Genetic evidence for the involvement of mismatch repair proteins, PMS2 and MLH3, in a late step of homologous recombination

MdMaminur Rahman, Mohiuddin Mohiuddin , Islam Shamima Keka, Kousei Yamada, Masataka Tsuda, Hiroyuki Sasanuma, Jessica Andreani , Raphael Guerois, Valerie Borde , Jean-Baptiste Charbonnier, and Shunichi Takeda* From the Department of Radiation Genetics, Graduate School of Medicine, Kyoto University, Yoshida-Konoe-cho, Kyoto, Japan, the Institute for Integrative Biology of the Cell (I2BC), Commissariat à l'Energie Atomique (CEA), CNRS, Université Paris-Saclay, Gif-sur-Yvette, France, and the Institut Curie, CNRS UMR3244, PSL Research University, Paris, France

Homologous recombination (HR) repairs DNA double-strand breaks using intact homologous sequences as template DNA. Broken DNA and intact homologous sequences form joint molecules (JMs), including Holliday junctions (HJs), as HR intermediates. HJs are resolved to form crossover and noncrossover products. A mismatch repair factor, MLH3 endonuclease, produces the majority of crossovers during meiotic HR, but it remains elusive whether mismatch repair factors promote HR in nonmeiotic cells. We disrupted genes encoding the MLH3 and PMS2 endonucleases in the human B cell line, TK6, generating null MLH3 2/2 and PMS2 2/2 mutant cells. We also inserted point mutations into the endonuclease motif of MLH3 and PMS2 genes, generating endonuclease death MLH3 DN/DN and PMS2 EK/EK cells. MLH3 2/2 and MLH3 DN/DN cells showed a very similar phenotype, a 2.5-fold decrease in the frequency of heteroallelic HR-dependent repair of restriction enzymeinduced double-strand breaks. PMS2 2/2 and PMS2 EK/EK cells showed a phenotype very similar to that of the MLH3 mutants. These data indicate that MLH3 and PMS2 promote HR as an endonuclease. The MLH3 DN/DN and PMS2 EK/EK mutations had an additive effect on the heteroallelic HR. MLH3 DN/DN /PMS2 EK/EK cells showed normal kinetics of g-irradiation-induced Rad51 foci but a significant delay in the resolution of Rad51 foci and a 3-fold decrease in the number of cisplatin-induced sister chromatid exchanges. The ectopic expression of the Gen1 HJ resolvase partially reversed the defective heteroallelic HR of MLH3 DN/DN /PMS2 EK/EK cells. Taken together, we propose that MLH3 and PMS2 promote HR as endonucleases, most likely by processing JMs in mammalian somatic cells.
A subclass of the MMR proteins is involved in double-strand break (DSB) repair. First, MutS complexes play a role in the rejection of heteroduplex DNA containing insertion/deletion mismatches when the nucleotide sequences of two partner DNAs are not identical (16,17). Second, MutSa may recognize mismatches within the heteroduplex region of the JMs and avoid recombination, collaborating with RecQ helicases (18). Third, MLH1 can affect nonhomologous end-joining (NHEJ) (19), which repairs 80% of the ionizing radiation-induced DSB in the G 2 phase (20). Fourth, a subset of MSH and MLH proteins promote meiotic HR, a function distinct from their MMR functions. MLH1-MLH3, which has a minor role in MMR, is critical for producing meiotic crossover products in mice and Saccharomyces cerevisiae (21,22). MLH1, MLH3, and PMS2 are essential for the progression of meiotic HR in mice (23)(24)(25)(26)(27)(28). The role played by the putative endonuclease activity of PMS2 in the resolution of meiotic HR intermediates has not yet been clarified in mice or humans, but recent studies have unveiled new insights into the molecular mechanisms of MLH1-MLH3 and the role of its endonuclease activity (22,29,30). Another unsolved question is whether MLH3 and PMS2 promote HR in mammalian somatic cells.
HR initiates DSB repair by resecting DSBs, leading to the formation of 39 single-strand overhangs, followed by polymerization of Rad51 on the single-strand DNA (31)(32)(33). The resulting Rad51 nucleoprotein filaments undergo homology search and pairing with the intact duplex DNA donor to form joint molecules (JMs) such as double Holliday junctions (dHJs) with the help of Rad54 (33)(34)(35). JMs are resolved into individual DNA duplexes to allow chromosomes to separate in the anaphase.
The separation is performed by two alternative processes, the dissolution and resolution pathways. The phenotypic analysis of meiotic HR indicates that only 10% of the DSBs (Mus musculus) form dHJs, and these are almost exclusively processed by the resolution pathway, involving the activity of MLH1-MLH3 (22). In somatic cells, the resolution of HJs is done by a number of structure-specific endonucleases, MUS81-EME1, SLX1-SLX4, XPF-ERCC1, and GEN1 (36)(37)(38)(39). Mice deficient in either MUS81-EME1 or SLX1-SLX4 or GEN1 are all viable, whereas mice deficient in both MUS81-EME1 and GEN1 are synthetic lethal (40)(41)(42)(43), suggesting a substantial functional overlap between the two nucleases. Although yeast genetic studies have precisely monitored the formation of HR intermediate molecules such as HJs over time upon DSB formation during both meiosis and mitosis (21,34,44,45), no equivalent phenotypic assays are available in the phenotypic analysis of HR in mammalian somatic cells.
There are two major DSB repair pathways in mammalian cells, HR and NHEJ. The two pathways differentially contribute to cellular tolerance to anti-malignant therapies. These pathways contribute to tolerance to radiotherapy with HR functioning in the S to G 2 phases and NHEJ functioning in the whole cell cycle (46). HR, but not NHEJ, repairs DSBs induced by camptothecin (Top1 poison) and olaparib (poly(ADP-ribose) polymerase poison). NHEJ plays the dominant role in repairing DSBs induced by ICRF-193 (catalytic inhibitor of Top2) (47,48). Thus, the sensitivity profile of DSB-repair mutants to these chemotherapeutic agents helps to discriminate which repair pathway is compromised in the mutants.
To investigate the role for MLH3 and PMS2 as nucleases in DSB repair of somatic mammalian cells, we inserted a point mutation into the DQHAX 2 EX 4 E motif of the endogenous MLH3 and PMS2 genes of the human TK6 B cell line (49) and generated MLH3 D1223N/D1223N and PMS2 E705K/E705K cells. These mutants exhibited increased sensitivities to camptothecin and olaparib, a few-fold decrease in the frequency of both sister chromatid exchange (SCE) and the heteroallelic HR, and delayed resolution of g-ray-induced Rad51 foci, indicating a defect in HR in later steps. Surprisingly, their role seems to be mostly independent of MLH1. We conclude that the MLH3 and PMS2 proteins promote DSB repair by HR, presumably by processing JMs in human cells.
The repair of g-ray-induced DSBs during G 2 phase is severely compromised in the PMS2 and MLH3 mutant cells To monitor DSB repair selectively during the G 2 phase when HR is active, we exposed cells to ionizing radiation and measured the number of chromosomal aberrations in mitotic chromosome spreads at 3 h after ionizing radiation (59). Only cells that were g-irradiated at the G 2 phase, but not the S phase, can enter the M phase within 3 h (60). This method allows for evaluating the capability of HR to repair DSBs with several times higher sensitivity than the analysis of the g-irradiation sensitivity of asynchronous cell populations (Fig. 1, A and B). Indeed, g-irradiation increased the number of chromosomal breaks by 1.0 per MUS81 2/2 cell and only 0.2 per WT cell (Fig. 2B). Remarkably, the total numbers of chromosome aberrations induced by g-rays were around 10 times higher in the PMS2 and MLH3 mutant cells compared with WT cells (Fig. 2B). The total number of mitotic chromosome aberrations was significantly higher in MLH1 2/2 cells, but not in MSH2 2/2 cells, compared with WT cells (Fig. 2B and Fig. S4K). We conclude that there is no significant contribution of canonical MMR involving MSH2 to DSB repair during the G 2 phase. The total numbers of g-ray-induced chromosome aberrations increased to very similar extents in the five mutants, PMS2 2/2 , PMS2 EK/EK , MLH3 2/2 , MLH3 DN/DN , and MLH3 EK/EK cells (Fig. 2B). These data suggest that PMS2 and MLH3 significantly contribute to DSB repair as endonuclease.
We counted the number of chromosome aberrations distinguishing chromatid-type breaks (where one of the two sister chromatids is broken), isochromatid-type breaks (where two sister chromatids are broken at the same sites), and radial chromosomes (which comprise the association of two or more chromatids) ( Fig. 2A). Ionizing irradiation of RAD54 2/2 cells caused a more significant increase in the number of chromatid-type breaks than that of isochromatid-type breaks (Fig. 2B). This observation agrees with the role of Rad54 in promoting strand exchange and JM formation. In contrast, MUS81 2/2 cells showed marked increases in the numbers of isochromatid-type MLH3 and PMS2 mutants are sensitive to camptothecin, g-rays, and olaparib. A, clonogenic cell survival assay following exposure of PMS2 mutants to camptothecin, g-rays, and olaparib (PARP inhibitor). The x axis represents the dose of the indicated DNA-damaging agent on a linear scale; the y axis represents the survival fraction on a logarithmic scale. Error bars, S.D. for three independent assays. Statistical analyses were performed by Student's t test (*, p , 0.01). B, clonogenic cell survival assay following exposure of MLH3 mutants to camptothecin, g-rays, and olaparib (PARP inhibitor). Cellular sensitivity is shown as in A. Statistical analyses were performed by Student's t test (*, p , 0.01). C, clonogenic cell survival assay following exposure of MUS81 2/2 PMS2 2/2 mutants to olaparib (PARP inhibitor). Cellular sensitivity is shown as in A. D, PMS2 EK/EK /MLH3 DN/DN double mutant cells show stronger HR defects than PMS2 EK/EK and MLH3 DN/DN cells. Shown is a clonogenic cell survival assay following exposure of PMS2 EK/EK , MLH3 DN/DN , and PMS2 EK/EK MLH3 DN/DN mutants to camptothecin, g-rays, and olaparib (PARP inhibitor). Cellular sensitivity is shown as in A. Statistical analyses were performed by Student's t test (*, p , 0.05).
breaks. Isochromatid-type breaks result from abnormal processing of JMs between broken and intact sister chromatids, as the persistent presence of JMs interferes with local chromosome condensation of both sister chromatids, leading to microscopically visible breakage of the two chromosomes at the same sites (39,61,62). Radial chromosomes may be caused by the abnormal separation of JMs containing two sisters, leading to inverted chromosome fusions. Like MUS81 2/2 cells, the PMS2 and MLH3 mutants showed significant increases in the numbers of both isochromatid-type breaks and radial chromosomes (Fig. 2B). These data suggest that PMS2 and MLH3 promote HR-dependent DSB repair after formation of JMs as does MUS81.
Surprisingly, the MLH1 2/2 phenotype is not as severe as expected from the phenotypes of the PMS2 and MLH3 mutants, particularly no significant alteration regarding isochromatid-type breaks. A moderate increase in the number of chromatid-type breaks in MLH1 2/2 cells suggests that the MLH1-MLH3 and MLH1-PMS2 heterodimers may play a minor role in NHEJ-mediated DSB repair, as suggested previously (19). One possible scenario is that an MLH1-independent alternative mechanism of PMS2 and MLH3 might be present in the process observed in this study. These data suggest that PMS2 and MLH3 promote HR-dependent DSB repair after the formation of JMs, as does MUS81.  for MLH3 DN/DN , and 14.3 h for PMS2 EK/EK /MLH3 DN/DN cells. The plating efficiency of these cells was 50-60% for all genotypes. PMS2 EK/EK /MLH3 DN/DN cells showed high sensitivity to camptothecin and g-rays, higher than MLH3 DN/DN and slightly higher than PMS2 EK/EK , suggesting a prominent role of PMS2 in these assays (Fig. 1D). PMS2 EK/EK /MLH3 DN/DN cells also showed a higher sensitivity to olaparib than did PMS2 EK/EK and MLH3 DN/DN cells (Fig. 1D). The number of g-ray-induced chromosomal breaks was more than 50% higher in PMS2 EK/EK /MLH3 DN/DN cells than in PMS2 EK/EK and in MLH3 DN/DN cells (Fig. 2). We therefore conclude that PMS2 and MLH3 contribute to HR as the endonuclease independently of each other.
We monitored DSB repair kinetics by measuring the number of gH2AX foci with time after g-irradiation (Fig. 3). The numbers of gH2AX foci were very similar among MLH3 DN/DN , PMS2 EK/EK , and WT cells at 2 h after ionizing irradiation. The numbers of gH2AX foci reduced more slowly in MLH3 DN/DN and PMS2 EK/EK cells compared with WT and MLH1 2/2 cells (Fig. 3). The delayed DSB repair kinetics observed more than 2 h after ionizing irradiation is consistent with the fact that HR needs a longer time to complete DSB repair than does NHEJ (20). PMS2 EK/EK /MLH3 DN/DN cells showed a more prominent delay in DSB repair at 8 h compared with MLH3 DN/DN and PMS2 EK/EK cells (Fig. 3). We conclude that PMS2 and MLH3 promote DSB repair independently of each other in an MLH1independent manner.

Resolution of g-ray-induced Rad51 foci is delayed in the PMS2 and MLH3 mutant cells
To evaluate whether PMS2 and MLH3 act in the early and late steps of HR, we analyzed the formation of Rad51 foci over time after g-irradiation (Fig. 4). The number of Rad51 foci peaked at 2 h after g-irradiation in WT TK6 cells (59,60). MLH3 DN/DN , PMS2 EK/EK , and PMS2 EK/EK /MLH3 DN/DN cells showed the same extent of Rad51 foci at 2 h as WT cells. Thus, PMS2 and MLH3 are dispensable for DSB resection and the polymerization of Rad51 on resected DSBs. Remarkably, MLH3 DN/DN , PMS2 EK/EK , and PMS2 EK/EK /MLH3 DN/DN cells showed a significant delay in the resolution of Rad51 foci compared with WT and MLH1 2/2 cells (Fig. 4). Both MLH3 DN/DN and PMS2 EK/EK single mutants showed a similar delay in the resolution of Rad51 foci compared with MUS81 2/2 cells, but this effect was more prominent in the PMS2 EK/EK /MLH3 DN/DN double mutant cells (Fig. 4B). All mutants were less sensitive than RAD54 2/2 cells. We therefore conclude that the PMS2 and MLH3 endonucleases promote HR-dependent DSB repair after the polymerization of Rad51 at DSBs.

MLH3 DN/DN and PMS2 EK/EK cells are deficient in heteroallelic HR
To assess the involvement of PMS2 and MLH3 in the resolution of HJs, we measured the frequency of heteroallelic recombination between the allelic thymidine kinase (TK) genes carrying compound heterozygous mutations (47, 50, 60, 63) (Fig. 5A). One of the two allelic TK genes carries an I-SceI site, and a mutation in the exon 5 localizes 108 nucleotides downstream of the I-SceI site. When I-SceI-induced DSBs are repaired by either the gene conversion (HR) that associates with crossover or long-tract gene conversion, it can restore an intact TK gene. These restoration events are detectable by counting the frequency of drug-resistant colonies (50). The HR frequency was 60% lower in MUS81 2/2 cells compared with WT cells (Fig. 5B), suggesting that a majority of the heteroallelic recombination events involve the formation of HJs. The PMS2 and MLH3 mutants, including MLH3 DN/DN and PMS2 EK/EK cells, showed 60-70% decreases, and PMS2 EK/EK /MLH3 DN/DN and MUS81 2/2 /PMS2 2/2 cells showed further declines in the frequency of crossover events when compared with WT cells (Fig. 5B). These observations suggest that the endonuclease activity of PMS2 and MLH3 may be involved in the resolution of HJs. This function of PMS2 and MLH3 is also independent of MLH1 and MSH2.
We further assessed the involvement of PMS2 and MLH3 in the resolution of HJs by measuring SCE events, crossover-type HR (62,64,65). To induce SCE, we treated cells with cisplatin, an interstrand cross-linking agent. The number of cisplatin-induced SCE events was measured by subtracting the number of SCEs before cisplatin treatment from the number of SCEs post-treatment (Fig. 5, C and D). The treatment increased the SCE frequency by 11 events per 100 mitotic WT cells (Fig. 5D). The number of induced SCEs was 50% smaller in MUS81 2/2 cells when compared with WT cells. MLH3 DN/DN and PMS2 EK/EK cells also showed 50% decreases, and PMS2 EK/EK /MLH3 DN/DN cells showed an 80% decrease in the SCE compared with WT cells (Fig. 5, C and D). In summary, PMS2 and MLH3 contribute to crossover formation most likely by promoting the resolution of HJs, as does MUS81, but MLH1 is not involved in this process.
The loss of MLH1 does not impair HR-dependent DSB repair MLH1 physically interacts with PMS2 and MLH3 as heterodimers and thereby stabilizes the two endonucleases (66). Here, to evaluate the role of MLH1 in the mitotic HR, we have employed five phenotypic assays: (i) sensitivity to camptothecin, g-irradiation, and olaparib ( Fig. 1A and Fig. S4J), (ii) g-rayinduced chromosome aberrations ( Fig. 2B and Fig. S4K), (iii) measuring the number of gH2AX and Rad51 foci over time after g-irradiation (Figs. 3 and 4), (iv) measuring the frequency of the heteroallelic recombination (Fig. 5, A and B), and (v) SCE induced by cisplatin (Fig. 5, C and D). Unexpectedly, as mentioned above, all of these phenotypic assays consistently showed that MLH1 2/2 cells were proficient in HR-mediated Role of PMS2 and MLH3 in homologous recombination DSB repair. We therefore conclude that MLH1 is dispensable for the functioning of human MLH3 and PMS2 in mitotic HR. It represents to our knowledge the first example where MLH1 is not required for the functioning of MLH3 and PMS2. Indeed, MLH1 has been shown to be required for the functioning of MLH3 and PMS2 in MMR and in meiotic HR in mice (24,25). We speculated that the MLH1-independent function of PMS2 and MLH3 in mitotic HR can be achieved through (i) homodimer formation, (ii) formation of a heterodimer with an unknown partner protein, which would stabilize the PMS2 and MLH3 proteins, and (iii) formation of an MLH3-PMS2 heterodimer. We could not examine these possibilities due to the lack of specific antibodies and no appropriate method of inserting functional tag sequences into PMS2 and MLH3. We therefore investigated PMS2 and MLH3 homodimer and heterodimer formation through 3D structure modeling using a standard homology modeling pipeline based on the HHpred and Roset-taCM methods (67)(68)(69). Structural analysis of the resulting models supports the potential homodimer and heterodimer formation (Fig. S8, A-C).
Significant rescue of the defective HR of the MLH3 and PMS2 mutants by ectopic expression of GEN1 We reason that if the PMS2 and MLH3 endonuclease activities promote HR by processing HJs, the mutant phenotype of MLH3 DN/DN and PMS2 EK/EK cells could be suppressed by ectopic expression of one of the resolvases described for HJs. We chose GEN1 as the HJ resolvase (70) and used the GEN1 transgene carrying mutations in its nuclear export signal (NES) and fused with the nuclear localization signal (NLS) (71) (Fig. 6A). We added the FLAG tag to this GEN1 transgene and inserted it into the pMSCV retroviral expression vector, which allows for the bicistronic expression of the GFP and GEN1 transgenes (72) (Fig. S7). We produced recombinant retrovirus and infected them into TK6 clones. To confirm the expression of the transgene, we performed Western blotting analyses using an anti-FLAG antibody (Fig. S7D). We measured the ionizing radiation sensitivity and calculated LD 50 , the dose of g-rays that reduced the survival of cells to 50% relative to nonirradiated cells (Fig. 6B). The expression of the GEN1 transgene reversed the ionizing radiation sensitivity of MUS81 2/2 cells, but not WT or RAD54 2/2 cells. Thus, the GEN1 transgene is able to selectively normalize the defective processing of JMs during HR-mediated DSB repair.
The GEN1 transgene restored the tolerance of MLH3 DN/DN , PMS2 EK/EK , and PMS2 EK/EK /MLH3 DN/DN cells to g-rays at least partially (Fig. 6B). The rescue effect of GEN1 transgene was more efficient in PMS2 EK/EK /MLH3 DN/DN double mutant cells compared with MLH3 DN/DN and PMS2 EK/EK cells. In agreement with this finding, the GEN1 transgene significantly reduced the total number of chromosomal aberrations in these mutants as well as MUS81 2/2 cells (Fig. 6C). Importantly, the GEN1 transgene expression reduced the number of isochromatid-type breaks to a considerably greater extent than that of chromatid breaks (Fig. 6C). The GEN1 transgene increased the frequency of heteroallelic HR in MUS81 2/2 cells by 60% but had no effect on that in WT or RAD54 2/2 cells (Fig. 6D), suggesting that a substantial fraction of heteroallelic HR involves HJ formation as HR intermediates. The GEN1 transgene restored heteroallelic HR in MLH3 DN/DN , PMS2 EK/EK , PMS2 EK/EK /MLH3 DN/DN , and MUS81 2/2 cells but not WT or RAD54 2/2 cells (Fig. 6D). In summary, the PMS2 and MLH3 endonuclease activities facilitate the separation of HJs.

Discussion
We demonstrate that human PMS2 and MLH3 promote DSB repair by HR in human somatic cells. Previous studies failed to uncover their role in the repair of X-ray-induced DSBs, presumably because murine primary cells deficient in PMS2 are slightly resistant to ionizing radiation due to defective MMR of damaged nucleotides (73). Strikingly, the defective HR phenotype of the PMS2 and MLH3 mutants derived from the TK6 cell line was as prominent as that of TK6 cells deficient in MUS81, an important endonuclease involved in the resolution of HJs (36,37) (Figs. 2B and 5 (B,  C, and D)). Furthermore, PMS2 EK/EK /MLH3 DN/DN cells displayed a significantly stronger phenotype than did MUS81 2/2 cells, including 15 times more mitotic chromosome breaks induced by g-irradiation at the G 2 phase (Fig. 2B) and an ;80% decrease in the number of cisplatin-induced sister chromatic exchanges (Fig. 5, C and D) compared with WT cells. The contribution of PMS2 and MLH3 to HR is totally independent of their functioning in MMR because MSH2 and MLH1 are required for MMR but dispensable for HR (Fig. 2). In summary, human PMS2 and MLH3 significantly contribute to the genome stability of somatic cells through at least two distinct mechanisms: MMR and DSB repair by HR.
The present study shows compelling genetic evidence for the requirement of the PMS2 and MLH3 endonuclease activity for the efficient resolution of HJs. MLH3 DN/DN and MLH3 2/2 cells showed the same phenotype in the defective HR (Fig. 2). Likewise, the phenotype of PMS2 EK/EK cells was very similar to that of PMS2 2/2 cells (Fig. 2). These data indicate that PMS2 and MLH3 promote HR as the endonuclease. In the MLH3 DN/DN and PMS2 EK/EK mutants, the initial kinetics of g-ray-induced Rad51 focus formation was normal, whereas its resolution was significantly delayed (Fig. 4). We therefore conclude that the PMS2 and MLH3 endonuclease activities promote a late step of HR, most likely after the formation of JMs. The MUS81 2/2 , MLH3 DN/DN , and PMS2 EK/EK mutants all showed a ;40% decrease in the frequency of cisplatin-induced SCEs (Fig. 5D). Furthermore, ectopic expression of GEN1, a typical HJ resolvase, reversed the defective heteroallelic HR of MUS81 2/2 , MLH3 DN/DN , and PMS2 EK/EK cells by 30-50% (Fig. 6). In conclusion, the endonuclease activity of PMS2 and MLH3 process HJs, generating both crossover and noncrossover products.
The PMS2-MLH1 and MLH3-MLH1 heterodimers are involved in both MMR and meiotic HR in S. cerevisiae and mice (21,24,25,66). Unexpectedly, we observed that only PMS2 and MLH3, and not MLH1, are involved in HR in human somatic cells. The PMS2 and MLH3 proteins may form homodimers and heterodimers when they are involved in HR in the same manner as the MutL homologs form heterodimer mediated by their C-terminal region (14). Indeed, homodimers of yeast Mlh1 have been reported, and an increase of their formation can inhibit MMR (74). In addition, in support of a possible heterodimer formation, a recent study in budding yeast found co-immunoprecipitation of Mlh3 with Pms1 (75). The crystal structure of the C-terminal region of human MLH1 (Protein Data Bank code 3RBN) showed that human MLH1 could form homodimers with the same residues involved in the heterodimer formation (14). The HHpred and RosettaCM analysis also suggested that both the homodimer and heterodimer formation are possible (Fig. S8). Future studies will demonstrate the homodimer and heterodimer formation as well as the interaction with an unidentified partner protein.
MLH3 and PMS2 have strong tumor suppressor activities, and it is believed that these activities are attributable exclusively to their function of MMR (76). The current study suggests that these endonucleases may contribute to tumor suppression also by promoting the resolution of HJs. The MUS81-EME1, SLX1-SLX4, and GEN1 endonucleases all play a critical role in genome maintenance, particularly when Bloom helicase is attenuated (62,77,78). Nonetheless, it remains unclear how much these endonucleases contribute to tumor suppression in humans. A defect in the resolution of HJs can pose a more serious threat to genome stability compared with the initial step of HR because the former deficiency not only leaves DSBs unrepaired but also can cleave intact sister chromatids (79). The following two mouse experiments suggest the critical role played by HJ resolvases in tumor suppression. SLX4 serves as a docking site for MUS81-EME1, SLX1-SLX4, and XPF-ERCC1 endonucleases and MSH2-MSH3 mismatch repair factor (79). SLX4 plays a dominant role in preventing carcinogenesis, as evidenced by the data indicating that the loss of SLX4 decreases the median survival time of mice to ;90 days due to enhanced tumorigenesis (80). The MUS81 null mutation reduces the life expectancy of p53 null mice by about 30% due to an increase in carcinogenesis (81). The critical role of MLH3 and PMS2 in the resolution of HJs emphasizes their strong tumor suppressor activities in addition to their function in MMR (21,22,82).
In this study, we characterized a major role of MLH3 and PMS2 in the DSB repair that is independent of MSH2 and MLH1. These results highlight an additional layer of the multifunctional role played by the MMR proteins (66). Studies on the molecular mechanisms of the process identified here will allow determination of whether this process is mediated by the homodimeric form of PMS2 and MLH3, a complex with not yet characterized partners, or a new pathway of DSB repair.

Cell clones
All of the clones used in this study are summarized in Table 1.

Generation of human PMS2 2/2 TK6 B cells
To generate a pair of TALEN expression plasmids against the PMS2 gene, we used a Golden Gate TALEN kit and a TAL effector kit (Addgene) (84,85). The TALEN target sites are shown in Fig. S1A. The gene-targeting constructs were generated from the genomic DNA of TK6 cells by amplifying with primers HindIII-flanked F1 and HindIII-flanked R1 for the 59arm and XbaI-flanked F2 and XbaI-flanked R2 for the 39-arm. The 59-arm and 39-arm PCR products were cloned into the cor-responding sites of the DT-ApA/puro or DT-ApA/hygro vectors. 10 mg of TALEN expression plasmids and 10 mg of linearized gene-targeting vectors were transfected into 10 3 10 6 TK6 cells using the Bio-Rad Gene Pulser II Transfection System at 250 V and 950 microfarads. After electroporation, cells were released into 20 ml of drug-free medium containing 10% horse serum. 48 h later, cells were seeded into 96-well plates with both hygromycin and puromycin antibiotics for 2 weeks. The gene disruption was confirmed by genomic PCR using primers P1, P2, P3, of P4 (Fig. S1B) and RT-quantitative PCR using primers P5 and P6 (Fig. S1C). All primers used in this study are shown in Table S1.

Generation of nuclease-dead human PMS2 E705K/E705K TK6 B cells
To generate nuclease-dead human PMS2 E705K/E705K TK6 B cells, we designed a guide RNA targeting intron sequence upstream of the 12th exon using the Zhang CRISPR tool (86) and gene-targeting constructs. The CRISPR target site is depicted in Fig. S1C. The gene-targeting constructs were generated using SLiCE (seamless ligation cloning extract). The genomic DNA was amplified with primers F3 and R3 from the PMS2 gene locus, and the PCR product was used as template DNA for amplifying the 59-arm. The 5'-arm was amplified using primers F4 and R4, where each primer shared 20-bp end homology with the insertion site of the vector. The sequence intended as the 39-arm of the PMS2-targeting construct was amplified by PCR as two fragments using overlapping primers (F5 and R5) and included a point mutation to change codon 705 from glutamic acid to lysine. The two fragments were then combined by chimeric PCR to yield the 39 targeting arm including the codon 705 mutation. The 39-arm was amplified using primers F6 and R6, where each primer shared 20-bp end homology with the insertion site of the vector. Both vectors, DT-ApA/neo and DT-ApA/hygro, were linearized with NotI and XbaI. All of the fragments of the vectors and inserts were purified using a QIAquick gel extraction kit (Qiagen, Venlo, Netherlands). The gene-targeting constructs were generated in a single reaction mixture containing DT-ApA/neo or DT-ApA/ hygro vectors, 59-and 39-arms, and 23 SLiCE buffer (Invitrogen) and incubated for 30 min at room temperature. 6 mg of CRISPR and 2 mg of each gene-targeting vector were transfected into 4 3 10 6 TK6 cells using the Neon Transfection System (Life Technologies, Inc.). After electroporation, cells were released into 20 ml of drug-free medium containing 10% horse serum. 48 h later, cells were seeded into 96-well plates for selection with both neomycin and hygromycin antibiotics for 2 weeks. The gene disruption was confirmed by RT-PCR using primers F7 and R7 followed by direct sequencing (Fig. S1, D  and E). The drug resistance markers are flanked by loxP sites and were thus excised from PMS2 E705K/E705K cells by transient expression of Cre recombinase, leading to the generation of PMS2 E705K/E705K cells.

Generation of human MLH3 2/2 TK6 B cells
To disrupt the MLH3 gene, we designed a guide RNA targeting the sixth exon using the Zhang CRISPR tool (86) and gene- targeting constructs. The CRISPR target site is depicted in Fig.  S2A. The gene-targeting constructs were generated using SLiCE. The genomic DNA was amplified with primers F8 and R8 from the MLH3 gene locus, and the PCR product was used as template DNA for amplifying the 59-and 39-arms. The 59arm was amplified using primers F9 and R9, and the 39-arm was amplified using primers F10 and R10, where each primer shared 20-bp end homology with the insertion site of the vector. Both vectors, DT-ApA/neo and DT-ApA/hygro, were linearized with AflII and ApaI. All of the fragments of the vectors and inserts were purified using a QIAquick gel extraction kit (Qiagen). The gene-targeting constructs were generated in a single reaction mixture containing DT-ApA/neo or DT-ApA/ hygro vectors, 59-and 39-arms, and 23 SLiCE buffer (Invitrogen) and incubated for 30 min at room temperature. 6 mg of CRISPR and 2 mg of each gene-targeting vector were transfected into 4 3 10 6 TK6 cells using the Neon Transfection System (Life Technologies). After electroporation, cells were released into 20 ml of drug-free medium containing 10% horse serum. 48 h later, cells were seeded into 96-well plates for selection with both neomycin and hygromycin antibiotics for 2 weeks. The gene disruption was confirmed by RT-PCR using primers F11 and R11 (Fig. S2B) and by Southern blotting analysis with a 0.6-kb probe amplified by PCR from genomic DNA using F12 and R12 (Fig. S2C). The genomic DNA of the candidate clones was digested with EcoRI for Southern blotting analysis.

Generation of human MSH2 2/2 TK6 B cells
To disrupt the MSH2 gene, we designed a guide RNA targeting the fourth exon using the Zhang CRISPR tool (86) and gene-targeting constructs. The CRISPR target site is depicted in Fig. S3A. The gene-targeting constructs were generated using SLiCE. The genomic DNA was amplified with primers F22 and R22 from the MSH2 gene locus, and the PCR product was used as template DNA for amplifying the 59-and 39arms. The 59-arm was amplified using primers F23 and R23, and the 39-arm was amplified using primers F24 and R24, where each primer shared 20-bp end homology with the insertion site of the vector. Both vectors, DT-ApA/neo and DT-ApA/puro, were linearized with AflII and ApaI. All of the fragments of the vectors and inserts were purified using a QIAquick gel extraction kit (Qiagen). The gene-targeting constructs were generated in a single reaction mixture containing DT-ApA/neo or DT-ApA/puro vectors, 59-and 39arms, and 23 SLiCE buffer (Invitrogen) and incubated for 30 min at room temperature. 6 mg of CRISPR and 2 mg of each gene-targeting vector were transfected into 4 3 10 6 TK6 cells using the Neon Transfection System (Life Technologies). After electroporation, cells were released into 20 ml of drug-free medium containing 10% horse serum. 48 h later, cells were seeded into 96-well plates for selection with both neomycin and puromycin antibiotics for 2 weeks. The gene disruption was confirmed by genomic PCR using primers F25, F26, and R25 (Fig. S3).

Generation of human MLH1 2/2 TK6 B cells
To disrupt the MLH1 gene, we designed a guide RNA targeting the 8th exon using the Zhang CRISPR tool (86) and genetargeting constructs. The CRISPR target site is depicted in Fig. S4A. The gene-targeting constructs were generated using SLiCE. The genomic DNA was amplified with primers F27 and R27 from the MLH1 gene locus, and the PCR product was used as template DNA for amplifying the 59-and 39arms. The 59-arm was amplified using primers F28 and R28, and the 39-arm was amplified using primers F29 and R29, where each primer shared 20-bp end homology with the insertion site of the vector. Both vectors, DT-ApA/neo and DT-ApA/puro, were linearized with AflII and ApaI. All of the fragments of the vectors and inserts were purified using a QIAquick gel extraction kit (Qiagen). The gene-targeting constructs were generated in a single reaction mixture containing DT-ApA/neo or DT-ApA/puroro vectors, 59-and 39-arms, and 23 SLiCE buffer (Invitrogen) and incubated for 30 min at room temperature. 6 mg of CRISPR and 2 mg of each gene-targeting vector were transfected into 4 3 10 6 TK6 cells using the Neon Transfection System (Life Technologies). After electroporation, cells were released into 20 ml of drug-free medium containing 10% horse serum. 48 h later, cells were seeded into 96-well plates for selection with both neomycin and puromycin antibiotics for 2 weeks. The gene disruption was confirmed by Southern blotting analysis (genomic DNA was digested with SphI) with a 0.52-kb probe amplified by PCR from genomic DNA using F30 and R30 (Fig. S4, B and C). The candidate clones were further confirmed by RT-PCR using primers F31 and R31 (Fig. S4D) and Western blotting analysis (Fig. S4E).

Generation of human MUS81 2/2 TK6 B cells
To generate a pair of TALEN expression plasmids against the MUS81 gene, we used a Golden Gate TALEN kit and a TAL effector kit (Addgene) (84,85). The TALEN target sites are shown in Fig. S5A. The gene-targeting constructs were generated from the genomic DNA of TK6 cells by amplifying with primers SacI-flanked F19 and BamHI-flanked R19 for the 59-arm and BamHI-flanked F20 and R20 for the 39-arm. The 39-arm PCR products were cloned into pCR-Blunt II-TOPO vector. The 59-arm PCR products were cloned into the SacI site of the pCR-Blunt II-TOPO vector containing the 39arm. The BamHI fragment containing either the bsr R or puro R gene was cloned into the BamHI site between the 39arm and the 59-arm in the pCR-Blunt II-TOPO vector. 10 mg of TALEN expression plasmids and 10 mg of linearized genetargeting vectors were transfected into 10 3 10 6 TK6 cells using the Bio-Rad Gene Pulser II Transfection System at 250 V and 950 microfarads. After electroporation, cells were released into 20 ml of drug-free medium containing 10% horse serum. 48 h later, cells were seeded into 96-well plates with both blasticidin and puromycin antibiotics for 2 weeks. The genomic DNAs of the isolated clones resistant to both hygromycin and puromycin were digested with DraI for Southern blotting analysis. A 0.6-kb probe was generated by PCR of genomic DNA using primers F21 and R21 (Fig. S5B).
Generation of nuclease-dead human MLH3 D1223N/D1223N and MLH3 E1229K/E1229K TK6 B cells To generate nuclease-dead human MLH3 EK/EK and MLH3 DN/DN TK6 B cells, we designed a guide RNA targeting intron sequence upstream of seventh exon using the Zhang CRISPR tool (86) and gene-targeting constructs. The CRISPR target site is depicted in Fig. S6. The gene-targeting constructs were generated using SLiCE. The genomic DNA was amplified with primers F13 and R13 from the MLH3 gene locus, and the PCR product was used as template DNA for amplifying the 59-arm. The 59-arm was amplified using primers F14 and R14, where each primer shared 20-bp end homology with the insertion site of the vector. The sequence intended as the 39-arm of the MLH3-targeting construct was amplified by PCR as two fragments using overlapping primers (F15 and R15 for MLH3 DN/DN and F16 and R16 for MLH3 EK/EK cells) that included a point mutation to change codon from aspartic acid to asparagine (MLH3 DN/DN ) and glutamic acid to lysine (MLH3 EK/EK ) subsequently. The two fragments were then combined by chimeric PCR to yield the 39 targeting arm including the mutation. The 39-arm was amplified using primers F17 and R17, where each primer shared 20-bp end homology with the insertion site of the vector. Both vectors, DT-ApA/neo and DT-ApA/hygro, were linearized with NotI and XbaI. All of the fragments of the vectors and inserts were purified using a QIAquick gel extraction kit (Qiagen). The gene-targeting constructs were generated in a single reaction mixture containing DT-ApA/ neo or DT-ApA/hygro vectors, 59-and 39-arms, and 23 SLiCE buffer (Invitrogen) and incubated for 30 min at room temperature. 6 mg of CRISPR and 2 mg of each gene-targeting vector were transfected into 4 3 10 6 TK6 cells using the Neon Transfection System (Life Technologies). After electroporation, cells were released into 20 ml of drug-free medium containing 10% horse serum. 48 h later, cells were seeded into 96-well plates for selection with both neomycin and hygromycin antibiotics for 2 weeks. The site-directed mutagenesis was confirmed by genomic PCR using primers F18 and R18 followed by direct sequencing (Figs. S6, C and D). The drug resistance markers are flanked by loxP sites and were thus excised from MLH3 DN/DN and MLH3 EK/EK cells by transient expression of Cre recombinase, leading to the generation of MLH3 DN/DN and MLH3 EK/EK cells.

Colony survival assay
To measure sensitivity, cells were treated with camptothecin (Topogen, Inc.) and olaparib (Funakoshi, Tokyo, Japan) and irradiated with ionizing radiation ( 137 Cs). Cell sensitivity to these DNA-damaging agents was evaluated by counting colony formation in methylcellulose plates as described previously (87).

Heteroallelic crossover analysis
The human lymphoblastoid cell line TSCER2 is a TK6 derivative with an I-SceI site inserted into the TK locus (50,51). TSCER2 cells are compound heterozygous (TK 2/2 ) for a point mutation in exons 4 and 5. A DSB occurring at the I-SceI site results in homologous recombination between the alleles and produces TK-proficient revertants (TK 1/2 ). 4 3 10 6 TK6 cells were transfected with 6 mg of I-SceI expression vector using the Neon Transfection System (Life Technologies) with 33 pulse at 1350 V and with 10-ms pulse width and released into 20 ml of drug-free medium containing 10% horse serum. After 48 h, cells were seeded as 1 3 10 6 cells/96-well plate, with 29-deoxycytidine (Sigma, D0776), hypoxanthine (Sigma, H9377), aminopterin (Sigma, A3411), and thymidine (Sigma, T9250) (CHAT for TK-revertants) medium. Drug-resistant colonies were counted 2 weeks later.

Chromosomal aberration analysis
TK6 cells were irradiated with 1-Gy IR. The cells were then treated with 0.1 mg/ml colcemid (GIBCO-BRL) and incubated at 37°C for 3 h. Experimental conditions for chromosomal aberration analysis were as described previously (72). Briefly, harvested cells were treated with 1 ml of 75 mM KCl for 15 min at room temperature and fixed in 5 ml of a freshly prepared 3:1 mixture of methanol/acetic acid. The cell suspension was dropped onto a glass slide and air-dried. The slides were stained with 5% Giemsa solution (Nacalai Tesque) for 10 min and air-dried after being rinsed carefully with water. All chromosomes in each mitotic cell were scored at 31000 magnification. A total of 50 mitotic cells were scored for each group using a microscope.

SCE analysis
TK6 cells were incubated with or without cisplatin (2 mM). After 1 h, cells were washed and released into bromodeoxyuridine (100 mM)-containing media. Cells were incubated for two more cell cycles and treated with colcemid (0.1 mg/ ml) for 3 h before being harvested. Metaphase chromosomes were prepared and assayed for SCEs as described previously (64).

Immunostaining and microscopic analysis
Cells were fixed with 4% paraformaldehyde (Nacalai Tesque) for 10 min at room temperature and permeabilized with 0.5% Triton X-100 (Sigma) for 30 min. Images were taken with a confocal microscope (TCS SP8, Leica Microsystems, Germany).
Construction of FLAG-tagged hGen1 with nuclear localization signal expressing TK6 cell lines FLAG-tagged hGen1-NES (4A)-NLS 1 -expressing TK6 cells were generated using a genetically modified retroviral vector as Role of PMS2 and MLH3 in homologous recombination described (Fig. S7) (72). Briefly, the coding sequence for hGen1-NES (4A)-3xNLS 1 -3xFLAG was cloned into the pMSCV retroviral expression vector (Clontech) (Fig. S7A). The newly engineered retroviral expression vector was co-transfected into human 293T cells with a helper plasmid (pClampho) expressing the viral Gag, Pol, and Env proteins to produce viral supernatant. The viral supernatant was collected after 48 h and used to transduce into WT, PMS2 EK/EK , MLH3 DN/DN , PMS2 EK/EK /MLH3 DN/DN , MUS81 2/2 , and RAD54 2/2 TK6 mutant strains (Fig. S7B). The efficiency of each step was assessed by quantifying the number of cells expressing GFP (Fig. S7C). The expression of hGen1-NES (4A)-3xNLS 1 -3xFLAG was further confirmed by Western blotting. Experimental conditions for Western blotting analysis were as described previously (88). Anti-FLAG antibody overnight at 4°C and anti-mouse IgG horseradish peroxidase-linked antibody for 1 h at room temperature were used as the primary and secondary antibodies, respectively (Fig. S7D).

Quantification and statistical analysis
For all statistical analyses with a p value, unpaired Student's t test was used. Error bars represent S.D., as indicated in the figure legends. We calculated the propagation of errors using the following formula: H((S.D. with IR treatment) 2 1 (S.D. without IR treatment)) 2

Data availability
All of the data described are contained within the article.