In Vivo Roles of Rad52, Rad54, and Rad55 Proteins in Rad51-Mediated Recombination

Molecular Cell, Vol. 12, 209–219, July, 2003, Copyright 2003 by Cell Press

In Vivo Roles of Rad52, Rad54, and Rad55 Proteins in Rad51-Mediated Recombination Neal Sugawara, Xuan Wang, and James E. Haber* Rosenstiel Center and Department of Biology Brandeis University Waltham, Massachusetts 02454

Summary Repairing a double-strand break by homologous recombination requires binding of the strand exchange protein Rad51p to ssDNA, followed by synapsis with a homologous donor. Here we used chromatin immunoprecipitation to monitor the in vivo association of Saccharomyces cerevisiae Rad51p with both the cleaved MATa locus and the HML␣ donor. Localization of Rad51p to MAT precedes its association with HML, providing evidence of the time needed for the Rad51 filament to search the genome for a homologous sequence. Rad51p binding to ssDNA requires Rad52p. The absence of Rad55p delays Rad51p binding to ssDNA and prevents strand invasion and localization of Rad51p to HML␣. Lack of Rad54p does not significantly impair Rad51p recruitment to MAT or its initial association with HML␣; however, Rad54p is required at or before the initiation of DNA synthesis after synapsis has occurred at the 3ⴕ end of the invading strand. Introduction Repair of a double-strand break (DSB) often occurs by gene conversion, a homologous recombination event. Homologous recombination depends on a search for homology by the ends of the broken chromosome to locate an intact donor sequence that could be used as a template for DNA repair. In budding yeast and other eukaryotes, the search for homology is facilitated by the DNA strand exchange protein, Rad51p, the homolog of the bacterial RecA protein. In vitro, Rad51p, like RecA, forms extensive Rad51p-DNA filaments, with 3 bp bound per monomer (Ogawa et al., 1993; Shinohara et al., 1992; Sung, 1994; Symington, 2002). The RecA or Rad51p filament can catalyze strand exchange between ssDNA and a homologous dsDNA (Bianco et al., 1998; Roca and Cox, 1997). This process is greatly facilitated by prior exposure of the ssDNA to a ssDNA binding protein, SSB in bacteria, or RPA in eukaryotes (Bianco et al., 1998; Song and Sung, 2000; Sugiyama and Kowalczykowski, 2002; Sung and Robberson, 1995). These ssDNA binding proteins appear to facilitate the polymerization of RecA/Rad51p across ssDNA regions that can form secondary structure. Incubation of RPA with ssDNA before the addition of Rad51p, however, reduces formation of Rad51 filaments. Either Rad52p or the Rad55p/Rad57p heterodimer can overcome this inhibition and mediate the *Correspondence: [email protected]

loading of Rad51p onto ssDNA (New et al., 1998; Sugiyama and Kowalczykowski, 2002; Sung, 1997a, 1997b). Rad54p shares sequence similarity with the Swi2/Snf2 family of chromatin-remodeling proteins and appears to act at a later step in strand exchange, to extend heteroduplex DNA, and to alter DNA conformation during the synaptic and/or postsynaptic phases of strand exchange (Kiianitsa et al., 2002; Mazin et al., 2000; Petukhova et al., 1998, 1999; Solinger and Heyer, 2001; Solinger et al., 2001, 2002; Tan et al., 2003; Van Komen et al., 2000). In vivo, Rad51p-mediated recombination in S. cerevisiae requires the participation of Rad52p, Rad54p, Rad55p, and Rad57p (Hays et al., 1995; Paˆques and Haber, 1999; Signon et al., 2001). Deletions of RAD52 are the most defective in spontaneous and DSBinduced recombination, as this protein is required both for Rad51p-mediated and Rad59p-mediated recombination events (Bai and Symington, 1996; Ira and Haber, 2002; Signon et al., 2001). The requirements for Rad55p and Rad57p are the least stringent, as defects are often seen only at low temperature. This defect can be suppressed by the overexpression of Rad51p (Hays et al., 1995) or by mutations within the RAD51 gene (Fortin and Symington, 2002). One of the best-studied homologous recombination events is the HO endonuclease-induced switching of the MATa locus, using HML␣ as the donor template during gene conversion (reviewed by Haber, 2002). A galactose-inducible GAL::HO gene provides the means to induce the DSB synchronously in all cells of the population. Physical monitoring of MAT DNA has shown that the DSB ends are first resected by 5⬘ to 3⬘ exonucleases (White and Haber, 1990), presumably so that the 3⬘-ended ssDNA can recruit the Rad51p strand exchange protein and undergo strand invasion of the donor and initiate new DNA synthesis. Although the formation of a strand invasion intermediate has not been directly demonstrated, the next step in the process, the use of the 3⬘ end of the invading strand as a primer to copy the template DNA, has been detected by PCR. This was observed to occur about 30 min after the appearance of the DSB and 30 min before the recombination event was completed (White and Haber, 1990). HO cleavage and 5⬘ to 3⬘ resection are normal in mutant cells lacking the auxiliary recombination proteins Rad52p, Rad54p, Rad55p, and Rad57p, but the primer extension step does not occur (Sugawara et al., 1995, and see below); however, in some strain backgrounds there is a small amount of MAT switching in rad54⌬ mutants (SchmuckliMaurer and Heyer, 1999). To determine more precisely the recombination steps that are prevented in the absence of the auxiliary proteins, we have used chromatin immunoprecipitation (ChIP) to follow the ability of Rad51p to associate with ssDNA formed at MAT and with DNA sequences at HML as well as a model for the synapsis of MAT and HML during gene conversion. This approach has enabled us to see distinct functions for Rad52p, Rad54p, and Radd55p/Rad57p in the Rad51p-mediated process. We conclude that Rad52p is required for the recruitment of

Molecular Cell 210

Rad51p to a DSB, whereas the rate of recruitment is reduced or delayed in the absence of Rad55p/Rad57p. At the step of strand exchange, Rad55p/Rad57p is required for recruitment of Rad51p to the donor whereas Rad54p is not, suggesting that synaptic association between MAT and the HML donor does not need Rad54p. Rad54p is required at a postsynaptic step to enable the completion of DNA repair. Results Binding of Rad51p to the Ends of a DSB in the Absence of Recombination In the first set of experiments, we used strains in which both HML and HMR were deleted, so that the DSB at MAT was not repaired and 5⬘ to 3⬘ resection would continue unimpeded for many hours (Lee et al., 1998). In a wild-type strain, within 30 min of adding galactose to the medium to induce expression of the GAL::HO gene, one can detect significant association of Rad51p to the MAT region, using a set of primers that amplify a region 189 to 483 nt centromere-distal to the DSB end (Figure 1B). This spans a region from within MAT-Z to the unique sequence distal to MAT. The intensity of the PCR product continued to increase for more than 2 hr (Figure 1C, circles). In these experiments the MAT IP signal was normalized to the ARG5,6 locus input signal and to the 0 hr MAT input signal (see Experimental Procedures). We did not normalize our data to the IP signal obtained prior to HO endonuclease induction, as the signal was too low to give a reproducible result. We also did not normalize to the PCR product of DNA immunoprecipitated by anti-Rad51 antibody from an independent locus both because Rad51p localization to undamaged regions was too low and because, as we demonstrate below, the number of Rad51p monomers is limited and would become selectively associated with damaged DNA after HO induction. Moreover, normalizing each time point IP signal to its respective input MAT signal was also problematic, as this region becomes single-stranded and the signal relative to the initial time point would decrease over time (Frank-Vaillant and Marcand, 2002; Lee et al., 1998; White and Haber, 1990). To validate this method of normalization, we repeated this experiment by using anti-Mif2 antibodies to immunoprecipitate CEN3 sequences as an internal control (Meluh and Koshland, 1997). Anti-Mif2 and anti-Rad51 antibodies were simultaneously added to cell extracts to precipitate both Rad51p and Mif2p complexes. PCR signals were obtained using primers from CEN3 and MAT, and the ratio was plotted as a function of time (Figure 1D). The ratio increases for at least 2–4 hr, confirming the kinetics of Rad51p recruitment to an unrepaired DSB at MAT. Binding of Rad51p Is Limited to Regions Near the DSB In strains lacking donor sequences, the region surrounding HO-cleaved MAT becomes progressively single-stranded (White and Haber, 1990). We used a series of primers that detect Rad51p localization along chromosome III on both sides of the DSB site as resection proceded (Figure 2A). The strongest signal was detected using a PCR primer pair immediately adjacent to the

DSB (20 to 204 nt from the DSB lying in MAT-Z). The signal is significantly lower at sites more distant from the DSB but is still significant at 3.6 kb from the DSB; however, at a site approximately 9.5 kb from the DSB, little, if any, chromatin-immunoprecipitated DNA was recovered after 7 hr. This region was expected to be single-stranded after approximately 2–3 hr since we have previously shown that 5⬘ to 3⬘ resection of DSB ends continues for at least 25 kb at a rate of approximately 4 kb/hr (Vaze et al., 2002). To confirm that ssDNA did indeed form at a site distant from the DSB, we examined the localization of the ssDNA binding complex, RPA, after HO induction (Figure 2D). By 2 hr, a region 9.5 kb proximal to the DSB can be immunoprecipitated by an anti-Rfa1 antibody, even though little, if any, Rad51p failed to immunoprecipitate at this distant site. We wondered whether the lack of Rad51p localization to more distant sites could be the result of a limitation in the amount of Rad51p present in the cells, so that long ssDNA had effectively titrated all the Rad51p. To address this, we expressed wild-type RAD51 under control of the PGK1 promoter on a centromere-containing plasmid, leading to an approximately 10-fold increase in Rad51p, based on Western blot analysis (data not shown). Under these conditions, we found that we could immunoprecipitate Rad51p with sequences 9 and even 17 kb proximal to the DSB site (Figure 2C). On the basis of these data we conclude that in most cells there is only enough Rad51p, even when the DNA damage checkpoint has been activated (Pellicioli et al., 2001), to cover fully less than 10 kb of ssDNA. This corresponds to about 3500 monomers of Rad51p. This estimate is identical to one reached by Mazin et al. (2000) based on biochemical purification of Rad51p. Effect of rad Deletion Mutations on Rad51p Recruitment in the Absence of Recombination To analyze the genetic requirements of loading Rad51p onto ssDNA, we used a set of isogenic strains lacking either rad52⌬, rad55⌬, or rad54⌬ and lacking a donor template. ChIP was performed using anti-Rad51p antibody, and DNA was then amplified with primers between 189 and 483 nt distal to the DSB. In contrast to wildtype cells, there was no significant immunoprecipitation of Rad51p-associated DNA in a rad52⌬ strain (Figures 3A and 3B). This result suggests that Rad52p is essential for Rad51p filament formation in vivo. In the case of rad55⌬, we found a statistically significant delay in the recruitment of Rad51p. This is evident at early time points from 0 to 2 hr (Figure 3C). Data collected for the 1 hr time points (Figures 3C and 3D) show that the difference between rad55⌬ and wild-type is statistically significant (Student’s t test, p ⫽ 0.014). This suggests that Rad55p, presumably as part of the Rad55p-Rad57p heterodimer, facilitates the loading of Rad51p onto ssDNA, which is consistent with models based on biochemical experiments (Sung, 1997b). Eventually the level of Rad51p IP reaches a level comparable to that of wild-type (Figure 3D and see below). Extensive association of Rad51p was observed when HO was induced at either 23⬚C or at 30⬚C (data not shown). The absence of a difference between the two temperatures is notable as in many recombination and repair events

In Vivo Assembly of Rad51 during Recombination 211

Figure 1. Localization of Rad51p to a DSB Created by HO Endonuclease A donorless strain, JKM139, incapable of repairing a DSB, was grown in YP lactate medium and induced to express HO endonuclease with 2% galactose. Samples were taken before induction and at 1, 2, 4, and 7 hr after addition of galactose. Polyclonal antibodies against Rad51p were used to precipitate Rad51p-bound chromatin. (A) Map of MAT shows the locations of the HO cut site and the primers (arrows), 189 to 483 bp distal to the DSB, used to PCR-amplify DNA from the immunoprecipitated DNA. As controls, primers to the ARG5,6 locus were used to amplify DNA from the immunoprecipitated chromatin and input DNA. PCR amplified DNA was run on ethidium bromide-stained gels (reverse images are shown). (B) Quantitated signals were graphed for the wild-type strain. IP represents the ratio of the Rad51p IP signal to input signal from an independent locus, ARG5,6 (see Experimental Procedures). Cells were harvested at 30 min intervals after HO induction. (C) Rad51p localization to MAT was followed either in a donorless strain where the DSB cannot be repaired or in a switching strain where the DSB can be repaired by gene conversion. Error bars show one standard error of the mean. (D) Whole-cell extracts were treated with anti-Rad51 antibody and with anti-Mif2 antibody to simultaneously precipitate both Rad51p-MAT and Mif2p-CEN3 complexes. PCR signals from the MAT distal site (above) were normalized to the PCR signals from CEN3 signal and plotted as a function of time.

rad55⌬ mutants only display a mutant phenotype at the lower temperature (Lovett and Mortimer, 1987), although MAT switching and interchromosomal gene conversion in rad55⌬ mutants is defective at both temperatures (Signon et al., 2001). We were also able to detect recruitment of Rad51p to MAT in a rad54⌬ background with kinetics similar to wild-type (Figure 3E). Although the standard errors in this experiment were nonoverlapping, the Student’s t test suggests that this difference was not statistically significant (p ⫽ 0.54, 0.51, 0.47, 0.64, and 0.40 for the 0, 1, 2, 4, and 7 hr time points, respectively). Moreover, in experiments where a DSB is repaired by gene conversion, shown later, Rad51p loading in the absence of Rad54p was not different from wild-type. We note that a reduced level of Rad51p localization in rad54⌬ would be consistent with a study suggesting that Rad54p may play an early role in the assembly of a strand exchange complex (Wolner et al., 2003); however, Rad54p’s predominant function seems to be in later steps of strand

invasion as studied in vitro (Mazin et al., 2000; Petukhova et al., 1998, 2000; Solinger and Heyer, 2001; Solinger et al., 2001; Van Komen et al., 2000). As shown below, we also find an important role for Rad54p at a later step in recombination in vivo. Association of Rad51p with the Donor and Recipient Sequences Using a second set of strains, we examined the repair of a DSB by gene conversion, using HML␣ as the donor template (HML␣ MATa hmr⌬::ADE1). GAL::HO was induced for 1 hr and then repressed by the addition of 2% glucose. Because HO endonuclease protein turns over rapidly, we could then examine the repair of the DSB at MATa by recombination with HML␣ (Connolly et al., 1988; White and Haber, 1990). In wild-type strains, within 1 hr, we find ChIP of Rad51p not only with the ssDNA tail at MAT but also with the HML donor sequences, as measured using HML-specific primers (Figure 4). There was no association of Rad51p with HML

Molecular Cell 212

observed that Rad51p localized to MAT before HML (Figure 4B). This difference was observed most clearly by carrying the reaction out at 20⬚C rather than at 30⬚C to slow the process down and has been confirmed using a different strain background (W303) at 30⬚C (data not shown). Localization to MAT took place with approximately the same kinetics as observed earlier (Figures 1B and 4A), whereas localization to HML was delayed by about 30 min. This delay should reflect the time it takes to search the genome to find a homologous partner. Association of Rad51p with HML occurs prior to the detection by PCR of a strand invasion intermediate in which new DNA synthesis has begun (Figure 5A). In this assay primers were selected to hybridize to MAT- and HML-specific sequences such that strand invasion and extension at HML will produce a PCR product. This is followed by completion of MAT switching as seen on a Southern blot (Figure 5B). This timing is consistent with a model in which a Rad51p-DNA filament from MAT synaptically associates with HML. Thus, we can see four distinct steps: the binding of Rad51p to ssDNA at MAT, the synapsis between MAT and HML, the initiation of new DNA synthesis to extend the 3⬘ end of the invading ssDNA, and the completion of gene conversion. There is one notable difference between the ChIP results when there is a homologous donor compared to its absence. When repair takes place, the extent of Rad51p association with MAT DNA is significantly lower than when repair cannot occur. Comparing the samples at 2 hr, each normalized to the ARG5,6 reference, there is approximately four times as much Rad51p bound in the absence of HML␣ as in its presence (Figure 1C). This result suggests that once a threshold amount of Rad51p is loaded onto ssDNA and the search for homology is successful, there is a downregulation of resection and/or removal of Rad51p from adjacent DNA that is not homologous to the donor. It should be noted that MAT and HML share only 320 bp of homology to the right of the DSB, so that the amount of Rad51p associated with the donor, HML, is limited. The extent of Rad51p association with HML is comparable to or perhaps less than the amount seen at MAT in this recombination-proficient strain.

Figure 2. Rad51p Localizes to Sites Near the DSB Rad51p-bound chromatin was precipitated from HO-induced strains before induction and at 1, 2, 4, and 7 hr after induction. DNA was PCR amplified from sites located proximal and distal to the DSB and then quantitated and graphed as described in Figure 1 and in the Experimental Procedures. The DSB is shown at 0 bp. HO was induced in a (A) donorless wild-type strain, JKM139 (B) JKM139 ⫹ pRS316, a URA3 centromere-bearing plasmid as a control, and (C) strain tNS2048 containing the RAD51-overexpressing plasmid (JKM139 ⫹ pNSU256). (D) Whole-cell extracts from tNS2048 were treated with anti-Rfa1, and precipitated chromatin was used to amplify a sequence 9.5 kb proximal to the DSB.

in a control strain in which HO endonuclease was induced but where the cleavage site at MAT was deleted (data not shown); hence, Rad51p’s association with HML␣ is dependent on the initiation of recombination by a DSB. When samples were harvested at 15 min intervals, we

In rad55⌬, Rad51p Associates with MAT but Not with HML Although rad55⌬ strains did allow delayed but extensive Rad51p association with MAT, they failed to exhibit an association of Rad51p with the HML donor (Figures 6A and 6C). This result is consistent with the failure of rad55⌬ strains to complete switching or to produce a primer-extended strand invasion intermediate (Sugawara et al., 1995). Thus, we see two defects in rad55⌬ strains; there is likely a delay and/or defect in the formation of a Rad51p filament and a failure to achieve synapsis with the donor. It is possible that, although Rad51p is loaded onto ssDNA without Rad55p, the strand exchange protein may not be in a functional filament structure (see Discussion). Rad54p Is Not Required for Synapsis but Is Necessary for Subsequent Steps in Recombination In rad54⌬ strains, Rad51p associated with the HML donor sequences nearly as efficiently as in the wild-type

In Vivo Assembly of Rad51 during Recombination 213

Figure 3. Effect of rad52⌬, rad54⌬, and rad55⌬ on the Localization of Rad51p to an Unrepairable DSB An unrepairable DSB was created in rad52⌬, rad54⌬, and rad55⌬ strains, and Rad51pbound chromatin was immunoprecipitated using anti-Rad51 antibodies. (A) PCR-amplified DNA from the MAT locus was run on ethidium bromide-stained gels. Reverse images of representative gels are shown. DNA signals were quantitated as described in the Experimental Procedures for (B) rad52⌬ (JKM166), (C) rad55⌬ (YLS52), (D) rad55⌬ (YLS52), and (E) rad54⌬ (YLS42) strains.

strain (Figures 6A and 6E). Moreover, the amount of Rad51p at the MAT locus in wild-type and rad54⌬ strains was not statistically distinguishable when the HML donor was present, although the mean values suggest that Rad51p localizes more to MAT in the rad54⌬ strain than in wild-type. Perhaps in the wild-type strain, Rad51p is limited to the region of homology shared with the donor, whereas in the rad54⌬ strain Rad51p has access to more ssDNA present due to the inability to complete gene conversion. It is possible that Rad51p recruitment at MAT and HML in the absence of Rad54p could be carried out by the Rad54p homolog, Tid1p (Rdh54p), as Tid1p has been shown to facilitate Rad51p’s strand exchange in vitro (Petukhova et al., 2000). We therefore examined the ability of Rad51p to associate with HML in a rad54⌬ tid1⌬ mutant and found that, just as with rad54⌬, antiRad51p antibody was able to chromatin immunoprecipitate both MAT and HML after HO induction (Figures 6A, 6D, and 6E). In spite of Rad51p’s association with HML in the absence of Rad54p, the next step in recombination, primer extension of the 3⬘ invading end, was not detected by the PCR assay (Figure 5A). These results suggest that

Rad51p can form at least a paranemic association of the MAT ssDNA with the HML␣ dsDNA, but that a key step—possibly the conversion into an interwound, plectonemic, joint—could not be accomplished. These results clearly show that Rad54p has a postsynaptic role in recombination (see Discussion). Discussion The use of physical monitoring of in vivo DSBs has allowed us to learn much about the process of homologous recombination, including the rate of resection of DSB ends (Lee et al., 1998; White and Haber, 1990), the kinetics of mismatch repair of heteroduplex DNA (Haber et al., 1993), and the formation of a primer-extended strand invasion intermediate (White and Haber, 1990). In strains with mutations of essential DNA replication genes, physical monitoring has also established the importance of many of the components of the normal DNA replication fork in the DSB repair process (Holmes and Haber, 1999; Umezu et al., 1998). An earlier study using ChIP revealed the recruitment of mismatch repair proteins to DSB ends (Evans et al., 2000). Now we have

Molecular Cell 214

Figure 5. rad54 Mutants Are Unable to Carry Out Primer Extension at HML after Strand Invasion Figure 4. Rad51p Recruitment to HML and MAT during DSBInduced Gene Conversion Strains carrying an HML␣ donor (HML␣ MATa hmr⌬::ADE1 ade3::GAL::HO) were treated with 2% galactose to induce HO endonuclease and then with 2% glucose after 1 hr to repress HO. AntiRad51 antibodies were used to immunoprecipitate Rad51p-bound chromatin. (A) PCR-amplified DNA, 189 to 483 bp distal to the DSB (MAT) and 188 to 467 bp from the uncleaved HO recognition site at HML, were prepared as described in Figure 1. Reverse images are shown. Quantitated signals were plotted for the MAT and HML loci in a wild-type strain (JKM161). One standard error is plotted. (B) Early time points were also examined at 15 min intervals in cells grown and induced at 20⬚C.

monitored the participation of the Rad51 strand exchange protein and its dependence on other recombination proteins. This has enabled us to visualize two key steps that must occur early in homologous recombination: the recruitment of Rad51p to ssDNA and the formation of a synaptic association between this nucleoprotein filament and the donor sequences. DSB-initiated gene conversion events, between homologous sequences on different chromosomes or when the

(A) Input DNA from wild-type (JKM161) and rad54⌬ (tNS2045) strains from the ChIP experiments in Figure 4 was used to PCR-amplify DNA using a unique primer distal to MAT and a primer within the Y␣ sequence from HML (White and Haber, 1990). PCR products from samples before and after HO induction were run on ethidium bromide-stained gels. Reverse images are shown. (B) Southern blot illustrates the physical monitoring of a DSBinduced gene conversion event as a wild-type strain, JKM161, switches from MATa to MAT␣. DNA was extracted from a time course culture used in Figure 4, digested with StyI, and prepared for Southern analysis using the MAT-specific probe (black box).

donor sequences are ⱖ100 kb away on the same chromosome as in MAT switching, are nearly completely eliminated in rad51⌬, rad54⌬, rad52⌬, and rad55⌬ mutants (Hays et al., 1995; Ira and Haber, 2002; Schmuckli-Maurer and Heyer, 1999; Signon et al., 2001; Sugawara et al., 1995). Previous analysis of DNA intermediates could not distinguish between the time of action of these different recombination proteins. From the present results we conclude that Rad52p acts earliest and its association with the DSB end is a prerequisite for the recruitment of Rad51p. These results are consistent with cytological studies showing that the formation of Rad51p foci depends on Rad52p (Gasior

In Vivo Assembly of Rad51 during Recombination 215

Figure 6. Effect of rad Mutations on Rad51p Localization at HML during DSB-Induced Gene Conversion Strains were induced to undergo DSB-induced gene conversion with 2% galactose followed by 2% glucose to repress HO expression. AntiRad51 antibodies were used to immunoprecipitate Rad51p-bound chromatin. (A) PCR-amplified DNA, 189 to 483 bp distal to the DSB (MAT) and 188 to 467 bp from the uncleaved HO recognition site at HML, were prepared as described in Figure 1. Reverse images are shown. Quantitated signals were plotted for (B) MAT and (C) HML in a rad55⌬ strain (tNS2042), and (D) MAT and (E) HML in a rad54⌬ strain (tNS2045). Error bars represent one standard error of the mean.

et al., 1998; Lisby et al., 2001). The degree to which Rad52p has been required in several in vitro studies of strand exchange ranges from essential to important, depending on the precise assay conditions (New et al., 1998; Shinohara and Ogawa, 1998; Sung, 1997a). Rad52p interacts directly with Rad51p, although it may not function solely through this interaction (Krejci et al., 2002). The human Rad52p binds strongly and precisely to the ends of ssDNA (Parsons et al., 2000; Van Dyck et al., 1999) but also can bind along ssDNA (Van Dyck et al., 1998). Both the human and yeast Rad52 proteins have been shown to facilitate the displacement of RPA from DNA as Rad51p filament formation occurs (Benson et al., 1998; New et al., 1998; Shinohara and Ogawa, 1998; Sugiyama and Kowalczykowski, 2002; Sung, 1997a). In vivo, the early steps of filament formation apparently

cannot proceed without Rad52p. We note that we cannot assess any possible roles for Rad52p in later stages of homologous recombination. It is possible that Rad52p is needed in the stabilization of strand invasion, acting as a strand-annealing protein, to engage the second end of the DSB (Mortensen et al., 1996; Sugiyama et al., 1998). On the basis of our results we suggest that Rad55p (and presumably its heterodimeric partner, Rad57p) might have two distinct roles in DSB repair: first in facilitating the loading of Rad51p onto ssDNA and later in promoting strand exchange. It is possible, however, that the delayed kinetics of Rad51p loading onto ssDNA in the absence of Rad55p reflects formation of a nonfunctional filament, as illustrated in the model in Figure 7. The failure to form a fully functional filament may also

Molecular Cell 216

Figure 7. Models of Rad52p, Rad54p, and Rad55p Participation in Rad51p-Mediated Homologous Recombination Following a DSB at MATa, each end is resected by 5⬘ to 3⬘ nucleases. In the wild-type cell, a Rad51p filament forms on ssDNA in the MAT-Z region (red) (top row) and then promotes strand invasion at HML␣ (bottom row). The sequences at HML that are not homologous with MATa are shown (Y␣ [gray] and HML-distal [hatched]). In the absence of Rad52p, no filament forms. Without Rad55p, filament formation may be incomplete and interrupted, perhaps because Rad55p and Rad57p are needed to remove RPA or facilitate binding to regions of secondary structure. The Rad51p that is bound to ssDNA in the absence of Rad55p is unable to facilitate strand invasion. Without Rad54p, formation of a Rad51p-DNA filament is normal and strand invasion occurs, but only to the point where a paranemic joint is formed. Formation of an interwound (plectonemic) joint that would allow the synthesis of new DNA, copying Y␣, requires Rad54p. Alternatively, Rad54p allows removal of Rad51p from the heteroduplex DNA permitting strand extension and subsequent recombination steps.

explain why Rad55p was found to be necessary to form cytologically visible Rad51p foci in DNA-damaged cells (Gasior et al., 1998). Nevertheless, the facts that overexpressing Rad51p in the absence of Rad55p partially suppresses the recombination defect (Hays et al., 1995; Johnson and Symington, 1995) and that some rad51 mutants also bypass rad55⌬ (Fortin and Symington, 2002) show that Rad55p’s role is not essential but still significantly important in effecting strand exchange. Support for a role of Rad51 paralogs after filament formation has recently been found for the Xrcc3p in mammalian recombination (Brenneman et al., 2002). One of the important differences between the DNA binding of bacterial RecA protein and yeast Rad51p is that the latter is much less cooperative; therefore, when Rad51p is limiting it will occupy several independent regions along ssDNA rather than “zippering up” as would the more cooperative RecA protein (De Zutter and Knight, 1999; Kiianitsa et al., 2002). Hence, it is conceivable that, without Rad55p, Rad51p binds to ssDNA in small patches rather than contiguously, and this is insufficient to support efficient strand exchange. One of the most striking findings is that the Rad54p is not required for synapsis of donor and recipient per se, but it is required for the creation of a mature strand invasion structure that can be used to initiate new DNA synthesis. While this study cannot shed direct light on the DNA structures involved, one possibility is that Rad54p may be important in converting a paranemic joint to an interwound plectonemic structure that will permit binding of PCNA and DNA polymerase and the initiation of DNA synthesis from the 3⬘ end of the invading strand (Figure 7). Rad54p could also be needed to remove Rad51p from the strand invasion complex formed after strand invasion and heteroduplex DNA formation (Kiianitsa et al., 2002; Solinger et al., 2002). Both of these roles are consistent with in vitro studies that show that Rad54p, in association with Rad51p, promotes DNA unwinding, homology searching, and heteroduplex DNA extension (Mazin et al., 2000; Petukhova

et al., 1998, 2000; Ristic et al., 2001; Solinger and Heyer, 2001; Solinger et al., 2001; Van Komen et al., 2000). ChIP analysis of protein association with DNA in recombination can be performed both qualitatively and quantitatively. It is evident from analyzing these data in an all-or-none fashion that Rad51p cannot associate with ssDNA in the absence of Rad52p. Moreover, qualitative analysis shows that Rad51p filament formation is possible without Rad55p or Rad54p; the subsequent step of synapsis requires Rad55p but not Rad54p. Qualitative analysis also allows us to conclude that the concentration of Rad51p in most cells—even after activation of the DNA damage checkpoint—is limited to less than approximately 3500 monomers because Rad51p fails to occupy a site 9 kb from the DSB unless the protein is overexpressed. Quantitative analysis allows us to draw further conclusions concerning the kinetics and extent of Rad51p localization in different circumstances. Conclusions with respect to Rad51p localization in rad54⌬ and rad55⌬ strains have been discussed earlier. In addition, we conclude that Rad51p localizes near an unrepaired DSB relatively slowly over a 2 hr period, that there is a limited amount of Rad51p per cell that mostly localizes within several hundred bp of the DSB, that there is less binding at MAT when a donor is present, and that there is delayed localization at HML. We recognize that there are complicating factors in analyzing ChIP data quantitatively. For example, we do not know the relationship between the number of Rad51p molecules bound to ssDNA and the efficiency with which crosslinked sequences are immunoprecipitated. The increase in Rad51p ChIP signal over a period of 2 hr, for example, may be the result both of Rad51p slowly displacing RPA and a requirement for the presence of a significant number of Rad51p monomers to create an IP signal. Alternatively, the increase in binding near the DSB may result from the delayed and asynchronous initiation of resection, leading to a growing subpopulation of cells with resected DNA, which is consistent with recent experimental observations (Frank-Vaillant and Marcand, 2002).

In Vivo Assembly of Rad51 during Recombination 217

We also recognize that the observations could reflect changes in the “crosslinkability” of Rad51p and/or in its accessibility to the anti-Rad51 antibody. For example, differences in Rad51p localization in rad52⌬, rad55⌬, or rad54⌬ strains may be due to changes in the conformation of Rad51p that reduce or enhance its ability to be immunoprecipitated. However, we point out that our observations are largely consistent with in vitro results pointing to a role for Rad52p and Rad55p in mediating loading of Rad51 and to a role for Rad54p in a postsynaptic step of gene conversion. That the ChIP assay is a valid means of investigating gene conversion is supported by the observations that Rad51p only binds after a DSB is formed and it binds to MAT before HML during gene conversion. Other ChIP experiments show that Rfa1p will also bind after a DSB is formed prior to Rad51p loading, and that loading is enhanced in a rad51⌬ strain and diminished in a RAD51-overexpressing strain (unpublished data), which are observations consistent with current models of DSB-induced gene conversion. The use of ChIP to study the early events in recombination has made it possible to identify distinct phenotypes for Rad52p, Rad51p, Rad54p, and Rad55p. Similar approaches should make it possible to examine the ways other proteins influence Rad51p function and to examine the participation of replication factors, topoisomerases, helicases, and other proteins needed to complete the repair of a DSB. Experimental Procedures Strains Donorless strains were derived from JKM139 (hml⌬::ADE1 MATa hmr⌬:: ADE1 ura3-52 leu2-3,112 trp1::hisG lys5 ade1-100 ade3::GAL::HO), except for JKM166 (rad52⌬) which was derived from JKM179 (hml⌬::ADE1 MAT␣ hmr⌬::ADE1 ura3-52 leu2-3,112 trp1::hisG lys5 ade1-100 ade3::GAL::HO) (Moore and Haber, 1996). Donorless strains YLS42 and YLS52 were derived from EI515 (hml⌬::ADE1 MATa hmr⌬::ADE1 ura3-52 leu2-3,112 lys5 ade1-100 ade3::GAL::HO) (Malkova et al., 1996) and contain rad54⌬::LEU2 and rad55⌬::LEU2, respectively. Strains capable of undergoing DSB-induced gene conversion were derived from JKM161 (HML␣ MATa hmr⌬::ADE1 ura352 leu2-3,112 trp1::hisG lys5 ade1-100 ade3::GAL::HO). JKM161 was transformed with pSTL11 (rad55⌬::LEU2) to create tNS2042 or with pLS101 (rad54⌬::URA3) to create tNS2045. The tid1⌬::kanMX (tNS2057) and mat⌬::kanMX (tNS2053) deletions were introduced by transformation (Wach et al., 1994). The plasmid pSL57 (gift from Sang Eun Lee) was previously derived from the centromere vector pRS316 and contains the rad51-K191A allele fused to the PGK1 promoter that results in an approximately 10-fold overexpression of Rad51p as assessed by Western blot analysis. The rad51-K191A allele was derived from pR51.3 (Sung and Stratton, 1996). The RAD51-overexpressing strain, tNS2046, was created from JKM139 by transforming pSL57, cleaved with Bsu36I and Bpu1102I, to gap repair the rad51-K191A allele to wild-type. Correct repair was confirmed by DNA sequencing, and the plasmid (designated pNSU256) was isolated in E. coli. A control strain, tNS2048, contains pRS316 in JKM139. DNA Analysis When cells were harvested for ChIP at intervals after induction of HO (see below), a second aliquot was collected for DNA extraction by a glass bead protocol and analyzed by Southern blots as described previously (Sugawara et al., 1995). The strand invasion/ primer extension assay in Figure 5 was previously described (White and Haber, 1990). The primers used were 5⬘-GCAGCACGGAATATG GGACT-3⬘ and 5⬘-ATGTGAACCGCATGGGCAGT-3⬘.

Chromatin Immunoprecipitation ChIP was carried out as described previously with minor modifications (Evans et al., 2000; Strahl-Bolsinger et al., 1997). Cells were grown to a density between 3 ⫻ 106 and 1 ⫻ 107 cells/ml in YP lactate medium, and HO endonuclease was induced by addition of 2% (w/v, final concentration) galactose. Strains undergoing DSBinduced gene conversion were treated after 1 hr with 2% (w/v) glucose to repress the GAL::HO gene. Absorbance readings (OD600) were taken before induction, and subsequent samples were diluted with medium to maintain this absorbance. This approximately maintains the protein:formaldehyde:antibody ratio throughout the time course. Proteins were crosslinked by the addition of 1% (final concentration) formaldehyde to 45 ml of culture for 10 min, followed by quenching with glycine (125 mM final concentration) for 5 min. Cells were lysed with glass beads, and the extracts were sonicated to shear the DNA to an average size of 0.5 kb. Extracts were divided into IP and input samples (7:1 ratio). IP samples were incubated with affinity-purified anti-Rad51 antibody (provided by P. Sung) or unpurified antibody (A. Shinohara) for 1 hr at 4⬚C and bound to protein A agarose beads for 1 hr at 4⬚C. Anti-Mif2 antibody was provided by D. Koshland (Meluh and Koshland, 1997) and anti-Rfa1 antibody by S. Brill. The protein bound beads were carried through a series of washes, followed by elution of the proteins and reversal of crosslinking (6 hr at 65⬚C) (Strahl-Bolsinger et al., 1997). Samples were treated with proteinase K followed by phenol extraction and ethanol precipitation.

PCR Amplification MAT-specific primers (5⬘-TCCCCATCGTCTTGCTCT-3⬘, 5⬘-GCATGG GCAGTTTACCTTTAC-3⬘) were used in Figures 1, 3, 4, and 6. The HMLspecific primers were 5⬘-TCCCCATCGTCTTGCTCT-3⬘ and 5⬘-CCC AAGGCTTAGTATACACATCC-3⬘. Primers used for the amplification of the sites proximal to the DSB (Figure 2) were ⫺29.8 kb, 5⬘-TCG TCGTCGCCATCATTTTC-3⬘, 5⬘-GCCCAAGTTTGAGAGAGGTTGC-3⬘; ⫺16.6 kb, 5⬘-CGTCTTCTCAGCGAACAACAGC-3⬘, 5⬘-GCAATAACC CACGGAAACACTG-3⬘; ⫺9.5 kb, 5⬘-TCAGGGTCTGGTGGAAGGAATG3⬘, 5⬘-CAAAGGTGGCAGTTGTTGAACC-3⬘; ⫺5.3 kb, 5⬘-ATTGCGACA AGGCTTCACCC-3⬘, 5⬘-CACATCACAGGTTTATTGGTTCCC-3⬘; ⫺3.6 kb, 5⬘-ATTCTGCCATTCAGGGACAGCG-3⬘, 5⬘-CGTGGGAAAAGT AATCCGATGC-3⬘; and ⫺1.6 kb, 5⬘-ATGTCCTGACTTCTTTTGAC GAGG-3⬘, 5⬘-ACGACCTATTTGTAACCGCACG-3⬘. Oligos used for the sites distal to the DSB were 0.2 kb, 5⬘-CCTGGTTTTGGTTTTGTA GAGTGG-3⬘, 5⬘-GAGCAAGACGATGGGGAGTTTC-3⬘; 2.1 kb, 5⬘-GCC TCTATGTCCCCATCTTGTCTC-3⬘, 5⬘-GTGTTCCCGATTCAGTTTGACG3⬘; 3.1 kb, 5⬘-TAACCAGCAATACCAAGACAGCAC-3⬘, 5⬘-TTTTACC TACCGCACCTTCTAAGC-3⬘; 5.7 kb, 5⬘-CCAAGGAACTAATGATC TAAGCACA-3⬘, 5⬘-ACCAGCAGTAATAAGTCGTCCTGA-3⬘; 9.5 kb, 5⬘-TGGATCATGGACAAGGTCCTAC-3⬘, 5⬘-GGCGAAAACAATGGC ACTCT-3⬘. Primers specific for the ARG5,6 locus were 5⬘-CAAG GATCCAGCAAAGTTGGGTGAAGTATGGTA-3⬘ and 5⬘-GAAGGATC CAAATTTGTCTAGTGTGGGAACG-3⬘. CEN3 was PCR amplified using the oligos 5⬘-GATCAGCGCCAAACAATATGG-3⬘ and 5⬘-AAC TTCCACCAGTAAACGTTTC-3⬘ (Meluh and Koshland, 1997). All PCR assays were accompanied by reactions using dilutions of the 0 hr input sample to assess the linearity of the PCR signal and to create a calibration curve. Samples were run on ethidium bromide-stained agarose gels (2%) and quantitated using an Innotech Alphaimager and Quantity One software (Biorad) which was also used to correct for minor deviations from a linear response in signal. Signals from the input (1:40 dilutions) and IP (undiluted) samples were normalized to the 0 hr signal from the input samples using the calibration curve. Previously, we normalized our samples to the 0 hr IP signal to measure the relative increase in IP signal (Evans et al., 2000). When using Rad51 antibody, we sometimes saw that the 0 hr IP signal was very low, making it difficult to obtain reproducible results when time point signals were normalized to this signal. The 0 hr IP signal may result from insufficient washing of IP samples or from Rad51p binding to spontaneous DNA lesions. For these reasons all IP samples were normalized to the input signal from an independent locus (ARG5,6) on chromosome V. This was accomplished by dividing each MAT or HML IP signal by the respective time point signal from the ARG5,6 input sample to correct for differing amounts of chromatin collected for each time point sample. We decided against

Molecular Cell 218

normalizing time point samples to the respective MAT input time point samples since its quantity decreases due to 5⬘ to 3⬘ nuclease degradation initiated at the DSB. We also considered normalizing against the ARG5,6 IP signals in a multiplex experiment by using ARG5,6 and MAT primers in the same PCR reactions. This approach was not used because the ARG5,6 IP signals were often very low relative to the MAT IP signal. Second, we were concerned that competition between MAT and ARG5,6 for Rad51p would alter its binding at ARG5,6 at later time points. Standard errors of the mean were calculated for independent HO inductions (three to eight time courses) except for Figure 3C where the range is shown for an average of two inductions and Figure 6D where rad54⌬ tid1⌬ represents one induction. Data collected from time courses of rad55⌬ strains at 23⬚C and 30⬚C were similar and were averaged. Student’s t test was used to assess statistical significance (paired, two-tailed, where n ⫽ 4 and 6 for the rad54⌬ and rad55⌬ donorless strains, respectively). Acknowledgments We thank E. Evans, T. Goldfarb, and E. Alani for helping us to set up ChIP assays, and R. Shroff and M. Lichten for sharing their results before publication. We are indebted to P. Sung and A. Shinohara for gifts of anti-Rad51p antibody. We also thank S. Brill and D. Koshland for antibodies against Rfa1p and Mif2p. We thank the persons above as well as S. Lovett, M. Jasin, L. Symington, B. Wolner, and C. Peterson and members of the Haber lab for helpful comments. This work was supported by NIH grant GM20056. Received: October 3, 2002 Revised: May 1, 2003 Accepted: May 8, 2003 Published: July 24, 2003

type genes. In Mobile DNA II, N.L. Craig, R.Craigie, M. Gellert and A.M. Lambowitz, eds. (Washington, DC: ASM Press), pp. 927–952. Haber, J.E., Ray, B.L., Kolb, J.M., and White, C.I. (1993). Rapid kinetics of mismatch repair of heteroduplex DNA that is formed during recombination in yeast. Proc. Natl. Acad. Sci. USA 90, 3363– 3367. Hays, S.L., Firmenich, A.A., and Berg, P. (1995). Complex formation in yeast double-strand break repair: participation of Rad51, Rad52, Rad55, and Rad57 proteins. Proc. Natl. Acad. Sci. USA 92, 6925– 6929. Holmes, A., and Haber, J.E. (1999). Double-strand break repair in yeast requires both leading and lagging strand DNA polymerases. Cell 96, 415–424. Ira, G., and Haber, J.E. (2002). Characterization of RAD51-independent break-induced replication that acts preferentially with short homologous sequences. Mol. Cell. Biol. 22, 6384–6392. Johnson, R.D., and Symington, L.S. (1995). Functional differences and interactions among the putative RecA homologs Rad51, Rad55, and Rad57. Mol. Cell. Biol. 15, 4843–4850. Kiianitsa, K., Solinger, J.A., and Heyer, W.-D. (2002). Rad54 protein exerts diverse modes of ATPase activity on duplex DNA partially and fully covered with Rad51 protein. J. Biol. Chem. 277, 46205–46215. Krejci, L., Song, B., Bussen, W., Rothstein, R., Mortensen, U.H., and Sung, P. (2002). Interaction with Rad51 is indispensable for recombination mediator function of Rad52. J. Biol. Chem. 277, 40132–40141. Lee, S.E., Moore, J.K., Holmes, A., Umezu, K., Kolodner, R., and Haber, J.E. (1998). Saccharomyces Ku70, Mre11/Rad50 and RPA proteins regulate adaptation to G2/M arrest DNA damage. Cell 94, 399–409.

References

Lisby, M., Rothstein, R., and Mortensen, U.H. (2001). Rad52 forms DNA repair and recombination centers during S phase. Proc. Natl. Acad. Sci. USA 98, 8276–8282.

Bai, Y., and Symington, L.S. (1996). A RAD52 homolog is required for RAD51-independent mitotic recombination in Saccharomyces cerevisiae. Genes Dev. 10, 2025–2037.

Lovett, S.T., and Mortimer, R.K. (1987). Characterization of null mutants of the RAD55 gene of Saccharomyces cerevisiae: effects of temperature, osmotic strength and mating type. Genetics 116, 547–553.

Benson, F.E., Baumann, P., and West, S.C. (1998). Synergistic actions of Rad51 and Rad52 in recombination and DNA repair. Nature 391, 401–404. Bianco, P.R., Tracy, R.B., and Kowalczykowski, S.C. (1998). DNA strand exchange proteins: a biochemical and physical comparison. Front. Biosci. 3, 570–603. Brenneman, M.A., Wagener, B.M., Miller, C.A., Allen, C., and Nickoloff, J.A. (2002). XRCC3 controls the fidelity of homologous recombination: roles for XRCC3 in late stages of recombination. Mol. Cell 10, 387–395. Connolly, B., White, C.I., and Haber, J.E. (1988). Physical monitoring of mating type switching in Saccharomyces cerevisiae. Mol. Cell. Biol. 8, 2342–2349. De Zutter, J.K., and Knight, K.L. (1999). The hRad51 and RecA proteins show significant differences in cooperative binding to singlestranded DNA. J. Mol. Biol. 293, 769–780. Evans, E., Sugawara, N., Haber, J.E., and Alani, E. (2000). The Saccharomyces cerevisiae Msh2 mismatch repair protein localizes to recombination intermediates in vivo. Mol. Cell 5, 789–799. Fortin, G.S., and Symington, L.S. (2002). Mutations in yeast Rad51 that partially bypass the requirement for Rad55 and Rad57 in DNA repair by increasing the stability of Rad51-DNA complexes. EMBO J. 21, 3160–3170.

Malkova, A., Ivanov, E.L., and Haber, J.E. (1996). Double-strand break repair in the absence of RAD51 in yeast: a possible role for break-induced DNA replication. Proc. Natl. Acad. Sci. USA 93, 7131–7136. Mazin, A.V., Bornarth, C.J., Solinger, J.A., Heyer, W.-D., and Kowalczykowski, S.C. (2000). Rad54 protein is targeted to pairing loci by the Rad51 nucleoprotein filament. Mol. Cell 6, 583–592. Meluh, P.B., and Koshland, D. (1997). Budding yeast centromere composition and assembly as revealed by in vivo cross-linking. Genes Dev. 11, 3401–3412. Moore, J.K., and Haber, J.E. (1996). Cell cycle and genetic requirements of two pathways of nonhomologous end-joining repair of double-strand breaks in Saccharomyces cerevisiae. Mol. Cell. Biol. 16, 2164–2173. Mortensen, U.H., Bendixen, C., Sunjevaric, I., and Rothstein, R. (1996). DNA strand annealing is promoted by the yeast Rad52 protein. Proc. Natl. Acad. Sci. USA 93, 10729–10734. New, J.H., Sugiyama, T., Zaitseva, E., and Kowalczykowski, S.C. (1998). Rad52 protein stimulates DNA strand exchange by Rad51 and replication protein A. Nature 391, 407–410. Ogawa, T., Yu, X., Shinohara, A., and Egelman, E.H. (1993). Similarity of the yeast RAD51 filament to the bacterial RecA filament. Science 259, 1896–1899.

Frank-Vaillant, M., and Marcand, S. (2002). Transient stability of DNA ends allows nonhomologous end joining to precede homologous recombination. Mol. Cell 10, 1189–1199.

Paˆques, F., and Haber, J.E. (1999). Multiple pathways of recombination induced by double-strand breaks in Saccharomyces cerevisiae. Microbiol. Mol. Biol. Rev. 63, 349–404.

Gasior, S.L., Wong, A.K., Kora, Y., Shinohara, A., and Bishop, D.K. (1998). Rad52 associates with RPA and functions with Rad55 and Rad57 to assemble meiotic recombination complexes. Genes Dev. 12, 2208–2221.

Parsons, C.A., Baumann, P., Van Dyck, E., and West, S.C. (2000). Precise binding of single-stranded DNA termini by human RAD52 protein. EMBO J. 19, 4175–4181.

Haber, J.E. (2002). Switching of Saccharomyces cerevisiae mating-

Pellicioli, A., Lee, S.E., Lucca, C., Foiani, M., and Haber, J.E. (2001). Regulation of Saccharomyces Rad53 checkpoint kinase during ad-

In Vivo Assembly of Rad51 during Recombination 219

aptation from DNA damage-induced G2/M arrest. Mol. Cell 7, 293–300.

ated by a Rad51-ssDNA nucleoprotein filament with polarity opposite to that of RecA. Cell 82, 453–461.

Petukhova, G., Stratton, S., and Sung, P. (1998). Catalysis of homologous DNA pairing by yeast Rad51 and Rad54 proteins. Nature 393, 91–94.

Sung, P., and Stratton, S.A. (1996). Yeast Rad51 recombinase mediates polar DNA strand exchange in the absence of ATP hydrolysis. J. Biol. Chem. 271, 27983–27986.

Petukhova, G., Van Komen, S., Vergano, S., Klein, H., and Sung, P. (1999). Yeast Rad54 promotes Rad51-dependent homologous DNA pairing via ATP hydrolysis-driven change in DNA double helix conformation. J. Biol. Chem. 274, 29453–29462.

Symington, L.S. (2002). Role of RAD52 epistasis group genes in homologous recombination and double-strand break repair. Microbiol. Mol. Biol. Rev. 66, 630–670.

Petukhova, G., Sung, P., and Klein, H. (2000). Promotion of Rad51dependent D-loop formation by yeast recombination factor Rdh54/ Tid1. Genes Dev. 14, 2206–2215. Ristic, D., Wyman, C., Paulusma, C., and Kanaar, R. (2001). The architecture of the human Rad54-DNA complex provides evidence for protein translocation along DNA. Proc. Natl. Acad. Sci. USA 98, 8454–8460.

Tan, T.L.R., Kanaar, R., and Wyman, C. (2003). Rad54, a jack of all trades in homologous recombination. DNA Repair 16, 787–794. Umezu, K., Sugawara, N., Chen, C., Haber, J.E., and Kolodner, R.D. (1998). Genetic analysis of yeast RPA1 reveals its multiple functions in DNA metabolism. Genetics 148, 989–1005. Van Dyck, E., Hajibagheri, N.M., Stasiak, A., and West, S.C. (1998). Visualisation of human Rad52 protein and its complexes with hRad51 and DNA. J. Mol. Biol. 284, 1027–1038.

Roca, A.I., and Cox, M.M. (1997). RecA protein: structure, function, and role in recombinational DNA repair. Prog. Nucleic Acid Res. Mol. Biol. 56, 129–223.

Van Dyck, E., Stasiak, A.Z., Stasiak, A., and West, S.C. (1999). Binding of double-strand breaks in DNA by human Rad52 protein. Nature 398, 728–731.

Schmuckli-Maurer, J., and Heyer, W.-D. (1999). The Saccharomyces cerevisiae RAD54 gene is important but not essential for natural homothallic mating-type switching. Mol. Gen. Genet. 260, 551–558.

Van Komen, S., Petukhova, G., Sigurdsson, S., Stratton, S., and Sung, P. (2000). Superhelicity-driven homologous DNA pairing by yeast recombination factors Rad51 and Rad54. Mol. Cell 6, 563–572.

Shinohara, A., and Ogawa, T. (1998). Stimulation by Rad52 of yeast Rad51-mediated recombination. Nature 391, 404–407.

Vaze, M.B., Pellicioli, A., Lee, S.E., Ira, G., Liberi, G., Arbel-Eden, A., Foiani, M., and Haber, J.E. (2002). Recovery from checkpointmediated arrest after repair of a double-strand break requires Srs2 helicase. Mol. Cell 10, 373–385.

Shinohara, A., Ogawa, H., and Ogawa, T. (1992). Rad51 protein involved in repair and recombination in S. cerevisiae is a RecA-like protein. Cell 69, 457–470. Signon, L., Malkova, A., Naylor, M., and Haber, J.E. (2001). Genetic requirements for RAD51- and RAD54-independent break-induced replication repair of a chromosomal double-strand break. Mol. Cell. Biol. 21, 2048–2056. Solinger, J.A., and Heyer, W.-D. (2001). Rad54 protein stimulates the postsynaptic phase of Rad51 protein- mediated DNA strand exchange. Proc. Natl. Acad. Sci. USA 98, 8447–8453. Solinger, J.A., Lutz, G., Sugiyama, T., Kowalczykowski, S.C., and Heyer, W.-D. (2001). Rad54 protein stimulates heteroduplex DNA formation in the synaptic phase of DNA strand exchange via specific interactions with the presynaptic Rad51 nucleoprotein filament. J. Mol. Biol. 307, 1207–1221. Solinger, J.A., Kiianitsa, K., and Heyer, W.-D. (2002). Rad54, a Swi2/ Snf2-like recombinational repair protein, disassembles Rad51: dsDNA filaments. Mol. Cell 10, 1175–1188. Song, B., and Sung, P. (2000). Functional interactions among yeast Rad51 recombinase, Rad52 mediator, and replication protein A in DNA strand exchange. J. Biol. Chem. 275, 15895–15904. Strahl-Bolsinger, S., Hecht, A., Luo, K., and Grunstein, M. (1997). SIR2 and SIR4 interactions differ in core and extended telomeric heterochromatin in yeast. Genes Dev. 11, 83–93. Sugawara, N., Ivanov, E.L., Fishman, L.J., Ray, B.L., Wu, X., and Haber, J.E. (1995). DNA structure-dependent requirements for yeast RAD genes in gene conversion. Nature 373, 84–86. Sugiyama, T., and Kowalczykowski, S.C. (2002). Rad52 protein associates with RPA-ssDNA to accelerate Rad51-mediated displacement of RPA and presynaptic complex formation. J. Biol. Chem. 277, 31663–31672. Sugiyama, T., New, J.H., and Kowalczykowski, S.C. (1998). DNA annealing by Rad52 protein is stimulated by specific interaction with the complex of replication protein A and single-stranded DNA. Proc. Natl. Acad. Sci. USA 95, 6049–6054. Sung, P. (1994). Catalysis of ATP-dependent homologous DNA pairing and strand exchange by yeast Rad51 protein. Science 265, 1241– 1243. Sung, P. (1997a). Function of yeast Rad52 protein as a mediator between replication protein A and the Rad51 recombinase. J. Biol. Chem. 272, 28194–28197. Sung, P. (1997b). Yeast Rad55 and Rad57 proteins form a heterodimer that functions with replication protein A to promote DNA strand exchange by Rad51 recombinase. Genes Dev. 11, 1111–1121. Sung, P., and Robberson, D.L. (1995). DNA strand exchange medi-

Wach, A., Brachat, A., Pohlmann, R., and Philippsen, P. (1994). New heterologous modules for classical or PCR-based gene disruptions in Saccharomyces cerevisiae. Yeast 10, 1793–1808. White, C.I., and Haber, J.E. (1990). Intermediates of recombination during mating type switching in Saccharomyces cerevisiae. EMBO J. 9, 663–673. Wolner, B., Van Komen, S., Sung, P., and Peterson, C. (2003). Recruitment of the recombinational repair machinery to a DNA double strand break in yeast. Mol. Cell 12, this issue, 221–232.