The interplay between homologous DNA recombination and mitotic progression is poorly understood. The five RAD51 paralogs (RAD51B, RAD51C, RAD51D, XRCC2 and XRCC3) are key enzymes for DNA double-strand break repair. In our search for specific functions of the various RAD51 paralogs, we found that inhibition of XRCC3 elicits checkpoint defects, while inhibition of RAD51B or RAD51C induces G2/M cell cycle arrest in HeLa cells. Using live-cell microscopy we show that in XRCC3-knockdown cells the spindle assembly checkpoint persists and there is a higher frequency of chromosome misalignments, anaphase bridges, and aneuploidy. We observed centrosome defects in the absence of XRCC3. While RAD51B and RAD51C act early in homologous recombination, XRCC3 functions jointly with GEN1 later in the pathway at the stage of Holliday junction resolution. Our data demonstrate that Holliday junction resolution has critical functions for preventing aberrant mitosis and aneuploidy in mitotic cells.
Genome stability relies on the accurate repair of double-strand breaks (DSBs) that arise during DNA replication or from exogenous DNA damaging agents. Homologous recombination (HR) is the major mechanism for the error-free homology-directed repair of DSBs. HR utilizes the presence of the sister chromatid as a template to repair DNA double-strand breaks (West, 2003). HR can be divided into three major steps: the presynaptic stage involving strand stabilization, the synapsic stage characterized by strand invasion and branch migration, and the postsynapsic stage which includes Holliday junction formation and resolution. Following DSB formation, the nuclease activity of MRE11 and CtIP (Mimitou and Symington, 2009; Nimonkar et al., 2008; Sartori et al., 2007) is required for the processing of DNA double-strand breaks to generate the replication protein A (RPA)-coated single-stranded (ss) DNA that is needed for ATR recruitment and phosphorylation of CHK1 (Jazayeri et al., 2006). Recombination mediators, such as the BRCA2 homolog Brh2 (Yang et al., 2005), and PALB2 (Buisson et al., 2010) overcomes the inhibitory effect of RPA and allows RAD51 to nucleate on the ssDNA to promote the presynaptic filament assembly (Yang et al., 2005). The five RAD51 paralogs (RAD51B, RAD51C, RAD51D, XRCC2 and XRCC3), promote binding of RAD51 to the ssDNA and stabilization of the nucleoprotein filament (Masson et al., 2001b; Sigurdsson et al., 2001). Then, the RAD51-bound single-stranded DNA invades a homologous molecule. This invasion leads to the formation of a Holliday junction (HJ), which is migrated and resolved to complete the exchange of genetic information and repair. DNA damage can also be repaired by synthesis-dependent strand annealing (SDSA), which does not involve Holliday junction formation (Nassif et al., 1994; Pâques and Haber, 1999).
There is much to learn about the biochemical properties and the biological functions of the paralogs. Two complexes of the RAD51 paralogs are present in cells. One complex, referred to as BCDX2, consists of RAD51B, RAD51C, RAD51D and XRCC2, whereas the CX3 complex contains RAD51C and XRCC3. The presence of additional sub-complexes, RAD51B–RAD51C, RAD51C–XRCC3, RAD51C–RAD51D and RAD51C–RAD51D–XRCC2, were also reported using yeast two- and three-hybrid assays and immunoprecipitation analyses (Liu et al., 2002; Masson et al., 2001a; Masson et al., 2001b; Miller et al., 2002; Wiese et al., 2002). The BCDX3 and CX3 complexes bind DNA in vitro (Masson et al., 2001a; Masson et al., 2001b), and the RAD51B–RAD51C heterodimer enhances Rad51-mediated strand exchange (Sigurdsson et al., 2001). Studies indicate that most RAD51-like proteins have a DNA-stimulated ATPase activity (Lio et al., 2003), like RAD51 itself. Many observations suggest that the RAD51 paralogs are involved in genome stability. In hamster, XRCC3 deficiency was shown to lead to a 25-fold decrease in error-free homology-directed repair using a I-SceI recombination reporter system (Pierce et al., 1999). Chicken DT40 cells with knocked-out RAD51 paralogs, although viable, were found to be sensitive to DNA cross-linking agents and to ionizing radiation besides exhibiting recombination/repair defective phenotypes such as reduced growth rates, chromosomal instability and spontaneous chromosomal breaks (Takata et al., 2001). RAD51D has been linked to telomere maintenance (Tarsounas et al., 2004; Suwaki et al., 2011). Recently, RAD51C and RAD51D have been found to be mutated in ovarian cancer (Loveday et al., 2011; Loveday et al., 2012), while a common variant in RAD51B was associated with male breast cancer risk (Orr et al., 2012).
The paralogs play roles at the early and late stages of homologous recombination. The inability of cell lines deficient in RAD51C (French et al., 2002; Godthelp et al., 2002), XRCC2 (O'Regan et al., 2001) and XRCC3 (Bishop et al., 1998) to form damage-induced RAD51 nuclear foci suggests that these three proteins are important in the homology search and strand invasion phase of homologous recombination. In support of this, Arabidopsis RAD51C is essential for the repair of Spo11-induced DSBs during prophase I of meiosis. In its absence, homologous chromosomes fail to synapse and become severely fragmented consistent with an early role for the RAD51 paralogs (Li et al., 2005). In addition, Arabidopsis xrcc3 mutants show a decrease in the number of SPO11-dependent AtRAD51 foci during meiosis (Vignard et al., 2007). These RAD51-like proteins also play important roles in later recombination events, perhaps by helping to stabilize heteroduplex DNA. Consistent with this idea, CX3 and BCDX2 ring structures bind Holliday junctions (Compton et al., 2010) and Holliday junction branch migration and resolution activities are dependent on the presence of the RAD51 paralogs (Liu et al., 2004). The XRCC3 mutant irs1SF hamster cell line showed a defect in HJ resolution activity while irs1 cells (defective in XRCC2), were normal (Liu et al., 2004). In addition, when the localization of RAD51C to meiotic chromosomes was investigated, RAD51C was detected at the pachytene stage, as one or two distinct foci associated with each synapsed bivalent. These foci are believed to represent the sites of crossovers, as RAD51C foci are markedly reduced in spermatocytes from Mlh1-deficient mice (Kuznetsov et al., 2007).
In this paper, we investigated the functions of the RAD51 paralogs during the cell cycle. Notably, we found that knockdowns of XRCC3 and RAD51C lead to opposite effects in cell cycle progression. Depletion of RAD51C by siRNA leads to a G2/M-phase arrest while XRCC3 knockdown leads to several mitotic problems possibly due to a deficiency in Holliday junction processing. To test this possibility, we analyzed the functions of the four known activities that can contribute to Holliday junction dissolution or resolution, GEN1 (Ip et al., 2008; Bailly et al., 2010; Rass et al., 2010), MUS81 (Chen et al., 2001), SLX1-SLX4 (Fekairi et al., 2009; Muñoz et al., 2009; Svendsen et al., 2009), and BLM helicase topoisomerase IIIa–RMI1/2 (Wu and Hickson, 2003) in cell cycle progression. We found that GEN1 and XRCC3 function in a common pathway and GEN1-knockdown cells displayed similar phenotypes to siXRCC3 cells. Our results support the concept that unresolved Holliday junctions lead to the missegregation of chromosomes in mitosis.
RAD51C and XRCC3 affect cell cycle progression differentially
To determine how human HR genes affect checkpoint and cell cycle progression, we analyzed the function of the RAD51 paralogs. We used a siRNA approach against the paralogs RAD51B (si51B), RAD51C (present in both paralog complexes) (si51C) and XRCC3 (siX3) (supplementary material Fig. S1) and immunofluorescence against the CENP-F protein as a marker for G2/M cells (Kao et al., 2001). We observed an increase in G2/M cells upon treatment with si51B or si51C, while cells treated with siCTL (which designates control scrambled siRNA) or siX3 were unaffected (Fig. 1A). These results suggested that RAD51 paralog deficiency affects cell cycle progression differentially. To confirm this result, HeLa cells were synchronized by a double thymidine block in G1/S, released for 9 and 12 hours and processed for FACS analysis (Fig. 1B,D). Between 9 and 12 hours, most of the siCTL and siXRCC3 cells had completed mitosis and cell division, as judged by the accumulation of cells with 2N DNA content. By contrast, most of si51C- and si51B-treated cells were stopped at the G2/M transition. To address this issue more precisely, we monitored cell cycle progression of siRNA knockdown cells using the fluorescent, ubiquitination-based cell cycle indicator (FUCCI) system (Sakaue-Sawano et al., 2008) according to the double thymidine synchronization procedure presented in Fig. 1B. FUCCI is a fluorescent protein-based sensor that employs a red (RFP) and a green (GFP) fluorescent protein fused to Cdt1 and geminin, respectively. Degradation of the green Geminin renders G1 cells red (expression of Cdt1–GFP), whereas destruction of the red Cdt1 protein in S, G2 and M renders cells green (expression of Geminin–RFP). The siRNA knockdowns were performed just prior to cell synchronization for optimal efficiency at the time of harvest. When cells were visualized 9 hours after release from the G1/S block, the majority had finished DNA replication as they exhibited green fluorescence. By 12 hours after release, however, siCTL cells had turned red (G1), while siX3 cells were mostly colorless (representative of the end of mitosis) and si51C and si51B cells were still green, suggestive of a S, G2 or early M phase block (Fig. 1C). These results suggested that siX3 cells do not undergo checkpoint arrest unlike si51C and si51B cells. However, they did not behave like control cells, as they progressed slower during the G2/M phase.
The cyclin-dependent CDK1 and WEE1 are important factors regulating mitosis (Watanabe et al., 1995). At 9 and 12 hours after release, we observed that the phosphorylation level of CDK1 was increased in si51C and si51B as compared to siCTL and siX3 cells (Fig. 1E). The level of histone H3 phosphorylation on serine 10, a G2/M marker, was inversely correlated to that of CDK1 in siCTL and siX3 versus si51C and si51B. Similar results were obtained with cells synchronized with thymidine-aphidicolin (data not shown). We reasoned that if induction of cell cycle arrest in RAD51B- and RAD51C-knockdown cells depends on the inhibitory phosphorylation of CDK1, then normal regulation of WEE1 might be changed in these cells. This was indeed evident in si51C and to a lesser degree, in si51B cells, where WEE1 upregulation, which is a sign of checkpoint engagement, was increased relatively to siCTL- and siX3-treated cells (Fig. 1E). To further investigate this hypothesis, we used a double knockdown by combining a siRNA against WEE1 and either siCTL, siX3, si51B and si51C under the same experimental protocol (Fig. 1B). Transfection with siWEE1 significantly reduced WEE1 protein levels (supplementary material Fig. S1), and increased the G1/S population of siRAD51B and RAD51C 12 hours after release from the thymidine block (Fig. 1F). Altogether, these results show that deficiency in RAD51B and RAD51C triggers a CDK1- and WEE1-dependent G2/M cell cycle arrest that can be overcome by abrogation of WEE1.
It could be argued that higher endogenous DNA damage in si51C and si51B cells lead to a more pronounced cell cycle arrest than in siX3. Cells depleted in RAD51B, RAD51C and XRCC3 accumulate similar level of DNA damage as judged by 53BP1 staining (supplementary material Fig. S2). Moreover, the number of RAD51 foci in siX3 cells was similar to control cells whereas si51B and si51C showed a significant decrease in RAD51 foci formation (supplementary material Fig. S3). Altogether, these results show that siX3 cells sustain DNA damage, as other paralog knockdowns, but do not show deficiency in activating RAD51. To test whether depletion of XRCC3 affect cell cycle progression in the presence of DNA damage, we monitored the cell cycle profile upon exogenous DNA damage induced by the radiomimetic drug neocarzinostatin (NCS). HeLa cells were synchronized, and once they had entered S-phase, they were treated for 2 hours with 50 ng/ml NCS, washed and released for cell cycle progression. Consistent with a G2/M checkpoint activation in response to DNA damage, siCTL, siX3, si51C and si51B cells all showed a delay entry into mitosis, as most cells were in G2/M at time 12 hours (Fig. 2B). However, siCTL and siXRCC3 cells were able to resume cell cycle progression such that after 24 hours, the majority were in G1, as opposed to most siRAD51C and siRAD51B cells that were still in G2/M. These results suggest that XRCC3-deficient cells, unlike RAD51C-deficient cells, can escape the G2/M arrest induced by DNA damage. These results were also observed in U2OS cells (supplementary material Fig. S4A).
Effect of XRCC3 on mitotic progression and centrosome stability
To analyze the effect of the checkpoint defects on mitosis in XRCC3-deficient cells, we used HeLa cells carrying a stable copy of histone H2B fused to GFP to allow live-cell visualization of mitotic division. This cellular approach, unlike FACS, allowed us to discriminate between a simply delayed entry into mitosis versus a delayed progression through that phase. Under all conditions tested, only a few si51C and si51B cells could escape the G2/M arrest and go through mitosis. When the duration of mitotic intervals (i.e. prophase to metaphase, and metaphase to anaphase) were analyzed, we found that X3-knockdown cells displayed a metaphase delay, taking on average 110 minutes to complete division (Fig. 3A). Hence, although XRCC3-knockdown cells do proceed in the cell cycle despite the presence of DNA damage, mitotic progression is extended.
Since siX3-treated cells exhibited a delay in the metaphase-to-anaphase transition, immunofluorescence was performed using antibodies against BubR1 and MAD2, two components of the spindle assembly checkpoint (SAC) that monitor spindle pole assembly at the metaphase–anaphase transition. During normal cell division, BubR1 and MAD2 localize to unattached kinetochores at prometaphase and decrease progressively in abundance during metaphase when proper bipolar attachment is achieved. We observed that in siXRCC3 cells, BubR1 staining remains detectable up to anaphase at levels higher than observed in siCTL (Fig. 3B). Consistent with this, we observed 60% of siX3 metaphase cells with multiple MAD2-positive kinetochores versus 12% for the siCTL (supplementary material Fig. S4B). These studies show that XRCC3 affects mitotic progression and SAC assembly and that XRCC3 defects will most likely lead to aneuploidy. Hence, the presence of mitotic aberrations (micronuclei, binuclei and multinuclei) was monitored in paralog knockdown cells. The number of micronucleated cells was particularly increased in XRCC3 knockdown, such aberrations originating from chromosome fragments or whole chromosomes that lag behind at anaphase during nuclear division (Fig. 4A; supplementary material Fig. S4C for U2OS cells). In contrast, si51C cells exhibited an increase in binucleated cells, suggesting a cytokinesis defect, highlighting again differences between XRCC3 and RAD51C.
A number of conditions can lead to SAC activation in metaphase cells, including centrosome fragmentation, which prevents stable microtubule–kinetochore attachment (Wong and Fang, 2006). Centrobin is a centriole-associated protein that is required for centriole duplication (Zou et al., 2005). It was recently shown that centrobin-depleted cells have compromised centrosome and spindle integrity, leading to persistent SAC activation (Jeffery et al., 2010). Since RAD51C-deficient hamster cells have an increased number of centrosomes in mitotic cells (Renglin Lindh et al., 2007), we verified whether centrosomes abnormalities might account for persistent SAC activation in siXRCC3 cells by pericentrin staining. XRCC3-, RAD51C-, and RAD51B-depleted cells show a twofold increase in the number of cells having more than two centrosomes (centrosome amplification), with some cells bearing up to seven centrosomes (Fig. 4B). Besides amplification, centrosomal fragmentation was also observed and was predominant in siX3 cells compared to si51C and si51B cells. Pericentrin interacts with the microtubule nucleation component γ-tubulin and is important for normal functioning of the centrosomes (Doxsey et al., 1994). To confirm that the pericentrin abnormality corresponds to actual centrosome defects, cells were stained for a second centrosome marker, γ-tubulin, which also shows abnormal amplification and fragmentation patterns (Fig. 4C).
Role of Holliday junction processing in XRCC3-associated cellular defects
We next endeavored to pinpoint the exact mechanism by which XRCC3 deficiency entails mitotic defects. XRCC3 plays roles in the recruitment of RAD51 (Bishop et al., 1998) as well as later in HR during Holliday junction processing (Brenneman et al., 2002; Liu et al., 2004). As downregulating RAD51C, which is also involved in the recruitment of RAD51 (Rodrigue et al., 2006), does not confer a phenotype similar to XRCC3 deficiency, we reasoned that the XRCC3-knockdown phenotypes observed in this study might be explained by the failure of resolving DNA recombination intermediates at later stages of HR. There are currently four known activities that can contribute to Holliday junction dissolution or resolution, GEN1 (Ip et al., 2008), MUS81 (Chen et al., 2001), SLX1–SLX4 (Fekairi et al., 2009; Muñoz et al., 2009; Svendsen et al., 2009), and BLM helicase topoisomerase IIIa–RMI1/2 (Wu and Hickson, 2003). We reasoned that if Holliday junction processing is indeed the mechanism leading to the phenotypes observed in XRCC3-knockdown cells, similar effects should be observed following knockdown of some of these enzymes (supplementary material Figs S5, S1). For the SLX1–SLX4 complex, we depleted SLX4 as it leads to a co-concomitant decrease of SLX1 protein levels (Muñoz et al., 2009). Using HeLa H2B–GFP cells, we analyzed the duration of mitosis in cells knockdown for HJ processing enzymes. siGEN1 cells were very similar to siX3 cells, with an average metaphase to anaphase transition increased by 50 minutes while siBLM showed ∼10 minute increase and siSLX4 cells progressed with normal kinetics (Fig. 5A,B). In contrast, siMUS81 cells entered anaphase after chromosome condensation without metaphase alignment. Given the similarity between siGEN1 and siX3 cells, we further analyzed the phenotypes associated with GEN1 depletion (Fig. 6; supplementary material Figs S5, S6). Remarkably, cells knocked down for GEN1, recapitulated the phenotypes observed in siX3 cells, namely (i) in contrast to si51C and si51B cells, siGEN1 cells escaped from the G2/M arrest (Fig. 6A); (ii) they displayed extended BubR1 staining and a SAC delay in mitosis (Fig. 6B); (iii) an increase in micronucleated cells (Fig. 6C); (v) an increase in centrosome fragmentation and amplification during mitosis (Fig. 6D,E).
To assess whether XRCC3 and GEN1 deficiencies lead to similar phenotypes related to HJ processing, we monitored the accumulation of DNA bridges by exploiting Plk1-interacting checkpoint helicase (PICH) staining as a marker (Fig. 7A) (Baumann et al., 2007). We found that both siX3 and siGEN1 cells had an increase of PICH threads representing ultrafine DNA bridges. To further define that the PICH threads we observed were indeed formed by unresolved Holliday junctions, we expressed active or catalytically inactive (D70N substitution) forms of the bacterial resolvase protein RusA in siXRCC3- and siGEN1-knockdown cells (supplementary material Fig. S7A,B). The RusA protein cleaves four-way Holliday junctions (Chan et al., 1997), and was shown to suppress the deficiencies in the Escherichia coli bacterial resolvase protein RuvC (Mandal et al., 1993), a late-stage meiotic recombination defect in S. pombe Mus81 mutants (Boddy et al., 2001), and improved both cell survival and the generation of recombination products following DNA damage in WRN-deficient cells (Giraud-Panis and Lilley, 1998; Saintigny et al., 2002). The number of PICH threads in these cells was not affected by expression of RusA D70N compared to the mock control. However, the number of PICH-positive siXRCC3 cells were reduced when wild-type RusA was expressed (Fig. 7B). Collectively, these results show that resolvase activity is necessary for the elimination of ultrafine bridges in siXRCC3 cells. Furthermore, these results show that defects in XRCC3, and particularly in Holliday junction resolution, lead to abnormal mitosis (Fig. 7C).
In 1964, Robin Holliday proposed that a cross-stranded DNA structure, named Holliday junction, explained meiotic gene conversion in fungi (Holliday, 1964). It is now known that Holliday junctions physically link homologous chromosomes in meiosis, as these four-stranded structures must be resolved into linear duplexes to complete meiosis I. However, the functional consequence of having unresolved HJ during mitosis in human cells is still poorly understood. This was hampered by the fact that many activities dissolving/resolving HJ junctions, including GEN1 and SLX1–SLX4, have been identified only recently.
The initial goal of the study was to find why the cell needs five different RAD51 paralogs to accomplish a similar function in DNA repair. Given that a deficiency of XRCC3, RAD51C, or RAD51B should affect homologous recombination, a knockdown of either one of these paralogs would be predicted to trigger a cell cycle arrest at the G2/M transition because of unrepaired DSBs. However, this is not the case, as XRCC3-knockdown cells did not exhibit increased CDK1 phosphorylation nor stabilization of WEE1 as opposed to si51C- and si51B-treated cells. Furthermore, although a pronounced delay was observed at the metaphase-to-anaphase transition, XRCC3-knockdown cells did not arrest at the G2/M transition as judged by FACS analysis. Hence, our study highlights a new separation of functions for XRCC3 compared to RAD51C.
The RAD51 paralogs are not listed in the centrosomeDB database (http://centrosome.dacya.ucm.es/centrosome) (Nogales-Cadenas et al., 2009), which contains human genes based on a compilation from the literature and homology to centrosome genes identified in various organisms. However, defects in the hamster RAD51 paralogs have been shown to promote centrosome defects (Griffin et al., 2000; Renglin Lindh et al., 2007). One possibility is that inhibition of XRCC3 or GEN1, leading to aberrant recombination or DNA repair, may indirectly lead to centrosome abnormalities. In fact, this phenomenon is not only restricted to our study, but also occurs in cells deficient for homologous recombination factors such as RAD51, RAD54 (Dodson et al., 2004), or BRCA1 (Starita et al., 2004).
The unrepaired DNA damage in cells depleted in XRCC3 and GEN1 might also impact on SAC. SAC prolongs mitosis until all kinetochores are stably attached to spindle microtubules. The persistent SAC activation in siGEN1 and siXRCC3 cells was detected using BubR1 and Mad2 staining, a clear indication that attachments are not fully established. This was suppressed by a knockdown of BubR1 (supplementary material Fig. S7C). It is known that SAC can be delayed by the number of chromosomes or centrosomes (Yang et al., 2009). Collectively, we propose that XRCC3-defective cells must accumulate DNA lesions, which impacts on centrosomal stability and SAC during mitosis. Our results also suggest that GEN1 and XRCC3 are functioning in the same pathway for centrosomal stability (supplementary material Fig. S8).
DSBs can also be repaired by non-homologous end-joining using DNA-PKcs, the Ku70/80 proteins, XRCC4, and DNA ligase IV and accessory factors (Chapman et al., 2012). Since Ku80 mutation does not lead to centrosomal defects (Griffin et al., 2000), we hypothesized that the DNA damage accumulated in XRCC3-knockdown cells must be of a different nature than solely unrepaired DSBs, but rather, different DNA repair intermediates. During genetic recombination, DNA molecules become linked at points of strand exchange by Holliday junctions. Branch migration and resolution of these crossovers allow the recombination process to be completed (West, 2003). We favor the possibility that XRCC3 important function in mitosis occurs through the accumulation of unresolved Holliday junctions. It was previously shown using in vitro HJ branch migration and resolution assays that both RAD51C and XRCC3 are required for Holliday junction processing in human cells (Liu et al., 2004), although the purified proteins did not possess intrinsic resolvase activity (Liu et al., 2007). Given these results, we originally predicted that knockdown of RAD51C would display the same phenotypes as a knockdown of XRCC3. This is not what we observed, as a knockdown of RAD51C, lead to the accumulation of cells arrested in G2/M. Our data indicate that cells depleted in RAD51C display a stronger phenotype, such as an enhanced reduction of RAD51 foci formation compared to siX3 cells, and they cannot reach the later stages of HR whereas XRCC3-knockdown cells do. As RAD51C is at the core of two different paralog complexes (BCDX2 and CX3), rather than one for XRCC3 (CX3), this may exacerbate its associated phenotype. Consistent with this, only a minor proportion of si51C cells could progress in mitosis, leading to an increase in binucleated cells. In contrast, most cells depleted for XRCC3 progressed to complete mitosis, albeit slower than wild-type cells.
Using live-cell analysis with HeLa GFP–H2B cells, we closely examined the mitotic progression of siX3 cells. We observed a higher number of DNA bridges between two separating daughter siX3 cells. We propose that mitosis cannot proceed efficiently in siX3 cells because recombining DNA molecules are still attached to each other leading to chromosome missegregation and aneuploidy, as observed in Blm-deficient cells (Chan et al., 2007). We also monitored an increase of PICH ultrafine bridges. PICH has been reported to connect centromeres and other chromosomal loci distinct from centromeres. The PICH helicase does not play a role in SAC, and the ultrafine bridges we observed are possibly due to mechanical linkages between sister chromatids. Although it has been mentioned that such structures might be transient (Wang et al., 2010), inhibition of XRCC3 and GEN1 leads to the accumulation of PICH structures in late anaphase cells which are likely to represent unresolved Holliday junctions. This is supported by the observation that expression of the E. coli RusA resolvase in siX3-knockdown cells clearly suppressed the number of PICH bridges. We consistently noted that the number of PICH structures in siGEN1 cells, were not suppressed as much as in siX3 cells. One possibility is that the recombination intermediates produced in siGEN1 cells might be different than in XRCC3 knockdown, leading to inefficient HJ cleavage by RusA. These results suggest that XRCC3 acts late in the HR process, to resolve or prevent the formation of HJ.
For more than 25 years, the Holliday junction has been defined as a central intermediate of homologous recombination. Until recently, several pieces of evidence hampered us to assess the role of Holliday junctions in human cells, such as detecting these intermediates in mitotically cycling eukaryotic cells (Bzymek et al., 2010) or identifying new enzymes acting on these substrates (Svendsen and Harper, 2010). Furthermore, the frequency of HJ in diploid mitotic cells is much lower compared to meiotic cells, because they form preferentially between sister chromatids and to a lesser extend between homologue chromosomes. Our study demonstrates that inactivating XRCC3 and GEN1 constitutes a common pathway for checkpoint dysfunction leading to aneuploidy. This is in accordance with the observation that depletion of MUS81 and GEN1 or SLX4 and GEN1 from Bloom syndrome cells result in chromosome instabilities (Wechsler et al., 2011). Our results support the concept that Holliday junction resolution is important for proper mitosis in human cells.
Materials and Methods
Cell culture and plasmids
HeLa cell lines stably expressing H2B–GFP (obtained from Randall King) or FUCCI (a gift from Atsushi Miyawaki) were maintained in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. U2OS human osteosarcoma cell line (ATCC, HTB-96) was cultured in McCoy's 5A medium containing 10% FBS and 1% penicillin/streptomycin.
Cell synchronization and neocarzinostatin treatment
Double thymidine synchronization was performed as previously published (Whitfield et al., 2002). In brief, HeLa H2B–GFP or FUCCI cells at 25–30% confluence were washed twice with PBS and incubated in DMEM containing 2 mM thymidine for 18 hours (first block). Cells were then washed twice with PBS to remove thymidine and were released into fresh medium. Nine hours after release, the medium was changed to DMEM supplemented with 2 mM thymidine for 17 hours (second block). Cells arrested at the G1/S boundary were then released into fresh medium and samples were processed for FACS analysis, western blotting, immunofluorescence or live-cell imaging. When indicated, cells were treated for 2 hours with 50 ng/ml NCS, 7 hours after release from the second thymidine block.
For thymidine-aphidicolin synchronization, cells were synchronized according to Zhu et al. with minor modifications (Zhu et al., 2000). HeLa cells were treated with 2 mM thymidine for 14 hours. The cells were washed with PBS and released into fresh media for 11 hours. Subsequently, cells were treated with aphidicolin at 1 µg/ml for 14 hours. Cells were then released in fresh media and harvested for cell cycle studies.
The siRNAs used in this study were synthesized by Dharmacon and directed against the following target sequences: siXRCC3NO. 1 (5′-CAGAAUUAUUGCUGCAAUUAA-3′); siXRCC3NO. 2 (5′-CAGCCAGAUCUUCAUCGAGCA-3′); siRAD51CNO. 1 (5′-CACCUUCUGUUCAGCACUAGA-3′); siRAD51CNO. 2 (5′-AAGAGAAUGUCUCACAAAUAA-3′); siRAD51BNO. 1 (5′-CCCAGUUAUCUUGACGAAUCA-3′) and (5′-ACGAGUGGGUUUAUCACAAGA-3′); siRAD51BNO. 2 (5′-GAGGUGUCCAUGAACUUCUAUGU-3′); siWEE1 (5′-AAGACCUUCCGCAAGCUGCGA-3′); siBubR1 (5′-GUCUCACAGAUUGCUGCCU-3′); siSLX4 (5′-AGCGGCAUUUGAGUCUGCAGGUG-3′).
The siRNA-mediated depletion of GEN1 was accomplished using a siGENOME SMARTpool reagent from Dharmacon (M-018757-00-0005), which consists in a combination of four different siRNA directed against GEN1, or with individuals siRNA, siGEN1NO. 1 (5′-UCUAAGACCUUUGGCUAUA-3′) and siGEN1NO. 2 (5′-GAAUCUAGUCAACCCAAUA-3′). Depletion of BLM and MUS81 was also performed using siGENOME SMARTpool reagents from Dharmacon (M-007287-02-0005 and M-016143-01-0005, respectively). As negative control, a non-silencing siRNA with target sequence 5′-GACGUCAUAUACCAAGCUAGUUU-3′ was used. Transfection of siRNAs was performed using Oligofectamine (Invitrogen), according to the manufacturer's protocol with minor modifications. In brief, cells were seeded in six-well plates at 1.5×105 cells/well, 16 hours prior to transfection. For each transfection, the cells were exposed to 50 nM (siGENOME SMARTpool) or 250 nM of the respective siRNA in serum- and antibiotic-free Opti-MEM (Invitrogen) during 4 hours and then supplemented with 500 µl of DMEM containing 30% FBS. Asynchronous or synchronized cells were fixed or harvested 48–60 hours following transfection.
Synchronized cells were collected by trypsinization, centrifuged and resuspended at 1×106 cells per ml in ice-cold (70% in PBS) while vortexing. Once fixed, cells were washed with PBS and stained with propidium iodide (0.1% sodium citrate, 0.3% Nonidet P40, 50 µg/ml propidium iodide, and 20 µg/ml RNaseA). Cell cycle analysis was performed on a Beckman Coulter Epics Elite model ESP using the Expo2 analysis software.
Western blot analysis
siRNA-transfected HeLa cells were harvested 9 and 12 hours following release from the second thymidine block, and soluble protein extracts were prepared as previously described (Rodrigue et al., 2006), resolved on 10% to 12% SDS-PAGE gels, and blotted onto nitrocellulose (Perkin-Elmer). After transfer, membranes were blocked for 1 hour in 5% skim milk/0.05% Tween in PBS and probed overnight at 4°C with mouse monoclonal antibodies against XRCC3 (10F1), RAD51C (2H11), RAD51B (1H3), GAPDH (Research Diagnostics), MUS81 (Novus, NB100-2064), BubR1 (Zymed, 41-2800) or rabbit polyclonal antibodies against phospho-Histone H3 (Ser10; Upstate Biotechnology, No. 06-570), WEE1 (Santa Cruz, sc-325), phospho-CDK1 (Tyr15; Cell Signaling, 9111), and BLM (Novus, NB100-66594). The primary monoclonal and polyclonal antibodies were diluted in 5% skim milk or 5% BSA/0.05% Tween 20 in PBS, respectively. Blotted proteins were revealed using enhanced chemiluminescence (Perkin-Elmer) following a 1 hour incubation with horseradish peroxidase-conjugated anti-mouse or anti-rabbit immunoglobulins.
For immunofluorescence studies of CENP-F, α-tubulin, 53BP1, and RAD51, siRNA-transfected cells grown on coverslips, were treated as detailed in Fig. 1A, Fig. 4A, supplementary material Figs S2, S3, respectively. Fixation, permeabilization and immunostaining was performed as previously described (Rodrigue et al., 2006) and cells were visualized using primary antibodies anti-CENP-F (Genetex, 30232), anti-α-tubulin DM1A (Abcam, ab7291), anti-53BP1 (Novus, NB 100-304A), or anti-RAD51 (Santa Cruz, sc-8349). Scoring of CENP-F-positive cells was performed as described in Fig. 1. For 53BP1 foci formation analysis, the number of cells with ≧10 53BP1 foci was determined and presented as mean percentage (± s.e.m.) from three independent experiments (n>200 per experiment). In the RAD51 foci formation analysis, the number of cells displaying no nuclear foci, 0–10 foci or more that 10 foci was scored and presented as mean percentages (± s.e.m.) from three independent experiments, with over 200 cells counted per experiment.
For aberrant mitosis detection, asynchronous cells were fixed 48 hours post-transfection and analyzed for micro-, bi- and multinucleated cells as determined by 4′,6-diamidino-2-phenylindole (DAPI) and α-tubulin staining. Over 500 nuclei were scored per sample, considering only those with well-defined boundaries.
Frequency of centrosomal aberrations was analyzed 48 hours following siRNA transfection using pericentrin and γ-tubulin immunostaining. First, cells were fixed with 2% paraformaldehyde in PBS for 10 minutes and fixed with methanol (−20°C) for 5 minutes. Next, cells were permeabilized with PBS containing 0.2% Triton X-100 for 5 minutes and quenched with 0.1% sodium borohydride for 5 minutes. After blocking in PBS containing 10% goat serum, 1% BSA and 0.02% NaN3 for 1 hour at RT, cells were incubated in the primary antibody anti-pericentrin (Abcam, ab4448) or anti-γ-tubulin (Sigma, T6557) diluted 1% BSA in TBS for 2 hour at RT. After multiple washing in TBS, cells were incubated with the appropriate secondary antibody conjugated to a fluorophore. For each sample, at least 300 interphase and 100 mitotic cells were examined with respect to the number and the integrity of the pericentrin and γ-tubulin spots. Centrosomes were considered abnormal if present in more than two per cell or when multiply fragmented.
The analysis of SAC activation and PICH threads was performed on HeLa H2B–GFP cells synchronized by double thymidine block and transfected with the appropriate siRNA as above. When cells reached mitosis (t = 10 hours), they were fixed and permeabilized simultaneously in 20 mM PIPES pH 6.8, 4% formaldehyde, 0.2% Triton X-100, 10 mM EGTA, 1 mM MgCl2 for 15 minutes. Cells were then blocked for 30 minutes at RT in PBS containing 3% BSA, incubated at 37°C for 3 hours in primary antibody(ies) followed by secondary antibody staining in the blocking solution. SAC components BubR1 and MAD2 were detected with anti- BubR1 (Zymed, 41-2800) and anti-MAD2 (Santa Cruz, sc-65492). The percentage of siCTL and siX3 cells with multiple MAD2-positive kinetochores were obtained from three independent experiments (n = 75 metaphase cells) while microtubules were stained with α-tubulin DM1A. Detection of PICH threads in siCTL, siXRCC3 and siGEN1 cells was performed using anti-PICH (Millipore, 04-1540). The number of PICH-positive bridges for each condition was determined by analyzing over 200 late anaphases.
In all cases, secondary antibodies included anti-mouse and anti-rabbit antibodies each conjugated with either Alexa Fluor 488, Alexa Fluor 568 or Alexa Fluor 647 (Molecular Probes). Coverslips were mounted onto slides with PBS–glycerol (90%) containing 1 mg/ml paraphenylenediamine and 0.2 mg/ml DAPI.
Fluorescence microscopy data was reported as average percent anomalies (± s.e.m.) from at least three independent experiments. For live cell imaging, outliers were first removed using Grubb's test with a P-value of 0.01. The significant difference between mean phase transition of the different siRNAs compare to the siCTL was assessed using t-test followed by a Bonferonni multiple testing correction. We considered as being significant only differences having a corrected P-value lower than 0.01. All the statistical analysis performed in this study were done using R (www.r-project.org).
HeLa cells stably expressing GFP-tagged histone H2B (H2B–GFP cells) were plated onto 35-mm dishes, transfected with the respective siRNA (control or directed against XRCC3, RAD51C, RAD51B, GEN1, BubR1, MUS81, SLX4, or BLM) and synchronized as described above. Eight hours after release from the second thymidine block, live-cell imaging was performed with the TE 2000 microscope (Nikon) using a Photometrics CoolSnap HQ camera. Cell culture dishes were placed onto the microscope with an environmental control chamber, maintaining the temperature at 37°C and the CO2 at 5%. The cells were imaged every 2 minutes for 15 hours with a Prior motorized stage for four dishes, and a stack of images with a Z-step size of 4.5 µm was collected. All images were collected with a Plan Fluor 40×lens (numerical aperture = 0.6) with a Chroma filter set 41001 configured for EGFP fluorescent protein tags [i.e. a set of filters for excitation atl (emission) = 488 nm and l (emission) = 500–550 nm]. Exposure times were 450 ms with a camera binning of 2. With a scanning stage, Z-stacks (three steps of 1.5 µm) of approximately nine cells were imaged at five different XY positions on four dishes during each experiment. The data were analyzed with MetaMorph software (Molecular Devices) and converted into .avi-format movie files. Owing to the large size of the movie files required to keep optimal resolution, all movies are not included in the supplementary material but are available on request.
We like to acknowledge Randall King for the gift of HeLa H2B–GFP cells and Ray Monnat for the RusA plasmids. We also like to thank Sabine Elowe, Steve West, Rajvee Shah, Josée N. Lavoie, Margit Fuchs, Richard Poulin and members of the Masson lab. for helpful comments, Carl St-Pierre of the Unité d'Imagerie Cellulaire for technical help, Nancy Roberge for FACS analysis, and Eric Paquet for statistical analysis.
This work was supported by the Canadian Institutes of Health Research to J.Y.M. and a doctoral graduate scholarship to A.R.