Direct visualization of transcription-replication conflicts reveals post-replicative DNA:RNA hybrids

Transcription-replication collisions (TRCs) are crucial determinants of genome instability. R-loops were linked to head-on TRCs and proposed to obstruct replication fork progression. The underlying mechanisms, however, remained elusive due to the lack of direct visualization and of non-ambiguous research tools. Here, we ascertained the stability of estrogen-induced R-loops on the human genome, visualized them directly by electron microscopy (EM), and measured R-loop frequency and size at the single-molecule level. Combining EM and immuno-labeling on locus-specific head-on TRCs in bacteria, we observed the frequent accumulation of DNA:RNA hybrids behind replication forks. These post-replicative structures are linked to fork slowing and reversal across conflict regions and are distinct from physiological DNA:RNA hybrids at Okazaki fragments. Comet assays on nascent DNA revealed a marked delay in nascent DNA maturation in multiple conditions previously linked to R-loop accumulation. Altogether, our findings suggest that TRC-associated replication interference entails transactions that follow initial R-loop bypass by the replication fork.

In a healthy cell, transcription and replication are spatially and temporally separated across most organisms [1][2][3][4] ; however, in pathological conditions, such as oncogene overexpression, this coordination is perturbed, leading to TRCs. Ultimately, TRCs contribute to genomic instability and drive tumorigenesis 5,6 . In bacteria, TRCs lead to increased mutagenesis and accelerated evolution 7,8 . TRCs can occur in two different orientations, depending on whether transcription and replication progress towards each other (head-on) or in the same direction (co-directionally), with head-on conflicts being far more deleterious than co-directional conflicts [9][10][11][12] . DNA:RNA hybrid formation and topological stress have emerged as major drivers of TRCs, especially in the head-on orientation, but the underlying molecular mechanisms remain elusive.
Head-on conflicts favor the formation of DNA:RNA hybrids, which can compromise genome integrity and are therefore counteracted by specialized helicases [13][14][15][16][17][18][19][20] and RNases [21][22][23] . DNA:RNA hybrids are generally thought to occur in the context of co-transcriptional R-loops, which are three-stranded nucleic acid structures, generated by the reinvasion of the nascent mRNA into the DNA duplex behind the RNA polymerase (RNAP). R-loops can stall RNAPs, creating a potential barrier to replication. In vitro studies have suggested that replisomes stall only temporarily at these R-loop-associated RNAPs. By evicting Article https://doi.org/10.1038/s41594-023-00928- 6 We confirmed the induction of transcription-dependent and ribonuclease H (RNase H)-sensitive R-loops by gel shift assay 41 (Extended Data Fig. 1b) and ensured that they remained stable during EM sample preparation (Extended Data Fig. 1c). Using native EM 39 , we detected RNase H-sensitive R-loop structures on roughly 40% of circular and linear pFC53 fragments (Fig. 1a). These R-loop structures consist of a duplex DNA:RNA hybrid, indistinguishable from the surrounding double-stranded DNA (dsDNA), and a displaced single-stranded DNA (ssDNA) strand (Fig. 1b). R-loop size ranges from 40 base pairs (bp) to 670 bp (Fig. 1c). To determine whether R-loops localized to the Airn gene, we focused on the linearized fragment, where the R-loop is positioned asymmetrically. As shown in Figure 1d, R-loops are strongly clustered in the region that corresponds to the Airn gene.
In addition to visualization of the R-loop itself, we combined EM with S9.6-gold labeling of DNA:RNA hybrids (immuno EM). Binding of the S9.6-gold conjugate was detected on roughly 50% of circular and linear pFC53 molecules (Fig. 1e,f), with S9.6-gold binding clustering within the same region in which R-loops had accumulated in native EM analysis (Fig. 1g). S9.6-gold binding was both transcription-dependent and sensitive to RNase H treatment, confirming its specificity for co-transcriptional DNA:RNA hybrids (Fig. 1e). However, competition experiments-in which we gradually diluted the R-loop-containing pFC53 with a differently sized plasmid that lacks an R-loop-suggest that S9.6-gold specificity and selectivity drop with decreasing R-loop frequency (Extended Data Fig. 1d-f). As such, this immuno-EM approach is inherently limited and is applicable only to conditions of high R-loop enrichment (see below). Overall, we conclude that R-loops can be visualized in vitro by both native and immuno EM, while maintaining R-loop stability.

R-loop visualization and quantification on human genomic DNA
We applied this EM approach to directly visualize R-loops within the human genome. We used the estrogen-dependent breast-cancer cell line MCF7, which has previously been shown to exhibit a strong enrichment of R-loops 42 . To induce R-loop formation, MCF7 cells were deprived of estrogen for 48 hours and then stimulated with 100 nM estradiol (E2) for 2 hours. This estrogen stimulation causes a strong transcriptional burst, which drives R-loop formation. To assess whether we could induce R-loop formation efficiently, we performed quantitative PCR with DNA-RNA immunoprecipitation (qPCR-DRIP) for selected loci, including genes that are transcriptionally activated by estrogen (SLC7A5 and GREB1), as well as constitutively active genes that are prone to R-loop formation (RPL13A). In agreement with previous results, we consistently observed a strong increase in DRIP signal after E2 induction at both inducible loci ( Fig. 2a and Extended Data Fig. 2a,b).
To visualize these R-loops by native EM, we extracted genomic DNA before and after the transcriptional burst. The extracted genomic DNA shows an RNaseH-sensitive S9.6 signal, which was increased after the addition of E2 (Fig. 2b,c and Extended Data Fig. 2c,d). Importantly, R-loops on extracted DNA remained stable across the multiple steps of EM sample preparation, including genomic DNA digestion (Extended Data Fig. 2e,f) and purification (Extended Data Fig. 2g,h). Using an automated high-throughput EM workflow (Extended Data Fig. 2i) 43 , we screened 1,200-1,650 Mb of this genomic DNA for R-loop structures ( Fig. 2d and Extended Data Fig. 3a). We found, on average, 0.19 and 0.26 R-loop structures/Mb genomic DNA before and after the transcriptional burst, respectively (corresponding with, on average, 614 and 830 R-loops/cell). The size of these R-loops ranged from 30 bp to 3,000 bp, with most R-loops being smaller than 200 bp and the smallest ones possibly escaping detection (Fig. 2e). Interestingly, although small (<300 bp) R-loops were not reproducibly induced by the transcriptional burst (Extended Data Fig. 3b), we observed a marked and consistent induction of longer R-loops (>300 bp) (Fig. 2f). The higher the RNAP, the replisome can continue its progression 24,25 ; this eviction step is facilitated by accessory helicases, such as Pif1 (ref. 26 ). A recent study suggests that the DNA:RNA hybrid ahead of the fork could remain intact during helicase bypass, thereby being transferred behind the fork 25 . Accordingly, factors that assist fork restart and DNA-damage tolerance behind the fork also limit DNA:RNA hybrid accumulation 27,28 . Interestingly, both prokaryotic 24,25 and eukaryotic replisomes 26 , although they travel on opposite strands of the fork 29 , have the capacity to bypass an R-loop-based roadblock on the lagging strand. Together, these studies suggest that R-loop-associated RNAPs ahead of the fork might be only temporary impediments, and that TRC-induced replication interference extends to post-replicative processes. However, whether this is the case in vivo remains unknown.
Topological stress, in the form of negative and positive supercoiling, is a natural byproduct of both transcription and replication and is regulated by topoisomerases [30][31][32] . Interestingly, a recent study has highlighted the requirement of topological stress for fork collapse at TRCs 33 . Limiting topological stress by gyrase or Topo IV in Bacillus subtilis 34 , or topoisomerase 1 in human cells 32,33 , is critical for TRC resolution. Mechanistically, topological constraints can arise in three ways: first, head-on progression of transcription and replication may lead to a build-up of excessive positive supercoiling in between the approaching machinery, impairing DNA unwinding 34 . Second, negative supercoiling behind the RNAP may favor R-loop formation, thereby stalling RNAPs and creating a barrier to replication 35 . Finally, topological constraints behind the replication fork-which are well documented, albeit largely ignored in mechanistic models 31 -may disturb proper coordination of DNA polymerases, primases, and Okazaki-fragment processing, blocking DNA synthesis. The relative importance of these components of topological stress-ahead and behind replication forks-in R-loop formation and TRC-associated replication interference is currently unclear.
The main limitation in our understanding of hybrid-associated TRCs is the lack of unambiguous techniques for studying R-loop formation within the context of ongoing replication. First, most available techniques rely on either the DNA:RNA-hybrid-specific antibody S9.6 (ref. 36 ) or mutant RNase H enzymes 37 , both of which cannot distinguish between an R-loop and a DNA:RNA hybrid structure 38 . Second, our current understanding of TRC is largely based on population assays, which correlate R-loop and hybrid formation and signs of replication fork progression, but fail to pin both events to the same DNA locus simultaneously. Here, we adapted a well-established workflow for the visualization of replication intermediates (RI) by electron microscopy (EM) 39 , used it to visualize TRC-associated DNA:RNA hybrids, and addressed how they affect replication fork progression and architecture. We provide the first direct visualization of R-loops at TRC regions on the genome of two highly diverse organisms, that is bacteria and humans, and report R-loop frequency and size unambiguously, on the basis of single-molecule observations. In both bacterial and mammalian cells, DNA:RNA hybrid accumulation is linked to evident signs of replication stress, such as delayed fork progression and fork reversal. Although we did not detect canonical R-loops accumulating ahead of these transcriptionally challenged forks, we did observe that DNA:RNA hybrids and persistent discontinuities in the nascent strand-clearly distinct from Okazaki fragmentsaccumulate on daughter duplexes, suggesting that novel molecular mechanisms are involved in transcription-dependent replication interference.

Direct EM visualization of in vitro-generated R-loops
To establish an EM-directed workflow that allows R-loop visualization, we used an in vitro transcription system based on the pFC53 plasmid 40 , which contains the R-loop-prone promoter region of the mouse Airn gene under the control of the T3 promoter (Extended Data Fig. 1a).
Article https://doi.org/10.1038/s41594-023-00928-6 levels of long R-loops in these conditions may also partially reflect increased stability or detectability of long hybrids in EM preparation and visualization.
Combining R-loop frequency and size information, we calculated the overall R-loop burden, that is the fraction of the genome that is involved in R-loop formation, assuming that R-loop formation was homogeneous within the cell population (Fig. 2g). R-loop burden reproducibly increased by twofold after the transcriptional burst, reaching a maximum of 0.009% of the human genome, meaning that about 290 kb of a cell's genome is involved in R-loop formation. Altogether, our EM approach provides the first direct visualization of R-loops in vivo and allows unambiguous quantification of R-loop frequency, size, and burden.

Fork slowing and delayed nascent strand maturation at TRCs
We next wondered whether and how R-loop formation affects replication fork progression. We used DNA fiber assays, in which ongoing DNA synthesis is labeled sequentially by addition of nucleotide analogs, that is CldU and IdU, to follow fork speed before and after the transcriptional burst (Fig. 3a). We observed a decrease in CldU tract length specifically upon transcriptional burst induction, indicative of a replication fork slowdown. Overexpression of RNase H1 labeled with green fluorescent protein (GFP) (Extended Data Fig. 4a) restored normal fork progression upon the transcriptional burst ( Fig. 3a and Extended Data Fig. 4b), linking the observed fork slowdown to accumulation of DNA:RNA hybrids. Moreover, downregulation of the DNA translocase ZRANB3 (Extended Data Fig. 4c)-previously implicated in the remodeling of replication forks into four-way junctions, that is reversed forks 44 -rescued fork speed upon E2-induced transcription ( Fig. 3b and Extended Data Fig. 4d). A similar rescue of fork speed was observed upon PARP inhibition (Extended Data Fig. 4e), which has previously been shown to impair reversed fork stability 45 , and was linked to a marked increase in γH2AX accumulation in stimulated cells (Extended Data Fig. 4f,g). Altogether, these data suggest that the transcriptional  Numbers below x-axes in a and e indicate the total number of molecules analyzed in all replicates. In d and g the position of the R-loop forming mAirn gene is indicated in light blue below the graph.
To assess whether this transcription-dependent replication interference is associated with any particular structural features, we conducted a thorough EM analysis of replication intermediates (RIs) after a transcriptional burst. We observed numerous small (<30 nt) ssDNA gaps that accumulated behind replication forks, at a close distance (<3 Kb) ( Fig. 3c and Extended Data Fig. 4h,i, blue dots), possibly reflecting short discontinuities that are physiologically associated with lagging strand synthesis and maturation. A fraction of the observed ssDNA gaps was unexpectedly large (>30 nt) and/or persisted at a notable distance from the fork regardless of E2 (Extended Data Fig. 4h,i, red dots), further suggesting that there is a form of endogenous replication stress in MCF7 cells. However, when we subjected this genomic DNA to in vitro RNase H treatment, we observed a marked increase in larger and/or persistent ssDNA gaps, which was specific to E2 addition (Fig. 3d,e). This effect was observed in both independent EM replicates ( Fig. 3e and Extended Data Fig. 3h,i) and suggested that-despite endogenous RNase H activity-RNA molecules remain embedded in the duplex DNA up to 15 kb behind the fork. In vitro RNase H treatment degrades this embedded RNA, generating larger post-replicative ssDNA gaps (Fig. 3d,e).
Because of restriction digestion of genomic DNA, the analysis of nascent strand discontinuities by EM is limited to a few kb behind    replication forks. Moreover, particularly small ssDNA gaps may escape detection. To overcome these limitations and to test the hypothesis that embedded DNA:RNA hybrids may lead to persistent discontinuities in nascent DNA, we optimized a recently published alkaline comet assay 46 to monitor nascent strand maturation, after pulse-labeling with EdU and a chase period of variable duration (0-1.5 hours; Extended Data Fig. 4j). In unperturbed conditions, our assay reveals physiological discontinuities during nascent strand synthesis, which are rapidly resolved upon nascent strand maturation (0.5-1.0 hour; Extended Data Fig. 4j). We reckoned that this assay may be used to reveal additional or persistent discontinuities, possibly associated with the accumulation of RNA molecules within nascent DNA (Fig. 3f). In agreement with previous assays, some nascent strand discontinuities persisted up to 0.5 hours in MCF7 cells, even in the absence of E2, likely marking endogenous replication stress (Fig. 3g). However, E2 addition drastically increased the half-life of nascent strand discontinuities, which were still detected 1 hour after EdU chase ( Fig. 3g and Extended Data Fig. 4k). We then applied the same assays to other conditions that have previously been linked to DNA:RNA hybrid-dependent fork slowing, such as TOP1 depletion 33 or TOP1 inhibition by camptothecin (CPT) 47 . Strikingly, in all tested conditions, the persistence of nascent strand discontinuities was markedly increased when compared with that in the matched control conditions (Fig. 3h,i and Extended Data Fig. 4l-m), with discontinuities clearly visible in CPT-treated cells even 1.5 hours after EdU removal (Fig. 3i). Altogether, our data suggest that post-replicative DNA:RNA hybrids and persistent nascent strand discontinuities associate with delayed fork progression upon a transcriptional burst and other conditions of DNA:RNA hybrid accumulation.

Fork slowing and reversal at locus-specific TRCs in bacteria
Our observations in MCF7 cells prompted us to explore a potential role for post-replicative DNA:RNA hybrids in replication interference at defined transcription-replication conflicts. To address this question, we took advantage of locus-specific head-on TRCs in the chromosome of the bacterial model system Bacillus subtilis. This B. subtilis strain carries an IPTG-inducible transcription unit that is oriented head-on to the bacterial origin (Fig. 4a). Previous work showed that expression of this engineered conflict increases hybrid formation and replication fork stalling within the same conflict region, in the absence of activity of RNase H III (Δrnhc mutant strain) 9,48 . Combining this locus-specific TRC model system with our EM approach, we explored how replication forks progress through this TRC.
To isolate the conflict region from the rest of the bacterial genomic DNA, we designed a digestion/gel-extraction-based protocol (Fig. 4a) and obtained RIs of amenable size to perform EM analysis 49 . We first confirmed by qPCR-DRIP that DNA:RNA hybrids accumulate in the conflict region and remain stable throughout the digestion/gel-extraction procedure. In agreement with published results, we detected a significant and RNase H-sensitive increase in hybrid formation in the Δrnhc strain (Fig. 4b). To assess replication fork progression through the conflict region, we analyzed the replicating conflict fraction of both wild-type (WT) and Δrnhc strains, with or without IPTG treatment, by EM in three independent biological replicates (Fig. 4c,d and Extended Data Fig. 5a, additional replicates in Extended Data Fig. 5b-g). We collected all detectable RIs and sorted them by fragment length (Fig. 4c). In the WT-IPTG strain, a condition in which forks are not expected to slow down at the conflict locus, we observed RIs of all sizes between 5 kb and 15 kb. The lack of enrichment for the conflict fragment in this control condition impairs the identification and reliable EM analysis of the conflict region. Conversely, the WT strain with IPTG, as well as Δrnhc with or without IPTG, showed a clear enrichment of RI with sizes matching the fragment of interest (marked in dark grey in Fig. 4c). This suggests that replication forks stall frequently in this conflict region when conflict is induced. The enrichment of conflict RIs in Δrnhc without IPTG is likely due to leaky transcription (which has been previously documented for the P spank(hy) promoter), in combination with perturbed clearance of hybrid accumulation. To assess where fork stalling occurs within this conflict region, we sorted the selected RIs and aligned them on the basis of the length of the daughter strand (Fig. 4d). In this analysis, replication fork stalling at a defined location should result in an accumulation of intermediates with a defined ratio of daughter (green) versus parental (gray) strand. In all three samples, however, we observed a uniform distribution of RIs across the conflict region. This suggests that transcription-replication interference does not block replication forks at a specific location, but rather leads to an overall slowing of replication fork progression throughout the conflict region, in line with published two-dimensional gel analysis 9 . Finally, we assessed replication fork remodeling. Conflict-induced RIs displaying reversed forks are shown in pink in Fig. 4d and sorted according to the position of the original junction (prior to reversal). We observed that hybrid accumulation in the Δrnhc sample treated with IPTG was associated with a significant increase in replication fork reversal (Fig. 4e). Altogether, our data suggest that TRCs lead to replication fork stalling and reversal across the conflict region.

Post-replicative DNA:RNA hybrids at head-on bacterial TRCs
Inspired by our evidence from the mammalian system (Fig. 3), we next investigated whether post-replicative DNA:RNA hybrids are present at TRC regions across species. We found that RIs from the WT bacterial strain display reproducibly higher levels of ssDNA gaps behind the fork than the Δrnhc mutant strain, which lacks endogenous RNase H III activity (Fig. 5a).
To test whether RNA remains embedded in the conflict DNA in the Δrnhc mutant-strain, we employed our S9.6-gold-based immuno EM approach (Fig. 1f) 49 . We compared S9.6-gold binding to the replicating conflict fraction with that to the linear conflict fraction. As a negative control, we included a bulk DNA fraction, which was extracted from a different region of the gel and is devoid of the transcriptional conflict; this fraction contained both linear and replicating DNA. As an internal control for S9.6-gold specificity, we spiked each extracted B. subtilis chromosomal fraction with an excess of R-loop-carrying, linearized pFC53 prior to S9.6-gold labeling and EM spreading. As depicted in Figure 5b (additional replicates are in Extended Data Figure 6a,b), B. subtilis DNA and pFC53 were identified by their respective sizes. S9.6-gold binding to pFC53 was constant between samples and replicates (Fig. 5c) and strongly reduced upon RNase H digestion, confirming that S9.6-gold labeling conditions were specific. In B. subtilis, S9.6-gold binding was observed on 30-40% of linear conflict intermediates, a reproducible increase compared with the bulk fraction. RNase H digestion markedly reduced S9.6-gold binding to the linear conflict intermediates and led to corresponding accumulation of ssDNA gaps (Fig. 5c-d and Extended Data Fig. 6c), suggesting that S9.6-gold does indeed label embedded DNA:RNA hybrids. Excitingly, S9.6-gold binding to B. subtilis RIs was significantly increased compared with both the linear conflict and bulk DNA, with 40-60% of RIs carrying at least one label (representative images, Fig. 5e). Importantly, S9.6-gold binding to bulk RIs is close to background levels and is not decreased upon RNase H treatment, excluding the concept that S9.6-gold binding reflects short DNA:RNA hybrids that are physiologically present at all RIs (for example, Okazaki fragments; Fig. 5c). S9.6-gold labels were found scattered throughout the conflict region ( Fig. 5f and Extended Data Fig. 6d,e) on both parental and daughter strands. S9.6-gold binding to the parental DNA strand was detected on roughly 20% of the RIs. Because our native EM analysis (Fig. 4) did not reveal any detectable R-loop structures on parental DNA-even though DNA:RNA hybrids remain stable throughout the gel extraction procedure (Extended Data Fig. 7a,b)-we propose that S9.6-gold labeling ahead of forks reflects R-loop structures that are processed by removal of the displaced ssDNA and thus are indistinguishable from duplex DNA Article https://doi.org/10.1038/s41594-023-00928-6 in native EM. However, the majority of S9.6-gold labels on RIs bound to the daughter strands ( Fig. 5f and Extended Data Fig. 6d,e), suggesting that there was extensive DNA:RNA hybrid formation on nascent DNA behind the replication fork. Altogether, these results strongly suggest that, in the absence of endogenous RNase H, RNA remains embedded in duplex DNA at transcription-replication conflicts.

Discussion
Using a direct EM-based visualization approach, we measured frequency and size of R-loops on genomic DNA upon a transcriptional burst. Moreover, we directly monitored DNA:RNA hybrids at replication forks progressing 'head-on' through transcriptionally active regions, where we found them to markedly accumulate as 'embedded' hybrids   on newly replicated DNA. This observation prompts us to propose a new, integrated model of head-on transcription-replication interference (Fig. 6). We propose that R-loops forming behind an RNAP that approaches the replisome head-on are either processed into DNA:RNA hybrids and/or efficiently bypassed by the replisome. The DNA:RNA hybrid is thereby transferred to the lagging strand behind the replisome, and is typically removed and processed quickly, to allow continuous DNA synthesis. In case of excessive accumulation or impaired removal of these DNA:RNA hybrids behind the fork, replication fork progression is impaired and is frequently associated with delayed nascent strand maturation and fork reversal. Accordingly, post-replicative processes-such as nascent DNA resection-are prevented in the absence of RNase H2 in both yeast and human cells (P. Pasero, personal communication), further suggesting that DNA:RNA hybrids may act behind the fork to interfere with DNA replication. Although excessive transcription and defective removal of transient DNA:RNA hybrids led to similar molecular phenotypes across species in our EM analyses, our data do not exclude that other specific defects in RNA metabolism may alter this sequence of events in drastically different ways, possibly leading to accumulation of different pathological intermediates.

R-loops form on human genomic DNA and can be visualized by EM
Using our optimized native EM approach 43 , we provide the first direct visualization of R-loops on genomic DNA and quantify their frequency and size unambiguously at the single-molecule level. In unchallenged MCF7 cells, we detected an average of 614 R-loops/cell, with a size range of 30-3,000 bp. Previous sequencing-based studies estimated R-loop frequency to be around 300 R-loops/cell in unperturbed HeLa cells 50 and R-loop size to range between 100 and 450 bp, with some R-loops reaching up to 2.7 kb in size 51 . Our S9.6-independent approach provides visual support for these previous estimations and allows simultaneous retrieval of information on R-loop frequency and size, providing mechanistic clues about R-loop-mediated genomic instability. Upon an estrogen-induced transcriptional burst, for example, we observed that long R-loops (>300 bp) are most reproducibly increased, suggesting that defined subsets of R-loops are differently affected by specific perturbations and may differentially impact DNA replication and genome stability.

DNA:RNA hybrids accumulation ahead of forks facing TRCs
Although our EM technique proved compatible with R-loop stability and could detect R-loops of sizes and frequencies comparable to those in previous studies, we did not detect an accumulation of R-loops ahead of replication forks, even at a locus-specific head-on TRC. However, S9.6-gold-dependent immuno EM of the conflict locus did reveal the presence of DNA:RNA hybrid structures ahead of forks, which are indistinguishable from duplex DNA in native EM. Our data show that, despite justified concerns about the use of S9.6 as an unambiguous readout of DNA:RNA hybrids 38 , rigorous internal controls for antibody specificity allow it to be used as a reliable, potentially revealing research tool. We propose that these DNA:RNA structures arise from R-loop processing by topoisomerases and/or structure-specific nucleases 18,32 . Torsional stress accumulating at sites of transcription-replication interference is reportedly addressed by topoisomerases [32][33][34] . By cleaving the displaced strand of the R-loop, topoisomerases/nucleases may locally relieve torsional constraints and convert the three-stranded R-loop into a DNA:RNA hybrid structure. Although R-loop incision may expose cells to the risk of fork collapse and chromosomal breakage 18,33 , it has recently been shown that cleavage-religation cycles are actively contributing to fork progression and restart at TRCs 47 . This suggests that nucleolytic processing may in fact be physiologically linked to R-loop bypass. We cannot exclude, however, that a fraction of R-loops escapes processing, yet remains undetected in our EM analysis. In line with in vitro observations 25 , these R-loops may be bypassed very efficiently by the replisome, preventing detectable accumulation of R-loops and forks in the same restriction fragment. Whether through processing or bypass, we consider it likely that an R-loop ahead of the fork can be efficiently transferred behind the fork as a DNA:RNA hybrid, without markedly affecting the fork progression rate (Fig. 6).

Origin and clearance of post-replicative DNA:RNA hybrids
At a bacterial locus-specific TRC impairing fork progression, S9.6-gold binding was most frequently observed behind the replication fork. Accordingly, in all tested genetic conditions that have previously been associated with DNA:RNA hybrid accumulation and hybrid-dependent fork slowing in human cells-that is, estrogen-dependent transcriptional burst 42 , TOP1 depletion 33 , and TOP1 inhibition 47 -we observed accumulation of nascent strand discontinuities, persisting well after replication. Combined, this evidence strongly suggests that post-replicative DNA:RNA hybrids are frequent and pathological intermediates at TRCs and may contribute to the observed impairment of fork progression. We envision three potential mechanisms through which post-replicative DNA:RNA hybrids can be generated (Extended Data Fig. 8a). First, transcription of a hybrid-prone sequence could be initiated behind the replication fork, and the elongating RNA may anneal to the unwound lagging strand template, as has recently been suggested 27 . Alternatively, the nascent RNA could invade the duplex daughter DNA strand, forming a post-replicative R-loop that is nucleolytically processed into a hybrid (Extended Data Fig. 8a, 1). Second, a hybrid ahead of the fork could remain bound and subsequently bypassed by the replisome (Extended Data Fig. 8a, 2). Finally, the annealed mRNA ahead of the fork may be displaced by replicative and/or accessory helicases and reanneal behind the replication fork (Extended Data Fig. 8a, 3). Because nascent chromatin is transcriptionally silenced 52 , we consider it unlikely-at least in the eukaryotic system-that the RNA is newly synthesized behind the fork, and currently favor the hypothesis that RNA might be inherited from preexisting hybrids ahead of the fork. The replicative helicase activity-which differs profoundly between prokaryotes and eukaryotes-may determine to what extent the hybrid remains intact; although it has been suggested that the RNA remains bound during prokaryotic TRCs 25 , it appears to be liberated during eukaryotic TRCs 26 . Interestingly, we observed comparable phenotypes (fork slowing and post-replicative hybrid formation) in both systems, suggesting that the impact of the bound RNA is independent of its origin. Whether post-replicative DNA:RNA hybrid formation requires specific RNA helicases and/or active transcription during replication will be fascinating questions for future studies. DNA:RNA hybrid resolving enzymes, such as SETX and RNase H2 (refs. 53,54 ), have been shown to travel with the replisome, suggesting that-in physiological conditions-DNA:RNA hybrids forming on the lagging strand are efficiently removed to allow unperturbed replication progression (Fig. 6). However, in case of excessive hybrid formation-for example, multiple or particularly long hybrids, as we observed upon transcriptional bursts or RNase H inactivation (Fig. 6)-these processing mechanisms might be insufficient or saturated. In that case, 'post poning' hybrid processing may be favorable, to enable restoration of fork progression and to limit excessive ssDNA accumulation, leading to post-replicative hybrids that remain embedded within the daughter duplexes long after fork passage. Intriguingly, several factors involved in fork restart and DNA-damage tolerance behind replication forks have recently been shown to contribute to hybrid resolution, through yet-elusive mechanisms 27,28 . Further investigations will be needed to address whether and how specific enzymes involved in DNA:RNA hybrid metabolism-some of which were implicated in human disease 17,55-60 -are specifically involved in resolving or preventing post-replicative DNA:RNA hybrids. Moreover, analogous analyses on co-directional conflicts may help clarify why-across species-head-on conflicts have more deleterious consequences than do co-directional ones 9,10 .

DNA:RNA hybrids, topological stress, and fork remodeling
We observed that excessive DNA:RNA hybrid accumulation at replication forks is associated with signs of replication stress, such as fork slowing and replication fork reversal. DNA:RNA hybrid accumulation behind the fork may impair replication fork progression through several different mechanisms (Extended Data Fig. 8b). First, unresolved DNA:RNA hybrids might remain tethered to the replisome to await processing by replisome-associated RNA nucleases or helicases (Extended Data Fig. 8b, 1). This could impose excessive DNA looping at ongoing forks, increase torsional constraints, and finally impair replisome function. Topological constraints have previously been linked to replication interference and genomic instability at TRCs, although they are mainly attributed to the build-up of positive supercoiling in between replication and transcription machinery on parental DNA 3,32 . Sterical exclusion of topoisomerases and/or excessive supercoiling ahead of the fork may impede resolution of torsional constraints, which may, however, be transferred behind the fork through fork rotation, leading to pre-catenane formation 31 . Importantly, excessive fork rotation per se may interfere with fork progression and prevent efficient post-replicative processing of DNA:RNA hybrids 61 , further exacerbating DNA:RNA hybrid accumulation. The topological constraints behind the replication fork might therefore be a combination of both the topological stress directly arising behind the fork and topological stress being transferred from ahead of the fork. Accordingly, bacterial topoisomerases have recently been shown to work both ahead and behind replication forks at TRCs 34 . Second, hybrid processing can expose stretches of excessive ssDNA, leading to RAD51 loading and fork reversal and thereby transiently preventing replication fork progression (Extended Data Fig. 8b, 2) 62 . Indeed, in line with recent evidence on drug-induced accumulation of DNA:RNA hybrids 47 , fork reversal emerged in our study as a critical response to endogenous transcription-replication conflicts. Besides ssDNA accumulation, increased topological stress 63 was shown to promote fork remodeling ( Fig. 6 and Extended Data Fig. 8b). The direct causative link between DNA:RNA hybrid formation and fork reversal, as well as the interplay with other processes, such as repriming, remain to be explored. Moreover, besides affecting fork remodeling and topology, accumulation of DNA:RNA hybrids may interfere with efficient chromatinization of the replicated duplexes (Extended Data Fig. 8b, 3), which has previously been shown to alter the rate of replication fork progression 64 . Finally, fork reversal has previously been shown to extend as a global nuclear response, even though lesions may locally affect only a small number of replication forks 65 . It is thus conceivable that-despite the relatively low number of DNA:RNA hybrids directly detected by our EM methodsfork remodeling may extend to a much higher number of replication forks than those directly challenged by hybrids and hybrid-associated topological constraints or chromatin-maturation issues.
To conclude, our study highlights post-replicative DNA:RNA hybrids as the most abundant intermediates accumulating at transcriptionally challenged forks, raising a plethora of new and intriguing questions to be explored. This may motivate researchers to revisit previous observations and orient future work, to expand our understanding of how formation of DNA:RNA hybrids interferes with ongoing replication and ultimately drives genomic instability.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, Under physiological conditions, this hybrid is rapidly resolved, allowing unperturbed fork progression. In case of excessive accumulation (that is transcriptional burst) or impaired removal (that is limited RNase H activity), DNA:RNA hybrids may be still efficiently bypassed, but their accumulation behind the fork would result in delayed fork progression and frequent fork reversal (see Extended Data Fig. 8b for potential mechanisms). Upon stress resolution, the reversed fork can be restarted to complete DNA replication.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Transfections
For RNase H1-GFP overexpression, MCF7 cells were transfected with pAIO RNH1-GFP using Lipofectamine 3000, according to manufacturer's instructions, 48 hours prior to estrogen stimulation. For ZRANB3 downregulation, MCF7 cells were transfected with siRNA targeted against ZRANB3 (40 nM) using RNAiMAX, according to the manufacturer's instructions, 96 hours prior to estrogen stimulation.

DNA fiber spreading analysis
Following estrogen depletion and stimulation, MCF7 cells were sequentially pulse-labeled with 30 µM CldU and 250 µM IdU for 30 minutes. The cells were collected by scraping in ice-cold PBS, washed, and resuspended in PBS at 3 × 10 5 cells/mL. Then, 3 µL of cells was mixed with 7 µlL of lysis buffer (200 mM Tris-HCl pH 7.5, 50 mM EDTA, 0.5% SDS) on a glass slide. After 5 minutes, the slides were tilted at 15-45°, and the resulting DNA spreads were air dried, fixed in 3:1 methanol/acetic acid overnight at 4 °C. The fibers were denatured with 2.5 M HCl for 1 hour, washed with PBS and blocked in 2% BSA/PBS/0.1% Tween for 40 minutes. The newly replicated CldU and IdU tracks were labeled with anti-BrdU antibodies recognizing CldU (1:500) and IdU (1:100) for 2.5 hours at RT, followed by 1 hour of incubation with secondary antibodies (anti-mouse-AlexaFluor 488, 1:300, and anti-rat-Cy3, 1:150) at RT in the dark. Fibers were mounted using ProLong Gold AntiFade, visualized (Leica DMI 6000; objective lenses: HC PL APO ×63, 1.40 numerical aperture (NA) oil; Leica Application Suite X 3.6.0.20104) and analyzed using ImageJ software (version 2.0.0-rc-43/1.51h). Of note, we noticed that MCF7 cells experience issues incorporating IdU and therefore considered the CldU tract length as the more reliable readout for these fiber experiments.

Flow cytometry
For flow cytometry analysis, MCF7 cells were labeled with 10 µM EdU for 30 minutes, collected by scraping into ice-cold PBS, and fixed with 4% PFA/PBS for 15 minutes at RT. In between all the following steps, cells were washed twice with 1% BSA/PBS at 500g for 5 minutes. For γH2AX detection, cells were incubated with primary antibody (ms anti-γH2AX, 1:1,000) in 1× saponin buffer for 2 hours at RT with occasional inversions and subsequently with secondary antibody (anti-mouse-Alexa Fluor 647, 1:125) for 30 minutes at RT. For all experiments, the EdU Click reaction was performed for 30 minutes at RT, according to the manufacturer's instructions. Finally, cells were resuspended in 1% BSA/PBS, containing 0.1 mg/ml RNase A and 1 µg/ml DAPI. Samples were measured on an Attune NXT flow cytometer (Beckman Coulter; Attune NTX Software Version 4.2.0.) and analyzed by the FlowJo software (version 10.4).

EdU alkaline comet assay
MCF7, U2OS or HeLa cells were treated as indicated and labeled with 10 µM EdU for the final 30 minutes of the treatment. Cells were washed with PBS and incubated in fresh medium without EdU for variable time periods (chase time of 0 hours, 0.5 hours, 1 hour, or 1.5 hours); the respective treatments were maintained during the entire chase time.

Nature Structural & Molecular Biology
Article https://doi.org/10.1038/s41594-023-00928-6 Cells were then collected by trypsinization, embedded in 0.8% Sea-Plaque low-melting point agarose on two-well comet slides, and lysed in lysis buffer (10 mM Tris pH 10, 2.5 M NaCl, 100 mM EDTA, 10% DMSO, 1% Triton X-100). Slides were then washed in PBS and incubated in denaturation buffer (300 mM NaOH, 1 mM EDTA) for 40 minutes, followed by electrophoresis for 20 minutes at 18 V and 300 mA. Afterwards, slides were washed in PBS, fixed in ice-cold ethanol for 10 minutes, and dried at 37 °C. Slides were then subjected to EdU Click IT reaction for 30 minutes at RT, washed with PBS, stained with SYBR for 15 minutes at RT, and washed again. The dried slides were imaged (Leica DMI 6000; objective lenses: HC PL FLUOTAR ×40/0.80) and analyzed for their olive tail moment using the Open Comet plugin (version 1.3.1) for ImageJ (version 2.0.0-rc-43/1.51h). To assess nascent strand maturation over time, median olive tail moments were normalized to the 0-hour chase time point. Half-lives of nascent strand discontinuities were calculated by fitting an exponential decay function to the normalized olive tail moments.

Conflict enrichment by gel electrophoresis
Fixed B. subtilis cells were collected by centrifugation at 5,500g for 10 minutes and washed with ice-cold PBS three times. Cells were then resuspended in PBS and cross-linked with 200 µg/mL 4,5′,8-trimethylpsoralen, followed by a 2-minute irradiation pulse with UV 365 nm monochromatic light. Cross-linked cells were washed with ice-cold PBS three times and subsequently resuspended in 0.5% low melt NuSieve GTG agarose, containing 125 µg/mL lysozyme (dissolved in water). The mixture was placed in an agarose plug mold until solidified. Agarose plugs were then incubated in 1 mL of 10 mM Tris (pH 7.5) 500 mM EDTA buffer at 37 °C overnight. After incubation, 400 µL of 5% sarkosyl 500 mM EDTA buffer and 100 µL of 20 mg/mL proteinase K was added. The plugs were then incubated at 50 °C for 5 hours and subsequently washed eight times (for 4 hours each wash step) with 2 mM Tris (pH 8) 1 mM EDTA buffer at 4 °C. For digestion, agarose plugs were equilibrated in 1× cutsmart buffer for at least 4 hours at RT and 1 hour at 37 °C and then digested with 180U BstBI, 180U NruI-HF, 180U PstPI and 90U BfaI in a wet chamber at 37 °C overnight. The next day the digestion mixture was renewed for an additional 3hrs at 37 °C. Plugs were poured into a 0.5% low-melting NuSieve GTG agarose gel in 1× TBE and run at 50 V for 21 hours at 4 °C without intercalating agents. The agarose gel was stained with ethidium bromide, and areas of interest excised with minimal UV exposure for gel extraction using the Silica Bead DNA Gel Extraction Kit according to manufacturer's instructions with minor modifications: Gel pieces were weighed and dissolved in 0.5 volumes/weight of TBE conversion buffer and 2 volumes/weight of binding buffer for a maximum of 5 minutes at 50 °C. Silica beads were added according to DNA content predictions, incubated at 50 °C for 5 minutes to allow DNA binding, and washed three times briefly with ice-cold washing buffer. During washing steps, the pellet was not resuspended to prevent DNA shearing. The beads were air dried for 10 minutes, and the DNA was eluted 2-4 times in TE pH 7.4 or dH 2 O; the volume for each elution was chosen to be equal to the volume of beads used in the reaction. If gel pieces exceeded the recommended volumes, the reaction was split into separate tubes after the gel was dissolved. In that case, eluates were pooled and subsequently concentrated by speed vac. A detailed protocol is available in ref. 49 .

MCF7 genomic DNA extraction and digestion
Following estrogen depletion and stimulation, MCF7 cells were collected by scraping in ice-cold PBS, washed, and lysed in TE buffer pH 7.4, containing 0.5% SDS and 0.6 mg/mL proteinase K at 37 °C overnight. The next day, 1.8 volumes of G2 buffer (800 mM guanidine-HCl, 30 mM Tris-HCl (pH 8.0), 30 mM EDTA (pH 8.0), 5% Tween-20, and 0.5% Triton X-100), containing 0.3 mg/ml proteinase K were carefully added to the samples and incubated at 37 °C for 1.5 hours. The DNA was subsequently purified using chloroform/isoamylalcohol (24:1) and precipitated in two volumes of 100% ethanol and 0.1 M sodium acetate. Finally, the DNA was washed with 70% ethanol, briefly dried at 37 °C, and resuspended in TE buffer pH 7.4 at 37 °C with slight agitation.
For dot blot and EM analysis, 10 µg of this genomic DNA was digested with the indicated restriction enzymes at 37 °C overnight. Samples were then digested with 0.25 mg/mL RNase A, 25 U/mL RNase T1, 25 U/mL RNase III ± 500 U/mL RNase H in NEB RNase H buffer, supplemented with 0.5 mM NaCl for 2.5 hours at 37 °C. The digested DNA was purified using the Silica Bead DNA Gel Extraction Kit, according to the manufacturer's instructions.

DNA:RNA immunoprecipitation
Extracted and purified genomic DNA from MCF7 cells was digested with HindIII-HF, EcoRI, BsrGI, XbaI, and SspI at 37 °C overnight, purified with phenol/chloroform/isoamylalcohol (25:24:1) and precipitated with 2 volumes 100% ethanol, 0.1 mM sodium acetate, and 75 µg/ml glycogen at −20 °C overnight. Finally, the DNA was washed with 70% ethanol, briefly dried at 37 °C and resuspended in TE buffer pH7.4 at 37 °C with slight agitation. RNase H controls were digested with 10U/ µg RNase H at 37 °C overnight. For the DNA:RNA immunoprecipitation (DRIP), 4.4 µg of digested and purified DNA was incubated with 10 µg of S9.6 antibody (Kerafast) in a total of 500 µL 1× binding buffer (100 mM NaPO 4 pH7.0, 1.4 M NaCl, 0.5% Triton X-100) at 4 °C overnight with steady rotation. Ten percent of the sample was kept aside as input control prior to antibody addition. To immobilize antibody-bound material, samples were incubated with 50 µL of magnetic protein A/G beads at 4 °C for 2 hours with steady rotation. Beads were washed three times for 10 minutes with 1× binding buffer and incubated in 250 µL 1× elution buffer (50 mM Tris-HCL pH 8.0, 10 mM EDTA pH 8.0, 0.5% SDS) with 0.6 mM proteinase K at 55 °C for 45 minutes with occasional inversion. Eluted DNA was purified by phenol/chloroform/isoamylalcohol (25:24:1) extraction and ethanol/sodium acetate/glycogen precipitation. qPCR was performed on a Roche LightCycler 480 Instrument II using SYBR-Green master mix (Bio-Rad Laboratories). Primers used for qPCR are listed in the Supplementary Table. For B. subtilis, digested genomic DNA was extracted from whole plugs using the Silica Bead DNA Gel Extraction Kit, as described above. RNase H controls were digested with 2.5 U/µg RNase H at 37 °C overnight. DRIP was performed as described above, with minor changes (500 ng of material, 5 µg of S9.6 antibody, 25 µg of beads).

Native electron microscopy
For native DNA spreading for EM analysis, 50-150 ng of DNA was mixed with benzyldimethylalkylammonium chloride (BAC) and formamide, spread on a water surface and loaded on carbon-coated 400-mesh copper grids, as previously described 39 . Subsequently, DNA was coated with platinum using the high vacuum evaporator MED 020 (BalTec). In vitro transcription samples and B. subtilis samples were imaged manually using a Tecnai G2 Spirit transmission electron microscope (FEI; LaB6 filament; high tension ≤120 kV; acquisition software: DigitalMicrograph Version 1.83.842 (Gatan)) with a side-mounted digital camera Gatan Orius 1000 (2,600 × 4,000 pixels). For MCF7 samples, imaging was automated using a Talos 120 transmission electron microscope (FEI; LaB6 filament, high tension ≤120 kV) with a bottom mounted CMOS camera BM-Ceta (4,000 × 4,000 pixels) and the MAPS software package (version 3.16, Thermo Fisher Scientific). Samples were annotated for molecules of interest using the MAPS Viewer software (version 3.16) and overlapping images for annotated regions were stitched together using the automated pipeline ForkStitcher (https://github.com/jluethi/ForkStitcher, Version 0.1.1), developed by J. Lüthi. The stitching was based on the ImageJ Stitching plugin (ref. 66 , version 3.1.6), used through the PyImageJ library. Total DNA content was determined using a DNA-content quantification algorithm (https:// github.com/roessler-f/DNAQuantification), developed by F. Roessler. A detailed protocol for both pipelines is available in ref. 43 .
Article https://doi.org/10.1038/s41594-023-00928-6 Extended Data Fig. 8 | Possible mechanisms of post-replicative hybrid formation and hybrid-dependent replication interference. a) Possible mechanisms of post-replicative DNA:RNA hybrid formation. Initiation of de novo transcription behind the replication fork generates nascent RNA, which can either anneal directly to exposed ssDNA on the lagging strand or reinvade the duplex daughter strand to form an R-loop, prone to nucleolytic processing (1). Alternatively, the RNA can be inherited from preexisting hybrids ahead of the fork either through replisome bypass (2) or through hybrid disassembly and subsequent RNA reannealing to the exposed ssDNA behind the fork (3). b) Possible mechanisms of replication interference by long/accumulating post replicative DNA:RNA hybrids. DNA:RNA hybrids forming on the lagging strand may remain tethered to the replisome through RNA-binding and/or -processing enzymes, creating torsional constrains through DNA looping (1). Alternatively, hybrid processing may expose ssDNA, which induces RAD51 loading and subsequent fork reversal (2). Finally, excessive RNA:DNA hybrid formation may interfere with chromatinization of the newly synthesized daughter strand (3). In both a and b, the alternative mechanisms may coexist, contributing to the observed structures.