ATP hydrolysis-driven structural transitions within the Saccharomyces cerevisiae Rad51 and Dmc1 nucleoprotein filaments

Abstract

Homologous recombination (HR) is essential for the maintenance of genome stability and for generating genetic diversity during meiosis. The eukaryotic protein Rad51 is member of the Rad51/RecA family of DNA recombinases and is responsible for guiding the DNA pairing reactions that take place in HR during mitosis. Dmc1 is a meiosis-specific paralog of Rad51 and is responsible for the DNA pairing reactions that take place in HR during meiosis. Rad51 and Dmc1 are both ATP-dependent DNA-binding proteins and both form extended helical filaments on ssDNA, which are key intermediates in HR. The stability of these nucleoprotein filaments is highly regulated and is also tightly coupled to nucleotide binding and hydrolysis. ATP binding promotes filament assembly, whereas the hydrolysis of ATP to ADP reduces filament stability to promote filament disassembly. Here, we present cryo-EM structures of the Saccharomyces cerevisiae recombinases Rad51 and Dmc1 in the ADP-bound states and provide a detailed structural comparison to the ATP-bound filaments. Our findings yield insights into the structural transitions that take place during the hydrolysis of ATP to ADP and suggest a new model for how these structural changes may be linked to nucleoprotein filament disassembly.

Homologous recombination (HR) is essential for the maintenance of genome integrity and is a major driving force in genome evolution (1, 2). HR plays important roles in dsDNA break repair (3, 4), the rescue of stalled or collapsed replication forks (5, 6), chromosomal rearrangements (7, 8, 9), horizontal gene transfer (10), and meiosis (11, 12, 13). The protein participants, nucleoprotein structures, and general HR reaction mechanisms are broadly conserved among all kingdoms of life (14, 15, 16, 17).

The DNA pairing reactions that take place in HR are promoted by the Rad51/RecA family of DNA recombinases, which are ATP-dependent proteins that form extended helical filaments on single stranded DNA, that are often referred to as presynaptic complexes (14, 15, 16, 17). During HR, an ssDNA serves as a platform for the assembly of the presynaptic complex which then pairs the bound ssDNA with the complementary strand from a homologous dsDNA from elsewhere in the genome, resulting in displacement of the noncomplementary strand (3, 14, 15, 16). These displacement loop intermediates can then be channeled through several mechanistically distinct pathways to restore the continuity of the originally damaged DNA (3, 14, 15, 16).

Most eukaryotes have two Rad51/RecA recombinases: Rad51, which is expressed constitutively, and Dmc1, which is only expressed in meiosis (11, 13, 18, 19, 20). Rad51 and Dmc1 are thought to have arisen from an ancient gene duplication event, and the emergence of Dmc1 as a separate lineage may have coincided with the emergence of meiosis and sexual reproduction (21, 22, 23, 24). These two proteins remain closely related; for instance, Saccharomyces cerevisiae Rad51 and Dmc1 share 45% sequence identity and 56% sequence similarity and both proteins perform the same basic biochemical function, namely the pairing of homologous DNA sequences (11, 13). Rad51 and Dmc1 were identified over 25 years ago (18, 25), yet we still do not fully understand of why most eukaryotes require two recombinases (11, 13). Prevailing hypotheses are that (i) each recombinase must interact with a specific subset of mitotic- or meiotic-specific accessory factors, (ii) there are biochemical differences between the recombinases making them uniquely suited to mitotic or meiotic HR (11, 13). There are examples of Rad51- and Dmc1-specific interacting factors (11, 13), and Rad51 and Dmc1 respond differently when presented with mismatch-containing HR intermediates (26, 27, 28), suggesting that aspects of both hypotheses may be correct.

It has long been known that DNA within the active ATP-bound presynaptic complex is stretched by approximately 50% relative to B-form DNA and crystal structures of RecA–ssDNA presynaptic and RecA–dsDNA postsynaptic complexes reveal that the DNA is organized into near B-form base triplets separated by ∼8 Å between adjacent triplets (29). This base triplet structural organization likely underpins homology recognition and the ability of the Rad51/RecA family of recombinases to promote DNA strand invasion in 3-nt steps (29, 30, 31, 32, 33). This structural organization of the DNA now recognized to be a conserved feature of the nucleoprotein filaments made by Rad51/RecA family members.

Presynaptic complex assembly and disassembly are linked to the ATP hydrolysis cycle (14, 34). Presynaptic complex assembly begins with an initial nucleation event with the recombinases in the high affinity ATP-bound state, followed by elongation as new ATP-bound protomers are added to the ends of the growing filaments (35, 36, 37, 38). Disassembly also takes place from the filament ends as the recombinase protomers hydrolyze ATP and are converted to the weakly bound ADP-state (39). Based upon early studies with bacterial RecA, it is thought that the transition between the ATP- and ADP-bound states is also linked to changes in filament structure and function. Filaments in the ATP-bound state are more elongated with a helical pitch ranging between approximately 90 Å and 130 Å, while filaments in the ADP-bound state are thought to be more compressed with a helical pitch ranging from approximately 60 Å to 80 Å (29, 40, 41, 42, 43). Given the link between the nucleotide-bound state and the structure and function of the presynaptic complex, these transitions represent a potentially important regulatory control mechanism.

We have previously reported the cryo-EM structures for the ATP-bound states of S. cerevisiae Rad51 and Dmc1, which correspond to the highly elongated states expected for biochemically active nucleoprotein filaments (44). Here, we present cryo-EM structures of S. cerevisiae Rad51 and Dmc1 nucleoprotein filaments in their ADP-bound states. Comparison of the ATP-bound and ADP-bound states show that the nucleoprotein filaments undergo a large structural transition resulting in weakened contacts between proteins within the nucleoprotein filaments and disordered electron density for the ssDNA. However, instead of undergoing compression, as expected based on studies of bacterial RecA (40), we find that the ADP-bound states of both Rad51 and Dmc1 were even more highly elongated, consistent with studies of human RAD51 (45, 46), suggesting that the observed structural transitions may be broadly conserved among the eukaryotic members of the Rad51/Dmc1 family members but are distinct from bacterial RecA. We suggest a model in which the extension of the Rad51 and Dmc1 nucleoprotein filaments upon ATP hydrolysis would require increased tension within the bound ssDNA to maintain proper register with the protein filament. This effect, together with weaker protein–protein contacts, may help to explain how the ATP hydrolysis cycle is linked to Rad51 and Dmc1 nucleoprotein filament disassembly.

Results

Cryo-EM structure of Rad51 and Dmc1 filaments in the presence of ADP

To solve the cryo-EM structures of S. cerevisiae Rad51 and Dmc1 in the ADP-bound states, we mixed the purified proteins with 96-nucleotide ssDNA oligonucleotide in the presence of 2 mM ADP. Under these conditions, Rad51 yielded filaments with a resolution of 3.37 Å (Fig. 1, A and B, Fig. S1, Fig S3, A and B, and Table S1). Interestingly, in contrast to our observations with Rad51, the Dmc1 filaments were interdigitated (Fig. S2), similar to what was recently observed for human RAD51 filaments in the presence of ADP (46). However, the only potential point of contact between the filaments was between residues E62 and R301, but the density of R301 is poor, suggesting that this does not represent a true interaction (Fig. S2F). At this time, we do not have any reason to believe that these interdigitated structures reflect a biologically meaningful intermediate. Therefore, the interdigitated filaments were computationally deconvolved into single helical filaments (Fig. S2), yielding well-defined structures with an overall resolution of 2.7 Å (Fig. 1, C and D, Fig. S2, Fig. S3, C and D, and Table S1).

Figure 1.

Overview of Rad51 and Dmc1 filament structures in the ADP-bound states. A, our previously published cryo-EM structure of S. cerevisiae Rad51 nucleoprotein filament in the ATP-bound state (PDB ID: 9D46) (44). A subsection of the nucleoprotein comprised of six Rad51 monomers is shown, the different protein monomers are highlighted in alternating cyan and light green, and the bound ssDNA is shown in black. B, cryo-EM structure of S. cerevisiae Rad51 bound in the ADP-bound state (PDB ID: 9NJK). A subsection of the nucleoprotein comprised of six Rad51 monomers is shown, and the different protein monomers are highlighted in alternating blue and green; the ssDNA is disordered within the structure. C, our previously published cryo-EM structure of S. cerevisiae Dmc1 nucleoprotein filament in the ATP-bound state (PDB ID: 9D4N) (44). A subsection of the nucleoprotein comprised of six Dmc1 monomers is shown, the different protein monomers are highlighted in alternating magenta and orange, and the bound ssDNA is shown in black. D, cryo-EM structure of S. cerevisiae Dmc1 bound in the ADP-bound state (PDB ID: 9NJR). A subsection of the nucleoprotein comprised of six Dmc1 monomers is shown, and the different protein monomers are highlighted in alternating purple and dark orange; the ssDNA is disordered within the structure.

We have previously reported cryo-EM structures for S. cerevisiae Rad51 and Dmc1 in the ATP-bound states (PBD IDs: 9D46 and 9D4N) (44), and here we use these structures as references for comparison to the ADP-bound states. There were extensive changes in the overall structures of the nucleoprotein filaments assembled in the ADP-bound states compared to the ATP-bound states (Figs. 1 & S3). Most notably, there was a dramatic increase in the overall contour length of the protein filaments in the ADP-bound states. In the case of Rad51, the ATP-bound filaments exhibited a helical pitch of 99 Å with 6.4 protein monomers per helical turn and 19.2 nucleotides per turn, each protomer was rotated 56.4° relative to its nearest neighbors, and the filament had a maximum width of 87.6 Å (Fig. 1A). In the Rad51 ADP-bound structure, the helical pitch increased to 132 Å, corresponding to an overall increase in the contour length of 33.3%, there were 6.5 monomers per turn and each protomer was rotated 51.6° relative to its neighbors, and the filament had a maximum width of 95.2 Å (Fig. 1B). In the case of Dmc1, the ATP-bound filaments exhibited a helical pitch of 98.4 Å with 6.35 protein monomers per helical turn and 19.05 nucleotides per turn, each protomer was rotated 55.5° relative to its neighbors, and the filament had a width of 87.6 Å (Fig. 1C). In the Dmc1 ADP-bound structure, the helical pitch increased to 133 Å, corresponding to an overall increase in the contour length of 35.2%, there were 6.6 monomers per turn and each protomer was rotated 47.4° relative to its neighbors, and the filament had a maximum width of 97.5 Å (Fig. 1D). The extended pitch of these filaments closely resembled the 130 Å pitch originally observed for the crystal structure of S. cerevisiae Rad51 (41) as well as the extended pitches observed for more recent cryo-EM structures of human RAD51 in the ADP-bound state (45, 46).

The extended helical pitches observed for the ADP-bound states of Rad51 and Dmc1 coincided with the disordered ssDNA density, consistent with a weakening of the protein–DNA contacts, as has been previously reported for human RAD51 (45, 46), strongly suggesting that this structural change is a conserved property of the eukaryotic members of the Rad51/RecA family (Fig. 1, B and D). It should be noted that formation of the Rad51 and Dmc1 filaments in the ADP-bound state required inclusion of the ssDNA substrate in the sample mixture, so although electron density for the ssDNA is not well resolved in the cryo-EM data, it seems likely that the ssDNA is present within the protein filaments.

Organization of the nucleotide-binding pocket

Rad51/RecA family members have bipartite nucleotide-binding pockets comprised of highly conserved amino acid side chains from two adjacent protomers (29, 41). The Walker A and Walker B nucleotide-binding motifs are contributed by a single protomer and reside on the 3′ facing side of each protomer (Fig. 2). Our cryo-EM structures of Rad51 and Dmc1 reveal well-defined density for the ATP and ADP molecules within the nucleotide-binding clefts in the ATP-bound and ADP-bound states, respectively (Fig. 2, A and B, D, E).

Figure 2.

Comparison of the nucleotide-binding pockets in the ATP and ADP-bound states. A, nucleotide-binding pocket of S. cerevisiae Rad51 in the ATP-bound state. The Rad51 protomer in which the Walker A and Walker B motifs interact with the bound nucleotide is shown in light green, and the second protomer is shown in cyan. The locations of important side chain residues and the two metal ions are indicated. B, nucleotide-binding pocket of S. cerevisiae Rad51 in the ADP-bound state. The Rad51 protomer in which the Walker A and Walker B motifs interact with the bound nucleotide is shown in green, and the second protomer is shown in blue. The locations of important side chain residues and the two metal ions are indicated. C, overlay of Rad51 in the ATP- and ADP-bound states. D, nucleotide-binding pocket of S. cerevisiae Dmc1 in the ATP-bound state. The Dmc1 protomer in which the Walker A and Walker B motifs interact with the bound nucleotide is shown in magenta, and the second protomer is shown in orange. The locations of important side chain residues and the two metal ions are indicated; here, the metal ions are indicated as Me2+ because we cannot distinguish between Mg²⁺ and Ca²⁺, both of which were present in samples with Dmc1. E, nucleotide-binding pocket of S. cerevisiae Dmc1 in the ADP-bound state. The Dmc1 protomer in which the Walker A and Walker B motifs interact with the bound nucleotide is shown in purple, and the second protomer is shown in dark orange. The locations of important side chain residues and the two metal ions are indicated. F, overlay of Dmc1 in the ATP- and ADP-bound states.

There is a highly conserved histidine residue on the 5′ face of each recombinase protomer, H352 in Rad51 and H289 in Dmc1, both located with within α helix 14 (Fig. S4), which interact in trans with the nucleotide-binding pocket of the neighboring protomer and helps to coordinate the gamma phosphate of the bound ATP molecule (Fig. 2, A and D). It has previously been suggested that these histidine residues act as sensors of ATP binding and help to coordinate communicate between adjacent recombinase protomers within the nucleoprotein filaments (47). Previous studies have shown, mutation of H352 within S. cerevisiae Rad51 to alanine, tyrosine, phenylalanine, lysine, or methionine, severely compromises both ATP hydrolysis activity and nucleoprotein filament assembly at physiological pH (47). Consistent with this model, we find that Rad51 H352 and Dmc1 H289 move away from the neighboring protomers when the nucleoprotein filaments are in the ADP-bound states, corresponding to a shift of 4.8 Å in the case of Rad51 and 5.3 Å for Dmc1 measured from the epsilon carbon if the histidine side chains (Fig. 2, C and F). Together with the previous study (47), this work further highlights the importance of H352 from Rad51 and the corresponding H289 from Dmc1 as important residues for sensing the nucleotide-bound state of the nucleoprotein filaments and because these interactions occur in trans they may contribute to allosteric communication within the nucleoprotein filaments.

For both Rad51 and Dmc1, we observe density for two metal ions within the nucleotide-binding cleft in the ATP-bound states (Fig. 2, A and D). In the ADP-bound states, however, there is a loss of the second metal ion (Fig. 2, B and E), a phenomenon that has also been reported for human RAD51 (45). For Rad51, the first metal ion is coordinated in cis by the side chains of residues S192 and D280, which are located between the beta and gamma phosphate of the bound ATP molecule as well as residue E221 which is near the gamma phosphate group of the ATP (Fig. 2A). In the case of Dmc1, the first metal ion is coordinated in cis by the sidechains of residues T128, D216, and E157, which are all positioned similarly to the equivalent residues from Rad51 (Fig. 2D). These contacts with the first metal ion do not undergo any large changes in the ATP- versus ADP-bound states, for either Rad51 (Fig. 2, A–C) or Dmc1 (Fig. 2, D–F). The second metal ion is coordinated in trans by a highly conserved aspartic acid residue, D374 from Rad51 and D311 from Dmc1, from the 5′ facing side of the adjacent protomer (Fig. 2, A and D). The second metal ion also interacts in with the amide group from the peptide backbone of residues S354 and T355 from Rad51 and residue S291 from Dmc1 (Fig. 2, A and D). Comparison of the ATP-bound and ADP-bound states reveals that S. cerevisiae Rad51 D374 and Dmc1 D311 move away from the neighboring protomer in the ADP-bound states, corresponding to a distance (measured from the alpha carbon) of 2.7 Å in the case of Rad51 and 3.7 Å for Dmc1 (Fig. 2, A–F). Importantly, this trans interaction with the second metal ion has also been reported for human RAD51 (D316) and archaeal RadA (D302), providing further evidence for the conservation of this mechanism (45, 48, 49).

Dmc1 requires the presence of Ca²⁺ for optimal in vitro activity (13, 50, 51, 52). Although we see clear density for two metal ions in the nucleotide-binding pocket of S. cerevisiae Rad51 and Dmc1, we cannot directly distinguish between Mg²⁺ and Ca²⁺ based on the electron density alone. In the case of Rad51, Mg²⁺ was the only divalent metal ion cofactor included in the filament assembly reactions, so it is reasonable to conclude that there are 2 Mg²⁺ ions bound within the Rad51 active site. However, for Dmc1, the filament assembly reactions required both Mg²⁺ and Ca²⁺, as has been well established in the literature (13, 50, 51, 52), so the bound metal ions could be Mg²⁺, Ca²⁺, or a mixture of both metal ions. Interestingly, superimposition of the Rad51 and Dmc1 structures from nucleoprotein filaments in the ATP-bound states reveals that the second divalent metal ion appears to be shifted by 2.1 Å in Rad51 compared to Dmc1 (Fig. S5, A–C). Comparison of our S. cerevisiae Rad51 (determined with Mg²⁺ only) and Dmc1 (determined with Mg²⁺ and Ca²⁺) structures to a recent structure for human RAD51 in the ATP-bound state on ssDNA in the presence of only Ca²⁺ revealed that the second metal ion in the human RAD51 (PBD ID: 8BQ2; Fig. S5D) (45) is shifted by 2.2 Å compared to the second metal ion in yeast Rad51 (Fig. S5E), but matches well with the second metal ion found in yeast Dmc1 (Fig. S5F). One possibility is that the differential positioning of the second metal ion in human RAD51 and S. cerevisiae Dmc1 reflects differential positioning due to the identity of the bound divalent metal ion, which might imply that Ca²⁺ is the second metal observed in our structure of S. cerevisiae Dmc1 even though Mg²⁺ was present in 13-fold molar excess over Ca²⁺ (20 mM MgCl₂ versus 1.5 mM CaCl₂).

The FxxA polymerization motif remains intact in the ADP-bound state

Eukaryotic and archaeal members of the Rad51/RecA family of proteins harbor a conserved FxxA polymerization motif that is essential for nucleoprotein filament formation (25, 53). This motif is found within the interdomain linker between the N-terminal five-helix bundle and core domain (Fig. 3, A and B, and Fig. S4) and is comprised of residues 144-FVTA-147 and 80-FIPA-83 for S. cerevisiae Rad51 and Dmc1, respectively (Fig. 3, C–H). For Rad51, phenylalanine residue F144 binds within a hydrophobic pocket of the adjacent protomer formed by residues L216, I218, A248, A250, L261, A264, and M268 from the adjacent protomer, and A147 fits into a hydrophobic pocket formed by F224, L244, and V247. Residue F224 is further stabilized by stacking interactions with Y249 and P226 in cis, and F150 in trans (Fig. 3C). For Dmc1, phenylalanine residue F80 binds within a hydrophobic pocket of the adjacent protomer formed by residues A152, I154, S184, A186, L197, and L201 from the adjacent protomer, and residue A83 fits into the hydrophobic pocket formed by F160, L180, and V183. Residue F160 is further stabilized by stacking interactions with Y185 and P162 in cis and Q86 in trans (Fig. 3F). Interestingly, the FxxA polymerization motifs remain fully intact upon transition to the ADP-bound state for both Rad51 and Dmc1, indicating that the disruption of this protein–protein interfacial contact does not directly contribute to nucleoprotein filament destabilization upon the hydrolysis of ATP to ADP (Fig. 3, E and H).

Figure 3.

The FxxA polymerization interface remains largely unaltered in the ATP and ADP-bound states. A, overlay of two adjacent Rad51 protomers in the ATP- and ADP-bound states, as indicated, highlighting the location of the FxxA polymerization motif in one protomer and its binding cleft on the adjacent protomer. B, overlay of two adjacent Dmc1 protomers in the ATP- and ADP-bound states, as indicated, highlighting the location of the FxxA polymerization motif in one protomer and its binding cleft on the adjacent protomer. C, close-up view of the Rad51 FxxA polymerization motif interaction in the ATP-bound state. D, close-up view of the Rad51 FxxA polymerization motif interaction in the ADP-bound state. E, overlay of the Rad51 FxxA polymerization motif interaction in the ATP- and ADP-bound states. F, close-up view of the Dmc1 FxxA polymerization motif interaction in the ATP-bound state. G, close-up view of the Dmc1 FxxA polymerization motif interaction in the ADP-bound state. H, overlay of the Dmc1 polymerization motif interaction in the ATP- and ADP-bound states.

Disruption of a Dmc1-specific interprotomer contact

In addition to the FxxA motif, Rad51 also harbors and highly conserved pair of stacked aromatic residues that is unique to the Rad51 lineage of the Rad51/RecA but absent in the Dmc1 lineage (44). In S. cerevisiae Rad51, these correspond to amino acids residues Y112 and Y253, whereas the equivalent residues in S. cerevisiae Dmc1 are S48 and L189, which do not appear to interact with one another (Fig. 3, C and F). We have previously reported deep mutagenesis experiments which demonstrated that nonaromatic residues are largely not tolerated at these positions with S. cerevisiae Rad51, whereas many different residue pairs retain biological function in S. cerevisiae Dmc1 (44).

In contrast to the stacked aromatic residues at the Rad51 protein–protein interface, Dmc1 residues S48 and L189 do not appear to interact with one another. Instead, residue S48 interacts with R248 through an intramolecular interaction, and residue L189 is stabilized by residues L47 and V78 via hydrophobic interaction (Fig. 3F). The equivalent arginine residue in Rad51 is R312 which does not contact Rad51 Y112 (Fig. 3C). The stacked aromatic residues of Rad51 are maintained upon transition to the ADP-bound state (Fig. 3D), while Dmc1 S48 maintains the interaction with R248 and therefore S48 is relieved from L189 in the ADP-bound state (Fig. 3G). Thus, this interprotomer contact appears to be different in Rad51 and Dmc1.

Disruption of DNA-binding loop L1 and α helix 13 interprotomer contacts

Comparison of the ATP- and ADP-bound structures for both Rad51 and Dmc1 reveal that transition to the ADP-bound state coincides with the rotation of a long α-helix (helix 13; Fig. S4) that is next to DNA-binding loop L1 and is involved in trans interactions with the L1 DNA-binding domain of the adjacent protomer (Fig. 4, A and B). Rotation of this helix coincides with the loss of DNA-binding loop L1 ssDNA contacts (Fig. 4, D and G and Fig. S6) and also disrupts protein–protein contacts between adjacent protomers (Fig. 4, C–H). In the case of Rad51, residues R308, R312, and D315 interact in trans with residues T288, D289, and R251 of the adjacent protomer in the ATP-bound state (Fig. 4C). In addition, residue K216 from one protomer contacts residue E295 in the L1 DNA-binding loop of the adjacent protomer (Fig. 4C). All of these contacts are disrupted in the ADP-bound state (Fig. 4, D and E). These interprotomer contacts are somewhat different in the case of Dmc1. For Dmc1, residues F244, R248, and E251 interact with V224 and R187 from the adjacent protomer in the ATP-bound state (Fig. 4F). Lastly, similar to Rad51, Dmc1 residue R52 from one protomer contacts residue E231 in the L1 DNA-binding loop of the adjacent protomer (Fig. 4C). As with Rad51, all of these contacts are disrupted in the ADP-bound state of Dmc1 (Fig. 4, G and H).

Figure 4.

Loss of trans L1-L1 contacts between adjacent protomers in the ADP-bound state. A, overlay of two adjacent Rad51 protomers in the ATP- and ADP-bound states, as indicated, highlighting the location of the L1 DNA-binding loop connect to helix 13. B, overlay of two adjacent Dmc1 protomers in the ATP- and ADP-bound states, as indicated, highlighting the location of the L1 DNA-binding loop connect to helix 13. C, close-up view of the Rad51 trans interactions between L1 and helix 13 interaction in the ATP-bound state. D, close-up view of the Rad51 trans interactions between L1 and helix 13 interaction in the ADP-bound state. E, Overlay of the Rad51 DNA-binding loop L1 and helix 13 in the ATP- and ADP-bound states. F, close-up view of the Dmc1 trans interactions between L1 and helix 13 interaction in the ATP-bound state. G, close-up view of the Dmc1 trans interactions between L1 and helix 13 interaction in the ADP-bound state. H, overlay of the Dmc1 DNA-binding loop L1 and helix 13 in the ATP- and ADP-bound states.

Disruption of DNA-binding loop L2 structure and contacts

For both Rad51 and Dmc1 in the ATP-bound state, the L2 DNA-binding loop makes crucial contacts with the ssDNA and also makes protein–protein contacts between adjacent protomers within the nucleoprotein filaments (Fig. 5). For Rad51 in the ATP-bound state, residues I345 and V328 form intramolecular hydrophobic interactions that help to organize the overall shape of the L2 DNA-binding loop (Fig. 5C). The peptide backbone of residue I345 also makes stabilizing intramolecular contacts with residue Q326 (Fig. 5C) and residue P344 provides a structural kink point near the end of L2 allowing it to transit back toward the core of the protein (Fig. 5C). Residues K343 and N348 both make contacts the ssDNA backbone and residue N348 also interacts in trans with residue K342 from the adjacent protomer (Fig. 5C). Lastly, residue H352 interacts in trans with Q326, which is itself stabilized in cis with the side chains of E221 and R287, and the backbone of residue I345 (Fig. 5C). These interactions are all weakened in the ADP-bound state as evidenced by the disordered electron density for both the ssDNA and many of the L2 amino acid residues (Fig. 5, D and E), which is consistent with prior findings for human RAD51 (45).

Figure 5.

Loss of L2 ssDNA-protomer and interprotomer contacts in the ADP-bound state. A, overlay of two adjacent Rad51 protomers in the ATP- and ADP-bound states, as indicated, highlighting the location of the L2 DNA-binding loop, α helix 14, and β strands 6 and 7. B, overlay of two adjacent Dmc1 protomers in the ATP- and ADP-bound states, as indicated, highlighting the location of the L2 DNA-binding loop, α helix 14, and β strands 6 and 7. C, close-up view of the Rad51 L2 contacts with the bound ssDNA substrate and interprotomer interactions in the ATP-bound state. D, close-up view showing the loss of Rad51 L2 contacts with the bound ssDNA substrate and interprotomer interactions in the ADP-bound state. E, overlay of the Rad51 region encompassing the L2 DNA-binding loop, α helix 14, and β strands 6 and 7 in the ATP- and ADP-bound states. F, close-up view of the Dmc1 L2 contacts with the bound ssDNA substrate and interprotomer interactions in the ATP-bound state. G, close-up view showing the loss of Dmc1 L2 contacts with the bound ssDNA substrate and interprotomer interactions in the ADP-bound state. H, overlay of the Dmc1 region encompassing the L2 DNA-binding loop, α helix 14, and β strands 6 and 7 in the ATP- and ADP-bound states.

In the case of Dmc1, in the ATP-bound state, residues I282 and Q264 form intramolecular contacts between their side chains as well as contacts between the peptide backbone atoms which help to organize the overall structure of the L2 DNA-binding loop (Fig. 5F). The proline side chain of P281 provides a structural kink that allows L2 to transit back toward the core of the protein (Fig. 5F). Residues H285, Q264, and K280 all make contacts with the ssDNA phosphate backbone (Fig. 5F). Residues H289 and Q262 interact in trans between adjacent protomers, and the position of Q262 is organized by cis interactions with the side chains of residues E157 and R223, and the backbone of residue I282. Similar to Rad51, these interactions are all weakened in the ADP-bound state of Dmc1 as evidenced by the disordered density for both the ssDNA and many of the L2 amino acid residues (Fig. 5, G and H).

Discussion

Our work shows that S. cerevisiae Rad51 and Dmc1 adopt a highly extended configuration when in the ADP-bound state. This large structural transition coincides with a weakening of DNA contacts, as evidenced as disordered electron density corresponding to the ssDNA and a weakening of contacts between adjacent protein subunits, supporting the hypothesis that ADP-bound state is an on path intermediate toward nucleoprotein filament disruption. These findings give insights into how the ATP hydrolysis cycle might be coupled to nucleoprotein filament disassembly and reveal interesting structural differences between the eukaryotic Rad51 and Dmc1 proteins compared to their bacterial homolog RecA, which may suggest that these proteins have distinct mechanistic approaches to regulating dissociation from ssDNA.

Comparison between S. cerevisiae Rad51 and Dmc1

S. cerevisiae Rad51 and Dmc1 both form similar nucleoprotein filament structures in the presence of ADP and in each case the filaments are highly elongated compared to those formed in the ATP-bound state. These findings are in good agreement with previous studies which have shown that nucleoprotein filaments prepared with human RAD51 in the presence of ADP are also in a highly elongated state compared to those formed with ATP (45, 46). Therefore, it is likely that this elongated ADP-bound state reflects a broadly conserved structural state for the eukaryotic Rad51 and Dmc1 recombinases.

It is notable that our structures of S. cerevisiae Rad51 and Dmc1 in the ADP-bound states bear a strong resemblance to the S. cerevisiae Rad51 crystal structure of S. cerevisiae Rad51, which represented a breakthrough in the field as it was the first reported high-resolution structure for Rad51 in a filament state (41). Interestingly, the ssDNA and the DNA-binding loops were not visible within the Rad51 crystal structure, the filaments were prepared with ATPγS but no nucleotide cofactor was visible within the active site, perhaps because the ATPγS had undergone hydrolysis during crystallization, and instead a sulfate ion was visible within the active site (41). Lastly, the helical pitch of 130 Å was observed in the S. cerevisiae Rad51 crystal structure (41), which closely corresponds to the helical pitch values of 132 Å and 133 Å that we report here for the ADP-bound states of S. cerevisiae Rad51 and Dmc1, respectively.

Comparison of Rad51 and Dmc1 to bacterial RecA

Rad51/RecA family members are ATP-dependent DNA-binding proteins and the transition from the ATP- and ADP-bound states is linked to nucleoprotein filament disassembly (14, 34). Thus, the ADP-bound forms of these nucleoprotein filaments should represent intermediates on the pathway toward protein dissociation. Based on structural studies of bacterial RecA, it had been thought that the active ATP-bound state was more elongated with a helical pitch ranging between approximately 90 Å and 130 Å, while inactive filaments in the ADP-bound state were thought to be more compressed with a helical pitch ranging from approximately 60 Å to 80 Å (29, 40, 41, 42, 43). Interestingly, based on the known high-resolution structures of eukaryotic recombinases in the ADP-bound states, including those presented here for S. cerevisiae Rad51 and Dmc1, and previous structures of human RAD51 (45, 46), the eukaryotic proteins may be behaving differently from bacterial RecA. Instead of transitioning to a compressed state, it appears as though Rad51 and Dmc1 undergo a structural conversion to a more elongated state which coincides with a loss of DNA contacts and disruption of contacts between the protomer–protomer interfaces. Indeed, we see no obvious evidence for the existence of compressed states for the ADP-bound nucleoprotein filaments for either Rad51 or Dmc1 under the conditions used for our sample preparation (Figs. S1 and S2). These observations raise the possibility that Rad51 and Dmc1 might follow a different pathway toward DNA dissociation compared to the bacterial RecA protein. In this regard, it is interesting to note that although the core ATP-binding domains of eukaryotic Rad51 and Dmc1 and prokaryotic RecA are all strongly conserved, the protein–protein interfaces between protomers within the filaments are markedly different (40). RecA has a C-terminal domain that contributes to the protein–protein interfaces and allosteric communication between adjacent proteins within a nucleoprotein filament, but this C-terminal domain is absent from the eukaryotic recombinases, whereas Rad51 and Dmc1 have an N-terminal domain that is absent from RecA. These differences in domain structure may be responsible for conferring distinct properties to the nucleoprotein filaments, which in this case manifest as structurally distinctive intermediates on the pathway toward nucleoprotein filament dissociation.

Model for Rad51 and Dmc1 nucleoprotein filament disassembly

Rad51 and Dmc1 filaments in the high-affinity ATP-bound state and bound to ssDNA that is held in a highly extended configuration. However, the extension is not isotropic, instead the base triplets are maintained in a near B-form DNA configuration, and the distance between base triplets is highly extended which gives rise to the overall extension of the bound ssDNA, as originally reported for bacterial RecA (29). In the case of S. cerevisiae Rad51, this corresponds to an axial rise of 7.2 to 7.3 Å between the last base of one triplet and the first base of the next triplet, and for S. cerevisiae Dmc1 these values are 7.0 to 7.2 Å (Fig. 6A). Our work shows that the Rad51 and Dmc1 nucleoprotein filaments are even longer in the ADP-bound state (Fig. 6B). The ssDNA molecules within the ATP-bound states of the Rad51 and Dmc1 nucleoprotein filaments are approximately 1.6 times the length of an equivalent B-form dsDNA molecule. With respect to the ADP-bound states, if one were to consider a hypothetical scenario where the base triplets had to remain in register with the Rad51 or Dmc1 monomers in the same way as is observed in the ATP-bound structures, and then the phosphodiester bonds between the base triplets would also have to increase up to ∼9.0 Å, which approaches the maximum extent to which the phosphodiester bond can be stretched and the bound ssDNA would have to be stretched to approximately 1.9 times the length of an equivalent B-form dsDNA (Fig. 6B). If one assumes that the isotropic stretching forces experienced by an ssDNA molecule under tension by optical tweezers can be used to roughly approximate the forces experienced by the extended ssDNA that is bound by Rad51 or Dmc1, then force exerted on the phosphodiester backbone between base triplets would be on the order of 20 to 30 pN (54). Making the same assumptions, the force needed to extend the phosphodiester backbone to match the extended state of the ADP-bound forms of the Rad51 or Dmc1 filaments would be on the order of ∼100 pN (54). Although the extension of the ssDNA within a Rad51 or Dmc1 filament is certainly not isotropic, these simple comparisons serve to illustrate that the increased extension on the ssDNA that would be needed to match to extension of the ADP-bound state would necessarily involve exerting increasing forces on the ssDNA which may in turn contribute to destabilizing the binding interactions.

Figure 6.

Model for Rad51 and Dmc1 nucleoprotein filament disassembly. A, Rad51 and Dmc1 in the ATP-bound states give exhibit extended nucleoprotein filaments and the bound ssDNA is organized into base triplets with an axial rise of ∼7.2 Å between the last base of one triplet and the first base of the next triplet. B, conversion to the ADP-bound state results in nucleoprotein filament length, which would require even greater extension of the phosphate backbone in order the base triples to maintain correct register with the protein protomers. C, the nucleoprotein filaments begin in the ATP-bound state and conversion of end-bound protomers leads to loss of contacts with the ssDNA and disruption of the protein–protein interfaces leading to protein dissociation from the ssDNA. Successive rounds of ATP hydrolysis can lead to complete disassembly of the nucleoprotein filaments.

Given these considerations, one potential model for how the ATP hydrolysis cycle is coupled to nucleoprotein filament disassembly is that conversion of the end-bound protein monomers, or small clusters of end-bound proteins (e.g., 1–3 monomers) (39) to the ADP-bound state is coupled to entry of these monomers into a highly extended configuration that is no longer compatible with ssDNA binding due to the increased tension that would be required to maintain the DNA in such a highly extended state (Fig. 6C). Conversion of the ADP-bound state also coincides with structural changes in the protein–protein interfaces between proteins within the filament which likely favor dissociation of these protomers from the ends of the nucleoprotein filaments (Fig. 6C) and loss of the second metal ion, as previously described (45).

In considering our model of ATP-hydrolysis coupled structural changes and nucleoprotein filament disassembly, it is important to highlight one caveat, namely, we are not assembling filaments with ATP or allowing them to go through the normal process of ATP hydrolysis. Instead, we are preparing the filaments in the ADP-bound state by providing ADP as the only nucleotide cofactor within the sample mixture. We cannot rule out the possibility that filaments undergoing the normal hydrolysis of ATP to ADP plus Pi may be structurally distinct from those that are assembled in the presence of ADP alone. Lastly, one must also recognize that our model is currently based solely upon structural data and would require further validation using alternative experimental approaches.

Single-molecule studies of recombinase filament dynamics

It should be noted that previous single-molecule experiments had suggested that the ADP-form of RAD51 could adopt a compressed state akin to that observed for RecA (35, 55). However, we see no evidence for a compressed state in our cryo-EM data for S. cerevisiae Rad51 and Dmc1 (Figs. S1 and S2) nor have other cryo-EM structural studies of human RAD51 in the ADP-bound state reported evidence for the existence of a compressed state (45, 46). We do not yet fully understand these apparent discrepancies, but there are several potential explanations that may warrant future experimentation. For example, it is possible that instead of monitoring changes in filament pitch, these studies may have been witnessing cycles of partial filament disassembly and reassembly. Alternatively, as stated above, Rad51 and Dmc1 may transition through a more compressed state when allowed to undergo their normal ATP hydrolysis cycles. It is also possible that the single-molecule experiments may have reflected the existence of a nucleotide-free state that is not yet reflected in the structural studies. Further work will be essential to help distinguish between these and other possibilities.

It has also been reported that applying tension to a dsDNA molecule bound by human RAD51 stabilizes the nucleoprotein filament and prevents disassembly of protein subunits within the nucleoprotein filaments even when in the ADP-bound state, with disassembly completely stalling at 50 pN of force (39). We speculate that the stalled disassembly seen at high tension on the DNA may have been caused by allowing the ADP-bound state to remain on the DNA by stretching the DNA so that it can maintain physical contact with the more extended ADP-bound state of RAD51. The disassembly that was observed upon reducing tension may have been triggered by a structural mismatch between the extended ADP-bound state of RAD51 now having a structural mismatch with the shortened DNA.

Lastly, another single-molecule study of human RAD51 suggested the existence of a force-dependent transition between two states for ATP-bound human RAD51, a more compacted state and a more extended state (43). This study also concluded that the ADP-bound RAD51 filaments adopted a single state, the length of which would be similar to the ATP-bound states (43). These authors also reported a crystal structure for human closely matching the yeast Rad51 filament with a helical pitch 128 Å (43). However, we note that unlike all other Rad51 structures in the ATP-bound state, the ssDNA was not visible in this structure and although the bound nucleotide was modeled as ATP, the electron density of the gamma phosphate was not observed, so it is possible that these studies may have also reflected the ADP-bound state.

Experimental procedures

Proteins purification

S. cerevisiae Rad51 was overexpressed in E. coli BL21 (DE3) Rosetta2 cells transformed with a plasmid encoding 6XHis-SUMO-Rad51. Cells were grown in 2 L LB media containing 100 μg/ml carbenicillin and 35 μg/ml chloramphenicol at 37 °C to an absorbance (A) ₆₀₀ = 0.6, then induced with 0.5 mM IPTG, and grown for 3 h at 37 °C. Cells were harvested by centrifugation and the resulting cell paste was resuspended in 50 ml lysis buffer (50 mM Tris–HCl [pH 7.5], 10% glycerol, 1 M NaCl, 1 mM DTT, 0.5 mM PMSF, 15 mM imidazole, 0.1% Tween 80, 1 protease inhibitor tablet [Roche, Cat No: 5892970001]) and then lysed by sonication. The lysate was centrifuged at 35,000 rpm for 45 min, followed by precipitating the supernatant with 12 g ammonium sulfate for 1 h. The precipitate was spun down at 10,000 rpm for 30 min. The ammonium sulfate pellet was dissolved in 50 ml binding buffer (25 mM Tris–HCl [pH 7.5], 10% glycerol, 200 mM NaCl, 0.1% Triton X-100, 15 mM imidazole, 5 mM beta-mercaptoethanol) and applied to a 5 ml HisPur nickel-nitrilotriacetic acid resin (Thermo Fisher Scientific) equilibrated with the same binding buffer. The protein was eluted with 10 ml elution buffer (25 mM Tris–HCl [pH 7.5], 10% glycerol, 200 mM NaCl, 200 mM imidazole, 0.1% Triton X-100). SUMO protease was added to the elution, followed by dialyzing for 16 h at 4 °C in dialysis buffer (50 mM Tris–HCl [pH 7.5], 200 mM NaCl, 10% glycerol, 15 mM imidazole, 1 mM DTT). The sample was reapplied to the 5 ml nickel-nitrilotriacetic acid resin equilibrated with the binding buffer and the flow-through was collected and concentrated to 80 μM using a spin concentrator (Vivaspin 6, 10 kDa molecular weight cut-off; Cytiva, Cat No: 28932296). The protein was flash-frozen in liquid nitrogen and stored at −80 °C.

S. cerevisiae Dmc1 was overexpressed in E. coli BL21 (DE3) Rosetta2 cells transformed with a plasmid encoding 6XHisDmc1. Cells were grown in 2 L LB media containing 100 μg/ml carbenicillin and 35 μg/ml chloramphenicol at 37 °C to an OD₆₀₀ = 0.8, then induced with 0.1 mM IPTG, and grown for 16 h at 16 °C. The cells were harvested by centrifugation and the cell paste was suspended in 100 ml lysis buffer (50 mM Tris–HCl [pH 7.5], 10% glycerol, 500 mM KCl, 0.01% Triton X-100, 1 mM DTT, 2 mM ATP, 2 mM MgCl₂, 1 mM PMSF, 1 protease inhibitor tablet [Roche, Cat No: 5892970001]) and then lysed by sonication. The lysate was centrifuged at 35,000 rpm for 45 min and the supernatant was applied to 5 ml Talon resin (Takara) equilibrated with the binding buffer (25 mM Tris–HCl [pH 7.5], 10% glycerol, 150 mM KCl, 0.01% Triton X-100, 2 mM ATP, 2 mM MgCl₂). The resin was washed with wash buffer (25 mM Tris–HCl [pH7.5], 10% glycerol, 500 mM KCl, 0.01% Triton X-100, 2 mM ATP, 2 mM MgCl₂), followed by re-equilibrating with the binding buffer. The protein was eluted with 10 ml elution buffer (25 mM Tris–HCl [pH 7.5], 10% glycerol, 150 mM KCl, 200 mM imidazole, 0.01% Triton X-100, 2 mM ATP, 2 mM MgCl₂), followed by dialyzing for 16 h at 4 °C in dialysis buffer (25 mM Tris–HCl [pH 7.5], 100 mM KCl, 10% glycerol, 0.01% Triton X-100, 2 mM MgCl₂, 0.5 mM EDTA). After dialysis, the sample was injected to a 1 ml heparin sepharose column (GE Healthcare) and fractionated with a 100 to 600 mM KCl gradient. The fractions containing Dmc1 were combined, concentrated to 80 μM with a spin concentrator (Vivaspin 6, 10 kDa molecular weight cut-off; Cytiva, Cat No: 28932296), then flash-frozen in liquid nitrogen, and stored at −80 °C.

Cryo-EM sample preparation

To prepare the S. cerevisiae Rad51 filaments in the ADP-bound state for cryo-EM analysis, purified Rad51 (20 μM) was mixed with 0.5 μM of a 96-mer ssDNA (IDT; 5′– AAT TCT CAT TTT ACT TAC CGG ACG CTA TTA GCA GTG AAA ATT TCC TGA TAG TCG TCA CCG CGT TTT GCG CAC TCT TTC TCG TAG GTA CTC AGT CCG–3′) in HR buffer (30 mM Hepes [pH 7.5], 50 mM KCl, 20 mM MgCl₂, 1 mM DTT) supplemented with 5 mM ADP and incubated at 30 °C for 10 min. A sample volume of 3.5 μl was applied to a glow-discharged UltrAuFoil grid (R 0.6/1, 300 mesh), blotted for 4.5 s, and plunge-frozen in liquid ethane using Vitrobot Mark IV (FEI) at 100% humidity and 4 °C.

To prepare the S. cerevisiae Dmc1 filaments in the ADP-bound state for cryo-EM analysis, purified Dmc1 (2.5 μM) was mixed with 0.125 μM of 96-mer ssDNA in HR buffer (as above) supplemented with 1.5 mM CaCl₂ and 5 mM ADP. The sample was incubated at 30 °C for 5 min. The samples were then supplemented with 8 mM CHAPSO (3-([3-cholamidopropyl]dimethyammonio)-2-hydroxy-1-propanesulfonate; Hampton Research, Cat No: HR2-406-87), prior to being applied to UltrAuFoil grid (R 0.6/1, 300 mesh), and plunge-frozen in liquid ethane, as described above for Rad51.

Cryo-EM data acquisition

Samples were initially screened using a Glacios (Thermo Fisher Scientific, 200 keV) at Columbia University Irving Medical Center. Grids selected for high-resolution data collection were imaged using a Titan Krios (Thermo Fisher Scientific) microscope equipped with a K3 direct electron detector (Gatan). Rad51 data were collected at National Cryo-Electron Microscopy Facility at National Cancer Institute with the microscope operated at 300 keV in electron counting mode, with a defocus range of −0.75 μm to −1.75 μm and magnification of 1,050,00x corresponding to 0.855 Å image pixel and a nominal dose of 50 e^-/Ų. A total of 14,085 micrographs were collected. Dmc1 data were collected at Columbia University Irving Medical Center with the microscope operated at 300 keV in electron counting mode, with a defocus range of −0.75 μm to −1.75 μm and magnification of 1,050,00x corresponding to 0.823 Å image pixel size and a nominal dose of 58 e^-/Ų. Total of 6390 micrographs were collected.

Rad51-ADP cryo-EM data processing

Raw movies were processed using cryoSPARC v4.3.1 (https://cryosparc.com) (56). The beam induced motion was corrected and contrast transfer function (CTF) was estimated by patch motion correction and patch CTF estimation jobs (57, 58). The aligned micrographs were manually examined for quality such as ice contamination, and 12,982 micrographs were selected for further processing (Fig. S1A). A total of 1,748,800 particles were picked by template-free blob picking and extracted with a box size of 360 × 360. After removing junk particles, such as denatured proteins and low-quality filaments, by three rounds of 2D classification, 615,261 particles were used to generate 4 ab initio 3D templates, followed by heterogenous refinement (Fig. S1, B–C). A class corresponding to the nucleoprotein filament with 428,741 particles was subjected for another round of 2D classification, and clean-up resulted in 330,601 final particles representing the Rad51-ADP density map (Fig. S1D). The density map was further polished by nonuniform refinement and CTF refinement. The nominal resolution of the density map of 3.37 Å was estimated by 0.143 gold standard Fourier shell correlation cut off (Fig. S1E).

Dmc1-ADP cryo-EM data processing

Raw movies were processed using cryoSPARC v4.3.1 (56). The beam induced motion was corrected and CTF was estimated by patch motion correction and patch CTF estimation jobs (57, 58). The aligned micrographs were manually examined for quality such as ice contamination, and 6044 micrographs were selected for downstream processing (Fig. S2A). A total of 1,184,484 particles were picked by template-free blob picking and extracted with a box size 360 × 360. During the imaging processing, we noticed that the nucleoprotein filaments showed bundled filaments in the micrographs and 2D class average images (Fig. S2B). After 2D classification, 1,124,382 particles were selected for generating 4 ab initio 3D templates, followed by heterogenous refinement (Fig. S2C). A class with 960,233 particles corresponding to nucleoprotein filament was examined for possible contacts between the bundled filaments, but no clear atomic contacts between the filaments were identified. Moreover, the same exact symmetry was observed between the filaments suggesting each was identical. Therefore, we applied a focus mask corresponding to a filament and then used this mask to generate a single filament map (Fig. S2D). Further particle subtraction and local refinement of the filament, resulting in the nominal resolution at 2.74 Å estimated by 0.143 gold standard Fourier shell correlation cut off (Fig. S2E).

Structure refinement

To refine the Rad51-ADP and Dmc1-ADP filament structures, our previously determined cryo-EM structures of Rad51-ATP bound state (PDB: 9D46) and Dmc1-ATP bound state (PDB: 9D4N) were used as starting models (44). Each protomer from the ATP-bound state structures were treated as a separate monomer and fitted into the corresponding cryo-EM density maps using ChimeraX (59). After initial rigid-body refinement in Phenix (https://phenix-online.org) (60), amino acid residue side chains were manually inspected and corrected for fitting into the density map using Coot (61). The ssDNA and disordered part of the DNA-binding loops were not modeled in the structures as the density is disordered. Fitted models were real space refined with the secondary structure and Ramachandran restraints in Phenix (60).

Structure analysis

The helical symmetry was not applied during 3D map reconstruction, and for the comparison of the protomer–protomer interface, protomers 3 and 4 were used. Globally, there are no large differences in the overall conformation between 6 protomers with a RMSD of 0.85 Å for Rad51 and 0.68 Å for Dmc1. Locally, some of residues show minor differences in positioning of side chains for the 6 protomers. However, in our study, we compared those interactions globally from ATP-bound state to ADP-bound state using protomers 3 and 4. Given our resolution, we would not want to assess small changes in amino acid side chain positions for each different protomer out of caution for overinterpreting the data. The pitch and monomers per turn values were analyzed by symmetry search utility function of cryoSPARC with search parameters of helical pitch between 10 Å and 150 Å and number of subunits per turn between 5 and 8 (56). The best local minima values are reported. Overall protomer–protomer contacts were initially identified by using Mapiya web service (62). The interactions were further investigated by contacts function in ChimeraX (59). All the interactions were verified by visual inspection. All the Figures were generated by ChimeraX.

Data availability

The refined structural models and cryo-EM density maps have been deposited in the Protein Data Bank (www.rcsb.org) and Electron Microscopy Databank (www.ebi.ac.uk/emdb) under accession numbers PDB: 9NJK and EMD-49485 for Rad51-ADP and PDB: 9NJR and EMD-49488 for Dmc1-ADP.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Patrick J. O'Brien