Structural and Kinetic Analysis of Miscoding Opposite the DNA Adduct 1,N6-Ethenodeoxyadenosine by Human Translesion DNA Polymerase η*

1,N6-Ethenodeoxyadenosine (1,N6-ϵdA) is the major etheno lesion formed in the reaction of DNA with epoxides substituted with good leaving groups (e.g. vinyl chloride epoxide). This lesion is also formed endogenously in DNA from lipid oxidation. Recombinant human DNA polymerase η (hpol η) can replicate oligonucleotide templates containing 1,N6-ϵdA. In steady-state kinetic analysis, hpol η preferred to incorporate dATP and dGTP, compared with dTTP. Mass spectral analysis of incorporation products also showed preferred purine (A, G) incorporation and extensive −1 frameshifts, suggesting pairing of the inserted purine and slippage before further replication. Five x-ray crystal structures of hpol η ternary complexes were determined, three at the insertion and two at the extension stage. Two insertion complexes revealed incoming non-hydrolyzable dATP or dGTP analogs not pairing with but instead in a staggered configuration relative to 1,N6-ϵdA in the anti conformation, thus opposite the 5′-T in the template, explaining the proclivity for frameshift misincorporation. In another insertion complex, dTTP was positioned opposite 1,N6-ϵdA, and the adduct base was in the syn conformation, with formation of two hydrogen bonds. At the extension stage, with either an incorporated dA or dT opposite 1,N6-ϵdA and 2′-deoxythymidine-5′-[(α,β)-imido]triphosphate opposite the 5′-A, the 3′-terminal nucleoside of the primer was disordered, consistent with the tendency not to incorporate dTTP opposite 1,N6-ϵdA. Collectively, the results show a preference for purine pairing opposite 1,N6-ϵdA and for −1 frameshifts.

Chemical or physical damage to DNA can lead to replication blockage and miscoding, which in turn can cause death or mutation of cells (1). Such changes can be associated with aging (2), cancer, and other diseases (3)(4)(5)(6). Chemical damage to DNA can be the result of either exogenous (e.g. vinyl chloride (7)) or endogenous origin (e.g. oxygen radicals (8)). Some specific types of DNA damage can result from either exogenous or endogenous sources, as exemplified by the etheno (⑀) adducts (9).
The etheno adducts are so named because they have an extra two carbons attached to DNA bases in an exocyclic arrangement (10). The history goes back to the chemistry of some unusual natural nucleosides found in tRNA, e.g. wybutosine and wyosine (11,12), and also synthetic work by Leonard and co-workers, who utilized fluorescent etheno analogs to monitor biochemical phenomena (12)(13)(14)(15). Barbin et al. (16) and Laib et al. (17) provided evidence that etheno adducts might be involved in the carcinogenicity of vinyl chloride, a known liver carcinogen in humans. Subsequent work showed the presence of etheno DNA adducts in rats that had never been exposed to vinyl chloride (or related vinyl monomers) (18) and that the adducts are the result of lipid peroxidation (19,20). The reaction of the primary vinyl chloride oxidation product 2-chloroxirane with DNA yields 1,N 6 -ethenodeoxyadenosine (1,N 6 -⑀dA) 5 as the major etheno adduct (21), although the repair of this adduct is faster than that of some others, such that the N 2 ,3ethenodeoxyguanosine levels become higher (22). 1,N 6 -⑀dA has been incorporated into extrachromosomal vectors (23) and studied in cells. The adduct was reported not to be very mutagenic in Escherichia coli (23,24), but this result may be due to extensive repair by the dioxygenase AlkB, which was later discovered to convert this lesion to deoxyadenosine (25). 1,N 6 -⑀dA has been reported to be mutagenic in simian COS7 (24) and human HeLa, HCT116, and HEK293 cells (26,27) and miscoding in human HeLa and XPV cell extracts (28). The bases A, C, and G have all been reported to be misinserted in these systems (24, 26 -28) as well as T (i.e. no miscoding). Structural studies (NMR and x-ray) have been reported on oligonucleotides (in the absence of polymerases), with non-planar pairing between 1,N 6 -⑀dA and dT and anti configurations of both bases (29). 1,N 6 -⑀dA:G pairing involved a syn configuration of 1,N 6 -⑀dA and anti configuration of G (30). An x-ray crystallography study of 1,N 6 -⑀dA:G pairing was interpreted to involve three H-bonds, with distortion of the oligonucleotide backbone to accommodate the pair (31).
Studies of misincorporation at 1,N 6 -⑀dA have yielded varying results with individual DNA polymerases (pols). Singer et al. (32) reported that E. coli pol I and avian myelovirus reverse transcriptase inserted G, to some extent, but that dTTP incorporation was not blocked. When 1,N 6 -⑀A was incorporated into an oligoribonucleotide, it was a complete block to avian myelovirus and Moloney murine leukemia virus reverse transcriptases (33). Human pol (hpol ) was reported to bypass this lesion (34).
Levine et al. (34) reported that hpol was 100-fold more active than hpol in replication past 1,N 6 -⑀dA. We analyzed incorporation events using steady-state kinetics and LC-MS analysis of primers extended opposite 1,N 6 -⑀dA by hpol , in different sequence contexts. The results show hpol bypass past 1,N 6 -⑀dA in a highly error-prone manner, with a proclivity for incorporation of purines and generation of frameshifts. We also describe five x-ray crystal structures of hpol ternary complexes, three containing 1,N 6 -⑀dA paired with dTTP, dAMPNPP, or dGMPNPP (non-hydrolyzable dATP/dGTP analogs) at the insertion stage and two further structures of complexes at the extension stage, with either dA or dT opposite 1,N 6 -⑀dA followed by a nascent pair between template A and an incoming dTMPNPP.

Results
Steady-state Kinetics of dNTP Incorporation Opposite A and 1,N 6 -⑀dA-To determine the efficiency and fidelity of translesion synthesis 1,N 6 -⑀dA adducts by hpol , steady-state kinetic analysis was performed. With an unmodified template (substrate B, Table 1), hpol incorporated a single dNTP opposite A in the order of preference T Ͼ Ͼ A Ͼ G Ͼ C (Table 2), as expected. However, incorporation of dTTP opposite 1,N 6 -⑀dA was very unfavorable, with a drastically decreased catalytic efficiency (1000-fold) compared with that of dTTP insertion opposite A (Table 2). In this study, dATP and dGTP were inserted more efficiently (3.7-and 2.5-fold, respectively, Table 2) than dTTP opposite 1,N 6 -⑀dA. These results indicated that hpol catalyzes translesion synthesis opposite 1,N 6 -⑀dA in a very error-prone manner. Another sequence context containing 1,N 6 -⑀dA (substrate C, Table 1), which had been used in a previous study by Levine et al. (34), was also included in our steadystate kinetic studies. Consistently, a high error rate for incorporation opposite 1,N 6 -⑀dA by hpol was observed, with preference for dGTP and dATP incorporation compared with dTTP (33-and 4.2-fold higher, respectively) ( Table 3 and supplemental Fig. S1).
Pre-steady-state Kinetics-dNTP incorporation opposite 1,N 6 -⑀dA was also examined in the sequence context of substrate A using a rapid quench method. However, no burst phase was observed with any of the four dNTPs, indicating that a step preceding nucleotidyl transfer is rate-limiting in all four cases (supplemental Fig. S2).

LC-MS/MS Analysis of Primer Extension
Products-In the steady-state kinetic analysis, hpol showed a high misinsertion frequency opposite 1,N 6 -⑀dA during single nucleotide extension (Tables 2 and 3). To gain insight into the ability of hpol to  extend beyond the lesion, an LC-MS/MS method previously developed in this laboratory (34 -36) and applied extensively (37)(38)(39)(40)(41)(42)(43) was used for sequence analysis of the extension products. dU-containing primers were extended by hpol in the presence of all four dNTPs, followed by treatment with UDG and piperidine to cleave the fully extended products into shorter fragments for LC-MS/MS analysis (35,36). Sequences of full-length extension products and relative yields were determined by LC-MS/MS analysis ( Fig. 1 and Tables 4 -6). The Ϫ1 frameshift products accounted for 37% of total products in a sequence with a T positioned 5Ј of the 1,N 6 -⑀dA (substrate D-U, Tables 1 and 4). The fraction of extension products with a G incorporated opposite the lesion was 27% and with an A incorporated opposite the lesion was 25%. Small amounts of products with T (11%) and C (1%) incorporated were also observed in the mass spectra. The formation of the Ϫ1 frameshift products may be the result of the incoming dATP skipping the lesion and pairing with the neighboring base T ( Table 4). The T on the 5Ј side of 1,N 6 -⑀dA in the template was replaced with a C, to evaluate the effect of substitution of the (5Ј) pyrimidine (substrate A-U, Table 1). LC-MS/MS results revealed 14 sequences of full-length products corresponding to nine peaks of M Ϫ 2H ions (Table 5). With this template, 70% of the products were due to Ϫ1 frameshifts, followed by 16% A and 14% T incorporation opposite the lesion. Extension products containing a G opposite 1,N 6 -⑀dA were not observed in this setting, presumably because the dGTP skipped 1,N 6 -⑀dA and paired with the C positioned 5Ј to the lesion to yield a frameshift. Human pol , like many other DNA polymerases (35), can also catalyze blunt-end addition of dATP or dGTP. Some extension oligonucleotides were added to the second nucleotide following the blunt end in this experiment (Table 5).
We also included a template containing 1,N 6 -⑀dA in another sequence context with a C in the 5Ј position (substrate C-U,  Table S1). Again we found Ϫ1 frameshifts (39%, dominated by an A in the next position) and incorporation of A (29%) and G (23%) opposite 1,N 6 -⑀dA, instead of T (8%) and C (Ͻ1%), consistent with our results from the steady-state study with this sequence (Table 3 and supplemental Fig. S1).
Crystal Structures of Insertion-stage Ternary Human pol Complexes-We determined five crystal structures of ternary hpol -DNA⅐dNTP complexes with template strands contain-ing 1,N 6 -⑀dA, three of which were trapped at the insertion stage. The resolution of these three structures varied between 2.12 and 2.26 Å (Table 7), and representative images of the quality of the final electron density are provided in Fig. 2, A-C. The first two complexes are of incoming dAMPNPP and dGMPNPP opposite 1,N 6 -⑀dA, the nucleoside triphosphates preferentially incorporated according to the steady-state kinetic data (⑀dA:dAMPNPP and ⑀dA:dGMPNPP, see Tables  1-3 and 7). The structures revealed that the purine bases of incoming nucleotides adopt a staggered orientation relative to 1,N 6 -⑀dA in the anti conformation. This arrangement leads to extensive cross-strand stacking between the adducted base and A or G (Fig. 3). Stacking interactions between 1,N 6 -⑀dA and adenines from the incoming nucleotide and 3Ј-adjacent template residue appear to be slightly more favorable (Fig. 3B) compared with the structure with incoming G (Fig. 3D). In the latter, the adduct base appears to have shifted slightly into the major groove, whereas the relative positions of the base portions of incoming nucleotides relative to the template A located 3Ј to the adduct are very similar in the two structures. In both, Gln-38 forms an H-bond to the sugar O4Ј of 1,N 6 -⑀dA (Fig. 3). In the structure with dGMPNPP, guanine N2 and Gln-38 are also engaged in an H-bond (Fig. 3B), although this interaction is absent in the insertion-stage complex with incoming dAMP-NPP (Fig. 3D). Shared properties of the two structures are the conformation of the template nucleotide situated 5Ј to 1,N 6 -⑀dA, in that this T is directed away from the active site and is not engaged in a stacking interaction with the adduct (Fig. 3, A and C). Furthermore, the A:T base pair at the Ϫ1 position displays strong buckling in both complexes, with the thymine plane tilted relative to 1,N 6 -⑀dA. However, this orientation of the 3Ј-terminal primer nucleotide still leaves its O3Ј at an optimal position to carry out a nucleophilic attack at the ␣-phosphate of the incoming nucleotide (AMPNPP or GMPNPP). The ␣and ␤-phosphate groups of the latter are engaged in an electrostatically favorable interaction with Arg-61 from the hpol finger domain.
An additional structure of an insertion stage hpol complex was determined for incoming dTTP opposite 1,N 6 -⑀dA (⑀dA: dTTP, Tables 1 and 7). In this complex, thymine and the adduct base are coplanar, with the former in the standard anti configuration and the adduct flipped into the syn conformation (Fig.  4). Unlike the structures with incoming purine nucleoside triphosphates, the Ϫ1 base pair is devoid of buckling and the T 5Ј-adjacent to the adduct is now rotated into the active site and stacked on 1,N 6 -⑀dA. However, the syn orientation of the latter results in diminished overlap with the 3Ј-adjacent template A compared with the above complexes with incoming AMPNPP or GMPNPP (Figs. 3 and 4). The syn-1,N 6 -⑀dA:anti-dT base pair features virtually equidistant spacings between the (⑀dA)N 7 ⅐⅐⅐(H)N 3 (T) and (⑀dA)N 6 (H)⅐⅐⅐O 4 (T) atom pairs. In the first contact, N 7 of 1,N 6 -⑀dA is an acceptor, and N 3 of T is a donor. Conversely, N 6 of 1,N 6 -⑀dA and O 4 of T are both acceptors. Rather than a clash between the two latter functionalities, the 2.8-Å separation might be indicative of protonation at N 6 of 1,N 6 -⑀dA. The pK a of N 6 in 1,N 6 -⑀dA is 4.1 (44,45), and under the conditions used for crystallization (pH 5.5), the N 6 nitrogen of the adduct may be at least partially protonated, consistent with the geometry of the nascent base pair at the active site.
Crystal Structures of Extension-stage Ternary hpol Complexes-Structures of ternary complexes with 1,N 6 -⑀dA opposite either primer dA or dT were determined at resolutions of 1.79 and 2.30 Å, respectively (⑀dA:dA extension and ⑀dA:dT extension, Tables 1 and 7; Fig. 2, D and E). Unlike the structure of the insertion complex with dTTP opposite the adduct, the latter adopts the energetically more favorable anti conformation in both extension complexes (Fig. 5). This orientation results in extensive stacking within the template strand between 1,N 6 -⑀dA and the base portions of 5Ј-and 3Ј-adjacent A and T nucleotides. In this fashion, the adduct is bracketed by Watson-Crick A:T(MPNPP) pairs at the 0 and Ϫ2 positions. As in the insertion complexes, Gln-38 establishes an H-bond to the sugar O4Ј of the adducted nucleotide, but the Arg-61 side chain has swung away from the phosphates of the incoming nucleotide, and its guanidino moiety is directed into the major groove ( Fig. 5). Similar to the dTTP insertion complex, the stacked template nucleotides include the T 5Ј-adjacent to Ap(1,N 6 -⑀dA) that is next in line to be replicated. The anti conformation of 1,N 6 -⑀dA leaves insufficient room for either dA or dT from the extended primer to be accommodated in the standard anti conformation (modeled nucleotides colored in gray in Fig. 5). Indeed, the electron density maps show good density for the phosphate group of the 3Ј-terminal primer nucleotide in the structures of the two extension-stage complexes but only partial density for its sugar moiety and no density even at the 0.7 level for the nucleobase portion (Fig. 2, D and E). The disordered state of the 3Ј-terminal T is consistent with the relatively low proclivity by hpol to insert T (or A when the 5Ј-adjacent  Table  S1 for assignments.

Discussion
The kinetic results presented here for insertion by hpol at the 1,N 6 -⑀dA adduct demonstrate that this lesion is significantly miscoding (Tables 2 and 3; supplemental Fig. S1). Under steady-state conditions, incorporation of T opposite 1,N 6 -⑀dA is strongly attenuated relative to insertion of T opposite A, and both purines are preferred over T (Tables 2 and 3). However, all four possible insertions opposite the adduct lack a burst phase as evidenced by our pre-steady-state analysis (supplemental Fig. S2). Therefore, it is unnecessary to invoke impediments during the extension step as the main reason for the low efficiency of bypass past the 1,N 6 -⑀dA adduct by hpol , and the LC-MS/MS results (Tables 4 -6) are consonant with the kinetic insertion results (Tables 2 and 3).
Our misincorporation results may be compared with previous literature. Incorporation of A, C, and G opposite 1,N 6 -⑀dA has been reported in different mammalian cell lines and extracts, in varying ratios (24, 26 -28). T incorporation is reflected in a lack of mutation, but the mutagenic frequency is a function of the DNA repair status of each cell line. Pandya and co-workers (24,27) reported a 70% mutation frequency for 1,N 6 -⑀dA in (monkey) COS7 cells, utilizing a single-stranded vector. Of these mutations, 63% corresponded to C incorporation (i.e. A 3 G transition), 6% to A incorporation, and 1% to G incorporation. No frameshift analysis was reported. In 2000, the same laboratory (26) utilized a double-stranded vector with HeLa, HCT116, and (HEK) 293 cells, derived from human cervical cancer, colon cancer, and embryonic kidney epithelium, respectively. High percentages of large deletions were reported in the 1,N 6 -⑀dA-plasmid cells (16 -89%), with deletion of several hundred to 2000 bases. The number of targeted single mutations was low in the HTC116 cells and higher in the HeLa cells (HEK293 not reported). Of the 25 HeLa cell mutants analyzed with 1,N 6 -⑀dA, 36% corresponded to G insertion, 16% correspond to C insertion, and 48% corresponded to A inser-

LC-MS analysis of products of hpol replication past template 1,N 6 -⑀dA (C 5-adjacent to adduct, Substrate A-U)
Products were cut at the U and begin at the 3' T. An underscore represents a frameshift (deletion). In this case the oligonucleotide pair included an upstream UT mispair. See Supplemental Data Fig. S3 for LC-MS data. The relative units differ from Tables 4 and 6 due to instrumental settings.

TABLE 6
LC-MS analysis of products of hpol replication past template 1,N 6 -⑀dA (T 5-adjacent to adduct, Substrate C-U) (34) Products were cut at the U and begin at the 3Ј AT. An underscore represents a frameshift (deletion). See Fig. 1 and Supplemental Data Table S1 for LC-MS data. tion. One-and two-base frameshifts were not reported. In 2008 Tolentino et al. (28) used a double-stranded vector containing 1,N 6 -⑀dA with extracts of HeLa cells and XPV cells. The misincorporation frequency was very low, and with the HeLa cells only 2-3 mutants were seen for A, G, and C insertion, making any conclusions about preferences untenable.
At the time that some of the previous cellular mutation results were reported (24,26,27), the repair of 1,N 6 -⑀dA by glycosylases (46,47) was known but repair by AlkB (25) and its mammalian orthologs was not recognized yet. The compositions of the DNA polymerase pools in these cell lines are largely uncharacterized, and the assignment of actions of the individual polymerases is not possible based on that information. As Levine et al. (26) state in comparing their results with cells of human origin with (monkey) COS7 cells (24): "…there are several differences in the design of these two studies, including single-stranded versus double-stranded vector, sequence context, location of the DNA adduct relative to replication origin, and host cells. One or several of these factors could have contributed to the differences observed." Ultimately, the contribution of hpol to mutations should be evaluated in cells equivalent in all respects except hpol (and the characterization of other DNA polymerases and repair enzymes established).
Our results with several sequences, including that used by Levine et al. (34), clearly show that hpol bypasses 1,N 6 -⑀dA in a highly error-prone manner, dominated by purine incorporations or generation of Ϫ1 frameshifts instead of insertion of T. This conclusion is based on both the kinetic analyses (Tables 2  and 3; supplemental Fig. S1) and LC-MS/MS analysis of the extended products (Tables 4 -6), including sequences used previously by others (substrate C, Table 1) (34). These misin-corporation patterns are also consistent with the results obtained in mammalian cells (26 -28), although there is considerable variation among cell lines (26). Although A, C, and G insertions have all been reported opposite 1,N 6 -⑀dA in many systems, the presence of frameshifts in cells replicating past 1,N 6 -⑀dA has not been addressed. Levine et al. (26) listed a Ϫ1 deletion probe to be used for detecting frameshifts in cells but did not report any analyses for such events. In that study, large deletions (several hundred to 2000 bp) were observed in 16 -89% of the mutants, depending on the cell line (26).
We found that the higher rate of incorporation of purine nucleoside triphosphates is in large part dependent on the availability of the matching template nucleotide 5Ј-adjacent to 1,N 6 -⑀dA (i.e. T and C for incoming dATP and dGTP, respectively). This purine incorporation is related in part to Ϫ1 frameshifts (Tables 4 -6). The structures of insertion-stage complexes with incoming non-hydrolyzable dNTP analogs (dAMPNPP and dGMPNPP) in the presence of Mg 2ϩ reveal that the 1,N 6 -⑀dA and adenine (guanine) bases are offset at the active site and, instead of direct pairing, engaged in stacking interactions across the template and primer strands (Fig. 3). Thus, unlike the structures of isolated oligonucleotide duplexes (30,31), where the adduct switches into a syn orientation to accommodate pairing with a purine, 1,N 6 -⑀dA remains in the favored anti orientation and the incoming purines adopt the observed staggered orientation.
From the states seen in the two insertion-stage complex structures, the replication process can proceed in two distinct ways. (i) The incoming purine is accommodated opposite the adduct (Tables 4 and 6). (ii) Alternatively, the template T(C) 5Ј-adjacent to the adduct, seen outside the active site in the Human pol and 1,N 6 -Ethenodeoxyadenosine JULY 1, 2016 • VOLUME 291 • NUMBER 27 structures (Fig. 3), rotates inward and above the adduct and then pairs with dAMPNPP (dGMPNPP), thus resulting in the Ϫ1 frameshift products that constituted the majority of the extended oligonucleotides according to the LC-MS/MS analysis. The actual configurations by which A or G are accommodated opposite 1,N 6 -⑀dA in the template⅐primer duplex remain unclear from the two structures at the insertion stage, although it is possible that the adduct and purine nucleotides are stacked in a cross-strand fashion. The structure of the extension complex with dA opposite 1,N 6 -⑀dA is useful in addressing this question. Rather than in the staggered configuration, the 1,N 6 -⑀dA:A pair is wedged  1 and 7). B, staggered arrangement of dGMPNPP and 1,N 6 -⑀dA; insertion stage (⑀dA:dGMPNPP, Tables 1 and 7). C, dTTP opposite syn 1,N 6 -⑀dA; insertion stage (⑀dA:dTTP, Tables 1 and 7). D, anti 1,N 6 -⑀dA opposite conformationally disordered primer dA, followed by dTMPNPP opposite template dA; extension stage (⑀dA:dA extension, Tables 1 and 7). E, anti 1,N 6 -⑀dA opposite conformationally disordered primer dT, followed by dTMPNPP opposite template dA; extension stage (⑀dA:dT extension, Tables 1 and 7). Selected active site residues are colored by atom with carbon atoms shown in maroon (1,N 6 -⑀dA), orange (incoming nucleotide), light blue (template nucleotide 5Ј-adjacent to 1,N 6 -⑀dA; D and E), purple (Arg-61 and Gln-38 from the finger domain), or magenta (Asp/Glu coordinating to Mg 2ϩ or Ca 2ϩ ). Base pairs at the Ϫ1 (A-C) or Ϫ2 (D and E) positions are shown in yellow, primer dA and dT in idealized stacked positions (that would result in a clash with anti 1,N 6 -⑀dA) in D and E, respectively, are shown in dark gray, and Mg 2ϩ and Ca 2ϩ ions are cyan and pink spheres, respectively.
between A:T(TP) pairs at the Ϫ2 and 0 positions that are separated by ϳ7 Å in an axial direction (Fig. 5, A and C). The staggered orientation following insertion can thus be discarded, but the nucleobase portion of primer dA is completely disordered, suggesting a position of the adduct partner either inside the major or the minor groove.  Even the smaller thymine moiety of dT is disordered in the second extension-stage complex (Fig. 5, C and D). However, the anti orientation of 1,N 6 -⑀dA is common to both complexes and is likely preferred because it allows for more optimal stacking interactions with flanking bases, even if it results in a suboptimal arrangement of the pairing partner and loss of H-bonds. This conclusion is supported by the structural and kinetic data for T incorporation opposite 1,N 6 -⑀dA. The structure of the corresponding insertion complex shows the adduct in the syn conformation and therefore has diminished stacking with the 3Ј-adjacent template base (A; Fig. 5). The incoming T and 1,N 6 -⑀dA are in the same plane, and it is possible, based on the relative orientation and taking into account pK a data for the adduct and the acidic pH of the crystallization solution, that two H-bonds are established between the pairing partners. However, the 1000-fold lower efficiency of T incorporation opposite 1,N 6 -⑀dA relative to T opposite A (steady-state results, Table 2) renders it unnecessary to envision an optimal structural context for the 1,N 6 -⑀dA:T pair at the insertion step. Moreover, the incoming C can form a bifurcated H-bond with its N4 amino group to the N6 and N7 acceptors of the adduct or, alternatively (envisioning a protonated state of C (pK a 4.6) under pH 5.5 crystallization conditions), two H-bonds. However, these assumptions are mute as the efficiency of C incorporation is further reduced compared with T (4-fold) and considerably below the levels of efficiency for A or G incorporation and frameshifting.
The structures of incoming dAMPNPP and dGMPNPP in a staggered orientation relative to 1,N 6 -⑀dA and the kinetic data manifesting preferred incorporation of purines opposite the adduct, accompanied with frameshifts offer some parallels to our observations of hpol -catalyzed bypass reactions opposite an abasic site. Specifically, we invoke a "purine rule," meaning that dATP and dGTP are preferred over dCTP and dTTP by this polymerase opposite an abasic lesion (48). Structures of the preferred dNTPs opposite the abasic site in ternary hpol complexes offered insight into this preference, in that the longer purine bases could be linked to the phosphate group of the abasic residue via water bridges. By comparison, the smaller pyrimidines were at a disadvantage, both because of the reduced level of stacking interactions and the inability to tether FIGURE 5. Active site conformations in two ternary hpol extension step complexes with primer dA or dT opposite 1,N 6 -⑀dA, followed by dTMPNPP across template dA. A, ternary extension stage complex with dA opposite 1,N 6 -⑀dA, viewed into the major groove, and B, rotated by 90°and viewed perpendicular to the best plane through the nascent base pair (⑀dA:dA extension, Tables 1 and 7). C, ternary extension stage complex with dT opposite 1,N 6 -⑀dA viewed into the major groove; D, rotated by 90°and viewed perpendicular to the best plane through the nascent base pair (⑀dA:dT extension, Tables  1 and 7). The 3Ј-terminal primer nucleoside (dA/dT) is disordered in both complexes. Residues shown with dark gray bonds represent idealized stacked positions of primer dA (A and B) and primer dT (C and D). The conformational disorder of these residues in the two complex structures might arise from a clash between dA or dT in the modeled orientation with anti 1,N 6 -⑀dA. Selected active site residues are colored by atom with carbon atoms shown in maroon (1,N 6 -⑀dA), orange (incoming nucleotide), light blue (template nucleotide 5Ј-adjacent to 1,N 6 -⑀dA), purple (Arg-61 and Gln-38 from the finger domain), or magenta (Asp/Glu coordinating to Mg 2ϩ ; cyan spheres).
them to the template strand via solvent molecules in the active site. Similarly, purines opposite the 1,N 6 -⑀dA adduct at the hpol active site exploit cross-strand stacking interactions that result in facilitated error-prone bypass. In addition, at least in the case of incoming dGMPNPP, the side chain of Gln-38 mediates an interaction between guanine (N2) and the backbone of the template strand (sugar ring of the adduct, Fig. 3, C and D). Finally, the structure of the third insertion-stage complex with dTTP opposite 1,N 6 -⑀dA and with the adduct adopting a syn orientation (Fig. 5) is reminiscent of the pairing between this lesion and incoming dTTP observed in the crystal structure of a ternary hpol complex (49). However, this Y-family pol is more efficient and less error prone than hpol in regard to bypass of 1,N 6 -⑀dA and incorporates a T opposite the lesion with only about 10-fold reduced efficiency relative to T opposite A, consistent with a more narrow active site that forces base pairs into the normally less favorable Hoogsteen configuration.
In summary, bypass of the 1,N 6 -⑀dA lesion by hpol proceeds with relatively low efficiency and in an error-prone fashion by misinsertion of A and G or generation of Ϫ1 frameshifts. Thus, the outcomes of hpol -catalyzed bypass of the bulky 1,N 6 -⑀dA lesion and an abasic site (48) are surprisingly similar.

Experimental Procedures
Materials-The catalytic core (amino acids 1-432) of hpol was expressed and purified as described previously (50). Unlabeled dNTPs, T4 polynucleotide kinase, and UDG were purchased from New England Biolabs (Ipswich, MA). A mixture of four dNTPs was purchased from Invitrogen. All non-hydrolyzable dNMPNPPs were obtained from Jena Bioscience (Jena, Germany). [␥-32 P]ATP (specific activity 3000 Ci/mmol) was purchased from PerkinElmer Life Sciences. Biospin columns were purchased from Bio-Rad. All oligonucleotides (purified by HPLC by the manufacturers) were obtained from Midland Certified Reagent Co. (Midland, TX), Integrated DNA Technologies (Coralville, IA), or TriLink Biotechnologies (San Diego). The oligonucleotides containing 1,N 6 -⑀dA were from TriLink.
Steady-state Kinetics-A 13-mer primer was 5Ј-5/6-carboxyfluorescein (FAM)-labeled and annealed to an 18-mer template at a 1:1.2 molar ratio ( Table 1). The same procedure was used for the 18/23-mer pair, annealing the primer and template in a 1:1 molar ratio. Enzyme concentrations and reaction times were optimized using time course assays to ensure Յ20% conversion of substrate to product. A typical assay solution included 4 -55 nM hpol , 5 M primer⅐template duplex, 40 mM Tris-HCl buffer (pH 7.5), 5 mM MgCl 2 , 10 mM DTT, 100 mM KCl, 5% glycerol (v/v), 100 g/ml bovine serum albumin, and varying concentrations of a single dNTP (0 -2 mM). Reactions were incubated at 37°C for 5-20 min and were terminated with 9 volumes of a quench solution (20 mM EDTA, 95% formamide (v/v), bromphenol blue, and xylene cyanol). Products were resolved by electrophoresis on 18% (w/v) polyacrylamide gels containing 7.5 M urea. The gels were scanned by a Typhoon Scanner (GE Healthcare) and quantified by fluorescence intensity using ImageJ software (National Institutes of Health). Data were fit to the Michaelis-Menten equation using Prism software (GraphPad, La Jolla, CA) to estimate k cat and K m values (non-linear regression).
Pre-steady-state Kinetics-Rapid quench experiments were performed using a model RQF-3 KinTek quench flow apparatus (KinTek, Austin, TX). The 13-mer primer was 32 P-labeled at the 5Ј end by T4 polynucleotide kinase/[␥-32 P]ATP and annealed to the 18-mer templates. Reactions were initiated by rapid mixing of 32 P-primer⅐template/polymerase mixtures with dNTP and Mg 2ϩ at 37°C. The final concentrations of the reactants were 50 nM hpol , 500 nM 32 P-labeled primer⅐template complex, and 0.5 mM dNTP. Other reaction conditions are the same as described for steady-state kinetics. Reactions were quenched with 0.5 M EDTA at times varying from 5 ms to 5 s. Products were separated on 18% (w/v) polyacrylamide gels and scanned and quantitated using a PhosphorImaging system (Bio-Rad, Molecular Imager FX) and Quantity One software as described previously (35).
LC-MS/MS Analysis of Full-length Extended Products-A 5Ј-FAM-labeled primer containing an appropriately positioned deoxyuridine (dU) (opposite A in some but not all cases) ( Table  1) was annealed to the 18-or 23-mer templates at a 1:1 molar ratio. Reaction conditions were similar to those used in steadystate kinetics assays, except that the final concentrations were 3 M hpol and 25 M primer⅐template duplex, in a total volume of 80 l. Reactions were initiated by addition of a mixture of 1 mM each of dNTPs (A, G, C, and T) and terminated by spincolumn separation to remove dNTP and Mg 2ϩ after 1 h of incubation (37°C). The resulting products were treated with 50 units of UDG and 0.25 M hot piperidine, following a previous protocol (35,36). The extent of the reaction was monitored by denaturing PAGE (Fig. 1A) prior to LC-MS analysis. The cleavage solution was lyophilized and reconstituted in 60 l of H 2 O.
LC-MS/MS analysis was performed on an Acquity ultraperformance liquid chromatography (UPLC) system (Waters Associates) coupled to a Thermo Finnigan LTQ mass spectrometer (Thermo Scientific, San Jose, CA) with an electrospray ionization source. Samples were separated on an Acquity UPLC BEH octadecylsilane (C18) column (1.7 m, 2.1 ϫ 100 mm) at a flow rate of 0.3 ml/min. The column temperature was maintained at 50°C. Eluent A contained 10 mM NH 4 CH 3 CO 2 in 98% H 2 O, 2% CH 3 CN (v/v), and eluent B consisted of 10 mM NH 4 CH 3 CO 2 in 90% CH 3 CN, 10% H 2 O (v/v). A gradient program was run as follows: 0 -3% B over 3 min, 3-20% B over 2 min, 20 -100% B over 1 min, held at 100% B for 2 min, 100 -0% B over 2 min, and held at 0% B for 3 min (all v/v). MS data were acquired in the negative mode and controlled by Xcalibur 2.1 software (Thermo). Electrospray ionization settings were as follows: source voltage 4 kV, source current 100 A, capillary voltage Ϫ49 V, capillary temperature 350, tube lens voltage Ϫ90 V. The most abundant species (Ϫ2 or Ϫ3 charged) were fragmented in the ion trap by collision-induced dissociation (CID) with a normalized collision energy of 35%. An activation Q of 0.25 and activation time of 30 ms were used. Oligonucleotide sequences can be identified by comparing the observed CID spectra and theoretical spectra of candidate oligonucleotide sequences calculated using Mongo Oligo Calculator 2.0 software (University of Utah, Salt Lake City). The relative yields of various DNA extension products were based on their respective peak areas in the extracted ion chromatograms.
Crystallization-Crystals were obtained by the hanging drop vapor diffusion technique at 18°C. 1,N 6 -⑀dA-modified DNA templates and primer sequences used in the crystallization experiments are listed in Table 1. DNA solutions were prepared by mixing template and primer strands in a 1:1 molar ratio and annealing the mixture in the presence of 10 mM sodium HEPES buffer (pH 8.0), 0.1 mM EDTA, and 50 mM NaCl at 85°C for 10 min, followed by slow cooling to room temperature. hpol protein was mixed with the DNA duplex in a 1:1.2 molar ratio in the presence of 50 mM Tris-HCl (pH 7.5) containing 450 mM KCl and 3 mM DTT, followed by addition of either 5 l of 100 mM MgCl 2 or 5 l of 100 mM CaCl 2 . Using a spin concentrator with an Amicon cutoff filter (Millipore, Billerica, MA), the complex was concentrated to a final concentration of ϳ2 mg of protein/ml. Either dTTP or one of the non-hydrolyzable nucleoside triphosphate analogs was added to the concentrated mixtures containing Ca 2ϩ or Mg 2ϩ . The ternary complex solution was mixed with an equal volume of reservoir solution containing 0.10 M sodium MES (pH 5.5), 5 mM MgCl 2 , and 16 -21% (w/v) PEG 2000 monomethyl ether and equilibrated against 500-l reservoir solutions. Crystals typically appeared after overnight incubation and were allowed to grow for 1 week. They were transferred to cryoprotectant solution containing reservoir solution along with 25% glycerol (v/v) and then frozen in liquid nitrogen for data collection.
X-ray Diffraction Data Collection, Structure Determination, and Refinement-X-ray diffraction data were collected at 100 K either on the 21-ID-F or the 21-ID-G beamline of the Life Sciences Collaborative Access Team at the Advanced Photon Source, Argonne National Laboratory (Argonne, IL). All data were processed with the program HKL2000 (51). (Data collection statistics are summarized in Table 7.) All structures were determined by molecular replacement in MOLREP (52,53), using the coordinates of the complex between hpol and native DNA (Protein Data Bank code 4O3N) as the search model. Structures were refined using PHENIX (54), and model building was carried out in COOT (55). Model statistics and geometric parameters are summarized in Table 7. Illustrations were generated with the program UCSF Chimera (56).
Author Contributions-A. P. crystallized the protein complexes and solved the structures; Q. Z. was responsible for the pre-steady-state and part of the steady-state kinetic analysis and part of the LC-MS measurements; Y. S. purified the enzyme and carried out part of the steady-state kinetic analysis and preparation of extended primers for some of the extended primers for LC-MS analysis; K. M. J. did part of the LC-MS measurements, and F. P. G. and M. E. conceived the studies and wrote the paper, along with Y. S.