The Structure of a High Fidelity DNA Polymerase Bound to a Mismatched Nucleotide Reveals an “Ajar” Intermediate Conformation in the Nucleotide Selection Mechanism*

To achieve accurate DNA synthesis, DNA polymerases must rapidly sample and discriminate against incorrect nucleotides. Here we report the crystal structure of a high fidelity DNA polymerase I bound to DNA primer-template caught in the act of binding a mismatched (dG:dTTP) nucleoside triphosphate. The polymerase adopts a conformation in between the previously established “open” and “closed” states. In this “ajar” conformation, the template base has moved into the insertion site but misaligns an incorrect nucleotide relative to the primer terminus. The displacement of a conserved active site tyrosine in the insertion site by the template base is accommodated by a distinctive kink in the polymerase O helix, resulting in a partially open ternary complex. We suggest that the ajar conformation allows the template to probe incoming nucleotides for complementarity before closure of the enzyme around the substrate. Based on solution fluorescence, kinetics, and crystallographic analyses of wild-type and mutant polymerases reported here, we present a three-state reaction pathway in which nucleotides either pass through this intermediate conformation to the closed conformation and catalysis or are misaligned within the intermediate, leading to destabilization of the closed conformation.

To achieve accurate DNA synthesis, DNA polymerases must rapidly sample and discriminate against incorrect nucleotides. Here we report the crystal structure of a high fidelity DNA polymerase I bound to DNA primer-template caught in the act of binding a mismatched (dG:dTTP) nucleoside triphosphate. The polymerase adopts a conformation in between the previously established "open" and "closed" states. In this "ajar" conformation, the template base has moved into the insertion site but misaligns an incorrect nucleotide relative to the primer terminus. The displacement of a conserved active site tyrosine in the insertion site by the template base is accommodated by a distinctive kink in the polymerase O helix, resulting in a partially open ternary complex. We suggest that the ajar conformation allows the template to probe incoming nucleotides for complementarity before closure of the enzyme around the substrate. Based on solution fluorescence, kinetics, and crystallographic analyses of wild-type and mutant polymerases reported here, we present a three-state reaction pathway in which nucleotides either pass through this intermediate conformation to the closed conformation and catalysis or are misaligned within the intermediate, leading to destabilization of the closed conformation.
DNA polymerases replicate DNA with remarkably lower error rates than would be expected solely on the basis of the free energy differences between correct and incorrect base pairing (1). Polymerases, therefore, actively contribute to the discrimination against mismatched bases. Four mechanisms by which DNA polymerases enhance intrinsic base pairing specificity to achieve high fidelity have been proposed; 1) hydrogen bonding between correctly paired bases, 2) water exclusion from the active site, 3) geometric selection of correct size and shape, and 4) conformational changes associated with dNTP binding (for review, see Ref. 2).
X-ray crystal structures of high fidelity DNA polymerases have revealed a large conformational change that is associated with binding to the complementary dNTP. The structure of DNA polymerases has been likened to a hand with a "fingers" subdomain encoding a solvent-exposed dNTP binding site and the "palm" and "thumb" grasping the nascent DNA chain (Fig.  1A) (3,4). For family A DNA polymerases, the template base is flipped out of the DNA base stack and resides in a "pre-insertion site" before encountering a dNTP. Upon binding a complementary dNTP, the fingers subdomain transitions from an open to a closed conformation, enclosing the nucleotide opposite the template base ( Fig. 1B) (3,(5)(6)(7)(8). A concerted dislocation of a highly conserved tyrosine in the O helix of family A enzymes enables the template to pair with the complementary dNTP in a newly formed "insertion site" and positions the triphosphate for in-line attack by the 3Ј-hydroxyl of the extending DNA strand. Although much is known about the conformational changes associated with binding of complementary nucleotides, mismatched substrates have been structurally less well characterized.
In the pool of all four dNTPs, DNA polymerases continually probe nucleotides and select the single complementary pair. It has been suggested that conformational changes play a critical role in the selection process. One model proposes that a dynamic equilibrium between the known open and closed conformations is sufficient to account for selectivity (9,10); only the complementary dNTP can shift the equilibrium toward the closed conformation. A second model posits that an additional, uncharacterized state enables the template to "preview" the incoming dNTP in the open ternary complex (9,11,12); only the complementary dNTP progresses beyond this putative conformation to the catalytically competent, closed conformation.
Here we report x-ray crystal structures of the large fragment of DNA polymerase I from Bacillus stearothermophilus (Bacillus fragment (BF)) 3 bound to DNA and a mismatched nucleoside triphosphate in the active site cleft. We used 2Ј,3Ј-dideoxyterminated primers to prevent nucleotide incorporation, similar to other studies (3,5,6,8,13). These structures reveal a previously unobserved intermediate conformation between the open and closed states that allows the template base to position the incoming nucleotide relative to the active site. In this conformation, mismatches are distinguished from Watson-Crick base pairs by a combination of effects from hydrogen bonding, active site water molecules, and geometric selection.
Crystallization and Structure Determination of Bacillus Fragment Bound to DNA and a Mismatched Nucleotide-The BF ternary complexes were crystallized in a similar fashion as described previously for a BF(D598A/F710Y)⅐DNA(dG)⅐ddCTP-Mg 2ϩ complex (15). For the BF(D598A/F710Y)⅐DNA(dG)⅐ ddTTP-Mg 2ϩ crystal, annealed template and primer DNA were mixed with purified D598A/F710Y BF, 2 mM ddGTP, 5 mM ddTTP, and 20 mM MgSO 4 . Insertion of ddGMP terminates the primer strand to trap the ddTTP in the active site. The use of a 2Ј,3Ј-deoxyribonucleoside substrate was intended to complement the F710Y mutation. Crystals were grown by the hanging drop vapor diffusion at 17°C from a MES (pH 5.8)buffered solution of ϳ50% saturated ammonium sulfate (3) and were transferred into cryoprotectant solutions with increasing concentrations of sucrose before plunging in liquid nitrogen. Our attempts to grow crystals with a chemically dideoxy-terminated primer, BF D598A, and dTTP resulted in an empty and open active site after soaking in cryoprotectant in the presence of dTTP (results not shown). Diffraction data were collected using 1.0 Å synchrotron radiation at the Advanced Photon Source (Argonne National Laboratory, Argonne, IL). For crystals of the V713P mutant BF, annealed T2 and PrimerddC were mixed with purified D598A/V13P BF, 5 mM dCTP or dTTP, and 20 mM MgSO 4 . Diffraction data were collected using 0.972 Å synchrotron radiation at the Advanced Light Source (Lawrence Berkeley National Laboratory, Berkeley, CA). Diffraction data were indexed, integrated, and scaled using XDS (16). All data sets were initially phased by rigid body refinement of the BF⅐DNA(dG)⅐ddCTP complex (PDB code 2HVI (15)). Crystal structures were refined using REFMAC5 (17) and rebuilt in Coot (18). Stereochemistry analysis using Molprobity (19) indicated that 98.0 and 2.0% of all residues in BF(D598A/F710Y)⅐ DNA(dG)⅐ddTTP-Mg 2ϩ and BF(D598A/V713P)⅐DNA(dG)⅐ dCTP-Mg 2ϩ were in the favored and allowed regions of the Ramachandran plot. No backbone torsions were in the disallowed region.
To crystallize BF(D329A/Y714S)⅐DNA(dG)⅐dTTP-Mg 2ϩ , complementary oligonucleotides T2 and PrimerddC were annealed and mixed with purified D329A/Y714S BF double mutant, 5 mM dTTP, and 20 mM MgSO 4 . Crystals were grown as described above. Diffraction data were processed using XDS and then initially phased by molecular replacement using Phaser (20) using a binary BF⅐DNA complex (PDB code 1L3U (3)) as the search model. After refinement using REFMAC5, the N and O helices of the fingers subdomain in the BF(D598A/ Y714S)⅐DNA(dG)⅐dTTP-Mg 2ϩ structure showed positive difference electron density near the modeled atoms due to a partial occupancy of the open conformation. The occupancy of the ajar conformation containing the dTTP substrate was estimated to be ϳ80% judging by the quality of the F o Ϫ F c difference electron density maps; the remaining ϳ20% was attributed to BF in the open conformation, devoid of substrate. Stereochemistry analysis using Molprobity (19) indicated that 97.8 and 2.2% of all residues in BF(D329A/Y714S)⅐ DNA(dG)⅐dTTP-Mg 2ϩ were in the favored and allowed regions of the Ramachandran plot. No backbone torsions were in the disallowed region.
Structural Analysis-The open conformation (PDB code 1L3U) and closed conformation (PDB code 2HVI) models of Bacillus fragment were aligned to the ajar conformation model using PyMol (Delano Scientific, Palo Alto, CA). Root mean square distances were calculated using PyMol. Solvent exposure of substrates and hydrophobic residues in the fingers subdomain (residues 678 -744) was calculated by determining the solvent accessible surface area (21) using StrucTools. A probe radius of 1.4 Å was used.
2-Aminopurine Fluorescence Spectroscopy-Annealed complementary oligonucleotides 2APTemplate and PrimerddC were mixed with an equimolar amount of wild-type BF (described below) and diluted in a cuvette to 500 nM in reaction buffer (50 mM Tris-HCl, pH 8.0, 50 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol). Fluorescence measurements were taken as dCTP or dTTP were titrated into the cuvette at 25°C ( ex ϭ 313 or 315 nm, em ϭ 330 -460 nm). Fluorescence from protein and buffer alone were subtracted from all spectra to measure fluorescence from the template 2-aminopurine (2AP). All spectra were normalized for the small volume changes due to the addition of dNTPs. Fluorescence was measured in an Aminco-Bowman Series 2 Luminescence Spectrometer (Spectronic Instruments, Rochester, NY).
Fluorescence Resonance Energy Transfer (FRET)-The wildtype Bacillus fragment open reading frame was mutated using the QuikChange multi-site-directed mutagenesis kit (Stratagene) in three locations to change cysteines 388 and 845 to serines and glutamine 691 to cysteine as directed by the manufacturer's instructions. The BF triple mutant (C388S/C845S/ Q691C) was expressed and purified as previously described (14). Concentrated protein was then treated with tris-(2-carboxethyl)phosphine to reduce the lone cysteine. Cy3-maleimide (Amersham Biosciences) was dissolved in dimethylformamide and mixed with reduced protein for 2 h at room temperature, then overnight at 4°C. Excess dye was separated from Cy3-labeled protein (Cy3BF) using a desalting column. Absorbance measurements at 280 and 550 nm indicated that 82% of the protein was labeled (data not shown). FRET was observed by comparing the fluorescence spectrum ( ex ϭ 475 nm, em ϭ 490 -640 nm) of a 500 nM mixture of Cy3BF and annealed FluorPrimer and T2 oligonucleotides (FlDNAddG) with the spectra of Cy3BF alone and FlDNAddG alone (see supplemental Fig. 3). Fluorescence measurements were taken as dCTP or dTTP were titrated into a cuvette containing 500 nM Cy3BF and 100 nM FlDNAddG in reaction buffer.
Presteady State Kinetics of BF V713P Mutant-Complementary oligonucleotides T75 (5Ј-TTACTTGACCAGATACAC-TGTCTTTGACACGTTGATGGATTAGAGCAATCACAT-CCAAGACTGGCTATGCACGAA-3Ј) and P67 (5Ј-(6carboxyfluorescein)-TCGTGCATAGCCAGTCTTGGATGT-GATTGCTCTAATCCATCAACGTGTCAAAGACAGTGT-ATCTGGT-3Ј) were synthesized (Operon Biotechnologies, Inc., Huntsville, AL) then dissolved and annealed as described above. P67 contains a fluorescent 6-carboxyfluorescein dye at the 5Ј end. Pilot experiments showed that mixing annealed T75: P67 DNA with BF and dCTP resulted in the extension of the fluorescently labeled P67 strand by one nucleotide, which could be separated from P67 and quantitated using capillary electrophoresis (data not shown). T75:P67 DNA and BF wild type or BF V713P were diluted with reaction buffer to 0.1 and 0.5 M, respectively. The BF:DNA complex was then mixed with an equal volume of dCTP or dTTP at various concentrations at room temperature and then quenched using 4 reaction volumes 95°C formamide. Reactions too fast to capture on the benchtop were performed in a KinTek RQF-3 Rapid Quench Flow instrument and quenched using 95% formamide, 25 mM EDTA. Reaction products were separated and quantitated using an ABI3100 Genetic Analyzer (Applied Biosystems, Foster City, CA). The fraction of P67 that had been extended was plotted for each time point, and the data were fitted to a hyperbolic curve to determine the rate at each dNTP concentration. Kinetic parameters were derived from curve fits of rates versus dNTP concentration plots.

RESULTS
The dG:ddTTP mismatch was captured in the BF polymerase double mutant F710Y/D598A by co-crystallization with a 3Ј-recessed double-stranded DNA containing a deoxyguanosine (dG) in the template overhang together with 2Ј,3Ј-dideoxythymidine triphosphate (ddTTP) as the mismatched incoming nucleotide. The Phe-710 to Tyr mutation promotes incorporation of dideoxynucleotides (22,23); the Asp-598 to Ala mutation destabilizes a crystal contact that predisposes BF to crystallize in the open conformation (3,15). To capture reaction intermediates, the primer strand was terminated by the incorporation of a ddGMP. Using this DNA substrate, we observed a misaligned dG:ddTTP complex in which the O helix adopts a previously unreported, bent conformation that has characteristics in between the fully open and closed state, which we term the ajar conformation.
Structure of a Distinct Intermediate Conformation in the Bacillus Fragment-We crystallized a ternary complex of Bacillus fragment bound to a 2Ј,3Ј-dideoxythymidine triphosphate opposite a template guanine and solved its structure to 1.81 Å resolution (BF(D598A/F710Y)⅐DNA(dG)⅐ddTTP-Mg 2ϩ ; r ϭ 21.0, R free ϭ 24.6; Fig. 1C and Table 1). The asymmetric unit contains two polymerases that are structurally indistinguishable from each other (r.m.s.d. (C␣) ϭ 0.40 Å). In these mismatched ternary complexes, we observed the thymine base of the incoming ddTTP directly interacting with the template guanine by forming specific hydrogen bonds. The dG:ddTTP mismatch forms a wobble pair ( Fig. 2A) with two hydrogen bonds between the bases. A water molecule positioned approximately in the base pair plane completes the hydrogen-bonding potential between the two bases ( Fig. 2A). This wobble pairing displaces the incoming ddTTP by 2.8 Å (root mean square deviation) relative to a complementary dG:ddCTP pair and thereby misaligns the ␣-phosphate for attack by the 3Ј-hydroxyl of the primer terminus (Fig. 2B). Seven water molecules fill the void resulting from this displacement (Fig. 2, B and C); ddTTP is, therefore, more solvated than the complementary ddCTP (solvent-accessible surface areas of 121 and 37 Å 2 , respectively) in the BF active site. Three of the extra water molecules in the ddTTP complex together with three triphosphate oxygen atoms chelate a magnesium (II) in an octahedral geometry (Fig.  1C). The non-bridging phosphate oxygens that are not coordinated by Mg(II) interact with His-682, Arg-702, and Lys-706 (Fig. 1C). We did not observe a second metal ion common in the active sites of polymerase enzymes (24) using anomalous scattering in the presence of manganese instead of magnesium (results not shown), likely due to the missing 3Ј-OH in the primer strand that chelates the metal ion. The template guanine displaces the conserved Tyr-714 from the insertion site and occupies a nearly identical position in both the complementary and mismatched nucleotide structures ( Fig. 2A).
The misaligned base pair is bound to Bacillus fragment in an ajar conformation of the fingers subdomain, which is intermediate between the open and closed conformations and contains elements of both ( To further probe the role of the tyrosine 714 steric gate in nucleotide discrimination, we solved the crystal structure of a D329A/Y714S double mutant of Bacillus fragment bound to a dG:dTTP mismatch. The homologous Y766S mutant of the Escherichia coli DNA polymerase I Klenow fragment (KF) is error-prone (26) and favors misinserting T opposite a template G (27). Like D598A, D329A destabilizes a crystal contact that predisposes BF to crystallize in the open conformation (3). The cocrystallization of the Y714S mutant with a dG⅐dTTP mismatch was performed with a synthesized oligonucleotide primer strand that contained a dideoxy nucleoside monophosphate at the terminus and thus did not require the F710Y mutation. The structure of BF(D329A/Y714S)⅐DNA(dG)⅐dTTP-Mg 2ϩ was solved to 1.75 Å resolution in a crystal form with only one molecule in the asymmetric unit and a smaller unit cell (r ϭ 20.1, R free ϭ 24.1; Table 1) and is nearly identical to the BF(D598A/F710Y)⅐DNA(dG)⅐ddTTP-Mg 2ϩ complex in the ajar conformation (see supplemental Fig. 2). There is no obvi-ous difference between the structures of mismatches with a Tyr-714 or a smaller Ser-714 that would explain the increase in misinserting nucleotides by the mutant. On the other hand, the similarity of these two crystal structures indicates that the crystal packing and the location of a single hydroxyl group, whether on the incoming nucleotide or on residue 710, do not have any appreciable impact on the structure of the mismatch or the enzyme active site.
The intermediate ajar conformation of BF observed in both crystals had not been observed when the complementary nucleotide is bound. It was, therefore, not clear if the ajar conformation arises only in the presence of a G⅐T mismatch or if it is an intermediate between the open and closed conformations in a more general mechanism for nucleotide binding and discrimination. We asked if the enzyme could be trapped in the ajar conformation in the presence of the complementary nucleotide. To capture the selection for Watson-Crick base pairing in the ajar conformation, we mutated valine 713 to proline, which should selectively stabilize the ajar conformation relative to the closed states. The proline mutation blocks the formation of a hydrogen bond between the carbonyl oxygen of residue 709 and the backbone nitrogen of residue 713. It also sterically hinders the formation of a contiguous O helix observed in the closed conformation. The cocrystallization of the V713P mutant was also performed with a synthesized oligonucleotide primer strand  (15)). This conformational change is characterized by rotation of the O helix (curved black arrow) in the fingers subdomain, thereby enclosing the correct ddCTP (stick and balls, blue) within the active site cleft. Red ovals, sites for key base pair interactions; yellow DNA, template strand; orange DNA, primer strand; gray spheres, Mg 2ϩ . The addition of a mismatched nucleotide results in an alternate structure (C) in which the template G pairs with the ddTTP (blue sticks and balls) in the insertion site but the fingers do not close. Residues that interact with the dG:ddTTP pair in the active site cleft are shown in sticks and labeled. Red spheres, active-site bound water molecules; black dashes, hydrogen bonds and magnesium-oxygen bonds; gray mesh, simulated-annealing omit map at 1.0 . The lower portion of panel C shows the sequence of DNA used in the mismatch crystallization experiment. The asterisk denotes the dideoxy-termination of the primer strand. Panels B and C are close-up views of the region of BF⅐DNA complex shown inside the black box in A. All models were generated using PyMol.
that contained a dideoxy nucleoside monophosphate at the terminus and thus did not require the F710Y mutation. As expected, a mismatched dG:dTTP pair induces the ajar conformation in the V713P mutant BF polymerase that is nearly identical to the BF(D598A/F710Y)⅐DNA(dG)⅐ddTTP-Mg 2ϩ complex (data not shown). The crystal structure of a complementary dG:dCTP pair in the V713P mutant BF polymerase (BF(D598A/V713P)⅐DNA(dG)⅐dCTP-Mg 2ϩ ; r ϭ 21.0%, R free ϭ 24.1%; Table 1) shows a complementary dCTP making a Watson-Crick base pair in the ajar conformation (Fig. 3A). Instead of enclosing the complementary nucleotide, the O helix of the V713P mutant BF is bent at Gly-711, and the bound dCTP is solvent-exposed (Fig. 3B). The triphosphate moiety is held far away from the active site, which is likely to slow catalysis.
The V713P Mutant Slows Catalysis-If the ajar conformation is indeed an intermediate in the enzymatic mechanism, obstacles to transitions from the ajar conformation to the closed conformation should affect the rates of catalysis for complementary nucleotides. We measured solution kinetics of complementary (dC) and incorrect (dT) nucleotide incorporation into an oligonucleotide by wild-type or V713P BF (Table 2), which should have a relatively unstable closed conformation. In comparison to wild-type BF, the V713P mutant BF exhibits a similar affinity for the complementary nucleotide but an ϳ10fold slower rate of polymerization (k pol ). The k pol for incorrect nucleotide incorporation is slowed over 100-fold by the mutation. As a result, the V713P mutant BF polymerase is generally slower than wild-type BF, whether binding the correct or incorrect nucleotide.

Conformational States in Solution Probed by Fluorescence-
We used FRET to probe whether complementary and mismatch base pairs induce different conformational changes (10,12). A fluorescein donor fluorophore was attached to the n-6 position of the primer strand of a dideoxy-terminated substrate DNA, and Cy3 acceptor was attached to a cysteine introduced at position 691 in the BF O helix (see supplemental Fig. 3). Consistent with previous work on homologous KF and T7 DNA polymerases that were fluorescently labeled in a similar location near the O helix (12,28), titration of dCTP (complementary nucleotide) or dTTP (mismatch) to this labeled complex resulted in a decrease and increase respectively in donor fluorescence emission intensity (Fig. 4, A and B). Because of the lack of knowledge about the relative orientations of the donor and acceptor fluorophores, we cannot relate distances between the fluorophores in the presence of the correct or incorrect nucleotides. The opposite directions of fluorescence intensity changes, however, indicate that the two nucleotides elicit distinct conformational changes in the enzyme, consistent with the inability of the mismatch to form the closed state. The data fit to a single-site binding isotherm (Fig. 4C) revealed that the mismatch binds ϳ30-fold more weakly than the complementary base.
The structure of the ajar conformation revealed that the incorrect nucleotide can also induce the template base to move from the preinsertion into the insertion site. This movement can be observed in solution by monitoring the fluorescence intensity emission of 2AP, an environmentally sensitive nucleoside analog, placed in the template acceptor position (29). 2AP, like adenine, can form two hydrogen bonds with thymine  (Fig. 4D), presumably as a result of the movement of 2AP into the insertion site. Titration of dCTP (mismatch) also decreases 2AP fluorescence emission intensity to a similar saturation point (Fig. 4E), supporting the hypothesis that a mismatched nucleotide also induces the template base to move into the insertion site. Furthermore, binding isotherms again show that the mismatch binds ϳ30-fold more weakly than the complementary nucleotide (Fig. 4F). Together, these equilibrium studies show that, although both mismatched and complementary nucleotides move the template base into the insertion site (Fig.  4, D-F), the dG:dTTP mismatch induces an O helix conformation that is distinct from that observed in the presence of the complementary nucleotide (Fig. 4, A-C). The ajar conformation observed in the crystal structure of BF when bound to a mismatch (Fig. 2, A and C) is consistent with both of these observations in solution phase.

DISCUSSION
Before chemical incorporation, DNA polymerases must test incoming nucleotides against the template base to distinguish between complementary and mismatch nucleotides. Experimental studies (11,12) suggest that this process involves at least one distinct conformational state that precedes the adoption of the closed state in family A DNA polymerases. We determined the crystal structure of a BF polymerase complex in which the template base has moved into the insertion site and is paired with a mismatched and misaligned nucleotide. Importantly, the template base occupies the same location when bound to a mismatched or a complementary nucleotide, indicating that the template is used to align or misalign incoming nucleotides relative to the active site using hydrogen bonds between bases. The mismatched and misaligned ternary complex appears unable to fully desolvate the incoming nucleotide, suggesting that active site water molecules play a role in nucleotide selection.
A key feature of the protein structure in the mismatched ternary complex is the pronounced kink in the mobile O helix near the active site. In the ajar conformation, the O helix is positioned at an angle intermediate to that adopted in the open or close state. In contrast to the open and closed states, the O helix is bent at Gly-711, allowing the template base to interrogate the incoming nucleotide without enclosing it. In addition, we trapped the enzyme complex bound to a complementary dNTP in the ajar conformation by mutating an O helix residue that inhibits the closed state. This observation demonstrates that the polymerase is capable of adopting the ajar conformation in the presence of a complementary or a noncomplementary nucleotide. Thus, this ajar conformation has the hallmarks expected for the structure of a reaction intermediate involved in previewing nucleotides in the polymerase active site.
The importance of the glycine hinge is further underlined by its conservation pattern among the members of the family of polymerases. This residue is conserved in T7 DNA polymerase, bacterial DNA polymerase I enzymes, human family A DNA polymerases ␥ (hPolG), which replicates mitochondrial DNA with high fidelity (30), and (hPolQ), a DNA repair DNA polymerase with conflicting reports on its fidelity (31, 32), but not in DNA polymerase (hPolNu), a low fidelity repair polymerase with a particular propensity for incorporating a T opposite a template G (33) (Fig. 5A). Low fidelity Y-family DNA polymerases that participate in DNA repair are hypothesized not to employ large-scale conformational changes during nucleotide binding of complementary or mismatched nucleotides (34) nor do they have an O-helix equivalent (35). The juxtaposition of these high and low fidelity family A DNA polymerases suggests that the existence of the ajar conformation contributes to fidelity in DNA replication.
Mismatch ternary complexes have also been captured in human DNA polymerase ␤ (Pol ␤), an X-family DNA polymerase, using non-hydrolyzable nucleotide analogues and manga-  nese in place of magnesium (36). Despite the similarity in DNA replication accuracy and the two-metal ion catalytic mechanisms between A-and X-family DNA polymerases, BF and polymerase ␤ recognize mismatches using distinct mechanisms (36). Polymerase ␤ does not contain the glycine hinge. Instead of forming a mismatched base pair as in BF, incorrect nucleotides in polymerase ␤ slow catalysis by displacing the primer terminus and template (36).
Recently, the conformations adopted by the homologous Klenow fragment from E. coli DNA polymerase I (KF) in the presence of DNA and nucleotides were studied by singlemolecule FRET between fluorophores on the thumb and fingers subdomains (37). When bound to mismatches (template A with dGTP or template T with dTTP), the apparent FRET efficiency slightly increased to an intermediate level between values for the open and closed conformations, suggesting that the O helix of KF adopted a conformation intermediate of the open and closed conformations. This alternate ternary complex is more similar to the open than the closed conformations and is presented as a candidate for a kinetic checkpoint before the closed conformation, in which the template tests the incoming nucleotide for complementarity. The ajar conformation we describe here using the dG:dTTP mismatch in BF, which is more similar to the open than the closed conformation (Fig. 2C), matches closely with the mis-matched ternary complexes observed in KF. It is quite possible that both enzymes use a similar mechanism to distinguish multiple mismatches.
In previously proposed structural reaction pathways, the template base pairs with an incoming dNTP either in the preinsertion site in the open conformation (37) or in the insertion site with the triphosphate bound to the active site aspartic acid residues instead of the O helix (11  In our interpretation (Fig. 5B), nucleotide selection operates by a modulation of the free energies of the closed state and relaxation back to the open state from the ajar conformation. A nucleotide binds along the O helix and initially interacts with the template base in the ajar conformation.
Our studies leave open the question of whether the incoming nucleotide induces the ajar conformation upon binding or the enzyme exists in equilibrium between the open and ajar states. A complementary dG:dCTP base pair passes through the ajar conformation to the low energy closed conformation (see supplemental Movie S1), stabilized by three hydrogen bonds between bases, proper geometric alignment, and binding of magnesium to the polymerase active site and to the incoming triphosphate. The closed conformation correctly aligns the chemical partners for catalysis. For the dG:dTTP mismatch, the closed state is destabilized due to having two instead of three interbase hydrogen bonds and to misalignment of the incoming nucleotide away from the active site. Misalignment of the mismatched nucleotide appears to impede its desolvation in the active site and may interfere with the proper chelation of catalytic magnesium ions by active site aspartates. The relatively high energy of the mismatched closed state results mostly in relaxation back to the open state and nucleotide release. FRET studies suggest that the closed state is partially occupied by Klenow fragment DNA polymerase in the presence of mismatches (37), and further discrimination of the mismatch may occur in the closed state or in additional states between ajar and closed.
In this model, the transitional ajar conformation is relatively unstable compared with the open and closed states, explaining why we are unable to observe the ajar conformation without the use of BF mutants (F710Y with dG⅐ddTTP, Y714S with dG⅐dTTP, or V713P with dG⅐dCTP) that potentially stabilize the ajar conformation selectively (see "Experimental Procedures"). The observation of the ajar conformation with three different mutations at three different residues suggests that the bending of the O helix is not the result of any single mutation. An additional factor in the instability of the observed ajar conformation is the missing divalent cation at metal "site A" (24). The use of a 2Ј,3Јdideoxy-terminated primer strand to prevent nucleotide insertion removes a chelating ligand for the site A metal, but it has been observed in the complementary ternary complex of Bacillus fragment (38). Recent solution fluorescence experiments on E. coli KF indicate that the site A metal does not bind until the fingers subdomain have closed around the complementary nucleotide (39). The absence of a site A metal in the BF ajar conformation correlates with the possibility that the ajar conformation precedes the closed conformation during nucleotide selection.
Although we have captured a mismatched nucleotide bound in the active site cleft of a high fidelity DNA polymerase, the structure does not answer how mismatches are incorporated. Because of a high energy barrier to incorrect nucleotide incorporation in the open conformation, it was hypothesized that "nucleotide misinsertions may occur from a partially open conformation" (40). The triphosphates of dG:dTTP mismatches we describe in the ajar conformation are distant from the location of the 3Ј-OH and not properly aligned for attack, suggesting that catalysis of mismatch incorporation in the ajar conformation would be very slow. In the context of our model for nucleotide selection, the observation that the V713P mutant that inhibits the closed state slows mismatched nucleotide incorporation more than complementary nucleotide incorporation ( Table 2) suggests that mismatched nucleotides must still transition from the ajar the closed conformation for catalysis.