Conserved Structural Chemistry for Incision Activity in Structurally Non-homologous Apurinic/Apyrimidinic Endonuclease APE1 and Endonuclease IV DNA Repair Enzymes*

Background: DNA apurinic/apyrimidinic (AP) sites are toxic and mutagenic if unrepaired by AP endonucleases. Results: Structural, mutational, and computational analyses of prototypic AP endonucleases APE1 and Nfo identify surprising similarities. Conclusion: APE1 and Nfo reveal functional equivalences illuminating their catalytic reaction. Significance: A conserved catalytic geometry is specific to AP site removal despite different enzyme structures and metal ions. Non-coding apurinic/apyrimidinic (AP) sites in DNA form spontaneously and as DNA base excision repair intermediates are the most common toxic and mutagenic in vivo DNA lesion. For repair, AP sites must be processed by 5′ AP endonucleases in initial stages of base repair. Human APE1 and bacterial Nfo represent the two conserved 5′ AP endonuclease families in the biosphere; they both recognize AP sites and incise the phosphodiester backbone 5′ to the lesion, yet they lack similar structures and metal ion requirements. Here, we determined and analyzed crystal structures of a 2.4 Å resolution APE1-DNA product complex with Mg2+ and a 0.92 Å Nfo with three metal ions. Structural and biochemical comparisons of these two evolutionarily distinct enzymes characterize key APE1 catalytic residues that are potentially functionally similar to Nfo active site components, as further tested and supported by computational analyses. We observe a magnesium-water cluster in the APE1 active site, with only Glu-96 forming the direct protein coordination to the Mg2+. Despite differences in structure and metal requirements of APE1 and Nfo, comparison of their active site structures surprisingly reveals strong geometric conservation of the catalytic reaction, with APE1 catalytic side chains positioned analogously to Nfo metal positions, suggesting surprising functional equivalence between Nfo metal ions and APE1 residues. The finding that APE1 residues are positioned to substitute for Nfo metal ions is supported by the impact of mutations on activity. Collectively, the results illuminate the activities of residues, metal ions, and active site features for abasic site endonucleases.

Non-coding apurinic/apyrimidinic (AP) sites in DNA form spontaneously and as DNA base excision repair intermediates are the most common toxic and mutagenic in vivo DNA lesion. For repair, AP sites must be processed by 5 AP endonucleases in initial stages of base repair. Human APE1 and bacterial Nfo represent the two conserved 5 AP endonuclease families in the biosphere; they both recognize AP sites and incise the phosphodiester backbone 5 to the lesion, yet they lack similar structures and metal ion requirements. Here, we determined and analyzed crystal structures of a 2.4 Å resolution APE1-DNA product complex with Mg 2؉ and a 0.92 Å Nfo with three metal ions. Structural and biochemical comparisons of these two evolutionarily distinct enzymes characterize key APE1 catalytic residues that are potentially functionally similar to Nfo active site components, as further tested and supported by computational analyses. We observe a magnesium-water cluster in the APE1 active site, with only Glu-96 forming the direct protein coordination to the Mg 2؉ . Despite differences in structure and metal requirements of APE1 and Nfo, comparison of their active site structures surprisingly reveals strong geometric conservation of the catalytic reaction, with APE1 catalytic side chains positioned analogously to Nfo metal positions, suggesting surprising functional equivalence between Nfo metal ions and APE1 residues. The finding that APE1 residues are positioned to substitute for Nfo metal ions is supported by the impact of mutations on activity. Collectively, the results illuminate the activities of residues, metal ions, and active site features for abasic site endonucleases.
Apurinic/apyrimidinic (AP) 4 sites are the most common DNA lesions in vivo, calculated to be generated at ϳ10,000 lesions/cell/day in humans (1)(2)(3). AP sites form spontaneously or as central DNA repair intermediates during base excision repair (BER) (4). AP sites can block replication and cause mutations, so its repair is critical for genetic integrity (5,6). Repair of AP sites is carried out via the BER pathway, which creates AP sites by uracil and alkylated base-specific monofunctional DNA glycosylases that excise the base lesion to produce AP sites (7). For the glycosylases, structures were key to revealing that specificity is encoded in nucleotide flipping and binding, as shown by MutY and uracil-DNA glycosylase (8,9). The AP sites produced by excision of the damaged base are the substrates of AP endonucleases. For AP endonucleases, there are primarily two distinct families that incise the phosphodiester backbone 5Ј to the lesion: the bacterial endonuclease IV (Nfo) family and the exonuclease III (Xth) family, which includes Xth in bacteria and AP endonuclease (APE1) in metazoan eukaryotes (10). These pivotal nucleases detect, recognize, and cleave the DNA phosphodiester backbone 5Ј of AP sites to create a free 3Ј-OH end for repair synthesis by a DNA polymerase. The AP endonucleases also act as 3Ј 3 5Ј exonucleases and catalyze nucleotide incision repair of particular oxidized bases (11)(12)(13)(14)(15)(16). These two prototypic families mediate the same activities, but their structural folds and metal dependence are different. In initial studies of Nfo, it was differentiated from Xth by its resistance to EDTA (17,18). The Xth family has a two-layered ␤-sheet core flanked by helices and is Mg 2ϩ -dependent (19). In contrast, Nfo has a TIM ␤ barrel core, surrounded by helices. It has three metal ions, either three Zn 2ϩ or two Zn 2ϩ and one Mn 2ϩ (20). In general, understanding of distinct metal ion binding and activities, even in microbial systems, has lagged far behind genome sequencing, and an increased knowledge of structure-function relationships is fundamental to more accurate metal ion prediction for responses to stress and DNA damage (21)(22)(23)(24)(25). Nfo is unusual among endonucleases in that it uses Zn 2ϩ ions and in that it uses three metals in its catalytic mechanism (26,27). Three-metal mechanisms have also been proposed for E. coli RV, RNase H, and microbial FEN1 (28 -30), but in Nfo, all three metals occupy the active site at the same time. Recent biochemical, structural, and molecular dynamics (MD) studies have defined the individual roles of each Zn 2ϩ : Zn1, Zn2, and Zn3 ( Fig. 1) (20,27,31). The initial AP site is flipped into the active site and is bound by Zn1 and Zn3. The attacking water is deprotonated by Glu-261, the one side chain directly involved in the catalytic mechanism. The resulting hydroxide ion is electrostatically stabilized by Zn1 and Zn2. All three Zn 2ϩ ions stabilize the pentacoordinated transition state. At the end, Zn3 moves to coordinate the phosphate oxygen (OPЈ) and help stabilize the developing negative charge of the leaving group. Interestingly, crystallographic and biochemical studies suggest that the Zn3 position may actually be occupied by Mn 2ϩ in Escherichia coli, suggesting further study.
Human APE1 (also called HAP1 and REF1) consists of a core nuclease domain that is conserved with E. coli Xth and a 61-residue N-terminal domain that is not conserved in bacterial proteins (32,33). Crystal structures have shown that APE1 binds to both major and minor grooves of the DNA and flips out the abasic deoxyribose phosphate (34 -37). In the original report of an APE1-DNA complex, we proposed a mechanism where Asp-210 activates the attacking water (35). The phosphate intermediate is stabilized by the Mg 2ϩ ion and contacts with His-309, Asn-174, and Asn-212. Protonation of the 3Ј-ribose oxygen leaving group is through water in the first hydration shell of the Mg 2ϩ (35), yet several subsequent papers with studies of active site mutants have proposed alternative enzyme mechanisms (38 -43). These experiments were done on different mutants in different laboratories using different techniques, so no clear consensus has been reached. Even the number of metal sites participating in catalysis is in question (44), which has hampered progress.
To help resolve these questions, we solved, analyzed, and compared new crystal structures of APE1 and Nfo. A 2.4 Å resolution crystal of wild type (WT) human APE1 in a product complex with Mg 2ϩ revealed a Mg 2ϩ -water cluster in the active site. We determined an extremely high resolution (0.92 Å) crystallographic structure for Thermotoga maritima Nfo with bound metal ions. Prompted by a common substrate, we superimposed tertiary structures of APE1 and Nfo and surprisingly observed that the scissile bond can be superimposed despite differences in protein fold, metal type, and number of metals. In fact, APE1 active site residues overlay onto metal positions in Nfo. Importantly, the geometric restraints implied by the Nfo superposition are consistent with only one of multiple proposed mechanisms. Our combined structural, biochemical, and computational analyses thus help to resolve mechanistic questions and support a unified excision geometry and mechanism for the two prototypic AP endonucleases despite their structural differences.
APE1 Crystallization and Data Analysis-WT APE1 (12 mg/ml) was incubated with 11-mer double-stranded DNA at a molar ratio of 1:1.2 for 10 min. The DNA contained a central tetrahydrofuran on one strand as described previously (35) and was purchased from Midland Inc. The protein and DNA were mixed 1:1 with 50 mM MES, pH 6.0, 200 mM LiSO 4 , and 25% Polyethylene glycol monomethyl ether 2,000. Crystallographic data were collected at Advanced Light Source beamline 5.0.1. Diffraction was observed to 2.1 Å, but resolution was truncated to 2.4 Å resolution due to high anisotropy and overlaps. Phases were determined by molecular replacement. Refinement was performed by CNS (48) and later by PHENIX (49). NCS restraints were combined with TLS during refinement in PHE-NIX. There were four molecules in the asymmetric unit. The A and C chains had the most well defined electron density, with both having B factor averages for all protein atoms of 42 Å 2 . B and D chains had significantly worse density, with average B factors of 73 and 58 Å 2 , respectively. The N terminus (residues 1-40) was evidently flexible because it lacked unambiguous electron density. Clear octahedral geometry for the Mg 2ϩ and coordinated waters were observable in the electron density for A, C, and D. Three other Mg 2ϩ sites based on coordination geometry and distances were assigned in density near A and C chains. However, coordination was to waters and to two bases in the DNA, and these other Mg 2ϩ ions are unlikely to affect the catalytic mechanism. The PDB code for APE1-product complex is 4IEM.
Nfo Crystallization and Data Analysis-T. maritima Nfo (10 mg/ml) was crystallized in hanging drops mixed 1:1 with 100 mM Tris-HCl, pH 9.0, 4 mM DTT, 1.5% saturated MgSO 4 , 16% ethylene glycol, 20% Polyethylene glycol monomethyl ether 2,000. Crystals were frozen with 25% ethylene glycol. Crystallographic data were collected at Stanford Synchrotron Radiation Laboratory beamline 9-1. Diffraction data were collected to 0.92 Å resolution. Refinement was consistent with a mixture of Zn 2ϩ and Mn 2ϩ at metal sites 2 and 3. Tests to identify manganese in the active site were not attempted at the time because occupancy of manganese was unexpected. Alternate metals could not be assigned the same number, and thus the metals have been renumbered consecutively. Phases were determined by molecular replacement using E. coli Nfo (PDB code 1QTW) as a search model with the program AMORE (50) and then refined using an improved model and the program EPMR (51) with resolution limits of 8 to 4.5 Å. The solution had an R factor of 0.317 and correlation coefficient (CC) of 0.725 as compared with the next solution, with an R factor of 0.52 and correlation coefficient of 0.241. Refinement was performed initially with CNS (48) and later with PHENIX (49). The PDB code for the T. maritima Nfo is 4HNO.
Single-turnover Kinetics-Specific endonuclease activity of APE1 WT and various APE1 mutants was analyzed by singleturnover kinetics under conditions of excess enzyme and limiting substrate, a 52-nt tetrahydrofuran-containing oligonucleotide duplex (position 30 nt) (52)(53)(54). The reaction mixture (100 l) contained 10 nM duplex oligonucleotide containing THF and 100 nM APE1. The 32 P-labeled oligonucleotide was diluted with unlabeled oligonucleotide to maximize the detection range. The reaction was performed at 10°C under final buffer conditions of 50 mM Tris-HCl, pH 8.0, 50 mM KCl, and 2 mM MgCl 2 . The reaction was initiated by adding enzyme to the reaction mixture, and 10-l aliquots were removed at 10 s, 20 s, 30 s, 1 min, 2 min, 5 min, 10 min, 30 min, and 60 min into an equal volume of 90% formamide denaturating loading dye containing 50 mM EDTA. The samples were heat-denatured at 94°C for 3 min. Product (30 nt) was separated from substrate (52 nt) by 20% acrylamide, 7 M urea gel. The radioactivity in these bands was quantitated in a PhosphorImager (Molecular Dynamics) using ImageQuant software.
MD of APE1 and Mutants-Models for the reactant APE1abasic DNA complex and select mutants (E96A, D210N, N212A, H309A, H309N, and Y171F) were based on the APE1product structure reported here and set up for classical MD using the AMBER PARM99SB force field with modified nucleic acid parameters (BSC0) (55,56) and TIP3P solvent (57). Na ϩ and Cl Ϫ ions were used for charge neutralization and to achieve a salt concentration of 0.1 M. The protonation states for histidine residues were determined by the WHATIF server. The catalytically important His-309 residue was set as protonated. After minimization (10,000 steps) and equilibration (4-ns dynamics with gradual scaling of positional restraints), we carried out fully unconstrained production runs for WT APE1 for 15 ns. The mutant simulation trajectories were of the same length and employed the same simulation protocol as the WT APE1 except that the mutated residue was allowed to move freely and equilibrate before releasing restraints on the other active site residues. All simulations were performed in the isothermal isobaric ensemble (NPT) at 1 atm and 300 K with the program NAMD 2.8 (58). An integration time step was used under a multiple time stepping scheme. The bonded and short range interactions were calculated every third step. A short range cut-off of 10 Å was used for the short range non-bonded interactions with a switching function at 8.5 Å. The long range electrostatic interactions were treated with a smooth particle mesh Ewald method (58). The r-RESPA multiple time step method (59) was adopted with a 2-fs time step for bonded interactions, 2 fs for short range non-bonded interactions, and 4 fs for long range electrostatic interactions. We used 2-fs time steps, keeping all bonds between hydrogen and heavy atoms constrained. We visualized trajectories and computed average properties with VMD (60). Structure figures were produced using PyMOL (Schrödinger, LLC, New York) and VMD (60).

RESULTS
Crystal Structure of Human WT APE1 Bound to DNA-We crystallized full-length protein with Mg 2ϩ and dsDNA containing a tetrahydrofuran. The structure was determined to 2.4 Å with an R factor and R free of 0.25 and 0.19, respectively (Table 1). This product structure is the highest resolution APE1-DNA complex known and a significant improvement compared with the published 3.0 Å Mn 2ϩ -product complex (35). As before, there is only one metal ion observed in the active site in the asymmetric unit. The Mg 2ϩ ion was identified by its coordination geometry and its distance to coordinating atoms ( Fig. 1) (61).
In the three well defined active sites, only one residue, Glu-96, directly coordinates the Mg 2ϩ ion. Mutation of Glu-96 reduces endonucleolytic activity 600-fold (39,62). The 3Ј-ribose oxygen and the phosphate from the DNA and waters complete the tetrahedral coordination. Three waters completed the coordination. Asn-68, Asp-70, and Asp-308 coordinate the waters in the magnesium-water cluster. Mutation of these residues reduces endonucleolytic activity: Asn-68, 200-fold (36); Asp-308, 5-fold (63,64); and Asp-70, 25-fold (39). Asp-70 is particularly intriguing because mutation enhances 3Ј-phosphodiesterase activity but reduces AP endonuclease activity (65). For both activities, maximum activity requires a higher Mg 2ϩ concentration, consistent with a role of Asp-70 in helping to coordinate the Mg 2ϩ ion through the water. Other active site waters act in indirect DNA interactions; e.g. Lys-98 has a water-mediated interaction with the nucleotide adjacent to the 3Ј-ribose oxygen. The position of the APE1 Mg 2ϩ ion, between the 5Ј-phosphate and the 3Ј-hydroxyl, is on the other side of the phosphate compared with many one-metal nucleases (66).
Ultrahigh Resolution Crystal Structure of T. maritima Nfo-To obtain a greater understanding of the multiple-metal ion Nfo catalytic mechanism in comparison with the single-metal ion seen in our APE1 structures, we crystallized a hyperthermophilic ortholog from T. maritima. The optimal growth of this thermophile is ϳ80°C (67). T. maritima Nfo has 33% identity with the E. coli enzyme, suggesting overall similarity. The protein structure was solved to 0.92 Å with an R factor and R free of 0.12 and 0.14, respectively ( Fig. 1 and Table 1). During our study, a structure of T. maritima Nfo with Zn 2ϩ and/or Cd 2ϩ at a lower resolution, 2.3 Å, was reported (68).
As in E. coli Nfo, we found three metals in the active site, with a root mean square deviation of 0.2 Å from E. coli Nfo positions (20,27). Previously, Mn 2ϩ had been found in the Zn3 ion position in E. coli Nfo (27). The coordination is also more octahedral in geometry, and Nfo with Mn 2ϩ is more active than with only Zn 2ϩ . In our 0.92 Å structure, we could not unambiguously assign the metal 2 and 3 positions as Zn 2ϩ and Mn 2ϩ because the metal-ligand distances and the geometry were not definitive (61). However during refinement, we did observe negative F o Ϫ F c difference density when Zn 2ϩ was in metal 2 or 3 position and positive F o Ϫ F c when Mn 2ϩ was in those posi-tions, suggesting either low occupancy of Zn 2ϩ or a mixture of Mn 2ϩ and Zn 2ϩ . In the Cd 2ϩ T. maritima Nfo structure, it was those two positions that were occupied by Cd 2ϩ (68). Refinement with both Mn 2ϩ and Zn 2ϩ resulted in the lowest R factor values and smallest difference between R work and R free , as compared with other refinement scenarios.
The catalytic Glu-261 and residues coordinating the three Zn 2ϩ ions were absolutely conserved with E. coli Nfo. There was one residue different in the active site pocket. Ala-30 in E. coli Nfo is replaced with Gln-30 in T. maritima Nfo. T. maritima Nfo Gln-30 appeared to exclude two water molecules. We postulate that this substitution is important for T. maritima Nfo to have optimal endonucleolytic catalysis at 80°C. To test this hypothesis, we compared the sequences from thermophilic and mesophic Nfo orthologs (Fig. 2). Although residue 30 was either alanine or glutamine in mesophilic Nfo, only glutamine was found at this position in thermophilic Nfo.
Superimposition of APE1 and Nfo Active Site Structures-The mechanism for Nfo has been comprehensively studied biochemically, structurally, and computationally by MD (27,31). To better understand APE1, we compared the APE1 and Nfo structures. Examination of the DNA binding grooves in terms of the local accessibility and for fractal dimension supported the overall similarity of the active sites (70,71) (Fig. 3). A shallow groove near the active site sterically imposes specificity for DNA lacking a bulky base. Notably, there is a significant pocket next to the active site that is present in both Nfo and APE1. We postulated that this pocket may allow the nucleotide incision repair activity and the requisite space for binding a base, albeit a non-canonical base, but simple modeling with a nucleotide incision repair substrate, an ␣-anomeric adenosine, did not position the base in that pocket (72). Perhaps this pocket has convergently evolved for a yet unrecognized functionality.
Because Nfo and APE1 are structurally dissimilar, we superimposed the tetrahydrofuran moiety of our 2.4 Å APE1-Mg 2ϩproduct complex with that from the reported E. coli structure of Nfo-Zn 2ϩ -product complex (PDB code 1QUM) (20). A superposition based on only the tetrahydrofuran moiety in the DNA oriented the respective scissile 5Ј-phosphates and 3Ј-ribose oxygen atoms to overlay on top of each other, supporting the hypothesis that the two enzymes have a similar mechanism (Fig. 4A). The root mean square deviation of the tetrahydrofuran atoms was 0.28 Å. The phosphates and ribose oxygens were 0.1 and 0.9 Å apart, respectively.
To better view any catalytic similarity of the active sites, we superimposed the scissile phosphates and ribose oxygens (Fig.  4, B and C). Because E. coli Nfo has three metal atoms with distinct catalytic roles (31), we were interested in which metal atom geometrically matched the Mg 2ϩ atom in APE1. Superimposing the scissile phosphate and the 3Ј-ribose oxygens from the two structures, the Zn 2ϩ atoms fall in three quadrants around the phosphate, with the Mg 2ϩ in the fourth quadrant R sym , the unweighted R value on I between symmetry mates. b R cryst ϭ ⌺ hkl ʈF o (hkl)͉ Ϫ ͉F c (hkl)ʈ/⌺ hkl ͉F o (hkl)͉. c R free , the cross-validation R factor for 5% of reflections against which the model was not refined. (Fig. 4C). Surprisingly, no one Zn 2ϩ ion overlaid the Mg 2ϩ ion. Notably, one Zn 2ϩ atom (Zn1) is ϳ1.3 Å from the imidizole moiety of His-309 in APE1. Mutation of His-309 reduces APE1 activity over 30,000-fold (73). Another Zn 2ϩ atom (Zn2) is located in between the side chains of Asn-212 and Asp-210. Zn2 also localizes ϳ1 Å from a second Mg 2ϩ site proposed for APE1 (44). The side chain of Asn-212 (APE1) is positioned similarly to Asp-261, which directly coordinates the attacking water in Nfo. Notably, the N212A APE1 mutant lacked detectable AP endonuclease activity (74).
The third Zn 2ϩ atom (Zn3) in Nfo is positioned close to APE1 Tyr-171. Zn3 is postulated to act in stabilizing the pentacoordinate intermediate. Mutation of APE1 Tyr-171 reduces AP catalytic activity more than 25,000-fold (42). Notably, because Zn3 is in a mirror position to the Mg 2ϩ relative to the axis of the reaction, replacement of Zn3 with Mn 2ϩ increases Nfo activity (27). No side chains from Nfo overlay in this metal position.
The one protein residue that contributes to the catalytic reaction in Nfo, Glu-261, superimposes to lie close to Asn-212 in APE1. Two Nfo residues that coordinated zinc atoms, Glu-145 and His-109, were 0.8 and 1.5 Å from Asp-210 and His-309 positions in APE1, respectively, the latter raising the possibility of the moving metal ion site. However, the measured pK a of over 8 for His-309 is inconsistent with the need for a non-protonated imidizole necessary for metal ion coordination (75). His-109 in Nfo is not in a His-Asp pair, as is His-309 in APE1, which has Asp-308 on one side and Asp-283 on the other.
The superposition of the DNA substrates and overlay with Nfo catalytic metals highlighted the significance of four APE1 residues, Asp-210, Asn-212, His-309, and Tyr-171. From this structural comparison, it appears that APE1 residues may substitute catalytically for Nfo metals, suggesting that mutational and computation testing was merited.
Single Turnover Kinetics on Active Site Mutants-To better compare the implicated catalytic residues that have been mutated and studied in separate studies with different methods, we did single turnover kinetic studies of single site mutations on the key catalytic residues Glu-96, Tyr-171, Asp-210, Asn-212,  and His-309 (Fig. 5). Interestingly, we found that mutation of Glu-96, the only residue that directly coordinates the Mg in the product structure, had the least effect, with activity down only by 15-fold.
In contrast, Y171F and H309N showed quite large decreases in activity of ϳ1200and 2500-fold, respectively. However, Tyr-171 and His-309 were not the most important catalytic residues because their mutation was not as severe as those of Asp-210 and Asn-212. Mutants D210N and N212A were decreased 10,000-and 7000-fold in activity, respectively. Notably, Asp-210 and Asn-212 are located similarly to the two Zn 2ϩ atoms (Zn1 and Zn2) found to coordinate the catalytic water in Nfo; they are positioned to coordinate a water that could make a linear attack on the phosphodiester. To test our model computationally and provide an informed basis as to why specific mutations are reducing catalytic activity, we did MD simulations of WT and mutant APE1: E96A ,  Y171P, H309A, N212A, D210N, and H309N. Model coordi-  nates are part of the Supplemental Materials. In WT, we observed the catalytic water tightly hydrogen-bonded to both Asp-210 and Asn-212 with average hydrogen bond distances of 2.65 Ϯ 0.15 and 2.83 Ϯ 0.10 Å, respectively (Fig. 6). Interactions with these two residues position the oxygen atom of the catalytic water within 3.51 Ϯ 0.22 Å of the scissile phosphate and orient the water for an inline attack (average O*-P-OЈ angle of 162.6 Ϯ 8.2°). Notably, the hydrogen bonding network (Fig. 6) between the catalytic water, Asp-210, Asn-212, His-309, and the abasic site phosphate is persistent throughout the MD tra-jectory. Indicating a less important catalytic role, mutations E96A, Y171P, and H309A did not affect the hydrogen-bonded triplet of Asp-210, Asn-212, and catalytic water, and the nucleophilic attack geometry is maintained (respective average O*-P distances of 3.52 Ϯ 0.25, 3.51 Ϯ 0.27, and 4.24 Ϯ 0.69 Å). In the N212A mutant, the catalytic water is displaced from the inline position, and Asp-210 instead accepts hydrogen bonds from Asn-68 and Ala-212 (backbone). In the D210N mutant, the catalytic water does not move. However, the mutated Asn-210 side chain cannot accept a proton to activate the water. The ability of both D210N and N212A mutants to disrupt the attacking water correlates with the experimentally observed rates.
Consistent with the new APE1 crystal structure, our MD simulation of WT reveals protonated His-309 stably hydrogenbonded to the abasic site phosphate (with average hydrogen bond distance of 2.84 Ϯ 0.13 Å and an almost linear angle of 8.9 Ϯ 4.7°). Experimentally, mutating His-309 has a large effect on the catalytic rate. We attribute this to electrostatic polarization and charge transfer from the positively charged His-309, which helps stabilize the developing negative charge on the phosphate moiety in the transition state. The H309N and H309A mutants would be much less effective in this role. Both mutants in MD are shown to create a water-filled cavity due to the smaller size of the Ala/Asn side chain compared with histidine. Although hydrogen bonding to the equatorial oxygen of the phosphate from nearby water molecules is still possible, neutral water molecules do not stabilize the transition state either electrostatically or through charge transfer.

DISCUSSION
The value of superimposing structurally homologous proteins to understand catalytic mechanisms is well appreciated. However, it is not as intuitive to superimpose proteins that have no structural homology and that furthermore have clearly distinct metal dependences in their catalytic mechanism. Nevertheless, for APE1 and Nfo, there was surprising agreement in positions of the tetrahydrofuran and leaving 3Ј-ribose oxygen as well as residues and metals important for the catalytic function. The case of the evolutionary convergent catalytic triad in FIGURE 5. Single turnover kinetics of WT and APE1 mutants shows the relative importance of Asp-210 and Asn-212. A, kinetics of product formation of three independent experiments for each enzyme at 10°C. B, rate constants of AP site cleavage (nmol of product formation/min) at 10°C. 52-nt THF-containing oligonucleotide duplex (10 nM) substrate and APE1 (100 nM) were used. serine proteases is analogous, except the proteins have converged on similar side chain chemistry.
For their biological functions, both APE1 and Nfo must recognize an AP site, and both flip the backbone at the AP site into a small pocket as a means of eliminating normal nucleotides with bases (10). For other DNA damage binding proteins, such as the alkylated guanine binding protein ATL, the adenine-guanine mispair glycosylase MutY, and the uracil DNA glycosylase, as well as structure-specific nucleases, such as the FEN1 superfamily and Mre11, specificity is provided by flipping of nucleotides and placing the target phosphate bond into the active site (8, 9, 76 -79). APE1 and Nfo both insert loop(s) from the minor groove side (10). Flipping the AP site ϳ180°into the active site pockets places the scissile phosphate into a position that is amenable for catalysis. Both endonucleases use an activated water to attack the phosphodiester bond, with an electronegative phospho-intermediate whose charge needs to be alleviated for efficient catalysis. Both endonucleases need to protonate the leaving 3Ј-ribose oxygen. We suggest that the requirements needed to catalyze the endonucleolytic reaction define the placement of the active site structural chemistry. It is these strict geometric requirements that made the superposition informative.
Superimpositions of APE1 and Nfo with other endonucleases that do not nucleotide-flip their substrates, such as type II restriction endonucleases, were not as informative. The bonds connecting the scissile atoms diverged in angle, and metal ions did not superimpose. Thus, the successful superposition between APE1 and Nfo is unique for the two prototypic AP endonucleases and defines them in their own class. The use of a magnesium-water cluster in APE1, clearly visible in this higher resolution structure of APE1-product, is not unprecedented. There have been several cases of magnesium-water clusters reported, among them mismatch VSR endonuclease, Serratia and Ppo1 homing endonucleases, Vibrio vulnificus periplasmic nuclease Vvn, endonuclease V that initiates deaminated adenine repair, and UVC nuclear excision repair endonuclease (80 -84), Possibly significant to their substrate specificity, APE1, Vvn, and Serratia homing endonuclease have been shown to be active on both RNA and DNA (85)(86)(87).
Given that the reaction and damaged DNA substrate have defined the geometry, the conservation of location of an active site residue in APE1 relative to the metals in the better structurally studied Nfo provides insight into the APE1 catalytic mechanism; it allows us to plausibly assign the function of individual residues in the mechanism. The resulting mechanism is consistent with our initial model (35) but not other proposed mechanisms (38 -43). Asp-210 and Asn-212 are positioned similarly to the key catalytic metals in Nfo, and their mutation was the most severe, suggesting that they coordinate the attacking water (Fig. 6).
There have been several papers published that have suggested alternative mechanisms that would be inconsistent with the geometric restraints. Because D210H was 15-fold more active than D210A or D210N, it was suggested that Asp-210 is donating a proton to the leaving group (38,39). Asp-210 is not close to the 3Ј-ribose oxygen, so it could not play this role. Histidines can activate the attacking water in nuclease mecha-nisms, and D210H could still activate an attacking water (40,41). It has also been suggested that Tyr-171 in a phenolate form attacks the scissile phosphate directly or generates the hydroxyl that will cleave the scissile phosphate (42). This mechanism would be similar to topoisomerases. However, the angle of attack would not be linear with the position of the 3Ј-ribose oxygen. In a different mechanistic model, His-309 has been suggested to generate the attacking nucleophile (43), but the geometry is also not consistent with this model. It is more than 3.5 Å from where the attacking water would be positioned. Moreover, evidence from NMR has suggested that His-309 in the APE1-DNA complex is protonated (75,88) and thus incapable of serving as a general base. Recently, MD studies suggested that there may be a second metal binding site B, where the Mg 2ϩ coordinates with Asp-210 and Asn-212, and that the one metal moves from site B to the experimentally observed metal site A during catalysis (44,69). Our crystallographic work with one metal in site A does not show any density in position B in our 2.4 Å electron density maps. Because this absence may be due to crystallographic conditions that prevent the B position from being occupied and taking into consideration the finding that site A was occupied, the absence in the electron density does not preclude the possibility that it exists. However, the superimposition with Nfo showed that site B was between Zn2 and the probable position of the attacking water, based on the product complex. Site B is too close (1.4 Å) to the attacking water, and its coordination with both carboxylate oxygen atoms of Asp-210 would prevent Asp-210 from taking the role of directly activating the water, as suggested from our computational work (Fig. 6). The authors also suggested that the retention of magnesium dependence in an E96Q mutant indicates that site A, coordinated by Glu-96, is not important (69). Our structure of the Mg 2ϩ water cluster does not suggest that an E96Q mutation would prevent Mg 2ϩ from occupying site A. Thus, our combined structural, mutational, and computation results do not support the existence of a second metal site B.
Here the determination and analysis of higher resolution APE1 and Nfo structures revealed a conserved active site structural chemistry despite major differences in metal ions and structural elements. In combination with mutational and computational analyses, the structures uncover functional equivalences and support a unified mechanism for AP site excision from DNA.