The extent of error-prone replication restart by homologous recombination is controlled by Exo1 and checkpoint proteins

Replicative stress can be caused by a wide variety of situations, including tightly bound protein–DNA complexes, clashes of the replication machinery with other cellular processes (i.e. transcription), the presence of non-canonical DNA structures and nucleotide precursor depletion (Lambert and Carr, 2013a). The intra-S-phase checkpoint acts within S phase and promotes cell survival and genome stability in response to replicative stress (Lindsay et al., 1998; Lopes et al., 2001; Tercero and Diffley, 2001) by stabilising arrested forks. In all organisms studied, the intra-S-phase checkpoint requires the activity of two kinases – a phosphoinositol-3 kinase-like kinase (PIKK), known in metazoans as ATR, that senses replication problems by interacting directly with single-stranded DNA (ssDNA)-binding proteins (Zou and Elledge, 2003) and a downstream checkpoint kinase that is directly activated by ATR through interactions with the mediator protein Claspin (Errico and Costanzo, 2012; Segurado and Tercero, 2009).

In the fission yeast Schizosaccharomyces pombe, the ATR homologue is known as Rad3^ATR and the downstream effector kinase for the intra-S-phase checkpoint is Cds1^Chk2 (Lindsay et al., 1998). As is the case in mammalian cells, but not for the budding yeast Saccharomyces cerevisiae, the heterotrimeric checkpoint clamp complex, composed of Rad9, Rad1 and Hus1 (9-1-1) is also essential for the intra-S-phase checkpoint (Errico and Costanzo, 2012). Activation of the intra-S-phase checkpoint results in phosphorylation of a wide range of replication proteins and DNA repair proteins (Bailis et al., 2008; Boddy et al., 2000; Chen et al., 2010; De Piccoli et al., 2012; Hu et al., 2012; Miyabe et al., 2009; Segurado and Tercero, 2009; Smolka et al., 2007). The precise details of how these phosphorylation events regulate DNA replication and DNA metabolism to stabilise the arrested fork remain largely obscure. In part, this is because the full range of phosphorylation events have not yet been fully mapped and phenotypically characterised. It is also because, in the absence of the intra-S-phase checkpoint, the DNA structures that are initially present at the replication fork in S phase are processed (Sogo et al., 2002) into different structures [i.e. into ssDNA and double-strand breaks (DSBs) (Sabatinos et al., 2012)]. These can both signal through the G2 DNA damage checkpoint and be repaired in a distinct manner from the original lesion. Further complicating genetic analysis, the DNA damage checkpoint requires the function of many of the same proteins as the intra-S-phase checkpoint, with the exceptions that the effector kinase and mediator proteins are replaced (Carr, 2002; Stracker et al., 2009).

Experimentally, replication stress is often imposed by treating cells with the ribonucleotide reductase inhibitor hydroxyurea to globally inhibit replication. During the subsequent dNTP depletion, the intra-S-phase checkpoint stabilises the slowed-down replication forks in a replication-competent state (Lopes et al., 2001). Here, we will refer to these stabilised structures as ‘stalled’ forks. Stalled forks can resume replication without intervention from additional mechanisms when the blockade is removed. By contrast, hydoxyurea treatment in the absence of the intra-S-phase checkpoint results in replication forks that cannot resume (Sabatinos et al., 2012; Sogo et al., 2002). We refer to these as ‘collapsed’ forks. Fork collapse likely occurs when the activities of the replicative helicase and the replicative polymerases are uncoupled, generating extensive stretches of ssDNA. Collapsed forks can also result from clashes between replisomes and the transcription machinery or tightly DNA-bound protein complexes. It has been reported that collapsed forks are no longer associated with components of the replisome, i.e. the replication machinery is not available for DNA synthesis (Cobb et al., 2003; Cobb et al., 2005; Katou et al., 2003; Lucca et al., 2004). However, this might be a simplification, and the machinery might still be present, but no longer able to resume replication (De Piccoli et al., 2012).

Irrespective of the precise nature of the replication machinery present at collapsed forks, it has been demonstrated that, in the absence of the intra-S-phase checkpoint, the genome of hydroxyurea-treated yeast cells is degraded by nucleases, including Mus81 (Boddy et al., 2000; Froget et al., 2008; Kai et al., 2005) and Exo1 (Cotta-Ramusino et al., 2005; Lopes et al., 2001; Segurado and Diffley, 2008; Sogo et al., 2002), and that this degradation is prevented, at least in part, by the phosphorylation of a range of replication proteins and proteins that process specific DNA structures (Chen et al., 2010; De Piccoli et al., 2012; Hu et al., 2012; Segurado and Tercero, 2009; Smolka et al., 2007). The precise physical consequences of fork collapse at the level of the resulting DNA structure remain largely unclear. It is also not clear whether the phenomenon of fork collapse is a cause or consequence of inappropriate DNA processing. However, once a fork does collapse, the DNA is exposed to recombination events that can potentially lead to genome instability (Alabert et al., 2009; Barlow and Rothstein, 2009; Iraqui et al., 2012; Lambert et al., 2010; Lambert et al., 2005; Lisby et al., 2004; Meister et al., 2005; Mizuno et al., 2009; Myung et al., 2001; Myung and Kolodner, 2002; Segurado and Diffley, 2008).

Nucleotide depletion is only one of many potential barriers to replication. Although avoiding fork collapse is one key function of the intra-S-phase checkpoint, in certain situations the collapse of an arrested replication fork might be preferable to its stabilisation. In other cases, replication fork collapse might be unavoidable; for example, when the replisome is blocked by an inter-strand crosslink there is not likely to be sufficient ssDNA to activate the intra-S-phase checkpoint. Interestingly, loss of the intra-S-phase effector kinase Cds1^Chk2 in fission yeast increases the resistance of otherwise wild-type cells to treatment with the DNA inter-strand crosslinking agent nitrogen mustard (Lambert et al., 2003). This suggests that the initial activation of the checkpoint effector kinase is detrimental to cell survival in these circumstances (reviewed in Lambert and Carr, 2013a). There is indirect evidence to suggest that the intra-S-phase checkpoint proteins regulate the use of recombination for the subsequent repair or restart of collapsed replication forks (Haghnazari and Heyer, 2004; Pandita et al., 2006; Sørensen et al., 2005). However, the mechanisms by which the checkpoint proteins might facilitate recombination at collapsed replication forks are not clear; for example, checkpoint proteins might function to directly promote recombination or they might favour certain recombination pathways over others (Haghnazari and Heyer, 2004; Kolodner et al., 2002).

The activation of the intra-S-phase checkpoint and its components has been largely characterised in response to acute replication stress (such as hydroxyurea treatment), whereas the cellular response to chronic and endogenous replication stress is less well characterised, despite the fact that this represents the main source of replication-induced genetic instability in pre-neoplastic lesions (Bester et al., 2011; Halazonetis et al., 2008). In this report, we have therefore used an established replication fork barrier (RFB) that induces a local and chronic replication stress to explore the ATR-dependent checkpoint response. The RFB we have exploited is the RTS1 sequence in fission yeast. RTS1 is a well-characterised polar RFB that requires a sequence specific Myb-domain DNA-binding protein, Rtf1, for its function (Lambert and Carr, 2005; Lambert et al., 2010). In wild-type S. pombe cells, RTS1 resides close to the mating-type (mat) locus and, when bound by Rtf1, functions to block replication forks passing in one direction while allowing them to pass unhindered in the opposite direction (Eydmann et al., 2008). Although RTS1:Rtf1 is not directly involved in mating-type switching, its barrier activity facilitates switching by preventing inappropriate forks moving through the switching region in the wrong direction (Codlin and Dalgaard, 2003; Dalgaard and Klar, 2001; Lee et al., 2004). In our experimental systems, either one or two copies of the 850-bp RTS1 sequence are positioned at the ura4 locus (Fig. 1A; Fig. 2B) and rtf1⁺, which is essential for RTS1 RFB activity, is under the control of a thiamine-repressible nmt promoter. Upon induction of rtf1⁺ transcription, forks arrest and rapidly collapse (Lambert et al., 2005). Recombination proteins are required for fork restart (Lambert et al., 2010), which occurs within 20 minutes (unpublished data). There is no cell cycle arrest resulting from this DNA processing (supplementary material Fig. S1) (Lambert et al., 2005), consistent with there being sufficient time within the normal cell cycle to restart the collapsed forks by homologous recombination.

Using a construct where two RTS1 sequences are integrated at the ura4 locus as inverted repeats flanking ura4⁺ (RuraR), we have previously demonstrated that, when forks arrest at RTS1, they are subject to recombination-mediated restart (Lambert et al., 2005). Greater than 94% of forks arrest when they encounter the RFB, and the vast majority of these restart correctly (Lambert et al., 2010; Mizuno et al., 2013) within 20 minutes and complete replication. In the RuraR system (Fig. 1A), cell viability is impaired when the recombination pathway, but not the Rad3^ATR checkpoint protein, is compromised (Lambert et al., 2005) and no cell cycle delay is observed when the RTS1-RFB is induced (Lambert et al., 2005; supplementary material Fig. S1). To extend this observation, we crossed the RuraR locus into backgrounds null for cds1, chk1 (downstream effector kinases), rad17 (checkpoint clamp loader) and rad9 (9-1-1 checkpoint clamp subunit). We used a micro-colony assay to establish the percentage of cells able to form colonies of greater than ten cells when replication arrest was either induced or not induced. We confirmed that (unlike recombination-defective RuraR control strains) viability was not significantly affected by checkpoint loss; rad3-, rad9-, rad17-, cds1- and chk1-null RuraR cells were all able to form micro-colonies as efficiently as the checkpoint-proficient RuraR strain (Fig. 1B). In order to rule out the possibility that checkpoint mutants are defective for RTS1-dependent RFB activity, we assayed replication intermediates in rad3-null and rad3⁺ control strains grown both with and without thiamine. The extent of fork pausing at RuraR in rad3-null cells (98.4%±1.2 of arrested forks, ±s.d.) was comparable to that observed for wild-type cells (97.3%±1.8) (Fig. 1C).

Thus, replication completion upon activation of the RTS1-RFB does not require checkpoint pathways or cell cycle delay. This is in contrast to acute replication stresses caused by agents such as hydroxyurea. These circumstances thus allow us to separate the known roles of the intra-S-phase and replication checkpoints – namely, promoting replication resumption by preventing fork collapse and delaying the cell cycle, respectively – from any potential functions in either regulating DNA metabolism at a collapsed fork or in regulating the ensuing choice of homologous recombination pathway. We thus set out to examine the recruitment of homologous recombination proteins to the collapsed fork at RTS1 and to observe potential changes to aberrant recombination outcomes caused by the loss of checkpoint proteins.

We demonstrated previously that induction of replication fork arrest at RTS1 leads to the recruitment of Rad52 to the RuraR locus (Lambert et al., 2005). To establish whether the extent of recruitment is subject to checkpoint regulation, we performed chromatin immunoprecipitation (ChIP) analyses against Rad52–GFP in checkpoint-mutated RuraR strains (Fig. 1D) following growth for 40 hours either with thiamine (arrest ‘off’) or without thiamine (arrest ‘on’). Transcription is induced by the nmt41 promoter ∼16 hours after thiamine removal (Maundrell, 1993). Thus, at 40 hours in the absence of thiamine, the culture is in a steady state, where forks arrest at RuraR during each S phase and replication of the locus is reliant on homologous-recombination-dependent fork restart. Consistent with this, we have demonstrated previously that fork arrest at RuraR is not detectable at 12 hours following thiamine removal, but occurs with similar efficiency at both 24 and 48 hours after thiamine removal (compare with figure 1 in Lambert et al., 2005).

In the wild-type strain background, ‘arrest on’ conditions resulted in Rad52 recruitment at, and immediately flanking, the RuraR locus (Fig. 1D, top panel, blue bars). As expected, enrichment was most prevalent at the right-hand [centromere (cen)-proximal] barrier, because the direction of replication fork movement is from right to left (Mizuno et al., 2013). Enrichment persisted for ∼2 kb cen-proximal to the RTS1 sequence. A second less-prevalent region of enrichment extended ∼1 kb from the left-hand barrier towards the telomere (tel). In the rad3^ATR-null background, Rad52 recruitment was significantly increased, both at RTS1 sequences and in the flanking regions, and spread further behind the fork arrest site, >3 kb cen-proximal to the RTS1 sequence (Fig. 1D, top panel, red bars). To better visualise the role of Rad3^ATR in regulating Rad52 association at the RuraR locus, Rad52 enrichment in rad3-d cells was calculated relative to that observed for the wild-type control (Fig. 1D, bottom panel). This confirmed that Rad52 was up to six times enriched at both tel- and cen-proximal regions flanking the RTS1-RFB in rad3-d cells when compared with its enrichment in rad3⁺ cells.

In contrast to the higher recruitment of Rad52 in the rad3-null strain, in the strains null for rad17 (clamp loader, Fig. 1E) and rad9 (9-1-1 complex subunit, supplementary material Fig. S2A,B), Rad52 recruitment was significantly reduced by approximately twofold at RTS1 sequences and at both the tel- and cen-proximal flanking sequences relative to its recruitment in wild-type cells. We also assessed the enrichment of Rad52 at the RuraR locus in the double rad3^ATR-rad17-null mutant. This mutant showed a Rad52 recruitment profile similar to that of rad17-null cells (Fig. 1E,F), indicating that the effect of losing Rad3^ATR function requires a functional Rad17/9-1-1 clamp. Interestingly, neither the chk1-null nor the cds1^Chk2-null strains showed a reproducible change in Rad52 recruitment when compared with that of wild-type cells (supplementary material Fig. S2C), a result consistent with there being no evidence for checkpoint-dependent cell cycle arrest (supplementary material Fig. S1). These data show that the checkpoint sensor Rad3^ATR prevents the extensive recruitment of Rad52 upstream of arrested forks, whereas 9-1-1 clamp loading, dependent on the sensor Rad17, promotes such Rad52 recruitment. Moreover, the function of checkpoint sensors in regulating Rad52 association at collapsed forks is independent of the downstream effector kinases Chk1 and Cds1^Chk2. To address whether the regulation of Rad52 association by the checkpoint proteins is related to the formation of ssDNA at blocked forks, we next investigated the role of nuclease activities in Rad52 association.

In response to a DNA DSB, the MRN complex functions to initiate resection (Mimitou and Symington, 2008; Raynard et al., 2008; Symington, 2002). The subsequent generation of ssDNA is largely Exo1-dependent, with a later contribution from Rqh1^RecQ and Dna2. When a fork collapses at RTS1 there is no DSB formation (Mizuno et al., 2009), and homologous-recombination-dependent replication restart occurs from a single-stranded gap (Lambert et al., 2010). To establish whether MRN or the Exo1 nuclease participate in DNA metabolism at RTS1-induced collapsed forks, we investigated the involvement of MRN and Exo1 in Rad52 loading. When the replication fork arrest was induced in a strain null for rad50 (an MRN component), Rad52 recruitment to RuraR was reduced relative to that of the wild-type strain (Fig. 2A, left panels). Similarly, the increased recruitment of Rad52 in the rad3-null background was also reduced. In contrast to rad50-null cells, cells carrying the rad32^mre11-D65N nuclease-deficient allele (Hartsuiker et al., 2009) did not display changes in Rad52 enrichment levels in either the rad3⁺ or rad3-null strains (Fig. 2A). Thus, an intact MRN complex, but not the nuclease activity of the MRN subunit Rad32^Mre11, is required for Rad52 recruitment, and the increased loading observed in a rad3-null background is similarly MRN-dependent.

Exo1 has been implicated with Mre11 in the resection of DSBs to generate ssDNA (Mimitou and Symington, 2008; Moreau et al., 1999; Moreau et al., 2001; Tsubouchi and Ogawa, 2000; Zhu et al., 2008). Exo1 has been reported to be negatively regulated by Mec1^ATR and Rad53^Chk2 in the generation of ssDNA at uncapped telomeres in budding yeast (Jia et al., 2004; Morin et al., 2008; Zubko et al., 2004). Again in S. cerevisiae, during DNA replication, Exo1 is proposed to travel with active replication forks and is known to participate in the instability of stalled forks in the absence of regulation by Rad53^Chk2 (Cotta-Ramusino et al., 2005; Segurado and Diffley, 2008). In our analysis, an exo1-null strain showed a significant decrease in Rad52 recruitment to RuraR, both in the rad3⁺ background and in combination with a rad3-null mutant (Fig. 2A, right panel). Thus, at collapsed forks, Exo1 is required for normal Rad52 recruitment to single-stranded gaps in checkpoint-proficient cells, and it mediates the extensive Rad52 loading observed in rad3-null cells.

Rad52 is known to bind to replication protein A (RPA)-coated ssDNA, replacing the RPA to initiate homologous recombination by nucleating Rad51 filaments (Krejci et al., 2012). Thus, Rad52 recruitment immediately upstream of the site of fork collapse is strongly indicative of DNA processing. We next verified that both Rpa1 and Rad51 were also recruited with similar profiles. To clearly distinguish the recruitment of proteins either upstream or downstream of the collapsed fork, we used the uraR construct in which a single RTS1 barrier is located cen-proximal to ura4⁺ (Lambert et al., 2010) (Fig. 2B). The use of the uraR locus simplifies the analysis because the RuraR locus, in addition to recombination-mediated fork restart at the site of fork collapse, is also subject to recombination events dependent on a template switch between the two inverted RTS1 repeats (Lambert et al., 2010). Such events could influence the location of Rad52 recruitment in the vicinity of the RTS1-RFB (Lambert et al., 2010).

Similar to the increased Rad52 association observed at RuraR, the binding of Rad52, Rad51 and RPA were all increased at uraR in rad3-null cells when compared with that of rad3⁺ controls, particularly upstream of the site of fork arrest (Fig. 2C). The association of these ssDNA-binding proteins with uraR was also dependent on Rad17 and Exo1, as observed for the RuraR construct. Moreover, the double rad17-exo1-null mutant showed a similar reduction in Rad52 binding upstream of the RTS1-RFB to that observed in each single mutant, suggesting that Rad17 and Exo1 act in the same pathway of Rad52 recruitment at arrested forks (Fig. 2C). These data are consistent with the idea that Rad3^ATR limits extensive Exo1-dependent resection behind collapsed forks, whereas the PCNA-like 9-1-1 complex promotes it.

Taken together, our data strongly imply that DNA is resected upstream of a collapsed replication fork in a manner that is dependent on Rad17, the 9-1-1 complex, MRN and Exo1 and is attenuated by the activity of Rad3^ATR. We have reported previously that the majority of arrested forks rapidly restart correctly by homologous recombination and complete replication, but that, in the RuraR system, 2–5% of cells in each generation undergo inappropriate template exchange with the nearby inverted repeat (Lambert et al., 2010; Lambert et al., 2005; Mizuno et al., 2009). This results in intra-chromosomal recombination, leading to either inversion of the ura4 gene or the formation of an acentric and dicentric chromosome (supplementary material Fig. S3). Both of these events can be monitored by Southern blotting and pulse-field gel analysis and, upon fork arrest, the frequency of the rearrangement-specific band increases with each generation. We thus analysed wild-type, checkpoint-null and exo1-null RuraR cultures at T₀ (fork arrest ‘off’) and after 48 hours of growth without thiamine, T₄₈ (fork arrest ‘on’), for fork-arrest-induced recombination intermediates [using two-dimensional gel electrophoresis (2DGE); supplementary material Fig. S3B,C] and for both recombination outcomes – acentric chromosome formation (supplementary material Fig. S3D) and ura4⁺ inversion (supplementary material Fig. S3E).

In rad3-, rad17- and exo1-null mutant backgrounds, recombination intermediates (D loops and structures containing Holliday junctions) occurred at a frequency equivalent to that of the wild-type control, and the recombination outcomes were unchanged. Thus, the perturbation in resection and in recombination protein loading seen in the checkpoint mutants or the exo1 mutant did not significantly influence the amount of homologous-recombination-dependent replication restart after fork collapse or the types of deleterious intra-chromosomal recombination events that occur due to faulty template exchange between RTS1 sequences at RuraR. Our data therefore suggest that a limited amount of recombination factors are sufficient to promote replication restart by template exchange.

Our data are consistent with a model whereby the extent of resection upstream of the collapsed fork is not rate-limiting for fork restart. However, extensive resection, such as that seen in the rad3-null mutant background, implies that restart must frequently occur a significant distance upstream of the point of the original fork collapse. To establish whether this is the case, we visualised the converging fork signal (Fig. 3A) by 2DGE analysis. In a wild-type background, ∼8% of the replication intermediates represented converging forks that we assume arise when an incoming replisome from the tel-proximal side approaches the arrested fork structure close to, or within, RTS1. In rad17- or exo1-null backgrounds, this signal remained constant. However, in rad3-null cells, where resection is proposed to be extensive, the signal was reduced by more than threefold (Fig. 3B,C). This is consistent with the expectation that resection beyond the restriction site cen-proximal to RTS1 (that defines the fragment being analysed by 2DGE, see the right-hand diagram in Fig. 3A) would result in a restart event upstream of the initial point of fork collapse and thus loss of the converging fork signal within the restriction fragment analysed by 2DGE. Concomitant loss of exo1 in the rad3-null mutant restored the converging fork signal to the wild-type level, consistent with the loss of the termination signal being a consequence of extensive resection.

The restart of the collapsed fork upstream of the initial site of arrest would result in more DNA being replicated by the restarted replication machine. We have shown previously that restarted replication forks are prone to replication slippage at sites of microhomology (Iraqui et al., 2012). Using an assay in which replication slippage removes a short direct repeat from the ura4-sd20 allele, we therefore tested for evidence that the region upstream of the blocked fork is more susceptible to such slippage errors when resection is extensive (Fig. 4). First, the spontaneous level of replication slippage (irrespective of whether Rtf1 was overexpressed or not) was similar in all the genetic backgrounds tested (RTS1-RFB is absent; Fig. 4, see construct 1). Second, fork arrest led to an equivalent increase in replication slippage downstream of the RTS1-RFB during replication restart in all genetic backgrounds (Fig. 4A, see construct 2). This observation confirmed that, whatever the level of recombination factors associated with the collapsed replication fork, replication restart occurred efficiently. Third, when the marker gene was placed upstream of the RTS1-RFB, fork arrest in checkpoint-proficient cells led to a threefold increase in replication slippage upstream of the site of fork arrest (P<1.7×10⁻⁵) (Fig. 4, see construct 3). This is consistent with occasional leading-strand degradation and subsequent re-synthesis by a homologous-recombination-restarted replication fork upstream of the initial site of fork arrest (Iraqui et al., 2012).

The induction of replication slippage upstream of the site of replication arrest is Exo1-dependent (P<0.0002). A 1.9-fold induction of replication slippage was observed (Fig. 4B) in rad17-null mutant cells, although this was not statistically significant (P>0.05). By contrast, fork arrest led to an 8.7-fold increase in replication slippage upstream of the arrest site in the rad3-null mutant (P<9.8×10⁻⁵), corresponding to a 2.9 times higher level when compared with that of the rad3⁺ strain (P<1.6×10⁻⁵). Replication slippage occurring upstream of the RTS1-RFB in rad3-null cells was dependent on both Exo1 and Rad17 (P<8.2×10⁻⁶). These data strongly support our model that efficient homologous-recombination-dependent restart does not require checkpoint activation, but frequently occurs upstream of the site of initial arrest when resection is extensive.

It can also be predicted that the generation of ssDNA behind the collapsed fork would increase opportunities for homologous recombination to occur erroneously upstream of the site of the initial fork arrest. To establish whether the increased DNA processing behind the fork increased non-allelic homologous recombination in this region, we turned to a direct repeat recombination assay (Ahn et al., 2005). It is proposed that recombination between the direct repeats requires nuclease activities to resect nascent strands until a homologous region is exposed as ssDNA (Sun et al., 2008). In this system, two ade6 heteroalleles are positioned as direct repeats that flank a his3⁺ marker and a single RTS1 barrier (Fig. 5A). Replication is predicted to run from right to left at this locus (Heichinger et al., 2006) and, consistent with this, it has been demonstrated that the RTS1 barrier orientation must arrest right-to-left forks to significantly elevate recombination rates (Ahn et al., 2005).

Ahn and colleagues assayed recombination in the presence of constitutive rtf1 expression (Ahn et al., 2005). In order to regulate fork arrest, we combined a thiamine-repressible nmt41-rtf1 allele with their ade6-heteroallele locus and scored ade6 recombination in the wild-type, rad3- and exo1-null backgrounds at T₀ (no induced arrest) and after 48 hours either with (arrest ‘off’) or without (arrest ‘on’) thiamine (Fig. 5B). It should be noted that, under the ‘arrest off’ conditions (+ thiamine), the cells retain a low, but significant, level of fork arrest (Lambert et al., 2010; Lambert et al., 2005), and thus these conditions do not fully reflect recombination in the complete absence of arrest. Nonetheless, rad3-null mutants in ‘arrest on’ conditions showed elevated levels of recombination (P≤0.0019) when compared with those of rad3⁺ cells (18.9 versus 11.79 recombinants per 10³ cells), and exo1-null cells show significantly reduced levels (P≤0.0004) of recombination (5.06 versus 11.79 recombinants per 10³ cells). Importantly, concomitant deletion of exo1 in the rad3-null background reduced the amount of recombination in the ‘arrest on’ conditions to levels approaching those of the exo1-null single mutant (6.88 versus 5.06 recombinants per 10³ cells). These data suggest that RTS1-induced ade6 heteroallele recombination is suppressed by Rad3^ATR and promoted by Exo1 activity, and that Rad3^ATR is inhibiting Exo1-dependent recombination.

The role of the intra-S-phase checkpoint in maintaining arrested replication forks in a replication-competent state is well documented, and the underlying mechanisms are beginning to be unravelled. In this report, we identify a new function for the intra-S-phase checkpoint at collapsed replication forks. Specifically, we show that the recruitment of RPA, Rad52 and Rad51 to the site of a collapsed fork is distinctively controlled by Rad3^ATR and the 9-1-1 checkpoint clamp, through the coordination of Exo1- and MRN-dependent resection (Fig. 5C). This checkpoint regulation of DNA processing acts to limit the extent of local replication errors that occur as a consequence of homologous-recombination-dependent replication restart. Moreover, our work reveals a role for the checkpoint sensors, independent of the downstream kinases, in limiting replication-induced genome instability in response to a chronic replication stress, thus contrasting with the classical analysis of checkpoint activation in response to acute replication stress.

An active replisome moves with the fork and closely couples DNA synthesis to the activity of the replicative helicase (Errico and Costanzo, 2012). The current model is that, if polymerisation is perturbed, the helicase initially moves ahead of the polymerases to expose an additional ∼100 bp of ssDNA (Sogo et al., 2002). This promotes the stimulation of Rad3^ATR and local activation of the intra-S-phase checkpoint. The checkpoint kinases (in S. pombe, Rad3^ATR and Cds1^Chk2) subsequently phosphorylate a range of replication and repair proteins. This protects the fork from collapse and retains the replisome in an active conformation (De Piccoli et al., 2012) while, at the same time, arresting cell cycle progression. By contrast, if helicase activity, as opposed to polymerisation, is perturbed, the initial exposure of ssDNA does not occur, the intra-S-phase checkpoint is not activated, the replisome cannot be held in an active conformation and the fork will collapse (Lambert and Carr, 2013a).

By analogy with the Escherichia coli Tus-ter site-specific RFB (Bastia et al., 2008) and the Reb1-dependent barrier at the S. pombe rDNA locus (Biswas and Bastia, 2008), we speculate that forks arrest at RTS1 because the replicative helicase is directly inhibited by the RTS1-associated proteins. Thus, because the helicase cannot move ahead of the polymerases, ssDNA is not formed, Rad3^ATR is not activated, and the fork collapses. By analysing the association of ssDNA-binding proteins with a specific collapsed replication fork, we have been able to show that the Rad3^ATR checkpoint is locally activated by fork collapse to ultimately control the activity of subsequent DNA processing events. Because forks arrested at RTS1 do not require the intra-S-phase checkpoint for their restart, we have been able to use our model systems to specifically examine the processing of DNA at the site of fork collapse, independently of the consequences of replisome stabilisation. Our data show that Rad3^ATR-dependent regulation of Exo1-dependent resection results in inappropriate DNA processing of the collapsed fork, but that this does not prevent homologous-recombination-dependent replication restart.

Previous work has identified Exo1 as a significant target of the intra-S-phase checkpoint when ATR is activated to stabilise intact replisomes; in S. cerevisiae, Rad53^Chk2 prevents Exo1-dependent replication fork breakdown in response to global replication stress (Cotta-Ramusino et al., 2005; Segurado and Diffley, 2008), and it has also been suggested that Rad53^Chk2 phosphorylates and regulates Exo1 at uncapped telomeres (Morin et al., 2008). In mammalian cells, Exo1 has been shown to be phosphorylated at 12 sites, of which three are induced by hydroxyurea treatment in an ATR-dependent manner (Bolderson et al., 2010; El-Shemerly et al., 2008). Thus, our identification of Exo1 as a key target of the ATR pathway at collapsed forks, as well as when forks are being stabilised, emphasises the importance of regulating this nuclease.

Mechanistically, we show that the PCNA-like 9-1-1 checkpoint clamp acts to promote MRN- and Exo1-dependent resection of DNA to extend a region of ssDNA upstream of the collapsed replication fork. This function for the clamp loader and clamp axis of the checkpoint is regulated by Rad3^ATR, but not by Chk1 or Cds1^Chk2. Thus, in the absence of Rad3^ATR function and the presence of a loaded 9-1-1 complex, repair-protein recruitment is likely to be increased. Unfortunately, we have been unable to generate reagents that are suitable for ChIP of Exo1. However, the likely explanation is that Rad3^ATR directly phosphorylates the clamp loader, clamp subunits and/or specific repair proteins recruited by the clamp (candidates include Mre11 and Exo1) to restrict resection. Such a mode of regulation would be fully consistent with the multiple phosphorylation events reported for these proteins.

Replication stress underlies a significant proportion of the genomic instability observed in model organisms and in cancer cells (Carr and Lambert, 2013; Lambert and Carr, 2013b; Segurado and Tercero, 2009). The intra-S-phase checkpoint is essential for maintaining the integrity of replication forks in the presence of such stress (Errico and Costanzo, 2012). Loss of the ability to maintain the replication-competent state of arrested or paused replication forks leads to their collapse, an event that has been linked to increased genome rearrangements in S. cerevisiae (Cha and Kleckner, 2002; Kaochar et al., 2010; Myung et al., 2001; Myung and Kolodner, 2002) and the expression of fragile sites in humans (Brown and Baltimore, 2003; Casper et al., 2002; Durkin et al., 2006). We have reported previously that, when a fork collapses, replication restart occurs through a ssDNA intermediate, not from a DSB (Mizuno et al., 2009). We found that fork restart is highly efficient, but is prone to non-allelic homologous recombination (NAHR), i.e. it has a ∼1–3% chance of restarting at the wrong place if a homologous sequence is nearby (Lambert and Carr, 2005; Lambert et al., 2010). We also demonstrated that, once restarted correctly, the restarted replication machinery is prone to replication slippage at sites of microhomology (Iraqui et al., 2012) or to performing a U-turn at closely spaced inverted repeats (Mizuno et al., 2013).

As a consequence of fork collapse and restart, NAHR (associated with the restart event) and replication slippage (associated with the restarted replisome) provide mechanisms for the genomic instability associated with replication stress. We examined the effects of the Rad3^ATR checkpoint on these mechanisms of genome instability that are specifically related to collapsed forks, as opposed to stalled forks, and their subsequent resumption of replication. Somewhat to our surprise, we found that recombination-mediated fork restart was independent of the checkpoint, and the frequency of the associated NAHR was unchanged. These data suggest that even limited association of recombination factors at collapsed replication forks is sufficient to ensure their efficient restart. However, the extensive resection we observed in the rad3-null mutant prompted us to explore whether this additional DNA processing resulted in more extensive genetic instability associated with the error-prone nature of the restarted replication fork and intra- or inter-sister-chromatid homologous recombination.

We observed that the extent of resection and subsequent recombination protein recruitment correlated directly with the promotion (rad3 null) or suppression (rad17 null) of inter- or intra-sister homologous recombination. These data implicate Rad3^ATR in limiting genome instability by regulating DNA metabolism and, thus, the activity of homologous recombination behind collapsed replication forks. We also observed a correlation between replication slippage after restart and the extent of resection. These data are entirely consistent with increased resection resulting in a larger region of DNA being replicated by an error-prone restarted fork. Thus, a function of Rad3^ATR is to limit the amount of DNA replicated by the restarted fork, which in turn reduces the likelihood of associated genetic instability.

It has become clear that the intra-S-phase checkpoint, acting through Rad3^ATR and Cds1^Chk2 in S. pombe, Mec1^ATR and Rad53^Chk2 in S. cerevisiae or ATR and Chk1 in human cells, prevents replication forks collapsing catastrophically, in part by phosphorylating replisome components and specific proteins affecting DNA metabolism, such as Exo1 (reviewed in Errico and Costanzo, 2012; Segurado and Tercero, 2009). A recent report also showed that moderate replication stress in ATR-depleted mammalian cells results in MRN-dependent ssDNA accumulation, the chromatin association of checkpoint sensors and the expression of common fragile sites (Koundrioukoff et al., 2013). Here, using a site-specific replication-arrest system, we have dissected the role of Rad3^ATR in fork stabilisation and checkpoint activation away from the roles it plays after the fork has collapsed. We reveal a subtle role for the core checkpoint kinase, ATR, and the 9-1-1 clamp in regulating recombination protein association at collapsed forks (Fig. 5C). We closely correlate this regulation with the restraint recombination and speculate that that it provides one of several roles by which ATR maintains genome stability.

The RuraR locus (Lambert et al., 2005) plus the uraR and Rura loci (Lambert et al., 2010) used in this study have been described previously, and the analysis of recombination outcomes was performed as described previously. Checkpoint deletions and alleles for the tagged proteins used were created and introduced by using standard molecular and genetic techniques (Bähler et al., 1998; Moreno et al., 1991; Watson et al., 2008). The strains used are listed in supplementary material Table S2. The ade6-M375 int::pUC8/his3⁺/RTS1(A2)/ade6-L469 locus was a gift from Matthew Whitby (Ahn et al., 2005). All strains were grown in 30 µM thiamine where indicated. The origins near RTS1 were renamed as ori3006/7. We have previously used ars3003/4 (Miyabe et al., 2011) and ars3004/5 (Lambert and Carr, 2005; Lambert et al., 2010; Mizuno et al., 2009; Mizuno et al., 2013).

Replication intermediates were analysed and quantified by 2DGE, as reported previously (Lambert et al., 2010). Briefly, zymolyase-treated cells were embedded in an agarose plug, treated with proteinase K and washed several times in Tris-EDTA. After restriction digestion by AseI, replication intermediates were enriched on benzoylated naphthoylated DEAE (BND)–cellulose columns, precipitated and separated by 2DGE using 0.35% and 0.9% agarose for the first and second dimensions, respectively. Quantification of replication intermediates was performed using a phosphor-imager (Typhoon-trio) to detect ³²P signal. Briefly, fork-termination and joint-molecules signals were quantified as the percentage of stalled fork signal. Chromosomal rearrangements were analysed by pulsed-field gel electrophoresis or Southern blotting as reported previously (Lambert et al., 2010; Lambert et al., 2005).

ChIP was performed as described previously (Lambert et al., 2005), but with sonication performed with a Diagenode Bioruptor at high setting (7 cycles; 30 seconds on +30 seconds off) to achieve a fragment size of 200–300 bp. The distance of primer pairs away from the RTS1 RFB is indicated on Fig. 2A and Fig. 2C. The primer sequences are given in supplementary material Table S1. Enrichment was normalised to an internal control (ade6 locus). Anti-GFP (rabbit polyclonal, Invitrogen), anti-Rpa2 (also known as Ssb2) [rabbit polyclonal; a gift from Hisao Masukata (Osaka University, Osaka, Japan)] or anti-human-Rad51 (H-92 rabbit polyclonal, Santa Cruz Biotechnology) were used at 1∶300, 1∶500 or 1∶100, respectively. Immunocomplexes were precipitated with Protein G Dynabeads (Invitrogen). For each ChIP experiment, wild-type and mutated strains were analysed in parallel. The data (except those shown in Fig. 1D) represent the relative enrichment of immunoprecipitated proteins in a given mutant relative to the enrichment observed in the corresponding wild-type control strain, either when the RTS1-RFB is active (arrest ‘on’) or inactive (arrest ‘off’).

Red colonies (ade⁻ cells) were picked from agar plates containing low adenine and no histidine, and were inoculated into 10 ml of rich medium, followed by overnight incubation. Cells were washed and split into two cultures each of 10 ml of Edinburgh Minimal Media (EMM) containing excess adenine and histidine, with or without thiamine. After 48 hours of logarithmic growth, cells from each culture were plated onto YE agar plates containing excess guanine. Cells were concurrently plated onto non-selective medium to determine the number of viable cells. After growth for 3 days, colonies from the plates lacking adenine were counted, to calculate the frequency of ade⁺ recombinants. Each experiment represents a median of 11 individual plates, and statistical significance was calculated by using Student's t-test.

Replication slippage was scored using the reporter allele ura4-sd20, which contains a duplication of 20 nt flanked by 5 nt of microhomology, as described previously (Iraqui et al., 2012). DNA synthesis associated with homologous-recombination-dependent fork restart is error prone, liable to replication slippage leading to the restoration of a functional ura4⁺ gene and, thus, an induction of ura⁺ colonies. Several single 5-FOA^R colonies were grown independently on uracil-containing plates with or without thiamine for 2–3 days, and then inoculated into uracil-containing medium with or without thiamine for 2 days at 30°C. Appropriate dilutions were plated onto supplemented minimal medium and uracil-free plates. Colonies were counted after incubation at 30°C for 5–7 days, and the frequency of ura⁺ colonies was determined. Statistical significance was determined using the nonparametric Mann-Whitney U test.

We thank Johanne M. Murray and Eva Hoffmann (both University of Sussex, Brighton, UK) for helpful suggestions, and members of the Carr and Murray laboratories for discussion and encouragement.