The Dynamics of Homologous Pairing during Mating Type Interconversion in Budding Yeast

Cells repair most double-strand breaks (DSBs) that arise during replication or by environmental insults through homologous recombination, a high-fidelity process critical for maintenance of genomic integrity. However, neither the detailed mechanism of homologous recombination nor the specific roles of critical components of the recombination machinery—such as Bloom and Werner syndrome proteins—have been resolved. We have taken a novel approach to examining the mechanism of homologous recombination by tracking both a DSB and the template from which it is repaired during the repair process in individual yeast cells. The two loci were labeled with arrays of DNA binding sites and visualized in live cells expressing green fluorescent protein–DNA binding protein chimeras. Following induction of an endonuclease that introduces a DSB next to one of the marked loci, live cells were imaged repeatedly to determine the relative positions of the DSB and the template locus. We found a significant increase in persistent associations between donor and recipient loci following formation of the DSB, demonstrating DSB-induced pairing between donor and template. However, such associations were transient and occurred repeatedly in every cell, a result not predicted from previous studies on populations of cells. Moreover, these associations were absent in sgs1 or srs2 mutants, yeast homologs of the Bloom and Werner syndrome genes, but were enhanced in a rad54 mutant, whose protein product promotes efficient strand exchange in vitro. Our results indicate that a DSB makes multiple and reversible contacts with a template during the repair process, suggesting that repair could involve interactions with multiple templates, potentially creating novel combinations of sequences at the repair site. Our results further suggest that both Sgs1 and Srs2 are required for efficient completion of recombination and that Rad54 may serve to dissociate such interactions. Finally, these results demonstrate that mechanistic insights into recombination not accessible from studies of populations of cells emerge from observations of individual cells.


Introduction
Double-stranded DNA breaks (DSBs) occur spontaneously during replication and by exposure to certain genotoxic chemicals or ionizing radiation. Efficient repair of these DSBs can be accomplished non-conservatively by non-homologous end joining (NHEJ) or with exact fidelity using homologous recombination. Failure to repair these DSBs accurately can perturb the normal cell cycle progression and increase genomic instability, thereby promoting tumorigenesis.
The timing, choreography, and genetic dependencies of steps during DSB repair by homologous recombination in the budding yeast have been examined on populations of cells following synchronous initiation of a DSB [1][2][3]. Moreover, microscopic observation of live cells expressing specific green fluorescent protein (GFP)-tagged proteins has provided information on the temporal sequence and dependencies in recruitment of repair and recombination proteins to nuclear repair foci that form following initiation of DSBs [4]. These studies demonstrated that the repair of DSBs by homologous recombination occurs in several steps. First, shortly after formation of a DSB, signal transduction and nuclease modules recognize and process the DSB. The Mre11, Rad50, Xrs2 (MRX) module in conjunction with the Tel1 kinase tether the ends of the newly formed DSB and promote efficient nucleolytic processing of the break to generate 39 single-stranded ends [4][5][6]. The single-strand binding complex, RPA, binds to the newly exposed single-stranded region. Rad51 then forms a nucleoprotein filament on the exposed single strands at the ends of the DSB. Rad52, Rad54, Rad55, and Rad57 also bind to the exposed single-strand region and promote formation and/or stabilization of the Rad51 nucleoprotein filament, in part by alleviating the inhibitory effect of RPA on Rad51 binding [2,3].
In the second step in DSB repair, the nucleoprotein filament performs a search for homology and associates with a donor sequence. Initial association likely occurs through a triple-strand association between the nucleoprotein filament with the duplex donor DNA, followed by isomerization to form a D-loop with the single-strand end of the DSB invading the duplex to pair with the template strand of the duplex [7]. Rad54 plays a role in strand invasion along with DNA polymerase holoenzyme [2], which extends the invading 39 strand. In yeast, the joint between invading and template strands is resolved 2 to 4 h after DSB formation. However, most of these studies report only on the average behavior of populations of cells. Our results described in this report indicate that the discrete dynamics of the recombination process in individual cells are quite different.
A variety of helicases participate in recombination. One such helicase in Saccharomyces, Sgs1, is a member of the RecQ helicase family that resembles the WRN and BLM helicases defective in Werner and Bloom genome instability syndromes in humans [8]. Deletion of SGS1 increases the frequency of gross chromosomal rearrangements and, in conjunction with deletion of a second related helicase gene, SRS2, causes a severe growth defect in yeast. This growth defect can be suppressed by mutations that inhibit initiation of homologous recombination, suggesting that Sgs1 and Srs2 prevent formation of aberrant, irreparable complexes that could arise during the process of homologous recombination [9]. Other studies have indicated that, despite the synthetic lethality of sgs1 and srs2, the two corresponding proteins act at different steps in recombination, with Sgs1 helping to resolve recombination intermediates and Srs2 functioning to counteract initiation of recombination [10]. Moreover, these proteins appear to play critical roles during replication fork stalling, with Sgs1 promoting resolution of recombinationdependent structures that form at damaged replication forks and Srs2 assisting to prevent the formation of such structures [11]. Finally, while these studies suggest that Sgs1 and Srs2 suppress recombination, sgs1 mutants are defective in DNA damage-induced heteroallelic recombination, suggesting that Sgs1 may promote recombination in certain contexts [9,12]. Consistent with this positive role of the yeast RecQ helicase in recombination, WRN protein is required to complete mitotic recombinants efficiently in unperturbed conditions in fibroblast cell lines [13]. Thus, we still do not have a clear conception of the function of RecQ helicases in recombination and a full appreciation of the mechanism underlying the pathological consequences of losing RecQ function in mammalian cells.
In order to address unresolved aspects of recombination, we have examined mating type switching in individual yeast cells. Mating type switching in yeast has been extensively used as a model system for homologous recombination [2,3,14,15]. Haploid Saccharomyces cells exist in one of two mating types, a or a, dependent solely on the particular allele present at the MAT locus on Chromosome III. Haploid cells can change mating type as often as every generation (reviewed in [16,17]) through recombination initiated by introduction of a DSB at the MAT locus, catalyzed by an endonuclease encoded by HO [18]. Switching then occurs by a gene conversion event, without associated crossover, that replaces the mating information at the MAT locus with the opposite mating information, a or a, present at either of two repository mating loci, HML and HMR, located at the opposite ends of Chromosome III. Mating type switching follows a precise developmental pattern that results from the intricate mode of transcriptional regulation of the HO gene and from a highly regulated interaction between donor and recipient loci [19][20][21]. Specifically, cell type dictates which donor locus is selected for participation in the gene conversion event. MATa cells predominantly select HML, which normally contains a mating information, whereas MATa cells select HMR, which normally contains a mating information. This selection process insures efficient mating type switching. We have created strains for observing the interaction of MAT and its preferred donor locus during mating type switching in single cells in order to assess the discrete dynamics of the pairing process. The results of this analysis, described below, indicate that the dynamics of gene conversion during mating type interconversion are quite different, and much more rapid, in individual cells than would be predicted from previous studies on populations of cells. Furthermore, the dynamics of interaction of donor and recipient loci in various mutant backgrounds suggest unexpected roles for several components of the recombination machinery.

Method for Observing Homologous Pairing in Live Yeast Cells
We developed strains and a method to observe the pairing of homologous DNA loci during recombination in live yeast cells by adopting previously described methods to visualize discrete chromosomal loci and applying them to the process of mating type switching [22,23]. As diagrammed in Figure  1A, we inserted an array of lac repressor binding sites adjacent to a donor locus, HML, and an array of tet repressor sites adjacent to the recipient locus, MAT, and then expressed both GFP-lacI and GFP-tetR fusions in the strain [23,24]. We performed these experiments in a cdc15-2 strain, which arrests at the non-permissive temperature just prior to exit from telophase. By shifting the strain from non-permissive to permissive temperature, cells synchronously enter G1 phase [25]. This insures that, at the beginning of our experiments, all cells have only one marked chromosome and are at the stage of the cell cycle in which switching normally occurs. Finally, the strain also carries the HO gene under control of the GAL10 promoter to allow regulated induction of a DSB at the MAT locus.
To monitor cells under controlled conditions, we affixed yeast cells to the coverslip of a thermally regulated flow cell in

Synopsis
Genetic recombination not only promotes genetic diversity in a population but also insures the integrity of an organism's genome. Inappropriate or inefficient recombination drives tumor formation in cancers and underlies certain premature aging diseases in humans, such as Bloom's and Werner's syndromes. To gain insight into the role of recombination in these diseases, the researchers developed a means of observing recombination in individual cells. They accomplish this by repeatedly monitoring the positions of the two chromosomal segments participating in recombination. Their observations show that interacting strands undergo repeated association and dissociation prior to fully completing recombination. This view contradicts the accepted model of recombination as an uninterrupted, continuous process, but explains, as has been observed, how sequences from different regions of the genome might arrive at a single site after recombination. Moreover, their results show that mutants lacking the Bloom's or Werner's genes have difficulty bringing the interacting strands of DNA together. Such delays could lead to errors in genome structure that could account for the disease characteristics. In short, observing recombination in individual cells reveals unexpected but important features of this process.
which we could control both temperature and medium while repeatedly interrogating by fluorescence microscopy multiple fields of cells on the coverslip. Cells freshly transformed with the P GAL1 -HO plasmid were pregrown in non-inducing medium at 23 8C and then shifted to 35 8C for 3 h. This arrests all the cells just prior to entry into G1 phase of the cell cycle. After attaching cells to the coverslip and centering the flow cell on the microscope, we initiated each experiment by simultaneously inducing the DSB at MAT and releasing cells from cell cycle arrest. This was accomplished by perfusing the flow cell with medium containing 2% galactose to induce HO and shifting the temperature of the flow cell to 23 8C. After DSBs were induced, we shut off HO expression to prevent subsequent rounds of recombination, by changing the medium flowing over the cells to one lacking galactose and containing 2% glucose. Since HO mRNA and HO protein have very short half-lives, no further DSB formation occurs much beyond this medium switch [26].
We confirmed that cells attached to the coverslip and treated as above switch efficiently and select the appropriate donor. We can determine by fluorescence microscopy at the single-cell level whether a cell has switched from MATa to MATa by including in the strain a P MFa1 -CFP construct, which consists of the CFP coding region under control of the MFa1 promoter. The MFa1 promoter is active only in MATa cells. Accordingly, a MATa cell prior to switching does not express CFP, but after it switches to MATa, it begins to accumulate CFP. We found that less than 1% of MATa P MFa1 -CFP cells without HO induction accumulated CFP at the end of the experiment, whereas approximately 50% of the cells that were induced for HO expression accumulated CFP ( Figure  1B). The switching rate of 50% agrees well with that determined by genetic assay with the same strain induced for 1 h in liquid culture, indicating that manipulations in the flow cell and observations by fluorescence microscopy did not affect the process of switching. The rate is somewhat less than the 85% switching rate observed with wild-type HO cells and reflects a slight reduction in the initiation of switching under our induction conditions, rather than any alteration in donor preference (unpublished data). This agrees with Southern analysis and plate-based mating assays of our test strain, from which we concluded that under our induction conditions, approximately half of our test cells sustain a DSB at MAT.

Pairing between Donor and Recipient Loci during Recombination Is Short and Reversible
To monitor pairing of MAT and HML following initiation of a DSB at MAT, we treated MATa cells as described above and then interrogated multiple cells every minute for 3 h. We determined in each cell, at each time point, whether the nucleus contained two GFP dots or a single GFP dot. In the former case, we assume that the MAT and HML loci are unpaired. In the latter case, we assume that the two loci are either paired or simply too close to be resolved. Representa- tive images from experiments in which HO was induced or not induced are shown in Figure 1C, demonstrating the resolution of the two loci in the test strains and indicating that all cells can be readily scored. In Figure 2A, we represent the histories of association of MAT and HML as determined by this method following HO induction for 11 wild-type MATa cells, which normally use HML as the preferred donor. In addition, we examined 11 wild-type MATa cells following HO induction. MATa cells normally use the untagged HMR locus as donor and thus MAT would not be expected to pair with HML in this strain. Finally, we examined 11 wild-type MATa cells in the absence of HO induction, in which no recombination should occur. In these representations, each line describes the history of a single cell; a white bar represents the presence of two dots in the cell's nucleus at that time point and a black bar represents the appearance of only a single dot.
These interrogations clearly show that, relative to the behavior of MATa cells in which HO was not induced, MATa cells in which HO was induced exhibited a statistically significant increase in the overall proportion of time MAT and HML were in close proximity. Moreover, cells from the induced culture divided into two discrete populations. In one group, cells exhibited a pattern of association that was indistinguishable from those of uninduced cells, whereas in the other group, cells exhibited a succession of extended associations between MAT and HML lasting from 7 to 15 min, starting approximately 80 min after initiation of HO expression. The fraction of cells exhibiting the extended association matched the fraction of cells that ultimately undergo gene conversion, and we believe that cells showing extended association were those in which cleavage at MAT had occurred.
A distinction between the association patterns in induced and uninduced cells is also evident by calculating the fraction of cells containing a single dot at each time point. As shown in Figure 2B, cells in which HO was not induced show a uniform, low level of association between HML and MAT throughout the experiment. On the other hand, MATa cells in which HO was induced exhibit increased association of HML and MAT starting from 80 min after initiation of induction of HO and lasting for approximately 1 1/2 h.
We also noted that, despite the fact that more than half of the cells expressing HO undergo gene conversion of MAT using HML as donor, continuous pairing between MAT and HML occurs for no longer than 17 min in any cell. We evaluated the number and duration of events in which HML and MAT appear to be in continuous association, i.e., cases in which a cell presents a single dot in an uninterrupted series of time points. As shown in Figure 2C, in uninduced MATa cells or induced MATa cells, the log of the number of association events of a given duration plotted versus the duration of association exhibits essentially a linear relation, consistent with a stochastic process in which the likelihood that HML and MAT are in close proximity at any time is independent of whether they were in close association at the previous or subsequent time point. Only one event longer than 6 min was observed in uninduced cells. In contrast, induced MATa cells exhibit a statistically significant excess of events of duration longer than 6 min. These extended associations occur for an average of 10 min, predominantly between 80 and 160 min postinduction ( Figure 2D). Unexpectedly, each cell undergoes  Figure 2A. The number of periods of a given duration divided by the total number of periods (probability density) is plotted versus the duration of apparent association for each strain shown in 2A. MATa þ HO (light blue circles); MATa À HO (green diamonds); and MATa þ HO (red squares). Error bars are estimated by dividing the data successively into random halves and computing the standard deviation among 20 random halves. The dashed line shows a fit of the combined values for MATa À HO and MATa þ HO cells to a normalized, exponential distribution. Under this assumption, the p-value for the likelihood that the data for MATa þ HO would arise by chance is less than 10 À5 . (D) Data from Figure 2A for MATa þ HO cells is replotted to show only those pairing events of duration longer than 6 min. DOI: 10.1371/journal.pgen.0020098.g002 more than one prolonged association event. Moreover, during the period between two extended associations, the average distance between the two spots (0.49 6 0.25 lm) is equivalent to that in uninduced cells (0.51 6 0.24 lm), which do not undergo switching. Thus, we conclude that the multiple associations of donor and recipient in a single cell are independent events, suggesting that the pairing of donor and recipient loci is readily reversible during recombination.
Finally, we note from the data in Figure 2 that, in contrast to the pattern seen with MATa cells, MATa cells, which do not use HML as the preferred donor, do not exhibit an enhanced association between MAT and HML over the course of the experiment. Rather, induced MATa cells present a pattern indistinguishable from that obtained in uninduced cells. This suggests that the cleaved MAT locus does not spend significant time in association with the non-preferred donor locus during mating type switching.

Sgs1 and Srs2 Are Required for Stable Pairing between Donor and Recipient Loci during Switching
We anticipate that assembly and resolution of recombination intermediates during switching would require the activity of DNA helicases, and several genes encoding helicases have been implicated in recombination and repair.
In an attempt to clarify the role of various helicases in recombination, we examined the extent of association between MAT and HML in strains deleted for individual helicase genes. As shown in Figure 3, the level of pairing between HML and MAT in a MATa sgs1D ( Figure 3A) or a MATa srs2D ( Figure 3B) mutant following HO induction was significantly reduced compared to that observed in wild-type cells. The level of pairing in the induced srs2D strain was slightly higher than that in the same strain lacking HO but was certainly significantly lower than that observed in the wildtype strain. More dramatically, the level of apparent pairing in the induced sgs1D strain was not significantly different than that in the same strain lacking HO. Thus, association of the DSB with the donor locus appears to be delayed in both mutants. These results would predict that sgs1 or srs2 mutants should be defective in mating type switching.
To confirm the roles of Sgs1 and Srs2 in repair of DSBs during mating type switching, we examined the effect of introducing these mutants into wild-type HO strains. A diploid HO/HO strain, which we used previously to determine switching frequencies under normal conditions [27], was rendered heterozygous for deletion of SGS1 or SRS2. After sporulation and dissection, we examined the behavior of cells and the switching pattern during outgrowth of spore clones. As seen in Table 1, HO SGS1 and HO SRS2 clones showed normal growth and switching patterns. Similarly, ho sgs1 and ho srs2 control strains grew indistinguishably from ho SGS1 and ho SRS2 strains (unpublished data and [9]). In contrast, HO srs2 spore clones were smaller than their sister HO SRS2 spore clones, and pedigree analysis of cells following germination indicated that mother cells in the lineage exhibited delayed cell cycle progression or failed to divide at all. This growth pattern is similar to, but less severe than, the pedigree of death that occurs in HO yeast strains incapable of completing HO-induced switching [28], which we observe with rad52 HO and rad54 HO spores (Table 1). Spore clones carrying an sgs1 deletion did not show as extensive delays in growth as did srs2 spore clones, but the viability of mother cells in pedigrees was clearly less than that seen with sister SGS1 spore clones (unpublished data). These results are consistent with the assumption that srs2 mutants, and to a lesser extent, sgs1 mutants, fail to efficiently heal DSB generated in lineages of HO haploid cells.
We also find that, although both srs2 and sgs1 mutants exhibited normal donor selection in a MATa background, both mutants showed essentially random selection of donor loci following switching in a MATa background (Table 1). This observation confirms that the mutants exhibit a delay in repairing the DSB during mating type switching. In cells in which repair of the HO-induced DSB at MAT is delayed, resection of the DSB into the adjacent coding region results in deletion of the MAT locus, rendering the cell temporarily mat-null. Since mat-null cells exhibit a MATa phenotype, donor selection in MATa cells would be unaffected by this delay-induced resection. However, the delay-induced resection in a MATa would render the cell phenotypically MATa, changing the donor preference from HMR to HML. Thus, the extent to which MATa cells select HML as donor reflects the extent of delay in the repair of the DSB. Accordingly, our results confirm that Srs2 and Sgs1 are both required for timely repair of DSBs.

Rad54 Is Not Required for Stable Pairing between Donor and Recipient Loci during Switching
RAD54 encodes a member of the Swi2/Snf2 family of DNAstimulated ATPases, and loss of RAD54 function substantially diminishes homologous recombination and renders cells sensitive to ionizing radiation. Consistent with previous reports that Rad54 is required for completion of DSB repair [29], HO rad54 spore clones yield microcolonies in our donor preference assay, resulting from extensive death of mother cells during growth of the clone (Table 1). Previous studies on the precise role Rad54 plays in recombination have been inconsistent, with some evidence pointing to its involvement in initial formation of the Rad51 presynaptic filament and other evidence suggesting that it functions after formation of the presynaptic complex [2,3]. To help resolve this issue, we examined pairing of MAT and HML following HO induction in a rad54D mutant.
As evident in Figure 4A, the two loci paired as efficiently in the rad54 mutant as in wild-type cells. Surprisingly, MAT and HML also paired efficiently in a MATa rad54 mutant strain ( Figure 4B), despite HML not being the preferred donor for switching in MATa cells. This result is quite distinct from that observed with HO-induced RAD54 MATa cells, which, as shown in Figure 2, show no prolonged association between MAT and HML. Therefore, association between donor and recipient loci proceeds efficiently in the absence of Rad54 function and, furthermore, association with the inappropriate donor occurs as frequently as does association with the preferred donor.

Discussion
To study the dynamics of recombination, we have developed a real-time assay for homologous pairing in individual living cells. The results from this assay correlate well with previous observations on the time course of the recombination reaction obtained with populations of cells, but also reveal unanticipated aspects of recombination not previously accessible through population studies. The system we examined-mating type interconversion following synchronous cleavage of the MAT locus by HO induction in Saccharomyces-has been used extensively as a model system for homologous recombination [2,3,14,15]. Previous studies Diploid strain Y2902 (MATa/MATa::HIS3 his3/his3 hmla1a2Dinc/hmla1a2Dinc hmra1Dĩínc/hmra1Dĩíinc HO/HO) was made heterozygous for the indicated deletion mutations and then sporulated and tetrads were dissected. Spores and spore clones were monitored microscopically over the first 24 h postdissection to determine the viability and growth pattern of the emerging colony. Spores segregating a high proportion of dead cells were scored as undergoing a ''pedigree of death'' [28], whereas those segregating cells showing extended delays in cell cycle progression were scored as ''delayed division.'' The predominant donor used in a spore clone was determined by PCR analysis as described and correlated with the MAT allele originally present in the spore clone, as determined by the His phenotype. Those spore clones that gave essentially equivalent signals for HML and HMR were not included in the totals.
a Spore clones showed significantly greater inviability of mother cells than did wild-type cells, but not as great as that seen with srs2 clones.  examining products and intermediates of recombination in vivo by PCR or Southern analysis indicate that strand invasion begins approximately 90 min following HO induction and ends approximately 90 min later [2,14]. Furthermore, by using chromatin immunoprecipitation to quantify in vivo association of recombination proteins with regions of the genome, several studies determined that Rad51 initially associates with the DSB at MAT approximately 30 min following HO induction-coincident with DSB formationand with the donor locus, HML, approximately 30 min later [2,3]. This association continues for up to 3 h. These results suggest that synapsis between the DSB and the donor locus begins approximately 60 min following HO induction and that the two loci remain physically associated for up to several hours. Our data are consistent with these population studies in that we observe an overall increase in the average association of MAT and HML beginning 80-90 min following HO induction and lasting for 60-90 min.
Monitoring the association of HML and MAT over time in individual cells revealed aspects of synapsis not evident from these population studies. First, we observed a statistically significant increase in the number of periods of extended association of MAT and HML in induced versus uninduced cells, although the durations of these associations were in general no longer than 17 min and averaged approximately 10 min. Thus, distinct periods of synapsis in individual cells are much shorter than would be expected from population studies. Second, individual cells exhibit several distinct periods of extended association following HO induction. Since HO protein is not present at these later times and since the same pattern occurs in rad54 mutants unable to complete recombination, these sequential associations do not represent sequential rounds of resolution and re-initiation of recombination. Also, these multiple associations are unlikely to be a single synaptic junction punctuated by dynamic stretching, since the distributions of distances between the two loci in the interval between two such extended associations are identical to those in uninduced cells. Rather, we would interpret these observations to suggest that the Rad51-mediated complex formed between the DSB and the template is reversible. For example, the initial Rad51mediated triple-strand complex consisting of the Rad51coated single-strand DNA bound to the homologous doublestrand region of the donor template could be resolved in either two ways: isomerization to yield a single-strand invasion creating a D-loop on the template, or disassociation to return to the pre-complex state. Alternatively, D-loop formation could be the reversible step, either before or after extension of the single-strand primer. Our results suggest that either or both of these initial interactions between the DSB and the template can occur several times before completion of recombination. Such cycles for strand invasion have been suggested to occur during repair of DSBs in Drosophila [30]. The results from those studies with Drosophila would suggest that dissociation could occur not only after Dloop formation but also after partial extension of the primer, as depicted in Figure 5. Finally, we note that the multiple, stochastic associations between the DSB and the donor locus in individual cells account for the apparent extended association between the two loci as measured on a population basis.
We explored some of the genetic requirements for homologous pairing in our system. We first examined the role of the helicases encoded by SGS1 and SRS2 and found that inactivation of either of these genes delayed sustained association between MAT and HML. We confirmed this result obtained from the physical pairing assay by showing that repair of DSBs during mating type switching is delayed in HO spore clones carrying sgs1 or srs2 mutations. Thus, each of these genes appears to be independently required for efficient homologous pairing in our system, an observation consistent with previous studies suggesting that, despite their synthetic lethality, Sgs1 and Srs2 play independent roles in recombination [9,12]. Our results are somewhat unexpected since previous genetic and biochemical studies have implicated Srs2 in inhibiting initiation of recombination by promoting disassembly of Rad51 filament complexes [31][32][33], whereas Sgs1 and other RecQ helicases are likely involved in resolution of recombination intermediates [10,[34][35][36][37][38]. Thus, we might have binds to the resected single-strand ends of the DSB and promotes their association with both HML and HMR via homology present at all three loci. Blue-and pink-colored regions at the mating type loci represent allele-specific DNA sequences, which are bracketed by sequences in common among all three loci (shown in brown). Dissociation of these initial complexes, suggested to be paranemic joints in this illustration, is stimulated by Rad54-promoted removal of Rad51. Since these associations are seen only in rad54 cells, we assume that they are too short-lived to be evident in wild-type cells, and thus are indicated in brackets. We propose that Rad54 also stimulates conversion of the initial joint complex into an initial D-loop structure, also by promoting removal of Rad51, allowing subsequent extension of the heteroduplex between the invading strand and donor locus, and elongation of the invading strand. These subsequent steps are promoted or stabilized by Srs2 and Sgs1. Our results suggest that one or more of the later-stage intermediates in recombination, perhaps even after partial extension of the invading strand, are also reversible prior to completion of the gene conversion event. The dissociated, but incompletely healed, MAT locus then would recycle through Rad51 filament formation and reassociation with the donor locus, ultimately yielding a completely reconstituted MAT locus. DOI: 10.1371/journal.pgen.0020098.g005 anticipated that sgs1 or srs2 mutants would exhibit prolonged association of MAT and HML in our assay rather than reduced association. However, previous studies have shown that, although sgs1 mutants exhibit a hyperrecombination phenotype under some conditions, they are hyporecombinogenic for DNA damage-induced heteroallelic recombination [9,12], a situation similar to DSB-induced mating type switching. In addition, Sgs1 is required for telomere elongation in the absence of telomerase, apparently through recombination via a break-induced repair process that is mechanistically similar to the DSB-induced gene conversion we have studied here [39]. Thus, Sgs1 could be required to stabilize the initial joint formed between a DSB and the donor by promoting formation of an extended D-loop or initiation of synthesis to extend the invading 39 end. In a similar vein, the effect of inactivation of SRS2 on recombination depends on the particular assay used. Perhaps most relevant to our observations, Aylon et al. [40] examined gene conversion of a ura3 allele from an ectopic site of homology following introduction of an HO-induced DSB within the ura3 allele, an assay quite similar to mating type switching. They found that inactivation of SRS2 substantially reduced the level of strand invasion and subsequent extension of the invading strand and concluded that only a small subpopulation of srs2 cells are able to complete recombination, whereas the majority of cells behaved as if no homology to the DSB were present in the cell. This is consistent with results from Pâ ques and Haber suggesting that Srs2 functions to extend and stabilize initial joints between the ssDNA-Rad51 nucleofilament and the donor locus [41]. Thus, both Sgs1 and Srs2 have been implicated independently in stabilizing initial joint formation following DSB-induced recombination, consistent with our current observations.
We also used our assay to examine the role of RAD54 in pairing in vivo. Rad54, a dsDNA-dependent ATPase of the Swi2/Snf2 family of chromatin remodeling enzymes, is required for resistance to ionizing radiation and repair of DSBs by homologous recombination [7,29]. Rad54 interacts with Rad51 in vivo, and addition of Rad54 to Rad51-mediated recombination reactions dramatically stimulates homologous pairing in vitro. In addition, Rad54 is essential for DNA strand invasion reactions using chromatin-packaged substrates [42]. Two studies recently used chromatin immunoprecipitation to examine the in vivo requirement of Rad54 for association of Rad51 with MAT and HML following initiation of a DSB at MAT. Wolner et al. [3] found that Rad51 association with either locus was essentially eliminated in a rad54D mutant background and provided evidence that Rad54 promotes formation of the Rad51 presynaptic filament. On the other hand, Sugawara et al. [2] found that Rad51 could, in fact, associate with MAT and HML in the absence of Rad54, but the synaptic complex could not proceed to a mature strand invasion complex in the absence of Rad54.
We find that MAT and HML associate as well in a rad54 mutant as they do in a wild-type strain. Moreover, we find that MAT associates frequently with HML even in a MATa strain, in which HML is the non-preferred donor. One possible explanation for these associations between MATa and HML is phenotypic switching of mating due to delay in resolution of the MAT DSB, as we proposed for sgs1 and srs2 mutants. However, since the timing of the observed associ-ations in rad54 mutants is the same as that seen in wild-type cells, in which no phenotypic switching occurs, we conclude that the inappropriate associations are not the result of phenotypic change in mating type. Rather, our results suggest that Rad54 plays a critical role in recombination at a step beyond initial synapsis. This conclusion is consistent with the observations above from Sugawara et al. [2] and also with subsequent work by Wolner and Peterson [43]. Although Rad54 may also have a role in vivo in promoting Rad51 nucleofilament formation, our results would suggest that this is not the rate-limiting step in recombination stimulated by Rad54. Our results can be explained by the proposal, based on the observation that Rad54 stimulates dissociation of Rad51 from dsDNA, that Rad54-stimulated removal of Rad51 promotes either dissociation of the paired loci or conversion of the initial synaptic intermediate into a D-loop and the subsequent extension of the heteroduplex region [44]. Thus, in the absence of Rad54, pairing between the DSB and either donor locus should still occur, but the formation of a D-loop and subsequent elongation of the invading strand would not. This model would account for the fact that we observe pairing of MAT and HML in the non-preferred mating background, reflecting formation of Rad51-mediated pairing of MAT with either donor locus via the common homology at all three loci. We would propose that these initial pairing events would be eliminated in a Rad54-dependent fashion, either reversing the initial pairing or promoting progression in the appropriate partner to productive recombination.
A model for recombination during mating type switching is presented in Figure 5. Our results suggest that following HOinduced cleavage of MAT, the locus associates readily and reversibly with both HML and HMR. Donor selection would then occur at a subsequent step, for instance in the isomerization from the initial synaptic intermediate to the D-loop, upon recruitment of DNA polymerase or with extension of the invading strand. Cell type-specific loading of factors promoting D-loop formation and strand extension at the preferred donor locus would then fix donor preference. Our results also suggest that this step is stimulated by both Sgs1 and Srs2. Our results further indicate that, even in the presence of Rad54, the association between donor and template are reversible. This would suggest that even after D-loop formation, the donor and recipient loci can dissociate. Moreover, given recent results suggesting that a doublestrand gap requires multiple cycles of strand invasion, synthesis, and dissociation of the nascent strand during repair of a DSB in Drosophila [30], this reversibility could occur after partial extension of the invading strand, as represented in Figure 5. Further analysis should resolve precisely at what steps recombination is reversible. However, these results clearly indicate that the recombination process is reversible at a much later stage than previously anticipated and raises a novel mechanism for generating genetic diversity.

Materials and Methods
Plasmids and strain construction. Plasmid pPJS218 (HIS3 TetR-GFP LacI-GFP) has been previously described [45]. Plasmid B2609 (LEU2 CEN4 ARS1 P GAL1 -HO P MFa1 -CFP) was constructed by in vivo recombination by co-transforming a leu2 trp1 yeast strain with plasmid B2686 (LEU2 TRP1 CEN4 ARS1 P GAL1 -HO) and two PCR fragments, one spanning the MFa1 promoter and one spanning the coding sequence of CFP from plasmid pDH3 (http://depts.washington. edu/;yeastrc). Both PCR fragments carried 50 base pair ends with homology to direct recombination with the appropriate adjacent fragments. Leu þ FAA-resistant transformants were selected and plasmid was recovered by transformation into Escherichia coli. Plasmid B2608 (LEU2 CEN4 ARS1 P GAL1 -HO P MFa1 -CFP) was constructed in an analogous fashion using the MFa1 promoter. Plasmid structures were confirmed by functional assays and restriction analysis.
Strains used in this study are listed in Table 2 and were all obtained from the S288C-derived haploid strain S150-2B. Construction of strains carrying DNA binding site arrays adjacent to MAT and HML have been described previously [45]. GFP fusions were introduced by integrative transformation of plasmid pPJS218, selecting His þ transformants. The cdc15-2 allele was introduced into S150-2B by first integrating plasmid B2400 (pRS406-cdc15-2) into the CDC15 locus and then selecting FOA-resistant revertants of selected transformants. FOA-resistant, temperature-sensitive isolates were scored microscopically for terminal arrest phenotype to confirm the presence of the cdc15-2 allele. The cdc15-2 allele was then introduced into the tagged strains by genetic crosses. Deletion alleles were introduced as needed by selecting transformants resistant to G418 (200 lg/ml, CalBiochem, San Diego, California, United States) following transformation with PCR products using DNA from the appropriate strain from the kanMX deletion collection, and primers 500 base pairs upstream, 59, and downstream, 39, of the gene.
Strains for donor preference were obtained by transformation of strain Y2902 using PCR products from the deletion collection as described above. Donor preference in spore clones of mutant strains were determined as described previously [27].
Microscopy methods. The yeast strains were freshly transformed with plasmid B2609 or B2608 and grown 3 d at 23 8C on SC-leu media. Three transformants of each strain were inoculated to 20-ml SC-leu broth þ 2% glucose and grown overnight at 23 8C. Cultures were then diluted into SC-leu broth þ 2% raffinose and lacking glucose to a density of 10 6 cells/ml and incubated for 3 h at 37 8C to arrest growth and render GAL1-HO readily inducible. Cells from each culture were examined by fluorescence microscopy to confirm that most exhibited two nuclear GFP dots and none expressed CFP. Cells from cultures with these properties were applied to a microscope coverslip (No.1 22360 mm, Corning, Corning, New York, United States) precoated with Concanavalin A (Sigma, St. Louis, Missouri, United States) as described (http://www.cgr.harvard.edu/thornlab/protocols/ConA.htm). Samples of test and control cultures were applied to the same coverslip but separated by a hydrophobic Pap pen (Daido Sangyo Co. Ltd., Tokyo, Japan) to preclude mixing. The cells were mounted on a flow cell (BioSurface Technologies, Bozeman, Montana, United States) and perfused with SC-leu medium þ 2% raffinose as the sole carbon source, maintained at 37 8C with a thermocouple detector and a feedback-modulated heated stage. Cells from several fields of each strain were repeatedly interrogated every 1 or 2 min using a Nikon Eclipse TE200 microscope with 1003 objective (1.4 aperture) and a planapochromatic light source (Nikon, Tokyo, Japan). Approximately 5 min after mounting the coverslip on the flow cell, the temperature was shifted to 23 8C to release the cdc15-2-imposed cell cycle arrest, and medium perfusing the flow cell was changed to SC-leu þ 2% galactose. This time was set as the zero time point, and galactose perfusion was maintained for 60 min, at which time medium was switched to SC-leu þ 2% glucose. At each interrogation, we captured 30 0.4-lm Z-sections of a 512 3 512 pixel image (0.1 lm per pixel), and prior to analysis, the images were processed by five cycles of deconvolution using Deltavision SoftWoRx v. 2.5 (Applied Precision, Issaquah, Washington, United States). The locations of GFP dots in each cell in a field were determined from the identification of the maximum intensity in the three-dimensional images. Distance between the separable dots was determined by applying the Pythagorean triangulation to the Cartesian coordinates of the maxima using Softworx pick points function and processed with the timedist and yeastparser programs (John Houston, Multimedia Gaming, Austin, Texas, United States). Those cells in which only a single intensity maximum was observed were defined as having a single dot, i.e., the two loci were closer than could be resolved by our microscopy.