During DNA replication, DNA polymerases carry out DNA synthesis in a semi-conservative manner to produce two copies of the existing double helix. Many additional factors work concurrently with the polymerases to ensure the high fidelity and processivity of DNA replication. To guarantee that only one round of DNA replication occurs in each cell cycle, pre-replicative complexes are formed in G1 phase to “license” certain genome positions as origins. In S-phase, CDKs activate these “licensed” complexes to fire only once. Upon licensing the MCM complex is converted to the active CMG (Cdc45-MCM-GINS) helicase to unwind DNA and to form replication forks (Masai et al., 2010). Elongation factors are then recruited, many of them via interactions with the core homo-trimeric sliding clamp PCNA (Mailand et al., 2013). The CMG helicase moves ahead of the forks and generates DNA supercoiling ahead of the fork as well as catenation of the newly synthesized DNA strands behind the fork. Topoisomerases cut double stranded DNA to relieve the supercoiling and to catalyze the decatenation of the DNA duplexes (Schalbetter et al., 2015).

The term “replication stress” refers to various impediments, which cause the slowing down or the pausing of the replication forks (Zeman and Cimprich, 2014; Khurana and Oberdoerffer, 2015; Nikolov and Taddei, 2016). Replication stress can be caused by DNA damage, by deprivation or imbalance of nucleotides or by insufficient supply of histones. In addition, replication stress can be induced by tightly bound non-histone proteins, the collision of replication and transcription complexes or secondary DNA structures (Lambert and Carr, 2013; Zeman and Cimprich, 2014; Khurana and Oberdoerffer, 2015). The latter impediments have been identified as transient replication pausing sites during the normal course of DNA replication (Ivessa et al., 2003; Makovets et al., 2004; Azvolinsky et al., 2009).

At transient pausing sites the forks are stabilized against topological stress and collapse by at least two factors, the FPC and the cohesin-like Smc5/6 complex (Menolfi et al., 2015; Bastia et al., 2016). In addition, components of the replisome and the histones in the vicinity of the stalled forks undergo specific PTM (Baker et al., 2010; Szilard et al., 2010; Bastia et al., 2016). DNA helicases are then recruited (or activated) to aid the removal of the impediments and help the resumption of elongation. In S. cerevisiae, the Rrm3p helicase is engaged in the displacement of tightly bound non-histone-proteins (Ivessa et al., 2003) while Pif1p and Srs2p are involved in the unwinding of G4 quadruplexes or DNA hairpins (Anand et al., 2011; Paeschke et al., 2013; Leon-Ortiz et al., 2014). These helicases are not essential as their destruction does not prevent the completion of S-phase; however, their loss leads to extended stalling and increased mutation rates (Putnam et al., 2012; Ang et al., 2016). Similar DNA helicases (FANCJ, WRN, BLM) are found in metazoans and again are linked to higher mutation rates and cancer (Sarkies and Sale, 2012a). Hence, the compromised pausing of replication forks generates mutations and genome instability.

Replication pausing also seems to interfere with the stable transmission of chromatin (Sarkies and Sale, 2012b; Khurana and Oberdoerffer, 2015), but our understanding of these processes is limited. For example, multiple protein-binding sites and G4 structures contribute to frequent fork pausing in the subtelomeric regions of S. cerevisiae that is exacerbated upon deletion of the RRM3 gene (Ivessa et al., 2003; Makovets et al., 2004; Azvolinsky et al., 2006). Interestingly, subtelomeric genes also undergo spontaneous epigenetic conversions, a phenomenon referred to as Telomere Position Effect (Gottschling et al., 1990; Yankulov, 2013). Similarly, the rRNA gene array of S. cerevisiae contains RFB (Replication Fork Barrier) sites, which arrest forks and prevent their collisions with transcription complexes (Kaplan and Bastia, 2009). The rRNA gene repeats are also subjected to spontaneous epigenetic conversions (Kaplan and Bastia, 2009; Yankulov, 2013). In chicken cells, epigenetic instability has been linked to the pausing of replication at G4 quadruplex-forming structures and can be exacerbated by depletion of nucleotides and by the deletion of the FANCJ gene (Schwab et al., 2013, 2015; Schiavone et al., 2014; Papadopoulou et al., 2015). It appears from this information that the pausing of replication forks is linked to epigenetic conversions, but the underlying mechanisms are not well understood.

Before we address the possible effect of transient replication pausing on the preservation of chromatin, we will briefly summarize the current knowledge on the transmission of epigenetic marks during the advancement of the forks. The transmission of the methylation marks on DNA has been reviewed by others (Alabert and Groth, 2012; Almouzni and Cedar, 2016) and will not be discussed here. The disassembly and reassembly of nucleosomes is mediated by a complex network of histone chaperones, nucleosome remodelers and histone modifying enzymes. Many of these factors work in close contact with the basal replication machinery (Alabert and Groth, 2012; Gurard-Levin et al., 2014; Almouzni and Cedar, 2016). The overall process of disassembly, histone chaperoning and reassembly seems highly conserved between eukaryotes. However, the histone modifying enzymes, PTM and the timing of their restoration differ between organisms (Alabert and Groth, 2012; Annunziato, 2015). In the following sections we focus on the interactions between histones and their chaperones and discuss how replication pausing can alter the reassembly of nucleosomes.

The advancing CMG helicase generates extensive negative supercoiling of the replicated DNA ahead of the fork. In turn, this torsional stress can force fork rotation and double-stranded catenanes behind the fork (Figure 2A). During elongation, the DNA supercoiling is relieved by topoisomerases, while fork rotation is suppressed at least in part by the so-called FPC (Figure 2A). FPC is composed of three proteins, Mrc1p, Tof1p and Csm3p (Bando et al., 2009; Komata et al., 2009) and is believed to be part of the elongating replisome (Gambus et al., 2006; Komata et al., 2009). TOF1 and CSM3 are required for the stabilization of replication forks in the presence of hydroxyurea (Bando et al., 2009) and at sites of tightly bound non-histone proteins (Mohanty et al., 2006; Tourriere and Pasero, 2007). Mrc1p is required for the normal progression rate of DNA replication forks and for the activation of checkpoints, but is dispensable for pausing (Tourriere et al., 2005; Mohanty et al., 2006; Hodgson et al., 2007; Petermann et al., 2008). Tof1p-Csm3p directly associate with the CMG helicase via Mcm2p while the association of Mrc1p is dependant on Tof1p-Csm3p (Bando et al., 2009) (Figure 2B).

A recent study has demonstrated increased catenation of the fork at several types of replication pausing sites (Schalbetter et al., 2015). It has been suggested that these effects are caused by the interference of the impediment (a tightly bound protein or a secondary DNA structure) with the activity of the topoisomerases. At these positions, Tof1p-Csm3p prevents further topological stress until the impediment (in this situation a non-histone protein) is removed by the displacement helicase Rrm3p (Figure 2B). In support of this idea, replication pausing is diminished upon the deletion of TOF1 or CSM3 (Hodgson et al., 2007) and fork rotation is exacerbated in the absence of TOF1, CSM3 and RRM3 (Schalbetter et al., 2015). Other factors are also involved in the coordinated activity of Rrm3p and Tof1p-Csm3p. It has been recently found that the cohesin-like Smc5/6 complex is enriched at sites of replication pausing and overlaps with sites of Rrm3p enrichment (Menolfi et al., 2015; Branzei and Menolfi, 2016). Notably, it was shown that Smc5/6 cooperates with Rrm3p and counteracts the prolonged pausing caused by Tof1p-Csm3p (Menolfi et al., 2015). Hence, the torsional fork rotation can lead to a quick resolution of the arrest, but also to complications due to fork distortion and catenations.

In summary, replication stalling is accompanied by topological distortion of the forks, which could lead to temporal perturbation of histone transmission and assembly. We discuss this possibility in Section “What Happens When the Fork Stalls?”.

Dbf4-Dependent Kinase is an essential kinase, which phosphorylates several MCM proteins (Figure 2B). Some of these phosphorylation events are critical for the firing of the origins (Sheu and Stillman, 2010). However, it has long been speculated that the function of DDK extends beyond the control of origins (Duncker and Brown, 2003). In budding yeast, DDK phosphorylates Tof1p-Csm3p and subunits of the MCM helicase at the stably arrested replication forks in the rRNA gene arrays (Kaplan and Bastia, 2009; Bastia et al., 2016) (Figure 2B). In turn, the phosphorylated Tof1p-Csm3p associates with MCM to inhibit its helicase activity and to counteract the activity of Rrm3p (Cho et al., 2013; Bastia et al., 2016) (Figure 2B). It is not known if the DDK-dependent phosphorylation of MCM is identical at origins and at the pausing site or if the phosphorylation of MCM precludes its possible activity as a histone chaperone (Clement and Almouzni, 2015; Huang et al., 2015). In human cells the homolog of Mrc1p, Claspin, directly associates with DDK (Yang C.C. et al., 2016). At this point it is uncertain if Claspin, which is recruited to stalled forks (Masai et al., 2010), could also engage DDK during fork pausing.

Dbf4-Dependent Kinase also phosphorylates CAF-I in both budding yeast and human cell extracts (Keller and Krude, 2000; Gerard et al., 2006; Jeffery et al., 2015) (Figure 2B). In human cell extracts the p150 subunit of CAF-I can form dimers (Gerard et al., 2006). The phosphorylation of p150 by DDK prevents this dimerization and stimulates its binding to PCNA (Gerard et al., 2006). However, both monomeric and dimeric forms of p150 seem to be a requirement for CAF-I activity. On the other hand, in S. cerevisiae the phosphorylation of p150 is not necessary for the loading of CAF-I to chromatin, suggesting that these events can take place at a later stage of DNA replication (Jeffery et al., 2015).

Finally, DDK phosphorylates the Histone H3T45 residue (Baker et al., 2010). This phosphorylation peaks in S-phase and H3-T45A mutations reduce the resistance of cells to hydroxyurea and camptothecin (an inhibitor of Topoisomerase I), but not to DNA alkylating agents (Baker et al., 2010). Importantly, H3T45 phosphorylation is not required for the initiation of DNA replication, but is necessary at a later step (Baker et al., 2010). Histone H2A is also phosphorylated at RRM3-dependent transient replication pause sites, but the actual kinase has not been identified (Szilard et al., 2010) (Figure 2B). Interestingly, in human cells gamma-H2AX accumulates at hydroxyurea-induced pause sites long before fork collapse and DNA damage, suggesting that this modification could be a regular event at transiently paused forks and not restricted to the DNA damage response (Sirbu et al., 2011).

It is conceivable that DDK is recruited to paused forks where it phosphorylates H3T45, Tof1p, Cac1p and the MCM complex (Figure 2B). The phosphorylation of Tof1p and MCM could prevent the premature resolution of pausing (Bastia et al., 2016) and the trafficking of old H3/H4 tetramers (Clement and Almouzni, 2015; Huang et al., 2015). DDK can also potentially alter the activity of CAF-I (Gerard et al., 2006; Jeffery et al., 2015). H3-T45 is positioned at the site where DNA enters and leaves the nucleosome (Baker et al., 2010). Its phosphorylation will almost certainly reduce the nucleosome-DNA contact and could facilitate the resumption of elongation, but could also affect the disassembly of the nucleosomes. All these events can promote a substantially different mode of H3/H4 handling and reassembly at the stalled fork.

The first key consideration is the fate of the old H3/H4 histones at paused forks. We know that the majority of old H3/H4 tetramers are not split during DNA replication (Xu et al., 2010; Katan-Khaykovich and Struhl, 2011) and that MCM alone can act as a chaperone for the H3/H4 tetramer (Huang et al., 2015). However, ASF1 can only interact with H3/H4 dimers and H3/H4 dimers in complex with CAF-I have been identified in nuclear extracts (Tagami et al., 2004; Natsume et al., 2007). Importantly, during replication stress ASF1 is found in complex with H3/H4 with typical parental PTMs (Groth et al., 2007; Jasencakova et al., 2010). Therefore, we have to consider that ASF1 may have different modes of action during elongation and at paused forks. Consequently, two models are possible. Both of them envisage some loss of epigenetic marks.

The simplest model suggests that during elongation whole tetramers are ferried behind the fork, but upon slowing-down and subsequent pausing the “old” tetramers are split into dimers by ASF1 (Figure 3A). This notion comes from the observed increase in the abundance of ASF1 in complex with “old” H3/H4 during replication stress (Jasencakova et al., 2010). The old H3/H4 dimers could be transferred behind the fork, complemented by new H3/H4 dimers and deposited on the new DNA strands as a mixed tetramer. In this case, the marks from the old histones could by copied on the new ones within the same nucleosome (Figure 3A). This model is in agreement with the previously proposed assembly of “mixed” nucleosomes in (Clement and Almouzni, 2015; Almouzni and Cedar, 2016) (Figure 1B). Under these conditions, the mere slowing-down of the fork could decrease the supply of old histones and enhance the assembly of nucleosomes from new histones. This situation can be exacerbated upon complete pausing.

Another possibility for loss of old H3/H4 could be the differential regulation of the MCM helicase during elongation and upon stalling (Figure 3B). In Section “DNA Replication” and Figure 2B we described the DDK-dependent phosphorylation of the MCM helicase and Tof1p-Csm3p at arrested replication forks. Recent studies also suggest that MCM2 can act as a chaperone for the H3/H4 tetramer (Clement and Almouzni, 2015; Huang et al., 2015). The phosphorylated Tof1p-Csm3p binds to MCM and is known to inhibit its helicase activity (Cho et al., 2013; Bastia et al., 2016), but it is unclear if the putative MCM H3/H4 chaperone activity is affected. It is open to conjecture that if MCM is responsible for the ASF1/CAFI-independent transmission of H3/H4 tetramers behind the fork (Clement and Almouzni, 2015; Huang et al., 2015), the loss of this activity will promote the assembly of nucleosomes from new histones only (Figure 3B).

In the second model, ASF1 has a similar action during elongation and pausing. It transiently destabilizes old H3/H4 tetramers, but they quickly re-form before being deposited (randomly or not) on one of the new strands (Figure 4). The increased abundance of ASF1 complexed with old H3/H4 during replication stress represents this transitional state of the old H3/H4 histones. Under this scenario, the slowing-down of the fork and/or the inhibition of “old” H3/H4 transmission would still promote the deposition of new H3/H4, but an additional complication would exist: the new H3/H4 tetramers need to copy the existing marks from another nucleosome with an old H3/H4 tetramer. This nucleosome could be the neighboring one on the same strand or the corresponding one on the other strand (Figure 4). Intuitively, one would expect that the latter mechanism is sensitive to topological distortion and could be less conservative in the preservation of epigenetic state. As already described, transient fork pausing is associated with rotation that induces topological alterations (Schalbetter et al., 2015). These alterations could temporarily suspend the communication between the two nucleosomes, leading to loss of the pre-existing histone marks at that position (Figure 4).

Both models assume that the pausing of the fork affects the transmission of existing histone marks and early “seeding” PTMs that take place in the immediate wake of the fork. These seeding marks could subsequently influence the maturation of chromatin and the establishment of the marks with delayed kinetics of reconstitution (Scharf et al., 2009a; Alabert et al., 2015).

It has been shown that both ASF1 and CAF-I remain associated with chromatin under conditions of transient arrest of DNA replication by hydroxyurea (Jasencakova et al., 2010; Petermann et al., 2010; Sirbu et al., 2011). While it is not given that this situation is identical at all kinds of pausing sites, it is likely that the key histone chaperones remain associated with the replisome at stalled forks. What would be the role of CAF-I at paused forks?

As discussed, it is assumed that CAF-I is responsible for the replication-coupled assembly of both old and new H3/H4 histones, but it is also possible that it works with new H3/H4 only (Sarkies and Sale, 2012b; Almouzni and Cedar, 2016). Consequently, two closely related mechanisms can describe the role of CAF-I at paused replication forks. Both mechanisms are based on the assumption that the increased abundance of ASF1 in complexes with old H3/H4, which are observed during replication stress (Jasencakova et al., 2010), reflects the decreased supply of old histones behind the fork. If CAF-I intercepts the old H3/H4 dimers from ASF1 (Figures 3A,B) and mixes them with new histones, the limiting supply of old histones would promote the assembly of nucleosomes with new histones only. If CAF-I does not intercepts old histones, it can continue to assemble new H3/H4 tetramers while the transmission of old H3/H4 by a CAF-I independent mechanism is temporarily suspended (Figures 3A,B). Both scenarios predict that in the absence of CAF-I the deposition of new histones, and concomitantly the probability for an epigenetic change, will be decreased. This conjecture is supported by the demonstrated reduction of epigenetic conversions upon the destruction of CAF-I in S. cerevisiae (Jeffery et al., 2013; Wyse et al., 2016).

It is also possible that the increased abundance of ASF1 complexes with old H3/H4 does not reflect a diminished delivery of old histones. In this situation the observed effects of CAF-I on epigenetic conversions could be explained by its altered activity upon stalling of the fork (Figure 2B). For example, we know that DDK phosphorylates CAF-I. In budding yeast this phosphorylation is not necessary for the association of CAF-I with chromatin, suggesting that DDK may act on CAF-I at a post-initiation event (Jeffery et al., 2015). Given the fact that DDK phosphorylates components of the stalled fork (Tof1p, MCM, H3T45, see above), it is not inconceivable that DDK specifically phosphorylates CAF-I at paused forks (Figure 2B). One possibility is that this phosphorylation event stimulates CAF-I activity toward new histones.

Another source of CAF-I modulation could be the PCNA-interacting proteins at stalled forks. Similarly to Cac1p, the budding yeast Rrm3p and Sgs1p helicases contain a PIP for the direct association with PCNA (Anand et al., 2011; Wyse et al., 2016). PCNA forms a homo-trimeric clamp capable of three PIP-mediated contacts, suggesting a complex communication between PCNA and its interacting partners (Mailand et al., 2013). It is not clear if Rrm3p is traveling with the fork or if it is recruited upon pausing (Calzada et al., 2005; Azvolinsky et al., 2006), however, it is assumed that it is activated on the lagging strand only (Ivessa et al., 2002) (Figure 5A). It is possible that the recruitment/activation of Rrm3p could alter the contact of PCNA with CAF-I or altogether displace CAF-I from PCNA (Figure 5A). This scenario puts forth the possibility that at paused forks Rrm3p imposes a different mode of reassembly on the lagging and leading strands and can grant the chance for a post-replicative epigenetic change on one of them only. In support of this idea, we have recently demonstrated that the deletion of RRM3 suppressed epigenetic changes in S. cerevisiae (Wyse et al., 2016). The involvement of Rrm3p and the hypothesized suppression of CAF-I on the lagging strand suggests that the existing chromatin state can be altered on the leading strand where CAF-I continues to operate, but preserved on the lagging strand where its activity is reduced (Figure 5A).

Preferential loss of histone marks on one of the two strands can also be caused by the uncoupling of the synthesis of DNA on the leading and lagging strands as suggested in (Sarkies and Sale, 2012b) (Figure 5B). For example, a G4 quadruplex can form on the leading strand and arrest DNA synthesis on it while the CMG helicase continues to operate. Under these conditions the lagging strand will continue to initiate close to the fork and will therefore be exposed to the pool of disassembled “old” H3/H4 histones. On the leading strand DNA synthesis can restart downstream of the G4-block and the resulting single-strand gap could be filled-in at a later stage. However, this later stage of DNA synthesis would take place away from the source of old histones and CAF-I will reassemble nucleosomes with new histones only (Sarkies and Sale, 2012b; Papadopoulou et al., 2015) (Figure 5B, bottom).

In both cases, the dissimilar assembly of chromatin on one of the two strands would provide a mechanism where replication pausing would enhance the loss of old histones on the leading strand and thus generate the possibility of a genetic change on one of the chromatids only. Stem cells could use such mechanisms to maintain pluripotency and at the same time to generate differentiating progeny.