Formation of higher-order nuclear Rad51 structures is functionally linked to p21 expression and protection from DNA damage-induced apoptosis

DNA double-strand breaks (DSBs), which may occur spontaneously or arise as a direct effect of ionizing radiation and many DNA-damaging agents, are potentially lethal for the cell. Recombinational repair of DSBs is an essential process for genome integrity that has been evolutionarily conserved from bacteria to man. Eukaryotic cells repair DSBs either by direct non-homologous end-joining of the broken ends or by homologous recombination (Jeggo, 1998). Homologous recombination is also responsible for the generation of genetic diversity during meiosis (Shinohara et al., 1992). Recent studies have shown that Rad51 recombinase plays an essential role in homologous recombination in mammalian cells. Similar to Escherichia coli RecA, both yeast and mammalian Rad51 proteins form nucleoprotein filaments on single-stranded (ss) DNA, mediating homologous pairing and strand-exchange reactions between ssDNA and homologous double-stranded DNA (Sung, 1994; Baumann et al., 1996; Gupta et al., 1997).

In normal, cultured mammalian cells, the Rad51 protein is detected in multiple discrete foci in the nucleoplasm of a low number of cells by immunofluorescent antibodies. After DNA damage, the percentage of cells with focally concentrated Rad51 protein increases in a time- and dose-dependent manner. Rad51-foci-positive cells are arrested during the cell cycle and undergo unscheduled DNA repair synthesis (Haaf et al., 1995; Haaf et al., 1999). Nuclear foci are formed at sites of DNA-damage-induced ssDNA (Raderschall et al., 1999) and contain the ssDNA-binding replication protein A (RPA) (Golub et al., 1998), which facilitates homologous pairing and DNA strand exchange, mediated by Rad51 (Baumann et al., 1996; Gupta et al., 1998). In addition, Rad foci may also contain Rad52 (Liu et al., 1999) and Rad54 (Tan et al., 1999), which belong to the same epistasis group as Rad51. It seems plausible that DNA-damage-induced Rad51 foci represent a repairosome-type assembly of Rad51 and other proteins that are essential for recombinational DNA repair.

In contrast to E. coli RecA and yeast ScRad51, mammalian Rad51 protein appears to be necessary for cell survival. Disruption of both Rad51 alleles conveys embryonic stem cell and early embryonic lethality in mice (Lim and Hasty, 1996; Tsuzuki et al., 1996). Rad51 is transcribed in dividing cell lines and, in general, its expression level in tissues correlates with the proportion of cycling cells (Shinohara et al., 1993; Yamamoto et al., 1996). This is consistent with a role for Rad51 protein in mammalian cell proliferation and/or DNA metabolism. In order to study the conserved and novel functions of mammalian Rad51 protein, we have both overexpressed and downregulated human Rad51 in various cell lines. Our data suggest that, in addition to its classical function in homologous recombination, mammalian Rad51 protein is involved in regulatory aspects of the cell cycle and apoptosis.

Primary human PPL and transformed rat TGR fibroblasts (Prouty et al., 1993) were grown in D-MEM medium supplemented with 10% fetal bovine serum and antibiotics. Plasmid pEG928.1 was made by inserting the whole coding sequence of HsRad51 into the HpaI site of retroviral vector pLXSH, which contains the hygromycin phosphotransferase marker (Miller et al., 1993). This plasmid was electroporated into the ectopic retrovirus Y2 packaging line (Mann et al., 2000). Viral supernatants were used to infect PPL and TGR cultures. Stably infected cells were then subjected to selection with 150 μg/ml hygromycin B. Individual clones were expanded into cell lines and screened for elevated Rad51 protein levels by western blotting. Two clonal lines, PPL928.1-2 and TGR928.1-9, showed significant Rad51 overexpression and were selected for further studies. To synchronize Rad51-overexpressing cells, subconfluent TGR928.1-9 cultures were starved for two days in medium containing 0.5% serum. Transition from G0 to G1 phase occurred several hours after feeding the cells with medium containing 10% serum.

Plasmid pEG915, which carries the HsRad51 coding sequence inserted in frame with the 5′-terminal sequence of vector pEBVHisB (Invitrogen), was used for transient expression of HsRad51 protein in mammalian cells (Haaf et al., 1995).

Cells from Xeroderma pigmentosum type A (XP-A) patients have defects in the enzyme that is responsible for DNA lesion recognition by nucleotide excision repair and, therefore, accumulate DNA damage. The unexcised DNA lesions stimulate intrachromosomal homologous recombination (Bhattacharyya et al., 1990). Compared with normal PPL fibroblasts, SV40-transformed XP-A fibroblasts exhibit an approximately 2.5-fold elevated Rad51 protein level and an increased number (i.e. 10-20%) of nuclei with Rad51 foci, even without the induction of DNA damage (Raderschall et al., 1999).

To induce DSBs in DNA, 4 μg/ml of etoposide was added to fresh culture medium for 24 hours. Topoisomerase II binds covalently to double-stranded DNA, then cleaves both strands and reseals the cleaved complex. Etoposide interferes with this breakage and re-joining cycle, trapping the enzyme in the cleaved complex. This results in irreparable DSBs (Mizumoto et al., 1994). Ionizing radiation, that is, exposure to a ⁶⁰Co irradiator at a dose rate of 9.13 Gy per minute or to a UV-C irradiator at a dose of 10 J/m², induces mostly single-strand breaks and oxidized apurinic and apyrimidinic sites. The abasic sites are hydrolyzed by cellular endonucleases, thereby producing DNA strand breaks (Demple and Harrison, 1994). In control experiments, cultures were treated with 1 μg/ml cycloheximide for 24 hours, which kills cells by inhibiting overall protein synthesis (Waring, 1990).

To quantify the number of radiation-induced chromosome aberrations, subconfluent TGR and TGR928.1-9 cells in Petri dishes were exposed to ⁶⁰Co γ ray doses of 1-7 Gy and immediately afterwards treated with 0.2 μg/ml colcemid. Metaphases were prepared at 16 hours after irradiation according to standard procedures. The chromosome number was determined in 150 metaphases each for TGR and TGR928.1-9 and the diploid status (2n=42) confirmed for both cell lines. For aberration analyses, metaphase slides were first coded and then screened in a double-blind manner for the presence of chromosome breaks (deletions and rings) and exchanges (dicentrics) as well as for chromatid breaks (gaps and fragments) and exchanges (triradials). For each radiation dose and cell line, 200 metaphases were evaluated. Mean values and standard deviations were determined from three (0 Gy, 5 Gy) or two (1 Gy, 3 Gy, 7 Gy) independent experiments.

Antisense phosphorothioate oligodeoxynucleotides (ODNs) for Rad51 (Rad51-AS, 5′-GGCTTCACTAATTCC-3′) (368-382) and scrambled Rad51 ODNs (Rad51-SC, 5′-TCGCGATCACCTTAT-3′) (MWG Biotech) were resuspended at 100 μM in 10 mM Tris-HCl, pH 7.5 and 1 mM EDTA (Taki et al., 1996). ODNs for p21 were complementary to the region of the initiation codon (p21-AS, 5′-CCCAGCCGGTTCTGACATGGCGCC-3′) (Yu et al., 1998). Scrambled p21 ODNs (p21-SC, 5′-CCGCACGGAGCGCTGC-GTTCTACC-3′) were used as controls. Subconfluent monolayer cultures were washed with phosphate-buffered saline (PBS: 136 mM NaCl, 2 mM KCl, 10.6 mM Na₂HPO₄, 1.5 mM KH₂PO₄, pH 7.3) and incubated for eight hours with D-MEM containing the indicated ODNs (final concentration 100 nM or 400 nM) and lipofectamine (Gibco BRL). Then the cultures were washed twice with medium and grown overnight at 37°C.

The custom cDNA microarrays used for expression analysis contain >900 ESTs (selected from public databases, i.e. http://www.ncbi.nlm.nih.gov/dbEST) for the investigation of approximately 300 genes. This chip includes genes involved in cell cycle, apoptosis, DNA recombination and repair, together with 10 housekeeping genes as controls. To bind PCR products covalently onto amino silane-coated glass slides, cDNAs were amplified from plasmids using amino-modified vector primers. Using a 96-well format, PCR products were spotted onto activated slides (Guo et al., 1994) using a commercially available robot (Beecher Instruments) that deposits 5 nl of DNA solution at each spotting site, resulting in spot areas of approximately 200 μm in diameter. Aliquots (25 μg) of poly(A) RNA from PPL and PPL928.1-2 cells were reverse transcribed with the Superscript II kit (Gibco BRL) using an oligo(dT) primer in the presence of either Cy3-dUTP or Cy5-dUTP. The differentially fluorescent-labeled targets were hybridized together on cDNA microarrays, and the fluorescent intensities for the two wavelengths at each spot were read by a laser scanner (GMS 418 array scanner). Image analysis was carried out with a custom-made software program that runs as an extension on IP Lab Spectrum software (Chen et al., 1997). The result file contains ratios of mean intensities per pixel for individual spots consisting of at least 20 pixels each. The background fluorescence was substracted from all spots.

HsRad51 protein, expressed in E. coli, was isolated and used for preparation of rabbit polyclonal antibodies (Haaf et al., 1995). Goat polyclonal antibodies against the entire human p53 protein (FL-393) and against the C-terminus of human p21 (C-19) were purchased from Santa Cruz Biotechnology. Cell extracts were resolved by electrophoresis on 12% SDS-polyacrylamide gels and then transferred to nitrocellulose membranes. The resulting filters were blocked overnight with 5% nonfat dried milk, incubated with the appropriately diluted primary antibodies for one hour, incubated with horseradish peroxidase-conjugated anti-rabbit or anti-goat IgG (Dianova) and washed. Antibody binding was visualized by chemiluminescence (ECL RPN 2209; Amersham). To compare the protein levels in different cell substrates, all filters were re-incubated with rabbit antibodies against β-actin (Sigma). The intensity of the Rad51 signals (and other primary antibody signals) was equilibrated to the intensity of the β-actin signals using PCbas2.0 software.

Quantification of the relative number of DNA breaks was based on random primer extension of 3′-OH ends by Klenow fragment polymerase (Basnakian and Jill James, 1996). 100,000 cells each of PPL and PPL928.1-2 were seeded in culture flasks and grown for 24 hours in medium containing 4 μg/ml etoposide. For each cell line, two samples, each containing 0.5 μg high-molecular weight DNA in a volume of 12 μl ddH₂O, were processed. 3′-OH end-containing DNA fragments generated in vivo through single-strand and double-strand breaks were separated by heat denaturation. After reassociation, these DNA fragments served as a primer and the excess high-molecular weight DNA as a template. In a random primer extension reaction, Klenow enzyme incorporated [γ-³²P]dCTP into newly synthesized DNA. Incorporation was linearly proportional to the number of DNA breaks present in the sample. The reaction mixture for 10 samples consisted of 25 μl of cold dNTPs (0.5 mM each of dATP, dGTP, and dTTP), 4.5 μl of 33 μM cold dCTP, 0.5 μl of [γ-³²P]dCTP (labeled to a specific activity of 3,000 Ci/mmol), 1 μl (5 Units) Klenow enzyme and 94 μl ddH₂O. Fifteen μl of this reaction mixture were added to each DNA sample and incubated at 16°C for 30 minutes. The reaction was stopped by adding 25 μl of 12.5 mM EDTA, pH 8.0. The radioactively labeled DNA fragments were purified with a PCR purification kit (Qiagen). Four 5 μl aliquots of each DNA sample were counted in a Packard liquid scintillation counter.

Harvested cells were washed and resuspended in PBS. Aliquots (10⁵ cells in 0.5 ml PBS) were centrifuged onto clean glass slides using a Shandon Cytospin. Immediately after cytocentrifugation, the preparations were fixed in absolute methanol for 30 minutes at –20°C and then rinsed in ice-cold acetone for a few seconds. Following three washes with PBS, the preparations were incubated at 37°C with rabbit anti-HsRad51 or goat anti-p21 antibodies diluted 1:100 with PBS in a humidified incubator for 30 minutes. The slides were then washed in PBS another three times for 10 minutes each and incubated for 30 minutes with fluorescein-isothiocyanate (FITC)- and/or Cy3-conjugated secondary antibodies, appropriately diluted with PBS. After three further washes with PBS, the preparations were counterstained with 1 μg/ml 4,6-diamidino-2-phenylindole (DAPI) in 2×SSC for one minute. The slides were mounted in 90% glycerol, 0.1 M Tris-HCl, pH 8.0, and 2.3% 1,4-diazobicyclo-2,2,2-octane.

Images were taken with a Zeiss epifluorescence microscope equipped with a thermoelectronically cooled charge coupled device camera (Photometrics CH250), which was controlled by an Apple Macintosh computer. Gray scale images were pseudocolored and merged using Oncor Image and Adobe Photoshop software.

Annexin V binds in a calcium-dependent manner to phosphatidylserine, which is translocated from the interior side of the plasma membrane to the outer leaflet during the early stages of apoptosis (Van Engeland et al., 1996). Fluorescein-conjugated annexin V-Fluos (Boehringer Mannheim) was added to the cell culture medium at a final concentration of 1.5 μg/ml for three minutes. Harvested cells were washed twice with fresh culture medium to remove excess annexin V, resuspended in 10 mM Hepes/NaOH, pH 7.4, 140 mM NaCl and 5 mM CaCl₂ at a density of approximately 10⁵ cells/ml and centrifuged onto glass slides. After fixation in methanol for 30 minutes, the preparations were incubated with rabbit anti-Rad51 antiserum and Cy3-conjugated anti-rabbit IgG.

In order to visualize cycling cells in situ, 10 μg/ml of 5-bromodeoxyuridine (BrdU) was added to the culture medium either two hours or 24 hours before cell harvesting. In place of thymidine, BrdU is incorporated into the DNA of replicating cells. After Rad51 protein staining, the immunofluorescent preparations were fixed overnight in a 3:1 mixture of methanol and acetic acid at –20°C. Since the anti-BrdU antibody only recognizes its epitope if the BrdU-substituted chromosomal DNA is in the single-stranded form, the slides were denatured in 70% formamide, 2×SSC for one minute at 80°C and then dehydrated in an alcohol series. BrdU incorporation was visualized by indirect anti-BrdU antibody (Boehringer Mannheim) staining. Only cells with intense BrdU labeling of the entire nucleus were considered BrdU-positive and scored as cycling cells.

Cells were grown on coverslips, fixed in methanol at –20°C for 30 minutes and then in acetone for five minutes. After drying, the cells were treated with 0.5% Triton X-100 in PBS, washed 3×5 minutes in PBS and incubated for one to four hours at 37°C with rabbit anti-Rad51 antibodies. After washing for 3×5 minutes with PBS, the cells were incubated overnight at 4°C with anti-rabbit IgG coupled to 12 nm colloidal gold particles. The cells were washed again for 10 minutes with PBS and fixed with 2% glutaraldehyde in 50 mM cacodylate buffer, pH 7.2, on ice, followed by treatment with 2% OsO₄ in H₂O for two hours at room temperature. Samples were dehydrated in an ethanol series and after propyleneoxide treatment embedded in Araldite (Agar Scientific). Ultrathin sections were stained for 20 minutes in 4% uranyl acetate and for 15 minutes in lead citrate and viewed with a Phillips CM100 electron microscope.

Human PPL fibroblasts and rat TGR fibroblasts were stably infected with plasmid pEG928.1 in which the retrovirus-based vector pLXSH carries the whole coding sequence of HsRad51. Western blotting showed an approximately 1.5-fold increase in total Rad51 protein in the human Rad51-overexpressing cell line, PPL928.1-2 (Fig. 1A) and an approximately 2.5-fold overexpression of Rad51 in rat TGR928.1-9 cells (Fig. 1B) compared with the respective parental controls.

The affinity-purified antibodies used for western analysis detected two Rad51 bands in human wild-type PPL and overexpressing PPL928.1-2 cells (Fig. 1A). The upper 39-kDa band had the same mobility as HsRad51 expressed in E. coli (data not shown) and, therefore, corresponds to the full-length Rad51 protein. The lower approximately 37-kDa band most probably represents a cleavage product following proteolysis, a mechanism that has been described for Rad51 protein in apoptotic cells (Flygare et al., 1998). As the net amount of full-length Rad51 protein decreased in etoposide-treated cells, we conclude that the formation of nuclear Rad51 foci in response to DNA damage is not due to de novo protein synthesis (Haaf et al., 1995; Haaf et al., 1999).

By immunofluorescence staining, 1-20% of cells in stably Rad51-overexpressing cell populations, PPL928.1-2 and TGR928.1-9, and in transiently tranfected PPL cultures showed nuclear foci which were indistinguishable from DNA-damage-induced Rad51 foci in wild-type cells. In addition, up to 20% of nuclei from these untreated cells displayed elongated higher-order structures up to 20-30 μm in length. Some nuclei were filled with a network of linear Rad51 structures (Fig. 2A). In contrast, linear Rad51 structures were only rarely (<0.1%) observed in wild-type PPL and TGR cells after etoposide treatment (data not shown). Immunoelectron microscopy of Rad51-overexpressing cells, which were labeled with colloidal gold coupled to anti-Rad51 antibody, demonstrated bundles of linear Rad51 filaments reminiscent of presynaptic Rad51 complexes (Fig. 2B). As these elongated bundle-like structures with a diameter of approximately 50 nm occurred in the absence of DNA damage, they are likely to reflect self-assembly of a multimeric form of Rad51 protein (Donovan et al., 1994; Krejci et al., 2001). Evidently, increased protein levels can induce a dramatic redistribution of Rad51 in the mammalian cell nucleus.

As Rad51-overexpressing cultures appeared to grow much slower than the corresponding parental lines, we analyzed the cell cycle progression of subconfluent cell populations by incorporation of the base analog BrdU into replicating cells and subsequent anti-BrdU antibody staining (Fig. 3A, red). When BrdU was added to exponentially growing (unsynchronized) cells two hours before cell harvest, PPL928.1-2 and TGR928.1-9 cultures exhibited a significantly lower percentage of BrdU-positive (replicating) cells than PPL and TGR (χ² test, P<0.001) (Fig. 3B). Significant differences (χ² test, P<0.001) in the number of cycling cells between Rad51 overexpressers and parental controls were also seen after etoposide treatment. Thus, we conclude that Rad51 protein imposes a growth delay in overexpressing cultures.

Simultaneous immunofluorescence staining of overexpressed Rad51 protein (Fig. 3A, green) and replicating DNA (Fig. 3A, red) demonstrated that Rad51 foci and higher-order structures almost exclusively occur in non-cycling cells. For both PPL928.1-2 and TGR928.1-9 cultures, several hundred nuclei each with prominent Rad51 foci, and several hundred nuclei each without cytologically detectable foci, were scored for DNA replication synthesis. When BrdU was incorporated into replicating DNA for 24 hours (approximately one population doubling), 92% and 83% of the nuclei of PPL928.1-2 and TGR928.1-9 with Rad51 protein structures were found to be BrdU-staining negative, respectively. In contrast, 88% and 94% of the cells without Rad51 structures from the same cultures showed BrdU incorporation, which is indicative of cycling cells.

Rat TGR cells are capable of normal physiological withdrawal into the quiescent (G0) phase of the cell cycle and resumption of growth following the appropriate stimuli (Prouty et al., 1993). In TGR928.1-9 cells, G0 arrest upon serum starvation dramatically induced nuclear Rad51 foci and higher-order structures (Table 1). Synchronous re-entry into the cell cycle after re-feeding reduced the percentage of Rad51-foci-positive cells to very low levels. New cell cycle arrest upon contact inhibition after three population doublings increased the number of cells with Rad51 foci again.

Both PPL928.1-2 and TGR928.1-9 express the ectopic HsRad51 gene from a vector-based retroviral LTR promoter. Although one might expect some variation in expression levels of the transfected gene between individual cells, it seems plausible that overexpressed Rad51 protein is present in all cells of the transfected clonal population, although it is not necessarily concentrated in foci in all or most cells of a culture. We conclude that Rad51-foci-positive cells represent a distinct subpopulation of Rad51-overexpressing cells that are arrested or delayed during the cell cycle, most probably during G1 phase.

Rad51 seems to be required for the elimination of chromosomal breaks that arise during DNA replication (Sonoda et al., 1998). To test whether overexpression of Rad51 protein decreases DNA damage in vivo, we quantified the relative number of DNA breaks in etoposide-treated PPL and PPL928.1-2 cells. The extent of radioactive nucleotide incorporation into DNA by random oligonucleotide-primed synthesis (Basnakian and Jill James, 1996) depends on the initial number of free 3′-OH ends generated by single-strand or double-strand-breaks in the living cells. The incorporated radioactivity (scintillation counts) was measured in four aliquots of two independent DNA samples from each cell line (Table 2). Rad51-overexpressing cells exhibited a significantly lower number of DNA breaks than untreated controls (t test, P<0.001).

A lower level of DNA breaks in Rad51-overexpressing cells is consistent with the observation of fewer chromosome aberrations in these cultures. In order to assess the effects of overexpressed Rad51 protein on chromosomal stability, both chromosome-type and chromatid-type aberrations were scored in TGR928.1-9 and parental TGR cells following treatment with different doses of γ irradiation. The number of chromosome breaks (gaps, deletions and rings) and exchanges (dicentrics) was the same in the two populations (Fig. 4). However, the number of chromatid breaks (gaps and fragments) and exchanges (triradials), which result from DSBs generated during or after DNA replication, was significantly reduced in Rad51-overexpressing TGR928.1-9 cells (t test, P<0.05).

Cells entering apoptosis were stained with annexin-V–fluorescein (Fig. 5A, green). Subsequently, Rad51 was detected in nuclei by anti-Rad51 antibody labeling (Fig. 5A, red). Surprisingly, only very few annexin-V-positive apoptotic cells contained nuclear Rad51 foci (Fig. 5B, black bars) following etoposide treatment or UV irradiation, indicating that the presence of Rad51 foci excludes or protects cells from undergoing apoptosis after DNA damage. This anti-apoptotic role for Rad51 foci was confirmed in Rad51-overexpressing lines, PPL928.1-2 and TGR928.1-9, which showed a significantly lower percentage of apoptotic cells (χ² test, P<0.001) (Fig. 5B, black and gray bars) than their respective parental lines after DNA damage. As a control, cells were exposed to the protein synthesis inhibitor cycloheximide, which induces apoptosis without causing DNA damage (Waring, 1990). There was no difference in the number of apoptotic cells in Rad51-overexpressing (13.0%) and wild-type cells (12.6%), indicating that Rad51 foci formation specifically increases resistance to DNA-damage-induced apoptosis. This suggests a cellular commitment after DNA damage to either apoptosis or Rad51-mediated protection through recombinational repair and/or cell cycle delay.

In order to further investigate the possible role of Rad51 in cell cycle control, we quantified the amount of the Cdk inhibitor p21 by western blotting. The net amount of p21 increased approximately twofold in Rad51-overexpressing PPL928.1-2 cells compared with wild-type PPL cells, even without induction of DNA damage (Fig. 6A). Increased p21 levels in Rad51-overexpressing cells were confirmed by immunofluorescence staining: 18% of subconfluent PPL cells versus 35% of PPL928.1-2 cells displayed strong diffuse nuclear anti-p21 antibody staining (data not shown). A two- to threefold increase of p21 protein was also observed in etoposide-treated wild-type cells, whereas the already elevated p21 levels in Rad51-overexpressing cells were not elevated further after etoposide treatment (Fig. 6A). Similar to the situation in human Rad51 overexpressers, we also found an approximately twofold p21 upregulation in untreated rat TGR928.1-9 cells (data not shown).

Since the tumor suppressor p53 is usually critical for p21 upregulation in response to DNA damage (Macleod et al., 1995), we analyzed whether enhanced p21 expression in Rad51-overexpressing cells is due to increased p53 protein levels. However, neither etoposide-treated wild-type PPL cells nor treated or untreated Rad51-overexpressing PPL928.1-2 cells showed clearly increased p53 levels (Fig. 6A). This suggests that the increased amount of p21 protein induced by Rad51 overexpression is independent of changes in p53 protein levels, as had been demonstrated previously in other systems (Gartel and Tyner, 1999).

In addition to western analyses of protein levels, custom-made cDNA microarrays were used to quantify mRNA expression. The mRNA levels were evaluated for p21 and several other DNA repair genes by calculating the ratio of spot (EST) intensities in PPL928.1-2 cells to those in wild-type PPL cells. For each gene analyzed, data from three independent ESTs (on the same chip) and three different chip hybridization experiments were averaged. Since the PPL928.1-2/PPL ratio for p21 (approximately 2.5) was significantly increased (t test, P<0.01) compared with housekeeping genes, such as G6pdh (1.3) and Nars (1.2) (Fig. 6B and data not shown), we conclude that p21 is upregulated by Rad51 at the transcriptional level. In contrast to increased p21 mRNA transcription and consistent with our immunoblotting experiments, p53 transcription (1.2) was not increased by Rad51. Thus, microarray analyses confirmed our western blotting data that p21 expression is specifically increased in Rad51-overexpressing cells.

Two genes on the chip, Rad52 (Benson et al., 1998) and c-Abl (Yuan et al., 1998), are known to interact with Rad51. However, only expression of the c-Abl oncogene (1.8) was increased in Rad51-overexpressing cells, whereas the Rad52 mRNA levels (1.0) were normal (Fig. 6B). Interaction of the c-Abl oncogene product with Rad51 may be required for the correct post-translational modification of Rad51 and the assembly of DNA repair protein complexes (Yuan et al., 1998; Chen et al., 1999).

Since Rad51 overexpression resulted in a net increase in p21 protein (Fig. 6A), we hypothesized that downregulation of endogenous Rad51 might decrease the amount of p21. Rad51 protein expression was reduced by 60-70% in wild-type PPL cells treated with 400 nM Rad51 antisense ODN compared to untreated or mock-treated cells (Fig. 7A, gray bars). This Rad51 suppression was associated with a dramatic decrease in the p21 protein level (Fig. 7A, black bars), thus supporting a regulatory link between Rad51 and p21 expression. We have demonstrated above that Rad51-foci-containing cells are excluded or protected from apoptosis. Therefore, we tested whether experimental suppression of endogenous Rad51 might sensitize cells to DNA-damaging agents. Indeed, using annexin V staining of Rad51 antisense-treated PPL cells, we observed a significantly increased number of apoptotic cells after etoposide treatment (Fig. 7B), supporting our hypothesis of a protective role for constitutitively overexpressed Rad51 protein in DNA-damage-induced apoptosis.

To establish a functional role for p21 in Rad51-mediated cellular processes, human wild-type PPL and Rad51-overexpressing PPL928.1-2 cells were treated with 400 nM p21 antisense or scrambled control ODNs. In addition, we analyzed XP-A cells which, owing to their inherent excision repair deficiency, exhibit increased amounts of Rad51 protein and Rad51 foci (Raderschall et al., 1999). Compared to untreated and mock-treated cells, p21 antisense-treated cultures showed a significantly reduced number of Rad51-foci-positive nuclei both without induction of DNA damage and after etoposide treatment (Table 3). In a conceptually related experiment, etoposide-treated p21^–/– fibroblast cultures from knockout mice (Deng et al., 1995) exhibited considerably fewer Rad51-foci-positive nuclei (4%) than normal mouse fibroblasts (23%). Since by western blot analysis, p21^–/– and wild-type cells exhibited similar Rad51 protein levels (data not shown), the p21-dependent defect in Rad51 foci formation is not caused by reduced amounts of Rad51 protein. We propose a functional role for p21 in the dynamic localization of Rad51; p21-mediated cell cycle delay may facilitate the redistribution of dispersed Rad51 protein to nuclear foci both in normal cells after DNA damage and in Rad51-overexpressing cells.

DSB repair in mammalian cells has been thought to occur primarily through non-homologous end-joining (Weaver, 1995). However, recent data demonstrate the essential contribution of homologous recombination to the repair of DNA damage (Liang et al., 1998; Takata et al., 1998). The prerequisite for homologous recombination is a DNA template with sequence similarity or identity to the damaged DNA molecule. In the majority of cases, the sister chromatid acts as a repair template in mammalian cells. Homologous recombination between (homologous or heterologous) chromosomes appears to be several orders of magnitude less frequent than that between sister chromatids of the same chromosome. This may explain why homologous recombination plays a predominant role in DSB repair in highly replicating cells. The outcome of homologous recombination is mainly gene conversion, which is not associated with crossing over (Johnson and Jasin, 2001). Rad51-deficient chicken lymphocytes in which a human Rad51 transgene was inactivated showed increases in chromosome breaks and cell death (Sonoda et al., 1998), whereas Rad51-overexpressing Chinese hamster ovary cells showed a stimulation of homologous recombination and increased resistance to ionizing radiation (Vispé et al., 1998). Our study demonstrates that constitutive upregulation of Rad51 protein increases DNA repair and cell survival after DNA damage, whereas Rad51 antisense inhibition sensitizes cells to DNA-damage-induced apoptosis.

DNA-damage-induced Rad51 protein foci are thought to function as repair complexes in mammalian cells (Haaf et al., 1995; Haaf et al., 1999). As Rad51 foci are located at ssDNA regions formed after DNA damage, these may be the nuclear areas where Rad51 forms filaments on ssDNA tails (Raderschall et al., 1999). The resulting ssDNA-Rad51 filaments are the key elements for promoting pairing and strand exchange between ssDNA and homologous double-stranded DNA (Sung, 1994; Baumann et al., 1996; Gupta et al., 1997). However, most Rad51 filaments that are formed as intermediates of homologous recombination may be too short (<1 kb) to be seen by immunofluorescence staining. In addition, Rad51 foci representing recombination intermediates should disappear after completion of homologous recombination. Thus, immunocytochemically detectable Rad51 foci may contain unusually large ssDNA-Rad51 filaments that did not find a partner for homologous recombination. On the other hand, DSB repair in mammalian cells may involve much larger chromatin domains than in yeast. In contrast to DNA-damage-induced Rad51 foci, the linear higher-order structures in Rad51-overexpressing cells are formed in the absence of ssDNA, most probably by self-interacting Rad51 molecules (Donovan et al., 1994; Krejci et al., 2001). Thus, Rad51 distribution in the mammalian cell nucleus does not only depend on the presence of ssDNA but also on protein stoichiometry. A twofold increase in the net amount of Rad51 protein in overexpressing cells is already sufficient to induce formation of nuclear foci and linear higher-order structures that are reminiscent of redistribution of endogenous Rad51 protein after DNA damage. In addition, we found that relatively moderate (two- to sixfold) upregulation of endogenous Rad51 in tumor cell lines is associated with Rad51 foci formation (Raderschall et al., 2002).

When Rad51 protein forms nuclear foci either by interacting with damaged DNA (in wild-type cells) or with itself (in overexpressing cells), cells appear to be unable to replicate. It is plausible that the repairosome-type assembly of Rad51 after DNA damage blocks, at least temporarily, the cell cycle and apoptosis in order to allow cells to try to repair DNA damage. Since the primary function of Rad51 is recombinational repair, Rad51 may be involved in signaling for DNA-damage-induced cell cycle arrest through p21. There are several possible explanations why overexpressed Rad51 protein forms higher-order structures and delays the cell cycle even in the absence of DNA damage. The constitutive overproduction of Rad51 in stably transfected cells or tumors might destroy the balance between different components of the DNA repair system. Tumor suppressor proteins such as p53 (Stürzbecher et al., 1996) and Brca2 (Sharan et al., 1997) interact with Rad51 directly and are thought to keep Rad51 in an inactive monomeric state. When Rad51 molecules are overexpressed in cells, they may form multimeric complexes because of potentially limiting concentrations of these interacting tumor suppressors. This artificial aggregation of Rad51 protein could disturb normal nuclear functions, including cell cycle progression and apoptosis. On the other hand, it is possible that unphysiologically high Rad51 concentrations in the nucleus induce binding of dispersed Rad51 protein to transcribing and/or replicating DNA, which may also be in a single-stranded form, and thereby signal cell cycle arrest. The biochemical association between Rad51 and RNA polymerase II transcription complexes (Maldonado et al., 1996) also argues in favor of the notion that Rad51-mediated recombination is coupled intimately with transcription.

Constitutive Rad51 upregulation is associated with a twofold overall increase of p21 mRNA and p21 protein levels (prior to DNA damage), whereas Rad51 downregulation dramatically decreases the p21 protein level. As only a subset (30-40%) of Rad51-overexpressing cells in a culture show strong p21 immunofluorescence, p21 induction may be much greater at the single cell level. At present it is not clear how Rad51 can, directly or indirectly, upregulate p21. Rad51 protein interacts with the tumor suppressor Brca1, which is involved in p21 upregulation (Somasundarm et al., 1997). Sequestration of Brca1 to Rad51 foci (Scully et al., 1997) may be essential for the activation of p21 transcription by Brca1 after DNA damage or in overexpressing cells. Although it is plausible that p21 transcription induced by Rad51 arrests the cell cycle, we cannot exclude the formal possibility that Rad51 causes arrest in some other way and p21 upregulation is a secondary phenomenon.

Rad51 foci are multi-component functional structures for DNA repair, which do not have a set stoichiometry, but interact (colocalize) dynamically with other mammalian proteins, including RPA (Golub et al., 1998), Rad52 (Liu et al., 1999), Rad54 (Tan et al., 1999), Brca1 (Scully et al., 1997), Brca2 (Sharan et al., 1997), Xrcc2 (O’Regan et al., 2001) and Xrcc3 (Bishop et al., 1998), which all promote Rad51-mediated homologous recombination. Previously, it has been shown that defects in Rad54 (Tan et al., 1999), Xrcc2 (O’Regan et al., 2001), Xrcc3 (Bishop et al., 1998) and other genes, which are required for assembly and stabilization of multimeric Rad51 protein complexes, interfere with Rad51 foci formation. Our experiments provide evidence that p21 is another such protein. When p21 is experimentally downregulated, formation of repairosome-type Rad51 foci is disrupted.

As p21 is a well-known inhibitor of G1 Cdks and a major mediator of G1 arrest after DNA damage (Harper et al., 1993; Deng et al., 1995), Rad51-foci-positive cells may be arrested during G1 phase. This would be consistent with the observation that G0/G1 arrest upon serum starvation induces formation of overexpressed Rad51 protein foci. However, two other studies (Tashiro et al., 1996; Scully et al., 1997) did not observe endogenous Rad51 foci in G1 phase cells. During G1 phase, Rad51-mediated recombinational repair requires pairing and strand exchange between the spatially separated homologous chromosomes. As this is far more difficult than homologous recombination between closely adjacent sister chromatids of G2 phase chromosomes, it is generally assumed that the recombination/repair function of mammalian Rad51 protein becomes more important in G2 phase (Takata et al., 1998; Johnson and Jasin, 2001). Indeed, UV microbeam and whole-cell irradiation experiments demonstrated preferential association of Rad51 foci with post-replicative chromatin during S phase (Tashiro et al., 2000). In addition, the reduction of chromatid-type aberrations (but not of chromosome-type aberrations) in Rad51-overexpressing cells is consistent with an increased DNA recombination and repair function of Rad51 in S/G2 phase. A similar function has been demonstrated for the Rad51 family member Xrcc3 (Liu et al., 1998). Nevertheless, it is possible that homologous recombination, which repairs DSBs with higher fidelity than non-homologous end-joining, also plays a role for the DNA damage repair in G1 phase cells.

Our finding that Rad51 protein expression influences cell cycle progression and apoptosis has important implications for both normal developmental and pathological cellular processes in mammals. Most strikingly, failed attempts to generate homozygous MmRAD51^–/– embryonic stem cells and an embryonic lethal phenotype in knockout mice demonstrate the requirement of Rad51 for normal cell proliferation and early development (Lim and Hasty, 1996; Tsuzuki et al., 1996). In contrast, yeast mutants carrying ScRad51 deletions are viable despite their sensitivity to DNA-damaging agents (Shinohara et al., 1992). Increased Rad51 protein levels in tumors (Maacke et al., 2000; Raderschall et al., 2002) and potential functional associations of Rad51 protein with the tumor suppressors Atm (Chen et al., 1999), p53 (Stürzbecher et al., 1996), Brca1 (Scully et al., 1997) and Brca2 (Sharan et al., 1997), the oncogene product c-Abl (Yuan et al., 1998; Chen et al., 1999), and the Blm helicase (Wu et al., 2001), all suggest that Rad51 upregulation confers an advantage(s) to tumor cells. Increased Rad51 protein levels could lead to uncontrolled recombination, genome instability and increased resistance of tumors to DNA-damaging agents. As downregulation of Rad51 protein by antisense ODNs or small molecules sensitizes cells to DNA-damaging agents, this may be a promising strategy for tumor therapy (Ohnishi et al., 1998).

It is important to emphasize that stable transfectants constitutively overexpressing Rad51 have adapted to the increased protein levels during clonal selection and long-term culturing. In view of this, the effects of Rad51 overexpression on cell proliferation and apoptosis in cell lines PPL928.1-2 and TGR928.1-9 may reflect the situation in transformed (Xia et al., 1997) and tumor cells (Maacke et al., 2000) much more closely than that in cells with acute ectopic Rad51 overexpression. Induction of a Rad51 transgene under the control of a repressible promoter resulted in a decreased growth rate in a dose-dependent manner (Flygare et al., 2001). However, in contrast to our study, abrupt (up to tenfold) Rad51 overexpression was associated with an increased apoptotic rate. These cytotoxic and cytokinetic effects of acute Rad51 overexpression may protect multicellular organisms from hyperrecombination and genomic instability. In clonally selected cells that become resistant to these negative effects, the increased Rad51 protein is most likely to be implicated in malignant transformation and/or tumor progression.

We thank Susanne Freier, Karen Stout, and Gerhild Lüder for excellent technical assistance, Sandra Prietz and Roland Kirchner for help with the microarray analysis and Dough Brash for providing p21^–/– fibroblasts. This study was supported by research grants Ha1374/4-3 (to T.H.) from the Deutsche Forschungsgemeinschaft and F14PCT950010 (to E.F.) from the European Community.