Repair of DNA Interstrand Cross-links During S Phase of the Mammalian Cell Cycle

Abstract

DNA interstrand cross-linking (ICL) agents are widely used in anticancer chemotherapy regimens, yet our understanding of the DNA repair mechanisms by which these lesions are removed from the genome remains incomplete. This is at least in part due to the enormously complicated nature and variety of the biochemical pathways that operate on these complex lesions. In this review we have focused specifically on the S phase pathway of ICL repair in mammalian cells, which appears to be the major mechanism by which these lesions are removed in cycling cells. The various stages and components of this pathway are discussed and a putative molecular model is presented. In addition, we propose an explanation as to how this pathway can lead to the observed high levels of sister chromatid exchanges known to be induced by ICLs.

Introduction

DNA interstrand crosslinking drugs, which have been widely used for over a half-century, remain among the most effective and potent anticancer agents available for treatment (Apostolopoulos et al. 2008; Kohn 1996). The ability of these agents to form covalent linkages between the two strands of the double helix is unquestionably the basis for their high cellular toxicity. For instance, it has been reported that 15 psoralen ICLs induce the same level of cell killing as do 2500 psoralen monoadducts or 18,000 UV photodimers (Averbeck et al. 1992; Bankmann and Brendel 1989). This molecular constraint of the helix by ICLs, if left unrepaired, prevents or hinders virtually all aspects of DNA metabolism such as replication, transcription, and recombination. It is this property of bifunctional crosslinking agents to link both strands of the helix that renders them among the most genotoxic as well as mutagenic of compounds (Mendelsohn et al. 1992; Vogel et al. 1998; Vogel et al. 1996). It is also this property that causes the removal of these adducts to be perhaps the most complicated of DNA repair processes particularly compared to agents that introduce only monoadducts. In outline, the repair of interstrand cross-links (ICLs) appears to take two general forms in many if not most organisms (Dronkert and Kanaar 2001; Legerski and Richie 2002; McHugh et al. 2001; Noll et al. 2006). In one pathway that is usually referred to as recombination-independent, the lesion is excised on one strand of the helix and the resulting gap is subsequently filled in by translesion bypass synthesis (Berardini et al. 1999; Berardini et al. 1997; Grossmann et al. 2001; Shen et al. 2006; Wang et al. 2001; Zheng et al. 2006; Zheng et al. 2003). The remaining monoadduct is then removed by a second round of excision repair. While this mechanism is well understood in E. coli, as it is carried out by bacterial nucleotide excision repair (NER) and Polymerase II (Berardini et al. 1999; Berardini et al. 1997), it remains to a large extent unexplicated in mammalian cells. In particular, the mechanism by which the excision of the ICL occurs is undefined although the NER pathway is clearly an essential component (Wang et al. 2001). This mechanism appears to be a minor pathway in cycling mammalian cells; however, in differentiated nonreplicating cells it may be the only method for removal of ICLs, and thus may represent an important component of the overall response to chemotherapeutic cross-linking drugs. Elements of this pathway are reviewed in other sections of this issue (Hlavin et al.; Vasquez; Wood). The second general pathway of ICL repair begins with an excision process followed by homologous recombination to complete the final repair steps. In E. coli and budding yeast the excision process is performed by NER (Cole 1973; Cole et al. 1976; Jachymczyk et al. 1981; McHugh et al. 1999; Miller et al. 1982), however, in cycling mammalian cells the NER pathway does not appear to play a major role since NER mutants are many fold less sensitive to ICL-inducing compounds than are, for instance, mutants defective in homologous recombination (Andersson et al. 1996; De Silva et al. 2000; Li et al. 1999; Zhang et al. 2002). This second pathway, which is the predominant mechanism in cycling mammalian cells, takes place during the S phase of the cell cycle, and is initiated by the encounter of a replication fork with an ICL. This repair pathway, which will be referred to as the S phase pathway, is the subject of this review.

Adducts Formed by DNA Interstrand Cross-links are Structurally Diverse

There are several excellent recent reviews on the formation and chemical nature of ICLs (Noll et al. 2006; Scharer 2005), and thus my purpose here is to briefly discuss the variety of structures and the nature of the distortions introduced into DNA by the more widely used ICL-forming agents. Obviously, bifunctional crosslinking compounds introduce both ICLs and monoadducts into DNA, thus complicating the interpretation of experimental findings. Typically, less than ten percent of total adducts introduced by these compounds are ICLs.

Mitomycin C (MMC) is a widely used experimental drug particularly for cellular cytotoxicity studies as well as an antitumor agent. It reacts with guanine residues through the minor groove in the sequence 5’CG and creates little distortion in the helix of either unwinding or bending (Norman et al. 1990; Rink et al. 1996). In contrast, drugs such as nitrogen mustards, cisplatin, and psoralen, all of which are also commonly used, produce significant distortions in the helix. Nitrogen mustards typically react with guanines in the sequence 5’GNC and in at least one case have been shown to introduce a bend of 14° in DNA (Rink and Hopkins 1995). Cisplatin forms both intrastrand cross-links and ICLs, and it remains unclear which of these adducts is the more cytotoxic (Cepeda et al. 2007). The cisplatin ICL reacts with guanines in 5’GC sequences and creates very large distortions in DNA in both unwinding (79°) and bending (45°) (Huang et al. 1995; Malinge et al. 1999; Paquet et al. 1996). Psoralen products upon activation by UVA have been used for several millennia for the treatment of psoriasis, and more recently for the treatment of cutaneous T cell lymphoma (Baron and Stevens 2003; Stern 2007). Psoralens react with thymines to form highly stable ICLs in both 5’TA and 5’AT sequences, and induce an unwinding of approximately 25° with little bending of the helix (Haran and Crothers 1988; Spielmann et al. 1995). Because of their high stability psoralen adducts have been widely used for the study of mechanisms of ICL repair particularly in vitro.

Recognition of Interstrand Cross-links in Mammalian Cells

As discussed above, ICLs constitute a formidable lesion for cells not only because of the linkage to both strands of the helix, but also because of the diversity of these adducts with respect to their chemical nature and the degree of structural alteration caused to DNA. Perhaps because of this structural diversity the processes by which ICLs are recognized as lesions have remained unclear. This is to be contrasted with the well-understood mechanisms for adduct recognition in NER and base excision repair (BER) pathways. In NER there are two basic mechanisms (Nouspikel 2009; Reardon and Sancar 2005). The global genomic repair (GGR) pathway XPC recognizes the single-strand character of the strand opposite the adduct (Camenisch et al. 2009; Maillard et al. 2007; Redondo et al. 2008), which indicates that lesions recognized by GGR require that they induce some degree of unwinding of the helix. In the transcription-coupled repair (TCR) pathway of NER a stalled RNA polymerase is the signal that initiates the repair process. In this case any lesion that blocks progression of the polymerase will induce a repair response. In the BER pathway that recognizes small chemical alterations to the bases, a host of different DNA glycosylases are required to recognize very specific types of lesions (Zharkov and Grollman 2005). However, in the case of ICLs the proteins that are specifically involved in recognizing ICLs have not been unequivocally identified in large part because genetic studies have not been particularly revealing.

As mentioned above, it is clear that there exists a recombination-independent pathway of ICL repair in mammalian cells that involves NER and translesion bypass synthesis. This pathway can function on nonreplicating plasmids in vivo, and thus does not require DNA replication for its activation (Wang et al. 2001; Zheng et al. 2003). In a very interesting recent study it has been shown that components of the Fanconi anemia (FA) pathway are also involved in a recombination-independent mechanism, perhaps to assist in the loading of a translesion bypass polymerase (Shen et al. 2009). Removal of the ICL can occur by both the GGR and TCR pathways. In the GGR pathway the presumption is that XPC (xeroderma pigmentosum group C) recognizes the lesion as it has been shown to do in the case of photolesions (Batty et al. 2000; Thoma and Vasquez 2003; Volker et al. 2001). However, there are a number of conflicting issues with this scenario. First, the binding of XPC to a photolesion occurs through its affinity for the unwound single-stranded DNA opposite the lesion with no interaction with the adducted base itself (Camenisch et al. 2009; Maillard et al. 2007; Redondo et al. 2008). This exact mechanism is obviously prevented by an ICL, although as indicated above, some adducts induce unwinding around the adduct, which could be a site of binding. In vitro, a reconstituted NER system has been shown to process psoralen ICLs, however, the dual incisions both occur upstream of the ICL rather than on either side, thus leading to a nonproductive reaction (Mu et al. 2000). Nevertheless, this result indicates that XPC can recognize psoralen ICLs, and this is confirmed by in vivo findings showing that the NER pathway is involved in the removal of both psoralen and MMC adducts (Muniandy et al. 2009; Wang et al. 2001; Zheng et al. 2003). Since MMC does not induce significant unwinding or distortion in the helix, it is unclear how these adducts are recognized by the recombination-independent pathway. Presumably there are additional factors that are involved in recognition and uncoupling of ICLs by this mechanism such as MutSβ (Zhao et al. 2009). Except for ERCC1 and XPF, XP genes do not exhibit a major role in mediating cellular resistance to cross-linking agents (Andersson et al. 1996; De Silva et al. 2000), thus the role, if any, of XPC in the processing of ICLs in the S phase of mammalian cells appears unlikely.

A promising candidate for ICL recognition during S phase is the mismatch repair heterodimer MSH2-MSH3 (MutSβ). MutSβ binds a wide variety of aberrant DNA structures including base-base mismatches, insertion/deletions, nicks, and alkylated bases (Hsieh and Yamane 2008; Li 2008; Wang et al. 2001; Wu and Vasquez 2008; Zheng et al. 2003). Using an in vitro assay (CRS assay) that employed a plasmid substrate with a defined psoralen ICL, we purified MutSβ from HeLa cell extracts as a necessary component of the repair processing reaction (Li et al. 1999; Zhang et al. 2002). We also showed by an electrophoretic gel shift assay that MutSβ strongly binds psoralen ICLs. Although the plasmid substrate was not capable of replication in the mammalian cell extracts, we were, nevertheless, able to observe uncoupling of the ICL. In addition to MutSβ, the reaction required ERCC1-XPF but not XPC or XPG. Interestingly, while MutSβ was required MutLα was not, indicating that the entire MMR pathway is not involved, which is a finding also reported by others (Wu et al. 2005); although MutLα has been shown to interact with FANCJ, and may thus have a function in the recombinational steps of ICL repair (Peng et al. 2007) (Cantor and Xie, this issue). A number of additional factors involved in the early processing of ICLs were also identified through the use of the CRS assay (Li et al. 2000; Zhang et al. 2005; Zhang et al. 2003; Zhang et al. 2000). These include WRN, the helicase deficient in Werner’s syndrome, replication protein A (RPA), PCNA, and the four-protein Pso4 complex consisting of PSO4/PRP19, CDC5L, PLRG1, and SPF27. Interestingly, consistent with current models that suggest that the FA pathway functions in homologous recombination and to promote translesion bypass synthesis (Thompson and Hinz 2009), we found no role for FA proteins in these early processing steps. The role of WRN in ICL repair may be to promote regression of the replication fork when it encounter an ICL since it has been shown to possess this activity in vitro (Machwe et al. 2007). Regression of the fork would be required to bring the 5’ end of the lagging strand near the fork junction to provide a substrate for cleavage by MUS81-EME1 (Osman and Whitby 2007) (see Fig. 1). WRN-deficient cells have, however, only a moderate level of hypersensitivity to ICLs (Poot et al. 2002; Poot et al. 2001; Swanson et al. 2004), thus, there may be redundant factors such as BLM or FANCM for this function. RPA is essential for uncoupling of ICLs in vitro, and presumably acts in vivo to stabilize stalled and collapsed replication forks, and to assist in the activation of the ATR-mediated S phase checkpoint (Zou and Elledge 2003). PCNA was not an essential factor in the plasmid-based ICL uncoupling assay, but apparently acts to enhance the binding of MutSβtotheadduct (Zhang et al. 2002). The Pso4 complex has a well-defined role in pre-mRNA splicing (Ajuh et al. 2000; Ajuh et al. 2001; Chan et al. 2003; Cheng et al. 1993), but more recently has also been demonstrated to function directly in DNA repair and cell cycle checkpoint signaling (Legerski 2009; Zhang et al. 2009c; Zhang et al. 2005). Its role in these latter processes is not clear, however, PSO4/PRP19 contains a U-box domain, and is thus a putative E3 ubiquitin ligase, although its target substrates have yet to be identified.

Figure 1.

Putative model for the repair of ICLs during S phase in mammalian cells. “Ub” indicates ubiquitin. See text for details.

A role for MSH2 in ICL repair processing has also been confirmed by in vivo studies. A number of studies have shown that MSH2-deficient cell lines are hypersensitive to various ICL agents (Aquilina et al. 1998; Fiumicino et al. 2000; Wu et al. 2005; Wu and Vasquez 2008). This sensitivity is not as great as is observed, for instance, in XPF mutants, however, MSH2 is also known to be involved in apoptotic signaling in response to alkylating agents and cisplatin (Toft et al. 1999; Topping et al. 2009). Thus, in the absence of MSH2 cellular sensitivity to ICLs would be comparatively reduced due to abrogation of apoptotic pathways. In addition, in the absence of MMR alternative tolerance pathways may be activated which may also reduce sensitivity (see Cantor and Xie, this issue). Lan et al. (Lan et al. 2004) have shown that MSH2 participates in a pathway of repair of cisplatin ICLs, but not monoadducts. Knockdown of either ERCC1 or MSH2 in an XPA background resulted in similar levels of additional sensitivity. Knockdown of both genes resulted in the same level of sensitivity as the single knockdowns. In addition, ERCC1-XPF interacts with MSH2 as determined by co-IP experiments. These results indicate that MSH2 cooperates with ERCC1-XPF in a common pathway of ICL repair. In addition, Wu et al. (Wu et al. 2005) showed that MSH2 was required for processing of psoralen ICLs in vivo, and that it functions in a nonmutagenic, presumably recombination-dependent, pathway of ICL repair. A similar finding was also made by Zheng et al. (Zheng et al. 2006) in that MSH2 is not required for recombination-independent repair. In fact, in the absence of MSH2 the activity of this pathway was enhanced suggesting that there is a competition between these pathways in the processing of ICLs. Using an in vivo plasmid-based assay in which homologous recombinational repair was stimulated by the presence of a psoralen ICL, we found that MSH2 mutants showed equal or greater deficiency in the assay as did ERCC1, XPF, FANCA, FANCG, and FANCD2 mutants (Zhang et al. 2007b). Only REV3 mutants exhibited greater deficiency. Thus, this assay, which is a direct measure of repair of psoralen ICLs in vivo, indicated that MSH2 was as critical to ICL processing as ERCC1-XPF or FA proteins. Recently, additional biochemical evidence has emerged for a role for MutSβ in ICL repair. Several groups have shown that MutSβ interacts with SLX4, which was shown in cellular studies to be required for ICL repair (Andersen et al. 2009; Fekairi et al. 2009; Munoz et al. 2009; Svendsen et al. 2009). In addition to MutSβ, SLX4 also interacts with ERCC1-XPF and MUS81-EME1, proteins that have previously been implicated in the uncoupling of ICLs. The SLX4 complex will be discussed in greater detail below.

Another mechanism by which ICLs may be sensed by cells is the stalling of replication forks when the replisome encounters an ICL. This mechanism would be analogous to the TCR pathway of NER in which a stalled RNA polymerase provides the signal to initiate repair processing. A stalled replication fork may recruit FANCM, which has an ATP-dependent translocase activity, and has been shown to bind and induce regression of replication forks in vitro (de Winter and Joenje 2008; Gari et al. 2008). FANCM is a component of the FA core complex and would therefore recruit the entire core complex for the subsequent monoubiquitylation of FANCD2 and FANCI, which would initiate downstream repair steps (Thompson and Hinz 2009). It is conceivable that both a stalled fork and the lesion itself may act as independent signaling events in the recruitment process. In this scenario the stalled fork would recruit the FA proteins and other components which are involved in fork stabilization and chromosome remodeling, translesion bypass polymerase recruitment, and ultimately homologous recombination. The ICL itself on the other hand would attract the SLX4 complex containing MutSβ, MUS81-EME1, and ERCC1-XPF, which is likely required for uncoupling of the adduct. While there does not appear to be evidence for a physical interaction between the SLX4 complex and FA proteins, it has been shown that ERCC1-XPF is required for efficient recruitment of ubiquitylated FANCD2 to sites of DNA damage (Bhagwat et al. 2009; McCabe et al. 2008). In addition, it has recently been reported that in Xenopus laevis extracts FANCD2 is required for the introduction of incisions at sites of ICLs (Knipscheer et al., Science, in press). However, these findings appear to conflict with earlier reports showing that the FA pathway is not required for the uncoupling of ICLs in either Xenpous extracts or in mammalian cells (Rothfuss and Grompe 2004; Sobeck et al. 2006).

Studies from the Lambert group have shown that the structural protein nonerythroid α spectrin (αII spectrin) can bind psoralen ICLs in vitro, and that it localizes at sites of ICL damage with ERCC1-XPF and FANCA (McMahon et al. 2001; Sridharan et al. 2003). Furthermore, depletion of αII spectrin by siRNA results in the attenuation of XPF nuclear foci upon exposure of cells to ICL-forming agents (McMahon et al. 2009). Thus, αII spectrin may have a role in the recognition of ICL damage, and to recruit downstream components of the ICL repair machinery.

Finally, Paul Miller and his colleagues have identified a novel pathway of ICL processing in mammalian cell extracts that does not require NER or MMR pathway proteins (Smeaton et al. 2008; Smeaton et al. 2009). A complete discussion of these findings can be found in this issue (Hlavin et al.)

S Phase Repair of Interstrand Cross-links

A number of investigators have proposed models for the repair of ICLs during S phase in mammalian cells (de Winter and Joenje 2008; Dronkert and Kanaar 2001; Li and Heyer 2008; McHugh et al. 2001; Niedernhofer et al. 2005; Patel and Joenje 2007; Thompson and Hinz 2009). The common theme among these models is that when a replication fork encounters an ICL it ultimately causes replication fork collapse and an incision in a template strand leading to the formation of a one-sided double strand break. This initial incision begins the process of the uncoupling of the ICL, which requires a second incision downstream of the lesion for completion. The uncoupling process creates essentially a monoadduct in a single-stranded region of the chromosome, which requires either translesion bypass synthesis or homologous recombination for gap repair. The requirement for a step requiring translesion bypass makes these models inherently error-prone particularly considering the many different types of ICLs that can form in mammalian cells. Once the double helix is restored by bypass synthesis the monoadduct can be removed possibly by either the NER or BER pathways (see Saparbaev in this issue), and finally restoration of the fork occurs by homologous recombination in a process generally referred to as break-induced replication (Kraus et al. 2001; McEachern and Haber 2006). Recently Raschle et al. have proposed a somewhat different mechanism based upon evidence derived from experiments performed in Xenopus lavis egg extracts (Raschle et al. 2008). In this model two converging forks encounter an ICL and uncoupling of the ICL leads to a two-sided double-strand break. The intact chromatid with the resulting monoadduct is repaired by translesion bypass synthesis and subsequent removal of the monoadduct. The chromatid with the double-strand break is repaired by homologous recombination using the repaired chromatid as a donor. Our group was actually the first to show complete repair of psoralen ICLs in a plasmid substrate in Xenopus laevis extracts (Lu et al. 2005). We also showed that this repair was highly mutagenic with a mutation frequency of approximately 18% composed primarily of point mutations indicating that translesion bypass synthesis was involved in this in vitro system.

All these models of ICL repair invoke a DSB as an obligatory intermediate in the repair process. MUS81-EME1 has been identified as a nuclease that mediates cleavage at stalled replication forks, and has been shown to be required for replication restart after exposure of cells to ICLs as well as other fork blocking lesions (Hanada et al. 2007; Hanada et al. 2006). Consistent with these results a knockout of MUS81 in the mouse revealed hypersensitivity to ICLs, and in one model accelerated tumorigenesis. However, this accelerated tumorigenesis was not observed in a second mouse model (Dendouga et al. 2005; McPherson et al. 2004). Nevertheless, the hypersensitivity of MUS81-deficient cells, and the extensive evidence both in vitro and in vivo that ICLs lead to DSBs strongly indicates that this is a necessary step in the uncoupling process (Bessho 2003; Raschle et al. 2008).

A second protein shown to be involved in mediating the formation of DSBs in response to ICLs is SNM1B/Apollo (Bae et al. 2008). SNM1B/Apollo is a member of a small gene family referred to as SNM1, whose members have been shown to possess nuclease activity located in the conserved metallo-β-lactamase and appended β-CASP (CPSF-Artemis-Snm1-Pso2) domains (Bonatto et al. 2005; Callebaut et al. 2002; Dominski 2007). Studies have shown in both human and chicken DT40 cells that SNM1B/Apollo-deficient cells are hypersensitive to ICLs (Bae et al. 2008; Demuth et al. 2004; Ishiai et al. 2004; Nojima et al. 2005). Some of these studies have also reported hypersensitivity to IR, while others have not. A defective activation of the ATM-mediated S phase checkpoint was also observed in SNM1B/Apollo-deficient cells, which is consistent with the failure of these cells to promote the formation of DSBs in response to ICLs (Bae et al. 2008). Initially when replication forks encounter an ICL the ATR-mediated S phase checkpoint is activated, which helps to protect and stabilize the stalled fork (Lambert and Carr 2005; Lambert et al. 2007; Pichierri and Rosselli 2004a; Pichierri and Rosselli 2004b). Upon attenuation of the ATR-mediated checkpoint, collapse of the replisome complex and formation of DSBs ensues resulting in activation of the ATM-mediated checkpoint (Bae et al. 2008; Harper and Elledge 2007). The collapse of the replication fork in mammalian cells is an extremely slow process, and appears to maximize at around 12 hours after exposure to the drug, although the timing of this process may be dependent on cell type and the agent used to introduce the ICL (Hanada et al. 2006; Niedernhofer et al. 2004).

The role of SNM1B/Apollo in DSB formation in response to ICLs is unclear at present, however, it has been shown to possess a 5’-3’ exonuclease activity (Lenain et al. 2006). In response to an ICL, replication forks are likely remodeled, and one outcome may be the formation of a regressed fork or chicken foot structure. This DNA structure is a very poor substrate for MUS81-EME1, thus, a possible function for SNM1B/Apollo may be to resect the lagging strand of fully regressed forks back to the fork junction, which would create an excellent substrate for MUS81-EME1(Osman and Whitby 2007) (Fig. 1). SNM1B/Apollo has been shown to interact directly with MUS81-EME1 suggesting that these nucleases act in a coordinated fashion to carry out DSB formation (Bae et al. 2008), although it is possible that not all stalled replication forks will require the activity of SNM1B/Apollo. In fact, in S. cerevisiae the 5’ exonuclease EXO1 has been shown to counteract regressed forks (Cotta-Ramusino et al. 2005), and is also known to have an overlapping function with the budding yeast SNM1 homolog during S phase repair of ICLs (Barber et al. 2005).

In vitro experiments have also implicated the ERCC1-XPF heterodimer in the formation of ICL-induced DSBs as this complex can incise both single-stranded and double-stranded fork structures containing an ICL near the junction (Fisher et al. 2008; Kuraoka et al. 2000). However, these reactions require virtually stochiometric levels of the nuclease complex, and thus may not be physiologically relevant. Indeed, in vivo results indicate that ERCC1-XPF is not required for the formation of DSBs, but is required for their resolution (De Silva et al. 2000; Niedernhofer et al. 2004). Although, ERCC1-XPF is unlikely to be required for the initial incision creating the DSB, evidence suggests that it is required for the second incision resulting in uncoupling of the ICL. Use of the comet assay has shown a requirement for ERCC1-XPF in uncoupling in vivo (Clingen et al. 2007; De Silva et al. 2000), and uncoupling of psoralen ICLs in vitro using plasmid-based substrates has also shown a requirement for this nuclease (Li et al. 1999; Zhang et al. 2002; Zhang et al. 2000). However, this model has been questioned and the extreme hypersensitivity of ERCC1 and XPF mutant cells to ICLs was attributed to a role in the recombination steps of ICL repair (Bergstralh and Sekelsky 2008). Nevertheless, as discussed above, the recent findings that ERCC1-XPF exists in a complex with SLX4, MUS81-EME1, and MutSβ strongly suggests that it functions in the uncoupling step of ICL repair (Andersen et al. 2009; Fekairi et al. 2009; Munoz et al. 2009; Svendsen et al. 2009). Very possibly ERCC1-XPF participates in both uncoupling of the ICL, and in homologous recombination mediated fork restoration (Al-Minawi et al. 2009; De Silva et al. 2002). In addition, ERCC1-XPF participates in the recombination-independent pathway of ICL repair mediated by NER (Wang et al. 2001). Thus, the involvement of ERCC1-XPF in diverse steps of ICL repair likely accounts for the extreme hypersensitivity of mutants deficient in this nuclease. A more complete discussion of the multiple roles of ERCC1-XPF in ICL repair can be found in a separate review in this issue (Rahn et al.).

The weight of current evidence indicates that DSBs are obligatory intermediates in the repair of ICLs during S phase of the cell cycle. As discussed above, Raschel et al. (Raschle et al. 2008) proposed a model based on evidence in Xenopus egg extracts using a plasmid-based substrate that replication forks converging from both sides of the ICL are processed to yield a two-sided double strand break. Although it is highly likely that both one-sided and two-sided DSB mechanisms occur in mammalian cells, several lines of evidence indicate that one-sided DSB may be the most common pathway. First, checkpoint pathways are known to both inhibit origin firing and to slow fork elongation (Kaufmann 2007; Lambert and Carr 2005; Lambert et al. 2007; Unsal-Kacmaz et al. 2007). Both of these mechanisms would operate to limit the possibility of two forks converging on an ICL. Second, the nonhomologous end-joining pathway is known to have a limited role in the repair of DSBs induced by ICLs (Muller et al. 2000); two-sided DSBs predicted by the Raschel model should be susceptible to repair by this pathway in contrast to one-sided DSBs, which require homologous recombination for fork restoration. Third, although uncommon particularly at low doses of drug, two ICLs occurring in the same replicon would clearly prevent the convergence of two replication forks, and require repair by a one-sided DSB mechanism. Fourth, and most convincing is the repeated finding that ICLs induce sister chromatid exchanges (SCEs) (Thompson and Hinz 2009). Two-sided DSBs created, for instance, by IR do not give rise to increased levels of SCE, while one-sided DSBs naturally give rise to these DNA exchanges (see Fig. 2). These considerations obviously do not rule out the possibility of ICL-induced two-sided DSBs, but do clearly indicate that one-sided DSBs occur as an intermediate during the course of ICL repair in mammalian cells.

Figure 2.

BIR leads to SCE. Schematic illustrating a mechanism by which recombination-mediated fork restart results in SCE in a pathway that is independent of BLM. The parental strand shown in red indicates labeling occurring during the previous S phase of the cell cycle. See text for details.

After uncoupling of the ICL a repair intermediate containing a gapped structure with an intervening monoadduct remains in the other chromatid in addition to the one-sided DSB (Fig. 1). This gap must be filled in to allow subsequent repair of the monoadduct by either NER or BER pathways (Couve et al. 2009; Couve-Privat et al. 2007) (see Saperbaev in this issue). Potentially, this gap can be filled in by either translesion bypass synthesis or homologous recombination with a homologous or non-homologous chromosome as donor. Repair of short regions of high homology by non-allelic homologous recombination (NAHR) is a well-established mechanism in human cells (Gu et al. 2008; Lupski and Stankiewicz 2005), as is recombination between homologous chromosomes (Stark and Jasin 2003). The latter possibility in particular would lead to nonmutagenic repair of ICLs. Nevertheless, in most systems examined mutants of translesion bypass polymerases such as the heterodimer composed of REV3 and REV7 (Pol ζ) or REV1 are extremely hypersensitive to ICL-inducing drugs (Niedzwiedz et al. 2004). An analysis of chicken DT40 cells showed that REV3 mutants were more sensitive to cisplatin or MMC than were mutants defective in either homologous recombination or the FA pathway (Nojima et al. 2005). Possibly this extreme level of hypersensitivity could be explained by results demonstrating that translesion bypass is also involved in the recombination-independent pathway of ICL repair indicating that this mechanism is utilized in multiple pathways. However, the in vivo findings are also supported by in vitro results in Xenopus laevis extracts showing that REV3 is required for ICL repair, and that this process is highly mutagenic (Lu et al. 2005; Raschle et al. 2008). Furthermore, in vivo studies showed that rev3^−/− mouse embryonic fibroblasts were among the most deficient in repairing an ICL located at a DSB in a pathway also requiring homologous recombination (Zhang et al. 2007a). In addition, an emerging paradigm for one function of the FA pathway is to effect loading or functioning of translesion bypass polymerases during S phase repair of ICLs (Niedzwiedz et al. 2004; Thompson and Hinz 2009). This is supported by findings showing that FA mutants have a decrease in point mutations in response to ICL drugs, but exhibit an increase in chromosomal translocations and large deletions. Thus, taken together these results indicate that translesion bypass synthesis, largely carried out by Pol ζ, is a critical feature of ICL repair in vertebrate cells, and acts to prevent chromosomal aberrations. Additional polymerases such as Pol κ and Pol θ have also been implicated in mediating resistance to ICLs (Harris et al. 1996; Minko et al. 2008; Muzzini et al. 2008).

The final stage of ICL repair requires restoration of the collapsed and broken replication fork presumably through a homology-dependent process referred to a break-induced replication (BIR) or recombination-mediated replication restart. BIR is well documented in budding yeast (Kraus et al. 2001; McEachern and Haber 2006), and although it has not been directly demonstrated in mammalian cells in response to ICLs, there is evidence that such a mechanism occurs in human cells as indicated by certain types of genomic rearrangements (Zhang et al. 2009a; Zhang et al. 2009b). Furthermore, the demonstration that PARP inhibitors are strongly cytotoxic to BRCA1 and BRCA2 mutant cells suggests that BIR occurs in mammalian cells (Bryant et al. 2005; Farmer et al. 2005). PARP is required for the repair of single-strand nicks, and unrepaired nicks are known to cause the formation of one-sided DSBs during DNA replication. In addition, elements involved in homologous recombination are almost uniformly highly sensitive to ICLs indicating a major role for these pathways in mediating repair of these lesions. The FA pathway is also clearly involved in this final stage of the ICL repair process as particularly indicated by the finding that BRCA2 (FANCD1) is an FA gene (Howlett et al. 2002). There are a number of excellent recent reviews of the FA pathway and its role in homologous recombination, and therefore this aspect of ICL repair will not be discussed here (de Winter and Joenje 2008; Mirchandani and D'Andrea 2006; Patel and Joenje 2007; Thompson and Hinz 2009). In passing though, it is noted that an additional function of the FA pathway may be to stabilize collapsed replication forks. A phenotype of FA cells is that in response to ICLs they exhibit an increase in radial chromosomes (Akkari and Olson 2004). Radial chromosomes occur largely as a result of attempted or abortive recombination between nonhomologous chromosomes (Newell et al. 2004). Destabilization of collapsed replication forks would permit freed broken chromosomes to attempt such abortive recombination with other chromosomes.

Molecular model for ICL repair processing during S phase

In the putative model shown in Fig. 1 it is proposed, based on current results, that a stalled replication fork would independently recruit the FA core complex via FANCM, and the SLX4 complex through binding of MutSβ directly to the lesion. The FA core complex in turn would recruit and monoubiquinate FANCD2 and FANCI whose exact function remains unresolved. SLX4 appears to act as a scaffold for the attachment of a number of nucleases including ERCC1-XPF, MUS81-EME1, and SLX1 as well as other proteins such as PLK1, which presumably acts as a regulatory kinase (Andersen et al. 2009; Fekairi et al. 2009; Munoz et al. 2009; Svendsen et al. 2009). Knockdown of SLX1 did not result in sensitivity to MMC or cisplatin, thus this nuclease may not have a role in ICL repair processing. The SLX4 complex is thus a type of molecular “Swiss Army knife” that may be used for a variety of DNA processing steps in different pathways of repair and recombination. FANCM and/or WRN would remodel or regress the fork to bring the 5’ end of the lagging strand near the fork junction to provide an optimal substrate for MUS81-EME1 (Osman and Whitby 2007). If this regression results in a chicken foot structure then resection of the lagging strand by the 5’ exonucleases SNM1B or EXOI would be required as a fully regressed fork is a poor substrate for MUS81-EME1. Cleavage of the replication fork by MUS81-EME1 would create a one-sided DSB and a gapped chromosome containing the ICL, which would be stabilized in a single complex by FA and other proteins. ERCC1-XPF can cleave at the site of an ICL at the junction of duplex DNA and single-stranded forks (Kuraoka et al. 2000), thus, resection by a 3’ exonuclease or unwinding by a helicase may be necessary to create a substrate for the uncoupling step. As discussed above, we have been able to demonstrate this stage of the reaction in vitro using mammalian cell extracts in the CRS assay, and have identified a number of the required proteins including WRN and the Pso4 complex (Legerski 2009; Zhang et al. 2005). After uncoupling, the resulting gap containing a monoadduct is filled in by translesion bypass synthesis with Polζ as the leading candidate for this step (Gan et al. 2008; Nojima et al. 2005; Raschle et al. 2008; Zhang et al. 2007a), which also requires FA proteins presumably to load the polymerase onto the chromosome (Niedzwiedz et al. 2004). Gap filling could also occur by homologous recombination with another chromosome either homologous or nonhomologous as donor. It has been assumed previously that the NER pathway would be required for the removal of the monoadduct, however, recent findings have shown that the DNA glycosylase NEIL1 can efficiently remove a psoralen monoadduct crosslinked to a short oligonucleotide ((Couve et al. 2009) also see Saperbaev this issue). NEIL1-deficient cells are also hypersensitive to ICL-inducing drugs (Couve-Privat et al. 2007). Finally, fork restoration occurs by invasion of the one-sided DSB into the repaired chromosome by homologous recombination via the BIR mechanism (Kraus et al. 2001; McEachern and Haber 2006). This step requires FA proteins including BRCA2, likely ERCC1-XPF, and other components of the homologous recombination pathway such as RAD51 and its paralogs. All of these proteins have been shown to be necessary to prevent extreme hypersensitivity to ICLs suggesting that there is little redundancy for the homologous recombination stage of ICL repair (Niedzwiedz et al. 2004; Thompson and Hinz 2009; Thompson et al. 2005). This appears to be in contrast to budding yeast in which there are two pathways of BIR one dependent and one independent of RAD51 (Malkova et al. 2005; Malkova et al. 2001). Additional factors such as the BLM helicase and TOPA IIIα are also known to have a role in replication fork restart (Wu 2007), as well as MutLα which interacts with FANCJ (Peng et al. 2007) (Cantor and Xie, this issue).

ICL repair is associated with an increase in SCEs (Thompson and Hinz 2009), which is a phenotype not commonly found in the repair of two-sided DSBs created by IR (San Filippo et al. 2008). As illustrated in Fig. 2, uncoupling of an ICL and fork restoration by BIR of a one-sided DSB can lead directly to an SCE event. In this scenario breakage of the template strand, and subsequent invasion of the broken chromosome causes the template strand to become the primer for extension of the lagging strand of the restored fork. Upon chromatid resolution, an exchange of the template strand occurrs between the sister chromatids. This form of SCE is distinct from the type created by a crossover event at a Holliday junction during the repair of two-sided DSBs, which are normally prevented by the BLM helicase (Wu and Hickson 2003). BLM-deficient cells do not exhibit an elevation in SCEs in response to ICL-inducing drugs (Hook et al. 1984), consistent with the model shown in Fig. 2, which does not require a Holliday junction as an intermediate. Rather, the cleavage of the template strand and the subsequent fork restoration by BIR results in an exchange of the template strand between sister chromatids. This mechanism thus accounts for the elevation in SCEs due to ICLs and supports a model in which one-sided DSBs are an intermediate in ICL repair.

Future directions

As depicted in Fig. 1, ICL repair is an extraordinarily complex DNA transaction involving many steps and numerous proteins at least some of which presumably remain unidentified. Nevertheless, ICL inducing drugs remain, and likely will remain for some time, a major component of anticancer chemotherapeutic regimens. Thus, a fuller understanding of ICL repair pathways is a crucial step in the improvement of this class of therapies. As proposed above (Fig. 1), the SLX4 complex is the likely mediator of the uncoupling steps of ICL repair, yet a thorough biochemical analysis of this complex in the context of an ICL substrate remains to be performed. Particularly, the efficiency of processing of different types of ICLs by this complex is an important issue to be addressed. Another outstanding issue that is being intensively studied is the recruitment and function of FANCD2 and FANCI at sites of ICL damage. The monoubiquitination of these proteins is required for their recruitment, but the receptors for this recruitment have yet to be identified, and the mechanisms by which they apparently participate in homologous recombination and translesion bypass synthesis have yet to be elucidated. The model postulated in Fig. 1 indicates that the SLX4 complex and the FA pathway proteins are recruited to the site of ICL damage independently, but possible molecular coordination between these complexes is also a subject for future investigation. The mechanism of BIR-mediated fork restoration, while firmly established in budding yeast, has not been directly demonstrated in mammalian cells. Thus, the evidence for the role of this pathway in ICL repair, although compelling, remains indirect. A fuller resolution of these issues should lead to the elucidation of detailed molecular mechanisms which can be utilized for drug targeting schemes that can be exploited to further selectively sensitize cancer cells to ICL-inducing drugs.