ABSTRACT
The secondary structures of the 5'-untranslated region (5'-UTR) of five different tymoviruses have been determined by structure probing, computer prediction and sequence comparison. Despite large sequence differences, there are remarkable similarities in the secondary structure. In all viruses two or four hairpins are found, most of which contain a symmetrical internal loop consisting of adjacent C-C or C-A mismatches. Since it is known that such mismatches can be protonated and protonated cytosines play an important role in RNA-protein interactions in tymoviral virions, the influence of pH on the conformation of the internal loop was studied. UV melting experiments and 1-dimensional proton NMR at varying pH values and salt concentrations confirm that the hairpins can be protonated under relatively mild conditions. The hairpin found in the 5'-UTR of erysimum latent virus, which has an asymmetrical internal loop consisting of cytosines and uridines, shows comparable behaviour. It is concluded that all tymoviral RNAs contain protonatable hairpins in the 5'-UTR. Binding experiments with empty viral capsids, however, do not yet establish a role in capsid protein binding.
Tymoviruses are a group of positive sense RNA plant viruses which consist of a monopartite genome containing three different genes and a shell of 180 identical coat protein subunits, arranged with T = 3 symmetry. The RNA genome of most tymoviruses contains a tRNA-like structure at the 3'-end, like some other plant viruses. This type of structure, which probably plays an important role in viral replication, has been studied in detail by several groups (reviewed in 1 ). The 5'-untranslated region (5'-UTR) of these viruses (or the 3'-end of the minus strand), which is ~100 nt in length, does not, however, mimic a tRNA or contain other proven conserved structural features, like the internal control region found in a number of other plant viruses (2 ). However, it is very likely that this 5'-UTR contains signals for viral replication as well. Beside a function of the 5'-UTR in viral replication, it may also be important for stimulating translation of the viral genome. The latter has been shown for the 5'-UTR of tobacco mosaic virus and several other plant viruses; for turnip yellow mosaic virus (TYMV) this is still an open question (3 ). Furthermore, the 5'-part of the viral RNA of tymoviruses contains two out-of-frame overlapping genes. These are 1888 and 5532 nt in length and code for a putative cell-to-cell transport protein (OP) and the replicase respectively. The start codons of these genes are separated by only 4 nt. It has been suggested that leaky scanning of ribosomes may lead to preferential initiation at the second start codon (4 -6 ). Since the context of the first start codon (of the OP gene) is rather poor, the replicase gene may be preferentially translated, but a role of the 5'-UTR in regulation of translation initiation cannot be excluded.
Sequences of the complete genome, including the 5'-UTR, have been determined for turnip yellow mosaic virus (TYMV; 7 ), ononis yellow mosaic virus (OYMV; 8 ), kennedya yellow mosaic virus (KYMV; 9 ), eggplant mosaic virus (EMV; 4 ) and erysimum latent virus (ErLV; 10 ). The 5'-UTRs of these tymoviruses are 78-172 nt in length and contain a methylated cap structure at their 5'-end. Interestingly, the base compositions of both the 5'- and 3'-UTRs are different from that of the coding part of the genome. The remarkable excess of cytosines (~40%) in the tymoviral genomes is not observed in the untranslated regions, which may have consequences for their secondary structure. In the 5'-UTR of several tymoviruses, a hairpin has been predicted just upstream of the AUG start codons or containing the first start codon (4 ), in analogy with several other viruses which have overlapping open reading frames (11 ). In the 5'-UTR of OYMV a direct repeat of 22 nt is found and in both regions a rather stable hairpin can be formed (8 ). Here we present an analysis of the secondary structure of the 5'-UTR of tymoviruses and present evidence for the existence of hairpins with conserved internal loops. The influence of pH and salt concentration on these structural elements was studied in UV melting experiments and by NMR measurements. The results give evidence for the presence of protonatable C-C and C-A base pairs. Biological implications of the finding of protonatable hairpins in the 5'-UTR of tymoviruses are discussed.
TYMV (type strain) was isolated from Chinese cabbage plants (Brassica pekinensis var. Witkrop) 3 weeks after inoculation, by the method of Dunn and Hitchborn (12 ). The viral RNA was isolated by a 2-fold phenol extraction and a single extraction with phenol/chloroform (50/50 v/v) and chloroform respectively, followed by ethanol precipitation. The virions of TYMV were purified by CsCl gradient centrifugation at an initial density of 1.45 g/ml for 20 h at 55 000 r.p.m. at 15oC. The band containing the B1 component (13 ), corresponding to intact virus particles, was isolated and dialysed against water or 10 mM Na acetate, pH 7.0. Artificial top component (ATC) was prepared by freezing TYMV for 2 min in liquid nitrogen and thawing at room temperature (14 ). The conditions used were 6-8 mg/ml TYMV and 30 mM Na acetate, pH 7.0, since ATC prepared under these conditions appeared as a single band on a 1% agarose gel, migrating with a rate that was slightly less than observed for virions (unpublished observations). The ATC was separated from RNA by CsCl gradient centrifugation at an initial density of 1.26 mg/ml for 20 h at 55 000 r.p.m. at 15oC.
Computer predictions of the secondary structure of the 5'-part of the genomes of TYMV, KYMV, OYMV, EMV and ErLV were performed with the Mfold program (15 ), which was used as a part of the Wisconsin GCG package, and the STAR program, including a genetic algorithm (16 ), which was run on a PC.
The enzymatic and chemical probing reactions of TYMV RNA with cobra venom nuclease, RNase T1, nuclease S1 and 1-cyclohexyl-3-(2-morpholinoethyl)carbodiimide metho-p-toluenesulfonate (CMCT) were performed as described before (17 ) and breaks or modifications were analysed by primer extension as described for rRNA (18 ) using three different primers.
A set of eight different RNA hairpins, which are shown in Figure 2 , were obtained by cloning two deoxyoligonucleotides with the sequences 5'-AGCTAATACGACTCACTATAGGCAAYYCTCGTAAGA(G/C)AATTGCCGG-3' and 5'-CTAGCCGGCAATT(G/C)TCTTACGAGRRTTGCCTATAGTGAGTCGTATT-3' in a pUC18 plasmid cut with HindIII and XbaI. Individual mutants were picked from single colonies. Digesting the resulting plasmid with MspI, followed by transcription with T7 RNA polymerase, yields RNA molecules of 25 nt in length. These RNA fragments were isolated from a 20% polyacrylamide gel. Radioactively labelled RNA was obtained by incorporation of [[alpha]-32P]ATP. T7 RNA polymerase was isolated as described (19 ). To obtain milligram amounts of the `wild-type' hairpin for NMR experiments, transcription with T7 RNA polymerase was performed from an oligonucleotide with the sequence 5'-GGCAATTGTCTTTCGAGGGTTGCCTATAGTGAGTCGTATT-3', hybridized to 5'-AGCTAATACGACTCACTATAG-3', resulting in an RNA hairpin with the sequence shown in Figure 6 . To avoid dimerization at high concentrations the sequence of the stable tetraloop in this case was GAAA.
Melting experiments of eight different hairpins were performed at various pH values. For pH 4.5-7.0, 50 mM Na acetate was used, whereas for pH 7.5 or 8.0, 50 mM Na phosphate was used. For measurements below 50 mM Na+ a lower concentration of buffer was used; for concentrations above 50 mM Na+, NaCl was added up to the desired concentration. All samples were heated at 90oC for 2 min and slowly cooled to room temperature prior to the UV melting measurements. Melting experiments were performed at a heating rate of 0.8oC/min. The absorbance at 260 nm was measured at every 0.1oC increase, using a Perkin-Elmer lambda 2 UV spectrophotometer coupled to a PTP6 temperature controller. The UV melting curves were recorded and analysed on a PC using the PETEMP and PECSS software. Melting temperatures were determined from the first derivatives of the absorption at 260 nm versus the temperature. In some cases further analysis of the curves was performed to obtain thermodynamic parameters, which were calculated by plotting lnK versus 1/T and dF/dT versus T (20 ).
Proton spectra of the RNA hairpin in 90% H2O/10% D2O were recorded at 10oC on a 500 MHz Varian Unity Plus spectrometer using the jump-return pulse sequence (21 ) for water suppression. Twelve thousand data points were collected over a sweep width of 12 001.2 Hz. The concentration of the 24 nt RNA fragment was 0.2 mM. No buffers were used and the sample was adjusted to pH with NaOD or DCl. The NMR data were processed and displayed on a SUN Sparc workstation running the NMRi software package (New Methods Research Inc., Syracuse, NY).
The identity of the imino proton resonances was determined by 1-dimensional NOE difference spectra using the jump-return pulse sequence. All imino proton resonances were irradiated for 1.0 s with low decoupler power to avoid spillover artefacts. Two thousand four hundred scans per imino proton resonance were recorded in an interleaved fashion.
A series of dilutions of ATC in H2O was prepared in the range 1 *10-12-1 * 10-8 mol/l. Aliquots of 450 [mu]l ATC solution and 50 [mu]l 10* C binding buffer (500 mM Na acetate, 100 mM MgCl2) were mixed and a few microliters (~10 000 c.p.m.) of a solution of radioactively labelled RNA was added. The mixture was incubated for 30 min at room temperature. The mixture was poured through a nitrocellulose filter (Schleicher & Schuell, 0.45 [mu]m) soaked in binding buffer for 30 min prior to the filtering. After binding the filter was washed twice with 500 [mu]l binding buffer. The amount of RNA bound to the filter was determined by Cerenkov counting. The Kd was determined from a plot of the concentration of ATC versus the fraction of RNA bound. Since the ATC particle consists of ~170 protein subunits and much more than one copy of RNA may bind to an ATC particle, we only measured an apparent Kd, which is indicated by Kd,app.
RNA isolated from the virions of TYMV was probed with the enzymes nuclease S1, RNase T1 and cobra venom nuclease and with CMCT, as described earlier for the 3'-part of the viral RNA (17 ). Sites of cleavage and modification in the 5'-part of the TYMV RNA were analysed by primer extension. A model of the secondary structure of the 150 nt at the 5'-end, based on the results of the structure probing, is shown in Figure 1 A. The hairpin loops are all sensitive to nuclease S1, whereas the symmetrical internal loops of the two hairpins in the 5'-UTR are not. This is probably due to the acidic pH at which the reaction with nuclease S1 was performed (see below). The 5'-UTR folds into two hairpins (hp1 and hp2), one of which (hp2) was proposed before (4 ). The first start codon is located immediately downstream of this latter hairpin, while the second start codon is part of a larger hairpin in the coding region. The two hairpins in the non-coding region have several elements in common: both contain a four-membered loop (in one case a stable tetraloop with the sequence GUAA), a symmetrical internal loop of 6 nt and two identical base pairs in between. The two internal loops are nearly identical.
UV melting experiments were performed with the resulting RNA molecules at different pH values in the range 4.5-8.0. A heating rate of 0.8oC/min was used. Neither decreasing the heating rate nor performing annealing instead of melting resulted in different Tm values, as was tested for the WT hairpin and some of the mutants. The relationship between the pH and the Tm for all eight hairpins is shown in Figure 2 B. Normal Watson-Crick base pairs have maximal stability at neutral pH, as is reflected in the behaviour of mutant G. At acidic pH the stability will decrease, since the bases A and C will become protonated, interfering with Watson-Crick base pair formation. The hairpin with the wild-type internal loop (`WT'), however, shows completely different behaviour: the melting temperature strongly increases upon lowering the pH and only below pH 5.0 is the helix destabilized. The difference in Tm between pH 7.5 and pH 5.0 is ~11oC. A similar behaviour is observed for most of the other TYMV-derived hairpins as well, although the hairpins have different Tm values under identical conditions, as a consequence of the different number of Watson-Crick base pairs. The Tm of mutant B, which has an internal loop identical to that in TYMV hp1 (Fig. 1 A), increases by ~8oC upon decreasing the pH from 7.5 to 5.0 and mutants A and C, which contain two neighbouring mismatches, also show a significant increase in Tm. Mutants D-F show a stabilization of ~2oC upon lowering the pH from 7.0 to 5.5. The latter mutants have only a single C-C or C-A mismatch. The pH for optimal stability apparently depends on the number of protonatable groups and is ~5.5 for the mutants having one C-C or C-A mismatch and ~5.0 for the mutants with two or three such mismatches.
Another way to visualize the differences between the various RNA molecules is shown in Figure 2 C. Here the gain in free energy (nnG) upon lowering the pH from 7.5 to the pH value at which the hairpins have a minimal nG is shown.
The influence of monovalent cations on the thermodynamic stability of the wild-type hairpin of TYMV and that of mutant G at different pH values is shown in Figure 3 . Mutant G behaves as do other small RNA hairpins (30 and references therein) and full-length TYMV RNA (31 ). The Tm shows an increase of 8.8 and 9.9oC for a 10-fold increase in Na+ concentration at pH 7.0 and at pH 5.0 respectively. The wild-type hairpin behaves completely differently. At pH 5.0, the slope of the dependence of Tm on salt concentration is much lower and decreases with increasing salt concentration. This may reflect poly(C)-like behaviour of the internal loop (see Discussion). At pH 7.0, where only a small fraction of the RNA molecules is protonated (Fig. 2 B), the dependence of Tm on salt concentration is more complex. At very low salt concentration the curve is comparable with that at pH 5.0. Above 10 mM Na+ the Tm decreases, leading to a minimum at 30 mM Na+. Above 300 mM the mutant hairpin behaves qualitatively identically to the mutant G hairpin.
A 24mer RNA hairpin, derived from TYMV hp2, was synthesized with T7 RNA polymerase and 1-dimensional proton NMR measurements were performed in 90% H2O/10% D2O. The RNA hairpin should as a maximum contain eight imino protons, one of which is formed by the G-A interaction in the stable tetraloop. At pH 5.0 and 10oC seven sharp peaks, one of which has double intensity, can indeed be observed in the imino proton region (Fig. 5 ). In addition one broad peak is observed at 11.30 p.p.m. NOE difference experiments were performed at pH 5.0. The resonances at 15.05 and 14.60 p.p.m. do not have NOEs with other imino protons, but we observed strong NOEs with an H2 proton of an adenosine at 7.96 and 7.84 p.p.m. respectively, indicating that they represent A-U base pairs. The large resonance at 13.16 p.p.m. has a NOE at 12.42 p.p.m., a strong NOE with an H2 proton of an adenosine at 6.79 p.p.m. and contacts with an amino proton resonance of a cytosine at 8.52 p.p.m. This means that both the imino proton of a C-G pair and the imino proton of an A-U pair resonate at this position. At 7.84 p.p.m. a NOE connectivity to the H2 of a neighbouring A-U base pair was also observed. Upon irradiation of the resonance at 12.42 p.p.m. NOEs were found at 13.16, 8.56 and 6.79 p.p.m. These are connectivities from an imino proton resonance to neighbouring imino protons, to the amino proton resonances of the cytidine within the base pair and to the H2 of the adenine within one of the A-U base pairs neighbouring the internal loop respectively. The imino protons resonating at 12.88 and 12.08 p.p.m. could not be assigned unambiguously. Assignments based on the NOE connectivities, temperature dependence of the resonances and pH titration (see below) are shown in Figure 5 . The identity of the weak resonance at 11.30 p.p.m. is not yet clear.
The pH dependence of the Tm of a set of RNA hairpins (Fig. 2 ) with varying internal loops indicates that protonation of C-C and C-A mismatches at low pH can be achieved in single mismatches as well as in a larger symmetrical internal loop. Both the increase in stability and the pH at which the optimum Tm is observed depend on the number of protonatable mismatches.
The results of the NMR experiments parallel the UV melting measurements (Fig. 6 and Table 1 ). The assignment of the imino protons, although not complete, makes it clear that at neutral pH at 10oC and 100 mM NaCl even the base pairs adjacent to the internal loop are not closed. Lowering the pH probably leads to a change in the structure of the internal loop, thereby also affecting the neighbouring Watson-Crick base pairs. The conformation of the protonated base pairs is not clear, but the C-A+ pair may adopt a wobble-type conformation (25 ,27 ), as has also been observed in DNA (38 ). The C-C+ pair may have a conformation comparable with a U-U mismatch, as was suggested for a symmetrical internal loop of four cytosines (26 ). Neither protonated C-A pairs nor protonated C-C pairs have imino protons, which makes NMR studies of these interactions difficult. The broad resonance at 11.30 p.p.m. may, however, result from the non-Watson-Crick base pairs formed at low pH. In the above mentioned duplex at pH 5.3 resonances were observed at 10.85, 10.05 and 9.2 p.p.m. (26 ). These observations may concern similar phenomena. The influence of NaCl on the spectra at the various pH values confirms the dual role of salt concentration in the stability of the RNA hairpin. At pH 6.0 the A-U base pairs neighbouring the internal loop are destabilized by 0.5 M NaCl, probably due to opening of the internal loop. At pH 7.0 these A-U base pairs are slightly stabilized by NaCl, parallel with what was measured by UV melting curves.
The base pair formation in the internal loops of TYMV RNA at low pH is also reflected in the nuclease S1 digestion pattern. Since digestion with nuclease S1 is performed at pH 5.5, the internal loops remain uncut.
The behaviour of the wild-type hairpin at varying salt concentrations also gives some evidence for the presence of protonated base pairs. Normal A- or B-type helices would be stabilized by cations because of neutralization of the phosphates, as is indeed observed for the mutant G hairpin (Fig. 3 ). Double-stranded poly(C) at pH 5.0 is destabilized by increasing salt concentrations (39 ). The protonatable hairpin of TYMV at pH 5.0 shows intermediate behaviour (Fig. 3 ). The shape of this curve, however, can be reconstructed from the sum of the mutant G curve and the poly(C) curve at a ratio of 7:3. This may reflect the seven Watson-Crick base pairs and the three protonated base pairs in TYMV hp2. Although in poly(C) parallel-stranded interactions are formed at low pH, differing from the interactions in the internal loop of the TYMV hairpin, the destabilizing influence of NaCl on the non-Watson-Crick base pairs seems to be comparable.
The influence of NaCl at neutral pH indicates that the protonated base pairs under these conditions can only be formed at very low salt concentration. At high salt concentration NaCl only has a stabilizing effect on the Watson-Crick base pairs. At very high or very low salt concentrations the influence of pH on the stability of the wild-type hairpin is much less than at 50 mM Na+, the concentration we used to compare the eight different hairpins (Fig. 2 ). It turns out that the stability of hairpins with C-A and C-C mismatches not only strongly depends on pH, but also on salt concentration. This makes it hard to extend the set of thermodynamic parameters used to predict the secondary structure of RNA with data for these kind of mismatches or internal loops. In 1 M salt, the concentration usually chosen for determining thermodynamic parameters, protonated mismatches are less likely than under physiological conditions.
The genomic RNA of TYMV has some physical properties in common with the wild-type hairpin described here. The melting temperature of intact TYMV RNA increases upon lowering the pH, with a difference of 10oC between pH 8.5 and pH 4.5 (40 ). Decreasing the pH also strongly influences the compactness of the RNA (31 ), also suggesting additional interactions, e.g. base pairing, at low pH. The protonation of cytosines and adenosines that was observed inside tymoviral virions by laser-Raman spectroscopy (29 ,41 ) may result from direct interaction with an acidic amino acid of the coat protein, as proposed earlier (28 ). However, the formation of protonated C-C and C-A pairs in the RNA inside the virion or in solution at low pH, in regions that are otherwise single-stranded, is also likely. Such protonated base pairs, interspersed by Watson-Crick base pairs, can be proposed for the long, C-rich single-stranded regions present in the coat protein gene of TYMV RNA (17 ). This study makes the latter more likely than the presence of i-motifs in TYMV RNA, as was proposed recently (42 ), as RNA is also probably unable to form i-motifs (43 ).
Since protonatable hairpins are a conserved feature of all tymovirus RNAs sequenced so far, a special function seems obvious. Conserved irregularities in RNA helices often point to a role in protein binding, as is the case for the TAR hairpin (44 ), the iron-responsive element (45 ) and the coat protein binding site of bacteriophage RNA (46 ). In tymoviral RNA protonated cytosines and adenosines are observed that interact with the shell of the virion (28 ,29 ). Since the 5'-UTR contains conserved hairpins with cytosines and adenosines that can become protonated far above their normal pK values, it is a very good candidate for interaction with the coat protein. Because packaging of plus strands into the virions starts during their synthesis by the replicase (47 ), the 5'-UTR may well function as the initiation site of assembly. Furthermore, binding of the coat protein to the 5'-UTR may also function as a repressor of non-structural genes. The observation that both the wild-type hairpin and the mutant G hairpin bind with similar affinity to the ATC particles does not confirm a role for the 5'-UTR in coat protein binding. The lack of specificity in binding small RNA molecules may be due to the test system and has also been observed in another virus. In turnip crinkle virus, a piece of RNA that was suggested to be the putative start site of assembly by in vivo experiments (48 ) did not show a high affinity for the coat protein in vitro and the binding of longer RNA molecules turned out to be highly cooperative (49 ). In TYMV something similar may be going on. We are currently investigating the putative role of the 5'-UTR as an origin of assembly by in vivo experiments. This may, in combination with other in vitro experiments, also shed light on the effect of the secondary structure of the viral leader on initiation of translation and plus strand synthesis.
We thank Dr H. Heus for critically reading the manuscript. This work was supported by grants from the Netherlands Foundation for Chemical Research (SON) and has been performed under the auspices of the BIOMAC Research School of the Leiden and Delft Universities.
REFERENCES
Return