Abstract
Homologous recombination (HR) repairs DNA double-strand breaks and stabilizes stressed replication forks, and HR deficiency promotes genome instability and cancer. HR requires assembly of RAD51 nucleoprotein filaments on single-stranded DNA (ssDNA), a process regulated by the human RAD51 paralogs RAD51C, XRCC3, RAD51D and XRCC2. Here, using cryo-electron microscopy, we find that the RAD51–XRCC3–RAD51C complex (RAD51–X3C) assembles into an octamer in which XRCC3 engages the RAD51 DNA-binding surface and RAD51 subunits adopt a misaligned configuration incompatible with filament formation. These features define an autoinhibited RAD51–X3C state that limits nonproductive RAD51 binding to double-stranded DNA or RNA–DNA hybrids while preserving RAD51 availability for ssDNA-dependent strand exchange. We further show that the RAD51D–XRCC2 paralog complex remodels RAD51–X3C into a pentameric RAD51–X3CDX2 assembly by engaging the exposed RAD51C surface and disrupting contacts that stabilize the octamer. This remodeling exposes the RAD51 DNA-binding interface, enhances RAD51–ssDNA filament assembly, and promotes strand exchange on RPA-coated ssDNA, and yields a filament-compatible paralog assembly that integrates into ssDNA-bound RAD51 filaments. Together, these findings establish paralog exchange as a mechanism that converts an autoinhibited RAD51–X3C octamer into an activated RAD51–X3CDX2 pentamer to regulate RAD51 filament formation during HR and replication fork preservation.
Similar content being viewed by others
Main
Homologous recombination (HR) is a conserved and high-fidelity DNA repair pathway that maintains genome stability by repairing DNA double-strand breaks (DSBs) and preserving damaged or stalled DNA replication forks1,2. In HR triggered by a DSB, DNA ends are resected nucleolytically to generate 3′-tailed single-stranded DNA (ssDNA) that serves as the template for assembly of a helical RAD51 filament3,4. The resulting RAD51–ssDNA nucleoprotein filament (presynaptic filament) works with several cofactors such as BRCA1–BARD1 and RAD51AP1–UAF1 (refs. 5,6,7) to conduct a homology search and invasion of homologous DNA sequence. Several mediators of RAD51 presynaptic filament assembly, including yeast Rad52, Rad55–Rad57 and human BRCA2–DSS1, have been described and thoroughly analyzed8,9,10,11.
Five RAD51 paralogs, namely, RAD51B, RAD51C, RAD51D, XRCC2 and XRCC3, have been implicated in RAD51 nucleoprotein filament assembly in mammalian cells12. Depletion of any one of these RAD51 paralogs renders cells sensitive to genotoxic agents and unable to assemble RAD51 foci upon occurrence of DNA damage13. Accordingly, mutations in RAD51 paralog genes are prevalent in persons with breast, ovarian and other cancer types14,15,16 and mutations in RAD51C (FANCO) and XRCC2 (FANCU) have also been associated with the cancer-prone disease Fanconi anemia17,18. To date, two complexes harboring RAD51 paralogs have been thoroughly studied, namely, the tetrameric RAD51B–RAD51C–RAD51D–XRCC2 (BCDX2) and dimeric XRCC3–RAD51C (X3C)19. The BCDX2 complex has been analyzed by cryo-electron microscopy (cryo-EM)20,21 and complementary biochemical and single-molecule DNA curtain analyses demonstrated that BCDX2 stimulates the nucleation and extension of RAD51 filaments in an ATPase-dependent manner and mediates RPA–RAD51 exchange on ssDNA14,20,22. A recent X-ray crystal structure of X3C from the extremophile metazoan Alvinella pompejana revealed a compact heterodimer with bound ATP analog and companion cell-based studies provided evidence for a crucial role of distinct X3C surfaces in DNA replication fork maintenance23. While these findings highlight X3C as a key replication stress response factor, its structural organization with RAD51 and the mechanisms by which it regulates RAD51 presynaptic filament assembly remain unknown.
Here, we report cryo-EM structures showing that the RAD51–X3C complex forms an octameric assembly in which XRCC3 occupies the canonical RAD51 DNA-binding surface and RAD51 subunits adopt a misaligned arrangement that disrupts filament continuity. This configuration limits nonproductive RAD51 interactions with duplex nucleic acids. In biochemical assays, RAD51–X3C retains ssDNA binding but is strongly impaired for double-stranded DNA (dsDNA) or RNA–DNA hybrid engagement and supports RAD51 loading onto ssDNA even in the presence of excess dsDNA. XRCC3 mutations that disrupt RAD51 interaction interfaces reduce this activity in biochemical assays and impair HR-related functions in cells. In the presence of the RAD51D–XRCC2 (DX2) paralog complex, RAD51–X3C converts from the octameric state into a pentameric RAD51–X3CDX2 assembly. In this remodeled architecture, the RAD51 DNA-binding interface that is occluded in the octamer becomes exposed. Biochemical analyses show that RAD51–X3CDX2 enhances RAD51 binding to ssDNA and promotes presynaptic filament assembly on RPA-coated ssDNA and cryo-EM analysis of the ssDNA-bound complex reveals how the remodeled paralog assembly integrates into an active RAD51 filament. This synergistic activity likely reflects a coordinated mechanism in which DX2 disrupts the autoinhibited RAD51–X3C conformation and promotes productive filament nucleation during HR. Together, our findings uncover a regulatory mechanism in which RAD51 paralog exchange activates a previously autoinhibited RAD51–X3C complex, thereby promoting efficient filament formation and recombination activity.
Results
Cryo-EM structure of the octameric RAD51–X3C complex
X3C is known to physically interact with RAD51 (refs. 23,24). To reconstitute this higher-order complex, purified X3C was incubated with RAD51 on ice for 30 min and the resulting RAD51–X3C complex was isolated by size-exclusion chromatography in the presence of ATP/Mg2+ (Fig. 1a). The elution profile showed a high-molecular-weight species (Fig. 1a), consistent with formation of a higher-order assembly. Notably, a similar RAD51–X3C complex was obtained upon coexpression of RAD51, XRCC3 and RAD51C in insect cells and purification in ATP/Mg2+ buffer (Extended Data Fig. 1a).
a, Top: size-exclusion chromatograms of RAD51–X3C and X3C complexes and molecular weight standards in the presence of ATP/Mg2+ using a Superdex 200 Increase 10/300 column. Bottom: Coomassie-stained SDS–PAGE gels of the indicated fractions are shown, representative of two independent experiments. Protein molecular weight standards: Tg, thyroglobulin; Fe, ferritin; Al, aldolase; Ca, conalbumin; Oa, ovalbumin. b, Cryo-EM map of the RAD51–X3C complex assembled in the presence of AMP-PNP. Individual subunits are labeled and color-coded. Multiple views are shown, related by the indicated rotations. c, Schematic cartoon illustrating the protomer and subunit organization within the octameric RAD51–X3C complex. d, Cartoon representation of the RAD51–X3C complex. Bound AMP-PNP molecules are shown as spheres.
Following buffer exchange to replace bound ATP with the nonhydrolyzable analog AMP-PNP, we determined a cryo-EM reconstruction of RAD51–X3C at an overall masked resolution of 3.25 Å (Fig. 1b, Table 1 and Extended Data Fig. 1b). Our cryo-EM map revealed an octameric assembly organized as two head-to-tail tetramers, with each tetramer exhibiting a 2:1:1 arrangement of RAD51, XRCC3 and RAD51C, respectively (Fig. 1b–d and Extended Data Fig. 1c–f). Each tetramer consists of two RAD51 molecules followed sequentially by XRCC3 and RAD51C (Fig. 1b–d and Extended Data Fig. 1f). We designate the two tetramers as protomer A (comprising RAD51.A1, RAD51.A2, XRCC3.A and RAD51C.A) and protomer B (with the analogous arrangement RAD51.B1, RAD51.B2, XRCC3.B and RAD51C.B). Each protomer contains well-ordered AMP-PNP molecules: one bound to the first RAD51 subunit and one positioned at the RAD51–XRCC3 and X3C interfaces (Fig. 1b–d and Supplementary Fig. 1). Within this octameric architecture, the four RAD51 molecules pack into a ring-like arrangement, with two X3C modules projecting outward like antennae. The extensive interaction networks observed within each tetramer, together with limited intertetramer contacts occurring primarily between XRCC3 of one tetramer and the terminal RAD51 of the adjacent tetramer, serve to hold the octamer together (Fig. 1b–d and Extended Data Fig. 2).
Structural organization of the RAD51–X3C complex
RAD51 filaments are stabilized by sequential stacking of protomers in a head-to-tail arrangement, in which the ATPase domain of one subunit engages the ATPase domain of its neighbor to form an asymmetric, directional interface (Fig. 2). ATP binding at this junction is coordinated by residues from both protomers and promotes filament assembly and stability (Fig. 2a)25. In addition, a conserved FXXA polymerization motif, most notably F86 and A89 in human RAD51, normally inserts into a hydrophobic pocket on the adjacent subunit, further reinforcing filament contacts (Fig. 2c,d)26. Although the N-terminal domain (NTD) is not a major contributor to intersubunit interactions in filaments, it may have regulatory roles through DNA binding or interactions with recombination mediators27.
a, Cartoon representations of the indicated complexes, with AMP-PNP (nonhydrolyzable ATP analog) and DNA shown as spheres. RAD51C from the BCDX2 complex and RAD51.1 from the presynaptic RAD51 filament were superimposed onto RAD51.A2 from the RAD51–X3C structure. XRCC3 from the A. pompejana crystal structure was superimposed onto XRCC3 from RAD51–X3C. b, Subcomplexes from the RAD51–X3C structure (RAD51C–XRCC3, RAD51.A2–XRCC3 and RAD51.A1–RAD51.A2) are shown side by side to compare subunit organization. Right: the RAD51 presynaptic filament is shown for reference. c, Structural analysis of the regions corresponding to the canonical RAD51 polymerization motif in the RAD51–X3C complex. d, Structure-based sequence alignment of the RAD51 polymerization motif. e, Structure-based sequence alignment of the RAD51 polymerization motif-binding pocket. Residues corresponding to the canonical RAD51 motif and binding pocket are boxed in red; divergent positions among RAD51 paralogs are indicated by stars above the alignment.
In contrast, BCDX2 does not rely on FXXA-mediated contacts for intersubunit interactions20,21. Its subunits lack the conserved FXXA sequence and the hydrophobic pocket found in RAD51 is occluded by bulky or charged residues (Fig. 2d,e). Instead, BCDX2 is stabilized by extensive interactions involving the ATPase and NTDs, which form a compact ordered tetrameric structure (Fig. 2a). The RAD51B, RAD51C and RAD51D NTDs adopt outward-facing conformations that contribute to unique intersubunit contacts in BCDX2. The ATPase domain of RAD51B is unresolved in available structures, suggesting a lack of stabilizing interactions with the RAD51C ATPase domain.
The RAD51–X3C complex adopts a distinct structural arrangement compared to both RAD51 filaments and BCDX2. Whereas the ATPase domain of RAD51B is unstructured and fails to contact RAD51C in BCDX2, the XRCC3 ATPase domain is well ordered and packs against the ATPase domains of RAD51C and RAD51.A2, forming a more continuous and ordered interface than is observed in BCDX2 (Fig. 2a). In RAD51–X3C, XRCC3 contributes a hydrophobic pocket that accommodates the FXXA motif of RAD51.A2, consistent with canonical FXXA-type interactions observed among RAD51 paralogs. On the opposite face of XRCC3, Ile62 (equivalent to F86 in RAD51) inserts into a shallow hydrophobic pocket on RAD51C, with additional contacts extending from A64 that form a short hydrophobic helix, resembling the local interaction pattern seen in BCDX2. (Fig. 2c,d and Extended Data Fig. 2b). These interactions are consistent with recent saturation genome editing (SGE) analyses of RAD51C, which showed that mutations impacting several of these interfacial residues are associated with reduced cellular fitness28. Canonical FXXA interactions are also retained between RAD51.A2 and RAD51.A1 (Fig. 2c,d and Extended Data Fig. 2a). Altogether, these contacts may help stabilize sequential stacking of subunits within the tetrameric RAD51–X3C protomers.
The organization of the NTDs further differs between the RAD51–X3C assembly and both RAD51 filaments and the BCDX2 complex. The XRCC3 NTD is ordered and contacts the RAD51C ATPase domain, in contrast to contacts observed between the RAD51B NTD and RAD51C in BCDX2 (Fig. 2a and Extended Data Fig. 2c). By contrast, the RAD51C NTD is not resolved, likely because of the absence of a neighboring ATPase domain to stabilize it, as is also seen in the A. pompejana X3C crystal structure (Fig. 2a)23. In canonical RAD51 filaments, the NTDs fold back toward their own ATPase domains and contribute minimally to intersubunit contacts (Fig. 2a,b)25. In RAD51–X3C, however, the RAD51 NTDs adopt distinct conformations and participate in additional contacts that may help stabilize the protomer (Fig. 2a,b and Extended Data Fig. 2a). The RAD51.A2 NTD contacts the ATPase domain of XRCC3, while the RAD51.A1 NTD contacts the ATPase domain of RAD51.A2, potentially contributing to protomer organization (Extended Data Fig. 2a,b). Although the NTDs of RAD51.A1, RAD51.A2 and XRCC3.A do not form the tripartite NTD–NTD interaction network observed in BCDX2, they nonetheless participate in local intersubunit contacts. Rather than forming an NTD stacking pathway, RAD51–X3C stabilizes local interfaces through diverse surface contacts within the NTDs (Extended Data Fig. 2).
Within the core of the RAD51–X3C protomer, RAD51.A2, XRCC3.A and RAD51C.A adopt the canonical head-to-tail protomer stacking arrangement characteristic of RAD51 filaments, with ATP molecules bridging adjacent ATPase domains (Fig. 2a,b). However, the interface between RAD51.A2 and RAD51.A1 deviates from this pattern. The ATPase domain of RAD51.A1 is rotated relative to its counterparts in other RAD51 or BCDX2 structures, creating a noncanonical interface that separates the two ATP half-sites and may restrict ATP binding at the junction (Fig. 2b). Contacts between the ATPase domains of these two RAD51 subunits are less extensive than those of RAD51 subunits within a presynaptic filament. As described above, the RAD51.A1 NTD adopts a distinct conformation that allows it to reinforce the RAD51.A2–RAD51.A1 interface through direct contact with the RAD51.A2 ATPase domain (Fig. 2b). As discussed further below, this noncanonical RAD51.A1–RAD51.A2 arrangement is further stabilized by contacts with the adjacent protomer in the octameric assembly.
XRCC3–RAD51 interactions and their biological importance
The subunit arrangement and interaction interfaces within the RAD51–X3C octamer underpin important functional implications (Fig. 3). In particular, our structure reveals a unique loop insertion in XRCC3 that reaches across the interprotomer interface to insert into a deep pocket on the surface of the terminal RAD51 subunit in the other protomer, an interaction that appears to be driven by shape complementarity and long-range electrostatic interactions (Fig. 3a,b). We analyzed the relevance of XRCC3–RAD51 interactions either through the FXXA motif of RAD51 or the loop insertion of XRCC3 in the assembly of the RAD51–X3C octamer. In affinity pulldown assays (Extended Data Fig. 3a), the X3C complex assembled with wild-type XRCC3 or a mutant lacking unique insertion loop residues R302–P311 (loopΔ) showed comparable interaction with RAD51 (Extended Data Fig. 3a). In contrast, XRCC3 with a compound quadruple substitution (quad; E182F, V194L, V198L and L201F), constructed on the basis of our cryo-EM structure of the RAD51–X3C octamer and predicted to disrupt association with the FXXA motif of RAD51 by substitution with bulkier residues (Fig. 2c and Extended Data Fig. 2b), exhibited a substantial reduction in RAD51 interaction (Extended Data Fig. 3a). We also combined the loopΔ and quad XRCC3 and found that the resulting X3C mutant was largely unable to interact with RAD51 (Extended Data Fig. 3a). Likewise, a RAD51 variant ablated for the FXXA motif (F86E;A89E) failed to interact with wild-type X3C (Extended Data Fig. 3b). To determine the relevance of the unique XRCC3 loop insertion, we conducted size-exclusion chromatography to examine the stoichiometry of RAD51–X3C formed by X3C harboring either wild-type or loopΔ mutant XRCC3. RAD51–X3loopΔ/C displayed an appreciably lower apparent molecular mass compared to the wild-type RAD51–X3C octamer (Extended Data Fig. 3c), consistent with a loss of octameric organization. Taken together, these structure-guided mutational analyses provide supporting evidence for the importance of the two RAD51 interaction interfaces in XRCC3 in stabilizing the octameric assembly.
a, The four RAD51 molecules of the RAD51–X3C octamer that pack into a loosely associated ring-like arrangement are shown as surface representations. The X3C modules that emanate from the RAD51 base are shown as cartoon representations. Insets: magnified views of the interaction between the XRCC3 loop insertion with RAD51 at both ends of the structure. b, Top left: cartoon representation of XRCC3 with L1, L2 and the loop insertion labeled. Top right: cartoon representations of RAD51 and RAD51 paralogs superimposed at their ATPase domains. Bottom: sequence alignment of the region corresponding to the XRCC3 loop insertion. c, XRCC3.B, RAD51.A1 and RAD51.A2 from the RAD51–X3C structure are shown as cartoons, colored as in a. The unique XRCC3 loop insertion is colored red. RAD51.A1 residues that interact with DNA in previously determined RAD51–DNA structures are shown as sticks and labeled. d, The ATPase domains of RAD51.A1 and RAD51.A2 from the RAD51–X3C structure and RAD51 from the presynaptic filament structure were superimposed. ATP and nonhydrolyzable ATP analogs are shown as spheres. e, The RAD51 filament was superimposed onto RAD51.A1 of the RAD51–X3C complex. The RAD51 filament is shown in surface representation and DNA is shown as sticks. XRCC3.B from the RAD51–X3C complex is shown as a cartoon. f, DNA from the alignment described in e is shown as sticks in the context of the RAD51–X3C complex. g, A RAD51 molecule from RAD51 filament (red) superimposed onto RAD51.A1 (left) and RAD51.A2 (right) from the RAD51–X3C structure.
X3C is dispensable for replication fork reversal but required for efficient fork restart upon replication stress occurrence29,30 and essential for HR13. To analyze the biological contribution of the two RAD51 interaction interfaces of XRCC3, we used DNA fiber combing to assess stalled forks and relative lengths of restarted forks following hydroxyurea treatment (Extended Data Fig. 3d). As expected, XRCC3 knockdown (Extended Data Fig. 3h) in HeLa cells increased the number of stalled forks (Extended Data Fig. 3e) and reduced fork restart, as measured by the IdU/CIdU ratio (Extended Data Fig. 3f). Expression of wild-type XRCC3 substantially corrected the fork restart defect, whereas stable expression of the loopΔ, quad or loopΔ/quad XRCC3 mutants did not rescue this phenotype (Extended Data Fig. 3d–f,h).
Next, we analyzed the XRCC3 mutants for the promotion of HR in U2OS cells using a well-established GFP-based sister chromatid repair (SCR) assay in which HR is triggered by the introduction of a site-specific DSB induced by the I-SceI endonuclease and HR efficiency is measured by the percentage of GFP-positive cells13,31. All three XRCC3 mutants displayed impaired HR, with the strongest phenotype observed for the loopΔ/quad mutant (Extended Data Fig. 3g,i). Altogether, these results indicate that both XRCC3 interaction interfaces contribute to the integrity and cellular function of X3C, consistent with their structural roles in stabilizing the RAD51–X3C assembly.
Autoinhibited architecture of the RAD51–X3C octamer
The arrangement of tetrameric protomers in the RAD51–X3C octamer also reveals structural insight into its regulatory role in HR (Fig. 3). The unique XRCC3 loop insertion (Fig. 3b) targets a surface on RAD51 that includes loops L1 and L2, which harbor key DNA-binding residues in presynaptic and postsynaptic RAD51 filaments (Fig. 3c). Superposition of the RAD51 subunit from a presynaptic filament onto RAD51.A1 of RAD51–X3C reveals steric clashes between the XRCC3.B loop insertion and one of the nucleotide triplet clusters of the presynaptic filament (Fig. 3e). Additional clashes between L2 of XRCC3.B and the ATPase domain of the neighboring RAD51 in presynaptic filament, which corresponds to RAD51.A2, are also observed in the RAD51–X3C structure (Fig. 3e). These latter clashes are circumvented in the RAD51–X3C structure because of the noncanonical ATP-independent interaction between RAD51.A1 and RAD51.A2, which projects RAD51.A2 away from XRCC3.B, creating physical separation that avoids steric conflict (Fig. 3f). This arrangement also disrupts the continuity of DNA-binding residues observed in filamentous RAD51, consistent with reduced accessibility of DNA-contacting residues (Fig. 3f). Together, these features support the conclusion that the RAD51–X3C octamer represents an autoinhibited conformation in which DNA binding may be limited by two structural features: direct occlusion of the canonical RAD51 DNA-binding surface by the XRCC3 loop insertion and misalignment of key DNA-contacting residues because of the noncanonical RAD51.A1–RAD51.A2 interface.
Further structural comparison reveals that the NTDs of RAD51 also have an active role in this regulatory arrangement. Superposition of RAD51.A1 or RAD51.A2 onto RAD51 from presynaptic filament structures shows that, while the ATPase domains align well, the NTDs adopt an array of different conformations (Fig. 3d). Superposition of presynaptic RAD51 onto RAD51.A1 or RAD51.A2 from the RAD51–X3C structure shows that the NTD would clash with the ATPase domain or NTD of the opposing subunit, respectively (Fig. 3g). These clashes arise from the noncanonical orientation of RAD51.A1 and RAD51.A2 in the complex, which diverges markedly from the geometry of RAD51 protomers in filamentous assemblies. In addition to preventing steric interference, these altered NTD conformations contribute productively to RAD51–X3C complex formation; the RAD51.A1 NTD packs against the ATPase domain of RAD51.A2, while the RAD51.A2 NTD engages the ATPase domain of XRCC3.A (Figs. 2a,b and 3g). Lastly, the unique RAD51.A1–RAD51.A2 interface places the hydrophobic pocket of RAD51.A1 in close proximity to the XRCC3.A loop insertion (Fig. 2a,b). This spatial arrangement suggests that loop engagement, ATPase domain rotation and NTD repositioning contribute to the autoinhibited architecture of the RAD51–X3C complex.
Promotion of ssDNA engagement by X3C
To test the functional relevance of this autoinhibited state, we next asked whether X3C might regulate the DNA-binding properties of RAD51. It is well established that, while RAD51 binds both dsDNA and ssDNA, only nucleoprotein filaments formed on ssDNA are catalytically active32. We used electrophoretic mobility shift assays (EMSAs) with 80-nt ssDNA or 80-bp dsDNA as substrates to test RAD51–X3C alongside RAD51 and X3C for DNA binding. ATP was included in the reaction buffer as it is needed for the assembly of catalytically active RAD51 presynaptic filaments. The results revealed that X3C binds ssDNA weakly and has negligible dsDNA-binding activity (Extended Data Fig. 4a)33. Our analysis further showed that RAD51–X3C binds ssDNA with an affinity comparable to that seen for RAD51 (Extended Data Fig. 4b). Similar results were obtained when we substituted ATP with the nonhydrolyzable analog AMP-PNP in the EMSA, revealing an apparent Kd of 39.8 ± 8.44 nM and 62.5 ± 15.27 nM for RAD51 and RAD51–X3C, respectively (Fig. 4a,b, top, and Extended Data Fig. 4d, top). In sharp contrast, RAD51–X3C displayed markedly reduced binding to dsDNA compared to RAD51 in the presence of both ATP and AMP-PNP. Specifically, in the presence of AMP-PNP, while RAD51 bound dsDNA with an apparent Kd of 68.4 ± 19.5 nM, the much-reduced affinity of RAD51–X3C for dsDNA prevented Kd estimation over the tested protein concentration range (Fig. 4a,b and Extended Data Fig. 4c,d). Importantly, mixing purified X3C with RAD51 also led to a marked attenuation of dsDNA binding but had little effect on ssDNA engagement (Extended Data Fig. 4e,f).
a,b, Gel images (a) and quantifications (b) of 80-nt ssDNA (top), 80-bp dsDNA (middle) and 60-bp RNA–DNA hybrid (bottom) binding of RAD51 and RAD51–X3C at the indicated RAD51 concentrations in the presence of AMP-PNP, assessed by EMSAs. Results from four independent experiments are plotted as the mean values ± s.d. c, Top: scheme of ssDNA EMSA for protein preincubated with an eightfold excess of phiX174 circular dsDNA (1.24 μM bp). The asterisk indicates Cy5 label on 80 nt ssDNA. Middle: ssDNA binding of RAD51 and RAD51–X3C at the indicated RAD51 concentrations in the presence of 1 mM AMP-PNP following preincubation with excess circular dsDNA. Bottom: results from three independent experiments are plotted as the mean values ± s.d. d, Top: reaction schematic. Middle: strand exchange with increasing concentrations of either RAD51 alone or RAD51–X3C in presence of AMP-PNP after preincubation with 8 μM bp pBlueScript plasmid DNA. The positions of the Cy5‑labeled (indicated by asterisk) 40‑bp dsDNA and the strand‑exchange product are annotated alongside the gel images. Bottom: results from four independent experiments are plotted as the mean values ± s.e. e, ssDNA binding of RAD51 preincubated with excess of circular dsDNA in the presence of increasing concentrations (12.5, 25 and 50 nM) of WT and indicated X3C mutants. Results from three independent experiments are plotted as the mean values ± s.d. f, Strand exchange in the presence of AMP-PNP with increasing concentrations of RAD51 as indicated in c. RAD51 alone or RAD51 and WT or indicated X3C mutants mixed in a 2:1 ratio after preincubation with 8 μM bp pBlueScript plasmid DNA. Results from three independent experiments are plotted as the mean values ± s.e. Gel images in c,e,f were derived from the same experiment and processed in parallel.
Because promiscuous RAD51 binding to genomic dsDNA can sequester RAD51 away from DNA repair sites, which can impede HR and lead to the accumulation of toxic RAD51–dsDNA complexes34, facilitated targeting of RAD51 to ssDNA is thought to be important for efficient homology search and strand invasion during HR32,35. To further determine the DNA substrate specificity of the RAD51–X3C complex, we performed EMSA in which RAD51 or RAD51–X3C was preincubated with an eightfold molar excess of dsDNA as a protein trap before the incorporation of labeled ssDNA (Fig. 4c). Under these conditions, RAD51 alone failed to bind ssDNA upon its preincubation with dsDNA, whereas ssDNA binding by RAD51–X3C remained largely unaffected (Fig. 4c and Extended Data Fig. 4g). These results again support the premise that X3C restricts dsDNA engagement by RAD51 while preserving access of RAD51 to ssDNA. We note that association of RAD51 with dsDNA in the genome or present at a DSB would sequester RAD51 from forming functional presynaptic filaments at resected DSBs32,35. Together, these findings support a model in which X3C limits nonproductive RAD51–dsDNA associations while preserving RAD51 access to ssDNA in conditions that favor presynaptic filament assembly.
Role of X3C in RAD51-mediated DNA strand exchange
We next tested the influence of X3C on RAD51-mediated DNA strand exchange using two distinct protocols. In the first reaction protocol, RAD51 and RAD51–X3C were preincubated with ssDNA before the commencement of DNA strand exchange (Extended Data Fig. 4h). Importantly, RAD51–X3C displayed significantly enhanced DNA strand exchange activity over RAD51 alone across the protein concentrations tested (Extended Data Fig. 4h). In the second reaction protocol, we first challenged RAD51 or RAD51–X3C with dsDNA as a protein trap before incubation with the ssDNA substrate and the DNA strand exchange commencement (Fig. 4d, top). Under these conditions, RAD51 alone failed to catalyze significant DNA strand exchange, while RAD51–X3C retained full DNA strand exchange capability (Fig. 4d, middle).
Next, we examined the functional impact of the XRCC3 loopΔ, quad, and the combined loopΔ/quad mutants on RAD51 behavior. The results revealed that all three mutant X3C complexes are compromised in their ability to target RAD51 to ssDNA in EMSA in the presence of a dsDNA trap (Fig. 4e) and facilitate RAD51-mediated DNA strand exchange challenged with (Fig. 4f) or without (Extended Data Fig. 4i) dsDNA trap. Taken together, these results support the conclusion that both RAD51 interaction interfaces in XRCC3 are important for efficient engagement of RAD51 with ssDNA in conditions that favor presynaptic filament assembly.
Restriction of RNA–DNA association by X3C
The accumulation of RNA–DNA hybrids at DSBs within transcriptionally active regions can hinder RAD51 loading and presynaptic filament formation, as RAD51 cannot bind ssDNA at these sites in the absence of RNA–DNA hybrid clearance by the RNA–DNA helicase SETX36,37. Moreover, the DNA polymerase α–primase complex is recruited to resected DSBs, where it deposits RNA primers and performs DNA synthesis to reduce the ssDNA character of processed DSB ends. These events are associated with the selection of nonhomologous end joining (NHEJ) and biasing the DSB repair pathway toward NHEJ38. Given the above, it was of interest to test whether X3C would also prevent RAD51 from engaging RNA–DNA hybrids as an attribute that would be conducive for presynaptic filament assembly when cellular mechanisms, such as those involving RNase H or SETX39,40, help to clear these nucleic acid hybrids. We found that, as expected41,42, while RAD51 alone readily binds an RNA–DNA substrate (apparent Kd of 178.9 ± 42.5 nM), RAD51–X3C is impaired in this binding (Fig. 4a,b, bottom, and Extended Data Fig. 4d). Thus, X3C also reduces the potentially nonproductive association of RAD51 with RNA–DNA hybrids.
Remodeling of RAD51–X3C by DX2 in a pentameric architecture
Although formation of the X3C and BCDX2 complexes in cells has been extensively documented19,23, the structural basis for how these assemblies might engage additional RAD51 paralog subcomplexes has remained unclear. RAD51C is a shared subunit of both complexes and, in the RAD51–X3C structure, occupies the terminal position of each protomer, leaving its RAD51D interaction surface fully exposed (Fig. 2a,b). This arrangement suggests that RAD51C remains competent for subsequent association with the DX2 dimer.
We performed proximity ligation assays (PLAs) to assess whether X3C can associate with DX2 to form a higher-order complex in human cells (U2OS). Importantly, PLA revealed robust interactions between GFP-tagged XRCC3 and HA-tagged RAD51D, XRCC2 and RAD51C (Fig. 5a), which were further enhanced by treatment of cells with ionizing radiation (IR). In contrast, no association between XRCC3 and RAD51B was detected even after IR treatment of cells (Fig. 5a). Consistent with the PLA results, we reconstituted stable X3CDX2 and RAD51–X3CDX2 complexes in vitro (Fig. 5b) and collected cryo-EM data on both assemblies. Of note, addition of DX2 to the RAD51–X3C octameric assembly led to an apparent decrease in the molecular weight of the complex, suggesting that DX2 drives the remodeling of the octameric RAD51–X3C complex (Fig. 5b).
a, Colocalization of GFP–XRCC3 and HA–RAD51B, HA–RAD51C, HA–RAD51D or HA–XRCC2 using PLA. Left: western blot showing the level of GFP–XRCC3 and HA-tagged RAD51B, RAD51D, XRCC2 and RAD51C. GFP-expressing empty vector (EV), GFP–RAD54 and HA-expressing EV were used as controls. Representative images (middle) and quantification (right) of colocalization of GFP–XRCC3 and HA–RAD51B, HA–RAD51C, HA–RAD51D or HA–XRCC2 in nonirradiated (no IR) and irradiated (after IR) (6 Gy, 2 h). GFP, GFP fusion proteins; PLA, GFP/HA (red); HA, HA fusion proteins. DAPI was used to identify nuclei (blue). Scale bar, 10 µm. Crossbars represent the mean ± s.e.m. calculated from indicated data points pooled from three independent experiments. ****P < 0.001 versus the control (EVDMSO). NS, not significant. n is total number of cells analyzed in each condition. b, Top: size-exclusion chromatogram of RAD51–X3C, RAD51–X3CDX2 and X3CDX2 complexes in the presence of ATP/Mg2+ using a Superdex 200 Increase 10/300 column. Bottom: Coomassie-stained SDS–PAGE gels of the indicated fractions, representative of two independent experiments. Chromatograms and gel images for RAD51–X3C and standards are reproduced from Fig. 1a. c, Cryo-EM map of the RAD51–X3CDX2 complex assembled in the presence of AMP-PNP. Individual subunits are labeled and color-coded. Multiple views are shown, related by the indicated rotations. d, Cartoon representation of the RAD51–X3CDX2 complex. Bound AMP-PNP molecules are shown as spheres. Multiple views are shown, related by the indicated rotations. e, The ATPase domains of RAD51.A1 and RAD51.A2 from the RAD51–X3C structure, RAD51 from the RAD51–X3CDX2 structure and RAD51 from the presynaptic filament structure were superimposed. ATP and nonhydrolyzable ATP analogs are shown as spheres. f, The RAD51–X3C and RAD51–X3CDX2 complexes are shown as cartoons, with ATP and nonhydrolyzable ATP analogs shown as spheres. g, DX2 complex superimposed onto RAD51C from the RAD51–X3C complex. Protomer B is shown in surface representation. Protomer A and the modeled DX2 are shown as cartoons. Inset: magnified view of DX2 clashes with subunits of the RAD51–X3C complex.
High-quality reconstructions were achieved at nominal resolutions of 3.51 and 3.16 Å (masked) for RAD51–X3CDX2 (Table 1 and Extended Data Fig. 5) and X3CDX2 (Supplementary Figs. 2 and 3), respectively. Remarkably, introduction of DX2 into the RAD51–X3C complex induced an architectural rearrangement from the octameric RAD51–X3C assembly into a pentameric RAD51–X3CDX2 assembly (Fig. 5c), providing a structural basis for the observed apparent decrease in molecular weight of the complex in analytical size-exclusion chromatography described above (Fig. 5b). As X3CDX2 does not exhibit notable conformational differences compared to its counterpart in RAD51–X3CDX2, we focus on the latter unless otherwise noted (Fig. 5 and Supplementary Fig. 4).
Within RAD51–X3CDX2, the CDX2 module closely resembles the structure of BCDX2 and the RAD51–X3C module is nearly identical to the RAD51.A2–X3C module from one protomer of the RAD51–X3C octamer (Fig. 5c–f and Extended Data Fig. 6). The pentamer is arranged in a sequential, head-to-tail fashion. Four AMP-PNP molecules are well resolved in the cryo-EM density, being bound at the RAD51–XRCC3, X3C, RAD51C–RAD51D and DX2 interfaces (Fig. 5d, Extended Data Fig. 7 and Supplementary Fig. 5). Consistent with our cryo-EM reconstructions, recently published SGE mapping of RAD51C highlighted residues at these interfaces as critical for RAD51C function and cancer risk, reinforcing the importance of this interface in RAD51 paralog assembly28. Lastly, it is notable that, in X3CDX2, a well-ordered AMP-PNP molecule is observed in the ATP-binding half-site of XRCC3 despite the absence of a partnering RAD51 subunit (Supplementary Fig. 4b,c). This unique arrangement is consistent with stable engagement of RAD51 by XRCC3, as compared to the transient interaction between RAD51 and RAD51B in BCDX2 (ref. 20).
To investigate how DX2 remodels RAD51–X3C into a pentamer rather than assembling into a larger complex, we modeled DX2 onto RAD51C from one protomer of the RAD51–X3C octamer. This analysis revealed extensive steric clashes between DX2 and subunits from the opposite protomer, including both XRCC3 and RAD51C (Fig. 5g). The steric hindrance is exacerbated when DX2 is modeled onto both RAD51C copies in the octamer, indicating that dual engagement by DX2 is also incompatible with the octameric assembly (Fig. 5g). Together, these structural findings provide a mechanistic rationale for how DX2 remodels RAD51–X3C into a lower-order complex. By targeting the exposed RAD51D-binding interface of RAD51C, DX2 disrupts the octameric arrangement and promotes formation of an alternative assembly, suggesting that remodeling by paralog exchange may represent a key regulatory step in HR.
Mechanistic testing of RAD51–X3CDX2
On the basis of the structural model, we next asked whether DX2 functionally disrupts the autoinhibited RAD51–X3C architecture, producing a pentameric RAD51–X3CDX2 complex in which the RAD51 DNA-binding surface becomes exposed (Fig. 6). Surface electrostatic analysis of pentameric RAD51–X3CDX2 revealed a highly basic surface extending from the DX2 module to the terminal RAD51 (Fig. 6a) and EMSA showed that X3CDX2 binds ssDNA with a higher affinity than either X3C or DX2 alone (Extended Data Fig. 8a,b). To define the biochemical properties of the RAD51–X3CDX2 complex, we evaluated effects of X3CDX2 on RAD51 binding to ssDNA. In EMSA, limiting concentrations of RAD51–X3CDX2 complex showed markedly enhanced ssDNA binding compared to RAD51 or X3CDX2 alone (Fig. 6b). Enhanced ssDNA binding was also observed upon mixing RAD51–X3C and DX2 (Extended Data Fig. 8c). We also note that the ssDNA-binding attribute of RAD51–X3CDX2 is compromised by the R266A substitution in RAD51D that affects DNA binding20 (Fig. 6b and Extended Data Fig. 8b).
a, The indicated RAD51 and RAD51 paralog complexes are shown as surface electrostatic representations. For RAD51–X3C, one protomer is shown in cartoon representation and the other is shown in surface electrostatic representation. b, ssDNA binding of RAD51 (20 nM) and indicated RAD51 paralog complexes (20 nM each) in the presence of AMP-PNP. Results from four independent experiments are plotted as the mean values ± s.d. Note that little or no binding of ssDNA by X3CDX2 occurred because of the low amount of protein complex tested. c,d, Box plots depicting RAD51 nucleoprotein filaments per frame (c) and length of filaments in nanometers (d) for measurements obtained from negative-stain EM in the absence or presence of indicated concentrations of X3CDX2 complex. Data depict quantifications from 49 electron micrographs for each sample obtained in two independent experiments. The P values for the significance of differences in median values were calculated using a two-sided Mann–Whitney–Wilcoxon test. For box plots, each box depicts the interquartile range (IQR) containing 50% of the data, intersected by the median, and whiskers extend up to 1.5× the IQR value. e, Reaction schematic of DNA strand exchange in the presence of RPA. f,g, DNA strand exchange in the absence (f) or in the presence (g) of 30 nM RPA, with 100 nM RAD51 alone or an equimolar mixture of RAD51 and indicated RAD51 paralog complexes. Reactions were performed with AMP-PNP as the nucleotide cofactor. The positions of the Cy5‑labeled (indicated by asterisk) 40‑bp dsDNA and the strand‑exchange product are annotated alongside the gel images. Results from three (f) and five (g) independent experiments are plotted as the mean values ± s.d. h, Model for RAD51–X3C and RAD51–X3CDX2 functions. X3C targets RAD51 to the ssDNA by restraining association of RAD51 with dsDNA and RNA–DNA hybrid to promote HR and fork restart. DX2 remodels RAD51–X3C into RAD51–X3CDX2 to accelerate RAD51 filament assembly and RPA/RAD51 exchange to promote HR.
We next used negative-stain EM to independently verify the influence of X3CDX2 on RAD51 presynaptic filament assembly on a 500-nt ssDNA substrate (Extended Data Fig. 8d). Importantly, the number of RAD51 presynaptic filaments increased significantly in the presence of increasing concentrations of X3CDX2 (Fig. 6c), indicating enhanced presynaptic filament assembly. However, X3CDX2 did not appreciably alter average filament length (Fig. 6d), suggesting that its primary effect may be on early steps of filament assembly rather than filament growth.
We also asked whether X3CDX2 could overcome the well-documented inhibitory effect of the ssDNA-binding protein RPA on RAD51-mediated strand exchange (Fig. 6e). In the absence of RPA, like X3C, X3CDX2 elevated DNA strand exchange activity relative to RAD51 alone (Fig. 6f). We note that neither X3C nor X3CDX2 alone has any DNA strand exchange activity (Fig. 6f). When RPA was preincubated with the ssDNA substrate, strand exchange activity was completely abrogated when RAD51, RAD51–X3C or DX2/RAD51 were tested (Fig. 6g, lanes 2–5). However, addition of X3CDX2 to RAD51 (Fig. 6g, lane 6) or DX2 to RAD51–X3C (Extended Data Fig. 8e) led to robust restoration of DNA strand exchange activity. This HR mediator function of X3CDX2 is dependent on the DNA-binding interface in RAD51D, as a mutant X3CDX2 complex that harbors the RAD51D-R266A mutant failed to overcome RPA inhibition (Fig. 6g, lane 7). Together, these data demonstrate that the RAD51–X3CDX2 complex functionally remodels RAD51 filament assembly by enhancing ssDNA binding and helping overcome RPA inhibition.
To evaluate the role of RAD51 interaction, we also generated X3CDX2 complexes containing either the XRCC3 loop deletion or the combined loop/quad mutant. Both mutant complexes bound ssDNA at levels comparable to wild-type X3CDX2, indicating that these alterations do not disrupt overall complex integrity (Extended Data Fig. 8f). However, the RAD51-interaction-defective XRCC3 loopΔ/quad mutant failed to support synergistic ssDNA binding with RAD51 or the ability to overcome RPA-mediated inhibition of strand exchange, whereas the XRCC3 loop deletion mutant retained intermediate activity (Extended Data Fig. 8g,h). These results confirm that RAD51 interaction is essential for X3CDX2 function and the XRCC3 insertion loop may contribute to activities beyond simple RAD51–X3C octamer assembly.
Cryo-EM structure of ssDNA-bound RAD51–X3CDX2 complex
Having successfully reconstituted and characterized the apo RAD51–X3CDX2 assembly, we sought to define how this paralog complex engages DNA within an active RAD51 filament. Using cryo-EM, we determined the structure of a poly(dT) ssDNA-bound RAD51–X3CDX2 complex at a nominal resolution of 2.95 Å (Fourier shell correlation (FSC) cutoff = 0.143; model-to-map resolution 3.28 Å at FSC cutoff = 0.5), allowing clear visualization of 23 nt within the bound ssDNA substrate (Fig. 7, Table 1 and Extended Data Figs. 9 and 10).
a, Cryo-EM map of the ssDNA-bound RAD51–X3CDX2 complex assembled in the presence of AMP-PNP. Individual subunits are labeled and color-coded. Two views related by a 90° rotation about the y axis are presented. b, Cartoon representation of the ssDNA-bound RAD51–X3CDX2 complex presented as in a (right). ssDNA and AMP-PNP are shown as spheres. c, Magnified view of ssDNA from the RAD51–X3CDX2 structure presented as in b, with ssDNA shown as sticks. The 5′ and 3′ ends of the ssDNA are indicated. d, Surface electrostatic representation of the ssDNA-bound RAD51–X3CDX2 structure, with DNA presented as green spheres. e, Superposition of the ‘apo’ RAD51–X3CDX2 structure (colored cyan) onto the ssDNA-bound RAD51–X3CDX2 structure (colored as in b).
The structure reveals a winding filament architecture in which XRCC2, RAD51D, RAD51C and XRCC3 cap the 5′ end of the ssDNA and are followed by a series of RAD51 protomers forming a canonical RAD51 filament (Fig. 7a,b). Consistent with previously reported RAD51–ssDNA structures, RAD51 engages ssDNA in a repeated triplet-binding pattern, being stabilized by predominantly positively charged contacts involving residues R229, R235, S239, R241, Q242, P286, N290 and I291 within each ssDNA-contacting RAD51 protomer (Fig. 7d and Extended Data Fig. 10). However, within the DNA segment engaged by the X3CDX2 paralog module, the canonical RAD51 triplet-binding pattern is locally disrupted (Fig. 7c). The final RAD51-bound triplet is stabilized not only by RAD51 but also by contributions from XRCC3 and RAD51C, including XRCC3 T287 and the RAD51C backbone amide of E261 (Fig. 7a–c and Extended Data Fig. 10). The triplet-binding pattern is subsequently interrupted near the 5′ terminus of the ssDNA, generating a nontriplet interface in which XRCC3, RAD51C and RAD51D collectively engage the DNA (Fig. 7c). At this interface, ssDNA contacts are mediated by XRCC3 residues R220, R231, L235 and Q236, RAD51C residues R249, R258, R260 and S304 and RAD51D residues T252, R255, A263, L264, G265 and R266 (Extended Data Fig. 10). Notably, the recently determined structure of the ssDNA-bound BCDX2 RAD51 paralog complex20 similarly revealed a paralog–ssDNA interface in which the canonical RAD51 triplet-binding pattern is disrupted, suggesting that alteration of triplet-based ssDNA engagement may represent a general feature of RAD51 -paralog assemblies.
Importantly, the overall architecture of the ssDNA-bound RAD51–X3CDX2 assembly closely matches that of the DNA-free structure, indicating that DNA engagement does not entail large-scale rearrangement of the paralog module (Fig. 7e). Instead, DNA binding reveals how DX2-mediated remodeling converts the autoinhibited RAD51–X3C octamer into a filament-compatible RAD51–X3CDX2 assembly that integrates seamlessly into the RAD51 filament while locally modifying the canonical RAD51–ssDNA interaction pattern. These results show that paralog exchange generates a filament-competent assembly whose structural organization is preserved upon DNA binding while simultaneously altering local DNA engagement within the paralog module.
Discussion
Our study identifies two structurally distinct complexes of RAD51 with X3C and X3CDX2 and provides mechanistic insights into how these assemblies regulate RAD51 presynaptic filament formation during HR (Fig. 6h). We show that the RAD51–X3C octameric architecture imposes constraints on nonproductive binding of RAD51 to dsDNA and RNA–DNA hybrids (Figs. 3a,c,e and 4a), reflecting a regulatory feature that promotes selective ssDNA engagement (Fig. 6h). We further show that DX2 remodels RAD51–X3C into a pentameric RAD51–X3CDX2 complex (Fig. 5), which accelerates RAD51 filament assembly and helps to overcome RPA inhibition, highlighting a structural switch that expands the functional repertoire of the RAD51 paralogs (Fig. 6). The ssDNA-bound RAD51–X3CDX2 structure also reveals how the remodeled paralog module integrates into the RAD51 filament while locally altering canonical RAD51–ssDNA interactions at the paralog interface (Fig. 7). These findings indicate that distinct RAD51 paralog assemblies differentially influence RAD51 filament assembly on ssDNA while avoiding nonproductive association with double-stranded nucleic acids.
Two recent studies published during review and revision of our manuscript have expanded the range of RAD51 paralog assemblies and their functional roles during HR. A cryo-EM structure of RAD51–X3CDX2 bound to 5′-tailed duplex DNA revealed how this assembly associates with RAD51 filaments and can cap filament ends while contributing to filament stability and organization43. An additional study of RAD51 paralogs demonstrated that all five paralogs can assemble into higher-order complexes, including a BCDX2–CX3–RAD51 supercomplex that functions as a filament-loading platform, as well as an RAD51B-independent X3CDX2 complex common to all three studies that can anchor and cap RAD51 filaments44. Together, these studies highlight the modular nature of paralog assemblies and their ability to support distinct stages of RAD51 filament dynamics.
In this context, our findings provide complementary insight by defining RAD51-bound paralog assemblies and identifying a remodeling transition between RAD51–X3C and RAD51–X3CDX2 that regulates filament-compatible states. While the RAD51–X3C octamer observed here likely represents a regulated or transient assembly, additional cellular studies will be required to establish its physiological context and relevance. Nevertheless, the structural and biochemical data presented here support a model in which paralog exchange enables dynamic reconfiguration of RAD51 paralog complexes, providing a mechanism to coordinate RAD51 filament assembly, stabilization and function across different stages of HR.
Consistent with these mechanistic insights, genetic studies show that, while deletion of RAD51C, RAD51D, XRCC2 or XRCC3 leads to cellular lethality12,13, loss of RAD51B is tolerated13. This differential requirement of RAD51 paralogs for cell viability maintenance suggests that the X3CDX2 complex can compensate for BCDX2 functions in promoting RAD51-mediated HR and replication fork stability. Consistent with this, pathogenic mutations in RAD51 paralogs, especially in RAD51C, RAD51D, XRCC2 and XRCC3, have been strongly linked to breast and ovarian cancers15,16,45. The biochemical and structural analyses established in this study provide a platform to assess how disease-associated mutations disrupt RAD51 paralog assembly, remodeling, and RAD51–ssDNA filament dynamics. By defining how distinct paralog assemblies tune RAD51 filament formation, our work advances the mechanistic understanding of genome instability syndromes and may inform future strategies targeting HR as a cancer therapeutic approach.
Lastly, building on the remodeling and assembly transitions described above, our study suggests a unified framework in which dynamic paralog exchange-driven remodeling serves as a general regulatory principle that determines when, where and which RAD51 paralog complexes form, enabling distinct assemblies to execute specialized functions. We view this regulatory logic as an important conceptual advance that extends current models of RAD51 paralog biology.
Methods
Protein expression and purification
Purifications of RAD51 using 6×His–SUMO–RAD51 construct, untagged RPA complex and DX2 (RAD51D and XRCC2–FLAG) subcomplex were carried out as previously described20. For purification of the X3C and RAD51–X3C complexes, Hi5 insect cells were infected for 48 h with baculoviruses expressing RAD51C and XRCC3–FLAG without or with RAD51 baculovirus, respectively. All purification steps were carried out at 0–4 °C. The crude cell lysate was prepared from 6–8 g of cell pellet (from 800 ml of insect cell culture) by sonication in 50 ml of T buffer (25 mM Tris-HCl pH 7.5, 10% glycerol, 0.5 mM EDTA, 1 mM DTT, 0.05% IGEPAL, 1 mM PMSF and protease inhibitors) containing 300 mM KCl, 5 mM ATP and 2 mM MgCl2. Lysates were clarified by centrifugation at 100,000g for 60 min. The supernatant was incubated with 1 ml of FLAG affinity resin (Pierce, A36804) for 2 h. The resin was sequentially washed with 100 ml of T buffer containing 300 mM KCl and then 50 ml of T buffer containing 100 mM KCl, with all wash buffers containing 2 mM ATP and 2 mM MgCl2. Bound proteins were eluted with 10 ml of T buffer containing 200 ng μl−1 FLAG peptide (Sigma, F3290), 100 mM KCl, 2 mM ATP and 2 mM MgCl2. Complexes were further fractionated by ion-exchange chromatography using a 1-ml HiTrap Q HP column with 30 ml of a 100–500 mM KCl gradient in T buffer containing 2 mM each of ATP and MgCl2. Peak fractions were pooled and concentrated to 0.4 ml in an Amicon 30 concentrator and subjected to size-exclusion chromatography in a Superdex 200 Increase 10/300 GL column in H100-ATP buffer (20 mM HEPES–KOH pH 7.5, 0.5 mM EDTA, 1 mM DTT, 2 mM ATP and 2 mM MgCl2). Peak fractions containing X3C and RAD51–X3C complexes in respective purifications were concentrated, snap-frozen in liquid nitrogen and stored at −80 °C.
For reconstitution of the RAD51–X3C complex, 250 μg each of purified X3C complex and RAD51 were mixed and incubated for 30 min on ice. Similarly, X3CDX2 complex was reconstituted by mixing 250 μg each of purified X3C and DX2 complexes. RAD51–X3CDX2 was reconstituted by mixing 250 μg of RAD51–X3C with 150 μg of DX2 complex. The mixed samples were subjected to the size-exclusion chromatography in a Superdex 200 Increase 10/300 GL column in H100-ATP buffer. Peak fractions containing desired complexes were concentrated, snap-frozen in liquid nitrogen and stored at −80 °C. For cryo-EM analysis, concentrated samples were incubated with otherwise similar HEPES buffer, but with ATP replaced by 1 mM AMP-PNP for 1 h; AMP-PNP was further enriched by buffer exchange during concentration.
Cryo-EM sample preparation and data acquisition
Freshly eluted RAD51–X3C, RAD51–X3CDX2 and X3CDX2 complexes from a gel filtration column (Superdex 200 Increase 10/300 GL) were buffer-exchanged to replace ATP with AMP-PNP and concentrated to around 6 µM. The samples were immediately vitrified using a Vitrobot Mark IV (Thermo Fisher Scientific). Then, 3 µl of sample was applied onto cryo-EM Au grids (UltrAuFoil 1.2/1.3 300-mesh, EM Sciences) that were glow-discharged at 20 mA for 30 s in a Quorum EMS glow discharger. Grids were subsequently blotted for 3 s with a blotting force of −10 and plunged into the liquid ethane. A temperature of 4 °C and 100% humidity were maintained in the Vitrobot chamber. For the ssDNA–RAD51–X3CDX2 complex, about 20 μM RAD51 and 5 μM apo X3CDX2 were mixed with 60 μM nucleotides of a 50-base poly(dT) oligonucleotide (5′-TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT-3′) in the presence of 2 mM AMP-PNP. The mixture was incubated for 30 min at 37 °C and then applied onto UltrAuFoil cryo-EM Au grids as for apo RAD51–X3C, RAD51–X3CDX2 and X3CDX2 complexes. The grids were glow-discharged for 60 s at 15 mA on each side of the grid using a Pelco easiGlow glow discharger before sample application. Frozen grids were prepared using a Leica EM GP2 plunge-freezer with a 3-s blot time at 4 °C and 95% humidity. All grids were transferred into liquid nitrogen and stored at cryogenic temperatures before clipping, screening and data collection.
Data for RAD51–X3C, RAD51–X3CDX2 and X3CDX2 complexes were collected at 200 kV on a Glacios cryo-TEM (transmission EM) instrument equipped with a Falcon IV camera and Selectris energy filter (slit width: 10 eV) at the University of Texas Health San Antonio (UTHSA) Cryo-EM center. Imaging was acquired using Thermo Fisher Scientific’s EPU software with an AFIS model. Image frames in the EER file format were recorded with a total dose of 50 e− per Å2 with a pixel size of 0.87 Å, a nominal magnification ×130,000 and a defocus range from −0.8 to −2.0 µm. Data for the ssDNA-bound RAD51–X3CDX2 complex were collected at 300 kV on a Krios equipped with a TEM Gatan BioContinuum imaging filter and K3 direct electron detector. Movies in TIF format were recorded with a total dose of 73 e− per Å2 with a pixel size of 0.8332 Å, a nominal magnification ×105,000 and a defocus range from −1.2 to −2.5 µm.
Cryo-EM data processing
After data collection, images were gain-corrected, dose-weighted and patch-motion-corrected using cryoSPARC (version 4.4.1)46. The CTF (contrast transfer function) of each micrograph was estimated using cryoSPARC ’s patch CTF estimation. After CTF estimation, all micrographs were curated using a CTF resolution cutoff value of 6 Å. This yielded curated sets of 12,106 micrographs for the RAD51–X3C octamer, 16,017 micrographs for RAD51–X3CDX2, 14,500 micrographs for X3CDX2 and 15,183 micrographs for the ssDNA-bound RAD51 filament with X3CDX2. Processing steps for each complex are provided in Extended Data Figs. 1, 5 and 9 and Supplementary Fig. 2. Briefly, the blob picker as implemented in cryoSPARC was used to pick particles between 80 Å and 150 Å extracted using box sizes of 320 pixels for RAD51–X3C octamer, 384 pixels for RAD51–X3CDX2, 256 pixels for X3CDX2 and 384 pixels for the ssDNA-bound RAD51 filament with X3CDX2. Box sizes were chosen to account for the relative dimensions of each complex and particle density on the grid. Serial two-dimensional classification was conducted for each complex to remove noise and dissociated particles before ab initio models created. The number of ab initio models generated for each particle stack was chosen by expected sample heterogeneity, with a higher number of models chosen for samples with greater heterogeneity. Ab initio model generation was followed by heterogeneous refinement with inputs ranging from all particles to best-class particles and all volumes to 2× the same volume chosen specifically to populate the best class and its volume from noise, low-resolution particles and heterogeneous particles. The best class of each was followed up by homogeneous refinement, nonuniform refinement and local refinement with the output of each used for the next step with the exception of the ssDNA-bound RAD51 filament with X3CDX2, where homogeneous refinement was followed by two rounds of local refinement. The final nominal resolution (FSC cutoff = 0.143), as reported by cryoSPARC, was 3.25 Å for RAD51–X3C octamer, 3.51 Å for RAD51–X3CDX2, 3.16 Å for X3CDX2 and 2.95 Å for the ssDNA-bound RAD51 filament with X3CDX2.
Model building and refinement
Models of RAD51, RAD51C, RAD51D, XRCC2 and XRCC3 were obtained from either AlphaFold2 (ref. 47) predictions or PDB 8FAZ, 8GBJ or 8GGA and rigid-body docked into each cryo-EM map using UCSF Chimera before manually deleting low-confidence and disordered regions. Rigid-body-docked models were brought through iterative rounds of manual building in Coot and real-space refinement in PHENIX. The structural model was validated using wwPDB (https://validate-rcsb-2.wwpdb.org/). Figures for the cryo-EM map and models were prepared with ChimeraX48 and PyMOL (Schrödinger). Final model-to-map resolutions (FSC cutoff = 0.5) achieved were 3.39 Å for RAD51–X3C octamer, 3.80 Å for RAD51–X3CDX2, 3.29 Å for X3CDX2 and 3.28 Å for the ssDNA-bound RAD51 filament with X3CDX2.
Affinity pulldown
All pulldown assays were carried out in T buffer supplemented with 100 mM KCl, 2 mM each ATP and MgCl2 and TurboNuclease. The indicated X3C complexes and RAD51 were incubated together for 30 min at 4 °C, followed by the addition of FLAG affinity resin (Pierce, A36804) and an additional 30-min incubation at 4 °C with gentle mixing. The resin was then separated from the flowthrough, washed twice with 50 µl of T buffer and retained for analysis. Both the flowthrough and resin-bound fractions, diluted in SDS–PAGE loading buffer, were resolved on 4–15% Mini-PROTEAN TGX precast gels (Bio-Rad), stained with Coomassie blue and imaged using a ChemiDoc Imaging System.
DNA binding
For DNA binding, 2 nM of 5′ Cy5-labeled 80-nt ssDNA (Cy5–oligo A; Supplementary Table 1), 80 bp of dsDNA (Cy5–oligo A + oligo B; Supplementary Table 1)9 or 60 bp of RNA–DNA hybrid dsDNA (Cy5–RNA–oligo E + oligo F; Supplementary Table 1) was incubated with the indicated concentration of purified RAD51, RAD51–X3C or indicated complexes in 10 μl of reaction buffer (50 mM Tris-HCl pH 7.5, 120 mM KCl, 1 mM DTT, 1 mM ATP or AMP-PNP, 1 mM MgCl2 and 100 ng μl−1 BSA) for 10 min at 37 °C.
In the DNA-binding assay with prior dsDNA challenge, the desired protein or protein complexes were first incubated with eightfold molar excess of phiX174 RFI circular dsDNA in the presence of AMP-PNP at room temperature for 10 min. Thereafter, 2 nM of 5′ Cy5-labeled 80-nt ssDNA (Cy5–oligo A) was added to the reaction and incubated for 10 min at 37 °C. Nucleoprotein complexes were resolved on the 5% polyacrylamide gels in Tris–borate buffer (45 mM each), gels were visualized using the ChemiDoc imaging system (Bio-Rad) and the proportion of bound versus unbound DNA was quantified using the Fiji ImageJ software (version 2.90/1.53t). The mean values were calculated from the data obtained from at least three independent experiments and plotted using the R (version 3.6.1) package ggpubr. Error bars indicate the s.d.
Negative-stain EM
For presynaptic filament assembly, RAD51 (1 μM) was incubated with 500 nt of ssDNA (3 μM nucleotides) (Supplementary Table 1) without or with indicated concentrations of the X3CDX2 complex in reaction buffer (25 mM HEPES pH 7.5, 25 mM KCl, 1 mM MgCl2 and 1 mM AMP-PNP) at 37 °C for 10 min. A droplet of 4 μl of a reaction mixture was blotted onto a glow-discharged carbon-coated 300-mesh copper grid (EMS CF300 cu) for 1 min. After blotting with filter paper, the grid was immersed in a 20-μl droplet of the staining solution (2% uranyl acetate). The stain was removed from the grid by blotting with filter paper and the grid was washed twice with water for 1 min. Following excess stain removal, grids were dried in air for about 3 min. Negatively stained samples of hRAD51 presynaptic filaments were examined by TEM. Images were processed using the Fiji ImageJ (version 2.90/1.53t) image analysis package and filaments were quantified using the ImageJ plugin Ridge detection. Data were plotted using the R (version 3.6.1) package ggpubr.
Homologous DNA pairing and strand exchange assay
In the standard strand exchange reaction (12.5 μl), indicated concentrations of RAD51, RAD51–X3C complex or RAD51 and indicated paralog combination were incubated with 80 nt of ssDNA (oligo A; 8 nM) in strand exchange buffer (35 mM Tris pH 7.5, 1 mM DTT, 0.1 μg μl−1 BSA and 1 mM MgCl2) containing 2 mM ATP or AMP-PNP at 37 °C for 10 min. The reaction was initiated by adding homologous 40 bp of 5′ Cy5-labeled dsDNA (Cy5–oligo C hybridized to oligo D, 8 nM; Supplementary Table 1) and 4 mM spermidine hydrochloride (pH 7.5) and incubated for 10 min at 37 °C. The reaction was stopped by adding stop mix to final concentration 0.1% SDS and 0.5 mg ml−1 proteinase K and 1× loading buffer (7% glycerol, 10 mM Tris pH 8.0, 1 mM EDTA and orange G) for 10 min at 37 °C. The samples were resolved in 8% native polyacrylamide gels in TGS buffer (160 mM Tris pH 8.3, 1.2 M Glycine and 0.22 mM SDS) at room temperature. Gels were visualized using the ChemiDoc imaging system (Bio-Rad) and quantified using the Fiji ImageJ software (version 2.90/1.53t) or ImageLab (version 6.1.0; Bio-Rad). To examine RPA/RAD51 exchange activity, the ssDNA was premixed with RPA (30 nM) in strand exchange reaction before the addition of RAD51 (100 nM) and the indicated RAD51 paralog combination. Incubation and analysis of the reaction mixtures were carried out as described above. Mean values were calculated from the data obtained from three to five independent experiments. Data were processed and plotted using GraphPad Prism8.4 or the R (version 3.6.1) package ggpubr. The s.d. or error was calculated and presented as error bars together with the mean values.
DNA fiber assay
Plasmids expressing either wild-type or mutant XRCC3 (codon-optimized for small interfering RNA (siRNA) resistance) were transfected into HeLa cells using Fugene HD transfection reagent (Promega) and Lipofectamine RNAiMAX (Invitrogen) following the manufacturer’s protocols. For siRNA knockdown, control or XRCC3-targeting siRNAs (Supplementary Table 1) were used in parallel. To assess fiber restart capability, cells were initially labeled with 100 μM CldU (Sigma-Aldrich, C6891) for 30 min, followed by treatment with 2.5 mM hydroxyurea for 3.5 h to stall the replication fork. Afterward, cells were labeled with 50 μM IdU (Sigma-Aldrich, I7125) for an additional 30 min to track replication restart. The fiber assay was conducted as outlined in the schematic (Extended Data Fig. 3d). After labeling, cells were collected and resuspended in trypsin–EDTA (Sigma-Aldrich, T3924) at a concentration of 2.5 × 105 cells per agarose plug. Agarose plug preparation, proteinase K treatment, washing, β-agarase treatment of the melted plug and fiber combing were carried out according to the protocols outlined in STAR Protocols49 and the Genomic Vision Manual. Immunodetection of the labeled DNA fibers was performed using the following primary antibodies: rat anti-BrdU (BU1/75 (ICR1); Abcam, ab6326) at a 1:40 dilution in blocking buffer (Invitrogen, B10710) for CldU detection and mouse anti-BrdU (clone B44; BD Biosciences, 347580) at a 1:10 dilution in blocking buffer for IdU detection. Secondary antibodies used were goat anti-rat 488 (Invitrogen, A11006) and donkey anti-mouse Cy3 (Invitrogen, A31570) at a 1:200 dilution. Fibers were visualized and imaged using the Olympus FV3000 confocal microscope and ImageJ software was used for subsequent quantification.
HR SCR GFP assay
U2OS CRISPR–Cas9 XRCC3-knockout (KO) cells were seeded in six-well plates at 200,000 cells per well and incubated overnight at 37 °C. U2OS CRISPR–Cas9 XRCC3-KO cells were then transfected with 1 µg of pC1-XRCC3 plasmids that express either WT XRCC3 or the indicated mutant (loop, quad and loop/quad), together with 1 µg of plasmid expressing I-SceI restriction enzyme. Then, 72 h after transfection, cells were harvested and fixed with 2% PFA. GFP-expressing cells were quantified using a BD Accuri C6 Plus personal flow cytometer at 10,000 events acquired for each condition. The experiment was performed in triplicate.
PLA
U2OS cells were cultured in DMEM supplemented with 10% FBS (Sigma), 100 μg ml−1 streptomycin and 100 U per ml penicillin (Gibco, 15140-122). The cells were tested for Mycoplasma contamination by a Mycoplasma PCR detection kit (Abcam, ab289834) following the manufacturer’s instructions. The U2OS cells were cotransfected with plasmids expressing GFP–XRCC3 and one of HA–RAD51B, HA–RAD51C, HA–RAD51D or HA–XRCC2 using PEI transfection reagent (Bio-Techne, 7854) and incubated for 48 h before analysis. The transfected cells were used for PLA and immunoblotting.
To analyze the colocalization of GFP–XRCC3 and HA–RAD51B, HA–RAD51C, HA–RAD51D or HA–XRCC2, PLA was carried out as per the manufacturer’s protocol (Duolink in situ starter kits; DUO92101, Sigma-Aldrich). U2OS cells were cultured in glass-bottom dishes (Cellvis, D35-14-1.5-N) and cells were transfected with the indicated plasmids for 48 h or treated with γ-irradiation (6 Gy, 2 h). To analyze colocalization of GFP or HA (Fig. 5a), cells were washed with PBS and fixed with 1% formaldehyde in PBS for 10 min, permeabilized in 0.5% Triton X-100 in PBS for 15 min at room temperature and then kept in PLA Duolink blocking buffer (0.1% Triton X-100 and 1% BSA in PBS) for 1 h at 37 °C. Next, to detect colocalization of signals, the cells were incubated with anti-GFP (Santa Cruz, sc-9996) or anti-HA (Cell Signaling Technologies, 3724S) in Duolink antibody dilution buffer at 4 °C overnight. Next, in situ PLA probes (anti-mouse plus and anti-rabbit minus) were diluted 1:5 in Duolink antibody diluent and incubated for 1 h at 37 °C. After washing with buffer A (10 mM Tris pH 7.4, 150 mM NaCl and 0.05% Tween-20) three times for 5 min each, dishes were incubated with Duolink ligase at 37 °C for 30 min and then washed with buffer A twice for 5 min each. Amplification mix prepared as per the manufacturer’s instructions was added into each dish and incubated at 37 °C for 100 min in the dark. Lastly, dishes were washed with buffer B (200 mM Tris pH 7.5 and 100 mM NaCl) three times for 5 min each and 0.1× buffer B once for 5 min. Duolink in situ mounting solution with DAPI was used to mount glass-bottom dishes for nuclear staining and applied to an Olympus FV3000 confocal microscope. The focus formation was determined using ImageJ software (version 1.53a; National Institutes of Health (NIH)).
Statistical analysis was conducted using the GraphPad PRISM software. An unpaired Student’s t-test was performed for comparison of PLA data. Data are presented as the means ± s.e.m. A P value < 0.05 was considered statistically significant.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Atomic coordinates for the RAD51–X3C, RAD51–X3CDX2, ssDNA-bound RAD51–X3CDX2 and X3CDX2 complexes were deposited to the PDB with accession codes 9OMY, 9OMZ, 9ZZR and 9ON2, respectively. The corresponding cryo-EM maps were deposited to the EM Data Bank with accession codes EMD-70624, EMD-70625, EMD-75014 and EMD-70627. Source data are provided with this paper.
References
Tarsounas, M. & Sung, P. The antitumorigenic roles of BRCA1–BARD1 in DNA repair and replication. Nat. Rev. Mol. Cell Biol. 21, 284–299 (2020).
Scully, R., Panday, A., Elango, R. & Willis, N. A. DNA double-strand break repair-pathway choice in somatic mammalian cells. Nat. Rev. Mol. Cell Biol. 20, 698–714 (2019).
Sigurdsson, S., Trujillo, K., Song, B., Stratton, S. & Sung, P. Basis for avid homologous DNA strand exchange by human Rad51 and RPA. J. Biol. Chem. 276, 8798–8806 (2001).
Ceccaldi, R. & Cejka, P. Mechanisms and regulation of DNA end resection in the maintenance of genome stability. Nat. Rev. Mol. Cell Biol. 26, 586–599 (2025).
Zhao, W. et al. BRCA1–BARD1 promotes RAD51-mediated homologous DNA pairing. Nature 550, 360–365 (2017).
Liang, F. et al. Promotion of RAD51-mediated homologous DNA pairing by the RAD51AP1–UAF1 complex. Cell Rep. 15, 2118–2126 (2016).
Prakash, R., Zhang, Y., Feng, W. & Jasin, M. Homologous recombination and human health: the roles of BRCA1, BRCA2, and associated proteins. Cold Spring Harb. Perspect. Biol. 7, a016600 (2015).
Sung, P. Yeast Rad55 and Rad57 proteins form a heterodimer that functions with replication protein A to promote DNA strand exchange by Rad51 recombinase. Genes Dev. 11, 1111–1121 (1997).
Gaines, W. A. et al. Promotion of presynaptic filament assembly by the ensemble of S. cerevisiae Rad51 paralogues with Rad52. Nat. Commun. 6, 7834 (2015).
Zhao, W. et al. Promotion of BRCA2-dependent homologous recombination by DSS1 via RPA targeting and DNA mimicry. Mol. Cell 59, 176–187 (2015).
Jensen, R. B., Carreira, A. & Kowalczykowski, S. C. Purified human BRCA2 stimulates RAD51-mediated recombination. Nature 467, 678–683 (2010).
Sullivan, M. R. & Bernstein, K. A. RAD-ical new insights into RAD51 regulation. Genes (Basel) 9, 629 (2018).
Garcin, E. B. et al. Differential requirements for the RAD51 paralogs in genome repair and maintenance in human cells. PLoS Genet. 15, e1008355 (2019).
Prakash, R. et al. Homologous recombination-deficient mutation cluster in tumor suppressor. Proc. Natl Acad. Sci. USA 119, e2202727119 (2022).
Meindl, A. et al. Germline mutations in breast and ovarian cancer pedigrees establish RAD51C as a human cancer susceptibility gene. Nat. Genet. 42, 410–414 (2010).
Kondrashova, O. et al. Secondary somatic mutations restoring RAD51C and RAD51D associated with acquired resistance to the PARP inhibitor rucaparib in high-grade ovarian carcinoma. Cancer Discov. 7, 984–998 (2017).
Jacquinet, A. et al. Expanding the FANCO/RAD51C associated phenotype: cleft lip and palate and lobar holoprosencephaly, two rare findings in Fanconi anemia. Eur. J. Med. Genet. 61, 257–261 (2018).
Shamseldin, H. E., Elfaki, M. & Alkuraya, F. S. Exome sequencing reveals a novel Fanconi group defined by XRCC2 mutation. J. Med. Genet. 49, 184–186 (2012).
Masson, J. Y. et al. Identification and purification of two distinct complexes containing the five RAD51 paralogs. Genes Dev. 15, 3296–3307 (2001).
Rawal, Y. et al. Structural insights into BCDX2 complex function in homologous recombination. Nature 619, 640–649 (2023).
Greenhough, L. A. et al. Structure and function of the RAD51B–RAD51C–RAD51D–XRCC2 tumour suppressor. Nature 619, 650–657 (2023).
Sigurdsson, S. et al. Mediator function of the human Rad51B–Rad51C complex in Rad51/RPA-catalyzed DNA strand exchange. Genes Dev. 15, 3308–3318 (2001).
Longo, M. A. et al. RAD51C–XRCC3 structure and cancer patient mutations define DNA replication roles. Nat. Commun. 14, 4445 (2023).
Schild, D., Lio, Y. C., Collins, D. W., Tsomondo, T. & Chen, D. J. Evidence for simultaneous protein interactions between human Rad51 paralogs. J. Biol. Chem. 275, 16443–16449 (2000).
Xu, J. et al. Cryo-EM structures of human RAD51 recombinase filaments during catalysis of DNA-strand exchange. Nat. Struct. Mol. Biol. 24, 40–46 (2017).
Pellegrini, L. et al. Insights into DNA recombination from the structure of a RAD51–BRCA2 complex. Nature 420, 287–293 (2002).
Subramanyam, S., Jones, W. T., Spies, M. & Spies, M. A. Contributions of the RAD51 N-terminal domain to BRCA2–RAD51 interaction. Nucleic Acids Res. 41, 9020–9032 (2013).
Olvera-Leon, R. et al. High-resolution functional mapping of RAD51C by saturation genome editing. Cell 187, 5719–5734 (2024).
Berti, M. et al. Sequential role of RAD51 paralog complexes in replication fork remodeling and restart. Nat. Commun. 11, 3531 (2020).
Bhattacharya, D. et al. RAD51 paralogs: expanding roles in replication stress responses and repair. Curr. Opin. Pharmacol. 67, 102313 (2022).
Pierce, A. J., Johnson, R. D., Thompson, L. H. & Jasin, M. XRCC3 promotes homology-directed repair of DNA damage in mammalian cells. Genes Dev. 13, 2633–2638 (1999).
Sung, P. & Robberson, D. L. DNA strand exchange mediated by a RAD51–ssDNA nucleoprotein filament with polarity opposite to that of RecA. Cell 82, 453–461 (1995).
Masson, J. Y., Stasiak, A. Z., Stasiak, A., Benson, F. E. & West, S. C. Complex formation by the human RAD51C and XRCC3 recombination repair proteins. Proc. Natl Acad. Sci. USA 98, 8440–8446 (2001).
Shah, P. P. et al. Swi2/Snf2-related translocases prevent accumulation of toxic Rad51 complexes during mitotic growth. Mol. Cell 39, 862–872 (2010).
Kiianitsa, K., Solinger, J. A. & Heyer, W. D. Terminal association of Rad54 protein with the Rad51–dsDNA filament. Proc. Natl Acad. Sci. USA 103, 9767–9772 (2006).
Saur, F. et al. Transcriptional repression facilitates RNA:DNA hybrid accumulation at DNA double-strand breaks. Nat. Cell Biol. 27, 992–1005 (2025).
Gomez-Gonzalez, B. & Aguilera, A. Break-induced RNA–DNA hybrids (BIRDHs) in homologous recombination: friend or foe? EMBO Rep. 24, e57801 (2023).
Mirman, Z., Sasi, N. K., King, A., Chapman, J. R. & de Lange, T. 53BP1-shieldin-dependent DSB processing in BRCA1-deficient cells requires CST–Polα–primase fill-in synthesis. Nat. Cell Biol. 24, 51–61 (2022).
Cohen, S. et al. Senataxin resolves RNA:DNA hybrids forming at DNA double-strand breaks to prevent translocations. Nat. Commun. 9, 533 (2018).
Parajuli, S. et al. Human ribonuclease H1 resolves R-loops and thereby enables progression of the DNA replication fork. J. Biol. Chem. 292, 15216–15224 (2017).
Yang, X. et al. RNA–DNA hybrids regulate meiotic recombination. Cell Rep. 37, 110097 (2021).
Garcia-Muse, T. & Aguilera, A. R loops: from physiological to pathological roles. Cell 179, 604–618 (2019).
Greenhough, L. A., Galanti, L., Liang, C. C., Boulton, S. J. & West, S. C. Cryo-electron microscopic visualization of RAD51 filament assembly and end-capping by XRCC3–RAD51C–RAD51D–XRCC2. Science 391, eaea1546 (2026).
Koo, C. W. et al. BCDX2–CX3 and DX2–CX3 complexes assemble and stabilize RAD51 filaments. Nature https://doi.org/10.1038/s41586-026-10314-z (2026).
Park, D. J. et al. Rare mutations in XRCC2 increase the risk of breast cancer. Am. J. Hum. Genet. 90, 734–739 (2012).
Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 (2017).
Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).
Pettersen, E. F. et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci. 30, 70–82 (2021).
Moore, G., Jimenez Sainz, J. & Jensen, R. B. DNA fiber combing protocol using in-house reagents and coverslips to analyze replication fork dynamics in mammalian cells. STAR Protoc. 3, 101371 (2022).
Acknowledgements
We thank members of the E.C.G., P.S., E.V.W. and S.K.O. laboratories for discussions. We thank B. Hunter at the UTHSA EM Laboratory of South Texas Reference Laboratories, Department of Pathology and Laboratory Medicine, for assistance with negative-stain EM. Research reported in this publication was supported by NIH grants R01 CA293655 (S.K.O.), NIH grants R01 CA168635, R01 ES007061, P01 CA92584 and R35 CA241801 (P.S.), NIH grant R00 GM140264 (E.V.W.), NIH grants R01 CA236606 and R35 GM118026 (E.C.G.), NIH grants R01 CA246807 and R01 CA258381 (S.B.), NIH grant R01 GM140127 (D.S.L.), NIH grants R01 GM136717, R01 CA237286 and R01 CA188347 and congressionally directed medical research programs BC191160 (A.V.M.), NIH grant R50 CA265315 (Y.K.), NIH grants R01 ES030335 and ES031796, American Cancer Society DBG-24-1243614-01-DMC and Department of Defense BC201356 (K.A.B.), NIH grants R01 CA239227 and R01 CA277498 (M.K.R.), NIH grant R01 CA139429 and Cancer Prevention and Research Institute of Texas (CPRIT) RP220269 (R.H.) and NIH grant PO1 CA275717 (S.B., P.S., Y.K., A.V.M., R.H., E.C.G. and D.S.L.). P.S. is the holder of the Robert A. Welch Distinguished Chair in Chemistry (AQ-0012) and the recipient of a CPRIT Recruitment of Established Investigators (REI) award (RR180029). A.V.M. is the holder of the Joe R. and Teresa Lozano Long Chair in Cancer Research and is a recipient of a CPRIT REI Award (RR210023). S.K.O. is the recipient of a CPRIT Rising Star award (RR200030) and holds the Cancer Therapy and Research Center 40th Anniversary Endowed Distinguished Professorship in Oncology. E.V.W. is the recipient of a CPRIT Recruitment of First Time Tenure Track Faculty Award (RR220068). The following grants also supported this work: American Cancer Society Postdoctoral Fellowship PF-22-034-01-DMC (C.M.R.), NIH T32CA279363 (S.S.), Paul and Cleo Schimmel Fellowship (F.Z.) and F30CA278370 and T32GM145432 (A.M.J.). The cryo-EM data in this study were collected at the UTHSA Cryo-EM Facility on a Glacios TEM equipped with a Falcon IV camera and Selectris energy filter purchased with the support of UT Science and Technology Acquisition and Retention awards 402-1288 (P.S.) and 402-1317 (S.K.O.). The high-performance computing cluster used for cryo-EM data processing was funded by the NIH Office of Research Infrastructure Programs Shared Instrument Grant S10OD036251. The surface plasmon resonance assay was conducted in the Center for Innovative Drug Discovery and the Mays Cancer Center Drug Discovery and Structural Biology Shared Resource supported by CPRIT Core Facility Award RP160844 (D.Z.) and NIH National Cancer Institute P30 CA054174 (S.K.O. and D.Z.), respectively. The content of this study is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
Author information
Authors and Affiliations
Contributions
Protein purification was conducted by Y.R., S.S. and M.T. Structural experiments and analysis were conducted by L.J., E.A.R., S.S.-M., E.V.W. and S.K.O. Y.R., M.T. and Y.K. conducted biochemical and negative-stain EM assays. J.-H.J., M.A.J. and Q.F. conducted cell-based experiments. S.Z., K.A.B., F.Z., M.J., Y.K., R.H., A.B.T., A.M.J., J.N.K., S.B., A.M.J., S.K., L.G., A.V.M., W.Z., E.C.G., C.M.R., D.S.L., H.K., L.S., M.K.R., E.D. and D.Z. assisted with experimental design and data interpretation. D.N. and C.M.S. assisted with structure analysis and figure preparation. The figures and paper were prepared by Y.R., Y.K., E.A.R., E.V.W., P.S. and S.K.O., with input from all authors.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Structural & Molecular Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Dimitris Typas, in collaboration with the Nature Structural & Molecular Biology team.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Cryo-EM data processing flowchart and map quality for the RAD51-X3C structure.
a, Coomassie-stained SDS–PAGE gel of the indicated fractions from size-exclusion chromatography of RAD51-X3C complex co-expressed in insect cells and purified in the presence of ATP/Mg2+. Data represents two independent experiments b, Auto-picked particles were first sorted by three rounds of 2D classification, six initial models were made first then followed by four rounds of heterogenous refinement. A final step of stringent heterogenous refinement and tight mask homogeneous refinement led to a final higher resolution map. All data analysis were performed using Cryosparc v4.4.1. c, Final Euler angle distribution of all particles used in refinement. d, Local resolution map, color coded from lower resolution (red) to higher resolution (blue). e, Detailed views of representative local cryo-EM density from the RAD51-X3C complex reconstruction. f, Cryo-EM map of the RAD51-X3C complex with two views related by a 180 degree rotation about the y axis. Maps of the individual subunits are shown to the right.
Extended Data Fig. 2 Subunit-subunit interaction networks within the RAD51-X3C complex.
a, Cartoon representation of the indicated RAD51.A1/RAD51.A2 interface with residues involved in intermolecular interactions shown as sticks. b-c, The indicated complexes are presented as in panel a with RAD51.A2/XRCC3 (panel b), XRCC3/RAD51C (panel c). d-e, Analysis of ATPase/ATPase domain interaction networks at the RAD51.A2/XRCC3 (panel d) and XRCC3/RAD51C (panel e) interfaces. AMP-PNP molecules are shown as spheres.
Extended Data Fig. 3 Biochemical and cell-based validation of X3C and RAD51 interaction.
a-b, Pull-down analysis of (a) WT or (b) RAD51 F86E/A89E mutant with either WT or indicated X3C (XRCC3-FLAG/ RAD51C) mutants using anti-FLAG resin. Results represent three independent experiments. c, Coomassie-stained SDS–PAGE gel of the indicated fractions from size-exclusion chromatography of RAD51-X3C complexes carrying either WT or loopΔ mutant of XRCC3. Results represent two independent experiments. d, Schematic of replication fork restart assay with CldU and IdU labelling is shown, along with representative images of fiber events from HeLa cells with or without XRCC3 knockdown and expressing either WT or indicated XRCC3 mutants. e, Box plot depicting the relative percentage of stalled fibers counted per micrograph for the indicated conditions. The P values for the significance of differences in median values were calculated by the unpaired two-sided Mann-Whitney-Wilcoxon test. For box plots, each box depicts the interquartile range (IQR) containing 50% of the data, intersected by the median, and whiskers extend up to 1.5 IQR value. n is total number of fibers analyzed in each condition. f, The ratio of the IdU to CldU tract length was measured and plotted for the indicated conditions. Statistical significance was determined using an unpaired two-tailed t-test. n is total number of restarted fibers analyzed in each condition. N.S. is not significant. g, SCR assay was performed in U2OS XRCC3 knockout cells expressing either WT or indicated XRCC3 mutants. The bar graph shows the percentage of GFP-positive cells, normalized to WT for indicated XRCC3 mutants. Significance was determined by unpaired two tailed student t-test from three independent experiments. Error bars represent standard deviation. h, Western blot analysis of HeLa cells with or without XRCC3 knockdown and expressing either WT or indicated XRCC3 mutants, Tubulin was used as a loading control. i, Western blot analysis of U2OS XRCC3 knockout cells expressing either WT or indicated XRCC3 mutants. Vinculin was used as a loading control. Western analysis in panels h and i were repeated twice.
Extended Data Fig. 4 Restriction of RAD51-dsDNA association by X3C.
a, 80 nt ssDNA and 80 bp dsDNA binding of X3C complex at indicated concentrations in the presence of AMP-PNP was assessed by electrophoretic mobility shift assays. Results from three independent experiments were plotted as mean values ± s.d. b, 80 nt ssDNA binding and c, 80 bp dsDNA binding of RAD51 and RAD51-X3C complex at indicated RAD51 concentrations in presence of ATP assessed by electrophoretic mobility shift assays. Results from three independent experiments were plotted as mean values ± s.d. d, Averaged and fitted (v=Vmax*[S]/{Kd + [S]}) line plots for (i) 80 nt ssDNA binding, (ii) 80 bp dsDNA and (iii) 60 bp RNA:DNA hybrids binding of RAD51 and RAD51-X3C complex at indicated RAD51 concentrations in the presence of AMP-PNP assessed by electrophoretic mobility shift assays. Results from three independent experiments were plotted as mean values ± s.d. e, ssDNA binding and f, dsDNA binding of RAD51 in the presence of increasing concentrations of X3C and 1 mM ATP. Results from three independent experiments were plotted as mean values ± s.d. g, ssDNA binding of RAD51 in the presence of increasing concentrations of X3C and 1 mM AMP-PNP following preincubation with excess circular dsDNA. Results from three independent experiments were plotted as mean values ± s.d. h, Reaction schematic and strand exchange with increasing concentrations of either RAD51 alone or X3C-RAD51 in the presence of AMP-PNP, results from four independent experiments were plotted as mean values ± s.e., i, Strand exchange with RAD51 alone or in the presence of X3C complexes carrying either WT or XRCC3 mutants. Reactions were performed in presence of AMP-PNP and results from four independent experiments were plotted as mean values ± s.e.
Extended Data Fig. 5 Cryo-EM data processing flowchart and map quality of RAD51-X3CDX2.
a, Auto-picked particles were first sorted by three rounds of 2D classification, six initial models were made first then followed by four rounds of heterogenous refinement. A final step of stringent heterogenous refinement and tight mask homogeneous refinement led to a final higher resolution map. All data analysis were performed using Cryosparc v4.4.1. b, Final Euler angle distribution of all particles used in refinement. c, Local resolution map, color coded from lower resolution (red) to higher resolution (blue). (c) Detailed views of representative local cryo-EM density from the RAD51-X3CDX2 complex reconstruction. d, Detailed views of representative local cryo-EM density from the RAD51-X3C complex reconstruction. e, Cryo-EM map of the RAD51-X3CDX2complex with two views related by a 180 degree rotation about the y axis. Maps of the individual subunits are shown to the right with RAD51 (light blue), XRCC3 (gold), RAD51C (magenta), RAD51D (dark blue), XRCC2 (salmon).
Extended Data Fig. 6 Comparison of ATPase/ATPase domain interactions in RAD51-X3CDX2.
a-d, Analysis of the amino acid interaction networks at the ATPase/ATPase interfaces of the RAD51/XRCC3 (panel a), XRCC3/RAD51C (panel b), RAD51C/RAD51D (panel c), and RAD51D/XRCC2 (panel d) subcomplexes of the RAD51-X3CDX2 structure. Structures are shown as cartoon representations with residues involved in intermolecular interactions shown as sticks. Two views of the complexes related by a 180 degree rotation about the y-axis are presented as indicated.
Extended Data Fig. 7 Subunit-subunit interaction networks within the RAD51-X3CDX2 complex.
a, Cartoon representation of the indicated RAD51/XRCC3 interface with residues involved in intermolecular interaction shown as sticks. An overview of the complex is shown on the left with magnified views of the indicated interfaces presented in the right panels. b-d, The indicated complexes are presented as in panel a with XRCC3/RAD51C (panel b), RAD51C/RAD51D (panel c), and RAD51D/XRCC2 (panel d).
Extended Data Fig. 8 X3CDX2 complex promotes RAD51 filament assembly.
a, ssDNA binding of BC, X3C and DX2 complexes and co-complexes of DX2 with BC (BCDX2) and X3C (X3CDX2). Results from 6 independent experiments were plotted as mean values ± s.d. b, ssDNA binding of WT X3CDX2 or X3CDX2 carrying RAD51D_R266A mutants at the indicated protein concentrations in the presence of AMP-PNP. Results from three independent experiments were plotted as mean values ± s.d. c, ssDNA binding of RAD51-X3C (20 nM) in the presence of increasing concentrations of DX2 complex and AMP-PNP. Results from three independent experiments were plotted as mean values ± s.d. d, Representative micrographs depicting RAD51 nucleoprotein filaments seen by negative stain electron microscopy in absence or presence of indicated concentrations of X3CDX2 complex. Scale is in nanometers. Data represent 49 electron micrographs for each sample captured in two independent experiments. e, Strand exchange in the presence of 30 nM RPA with 100 nM RAD51-X3C alone or in the presence of increasing concentrations of DX2 complex. Reactions were performed in the presence of AMP-PNP. Results from three independent experiments were plotted as mean values ± s.d. f, ssDNA binding of X3CDX2 or X3C complexes carrying either WT or indicated XRCC3 mutants in the presence of AMP-PNP. Results from three independent experiments were plotted as mean values ± s.d. g, ssDNA binding of RAD51 (20 nM) alone or and indicated conc. of X3CDX2 complexes carrying either WT or XRCC3 mutants in the presence of AMP-PNP. Results from three independent experiments were plotted as mean values ±. s.d. h, Strand exchange in the presence of 30 nM RPA with 100 nM RAD51 alone or in the presence of equimolar X3CDX2 complexes carrying either WT or XRCC3 mutants. Reactions were performed in presence of AMP-PNP. Results from four independent experiments were plotted as mean values ± s.d.
Extended Data Fig. 9 Cryo-EM data processing flowchart and map quality of ssDNA-bound RAD51-X3CDX2.
a, Auto-picked particles (2,424,208) were first sorted by four rounds of 2D classification. Next, three initial models were generated, followed by another two initial models from the largest class. Two further rounds of heterogenous refinement followed by a subsequent homogeneous refinement of a 123,502 particle stack resulted in a map with 3.1 Å resolution. Two final rounds of local refinement yielded a final map with a 2.95 Å nominal resolution. All data processing was performed using CryoSparc v 4.2.0. b, Final Euler angle distribution of all particles used in refinement. c, Local resolution map color coded from lower resolution (red) to higher resolution (blue). d, Shell Correlation (FSC) plot from two half-maps with an overall resolution of 2.95 Å as determined by 0.143 criterion. e, Local EM-densities for bound AMP-PNP, side-chains and ssDNA. f, Cryo-EM maps for the overall XRCC3-RAD51C-RAD51D-XRCC2 paralog and individual subunits.
Extended Data Fig. 10 Intermolecular contacts to ssDNA in the RAD51-X3CDX2 complex.
a, The ssDNA-bound RAD51 filament/X3CDX2 complex (cartoon). ssDNA (green) and AMP-PNP (magenta) are shown as spheres. b, Interface between RAD51 subunits and ssDNA. Interacting residues and ssDNA (green) are rendered as sticks. Hydrogen bonds are dashed lines.
Supplementary information
Supplementary Information (download PDF )
Supplementary Figs. 1–5 and Table 1.
Source data
Source Data Fig. 1 (download PDF )
Unprocessed gels.
Source Data Fig. 1 (download XLSX )
Statistical source data.
Source Data Fig. 4 (download PDF )
Unprocessed gels.
Source Data Fig. 4 (download XLSX )
Statistical source data.
Source Data Fig. 5 (download PDF )
Unprocessed gels, uncropped western blots and microscopy images.
Source Data Fig. 5 (download XLSX )
Statistical source data.
Source Data Fig. 6 (download PDF )
Unprocessed gels.
Source Data Fig. 6 (download XLSX )
Statistical source data.
Source Data Extended Data Fig. 1 (download PDF )
Unprocessed gels.
Source Data Extended Data Fig. 3 (download PDF )
Unprocessed gels, uncropped western blots and microscopy images.
Source Data Extended Data Fig. 3 (download XLSX )
Statistical source data.
Source Data Extended Data Fig. 4 (download PDF )
Unprocessed gels.
Source Data Extended Data Fig. 4 (download XLSX )
Statistical source data.
Source Data Extended Data Fig. 8 (download PDF )
EM micrographs and unprocessed gels.
Source Data Extended Data Fig. 8 (download XLSX )
Statistical source data.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Rawal, Y., Kwon, Y., Jia, L. et al. Structural insight into how RAD51 paralog exchange regulates RAD51 filament formation. Nat Struct Mol Biol (2026). https://doi.org/10.1038/s41594-026-01796-6
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1038/s41594-026-01796-6
