BRCA2 BRC motifs bind RAD51–DNA filaments

Abstract

Germ-line mutations in BRCA2 account for approximately half the cases of autosomal dominant familial breast cancers. BRCA2 has been shown to interact directly with RAD51, an essential component of the cellular machinery for homologous recombination and the maintenance of genome stability. Interactions between BRCA2 and RAD51 take place by means of the conserved BRC repeat regions of BRCA2. Previously, it was shown that peptides corresponding to BRC3 or BRC4 bind RAD51 monomers and block RAD51–DNA filament formation. In this work, we further analyze these interactions and find that at lower molar ratios BRC3 or BRC4 actually bind and form stable complexes with RAD51–DNA nucleoprotein filaments. Only at high concentrations of the BRC repeats are filaments disrupted. The specific protein–protein contacts occur in the RAD51 filament by means of the N-terminal domain of RAD51 for BRC3 and the nucleotide-binding core of RAD51 for BRC4. These observations show that the BRC repeats bind distinct regions of RAD51 and are nonequivalent in their mode of interaction. The results provide insight into why mutation in just one of the eight BRC repeats would affect the way that BRCA2 protein interacts with the RAD51 filament. Disruption of a single RAD51 interaction site, one of several simultaneous interactions occurring throughout the BRC repeat-containing exon 11 of BRCA2, might modulate the ability of RAD51 to promote recombinational repair and lead to an increased risk of breast cancer.

Mutations in BRCA2 are associated with an increased risk of breast and ovarian cancers, and almost half the cases of inherited early onset breast cancers have been linked to mutations in BRCA2. The product of this gene, BRCA2 protein (384 kDa), interacts with RAD51, and it has been shown that both proteins are essential for homologous recombination, DNA repair, and the maintenance of genome stability. There appear to be a number of different interactions between BRCA2 and RAD51. One such interaction involves a C-terminal region of BRCA2 (1). In addition, BRCA2 interacts with RAD51 through the eight conserved BRC repeats in BRCA2 (2, 3), and mutations within these repeats are associated with cancer predisposition. The deletion of several BRC repeats in mice leads to cancer (4), and somatic mutations in BRC repeats have been found to be associated with breast cancer (5). Although there are eight BRC repeats in human BRCA2, in other organisms the number of BRC repeats is quite variable, with 15 found in a Trypanosoma BRCA2-like protein (6) and only one found in the Ustilago Brh2 protein (7).

Many details of the interactions between BRCA2 and RAD51 are unclear, such as whether BRCA2 binds to the N-terminal domain of RAD51 (8) or to the nucleotide-binding core (3, 9). Recently, a fusion protein containing the nucleotide-binding core of RAD51 and a single BRC repeat was crystallized. Analysis of the x-ray structure of the fusion led to the proposal that the BRC repeat mimics a RAD51 oligomerization motif and thereby blocks RAD51–DNA nucleoprotein filament formation (9, 10). However, other seemingly contradictory studies have suggested that BRCA2 may play a role in the nucleation of RAD51 filaments (11, 12).

We show here that peptides corresponding to the BRC3 and BRC4 repeats of BRCA2 bind to RAD51–DNA filaments. The binding of BRC3 is with the N-terminal domain of RAD51, a region of RAD51 that was absent from the crystal structure. These observations indicate how BRCA2 might interact with monomers and filaments of RAD51 in different manners. Because the BRC RAD51-binding module can interact with filamentous RAD51, and not just monomeric RAD51, our observations suggest why mutations in only one of the eight BRC repeats can lead to an increased risk of breast cancer. In the filamentous binding mode, the proper interaction of all of the BRC repeats with RAD51 may be crucial for the BRCA2-mediated nucleation of a RAD51 filament at a site of DNA damage.

Methods

Proteins and Peptides. Human RAD51 (13), the BRC3 and BRC4 peptides (10), and the GST fusion protein B2-4 (14) were prepared essentially as described. The peptides were resuspended in buffer A (30 mM Tris·HCl, pH 8.0/10% glycerol/0.5 mM EDTA/0.5 mM DTT/0.2 M KCl).

Gel Analysis. RAD51 protein (1 μM) was preincubated with the indicated amounts of BRC3 peptide for 15 min at 37°C in buffer A. The mixture then was supplemented with binding buffer [50 mM triethanolamine, pH 7.5/0.5 mM Mg(OAc)₂/2 mM ATP/1 mM DTT/100 μg/ml BSA] followed by AvaI-linearized ³²P-end-labeled ϕX174 duplex DNA (2 μM in nucleotides). The mixture then was incubated for 60 min at 37°C. In some reactions, ATP was replaced by adenosine 5′-[β,γ-imido]triphosphate (AMP-PNP) (0.5 mM). The products were analyzed by electrophoresis without fixation through 0.5% agarose [5 h at 4°C in TAE buffer (40 mM Tris·acetate/1 mM EDTA, pH 8.3)]. Labeled products were visualized by autoradiography.

Electron Microscopy (EM) and Image Analysis. The RAD51–dsDNA–A MP-PNP complex was formed in 25 mM triethanolamine·HCl buffer (pH 7.2) and incubated at 37°C for 15 min, with 7 μM hRAD51 [a RAD51 to DNA ratio of 40:1 (wt/wt)], 1.25 mM AMP-PNP (Sigma), and 2.5 mM magnesium acetate. Sheared calf thymus dsDNA (Sigma) was used for all complexes. Then, BRC3 (7 μM), BRC4 (7 μM), or B2-4 (3.5 μM) was added and further incubated at 37°C for 15 min. Additional experiments were conducted by preincubating BRCA2 fragments before filament formation, but no differences were seen. Samples were applied to carbon-coated grids and stained with 2% uranyl acetate (wt/vol). Images were recorded on film using a Tecnai 12 electron microscope (FEI, Hilsboro, OR) operating at 80 keV with a nominal magnification of ×30,000. Negatives were scanned with a Leaf 45 densitometer (Scitex, Tel Aviv), at a raster of 3.9 Å/pixel. Segments of filaments were extracted into 60 × 60-pixel boxes. Model volumes having a pitch from 75 to 125 Å, in steps of 5 Å, were used for sorting the segments by pitch. The volumes were azimuthally rotated in steps of 9° to generate 40 projections from different orientations of each model volume. Segments of pure RAD51–DNA filaments (n = 12,448), RAD51–DNA–B2-4 (n = 14,902), RAD51–DNA–BRC4 (n = 16,719), and RAD51–DNA–BRC3 (n = 17,826) were cross-correlated with the 440 projections of these reference volumes to assign a pitch to each segment. Subsequent sorting, when used, involved selecting segments having the greatest projected density at a particular radius from the helical axis. As a control, similar experiments were conducted with nucleoprotein filaments formed by the Escherichia coli RecA protein. No specific binding of BRC3 to RecA was found.

The reconstruction of a control RAD51–DNA–AMP-PNP filament (see Fig. 3a) involved 2,000 segments, and had 6.4 subunits per turn of a 100 Å pitch helix. The BRC3–RAD51–DNA–AMP-PNP filament reconstruction (Fig. 3b) was first generated by selecting a subset of segments with a pitch from 120 to 125 Å (n = 3,161) and then sorting these by both twist (number of subunits per turn) and projected density. The class shown (n = 350) had 6.4 subunits per turn of a 125-Å pitch helix. The B2-4 reconstructions (Fig. 3c) were generated by a similar method, and the class shown (n = 436 for Fig. 3c) had 6.3 subunits per turn of a 130-Å pitch helix.

Fig. 3.

Three-dimensional reconstructions of BRC3 (b) and B2-4 (c) peptides binding to RAD51–DNA filaments, with a pure RAD51–DNA filament reconstruction (a) shown as a control. The crystal structure of an intact ScRad51 subunit (17) is fit into the filaments. In the presence of BRC3 (b) or B2-4 (c), additional mass can be seen (arrows) associated with the N-terminal domain. A portion of the α-helix containing residues 146–156 in ScRad51 projects outside of the reconstructions (a, blue arrow), consistent with the suggestion that this region might be flexible in the filament (17). The concentrations used for these reconstructions were [RAD51] = 7 μM, [BRC3] = 7 μM(b), and [B2-4] = 3.5 μM (c).

The BRC4–RAD51 filament reconstruction in Fig. 4c was generated by first grouping together segments having a pitch from 85 to 100 Å (n = 8,138) and then sorting these segments by cross-correlation against references having a regular or disordered N-terminal domain. These segments were subsequently sorted by twist. The volume shown in Fig. 4c (n = 1,064) had 6.6 subunits per turn of a 92-Å pitch helix. The reconstruction in Fig. 4d (n = 1,044) involved only sorting by pitch and had 6.4 subunits per turn of a 129-Å pitch helix. The reconstruction in Fig. 4d was made by using segments selected as having a pitch of 125 Å and converged to 130-Å pitch.

Fig. 4.

Reconstructions of RAD51–DNA filaments ([RAD51] = 7 μM) in the presence of the BRC4 peptide ([BRC4] = 7 μM) (c and d), a control reconstruction of a pure RAD51–DNA filament (b), and a crystal structure of a yeast Rad51 filament (a) (17). The two reconstructions in the presence of BRC4 (c and d) were generated by sorting filament segments into different classes based on similarity. The main difference between these reconstructions involves RAD51's N-terminal domain, which is not seen in c, and shifted considerably in d from the position seen in the control filament (b). In b and d the RAD51 ATP-binding core is shown in magenta, and the N-terminal domain is in red. The rotated position of an N-terminal domain (19), used to fit a compressed RadA filament reconstruction (18), is shown in green (b). The BRC4 crystal fragment (9) is shown in blue (c and d).

Results

In previous studies, it was shown that peptides corresponding to BRC3 or BRC4 interact with RAD51 monomers to prevent nucleoprotein filament formation (10). In these experiments, RAD51–DNA complexes were analyzed by electrophoresis after fixation by treatment with glutaraldehyde. To further analyze interactions between BRC repeats and RAD51 and the effect that these interactions have on RAD51–DNA nucleoprotein filament formation in the presence of ATP, we have now analyzed these interactions by using electrophoretic mobility shift assays of unfixed complexes (Fig. 1a). We found that a 69-aa peptide corresponding to BRC3 (BRCA2 residues 1415–1483) formed a stable complex with RAD51–ATP–DNA filaments, seen as a supershift when analyzed by agarose gel electrophoresis (Fig. 1a, lanes f and g). The peak of this supershift occurs at a molar ratio of ≈5:1 (BRC3:RAD51), and it is only at higher molar ratios (≈8:1) that the RAD51–DNA filaments are eliminated. The hydrolysis of ATP by RAD51–DNA filaments will lead to subunit turnover within these filaments and dynamic structural changes that result from the ATPase cycle. We therefore looked at conditions where more stable structures are formed. In contrast to what was observed with ATP, preincubation of an 8:1 molar ratio of the BRC3 peptide with RAD51 did not inhibit filament formation on DNA in the presence of the nonhydrolyzable ATP analog AMP-PNP (Fig. 1b, lane j). In the presence of AMP-PNP, the supershift of the RAD51–DNA filaments saturates at a molar ratio of ≈7:1 (BRC3 to RAD51).

Fig. 1.

Effect of BRC peptides on RAD51–DNA nucleoprotein filament formation. BRC3 (concentrations indicated) and RAD51 (1 μM) were incubated together before the addition of ³²P-labeled linear dsDNA by using either ATP (a) or AMP-PNP (b and c). Various BRC peptides or mutant derivatives of BRC3 were incubated with RAD51, before addition of the labeled DNA, in the presence of AMP-PNP After 60-min incubation at 37°C, protein–DNA complexes were analyzed by agarose gel electrophoresis followed by autoradiography.

We tested the ability of peptides containing mutations within BRC3, as well as other BRC repeats, to bind to RAD51–DNA filaments. The D1420Y mutation within the BRC3 region has been found in many cases of familial breast cancer, and a peptide containing this mutation was shown to block RAD51–ATP–DNA filament formation (10). However, the BRC3 D1420Y peptide containing this mutation retained the ability to bind RAD51–AMP-PNP–DNA filaments (Fig. 1c, lanes e and f) in a similar manner as the wild-type peptide. In contrast, a T1430A mutation within the BRC3 peptide was shown to eliminate the activity of BRC3 in blocking RAD51–ATP–DNA filament formation (10), and we also found that the peptide containing this mutation failed to form stable complexes with the RAD51–AMP-PNP–DNA filaments (Fig. 1c, lanes g and h).

Like BRC3, a BRC4 peptide (BRCA2 residues 1511–1579) was found to interact with RAD51–DNA filaments at a 4:1 molar ratio (Fig. 1c, lane i) but inhibited RAD51–AMP-PNP–DNA filament formation when present at a molar ratio of 8:1 (Fig. 1c, lane j). In contrast, the BRC3 peptide did not inhibit nucleoprotein filament formation under similar conditions (Fig. 1c, lane d). The BRC7 peptide, which only partially reduced the ability of RAD51 to form nucleoprotein filaments in the presence of ATP (10), fails to bind RAD51–AMP-PNP–DNA filaments under these reaction conditions (Fig. 1c, lanes k and l).

We interpret these data as arising from two different sites of interaction between the BRC repeats and RAD51. One interaction occurs between the filamentous form of RAD51 and the BRC repeat, whereas a second interaction occurs between monomeric RAD51 and the BRC repeat. It is the second interaction that prevents RAD51 filament formation. The DNA supershift, arising from the binding of the BRC repeats to a RAD51–DNA filament, can only be seen in the absence of fixation because the presence of glutaraldehyde fixes large aggregates containing many RAD51–DNA filaments. Because these aggregates are so massive, a supershift cannot be seen when the BRC peptides bind to the filaments.

As suggested by the gel supershift in the absence of fixation, electron micrographs show that RAD51–DNA filaments (Fig. 2a) look different after incubation with the BRC3 peptide (Fig. 2b). The main difference appears to be that the modulation of density due to the helical groove is diminished. The difference in morphology between RAD51–DNA filaments in the presence or absence of BRC3 did not depend on whether the BRC3 peptide is added before or after the formation of RAD51 filaments on DNA (data not shown). When the BRC3 peptide is present at lower ratios (1:3 and 1:6, BRC3:RAD51), we do not see distinct patches within filaments, which would occur if there were a highly cooperative binding. Such a cooperative mode of binding was shown for very low ratios of myosin to actin, where small segments of heavy decoration were seen surrounded by largely naked actin (15).

Fig. 2.

EM shows morphological differences in RAD51–DNA filaments after incubation with BRC3. Shown are electron micrographs of a control RAD51–AMP-PNP–DNA filament ([RAD51] = 7 μM) (a) and BRC3–RAD51–AMP-PNP–DNA filaments ([RAD51] = 7 μM, [BRC3] = 7 μM) (b). The strong striations in the RAD51–DNA filament (a) correspond to the helical pitch of this filament (≈75–130 Å). The binding of the BRC3 peptide to the RAD51 filament is suggested by the great reduction in the appearance of these striations (b). (Scale bar: 1,000 Å.)

Image analysis has been used to characterize the interaction between BRC3 and the RAD51–DNA filaments. In these experiments, we used linear duplex DNA and the nonhydrolyzable ATP analog AMP-PNP. We extracted 17,826 segments (each ≈235-Å long) from the RAD51–DNA–BRC3 filaments and 12,448 segments from control RAD51–DNA filaments. Image analysis shows that the most ordered BRC3 binding occurs to RAD51 filament segments having a pitch in the region of ≈120–125 Å, which is greater than the mean pitch of pure RAD51–DNA filaments (≈100 Å). A 3D reconstruction, using the iterative helical real space reconstruction method (16), has been generated for RAD51–BRC3–DNA filament segments with this extended pitch (Fig. 3b), and can be compared with a reconstruction of pure RAD51–DNA filament segments (Fig. 3a). The reconstruction shows that the additional mass due to BRC3 in the RAD51–DNA filament is associated with the N-terminal domain of RAD51 (Fig. 3b, arrow) rather than with the nucleotide-binding core of RAD51.

Could the extra mass seen at the N-terminal domain actually be due to a conformational change in RAD51, with BRC3 bound elsewhere? This possibility would not appear to be reasonable, because the crystal structure of a RAD51 subunit can be fit into filaments that either have BRC3 bound (Fig. 3b) or do not have BRC3 bound (Fig. 3a). In addition, the excellent agreement between a crystal filament of a Rad51 protein (17) and our EM reconstruction (Fig. 4 a and b) shows that the assignment of the N-terminal domain is unambiguous. The position of the additional mass due to BRC3 provides a simple explanation for why the most regular binding is found in segments having a greater than normal pitch: with a smaller pitch, steric hindrance would exist with subunits in the turn below, and the BRC3–N-terminal domain complex might be forced into multiple conformations to eliminate the steric clashes.

We also analyzed the interaction of a recombinant GST–BRCA2 fusion protein (amino acids 1338–1781, designated B2-4) containing the repeats BRC3, BRC4, and BRC5 (14). We found that the B2-4 protein also bound to the N-terminal domain of the RAD51 filament (Fig. 3c, arrow). However, with this large protein containing an N-terminal glutathione S-transferase tag, we failed to observe most of its mass, and only the portion proximal to RAD51 was visualized. However, more mass was seen associated with RAD51's N-terminal domain when GST–B2-4 was bound (Fig. 3c) than when BRC3 was bound (Fig. 3b), consistent with the larger mass of this protein. Again, the most regular binding of this fragment to RAD51 was observed for filaments having the longest pitch. The reconstruction shown (Fig. 3c) was generated by sorting filament segments based on pitch and has a pitch of ≈130 Å.

The BRC4 peptide, examined in the crystal complex with the RAD51 nucleotide-binding core (9), also binds to RAD51–DNA filaments, as judged by the band shift analysis (Fig. 1c, lane i) and by the shift in the pitch distribution from pure RAD51–DNA filaments after incubation with this peptide (data not shown). We analyzed 16,719 segments extracted from RAD51–AMP-PNP–DNA filaments incubated with the BRC4 peptide. Because of the heterogeneity in the segments arising from variable pitch, twist, binding of BRC4, etc., segments could be sorted in different ways to generate more homogeneous subsets. Reconstructions from two such subsets are shown in Fig. 4 c and d. One subset, Fig. 4c, did not display density due to the N-terminal domain. We interpret this result as due to extensive disorder in this domain, so that after averaging (both within segments and between segments) this mass is no longer visualized. Such disorder of the N-terminal domain already has been described for an archaeal Sulfolobus solfataricus RadA filament (a homolog of RAD51), where this domain was visualized in filaments prepared with adenosine 5′-[γ-thio]triphosphate but absent in filaments prepared with ATP and aluminum fluoride (18). In addition, a similar disordering of the N-terminal domain was seen in a crystal structure of the archaeal Pyrococcus furiosus RadA protein (PfRad51), where in a heptameric ring only one of the seven N-terminal domains was resolved (19). In a crystal structure of the homologous human Dmc1, no N-terminal domains are seen within an octameric ring of this protein because of substantial disorder (20).

The crystal structure of the RAD51 nucleotide-binding core fused with a BRC4 peptide (9) can be fit quite well into the density shown in Fig. 4c. The RAD51 N-terminal domain was not present in this crystal structure, and it was subsequently shown that the N-terminal domain seen in an intact PfRad51 subunit would sterically clash with the BRC4 peptide as it was observed bound to the core (19). We suggest, therefore, that the binding of the BRC4 peptide to the filament induces large shifts of the N-terminal domain, so that it is not visualized in the reconstruction of Fig. 4c. Consistent with this result, the other subset shown (Fig. 4d) displays a large displacement of the N-terminal domain away from the position seen in the control filament (Fig. 4b). The shift of the N-terminal domain visualized in Fig. 4d would remove any steric clashes between the N-terminal domain and BRC4. A rotation of the N-terminal domain by ≈60° and translation by ≈25 Å (Fig. 4b) was previously used to fit a crystal structure of RadA into a filament reconstruction (19). Random shifts and rotations of such a magnitude would be sufficient to eliminate the density due to the N-terminal domain in an averaged 3D reconstruction (e.g., subset shown in Fig. 4c). As discussed below, the disordering and displacements of the N-terminal domain by BRC4 explain the observation that this peptide can inhibit RAD51 nucleoprotein filament formation in the presence of AMP-PNP, whereas the BRC3 peptide cannot inhibit such filament formation under similar conditions (Fig. 1c).

If the BRC3 and BRC4 peptides bind RAD51 differently, one would not expect simple competition if both are added to RAD51. We have confirmed that this hypothesis is the case. When one adds a large molar excess of BRC4 to RAD51 filaments formed on DNA in the presence of ATP ([RAD51] = 5 μM, [BRC4] = 50 μM), very few filaments are seen by EM, in contrast to the abundant filaments seen in the absence of BRC4 (data not shown). If the RAD51–DNA–ATP filaments are first incubated with BRC3 ([RAD51] = 5 μM, [BRC3] = 30 μM), and then a large excess of BRC4 is added ([BRC4] = 50 μM), abundant filaments are seen. It was shown that the position of the N-terminal domain in a RadA crystal (19) was incompatible with the binding of BRC4 seen in the crystal structure of a RAD51 fragment (9). We interpret our result as showing that when BRC3 is bound to the N-terminal domain, this domain is stabilized, and the stabilized N-terminal domain sterically blocks BRC4 from binding to the core of RAD51.

Discussion

We have directly observed the binding of three BRCA2 fragments (BRC3, BRC4, and B2-4) to RAD51–DNA filaments. These results therefore provide a description of a complex between RAD51–DNA filaments and BRCA2 fragments and contrasts with previous reports suggesting that the association between BRCA2 and RAD51 can only prevent RAD51 filament formation on DNA. A direct interaction between BRCA2 and RAD51–DNA filaments helps resolve a crucial paradox: why do BRCA2 fragments inhibit RAD51 filament formation, whereas BRCA2 is known to be required for the association of RAD51 at sites of DNA damage in situ (11)? A specific role for a BRCA2 homolog, Brh2, in nucleating RAD51 filament formation has now been shown (12). A more general role for BRCA2-like proteins has been suggested by the observation in Arabidopsis that the Brca2 protein interacts with both Rad51 and Dmc1 and that Brca2 within Arabidopsis during meiosis might be targeting Dmc1 to sites of double strand breaks (21). Because we have shown that the RAD51-binding modules of BRCA2 can interact with a RAD51–DNA filament, we suggest that the biologically important interaction may be between BRCA2 and RAD51 subunits that are assembling on DNA at the site of DNA damage. RAD51 foci form normally during S-phase in BRCA2-defective CAPAN-1 cells (22), so it is apparent that BRCA2 is not necessary for RAD51 filament formation per se. The role of BRCA2 therefore may involve stabilization of filaments and regulation of the interaction between RAD51 filaments and other proteins.

In this context, an interesting parallel may exist between BRCA2 and the bacterial DinI protein. Although DinI has been shown to destabilize and inhibit RecA filaments when present at very high concentrations (superstoichiometric with respect to RecA concentrations) (23–25), DinI actually stabilizes RecA filaments when present at stoichiometric concentrations (26). A further parallel might exist because the C-terminal domain of RecA has been shown to be involved in the interaction with DinI (26), whereas we have shown that the N-terminal domain of RAD51 is the primary site of interaction with BRC3 and B2-4. Both domains form pendulous lobes on the corresponding filaments, and both domains also have been shown to bind DNA (27, 28).

Although we have not observed a role for the BRCA2 fragments in nucleating RAD51 filament formation on DNA (due to the fact that under the in vitro conditions we are using, RAD51 filaments nucleate and polymerize on DNA by themselves quite efficiently), a comparison with other eukaryotic protein polymers is instructive. Actin, typically the single most abundant protein in eukaryotic cells, can nucleate itself and polymerize in vitro. However, a large number of proteins are present in vivo that function to either nucleate or depolymerize actin filaments (29). As far as we understand, no actin filaments are formed within the cell by spontaneous self-nucleation. The fact that actin-nucleating proteins, such as the bacterial virulence protein SipA (30), also bind to preformed actin filaments in vitro is consistent with a role for BRCA2 functioning in either RAD51 filament nucleation, stabilization, or regulation.

The role of the previously described interactions between BRCA2 fragments and RAD51 monomers that block RAD51 filament assembly on DNA remain unclear. It is possible that these in vitro interactions do not reflect the complexes that form in the cell, because they are only found when there is a large excess of BRCA2 fragments over RAD51 (10) or when a BRCA2 fragment is covalently fused to the RAD51 core (9). Conversely, it is possible that BRCA2 could play a dual role in both polymerizing and depolymerizing RAD51 filaments. Again, a comparison with actin-binding proteins may be informative, because proteins in the actin depolymerizing factor (ADF)/cofilin family (31, 32) appear to play a role in both nucleating and depolymerizing actin filaments. The actin-binding protein gel-solin has been known to play a role in very different activities, involving severing, capping, and nucleating actin filaments (33).

Our observations complement previous studies that have shown nonequivalent interactions between the different BRC repeats in BRCA2 and RAD51. For example, the BRC5 and BRC6 repeats were not observed to interact with RAD51, as judged by yeast two-hybrid experiments (3). Similarly, the BRC7 repeat was much less efficient at blocking RAD51–DNA filament formation than the BRC3 and BRC4 repeats (10). We have observed that the BRC3 and BRC4 repeats behave differently with respect to their interactions with RAD51–DNA filaments: the main interaction of the BRC3 repeat is with the N-terminal domain of RAD51, whereas the main interaction of the BRC4 repeat appears to be with the nucleotide-binding core of RAD51. At high concentrations, BRC3 may bind to a lower affinity site on the nucleotide-binding core and block RAD51 filament formation.

The eight nonequivalent BRC repeats in BRCA2 also may suggest a parallel with several proteins that bind actin polymers, each containing multiple repeats of an actin-binding module. A tropomyosin molecule, for example, has seven actin-binding repeats. Tropomyosin plays a role in both regulation of the acto–myosin interaction in muscle (34), as well as in stabilization of actin filaments (35, 36). A mutational analysis of the actin-binding repeats in tropomyosin concluded that these repeats “are important for specific functions and are not quasiequivalent” (37), just as the BRC repeats in BRCA2 have been shown to be nonequivalent. The giant muscle protein nebulin has ≈200 repeats of a 35-residue actin-binding motif. It has been shown that the interaction of these modules with F-actin involves both high- and low-affinity binding sites on nebulin (38).

If all BRC repeats bound RAD51 monomers as part of a “sequestering” mechanism that prevented RAD51 filament assembly, it would be difficult to explain the mutational data that indicates that a point mutation within a single BRC repeat can lead to a greatly increased susceptibility to cancer. Conversely, if multiple BRC repeats are binding multiple RAD51 protomers within a nascent RAD51–DNA filament, improper contacts within a single such interaction could play a dominant negative role by disrupting the entire complex of BRCA2 with a RAD51–DNA filament. We therefore suggest that direct interactions between BRCA2 and RAD51 filaments, rather than RAD51 monomers, provide the simplest interpretation of the mutational data.

Comparison with a Previous Interpretation

The binding of BRC4 to the RAD51 core within a RAD51–DNA filament does raise a complication. If the binding occurs as seen in the crystal structure of the RAD51–BRC4 fusion protein (9), then the “polymerization motif” that was suggested is not taking part in the subunit–subunit interface within these filaments when BRC4 is bound. It was suggested that the BRC repeats prevent polymerization of RAD51–DNA filaments by mimicking a “highly conserved” oligomerization motif found in RecA, RadA, and RAD51 proteins (9). However, quantitative analyses fail to find any evidence for homology within this region between RecA, on the one hand, and BRCA2 and RAD51-like proteins, on the other (39, 40). This small oligomerization motif is part of the subunit–subunit interface in a crystal filament of RecA, formed in the absence of DNA and ATP (41). Because this region of RAD51 was not present in the first RAD51 crystal structure (9), it was argued that there was a structural homology between this region of BRCA2 (residues 1,523–1,529) and human RAD51 (residues 85–91) based on a similarity with the RecA crystal structure in this region (residues 25–31). The crystal structures of the archaeal Pyrococcus furiosus RadA (19), the archaeal Methanococcus voltae RadA (42), and ScRad51 (17) both support and undercut this argument of structural homology, because a portion of this region is an α-helix in RadA and ScRad51, whereas it is a coil and very short β-strand in both RecA and BRC4. Nevertheless, it is likely that this short motif is part of the subunit–subunit interface in both RAD51 and RecA filaments, and the binding of the BRC repeats to this site blocks filament formation. In fact, a portion of this motif was part of the subunit–subunit interface in an EM-based model for the active RecA–DNA filament (43). This model for a RecA–DNA–ATP filament differed from the filament seen in a crystal of RecA alone (in the absence of both DNA and ATP) by a rotation of subunits through ≈30°. A crystal structure of a yeast Rad51 filament (17) and an archaeal RadA filament (42) show a similar rotation of subunits as that predicted for the active RecA–DNA filament, even though the oligomerization motif is present as a small part of the subunit–subunit interface. This finding shows a large plasticity of this overall interface and suggests that the small oligomerization motif alone may not determine the remainder of the interface.

Our results with the BRC4 peptide in the presence of AMP-PNP provide partial support for the existing picture of how BRC4 binds to RAD51. It was shown that BRC4 cannot bind to the full-length PfRad51 subunit conformation seen in a crystal because of steric clashes with PfRad51's N-terminal domain (19). This N-terminal domain was missing in the RAD51–BRC4 crystal structure (9) and therefore could not provide any steric interference in that complex. We observed that when the BRC4 peptide is bound to preexisting RAD51–AMP-PNP–DNA filaments, the N-terminal domain of RAD51 is either disordered or significantly shifted. This binding of BRC4 to filaments also is seen by the gel shift (Fig. 1c, lane i). In contrast, we suggest that the BRC3 peptide under similar conditions cannot access the core-binding site because the N-terminal domain is providing steric hindrance and does not interfere with filament formation in the presence of AMP-PNP. When ATP is used rather than AMP-PNP (Fig. 1a), the N-terminal domain becomes dynamic because of the hydrolysis of ATP and existence of different nucleotide states (i.e., ATP, ADP–Pi, and ADP), and both BRC3 and BRC4 can inhibit filament formation. The dynamics of the N-terminal domain as a function of different nucleotide states has been shown for both archaeal RadA (18) and human RAD51 (44).

Because the polymerization motif would be a small part of the entire subunit–subunit interface that must exist in RAD51–DNA filaments (involving many other residues), it is possible that this part of the interface can be disrupted without depolymerizing filaments. Another possibility is that this interface can be disrupted in some subunits within a filament, but filaments are unable to assemble when the loss of this interface is stoichiometric. This effect has been seen with a modification of actin, such that actin filaments are unable to assemble when every subunit is modified but can assemble into copolymers with unmodified actin when up to 50% of the subunits are so modified (45). Alternatively, it is possible that the polymerization motif is making contacts with adjacent subunits, and a portion of BRC4 is displaced from making the contacts that were seen in the crystal. Our present results cannot distinguish among these possibilities.