ABSTRACT
Spontaneous cleavage of the less abundant form of tobacco ringspot virus satellite RNA is readily reversible. Capitalizing on earlier observations by Feldstein and Bruening that small `mini-monomer' RNAs derived from this molecule and containing little more than covalently attached ribozyme and substrate cleavage products are able to efficiently circularize, we have constructed a series of self-circularizing RNAs of precisely known size. Mixtures of linear and circular RNAs synthesized in vitro and containing 225-1132 nt could be completely resolved using a novel two-dimensional denaturing polyacrylamide gel electrophoresis system. Similar analyses of a complex mixture of coconut cadang-cadang viroid RNAs revealed the presence of relatively large amounts of a previously undescribed `fast-slow' heterodimeric RNA species in infected palms. Only a single DNA template is required to prepare each pair of circular and linear RNA markers.
Although of relatively recent discovery, circular RNAs are now known to play an important role in many biological processes. The first naturally occurring circular RNA to be described, potato spindle tuber viroid, is a small, autonomously replicating pathogenic RNA that is able to infect a variety of higher plants (1 ,2 ). Other covalently circular subviral RNAs include the satellite RNA of tobacco ringspot virus (satTRSV RNA; 3 ) and the genome of hepatitis delta virus (4 ). Replication of these pathogenic RNAs is thought to proceed via a `rolling circle' mechanism involving one or more circular templates (5 ).
Synthesis and processing of cellular mRNAs and rRNAs also generates a variety of non-linear RNAs. Following self-excision from pre-rRNA the group I intron of Tetrahymena thermophila circularizes via attack of its 3'-terminal hydroxyl group on a phosphodiester linkage near the 5'-terminus (see for example 6 ). Spliceosome-mediated maturation of eukaryotic mRNA as well as self-excision of group II introns releases a branched or `lariat' RNA in which the 5'-terminus of the intron is joined to the free 2'-OH of an adenylate residue ~25 nt upstream from the 3' splice site (reviewed in 7 ,8 ). In eukaryotic cells aberrant interactions of splice donor sites with a 3' rather than a 5' acceptor site result in a variety of unusual RNA molecules, i.e. RNAs containing scrambled exons (9 ,10 ) as well as circularized transcripts (11 ,12 ). Splice site pairing across an exon can also result in exon circularization in vitro (13 ). These circularized RNAs are more resistant to degradation than the corresponding linear forms (14 ).
During polyacrylamide gel electrophoresis under denaturing conditions, circular RNAs usually migrate more slowly than the corresponding linear molecules. A number of electrophoresis systems capable of resolving circular and linear RNA have been described (see for example 15 -18 ), but preparation of the molecular standards necessary to calibrate such gels can be laborious. Several small RNAs, such as the Tetrahymena rRNA group I intron (413 nt) or satTRSV(-)RNA (359 nt), undergo spontaneous self-cleavage and ligation in vitro. In vitro rearrangement of certain group I or group II introns results in subsequent release of circularized exons (19 ,20 ) and virtually any RNA can be circularized by oligodeoxynucleotide-directed ligation (21 ,22 ).
Here we describe construction of a set of RNAs containing sequences derived from the paperclip (or hairpin) ribozyme found in satTRSV(-)RNA. For each such RNA the readily reversible nature of the cleavage reaction generates a mixture of linear and circular molecules of precisely known length. Using an improved two-dimensional gel electrophoresis system circular and linear RNAs containing 225-1132 nt could be conveniently resolved. Analysis of coconut cadang-cadang viroid (CCCVd) RNAs of known size and sequence (23 ) confirmed the utility of this system for determining the sizes of naturally occurring circular RNA molecules.
All DNA inserts were ligated into pT7(-)miniMn = 36, a plasmid containing a cDNA copy of a specially modified version of the paperclip ribozyme from satTRSV(-)RNA, the less abundant form of satTRSV RNA (24 ). Clones designed to yield mini-monomer RNAs containing <400 nt were generated by ligating Sau3A fragments of [lambda] DNA into SalI-digested vector. Prior to ligation inserts and vector were made compatible by two base fill-in reactions using the Klenow fragment of Escherichia coli DNA polymerase I and dGTP/dATP or dTTP/dCTP respectively. Clones yielding mini-monomer RNAs containing 420-1132 nt were generated by ligating MspI-digested pBR322 (Pharmacia) directly into AccI-digested pT7(-)miniMn = 36.
Clones were screened for insert size using a PCR colony miniprep procedure and the M13 universal forward (5'-GTAAAACGACGGCCAGT-3') and reverse-5 (5'-CAGGAAACAGCTATGA-3') sequencing primers (24 ,25 ). PCR products were analyzed on 5% polyacrylamide gels containing 1× TBE buffer (89 mM Tris, 89 mM borate, 2.5 mM Na2EDTA, pH 8.3) and visualized by staining with silver nitrate. Insert size was verified by automated dye terminator sequencing (Applied Biosystems).
32P-Labeled mini-monomer RNAs were prepared by in vitro transcription of selected PCR products (2.0 µl) in 10 µl reaction mixtures containing 1× transcription buffer (10 mM Tris, 15 mM MgCl2, 2 mM spermidine-HCl, pH 7.5), 10 mM DTT, 10 U RNasin (Promega), 2.5 mM rNTPs (rGTP, rATP and rUTP), 100 µM rCTP, 5-10 µCi [[alpha]-32P]rCTP (800 Ci/mmol; Amersham) and 20 U T7 RNA polymerase (New England Biolabs). Reactions were incubated for 2 h at 37°C and then heated for 2 min at 100°C following addition of 5 µl sequencing stop buffer (10 mM NaOH, 95% formamide, 0.05% bromophenol blue, 0.05% xylene cyanol).
Denatured RNAs were loaded onto a pre-run 5% polyacrylamide gel containing 1× TBE buffer, 7 M urea and electrophoresed until the xylene cyanol dye had reached the bottom of the gel. Bands containing the desired 32P-labeled mini-monomer RNAs were identified by autoradiography and recovered by the crush and soak method (26 ). Gel-purified linear RNAs were allowed to circularize by addition of an equal volume of 2× transcription buffer and incubation at 37°C (10 min) and 0°C (30 min). The reaction was stopped by addition of an equal volume of sequencing stop buffer.
One plate from each pair of 5 mm thick glass plates used to cast polyacrylamide slab gels was larger than the other, i.e. 183 × 183 mm versus 176 × 176 mm. The inward facing surface of the smaller plate was siliconized and the plates were arranged as shown in Figure 1 A. Two plexiglass spacers (8 mm wide, 1 mm thick) were inserted in alignment with the flush edges of the plates and a third spacer was placed with its outer edge 22-25 mm from the edge of the smaller plate. A 5% polyacrylamide gel (acrylamide:N,N'-methylene bis-acrylamide 38:2) containing 1× TBE, 7 M urea was then cast to a level 8 mm below the edge of the smaller plate. Following polymerization the third spacer was moved to the edge of the smaller plate and a 5% polyacrylamide gel (acrylamide:N,N'-methylene bis-acrylamide 39:1) containing 7 M urea and 1× TBE buffer was then cast in the resulting space. As shown in Figure 1 B, the top edge of this gel contained a single sample well that is ~10 mm deep.
Feldstein and Bruening (24 ) have described modifications of satTRSV(-)RNA that result in so-called `mini-monomer' RNAs, molecules containing little more than covalently attached ribozyme and substrate cleavage products separated by a polylinker sequence. Like naturally occurring satTRSV(-)RNA, several such RNAs (e.g. miniMn = 36) not only undergo spontaneous cleavage in vitro but also readily ligate to form both linear dimers and circular molecules. Relevant structural features of miniMn = 36 are schematically illustrated in Figure 2 .
One-dimensional electrophoresis under denaturing conditions (i.e. 1× TBE buffer, 7 M urea at 50°C) revealed differences in the processing efficiencies of these eight mini-monomer RNAs (results not shown). Molecules containing 225, 320 or 428 nt processed efficiently and the fully processed linear mini-monomers were excised from the gel, eluted in 0.5 M NH4 acetate, 0.1% SDS and recovered by ethanol precipitation. For those RNAs which processed less efficiently mixtures of the fully and partially processed mini monomer (see Fig. 2 ) were recovered instead. After incubation under conditions suitable for cleavage and circularization, mini-monomer RNAs were mixed and fractionated by two-dimensional polyacrylamide gel electrophoresis.
As shown in Figure 1 A, both dimensions of our gel system contain 5% polyacrylamide, but with different degrees of crosslinking. The behavior of a mixture of six pairs of linear and circular RNAs in such a gel system is illustrated in Figure 1 B. In the first dimension migration rates of single-stranded RNAs are determined by both their size and conformation. Note that the linear form of each RNA migrates faster than the circular form. In the second dimension this effect leads to the appearance of each linear RNA in a position diagonally below its corresponding circular form. This effect is shown more clearly in Figure 4 .
Extracts from coconut palms infected with CCCVd contain a complex mixture of monomeric and multimeric viroid RNAs (28 ). As the disease progresses the `fast' forms of CCCVd which predominate during the early stages of infection are replaced by `slow' forms (29 ). Nucleotide sequence analysis suggests that the slow (i.e. 296-297 nt) form of monomeric CCCVd RNA is directly derived from the corresponding 246-247 nt fast form by a sequence duplication event involving the right hand portion of the molecule (23 ).
Figure 5 shows the resolution of linear and circular (+)CCCVd RNAs possible using our two-dimensional gel system. Note the presence of streaks extending first horizontally, then vertically from the circular forms of CCCVdF and CCCVdS to their respective linear forms. These streaks (or cross-peaks) contain linear molecules produced by cleavage during electrophoresis in either the second or first dimension. Longer autoradiographic exposures revealed similiar streaks connecting the more slowly moving fast-fast (CCCVdFF), fast-slow (CCCVdFS) and slow-slow (CCCVdSS) circular dimeric RNAs with their respective linear forms (data not shown). Likewise, the continuous `diagonal' connecting the linear CCCVd RNAs can be explained by random degradation, either before electrophoresis or during the first dimension. Long autoradiographic exposures revealed the presence of similar cross-peaks and diagonals in gels containing 32P-labeled mini-monomer RNAs synthesized in vitro (results not shown). Contamination with trace amounts of host nuclease(s) or metal ions may explain the more extensive nicking of the CCCVd RNAs.
Feldstein and Bruening (24 ) have previously shown that certain oligoribonucleotides derived from satTRSV(-)RNA and containing little more than covalently attached ribozyme and substrate cleavage products retain the ability to spontaneously circularize in vitro. The construct designated miniMn = 36 contains only 102 nt and we have taken advantage of overlapping SalI/AccI sites present in the polylinker sequence separating the ribozyme and substrate portions of this molecule to create a series of self-circularizing RNAs containing 225-1132 nt. Following purification by preparative PAGE, the linear mini-monomer RNAs can be circularized by a brief incubation in the same Tris, Mg2+, spermidine buffer used for the inital T7 transcription and ribozyme cleavage.
In contrast to oligodeoxynucleotide-directed RNA ligation (21 ,22 ), interconversion of the linear and circular forms of the modified satTRSV(-)RNAs is both rapid and spontaneous. No lengthy incubation with DNA or RNA ligase or subsequent purification to remove template oligonucleotide is required. Unlike methods involving in vitro rearrangement of group I or group II introns (see for example 19 ,20 ), only a single DNA template is required to produce each pair of linear and circular RNAs. The eight RNAs selected for use in our studies of viroids and other subviral RNAs range in size from 225 to 1132 nt, but it should be possible to construct even larger molecules that retain the ability to spontaneously circularize. An important practical limitation is the ability of polyacrylamide gel electrophoresis to resolve large RNA molecules.
Electrophoresis in polyacrylamide gels containing high concentrations of urea is widely used for both the analysis and purification of RNA under denaturing conditions (15 ). One dimensional electrophoresis under such conditions is sufficient to separate the 359 nt circular and linear forms of potato spindle tuber viroid (2 ), but a variety of other two-dimensional (see for example 30 ) and bidirectional (17 ,18 ) gel systems have also been described for fractionation of complex mixtures of RNAs to high levels of purity. Most two-dimensional systems are based on the effects of increasing the concentration of either acrylamide or urea between the first and second dimension. Increases in acrylamide concentration (usually in the ratio of 1:2) have often been used to separate RNAs containing 50-500 nt, but changes in the concentrations of both acrylamide and urea can also be combined in a single system (31 ).
Our two-dimensional gel system relies upon an increase in the degree of crosslinking rather than acrylamide concentration to preferentially retard migration of circular molecules. In its present configuration (i.e. 5% polyacrylamide with an acrylamide:bisacrylamide ratio of 38:2 in the second dimension) this gel system is able to resolve circular and linear RNAs containing as many as 1100-1200 nt. For many two-dimensional systems it is necessary to excise the relevant portion of the first dimension gel before polymerization of the second dimension gel. Because the buffer conditions are identical (i.e. 1× TBE, 7 M urea) we have found it more convenient and reproducible to cast both dimensions of our gel system before applying the sample and then simply rotate the gel between the first and second dimensions. Analysis of larger RNAs will require gels with increased pore size that still retain adequate mechanical strength to allow visualization of the various molecular species by autoradiography or staining. Agarose gel electrophoresis of glyoxal-denatured RNAs (32 ) provides one possible approach.
The use of denaturing conditions in both dimensions of our gel system allows certain linear RNAs to be identified as products of intragel cleavage of their respective circular forms. The resulting cross-peak phenomenon allows identification of sets of linear and circular RNAs and can provide a considerable analytical advantage over other gel systems where the first dimension is non-denaturing and the second denaturing. For example we have identified the linear and circular forms of a previously undescribed fast-slow heterodimer among the CCCVd RNAs analyzed in Figure 5 . The generally accepted rolling circle model for viroid replication (33 ) readily explains the presence of fast-fast and slow-slow CCCVd homodimers in infected trees, but this model does not predict synthesis of large amounts of heterodimeric fast-slow CCCVd RNA (see 23 ). The possible role of these heterodimers in formation of the monomeric slow form of CCCVd remains to be determined.
Finally, one feature of our gel system, the transitory decrease in gel temperature associated with the partial disassembly and rotation of the gel assembly between the first and second dimensions, deserves special mention. We have attempted to minimize the magnitude and duration of this effect, but controlled temperature fluctuations could also faciliate certain intramolecular rearrangements. Previous efforts to detect ribozyme activity in CCCVd and related viroids have been unsuccessful (reviewed in 34 ). Under the proper conditions our two-dimensional gel electrophoresis system might be used to activate a latent ribozyme activity.
We thank S.M.Thompson for technical assistance, Su-hwei Zhao for carrying out the automated DNA sequence analyses and S.Dube and J.M.Kaper for helpful discussions and critical review of the manuscript. P.A.F. was supported by funds provided by the USDA/NRI Competitive Grants Program (grant 91-373303-6649).
REFERENCES
§Plant Methods Development Laboratory, USDA/APHIS, Building 580-BARC East, Beltsville, MD 20705, USA