R‐loop formation during S phase is restricted by PrimPol‐mediated repriming

Instability of BU‐1 expression monitors replication delay at (GAA)_n

We have previously shown that expression instability of the BU‐1 locus in chicken DT40 cells provides a sensitive readout for replication delay at G4 motifs (Schiavone et al, 2014). The wild‐type locus contains a G4 motif 3.5 kb downstream of the TSS (the +3.5 G4) towards the end of the second intron (Fig 1A), which is responsible for replication‐dependent instability of BU‐1 expression under conditions in which G4 replication is impaired (Sarkies et al, 2012; Schiavone et al, 2014; Guilbaud et al, 2017). Failure to maintain processive replication through the +3.5 G4 motif leads to uncoupling of DNA unwinding and DNA synthesis, interrupting normal histone recycling at the fork and the accurate propagation of epigenetic information carried by post‐translational modifications on histone proteins (Fig 1A). This leads to replication‐dependent instability of BU‐1 expression manifested as stochastic conversion of the normal “high” expression state to a lower expression level as cells divide (Sarkies et al, 2012; Schiavone et al, 2014). This expression instability can be readily monitored by flow cytometry analysis of surface Bu‐1 protein (Sarkies et al, 2012; Fig EV1), providing a simple method to cumulatively “record” episodes of interrupted DNA synthesis at the +3.5 G4.

To model the replication of (GAA)_n repeats, we started with DT40 cells in which the BU‐1 +3.5 G4 motif had been deleted in both alleles (Schiavone et al, 2014). (GAA)_n repeats of lengths between n = 10 and n = 75 were constructed either by synthesis for n ≤ 30 or, for the longer tracts, using a cloning strategy for highly repetitive sequences (Fig EV2). The repeats were then introduced into the BU‐1A allele by gene targeting, as previously described (Schiavone et al, 2014), to create BU‐1A ^(GAA)n cells. Following selection cassette removal, cells carrying (GAA)₁₀ and (GAA)₂₀ in BU‐1A exhibited wild‐type expression levels (Fig 1B). (GAA)₃₀ reduced expression of BU‐1A, while (GAA)₅₀ and (GAA)₇₅ essentially abrogated expression of the gene (Fig 1B). The reduced expression in cells carrying (GAA)_30–75 affects the entire population and thus appears to be distinct from the stochastic, replication‐dependent loss of expression we have previously reported to be induced by G4 motifs in cells lacking enzymes involved in G4 replication (Sarkies et al, 2012; Schiavone et al, 2014) or in which G4s are stabilised (Guilbaud et al, 2017). Rather, these longer repetitive tracts resulted in the accumulation of chromatin‐associated nascent RNA (ChrRNA; Nojima et al, 2016) within the locus (Fig EV3A), consistent with impaired expression being due to reduced processivity of RNAPII. As the global reduction of BU‐1 expression in (GAA)_30–75 alleles precluded the detection of stochastically generated loss variants, we focussed our subsequent analyses on (GAA)₁₀ and (GAA)₂₀.

Fluctuation analysis for the generation of Bu‐1a loss variants confirmed that the presence of (GAA)₁₀ at the +3.5 kb position did not affect the stability of BU‐1 expression in a wild‐type background (Fig 1C). However, (GAA)₂₀ induced modest, but significant, formation of Bu‐1a loss variants (Fig 1C), suggesting that this repeat is able to impede replication even in wild‐type conditions. We next examined the effect of deleting PrimPol to explore the extent to which repriming mitigates the replication impediment posed by these sequences. The results were striking: the rate at which Bu‐1a loss variants were generated in primpol cells increased significantly, both for (GAA)₁₀ and for (GAA)₂₀ (Fig 1C). These observations are consistent with repriming preventing significant uncoupling of DNA unwinding and DNA synthesis at these short repeats and thus that they form frequent impediments to otherwise unperturbed DNA replication.

The BU‐1 instability seen in primpol cells carrying (GAA)₁₀ is comparable to that observed in primpol cells harbouring the endogenous +3.5 G4 (Schiavone et al, 2016), and similar Bu‐1a^medium and Bu‐1a^low expression states, characterised by loss of H3K4me3 and additional DNA methylation respectively, were isolated (Fig EV3B–E). Genetic instability, at a level that could explain the observed formation of Bu‐1a loss variants, was not detected (Fig EV3F). The effect of the repeat was orientation dependent (Fig 1D), only producing instability when knocked in such that the purine‐rich strand formed the leading strand template for a fork entering from the 3′ end of the locus. Together, these observations suggest that the (GAA)₁₀ sequence causes epigenetic instability through the same replication‐dependent mechanism that we have previously described for G4s.

The RPA‐binding and repriming functions of PrimPol are required to ensure processive replication at the BU‐1 (GAA)₁₀ repeat

While PrimPol can perform some translesion synthesis, considerable evidence now supports its main in vivo role being repriming (Mouron et al, 2013; Keen et al, 2014; Kobayashi et al, 2016; Schiavone et al, 2016). The repriming function of PrimPol requires the C‐terminal zinc finger and RPA‐binding motif A (RBM‐A), which mediates an interaction with the single‐stranded binding protein replication protein A (Wan et al, 2013; Guilliam et al, 2015, 2017). To confirm that the primary in vivo role played by PrimPol in the replication of (GAA)₁₀ is indeed repriming, we performed a complementation study by ectopically expressing YFP‐tagged human PrimPol in primpol cells (Fig 2). Expression of full‐length human PrimPol completely restored the stability of BU‐1A expression in primpol cells carrying the (GAA)₁₀ repeat. However, neither catalytically inactive PrimPol (D114A, E116A or hPrimPol[AxA]) nor a repriming‐defective Zn‐finger mutant (C419A, H426A or hPrimPol[ZfKO]) was able to prevent instability of BU‐1 expression (Fig 2). As noted previously, both these constructs confer a growth disadvantage when expressed in DT40 (Schiavone et al, 2016), meaning that cells retaining the transgene do not go through as many cell cycles in the course of the experiment resulting in a lower frequency of Bu‐1 loss variants (Schiavone et al, 2016). Recent work has demonstrated that repriming by PrimPol also requires RPA binding, which is mediated by two RPA‐binding motifs, RBM‐A and RBM‐B. While both RBM‐A and RBM‐B can bind the same basic cleft in RPA70N in vitro, RBM‐A appears to play a dominant role in vivo (Guilliam et al, 2017). Consistent with this observation, expression of hPrimPol[∆RBM‐B] was much more effective at suppressing BU‐1 instability than hPrimPol[∆RBM‐A].

REV1, a Y‐family DNA polymerase, is required for maintaining stability of BU‐1 expression when the sequence at the +3.5 kb position is a G4 motif (Sarkies et al, 2012; Schiavone et al, 2014), which reflects a direct role for REV1 in G4 replication (Sarkies et al, 2010; Eddy et al, 2014). Replacing the +3.5 G4 in rev1 cells with (GAA)₁₀ repeats did not result in significant destabilisation of BU‐1 expression (Appendix Fig S1), while (GAA)₂₀ results in a modest destabilisation of BU‐1 expression, as in wild‐type cells. This suggests that the role REV1 plays in maintaining epigenetic stability of BU‐1 is specific to G4 motifs in contrast to PrimPol, the ability of which to reprime is required at both types of secondary structure.

PrimPol limits R‐loop formation around a (GAA)₁₀ repeat

The orientation dependence of the GAA tract with respect to BU‐1 instability is in line with the predicted formation of triplex DNA when a polypurine tract is transcribed as the coding strand, but not as the template strand (Grabczyk & Fishman, 1995; Grabczyk et al, 2007). In vitro studies have shown that the formation of triplexes at (GAA)_n repeats occurs concurrently with the formation of a stable DNA:RNA hybrid between the TTC‐rich template strand and the nascent GAA‐containing RNA strand (Grabczyk et al, 2007). Furthermore, pathological formation of R‐loops has been reported at long (GAA)_n repeats (n ≥ 650) in immortalised lymphoblasts derived from Friedreich's ataxia patients (Groh et al, 2014). These reports, together with the results presented thus far, prompted us to investigate whether R‐loops contribute to replication stalling induced by a short GAA tract in BU‐1.

R‐loops can be detected using a DNA:RNA hybrid‐specific antibody, S9.6 (Boguslawski et al, 1986). We first examined R‐loop formation in the (GAA)₁₀‐containing BU‐1 locus of wild‐type cells using DNA:RNA immunoprecipitation (DRIP) followed by quantitative PCR (Fig 3A). In the body of BU‐1 in wild type, the presence of (GAA)₁₀ at the +3.5 kb position correlates with a very modest DRIP signal in the vicinity of the repeat. In contrast, primpol cells exhibit a highly significant increase in R‐loop signal around the repeat. This signal is reduced by treatment of the extracted nucleic acids with RNase H, but not RNase III (Appendix Fig S2), supporting that the detected signal corresponds to RNA:DNA hybrids, rather than dsRNA, which has been reported to cross‐react with the S9.6 antibody and to confound R‐loop analysis (Phillips et al, 2013; Hartono et al, 2018). Somewhat surprisingly, the R‐loop signal is detected on both sides of the (GAA)₁₀ (Fig 3A). While this may, in part, reflect the resolution of the DRIP‐qPCR method, it is also consistent with the repeat promoting so‐called sticky behaviour, the accumulation of R‐loops across the gene body observed in about a quarter of human loci (Sanz et al, 2016). Indeed, analysis of RNA DIP‐seq data covering BU‐1 in wild‐type cells reveals a constitutive coding strand R‐loop signal across the locus (Appendix Fig S3). Crucially, the increased gene body R‐loop signal in primpol cells is abrogated when the +3.5 (GAA)₁₀ repeat is deleted (Fig 3A). Together, these data show that PrimPol suppresses R‐loop formation associated with this sequence element, rather than playing a more general role in controlling R‐loop formation during transcription.

A further striking feature to note in Fig 3A is the strong S9.6 DRIP signal at +11.5 kb, which is in the vicinity of the transcription termination site (Fig 3A). While the presence of this signal is consistent with the previously described formation of R‐loops as part of the mechanism of transcription termination in a subset of genes (Skourti‐Stathaki et al, 2014), it is noteworthy that the signal is increased significantly in the primpol mutant. This may be explained by the fact that the region harbours a number of sequences with significant secondary structure‐forming potential (Appendix Fig S4).

R‐loop formation is required for (GAA)₁₀ to induce expression instability of BU‐1 in PrimPol‐deficient cells

We next asked whether formation of a replication block at (GAA)₁₀ requires an R‐loop. To address this, we overexpressed YFP‐tagged chicken RNase H1 carrying a disrupted mitochondrial localisation sequence. RNase H1 degrades R‐loops (Stein & Hausen, 1969), and we have previously shown this protein to be stably expressed and active in DT40 cells (Romanello et al, 2016). This RNase H1 construct was stably expressed in primpol BU‐1A ^(GAA)10 cells (Appendix Fig S5). This reduced the R‐loop signal in the vicinity of the repeat (+3 kb), as did complementation with human PrimPol (Fig 3B). Strikingly, RNase H1 overexpression completely prevented the formation of Bu‐1a loss variants in three separate clones of primpol BU‐1A ^(GAA)10, an effect comparable to removing the (GAA)₁₀ repeat itself (Fig 3C). This suggests that DNA:RNA hybrid formation makes a crucial contribution to the ability of (GAA)₁₀ to act as a replication impediment and to induce BU‐1 expression instability.

R‐loop stabilisation converts the (GAA)₁₀ sequence into a replication impediment

This R‐loop dependence of BU‐1 expression instability in primpol mutants led us to predict that enforced stabilisation of R‐loops might lead the (GAA)₁₀ repeat to induce BU‐1 expression instability even in wildtype cells. To achieve this, we overexpressed the 52 amino acid DNA:RNA hybrid binding domain (HBD) of human RNase H1, a fragment previously shown to co‐localise with and stabilise DNA:RNA hybrids in vivo (Bhatia et al, 2014), fused in frame with mCherry separated by a flexible GSGSG linker (Fig 3D). The resulting fusion protein could be stably expressed in DT40 cells as monitored by mCherry fluorescence and Western blotting (Fig 3D and Appendix Fig S6). Expression of the HBD in cells lacking a structure‐forming sequence at the +3.5 kb position of BU‐1A (DT40 BU‐1A ^∆G4) did not induce statistically significant destabilisation of BU‐1 expression compared with the control (Fig 3E). However, when the (GAA)₁₀ repeat was present at the +3.5 kb position, we observed significantly greater expression instability. This observation provides further evidence that R‐loops are causal in promoting a (GAA)₁₀ motif to become a replication block.

PrimPol curtails R‐loop formation during S phase

Since the activity of PrimPol is intimately linked with replication, we hypothesised that specifically removing the R‐loops in S phase would suppress (GAA)₁₀‐induced BU‐1 expression instability. We therefore expressed YFP‐tagged chicken RNase H1 fused to a degron sequence from geminin, which ensures protein expression is restricted to S phase (Sakaue‐Sawano et al, 2008; Fig 4A). Expression of this construct was able to prevent instability of BU‐1 expression in primpol BU‐1A ^(GAA)10 (Fig 4B), confirming that R‐loops present during S phase are indeed responsible for the (GAA)₁₀‐dependent destabilisation of BU‐1.

Next, we asked whether the accumulation of R‐loops at the BU‐1 locus of primpol cells (Fig 3A) does indeed occur during S phase. Wild‐type BU‐1A ^(GAA)10 and primpol BU‐1A ^(GAA)10 cells were synchronised in G1 by double thymidine block and released with samples taken over the ensuing 6 h (Fig EV4) for monitoring steady state DNA:RNA hybrids by DRIP‐qPCR (Fig 4C). This analysis showed the level of DNA:RNA hybrids in the gene body in wild‐type cells to be essentially stable through S phase. In contrast, gene body DNA:RNA hybrids increase significantly in primpol cells, peaking an hour into S phase. This corresponds to the estimated time that the locus is replicated (Schiavone et al, 2014). Striking also is the increase in DNA:RNA hybrid signal at the 3′UTR of the primpol mutant between 0.5 and 2 h. As noted above, this may be due to a group of structure‐forming DNA sequences in the vicinity of the +11.5 kb position. This does not appear to be a general feature of 3′UTRs as a selection of genes that do not contain identifiable structure‐forming sequences in their 3′UTR do not exhibit this behaviour (Appendix Fig S7).

The observed increase in the R‐loop levels is unlikely to be related to a direct activity of PrimPol on R‐loop dissolution as PrimPol lacks the key nucleolytic or helicase activities found in proteins known to directly disrupt R‐loops. As the half‐life of R‐loops in vertebrate cells has been estimated to be around 10 min (Sanz et al, 2016), we instead hypothesised that the increase in steady state DNA:RNA hybrids observed in primpol cells in S phase was likely due to increased synthesis around the time the locus is replicated. To test this, we used metabolic labelling of nascent RNA with 4‐thiouridine (4‐SU) to examine active DNA:RNA hybrid formation through S phase (Fig 4D). DNA containing R‐loops was gently extracted, and free cytoplasmic and nuclear RNA was degraded with RNase I leaving behind the RNA moiety of R‐loops. After heat denaturation and DNase I treatment, surviving 4‐SU‐labelled RNA was biotinylated and captured with streptavidin beads and its abundance analysed by RT–qPCR (Fig 4D). We observed higher levels of captured nascent DNA:RNA hybrids in the BU‐1 locus of an asynchronous primpol culture compared to wild type (Fig 4E), with the increase consistent with higher levels of R‐loops detected with S9.6 (Fig 3A). Importantly, treatment with RNase H completely abrogated the signal across the BU‐1 locus. This method to analyse nascent DNA:RNA hybrids also provides support for our central hypothesis free of the potential concerns surrounding the specificity of the S9.6 antibody (Vanoosthuyse, 2018).

Applying this technique to thymidine synchronised cultures (Figs 4F and EV4) revealed a striking spike in R‐loop synthesis in the vicinity of the (GAA)₁₀ repeat in primpol cells an hour after release from G1 arrest (Fig 4G). As noted above, this corresponds closely with the estimated time the locus is replicated (Schiavone et al, 2014). Together, these observations are consistent with a model in which excessive R‐loop formation during S phase in the absence of PrimPol results from failure to restrict the exposure of single‐stranded DNA gaps produced during interruptions of DNA synthesis.

PrimPol suppresses R‐loop formation in the vicinity of secondary structure‐forming sequences throughout the genome

To explore whether our observations at BU‐1 could be extended to the whole genome, we performed quantitative high‐throughput sequencing of S9.6 immunoprecipitated DNA (DRIP‐seq) from wild‐type and primpol DT40 cells. DRIP was performed following treatment of the isolated nucleic acid with RNase A, and the specificity of the DRIP‐seq signal was confirmed by pre‐treatment of the precipitated nucleic acids with RNase H. To allow quantitation of the DRIP signal, the DT40 samples were spiked with a fixed proportion of Drosophila S2 cells to provide an internal control (Orlando et al, 2014). Inspection of the distribution of spike‐normalised reads, shown across two representative genes in Fig 5A, revealed a strong correlation between wild‐type and primpol cells, but with higher numbers of reads in peaks in primpol. Following peak calling, we examined the global distribution of R‐loops across the DT40 genome and found it to be enriched in promoters and terminal regions of genes (Fig 5B), in agreement with previous experiments in human cells (Ginno et al, 2013; Skourti‐Stathaki et al, 2014; Sanz et al, 2016).

Peak heights in the wild‐type and primpol samples were normalised to the mean number of Drosophila reads. This revealed a highly significant increase in the height of the DRIP peaks in primpol cells (Fig 5C). Between wild type and primpol, 84% of peaks were shared suggesting that the loss of PrimPol does not result in the appearance of new peaks, but for any given peak there is a greater DRIP signal, suggesting a higher steady state level of R‐loops (Fig 5D). 41% of DRIP peaks overlapped with genes, and 83% of genes with DRIP peaks were shared between wild type and primpol. The degree of overlap in the two conditions is not due to the observed change in peak width as the correlation is still observed when allowing 1 kb separation, a much greater distance than the peak width increase. (Appendix Figs S8 and S9). We next asked whether genes with DRIP peaks are enriched for H‐DNA motifs. To identify potential H‐DNA motifs, we employed the “Triplex” R package (Hon et al, 2013), which adopts an approach that allows the identification of sequences with H‐DNA‐forming potential despite the presence of small imperfections in the sequence as it uses a scoring system based on models of the structures of triplex DNA. The distribution of H‐DNAs identified in the chicken genome with this algorithm is shown in Appendix Fig S10. H‐DNA sequences were identified as overlapping just under 8% of all genes (see Materials and Methods for further details for determining overlaps). The subset of genes harbouring DRIP peaks was significantly enriched for these sequences, with c. 15% of these genes overlapping sequences with H‐DNA‐forming potential (Fig 5E). Within this set of genes, there was a significant increase in peak height in the primpol mutant (Fig 5F). A similar degree of overlap is seen in extragenic DRIP peaks, with 11% of non‐genic peaks falling within 1 kb of an H‐DNA motif in wild‐type cells, and 12% in primpol.

Our previous work has shown that G4s are able to induce similar epigenetic instability to the (GAA)₁₀ repeat that has been the focus of this study. Further, G4 motifs have been linked to R‐loop formation (Duquette et al, 2004). We therefore used the regular expression [G_3‐5N_1‐7G_3‐5N_1‐7G_3‐5N_1‐7G_3‐5] (Huppert & Balasubramanian, 2005) to identify a core subset of G4 motifs in the chicken genome (Appendix Fig S9). G4 motifs identified with this approach overlap with 59% of all genes, while 76% of genes with DRIP peaks overlapped motifs, a significant enrichment (Fig 5G), and again, the heights of the peaks in the primpol data set were significantly increased (Fig 5H).

The striking increase in steady state R‐loop accumulation in genes containing G4 motifs and our previous demonstration that G4 motifs also potently destabilise BU‐1 expression in primpol cells (Schiavone et al, 2016) prompted us to ask whether R‐loops also promote G4 motifs to become replication impediments. Forcing RNase H1 expression resulted in a highly significant reduction in BU‐1 instability induced by the natural +3.5 G4 in four separate RNase H1‐expressing clones (Fig EV5), demonstrating that R‐loop formation increases the probability of this G4 forming a significant replication impediment, but that it is not essential for it to do so.

Finally, we asked whether loss of PrimPol also affected R‐loop levels in human cells. PrimPol was disrupted using CRISPR/Cas9 editing in the induced pluripotent stem cell line BOBSC (Yusa et al, 2011). Genome‐wide R‐loops were isolated with S9.6 immunoprecipitation followed by a modified sequencing protocol, RNA DIP‐seq, that monitors the RNA moiety within the DNA:RNA hybrids. An example of the signal, normalised to a spike‐in control of DT40 cells, across the SKI locus in wild type and primpol BOBSC (Fig 6A) shows the same pattern of R‐loop enrichment observed in a previous study (Sanz et al, 2016). It also demonstrates the overlap of sites of R‐loop formation between wild type and primpol and the increase in peak heights in primpol. Following peak calling, we determined that the overall distribution of R‐loops across a metagene is similar to that in DT40 (Fig 5B) and that described previously for human cells, with an enrichment at promoter and terminus regions of genes (Fig 6B; Sanz et al, 2016). The increase in the height of existing peaks in primpol cells, evident in the SKI locus (Fig 6A), is confirmed by genome‐wide analysis. 66% of wild‐type and primpol peaks overlap with a highly significant increase in peak height in the primpol mutant (Fig 6C and D). Again, this correlation is independent of the observed increase in peak width (Appendix Fig S8). As in DT40, the genes containing peaks were associated with H‐DNA and G4 motifs (Fig 6E and F), the length distribution of which is comparable between the two species (Appendix Fig S10). In both cases, the mean peak height was significantly higher in primpol cells (Figs 6G and H), demonstrating that loss of PrimPol also results in increased R‐loop formation in the vicinity of DNA secondary structures in human cells.

A requirement for PrimPol reveals that (GAA)₁₀ forms a replication impediment

The (GAA)₁₀ repeat upon which this study has focussed is typical of widespread short tandem repeats found throughout vertebrate genomes (Willems et al, 2014). Repeats of this length have not been previously linked to detectable disturbances in replication or transcription in vitro (Bidichandani et al, 1998; Ohshima et al, 1998) despite their potential to form triplex structures at physiological pH (Potaman et al, 2004). Our previous work supports a model in which instability of BU‐1 expression induced by G4s results from uncoupling of DNA unwinding from leading strand DNA synthesis (Sarkies et al, 2010; Schiavone et al, 2014; Šviković & Sale, 2017). This uncoupling can extend up to c. 4.5 kb (Schiavone et al, 2014), consistent with earlier observations in both mammalian and yeast cells (Lehmann, 1972; Lopes et al, 2006), and is mitigated by PrimPol‐dependent repriming (Schiavone et al, 2016). We now show that this repriming activity is also frequently deployed at a model short tandem repeat, of a form found commonly in vertebrate genomes, demonstrating that these sequences can indeed form replication impediments.

The nature of the replication impediment formed by (GAA)₁₀

(GAA)_n repeats, in common with other polypurine–polypyrimidine tracts, are capable of forming triplex secondary structures in which a third strand anneals through Hoogsteen base pairing. This tendency has been linked to the detrimental effect of long GAA repeats on transcription elongation (Bidichandani et al, 1998; Punga & Buhler, 2010) through the trapping of transcribing RNA polymerase II (Grabczyk & Fishman, 1995). However, the very act of transcription also promotes formation of secondary structures (Lilley, 1980; Kouzine et al, 2017), including triplexes (Grabczyk & Fishman, 1995; Kouzine et al, 2004). This is likely driven by structure formation releasing the negative supercoiling generated in the wake of translocating RNA polymerase (Liu & Wang, 1987; Levens et al, 2016). Similar topology‐induced structure formation could also contribute to leading strand secondary structure formation behind the replicative helicase (reviewed in Kurth et al, 2013; Yu & Droge, 2014).

We show that (GAA)₁₀ requires an RNase H1‐sensitive R‐loop in order to create a replication impediment that requires PrimPol‐dependent repriming. It is well established that formation of DNA:RNA hybrids coincides with the sequences able to adopt non‐B DNA structures (Duquette et al, 2004; Grabczyk et al, 2007) and is favoured in a negatively supercoiled DNA template (Roy et al, 2010). R‐loops have been implicated as a major factor in the severity of head‐on collisions between the replication and the transcriptional machinery (Hamperl et al, 2017). However, a direct head‐on collision with transcribing RNA polymerase is likely to halt the entire replisome (Pomerantz & O'Donnell, 2010), precluding the displacement of parental nucleosomes caused by the uncoupling between the replicative helicase and DNA synthesis. It is difficult to reconcile this type of stall with the involvement of PrimPol. Specifically, the DNA‐ and RPA‐binding activities of the enzyme suggest the transient formation of ssDNA, which most likely arises as a result of uncoupling of replicative helicase from the replicative polymerases, and which is the basis for BU‐1 expression instability.

How then can a (GAA)_n repeat generate the uncoupling of DNA unwinding and leading strand DNA synthesis necessary to induce expression instability of BU‐1? We propose that transcription of the (GAA)₁₀ repeat generates an R‐loop (Fig 7). During replication, the approaching replicative helicase traverses the transcription complex by displacing the RNA polymerase (Pomerantz & O'Donnell, 2010) or by reorganising the helicase itself (Huang et al, 2013; Vijayraghavan et al, 2016). Biophysical calculations show that DNA:RNA hybrids are sufficiently thermodynamically stable to survive the accumulation of positive supercoiling generated ahead of the replicative helicase (Belotserkovskii et al, 2013). Since the eukaryotic replicative helicase tracks on the leading strand (Douglas et al, 2018), we suggest that the DNA:RNA hybrid could remain intact on the lagging strand during passage of the helicase. Behind the replicative helicase, the persistent lagging strand DNA:RNA hybrid may re‐trap the purine‐rich leading strand through triplex formation. The resulting R:R•Y hybrid triplex, recently termed an H‐loop (Neil et al, 2018), could then block leading strand synthesis (Samadashwily & Mirkin, 1994). This model is consistent with the observation that the depletion of DNA:RNA hybrids through overexpression of RNase H1 completely abolishes (GAA)₁₀‐dependent BU‐1 expression instability in PrimPol‐deficient cells. An alternative explanation for the creation of a leading strand impediment is the formation of a DNA triplex stabilised by an adjacent DNA:RNA hybrid, of the form proposed by Grabczyk and Fishman (1995). In either event, continued helicase activity would result in exposure of ssDNA ahead of the stalled replicative polymerase, which through being bound by RPA promotes the recruitment of PrimPol. Repriming close to the structure then allows DNA synthesis to remain coupled to unwinding leaving the triplex in a small gap to be disassembled post‐replicatively (Fig 7). Whether loss of PrimPol completely disables leading strand repriming or whether PRIM1 (the primase associated with Pol α) can substitute to some extent remains unclear. Recent work with a reconstituted yeast replisome suggests that leading strand repriming by PRIM1 is intrinsically inefficient (Taylor & Yeeles, 2018), suggesting that the ssDNA generated by the helicase–polymerase uncoupling event will ultimately be replicated either by restart of the stalled fork or by a fork arriving from the opposite direction.

Loss of PrimPol‐mediated repriming at structured DNA promotes S phase R‐loop accumulation

In the absence of PrimPol, continued unwinding of the parental duplex by the replicative helicase in the context of a continued stalling of DNA synthesis would create a more extensive region of ssDNA. That this results in increased R‐loop formation in primpol cells during the time the locus is replicated implies that RNAPII continues to transcribe despite its template remaining single‐stranded. This idea is consistent with both biochemical (Kadesch & Chamberlin, 1982) and in vivo reports (Ohle et al, 2016; Michelini et al, 2017) demonstrating that DNA:RNA hybrid formation can occur at ssDNA generated by resection of DNA ends at double‐stranded DNA breaks. We suggest that ssDNA generated as a consequence of helicase–polymerase uncoupling, unmitigated by repriming, could also act as a substrate for unscheduled RNAPII transcription and DNA:RNA hybrid formation (Fig 7).

The results we present here establish two important mechanistic points concerning the relationship between R‐loops and impeded replication. First, we show that R‐loops are able to promote short sequences with structure‐forming potential to become replication impediments, requiring the repriming activity of PrimPol to maintain their processive replication. Second, failure to reprime at these sequences increases R‐loop formation. We suggest this is due to exposure of excessive single‐stranded DNA during S phase, potentially increasing unscheduled access of RNAPII. An increase in R‐loops associated with structure‐forming DNA sequences is seen throughout the genome in PrimPol‐deficient cells, but principally in regions in which R‐loops are already formed in wild‐type cells, particularly in transcribed regions. This suggests that repriming plays a particularly important role in allowing cells to manage the complex challenges created by clashes between transcription and replication.

Cell culture and transfection

DT40 cell culture, cell survival assays, fluctuation analysis for generation of Bu‐1 loss variants (Fig EV1), and genetic manipulation of the BU‐1 locus were performed as previously described (Simpson & Sale, 2003, 2006; Schiavone et al, 2014). Drosophila S2 cells were grown in Insect‐XPRESS Protein‐Free Insect Cell Medium with l‐glutamine (Lonza), supplemented with 1% penicillin–streptomycin at 27°C, ambient CO₂ with 105 rpm agitation. BOBSC human induced pluripotent stem (hiPS) cells (Yusa et al, 2011) were cultured feeder‐free on dishes coated with Vitronectin XF (07180; Stem Cell Technologies) in Essential 8 Flex media (A2858501; Thermo Fisher Scientific) at 37°C and 5% CO₂. Cells were split 1:10–1:15 every 3–4 days depending on confluence. All cells tested negative for mycoplasma.

CRISPR/Cas9‐mediated gene disruption in human cells

Guide RNA sequences used for disrupting PRIMPOL in BOBSC iPS cell lines are listed in the Appendix. Each gRNA sequence was cloned into pX458 (Ran et al, 2013). A targeting construct carrying puromycin selection marker was constructed by Gibson assembly using PCR‐amplified 5′ and 3′ homology arms (see Appendix Table for all oligonucleotides used in this study). Equimolar amounts of targeting construct, gRNA expression vectors and Cas9 expression vectors were delivered by Amaxa electroporation. Puromycin‐resistant clones were genotyped by PCR and Sanger sequencing (Shen et al, 2014).

Molecular cloning and transgene constructs

For transgene expression, cDNA was cloned in frame with fluorescent protein (mCherry or YFP) in the polylinker of pXPSN2 (Ross et al, 2005). The expression module was released with SpeI digestion and subcloned into pBluescript‐based vectors containing a loxP flanked puromycin or blasticidin S selection cassettes (Arakawa et al, 2001), which were transfected into DT40 via electroporation. Primers used for molecular cloning are listed in the Appendix. The hPrimPol‐YFP, hPrimPol [AxA]‐YFP and hPrimPol[ZfKO]‐YFP constructs for complementation of primpol DT40 were previously described (Schiavone et al, 2016). cDNAs for PrimPol RPA binding mutants (∆RBM‐A and ∆RBM‐B) were PCR‐amplified from previously described vectors (Guilliam et al, 2017) with primers listed in the Appendix. Similarly, chicken RNase H1 (lacking the mitochondrial localisation sequence) was PCR‐amplified from DT40 cDNA and cloned in frame with YFP on the C‐terminus. To produce a cell cycle‐regulatable RNase H1, a fragment of human geminin corresponding to amino acids 1–110 was PCR‐amplified from pLL3.7m‐Clover‐Geminin(1‐110)‐IRES‐mKO2‐Cdt(30‐120) from the Fucci4 system (Bajar et al, 2016) and cloned in frame to the C‐terminus of chicken RNase H1 ΔMLS‐YFP. The hybrid binding domain was amplified from human cDNA using primers previously published (Bhatia et al, 2014) and fused to C‐terminal mCherry via flexible GSGSG linker. Uninterrupted GAA tracts were created using a previously developed strategy (Scior et al, 2011); see Fig EV2 for details. The acceptor plasmid was created by inserting a linker containing a (GAA)₂₀ repetitive tract flanked by recognition sites for the restriction enzymes BbsI, BsmBI and NcoI into pBluescript SK(+). Plasmids containing the repeats were transformed into a strain of Escherichia coli lacking the SbcCD nuclease, DL733 (Connelly et al, 1997), a kind gift from David Leach, to avoid excision of secondary structures caused by the repeat. The confirmed uninterrupted repeats with minimal flanking sequence were released by MluI digest, subcloned into the BU‐1 targeting construct and screened for orientation by Sanger sequencing.

Chromatin immunoprecipitation

Chromatin immunoprecipitation (ChIP) was performed as previously described (Schiavone et al, 2014). Solubilised chromatin was diluted and immunoprecipitated overnight using the following antibodies: histone H3 (ab1791; Abcam), H3K4me3 (9727; Cell Signaling), H3K36me3 (ab9050; Abcam) and normal rabbit IgG (2729; Cell Signaling). Primer sequences are provided in the Appendix. Enrichment was normalised to H3 and reported relative to wild type.

Bisulphite sequencing

Bisulphite conversion of genomic DNA was performed using EZ DNA Methylation‐Gold Kit (Zymo Research) as per the manufacturer's instructions. PCR was performed with ZymoTaq (Zymo Research) and primers compatible with bisulphite‐converted DNA (Appendix) for 40 cycles according to the manufacturer's instructions. PCR products were purified, digested with NotI and SacI, size‐selected by gel extraction and cloned into pBK CMV. Plasmids from individual bacterial colonies were sequenced by GATC Biotech.

Chromatin‐associated RNA (ChrRNA) extraction

RNA associated with the chromatin was extracted as described previously (Nojima et al, 2016). Briefly, DT40 cells were lysed in HLB + N [10 mM Tris–HCl (pH 7.5), 10 mM NaCl, 2.5 mM MgCl₂ and 0.5% (vol/vol) NP‐40] and passed through a 10% sucrose cushion. Nuclei were then resuspended in 125 μl NUN1 buffer [20 mM Tris–HCl (pH 7.9), 75 mM NaCl, 0.5 mM EDTA and 50% (vol/vol) glycerol], mixed with 1.2 ml NUN2 [20 mM HEPES–KOH (pH 7.6), 300 mM NaCl, 0.2 mM EDTA, 7.5 mM MgCl₂, 1% (vol/vol) NP‐40 and 1 M urea], vortexed vigorously and spun for 10 min at 16,000 g. Chromatin pellets were digested twice with DNase I (NEB, M0303) and Proteinase K. ChrRNA was extracted with QIAzol (QIAGEN) and converted to cDNA using QuantiTect Reverse Transcription Kit (QIAGEN) as recommended by the manufacturer. The enrichment of RNA across the BU‐1 locus was analysed by qPCR and the signal normalised to GAPDH.

DNA:RNA immunoprecipitation (DRIP)

Extraction of R‐loops is largely based on methods previously described (Groh et al, 2014). Ten to 30 million DT40 cells were harvested by centrifugation, washed in 25 ml cold PBS and lysed in cell lysis buffer (85 mM KCl, 5 mM PIPES (pH 8.0), 0.5% NP‐40) for 10 min on ice. Nuclei were gently pelleted (1,000 g, 10 min), equilibrated in nuclei lysis buffer (50 mM Tris–HCl (pH 8.0), 1.2 mM EDTA, 1% SDS) and then incubated overnight at 37°C upon addition of Proteinase K (Thermo Fisher). SDS and contaminating proteins were removed by adding 5 M KOAc (pH 5.5) and centrifuging at high speed for 15 min. DNA was precipitated from the supernatant with glycogen (Santa Cruz Biotechnology) and isopropanol. DNA was pelleted, gently washed several times with 70% EtOH and rehydrated in 10 mM Tris–HCl (pH 8.0).

Genomic DNA containing DNA:RNA hybrids was digested overnight with a restriction enzyme cocktail containing BamHI, NcoI, PvuII, ApaLI and NheI, yielding an average fragment size of 1 kb. Samples were subsequently diluted to 5 ml with IP dilution buffer [16.7 mM Tris–HCl (pH 8.0), 1.2 mM EDTA, 167 mM NaCl, 1.1% Triton X‐100, 0.01% SDS], pre‐cleared for 2 h with 30 μl Protein G Sepharose beads (Dharmacon) and immunoprecipitated with 10 μg S9.6 antibody overnight at 4°C. Subsequent steps are essentially the same as for ChIP. Briefly, captured immunocomplexes were washed with low‐salt [0.1% SDS, 1% Triton X‐100, 20 mM Tris–HCl (pH 7.5), 165 mM NaCl, 2 mM EDTA], high‐salt [0.1% SDS, 1% Triton X‐100, 20 mM Tris–HCl (pH 7.5), 500 mM NaCl, 2 mM EDTA] and LiCl [1% NP‐40, 1% deoxycholate, 10 mM Tris–HCl (pH 7.5), 250 mM LiCl, 1 mM EDTA] wash buffers and TE buffer [10 mM Tris–HCl (pH 7.4), 1 mM EDTA]. DNA:RNA hybrids were eluted for 2 h at 65°C in elution buffer (1% SDS, 100 mM NaHCO₃) and purified with PCR Purification Kit (QIAGEN). The specificity of the pull‐down was tested with RNase H and RNase III treatments prior to immunoprecipitation: one‐third of the digested material was treated with 25 U of RNase H (NEB, M0297), or with 10 U of RNase III (Ambion, AM2290) in appropriate buffers overnight at 37°C, with the subsequent steps performed as described above. The signal across BU‐1 locus was normalised to RNase H background signal and baselined to 28S rDNA.

Cell cycle synchronisation and 2D cell cycle analysis

G1 phase synchronisation of DT40 cells was achieved by double thymidine block. Cells were treated overnight with 2 mM thymidine, released for 9 h and again treated with thymidine overnight, after which cells were released into medium containing 0.2 μM nocodazole to prevent cells entering mitosis. Upon release from thymidine block, cells in different cell cycle phases were harvested to be analysed or pulse‐labelled with BrdU. Five to 10 million DT40 cells were pulse‐labelled with 50 μM BrdU for 30 min in complete medium at 37°C. BrdU staining was performed as previously described (Frey et al, 2014).

Capture of 4‐SU‐labelled nascent DNA:RNA hybrids

To label nascently formed DNA:RNA hybrids, 150–250 million DT40 cells were resuspended in 10 ml of warm complete medium supplemented with 100 μM 4‐thiouridine (4‐SU; Sigma‐Aldrich) and incubated for 30 min at 37°C. 4‐SU incorporation was terminated by adding ice‐cold PBS, after which nuclei were extracted as described for ChrRNA extraction. The nuclear pellet was divided, and equivalent of 20–50 million nuclei were lysed in 700 μl of nuclear lysis buffer (25 mM Tris–HCl (pH 7.4), 1% SDS, 5 mM EDTA and 0.125 mg/ml Proteinase K) overnight at 37°C with agitation. SDS and digested proteins were removed with 1 M potassium acetate (pH 5.5) and nucleic acids precipitated with isopropanol. Any soluble ssRNA was degraded by treating the nucleic acids with 25 U of RNase I (Ambion) in TNE buffer (10 mM Tris–HCl (pH 7.4), 100 mM NaCl, 1 mM EDTA) for 30 min at 37°C; to remove RNase I (Ambion, AM2294), 5 μg of Proteinase K was added and reaction incubated for a further 2 h, after which the nucleic acids were purified with phenol:chloroform:isoamyl alcohol (25:24:1) and precipitated with 0.3 M NaCl. Nucleic acids were resuspended in water and fragmented using Bioruptor Plus (Diagenode) to average size of 500 bp (30 cycles, 30″ ON, 30″ OFF at high output). Appropriate specificity controls were performed at this point identically to the DRIP protocol.

Sheared nucleic acids were supplemented with 1× DNase I Reaction Buffer (New England Biolabs), denatured for 5 min at 95°C and snap‐cooled on ice to release RNA moiety of DNA:RNA hybrids. To digest the DNA component, denatured nucleic acids were incubated with 15 U of RNase‐free DNase I (NEB) at 37°C. RNA moiety of DNA:RNA hybrids was extracted with QIAzol (QIAGEN) as per the manufacturer's instructions and precipitated with glycogen at −80°C. RNA was collected by centrifugation, washed with 70% ethanol and resuspended in 217 μl of RNase‐free water. 2% of recovered RNA was reserved as total input for normalisation. 4‐SU‐containing RNA was further labelled with thiol‐specific and reversible biotinylation reagent MTSEA biotin‐XX: the remainder of RNA was mixed with 25 μl of 10× biotinylation buffer (100 mM HEPES (pH 7.4), 10 mM EDTA) and 12.5 μl MTSEA biotin‐XX (1 mg/ml in N,N‐dimethylformamide) and incubated for 30 min at room temperature in the dark with gentle rotation (Duffy et al, 2015). Following completion of the labelling reaction, free biotin was removed by chloroform:isoamyl alcohol (24:1) extraction. Biotinylated RNA was captured with 60 μl of Dynabeads MyOne Streptavidin C1 (Invitrogen, 65001) according to the manufacturer's instructions (including steps required for RNA application). To remove any unbound nucleic acids, streptavidin:biotinylated RNA complexes were washed twice with 1× B&W buffer (5 mM Tris–HCl (pH 7.4), 0.5 mM EDTA, 1 M NaCl). RNA was released by cleaving the disulphide bond previously formed between 4‐SU and MTSEA biotin‐XX with 100 mM DTT at room temperature. Eluted RNA was precipitated with glycogen, 0.3 M NaCl and isopropanol overnight at −20°C, followed by high‐speed centrifugation and 70% ethanol wash. RNA was resuspended in RNase‐free water, converted to cDNA using QuantiTect RT kit (QIAGEN) and analysed with qPCR as previously described.

DRIP‐seq

Sample preparation for DRIP‐seq was essentially performed as described above, but with some minor changes. All the samples were spiked (Orlando et al, 2014) with the same batch of Drosophila S2 cells in 1:4.2 ratio to DT40 cells. Digested DNA was treated with 100 μg RNase A in the presence of 0.5 M NaCl for 2 h at 37°C. Elution from magnetic beads was performed for 1 h at 37°C in 300 μl elution buffer supplemented with 0.1 mg/ml RNase A. To ensure complete elution, 10 μg Proteinase K was added and incubated for a further 90 min at 37°C. DNA was purified by phenol:chloroform:isoamyl alcohol extraction, quantified with Qubit dsDNA HS Assay Kit (Invitrogen), diluted with ultra‐pure water to 55 μl and sheared with Covaris M220 Focused‐ultrasonicator and Holder XTU to average size of 300 bp in microTUBE‐50 AFA. DNA libraries were built using NEBNext Ultra II DNA Library Prep Kit (New England Biolabs) as per the manufacturer's instructions.

RNA DIP‐seq

Between 60 and 100 million human cells were spiked in with DT40 cells (1/10 ratio), harvested and washed with cold PBS and nuclei isolated by lysing cells in HLB + N [10 mM Tris–HCl (pH 7.5), 10 mM NaCl, 2.5 mM MgCl₂ and 0.5% (vol/vol) NP‐40] and passing it through a 10% sucrose cushion. Collected nuclei were lysed overnight in NLB [25 mM Tris–HCl (pH 7.4), 1% SDS, 5 mM EDTA, 0.125 mg/ml Proteinase K] with agitation at 37°C. Nucleic acids were purified with 1 M potassium acetate (pH 5.5), precipitated and treated with 1 U of RNase I (Ambion, AM2294) per 90 μg of DNA (15 min at 37°C) to degrade soluble RNA. DNA was purified with phenol:chloroform, diluted with the IP dilution buffer and sheared with a Bioruptor Plus (Diagenode) to average size of 300 bp. DNA:RNA hybrids were immunoprecipitated with S9.6 mAb (1 μg antibody for each 2 μg of DNA) overnight. Immunocomplexes were captured with Protein G beads and washed as for ChIP and DRIP preparation. DNA:RNA hybrids were eluted by incubating the sample with Proteinase K for 2 h at 42°C. Nucleic acids were cleaned up with phenol:chloroform:isoamyl alcohol, precipitated with glycogen and resuspended in water, denatured for 5 min at 90°C and immediately placed on ice. DNA moiety of DNA:RNA hybrids was removed with 4 U DNase I for 30 min at 37°C. RNA was extracted with QIAzol, precipitated overnight and dissolved in RNase‐free water. Strand‐specific Illumina‐compatible libraries were prepared with NEBNext Ultra II Directional RNA Library Prep Kit (NEB, E7760) with 100 ng input. Libraries were quality‐checked as before and sequenced on a NextSeq 500 (Illumina).

Quantification, display and statistical analysis of deep sequencing data

DRIP‐seq libraries were sequenced on an Illumina HiSeq 4000, and RNA‐DIP libraries were sequenced on an Illumina NextSeq. Reads were trimmed and quality‐filtered using Trim Galore (version 0.4.4; https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), and then aligned to genomes with bowtie2 (version 2.26; Langmead & Salzberg, 2012) using default settings. DT40 reads were aligned to Ggal 5.0 and Dmel r6.18, and BOBSC reads were aligned to GRCh38 and Ggal 5.0. Alignments were filtered for uniquely matching reads and separated into sample and spike‐in. Peaks were called on filtered alignments using MACS2 (version 2.1.1.20160309) with the default settings and −g 1.87e9 (or −g hs for human)—broad (Feng et al, 2012). Peak heights were normalised to the read number of the spike‐ins and compared using the Mann–Whitney U‐test. Overlaps between peaks were calculated using bedtools2 closest (version 2.27.1) with default settings (Quinlan & Hall, 2010). Peaks were considered to be overlapping if at least 1 bp overlapped. Sequences with H‐DNA‐forming potential were identified with the Triplex R package (Hon et al, 2013). G4 motifs were identified using the Quadparser algorithm (Huppert & Balasubramanian, 2005) with the regex [G_3‐5N_1‐7G_3‐5N_1‐7G_3‐5N_1‐7G_3‐5]. The .bed files containing the positions of the identified H‐DNA‐ and G4‐forming sequences have been deposited. Enrichment testing for secondary structures was performed using the hypergeometric test.

To generate the normalised profiles presented in Figs 5A and 6A, the number of uniquely mapped reads per 100‐bp windows along Galgal5 genome was determined and normalised by the total number of uniquely mapped reads for each experiment. Values for wild type and primpol were then normalised to the relative abundance of the spiked genome (Drosophila melanogaster from S2 cells in the case of the DT40 DRIP‐seq data and Gallus gallus from DT40 cells in the case of the BOBSC RNA DIP‐seq data). For RNase H‐treated controls (in which both the experimental and spike genome signal will be reduced), the read height was further corrected to reflect the amount of nucleic acid retrieved after immunoprecipitation, which was at least eightfold less following RNase H treatment. In the case of RNA DIP‐seq, it was necessary to pool all RNase H‐treated samples to obtain sufficient material to build a sequencing library.