA with its various partners, one can gain information on which of these interactions are employed by the plant viral TLSs. In other words, the functions of the TLSs may be deduced from the selection pressures that they have been subject to.

In the present paper, we focus on a highly conserved element of tRNA structure that appears to be partly, but not totally, mimicked by the valine-accepting plant viral RNAs. In virtually all known cytoplasmic tRNAs, the outer corner of their L-shape (the elbow) is formed by a 2 bp interaction between the D-loop and the T-loop (see Fig. 4). This interaction consists of a somewhat distorted Watson–Crick G–C pair (G₁₉–C₅₆) and a bifurcated G·Ψ pair (G₁₈·Ψ₅₅). The two (parallel) loop–loop base pairs form an extension of the acceptor arm by stacking on the intraloop pair T₅₄–A₅₈, which in turn stacks upon the T-stem (5). The conserved purine at position 57 intercalates between the two loop–loop base pairs. Studies on in vitro transcribed tRNAs have shown that pseudouridine Ψ₅₅ can be replaced with a uridine without significant consequences for the structure and the function of the tRNA (6,7).

Figure 4.

Crystal structure of the elbow region of yeast tRNA^Phe (PDB ID code 1EHZ). (Left) Overall structure of the tRNA backbone (blue, 5′ half; green, 3′ half) with nucleotide structures shown in the elbow region only. (Right) Blow-up of the elbow region, showing from left to right the base stacking of G₁₉–C₅₆, G₅₇, G₁₈·Ψ₅₅ (yellow), T₅₄·m¹A₅₈ and G₅₃–C₆₁. The D-loop backbone is shown in blue, the T-loop backbone in green.

By aligning the sequences of the valine-accepting plant viral TLSs, we show that the possibility to form the equivalent of G₁₉–C₅₆ is universally conserved among these viruses. The equivalent of G₁₈·Ψ₅₅, on the other hand, is found as either G·U or A·G. These two base–base combinations were not previously found as a covariation in other RNA molecules, nor was it a priori evident that they can form isosteric base pairs. We demonstrate their functional equivalence in vivo by showing that the development of symptoms in plants inoculated with mutant turnip yellow mosaic virus (TYMV) RNA is dependent on this pair being either A·G (‘wild-type’) or G·U (‘even more tRNA-like’). Other mutants either never developed symptoms or reverted to one of the two functional genotypes. We discuss the structural and functional implications of these findings.

MATERIALS AND METHODS

Sequences and structures

Sequences of plant viral RNAs and structures of tRNAs were collected from the NCBI nucleotide database and structure database, respectively (http://www.ncbi.nlm.nih.gov/); Arabidopsis tRNA sequences were collected from the AGR database at UK Cropnet (http://ukcrop.net/agr/). Alignment programs CLUSTALW and DIALIGN2 were run on the public servers of Infobiogen (http://www.infobiogen.fr/services/analyseq/cgi-bin/clustalw_in.pl) and the Pasteur Institute (http://bioweb.pasteur.fr/seqanal/interfaces/dialign2-simple.html). The occurrence of various base–base interactions was found in the Fox Lab non-canonical base pair database (http://prion.bchs.uh.edu/bp_type/). Modeling was based on the 1.93 Å crystal structure of yeast tRNA^Phe, labeled 1EHZ in the Protein Data Bank (8), and performed with MolScript v.2.1.2 (9). Structures were rendered with Raster3D v.2.5d (10).

Viability and evolution of mutant TYMV

Site-directed mutagenesis by PCR was performed according to standard procedures on cDNA of the ‘Blue Lake’ strain of TYMV. The template (pACBL16) was constructed by cloning the XbaI–HindIII fragment from the cDNA plasmid pBL16 (a kind gift of Dr A. Gibbs) (11) into pACYC184, from which the 2069 nt Bsu36I–HindIII fragment had been deleted. pACBL16 was more stable in Escherichia coli than the original pBL16. The PCR mutagenesis also converted the NdeI site at the 3′ end of the TYMV cDNA into an Eco72I site. T7 transcription of the resulting plasmids, linearized with Eco72I, yields full-length TYMV RNA with one additional uridine encoded at the 3′ end. Three-week-old plants of Chinese cabbage (Brassica rapa subsp. pekinensis) were inoculated with in vitro capped transcripts as described (12). Viral RNA obtained by phenol extraction of bentonite-treated sap was used for RT–PCR according to the procedures recommended by the suppliers (Sigma-Aldrich and Eurogentec), using one biotinylated primer. The region corresponding to the TLS was analyzed by dideoxy sequencing after strand separation with Dynabeads (Dynal).

RESULTS

Strict non-Watson–Crick covariation in the potential D-loop–T-loop interaction of valine-accepting TLSs

Figure 2 presents an alignment of all valine-accepting TLSs available in the NCBI nucleotide database, i.e. most tymoviruses, SHMV (an unusual Tobamovirus that has a typical valine-accepting TLS) (13) and all pecluviruses, pomoviruses and furoviruses. Not included are the tymoviruses ErLV, which has a different, much less tRNA-like structure (14), DuMV, which carries a poly(A) tail instead of a TLS, and ChMV, the sequence of which either has no TLS or is incomplete.

Figure 2.

Alignment of all valine-accepting plant viral TLSs present in the NCBI nucleotide database. Columns show, respectively: genus name; database accession number; virus acronym (where applicable with the strain acronym and the number of the genomic RNA species); aligned sequence (hyphen, no nucleotide; vertical bar, separation between stacked stems; dot, unknown sequence). The top line indicates the loops, the pseudoknot region and the anticodon (AC). The D-loop is boxed in blue, the T-loop in green. Nucleotides corresponding to G₁₈, G₁₉, Ψ₅₅ and C₅₆ are shown in black print on yellow when identical to canonical tRNA and in red print on orange when different. Underlined nucleotides are proposed to be base paired. Consensus denotes a consensus sequence of RNA 3 from 18 variants of BSBV (K = G or U, Y = C or U) (47). For comparison, the bottom lines show the sequences of cytoplasmic and chloroplast tRNA^Val from Arabidopsis thaliana, collected from the AGR database at UK Cropnet (sequences from Brassica are not available). Grey brackets indicate the three potential tertiary base pairs discussed in the text.

While the similarities in TLS organization are striking, these are not always paralleled by similarities in the coat or RdRp proteins. In agreement with the accepted classification of these viruses, phylogenetic analysis using the alignment programs CLUSTALW and DIALIGN2 indicates, for example, that the tymoviral proteins are not more similar to the furoviral ones than to the bromoviral or tobamoviral ones (not shown). This suggests that the overall shape of the TLS is not enforced by the viral proteins. Direct interaction with the virus-encoded RdRp or with the coat protein, as proposed for TLSs from certain other genera, therefore seems unlikely to be the main driving force behind the tRNA mimicry in these viruses. Rather, the crucial interaction partners should probably be sought amongst the host proteins.

Focusing on the D-loop–T-loop interaction, the equivalents of G₁₈, G₁₉, Ψ₅₅ and C₅₆ are easily pinpointed in the majority of the furoviruses, where the entire consensus T-loop (TΨCRANY) is present as UUCRANY (boxed in green) and a conserved pair of G residues is found in the loop that mimics the tRNA D-loop (boxed in blue). The alignment of all the TLSs with the various species of plant tRNA^Val now shows that a G–C pair equivalent to G₁₉–C₅₆ in tRNA is possible in all cases, the two nucleotides being absolutely conserved (boxed in yellow). The G·Ψ pair, on the other hand, appears to be present as either G·U or A·G. Two possible explanations for this strict limitation to two base–base combinations spring to mind. First, it could represent a genuine non-Watson–Crick covariation, implying that the two bases interact as in tRNA (see Fig. 1). Alternatively, it could be the result of an evolutionary dichotomy, where the two nucleotides have become fixed independently of each other for reasons other than a direct interaction. Several facts argue against the latter possibility. First, the switch between G·U and A·G appears to have occurred twice independently, namely in the tymoviruses (OYMV) and in the furoviruses (OGSV). This would be very unlikely if there were no direct requirement for the two nucleotides to change simultaneously. Second, the genome of the Japanese strain of soil-borne wheat mosaic virus (SBWMV-JT) consists of two RNAs, one of which has the G·U pair, while the other has A·G in an otherwise almost identical context. This suggests that G·U and A·G are (quasi-) equivalent pairs in the context of the D-loop–T-loop interaction of the valine-accepting TLSs.

In addition to this covariation in the D-loop–T-loop interaction, we note a covariation from U–A to A–U in the intraloop base pair within the T-loop of SHMV (indicated by the third grey bracket in the bottom part of Fig. 2). This is surprising, because the equivalent T–A pair in tRNA has a trans Watson–Crick/Hoogsteen geometry and is not self-isosteric (5,15).

Functional evidence for a D-loop–T-loop interaction in TYMV

In order to examine the equivalence of G·U and A·G, as opposed to other base–base combinations, we have mutated the two relevant nucleotides in a cDNA copy of the genome of the ‘Blue Lake’ strain of TYMV (11). Twelve different base–base combinations were obtained, including the wild-type (A·G). Transcripts of these mutant genomes were produced by T7 transcription and used to inoculate four plants of Chinese cabbage (B.rapa subsp. pekinensis) each. The results of these experiments were monitored in two ways. First, the appearance of local and systemic symptoms was recorded for each of the plants, giving a direct indication of the viability of the mutant viruses. Second, young leaves were harvested after sufficient development of systemic symptoms and the virus progeny extracted from these were examined for secondary mutations in the TLS. This would reveal whether reversions or pseudoreversions are required for the development of visible systemic symptoms.

The horizontal bars in Figure 3 summarize the results concerning the appearance of symptoms. Dashed bars indicate local symptoms on the leaves that were inoculated; solid bars indicate the presence of systemic symptoms on young leaves; open bars indicate the absence of either type of symptom. The only two variants that consistently produced local symptoms within 5 days after inoculation and systemic symptoms during the second week (8–12 days) are the A·G wild-type and the ‘even more tRNA-like’ G·U variant. This, by itself, already confirms that the two combinations work similarly well, even in the context of a single viral species. The fact that symptoms from other combinations appeared with a delay indicates that the interaction of these two bases is not only a physical reality, but that it is of great importance for the viability of the virus. Consistently early symptoms, although slightly delayed relative to A·G and G·U (6–9 days after inoculation), were further only seen with the combination G·C. The fact that for many other variants the results differed from plant to plant suggested that their survival depended on chance events, like spontaneous mutations.

Figure 3.

Results of inoculation of plants with mutant TYMV RNA. The first column shows the nucleotides present in the starting mutants at the positions equivalent to tRNA G₁₈ and Ψ₅₅. Dashed bars indicate the appearance of local symptoms, solid bars represent systemic symptoms. Fully open bars indicate plants that did not reveal symptoms at any time during the experiment. Asterisks indicate the time points where leaves were harvested, virus extracted and its TLS sequenced. The final column shows the nucleotides present at the mutated positions at the time of harvest.

At the time points indicated with asterisks in Figure 3, virus was isolated from young, systemically infected leaves and the TLS examined by RT–PCR sequencing. Spontaneous mutations were found, but only at the positions that had originally been altered. As shown in the rightmost column, the vast majority of the viruses that gave clear systemic symptoms were now of the A·G or G·U type. G·C was found on two occasions, once as a mixed sequence and once exclusively in the young leaves, while the inoculated leaves contained the G·U revertant. We conclude that the G·C variant is by itself capable of producing symptoms, but is nonetheless soon replaced by G·U. One plant containing C·U showed early and strong symptoms while no (pseudo)reversion could be detected. We do not understand this observation, especially since two other plants did not produce systemic symptoms while a fourth one contained exclusively the G·U pseudorevertant. We also analyzed some young leaves of plants that did not show systemic symptoms, to examine whether perhaps the virus was multiplying and spreading without causing visible symptoms. No RT–PCR product was obtained from these leaves.

Taken together, these findings support our hypothesis that a D-loop–T-loop interaction exists in the valine-accepting plant viral RNAs. Moreover, this interaction seems to be of pivotal importance for multiplication, survival or spread of the virus in the plant. G·U and A·G appear to be equivalent base pairs in the context of the valine-accepting TLS, as predicted from the phylogenetic analysis above.

G·U and A·G are equivalent base pairs only in the context of the D-loop–T-loop interaction

Before discussing the structural implications of the observed covariation, we must mention that we do not know whether the U of the G·U pair is modified in the plant viral TLSs concerned. When the G of the A·G pair was replaced by U in a 3′-terminal fragment of TYMV RNA, this was efficiently modified to Ψ by a yeast extract (16). However, while this extract modifies various nucleotides in the wild-type fragment in a manner analogous to tRNA, no modifications of any kind have been detected in TYMV RNA extracted from infected plants (17). Structurally, U and Ψ are nearly equivalent, and this modification only marginally stabilizes the structure of canonical tRNA. The modification would probably also be ignored during viral replication, as Watson–Crick Ψ–A pairs have the same structure and stability as U–A (18). In fact, the corresponding U in an internal T-loop sequence in BMV is modified to Ψ in vivo (19).

Covariation between G·U and A·G has so far not been found outside the world of the tRNAs and tRNA-like molecules. The G·Ψ pair in tRNA has a radically different structure to the common ‘wobble’ G·U pair as found in regular RNA helices. X-ray crystallography of various tRNAs has indicated this pair to be of the ‘GU imino:amino-2-carbonyl bifurcated’ type, in which the O2 of U₅₅ can accept a hydrogen from either N1 or N2 of G₁₈ (20). Such bifurcated hydrogen bonds confer more stability to the interaction than single bonds. It is not immediately obvious which of the possible A·G base pairing schemes would be isosteric with the bifurcated G·U pair, nor is it clear why it cannot be replaced by any other base–base combination. We therefore envisaged the possibility that the choice of base pair is severely limited by the compact nature of the elbow region of the L-shaped molecule. Figure 4, right, focuses on this region as present in the crystal structure of tRNA^Phe from yeast (8).

To examine the possibility of fitting various base–base combinations into the context of the TLS from TYMV, we initially turned to the three-dimensional model proposed in 1987 (21). Unfortunately, however, the elbow corner of the molecule could not be modeled with sufficient confidence to be suitable for the fitting of alternative bases. Because of the high structural similarity between the valine-accepting TLSs and canonical tRNAs, we decided to start instead from the most recent update of the crystal structure of yeast tRNA^Phe (8). Further justification for this choice came from the fact that the same G·U versus A·G covariation was subsequently found to occur in tmRNAs (22) and in certain mitochondrial tRNAs (23; M.H.de Smit and C.W.A.Pleij, manuscript in preparation).

Keeping the backbone in place, we simply replaced the bases of the original G·Ψ pair with all other combinations and judged these for (i) steric fit without distortion of the backbone, (ii) preservation of the strong stacking interactions with neighboring bases and (iii) the possibility of hydrogen bond formation. Figure 5 summarizes the results. Eight of the possible combinations were immediately ruled out because a pyrimidine in place of G₁₈ would cause a large loss in stacking energy both with the intercalating single purine from the T-loop (G₅₇) and with A₅₈ of the intraloop base pair. Indeed, both G·U and A·G have a purine at this position. G·A and G·G cannot be fitted in without serious distortion of the backbone, which would probably also cause a loss of stacking at other positions in the region. Of the remaining combinations, A·U, A·C and A·A are not predicted to form base–base hydrogen bonds in their present conformation. They could only do so if the N1 positions of the adenosines were protonated, but we have no reason to assume that this occurs to a significant degree. Indeed, such interactions have not been found in other RNA molecules (20). This leaves A·G, G·U and G·C as the only three combinations that are predicted to form hydrogen bonds when fitted into the compact elbow structure of tRNA^Phe from yeast. In the case of A·G, only a single hydrogen bond is predicted between N2 of guanine and N1 of adenine. We note, however, that a slight sideways displacement of the bases would allow the formation of a second hydrogen bond between N1 and N6. The resulting pair would be of the trans Watson–Crick/sugar edge type (15). This conformation, proposed in the original three-dimensional model for TYMV RNA (21), has been found at various positions in rRNAs (20).

Figure 5.

Fitting of various base–base combinations in place of G₁₈·Ψ₅₅ of tRNA^Phe. Predicted hydrogen bonds are shown as yellow dotted lines, with their lengths in angstroms. The hydrogen bond in green (A·G pair) is only possible after slight readjustment of the bases (see text). The unfavorable hydrophobic interaction in G·A and the steric clash in G·G are indicated in orange, with atomic distances in angstroms.

The outcome of our ‘naive’ modeling is quite striking, especially considering the fact that in our experiments G·C was the only non-natural combination that allowed near-wild-type viability and symptom formation. This suggests that the covariation found in the valine-accepting TLSs may be enforced mainly by the compact nature of the D-loop–T-loop interaction. Why G·C is not found as a natural variant we do not know, but the suggested bifurcated G·C pair has so far not been found elsewhere either (20). A 60° turn of the cytosine in its plane would convert it to the more common reverse Watson–Crick (trans Watson–Crick/Watson–Crick) pair forming two distinct hydrogen bonds, but this would presumably be incompatible with the elbow structure.

DISCUSSION

The possibility of an unusual D-loop–T-loop interaction in TYMV has been proposed before, on the basis of the nucleotide composition of the two loops (4). However, no phylogenetic or experimental evidence had been presented to corroborate the existence of an A·G pair as a functional equivalent of the G·Ψ pair of canonical tRNAs. In this report, we provide both types of evidence, confirming the existence and functional importance of a D-loop–T-loop interaction in the tymoviruses. Moreover, the phylogenetic support extends to the other valine-accepting TLSs as well.

The existence of this interaction was not a priori obvious, as the D-loop of the TLSs shows some marked differences from the classic tRNA D-loop. Although D-loops in cytoplasmic tRNAs vary in size, they always contain at least 7 nt, with the GG nucleotides that pair with the T-loop located in the middle or somewhat to the 3′ side (see, for example, the various species of tRNA^Val in Fig. 2). The TLS equivalents, in contrast, mostly consist of only 4–6 nt, with the GG or AG sequence often on the 5′ side of the loop (Fig. 2). There is little evidence for specific structures within the D-loop that are required to expose the GG nucleotides to the solvent. In yeast tRNA^Phe, the D-loop forms a binding pocket for a Mg²⁺ ion, which may help to push the GG outwards (8,24). However, this same Mg²⁺-binding pocket is not present in all tRNAs and it apparently does not impose a strong conservation of the overall conformation of the loop. It thus appears that the T-loop nucleotides manage to find their counterparts in a variety of different D-loops and this probably explains their ability to interact with the smaller loops in the TLSs as well.

The precise structural consequences of the replacement of the T-loop (pseudo)uridine of tRNA by a guanosine are unclear. The Ψ in tRNA is important in forming the U-turn that is characteristic of the T-loop. To keep this turn intact, the intraloop U–A pair had to be sacrificed in the three-dimensional model of the TYMV TLS (21), but the conservation of both nucleotides and the covariation in SHMV now render this choice disputable. An earlier NMR structure analysis of the TYMV acceptor arm provides no answer, because it was performed with a slightly altered molecule where the UGC sequence in the T-loop had been replaced by CUC (25). However, the phylogenetic and experimental data indicate that the (pseudo)uridine is not necessary for the T-loop to be able to capture its pairing partners in the D-loop. Further experiments to determine the detailed structure of the TYMV TLS by NMR are presently underway.

From an evolutionary viewpoint, it is surprising to find that the two RNAs of SBWMV-JT differ in the nature of their D-loop–T-loop interactions. Usually, the TLSs of the component RNAs of a multipartite virus are highly similar or identical, which is taken as an indication of a rather frequent exchange of domains by RNA recombination. The alignments in Figure 2 suggest that the conversion between G·U and A·G is a relatively rare event that probably occurred only once in the tymoviruses and once in the furoviruses. It seems that in the case of SBWMV-JT we see the virus just shortly after this conversion, before the two RNAs have had the chance to match their 3′ ends. In this context, it would be interesting to know whether RNA2 of OGSV carries the A·G pair, as does its RNA1. The apparent rarity of the conversion is not surprising, since it requires a transition and a transversion to occur simultaneously. In our evolution experiment we found, as expected, that transitions were preferred over transversions and that double transitions were more rare than single transversions (Fig. 3). In addition, the double mutation from A·G to G·U or vice versa would not yield a selective advantage, so any such mutant that might arise would not overgrow the population.

The strong requirement for an intact G·U or A·G pair for the survival of TYMV suggests that by preventing the non-canonical pairing at the heart of this compact region (see Fig. 4) the two loops may become disconnected. But why is an undisturbed pairing between the D-loop and the T-loop so important to TYMV? In tRNA, the generally accepted idea is that this interaction stabilizes the L-shape of the molecule and, particularly, fixes the angle between the two arms (26). This, of course, does not tell us why the structure must be stabilized or the angle fixed. In fact, many of the enzymatic reactions that tRNA undergoes (at least when off the ribosome) do not require the rigid L-shape.

Concerning aminoacylation, the requirements appear to vary between the synthetases (27). Charging of tRNA^Ala by E.coli AlaRS depends heavily on the G₁₈·U₅₅ pair and only a partial restoration is seen with A·G (22). In contrast, mutation of G₁₈ or G₁₉ to A did not impair the charging of tRNA^Ile by E.coli IleRS (28). Even the entire T-arm could be removed from this tRNA, or from yeast initiator tRNA^Met, with little effect on aminoacylation (28,29). Importantly, ValRS belongs to the same subclass of synthetases as IleRS and MetRS (30). In a preliminary experiment, we valylated 3′-terminal fragments of 102 nt from the mutant TYMV RNAs with yeast ValRS. As expected, we found only minor variations in valylation efficiency (V_max/K_m reduced by 20–50% in a single experiment) which did not parallel the in vivo viability. Since reductions by up to 92% in V_max/K_m did not affect the accumulation of mutant TYMV and the development of symptoms in an earlier study (31), it is unlikely that the reduced viability of our mutants is due to a defect in aminoacylation.

Increased flexibility of the elbow region might also increase misaminoacylation. While increased mischarging of tRNAs would be deleterious to the cell, it cannot explain the strong reduction in viability of the TYMV mutants. In fact, a TYMV mutant that was methionylated instead of valylated was perfectly viable, indicating that misaminoacylation does not inactivate the virus (31).

Regarding the proposed function of the TLS as a replication signal, it is important to note that while we do not know the replication efficiency of our various mutants, most of them produced revertants. This indicates that replication must have taken place at a considerable level. Partially purified RdRp preparations from TYMV do not rely on an intact TLS for minus strand synthesis in vitro (32,33). Moreover, TYMV carrying the heterologous and only marginally tRNA-like 3′ end from ErLV is viable, underlining the fact that the tymoviral RdRp does not even require a full TLS in vivo (34,35). The latter result also renders an essential function of the TYMV TLS in packaging unlikely, contrary to recent findings for BMV (36).

The telomere-like function of the TLS as a substrate for ATP/CTP:tRNA nucleotidyltransferase, as shown to exist in BMV (for an overview see 2), is probably sensitive to mutations in the T-loop, but not to mutations in the D-loop (37,38). Functional compensation of a mutation in the T-loop by a D-loop mutation, as found in our double mutant G·U, would therefore not be possible if the viability of the virus was mainly determined by this activity.

Apart from the various enzymes mentioned above, plant viral TLSs also interact efficiently with the translational elongation factor eEF1A (formerly EF-1α) (39). A recent hypothesis on the function of the TLS is that the binding of eEF1A may block replication and thus control the switch between the translatable (mRNA) state and the replicable (genome) state of the viral RNA (35). A similar switch has been demonstrated experimentally in alfalfa mosaic virus, where it is induced by the viral coat protein (40). It is as yet unclear how an abundant host protein like eEF1A might induce a switch, but an attractive hypothesis may be found in the physical separation of translation and replication of TYMV RNA (41). While eEF1A would bind to the RNA in the cytoplasm and keep it in translation mode, it might dislodge, due to altered conditions, in the chloroplast vesicles where replication takes place (42,43).

The crystal structure of the bacterial ternary complex, EF-Tu·GDPNP·Phe-tRNA^Phe, shows no direct interaction between EF-Tu and the loop nucleotides of the tRNA (44). However, it was recently found that the binding of EF-Tu·GTP to tRNA^Ala became seriously impaired upon disruption of the G₁₈·U₅₅ pair (4- to 10-fold increased K_d for A·U and G·G, respectively) (22). Crucially, this effect was for the most part reversed (to a K_d only 2-fold higher than wild-type) when the G·U pair was replaced by A·G, thus mimicking the covariation in the plant viral TLSs. Apparently, the loop–loop interaction somehow optimizes the structure of the acceptor arm for EF-Tu binding. In addition, the combination A·G turned up as the predominant outcome of a SELEX experiment where tRNAs, randomized in their D- and T-loops, were selected for efficient phenylalanylation and subsequent EF-Tu binding (45). Phe-tRNA^Phe in which G₁₈·U₅₅ was mutated to A·C bound to EF-Tu·GTP with a 3.5-fold increased K_d, but mutation to G·C had no effect (46). This latter observation is in agreement with both our structure modeling and our observation that the G·C mutant of TYMV gives clear systemic symptoms and is replaced only gradually by the G·U pseudorevertant.

We therefore suggest that the most likely explanation for the need for an intact D-loop–T-loop interaction is that it stimulates binding of eEF1A and thus plays an important role in controlling the life cycle of the virus. Decreased binding of eEF1A to the mutant RNAs would reduce their availability for translation, leading to a decreased production of viral proteins and, ultimately, to the absence of visible symptom development. Replication, on the other hand, would not be impaired directly and a decreased amount of RdRp may still allow the generation of the observed revertants. The present collection of mutant TYMV RNAs and their associated strong phenotypes may form a useful starting point for further examination of this and alternative hypotheses.