Structure-Specific Endonucleases Xpf and Mus81 Play Overlapping but Essential Roles in DNA Repair by Homologous Recombination

Homologous recombination (HR) mediated double-strand break (DSB) repair is initiated by resection of DSBs and formation of 3′-single-strand overhangs, followed by polymerization of Rad51 (1, 2). The resulting nucleoprotein filaments, consisting of the 3′-single-strand tail and the polymerized Rad51, undergo homology search and pairing with the intact duplex DNA donor to form a displacement (D)-loop structure. Extensive strand exchange of the D-loop leads to the generation of HR intermediates. HR intermediates are processed into either crossover products or non-crossover products (3). HR intermediates including Holliday junction (HJ) are hereafter called joint molecules.

The processing of joint molecules is carried out by dissolution and resolution pathways. In the dissolution pathway, Sgs1 (Blm, an ortholog of Sgs1 in human), topoisomerase III, and RMI1/2 collaboratively catalyze the decatenation of HJs and generate non-crossover products (4–6). In the resolution pathway, structure-specific endonucleases cleave the HJs and produce crossover and non-crossover products, depending on the choice of cleaved strands at the 4-way junction (7). In Escherichia coli (E. coli), the RusA nuclease and the RuvABC complex cleave HJs symmetrically and play a key role in the resolution pathway (8–10). Recent work with eukaryotes has revealed 2 resolvases, Gen1 (11) and the Slx1–Slx4 complex (12–15), both of which can symmetrically cleave the 4-way junction. A third enzyme called Mus81, together with its partner Eme1/Mms4, has also been implicated in resolving HJs and the formation of crossover recombinants in both mitotic and meiotic recombination in yeast and mammals (16, 17). In vitro data show an ability of Mus81 to incise HJs asymmetrically; however a greater predilection for cleaving D-loops and nicked HJs suggests that, unlike “classical” HJ resolvases, Mus81 may process joint molecules before they mature into fully ligated HJs (16).

Mus81 is a member of the Xpf family of structure-specific endonucleases. Human Xpf is best known for its role in nucleotide excision repair (NER), together with its partner Ercc1. Moreover, Xpf, but not the other NER factors, is involved in DSB repair including single-strand annealing (18, 19), incision at interstrand crosslinks (20), and gene targeting (21). However, no studies have reported a functional overlap between Mus81 and Xpf in any DNA repair or recombination reactions, or a role of Xpf in HR after the formation of joint molecules. Intriguingly the Xpf orthologs in both Drosophila and Caenorhabditis elegans (C. elegans) have been implicated in joint molecules processing and the formation of crossovers during meiosis. Whether Xpf is similarly involved in joint molecules processing in vertebrates alongside enzymes like Mus81 is currently unknown.

To investigate the role of Xpf in HR, we conditionally disrupted the XPF gene in the chicken B lymphocyte line DT40 (22). Because this line does not possess the MUS81 gene, we propose that the loss of Xpf in DT40 cells is equivalent to mammalian cells deficient in both Xpf and Mus81. Deletion of XPF caused extensive chromosomal aberrations and cell death. However, this lethality was substantially reversed by ectopic expression of human Mus81 together with Eme1 (HsMus81-Eme1), indicating a compensatory relationship between Xpf and Mus81 in the maintenance of chromosomal DNA. The phenotypic analysis of Xpf-depleted DT40 cells indicated that a marked genomic instability may result from the defective processing of joint molecules. Our data uncovered that the Xpf and Mus81 endonucleases play overlapping and essential roles in completion of HR.

Chicken DT40, mouse embryonic stem [ES; wild-type (IB10) and MUS81^−/−], and HeLa cells were cultured as described previously (23–25). DT40 or ES cells have been maintained by S. Takeda or J. Essers since 1991 or 2004, respectively (22, 24). HeLa cells were obtained from K. Myung (National Human Genome Research Institute, Bethesda, MD) in 2007 (25). All cell lines were tested routinely for various criteria such as morphology, growth rate, and karyotype. Details of plasmid constructs and siRNAs are provided in Supplementary Materials and Methods.

After bromodeoxyuridine (BrdUrd) pulse-labeling in the cells exposed to tamoxifen, cell-cycle distribution was measured as described previously (26).

Chromosomal aberrations were measured as described previously (23). Briefly, the cells were exposed to tamoxifen for 1, 2, or 3 days, with 0.1 μg/mL of colcemid added for the last 3 hours of incubation before harvesting. To measure ionizing radiation (IR)-induced chromosomal aberrations, the cells were exposed to tamoxifen for 24 hours, and colcemid was added immediately after cells were irradiated with 0.3 Gy γ-rays. To test the response to camptothecin, cells were continuously exposed to tamoxifen for 33 hours; these cells were also treated with 100 nmol/L camptothecin for the last 9 hours and with colcemid for the last 3 hours before harvesting of mitotic cells.

Before measurement of Rad51 subnuclear foci, cells were exposed to tamoxifen for 2 days. Rad51 foci were visualized with anti-Rad51 antibody in untreated cells and in cells γ-irradiated with 4 Gy, 3 and 6 hours after treatment as previously described (26).

Sister-chromatid exchange (SCE) levels were measured as described previously (27). Prior to BrdUrd labeling, cells were exposed to tamoxifen for 30 hours. Cells were cultured in the presence of 10 μmol/L BrdUrd for 21 hours and treated with 0.1 μg/mL of colcemid for the last 3 hours before harvesting. Camptothecin (5 or 100 nmol/L) was added 9 hours before harvesting.

Chromosomal aberrations were measured as described previously (23). Briefly, transfection of siRNA into HeLa or mouse ES cells using lipofectamine RNAiMAX or lipofectamine 2000 was carried out according to the manufacturer's instructions, respectively. Forty-five hours after the transfection, HeLa cells were incubated for 1 hour with 0.2 μg/mL colcemid; after this, metaphase cells were collected by mitotic shake-off. Alternatively, 72 hours after the transfection, mouse ES cells were incubated for 2 hours with 0.1 μg/mL colcemid, after which metaphase cells were collected by mitotic shake-off.

We conducted 3 independent experiments for all of the data sets. The results were expressed as mean ± SD (growth curves and sensitivity to the genotoxic agents) or mean ± SEM (chromosomal aberrations and SCEs). Differences among the data were tested for statistical significance using the t test.

Additional details are provided in Supplementary Materials.

Chicken XPF cDNA encodes a putative 903 amino acid proteins, compared with the 905 amino acids of the human Xpf (Supplementary Fig. S1). The sequence identity between the 2 proteins is 76.8%. As expected, immunoprecipitants of tagged Xpf included Ercc1 (Supplementary Fig. S2A and S2B), indicating that Xpf associates with Ercc1 in DT40 cells. As there is an ortholog of Eme1 but not Mus81 registered in the chicken genome database (Supplementary Fig. S3A and S3B; ref. 28), we analyzed proteins that interact with Eme1, which forms a heterodimer with Mus81 in mammalian cells. Immunoprecipitants of tagged Eme1 included Xpf, but not Mus81 (Supplementary Fig. S4A), indicating the absence of a functional Mus81–Eme1 complex in DT40 cells. To verify the absence of Mus81–Eme1, we disrupted the EME1 gene (Supplementary Fig. S5A). EME1^−/− DT40 cells were able to proliferate and showed moderate sensitivity only to cisplatin (Supplementary Fig. S5B and S5C), whose phenotype is in marked contrast with the hypersensitivity of mouse EME1^−/− and MUS81^−/− ES cells to a wide variety of DNA-damaging agents (29, 30). We, therefore, concluded there was no Mus81 ortholog in DT40 cells.

The data indicate that the relationship among Xpf, Mus81, Eme1, and Ercc1 differs between DT40 and mammalian cells. The moderate sensitivity of EME1^−/− DT40 cells to cisplatin (Supplementary Fig. S5C) indicates that Xpf might form a functional complex with Eme1, whereas mammalian Mus81 and Xpf form a heterodimer with Eme1 and Ercc1, respectively. This theory is supported by the data that, consistent with a previous report (31), interaction of Xpf with Eme1 was confirmed by co-immunoprecipitation of recombinant proteins (Supplementary Fig. S4B and S4C). In addition to the presence of Xpf–Eme1, it should be noted that we could not formally exclude the possibility that a functional homolog of Mus81 is present in chicken cells. In conclusion, Xpf–Ercc1, but not Xpf–Eme1, plays the major role in DNA damage response in chicken cells.

We generated XPF gene-disruption constructs, which deleted amino acid coding sequences 1 to 148 together with the transcriptional promoter sequences (Fig. 1A; Supplementary Fig. S6A). Because we failed to establish XPF^−/− cells from XPF^+/− cells, we generated conditional Xpf-deficient cells using a chicken XPF transgene flanked by loxP-signal sequences (GdXPF-loxP), which is excised by the chimeric Cre recombinase carrying the tamoxifen-binding domain (Supplementary Fig. S6B). As the GdXPF-loxP transgene also carries a marker gene encoding the GFP, tamoxifen-mediated excision of the GdXPF-loxP transgene can be evaluated by monitoring the loss of the GFP signal (Supplementary Fig. S6C and S6D). We also confirmed the depletion level of XPF mRNA by reverse transcription (RT)-PCR (Supplementary Fig. S6E). We generated XPF^+/−GdXPF-loxP clones, and subsequently generated 2 XPF^−/−GdXPF-loxP clones, which displayed indistinguishable phenotypes.

We analyzed the proliferation kinetics of the cells with and without treatment with tamoxifen (Fig. 1B and C). The XPF^+/+, XPF^+/−GdXPF-loxP, and XPF^−/−GdXPF-loxP cells without the tamoxifen treatment divided every 8 hours (Supplementary Fig. S6F). By contrast, at approximately 2 days after treatment with tamoxifen, the XPF^−/− cells ceased to proliferate (Fig. 1B and C). To explore the cause of this mortality, we examined spontaneously arising chromosomal aberrations in mitotic cells. It should be noted that chromosomal breaks, hereafter, represent discontinuities, which appear on chromosomes in metaphase spreads as regions unstained by Giemsa and do not always result from DSBs. We found a dramatic increase in the number of chromosomal aberrations prior to cell death in the Xpf-deficient cells (Fig. 1D), an occurrence that is consistently observed in cells that have a severe defect in HR (26, 32–34), but is not observed in DT40 cells deficient in the XPA or XPG genes involved in NER (35, 36). To test whether the nuclease activity of Xpf is required for cellular viability, we created 2 cDNAs mutated at the catalytic center of Xpf, GdXPF(D674A) and GdXPF(D702A) [Supplementary Figs. S1 and S6G]. In contrast to a wild-type Xpf transgene, these mutant transgenes yielded no stable clones (Supplementary Fig. S6G), suggesting that the mutant Xpf proteins may interfere with the endogenous Xpf by competing for the association with Ercc1.

To analyze the role of Xpf in HR, we assessed the capability of HR-mediated DSB repair at 24 hours after tamoxifen treatment, when the cells were still able to proliferate exponentially. To selectively evaluate HR-mediated DSB repair, we measured the number of chromosomal aberrations in mitotic cells following exposure of cells to camptothecin, a DNA topoisomerase I poison, and to γ-rays in the G₂ phase. Camptothecin induces single-end breaks during replication, and the restart of replication requires HR with the intact sister chromatid (37). Similarly, γ-ray–induced chromosomal aberrations are repaired exclusively by HR in the G₂ phase in DT40 cells (38).

Xpf-depleted cells had a greater number of camptothecin-induced chromosomal aberrations than did Xpf-expressing cells (Fig. 2A). We next exposed an asynchronous population of cells to γ-rays and measured the number of chromosomal aberrations in cells that entered the M phase within 3 hours after irradiation. This protocol allows for the evaluation of DSB repair selectively during the G₂ phase, where HR plays a dominant role in DSB repair (38). Following irradiation, the total number of chromosomal aberrations in Xpf-deficient cells increased to 0.26 per cell, whereas the total number in XPF^+/+ cells increased only to 0.08 (Fig. 2B). Taken together, these results indicate that Xpf plays a key role in HR-mediated DSB repair.

To differentiate between the early and late steps of HR, we analyzed the formation of γ-ray–induced Rad51 foci (Fig. 2C). At 3 hours after IR, there was no change in the rate of Rad51 focus formation for Xpf-expressing or Xpf-depleted cells. However, the Rad51 foci continued to be formed by 6 hours in Xpf-depleted cells, whereas the foci were decreased in Xpf-expressing cells. This suggests that Xpf is required for the completion of HR after formation of the Rad51 nucleofilament. To assess the role of Xpf in the late steps of HR, that is, during joint molecules formation and processing, we measured the number of SCEs, which represent crossover-type HR (27). The number of camptothecin-induced SCEs in Xpf-depleted cells was only 40% of the number in Xpf-expressing cells (Fig. 2D), indicating that Xpf is required for crossover-type HR.

If the mortality of the XPF^−/− cells is caused by impaired completion of HR, it might be reversed by blocking the initiation of HR. To test this hypothesis, we disrupted the XRCC3 gene in XPF^−/−GdXPF-loxP cells. Xrcc3 facilitates an initial step of HR by promoting the polymerization of Rad51 at DNA lesions (39). To generate XPF^−/−/XRCC3^−/− cells, we exposed the resulting XPF^−/−GdXPF-loxP/XRCC3^−/− cells to tamoxifen. This mutant displayed lower levels of cell death (Fig. 3A) and a significant decrease in the number of chromosomal aberrations (Fig. 3B), compared with Xpf-depleted cells, although the XRCC3^−/− cells did display moderate genomic instability (23, 39). These observations support the conclusion that Xpf plays a critical role in the completion of HR-mediated repair after the polymerization of Rad51 at DNA lesions.

We next considered that if Xpf contributes to HR after the formation of joint molecules, the severe phenotype of Xpf-depleted cells might be suppressed by ectopic expression of E. coli RusA resolvase (40). To generate XPF^−/− cells stably expressing RusA, XPF^−/−GdXPF-loxP cells were transfected with RusA cDNA, and the resulting XPF^−/−GdXPF-loxP/RusA cells were exposed to tamoxifen. The RusA expression improved cellular viability (Fig. 3C) and decreased the number of spontaneous chromosomal aberrations in Xpf-depleted cells (Fig. 3D). Intriguingly, the reduction in chromosome-type breaks is associated with an increase in chromatid-type breaks. Similarly RusA expression resulted in a decrease in γ-ray–induced chromosome-type breaks without affecting the number of chromatid-type breaks (Fig. 3E). These data are consistent with the idea that chromosome-type breaks result from a failure to complete HR after the formation of joint molecules in Xpf-deficient cells.

The absence of both Xpf and Mus81 in XPF^−/− DT40 cells provides us with the novel opportunity of investigating the role of human Xpf–Ercc1 and Mus81–Eme1 in HR. To generate XPF^−/−/HsXPF-ERCC1 and XPF^−/−/HsMUS81-EME1 cells, XPF^−/−GdXPF-loxP cells were transfected with HsXpf-Ercc1 or HsMus81-Eme1 cDNA, and XPF^−/−GdXPF-loxP cells stably expressing HsXpf-Ercc1 or HsMus81-Eme1 were exposed to tamoxifen. Note that we failed to generate XPF^−/−/HsMUS81 cells, probably due to instability of HsMus81 as a consequence of poor association with GdEme1. HsXpf-Ercc1 reversed the mortality in XPF^−/− DT40 cells (Fig. 4A). Remarkably, HsMus81-Eme1 also significantly restored the cellular proliferation of XPF^−/− DT40 cells to a level comparable with the cells complemented with HsXpf-Ercc1 (Fig. 4C), although the amino acid sequence identity between GdXpf and HsMus81 is only 9.7%. Next, we analyzed XPF^−/−/HsXPF-ERCC1 and XPF^−/−/HsMUS81-EME1 cells by counting both spontaneous (Fig. 4B and D) and γ-ray–induced chromosomal aberrations (Fig. 5A and B). In all cases, the ectopic expression of the human enzymes suppressed the chromosomal aberrations caused by Xpf deficiency and, particularly, chromosome-type breaks. These observations indicate that HsMus81–Eme1 and HsXpf–Ercc1 have very similar functions in HR-mediated DNA repair and genome maintenance.

To determine whether HsMus81–Eme1 promotes crossover-type HR, we counted the number of SCEs in XPF^−/−/HsMUS81-EME1 cells. The expression of HsMus81–Eme1 restored the number of camptothecin-induced SCEs (Fig. 5C) to the number of SCEs in GdXpf-expressing cells (Fig. 2D). We conclude that Mus81 and Xpf have very similar functions in promoting crossover-type HR.

Finally, we tested whether Xpf and Mus81 compensate for each other in the completion of HR in human and mouse cell lines. We depleted Xpf and Mus81 in human HeLa cells or depleted Xpf in mouse MUS81^−/− ES cells, using siRNA (Supplementary Fig. S7A and 7B). In HeLa cells, the depletion of both Xpf and Mus81, but not the depletion of either Xpf or Mus81 alone, increased the number of spontaneous chromosome-type breaks (Fig. 6A). Similarly, depletion of Xpf in mouse MUS81^−/− ES cells greatly increased the number of spontaneous chromosome-type breaks, compared with Xpf-depleted wild-type and mock-transfected MUS81^−/− ES cells, while they showed only a moderate increase in the number of spontaneous chromosome-type breaks (Fig. 6B). These results indicate that the role of Xpf in HR is conserved in mammalian and DT40 cells.

We have shown that the Xpf and Mus81 structure-specific endonucleases compensate for each other in the maintenance of chromosomal DNA. This compensatory relationship is verified by the following phenotypic analysis of Xpf-deficient DT40 cells, which lack a MUS81 ortholog. First, the expression of HsMus81–Eme1 as well as HsXpf–Ercc1 reversed the mortality of Xpf-deficient DT40 cells (Figs. 4 and 5). Second, the mortality of Xpf-deficient DT40 cells is in marked contrast to the viability of Xpf-deficient mammalian cells (41, 42) and to the normal development of Mus81-deficient mice (24, 43). Third, although chicken Xpf and Eme1 physically interact with each other (Supplementary Fig. S4B and S4C), GdXpf-Eme1 has a minor role in genome maintenance (Supplementary Fig. S5), and does not account for mortality of Xpf-deficient DT40 cells. Finally, our findings using DT40 cells are applicable to mammals as evidenced by the fact that inactivation of both Xpf and Mus81 caused increased the number of chromosome-type breaks in human HeLa and mouse ES cells (Fig. 6). These observations show that the 2 structure-specific endonucleases have substantially overlapping functions in the maintenance of genomic DNA.

We have provided a few different lines of evidence that collectively indicate that Xpf contributes to HR-dependent DSB repair, probably after the formation of joint molecules made of broken sister chromatids and the other intact ones. First, the inactivation of Xpf resulted in prolonged Rad51 foci formation (Fig. 2C). Second, the expression of Xpf and Mus81 enhanced the formation of SCEs (Figs. 2D and 5C). Third, preventing the initiation of HR by deleting XRCC3 suppressed the mutant phenotype of Xpf-deficient DT40 cells (Fig. 3A and B). These results indicated that Xpf plays a role in HR at a late step. Conceivably, the inactivation of Xpf may cause a defect in the separation of joint molecules before mitosis that is more toxic than the defective formation of joint molecules due to the absence of XRCC3.

The loss of viability in Xpf-deficient DT40 cells is associated with chromosome aberrations, which appear on metaphase spreads as regions unstained by Giemsa. Throughout, we have referred to these as chomatid-type breaks and chromosome-type breaks depending on whether the discontinuity in staining affects 1 or both sister chromatids, respectively. In the case of chromatid-type breaks, we suspect that the absence of Giemsa staining actually represents a DSB in the sister chromatid as these aberrations are also observed in mutants that are defective for early steps of HR (44). However, we think that the chromosome-type breaks represent regions of uncondensed DNA rather than actual broken chromosomes similar to what was recently reported (45). It is proposed that regions devoid of Giemsa staining occur at sites of sister chromatid entanglement, which inhibits chromosome condensation. Importantly, the chromosome-type breaks that we observe in Xpf-deficient cells both spontaneously and following γ-ray exposure are suppressed by ectopic expression of the HJ resolvase RusA (Fig. 3D and E), providing strong evidence that they result from unprocessed joint molecules. The fact that ectopic expression of HsXpf–Ercc1 and HsMus81–Eme1 also suppress chromosome-type breaks in Xpf-deficient DT40 cells suggests that there is a significant overlap in function between these enzymes that is most likely related to processing joint molecules such as D-loops and nicked single HJs.

Cells deficient in Xpf–Ercc1 are considerably more sensitive to chemical cross-linking agents than cells deficient in the other NER factors, indicating the critical role for Xpf–Ercc1 in interstrand cross-link (ICL) repair. Although there is compelling evidence that the critical role played by Xpf is carried out by introducing single-strand breaks at cross-links to initiate ICL repair (20, 46), Xpf may function in ICL repair also by facilitating HR. We recently found that Slx4 links the FA-dependent ICL pathway with both Mus81 and Xpf, as Slx4 serves as docking sites for the 2 nucleases and the ubiquitinated FancD2 protein, which agrees with recent reports showing FA patients carrying mutations in the SLX4 gene (47–49). Taking into account the fact that Slx4 also binds to the Slx1 5′-flap endonuclease in addition to the Mus81–Eme1 and Xpf–Ercc1 3′-flap endonucleases (Supplementary Fig. S2A and S4D; refs. 12–15), these endonucleases may collaboratively work in ICL repair by promoting the completion of HR. In this scenario, it is not surprising that the 2 3′-flap endonucleases are complementary to each other. Future studies will analyze the interdependent and complementary relationships of multiple endonucleases when they carry out a variety of DNA repair reactions.

No potential conflicts of interest were disclosed.

Conception and design: K. Kikuchi, R. Kanaar, Y. Taniguchi, S. Takeda

Development of methodology: K. Kikuchi, M.C. Whitby

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Kikuchi, T. Narita, J. Iijima, K. Hirota, I.S. KeKa, M. Mohiuddin, K. Okawa, T. Fukagawa, J. Essers, Y. Taniguchi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Kikuchi, T. Narita, V.T. Pham, I.S. KeKa, M. Mohiuddin, K. Okawa, R. Kanaar, S. Takeda

Writing, review, and/or revision of the manuscript: K. Kikuchi, T. Narita, V.T. Pham, R. Kanaar, M.C. Whitby, S. Takeda

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Kikuchi, T. Hori, K. Sugasawa, S. Takeda

Study supervision: K. Kikuchi, R. Kanaar, Y. Taniguchi, K. Kitagawa, S. Takeda

The authors thank R. Ohta and Y. Satoh for their technical assistance. The authors also thank Dr. M. Lisby for sharing unpublished data.

Financial support was provided in part by a grant-in-aid from the Ministry of Education, Culture, Sports, Science and Technology of the Japanese Government (MEXT; grant no. 20241012 to S. T. and grant no. 22700881 to K. Kikuchi), a grant from The Kanae Foundation (K. Kikuchi), grants from the European Community's Seventh Framework Programme (FP7/2007–2013) under grant agreement no. HEALTH-F2-2010-259893, and from the Netherlands Genomics Initiative/Netherlands Organization for Scientific Research (R. Kanaar), and NIH grants GM68418 and CA133093 (K. Kitagawa). Funding was also provided by The Wellcome Trust (090767/Z/09/Z; M.C. Whitby).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.