Fission yeast Rad8/HLTF facilitates Rad52-dependent chromosomal rearrangements through PCNA lysine 107 ubiquitination

Gross chromosomal rearrangements (GCRs), including translocation, deletion, and inversion, can cause cell death and genetic diseases such as cancer in multicellular organisms. Rad51, a DNA strand exchange protein, suppresses GCRs by repairing spontaneous DNA damage through a conservative way of homologous recombination, gene conversion. On the other hand, Rad52 that catalyzes single-strand annealing (SSA) causes GCRs using homologous sequences. However, the detailed mechanism of Rad52-dependent GCRs remains unclear. Here, we provide genetic evidence that fission yeast Rad8/HLTF facilitates Rad52-dependent GCRs through the ubiquitination of lysine 107 (K107) of PCNA, a DNA sliding clamp. In rad51Δ cells, loss of Rad8 eliminated 75% of the isochromosomes resulting from centromere inverted repeat recombination, showing that Rad8 is essential for the formation of the majority of isochromosomes in rad51Δ cells. Rad8 HIRAN and RING finger mutations reduced GCRs, suggesting that Rad8 facilitates GCRs through 3’ DNA-end binding and ubiquitin ligase activity. Mms2 and Ubc4 but not Ubc13 ubiquitin-conjugating enzymes were required for GCRs. Consistent with this, mutating PCNA K107 rather than the well-studied PCNA K164 reduced GCRs. Rad8-dependent PCNA K107 ubiquitination facilitates Rad52-dependent GCRs, as PCNA K107R, rad8, and rad52 mutations epistatically reduced GCRs. In contrast to GCRs, PCNA K107R did not significantly change gene conversion rates, suggesting a specific role of PCNA K107 ubiquitination in GCRs. PCNA K107R enhanced temperature-sensitive growth defects of DNA ligase I cdc17-K42 mutant, implying that PCNA K107 ubiquitination occurs when Okazaki fragment maturation fails. Remarkably, K107 is located at the interface between PCNA subunits, and an interface mutation D150E bypassed the requirement of PCNA K107 and Rad8 ubiquitin ligase for GCRs. These data suggest that Rad8-dependent PCNA K107 ubiquitination facilitates Rad52-dependent GCRs by changing the PCNA clamp structure.

Introduction Faithful repair of DNA damage, such as DNA double-strand breaks (DSBs), is critical to maintaining genome integrity [1,2]. Homologous recombination is considered an error-free DSB repair mechanism, as it uses intact DNA as the template. However, when accompanied by crossover or break-induced replication, nonallelic homologous recombination between repetitive sequences in a genome results in gross chromosomal rearrangements (GCRs), including translocation, deletion, and inversion [3,4]. Isochromosomes, whose arms mirror each other, are the GCR products formed by recombination between inverted repeats around centromeres [5]. Replication fork stalling can cause GCR events without DSB formation [6]. Replication forks reassembled by homologous recombination following fork stalling are prone to switch template strands and form acentric and dicentric isochromosomes using nearby inverted repeats [7]. GCRs can cause cell death and genetic diseases, including cancer [8]. On the other hand, GCRs can also foster evolution of living organisms. However, the detailed mechanism of GCRs remains unclear.
Rad51 and Rad52 are evolutionarily conserved recombination enzymes that have pivotal roles in distinct recombination pathways [9,10]. Rad51 preferentially promotes conservative recombination, gene conversion [11][12][13]. Rad51 binds single-stranded DNA (ssDNA) and catalyzes DNA strand exchange with homologous double-stranded DNA (dsDNA) [14]. Mammalian BRCA1 and BRCA2 facilitate Rad51-dependent recombination. BRCA gene mutations increase GCRs and predispose the carriers to hereditary breast and ovarian cancer [15], demonstrating that Rad51-dependent recombination safeguards genome integrity and prevents tumorigenesis. Although yeast Rad52 mediates Rad51 loading onto RPA-coated ssDNA, human Rad52 has no such activity, and BRCA2 mediates Rad51 loading [14]. In contrast to the mediator function, both yeast and human Rad52 proteins catalyze single-strand annealing (SSA) by which complementary ssDNA strands are annealed to form dsDNA [16][17][18]. For the sake of simplicity, Rad52-dependent SSA that occurs independently of Rad51 is designated Rad52-dependent recombination in this paper. Rad52 knockout mice show only a mild defect in DNA recombinational repair and are not predisposed to cancer [19]. However, Rad52 deficiency is synthetic lethal with BRCA mutations [20,21]. Rad52 N-terminal domain that retains SSA activity is sufficient to restore cell growth of the double mutants [22], showing the importance of Rad52-dependent recombination in cells defective in Rad51-dependent recombination. Consistent with this idea, yeast rad52Δ cells exhibit more severe DNA repair and recombination defects than rad51Δ cells [3]. Previously, we have shown in fission yeast that loss of Rad51 reduces gene conversion between centromere inverted repeats and increases isochromosome formation in a manner dependent on Mus81 crossover-specific endonuclease [11,23]. Loss of Rad52 partially reduces isochromosome formation in rad51Δ cells [24], showing that isochromosomes are produced in both Rad52-dependent and Rad52-independent manners. In budding yeast, mutating the evolutionarily conserved arginine 70 (R70) in the Nterminal DNA-binding domain of Rad52 impairs in vitro SSA activity but not the mediator function [25]. Consistent with this, the rad52-R70K mutant phenotype in DNA recombination and repair is less severe than a rad52 null mutant [26]. Fission yeast Rad52 R45 is the counterpart of budding yeast Rad52 R70. The fission yeast rad52-R45K mutation also impairs in vitro SSA activity and causes less severe phenotypes than rad52 deletion [24]. The rad52-R45K mutation reduces isochromosome formation in rad51Δ cells to the same extent as rad52Δ [24], suggesting that the SSA activity of Rad52 is responsible for isochromosome formation. At fission yeast centromeres, Rad51-dependent recombination predominates, and Rad52-dependent recombination is suppressed [27]. Intriguingly, a mutation in DNA polymerase α (Pol α), required for lagging-strand synthesis, increases Rad52-dependent recombination without changing the total rate of recombination at centromeres [24], suggesting that the formation of ssDNA gaps during DNA replication might be a rate-limiting step of Rad52-dependent recombination. Rad52-dependent recombination leads to not only GCRs but also gene conversion in the absence of Rad51. Msh2-Msh3 MutS-homologs [28] and Mus81 crossover-specific endonuclease [29,30] are required for the GCR but not gene conversion branch of Rad52-dependent recombination that occurs in the absence of Rad51 [11,24], suggesting that Msh2-Msh3 and Mus81 resolve joint molecules formed by Rad52 specifically into GCR products. However, how Rad52-dependent recombination differentiates into GCRs is still unclear.
Here, we provide genetic evidence that fission yeast Rad8 (E3) and Mms2-Ubc4 (E2) ubiquitinate PCNA at K107 to facilitate Rad52-dependent GCRs. Loss of Rad8 reduced isochromosome formation but not chromosomal truncation in rad51Δ cells. Mutation in Rad8 HIRAN or RING finger but not Snf2/Swi2 translocase domain reduced GCRs. mms2 and ubc4 but not ubc13; PCNA K107R but not K164R reduced GCRs in rad51Δ cells, suggesting that the ubiquitination of K107 but not a canonical site K164 is required for GCRs. The epistatic analysis showed that Rad8 and PCNA K107 play a crucial role in the GCR branch of Rad52-dependent recombination. K107 is located at the interface between PCNA subunits, suggesting that K107 ubiquitination weakens the PCNA-PCNA interaction to cause GCRs. Remarkably, an interface mutation D150E [46,47] indeed bypassed the requirement of Rad8 ubiquitin ligase and PCNA K107 for GCRs. These data suggest that PCNA K107 ubiquitination changes the PCNA clamp structure to facilitate Rad52-dependent GCRs.

Fission yeast Rad8 facilitates isochromosome formation but not chromosomal truncation
Loss of Rad8 increases DNA damage sensitivity of rad51Δ cells [39,48], raising the possibility that Rad8 is involved in GCRs that occur independently of Rad51. To test this, we disrupted the rad8 gene in the fission yeast cell harbouring the extra-chromosome ChL C derived from chromosome 3 (chr3) (Fig 1A). ChL C retains an entire region of centromere 3 (cen3) that consists of a unique sequence, cnt3, surrounded by inverted repeats: imr3, dg, dh, and irc3, and contains three genetic markers: LEU2, ura4 + , and ade6 + [11,23]. As ChL C is dispensable for cell growth, one can detect otherwise lethal GCRs using ChL C . To detect Leu + Ura − Ade − GCR clones that have lost both ura4 + and ade6 + , cells were grown in Edinburgh minimum media supplemented with uracil and adenine (EMM+UA) and plated on FOA+UA media containing 5-fluoroorotic acid (5-FOA) which is toxic to ura4 + cells. Leu + Ura − colonies formed on FOA +UA plates were transferred to EMM+U to inspect adenine auxotrophy. Fluctuation analysis showed that rad8Δ did not significantly change spontaneous GCR rates in wild-type cells ( Fig  1B). However, rad8Δ partially but significantly reduced GCRs in rad51Δ cells (Fig 1B), indicating that Rad8 facilitates GCRs that occur independently of Rad51.
Three types of GCRs have been detected using the extra-chromosome: translocation, isochromosome, and truncation, which can be distinguished by their lengths [11,23,24,[49][50][51] ( Fig 1A). Among them, isochromosomes are formed by recombination between inverted repeats at centromeres. To determine GCR types Rad8 causes, chromosomal DNAs were prepared from the parental and independent GCR clones, resolved by pulsed-field gel electrophoresis (PFGE) under two different conditions (broad-and short-range PFGE), and stained with ethidium bromide (EtBr) (Figs 1C and S1; results for wild-type and rad8Δ and additional results for rad51Δ and rad8 rad51Δ). As previously observed [24], isochromosomes and a small number of translocations were detected in wild-type cells (Table 1). rad8Δ did not significantly change the GCR types in wild-type cells (P = 0.6, the two-tailed Fisher's exact test). As previously observed [23,24,50], rad51Δ cells produced isochromosomes and a small number of truncations ( Fig 1C, sample #4). Given elevated GCR rates (Fig 1B), these results show that rad51Δ increases isochromosome formation and chromosomal truncation. In rad51Δ cells, rad8Δ increased the proportion of truncations from 10 to 37% (Table 1; P = 0.030, the twotailed Fisher's exact test). To obtain the rate of each GCR type (Fig 1D), we multiplied the total GCR rate (Fig 1B) by the proportion of each GCR type ( Table 1). Loss of Rad8 eliminated 75% of the isochromosomes produced in rad51Δ cells. On the other hand, Rad8 was dispensable for chromosomal truncation. These data demonstrate that Rad8 is specifically involved in the major pathway of isochromosome formation in rad51Δ cells. The ChL C chromosome contains an entire region of centromere 3 (cen3) and three genetic markers: LEU2, ura4 + , and ade6 + . Gross chromosomal rearrangements (GCRs) associated with loss of both ura4 + and ade6 + result in the formation of Leu + Ura − Ade − cells. The structure and the length of three GCR types: translocation, isochromosome, and truncation are shown. (B) GCR rates of the wild-type, rad8Δ, rad51Δ, and rad8Δ rad51Δ strains (TNF5369, 5549, 5411, and 5644, respectively). Each dot represents an independent experimental value obtained from an independent colony. Black bars indicate medians. Rates relative to that of wild type are shown on the top of each cluster of dots. Statistical analyses between the wild-type and mutant strains, and that between rad51Δ and rad8Δ rad51Δ strains were performed by the two-tailed Mann-Whitney test. ���� P < 0.0001. (C) Chromosomal DNAs prepared from the parental (P) and independent GCR clones of rad51Δ and rad8Δ rad51Δ strains were separated by broad-and short-range PFGE (top and bottom rows, respectively). Positions of chr1, chr2, chr3, and ChL C (5.7, 4.6,~3.5, and 0.5 Mb, respectively) are indicated on the left of broad-range PFGE panels. Sizes of Lambda (λ) ladders (ProMega-Markers, Madison, Wisconsin, G3011) are displayed on the left of short-range PFGE panels. Sample number of truncations are highlighted in red. (D) Rates of each GCR type. Rates relative to those of rad51Δ are indicated. Numerical source data underlying the graphs shown in (B) and (D) are provided respectively in Tables A and B in S1 File. It has been shown that truncation ends are healed by de novo telomere addition [52][53][54]. To determine the chromosomal sites to which telomere sequences have been added, we recovered chromosomal truncations from agarose gels and amplified the breakpoints using a pair of ChL C and telomere primers: M13R-C19 or M13R-T1 [55]. DNA sequencing of the PCR products revealed that telomere sequences (G 2-5 TTAC (A) (C)) [56] were added either within or outside centromere repeats (Tables 1 and 2). Only 0~3-bp overlaps were detected between ChL C and telomere sequences around the breakpoints, suggesting that no extensive annealing with telomere RNA is required to initiate telomere addition. No apparent differences were detected between rad51Δ and rad8Δ rad51Δ. Together, these results show that Rad8 is required for homology-mediated GCRs resulting in isochromosome formation but dispensable for de novo telomere addition resulting in chromosomal truncation.

Rad8 HIRAN and RING finger domains are required for GCRs
Rad8 and its homologs contain HIRAN, Snf2/Swi2 translocase, and RING finger domains (Fig  2A, top). To gain insight into how Rad8 facilitates GCRs, we introduced alanine (A) Telomere repeat sequences are bold. The nucleotides overlapping between ChL C and telomere sequences are placed between lines. The breakpoint positions are substitution into each domain (Fig 2A, bottom). The rad8-HIRAN mutation alters the residues forming the 3' DNA-end binding pocket [41][42][43]. The rad8-ATPase mutation alters the conserved residues in the ATP-binding Walker A motif [57]. The rad8-RING mutation changes a residue involved in the interaction with ubiquitin-conjugating enzymes [58,59]. rad8-ATPase did not significantly change GCR rates in both wild-type and rad51Δ cells (Fig 2B), indicating that Rad8 translocase activity is not essential for GCRs. On the other hand, rad8-HIRAN and rad8-RING reduced GCR rates, suggesting that Rad8 facilitates GCRs through 3' DNA-end binding and ubiquitin ligase activity. rad8-HIRAN and rad8-RING exhibited slightly more pronounced effects on GCR rates than rad8Δ (Figs 1B and 2B). In addition to Rad8-dependent GCR events, the mutant proteins may interfere with the GCR events that occur when Rad8 proteins were absent.
Rad52 promotes gene conversion as well as GCRs in the absence of Rad51 [24]. However, Msh2-Msh3 and Mus81 are not required for gene conversion branch of Rad52-dependent recombination. To determine whether PCNA K107 ubiquitination is also involved in gene conversion, we determined the rate of gene conversion between ade6B and ade6X heteroalleles that results in adenine prototrophs [27] (Fig 4D and 4E). As previously observed [24], rad52-R45K reduced gene conversion only in the absence of Rad51, suggesting the role of Rad52 SSA in Rad51-independent gene conversion. Contrary to rad52-R45K, pcn1-K107R did not reduce gene conversion even in the absence of Rad51. PCNA K107 was also dispensable for cells to survive exogenous DNA damage (Fig 4F). pcn1-K107R did not increase the sensitivity of rad51Δ cells to hydroxyurea (HU), methyl methanesulfonate (MMS), and camptothecin (CPT), while pcn1-K164R increased the sensitivity [39,48]. Together, these results suggest that, like Msh2-Msh3 and Mus81 [11,24], PCNA K107 ubiquitination is specifically required for nonallelic crossover recombination that results in GCRs but not gene conversion branch of Rad52-dependent recombination, which may play a role in repairing exogenous DNA damage.
In budding yeast, PCNA K107 ubiquitination is induced and required for cell growth in DNA ligase I mutant strains [44]. pcn1-K107R also exhibited synthetic growth defects in the fission yeast DNA ligase I cdc17-K42 mutant [62] at semipermissive temperatures (Fig 4G), suggesting that PCNA K107 ubiquitination occurs when Okazaki fragment maturation fails.

Discussion
Here, we provided genetic evidence that fission yeast Rad8 facilitates Rad52-dependent GCRs through PCNA K107 ubiquitination. Loss of Rad8 reduced isochromosome formation in rad51Δ cells. Mutations in Rad8 HIRAN and RING finger but not Snf2/Swi2 domain reduced GCRs. mms2 and ubc4 but not ubc13 mutations; PCNA K107R but not K164R reduced GCRs in rad51Δ cells. The epistatic analysis suggests that PCNA K107 ubiquitination plays a crucial role in the Rad52-dependent GCR pathway that involves Msh2-Msh3 and Mus81. PCNA K107 is located at the interface between PCNA subunits, suggesting that its ubiquitination affects the PCNA-PCNA interaction to cause GCRs. Indeed, an interface mutation D150E bypassed the requirement of PCNA K107 ubiquitination for GCRs. These data suggest that Rad8-dependent K107 ubiquitination changes the structure of the PCNA clamp to facilitate Rad52-dependent GCRs.
Our data suggest that Rad8 facilitates GCRs through 3' DNA-end binding and ubiquitin ligase activity (Fig 6). Rad8 ubiquitin ligase acts with Mms2-Ubc13 ubiquitin-conjugating E2 enzymes to cause template switching [39]. However, in the case of GCRs, Rad8 functions with Mms2-Ubc4, as mms2 and ubc4 but not ubc13 mutations reduced GCRs in rad51Δ cells. That is further supported by the fact that rad8-RING and ubc4 mutations epistatically reduce GCRs. Ubc4 and Ubc13 contain the cysteine residue critical for the E2 activity, while Mms2 is an E2 variant that lacks the active site [64]. Ubc13 but not Mms2 interacts with budding yeast Rad5 [58]. When Mms2-Ubc4 interacts with Rad8, Ubc4 rather than Mms2 may be the protein that directly binds Rad8 RING finger. E2 complexes determine substrate specificity for Rad8 ubiquitination: Mms2-Ubc4 promotes PCNA K107 while Mms2-Ubc13 causes PCNA K164 ubiquitination [37,44]. It is unknown how Rad8 chooses the E2 partner. The Snf2/Swi2 but not HIRAN domain is essential for PCNA K164 ubiquitination [43,65,66]. In contrast, the HIRAN but not Snf2/Swi2 domain was required for GCRs. It is tempting to suggest that DNA binding via the HIRAN domain facilitates the RING finger to accommodate Mms2-Ubc4 rather than Mms2-Ubc13. However, it is also possible that the rad8-HIRAN mutation directly or indirectly interferes with Rad8-PCNA interaction [67] and thereby reduces PCNA K107 ubiquitination and GCRs. It is also possible that DNA binding of the HIRAN domain has a ubiquitination-independent role to cause GCRs [43,68,69]. Further study is required to understand how Rad8 interacts with Mms2-Ubc4 and Mms2-Ubc13 to ubiquitinate different substrates.
In the absence of Rad51, isochromosomes are formed in Rad52-dependent and Rad52-independent manners [24]. Several lines of evidence suggest that Rad8-dependent PCNA K107 ubiquitination plays a crucial role in Rad52-dependent GCRs. Firstly, both Rad8 and Rad52 are involved specifically in homology-mediated GCRs that result in isochromosome formation. In rad51Δ cells, loss of Rad8 and Rad52 eliminated 75 and 90% of isochromosomes, respectively, but did not reduce chromosomal truncations [24]. Secondly, rad52Δ and PCNA K107R epistatically reduced GCRs. In addition, the rad52-R45K mutation that impairs SSA activity [24] epistatically reduced GCRs with rad8Δ as well as PCNA K107R. Thirdly, PCNA K107R reduced GCRs epistatically with loss of Msh3 or Mus81, both of which have been implicated in Rad52-dependent GCRs [24]. Collectively, it seems that Msh2-Msh3 MutS-homologs, Mus81 endonuclease, and Rad8-dependent PCNA K107 ubiquitination are involved in Rad52-dependent GCR pathway. In addition to GCRs, Rad52 facilitates gene conversion between inverted repeats in the absence of Rad51. However, like Msh2-Msh3 and Mus81, PCNA K107 was dispensable for the gene conversion in rad51Δ cells, suggesting that PCNA K107 ubiquitination is specifically involved in the GCR branch of Rad52-dependent recombination. Interestingly, human K110 is also located at the PCNA-PCNA interface (Fig 5B), and its ubiquitination has been detected in cultured cells [70], suggesting that human K110 is the counterpart of yeast K107.
Homology-mediated GCRs occur even in the absence of Rad52. Isochromosome formation is increased in rad52Δ rad51Δ and rad8Δ rad51Δ cells compared to wild-type cells ( [24] and this study). The fission yeast Fbh1 helicase disrupts Rad51 nucleoprotein filaments [61], and a mutation in Fbh1 suppresses rad52Δ mutant phenotypes [30,[71][72][73], raising the possibility that a residual level of GCRs in rad52Δ rad51Δ cells is caused by a spontaneous mutation in the fbh1 gene. However, that is unlikely, because Rad51 is required for the fbh1 suppression [71]. Indeed, DNA sequencing showed no fbh1 mutations in the parental cells of the rad52 strains used in this study (S2 File) and in two independent GCR clones of the rad52Δ rad51Δ strain (S3 File). Recombination between centromere inverted repeats may occur even in the absence of Rad52 and Rad51, resulting in isochromosome formation. Interestingly, in budding yeast, most of the heteroallelic recombination in rad52Δ rad51Δ cells are crossover recombination while noncrossover recombination in wild-type cells [74,75]. Rad51 and Rad52-independent recombination has been reported in other repetitive regions of chromosomes. In budding yeast, Rad51 and Rad52 are dispensable for repairing DSBs formed at the replication fork barrier site in rDNA repeats in wild type [76]. Alternative lengthening of telomeres (ALT) is a telomerase-independent but recombination-dependent process that maintains telomeres. Rad51 is dispensable for ALT in human cells, and there are Rad52-dependent and Rad52-independent ALT pathways [77]. POLD3 and POLD4 subunits of DNA polymerases δ (Pol δ) are required for ALT in Rad52-knockout cell lines [77], suggesting the role of Pol δ in Rad52-independent recombination. Interestingly, budding yeast Pol δ is also required for microhomology/microsatellite-induced replication (MMIR) or microhomology-mediated repair (MHMR) that occurs in the absence of Rad52 [78,79], suggesting the role of Pol δ in stabilizing otherwise unstable joint molecules. Break-induced replication [80] and DSB-independent replication template exchange [6,81] also require Pol δ. Although a mutation in a Cdc27/ Pol32/POLD3 subunit of Pol δ did not reduce GCRs in rad51Δ cells [11], Pol δ could be required for isochromosome formation in rad51Δ rad52Δ cells [82]. One of the future directions is to understand how homology-mediated GCRs occur independently of Rad52.
Post-translational modifications of PCNA at K164 affect protein interaction. K164 ubiquitination facilitates PCNA binding to translesion synthesis polymerases and Mgs1/ZRANB3 helicase [36]. In budding yeast, Small ubiquitin-like modifier (SUMO) protein is also attached to PCNA K164. K164 SUMOylation causes PCNA interaction with Srs2 helicase that suppresses Rad51-dependent recombination by dissociating Rad51 protein from ssDNA [83]. However, the C-terminal domain of budding yeast Srs2 that interacts with SUMOylated PCNA is not conserved in fission yeast Srs2 [39]. Nonetheless, loss of Rad51 resulted in slow growth phenotypes and increased sensitivities to DNA damaging agents in pcn1-K164R but not in pcn1-K107R cells [39] (Fig 4F). PCNA K164R slightly but significantly reduced GCRs in wild-type but not in rad51Δ cells ( Fig  3C). PCNA K164R may channel DNA repair into Rad51-dependent gene conversion pathway.
How does PCNA K107 ubiquitination cause GCRs? Our data suggest that K107 ubiquitination changes the PCNA clamp structure to facilitate GCRs (Fig 6). PCNA subunits interact with each other to form ring-shaped homotrimers. K107 is located at the interface between PCNA subunits. An interface mutation D150E [46,47] bypassed the requirement of Rad8 RING finger and PCNA K107 for GCRs, showing that the ubiquitin or ubiquitin chain at K107 is not essential for GCRs and that weakening the PCNA-PCNA interaction is sufficient to cause GCRs. PCNA K107R mutation did not significantly change the level of Rad52 foci accumulated in the absence of Rad51 [84] (S5 Fig), suggesting that PCNA K107 ubiquitination is not required for Rad52 to localize to damage sites. However, the PCNA clamp present at the end of Okazaki fragments can be a structural hinderance of Rad52-dependent SSA. Consistent with this idea that PCNA K107 ubiquitination occurs when PCNA is bound to Okazaki fragments, PCNA K107R showed synthetic growth defects in DNA ligase I cdc17 mutant cells. In budding yeast, PCNA K107R also causes growth defects in DNA ligase I mutants, and K107 ubiquitination facilitates checkpoint activation that depends on the assembly of checkpoint proteins onto ssDNA [44,45]. RFC-like complexes containing Elg1 unload PCNA from DNA after Okazaki fragment ligation [34,35]. Elg1 promotes Rad51 and Rad52 proteins' recruitment to stalled replication forks and facilitates nearby recombination [63], suggesting that PCNA unloading facilitates recombination around stalled forks. An interface mutation D150E also bypasses the role of Elg1 in PCNA unloading and promoting recombination around stalled replication forks [46,63]. However, unlike PCNA K107R, loss of Elg1 did not reduce GCRs, suggesting that Elg1-dependent PCNA unloading is not required to cause GCRs. However, we do not exclude the possibility that K107 ubiquitination facilitates PCNA unloading, by itself or with the factors other than Elg1. Like Msh2-Msh3 MutS-homologs and Mus81 endonuclease [24], it appears that PCNA K107 ubiquitination is specifically required for the GCR branch of Rad52-dependent recombination. It has been reported that the PCNA clamp interacts with Msh2-Msh3 [85] and Mus81 to enhance endonuclease activity [86]. The structural change of the PCNA clamp induced by K107 ubiquitination might facilitate the interaction with Msh2-Msh3 and Mus81 and result in GCRs (Fig 6). Our Western blot failed to detect PCNA ubiquitination that clearly depends on K107, suggesting that K107 ubiquitination occurs in a limited fraction of PCNA molecules (S6 Fig). In the future, biochemical study is required to detect PCNA K107 ubiquitination and address the effect of K107 ubiquitination on the structure and function of PCNA.
This study has uncovered the role of Rad8 ubiquitin ligase and PCNA K107 in Rad52-dependent GCRs. We also suggest that human PCNA K110 is the counterpart of yeast PCNA K107. Interestingly, HLTF, the human homolog of Rad8, is often amplified and overexpressed in cancer [68], suggesting that HLTF causes GCRs and facilitates tumorigenesis.

Genetic procedures and yeast strains
The fission yeast strains used in this study are listed in S1 Table. Standard genetic procedures and fission yeast media were used as previously described [11]. Pombe minimal glutamate (PMG) medium is identical to Edinburgh minimal medium (EMM), except containing 3.75 g/ l monosodium L-glutamate instead of 5 g/l ammonium chloride. Yeast nitrogen base (YNB) medium contained 1.7 g/l yeast nitrogen base (BD Biosciences, San Jose, California, Difco 233520), 5 g/l ammonium sulphate (Nacalai Tesque, Kyoto, Japan, 02619-15), and 2% glucose. YE, YNB, EMM, and PMG contain 225 mg/l of each amino acid when indicated. FOA+UA is a YNB derivative supplemented with 56 mg/l uracil, 225 mg/l adenine, and 1 g/l 5-fluoroorotic acid monohydrate (Apollo Scientific, Stockport, United Kingdom, PC4054). Yeast cells were grown at 30˚C. When we inoculated liquid media using slow-growing mutant strains, we did not take exceptionally large colonies, if any, because they might contain a mutation that suppresses the growth and other defects. S7 Fig shows a picture of rad52Δ rad51Δ colonies formed on EMM+UA plates.
The rad8-HIRAN mutant strain was created by the pop-in/pop-out gene replacement [87]. pTN1192 plasmid containing ura4 + and rad8-HIRAN was linearised by BglII digestion at a unique site in the rad8 region and introduced into ura4-D18 cells. Ura + transformants were selected on EMM. After confirmation of the correct integration by PCR and DNA sequencing, the rad8:ura4 + :rad8-HIRAN strain was grown in YE media supplemented with uracil (YE+U) and then plated on FOA+U media to select Ura − colonies, resulting from "pop-out" of the ura4 + marker. DNA sequencing confirmed the retention of the rad8-HIRAN mutation in the Ura − cells. rad8-ATPase and rad8-RING mutant strains were created essentially in the same way, but BglII-digested pTN1191 plasmid containing rad8-ATPase and PacI-digested pTN1193 plasmid containing rad8-RING were used, respectively.
We created the ubc4-P61S mutation as follows. Two partially overlapping fragments were amplified separately: a 0.7 kb fragment using ubc4-F1 and ubc4-P61S-B primers, and a 1.5 kb fragment using ubc4-P61S-T and ubc4-R4 primers. ubc4-P61S-B and ubc4-P61S-T contain the ubc4-P61S mutation. The two PCR products were mixed and connected by the second round of PCR in the presence of ubc4-F1 and ubc4-R4 primers, resulting in the formation of a 2.2 kb fragment containing the ubc4-P61S mutation. To create the ubc4-P61S strain, we first introduced the ura4 + gene 0.6 kb downstream of the ubc4 + gene in ura4-D18 cells, making ubc4 + :ura4 + cells. Then, the 2.2 kb PCR fragment that contains the ubc4-P61S mutation and encompasses the ura4 + integration site was introduced into ubc4 + :ura4 + cells. Ura − transformants were selected on FOA+U plates. DNA sequencing confirmed the correct integration of ubc4-P61S. Sequences of PCR primers used are listed in S2 Table.

Plasmids
We constructed the plasmid pTN1192 containing ura4 + and rad8-HIRAN as follows. Two partially overlapping fragments were amplified separately from yeast genomic DNA: a 1.4 kb fragment using rad8-KpnI and rad8-HIRAN-B primers, and a 1.0 kb fragment using rad8-HIRAN-T and rad8-SalI primers. rad8-HIRAN-B and rad8-HIRAN-T contain the rad8-HIRAN mutation. The two PCR products were mixed and connected by the second round of PCR in the presence of rad8-KpnI and Rad8-SalI primers. A 2.3 kb KpnI-SalI restriction fragment of the PCR product containing the rad8-HIRAN mutation was introduced between KpnI-SalI sites of pTN782 [88], which contains a 1.5 kb HindIII-SspI fragment containing the ura4 + gene between HindIII-EcoRV sites of pBluescript II KS + (Agilent, Santa Clara, California).

GCR rates
Spontaneous rates of GCRs that result in loss of ura4 + and ade6 + markers were determined essentially as previously described [24]. Yeast cells harbouring ChL C were incubated on EMM +UA plates for 6-8 days. 10 ml EMM+UA was inoculated with a single colony from EMM +UA plates and incubated for 1-2 days. Cells were plated on YNB+UA and FOA+UA plates and incubated for 5-8 days. Leu + and Leu + Ura − colonies formed on YNB+UA and FOA+UA plates, respectively, were counted. Leu + Ura − colonies were streaked on EMM+UA plates to examine the colony size and transferred to EMM+U plates to inspect adenine auxotrophy. The number of Leu + Ura − Ade − was obtained by subtracting the number of Leu + Ura − Ade + from that of Leu + Ura -. Rates of GCR per cell division were calculated [89], using the numbers of Leu + cells and Leu + Ura − Ade − cells, using Microsoft Excel for Mac 16 (Microsoft, Redmond, Washington). The GCR strains were constructed in the h-(smt0) background [90] (S1 Table), in which no DSBs are formed at the mat1 locus.

Gene conversion rates
Spontaneous rates of gene conversion that occurs between ade6B and ade6X heteroalleles integrated at the ura4 locus [27] were determined. 10 ml PMG+UA was inoculated with a single colony from PMG+UA plates and incubated for 1-2 days. Cells were plated onto PMG+UA and PMG+U media. After 4-7 days' incubation of the plates, colonies formed on PMG+UA and PMG+U were counted to determine the number of colony-forming units and Ade + prototrophs, respectively. The rates of gene conversion per cell division were calculated [89], using Microsoft Excel for Mac 16.

Cell imaging
Exponentially growing cells in EMM were collected, seeded on glass-bottom dishes (Matsunami Glass, Osaka, Japan, D11130H), and observed using a DeltaVision Personal fluorescence microscopy system (GE Healthcare), which is based on an Olympus wide-field IX71 fluorescence microscope equipped with a CoolSNAP HQ2 CCD camera (Photometrics, Tucson, Arizona) and an oil-immersion objective lens (UAPO 40×; NA = 1.35; Olympus, Tokyo, Japan).

Statistical analyses
Two-tailed Mann-Whitney tests and Fisher's exact tests were performed using GraphPad Prism for Mac version 8 (GraphPad Software, San Diego, California). Two-tailed student's ttests were performed using Microsoft Excel for Mac 16.