Mgm101 Is a Rad52-related Protein Required for Mitochondrial DNA Recombination*

Background: The mechanisms of homologous recombination and recombinational repair of double-stranded DNA breaks in mitochondria are poorly understood. Results: The yeast mitochondrial nucleoid protein, Mgm101, shares biochemical, structural, and functional similarities with the Rad52 family proteins. Conclusion: Mgm101 is a Rad52-related molecular component involved in mtDNA recombination Significance: This finding helps in understanding the mechanism of homologous recombination in mitochondria. Homologous recombination is a conserved molecular process that has primarily evolved for the repair of double-stranded DNA breaks and stalled replication forks. However, the recombination machinery in mitochondria is poorly understood. Here, we show that the yeast mitochondrial nucleoid protein, Mgm101, is related to the Rad52-type recombination proteins that are widespread in organisms from bacteriophage to humans. Mgm101 is required for repeat-mediated recombination and suppression of mtDNA fragmentation in vivo. It preferentially binds to single-stranded DNA and catalyzes the annealing of ssDNA precomplexed with the mitochondrial ssDNA-binding protein, Rim1. Transmission electron microscopy showed that Mgm101 forms large oligomeric rings of ∼14-fold symmetry and highly compressed helical filaments. Specific mutations affecting ring formation reduce protein stability in vitro. The data suggest that the ring structure may provide a scaffold for stabilization of Mgm101 by preventing the aggregation of the otherwise unstable monomeric conformation. Upon binding to ssDNA, Mgm101 is remobilized from the rings to form distinct nucleoprotein filaments. These studies reveal a recombination protein of likely bacteriophage origin in mitochondria and support the notion that recombination is indispensable for mtDNA integrity.

Homologous recombination is a conserved molecular process that has primarily evolved for the repair of double-stranded DNA breaks and stalled replication forks. However, the recombination machinery in mitochondria is poorly understood. Here, we show that the yeast mitochondrial nucleoid protein, Mgm101, is related to the Rad52-type recombination proteins that are widespread in organisms from bacteriophage to humans. Mgm101 is required for repeat-mediated recombination and suppression of mtDNA fragmentation in vivo. It preferentially binds to single-stranded DNA and catalyzes the annealing of ssDNA precomplexed with the mitochondrial ssDNA-binding protein, Rim1. Transmission electron microscopy showed that Mgm101 forms large oligomeric rings of ϳ14fold symmetry and highly compressed helical filaments. Specific mutations affecting ring formation reduce protein stability in vitro. The data suggest that the ring structure may provide a scaffold for stabilization of Mgm101 by preventing the aggregation of the otherwise unstable monomeric conformation. Upon binding to ssDNA, Mgm101 is remobilized from the rings to form distinct nucleoprotein filaments. These studies reveal a recombination protein of likely bacteriophage origin in mitochondria and support the notion that recombination is indispensable for mtDNA integrity.
Mitochondrial DNA (mtDNA) 3 encodes integral components of the energy-transducing oxidative phosphorylation pathway. It is organized in vivo as protein-DNA complexes, known as mitochondrial nucleoids, with each nucleoid containing ϳ2-10 copies of the genome (1)(2)(3)(4). mtDNA is subject to constant attack by reactive oxygen species that are produced through metabolism in mitochondria. This is generally believed to be responsible for the higher mutational load in mtDNA compared with the nuclear genome (5,6). In addition to oxidative damage, it has been suggested that replicative errors also contribute significantly to the overaccumulation of mutations in mtDNA during the lifetime of the cell (7). If unrepaired, the overaccumulated mutations in mtDNA can directly affect mitochondrial energy production and lead to severe physiological consequences, which are manifested by many human degenerative diseases (8,9). Unfortunately, in contrast to what we have learned from the nuclear genome, our understanding of how damaged DNA is repaired in mitochondria is rather limited. So far, only a few enzymes involved in base excision repair have been documented inside the organelle (10,11).
Homologous recombination (HR) is a DNA repair mechanism conserved from bacteriophage to humans that plays a critical role in the error-free repair of double strand breaks (DSBs) and stalled or collapsed replication forks. It is catalyzed by the Rad52 epistasis group proteins (12,13). In conventional HR, DSBs are first processed by a nuclease into 3Ј ssDNA tails. The 3Ј ssDNA tails are subsequently coated by single strand DNAbinding proteins to prevent the formation of secondary structures. A recombination mediator, as exemplified by the bacterial RecO and yeast Rad52, displaces the single strand DNAbinding protein and recruits the core ATP-dependent recombinase (such as RecA in prokaryotes and its eukaryotic orthologs, Rad51) to form helical nucleoprotein filaments. The filaments then engage in homology search, strand invasion, and homologous pairing within duplex DNA templates. As a result, the invading 3Ј-end is used to prime DNA replication to copy the genetic information missing from the dsDNA breaks. The genome integrity is consequently restored following the resolution of recombination intermediates by different molecular strategies with or without involving DNA strand crossover (13)(14)(15). In canonical DSB repair, Rad52 has dual functions. In addition to its role as a recombination mediator for Rad51 recruit-ment, Rad52 also participates in ssDNA annealing, an activity critical for the capture of the second DNA end at the recombination site that generates Holliday junctions (16,17).
Does HR, an almost universal DNA repair mechanism in all domains of life, also occur in mitochondria? Recombination between DNA markers has been well documented in yeast mitochondria (18). Shibata and co-workers (19 -21) have identified the MHR1 gene that encodes a mitochondrial protein affecting gene conversion and the repair of oxidatively damaged mtDNA. Based on its ability in promoting homologous pairing between ssDNA and dsDNA duplex in vitro, Mhr1 was proposed to be an ATP-independent recombinase (22,23). Because mutations in MHR1 have only a slight effect on the crossing over of unlinked mitochondrial genetic loci, it has been suggested that there is another pathway for homologous recombination in mitochondria (19).
We have previously identified the MGM101 gene in a genetic screen for temperature-sensitive (ts) mutants affecting mtDNA maintenance in Saccharomyces cerevisiae (24). Mature Mgm101 is a positively charged protein of 247 amino acids (25). Nunnari and co-workers (26,27) have shown that Mgm101 is associated with actively replicating mitochondrial nucleoids and that mtDNA in mgm101 mutants is hypersensitive to several DNA-damaging agents, including ultraviolet, ␥-ray irradiation, and hydrogen peroxide. These observations strongly suggested a role for Mgm101 in mtDNA repair. We have also found that the mgm101 mutant rapidly loses the wild-type ( ϩ ) but not the hypersuppressive Ϫ mtDNA (28). These Ϫ genomes contain highly recombinogenic GC-rich repeats, which may spare the requirement for Mgm101. This finding further suggested that Mgm101 may play a critical role in mtDNA repair via a recombination-based process. In this report, we show that Mgm101 is a Rad52-related protein of bacteriophage origin. The data suggest the presence of an evolutionarily conserved HR component in mitochondria that is essential for the maintenance of mtDNA integrity.
In Vivo Repeat-mediated Deletion/Recombination Assay-The fluctuation assay for determining the rate of repeat-mediated deletions in mtDNA has been described by Sia and coworkers (29). Cells were preselected on minimal medium for Arg ϩ for 3 days and plated for colony formation on YPD medium for 3 days. Cells from 20 independent Arg ϩ colonies were then resuspended in water, and appropriate dilutions were plated on YPGE and YPD. Respiratory competent colonies (or recombinants) on YPGE were scored after growth at 30°C for 7 days. The number of viable colonies on the YPD plates was used to deduce the total cell number plated on YPGE. The rate of recombination per cell division was estimated using the p 0 method (30). For mtDNA instability and recombination rate, a two-sample t test (two-tailed) was used for statistical evaluation between the control and the mutant cells.
Expression and Purification of Mgm101, Rim1, and Mhr1 from Escherichia coli-The MGM101 ORF lacking the first 22 codons was amplified by PCR using the primers MGM101P1 (5Ј-CGGGGATCCGTGGTGAGTACCGGCACTAG-3Ј) and MGM101P2 (5Ј-AAAACTGCAGCTATTTATAAGGATAT-TCAACTTG-3Ј). The BamHI and PstI sites incorporated in the primers (underlined) allowed the cloning of the gene downstream of the malE sequence encoding the maltose-binding protein (MBP) in the expression vector pMal-c2e (New England Biolabs). This created the MBP-Mgm101 fusion with a linker containing a cleavage site for the Prescission TM protease (GE Healthcare). The fusion plasmid pMalc2e-MGM101 was introduced into the E. coli strain BL21-CodonPlus(DE3)-RIL (Stratagene). Expression of fusion protein was induced with isopropyl ␤-D-1-thiogalactopyranoside at 0.3 mM in LB medium supplemented with 0.2% glucose, ampicillin (100 g/ml), and chloramphenicol (50 g/ml). Cells were cultured at 30°C for 5 h with an A 600 of ϳ0.5. After centrifugation, the cell pellet was resuspended in the column buffer (20 mM Tris-HCl, pH 7.4, 50 mM glucose, and 1 mM EDTA) containing protease inhibitors (25 M leupeptin, 10 M pepstatin, and 0.5 mM phenylmethanesulfonyl fluoride). The cell suspension was sonicated for 30 s five times, and DNase I (50 g/ml) was added to cell lysate and incubated at 4°C for 2 h. The final concentration of NaCl in the crude lysate was adjusted to 500 mM, and cell debris was removed by centrifugation at 10,000 ϫ g at 4°C. The cell lysate was then incubated with amylose resin at 4°C overnight. After several washes with column buffer, MBP-Mgm101 was eluted with 10 mM maltose in column buffer. The fusion protein was cleaved with the Prescission TM protease (0.33 units/100 g) overnight to release Mgm101. The recombinant Mgm101 thus has the sequence GPKVPEFGS added to the N terminus of the native protein. After dialysis against 10 mM sodium phosphate (pH 7.2, 4°C, 100 mM NaCl), Mgm101 was separated from MBP and purified to homogeneity by cation exchange chromatography using a Bio-Scale TM Mini Prep UNOsphere TM cartridge (Bio-Rad). Mgm101 was further purified by size exclusion chromatography on a calibrated Superose 6 column equilibrated with 20 mM MOPS, pH 7.0, 150 mM NaCl, 5 mM ␤-mercaptoethanol, 1 mM EDTA, 0.2 mM phenyl-methanesulfonyl fluoride. Peak fractions were analyzed by SDS-PAGE, concentrated, and stored at Ϫ80°C.
Rim1 was expressed in an MBP-fused form. After proteolytic cleavage by Prescission Protease (GE Healthcare), Rim1 was released from MBP. Rim1 was then bound to single strand DNA cellulose. After washing six times with 20 Tris, pH 7.4, 500 mM NaCl, 1 mM EDTA, 10% glycerol, 2.5 mM ␤-mercaptoethanol, and 1 mM PMSF, Rim1 was eluted with the same buffer but containing 1.5 M NaCl. Mhr1 was expressed in a His 6 -tagged form as described by Ling and Shibata (22). After the removal of His 6 with Prescission Protease (GE Healthcare), Mhr1 was further purified to homogeneity by size exclusion chromatography on a calibrated S75 column equilibrated in 20 mM Tris, pH 7.0, 150 mM NaCl, 10 mM MgCl 2 , 5 mM ␤-mercaptoethanol, 1 mM EDTA, and 0.2 mM PMSF.
For the single strand annealing assay, the 51-base HS40AT-F (see above) oligonucleotide was end-labeled with [ 32 P]ATP. A 200 nM concentration of the labeled HS40AT-F and 1.75 M Mgm101 were added to the annealing buffer containing 50 mM triethanolamine-HCl, pH 7.5, 0.5 mM Mg(OAc) 2 , 1 mM DTT, and BSA at 100 g/ml. After incubation for 5 min at 25°C, the reactions were initiated by adding a 200 nM concentration of the unlabeled complementary oligonucleotide HS40AT-R and incubated at 25°C for the designated time. The reactions were stopped by the addition of 2 M cold HS40AT-F, 0.5% SDS, and proteinase K at 1 mg/ml. The deproteination was continued at 25°C for 15 min, and the samples were loaded on a 10% native PAGE, which was run at 4°C in Tris-acetate-EDTA buffer.
Native Agarose Gel Electrophoresis and Dithiobis(Succinimidyl Propionate) (DSP) Cross-linking-Native agarose gel electrophoresis was performed using a 0.8% agarose gel. The gel was run in TA buffer (40 mM Tris-base, 20 mM acetic acid, pH 8.25) for ϳ1 h on ice using an applied voltage of 100 V. The gel was then fixed in fixing solution (25% isopropyl alcohol, 10% acetic acid) for 30 min, washed in 95% ethanol two times for 15 min and once overnight, blot-dried, and dried in a vacuum slab drier for 1 h at 80°C. The gel was subsequently stained in Coomassie Blue and destained in fixing solution to visualize bands. For DSP-mediated cross-linking, 73 g of purified Mgm101 at a concentration of 2.02 mg/ml in sodium phosphate buffer (10 mM sodium phosphate, pH 7.2, 500 mM NaCl) was diluted in water to a final volume of 200 l. DSP was added from a 1.0 mM stock to obtain a final concentration of 0.5 mM. Prior to the addition of DSP, a 10.7-l aliquot of Mgm101 was removed as uncross-linked control and treated with glycine at a final concentration of 100 mM. Once cross-linker was added, the protein was incubated on ice for 1 h 20 min, and aliquots of 10.7 l were removed at distinct time points. The cross-linking reaction in each aliquot was quenched with glycine (100 mM) at the point of removal and analyzed on a 12% non-reducing SDS-polyacrylamide gel. The cross-linked products were visualized by Western blot analysis using an anti-Mgm101 antibody.
Sedimentation Velocity Analytical Ultracentrifugation-We used a Beckman Coulter ProteomeLab TM XL-A analytical ultracentrifuge equipped with absorbance optics and an 8-hole An-50 Ti analytical rotor. Sedimentation velocity experiments were conducted at 40,000 rpm (128,000 ϫ g) at 10°C using 3-mm two-sector charcoal-filled Epon centerpieces. Protein samples were run in 10 mM sodium phosphate, pH 7.2, 100 mM NaCl, and 0.1% sodium azide. 300 scans were collected with the time interval between scans set to zero. Sedimentation boundaries were analyzed by the continuous distribution (c(s)) method using the program SEDFIT (31). The program SEDN-TERP version 1.09 (32) was used to correct the experimental S value (s*) to standard conditions at 20°C in water (s 20,w ) and to calculate the partial specific volume of each protein.
Transmission Electron Microscopy and Image Analysis-For negative stain electron microscopy, Mgm101 (with or without DNA) was diluted to between 20 and 100 g/ml and spread on carbon-coated copper grids. For visualizing Mgm101-ssDNA interaction, 2 M Mgm101 was incubated with ssDNA oligonucleotide (8 M) or M13mp18 ssDNA (5 M) at 37°C for 10 min before being applied to the grids. The buffer contained 20 mM Tris, pH 7.0, 10 mM MgCl 2 , 1 mM EDTA, and 0.2 mM PMSF. The grids were washed once with water and stained with 1% uranyl acetate. For electron cryomicroscopy, protein at 0.2-1 mg/ml was applied to holey carbon grids (C-flat), blotted with filter paper and flash-frozen by plunging into liquid ethane. Negative stained and frozen hydrated samples were visualized in a JEM-2100 transmission electron microscope (JEOL) operating at 200 kV.
Electron micrographs were recorded on a 4096 ϫ 4096 pixels CCD camera (TVIPS F415-MP) in minimum dose mode at an electron optical magnification of 40,000ϫ and a defocus of Ϫ1.5 m. Micrographs were displayed using the boxer program of the EMAN software package (33), and data sets of 4,500 top and 930 side view images were collected using the "autobox" function (for the top views) or manually (for the side views). All subsequent image analyses were done with the IMAGIC 5 package of programs (34). Images were normalized and were band pass-filtered to remove low (Ͻ0.05 Å Ϫ1 ) and high (Ͼ0.16 Å Ϫ1 ) spatial frequencies, and a circular mask was applied. Data sets were treated by multivariate statistical analysis and sorted into 16 -36 initial classes by hierarchical ascendant classification. The best averages of the classes as determined by IMAGIC were then used as references for multireference or single reference alignment, and the multivariate statistical analysis/classification was repeated until stable results were obtained.

Mgm101 Shares Functionally Conserved Sequence
Signatures with Rad52-Rad52 is a central recombination protein best known as a mediator of the Rad51 recombinase in S. cerevisiae. It is evolutionarily related to a highly diverged superfamily of recombination proteins that are widely present in temperate bacteriophages (35)(36)(37). These viral proteins, as exemplified by Red␤, Erf, and Sak, have single strand annealing (SSA) activity. The eukaryotic Rad52 has retained the SSA activity on its N terminus, but it has also acquired a large C-terminal domain that recruits Rad51 onto ssDNA (38,39). Yeast cells also express Rad59, a Rad52 paralogue that lacks the C-terminal Rad51-interacting domain (40). Rad59 functionally overlaps with Rad52 in DSB repair, but it also appears to have unique functions, including stimulation of Rad52-mediated single strand annealing (41)(42)(43).
The central domain of Mgm101 between residues 108 and 241 shares 17 and 15% sequence identities with the N-terminal ssDNA-annealing domain of yeast Rad52 (25) (supplemental Fig. S1) and with Rad59, respectively (Fig. 1A). Sequence conservation is hardly recognizable between Mgm101 and the Rad52 orthologs in bacteriophage, such as Red␤, Erf, and Sak. To test whether Mgm101 is a Rad52-related protein, we first asked whether highly conserved amino acids critical for Rad52 activity are required for Mgm101 function in vivo. Previous studies have shown that the nuclear genome of yeast cells expressing the N58A, F61A, and F140A alleles of Rad52 are hypersensitive to ␥-ray irradiation (44). These residues are also conserved in Rad59, which correspond to Asn-87, Phe-90, and Phe-180. Therefore, we mutated the equivalent amino acids in Mgm101 (Asn-150, Phe-153, and Phe-235) to alanine. The mutant alleles were examined for their activity in supporting mtDNA maintenance in two series of experiments. As shown in Fig. 1B, the mgm101-1 ts mutant harboring the control vector is completely converted into petite colonies lacking mtDNA at both 35 and 37°C. Petite formation was largely suppressed by the wild type MGM101. At 37°C, the mgm101 N150A , mgm101 F153A , and mgm101 F235A alleles expressed from a centromeric vector failed to suppress petite formation, suggesting that the mutant proteins are functionally defective in vivo. The N150A allele was found to be relatively milder compared with F153A and F235A. It partially suppressed the petite phenotype when cells were incubated at 35°C.
To exclude possible interallelic complementation and the synthetic defects that might occur with mgm101-1 ts , we generated meiotic segregants expressing only mgm101 N150A , mgm101 F153A , or mgm101 F235A (Fig. 1C). The mutant alleles were introduced into a diploid strain in which one copy of the  (44), are conserved in Mgm101 (corresponding to Asn-150, Phe-153, and Phe-235) and Rad59 (corresponding to Asn-87, Phe-90, and Phe-180). B, stability of the ϩ genome in the mgm101-1 ts mutant transformed with mutant alleles of MGM101. The mgm101-1 ts mutant was transformed with a centromeric vector expressing mgm101 N150A , mgm101 F153A , and mgm101 F235A . Ura ϩ transformants were plated on YPD medium and incubated at 35 and 37°C. Petites were scored as colonies unable to grow on non-fermentable carbon sources after replica plating. The data are the averaged petite frequency of at least three independent transformants. Error bars, average deviations. C, meiotic analysis showing that mgm101 N150A , but not mgm101 F153A and mgm101 F235A , complements the null mgm101⌬::kan allele for respiratory growth on ethanol plus glycerol medium at 30°C. Four complete tetrads independently segregating the null mgm101⌬::kan and the mgm101 N150A , mgm101 F153A , or mgm101 F235A allele, were dissected on complete glucose medium. The meiotic segregants were then replica-plated onto G418, ϪUra, and ethanol plus glycerol medium. G418 R marks the presence of the null mgm101⌬::kan allele. Ura ϩ marks the mgm101 N150A , mgm101 F153A , or mgm101 F235A allele, as a wild-type URA3 gene was introduced together with mutant alleles into the ura3 locus after chromosomal integration. Circled are the meiotic spores that cosegregate mgm101⌬::kan with mgm101 N150A , mgm101 F153A , or mgm101 F235A . D, the respiratory growth of cells expressing only the mgm101 N150A allele is completely inhibited at 37°C but partially inhibited at 35°C. DECEMBER 9, 2011 • VOLUME 286 • NUMBER 49

JOURNAL OF BIOLOGICAL CHEMISTRY 42363
wild type MGM101 is disrupted by the insertion of the kan (G418 R ) marker. The mutant alleles were placed adjacent to the URA3 gene, which facilitated the integration into the ura3 locus by selecting for Ura ϩ transformants. After sporulation and tetrad dissection, G418 R and Ura ϩ segregants were examined for respiratory competency. We found that the mgm101 F153A and mgm101 F235A alleles are unable to support respiratory growth on ethanol plus glycerol medium. The meiotic segregants expressing only mgm101 N150A are respiratory competent at 30°C. However, these cells exhibited partial and complete loss of respiratory growth at 35 and 37°C, respectively (Fig. 1D). Taken together, the data demonstrated that Asn-150, Phe-153, and Phe-235 in Mgm101 are functionally critical for in vivo function, supporting the idea that Rad52, Rad59, and Mgm101 may share similar molecular functions despite a weak similarity in their primary sequences.
Mgm101 Is Required for mtDNA Recombination and Suppression of mtDNA Fragmentation in Vivo-We used a reporter system developed by Sia and co-workers (29) to determine whether Mgm101 affects mtDNA recombination in vivo ( Fig.  2A). In this system, expression of the mitochondrial COX2 gene, encoding subunit 2 of cytochrome c oxidase, is interrupted by the insertion of ARG8 m , which has been recoded for expression in mitochondria. ARG8 m is expressed via an in-frame translational fusion with COX2, which complements the nuclear arg8 mutation (45). ARG8 m is flanked by 96-bp direct repeats. Recombination events that occur between the repeats following dsDNA breaks result in the deletion of ARG8 m and the restoration of COX2 expression. Thus, the number of respiratory competent and Arg Ϫ colonies that arise on a non-fermentable carbon source can be used as an indicator for the frequency of recombination-based ARG8 m deletions.
We took advantage of the relatively mild N150A allele of MGM101 (see Fig. 1) to evaluate its effect on repeat-mediated mtDNA deletion/recombination. The choice of such a mild mutant is critical to avoiding possible nonspecific effects on recombination in severely affected mutants having a highly destabilized mitochondrial genome and reduced mtDNA copy number. The N150A allele was introduced into the chromosome, and the endogenous wild-type MGM101 was subsequently disrupted. At 30°C, haploid cells expressing mgm101 N150A had only a mildly increased Ϫ and °frequency, which was 26.7% compared with 13.6% in the MGM101 control strain (Fig. 2B). However, we found that the rate of repeatmediated mtDNA deletions in the mutant is reduced by 8.1fold compared with the control (Fig. 2C). Haploid cells co-expressing MGM101 and mgm101 N150A also had a reduced recombination rate, suggesting a partially dominant effect of the mutant allele despite its slight reduction in the petite formation. We speculate that a mild decrease in recombination may partially stabilize instead of destabilizing the mitochondrial genome under unstressed conditions by lowering ectopic recombinations, which give rise to spontaneous petites. In summary, these data demonstrate that defects in Mgm101 function affect recombination between non-tandem repeats in the mitochondrial genome.
To learn whether reduced recombination in the mgm101 N150A mutant impairs mtDNA repair, we directly examined the sensitivity of the mutant cells to hydrogen peroxide on non-fermentable carbon sources. As shown in Fig. 2D, the growth of the mgm101 N150A mutant is partially reduced on ethanol plus glycerol medium (YPGE) at the semipermissive temperature (35°C), but cell growth is inhibited with an increasing level of hydrogen peroxide. Consistent with previous observations (26), these results further support the idea that Mgm101 is required for the repair of oxidatively damaged mtDNA. Furthermore, we found that mtDNA fragmentations can be readily detected in mgm101 mutants even under normal growth conditions. We have previously shown that in a nascent null mutant of MGM101, the copy number of the 430-bp N1 Ϫ genome is reduced by ϳ4 -5-fold compared with that in cells expressing the wild type (28). The partial retention of mtDNA permitted us to determine possible molecular changes to mtDNA brought by MGM101 inactivation. We performed Southern blot analysis of the NspI-digested DNA samples, which generates a single cut in the N1 Ϫ genome. It was found that in addition to the expected 430-bp N1 repeats, several fragments of different subgenomic sizes were frequently detected (see Fig. 2E). These subgenomic fragments, which probably arise from replication of broken DNA, were not seen in cells expressing the wild-type MGM101. The capture of these subgenomic mtDNA species, together with a role of Mgm101 in repeat-mediated recombination and the hypersensitivity of the mgm101 N150A mutant to hydrogen peroxide, strongly supports the idea that Mgm101 may be involved in the repair of dsDNA breaks, probably via a recombination-based mechanism.
Mgm101 Preferentially Binds to Single-stranded DNA and Promotes Annealing of Rim1-coated Single-stranded DNA-To characterize Mgm101 biochemically, we attempted to produce and purify the recombinant yeast protein from E. coli. Direct expression of Mgm101 in bacteria has been unsuccessful, probably due to its cellular toxicity and/or poor solubility in bacteria. However, we succeeded in producing Mgm101 at high yield in an MBP-tagged form. Cell lysates were treated with DNase I to minimize contamination by bacterial DNA before amylose affinity chromatography. After proteolytic cleavage by the Prescission TM protease, Mgm101 was released from MBP and purified to homogeneity in a soluble form by cation exchange and size exclusion chromatography (Fig. 3A). We compared the quality of Mgm101 with regard to DNA contamination in preparations with or without DNase I treatment. We found that this treatment improves the A 280 /A 260 ratio from ϳ1.1 to 1.55-1.63, suggesting that the final preparations have minimal DNA contamination.
Using electrophoretic mobility shift assays, we found that Mgm101 binds to ssDNA with an affinity 5.5 times higher than to dsDNA (Fig. 3B), with a K d value of 192 nM for ssDNA versus 1.068 M for dsDNA (p ϭ 0.005). This property is reminiscent  DECEMBER 9, 2011 • VOLUME 286 • NUMBER 49

JOURNAL OF BIOLOGICAL CHEMISTRY 42365
of Rad52 family proteins (46,47). The binding affinity of Mgm101 to ssDNA is roughly comparable with that of Rad52. For comparison, we purified and analyzed the DNA-binding affinity of Mhr1, a protein previously known to be required for mtDNA recombination (22). We found that Mgm101 binds to ssDNA with an affinity ϳ8-fold higher than Mhr1 (Fig. 3, B and  C).
To know whether Mgm101 is functionally analogous to Rad52, we determined whether Mgm101 catalyzes ssDNA annealing. In this assay, a 32 P-labeled 51-base oligonucleotide (HS40AT-F) was incubated with Mgm101. The annealing reactions were initiated by the addition of the complementary strand (HS40AT-R) and stopped by a 10-fold excess of the unlabeled HS40AT-F. After deproteination, the reaction products were resolved on a native polyacrylamide gel. As shown in Fig.  3D, spontaneous annealing between the two oligonucleotides occurred slowly in a time-dependent manner. The annealing process was drastically accelerated by Mgm101. The oligonucleotides were almost completely annealed to form the slowly migrating dsDNA after only 30 s of incubation in the presence of Mgm101.
The in vivo substrates for Rad52 are ssDNA that are coated by their cognate single strand binding proteins to prevent the formation of secondary structures (48,49). The yeast mitochondrial single strand binding protein is encoded by the RIM1 gene, which contains an intron on its 5Ј moiety (50). We completely synthesized the intronless open reading frame of RIM1. After expression in E. coli, the protein was purified to homoge-neity (supplemental Fig. S2). To assess whether Mgm101 anneals Rim1-complexed ssDNA, the 32 P-labeled oligonucleotides were preincubated with Rim1. Annealing reactions were then initiated with or without the addition of Mgm101. As predicted, Rim1 alone inhibited the self-annealing of the complementary oligonucleotides (Fig. 3D). However, Mgm101 considerably stimulated strand annealing in the presence of Rim1, although the rate of annealing is significantly lower compared with that in the reaction lacking Rim1. The data provide strong support for a role of Mgm101 in promoting single strand annealing under physiologically relevant conditions. Mgm101 Forms Homo-oligomers-Several lines of evidence suggest that Mgm101 forms homo-oligomers. First, recombinant Mgm101 migrates poorly to the cathode in a native agarose gel at pH 5.5, 7.4, and 8.25 despite an estimated isoelectric point of 9.09 (Fig. 4A). Second, treatment of purified Mgm101 with DSP generated oligomeric species with a molecular mass above 180 kDa (Fig. 4B). Third, size exclusion chromatography clearly showed that purified Mgm101 was eluted as a single peak with an estimated molecular mass of ϳ400 kDa (Fig. 4C). The tagging by MBP does not appear to affect the oligomerization of Mgm101, because the MBP-Mgm101 fusion was eluted as an expected molecular species of ϳ940 kDa. Note that no protein fractions corresponding to the monomeric form were detected with Mgm101 or MBP-Mgm101.
To more precisely determine the molecular weight of Mgm101 oligomers, we characterized the recombinant protein by sedimentation velocity analytical ultracentrifugation. The majority of Mgm101 sedimented as a monodispersed species with an estimated sedimentation coefficient (s*) of 8.0 s (8.3 s for the standard sedimentation coefficient, s 20,w , after correcting for water at 20°C) ( Fig. 4D and supplemental Fig. S3). The native molecular mass calculated from the s 20,w value is ϳ397 kDa, which is consistent with the presence of ϳ14 subunits in the Mgm101 oligomers as estimated with size exclusion chromatography.
Mgm101 Oligomeric State Is Affected by the F153A, F235A, and N150A Mutations-Interestingly, we found that the oligomeric state of Mgm101 is dramatically altered by the functionally defective F153A, F235A, and N150A mutations. We prepared the mutant proteins by the MBP-tagging strategy. However, after the proteolytic release from MBP, no mutant Mgm101 was recovered due to poor solubility. Direct analysis of the fusion proteins revealed that the mutant proteins are aggregated. By size exclusion chromatography, we found that MBP-Mgm101 F153A and MBP-Mgm101 F235A eluted in the void volume fraction, in contrast to MBP-Mgm101, which formed a distinct peak of ϳ940 kDa as expected (Fig. 4C). This suggests that Mgm101 F153A and Mgm101 F235A form large aggregates. Consistent with its mild phenotype, Mgm101 N150A behaved differently. A large fraction of Mgm101 N150A eluted in the void volume, suggesting aggregate formation (denoted as fraction A). In addition, Mgm101 N150A forms a second peak with an estimated molecular mass of ϳ1.4 MDa, which remains soluble (denoted as fraction S), suggesting the formation of oligomers significantly larger than MBP-Mgm101. The molecular species in the void volume appeared to result from protein aggregates rather than from the formation of higher order structures. On SDS-PAGE, MBP-Mgm101 F153A , MBP-Mgm101 F235A , and the aggregated fraction of MBP-Mgm101 N150A (N150A (A)) formed SDS-resistant aggregates that mainly remained in the wells, whereas the soluble fraction of MBP-Mgm101 N150A (N150A (S)) was readily resolved into monomers like MBP-Mgm101 (Fig. 4E).
Electrophoretic mobility shift assays showed that in comparison with MBP-Mgm101, the DNA binding activity of MBP-Mgm101 F153A and MBP-Mgm101 F235A is drastically reduced, providing an explanation for the loss of their function in vivo (see Fig. 1, B and C). Intriguingly, the aggregated fraction of MBP-Mgm101 N150A retains a DNA binding activity comparable with that of MBP-Mgm101 at 25°C, whereas the soluble fraction of MBP-Mgm101 N150A has an increased DNA binding activity (Fig. 4F). It is important to note that despite its DNA binding activity in vitro, Mgm101 N150A is mainly present in aggregates (see Fig. 4C). We speculate that this may be further accentuated when cells are grown at the non-permissive temperature (37°C). The aggregated protein is either inactive in DNA transactions after ssDNA binding or unable to interact with other components in the recombination pathway, which ultimately results in mtDNA disintegration in the mgm101 N150A mutant. In summary, the ensemble of the data demonstrates that the oligomerizing property of Mgm101 is altered by the F153A, F235A, and N150A mutations, thereby leading to protein aggregation and loss of in vivo function.
Higher Order Structural Organization of Mgm101 in the Absence and Presence of DNA-We used transmission electron microscopy to directly visualize the oligomeric structure of Mgm101. We found that Mgm101 shares some higher order structural features similar to the Rad52 family proteins, especially those from bateriophages. Analysis of the negatively stained specimen showed that Mgm101 forms homogeneous ring-shaped structures (Fig. 5A). Alignment and classification of a data set of ϳ4,500 molecular images (without imposing a symmetry) revealed an estimated diameter of ϳ200 Å for the rings and ϳ60 Å for the central channel (Fig. 5B).
Mgm101 also exhibited dynamic structural organization like several known recombination proteins. The Mgm101 rings interacted laterally to form honeycomb-like structures (Fig.  5E). Short filaments were frequently observed in freshly prepared samples (Fig. 5A, boxed). Interestingly, these filaments became the dominant molecular species in samples after storage for 4 weeks at 4°C (Fig. 5C). We found that there is no significant difference between the ring-and the filament-enriched Mgm101 samples in binding to ssDNA (our unpublished observation). Further inspection of aged samples by electron cryomicroscopy revealed that some filaments can reach a length of several hundred nm (Fig. 5F).
To visualize in greater detail the structures of the Mgm101based filaments, we analyzed 930 images from negatively stained samples by the IMAGIC 5 software. The data suggested that the filaments contain compressed helical structures with a pitch of ϳ50 Å (Fig. 5D). The filaments are therefore formed by following a distinct structural pattern and not simply by the stacking of conventional oligomeric rings.
Finally, we used negative staining to visualize Mgm101-DNA interaction. Incubation with an oligonucleotide of only 66 bases gave rise to distinct nucleoprotein filaments (Fig. 5G). In the presence of the 7.25-kb M13mp18 ssDNA, the Mgm101-DNA complexes were mostly condensed into thick nucleoprotein filaments (Fig. 5H). Neither ring nor helical structures could be discerned in the Mgm101-based nucleoprotein filaments.

DISCUSSION
In this report, we show that the yeast mitochondrial nucleoid protein, Mgm101, shares similar biochemical, structural, and functional properties with several Rad52 family proteins involved in DNA recombination. The work reveals the presence of an evolutionarily conserved recombination protein in mitochondria that is essential for the maintenance of mtDNA integrity. The Rad52-type proteins are widespread among bacteriophages. Based on structure-based modeling, these viral recombination proteins are grouped into the Rad52-, RecA/ Rad51-, and Gp2.5-type subfamilies (37). The Rad52 family proteins are structurally characterized by a typical ␤-␤-␤-␣ fold, despite very limited similarity in their primary sequences. The best characterized Rad52-type proteins include Red␤ and Erf from the bacteriophages and P22, RecT from the prophage rac, and Sak from the lactococcal phage ul36. These proteins have a higher order structural organization similar to the ssDNA binding domain of Rad52, all forming homo-oligomeric rings of 10 -14-fold symmetry (35,47,(51)(52)(53). Mgm101 shares similar features with these proteins, by forming rings of ϳ14-fold symmetry with a diameter of ϳ200 Å, compared with ϳ130, 145, and 155 Å for Rad52, Red␤, and Sak, respectively.
Biochemical data suggest that the ring structure is functionally important in vivo. The F153A, F235A, and N150A alleles affecting oligomerization are all defective in maintaining mtDNA. By using several techniques, including native gel electrophoresis, size exclusion chromatography, and sedimentation velocity analytical ultracentrifugation, we did not observe the presence of protein fractions corresponding to monomeric Mgm101. Even in the cases of the aggregation-prone MBP-Mgm101 F153A , MBP-Mgm101 F235A , and Mgm101 N150A , the monomeric form of the proteins was undetectable. It is possible that the monomers are unstable in solution in the absence of DNA. The F153A, F235A, and N150A mutations probably induce aggregation by disassembling the ring structure into the unstable monomeric components. Thus, the ring structure may play a critical role in stabilizing Mgm101.
Although Mgm101 shares hardly recognizable sequence similarity with the bacteriophage enzymes, its structural organization appears to be more close to the phage proteins than to Rad52. For example, Mgm101 forms highly compressed helical filaments like Red␤ (35). The functional implications of these structures are not well understood. We speculate that the higher order structures, including rings and filaments, may serve as storage forms of Mgm101 in the nucleoids. This is particularly relevant in view of the fact that monomers are unstable and prone to aggregation. In response to DNA damage, the protein can therefore be rapidly remobilized from the stores to bind and repair damaged mtDNA that is present in the vicinity. However, future in vivo experiments are required to validate this hypothesis.
Although Mgm101 and Red␤ form similar helical filaments, they display structural differences when bound to DNA. In contrast to Red␤, which to some extent retains the helical organization (35), Mgm101 forms distinct nucleoprotein filaments on ssDNA. No ring structure is recognizable in the nucleoprotein filaments. This type of organization is rather close to that observed with the Sak protein in lactococcal phage ul36 (47). The unique DNA condensing activity may provide a mechanism for bringing the substrates into close proximity, thereby facilitating a homology search. Because neither the DNA-free helical structure nor the condensed nucleoprotein filaments have been seen with Rad52, Mgm101 would have most likely arisen from a viral protein. From an evolutionary perspective, this may not be surprising because several mtDNA-transacting enzymes have previously been known to have a viral origin (e.g. mitochondrial DNA and RNA polymerases) (54,55).
We have demonstrated that defects in Mgm101 function lead to mtDNA fragmentation and reduce repeat-mediated mtDNA deletions. Together with the hypersensitivity of the mgm101 N150A mutant to hydrogen peroxide and other DNAdamaging agents (26), the data strongly support a role of the protein in recombinational repair in vivo. Recombination between non-tandem direct repeats generally results from recombinational repair of dsDNA breaks, which are mediated either by non-allelic strand invasion between the repeats or by annealing of non-allelic repeated sequences after exonuclease resection of dsDNA breaks. Because Mgm101 lacks the large C-terminal domain that is present in Rad52 for the recruitment of Rad51 (see Fig. 1A), it is unlikely that Mgm101 operates in the classic strand invasion mode by acting as a mediator for a RecA/ Rad51-type recombinase. Based on its single strand annealing activity in vitro, the simplest model would be that Mgm101 operates by the strand annealing mode. In this regard, Mgm101 shares similar molecular organization with Rad59. Rad59 has been reported to promote Rad51-independent inverted repeat recombination by a strand annealing mechanism (56).
It is important to note that, unlike the strand invasion mechanism, repair of dsDNA breaks by the single strand annealing mode is error-prone and generates deletions that could be detrimental for the functionality of the genome. Therefore, a more conciliatory model would predict that Mgm101 may promote strand invasion and error-free recombinational repair by an unconventional deployment of its strand annealing activity. mtDNA is rich in ssDNA stretches (57). Instead of invading dsDNA duplex, which is dependent on Rad51/RecA, the Mgm101-ssDNA nucleoprotein filaments may be directly annealed to these homologous single-stranded donor sequences. In this way, although Mgm101 primarily operates in the strand annealing mode, strand invasion could be one of the recombination products during mtDNA repair in vivo. The use of the annealing activity for strand invasion bypasses the requirements for ATP-dependent ssDNA binding and dsDNA remodeling in a conventional recombination reaction.
Shibata and co-workers (19,22) have identified the MHR1 gene, which affects mitochondrial gene conversion after mating of yeast cells carrying different mitochondrial genotypes. Mhr1 is a DNA-binding protein of 27 kDa that lacks recognizable sequence homology to the ATP-dependent Rad51/RecA recombinases. Unlike the currently known DNA recombinases that all form oligomers, Mhr1 is clearly monomeric in solution (supplemental Fig. S4). Mhr1 has been shown to promote ATPindependent D-loop formation in vitro (22), but the implications for mtDNA recombination in vivo need to be further evaluated. Interestingly, Mhr1 has recently been shown to share sequence similarity with Rad54 (58), a dsDNA-dependent helicase that promotes dsDNA unwinding, stimulates homologous pairing, stabilizes the Rad51-ssDNA nucleoprotein filament, and contributes to the late phase of homologous recombination, especially the branch migration of Holliday junctions (13,59). Further studies are required to determine the functional relationship between Mgm101 and Mhr1.
Although the Rad52-type recombination proteins are only conserved at the tertiary and quaternary structural levels rather than at primary sequences (37), close orthologs of Mgm101 are readily identified in lower animals (e.g., Nematostella venctensis) and in Dictyostelium discoideum (see supplemental Fig. S5). For a long time, animal mtDNA has been believed not to recombine. On the other hand, extensive evidence has also been provided that supports mtDNA recombination in species across the animal kingdom, including humans (e.g. see Refs. 60 and 61). Notably, recent electron microscopic studies have revealed the presence of abundant four-way and three-way junctions suggestive of recombination intermediates in the oxyradical-rich human adult heart and brain mitochondria (62)(63)(64). A low level of recombination, which is probably limited between intranucleoidal mtDNA molecules (65), may be in place in animal mitochondria from specific tissues as an effective mechanism for mutational clearance while minimizing the invasion by selfish DNAs (66). RecA orthologs have been identified in plant mitochondria (67). Because fungi, dictyostelids, and animals belong to the unikonts clade on the eukaryotic tree, the Mgm101 orthologs in these species may have arisen from a common ancestral gene during evolution. In light of the present work, it would be interesting to see whether the evolutionary lineage of the Rad52-type proteins, with or without primary sequence similarity to Mgm101, can be extended to the mitochondria of higher animals. More recently, a co-fractionation of the Rad51-type proteins with mitochondria has been reported in carcinoma cell lines (68). This would stipulate the need for the presence of a Rad52-type activity as a recombination mediator in mitochondria. In summary, the data reported in the current work could set the stage for studying Rad52-type proteins in mtDNA recombination in different species.