In Vitro Transcription by the Turnip Yellow Mosaic Virus RNA Polymerase: a Comparison with the Alfalfa Mosaic Virus and Brome Mosaic Virus Replicases

Abstract

Recently, we showed that the main determinant in the tRNA-like structure of turnip yellow mosaic virus RNA to initiate minus-strand synthesis in vitro is the 3′ ACCA end. By mutational analysis of the 3′-terminal hairpin, we show here that only a non-base-paired ACCA end is functional and that the stability of the wild-type 3′-proximal hairpin is the most favorable, in that it has the lowest ΔG value and a high transcription efficiency. With a nested set of RNA fragments, we show that the minimum template length is 9 nucleotides and that transcription is improved with increasing the length of the template. The results also suggest that proper base stacking contributes to efficient transcription initiation. Internal initiation is shown to take place on every NPyCPu sequence of a nonstructured template. However, the position of the internal initiation site in the template is important, and competition between the different sites takes place. Internal initiation was also studied with the RNA-dependent RNA polymerase of brome mosaic virus (BMV) and alfalfa mosaic virus (AlMV). The BMV polymerase can start internally on ACCA sequences, though inefficiently. Unexpectedly, the polymerases of both AlMV and BMV can start efficiently on an internal AUGC sequence.

An important step in virus multiplication is replication of the virus genome. In the case of positive-strand RNA viruses, efficient replication is mediated by a virus-encoded polymerase which specifically interacts with the viral genome. Specificity can be obtained in different ways. First, the polymerase itself binds to a specific sequence or structure in the genome, followed directly by initiation of transcription. Second, a different protein recognizes the initiation site and by protein-protein interactions enables the polymerase to start transcription at the correct position. Third, a primer binds to the initiation site, and by interaction of the polymerase with the primer, transcription is activated. In many cases the polymerase is part of a replication complex consisting of viral and host proteins, suggesting that specificity is obtained in more than one way.

Different experimental approaches have been used to determine the specific sequences required for replication of the viral RNA by its RNA-dependent RNA polymerase (RdRp). Defective interfering RNAs (22) and virus-associated satellite RNAs (15, 33, 34) were shown to be useful tools in determining the template requirements for replication by the RdRp of turnip crinkle virus (TCV). Studies on the subgenomic promoter regions of brome mosaic virus (BMV) RNA3 (1, 2, 12, 23, 30), Sindbis virus (21), and cucumber necrosis virus (17) also contributed to the knowledge of RNA replication of these viruses. However, none of these approaches has led to a complete understanding of initiation of RNA replication.

RNA replication of positive-strand RNA viruses starts at the 3′ end of the genome. For members of the Tymovirus, Tobamovirus, Hordeivirus, Furovirus genera and some members of the Bromovirus genus, the 3′ end of the RNA is folded in a tRNA-like structure. The nucleotide sequences and secondary structures of these tRNA-like structures are not conserved, indicating that each group has independently acquired the structure via convergent evolution (16). This suggests that the tRNA-like structure is a way to fulfill one function or a set of defined functions that are of vital importance for the virus. For BMV, a bromovirus, it was shown that the tRNA-like structure is involved in RNA replication (11, 24). A detailed investigation on initiation of minus-strand synthesis of one of these viruses not only would add greatly to our knowledge of RNA replication by a viral RdRp but also could be useful to define the function of the tRNA-like structures at the 3′ ends of viral genomes.

The simplest and possibly the best-defined tRNA-like structure is that of turnip yellow mosaic virus (TYMV), a tymovirus (29). TYMV already has a long history of in vitro studies on RNA replication (26). However, for many years it was known only that initiation of minus-strand synthesis could be inhibited by RNA fragments containing the 3′-terminal 108 nucleotides (nt) of the genomic RNA, including the 3′-terminal tRNA-like structure (25). Later, it was shown that competition was also obtained with an RNA fragment consisting of the 3′-terminal 38 nt of the genomic RNA (13). Only recently, a simple, highly reproducible method for the isolation and partial purification of the replication complex of TYMV enabled the continuation of investigations on RNA replication in vitro (10). By using this RdRp preparation, shown to be of viral origin and specific for TYMV RNA (10), an efficient transcription of an RNA fragment consisting of the 3′-terminal 83 nt, including the tRNA-like structure, was obtained (9, 10, 31). However, a similar result was obtained with a fragment consisting of only the 3′-terminal 28 nt involved in the formation of a pseudoknot structure (Fig. 1A) (9, 31). This indicates that the complete tRNA-like structure of TYMV RNA is not required for initiation of minus-strand synthesis in vitro. Recently, the results of an extensive mutational analysis of part of the tRNA-like structure upstream of the pseudoknot region confirmed this conclusion (32). Even with an RNA fragment representing the 3′-terminal hairpin, only a twofold reduction in efficiency was obtained (9).

FIG. 1.

Base pairing of the 3′ ACCA end and stabilization of the 3′ hairpin stem of TYMV RNA inhibits transcription. (A) Secondary structure of the 3′ hairpin mutants in which the 3′ ACCA end is completely base paired (Closed-1) and the stem region is destabilized (Closed-2 to -4) or in which the 3′ ACCA end is partially base paired (Closed-5) or in which the 3′ ACCA end is free (Stable-1) and the stem region is stabilized (Stable-2 to -4). Boxes indicate the mutations compared to Closed-1. The ΔG values as calculated by the program Mfold (39) are indicated. The 28-nt fragment consists of the 3′-terminal 28 nt of TYMV RNA and is used as reference RNA. The pseudoknot formation is presented by dotted lines, and the stem regions (S1 and S2) are indicated. (B) Autoradiography of the ³²P-labeled products obtained with the various RNAs. Positions of the template RNA (21 nt) and the reference RNA (28 nt) are indicated. Transcription efficiencies compared to Stable-1 and corrected for the varying number of [³²P]UMP incorporated are presented. The product of Stable-4 migrates faster, probably due to incomplete denaturation of its stable structure.

By mutational analysis of the 3′-terminal hairpin, it was shown that only the two C residues of the non-base-paired ACCA end were specifically involved in the interaction with the RdRp (8). Initiation was shown to occur de novo with incorporation of a GTP, as was previously shown for BMV RdRp (18, 24), indicating that the 3′-terminal A residue is not transcribed (8, 31). However, the 3′-terminal A residue is required for efficient transcription (8). Besides nucleotide specificity, some investigations were performed on the effect of the RNA structure on template efficiency. Base pairing of the 3′ ACCA end resulted in a drastic drop in efficiency, while reducing the length of the hairpin stem also resulted in a decrease of transcription efficiency. In addition, it was suggested that the RdRp is able to initiate internally on an NCCN or NUCN sequence (8, 32).

In this report, we present the results of a detailed investigation of the effect of base pairing and length of the template on transcription efficiency. The new data have improved our understanding of what is required for in vitro transcription initiation by the RdRp of TYMV at both the 3′ terminus and internal initiation sites of the template.

Internal initiation was also observed with the RdRps of BMV and alfalfa mosaic virus (AlMV). Unexpectedly, the RdRps of both AlMV and BMV could use AUGC as an internal initiation site in a very efficient manner.

MATERIALS AND METHODS

TYMV RdRp preparation from Chinese cabbage plants and in vitro transcription assay.

The RdRp was isolated from Chinese cabbage leaves 10 days after inoculation with TYMV and was purified up to and through the glycerol gradient centrifugation step as described previously (10). Twenty microliters of the glycerol gradient fraction containing the highest RdRp activity was treated with micrococcal nuclease, and in vitro transcription was performed in 100 μl containing 40 mM Tris-HCl (pH 9.0), 8.0 mM MgCl₂, 2.5 mM dithiothreitol, 0.8 mM ATP, GTP, and CTP, 10 μCi of [α-³²P]UTP (800 Ci/mmol; ICN), 2% ethanol, 125 ng of actinomycin D, RNAguard (1 U/μl; Pharmacia), and an equimolar amount of the various RNA fragments, as described previously (10). The samples were incubated at 29°C for 1 h. The reaction products were phenol extracted, precipitated, preheated for 1.5 min at 95°C in formamide loading buffer, and analyzed by gel electrophoresis on a 9.5 M urea–20% polyacrylamide gel under denaturing conditions. These strong denaturing conditions are required to denature the stable double-stranded RNA products. After electrophoresis, the gel was stained with o-toluidine blue to determine the positions of the template RNAs and to ensure that no RNA degradation had taken place during the incubation. The incorporation of [³²P]UMP was determined by Cerenkov counting of the reaction product in the gel. The relative transcription efficiency was obtained by comparing the [³²P]UMP incorporation, corrected for the number of UMP residues in the RdRp transcript, with that obtained for the reference template.

BMV RdRp preparation.

The RdRp of BMV was isolated from BMV-infected barley leaves and partially purified up to and through the sucrose gradient centrifugation step as described by Bujarski et al. (5), using dodecyl-β-d-maltoside as the detergent. Ten microliters of the sucrose gradient fraction containing the highest RdRp activity was treated with micrococcal nuclease and used for in vitro transcription.

AlMV RdRp preparation.

The RdRp of AlMV was isolated from AlMV 425L-infected tobacco (Nicotiana tabacum L Samsun NN) leaves and partially purified up to and through the glycerol gradient centrifugation step as described by Quadt et al. (28). Ten microliters of the glycerol gradient fraction containing the highest RdRp activity was treated with micrococcal nuclease and used for in vitro transcription.

T7 DdRp and AMV RT.

About 65 U of commercially available T7 DNA-dependent RNA polymerase (DdRp) (Pharmacia) and 50 U of avian myeloblastosis virus (AMV) reverse transcriptase (RT) (Promega) were used in the in vitro transcription assays.

Preparation of closed, stable, and internal RNA constructs and mutants.

Oligonucleotides corresponding to the desired RNA fragments (see Results) were ligated downstream of a T7 promoter in the vector pUC19 as previously described (9, 10). The DNA was digested with MvaI (MBI Fermentas), and runoff T7 transcription was performed. The RNA was purified by electrophoresis on an 8 M urea–20% polyacrylamide gel as described before (9, 10).

Preparation of single-strand and hairpin RNA fragments.

RNA fragments of 8, 9, 10, 11, 14, and 18 nt, respectively, were synthesized on a Gene Assembler Special DNA synthesizer (Pharmacia LKB).

Nuclease S1 digestion.

After phenol extraction and precipitation of the reaction products, the pellet was dissolved in 10 μl containing 50 mM sodium acetate (pH 4.6), 200 mM NaCl, 2 mM ZnSO₄, and 10 U of nuclease S1 (Pharmacia). Incubation for 30 min at 37°C was performed as described previously (10). After phenol extraction and precipitation, the products were analyzed by electrophoresis.

RESULTS

A non-base-paired ACCA is absolutely required for efficient transcription initiation in vitro.

Previously, we concluded that the only determinant in the tRNA-like structure of TYMV RNA for initiation of minus-strand synthesis by the viral polymerase in vitro is the 3′ ACCA end (8). Base pairing of this ACCA end resulted in an appreciable loss in transcription efficiency (8). To test whether this was due to the increased stability (i.e., decreased ΔG value) of the 3′ terminal hairpin by the addition of one to three extra base pairs or whether the RdRp can recognize only a non-base-paired ACCA end, we constructed RNA fragments in which the ACCA end is base paired but in which the stem region is less stable than the wild-type 3′ hairpin stem as indicated from their calculated ΔG values (Closed-1 to -4 [Fig. 1A]). To avoid the formation of alternative conformations, the wild-type loop region with three consecutive G residues was replaced by a stable tetraloop (UNCG) which was previously shown to contribute to transcription efficiency (8). None of the fragments in which the ACCA end is base paired could be used as a template by the RdRp (Fig. 1B). Base pairing of only the 5′ A residue of the 3′ ACCA end (Closed-5) results in a drop in efficiency to 32% compared to a non-base-paired 3′ ACCA end (Stable-1), which is in agreement with previous results (8, 32). These results unambiguously prove that a non-base-paired ACCA is required for transcription initiation.

Transcription is reduced by stabilization of the 3′ hairpin stem.

To test the effect of increasing the stability of the wild-type stem region, the AU base pairs of Stable-1 were changed to GC base pairs (Stable-2 to -4 [Fig. 1A]). In Fig. 1B it is shown that increasing the stability results in a decrease of template efficiency. Earlier work showed that destabilization of the stem region does not influence the efficiency (8). We therefore conclude that the calculated ΔG value of a hairpin directly upstream of the ACCA end, consisting of five base pairs, should not be lower than −8.2 ± 0.5 kcal/mol for efficient transcription by the RdRp of TYMV.

The minimal template length to start transcription in vitro is 9 nt.

Previous work suggested that a secondary structure element like the 3′-terminal hairpin contributes to efficiency, although it is not absolutely required to initiate transcription (8, 32) (see below). To investigate the effect of template length on transcription efficiency without a possible effect of a secondary structure element, RNA fragments of increasing length and without a secondary structure, SS1 to SS6, were constructed (Fig. 2A). The sequence of the fragments, with a high G+A content, was chosen in such a way as to both prevent internal initiation on a C or U residue (see below) and prevent hairpin formation. Except for SS1, all constructs could be used as a template by the RdRp (Fig. 2B). However, besides the full-length product, a high amount of shorter products is obtained. As determined by staining with o-toluidine blue, degradation of the template RNAs during the assay had not taken place. Two main short products were determined to be 6 and 9 nt in size, respectively. Coincidentally, the sequences of SS1, SS2, SS3, and the 3′-terminal 10 nt of SS4 to SS6 appeared to be identical to the purine-rich 3′ end of BMV RNA. Depending on the nucleotide sequence of BMV RNA, abortive products of 6 and 8 nt were shown to be synthesized by the RdRp of BMV (35–37). Since the exact sizes of the abortive products are hard to determine, those obtained with the RdRp of BMV could very well be identical to the short products that we obtained with the RdRp of TYMV. As we have never detected abortive products when using structured as well as other nonstructured templates derived from TYMV RNA (8) (see below), the production of these putative abortive products is probably also due to the specific nucleotide sequence of the template for reasons we do not know. Besides these putative abortives products, products 1 or 2 nt smaller than the expected full-length product are obtained with SS5 and SS6. Again the 5′ purine-rich sequences of these fragments could be the reason for the synthesis of these shorter products.

FIG. 2.

Effect of template length on transcription efficiency. (A) Nested set of purine-rich single-stranded RNA fragments with 3′ ACCA ends (SS1 to -6). In HP5, the 3 nt at the 5′ end of SS5 are changed and base pair with the downstream GAG sequence, thereby forming a hairpin with a stable GAAA tetraloop. Base pairing is indicated by blocks connected by dashed lines. Tetra-G, previously shown to be transcribed as efficient as the 28 nt-fragment (8), is used as reference RNA. (B) Autoradiography showing ³²P-labeled products obtained with the various templates. Positions of the template RNAs (9, 10, 11, 14, 18, and 19 nt) and of the two main abortive products (6 and 9 nt) are indicated. A 10-nt fragment with a sequence complementary to the 5′-terminal 10 nt of SS4 migrated at the same position as SS3. A low amount of product is obtained from a degradation product of tetra-G and is expected to be 11 nt. The product of HP5 migates faster than the full-length product of SS5, probably due to incomplete denaturation of its structure. Transcription efficiencies compared to that of SS6 are indicated.

The quantity of the products, including the putative abortive products, was determined and presented as a function of the length of the template. Figure 2B shows that an RNA fragment of 9 nt can be used, although very inefficiently, as the template and that transcription efficiency increases when the length of the template is increased.

Interestingly, with a fragment in which part of the BMV-derived sequence is involved in the formation of a short hairpin structure, closed by the stable tetraloop GAAA (HP5 [Fig. 2A]), no abortives were obtained, indicating that base pairing in some way prevents generation of abortive products. However, the transcription efficiency of this fragment is twofold lower than that of the nonstructured SS5, a fragment of exactly the same length. The efficiency is comparable with that of the smaller SS3, a 10-nt fragment (Fig. 2B). We here note that the stability itself of the hairpin stem of HP5 should not hamper the polymerase (see above), indicating that the relatively short length of the stem region is responsible for the reduced efficiency.

Internal initiation takes place on an NPyCPu sequence.

Unexpectedly, the RdRp of TYMV was shown to be able to start transcription internally on a non-base-paired NCCN or NUCN sequence (8). We designed and constructed a new RNA fragment (Intern [Fig. 3B]) with four ACCA sequences (ACCA-1 to -4) in order to examine internal initiation in more detail. The four ACCA sequences are positioned in such a way that the products obtained could be easily separated and quantified by gel electrophoresis and the effects of the length of the parts upstream and downstream of the initiation site could be tested.

FIG. 3.

Internal initiation by the RdRp of TYMV. (A) Autoradiography showing α-³²P-labeled products obtained with the Intern or InternΔ fragment (Fig. 3B) before (−) or after (+) nuclease S1 treatment. An alkaline digestion of α-³²P-labeled Intern fragment (L) is used as a size marker. Sizes of the various products are presented. The corresponding initiation sequences (ACCA-1 to -3 and AUGC) of both fragments and their transcription efficiencies as a percentage of the total amount of product are indicated. In previous work we discussed problems with denaturation of double-stranded RNA products in sequencing gels (8–10). Only products obtained with small (19-nt) structured templates could be completely denatured. It was shown that complete denaturation is enhanced after nuclease S1 treatment (10). Therefore, it is assumed that the faint band migrating more slowly than the template RNA and disappearing after nuclease S1 treatment is still some nondenatured product (ND), despite the improved denaturing conditions (see Materials and Methods). (B) The Intern fragment. Nucleotide numbering is from the 5′ end. The ACCA sequences (bold) are numbered from the 3′ end (ACCA-1 to -4). The proposed initiation sites are represented by hooked arrows. In the InternΔ fragment, the AAUA sequence at the 5′ end is deleted (Δ). Base substitution in the ACCA-2 sequence is indicated by arrows. The only AUGC sequence is boxed. (C) Autoradiography of ³²P-labeled products obtained with the Intern fragment mutated in ACCA-2. The ACCA initiation sites corresponding to the various products are indicated. Transcription efficiencies corresponding to the products obtained from the mutated ACCA-2 as a percentage of the total amount of product and the product obtained from the nonmutated ACCA-2 are presented.

With the Intern fragment, four products corresponding to fragments of 32, 28, 23, and 15 nt are obtained in a denaturing gel system after treatment with nuclease S1, indicating that these fragments are derived from partially double stranded reaction products. With a mutant of Intern in which nt 2 to 5 from the 5′ end are deleted (InternΔ [Fig. 3B]), a similar pattern of bands that migrates faster in the gel is obtained (Fig. 3A). The products appeared to be 4 nt smaller than those obtained with the Intern fragment, proving that they are the result of internal initiation of transcription.

From the sizes of the different products, it was determined that internal initiation takes place on the designed ACCA sequences (ACCA-1 to -3). The fourth ACCA sequence (ACCA-4) is not used as an internal initiation site, in agreement with the results of the minimal template length determination as discussed above. Interestingly, ACCA-3, resulting in a 15-nt product, appears to be a very good initiation site in the Intern fragment. About 65% of the total amount of product is obtained from this site, while only 23 and 8% are obtained from ACCA-2 and ACCA-1, respectively. In the InternΔ fragment this ACCA-3 site, resulting in an 11-nt product, is used less efficiently: 47% of the total amount of product. Again this is in agreement with the shorter distance of the initiation site with respect to the 5′ end of the template. Interestingly, transcription efficiency of the ACCA-2 of this construct is now 41% of the total amount of product.

Remarkably, another product with a size of 28 nt, resulting from internal initiation on an AUGC sequence, is obtained (Fig. 3B). This sequence was designed in the Intern fragment to investigate internal initiation with the RdRp of AlMV (see below). For both constructs, the transcription efficiency of this site is 4% (Fig. 3A).

To study the specificity of the ACCA sequence in more detail than in the past (8), both C residues and the 3′ A residue of the ACCA-2 sequence were mutated one by one (Fig. 3B). The 5′ A residue was previously shown not to be specifically involved in initiation of minus-strand synthesis (8) and therefore was not mutated. Both C residues of the ACCA site are important for the interaction with the RdRp (Fig. 3C). However, the 5′ C residue is less specific than the 3′ C residue, since an AUCA sequence can also efficiently be used as an initiation site (Fig. 3C), again in agreement with previous results (8). A low amount of product, 21% compared to the ACCA sequence, was obtained from an AGCA sequence, in agreement with the use of the downstream AUGC(A) sequence as an internal initiation site (see above). In the case of an ACCC(A) sequence, the length of the product is increased by 1 nt, showing that internal initiation starts at the most 3′ C residue. This is again in agreement with what was previously published (31). The best internal initiation is obtained on a ACCG sequence, with an efficiency of more than 200%. On the other hand, ACCU is not a good sequence for internal initiation, indicating that the 3′ residue of ACCN should be a purine.

Altogether, these results show that efficient internal initiation takes place on a single-stranded NPyCPu sequence. Interestingly, the efficiency of an internal initiation site is dependent on its position in the template and on the efficiency of the other internal initiation sites, which means that competition takes place between the different internal initiation sites.

The RdRp of BMV can start internally on ACCA sequences, and the RdRps of both AlMV and BMV initiate internally on an AUGC sequence.

To investigate whether this phenomenon of internal initiation is characteristic for our TYMV RdRp preparations, we isolated the RdRps of AlMV according to the method of Quadt et al. (28) and of BMV as described by Bujarski et al. (5). Various laboratories successfully use these procedures in the investigation of RNA replication. Both preparations were shown to be specific for their own templates. Like the genomic RNA of TYMV, those of BMV possess a tRNA-like structure with an ACCA 3′ terminus. For a long time, the 3′ terminus of the genomic RNAs of AlMV was believed to have a different non-tRNA-like structure with an AUGC 3′ end. However, recently it was suggested that also this 3′ terminus can be folded in a tRNA-like structure (27). The RdRps were tested for transcription of the Intern fragment (Fig. 3B); commercially available AMV RT and T7 DdRp were used as controls. For the RdRp of AlMV, a product that migrates faster than the template RNA is obtained (Fig. 4A). A 4-nt-smaller product is obtained with the InternΔ fragment, proving that the products are synthesized from an internal initiation site (Fig. 4B). By using the products obtained with the TYMV RdRp as a marker, internal initiation is determined to take place on the AUGC sequence. Surprisingly, the RdRp of BMV was also able to initiate transcription very efficiently at the same position. In addition, the BMV RdRp can start internally on the ACCA-2 and ACCA-3 sequences (Fig. 4), though less efficiently than the RdRp of TYMV. This difference in efficiency could be due to the competition with the internal AUGC site. These results indicate that the RdRp of BMV can initiate transcription on two different sequences, though it prefers the AUGC site. The products obtained from internal initiation could not be degraded with nuclease S1, indicating that they are double stranded under the conditions used.

FIG. 4.

Comparison of internal initiation by the RdRps of TYMV, BMV, and AlMV. (A) Autoradiography showing ³²P-labeled products obtained with the Intern fragment before (−) and after (+) treatment with nuclease S1. The initiation sites (ACCA-1 to -3 and AUGC [Fig. 3B]) corresponding to the products are indicated. As for TYMV, some nondenatured product (ND) is detectable for BMV and AlMV before S1 treatment of the products (see the legend to Fig. 3). (B) Autoradiography showing ³²P-labeled products obtained with the Intern (i) and InternΔ (Δ) fragments after (+) treatment with nuclease S1. The initiation sites (ACCA-1 to -3 and AUGC) corresponding to the products obtained with the Intern fragment are indicated.

Both the AMV RT and the T7 DdRp could not use the Intern fragment as a template (result not shown).

DISCUSSION

A non-base-paired 3′ ACCA end downstream of at least five nucleotide residues is required for efficient initiation of transcription.

To understand the initiation of minus-strand synthesis by the RdRp of TYMV, an in vitro transcription assay was used to examine which RNA fragments derived from the 3′ end of the viral RNA can be used as template and what determines their transcription efficiency. Previous work showed that the main determinant for efficient initiation of transcription in vitro is the 3′ ACCA end of the tRNA-like structure of the viral RNA (8). More recent work by Singh and Dreher (32) confirmed this conclusion. Base pairing of the ACCA sequence led to a strong decrease in transcription efficiency, but at that time it was not possible to discriminate between an effect on the accessibility of these nucleotides for the incoming triphosphates or an effect of the increased stability of the 3′ terminal hairpin (8). We here prove that only a non-base-paired 3′ ACCA end is functional, as destabilizing the 3′-terminal hairpin has no effect on transcription efficiency as long as the ACCA end remains base paired.

A nested set of RNA fragments was used to determine the minimal template length required to initiate transcription. The results show that transcription efficiency increases upon increasing the length of the template and that five nonspecific nucleotides upstream of the ACCA terminus are minimally needed for detectable transcription. This total number of 9 nt is interesting, as it was recently reported that for poliovirus the minimal size of RNA for polymerase binding is 10 nt (3).

The results in this report show that stabilizing the stem of the 3′-terminal hairpin by replacing the A-U base pairs with G-C pairs is detrimental to transcription. On the other hand, previous results showed that destabilizing this 5-bp stem has no effect (8). This suggests that the stability of this hairpin stem should not exceed the calculated ΔG value of −8.2 kcal/mol for efficient transcription by the RdRp of TYMV. Previous work showed that an extension of the stem region to 6 bp with one G-C pair leads to a twofold increase in efficiency (8), whereas a further increase in stability by replacing the six-membered loop by a stable tetraloop resulted again in a drop in efficiency (8). Interestingly, a very delicate balance appears to exist between the positive effect on transcription of an increase in template length and of stem length and the negative effect of an increase in the stability of the stem, the latter as a consequence of extension of the stem. We here propose that an extension of the stem will help in binding of the template to the replicase in the same way as achieved by an increase in the size of a nonstructured single-stranded template. Optimal binding requires more than five to six nucleotides upstream of the ACCA sequence, independent whether they are in single-stranded conformation or part of an RNA A-type helix. If these nucleotides are in a structured form, like in the original 3′-terminal hairpin, this is not a problem as long as the stability of the stem does not exceed a certain value. In principle, the size per se of the 20-nt hairpin-containing template is more than sufficient, but the looping back of the 5′-proximal nucleotides (or phosphate groups) for stem formation makes these nucleotides no longer available for binding to the replicase. This is also the reason why a template containing a shorter stem of 4 (8) or 3 (3) (see above) bp shows a drop in transcription efficiency despite the lowered stability of the remaining hairpin. As shown above, the problem is not this stability but rather the lesser amount of nucleotides available at the 3′ half of the stem, binding in the active site of the replicase. This is also illustrated by the result obtained after introduction of a short stem in the HP5 template, consisting of 14 nt (Fig. 2A). Transcription efficiency now drops twofold compared to a single-stranded template of the same size (SS5) and reaches the same value as obtained for the 4-nt-shorter SS3 template. This weaker binding may be caused by the looping back of the 5′ side as discussed above, leading to an effective template size of 9 to 11 nt. We here emphasize that the extent to which hairpin loop residues (or their phosphates) contribute to replicase binding will depend on the actual conformation of the loop nucleotides. In other words, the presence of a small hairpin stem just upstream of the 3′ ACCA end can even inhibit transcription. Earlier studies showed that replacement of the 3′-terminal AUGC sequence of AlMV RNA by an ACCA sequence did not result in an active template for the RdRp of TYMV in vitro (31). This could be due to the short hairpin just upstream of the 3′ AUGC end of AlMV RNA, which has a stem region of only 3 bp and is not stabilized by a tetraloop.

It is noteworthy that just the pseudoknot structure presents a solution for the problem of the opposing effects of stem length and stability for optimal binding of the template. The three G-C base pairs forming the pseudoknot and stacking coaxially with the five base pairs of the 3′-terminal hairpin extend this quasi-continuous helix by three extra nucleotides, while the stability is only marginally increased by a few kilocalories per mole (19).

A non-base-paired NPyCPu sequence will allow internal initiation by the RdRp of TYMV in vitro.

Our previous results suggested that the RdRp of TYMV is able to initiate minus-strand synthesis internally on NCCN and NUCN sequences (8). By mutational analysis of one of the internal ACCA sites in a specially designed RNA template, we have shown in this report that efficient transcription occurs only when the 3′ N residue of the NCCN or NUCN sequence is a purine, with a preference of the G residue over the A residue. The nature of the 5′ N residue is of no importance (8), leading to the refined conclusion that a non-base-paired NPyCPu sequence will allow internal initiation by the RdRp of TYMV in vitro. These findings are in agreement with the results recently presented by Singh and Dreher (32).

The position of the initiation site in this 33-nt template also determines the efficiency of transcription. An ACCA sequence 6 to 9 nt downstream of the 5′ end of the template is not used to start transcription, in agreement with our results obtained in the determination of the minimal template length (Fig. 2). The best transcription was obtained from the site 13 to 16 nt downstream of the 5′ end. Sites further downstream were recognized less well, strongly suggesting that in addition to the upstream part, the region downstream of the initiation site contributes to efficient internal initiation. This is in agreement with results shown previously (32). Furthermore, it is shown that optimization of one of the initiation sites results in a drop in efficiency of the other initiation sites, implying that competition takes place between the different sites.

The minimal template requirements for initiation of minus-strand synthesis for the RdRp of BMV resemble those found for the RdRp of TYMV.

Having investigated initiation of transcription by the RdRp of TYMV in such detail, we wondered whether these results are characteristic for the RdRp of TYMV or whether in vitro initiation of transcription by other plant viral polymerases proceeds in a comparable way.

To answer this question, the transcription activity of two other plant viral polymerases was tested in vitro. Like the RdRp of TYMV, the RdRp of BMV is able to initiate transcription on an internal ACCA sequence, although less efficiently than the RdRp of TYMV. This internal initiation event is in agreement with the results reported earlier for the RdRp of BMV, showing that specific initiation and normal template activity were retained for RNAs with 3′ extensions (24). Like for TYMV replicase, sequences upstream as well as downstream of the internal initiation site contribute to transcription efficiencies. In addition, this result suggests that as for TYMV, the tRNA-like structure present at the 3′ end of their viral RNAs is not required for initiation of transcription in vitro. Previously it was shown by mutational analysis that three different parts of the tRNA-like structure of BMV RNA, i.e., the pseudoknot structure in the aminoacyl acceptor arm, the C residue adjacent to the 3′ terminus, and a hairpin structure 49 to 79 nt upstream of the 3′ terminus (arm C), are essential for optimal promoter activity (11). The pseudoknot was considered to be of structural importance, and the loop regions of arm C were believed to be base specifically involved in the interaction with the replicase (11). More recently it was shown that the pseudoknot structure is not required for transcription initiation and that an RNA fragment consisting of stem C with an accessible 3′ ACCA terminus could be used as a template by the RdRp in vitro (6). The C residue adjacent to the 3′ terminus was shown to be the transcription initiation site for the RdRps of both BMV (18, 24) and TYMV (31) and to be present in the internal initiation sites. Although the pseudoknot structure and arm C of the tRNA-like structure of BMV RNA are important for optimal promoter activity, our results show that in addition to the pseudoknot, also arm C is not strictly required for transcription initiation in vitro. Altogether, these results strongly suggest that the minimal template requirements for initiation of minus-strand synthesis by the RdRp of BMV resemble those found for the RdRp of TYMV.

The RdRps of both AlMV and BMV start internally on an AUGC sequence.

Surprisingly, the RdRps of both AlMV and BMV appear to start transcription very efficiently on an internal AUGC sequence. The 3′ terminus of AlMV RNA consists of a non-base-paired AUGC sequence. Previously it was shown that a construct, derived from AlMV RNA3, in which the AUGC sequence is changed into AGGC could still, although at a lower level, be transcribed by the AlMV RdRp in vitro (38). Because of this result, the authors concluded that the AUGC sequence is not involved in the recognition of RNA 3 by the viral RdRp in vitro. We would like to refine this conclusion in that the U residue of the AUGC sequence is not specifically involved in RdRp binding.

No AUGC sequence is present at the 3′ terminus of either the positive- or the negative-stranded RNA of BMV. The initiation sequence for subgenomic RNA synthesis by the RdRp of BMV is AUAC, in which initiation starts at the C residue (7). However, changing this sequence to AUGC resulted in a drop in efficiency to 4% compared to the wild-type situation, leading to the conclusion that this A residue at position +2 compared to the initiation site is important for subgenomic RNA synthesis (2). A check of the secondary structures of the various RNA fragments used in those studies by using the program MFold (39) showed that although the wild-type fragment and almost all of the mutants are largely nonstructured, the initiation site of the fragments with a mutation at position +2 becomes involved in base pairing. The AUGC sequence in our fragment is not involved in base pairing, which may be the explanation for the contradictory results. Anyhow, it still is surprising that the AUGC sequence can be used for internal initiation without the presence of the four nucleotides upstream of the initiation site that were shown to be essential for subgenomic promoter activity (30). This suggests that the mechanism used for subgenomic RNA synthesis is different from that responsible for the internal initiation on the AUGC sequence.

Both BMV and AlMV belong to the Bromoviridae, family and the modes of expression and lineage of their RdRps are the same (4). From an evolutionary point of view, it is understandable that both RdRps recognize the same sequence in vitro.

Also, the results published for TCV satellite C RNA are in line with the results presented in this report. A 22-nt hairpin and 6-nt single-stranded tail located at the 3′ terminus were previously identified as the promoter for minus-strand initiation (33). However, while a hairpin is required, the primary structure of the loop and stem regions are of limited importance both in vitro (33) and in vivo (34), although the 6-nt tail is not needed for biological active promoters in vivo (34). Furthermore, the TCV RdRp can recognize and even start internally on nonstructured short pyrimidine sequences (15).

Altogether, results obtained with the small and nonstructured RNA fragments used in these studies show that only a short specific non-base-paired sequence is needed for in vitro transcription by RdRp preparations of at least three different plant viruses. In the case of the BMV polymerase, however, two different promoter sequences can be recognized. The nonspecific residues or phosphates upstream and downstream of this initiation site are probably necessary for optimal binding of the replicase. The latter might be a general feature of all kinds of RNA and DNA polymerases.

Template specificity must be obtained in more than one way.

What else determines the specificity of the RdRp for the viral RNA in vivo if there are only a few template requirements in vitro? More and more observations suggest that for RdRps as for DdRps, cellular factors are necessary for template-specific RNA synthesis and that the core viral polymerase carries only the basic RdRp activity but not the determinant for their template specificity (for a review, see reference 20). It is this basic activity of our RdRp preparations which is observed in our in vitro experiments. Most cellular factors identified so far are subverted from the transcription or translation machineries of host cells. The presence of a tRNA-like structure offers a perfect binding site for tRNA-specific enzymes. The fact that the genomic RNAs of TYMV, TMV, and BMV interact with nucleotidyltransferase and can be aminoacylated proves that this structure is indeed recognized by these proteins (for a review, see reference 16). Also elongation factors were shown to interact with the aminoacylated RNAs of these viruses. It remains to be seen whether these proteins indeed contribute to the specific interaction of the polymerase and the viral RNA (14).