Polymerase δ replicates both strands after homologous recombination-dependent fork restart

Abstract

To maintain genetic stability DNA must be replicated only once and replication completed even when individual replication forks are inactivated. Because fork inactivation is common, the passive convergence of an adjacent fork is insufficient to rescue all inactive forks. Thus, eukaryotic cells have evolved homologous recombination-dependent mechanisms to restart persistent inactive forks. Completing DNA synthesis via Homologous Recombination Restarted Replication (HoRReR) ensures cell survival, but at a cost. One such cost is increased mutagenesis caused by HoRReR being more error prone than canonical replication. This increased error rate implies that the HoRReR mechanism is distinct from that of a canonical fork. Here we exploit the fission yeast Schizosaccharomyces pombe to demonstrate that a DNA sequence duplicated by HoRReR during S phase is replicated semi-conservatively, but that both the leading and lagging strands are synthesised by DNA polymerase delta.

Introduction

Accurate and complete DNA replication is prerequisite for maintaining genetic stability. Perturbations to replication underpin a range of genetic alterations, including the translocations and copy number variations typical of human genomic disorders and cancer^1-6. The completion of DNA replication is routinely challenged by a range of replication fork barriers (RFBs) that can interfere with DNA synthesis. These include multiple forms of DNA damage, non-histone protein:DNA interactions, DNA secondary structures, clashes with the transcription machinery, DNA:RNA hybrids, programmed RFBs and DNA topology^7,8. Replication also pauses stochastically and in response to the incorrect regulation of dNTP pools⁷.

Arrested forks are initially stabilized by the intra-S phase checkpoint to allow later continuation⁹. We term such forks as “paused” and describe their subsequent continuation as “resumption”. However, some arrested forks are not able to resume replication, either due to the initial nature of their arrest¹ or the stochastic failure of the intra S phase checkpoint to stabilize them. We term forks that cannot resume as “collapsed”. A collapsed fork cannot, by definition, resume replication but it can be actively “restarted”. In eukaryotic cells, the known mechanisms of fork restart require the homologous recombination (HR) machinery.

The preferred mechanism for dealing with an arrested replication fork is to stabilize it using the intra-S phase checkpoint, thus preventing inappropriate DNA transactions^10,11. This stalled fork can then either resume replication when the original problem has been resolved, or await a converging fork that will merge with it and complete replication of the locus⁹. Collapsed forks cannot resume replication, but like stalled forks that cannot bypass the original problem, they can await merger with a converging fork. To help promote such fork merging (and thus replication completion) dormant DNA replication origins are present throughout the genome and these can fire in response to local replication delays¹². However, if a fork collapses in a region with low origin density, or at a locus where replication is unidirectional (such as a telomere, within the rDNA or proximal to a second collapsed fork) a converging fork may not be available to overcome the problem. In such circumstances, restart of the replication machinery using HR provides an additional opportunity to complete DNA synthesis¹³.

Several site-specific experimental systems have been developed to characterise the mechanisms by which HR initiates DNA replication. In Saccharomyces cerevisiae the generation of a one-ended DSB outside of S phase has been used to initiate replication and thus model fork restart from a DSB¹⁴. This is defined as Recombination Dependent Replication (RDR) by break-induced replication (BIR). Experimentally, BIR is instigated outside of S phase and it takes several hours from the initial strand invasion to the onset of DNA synthesis¹⁵. This is likely because the recombination execution checkpoint (which usually promotes second-end capture) must first be overcome¹⁶. In Schizosaccharomyces pombe the RTS1 replication fork barrier (RFB) has been used to initiate HR-dependent replication. Experimentally, RTS1 generates a collapsed fork within S phase that is rapidly processed by HR proteins and initiates Homologous Recombination Restarted Replication (HoRReR) during S phase^17-19. Importantly, while it requires the HR machinery, in the RTS1 system HoRReR does not initiate through a DSB intermediate^17,20, formally distinguishing it from RDR by BIR.

While HoRReR allows the completion of S phase under circumstances where forks cannot resume and dormant origins are not available, there are costs associated with completing replication in this way. First there is an increase in non-allelic HR (NAHR) during the restart event itself^13,17 and second it has recently become clear that, once restarted, the resulting replication machine is error-prone. This has been characterized for RDR initiated by a DSB outside of S phase using BIR systems in S. cerevisiae^21-23 and for HoRReR initiated during S phase without a DSB intermediate from the RTS1 RFB using S. pombe^24,25. These increased error rates imply that RDR and HoRReR forks are non-canonical. To begin to understand the nature of a HoRReR fork, we exploited the RTS1 RFB to explore which DNA polymerases synthesize the DNA during HoRReR.

Results

RTS1 is a ~850 bp DNA sequence that acts as a polar (i.e. direction-specific) replication fork barrier in fission yeast. RTS1 contains several binding sites for a Myb-domain protein, Rtf1, which is necessary for fork arrest²⁶. In the absence of the rtf1⁺ gene, RTS1 is replicated normally: there is no intrinsic feature of its sequence that is difficult to replicate and the barrier activity is entirely dependent on Rtf1 binding. To establish the nature of HoRReR following RTS1-induced replication arrest and HR-dependent replication restart we exploited the T45R construct (Fig. 1a; see methods) where RTS1 is integrated into the genome such that a replication fork is initiated from ars3004/3005 and proceeds towards RTS1 where it is inactivated and subsequently restarted by HR (Fig. 1a). To minimize fork convergence from the distal side, we introduced multiple copies of TER2/3, an rDNA fork barrier. Unlike RTS1, TER2/3 barriers slow forks down, but do not collapse or inactivate them^25,27. The presence of the TER2/3 barriers ensured that in 75-80% of cells the fork inactivated at RTS1 restarts by HR and the 2.9 kb ura5-ura4 region is replicated by HoRReR²⁵.

Figure 1. Polδ, but not Polε participates in bulk DNA synthesis after HR-restart.

(a) Schematic representation of the construct containing RTS1 and repetitive TER2/3 (T45R). The brown and black region is always replicated by a canonical fork. The red and blue region is replicated canonically when barrier activity is OFF, while it is replicated by Homologous Recombination Restarted Replication (HoRReR) when the barrier activity is active ON. (b) Delayed timing of DNA synthesis after HR-restart. Genomic DNA was extracted at the indicated time points from G2 arrested cdc25-22 cells released synchronously into the cell cycle. Replication of the indicated loci, U5 and L3, was monitored by quantitative PCR. (c) The contribution of Polε to DNA synthesis during HoRReR. (d) Equivalent analysis of Polδ contribution to DNA synthesis during HoRReR.

First we determined the timing of HoRReR (Fig. 1b). Cells arrested by the cdc25-22 temperature sensitive mutation were synchronously released into the cell cycle and DNA duplication was followed by quantitative PCR using the indicated primers. Fork arrest was controlled¹³ by the presence (ON) or absence (OFF) of rtf1⁺. When the barrier was inactive, loci either side of RTS1 replicated together. In the presence of barrier activity, duplication of the downstream loci was delayed ~20 minutes. This is consistent with data demonstrating that recombination proteins rapidly associate with RTS1 when forks collapse¹⁹ and that the recombination structures associated with HoRReR that can be visualized by 2D-gel analysis are resolved before mitosis¹⁸.

Replicative polymerase usage during HoRReR

During canonical DNA replication leading and lagging strand synthesis are performed by Polε and Polδ respectively. To establish which polymerases replicate the leading and lagging strands during HoRReR, we exploited alleles (Polε; cdc20-M630F and Polδ; cdc6-L591G) that incorporate excess ribonucleotides^28,29. DNA strands replicated by one or other of these mutant polymerases will harbor extensive rNTPs that, in the absence of RNaseH2 activity, allows the use of a strand-specific alkali liability assay to infer which polymerase replicated either the Watson or Crick strand of any specific locus²⁸.

We examined two regions (Fig. 1a): a control region between ars3004/3005 and RTS1 that should not be affected by RTS1 activity and the experimental ura5-ura4 region immediately downstream of RTS1. To determine the difference between canonical replication (barrier activity OFF) and HoRReR (barrier activity ON), strand-specific alkali sensitivity was assayed in either the absence (rtf1Δ) or the presence (rft1⁺) of the Rtf1 protein cofactor. Two further control conditions were included: first, the mutant polymerases were compared to the wild-type polymerases. Second, the presence or absence of RNaseH2 (rnh201Δ) was matched (in the presence of RNaseH2 ribonucleotides are rapidly excised).

The Polε and Polδ mutations were analysed separately (Fig. 1c,d). As expected, in strains where the polymerases were wild-type, we did not observed high levels of alkali sensitivity for any strand. Similarly, when RNaseH2 was active, we did not observe alkali sensitivity. In the rnh201Δ background, the control region that is replicated canonically showed leading strand-specific alkali sensitivity in the presence of cdc20-M630F (Polε; Fig. 1c; c.f. lanes 20,24), irrespective of barrier activity. Similarly, we observed lagging strand-specific alkali sensitivity in the presence of cdc6-L591G (Polδ; Fig. 1d; c.f. lanes 28,32). Thus, rtf1⁺ expression does not affect normal replication. In contrast, for the experimental region that is replicated by HoRReR in the presence of rtf1⁺, in the cdc20-M630F (Polε) background leading strand alkali sensitivity was only observed when RTS1 was inactive (Fig. 1c; c.f. lanes 4,8). Thus, Polε is not involved in catalyzing synthesis of either strand when DNA replication is restarted by HR. For the cdc6-L591G allele (Polδ), we observed lagging strand alkali sensitivity irrespective of barrier activity (Fig. 1d; c.f. 12,16). We found that the leading strand also became sensitive in the presence of RTS1 barrier activity (Fig. 1d; c.f. lanes 4,8). Thus, both strands are synthesised by Polδ when replication is restarted by HR (HoRReR).

We have previously demonstrated that the error prone nature of HoRReR forks declined to approximately 25% over the first ~1 kb of DNA replication and then stabilized at a high level over the subsequent ~1.5 kb²⁵. To eliminate the possibility that rNMPs were only incorporated into both strands over this first ~1kb we re-analysed the experimental region in two parts (Fig. 2a). We observed broadly equivalent leading and lagging strand degradation for both fragments, indicating high levels of ribonucleotide incorporation across the region.

Figure 2. Polδ usage is relatively uniform and Polα usage decreased.

(a) Polδ extends a minimum of several kb of the leading strand during homologous recombination restarted replication. The ura5 (ClaI-BlpI) and ura4 (HindIII-ClaI) regions were analysed separately. Genomic DNA from the strains indicated was digested, alkali treated and separated on an alkaline gel. The Watson and Crick strands were visualized following alkali treatment using single-stranded probes as indicated by colors of frames (c.f. Fig. 1a). Non-specific band; *. (b) Spontaneous mutation rates for indicated strains in a swi7-H4 (Polα) background. ON and OFF represent rtf1⁺ and rtf1-d, respectively. Values: mean of three independent experiments, each with 11 independent cultures. Error bars: s.d.

Polα initiates each DNA synthesis event and thus contributes substantially to lagging strand synthesis. To address the role of Polα during HoRReR we used a genetic assay that exploits a highly mutagenic allele of Polα, swi7-H4, that increases the mutation frequency³⁰ by a factor of approximately 250. Since ura4⁺ and ura5⁺ are the only genes in S. pombe that confer 5-fluroorotic acid (5FOA) resistance when inactivated by mutation, and these both reside within the experimental region replicated by HoRReR when the RTS1 barrier is active, we used the mutation rate to 5FOA resistance as an indication of the contribution of Polα to DNA synthesis of this region (Fig. 2b). When the ura4-ura5 locus was replicated by HoRReR (barrier ON) in the presence of wild type Polα (swi7⁺) we observed that the mutation rate of the ura4 and ura5 at the T45R locus increased 4.3 times compared to when the barrier was OFF (from 1.1 to 5.2 × 10⁷). This is consistent with previous reports that HoRReR is prone to replication slippage errors²⁴. In the presence of the mutagenic swi7-H1 (Polα) allele and when the RTS1 barrier activity was OFF (canonical replication), we observed that the mutation frequency increased by a factor of 250 when compared to swi7⁺ cells (1.1 to 300 × 10⁷). This reflects the substantial usage of Polα during canonical replication. However, when the barrier was active in the swi7-H1 (Polα) strain background (the ura4 and ura5 genes are replicated by HoRReR), we observed that the mutation rate decreased by a factor of 5.2 (300 to 58 × 10⁷) when compared to the frequency seen when replication was canonical in the swi7-H1 (Polα) background (barrier OFF). This implies that, when compared to unperturbed replication, Polα is polymerizing less DNA during HoRReR.

HoRReR forks are replicated semi-conservatively

BIR initiates replication from a DNA DSB and proceeds via a migrating D-loop^31-33. A similar mechanism could explain the lack of Polε usage during RTS1-dependent HoRReR. D-loop replication is conservative, i.e. the two newly synthesised strands are present in the same sister duplex (Fig. 3a). We thus established (Fig. 3b) the mode of replication of the ura4-ura5 locus using density substitution³⁴. Arrest ON (rtf1⁺) or arrest OFF (rtf1Δ) cdc6-L591G (Polδ) rnh201Δ cells were first synchronized in G2 and the cultures then divided into two aliquots. One aliquot from both the ON and OFF cultures was harvested immediately, representing a “light-light” isotope control (Fig. 3c). The other aliquot of each culture was transferred into media containing heavy isotope and allowed to proceed through mitosis and enter and complete S phase before harvesting. DNA was then prepared, digested with HindIII and BlpI to release the ura4-ura5 fragment and subjected to density gradient centrifugation (Fig. 3c). At time zero, we observed that the ura4-ura5 DNA peaked as light-light, as expected. Following either unperturbed replication (OFF), or HoRReR subsequent to fork arrest and restart (ON), we observed that the ura4-ura5 DNA peaked as heavy-light. Importantly, no heavy-heavy DNA was detected. By subjecting an aliquot of DNA to alkali electrophoresis and probing separately for the leading and lagging strands (Fig. 3d), our alkali-sensitivity assay confirmed that the rtf1⁺ (arrest ON) strain replicated the leading strand using Polδ. Thus, replication after HR-restart is semi-conservative and mechanistically distinct from BIR.

Figure 3. HR-restarted replication is semi-conservative.

(a) Possible models for progression of Homologous Recombination Restarted Replication (HoRReR). Top: following strand invasion, replication occurs via a migrating D-loop. The newly synthesised strand is used as a template for the “lagging” strand, resulting in conservative replication. Bottom: Following strand invasion and D-loop formation, the replication fork is reset such that both parental strands template new synthesis, resulting in semi-conservative replication. (b) Schematic representation of DNA fragment analysed. (c) Density substitution analysis of HR-restarted replication. Sample are digested DNA from rnh201-d polδ-L591G cells at (T0) or after 150 min incubation in a heavy medium (T150). The distribution of the ura4-ura5 fragment after CsCl gradient centrifugation is shown. Upper panel: rtf1⁺ (RTS1 barrier activity ON). HH+LL indicates distribution of control DNA (mixture of heavy:heavy and light:light plasmid DNA). Lower panel: rtf1-d (RTS1 barrier activity OFF). Error bars: s.d. n=3. (d) For density substitution experiments, the leading strand is synthesised by Polδ after HoRReR (see Fig. 1d). The leading and lagging strands of ura4-ura5 were detected by single-stranded probes. Density substitution experiments in a cdc6⁺ (Polδ⁺) rnh2⁺ background (where HoRReR cannot be verified at the time of the experiment) show equivalent results.

HoRReR errors are not intrinsic to δ:δ replication

HoRReR forks are prone to error. In the RTS1 system HoRReR causes an increase in template switches at micro-homology²⁴ and a high rate of fork U-turn between inverted repeats²⁵. One possibility is that these errors are simply a consequence of using Polδ to replicate both DNA strands. It has been previously shown that the catalytic activity of Polε is not essential for cell viability³⁵ and it is known that, during replication driven by the SV40 T antigen, Polδ can replicate both strands³⁶. Thus, in the absence of the Polε catalytic domain it has been proposed that Polδ replicates both the leading and the lagging strand. We therefore assayed replication slippage and fork U-turn in an S. pombe cdc20-ΔN strain in which the catalytic domain of Polε is deleted³⁷. First, we established that the bulk of DNA synthesis was indeed performed by Polδ in cdc20-ΔN cells. The alkali sensitivity of the previously characterized ura4 locus was assessed in cdc6-L591G (Polδ) cdc20-ΔN rnh201Δ cells (and relevant controls) using strand-specific probes (Fig. 4a). We observed that both the leading and the lagging strands were degraded (and thus synthesised by Polδ; see lane 4, leading and lagging strand Southern blots) when Polε was catalytically inactive, confirming its expected substitution by Polδ.

Figure 4. Instability resulting from homologous recombination restarted replication is not intrinsic to leading strand synthesis by Polδ.

(a) Both leading and lagging stands are synthesised by Polδ in the Polε N-terminal deletion strain. Alkali sensitivity in the indicated strains was assessed by Southern blot using strand-specific probes. (b) Schematic of the Rura4-sd20 locus for the microhomology mediated strand exchange assay. ura4-sd20 contains a 20bp duplication flanked by microhomology and is phenotypically ura⁻ (c) Replication slippage at Rura4-sd20 measured in a cdc20⁺ (Polε⁺) and a cdc20-ΔN strain where the catalytic domain of Polε is deleted. Replication forks were either not arrested (OFF: rtf1Δ) or arrested (ON: rtf1⁺) at RTS1. Error bars; s.d. n=3, each with 11 independent cultures. (d) Schematic of the TpalR locus used for the Gross Chromosomal Rearrangement (GCR) assay. Homologous recombination restarted replication results in fork U-turn at the inverted repeat center, causing acentric and dicentric chromosome formation. (e) Southern blot of normal (9.9 kb) and dicentric (14.4 kb) chromosomes in three independent isolates of TpalR cdc20⁺ (Polε⁺) and TpalR cdc20-ΔN in an rtf1Δ background (no arrest). A low level of uncharacterized higher molecular weight species is evident in cdc20-ΔN cells (vertical bar: ?). Right: quantification of 14.4 kb (GCR) band, or all high molecular weight species. Error bars; s.d. n=3 (f) Southern blot of normal (9.9 kb) and dicentric (14.4 kb) chromosomes at T= 0, 3.5 and 7 hours after induction of rtf1⁺. Right: quantification of three experiments. Error bars: s.d. The increased GCRs seen in panel f at T=0 for cdc20-ΔN likely represents increased leakiness of the P_urg1 promoter used.

To measure the level of microhomology-mediated template exchange (replication slippage) in cells where the Polε catalytic domain is deleted we exploited the Rura4-sd20 loci (Fig. 4b). This contains a 20 bp duplication within the ura4 ORF which, when deleted by replication slippage, restores uracil prototrophy²⁴. As previously reported, in the cdc20⁺ (Polε⁺) background, we found that microhomology mediated slippage at ura4-sd20 increased by an order of magnitude when ura4-sd20 was replicated by HoRReR forks (Fig. 4c). In the cdc20-ΔN (Polε catalytic deletion) background, replication without fork arrest and restart (arrest OFF) we observed an increase in slippage of ~5 times more when compared to the cdc20⁺ (Polε⁺) background. This demonstrates that intrinsic replication by Polδ on both strands is prone to replication slippage. However, when replication of the locus was performed by HoRReR forks (arrest ON) in the cdc20-ΔN background, we found that slippage increased a further factor of 2, becoming comparable to Polε⁺ arrest ON cells. This demonstrates that, while intrinsic replication by Polδ on both strands is prone to replication slippage, this error frequency is further increased when replication occurs by HoRReR and that there must therefore be additional features of the HoRReR machinery that cause these replication errors.

To assess gross chromosomal rearrangements (GCRs) caused by fork U-turn at an inverted repeat during intrinsic replication (arrest OFF) in the cdc20-ΔN (Polε catalytic deletion) background, we exploited the TpalR loci (Fig. 4d). This construct has a short inverted repeat downstream of the RTS1 barrier that is highly prone to generating acentric and dicentric chromosomes when replicated by HoRReR²⁵. These GCRs are detected by Southern blot: the parental chromosome is represented by a 9.9 Kb fragment, while the dicentric chromosome (the product of the U-turn) is represented by a 14.4 kb fragment. In the absence of replication arrest (rtf1Δ) we observed a similar low level of GCRs (Fig. 4e) when comparing cdc20⁺ (Polε⁺) and cdc20ΔN (Polε catalytic deletion). This indicates that replication of both strands by Polδ is not intrinsically prone to fork U-turn and the generation of GCRs at this locus. We note a small but reproducible increase in higher molecular weight products (greater than the 14.4 kb dicentric fragment; marked with a ? in Fig. 4e) that we are not currently able to characterise.

When replication fork arrest was induced (P_urg1-rtf1) and cells grown for approximately 3 cell cycles (Fig. 4f), we observed that the level of the rearrangement increased over the time course for both cdc20⁺ (Polε⁺) and cdc20-ΔN (Polε catalytic deletion) backgrounds. Thus, intrinsic replication by Polδ on both strands is not prone to generating GCRs at inverted repeat sequences and such rearrangements are increased in the cdc20-ΔN (Polε catalytic deletion) background when replication occurs by HoRReR. We attribute the elevated levels of rearrangement seen at time zero in the cdc20-ΔN P_urg1-rtf1 background (Fig. 4f) to an increased promoter leakiness in these cells when compared to cdc20⁺ P_urg1-rtf1. The important point is that this elevated frequency is not seen in the rtf1Δ background (Fig. 4e).

Biased Polδ usage may indicate chromosome fragility

If leading and lagging strand replication by Polδ is a common feature of HoRReR, some genomic fragile sites may show a relative increase in Polδ usage for duplex DNA. This predicts that Polδ usage enriched regions in unperturbed cells could reflect intrinsic fragile sites. We have previously reported that many loci are slightly enriched for Polδ usage when compared to the genome average²⁹ and that the majority of these corresponded to efficient replication origins. We postulated that this may reflect occasional leading strand initiation by Polδ and the subsequent replacement of Polδ on the lagging strand by Polε as replication progresses²⁹.

After eliminating origin-associated regions that show a relative enrichment for Polδ usage, plus loci with heterochromatic and repetitive features such as centromeres, telomeres and the mating type locus, several Polδ-enriched regions remained that show no obvious correlative features. We chose four such regions (Fig. 5a) and, avoiding annotated features such as transcription units, integrated a 1.2kb inverted repeat at several positions, either where Polδ was relatively enriched (blue) or at unenriched flanking controls (red). When replicated by HoRReR forks the inverted repeat is highly prone to U-turn and concomitant GCR formation that can be physically assayed by quantification of Southern blots²⁵. We observed that two sites (1 and 3; Fig. 5b) showed a statistically significant increase in the level of GCRs associated with Polδ enrichment when compared to the control sites (p=0.037 and 0.018). A third site showed a similar trend that did not reach statistical significance, while a fourth site showed no difference.

Figure 5. Polδ usage correlates with fragile sites.

(a) The positions at which a the small inverted repeat (top) was integrated. Theses correspond to sites around four euchromatic regions that are not associated with a replication origin (open circles), but which showed an intrinsic relative increase in Polδ usage (Polδ:Polε ratio) for the duplex DNA. Blue: relatively increased Polδ, red: relatively increased Polε. Numbers: kb on indicated chromosome (b). Gross Chromosomal Rearrangements (GCRs) resulting from a fork U-turn at the center of the inverted repeat were assayed. Top: a representative Southern blot. Bottom: the % of GCRs (dicentric + acentric as a proportion of total signal) in the population quantified in 3 independent experiments. * = significant p-values (two-tailed Student’s t test): region 1, I-3870/I-3920, 0.037 and I-3890/I-3920, 0.018. Region 3, II-1165/II-1190, 0.010.

Discussion

To understand the mechanisms of genetic instability underpinning genomic disorders and carcinogenesis it is important to understand the impediments to replication, the cellular responses to these impediments and the mechanism by which these operate to ensure complete and faithful replication. Despite the function of the intra-S phase checkpoint, cells cannot avoid replication breakdown and use homologous recombination (HR) to both protect³⁸ and restart replication forks that are inactivated^39,40. The nature of HR is such that it becomes a double edged sword; the ability to resume replication arises at the cost of potential NAHR¹³, which causes chromosomal rearrangements. Further errors are introduced during the recombination dependent replication that ensues after fork restart^22,24,25.

The nature of HR-restarted replication

In S. cerevisiae, studies of the in vivo mutation spectra resulting from specific Polε and Polδ mutations with characterized in vitro substitution biases lead to a model by which the labor of replication was divided: Polε synthesised the leading strand and Polδ the lagging strand⁴¹. This model received significant support from a separate study using S. pombe where an equivalent genetic experiment assigned Polδ to the lagging strand²⁸. Because a Polε mutant that exhibited a biased mutational spectra could not be identified in S. pombe, a physical assay was used to assign Polε to the leading strand. This assay exploited the propensity of a specific Polε mutant to incorporate rNTPs at elevated frequency²⁸. Subsequent, the model was supported by biochemical characterization of the Polδ and Polε holoenzymes in the context of the CMG helicase⁴². Finally, whole genome strand-specific profiling of rNTP incorporation in strains containing either Polε, Polδ or Polα mutants that incorporated elevated levels rNTPs demonstrated that the division of labor between Polε and Polδ pertained across the entirety of both yeast genomes^29,43,44.

Canonical DNA replication is highly accurate⁴⁵, whereas RDR and HoRReR are surprisingly error prone^22,24,25. Understanding the nature of the restarted replication machine is therefore of interest for understanding the causes of genetic instability and its contribution to human disease. However, it remains unclear how the different systems used to characterise RDR and HoRReR, mainly BIR in S. cerevisiae and RTS1 in S. pombe respectively, are mechanistically related. During S phase fork collapse will result in both single-ended DNA DSBs (for example, when a fork encounters a nick in the template strand) and structures where the three duplexes remain connected by two regions of single stranded DNA. While these latter structures can be processed to form a single-ended DSB, this is not a requirement for HoRReR^20,39.

The RTS1 model for HoRReR initiates without a DSB²⁰, occurs within S phase^18,19, initiates replication rapidly (Fig. 1), utilizes Polδ to synthesize both strands (Fig. 1), uses Polα less frequently than canonical replication (Fig. 2) and proceeds via a semi conservative process (Fig. 3). Semi-conservative replication is incompatible with a D-loop migration mechanism, but the use of Polδ on both strands and the fact that this cannot explain all of the elevated replication errors observed during HoRReR (Fig. 4) strongly implies a mode of replication that is distinct from a canonical fork. The further distinctions between a canonical replication and RTS1-induced HoRReR remain to be determined.

In contrast to the RTS1 system studied here, The BIR systems used to study RDR in S. cerevisiae initiates replication from a single-ended DSB outside of S phase and there is a significant delay to the onset of DNA synthesis¹⁵. Once RDR initiates, replication proceeds conservatively via a migrating D-loop^31-33 that often replicates to the end of the template molecule and does not meet an adjacent fork. Thus, the two systems are distinct, although it is unknown if this reflects differences between the two organisms, differences between RDR being initiated by a DSB as opposed to HoRReR being initiated via a non-DSB intermediate, or if it reflects differences in initiating RDR during G2 phase as opposed to HoRReR within S phase.

Non-canonical restarted forks may contribute to genome instability

Oncogene activation, an early step in carcinogenesis, results in unbalanced replication which manifests as replication stress^46,47. Such oncogene-induced stress provides a barrier to carcinogenesis through the activation of the DNA structure checkpoints⁴⁸. When these checkpoints are subverted to allow cancer development, continued oncogene-induced stress is thought to underlie much of the characteristic genetic instability of cancers. The mechanisms by which oncogene activation promotes replication stress remain obscure, but low levels of dNTPs⁴⁹, increased replication origin firing^50-52, up-regulated transcription⁵¹ and premature activation of structure-specific nucleases⁵³ have all been implicated.

The GCRs characterized in cancer are often associated with specific genomic regions such as common fragile sites (CFS) and early replicating fragile sites (ERFS). The innate fragility of some loci, when combined with oncogene-induced replication stress, likely results in increased abundance of replication fork collapse and – particularly in regions with a paucity of origins – this may increase reliance on replication restart and HoRReR. Because HoRReR in fission yeast results in the use of Polδ to replicate both strands, we reasoned that, in a population of cells, loci prone to occasional replication fork collapse and subsequent HoRReR may manifest as regions where ensemble mapping of replication polymerase usage on the duplex DNA reveals a bias towards Polδ when compared to Polε.

Analyzing our previously published data²⁹ we identified 4 sites such sites with unexplained increases in apparent Polδ usage and integrated a reporter for HoRReR signature errors. Overall, we cannot draw the direct conclusion that these sites represent previously unidentified fragile sites, but the fact that two showed a statistically significant increase in instability (p=0.037 and 0.018) when compared to matched control loci supports the possibility that some of the genomic instability associated with fragile sties in cancer development may result directly from forks restarted by HR within S phase that are non-canonical and thus prone to error.

Online Materials and Methods

S. pombe Methods

Cells were grown in Yeast Extract (YE) or EMM2 minimal medium. Standard genetic and molecular procedures were employed as described previously⁵⁴. The cdc20-ΔN (cdc20::hphMX6-Pnmt1-cdc20CTD) strain³⁷ was grown in EMM2 without thiamine.

Construction of TER2/3-ura4/5-RTS1 (T45R)

The previously described TpalR (TER2/3-ura4 palindrome-RTS1) construct²⁵ was modified as below. The centromere-proximal ura4 gene in TpalR was replaced with AseI-HindIII fragment containing ura5 gene. To increase the ratio of restarted forks, 10 copies of TER2/3 were inserted ~500 bp telomere-proximal of TER2/3 in the construct by recombination-mediated cassette exchange (RMCE)⁵⁵. The original ura5 gene on chromosome II was deleted.

Detection of alkali-sensitive sites in genomic DNA

Genomic DNA, digested with appropriate restriction enzymes, was incubated in 0.3N KOH at 55°C for 2 hours and subjected to 1% alkaline agarose gel electrophoresis. Each strand of ura4-ura5 fragment was detected by Southern blot using a single-stranded probe as described previously²⁸. All uncropped blots are shown in supplementary dataset 1.

Monitoring replication timing

cdc25-22 cells were arrested at G2 at 36°C for 3.5 hours and released into cell cycle at 25°C. Cells were collected every 15 minutes and genomic DNA was extracted. Replication was monitored by quantitative PCR (qPCR).

Determination of spontaneous mutation rates

Uracil prototrophic cells were spread on YEA (YE agar) plates and incubated at 30°C for 3 days. 11 independent single colonies were inoculated in 5ml YE and grown to saturated phase at 30°C. Cells were diluted appropriately and plated on YEA or YEA containing 0.1% 5-fluoroorotic acid (5-FOA). Colonies were counted after 4 days incubation at 30°C. Mutation rates were calculated by the method of median⁵⁶.

Detection of chromosomal rearrangements

For P_urg1-rtf1⁺ expression⁵⁷, cells were grown to log phase in EMM2 without uracil, then 250 μg/ml of uracil was added to induce expression of the rtf1⁺ gene. Genomic DNA was prepared in agarose plugs, digested with an appropriate restriction enzyme after equilibration with the reaction buffer and subjected to agarose gel electrophoresis. Chromosomal rearrangements were detected by Southern blot using an appropriate probe. Band intensities were quantified using ImageJ. All uncropped blots are shown in supplementary dataset 1.

Replication slippage assay using ura4-sd20

The ura⁺ reversion assay using ura4-sd20 was performed as described previously²⁴ with some modifications. Briefly, 5-FOA resistant cells were grown in uracil-containing media to saturation. Cells were washed, appropriately diluted and plated on minimal medium plates with or without uracil. Colonies were counted after 4-5 days incubation at 30°C to determine the reversion frequency.

Determination of the mode of replication by density substitution

Cells were grown, synchronized and treated as described in⁵⁸. Briefly: rnh201-d polδ-L591G cells were grown in EMM2⁵⁴ containing 0.2% NH₄Cl and 1% glucose were harvested and loaded on lactose gradients. After centrifugation, small (G2) cells were collected and split into two aliquots. One was centrifuged immediately (T0), the other was washed and transferred into “heavy” EMM2 containing ¹⁵N NH₄Cl and ¹³C glucose. Cells were harvested after 150 min incubation at 30°C (T150). Genomic DNA was extracted and digested with HindIII and BlpI and split into two aliquots. One was treated with alkali as described above, followed by alkaline gel electrophoresis. The other was loaded on a CsCl solution gradient and density gradient centrifugation³⁴ was performed at 30,000 rpm for 67 hours. The density gradient was fractionated and distribution of the ura4-ura5 fragment was quantified by slot-blot using ura5 probe. Relative intensity of each fraction to the peak in T0 was plotted.