Sensing and Processing of DNA Interstrand Crosslinks by the Mismatch Repair Pathway

MutSα (MSH2–2013;MSH6) senses DNA interstrand crosslinks

The MutSα complex recognizes a diverse array of aberrant DNA structures including base-base mispairs and chemically modified DNA adducts. We hypothesized that the ability of MutSα to recognize distortions within DNA might also extend to interstrand crosslinks (ICLs), and that sensing of ICLs by MutSα could initiate ICL repair by a set of reactions analogous to MMR. To test this hypothesis, we first asked whether MutSα binds preferentially to ICLs in non-replicating Xenopus extracts (High Speed Supernatant - HSS) using a plasmid pull-down assay. Notably, these cell-free extracts support ICL repair and restore the integrity of both DNA strands (Williams et al., 2012).

We used SJG-136, a rationally designed crosslinking drug, to generate plasmids with a single site-specific ICL lesion. SJG-136 preferentially forms ICLs between guanine residues at 5′-purine-GATC-pyrimidine-3′ sequences (Figure S1A) (Gregson et al., 2001). We treated oligonucleotide duplexes containing a single SJG-136 reaction site (Figure S1B), and purified crosslinked duplexes by denaturing PAGE. These oligos were ligated into a small plasmid (pBS: pBlueScript), and further purified by cesium chloride density ultracentrifugation.

Following incubation in HSS, we isolated plasmid and plasmid-bound proteins from the extract using biotinylated lac repressor protein and streptavidin beads, as previously described (Williams et al., 2012). Western blot analysis revealed that the crosslinked plasmid was enriched in bound MSH2 and MSH6 when compared to identical undamaged control plasmids (Figure 1A).

In order to study the dose-dependent response of MutSα to crosslinks, we conducted a similar experiment with plasmids containing multiple ICLs by treating double-stranded pBS plasmids in vitro with increasing doses of SJG-136 (Figure S1C–E). We observed dose-dependent increase of MSH2 and MSH6 binding to damaged plasmids (Figure 1B), after normalizing for plasmid recovery using qPCR (Figure S1F). We also observe the recruitment of PCNA and RPA to these plasmids in a dose-dependent manner, suggesting that the cross-linked plasmids were undergoing processing and repair. Consistent with the accumulation of RPA and with our previous studies (Ben-Yehoyada et al., 2009; Williams et al., 2012), incubation of the SJG-136-treated plasmids in HSS triggered phosphorylation of cytosolic Chk1 (Figure S1G), a marker for ATR checkpoint activation.

Next, we asked whether MutSα is required for ICL repair. ICL plasmids harboring a single, site-specific trimethylene ICL lesion (5′GpC-ICL), for which an NMR structure has been described were used as substrates (Protein Data Bank: 2KNL) (Dooley et al., 2003). In contrast to the temperature-sensitive ICLs generated by SJG-136, the trimethylene crosslink is extremely stable at high temperature. We can therefore monitor ICL repair by quantitative PCR by comparing the increase in amplification of the damaged “X” region to that of an undamaged “C” region on the plasmid backbone over time (Ben-Yehoyada et al., 2009; Williams et al., 2012) (Figure S1B, right).

Incubation of cytosol with MSH2 antibodies raised against Xenopus MSH2 protein depleted both MSH2 and MSH6 subunits of MutSα (Figure 1C, left), but did not reduce the cytosolic levels of other MMR proteins including MLH1 and EXO1 (Figure 1C, right). Compared to mock-depleted extracts, depletion of MutSα significantly decreased ICL repair (Figure 1D). Importantly, addition of recombinant MutSα^WT to MSH2-depleted extracts restored ICL repair to the level of mock-depleted extracts. Moreover, over-expression of MutSα^WT in undepleted HSS increased the ICL repair efficiency (Figure 1E), suggesting that lesion recognition by MutSα is a rate-limiting step during RIR. These data establish that MutSα is required for ICL repair in the absence of DNA replication and transcription.

Mechanism of ICL recognition by MutSα

MutSα discriminates aberrant DNA substrates amid a vast excess of normal DNA bases. The MSH6 subunit contributes critically to this process in at least two ways. First, the MSH6 subunit contains a highly conserved FXE motif (Figure S2A,B) that functions during damage recognition (Lamers et al., 2000; Malkov et al., 1997): the aromatic ring of the phenylalanine residue stacks with the mispaired base, whereas the carboxyl group of the glutamic acid residue hydrogen bonds with the mispaired base (Warren et al., 2007). A single amino acid substitution of this phenylalanine residue prevents efficient recognition of mispaired bases in yeast (Bowers et al., 1999), and results in defective repair in human cell extracts (Dufner et al., 2000).

To test the possible role of the FXE motif in MutSα-dependent repair of ICL in Xenopus extracts, we cloned and purified Xenopus mismatch binding-deficient MutSα^FXE complex (Figure S2A,B,D). First, we asked if the F411A substitution affected mismatch repair using a plasmid-based MMR assay in HSS extracts. We used plasmids containing a single A:C mismatch and a 15 nucleotide gap on the 3′ side of the A-strand (pMM1^AC) to trigger gap-directed strand-specific MMR (Figure 2A), as previously described (Kawasoe et al., 2016). Repair of this plasmid occurs preferentially on the A-strand, and the correction of the A:C site to a G:C pair generates a BamHI restriction site (Figure 2A, top). Background levels of repair on the C-strand is monitored by digestion with XhoI (Figure 2A, bottom). 60 minutes after incubation in mock-depleted extracts, 37% of the A:C mismatches were repaired to G:C. As anticipated, MMR was abolished in MSH2-depleted extracts. Addition of recombinant MutSα^WT, but not MutSα^FXE restored repair (Figure 2A, top). The requirement for the FXE motif for MMR is therefore conserved in Xenopus.

Next, we determined if the F411A substitution ablated ICL repair. As shown in Figure 2B, recombinant MutSα^FXE was unable to support repair of ICLs in MSH2-depleted extracts. The FXE domain is, therefore, critical for ICL damage recognition.

In addition to recognizing a mismatched base, MSH6 also interacts with proliferating cell nuclear antigen (PCNA). Studies using fluorescently-tagged proteins show that the interaction between MSH6 and PCNA is important for the colocalization of the MutSα complex to replication factories in S. cerevisiae (Hombauer et al., 2011) and in human cells undergoing MMR (Kleczkowska et al., 2001). Yet, disruption of the PCNA-MSH6 interface results in a modest MMR phenotype (Shell et al., 2007), and other PCNA-independent recruitment of MutSα to mismatches via histone methylation has also been described (Li et al., 2013).

We used plasmid pull-down assays to ask whether MutSα was recruited to ICLs in a PCNA-independent manner. As shown in Figure 1B, the MutSα complex and PCNA were recruited to SJG-treated plasmids in HSS extracts. MutSα recruitment was not dependent on PCNA; MSH2 and MSH6 were efficiently recruited to crosslinked plasmids in PCNA-depleted extracts. Furthermore, the extent of MutSα recruitment was unaffected by supplementation of extracts with recombinant PCNA protein (Figure 2C). This experiment strongly suggests that MutSα can be recruited to ICLs independently of PCNA.

The interaction between MSH6 and PCNA is mediated by a PCNA interaction motif (PIP box) in MSH6 (Clark et al., 2000; Flores-Rozas et al., 2000). We constructed recombinant MutSα complex with a triple amino acid substitution in its conserved PIP box motif: Q(X)₂LI(X)₂FF (Figure S2A,C,D) that disrupts PCNA interaction with MutSα in vitro (Figure S2G). We tested the ability of this MutSα^PIP to support ICL repair in MutSα-depleted extracts. We find that MutSα^PIP is unable to support ICL repair (Figure 2B). This indicates that whereas MSH6-PCNA interaction is not required for MutSα recruitment to ICLs, and presumably for ICL sensing, the interaction between the two is essential for coordinating downstream repair events.

ICLs are processed by MutLα and EXO1 nucleases

The earliest catalytic events during ICL repair are thought to be dual endonucleolytic incisions flanking the ICL lesion in a process referred to as “unhooking”. However, neither the exact mechanism nor the nuclease(s) involved in the process are known for RIR.

We engineered the ICL-containing plasmids with 4 unique nicking sites flanking the ICL (Figure 3A and Figure S3A). We hypothesized that presenting the extract with a pre-nicked plasmid might stimulate ICL repair if the substrate resembled a physiologically relevant ICL repair intermediate. Indeed, introduction of a single nick 19 base pairs (bp) 5′ to the ICL lesion stimulated repair by 4-fold. In contrast, nicks 15 bp or 30 bp 3′ to the ICL or 34 bp 5′ to the ICL failed to stimulate repair (Figure 3B). We conclude that incision 5′ to the ICL is a rate-limiting step during the repair process. It further suggests that a 5′ incision close to the ICL (19 bp vs. 34 bp), is optimal for repair and could reflect the physiological site of incision. Finally, the asymmetry between 3′ and 5′ incisions could reflect the action of 2 distinct nucleases.

Our evidence that MutSα is recruited to ICLs and is required for their repair prompted us to investigate the possible role of the mismatch-associated MutLα (MLH1-PMS2 heterodimer) endonuclease in ICL repair. Recruitment of MutLα to SJG-136 treated plasmids was observed following incubation of SJG-136-treated plasmid in extracts (Figure 3C). Notably, MLH1 recruitment was critically dependent upon MutSα, since MLH1 binding to the ICL-containing plasmid was abrogated in MSH2-depleted extracts (Figure 3C).

Specific antibodies against Xenopus MLH1 quantitatively depleted both MLH1 and PMS2 from extracts (Figure 3D, left). Importantly, MLH1 depletion did not affect the levels of MSH2 and MSH6 in the cytosol (Figure 3D, right), and MSH2 and MSH6 were still enriched on SJG-136 treated plasmids in MLH1-depleted extracts (Figure S3E, left). Removal of MLH1/PMS2 reduced ICL repair by 50%, and addition of recombinant MutLα^WT to MLH1-depleted extract restored repair (Figure 3E). Therefore, MutLα is not only recruited by MutSα to ICL plasmids, it is also necessary for efficient ICL repair.

The PMS2 subunit of MutLα complex contains a conserved endonuclease motif DQHA(X)₂E(X)₄E. Substitution of E707 to K abolishes the endonucleolytic activity associated with PMS2 and yields a strong mutator phenotype in S. cerevisiae (Smith et al., 2013). Mutation of the corresponding residue (PMS2-E702K) in mice increases genomic mutation rates and cancer predisposition (van Oers et al., 2010). We generated the equivalent mutation in Xenopus MutLα (MLH1-PMS2^E674K) (Figure S3B–D). Recombinant MutLα lacking endonuclease activity did not support ICL repair in MLH1-depleted extract (Figure 3E). Furthermore, overexpression of nuclease deficient MutLα complex significantly reduced ICL repair, possibly by acting as a dominant-negative (Figure 3F). Overexpression of recombinant MutLα complexes did not interfere with MutSα recruitment to SJG-136 treated plasmids (Figure S3E, right). These results indicate that PMS2-associated nuclease activity is required for ICL repair.

The dual incision flanking the ICL lesion during “unhooking” produces an oligonucleotide that remains covalently attached to DNA by the ICL adduct and is base-paired on each side of the ICL. During replication-coupled ICL repair, this DNA fragment is degraded with 5′→ 3′ polarity by the SNM1A exonuclease, presumably allowing DNA polymerases to be loaded (Wang et al., 2011), and for other downstream repair reactions to take place.

We speculated that the 5′→ 3′ EXO1 exonuclease could fulfill a similar function in RIR. In an in vitro reconstituted vertebrate MMR system, EXO1 degrades the single-strand oligonucleotide harboring the mismatch (Genschel et al., 2002). We found that EXO1 was recruited to SJG-136 treated plasmids. As is the case for MutLα, EXO1 recruitment was dependent on MutSα: EXO1 binding to the ICL-containing plasmid was reduced in MSH2-depleted extracts (Figure 4A). A specific antibody generated against Xenopus EXO1 quantitatively depleted EXO1 from extracts (Figure 4B). ICL repair was significantly reduced in EXO1-depleted HSS. This defect could be rescued by addition of EXO1^WT, but not catalytically inactive EXO1^D173A (Figure 4C) (Liao et al., 2011). This demonstrates that EXO1 and its nuclease activity are required for ICL repair in the absence of DNA replication and transcription.

To evaluate the specific contribution of EXO1 during ICL repair, we conducted in vitro nuclease assays (Wang et al., 2011). First, we purified the catalytic domain of human EXO1 (EXO1₃₅₂) from E. coli as previously described (Shi et al., 2017). EXO1₃₅₂ harbors robust nuclease activity in vitro, but lacks its C-terminal region (Figure S4A), important for making connections with various protein binding partners including MSH2, MLH1, PCNA, and RPA (Shi et al., 2017).

Duplex oligonucleotides containing a single SJG-136 ICL lesion are easily distinguished by size from uncrosslinked duplex on a denaturing polyacrylamide gel (Figure S4B). Similarly, the reaction products upon incubation with EXO1₃₅₂ can be visualized and identified. We generated control duplex DNA with only one free 5′ end available for processing (the 5′ end of the other strand blocked by addition of a biotin moiety). This template is efficiently processed by EXO1₃₅₂ from the biotin-free 5′ end in the 5′ to 3′ direction as anticipated (Figure 4D, left). Exonucleolytic processing was abolished when both ends of the duplex are blocked with biotin, confirming that under these experimental conditions, only the exonuclease activity contributes to the processing of the DNA substrates. When EXO1₃₅₂ was incubated with similar ds-DNA oligonucleotides containing a single SJG-136 ICL lesion, EXO1₃₅₂ was able to initiate processing of the substrate from an available 5′ end. However, EXO1 processing was blocked by the ICL lesion (Figure 4D, right). EXO1₃₅₂ is therefore unable to bypass an ICL under these conditions.

Incubation of plasmids harboring ICL lesions activates the RPA-ATR-pChk1 branch of the DNA damage response (Figure S1G). Next, we assessed the role of replication protein A (RPA) in ICL repair. Trimeric RPA protein complex was depleted using an antibody generated against RPA1. Depletion of RPA from HSS extracts completely abolished ICL repair. ICL repair was restored by addition of recombinant Xenopus trimeric RPA complex (Figure S4C–E). This is consistent with our previous studies in which we reported a requirement for RPA in ICL repair in LSS extracts (Low Speed Supernatant) treated with geminin and roscovitine to inhibit replication (Ben-Yehoyada et al., 2009).

We predicted that RPA binding to crosslinked plasmids during ICL repair would occur primarily after ICL recognition and initiation of ICL unhooking events. We found that MSH2, MSH6, and MLH1 were loaded to similar levels to SJG-136 treated plasmids following incubation in mock- and RPA-depleted extracts (Figure S4F). This result supports a model in which MutSα and MutLα play early roles during ICL repair, preceding nucleolytic events during the repair.

The efficiency of RIR is influenced by ICL structure

In replicating extracts, ICLs are sensed by a translocating replisome that stalls at the ICL. In RIR, we suggest ICL sensing is dependent on distortion of the DNA helix at the lesion. To test this idea, we compared the kinetics of repair between chemically related trimethylene ICLs in two distinct structural conformations: the more distorting 5′-GpC-ICL (Dooley et al., 2003) and the less distorting 5′-CpG-ICL (Dooley et al., 2001) (Figure 5A and Figure S5A). Both ICLs were ligated into a pEGFP vector backbone (Figure S5B). We found that the more distorting ICL induced a stronger ATR checkpoint activation (Figure 5B). Furthermore, repair of the more distorting lesion was 40% more efficient than the less distorting lesion (Figure 5C). We next asked whether repair of these ICLs was dependent upon MutSα. As shown in Figure 5D, repair of both trimethylene crosslinks displayed a similar requirement for MSH2. Thus, MutSα is able to recognize structurally distinct ICL lesions.

Finally, we monitored repair of a chemically distinct nitrogen mustard-like ICL lesion (Figure 5E and Figure S5A). This NM-ICL is generated within the major groove with minimal distortion of the DNA helix, allowing the DNA to maintain B-DNA conformation (Guainazzi et al., 2010; Mukherjee et al., 2014). Interestingly, even after 3 hours of incubation there was no detectable repair of this NM-ICL (Figure 5E). This supports our notion that ICL recognition in RIR depends on DNA distortion, and is consistent with a previous report that repair of NM-ICLs are entirely dependent on replication (Räschle et al., 2008).

MutSα is a bona fide ICL sensor

The mechanism(s) by which ICLs are sensed and repaired in non-dividing cells has been difficult to elucidate. In this study, we used Xenopus HSS extracts, which do not replicate or transcribe, to identify eukaryotic mismatch repair complex, MutSα (MSH2-MSH6), as a sensor of ICL lesions. Our conclusion is supported by the following observations. First, MutSα was recruited to plasmids containing ICL lesions (Figure 1A,B). Second, MutSα carrying its mismatch binding FXE motif was required for repair of trimethylene ICLs (Figure 2B). Third, binding of MutSα to ICL was required for recruitment of downstream repair proteins including MutLα (Figure 3C) and EXO1 (Figure 4A).

We conclude that binding of MutSα to ICLs precedes nucleolytic processing, and thus acts at an early step in ICL repair. Moreover, overexpression of MutSα in extracts enhanced ICL repair efficiency (Figure 1E), further suggesting MutSα senses ICLs, and that such sensing is a rate-limiting step in repair.

We observe that MutSα-dependent ICL repair has slower kinetics than the repair of mismatches in Xenopus extracts (Kawasoe et al., 2016; Radman, 2016). We surmise that the complexity of the ICL lesion, which requires two rounds of repair synthesis during RIR, could account for this difference.

Importantly, we find that structurally distinct ICLs were repaired with varying efficiencies. Trimethylene-ICLs, which significantly distort DNA were repaired with robust efficiency, in contrast to non-distorting NM-like ICLs, which had no detectible repair (Figure 5C,E). This indicates that ICL repair by RIR is critically dependent on ability of the ICL lesion to be recognized by DNA damage sensor proteins. Thus, repair of ICLs in G0/G1 utilizes a fundamentally different mechanism than ICL repair during S-phase, when lesion sensing occurs by replisomes stalled at ICLs irrespective of crosslink structure. Moreover, ICL recognition by MutSα during RIR may also be influenced by sequence context, as has been proposed for MMR (Mazurek et al., 2009).

Of note, MutSα (MSH2-MSH6) is one of two mismatch repair sensors that operate during MMR in vertebrates. The other is MutSβ (MSH2–MSH3), which is about 10 times less abundant than MutSα in mammalian cells and has higher specificity for recognizing larger insertion-deletion loops (IDLs) and branched DNA substrates (Genschel et al., 1998). Depletion of MSH2 from extracts quantitatively depleted MSH6 and MSH3. Our depletion-rescue experiments with recombinant MutSα, which fully rescues ICL repair, helps assign the specificity of ICL recognition to MutSα (Figure 1D). In mammalian cells, MutSβ along with the global genome nucleotide excision repair sensor, XPC (xeroderma pigmentosum complementation group C) have been linked to psoralen ICL repair (Muniandy et al., 2009; Thoma et al., 2005; Zhang et al., 2002; Zhao et al., 2009). However, how these complexes contribute to the repair of ICLs remains unclear.

We were unable to detect direct binding of MutSα to SJG-136 treated oligonucleotides using EMSA experiments. This could be due to the design of the oligo duplex itself, which is restricted in length (~20 bases) to ensure SJG-136 generates a single ICL and to avoid the formation of secondary structures, since the duplex must be almost exclusively be composed of A and T nucleotides. While our data using the mismatch binding deficient (FXE) mutant MutSα (Figure 2B) strongly suggest that MutSα binding to DNA is critical for ICL repair, we cannot formally rule out that additional proteins participate in recruiting MutSα to ICLs.

The role of PCNA during Replication Independent Repair

Previously we reported that PCNA is an essential RIR factor. We showed that PCNA ubiquitinylated at lysine (K164) is required to recruit Polκ to ICL sites for repair synthesis (Williams et al., 2012). The experiments described here indicate that PCNA is not essential for MutSα recruitment to ICLs (Figure 2C). However, we find a requirement for interaction between MutSα and PCNA for ICL repair (Figure 2B). This suggests that PCNA may help retain MutSα at damage sites, or perhaps enhance MutSα’s damage recognition specificity, as has been suggested during MMR (Clark et al., 2000; Flores-Rozas et al., 2000; Kleczkowska et al., 2001). Our data show that PCNA plays an upstream role during ICL repair, in addition to supporting Polκ recruitment for repair synthesis.

We also probed the requirement for binding between PCNA and components of the MMR machinery by generating mutations in the interdomain connector loop (IDCL) of PCNA. Mutations in this region abrogate interactions with MSH6 and PMS2 in vitro, and cause MMR defects in yeast (Lee and Alani, 2006). We made the equivalent mutations in Xenopus PCNA^L126A,I128A (Figure S2E,F). However, we were not able to determine the contribution of the PCNA:MutSα interface to ICL repair using the this mutant, since binding between MutSα^WT and PCNA^L126A,I128A was enhanced, as measured by co-immunoprecipitation, compared to binding between MutSα^WT and PCNA^WT (Figure S2G). Accordingly, both PCNA^WT and PCNA^L126A,I128A were able to rescue ICL repair defects in PCNA-depleted extracts (Figure S2H,I).

Insights into nucleolytic processing of ICLs by MutLα and EXO1

Following sensing, an ICL is thought to be processed in two consecutive nucleolytic steps. First, dual incisions surrounding the ICL are made to produce an “unhooked” oligonucleotide, which is covalently tethered to the DNA by the ICL adduct. The unhooked lesion is then resected in a “trimming” reaction that facilitates synthesis past the adduct by translesion synthesis polymerases by eliminating the need for displacement synthesis (Roy et al., 2016). In replication-coupled ICL repair the XPF-ERCC1 endonuclease and SLX4 promote unhooking of ICLs (Klein Douwel et al., 2014), and the SNM1A exonuclease has been proposed to process unhooked ICL lesions (Wang et al., 2011).

We investigated the role of these nucleolytic reactions in RIR. We used nicked plasmids to identify that a 5′ incision 19 base pairs away from an ICL lesion was able to stimulate repair by 4-fold. This is in contrast to nicks placed 15 bp or 30 bp 3′ to the ICL or 34 bp 5′ to the ICL, all of which failed to stimulate repair (Figure 3B). This experiment demonstrates a striking mechanistic difference between RIR and MMR. MMR is stimulated symmetrically by either a 5′ or a 3′ nick in vitro (Constantin, 2005; Varlet et al., 1996). Our results also suggest that 5′ incision is a rate-limiting step during RIR, and that 2 distinct nucleases are required during ICL repair. Depletion-rescue experiments with MutLα strongly suggest that MutLα promotes one or both of these incisions. However, because it is generally thought that MutLα requires a pre-existing nick and interaction with PCNA to stimulate its otherwise latent endonuclease activity, it is possible that another, as yet unidentified nuclease, is responsible for the initial incision.

Our discovery that MutSα and MutLα participate in ICL repair highlights the versatile roles DNA repair proteins play in genome maintenance. This idea is further exemplified by our finding that EXO1 is required during RIR. EXO1 is a 5′→ 3′ exonuclease that is not only involved in MMR, but also plays critical roles in double-strand break repair and telomere maintenance (Tran et al., 2004). In all of these roles, EXO1 catalyzes digestion of DNA by hydrolyzing phosphodiester bonds between adjoining normal nucleotides. The requirement for EXO1’s nuclease activity during ICL repair (Figure 4C) suggests that EXO1 could process complex, chemically modified DNA substrates, including crosslinked bases. However, our in vitro experiments in which EXO1₃₅₂ was unable to bypass an ICL lesion strongly suggests that EXO1 alone cannot efficiently process ICLs (Figure 4D). However, it is conceivable that ICL processing by EXO1 requires association with other proteins through its C-terminus.

During replication-coupled ICL repair, SNM1A has been proposed to operate similarly to EXO1 in processing unhooked ICL lesions (Wang et al., 2011). In addition, there is evidence that the FAN1 nuclease can digest past ICL-containing oligonucleotides in vitro (Pizzolato et al., 2015; Wang et al., 2014). However, we find that FAN1 depletion in Xenopus HSS extracts has no effect on ICL repair (unpublished data).

Conclusions

DNA crosslinking agents are among the most widely prescribed anti-cancer drugs used in the clinic. Yet, acquired resistance remains is a significant limitation of these drugs (Deans and West, 2011). It seems likely that alterations to pathways that contribute to the repair of ICLs, the principle cytotoxic lesion generated by these drugs, may underlie acquired resistance. It is therefore critical to fully understand how cells respond to and repair ICLs.

We provide compelling evidence that the MMR pathway contributes to ICL repair. This is in agreement with a recent study that used an unbiased proteomic approach to study protein recruitment to psoralen-ICLs using quantitative mass spectrometry. This study revealed that MSH2, MSH6, PMS2, and EXO1 associate with cross-linked chromatin (Raschle et al., 2015). Importantly, recruitment of these proteins was resistant to geminin, suggesting that MMR proteins may play a role in ICL repair both in and beyond G0/G1 phase. Indeed, requirement for MSH2 and EXO1 has been reported in repair of nitrogen mustard ICLs in S-phase yeast (Barber et al., 2005), and in S-phase mammalian cells defective in the Fanconi Anemia pathway (Huang et al., 2011).

In this study, we provide mechanistic insights into how MMR proteins cooperate to accomplish ICL repair (Figure 6). We propose that ICLs that are recognized by MutSα are processed by MutLα and EXO1 in an incision-excision reaction, followed by DNA synthesis process that requires PCNA and Polκ. Further, our work stresses the importance of distinguishing between structurally distinct ICLs, which may be repaired in overlapping and/or divergent ways. Future experiments should address whether the sequence context of ICLs influences ICL recognition and repair.

Xenopus cell-free extract and depletions

Xenopus laevis frogs were handled in accordance with guidelines provided by the Institutional Animal Care and Use Committee at Columbia University, protocol AAAK0551. For details regarding the preparation of cell-free extracts and depleted extracts see Supplemental Experimental Procedures.

Preparation ICL plasmids

SJG-136 treated plasmids were prepared by incubating pBS with indicated concentrations of SJG-136 (NCI/Spriogen LTD) in 50 mM triethanolamine and 2 mM EDTA, overnight at 37°C. Plasmids were ethanol precipitated and resuspended in water. The quality and quantity of plasmids recovered were analyzed by agarose gel electrophoresis with ethidium bromide, Nanodrop measurement, and qPCR. The amount of plasmid used in plasmid pull-down experiments was normalized accordingly. The preparation of plasmids containing a single SJG-136, trimethylene, or NM-like ICL lesion is described in detail in Supplemental Experimental Procedures.

ICL repair and lac repressor plasmid pull-down experiments in Xenopus extracts

Lac repressor plasmid pull-down assays in un-depleted and depleted HSS extracts were performed as described previously (Williams et al., 2012). ICL repair assays in HSS were performed essentially as describe previously (Ben-Yehoyada et al., 2009; Williams et al., 2012). Each of these assays, as well as the MMR assay of a site-specific A:C mispair is described in detail in Supplemental Experimental Procedures.

Statistical Analysis

All statistical analyses were performed using GraphPad Prism 7 software using paired Student’s T-test or ANOVA. Data annotated as ns=p>0.05; (*)=p≤0.05; (**)=p≤0.01; (***)=p≤0.001

Protein Expression and Purification

Recombinant MutSα, MutLα, RPA, PCNA, and EXO1 were purified as described in Supplemental Experimental Procedures.

Reagents

All antibodies and reagents are listed in Supplemental Experimental Procedures.