Structural Basis of Homology-Directed DNA Repair Mediated by RAD52.

RAD52 mediates homologous recombination by annealing cDNA strands. However, the detailed mechanism of DNA annealing promoted by RAD52 has remained elusive. Here we report two crystal structures of human RAD52 single-stranded DNA (ssDNA) complexes that probably represent key reaction intermediates of RAD52-mediated DNA annealing. The first structure revealed a "wrapped" conformation of ssDNA around the homo-oligomeric RAD52 ring, in which the edges of the bases involved in base pairing are exposed to the solvent. The ssDNA conformation is close to B-form and appears capable of engaging in Watson-Crick base pairing with the cDNA strand. The second structure revealed a "trapped" conformation of ssDNA between two RAD52 rings. This conformation is stabilized by a different RAD52 DNA binding site, which promotes the accumulation of multiple RAD52 rings on ssDNA and the aggregation of ssDNA. These structures provide a structural framework for understanding the mechanism of RAD52-mediated DNA annealing.

Introduction

DNA double-strand breaks (DSBs) are formed by reactive oxygen species and ionizing radiation, as well as by replication over single-strand nicks and inadvertent actions of nuclear enzymes (Lieber, 2010, Mehta and Haber, 2014). Homology-directed repair (HDR) is an important mechanism for repairing DSBs (Ceccaldi et al., 2016, Pâques and Haber, 1999). Properly functioning HDR is critical for genomic stability. In eukaryotes, the failure of HDR can lead to gross changes in the chromosome structure, including an abnormal number of chromosomes (aneuploidy) and the loss of heterozygosity. These chromosomal alterations can have serious consequences, such as tumorigenesis (Jackson and Bartek, 2009, Moynahan and Jasin, 2010). The importance of HDR is underscored by the high conservation, from yeast to humans, of the proteins that catalyze the reaction (San Filippo et al., 2008, Krogh and Symington, 2004).

Multiple reaction pathways have been proposed for HDR (Heyer et al., 2010). They include double-strand break repair (DSBR), synthesis-dependent strand annealing (SDSA), break-induced replication (BIR), single-strand annealing (SSA), and alternative end joining (alt-EJ). Most of them involve the formation of a D-loop structure, which is a duplex DNA containing an invading homologous single-stranded DNA (ssDNA). The RecA/Rad51 recombinase family plays a central role in catalyzing this reaction. The formation of base pairs at the proper location between the invading ssDNA and the double-stranded DNA (dsDNA) is critical for the accurate repair. Depending on the HDR pathway, DNA annealing between complementary ssDNA regions can also influence the outcome of HDR (Bhargava et al., 2016, Ivanov et al., 1996). DNA annealing may occur spontaneously, but several proteins, such as Rad52, RecO, UvsY, RecT, Redβ, and ICP8, catalyze the reaction (Morrical, 2015).

Rad52 plays a multi-faceted role in HDR, and functions in both Rad51-dependent and Rad51-independent pathways. In yeast, mutations in Rad52 result in a severe recombination-deficient phenotype, indicating the importance of Rad52 in HDR (Game and Mortimer, 1974). Yeast Rad52 is believed to play two roles in HDR: facilitating the replacement of RPA with Rad51 on ssDNA (Song and Sung, 2000, Sugiyama and Kowalczykowski, 2002) and annealing cDNA strands (Mortensen et al., 1996, Sugiyama et al., 1998). The former role is correlated with the polymerization of Rad51 on ssDNA before D-loop formation and DNA strand exchange (Sung, 1997, Shinohara and Ogawa, 1998, New et al., 1998). By contrast, the latter role is probably more widely utilized in various HDR pathways, including DSBR, SDSA, and SSA. In humans, it is not clear whether RAD52 functions as a mediator of RAD51 (San Filippo et al., 2008). However, like its yeast counterpart, RAD52 promotes DNA annealing (Reddy et al., 1997), which indicates that DNA annealing is a more widely conserved function among Rad52 orthologs. Human RAD52 is also capable of catalyzing the formation of D-loops in vitro (Kagawa et al., 2001), which may be a relevant activity in certain RAD51-independent pathways. Recently, RAD52 was demonstrated to function in the repair of collapsed DNA replication forks, via RAD51-independent BIR (Sotiriou et al., 2016, Bhowmick et al., 2016, Ciccia and Symington, 2016). Furthermore, RAD52 was implicated in promoting DNA repair in cancer cells, in which the RAD51-dependent repair pathway is defective (Feng et al., 2011, Lok et al., 2013). Although the precise role of RAD52 in these repair pathways remains obscure, understanding the molecular details of RAD52-mediated DNA annealing will be important for clarifying its functions in HDR (Hanamshet et al., 2016).

Yeast and human Rad52 form oligomeric rings (Shinohara et al., 1998, Stasiak et al., 2000, Ranatunga et al., 2001). Electron microscopic visualizations of Rad52 in the presence of ssDNA revealed that Rad52 forms aggregates on ssDNA and suggested that Rad52 rings engage in side-by-side interactions (Van Dyck et al., 1998). A positively charged groove running around the ring structure was identified from the crystal structures of the human RAD52 protein (Kagawa et al., 2002, Singleton et al., 2002). Mutagenesis studies suggested that this groove is a potential ssDNA binding site (Kagawa et al., 2002, Lloyd et al., 2005). Subsequent biochemical studies revealed that a region outside of the groove is a second DNA binding site (Kagawa et al., 2008). Mutations in this site impair the abilities of RAD52 to anneal complementary ssDNA strands, promote D-loop formation, and induce positive supercoils in DNA. Although these observations, along with DNA footprinting studies (Parsons et al., 2000, Singleton et al., 2002), are consistent with the view that the ssDNA wraps around the RAD52 ring, the precise path and structure of the ssDNA along the RAD52 surface is unclear. Furthermore, two DNA annealing mechanisms are conceivable: a cis mechanism, in which the complementary ssDNA strands bind and anneal on a single RAD52 ring (Kagawa et al., 2008), and a trans mechanism, in which annealing is promoted between two RAD52-ssDNA complexes (Rothenberg et al., 2008, Grimme et al., 2010). Further insights into the mechanism of the RAD52-mediated homology search require the structural details of the RAD52-ssDNA interactions.

In the present study, we determined two complex structures of human RAD52 and ssDNA, one in which the ssDNA is bound to the inner DNA binding site and the other wherein the ssDNA interacts with the outer DNA binding site of RAD52. The structures revealed clear differences in the ssDNA binding modes of the two DNA binding sites and suggest the means by which the inner and outer DNA binding sites participate in DNA annealing.

Results

Structure of ssDNA Bound to the Inner DNA Binding Site of RAD52

The inner and outer DNA binding sites were previously identified from structural and mutagenesis studies of the N-terminal half of RAD52 (Kagawa et al., 2002, Kagawa et al., 2008). To understand the roles of these two DNA binding sites in RAD52-mediated DNA annealing, we first crystallized a 40-nucleotide ssDNA bound to the inner DNA binding site of the RAD52 homo-oligomeric ring and determined the structure of the complex by X-ray crystallography (Table 1). For crystallization, we used a truncated construct of the human RAD52 protein (referred to as RAD52 hereafter), lacking the C-terminal half (amino acid residues 213–418). The C-terminal half is structurally unstable, as revealed from previous limited proteolysis experiments (Kagawa et al., 2001), and prevented the full-length RAD52 from crystallizing. Importantly, multiple studies have shown that the full-length and C-terminally truncated RAD52 proteins display highly similar DNA binding properties (Parsons et al., 2000, Singleton et al., 2002, Kagawa et al., 2001, Kagawa et al., 2002, Lloyd et al., 2005). We incorporated two alanine substitutions distant from the inner DNA binding site (RAD521−212 K102A/K133A), which were essential for obtaining crystals that yielded interpretable electron density maps of the ssDNA (see Transparent Methods for details).

Table 1

Summary of Data Collection and Refinement Statistics

	RAD52-ssDNA (Outer DNA Binding Site) (PDB: 5XS0)	RAD52-ssDNA (Inner DNA Binding Site) (PDB: 5XRZ)
Data Collection
Space group	P21	P1
Cell dimensions
a, b, c (Å)	100.22, 164.41, 166.09	67.62, 101.42, 101.71
α, β, γ (˚)	90.0, 90.19, 90.0	91.9, 108.7, 108.2
Resolution (Å)	48.5–3.0 (3.05–3.00)a	95.3–3.6 (3.7–3.6)a
Rmerge	0.073 (0.596)	0.117 (0.497)
I/σ(I)	11.8 (1.8)	7.5 (2.7)
CC1/2	0.997 (0.733)	0.972 (0.597)
Completeness (%)	99.9 (98.7)	96.9 (97.1)
Redundancy	3.8 (3.8)	3.3 (3.4)
Refinement
Resolution (Å)	48.5–3.0	45.7–3.6
No. reflections	107,485	26,956
Rwork/Rfree	0.221/0.251	0.217/0.257
No. atoms
Protein	31,852	15,862
DNA	615	794
K+	0	10
B factors
Protein	78	121
DNA	90	143
K+	–	118
RMS deviations
Bond lengths (Å)	0.003	0.002
Bond angles (°)	0.571	0.480

One crystal was used for each structure.

RMS, root mean square.

Values in parentheses are for highest-resolution shell.

As predicted in previous studies, the ssDNA was accommodated inside the groove (Figure 1A). The electron density for the entire length of the ssDNA was clearly observed (Figure S1). The 40 nucleotides of ssDNA spanned across 10 subunits of the RAD52 ring (Figure 1B), and thus each subunit accommodated four nucleotides of the ssDNA. The phosphate backbone of the ssDNA was near the bottom of the groove, and the bases faced outward toward the solvent (Figures 1C–1E). This DNA binding mode is consistent with the fact that the deepest part of the groove is highly positively charged and appears to be suited for electrostatic interactions with ssDNA (Kagawa et al., 2002, Singleton et al., 2002). Inside the groove, the distances between the side chain amino or guanidinium groups of Lys152, Arg153, and Arg156 and the phosphate backbone of the ssDNA were between 2.3 and 3.3 Å, indicating direct interactions between these residues and the ssDNA (Figure 1E).

Figure 1

“Wrapped” Configuration of the ssDNA Around the Oligomeric RAD52 Ring

(A) Side view of the RAD52-ssDNA complex. To clearly depict the inner DNA binding site, the oligomeric RAD52 ring is shown in a surface representation (light blue). The ssDNA (red stick representation) is bound deeply within the cleft of RAD52 and occupies most of the space inside the groove.

(B) The RAD52-ssDNA complex, viewed down the central channel of the ring. The 40-nucleotide ssDNA (5′ to 3′ counterclockwise orientation) spans across 10 RAD52 subunits. Each subunit is colored differently.

(C and D) Two views of three consecutive RAD52 subunits and the 12-nucleotide ssDNA region that spans the subunits.

(E) A close-up view of the electrostatic interactions of K152, R153, and R156 with the phosphate backbone of the ssDNA. Dashed lines (magenta) depict potential hydrogen bonds.

(F) Hydrophobic stacking interactions that sandwich the four-nucleotide repeats. The β-hairpin structure of RAD52 (amino acid residues 52–66) is located between the four-nucleotide repeats. The ssDNA bases are sandwiched between R55 and V63 of the β-hairpin. For clarity, the DNA bases and the side chains of R55 and V63 are shown in sphere representations.

See also Figures S1 and S2.

The oligomerization state of the RAD52 ring was unaffected by ssDNA binding. Like the DNA-unbound RAD52, the DNA-bound RAD52 rings contained 11 subunits. The main chain conformations of the DNA-bound and DNA-unbound forms of RAD52 were also quite similar to each other (Figure S2A). The root-mean-square deviation (RMSD) was 0.538 Å for the superimposed RAD52 structures. The largest Cα to Cα distances were observed in the β-hairpin region (amino acid residues 53–65) (Figure S2B). These observations suggest that the DNA binding groove formed by the RAD52 ring is in a “ready state” for ssDNA binding, in which ssDNA can be accommodated without large conformational changes in the ring structure.

B-Form-like Structure of the ssDNA

The ssDNA bound to RAD52 exhibited a periodic structure along the entire length, with a stretched phosphate backbone region appearing every four nucleotides (Figure 2A). The four bases between the stretched regions had similar conformations to each other (Figure 2B). These structural features are consistent with the four-base periodicity observed in the previous hydroxyl radical footprinting studies of ssDNA bound to RAD52 (Singleton et al., 2002, Parsons et al., 2000). Interestingly, the DNA conformation of the four consecutive nucleotides between the stretched phosphate backbone was close to B-form (Figures 2B, 2C, and S3). The bases of the B-form-like segments of the ssDNA were sandwiched between Arg55 and Val63, likely by hydrophobic stacking interactions (Figure 1F). In particular, Arg55 straddled the stretched region of the phosphate backbone and appeared to anchor the ssDNA to the inner DNA binding site. Thus these residues seem to stabilize the B-form-like conformation of the ssDNA through hydrophobic stacking interactions with its bases.

Figure 2

Conformation of the ssDNA Bound to the Inner DNA Binding Site

(A) The RAD52-ssDNA complex, viewed down the central channel of the ring. The four-nucleotide repeats in the 40-nucleotide ssDNA are colored differently.

(B) Superimposed structures of the ssDNA repeating units depicted in (A).

(C) Ideal B-form DNA.

See also Figure S3.

We also found spherical electron densities near the stretched regions of the phosphate backbone (Figure 3A). These electron densities are probably derived from either potassium or calcium ions, because they were present in relatively high concentrations during the crystallization of the complex (25 mM potassium ion and 50 mM calcium ion). Potassium ions were modeled in the final complex structure, based on the fact that the average distance between the putative metal ion and the side chain oxygen atom of the nearest interacting amino acid residue (Glu140) was close to those reported for the coordinated potassium ion (Figure 3B; Zheng et al., 2008, Zheng et al., 2017). The putative metal ion was also in close proximity to the phosphate backbone of the ssDNA and the acidic residues, Glu145 and Asp149, of RAD52, and appeared to bridge the interaction between the two. The observed interaction between the putative metal ions, ssDNA, and RAD52 may explain the elevated ssDNA binding activities of RAD52 previously observed in the presence of divalent ions (Benson et al., 1998, Brouwer et al., 2017).

Figure 3

Putative Metal Ions Located between the Four-Nucleotide Repeats

(A) Locations of the putative metal ions (pink).

(B) Close-up views of the putative metal ion environment (boxed region) in (A). The putative metal ion (potassium ion) is surrounded by E140, E145, D149, and the phosphate group of the ssDNA (indicated with arrows). The table below shows the distances between the potassium ion and the potential coordinating ligands (side chain oxygen atoms from the acidic amino acid residues or oxygen atoms from the phosphate backbone of the ssDNA). The distances are averages over the 10 potassium ion-ligand interaction sites.

Structure of ssDNA Bound to the Outer DNA Binding Site of RAD52

We next determined the crystal structure of a RAD52-ssDNA complex, in which the ssDNA is bound to the outer DNA binding site. The crystal structure revealed a “trapped” ssDNA segment between two RAD52 rings (Figures 4A and 4B). The ssDNA had a markedly different structure from that bound to the inner DNA binding site. The ssDNA was mostly buried between the two RAD52 rings and formed a compact, right-handed helix containing intra-strand base-backbone contacts (Figures 4C and S4A–S4C). Consistent with a previous alanine-scanning mutagenesis study (Kagawa et al., 2008), Lys102 and Lys133 were the primary residues outside of the inner DNA binding site that were involved in the interactions with the ssDNA (Figures S4D–S4F).

Figure 4

“Trapped” Configuration of ssDNA between Two RAD52 Rings

(A) Crystal structure of ssDNA bound to the outer DNA binding site of RAD52. The ssDNA (red) is “trapped” between two RAD52 rings (shown in a light blue surface representation). A schematic diagram of the complex is shown in the upper right.

(B and C) Close-up views of the ssDNA (boxed region) in (A). The helical structure of the ssDNA is stabilized by K102 and K133, which associate with the helical groove of the ssDNA.

See also Figures S4 and S5.

A notable feature of the crystal structure is the close association between the RAD52 rings. The “trapped” ssDNA brings the DNA binding surfaces of the two RAD52 rings in close proximity (Figure 4A, schematic diagram). This results in a trans interaction between the β-hairpin (β1 and β2, Figure S2B) of one RAD52 ring and the outer DNA binding site (loop region between β3 and β4; loop region between β5 and α3, Figure S2B) of the second RAD52 ring. The associated region forms a confined space that includes both the inner and outer DNA binding sites and may provide a platform for DNA annealing. These structural features suggest a role of the outer DNA binding site in facilitating ring-ring associations on ssDNA.

In the crystal, a similar ring-ring association was observed between neighboring, symmetry-related RAD52-ssDNA complexes (Figures S5A and S5B). On each RAD52 ring, four “trapped” ssDNA segments with lengths ranging from 6 to 10 nucleotides were observed (Figure S5C). These interactions extend throughout the crystal, resulting in a RAD52-ssDNA network (Figure S5B). The observed crystal packing interactions may reflect the ability of RAD52 to effectively nucleate on ssDNA, as observed in previous electron micrographic studies of RAD52-ssDNA complexes (Van Dyck et al., 1998, Van Dyck et al., 2001, Kagawa et al., 2001).

ssDNA Aggregation Promoted by the Outer DNA Binding Site

The “trapped” configuration of the ssDNA between two RAD52 rings suggests a role of the outer DNA binding site in promoting the accumulation of multiple RAD52 rings on ssDNA. To examine whether RAD52 aggregation on ssDNA is dependent on the outer DNA binding site, we incubated different concentrations of wild-type RAD52 or the K102A mutant with a circular ssDNA (ФX174; 5,386 bases) and then centrifuged and analyzed the amounts of ssDNA in the upper and lower fractions of the centrifuged sample (Figure 5A). We found that as the concentration of the wild-type RAD52 increased, the amount of ssDNA in the lower fraction increased, indicating that a large network of RAD52-ssDNA complexes formed (Figure 5B). By contrast, the amounts of ssDNA in the upper and lower fractions of the centrifuged sample were nearly the same for the K102A mutant, regardless of the protein concentration (Figure 5C). The K102A mutant retained both ssDNA and dsDNA binding activities (Figure S6), indicating that the absence of aggregation activity was not because the mutant was completely defective in binding to DNA. These results suggest that the outer DNA binding site promotes the co-aggregation of RAD52 and ssDNA. Previously, we demonstrated that alanine substitutions of Lys102 and Lys133 impair the ability of RAD52 to promote DNA annealing (Kagawa et al., 2008). Thus RAD52-mediated DNA annealing likely involves the aggregation of multiple RAD52 rings on ssDNA, which is promoted by the outer DNA binding site.

Figure 5

Accumulation of the Full-Length RAD52 Protein on ssDNA Is Promoted by the Outer DNA Binding Site

(A) A schematic diagram of the assay for examining the ssDNA aggregation promoted by the full-length RAD52.

(B and C) ssDNA aggregation activities of the wild-type and the mutant RAD52 (K102A). A circular, plasmid-sized ssDNA (ФX174, 5,386 bases) was used as the DNA substrate. The ssDNA was fractionated by agarose gel electrophoresis and visualized by ethidium bromide staining. Error bars indicate standard deviations.

See also Figure S6.

ssDNA Binding Affinities of the Two DNA Binding Sites

To further characterize the inner and outer DNA binding sites of RAD52, we investigated the ssDNA binding affinities of each site, using isothermal titration calorimetry (ITC). The K133A mutant, which contains a mutation in the outer DNA binding site, was utilized to determine the ssDNA binding affinity of the inner DNA binding site (Figure 6A). Similarly, the R55A/K152A mutant, which contains two mutations in the inner DNA binding site, was utilized to determine the ssDNA binding affinity of the outer DNA binding site (Figure 6B). The ITC experiments were performed by titrating a 40-mer ssDNA oligonucleotide into solutions of the mutant proteins. Analysis of the data revealed strikingly different thermodynamic profiles between the two mutants (Table S1). Heat absorption was observed for the K133A mutant (endothermic reaction; ΔH > 0), whereas heat release was observed for the R55A/K152A mutant (exothermic reaction; ΔH < 0). These results suggest that the ssDNA binding modes of the two sites are different, a conclusion that is consistent with the structures of the RAD52-ssDNA complexes.

Figure 6

ITC Isotherms for the Binding of ssDNA to RAD52

The upper panel shows the raw titration data, plotted as the heat signal (microcalories per second) versus time (minutes), obtained for 24 injections (12 μL each) of the 40-mer ssDNA oligonucleotide (polydeoxythymine) into a solution containing RAD52. The lower panel shows the ITC data from the upper panel in the form of a titration isotherm, where the integrated heat responses per injection were normalized to the moles of injected ssDNA and plotted versus the total ssDNA to 11-mer RAD52 ratio. The red patches on the schematic diagrams of the RAD52 subunits indicate the mutation sites.

(A) Calorimetric titration of ssDNA (200 μM) into the RAD521−212 K133A mutant (20 μM 11-mer). The continuous curve depicts the best fit of the data to a one-site model.

(B) Calorimetric titration of ssDNA (100 μM) into the RAD521−212 R55A/K152A mutant (10 μM 11-mer). The continuous curve depicts the best fit of the data to a one-site model.

(C) Calorimetric titration of ssDNA (100 μM) into RAD521−212 (10 μM 11-mer). The continuous curve depicts the best fit of the data to a two-site model.

See also Table S1.

The apparent dissociation constant (Kd) of the complex between the R55A/K152A mutant and ssDNA (∼200 nM) was significantly lower than that of the K133A mutant (∼24,000 nM). This indicates that the outer DNA binding site has higher affinity for ssDNA than the inner DNA binding site. To examine whether the two DNA binding sites interact with ssDNA in an independent or cooperative manner, we next performed an ITC experiment using the RAD52 protein without mutations. Like the K133A mutant, the RAD52 protein without mutations exhibited an initial, endothermic phase, which was indicative of ssDNA binding to the inner site (Figure 6C). However, RAD52 displayed a higher affinity toward ssDNA (∼6 nM). This result suggests that the outer DNA binding site facilitated the ssDNA binding at the inner site. Thus in the initial endothermic phase, the ssDNA probably bound to both the inner and outer DNA binding sites, where the sum of the binding heats is predominantly endothermic. Moreover, the outer DNA binding site may be the initial contact site for ssDNA. Our findings demonstrate that the inner and outer binding sites cooperatively interact with ssDNA, which is likely important for the DNA annealing mediated by RAD52.

Discussion

The crystal structures of the two RAD52-ssDNA complexes revealed how the two DNA binding sites may participate in RAD52-mediated DNA annealing. Both structures are compatible with the proposed model for SSA, in which RAD52 binds to resected DSBs (Van Dyck et al., 2001). The ssDNA bound to the inner DNA binding site was B-form-like, with the base-pairing edges of the bases exposed to the solvent. This conformation appears well suited for a homology search and base pair formation with a second ssDNA. By contrast, the ssDNA bound to the outer DNA binding site formed a compact, helical structure, and was buried at the ring-ring interface of RAD52. This interaction may be important for promoting ring-ring associations on ssDNA, thereby increasing the local concentration of RAD52 rings on ssDNA.

Collectively, the two crystal structures and the various analyses presented here help to define a framework for understanding how RAD52 promotes DNA annealing (Figure 7). In our proposed model, RAD52 would initially contact the ssDNA at the outer DNA binding site. Our ITC experiments revealed that the outer binding site has higher affinity for ssDNA than the inner binding site. The “trapped” configuration would promote the accumulation of multiple RAD52 rings on both complementary ssDNAs. This is consistent with the previous report of the accumulation of RAD52 at DSBs, in an ionizing radiation-dependent manner (Wray et al., 2008). At some point, the “trapped” ssDNA would move into the inner DNA binding site, where the conformation becomes B-form. The ssDNA bound to RAD52 would then undergo a homology search with another ssDNA bound to a different RAD52 ring (Rothenberg et al., 2008, Grimme et al., 2010), either by random collision or through directional movement. The B-form conformation induced in the bound ssDNA would facilitate the homology search. When homology is found, the inner DNA binding site would release the base-paired regions of the ssDNA. We previously reported that the outer DNA binding site displays dsDNA binding activity and introduces positive supercoils into dsDNA (Kagawa et al., 2008). Thus after homology has been found, the paired ssDNA region may move to the outer DNA binding sites of the RAD52 rings that promoted the base pair formation, or bind to the outer DNA binding site of a different pair of RAD52 rings, where the duplex DNA would be stabilized.

Figure 7

A Model for RAD52-Mediated DNA Annealing

Close-up views of the RAD52-ssDNA interactions leading to the homology search. Initially, both cDNA strands (colored blue and magenta) are “trapped” by RAD52 rings. The ssDNA moves into the inner DNA binding site of either one of the RAD52 rings that is “trapping” the ssDNA, based on the 5′ to 3′ orientation. The magenta-colored ssDNA is bound to the bottom RAD52 ring, whereas the blue-colored ssDNA is bound to the top RAD52 ring. The RAD52 rings accommodating the ssDNA associate with each other to facilitate the homology search.

See also Figures S7 and S8.

The structure of the RAD52-ssDNA complex and the proposed mechanism for RAD52-mediated DNA annealing share several key features to those of the bacterial DdrB, the only SSA protein for which high-resolution insight into the mechanism of protein-assisted DNA annealing is currently available (Sugiman-Marangos et al., 2016). Like RAD52, DdrB forms a homo-multimeric ring structure, and accommodates ssDNA around the ring structure. The ssDNAs bound to RAD52 and DdrB are oriented such that the bases are accessible to the incoming homologous DNA strand. The two proteins are also similar in that the proposed models for DNA annealing involve the interface between two multimeric ring structures. RAD52 and DdrB also exhibit some differences, such as the number of ssDNA bases bound to each subunit and the ring orientations in the multi-ring complex. These differences may reflect the distinct mechanisms utilized by each protein. Nevertheless, the present findings appear consistent with previous observations of other SSA proteins. Thus the proposed DNA annealing model is probably valid for ancestors of RAD52 in lower organisms and bacteriophages, such as the Sak protein from the Lactobacillus lactis phage ul36 (Ploquin et al., 2008, Scaltriti et al., 2010).

We note that the four-nucleotide, B-form-like segment of the ssDNA bound to the inner DNA binding site is not fully accessible for base pairing with the cDNA strand, owing to steric clashes between the complementary strand and the inner walls of the DNA binding groove (Figures S7A–S7C). We speculate that, during the search for DNA homology, the ssDNA bound to the inner DNA binding site detaches from the bottom of the groove, to make the bases more accessible for base pairing (Figure S7D). This could be facilitated if Arg55, which appears to act as an entry/exit gate for ssDNA, “opens” so that it no longer straddles the phosphate backbone of the ssDNA. The “opening” of the Arg55 gate may result from conformational changes in the β-hairpin, when the DNA binding sites of two RAD52 rings come in close contact.

Importantly, the present structural findings are consistent with the DNA annealing activities displayed by the various point mutants of full-length RAD52 (Figure S8). Alanine substitutions of amino acid residues at the inner and outer DNA binding sites (Lys102, Lys133, Lys152, Arg153, and Arg156), which are important for the stable association with ssDNA, significantly impaired the ability of RAD52 to efficiently promote the annealing of complementary 50-mer oligonucleotides (Figures S8B and S8D). By contrast, the alanine substitution of Arg55 resulted in a milder defect in DNA annealing (Figure S8C), when compared with those of Lys152, Arg153, and Arg156, implying that Arg55 may have a distinct role in DNA annealing. This observation is consistent with the unique role of Arg55 revealed by the present structural studies. The amino acid residues that constitute the two DNA binding sites are well conserved among RAD52 orthologs (Figure S8E), suggesting that the mechanisms to promote DNA annealing are also conserved.

Intriguingly, the mechanisms to induce the B-form-like conformation of ssDNA do not appear to be specific to RAD52. The B-form-like conformation of the ssDNA was stabilized by the electrostatic interactions between the phosphate backbone and the basic residues at the bottom of the DNA binding groove, as well as by the hydrophobic stacking interactions between the ssDNA bases and the Arg55 and Val63 residues of the β-hairpin. The latter hydrophobic interaction resulted in a stretched phosphate backbone conformation (with respect to B-form) and the sandwiching of four consecutive bases. The stretching of the phosphate backbone and the sandwiching of bases are also observed in the ssDNA structure bound to the bacterial RecA recombinase (Chen et al., 2008), which lacks apparent sequence homology to RAD52 and exhibits a completely different oligomerization structure. Thus the stretching and sandwiching may be common interaction modes utilized by proteins that expose the base pairing edges of ssDNA to the solvent, to facilitate base pair formation with a cDNA strand.

In summary, the crystal structures of the RAD52-ssDNA complexes revealed in this study are important for understanding the molecular mechanisms of HDR systems, especially those that do not involve RAD51. Several types of cancers have defects in the genes that function in RAD51-dependent HDR pathways, such as BRCA1, BRCA2, and PALB2. RAD52 is synthetically lethal with these genes (Feng et al., 2011, Lok et al., 2013), suggesting that RAD52 is a potential target for cancer therapeutics (O’Connor, 2015). The present complex structures may be useful in designing inhibitors that target RAD52 for cancer treatment.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.