High-resolution Digital Mapping of N-Methylpurines in Human Cells Reveals Modulation of Their Induction and Repair by Nearest-neighbor Nucleotides*

Background: N-Methylpurines are repaired by the base excision repair pathway. Results: Induction and repair of N-methylpurines in human cells is significantly affected by nearest-neighbor nucleotides. Conclusion: Modulation of N-methylpurine repair by nearest-neighbor nucleotides is primarily achieved by affecting the initial step of the base excision repair process. Significance: Excision of N-methylpurines by alkyladenine glycosylase is most dramatically affected by sequence context. N-Methylpurines (NMPs), including N7-methylguanine (7MeG) and N3-methyladenine (3MeA), can be induced by environmental methylating agents, chemotherapeutics, and natural cellular methyl donors. In human cells, NMPs are repaired by the multi-step base excision repair pathway initiated by human alkyladenine glycosylase. Repair of NMPs has been shown to be affected by DNA sequence contexts. However, the nature of the sequence contexts has been poorly understood. We developed a sensitive method, LAF-Seq (Lesion-Adjoining Fragment Sequencing), which allows nucleotide-resolution digital mapping of DNA damage and repair in multiple genomic fragments of interest in human cells. We also developed a strategy that allows accurate measurement of the excision kinetics of NMP bases in vitro. We demonstrate that 3MeAs are induced to a much lower level by the SN2 methylating agent dimethyl sulfate and repaired much faster than 7MeGs in human fibroblasts. Induction of 7MeGs by dimethyl sulfate is affected by nearest-neighbor nucleotides, being enhanced at sites neighbored by a G or T on the 3′ side, but impaired at sites neighbored by a G on the 5′ side. Repair of 7MeGs is also affected by nearest-neighbor nucleotides, being slow if the lesions are between purines, especially Gs, and fast if the lesions are between pyrimidines, especially Ts. Excision of 7MeG bases from the DNA backbone by human alkyladenine glycosylase in vitro is similarly affected by nearest-neighbor nucleotides, suggesting that the effect of nearest-neighbor nucleotides on repair of 7MeGs in the cells is primarily achieved by modulating the initial step of the base excision repair process.

Alkylation damage to DNA can be caused by environmental or chemotherapeutic alkylating agents, or by natural cellular methyl donors such as S-adenosylmethionine (1). Simple S N 2 methylating agents, such as dimethyl sulfate (DMS), 2 produce a variety of damaged bases in DNA, of which N-methylpurines (NMPs), including N 7 -methylguanine (7MeG) and N 3 -methyladenine (3MeA), constitute ϳ90% of the lesions (2). NMPs are repaired by the multi-step base excision repair (BER) pathway. In human cells, the BER process is initiated by human alkyladenine glycosylase (hAAG), which recognizes and excises the methylated purine bases (1). The apurinic/apyrimidinic (AP) sites formed are typically recognized by an AP endonuclease that cleaves the DNA on the 5Ј side of the AP sites. The repair process concludes following DNA repair synthesis and ligation. BER of NMPs plays an important role in protecting the genome, and at the same time confounds cancer alkylation therapies, by excising the cytotoxic lesions.
The efficiency of BER can be affected by local DNA sequences (3). The different efficiencies may have implications for genome stability as lesions that persist longer in DNA have a greater chance of leading to mutagenesis. It was demonstrated that the rates of NMP repair in the PGK1 gene of human cells are highly heterogeneous and can be affected by some DNA sequence contexts (4). However, until now, the exact nature of the sequence contexts that affect the repair of NMPs in human cells has been poorly understood. Also, the excision kinetics of NMP bases have been difficult to measure in vitro because the methylated bases are labile to spontaneous depurination and the AP sites formed after excision of the methylated bases are labile to spontaneous strand cleavage. This difficulty has made it impossible to unambiguously address the question as to whether the heterogeneity of NMP repair is caused by the excision of the methylated bases or by a later step of the BER process.
Multiple methods have been developed to map the formation and repair of DNA adducts in living cells (5,6). Broadly speaking, these methods can be grouped into the genome overall level, gene/DNA fragment level, and nucleotide level. The nucleotide-level methods can be powerful for delineating the modulation of DNA adduct formation and repair by various cellular elements, such as DNA sequence context, chromatin structure, epigenetic modifications, and certain cellular processes (e.g. transcription) (7). However, most of the nucleotidelevel methods are only sensitive enough for mapping DNA adduct formation and repair in prokaryotes and lower eukaryotes (i.e. organisms with small genomes) or in multicopy sequences (e.g. those in mitochondria) of mammalian cells. Indeed, until very recently, ligation-mediated PCR had been the only method that allowed nucleotide-level mapping of DNA lesions in single-copy sequences in mammalian cells (8). However, the ligation-mediated PCR technique is unsuitable for large scale analyses of DNA damage and repair in the cell. Also, the gel bands corresponding to DNA lesions cannot always be well separated, and the lesions cannot be precisely counted. Very recently, a high-throughput method, called XR-Seq (excision repair sequencing), which allows mapping of nucleotide excision repair (NER) of UV photoproducts (cis-syn cyclobutane pyrimidine dimers and 6-4 photoproduct) in human cells, was developed (9). This method can provide a snapshot measurement of UV photoproduct-containing oligonucleotides (ϳ30 nucleotides) that are excised during NER. As it relies on immunoprecipitation of the excised oligonucleotides that transiently exist in the cell, the XR-Seq method is not suitable for accurately measuring the kinetics of NER. Also, the XR-Seq method cannot be used for mapping the formation and repair of DNA lesions that are not NER substrates.
Here, we report the development of a sensitive method, LAF-Seq (Lesion-Adjoining Fragment Sequencing), which allows nucleotide-resolution digital mapping of DNA damage and repair in multiple genomic fragments of interest in human cells. This method is suitable for mapping any type of DNA lesions, provided that they can be converted into DNA single or double strand breaks after isolation of total genomic DNA. We also developed a strategy that allows accurate measurement of the excision kinetics of NMP bases in vitro. We demonstrate that 3MeAs are induced to a much lower level by DMS and repaired much faster than 7MeGs in human fibroblasts. Induction of 7MeGs is affected by nearest-neighbor nucleotides, being enhanced at sites neighbored by a G or T on the 3Ј side, but impaired at sites neighbored by a G on the 5Ј side. Repair of 7MeGs is also affected by nearest-neighbor nucleotides, being slow if the lesions are between purines, especially Gs, and fast if the lesions are between pyrimidines, especially Ts. Furthermore, we show that the excision of 7MeG bases by hAAG in vitro is similarly affected by nearest-neighbor nucleotides, suggesting that the effect of nearest-neighbor nucleotides on repair of 7MeGs in human cells is primarily achieved by modulating the initial step of the BER process, namely the excision of 7MeG bases by hAAG.

Experimental Procedures
Cell Culture, DMS Treatment, and DNA Isolation-Telomerase-immortalized human foreskin fibroblast R2F/TERT cells (10) were cultured in DMEM/F12 medium supplemented with 15% newborn calf serum and 10 ng/ml epidermal growth factor. A fraction of the cultured cells was saved as control, and the rest were treated with 0.005% (v/v) DMS for 10 min. This treatment induced ϳ1 NMP/10 kb of genomic DNA and caused no obvious cell killing. The cells were washed twice with PBS containing 0.005% (v/v) ␤-mercaptoethanol, washed twice with PBS, and then replenished with fresh medium and incubated at 37°C. At 0, 6, 12, and 24 h of the incubation, an aliquot of the cells was harvested. Each aliquot (10 million cells) was resuspended in 2 ml of nuclei isolation buffer (10 mM Tris-Cl, 10 mM NaCl, 3 mM MgCl 2 , 0.5% Nonidet P-40, pH 7.9) and incubated on ice for 30 min. The nuclei were pelleted by centrifugation, resuspended in 10 ml of lysis buffer (10 mM Tris-HCl, 1 mM EDTA, 400 mM NaCl, 1% SDS, 10 g/ml proteinase K, pH 8.0), and incubated at 65°C for 12 h. In addition to releasing DNA, this 12-h incubation at 65°C caused complete depurination of NMP bases from the DNA backbone. The AP sites formed might be further cleaved by spontaneous ␤ elimination or ␤ and ␦ eliminations, leaving an unsaturated sugar or a phosphate at the 3Ј ends of the cleaved sites. Each of the samples was mixed with 5 ml of 5 M NaCl and incubated on ice overnight. The samples were centrifuged at 12,000 rpm for 30 min at 4°C, and the supernatant was collected. The genomic DNA was precipitated with isopropanol.
Construction and Sequencing of Libraries of DNA Fragments of Interest-Six g of genomic DNA of each sample (DMStreated and untreated) were digested with 12 units of MseI and 3 units of endonuclease IV (New England Biolabs) in 100 l of NEBuffer 3 (50 mM Tris-HCl, 100 mM NaCl, 10 mM MgCl 2 , 1 mM DTT, 100 g/ml BSA, pH 7.9) overnight. MseI digestion releases the three fragments encompassing exons 11 (281 bp) and 15 (264 bp) of the BRAF gene and the coding exon 1 (228 bp) of the NRAS gene. In addition to incising the DNA on the 5Ј side of the AP sites formed after spontaneous depurination of NMPs, endonuclease IV also removes unsaturated sugar or phosphate groups from the 3Ј ends of the DNA (11). Each of the samples was added with NaCl to a final concentration of 1 M and a mixture (0.5 pmol each) of biotinylated oligonucleotides (see Table 1). The samples were heated at 95°C for 5 min and then incubated at 50°C for 30 min. Ten l of streptavidin magnetic beads (Dynabeads M-280 streptavidin, Life Technologies) were added to each of the samples and incubated at room temperature for 30 min. The beads were washed with STES (10 mM Tris, 1 mM EDTA, 100 mM NaCl, 0.5% SDS, pH 7.5) at room temperature and with STE (10 mM Tris, 1 mM EDTA, 100 mM NaCl, pH 7.5) at 55°C. To estimate the efficiencies of fishing out the fragments of interest, a small fraction of the beads was resuspended in 25% ammonium hydroxide. The eluted fragments of interest (contained in the supernatants) were vacuum-dried and quantified by real-time PCR (using SYBR Select Master Mix from Life Technologies).
To ligate adapter RC-Ai (see Table 1) to the 3Ј ends of the fragments of interest, the beads were resuspended in 10 l of CircLigase buffer (50 mM MOPS, pH 7.5, 20% PEG 8000, 10 mM KCl, 5 mM MgCl 2 , 0.5 M betaine, 20 M ATP, 2.5 mM MnCl 2 , 200 g/ml BSA, 1 mM DTT), which we have extensively optimized to allow efficient ligation of single-stranded DNA molecules by CircLigase (Epicentre) (12). Two hundred pmol of RC-Ai and 10 units of CircLigase were added to each of the samples and incubated at 55°C for 1 h. The beads were washed with STES at room temperature, binding and washing buffer with SDS (10 mM Tris, 1 M NaCl, 1 mM EDTA, 0.5% SDS, pH 8.0) at 55°C, and Tris-EDTA buffer (pH 8.0) at room temperature. The RC-Ailigated fragments were eluted from the beads by using 25% ammonium hydroxide and vacuum-dried.
To ligate adapter P1 to the 5Ј ends of the fragments of interest, the DNA samples were dissolved in 50 l of T7 DNA ligation buffer (66 mM Tris-HCl, 10 mM MgCl 2 , 1 mM ATP, 1 mM DTT, 7.5% PEG 6000, pH 7.6) containing a mixture (0.25 pmol each) of the biotinylated oligonucleotides and 20 pmol of adapter P1 (see Table 1). The samples were heated at 95°C for 5 min and at 50°C for 30 min and then cooled to room temperature. The samples were added with 3 units of T7 DNA ligase (Molecular Cloning Laboratories) and incubated at room temperature for 20 min. Each of the samples was added with 50 l of 2ϫbinding and washing buffer (20 mM Tris, 2 M NaCl, 2 mM EDTA, pH 8.0) and 10 l of streptavidin magnetic beads and incubated at room temperature for 30 min. The beads were washed with STES, Tris-EDTA buffer (pH 8.0), and H 2 O at room temperature. The adapter-ligated fragments were eluted Dashed line box indicates a fragment that can be released by restriction digestion. Gray circles marked with B denote the biotin group. The 3Ј (pink) and 5Ј (black) regions of the biotinylated oligonucleotides are complementary to adapter P1 and the 5Ј end region of fragments of interest, respectively. Adapter RC-Ai ("i" stands for a barcode number) is an oligonucleotide whose 5Ј end is phosphorylated and 3Ј end blocked by an inverted dT. The 5Ј region (cyan) of RC-Ai is a barcode sequence (i). The 3Ј region (brown) of RC-Ai is complementary to adapter A. See Table 1 for sequences of the oligonucleotides.

Digital Mapping of N-Methylpurines in Human Cells
from the beads by using 25% ammonium hydroxide and vacuum-dried. The efficiencies of adapters RC-Ai and P1 ligations were estimated by real-time PCR using a small fraction of the eluted fragments as templates.
The libraries of DNA fragments were amplified by 10 -14 cycles of PCR, using Herculase II fusion DNA polymerase (Agilent Technologies) and adapters A and P1 (see Table 1) as primers. The qualities and quantities of the libraries were analyzed by using a Bioanalyzer (Agilent). The barcoded libraries were then pooled, and the sequencing templates were prepared on Ion Sphere particles by using the Ion OneTouch 2 system. The templates were loaded onto Ion 318 chips and sequenced from the RC-Ai-ligated ends (by "reading" the complementary strand in the 5Ј to 3Ј direction) on an Ion Torrent personal genome machine.
Analysis of Sequencing Data-The sequencing reads were sorted according to their barcodes by using the Torrent Suite software. After being trimmed of barcode sequences, the reads were aligned to the sequences of the fragments of interest by using Bowtie 2 in Galaxy (The Galaxy Project). As the NMPs were induced to a relatively low level (ϳ1 NMP/10 kb of DNA), the majority (Ն97%) of the reads correspond to the complementary strands of the full-length (undamaged) fragments of interest. A small fraction of the reads (Յ3%) corresponds to the complementary strands of the fragments of interest adjoining the NMP sites. The total numbers of reads from the DMS-treated samples were normalized to those from the control (not treated with DMS) samples. The 5Ј ends of the reads that align to the internal sites of the fragments of interest would correspond to the NMPs induced or remaining (at different times of repair) at sites on the complementary strands. The reads whose 5Ј ends align to the same sites of the fragments of interest were counted. To remove background signals, the read counts aligned to different sites of the fragments of interest from the DMS-treated samples were deducted by the corresponding counts from the control samples. Only those sites with Ͼ10 counts of lesions at 0 h of repair were analyzed for NMP repair.
Excision of 7MeG Bases by hAAG in Vitro-Eighty pmol of each of the substrates (see Fig. 9A) was treated with 0.6% (v/v) DMS in a total volume of 100 l of Tris-EDTA buffer (pH 8.0) at room temperature for 15 min, which induced ϳ0.05 7MeG per fluorescein (FAM)-labeled fragment of the substrates. The DMS-treated substrates were purified and dissolved in 80 l of ThermoPol buffer (20 mM Tris-HCl, 10 mM (NH 4 ) 2 SO4, 10 mM KCl, 2 mM MgSO 4 , 100 g/ml BSA, 0.1% Triton X-100, pH 8.8, at 25°C). Two pmol of purified hAAG (New England Biolabs) was added to each of the samples and incubated at 37°C. At different times of the incubation, an aliquot of 10 l (10 pmol of DNA) was taken, and the DNA was purified. Each of the aliquots was then treated with 100 mM O-(tetrahydro-2H-pyr an-2-yl)hydroxylamine (OTX) (Sigma) in a total volume of 25 l of STE at 37°C for 1 h. Each of the aliquots (10 pmol of DNA) was then treated with 1.2 pmol of hAAG and an excess amount of AP endonuclease 1 (APE1) (New England Biolabs) at 37°C for 3 h to excise all the remaining NMPs and completely cleave the DNA at the resulting unprotected AP sites.
To analyze the samples on a 3130xl Genetic Analyzer (Applied Biosystems), 0.1 pmol of each of the samples was loaded into an 80-cm capillary column filled with POP-6 polymer. The FSA binary files generated were converted to text files by using BatchExtract (ftp://ftp.ncbi.nih.gov/pub/forensics/ BATCHEXTRACT). The fluorescent signals were then analyzed by using Microsoft Excel.

Results
Development of LAF-Seq-We reasoned that DNA lesions at specific sites of the genome could be digitally mapped if singlestranded DNA fragments adjoining the lesions could be fished out, sequenced, and counted. Exons 11 and 15 of the BRAF gene and the coding exon 1 of the NRAS gene contain frequent mutations in cutaneous melanoma and other human cancers (13). We chose to develop the method by mapping NMP induction and repair in both strands of genomic DNA fragments encompassing these exons.
Human fibroblasts were treated with DMS to induce a low level of NMPs in the genomic DNA (ϳ1 NMP/10 kb of DNA). Following different times of repair incubation, aliquots were taken and total genomic DNA was isolated. DNA samples from the DMS-treated cells at different times of repair and from the untreated cells was separately cut with a restriction enzyme to release the fragments of interest and cleaved at the NMP sites (Fig. 1). The fragments of interest were simultaneously fished out from each of the samples by using excess copies of a mixture of biotinylated oligonucleotides that are complementary to the 5Ј end region of the fragments of interest ( Table 1). The efficiency of fishing out the genomic fragments of interest from the total genomic DNA was around 70% (Fig. 2). To allow sequencing by the Ion Torrent personal genome machine, adapter RC-Ai (reverse complement of adapter Ai, "i" stands for a barcode number) and adapter P1 were ligated to the 3Ј and 5Ј ends of the fragments of interest, respectively. Each of the samples was ligated with specifically barcoded RC-Ai. Under our optimized conditions, over 90% of the fished-out fragments of interest were ligated to adapters RC-Ai and P1 (Fig. 3). The adapter-ligated fragments with different barcodes were pooled and sequenced simultaneously from the adapter RC-Ai ligated ends (by reading the complementary strand in the 5Ј to 3Ј direction) on Ion Torrent 318 chips, which typically generate 5-6 million high quality sequencing reads each run (Fig. 4). Most of the sequencing reads can be aligned to the reference sequences of the fragments of interest (94 and 99% for the two sequencing runs, respectively). The low read counts aligned to specific sites of the fragments of interest from the control (untreated with DMS) samples reflect background signals, which were deducted from the corresponding sites from the DMS-treated samples. NMPs induced (at 0 h of repair) or remaining (at different times of repair) at specific sites of the fragments of interest were then counted by tallying the sequencing reads whose ends align to the respective damaged sites.
Induction and Repair of NMPs in Human Cells-Lesion counts at different sites of the fragments analyzed are shown in Figs. 5-7. As can be seen, most (ϳ90%) of the NMPs induced by DMS were 7MeGs, and a small fraction (ϳ10%) was 3MeAs. For five out of six strands of the three fragments analyzed, NMPs that can be detected are located in sites that are Ն2-9 nucleotides from the regions that are complementary to the biotinylated oligonucleotides used to fish out the fragments (Fig. 1). This is consistent with the fact that the CircLigase we used to ligate adapter RC-Ai to the 3Ј end of the fragments of interest can only ligate single-stranded but not double-stranded DNA (12). However, we detected 7MeGs at a site of the region that is complementary to BRAF11-2 (Table 1). This site is complementary to the 5th nucleotide (from the 5Ј end) of BRAF11-2 and is located in the transcribed strand of the fragment encompassing exon 11 of the BRAF gene (Fig. 5, see the "g" site that is the second nucleotide from the 5Ј end of the transcribed strand). Presumably, the fragment adjoining 7MeG at this site was ligated to adapter RC-Ai due to "breathing" of doublestranded DNA. The "breathing" at 55°C (the temperature we used for ligation of RC-Ai) might be frequent but did not cause detachment of the fragment from the biotinylated BRAF11-2.
In addition to inducing the predominant NMPs, DMS induces a small fraction (0.3%) of O 6 -methylguanine, which is stable (2) and cannot be converted to a strand break and detected by the LAF-Seq method. DMS also induces ϳ10 other minor adducts (e.g. N 1 -methyladenine, N 7 -methyladenine, N 3 -methylguanine, and O 2 -methylcytosine), some of which are heat-labile (2) and may be detected by the LAF-Seq method. However, as together they comprise Ͻ5% of the total lesions induced by DMS (2), the levels of the minor adducts were expected to be marginal when compared with NMPs we detected here.
The induction levels and repair rates of NMPs were highly heterogeneous. However, the heterogeneity did not seem to correlate with the hotspots of carcinogenic mutations in these fragments documented in the Catalogue of Somatic Mutations in Cancer (COSMIC) (Figs. 5-7), suggesting that NMPs may not particularly cause more mutations at these hotspots. According to data deposited in the UCSC Genome Browser, some weakly positioned nucleosomes are present in the fragments we analyzed. The repair rates did not seem to signifi- cantly correlate with these weakly positioned nucleosomes (not shown).
7MeGs were induced to higher levels at sites of aGg, aGt, tGg, and tGt, but to lower levels at sites of gGa, gGc, gGg, and gGt (Gs are the sites of 7MeG formation, and lowercase letters denote nearest-neighbor nucleotides) (Fig. 8A, compare 0-h lesion levels at the different sites). It appears that 7MeG induction was enhanced at sites neighbored by a G or T on the 3Ј side, but impaired at sites neighbored by a G on the 5Ј side. In G tracts (with Ն2 continuous Gs), 7MeG induction tended to be  SEPTEMBER 18, 2015 • VOLUME 290 • NUMBER 38 high at the 5Ј-most G site but low at the other G sites (Figs. 5-7 and 8A).

Digital Mapping of N-Methylpurines in Human Cells
The 7MeG counts at some sites, especially those between purines (As or Gs), were higher after certain times of repair than those at 0 h of repair (Figs. 5-7 and 8, A and B), presumably reflecting certain levels of continued induction of the lesions after DMS was removed from the medium and slow repair at these sites. Indeed, repair of 7MeGs appeared to be significantly affected by nearest-neighbor nucleotides, being slow for the lesions between purines, especially Gs, and fast for the lesions between pyrimidines, especially Ts (Fig. 8, A-D). 7MeGs neighbored by a pyrimidine on one side and a purine on the other were repaired at intermediate rates. In contrast, the second-or third-nearest nucleotides did not seem to significantly affect the rate of 7MeG repair (data not shown). The average repair rate of all 3MeAs in these fragments was much faster than that of 7MeGs (Fig. 8E). Indeed, the average half-life of 3MeAs was ϳ7 h, whereas that of 7MeGs was ϳ22 h.
Excision of 7MeG Bases by hAAG in Vitro-The rates of 7MeG repair we measured in the cells reflect the speeds of the whole BER process. We wondered whether the heterogeneity of 7MeG repair in the cells was caused by the initial step (i.e. the excision of 7MeG bases) of the BER process. The excision kinetics of 7MeG bases by purified hAAG had been difficult to measure because 7MeG bases are labile to spontaneous depurination and the AP sites formed after excision of the 7MeG bases are labile to spontaneous strand cleavage, especially at an ele-vated temperature. We therefore developed a new strategy for the measurement (Fig. 9). All the DNA substrates we used were identical except for the 2 nucleotides neighboring the G in the middle of the 5Ј-FAM-labeled strand (Fig. 9A). The substrates were treated with DMS to induce ϳ0.05 7MeG per FAM-labeled fragment of the substrates. The inductions of 7MeGs by DMS are largely random events, although they may be modulated by certain DNA sequences. At an induction level of 0.05 7MeG/fragment, the chance that the fragment molecules may contain Ͼ1 lesion is very low (0.12%). The DMS-treated substrates were incubated with a limited amount of hAAG. At different times of the incubation, aliquots were taken and the AP sites formed after excision of the damaged bases were stabilized by treatment with OTX. AP sites stabilized by OTX are resistant to cleavage by heat and the human AP endonuclease APE1 (14,15). The DNA was then completely cleaved at the remaining 7MeG sites by extended incubation with excess amounts of hAAG and APE1. The DNA fragments were analyzed by capillary electrophoresis on a 3130xl Genetic Analyzer, which is FIGURE 8. NMPs remaining at different times of repair. A, 7MeG counts (means ϩ standard deviation) at sites with different nearest-neighbor nucleotides. Gs indicate 7MeG sites, and lowercase letters denote nearest-neighbor nucleotides. The levels of 7MeG at the cGt site are not shown because Ͻ10 7MeGs were induced at the sole cGt site in all the three fragments analyzed. No error bars are shown for the cGc site as Ͼ10 7MeGs were induced at only one of three cGc sites in all the three fragments analyzed. The levels of 7MeG induced (0 h of repair) at aGg, aGt, tGg, and tGt sites were significantly higher than those at gGa, gGc, gGg, and gGt sites (p Ͻ 0.01, Student's t test). B, the percentage of 7MeGs remaining (mean) at sites with different nearest-neighbor nucleotides. The values at sites of fast repair (tGt and tGc) are significantly different from those at sites of slow repair (aGg, gGg, gGa, aGa) at all the time points of repair (p Ͻ 0.05, Student's t test). C, 7MeG counts (mean ϩ standard deviation) at sites neighbored by purines on both sides (rGr), by a purine on one side and a pyrimidine on the other (rGy/yGr), and by pyrimidines on both sides (yGy). D, the percentage of 7MeGs remaining (mean) at sites of rGr, rGy/yGr, and yGy. The value at each of the repair time points is significantly different from one context of the neighbor nucleotides to any of the other two (p Ͻ 0.01, Student's t test). E, the percentage of 3MeAs and 7MeGs remaining (mean). The percentage of 3MeAs remaining is significantly different from the percentage of 7MeGs remaining at each of the repair time-points (p Ͻ 0.01, Student's t test).
commonly used for routine DNA sequencing. We found that, when compared with DNA sequencing gels, the capillary electrophoresis has a much higher resolution in separating bands of DNA fragments and allows much more sensitive detection (the limit is ϳ0.1 fmol or 6 ϫ 10 7 of FAM-labeled molecules for detecting a band) and accurate quantification of the bands.
As can be seen in Fig. 10, 7MeG bases between purines, especially Gs, were excised much more slowly than those between pyrimidines, especially Ts. The trend of nearest-neighbor nucleotide effect on excision of 7MeG bases by hAAG in vitro resembles that on 7MeG repair in human cells. The overall excision rates of 7MeG bases in vitro under our conditions were much faster than those of 7MeG repair in the cells, presumably because 1) a relatively low level of hAAG existed in the cell and/or 2) the rates of 7MeG repair we measured in the cell reflected the whole process of BER, rather than just the excision of the damaged bases. The scattered distribution of 7MeGs and the chromatin structure might also contribute to the slow repair of 7MeGs in the cell. These results support the idea that the effect of nearest-neighbor nucleotides on repair of 7MeGs in human cells is primarily achieved by modulating the initial step of the repair process, namely the excision of the damaged bases by hAAG.

Discussion
We developed the LAF-Seq method for high-resolution digital mapping of NMP induction and repair in human cells. When compared with currently available gel-based methods, the LAF-Seq method has several advantages. First, the LAF-Seq method is highly sensitive. We mapped lesions that were induced to a level of ϳ1 per 10 kb of DNA. Second, the LAF-Seq method is much less labor-intensive, especially for simultaneously mapping DNA damage and repair in multiple sequences of the genome. Although we have tested fishing out and mapping three fragments of interest, there is no reason why more fragments cannot be fished out and analyzed simultaneously. Third, the LAF-Seq method can be used to achieve true nucleotide resolution mapping and digital quantification of DNA lesions. This is in contrast to gel-based methods where bands cannot always be well separated and lesions cannot be precisely counted. Furthermore, the LAF-Seq method should be able to be used for mapping other types of DNA lesions if they can be converted to single or double strand breaks after the genomic DNA is isolated from the cells. Many enzymes are available that can specifically incise DNA at the sites of different lesions. For example, the bifunctional DNA glycosylases formamidopyrimidine DNA glycosylase (FPG) and endonuclease VIII can excise oxidized purine and pyrimidine bases, respectively, and incise the DNA 3Ј (through ␤-elimination) and 5Ј (through ␦-elimination) to the resulting AP sites, leaving a phosphate group at the 3Ј end (16). On the other hand, T4 endonuclease V can incise DNA at cyclobutane pyrimidine dimers, resulting in an unsaturated sugar at the 3Ј end (17). To make the 3Ј ends ligatable to adaptor RC-Ai, the lesion-incised DNA samples can be treated with endonuclease IV, which seems to be able to remove essentially any non-OH groups at the 3Ј ends of DNA formed after incision by different damage-specific enzymes (11). However, certain fragments cannot be fished out from the genomic DNA due to the lack of appropriate restriction sites. Therefore, although it can be used for simultaneous mapping of DNA lesions in multiple restriction fragments, the LAF-Seq method is not suitable for genome-wide mapping of DNA lesions.
The N 7 position of Gs faces the major groove of DNA and has a high electrostatic potential, which is the major reason why this position is preferentially methylated by such S N 2 methylating agents as DMS (2). It was reported that DMS preferentially methylates the 5Ј end G in G runs, presumably due to the highest electrostatic potential at the N 7 position of the 5Ј end G (18). Our result that 7MeG induction is high at the 5Ј-most G site but low at the other G sites in a G tract agrees well with the previous study. However, unlike the previous study, which showed that an A present in a G run did not affect the damage pattern, we found that an A present in a G run will disrupt the pattern (i.e. N 7 -methylation of a G 3Ј to the A is not impaired by the G(s) 5Ј to the A (Figs. 5-7)). We also found that 7MeG induction was enhanced at sites neighbored by a G or T on the 3Ј side. The effects of the 3Ј neighbor nucleotides on the 7MeG induction may also be due to modulation of the electrostatic potential. It can be predicted that the electrostatic potential at the N 7 position of a G can be reinforced by a 3Ј-G or -T (19,20).
Each human cell contains thousands of copies of mitochondrial DNA (21). Using a gel-based method that is only sensitive enough for mapping DNA damage in multi-copy sequences in the human genome, we previously found that repair of 7MeGs in human mitochondria is affected by nearest-neighbor nucleotides, being slow for the lesions between purines and fast for the lesions between pyrimidines (22). It was found recently that hAAG is present in human mitochondria and associates with mitochondrial single-stranded DNA-binding protein (23). Taken together with our results, it appears that excision of 7MeG bases by hAAG may be similarly affected by neighbor nucleotides in the nucleus and mitochondria, although the local environments can be quite different.
The crystal structure of hAAG (1BNK) showed that a damaged base is flipped from the DNA base stack into a sequestered active site pocket where the glycosidic bond between the damaged base and the deoxyribose is hydrolyzed (24). The stability of base stacking between two adjacent bases follows the order: purine-purine Ͼ Ͼ purine-pyrimidine Ͼ pyrimidine-purine Ͼ pyrimidine-pyrimidine, with a 2-kcal/mol free energy spread between the most stable purine-purine and the least stable pyrimidine-pyrimidine base stacks (25). Under the physiological pH, a 7MeG base can be positively charged (protonated) (26). The positive charge may increase base stacking through the cationinteraction (27). To date, the free energies of base stacking between 7MeG and normal purines or pyrimidines have not been documented. It is likely that a 7MeG base has more stable stacking with purines than with pyrimidines. The different base-stacking stabilities may affect the flipping of the 7MeG base by hAAG, thereby modulating the excision of the damaged bases. Indeed, it has been shown that hAAG is exquisitely sensitive to the structural context of a deoxyinosine lesion (an uncharged substrate of hAAG), and an inverse correlation between duplex stability and catalytic efficiency was observed (28).
The overall repair of NMPs in the yeast Saccharomyces cerevisiae (29,30) appears to be much faster than that in the human cells we reported here. Interestingly, however, repair of 7MeGs in the yeast has also been found to be affected by nearest-neighbor nucleotides, being slow if they are between purines and fast if they are between pyrimidines (29,30). Repair of NMPs in the yeast is solely initiated by Mag1, the homolog of hAAG (29). The crystal structure of the yeast Mag1 complexed with its substrate has not been reported. The nearest-neighbor nucleotides appear to affect hAAG and Mag1 similarly in excising the damaged bases.
Similar to hAAG, most DNA glycosylases have converged on a single mechanistic solution for damaged base recognition and excision: flip of the damaged base from the DNA base stack into a sequestered active site pocket (31). The activities of some but not all glycosylases have been shown to be modulated to some extent by DNA sequence context (for a review, see Ref. 3). Interestingly, altering either the global DNA sequence or the 5Ј-flanking base pair failed to influence the excision of 7,8-dihydro-8-oxoguanine (8-oxoG) by human 8-oxoguanine glycosylase (OGG1) in vitro (32). However, an 8-oxoG located in the CAGGGC [8-oxoG]GACTG motif is poorly excised by OGG1 (33). Therefore, the excision of 8-oxoG by OGG1 does not seem to be significantly affected by base stacking with nearest-neighbor nucleotides. It seems that the activities of alkyladenine glycosylases (e.g. hAAG and Mag1) are more dramatically modulated by nearest-neighbor nucleotides than those of any other DNA glycosylases analyzed so far.
Author Contributions-S. L. and M. L. conceived and designed the studies. M. L., S. L., and T. K. conducted the experiments. M. L. and S. L. analyzed the data and wrote the paper. All authors approved the final version of the manuscript.