Oxidatively Generated Guanine(C8)-Thymine(N3) Intrastrand Cross-links in Double-stranded DNA Are Repaired by Base Excision Repair Pathways*

Background: Base excision repair (BER) is the major pathway for repair of single oxidized nucleobases. Results: Bifunctional DNA glycosylases and AP endonucleases are able to remove cross-linked guanine in guanine(C8)-thymine(N3) intrastrand cross-links. Conclusion: BER pathways can repair the intrastrand cross-links. Significance: Oxidatively generated intrastrand cross-linked DNA lesions can be repaired in HeLa cell extracts not only by nucleotide excision repair, but also by multiple BER pathways. Oxidatively generated guanine radical cations in DNA can undergo various nucleophilic reactions including the formation of C8-guanine cross-links with adjacent or nearby N3-thymines in DNA in the presence of O2. The G*[C8-N3]T* lesions have been identified in the DNA of human cells exposed to oxidative stress, and are most likely genotoxic if not removed by cellular defense mechanisms. It has been shown that the G*[C8-N3]T* lesions are substrates of nucleotide excision repair in human cell extracts. Cleavage at the sites of the lesions was also observed but not further investigated (Ding et al. (2012) Nucleic Acids Res. 40, 2506–2517). Using a panel of eukaryotic and prokaryotic bifunctional DNA glycosylases/lyases (NEIL1, Nei, Fpg, Nth, and NTH1) and apurinic/apyrimidinic (AP) endonucleases (Apn1, APE1, and Nfo), the analysis of cleavage fragments by PAGE and MALDI-TOF/MS show that the G*[C8-N3]T* lesions in 17-mer duplexes are incised on either side of G*, that none of the recovered cleavage fragments contain G*, and that T* is converted to a normal T in the 3′-fragment cleavage products. The abilities of the DNA glycosylases to incise the DNA strand adjacent to G*, while this base is initially cross-linked with T*, is a surprising observation and an indication of the versatility of these base excision repair proteins.

Reactions of reactive intermediates such as free radicals and oxidizing agents with DNA can give rise to interstrand and intrastrand cross-linked DNA lesions. Interstrand cross-linked DNA lesions (ICL) 6 that result from reactions of a variety of bifunctional agents such as cisplatin (1) with DNA, have received much attention because they are difficult to remove by DNA repair mechanisms and are therefore highly genotoxic (2,3). While ICL lesions arise from the covalent coupling of two nucleotides on opposite DNA strands, the coupling of two nucleotides on the same strand gives rise to intrastrand crosslinked (IntraCL) lesions. Well known examples are the UV photo-induced cyclobutanepyrimidine dimers (CPD) and 6 -4 photoproducts (4). Oxidatively generated IntraCL lesions include cross-links between the C8-atom of guanine and a 3Ј-adjacent 5-methyl group of thymine (G[8 -5m]T) (5), and the analogous G [8 -5]C (6), and G[8 -5m]C (7) intrastrand tandem lesions. Other forms of intrastrand lesions are the intranucleotide 8,5Ј-cyclo-2Ј-deoxypurine lesions that are characterized by covalent linkages between the C5Ј deoxyribose and C8 carbon atoms of adenine (cdA) or guanine (cdG); these types of lesions were first discovered in dilute DNA solutions exposed to ␥-radiation (8,9).
Recently, the novel guanine(C8)-thymine(N3) tandem lesions (G*[C8-N3]T*) were identified in vitro (10) and detected in human HeLa cells by isotope dilution LC-MS/MS methods (11). In these IntraCLs, guanine and thymine bases are either adjacent to one another (G*T*) or separated by one intervening cytosine (G*CT*) linked by a covalent bond between the C8(G) and N3(T) atoms (Fig. 1A). Like the cdG and cdA lesions embedded in double-stranded DNA (12)(13)(14)(15)(16), the G*T* and G*CT* lesions are moderate to good substrates, respectively, of the human nucleotide excision repair (NER) system in extracts from HeLa cells (17). Furthermore, evidence was presented that both types of IntraCL lesions are also incised at the sites of the lesions. The origins of these incisions were not further investigated (17), but suggested that a base excision repair (BER) pathway could have been responsible for these incisions. However, BER repair enzymes are not known to incise intrastrand DNA cross-links. Here, we considered whether conventional BER pathways (18 -20) and/or the nucleotide incision repair (NIR) pathway (21,22) could account for the non-NER incisions reported earlier by Ding et al. (17).
To gain insights into these mechanisms of incision of the G*CT* and G*T* IntraCLs embedded in double-stranded DNA, we incubated the 17-mer double-stranded constructs with different Escherichia coli and human DNA glycosylases/AP lyases Nei and NEIL1 (endonuclease VIII and oxidized pyrimidinespecific DNA glycosylase), Nth and NTH1 (endonuclease III and thymine glycol-DNA glycosylase), respectively, E. coli Fpg (8-oxoguanine DNA glycosylase), and E. coli Nfo, Saccharomyces cerevisiae Apn1 and human APE1 proteins, and monitored the formation of incision products. We demonstrate here that the bacterial, yeast, and human bifunctional DNA glycosylases and AP endonucleases cleave the strands adjacent to the G*CT* and G*T* IntraCL (Fig. 1A) embedded in site-specifically modified oligonucleotide duplexes. Analysis of the cleavage products by denaturing polyacrylamide gel electrophoresis (PAGE) and MALDI-TOF/MS methods showed that the DNA glycosylases/AP lyases excise the cross-linked guanine and cleave the resulting abasic sites via ␤and ␤,␦-elimination mechanisms. In turn, the AP endonucleases of E. coli Nfo, yeast Apn1 and human APE1 cleave the duplex DNA containing G*[C8-N3]T* lesions 5Ј next to cross-linked guanine initiating the nucleotide incision repair (NIR) pathway (22).

Experimental Procedures
G*T* and G*CT* Duplexes and Proteins-The 17-mer oligonucleotides containing the G*T* or G*CT* lesions were synthesized as described by Crean et al. (10). Before the enzymatic assays, oligonucleotides were either 5Ј-end-labeled by T4 polynucleotide kinase (New England Biolabs) in the presence of [␥-32 P]ATP (3,000 Ci/mmol, PerkinElmer-Life Science Research), or 3Ј-end-labeled by terminal deoxynucleotidyl transferase (TdT, New England Biolabs) in the presence of 3Ј-[␣-32 P]dATP (cordycepin 5Ј-triphosphate, 5,000 Ci/mmol) employing the standard protocols. The radioactively labeled oligonucleotides were desalted on a Sephadex G-25 column equilibrated in water and then annealed to the required complementary strand for 3 min at 65°C in a buffer containing 20 mM Hepes-KOH (pH 7.6) and 50 mM KCl, and then slowly cooled to room temperature. The purified E. coli uracil-DNA glycosylase (Ung), Fpg, Nth, and human NTH1 and APE1 proteins were from laboratory stock and prepared as described (23). Purifications of the E. coli Nfo and S. cerevisiae Apn1 proteins were performed as described (22,24). The expression vectors phNEIL1 (25) and pET24b-EndoVIII (26) were generously provided by Drs. Hiroshi Ide (Hiroshima University, Japan) and Dmitry Zharkov (ICBFM, Novosibirsk, Russia), respectively. The E. coli Nei and full-length native NEIL1 proteins were purified as described (25,26).
Repair Assays-The standard reaction mixtures (20 l) contained 5 nM concentrations of the 32 P-labeled G*T* or G*CT* duplexes and 50 nM of the purified repair protein, were incubated for 30 min at 37°C unless stated otherwise. The DNA repair activities of APE1 protein were tested either in the NIR buffer, which is optimal for the nucleotide incision activity and contained 50 mM KCl, 20 mM Hepes-KOH (pH 6.9), 0.1 mg/ml BSA, 1 mM DTT, and 0.1 mM MgCl 2 , or in the BERϩMg 2ϩ buffer, which is optimal for the AP endonuclease activity, containing 50 mM KCl, 20 mM Hepes-KOH (pH 7.6), 0.1 mg/ml BSA, 1 mM DTT, and 5 mM MgCl 2 . The same buffer was used for the S. cerevisiae Apn1 protein whereas, for the E. coli Nfo protein, MgCl 2 was omitted. The incision activities were deter-G*CT* 5'-CCACCAACG*CT*ACCACC-3' 3'-GGTGGTTGC GA TGGTGG-5' Base excision repair (BER) Nucleotide incision repair (NIR) mined from the amount of cleaved oligonucleotide substrates. The DNA glycosylase activity was measured in the BERϩEDTA buffer containing 5 nM of an oligonucleotide duplex, 50 mM KCl, 20 mM Hepes-KOH (pH 7.6), 0.1 mg/ml BSA, 1 mM DTT, 1 mM EDTA, and 50 nM of the required purified protein for 30 min at 37°C unless stated otherwise. All reactions were terminated by adding 10 l of a stop solution containing 0.5% SDS and 20 mM EDTA, and then desalted by hand-made spin-down columns filled with Sephadex G25 (Amersham Biosciences) equilibrated in 7.5 M urea. The purified reaction products were separated by electrophoresis in denaturing 20% (w/v) polyacrylamide gel (7.5 M urea, 0.5ϫ TBE, 42°C). The gels were exposed to a Fuji FLA-3000 Phosphor Screen, then scanned with Fuji FLA-3000 or FLA-9500, and analyzed using Image Gauge V4.0 software. MALDI-TOF/MS Analysis of the Cleavage Products-Mass spectrometry measurements were performed as described previously (45). Typically, 10 pmol of the non-labeled lesion-containing oligonucleotide duplexes (in 100 l) were incubated with repair proteins (100 nM) in the relevant buffer solution at 37°C either for 1 h with APE1 or for 15 min with NEIL1 and NTH1. The reaction products were precipitated with 2% lithium perchlorate in acetone, desalted, and then dissolved in water prior to analysis by MALDI-TOF/MS. The mass spectra were obtained in the negative mode on a time-of-flight Microflex mass spectrometer (Bruker), equipped with a 337-nm nitrogen laser and pulsed delay source extraction. The matrix was prepared by dissolving 3-hydroxypicolinic acid in 10 mM ammonium citrate buffer. Prior to MALDI-TOF mass analysis, oligonucleotide solutions were purified and concenterd on Zip-Tip pipette tips (Millipore). A mixture of purified DNA sample (10 pmol; 1 l) was added to matrix (1 l) and spotted on a polished stainless target plate using the dried droplet method. Spectra were calibrated using reference oligonucleotides of known masses.

Results
Characterization of Duplexes with G*[C8-N3]T* Crosslinks-Such duplexes are referred to as G*CT* and G*T*, depending on the nature of the cross-linked lesion (Fig. 1A). The 17-mer sequences used are defined in Fig. 1A. The differences in molar mass between the G*CT* and G*T* duplexes and the normal duplexes without these cross-links are only 2 Da. Since the absorption spectra of these cross-linked and normal duplexes are identical, it is not straightforward to verify the presence of these lesions (10). While mass spectrometry combined with two-dimensional 13 C NMR yields definitive characterization (10), a more simple approach is to monitor the stepwise degradation of the single-stranded oligonucleotides using exonucleases that stall at the sites of the cross-linked nucleotides (27). The MALDI-TOF/MS analysis of the digestion products of G*CT* and G*T* duplexes generated by snake venom phosphodiesterase 1 (digestion from the 3Ј-end) and calf spleen phosphodiesterase 2 (digestion from the 5Ј-end) clearly demonstrated the presence and integrity of the intrastrand crosslinks in these duplexes (Fig. 2).
Overall Approach-In this work, the 17-mer duplexes containing the G*CT* and G*T* lesions were exposed to either bifunctional DNA glycosylases (BER) or AP endonucleases that are involved in the NIR pathway. In the case of BER, DNA glycosylases recognize the lesions and then cleave the N-glycosyl bond releasing the damaged base, thus producing abasic sites (18) (Fig. 1B). Bifunctional DNA glycosylases, in addition to base excision, also exhibit AP lyase activity and cleave the abasic sites formed, leaving single-strand breaks flanked by 5Ј-phosphates and either 3Ј-phosphate groups (generated by ␤,␦-elimination), or a 3Ј-␣,␤-unsaturated aldehyde (PUA) group (␤-elimination). In the NIR pathway, AP endonucleases incise the phosphodiester bond adjacent to and on the 5Ј-side of the damaged bases in a DNA glycosylase-independent manner, leaving single-strand breaks flanked by proper 3Ј-OH termini for subsequent primer extension, and 5Ј-damaged nucleotides in the 3Ј-downstream cleavage fragments (21) (Fig. 1B). These cleavage products, generated by BER or NIR mechanisms, were distinguished by denaturing PAGE and identified by MALDI-TOF/MS methods (18,28).
In the case of the 3Ј-labeled G*CT* duplexes, the modified strands are longer by one nucleotide because of the 3Ј-endlabling method. The bifunctional DNA glycosylases NEIL1, Nei, and Fpg yield fast-migrating ϳ9-mer fragments with a 5Ј-phosphate end that are consistent with cleavage of the G* site (Fig. 3C, lanes 1-3). The NTH1 and Nth glycosylases also yield the same fast-migrating 3Ј-end-labeled cleavage fragments (Fig. 3B, lanes 7 and 8) since the bulky PUA groups remain on the 5Ј-unlabeled fragment rather than on the 3Ј-endlabeled one (29) according to the cleavage mechanism (Fig. 1B). As expected from the established mechanism of action of NIR-AP endonucleases, Apn1, APE1, and Nfo generate slower migrating DNA cleavage fragments (Fig. 3C, lanes 4 -6). Based on their mobilities, these fragments could be identified either as 10-mers with a phosphorylated damaged guanine residue G* at its 5Ј-end (21), or as slower migrating 9-mer fragments with a 5Ј-OH end, which is more consistent with the MALDI-TOF/MS results (see below). It should be noted that all three NIR-AP endonucleases exhibit nonspecific 3Ј35Ј exonuclease activity (32,33), which degrades 5Ј-and 3Ј-end-labeled 17-mer and 18-mer oligonucleotides, respectively (Fig. 3, A and C). This activity accounts for the dark bands on the bottom of the gels in lanes 4 -6 ( Fig. 3, C and E) that are due to cordycepin monophosphate 3Ј-dAM 32 P, which are products of the 3Ј35Ј exonuclease activity of AP endonucleases. Similarly, some degradation of the 5Ј-end-labeled 17-mers is observable in Fig. 3, A  and D (lanes 4 and 5). G*T* Oligonucleotide Duplexes-Incubation of the G*T* 17-mer duplexes (Fig. 3, D and E) with the same enzymes as the G*CT* 17-mer duplexes (Fig. 3, A and C) yields similar results. Bifunctional DNA glycosylases excise the G* residue and the resulting AP sites thus formed are subsequently cleaved on their 3Ј-sides generating a 3Ј-deoxyribose-phosphate group at the 5Ј-cleavage fragment. The AP endonucleases on the other hand, incise the damaged strand on the 5Ј-side and adjacent to G*.
MALDI-TOF/MS Analysis of Cleavage Fragments-Further insights into the identities of the cleavage fragments generated by BER and NIR pathways were obtained using negative mode mass spectrometric methods. Here we focus on the results obtained with G*CT* duplexes (Fig. 4), since the G*T* and G*CT* duplexes yield similar cleavage products (Fig. 3). Results obtained with untreated G*CT* 17-mer duplexes are depicted in Fig. 4A. The two peaks shown are due to the G*CT* IntraCLcontaining strand and the complementary strand, and the lack of other smaller molecular mass products indicates that the starting material is not degraded (Fig. 4A). On the other hand, after treatment with NEIL1, two additional, closely spaced fragments are observed at m/z 2402.1 and 2393.3 (Fig. 4B). These peaks are identified as the 5Ј-CCACCAAC-p and p-CTAC-CACC-3Ј 8-mer fragments that are the expected cleavage products induced by bifunctional glycosylase (Fig. 1B). These results are also consistent with the gel electrophoresis results (Fig. 3, A and C, lane 1) and the conclusion that the cross-linked base G* is excised.
After treatment with NTH1, one of the cleavage fragments is p-CTACCACC-3Ј (m/z 2393.3 (Fig. 4C) which is also observed in the case of NEIL1 (Fig. 4B). The other one at m/z 2518.5 is the fragment 5Ј-CCACCAAC-3Ј-p-PUA H (Fig. 4C), as expected for NTH1 (Fig. 3A, lane 8). In this fragment the PUA H is the hydrated form (-CH 2 CHOHCHOHCH 2 CHO, M ϭ 117.0 Da) of the aldehyde since its mass is 18 Da higher than the mass of the ␣,␤-unsaturated aldehyde (-CH 2 CHOHCH ϭ CHCHO, M ϭ 99.0 Da) (30,31). Again the G* base is missing from all of these oligonucleotide fragments; furthermore the former T* base is present in its intact form T in all 3Ј-downstream cleavage fragments.
The negative mass spectra of the products of APE1-catalyzed incision of G*CT* duplexes exhibits a series of molecular ions (Fig. 4D): 1) The cleavage fragment at m/z 2323.2 is 5Ј-CCAC-CAAC generated by the cleavage adjacent to and on the 5Ј-side of G*, as expected from Fig. 3A (lane 5). Two other fragments at m/z 2034.4 and 1720.2, corresponding to the 7-mer oligonucleotides 5Ј-CCACCAA-3Ј and the 6-mer 5Ј-CCACCA-3Ј, respectively, are attributed to degradation catalyzed by the known 3Ј35Ј exonuclease activity of APE1 (32,33). 2) The APE1-cleavage fragments at m/z 2394.7 and 2314.9, assigned to the 8-mer oligonucleotide 5Ј-p-CTACCACC-3Ј and its dephosphorylated form 5Ј-CTACCACC-3Ј, respectively, both released from the cleavage of the strand on the 3Ј-side of G* in the G*CT* duplexes. As in the case of 5Ј-upstream cleavage fragments, these are shorter by one nucleotide than a 9-mer with a G (or G*) at the 5Ј-end of the 3Ј-downstream cleavage fragment (21,22).
Taken together, these results suggest that all the DNA glycosylases tested excise the guanine G* in G*CT* and G*T* crosslinks and then cleave the DNA strands on the 3Ј-side of the apurinic site generated by the removal of G*, either via ␤ or ␤,␦-elimination mechanisms (Fig. 1B). The AP endonucleases tested, cleave the modified strands on the 5Ј-side of G* of the G*T* and G*CT* cross-links and induce further degradation of the cross-link with the release of the G* base that is missing from the 3Ј-side cleavage fragments analyzed by mass spectrometric methods.
It is well established that the 8,5Ј-cyclo-2Ј-deoxypurine lesions are not repaired by DNA glycosylase-mediated BER mechanisms, but are excellent substrates of mammalian NER (12,13,16,41,42) and prokaryotic NER repair pathways (43). However, a limited amount of information is available about IntraCLs that involve covalent bonds between two different nucleotides on the same strand. It has been shown that the G[8 -5m]C, G [8 -5mT] and G[8 -5m]C IntraCLs are substrates of the prokaryotic UvrABC system (44,45). Based on differences of levels of these lesions in NER-deficient and proficient mammalian cells and tissues, it was concluded that the G[8 -5m]T IntraCL is a substrate of NER in vivo (15). The G*CT* and G*T* IntraCLs embedded in 135-mer DNA were found to be excellent-to-modest NER substrates, respectively, in human cell extracts as shown in Fig. 5 (17). In the same cell extract incubation experiments, substantial amounts of 67-mer cleavage fragments, corresponding to cleavage at the sites of the lesions, were also noted, but not further investigated (17).
Our in vitro biochemical studies suggest that all the DNA glycosylases/AP lyases tested excise the guanine G* in G*CT* and G*T* cross-links and then cleave the DNA strands on the 3Ј-side of the apurinic site resulting from the removal of G*, via either ␤ or ␤,␦-elimination mechanisms (Fig. 1B). All of the NIR-specific AP endonucleases tested, cleave the damaged strands on the 5Ј-side of G* and appear to generate cordycepinlabeled ϳ9-mers with either phosphate or OH-groups at the 5Ј-ends, the latter exhibiting a mobility similar to that of a 10-mer with a phosphate residue at its 5Ј-end. However, the MALDI-TOF/MS results are consistent with the 9-mer interpretation. Nevertheless, incision at the 3Ј-side of a lesion by NIR mechanisms is not supported by previous studies (21) (Fig.  1B), and we cannot exclude that the initially formed 3Ј-fragment does contain a G* residue at its 5Ј-end and that AP endonucleases induce further degradation of DNA with the loss of the G* nucleotide to yield the observed 5Ј-p-CTACCCCACC and 5Ј-CTACCACC fragments (Fig. 4D, peaks 9 and 10), under conditions used to prepare samples for MALDI-TOF/MS analysis.
It is interesting to note that the bifunctional DNA glycosylases and AP endonucleases studied are capable of excising the cross-linked G* in the G*T* and G*CT* duplexes, but not the originally cross-linked T* according to the denaturing gel elec- trophoresis results (Fig. 3) and MALDI-TOF/MS analysis (Fig.  4). In contrast, denaturing gel electrophoresis showed that the standard hot piperidine treatment cleaves 5Ј-CCA-TCG*CT*ACC at G* and T* sites with the release of 5Ј-CCATCp and 5Ј-pACC fragments according to MALDI-TOF/MS analysis (27). This is clear evidence that hot alkali is unable to hydrolyze the G*[C8-N3]T* bond to form the intact T that is generated by the bifunctional DNA glycosylases and AP endonucleases (Figs. 3 and 4). The G*[C8-N3]T* bond is also resistant to nuclease P1, which generates cross-linked dinucleotide d(G*-T*) and d(G*pT*) fragments after complete hydrolysis of the single-stranded DNA containing G*CT* and G*T* IntraCL, respectively (10,27). The phosphodiesterases 1 and 2 do not cleave the phosphodiester bonds between the crosslinked nucleotides (Fig. 2), and the combined action of these enzymes generates only d(G*pCpT*) and d(G*pT*) fragments (10,27).
The cross-linked guanine has two covalent N-C bonds, one is the normal G*[N7-C1Ј] glycosydic bond, and the second involves the C8 atom on the same imidazole ring linked to N3 of thymine in G*T* or to the T on the 3Ј-side of C in G*CT* duplexes (Fig. 1A). Based on our data we propose that the bifunctional DNA glycosylases and AP endonucleases are able to cleave both the G*[C8-N3]T* and the G*[N7-C1Ј] bonds.
To summarize, we have shown that the cleavage observed in cell extracts at the sites of the G*T* and G*CT* intrastrand lesions in double-stranded DNA (17) can be attributed to BER mechanisms. It should be noted that in the same experiments both G*CT* and G*T* lesions are also removed by NER mechanisms as shown in Fig. 5 (17). It is noteworthy that the higher yield of NER products in the case of G*CT* is accompanied by a lower yield of BER products (Fig. 5A). By contrast, the yield of NER products is ϳ5 times smaller in the case of G*T* while the BER yield is significantly higher (Fig. 5B). This inverse correlation between BER and NER product yields suggests that these two processes may be competing with one another. The origins of these effects are presently under investigation. It is remarkable that both BER and NER can function in parallel in incising the G*T* and G*CT* intrastrand cross-links in DNA in the same human cell extract experiments (17). These observations suggest that these two pathways may complement one another in cellular environments.