Orchestrating the nucleases involved in DNA interstrand cross-link (ICL) repair

Mammalian ICL Repair

Data have been presented on mammalian systems that suggest pathways similar to those utilized by prokaryotes and lower eukaryotes may operate, especially in non-replicating cells.^3–6 However, a key aspect of mammalian ICL repair is suggested to be its predominant occurrence during S phase, coupled to replication.^1,7–10 Such replication-dependent repair was first demonstrated when photoactivation of a psoralen lesion arrested synchronized human skin fibroblasts cells exclusively on passage through S phase, regardless of the cell cycle stage at which cross-linking was induced,¹¹ supported by the observation that repair of ICLs during S phase results in generation of double-strand breaks (DSBs), which is not evident in stationary yeast or Chinese hamster ovary cells.^12,13 Additionally, there is a key difference between lower eukaryotes and mammalian cells as mammalian cells with a defective XPF-ERCC1 structure-specific endonuclease complex demonstrate increased ICL sensitivity,^12,14–16 suggesting a role for XPF-ERCC1 in ICL repair that is largely independent of its function in NER, also supported by the fact that XPF and ERCC1 defective cells accumulate excess replication-associated DNA DSBs when treated with cross-linking agents.¹⁷

The replication-coupled model of ICL repair comes in two variations. The first one, modified from Niedernhofer et al.⁸ proposes a single replication fork stalling at an ICL lesion, followed by incision by an (unspecified) endonuclease and a subsequent incision by XPF-ERCC1 to complete ICL unhooking. This allows TLS to occur past the tethered oligonucleotide, followed by a dual endonucleolytic incision and removal of the ICL from the second strand, and by gap-filling and HR steps to restore the replication fork (Fig. 1A). More recently, an alternative model for replication-dependent ICL repair has been postulated to occur when two forks converge on a single ICL^18,19 (Fig. 1B). The physical relevance of this model was demonstrated using electron microscopy where two replication forks were observed to converge on a plasmid-based ICL in the presence of Xenopus egg extract. Repair in this context involves initial stalling of both replication forks 20–40 nucleotides from the lesion, followed by one of the leading strands advancing to within one nucleotide from the ICL. Subsequently, dual incisions of the ICL result in the formation of a two-sided DSB, which can be followed by lesion bypass synthesis and HR to restore the replication fork. Regardless of whether replication-coupled ICL repair occurs at a one-sided or a two-sided replication fork, the involvement and coordination of the activities of structure-specific endonucleases is essential during the unhooking steps of ICL repair.

Structure-Specific Endonucleases and Their Substrate Preference

As implied above, incision around the ICL is a key step in initiating the repair process, as it is a prerequisite for the processing of the repair intermediates by downstream pathways. A number of structure-specific endonucleases have been implicated in the incision events of ICL repair, including the canonical NER endonuclease, XPF-ERCC1 heterodimer,^12,17,20,21 the MUS81-EME1 heterodimer,^22,23 the SLX1-SLX4 heterodimer,^24–30 and the Fanconi anemia-associated nuclease 1 (FAN1).^31–33 Biochemically, the activities of these nucleases have been studied with non-cross-linked substrates and their activities described. The activities of MUS81-EME1 and XPF-ERCC1 have been recently reviewed in reference ³⁴, and are illustrated in Figure 2A and B, respectively. Briefly, vertebrate XPF-ERCC1 has a preference for cleaving splayed-arm, bubble and stem-loop structures with a 3′-flap polarity, with hydrolysis occurring in the double-stranded (ds) DNA region close to the junction between ds and single-stranded (ss) DNA.^36–38 The nuclease activity is dependent on residues that reside in the ERCC4 domain of the XPF protein^39,40 and ERCC1 has been postulated to play a role in directing the XPF-ERCC1 complex to its target site during NER.^41,42 Meanwhile, MUS81-EME1 and its yeast homologs, while still cleaving a 3′-flap like structure, have a distinct specificity and preferentially cleave 3′-flap and replication fork as compared with splayed arm structures,^43–45 together with some activity toward Holliday junction (HJ) resolution intermediates.^34,35

Purified Slx1-Slx4 from yeast has been shown to be an endonuclease on 5′-flap, splayed arms and replication fork structures, with a lower level of activity toward static HJs.⁴⁶ Importantly, yeast Slx1 on its own is inactive, suggesting that Slx4 is required for supporting endonucleolytic activity, in analogy with MUS81-EME1 and XPF-ERCC1 heterodimers.^46,47 While mammalian homologs of yeast Slx1 have been apparent, early attempts to identify Slx4 homologs were largely unsuccessful,⁴⁶ and these have been only recently described by several independent groups using more sophisticated PSI-BLAST methods.^24,26,28,30 These groups have demonstrated that SLX1 and SLX4 interact to form a structure-specific endonuclease, with SLX1 being the catalytic and SLX4 the regulatory subunit. The human SLX4-SLX1 heterodimer exhibits hydrolytic activity toward 5′-flap, stem loop and replication fork structures, as well as being shown to be a HJ resolvase,^26,28,30 illustrated in Figure 2C. When SLX4-SLX1 was first described and its preference for cleaving 5′-flap structures demonstrated, in opposing polarity to XPF-ERCC1 and MUS81-EME1, it was proposed that SLX4-SLX1 might constitute a nuclease activity that could act together with XPF-ERCC1 (or MUS81) to unhook ICLs. However, since SLX4 interacts with both XPF-ERCC1 and MUS81-EME1, this picture and interpretation of cellular data has become more complex (see below).

The specificity of the recently identified FAN1 nuclease, illustrated in Figure 2D, has been studied by three independent groups.^31–33 All suggest that FAN1 possesses 5′-exonuclease and structure-specific endonuclease activities. In terms of the endonuclease activity, Mackay et al. showed preferential activity on 5′-flap structures, with weaker activity on replication fork structures, in agreement with Smogorzewska et al.³³ with the latter activity not observed in Kratz et al. In addition, Smogorzewska et al. demonstrate endonuclease activity on nicked substrates opposite the nick, and also label the opposite strand to show that the major endonuclease activity cleaves the strand opposite the 5′-flap, with the equivalent strand also being hydrolyzed more readily within the replication fork structure. All three groups demonstrate a 5′-exonuclease activity, with a preference for ds over ssDNA and with an ability to digest DNA from a nicked or a recessed end.

Structure-Specific Endonucleases in ICL Incision

Despite progress to define the biochemical activities of the structure-specific endonucleases discussed above, only purified XPF-ERCC1 has been shown directly to incise ICL-containing substrates in vitro.^20,21,48 The other endonucleases have been implicated in ICL repair based on their depletion increasing sensitivity of cell and animal models to cross-linking agents.

MUS81-EME1 involvement in ICL repair is suggested as mouse embryonic stem cells disrupted of MUS81 or EME1 are sensitive to mitomycin C (MMC) ^23,49 and fail to form DSBs in response to long exposure of MMC, suggesting MUS81mediated DSB formation.²³ However, the same study found formation of DSBs in response to pulsed treatment of MMC to be apparently independent of MUS81, and another conflicting report implies normal DSB formation and persistent Rad51 foci in mus81^-/- cells, suggesting a role for MUS81 in later stages of repair.⁵⁰ In our recent study using human cell lines, we propose a role for MUS81-mediated DSB formation as a late response in repair during S phase, which is important for completion of repair and replication when the XPF-hSNM1A-dependent pathway fails.⁴⁸

Together with the biochemical data implicating XPF-ERCC1 in ICL repair,^20,21,48 this endonuclease is implicated in ICL repair by increased sensitivity of Chinese hamster ovary cells defective in XPF or ERCC1 to nitrogen mustard (HN2) and MMC as compared with those defective in other NER factors.^12,14 This hypersensitivity is associated with an unhooking defect,¹² consistent with findings from the biochemical studies showing that purified XPF-ERCC1 can make incisions on both the 5′ and 3′ sides of a psoralen ICL placed on a duplex with splayed arms structure.^20,21 Although these studies imply that incisions do not take place in the context of a linear substrate, other work suggested that XPF-ERCC1 can incise ICLs within linear DNA.^51–53 In our recent study, we found that XPF-ERCC1 was able to incise 5′ to a site-specific SJG-136 ICL within a linear duplex, albeit with low efficiency, a reaction that does not occur on non-crosslinked substrate.⁴⁸ The low efficiency of purified XPF-ERCC1 activity in vitro may be due to the need for additional co-factors, for example XPF-ERCC1 enzymatic activity has been shown to be stimulated by a scaffold protein, SLX4.²⁸

As discussed below, the importance of SLX4 in ICL repair is likely to be related to its role as a scaffold protein rather than exclusively to its role as the regulatory co-factor for the SLX1 nuclease. It has been shown that SLX4 contains XPF-ERCC1 and MUS81-EME1 interacting domains in addition to interacting with SLX1^26,28,30 and other proteins, including TRF2/RAP1, the MSH2-MSH3 mismatch repair complex and the polo-like kinase PLK1,³⁰ and has thus been postulated to act as a scaffold protein for recruiting the XPF-ERCC1 and MUS81-EME1 structure specific endonucleases to the sites of ICLs.^24–30 SLX4 is implicated in ICL repair, since cells depleted of SLX4 but not SLX1 are hypersensitive to cross-linking agents.^24,26,28,30 Cells deficient in SLX4 have also been shown to be defective in repair of ICL-induced DSBs, as marked by persistence of γH2AX activation following cisplatin treatment,^26,28 but this could also be related to defective HJ resolvase activity of the SLX4-SLX1 complex and, thus, impaired HR in later stages of ICL-induced DSB resolution.

Finally the Fanconi anemia (FA)-associated nuclease, FAN1, was identified recently in a genome-wide screen for MLH1-interacting factors and is required for resistance to MMC.^{31–33,54,55} FAN1 was shown to physically interact with the monoubiquitinated form of FANCD2 via the UBZ domain,^31–33,55 and size-exclusion chromatography implicated FAN1 in two sub-complexes,³² one containing MLH1 and the other FANCD2/FANCI. FAN1 is likely involved in ICL repair, since FAN1-depleted cells show specific sensitivity to MMC and cisplatin,^31–33 although one study also reports sensitivity to camptothecin.³³ The role of FAN1 in ICL repair appears to be evolutionarily conserved as the C. elegans FAN1 mutant (tm423) showed increased embryonic lethality compared with wild-type following MMC, cisplatin or HN2 treatment.^31–33 FAN1-depleted cells showed persistence of γH2AX, replication protein A and Rad51 foci following cisplatin treatment, implicating FAN1 in DSB resolution.^31–33 Finally, FAN1-depleted cells are also defective in HR repair as directly measured by the GFP reporter assay, suggesting a role for FAN1 in HR-mediated DSB resolution during the final stages of ICL repair.^31,32

Trimming Activities at an ICL

Regardless of the steps involved in the incision of an ICL and whether a replication-dependent or replication-independent repair model operates, one of the downstream processing steps involves TLS to synthesize new DNA opposite the damaged base that remains tethered to the unhooked oligonucleotide. The role of the different translesion polymerases in ICL repair has been reviewed recently in reference ⁵⁶, but a key observation that has been made suggests that TLS is more efficient when the oligonucleotide tethered to the unhooked cross-link has been resected.^57,58 Moreover, direct evidence of trimming to a single tethered nucleotide was obtained in the recent studies of replication coupled ICL repair in the Xenopus cell-free system.¹⁹ Together, this strongly suggests a need for an enzyme to carry out such exonucleolytic processing of the tethered oligonucleotide, as illustrated by Figure 3.

A good candidate for taking this role in human cells is hSNM1A, a homolog of the Pso2 factor, first identified in two independent screens in budding yeast for mutants sensitive to cross-linking agents.^59–61 Homologs of yeast Pso2 in higher eukaryotes have been identified as SNM1A, SNM1B (also known as Apollo) and SNM1C (Artemis), structurally all members of the β-CASP family of metallo-β-lactamase-related nucleases, reviewed in references 62 and 63. Human SNM1A has been implicated in ICL repair due to its ability to partially rescue the pso2 defect in yeast cells, unlike hSNM1B and hSNM1C.⁶⁴ Crucially for the functional relevance of hSNM1A in ICL repair, both yeast Pso2 ⁶⁵ and human SNM1A^64,66 display 5′-exonuclease activity in vitro. In our recently published study, we have confirmed this activity and have shown, in addition, that hSNM1A can hydrolyze cross-linked substrates past a site-specific SJG-136 cross-link in vitro, with an ability to start digestion from a nick provided by XPF-ERCC1 endonuclease, suggesting hSNM1A's involvement as an ICL “trimming” exonuclease.⁴⁸

In terms of cellular sensitivity, three mice knockout models of SNM1A have been published to date, all suggesting a modest increase in sensitivity to MMC, but not to melphalan, cisplatin, IR or UV, of snm1a^-/- mouse embryonic fibroblasts (MEFs),^67–69 which could be complemented by hSNM1A.⁶⁸ In support, hypersensitivity to a range of cross-linking agents was observed in snm1a^-/- chicken DT40 cells,⁷⁰ although interestingly in the DT40 model, deletion of either SNM1A or SNM1B results in sensitivity to cisplatin and MMC, suggesting a possible redundancy between the two metalloβ-lactamase structural family enzymes. In contrast, SNM1C-depleted cells were found to be sensitive to IR but not to cross-linking agents.

Biochemically, the related exonuclease hSNM1B/Apollo, another member of the β-CASP family of metallo-β-lactamase related nucleases, has also been shown to posses 5′-exonuclease activity,^71,72 which is stimulated by the TRF2 telomere-associated factor, although this activity has been postulated to be involved in telomere processing. In cellular studies, as mentioned above, SNM1B disruption in chicken DT40 cells results in hypersensitivity to cross-linking agents,⁷⁰ but the role of human SNM1B in DNA repair is not entirely resolved, with some studies indicating sensitivity of hSNM1B-depleted cells to cross-linking agents and IR and others a much more drastic increase in sensitivity to MMC but not to IR.^73–76 The function of hSNM1B in ICL repair has also been attributed to a possible role in mediating checkpoint signaling, although conflicting views exist on which aspects of signaling might be linked to hSNM1B.^73,75,77,78 It remains of interest whether the exonuclease activity of hSNM1B is able to digest cross-linked DNA past a lesion in analogy to hSNM1A, and the possible redundancy between the hSNM1A and hSNM1B cross-link trimming activities remains an area for investigation. Although hSNM1C/Artemis has previously been proposed to also act as a 5′-exonuclease^79,80 as well as a structure-specific endonuclease in nonhomologous end-joining and V(D)J recombination, this was shown more recently to have been a contaminating activity and a consequence of the purification method.⁸¹ It is worth noting that the entire SNM1 family has also been proposed to have roles independent of their nuclease activity, including in mitotic cell cycle checkpoints, and this remains a topic for investigation.^63,82

Finally, the recently identified FAN1 nuclease has been shown to have a 5′-exonuclease activity with a high specificity for nicked substrates,^31–33 which would clearly be a good candidate for trimming the tethered oligonucleotide prior to TLS. However, the FAN1 activity has so far been only demonstrated on unmodified substrate, so it remains of interest to study this activity on substrates containing ICLs. To this end, it is helpful that methods have been developed recently to make ICL-containing substrates with high efficiency and high specificity, as discussed in our work with relation to the SJG-136 compound,^48,83 and in the work of the Schärer and Miller groups, who have developed methods to generate DNA ICLs using post-synthetic reductive amination or an approach combining solid phase oligonucleotide synthesis with selective removal of protecting groups of a cross-linked thymidine dimer.^57,84–86 Although there is a possible role for FAN1 in trimming a tethered oligonucleotide, it is worth noting that the groups that have identified FAN1 propose a possible role for its 5′-exonuclease activity in processing DNA strands to generate 3′ overhangs to allow strand invasion and HR, required for the restart of the replication fork once the cross-link is removed.^31–33 It is also worth noting that rodents may use FAN1 and SNM1A redundantly, explaining the relatively mild phenotype of SNM1Adeficient mice.

Coordinating the Nucleases at an ICL

A key question to answer is whether all of the different endonucleases are involved in the incision step, or whether they are recruited specifically and sequentially to ICL sites depending on the context in which the ICLs are recognized at different stages of the cell cycle. Second, it remains to be resolved how the exo and endonucleases are coordinated spatially and temporally during ICL repair. Based on the observation that mus81^-/- MEFs appear to have suppressed DSB formation,²³ while ercc1^-/- MEFs accumulate DSBs,¹⁷ the model for replication-dependent ICL repair illustrated in Figure 1A has previously been proposed. Here the initiating incision is made by MUS81, possibly on the leading strand template of the replication fork, followed by a second XPF-ERCC1-dependent incision 5′ to the ICL, which results in ICL unhooking, accepted to be a prerequisite for downstream repair processes such as TLS and HR.^17,23,87 However, evidence for the order in which XPF-ERCC1- and MUS81-mediated incisions occur had remained inconclusive. In our recent study, we now provide evidence supporting a differential employment of incision activities of XPF-ERCC1 and MUS81-EME1 depending on the context in which the ICL is detected by the replication fork. We found that depleting ERCC1 results in an increase in DSB formation, which was MUS81-dependent, a result reminiscent of hSNM1A and SLX4/FANCP depletion, implying that MUS81 acts on repair intermediates that persist when the XPF-ERCC1/SLX4/hSNM1A controlled pathway fails.⁴⁸ This suggested that MUS81-mediated incision is not the initiating event during ICL repair normally, but has increased importance when ICLs persist, or in late S phase when completion of replication requires DSB ends for HR, as illustrated by our proposed context-dependent ICL repair models in Figure 4A.

To date, three important regulators of ICL repair have been identified, FANCD2-FANCI, SLX4 and RAD18. It appears that the FANCD2-FANCI complex plays a pivotal role in controlling the incision step.^18,88 FANCD2 ubiquitination occurs rapidly in response to cross-linking agent treatment, as a result of replication fork stalling triggering the ATR signaling, as evident by similar kinetics to Chk1 phosphorylation. However, whether the activation of FANCD2 is in response to replication fork stalling or to the presence of ICL per se is unclear. Furthermore, how FANCD2 controls incision events remains to be determined, although we have previously shown that XPF-ERCC1 influences dynamics of FANCD2 chromatin association.⁸⁷ Additionally, the recent structure of the FANCD2-FANCI complex implicates monoubiquitination, as well regulatory phosphorylation, in stabilization of the FANCD2-FANCI heterodimer.⁸⁹ The identification of the FAN1 nuclease that interacts with monoubiquitinated FANCD2 may provide the missing link between FANCD2 and incision. All three groups that identified FAN1 propose the possibility that FAN1 is involved in an early incision step at the ICL, acting in concert with but with opposite polarity to (5′-flap rather than 3′-flap specificity) MUS81 (or XPF-ERCC1) to cleave the same strand on the opposite side of the ICL (Fig. 4B). However, it is acknowledged that there are problems with this proposal, since it is difficult to see how MUS81 (or XPF-ERCC1) would only cleave one of the two leading strands of the converged replication forks.^31,32 Cleavage of both forks would result in two DSBs and a linear duplex containing an ICL, and from the proposed biochemistry of the enzyme this is unlikely to be a substrate for FAN1. Additionally, Mackay et al. and Kratz et al. also investigated how depletion of FAN1 affects the appearance and persistence of DSBs in response to cisplatin or MMC treatment and show that reducing FAN1 does not prevent DSB formation but slows down DSB disappearance.^31,32 Since DSBs formed in ICL repair are removed predominantly by HR, this may implicate FAN1 in late stages of ICL repair, potentially in DSB resection through its 5′ exonuclease activity.^31–33

SLX4 has been suggested to be an important regulator of incision as it is found to interact with XPF-ERCC1 and MUS81-EME1 as well as SLX1, which has led to a proposed role for SLX4 in controlling the nuclease activity at ICL-stalled replication forks. Our data indicate that SLX4 acts in the same pathway as XPF-ERCC1 and hSNM1A,⁴⁸ and the fact that chromatin recruitment of XPF-ERCC1 is dependent on SLX4 ^25,27,29 makes SLX4 an attractive candidate for acting as a regulator of the recruitment of repair nucleases, particularly XPF-ERCC1. However, SLX4-deficient cells also accumulate ICL-associated DSBs,⁴⁸ suggesting that MUS81 recruitment to stalled replication forks may also be controlled by another factor independently of the SLX4-MUS81 interaction. Crucially, SLX4 has been recently shown to be a new Fanconi anemia factor, FANCP, providing another key link between the FA pathway and ICL repair.^25,27,29 However, SLX4 has been termed as the downstream effector of the FA pathway, since SLX4/FANCP patient cells show normal ubiquitination of FANCD2.^25,27,29 It remains to be determined if FANCD2 controls ICL repair through SLX4, or if SLX4 acts in parallel with FANCD2-FANCI, providing another layer of regulation of the ICL repair process. Finally, it has been recently suggested that RAD18 and ubiquitinated PCNA might be involved in targeting hSNM1A to damaged replication forks,⁹⁰ possibly providing a means of targeting this protein to repair intermediates requiring exonucleolytic trimming. In this regard, several recent publications suggest that RAD18 contributes to the FA-mediated repair, as PCNA K164 monoubiquitination appears to promote full activation of the FA pathway.^91–93 This would potentially provide an attractive means of co-opting hSNM1A into FA-controlled repair processes, and coordinating this with TLS-mediated downstream steps also involving monoubiquinated PCNA. In fact, RAD18 has also been recently implicated in chromatin localization of FANCD2-FANCI,⁹⁴ further supporting its role as a coordinator involved in ICL repair.

Future Directions

Work published recently by our group and others suggests that there is more than one system for the repair of ICLs.^{6,19,31–33,48,95} Clearly, the regulation of the choice of repair mechanism represents the next major challenge in the field. While our data suggest that SLX4 acts in the same pathway as XPF-ERCC1 and hSNM1A, which together with the known interaction of SLX4 and XPF-ERCC1 indicates a possible role for SLX4 as a regulator of the pathway, a number of outstanding questions remain to be addressed in the proposed model of ICL repair. In particular, delineation of the order of recruitment of different nucleases to the chromatin will be critical. We have attempted to examine the chromatin recruitment of XPF-ERCC1 and MUS81-EME1 nucleases following cross-linking agent treatment, but this has proven technically difficult as the nucleases appear to be constitutively associated with chromatin (Wang AT and McHugh PJ, unpublished data). Thus examining the localization of fluorochrome-tagged nucleases as well employing proximity ligation assays following cross-linking agent treatment could provide an alternative approach for determining the spatial and temporal relationships of key factors deployed in ICL repair. It would be of particular interest to investigate whether or not depletion of SLX4, the putative nuclease coordinator, affects the timing and location of nuclease recruitment.

Further understanding of the mechanistic details of ICL repair could be derived from a comprehensive analysis of the substrate specificity for each of the nucleases, with particular emphasis on using crosslinked substrates, such as in the studies discussed previously in references 20 and 21. It will be helpful to utilize the recently described methods for making crosslinked substrates with high specificity and high efficiency,^{48,57,84–86} introducing a site-selective ICL in the context of a variety of replication fork-like substrates, including leading and lagging strands and converged replication forks. In particular, the interplay between the various nucleases, which have been highly purified in recombinant form^{31–33,40,48,66} and the SLX4 scaffold protein^28,30 in vitro will be of interest as it is likely that a number of the proteins require the presence and/or activity of other proteins to support their function.

Although it may suggest which steps of ICL the nucleases participate in, the in vitro approach would clearly need to be validated in the in vivo setting. Therefore, it will be important to identify and confirm direct interactions between the nucleases and associated coordinator factors in order to better understand the dynamics of ICL repair. The interaction of SLX4 with both XPF and MUS81 ^24,26,28,30 is a key area for investigation, especially since the two nucleases appear to have distinct roles during ICL repair, and so it will be interesting to examine the physical interaction of SLX4 with XPF/MUS81, and how it varies during the different stages of ICL repair. Furthermore, the significance of the interaction between FANCD2 and FAN1 during ICL repair, and the interplay between this interaction and the SLX4-XPF/MUS81 interactions remains to be determined. Finally, how hSNM1A is targeted to ICLs has only recently begun to be examined, and this will shed further light on the control of this important repair pathway.