Structures of ATP-bound DNA ligase D in a closed domain conformation reveal a network of amino acid and metal contacts to the ATP phosphates

DNA ligases are the sine qua non of genome integrity and essential for DNA replication and repair in all organisms. DNA ligases join 3′-OH and 5′-PO4 ends via a series of three nucleotidyl transfer steps. In step 1, ligase reacts with ATP or NAD+ to form a covalent ligase-(lysyl-Nζ)–AMP intermediate and release pyrophosphate (PPi) or nicotinamide mononucleotide. In step 2, AMP is transferred from ligase-adenylate to the 5′-PO4 DNA end to form a DNA-adenylate intermediate (AppDNA). In step 3, ligase catalyzes attack by a DNA 3′-OH on the DNA-adenylate to seal the two ends via a phosphodiester bond and release AMP. Eukaryal, archaeal, and many bacterial and viral DNA ligases are ATP-dependent. The catalytic core of ATP-dependent DNA ligases consists of an N-terminal nucleotidyltransferase domain fused to a C-terminal OB domain. Here we report crystal structures at 1.4–1.8 Å resolution of Mycobacterium tuberculosis LigD, an ATP-dependent DNA ligase dedicated to nonhomologous end joining, in complexes with ATP that highlight large movements of the OB domain (∼50 Å), from a closed conformation in the ATP complex to an open conformation in the covalent ligase-AMP intermediate. The LigD·ATP structures revealed a network of amino acid contacts to the ATP phosphates that stabilize the transition state and orient the PPi leaving group. A complex with ATP and magnesium suggested a two-metal mechanism of lysine adenylylation driven by a catalytic Mg2+ that engages the ATP α phosphate and a second metal that bridges the ATP β and γ phosphates.

Polynucleotide ligases join 3Ј-OH and 5Ј-PO 4 ends via a series of three nucleotidyl transfer steps. In step 1, ligase reacts with ATP or NAD ϩ to form a covalent ligase-(lysyl-N)-AMP intermediate and release pyrophosphate (PP i ) or nicotinamide mononucleotide. In step 2, AMP is transferred from ligase-adenylate to the 5Ј-PO 4 DNA or RNA end to form a DNA-adeny-late or RNA-adenylate intermediate (AppDNA or AppRNA). In step 3, ligase catalyzes attack by a DNA or RNA 3Ј-OH on the polynucleotide-adenylate to seal the two ends via a phosphodiester bond and release AMP. All steps in the ligase pathway require a divalent cation cofactor.
The step1 auto-adenylylation reaction of polynucleotide ligases relies on a conserved nucleotidyltransferase (NTase) 2 domain that includes six defining peptide motifs that form the nucleotide-binding pocket (1,2). Motif I (KXXG) contains the lysine that becomes covalently attached to the AMP. RNA ligases and DNA ligases have different nucleic acid substrate preferences for step 2 of the end-joining pathway that are, to a first approximation, dictated by their distinctive structural modules appended to the NTase domain. DNA ligases comprise a coherent family insofar as they all have a core catalytic unit composed of an NTase domain fused at its C terminus to an OB domain.
ATP-dependent DNA ligases are present in all eukarya and archaea as well as in many bacteria and DNA viruses. These DNA ligases play key roles in DNA replication, repair, and recombination. They vary greatly in their domain complexity, from minimal versions that consist of little more than the NTase-OB core to larger forms with various additional domains that abet DNA recognition and protein-protein interactions. Within a given taxon, two or more DNA ligase enzymes may coexist, often with a division of labor between them. Examples of minimal ATP-dependent ligases include bacteriophage T7 DNA ligase (3), Chlorella virus DNA ligase (4,5), bacterial nonhomologous end-joining (NHEJ) ligases LigD and LigC (6)(7)(8), and bacterial DNA ligase LigE (9,10). The more complex multidomain architectures of ATP-dependent DNA ligases are exemplified by mammalian enzymes Lig1, LigIII, and LigIV (11)(12)(13)(14).
Our understanding of DNA ligase mechanism and specificity has been advanced via crystal structures of exemplary enzymes, captured variously as the apoenzyme, the covalent ligase-(lysyl-N)-AMP intermediate (step 1 product), ligase-AMP bound to a DNA nick (step 2 substrate), ligase bound to an AppDNA nick (step 2 product), ligase bound to a nick that was sealed in crystallo (step 3 product), or ligase bound noncovalently to AMP (3-5, 8 -20). A salient theme from the available structures is that the position of the OB domain (and other flanking domains) relative to the NTase domain is variable according to the functional state in which the ligase is crystallized. This has prompted the suggestion, supported by NMR studies (21), that the ligase domains are flexible in solution and that specific movements accompany or orchestrate the steps of the catalytic cycle. In particular, the OB domain moves when ligase engages its DNA substrate such that the enzyme forms a C-shaped clamp at the nick in which the OB domain engages the minor groove (4, 10 -12, 14, 15).
Two aspects of the ATP-dependent ligase mechanism remain poorly understood in structural terms: (i) the enzymic interactions and catalytic roles of the essential divalent cation cofactor(s) and (ii) the nature of the step 1 Michaelis complex of DNA ligase with ATP and metals. Most structures of ATP-dependent DNA ligases have no divalent cation in the refined model, even in cases in which an active metal cofactor was included during crystallization. Structures have been reported for T7 DNA ligase and human LigIV in complex with ATP (3,13), but these are "off pathway" with respect to the lysine adenylylation reaction because the PP i leaving group (comprising the ATP ␤ and ␥ phosphates) is oriented orthogonal to the motif I lysine nucleophile (N-P␣-O3␣ angle of Յ90°), a situation inimical to an in line-attack by lysine on the ATP ␣ phosphate. In these off-pathway ligase⅐ATP complexes, which lack a metal cofactor, the OB domain is splayed out away from the NTase domain, and there are no enzymic contacts to the ATP ␤ and ␥ phosphates. In the structure solved for crystals of Sulfolobus DNA ligase that had been soaked in ATP, the PP i leaving group is apical to the lysine nucleophile (N-P␣-O3␣ angle ϭ 155°), but there is no reaction in crystallo in the absence of a metal cofactor (17); in addition, the Sulfolobus OB domain is reflected away from the NTase domain. It is thought that step 1 catalysis of lysyl-NMP formation is driven by the presence of metal cofactor(s) and by adoption of a closed conformation of the OB domain in a step 1 Michaelis complex such that the OB domain engages the PP i leaving group and helps place it apical to the lysine nucleophile (22). However, to our knowledge, there are as yet no structural snapshots of an ATP-dependent DNA ligase with ATP and metal cofactor in an on-pathway conformation.
The structure of the MtuLigD-LIG domain, solved previously as the covalent ligase-(lysyl-N)-AMP intermediate (8), resembles other ATP-dependent DNA ligases with respect to its component NTase and OB domains and the conservation of its catalytic motifs. By mutating the motif I lysine to methionine to prevent covalent adenylylation, we now solve a 1.4 Å structure of a LigD-LIG⅐ATP complex and a 1.8 Å structure of a LigD-LIG⅐ATP⅐Mg complex. In both ATP-bound structures, the OB domain is in a closed conformation with respect to the NTase domain. Our results indicate that LigD-LIG employs a two-metal mechanism of lysine adenylylation, driven by: (i) a catalytic Mg 2ϩ that engages the ATP ␣ phosphate and (ii) a second metal that bridges the ATP ␤ and ␥ phosphates. Enzymic contacts to the ATP ␥ phosphate orient the PP i leaving group favorably for in-line catalysis.

M. tuberculosis LigD-LIG prefers to seal a 3-OH monoribonucleotide nick
Many bacterial taxa (e.g. Mycobacterium, Pseudomonas, Agrobacterium) have a NHEJ system of DNA double-strand break (DSB) repair driven by Ku and one or two dedicated ATPdependent DNA ligases (LigD and LigC) (30). LigD is a multifunctional enzyme composed of a ligase (LIG) domain fused to two other catalytic modules: a polymerase (POL) that preferentially adds ribonucleotides to DSB ends and a phosphoesterase that trims 3Ј oligoribo tracts until only a single 3Ј ribonucleotide remains. LigD and LigC are conspicuously feeble at sealing 3Ј-OH/5Ј-PO 4 DNA nicks in vitro. This property distinguishes them from LigA, the essential replicative bacterial ligase. Previous studies showed that efficient nick sealing by Pseudomonas and Agrobacterium LigD and LigC enzymes requires the presence of a single ribonucleotide at the broken 3Ј-OH end (29). The ribo effect on those LigD and LigC enzymes is specific for the 3Ј terminal nucleotide and is either diminished or abolished when additional vicinal ribonucleotides are present. In vitro repair of a DSB by LigD requires the POL module and results in incorporation of an alkali-labile ribonucleotide at the repair junction (29). Those results illuminated an underlying logic for the organization of LigD, whereby POL and phosphoesterase can heal the broken 3Ј end to generate the monoribo terminus favored by the NHEJ ligases.
It was of interest here to query whether the property of 3Ј monoribonucleotide specificity applies to the LIG component of M. tuberculosis LigD. The autonomous LIG domain of the 759-amino acid MtuLigD protein extends from amino acids 452 to 759 (8). We reacted purified recombinant MtuLigD-LIG with two different nicked 36-bp duplexes (5Ј 32 P-labeled on the nick 5Ј-PO 4 strand) in which the unlabeled 18-mer 3Ј-OH strand consisted of either 18 deoxynucleotides (D18) or 17 deoxynucleotides and 1 ribonucleotide (D17R1) (Fig. 1). An enzyme titration experiment revealed that the specific activity of MtuLigD-LIG was 3.5-fold higher on the D17R1 nick (5.8 fmol nicks sealed per fmol LIG) than on the D18 nick (1.65 fmol nicks sealed per fmol LIG) (Fig. 1A). Both substrates were sealed to an equivalent extent at saturating enzyme. An analysis of the kinetics of nick sealing under conditions of enzyme excess is shown in Fig. 1B. The apparent ligation rate constant for D17R1 nick ligation (0.128 Ϯ 0.0059 s Ϫ1 ) was 12-fold faster than the rate constant for D18 nick ligation (0.011 Ϯ 0.00098 s Ϫ1 ). Thus, the introduction of a single 3Ј ribonucleotide at the break stimulates nick sealing by MtuLigD-LIG.

Structure of MtuLigD-LIG in a complex with ATP
The structure of the MtuLigD-LIG domain, solved previously as the covalent ligase-(Lys481-N)-AMP intermediate (8), resembles prototypal ATP-dependent DNA ligases with respect to its core NTase and OB domains and the conservation of its NTase motifs (I, Ia, III, IIIa, IV, V, and VI), all of which are shared with NHEJ ligases D and C from other bacteria (Fig. 2). In the LigD-LIG-AMP structure, the NTase and OB domains are splayed wide apart in an "open" conformation that exposes the AMP phosphate on the surface of the NTase domain (Fig.  3B). An open conformation allows binding of the DNA substrate so that a reactive 5Ј-PO 4 DNA end is placed next to the AMP phosphate in preparation for step 2 adenylate transfer to form the AppDNA intermediate.
Our aim here was to capture a complex of MtuLigD-LIG with ATP in a state poised for catalysis of step 1 lysine adenylylation.
To that end, we exploited a mutated version, K481M, in which the Lys 481 nucleophile was replaced with methionine to preclude formation of the LIG-AMP intermediate. LIG-K481M was preincubated with 1 mM ATP and 5 mM MgCl 2 prior to crystallization by vapor diffusion against precipitant solution containing 0.1 M MES and 25% PEG 10000. Crystals diffracting to 1.4 Å resolution were in space group P2 1 with one LIG protomer in the asymmetric unit. The refined 1.4 Å structure (R work /R free ϭ 0.188/0.198; Table 1) consisted of two polypeptide segments, from 453 to 652 and 659 to 759, punctuated by a disordered 6-amino acid surface loop. Electron density for ATP and MES, the buffer in the precipitant solution, was evident in the active site (Fig. 4); no density consistent with a metal ion was seen. In the ATP-bound ligase, the N-terminal NTase domain (aa 453-639) and the C-terminal OB domain (aa 640 -759) adopt a "closed" conformation in which the OB domain undergoes a rigid body movement that brings it closer to the ATP phosphates (Fig. 3A). The conformational switch occurs around a pivot point at the interdomain junction (denoted by the arrow in

Active site with ATP and MES
Fig. 5A shows a stereo view of the active site of the LIG-K481M⅐ATP⅐MES complex, highlighting atomic interactions with the ATP ribose and phosphates, the nearby MES ligand, and pertinent inter-NTase-OB domain contacts. The ATP ribose O2Ј receives a hydrogen bond from Arg 486 (motif I). Arg 745 (motif VI) makes bifurcated hydrogen bonds to the ATP ␣ phosphate and a bidentate salt bridge to Asp 483 (motif I). Arg 748 (motif VI) makes a bidentate interaction with the ATP ␤ phosphate. Glu 727 (a conserved residue in the OB domain; Fig.  2) forms a bifurcated salt bridge to Arg 745 and Arg 748 . Lys 635 (motif V) makes bifurcated hydrogen bonds to the ATP ␣ and ␥ phosphates. The ␥ phosphate also receives bidentate hydrogen bonds from Arg 501 (motif Ia) and Arg 629 (a conserved residue preceding motif V; Fig. 2) and a hydrogen bond from the ATP ribose 3Ј-OH. The MES morpholino ring makes van der Waals contacts with the ATP ribose and ␤ phosphate, and the MES sulfonate oxygens receive hydrogen bonds from Arg 501 and the Arg 486 main-chain amide (Fig. 5A).

Structure of MtuLigD-LIG in a complex with ATP and magnesium
The unanticipated affinity of MES for the LigD-LIG active site (driven, we suspect by enzymic interactions with the MES

DNA ligase mechanism
sulfonate) prompted us to cocrystallize LIG-K481M with ATP and magnesium under different conditions that might be more conducive to metal occupancy of the active site. LIG-K481M was preincubated with 1 mM ATP and 20 mM MgCl 2 prior to crystallization by vapor diffusion against precipitant solution containing 0.1 M ammonium fluoride and 20% PEG 3350. Crystals diffracting to 1.8 Å resolution were in space group P2 1 with one LIG protomer in the asymmetric unit. The refined model (R work /R free ϭ 0.206/0.244; Table 1) included ATP and two magnesium ions in the active site. Superposition of the LIG⅐ATP⅐Mg and LIG⅐ATP⅐MES structures with respect to their NTase domains (Fig. 6A) showed that whereas LIG⅐ATP⅐Mg also adopted a closed NTase-OB conformation, the position of its OB domain was shifted in the LIG⅐ATP⅐Mg structure, such that the OB domain was slightly further away from the NTase domain compared with LIG⅐ATP⅐MES. The extent of the movement was as little as 1 Å for some segments of the OB domain and as much as 5.3 A for other OB domain elements.

DNA ligase mechanism
the "catalytic" metal ion for the lysine adenylylation reaction, directly engages one of the nonbridging ␣ phosphate oxygens at a distance of 2.5 Å. Mg2 is a "noncatalytic" metal that bridges the ATP ␤ and ␥ phosphates, at a distance of 2.6 Å to the ␤ phosphate oxygen and 2.8 Å to the ␥ phosphate oxygen. Mg1 and Mg2 occupy active site positions that overlap the MES molecule in the K481M⅐ATP⅐MES complex (Fig. 5, compare A and B). Mg1 interacts, directly and via a water, with Glu 613 (motif IV). Mg2 interactions include, in addition to the ␤ and ␥ phosphates oxygens, two waters at distances of 2.5 and 2.7 Å. One of these waters bridges Mg2 and Mg1 (Fig. 5B). Mg2 is situated 3.4 Å from the ATP ribose O3Ј.
Comparison of the ATP⅐Mg and ATP⅐MES complexes shows that there are notable changes, particularly with respect to contacts around the ATP ␣ phosphate. Glu 613 severs its salt bridge to Arg 729 (seen in the ATP⅐MES complex) and reorients to engage Mg1 that binds the ␣ phosphate. The salt-bridge network involving Asp 483 , Arg 745 , Glu 727 , and Arg 748 remains intact, but the direct contacts of Arg 745 and Arg 748 with the ATP ␣ and ␤ phosphates (seen in the ATP⅐MES complex) are relinquished, because of the slight movement of the OB domain away from the NTase domain. In the ATP⅐Mg complex, the ␣ phosphate acquires new contacts (vis-à-vis ATP⅐MES), from Lys 637 (motif V) and His 465 . The bidentate contacts of Arg 501 and Arg 629 with the ATP ␥ phosphate are retained in the ATP⅐Mg complex, as is the bifurcated interaction of Lys 635 with the ␣ and ␥ phosphates and the hydrogen bond from Arg 486 to the ribose O 2 Ј. Fig. 6B shows a laterally displaced superposition of the ATP substrate in the K481M⅐ATP⅐MES and K481M⅐ATP⅐Mg structures and the Lys 481 -AMP adduct in the WT LIG-AMP structure solved previously. The adenosine nucleoside is in the anti conformation in each case. As superimposed, the Lys 481 -N is situated 3.6 Å from the ATP ␣ phosphorus in the ATP⅐Mg complex and makes an N-P␣-O3␣ angle of 137°with respect to the pyrophosphate leaving group. Consistent with a singlestep in-line mechanism, we see that the ␣ phosphate undergoes stereochemical inversion during the transition from ATP⅐Mg complex to lysyl-AMP intermediate (Fig. 6B). We envision that (i) the catalytic Mg1 metal ion stabilizes a pentavalent transition state of the ATP ␣ phosphate during the lysine adenylylation reaction, abetted by the two conserved motif V lysines (Lys 635 and Lys 637 ) that contact the ␣ phosphate oxygens, and (ii) the Mg2 metal ion engages the pyrophosphate leaving group and, with the assistance of basic side chains Arg 501 , Arg 629 , and Lys 635 that coordinate the ATP ␥ phosphate, orients it for inline displacement by the Lys 481 nucleophile.

Structure-guided mutagenesis
Previous mutational analysis of MtuLigD had shown that alanine mutations of Asp 483 (motif I), Glu 530 (motif III), Glu 613 (motif IV), and Lys 637 (motif V) abolished DNA nick-sealing activity in vitro (8). Alanine mutation of Lys 635 (motif V) reduced DNA ligation specific activity to 20% of the WT value (8). Here we conducted a new round of alanine scanning of LigD-LIG, guided by the structures of the ATP complexes, targeting the following conserved amino acids: Arg 501 and Arg 629 in the NTase domain that coordinate the ATP ␥ phosphate; and Arg 745 , Glu 727 , and Arg 748 in the OB domain that form a saltbridged network with Asp 483 , the arginines of which interact with the ATP phosphates in the ATP⅐MES complex. The recombinant LigD-LIG-Ala proteins were produced in Esche-

DNA ligase mechanism
richia coli and purified from soluble bacterial lysates in parallel with the WT LigD-LIG as described under "Experimental procedures." SDS-PAGE affirmed that the preparations were similarly enriched with respect to the LIG polypeptide (Fig. 7A). The WT LigD-LIG and LigD-LIG-Ala mutants were titrated for ligation of the D17R1 nick substrate (Fig. 7B). Specific activities were derived by linear regression fitting of the data and normalized to the WT (defined as 100%). The R501A and R629A proteins were 1.2 and 7.2% as active as WT, respectively. We surmise that the Arg 501 interaction with the ATP ␥ phosphate is especially crucial for ligase activity. The E727A, R745A, and R748A mutants were 0.8, 4.3, and 6.2% as active as WT, respectively. It appears that the loss of the Glu 727 salt bridge from Arg 745 to Arg 748 is more deleterious than subtraction of either of the two individual arginine side chains.

Mutational effects on sealing at a preadenylylated nick
WT LigD-LIG and the LigD-LIG-Ala mutants were assayed by enzyme titration for their ability to seal a preadenylylated D17R1 AppDNA nicked substrate in the absence of ATP (Fig.  8), the rationale being that mutational defects in either step 1 or step 2 of the ligase pathway might be bypassed by presenting the mutant enzyme with a preadenylylated nick. The 32 P-labeled AppDNA strand was prepared reaction of the 5Ј 32 P-labeled 18-mer pDNA oligonucleotide with E. coli RtcA and ATP (31). The AppDNA strand was gel-purified and annealed with the D17R1 3Ј-OH strand and complementary 36-mer DNA template strand to form the substrate shown in Fig. 8. AppDNA ligation specific activities were derived by linear regression fitting of the data and normalized to the WT (defined as 100%) as follows: R501A (130%), R629A (140%), E727A (170%), R745A (100%), and R748A (72%). These results signify that Arg 501 , Arg 629 , Glu 727 , Arg 745 , and Arg 748 are not critical for the isolated step 3 reaction of the LigD sealing pathway.

DNA ligase domain dynamics
The covalent lysine nucleotidyltransferase superfamily embraces ATP-dependent DNA ligases, ATP-dependent RNA ligases, NAD ϩ -dependent DNA ligases, and GTP-dependent mRNA capping enzymes (1). The superfamily is defined by (i) the formation of a lysyl-NMP intermediate en route to the nucleotidylation of nucleic acid 5Ј ends and (ii) a conserved NTase catalytic domain and a set of NTase motifs that forms the NMP-binding pocket. Six distinct families of ATP-dependent RNA ligase (Rnl) have been characterized, each consisting of an NTase domain fused to a structurally unique flanking domain that defines the Rnl family (32)(33)(34)(35)(36)(37)(38)(39). By contrast, ATPdependent DNA ligases, NAD ϩ -dependent DNA ligases, and GTP-dependent mRNA capping enzymes share a core structure composed of a proximal NTase domain fused to a distal OB-fold domain (1). Crystal structures of DNA ligases and capping enzymes have revealed considerable variability in the position of the OB domain relative to the NTase domain. In the few cases where structures of the same enzyme have been captured at different stages along the reaction pathway, it seems that OB domain movements occur in sync, or at least correlate, with substrate binding and product release (1, 4, 5, 9, 10, 13-15, 22, 33). In general terms, it is thought that the DNA ligase OB domain is in an open conformation to allow ATP binding, undergoes domain closure for catalysis of lysine adenylylation, and then reopens to allow release of the PP i product and exposure of the DNA nick-binding site overlying the lysyl-AMP phosphate. DNA binding is coupled to reorientation of the OB domain to form a C-shaped clamp around the nick. After step 3 phosphodiester synthesis, the clamp opens to allow release of the ligated DNA and AMP products.
The suite of available structures of LigD-LIG now includes two new complexes with ATP, both with the OB domain in a closed conformation, and a prior structure of the LIG-AMP covalent intermediate, wherein the OB domain is splayed widely open (8). It had been proposed previously, based on the structure of Chlorella virus capping enzyme in a reactive complex with GTP (22), that a consequence of OB domain closure upon NTP binding would be to bring NTase motif VI (located in the OB domain; Fig. 2) into the active site for lysine adenylylation, wherein motif VI basic amino acids would engage the NTP ␤ and ␥ phosphates and orient the PP i leaving group for in-line catalysis. In the capping enzyme⅐GTP complex, the ␥ phosphate is additionally coordinated by three amino acids in the NTase domain: Arg 106 (in motif Ia), Lys 234 (motif V), and Arg 228 (located 6 amino acids upstream of the first motif V Lys) (Fig. 9A). A similar network of contacts to the ATP ␥ phosphate is seen presently in the LigD-LIG⅐ATP complex and in the

DNA ligase mechanism
Sulfolobus ssLig⅐ATP complex (17), involving the equivalent amino acids: LigD Arg 501 and ssLig Arg 280 (motif Ia); LigD Lys 635 and ssLig Lys 433 (motif V); and LigD Arg 629 and ssLig Arg 427 (located 6 amino acids upstream of the motif V Lys 635 and Lys 433 ) (Fig. 9A). We showed here by alanine scanning that Arg 501 is essential and Arg 629 is important for LigD-LIG activity in nick sealing in vitro.
In the closed conformation of the capping enzyme⅐GTP complex, motif VI Lys 298 ( 295 RSDK 298 ) engages the ␥ phosphate, and motif VI Arg 295 coordinates the ␤ phosphate (Fig.  9B). Arg 295 also participates in salt bridges to Asp 297 and Glu 281 . The corresponding amino acids and secondary structure elements of the LigD-LIG OB domain superimpose well on those of the capping enzyme (Fig. 9B), even though their interactions with the ATP nucleotide are different in some respects. To wit, (i) in the relatively more closed domain conformation in the LIG⅐ATP⅐MES structure, motif VI Arg 748 engages the ␤ phosphate, but Lys 751 makes no contacts to ATP; and (ii) in the LIG⅐ATP⅐Mg complex, motif VI has shifted position and makes no direct contacts to ATP (Fig. 5). (Note that LigD Arg 745 , which contacts the ␣ phosphate in the ATP⅐MES complex and is part of the salt-bridge network with motif VI, is not conserved in capping enzymes (40).) In light of the importance of motif VI Arg 748 for nick sealing activity (Fig. 8), we take the view that the present structures of LigD-LIG in complex with ATP sample different states in the trajectory of domain closure en route to lysine adenylylation. We keep open the prospect that an even more closed conformation than what we report here is attained in a true step 1 Michaelis complex, wherein motif VI does directly engage the ATP ␤ and ␥ phosphates.

A two-metal mechanism of LigD-LIG lysine adenylylation
Recent studies have highlighted a conserved mechanism of metal-dependent lysine adenylylation by ATP-dependent RNA ligases and NAD ϩ -dependent DNA ligases, whereby an octahedral catalytic metal-(H 2 O) 5 complex stabilizes the transition state of the ATP or NAD ϩ ␣ phosphate (32,33,35,38,39). Three side chains of NTase motifs I, III, and IV bind the pentahydrated metal cofactor via waters. Structures of the Michaelis complex of RNA ligases from three different families reveal a two-metal mechanism of lysine adenylylation whereby a second octahedral noncatalytic metal complex bridges the ATP ␤ and ␥ phosphates and orients the PP i leaving group for in-line attackbythelysinenucleophile (32,33,39).Bycontrast,NAD ϩ -dependent DNA ligase employs a one-metal mechanism of lysine adenylylation and relies on direct enzymic contacts to orient the nicotinamide mononucleotide leaving group (33).
An account of the metal requirement for lysine adenylylation by ATP-dependent DNA ligases was lacking. The present study of LigD-LIG points to a two-metal mechanism driven by a catalytic magnesium at the ATP ␣ phosphate and a noncatalytic magnesium bridging the ␤ and ␥ phosphates. Although the roles imputed to the two metals in the LigD-LIG⅐ATP structure are analogous to those of the ATP-dependent RNA ligases and although the positions of the motif I, motif III, and motif IV side chains that bind the catalytic metal in the Rnl5 family Michaelis complex (32) are virtually identical to the corresponding Asp 483 , Glu 530 , and Glu 613 residues in the LigD-LIG⅐ATP⅐Mg complex, the specifics of the LigD-LIG metal coordination complexes are different. The LigD catalytic Mg1 coordination complex has only four occupants. The direct Mg1 contact to the ATP ␣ phosphate and a water-mediated contact to Glu 613 are features shared with other ligases. The second Mg1 contact to Glu 613 is direct, rather than water-mediated as in RNA ligases and NAD ϩ -dependent DNA ligase. The motif I Asp 483 is absent from the LigD Mg1 coordination complex, being instead engaged in a salt bridge to Arg 745 . Motif III Glu 530 is also absent. These differences raise two possible scenarios: (i) the catalytic metal in LigD-LIG is engaged to the enzyme in a manner distinct from that employed by RNA ligases and NAD ϩ -dependent DNA ligase; or (ii) the structure of the LIG⅐ATP⅐Mg complex, although solved at high resolution, approximates but does not fully reflect the state of the step 1 Michaelis complex.
In favor of the latter scenario is the structure of Pyrococcus furiosus DNA ligase (PfuLig) in a noncovalent complex with AMP, with a single magnesium ion in the active site and the OB domain in a closed conformation over the NTase domain (18). In the PfuLig⅐AMP structure, a hexahydrated magnesium complex with octahedral geometry is engaged to the enzyme via water-mediated contacts to motif I, motif III, and motif IV carboxylate side chains in a manner identical to that of the catalytic metal in the RNA ligase and NAD ϩ -dependent DNA ligase

DNA ligase mechanism
structures, the key difference being that the AMP phosphate in the PfuLig⅐AMP complex is not part of the metal coordination complex (its place being taken by a sixth water molecule).

Recombinant MtuLigD-LIG-K481M
The ORF encoding the LigD-LIG domain (aa 452-759) was PCR-amplified from M. tuberculosis genomic DNA with primers that introduced a BamHI site 5Ј of the start codon and a XhoI site 3Ј of the stop codon. The BamHI-XhoI fragment was inserted into pET28b-His 10 Smt3 to generate pET28b-His 10 Smt3-MtuLigD-LIG plasmid. The resulting expression plasmid encodes the MtuLigD-LIG polypeptide fused to an N-terminal His 10 Smt3 tag under the transcriptional control of a T7 RNA polymerase promoter. A K481M coding change was introduced into MtuLigD-LIG by PCR using mutagenic primers. The expression plasmid inserts were sequenced completely to exclude the acquisition of unwanted changes during PCR amplification and cloning. The pET28b-His 10 Smt3-MtuLigD-LIG -K481M plasmid was introduced into E. coli BL21(DE3) cells. A 2-liter culture was grown at 37°C in Luria-Bertani medium containing 0.1 mg/ml kanamycin until the A 600 reached 0.6. The culture was then adjusted to 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside, followed by incubation for 17 h at 17°C with constant shaking. The cells were harvest by centrifugation and stored at Ϫ80°C. All subsequent procedures were performed at 4°C. The bacteria were suspended in 50 ml of buffer containing 50 mM Tris-HCl, pH 8.0, 250 mM NaCl, 10% sucrose and then lysed by sonication for 10 min. Insoluble material was removed by centrifugation for 1 h at 16,000 rpm. The supernatant was applied to a 10-ml nickel-nitrilotriacetic acid-agarose column (Qiagen) that had been equilibrated with buffer A (50 mM Tris-HCl, pH 8.0, 250 mM NaCl, 10% glycerol). The column was washed with buffer A, and bound material was eluted stepwise with 20, 100, 300, and 1000 mM imidazole in buffer A. The polypeptide compositions of the eluate fractions were monitored by SDS-PAGE. The His 10 Smt3-MtuLigD-LIG-K481M protein was recovered predominantly in the 300 mM imidazole eluate fractions, which were pooled, supplemented with Smt3-specific protease Ulp1 (at a MtuLigD-LIG: Ulp1 ratio of ϳ500:1), and then dialyzed overnight against buffer A with 20 mM imidazole. The LigD-LIG-K481M protein was separated from the cleaved His 10 Smt3 tag by passage over a second nickel-agarose column equilibrated in buffer A with 20 mM imidazole, whereby the tag-free LigD-LIG was recovered in the flow-through fraction. LigD-LIG-K481M was then concentrated by centrifugal ultrafiltration and further purified by gel filtration through a column of Superdex-200 equilibrated in 10 mM Tris HCl, pH 8.0, 100 mM NaCl. The peak fractions of LigD-LIG-K481M (which eluted as a monomer) were pooled, concentrated by centrifugal ultrafiltration, and stored at Ϫ80°C.

Crystallization of LIG-K481M
A solution of LIG-K481M (10 mg/ml) was adjusted to 1 mM ATP and 5 mM MgCl 2 and incubated for 15 min on ice before aliquots (1 l) were mixed on a coverslip with an equal volume of precipitant solution containing 0.1 M MES, pH 6.5, 25% PEG 10000. Crystals were grown at room temperature by hanging-drop vapor diffusion against a reservoir of the same precipitant solution. Single crystals appearing after 2-3 days were harvested; cryoprotected by transfer to 0.1 M MES, pH 6.5, 1 mM ATP, 5 mM MgCl 2 , 25% PEG 10000, 15% PEG 400; and then flash-frozen in liquid nitrogen.
Alternatively, a solution of LIG-K481M (10 mg/ml) was adjusted to 1 mM ATP and 20 mM MgCl 2 and incubated for 15 min on ice before aliquots (1 l) were mixed on a coverslip with an equal volume of precipitant solution containing 0.1 M ammonium fluoride, 20% PEG 3350. Crystals were grown at room temperature by hanging-drop vapor diffusion against a reservoir of the same precipitant solution. Single crystals appearing after 2-3 days were harvested; cryoprotected by transfer to 0.1 M ammonium fluoride, 20% PEG 3350, 1 mM ATP, 20 mM MgCl 2 , 15% glycerol; and then flash-frozen in liquid nitrogen.

Diffraction data collection and structure determination
X-ray diffraction data were collected from single crystals at the Advanced Photon Source Beamline 24ID-C. Indexing and merging of the diffraction data were performed in HKL2000 (41). The phases were obtained by molecular replacement in MOLREP (42) using the NTase and OB domains of PDB entry 1VS0 separately as the search models. Interactive model building was performed in O (43). Refinement was accomplished with PHENIX (28). Data collection and refinement statistics are summarized in Table 1. The model of LIG-K481M⅐ATP⅐MES was refined to 1.4 Å resolution (R work /R free ϭ 0.188/0.198) and the model of LIG-K481M⅐ATP⅐Mg to 1.8 Å resolution (R work / R free ϭ 0.206/0.244). Both models comprised one LIG protomer in the asymmetric unit.

Accession numbers
Structure coordinates have been deposited in the Protein Data Bank database under accession codes 6NHX and 6NHZ.

Structure-guided mutagenesis
Alanine mutations were introduced into the LIG ORF of pET28b-His 10 Smt3-MtuLigD-LIG by PCR using mutagenic primers. The plasmid inserts were sequenced completely to exclude the acquisition of unwanted changes during PCR amplification and cloning. WT LigD-LIG and mutant LigD-LIG-Ala proteins were produced in E. coli and purified through the second nickel-agarose step as described above for LigD-LIG-K431M. Protein concentrations were determined by using the Bio-Rad dye reagent with BSA as the standard.