Insight into mechanisms of 3′-5′ exonuclease activity and removal of bulky 8,5′-cyclopurine adducts by apurinic/apyrimidinic endonucleases

Significance

Oxidative DNA damage has been postulated to play an important role in human neurodegenerative disorders and cancer. 8,5′-cyclo-2′-deoxyadenosine (cdA) is generated in DNA by hydroxyl radical attack and strongly blocks DNA replication and transcription. Here we demonstrate that cdA adducts at 3′ termini of DNA can be removed by 3′-5′ exonuclease activity of the apurinic/apyrimidinic (AP) endonucleases: Escherichia coli Xth and human APE1. The crystal structure of bacterial AP endonuclease in complex with DNA duplex provides insight into the mechanism of this activity. This new repair function provides an alternative pathway to counteract genotoxic effect of helix-distorting DNA lesions.

Abstract

8,5′-cyclo-2'-deoxyadenosine (cdA) and 8,5′-cyclo-2'-deoxyguanosine generated in DNA by both endogenous oxidative stress and ionizing radiation are helix-distorting lesions and strong blocks for DNA replication and transcription. In duplex DNA, these lesions are repaired in the nucleotide excision repair (NER) pathway. However, lesions at DNA strand breaks are most likely poor substrates for NER. Here we report that the apurinic/apyrimidinic (AP) endonucleases—Escherichia coli Xth and human APE1—can remove 5′S cdA (S-cdA) at 3′ termini of duplex DNA. In contrast, E. coli Nfo and yeast Apn1 are unable to carry out this reaction. None of these enzymes can remove S-cdA adduct located at 1 or more nt away from the 3′ end. To understand the structural basis of 3′ repair activity, we determined a high-resolution crystal structure of E. coli Nfo-H69A mutant bound to a duplex DNA containing an α-anomeric 2′-deoxyadenosine:T base pair. Surprisingly, the structure reveals a bound nucleotide incision repair (NIR) product with an abortive 3′-terminal dC close to the scissile position in the enzyme active site, providing insight into the mechanism for Nfo-catalyzed 3′→5′ exonuclease function and its inhibition by 3′-terminal S-cdA residue. This structure was used as a template to model 3′-terminal residues in the APE1 active site and to explain biochemical data on APE1-catalyzed 3′ repair activities. We propose that Xth and APE1 may act as a complementary repair pathway to NER to remove S-cdA adducts from 3′ DNA termini in E. coli and human cells, respectively.

Oxidative damage to DNA caused by reactive oxygen species is believed to be a major type of endogenous cellular damage. If unrepaired, the damage will tend to accumulate and lead to premature aging, neurodegenerative disorders, and cancer (1). More than 80 different oxidative modifications of DNA bases and sugar backbone have been identified to date (2). Diastereoisomeric (5′S)- and (5′R)-8,5′-cyclo-2′-deoxyadenosine (cdA) and 8,5′-cyclo-2′-deoxyguanosine (cdG) are generated by endogenous oxidative stress and ionizing radiation among other oxidized bases (Fig. 1A). 8,5′-cyclo-2′-deoxypurines (cdPu) are generated by hydroxyl radical attack at C5′ sugar by H-abstraction resulting in formation of C5′-centered sugar radical, which then reacts in the absence of oxygen with the C8 of the purine. Subsequent oxidation of the resulting N7-centered radical leads to intramolecular cyclization with the formation of a covalent bond between the C5′- and C8-positions of the purine nucleoside. When present in DNA duplex cdA causes large changes in backbone torsion angles, which leads to weakening of base pair hydrogen bonds and strong perturbations of the helix conformation near the lesion for both diastereoisomers. Interestingly, the glycosidic bond in S-cdA is approximately 40-fold more resistant to acid hydrolysis compared with regular dA, implying that this base lesion would be resistant to DNA glycosylase action (3).

Fig. 1.

Repair of 3′-blocking bulky adducts by AP endonucleases-catalyzed 3′→5′ exonuclease activity. (A) Chemical structures of 5′S- (S-cdA) and 5′R- (R-cdA) diastereoisomers of cdA. (B) Denaturing PAGE analysis of repair products generated by the AP endonucleases when acting upon 3′-terminal S-cdA nucleotide. 5′-[³²P]-labeled A•T^rec, cdA•T^rec, and cdAA•T^rec duplex oligonucleotides were incubated with various AP endonucleases of different origins (5 nM of APE1, Nfo, and Apn1 or 10 pM of Xth) for 10 min at 37 °C. (C) APE1-catalyzed 3′ repair activities toward recessed, gapped, and nicked duplex DNA. 5′-[³²P]-labeled oligonucleotide duplexes were incubated in the presence of 5 nM Ape1 for 10 min at 37 °C. The “X” denotes the position of S-cdA nucleotide. Details in Materials and Methods.

The cdA adducts in DNA are a strong block to various DNA polymerases, such as T7, δ, and η (4). Interestingly, translesion DNA polymerase η can perform lesion bypass synthesis on the R-cdA but not on S-cdA (5). Both diastereomers of cdA also inhibit DNA transcription by blocking primer extension by T7 DNA polymerase, and S-cdA inhibits binding of the TATA box protein in vitro and strongly reduces gene expression in vivo (6). In addition, in vivo human RNA polymerase II generates mutated RNA transcripts when using DNA template containing S-cdA (7). Given the strong genotoxic effect of cdA adducts on DNA metabolism, cells should have a repair mechanism to remove these helix-distorting DNA adducts. Indeed, it was shown that the nucleotide excision repair (NER) pathway can remove cdA adducts with efficiency comparable to that of T = T cyclobutane dimers and exhibits higher activity in excising the R-isomer (4, 8). In agreement with the biochemical data, it was shown that cdPu adducts accumulate in keratinocytes from xeroderma pigmentosum group C and Cockayne syndrome (CS) group A patients exposed to X-rays and potassium bromate (KBrO₃) (9, 10) and also in organs of CS group B knockout mice (11). Importantly, cdA and cdG lesions accumulate spontaneously in nuclear DNA of WT mice with age, suggesting that DNA repair is unable to keep the steady-state level of these complex DNA lesions over the lifespan of an organism (12). Interestingly, S-cdA diastereoisomers are removed in the NER pathway much less efficiently than the corresponding 5′R-cdA ones and are also present at a higher level in nontreated mice organs (4, 12).

At present, NER is the only known DNA repair pathway to remove cyclopurine adducts in duplex DNA. However, removal of lesions located in close proximity to DNA strand breaks by NER has not been reported except in one study by Plunkett and coworkers (13). These authors demonstrated that a 3′-terminal cytosine analog present at a single-strand break is not a substrate for the human global genome NER pathway, but transcription-coupled NER may participate in cleansing single-strand breaks from this 3′ adduct (13). Therefore, it is unlikely that NER could be able to efficiently remove cdPu located at 3′ termini of a single-strand break. Recently it has been demonstrated that 5′R and 5′S isomers of cdATP could be incorporated with low efficiency by replicative DNA polymerases and then inhibit further DNA synthesis, thus potentially generating gapped DNA with a cdA adduct at the 3′ end (14). Furthermore, in the absence of ionizing radiation and/or drugs, cdPu could arise at 3′ termini as a result of 3′→5′ exonuclease degradation of DNA, for example by TREX1 (5).

The majority of oxidatively damaged DNA bases are substrates for two overlapping pathways: DNA glycosylase-initiated base excision repair (BER) and apurinic/apyrimidinic (AP) endonuclease-mediated nucleotide incision repair (NIR) (15). In the NIR pathway, an AP endonuclease makes an incision 5′ to a damaged nucleotide and then extends the resulting single-strand break to a gap by a nonspecific 3′→5′ exonuclease activity (16, 17). AP endonucleases are multifunctional DNA repair enzymes that possess AP site nicking, 3′ repair diesterase, NIR, and 3′→5′ exonuclease activities and are divided into two distinct families based on amino acid sequence identity to either Escherichia coli exonuclease III (Xth) or endonuclease IV (Nfo) (18). Human APE1 is homologous to Xth, whereas Saccharomyces cerevisiae Apn1 is homologous to Nfo. Previously it was shown that AP endonuclease-catalyzed 3′→5′ exonuclease activity could serve as a 3′ editing function for removing mismatched and oxidized bases at 3′ termini of DNA duplex (19–21). However, the detailed mechanisms for those 3′ editing repair activities are not yet clearly understood. Although Xth and Nfo AP endonuclease families share common DNA substrate specificities, they are distinguished by their modes of DNA damage recognition. Indeed, cocrystal structures of Nfo bound to an AP site analog, tetrahydrofuran (THF), showed that the enzyme drastically distorts the DNA helix by ∼90° bending and flips out not only the target AP site but also its opposing nucleotide out of the DNA base stack (22, 23). Interestingly, the Nfo active site pocket sterically excludes binding of normal β-configuration nucleotides, but it can fit α-anomeric nucleotides. In contrast, cocrystal structures of APE1 bound to abasic site-containing DNA show that the enzyme kinks the DNA helix by only 35° and binds a flipped-out AP site in a pocket that excludes DNA bases, whereas the opposite base remains stacked in the duplex (24). Importantly, the DNA substrate specificity of APE1 but not that of Xth varies depending on reaction condition (25).

Here we investigate whether S-cdA adducts in DNA can be removed in alternative to NER pathways by AP endonucleases. Our results show that in contrast to Nfo and Apn1, Xth and APE1 remove S-cdA adducts when present at 3′ termini of a recessed, nicked, or a gapped DNA duplex. However, S-cdA adducts located at 1 or more nt away from the 3′ end are not substrates for AP endonucleases. Using the high-resolution crystal structure of Nfo-H69A:DNA complex, which provides a picture of the position of the DNA for exonuclease activity, we were able to model the DNA exonuclease conformation into APE1’s active site and provide insight into the mechanism for repair of S-cdA adducts. The structural basis and potential biological importance of the reported substrate specificity of Xth and APE1 in cleansing genomic DNA of highly cytotoxic lesions are discussed.

Results

Xth/APE1 but Not Nfo/Apn1 Remove Bulky S-cdA Adducts at 3′ Termini of Recessed, Nicked, and Gapped Duplex DNA.

First we examined whether an S-cdA nucleotide in duplex DNA is a substrate for NIR-AP endonucleases E. coli Nfo, yeast Apn1, human APE1, or the human endonuclease VIII-like 1 (NEIL1) DNA glycosylase. For this, we incubated a 3′-[³²P]-labeled 42-mer duplex oligonucleotide referred to as cdA⁴²•T⁴² with an excess of enzyme and then analyzed the reaction products by denaturing PAGE. No cleavage of cdA⁴²•T⁴² by any of these enzymes was observed, indicating that bulky S-cdA adduct is not a substrate for either NIR or BER pathways (Fig. S1).

It was previously shown that S-cdA nucleotide in duplex DNA is a strong block to 3′→5′ exonuclease activities of T4 DNA polymerase, mammalian 3′ repair exonuclease 1 (TREX1), E. coli Xth, snake venom phosphodiesterase, and nuclease P1 (5). Because AP endonucleases contain robust 3′→5′ exonuclease activity that removes regular and modified nucleotides in duplex DNA (20), we examined whether APE1, Nfo, Apn1, and Xth could remove S-cdA monophosphate (S-cdAMP) when placed in close proximity to 3′ termini of the recessed duplex oligonucleotide cdA•T^rec and cdAA•T^rec (Table 1). As shown in Fig. 1, all AP endonucleases tested efficiently degrade regular dA•T^rec duplex oligonucleotide containing normal dA nucleotide at the 3′ termini of a gap (Fig. 1B, lanes 2, 7, 10, and 14). Intriguingly, APE1 and Xth, but not Nfo and Apn1, can remove 3′-terminal S-cdA nucleotide in cdA•T^rec with high efficiency (lanes 4 and 15 vs. 8 and 11) and then continue to degrade DNA further in a processive manner by their nonspecific 3′→5′ exonuclease activity (Fig. 1B, lanes 2, 7, 10, and 14). Interestingly, the recessed duplex cdAA•T^rec with the S-cdA adduct placed at the second position from the 3′ terminus completely blocks the 3′→5′ exonuclease activity of APE1, Nfo, and Xth (lanes 6, 9, and 16). However, Apn1 was able to remove 3′-terminal dAMP in cdAA•T^rec duplex with very low efficiency but then was completely blocked by the remaining S-cdA adduct (lane 12).

Table 1.

Sequence of the oligonucleotides used in the study, where X is S-cdA

Oligonucleotide name	Sequence 5′→3′	Source
Upstream strands
Exo20A, 21 mer	d(GTGGCGCGGAGACTTAGAGAA)	20
Exo20THF, 21 mer	d(GTGGCGCGGAGACTTAGAGA-THF)	20, 30, 35
Exo20PG, 21 mer	d(GTGGCGCGGAGACTTAGAGA-PG)	This study
Exo20-ScdA, 21 mer	d(GTGGCGCGGAGACTTAGAGAX)	This study
Exo19-ScdAA, 21 mer	d(GTGGCGCGGAGACTTAGAGXA)	This study
cdA42, 42 mer	d(AGAAACAACAGCACTACTGTACTCATGXATTCTATTCCAGCA)	This study
Downstream strands
3′-pExo19, 19 mer	pd(ATTTGGCGCGGGGAATTCC)	16, 17, 20, 28, 30
3′-pExo18, 18 mer	pd(TTTGGCGCGGGGAATTCC)	16
Template strands
Rex-T, 40 mer	d(GGAATTCCCCGCGCCAAATTTCTCTAAGTCTCCGCGCCAC)	20, 35
T42, 42 mer	d(TGCTGGAATAGAATTCATGAGTACAGTAGTGCTGTTGTTTCT)	This study
T55, 55 mer	d(CGAGGACAGACACTGCTGGAATAGAATTCATGAGTACAGTAGTGCTGTTGTTTCT)	This study
Duplex name	Oligonucleotides hybridized
A•Trec (nick/gap)	Exo20A, Rex-T and 3′-pExo19 (nick) or 3′-pExo18 (gap)	20, 35
THF•Trec	Exo20THF and Rex-T	17, 35
PG•Trec	Exo20PG and Rex-T	This study
cdA•Trec (nick/gap)	Exo20-ScdA, Rex-T and 3′-pExo19 (nick) or 3′-pExo18 (gap)	This study
cdAA•Trec (nick/gap)	Exo19-ScdAA, Rex-T and 3′-pExo19 (nick) or 3′-pExo18 (gap)	This study

To precisely determine the distance from the 3′ end to S-cdA nucleotide at which the adduct starts to inhibit 3′→5′ exonuclease activity of AP endonucleases, we constructed recessed DNA duplex cdA⁴²•T⁵⁵ containing S-cdA nucleotide on the recessed strand 14 nt away from the 3′ end. As expected, the presence of S-cdA adduct strongly blocks the 3′→5′ exonuclease activities of all AP endonucleases tested (Fig. S2). APE1 and Xth exonuclease activities stop 2 nt before the S-cdA adduct (in the context 5′-DNA-cdA-A-T-3′), whereas Nfo exonuclease slows down 3 nt and completely blocked 1 nt before the lesion, and Apn1 is blocked at the lesion site (Fig. S2). Taken together, these results suggest that a bulky S-cdA adduct placed in the middle of a DNA duplex cannot be removed by AP endonuclease-catalyzed NIR or by exonuclease activities. Nevertheless, human APE1 and E. coli Xth can efficiently eliminate S-cdA adduct at the 3′ termini of a recessed DNA duplex.

Next we examined APE1 activities on the recessed, nicked, and gapped DNA duplexes containing either regular dA or S-cdA nucleotide in close proximity to the 3′ end. As shown in Fig. 1C, whereas APE1 exhibits highly processive 3′→5′ exonuclease activity on nondamaged recessed A•T^rec duplex (lane 2), it is strongly inhibited on the gapped A•T^gap duplex and almost blocked on the nicked A•T^nick duplex (lanes 3 and 4). These results indicate that APE1 requires an extended single-stranded DNA region for its processive exonuclease function. It should be noted that previous studies showed either similar efficiency of APE1-catalyzed exonuclease toward the recessed, nicked, and gapped DNA (26, 27) or even lowest efficiency on the recessed DNA duplex (28). These apparent discrepancies between our data and the previous studies could be explained by several factors, such as (i) different reaction conditions used, (ii) different sequence context of DNA substrates used, and (iii) the use of histidine-tagged APE1 instead of native form (19, 28). Interestingly, APE1 efficiently eliminates S-cdA nucleotide at the 3′ end of all recessed, nicked, and gapped DNA duplexes (lanes 6–8), suggesting that a single-strand break at the 3′ side of the S-cdA adduct in duplex DNA is sufficient for its removal. After removing the 3′-terminal S-cdA nucleotide in the recessed DNA duplex, APE1-catalyzed 3′→5′ exonuclease activity continues to degrade DNA in a nonspecific manner (lane 6), whereas it is inhibited on the gapped and nicked DNA duplexes (lanes 7 and 8). Strikingly, when acting upon nicked DNA duplexes APE1 efficiently removes a 3′-terminal S-cdAMP (lane 8) but not a 3′-terminal regular dAMP (lane 4). This result strongly suggests that APE1 recognizes S-cdA adduct with a high degree of specificity when present at the 3′ end. However, APE1 exonuclease function is totally inhibited on the recessed cdAA•T^rec, nicked cdAA•T^nick, and gapped cdAA•T^gap DNA duplexes where S-cdA is located 1 nt before the 3′ end (lanes 10–12), in agreement with the results described above.

Characterization of APE1 Interactions with DNA Duplex Containing 3′-Terminal S-cdA Adduct.

To characterize the specificity of APE1 interactions with S-cdA adducts, we measured the kinetic parameters of excision of 3′-terminal S-cdA adduct by APE1 under steady-state conditions. Comparison of the kinetic constants for recessed DNA substrates containing different DNA lesions on 3′ termini (Table 2) showed that the k_cat/K_M value for the WT APE1-catalyzed excision of S-cdA adduct (1,700 min⁻¹ µM⁻¹) was two- and fivefold higher compared with that for 3′-terminal THF residue (780 min⁻¹ µM⁻¹) and regular 3′-terminal dA nucleotide (360 min⁻¹ µM⁻¹), respectively. These results indicate that the efficiency of APE1 3′ cleansing activity for the S-cdA adduct was comparable to that for 3′ sugar-phosphate group and significantly higher than for a regular deoxynucleotide in the recessed DNA duplex.

Table 2.

Steady-state kinetic parameters of the WT and D308A mutant APE1 proteins

Substrates	APE1 WT			APE1 D308A			Fold decrease of kcat/KM value, WT/D308A	Source
	KM, nM	kcat, min−1	kcat/KM, min−1 M−6	KM, nM	kcat, min−1	kcat/KM, min−1 M−6
cdA•Trec	5.4	9.3	1700	15	5.5	360	4.7	This study
A•Trec	2.4	0.86	360	21	0.027	1.3	280.0	35
THF•Trec	8.2	6.4	780	5.4	0.56	104	7.5	35

SDs for K_M and k_cat values varied within 20–40%. Details in Materials and Methods.

Next, to understand the mechanism of inhibition of APE1 3′→5′ exonuclease activity on the recessed DNA duplex containing an S-cdA adduct located 1 nt away from the 3′ end, we studied the interactions between APE1 and 5′-[³²P]-labeled A•T^rec, cdA•T^rec, and cdAA•T^rec duplexes using an EMSA. APE1 binds to recessed duplex cdA•T^rec with 3′-terminal S-cdA more efficiently than to a regular A•T^rec duplex and essentially fails to form stable DNA–protein complexes with cdAA•T^rec duplex in which S-cdA nucleotide is located at the second position from the 3′ end of a gap (Fig. S3A). These results suggest that the lack of APE1 activity on recessed DNA duplexes containing an S-cdA adduct 1 or more nt away from the 3′ end might be due to the loss of enzyme affinity to the DNA substrate.

Structure of E. coli Nfo-H69A Mutant Bound to a Cleaved DNA Duplex Reveals the Mechanism of Exonuclease Activity.

To gain insight into the structural basis of substrate specificity of E. coli Nfo and human APE1, we performed crystallographic studies using a 15-mer DNA duplex containing a single α-anomeric 2′-deoxyadenosine (αdA) nucleotide. The sequence context was taken from previous study by Garcin et al. (23) [Protein Data Bank (PDB) code 2NQJ] of the catalytically inactive mutant E261Q but having an αdA•T pair instead of a AP•G pair at position 7. Unfortunately, all attempts to cocrystallize these two WT AP endonucleases with αdA•T oligonucleotide duplex were unsuccessful because both enzymes contain a nonspecific 3′→5′ exonuclease activity (29). Previously we have isolated Nfo-H69A mutant that contains reduced metal content and lacks NIR and 3′→5′ exonuclease activities in the absence of divalent cations. It also contains lower AP endonuclease/3′-diesterase activities compared with WT Nfo (30). Therefore, we decided to use this Nfo mutant for cocrystallization with DNA because it should not degrade DNA in a nonspecific manner (17).

We succeeded in obtaining a 1.9-Å resolution X-ray structure of Nfo-H69A 15-mer αdA•T complex (Table 3). The electron density maps are of very good quality, and the asymmetric unit contains two identical DNA-Nfo complexes with an rmsd value of 0.13 Å between 279 Cα atoms. The small differences concern four additional residues at the C terminus of molecule A (283 residues) and the disordered terminal DNA base pair in molecule B, whereas the full DNA fragment is well defined in molecule A. Additionally, this structure resembles previously published Nfo:DNA containing AP site structures, such as 1QUM (22) and 2NQJ (23), with respective rmsd values of only 0.32 and 0.27 Å for all defined Cα atoms. Noticeably, DNA fragments present the same large distortion (Fig. 2A), meaning that the WT enzyme, E261Q, and H69A mutants bend the DNA identically by making similar protein–DNA interactions (Fig. 2E). However, the state of the bound DNA in the Nfo-H69A structure was unexpected. Instead of showing the base of αdA placed in the solvent-accessible pocket on the enzyme surface to accommodate its 5′ phosphate in the active site for a catalytically competent complex, as proposed by Hosfield et al. (22), electron density maps show that the αdA site 5′ phosphate is not connected to the 3′ hydroxyl of the preceding nucleotide but is 13.4 Å away from the cytosine C6 (Fig. 2B and Fig. S4). The phosphodiester bond cleavage proves that the Nfo-H69A mutant was able to cleave the DNA backbone 5′ of αdA in the crystallization solution, and thus it can recognize the αdA nucleotide as a target base (Fig. S5A). This cleavage occurred before the formation of the crystal because the structure represents a bound NIR product in a catalytically abortive complex for exonuclease activity (Fig. S5B).

Table 3.

Crystallographic and refinement data for NfoH69A–DNA complex

PDB code	4K1G
Data collection
Beamline	PROXIMA 1 (SOLEIL)
Wavelength (Å)	0.98
Space group	C2221
Cell parameters (Å)	a = 117.9, b = 136.6, c = 112.4
Resolution (Å)	40–1.9 (2.02–1.9)
No. of observed reflections	429,728 (68,483)
No. of unique reflections	71,003 (11,229)
Rsym (%)*	9.5 (68.1)
Completeness (%)	99.7 (98.8)
I/σ	13 (2.5)
Redundancy	6
Wilson B factor	26.58
Refinement statistics
Rcryst (%)†	17.6
Rfree (%)‡	21.2
rms bond deviation (Å)	0.01
rms angle deviation (°)	1.1
Average B (Å2), molecules A; B (no. of atoms)
Protein	24.5; 26.8 (4,366)
Zn ions	20; 23.6 (4)
DNA	46.8; 48.9 (1,182)
Solvent	36.3 (610)
Ramachandran statistics (%)
Preferred regions	98.92
Allowed regions	1.08
Outliers	0

Numbers in parentheses are for the highest-resolution range.

^*Rsym = Σ_hkl Σ_i|I_i(hkl) -<I(hkl)> |/Σ_hkl Σ_i I_i (hkl), where I_i(hkl) is the i^th observed amplitude of reflection hkl and <I(hkl)> is the mean amplitude for all observations i of reflection hkl.

^†R_cryst = Σ ||Fobs| − |Fcalc||/Σ |Fobs|.

^‡5% of the data were set aside for free R-factor calculation.

Fig. 2.

Crystal structure of Nfo:αdA•T-DNA complex. (A) Superposition of the DNA fragment shown in ribbon and bound to Nfo: in magenta the 15 mer containing an αdA•T site in the H69A structure with the 3′-terminal cytosine is cyan, in orange the same 15 mer containing an AP•G site in the inactive E261Q structure (2NQJ.pdb). Nfo is shown in gray surface representation. (B) Close-up view showing the two unpaired C6:G25 in cyan. The αdA•T pair shown in pink is well stacked, making Watson–Crick-like interactions. The cleavage separates the O3′ atom of the cytosine C6 13.4 Å away from its phosphodiester bond. Hydrogen bonds are shown as black dashes. (C) Close-up view of the cytosine C6 in an abortive complex for the intrinsic exonuclease activity of Nfo. Superposition around the DNA cleavage site of the H69A:DNA complex (in magenta except the cytosine C6 in cyan) and the E261Q:DNA complex containing an AP site (in orange). The ribose O3′ atom of C6 rotates by 180° around the phosphate group, moving it 6.4 Å away from its intrahelical position, and points toward the mutated residue H69A. The two present Zn ions are shown in green spheres and that lost in the H69A mutant in a red sphere. Hydrogen bonds are shown as black dashes. (D) Model of S-cdA at the position of the C6 shows steric hindrance: it is in close contact (less than 2 Å) with the phosphate group of the preceding nucleotide. (E) Schematic diagram of NfoH69A–DNA interactions. Polar interactions of DNA–protein side chains and DNA–backbone atoms are shown in blue and black arrows, respectively. The active site Zn²⁺ ions, which bind the phosphate of the C6 nucleotide, are shown as green spheres.

The base pair αdA7•T24 is well stacked in the DNA (Fig. 2B), and the phosphate group of αdA points toward the solvent. The αdA base, which is 180° rotated around the C1’-N9 axis compared with an adenine base makes modified Watson–Crick contacts. Its N6 atom interacts with the O2 and not O4 atom of the partner thymine (Fig. 2B). Its N6 and N7 atoms make hydrogen bonds with Asn35 side chain. In contrast, the preceding base pair (C6:G25) is now unpaired, with both bases being extrahelical. Although G occupies the previously observed position of the flipped-out AP site opposite base in previously published Nfo•DNA complexes, the cytosine C6 occupies the position of the flipped-out AP site in a compact conformation (Fig. 2B). A 2-Å rearrangement of Tyr72 side chain coupled to that of Gln36 allows C6 to adopt this particular position. The base is maintained by two hydrogen bonds; its N4 atom interacts with the main chain carbonyl group of the catalytic E261 residue and with a 5′ phosphate oxygen of the base upstream C5 (Fig. 2C). The C6 ribose O3′ atom, rotated by 180° around the phosphate group, moving it 6.4 Å away from its intrahelical position, points toward the mutated residue H69A and interacts with the E261 side chain (Fig. 2C). The phosphate group is 1.6 Å away from its position to be cleaved. Moving this phosphate group to a position amenable for catalysis leads to a displacement of the whole cytosine compatible with the presence of the His69 (Fig. S6). Thus, in the WT enzyme, we expect that the ribose O3′ atom would make an additional interaction with the His7 side chain, whereas the N4 atom would preserve the interactions observed in the H69A mutant (Fig. S6). Replacing the C6 base by dT, dA, or dG shows that any natural DNA base would be in contact with the 5′ phosphate group of the preceding nucleotide and the E261 CO group.

To assess the effect of S-cdA, we used data from a recently published NMR structure (PDB 2LSF) of DNA containing the lesion (31). In contrast to natural DNA bases, replacing the C6 nucleotide with a S-cdA adduct (referred to as 02I in PDB 2LSF of the NMR in structure) shows that the base would be less than 2 Å from the phosphate group of the preceding nucleotide, resulting in a clash (Fig. 2D). The coordinates of all models are part of the Supporting Information, where Nfo coordinates for Fig. 2D and Fig. S6 presented as Dataset S1; APE1 coordinates for Fig. 3A and Fig. 3C as Dataset S2; S-cdA model for Fig. 2D as Dataset S3; Nfo DNA model for Fig. 3A as Dataset S4; S-cdA model for Fig. 3C as Dataset S5; C6 model for Fig. S6 as Dataset S6.

Fig. 3.

Model of APE1 with 3′-terminal nucleotide. (A) Superposition of the DNA fragment of APE1:substrate (1DEW.pdb in green) and Nfo-H69A:exo (in magenta). Only the part of the DNA fragment of Nfo-H69A that overlays is shown. APE1 is shown in gray surface representation. Because the cytosine C6 (shown in cyan) in exonuclease position in Nfo makes steric clashes in APE1, the cytosine has been rotated around its phosphate group to find an adequate position in APE1 shown in the figure. (B) Close-up view of the modeled cytosine in exonuclease conformation in APE1. Hydrogen bonds are shown as black dashes. (C) Close-up view of a modeled S-cdA adduct in exonuclease position in APE1. Hydrogen bonds are shown as black dashes.

These structural considerations can explain why E. coli Nfo-H69A mutant cannot bind 3′-terminal S-cdAMP nucleotide in cdA•T^rec duplex contrary to nondamaged 3′-terminal dAMP nucleotide in A•T^rec duplex. To test this prediction, we performed an EMSA using the mutant Nfo-H69A protein and 5′-[³²P]-labeled A•T^rec and cdA•T^rec duplexes. The results showed that under the conditions used Nfo-H69A forms a stable DNA–protein complex only with regular A•T^rec duplex but not with a damaged cdA•T^rec one (Fig. S3B). These results are in agreement with our structural model demonstrating that the Nfo-H69A mutant can bind regular nicked DNA duplex in a nonproductive enzyme/substrate complex but loses its affinity to DNA duplex containing 3′-terminal S-cdA adduct at a nick site.

Superimposition of APE1 and Nfo Active Site Structures: Model of Human APE1–3′-Terminal S-cdA Adduct Interactions and Effect of D308A Mutation.

Although Nfo and APE1 have distinct structures, comparison of their active site conformations reveals strong geometric conservation of the catalytic reaction, supporting a unified mechanism for AP site removal from DNA (32). Indeed, both enzymes flip-out the AP site in a similar conformation into the active site pocket. To determine whether the conformation and the position of the 3′-terminal nucleotide for exonuclease activity were the same in Nfo and APE1, we used a structural analysis similar to that described by Tsutakawa et al. (32). Using the common AP site substrate, we superimposed tertiary structures of APE1 (1DEW) (24) and Nfo (2NQJ) (22) in complex with DNA. As described, the superposition based on only the THF moiety in the DNA oriented the respective scissile 5′ phosphates and 3′ ribose oxygen atoms to overlay on top of each other. A portion of the DNA in Nfo:DNA complex (the strand from the 5′-αdA nucleotide to the downstream end and its partner strand) can be overlaid with that in APE1, and the scissile bond superimposes well (Fig. 3A and Fig. S7A). The superposition of our present DNA structure describing the exonuclease activity of Nfo on DNA in APE1 was then straightforward (Fig. S7A). This superposition reveals that the 3′-terminal cytosine C6 cannot be orientated and positioned in APE1 as it is in Nfo because of steric hindrances (Fig. S7B). Indeed, the ribose O3′ atom clashes with the phenol ring of Phe266, the O2 atom is less than 2 Å to the Trp280 side chain, and the N4 atom is at similar distances (less than 2 Å) from the carbonyl of N212 and the Cα of Gly231. Rotating the cytosine around the phosphate group (180° along the P-O5′ axis) leads to a unique position of C6 for APE1 exonuclease activity (Fig. 1A and Fig. S7C). To optimize this position, the Phe266 side chain needs to slightly reorient. Our model confirms the biochemical data showing that Phe266 mutations increase APE1-catalyzed 3′→5′ exonuclease activity (33). In our model, the ribose O3′ atom is shifted by 3.3 Å away from its intrahelical position and interacts with the Asn174 side chain and the carbonyl group of Ala130 (Fig. 3B). The N4 and O2 atoms of C6 hydrogen bond the Asp308 and Arg177 side chains, respectively. Replacing the C6 base by dA or dG shows that a regular purine could be adapted with a rearrangement of the Arg177 side chain and would be involved in a contact with Thr268. An S-cdA nucleotide at this position can also be accommodated making the same protein interactions (Fig. 3C). However, a close contact (around 2 Å) with Asp308 would force the side chain to slightly move. Therefore, Asp308 would have little or no effect on the removal of S-cdAMP. In agreement with this, the D308A mutant is capable of removing the 3′-terminal S-cdA, THF, and phosphoglycolate (PG) adducts almost as efficiently as WT APE1 (Table 2 and Fig. S8). As shown in Table 2, the k_cat/K_M values for the APE1 D308A-catalyzed excision of 3′-terminal S-cdA, THF, and dA nucleotides were 5-, 7-, and 280-fold lower compared with that of WT APE1, respectively, indicating that D308 residue is very important for the efficient removal of 3′-terminal regular nucleotides but not that of S-cdA in agreement with our structure-based model of APE1-DNA exo interactions.

In Vitro Reconstitution and Repair of DNA Containing S-cdA Adducts in E. coli, Yeast, and Human Cell-Free Extracts.

Removal of 3′-terminal S-cdAMP by AP endonuclease-catalyzed hydrolysis of the phosphodiester bond on the 5′ side of the damaged nucleotide should generate 3′-OH termini and enable DNA synthesis. To examine this, we reconstituted the repair of 5′-[³²P]-labeled A•T^rec and cdA•T^rec oligonucleotide duplexes in the presence of the purified APE1 and DNA polymerase β (POLβ) and dNTPs. Incubation of A•T^rec with POLβ generated elongated products up to the full-length 40-mer product after 30 min (Fig. S9). Interestingly, addition of APE1 resulted in the appearance of degradation products with size less than 21 mer owing to its processive 3′→5′ exonuclease activity. The presence of 3′-terminal S-cdA adduct in cdA•T^rec duplex strongly inhibited the first nucleotide insertion and completely blocked POLβ-catalyzed strand elongation from 3′-terminal S-cdAA. As expected, addition of APE1 resulted in the removal of 3′-blocking S-cdAMP and allowed the elongation and restoration of the damaged recessed DNA duplex by POLβ (Fig. S9). These results indicate that efficient elongation of DNA primers containing 3′-blocking bulky S-cdA nucleotide can be reactivated by the 3′-repair cleansing function APE1.

Data obtained with the purified AP endonucleases show that S-cdA nucleotide at the 3′ end are efficiently repaired by E. coli Xth and human APE1 enzymes but not by E. coli Nfo and yeast Apn1. To address the role of the AP endonuclease-catalyzed 3′ repair cleansing activity in a more physiological context, we examined repair of 5′-[³²P]-labeled cdA•T^rec duplex in cell-free extracts from E. coli, S. cerevisiae, and HeLa cells lacking Nfo/Xth, Apn1, and APE1, respectively (Fig. 4). As expected, we observed robust nonspecific 3′→5′ exonuclease activities on regular A•T^rec duplex in cell-free extracts from HeLa (Fig. 4, lane 2), S. cerevisiae WT (lane 4), and E. coli WT (lane 17). The nonspecific exonuclease activity is strongly decreased in S. cerevisiae apn1 (lane 5) and E. coli xth (lane 19) but not in siRNA-APE1 silenced HeLa cells (lane 3) and nfo (lane 18) compared with extracts from WT cells. As expected, cell-free extracts from HeLa and E. coli WT and nfo efficiently remove 3′-terminal S-cdA nucleotide in cdA•T^rec duplex (lanes 7, 21, and 22). Importantly, extracts from siRNA-APE1 silenced HeLa cells and E. coli xth still contain exonuclease activity on A•T^rec (lanes 3 and 19) but have lost the activity on cdA•T^rec duplexes (lanes 8 and 23). No activity was observed on cdAA•T^rec duplex in the extracts from HeLa and E. coli cells (lanes 12–13 and 25–27). Interestingly, the extracts from S. cerevisiae WT can remove with very low efficiency S-cdAMP in cdA•T^rec duplex (lane 9) and 3′-terminal dAMP in cdAA•T^rec duplex (lane 14), and this weak activity was not observed in apn1 extracts (lanes 10 and 15), thus confirming results obtained with the purified protein. These results suggest that in yeast extracts S-cdA adducts present in DNA duplex could be removed, albeit inefficiently, by combined action of Apn1 and nonidentified factors. Taken together the results obtained with cell-free systems confirm the role of APE1, Apn1, and Xth proteins in the repair of bulky S-cdA adduct when present next to a DNA strand break in human, yeast, and E. coli cells.

Fig. 4.

3′ repair activities in cell-free extracts of human, yeast, and E. coli cells. 5′-[³²P]-labeled recessed oligonucleotide duplexes (5 nM) were incubated with 1 µg of cell-free extract for 10 min at 37 °C. The “X” denotes the position of S-cdA nucleotide. Details in Materials and Methods.

Discussion

Structurally unusual helix-distorting cdA and cdG adducts in DNA are biologically important owing to their strong inhibitory effect on DNA replication and transcription. The presence of C8-C5′ covalent bond prevents repair of cdPu by DNA glycosylase-mediated excision and direct damage reversal. Therefore, cells use the NER system to remove cdPu adducts in vitro and in vivo (4, 8, 9). However, ionizing radiation, breakage of replication forks stalled at bulky DNA lesions, and/or misincorporation of oxidatively damaged precursors during DNA synthesis can generate cdPu lesions located in close proximity to strand breaks, making them resistant to NER. Therefore, alternative repair pathways may exist to remove these endogenous helix-distorting DNA lesions at DNA termini.

Here, we investigated the mechanism of 3′→5′ exonuclease activity of AP endonucleases involved in the BER and NIR pathways and their ability to recognize S-cdA adducts in DNA duplex. We have shown that the S-cdA nucleotide, when present in fully duplex DNA, is not a substrate of the AP endonuclease-catalyzed NIR activity, but it can be a substrate of AP endonuclease-catalyzed 3′→5′ exonuclease activity when present at the 3′ end of single-stranded DNA break. E. coli Xth and human APE1, but not E. coli Nfo and yeast Apn1, remove S-cdA adducts at 3′ termini of recessed DNA duplex with high efficiency. However, when the S-cdA nucleotide is located 1 or more nt away from the 3′ termini of a DNA duplex, it strongly blocks the 3′→5′ exonuclease activity of all AP endonucleases tested. Our EMSA data suggest that APE1 fails to bind DNA duplexes with S-cdA adduct located at the second position from the 3′ end of a gap. Taken together these results establish that 3′-terminal S-cdA adduct in gapped DNA duplex can be removed by an alternative mechanism distinct from NER that involves 3′→5′ exonuclease function of the Xth family AP endonucleases.

It should be stressed that Xth and APE1 contain two distinct activities toward DNA strand breaks: a 3′ repair diesterase function that catalyzes removal of 3′-dRP groups (34), and 3′→5′ exonuclease, which catalyzes nonspecific removal of regular and oxidized bases (19, 20). Comparison of APE1 exonuclease activities on recessed, gapped, and nicked DNA duplexes as well as kinetic data for recessed DNA substrates revealed that APE1 was much more efficient on 3′-terminal S-cdA compared with a regular dA nucleotide. This strongly suggests that APE1-catalyzed 3′ cleansing activities are highly specific to damaged DNA. Interestingly, APE1 also removes mismatched nucleotides from the 3′ terminus of DNA much more efficiency than nucleotides from matched pairs (26–28). It was suggested that the efficiency of APE1’s proof-reading exonuclease activity depends primarily on the melted conformation of the 3′ end of DNA duplex at the site of mismatch (26). Therefore, we propose that helix-distorting S-cdA adduct at the 3′ end would resemble a melted duplex conformation similar to that of a 3′-terminal mismatch base pair and thus would facilitate the recognition of the adduct by APE1.

At present there are no structural data available on AP endonuclease-catalyzed 3′→5′ exonuclease function. Here, we solved the crystal structure of Nfo mutant with nicked DNA duplex. Initially, the crystal structure of the Nfo-H69A mutant in complex with a 15 mer containing a αdA•T base pair at position 7 was expected to reveal the molecular mechanism of the NIR activity because the H69A mutant was previously shown to be inactive for both NIR and 3′→5′ exonuclease functions owing to a loss of one Zn atom (30). Surprisingly, the cocrystal structure showed a cleaved αdA•T duplex (the expected product of WT Nfo). The NIR product with the 3′-terminal cytosine C6 bound to the enzyme’s active site forms a nonproductive complex with the Nfo-H69A mutant, which is unable to perform exonuclease activity. This structure of the Nfo-H69A:DNA complex with the 3′-terminal cytosine C6 bound to the enzyme active site is very interesting and informative for understanding the mechanism of 3′→5′ exonuclease activity of the WT Nfo, because both mutant and WT enzymes behave identically. Indeed, both Nfo proteins recognize an AP site and 3′-terminal nucleotide by binding and distorting two DNA substrates by 90° and flipping-out the base opposite the target AP site or the 3′-terminal base (Fig. 2 and Fig. S6). Because the 3′-end of C6 nucleotide overlaps with the AP site, a slight shifting of the C6 nucleotide to bring it in an amenable position for catalysis was accomplished by superimposing the complexed H69A and WT Nfo structures. It is important to note that the catalytic 5′-phosphate group in the resulting model stays in the same scissile position for both AP endonuclease and 3′-exonuclease activities of Nfo, ensuring the hydrolysis of phosphodiester bond. Replacing C6 by either A, G, or T nucleotide was also straightforward. All these regular nucleotides can be accommodated in the WT Nfo active site, thus explaining the structural basis of nonspecific 3′ exonuclease function of Nfo. It should be stressed that the compact and curved position of the 3′-terminal C6 nucleotide, which interacts with the phosphate group of an upstream C5 nucleotide, is not compatible with the presence of a S-cdA, which will generate steric hindrance in the enzyme’s active site. Therefore, the inability of Nfo to remove 3′-terminal S-cdA adduct comes from an inadequate accommodation of this nucleotide in the enzyme active site pocket. In agreement with previous data (30), the complex of Nfo with cleaved DNA well illustrates that after incision of the duplex 5′ next to the lesion site by NIR activity, Nfo proceeds further to degrade DNA by its 3′→5′ exonuclease activity at the site of the nick and that this latter function fails when the imidazole group of His69 and Zinc-1 atom are absent.

Recent work by Tsutakawa et a. (32) demonstrated that tertiary structures of APE1 and Nfo in complex with DNA can be superimposed. Therefore, the structure of Nfo-H69A:DNA complex can serve as an excellent template for modeling of interactions between active site of APE1 and DNA to provide the structural basis for the enzyme's 3′ repair activities. Using superimposed active site structures of APE1:AP site-DNA and Nfo:AP site-DNA complexes, we have been able to observe how the C6 nucleotide in Nfo would be positioned in APE1. The conformation of C6 nucleotide adopted in the Nfo’s active site results in steric clashes with F266, W280, G231, and CO Asn212 amino acid residues of APE1. To avoid steric hindrance and to accommodate C6 into the APE1 active site, we rotated the nucleotide around its 5′ phosphate group, fixing it in catalytic position. In the resulting model, N174 interacts with the ribose O3′ atom of C6, whereas D308 interacts with the C6 base moiety and maintains the 3′-terminal nucleotide in a position favorable for exonuclease activity of APE1. Most interestingly, replacing the 3′ end C6 by an S-cdA nucleotide creates no steric hindrance except a small rearrangement of D308 side chain to accommodate the bulky adduct. The present model of APE1 interactions with 3′-terminal nucleotide adducts provides the structural basis for the observed 3′ repair exonuclease activity toward regular nucleotides and S-cdA adduct and reveals the role of D308 residue. In agreement with this model, we showed that APE1-D308A mutant can remove S-cdA adduct in cdA•T^rec duplex with good efficiency, similar to that of WT APE1 (Table 2), indicating that APE1 removes the damaged bulky nucleotide by its 3′ repair cleansing function. Furthermore, we have previously demonstrated that D308A mutant has dramatically decreased 3′→5′ exonuclease activity compared with WT APE1 (35).

The lack of exonuclease activity toward S-cdA adduct in cdAA•T^rec duplex in all AP endonuclease tested observed in our studies is very intriguing. To examine the structural basis of this exonuclease resistance, we used the recently published NMR structure of duplex DNA containing S-cdA nucleotide (31) to superimpose it onto DNA bound to APE1. As described, the presence of S-cdA nucleotide at position 6 in DNA duplex results in the decrease of the phosphate oxygen atom (from residue at position 7)-O5′ (from S-cdA) bond distance that shortens by approximately 1 Å (31), which could explain the lack of exonuclease activity toward S-cdA nucleotide in cdAA•T^rec duplex in all AP endonucleases tested. Therefore, the shorter distance of 6.3 Å between the phosphate group of S-cdA and that of the 3′ nucleotide instead of 7.05 Å between two regular nucleotides probably leads to an incorrect positioning of the scissile phosphodiester bond in the AP endonuclease active site and consequently to the loss of both 3′ repair activities and stable binding to cdAA•T^rec duplex. Finally, although the conserved structural chemistry of active sites of APE1 and Nfo supports a unified mechanism for the AP site cleavage in DNA, our structural models reveal that this mechanism cannot be extended to APE1 and Nfo exonuclease activities. The conformation and position of the 3′-terminal nucleotide for exonuclease activity differs between these two enzymes and thus involves different protein interactions. This also leads to a different DNA substrate specificity of their respective exonuclease functions. In conclusion, the structural data described in the present work provide insight into the mechanism of 3′ cleansing activity of the Nfo and Xth families of AP endonucleases.

Ionizing radiation and certain anticancer drugs generate complex or “dirty” DNA strand breaks containing 3′ end proximal damaged bases, which are poorly repaired by classic BER and NER pathways (21, 36). Here we demonstrate that APE1-catalyzed removal of 3′-terminal S-cdA nucleotide in recessed DNA duplex enables otherwise blocked DNA polymerase synthesis in vitro, pointing to a possible role of APE1 in cleansing of complex and/or dirty DNA strand breaks (Fig. 2). To further substantiate physiological relevance of this specific repair function of AP endonucleases, we demonstrated the presence of the 3′ repair activities in cell-free extracts toward DNA duplexes containing a 3′-terminal S-cdA adduct (Fig. 4). Importantly, under the experimental condition used we did not observe significant repair of cdAA•T^rec duplex in any cell-free extracts tested. Interestingly, we have identified a weak Apn1-independent activity on cdA•T^rec duplex in yeast extracts and also found that extracts from WT cells and the purified Apn1 can remove a regular dA nucleotide in cdAA•T^rec duplex, albeit with low efficiency. Thus, S. cerevisiae contains two enzymes: Apn1 and unknown exonuclease that can remove S-cdA adducts when located 1 or more nt away from the 3′ end of strand breaks.

Tirapazamine is an experimental bioreductively activated anticancer drug that selectively kills cells under hypoxia (37). It has been demonstrated that in vitro tirapazamine mediates formation of 8,5′-cyclopurine-2'-deoxynucleosides in DNA under hypoxic condition, suggesting that these lesions may contribute to the drug’s cytotoxicity (38). It is tempting to speculate that tirapazamine treatment may generate DNA strand breaks containing cdPu adducts and that this specific APE1 activity on the 3′-terminal lesions described in the present work could provide a rationale for the use of APE1 inhibitors to enhance effectiveness of tirapazamine and perhaps other anticancer drugs. Previously, it was shown that APE1 removes β-l-Dioxolane-cytidine (l-OddC), a nonnatural stereochemical l-nucleoside analog, when incorporated at the 3′ terminus of duplex DNA by DNA polymerases (39). APE1 was also shown to remove 3′-tyrosyl residues from the recessed and nicked DNA duplexes, suggesting its potential role in the processing of covalent topoisomerase I:DNA complexes generated by anticancer drugs (27). Recently it has been demonstrated that 5′S isomer of cdATP could be incorporated more efficiently than the 5′R isomer by replicative DNA polymerases during DNA synthesis (14). On the basis of the results reported here, we propose that APE1 may act in a similar manner on above adducts and also on 3′-terminal S-cdA nucleotides resulting from DNA polymerase catalyzed misincorporation. Such an activity would be of relevance to the proposal that S-cdA could be used as an anticancer and/or antiviral drug (14).

Finally, recent studies from the Dizdaroglu laboratory (40) have found that both free R-cdA and S-cdA nucleosides can be detected in human urine. Whether this material derived from nuclease digestion of cdA-containing oligonucleotides resulting from NER, and/or APE1-catalyzed repair, and/or digestion of DNA from dead cells remains to be determined. However, on the basis of the current results, an intriguing possibility is that AP endonucleases play a key role in the generation of urinary cdA, which could be a biomarker of endogenous oxidative stress to DNA.

Materials and Methods

Oligonucleotides, Proteins, and Antibodies.

Sequences of the oligonucleotides and their duplexes used in the present work are shown in Table 1. All oligonucleotides were purchased from Eurogentec, including regular oligonucleotides and those containing S-cdA, αdA, THF, and PG. Before enzymatic assays oligonucleotides were either 5′-end-labeled by T4 polynucleotide kinase (New England Biolabs, Ozyme) in the presence of [γ-³²P]-ATP (3,000 Ci/mmol-1) (PerkinElmer), or 3′-end-labeled by terminal deoxynucleotidyl transferase (New England Biolabs) in the presence of [α-³²P]-3′-dATP (Cordycepin 5′-triphosphate, 5,000 Ci/mmol-1) (PerkinElmer) as recommended by the manufacturers. Radioactively labeled oligonucleotides were desalted with a Sephadex G-25 column equilibrated in water and then annealed with corresponding complementary strands for 3 min at 65 °C in a buffer containing 20 mM Hepes-KOH (pH 7.6) and 50 mM KCl.

The sequence of the 15-mer DNA duplex used for crystallization assays is d(GCGTCCXCGACGACG)/d(CGTCGTCGTGGACGC), where X is αdA. The oligonucleotides were hybridized by mixing equal concentrations (10 mM) in 2 mM Tris·HCl (pH 7.0) heated to 65 °C for 3 min and cooled down to room temperature over 2 h. The MALDI-TOF mass spectrometry analysis of the oligonucleotides performed by the manufacturer validated their size and homogeneity. In addition the purity and integrity of the oligonucleotide preparations were verified by denaturing PAGE. The siRNA sequences used to decrease APE1 in HeLa cells have been taken from previously described studies (41). The siRNA specific to mouse major AP endonuclease, mAPEX, was used as negative control in both cases.

All AP endonucleases, their mutants, and human DNA glycosylase Neil1 used in the present study were expressed and purified in their native form without tags or other modifications as described previously (16, 17). The purified human DNA POLβ and T4 DNA polymerase were purchased from Trevigen and New England Biolabs, respectively.

Strains, Extract Preparation, Cell Culture, and Silencing of APE1 Expression.

AB1157 [IeuB6 thr-1 Δ(gpt-proA2) hisG4 argE3 lacY1 gaIK2 ara-14 mtl-1 xyl-5 thi-1 tsx-33 rpsL31 supE44 rac] (WT) and its isogenic derivatives BH130 (nfo::kan^R) and BH110 (nfo::kan^R [Δ(xth-pncA)90 X::Tn10]) were from the laboratory stock (42). S. cerevisiae FF18733 WT strain (MATa his7-3 leu2-1,112 lys1-1 trp1-289 ura3-52) and its isogenic derivative BG1 (apn1Δ::HIS3) were kindly provided by S. Boiteux (French Alternative Energies and Atomic Energy Commission, France).

Crude cellular extracts from E. coli, S. cerevisiae, and HeLa cells with down-regulated APE1 expression were prepared as described previously (16, 20, 43).

DNA Repair Assays.

The 3′-phosphodiesterase/exonuclease activity assay of APE1 was performed in the standard reaction mixture (20 μL) containing 5 nM of [³²P]-labeled oligonucleotide duplexes, 50 mM KCl, 20 mM Hepes·KOH (pH 6.8), 0.1 mg/mL BSA, 1 mM DTT, 1 mM MgCl₂, and a limited amount of pure protein or extract; when measuring repair activities in human cell-free extracts BSA and DTT were omitted. For bacterial cell-free extracts and the purified Xth and Apn1 proteins, the standard reaction mixture (20 μL) contained 5 nM of [³²P]-labeled DNA substrate, 50 mM KCl, 20 mM Hepes·KOH (pH 7.6), 0.1 mg/mL BSA, and 5 mM MgCl₂, except when incubating with Nfo, when MgCl₂ was omitted from the buffer.

The reactions were stopped by adding 10 μL of a solution containing 0.5% SDS and 20 mM EDTA and then desalted by hand-made spin columns filled with Sephadex G25 (Amersham Biosciences) equilibrated in 7.5 M urea. Purified reaction products were separated by electrophoresis in denaturing 20% (wt/vol) polyacrylamide gels (7 M urea, 0.5× tris-borate-EDTA buffer, 42 °C). Gels were exposed to a Fuji FLA-3000 Phosphor Screen and analyzed using Image Gauge V3.12 software.

The kinetic parameters for exonuclease activity of APE1 were measured as described previously (35). Briefly, 2–1,000 nM of duplex oligonucleotide substrate was incubated with 0.2 nM APE1 under standard reaction conditions, the linear velocity was measured, and the K_M and k_cat constants were determined from Lineweaver-Burk plots. All biochemical experiments were performed independently and repeated at least three times.

Crystallographic Analysis.

The E. coli Nfo-H69A mutant was cloned, expressed, and purified as previously described (17). In crystallization trials, the 15-mer DNA duplex was mixed with Nfo-H69A used at a concentration of 18 mg/mL in a buffer containing 50 mM KCl and 20 mM Hepes·KOH (pH 7.6) in a 2:1 stoichiometry. Commercial crystallization solutions (Qiagen kits) were screened in sitting-drop vapor diffusion experiments using a nanodrop Cartesian robot (Proteomic Solutions) at 293 K. One condition [number 88: 30% (wt/vol) PEG 4000, 0.1 M Tris·HCl (pH 8.5), and 0.2 M MgCl₂] in the Classics suite was manually optimized with home-made solutions in hanging drops composed of 1:1 volume ratio of crystallization solution and of Nfo-H69A:DNA complex. Crystals obtained in 0.1 M Tris·HCl (pH 8.0), 12% (wt/vol) PEG 4000, and 200 mM MgCl₂ were flash-frozen in a cryo-protecting solution consisting of the mother solution supplemented with 20% (wt/vol) PEG 400.

X-ray diffraction data were collected at 100 K on beamline PROXIMA I at SOLEIL, and intensities were integrated using XDS¹⁹. The asymmetric unit can contain two complexes of Nfo-H69A:DNA, corresponding to a Matthews coefficient (44) of 2.78 ų Da⁻¹ and a solvent content of 55.8%. The phase problem was solved by molecular replacement using the program PHASER (45) and an Nfo mutant:DNA structure (PDB code 2NQJ) as a search model. The resulting atomic model was refined using BUSTER (46) and manually improved using COOT²³. Data collection and refinement statistics are given in Table 3. His-109 is not well defined in the electron density owing to the absence of the Zn1 ion.