An Enzyme-Catalyzed Multistep DNA Refolding Mechanism in Hairpin Telomere Formation

Crystal structures reveal catalysis of DNA refolding in the molecular mechanism underlying generation of bacterial hairpin telomeres.


Introduction
Telomeres at the termini of linear chromosomes protect the DNA ends from degradation or aberrant repair reactions including end-fusion, while allowing complete replication of the terminal sequences [1]. The simplest form of telomere is a covalently closed hairpin loop found in bacteria carrying linear chromosomes, including Borrelia spirochetes-the causative agents of Lyme disease and relapsing fever [2,3], Agrobacterium [4,5], and cyanobacteria. Replication of such bacterial linear chromosomes with hairpin telomeres starts from an internal origin and proceeds bi-directionally, yielding a circular intermediate composed of a head-to-head, tail-to-tail dimer of chromosomes ( Figure 1A) [6,7]. The circular dimeric chromosome is then resolved at the two inverted-repeat junctions, formed as replication traverses through the telomeres, by a dedicated DNA cleavage-rejoining enzyme called protelomerase (also known as telomere resolvase) to regenerate unit-length chromosomes with hairpin telomeres [8][9][10][11].
The protelomerase enzymes function as a dimer, making staggered cleavages of both strands of DNA to form covalent 39phosphotyrosine linkages, then exchanging the released 59-ends while holding onto the 39-ends, and finally sealing the broken DNA strands to generate two hairpin ends ( Figure 1B) [12,13]. The phosphotyrosine-mediated DNA cleavage-rejoining reaction is chemically isoenergetic with no intrinsic directional bias, in a similar fashion to the reactions catalyzed by topoisomerases [14] and tyrosine-recombinases [15] that interconvert between linear double-helical DNA substrates and products. However, protelomerase is unique in that it converts the canonical duplex conformation of DNA into strained hairpin structures. Although the crystal structure of a bacteriophage-derived protelomerase TelK bound to a linear duplex substrate DNA was reported ( Figure S3, top panel) [16], the TelK-DNA complex structure provided little information regarding events following the strand cleavages or structure of the hairpin telomere products. Thus, mechanisms by which protelomerase drives the hairpin formation reaction forward without an exogenous input of energy or getting trapped in a DNA cleavage-religation equilibrium are not well understood.
We describe here a series of crystal structures of a bacterial protelomerase bound to reaction intermediates and hairpin products, which reveal that the enzyme dimer actively stabilizes both the tightly folded hairpin products and a transition state of DNA refolding pathway following the DNA strand cleavages. Furthermore, we show that the hairpin telomere formation by protelomerase is highly sequence-dependent, in line with the multistep strand-refolding mechanism suggested by the crystal structures. Thus, our collective results suggest that protelomerase catalyzes not only the chemical reactions of DNA strand cutting and rejoining but also the ordered transition between different DNA conformations to guide refolding of a DNA strand. Since DNA hairpins are formed as key intermediates in transposition [17][18][19][20] and the use of transposon-type motif by the Borrelia protelomerase/telomere resolvase ResT has been reported [21], our findings on protelomerase could provide insights into the mechanisms of transposon-related DNA remodeling reactions, including the V(D)J recombination responsible for diversifying antigen-receptor genes in higher vertebrates [22][23][24].

Results and Discussion
Overall Structure of the TelA-DNA Complex To better understand how protelomerase functions, we have crystallized the full-length protelomerase from the plant pathogen Agrobacterium tumefaciens C58 (TelA) in complex with DNA substrates containing the terminal sequences of the Agrobacterium linear chromosome. We used several types of DNA substrates ( Figure S1) resulting in crystal structures of TelA-DNA complexes that differ in the DNA conformation for the region ultimately forming the hairpin turn (shown by red letters in Figure S1). Phases for a parental TelA-DNA complex were obtained by the selenomethionine SAD phasing method, and all crystal structures have been refined to 2.2,2.4 Å resolution. Due to strongly anisotropic diffraction, the resolution limits for the refinement were set differently along the three principal axes (Table 1). In the crystals, the asymmetric unit contains a single TelA molecule bound to a hairpin telomere sequence, and the crystallographic dyad generates the TelA homodimer responsible for resolution of a replicated hairpin telomere ( Figure 2). Our models consist of residues 102 to 421 of the full-length 442 residues TelA protein bound to a half-site DNA substrate consisting of a 13 or 14 bp double-stranded stem and a 59-overhang ( Figure S1). The electron density for the N-terminal ,100 residues was very weak, likely reflecting high flexibility. This region is not required for in vitro hairpin formation by TelA [5].
The TelA monomer consists of two structural domains, the catalytic domain and an a-helical bundle domain, that together clamp down the DNA substrate ( Figure 2). While TelA shows significant sequence homology to the bacteriophage-derived protelomerase TelK [16,25] only in the ,190-residue region of the catalytic domain including the active site residues (,25% identity), the three-dimensional structure of TelA closely resembles that of the core part of the larger TelK protein ( Figure S2). Each of the two domains of TelA interacts with DNA in both the major and minor grooves, with a total of 43 residues positioned in close proximity to DNA (,3.6 Å ). Six base-pairs of the binding site are recognized by direct hydrogen bonding interactions, with additional base-pairs involved in water-mediated hydrogen bonds or van der Waals contacts (Figures 3, S4, and S5). The a-helical linker (Ala198-Gly217) connecting the two domains harbors several residues that make DNA contacts. Among these residues, Tyr201 and Arg205 play key roles in refolding the duplex DNA substrate into hairpin products as discussed later. The catalytic domains in the TelA dimer interact extensively with one another, burying 1,147 Å 2 of surface area per subunit. The two protein molecules in the TelA dimer bound to DNA are positioned such that there is a large (.10 Å ) offset in the DNA helical axes across the dimer interface, similar to the arrangement observed in the TelK-DNA complex ( Figures 2B and S3) [16].

Hairpin Product Complex
By crystallizing the wild-type TelA bound to a palindromic target DNA nicked at the scissile positions 6 bp apart (DNAa in Figure S1) in the presence of orthovanadate, we obtained a TelA dimer bound to the covalently closed hairpin products. The selfcomplementary six-base overhangs (T 1 C 2 A 3 T 4 G 5 A 6 ) with the native sequence of the Agrobacterium chromosome terminus form sharp hairpin turns in two alternating conformations, and are packed tightly between the two active sites in the TelA dimer ( Figure 4). The vanadate moiety links the 59-and 39-OH groups across the DNA nick as well as the Tyr405 nucleophile in the active site, mimicking the pentavalent transition state of the DNA ligation reaction ( Figure 5). The DNA strands take an extremely compact hairpin conformation in which all bases except two at the apex form the canonical Watson-Crick base-pairing. The two unpaired bases (Ade3 and Thy4) remain intrahelical and stack on the base-paired stem of the hairpin DNA, with rises per base steps comparable to that in the B-form DNA (,3.4 Å ) ( Figure 4A,B).
The compact di-nucleotide hairpin structure is stabilized by TelA through a number of interactions. The guanidinium group of Arg205 makes a cation-p stacking interaction on the thymine base (Thy4) at the tip of the hairpin turn ( Figure 4A,E). The stacking interaction by Arg205 is made in trans-that is, the interdomain linker contributing Arg205 to cap a hairpin DNA product connects the catalytic and DNA-binding domains clamping down the other hairpin half-site in the TelA dimer. The a-helical interdomain linker segment completely blocks the helical path of each DNA half-site, suggesting that the linear duplex substrate DNA would have to be severely distorted to fit in the DNAbinding path of the TelA dimer. The backbone phosphate group of Cyt2, Ade3, and Ade6 in the hairpin turn each forms both direct and water-mediated hydrogen bonds/salt bridges with TelA ( Figure 3). Sequence-specific contacts are made by the carboxyl group of Glu400 that interacts with the unpaired Ade3 base in both of its two alternate sidechain conformers ( Figure 4A), Lys208 that hydrogen bonds with the O4 atom of Thy1 in the major

Author Summary
Linear chromosomes capped by hairpin telomeres are widespread in prokaryotes and are found in important bacterial pathogens. However, three-dimensional structure of the hairpin telomere, as well as the molecular mechanisms underlying its generation, has remained poorly understood. In this work, we investigated how the enzyme responsible for generating the bacterial hairpin telomeres (protelomerase, also known as telomere resolvase) transforms a linear double-stranded DNA molecule into sharp hairpin turns. Our X-ray crystallographic and biochemical data collectively suggest that protelomerase employs a multistep DNA strand-refolding mechanism as described below. Protelomerase first cleaves both strands of a double-helical DNA substrate and reshapes the DNA strands into a transition state conformation (refolding intermediate) stabilized by specific protein-DNA and DNA-DNA interactions including noncanonical (non-Watson-Crick) base-pairs. The DNA strands are then refolded into extremely compact hairpin products, stabilized by a set of interactions distinct from those stabilizing the refolding intermediate. We believe that an enzyme ''catalyzing'' not only the chemical reactions of DNA strand cutting/rejoining but also the ordered transition between different DNA conformations to guide refolding of the DNA strand is a novel concept, and we suspect that similar mechanisms may be employed by other enzymes involved in conformational changes/refolding of biological macromolecules.
groove, and Lys288 that interacts with the O2 atoms of Thy1 and Cyt2 in the minor groove ( Figure 6).
Refinement of the atomic model of the TelA-hairpin DNAvanadate complex at 2.2 Å resolution allowed two alternate conformations of DNA strand in the hairpin turn to be resolved (indicated by the red arrows in Figure 4A and 4C). The backbone phosphate group of Gua5 at the apex of the hairpin turn as well as the flanking bases Thy4 and Gua5 take two distinct positions with approximately equal partial occupancies. The terminal base Thy4 is stacked more closely with Arg205 in trans in one conformer, while the backbone phosphate of Gua5 in the other conformer forms a salt bridge with the same Arg205 in cis. Thus, Arg205 plays dual roles in stabilizing the tight hairpin turn bound to the TelA dimer. For steric reasons, the two juxtaposed hairpin ends in a TelA dimer must be in alternative conformations, which introduces asymmetry into the otherwise 2-fold symmetric TelA-DNA complex ( Figure 4E). The two hairpin ends have backbone phosphorus atoms positioned only 5.9 Å apart across the 2-fold axis (intrastrand phosphorus distance in the B-form DNA is ,6.8 Å ), highlighting the tight packing of hairpin turns in the TelA dimer.
Having revealed the structure of the hairpin telomere, we then asked whether stabilization of the extremely compact DNA hairpin conformation by TelA is dependent on covalent closure of the DNA strand. To address this question, we used TelA with a point mutation in one of the active site residues Arg255 that plays an essential role in coordinating the scissile phosphate ( Figure 4C and Figure 5). TelANR255A was crystallized in complex with the palindromic target DNA sequence nicked at the scissile positions (DNAa in Figure S1). The resulting 2.3 Å resolution crystal structure of the TelANR255A-DNA complex shows the exact same hairpin DNA conformation as observed in the wild-type TelAhairpin DNA-vanadate complex described above (r.m.s.d. of 0.41 Å for the six nucleotides in the hairpin), except that the scissile phosphate-binding site is empty and nothing bridges the 59 and 39-OH groups of DNA ( Figure 4D). As the DNA hairpin region is free from crystal lattice contacts ( Figure S6), the result suggests that the hairpin conformation observed in our crystal structures is the most thermodynamically preferred conformation of the telomere DNA sequence when bound to TelA, even in the absence of a covalent phosphodiester linkage. This argues against a model in which strand ligation captures the hairpin conformation of an otherwise flexible DNA strand during hairpin telomere formation by TelA.

Refolding Intermediate
The crystal structures of the hairpin-DNA-TelA complexes suggest that a linear duplex substrate bound to the TelA dimer  Table 1. X-ray data collection, phasing, and refinement statistics.  would be in a severely distorted conformation ( Figure S3), as observed for the structure of TelK bound to a duplex substrate DNA [16], and is thus transformed into the thermodynamically more favorable hairpin form once the DNA strands are cleaved. Despite the overall favorable reaction energetics, refolding of a duplex substrate into two hairpin products within the TelA dimer may not readily proceed due to steric and/or electrostatic interferences. To gain insights into how the 59-ends are exchanged, refolded, and packed tightly within the partially buried interior of the stable TelA dimer, we determined crystal structures of the covalent phosphotyrosine TelA-DNA intermediate trapped using suicide DNA substrates. The first type of suicide DNA substrate used in our studies has nicks one base 39 to the scissile positions (DNAb or DNAc in Figure S1) [26]. Strand cleavage by TelA releases a single nucleotide (Thy1) between the nick and the newly formed phosphotyrosine bond, removing the 59-OH group necessary for strand ligation. The resulting DNA has five-base 59-overhangs, one base short of the natural six-base 59-overhang. Crystal structure of this phosphotyrosine complex refined at 2.2 Å resolution ( Figure 7A-C) shows a unique ''open'' conformation of the DNA strand distinct from that observed for the compact hairpin telomere product ( Figure 7E). In this open conformation, the first base into the overhang region to be refolded (Ade6) completely swings out into an extrahelical position and stacks against the Tyr201 sidechain. The positioning of Tyr201 is determined by water-mediated hydrogen bonds with the main chain carbonyl group of Ile170, Thr174 sidechain, and a DNA backbone phosphate, whereas the swung-out conformation of Ade6 is stabilized by those involving Asp202 and Arg205 ( Figure 7A). The base plane of Ade6 has a ,45u tilt with respect to the base planes in the duplex region. The second base, Gua5, is also flipped out and is stacked on Ade6. The third base, Thy4, is positioned similarly to how it is positioned in the hairpin product structure ( Figure 7E) where the Arg205 guanidinium group makes a cation-p stacking on its nucleobase moiety, while the other face of the Thy4 base stacks on its symmetry-mate across the 2-fold axis of the TelA complex. The sugar moiety of Thy4 stacks on the Gua5 base. In addition to these stacking interactions, the positioning of Gua5 and Thy4 is further stabilized by G-T wobble base-pairs ( Figure 7A,C). The syn conformation of the Gua5 base is well supported by simulated annealing omit difference electron density ( Figure 7C). While clear electron density was observed for the bases Thy4, Gua5, and Ade6, the density was weaker and less continuous for the bases Cyt2 and Ade3, suggesting higher flexibility near the 59-terminus of the DNA strand ( Figure S7).
In the open DNA conformation observed in the phosphotyrosine complex, the trajectory of the linearly stacked Tyr201 sidechain, Ade6, Gua5, and Thy4 bases is completely blocked by the interdomain a-helix harboring Arg205, and the DNA backbone makes a sharp turn to reverse the chain direction ( Figure 7A,B and Figure S7). The irregular conformation of DNA with flipped-out bases, sharply bent backbone, and the flexible 59terminus suggests that it is an intermediate step in the DNA strand refolding pathway ( Figure 1B). Stabilization of the flipped-out bases Ade6 and Gua5 in extrahelical positions would help disrupt the original base-pairing in the duplex substrate, as well as clear space for exchanging the 59-ends. The path of the DNA strand is set by the stacking interaction made by Tyr201 that orients the first base Ade6 and the capping interaction made by Arg205 that shapes the nascent hairpin loop structure. Point mutants TelANY201A and TelANR205A are capable of cleaving DNA to form the phosphotyrosine complex, but they are completely inactive in producing hairpin products in vitro ( Figure 8A and Figure S8). This underscores the important roles of Tyr201 and Arg205 in refolding DNA into a hairpin telomere. Consistent with the critical role of Tyr201 in orienting the tri-nucleotide stretch Ade6, Gua5, and Thy4, a crystal structure of TelANY201A-DNA complex shows that Ade6 has swung out further toward the protein to partially occupy the space taken by Tyr201 in the wildtype enzyme, leaving other bases in the 59-overhang flexible ( Figure 8B). Similarly, a crystal structure of TelANR205A complexed with DNA carrying the natural self-complementary six-base 59-overhang (DNAb6 in Figure S1) shows an extended DNA conformation without hairpin loop formation or flipping of the Gua5 base into the syn conformation ( Figure 8B).
Interestingly, Tyr319 and Arg322 of Tn5 transposase are part of a conserved ''YREK'' sequence motif found in the IS4 family of transposases and are involved in stabilizing the hairpin DNA conformation [18]. However, the precise roles of the Tyr and Arg sidechains appear to be different between Tyr319/Arg322 of Tn5 and Tyr201/Arg205 of TelA. Tyr201 of TelA rather plays an analogous role to Trp298 of Tn5 that stacks against a flipped-out DNA base in an extrahelical position [18,27]. In any case, a critical difference between the transposon and the hairpin telomere systems is that while base-flipping occurs in the transposase-bound hairpin DNA [18,28], it is observed only in the refolding intermediate ( Figure 7A,B) and not in the hairpin telomere product (Figure 4A,B)   or DNAc in Figure S1). We reasoned that if it is indeed a transition state structure, the DNA strand could be trapped in the same conformation by introducing a mismatch in the natural palindromic six-base 59-overhang to inhibit formation of the fully basepaired hairpin loop. Thus a second type of suicide DNA substrate was designed, in which the six-base overhang has cytosine in place of Thy1 to block base-pairing (DNAd in Figure S1). The Thy to Cyt substitution should also prevent the major groove interaction by Lys208 (Figure 6). Wild-type TelA mixed with this mismatched DNA substrate was trapped in a phosphotyrosine complex as  expected. The crystal structure refined at 2.4 Å resolution shows that the six-base overhang adopts a conformation very similar to that observed for the nicked suicide substrate, with an r.m.s.d. of 0.78 Å for atoms in the tri-nucleotide stretch ( Figure 7D). We therefore concluded that the open DNA conformation observed in the trapped phosphotyrosine complexes represents a metastable state of the 59-overhang, which precedes stable capturing of the 59end into the fully base-paired hairpin form. We propose that this ''refolding intermediate'' conformation allows TelA to overcome steric and/or electrostatic interferences between the DNA strands during refolding of a duplex substrate into the compact hairpin products ( Figure 1B). The mechanism is reminiscent of how enzymes in general catalyze chemical reactions by stabilizing intermediates, lowering energy barriers along reaction pathways [29].

DNA Sequence Requirements
To further validate our model on the mechanism of hairpin telomere formation, we examined effects of sequence variation in the DNA substrate. Based on the observed sequence-specific interactions that stabilize the hairpin product or the refolding intermediate including the G-T wobble pair, one would expect a strict requirement in hairpin telomere formation for the specific 6 bp DNA sequence between the two scissile positions. We therefore tested in vitro hairpin formation by TelA on a series of duplex DNA substrates with various palindromic sequences in the central 6 bp ( Figure 8C). Most of the nonnatural target sequences yielded no hairpin products, including 59-TGATCA-39 that corresponds to a natural target site after swapping Gua5 and Cyt2 ( Figure 8C, Lanes 3 and 12). Nonetheless, these nonnative substrates were cleaved by TelA (unpublished data). The results are consistent with our model that the DNA strands are refolded into hairpins guided by a set of specific protein-DNA and DNA-DNA interactions. It would be fair to note, however, that there are nonnative sequences that still support in vitro hairpin formation (e.g., Figure 8C, Lane 6), and we found sequence changes at the tip of the hairpin turn (Ade3 and Thy4) to be more tolerated in general [5]. Alternative base-pairing schemes, a G-G base-pair for instance [30], may allow for formation of a refolding intermediate similar to that formed by the natural target sequence in such cases.
The fairly elaborate mechanism of strand refolding by TelA described in this work raises a question as to how well it is conserved among protelomerase enzymes from different organisms. The basic architecture of the protelomerase dimer with a sharp disjunction in the DNA-binding path has been observed for both TelA and the bacteriophage-derived TelK systems ( Figure  S3) [16] and is likely a common feature for this family of proteins. In addition, all known protelomerase enzymes make staggered DNA cleavages 6 bp apart [12,25,31], where the conserved spacing between the scissile phosphates may reflect similarities in general mechanisms of hairpin telomere formation. On the other hand, the amino acid residues of TelA outside its catalytic domain, including those important in DNA refolding (Tyr201 and Arg205), do not appear to be particularly well conserved among   protelomerase enzymes from different systems [32]. Moreover, DNA sequences of the hairpin telomeres from different organisms do not show strong similarities and each protelomerase system has unique target sequences. We have shown that a modified palindromic DNA substrate with 59-CGCGCG-39 between the scissile positions, as found in the native target sequence for TelK [25], is not processed into hairpin products by TelA ( Figure 8C, Lane 9). Similarly, TelK cleaves efficiently but does not produce hairpin products on a modified DNA substrate with 59-GTATAC-39, as found in the presumed target sequence for the protelomerase crystal structures is either missing or replaced by cytosine to block ligation and trap the phosphotyrosine bond. (C) Simulated annealing omit Fo-Fc electron density contoured at 3.0 s for the G-T wobble pair. (D) 59-overhang conformation observed in the phosphotyrosine complexes (the refolding intermediate) trapped using two different types of suicide DNA substrates. The structure obtained with the nicked suicide substrate (DNAb/DNAc) is shown in yellow, while that obtained with a mismatch-based suicide substrate (DNAd) is shown in purple. The tri-nucleotide stretch Thy4, Gua5, and Ade6 is superimposable with an r.m.s.d. of 0.78 Å . The deviation comes mostly from the phosphate backbone atoms. (E) Comparison of the DNA structures between the refolding intermediate (yellow) and the hairpin telomere product (alternate conformations shown in black and grey). The superposition is based on the double-stranded stem region of the DNAs. doi:10.1371/journal.pbio.1001472.g007 from phage VHML [16,33]. The strict but distinct sequence requirements imply that stabilization of DNA refolding intermediates may be a common strategy employed by protelomerase enzymes, while the detailed refolding mechanisms, including specific protein-DNA and DNA-DNA interactions, likely differ among different organisms.

Conclusion
Enzymes that rearrange/recombine DNA play important roles in diverse biological contexts [34]. A common strategy employed by these enzymes is to bind tightly to and stabilize the final products, thereby driving the reactions forward by virtue of DNAbinding free energies [35][36][37]. We have shown that this is indeed the case for the protelomerase TelA, a DNA rearrangement enzyme involved in the maintenance of Agrobacterium linear chromosome. Unexpectedly, however, we found that TelA facilitates the duplex-to-hairpin conversion by stabilizing not only the hairpin telomere product but also a transient strand-refolding intermediate to guide the DNA strand refolding process. We believe that the enzyme-catalyzed, multistep DNA refolding described in this study is a novel mechanism, and we suspect that similar strategies may be employed by other protein machineries that facilitate conformational changes/refolding of DNA or other macromolecules in various biological contexts.

Crystallization and Structure Determination
The TelA protein and its mutants used in the present study have a 20-residue N-terminal His-tag attached to the full-length Agrobacterium tumefaciens protelomerase protein (gene locus: Atu_2523). The His-tagged TelA was overexpressed in E. coli strain BL21 under the control of the arabinose-inducible pBAD promoter and purified using the Ni-NTA and Heparin-Sepharose columns. Selenomethionine-labeled TelA was overexpressed in the methionine auxotroph strain B834(DE3). All oligonucleotides were purchased from IDT in the standard desalting grade and used without further purification. The sequences of the oligonucleotides used in the crystallization experiments are available in Figure S1. The target sequence for TelA was derived from either the left terminus (for DNAa, DNAb, DNAd, and DNAb6 in Figure S1) or the right terminus (for DNAc used to grow SeMet crystals) of the Agrobacterium tumefaciens C58 linear chromosome (Genbank #AE007870.2).
TelA-DNA complexes used in crystallization were prepared by mixing equal moles of protein monomer and DNA half-site at an approximate protein concentration of 0.2 mM and a NaCl concentration of ,0.5 M. The hairpin DNA-TelA-vanadate complex was assembled in the presence of 10 mM sodium orthovanadate using a DNA substrate containing the six-base TCATGA overhang (DNAa in Figure S1). The R255A TelAhairpin DNA complex was prepared using the same DNA in the absence of sodium orthovanadate. To prepare the phosphotyrosine complexes with cleaved DNA, the purified protein was mixed with either the CATGA (nicked: DNAb, DNAc in Figure  S1) or CCATGA (mismatched: DNAd in Figure S1) suicide DNA substrate. Crystals of the TelA-DNA complexes were obtained by the hanging drop vapor diffusion method at 20uC. The well solution consisted of 5% (w/v) PEG 4000, 10 mM Tris-HCl (pH 7.4), and 300 mM NaCl, and the hanging drops were formed by mixing the TelA-DNA complex with the well solution at a 1:1 volume ratio. All TelA-DNA complexes were crystallized in the same crystal form under similar conditions, though the time it took for the crystals to appear varied depending on the DNA substrates used. The crystals were cryoprotected by gradually introducing glycerol to a final concentration to 25%, then flash frozen in liquid nitrogen.
X-ray diffraction data were collected at the beamlines 14ID-B, 14BM-C, and 24ID-C of the Advanced Photon Source (Argonne, IL). The structure of the TelA phosphotyrosine complex formed with a 1 bp longer suicide DNA substrate than the other DNA substrates (15 mer+19 mer: DNAc in Figure S1) was determined by the single-wavelength anomalous dispersion (SAD) method using the selenomethionine drivative. Two single-path scan datasets were collected at the selenium K-edge, one at a lower resolution (3.0 Å ) and one to a higher resolution (2.4 Å ). Indexing, integration, and scaling of the collected diffraction frames were done using HKL2000 [38] or XDS [39]. Eleven selenium sites per TelA monomer were located with the lower resolution dataset (3.0 Å ) using SOLVE [40]. Resolution of the experimental phases was then improved by combining the two datasets, and density modification by RESOLVE generated an interpretable map. Automated model building by RESOLVE built ,50% of the protein residues. Iteration of phase-restrained refinement using REFMAC5 [41] and manual model building using COOT [42] eventually generated a model consisting of 320 amino acids covering the TelA residues 102,421, and all DNA residues. The 121 N-terminal amino acids including the 20-residue His-tag, and the 21 C-terminal amino acids are disordered. Structures of all other complexes were determined by molecular replacement using PHASER [43]. All crystallographic models were finally refined using PHENIX [44] with the TLS refinement. X-ray diffraction data, phasing, and model refinement statistics are summarized in Table 1. Severely anisotropic data were subjected to ellipsoidal truncation and anisotropic scaling [45] prior to structure refinements. Figures were produced using PYMOL (www.pymol. org). Buried protein surface area was calculated using CNS [46]. Curves [47] and 3DNA [48] were used for DNA geometry analyses.

In Vitro Hairpin Formation and DNA Cleavage Assays
A 50 bp region from Agrobacterium chromosome terminus centered on the target sequence of TelA, or its variant sequences, was cloned into the pSK plasmid to generate pAgSK54. pAgSK54 linearized with the restriction enzyme AlwNI was purified using a Qiagen spin-column and was employed as the substrate in the in vitro resolution assay. The resolution reaction was performed with 10 nM linearized plasmid DNA and 6.7 mM TelA at 22uC for 2 h in 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, and 1 mM DTT ( Figure 8A) or at 30uC for 30 min in 20 mM Tris-HCl (pH 7.5), 50 mM potassium glutamate, and 1 mM DTT ( Figure 8C). The reactions were quenched by addition of proteinase K or SDS. The DNA products extracted by phenol extraction were separated on 1% argarose gel and visualized by ethidium bromide staining. The phosphotyrosine complex formation of the wild-type and mutant TelA proteins ( Figure S8) was monitored on a 44 bp nicked suicide substrate assembled from 24 mer and 20 mer oligonucleotides containing the 59-CATG-39 overhang. Protein at 10 mM was incubated with 26 molar excess of the suicide substrate DNA in the same buffer condition used for the resolution reaction. The reaction was quenched at each time point (1 h, 2 h, 1 d, 2 d, 3 d, and 1 wk) by addition of SDS and the samples were analyzed by SDS-PAGE.

Accession Codes
The atomic coordinates and the structure factors for all crystal structures reported here have been deposited in the Protein Data Bank with accession codes 4E0G, 4E0J, 4E0P, 4DWP, 4E10, 4E0Z, and 4E0Y. Figure S1 DNA substrates used in the crystallographic studies. Each substrate was assembled from two oligonucleotides of different lengths to form a 59-overhang. The red letters correspond to the region between the two scissile phosphates. DNAc is based on the chromosome right terminal sequence, whereas all others are based on the left terminal sequence shown at the top. The differences between the left and right telomere sequences are limited to the positions highlighted by the green boxes. (TIF) Figure S2 Comparison between Agrobacterium tumefaciens TelA and the protelomerase TelK from bacteriophage WKO2. (Top) Superposition of the two crystal structures (DNA was omitted for clarity). TelK has a large insertion (muzzle) within the a-helical bundle domain opposite the catalytic domain and an additional DNA-binding domain in the C-terminus (stirrup) [16], both of which are absent from TelA. (Bottom) Amino acid sequence alignment for the region of the catalytic domain harboring the active site residues. TelA residues highlighted by the colored boxes are Arg255, Lys286, Tyr363, Arg366, His394, and Tyr405.