DNMT3b protects centromere integrity by restricting R-loop-mediated DNA damage

This study used DNA methyltransferase 3b (DNMT3b) knockout cells and the functional loss of DNMT3b mutation in immunodeficiency-centromeric instability-facial anomalies syndrome (ICF) cells to understand how DNMT3b dysfunction causes genome instability. We demonstrated that R-loops contribute to DNA damages in DNMT3b knockout and ICF cells. More prominent DNA damage signal in DNMT3b knockout cells was due to the loss of DNMT3b expression and the acquirement of p53 mutation. Genome-wide ChIP-sequencing mapped DNA damage sites at satellite repetitive DNA sequences including (peri-)centromere regions. However, the steady-state levels of (peri-)centromeric R-loops were reduced in DNMT3b knockout and ICF cells. Our analysis indicates that XPG and XPF endonucleases-mediated cleavages remove (peri-)centromeric R-loops to generate DNA beaks, causing chromosome instability. DNMT3b dysfunctions clearly increase R-loops susceptibility to the cleavage process. Finally, we showed that DNA double-strand breaks (DSBs) in centromere are probably repaired by error-prone end-joining pathway in ICF cells. Thus, DNMT3 dysfunctions undermine the integrity of centromere by R-loop-mediated DNA damages and repair.


INTRODUCTION
DNA methyltransferase DNMT3b carries out de novo DNA methylation and is essential for mammalian development [1,2]. The murine embryonic fibroblasts derived from DNMT3b knockout embryo display DNA damage and chromosome instability [1,3], suggesting the critical function of DNMT3b in genome stability. Loss-of-function mutation in DNMT3b is specifically found in a rare human genetic disorder, immunodeficiency-centromeric instability-facial anomalies (type-1 ICF) syndrome [4][5][6]. It has been shown that DNMT3b is recruited to GC-rich (peri-)centromere regions by interacting with centromere protein CENP-C for maintaining chromosome stability [7]. In accordance, the major genome regions affected by the loss of DNMT3b function in ICF are non-coding repetitive elements surrounding centromeres, where GC regions are hypo-methylated [8], coinciding with centromeric DNA breaks observed in ICF cells [9,10]. However, it is still an open question how DNMT3b dysfunction increases DNA damage and centromere instability.
R-loop is a three-stranded nucleic acid structure consisting of an RNA-template DNA hybrid and a non-template single-stranded DNA [11]. In mitosis, centromeric domain remains transcriptionally active [12]. It has been reported that centromeric R-loop has a function in mediating ATR binding for checkpoint control in mitosis [13], suggesting the beneficial role of R-loops in a cell. However, R-loops are also considered the source of DNA damage [14,15]. R-loops can be resolved by helicases, translocase, or removed by RNase H [16][17][18]. Without a resolution, R-loops lead to DSBs through replication fork collapse [14] or DNA cleavage by endonucleases XPG and XPF [19,20]. In addition, it has been shown that chromatin modifications regulate R-loop-induced genome instability [21][22][23][24]. Since DNMT3b is an epigenetic factor that regulates transcription, we asked the question of whether the levels and processing of R-loops are involved in DNMT3b dysfunction-mediated chromosome instability.
In this study, we addressed the question by using human colon cancer HCT116-BKO cells in which two alleles of DNMT3b disrupted by homologous recombination and ICF cells that carry loss-of-function mutations in DNMT3b. Our analysis demonstrated that R-loops contribute to prominent DNA damage signals observed in both DNMT3b knockout in HCT116 (BKO) and the loss-of-function mutation of DNMT3b in ICF lymphocytes. DNA damage sites in BKO cells were mapped to repetitive satellite sequences and rDNA genes. We focused on understanding the role of DNMT3b in centromere instability. In BKO and ICF cells, (peri-)centromeric R-loops are cleaved and removed by endonucleases XPG and XPF. Depletion of XPG and XPF elevated R-loops while reduced γH2AX associated with (peri)-centromeric DNA sequences in BKO and ICF cells. Finally, we examined the choice of DNA double-strand break (DSB) repair pathway in centromeric breaks in ICF cells. Our data indicate that error-free homologous recombination (HR) of DSB repair at centromere sites is not activated in ICF cells. Instead, the non-homologous end-join process (NHEJ), which might cause error-prone repair [25]. Here, we proposed that DNMT3b dysfunction promotes XPG/XPFmediated DNA breaks at (peri-)centromeric R-loops sites, where the repair of DSBs via NHEJ increases centromeric shortening and fusion that undermine centromere stability in ICF cells.

RESULTS
DNMT3b dysfunctions increase R-loops-dependent DNA damage To assess the importance of DNMT3b in genome instability, the levels of DSBs in HCT116 and BKO cells were compared by γH2AX IF staining. Data showed that the γH2AX signal was very prominent all over the nuclei in BKO cells (Fig. 1a), so as other DNA damage response signals, including phospho-ATM, -Chk2, and -Chk1 (Fig. 1b). Enforced expression of HA-RNase H1 significantly reduced the γH2AX signal in BKO cells (Fig. 1c). In contrast, the expression of HA-RNase H-D209N, a catalytic-dead mutant, had no effect (Fig. 1c). Since R-loops are the results of transcription, we then treated BKO cells with α-Amanitin (20 μg/ml) or cordycepin (50 μM) for 6 h and found that γH2AX intensity was clearly reduced (Fig. S1). These results suggest that R-loops contribute to the increase of DNA damage in BKO cells.
A previous report has shown that transcription-coupled nucleotide excision repair (TC-NER)-mediated endonuclease cleavage of unprocessed R-loops promotes genome instability [19,20]. We then depleted Cockayne Syndrome B (CSB) protein, a critical TC-NER factor in BKO cells. The overall γH2AX intensity was reduced in BKO cells after the knockdown of CSB, suggesting the involvement of NER in making DSBs in these cells (Fig. S2). It has been suggested that in the S phase, DNA gaps generated by TC-NER processing cause DNA replication fork collapses and DSBs, which activate γH2AX [26]. The flow cytometric analysis, showed that BKO cells contained more S phase cells (Fig. S3a). The γH2AX signal associated with the EdU incorporation-positive cells indicated by Click-iT conjugation was obviously higher in BKO cells (Fig. S3b). The DNA fiber analysis of replicating DNA further revealed that the lengths of replication tracks were much shorter in BKO than those in HCT116 cells (Fig. S3c). Altogether, these data suggest that the replication stress in the S phase is associated with DNA damage in BKO cells.
Next, we tested whether R-loops also cause DNA damage in ICF cells. To this end, we first compared γH2AX IF staining in EBVimmortalized lymphoblastoid cell lines (LCLs) from wild-type and a type-1 ICF patient carrying DNMT3b mutation (Fig. 1d). The results showed that the levels of γH2AX foci were higher in ICF than those in wild-type LCLs. We then used retroviral infection to express RNase H1 in ICF LCLs. The amounts of γH2AX foci were clearly decreased by wild-type but not catalytic-dead RNase H1 in ICF cells (Fig. 1e). In conclusion, either deficiency or functional loss of DNMT3b increases R-loops-dependent DNA damage.
It is worthy of noting that the intensity of γH2AX was much more prominent in BKO than in ICF cells. Despite of high DNA damage signal, BKO cells still grew well in the culture. The sequencing of a panel of genes associated with cancer indicated p53 mutation in BKO cells (Fig. 2a). Since BKO cells were generated by flox-mediated knockout from HCT116, which is a p53 proficient line, it is likely that R-loop-mediated DNA damages drive the acquirement of p53 mutation in BKO cells. We found that tet-on induced expression of p53 by doxycycline treatment caused BKO cell growth arrest and cell death (Fig. 2b). Thus, the selection of p53 mutation allows BKO cell survival and proliferation. In addition, re-expression of both DNMT3b and p53 reduced the DNA damage signal in BKO cells (Fig. 2c). Very likely, the loss of functional p53 and DNMT3b expression together contribute to sustaining a prominent level of DNA damage in BKO cells. As for ICF cells, the intensity of the DNA damage signal was less, probably because p53 was still functional and DNMT3b gene was not truncated. Knockdown of p53 in ICF cells was performed to test the contribution of functional p53 to the level of endogenous DNA damage. The results showed that p53 knockdown clearly increased γH2AX foci in ICF cells (Fig. 2d). Thus, p53 plays a role in determining the extent of DNA damage induced by the loss of DNMT3b function.
R-loops-dependent DNA damages at (peri-)centromere To map DNA damage sites in BKO cells, we performed a ChIPsequencing analysis using the γH2AX antibody. This genome-wide search showed higher levels of γH2AX associated with satellite regions of telomere, centromere, peri-centromere, and rDNA in BKO cells (Fig. 3a). These data are correlated with transcription abnormalities previously observed at (peri-)centromere in DNMT3bdepleted HCT116 cells [7], telomeres in ICF cells [27], and rDNA in DNMT1/DNMT3b-disrupted HCT116 cells [28]. We further validated ChIP-seq data by ChIP-qPCR of these repetitive sequences. Several high complexity regions of (peri-)centromere were chosen for primer design. The results confirmed more DSBs in satellite regions of rDNA, telomere, and (peri-)centromere in BKO cells (Fig. 3b).
We also performed a ChIP-qPCR analysis to know whether ICF cells have more γH2AX binding to (peri-)centromere. The ChIP data revealed higher levels of γH2AX associated with centromere core sequences of chromosomes 1 and 4 and peri-centromeric sequences of chromosomes 1 and 22 in ICF LCL than in normal LCL (Fig. 3c). SNRPN locus has been commonly used as a negative control of R-loop hybrid [13,29]. No difference in γH2AX ChIP at SNRPN was found in WT and ICF LCLs. To test the causal role of R-loops in (peri-)centromeric DSBs, we then expressed RNase H1 by retroviral infection in ICF cells. γH2AX-ChIP analysis showed that expression of wild-type of RNase H1 significantly reduced γH2AX association with (peri-)centromeric sequences in ICF cells, while catalytic-dead RNase H1 had little effect (Fig. 3d). These data suggest that R-loop formation increases γH2AX associated with (peri-)centromeric sequences.
XPG-and XPF-mediated cleavages remove (peri-)centromeric R-loops and generate DNA damages in DNMT3b defective cells XPF and XPG are the endonucleases of TC-NER. We then depleted XPF and XPG, by which γH2AX IF staining in BKO cells was abolished (Fig. 4a). Consistently, ChIP-qPCR confirmed that γH2AX binding at (peri-)centromeric sites was also reduced by XPG and XPF knockdown (Fig. 4b). In ICF LCLs, depletion of XPG and XPF also abolished γH2AX foci (Fig. 4c). XPG/XPF knockdown did not affect the protein levels of 53BP1 and H2AX, excluding the possibility of their expression regulated by XPG/XPF. The comet assay also showed the reduction in the tail moment by XPG/XPF knockdown (Fig. S4). The γH2AX-ChIP-qPCR analysis also showed higher levels of γH2AX association with (peri-)centromeric sequences in ICF cells were also brought down by XPG/XPF depletion (Fig. 4d).
Centromere integrity is important in chromosome segregation. We then asked whether XPG/XPF knockdown can affect chromosome segregation in ICF cells. To this end, ICF cells were treated with nocodazole overnight. By washing out nocodazole, cells were allowed to have mitotic progression for chromosome segregation analysis. The amounts of lagging and bridge chromosome in mitosis were evaluated. The population of cells with lagging and bridge chromosomes in ICF cells was diminished by XPG/XPF knockdown (Fig. 4e). Thus, the loss of DNMT3b function promotes R-loop-mediated DNA damage via XPG/XPF cleavage, which contributes to chromosome instability.
NER-mediated removal of (peri-)centromeric R-loops and Pol II stalling in DNMT3b knockout and ICF cells Next, the levels of (peri-)centromeric R-loops were compared in HCT116 and BKO cells. We then performed DRIP analysis using  DNA-RNA hybrid recognition antibody S9.6 followed by qPCR of (peri-)centromeric region. After nucleic acid extraction, all samples were immuno-precipitated by S9.6 antibody. Samples pretreated with RNase H were used as the control. DRIP readouts normalized by those with RNase H pretreatment indicate the level of R-loop.
Unexpectedly, the analyses revealed that the steady-state levels of (peri-)centromeric R-loop in HCT116 was indeed higher than those in BKO cells (Fig. 5a). DRIP analysis showed that XPG/XPF depletion brought up the level of centromeric R-loop in BKO but not WT HCT116 cells (Fig. 5b). Thus, in BKO cells, (peri-) H.-T. Shih et al.
centromeric R-loops are removed by NER-mediated cleavage to generate DNA damages. These results explained why the steadystate levels of (peri-)centromeric R-loops was lower in BKO cells.
Considering the recruitment of NER factors involves the stalling of RNA polymerase II (pol II), we then examined the association of pol II with (peri-)centromere sequence. The results showed that BKO cells had higher amount of pol II binding to (peri-) centromeric sequence (Fig. 5c). Taken together, these results suggest that the loss of DNMT3b expression causes pol II stalling in (peri-)centromeric regions, which not only generates R-loops but also increases NER recruitment to give DNA breaks.
Like WT and BKO HCT116 cells, the levels of (peri-)centromeric R-loops were higher in normal than those in ICF LCL cells (Fig. 6a). Of note, the DRIP assay for evaluating the level of DNA-RNA hybrid at the (peri-)centromeric sequences was confirmed by the inverse effect by ectopic expression of WT and catalytic-dead RNase H1 in ICF cells (Fig. S5). The level of DNA-RNA hybrid at the (peri-) centromeric sequences was also elevated by XPG/XPF knockdown in ICF but not normal cells (Fig. 6b). However, it was noted that the levels of (peri-)centromeric DNA-RNA hybrid in normal and ICF LCLs after XPG/XPF knockdown were similar. In addition, pol II ChIP analysis showed that no difference in pol II association in these sequences in normal and ICF cells (Fig. 6c). These data suggest that DNMT3b functional defect in ICF cells probably does not increase R-loop formation but only causes (peri-)centromeric R-loops more susceptible to NER-mediated cleavage.
The lack of HR repair in centromere It has been reported that DNMT3b mutation in ICF leads to centromere shortening [9,10]. Presumably, endogenous DSBs are constantly repaired to prevent genome deterioration. We suspected that the repair of DNA breaks might lead to centromere shortening. Fig. 2 The loss of DNMT3b and the acquired p53 mutation cause prominent DNA damage in HCT116 cells. a NGS targeting sequencing of 54 genes in HCT116 and BKO cells. HCT116 and BKO shared common mutations in four genes. Additional mutations at p53 (TP53) were found in BKO cells. b BKO cells were infected with Tet-on Flag-p53-IRES-GFP virus followed by treatment with or without 2 μM doxycycline for 3 days for western blot (left) and viability assay (right, means ± SEM, n = 3, *** indicated P < 0.001 by two-tailed unpaired Student's t-test. ns. P > 0.05). c BKO cells stably expressing Flag-DNMT3b were selected and infected with Flag-p53-IRES-GFP virus. Afterward, cells were treated with or without 1 μM doxycycline for 48 h and fixed for γH2AX IF staining (scale bar, 10 μm, left). Relative intensity of γH2AX in cells (n > 80), *** P < 0.001 by Mann-Whitney test (upper right). Western blots of Flag-DNMT3b, p53, p21, and Actin (bottom right). d ICF LCLs were transfection with 2 μg control siRNA or p53 siRNA using Amaxa® Cell Line Nucleofector® Kit V. After post-transfection at 48 h, cells were fixed for γH2AX IF staining (scale bar, 10 μm, left). Fluorescent intensity of γH2AX in cells (n > 100) was quantitated by Image J and relative intensity is expressed, ***P < 0.001 by the Mann-Whitney test (middle). Western blots of p53 and GAPDH (right).

Fig. 3
R-loops-dependent DNA damages at repetitive sequences. a Genome-wide sequencing of DNA damage sites in HCT116 and BKO cells. ChIP by γH2AX antibody was used for sequencing analysis. The plot shows the relative ChIP-seq reads of γH2AX at indicated genomic regions in BKO versus control HCT116 cells. Data were from one experiment. b HCT116 and BKO cells were used for γH2AX-ChIP-qPCR analysis. Data are shown as the percentage of input DNA in γH2AX antibody at the sequences of (peri-)centromere, rDNA, telomere, and control p21 (CDKN1A) regions (mean ± SEM, n = 3, ***P < 0.001, ns: no significant by two-tailed unpaired Student's t-test). c γH2AX-ChIP-qPCR in wild-type and ICF LCL cells. γH2AX-ChIP-qPCR was analyzed at the sequences of (peri-)centromere of chromosome and intergenic region downstream of SNRPN. Data are expressed as percentage of input DNA (mean ± SEM, n = 3, *, *** P < 0.05, 0.001, ns: no significant by twotailed unpaired Student's t-test). d γH2AX-ChIP-qPCR analysis in ICF LCLs after infection with retrovirus of empty HA-vector, HA-RNaseH1-WT, or HA-RNaseH1-D209N. Data are expressed as percentage of input DNA (mean ± SEM, n = 3, *, ***P < 0.05, 0.001, ns: no significant by twotailed unpaired Student's t-test).
DSBs are repaired by homologous recombination (HR) or nonhomologous end-join (NHEJ) repair pathways [30]. HRR pathway involves an extended BRCA1-dependent end resection and loading of Rad51 for strand invasion to give error-free repair [31]. 53BP1 recruitment to DSB sites antagonizes the HRR pathway and favors NHEJ repair [31][32][33]. By confocal microscopic analysis, γH2AX and 53BP1 were well co-localized with the IF signal of anti-centromere antibody (ACA) that marks centromere sites. In contrast, little Rad51 foci were co-localized with ACA signal in ICF cells (Fig. 7a), suggesting the lack of ongoing HRR at centromeric sites. We then tested NHEJ in the repair of DNA damage in ICF cells. Two NHEJ pathways, conventional NHEJ (c-NHEJ) and alternative-end joining (Alt-EJ), participate in the repair of two-ended and one-ended DSBs [34]. NU7441, an inhibitor of DNA-PK, specifically suppresses c-NHEJ. AZD2281, a PARP1 inhibitor, suppresses multiple repair pathways including alt-EJ and c-NHEJ. ICF cells were then treated with NU7441 and AZD2281 to distinguish the contribution of these two NHEJ pathways in repairing endogenous DSBs. AZD2881 treatment significantly increased overall 53BP1 foci. The amounts of these foci co-localized with ACA were also increased. NU7441 treatment gave a much lesser effect on 53BP1 foci and their co-localization with ACA (Fig. 7b). Comet assay also showed that AZD treatment increased the length of the tail moment in ICF cells, confirming the contribution of PARP1 activity in reducing DNA damage in ICF cells (Fig. 7c). We further performed ChIP analysis by 53BP1 antibody in ICF LCLs with and without AZD treatment. The result showed that inhibition of PARP1 significantly increased 53BP1 binding at the centromere site of chromosome 1 (Fig. 7d), suggesting that the centromere DNA breaks are constantly repaired in a PARP1dependent manner. As such, AZD treatment decreased the viability of ICF cells, whereas XPG/XPF knockdown ICF cells were less susceptible to this treatment (Fig. 7e). Likely, inhibition of PARP1 accumulates DSBs in the centromere. This could also occur in the repetitive regions such as rDNA and telomere, leading to cell death in ICF cells. These data suggest that DNA lesions at centromeric sites might be frequently repaired by the PARP1-mediated DNA repair pathway. Given the lack of error-free HRR, our data imply that XPG/ XPF-induced DSBs in centromere and the participation of NHEJ repair leads to centromere instability in ICF cells.

DISCUSSION
This study investigated DNA damage in DNMT3b deficient HCT116 cells and the loss of DNMT3b function in ICF cells. Despite prominent differences in the intensity of DNA damage signals in type-1 ICF and BKO cells, the signals in these cells were R-loops-dependent. It has been found that DNMT3b dysfunction mainly causes hypomethylation in repetitive sequences rather than promoters of coding genes [1,35]. Consistently, we found that R-loop-mediated DNA damages are spreading over these repetitive sequences at the centromere, telomere, and rDNA regions. Since XPG/XPF knockdown was able to abolish DNA damage signal in either BKO or ICF cells, it is apparent that R-loop-mediated DNA breaks in these sequences are the results of DNA cleavages by these two DNA endonucleases in the NER process. Moreover, our data also point out that a prominent DNA damage signal observed in BKO cells involves the loss of DNMT3b together with the acquired p53 mutation.
DNA hypomethylation not only affects transcription but also histone modification and recruitment of chromatin factors [36]. Our findings further addressed the question of whether the induction of R-loop-mediated DNA damages by DNMT3b dysfunction is through generating too many R-loops or having more DNA cleavages by XPG/ XPF. The steady-state levels of (peri-)centromeric R-loops are higher in HCT116 and normal LCLs as compared to those in BKO and ICF cells, respectively. After XPG/XPF depletion, the centromeric R-loops were significantly increased in BKO and ICF but not HCT116 and normal LCLs. Thus, it is the NER-mediated cleavage in BKO and ICF cells that decrease the levels of centromeric R-loops. Depletion of XPG and XPF caused the level of centromeric R-loop in BKO cells higher than in HCT116 cells. The amount of Pol II binding to these regions was also higher in BKO cells. Contrarily, in normal and ICF LCLs after XPG/XPF knockdown, the levels of centromeric R-loops were similar. Therefore, the loss-of-function DNMT3b mutation increases the removal of centromeric R-loop by the NER process, rather than having excessive amounts of R-loop formation. In addition, we did not find an increase of pol II binding in ICF cells. Accordingly, we proposed that DNA methylation by DNMT3b at (peri-)centromere R-loop sites mediates an as-yet-unknown process in restricting the recruitment of NER factors, thus preventing R-loops in centromere from XPG/XPF-mediated processing. ICF cells defective in DNMT3b function, therefore, have more NER-dependent DSBs in the centromere, leading to centromere instability.
This study also investigated the repair of these R-loops-mediated DSBs in centromere. We did not find Rad51 foci, an HHR marker, localized at centromere sites. In contrast, 53BP1 foci were localized at centromeric sites, which would prevent HRR but favor NHEJ in DNA repair. This is consistent with a previous study showing that the HR repair factor is excluded from the heterochromatin domain to prevent abnormal recombination of repetitive sequences unless the DSB site moves outside the domain [37]. Although conventional NHEJ repair can cause small deletion to threaten the integrity of the coding gene, this problem might not be severe enough to affect repetitive satellite DNA sequences that span over the mega-base range. PARP1 is critical for many DNA repair pathways including c-NHEJ and alt-EJ pathways, while DNA-PK is required for c-NHEJ but not alt-EJ repair [38][39][40]. This study showed that inhibition of DNA-PK had a rather moderate effect on the overall 53BP1 foci, whereas PARP1 inhibition markedly increased 53BP1 foci associated with centromere. Furthermore, the ChIP analysis showed 53BP1 binding at centromeric sites increased by PARP1 inhibition. This is consistent with a recent report showing that the cell viability of two DNMT3b knockout clones of HEK293 cells is not affected by DNA-PK inhibitor treatment, while the cell viability of one clone was affected by PARP1 inhibitor [41]. Altogether, we conclude that PARP1dependent repair of DNA lesions is constantly activated in type-1 ICF cells. Given the lack of error-free HHR in the centromere, we hypothesize that the repair of the centromeric DSBs generated by XPG/XPF-mediated removal of R-loops by the end-joining process might lead to centromere shortening, metaphase quadriradials with mixed chromosomal arms as common features observed in ICF due to DNMT3b dysfunction [9,10].

MATERIALS AND METHODS Cell culture, transfection, and infection
HCT116 and BKO cells were kindly provided by Bert Vogelstein (Johns Hopkins University School of Medicine, MD, USA) and were maintained in McCoy's 5A medium supplemented with 10% fetal bovine serum (FBS), 100 U/ml penicillin and 10 mg/ml streptomycin. BKO cells were generated by disrupting the alleles of DNMT3b by homologous recombination in HCT116 cells [42]. Type-1 ICF cells, pGM08714 (lymphoblastoid cell lines, LCLs) was obtained from Coriell Cell Repositories. In pGM80714 (ICF) cells, one allele of DNMT3b has a nucleoside mutation at G1807 (1807G > A), which gives a substitution at codon 603 (A603T), and another allele has a 9-bp insertion, which results in three amino acids insertion at codon 744 (744ins3). Both mutations occur within the catalytic domain of DNMT3b. Normal (WT) LCLs were kindly provided by Prof. Ching-Hwa Tsai (Graduate Institute of Microbiology, National Taiwan University). LCLs were maintained in RPMI supplemented with 15% FBS, 2 mM glutamine, 100 U/ml penicillin, and 10 mg/ml streptomycin. HEK293T and Platinum-A retroviral packaging (Cell Biolabs, CA, USA) cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% FBS, 100 U/ml penicillin and 10 mg/ml streptomycin. HEK293T cells were cotransfected with pCMVdeltaR8.91, pCMV VSVG, and targeted plasmids including PL-SIN-5TO-Flag-p53-IRES-GFP, ERCC4 (XPF) shRNA (TRCN0000078587) and ERCC5 (XPG) shRNA (TRCN0000358878) to produce lentivirus. For LCL cells, individual shRNA viral infection was performed by incubation of cells with viral supernatant followed by centrifugation at 800 x g for 1 h and recovery for 2 days. For double knockdown by shXPG and shXPF viral infection, cells were infected with shXPG virus and selected with puromycin (1 μg/ml), followed by subsequent shXPF viral infection. After recovery for 2 days, cells were incubated in the growth medium containing puromycin (1 μg/ml). For retrovectors infection, pMXs-3XHA-mouseRNaseH1, pMXs-3XHA-mouseR-NaseH1-D209N, and pQFlag-hDNMT3b were transfected into Platinum-A cells to produce retrovirus. After transfection for 48 h, the supernatants containing the virus were collected for infection.

Immunofluorescence staining
Cells were fixed with 4% paraformaldehyde for 15 min at room temperature (RT), followed by blocking in 5% BSA/TBS for 1 h at RT and staining with primary antibody overnight at 4°C. After incubation with a secondary antibody for 1 h at RT, slides were mounted in Fluoro-gel mounting oil (EMS, #17985-10) and were analyzed by an OLYMPUS BX53 and LSM 700 laser scanning confocal microscope (Carl Zeiss).

Comet assay
Comet assay was performed by Electrophoresis Assay (Trevigen, Inc) according to the manufacturer's protocol. DNA was stained with ethidium bromide and analyzed with Image J (v 1.47) for measuring tail length (TL).

Fig. 5
The regulation of (peri-)centromeric R-loops by XPG/XPF and transcription in BKO cells. a, b Levels of (peri-)centromeric R-loops by DRIP analysis. a HCT116 and BKO cells and b cells with or without XPG/XPF knockdown. All DNA samples were untreated (-) or treated (+) with RNase H before immunoprecipitation by S.9.6 antibody. qPCR values of DNA-RNA hybrids at (peri-)centromere sequences of chromosomes are normalized by IgG control. Data are expressed relative to that in HCT116 cells and presented as mean ± SEM of three independent experiments (*, **P < 0.05, 0.01, ns: no significant by two-tailed unpaired Student's t-test, RNH: RNase H). c HCT116 and BKO cells were used for Pol II-ChIP-qPCR analysis. Data are shown as the percentage of input DNA in Pol II antibody versus IgG control at (peri-)centromere sequences of chromosomes regions (mean ± SEM, n = 3, *, **, ***P < 0.05, 0.01 0.001, ns: no significant by two-tailed unpaired Student's t-test).
The comet tail (TM = %DNA in tail × TL/100) according to the manufacturer's suggestion.

Chromatin immunoprecipitation (ChIP)-sequencing
Chromatin immunoprecipitation using the anti-gamma H2A.X (Abcam, ab2893) was carried out according to a previously described protocol [43]. Libraries were constructed and bar-coded using TruSeq RS Cluster kit-HS (Illumina) and single-end sequencing (50 bp) was performed using an Illumina HiSeq-2500 sequencer at the Genomic Research Center of National Yang-Ming University (Taiwan) according to the manufacturer's instruction. All reads were mapped to the human genome (hg38) using the bowtie2 alignment software [44]. The alignment results were used to call peaks by MACS [45]. The results of the peak signal were subject to the Integrative Genomics Viewer (IGV, Broad Institute) [46] for further comparison and analysis. We used UCSC hg38 and Rfam v11 to identify and annotate the ChIP-seq peak regions. Fig. 6 The functional mutation of DNMT3b in ICF increases XPG/XPF accessibility to (peri-)centromeric R-loops. a, b Levels of (peri-) centromeric R-loops by DRIP analysis. a Wild-type and ICF LCL cells. b Cells with or without XPG/XPF knockdown. All DNA samples were untreated (-) or treated (+) with RNase H followed by S.9.6 antibody immunoprecipitation for qPCR analysis at (peri-)centromere sequences of chromosomes. qPCR values of DNA-RNA hybrids are normalized by IgG control. Data are expressed relative to that in wild type cells and presented as mean ± SEM of three independent experiments (*, **P < 0.05, 0.01, ns: no significant by two-tailed unpaired Student's t-test, RNH: RNase H). c Pol II-ChIP-qPCR analysis in wild-type and ICF LCLs. Data are shown as % of input DNA in Pol II antibody versus IgG control at (peri-) centromere sequences of chromosomes regions (mean ± SEM, n = 3, *P < 0.05, ns: no significant by two-tailed unpaired Student's t-test).
DNA-RNA immunoprecipitation (DRIP) assay DRIP was performed as described in [47]. Briefly, DNA was extracted carefully by Wizard® Genomic DNA Purification Kit (Promega, A1125), washed with 70% EtOH, and resuspended in TE buffer. The purified nucleic acids were sonicated in a buffer containing 10 mM Tris-HCl pH 8.5 and 300 mM NaCl to yield an average DNA fragment size of ∼300 bp. Sonicated DNA (6 μg) was added in the IP buffer (50 mM HEPES/KOH at pH 7.5; 0.14 M NaCl; 5 mM EDTA; 1% Triton X-100; 0.1% Na-Deoxycholate), followed by addition of S9.6 antibody (10 μg) for incubation overnight at 4°C. For removal of R-loops as a negative control, 6 μg of sonicated DNA was pretreated with 12 μl of RNase H (5000 U/mL; NEB, M0297s) for incubation at 37°C for 6 h prior to immunoprecipitation by S9.6 antibody. Protein G Sepharose beads that were pre-blocked with PBS containing 0.5% BSA, were added for 4 h to pull down immunocomplex. Bound beads were recovered and washed with 1 ml of low-salt buffer (50 mM HEPES/KOH, pH 7.5, 0.14 M NaCl, 5 mM EDTA pH 8, 1% Triton X-100, 0.1% Na-

Chromosome segregation analysis
Cells were treated with 0.2 μg/ml nocodazole for 16 h. After replacing with fresh medium for 1 h, cells were fixed with 4% paraformaldehyde for 15 min at RT, followed by blocking in 5% BSA/TBS for 1 h at RT and staining with mouse antibody to β-tubulin (1:200) and rabbit antibody to phosphoSer10-H3 (1:500) overnight at 4°C. Cells were incubated with secondary antibodies in the presence of Hoechst 33342 for 1 h at RT. Cells were mounted in Fluoro-gel mounting oil (EMS, #17985-10) for observation by OLYMPUS BX53 and confocal microscope (ZEISS, LSM700).

Confocal microscope and image quantification
Co-localization of two proteins of IF staining was acquired by LSM 700 laser scanning confocal microscope (Carl Zeiss). The conditions of IF staining, laser power, and pinhole sizes by confocal microscopy were identical among groups. Pixel number, pixel intensity, and area were provided by built-in software in LSM 700. Manders' coefficient was calculated for the co-localization of two proteins in the nuclear area. The co-localization coefficient values are ranged from 0 (segregation) to 1 (complete co-localization).

Cell viability assay
Cells were plated at 1,000 cells per well in a 96-well plate. Cells were treated with NU7441 (5 μM) or AZD2281 (5 μM) for 72 h, after which 10 μl of the Cell Counting Kit-8 (CCK-8) solution was added into each well for an additional 2 h at 37°C. The absorbance at 450 nm was measured using a Tecan Spark multimode microplate reader and the percentage of cell viability was calculated.

Statistics
Statistical analysis was performed using an unpaired two-tailed Student's ttest and Mann-Whitney test. Differences were considered statistically significant when P-values < 0.05, 0.01, and 0.001 indicated by *, **, and ***, respectively. Error bars represent the SEM of at least three independent experiments.

DATA AVAILABILITY
Sequencing data have been deposited at the GEO under the accession number GSE142376 and linked to UCSC genome browser session of DNMT3B study. https:// genome.ucsc.edu/s/chenwy/rH2AX_ChIP%2Dseq_HCT116