RECQL5/Recql5 helicase regulates homologous recombination and suppresses tumor formation via disruption of Rad51 presynaptic filaments

Recql5-deficient mice are highly cancer prone

To examine whether Recql5 suppresses tumorigenesis, we generated large cohorts of Recql5 knockout mice and their wild-type littermates and monitored their phenotypes. No significant differences between the two groups were observed after 1 yr. However, as these animals aged further, a significantly higher percentage of Recql5 knockout mice developed cancer. Specifically, by 22 mo of age, 23 out of 50 (46%) of the knockout mice and only two out of 32 (6%) wild-type animals developed cancer (Fig. 1A–F). The only two cancer cases in the wild-type were both lymphomas. In mutant mice, approximately half of the cancer incidences were lymphomas, with late-onset follicular lymphoma, which is a common type of malignant lymphoma in humans, accounting for 36% (five out of 14) of the total lymphoma incidences. The rest of the cancers were solid tumors of different tissue origins. Intriguingly, lung adenocarcinomas predominated among them (six out 12, or 50%) (Fig. 1B,D). Three Recql5 knockout mice had two different types of primary cancers each (Fig. 1B). Interestingly, aside from the striking cancer susceptibility phenotype, Recql5 knockout mice were indistinguishable from their wild-type siblings. In particular, except for those animals that succumbed to cancer, Recql5 knockout mice had a lifespan that was not different from wild-type animals. Collectively, these data show that Recql5 is not required for normal embryonic development or for postnatal life in mice, but it has a potent tumor suppression function.

Recql5 deficiency is not associated with an elevated frequency of LOH

As Blm and Recql5 have nonredundant roles in suppressing SCEs in mitotic cells (Hu et al. 2005), our finding that Recql5 knockout mice are highly cancer prone suggests that misregulation of HR may be the underlying basis for the cancer phenotype in these animals. Mutations in both human BLM and mouse Blm result in elevated rates of LOH and engender cancer susceptibility (German 1993; Luo et al. 2000; Goss et al. 2002; Hickson 2003). Thus, we first examined whether the elevated SCEs in Recql5-deficient cells is also associated with an increased frequency of LOH. Consistent with a previous report (Luo et al. 2000), we found that Blm knockout embryonic stem (ES) cells exhibited a significant increase in LOH at the Rad50 locus compared with their wild-type counterparts (Supplementary Fig. S1A,B). In contrast, Recql5 knockout ES cells had similar frequencies of LOH as wild-type ES cells at both the Rad50 locus (Supplementary Fig. S1A,C) and the Recql5 locus (Supplementary Fig. S1D,E). This finding suggests that alteration in LOH is not the primary underlying mechanism of cancer susceptibility in Recql5 knockout mice. It also suggests that the main function of Recql5 with respect to suppressing SCEs does not lie at the dissolution of DHJ, a deduction that is consistent with our previous finding that Blm and Recql5 have nonredundant roles in suppressing SCEs in mitotic cells (Hu et al. 2005).

Recql5-deficient cells exhibit an increased frequency of Rad51 and γ-H2AX foci

HR can also be regulated at an early step—for instance, during the assembly of presynaptic filaments of the Rad51 recombinase (Sung and Klein 2006). To determine whether Recql5 functions at this stage, we examined the effect of Recql5 ablation on the kinetics of HR induced by DSBs. We reasoned that if Recql5 suppresses SCEs by affecting the channeling of DSBs into HR, deletion of Recql5 should result in an elevated frequency of HR. First, γ-H2AX and Rad51 foci were used as indicators of DSB induction and HR activation, respectively, whereas the colocalization of γ-H2AX and Rad51 foci was used to monitor HR-mediated DSB repair (Paull et al. 2000; Rothkamm et al. 2003). We found that in mouse embryonic fibroblasts (MEFs) derived from Recql5 knockout mice, the percentage of cells with >10 Rad51 foci was significantly higher than wild-type MEFs (Fig. 2A; Supplementary Fig. S2A). Unexpectedly, Recql5-deficient MEFs also had elevated frequencies of cells with >10 spontaneous γ-H2AX foci (Fig. 2B). No significant difference in the frequency of S-phase cells or tetraploid cells between the wild-type and Recql5-deficient MEFs was seen by flow cytometry analysis (data not shown), ruling out the possibility that the increased percentages of cells with >10 γ−H2AX or Rad51 foci were due to differences in the cell cycle distributions. The majority of the spontaneous γ-H2AX foci were detected in PCNA positive cells (data not shown), suggesting that they arise during DNA replication. Consistent with this hypothesis, we found that after exposure to a sublethal dose of camptothecin (CPT), a potent S-phase-specific DSB inducer (Pommier et al. 2003), the proportion of cells with Rad51 foci increased dramatically in both mutant and wild-type MEFs, with the fraction of such cells being much higher in the mutant than wild-type (Fig. 2C,D). Importantly, the CPT-induced Rad51 foci persisted much longer in the mutant cells as compared with wild-type cells (Fig. 2D). The Recql5 mutant cells also exhibited more γ-H2AX foci and for a much longer length of time than wild-type cells (Fig. 2C,E). Moreover, γ-H2AX and Rad51 foci colocalization was rarely observed in wild-type MEFs, even after CPT treatment. In comparison, there were a significantly higher percentage of mutant cells with >10 γ-H2AX foci, of which at least 80% of the foci colocalized with Rad51. Parallel experiments using ES cells showed a similar trend (Supplementary Fig. S2B). In addition, Western blotting showed that chromatin-bound Rad51 was enriched in mutant ES cells compared with the wild-type control following CPT treatment (Supplementary Fig. S3).

Recql5 suppresses DSB-induced HR

The finding that deletion of Recql5 leads to a dramatic increase in γ-H2AX foci-positive cells strongly suggests that Recql5 has a role in suppressing the accumulation of DSBs. Therefore, it seemed possible that the elevated Rad51 foci and Rad51/γ-H2AX colocalization in Recql5-deficient cells could be an indirect consequence of the increased accumulation of DSBs, rather than an inability to suppress HR. To distinguish between these two possibilities, we examined whether Recql5 deficiency affects the efficiency of HR-mediated repair of I-SceI-induced DSBs. An SCneo cassette for monitoring HR-mediated repair (Johnson et al. 1999) was introduced into the Rb locus in both wild-type and Recql5 knockout ES cells by gene targeting. Within this reporter system, a single DSB could be induced in the mutated version of a G418 resistance selection marker gene by the I-SceI endonuclease. The DSB triggers HR-mediated repair with a nonmutated fragment of the neomycin gene to generate G418-resistant clones (Fig. 3A; Johnson et al. 1999). Since Recql5 deletion does not seem to affect either HJ resolution (Supplementary Fig. S1) or nonhomologous end joining (Hu et al. 2005), differences in the proficiency of reconstructing a functional G418 resistance gene between wild-type and Recql5-deficient ES cells likely reflect differences in the channeling of DSBs into HR. The result showed that the frequency of G418-resistant colonies arising from Recql5-deficient ES cells was almost 3.5-fold higher than that from wild-type ES cells (Fig. 3B). Taken together, these data indicate that Recql5 deficiency is associated with increased DSB repair by HR.

Recql5 is required for genome stability in response to replicative stress

Both the accumulation of DSBs and unchecked inappropriate HR events that accompany Recql5 deletion can result in genome instability. To address this important issue, we examined metaphase chromosome spreads from both wild-type and Recql5-deficient ES cells. As reported previously, spontaneous GCRs are very rare in both wild-type and Recql5-deficient cells (Hu et al. 2005) and therefore difficult to quantify. Thus, to better assess the impact of Recql5 deficiency on GCRs, we compared the frequency of chromosomal aberrations between wild-type, Blm, and Recql5 mutant ES cells following exposure to a sublethal dose (50 nM) of CPT, a potent inducer of GCRs (Pommier et al. 2003). This treatment had only a modest impact on genome stability in the wild-type and Blm knockout ES cells, but it resulted in a dramatic increase in the incidence of chromosomal aberrations in the Recql5^−/− mutant cells. The aberrations included chromatid and chromosome gaps, as well as multiradial structures (Fig. 4A–D). These data provide evidence that Recql5 deficiency elevates the susceptibility for GCRs.

Human RECQL5 physically interacts with Rad51 and suppresses Rad51-mediated D-loop formation

To delineate how Recql5 regulates HR at the mechanistic level, we expressed human RECQL5 in Escherichia coli as a C-terminal fusion with a self-cleaving affinity tag composed of an Mxe-intein fragment and a chitin-binding domain (CBD) (Supplementary Fig. S4A; Garcia et al. 2004). RECQL5 was purified to near homogeneity using two chromatographic fractionation steps coupled with affinity chromatography on chitin beads (Supplementary Fig. S4B). Three independently purified RECQL5 preparations gave the same results in all the biochemical analyses.

Since human RECQL5 physically interacts with both Topo IIIα and its isoform, Topo IIIβ (Shimamoto et al. 2000), we first considered the possibility that RECQL5 helps mediate DHJ dissolution as has been shown for BLM (Wu and Hickson 2003). However, using a previously described DHJ dissolution assay (Wu and Hickson 2003), we could find no evidence of RECQL5 and Topo IIIα or Topo IIIβ being capable of dissolving the DHJ. The same negative result was obtained with the inclusion of BLAP75 in the reaction (data not shown).

We then entertained the possibility that RECQL5 regulates the activity of the human Rad51 recombinase. To test this premise, we first examined whether RECQL5 is capable of associating with Rad51. Interestingly, we found that Rad51 can be coimmunoprecipitated with RECQL5 from U2OS cell extracts (Supplementary Fig. S5A). Likewise, Rad51 in HEK293 cell extracts bound to chitin beads coated with purified RECQL5-CBD protein, but not to control beads containing E. coli McrA-CBD protein (Supplementary Fig. S5B). This interaction is direct, as purified RECQL5 and Rad51 can be efficiently coimmunoprecipitated (Fig. 5A).

A D-loop assay (Fig. 5B; Petukhova et al. 1998; Mazin et al. 2000) was used to test whether RECQL5 has an effect on the recombinase activity of Rad51. We initially used the Rad51 K133R protein, which binds ATP but is greatly attenuated for ATP hydrolysis, in the D-loop reaction. Since ATP hydrolysis prompts the dissociation of Rad51 from DNA, Rad51 K133R forms a highly stable presynaptic filament (Chi et al. 2006), and its use facilitated our analyses. Coimmunoprecipitation verified that RECQL5 has the same affinity for Rad51 K133R as Rad51 (data not shown). To enhance the efficiency of the D-loop reaction, the accessory factor Hop2-Mnd1 (Petukhova et al. 2005) was included. As expected, Rad51 K133R, and Hop2-Mnd1 catalyzed efficient D-loop formation both in the presence and absence of the ssDNA-binding factor, RPA (Fig. 5C, lanes 2,3). The addition of a catalytic quantity of RECQL5 with Rad51 K133R strongly inhibited D-loop formation when RPA was present (Fig. 5C,E). As little as 15 nM of RECQL5 suppressed the D-loop reaction by four- to fivefold. Interestingly, no significant inhibition by RECQL5 was observed upon the omission of RPA (Fig. 5C, lane 7). Exactly the same results—i.e., RPA-dependent inhibition of the D-loop reaction by RECQL5—were obtained when wild-type Rad51 was used (Supplementary Fig. S6A). To address the specificity of the reaction, we substituted RECQL5 with another human RecQ helicase, WRN, which was previously reported to directly interact with Rad51 (Otterlei et al. 2006). WRN, even at 45 nM, did not affect the D-loop reaction catalyzed by either Rad51 K133R (Fig. 5D,E) or Rad51 (Supplementary Fig. S6A, lanes 5,6). However, this specific WRN preparation was quite adept at dissociating a HJ substrate—in fact, much more so than RECQL5 (Supplementary Fig. S7A,B).

D-loop inhibitory action is dependent on ATP hydrolysis by RECQL5 and order of addition of proteins

We used the ATPase-defective RECQL5 K58R mutant (Garcia et al. 2004) to ask whether suppression of D-loop formation requires ATP hydrolysis by RECQL5. RECQL5 K58R retains the ability to interact with Rad51 and Rad51 K133R (data not shown), but has no effect on the D-loop reaction, indicating that the RECQL5 ATPase activity is required for inhibition (Fig. 5F). Rad51 forms a functional presynaptic filament in the presence of the nonhydrolyzable ATP analog AMP-PNP (Chi et al. 2006). The D-loop reaction catalyzed by Rad51 with AMP-PNP as the nucleotide cofactor is impervious to RECQL5 (data not shown), a result that further supports the conclusion that ATP hydrolysis by RECQL5 is needed for Rad51 inhibition.

A series of order-of-addition experiments were conducted to test the possibility that RECQL5 might use its helicase activity to dissociate the D-loop product. Interestingly, for RECQL5 to inhibit the D-loop reaction, it must be added with RPA during the assembly of the presynaptic filament (Fig. 5G; Supplementary Fig. S6B). This finding argues that RECQL5 acts not by unwinding the D-loop product, but by interfering with the presynaptic filament assembly process.

RECQL5 disrupts the Rad51 presynaptic filament

We employed a topoisomerase-linked assay previously used to characterize the presynaptic filament disruption function of Srs2 (Krejci et al. 2003) to enquire whether RECQL5 can catalyze the transfer of Rad51 K133R from ssDNA to a double-stranded DNA (dsDNA) trap. A preassembled presynaptic filament of Rad51 K133R was treated with RECQL5 with or without RPA and, after a brief incubation, an excess of topologically relaxed dsDNA was added to trap Rad51 K133R molecules displaced from the ssDNA. The binding of Rad51 K133R to topologically relaxed dsDNA induces lengthening of the DNA (Ogawa et al. 1993; Sung and Robberson 1995) that can be monitored as a change in the DNA linking number upon incubation with calf thymus topoisomerase I (Fig. 6A; Krejci et al. 2003). The product, Form U DNA, is well separated from the topologically relaxed substrate in an agarose gel (Fig. 6B; Krejci et al. 2003). Control experiments show that RPA and RECQL5 do not make Form U DNA (Fig. 6B, lane 8). The addition of a catalytic amount of RECQL5 resulted in the generation of Form U, indicating that it promotes the transfer of Rad51 K133R from the presynaptic filament to the dsDNA trap (Fig. 6C). The RECQL5-mediated release of Rad51 K133R from the ssDNA is enhanced by RPA (Fig. 6C, cf. lanes 7,8 and 4,5). In contrast, WRN is incapable of mediating the release of Rad51 K133R from the presynaptic filament, regardless of whether RPA is included (Fig. 6C, lane 11) or not (data not shown). The disruptive action of RECQL5 on the presynaptic filament is dependent on its ATPase activity, as the RECQL5 K58R mutant protein is completely inactive in this regard (Fig. 6C, lanes 9,10).

Electron microscopy was used to visualize disruption of the presynaptic filament by RECQL5. Incubation of Rad51 K133R with a 150-mer oligonucleotide and ATP produced abundant presynaptic filaments with distinctive striations (Fig. 7A). Nucleoprotein complexes of RPA with ssDNA were nondescript (Fig. 7B). The dissociation of the Rad51 K133R presynaptic filaments was quantified by determining the relative frequencies of filaments and RPA–ssDNA complexes. At least 2500 nucleoprotein complexes were counted for each reaction. In congruence with the biochemical data, incubation of the Rad51 K133R presynaptic filaments with RECQL5 in the presence of ATP led to its replacement by RPA–ssDNA complexes (Fig. 7C). As anticipated, the Rad51 K133R filaments were impervious to RECQL5 K58R and WRN (Fig. 7C). RECQL5, but not RECQL5 K58R or WRN, also catalyzed the dismantling of presynaptic filaments that comprise Rad51 protein (Supplementary Fig. S8A–C). Thus, the electron microscopic analyses provide clear validation that RECQL5 catalyzes disassembly of the Rad51 presynaptic filament in an ATPase-dependent fashion.

Establishment and culture of mouse ES cells and MEFs

Culture and genetic manipulation of mouse ES cells was performed as described (Hu et al. 2005). The detailed information for individual genetically modified ES cell lines used in this study is described in the Supplemental Material. Primary MEF cultures were derived from 13.5-d-post-coitus (13.5-dpc) embryos of various genotypes, as described (Hu et al. 2005; Mann et al. 2005), except that a low oxygen condition (3% O₂, 5% CO₂, and 92% N₂) was used to reduce oxidative damage-induced stress (Parrinello et al. 2003). All MEF cells used in experiments were under passage number 4.

Mouse and pathological analysis of tumors

Recql5^+/+ and Recql5^−/− mice were generated by crossing between Recql5^+/− mice in a mixed genetic background of 129Sv/Ev and C57BL/6. Mice were maintained in a standard transgenic mouse facility and monitored daily. Individual mice were autopsied, and potential lesions were recorded and fixed in 10% phosphate-buffered formalin. After the fixation, an experienced pathologist examined each animal again. All lesions identified were processed and subjected to standard histological and pathological analyses.

Immunofluorescence

Anti-γ-H2AX (1:200, Upstate Biotechnology; 1:100, Trevigen), anti-Rad51 (1:100, Santa Cruz Biotechnology), and anti-PCNA (1:200, Santa Cruz Biotechnology) antibodies were used in immunofluorescence experiments. Passage 3 MEFs were seeded on glass coverslips for 24 h before drug treatment. After treatment, coverslips were washed briefly in PBS and processed by treating in 0.5% Triton X-100 in PBS (pH 7.1) for 5 min, followed by fixation with 3.7% paraformaldehyde (pH 7.1) for 10 min. Coverslips were incubated with primary antibodies overnight at 4°C and then with Alexa Fluor-conjugated secondary antibodies (Molecular Probes) for 1 h at room temperature, and then were counterstained with DAPI (Vector Laboratories). Images were captured using a confocal microscope (TCS SP2, UV/spectral confocal laser scanner, Leica) and analyzed using Adobe Photoshop (Adobe Systems).

Repair of I-SceI-induced DSBs in mouse ES cells

The SCneo substrate was introduced into the Rb locus in both wild-type and Recql5 knockout ES cells by gene targeting essentially as described (Stark and Jasin 2003), except that the Hyg selection marker was replaced by a Puro marker and puromycin was used to select for transfected cells. After the correctly targeted ES cell lines were obtained, individual cell lines containing the SCneo cassette were transfected with either the expression vector for I-SceI or the corresponding empty vector by electroporation. The transfected cells were divided equally into several plates and then cultured in medium with or without 180 μg/mL G418. The number of colonies from each plate was scored at 10 d after transfection. For each data point, the average frequency of recovering G418-resistant colonies was calculated by dividing the number of G418-resistant colonies by the estimated total number of cells plated after adjusting to the corresponding plating efficiencies. Each experiment was repeated three times.

Chromosome and mitotic abnormalities

Conventional cytogenetic protocols were used for analyzing chromosomal abnormalities as described (Hu et al. 2005; Mann et al. 2005).

RECQL5–Rad51 coimmunoprecipitation

All the incubation steps were performed at 4°C. Purified RECQL5 or RECQL5 K58R and Rad51 or Rad51 K133R, 4 μg each, were incubated in 30 μL of K buffer (20 mM KH₂PO₄ at pH 7.4, 10% glycerol, 1 mM DTT, 0.5 mM EDTA, 0.01% Igepal) containing 150 mM KCl and 100 μg mL⁻¹ BSA for 10 min, and then rabbit polyclonal anti-RECQL5 antibody (0.1 μg) (Kanagaraj et al. 2006) was added, followed by a 30-min incubation. To capture RECQL5 and associated Rad51, the reactions were incubated for 30 min with 10 μL of Protein-G-coupled magnetic beads (Dynal Biotech) with frequent agitation. The magnetic beads and bound proteins were isolated using a magnet and, after washing the beads twice with 15 μL of K buffer containing 150 mM KCl, the bound proteins were eluted with 30 μL of SDS-PAGE sample loading buffer. The supernatant (S) that contained unbound proteins, the wash (W), and the SDS eluate (E), 10 μL each, were subject to 10% SDS-PAGE and staining with Coomassie blue to visualize proteins.

D-loop reaction

Buffer R (25 mM Tris-HCl at pH 7.5, 2 mM ATP, 1 mM MgCl₂, 50 mM KCl, 1 mM DTT, 100 μg mL⁻¹ BSA, containing an ATP-regenerating system consisting of 20 mM creatine phosphate and 20 μg mL⁻¹ creatine kinase) was used for the D-loop reactions, and all the incubation steps were performed at 37°C. Rad51 K133R (1 μM) was incubated with the 5′-end-labeled 90-mer oligonucleotide D1 (3 μM nucleotides) in 11 μL of buffer R for 5 min, followed by the incorporation of Hop2-Mnd1 (300 nM) in 0.75 μL of K buffer and a 1-min incubation. The reaction was initiated by adding pBluescript replicative form I DNA (50 μM base pairs) in 0.75 μL of water. The reactions were terminated after 6 min by the addition of 0.8 μL each of 10% SDS and proteinase K (10 mg mL⁻¹). Following a 3-min incubation, the reaction mixtures were resolved in 0.9% agarose gels in TAE buffer (40 mM Tris acetate at pH 7.4, 0.5 mM EDTA). The gels were dried and subjected to phosphorimaging analysis. When present, RPA (135 nM) and RECQL5 (15–45 nM), WRN (30 and 45 nM), or RECQL5 K58R (45 nM) were added to the preassembled Rad51 filament, followed by a 4-min incubation before the incorporation of Hop2-Mnd1 and replicative form I DNA.

Topoisomerase I-linked DNA topology modification

Buffer R was used for the reactions and all the incubation steps were performed at 37°C. Rad51 K133R (1.5 μM) was incubated for 5 min with pBluescript (−) strand (6 μM nucleotides) in 10.5 μL of buffer R. RECQL5, RECQL5 K58R, or WRN (70 and 140 nM each) and RPA (150 nM) were each added in 0.5 μL of K buffer, followed by a 4-min incubation. Topologically relaxed ϕX174 (7 μM base pairs) in 0.5 μL of water and 2.5 U calf thymus topoisomerase I (Invitrogen) were then incorporated to complete the reaction. The reaction mixtures were incubated for 6 min, terminated, and analyzed as above.

Electron microscopy

Buffer R was used for the reactions and all the incubation steps were performed at 37°C. Rad51 K133R (4 μM) was incubated with 150-mer ssDNA (12 μM nucleotides) for 5 min in 11.5 μL of buffer R, followed by addition of RPA (0.55 μM) in 0.5 μL of K buffer and either buffer, RECQL5, RECQL5 K58R, or WRN (90 nM each) in 0.5 μL of K buffer. After a 4-min incubation, the reactions were diluted eightfold with buffer, and a 4-μL aliquot was applied to 400-mesh grids coated with carbon film that was glow-discharged in air. After staining for 30 sec with 2% uranyl acetate, the samples were examined in a Tecnai 12 Biotwin electron microscope (FEI Company) equipped with a tungsten filament at 100 keV. Digital images were captured with a Morada (Olympus Soft Imaging Solutions) charge-coupled device camera at a nominal magnification of ×87,000.