
==== Front
Proc Natl Acad Sci U S A
Proc Natl Acad Sci U S A
PNAS
Proceedings of the National Academy of Sciences of the United States of America
0027-8424
1091-6490
National Academy of Sciences

38466836
202316491
10.1073/pnas.2316491121
research-articleResearch Articlebiophys-bioBiophysics and Computational Biology408
Biological Sciences
Biophysics and Computational Biology
Structure of RADX and mechanism for regulation of RAD51 nucleofilaments
Balakrishnan Swati a b
Adolph Madison b https://orcid.org/0000-0002-4761-7388

Tsai Miaw-Sheue c
Akizuki Tae a b
Gallagher Kaitlyn a b
Cortez David b https://orcid.org/0000-0003-0154-140X

Chazin Walter J. walter.chazin@vanderbilt.edu
a b d 1 https://orcid.org/0000-0002-2180-0790

aCenter for Structural Biology, Vanderbilt University, Nashville, TN 37240
bDepartment of Biochemistry, Vanderbilt University School of Medicine, Nashville, TN 37237
cBiological Systems and Bioengineering Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720
dDepartment of Chemistry, Vanderbilt University, Nashville, TN 37235
1To whom correspondence may be addressed. Email: walter.chazin@vanderbilt.edu.
Edited by James Berger, Johns Hopkins University, Baltimore, MD; received September 27, 2023; accepted February 7, 2024

11 3 2024
19 3 2024
11 9 2024
121 12 e231649112127 9 2023
07 2 2024
Copyright © 2024 the Author(s). Published by PNAS.
2024
https://creativecommons.org/licenses/by-nc-nd/4.0/ This article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND).

Significance

Despite the central role of RAD51 in DNA replication and repair processes, the mechanisms of action of its many modulators are poorly understood. Here, we combine structural and biophysical data to determine how the negative regulator RADX (RPA-related RAD51-antagonist on the X chromosome) functions. We show that RADX oligomerizes upon binding DNA, and caps RAD51 filaments at the ends to prevent extension. This work advances knowledge of how RAD51 filaments can be modulated to regulate replication fork reversal and maintain genomic stability.

Replication fork reversal is a fundamental process required for resolution of encounters with DNA damage. A key step in the stabilization and eventual resolution of reversed forks is formation of RAD51 nucleoprotein filaments on exposed single strand DNA (ssDNA). To avoid genome instability, RAD51 filaments are tightly controlled by a variety of positive and negative regulators. RADX (RPA-related RAD51-antagonist on the X chromosome) is a recently discovered negative regulator that binds tightly to ssDNA, directly interacts with RAD51, and regulates replication fork reversal and stabilization in a context-dependent manner. Here, we present a structure-based investigation of RADX’s mechanism of action. Mass photometry experiments showed that RADX forms multiple oligomeric states in a concentration-dependent manner, with a predominance of trimers in the presence of ssDNA. The structure of RADX, which has no structurally characterized orthologs, was determined ab initio by cryo-electron microscopy (cryo-EM) from maps in the 2 to 4 Å range. The structure reveals the molecular basis for RADX oligomerization and the coupled multi-valent binding of ssDNA binding. The interaction of RADX with RAD51 filaments was imaged by negative stain EM, which showed a RADX oligomer at the end of filaments. Based on these results, we propose a model in which RADX functions by capping and restricting the end of RAD51 filaments.

DNA replication
RAD51
replication fork
cryo-EM
HHS | National Institutes of Health (NIH) 100000002 R35 GM118089 Swati BalakrishnanMadison Brett AdolphMiaw-Sheue TsaiTae AkizukiKaitlyn GallagherDavid CortezWalter J. Chazin HHS | National Institutes of Health (NIH) 100000002 P01 CA092584 Swati BalakrishnanMadison Brett AdolphMiaw-Sheue TsaiTae AkizukiKaitlyn GallagherDavid CortezWalter J. Chazin HHS | National Institutes of Health (NIH) 100000002 R01 GM116616 Swati BalakrishnanMadison Brett AdolphMiaw-Sheue TsaiTae AkizukiKaitlyn GallagherDavid CortezWalter J. Chazin
==== Body
pmcReplication fork reversal is a crucial aspect of genome maintenance, allowing for either replication coupled repair or damage bypass upon encounter of a DNA lesion (1). Fork reversal is tightly controlled, with multiple proteins regulating both the formation and stabilization of reversed forks. Unregulated fork reversal in humans is seen to slow down the process of replication, increase the incidence of double-strand breaks, and lead to nascent strand degradation, all of which can result in genome instability (2).

The single-strand DNA binding recombinase RAD51 has a central role in this process, promoting fork reversal (3) and preventing nascent strand degradation at persistently stalled forks (4, 5). It may also promote strand invasion to restart stalled forks (6). Upon replication fork stalling, the exposed single strand DNA (ssDNA) is protected by RPA (replication protein A), the “first-responder” single-strand DNA binding protein (SSB). RAD51 must then replace RPA, a process driven by mediators such as BRCA2 that load and stabilize RAD51 nucleoprotein filaments (7). Inappropriate replication fork reversal can be deleterious to genome stability, so RAD51 action is tightly regulated by multiple proteins. BRCA2 and RAD51 paralog proteins are viewed as positive regulators of formation of RAD51 filaments on ssDNA, even though the exact mechanism of action for the RAD51 paralogs remains unknown (8, 9). RADX (RPA-related RAD51-antagonist on the X chromosome) was recently identified as a modulator of RAD51 at replication forks (10). Its functional relevance is underscored by the observation that RADX expression levels are inversely correlated to the poly ADP-ribose polymerase (PARP)-inhibitor resistance of BRCA2-deficient cells and that higher levels of RADX indicate better outcomes in breast and lung cancer patients (10). Interestingly, patients with high levels of RAD51 fare poorly when afflicted with the same cancer types, suggesting that the modulation of RAD51 function by RADX can impact responses to treatments directed toward DNA replication and repair. This also makes RADX a potential target for the development of therapeutics.

Previous investigations into RADX function show that RADX allows cells to maintain a high capacity for homologous recombination, while buffering RAD51 between fork reversal and protection to ensure replication integrity (10–13). Additionally, pull-down and electrophoretic mobility shift assays have identified that RADX binds to DNA and RAD51, and mutants have been designed based on sequence homology, structural modeling and biochemical approaches to parse the importance of these interactions for RADX function (10, 12). Two models for the mechanism of action of RADX have been proposed as a result of these studies. Single molecule fluorescence experiments led to a model proposing sequestration of DNA by RADX as a mechanism of action (14). The second model is based on our biochemical analysis, which showed that the ATP hydrolysis rate of RAD51 increases in the presence of RADX (12). Since hydrolysis of ATP by RAD51 leads to release from DNA, this observation implies that RADX functions by promoting filament disassembly.

In order to better understand how RADX participates in fork remodeling and in particular, modulates formation and disassembly of RAD51 filaments, we set out to determine the structural and molecular mechanisms of RADX function. It has been proposed that RADX oligomerization may be crucial to its function (15); we used mass photometry to test this hypothesis and systematically characterize the concentration and ssDNA dependence of the population distribution of oligomeric states of RADX. We went on to determine high resolution cryo-electron microscopy (cryo-EM) structures of the predominant RADX trimer and of the secondary population of RADX tetramer, both in the presence of ssDNA. We also used EM to investigate the binding of RADX to RAD51 filaments. Together, these results, integrated with all previous data, provided the basis for proposing that RADX function in fork remodeling involves RADX oligomers binding to and capping the ends of RAD51 filaments.

Results

RADX Oligomerization Is Stabilized and Preferentially Forms Trimers upon Binding ssDNA.

RADX forms homo-oligomers and this property is required for full function (14, 15). RADX mutants that cannot properly oligomerize were seen to cause signs of replication stress in cells, despite retaining the ability to bind ssDNA and localize to replication forks at levels comparable to wild-type RADX. The experiments performed previously demonstrated that RADX forms oligomers but did not specifically characterize which oligomeric states are present.

Mass photometry is a powerful method to characterize the mass of particles present in a solution and was applied to quantify RADX oligomerization under different conditions. RADX alone was found to be primarily monomeric at a concentration of 50 nM, but a modest twofold increase in concentration to 100 nM resulted in a reduction of monomer in favor of the dimer and larger oligomeric states (Fig. 1A and SI Appendix, Fig. S1 A and B). Importantly, the higher-order oligomeric states (trimer and higher) are not discrete well-defined states, but rather appear as an amorphous histogram (SI Appendix, Fig. S1B). This is indicative of rapid transient association/dissociation on the timescale of the mass photometry measurement, where no specific oligomer is stable enough to be cleanly isolated. Thus, RADX alone exists in a concentration-dependent equilibrium between a wide range of oligomeric states.

Fig. 1. RADX oligomerization is stabilized and preferentially forms trimers upon binding ssDNA. (A) RADX is primarily monomeric at 50 nM, but dimerizes at 100 nM. (B) Addition of ssDNA to monomeric RADX leads to a significant increase in the population of trimers, from 4 to 33%. (C and D) Addition of different lengths of ssDNA leads to varying distribution of oligomers. (E) DNA binding affinity of RADX determined by fluorescence polarization anisotropy show there is no dependence on the length of the substrate. (F) Tryptophan quenching assay measuring the DNA footprint of RADX shows a footprint length of 19 to 27 nucleotides.

To elucidate the effect of DNA binding on RADX oligomerization, mass photometry data were acquired for 50 nM RADX in the presence of a variety of ssDNA oligomers. Binding of dT25 had a significant effect on the oligomerization of RADX, producing a prominent trimer along with discrete dimers and tetramers (Fig. 1B and SI Appendix, Fig. S1 A and C). To ensure the results were not unique to this polypyrimidine oligomer, additional experiments for mixed sequence 25-mer ssDNA were acquired. The data obtained for all 25-mers were strikingly similar (SI Appendix, Fig. S1H), which shows that RADX binds ssDNA with no sequence specificity. Thus, although RADX exists as a large and heterogeneous equilibrium distribution of oligomerization states in the absence of ssDNA, the addition of ssDNA significantly stabilizes specific oligomerization states.

In order to determine whether this observation was dependent on the length of the substrate, additional mass photometry data were acquired for 50 nM RADX bound to varying lengths of ssDNA (Fig. 1 C and D and SI Appendix, Fig. S1 C–G). The data show that the distribution of oligomeric states is dependent on the length of DNA available for binding. RADX primarily forms trimers when bound to dT25 and dT20, but the trimers are apparently destabilized as the ssDNA is shortened as dimers become the dominant states in the presence of dT15 (Fig. 1C) and dT11 shifts the equilibrium further toward the monomer. The presence of longer ssDNA lengths such as dT40 and dT60 lead to the formation of larger RADX–DNA complexes with more (e.g., four, five, six) RADX molecules bound to the ssDNA (Fig. 1D). A striking observation in these experiments was that even in the presence of these longer lengths of ssDNA the trimer remains the most abundant state. These results show that the trimer is more stable than other oligomeric states and indicate the higher mass states are not higher-order oligomers but rather two or more lower order oligomers (e.g., 2× trimers) loaded on the ssDNA.

To further ensure that any effects observed were not due to differences in the mode of binding ssDNA, we measured the affinity of RADX using a fluorescence anisotropy assay with fluorescein labeled dT20, dT60, and dT100 (Fig. 1E). The Kd values measured were all very similar (38 ± 5 to 49 ± 6 nM) indicating a consistent ssDNA binding mode. Additionally, the value of the Hill coefficient derived from these data was above 1 even for dT20, indicating cooperativity in RADX oligomerization and binding of ssDNA. The energetic coupling of the two binding phenomena complicates the interpretation of these apparent Kd values and Hill coefficients. Nevertheless, the similarity in these values imply the mode of binding is the same. We also measured the DNA footprint to be 19 to 27 nucleotides by monitoring the fluorescence quenching of tryptophan residues in RADX binding to poly-dT (Fig. 1F). Based on the mass photometry data showing that the trimer and tetramer dominate, the observed site size of 19 to 27 nucleotides corresponds to approximately six nucleotides per protomer for the trimer and tetramer, respectively.

Mass photometry, while powerful, is restricted to measurements in the nM range of concentrations. To further explore RADX oligomerization at higher concentrations, we performed small angle X-ray scattering (SAXS) experiments (SI Appendix, Fig. S2). It is not possible to concentrate RADX above 5 µM in the absence of ssDNA and as a result we were unable to collect high quality data to perform SAXS experiments on the free protein. However, RADX is more readily concentrated in the presence of ssDNA and has a much lower tendency to aggregate, so SAXS experiments could be performed for the complex with dT25, albeit at the lower limit of protein concentration for high quality data. Linearity in the Guinier region and the ability to include data points to low values of q (nmin = 2) indicated the data were of acceptable quality to extract parameters relating to the mass and oligomerization state. The P(r) function indicates the protein is globular in nature. The calculated mass of 640 kDa for the complex of RADX with dT25 from the SAXS data corresponds to a tetrameric state (SI Appendix, Fig. S1C). The 300-fold increase in concentration from 100 nm to 30 μM appears to cause a shift from a preference for trimer to tetramer, which is consistent with the trend of higher-order oligomers with increasing protein concentration.

High Resolution Structure Shows RADX Is Comprised of Four Independent OB-Fold Domains.

While significant data have been accumulated about RADX function, the lack of a three-dimensional (3D) structure has limited the ability to draw firm conclusions about its mechanism of action. A previous analysis of the RADX sequence proposed that RADX contains three N-terminal OB-fold domains and a dual motif C-terminal domain (10). However, the lack of RADX orthologs and its poor sequence homology to other proteins with known structures precludes reliable prediction of its 3D structure. We note that AlphaFold does produce a structural model for monomeric RADX with relatively high confidence for 75% of the protein but was unable to produce a viable structure for a RADX trimer or tetramer. We therefore set out to determine the structure of RADX by cryo-EM.

Initial screening was performed by selecting specific fractions of the complex of RADX with dT25 purified by size exclusion chromatography. Negative stain imaging of these fractions produced two-dimensional (2D) class averages with different sizes and shapes of particles (SI Appendix, Fig. S3A). Upon screening the same sample in cryo-EM, the distribution of individual particles was more readily determined, but the resulting 2D class averages remained poorly resolved despite extensive attempts to optimize the data analysis parameters (SI Appendix, Fig. S3B). From these results, we surmised that RADX has significant inter-domain flexibility under these conditions.

To overcome this barrier, samples of the RADX–dT25 complex were chemically cross-linked with BS3, which resulted in significant improvement in the resolution of the 2D classes (SI Appendix, Fig. S3C). Particles in these grids were still heterogeneous, which was attributed to population of different oligomeric states. After extensive 2D classification, it became clear that trimer and tetramer particles were dominant and in sufficient quantity to warrant pursuing refinement of their respective structures (SI Appendix, Fig. S4 and Table S1). Trimer and tetramer particles were then separated out based on both 2D and 3D classification. The distribution of particles from this analysis was ~70% trimer and ~30% tetramer, consistent with the ratio observed in the mass photometry data on the non-crosslinked protein (Fig. 1A). This suggests that the trimer and tetramer particles are not artifacts of cross-linking.

The structures of the trimer and tetramer were each determined using a combination of non-uniform refinement and local refinement with masking (SI Appendix, Fig. S4). Local refinement was used to circumvent the loss of resolution due to motion between each RADX protomer relative to the others in the oligomer. Thus, separate local refinements were performed on the A-B and B-C RADX pairs within the trimer (Fig. 2A). The maps were then merged to generate the final composite map, which has an average resolution of 2.9 Å (Fig. 2A and SI Appendix, Fig. S5 A and B). The data for the tetramer were processed in the same manner with separate local refinements performed on the A-B, B-C, and C-D pairs. The density for the fourth RADX protomer (D) was relatively poor compared to A, B, and C and when combined with the smaller number of particles, resulted in the average resolution of the final map for the tetramer being 3.7 Å (Fig. 2B and SI Appendix, Fig. S5 C and D). Importantly, density for the ssDNA is well resolved in both maps (SI Appendix, Fig. S5E). Subsequent analyses are based on the structure of the trimer as the resolution of the final map was higher.

Fig. 2. High resolution structure of RADX. (A and B) EM map and ribbon representation of the structure of the RADX trimer bound to ssDNA (red) with the three protomers in cyan, blue and purple respectively. (C and D) EM map and ribbon representation of the structure of the RADX tetramer bound to ssDNA (red) with the four protomers in cyan, blue, purple and violet respectively.

RADX is found to consist of four OB-fold domains (OB1-OB4) (SI Appendix, Fig. S6), with a 110-residue disordered insertion within the fourth domain spanning S566–I676, for which no density was observed. All four domains are canonical OB-folds with a core β-sheet organized into two three-stranded anti-parallel sub-sheets that are each composed of a central curved β-strand with anti-parallel strands on either side. Each domain also includes 2 or 3 α-helices, with the first helix packed at the bottom of the β-sheet, completing the canonical OB-fold domain structure (16). There is clear density for the ssDNA in the RADX trimer, which contacts 16 of the 25 nucleotides. Notably, the density of the fourth protomer (RADX-D) in the tetramer is insufficient to visualize all the ssDNA. A key feature of both the trimer and tetramer structures is the lack of symmetry with respect to the three or four protomers within in each structure.

A comparison of RADX to known structures using the programs DALI (a distance-matrix alignment algorithm) (17) and FoldSeek (18) show that RADX has a unique protein fold, with no alignment matches for all four domains. Many matches are seen for the single domains, such as the alignment of RADX OB1 with RPA 70N (SI Appendix, Fig. S7) or OB3 with POT1. However, OB4 stands out as it does not have any well-matched homologs. All potential alignments to OB4 show very large RMSDs (over 10 Å) and very low Z-scores (below 3). The unique combination of domains as reflected in the lack of well-matched homologs make our RADX structure a valuable addition to the evolving understanding of protein structure.

RADX Oligomerization Is Stabilized by Multiple Inter-Domain Interfaces.

Oligomerization is a fundamental biochemical property of RADX, therefore elucidating the driving force for RADX oligomerization is important for understanding its function. To this end, we analyzed the interfaces between RADX protomers in the high-resolution structure of the trimer using the PDBePISA server (19). The primary interface mediating oligomerization is between OB1 and OB4 on adjacent protomers and a secondary interface is provided by successive OB2 domains.

In the primary interface OB4 of RADX-A interacts with OB1 of RADX-B, OB4 of RADX-B interacts with OB1 of RADX-C, and so on. Although the interfaces involve the same domains, differences in the relative orientation of the protomers are required so that each can align successively for oligomerization (Fig. 3). On average, the three OB1–OB4 interfaces in the trimer have a buried solvent accessible surface area of ~600 Å2 and six hydrogen bonds or salt bridges (Fig. 3). Key hydrogen-bonding residues include E526, L529, Q553, N759, and E761 from OB4 and R58, Y138, E140, K141, and R142 from OB1. The importance of this interface for oligomerization is underscored by mass photometry experiments acquired for an N-terminal construct spanning R43-K560 (RADX-N), which lacks OB4 (SI Appendix, Fig. S8A). These data show that RADX-N does oligomerize, albeit almost exclusively to the dimeric state. Notably, this RADX-N construct does not form stable higher-order oligomers even in the presence of ssDNA (SI Appendix, Fig. S8B). We attribute the ability of RADX-N to oligomerize to the secondary OB2–OB2 interface. Although this interface is smaller than the OB1–OB4 interface, averaging ~100 Å2 in buried surface area and with only 1 or 2 hydrogen bonds involving residues R232 and Y307, it is sufficient to support formation of RADX-N dimers.

Fig. 3. RADX oligomerization is stabilized by multiple inter-domain interfaces. The OB1–OB4 interface (in gold) between two RADX protomers A and B (teal and blue, respectively). RADX oligomerizes primarily via hydrogen bond and salt-bridge interactions between domains 1 and 4. The Left Inset shows the residues involved in hydrogen bonding colored according to the protomer of origin, with the hydrogen bonds shown as red dashed lines. The Right Inset shows the three sites of RADX oligomerization mutations (residues in red) in the central beta sheet of OB4. Mutation of these residues are likely to perturb the beta sheet and the entire D4 domain, which could contribute to effects observed in functional assays.

A previous investigation of RADX function surmised that the fourth predicted domain in the C terminus was important for oligomerization, and three single-site mutations (E842K, K847E, and Y848A) were investigated to test this hypothesis (15). These variants inhibited oligomerization, retained DNA and RAD51 binding, and localized to replication forks. Remarkably, the structure shows that these three residues are not located at the OB1–OB4 oligomerization interfaces. Rather, all three form part of a β-strand that is critical to the central β-sheet of OB4, which suggests these mutations may disrupt the structural stability of OB4.

To more definitively validate the RADX oligomerization interfaces, we designed, expressed and purified a RADX variant that contains five key mutations that span both the OB1–OB4 and OB2–OB2 interfaces: R58E/K141A/R142E/R232E/Y307A (RADX-OLM) (Fig. 3). This variant had expression levels comparable to wild-type RADX and an identical profile from size-exclusion chromatography, suggesting the mutations had not caused significant changes to the structure or stability of the protein. The oligomerization capacity of this mutant was characterized using mass photometry. The data show that like the wild-type protein, the RADX-OLM alone is primarily monomeric. However, unlike the wild-type protein, there is no change upon addition of ssDNA (Fig. 5 A and B). These data confirm the key role of the inter-domain contacts in RADX oligomerization.

RADX Binding of ssDNA Is Coupled to Oligomerization.

The discovery and initial analyses of RADX revealed homology to the large subunit of RPA70 and tight binding of ssDNA (10). RPA binds ssDNA using multiple OB-fold domains (20), including three in RPA70 (RPA70ABC). The initial prediction of three OB-fold domains in RADX suggested that RADX uses the same multi-valent mode of binding as RPA. Subsequent mutational analysis revealed the central role of OB2 (10), but exactly how RADX binds ssDNA remained poorly understood.

The high-resolution structure of RADX shows that it does bind ssDNA in a multi-valent mode, but not in the same manner as RPA. Instead, RADX binds ssDNA almost exclusively through OB2 and oligomerizes to generate multi-valency and attain high affinity. In the RADX trimer, all three protomers are tightly packed in together in a manner that allows for binding of the ssDNA via their respective OB2 domains with direct contacts to 16 nucleotides. This observation is consistent with the 19 to 27 DNA footprint obtained for poly-dT, given that occluded site size determined by tryptophan quenching is usually larger than the number of residues in direct contact and that the RADX tetramer bound to the DNA is also significantly populated. The high affinity for ssDNA is reflected in the total of ~2,100 Å2 of buried surface across the entire interface. Consistent with the tight packing of OB2 domains and the lack of symmetry within the trimer, the ssDNA binding interfaces are not identical for each protomer. The ability of the protomers to adapt to bind the substrate, and the fact that on average only ~5 nucleotides are bound by each, provides a potential explanation for why a substantial population of RADX tetramers bound to dT25 is also observed in addition to the dominant trimers.

The interaction with ssDNA is driven to a great extent by a large positively charged surface in OB2 that complements the ssDNA polyanion (Fig. 4). A few additional contacts with the ssDNA are contributed by residues in one loop of OB3 in protomers A and B, but not the OB3 loop in protomer C. Correspondingly, the ssDNA binding interfaces of the protomers are not identical: Protomers A and B each contact six nucleotides and bury ~750 Å2 of solvent exposed surface, whereas protomer C contacts 5 nucleotides and buries only ~600 Å2 due to the absence of the contribution from the OB3 loop (Table 1).

Fig. 4. RADX binding of ssDNA is coupled to oligomerization. The Top panel shows the ribbon representation (Left) of a RADX protomer (blue) with ssDNA bound (red), with the corresponding map of the surface electrostatic charge showing the positively charged DNA binding surface in blue (Right). The Top Inset shows the consensus residues involved in hydrogen bonding or pi-stacking interactions (blue) with the DNA (red). The Lower Inset shows the sites of the Ob2m mutations (green and orange). Among the mutations, only the two residues with orange labels (W279, K305) have any overlap with the DNA binding site.

Table 1. Analyses of the DNA binding surfaces of the three protomers in the RADX trimer

Protomer	Interface area (A2)	*NSB+NHB+Npi	⍙G (kcal/mol)	Residues† involved in H-bonds, salt bridges and pi-stacking	
RADX-A	744.5	10	−11.5	Y250, Q262, F264, W279, Y307, F309, R333, R396‡	
RADX-B	753.4	9	−7.8	R248, Y250, F264, W279, Y301, K304, Y307, F309	
RADX-C	579.6	9	−8.4	R232, R248, Y250, Q262, F264, W279, Y307, N331	
*Number of salt bridges + number of hydrogen bonds + number of pi-stacking interactions.

†Underlined residues participate in hydrogen bonds.

‡This residue is in OB3.

The RADX–DNA interaction is also supported by pi-stacking interactions and hydrogen bonds. The aromatic residues from OB2 at the interaction interface such as W279, Y307, and F309 contribute pi-stacking interactions in each protomer, but the identity and number of the hydrogen-bonding residues varies from protomer to protomer. For example, OB2 residues Q262 and R333 and OB3 residue R396 in protomer A form hydrogen bonds with the ssDNA, whereas only OB2 residues R248 and K304 are engaged in hydrogen bonds in protomer B (Table 1 and SI Appendix, Fig. S9).

To obtain additional insights, the free energy of RADX binding ssDNA was estimated using PDBePISA (19). The predictions ranged between −8 and −12 kcal/mol for the three different RADX protomers, much higher than the values of −1 to −2 kcal/mol determined for contributions to oligomerization. The higher values are fully consistent with the observation in the mass photometry data that binding to ssDNA stabilizes discrete oligomer populations (i.e., trimers and tetramers) (Fig. 1).

RADX variants to probe the functional relevance of DNA binding were designed previously based on sequence homology to RPA70, structural predictions and biochemical experiments. Two DNA-binding deficient variants containing multiple mutations in the predicted DNA binding surface of the OB2 domain were generated: one with nine mutations (R240E/R248E/K252E/K255E/K256E/W279A/K304E/R310E/E327A) (10, 12) referred to as RADX-Ob2m, and the other with two (K304A/E327A) (13) mutations. Both variants attenuated but did not abrogate DNA binding, which was interpreted as indicating other RADX domains must be involved in binding DNA. With the availability of the structure, we can now see that only two of the eight mutated residues (W279 and K304) directly contact the DNA (Fig. 4). This explains why these variants reduced but did not eliminate binding to ssDNA.

To validate the structural analysis of the driving force for RADX binding of ssDNA, a RADX variant was designed with five key mutations in the DNA binding interface: R248E/Y250A/Q262A/R333E/R396E (RADX-DBM). The expression and purification of this mutant was comparable to wild-type RADX and RADX-OLM, suggesting no large structural disruption or localized unfolding were induced by mutating these residues. We measured the ssDNA binding of RADX-DBM using fluorescence anisotropy, and, as anticipated, the affinity of RADX-DBM was drastically reduced, exhibiting no detectable binding of ssDNA. To ensure that the length dependence of DNA binding was not altered, the substrate length was varied from 20 to 100 nt but no detectable binding was observed (Fig. 5 E–G). These data confirm the critical role of the contacts observed in the structure for RADX binding of ssDNA. We had previously observed for RADX-OLM that its state of oligomerization was not altered upon the addition of ssDNA. This suggested there is an interplay between oligomerization and DNA binding. To further explore this hypothesis, we measured the ssDNA binding affinity of RADX-OLM using fluorescence anisotropy. The data show RADX-OLM has significantly weaker affinity than wild-type RADX (Fig. 5 E–G). Saturation of binding is not achieved due to the limitations of concentrating RADX, so a Kd value could not be extracted. However, these data are sufficient to establish that the affinity for ssDNA of the RADX-OLM is reduced greater than 100 fold compared to RADX. Similarly, due to the low affinity for ssDNA, stoichiometric binding of RADX-OLM to a poly-dT substrate could not be attained, which limits the ability to determine the footprint of RADX-OLM (Fig. 5H). Having observed the effect of the OLM mutations on DNA binding affinity, we asked whether the RADX-DBM mutations had an effect on RADX oligomerization. The mass photometry profile of RADX-DBM alone mimicked the profile of wild-type RADX with a wide spectrum of species and no well-resolved oligomeric state (Fig. 5C and SI Appendix, Fig. S1B). However, in contrast to wild-type RADX, no changes in the profile were observed upon the addition of ssDNA (Fig. 5D). Together, these data confirm that oligomerization is not only stabilized by binding to ssDNA, but that in turn, oligomerization is crucial to binding ssDNA with high affinity. Thus, this strong energetic coupling of DNA binding and RADX oligomerization lies at the core of RADX function.

Fig. 5. RADX variants show coupling of oligomerization and DNA binding. (A–D) RADX-OLM and RADX-DBM do not oligomerize in the presence of ssDNA. (E–G) DNA binding affinity of RADX-OLM is significantly lower than wt RADX, while RADX-DBM shows no detectable binding to ssDNA. (H) Measurement of tryptophan quenching of RADX-OLM does not show stoichiometric binding, as compared to wt RADX.

RADX Is Unique but Well Conserved.

RADX does not share significant sequence identity with proteins apart from other RADX orthologs. The closest match is to the large subunit of RPA (RPA70), although there is only 20% sequence identity between them. In contrast, the protein sequence is well conserved among RADX orthologs. Multiple sequence alignment shows that human RADX has a sequence identity of 75 to 100% with mammalian RADX orthologs and at least ~50% sequence identity with RADX orthologs outside the Mammalia family (SI Appendix, Fig. S10 and Table S2 A and B). The secondary structure elements present in the four domains of human RADX are seen to be well conserved across all the families where RADX is present. The highest variability is in the unstructured loop of human RADX. The availability of an experimental structure provides a window to view sequence conservation in a structural and functional context and enables a search for structural homologs with similar folds.

We were particularly interested in investigating whether sequence conservation analysis would reveal residues in RADX whose structural and/or functional importance had yet to be investigated. Sequence conservation was assessed using the ConSurf (21) server, which revealed as expected the structured regions of OB1, OB2 and OB3 are highly conserved (SI Appendix, Fig. S11A). The key hydrogen-bonding residues involved in the oligomerization of RADX are also shown to be highly conserved, as are a majority of the residues of the DNA binding interface. There are two exceptions, F309 and R232, that are not conserved, the origin of which is not clear. In contrast to OB1-OB3, OB4 has significant regions of variability, particularly in the large unstructured loop (S566-I676) inserted within the globular OB-fold domain (SI Appendix, Fig. S11B). Multiple serine and threonine residues (e.g., T576, S584, T585, and S586) in this large loop are particularly intriguing as they are potential phosphorylation sites that could modulate RADX activity (22, 23).

RADX Modulation of RAD51 Filaments Correlates with End-Binding.

We have shown previously that binding of ssDNA and interaction with RAD51 are required for RADX to inhibit RAD51 nucleoprotein filament formation on ssDNA in vitro (12). We also demonstrated that RADX reduces the length of RAD51 filaments. Potential models for how RADX alters RAD51 filament formation include acceleration of the ATP hydrolysis rate of RAD51, sequestration of the ssDNA by RADX to prevent RAD51 binding or a combination of the two. To further explore the inhibition mechanism, define the RAD51 interaction interface, and address how RADX modulates RAD51 filaments, we turned to electron microscopy.

In our previous study of RADX modifications of RAD51 filaments monitored by negative stain EM (12), we were unable to locate RADX molecules due to their small size and the limited amount of RADX in the specimen. To overcome this limitation, an anti-RADX antibody attached to a 20 nm gold nanoparticle was used as a fiducial marker to locate RADX molecules, an approach used previously to locate BRCA2 bound to RAD51 filaments (24). RAD51 filaments were prepared with ATP and RADX attached to nanoparticles was added at a 1:100 ratio to mimic cellular conditions. The low level of RADX is also needed because the EM data are collected during the course of “live” RAD51 filament formation; as a negative regulator, high levels of RADX would completely abrogate the formation of filaments.The negative stain EM micrographs obtained with nanogold labeling revealed RADX is binding to one end of RAD51 filaments (Fig. 6 A, Left). Armed with this knowledge, negative stain EM data were collected on RADX bound to RAD51 filaments stabilized by chemical-crosslinking using BS3. These data could then be analyzed by leveraging the availability of the high-resolution structure of RADX, allowing us to build low-resolution templates and find RADX more effectively in micrographs. The resulting particles from template-picked micrographs were assigned into 2D classes, which included free RAD51 filaments, free RADX molecules and RADX bound to the ends of RAD51 filaments (Fig. 6 A, Right). The RADX bound filaments were distinguished on the basis of volumes and comparisons to previous structures of RAD51 filaments (25). All 2D classes containing RAD51 filaments were isolated and grouped for ab-initio reconstruction and 3D classification with ~4,000 particles. This led to the generation of two low-resolution reconstructions at ~20 Å, one containing only a RAD51 filament and the second in which the RAD51 filament is clearly visible with the additional unique volume at the end assigned to RADX (Fig. 6B). These results imply that RADX functions by binding to the ends of growing RAD51 filaments and capping further growth.

Fig. 6. RADX modulation of RAD51 filaments correlates with end-binding. (A) (Left) Insets from negative stain EM micrographs showing anti-RADX Ab bound gold nanoparticles localized at the ends of RAD51 filaments. (Right) 2D class averages containing RADX crosslinked to RAD51 filaments, which reveal populations of free RADX and RADX bound to RAD51 filaments. (B) 3D reconstructions of free RADX and RADX bound to RAD51 filaments. The extra volume assigned to RADX is highlighted by the red circle. (C) The residues mutated in the RAD51 binding-deficient QVPK mutant displayed on the structure of RADX. These residues form a surface-exposed patch, consistent with the proposal that these are in the RAD51 binding site. (D) Model of a RADX trimer bound to a RAD51 filament generated via manual docking. The RADX is colored green, the RAD51 filament (8GYK) purple, and the ssDNA red.

RADX variants to probe RAD51 binding have been designed based on homology modeling and biochemical analysis. The RAD51 binding-deficient variant of RADX (RADX-QVPK) contained alanine substitutions for four consecutive residues in the OB3 domain (Q451-K454). In the high-resolution structure of the trimer, these residues are in a loop region and surface exposed (Fig. 6C), consistent with participation in a protein–protein interaction. This RADX QVPK variant led to elevated levels of the DNA-damage marker protein γH2AX in U2OS cells, and was unable to rescue slow rates of replication fork elongation in RADXΔ cells (12). These observations show that the RAD51-RADX interaction is essential to maintain replication fork stability in cells.

To gain further insight into the RAD51-RADX interaction, we generated a model of a RAD51 filament with a RADX trimer bound to one end by docking the structures manually and with HADDOCK (26) (Fig. 6D). The active residues specified for docking were the previously identified Q451-K454 for RADX, while CPORT was used to predict RAD51 active residues since no experimental information is available. Multiple HADDOCK-generated models and the manually docked model both fit remarkably well into our 3D reconstruction of the RAD51-RADX complex as reflected in the goodness of fits assessed using the correlation coefficient from Chimera. In these docked models, the DNA binding site of RADX-A OB2 is remarkably well set up in a position to bind free ssDNA exposed at the end of the filament. There are a large number of structures of RAD51 filaments and the best fit to the density was obtained for RAD51-ADP double filaments(8GYK) (27). Since the experiment was conducted in the presence of ATP and cross-linker over the course of 30 min, RADX would be expected to be released when ATP was hydrolyzed but the complex could be trapped by cross-linking before RADX diffuses away and then the structure cross-linked. The most likely scenario is there is a mixed population of different filaments with RADX bound with a majority being RAD51-ADP double filaments. However, with the limited dataset available it is not possible to separate out the different types of RAD51 filaments.

We prefer a model of RADX binding to RAD51 filaments on the 3′ end over the alternative model of 5′-end binding because in vitro, in the absence of other filament modulators, RAD51 filaments on ssDNA are seen to extend with a slight preference for the 5′-3″ direction (28). This is also consistent with the observation that BRCA2 BRC motifs compete with RADX for binding to RAD51 (12), as BRCA2 binds the 3′ end of the filament as well. In this model, capping on the 3′ end would directly inhibit filament extension. More detailed analysis of the RAD51–RADX interface would be of great interest but was not possible due to the low resolution of the negative stain EM reconstructions and the scarcity of information about the RADX interaction surface on RAD51.

Discussion

RADX is required to maintain replication fork stability and is a direct regulator of RAD51 that counters the effect of positive regulators such as BRCA2. Knowledge of its structure and mechanism of action is of significance because tight regulation of RAD51 is needed to maintain genome stability. Thus, the high resolution cryo-EM structure of RADX opens up new horizons for exploring its function(s) as a modulator of RAD51 nucleofilament formation. RADX is structurally unique, with no matching protein folds in structural databases. Knowledge of the interfaces mediating oligomerization and DNA binding, energetically coupled and crucial to its function, facilitates precise design of separation of function mutants, which will enable more accurate definition of the roles of RADX in fork stalling, reversal and stabilization.

The absence of density for the large 111 residue loop (S566-I676) inserted within the OB4 domain strongly suggests it is disordered, consistent with predictions based on sequence. Hence, the functional role(s) of this loop remains a mystery. It is possible that the loop mediates interactions with as-yet-unknown binding partners. Alternatively, it may function in regulation of RADX activity. For example, there are number of highly conserved residues in the loop, including multiple serine and threonine residues that are potential sites of phosphorylation. Poly-phosphorylation has been shown to serve as a DNA mimic that can block DNA binding sites (22, 23), and could play such a role in RADX. Further biochemical and functional analysis of corresponding RADX loop mutants are required to test this hypothesis.

RADX modulation of RAD51 filaments has a parallel in RecX modulation of RecA filaments in bacteria. Like RAD51, RecA is a DNA dependent ATPase that promotes replication fork reversal and prevents degradation of ssDNA by nucleases (29). Both proteins also cycle between the ATP-bound form which favors filament formation, and the ADP bound form which shifts the equilibrium toward dissociation from DNA and filament breakdown (30). However, RAD51 and RecA have opposite polarity for strand exchange (31), and do differ with respect to filament extension in vitro: RecA retains its 5′-3′ polarity, but RAD51 is bidirectional with a slight preference for 5′-3′ (28, 32, 33). Both proteins are modulated by a number of regulatory proteins. Like RADX, RecX has been identified as a negative regulator of RecA. While RecX is not a homolog of RADX, certain parallels can be seen in their mechanisms of action. RecX binds ssDNA and interacts with DNA-bound RecA (34). RecX inhibits RecA nucleoprotein filament formation by a mechanism involving capping RecA filaments through a RecX-RecA interaction, which prevents filament extension. However, unlike RADX, RecX reduces the rate of ATP hydrolysis by RecA (35). RecX functions to impede homologous recombination at double strand breaks in bacteria. In contrast, it is unclear whether RADX has a role in recombination as depletion of RADX has a limited effect on homologous recombination except when overexpressed in BRCA2-deficient cells (10). The lack of sequence similarity, opposite effects on ATP hydrolysis and differences in function indicate that despite some parallels, RecX and RADX are not homologs.

We have determined that RADX oligomerizes in a concentration and ssDNA-dependent manner, and that DNA binding drives the equilibrium to the trimeric and tetrameric states. Basal cell expression levels of RADX are in the nanomolar concentration range, where mass photometry shows RADX preferential formation of trimers, supporting the idea that this is a functionally relevant state. Mass photometry also shows that the preference for trimers remains independent of the length of ssDNA and that trimer particles form the majority population in cryo-EM micrographs. The highest masses observed for RADX on dT25, dT40 and dT60 correspond to a tetramer, pentamer and septamer, respectively. Thus, RADX oligomerization does not scale linearly with the increase in substrate length. Since mass photometry only reports on the total mass of a complex, if RADX formed single filaments, the oligomer would grow by one protomer for each 5 to 6 nucleotides in additional length and one would expect a predominance of septamers and undecamers for dT40 and dT60, respectively, but this is not observed. Rather, the results for dT40 and dT60 are consistent only with combinations of discrete lower order dimers, trimers and tetramers spaced apart on the substrate, not a single large oligomer. The results also align with the DNA footprint for poly-dT of 19 to 27 residues, a range that corresponds to trimers and tetramers and not higher-order RADX-DNA oligomers.

Fig. 7 shows a diagrammatic representation of our model of how RADX functions. Upon exposure of ssDNA at stalled replication forks, RPA binds and protects the DNA from degradation, and recruits a range of damage repair factors. RPA is replaced by RAD51 via the action of a mediator such as BRCA2. RADX is localized to the vicinity of RAD51 through its protein interaction surface and can bind exposed ssDNA at the termini of the expanding RAD51 filament. RADX binding to ssDNA is energetically coupled to oligomerization and forms stable caps at the end of the filament, effectively blocking further extension. We have previously shown that the rate of ATP hydrolysis by RAD51 increases in the presence of RADX, and that binding of BRCA2 to RAD51 slows the ATP hydrolysis rate (12). Therefore, the mechanism by which RADX disassembles RAD51 filaments is likely to consist of two parts: end capping of the RAD51 filament by RADX oligomers to block extension, and a conformational change in RAD51 upon interacting with the terminal RADX that accelerates ATP hydrolysis and therefore release of RAD51 from DNA.

Fig. 7. Model for the mechanism of RADX action. (A) Upon replication fork stalling, single stranded DNA is exposed, (B) which is then bound by RPA to protect the ssDNA from damaging agents. (C) RPA is replaced on ssDNA with RAD51 by the action of mediator protein BRCA2. RAD51 binds cooperatively to form filaments on ssDNA. (D) RADX is recruited to ssDNA where it binds the ends of the growing RAD51 filament. (E) RADX promotes filament disassembly by either i) blocking expansion at the end of the filament as RAD51 hydrolyses ATP and detaches from ssDNA or ii) both blocking filament expansion and accelerating the ATP hydrolysis rate of RAD51.

Our model places RADX at the 3′ end of the filament. In isolation, RADX will be at either end of a RAD51 filament. However, at a stalled replication fork in the cell, other modulators such as BRCA2 and the RAD51 paralog proteins will also be present (5, 36). While BRCA2, which binds at the 3′ end, is seen as primarily a RAD51 filament nucleation factor, the RAD51 paralog proteins bind the 5′ end of the filament and stabilize it, promoting 3′-5′ filament extension (9). Therefore, while it is possible that RADX caps either or both the 3′ and the 5′ ends of the RAD51 nucleoprotein filament depending on the other proteins acting at the replication fork at that time, we have chosen to place RADX at the 3′ end of the filament due to the data showing its ability to compete with BRAC2.

Formation of RAD51 filaments is a co-operative process (37) with the rate of growth of filaments on ssDNA measured as three orders of magnitude higher than the nucleation rate. Regulators like BRCA2 play a crucial role during nucleation, first by making DNA available by removing RPA and then stabilizing an oligomer of RAD51 on ssDNA. Once filament formation has begun cooperativity results in rapid elongation of the filament, which would limit the efficacy of an inhibition mechanism that relies solely on out-competing RAD51 for ssDNA. A mechanism that physically blocks RAD51 and shifts the equilibrium of DNA binding to disassembly is both plausible and supported by in vitro experimental evidence, including the disassembly of RAD51 filament by RADX and corresponding increase in the hydrolysis rate of RAD51 bound ATP (12), as well as the localization of RADX at the termini of filaments shown in this study. Additional studies using our separation-of-function mutants designed based on the RADX structure as well as determination of a high-resolution structure of RADX bound to RAD51 will test and refine this proposed mechanism of RADX function.

Materials and Methods

Expression and Purification of RADX.

RADX, RADX-OLM, RADX-DBM, and a C-terminal truncation construct (RADX 43-560, RADX-N) were expressed and purified from three different expression systems, Escherichia coli, Sf9 insect cells (Expression Systems), and Expi293F mammalian cells (Thermo Fisher). Final EM and mass photometry data were collected for protein expressed in mammalian cells. Details of expression and purification are given in SI Appendix.

Expression and Purification of RAD51.

RAD51 was expressed in BLR(DE3)pLysS E.coli cells, a RecA deficient cell line. Details of expression and purification are given in SI Appendix.

Mass Photometry.

Mass photometry studies were performed using a TwoMP system from Refeyn (38). Details of sample, buffer conditions, and experimental parameters are given in SI Appendix.

Steady State Rotational Anisotropy and Fluorescence Quenching Measurements.

Rotational anisotropy was used to measure the affinity of RADX for various lengths of fluorescein-labeled ssDNA, using a filter-based Biotek plate reader. Experimental details and analysis procedures are given in SI Appendix.

SEC-SAXS Data Collection and Analysis for RADX-dT25.

The SAXS profiles for the RADX-dT25 complex were collected in SEC-SAXS mode at the ALS beamline 12.3.1 at the Lawrence Berkely National Laboratory in Berkeley, CA (39). Details of data collection and analysis are given in SI Appendix.

Cryo-EM Sample Preparation and Data Collection.

RADX was incubated with a twofold molar excess of dT25 and then purified by size-exclusion chromatography (SEC). The final grids were prepared by taking a 20 μL aliquot adjusted to 1 μM RADX then incubating with 10 μM BS3 for 2 h on ice. Grids were screened using a Glacios 200 kV transmission electron microscope (TEM, ThermoFisher) equipped with a Falcon4 direct electron detector. Large-scale datasets were collected using a Titan Krios G4 300 kV TEM (ThermoFisher) equipped with a Gatan K3 BioQuantum direct electron detector functioning in counting mode. Details on grid preparation and data acquisition parameters are given in SI Appendix.

Cryo-EM Data Processing and Analysis.

Data processing and analysis was performed using cryoSPARC (40). The initial model was then further refined using both the real space refine package and secondary structure/stereochemical restraints in Phenix (41) and visualized using Coot (42). Details of the analysis, refinement, and generation of figures are given in SI Appendix.

RADX Localization and Negative-Stain EM Studies.

To determine localization of RADX on RAD51 nucleoprotein filaments, 20 nm gold nanoparticles from an antibody conjugation kit were conjugated to an anti-RADX mouse monoclonal Ab (CXorf57, Santa Cruz Biotechnology, sc-514563). The conjugated antibody was then incubated with an excess of RADX and added to pre-formed RAD51 filaments. This sample was fixed on grids and imaged using negative stain EM. The details of the kit, protocol for filament formation, details of the data collection and analysis are given in SI Appendix.

Computational Analyses to Determine Sequence Conservation, Structural Homology, and Docking of RADX and RAD51.

The PDBePISA server was used to analyze the residues involved at the interaction interfaces between RADX protomers, the interaction interfaces between RADX and ssDNA, and the energetics of the interactions. The HADDOCK (26) server was used for docking RADX with RAD51 filaments, while sequence-based homology and alignment were analyzed using BLAST (43) and ALIGN (44). Sequence conservation was assessed using ConSurf (21). Analysis of similarity to known structures was performed on the DALI (17) and FoldSeek (18) servers. Details of the docking parameters, sequences used for sequence based homology, and alignment as well as parameters for the ConSurf, DALI and FoldSeek analyses are given in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

We thank Prof. Melanie Ohi, Dr. Jason Porta, Dr. Heather Kroh, Dr. Elad Binshtin and Dr. Elwood Mullins for advice and assistance with cryo-EM data analysis. Dr. Carl Schiltz provided advice on optimization of protein cross-linking. This work was supported by grants from the US NIH: R35 GM118089 (W.J.C.); P01 CA092584 (M.-S.T., D.C., and W.J.C.), R01 GM116616 (D.C.). The EM data were acquired at the Vanderbilt Center for Structural Biology Cryo-EM Facility with the support of NIH grant S10 OD030292 for the purchase of the Glacios cryo-TEM.

Author contributions

S.B., M.A., D.C., and W.J.C. designed research; S.B., M.A., T.A., and K.G. performed research; M.-S.T. contributed new reagents/analytic tools; S.B., M.A., and W.J.C. analyzed data; D.C. and W.J.C. funding and supervision of research groups; and S.B. and W.J.C. wrote the paper with input from all authors.

Competing interests

The authors declare no competing interest.

Data, Materials, and Software Availability

Structural data have been deposited to the PDB and EMDB. This includes structural co-ordinates, cryo-EM maps, and details of data collection. Accession codes for the RADX trimer are PDB ID 8U5Y/EMD ID 41939 (45, 46) and tetramer PDB ID 8U61/EMD ID 41940 (47, 48).

Supporting Information

Preprint: bioRxiv DOI: https://doi.org/10.1101/2023.09.19.558089.

This article is a PNAS Direct Submission.
==== Refs
1 M. Berti, D. Cortez, M. Lopes, The plasticity of DNA replication forks in response to clinically relevant genotoxic stress. Nat. Rev. Mol. Cell Biol. 21 , 633–651 (2020).32612242
2 D. Cortez, Replication-coupled DNA repair. Mol. Cell 74 , 866–876 (2019).31173722
3 R. Zellweger , Rad51-mediated replication fork reversal is a global response to genotoxic treatments in human cells. J. Cell Biol. 208 , 563–579 (2015).25733714
4 Y. Hashimoto, A. R. Chaudhuri, M. Lopes, V. Costanzo, Rad51 protects nascent DNA from Mre11-dependent degradation and promotes continuous DNA synthesis. Nat. Struct. Mol. Biol. 17 , 1305–1311 (2010).20935632
5 K. Schlacher , Double-strand break repair-independent role for BRCA2 in blocking stalled replication fork degradation by MRE11. Cell 145 , 529–542 (2011).21565612
6 S. Sarbajna, S. C. West, Holliday junction processing enzymes as guardians of genome stability. Trends Biochem. Sci. 39 , 409–419 (2014).25131815
7 A. M. Kolinjivadi , Moonlighting at replication forks–A new life for homologous recombination proteins BRCA1, BRCA2 and RAD51. FEBS Lett. 591 , 1083–1100 (2017).28079255
8 M. R. Sullivan, K. A. Bernstein, RAD-ical new insights into RAD51 regulation. Genes (Basel) 9 , 629 (2018).30551670
9 M. R. G. Taylor , A polar and nucleotide-dependent mechanism of action for RAD51 paralogs in RAD51 filament remodeling. Mol. Cell 64 , 926–939 (2016).27867009
10 H. Dungrawala , RADX promotes genome stability and modulates chemosensitivity by regulating RAD51 at replication forks. Mol. Cell 67 , 374–386.e5 (2017).28735897
11 K. P. Bhat , RADX modulates RAD51 activity to control replication fork protection. Cell Rep. 24 , 538–545 (2018).30021152
12 M. B. Adolph , RADX controls RAD51 filament dynamics to regulate replication fork stability. Mol. Cell 81 , 1074–1083.e5 (2021).33453169
13 L. Schubert , RADX interacts with single-stranded DNA to promote replication fork stability. EMBO Rep. 18 , 1991–2003 (2017).29021206
14 H. Zhang, J. M. Schaub, I. J. Finkelstein, RADX condenses single-stranded DNA to antagonize RAD51 loading. Nucleic Acids Res. 48 , 7834–7843 (2020).32621611
15 T. Mohamed, M. B. Adolph, D. Cortez, Oligomerization of DNA replication regulatory protein RADX is essential to maintain replication fork stability. J. Biol. Chem. 298 , 101672 (2022).35120927
16 D. L. Theobald, R. M. Mitton-Fry, D. S. Wuttke, Nucleic acid recognition by OB-fold proteins. Annu. Rev. Biophys. Biomol. Struct. 32 , 115–133 (2003).12598368
17 L. Holm, DALI and the persistence of protein shape. Protein Sci. 29 , 128–140 (2020).31606894
18 M. v. Kempen , Fast and accurate protein structure search with Foldseek. Nat. Biotechnol. 42 , 243–246 (2024).37156916
19 E. Krissinel, K. Henrick, Inference of macromolecular assemblies from crystalline state. J. Mol. Biol. 372 , 774–797 (2007).17681537
20 C. A. Brosey , Functional dynamics in replication protein A DNA binding and protein recruitment domains. Structure 23 , 1028–1038 (2015).26004442
21 B. Yariv , Using evolutionary data to make sense of macromolecules with a “face-lifted” ConSurf. Protein Sci. 32 , 1–12 (2023).
22 W. Zhao , Promotion of BRCA2-dependent homologous recombination by DSS1 via RPA targeting and DNA mimicry. Mol. Cell 59 , 176–187 (2015).26145171
23 G. Kaur , Allosteric autoregulation of DNA binding via a DNA-mimicking protein domain: A biophysical study of ZNF410–DNA interaction using small angle X-ray scattering. Nucleic Acids Res. 51 , 1674–1686 (2023).36660822
24 T. Shahid , Structure and mechanism of action of the BRCA2 breast cancer tumor suppressor. Nat. Struct. Mol. Biol. 21 , 962–968 (2014).25282148
25 J. Xu , Cryo-EM structures of human RAD51 recombinase filaments during catalysis of DNA-strand exchange. Nat. Struct. Mol. Biol. 24 , 40–46 (2017).27941862
26 C. Dominguez, R. Boelens, A. M. J. J. Bonvin, HADDOCK: A protein−protein docking approach based on biochemical or biophysical information. J. Am. Chem. Soc. 125 , 1731–1737 (2003).12580598
27 S.-C. Luo , A RAD51–ADP double filament structure unveils the mechanism of filament dynamics in homologous recombination. Nat. Commun. 14 , 4993 (2023).37591853
28 M. Spírek , Human RAD51 rapidly forms intrinsically dynamic nucleoprotein filaments modulated by nucleotide binding state. Nucleic Acids Res. 46 , 3967–3980 (2018).29481689
29 J. P. Menetski, D. G. Bear, S. C. Kowalczykowski, Stable DNA heteroduplex formation catalyzed by the Escherichia coli RecA protein in the absence of ATP hydrolysis. Proc. Natl. Acad. Sci. U.S.A. 87 , 21–25 (1990).2404275
30 M. T. J. van Loenhout, T. van der Heijden, R. Kanaar, C. Wyman, C. Dekker, Dynamics of RecA filaments on single-stranded DNA. Nucleic Acids Res. 37 , 4089–4099 (2009).19429893
31 P. Baumann, S. C. West, The human Rad51 protein: Polarity of strand transfer and stimulation by hRP-A. EMBO J. 16 , 5198–5206 (1997).9311980
32 J. C. Register, J. Griffith, The direction of RecA protein assembly onto single strand DNA is the same as the direction of strand assimilation during strand exchange. J. Biol. Chem. 260 , 12308–12312 (1985).3900072
33 C. Joo , Real-time observation of RecA filament dynamics with single monomer resolution. Cell 126 , 515–527 (2006).16901785
34 J. C. Drees, S. L. Lusetti, S. Chitteni-Pattu, R. B. Inman, M. M. Cox, A RecA filament capping mechanism for RecX protein. Mol. Cell 15 , 789–798 (2004).15350222
35 R. Venkatesh , RecX protein abrogates ATP hydrolysis and strand exchange promoted by RecA: Insights into negative regulation of homologous recombination. Proc. Natl. Acad. Sci. U.S.A. 99 , 12091–12096 (2002).12218174
36 M. Berti , Sequential role of RAD51 paralog complexes in replication fork remodeling and restart. Nat. Commun. 11 , 3531 (2020).32669601
37 A. Candelli , Visualization and quantification of nascent RAD51 filament formation at single-monomer resolution. Proc. Natl. Acad. Sci. U.S.A. 111 , 15090–15095 (2014).25288749
38 G. Young , Quantitative mass imaging of single molecules HHS Public Access. Science 360 , 423–427 (2018).29700264
39 S. Classen , Implementation and performance of SIBYLS: A dual endstation small-angle X-ray scattering and macromolecular crystallography beamline at the Advanced Light Source. J. Appl. Crystallogr. 46 , 1–13 (2013).23396808
40 A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, CryoSPARC: Algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14 , 290–296 (2017).28165473
41 D. Liebschner , Macromolecular structure determination using X-rays, neutrons and electrons: Recent developments in Phenix. Acta Crystallogr. D Struct. Biol. 75 , 861–877 (2019).31588918
42 P. Emsley, B. Lohkamp, W. G. Scott, K. Cowtan, Features and development of Coot. Acta Crystallogr. D Biol. Crystallogr. 66 , 486–501 (2010).20383002
43 S. F. Altschul, W. Gish, W. Miller, E. W. Myers, D. J. Lipman, Basic local alignment search tool. J. Mol. Biol. 215 , 403–410 (1990).2231712
44 F. Sievers , Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Mol. Syst. Biol. 7 , 539 (2011).21988835
45 S. Balakrishnan, W. J. Chazin, Human RADX trimer bound to ssDNA. PDB. 10.2210/pdb8u5y/pdb. Deposited 13 September 2023.
46 S. Balakrishnan, W. J. Chazin, Human RADX trimer bound to ssDNA. EMDB. https://www.ebi.ac.uk/emdb/EMD-41939. Deposited 13 September 2023.
47 S. Balakrishnan, W. J. Chazin, Human RADX tetramer bound to ssDNA. PDB. 10.2210/pdb8u61/pdb. Deposited 13 September 2023.
48 S. Balakrishnan, W. J. Chazin, Human RADX tetramer bound to ssDNA. EMDB. https://www.ebi.ac.uk/emdb/EMD-41940. Deposited 13 September 2023.
