MSH2-MSH3 promotes DNA end resection during homologous recombination and blocks polymerase theta-mediated end-joining through interaction with SMARCAD1 and EXO1

Abstract DNA double-strand break (DSB) repair via homologous recombination is initiated by end resection. The extent of DNA end resection determines the choice of the DSB repair pathway. Nucleases for end resection have been extensively studied. However, it is still unclear how the potential DNA structures generated by the initial short resection by MRE11-RAD50-NBS1 are recognized and recruit proteins, such as EXO1, to DSB sites to facilitate long-range resection. We found that the MSH2-MSH3 mismatch repair complex is recruited to DSB sites through interaction with the chromatin remodeling protein SMARCAD1. MSH2-MSH3 facilitates the recruitment of EXO1 for long-range resection and enhances its enzymatic activity. MSH2-MSH3 also inhibits access of POLθ, which promotes polymerase theta-mediated end-joining (TMEJ). Collectively, we present a direct role of MSH2-MSH3 in the initial stages of DSB repair by promoting end resection and influencing the DSB repair pathway by favoring homologous recombination over TMEJ.


INTRODUCTION
Genome integrity is constantly challenged by DNA replication errors and di v erse dama ging a gents, such as oxidati v e stress and environmental radiation ( 1 , 2 ). To maintain genomic stability, cells possess DNA repair anism by which the MSH2-MSH3 heterodimer is recruited to DSBs and how it contributes to the downstream steps of HR remain unclear. A possible connection may be the ' S WI / SNF-related M atrix-Associated A ctin-Dependent R egulator of C hromatin Subfamily A containing D EAD / H Box1 (SMARCAD1) protein. Resection of DSB ends is promoted by SMARCAD1 (Fun30 in yeast) in human cells ( 30 , 31 ). SMARCAD1 interacts with MSH2-MSH6 for proper MMR ( 32 , 33 ). Howe v er, it is still unclear how the interaction between SMARCAD1 and MMR proteins modulates MMR and HR. Another protein involved in MMR and HR is EX O1. How EX O1 is activ ated b y interaction with MSH2 in MMR is kno wn. Ho we v er, it is unclear whether this interaction may play a role in HR (34)(35)(36)(37)(38).
Here, we initially aimed to uncover the role of MSH2-MSH3 in HR. Findings re v eal that MSH2-MSH3 contributes to HR through its interaction with SMARCAD1 and EXO1. We demonstrate that SMARCAD1, MSH2-MSH3, and EXO1 are sequentially recruited to the DSB to initiate DNA end r esection. MSH2-MSH3 pr e v ents POL recruitment to broken DNA and inhibits the sealing of broken DNA ends through TMEJ by blocking POL polymerase activity on annealed microhomology sequences carrying DNA mismatch. Blockage of TMEJ facilitates errorfree HR via e xtensi v e EXO1-dependent end resection.

Cell culture and treatment
U2OS, HEK293T and HeLa cells were purchased from American Type Culture Collection and maintained in high glucose Dulbecco's modified Eagle's medium (DMEM; Hyclone) with 10% fetal bovine serum (FBS; Millipore) and 1% penicillin and streptomycin (Invitrogen) at 37 • C and 5% CO 2 . For DNA repair assays, U2OS cells stably expressing DR-GFP (HR), SA-GFP (SSA), EJ2-GFP (TMEJ) and EJ5-GFP (NHEJ) (39)(40)(41)(42) were grown in DMEM (Gibco) containing 10% FBS (Merck) and 2 g / ml puromycin (Invitrogen). Human MSH2 cDNA was PCR-amplified from human cDNA isolated from HeLa cells using TRIzol (Invitrogen) and cloned into the EGFP-C2 vector using Sal I and BamH I restriction sites, and the pcDNA3.1 myc-His A vector using BamH I and Apa I sites. SMARCAD1 cDNA ( 43 ) that was a gift from Tej K. Pandita, Baylor College of Medicine, was cloned into the EGFP-C2 vector using Sal I and BamH I restriction sites and into the pcDNA3.1 myc-His A vector using BamH I and Xba I sites. EXO1 cDNA ( 44 ) that was a gift from Zhongsheng You, Washington Uni v ersity School of Medicine, was cloned into the EGFP-C2 vector using Sal I and BamH I restriction sites and the pcDNA3.1 myc-His A vector using BamH I and Apa I sites. All cDNAs were confirmed by sequencing. The plasmid for expressing the active DN A pol ymer ase fr agment of POL (Sumo3 POLQM1) was a gift from Sylvie Doublie and Susan Wallace, Uni v ersity of Vermont (Addgene plasmid # 78462) ( 45 ). Full-length POL without a stop codon was cloned into a pcDNA-DEST47 plasmid (Invitrogen), resulting in a GFP-tagged protein at the C-terminus.

Laser microirradiation
U2OS cells 3 × 10 5 were plated in confocal dishes (SPL) and incubated them for one day. Then, 2 g of each plasmid expressing GFP-MSH2, GFP-SMARCAD1, or GFP-EXO1 was transfected using Lipofectamine 3000 according to the manufacturer's instructions. Media containing plasmids with Lipofectamine were replaced with media containing 10 M 5-bromo-2'-deoxyuridine after 4 h and incubated for 24 h. A 355 nm ultraviolet A laser was used for laser microirradiation, followed by incubation of the cells in a 37 • C chamber in an atmosphere of 5% CO 2 . After each laser microirradiation, cell images were obtained e v ery 10 s for 5 min using an LSM880 confocal microscope (Carl Zeiss). The intensity of each laser stripe was determined using Zen Blue software (Carl Zeiss). The values were normalized to baseline values. At least 10 cells were used for quantification.

F okI assa y
FokI-U2OS cells, a stable cell line with a Fok I restriction enzyme site, were plated in a four-well plate and transfected with control, MSH2, or MSH3 siRNA. The next day, Lipofectamine 3000 was used to transfect the cells with LacI-mCherry-FokI expression plasmid and GFP-EXO1 or mNeon-MRE11 . After 48 h, the transfected cells were stained with Hoechst for 15 min to visualize the nuclei. Li v e cell images were obtained using a model LSM880 confocal microscope (Carl Zeiss).

Cell cycle analysis
U2OS cells were transfected in a 60-mm diameter plate with the indicated siRNAs. After 48 h, the cells wer e fix ed with 70% (v / v) ice-cold ethanol and incuba ted a t -20 • C for 1 h. Cells were washed once with ice-cold PBS and stained with propidium iodide in fluorescence-activated cell sorting (FACS) buffer (1 × PBS, 0.1% Triton X-100, 0.2 mg / ml RNase A) at 37 • C for 30 min. The stained cells were analyzed using a Becton Dickinson FACSVerse flow cytometer.

Immunofluorescence assay
Cell samples were prepared as previously described ( 46 ). Briefly, U2OS cells plated on LabTek chamber slides (Thermo Fisher Scientific) were incubated in CSK buffer for 10 min. The cells wer e fix ed with 4% paraformaldehyde for 20 min. Cells were incubated with the anti-Rad51 antibod y (ca t no. 8875; Cell Signaling Technology) or anti-RPA antibod y (ca t no. ab2175; Abcam) a t 4 • C overnight. After 30 min incubation with Alexa Fluor-conjugated secondary antibody, cells were mounted with ProLong Gold antifade reagent (Vector Laboratories). Confocal images were obtained using an LSM880 confocal microscope (Carl Zeiss). The images were analyzed using ZEN2.1 software.

HR, SSA, NHEJ, and TMEJ assays
SceI (pCAGGS-I-SceI, also denoted pCBASce), empty vector (pCAGGS-BSKX), and dsRed vector (a gift from Nucleic Acids Research, 2023, Vol. 51, No. 11 5587 Jeremy Stark) were prepared as previously described (39)(40)(41). U2OS cells stab ly e xpressing DR-GFP , SA-GFP , EJ2-GFP, or EJ5-GFP plasmids were plated on a 12-well plate (1 × 10 5 cells / well). The following day, the cells were transfected with 20 nM siRNA duplex mixed with RNAiMAX (Invitrogen) in Opti-MEM. After 24 h, a second round of transfection was performed. The following day, the cells were co-transfected with 0.5 g of either I-SceI expression vector or empty vector, and 0.1 g of dsRED vector (used as a transfection control) in 0.1 ml Opti-MEM containing 3 l of Lipofectamine 3000 (Invitrogen). After 6 h, the medium was removed and replaced with the growth medium. Two days after I-SceI transfection, the percentage of GFP-positi v e (GFP+) cells was analyzed using a Becton Dickinson FACSVerse flow cytometer. DNA repair efficiency was calculated as described previously ( 42 ). The experiments were repea ted a t least three times.

End resection assay
ER-Asi SI U2OS cells wer e pr epar ed as pr eviously described ( 47 ). Trypsinized cells wer e r esuspended with 0.6% lowmelting agarose (Bio-Rad) at a concentration of 1.2 × 10 7 cells / ml. Fifty microliters of cell suspension was used to make an agar ball, which was incubated with ESP buffer (0.5 M EDTA, 2% N-lauroylsarcosine, 1 mg / ml proteinase K, 1 mM CaCl 2 , pH 8.0) at 16 • C for 20 h. The agar ball was treated with HS buffer (1.85 M NaCl, 0.15 M KCl, 5 mM MgCl 2 , 2 mM EDTA, 4 mM T ris, 0.5% T riton X-100, pH 7.5) at 16 • C for 20 h. Melted agar balls were incubated overnight with restriction enzyme ( BsrG I or Hind III-HF; New England Biolabs). Real-time PCR was performed with restriction enzyme-treated or non-treated samples. The percentage of single-strand DN A (ssDN A) was calculated as described previousl y ( 47 ). Briefly, the Ct value was calculated by subtracting the Ct value of an untreated sample from that of a sample treated with the restriction enzyme. The ssDNA fraction was calculated ( 47 ) as s s DN A f racti on (%) = (1 / ( 2 ( Ct−1 ) + 0 . 5 )) × 100.

Replication protein A (RPA) retention assay
Cells were treated with 5 M camptothecin (CPT) for 1 h, or 62.5 M baicalein for 24 h. The trypsinized cells were transferred to a 1.5 ml tube, washed with PBS, and permeabilized with 100 l of 0.2% Triton X-100 in PBS for 10 min on ice. After washing with 1 × PBS containing 1 mg / ml bovine serum albumin (PBS-BSA), the cells were fixed and permeabilized with 100 l BD Cytofix / Cytoperm buffer (BD Biosciences) at room temperature for 15 min. The fixed cells were washed with 0.5 ml of 1 × BD Perm / Wash buffer (BD Biosciences) and suspended in 0.5 ml of 1 × BD Perm / Wash buffer and sequentially incubated with anti-RPA2 and Alexa Fluor 488-secondary antibodies. Nuclei in the cells were visualized by propidium iodide staining for 15 min and analyzed using a Becton Dickinson FACSVerse flow cytometer.

DNA pr epar ation f or in vitro experiments
All DNA oligomers were chemically synthesized (Bioneer) and are listed in Supplementary Table S1. Each set of oligomers was annealed by heating at 95 • C for 20 min followed by slow cooling to 23 • C.

Electrophoretic mobility shift assay (EMSA)
The EMSA for MSH2-MSH3 was performed as previously described ( 48 ) All EMSA with competitors were performed with 50 nM unlabeled 40 bp homoduplex.

Protein purification
Full-length human MSH2-MSH3, EXO1 WT, Mut EXO1-D173A, SMARCAD1, and MSH2-MSH6 were obtained by infecting Hi5 insect cells with amplified baculoviruses. To enhance the protein solubility of EXO1, Mut EXO1-D173A, and SMARCAD1, we added a maltose-binding protein tag to the N-terminus of the protein. After  For MSH2-MSH6 purification, cells were harvested and resuspended in buffer C containing 25 mM HEPES [pH 7.5], 150 mM KCl, 0.1 mM EDTA, 10% glycerol, and 1 mM DTT with 0.1% phen ylmethyl sulf on yl fluoride (PMSF) and protease inhibitor cocktail (cat. no. 11873580001; Roche). The supernatant was applied to a HiTrap Heparin HP column (cat. no. 17040601; Cytiva), and proteins were eluted with a linear salt gradient in buffer C up to 650 mM KCl. After adjusting the salt concentration of the collected sample to 150 mM KCl, we applied proteins to a HiTrap Q HP column (cat. no. 17115301; Cytiva) and eluted them with a linear salt gradient similar to that of the heparin column. Protein peaks were collected and concentrated using an Amicon ultra-15 50 K centrifugal filter. Concentrated proteins were then applied to a HiLoad 26 / 600 Super de x 200 pg column (cat. no. 28989336; Cytiva) equilibrated in buffer consisting of 25 mM HEPES [pH 7.5], 100 mM KCl, 0.1 mM EDTA, 10% glycerol, and 1 mM DTT. The fractionated protein peak from each step was confirmed by SDS-PAGE. Protein concentrations were measured using the Bradford assay.

Deter mination of ter mination probability and amount of fulllength extension
The termination probability at position N was defined as the band intensity at N divided by the total intensity of all bands ≥ N , as previously described ( 50 ). The quantification of full-length extension products was defined as the fully extended band intensity divided by the intensity of all bands ≥ N 0 (primer position).

Depletion of MSH2 and MSH3 decreases HR
We previously found that the natural compound baicalein inhibits MMR ( 52 ) and selecti v ely kills MMR-deficient cancer cells. Baicalein is deri v ed from Scutellaria baicalensis and is widely used in traditional Chinese medicine ( 53 ).
Gi v en the reported interconnection between the MMR and DSB repair pathways, we hypothesized that baicalein influences DSB r epair. To explor e this, U2OS cells wer e tr eated with increasing concentrations of baicalein, and the frequencies of HR, SSA, NHEJ, and TMEJ were measured. Using established reporter assays based on the restoration of GFP expression (39)(40)(41), we observed that the frequencies of HR and SSA wer e decr eased by baicalein treatment in a dose-dependent manner. In contrast, NHEJ and TMEJ were not significantly affected (  Figure S1D). These data suggest that MSH6 knockdown does not change the le v el of the MSH2-MSH3 complex, but does affect the MSH2-MSH6 complex. In summary, consistent with a pr evious r eport ( 25 )

Depletion of MSH2 and MSH3 suppresses DNA end resection
Gi v en that HR and SSA wer e decr eased upon MSH2 or MSH3 depletion, we determined which step in the HR pathway depended on MSH2 or MSH3. We measured the extent of DNA end resection by assessing RPA2 loading onto r esected ssDNA upon tr eatment with camptothecin using FACS analysis ( 55 ) Figure S1E). To directly measure the efficiency of DNA end resection, we used ER-Asi SI U2OS cells tha t genera te DSBs by the induction of the Asi SI restriction nuclease upon 4-OHT treatment ( 47 ). The extent of resection was measured by qPCR to assess the amplification from the r esected ssDNA compar ed to the corresponding dsDNA. The resection assay can measure resected ssDNA because un-resected double-stranded DNAs digested with restriction enzyme can no longer be used as a template for PCR amplification. Thus, only resected ss-DNA r esistant to r estriction enzymes can be detected by real-time PCR amplification. We obtained the cycle threshold (Ct) value for each sample using real-time PCR. The calculation, a Ct value was calculated by subtracting the Ct value of an untreated sample from the Ct value of a sample treated with the restriction enzyme. We then calculated the ssDNA fraction (%) using the following equation: Figure S1F) ( 47 ). Induction of Asi SI expr ession r esulted in e xtensi v e end resection, as measured by the amplification of 335 and 1618 bp fragments, which was blocked by baicalein tr eatment (Figur e 2 C). Consistent with the baicalein results, end resection was significantly reduced in MSH2 or MSH3 depleted cells. Consistent with our HR analysis, no effect was observed upon MSH6 (MutS ␣) depletion (Figure 2 D).

MSH2-MSH3 interacts with EXO1 to promote end resection activity
We next monitored the recruitment of MRE11 and EXO1 to DSB to determine the step of DNA end resection facilitated by MSH2-MSH3. DSBs are induced and visualized using a fusion protein comprising Fok I, the lac r epr essor, and mCherry, wher e br eaks can be targeted to lac opera tor repea ts in U2OS cells ( 56 ). MRE11 recruitment to DSBs was not affected in control, MSH2, or MSH3 siRNAtreated cells (Figure 2 E), suggesting that MSH2-MSH3 is not essential for MRE11 recruitment to DSBs. In contrast,  EXO1 recruitment to DSB sites was significantly decreased in MSH2 or MSH3 knockdown cells (Figure 2 F). MRE11 and EXO1 recruitment to DSB was not observed when a catal yticall y inacti v e Fok I D450A nuclease was used (Figures 2 E, F). Collecti v ely, the MSH2-MSH3 complex is required for recruiting EXO1, but not MRE11, to DSBs. EXO1 interacts with MSH2 through the C-terminal region of EXO1 ( 38 ). We first confirmed this interaction by immunoprecipitation of endogenous proteins. As previously observed, MSH2 and EXO1 interacted with each other (Figure 3 A). This interaction was not changed by ionizing radiation treatment (Figure 3 A). Using a series of GFP-tagged EXO1 deletion mutants (EXO1 D1 to D4) spanning the entire protein and myc-tagged MSH2, we confirmed that the C-terminal amino acid (aa) residues 600-846 of EXO1 are required for MSH2 binding (Supplementary Figure S2A) ( 38 ). To further investigate how MSH2 regulates EXO1 recruitment to DSB sites, we used a series of small C-terminal deletions and narrowed down the minimal MSH2 binding domain of EXO1 to aa residues from 801 to 807 (Figure 3 B). Conversely, the minimal MSH2 domain required for EXO1 binding was determined to be aa residues from 306 to 623 (Supplementary Figure S2B, Figure 3 C). This domain contains the MutS core domain and is a part of the previously annotated EXO1 binding domain ( 38 ). The r equir ement of the EXO1 C-terminal residues from 801 to 807 for MSH2 binding in vivo was further confirmed by CU-PID assays ( 57 ). For this assay, a PKC-␦ domain was fused to mRFP-MSH2 and employed to tether the fusion protein to the nuclear membrane upon phorbol 12-myristate 13-aceta te trea tment, which resulted in the localization of EX O1, but not EX O1 D16 ( 801-807) to the nuclear membrane (Supplementary Figure S2C).
To investigate the direct interactions between MSH2-MSH3 and EXO1 in vitro , all proteins were purified (Supplementary Figure S2D), and the activities of the purified proteins were tested (Supplementary Figures S2E, F). We then confirmed that purified MSH2-MSH3 and EXO1 directly bind each other in vitro (Figure 3 D). Consistent with a previous work, MSH2-MSH3 showed a higher binding affinity to an oligonucleotide substrate carrying an 8nt loop (+8-loop DNA) compared to a corresponding homoduplex oligonucleotide substrate ( Supplementary Figure S2E) ( 48 ). Wild-type EXO1 (WT EXO1) displayed the e xpected nuclease acti vity when serv ed with a 40 bp DNA double-str anded substr ate carrying a 4-nt single-stranded 3 overhang (Supplementary Figure S2F). However, WT EXO1 did not show e xonuclease acti vity for blunt end DNA compared to the 3 overhang DNA (Supplementary Figure  S2F). In addition, the EXO1 nuclease mutant (Mut EXO1-D173A) did not digest any type of DNA (Supplementary Figure S2F).
Having confirmed the functionality of in vitro purified MSH2-MSH3 and EXO1 proteins, we investigated whether MSH2-MSH3 can facilitate EXO1 recruitment to DNA substrates. Supershift assays were performed by adding EXO1 to a +8-loop-containing oligonucleotide substrate bound to MSH2-MSH3, and a dose-dependent supershift was observed (Figure 3 E). Supershift by WT EXO1 was observed at a lower concentration ( ∼15 nM) in the presence of MSH2-MSH3, while WT EXO1 bound to the same DN A substrate onl y a t a higher concentra tion ( ∼80 nM). For Mut EXO1-D173A, the supershift occurred at a higher concentration ( ∼40 nM) than that for WT EXO1, whereas Mut EXO1-D173A alone did not bind to the DNA. Taken together, MSH2-MSH3 promoted the association of WT EXO1 and Mut EXO1-D173A with DNA ( Figure 3 E and Supplementary Figure S2G).
We then investigated how the interaction between MSH2-MSH3 and EXO1 contributes to DNA end resection. It has been reported that MSH2-MSH6 enhances EXO1 activity in MMR ( 58 ). Thus, we tested whether MSH2-MSH3 also enhances EXO1 nuclease activity. In the presence of ATP, MSH2-MSH3 enhanced DNA degradation by WT EXO1 (Figure 3 F). In addition, as the MSH2-MSH3 concentration increased at a fixed WT EXO1 concentration, more DNA was digested by WT EXO1 (Figure 3 G). In contrast, in the absence of ATP, DNA degradation by WT EXO1 was slightly increased by MSH2-MSH3, indica ting tha t ATP is important for the enhancement of WT EXO1 nuclease activity (Supplementary Figure S2H). MSH2-MSH3 did not enhance DNA digestion of catalytically inacti v e Mut EXO1-D173A, regardless of ATP (Figures 3 F-G, and Supplementary Figure S2H). Collecti v ely, our data suggest that EXO1 recruitment by MSH2-MSH3 enhances end resection (Figure 3 F, G).

SMARCAD1 directly interacts with MSH2-MSH3
MSH2-MSH3 pr efer entially r ecognizes small loop structur es. Since r esected DSBs do not have small loops, we investigated how MSH2-MSH3 might be recruited to DSB sites to facilitate EXO1 recruitment for end resection. We hypothesized that the protein(s) interacting with MSH2-MSH3 would help recruit MSH2-MSH3 to DSBs. The chromatin remodeler SMARCAD1 interacts with MSH2-MSH6 ( 32 , 33 ) and is reported to be localized at DSBs and facilitate end resection in yeast and human cells ( 30 , 43 ). We first re-examined whether SMARCAD1 interacts with MSH2 in HEK293T cells. Reciprocal pull-down of endogenous MSH2 and SMARCAD1 was observed using immunopr ecipitation (Figur e 4 A), which was not changed by ionizing radia tion trea tment. To identify the MSH2 binding domain in SMARCAD1, we generated GFP-tagged SMARCAD1 WT and a series of deletion mutants (D1 to D4) spanning the entire protein and assessed its association with myc-MSH2 by immunoprecipita tion. Bioinforma tic analysis predicted that SMARCAD1 has a potential MSH2 binding domain, termed SHIP box, in its N-terminus (aa residues 5-11) ( 37 ). Our domain analysis experimentally confirmed that the N-terminus of SMARCAD1 is r equir ed for MSH2 binding, with the SMARCAD1 D1 ( 1-156) deletion being unable to bind MSH2 (Figure 4 B). Conversely, in co-transfection experiments with a series of MSH2 deletion mutations, we narrowed down the minimal SMARCAD1 binding domain in MSH2 to aa residues from 306 to 623 (Supplementary Figure S3A and Figure  4 C), which were the same region of MSH2 interacting with EXO1 (Figure 3 C), suggesting that SMARCAD1 and EXO1 bind to the same region of MSH2. MSH2 D6 to D9 deletion mutants spanning aa residues from 306 to 623 lost their ability to interact with EXO1 and SMARCAD1.   These MSH2 deletion mutants were still able to assemble an MSH2-MSH3 heterodimer complex with MSH3 (Supplementary Figure S3B). No interaction between SMAR-CAD1 and EXO1 was evident by endogenous immunoprecipitation (Supplementary Figure S3C). The interaction between MSH2 and SMARCAD1 in vivo was confirmed by CUPID analysis (Supplementary Figure S3D).
To determine whether SMARCAD1 directly interacts with MSH2-MSH3, in vitro IP assays using purified proteins wer e performed, wher e purified SMARCAD1 and  Figure S3E). Similarly, SMARCAD1 enhanced the recruitment of MSH2-MSH3 to homoduplex DNA (Supplementary Figure S3E). Thus, the binding affinity of MSH2-MSH3 to DNA substrates was enhanced by a pproximatel y eight times in the presence of SMARCAD1. The finding supports the view that SMARCAD1 recruits MSH2-MSH3 and forms a complex on DNA regardless of the type of DNA substrate.

Interdependence of SMARCAD1, MSH2, and EXO1 localization at DNA damage sites
To determine the interdependency of SMARCAD1, MSH2, and EXO1 recruitment to DNA damage sites, GFP-tagged versions of these genes were transfected into U2OS cells and their recruitment to stripes irradiated with a 355 nm laser was determined. GFP-MSH2 accumula ted a t microirradiation sites within 1 min of irradiation. The accumulation was compromised by deletion of SMARCAD1, but not EXO1 (Figure 5 A). Depletion of MSH2 or EXO1 did not alter the recruitment of SMARCAD1 to sites of damage (Figure 5 B). Finally, EXO1 recruitment was dramatically reduced in the absence of MSH2, MSH3, or SMAR-CAD1, but not in MSH6 knockdown cells ( Figure 5 C and Supplementary Figure S4A) Figure S4B). Taken together, these da ta indica te tha t SMARCAD1 is r equir ed for MSH2 r ecruitment, which in turn is needed for EXO1recruitment to DSB sites.
Since MRE11 recruitment to DSB was not dependent on MSH2 (Figure 2 E), we examined the interaction between MRE11 and SMARCAD1. MRE11 recruitment to the microirradiation-induced DSB sites occurred normally in control, MSH2, or SMARCAD1 knockdown cells (Supplementary Figure S4C), indica ting tha t MRE11 recruitment to DSB occurs independently of SMARCAD1 and MSH2. MMR proteins have been suggested to function in rejecting heteroduplex DNA with imperfect matches during later stages of HR (26)(27)(28). Formation of heteroduplex DNA r equir es RAD51-dependent strand invasion ( 59 , 60 ). Thus, we investigated whether MSH2-MSH3 recruitment to DSB depends on RAD51. Cells were treated with the RAD51 inhibitor B02 for 4 h and MSH2 recruitment to microirradiation-induced DSB was monitored. MSH2 accumulation to DSBs was not influenced by the RAD51 inhibitor and RAD51 depletion (Supplementary Figure  S4D), indicating that MSH2 acts upstream of RAD51 and locates to DSBs before the strand invasion stage of the HR. Knockdown of MLH1 did not affect HR, end resection, and EXO1 recruitment to DSB, suggesting that MSH2-MSH3 in MMR is a major protein involved in the recruitment of EXO1 for the end resection of HR repair (Supplementary Figure S4E). Recently, MLH1 deficiency leads to the hyperactivation of EXO1 resulting in e xcessi v e longrange resection ( 61 ). In the previous study, ER-Asi SI and RPA f oci assa ys were used for end resection with IR, while we used RPA f oci assa y with IR or CPT ( Supplementary  FigureS4E). For RPA foci assay, we measured RPA foci that were generated in 2 h after IR or 1 h after CPT treatment, whereas the previous study measured RPA foci that in 24 h after IR treatment. Our result provides the initial step of end resection for recruitment of EXO1 by MSH2-MSH3, which is different from the role of MLH1 in the termination of end resection.

SMARCAD1-MSH2-EXO1 recruitment is important for HR
Next, we assessed the r equir ement of various SMARCAD1 domains for recruitment to the microirradiated sites. The SMARCAD1 D1 deletion m utant, w hich does not interact with MSH2, was localized to sites of DNA damage, similar to WT (Figure 6 A). Interestingly, recruitment of the SMARCAD1 D3 deletion m utant, w hich lacks the DNA helicase and ATP binding domain, was decreased. The finding suggests that DNA helicase activity is important for DNA binding (Figure 6 A). To directly test whether SMARCAD1 is r equir ed for HR, reporter-based HR assa ys were perf ormed (see Figure 1 ). SMARCAD1 depletion decr eased HR fr equency (Figur e 6 B), which was complemented by transfection with an RNA interference (RNAi)resistant full-length SMARCAD1, but not with the siRNAresistant SMARCAD1 D1 mutant (Figure 6 B). Consistent with the results of the HR assay, RAD51 foci were formed after irradiation with 10 Gy in the presence of WT SMAR-CAD1. Howe v er, f oci f orma tion was a ttenua ted in the absence of SMARCAD1 or in the presence of the MSH2-interaction mutant D1 of SMARCAD1 (Figure 6 C). These data show that SMARCAD1-dependent MSH2 recruitment is r equir ed for a proficient HR.
We next examined the recruitment of WT and a series of deletion mutants of GFP-MSH2, including mutants D6 to D9, spanning aa residues from 306 to 623 of MSH2 required for SMARCAD1 and EXO1 binding. WT MSH2 was recruited to microirradiated stripes, but EXO1-and SMARCAD1-binding defecti v e MSH2 D6 to D9 mutants were not recruited (Figure 6 D). MSH2 D6 to D9 deletion proteins could move into the nucleus with a nuclear localization signal, excluding the possibility that D6 to D9 MSH2 failed to move to the DSB due to its incapability to enter the nucleus (Supplementary Figure S5A). Thus, MSH2 recruitment to DSBs depends on SMAR-CAD1. By measuring the effect of MSH2 depletion on HR activity and RAD51 foci formation upon irradiation, it was confirmed that HR was reduced by MSH2 depletion (Figure 6 E). MSH2 depletion could be rescued by expr essing siRNA-r esistant full-length MSH2, but not by the SMARCAD1 binding defecti v e MSH2 D9 mutant (Figures 6 E, F). Collecti v ely, MSH2 recruitment to DNA damage depends on SMARCAD1 and is important for a proficient HR.
Lastl y, to determine w hether EXO1 recruitment to DSBs r equir es MSH2, we examined GFP-EXO1 recruitment to microirradiated stripes, HR activity, and RAD51 f oci f ormation with GFP-tagged WT EXO1, MSH2-binding proficient EXO1 mutants (D11, D14, and D15), and MSH2binding deficient EXO1 mutants (D12, D13, and D16) (Figure 3 B and Figures 6 G-I). WT, D11, D14, and D15 GFP-EXO1 wer e r ecruited to microirradiated sites, wher eas D12, D13 and D16 were not (Figure 6 G). Consistently, EXO1-D16 expressing cells displayed reduced HR (Figure 6 H) and RAD51 foci formation (Figure 6 I), compared to WT EX O1. Conversely, the decreased EX O1 recruitment to microirradiated sites in MSH2 depleted cells was rescued by transfection with siRNA-resistant WT MSH2, but not by the EXO1-interaction defecti v e D9 mutant (Supplementary Figure S5B), suggesting that EXO1 recruitment to DSB requires MSH2. Since the MSH3 binding domain of EXO1 is w ell-characterized ( 38 ), w e tested the recruitment of EXO1 to microirradiated sites using the MSH3 binding defecti v e mutant, GFP-EXO1 D2, and the EXO1 D2 was not recruited to the DSB (Supplementary Figure S5C-D), suggesting that MSH3 is involved in EXO1 recruitment to the DSB. Taken together, EXO1 recruitment to DSB requires the MSH2-MSH3 complex, which in turn requires SMARCAD1.
To exclude the possibility that RNAi depletion alters the cell cycle profile to affect HR activity, we monitored the cell cycle profiles by FACS analysis. Depletion of MSH2, MSH3, SMARCAD1, or EXO1 did not significantly change the cell cycle profile (Supplementary Figure S5E).

MSH2-MSH6 does not affect end resection
To investigate the effect of MSH2-MSH6, we purified human MSH2-MSH6 and performed biochemical assays in the same manner as for MSH2-MSH3. We confirmed that    Figure S6F, G). When the MSH2-MSH6 concentration was fixed and EXO1 was titrated, the EXO1 nuclease activity was slightly increased. When MSH2-MSH6 was titrated at a fixed EXO1 concentration, no dramatic change in EXO1 nuclease activity was observed, indica ting tha t MSH2-MSH6 slightly enhanced EXO1 nuclease activity. We suspect that the slight effect of MSH2-MSH6 on EXO1 nuclease activity is due to the structure of the flap DNA. A prior study reported that MSH2-MSH6 pr efer entially binds to +12 or +14 bp palindromic insertions in vitro , but does not repair them in vivo ( 62 ).

MSH2 inhibits POL -mediated end-joining
Heteroduplex DNAs are rejected by mismatch repair proteins during HR and SSA ( 63 ). DN A pol ymerase (POL )mediated end-joining (TMEJ) uses the pairing of short homologous sequences (2-6 bp microhomology) of resected ssDNA to repair DSBs at the cost of generating small deletions. We hypothesized that MSH2-MSH3 might act on resected DNA to pre v ent POL recruitment, allowing for further DNA end resection and facilitating error-free HR. Such a mechanism would r equir e POL to prime mismatched heteroduplex DNA. We did not observe any drastic effect on TMEJ after MSH2 knockdown (Figure 1 B).
Howe v er, this could be due to the sensitivity of the assay. We decided to study the relationship between POL and MSH2-MSH3 more directly. We tested whether POL could extend primers carrying a 2 bp mismatch at an internal or terminal position (Supplementary Figure S7). In a control experiment, the polymerase fragment of POLca talyzed templa te-dependent DNA synthesis from a perfectly annealed primer pair was similar to the exonucleasedeficient Esc heric hia coli pol I Klenow Fragment (Kf exo-), another A-famil y DN A pol ymerase that served as a control (Supplementary Figure S7A) ( 45 , 64 ). Importantly, annealed primer pairs, which was carrying 2 bp mismatches tha t loca ted 1-2 bp or 3-4 bp upstream from the 3 primer end (MM-1, 2 and MM-3, 4), could be much more efficiently extended by POL than by Kf exo-(Supplementary Figures S7B, C). In other words, such extension by POL was particularly strong when the mismatch occurs at the primer junction (Supplementary Figure S7B) Figure S2E)

DISCUSSION
In the present study, we show that the MSH2-MSH3 heter odimer pr omotes err or-free HR for DSB repair via two complementary mechanisms (Figure 7 I). MSH2-MSH3 is recruited to the DSB after the initial stages of DNA end resection by SMAR CAD1. SMAR CAD1 and MSH2-MSH3 dependent EXO1 recruitment promotes further resection of HR. Sim ultaneousl y, the MSH2-MSH3 complex also inhibits POL priming and extension from mismatched DNA to pre v ent mutagenic TMEJ.
Here, we show that MSH2-MSH3 has a more direct role in HR by facilitating DNA end resection. The sequential recruitment of SMARCAD1, MSH2-MSH3, and EXO1 observed in the present study, together with the r equir ement of these proteins for RAD51 loading and proper HR, clearly demonstra ted tha t MSH2-MSH3 plays an acti v e role in the early stages of HR. In addition to inhibiting TMEJ by rejecting POL , MSH2-MSH3 facilitates EXO1 recruitment and long-range DNA end resection, thus funneling pathway choice towards error-free HR. HR primarily occurs in the S and G2 phases of the cell cycle. These processes are available for actively proliferating cells.
In addition to facilitating EXO1 recruitment, how could MSH2-MSH3 further aid EXO1 in DNA end resection? When MSH2-MSH3 binds to loop structures, the DNA  3' The termination probability at position N3 is defined as the band density at N3 divided by the intensity of ≥N3. ( C ) The quantity of full-length extension products was calculated as the fully extended band density divided by the intensity ≥ N0 (primer position). MSH2-MSH3 on non-mismatched substrates did not disturb POL extension ( D-F ). Data are presented as mean ± standard deviation ( n = 3). P -values were calculated using two-tailed Student's t -test. ( G ) U2OS cells were transfected with control, MSH2, MSH3, or MSH6 siRNAs. After 24 h, cells were transfected with GFP-POL . Data are presented as mean + standard deviation ( n = 10). ( H ) Mutation signatures at the CEL locus upon CRISPR-Cas9-induced DSB were compared those in control and MSH2 knockdo wn HEK293T cells. Bo xplot sho wing the frequency of deletion mutations harboring microhomology longer than two nucleotides at the DNA junction out of the total deletion mutations induced by CRISPR-Cas9 targeting the CEL locus in control and MSH2 knockdown HEK293T cells. P -values were calculated using an unpaired two-tailed t -test ( n = 7) (left). DNA deletion spectrum associated with microhomology longer than four nucleotides induced by CRISPR-Cas9 targeting the CEL locus in control and MSH2 knockdown HEK293T cells. P -values were calculated using a paired two-tailed t -test ( n = 7) (right panel). ( I ) A model of how MSH2-MSH3 acts in the early stages of HR. Processing of only one end of the DSB is shown in the figure. bound by MSH2-MSH3 is bent for proper recognition by downstream proteins ( 67 ). Thus, it is possible that MSH2-MSH3 recruited to DSB sites could bend DNA to provide better access of EXO1 to DNA. MRE11-RAD50-NBS1 generates a nick and degrades ssDNA with 3 to 5 polarity ( 68 ). Small single-stranded gaps structurally resemble small loop structures, with the ssDNA stretch being extruded. MSH2-MSH3 recognizes such a structure ( 48 ) and bends this small-ga pped DN A to provide an entry platf orm f or EXO1 to generate long ssDNA. The recent report showing no effect of MSH3 on the formation of RPA foci differs from our results (Figure 2 B and Supplementary Figure S1E). We measured RPA accumulation by two different methods, confocal microscopy and FACS. Both approaches yielded similar results. MSH2-MSH3 and XPF / ERCC1 are known to have a critical function in 3 nonhomologous tail removal (3 NHTR) during HR ( 69 ). Knockdown of ERCC1 reduced the frequencies of HR and SSA, but did not affect end resection (Supplementary Figure S7P). It suggests that in addition to the initial end resection by MSH2-MSH3 observed in this study, 3 NHTR function of MSH2-MSH3 could also affect HR.
Chromatin r emodeling complex es play important roles in DSB repair ( 43 ). The budding yeast Fun30 protein, an ortholog of human SMARCAD1, is a major nucleosome remodeler that enhances Exo1 and Sgs1 dependent end resection during HR repair ( 30 ). Mammalian SMARCAD1 has been suggested to play a role in HR ( 43 ). Fun30 also functions in MMR through its interaction with MSH2 ( 33 , 37 ). SMARCAD1, as a chromatin remodeler, may unwind chromatin structures near DSBs to help recruit MSH2-MSH3 at the early stages of DSB processing. Gi v en the conserved interactions between SMARCAD1 and MSH2 and between MSH2 and EXO1, these interactions are conserved throughout evolution to facilitate HR. Our results clearly support a conserved mechanism by which SMARCAD1 interacts with MSH2-MSH3 and enhances the DNA binding affinity of MSH2-MSH3 (Figure 4 D, E). Additionally, MSH2-MSH3 pre v ents the access of POL , the key enzyme facilitating err or-pr one TMEJ, to DNA damage sites and the subsequent function of POL . Ther efor e, when MSH2 or MSH3 was depleted, POL recruitment to DNA damage sites was enhanced.
MSH2-MSH3 (Muts ␤) and POL promote CAG repeat expansion during DNA replication (70)(71)(72)(73), which may be different from DN A DSB repair. DN A replication slippage is the major mechanism for CAG repeat expansion and can be promoted by MMR and POL . Our observations of more POL recruitment to laser stripes under MSH2-MSH3 deficient conditions suggest that MSH2-MSH3 competes with POL at DNA DSB sites, which could be different from CAG expansion.
We expected that Lynch syndrome patients with mutations in mismatch repair proteins would have more POLmedia ted muta tion signa tures in the genome. Howe v er, we did not find significant enrichment of POL -mediated muta tion signa tures in the genome of Lynch syndrome patients. It is possible that strong MMR defect signatures could be dominant in the genome of Lynch syndrome patients compared with the POL -mediated mutation signatures.
Analogous to its role in MMR repair ( 37 ), MSH2-MSH3 recruits EXO1 to promote DNA end resection. Interestingly, SMARCAD1 binding domains in MSH2 are shared with EXO1 binding domains. Although we did not observe competition between SMARCAD1 and EXO1 for MSH2 binding in ov ere xpression e xperiments (data not shown), it is possible that this region is critical for the handover mechanism from SMARCAD1 to EXO1 facilitated by MSH2.
From an evolutionary point of view, it makes sense that key pathway proteins, such as MSH2-MSH3 dependent recruitment of EXO1, are used by more than one repair pathwa y. Shared modalities ma y also facilitate crosstalk between pathwa ys f or more ef ficient and coordina ted repair. In the case of HR, where a single persistent DSB might lead to lethality, crosstalk may ensure that all lesions are mended. It is likely that MSH2-MSH3 is important in pre v enting err or-pr one repair by POL . The collecti v e findings provide a mechanistic explanation for how the MSH2-MSH3 complex facilitates efficient DSB repair by promoting HR via recruitment of EXO1 and by pre v enting err or-pr one TMEJ by blocking POL access.

DA T A A V AILABILITY
All da ta genera ted or analyzed during this study are included in this published article (and its supplementary information files). Sequencing data are available on the SRA (BioProject ID: PRJNA772898).