An RNA Ligase from Deinococcus radiodurans*

Although DNA repair pathways have been the focus of much attention, there is an emerging appreciation that distinct pathways exist to maintain or manipulate RNA structure in response to breakage events. Here we identify an RNA ligase (DraRnl) from the radiation-resistant bacterium Deinococcus radiodurans. DraRnl seals 3′-OH/5′-PO4 RNA nicks in either a duplex RNA or an RNA: DNA hybrid, but it cannot seal 3′-OH/5′-PO4 DNA nicks. The specificity of DraRnl arises from a requirementfor RNA on the 3′-OH side of the nick. DraRnl is a 342-amino acid monomeric protein with a distinctive structure composed of a C-terminal adenylyltransferase domain linked to an N-terminal module that resembles the OB-fold of phenylalanyl-tRNA synthetases. RNA sealing activity was abolished by mutation of the predicted lysine adenylylation site (Lys-165) in the C-terminal domain and was reduced by an order of magnitude by deletion of the N-terminal OB module. Our findings highlight the existence of an RNA repair capacity in bacteria and support the hypothesis that contemporary DNA ligases, RNA ligases, and RNA capping enzymes evolved by the fusion of ancillary effector domains to an ancestral catalytic module involved in RNA repair.

Polynucleotide ligases seal broken 3Ј-OH and 5Ј-PO 4 ends in DNA or RNA. The ubiquity and essentiality of DNA ligases attest to the premium placed on the avoidance of strand breaks in the DNA genome. It is sensible to think that polynucleotide ligases initially evolved to catalyze RNA repair/recombination reactions during an aboriginal RNA-protein world. There are a limited number of documented RNA repair pathways in current biology that exploit RNA ligases, including bacteriophage tRNA restriction-repair, fungal tRNA splicing, nonspliceosomal mRNA splicing in the yeast unfolded protein response, and kinetoplastid mRNA editing (1)(2)(3)(4).
The discovery of these exemplary RNA breakage and rejoining pathways was an unexpected dividend of basic research in diverse fields (phage genetics, kinetoplast gene expression, signaling from the ER to the nucleus, tRNA gene structure) rather than a targeted effort to mine the subject of RNA repair. Indeed, it is not clear if these few cases comprise sporadic relics of an ancient RNA repair system or whether RNA repair is a widespread phenomenon in current biology. Our hypothesis is that RNA repair is "hiding in plain sight," if only we knew how to see it. Specifically, we posit that RNA repair enzymes exist in diverse niches and species and that they can be uncovered by integrating recent biochemical and structural insights to ligase mechanism with phylogenetic analyses of cellular and viral proteomes.
RNA ligases join 3Ј-OH and 5Ј-PO 4 RNA termini via a series of three nucleotidyl transfer steps. (i) RNA ligase reacts with ATP to form a covalent ligase-(lysyl-N)-AMP intermediate plus pyrophosphate. (ii) AMP is transferred from ligase-adenylate to the 5Ј-PO 4 RNA end to form an RNA-adenylate intermediate (AppRNA); and (iii) ligase catalyzes attack by an RNA 3Ј-OH on the RNA-adenylate to seal the two ends via a phosphodiester bond and release AMP (5)(6)(7). There are two different branches of the RNA ligase family exemplified by bacteriophage T4 RNA ligase 1 (Rnl1) and RNA ligase 2 (Rnl2), respectively (5,8). Rnl2-like proteins are found in all three phylogenetic domains, although few have been characterized. RNA ligase activity has been demonstrated for coliphage T4 and vibriophage KVP40 Rnl2 and for the RNA-editing ligases REL1 and REL2 of Trypanosoma and Leishmania (8 -11). Putative Rnl2-like RNA ligases are encoded by certain mycobacteriophages and eukaryotic viruses and by many species of Archaea (8,12). Among the Rnl2-like enzymes, a biological function is firmly established only for the protozoan RNA-editing ligases, which repair programmed breaks in mitochondrial mRNAs that direct insertion or removal of uridylates in order to establish a proper translational reading frame (13)(14)(15). Rnl1-like ligases are few in number, and they have a relatively narrow phylogenetic distribution that is limited, as far as we know, to bacteriophages, fungi, and baculoviruses (16 -19). Rnl1 functions in vivo to repair a break in the anticodon loop of Escherichia coli tRNA Lys triggered by phage-activation of a host-encoded anticodon nuclease (1). Fungal tRNA ligase is an Rnl1-type enzyme that mediates the sealing steps of both tRNA splicing and mRNA splicing in the unfolded protein response (2,3).
Although bacteriophage tRNA restriction provided the initial insights to RNA repair biology, little attention has been paid to the issue of whether bacteria themselves (as opposed to phage) might specify enzymes that catalyze RNA repair reactions. Here we identify and characterize an Rnl2-like RNA ligase from Deinococcus radiodurans, a bacterium that is famous for its extreme resistance to ionizing radiation (20). The D. radiodurans RNA ligase (henceforth DraRnl) 1 has a novel domain structure and distinctive substrate and cofactor specificities in sealing RNA strands.

EXPERIMENTAL PROCEDURES
Recombinant DraRnl-A DNA fragment containing the D. radiodurans DRB0094 gene was amplified by PCR from total genomic DNA with oligonucleotide primers designed to introduce an NdeI restriction site at the translation start codon and a BamHI site 3Ј of the stop codon. The PCR product was digested with NdeI and BamHI and inserted into pET16b (Novagen) to generate the expression plasmid pET-DraRNL. Amino-terminal truncation alleles DraRNL-(106 -342), DraRNL-(126 -342), and DraRNL-(136 -342) were generated by PCR amplification using sense primers that introduced an NdeI site and a Met codon in lieu of the codons for Phe-105 or Pro-135 or an NdeI site at the Met-126 * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
‡ To whom correspondence should be addressed. E-mail: s-shuman@ ski.mskcc.org. codon. Alanine substitution mutations K165A and E278A were introduced by PCR using the two-stage overlap extension method. The truncated and/or mutated DraRNL DNAs were digested with NdeI and BamHI and then inserted into pET16b. The inserts of the wild-type and mutant RNL plasmids were sequenced completely to exclude the acquisition of unwanted changes during amplification and cloning.
Wild-type and mutant pET-DraRNL plasmids were transformed into E. coli BL21(DE3). Cultures (1 liter) of E. coli BL21(DE3)/pET-DraRNL were grown at 37°C in Luria-Bertani medium containing 0.1 mg/ml ampicillin until the A 600 reached ϳ0.6. The cultures were chilled on ice for 30 min, adjusted to 0.1 mM isopropyl-␤-D-thiogalactopyranoside and 2% ethanol, and then incubated at 17°C for 16 h with continuous shaking. Cells were harvested by centrifugation, and the pellet was stored at Ϫ80°C. All subsequent procedures were performed at 4°C. Thawed bacteria were resuspended in 30 ml of buffer A (50 mM Tris-HCl (pH 7.5), 0.5 M NaCl, and 10% sucrose). Lysozyme, phenylmethylsulfonyl fluoride, and Triton X-100 were added to final concentrations of 1 mg/ml, 1 mM, and 0.1%, respectively. The lysates were sonicated to reduce viscosity, and insoluble material was removed by centrifugation. The soluble extracts were applied to 1-ml columns of nickel-nitrilotriacetic acid-agarose (Qiagen) that had been equilibrated with buffer A. The columns were washed with 8 ml of the same buffer and then eluted stepwise with 4-ml aliquots of 25, 50, and 200 mM imidazole in buffer B (50 mM Tris HCl (pH 8), 0.5 M NaCl, and 10% glycerol). The polypeptide compositions of the column fractions were monitored by SDS-PAGE. The His 10 -DraRnl proteins adsorbed to the column and were recovered predominantly in the 200 mM imidazole eluates. The eluates were dialyzed against buffer containing 50 mM Tris-HCl (pH 8), 200 mM NaCl, 2 mM DTT, 2 mM EDTA, 10% glycerol, and 0.1% Triton X-100 and then stored at Ϫ80°C. Protein concentrations were determined by SDS-PAGE analysis of serial dilutions of the DraRnl preparations in parallel with serial dilutions of a BSA standard. The gels were stained with Coomassie Blue, and the staining intensities of the DraRnl and BSA polypeptides were quantified using a digital imaging and analysis system from Alpha Innotech Corporation.
RNA Ligase Substrates-Oligoribonucleotides were purchased from Dharmacon (Lafayette, CO) and deprotected as instructed by the vendor. Oligodeoxyribonucleotides were purchased from BIOSOURCE. The RNA or DNA strands were 32 P-5Ј-end-labeled using T4 polynucleotide kinase and [␥-32 P]ATP and then gel-purified. To form the RNA-labeled tandemly nicked RNA:DNA hybrid, the 32 P-labeled RNA strand was annealed to a 2-fold molar excess of an unlabeled 5-OH-terminated DNA strand in 150 mM NaCl, 10 mM Tris-HCl (pH 8), and 1 mM EDTA by incubation for 10 min at 65°C, followed by incubation for 15 min at 37°C and then for 30 min at 22°C. To form the singly nicked substrates, mixtures of a 5Ј-PO 4 strand, a 3Ј-OH strand, and a complementary template strand were annealed at a molar ratio of 1:5:2. Substrates containing gaps and flaps were prepared as described (21).
RNA Ligase Assay-Reaction mixtures (10 l) containing components as specified were incubated for 15 min at 37°C. The reactions were quenched by adding 15 l of 80% formamide and 100 mM EDTA. The samples were heated for 5 min at 95°C and then analyzed by electrophoresis through a 14-cm 8% polyacrylamide gel containing 7 M urea in 45 mM Tris borate, and 1 mM EDTA. The ligation products were visualized by autoradiography. Ligation activity with the tandemly nicked dsRNA:DNA hybrid substrates was quantified by scanning the gel with a Fuji BAS-2500 imaging apparatus. The radioactivity signal was determined for each "n-mer" (e.g. n ϭ 1 for the input substrate strand, n ϭ 2 for linear dimer, etc.) that was visualized. The radioactivity for individual n-mers in each reaction was summed, and each n-mer species was expressed as the fraction of the total. The amount (in femtomoles) of 5Ј 32 P-labeled 24-mer comprising each labeled n-mer was determined by multiplying this fraction by the known amount of input 24-mer substrate strands. Each linear n-mer is necessarily generated as a consequence of "n Ϫ 1" ligation events. Thus, the amount (in femtomoles) of ligation required to form each n-mer product was calculated for each n-mer by the following equation: fmol (ligation) ϭ fmol (24-mer) ϫ [(n Ϫ 1)/n]. The total amount of strand ligation (fmol of 5Ј-ends joined) for each reaction was then determined by summing the ligation events for each n-mer product in the ladder. were subjected to mutational analysis in the present study are indicated by a vertical bar (͉) (essential) and a plus sign (ϩ) (nonessential), respectively. The translation start sites of serial N-terminal deletion mutants of DraRnl are indicated by arrows above the sequence.

A Candidate Deinococcus RNA Ligase with a Novel Domain
Structure-The genome of the radiation-resistant bacterium D. radiodurans consists of 3,195 predicted genes contained on two chromosomes, one megaplasmid, and one plasmid (20). The DRB0094 gene on the megaplasmid encodes a 342-aa polypeptide, DraRnl, that we classify as a putative RNA ligase. Homologous proteins of similar size are encoded by the bacterium Streptomyces avermitilis (355-aa) and the bacteriophage 44RR.8t (366-aa). An alignment of these three putative ligases reveals that their primary structures are conserved throughout their lengths, with 84 positions of side chain identity/similarity (Fig. 1). The C-terminal portions of DraRnl and its homologs contain putative counterparts of the five motifs (I, III, IIIa, IV, and V) that define the covalent nucleotidyl transferase enzyme superfamily (Fig. 1). The members of this superfamily, consisting of DNA ligases, Rnl2-like RNA ligases, and mRNA capping enzymes, all catalyze NMP transfer to polynucleotide 5Ј ends via a covalent enzyme-(lysyl-N)-NMP intermediate (22).
DNA ligases and mRNA capping enzymes have a common tertiary structure composed of a proximal nucleotidyl transferase domain fused to a distal OB-fold domain (23)(24)(25)(26)(27). The NMP-binding pocket of the nucleotidyltransferase domain is composed of a cage of ␤-strands that includes motifs I, III, IIIa, IV, and V. Motif I, KX(D/H)G, contains the lysine nucleophile to which the NMP becomes covalently linked (28,29). Motifs III, IIIa, IV, and V contain conserved side chains that contact AMP or GMP and play essential roles in one or more steps of the ligation/capping pathways (30 -33). The distal OB domain consists of a five-stranded antiparallel ␤ barrel. The OB domain is essential for DNA ligase and RNA capping activities (33-36). T4 RNA ligase 2 is composed of a structurally homologous N-terminal nucleotidyltransferase domain fused to a distinctive essential C-terminal domain that has no recognizable similarity to the OB-folds of DNA ligases and capping enzymes (37,38).
The striking features of the putative Deinococcus RNA ligase polypeptide and its homologs are as follows. (i) They terminate immediately distal to nucleotidyltransferase motif V and therefore have no equivalent of the C-terminal domains that are essential for known ligases and capping enzymes. (ii) They include N-terminal domain modules that have no apparent primary structural similarity to segments of any known RNA ligases, DNA ligases, or mRNA capping enzymes; and (iii) their nucleotidyltransferase domains are structurally similar to the Rnl2-like RNA ligases (21,37,38), not the Rnl1-like enzymes. Thus, DraRnl and its homologs have a unique domain architecture that sets them apart from previously described covalent nucleotidyltransferases.
Purification and Adenylyltransferase Activity of DraRnl-We expressed DraRnl in E. coli as a His 10 -tagged fusion and purified the 40-kDa recombinant protein from a soluble bacterial extract by adsorption to nickel-agarose and step elution with imidazole (Fig. 2a). The native size of DraRnl was gauged by sedimentation through a 15-30% glycerol gradient. Marker proteins catalase (248 kDa), BSA (66 kDa), and cytochrome c (13 kDa) were included as internal standards. DraRnl sedimented as a single discrete peak between BSA and cytochrome c (Fig. 3a). An S value of 3.2 was determined for DraRnl by interpolation to the internal standard curve. We surmise that DraRnl is a monomer in solution.
An adenylyltransferase activity of recombinant DraRnl was evinced by label transfer from [␣-32 P]ATP to the DraRnl polypeptide to form a covalent enzyme-adenylate adduct (Fig.  2b). The adenylyltransferase activity profile during glycerol gradient sedimentation paralleled exactly the sedimentation profile of the DraRnl polypeptide (Fig. 3b).
The adenylyltransferase activity of DraRnl was characterized in greater detail using the nickel-agarose enzyme fraction. The extent of ligase-adenylate formation was proportional to input DraRnl (Fig. 4a). DraRnl adenylylation was optimal at pH 9 -9.5 and declined sharply at and below neutral pH (Fig.  4b). DraRnl required a divalent cation cofactor to form the covalent adduct. MgCl 2 and MnCl 2 supported approximately equivalent activity at the optimal concentrations of 10 -20 and 2.5-5 mM, respectively (Fig. 4c). Other divalent cations (calcium, cobalt, cadmium, copper, and zinc) were ineffective in supporting enzyme adenylylation at 5 mM concentration (data not shown). The yield of DraRnl-AMP complex reached saturation at Ն12 M ATP (Fig. 4d). Half-saturation was achieved at ϳ4 M ATP. We calculated that ϳ7-10% of the DraRnl protein preparation could be adenylated in vitro with 32 P-AMP. DraRnl was unreactive with [␣-32 P]GTP (data not shown).
RNA Ligase Activity of DraRnl-Rnl2 and protozoan RNAediting ligases are particularly adept at joining RNA termini splinted together by a bridging template strand (10,11,21). In the case of Rnl2, sealing activity is optimal when the reactive 3Ј-OH and 5Ј-PO 4 RNA ends are opposed at a nick in a doublestranded RNA or an RNA:DNA hybrid (21). Here, we prepared a nicked dsRNA:DNA substrate by annealing a 32 P-5Ј-endlabeled 24-mer RNA oligonucleotide to an unlabeled 24-mer DNA oligonucleotide with overlapping complementarity (Fig.  2c). The annealed strands form a 12-bp duplex with complementary 12-nucleotide 5Ј-tails that promote self-assembly into an RNA:DNA hybrid duplex containing staggered 3Ј-OH/5Ј-PO 4 RNA nicks on one strand (note that the DNA strand of the RNA:DNA hybrid contains staggered 3Ј-OH/5Ј-OH nicks, which cannot be sealed by polynucleotide ligases). DraRnl was reacted with the annealed 32 P-5Ј-end-labeled RNA:DNA hybrid (20 nM), and the products were analyzed by gel electrophoresis under denaturing conditions. DraRnl generated a ladder of sealed 32 P-labeled RNAs comprising a perfectly spaced series of n-mers, with n values ranging from 2 to Ն12 (Fig. 2c). The RNA sealing activity profile of DraRnl during zonal velocity sedimentation coincided with the adenylyltransferase activity profile and the abundance of the DraRnl polypeptide (Fig. 3c).
The extent of the ligation of the tandemly nicked dsRNA: DNA hybrid was proportional to input DraRnl (Fig. 2c). Equivalent concentrations of DraRnl were incapable of joining a tandemly nicked dsDNA:DNA substrate of identical sequence and complementarity (data not shown). However, a tandemly nicked dsRNA:RNA substrate of identical sequence and complementarity was sealed by DraRnl (data not shown). A control reaction confirmed that the nicked dsDNA substrate that was unreactive with DraRnl was readily sealed by T4 DNA ligase (not shown). DraRnl was unable to seal (either intermolecularly or intramolecularly) the 32 P-labeled 24-mer RNA single-strand by itself; a control experiment showed that the 32 Plabeled 24-mer ssRNA was circularized nearly quantitatively by baculovirus RNA ligase, an Rnl1-type enzyme (data not shown). These results indicate that Deinococcus RNA ligase specifically seals RNA strands at a duplex nick.
Sealing of the nicked dsRNA:DNA hybrid required a divalent cation cofactor. Activity was optimal in the presence of 5 to 10 mM manganese (Fig. 5a). Whereas cobalt (5 mM) was half as effective as manganese, magnesium and cadmium were only 5% as effective as manganese (Fig. 5b). Neither calcium, copper, nor zinc was able to support RNA-joining activity at a 5 mM concentration. RNA strand joining was optimal at pH 5.5-7 and declined at alkaline pH (Fig. 5c). The standard ligation reaction mixtures contained Յ10 mM NaCl contributed by the enzyme and RNA substrate solutions. The effect of increasing ionic strength on RNA ligation was gauged by supplementing the reactions with NaCl, KCl, or potassium glutamate. Activity was reduced by each of these monovalent salts in a concentrationdependent manner; 85-95% inhibition was seen at 100 mM added salts (Fig. 5d).
Enzyme titrations in the presence or absence of 0.2 mM ATP showed that RNA sealing activity was stimulated 2-fold by ATP (not shown). ATP-independent sealing activity is attributable to the presence of pre-adenylated DraRnl in the enzyme preparation. From the slope of the ATP-independent ligase titration curve, we estimated that one-third of the enzyme preparation was ligase-proficient DraRnl-AMP. The relatively modest ATP stimulation of sealing likely reflects the fact that the pH optimum values for the ligase adenylylation reaction and the overall RNA sealing reaction are significantly different.
Ligation of RNA Ends at a Nick versus a Gap or Flap-The DNA strand of the RNA:DNA hybrid was altered by adding one or two deoxyadenosine nucleotides in the center of the oligonucleotide such that, upon annealing to the 24-mer 32 P-labeled RNA strand, the 3Ј-OH RNA end was separated from the 5Ј-PO 4 end by a one-or two-nucleotide gap (21). The specific activities of DraRnl in sealing the one-and two-nucleotide gap substrates were 7.2% and 0.3%, respectively, of its activity at a nick (Fig. 6a). We then removed one nucleotide in the center of the DNA strand of the RNA:DNA hybrid such that, upon annealing to the 24-mer 32 P-labeled RNA strand, either the 3Ј-OH RNA end or the 5Ј-PO 4 end protruded from the bridging template strand as a one-nucleotide flap (21). DraRnl-specific activities on the 5Ј-flap and 3Ј-flap substrates were 1.4 and 0.7%, respectively, of the specific activity on the nicked substrate (Fig. 6b). We surmise that DraRnl sealing activity is optimal when the reactive termini are approximated at a nick.
RNA Specificity Is Dictated by the 3Ј-OH Strand-To understand the basis for the RNA specificity of DraRnl, we tested its activity with three-piece singly nicked duplex substrates in which two of the three components strands were DNA (Fig. 7). First, we annealed a 32 P-5Ј-end-labeled 24-mer DNA oligonucleotide and an unlabeled 18-mer 3Ј-OH RNA oligonucleotide to a 24-mer DNA template. DraRnl readily joined the 3Ј-OH RNA strand to the radiolabeled DNA strand to form a labeled 42-mer product. Indeed the yield of the RNA-to-DNA ligation product was higher than that of the RNA-to-RNA product (Fig. 7). Next, we annealed a 32 P-5Ј-end-labeled 24-mer RNA oligonucleotide and an unlabeled 18-mer 3Ј-OH DNA oligonucleotide to a 24mer DNA oligonucleotide template. DraRnl failed to join the 3Ј-OH DNA strand to the radiolabeled RNA strand at levels of input enzyme that were saturating for RNA-to-DNA ligation (Fig. 7). We came to the following conclusions. (i) DraRnl does not discriminate between RNA and DNA on the 5Ј-PO 4 side of the nick; and (ii) RNA specificity is dictated by a requirement for RNA on the 3Ј-OH side of the nick.
The RNA:DNA hybrid substrates used in the experiments shown in Fig. 7 have single strand tails protruding from a central 24-bp nicked duplex. We found that trimming the 5Ј single strand extension of the RNA strand on the 3Ј-OH side of the nick to form a blunt-ended 12-bp RNA:DNA hybrid on the 3Ј-OH side of the nick and trimming the 3Ј single-strand extension of the 5Ј-PO 4 DNA strand to form a blunt-ended 12-bp DNA duplex on the 5Ј-PO 4 side of the nick did not impede the RNA-to-DNA strand sealing reaction of DraRnl (Fig. 8a). However, an all-DNA 24-bp nicked duplex of identical sequence was unreactive with DraRnl (Fig. 8a). A control reaction showed that the all-DNA nicked duplex was readily sealed by a Chlorella virus DNA ligase (Fig. 8a).
Variants of the nicked 24-bp substrate were then prepared in which the 12-mer 3Ј-OH strand consisted of either nine deoxyribonucleotides and three ribonucleotides, 10 deoxyribonucleotides and two ribonucleotides, or 11 deoxyribonucleotides and one ribonucleotide (Fig. 8b). DraRnl joined the mixed DNA/RNA 3Ј-OH strand to the 5Ј-PO 4 DNA strand when only three ribonucleotides were present at the 3Ј-OH side of the nick  (Fig. 8b). The specific activity in sealing the three-ribonucleotide substrate was 38% of the specific activity with the substrate containing an all-ribonucleotide 3Ј-OH strand. Sealing activity declined further (to 6% of the all-ribonucleotide control) when the RNA content was diminished so that only two ribonucleotides were present at the 3Ј-OH side of the nick (Fig.  8b). DraRnl was unable to seal the nicked duplex when only a single 3Ј-OH ribonucleotide was present at the nick (Fig. 8b).
Adenylyltransferase and Ligase Activities Are Abolished by Mutation of the Motif I Lysine-The 164 EKLHGTQ 170 sequence of DraRnl, which embraces the predicted motif I lysine nucleo-phile, is most closely related to the 34 EKIHGTN 40 element of T4 Rnl2. The contributions of motif I Lys-165 to the activities of DraRnl were surmised from the effects of a single alanine substitution, K165A. As a control, we introduced an alanine change in lieu of Glu-278, a nonconserved residue within the predicted counterpart of nucleotidyltransferase motif IIIa (indicated by a plus sign (ϩ) in Fig. 1). Mutant proteins K165A and E278A were produced in bacteria and purified from soluble lysates by nickel-affinity chromatography (Fig. 2a). The K165A mutant was inert in enzyme-adenylate formation (Fig. 2b) and nick joining (Fig. 2d). These findings are consistent with Lys- 165 being the site of covalent adenylation. The E278A protein retained adenylyltransferase and RNA ligase activities (Fig. 2,  b and d).
The N-terminal Domain of DraRnl Promotes RNA Ligase Activity but Is Not Essential-The function of the distinctive N-terminal segment of the DraRnl protein was probed by analysis of incrementally truncated deletion mutants, namely DraRnl-(106 -342), DraRnl-(126 -342), and DraRnl-(136 -342). The N termini of the DraRnl⌬ mutants are indicated by arrows in Fig. 1. The recombinant His 10 -DraRnl⌬ proteins displayed the expected incremental increases in electrophoretic mobility (Fig. 2a). DraRnl-(106 -342) and DraRnl-(126 -342) reacted with [␣-32 P]ATP to form covalent protein-[ 32 P]adenylate adducts of the appropriate size, whereas the DraRnl-(136 -342) mutant was apparently inert (Fig. 2b). The DraRnl-(106 -342) and DraRnl-(126 -342) proteins retained RNA sealing activity (albeit less than full-length DraRnl), but DraRnl-(136 -342) was unable to seal RNA strands. We conclude the following. (i) The C-terminal nucleotidyltransferase containing as few as 39-aa proximal to the motif I lysine nucleophile is competent for RNA ligation; and (ii) the peptide segment from aa 126 to 136 is essential for adenylyltransferase activity, and, hence, RNA sealing. The upstream margin of the minimal catalytic domain of DraRnl relative to motif I is consistent with the location of the adenylylation site of T4 Rnl2 at Lys-35 (8). Although the 126-aa N-terminal portion of DraRnl is not strictly essential for catalysis, it does promote RNA sealing activity. Protein titration experiments showed that the specific activity of DraRnl-(126 -342) in sealing the nicked dsRNA:DNA hybrid substrate was 10% of the specific activity of full-length DraRnl (data not shown). The specific activity of DraRnl-(126 -342) in sealing the blunt-ended 24-bp nicked substrate with the all-RNA 3Ј-OH strand was 9% of the specific activity of fulllength DraRnl (data not shown).

DISCUSSION
Our demonstration that D. radiodurans encodes an RNA ligase with a distinctive domain architecture illuminates a potential for RNA repair reactions in bacteria. We speculate that RNA repair is advantageous for an organism like D. radiodurans that is exceptionally resistant to the effects of ionizing radiation. Although the repair of radiation damage to DNA has been studied intensively for decades, little is known about the biological impact of radiation damage to RNA. It is therefore noteworthy that expression of the DRB0094 gene encoding DraRnl has been reported to be transiently up-regulated in vivo during recovery from radiation exposure (39).
RNA repair had been documented previously in T4 phageinfected E. coli where two viral enzymes, RNA ligase 1 and polynucleotide kinase/phosphatase, function to evade an antiviral response predicated on the breakage of a specific tRNA by a host anticodon nuclease, PrrC (1). Homologs of PrrC are present in the proteomes of many bacterial species, albeit not in Deinococcus. tRNA restriction is not limited to antiviral responses; it also affords an RNA-based immune mechanism for individual bacteria to distinguish self from non-self in their niche. For example, the secreted bacterial toxins colicins D and E5 are anticodon nucleases that incise specific tRNAs in the anticodon loop to yield 2Ј,3Ј-cyclic phosphate and 5Ј-OH termini (40). After being taken up by susceptible bacteria, these colicins deplete the pool of tRNA, thereby killing the target bacterium via protein synthesis inhibition. Other bacterial toxins have been characterized that specifically incise mRNA (41)(42)(43) or rRNA (44,45). The atomic structure of colicin D is unlike that of any other structurally defined RNase, and it is suggested that colicin D, colicin E5, and PrrC-like proteins comprise distinct families of anticodon nucleases (46), in which case there might be a plethora of RNA-damaging enzymes and RNA-damage response pathways of which we currently have no clue. Colicin-producing bacteria are not altruistic; they protect themselves from the tRNA restriction enzyme by coproducing an antitoxin immunity protein that binds tightly to the anticodon nuclease and occludes its active site (46). Immunity from rRNA restriction is conferred by steric hindrance with the access of the nuclease to the ribosome (44). Endogenous RNA repair enzymes such as DraRnl provide a possible alternative mechanism to confer immunity to toxic RNases.
The nucleotidyl transferase domain of D. radiodurans RNA ligase is structurally similar to Rnl2-like RNA ligases, and its apparent specificity for template-directed RNA strand joining resembles that of the Rnl2-like enzymes (21). In particular, DraRnl (see above) and T4 Rnl2 (55) are able to ligate RNAto-RNA and RNA-to-DNA, but they are unable to join a 3Ј-OH DNA strand to a 5Ј-PO 4 RNA strand. DraRnl differs from other polynucleotide ligases in its reliance on manganese as a cofactor. Several studies suggest that manganese exerts unique physiological effects on Deinococcus growth, metabolism, and radiosensitivity (47,48) and that manganese is bound to the Deinococcus genome (49). Variations in extracellular manga- nese concentration affect genome condensation (49,50), and at least one other nucleic acid-modifying enzyme from Deinococcus, an endonuclease that incises UV damaged DNA, displays a unique requirement for manganese (51).
DraRnl has several remarkable structural features compared with other members of the ligase/capping enzyme superfamily. First, DraRnl is devoid of any counterpart of the Cterminal OB-fold domains found in DNA ligases and capping enzymes or of the distinctive C-terminal domain found in the Rnl2/REL class of RNA ligases. Second, DraRnl and its homologs have their own distinctive N-terminal domain. A PSI-BLAST search revealed that the N-terminal domain of DraRnl is similar in primary structure to the so-called B2 domain of the ␤-subunit of bacterial phenylalanyl-tRNA synthetases (Fig. 9). The B2 domain of Thermus thermophilus phenylalanyl-tRNA synthetase comprises a ␤-barrel with the topology of an OB-fold (52,53). Although the B2 domain does not interact directly with tRNA Phe , a structurally homologous OB-fold in aspartyl-tRNA synthetase binds the tRNA Asp anticodon loop and thereby contributes to substrate specificity (54). Although we show that the N-terminal OB-like domain is not strictly essential for adenylyltransferase or RNA sealing by DraRnl, its deletion results in significantly diminished strand joining activity with a synthetic model substrate as compared with the native protein. We hypothesize that the OB-domain either confers selectivity for certain RNA substrates, as yet undefined, that are relevant to DraRnl activity in vivo or else mediates interactions of DraRnl with other bacterial proteins involved in RNA repair.
Ho et al. (37) proposed that contemporary members of the covalent nucleotidyl transferase superfamily evolved by the fusion of ancillary domains to an Rnl2-like ancestral catalytic domain module involved in RNA ligation. The domain architecture of DraRnl supports this hypothesis by showing that the domain order is clearly not fixed and that different domains and even different flavors of a ubiquitous domain (e.g. the tRNA synthetase-like OB-fold of DraRnl versus the DNA ligase/capping enzyme-like OB-fold) can be appended to either end of the nucleotidyl transferase domain.