Resolution of R-loops by INO80 promotes DNA replication and maintains cancer cell proliferation and viability

Collisions between the DNA replication machinery and co-transcriptional R-loops can impede DNA synthesis and are a major source of genomic instability in cancer cells. How cancer cells deal with R-loops to proliferate is poorly understood. Here we show that the ATP-dependent chromatin remodelling INO80 complex promotes resolution of R-loops to prevent replication-associated DNA damage in cancer cells. Depletion of INO80 in prostate cancer PC3 cells leads to increased R-loops. Overexpression of the RNA:DNA endonuclease RNAse H1 rescues the DNA synthesis defects and suppresses DNA damage caused by INO80 depletion. R-loops co-localize with and promote recruitment of INO80 to chromatin. Artificial tethering of INO80 to a LacO locus enabled turnover of R-loops in cis. Finally, counteracting R-loops by INO80 promotes proliferation and averts DNA damage-induced death in cancer cells. Our work suggests that INO80-dependent resolution of R-loops promotes DNA replication in the presence of transcription, thus enabling unlimited proliferation in cancers.

I n proliferating cells, conflicts between DNA replication and transcription are one of the greatest threats to genome stability. Failure to resolve transcription-replication interference can lead to replication stress, which is characterized by stalling of the replication fork and induction of DNA breaks, with detrimental effects to cell proliferation and homeostasis. Mounting evidence indicates that co-transcriptional RNA:DNA hybrid structures known as R-loops 1 are a major obstacle to replication fork progression. While R-loops play a regulatory role in transcription 1 , encounters of forks with R-loops are a potent source of replication stress [2][3][4] and are particularly genotoxic when they occur in a head-on orientation 5 . Recent studies indicate that R-loops are highly abundant and induce replication stress in cancer cells 6,7 . This raises the question how cancer cells sustain sufficient DNA synthesis rates in the presence of increased transcription-replication conflicts. Several factors have been shown to prevent the formation or promote resolution of R-loops [reviewed in refs. 2,3,8 ]. However, the molecular pathways protecting forks from collisions with R-loops in cancer cells are poorly understood.
Emerging evidence indicates a role for chromatin structure in R-loop control. Mutations in the core nucleosomal histones H3 and H4 lead to accumulation of R-loops 9 . The histone chaperone complex FACT, which promotes nucleosomal integrity 10 and facilitates transcription in the presence of chromatin 11 , prevents R-loop accumulation and promotes resolution of transcriptionreplication conflicts 12 . Nevertheless, our understanding of the role of chromatin regulation in R-loop modulation remains critically limited. Importantly, whether the role of chromatin is that to solely suppress R-loop formation, or to also promote R-loop resolution, remains unknown.
The chromatin landscape is shaped by the action of ATPdependent chromatin remodelling enzymes that alter the structure, composition or position of nucleosomes 13 . The INO80 complex, which contains the INO80 ATPase, is a structurally and functionally evolutionary conserved chromatin remodelling complex 14 . In yeast, INO80 has been shown to facilitate degradation of RNA Polymerase II during replication stress conditions in order to preserve genome stability 15,16 . Mammalian INO80 is required for replication fork stability and recovery following replication stress 17 , whilst it also promotes DNA replication in unperturbed conditions 18 . Nevertheless, the role of INO80 in DNA replication remains unclear.
Several INO80 subunits are overexpressed in different cancers such as breast 19 , neuroendocrine prostate cancer 20 and melanoma 21 , and their expression levels can correlate positively with bad prognosis 21,22 . Depletion of INO80 in cancer cells decreases oncogenic transcription, compromises cell proliferation and subsequent tumour growth 21 . These observations suggest a critical function for INO80 in sustaining cancer development and progression. However, the role of INO80 in cancer cell proliferation remains largely elusive.
Here we investigate the mechanisms that protect cells from replication-associated DNA damage. By using oncogene-driven prostate cancer PC3 cells, we find that INO80 prevents replication stress-induced DNA damage and promotes proper and efficient DNA synthesis by counteracting accumulation of Rloops. INO80 is recruited to R-loop-enriched sites across the genome independently of gene expression levels. Artificial tethering of INO80 at a genomic site enriched in R-loops results in turnover of R-loops in cis. Notably, removal of R-loops by overexpression of the RNA:DNA endonuclease RNAse H1 rescues the growth defects caused by INO80 depletion in PC3 cells, NRAS-dependent melanoma WM1361 cells and estrogendependent breast cancer MCF7 cells, while inhibition of the BER pathway sensitizes INO80-depleted cancer cells to lethality.
Our results suggest that R-loop resolution driven by INO80 prevents genotoxic collisions between transcription and replication, enabling unlimited proliferation of cancer cells.

INO80 promotes DNA replication by counteracting R-loops.
We sought to understand the underlying cause for defective DNA replication in human cells lacking INO80. We questioned whether the role of INO80 in promoting DNA replication is dependent on transcription. siControl and siINO80 PC3 cells were treated with the transcriptional inhibitors α-amanitin or cordycepin and analysed for DNA synthesis rates by CldU/IdU DNA fibre pulse labelling assay (Fig. 1a). INO80 depletion led to significantly decreased DNA synthesis (Fig. 1b, c), as expected 17,18 . Total labelling of DNA fibres indicated that DNA fragmentation was not the cause of reduced fibre length ( Supplementary Fig. 1). Ectopic expression of siRNA immune INO80 cDNA rescued the replication defect in siINO80 cells, indicating that the replication defect is specifically due to loss of INO80 ( Supplementary  Fig. 2a-d). Treatment with transcription inhibitors rescued partially but significantly the DNA synthesis defect of siINO80 cells (Fig. 1b, c and Supplementary Fig. 1a), suggesting that transcription impedes DNA replication in the absence of INO80.
We asked whether R-loops are the cause of the DNA replication defect in siINO80 cells. siControl and siINO80 cells were co-transfected with either an empty vector (EV) or a plasmid overexpressing the endonuclease RNAse H1 (RNAseH1 o/e), which specifically targets and removes RNA:DNA hybrids from the genome, and DNA synthesis rates were evaluated using DNA fibre labelling as before (Fig. 1d-f and Supplementary  Fig. 2g, h). RNAse H1 overexpression reduced DNA synthesis in control cells and slightly increased the number of cells in S-phase as indicated by FACS cell cycle profile analysis ( The ACTR8/Arp8 subunit of INO80 is required for the complex's chromatin remodelling activity 23 . Depletion of ACTR8 led to a significant decrease in DNA synthesis rates (Supplementary Fig. 3e-h), while overexpression of RNAse H1 in siACTR8 cells rescued the replication defect ( Supplementary Fig. 3g, h). We tested whether chromatin relaxation in cells depleted of INO80 would rescue DNA replication similar to the rescue observed upon RNase H1 overexpression. The histone deacetylase (HDAC) inhibitor SAHA/Vorinostat induces hyperacetylation of the Nterminal tails of histones H3 and H4, creating a more open nucleosomal structure. Addition of Vorinostat in control cells reduced DNA synthesis rates 24 , while it significantly rescued the DNA synthesis rates of siINO80 cells (Fig. 1i,). When Vorinostat was added in siINO80 cells overexpressing RNAse H1, it did not further increase DNA synthesis rates compared to untreated siINO80 cells overexpressing RNAse H1 (Fig. 1i). This supports an epistatic functional relationship between RNase H1 and Vorinostat in rescuing DNA synthesis in the absence of INO80. Together these data suggest that chromatin regulation by INO80 counteracts R-loops to promote replication fork progression.
RNAse H1 overexpression rescues stalled forks in siINO80. To distinguish between slower fork movement, or increased fork stalling by R-loops in siINO80 cells, we analysed the progression of sister replication forks (Fig. 2a-d). In control cells, the majority of sister forks progressed at a similar rate from a given origin and generated symmetrical patterns of IdU/CldU incorporation (Fig. 2b, c). However, in siINO80 cells, 89% of the forks were asymmetrical with a greater than 2-fold difference in DNA synthesis rates between the two sister forks compared to control cells ( Fig. 2b-d). This indicates that forks stall more frequently in cells lacking INO80. ssDNA fibre analysis showed intact DNA fibres at non-symmetrical sister replication forks in siINO80 cells ( Supplementary Fig. 4), ruling out the possibility that the replication fork asymmetry in siINO80 cells is caused by DNA damage. When RNAse H1 was overexpressed in siINO80 cells, fork symmetry was recovered (Fig. 2), strongly suggesting that Rloops cause increased replication fork stalling in the absence of INO80.  25 . Chk1-Ser345, a downstream target of the checkpoint kinase ATR 26 , was phosphorylated in siINO80 cells (Fig. 3a), suggesting increased DNA damage in unperturbed conditions. To test whether DNA damage during S-phase in the absence of INO80 is associated with Rloops, siControl and siINO80 cells were co-transfected with either the empty vector (EV) or the RNAse H1 o/e plasmid and grouped in replicating and non-replicating populations based on positive and negative EdU staining respectively (Fig. 3b). Quantitative immunostaining analysis was conducted against γH2A.X, a marker of DNA damage (Fig. 3c-e). Overexpression of RNAse H1 in control cells partially activated the S-phase checkpoint as expected 27 , without affecting γH2A.X levels (Fig. 3a, c-e). Depletion of INO80 significantly increased γH2A.X signal in S-phase cells but not in non-S-phase cells (Fig. 3c-e). RNase H1 overexpression in siINO80 cells decreased the levels of Phospho-Chk1-S345 to levels comparable to control cells overexpressing RNAse H1 and significantly reduced the intensity of γH2A.X (Fig. 3a, c-e). This results indicate that R-loops induce DNA damage and activate the S-phase checkpoint in the absence of INO80.
We further tested whether DNA damage induced by replication stress in INO80-depleted cells 17 is dependent on R-loops. Control and siINO80 cells overexpressing RNAse H1 were treated with hydroxyurea (HU), a drug that depletes the dNTP pools, leading to replication stress and DNA damage 28,29 . Loss of INO80 led to an increase in γH2A.X intensity in HU-treated Sphase cells, but not in non-S-phase cells ( Supplementary  Fig. 5b-d). RNase H1 overexpression significantly reduced both the intensity of γH2A.X signal and the percentage of γH2A.Xpositive cells in control and siINO80 cells ( Supplementary  Fig. 5b-e). Likewise, inhibition of transcription by α-amanitin resulted in a significant decrease in the intensity of γH2A.X inside S-phase in INO80-depleted cells either with or without HU ( Supplementary Fig. 6). These results suggest that INO80 counteracts co-transcriptional R-loops to suppress replicationassociated DNA damage in human cells.
R-loops accumulate genome-wide in the absence of INO80. To evaluate the levels of R-loops in the absence of INO80, immunofluorescence analysis was conducted in control and siINO80treated PC3 cells using the S9.6 antibody (Fig. 4a, b). Consistent with other reports 30 , S9.6 puncta were detected in the cytoplasm and the nucleus. The nuclear S9.6 signal was diminished upon RNase H1 overexpression ( Supplementary Fig. 7a). Quantification of the nuclear S9.6 signal intensity revealed a significant increase in R-loops upon INO80 depletion (Fig. 4c, d). The intensity of S9.6 signal was also increased in cells depleted for INO80 by viral shRNA compared to a non-targeting shScrambled Control ( Supplementary Fig. 7b-d). In vitro treatment with recombinant RNAse H eliminated the increased S9.6 signal in siINO80 cells ( Supplementary Fig. 7d). Pulse labelling of cells with the uridine analogue 5-ethynyluridine (EU), which is incorporated into newly synthesized RNA 31 , revealed no significant changes in transcriptional activity between control and siINO80 cells ( Supplementary Fig. 7e), indicating that the increase in R-loop abundance in the absence of INO80 is not due to elevated transcription rates.
We asked whether depletion of INO80 leads to an increase in R-loops inside S-phase. Analysis of R-loop intensity in EdU positive and negative cells showed enhanced accumulation of Rloops in siINO80 cells both outside and inside S phase (Fig. 4e). Therefore, R-loops that accumulate throughout the cell cycle in the absence of INO80 can be a potential source of genome instability during DNA replication.
To test whether INO80 prevents accumulation of R-loops formed at specific genomic loci, DNA:RNA immunoprecipitation (DRIP) assay was conducted in control and INO80-depleted cells by four independent lentiviral shRNAs ( Fig. 4f, g). The promoterproximal and termination regions of the beta-actin gene (in1 and pause regions) and the EGR1 gene are sites prone for R-loops formation 32,33 . Loss of INO80 induced a reproducible increase in R-loop enrichment at the beta-actin in1 and pause regions, as well as in the EGR1 gene (Fig. 4g). In contrast, no increase in R-loops was observed at the 5′ region upstream the beta-actin gene promoter in INO80-depleted cells when compared to control cells (Fig. 4g). The increase in DRIP-qPCR signal observed at the betaactin and EGR1 genes upon INO80 depletion was diminished upon treatment of the genomic DNA with recombinant RNAse H prior to DRIP ( Supplementary Fig. 7f). These results suggest that INO80 counteracts accumulation of R-loops forming at R-loop prone sites.
Nuclear colocalization of INO80 with R-loops. We asked whether INO80 associates with nuclear R-loops. Immunofluorescence (IF) samples stained with the S9.6 antibody (Rloops) and anti-INO80 were imaged using STED nanoscopy and analysed for colocalization between INO80 and R-loops (Fig. 5a). The increased resolution of STED at~50 nm in our conditions, compared to confocal imaging (~250 nm), allows discrimination between 'true' and 'false' colocalization events with high level of certainty. Colocalization between INO80 foci and R-loop foci by STED was readily observed (Fig. 5a), while multiple colocalization events between INO80 and R-loops visualised by confocal were Fig. 1 R-loops slow replication rate in INO80-depleted cells. a Schematic representation of the experimental approach. PC3 cells were transfected with esiRNAs against either GFP or INO80. Three days later, cells were treated with α-amanitin (α-a) or cordycepin (crd) for 3 h or left untreated as control and then subjected to fibre labelling analysis. b Representative images of spread fibres from each condition. Similar results were obtained in five independent experiments (cordycepin treatment-2). c Distribution of fork speed rates in INO80-proficient (siGFP) and INO80-deficient (siINO80). Data is from five independent experiments (in cordycepin-treated cells-2), at least 250 fibres were measured per condition in each experiment. ****p-value < 0.0001, (two-tailed unpaired Student's t-test). d Schematic representation of the experimental setup used. PC3 cells were co-transfected with esiRNAs against either GFP (siGFP) or INO80 (siINO80) along with either a control (CTRL) or RNase H1-overexpressing (RNAseH1) vector. Two days later RNAse H1 expression was induced by doxycycline for 24 h. Cells were labelled with CldU for 5 min followed by IdU pulse for 20 min and subjected to DNA fibre labelling analysis. e Representative images of spread fibres from each condition. Similar results were obtained in four independent experiments. f Distribution of fork speed rates (kilobase/min) in siGFP and INO80-deficient cells transfected with control or RNAse H1 overexpression plasmids. Data are from 4 independent experiments, at least 250 fibres were measured per condition in each experiment. ****p-value < 0.0001, (two-tailed unpaired Student's t-test). g Schematic representation of the experimental setup. Cells were co-transfected and induced as in d and prior to labelling were treated with 5 µM vorinostat for 6 h. h Representative images of spread fibres from each condition. Similar results were obtained in three independent experiments. i Distribution of fork rates from (h, at least 250 fibres were measured per condition in each experiment; ns non-significant, ****p-value < 0.0001, *p-value < 0.05, (two-tailed unpaired Student's t-test). c, f, i Kruskal-Wallis test p-value was < 0.0001. Data is presented as Tukey boxplot (box representing first quartile, median and third quartile, whiskers 1.5 times interquartile range). ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-18306-x found to be separate, distinct foci when resolved by STED (Supplementary Fig. 8). To distinguish between random and nonrandom co-localization events, we conducted a Van Steensel's cross-correlation function analysis (CCF) 34 . Co-localization events between the STED imaged channel (INO80 or S9.6) and confocal imaged EdU were random, as expected (Supplementary Table 1).
Contrary, the global colocalization between the STED INO80 and STED S9.6 R-loop signals was not random (Supplementary Table 1). This suggests true R-loop:INO80 colocalization events.
Only a subset of INO80 foci co-localized with R-loops. To quantify the percentage of R-loop objects with overlapping volume with INO80 objects, we created 3D volume objects for   Fig. 2 Fork asymmetry in Ino80-depleted cells depends on R-loops. a Schematic representation of the experimental setup used. PC3 cells were cotransfected with esiRNAs against either GFP (siGFP) or INO80 (siINO80) along with either a control (CTRL) or RNAse H1-overexpressing (RNAse H1) vector. Two days later RNAse H1 expression was induced by doxycycline for 24 h. Cells were labelled with CldU for 20 min (red) followed by IdU pulse for 20 min and subjected to DNA fibre labelling analysis. b Representative pairs of sister replication forks were assembled from different fields of view and were arbitrarily centred on the position of origin. Scale bar 5 μm. Similar distribution of paired forks was observed in three independent experiments. c Scatter plots of the distances covered by right-moving and left-moving sister forks during the CldU pulse in Ino80-proficient or deficient cells expressing or not RNAse H1. The central areas delimited with grey lines contain sister forks with less than a 25% length difference. The percentage of symmetric forks is indicated. d Relative fork asymmetry. Fork asymmetry is expressed as the ratio of the longer arm to the shorter one for each pair of sister replication forks, ****p-value < 0.0001; n.s non-significant (two-tailed unpaired Student's ttest). Numbers above boxes indicate the median of the ratio of the longer to shorter arm. Data is presented as Tukey boxplot.  R-loops and INO80 in the nucleus (Fig. 5b). The number of nuclear R-loops counted per cell in our analysis varied from 164-682 (Fig. 5c, upper panel). The percentage of R-loops which had some co-localizing volume with INO80 ranged between 25 and 52% (Fig. 5c, bottom panel), suggesting that a significant proportion of R-loops are bound by INO80.
To determine whether INO80 colocalizes with specific R-loops, we analysed the fluorescence intensity, volume and length properties of the 3D objects. All properties were significantly different between INO80 and R-loop objects ( Supplementary  Fig. 9a). For all cells analysed, the R-loops colocalized with INO80 were significantly more intense, had greater volume and greater length than their non-colocalizing counterparts (Fig. 5d-f and Supplementary Fig. 9b-f). These data suggest that the INO80 complex associates with the largest, most enriched, R-loop domains in the nucleus.
Chromatin association of INO80 is promoted by R-loops. To gain insight into the co-enrichment of INO80 and R-loops across the genome, we analysed the genome-wide association of INO80 and R-loops using published ChIP-seq data for INO80 and DRIPseq data for R-loops in mouse embryonic stem cells (mESCs) 35,36 . Comparative analysis for INO80 and R-loop enrichment at protein-coding gene bodies revealed a significant positive correlation (p = 1.1 × 10 −206 , Fig. 6a), suggesting that INO80 is recruited to transcribed genes with high propensity to generate Rloops. Partial correlation analysis performed on the INO80 ChIPseq and DRIP-seq data over RNA-seq data, indicated that the correlation between INO80 and R-loop enrichments remained highly significant when controlled for gene expression levels (p = 5.5 × 10 −160 , Supplementary Fig. 10a-c). Therefore, the positive correlation detected between INO80 and R-loop enrichment is not indirectly due to their mutual association with transcription. Visualisation of overlayed INO80 ChIP-seq and DRIP-seq reads confirmed that INO80 and R-loops are co-enriched at the betaactin gene and other mRNA genes ( Fig. 6b and Supplementary  Fig. 10d).
To illuminate the genome-wide association of INO80 with Rloops, we analysed INO80 and R-loop enrichment across mESC chromatin states at 200 bp resolution. We segmented the genome into 20 chromatin states based on the combinations of 14 histone marks, 3 cytosine modifications and INO80 peaks 37 (Fig. 6c). The maximum enrichment of INO80 was found to be in states 10 and 15 (Fig. 6c). State 10 is characterised by high levels of histone marks such as H3K4me3 and H3K9ac which are associated with active promoters and characterise the transcription start site (TSS) of genes. State 15 is solely defined by INO80 enrichment (Fig. 6c). When the enrichment of R-loops with or without INO80 was analysed, R-loops were primarily enriched without INO80 in most of the chromatin states ("DRIPseq+INO80" and "DRIPseq" peaks respectively, Fig. 6d). However, the greatest association of R-loops with INO80 as well as the highest enrichment in R-loops were observed in state 15 (Fig. 6d). These results indicate that INO80 and R-loops strongly associate even outside annotated TSS and gene bodies and suggest that INO80 is recruited at genomic regions enriched in R-loops.
The Ruvbl1 and Ruvbl2 subunits of INO80 bind to RNA:DNA hybrid structures in vitro 38 . We, therefore, tested whether R-loops promote the binding of INO80 to chromatin. PC3 cells stably expressing either a control plasmid or the doxycycline-inducible RNAse H1 plasmid were subjected to differential salt fractionation after addition of doxycycline (Fig. 6e). Analysis of the different subcellular fractions showed that the amount of INO80 found in the high salt fraction, which represents soluble nuclear proteins and loosely-associated chromatin proteins, was increased upon overexpression of RNAse H1 (Fig. 6f, g). Concurrently, the fraction of INO80 tightly bound to chromatin was significantly reduced by approximately two-fold (Fig. 6f, g). This suggests that R-loops promote stabilization of INO80 on chromatin.
Artificial tethering of INO80 promotes R-loop resolution. We hypothesized that INO80 may promote resolution of R-loops. Employing U2OS cells carrying the 256x-LacO tandem array 39 , we devised an assay to monitor changes in R-loop enrichment upon artificial tethering of the INO80 complex. The LacO-LacI array has been reported to be a site of replication stress 40 . LacIfused proteins bind LacO, while incorporating eGFP in the system enables visualisation of the LacO locus (Fig. 7a). S9.6 IF in LacO-U2OS cells expressing LacI-GFP demonstrated that the LacO array is enriched in R-loops (Fig. 7b). We next expressed LacI-eGFP tagged versions of RNAse H1 or the INO80E subunit of the INO80 complex in LacO-U2OS cells and the S9.6 signal overlapping with the eGFP-LacI signal was quantified. Tethering of LacI-eGFP-RNAse H1 to the LacO array led to a significant decrease in the intensity of the underlying R-loop signal (Fig. 7b,  c). This indicates that R-loops at the LacO site can be suppressed by artificial recruitment of factors promoting their resolution. Tethering of LacI-eGFP-INO80E reduced the R-loop signal to levels similar to LacI-eGFP-RNAse H1 (Fig. 7b, c). This suggests that INO80 is directly involved in the downregulation of R-loops. Depletion of INO80 in LacO-U2OS cells expressing LacI-eGFP-INO80E resulted in increased enrichment of R-loops at the LacO site, suggesting that intact INO80 complex is required for the suppression of R-loops ( Supplementary Fig. 11).
To understand how INO80 regulates R-loops, we monitored the dynamics of R-loops at the lacO locus in live cells. LacO-U2OS cells were transfected with a plasmid expressing the RNA Binding Domain of RNAse H1 fused to DsRed (RBD-DsRed) (Fig. 7d). The RBD construct allows monitoring R-loop enrichment 41 in live cells. In agreement with our S9.6 IF results, the RBD-DsRed signal also accumulated at the LacO locus (Fig. 7e). We co-transfected cells with RBD-DsRed and either LacI-eGFP or LacI-eGFP-INO80E and performed time-lapse imaging every 6 min for 25 h, 24 h after transfection ( Fig. 7f; Supplementary movies 1 and 2). The RBD-DsRed intensities relative to the colocalizing eGFP signal were quantified throughout the time-course of the experiment in single cells. Changes in R-loop signal intensity were analysed by calculating the Fold-Change in the relative DsRed fluorescence Intensity (FC-I) between every time point and its previous one (FC-I = I t2 /I t1 ) , and plotted in log2 scale as FC-I(log2) (Fig. 7g). The fold-change in R-loop intensity indicates the number of R-loops created minus the R-loops resolved during the specific time period. A positive FC-I(log2) value (I t2 > I t1 ) suggests net formation of new R-loops across the LacO site between the two time points. A negative FC-I (log2) value (I t2 < I t1 ) indicates that resolution of Rloops is greater than formation for the specific time period. In both LacI-eGFP and LacI-INO80 the mean FC-I(log2) values were positive, indicating that R-loops are constantly formed at the LacO site. However, the mean FC-I(log2) value in cells transfected with LacI-INO80E was significantly smaller than the mean FC-I(log2) value for LacI-eGFP (Fig. 7g, Total), indicating INO80 actively counteracts R-loop formation. Tethered INO80 could affect R-loop dynamics through either suppressing their formation, or promoting their resolution. To distinguish between these possibilities, positive and negative FC-I(log2) values were clustered separately in LacI-eGFP and LacI-INO80E cells. The positive FC-I(log2) values in the LacI-INO80E cells were significantly smaller than in the LacI-eGFP (Fig. 7g). If INO80 affects R-loop formation but not R-loop resolution, then the same 'amount' of resolution should be applied to both the LacI-eGFP and LacI-INO80E datasets. In that case, because LacI-INO80E cells have a lower mean positive FC-I(log2) value than LacI-eGFP, the negative FC-I(log2) values upon tethering INO80 should be greater. However, when the negative FC-I(log2) values were analysed, no significant difference was found between LacI-INO80E and LacI-eGFP (Fig. 7g). This contests the hypothesis that resolution of R-loops is independent of INO80, and suggests that tethering INO80 does not impact R-loop synthesis. Thus, the dynamics of the R-loop signal at the LacO locus observed upon artificial recruitment of INO80 suggest a role for INO80 in promoting turnover of R-loops.

R-loop resolution by INO80 promotes cancer cell proliferation.
We asked whether the role of INO80 in cancer cell growth 14 is associated with removal of R-loops. Proliferation was monitored in cells that were either co-transfected with the RNAse H1 overexpression plasmid or not and depleted for INO80. siRNase H1 cells were also analysed to assess the effect of R-loop accumulation in cancer cell proliferation ( Supplementary  Fig. 12a). In addition, we calculated the expected change in growth predicted in the case that the two factors are not functionally associated.
Silencing of either INO80 or RNase H1 did not affect the proliferation of human embryonic kidney HEK293 cells, however  both depletions compromised the growth of PC3 cells (Fig. 8a, b and Supplementary Fig. 12b, c). RNAse H1 overexpression rescued the growth of siINO80 PC3 cells by almost four-fold compared to the expected growth value (Fig. 8b), indicating that the proliferation defect caused by INO80 depletion is rescued by removal of R-loops. INO80 has been reported to promote growth of NRAS oncogene mutant-driven melanoma cells 21 . Depletion of INO80 in the NRAS mutant WM1361 melanoma cell line compromised cellular growth, while RNAse H1 overexpression in the siINO80 WM1361 cells rescued growth by approximately three-fold (Fig. 8c). R-loop-induced replication stress is a reported hallmark of E2 estrogen-positive MCF7 cancer cells 7 . Disruption of INO80 led to a severe growth defect in MCF7 cells (Fig. 8d). Notably, overexpression of RNAse H1 in INO80-depleted MCF7 cells strongly rescued their proliferation defect (Fig. 8d). These results suggest that INO80-dependent resolution of R-loops is required for proliferation of cancers characterized by dysregulated transcription.
Although proliferation of PC3 cells is severely compromised and DNA damage accumulates upon depletion of INO80 or RNase H1 42 , loss of either factor led to a minor increase in cell lethality ( Fig. 8e and Supplementary Fig. 12d). We therefore hypothesized that R-loop-induced DNA damage is efficiently repaired in cancer cells, thus averting cell death. DNA damage associated with R-loops is repaired by the base excision repair (BER) pathway 43 , which requires the AP endonuclease APE1/ yAPN1 and the homologous recombination repair factor Rad52 44 . We therefore tested whether combined inhibition of APE1 and Rad52 sensitizes cancer cells lacking INO80 or RNase H1 to death. PC3 and HEK293 cells depleted of either INO80 or RNase H1 were treated with the APE1 and Rad52 inhibitor 6hydroxy-DL-Dopa 45,46 (DL-Dopa) and assessed for lethality. None of the treatment combinations induced increased cell death in HEK293 cells (Supplementary Fig. 12e). A 15-20-fold increase in lethality was observed in siINO80 and siRNAse H1 PC3 cells treated with DL-Dopa ( Fig. 8e and Supplementary Fig. 12f). In contrast, inhibition of the DNA damage checkpoint factor ATR which also safeguards against genotoxic R-loops 47 , did not induce further cell death in siINO80 PC3 cells (Supplementary Fig. 12g). This synthetic lethality phenotype suggests that proliferating cancer cells with unresolved R-loops rely on the BER pathway for their viability. Taken together, our results suggest that R-loop resolution facilitated by INO80 ameliorates DNA damage at sites of transcription-replication conflicts to promote cancer cell proliferation and prevent cell death.

Discussion
Here, we elucidate a role of the human INO80 complex in DNA replication. Our study reveals that by counteracting accumulation of R-loops, INO80 prevents genotoxic conflicts between transcription and the replication machinery and promotes efficient DNA synthesis. Our study supports the idea that INO80 defines a pathway for the removal of R-loop structures from chromatin that is critical for maintenance of genome integrity and cancer cell proliferation.
A recent study using an in vitro eukaryotic DNA replication system on a nucleosomal template has suggested that INO80 promotes replisome progression through chromatin in the absence of transcription 48 . While this possibility cannot be ruled out by our study, our in vivo evidence suggests that INO80 facilitates DNA replication by averting collisions between the fork and co-transcriptional R-loops: Firstly, RNAse H1 overexpression rescued DNA replication progression and suppressed fork stalling in INO80-depleted cells (Figs. 1 and 2). Secondly, the replicationassociated DNA damage of INO80-depleted cells was significantly relieved by overexpression of RNAse H1 or chemical inhibition of transcription (Fig. 3). These results support a model where, by counteracting R-loops, INO80 removes a critical barrier to DNA replication and suppresses replication-associated DNA damage, rather than facilitating the repair of DNA damage.
While impaired fork progression in siINO80 cells is rescued by overexpression of RNAse H1 in normal conditions (Fig.1), we do not anticipate that every fork encounters R-loops in the absence of INO80, as the increase in fork asymmetry also indicates (Fig. 2). Loss of INO80 leads to constitutive activation of the ATR/Chk1 DNA synthesis checkpoint pathway in normal conditions (Fig. 3), which reduces global DNA synthesis rates 49 and slows down elongation even at unchallenged forks 50 . It is thus likely that stalling of forks following collisions with R-loops in the absence of INO80 induces activation of the ATR/Chk1 pathway, triggering an overall slowdown of replication fork movement. In addition, our observation that R-loops contribute to the high levels of DNA damage observed in HU conditions, in both normal and INO80-depleted cells ( Supplementary Fig. 5), makes it plausible that INO80 is not indiscriminately targeting R-loops, but it is specifically required to suppress those R-loops that can potentially interfere with forks in a genotoxic, head-on orientation 5,51 .
Several data indicate a direct involvement of INO80 in downregulation of pre-existing R-loops. INO80 reduces the enrichment of R-loops formed at specific genomic sites, such as the beta-actin gene (Fig. 4). STED nanoscopy revealed that  (Fig. 6). Taken together, our single cell, biochemistry and genomic analyses suggest the presence of a regulatory mechanism for recruitment of INO80 by R-loops across the genome.
Unexpectedly, our genomics analysis revealed an uncharacterized chromatin state that is defined by INO80 and is highly enriched in R-loops but is not associated with neither the R-loop- enriched 5′ and 3′ ends of gene bodies 52 , nor with the gene body itself (Fig. 6). Characterisation of this chromatin state is an important step towards elucidating the cellular mechanisms controlling R-loop metabolism and promoting genomic stability. Our observation that artificial tethering of INO80 at the lacO array led to reduced enrichment of R-loops in cis suggests that INO80 binds to genomic regions enriched for R-loops in order to promote their removal (Fig. 7). Time-lapse analysis of R-loops at the lacO site suggested that onsite recruitment of INO80 did not suppress formation of R-loops but instead triggered their turnover. Although we cannot formally exclude that binding of the RBD-DsRed construct at the lacO array is compromised in the presence of LacI-INO80E, the results obtained from our kinetics analysis (Fig. 7g) argue against this possibility. If RBD-dsRd binding was adversely affected by LacI-INO80E, and the resolution kinetics remained the same upon binding of either LacI-GFP or LacI-INO80E, we would expect to see an increase in negative values of RBD intensity change in the LacI-INO80E compared to lacI-GFP. However, we observe similar negative values in the LacI-GFP and LacI-INO80E cells (Fig. 7g). Moreover, the welldocumented role for mammalian INO80 in transcriptional activation 21,35 , makes it unlikely that INO80 decreases the abundance of R-loops at the lacO site by repressing transcription. The INO80 complex has been reported to physically interact with RNA:DNA helicases such as DDX5 or DDX59 [53][54][55] . Given that INO80 promotes extraction of ubiquitinated RNA Polymerase II from chromatin 16 , it is plausible that INO80 coordinates resolution of R-loops with removal of stalled RNA Polymerase II.
Human INO80 has been linked to opening up chromatin structure 14 . Evidence suggests that the chromatin surrounding Rloops adopts a compacted nucleosomal structure 56,57 . We observed that chemically induced decompaction of chromatin by SAHA/Vorinostat rescued the DNA replication defect of INO80depleted cells in an epistatic manner with RNase H1 overexpression (Fig. 1). This supports the possibility that the chromatin remodelling activity of INO80 facilitates decompaction of the repressive chromatin landscape at R-loop enriched sites, revealing an intriguing aspect of R-loop regulation by chromatin.
Recent reports have shown that oncogenic and hormonedependent transcription, in HRAS overexpressing cells and in breast cancer MCF7 cells respectively, leads to enhanced formation of R-loops and increased R-loop-dependent DNA damage during DNA replication 6,7 . The fact that these cancer cells are able to sustain sufficient DNA synthesis rates for their proliferation under such highly genotoxic conditions suggests that cancer cells have established mechanisms to cope with the increased occurrence of genotoxic transcription-replication conflicts. Our findings that (i) INO80 counteracts genotoxic R-loops to promotes proliferation of prostate, breast and melanoma cancer cells (Fig. 8), and (ii) INO80 depletion is synthetically lethal with Rad52/APE1 inhibition suggest a chromatin-based Rloop resolution mechanism in cancer cells that suppresses their inherent predisposition for DNA damage during S-phase. Whether resolution of R-loops by INO80 regulates oncogenic transcription and enables coordination of dysregulated gene expression with DNA replication in cancer cells is an exciting possibility.
In conclusion, by identifying INO80 as a molecular link between cellular proliferation and silencing of R-loops, our study provides insight into how cancer cells balance transcription with replication, enabling unlimited growth in the presence of inherent replication stress conditions.

Methods
Cell culture, transfections and treatments. Human PC3 cells were grown in DMEM (Gibco) supplemented with L-glutamine, 10% fetal bovine serum, 1 mM sodium pyruvate and antibiotics in 5% CO 2 atmosphere at 37°C. Cells were treated with 0.5 mM hydroxyurea to induce replication stress. To inhibit transcription αamanitin (2 µg/ml) and cordycepin (50 µM) were used. Induction of the Tet-ON promoter was achieved with 1 µg/ml doxycycline for 24 h. 5-Ethynyl-2′-deoxyuridine (EdU)to label S-phase cells was used at 25 µM final concentration and 5ethynyl uridine (EU) was used at 1 mM to assess the overall nascent transcription.
EsiRNAs targeting the coding regions of human INO80 (3440-3894, transcript NM_017553.1) or EGFP (132-591) were synthesized following standard procedures 58,59 . Primers used to amplify the targeted regions were selected using Riddle database 60   The inducible expression of RNAse H1 has been achieved using pEBTet-BLAST-RNAse H1-myc/His and pEBTet-EGFP-BLAST was used as control.
The RBD-DsRed plasmid was constructed by PCR cloning the HB domain of RNAse H1 into the pDsRed-Express-C1 vector (Clontech) using the following primers: 41 RNH1_HBF (5′-ACTCA GATCTGGGATGTTCTATGCCGTGAGG-3′) RNH1_HBR (5′-ATTGAG TCGACGCTTGCTGATTTCCTGAC-3′) Immunofluorescence and image analyses. The S9.6 antibody was purified from the S9.6 mouse hybridoma cell line (ATCC ® HB-8730 ™ ), at the Protein Expression and Purification Core Facility in Institut Curie, France using the ATCC recommended growth conditions. For immunofluorescence, the protocol from ref. 57 was used with the following modifications. Cells were grown on coverslips for 24-72 h. Cells were fixed using ice-cold MeOH and stored overnight at −20 o C. Coverslips were then washed 1x in PBS 1′, followed by 10′ wash in 50 mM NH 4 Cl in PBS for 10′ at room temperature. Cells were washed 1x in PBS, then 1x in PBS-Triton X 100 0.1% for 5′ each. Cells were blocked for 30′ in 3% BSA in 0.1% PBS-TX. S9.6 antibody was used at 1:1000, anti-INO80 antibody (Abcam ab118787) was used at 1:1000 both diluted in 1% BSA in 0.1% PBS-TX. Secondary antibodies for confocal microscopy were from Jackson Immunoresearch Alexa Fluor® 647 AffiniPure Donkey Anti-mouse IgG (1:500-1500); Alexa Fluor® 594 AffiniPure Donkey Anti-Rabbit IgG (for INO80 confocal and STED) or for S9.6 STED, ATTO 647N (STED/GSD) Goat anti-mouse IgG (Active Motif Catalog No. 15038) at 1:100. DNA was labelled using Hoescht. Coverslips were mounted using Prolong Diamond (Thermo Fisher) for confocal imaging or Mowiol 4.88 (Calbiochem) mounting media was prepared and used for STED. Mowiol mounted slides were allowed to cure for 48 h prior to imaging. Using this protocol, the non-specific S9.6 signal was depleted from the nucleolus. Images were deconvolved prior to quantification. Deconvolution was performed with Huygens 18.04 from SVI (www.svi.nl). Quantifications of both total nuclear R-loops and EdU associated R-loops were carried out using 3D volumes with IMARIS image analysis software. Maximum intensity stack projections were used for presentation in Fig. 2.
Recombinant RNAse H (NEB #M0297) was used following standard protocol 66 with the modification that fixed cells on coverslips were digested with RNase H in RNAse H buffer for 2 h at 37°C, after which coverslips were washed 3 ×10′ in PBS prior to immunofluorescence staining. Control samples were mock-treated with RNase H buffer.
To stain for γH2AX, cells were grown on coverslips, washed in PBS, fixed with ice-cold methanol for 7 min at −20°C, permeabilized with 0.5% Triton X-100 in PBS for 5 min, washed with PBS and blocked in 5% bovine serum albumin (BSA) in PBS containing 0.05% Tween (PBS-T) for 1 h. Staining was done using mouse anti-γH2AX Ab (BioLegend) diluted 1:200 using the same dilution overnight at 4°C. Slides were then washed 3 ×5 min in PBS-T and secondary IgG DyLight 594 were used at 1:500 dilution for 1 h at room temperature.  Fig. 8b p-value = siGFP vs siINO80 = 0.047, p-value siGFP vs siINO80 = 0.007; p-value siINO80-siRNaseH1 vs predicted = 0.05. Figure 8c: p-value siGFP vs siINO80 = 0.009, siINO80-siRNaseH1 vs predicted = 0.046 Fig. 8d: p-value siGFP vs siRNaseH1 = 0.0005, p-value siGFP vs siINO80 > 0.0001, siINO80-siRNaseH1 vs predicted 0.0245 *p value < 0.05; **p-value < 0.01, ***p-value < 0.005; two-tailed unpaired Student's t-test. Data are presented as mean values ± SEM e Cell death analysis. PC3 cells transfected with control (siGFP), RNase H1-targeting (siRNaseH1) or INO80-targeting (siINO80) siRNAs for 24 h were treated with increasing concentrations of DL-Dopa for further 7 days incubation and analysed for cell growth (Supplementary Fig. 9f) and cytotoxicity. Cell death was calculated for cytotoxicity fluorescence values normalized to the respective relative cell growth. Fold change cell death values were calculated by setting untreated control cells arbitrarily to 1. Concentrations used for DL-Dopa inhibitor: non-treated (−), 1 μM, 2 μM and 5 μM. Data are presented as mean values ± SD, measure of centre is mean. Data are from three independent experiments. P values calculated by unpaired two-tailed t-test. (siGFP v siINO80 P value **p = 0.0036, siGFP 2 μM v siINO80 2 μM P value *p = 0.0464, siGFP 5 μM v siRNAseH1 5 μM P value *p = 0.0156. Confocal Microscopy and STED Nanoscopy. Super-resolution methodology was used for the visualization of R-loops and INO80. STED overcomes the diffraction limit of conventional confocal microscopy (*). This yields resolutions of ≥200 nm for visible light in the lateral dimensions (in x-y) and ≥500 nm, in the axial direction (in Z) (65). STED resolution is typically approximately 50 nm in XY and 150 nm in Z. Images were acquired on a Leica TCS SP8 STED 3X point scanning confocal nanoscopy with while light super continuum lasers and three STED depletion lasers (592, 660 and 775 nm) using STED WHITE HC PL APO CS2 100×/1.40 OIL lens. The DAPI and AF488 channels where acquired in confocal mode while the AF594 and ATTO647 channels where acquired in confocal and STED mode. Colocalization analysis were performed following the protocol in 67 . Images were deconvolved prior to quantification. Deconvolution, colocalization and particle analysis was performed with Huygens 18.04 from SVI (www.svi.nl). Cell proliferation. For proliferation analysis cells were seeded in 96-well plates 2000 cells/well. IncuCyte measurements of cellular occupation of the wells were taken every 3-6 h. Cell growth rate was normalised to the time point zero and additionally in a separate set of experiments cell numbers were counted at 96 h to assess cellular proliferation.
Cytotoxicity assay and genotoxic agents. After removal of 96-well plates from the IncuCyte, cytotoxicity was evaluated using CellTox Green® cytotoxicity assay (Promega) following the manufacturer's guidelines. CellTox green dye was diluted to 1:2000 in assay buffer and 25 μL added to each well of cells followed by incubation for 30 min at 37°C under a humidified atmosphere with 5% CO 2 . Fluorescence was measured at Ex:485, Em:520 nm in POLARstar® Omega microplate reader (BMG LABTECH).
Statistical tests. GraphPad PRISM version 7 and 8 were used to calculate significance and prepare graphs for presentation in figures.
DRIP-qPCR. DRIP assay was performed by performing IP with the S9.6 antibody in 2.5 μg genomic DNA and following the protocol of experiment 5 in ref. 66