Perturbation of base excision repair sensitizes breast cancer cells to APOBEC3 deaminase-mediated mutations

Abundant APOBEC3 (A3) deaminase-mediated mutations can dominate the mutational landscape (‘mutator phenotype’) of some cancers, however, the basis of this sporadic vulnerability is unknown. We show here that elevated expression of the bifunctional DNA glycosylase, NEIL2, sensitizes breast cancer cells to A3B-mediated mutations and double-strand breaks (DSBs) by perturbing canonical base excision repair (BER). NEIL2 usurps the canonical lyase, APE1, at abasic sites in a purified BER system, rendering them poor substrates for polymerase β. However, the nicked NEIL2 product can serve as an entry site for Exo1 in vitro to generate single-stranded DNA, which would be susceptible to both A3B and DSBs. As NEIL2 or Exo1 depletion mitigates the DNA damage caused by A3B expression, we suggest that aberrant NEIL2 expression can explain certain instances of A3B-mediated mutations.

Although A3B is a major source of A3 mutations in some cancers (Burns et al., 2013a;Burns et al., 2013b) and can be overexpressed as a function of cell proliferation in breast cancers (Cescon et al., 2015), the frequency of A3B mutations is not always correlated with A3B expression (Cescon et al., 2015;Nik-Zainal et al., 2014;Roberts and Gordenin, 2014). Furthermore, the aforementioned single-stranded A3 substrates are also present in non-cancerous cells, and A3 expression is rather ubiquitous (Refsland et al., 2010). Therefore, a major issue is why some cancers become sensitized to the activity of APOBEC enzymes and whether it is related to dysregulation of DNA repair.
We had earlier shown that the repair of plasmid-borne mismatches can induce flanking A3-mediated mutations in HeLa cells. Although the mismatches (e.g., U/G) were invariably repaired by base excision repair (BER), this process was sometimes hijacked by non-canonical mismatch repair (MMR), which generates single-stranded APOBEC substrates (Chen et al., 2014). In its simplest form, BER consists of a concerted series of reactions: removal of U by a Uracil-DNA Glycosylase (UDG, UNG in mammals) to generate an abasic (AP) site; scission of the AP site by apurinic/apyrimidinic endonuclease 1 (APE1) to generate a nick with a 3'OH and a 5' deoxyribophosphate (5'dRP); insertion of the complementary base and removal of the 5'dRP by polymerase b (Polb); ligation of the nick by DNA Ligase I or Ligase III-XRCC1 (Beard et al., 2006;Robertson et al., 2009). Because some of these intermediates can be highly mutagenic, these reactions are tightly coupled and sequestered (Fu et al., 2012;Prasad et al., 2010). Nonetheless, under certain conditions MMR can access the 3'OH terminated nick, leading to an Exo1-mediated resection that exposes single-stranded DNA opposite to the nicked strand (Kadyrov et al., 2006;Peña-Diaz et al., 2012;Pluciennik et al., 2010;Schanz et al., 2009). Given that thousands of U/G mismatches and AP sites are produced daily (Atamna et al., 2000;Barnes and Lindahl, 2004;Frederico et al., 1993), perturbations that compromise the integrity of the BER complex could lead to DNA damage (Fu et al., 2012).
Here we demonstrate in breast cancer cell lines that elevated expression of NEIL2, a bifunctional glycosylase normally involved in oxidative base excision repair (Chakraborty et al., 2015;Das et al., 2006;Hazra et al., 2002;Wiederhold et al., 2004), facilitates A3B-mediated mutations during U/G mismatch repair and induces double-strand breaks (DSBs) in genomic DNA. We further show that purified NEIL2 disrupts canonical BER by outcompeting APE1 for AP sites, thereby providing a possible mechanistic explanation for how this instance of DNA repair dysregulation contributes to the mutational landscape in breast cancer cells.

A3B activity is not the only determinant of repair-induced mutations
To examine U/G mismatch repair-induced effects in breast cancer cell lines, we transfected shuttle vectors containing no or a U/G mismatch into four established breast cancer cell lines: MCF7, HCC1569, Hs578T and MDA-MB-453, and screened for mutations in the reporter region that flanked the U/G. The reporter region consists of the E. coli SupF gene and its promoter on the shuttle vector pSP189-SnA ( Figure 1A and Figure 1-figure supplement 1A). Inactivating mutations of the SupF region induced by U/G repair cannot suppress the mutated b galactosidase gene in the MBM7070 E. coli strain, resulting in white colonies on the indicator plates ( Figure 1A, bottom row). U/G-repair did not induce mutations in MDA-MB-453, but it did so in Hs578T ( Figure 1B, bottom bar graph), despite similar levels of A3B transcripts ( Figure 1B, upper bar graph) and comparable nuclear TCspecific deaminase activity ( Figure 1C and Figure 1-figure supplement 1B,C) in these cell lines. The discrepancy between statistically significant amounts of repair-induced mutations and A3B expression also occurred in other cell lines ( Figure 1B). We sequenced the mutated reporter regions of plasmids from all the white colonies, and essentially all of the repair-induced mutations in Hs578T and HCC1569 exhibited an A3 signature, displayed here on the complement of the TC-containing strand -thus, G was the most frequently mutated nucleotide and >70% of mutated bases in Hs578T cells and >50% in HCC1569 cells involved AGA, CGA, or TGA ( Figure 1D,E and Figure 1-figure supplement 1D).
Among the seven A3 enzymes, A3A, A3B, A3C, and A3H localize to the nucleus and prefer TC sites (Lackey et al., 2013). Quantitative real-time PCR (qRT-PCR) showed that only A3B and A3C were expressed in Hs578T cells (  qRT-PCR of A3B relative to the housekeeping gene TBP. Lower panel: mutation rate (scored as % of white/total colonies) induced by U/G mismatch repair in MCF7, HCC1569, Hs578T, and MDA-MB-453 breast cancer cell lines. 0 MM, no mismatch; U/G MM, U/G mismatch. Error bars represent s.d., n = 2 for MCF7, HCC1569 and MDA-MB-453 cells; n = 5 for Hs578T cells. ** P < 0.01; *** P < 0.001; n.s., no significant difference by two-tailed unpaired Student's t test. (C) Concentration gradient of in vitro deaminase assay using nuclear extracts from Hs578T and MDA-MB-453 cells against a -TCT-containing fluorescein-labeled single strand oligonucleotide (39 nt). The amounts of total protein used are listed on top of the gel. The right panel shows quantification of the deamination percentage. The deamination activity is specific for -TCT-(Figure 1-figure supplement 1B). The time course deamination is shown in Figure 1-figure supplement 1C. S, substrate; P, product. (D and E) Mutation matrices and 5'-Trinucleotide context of mutations induced by U/G MM repair in Hs578T (D) and HCC1569 (E) cells. C is the most frequently mutated base and 70% of the mutated bases are in a 5'-GA (reverse complement of 5'-TC) motif. (F) A3B deficiency decreases U/G mismatch repair-induced mutagenesis. 0 MM, no mismatch; U/G MM, U/G mismatch. Error bars represent s.d., n = 3. *** P < 0.001 by twotailed unpaired Student's t test. The online version of this article includes the following figure supplement(s) for figure 1:  mutations in Hs578T cells. That non-mutagenic MDA-MB-453 cells contained a similar level of C-deaminase activity as the mutagenic Hs578T cells ( Figure 1C and Figure 1-figure supplement 1C) indicates that A3 activity per se is not sufficient to cause repair-induced mutations.

NEIL2 facilitates repair-induced mutagenesis
To explore whether dysregulation of DNA repair sensitized cells to A3B-mediated mutations, we compared the expression of 84 DNA repair enzymes in Hs578T and MDA-MB-453 cells using RT 2 Profiler PCR array. Consistent with previous RNA-seq data (Klijn et al., 2015), most of the tested genes were expressed at lower levels in Hs578T than in MDA-MB-453 (Figure 2-source data 1). However, two genes, NEIL2 and TREX1 were significantly upregulated ( Figure 2A). TREX1 is a 3'À5' exonuclease that degrades cytosolic single and double-stranded DNA, which can illicit an inflammatory innate immune response (Crow et al., 2006). NEIL2 is a bifunctional glycosylase involved in BER of oxidized bases (Chakraborty et al., 2015;Hazra et al., 2002;Mandal et al., 2012;Wiederhold et al., 2004) and methyl-cytosine demethylation (Schomacher et al., 2016). NEIL2 catalyzes both base removal and scission of the ensuing AP sites but leaves a 3'-phosphate (3'P) that is removed by PNKP (a polynucleotide 3'kinase and phosphatase) to generate a 3'OH that primes Polb (Das et al., 2006;Wiederhold et al., 2004).
In addition, we found a positive correlation between U/G repair-induced mutations and the relative NEIL2 expression (qRT-PCR data) in the four breast cancer cell lines ( NEIL2 participates in A3B-mediated genomic DNA damage The above results indicate that the elevated level of NEIL2 in Hs578T cells sensitizes them to the single strand deaminase activity of A3B during DNA repair. Single-stranded DNA is prone to further damage including DSBs, which can be detected as gH2AX foci (Bonner et al., 2008;Burns et al., 2013a;Landry et al., 2011;Morel et al., 2017;). Therefore, we used gH2AX foci as a proxy for NEIL2-facilitated, A3B-induced genomic damage. Expression of exogenous A3B generated a statistically significant increase in gH2AX foci, which was markedly decreased in NEIL2-depleted Hs578T cells ( Figure 3A). This was not the result of diminished deaminase activity as A3B deaminase activity generated from the A3B-HA expression vector was unaffected in NEIL2-depleted cell lines ( Figure 3B and  Figure 3D). These results indicate that NEIL2 is involved in A3B-induced genomic DNA damage.
(D) U/G repair-induced mutation using the shuttle vector assay in NEIL2-stable-knockdown Hs578T cell line. NEIL2 depletion decreased U/G MM repair-induced mutagenesis. 0 MM, no mismatch; U/G MM, U/G mismatch. Error bars represent s.d., n = 3. * P < 0.05 by two-tailed unpaired Student's t test. (E) Rescue of NEIL2 by NEIL2-HA overexpression vector restores U/G mismatch repair-induced mutation rate in NEIL2-stable-knockdown Hs578T cell line shNEIL2#1 (targets NEIL2 3'UTR). Western blot (left panel) shows NEIL2-HA overexpression for rescue of NEIL2 in shNEIL2-#1 Hs578T cell line. Lamin B1 serves as a loading control. Error bars represent s.d., n = 3. * P < 0.05; ** P < 0.01 by two-tailed unpaired Student's t test. (F) Overexpression of NEIL2-HA (pPM-NEIL2-3'HA) in Figure 2 continued on next page (Hazra et al., 2002), NEIL2 exhibited robust glycosylase/lyase activity on hydroxyl-U-containing single-stranded oligonucleotide (ssDNA-OHU), but only trace activity on double-stranded oligonucleotide (dsDNA-OHU/G) ( APE1 is the conventional BER AP lyase. Therefore, we determined if NEIL2 collaborates or competes with APE1 for AP sites. APE1 and NEIL2 generate respectively 3'OH and 3'P-terminated fragments ( Figure 4A and Figure 4-figure supplement 2), which migrate differently on Urea-PAGE (Schomacher et al., 2016). In a reaction with both NEIL2 and APE1, the NEIL2 product prevailed over the APE1 product (lane 5 of Figure 4B). Here we used amounts of the proteins just sufficient to completely digest the UDG-generated AP substrates from ss-U oligonucleotides ( In contrast to its activity on ssDNA, NEIL2 generated an unexplained double-band product from the dsDNA-U/G oligonucleotides in the concentration gradient assay ( Figure 4C). Double band NEIL2 products have been previously observed (Hazra et al., 2002) and in our case were related to the length and composition of the substrate (35 nt, Figure 4C and 51nt, Figure 4-figure supplement 3C). Though unexplained, these distinct banding patterns provided a convenient way for distinguishing the NEIL2 and APE1 products. To determine whether NEIL2 can compete with APE1 at AP sites generated from dsDNA-U/G oligonucleotides, we gradually increased the concentration of NEIL2 in reactions that contained constant amounts of UDG and APE1. When APE1 was somewhat limiting (0.005U, Low APE1 panel, lanes 4-8, Figure 4D), the amount of NEIL2 products increased with increasing levels of NEIL2. Notably, the same results were found with excess APE1 (0.02U, High APE1 panel, lanes 9-13, Figure 4D). These results strongly suggest that NEIL2 supplants APE1 at AP sites. If NEIL2 was merely inhibiting APE1 activity, then the NEIL2 products would likely prevail only at the lower APE1 concentration.
The NEIL2 product is a poor polb substrate As the Polb reaction is a rate-limiting step during BER (Srivastava et al., 1998), we compared the NEIL2 and APE1 products as substrates for this reaction using His-tagged Polb purified from E. coli (Figure 4-figure supplement 4A). The NEIL2 product retains a 3'P that can be removed by the phosphatase activity of PNKP (Wiederhold et al., 2004). Although Polb showed robust activity on the APE1 product (Figure 4-figure supplement 4B), it was less active on the NEIL2 product, which depended on PNKP ( Figure 4E and illustration in Figure 4-figure supplement 2). As expected from the usurpation of AP sites by NEIL2 ( Figure 4B,D), Polb incorporation was attenuated by NEIL2 ( Figure 4F).
To confirm that our preparation of PNKP contained robust 3'-phosphatase activity, we prepared an oligonucleotide substrate that would yield a 3'-32 P-terminated product after completion of the coupled glycosylase/scission reactions of bifunctional glycosylases (Wiederhold et al., 2004)   Lamin B1 serves as a loading control. Error bar represents s.d., n = 3. * P < 0.05 by two-tailed unpaired Student's t test.
The online version of this article includes the following source data, source code and figure supplement(s) for figure 2: Source code 1. Python script for generating the volcano plot in Figure 2A. Source data 1. Original data for DNA repair enzymes screening.   Immunostaining of gH2AX foci in NEIL2stable-knockdown Hs578T cell lines (shNEIL2#1 and shNEIL2#2) transfected with A3B-3HA. NEIL2 knockdown decreases the A3B-mediated gH2AX foci. EV, empty vector; shSCR, scramble shRNA. Scale bar, 50 mm. Right panel: Percentage of gH2AX foci, showing mean ± s.d., in at least 10 randomly selected microscopic fields in two replicate experiments for each condition. *** P < 0.001 by two-tailed unpaired Student's t test. (B) In vitro deamination assay of nuclear extracts from NEIL2-stable-knockdown Hs578T cell lines with or without A3B-3HA expression. The substrate was a fluorescein-labeled single-stranded oligonucleotide (39 nt) containing -TCT-or -ACT-(negative control). Nuclear extract from HEK293T expressing A3B-3HA (A3B OE) was used as a positive control. NEIL2 knockdown does not affect A3B deaminase activity. Right panel: Quantifications of the cleaved products relative to total DNA loaded onto gel. S, substrate; P, product. (C) Quantification of the percentage of cells with gH2AX foci in NEIL2-stable-knockdown Hs578T cell line (shNEIL2#1) in the absence or presence of a NEIL2 expression vector pcDNA3.1(+)-NEIL2-3'HA. NEIL2 restoration increases A3B-triggered gH2AX foci. Data are represented as mean ± s.d. (n = 10 randomly selected microscopic fields in two replicate experiments). *** P < 0.001 by two-tailed unpaired Student's t test. The corresponding images of gH2AX foci are shown in Data are represented as mean ± s.d. (n = 10 randomly selected microscopic fields in two replicate experiments). ** P < 0.01; n.s., no significant difference by two-tailed unpaired Student's t test.
The online version of this article includes the following figure supplement(s) for figure 3:  . The NEIL2 product retains a 3'P and migrates faster than 3'OH-terminated APE1 product. When both NEIL2 and APE1 are present, only the NEIL2 product is generated (lane 5). S, substrate; P, product. (C) Concentration gradient and product accumulation curves of APE1 and NEIL2 on 5'-[ 32 P]-U-containing dsDNA (35 nt) in the presence of UDG. The volumes of NEIL2 (68 ng/ml) and APE1 (0.005 U/ml) used are listed in the figure. (D) NEIL2 competes with APE1 on 5'-[ 32 P]-U-containing dsDNA (35 nt) in the presence of UDG. The reactions contained either 0.005 U (Low APE1) or 0.02 U APE1 (High APE1) and increasing amounts of NEIL2-His 6 . Lanes 2 and 3 contain respectively 4 ml NEIL2 (68 ng/ml) and 0.02 U APE1. The APE1 cleavage pattern was converted to the NEIL2 pattern with increasing amounts of NEIL2. S, substrate; P, product. (E) Incorporation of [a-32 P]-dCTP by Polb for products generated by APE1, NEIL2, and NEIL2 and PNKP on di-deoxynucleotide (ddC)-modified oligonucleotide in the presence of UDG. P, product. (F) Incorporation of [a-32 P]-dCTP by Polb in the presence of UDG and APE1 as a function of NEIL2 (68 ng/ml, 0, 1, 2, 5, 10 ml) and constant PNKP (127 ng/ml, 2 ml). As the NEIL2 product increasingly dominates the reaction, Polb incorporation decreases. The number under each band gives the intensity relative to that in the first lane. P, product. The online version of this article includes the following figure supplement(s) for figure 4:  Exo1 generates single-stranded substrates vulnerable to A3B and DNA damage from NEIL2 products We previously showed that hijacking of BER by mismatch repair (MMR) provided the wherewithal for Exo1 to generate single-stranded DNA for A3B (Chen et al., 2014). However, while NEIL2 has no glycosylase activity toward U/G (Figure 4-figure supplement 1B), it can interrupt normal BER by displacing APE1 at AP sites. As Exo1 can be recruited to endonucleolytic nicks generated by various means (Wang et al., 2018), we determined whether Exo1 was involved in NEIL2-mediated A3Binduced mutations or gH2AX foci. Exo1 siRNA knockdown ( Figure 5A) reduced production of gH2AX foci in Hs578T cells but had no effect in the NEIL2-stable-knockdown cell lines ( Figure 5B,C), indicating that Exo1 activity occurs downstream of NEIL2. In addition, U/G repair-induced mutations were also decreased in Exo1-knockdown Hs578T cells ( Figure 5D). Thus, knockdown of Exo1 recapitulated the effects of NEIL2 knockdown. Furthermore, the nicked AP products generated by APE1 or NEIL2 were equally good substrates for Exo1 despite the different 5' termini of the nicked DNA ( Figure 5E, 5'dRP for APE1 and 5'P for NEIL2). These results indicate that NEIL2 can divert a BER intermediate to an Exo1-generated single-stranded DNA that is susceptible to A3B deaminase activity, DSBs, or other DNA damage ( Figure 5F).

Discussion
A major unresolved issue in cancer biology is why some cancers become vulnerable to the mutagenic effect of A3 single-stranded deaminases (Alexandrov et al., 2013b;Cescon et al., 2015;Helleday et al., 2014;Petljak et al., 2019;Roberts and Gordenin, 2014), and even more intriguing, that it can be sporadic over clonal lineages of the same tumor (Petljak et al., 2019). Although it may be unlikely that a single mechanism would explain every instance of A3-mediated mutations in cancers, we report here that elevated expression of the bifunctional glycosylase, NEIL2 (Figure 2A), sensitizes Hs578T breast cancer cells to two A3B-mediated effects, repair-induced mutations ( Figure 2B-F) and DNA damage revealed by gH2AX foci (Burns et al., 2013a;Landry et al., 2011) ( Figure 3). NEIL2 depletion mitigates both effects, but they are induced by its overexpression. In vitro BER experiments using purified components show that NEIL2, though unable to process U/G mismatches, subverts the BER process that repairs these lesions by usurping the normal BER endonuclease, APE1, at AP sites (the first BER product) (Figure 4). Although we have not determined why the NEIL2 scission product is a poor substrate for Polb, the important issue is that it is vulnerable to the 5'À3' exonuclease Exo1 ( Figure 5E), which generates single-stranded DNA susceptible to A3 deaminases. In vivo experiments show that Exo1 and NEIL2 double-knockdown mimics the effect of NEIL2 depletion, indicating that NEIL2 acts upstream of Exo1 ( Figure 5A-C).
Therefore, the most parsimonious explanation of our results is that NEIL2 diverts BER to Exo1generation of single-stranded DNA that would be vulnerable to A3B deaminase ( Figure 5F). It is also important to stress that the eventual outcome at a given AP site is not likely to be only a function of the relative intracellular concentrations of each protein. APE1 does have other binding partners (Bazlekowa-Karaban et al., 2019;Madlener et al., 2013;Tell et al., 2009;Thakur et al., 2014) and we presume this is likely also to be true of NEIL2 (Das et al., 2007). In addition, our results shows that NEIL2, even at a concentration of APE1 that is in four-fold excess of the amount needed to completely digest an AP site, can enzymatically outcompete APE1 ( Figure 4D). The inherent reductive power of a biochemical experiment using purified components is its ability to reveal the prevailing baseline conditions of a process ( Figure 5F).
we used, and only the non-mutagenic, low NEIL2-exspressing MDA-MB-453 contained a SNP (rs804271) that had been previously correlated with elevated NEIL2 expression and DNA damage in a BRCA1/2 background (Benítez-Buelga et al., 2017). Thus, additional factors contribute to the regulation of NEIL2 in these cells. We found a positive correlation between U/G repair-induced mutations and relative NEIL2 expression (Figure 2-figure supplement 2A,B), but no relationship to endogenous A3B expression ( Figure 1B), in four breast cancer cell lines. Interestingly, it was recently reported that clonal lineages of MDA-MB-453 can undergo sporadic episodes of A3-mediated mutations (Petljak et al., 2019). Thus, it seems that our ATCC isolate of this line (HTB-131) was in a low mutation phase. However, both high and low mutagenic clonal lines from the same lineage of MDA-MB-453 are available (Petljak et al., 2019) and would seem to provide the ideal experimental material to investigate by the approaches and methods we report in this paper. Given the heterogeneity of tumor samples, cell lines derived from the breast cancer tissues should also be useful sources of experimental material for investigating the relationship between NEIL2 expression and the mutational processes in tumors.

Mismatch plasmid construction
Mismatch plasmids were constructed as previously described (Chen et al., 2014). For preparation of the gapped vector, we digested 60 mg pSP189-SnA plasmid (Chen et al., 2014) with 120 U nicking endonuclease, Nt.BbvCI (NEB), at 37˚C overnight, followed by hybridization with 1,200 pmol biotinylated-complementary oligo at 37˚C for 1 hr. The hybridized product was captured by 3 mg Streptavidin Magnetic Particles (Roche, 11641786001) with rotation at 37˚C for 2 hr. The gapped vectors were then purified by phenol/chloroform/isoamyl alcohol (25:24:1, PCI) extraction and ethanol precipitation. For 0 MM and U/G MM reconstitution, 540 pmol C-or U-containing oligonucleotides were annealed at 100-fold molar excess to 18 mg gapped vectors in annealing buffer (100 mM KOAc, 30 mM HEPES, pH 7.5) by incubation at 95˚C for 3 min, 40˚C for 4 hr, 35˚C for 30 s, 30˚C for 30 s, and then kept at 25˚C. Annealed samples were ligated with T4 DNA ligase (NEB, M0202S) at room temperature (RT, 25˚C) for 2 hr, followed by incubation with 0.2 U Klenow Fragment (3'exo -,NEB, M0212), 100 mM dNTP, and 1Â NEBuffer 2 at 37˚C for 10 min to repair any remaining gapped plasmids, which are highly mutagenic (Chen et al., 2014). The mismatch plasmids were purified by PCI extraction and ethanol precipitation. The gapping and ligation efficiencies were monitored by KpnI digestion (one of the two sites is lost after gapping but restored after ligation, Figure 1-figure supplement 1A).

Determination of repair-induced mutations
Control (0 MM) and mismatch (U/G MM)-containing plasmids (1.5 mg per well in 6-well plate) were transfected into breast cancer cells, extracted 48 hr later using the Wizard Plus SV Miniprep kit (Promega, A1460), and treated with DpnI (NEB, R0176) for 15 min at 37˚C. Plasmids (5 ng) were electroporated into 20 mL MBM7070 competent cells. After transformation, the cells were recovered in S.O.C. medium (Invitrogen, 15544-034) for 1 hr at 37˚C and plated on LB agar plates containing 100 mg/mL carbenicillin (Sigma-Aldrich, C1389), 1 mM IPTG (Invitrogen, 15529-019), and 0.03% (w/ v) Bluo-Gal (Invitrogen, 15519-028). After incubation at 37˚C overnight, the plates were stored at 4C in the dark to allow color development. The percentage of white to total colonies (2,000-3,000) per sample was calculated as 'mutation rate'. For mutation fate and trinucleotide context analysis shown in Figure 1D,E, we sequenced the reporter region ( Figure 1A) of the pSP189-SnA shuttle vector in all white colonies (ACGT, Inc) using primer R250 (TTTTTGTGATGCTCGTCAGG) (Chen et al., 2014). Mutations were tabulated from the alignments between these sequences and the reference sequence of the starting reporter region.

DNA repair enzymes screening and analysis
Breast cancer cells were harvested and cDNA was generated as described in the section of 'RNA isolation and qRT-PCR'. The cDNA was quantified by quantitative real-time PCR (qRT-PCR) using RT 2 Profiler PCR array Human DNA Repair (Qiagen, PAHS-042Z). Raw Ct values were normalized to two housekeeping genes, GAPDH and RPLP0, to determine DCt values. MDA-MB-453 DCt values were subtracted from Hs578T DCt values to obtain DDCt values. Log 2 (fold change) (calculated from log 2 (2 -DDCt )) was plotted as a function -log 10 (p-value) to generate the volcano plot (Figure 2A). These values were calculated from DCt values (n = 4 replicates) using the ttest_ind function from the Python scipy stats package. Only log 2 (fold change) values above 2 or below À2 and -log 10 (p-value) > 2 were counted as significant. The python script for generating the volcano plot in Figure 2A has been uploaded as an additional data file.

RNAi
As preliminary experiments showed no difference in the reduction of NEIL2 transcripts between 10, 20 or 50 nM siRNA transfected with Lipofectamine RNAiMAX (ThermoFisher Scientific, 13778030) using the manufacturer's protocol (Figure 2-figure supplement 1B), we routinely used 10 nM siRNA unless stated otherwise. Cells were harvested 72 hr post-transfection for Western Blot or qRT-PCR analysis. The sequence information for siSCR, siA3B, siA3C, siNEIL2, siTREX1, and siExo1 can be found in the key resources table.

RNA isolation and qRT-PCR
Total RNA was extracted using the PureLink RNA Mini Kit (Invitrogen, 12183018A), and reverse transcribed using the SuperScript III First-Strand Synthesis System (Invitrogen, 18080-051) following the manufacturer's instruction. The qRT-PCR was performed on StepOnePlus Real-Time PCR System (Applied Biosystems) using TaqMan gene expression assays (listed in the key resources table) and TaqMan Fast Universal PCR Master Mix (Applied Biosystems, 4352042). The 2 -DDCt method (Livak and Schmittgen, 2001) was used to analyze the data.

Lentivirus packaging and generation of stable knockdown cell lines
We used the 3 rd generation packaging system following the addgene pLKO.1-TRC Vector protocol. LentiX-293T cells were transfected with 4 mg pLKO.1 vector containing short hairpin RNA (shRNA) in a 10 cm dish plate along with three packaging plasmids: pMDLg/pRRE (4 mg), pRSV-Rev (2 mg), and pMD2.G (2 mg). Media were replaced with 6 mL fresh complete media 24 hr post-transfection. After another 48 hr, media were harvested and centrifuged at 1,500 Âg for 10 min to remove cells and debris. The viral particles-containing supernatant and 7.5 mg/ml polybrene (Millipore, A-003-E) were added to infect Hs578T cells cultured in 6-well plate (100 ml per well). Media were replaced 24 hr after infection, and fresh media with 2 mg/ml puromycin (Sigma-Aldrich, P8833) were added another 24 hr later to select transfected cells.
The underlined nucleotides denote restriction sites.

Immunofluorescence
Hs578T cells seeded in 6-well plates were transfected with 1.5 mg phAPOBEC3B-HA (A3B-3HA) plasmid, and 24 hr post-transfection, the cells were seeded into Lab-Tek II Chamber Slide System (Nunc, 177380). After another 24 hr, the cells were washed with pre-warmed PBS, and fixed in 4% Formaldehyde (ThermoFisher Scientific, 28906, diluted with PBS) at 37˚C for 15 min. Cells were washed three times with PBS and then permeabilized in 0.1% Triton X-100 (Sigma-Aldrich, T8787) at RT for 10 min, followed by another wash with PBS. For immunostaining, cells were blocked with 1% BSA (in PBS) at RT for 30 min, and incubated with anti-gH2AX primary antibody overnight at 4˚C. After washing with PBS three times, cells were blocked with 1% BSA (in PBS) at RT for 30 min and then incubated with Alexa Fluor 568 anti-Rabbit IgG secondary antibody at RT for 2 hr. Cells were washed with PBS three times, mounted with Prolong Diamond Antifade Mountant with DAPI (Life technologies, P36962) and cured overnight at RT. Slides were imaged using a Keyence Digital Microscope and images were analyzed with Fiji (ImageJ) software using identical acquisition parameters for all images.
Effect of expressing NEIL2-HA in NEIL2-stable-knockdown Hs578T cells NEIL2-stable-knockdown (shNEIL2#1) and the negative control scramble shRNA-stable Hs578T cell lines were transfected with 1 mg pcDNA3.1(+)-NEIL2-3'HA or empty vector on the same day of cell seeding in 6-well plates. After 16 hr, the cells were transfected with 1.5 mg phAPOBEC3B-HA. The cells were re-seeded into a Nunc Lab-Tek II Chamber Slide System and underwent the gH2AX immunofluorescence procedure 24 hr later as described above.

DNA deaminase activity
Deamination assays were performed as previously described (Byeon et al., 2016;Mitra et al., 2014). For nuclei isolation, breast cancer cells were harvested, washed with 1Â PBS. The cells were resuspended in hypotonic buffer (20 mM Tris, pH7.5, 0.1 mM EDTA, 2 mM MgCl 2 ) and incubated at RT for 2 min and then on ice for 10 min, followed by adding 1/10 vol of 10% NP-40 (ThermoFisher Scientific, 28324) and centrifugation at 1,000 Âg for 5 min at 4˚C to pellet the nuclei. Extracts of HEK293T cells expressing A3B-3HA or isolated breast cancer nuclei were prepared by lysis in M-PER mammalian protein extraction reagent (ThermoFisher Scientific, 78501), supplemented with protease inhibitor cocktail (Roche, 04693159001) and 100 mM NaCl (final concentration). After lysis, glycerol (final concentration, 10%, vol/vol) was added and the resulting mixture was centrifuged at 13,000 Âg for 10 min at 4˚C. Total protein concentration was quantified by Pierce BCA Protein Assay Kit (ThermoFisher Scientific, 23225). Whole cell or nuclear extracts were incubated with RNase A (ThermoFisher Scientific, EN 0531) at 37˚C for 15 min before adding 500 nM fluorescein-labeled oligonucleotides (39 nt, synthesized by the Midland Certified Reagent Company, Inc) and 2U UDG (NEB, M0280) in 10 mM Tris-HCl pH 8.0, 50 mM NaCl, 1 mM DTT, 1 mM EDTA. Reactions were incubated at 37˚C for up to 5 hr followed by treatment with 150 mM NaOH at 37˚C for 20 min. After heating at 95˚C for 3 min, samples were immediately chilled on ice and purified by PCI extraction and ethanol precipitation. Samples were heated for 3 min at 95˚C with an equal volume of 2Â Novex TBE-Urea Sample Buffer (Invitrogen, LC6876), separated by 12% 8M Urea PAGE gels, and imaged using the Fujifilm FLA-5100 (Fujifilm Life Science). 39 nt -TCT-:5'-fluorescein-AATAATAATAATAATAATTCTAATAATAATAATAATAAT-3' 39 nt -ACT-:5'-fluorescein-AATAATAATAATAATAATACTAATAATAATAATAATAAT-3'
For competition assays between NEIL2 and APE1, cleaved products were separated using 20% 8M Urea PAGE. For double-stranded DNA (dsDNA), 1 pmol 32 P-labeled dsDNA (35 nt) was mixed on ice with UDG (1 U/ml, 1 ml), various concentrations of purified NEIL2 or APE1 (indicated in figures and legends), and 1Â NEBuffer 4 in 20 ml reactions. After incubation at 37˚C for 30 min, the reaction products were extracted by PCI followed by ethanol precipitation. The samples were treated with an equal volume of 2Â Novex TBE-Urea Sample Buffer (Invitrogen), heated at 95˚C for 3 min, and chilled on ice before subjecting to 20% 8 M Urea PAGE. 51-nt single strand U:

3' phosphatase activity of PNKP
A 26 nt oligonucleotide with U at the 5' end was labeled with 32 P by T4 PNK (NEB) as described above. As illustrated in Figure 4-figure supplement 4C, the oligonucleotide was annealed to a 51 nt complementary oligonucleotide, and then annealed to a 25 nt oligonucleotide in annealing buffer (10 mM Tris, pH 7.5, 50 mM NaCl, 1 mM EDTA), followed by ligation with T4 DNA ligase (NEB) to form an internally-labeled duplex oligonucleotide (designated S, Figure 4-figure supplement 4C).

Exo1 nuclease assay
The dsDNA substrates for the resection assay were 85 nt oligonucleotides, which contained a 5'phosphorothioate modification to block end resection by exonuclease. The 85 nt U-containing (U at position 43) oligonucleotides were 3'-labeled with 32 P-cordycepin and then annealed with the bottom strand before treatment with UDG and APE1 or NEIL2 as described above. After PCI purification and ethanol precipitation, 0.1 pmol dsDNA substrates were incubated with Exo1 (human Exo1 protein, kindly provided by Dr. Tanya Paull at the University of Texas at Austin) (Myler et al., 2016) resection assay in a 10 ml reaction as previously described (Keijzers et al., 2015). The reactions were carried out at 30˚C for 15 min and then purified by PCI extraction and ethanol precipitation. The products were mixed with an equal volume of 2Â Novex TBE-Urea Sample Buffer (Invitrogen) and heated at 90˚C for 3 min. Samples were subjected to 20% Urea PAGE and exposed to X-ray film. Top_U_5'phosphorothioate_85: 5'G*A*C*AGGATCCGGGCTAGCATCTTCATACGCCCTGCAGG TCGAUTCTAGAGGATCCCCGGGTACCTTCATACATAGTTGTCACTGG3' Bottom_G_5'phosphorothioate_85: 5'C*C*A*GTGACAACTATGTATGAAGGTACCCGGGGATCC TCTAGAGTCGACCTGCAGGGCGTATGAAGATGCTAGCCCGGATCCTGTC3' * denotes phosphorothioate

Quantification and statistical analysis
Statistical details, including the statistical methods, n values, definition of significance, and definition of mean value and dispersion were indicated in the figure legends. Statistical analyses were carried out using GraphPad Prism 8. Additional files

Supplementary files
. Transparent reporting form

Data availability
All data generated or analysed during this study are included in the manuscript and supporting files.