Characterization of the Escherichia coli RNA 3′-Terminal Phosphate Cyclase and Its ς54-Regulated Operon*

The RNA 3′-terminal phosphate cyclase catalyzes the ATP-dependent conversion of the 3′-phosphate to the 2′,3′-cyclic phosphodiester at the end of various RNA substrates. Recent cloning of a cDNA encoding the human cyclase indicated that genes encoding cyclase-like proteins are conserved among Eucarya, Bacteria, and Archaea. The protein encoded by the Escherichia coli gene was overexpressed and shown to have the RNA 3′-phosphate cyclase activity (Genschik, P., Billy, E., Swianiewicz, M., and Filipowicz, W. (1997) EMBO J. 16, 2955–2967). Analysis of the requirements and substrate specificity of the E. coli protein, presented in this work, demonstrates that properties of the bacterial and human enzymes are similar. ATP is the best cofactor (K m = 20 μm), whereas GTP (K m = 100 μm) and other nucleoside triphosphates (NTPs) act less efficiently. The enzyme undergoes nucleotidylation in the presence of [α-32P]ATP and, to a lesser extent, also in the presence of other NTPs. Comparison of 3′-phosphorylated oligoribonucleotides and oligodeoxyribonucleotides of identical sequence demonstrated that the latter are at least 300-fold poorer substrates for the enzyme. The E. coli cyclase gene, namedrtcA, forms part of an uncharacterized operon containing two additional open reading frames (ORFs). The ORF positioned immediately upstream, named rtcB, encodes a protein that is also highly conserved between Eucarya, Bacteria, and Archaea. Another ORF, called rtcR, is positioned upstream of thertcA/rtcB unit and is transcribed in the opposite direction. It encodes a protein having features of ς54-dependent regulators. By overexpressing the N-terminally truncated form of RtcR, we demonstrate that this regulator indeed controls expression of rtcA andrtcB in a ς54-dependent manner. Also consistent with the involvement of ς54, the region upstream of the transcription start site of the rtcA/rtcBmRNA contains the −12 and −24 elements, TTGCA and TGGCA, respectively, characteristic of ς54-dependent promoters. The cyclase gene is nonessential as demonstrated by knockout experiments. Possible functions of the cyclase in RNA metabolism are discussed.

The 2Ј,3Ј-cyclic phosphate termini are produced during RNA cleavage by many different endoribonucleases. For most of the known enzymes, among them many secretory degradative nucleases such as RNases A or T1, cyclic phosphates are formed as intermediates that are subsequently opened into 3Ј-phosphomonoesters (1,2). For some enzymes, such as tRNAsplicing endonucleases from Eucarya and Archaea, the cyclic phosphate is a final product of the cleavage reaction (4 -6). In addition, the type I topoisomerase has recently been shown to have endoribonuclease activity yielding 2Ј,3Ј-cyclic phosphate termini (7). Like some protein enzymes, ribozymes such as hammerheads, hairpins, or the hepatitis delta ribozyme generate 2Ј,3Ј-cyclic phosphate and 5Ј-hydroxyl ends during the RNA cleavage reaction (8).
That the 2Ј,3Ј-cyclic phosphate has an anabolic function in RNA metabolism emerged when it was found that eukaryotic RNA ligases require 2Ј,3Ј-cyclic ends for RNA ligation (9 -17). This requirement applies to both of the known non-organellar RNA ligases, one ligating RNA ends by the 3Ј,5Ј-phosphodiester, 2Ј-phosphomonoester linkage and the other joining the ends by the ordinary 3Ј,5Ј-phosphodiester (reviewed in Refs. 18 -20). The two ligases were shown to be involved in nuclear pre-tRNA splicing (12,14,(21)(22)(23)(24)(25)(26). In addition, the ligase generating the 3Ј,5Ј-phosphodiester, 2Ј-phosphomonoester linkage functions in splicing the unusual intron present in HAC1 pre-mRNA in yeast (27,28) and may also be involved in ligation of virusoid and viroid RNAs in plants (29 -31). Interestingly, the only known cellular RNA ligase in eubacteria, which joins RNA ends via the 2Ј,5Ј-phosphodiester, also requires 2Ј,3Јcyclic ends for ligation (32,33). Another finding uncovering a potential role for the 2Ј,3Ј-cyclic phosphate in RNA metabolism was the demonstration that the spliceosomal U6 snRNA in many organisms has a cyclic 2Ј,3Ј-phosphodiester at the terminus. The mechanism and enzymes responsible for this modification and its biological function are not known (34 -36).
In light of the importance of cyclic termini in RNA metabolism, it was interesting to discover that endonucleolytic cleavage is not the only way to generate RNA molecules bearing 2Ј,3Ј-cyclic phosphates. Such molecules can also be produced by the action of the RNA 3Ј-terminal phosphate cyclase, an enzyme that catalyzes ATP-dependent conversion of a 3Ј-phosphate at the end of RNA to the 2Ј,3Ј-cyclic phosphodiester (11). The cyclase has been purified from HeLa cells and its mechanism of action established (37)(38)(39)(40). The cyclization occurs in three steps: (a) Enzyme ϩ ATP 3 Enzyme-AMP ϩ PP i , (b) RNA-N 3Ј p ϩ Enzyme-AMP 3 RNA-N 3Ј pp 5Ј A ϩ Enzyme, and (c) RNA-N 3Ј pp 5Ј A 3 RNA-NϾp ϩ AMP.
Evidence for the initial two steps were the identification of the covalent cyclase-AMP intermediate complex (37)(38)(39)41) and the demonstration of the RNA-N 3Ј pp 5Ј A molecule accumulation when the ribose at the RNA 3Ј terminus is replaced with the 2Ј-deoxy-or 2Ј-O-methylribose (37). Reaction (c) probably occurs non-enzymatically as the result of nucleophilic attack by the adjacent 2Ј-OH on the phosphorus in the phosphodiester linkage (40).
To investigate the biological role of the cyclase, we have recently cloned a cDNA encoding the human enzyme (41). The cyclase mRNA was shown to be expressed in all tissues and cell lines analyzed. The protein is localized to the nucleus, consistent with its postulated role in RNA processing. The sequence of the human cyclase has no apparent motifs in common with any protein of known function. However, genes encoding proteins having strong similarity to the cyclase were identified in organisms belonging to all three kingdoms, Eucarya, Bacteria, and Archaea. The protein encoded by the Escherichia coli gene was overexpressed and purified and was shown to have RNA 3Ј-terminal phosphate cyclase activity (41). Conservation of the cyclase among eukaryotic and prokaryotic organisms suggests that the enzyme performs an important function in RNA metabolism.
In this work, we describe the characterization of the E. coli cyclase and demonstrate that its gene forms part of a so far uncharacterized operon that belongs to the class of operons controlled by the alternative 54 factor.

MATERIALS AND METHODS
General Procedures-Unless stated otherwise, all techniques for manipulation of DNA and RNA, commonly used buffers, and media were as described (42,43).
Plasmid pRtcB, containing the rtcB gene and its flanking sequences, was obtained by cloning of the PCR 1 -amplified E. coli DNA fragment into the EcoRV site of pBluescript II KSϩ; oligonucleotides GGCAC-GACGGTTGCAATTATCAGG and CAGCGCAATCATCCTTTTCATC were used as amplification primers. Plasmids pRtcR and pRtcR⌬N, expressing the RtcR activator and its N-terminally truncated version, respectively, were constructed as follows. The rtcR gene with flanking regions was amplified by PCR using as a template phage DD765 DNA (kindly provided by Drs. G. Plunkett and F. Blattner, University of Wisconsin, Madison, WI) and cloned into pBluescript II KSϩ, yielding plasmid pKD4. HindIII and BamHI sites were introduced by oligonucleotide-directed PCR mutagenesis at the 5Ј and 3Ј ends of the rtcR gene, respectively, using pKD4 as a template. The PCR product was cloned into HindIII and BamHI sites of the pHSG765 vector containing a chloramphenicol resistance marker (47), yielding plasmid pRtcR. In this plasmid, the rtcR gene is expressed under the control of the lacZ promoter as a translational fusion with six amino acids of ␤-galactosidase at the N terminus. pRtcR⌬N, expressing the N-terminally truncated mutant of RtcR, was constructed similarly except that the HindIII site was introduced 21 nucleotides upstream of the beginning of the region encoding the central domain of RtcR (the translation product starts with MTMITPLDFLKSG . . . ; the first L residue shown in italics corresponds to amino acid 179 of RtcR). Identity of inserts in pRtcB, pRtcR, and pRtcR⌬N was verified by sequencing.
Plasmid pMAK705cyc::Km r used for gene disruption was constructed as follows. The 1,324-bp SmaI-HindIII fragment from pREP4, containing the kanamycin resistance gene, was first cloned into SmaI-HindIII sites of pBluescript KS (Stratagene), yielding pBSKm r . Fragments corresponding to the 5Ј-flanking (620 bp) and 3Ј-flanking (669 bp) regions of the cyclase gene (obtained by PCR using the E. coli genomic DNA as a template and oligonucleotides TGCGCCATCGATCGCAATCATC-CTTTTCATC and CTTACTTTGTCGACCTGGCACAAAAAGAGATG as the 5Ј-region-specific primers and oligonucleotides ATCTCTAGAGTA-ACCTGTTGCTGCTTAATC and GGTCGCGGATCCCTCATGCCATCT-GCTGAC as the 3Ј-region-specific primers) were then cloned stepwise into pBSKm r , using the ClaI-SalI and XbaI-BamHI sites, respectively, yielding pBScyc5Ј3Ј/Km r . Finally, the 2.6-kilobase SalI-XbaI fragment from pBScyc5Ј3Ј/Km r , encompassing the kanamycin resistance gene (Km r ) and the rtcA gene upstream and downstream regions, was cloned into XbaI-SalI sites of pMAK705, yielding pMAK705cyc::Km r .
Preparation of the E. coli Cyclase and Its Substrates-The cyclase was overexpressed as His-tagged protein in the E. coli strain BL21(DE3) and purified on an Ni-nitrilotriacetic acid column as described previously (41). Cyclase purity was Ͼ95% as judged after Coomassie Blue staining of the gel (41). Protein concentration was measured by the method of Bradford using the reagent obtained from Bio-Rad and bovine serum albumin as a standard.
The oligoribonucleotide substrates, CCCCACCCCGp* and AAAAUAAAAGp*, were prepared as described previously (41). For use in some experiments, the substrates were additionally purified on a 10% polyacrylamide, 8 M urea gel. Aliquots of radioactive substrates were analyzed by digestion with RNase T2, nuclease P1, or calf intestine phosphatase (9), followed by TLC on cellulose plates in solvent A (see below). Over 90% of the label was always present as the terminal G 3Ј p*. The E. coli 5 S rRNA (Boehringer Mannheim) and the in vitro transcribed human U14 snoRNA (kindly provided by F. Dragon of this laboratory) were 3Ј-terminally labeled using [5Ј-32 P]pCp and T4 RNA ligase as described (41). Preparation of the unlabeled competitors, AAAAUAAAAG 3Ј p (referred to as RNA 3Ј p) and AAAATAAAAG 3Ј p (referred to as DNA 3Ј p), and their 3Ј-OH-terminated counterparts (referred to as RNA 3Ј OH and DNA 3Ј OH, respectively) as well as preparation of CCCCACCCCG 3Ј p and (dN) n pdN 3Ј p (representing a mixture of 3Ј-phosphorylated oligodeoxyribonucleotides (n ϭ 8 -14); referred to as DNA 3Ј p (m.n.)), was as described (41). TLC was in solvent A (isobutyric acid:concentrated NH 3 :H 2 O (66:1:33)) or solvent B (saturated (NH 4 ) 2 SO 4 :3 M sodium acetate:isopropyl alcohol (80:6:2)).
Cyclase Assays-Cyclase activity was assayed by the Norit method as described before (40,41). Unless indicated otherwise, 10-l assays K m values for ATP and GTP were calculated from Lineweaver-Burk plots. Assays were performed under standard conditions in the presence of five different concentrations of ATP (2.5-200 M) or GTP (7.4 M-1 mM). Velocities were calculated from the initial linear rates.
Labeling of the cyclase with different [␣-32 P]NTPs was performed under cyclase assay conditions except that the AAAAUAAAAGp* substrate was omitted. Reactions (10 l) contained 1.2 M [␣-32 P]NTPs (specific activity 800 Ci/mmol) or 0.33 M [␣-32 P]dATP (specific activity 3,000 Ci/mmol) and indicated amounts of the cyclase and were incubated at 25°C. The reactions were analyzed by SDS-polyacrylamide gel electrophoresis and autoradiography. Immediately before gel electrophoresis, samples were supplemented with respective unlabeled NTPs or dATP to a final concentration of 10 mM.
For AMP release assays, 1.8 ng of the cyclase was first adenylylated under standard conditions in the presence of 1.7 M [␣-32 P]ATP (specific activity 800 Ci/mmol), with no substrate added, for 2 h at 25°C. The reactions were then diluted 7-fold with cyclase assay buffer, different amounts of competitors were added, and incubations were continued for 15 min at 25°C. The reactions were analyzed as described above.
RNA Isolation from E. coli Cells-Bacterial cultures were grown in LB medium containing, when appropriate, 100 mg/liter ampicillin, 20 mg/liter kanamycin, or 10 mg/liter chloramphenicol, to an A 600 of 1.0 (49). When required, isopropyl-1-thio-␤-D-galactopyranoside was added to 1 mM concentration for the last 30 min of growth. Bacteria were harvested by centrifugation. Total RNA was isolated with RNeasy spin columns (Qiagen) according to the manufacturer's recommendations, followed by DNase treatment (20 min at 37°C, 0.1 unit/l RNase free DNase (Promega) in 50 mM Tris-HCl, pH 7.5, 1 mM MgCl 2 ). The column fractionation and DNase treatment were repeated twice. After the second DNase treatment, the enzyme was inactivated by heating for 5 1 The abbreviations used are: PCR, polymerase chain reaction; N, any of the four (A, G, C, U) nucleosides; Np, nucleoside 3Ј-phosphates; pNp or pN 3Ј p, nucleotide 5Ј,3Ј-bisphosphates; pN 2Ј  min at 72°C, and RNA was directly used for cDNA synthesis.
RT-PCR-cDNA synthesis was performed in a 40-l volume. Reactions contained the first-strand synthesis buffer for Superscript II reverse transcriptase (Life Technologies, Inc.), 10 mM dithiothreitol, 1 mM of each dNTP, 2.5 M random hexamers (Promega), 50 ng/ml RNA, and 100 units of RNase H Ϫ SuperScript II reverse transcriptase (Life Technologies, Inc.). Incubations were for 12 min at 21°C, followed by 45 min at 42°C. Reverse transcriptase was inactivated by heating for 5 min at 95°C.
PCR reactions were performed in a 50-l volume. They contained the native Taq polymerase buffer (Stratagene), 0.5 M primers, 4 l of the cDNA synthesis reaction, and 1.5 units of native Taq polymerase (Stratagene). In addition, reactions contained 2 nmol (ϳ50,000 cpm) of one of the primers that had been 5Ј-end labeled using [␥-32 P]ATP (3,000 Ci/ mmol; Amersham Pharmacia Biotech) and T4 polynucleotide kinase and purified on a 20% acrylamide, 8 M urea gel. 21 PCR cycles were performed (1 min, 95°C; 1 min, 65°C; 1 min, 72°C). Primers for the following genes were used: rtcA (TGAGCCTGTCGATGATAACC and AGCACCAGCGTACAACTTCC), rtcB (GTGAAGAGATCTACGTGACG and GATAACTTCCACCAGATCGC), rtcR N terminus (CGTTGGT-CATCGATCGACTG and GCTTCTGCCAGCAGAAACCA), rtcR C terminus (GCATACGGTCGAAGAGATCG and TCTCTGGACCTTCAC-CCTGC), rpoN (AGGTTAACTTGCTCTCGCTC and GCCAAATGGTT-GATCAAGAG), and ribosomal protein S5 (rpsE) gene (CAACGGATT-TACCACGCTTGGCA and TTCCAGCAGCGATCCAGAAAGC). The unique specificity of each primer pair was verified using E. coli genomic DNA as a template. Furthermore, in additional control experiments, it has been demonstrated that formation of the RT-PCR products with all pairs of primers was dependent upon the reverse transcription step, indicating that RNA samples were not contaminated with chromosomal or plasmid DNA (data not shown). PCR reactions with different pairs of primers were always performed with the same batch of cDNA. Products of PCR reactions were separated on 2% agarose gels. Gels were dried on the DEAE-cellulose paper (Whatman) support and exposed to an x-ray film. It has been found with the N-terminal rtcR pair of primers that within an 18-to 24-cycle range, there is a linear relationship between the number of PCR cycles and the amount of amplified product.
Primer Extension-An oligodeoxynucleotide complementary to rtcB mRNA, CCTTTGGTCCACATTTTTACC, was 5Ј-end 32 P-labeled and purified as described above. Annealing reactions contained 20 pmol of the primer and 20 g of RNA in 160 mM Hepes-KOH, pH 7.7, 1 M NaCl, and 1 mM Na 2 -EDTA. After heating at 95°C for 2 min and annealing overnight at 48°C, the mixture was ethanol precipitated and dissolved in 20 l of the first-strand synthesis buffer for Superscript II reverse transcriptase containing 10 mM dithiothreitol, 1 mM of each dNTP, and 400 units of Superscript II reverse transcriptase. After 1 h at 42°C, the reaction was processed for analysis on a 10% acrylamide, 8 M urea gel.
Cyclase Gene Replacement-Chromosomal insertion and excision steps with pMAK705cyc::km r , resulting in the replacement of the nearly complete rtcA (the remaining rtcA regions encode the 7 N-terminal and 34 C-terminal amino acids) by km r , were performed in the E. coli strain MC1061 as described (46). Southern blot analysis and PCR reactions, using primers specific for the cyclase coding and flanking regions, and for the kanamycin resistance gene, performed with DNA isolated from both the knockout and control MC1061 strains, confirmed that the gene replacement had taken place.
Characterization of the Human cDNA Encoding an RtcB-like Pro-tein-A human cDNA clone (GenBank R61436) encoding RtcB-like protein was obtained from the I.M.A.G.E. Consortium (Livermore, CA). The clone was sequenced on both strands, using appropriate oligonucleotide primers. The sequence encoding 29 N-terminal amino acids missing in this clone was obtained by RT-PCR using sequence information derived from the highly conserved mouse cDNA clone (GenBank W42119) to design the forward PCR primer. Poly(A) ϩ RNA from HeLa cells, kindly provided by P. Pelczar of this laboratory, was used to generate single-stranded cDNA as described above. The PCR reaction was performed on 200 ng of cDNA using as a forward primer the oligonucleotide ATGAGTCGTAACTACAACGATG. This sequence encompasses the ATG initiation codon (shown in italics) and the adjacent 5Ј-untranslated region nucleotides of the mouse cDNA. The backward primer, ATCATTCACATAGAAAACACC, was complementary to the human cDNA sequence, 120 nucleotides downstream of the ATG codon. The PCR-amplified fragment was cloned into the EcoRV site in pBluescript II KSϩ, and the insert was sequenced. The sequence of the 5Ј-terminal part of the cDNA encoding human RtcB-like protein was further confirmed by inspection of two recent expressed sequence tag entries (GenBank AA090429 and AA 232068).
Computer Analysis-Unless indicated otherwise, sequence management and analysis were performed with the GCG Wisconsin package of programs (Genetics Computer Group, Madison, WI) on the UNIX platform. Protein sequence alignments were performed with the ClustalW 1.5 program (50), using default parameters; the alignment was improved manually. Shading of amino acids was performed with the BOX-SHADE program at the BOXSHADE WWW server at the University of Lausanne (http://ulrec3.unil.ch/software/boxshade/boxshade.html). Phylogenetic analysis was performed with the PHYLIP package (51). The ClustalW 1.6 multiple-sequence alignment served as an input file for the PROTDIST program, which generated the distance matrix. The distance matrix was further analyzed by the NEIGHBOR program to obtain an evolutionary tree. The statistical significance of phylogenetic relationships was assessed by bootstrapping analysis with the BOOTSTRAP program; 500 data sets were analyzed by the PROTDIST program, generating distance matrices for all these data sets. Phylogenetic trees were generated from these matrices with the NEIGHBOR program. A consensus tree was generated with the CONSENSE program. Helix-turn-helix motifs were predicted with the helix-turn-helix program version 1.0.5 (52).

RESULTS
Requirements of the E. coli Cyclase-To compare the human and E. coli enzymes, we have studied requirements of the overexpressed and purified bacterial protein using oligoribonucleotides specifically labeled at the 3Ј-terminal phosphate, AAAAUAAAAGp* or CCCCACCCCGp*, as substrates. The cyclization reaction with the E. coli enzyme showed a pH optimum of 8.0 -8.5 (Fig. 1A). The reaction required the presence of divalent cations, either Mg 2ϩ or Mn 2ϩ . In the presence of Mn 2ϩ , the enzyme activity was 50 -70% higher than in the presence of Mg 2ϩ . With both cations, a broad optimum was found at 1-4 mM. No activity was seen when 2 mM Mg 2ϩ or Mn 2ϩ was replaced with 2 mM Ca 2ϩ , Zn 2ϩ , or Cu 2ϩ (Table I and  guishes the E. coli and human enzymes. In the presence of Mn 2ϩ ions, the human protein, either purified from HeLa cell extracts (39) or bacterially overexpressed (Table I), showed only 5 and 17%, respectively, of the activity seen with Mg 2ϩ . Activity of the E. coli enzyme was similar at 0 and 0.1 M NaCl (in addition to 30 mM Hepes-KOH). Addition of NaCl to 0.2 or 0.4 M inhibited cyclization by 30 and 70%, respectively. At 20 or 50 mM, sodium phosphate did not inhibit the reaction, but sodium pyrophosphate was strongly inhibitory (86 and 98% inhibition, respectively) ( Table I and data not shown).
The activity of different nucleoside triphosphates as cofactors in the cyclization reaction was compared. ATP was found to be the most efficient cofactor, and considerable activity was also seen with GTP. UTP, CTP, and dATP were much less active (Fig. 1B). K m values for ATP and GTP were 20 and 100 M, respectively (see "Materials and Methods"). Previously determined values for the human enzyme are 6 M (ATP) and 200 M (GTP) (38,39). ADP and AMP were not active as cofactors (Table I). Likewise, ATP could not be replaced by either ␣,␤methylene (AMPCPP), ␤,␥-methylene (AMPPCP), or ␤,␥-imido (AMPPNP), all nonhydrolyzable analogs of ATP. ATP␥S was about 20% more active than ATP (Table I). Similar observations were previously made for the human cyclase (37,38,40).
Substrate Specificity-The foregoing experiments have shown that the E. coli enzyme can catalyze the cyclization of the 3Ј-terminal phosphate in synthetic oligoribonucleotides such as CCCCACCCCGp, AAAAUAAAAGp, and AAAAUAAA-AGCp (41). We have tested the ability of two natural RNAs, the E. coli 5 S rRNA and human U14 snoRNA, to act as substrates. The RNAs were modified by ligation of [5Ј-32 P]pCp to the 3Ј terminus. Incubation with the E. coli cyclase resulted in the formation of 2Ј,3Ј-cyclic phosphate termini in both RNAs as determined by digestion with nuclease P1, followed by TLC on cellulose plates (data not shown; see "Materials and Methods").
For the human cyclase, we have previously demonstrated that prolonged incubation of the 3Ј-phosphorylated oligodeoxyribonucleotides with an excess of the enzyme generates low amounts of products bearing dN 3Ј pp 5Ј A at the 3Ј terminus. Competition experiments have shown that 3Ј-phosphorylated oligodeoxyribonucleotides are ϳ500-fold poorer substrates than oligoribonucleotides (41). Two different assays were used to compare the ability of the 3Ј-phosphorylated oligoribonucleotide AAAAUAAAAG 3Ј p, referred to as RNA 3Ј p, and the oligodeoxyribonucleotide of equivalent sequence, AAAATAAA-AG 3Ј p, referred to as DNA 3Ј p, to act as substrates for the E. coli cyclase. Using a competition assay (Fig. 3A), it was found that RNA 3Ј p and another oligoribonucleotide, CCCCACCCCG 3Ј p, are approximately 1,000-fold better competitors than DNA 3Ј p in the cyclization reaction carried out with the radiolabeled AAAAUAAAAG 3Ј p* as a substrate. A mixture of 3Ј-phosphorylated oligodeoxyribonucleotides ((dN) n pdN 3Ј p, n ϭ 8 -14), obtained by limited digestion of a synthetic 80-mer oligodeoxyribonucleotide with micrococcal nuclease and referred to as DNA 3Ј p(m.n.), was also an about 300-fold poorer competitor than RNA 3Ј p. 3Ј-Hydroxyl-terminated oligoribonucleotides and oligodeoxyribonucleotides, RNA 3Ј OH and DNA 3Ј OH, did not compete with the cyclization of AAAAUAAAAG 3Ј p* even when added at 10,000-fold excess (Fig. 3A).
In the second assay, the oligoribonucleotides and oligodeoxyribonucleotides were compared for their ability to release AMP from the performed adenylylated enzyme complex. The complex was formed by preincubation of the protein with [␣-32 P]ATP. Incubations were then continued in the presence of increasing quantities of different oligonucleotides. Addition of 33 fmol of RNA 3Ј p decreased the amount of the complex by more than 50% (Fig. 3B, lane b), and no complex was detected when 330 or 3,300 fmol of RNA 3Ј p was added in the second incubation (lanes c and d). In contrast, incubation in the presence of 330 or 3,300 fmol of RNA 3Ј OH, DNA 3Ј p, or DNA 3Ј OH (lanes c and d) did not result in the release of the label from the preformed complex. In the presence of even higher amounts (20 and 200 pmol), DNA 3Ј p but not DNA 3Ј OH resulted in AMP release from the complex (Fig. 3B, lanes e and f).
Competition experiments, similar to those shown in Fig. 3A, were also carried out using nucleoside 3Ј-monophosphates (Np) and nucleoside 5Ј,3Ј-bisphosphates (pN 3Ј p) or nucleoside 5Ј,2Јbisphosphates (pN 2Ј p) as competitors. No significant competition for the cyclization of AAAAUAAAAG 3Ј p* was observed  2. Labeling of the cyclase with ␣-32 P-labeled ribonucleoside triphosphates and [␣-32 P]dATP. Assays contained 12 ng (A) or 1.2 g (B) of the cyclase and were incubated in the presence of indicated labeled triphosphates for the time shown above the autoradiograms. Positions of protein size markers (in kDa) are indicated. when 6,000-fold molar excess of each compound was used (Fig.  3A, inset). A small inhibitory effect of pC 3Ј p and pdCp was probably unspecific as incubation of radiolabeled [5Ј-32 P]pC 3Ј p with an excess of the cyclase for 5 h at 25°C did not result in detectable formation of pCϾp as verified by TLC (data not shown). Identical inhibition with pC 3Ј p and pdCp also argues against these compounds acting as cyclase substrates because oligodeoxyribonucleotides or 2Ј-deoxy-terminated oligoribonucleotides were found to be much less efficient substrates for the enzyme (Fig. 3 and data not shown).
Taken together, the results presented in this section indicate that E. coli cyclase can efficiently use as substrates 3Ј-phosphate-terminated RNA molecules of different sequence and base composition and that ribonucleoside 3Ј-monophosphates and 5Ј,3Ј-bisphosphates do not act as substrates. Furthermore, 3Ј-phosphate-terminated DNA molecules are two to three orders of magnitude poorer substrates than RNAs.
Structure of the Cyclase Operon-The gene encoding the cyclase, named rtcA (for RNA terminal phosphate cyclase orfA), is positioned at 76 min on the E. coli K12 chromosome. rtcA probably forms part of an uncharacterized operon, the structure of which is schematically shown in Fig. 4. Another ORF transcribed in the same direction, named rtcB, is present upstream of rtcA. As the termination codon of rtcB is immediately followed by the AUG of rtcA, it is very probable that the two genes are transcribed into a dicistronic mRNA. Inspection of the region positioned upstream of rtcB suggested that transcription of the rtcB/rtcA unit may involve the alternative 54 factor. The region likely to correspond to the rtcB/rtcA promoter contains TTGCA and TGGCA elements, the sequences and spacing of which are identical with the Ϫ12 and Ϫ24 elements constituting recognition signals for 54 (53)(54)(55). Moreover, there are two putative binding sites for the integration host factor, known to be involved in transcription of many 54 -specific promoters (56 -59), further upstream. Finally, the ORF positioned upstream of rtcB, named rtcR, which is transcribed in the opposite direction, encodes a protein having all the features of 54 -dependent regulators ( Fig. 4; see below). Initiation of transcription by 54 -RNA polymerase holoenzyme requires additional activator proteins that bind to enhancerlike sequences typically positioned 100 -200 bp upstream from the transcription start site (reviewed in Refs. 58, 60 -62). A schematic structure of the rtcR-encoded protein is shown in Fig. 5A. Generally, 54 -specific regulators comprise three different domains: the N-terminal regulatory or sensory domain, the highly conserved central domain responsible for ATP hydrolysis and interaction with 54 , and the C-terminal DNA binding domain, which contains a helix-turn-helix motif (reviewed in Refs. 58, 60 -62). The central domain of RtcR shows 28 -39% identity and 56 -61% similarity with counterparts in other regulators of this class and contains conserved regions C1-C7 found in other members of the family (61). The Nterminal domain does not have significant sequence similarity to any of the known 54 class regulators or other proteins deposited in the data bases; it also does not contain conserved Asp residues characteristic of the members of two component systems (61,63,64). Alignment of C-terminal domains of RtcR and a representative selection of known 54 regulators (Fig. 5B) suggests that RtcR has an atypical DNA binding domain with the helix-turn-helix motif containing a 20-amino acid-long turn. Helix-turn-helix motifs with turns as long as 20 amino acids have been identified in some structurally characterized DNA binding proteins (65). Phylogenetic analyses, performed with the PHYLIP program, indicated that within a family of 54 class activators, RtcR does not resemble any particular known activator more than others (Fig. 5C). A similar conclu- sion was reached when phylogenetic analyses were carried out separately for each of the three domains of the regulator (data not shown).
Involvement of RtcR and 54 in Transcription of the Cyclase Operon-54 -specific regulators are usually constitutively expressed but not constitutively active (reviewed in Refs. 58,61,62,64). For some regulators, it has been shown that deletion of the sensory domain derepresses ATPase activity of the regulator and makes it constitutively active (64). To investigate whether the RtcR regulator plays a role in expression of the cyclase operon, we have tested the effect of overexpression of wild-type and N-terminally truncated forms of RtcR on transcription of rtcA and rtcB in the strain YMC10 and its derivative YMC22, containing the inactivated 54 gene. RNA isolated from either control bacteria or from bacteria transformed with plasmids expressing the full-length RtcR or its N-terminally truncated form, RtcR⌬N, was reverse transcribed using random hexamers as primers. Resulting cDNAs were utilized as templates for PCR, using pairs of oligodeoxynucleotide primers specific for rtcA and rtcB and, as controls, the genes encoding 54 (rpoN) and a ribosomal protein S5 (rpsE). Expression of the rtcR gene was monitored with two pairs of primers, one specific for the region encoding the N-terminal part of RtcR and another specific for the C-terminal part.
Results of RT-PCR analysis are shown in Fig. 6. As expected, overexpression of the mRNA encoding the wild-type RtcR could be visualized with both the rtcR N-terminal and C-terminal pairs of primers (lanes 4), whereas that of RtcR⌬N mRNA could be visualized with only the C-terminal primers (lanes 5-7). Expression of rtcR in nontransformed bacteria was found to be very low and could only be visualized when more PCR cycles were performed (data not shown). Transcription of rtcA and rtcB was observed in YMC10 cells expressing the N-terminally truncated (lanes 5) but not wild-type (lanes 4) RtcR. Neither rtcA nor rtcB was expressed in the YMC22 54 knockout strain transformed with pRtcR⌬N alone (lanes 6). However, transcription of both genes took place when YMC22 was additionally cotransformed with pTH7, which encodes 54 (lanes 7). Activation of rtcA and rtcB transcription by the N-terminally truncated but not wild-type RtcR was also observed in E. coli BL21(DE)pLysS cells transformed with plasmids in which expression of the regulators is driven by the T7 promoter (data not shown).
To obtain additional evidence that expression of the cyclase operon involves 54 , the transcription start site of the rtcB/rtcA mRNA was determined by primer extension. The analysis was performed with RNA isolated from both YMC10 and the YMC22 54 knockout mutant, each overexpressing the N-terminally truncated form of RtcR. Consistent with the experiments presented in Fig. 6, the primer extension product was only identified when RNA from YMC10 was used as a template (Fig. 7). The 5Ј-end of the RNA mapped to the C residue positioned 11 nucleotides downstream of the putative Ϫ12 TT-GCA box present in the rtcB upstream region.
Taken together, the results presented above indicate that both 54 and the 54 -specific regulator RtcR are involved in expression of the cyclase operon.
rtcA Is a Nonessential Gene-To investigate whether the rtcA gene product is essential for E. coli growth, the central 90% of the rtcA coding region was replaced by homologous recombination with a gene coding for kanamycin resistance in the strain MC1061 (46). The replacement was confirmed by Southern and PCR analyses (data not shown; see "Materials and Methods"). No differences in growth of the parent and the rtcA null strain FIG. 4. Structure of the cyclase operon and the sequence of the intergenic promoter region between rtcB and rtcR. Genes constituting the cyclase operon are shown as filled black arrows. The rtcB transcription start site, as established by primer extension analysis (Fig.  7), is indicated by a bent arrow. Translation start sites for rtcB and rtcR are marked with arrows, and putative Shine-Dalgarno sequences are underlined. The Ϫ12 and Ϫ24 consensus sequences characteristic of 54 promoters are boxed. Sequences resembling the most highly conserved regions of the consensus binding site, AATCAAN 4 TTA, for the integration host factor (IHF) are indicated by boxes connected by the dotted lines. The boxed sequences are flanked by AT-rich regions, which contain at many positions nucleotides conforming with the extended integration host factor consensus (80,81). Inspection of the upstream region of the rtcA/rctB promoter did not identify obvious sequence repeats that could act as likely targets for RtcR. on either LB or a minimal M9 medium were observed (data not shown).
Genes Similar to rtcB Are Present in Archaea and Eucarya-Genes encoding cyclase-like proteins are conserved among Eucarya, Bacteria, and Archaea (41). By searching sequence data bases, we have found that also genes similar to rtcB, the gene likely to be cotranscribed with the cyclase gene in E. coli, are present in other organisms such as Methanococcus jannaschii and Caenorhabditis elegans, representing two other kingdoms. Expressed sequence tags encoding RtcB-like proteins in humans have also been identified. The human cDNA was fully sequenced, and the protein encoded by it is included in the alignment shown in Fig. 8 (for sequences of RtcB-like proteins in other species, see legend). The E. coli RtcB and other related proteins listed in Fig. 8 show no significant sequence similarity or motifs in common with proteins of known function deposited in public data bases. The noteworthy feature of RtcB proteins is the presence of six conserved histidine residues, suggesting possible involvement of a metal ion in RtcB function.
The RtcB protein was overexpressed in E. coli as a C-terminal His 6 -tag fusion and purified on the Ni-nitrilotriacetic acid column. The recombinant protein contained no detectable RNase or RNA ligase activities (the latter assayed with 5Ј-hydroxyl and cyclic-phosphate-terminated RNA substrates). Likewise, it had no phosphodiesterase activity opening the cyclic phosphate in RNA or pGϾp to either a 3Ј-or 2Ј-monoester or a phosphatase activity removing the 3Ј-phosphate from RNA. Addition of RtcB had no effect on the cyclization reaction catalyzed by the E. coli RNA 3Ј-phosphate cyclase (data not shown).

DISCUSSION
Recent cloning of a cDNA encoding the RNA 3Ј-terminal phosphate cyclase from humans led to the identification of cDNAs and/or genes encoding cyclase-like proteins in diverse eukaryotes and also the bacterium E. coli and the archeon M. jannaschii. The protein encoded by the E. coli gene was overexpressed and shown to have RNA 3Ј-phosphate cyclase activity (41). In this work, we have studied the requirements and substrate specificity of the overexpressed E. coli cyclase. We have also established that the cyclase gene forms part of a so far uncharacterized operon, expression of which is controlled by 54 and the transcriptional regulator RtcR. The requirements and other enzymatic properties of the overexpressed E. coli cyclase are generally similar to the properties of the human protein, although some important differences are apparent. One of them is the ability of Mn 2ϩ ions to replace Mg 2ϩ in reactions catalyzed by the E. coli but not by the human cyclase. Both enzymes preferentially use ATP as a cofactor but are also able to utilize GTP and, much less efficiently, other ribonucleoside triphosphates. GTP is a relatively more efficient cofactor with the E. coli than with the human enzyme. Apparent K m values for ATP and GTP are, respectively, 20 and 100 M for the E. coli protein and 6 and 200 M for the human counterpart. Like the human protein (37)(38)(39)(40)(41), the E. coli enzyme undergoes adenylylation, and the adenylyl group can be released from the preformed cyclase-AMP complex protein upon incubation with the 3Ј-phosphorylated RNA but not the 3Ј-OH-terminated RNA. Covalent labeling of the bacterial cyclase with [␣-32 P]GTP and, much less efficiently, with [␣-32 P]CTP, [␣-32 P]UTP, and [␣-32 P]dATP was also observed. Although we have not directly validated the identity of nucleotidyl groups attached to the E. coli enzyme, previous demonstration that the human cyclase can undergo guanylylation, cytidylylation, and uridylylation (39) argues that the same is also true for the E. coli enzyme. Altogether, the results discussed above strongly suggest that the mechanism of 3Јphosphate cyclization by the E. coli enzyme is similar to that FIG. 6. Transcription of rtcA and rtcB genes in response to overexpression of the wild-type and N-terminally truncated RtcR. Bacterial strains and plasmids used for transformations are indicated at the top. cDNA prepared from either nontransformed bacteria or bacteria transformed with indicated plasmids was used as a template for PCR, using pairs of oligodeoxynucleotide primers (one of them being 32 P-labeled) specific for rtcA, rtcB, rtcR (two pairs of primers, specific for regions encoding N-terminal and C-terminal parts of the gene, respectively), and genes coding for 54 40). This conclusion is further supported by the observation that E. coli cyclase, like its human counterpart (37,41), can inefficiently convert the 3Ј-terminal phosphate in the oligodeoxyribonucleotide AAAATAAAAGp into the product terminated with dN 3Ј pp 5Ј A. 2 Substrate specificities of the bacterial and human enzymes seem to be similar. Both enzymes can use a variety of 3Јphosphorylated RNAs and oligoribonucleotides bearing either purine or pyrimidine 3Ј-terminal nucleotides (Refs. 11, 38, 40, and 41 and this work). It was demonstrated previously that nucleoside 3Ј-phosphates and nucleoside 5Ј,3Ј-diphosphates do not act as substrates for the human cyclase (11,38,40). We have found that these compounds also do not act as substrates for the E. coli enzyme. Using the AMP release and competition assays (Fig. 3), we have compared the ability of oligoribonucleotides and oligodeoxyribonucleotides of identical sequence to act as substrates for the bacterial cyclase. As with the human protein (41), the oligodeoxyribonucleotides were found to be at least 300-fold poorer substrates than the oligoribonucleotides. Hence, RNA and not DNA molecules are the most likely physiological substrates for the bacterial enzyme.
The E. coli cyclase gene, rtcA, forms part of the previously uncharacterized operon containing two additional ORFs. The ORF positioned immediately upstream, named rtcB, encodes a protein of unknown function that, like the cyclase, is also highly conserved among Eucarya, Bacteria, and Archaea. Inspection of sequences of RtcB and related proteins from other organisms, as well as preliminary experiments aimed at identifying an enzymatic activity potentially associated with RtcB, did not provide any clue as to the function of the protein.
Another ORF of the cyclase operon, named rtcR, is positioned upstream of the rtcA/rtcB unit and is transcribed in the opposite direction. It encodes a protein having features of 54 -dependent regulators ( Fig. 5; reviewed in Refs. 58,61,64). By overexpressing the N-terminally truncated form of RtcR, we have demonstrated that this regulator controls expression of rtcA and rtcB in a 54 -dependent manner. Further evidence that 54 is involved in expression of rtcA and rtcB is provided by the presence of the TTGCA and TGGCA elements centered 13 and 24 bp upstream of the rtcA/rtcB transcription start site established by primer extension (Figs. 4 and 7). Both the sequence and position relative to the transcription start site of these two elements conform with the consensus Ϫ12 and Ϫ24 boxes characteristic of all studied 54 -dependent promoters (53)(54)(55).
The observation that overexpression of the N-terminally truncated but not full-length RtcR induces transcription of rtcA and rtcB suggests that, as in the case of several previously characterized 54 -specific regulators (e.g. XylR, DmpR, DctD, and LevR (66 -69)), the N-terminal domain represses activity of RtcR, and its deletion makes the regulator constitutively active. Physiologically, activation of most 54 -specific regulators is brought about either by phosphorylation or by binding of specific effector molecules to the sensory domain that, in the uninduced state, represses the ATPase activity of the protein required for isomerization of the promoter complex (reviewed in Refs. 55, 58 -61, 64). RtcR is not a member of the twocomponent system family of regulators (63,64), so its activation is unlikely to involve phosphorylation. A potential effector interacting with RtcR remains to be identified. With the exception of Myxococcus xantus in which rpoN null mutants are 2 P. Genschik and W. Filipowicz, unpublished results. nonviable (70), in all bacteria studied to date, 54 and consequently also genes or operons dependent on it were found not to be essential for growth. Products of genes controlled by 54 factors have very diverse physiological functions. They are involved in specialized metabolic processes such as utilization of alternative carbon and energy sources or assimilation and fixation of nitrogen, in the production of extracellular structures such as flagellum and pilus, or in the synthesis of virulence determinants (54,55,59,64,71). In Caulobacter crescentus and M. xantus, some important cell differentiation decisions involve 54 and/or 54 -specific transcription regulators (72)(73)(74)(75). Identification of physiological conditions leading to activation of the rtcA operon in E. coli would greatly help in establishing a biological function of the RNA 3Ј-phosphate cyclase in bacteria.
The findings that two different eukaryotic RNA ligases (18 -20) and also the RNA ligase identified in E. coli and some other bacteria (32,33) all require 2Ј,3Ј-cyclic phosphate termini suggested that cyclases may be involved in RNA ligation pathways by generating (or regenerating) cyclic phosphate ends (40,41). The E. coli RNA ligase studied by Abelson and co-workers (32,33) preferentially uses yeast tRNA half molecules as substrates in vitro and ligates them via an unusual 2Ј-5Ј-phosphodiester bond. Physiological substrates of this ligase and also enzymes responsible for generation of cyclic ends required for ligation remain unknown. Like rtcA, the gene encoding the RNA ligase in E. coli is not essential (33). In the archeon Desulfurococcus mobilis, the 23 S pre-rRNA contains an intron, and its cleavage by the endonuclease seems to generate splicing intermediates containing 3Ј-phosphomonoesters (76). It is not known whether the terminal phosphate has to undergo cyclization prior to the ligation step. In eukaryotes, the spliceosomal U6 snRNA and also a small fraction of some other low molecular weight RNAs have 2Ј,3Ј-cyclic phosphate termini (34,36); it is possible that a cyclase is involved in formation of these structures. No RNA molecules bearing cyclic phosphate ends have been as yet identified in prokaryotes. In certain E. coli strains, phage T4 induces cleavage of the host tRNA Lys by anticodon nuclease, which produces 2Ј,3Ј-cyclic phosphate and 5Ј-OH ends. The damaged tRNA Lys is normally repaired by the action of two phage-encoded enzymes: polynucleotide kinase and RNA ligase (77,78). Although previous experiments have already indicated that anticodon nuclease itself generates cyclic termini (77) and that T4 RNA ligase requires 3Ј-OH and not a cyclic phosphate for ligation (79), we have tested for a possible effect of cyclase gene disruption on T4 growth in the absence and presence of anticodon nuclease expressed from the cosmid. Both the parent and the cyclase mutant strains supported T4 growth and showed normal restriction of the ligase and kinase mutants of T4 in the presence of anticodon nuclease. 3 Further experiments are required to establish the biological role of the RNA 3Јterminal phosphate cyclase in E. coli and other organisms.