Mechanisms of direct replication restart at stressed replisomes

DNA replication is a highly regulated process during which the genetic information of the cell is duplicated. The replication machinery encounters many challenges that threaten the transmission of genetic information to daughter cells. The term “replication stress” encompasses the diverse ways that progression of the replisome may be hindered, slowed or completely stalled (reviewed in [1]). The inability to deal with these challenges in a timely manner destabilizes the genome, which can lead to diseases, such as cancer, hematopoietic stem cell failure, inappropriate innate immune responses, and aging. Targeting DNA replication and induction of replication stress response has also been a mainstay of cancer therapies. This review will discuss sources of replication stress that threaten the fidelity of DNA replication and the mechanisms that mammalian cells use to efficiently restart stalled replication. For mechanisms of replication fork restart in prokaryotes, we direct readers to the following reviews [2, 3].

Sources of replication stress and S-phase specific DNA damage response

There are many endogenous lesions that may pause the replicative polymerases and block DNA replication. They include unrepaired damaged bases or base modifications, transcribing RNA polymerases, DNA with secondary structures, RNA-DNA hybrids, intrastrand and interstrand crosslinks (ICLs), and DNA-protein complexes (DPCs). While some lesions, such as ICLs and DPCs, were thought to also block the replicative helicase, CDC45-MCM2–7-GINS (CMG), evidence is accumulating that the CMG helicase can bypass these lesions [4–6], though the DNA polymerases may remain blocked.

Replication stress can also be exogenously induced by UV radiation, which induces intrastrand crosslinks, and and by variety of compounds, including hydroxyurea (HU), which depletes nucleotide pools by inhibiting ribonucleotide reductase [7–9], aphidicolin, which directly inhibits B family polymerases [10]; or Mitomycin C (MMC), and cisplatin, which both induce ICLs [11, 12]. Responses to these insults during DNA replication depend on the offending lesion and the specific DNA repair pathways necessary to remove it. However, besides lesion removal, the DNA surrounding the lesion needs to be replicated to faithfully transfer the DNA to the next generation.

A universal response to replication fork stalling is the formation of stretches of single-stranded DNA (ssDNA). This ssDNA may be a result of functional polymerasehelicase uncoupling that occurs when the polymerase stalls and the replicative helicase continues to unwind dsDNA [13]. RPA binds ssDNA and serves as a recruitment platform for many factors, including Ataxia Telangiectasia Mutated and Rad3 Related-interacting Protein (ATRIP), which facilitates the association of ATR [14], a member of the phosphoinositide-4,5-bisphosphate 3- kinase (PI3K)-related protein kinases (PIKKs) family, which also includes ataxia-telangiectasia mutated (ATM) and DNA dependent protein kinase catalytic subunit (DNA-PKcs) (reviewed in [15]).

The persistence of ssDNA facilitates downstream activation of ATR. Its activation is facilitated by multiple proteins, including DNA Topoisomerase 2-binding Protein 1 (TOPBP1) and Ewing’s Tumor-associated Antigen (ETAA1). The RAD9-RAD1-HUS1 (9-1-1) complex binds at the 5’-ended ssDNA-dsDNA junction and recruits TOPBP1 for activation [16, 17]. Long patches of ssDNA facilitate recruitment of ETAA1 (through RPA binding) [18, 19]. ETAA1 and TOPBP1 function in parallel pathways to activate ATR, but recent data suggest that ETAA1 is also important for controlling cell division processes unrelated to DNA damage response pathways, such as regulation of the spindle assembly checkpoint [20]. ATR orchestrates multiple downstream events in response to replication stress, including cell cycle arrest, global suppression of origin firing, DNA repair, and replication fork restart (reviewed in [15]). These downstream responses are largely dependent on the activation of CHK1 by ATR-dependent phosphorylation [21, 22].

Direct replication fork restart

Here, we will refer to direct replication restart as the ability to restart replication without disassembly of the replisome and without DNA breakage. Several pathways have been proposed to promote direct replication restart of the compromised replication fork, including re-priming downstream of the lesion, lesion bypass with the use of translesion synthesis (TLS) polymerases, or lesion bypass via template switching (TS) (reviewed in [23–26]). Fork reversal at a stalled fork may also assist in replication restart (Figure 1). It should be mentioned that the replication of a permanently stalled fork can be rescued by a replication fork traveling towards the lesion from a neighboring origin or activation of a “dormant origin” allowing for a passive DNA replication around the lesion [27] (Figure 1A). Alternatively, a stalled replication fork may be processed by structure specific nucleases and the resulting breaks are repaired using homology directed repair (HDR).

Figure 1. Mechanisms of direct replication fork restart defined as restart without disassembly of the replisome or replication fork breakage.

A compromised replication fork can be passively rescued by an adjacent ongoing replication fork or newly activated ‘dormant’ origin (A) [27]. Direct restart includes AEP primase-dependent re-priming downstream of the lesion (B), lesion bypass with the use of translesion (TLS) polymerases (C), and lesion bypass via template switching (TS) (D) (reviewed in [23–26]). Fork reversal generates a substrate, resembling a “chicken foot,” that allows for lesion bypass via TS and fork restoration (E) [120].

Replication restart can be assessed using a single-molecule DNA fiber assay in which replicating DNA is visualized using incorporation of nucleotide analogs followed by recognition of these analogs using immunofluorescently-labeled antibodies. Replication track lengths can be evaluated with or without drug treatments and replication restart can be studied by assessing whether replication resumes upon removal of the drug (measuring % of restarting forks) or comparing the replication tract length upon drug removal to the length before drug addition (measuring restart efficiency). Table 1 highlights the mammalian proteins that have been implicated in replication restart as judged by the replication fiber assays.

Table 1.

Mammalian factors implicated in replication restart as assessed by DNA fiber assays

Mammalian factors shown to promote restart
Factor	Agents	Defect*	Cell Type	Citation
Re-priming
PRIMPOL	UV	% Restart, Restart Efficiency	HeLa, MEFs	[40],[41]
PRIMPOL	HU	% Restart, Restart Efficiency	HeLa, MEFs	[41], [42]
PRIMPOL	BPDE	% Restart	U2OS	[53]
TLS
Pol κ	HU	% Restart, Restart Efficiency	293T, U2-OS, RPE-1	[67]
Rev1	UV	Restart Efficiency	XP-C fibroblasts (NER-deficient)	[183]
Pol η	UV	% Restart, Restart Efficiency	XP30RO fibroblasts (Pol η -deficient)	[66],[65]
TS
SMARCAL1	Aphidicolin	% Restart	U2OS	[185]
ZRANB3	HU	% Restart	U2OS	[122]
RECQ1	CPT	Restart Efficiency	U2OS	[132]
PARP1	HU	% Restart	U2OS	[186]
WRN	HU	% Restart	GM639, U2OS	[137],[133]
WRN	CPT	% Restart	WRN-deficient fibroblasts	[187]
DNA2	HU, CPT, MMC	% Restart	U2OS	[133]
RIF1	HU	Restart Efficiency	MEFs	[138]
BLM	HU	% Restart	GM639, PSNG13 fibroblasts (BLM-deficient)	[136], [137]
BLM	Aphidicolin	% Restart	PD20 fibroblasts (FANCD2-deficient)	[136], [188]
RAD51	HU	% Restart	U2OS, GM639	[119], [137], [177]
RAD52	HU (6 and 24 hour treatments)	% Restart	U2OS	[147]
Pathway Choice
USP1	HU	Restart Efficiency	HeLa	[189]
PCNA-Ub (K164R)	HU	% Restart	U2OS	[67]
FANCD2	HU	% Restart, Restart Efficiency	U2OS	[67]
Other Pathways
BAP1	HU	% Restart, Restart Efficiency	HT29	[166]
INO80	HU	% Restart, Restart Efficiency	HT29, PC3	[166], [167]
EXD2	HU	Restart Efficiency	HeLa	[169]
DDI1, DDI2	HU	% Restart, Restart Efficiency	U2OS	[170]
Mammalian factors that may inhibit restart
Factor	Agents	Defect	Cell Type	Citation
TS
ZRANB3	CPT, MMC, UV	Fork Slowing	U2OS	[124]
HLTF	HU	Restart Efficiency	U2OS, HCT116	[127], [190]
Pathway Choice
PCNA-Ub (K164R)	CPT, MMC	Fork Slowing	U2OS	[124]
UBC13	CPT, MMC, UV	Fork Slowing	HCT116	[124]
Mammalian factors shown to not play a role in restart by DNA fiber assays
Factor	Agents	No defect	Cell Type	Citation
Re-priming
PRIMPOL	HU	% Stalled	RPE hTERT	[54]
TLS
Pol η , Pol ι , Rev1, Rev7, Rev 3L	HU	% Stalled, Restart Efficiency	RPE-1, XP-C fibroblasts (NER-deficient)	[67], [183]
Pol κ	Aphidicolin	% Stalled	RPE-1	[67]
TS
RAD51	HU	% Stalled	RPE-1	[67]
BRCA2	HU	% Stalled	RPE-1, U2OS	[67], [177], [191]
RAD52	HU	Restart Efficiency	U2OS	[148]

There exist caveats to the fiber assay especially as it pertains to replication restart. To assess restart, replication forks must be completely stalled. A uniform block across all replication forks can be visualized upon incubation with high doses of HU or aphidicolin making it possible to assess restart upon removal of these drugs. The response upon incubation with lower doses of HU, aphidicolin, or use of other DNA damaging agents including UV, CPT, and MMC, may not fully stall replication, making it impossible to properly assess replication restart. Additionally, the majority of replication forks in cells treated with some agents, including MMC, will not be impeded by the lesion itself. Even with high doses of MMC, there are too few lesions to physically block most replication forks. Instead, changes at many forks will be driven by a poorly-understood global responses which includes fork reversal [28]. Without visualization of the forks stalled specifically at the lesion [5] one cannot assess lesion-specific restart at all.

Re-priming

Re-priming, the process of synthesizing a primer downstream of a lesion, was first proposed in studies from Escherichia coli that found an accumulation of ssDNA in nucleotide excision repair (NER)-deficient cells that continued to replicate following UV irradiation [29, 30]. Re-priming is an effective mechanism to resume synthesis directly at the replication fork, but it also leaves behind gaps of ssDNA that need to be repaired post-replicatively. Re-priming is dependent on the Archaeo-Eukaryotic Primase (AEP) superfamily of proteins, which are responsible for synthesizing new primers (Figure 1B). In mammals, there are only two identified members of the AEP superfamily: PRIM1 (the primase catalytic subunit of DNA Polymerase α (Pol α)-primase and PRIMPOL. The factors determining the usage of PRIM1 or PRIMPOL for re-priming may be complex, but a possible model is that usage differs depending on the location of the lesion in either the lagging- or leading-strand template.

Evidence suggests that re-priming occurs efficiently on the lagging-strand for lesion bypass. In bacteria, abasic sites on the lagging-strand do not impede replication in vitro [31, 32], with synthesis of new Okazaki fragments offering an efficient method of lesion bypass. In Saccharomyces cerevisiae, reconstitution of DNA replication on templates containing blocking lesions shows that lesions on the lagging-strand, but not the leading-strand, can be efficiently re-primed by Pol α-primase [33]. In Xenopus laevis egg extracts, Pol α-primase is hyper-loaded at stalled replication forks by TOPBP1 [34, 35], resulting in a slight accumulation of short nascent DNAs [36].

The first evidence of re-priming on the leading-strand comes from in vitro reconstituted DNA replication studies showing that the E. coli replisome has the inherent capacity to “skip” over multiple leading-strand lesions [37, 38]. Unlike bacteria, biochemical studies in yeast show that re-priming by the replication fork-associated proteins, specifically Pol α-primase, on the leading-strand is inefficient [33], although it can occur at a subset of stalled forks [33, 39]. Since these studies rely on biochemical reconstitution of DNA replication, it is possible that there exist cellular factors that promote efficient re-priming on the leading-strand in yeast.

In human cells, PRIMPOL has emerged as a primase and polymerase that can re-prime replication and its function seems to be the most relevant for leading-strand lesion bypass [40–44]. PRIMPOL is a 560 amino acid (AA) protein that contains an N-terminal AEP-like catalytic domain with the conserved I, II, III motifs and a UL52-like zinc finger that support its primase and polymerase activities [45, 46]. In addition to retaining the classical primase activity to synthesize primers using rNTPs, PRIMPOL is unique in its ability to also synthesize primers using dNTPs [40, 43]. The primase activity of PRIMPOL is dependent on a template T and on rATP or dATP as a starting nucleotide [40]. In vitro studies also demonstrate that PRIMPOL has TLS polymerase activity [40, 41, 43, 47–49], and exhibits low fidelity DNA synthesis, lacking 3’–5’ exonuclease activity [48–51].

The biological relevance of PRIMPOL activity remains an active area of research, especially in regard to re-priming at stalled replication forks. PRIMPOL is present in cytosolic, nuclear and mitochondrial matrix compartments of the cell [43]. Cells are viable without PRIMPOL, but exhibit a growth defect [41]. Loss of PRIMPOL induces sensitivity to replication stress-inducing agents, including MMS, cisplatin, and UV [40, 52], suggesting it is important in the replication stress response. In mammalian cells, PRIMPOL is crucial for restarting replication forks after exposure to UV radiation and benzo(a)pyrene diol epoxide (BPDE) treatment [40, 41, 53]. Data on the role for PRIMPOL in replication fork restart after treatment with HU is contradictory in the literature with the Wan et al. and Mouron et al. studies, but not the Ercilla et al. study, showing a role for PRIMPOL in replication restart after HU-induced stalling [41, 42, 54].

Separation of function mutations for PRIMPOL’s primase and polymerase activities have been helpful to decipher PRIMPOL’s biological role in vivo. A double mutation in the zinc-finger element in the CTD (C419G/H426Y) abolishes its primase activity while preserving polymerase function [41]. Mutation of the two catalytic carboxylate residues (D114A/E116A) disrupts both catalytic activities [40, 41, 43]. A point mutation of Y89D retains primase activity while reducing polymerase activity [55, 56]. A mutation that retains primase activity while completely abolishing polymerase activity has not yet been identified [57]. Complementation of PRIMPOL-deficient cells with mutant PRIMPOL cDNA constructs suggests that PRIMPOL’s re-priming function, not its TLS polymerase function, is biologically relevant in regard to the replication stress response. For instance, the expression of the primase-deficient PRIMPOL constructs cannot rescue phenotypes associated with loss of PRIMPOL, including increased cell death upon treatment with replication stress-inducing agents and the inability to restart replication after replication stress [41, 44, 52, 57, 58]. These data are consistent with a model where PRIMPOL restarts stalled replication forks by re-priming downstream of a lesion.

PRIMPOL does not seem to travel with the replisome as assessed by iPOND [59]. The absence of PRIMPOL as a component of the replisome may prevent unrestrained PRIMPOL activity that could be mutagenic and suggests PRIMPOL’s recruitment to the replication fork upon replication stress is regulated. PRIMPOL may be recruited to sites of replication stress through interaction with RPA [60]. It has also been shown that binding partners, such as POLDIP2, may help to enhance PRIMPOL activity [61]. Recent data suggest that PRIMPOL may be regulated transcriptionally, as PRIMPOL mRNA transcripts increase in response to treatment with cisplatin [58]. PRIMPOL’s removal from the replication fork could be facilitated by its weak DNA binding affinity and its zinc finger domain, which acts to negatively regulate PRIMPOL’s processivity [57].

Translesion Synthesis

Translesion Synthesis (TLS) is a process where the replicative polymerases are replaced with TLS polymerases to insert and extend bases opposite of a DNA lesion (Figure 1C). Mammalian TLS polymerases include Y-family polymerases (Pol η, Pol ι, Pol κ, Rev1), and B-family polymerase, pol ζ, composed of Rev3, Rev7, PolD2, and PolD3 [62]. Once the Y- and B-family polymerases bypass the lesion, they are replaced with replicative polymerases which resume high fidelity replication. Although TLS provides a mechanism to restart replication, it can be a mutagenic process. The error rate of TLS polymerases can be up to 1 in every 10² incorporated nucleotides compared to the replicative polymerases that make 1 error per 10⁴-10⁵ incorporated nucleotides (reviewed in [63]). It has been proposed that TLS can occur at or behind the replication fork, though the knowledge of which factors function directly at the replication fork is still limited (Table 1) [58, 64–67](reviewed in [68]).

Translesion bypass is a two-step process that includes: (1) insertion of a nucleotide by the Y-family polymerase and (2) extension by pol ζ [62, 69] and to a lesser extent, Pol κ [70–72]. During nucleotide insertion, the TLS polymerase accommodates the lesion in its active site to direct dNTP incorporation. Y-family polymerases can tolerate bulky lesions due to their capacious active sites and the absence of proofreading activity [73]. Y-family polymerases are remarkably diverse in their substrate specificity and respond to particular cognate lesions (reviewed in [74]). For example, biochemical evidence shows Pol η can synthesize across the cyclobutane pyrimidine dimers and intra-strand crosslinks [75–78], Pol κ across from polycyclic aromatic hydrocarbon-induced adducts, [79–81], and Pol ι across from 8-oxo-guananine [82–84], among others. Rev1 is a DNA-template dCMP transferase that can place a dCMP opposite of an abasic site [85]. TLS polymerases generally tolerate bulky adducts, but Pol κ has also been shown to be required for efficient replication fork restart in response to hydroxyurea [67]. The mechanistic details regarding how the correct TLS polymerase is selected for the target lesion are still being resolved, but this process could be dependent on the identity of the DNA lesion, specific interactions with other factors at the site of the lesion, or the availability of TLS polymerases present at a particular lesion. During extension, the TLS polymerase accommodates the abnormal lesion-dNTP duplex for further synthesis.

Though TLS can promote error-free repair, such as in the case of Pol η in repair of UV-induced lesions, the potential mutagenicity of the TLS polymerases necessitates their regulation, both at the recruitment and displacement level. Central to TLS regulation is the RAD18- and RAD6-dependent mono-ubiquitination of PCNA on Lys164, which increases PCNA’s affinity for TLS polymerases at stalled replication forks [86–88]. CRL4^Cdt2 and RNF8 can also catalyze mono-ubiquitination of PCNA, but may play a minor role [89, 90]. PCNA is de-ubiquitinated by USP1, USP7, and USP10-dependent mechanisms [91–93]. PCNA^K164R/K164R mutant mouse embryonic fibroblasts can still support 25–30% of TLS synthesis, suggesting there are other mechanisms to control the use of TLS polymerases [94]. This includes post-translational modifications of the TLS polymerases themselves [95, 96] and protein-protein interactions, such as that between PAF15 and Pol η or REV7 and TRIP13 [97, 98]. TLS polymerases have low processivity, which likely facilitates their removal and replacement by the replicative polymerases. For a more comprehensive overview of the mechanisms regulating TLS polymerase usage, readers are directed to a recent review [99].

Template Switching

Template switching (TS), first proposed in 1976 by Higgins et. al., is a homologous recombination-based method of lesion bypass that allows the replisome to use an alternate DNA template, most commonly the newly synthesized daughter strand on the sister chromatid, as a template instead of the damaged parental strand (Figure 1D) [100]. Replication restart by TS has the potential to be error-free, but can also be a template of complex genomic rearrangements when a nascent strand invades a distal replication bubble rather than its sister chromatid [101].

Site-specific replication fork barriers, such as the Replication Termination Sequence 1 (RTS1) site in Schizosaccharomyces pombe, have been useful to determine the molecular mechanisms for direct replication fork restart via TS, also referred to as replication-dependent recombination (RDR) or homologous recombination-restarted replication (HoRReR) [102–107]. Replication restart at the RTS1 occurs within S phase without the generation of a double-strand break (DSB) [103, 107, 108], but this type of restart can still be mutagenic [103, 109].

The data from experiments using the RTS1 site have generated two models of replication fork restart via TS. The first model proposes the stalled replication fork is regressed into a reversed fork (see below). In a second model, the replisome backtracks without the annealing of the nascent strands. Nucleolytic activities of Mre11-Rad50-Nbs1-Ctp1 and Exo1 can process either of these structures, exposing a ssDNA tail bound by RPA [110]. Rhp51^Rad51 can then promote invasion of the ssDNA tail into a homologous duplex to form a D-loop and subsequently, DNA Polymerase δ extends the 3’ end of the invading strand [102]. In yeast, Rad22^Rad52 functions to displace RPA and facilitates loading of Rhp51^Rad51. Rad22^Rad52 also possess its own strand annealing activity that can mediate Rhp51^Rad51-independent recombination [103]. DNA helicases, Pfh1 and Srs2, are also required for replication restart at the RTS1 [103, 106, 111].

Mechanisms of TS dependent on Rad51 and Rad52 also occur in Saccharomyces cerevisiae [112, 113], but experimental models often focus on TS that ensues behind the fork, necessitating earlier bypass of the lesion by another mechanism such as re-priming [114–116]. If TS exists as a co-replication mechanism, it is likely dependent on Rad5-generated intermediates [117, 118]. In mammalian cells, a RAD51-dependent mechanism of rapid fork restart devoid of generation of a DSB has been described [119]. It has been proposed that fork reversal and subsequent fork restoration largely accounts for TS-dependent replication restart.

Fork Reversal

A reversed fork is a replication intermediate that may form at a stalled replication fork. It forms by active remodeling of the fork DNA, during which the nascent strands anneal to each other to form a “chicken foot” structure [120] (Figure 1E). Reversed forks may assist in replication fork restart upon treatment with genotoxic agents without fork breakage. Fork reversal may stabilize the stalled replication fork until a distal origin can rescue it. Reversed forks may also provide the ability to restart replication by placing a blocking lesion in the context of dsDNA where repair pathways can recognize and remove the lesion resulting in efficient restart of replication. Finally, fork reversal may facilitate TS to bypass the lesion and subsequently restart replication.

Electron microscopy data indicate that fork reversal occurs at about ~25% of replication forks upon exposure to a variety of replication stress-inducing agents [121]. Fork reversal can be catalyzed by SWI/SNF Related, Matrix Associated, Actin Dependent Regulator Of Chromatin, Subfamily A Like 1 (SMARCAL1), Zinc Finger RANBP2-type Containing 3 (ZRANB3), and Helicase-like Transcription Factor (HLTF) [122–128], and FBH1 helicase [129]. It is promoted by RAD51 recombinase [121].

There likely exists a balance between replication fork reversal and restoration, but the details of these processes are still being resolved (Figure 2). It is possible that the replication fork remodelers, SMARCAL1, ZRANB3, HLTF, can restore a reversed fork into a replication fork as has been shown in vitro [130, 131]. While SMARCAL1 can restore both leading- and lagging-strand gapped forks, RPA stimulates and enforces the specificity of SMARCAL1 to only restore the fork when the nascent leading-strand is longer than the nascent lagging-strand [131]. ZRANB3 and HLTF show preference for restoration of forks where the nascent lagging-strand is longer than that of the leading-strand, but this activity is inhibited by RPA [130, 131]. This model also predicts that 5’–3’ nucleolytic resection is required for SMARCAL1-dependent replication fork restoration [131].

Figure 2. Mechanisms of replication fork restoration from a reversed fork.

To resume DNA replication, a reversed fork can be restored to allow for bypass of a lesion. (A) Fork remodeling proteins responsible for creation of a reversed fork, can also catalyze fork restoration in vitro. When the nascent leading-strand is longer than the nascent lagging-strand, SMARCAL1-dependent fork restoration is stimulated by RPA [131]. HLTF and ZRANB3 can only weakly catalyze this substrate, and while they can catalyze fork restoration when the nascent lagging-strand is longer than the nascent leading-strand, this reaction is inhibited by the presence of RPA [130, 131]. (B) The RECQ family helicases can also promote fork restoration [132]. RECQ1, which is inhibited by PARP1-dependent poly-ADP-ribosylation, has been shown to be active when the nascent lagging-strand is longer than the nascent leading-strand or both nascent strands are of equal length [132], but its activity has yet to be determined on substrates where the nascent leading-strand is longer than the nascent-lagging strand. RECQ1 also protects forks from nascent degradation by DNA2 [133]. WRN helicase function is epistatic with DNA2 nuclease to promote replication restart [133]. A reversed fork processed by WRN and DNA2 can be restored through branch migration or homology-directed restoration. BLM promotes fork restoration [134–137], although the details of its involvement need to be furthered studied. (C) Homology-directed restoration of replication may rely on classical homologous recombination factors: BRCA2 and RAD51. BRCA2 functions to load RAD51 onto the ssDNA overhang and subsequently RAD51 mediates strand-invasion and D-loop formation [119]. Whether RAD52 plays a role in fork restoration at an intact stalled replication fork is unknown.

The RecQ family of helicases, RECQ1, WRN, and BLM, may also promote replication fork restoration. RECQ1’s helicase activity promotes the restart of replication forks reversed by various types of replication stress [121, 132]. Poly-ADP-ribosylation by PARP1 inhibits RECQ1 fork restoration activity to prevent premature restart of the regressed forks [132]. Another RecQ family helicase, WRN, is also required for restoration of a reversed fork, and this restoration is dependent on DNA2-mediated processing of the reversed fork [133, 134]. RECQ1- and WRN-dependent restart likely function in independent pathways of replication fork restoration as RECQ1 limits DNA2-mediated nucleolytic resection shown to be required for WRN-mediated fork restoration [133]. BLM may also function in fork restoration as it can catalyze fork restoration in vitro and is required for fork restart in vivo [134–137]. Other factors may also contribute to directing RECQ-dependent fork restoration. For example, RIF1 was found to be enriched at stalled replication forks to prevent degradation of reversed forks by DNA2 and promote replication restart [138]. Although it has been proposed that this mechanism preserves the reversed fork as a substrate for RECQ1-dependent fork restoration, it remains to be tested whether RECQ1 and RIF1 are epistatic with each other. It is unclear exactly how the RecQ family helicases restart a reversed replication fork. RECQ1 may unwind the daughter strand duplex to promote branch migration-assisted re-establishment of a functional replication fork [132]. BLM and WRN may also promote branch migration and HDR [133, 137, 139–141]. WRNIP1, WRN and DNA Polymerase δ, may form a complex to promote DNA synthesis [142, 143].

Restoration of reversed forks is also catalyzed by proteins required for HDR, such as BRCA2, RAD51, and RAD52. Since BRCA2 and RAD51 have multiple functions at reversed fork substrates, including promoting fork reversal (RAD51) [121] and protecting the regressed nascent strands (BRAC2 and RAD51) [144, 145], teasing out their exact role in fork restoration is difficult. HDR requires much higher levels of RAD51 than those required for fork reversal or nascent strand protection [146]. RAD52 has been shown to promote replication fork restart after extensive (>6 hours) treatment with replication stress-inducing agents [147], but not with shorter (≤2 hours) treatments [148, 149] and it appears unlikely that RAD52 plays a role in HDR-dependent restoration of reversed forks without the generation of a DSB.

Direct Replication Restart Pathway Choice

In mammalian cells, evidence is accumulating that re-priming, TLS, and TS are independent pathways. For example, co-depletion of PRIMPOL and TLS polymerases (Pol η and pol ζ) leads to a synergistic increase in sensitivity to replication stress-inducing agents, suggesting re-priming and TLS function independently [40, 52]. In addition, fork reversal (and TS) is independent of re-priming and TLS [53, 58, 114]. Consistent with this, Pol κ-deficiency results in nascent strand degradation in response to HU suggesting that the replication fork reversal (and TS) occurs in the absence of TLS [67]. In the absence of PRIMPOL and Pol η, there is an accumulation of chromatin-bound RAD51, suggesting that either fork reversal or HDR occurs in the absence of re-priming or TLS [40]. Loss of POLDIP2, implicated in regulation of PRIMPOL, can also decrease the usage of other TLS polymerases and increase the usage of TS [150]. While it is clear that re-priming, TLS, and TS can proceed independently, the extent to which the pathways are interchangeable as well as the factors controlling pathway usage require further investigation.

The molecular mechanisms of pathway choice between re-priming, TLS, and TS are likely complex (Figure 3). Pathways may be engaged differently when the lesion is located on the leading- or lagging-strand. Re-priming by Pol α-primase theoretically should be able to account for the majority of lagging-strand restart as DNA Polymerase α-primase is readily available as part of the replisome and the discontinuous DNA replication on the lagging strand is part of a normal replication. However, specific evidence of this dependence on Pol α-primase in mammalian cells has yet to be reported. On the leading strand, the decision whether to re-prime with PRIMPOL, bypass the lesion with a specific TLS polymerase, or bypass the lesion by fork reversal and TS is even more complex.

Figure 3. DNA replication restart pathway choice.

When a replication fork encounters replication stress, DNA synthesis can resume through several processes to preserve genome integrity. The manner with which DNA replication resumes may be dependent on the location of the lesion. (1) Re-priming, TLS or TS can bypass lesions on the leading-strand. (1A) The leading-strand can be re-primed with PRIMPOL, which is transcriptionally upregulated in response to replication stress, can be recruited to the replication fork through interaction with RPA, and is stimulated by binding partner, POLDIP2 [61]. PRIMPOL has been primarily implicated in response to UV [40], [41] and BPDE [53] while its role in response to HU is still unclear [41], [42]. (1B) PCNA mono-ubiquitination on K164 can direct leading-strand restart through the use of TLS polymerases to bypass the lesion [86–88]. Thus far, only Pol η [66],[65], Rev1 [183] and Pol κ [67] have been implicated in acting directly at the replication fork in response to UV and HU. (1C) PCNA poly-ubiquitination on K164 [122] and PCNA SUMOylation [165] can direct leading-strand restart through fork reversal and TS. Fork reversal is a ubiquitous response to all types of replication-stress inducing agents. However, functional consequences of lack of this response has not been studied for many agents and for some, including MMC, fork reversal is not necessary for cellular survival [184] (2) Lesions on the lagging-strand are likely easily bypassed with DNA Polymerase α-primase-dependent re-priming as has been demonstrated in biochemical reconstitution experiments using yeast proteins [33].

When a lesion is present on the leading strand, PCNA modification, ubiquitination or small ubiquitin-like modification (SUMOylation), might be a central determinant for which factors are recruited to the replication fork. PCNA mono-ubiquitination promotes TLS, whereas PCNA poly-ubiquitination promotes TS. In yeast, PCNA poly-ubiquitination requires Rad6-Rad18 and Ubc13-Mms2-Rad5 [88, 151, 152]. Similarly, RAD6-RAD18, UBC13-MMS2 and the human ortholog of Rad5, HLTF, as well as SNF2 histone linker PHD RING helicase (SHPRH), are responsible for poly-ubiquitination of PCNA in mammalian cells [153–157] and this PCNA poly-ubiquitination drives replication fork reversal [122]. Recent biochemical analysis suggests a mechanism for pathway choice via distinct modes of PCNA poly-ubiquitination, including both en bloc transfer of a pre-formed poly-ubiquitin chain and step-by-step ubiquitin elongation processes [158]. Other factors have been shown to stimulate PCNA ubiquitination in response to specific types of damage, including SIVA1, NBS1, FANCD2, RAD51, PARP10, and chromatin remodelers, BAF180 and ZBTB1 [159–164]. PIAS1- and PIAS4-dependent SUMOylation of PCNA can also direct repair to TS [165]. The wide variety of responses that occur at stalled replication forks may be dependent on the timing of recruitment of the proteins involved and the type, location, and duration of replication stress. More work is required to understand if and how a stalled replication fork favors a certain response.

Other Pathways of Direct Restart

Single-molecule DNA fiber analysis suggests that many additional factors function in direct replication restart, but how these factors contribute to the pathways of re-priming, TLS, or TS is currently unknown (Table 1). BAP1 promotes replication restart by recruiting INO80 as INO80 is required for replication restart [166, 167]. In yeast, Ino80 facilitates PCNA ubiquitination [168], but whether INO80 has a similar function in mammals requires further examination. EXD2 was discovered to inhibit fork reversal and shift the balance towards an active replication fork [169]. EXD2-deficient cells exhibit BRCA-independent, MRE11-dependent nascent strand degradation and inefficient fork restart. Both of these phenotypes are rescued by inhibition of SMARCAL1 or inhibition of PARP1 (promoting RECQ1-dependent restoration) [169]. It is unknown how exactly EXD2 facilitates fork restart. A study from our laboratory implicated DNA Damage Inducible 1- and 2- (DDI1/2) mediated removal of Replication Termination Factor 2 (RTF2) from the replication fork in replication restart in response to HU [170]. When the proteasomal shuttle proteins DDI1 and DDI2 are depleted, cells exhibit an abnormal replication stress response due to the stabilization of RTF2 at the stalled replication fork [170]. How removal of RTF2 promotes restart is currently being investigated.

Fork collapse and replication restart

Fork collapse is defined as the formation of a DSB at the replication fork. At the replication fork, single-strand breaks (SSBs) can be converted to DSBs after fork passage [171, 172]. Also, in the absence of appropriate and timely restart of a stalled replisome, the replication fork is susceptible to breakage and DSBs can be generated from nucleolytic cleavage [173]. This includes the cleavage of protected and de-protected reversed replication forks by structure-specific endonucleases [144, 174–177]. Replication forks can be restarted after breakage by homologous recombination-mediated mechanisms, including break-induced replication (BIR). BIR, a form of TS requiring replication fork breakage, resumes DNA synthesis using a modified replisome assembled on a migrating D-loop. BIR requires the presence of accessory proteins, including POLD3 and RAD52 [147, 178, 179].

Concluding Remarks

Re-priming, TLS, and TS are all potential mechanisms of replication fork restart occurring directly at the replication fork and we are continuing to learn more about the involved proteins. Additionally, still little is known about the fate of the replisome components in most models of replication fork restart, though one study showed that the CMG helicase loses its GINS subunit upon fork breakage and replication is only restored upon reloading of GINS and Pol ε [171]. Single molecule DNA fiber analysis has been useful to determine factors required for direct replication fork restart in mammalian cells, but the picture of replication restart obtained from these studies is limited. Besides the caveats discussed above, replication fork breakage or fork slowing may phenotypically present similarly to a replication restart defect by single-molecule assays. The field is beginning to uncover the mechanisms of replication restart in the mammalian system and would greatly benefit from the development of new experimental model systems, including but not limited to biochemical reconstitution of mammalian DNA replication combined with live imaging techniques and reporter assays (such as a RTS1 plasmid system) in mammalian cells to observe restart by-products in vivo. Systems, like the Tus-Ter, recently adapted for the mammalian cells [180–182], need to be developed and employed more widely to study replication restart at specific sites in the genome.