Novel 3'-ribonuclease and 3'-phosphatase activities of the bacterial non-homologous end-joining protein, DNA ligase D.

Pseudomonas aeruginosa DNA ligase D (PaeLigD) exemplifies a family of bacterial DNA end-joining proteins that consist of a ligase domain fused to a polymerase domain and a putative nuclease module. The LigD polymerase preferentially adds single ribonucleotides at blunt DNA ends and, as we show here, is also capable of adding up to 4 ribonucleotides to a DNA primer-template. We report that PaeLigD has an intrinsic ability to resect the short tract of 3'-ribonucleotides of a primer-template substrate to the point at which the primer strand has a single 3'-ribonucleotide remaining. The failure to digest beyond this point reflects a requirement for a 2'-OH group on the penultimate nucleoside of the primer strand. Replacing the 2'-OH by a 2'-F, 2'-NH2, 2'-OCH3, or 2'-H abolishes the resection reaction. The ribonucleotide resection activity resides within a 187-amino acid N-terminal nuclease domain and is the result of at least two component steps: (i) the 3'-terminal nucleoside is first removed to yield a primer strand with a ribonucleoside 3'-PO4 terminus, and (ii) the 3'-PO4 is hydrolyzed to a 3'-OH. The 3'-ribonuclease and 3'-phosphatase activities are both dependent on a divalent cation, specifically manganese. PaeLigD preferentially remodels the 3'-ends of a duplex primer-template substrate rather than a single strand of identical composition, and it prefers DNA primer strands containing a short 3'-ribonucleotide tract to an all-RNA primer. The nuclease domain of PaeLigD and its bacterial homologs has no apparent structural or mechanistic similarity to previously characterized nucleases. Thus, we surmise that it exemplifies a novel phosphoesterase family, defined in part by conserved residues Asp-50, Arg-52, and His-84, which are essential for the 3'-ribonuclease and 3'-phosphatase reactions.

DNA double-strand breaks can be repaired either via homologous recombination or non-homologous end joining (NHEJ) 1 (1). Homologous recombination and NHEJ pathways coexist in eukaryotic cells. Eukaryal NHEJ requires a DNA end-binding protein (Ku) and a specialized ATP-dependent DNA ligase (LigIV) (2,3). Until recently, it had been thought that bacteria rely exclusively on homologous recombination to repair doublestrand breaks. Although this is the case for Escherichia coli, the detection of homologs of Ku in a subset of bacterial proteomes raised interest in a putative bacterial NHEJ pathway (4 -6). There is now direct evidence that Mycobacteria have a vigorous NHEJ system that requires Ku and a specialized polyfunctional ATP-dependent DNA ligase (LigD) (7,8). Mycobacterial NHEJ is highly mutagenic (ϳ50% error rate), even when repairing complementary 5Ј-overhang ends. An analysis of the recombination junctions from individual NHEJ events revealed the participation of several DNA end-remodeling activities, including template-dependent fill-in of 5Ј-overhangs, non-templated addition of single nucleotides at blunt ends, and nucleolytic resection (8). The findings that a LigD deletion suppresses the overall frequency of NHEJ and increases the fidelity of the LigD-independent residual repair events are consistent with LigD serving as a direct catalyst of error-prone repair in vivo (8).
Bacterial LigD proteins are multifunctional enzymes composed of an ATP-dependent DNA ligase domain fused to a polymerase domain and a putative nuclease domain (6 -10). The domain order varies among bacterial LigD proteins, e.g. Mycobacterium tuberculosis (Mtu) and Mycobacterium smegmatis LigD consist of an N-terminal polymerase domain, a central putative nuclease domain, and a C-terminal ligase domain, whereas Pseudomonas aeruginosa (Pae) LigD is composed of an N-terminal putative nuclease domain, a central ligase domain, and a C-terminal polymerase domain (Fig. 1). The C-terminal segment of MtuLigD (amino acids 412-759) and the central segment of PaeLigD (amino acids 188 -527) are autonomous ligase domains capable of DNA nick sealing and ATP-dependent autoadenylylation in vitro (7,9). The ligase domain includes the six defining nucleotidyl transferase motifs of the ATP-dependent DNA ligase family (7,11).
The polymerase function of MtuLigD is localized to an autonomous N-terminal domain LigD-(1-299), whereas the Pae-LigD polymerase activity resides within an autonomous Cterminal segment, LigD-(533-840) (8 -10). The Pol domain of LigD has primary structure and functional similarities to members of the Pol X family of nucleic acid polymerases (12)(13)(14). The polymerase activities of Pae-and MtuLigD are manifest either as non-templated nucleotide addition to a blunt-ended duplex DNA primer or templated extension of a 5Ј-tailed duplex DNA primer-template (8 -10). Non-templated blunt-end addition in vitro is limited to the incorporation of only 1 or 2 nucleotides at the primer terminus. During templated synthesis, the primer is elongated to the end of the template strand and is then further extended with a single non-templated nucleotide. Fill-in and addition of a single non-templated nucleotide are the molecular signatures of mycobacterial NHEJ in vivo at 5Ј-overhang double-strand breaks (DSB) and blunt-end DSB, respectively (8). It is notable that rNTPs are preferred over dNTPs as substrates for non-templated blunt-end addition. This property led to speculation that the initial insertions preceding the strand-sealing step of NHEJ of a blunt DSB might involve rNMP incorporation and that the ability of LigD to use rNTPs as substrates might be advantageous for the repair of chromosomal double-strand breaks that arise in quiescent cells, insofar as the dNTP pool might be limiting when bacteria are not actively replicating.
There is less known at present about the nuclease function of bacterial LigD. Fig. 1 shows an alignment of the amino acid sequences of the putative nuclease domains of the LigD proteins from eight species of bacteria (P. aeruginosa, Agrobacterium tumefaciens, Bordetella bronchiseptica, M. tuberculosis, M. avium, Bradyrhizobium japonicum, Mezorhizobium loti, and Nocardia farcinica); the native LigD proteins range in size from 759 to 892 amino acids. Also included in the alignment are two archaeal homologs from Methanosarcina mazei and Methanosarcina acetivorans, which are single-domain proteins of 152 and 156 amino acids, respectively. The side chains conserved in all of the aligned proteins are denoted by dots above the sequence in Fig. 1. Initial informatics analysis focused on the presence of clusters of conserved histidines and aspartates that were predicted to form a metal-coordinating cluster composed of ␤-strands. On this basis, a deoxynuclease function was suggested (4). Della et al. (10) have reported that MtuLigD has an associated metal-dependent DNA 3Ј-exonuclease activity that digests either single-stranded DNA, a 3Ј-tailed duplex DNA, or a DNA flap structure with a 3Ј-tail.
Here we describe a nuclease activity intrinsic to Pseudomonas LigD. We discovered the nuclease activity while analyzing the capacity of PaeLigD to incorporate ribonucleotides at the 3Ј-end of a DNA primer-template. We show that PaeLigD and its isolated N-terminal nuclease domain catalyze the removal of 3Ј-ribonucleotides from a 5Ј-DNA-RNA-3Ј primer-template substrate via a novel metal-dependent mechanism entailing hydrolysis of the terminal DNA-(rN)p(rN) phosphodiester to yield a DNA-(rN)p species, which is subsequently converted to DNA-(rN) OH product. Consistent with the proposed reaction scheme, the PaeLigD nuclease domain has an intrinsic metaldependent 3Ј-phosphatase activity on a DNA-(rN)p primertemplate substrate. We discuss the potential role of the LigD 3Ј-ribonuclease and 3Ј-phosphatase activities in healing damaged ends via ribonucleotide incorporation at NHEJ junctions.

EXPERIMENTAL PROCEDURES
Nuclease Domain of PaeLigD-A gene fragment encoding LigD-(1-187) (Nuc domain) was amplified by PCR with a sense strand primers that introduced an NdeI restriction site at the beginning of the open reading frame and an antisense primer that introduced a new stop codon and a flanking BamHI site. The PCR product was digested with NdeI and BamHI and inserted into pET16b to generate an expression plasmid encoding a His 10 -tagged version of the putative Nuc domain of PaeLigD. Alanine mutations were introduced into the pET-PaeLigD-(1-187) plasmid by the two-stage overlap extension PCR method. The inserts of the wild-type and Ala-mutant pET-PaeLigD-(1-187) plasmids were sequenced completely to exclude the acquisition of unwanted changes during amplification and cloning. The pET-PaeLigD-(1-187) plasmids were transformed into E. coli BL21(DE3). Induction of protein expression with isopropyl 1-thio-␤-D-galactopyranoside, preparation of soluble bacterial lysates, and initial purification of the recombinant wild-type LigD-(1-187) protein by Ni-agarose affinity chromatography were performed as described previously (9) for full-length PaeLigD and the LigD Pol domain. The affinity column was eluted stepwise with 50, 100, 200, and 500 mM imidazole in buffer A (50 mM Tris-HCl, pH 7.5, 400 mM NaCl, 10% glycerol). The polypeptide compositions of the fractions were monitored by SDS-PAGE. LigD-(1-187) was recovered predominantly in the 200 and 500 mM imidazole eluate fractions, which were pooled and diluted with an equal volume of buffer B (50 mM Tris-HCl, pH 7.5, 1 mM DTT, 10% glycerol). A protein-containing precipitate that formed upon dilution was removed by centrifugation. The clarified supernatant was applied to an 8-ml Source 15S (Amersham Biosciences) cation exchange column that had been equilibrated with buffer B. The adsorbed LigD-(1-187) protein was eluted with a 250-ml linear gradient of 0 -1.0 M NaCl in buffer B. The peak fractions were pooled, concentrated by centrifugal ultrafiltration, and stored at Ϫ80°C. The yield at the Source S step was ϳ6 mg of LigD-(1-187) protein from a 1-liter bacterial culture. The LigD-(1-187)-Ala mutants were purified through the Ni-agarose step. Protein concentrations were determined by using the Bio-Rad dye reagent with bovine serum albumin as the standard.
Glycerol Gradient Sedimentation-An aliquot (40 g) of the Ni-agarose preparation of the PaeLigD Nuc domain was mixed with catalase (30 g), bovine serum albumin (30 g), and cytochrome c (30 g). The mixture was applied to a 4.8-ml 15-30% glycerol gradient containing 50 mM Tris-HCl, pH 8.0, 0.2 M NaCl, 1 mM EDTA, 2.5 mM DTT, 0.1% Triton X-100. An aliquot (50 g) of the phosphocellulose preparation of the full-length PaeLigD (9) was applied to a separate glycerol gradient without internal markers. The gradients were centrifuged for 18 h at 4°C in a Beckman SW50 rotor at 50,000 rpm. Fractions (ϳ0.2 ml) were collected from the bottoms of the tubes. The polypeptide compositions of the gradient fractions were analyzed by SDS-PAGE. Aliquots of the fractions were assayed for activity as specified in the figure legends.
Primed RNA Polymerase Assay-Activity was gauged by NTP-dependent extension of a 12-mer 5Ј 32 P-labeled DNA, RNA, or mixed DNA-RNA primer. Oligoribonucleotides were purchased from Dharmacon (Lafayette, CO) and deprotected as instructed by the vendor. Oligodeoxyribonucleotides were purchased from Biosource International (Camarillo, CA). The 12-mer primer strands were end-labeled by reaction with T4 polynucleotide kinase and [␥ 32 P]ATP, purified by native gel electrophoresis, and annealed to a 4-fold excess of an unlabeled complementary 24-mer DNA strand to generate the 5Ј-tailed substrates depicted in Fig 48% formamide. The products were resolved by electrophoresis through a 15-cm 18% polyacrylamide gel containing 7 M urea in TBE (90 mM Tris borate, 2.5 mM EDTA). The products were visualized by autoradiography. The fraction of input primer extended was determined by scanning the gel with a Fujifilm BAS-2500 imaging apparatus.
Nuclease Assay-Reaction mixtures (10 l) containing 50 mM Tris-HCl, pH 7.5, 5 mM DTT, 0.5 mM MnCl 2 , 0.5 pmol 5Ј 32 P-labeled D10R2 or other primer-template as specified and enzyme were incubated at 37°C for 20 min. The reactions were quenched by adjusting the mixtures to 7 mM EDTA and 31% formamide. The products were resolved by electrophoresis through a 40-cm 18% polyacrylamide gel containing 7 M urea in TBE. The products were visualized by autoradiography and quantified by scanning the gel with a Fujifilm BAS-2500 imaging apparatus.
3Ј-Phosphatase Assay-The 5Ј 32 P-labeled, 3Ј-phosphate-terminated D10R1-p and D9R1-p strands were prepared by digesting the 5Ј 32 P-labeled D10R2 and D9R3 primer oligonucleotides with ribonuclease A. Reaction mixtures (100 l) containing 50 pmol radiolabeled D10R2 or D9R3 oligonucleotide, 5 mM DTT, and 20 g RNase A were incubated for 20 min at 37°C. The mixtures were extracted with phenol-chloroform. The oligonucleotides were precipitated with ethanol in the presence of glycogen carrier and then annealed to a 4-fold excess of an unlabeled complementary 24-mer DNA strand to generate the 5Ј-tailed substrates depicted in Fig. 4C. The 3Ј-phosphatase reaction mixtures (10 l) containing 50 mM Tris-HCl, pH 7.5, 5 mM DTT, 0.5 mM MnCl 2 , 0.25 pmol 5Ј 32 P-labeled D10R1-p or D9R1-p primer-template and enzyme as specified were incubated at 37°C for 20 min. The reactions were quenched by adjusting the mixtures to 7 mM EDTA and 31% formamide. The products were resolved by electrophoresis through a 40-cm 18% polyacrylamide gel containing 7 M urea in TBE. The products were visualized by autoradiography. The extent of conversion of the 3Ј-PO 4 oligonucleotide to the slower migrating 3Ј-OH product was determined by scanning the gel with a Fujifilm BAS-2500 imaging apparatus.

Primer-template-directed RNA Polymerization by the Pol Domain of PaeLigD-Previous
studies showed that rNTPs are preferred over dNTPs as substrates for non-templated bluntend addition by PaeLigD (9). Here we queried whether the PaeLigD polymerase could incorporate ribonucleotides during a templated fill-in synthesis reaction. The DNA primer-template used was composed of a 5Ј 32 P-labeled 12-mer DNA strand annealed to a complementary 24-mer strand to form the 5Ј-tailed molecule shown in Fig. 2A. Control experiments showed that the isolated PaeLigD Pol domain catalyzed dNTPdependent extension of the primer strand until the overhang was filled in completely, after which an extra non-templated dNMP was added to the blunt 3Ј-end (Ref. 9 and data not shown). When dNTPs were replaced by rNTPs, the efficiency of primer utilization (defined as the percent of the primer strands elongated by at least one nucleotide) remained high, but the number of rNMPs added was reduced. At enzyme concentrations sufficient to elongate 85-95% of the DNA substrate, rNMP incorporation was limited to 1-3 nucleotides. Further increases in the enzyme concentration resulted in the accumulation of a predominant product with four added rNMPs. Only a very low fraction of the primers was extended beyond the nϩ4 template position. This result suggested that the LigD polymerase might be sensitive to the pentose sugar composition of the primer strand.
To address this point, we used a substrate consisting of a 5Ј 32 P-labeled 12-mer RNA hybridized to the 24-mer DNA template strand (Fig. 2E). The efficiency of RNA primer utilization for RNA synthesis was reduced compared with the DNA primer, e.g. only 12-14% of the ends were extended by levels of polymerase that sufficed to extend 95% of the DNA primer ( Fig.  2E versus A). Raising the enzyme concentration resulted in up to 49% primer utilization, but only 1 rNMP was added to the RNA primer at all levels of input enzyme tested. This finding shows that PaeLigD polymerase strongly prefers a DNA primer terminus versus an RNA primer.
The basis for the RNA interference effect was revealed by testing a series of 12-mer primers consisting of 11, 10, or 9 deoxynucleotides at their 5Ј-ends plus 1, 2, or 3 ribonucleotides at their 3Ј-ends. We observed that substitution of even a single ribonucleotide at the 3Ј-OH terminus (in the D11R1 substrate shown in Fig. 2B) reduced the efficiency of primer utilization compared with the all-DNA D12 primer. Moreover, most D11R1 primers that were extended at the lower enzyme concentration were elongated by 1 nucleotide, and the product distribution at the highest enzyme concentrations included a mixture of 13-, 14-, 15-, and 16-mer species, but nothing larger (Fig. 2B). When the primer contained two terminal ribonucleo- The products were resolved by denaturing PAGE and visualized by autoradiography. The extents of the end-addition reactions, expressed as the percent of the input 12-mer primer that was elongated, are indicated in italics below the lanes. tides (D10R2 in Fig. 2C), the primer utilization efficiency declined again, and the major product was the nϩ1 species, even at high enzyme concentrations. Increasing the RNA tract to 3 nucleotides suppressed primer utilization even further, to the point that the D9R3 substrate phenocopied the all-RNA R12 primer-template construct (Fig. 2D). Together, these data explain the restricted addition of only 4 nucleotides at a DNA primer terminus, as follows: (i) productive use by the PaeLigD Pol domain of the primer 3Ј-OH terminus depends on contacts with deoxynucleotides at the three terminal positions of the primer strand; (ii) each serial rNMP incorporation event diminishes the ability of the ribo-extended primer to undergo the next round of reaction with an rNTP; and (iii) after 3 ribonucleotides are added, activity is suppressed strongly and effectively limited to only one more cycle of extension under the conditions analyzed. Thus, the impediment to the RNA polymerase activity of LigD is imposed by the primer, not the rNTP substrate. Consistent with this model, we found that RNA tracts at the 3Ј-primer terminus also suppressed the efficiency of deoxynucleotide incorporation by the LigD Pol domain (data not shown). The implication is that LigD utilization of ribonucleotides for fill-in synthesis during NHEJ is plausible only for short repair tracts. This could include the fill-in events observed in vivo in M. smegmatis at double-strand breaks containing 4-nucleotide 5Ј-overhangs (8).
Removal of a 3Ј-Ribonucleotide by PaeLigD-An initial experiment showed that reaction of full-length PaeLigD with the D10R2 primer-template construct in the absence of nucleotides resulted in the shortening of the 5Ј-radiolabeled primer strand to yield a predominant product that migrated as an 11-mer (see below). No shorter products were detected when PaeLigD was reacted in parallel with the D12 or D11R1 primer-templates, suggesting that the reaction required a diribonucleotide 3Ј terminus. To test whether this putative 3Ј-ribonuclease was intrinsic to PaeLigD, we tracked the activity after sedimenting PaeLigD in a glycerol gradient. We showed previously that ligase and polymerase activities cosedimented with the Pae-LigD protein, which was judged to be a monomer compared with internal standards (9). When sedimented by itself, the 97-kDa LigD polypeptide comprised a single discrete peak centered at gradient fractions 13-15 (Fig. 3A). The products of the reaction of the glycerol gradient fractions with the D10R2 primer template are shown in Fig. 3B. Fractions 13-15 contained the peak of activity that removed the 3Ј-terminal ribonucleotide. The peak fractions converted the input 12-mer D10R2 strand to a major species corresponding to an 11-mer D10R1, whereas the flanking fractions 11 and 17 generated a second more rapidly migrating species, which we will identify below as the terminal-phosphorylated 11-mer, D10R1-p. Based on cosedimentation, we would attribute the 3Ј-ribonuclease activity to the recombinant PaeLigD protein.
An Autonomous 3Ј-Ribonuclease Domain of PaeLigD-We produced an N-terminal fragment, the LigD polypeptide spanning amino acids 1-187 (the putative nuclease domain) in E. coli, as a His 10 fusion and purified the recombinant Nuc domain from a soluble bacterial lysate by Ni-agarose and cation exchange chromatography (Fig. 4A). We also produced in parallel a mutated version of the nuclease domain in which the invariant His-84 side chain was replaced by alanine. The H84A protein was purified by Ni-agarose chromatography. The wild-type and mutant LigD-(1-187) preparations were nearly homogeneous with respect to the ϳ29 kDa Pae polypeptide (Fig. 4A). The wild-type Nuc domain displayed the same 3Ј-processing activity on the D10R2 primer-template that was noted above for full-length PaeLigD (Fig. 4B). Moreover, the 3Ј-processing function was abolished by the H84A mutation, arguing that the observed activity was intrinsic to the Nuc domain (Fig. 4B).
The quaternary structure of the Nuc domain was examined by zonal velocity sedimentation in a 15-30% glycerol gradient (Fig. 5). Marker protein catalase (native size 248 kDa), bovine serum albumin (66 kDa), and cytochrome c (12 kDa) were included as internal standards. After centrifugation, the polypeptide compositions of the odd-numbered gradient fractions were analyzed by SDS-PAGE. The Nuc domain sedimented as a discrete peak (fractions 19 -23) between bovine serum albumin and cytochrome c (Fig. 5A), as did the activity responsible for 3Ј shortening of the D10R2 substrate (Fig. 5B). These results are consistent with a monomeric structure for the isolated Nuc domain.
Sequential Mechanism of 3Ј-Ribonucleotide Removal-A kinetic analysis of the reaction of the Nuc domain with the D10R2 primer-template is shown in Fig. 6A (and, in a separate experiment, in Fig. 8B). The 5Ј-labeled primer was rapidly converted to a fast migrating species (with apparent size of 10 nt) that comprised 52% of the total labeled material after 1-2 min and decayed steadily at 5, 10, 20, and 30 min. A second product, migrating as an 11-mer, comprised 15% of the total label after 1 min and increased steadily in abundance thereafter, concomitant with the decay of the "10-mer-sized" species. The kinetic data (plotted in Fig. 8C) support a precursor product relationship between the more rapidly migrating species and the 11-mer end product. Because chain length obviously cannot increase with time in the absence of added nucleotides, we suspected that the initial faster migrating intermediate was an 11-mer D10R1-p molecule containing a phosphate at the processed 3Ј-end and that this intermediate was subsequently converted to an 11-mer D10R1 strand with a 3Ј-OH terminus. It is well established that a 3Ј-PO 4 -terminated oligonucleotide migrates faster during denaturing PAGE than a 3Ј-OH polynucleotide of identical length and sequence. The kinetic data in Fig. 8C were fit using the CKS kinetic simulation program (version 1.0; IBM Corp.) to a unidirectional two-step reaction mechanism with rate constants of 0.02 and 0.0055 s Ϫ1 for the ribonuclease and phosphatase steps, respectively.
If the proposed sequential mechanism is valid, then the PaeLigD Nuc domain should have a 3Ј-phosphatase activity  Reaction mixtures (80 l) containing 50 mM Tris-HCl, pH 7.5, 5 mM DTT, 0.5 mM MnCl 2 , 9.6 g of (400 pmol) Nuc domain, and either 4 pmol 32 P-labeled D10R2 primer-template (A), 4 pmol 32 Plabeled D9R3 primer-template (B), or 2 pmol 32 P-labeled D10R1-p primer-template (C) were incubated at 37°C. Aliquots (10 l) were withdrawn at the times indicated and quenched immediately with EDTA/ formamide. The time 0 sample was withdrawn prior to adding Nuc domain. The products were resolved by PAGE and visualized by autoradiography. uncoupled from prior resection of the 3Ј-terminal ribonucleoside. To measure this activity, we generated a 5Ј-labeled D10R1-p strand by digesting D10R2 with RNase A. D10R1-p was then annealed to the complementary 24-mer DNA template strand to form the tailed duplex substrate shown in Fig.  4C. The D10R1-p strand (which comigrates with the initial intermediate generated during the D10R2 3Ј-processing reaction) was converted quantitatively by the wild-type Nuc domain to a more slowly migrating 11-mer species corresponding to D10R1-OH (Fig. 4B). Note that the product of the 3Ј-phosphatase reaction comigrated with the 11-mer end product of the D10R2 processing reaction analyzed in the same gel (Fig.  4B). The conversion of D10R1-p to D10R1-OH was abolished by the H84A mutation of the Nuc domain (Fig. 4B), which signifies that the 3Ј-phosphatase activity is intrinsic to LigD. This conclusion is supported by the findings that the 3Ј-phosphatase activity cosedimented with full-length LigD (Fig. 3C) and with the isolated Nuc domain (data not shown).
Further insights to the mechanism of 3Ј processing emerged from analysis of the reaction of the wild-type Nuc domain with the D9R3 primer-template (Fig. 4C), which was converted to two products that migrated faster (by 1-nucleotide steps) than the D10R2 reaction products analyzed in parallel (Fig. 4B). We surmise that the major end product is D9R1-OH and the minor species is a D9R1-p reaction intermediate. The H84A mutation abolished formation of both products derived from the D9R3 substrate (Fig. 4B). To verify this reaction scheme, we generated a 5Ј-labeled D9R1-p strand by RNase A digestion of D9R3 and annealed the D9R1-p strand to the 24-mer DNA to form the substrate shown in Fig. 4C. The D9R1-p strand (which comigrates with the faster species generated during the D9R3 end resection reaction) was converted by the wild-type Nuc domain to a more slowly migrating species, D9R1-OH , which comigrated with the major product of the D9R3 reaction (Fig.  4B). Again, the conversion of D9R1-p to D9R1-OH was eliminated by the H84A mutation (Fig. 4B).
A kinetic analysis of the 3Ј-phosphatase reaction with the D10R1-p substrate is shown in Fig. 6C. The dephosphorylated product accumulated steadily with pseudo-first order kinetics and an apparent rate constant of 0.0046 s Ϫ1 . This value agrees with the rate of 0.0055 s Ϫ1 estimated for the phosphatase step of the 3Ј-processing reaction with D10R2 substrate.
The kinetic analysis of the processing of the D9R3 substrate shown in Fig. 6B is complicated by the fact that one of the reaction intermediates, D9R2-p, comigrates with the end-product, D9R1-OH . The results are compatible with the scheme depicted in Fig. 6B whereby the Nuc domain incises the phosphodiester of the terminal (rC)p(rC) dinucleotide to give the 11-mer D9R2-p, which is then converted to D9R1-p, which is in turn dephosphorylated to form the D9R1-OH end product. A longer exposure of the gel revealed only trace amounts of a D9R2-OH species at 5 and 10 min, which disappeared at later times (not shown).
Divalent Cofactor Requirement-The 3Ј-ribonucleotide resection activity of the PaeLigD Nuc domain was optimal at pH 5.5-8.0 in 50 mM Tris acetate or Tris-HCl buffer (data not shown) and was strictly dependent on a divalent cation, specifically manganese (Fig. 4D). A manganese titration experiment showed that the 3Ј-ribonuclease activity was optimal at 0.2-0.8 mM MnCl 2 (not shown). Magnesium, zinc, and calcium were ineffective at 0.5 mM concentration (Fig. 4D). Higher concentrations of magnesium (1, 2.5, 5, and 10 mM) were also ineffective (not shown). Cadmium, cobalt, and copper (0.5 mM) were capable of sustaining reduced activity that resulted in the formation of the initial D10R1-p product, little of which was converted to the D10R1-OH end product (Fig. 4D). The 3Ј-phosphatase activity also required a divalent cation, again specifically manganese (Fig. 4D). Magnesium, zinc, and cadmium were inactive, whereas calcium, cobalt, and copper supported low levels of dephosphorylation (Fig. 4D).
Requirement for the Template DNA Strand-The dependence of the 3Ј-ribonuclease activity on the DNA template strand was gauged by comparing product formation as a function of input Nuc domain for the D10R2 primer-template substrate versus the D10R2 single-stranded oligonucleotide (Fig. 7). The extent of product formation on the primer-template substrate increased with Nuc domain concentration. At enzyme levels sufficient to consume 20 -70% of the input-labeled strand, the products consisted of a mixture of D10R1-p and D10R1-OH strands (Fig. 7, A and D). At the highest level of enzyme, nearly all the substrate was converted to D10R1-OH . Activity on the single-stranded D10R2 substrate was clearly diminished, and there was a strong skew in the product distribution toward accumulation of the D10R1-p intermediate, with little conversion to the D10R1-OH end product (Fig. 7, B and D). The specific activity of the initial ribonucleoside excision reaction, gauged by the consumption of the D10R2 strand as function of enzyme in the linear range, was about 5-fold higher for the primertemplate than for the primer strand alone. However, the specific activity in generating dephosphorylated D10R1 was about 35-fold higher for the primer-template than for the primer alone. These results showed that the PaeLigD nuclease displays a strong bias for a 3Ј terminus within a duplex nucleic acid compared with a single-stranded oligonucleotide and hinted that the 3Ј-phosphatase might be especially sensitive to the secondary structure of the substrate.
The latter issue was addressed directly by comparing the 3Ј-phosphatase activity of the Nuc domain on a D10R1-p primer-template versus a D10R1-p single strand (Fig. 7C). The 3Ј-phosphatase-specific activity on the primer-template substrate was ϳ200-fold greater than on the primer alone. Thus, the 3Ј-phosphatase reaction displays a fairly stringent requirement for the template DNA strand.
Requirement for a 2Ј-OH at the Incised Phosphodiester-At levels of input Nuc domain that sufficed to convert all of the input D10R2 primer-template to the D10R1 product, we observed virtually no shortening of either the all-DNA D12 primer-template or the D11R1 primer-template containing only a single 3Ј-ribonucleotide (Fig. 8A). These results indicate that a ribose sugar is required at the penultimate nucleotide position. The role of the 2Ј-OH at the penultimate position was probed using a series of chemically modified primer-templates in which the 12-mer 5Ј-PO 4 strand consisted of 10 deoxynucleotides at the 5Ј-end, 1 ribonucleotide at the 3Ј terminus, and either a 2Ј-O-methyl, a 2Ј-amine, or a 2Ј-fluorine nucleotide at the penultimate position. The instructive finding was that all of the 2Ј modifications of the penultimate nucleoside abolished the 3Ј-ribonuclease activity of PaeLigD (Fig. 8A). These results argue for a direct catalytic role of the 2Ј-OH at the incised phosphodiester, which could entail (i) its action as a nucleophile to attack the phosphorus center to yield a 2Ј,3Ј-cyclic phosphate intermediate, which is subsequently hydrolyzed to a 3Ј-phosphate; (ii) coordination of the divalent cation cofactor; or (iii) assistance in activation of a nucleophilic water in its attack on the phosphorus center. The prospect that the 2Ј-OH could act only indirectly by affecting the sugar pucker of the penultimate nucleoside appears unlikely, given that a 2Ј-F is effective in promoting an RNA-like C3Ј-endo conformation, even more so than is a 2Ј-OCH 3 (15).
The findings that LigD efficiently removes one or two ribonucleotides from the 3Ј-end of a DNA primer-template (Figs. 4A and 8A) raises the question of whether and how LigD would process an all-RNA strand in a primer-template substrate. We observed that the Nuc domain was less active in shortening the R12 primer-template than it was in resecting either the D10R2 or D9R3 substrates, as gauged by the extent of decay of the radiolabeled input strand (Fig. 8A). Moreover, whereas the enzyme resected the D10R2 and D9R3 substrates to the last remaining 3Ј-ribonucleotide, the products of the reaction with the R12 substrate were limited to only two major species, migrating at positions of an 11-and 10-mer, respectively (Fig.  8A). (Note that the all-RNA 12-mer strand migrates more slowly than the DNA primer strands of the same length and base composition.) A kinetic analysis of the processing of the R12 primer-template in shown in Fig. 8, B and C, with a parallel analysis of the reaction with D10R2 serving as the positive control. The "10-mer" RNA species was the first to appear and accumulated over 20 min to an extent of about 27% of the total labeled material. The longer "11-mer" species accumulated after an initial lag phase. We surmise that the initial product is a 3Ј-phosphorylated R11-p strand, which is subsequently dephosphorylated to a more slowly migrating R11-OH product. Only trace amounts of product shorter than the 10mer were formed at the 20 and 30 min times (comprising 3-6% of the total label). Two relevant points emerge from this analysis: (i) the sequential reaction mechanism at the terminal (rN)p(rN) dinucleotide applies whether or not the 5Ј-segment of the primer strand is DNA or RNA, and (ii) the initial increment of resection is a mononucleoside, even when there are multiple internal (rN)p(rN) linkages available. The product distributions of the D10R2 and R12 reactions are plotted in Fig. 8C. A kinetic simulation of the R12 data yielded estimated rate constants of 0.00055 and 0.0008 s Ϫ1 for the first ribonuclease and phosphatase steps, respectively. Comparison to the simulation of the D10R2 kinetic data indicates that ribonuclease and phosphatase reactions are ϳ35and 7-fold faster on the D10R2 primer than on the all-RNA strand.
Mutational Inactivation of the 3Ј-Ribonuclease and 3Ј-Phosphatase-The nuclease domain of PaeLigD displays primary structure similarity to other bacterial LigD proteins, but it does not score against any biochemically established nuclease families when subjected to a PSI-BLAST search at NCBI. Thus, we surmise that it exemplifies a novel phosphoesterase family. We showed above that activity is strictly dependent on His-84, which is conserved in all of the LigD Nuc-like polypeptides aligned in Fig. 1. To identify other candidate components of the active site, we extended the alanine scan to 4 other invariant amino acids, Asp-50, Arg-52, Glu-54, and Glu-82 (highlighted in shaded boxes in Fig. 1). The mutant Nuc domains were produced in E. coli and purified from soluble bacterial extracts by Ni-agarose chromatography. SDS-PAGE analysis revealed comparable purity for the wild-type and mutant proteins (Fig.  9A). The 3Ј-ribonuclease activities were assayed in parallel using the D10R2 primer-template substrate at a level of input enzyme (1.2 g) that was saturating for the wild-type Nuc domain; this allowed for ready detection of the most severe catalytic defects. We found that the D50A and R52A proteins (as well as the H84A mutant) were virtually inert in generating either the D10R1-p or D10R1-OH reaction products (Fig. 9B). On the other hand, the E54A protein generated the D10R1-OH end product to a similar extent as the wild-type Nuc domain (Fig. 4B). Whereas the E82A mutant was active in 3Ј-ribonucleoside resection, as gauged by consumption of the input strand, the product distribution was strongly skewed toward accumulation of the D10R1-p species (Fig. 9B), suggesting that the mutation selectively affected the 3Ј-phosphatase activity. Direct measurement of 3Ј-phosphate removal from the D9R1-p primer-template at a level of input enzyme saturating for wild-type Nuc showed that activity was virtually abolished by the D50A, R52A, E82A, and H84A mutations, whereas the E54A change was benign (Fig. 8C). These results suggest that the active sites for the ribonuclease and phosphatase functions are likely to overlap, with both activities being reliant on Asp-50, Arg-52, and His-84. However, the finding that the loss of Glu-82 selectively impaired the 3Ј-phosphatase hints that the ribonuclease and phosphatase active sites might not be identical. A kinetic analysis of the reactions of E82A with the D10R2 primer-template verified the accumulation of high levels of the initial D10R1-p intermediate, with diminished conversion to the D10R1-OH end product (Fig. 9D).

DISCUSSION
Pseudomonas LigD has a complex domain architecture composed of nuclease, ligase, and polymerase modules. It is homologous to mycobacterial LigD, which plays a critical role in an NHEJ pathway of DNA repair (8). Here we show that the Pol domain of PaeLigD is able to add ribonucleotides to a DNA primer-template, but this reaction is limited to incorporation of only a few rNMPs because the primer-template is rendered progressively less active as ribonucleotides accumulate at the 3Ј-end. We found previously that LigD prefers rNTPs when performing blunt-end additions in vitro, which is a signature feature of bacterial NHEJ in vivo (8,9). The present study hints that ribonucleotides might also be utilized for short-patch fill-in synthesis prior to ligation. Bebenek et al. (16) have recently reported that yeast DNA polymerase IV (a Pol X family member implicated in yeast NHEJ) is also adept at adding a ribonucleotide to a DNA primer-template.
The major finding here is that PaeLigD has an intrinsic ability to resect the short tract of 3Ј-ribonucleotides of a primertemplate substrate to the point at which the primer strand has a single 3Ј-ribonucleotide remaining. The 3Ј-resection activity is resident within an autonomous 187-amino acid N-terminal nuclease domain of PaeLigD. The 3Ј-ribonuclease activity is the composite of at least two component steps: (i) the 3Ј-terminal ribonucleoside is first removed to yield a primer strand with a FIG. 8. RNA specificity and the role of the 2-OH of the penultimate nucleotide. A, nuclease reaction mixtures (10 l) containing 50 mM Tris-HCl, pH 7.5, 5 mM DTT, 0.5 mM MnCl 2 , 0.5 pmol 32 P-labeled primer-template as specified, and either no enzyme (lanes Ϫ) or 1.2 g (ϳ50 pmol) of Nuc domain (lanes ϩ) were incubated for 20 min at 37°C. The products were resolved by PAGE and visualized by autoradiography. The structures of the 2Ј-modified primer-templates (D10R2*) are illustrated below the gel with the 3Ј-ribonucleotide shaded in gray and the flanking nucleotide containing different 2Ј-sugar substituents indicated by ? B, reaction mixtures (80 l) containing 50 mM Tris-HCl, pH 7.5, 5 mM DTT, 0.5 mM MnCl 2 , 9.6 g (ϳ400 pmol) of Nuc domain, and 4 pmol 32 P-labeled D10R2 or R12 primer-template were incubated at 37°C. Aliquots (10 l) were withdrawn at the times indicated and quenched immediately with EDTA/formamide. The time 0 sample was withdrawn prior to adding Nuc domain. C, the product distributions are plotted as a function of time.
FIG. 9. Effects of alanine mutation on ribonuclease and phosphatase activity. A, aliquots (5 g) of the Ni-agarose preparations of wild-type Nuc domain and the indicated Ala mutants were analyzed by SDS-PAGE. Polypeptides were visualized by staining the gel with Coomassie Blue dye. The positions and sizes (in kDa) of marker polypeptides are indicated on the left. B, nuclease reaction mixtures containing 0.5 pmol 32 P-labeled D10R2 primer-template and 1.2 g (ϳ50 pmol) of wild-type (WT) or mutant protein as specified were incubated for 20 min at 37°C. C, phosphatase reaction mixtures containing 0.25 pmol 32 Plabeled D9R1-p primer-template and 1.2 g of wild-type or mutant protein as specified were incubated for 20 min at 37°C. D, a reaction mixture (100 l) containing 12 g (ϳ500 pmol) of E82A protein and 5 pmol 32 P-labeled D10R2 primer-template was incubated at 37°C. Aliquots (10 l) were withdrawn at the times indicated and quenched immediately with EDTA/formamide. ribonucleoside 3Ј-PO 4 terminus, and (ii) then the 3Ј-PO 4 is hydrolyzed to a 3Ј-OH. This sequential mechanism is supported by (i) kinetic evidence for a precursor-product relationship of the 3Ј-PO 4 and 3Ј-OH strands, (ii) the capacity of the Nuc domain to directly hydrolyze a primer-template with a ribonucleoside 3Ј-PO 4 end at a rate similar to the 3Ј-phosphatase step of the composite ribonucleotide resection reaction, and (iii) the fact that alterations of the substrate or the enzyme can selectively impair the 3Ј-phosphomonoesterase step and thereby result in accumulation of the 3Ј-PO 4 strand.
The 3Ј-ribonuclease and 3Ј-phosphatase activities are both dependent on a divalent cation, specifically manganese. Magnesium is conspicuously unable to support either activity. This metal specificity is notable given that the polymerase activity of the PaeLigD is also manganese-dependent (9) and in light of the recent report by Daly et al. (17) that the accumulation of intracellular manganese by Deinococcus facilitates the extreme resistance of this bacterium to ionizing radiation.
The 3Ј-nuclease activity of PaeLigD ceases when a single ribonucleotide remains at the 3Ј terminus of the primer-template. The failure to digest beyond this point reflects the stringent requirement for a 2Ј-OH group on the penultimate nucleoside of the primer strand. Replacing the 2Ј-OH by a 2Ј-F, 2Ј-NH 2 , 2Ј-OCH 3 , or 2Ј-H abolishes the phosphodiesterase reaction. A simple explanation for these modification interference effects is that the enzyme catalyzes scission of the phosphodiester via a 2Ј,3Јcyclic phosphate intermediate. Alternatively, the 2Ј-OH could be required for coordination of the metal cofactor. Further studies will be required to discriminate the two models.
To our knowledge, there is no known nuclease activity with the properties we describe here for PaeLigD, to wit: (i) it catalyzes metal-dependent scission of the P-O5Ј-bond of a 3Ј-phosphodiester to yield a 3Ј-PO 4 , (ii) it separately catalyzes metaldependent hydrolysis of a P-O3Ј-bond of a phosphomonoester, (iii) it acts preferentially on a duplex primer-template substrate rather than a single strand of identical composition, and (iv) it prefers DNA primer strands containing a short 3Ј-ribo tract to an all-RNA primer.
The LigD ribonuclease reaction is unlike that catalyzed by the familiar ribonucleases A, T1, and T2. The latter enzymes also generate a 3Ј-PO 4 -terminated product and do so via a cyclic intermediate (18). However, unlike PaeLigD, RNase A-type enzymes act endonucleolytically and do not rely at all on a divalent cation. Their catalytic power is derived principally from two histidine general acid-base catalysts and a lysine that stabilizes the transition state of the scissile phosphodiester (reviewed in Ref. 18). Although PaeLigD Nuc activity depends on at least one conserved histidine (His-84), there is no primary structural similarity between RNase A-type enzymes and LigD to suggest that their catalytic mechanism is similar. Moreover, RNase A-type enzymes do not hydrolyze 3Ј-phosphomonoesters.
LigD bears no overt resemblance, structurally or mechanistically, to any of the several families of metal-dependent 3Ј-exoribonucleases classified by Deutscher and co-workers (19,20), which include the RNaseII-RNaseR group, the DEDD superfamily, and the RBN family. Though metal dependent and 3Ј directional, these enzymes differ from Pae-LigD in that they all hydrolyze the P-O3Ј-bond of the scissile phosphodiester to yield a 3Ј-OH end directly and liberate 5Ј-nucleoside monophosphates. Although the 3Ј-exoribonuclease enzymes may rely on aspartates, glutamate, and histidine side chains as catalysts, e.g. in coordinating the metal ions and the NMP 5Ј-phosphate (21), their signature motifs are not present in the nuclease domain of LigD and its homologs. Rather, our mutational analysis highlights 4 can-didate catalytic residues located within two conserved motifs in PaeLigD. The proximal peptide motif 47 LHYDFRLE 54 , which contains the essential Asp-50 and Arg-52 side chains, consists of alternating charged residues separated by alternating hydrophobic side chains, an arrangement suggestive of a ␤-strand. The distal charge cluster motif 82 EDH 84 is invariant and includes the essential residue His-84 and a Glu-82 side chain implicated in the 3Ј-phosphatase reaction.
LigD is also functionally distinct from exonuclease III, a metal-dependent DNA 3Ј-exonuclease that, like PaeLigD, has an intrinsic polynucleotide 3Ј-phosphatase activity but, unlike PaeLigD, hydrolyzes the P-O3Ј-bond of the scissile phosphodiester to yield a 3Ј-OH end directly and liberate 5Ј-deoxynucleoside monophosphates, with no requirement for a 2Ј-OH (22,23).
The 3Ј-phosphatase activity of LigD differs from that of the bacteriophage T4, mycobacteriophage, and baculovirus polynucleotide kinase-phosphatase proteins, insofar as the viral enzymes can use magnesium as the divalent cation cofactor and they readily dephosphorylate single-stranded oligonucleotides (or 3Ј-nucleoside monophosphates) with no dependence on a template strand (24 -26). The 3Ј-phosphatase domains of the viral polynucleotide kinase-phosphatase enzymes belong to the DxDxT superfamily of acylphosphatases (25)(26)(27)(28), to which the Nuc domain of LigD bears no structural similarity.
Our results indicate that PaeLigD exemplifies a novel class of 3Ј-remodeling enzyme. The ribonucleotide resection activity is potentially relevant to LigD function in bacterial NHEJ, insofar as we find that the strand-joining activity of PaeLigD is stimulated by the presence of a single ribonucleotide at the 3Ј-OH terminus. 2