The Role of Replication Bypass Pathways in Dicentric Chromosome Formation in Budding Yeast

Abstract

Gross chromosomal rearrangements (GCRs) are large scale changes to chromosome structure and can lead to human disease. We previously showed in Saccharomyces cerevisiae that nearby inverted repeat sequences (∼20–200 bp of homology, separated by ∼1–5 kb) frequently fuse to form unstable dicentric and acentric chromosomes. Here we analyzed inverted repeat fusion in mutants of three sets of genes. First, we show that genes in the error-free postreplication repair (PRR) pathway prevent fusion of inverted repeats, while genes in the translesion branch have no detectable role. Second, we found that siz1 mutants, which are defective for Srs2 recruitment to replication forks, and srs2 mutants had opposite effects on instability. This may reflect separate roles for Srs2 in different phases of the cell cycle. Third, we provide evidence for a faulty template switch model by studying mutants of DNA polymerases; defects in DNA pol delta (lagging strand polymerase) and Mgs1 (a pol delta interacting protein) lead to a defect in fusion events as well as allelic recombination. Pol delta and Mgs1 may collaborate either in strand annealing and/or DNA replication involved in fusion and allelic recombination events. Fourth, by studying genes implicated in suppression of GCRs in other studies, we found that inverted repeat fusion has a profile of genetic regulation distinct from these other major forms of GCR formation.

ALL organisms are prone to large-scale changes (gross chromosomal rearrangements, GCRs) to their genomes that include deletions, inversions, and translocations. These large-scale changes are thought to drive evolutionary events, such as speciation, and contribute to human pathology such as Pelziaeus-Merzbacher syndrome and other genetic disorders (Lee et al. 2007; Stankiewicz and Lupski 2010). Thus, a firm understanding of how cells normally prevent such rearrangements, and how they accumulate, is critical to our understanding of both evolution and pathology.

GCRs arise by many different mechanisms, and there is growing evidence that errors during DNA replication are a major source (Myung et al. 2001; Admire et al. 2006; Mizuno et al. 2009). Errors are thought to arise when replication forks encounter “lesions” on the template strand. Lesions can consist of protein complexes bound to DNA or lesions in the DNA itself. Replication forks bypass lesions by several different mechanisms that are still poorly understood (Atkinson and McGlynn 2009; Weinert et al. 2009). We believe that understanding lesion bypass mechanisms is central to understanding both how GCRs are prevented and how they form when lesion bypass mechanisms fail.

All lesion bypass pathways utilize sequence homology to restart replication (Atkinson and McGlynn 2009; Weinert et al. 2009). Use of sequence homology during restart may limit the frequency of GCRs, as it lowers the probability of annealing to nonallelic sequences. Repetitive sequences present a problem because lesion bypass at sites near repetitive sequences may lead to annealing of nonallelic sequences and thus to GCR formation (Lemoine et al. 2005; Narayanan et al. 2006; Argueso et al. 2008). Indeed in yeast and in other organisms, GCRs occur frequently in repeat sequences (Dunham et al. 2002; Argueso et al. 2008; Di Rienzi et al. 2009). Some rearrangements do occur between so-called “single-copy sequences” with either no homology or limited homology (microhomologies of 5–9 bp; Myung et al. 2001; Kolodner et al. 2002; Putnam et al. 2005) though evidence suggests these rearrangements occur less frequently than rearrangements between repetitive sequences (Putnam et al. 2009). Interestingly, it has been shown that some genes are required to prevent the fusion of repetitive elements yet have no effect on rearrangements between single-copy sequences (Putnam et al. 2009). Currently it is not clear how these pathways act to suppress repeat-mediated events and why they are not required to prevent rearrangements between single-copy sequences.

Our current understanding of the mechanisms underlying GCR formation is mostly derived from assays designed to detect specific changes to yeast chromosomes (Chen and Kolodner 1999; Myung et al. 2001; Huang and Koshland 2003; Lambert et al. 2005; Rattray et al. 2005; Admire et al. 2006; Narayanan et al. 2006; Schmidt et al. 2006; Smith et al. 2007; Pannunzio et al. 2008; Payen et al. 2008; Paek et al. 2009; Mizuno et al. 2009). Previously we reported on GCR formation in the budding yeast Saccharomyces cerevisiae using an assay we developed. We found that a major source of genome instability involves the fusion of nearby inverted repeats (with ∼20–200 bp of sequence homology, separated by 1–5 kb) to form either dicentric or acentric chromosomes (Figure 1D; Paek et al. 2009). We also found that fusion of inverted repeats is general: fusion occurred between inverted repeats at all five different locations tested on four different yeast chromosomes, as well as between synthetic inverted repeats (Paek et al. 2009). Genetic data suggest that these events most likely occur during replication of DNA (Admire et al. 2006). Further genetic analysis suggested that the mechanism of inverted repeat fusion differed from that of direct repeat recombination, in that inverted repeat fusion did not require genes involved in homologous recombination (HR) or single-strand annealing (SSA) pathways (Paek et al. 2009). In addition, fusion events are unlikely to involve double-strand breaks (DSBs), as genes in the nonhomologous end joining (NHEJ) and microhomology-mediated end joining (MMEJ) are not required for fusion events (Paek et al. 2009). Indeed gene knockouts in the HR (RAD52, RAD51, and RAD59), SSA (RAD52 and RAD1) and postreplication repair (PRR) (RAD18) pathways actually increased the frequency of fusion of an inverted repeat on chromosome (Chr) VII (Paek et al. 2009); these pathways normally suppress inverted repeat fusion.

Figure 1.—

Experimental setup for the detection of inverted repeat fusion and chromosome instability. Objects are not drawn to scale. (A) The starting strain has two copies of Chr VII. One copy contains the CAN1 gene, ADE6, ade3, while the other copy is ade6, ADE3. Cells are plated to canavanine, and three types of colonies are formed: (B) Allelic recombinants are round in appearance and are Ade⁺; (C) colonies that form by loss of Chr VII are round in appearance and Ade⁻; and (D) cells that contain unstable dicentric chromosomes form by the fusion of inverted repeats. One specific case of this fusion (the S2/S3 dicentric) is shown within braces. Cells with dicentrics form mixed colonies, which contain allelic recombinants, chromosome loss events, as well as a translocation between D7 and D11. The bar in the S2/S3 repeat represents a fusion junction. (E) The specific dicentric is detected by dicentric primers DP1 and DP2 and (F) a monocentric translocation that is detected with translocation primers TP1 and TP2.

To further our previous studies, we analyzed three groups of genes implicated in the maintenance of genome stability. We tested how these genes affect the overall stability of Chr VII, focusing on the fusion of nearby inverted repeats to form a specific dicentric Chr VII and the resolution of the dicentric into a monocentric translocation (which we term the 403–535 translocation; Figure 1, D–F). First, we analyzed several genes in the PRR pathway and found that error-free bypass, but not translesion synthesis, is required for the prevention of inverted repeat fusion. Surprisingly, we found that siz1 mutants, which are defective for Srs2 recruitment to replication forks, and srs2 mutants had opposite effects on instability. This may reflect separate roles for Srs2 in different phases of the cell cycle. Second, we analyzed several mutations in genes that are associated with replication forks. We found that mutants in POL3 (polymerase delta) and MGS1 (encoding a single-strand annealing protein, which binds polymerase delta) significantly reduced the frequency of dicentric formation and allelic recombinants that arise in the checkpoint mutant rad9 (Giot et al. 1997; Hishida et al. 2001; Paek et al. 2009). Finally we studied genes associated with rearrangements involving repeats or single-copy sequences, as well as a subset of mutants involved in recombination. Generally, we find that the mechanisms of nearby inverted repeat fusion are distinct from mechanisms fusing longer repeats or single-copy sequences.

MATERIALS AND METHODS

Yeast strains:

All strains are derived from TY200 and TY206, which are derivatives of the A364a yeast strain (Weinert and Hartwell 1990; Admire et al. 2006). TY200 is a Chr VII disome that is MATα HXK2lys5cyh2trp5leu1 cenVII ade6ADE3/hxk2:CAN1LYS5CYH2TRP5LEU1 cenVII ADE6ade3. TY206 contains a rad9 null mutation, rad9∷ura3, which was made as described in Weinert and Hartwell (1990). It is a RAD9 deletion marked with URA3, which was then subjected to FOA selection to isolate an ura3 mutant cell to allow use of the URA3 gene for further genetic changes. Additional null mutants were made by transformation of PCR products derived from Euroscarf strains with appropriate PCR primers and selection for resistance to geneticin (Table S1; Wach et al. 1994). Mutations were verified by PCR and phenotypic assays (i.e., damage sensitivity) where appropriate. The pol3-13 mutation was introduced using the two-step gene replacement strategy (Giot et al. 1997). First a pY9 plasmid (gift of Dmitry Gordenin) bearing the pol3-13 mutation and URA3 was digested with Hpa1, and Ura⁺ colonies were selected. The native POL3 gene was removed by selecting for loss of URA3 by plating to 5-FOA, and candidates were verified by sequencing. A similar strategy using the Ycplac22–POL30 (gift of Anja-Katrin Bielinsky) plasmid bearing various POL30 mutations was used to generate the proliferating cell nuclear antigen (PCNA) mutations (Hoege et al. 2002). For all mutants reported, two separate mutants were made and analyzed.

Chromosome instability assays:

All assays were performed as described in Admire et al. (2006). Cells were grown on YEP plates supplemented with 2% dextrose (YEPD) for 2–3 days at 30° (unless otherwise noted). Single colonies were then resuspended in water, counted, and plated onto selective medium with 60 μg/ml canavanine and all essential amino acids except arginine and serine (Can plates), and in addition, the same selective medium yet lacking adenine (Can −Ade plates). Colonies that grew on Can plates and were Ade⁻ were scored as loss events. Can^R Ade⁺ colonies that were round in appearance were scored as allelic recombinants, while those with a serrated or sectored appearance (see Figure 1) were scored as “mixed colonies.” We identify colonies as “sectored” visually, and verify their genetic content as mixed by lineage assays. We verified by a lineage assay (Admire et al. 2006) that each round colony consisted of cells with the same genotype (>95%), while each mixed colony contained cells of multiple genotypes. In rare cases where this was not the case, frequencies were corrected accordingly. For example, if 1 out of 10 sectored colonies proved to have a stable genotype (i.e., all cells were of one genotype), then the mixed colony frequency would be adjusted to 90% of the sectored colony frequency observed. The lineage assay also provides a means of determining where events in the mixed colonies had occurred. We can determine where events occurred by assessing the markers on Chr VII (see genotype of Chr VII above). In all strains tested in this study, the distribution of recombination events in mixed colonies are similar to results found in Rad⁺ and rad9 strains. That is, the majority of events (∼60%) occurred in the E2 genetic region linked to the 403 fragile site (data not shown). Frequencies were determined from analysis of at least 12 colonies, 6 from each of two independently derived mutants. Statistical tests were performed using the Kruskal–Wallis method (Kruskal and Wallis 1952).

Detection of the Chr VII^403–535 translocation:

For each strain we screened for the presence of the Chr VII^403–535 translocation in both round and mixed Can^R Ade⁺ colonies (see Figure 1F). Cells from Can^R Ade⁺ colonies were grown in 3 ml YEPD, and genomic DNA was prepared and subjected to PCR amplification using primers that detect the Chr VII^403–535 translocation (see Figure 1F). The coordinates for these primers in the Saccharomyces Genome Database (SGD) are VII 406,041–406,022 and VII 535,582–535,562. The 1.1-kb translocation band was sequenced in a previous publication (Admire et al. 2006). For each strain at least six independent Can^R Ade⁺ colonies were analyzed by PCR to judge whether the translocation was generated.

Quantification of a specific dicentric chromosome by real-time PCR:

To determine the frequency of inverted repeat fusion to form a dicentric chromosome at the site on Chr VII that is 403 kb from the left telomere (“the 403 fragile site”), we employed a quantitative PCR assay explained previously (Paek et al. 2009). Briefly, cells that initially contained an intact Chr VII were grown on YEPD for 2 days at 30° to allow formation of dicentrics (the pol3-13 mutants were grown at different temperatures, as noted). Single colonies were then picked and grown overnight in YEPD liquid cultures to generate enough cells to isolate genomic DNA. Two separate qPCR assays were carried out on each genomic DNA preparation, one to a “normal” unrearranged DNA region of Chr V (SGD coordinates Chr V 410,165–410,189 and Chr V 412,617–412,639) to determine the relative amount of genomic DNA in the sample and the second qPCR detected the specific dicentric (SGD coordinates VII 405,579–405,606 and VII 402,296–402,319, see Figure 1E). Each qPCR reaction was done in triplicate, and for each strain six independent colonies were analyzed. Average cycle number was calculated and compared to standards that consisted of purified PCR product of known concentration. By comparing standard concentrations to samples, we were able to determine the number of copies of the specific dicentric in each sample.

RESULTS

The system:

The genetic system that we use is a haploid strain of yeast containing two copies of Chr VII (a Chr VII disome, Figure 1A; Carson and Hartwell 1985). Having an extra copy of Chr VII allows for changes to occur on one copy without affecting cell viability. To identify cells with an altered copy of Chr VII, we used a strain that has a copy of CAN1 on one left arm (Figure 1A; Carson and Hartwell 1985). We identified changes to this chromosome by selecting for loss of CAN1 by plating on medium containing canavanine. Canavanine is a toxic analog of arginine that is imported into the cell by the Can1 protein (arginine permease), causing cell death (Walker 1955). Therefore, to survive on canavanine medium, cells must lose the CAN1 gene. This is predominantly done in one of three ways; allelic recombination (Figure 1B), chromosome loss (Figure 1C), or nonallelic recombination that results in formation of mixed colonies (see mixed colony in Figure 1D; Paek et al. 2009). Chromosome loss frequency is measured by the proportion of Can^R Ade⁻ cells in a colony. Allelic recombinants are those that form Can^R Ade⁺ colonies and are round in appearance. Mixed colonies are those that are Can^R Ade⁺ and have a sectored appearance (see mixed colony in Figure 1D; materials and methods).

Each mixed colony consists of cells with multiple genotypes (Figure 1D). We showed previously, using a genetic test, that mixed colonies arise from cells with dicentrics (Paek et al. 2009). At least one dicentric that is associated with mixed colonies is formed by the fusion of one set of nearby inverted repeats on Chr VII (Figure 1E; S2 and S3 repeats in the 403 fragile site; Paek et al. 2009). S2 and S3 are fragments of the ends of retrotransposons (LTR fragments). The S2/S3 fusion event leads to the formation of an unstable dicentric chromosome. The dicentric can be resolved by a chromosome loss event, allelic recombination, or a nonreciprocal translocation termed the 403–535 translocation (Figure 1F; Admire et al. 2006). We believe that any of several inverted repeats may fuse to form dicentrics, though only the S2–S3 dicentric has been positively identified, and the frequency of S2–S3 dicentrics correlates with the frequency of mixed colonies (Paek et al. 2009).

For each mutant strain we measured the frequency of chromosome loss, allelic recombination, and dicentric chromosome formation. We test every gene in two separate mutant isolates, and the results from pairs of strains were in all cases concordant. In general we can infer the frequency of dicentric formation from the frequency of mixed colonies, because in most mutants the frequency of mixed colonies does correlate with the formation of a specific dicentric chromosome (Paek et al. 2009). Formation of a mixed colony from a cell containing a dicentric chromosome requires the resolution of the dicentric (Figure 1, D–F), and it is therefore possible that mutants that are defective for mixed colony formation still form dicentrics but cannot resolve them. Therefore, for some mutants we also quantified directly the frequency of the specific dicentric chromosome, present in cells before selection, using a quantitative PCR technique (Figure 1E; material and methods). In addition we determined whether the 403–535 translocation was present, in mixed colonies, by a qualitative PCR technique (Figure 1F); the translocation most likely arises from recombination of the specific dicentric chromosome. In all cases thus far, mutants that form the translocation in fact form the dicentric. In all mutants reported here, strains that formed the specific dicentric also formed the specific translocation. We therefore focus here on mechanisms involved in inverted repeat fusion to form dicentrics. (We previously reported a single mutant, rad59, that formed dicentrics but failed to form the translocation; Paek et al. 2009).

In our analyses, we first determine how fusion events are normally prevented by measuring, in single mutants, the frequency of mixed colonies and dicentrics (of GENEX, say). If a geneX mutant has an increased frequency of instability (of specific dicentric or mixed colonies), then we infer that GENEX normally prevents fusion events. If the single mutants are stable, we conclude that GENEX is not needed to prevent instability (though GENEX may prevent instability by a mechanism redundant with another gene). Finally, to test whether GENEX has a role in forming rearrangements, we tested GENEX function in a rad9 checkpoint mutant background, which has a high level of instability, and determined whether events still occurred in the rad9 geneX double mutant (Admire et al. 2006). If events still occur in the double mutant, we conclude that GENEX has no detectable role in forming the rearrangements, while if events do not occur (or occur to a lesser extent), we conclude GENEX has a role in forming the rearrangement.

The error-free branch of postreplication repair, but not translesion synthesis, prevents the fusion of nearby inverted repeats:

The PRR pathway is one of the major mechanisms the cell uses to bypass lesions during DNA replication (Lee and Myung 2008). There are at least two ways in which the PRR pathway is thought to bypass barriers to replication forks (Figure 2A). These separate functions are commonly referred to as translesion synthesis and error-free bypass, and the choice between these two pathways is tightly regulated by the ubiquitylation of PCNA by Rad6, Rad18, Ubc13, and Mms2 (Hoege et al. 2002; Stelter and Ulrich 2003; Lee and Myung 2008). When PCNA is monoubiquitylated by a Rad6–Rad18 heterodimer, translesion DNA polymerases such as Rev3 and Rad30 are recruited to the lesion sites and are able to replicate past these lesions often in an error-prone manner, leading to frequent point mutations (Figure 2A; Stelter and Ulrich 2003). In contrast, the polyubiquitylation of PCNA by Rad5, Ubc13, and Mms2 is thought to be required for error-free bypass, a poorly understood pathway that may involve template switching, and does not lead to an increase in point mutations (Stelter and Ulrich 2003; Branzei et al. 2008).

Figure 2.—

Model of the postreplication repair pathway and regulation of Srs2. (A) Ubc9 and Siz1 sumoylate PCNA at lysines 164 and 127. SUMOylation recruits Srs2 to PCNA and prevents Rad51 from accessing forks. In addition, PCNA is monoubiquitylated by a Rad6–Rad18 heterodimer at lysine 164, which recruits translesion polymerases (Rev3 and Rad30) to sites of damage. Rad5, Ubc13, and Mms2 polyubiquitylate PCNA to allow for error-free bypass. (B) Srs2 inhibits Rad51; Rad51 promotes allelic recombination and prevents mixed colonies from forming. Thus srs2 mutants show an increase in allelic recombinants and a decrease in mixed colony formation.

We previously showed that PRR-deficient rad18 mutants have an increased frequency of mixed colony formation (Paek et al. 2009). To determine whether the effect we see in rad18 mutants was due to translesion synthesis and/or error-free bypass, we analyzed mutants that are reported to be involved in either branch but not both (Stelter and Ulrich 2003). We found that mutants in the translesion synthesis pathway, such as rad30 and rev7, had no effect on the frequency of mixed colony formation, either in cells with a RAD9 defect or in cells with functional RAD9 (Table 1). Translesion polymerases therefore do not cause instability when defective and play no detectable role in rearrangement events themselves. In contrast, we found that rad5 and ubc13 single mutants both showed a dramatic increase in mixed colonies (Table 1). (The phenotype of these mutants with a rad9 mutation is addressed below.) Taken together these data imply that the error-free bypass branch of PRR (regulated by Rad5 and Ubc13), but not the translesion synthesis branch, acts to prevent the fusion of nearby inverted repeats. These data also suggest that neither the error-prone polymerases nor the error-free proteins are involved in the proposed faulty template switch events that form dicentrics.

TABLE 1.

Mutants in the postreplication repair pathway

	Genotype	Mixed colonies (×10−5)	Translocation	Chromosome loss (×10−5)	Allelic recomb. (×10−5)
Master regulators	RAD+ (wild type)	2.3 ± 1.3 (1.0)	6/6a	8.5 ± 5.8 (1.0)	9.0 ± 3.9 (1.0)
	rad18Δc	84 ± 33 (37)***	1/6	110 ± 150 (13)**	7.8 ± 8.9 (.87)
	pol30-K164R	52 ± 13 (23)***	6/6	140 ± 63 (16)***	42 ± 23 (4.7)***
Error-free bypass	rad5Δ	61 ± 20 (27)***	1/6	520 ± 530 (61)***	43 ± 26 (4.8)**
	ubc13Δ	13 ± 5.3 (5.7)***	5/6	60 ± 37 (7.1)***	19 ± 14 (2.1)*
Translesion synthesis	rev7Δ	2.8 ± 2.0 (1.2)	6/6	23 ± 24 (2.7)*	17 ± 11 (1.9)*
	rad30Δ	3.2 ± 0.78 (1.4)	6/6	27 ± 25 (3.2)**	7.9 ± 4.9 (0.88)
HR regulators	pol30-K127R	1.6 ± 1.0 (0.70)	6/6	28 ± 35 (3.3)**	10 ± 11 (1.1)
	siz1Δ	13 ± 9 (5.7)**	6/6	63 ± 55 (7.4)***	16 ± 6.6 (1.8)*
	srs2Δc	3.5 ± 1.7 (1.5)	0/6	240 ± 190 (28)***	190 ± 92 (21)***
	rad9Δ	49 ± 22 (21)***	13/14b	280 ± 200 (33)***	11 ± 9 (1.2)
Double mutants	rad9Δ rad18Δc	290 ± 170 (126)***	11/12	1620 ± 1270 (190)**	16 ± 12 (1.8)
	rad9Δ pol30-K164R	310 ± 110 (130)***	5/6	2900 ± 1200 (340)***	38 ± 10 (4.2)***
	pol30-K164R rad5Δ	48 ± 12 (21)	4/6	110 ± 52 (13)	40 ± 11 (4.4)***
	pol30-K164R ubc13Δ	42 ± 14 (18)	6/6	97 ± 54 (0.71)	35 ± 17 (3.9)***
	rad9Δ rad5Δ	110 ± 40 (48)*	1/6	590 ± 450 (69)*	2.1 ± 1.7 (0.23)*
	rad9Δ ubc13Δ	740 ± 540 (320)***	3/6	6700 ± 4800 (790)***	3.2 ± 6.7 (0.36)**
	rad9Δ rev7Δ	56 ± 24 (24)	6/6	230 ± 110 (27)	5.4 ± 3.7 (0.6)
	rad9Δ rad30Δ	55 ± 31 (24)	5/6	180 ± 60 (21)	9.6 ± 6.9 (1.1)
	rad9Δ pol30-K127R	43 ± 12 (19)	6/6	410 ± 350 (48)	14 ± 9.2 (1.6)
	rad9Δ siz1Δ	330 ± 260 (140)***	5/6	110 ± 120 (13)**	34 ± 31 (3.8)*
	rad9Δ srs2Δc	11 ± 5.1 (5.0)**	0/6	480 ± 250 (56)	270 ± 140 (30)***

Recomb., recombination. Statistically significant in boldface type. Kruskal–Wallis test, *P < 0.05, **P < 0.01, ***P < 0.001. Single mutants to RAD⁺ or double mutants to rad9 fold change. Frequencies were normalized to RAD⁺.

^aTranslocation for Rad⁺ was previously reported (Admire et al. 2006).

^bTranslocation for rad9 was previously reported (Admire et al. 2006).

^crad18 data was previously reported (Paek et al. 2009).

Mutants in the error-free branch of postreplication repair show different levels of instability:

Both Rad5 and Ubc13 are required to polyubiquitylate PCNA at lysine 164 to promote error-free bypass (Figure 2A; Hoege et al. 2002). Thus it was puzzling as to why rad5 mutants formed mixed colonies much more frequently than did ubc13 mutants (61 × 10⁵ and 13 × 10⁵, respectively; Table 1). The different phenotypes seen in these single mutants might be explained by either of two separate models. First Rad5 may act in both PRR and in a pathway separate from PRR, and both pathways act independently to prevent inverted repeat fusion. There is evidence for this, as Rad5 is required for ubiquitylation of lysine 107 of PCNA, while Ubc13 is not (Das-Bradoo et al. 2010). Alternatively, the suppression of mixed colonies by Rad5 and Ubc13 might be exclusively through error-free bypass, yet Rad5 is essential for PRR and Ubc13 plays some lesser role.

To formally distinguish between these two models, we made use of an allele of PCNA (pol30-K164R), which is unable to be ubiquitylated on this residue and therefore is proposed to be defective for all branches of PRR (Figure 2A; Hoege et al. 2002; Stelter and Ulrich 2003). If the first model is correct, where Rad5 acts in PRR and in a pathway separate from PRR, then a pol30-K164R rad5 double mutant should form mixed colonies more frequently than a pol30-K164R single mutant. In contrast, if the second model is correct and both RAD5 and UBC13 are suppressing mixed colonies through PRR, then pol30-K164R, rad5pol30-K164R, and ubc13pol30-K164R should show the same frequencies of mixed colony formation. This was indeed the case; POL30 is epistatic to both RAD5 and UBC13 (Table 1). Together these data suggest that Pol30, Rad5, and Rad18 all act through one pathway, that Rad5 and Rad18 have greater roles in PRR than does Ubc13, and that all these activities suppress inverted repeat fusion. We suggest that Ubc13 may simply have a lesser role in ubiquitylation than Rad5 and Rad18 (see discussion).

We also found that rad9rad18, rad9ubc13, and rad9pol30-K164R double mutants showed a synergistic (more than additive) increase in both mixed colony formation and chromosome loss when compared to corresponding single mutants (Table 1). In contrast, rad9rad5 mutants have approximately additive levels of mixed colony formation (Table 1). It is unclear why the first three genes have different genetic interactions with RAD9 than does RAD5.

The role of Srs2, Siz1, and PCNA SUMOylation in allelic recombination and inverted repeat fusion (mixed colony formation):

In a previous study, we provided evidence for a genetic interaction between SRS2 and RAD51 in mediating instability of Chr VII (Paek et al. 2009). Srs2 is a DNA helicase that disrupts the Rad51 presynaptic filament (Klein 2001; Krejci et al. 2003; Veaute et al. 2003). Previously, we showed that a srs2 mutant suppressed the high frequency of mixed colony formation (shown in a rad9 background; Paek et al. 2009). Mutants in SRS2 also increased the frequency of allelic recombinants, as expected from previous studies (Table 1; Paek et al. 2009). We further showed that rad51 mutants have a high frequency of instability, suggesting that srs2 mutants decrease instability by increasing Rad51 activity. In accordance with this model, rad51 mutants are epistatic with srs2 mutants, that is, both srs2rad51 double mutants and rad9srs2rad51 triple mutants have the same high frequency of mixed colony formation as rad51 and rad9rad51 mutants, respectively (Figure 2B; Paek et al. 2009).

We therefore decided to examine regulators that are known to permit Srs2 association with the replication fork. Srs2 is recruited to the replication fork in part via Siz1-dependent SUMOylation (small ubiquitin-related modifier) of PCNA (Figure 2A; Papouli et al. 2005; Pfander et al. 2005). During S phase, PCNA (Pol30) is SUMOylated by Siz1 at lysines 127 and 164 (Hoege et al. 2002). SUMOylated PCNA then recruits Srs2 to replication forks, which prevents Rad51 from acting at these forks (Papouli et al. 2005). The absence of Rad51 at the fork may prevent ssDNA from engaging in other reactions (i.e., allelic and nonallelic recombination) and may give PRR proteins unfettered access to regulate stalled forks. We therefore tested the instability of a siz1 mutant, expecting a siz1 mutant would fail to recruit Srs2 to forks, and thus siz1 and srs2 would have similar instability phenotypes. Unexpectedly, we found that siz1 mutants showed more instability than wild-type cells, not less as in srs2 mutants, and there was a dramatic synergistic increase in rad9siz1 double mutants, not decrease as in rad9srs2 double mutants (Table 1). We also found that siz1 mutants had only a modest increase in allelic recombinants, not 20-fold more as in srs2 mutants (Table 1). Similar results (that siz1 and srs2 have different rates of homologous recombination) have been reported previously (Papouli et al. 2005).

In addition to analyzing siz1 mutants, we also assessed the role of PCNA SUMOylation directly. To do this, we utilized pol30-K127R and pol30-K164R mutants (discussed above), which are defective for PCNA SUMOylation and thus are defective for Srs2 recruitment (Figure 2A; Hoege et al. 2002; Pfander et al. 2005). Since both residues may be SUMOylated, neither allele is completely deficient for SUMOylation of PCNA. It was previously shown that the pol30-K164R allele is severely defective for Srs2 recruitment, while the pol30-K127R allele has a milder phenotype (Pfander et al. 2005). The pol30-K127R allele did not show any instability, while the pol30-K164R allele showed a high frequency of instability (Table 1). Furthermore, pol30-K127R mutants did not alter the phenotype of rad9 single mutants, while pol30-K164R synergized with instability of rad9 single mutants (Table 1). This suggests that Pol30 K164 modification, by SUMOylation or ubiquitylation, is critical for preventing instability. We favor the idea that Pol30 K164 ubiquitylation is more critical for preventing instability than SUMOylation, as rad5 mutants (that can SUMOylate PCNA, yet cannot polyubiquitylate PCNA) have a high-instability phenotype similar to pol30-K164R mutants (deficient for SUMOylation and ubiquitylation of PCNA; Table1).

The fact that siz1 and pol30-K164R (high instability, deficient for Srs2 recruitment to forks) mutants have such different phenotypes than srs2 mutants (low instability, deficient for all Srs2 activity) may suggest separate roles of Srs2 in S phase vs. other cell cycle stages. In the discussion we speculate on what these separate roles might be.

Analysis of DNA polymerase-associated mutants for roles in faulty template switching:

From our previous studies, we proposed that nearby inverted repeat fusion occurs by a faulty template switch reaction (Figure 1; Paek et al. 2009). To date we know of few mutants defective for this reaction (Paek et al. 2009); most mutants we have tested either have no phenotype or have an increase in instability (Admire et al. 2006; Paek et al. 2009). In an attempt to identify additional mutants defective in the faulty template switch reaction, we turned our attention to genes that are involved in DNA replication itself.

We first examined the role of nonessential subunits of DNA polymerase epsilon (DPB3, DPB4 leading strand polymerase; Araki et al. 1991; Ohya et al. 2000). Deletion of either of these subunits leads to an approximately sixfold increase in the formation of mixed colonies in a Rad⁺ background (compare Rad⁺ to dpb3, dpb4) and a further twofold increase in a rad9 background (compare rad9 to rad9dpb3, rad9dpb4; Table 2). A similar effect was seen previously in pol32 mutants, a nonessential subunit of DNA polymerase delta (Paek et al. 2009). Thus, deleting nonessential subunits of DNA polymerase epsilon (dpb3, dpb4) and delta (pol32) increased instability. So disrupting DNA replication increases inverted repeat fusion, yet these mutants do not shed light on how events occur.

TABLE 2.

Polymerase mutants

Genotype	Mixed colonies (×10−5)	Trans location	Dicentricc (×10−5)	Chromosome loss (×10−5)	Allelic recomb. (×10−5)
RAD+ (wild type)	2.3 ± 1.3 (1.0)	6/6a	5.7 ± 2.0 (1.0)	8.5 ± 5.8 (1.0)	9.0 ± 3.9 (1.0)
RAD+ 23°	5.1 ± 2.0 (2.2)*	ND	ND	310 ± 330 (36)**	8.9 ± 4.1 (1.0)
RAD+ 36°	2.5 ± 1.1 (1.1)	ND	ND	5.8 ± 2.3 (0.68)	6.5 ± 2.3 (0.72)
mgs1Δ	1.6 ± .88 (0.70)	2/6	(5.6)**	15 ± 20 (1.8)*	11 ± 6.4 (1.2)
dpb3Δ	16 ± 19 (7.0)**	7/7	ND	380 ± 470 (45)**	18 ± 17 (2.0)*
dpb4Δ	17 ± 13 (7.4)**	6/6	ND	300 ± 370 (35)**	51 ± 14 (5.7)*
pol3-13 23°	2.9 ± 2.2 (1.3)	ND	(8.8)**	38 ± 37 (4.5)**	8.2 ± 5.3 (.91)
pol3-13 36°	1.5 ± .44 (.65)	2/6	(2.5)**	120 ± 140 (14)**	5.1 ± 1.3 (.57)**
mgs1Δ pol3-13 23°	4.1 ± 2.1 (1.8)	5/6	ND	25 ± 10 (2.9)**	23 ± 22 (2.6)**
mgs1Δ pol3-13 36°	13 ± 9.8 (5.7)*	10/10	ND	11 ± 5.6 (1.3)	3.2 ± 3.1 (0.36)**
rad9Δ	49 ± 22 (21)***	13/14b	(17)***	280 ± 200 (33)***	11 ± 9 (1.2)
rad9Δ 23°	60 ± 34 (26)	ND	ND	240 ± 160 (28)	9.1 ± 4.7 (1.0)
rad9Δ 36°	38 ± 5.4 (17)	ND	ND	370 ± 260 (44)	4.5 ± 2.6 (0.50)*
rad9Δ mgs1Δ	20 ± 11 (8.7)***	2/6	(6.1)**	160 ± 180 (19)	14 ± 8.2 (1.6)
rad9Δ dpb3Δ	72 ± 34 (31)*	6/6	ND	1800 ± 1500 (210)***	18 ± 15 (2.0)*
rad9Δ dpb4Δ	94 ± 56 (41)*	6/6	ND	1200 ± 940 (140)***	21 ± 14 (2.3)*
rad9Δ pol3-13 23°	25 ± 8.6 (11)***	ND	(9.7)	270 ± 260 (32)	3.8 ± 2.1 (0.42)**
rad9Δ pol3-13 36°	17 ± 4.8 (7.4)***	2/6	(5.0)**	330 ± 170 (39)	3.5 ± 1.6 (0.39)**
rad9Δ pol3-13 mgs1 23°	26 ± 7.2 (11)***	4/6	(5.6)**	280 ± 270 (33)	8.7 ± 6.9 (0.97)
rad9Δ pol3-13 mgs1 36°	1.8 ± 3.7 (0.80)***	0/6	(3.5)**	950 ± 1500 (110)*	.04 ± .16 (0.004)***
rad9Δ pol3-13 mgs1 23°–36°	0.94 ± 1.5 (0.41)***	ND	ND	170 ± 260 (20)*	.05 ± .13 (0.006)***
rad9Δ pol3-13 mgs1 36°–23°	57 ± 20 (25)	ND	ND	728 ± 420 (86)*	7.7 ± 3.9 (0.86)

Recomb., recombination. Statistically significant in boldface type. Kruskal–Wallis test, *P < 0.05, **P < 0.01, ***P < 0.001. Single mutants to RAD⁺ or double mutants to rad9 fold change. Frequencies were normalized to RAD⁺.

^aTranslocation for Rad⁺ was previously reported (Admire et al. 2006).

^bTranslocation for rad9 was previously reported (Admire et al. 2006).

^cDicentric frequencies were determined by PCR and normalized to Rad⁺ frequency. Actual frequencies are: Rad⁺ 5.7 + 2.0 × 10⁻⁵, mgs1 32 ± 9 × 10⁻⁵, pol3-13 23° 50 ± 13 × 10⁻⁵, pol3-13 36° 14 ± 7 × 10⁻⁵, rad9 98 ± 66 × 10⁻⁵, rad9 mgs1 35 ± 6 × 10⁻⁵, rad9 pol3-13 23° 56 ± 8 × 10⁻⁵, rad9 pol3-13 36° 29 ± 6 × 10⁻⁵, rad9 pol3-13 mgs1 23° 56 ± 11 × 10⁻⁵, rad9 pol3-13 mgs1 36° 20 ± 2 × 10⁻⁵.

We then turned to analyzing DNA polymerase delta (Pol3) itself, using a temperature-sensitive allele (pol3-13) and Mgs1, a single-strand binding protein that has a strong genetic interaction with Pol3 (Giot et al. 1997; Hishida et al. 2001, 2002; Vijeh Motlagh et al. 2006). We studied each single mutant, pol3-13 mgs1 double mutants, as well as each single and the double mutant in combination with a highly unstable rad9 mutant (Table 2). The strongest phenotype is observed in the rad9pol3-13 mgs1 triple mutant. We found that, relative to unstable rad9 single mutants, rad9pol3-13 mgs1 mutants grown at the semirestrictive temperature (36°) had a 20-fold decrease in mixed colonies, a 5-fold decrease in dicentric chromosome formation, and a 200-fold decrease in allelic recombinants (Table 2; compare rad9 36° to rad9pol3-13 mgs1 36°).

The strong phenotype of the triple mutant suggests Pol3 and Mgs1 have roles in forming rearrangements. Analyses of single and double mutants are largely consistent with this view (Table 2). Both pol3-13 and mgs1 single mutants had an increase in the formation of dicentric chromosomes (compared to Rad⁺), yet rad9pol3-13 and rad9mgs1 strains had a modest two- to threefold decrease in both dicentric chromosome and mixed colony formation (Table 2; compared to rad9). In the discussion, we speculate as to why the single mutants show an increase in dicentric chromosomes, while the double and triple mutants show a decrease in these events.

The triple mutant is not only defective in the formation of dicentrics, but also in resolving them, as the decrease in mixed colony formation (20-fold) is greater than the decrease in dicentrics (5-fold). We propose, therefore, that the triple mutant also has a defect in resolving dicentric chromosomes, which is required to form a mixed colony (Table 2). We suggest that, in particular, the triple mutant has a profound defect in completing normal allelic recombination (Table 2; Figure 3). In sum, the triple mutant fails to form allelic recombinants either from a dicentric intermediate (which occurs in mixed colonies) or through a DSB formed independently of dicentrics (200-fold decrease in allelic recombinants; Table 2). The defect in allelic recombination in the triple mutant is addressed in discussion.

Figure 3.—

Model for defects in rad9 pol3-13 mgs1 strains at the restrictive temperature. (1) When a double-strand break occurs on a dicentric or monocentric chromosome, (2) the ends are resected and (3) strand invasion and second end capture occur. (4) This is followed by Pol Delta and Mgs1-dependent polymerization, which by this model proceeds slowly in rad9 pol3-13 mgs1 mutants. (5a) In a RAD9 cell, this signals a checkpoint response, which prevents chromosome segregation. However in a rad9 cell mitosis proceeds, which leads to chromosome missegregation. (5b) A normal cell is able to form and (6) resolve a Holliday junction, leading to allelic recombinants.

The model that the triple mutant fails to complete allelic recombination makes a simple and testable prediction. We suggest that rad9pol3-13 mgs1 mutants may start to make allelic recombinants at 36°, but cannot resolve them at 36°, yet can resolve them at 23°. In support of this hypothesis, we found that cells grown in rich medium at 36° but plated to selective medium at 23° readily form allelic recombinants, while cells grown in rich medium at 23° and plated to selective medium at 36° could not (Table 2; rad9pol3-13 mgs1 36°–23°, rad9pol3-13 mgs1 23°–36°). Formation of mixed colonies followed the same trend; when grown at 36° in rich medium and plated to selective medium at 23°, mixed colonies formed, but cells grown at 23° in rich medium and plated to selective medium at 36° did not form mixed colonies (Table 2; rad9pol3-13 mgs1 36°–23°, rad9pol3-13 mgs1 23°–36°). Therefore early events in formation of either allelic recombinants or mixed colonies (dicentrics) can form at 36°, but their resolution requires full Pol3 function (see discussion and supporting information, Figure S1).

Additional genes involved in chromosomal instability:

In separate genome instability assays developed by Kolodner and colleagues, fusion events occur between sequences with either limited homology (5–9 bp; Chen and Kolodner 1999; Myung et al. 2001; Kolodner et al. 2002; Putnam et al. 2005) or with larger regions of homology (∼4.2 kb; Putnam et al. 2009, 2010). They refer to these rearrangements as single-copy and duplication-mediated rearrangements, respectively, and interestingly, they found that certain genes were required to prevent duplication-mediated rearrangements and not single-copy rearrangements, and vice versa (Putnam et al. 2009, 2010). In the reactions that we study, inverted repeat fusions seem in general to require more homology than single-copy reactions (though see discussion) and less homology than duplication-mediated events (Paek et al. 2009; Putnam et al. 2009). To better understand the mechanisms underlying inverted repeat fusion, we next analyzed a subset of genes that affect single-copy and/or duplication-mediated rearrangements for their roles in inverted repeat fusion.

First we analyzed RAD27, the ortholog of human FEN-1, a 5′ flap endonuclease (Kolodner and Marsischky 1999). Rad27 is required for Okazaki fragment processing and maturation during replication. Rad27 also has roles in base-excision repair and NHEJ (Wu et al. 1999; Liu et al. 2004). Previously it was found that rad27 mutants have an ∼1000-fold increase in single-copy GCRs and a 140-fold increase in duplication-mediated GCRs (Chen and Kolodner 1999; Putnam et al. 2009). In our assay, although rad27 mutants have an increased frequency of mixed colony formation, the increase was far less dramatic (∼10-fold; Table 3); thus Rad27 has a more dramatic role in preventing both single-copy and duplication-mediated events than in preventing inverted repeat fusion. (rad27 mutants also had a significant increase in allelic recombinants and chromosome loss, similar to what others have seen; Vallen and Cross 1995.)

TABLE 3.

Frequencies, translocation, and dicentric analysis of chromosome stability mutants

	Genotype	Mixed colonies (×10−5)	Translocation	Chromosome loss (×10−5)	Allelic recomb. (×10−5)
Single mutants	RAD+ (wild type)	2.3 ± 1.3 (1.0)	6/6a	8.5 ± 5.8 (1.0)	9.0 ± 3.9 (1.0)
	rad27Δ	24 ± 19 (10)***	ND	260 ± 230 (31)***	400 ± 480 (44)***
	rtt101Δ	60 ± 14 (26)***	6/6	140 ± 92 (16)**	29 ± 34 (3.2)*
	rtt107Δ	170 ± 41 (74)***	6/6	690 ± 270 (81)***	22 ± 7.8 (2.4)*
	slx5Δ	20 ± 6.9 (8.7)***	ND	310 ± 400 (36)***	25 ± 16 (2.8)*
	yen1Δ	6.0 ± 2.6 (2.6)**	1/6	32 ± 18 (3.8)**	7.9 ± 3.1 (.88)
RAD9 double mutants	rad9Δ	49 ± 22 (21)***	13/14b	280 ± 200 (33)***	11 ± 9 (1.2)
	rad9Δ rtt101Δ	310 ± 140 (130)***	6/6	4700 ± 1300 (550)***	47 ± 23 (5.2)***
	rad9Δ rtt107Δ	250 ± 47 (110)***	6/6	2700 ± 1400 (320)***	24 ± 16 (2.7)**
	rad9Δ slx5Δ	63 ± 12 (27)	ND	2400 ± 2200 (280)***	28 ± 11 (3.1)**
	rad9Δ yen1Δ	54 ± 13 (23)	3/6	480 ± 380 (56)	8.2 ± 2.7 (.91)

Recomb., recombination. Statistically significant in boldface type. *P < 0.05, Kruskal–Wallis test, **P < 0.01, ***P < 0.001. Single mutants to RAD⁺ or double mutants to rad9 fold change. Frequencies were normalized to RAD⁺.

^aTranslocation for Rad⁺ was previously reported (Rad+ et al. 2006).

^bTranslocation for rad9 was previously reported (Admire et al. 2006).

The Slx5–Slx8 complex is a SUMO-dependent ubiquitin ligase complex thought to bind and ubiquitylate SUMOylated proteins and has been shown to have a role in the DNA damage response (Ii et al. 2007; Rouse 2009). Interestingly, slx5 mutants were shown to have a high rate of duplication-mediated GCRs (25-fold increase) but not single-copy GCRs (Putnam et al. 2009). Consistent with the role of Slx5 in preventing repeat-mediated rearrangements, we saw an increase in mixed colony formation in slx5 single mutants compared to wild type (Table 3), yet again, as with rad27 mutants, the increase in mixed colony formation in slx5 mutants in our assay was less dramatic (9-fold increase) than in the duplication-mediated assay (25-fold increase; Table 3; Putnam et al. 2009). (slx5 strains also showed a significant increase in chromosome loss and allelic recombinants in our assay. The rad9 slx5 double mutant displayed a synergistic increase in chromosome loss compared to either single; rad9slx5, 2350 × 10⁻5; rad9, 273 × 10⁻5; slx5, 313 × 10⁻5; Table 3).

Next, we examined the Cullin Rtt101 protein that is involved in replication through natural and MMS-induced pause sites and has been shown to increase the frequency of single-copy GCRs (to our knowledge it has not yet been tested in the duplication-mediated GCR assay; Luke et al. 2006). Rtt101 is a multisubunit E3 ubiquitin ligase that, together with Mms1 and Mms22, aids in the replication of damaged DNA (Zaidi et al. 2008). At least one way Rtt101 is thought to do this is through the recruitment to stalled forks of Rtt107, a BRCT domain protein implicated in replication restart following DNA damage (Rouse 2004; Chin et al. 2006). Rtt101 and Rtt107 likely act independently as well, because rtt101 rtt107 double mutants have a higher degree of MMS sensitivity than either single (Roberts et al. 2008; Zaidi et al. 2008). Interestingly we see a very high frequency of mixed colony formation in both rtt101 and rtt107 single mutants, as well as in rad9 rtt101 and rad9 rtt107 doubles (Table 3). rtt107 single mutants had approximately three times more mixed colonies and five times more chromosome loss than rtt101 single mutants. Taken together, these data imply that both Rtt101 and Rtt107 act to prevent nearby inverted repeat fusion and chromosome instability, and that Rtt107 has a greater role than does Rtt101.

Finally, we analyzed yen1 mutants. Yen1 acts as a Holliday junction resolvase, though there are several other proteins that also have resolvase activity (Ip et al. 2008; Svendsen and Harper 2010). To our knowledge yen1 mutants have not been analyzed by either the single-copy or duplication-mediated GCR assays; however, we feel this is an important mutant as one version of inverted repeat fusion occurs by a potential Yen1-dependent resolution of a regressed fork (chicken foot; Paek et al. 2009). yen1 mutants have a very modest to negligible phenotype in our instability assay, arguing against a role for regressed forks (Table 3).

DISCUSSION

Postreplication repair:

The postreplication repair pathway is thought to be one of the predominant mechanisms that cells utilize to bypass lesions on the template strand during DNA replication (Zhang and Lawrence 2005; Lee and Myung 2008). Using mutants that are deficient for translesion synthesis (rad30 and rev7), we have shown that this subpathway of PRR does not prevent the fusion of inverted repeats or do the error-prone polymerases appear to have a role in rearrangements (Table 1). In contrast, mutants in the error-free bypass branch (rad5 and to a lesser extent ubc13) all increased the frequency of mixed colonies. Indeed, rad5 mutants had a phenotype strikingly similar to that of pol30-K164R mutants that are deficient for both branches of PRR (Table 1). Thus it is likely that the chromosome instability phenotypes that we see in these mutants are solely due to a defect in the error-free bypass branch of postreplication repair. We propose that PRR prevents fusion of inverted repeats by promoting a template switch of the stalled nascent strand to the correct sequence on its sister chromatid, forming a hemicatenane as proposed by others (Figure 4, IIA and IIB; Branzei et al. 2008). PRR may thus prevent a “faulty” template switch event to the nonallelic inverted repeat (Figure 4).

Figure 4.—

Model of faulty template switching events, red arrows indicate inverted repeats, the blue asterisk indicates a “lesion” in the DNA. The model is drawn for a leading strand stall, but events could initiate on lagging strand instead. (IA) When a replication fork stalls at a lesion, (IB) Srs2 promotes replication restart downstream of the stall site, by preventing Rad51 from acting. A faulty template switch event is also possible at this step (see Figure 1D). (IIA) In G2 Rad18 and other PRR proteins can promote a “normal” template switch to the nascent strand of the sister chromatid, which (IIB) forms a hemicatenane. Alternatively, after resection of DNA by a nuclease, (IIIA) a faulty template switch event initiates when Mgs1 anneals the stalled nascent strand to the nonallelic repeat sequence on the template strand. (IIIB) Pol Delta and Mgs1-dependent DNA synthesis leads to the formation of a dicentric chromosome. (IVA) In srs2 cells Rad51 gains access to ssDNA and promotes strand invasion and second end capture. (IVB) Resolution leads to formation of an allelic recombinant.

Individual mutants in the error-free branch of PRR had different levels of inverted repeat fusion, implying that the wild-type proteins play slightly different roles (Table 1). That is, rad5 mutants had a much higher level of mixed colony formation than ubc13 mutants, though both mutants increased mixed colony formation to some degree. A similar trend was seen when these mutants were analyzed for single-copy GCR formation, and also for MMS sensitivity (Xiao et al. 2000; Motegi et al. 2006; Putnam et al. 2010). Since pol30-K164R mutants are epistatic to rad5 and ubc13 for mixed colony formation, this implies that mixed colonies in these mutants are due to aberrant PRR and not some other (PRR-independent) pathway that these genes might be involved in (Table 1; Figure 2). How can we explain the observation that instability is higher in rad5 mutants than in ubc13 mutants? Both Rad5 and Ubc13 have been shown to be required for polyubiquitylating PCNA, however the polyubiquitylation assays were not quantitative; it may be that Rad5 is required for full polyubiquitylation, while Ubc13 plays a lesser role (Hoege et al. 2002). Thus it could be that in ubc13 mutants, PCNA is polyubiquitylated to some extent, allowing a greater amount of PRR function than present in rad5 or rad18 mutants (Hoege et al. 2002).

Another interesting facet in PRR regulation is the SUMOylation of PCNA at lysine 127 and lysine 164 by the Siz1 SUMO ligase (Hoege et al. 2002; Stelter and Ulrich 2003; Papouli et al. 2005; Pfander et al. 2005). PCNA SUMOylation recruits the Srs2 helicase that removes Rad51 from DNA and thus may provide access to proteins in the PRR pathway (Papouli et al. 2005; Pfander et al. 2005). We found that mutants that are deficient for Srs2 recruitment to replication forks (siz1 and pol30-K164R) had a high level of instability, whereas srs2 mutants had a low level of instability (Table 1; Paek et al. 2009). How can we explain that srs2 and siz1 mutants have opposite phenotypes, if Siz1-dependent PCNA SUMOylation is required for Srs2 recruitment to a fork? We suggest that when Srs2 is recruited to stalled replication forks via Siz1-dependent PCNA SUMOylation, Srs2 prevents improper recombination events between sisters (by blocking Rad51 assembly). We suggest, however, that Srs2 may also prevent recombination between sisters in G2 (also by blocking Rad51 assembly), independent of the replication fork. And, Siz1 may promote PRR independent of Srs2 function (by a mechanism that is not yet known). Therefore, in a siz1 mutant PRR fails to some extent, and some inverted repeat fusion ensues in G2 (Figure 4). In srs2 mutants, replication forks may fail, yet when cells reach G2 the lesions are repaired by Rad51-dependent homologous recombination and not faulty template switching (Figure 4). In support of this model, it has been shown that PCNA is SUMOylated mainly in S phase, while ubiquitylation mainly occurs in G2 (Hoege et al. 2002). In addition, it has been shown recently that PRR can operate efficiently in the G2/M phase of the cell cycle, and PRR's predominant role might be at single-stranded gaps left after replication fork restart (Daigaku et al. 2010; Karras and Jentsch 2010).

Polymerase Delta and Mgs1 promote fusion events in rad9 mutants, and Rad9, Pol Delta, and Mgs1 are required for the propagation of allelic recombinants and cells with dicentric chromosomes:

Surprisingly we found that rad9pol3-13 mgs1 mutants grown at the restrictive temperature showed a dramatic 20-fold decrease in the formation of mixed colonies and a 200-fold decrease in allelic recombinants (Table 2). We propose that the decrease in mixed colony formation in rad9pol3-13 mgs1 cells is due to a defect both in the formation of dicentric chromosomes (dicentrics down 5-fold relative to rad9 single mutants) and in an inability to properly resolve dicentrics by allelic recombination once dicentrics are formed (Figures 3 and 4).

Both mgs1 and pol3-13 mutants decreased the frequency of dicentric chromosome formation in a rad9 mutant background, yet these mutations caused an increase in a wild-type background (Table 2). We suggest that as single mutants (pol3-13 36°, mgs1), defects in replication lead to an increased number of defective forks, which then lead to dicentrics through the use of the other intact gene products. In double mutants (rad9pol3-13 36°, rad9mgs1), the increase in fork defects (caused by a rad9 mutation) is offset by defects in inverted repeat fusion (by Mgs1 and Pol3 defects) needed to complete events. Finally, in the triple mutant (rad9pol3-13 mgs1 36°), most lesions are not converted into dicentrics, as we see a fivefold decrease in dicentric chromosomes.

There seems to be a synergistic interaction between all three genes, an interaction that occurs to a lesser extent in any double mutant (Table 2). Therefore a plausible molecular explanation requires three interactions, and all three may have some role in replication fork behavior. Mgs1 is a DNA-dependent ATPase with single-strand annealing activities, interacts with pol delta (Hishida et al. 2001; Branzei et al. 2002), and has been proposed to promote replication restart (Vijeh Motlagh et al. 2006). Thus we propose that Mgs1 is partly responsible for annealing of the nascent strand to the nonallelic template strand (in inverted repeat fusion; Figure 4, IIIA). After annealing, pol delta is likely required for replication to the end of the neighboring Okazaki fragment (Figure 4, IIIB) thus fusing the two sisters in inverted repeat fusion or completing structures in allelic recombination (Figure 3, step 4; Maloisel et al. 2008). Mgs1 could act at these polymerization steps as well, as it has been shown that Mgs1 binds to Pol delta, and is required to stabilize temperature-sensitive Pol delta strains (Figure 4, IIIB; Branzei et al. 2002; Vijeh Motlagh et al. 2006). Rad9's role is either in a cell cycle delay (Figure 3, step 5a) or in promoting fork stability (via Chk1; Segurado and Diffley 2008). Thus proper annealing and synthesis, be it to nonallelic or allelic sites, requires the combined action of Mgs1, Pol3, and Rad9.

Although speculative, this model does have additional support. First, it has been suggested that Mgs1 stabilizes replication forks in strains with pol delta defects (Vijeh Motlagh et al. 2006). Second, genetic data supports the idea that pol delta is the major polymerase during mitotic recombination (Giot et al. 1997; Maloisel et al. 2008). Third, temperature shift experiments showed that when the triple mutant is grown in rich medium at the restrictive temperature, then plated to selection medium at the permissive temperature, mixed colony and allelic recombinants occur. We suggest that resolution of dicentrics requires events similar to those required for allelic recombination. A surprising prediction of this model is that most allelic recombinants in the triple mutant are not resolved before cells are subjected to selection, but rather are resolved on selective medium (Table 2). We feel that wild-type cells probably do resolve recombination events before selection and provide a model that explains this phenomenon in rad9pol3-13 mgs1 cells (Figure S1). Finally we note that genes that have defects at other stages of recombination do not show a decrease in mixed colony formation; mutants with defects in the resection (mre11), pairing (rad51 and rad52), and Holliday junction resolution (sgs1, mus81, and yen1) stages of homologous recombination do not show decreases in mixed colony formation (Krogh and Symington 2004; Paek et al. 2009; S. Kaochar and T. Weinert, unpublished data; this study). Thus we infer a specific defect in resolution of allelic recombinants in rad9pol3-13 mgs1 triple mutants that these other single mutants are proficient in.

Relationship of single-copy and repeat-copy rearrangements to inverted repeat fusion:

Finally, to compare inverted repeat fusion and dicentric chromosome formation to rearrangements seen in other GCR assays, we analyzed genes with roles in single-copy or repeat-mediated rearrangements (Table 3). We saw little correlation between the increase in dicentric chromosome formation and either single-copy or duplication-mediated GCRs. For example, rad27 mutants have a 1000-fold increase in single-copy GCR formation, and a 140-fold increase in duplication-mediated GCRs (Chen and Kolodner 1999; Putnam et al. 2009). In contrast, slx5 mutants have only a modest (∼10-fold) increase in inverted repeat fusion events (Table 3). Analysis of PRR mutants also distinguish both single-copy GCRs and duplication-mediated GCRs from inverted repeat fusion events (Motegi et al. 2006; Putnam et al. 2010). This implies that inverted repeat fusions, single-copy, and repeat-sequence rearrangements are likely formed by separate mechanisms. Indeed, duplication-mediated GCRs occur between separate chromosomes, between sequences with 4.2 kb of homology, and appear to be Rad52 dependent (Putnam et al. 2009). In contrast, the GCRs in this study arise from inverted repeat fusion, occur on the same chromosome (likely by sister chromatid fusion) between 300 bp repeats, and occur independently of Rad52 (Admire et al. 2006; Paek et al. 2009).