Members of the RAD52 Epistasis Group Contribute to Mitochondrial Homologous Recombination and Double-Strand Break Repair in Saccharomyces cerevisiae

Mitochondria contain an independently maintained genome that encodes several proteins required for cellular respiration. Deletions in the mitochondrial genome have been identified that cause several maternally inherited diseases and are associated with certain cancers and neurological disorders. The majority of these deletions in human cells are flanked by short, repetitive sequences, suggesting that these deletions may result from recombination events. Our current understanding of the maintenance and repair of mtDNA is quite limited compared to our understanding of similar events in the nucleus. Many nuclear DNA repair proteins are now known to also localize to mitochondria, but their function and the mechanism of their action remain largely unknown. This study investigated the contribution of the nuclear double-strand break repair (DSBR) proteins Rad51p, Rad52p and Rad59p in mtDNA repair. We have determined that both Rad51p and Rad59p are localized to the matrix of the mitochondria and that Rad51p binds directly to mitochondrial DNA. In addition, a mitochondrially-targeted restriction endonuclease (mtLS-KpnI) was used to produce a unique double-strand break (DSB) in the mitochondrial genome, which allowed direct analysis of DSB repair in vivo in Saccharomyces cerevisiae. We find that loss of these three proteins significantly decreases the rate of spontaneous deletion events and the loss of Rad51p and Rad59p impairs the repair of induced mtDNA DSBs.


Introduction
Mitochondria are dynamic organelles that function as the primary source of ATP. Mitochondria contain an autonomously replicating genome that is packaged into DNA-protein structures called nucleoids. This multi-copy genome encodes subunits of electron transport chain (ETC) complexes and the rRNAs and tRNAs that are specifically required for the expression of these proteins. The number and types of proteins encoded by the mitochondrial genome are highly conserved among eukaryotes, even though the overall size and structure of the genome varies widely between species. The remaining proteins required for the ETC, as well as those required for maintenance and repair of the mitochondrial genome are encoded in the nucleus and imported into mitochondria from the cytoplasm.
Stability of the mitochondrial genome is required for the survival of most eukaryotic cells. Mutations in the mitochondrial genome cause metabolic defects that underlie multiple diseases, both inherited and spontaneous [1][2][3][4][5][6]. Accumulation of both point mutations and deletions in mtDNA are also proposed to contribute to the normal aging process [7]. Consistent with this hypothesis, transgenic mice expressing a proofreading-defective Pol γ, the replicative mitochondrial polymerase, display accelerated aging phenotypes due to the accumulation of somatic mtDNA mutations [8]. Despite its profound importance to metabolic control and cell maintenance, the mechanisms of mitochondrial mutagenesis and repair are poorly understood, particularly double-strand break repair (DSBR) [9]. Though homologous recombination (HR) in yeast, plant and mammalian mtDNA has been reported [10][11][12][13][14][15], few specific proteins involved in mitochondrial recombination have been identified [16][17][18][19][20][21][22]. For those proteins that have been implicated in mitochondrial HR, the inability to generate and visualize specific, controlled mitochondrial double-strand breaks has precluded a direct analysis of their role in DSBR.
Members of the RAD52 epistasis group are essential for HR in the nucleus. Rad51p, a RecA homologue, forms a nucleoprotein filament that promotes the homologous pairing and strand exchange required for DSBR, synthesis dependent strand annealing and break induced replication [23][24][25]. Rad52p is required for almost all forms of HR in yeast, by promoting the exchange of RPA for Rad51p on the single-stranded ends that are generated at DSBs following resection [26]. HR is often thought of as an error-free repair pathway but there are several HR mechanisms that can lead to the generation of deletions between repetitive elements. For example single strand annealing (SSA), which is highly dependent on Rad52p and Rad59p, is an error-prone recombination pathway that occurs when a DSB or a lesion that results in a DSB arises between two repetitive sequences [27]. The annealing of these repetitive sequences after resection leads to the deletion of one repeat and the intervening sequence (Fig 1) [26,28]. A DSB that arises within a repeat can also generate a deletion if the repetitive sequences are misaligned during unequal crossing over (Fig 1) [29]. Deletions may also arise between repeated sequences due to errors in replication such as slippage or template switching (Fig 1) [30].
Deletions have been observed in yeast, plants, flies, and mammalian mtDNA [10,11,31,32], and often these deletions involve sequences originally flanked by direct repeats. For example, in human cells nearly 90% of deletions in mtDNA are flanked by either perfect or imperfect repeats. This suggests that recombination is a possible mechanism for the generation of mtDNA deletions, but the exact mechanisms and proteins involved are currently unknown. [33,34].
Recent studies have localized members of the RAD52 epistasis group to mitochondria. In plants, a mitochondrial-specific isoform of Rad52p has been identified, and demonstrated to promote annealing of complementary DNA sequences in vitro, consistent with a role in recombination in mitochondria [35]. In human cell lines, Rad51, and two of its paralogs Rad51C and XRCC3 were found to localize to mitochondria and promote mtDNA replication upon oxidative stress [36,37]. These studies strongly suggest that at least a subset of the nuclear recombination proteins can localize to mitochondria and that once there, they may facilitate processes that require homologous sequences such as HR. Mammalian systems do not lend themselves to mechanistic studies because these organisms require intact mtDNA molecules to survive. In addition, it is not currently possible to introduce exogenous DNA reporters in vivo in these systems. Saccharomyces cerevisiae is an ideal model system for these studies due to the fact that these repair proteins are highly conserved, mtDNA is not required for cell survival, and it is possible to introduce exogenous reporter constructs directly into the mitochondrial genome.
We previously developed a reporter system for quantitatively measuring the occurrence of direct repeat mediated deletions (DRMD) in mtDNA [38][39][40]. This reporter introduces a unique KpnI restriction recognition site between direct repeats. By expressing mitochondriallytargeted KpnI, induction of a directed mtDNA DSB can be precisely controlled. In the work presented here, we use a modified version of this endonuclease fusion to demonstrate that loss Rad51p, Rad52p and Rad59p significantly decreases the generation of both spontaneous and DSB induced mitochondrial DRMDs. Improved control of DSB induction has also allowed us to study the kinetics of mtDNA DSBR using both genetic and molecular methods. Southern blot analysis confirms that loss of these proteins leads to delays in break processing and/or generation of deletion products. The localization of Rad51p and Rad59p to yeast mitochondria Models for the generation of deletions between directly repeated sequences. Repair of a doublestrand break (DSB) located between directly repeated sequences (grey boxes) can result in the deletion of one of the repeats and all of the intervening sequence. 5' to 3' resection at the DSB reveals the direct repeats. Rad52p alone or in conjunction with Rad59p, promotes the annealing of the repeats. Template switching can occur if a lesion (yellow star) is encountered during replication. At a stalled fork, the nascent strand may invade at the incorrect repeat, leading to the generation of a deletion. Unequal exchange can occur when HR occurs between misaligned repeats leading to a deletion. Both template switching and unequal exchange can happen intramolecularly or intermolecularly. Arrowheads indicate 3' end. argue that these effects are direct. Analysis of double mutants indicates that the contribution of these proteins to the repair process differs between the mitochondrial and the nuclear genomes.

Rad51p and Rad59p localize to the mitochondrial matrix
Previous studies have shown that Rad51, Rad51C and XRCC3 are found in the mitochondria of human cells [36]. In order to determine if this dual localization of Rad51 to both the nucleus and mitochondria is conserved in yeast, mitochondria were purified from cells expressing a Cterminal Rad51-HA tag expressed from the endogenous locus. We were able to detect at least four Rad51-HA specific bands in whole cell extracts and associated with the mitochondrial fraction (Fig 2A). These bands are specific to the strain expressing the Rad51-HA as they cannot be detected in whole cell extracts or mitochondrial extracts from an untagged control strain (S3 Fig). In order to determine the precise mitochondrial localization a protease protection assay was performed. Isolated mitochondria were subjected to proteinase K treatment in the presence or absence of 20mM Hepes, pH 7.4 to remove the outer mitochondrial membrane, or Triton X-100 to rupture both membranes. The inter membrane space protein, cytochrome b (Cob), is sensitive to digestion when the mitochondria are treated with 20mM Hepes (Fig 2A). The~55 kDa band of Rad51p is also sensitive to digestion when the mitochondria are treated with 20mM Hepes suggesting this isoform is also found in the inter membrane space (Fig 2A,  lanes 1 and 2). It is possible this isoform is also found in the matrix but is not detectable with our current system. Citrate synthase (Cit1p), a matrix protein (Fig 2A) and the~60 kDa band of Rad51 remains protected from proteinase K unless treated with Triton X-100, confirming Rad51p's localization in the matrix of the mitochondria in yeast (Fig 2A, lane 4). The specific modification(s) responsible for the larger Rad51p mitochondrial isoform(s) is currently unknown. Mitochondria were also purified in the same manner from cells expressing Rad59-HA from the endogenous locus. A protease protection assay confirmed the presence of Rad59 in the matrix of the mitochondria (Fig 2A and S3 Fig).
In order to further confirm the mitochondrial localization of Rad51p and to determine if it binds directly to mtDNA, chromatin immunoprecipitation (ChIP) was performed using an antibody to the native untagged Rad51 protein. Using primers that anneal to a region previously demonstrated to be near a recombination hotspot in the yeast mitochondrial genome, we were able to detect a significant 4-fold (p = 0.008) increase in mtDNA signal in the Rad51p IP compared to the mock IP ( Fig 2B) [41]. The western blot and ChIP data together clearly demonstrate the Rad51p is localized to the mitochondria of S. cerevisiae.

Rad51p, Rad52p, and Rad59p promote spontaneous mitochondrial deletions
Rad51p, Rad52p, and Rad59p participate in spontaneous DRMD events and the repair of induced DSBs in the nucleus [26]. To determine whether they also contribute to spontaneous mitochondrial DRMD, we generated deletions of these genes both individually and in combination, in yeast strains containing genetic reporters for nuclear and mitochondrial DRMD events [38][39][40].
The mitochondrial reporter consists of the ARG8 m gene, a mitochondrial derivative of the nuclear ARG8 gene that has been recoded to reflect the codon usage and bias of a mitochondrial gene [42]. The ARG8 m gene is inserted 99 bp into the COX2 gene followed by the entire COX2 gene lacking the start codon ( Fig 3B). This generates 96 bp of directly repeated sequence flanking ARG8 m . Strains that contain an intact reporter are phenotypically Arg + and respiration deficient. When a mitochondrial DRMD event occurs, the strain becomes respiration proficient and will become Argfollowing sorting to homoplasmy, which occurs rapidly in yeast. The nuclear reporter is similarly constructed, with the URA3 gene inserted 99 bp into the TRP1 gene, followed by the entire TRP1 gene, generating 96bp of directly repeated sequence (Fig 3A) [39]. Strains with an intact reporter are phenotypically Ura + and Trp -. If a nuclear DRMD event occurs, the strain becomes Uraand Trp + . Plating of cells on the appropriate media allows selection of both mitochondrial and nuclear deletions from the same culture.
As we have previously shown, the mitochondrial deletion rate is approximately160-fold greater than the nuclear rate for similar repeats in wild-type cells. It is important to note that nuclear and mitochondrial rates should not be compared directly, since the multicopy nature of the mitochondrial genome, and its subsequent replication and segregation into daughter cells will impact the rate. Rates within each compartment are compared between wild-type and mutant strains. Deletion of RAD52 or RAD59 resulted in similar decreases of both nuclear and mitochondrial DRMDs (Fig 3C and 3D). Spontaneous nuclear and mitochondrial DRMDs decreased 3-fold in the rad52-Δ strain (p = 0.001, in each case) and 2-fold in the rad59-Δ strain (nuclear p = 0.001, mitochondrial p = 0.002) relative to wild-type. This suggests that the contribution of Rad52p and Rad59p to the generation of spontaneous deletions is equally as important in both the nucleus and mitochondria. In contrast, RAD51 deletion resulted in different nuclear and mitochondrial phenotypes, with a 6-fold increase in nuclear DRMD (p = 2 x 10 -8 ) and an approximately 2-fold decrease in mitochondrial DRMD (p = 0.01) (Fig 3C and 3D). The nuclear spontaneous DRMD results for this strain are consistent with the current model of SSA, however, the decrease measured for mitochondrial DRMD is not. According to the current nuclear model, SSA is dependent on Rad52p and Rad59p and independent of Rad51p. Deletions generated by SSA increase in the absence of accurate repair mediated by Rad51p [28,43]. Thus, the observed decreases in nuclear DRMD events in the rad52-Δ, rad59-Δ, and rad51-Δ rad52-Δ strains, and the increase in nuclear DRMD events in rad51-Δ strains were as expected [27,28]. The decrease in mitochondrial DRMD events observed in the rad51 null mutants indicates that in contrast to the nuclear DRMD events, spontaneous mitochondrial DRMDs are at least partially dependent on Rad51p, suggesting a unique functional role for Rad51p in mitochondria.
To confirm that the decreases in mitochondrial DRMD rates in the rad mutants are due to specific deficiencies in mitochondrial pathways leading to deletions, and not due to general mitochondrial genome instability, the spontaneous frequency of petites was measured in each of the above mutants. Spontaneously arising petite colonies (ρ -) are non-respiring, usually due to large-scale deletions and rearrangements of mtDNA, and do not grow on non-fermentable carbon sources such as glycerol [10]. The mechanism by which these deletions and rearrangements occur has not been defined, although homologous recombination has been proposed [10]. None of the deletions we tested exhibited any significant change from wild-type in the frequency of respiration loss (S2 Table). This suggests that the phenotypes we observe do not result from a general increase in mtDNA instability, and these proteins are not involved in the mechanism that generates spontaneous petite colonies.

Specific mitochondrial double-strand breaks increase the frequency of mtDNA deletions
The ARG8 m sequence used in our mitochondrial reporter contains a unique KpnI restriction site. Cleavage at this site generates a specific DSB 370 bp downstream of the ARG8 m start codon, with 4 bp 3' single-stranded complementary overhangs at the broken ends ( Fig 4A). We previously reported that the galactose-inducible promoter regulating the expression of our mtLS-KpnI was leaky, which stimulated the generation of deletion products prior to the addition of galactose [40]. In order to ensure that we did not have premature expression of KpnI endonuclease, a temperature sensitive intein was inserted into the active site of KpnI [44]. This intein self-cleaves at 20°C generating an active KpnI enzyme. A control strain was also constructed utilizing a nonfunctional intein (int dead ) that cannot self-cleave at any temperature [44].
This modification allows tight control of the activity of mtLS-KpnI. Southern blot analysis shows no detectible digestion at the mitochondrial KpnI recognition sequence, and no evidence of deletion product prior to induction of mtLS-KpnI-int ts ( Fig 4B, lane P). KpnI cleavage is achieved by inducing mtLS-KpnI-int ts expression with galactose and reducing the incubation temperature to 20°C (Fig 4B, lane 0). Furthermore, the frequency of DSB-induced DRMDs increases linearly with extended incubation time at 20°C prior to assaying at 30°C, suggesting that KpnI activity is largely restricted to 20°C (S1 Fig). Longer incubations at 20°C generated a larger proportion of DSBs, but never exceeded 60% of the total COX2 signal. This result may suggest that some genomes may not be accessible to the KpnI enzyme or that alternative repair processes are rapidly repairing the break (S4 Fig). In subsequent experiments, 16-18 hours incubation at 20°C was used to ensure significant visible DSBs were generated.
Quantification of COX2-hybridizing species from independent experiments reveals that an average of 30.2 ± 16.7% of the COX2 signal appears as a specific KpnI digested fragment after 16-18 hours incubation at 20°C, and no measurable deletions are observed (Fig 4B and 4D). Two hours after cultures are returned to 30°C, deletion products can be observed, and their concentration continues to increase over the course of the experiment to represent, on average, 16.2% of the COX2 signal (Fig 4B and 4D). Thus, the percent of the COX2 signal that hybridizes to DRMD products after 8 hours at 30°C is approximately equivalent to two-thirds of the amount observed in the DSB at time 0. By 8 hours after shift to 30°C the cultures have reached saturation, and no further increase in deletion product was seen when cultures were monitored for an additional 4 hours (S4 Fig), suggesting that repair is complete by 8 hours in the wildtype strain. In addition, while breaks are stimulated at 20°C, the bulk of DRMD events occur after the cells are shifted back to 30°C, since deletions are not detected by Southern blot analysis at time of shift back to 30°C. In control cultures, expressing the mtLS-KpnI-int dead , no stimulation of DRMD was observed, indicating that no significant cleavage of the int dead occurs.
We included two additional probes in our Southern blot analysis to quantify potential changes in mtDNA content relative to nuclear DNA. The first, a mtDNA probe, hybridizes to the 21S rRNA gene approximately 10 kb from the site of the induced DSB. The second is a nuclear probe that hybridizes to the 25S rRNA gene. These probes were used to measure the ratio of mitochondrial to nuclear DNA during the course of these experiments. We find that shifting the cells to 20°C caused a slight decrease in mtDNA content relative to the nuclear control in both the mtLS-KpnI::int ts strain and mtLS-KpnI::int dead strain indicating that this decrease is due to the temperature shift and not the expression of active KpnI (Fig 4C). The mtLS-KpnI::int ts expressing strains recovered significantly faster than the control strain, however, both the control and mtLS-KpnI::int ts strains fully recovered to pre-induction levels of mtDNA after eight hours of growth at 30°C (Fig 4C).
It has been reported that mitochondrial DNA can move to the nucleus, although this occurs rarely [45]. In order to ensure that the signal we are detecting in the Southern blots is originating from the mitochondrial genome, both pre-and post-induction wild type cultures were treated with ethidium bromide. Growth of yeast in culture medium in the presence of ethidium bromide has been demonstrated to suppress mtDNA replication [46]. Southern blot analysis confirms that greater than 98% of the signal from the COX2 and the 21S rRNA probes is lost when the cells are subjected to ethidium bromide treatment. (S2 Fig).
Eight hours after shifting to 30°C, approximately 16% of the mtDNA detected contains a deletion of the ARG8 m reporter (Fig 4B and 4D). However, the distribution of the 16% of deleted genomes within the population of cells is unclear. For example, many deletions may be concentrated in a small subset of the cell population, or distributed throughout, with most cells containing one or more mitochondrial genomes with deletions. By assaying for respiring colonies, we can directly distinguish between these possibilities. Cells from time 0 in the above experiments ( Fig 4B) were plated on non-selective medium, allowed to grow, and individual colonies were tested for the presence of the intact reporter (Arg + ) or the deletion (respiration proficient) by replica plating. We found that 50% of the population is respiration competent, consistent with a wide distribution of mitochondrial genomes containing deletions (Fig 5). The majority of these colonies also contain Arg + cells, indicating heteroplasmy at the time of (B) Representative Southern blot of AvaII-digested DNA extracted from EAS748 containing pEAS115 (mtLS-KpnI-intein dead ) or pEAS114 (mtLS-KpnI-intein ts ). Both strains were grown in SRaffinose-Arg-Ura until OD 600 = 0.150. Galactose was added to a final concentration of 2%, and the strains were shifted to 20°C to induce the either mtLS-KpnI-intein ts or mtLS-KpnI-intein dead . Genomic DNA in lanes labeled P are from preinduced cultures. Cultures were incubated for 16-18 hours at 20°C and samples were taken for the time 0 timepoint. The cultures then were shifted to 30°C to allow repair. Lane I contains DNA from the t = 0 timepoint of the mtLS-KpnI-intein ts -containing strain digested in vitro with KpnI and AvaII to demonstrate the migration of DNA with a DSB. The 21S rRNA gene was probed to detect total mtDNA. The 25S rRNA gene was probed to detect total nuclear DNA. (C) Relative mtDNA content was determined by taking the ratio of the 21S rRNA signal and the 25S rRNA signal, then normalizing to the t = 0 sample for each strain. (D) The intensity of the COX2-hybridizing bands were measured using Image Lab software (www.biorad.com) The proportions of the total COX2 signal present in the DSB and deletion products were calculated as a percent of the total COX2 signal. plating. Significantly, we did not observe DSB stimulation of ρor ρ 0 (lacking mtDNA) petites, which would be both Argand non-respiring. After cutting, 1.5% of the colonies from wildtype cells were both Argand respiration deficient. This is comparable to the percentage of spontaneous petites seen in wild-type cultures prior to KpnI-induced mtDNA DSBs, indicating the induction of the DSB does not lead to general mitochondrial genome instability (S3 Table).

Rad51p, Rad52p, and Rad59p influence mitochondrial DSBR
To determine the effect of Rad51p, Rad52p, and Rad59p on the repair of mitochondrial DSBs, plasmids expressing either mtLS-KpnI::int ts (pEAS114) or mtLS-KpnI::int dead (pEAS115) were introduced into the mutant strains. Mitochondrial DSBs were induced as described above, and Southern blots reveal that the level of cutting by mtLSKpnI-int ts at the time of shift to 30°C in all of the mutant strains is equivalent to wild-type. (Table 1).
In our genetic assay for induced deletions, 50% of colonies from wild-type cells had undergone at least one selectable deletion event (Fig 5), and each of the RAD single mutants exhibited a modest, yet significant, decrease in selectable deletion events when compared to wild-type. In the rad51-Δ, rad52-Δ, rad59-Δ strains 30% (p = 4.2 x 10 -5 ), 31% (p = 0.03), and 36% (p = 0.03)  of the cells were able to give rise to respiring colonies respectively, after 16-18 hours of inducing DSBs (Fig 4). As with the wild-type strain, non-respiring and Argpetites did not significantly increase after DSB induction in these strains (S3 Table). Southern blots analysis of the mtDNA in rad51-Δ and rad59-Δ strains revealed that the same proportions of the mitochondrial genomes are being cut (31% and 25%, respectively) compared to the wild-type strain (Table 1). However, consistent with the genetic data, these cut genomes are not converted to deletion products to the same extent as wild-type (Figs 6A  and 6C and 7). In the rad51-Δ strain after 8 hours of recovery only 14.4% ± 7.9 (p = 0.001) of the cut COX2 signal had been converted to a deletion product compared to 66.8% ± 7.4 in the wild-type strain. In the rad59-Δ strain, 36.7% ± 14.9 (p = 0.01) of the cut COX2 signal is converted to a deletion product. The cut band is completely lost in all mutant and wild-type strains after 8 hours of recovery, suggesting that DSB processing is not being eliminated but the pathway choice is altered in the absence of Rad51p or Rad59p.
Analysis of the rad52-Δ strain revealed no significant change in cut COX2 signal at time 0, 40.6% ± 2 (p = 1.0) compared to wild-type (Table 1). Surprisingly, in contrast to the genetic data, the proportion of the cut COX2 signal converted to deletion product, 55.9% ± 26.3 (p = 0.5) is equivalent to wild-type. We know that the distribution or maintenance of these genomes in the population of cells is not equivalent to wild-type because only 31% of the colonies contain respiring cells, while 50% of wild-type colonies contain respiration-competent cells after 3 days growth on solid medium.
The rad51-Δ rad52-Δ double mutant exhibited the same decrease in deletion events as the rad51-Δ and rad52-Δ single mutants, with 29% (p = 4.5 x 10 -5 ) of colonies capable of respiration after induction of mtLS-KpnI::int ts (Fig 5). Interestingly, the mtDNA in this mutant displayed a similar significant decrease in the proportion of cut COX2 signal converted to deletion product, 7.6% ± 5.7 (p = 0.0004), as the rad51-Δ strain (p = 0.3) (Figs 6D and 7). This would suggest that during the initial generation of deletion products after induction of a DSB that RAD51 is epistatic to RAD52 in the mitochondria.
We considered the possibility that the relative levels of the different COX2-hybridizing species in all cases might be affected by differences in growth rates of the mutant and wild-type strains under the conditions of the experiment. Monitoring the growth of the cultures by optical density and total cell counts throughout the protocol revealed that the rad52-Δ and rad59-Δ strains displayed a 20% increase in doubling time at 20°C, while the rad51-Δ mutant strain was comparable to the wild-type. All strains undergo one population doubling during the eighthour recovery period. It is not likely then that the wild-type levels of deletion product generated in the rad52-Δ strain are dependent on this growth defect, considering the rad59-Δ strain also has inhibited growth at 20°C but clearly shows impaired deletion product formation. We believe there is no significant correlation of differences in growth rate with differences in DSBs or deletions (Fig 6B).

Single-stranded DNA is present at break after induction of KpnI
In the nucleus, the resection of the 5' ends at a DSB commits the cell to repairing the lesion by HR. In order to determine if the DSBs detected by Southern blot in mitochondria are being processed, such that 3' ssDNA is generated, the mtDNA was blotted and probed with a ssDNA probe that would anneal to 3' single-stranded overhangs approximately 200 nucleotides from the site of the break. All of the strains exhibited a significant 3-4.5 fold increase in ssDNA after 16 hours of cleavage ( Table 2). The increase in ssDNA at the site of a DSB suggests that the ends of mtDSBs are resected in a manner consistent with known nuclear HR models. The presence of similar levels of 3' ssDNA in the wild-type, rad51-Δ, rad52-Δ, rad59-Δ, and rad51-Δ rad52-Δ strains is also consistent with HR models that show these proteins function in the repair of DSBs after resection has occurred (Table 2).

Discussion
The faithful maintenance of the mitochondrial genome is required for eukaryotic cells to have a functional electron transport chain. Mutations and deletions in the mitochondrial genome can compromise energy production for the cell. However, studying mtDNA repair has been difficult, due to the necessity for intact mitochondrial genomes in most eukaryotic systems. Thus, the extent of mtDNA repair has remained unclear. For years it was believed that since there are multiple copies of the mitochondrial genome, the damaged mtDNA was simply degraded in lieu of repair [47]. While there is evidence that mitochondrial genomes are degraded under high levels of chronic damage, it is unclear if this is the response under more moderate damaging conditions [48].
Mitochondrial base excision repair has been more extensively studied [9,49,50], indicating that at least some highly conserved repair pathways exist in mitochondria. In addition, there is evidence for homologous recombination (HR) repair pathways in mitochondria of several different species including yeast, Drosophila, and mice [40,51,52]. In this study, we have shown that the HR proteins Rad51p and Rad59p localize to the matrix of the mitochondria. We have also demonstrated that Rad51p, Rad52p and Rad59p impact the rate of spontaneous mitochondrial DNA deletions and the repair of induced mitochondrial DSBs in yeast. This is the first study to demonstrate that Rad51p, Rad52p, and Rad59p impact the repair of specific mtDNA lesions.
In the present study, we have greatly improved our reporter system to introduce a maximum of a single DSB per mitochondrial genome to identify the proteins involved in the repair of these breaks. We had previously generated a plasmid expressing a galactose-inducible, mitochondrially-targeted KpnI endonuclease that showed evidence of expression under uninduced conditions [40]. To add an additional level of control over endonuclease activity, we have inserted a temperature-sensitive intein into the active site of the KpnI enzyme. This modified expression construct does not result in efficient cleavage of the mitochondrial genome, since we must incubate at the permissive temperature for 16-18 hours to observe digestion of approximately 30% of the mtDNA by Southern blot analysis (Fig 3). However, it is likely that a larger proportion of breaks are present, but these breaks are not detectible due to the end resection that occurs at the site of DSBs. We have demonstrated that 3' single-stranded ends are present at the site of the break following cleavage. The heterogeneous nature of these intermediates would render them undetectable by our current Southern blot techniques. We would also be unable to detect any breaks that occurred and were subsequently ligated with fidelity. Conversion of DSBs to deletion product is impaired in HR mutants. Experiments were performed as in Fig 4. The ratios of the amount of COX2 signal in the cut band at t = 0 and the amount of COX2 signal in the deletion product at t = 8 were calculated. The intensity of the COX2-hybridizing bands was measured using Image Lab software (www.biorad.com). The proportions of the total COX2 signal present in the DSB and deletion products were calculated as a percent of the total COX2 signal. Error bars indicate SD. Asterisks indicate significant differences between the mutant and wild-type rates (* = p 0.05, *** = p 0.001).  This system is an improvement over our previously published mtLS-KpnI construct, as there is no molecular or genetic evidence of mtDNA cleavage prior to induction, nor is there selection for cleavage-resistant genomes in our cultures. In addition, this level of digestion is sufficient to observe DNA repair over the course of 8 hours after shifting back to the non-permissive temperature for intein cleavage. It bears noting that deletion products do not appear after extensive incubation at the permissive temperature for intein cleavage, although they are clearly visible by 2 hours after shift back to the non-permissive temperature. It is unlikely that this delay simply results from slow repair, because no deletions are visible on a Southern blot after 24 hours of incubation at the permissive temperature (S1 Fig). Rather, this observation suggests that at least one step of the deletion process is inefficient or non-functional at 20°C. Deletion formation does not appear to be the only temperature-dependent process. We observed a slight decrease in relative mtDNA copy number upon shifting the cultures to 20°C. However, a decrease was also seen in our controls expressing mtLS-KpnI-int dead , leading us to conclude that temperature shift alone is cause of this temporary decrease in mtDNA copy number (Fig 4C). Consistent with this hypothesis, both strains recover by 8 hours after shift back to 30°C. We conclude that in our system, repairable DSBs do not stimulate mtDNA loss from the population of cells. Our genetic assay supports this observation, as induction of a specific mitochondrial DSB does not stimulate an increase in the loss of functional mtDNA. Therefore, mechanisms exist in cells to maintain functional mtDNA. It is important to note that this selection cannot be applied at the level of respiratory function, or ATP production, as the genomes containing the intact DRMD reporter confer an Arg + phenotype, and have active mitochondrial protein synthesis, but do not assemble a functional electron transport chain. In yeast, there is evidence for a mitochondrial G1-to-S phase checkpoint that is activated in response to the loss of mtDNA and not linked to respiratory capacity [53], suggesting that the cell has a mechanism for sensing mtDNA. Clearly, although mtDNA can be eliminated by treatment of cells with ethidium bromide and spontaneous petite mutants arise in yeast cultures [10] this should not be taken as evidence that mechanisms do not exist to maintain an intact genome.
Our results demonstrate that Rad51p, Rad52p, and Rad59p are involved in a portion of DSBR events. In the nucleus, SSA is the primary HR pathway that generates DSB-induced deletions between repeated sequences [26], and it seems likely to be an important pathway for spontaneous events at direct repeats as well. This mutagenic DSB repair pathway leads to the deletion of one of the repeats and any intervening sequence. In the nucleus, SSA is independent of Rad51p, but requires Rad52p and Rad59p [43]. Previous studies have shown [26,27], and this study confirms, that in the nucleus, deletion of RAD51 leads to an increase in deletions generated by a Rad52p-dependent pathway, likely SSA (Fig 3B). In contrast, in mitochondria, the deletion of RAD51 caused a significant decrease in the frequency of spontaneous deletions and those generated after inducing a DSB. Genetically, the contribution of Rad51p to the generation of DRMDs was equivalent to Rad52p and Rad59p (Figs 3D and 5). This would imply that both compartments use at least some of the same proteins to generate these deletions, but that the pathways these proteins are participating in are unique compared to the nucleus. In the absence of the full suite of nuclear HR proteins in the mitochondria we speculate that HR proteins, such as Rad51, have been co-opted to perform their HR functions in contexts that are not necessary in the nucleus.
In regard to the generation of spontaneous deletions, one possible explanation is that the initiating event that generates spontaneous mtDNA deletions is a replication error such as slippage, or a stalled fork that triggers template switching, and thus relies on a different suite of proteins, but is still dependent on the presence of homologous sequences and thus requires HR proteins, such as Rad51p. In our assay where we induced a DSB, an explanation for the increased dependence on Rad51p for deletion formation, is that the AT-rich mitochondrial genome requires further stabilization of the annealed intermediate and thus utilizes Rad51p uniquely in this manner. However, at this time, we cannot rule out the existence of a novel Rad51p dependent mechanism that generates deletions solely in the mitochondria.
The localization of Rad51p and Rad59p to mitochondria strongly suggests that the effects we observe on deletion product formation in the mutant strains are direct, for several reasons. First, we demonstrated that mitochondrial DSB repair and homology-dependent deletions were specifically affected, while mitochondrial genomes were stably maintained, as indicated by the maintenance of wild-type copy number and the lack of additional ρin our mutant strains. In addition, the epistatic relationships between RAD51, RAD52, RAD59 with respect to mitochondrial phenotypes were not the same as the nuclear phenotypes, most notably in the case of Rad51p, which had a similar impact on mitochondrial DRMD's as both Rad52p and Rad59p. Finally, their action at DSBs is consistent with their known function. Identification of mitochondrial-specific factors that interact either physically or genetically with Rad51p, Rad52p, and Rad59p will help to elucidate the mitochondrial specific repair pathways or pathway components.

Media and strains
Media used in this study were previously described [39]. S. cerevisiae strains used in this study are listed in S1 Table and are isogenic with DFS188 except NPY66, which is derived from DFS160 [38,54]. Single-deletion strains were constructed by one-step gene replacement of the wild-type gene with a kanMX marker using standard protocols [55]. Strains containing two gene replacements were constructed by mating isogenic haploid strains, dissecting diploids, and screening by PCR for the desired genotypes. The spontaneous DRMD reporter strain (LKY196) was constructed previously [39]. Gene deletions in this strain were constructed by mating and dissection. The desired genotypes were screened for by PCR. These strains were then cured of their mtDNA by treatment with ethidium bromide, and cytoduced with the mitochondrial reporter from NPY66 [56].
Mutant alleles of the VMA1 intein from yeast were obtained from C. Tan (University of Missouri) on pUGal4int dead and pS5Gal4intein TS303 [44]. Inteins were inserted into the mtLS-KpnI endonuclease gene by gap repair in pEAS101. The TS303 and intein-dead alleles were amplified from the above plasmid with 50 bp of KpnI flanking sequence using the same forward primer 5'-CAATAGATCCCGCAACTGATGTAAATGATCCTAAAATGTGGCAAG CATTGtgctttgccaagggtaccaat-3' and reverse primers 5'-CCATTATTTGAATCCCAATAATT TTTTTTCATAACTTGATGACGTCCACAattatggacgacaacctggttggc-3' and 5'-CCATTATTT GAATCCCAATAATTTTTTTTCATAACTTGATGACGTCCACActgatggacgacaacctgcttggc-3', respectively. pEAS101 was digested with ZraI and contransformed into EAS748 [38] with each of the amplified intein cassettes. The primers are designed to insert the intein just before the cysteine at position 198 of KpnI. Transformants were selected on SD-Ura medium. Genomic DNA was collected from the transformants and plasmids were rescued by the transformation of E. coli Top10F' cells and selection on LB+ampicillin. Plasmids containing the intein were screened by digestion with BamHI and BglI and confirmed by sequencing. pEAS114 contains mtLS-KpnI+intein ts303 . pEAS115 contained mtLS-KpnI+intein dead .

Mitochondrial isolation and protease protection assay
Mitochondria were isolated as previously described in Diekert et al., 2001 [58]. The protease protection assay was performed as previously described in Diekert et al., 2001 with the following modification. 200μg of isolated mitochondria were used in each protease protection assay for both the ASY092 (Rad51-HA) and ASY127 (Rad59-HA) strains. 100μg of isolated mitochondria were used in the protease protection assay for DFS188 (WT). Samples were subjected to western blot analysis on 10% SDS-PAGE gels. Antibodies used for detection were as follows: 1:2500 dilution of anti-HA-HRP (Santa Cruz Biotechnology), 1:4000 anti-Cytb2 (generous gift from Dr. Tom Fox), 1:4000 anti-Cit1 (generous gift from Dr. Tom Fox), and 1:25,000 anti-POR1 (Invitrogen).

Chromatin immunoprecipitation
A 150 ml culture (RCY248) was grown to OD 600 = 0.800 in YPD and crosslinked in 1% formaldehyde for 20 minutes at room temperature. Cells were pelleted and washed twice in 1X TBS. Pellets were resuspended in 1ml of lysis buffer (50 mM Hepes-KOH, pH7.5, 150 mM NaCl, 1 mM EDTA, 1% TritonX-100, 0.1% sodium deoxycholate, 0.5% SDS) and 0.4 g of glass beads. Cellular lysates were obtained by bead beating at 4°C for 2.5 minutes. The lysates were then sonicated to generate fragments between 400-700bp in size. Lysates were precleared by incubating with 20μl of Protein A/G magnetic beads (Thermo Scientific). One milliliter of lysate was incubated over night with either 1 μg of anti-Rad51 antibody (Abcam) or anti-V5 (mock) (Invitrogen) rocking at 4°C. Immune complexes were precipitated by incubating with 20 μl of Protein A/G magnetic beads (Thermo Scientific) for 1 hour at 4°C. Beads were washed twice with lysis buffer, once with LiCl buffer (, and twice with 1X TE. Immune complexes were eluted twice with 150μl of elution buffer (1% SDS, 0.1M NaHCO 3 ). Crosslinks were reversed by adding 0.2M NaCl and heating overnight at 65°C. DNA was purified and equal amounts of DNA were analyzed by qPCR using SYBR green mix (Biorad) with the following primer pair: 5'-GGTAAATAGCAGCCTTATTATG-3', 5'-CGATCTATCTAATTACAGTAAAGC-3'.

Measuring nuclear and mitochondrial DRMDs and respiration loss
The rates of nuclear and mitochondrial spontaneous DRMDs and the average frequency of respiration loss were determined as previously described [39,40]. Fluctuation analyses consisted of 10-20 independent samples each. Rates were determined using the method of the median [59], and each experiment was performed a minimum of 3 times. Induced DRMDs were measured by growing strains to saturation in SRaffinose-Arg-Ura. These cultures were then diluted back to an OD 600~0 .08 and grown to OD 600 of 0.15-0.20. A pre-induction sample was removed and an appropriate dilution was plated on YPD and incubated for 3 days at 30°C. A final concentration of 2% galactose was added to the remaining culture, which was then incubated at 20°C for 16-18 hours. A post-induction aliquot was then removed and the appropriate dilution was plated on YPD and incubated for 3 days at 30°C. All colonies were replica plated to YPG to score for deletion events, and SD-Arg to score for intact reporters. The average frequency and standard deviation of three or more independent experiments is shown. P-values were determined by un-paired, two-tailed t-tests using Microsoft Excel for Macintosh 2011.

Southern blot analysis
DNA for Southern blot analysis was prepared from cells grown as stated above. A pre-induction sample was obtained prior to induction with 2% galactose and shifting to 20°C. After 16-18 hours of cutting, the cultures were shifted back to 30°C and samples were collected after 0, 2, 4, 6, and 8 hours. Total cellular DNA was extracted using standard procedures and transferred to Amersham Hybond-XL (GE Healthcare) nylon membranes following standard protocols. DNA was obtained from three independent experiments for each strain. Representative blots are shown in Figs 3, 5 and 6. The COX2, nuclear 25S rRNA, and mitochondrial 21S rRNA probes were constructed as previously described [40]. Quantification of Southern blots was performed using Image Lab software (www.biorad.com). The percentage of the COX2 signal represented by deletion product and break product was calculated by measuring the intensity of each band and dividing by the total signal for all 3 COX2-hybridizing bands.
In order to confirm that the signal from the COX2 probe was originating from the mtDNA the mitochondrial DNA from both pre-induction and time 0 cultures were diluted 1:1000 into minimal media containing 0.125mg of ethidium bromide and grown to saturation [46]. The saturated cultures were diluted 1:1000 into minimal media containing 0.125mg of ethidium bromide and grown to saturation. DNA was obtained from these cultures using standard protocol. The ethidium bromide treated cultures as well as DNA collected from these cultures prior to the ethidium bromide treatment were analyzed by Southern blot as described above.

Slot blot analysis of 3' ssDNA
Samples collected at time 0 were also subjected to slot blot analysis in order to detect 3'-ssDNA near the KpnI cleavage site. Total cellular DNA was extracted using standard procedures. Samples were prepared as previously described [60] with the following modifications. Native genomic DNA (1μg) was prepared in 10X SSC and denatured samples (0.15μg) were prepared in 0.4N NaOH for 5 minutes. The samples were neutralized by the addition of 3M sodium acetate. Native and denatured samples were transferred to Amersham Hybond-XL (GE Healthcare) nylon membranes with a slot blot manifold following the manufacturers suggested protocol (Schleicher & Schuell). The DNA was UV cross-linked to the membrane. The membrane was prehybridized for a minimum of 1 hour in prehybridization solution (6X SSC, 5X Denhardt's, 0.05% sodium pyrophosphate, 3 mg boiled fish sperm) at 37°C. The membrane was probed with a small synthetic end-labeled oligo (5' CAGCAATACCAGCTGTGAAATCG 3') overnight at 48°C. This probe anneals 213 nucleotides up stream of the KpnI cleavage site. The membrane was washed in 6X SSC, 0.05% sodium pyrophosphate at room temperature. Quantification of slot blots was performed using Image J software (http://rsbweb.nih.gov/ij/). The native samples were normalized to the appropriate denatured control. p-values were determined by un-paired, two-tailed t-tests using Microsoft Excel for Macintosh 2011. Three independent experiments were performed for each strain.  Fig. COX2 signal is dependent on the presence of mtDNA. All samples labeled pre were obtained prior to the induction of mtLS-KpnI ts . All samples labeled 0 were obtained after 16 hours of induction of mtLS-KpnI ts . All samples labeled + EtBr were grown to saturation in ethidium bromide twice in order to remove the mtDNA. Lane I contains DNA from the wildtype t = 0 timepoint of the mtLS-KpnI-intein ts -containing strain digested in vitro with KpnI and AvaII to demonstrate the migration of DNA with a DSB.  Table. Spontaneous respiration loss frequency in the rad mutants. The median frequency of spontaneous respiration loss was determined for each strain using 10-20 independent cultures. The average median frequencies for at least three experiments are presented. Mutant frequencies are compared to wild-type using unpaired two-tailed t-tests, and p-values are provided. (DOCX) S3 Table. The average difference in petite frequency following double-strand break induction. The samples analyzed in Fig 4 were also assessed for petites (Arg-, non-respiring colonies). For each experiment, we calculated the difference in the frequency of petites from the same culture before and after DSB induction. The average difference for all experiments is shown. The unpaired two-tailed t-tests were used to compare the frequencies for each strain before and after break induction, and p-values are presented. (DOCX)