Structure of bacterial LigD 3′-phosphoesterase unveils a DNA repair superfamily

Abstract

The DNA ligase D (LigD) 3′-phosphoesterase (PE) module is a conserved component of the bacterial nonhomologous end-joining (NHEJ) apparatus that performs 3′ end-healing reactions at DNA double-strand breaks. Here we report the 1.9 Å crystal structure of Pseudomonas aeruginosa PE, which reveals that PE exemplifies a unique class of DNA repair enzyme. PE has a distinctive fold in which an eight stranded β barrel with a hydrophobic interior supports a crescent-shaped hydrophilic active site on its outer surface. Six essential side chains coordinate manganese and a sulfate mimetic of the scissile phosphate. The PE active site and mechanism are unique vis à vis other end-healing enzymes. We find PE homologs in archaeal and eukaryal proteomes, signifying that PEs comprise a DNA repair superfamily.

DNA ligase D (LigD) is the key agent of the bacterial nonhomologous end-joining (NHEJ) pathway of DNA double-strand break (DSB) repair (1). LigD is a single polypeptide consisting of three autonomous catalytic domain modules: an ATP-dependent ligase (LIG), a polymerase (POL), and a 3′-phosphoesterase (PE). The POL domain incorporates dNMP/rNMPs at DSB ends and gaps prior to strand sealing by the LIG domain (2–6) and is responsible, in large part, for the mutagenic outcomes of bacterial NHEJ in vivo (7). The PE domain provides a 3′ end-healing function, whereby it cleans up “dirty” DSBs with 3′-phosphate ends (8). PE also trims short 3′-ribonucleotide tracts (produced by POL) to generate the 3′ monoribonucleotide ends that are the preferred substrates for sealing by bacterial NHEJ ligases (3, 8). The biochemical properties and atomic structures of the LIG and POL domains highlighted their membership in the covalent nucleotidyltransferase and archaeal/eukaryal primase-polymerase families respectively (5, 9, 10). By contrast, the PE domain appears to be sui generis.

The properties of the PE domain elucidated initially for Pseudomonas LigD also apply to the PE modules of Agrobacterium and Mycobacterium LigD (8, 11, 12). Specifically, PE displays a distinctive manganese-dependent 3′-ribonuclease/3′-phosphatase activity, entailing two component steps: (i) the 3′-terminal nucleoside is removed to yield a primer strand with a ribonucleoside 3′-PO₄ terminus; (ii) the 3′-PO₄ is hydrolyzed to a 3′-OH (Fig. 1A). PE activity is acutely dependent on the presence and length of a 5′ single-strand tail on a duplex primer-template substrate, thus implicating PE in 3′ end repair at gaps or recessed DSBs. Structure probing of Pseudomonas PE in solution revealed an apparently disordered N-terminal 29-aa segment, punctuated by a cluster of trypsin- and chymotrypsin-sensitive sites (Fig. 1B), flanking a seemingly well folded (i.e., protease insensitive) C-terminal domain (13). Deletion of the protease-sensitive N-terminal peptide had no effect on the phosphodiesterase activity of PE, though monoesterase activity was reduced. Mutational analyses identified an ensemble of conserved side chain functional groups within the protease-resistant module that were essential for phosphoesterase activity (Fig. 1B, highlighted in yellow) and thus candidates to comprise the active site (8, 11, 13). Because LigD PE has no apparent primary structural or mechanistic similarity to any other proteins that remove nucleotides or phosphates from DNA or RNA, it was proposed to exemplify a unique family of 3′ end-healing enzymes. Here we put this idea to the test and determined the atomic structure of the Pseudomonas PE domain.

Fig. 1.

The bacterial LigD PE family of 3′ end repair enzymes. (A) Two-step 3′ end-healing activity of the LigD PE domain. The reaction mechanism entails an initial phosphodiesterase step that incises the 3′-diribonucleotide linkage of the primer template to generate a monoribonucleotide-3′-phosphate-terminated primer strand and release a monoribonucleoside. The phosphomonoesterase activity then releases inorganic phosphate and leaves a 3′-monoribonucleotide primer-template end-product. Both steps depend on manganese. The time course of the reaction of the Pseudomonas PE domain with a diribonucleotide-terminated primer-template (ribonucleotides colored red) is shown at left. The D10R2 primer strand (composed of 10 deoxynucleotides and two ribonucleotides) is labeled at the 5^′-PO₄ end with ³²P (indicated by the asterisk). The radiolabeled reaction products were analyzed by denaturing gel electrophoresis. The experiment (8) establishes the precursor-product relationship between the fast migrating D11R1p species and the D11R1_OH end-product. (B) Conservation of primary and secondary structure among bacterial LigD PE domains. The amino acid sequence of the Pseudomonas aeruginosa (Pae) LigD PE domain is aligned to the PE domains of two Agrobacterium tumefaciens (Atu) LigD paralogs (D1 and D2) and the PE domain of Mycobacterium tuberculosis (Mtu) LigD. Positions of amino acid side chain identity/similarity in all four proteins are denoted by solid black dots above the sequence. Six conserved residues shown previously to be essential for PaePE activity are highlighted in yellow. Protease-sensitive sites in the N-terminal peptide of PaePE are denoted by blue arrows. The secondary structure elements of the PE fold are shown above the sequence, with β strands depicted as arrows and helices as cylinders. The eight strands that comprise the PE β barrel (Fig. 2A) are numbered. Conserved residues located in the hydrophobic interior of the β-barrel are denoted by black arrowheads below the sequence.

Results and Discussion

Crystallization and Structure Determination.

Whereas our attempts to crystallize the full-length PE domain (aa 1–187) were unsuccessful, we noted that inclusion of trace amounts of chymotrypsin in the precipitant solution resulted in growth of tiny crystals. Taking this cue, we conducted crystallization trials with recombinant PE (17–187), a catalytically active phosphodiesterase initiating at a chymotryptic cleavage site. Crystals grew in the presence of PEG, ammonium sulfate, and manganese, but diffracted poorly. Additive screening identified yttrium as uniquely effective in promoting growth of diffraction quality crystals. Diffraction data at 1.9 Å and 2.5 Å resolution were collected for native and methylmercury-treated crystals in space group P1. Phases were obtained by single isomorphous replacement with anomalous Hg/Y/Mn scattering (see SI Methods in SI Appendix). The structure was ultimately refined at 1.92 Å resolution with R/R_free values of 0.148/0.199 and excellent geometry (SI Appendix: Table S1). The P1 crystals contained two PE protomers in the asymmetric unit. The model of the A protomer comprised aa 34–185 (with a 4-aa gap at residues 98–101); the B protomer consisted of aa 30–187 (with a single aa gap at residue 99). Absence of an extensive protomer interface in the crystal was consistent with the monomeric state of PE in solution (8). The salutary effects of yttrium on PE crystallization were evident in the refined structure, which revealed four yttrium atoms bridging the A protomer to two vicinal PE protomers in the lattice (SI Appendix: Fig. S1). The yttrium cations were chelated by pairs or trios of carboxylate side chains located on the surfaces of adjacent PE protomers (SI Appendix: Fig. S1). The A and B protomers in the P1 lattice had virtually identical tertiary structures (rmsd of 0.33 Å at 145 α carbon positions). We also collected 2.3 Å diffraction data on a crystal in space group C2 and solved that structure by molecular replacement (SI Appendix: Table S1). There were no significant differences in the PE structure in space group C2 vis à vis the P1 structure (rmsd of 0.27 Å at 145 Cα positions). The descriptions of the PE structure and active site that follow are those of the A protomer in the P1 crystal.

Overview of the PE Tertiary Structure.

The PE domain comprises a central 8-stranded antiparallel β barrel surrounded by two α helices and a 3₁₀ helix (Fig. 2A). The secondary structure elements are displayed above the PE amino acid sequence in Fig. 1B and the folding topology is diagrammed in SI Appendix: Fig. S2. The PE β barrel has a narrow aperture filled with hydrophobic residues (Fig. 2B) that are conserved in other LigD PE domains (these are denoted by black arrowheads in Fig. 1B). Among these are a constellation of tryptophans that, in addition to their van der Waals contacts, donate hydrogen bonds from their indole nitrogens to PE main chain carbonyls (e.g., Trp108, Trp113, and Trp141) and form π-cation pairs with overlying lysine/arginine side chains (e.g., Trp108-Lys136, Trp155-Arg145, and Trp141-Lys159 pairs) (Fig. 2B), said pairs being conserved among bacterial PE domains (Fig. 1B). Thus, there appears to be a characteristic pattern of amino acid side chain and main chain interactions that stabilize the PE β barrel.

Fig. 2.

Tertiary structure of the PE domain. A stereo view of the Pseudomonas LigD PE structure is show in A. The fold comprises an 8-stranded antiparallel β barrel flanked by a 3₁₀ helix and two α-helices. The N and C termini (residues 34 and 185, respectively) are indicated. The β strands are numbered as in Fig. 1B. The active site, demarcated by protein-bound manganese (magenta sphere) and sulfate (stick model) ligands, is located in a crescent-shaped groove on the outer surface of the β-barrel. B shows a detailed view of the hydrophobic interior of the PE β-barrel, viewed from the “back” opening relative to the view in A (note the manganese is now on the left). Selected amino acid side chains and main chain atoms are rendered as stick models with beige carbons. Hydrogen bonds are indicated by dashed lines.

Indeed, a three-dimensional homology search of the protein data bank using DALI (14) highlights the PE domain as a unique protein fold. The five top DALI “hits,” with relatively feeble Z scores of 3.6 to 4.2, were members of the triphosphate tunnel metalloenzyme (TTM) superfamily of phosphohydrolases, which are composed of an 8-stranded antiparallel β barrel of quite different folding topology (15, 16). The TTM barrel has a wider aperture and an extremely hydrophilic interior that binds the divalent cation and polyphosphorylated substrate on which the TTM enzymes act (17). Whereas the TTM active site is inside the β barrel, the PE active site is clearly located on the solvent-exposed surface of the β barrel, as demarcated by the positions of closely spaced manganese and sulfate ligands (Fig. 2A). The PE active site architecture is discussed in detail below.

Active Site and Catalytic Mechanism.

The PE β barrel supports a crescent-shaped hydrophilic active site on its outer surface (Fig. 2A) that binds the essential manganese cofactor and a sulfate anion that we regard as a steric and electrostatic mimetic of the scissile phosphate. The active site architecture and pertinent atomic interactions are depicted in Fig. 3. The electron density map highlights an octahedral coordination complex about the manganese ion, which is filled by His42-Nδ, His48-Nϵ, Asp50-Oϵ, a sulfate oxygen, and two waters (Fig. 3B). The structure teaches us that a likely catalytic role of the metal ion is to aid in substrate binding and then stabilize the developing negative charge on the scissile phosphate in the presumptive associative transition state. The structure also accounts for the characteristic specificity of PE for manganese (and cobalt, cadmium, or copper) as the metal cofactor, and its inability to utilize magnesium (8), insofar as the reliance on two histidine nitrogens as metal ligands will favor “soft” metal interactions in contrast to the “hard” oxygen-based contacts preferred by magnesium. The functional relevance of the three manganese-binding side chains is already established by mutational studies, to wit: H42A, H48A, and D50A mutants of Pseudomonas PE are catalytically inert (8, 13). Moreover, each metal-binding residue is strictly essential, insofar as conservative mutants H42N, H42Q, H48N, H48Q, D50E, and D50N are also unreactive (11).

Fig. 3.

PE active site. (A) Stereo view of the active site. The main chain ribbon is colored green. Selected amino acid side chains and main chain atoms, and the sulfate anion (a putative mimic of the scissile phosphate), are rendered as stick models with beige carbons. Manganese and waters are depicted as magenta and red spheres, respectively. Interatomic contacts are indicated by dashed lines. (B) Stereo view of a finely sampled 2.0 Å composite omit density map of the PE active site contoured at 1.1 σ (black mesh; grid spacing 0.25 Å). The red mesh is the anomalous difference density for the manganese ion contoured at 15 σ.

The sulfate anion is bound in the active site via a network of direct and water-mediated contacts to the metal and to amino acid side chains His84, Arg52, and Tyr88 (Fig. 3A). His84-Nϵ makes a direct contact to the sulfate (Fig. 3A). His84 is strictly essential for PE function, i.e., the H84A, H84N, and H84Q mutants are inert (8, 13). We surmise that His84 promotes catalysis via transition state stabilization. Arg52 is poised at the center of a hydrogen bonding network involving all three guanidinium nitrogens, which variously coordinate three of the sulfate oxygens (via two waters) and directly engage Ser61-Oγ and Gln40-Oϵ (Fig. 3A). Arg52 is strictly essential for PE function, i.e., the R52A, R52K, and R52Q mutants are inert (8, 11). Although we surmise that Arg52 participates directly in phosphoester hydrolysis, it is also likely to play an additional structural role via the Ser61-Arg52-Gln40 triad that tethers the secondary structure elements comprising the active site (Fig. 3A). Interruption of this triad by mutating Gln40 to alanine reduced phosphodiesterase activity by 2.5-fold, and activity was virtually abolished (< 1% of wild type) by replacing Gln40 with glutamate (11); the latter mutation would antagonize the native hydrogen-bonding contacts of Gln40-Nϵ to the Tyr88 and Phe91 main-chain carbonyls) (Fig. 3A).

The Tyr88 OH coordinates a sulfate-bound water in the PE active site (Fig. 3B). Changing Tyr88 to alanine suppressed PE diesterase and monoesterase activities, to 7% and 1% of wild type, respectively (13). It is noteworthy that the conservative Y88F mutant retained activity as a 3′-phosphodiesterase, but was inactive as a 3′-phosphomonoesterase (11). This separation of function suggests that the water-bridged contact between Tyr88 OH and the scissile phosphate is specifically essential for monoester hydrolysis. Accordingly, we speculate that the water interposed between Tyr88-OH and two of the sulfate oxygens is the immediate nucleophile in the 3′ phosphomonoesterase reaction and it relies on Tyr88 for proper orientation during an in-line attack. Also, we surmise that the drastic loss of both phosphoesterase activities when the Tyr88 side chain is subtracted (in Y88A or Y88S) reflects the structural contributions of the many van der Waals contacts (3.5–3.8 Å) between the Tyr88 phenyl ring carbon atoms and the atoms of the surrounding Arg52, His84, Gln40, and Phe91 side chains.

In sum, the PE crystal structure localizes the previously defined essential side chains at the active site and suggests catalytic and/or structural roles for each of the active site residues consistent with available structure-function data. The structure also hints at a plausible model for how PE might bind to a DNA primer-template substrate. The image in SI Appendix: Fig. S3A showing the molecular surface and electrostatics of the PE domain highlights two points: (i) the active site occupies a concave depression on the protein surface that might accommodate the duplex segment of the primer-template; and (ii) the positive electrostatic potential on the protein surface in front of the sulfate might complement that of the electronegative DNA phosphodiester backbone. Using an idealized B-form DNA duplex with a 3′-phosphomonoester terminus on the primer strand and a 5′ single-stranded extension on the template strand, we manually docked the DNA ligand in the active site depression, by superimposing the 3′-phosphate of the primer strand on the sulfate anion. As seen in SI Appendix: Fig. S3B, the resulting docked model entailed no gross steric clashes. Also, this orientation of the primer-template positioned the Tyr88-coordinated water nearly apical to the primer strand terminal O3′, the leaving atom in the 3′ phosphomonoesterase reaction.

Uniqueness of LigD PE Versus Other End-Healing Enzymes.

Although LigD PE displays no overt amino acid sequence similarity to any known polynucleotide end-healing enzymes, it was unclear if the PE domain truly defined a new repair enzyme family. This situation now appears to be the case, given its unique fold, distinctive active site, and biochemical specificities vis à vis other phosphoesterases that repair DNA/RNA ends. Table 1 compares and contrasts PE to the several families of 3′ end-healing enzymes that have been characterized biochemically and structurally, and are variously dedicated to DNA or RNA repair pathways. For example, bacteriophage and mammalian polynucleotide 5′-kinase/3′-phosphatase (Pnkp) are members of the DxD acylphosphatase superfamily that repair the 2′,3′ cyclic phosphate and 3′-phosphate ends of broken RNA and DNA strands, respectively, via a covalent aspartyl-phosphoenzyme intermediate (18–20). Whereas phage and mammalian Pnkps are bifunctional 5^′/3^′ end-healing enzymes, the yeast DNA 3′-phosphatase exemplifies a stand-alone acylphosphatase module (21). Exonuclease III and its homologs are members of the DNase I superfamily; they remove 3′-monophosphates from DNA ends and also have 3′ exonuclease and abasic nuclease activities (22, 23). Tyrosine-DNA-phosphodiesterase, a member of the phospholipase D superfamily, hydrolyzes trapped covalent topoisomerase IB-DNA adducts and DNA 3′-phophoglycolate lesions via a covalent enzyme-(histidinyl-Nϵ)-3′-phosphoryl-DNA intermediate (24).

Table 1.

Distinct families of polynucleotide 3′ end-healing enzymes

Enzyme	Superfamily	Metal cofactor / ligands	DNA/RNA repair
LigD PE	PE	Mn / 2His, Asp	DNA
Bacteriophage Pnkp	DxD acylphosphatase	Mg / 3Asp	RNA
Mammalian Pnkp yeast DNA 3′-phosphatase	DxD acylphosphatase	Mg / 3Asp	DNA
Bacterial Pnkp	Dinuclear metallophosphoesterase	2Mn / 3His, 3Asp, Asn	RNA
Exonuclease III	DNase I	2Mg / Glu, Asp, 2Asn	DNA
tRNA ligase (CPD)	2H	none	DNA
TDP1	Phospholipase D	none	DNA

The bacterial clade of Pnkps can be viewed as the RNA repair analogs of the polyfunctional bacterial LigD DNA repair enzyme, insofar as bacterial Pnkps comprise three catalytic modules in a single polypeptide: a 3′ healing enzyme (belonging to the dinuclear metallophosphoesterase superfamily), a 5′ end-healing enzyme, and an RNA ligase (25, 26). Yeast and plant tRNA ligases are also multidomain RNA repair proteins with ligase and dual end-healing modules, including a 3′ end-healing enzyme (2′, 3′ cyclic phosphodiesterase) that belongs to the 2H superfamily (27).

Most remarkably, even though many of the known 3′ repair enzymes are metal-dependent (or specifically manganese-dependent, in the case of bacterial Pnkp), none of their active sites and metal coordination complexes resemble the apparently unique configuration in LigD PE (Table 1). Thus, we see no evidence to indicate convergence of PE with any known phosphoesterase catalytic site.

A PE Superfamily in Bacteria, Archaea, and Eukarya.

Having installed LigD PE as the founder of an enzyme family, we queried whether the PE module might have achieved broader use in nature—in functional and phylogenetic contexts outside the polyfunctional LigD polypeptide that anchors the NHEJ systems found in scores of bacterial taxa. Reasoning that PE might have existed as a free standing catalytic module prior to its fusion with LIG and POL to form LigD, we surveyed bacterial proteomes for PE homologs that were similar in size to the Pseudomonas PE module. We thereby identified 21 stand-alone PE enzymes that contained the full set of active site residues. The bacterial taxa encoding stand-alone PE enzymes span eight different phyla (SI Appendix: Table S3).

PE homologs are not confined to bacteria. For example, we detected stand-alone PE domains in the proteomes of seven archaeal species, representing five genera of the phylum Euryarchaeota (i.e., Methanoculleus, Methanocella, Methanosarcina, and Archaeoglobus, and uncultured methanogen RC-1) (Fig. 4A). The archaeal PE proteins are similar in size (121–199 aa) to the autonomous bacterial PE module, they have strictly conserved counterparts of the six PE active site residues (highlighted in yellow in Fig. 4A), and they have conserved counterparts of the signature residues in the interior of the PE β barrel (denoted by arrowheads in Fig. 4A). Six of the archaeal PE proteins also have conserved counterparts of the N-terminal PE peptide, suggesting that they are likely to possess 3′ phosphodiesterase and monoesterase activities à la the bacterial LigD PE domains. The Archaeoglobus profundus PE lacks the N-terminal peptide, and is thereby likely to be proficient as a 3′ phosphodiesterase, with weaker monoesterase activity, similar to the N-terminal deletants of Pseudomonas PE (13).

Fig. 4.

PE family members in archaea and eukarya. (A) Archaeal PE homologs. PaePE is aligned to the homologous polypeptides encoded by Methanoculleus marisnigri (Mma), Methanocella paludicola (Mpa), uncultured methanogenic archaeon RC-I (RC-I), Methanosarcina barkeri (Mba), Methanosarcina mazei (Mma), Methanosarcina acetivorans (Mac), and Archaeoglobus profundus (Apr). (B) Eukaryal PE homologs. PaePE is aligned to the homologous polypeptides encoded by Aspergillus nidulans (Ani), Aspergillus clavatus (Acl), Neosartorya fischeri (Nfi), Talaromyces stipitatus (Tst), Penicillium marneffei (Pma), Penicillium chrysogenum (Pch), Coccidioides immitis (Cim), Phaeosphaeria nodorum (Pno), Verticillium albo-atrum (Val), and Cryptococcus neoformans (Cne). Positions of amino acid side chain identity/similarity are denoted by dots above the aligned sequences. Conserved PE active site residues are highlighted in yellow. The secondary structure elements of the PE fold are shown above the sequences. Conserved residues located in the hydrophobic interior of the PaePE β-barrel are denoted by black arrowheads below the aligned sequences.

We inspected each archaeal genomic PE locus for potential clustering of genes involved in DNA repair or modification. The Methanoculleus, Methanosarcina, and methanogen RC-I PEs had no such neighbors. The Archaeoglobus PE gene is immediately downstream of a cooriented ORF encoding a putative DnaG-type DNA primase, which could signify a functional relationship between these two enzymes. Notably, the Methanocella paludicola PE is encoded by the proximal ORF of a cooriented three ORF cluster comprising PE, an ATP-dependent DNA ligase, and a LigD-like polymerase. We surmise that the archaeon M. paludicola has a three-component equivalent of the trifunctional bacterial LigD enzyme.

Equally remarkable was the appreciation that PE homologs are extant in eukarya, specifically in fungi. The PE-encoding fungi represent seven genera from phylum Ascomycota (Aspergillus, Neosartorya, Talaromyces, Penicillium, Coccidioides, Phaeosphaeria, and Verticillium) and a single taxon from phylum Basidiomycota (Cryptococcus neoformans). An alignment of Pseudomonas PE to its fungal homologs highlights strict conservation of five of the active site constituents and most of the signature residues in the interior of the PE β barrel. The Tyr88 active site moiety of bacterial PE is conserved in Cryptococcus PE, but substituted by asparagine in the other fungal PE homologs (Fig. 4B). The activity and physiology of the fungal PE homologs merits further attention, especially as several of the PE-encoding fungi are human pathogens.

In conclusion, our elucidation of the atomic structure of LigD PE has revealed a DNA repair enzyme family that is distributed broadly among taxa in all three phylogenetic domains.

Materials and Methods

Detailed methods are provided in SI Appendix and are summarized briefly here. The Pseudomonas PE domain was produced in Escherichia coli as a His₁₀Smt3 fusion and then isolated from a soluble bacterial lysate by Ni-affinity chromatography. After tag removal by treatment with Smt3 protease Ulp1, the tag-free PE domain was purified by phosphocellulose and gel filtration chromatography. PE crystals were grown by vapor diffusion at room temperature against buffer containing 30% PEG-5000-MME, 100 mM MES (pH 7.0), 200 mM ammonium sulfate, 10 mM yttrium chloride, 2 mM MnCl₂. Diffraction data at 1.9 to 2.5 Å resolution were collected at National Synchrotron Light Source (NSLS), Brookhaven NY. Phases for a Hg-soaked triclinic crystal were determined by using a modified Single Isomorphous Replacement with Anomalous Scattering (SIRAS) method that included contributions from anomalous scatterers in a native P1 crystal dataset. Electron density maps revealed two PE molecules in the unit cell. The final refined model at 1.92 Å resolution (R/R_free = 0.148/0.199) included 305 amino acids from the two PE protomers, with excellent geometry (SI Appendix: Table S1). We also solved via molecular replacement the 2.3 Å structure of PE crystallized in space group C2 (R/R_free = 0.212/0.263). The crystallographic data and refinement statistics are compiled in SI Appendix: Table S1. The coordinates for the P1 and C2 structures have been deposited in PDB under ID codes 3N9B and 3N9D.