INTRODUCTION

RNA splicing reactions can generate circular exon sequences when the donor (5) splice site is 3 of the acceptor (3) splice site. That is, in a single exon ligation event, if the 3 end of the exon is joined to a splice site at an upstream rather than a downstream position, the exon will be circularized (1, 2, 3, 4). Circular RNAs generated by splicing have been demonstrated with in vitro manipulated group I and group II intron sequences (2, 3, 4). In those examples the fundamental mechanism of the splicing reaction remains the same, and the circular product is novel because the linear order of the splice sites is different. No natural examples of group I or group II introns producing circular exons have been documented; however, naturally occurring examples of circular exons generated by spliceosomal splicing have been observed, indicating that circular RNAs from nuclear pre-mRNAs can be produced in vivo at low levels by mis-splicing (5, 6). In that circles generated by pre-mRNA splicing are rarely detected, it is likely that there exists a mechanism to minimize such events in spliceosomal splicing. At this time there is little reason to think that these naturally occurring circular exons have a specific function.

Examples of RNA circles are found in nature among certain plant pathogens, the viroids, virusoids, and satellite viruses where the circular structure of the RNA plays an essential role in the replication of the genome (7, 8, 9). In mammals, the hepatitis delta virus (HDV)¹ is an example of a replicon consisting of a circular RNA (10). In these pathogens, the RNA appears to be circularized as a result of end to end ligation of a linear monomer processed from a replication intermediate longer than unit length (11).

Circular intron RNA in the form of either true circles or branched circles (lariats) are generated naturally either post-splicing (some group I introns and archaeal introns) (12, 13, 14) or as an integral part of the splicing mechanism (group II and nuclear pre-mRNA introns) (15, 16, 17). Studies have shown that the circular group I introns have an enhanced half-life relative to linear RNA in Xenopus developing embryos (18) and in Escherichia coli (19), suggesting that the circular RNAs are more stable in some in vivo systems. One might predict that circular exons could be more stable than the intron circles or lariats. This is because the circularized group I introns are capable of autocatalyzed hydrolytic cleavage at the site of ligation and as a result are relinearized (20). A similar fate is realized with branched circles (lariats) since they are targets for the debranching enzyme (21). Circular exons produced by splicing reactions will contain no autolytic sequence and, unless by design, will not contain specific sequences or structures that will hasten their demise as circles.

A general method for circularizing RNA in vitro and in vivo may be useful because circular nucleic acids have unique features that make them valuable for a variety of studies (22, 23, 24, 25). We are particularly interested in the potential for in vivo applications toward RNA-based control of gene expression and therapeutics (26, 27, 28, 29, 30). One reason for this interest is that circularized RNAs could be more resistant to nuclease degradation than linear counterparts of the same sequence. For applications that may include antisense (27, 31, 32), decoy inhibitors (33, 34, 35), triplexes (26, 28, 36, 37), or ribozymes (29, 38, 39, 40, 41, 42, 43), circularized RNAs may prove beneficial if they have a longer half-life without an associated loss of the desired activity relative to the linear RNA. We have previously shown that it is possible to generate ribozyme circles in vitro that are active and resistant to degradation in HeLa cell nuclear and cytoplasmic extracts (44, 45). Even if circular forms of natural RNA are less stable than the chemically modified synthetic RNA (46), for therapeutic uses, advantages of unmodified RNAs may include lower toxicity and antigenic properties. Perhaps more important, the advantage of stabilizing RNA through circularization by splicing would be the potential for expression in vivo, something that cannot be done with chemically modified RNA. Before many of these questions can be addressed it is necessary to demonstrate efficient circularization of the RNAs of interest in vivo.

In this study we have investigated the possibility of generating circular trans-acting ribozymes in vivo using group I-permuted intron-exon (PIE) sequences (Fig. 1). Ford and Ares (4) have shown that a permuted T4-derived group I intron can generate circular mRNA sequences in E. coli and in yeast, demonstrating that the splicing reaction can be used to generate a circular messenger RNA in cells. We were interested in determining if ribozymes produced as circles in E. coli using the Anabaena group I PIE sequence would retain function. We chose to produce ribozymes as circles because the reaction catalyzed by the ribozyme would provide a simple assay for function. In previous in vitro studies, we have shown that a circular form of the trans-acting hepatitis delta virus ribozyme (RC1) is active as the circle (44). More recently, we have demonstrated that a circular form of the Bacillus subtilis RNaseP RNA (C-PRNA) is fully active in vitro (45). Both of these were produced using Anabaena group I PIE sequences (2). In this study, we provide evidence that both of these trans-acting ribozymes can be expressed as circles in E. coli. These results suggest that circularization could be a viable means of producing active ribozymes in vivo. We think the method could be generalized to produce a variety of RNAs that can function as circles.

EXPERIMENTAL PROCEDURES

Construction of the plasmids pRC1 and pRNP-RNA1 has already been described (44, 45). These plasmids generate T7 transcripts in vitro capable of splicing to produce circular HDV ribozyme and circular RNaseP RNA, respectively. pET-RC1 and pET-PRNA1 were prepared by recloning the PIE-ribozyme sequences into pET3a. pRC1 was cut with EcoRI and BamHI, and pRNP-RNA1 was cut with KpnI and BamHI. The PIE sequences containing the HDV ribozyme and the PRNA sequences were isolated, and the ends were made blunt using T4 DNA polymerase. These fragments were ligated into BamHI-digested and end-filled pET3a vector. A hybridization probe for RNaseP was prepared by ligating a 391-bp PCR product of PRNA into SmaI-digested pTZ18U. Plasmid DNA was isolated and purified by CsCl density gradient centrifugation (2).

Cells transformed with the PIE-ribozyme plasmids were grown in LB medium containing 20 µg/ml ampicillin to an A₆₀₀ of 0.6 and induced with 0.4 mM isopropyl-1-thio--D-galactopyranoside. After an additional 2 h, total RNA was isolated by the acid guanidinium/phenol/chloroform method (47). Cells from 20-ml cultures were resuspended in 1 ml of lysis buffer (4 M guanidinium thiocyanate, 25 mM sodium citrate (pH 7.0), 0.5% sarcosyl, and 0.1% -mercaptoethanol) followed by the addition of 0.1 ml of 2 M sodium acetate (pH 4.0), 1 ml of water-saturated phenol, and 0.2 ml of chloroform/isoamyl alcohol. The contents were mixed thoroughly with each addition. The homogenate was incubated on ice for 15 min and centrifuged at 14,000 rpm in a microcentrifuge for 20 min at 4 °C. The RNA was precipitated twice with isopropyl alcohol (1 ml), washed with 75% ethanol, dried, and dissolved in TE (10 mM Tris (pH 8), 1 mM EDTA).

Cellular RNA (10 µg) from E. coli was fractionated by two-dimensional polyacrylamide gel electrophoresis under denaturing conditions. The first dimension was 5 or 7% polyacrylamide, and the second dimension was 7 or 12% polyacrylamide (percentage depended on the RNA sample used). To visualize total RNA, the gels were stained with Stains-all (USB). For Northern analyses, the RNA was electroblotted onto a nylon membrane (Hybond-N⁺, Amersham Corp.) using a Trans-Blot Semi-Dry transfer unit (Bio-Rad). The filters were rinsed with 2 × saline sodium citrate (SSC) buffer, air-dried, and UV-irradiated for 5 min to cross-link the RNA onto the filter. Filters were prehybridized in 6 × SSC, 5 × Denhardt's solution (0.1% Ficoll, 0.1% polyvinylpyrrolidone, and 0.1% bovine serum albumin), 5 mM EDTA, 0.5% sodium dodecyl sulfate (SDS), and 100 µg/ml denatured salmon sperm DNA for 1 h at 65 °C. Denatured radiolabeled probe was added and hybridized at 65 °C for 16 h. Filters were washed with 2 × SSC containing 0.5% SDS and then with 1 × SSC containing 0.5% SDS for 15 min each at room temperature, and finally with 0.1 × SSC containing 0.5% SDS at 65 °C for 15 min. The filters were air-dried and autoradiographed or imaged using a Molecular Dynamics Phosphorimager.

Circular and linear ribozymes (HDV ribozyme and PRNA) and the complementary strand of PRNA were synthesized by in vitro transcription of BamHI-digested DNA of pRC1, pRNP-RNA1, and pPRNA(AS), respectively, using T7 RNA polymerase at 37 °C for 2 h as described previously (44). Precursor tRNA^Asp was prepared using BstNI-cut pDW152 (48). To label the RNA, [-³²P]CTP was included in the transcription reaction. The RNAs were purified on denaturing polyacrylamide gels and eluted.

Oligonucleotide 13-mer substrates, DHS1 (5 UUC·GGGUCGGCAU) and DHS3 (5 UUC·GGCACGGCAU) (49), used for the HDV ribozyme trans-cleavage reactions were 5 end-labeled using [-³²P]ATP and T4 polynucleotide kinase and purified on a 20% polyacrylamide gel.

Trans-cleavage activity of the HDV ribozyme was assayed by incubating either the gel-purified C-HDV ribozyme or E. coli total RNA containing the in vivo expressed C-HDV ribozyme with 5 end-labeled oligonucleotide substrate (DHS1 or DHS3) in a 20-µl reaction containing 40 mM Tris-HCl (pH 8.0), 11 mM MgCl₂, and 1 mM EDTA as described previously (44). The 5 end-labeled trinucleotide product was separated from the substrate on polyethyleneimine-TLC plates using 1 M LiCl as solvent (49).

RNaseP activity was assayed by incubating C-PRNA with uniformly labeled pre-tRNA^Asp in high salt buffer (100 mM MgCl₂, 800 mM NH₄Cl, 50 mM Tris-HCl (pH 8.0), 0.05% Nonidet P-40, and 0.1% SDS) as described (50). C-PRNA (enzyme) and the pre-tRNA^Asp (substrate) were pre-heated separately, and the reactions were initiated by mixing. Aliquots were quenched with an equal volume of stop mix (10 M urea, 200 mM EDTA and loading dyes). Reaction products were separated on an 8% denaturing polyacrylamide gel, and the radioactivity was quantified using the phosphorimager.

About 1 µg of the total RNA containing the in vivo expressed C-PRNA or 100 ng of the gel-purified C-PRNA was used as a template for reverse transcription (RT). The RT reaction was carried out using the GeneAmp RNA PCR kit (Perkin-Elmer) and a primer complementary to positions 62-40 of the B. subtilis PRNA (5-AGCATGGACTTTCCTCTAC) (51). After 15 min at 45 °C, the enzyme was inactivated by heating at 95 °C for 5 min. The cDNA was amplified by the PCR (35 cycles; 94 °C for 30 s, 42 °C for 30 s, and 72 °C for 30 s) using a second primer complementary to positions 175-196 (5-TGAAAGTGCCACAGTGACGAAG). The 290-bp PCR-amplified DNA fragment was purified on agarose gel and sequenced.

One microgram of gel-purified 290-bp RT-PCR-amplified DNA fragment was sequenced using a 5 ³²P-labeled primer that is complementary to positions 62-40 of B. subtilis PRNA (5-pAGCATGGACTTTCCTCTAC, where p is a 5 monophosphate group) and modified T7 DNA polymerase (Sequenase). The products were separated on a 6% sequencing gel.

To sequence the circular HDV ribozyme, about 0.25 µg of purified C-HDV ribozyme was annealed to a 5-³²P-labeled primer (5-pCTAGCCCAGGTGGGCCGCGAGGAGGCTGGCCGAAGCCATTCGC) at 85 °C for 3 min, followed by slow cooling to room temperature. The primer was then extended in the presence of dideoxynucleotide triphosphates using avian myeloblastosis virus reverse transcriptase (USB) by incubating at 45 °C for 15 min, and the reaction products were separated on a 10% sequencing gel.

RESULTS

The Anabaena group I PIE sequence used in previous studies (2) was modified to generate a circular form of PRNA in vitro (Fig. 2A). The circular form of the B. subtilis PRNA was shown to be as active as the linear B. subtilis PRNA (45). For expression in E. coli, the PIE-PRNA sequence was recloned from pRNP-RNA1 into the T7 expression vector, pET3a. The resulting plasmid, pET-PRNA1, was transformed into BL21(DE3); T7 RNA polymerase expression was induced with isopropyl-1-thio--D-galactopyranoside, and total RNA was isolated by the acid guanidinium/phenol/chloroform procedure (47). The RNA was fractionated by polyacrylamide gel electrophoresis under denaturing conditions. On a 6% gel there was a prominent band at the top of the lane that was not present in the control (RNA isolated from cells transformed with pET3a). RNA isolated from this part of the gel was found to have RNaseP activity (data not shown). To better characterize the in vivo produced RNA with RNaseP activity, denaturing two-dimensional gels were used (5 and 7% acrylamide in the first and second dimension, respectively). In the second dimension, nonlinear species of RNA often display altered mobility relative to the linear species with which they comigrated in the first dimension. This causes the nonlinear products to migrate off the diagonal of linears. A prominent spot, near the top of the gel, was observed to migrate off the diagonal to a position that corresponded to a C-PRNA marker prepared in vitro (Fig. 3A). This spot (labeled C) was found to hybridize strongly to a probe for B. subtilis PRNA (Fig. 3B). This result suggested that the spot corresponds to a circular form of the PRNA. A minor spot that comigrated with circle in the first dimension but with the linear marker in the second dimension would correspond to circular RNA that was nicked between running of the first and second dimension. A spot that migrated with a linear marker and on the diagonal would correspond to a linear form in both dimensions. It appeared from these data that a large fraction of the PRNA was circular.

A major concern in detecting circle production in E. coli was that splicing could occur during the RNA isolation, rather than in the cells prior to isolation. In a series of experiments using purified precursors in extraction buffer, no splicing product was detected (data not shown). Splicing requires both Mg²⁺ and guanosine, and although the RNA extraction buffer (47) introduced neither, it was possible that small amounts of each could have been released along with the RNA and thus allowed splicing. To rule out such a possibility, precursor RNA prepared in vitro was incubated in extraction buffer with 5 or 25 mM Mg²⁺ and with or without 0.1 mM GTP. Time course experiments indicated that even after 2 h in the presence of Mg²⁺ and guanosine substrate no splicing was detected in the extraction buffer (data not shown). This suggested that the extraction buffer conditions (denaturant and low pH) were sufficient to prevent splicing.

To examine the accuracy of the splicing reaction, a cDNA copy of the C-PRNA ligation junction was sequenced. Using total RNA from the induced cells and a primer complementary to positions 62-40 of the B. subtilis PRNA (51), a cDNA that spans the putative ligation junction of the circle was synthesized and amplified by the PCR using a second primer (positions 175-196). An expected 290-bp fragment of DNA was isolated from an agarose gel and sequenced. The same procedure was followed using a gel-purified PRNA circle prepared in vitro. Both the in vitro and in vivo generated RNA sequences were identical and unambiguously showed the expected splice junction sequence (Fig. 4). Thus, the sequencing data together with the two-dimensional gel results allowed us to conclude that the RNA was the predicted circle.

To confirm that the circular PRNA produced in vivo had ribozyme activity, RNA was eluted from the portion of the gel that contained the circular species and incubated with pre-tRNA substrate. The in vivo synthesized C-PRNA cleaved the pre-tRNA substrate completely (Fig. 5A), but the rate of cleavage was slightly less than a similar amount of in vitro transcribed and purified C-PRNA (Fig. 5B). It is likely that, despite gel fractionation, the in vivo prepared RNA was less pure than the in vitro synthesized ribozyme, and the observed rate could reflect lower enzyme concentrations and the presence of other RNA. No RNaseP activity was detected in the RNA eluted from other regions of the gel.

Increased RNaseP activity was detected in the total RNA isolated from cells expressing the PRNA-containing PIE precursor (Fig. 5, C and D). Reconstitution experiments in which purified PRNA circle was added to the total RNA isolated from non-ribozyme expressing cells indicated that the rate of the reaction was proportional to the ribozyme concentrations. Using this assay, it was estimated that the active C-PRNA ribozyme constitutes about 0.8% of the total isolated RNA mass (Fig. 6). By the same method, it was found that expression in E. coli of a linear form of PRNA resulted in levels of PRNA activity similar to that seen with the PIE construct (data not shown). This would suggest that for PRNA, both a circular and linear form of the ribozyme were expressed efficiently in vivo. It is possible that the B. subtilis PRNA may be processed to an active form by endogenous enzymes and perhaps stabilized by association with the protein component of E. coli RNaseP (52).

An HDV trans-acting ribozyme has been derived from the self-cleaving antigenomic sequence of hepatitis delta virus (49). In vitro, the circular form of this ribozyme (Fig. 2B) behaves in a manner very similar to the linear forms, binding and cleaving a single-stranded RNA substrate that can base pair with a seven nucleotide sequence in the ribozyme (44). The PIE sequence that is able to generate a C-HDV ribozyme was recloned from pRC1 into the pET3a vector for expression in E. coli. Total RNA was isolated from the induced cells and fractionated on a two-dimensional polyacrylamide gel containing urea. In this case no band corresponding to the circular ribozyme was apparent by staining (Fig. 7A). However, when probed with an oligonucleotide complementary to the HDV ribozyme, a distinct spot was detected running off the diagonal of linear fragments and appearing to co-migrate in the second dimension with in vitro synthesized circle (Fig. 7B). The in vitro product has previously been characterized and demonstrated to be a circle with HDV ribozyme trans-cleavage activity (44). A minor spot was also consistently seen in the high molecular weight region off the diagonal (Fig. 7B); the identity of this species has not been established, but it may represent another circular splicing product.

Although small C-HDV ribozyme produced in vivo could not be isolated, the sequences across the ligation junction was confirmed by primer extension sequencing of unfractionated RNA from cells expressing the purported C-HDV ribozyme. Due to unspliced precursor in the total RNA, it was anticipated that sequence of the splice site in precursor would be apparent along with the sequence of the ligation junction of spliced RNA. To facilitate the analysis, in vitro transcribed gel-purified C-HDV ribozyme was also sequenced to provide a reference (Fig. 8A). Sequencing of the total E. coli RNA revealed the expected mixture of sequences corresponding to the ligation junction of the C-HDV ribozyme and the precursor RNA (Fig. 8B). We interpret this result to indicate that a splicing reaction occurs in vivo to generate the predicted ligation junction.

Total RNA from the cells expressing C-HDV ribozyme was incubated with a 5 end-labeled 13-mer oligonucleotide (DHS1) that is a substrate for the antigenomic derived ribozyme. Cleavage of DHS1 releases a 5 ³²P-labeled trinucleotide that can be detected either by gel electrophoresis (49) or by TLC on PEI plates. A control substrate (DHS3) was also used. DHS3 is the same size as DHS1 but differs in having two base changes such that it is not cleaved in vitro by a ribozyme with the specificity of RC1. However, it is cleaved by a ribozyme with an altered binding sequence (49). The total RNA contained an activity able to cleave DHS1 but not DHS3 (Fig. 9) that strongly suggested that the ribozyme was active and showed normal specificity. Cleavage of these substrates was not detected in RNA isolated from non-ribozyme expressing cells. Reconstitution experiments as described above for C-PRNA indicated that about 0.2% of the isolated RNA mass was active ribozyme (Fig. 10).

Can a linear HDV ribozyme be expressed in E. coli? A linear version of an HDV ribozyme (PDC7) (53) was cloned into pET3a to generate pET-PDC7. PDC7 is very similar to RC1, differing only in that the end of P4 is not closed with a loop. When PDC7 was expressed in vivo, about 104 and 110 nucleotides of vector-derived sequence would be expected to flank the 5 and 3 ends of the ribozyme sequence, respectively. Activity was not detected in the RNA isolated from cells nor in the in vitro transcripts. Failure to detect the activity of the ribozyme was not due to total degradation or absence of the ribozyme in the RNA because expression of a high molecular weight RNA containing the HDV ribozyme could be demonstrated by hybridization (data not shown). The linear ribozyme (PDC7) is active when synthesized in vitro as a small ribozyme (53); therefore, we attributed the lack of activity to the flanking sequences. No further attempt was made to express the linear ribozyme devoid of flanking sequences in vivo. These results indicated that the group I PIE construct offers a promising method for producing a small artificial ribozyme in an active form from a larger transcript in vivo.

DISCUSSION

The concept of manipulating RNA to artificially regulate gene expression has been around for more than a decade. The particulars of the approach can take several forms, decoys or competitors (33, 35), antisense (31), ribozymes (38, 39, 42), and triplex forming sequences (28). While each of these strategies holds promise (27), undoubtedly there is room for technical improvements. Uncertainties in the usefulness of these approaches stem from real and perceived problems associated with RNA delivery, expression, and intracellular stability. If the nucleic acid is to be delivered exogenously, some measure must be taken to protect the nucleic acid from degradation in the intracellular environment. To this end, chemical modifications during synthesis appear to be effective (46, 54, 55, 56). However, if the sequence is to be expressed as an RNA in the cell, then it must be expressed in an active and preferably stable form. This approach presents a different challenge because chemical modification is not an option. Strategies directed at stabilizing transcripts tend to rely on incorporation of terminal sequences forming hairpin-loop structures thought to enhance resistance to degradation (57, 58). As we learn more about control of RNA stability, the outcome of these approaches will become more predictable; yet sequences added to the termini will continue to have the potential for interfering with the function of the RNA that is being expressed.

We are interested in applying a group I splicing reaction that generates active circular ribozymes in vitro (44, 45) to the production of ribozymes as well as other RNAs of therapeutic potential in vivo. Regardless of how they are generated, it is possible that circular RNAs will be more resistant to degradation due to the lack of free ends and specific sequences and structures normally associated with the ends. In addition, because the circles are excised from longer transcripts, the effects of sequences contributed by the promoter, termination, and 3 end processing signals will be removed. On the other hand, circularization may be a source of other problems; for example, it might be expected that circularization could interfere with the structure, activity, or pairing possibilities of the RNA. The ability to produce circles easily in vitro and in vivo will allow us to address questions of function and stability. As an early step in this direction, we investigated whether active circular ribozymes can be expressed in vivo with the group I PIE splicing reaction.

Using two entirely different trans-acting ribozymes, we have asked if these sequences can be expressed as circles and, if so, will they be active as ribozymes? The data indicated that both PRNA and HDV ribozyme were produced as circles in vivo. The circles can be detected in RNA isolated from the cells, and it appears unlikely that splicing is occurring during the isolation of the RNA. If splicing had occurred during RNA isolation it would have had to occur under conditions in which we do not otherwise detect any splicing activity, and moreover, to account for the extent of circle formation, the splicing would have to be much faster than we have otherwise observed. An in vivo assay for circle formation would be desirable, and although it was not possible to do such an experiment with the ribozymes used in this study, Ford and Ares (4) reported expression of a circularized message in E. coli generated with a permuted T4 group I intron. The circular nature of the RNA was confirmed by the finding that the products of the HDV ribozyme splicing reaction comigrated with in vitro characterized circular HDV ribozyme (44) and displayed an enhanced electrophoretic retardation when the percentage of the acrylamide in the gel was increased. The PRNA circle also showed an anomalous migration pattern in the gels, and the sequence of the expected ligation junction was demonstrated. Finally, total RNA isolated from cells expressing the ribozyme sequences contain ribozyme activity. In the case of the PRNA, RNA isolated from the portion of the gel containing the circle contained PRNA activity, so we conclude that the circular ribozyme was produced in E. coli. Although we have not demonstrated that the circular ribozyme is active in E. coli, it may be worth emphasizing the obvious; because a ribozyme is used to generate the circle, at least some artificial ribozymes are active in E. coli, otherwise no circles would be produced.

An important question that is still to be answered is that of stability of RNA circles relative to RNA linears. Previous studies suggested that circular RNAs have longer half-lives in vivo (18, 19). With B. subtilis PRNA, our experiments had indicated that linear and circular forms expressed in E. coli were active to similar levels; however, it is possible that this is not a good test for expressing nonendogenous RNA since the protein and RNA components of RNaseP are interchangeable between these bacteria (52). The HDV ribozyme presented a different set of problems since it was not possible to design a construct capable of expressing high levels of the linear ribozyme devoid of flanking sequences. The flanking sequences appeared to interfere with the activity of the ribozyme, and if this was the case, additional steps would be necessary to process the ribozyme sequence. So while these experiments did not address the question of relative stability of small linears versus circles, they did point out the advantages in using autoprocessing group I PIE sequences to produce active ribozymes in vivo.