Gly-103 in the N-terminal domain of Saccharomyces cerevisiae Rad51 protein is critical for DNA binding.

Rad51 is a homolog of the bacterial RecA protein and is central for recombination in eukaryotes performing homology search and DNA strand exchange. Rad51 and RecA share a core ATPase domain that is structurally similar to the ATPase domains of helicases and the F1 ATPase. Rad51 has an additional N-terminal domain, whereas RecA protein has an additional C-terminal domain. Here we show that glycine 103 in the N-terminal domain of Saccharomyces cerevisiae Rad51 is important for binding to single-stranded and duplex DNA. The Rad51-G103E mutant protein is deficient in DNA strand exchange and ATPase activity due to a primary DNA binding defect. The N-terminal domain of Rad51 is connected to the ATPase core through an extended elbow linker that ensures flexibility of the N-terminal domain. Molecular modeling of the Rad51-G103E mutant protein shows that the negatively charged glutamate residue lies on the surface of the N-terminal domain facing a positively charged patch composed of Arg-260, His-302, and Lys-305 on the ATPase core domain. A possible structural explanation for the DNA binding defect is that a charge interaction between Glu-103 and the positive patch restricts the flexibility of the N-terminal domain. Rad51-G103E was identified in a screen for Rad51 interaction-deficient mutants and was shown to ablate the Rad54 interaction in two-hybrid assays (Krejci, L., Damborsky, J., Thomsen, B., Duno, M., and Bendixen, C. (2001) Mol. Cell. Biol. 21, 966-976). Surprisingly, we found that the physical interaction of Rad51-G103E with Rad54 was not affected. Our data suggest that the two-hybrid interaction defect was an indirect consequence of the DNA binding defect.

The DNA within our cells is constantly exposed to DNAdamaging agents, such as ultraviolet light, ionizing radiation (from natural radiation sources or diagnostic and therapeutic medical procedures), environmental chemicals, free radicals generated by cellular metabolism, and mechanical stress on chromosomes during mitosis and meiosis (1). These agents can induce different kinds of DNA damage. One of the most detrimental DNA damages is DNA double-stranded breaks (DSBs), 1 which threaten chromosomal integrity and genetic stability. Failure to accurately repair DSBs may result in chromosomal aberrations, which can lead to cell death (apoptosis) or uncontrolled cell growth (cancer) in mammalian cells (2,3). Several pathways have evolved to repair DSBs to maintain the stability and integrity of the genome: homologous recombination (HR), non-homologous end joining, and single-strand annealing (SSA). Only the HR pathway is inherently error-free, whereas the other pathways often involve sequence changes at the DSB junction (4,5).
The RAD52 epistasis group genes encode the core of the HR pathway (4 -6). After DSB formation redundant pathways involving Exo1, Mre11-Rad50-Xrs2, and possibly other factors process the break to generate 3Ј-OH ssDNA tails. Rad51, Rad52, Rad55-Rad57 heterodimer, and the eukaryotic ssDNAbinding protein complex replication protein A (RPA) cooperate to form the Rad51 pre-synaptic filament on the ssDNA tails. Rad51 filaments recognize homologous dsDNA and perform DNA strand exchange, which are the central steps in HR. Rad54 augments the activities of Rad51 in joint molecule formation and heteroduplex DNA extension (branch migration) in vitro. Genetic and biochemical data show that the Rad51 and Rad54 proteins are key components in recombinational DNA repair since mutants in either gene exhibit an extreme sensitivity to DSBs. In chicken and mammalian cell lines, disruption of Rad51 is lethal (7)(8)(9). Rad51 is a homolog of the Escherichia coli recombination protein RecA and possesses, like RecA, DNAdependent ATPase activity (4 -6, 10). Dmc1 is a meiosis-specific paralog of Rad51 and is specifically required for meiotic recombination (4 -6). Saccharomyces cerevisiae Rad51 protein contains 400 amino acid residues. Its central ATPase domain (core domain) is highly conserved both in primary sequence and three-dimensional structure from E. coli to higher eukaryotes. The N terminus of Rad51 contains a portion that is conserved in Archaea and eukaryotes but is absent from bacterial RecA proteins (10 -12). This N-terminal part of Rad51 folds into a compact, globular domain that functions in dsDNA and ssDNA binding (12,13). In RecA proteins a similar DNA binding domain is located at the C terminus (12,14).
RecA, Rad51, and Dmc1 function in HR by forming nucleoprotein filaments on DNA that are active in homology search and DNA strand exchange. In the absence of DNA and nucleotides, RecA forms either ring structures (15) or compressed filaments with a pitch of ϳ85 Å by self-polymerization as demonstrated by x-ray crystallography (16). In the presence of either ssDNA or dsDNA, RecA forms extended, active nucleoprotein filaments, as characterized by electron microscope (EM) and linear dichroism polarized-light spectroscopy studies (17,18). On dsDNA, the filament contains 6.2 monomers per turn and has a pitch of ϳ95-100 Å (up to 130 Å) (17)(18)(19). ATP or non-hydrolyzable analogs (e.g. ATP␥S, ADP-AlF 4 Ϫ ) are essential cofactors in filament formation. ATP and ADP significantly regulate the conformation of the filament and affect its pitch (18,20). Similar to RecA, eukaryotic and archaeal Rad51/ Dmc1 also adopt ring or filament structures in solution depending on the presence of DNA and ATP (21)(22)(23)(24)(25). The inactive ring (26,27) and active filament structure (13,28,29) of Rad51 have been solved in atomic resolution. The conformational change between active and inactive forms of the RecA filament is caused by dramatic reorientation of subunits (17,18) or a large movement of the C-terminal globular lobe (30). The N-terminal globular domain of Rad51 is connected through an extended elbow linker to the core ATPase domain. This linker ensures the flexibility of the N-terminal domain. Structural alignment of one Rad51 subunit from an active filament (13,28,29) and the first subunit of an inactive ring (26) indicated that the orientation of the N-terminal domain is different in the two different states. The flexibility of the N-terminal domain may be critical for Rad51 function.
Rad54 protein is a member of the SWI2/SNF2 family of DNA-stimulated/dependent ATPases and is conserved in all eukaryotes investigated (31). ATP hydrolysis provides energy for its topological activity on duplex DNA, which is consistent with Rad54 protein tracking along dsDNA (31). Several models have been proposed to account for the stimulation of Rad51 protein-mediated DNA strand exchange by Rad54 protein, including chromatin remodeling, Rad51 filament formation/stabilization, synapsis, and post-synapsis (31). S. cerevisiae Rad54 protein displays chromatin remodeling activities (32,33), although Rad54 protein stimulates the DNA strand exchange activity of Rad51 protein also on non-chromatin templates (34 -36). Rad54 stabilizes the Rad51 filaments on ssDNA (37) and augments the activities of Rad51 in joint molecule formation and heteroduplex DNA extension (branch migration) in vitro (34 -36, 38). Rad54 specifically dissociates Rad51 protein from duplex DNA and may have a role in the turnover of Rad51 filaments stuck on the heteroduplex DNA product formed during HR (39,40). Common to all models is the notion of specific Rad51-Rad54 protein interactions, when Rad51 is bound to ssDNA and/or dsDNA. Interestingly, Rad54 homologs only exist in eukaryotes and possibly some Archaea but not in bacteria.
Rad51 is a multifunctional protein that interacts with DNA, itself, Rad52, Rad55-Rad57 heterodimer, RPA, Rad54, and with Brca2 in humans (6,41). The protein interaction surface on Rad51 may be shared by several proteins, and the interface exchange may control transitions between major events in HR (42). Interaction between the Rad51 and Rad54 proteins has been demonstrated in vitro and in vivo (43)(44)(45). In vitro, Rad54 assists Rad51-mediated recombination by interacting with Rad51 filaments, stimulating the pairing between the nucleoprotein filaments and dsDNA, enhancing the homology search (35,36,46). Likewise, the Rad54 ATPase activity is specifically stimulated by Rad51 protein bound to dsDNA (40). The Rad51 interaction surface is mainly limited to the N-terminal region of yeast and human Rad54 protein (43)(44)(45)47). The Rad54 interaction surface on Rad51 protein is complex and may be involved in multiple protein interactions, including the Rad51 self-interaction and the interaction of human Rad51 with the tumor suppressor protein Brca2 (48 -50). It has been proposed that consecutive protein interactions involving common and overlapping interfaces are critical for the functioning of multistep DNA repair pathways (51)(52)(53). It appears that the Rad51 interaction surface controls multiple, probably mutually exclusive protein interactions that are critical for HR (42).
In budding yeast, two-hybrid assays have identified several point mutations in RAD51, including rad51-G103E, which ablate the Rad51-Rad54 two-hybrid interaction (54). However, the precise molecular deficit of the interaction defect has not been elucidated. Here we have reconstructed the rad51-G103E mutant, overexpressed the protein in budding yeast cells, and purified the Rad51-G103E protein to apparent homogeneity ( Fig. 1). Biochemical analysis showed that Rad51-G103E is defective in DNA strand exchange, exhibiting very low ATPase activity due to a general DNA binding defect. Contrary to the expectation from the two-hybrid analysis, the physical interaction of the mutant protein with wild type Rad54 in solution was unaffected. Our data and structural modeling suggest that the flexibility of the N-terminal domain of Rad51 protein is important for DNA binding. The two-hybrid interaction defect with Rad54 appears to be an indirect consequence of the inherent DNA binding defect.

EXPERIMENTAL PROCEDURES
Yeast Media, Strains, and Plasmids-Standard yeast media were used for cell growth. S. cerevisiae WDHY1930, a rad51 deletion mutant (his3-⌬1 leu2-3,112 trp1 ura3-52 pep4-3 rad51-⌬::KanMX), was created by using pFA-KanMX6 to replace the RAD51 open reading frame from amino acids 5-394 by the KAN resistance marker and used as a host for overexpression of wild type and mutant Rad51 proteins (55). A Rad51-G103E-expressing plasmid was generated by in vitro mutagenesis of plasmid pR51.3 (kind gift of Patrick Sung, Yale University (56)), in which the constitutive phosphoglycerol kinase promoter controls the expression of the wild type RAD51 gene. Mutagenesis was performed using QuikChange site-directed mutagenesis (Stratagene) with the forward primer, 5Ј-CTAAGGGAGAGTGAGCTTCACACTGCTGAAGC-GG-3Ј, and the reverse primer, 5Ј-CCGCTTCAGCAGTGTGAAGCTCA-CTCTCCCTTAG-3Ј. The mutation was confirmed by DNA sequencing.
Protein Overexpression and Purification-S. cerevisiae GST-Rad54 and RPA were purified as described previously (36). The wild type Rad51 protein was purified as described (57). The Rad51-G103E protein was purified from 116 g of yeast cells obtained from 33 liters of culture. The cells were lysed in 100 ml of buffer A (20 mM Tris-HCl, pH 7.5, 1 mM EDTA, 10% glycerol, 10 mM ␤-mercaptoethanol) with 0.5 M KCl and protease inhibitors (1 mM phenylmethylsulfonyl fluoride, 1 M pepstatin, 1 M leupeptin, and 2 mM benzamidine). The lysate was centrifuged for 1 h at 30,000 ϫ g. The supernatant was loaded onto a G-Sepharose column (equilibrated with buffer A with 0.5 M KCl) and washed with buffer A with 1 M KCl. Ammonium sulfate was added to the flowthrough to 40% and stirred for 30 min at 4°C. The Rad51-G103E protein was sedimented by centrifugation at 40,000 ϫ g for 1.5 h. The pellet was resuspended in 150 ml of buffer P (20 mM potassium phosphate, pH 7.5, 1 mM EDTA, 10% glycerol, 10 mM ␤-mercaptoethanol) with 50 mM KCl. The sample was loaded onto a Cibacron blue column equilibrated with buffer P containing 200 mM KCl and eluted with buffer P containing 1 M KCl. Rad51-G103E was in the flow-through fraction from the Cibacron blue column and dialyzed against buffer H (20 mM MES, pH 6.5, 10% glycerol, 1 mM EDTA, and 40 mM KCl) and loaded on a Macro-Prep ceramic hydroxyapatite (HAP, Bio-Rad) column. The HAP column was washed with buffer H containing 20 mM phosphate, and proteins were eluted with buffer H containing a stepped phosphate gradient (20 -200-500 mM). The fractions containing Rad51 were pooled and dialyzed against buffer A containing 100 mM KCl before loading onto a fast protein liquid chromatography (FPLC) Mono Q column (5 ml, Amersham Biosciences, equilibrated with buffer A containing 100 mM KCl). The column was washed with buffer A containing 100 mM KCl and eluted with buffer A containing a 100 -600 mM KCl gradient. The Rad51-containing fractions were dialyzed against Rad51 storage buffer (20 mM Tris acetate, pH 7.5, 0.5 mM EDTA, 10% glycerol, 1 mM DTT, and 100 mM KCl) and kept at Ϫ80°C.
DNA Substrates-RFI X174 and single-stranded poly(dA) were used for the wild type Rad51 and Rad51-G103E ATPase assay. Linearized X174 dsDNA and short ssDNA (621 nt) were used for DNA binding analysis of the Rad51 proteins. Linearization and labeling of RFI X174 DNA was performed as described (36). The 621-nt linear ssDNA fragment was synthesized by run-off PCR from a PstI-linearized X174 DNA template in 40 cycles of non-exponential Taq polymerase extension from a single primer (5Ј-GTC TTC ATT TCC ATG CGG TG-3Ј). Fragments were purified using a PCR purification kit (Qiagen) and radiolabeled to high specific activity using T4 DNA polynucleotide kinase. The activity of radiolabeled DNA substrate was adjusted by spiking to 10,000 -20,000 cpm per reaction with a known molar amount of unlabeled DNA substrate. DNA concentrations are expressed in mol of nt for ssDNA and mol of bp for dsDNA. ATPase Activity-A charcoal-based ATPase assay was used as described (36) with minor modifications. The reaction buffer contained 33 mM Tris-HCl, pH 7.5, 13 mM MgCl 2 , 1.8 mM DTT, 1 mM ATP, 90 g/ml bovine serum albumin, and 2 M Rad51/Rad51-G103E. DNA concentration was varied as indicated. A 50-l reaction system contained 0.1 Ci of [␥-32 P]ATP. The reaction was started by transferring the sample tubes from ice to a 30°C water bath and incubated for 30 min. The reaction was stopped by adding 0.5 ml of Norite charcoal solution, which contained 5% Norite charcoal (Sigma, C-5260) and 50 mM KH 2 PO 4 . The samples were kept on ice for 5 min, vortexed once during the incubation. Then the samples were centrifuged at maximum speed using an Eppendorf bench top centrifuge, and 0.4 ml of the supernatant was thoroughly mixed with 4 ml of Aquasol. The free ␥-32 P from [␥-32 P]ATP hydrolysis was counted on a Beckman LS6500 multipurpose scintillation counter. The ATPase activity was expressed as the ratio of radioactivity in free phosphate relative to the total radioactivity added in the sample.
DNA Strand Exchange Assay-The DNA strand exchange reaction was performed as described (36). The reaction buffer contained 30 mM Tris acetate, pH 7.5, 1 mM DTT, 20 mM ATP, 20 mM magnesium acetate, 20 mM phosphocreatine, 0.1 g/l creatine kinase, 50 g/ml bovine serum albumin, 2.37 mM spermidine. The reaction was started with a buffer containing 33 M (nt) X174 ssDNA and 10 M Rad51/Rad51-G103E and a total volume of 10.5 l. After incubation at 30°C for 15 min, 1.8 M RPA was added, and incubation was continued at 30°C for another 30 min. At this point either linearized X174 dsDNA (16 M bp) only or the linearized X174 dsDNA together with Rad54 (0.2 M) was added, and incubation was continued at 30°C for 4 h. The reaction was stopped by adding 2 l of stop buffer (0.714% SDS, 357 mM EDTA, and 4.3 mg/ml proteinase K) and incubated at 30°C for another 30 min. Finally, the sample was separated on a 0.8% agarose gel, stained with ethidium bromide, visualized on Eagle Eye II (Stratagene), and analyzed with ImageQuant 5.2 software (Amersham Biosciences).
DNA Binding Assay-A nucleoprotein gel assay was used to analyze the DNA binding activity of the Rad51 proteins. The method is based on a difference in electrophoretic mobility of glutaraldehyde-fixed DNA-Rad51 complexes, which is related to the nucleoprotein filament saturation (40). DNA-Rad51 filaments were formed with the indicated amounts of Rad51 protein and DNA (30 M bp with full-length linear X174 or 2 M nt of 5Ј-[ 32 P]-end-labeled linear ssDNA (621 nt long) in buffer containing 25 mM triethanolamine acetate, pH 7.5, 13 mM magnesium acetate, 1.8 mM DTT, 5 mM ATP, 100 g/ml bovine serum albumin by incubation at 30°C for 30 min (15 min for oligonucleotides). Samples were then fixed with glutaraldehyde (final concentration 0.25%) for 30 min at 30°C (10 min for oligonucleotides), loaded onto agarose gels (1%, 1ϫ Tris-buffered EDTA), and run for 2 h (90 min for short oligos) at 80 V (4 V/cm). The DNA in agarose gels was either stained with ethidium bromide and recorded using Eagle Eye II (Stratagene) or dried and visualized using a PhosphorImager (Amersham Biosciences Storm 840). The quantification was performed with Image-Quant 5.2 software (Amersham Biosciences).
Physical Interaction of Rad51 and Rad54 -Reactions containing 0.25 M Rad51 and 0.125 M GST-Rad54 were incubated at room temperature (23°C) for 1 h in 250 l of reaction buffer containing 25 mM Tris-HCl, pH 7.5, 50 mM NaCl, 1 mM DTT, 10% glycerol, and 0.05% Nonidet P-40. Glutathione-Sepharose 4B (50% v/v, 40 l) slurry was then added that had been pretreated with 0.5 mg/ml bovine serum albumin for 1 h at room temperature. The reaction containing the slurry was incubated at room temperature for another hour. The slurry and supernatant were separated by centrifugation (Eppendorf bench top centrifuge, 2300 rpm, 5 min). Proteins in both fractions were monitored by SDS-PAGE. Proteins in the supernatant were precipitated in 15% trichloroacetic acid before solubilization in Lä emmli buffer. The Coomassie-stained gel was dried and scanned, and the Rad51 and Rad54 bands were quantified with ImageQuant 5.2 software (Amersham Bio-sciences). The interaction between Rad51 and Rad54 is expressed as the ratio of Rad51 to Rad54 signal in the pellet fraction, making the quantitation of the interaction independent of absolute protein amounts.
Structure Analysis and Computer Modeling-The x-ray crystal structures of S. cerevisiae Rad51-⌬(1-79)-I345T (PDB code 1SZP) (13), archaeal Pyrococcus furiosus PfRad51 (PDB codes 1PZN) (26), and human Rad51 (PDB codes 1N0W) (48) were used in structural alignment in ICMLite (www.molsoft.com). The subunit folding of the core domain of the three structures is highly conserved. Although the folding of the N-terminal domain is also conserved, its orientation shows slight variation between the active and inactive forms. 1SZP represents an active Rad51 filament. 1PZN represents an inactive structure of Rad51. Based on 1PZN subunit A, a structural model of S. cerevisiae Rad51-G103E (Fig. 8B) was created by homology modeling in Modeller 7v7 (58,59). The main difference between the active and inactive structure is the orientation of the flexible N-terminal portion. The quality of the model was examined by Procheck (60), Whatif (61), ProsaII (62), and Verify_3D (63,64). The image of the structural model was generated using PyMol (www.pymol.org; Ref. 65) (Fig. 8, B-D).

RESULTS
Residue glycine 103 of budding yeast Rad51 protein maps in the N-terminal domain that was proposed to bind to DNA based on studies with the N-terminal domain of human Rad51 protein (12). The Rad51-G103E mutant was shown to be deficient in its interaction with Rad54 in two-hybrid analysis (54). To understand the function of S. cerevisiae Rad51 protein and its interaction with Rad54 protein, we reconstructed the mutant, overexpressed the mutant protein in yeast, and purified the protein to apparent homogeneity (Fig. 1). Purified wild type and Rad51-G103E mutant protein were analyzed in side-byside biochemical assays to determine the function of glycine 103 in Rad51 protein and the molecular defects of the Rad51-G103E mutant protein.
Rad51-G103E Lacks DNA Strand-exchange Activity-Rad51 protein catalyzes the DNA strand exchange reaction, which recapitulates central steps in homologous recombination ( Fig.  2A). Rad51 forms a nucleoprotein filament of circular ssDNA that searches for homology on duplex DNA leading to strand invasion and DNA strand exchange with the linear duplex DNA partner. Joint molecule intermediates and nicked circle products are formed optimally at a ratio of 1 Rad51 protomer per 3 nucleotides, representing a fully saturated nucleoprotein filament. The DNA strand exchange activity of Rad51 protein is stimulated by Rad54 protein involving specific protein interactions (34,36,39). We expected a Rad54 interaction-deficient Rad51 mutant protein to perform DNA strand exchange in vitro but lack the stimulation by Rad54 protein. To our surprise, the purified Rad51-G103E did not show any DNA strand-exchange activity under a variety of protein to DNA stoichiometries ranging from oversaturation (1 Rad51 per 1.5 nucleotides) to subsaturation (1/12 nucleotides) (Fig. 2B). The addition of Rad54 protein did not recover DNA strand exchange activity by Rad51-G103E protein, whereas Rad54 stimulated the DNA strand exchange activity of wild type Rad51 protein as expected (Fig. 2C). We conclude that Rad51-G103E lacks DNA strand exchange activity.
The G103E Mutation Impairs DNA-dependent ATPase Activity of Rad51-To understand the defect of the Rad51-G103E mutant protein in DNA strand exchange, we examined its ATPase activity. Wild type Rad51 protein is a DNA-dependent ATPase that exhibits ATPase activity when bound to ssDNA or dsDNA (56). ATP binding and hydrolysis are important for the activity of Rad51 protein in HR (10). The ATPase activity of the purified Rad51-G103E mutant protein was analyzed in the presence of ssDNA (poly(dA)) ( Fig. 3A) and dsDNA (RFI X174) (Fig. 3B) in direct comparison with wild type Rad51 protein. Although both forms of DNA support the ATPase activity of wild type Rad51, the ssDNA-based activity is 3.75-10 times higher than that of dsDNA-based activity, as previously shown (56). The Rad51-G103E mutant protein significantly inhibits the DNA-dependent ATPase activity over a wide range of protein-to-DNA ratios. We conclude that Rad51-G103E protein is severely defective in its ATPase activity.
The G103E Mutation Affects DNA Binding by Rad51-Glycine 103 resides outside the ATPase core domain in the Nterminal domain that was proposed to function in DNA binding (12). Hence, we suspected that a DNA binding defect might be the root cause of the DNA strand exchange and ATPase defect of Rad51-G103E protein because the Rad51 ATPase activity is dependent on DNA binding. DNA binding was examined in nucleoprotein gel assays with dsDNA (linearized X174) (Fig.  4A) or ssDNA (a 621 nucleotide linear PCR product) (Fig. 4B). Wild type Rad51 protein forms nucleoprotein filaments on ssDNA and duplex DNA that require stabilization by glutaraldehyde cross-linking to be visualized by gel electrophoresis on agarose gels (Fig. 4). DNA binding saturates at 1 Rad51 protomer per 3 nt or bp but wild type Rad51 protein forms partial filaments under substoichiometric conditions. For example, at 1 Rad51 protomer per 24 bp, 80% of the input duplex DNA was protein-bound (Fig. 4A), and at 1 protomer per 8 nt more than 70% of the ssDNA was protein-bound (Fig. 4B). A protein titration of the Rad51-G103E mutant protein revealed a defect in binding to duplex DNA (Fig. 4A) and ssDNA (Fig. 4B). Much higher protein concentrations were required to initiate the formation of protein-DNA complexes. Moreover, the electrophoretic mobility of the Rad51-G103E-duplex DNA complexes at sub-saturating conditions (1/24 -1/6 bp) was slower than that of the wild type complexes or the fully saturated mutant complex, suggesting that the protein DNA filament formed under these conditions may have adopted a variant structure. With ssDNA, very high Rad51-G103E concentrations (2.5 protomer per nt) were required to efficiently form nucleoprotein filaments. Rad51 binding to DNA is cooperative. The lack of partial filaments under conditions where not all substrate DNA was bound (dsDNA, 1/12, 1/8, 1/6 bp; ssDNA, 1/0.4 nt; Fig. 4) suggests that Rad51-G103E retained the cooperative binding typical for the wild type Rad51 protein.
Another method to evaluate DNA binding is determining the salt titration mid-point (66). The salt titration mid-point is the salt concentration at which 50% of the DNA exists in a protein- bound state. We determined the salt titration mid-point of protein-DNA complex formation with linearized X174 dsDNA under a saturating stoichiometry of 1 Rad51 protomer per 3 bp that allowed complete binding of all available duplex DNA by wild type and Rad51-G103E mutant protein (Fig. 4). The salt titration mid-point for wild type Rad51 was determined to be 168 mM NaCl, which was significantly decreased to 51 mM for the Rad51-G103E protein (Fig. 5). The results of the DNA binding experiments demonstrate a general DNA binding defect of the Rad51-G103E protein. The formation of protein-DNA complexes requires higher protein concentration, is more salt-sensitive, and appears to generate aberrant complexes (with dsDNA) as judged by their electrophoretic mobility.
The ability of Rad51-G103E-DNA Filaments to Stimulate Rad54 ATPase Activity Is Altered-Partial filaments of wild type Rad51-DNA filaments stimulate Rad54 ATPase activity up to 6-fold (40). The extent of the stimulation depends on the saturation of DNA and has an optimal stoichiometry of ϳ40 bp per Rad51 protomer. Under these conditions, short Rad51 filaments are formed, leaving significant segments of the duplex DNA protein free (40). The stimulation of the Rad54 ATPase is mediated by specific protein interactions, and the contribution of the Rad51 ATPase to the overall ATPase activity is negligible under these conditions (40). Here, we analyzed the interaction of Rad51-G103E with Rad54 (Fig. 6). In the reaction system Rad54 and DNA concentrations were fixed, and the amount of Rad51 was gradually increased. The data for the wild type Rad51 protein were in accordance with the previous analysis (40). The Rad51-G103E did not show measurable stimulation of Rad54 ATPase until a stoichiometry of 1 Rad51-G103E protomer per 6 bp was reached. At very high concentrations of Rad51-G103E (1 protomer per bp), the ATPase activity of the system increased almost to wild type levels. These data are consistent with the DNA binding data of Fig. 4 showing that Rad51-G103E forms protein-DNA complexes at high protein concentrations. The stimulation of the Rad54 ATPase suggests that the interaction between Rad51 and Rad54 is largely intact when Rad51 is bound to dsDNA.
The G103E Mutation Does Not Affect the Physical Interaction of Rad51 and Rad54 in Solution-The physical solution interaction between Rad51-G103E and Rad54 was investigated using a GST pull-down assay (Fig. 7A). Purified GST-Rad54 and  Rad51-G103E were incubated at room temperature to allow complex formation between the GST-Rad54 and Rad51 proteins. Glutathione-Sepharose beads were subsequently added to the system, and incubation was continued for another hour. The Sepharose-GST-Rad54-Rad51 complexes formed during incubation were pulled down by centrifugation. The complexes were washed three times with buffers containing different salt concentrations to determine the stability of Rad51-Rad54 complexes to salt. The amount of Rad51/Rad51-G103E found in complex with GST-Rad54 was quantified and showed no difference between the wild type and mutant Rad51 proteins (Fig. 7, B and C). A control experiment demonstrated that the GST part of the Rad54 fusion protein did not interfere with the experiment (Fig. 7B). Both the wild type and Rad51-G103E mutant protein bound Rad54 tightly, and 1 M NaCl did not dissociate the complexes completely. These results indicated that the G103E mutation did not measurably affect the physical interaction between the Rad51 and Rad54 proteins in solution.
G103E May Affect the Flexibility of the Rad51 N-terminal Domain-Eukaryotic Rad51 protein is a homolog of the E. coli RecA protein. RecA protein has a central ATPase domain (core domain) that binds and hydrolyzes ATP and binds to ssDNA. At the RecA N terminus there is an extended helical domain that is mainly involved in the subunit-subunit interaction. The C-terminal portion contains a compact globular domain that is thought to be involved in dsDNA binding. Analysis of Rad51/ RecA sequences from different organisms (10,11) and structures of RecA (16) and Rad51 homologs from Archaea (P. furiosus PfRad51 (26) and Methanococcus voltae (28)), S. cerevisiae (13), and human (27,48) showed that the core domain is highly conserved from primary sequence to tertiary structure, as noted previously (10) (Fig. 8B). In eukaryotic cells a highly conserved N-terminal portion forms a compact domain, which seems to play a similar role as the C-terminal compact domain of RecA (Fig. 8A). This domain has been shown to bind ssDNA and dsDNA in human Rad51 (12). The N terminus and the core domain are connected through an extended elbow linker that is composed of a short helical segment flanked by a ␤-strand and random coil on both sides (Fig. 8, A  and B). In the active filament (S. cerevisiae Rad51, 1SZP) and inactive ring structure of PfRad51 (1PZN), this flexible linker assumes slightly different orientations, whereas the N terminus undergoes a significant shift (13,26,29). We built a Rad51 model based on the A subunit of the inactive PfRad51 ring structure (1PZN) (Fig. 8C). In this model the N-terminal domain is spatially close to the core domain, facing a subdomain formed by double helices on the core domain. At least three positively charged residues (Arg-260, His-302, Lys-305) sit closely on the double helical subdomain (Fig. 8, C and D). Gly-103 is almost absolutely conserved in the N terminus of all Rad51 proteins, with Drosophila Rad51 being the only exception having a serine at this position (Fig. 8A). Gly-103 resides on the surface of the N-terminal domain, which directly faces the positively charged area. A charge-charge interaction between the G103E residue of the N-terminal domain and the positive charge cluster of the core domain of Rad51 protein may be possible. Such an interaction may restrict the flexibility of the N-terminal domain, which in turn might affect the dynamic equilibration between active and inactive Rad51, possibly affecting DNA binding of the mutant protein.

DISCUSSION
Glycine 103 of the S. cerevisiae Rad51 protein is a highly conserved residue in the N-terminal domain of Rad51 proteins with the exception of the Drosophila melanogaster Rad51 protein, which has a serine residue at that position (Fig. 8A). Using purified proteins and side-by-side biochemical assays, we identified that the Rad51-G103E protein is defective in DNA binding , which leads to a defect in the Rad51 ATPase activity (Fig. 3) and complete absence of DNA strand exchange activity (Fig. 2). These data show that Gly-103 and the N-terminal domain are critical for the biochemical functions of Rad51. This is consistent with and extends the previous analysis of the N-terminal domain of the human Rad51 protein, which showed that the N-terminal domain could bind ssDNA and dsDNA (12). The biochemical defect of the Rad51-G103E protein can explain the in vivo defect observed with a plasmidborne copy of rad51-G103E, which failed to complement the MMS sensitivity of a rad51⌬ mutant (54).
The DNA binding defect of Rad51-G103E appears to be the root cause for the ATPase and DNA strand exchange deficiency. With bacterial RecA protein, nucleation (binding of the first subunit) of the filament is the rate-limiting step in DNA binding, and the subsequent cooperativity in DNA binding results in efficient filament formation (10). Rad51 protein also binds DNA in a cooperative fashion (see Figs. 4 and 5), although its cooperativity appears to be lower than that of RecA protein (10,40). The analysis of Rad51-DNA complexes on agarose gels demonstrated that much higher Rad51-G103E concentrations were needed for complex formation but that cooperative binding was maintained, as evidenced by the lack of protein-DNA complexes with low saturation running with intermediate (between protein-free and fully saturated filaments) electrophoretic mobility (Fig. 4). Rad51-G103E-DNA complex formation was also found to be more saltsensitive, as shown by a significantly lower (3-fold) salt titration midpoint than wild type protein (Fig. 5). The profile of the protein-DNA complexes formed under various salt concentrations showed that cooperative DNA binding was maintained in Rad51-G103E protein. The electrophoretic profile of the Rad51-G103E-dsDNA complexes was aberrantly slow compared with wild type complexes under substoichiometric (Ͼ6 bp per Rad51 protomer, Fig. 4A) conditions, suggesting that the filaments formed were abnormal. Moreover, under conditions where Rad51-G103E formed complexes with dsDNA that were quantitatively and qualitatively indistinguishable from wild type complexes (3 bp per Rad51 protomer) in their electrophoretic profile, Rad51-G103E-dsDNA complexes still did not exhibit dsDNA-dependent ATPase. We conclude that Rad51-G103E affects DNA binding and forms non-functional Rad51 filaments. Molecular modeling of the Rad51-G103E mutant protein based on the available Rad51 crystal structures (Fig. 8) suggests a structural explanation for the observed DNA binding defect and underlines an important facet in the structural organization of the Rad51 protein, which is the flexible connection between the Rad51 N-terminal domain and the core ATPase domain (13). The major difference in primary structure between eukaryotic Rad51 and RecA is that the eukaryotic Rad51 contains an N-terminal domain, which is absent in RecA. Structurally, the Rad51 Nterminal domain forms a globular lobe as analyzed by EM image reconstruction, which is very similar to that formed by the RecA C-terminal domain (30). The N-terminal domain of Rad51 is also functionally related to the C-terminal domain of RecA, as both are involved in DNA binding (12,30,(67)(68)(69). The RecA/Rad51 family proteins form two kinds of multisubunit structures, either inactive ring complexes or active nucleoprotein filaments (10, 15, 18, 21, 24, 25, 68 -70). Two ring structures (from P. furiosus PfRad5 (26) and human (27)) and three filament structures (from M. voltae (28), S. cerevisiae (13), and Sulfolobus solfataricus (29)) have been solved in atomic resolution. The crystal structures show that in each subunit of Rad51, its N-terminal domain and the central ATPase domain are connected through an extended elbow linker, which is flexible and makes the N-terminal domain movable. Comparison of the subunits from the inactive ring complex and the active filament revealed that the difference between the two structures is the localization of the N-terminal globular domain relative to the core ATPase domain. In the inactive form, the N-terminal domain is much closer to the core domain and directly sitting on top of the double helical subdomain of the core domain (26). In the S. solfataricus RadA x-ray crystal structure, FIG. 7. Physical interaction with Rad54. A, scheme of the GST pull-down assay. Purified GST-tagged Rad54 protein was incubated with purified Rad51 or Rad51-G103E protein at room temperature for 1 h to allow the formation of protein complexes. Glutathione-Sepharose 4B beads were added to the reaction, and the incubation was continued for another hour. Finally, the Rad51-Rad54 complexes were isolated by centrifugation and washed with buffers containing 50 mM to 1 M NaCl. The amount of Rad51 or Rad51-G103E protein in complexes with Rad54 protein was determined after 10% SDS-PAGE and Coomassie staining. B, Coomassie-stained 10% SDS-PAGE of Rad51/Rad51-G103E-Rad54 complexes (a and b, respectively). c, GST control showing no binding between GST and Rad51. The experiment was performed under the same condition as in a and b but washed with buffer containing 50 mM NaCl. C, quantitation of the gels shown in B with the means and S.D. from three experiments. Black bars, wild type Rad51; gray bars, Rad51-G103E. the N-terminal domain is shifted very far away from the ATPase domain, and the double helical subdomain on the ATPase domain is completely open, forming a filament with only three subunits per turn (29). The elbow linker domain is composed of a short ␤-strand and an ␣-helix (Fig. 8), which is involved in the subunitsubunit interaction (13,26,28,29). During filament formation, the linker domain itself appears rigid (29). However, the Nterminal domain appears to undergo a large shift. The N-terminal movement can either close (inactive state) or open the double helical subdomain on the ATPase domain. The latter seems the favorable state for filament formation. Three positively charged residues, Arg-260, His-302, and Lys-305, form a positively charged patch on the double helical subdomain that is accessible to the N-terminal domain. When the N-terminal domain moves to the vicinity of the positive charges (as shown in the inactive model) (Fig. 8, C and D), the Glu-103 residue may interact with the positive charges and keep the monomer protein in a closed state. Therefore, the protein is "frozen" in an inactive state and can no longer respond to DNA and ATP. This may explain the defect in DNA binding of Rad51-G103E and why the Rad51-G103E filament that is formed under high protein concentration is nonfunctional.  (73), and the picture was generated in Bioedit (74). The S. cerevisiae is highlighted by a dot on the left-hand side. B, superimposed Rad51 subunits from S. cerevisiae (1SZP (yellow)), human (1N0W (green)), and the archael P. furiosus (1PZN, magenta). The tertiary structure of the ATPase core domain of the Rad51 proteins from different organisms (including E. coli RecA, not shown) is highly conserved. The N-terminal DNA binding domain, which is connected to the ATPase core domain by an elbow linker, is conserved in eukaryotic and archaeal Rad51 proteins but not in bacterial RecA protein (A). Superimposition was done in ICMlite (75). C, a homology model of the S. cerevisiae Rad51-G103E was built in Modeller 7v7 (58,59)  Rad51-G103E was initially identified in a screen for rad51 mutants that exhibited a protein interaction defect in a twohybrid system (54). Rad51-G103E exhibited a highly specific interaction defect with Rad54 protein but wild type interactions with wild type Rad51, Rad52, and Rad55 (54). We expected Rad51-G103E protein to be defective in its interaction with Rad54 but found that the solution interaction between Rad51-G103E and Rad54 was as efficient and resistant to increasing ionic strength as the interaction between the wild type proteins (Fig. 7). This interpretation is supported by the finding that Rad51-G103E protein, when bound to DNA (i.e. at excess protein concentration; Fig. 6), was able to stimulate the Rad54 dsDNA-dependent ATPase nearly as efficiently as partial filament of the wild type Rad51 protein. These data suggest that the G103E mutation does not disrupt the Rad54-Rad51 protein interaction. The yeast two-hybrid system has been invaluable in identifying biologically relevant protein interactions (71) but has been marred with high frequencies of false negatives and false positives (72). False-negative two-hybrid results are known to be caused by defective protein folding or nuclear import. Based on the data described above, we conclude that Rad51-G103E is a DNA binding-deficient mutant and that the Rad54 interaction-deficient phenotype identified in twohybrid analysis was an indirect outcome of DNA binding deficiency, suggesting that the initial interaction Rad51-Rad54 interaction in the two-hybrid system was dependent on Rad51 DNA binding.