Synthetic lethal genetic interactions between Rad54 and PARP-1 in mouse development and oncogenesis

Mutations in DNA repair pathways are frequent in human cancers. Hence, gaining insights into the interaction of DNA repair genes is key to development of novel tumor-specific treatment strategies. In this study, we tested the functional relationship in development and oncogenesis between the homologous recombination (HR) factor Rad54 and Parp-1, a nuclear enzyme that plays a multifunctional role in DNA damage signaling and repair. We introduced single or combined Rad54 and Parp-1 inactivating germline mutations in Ptc1 heterozygous mice, a well-characterized model of medulloblastoma, the most common malignant pediatric brain tumor. Our study reveals that combined inactivation of Rad54 and Parp-1 causes a marked growth delay culminating in perinatallethality, providing for the first time evidence of synthetic lethal interactions between Rad54 and Parp-1 in vivo. Although the double mutation hampered investigation of Rad54 and Parp-1 interactions in cerebellum tumorigenesis, insights were gained by showing accumulation of endogenous DNA damage and increased apoptotic rate in granule cell precursors (GCPs). A network-based approach to detect differential expression of DNA repair genes in the cerebellum revealed perturbation of p53 signaling in Rad54-/-/Parp-1-/-/Ptc1+/-, and MEFs from combined Rad54/Parp-1 mutants showed p53/p21-dependent typical senescent features. These findings help elucidate the genetic interplay between Rad54 and Parp-1 by suggesting that p53/p21-mediated apoptosis and/or senescence may be involved in synthetic lethal interactions occurring during development and inhibition of tumor growth.


INTRODUCTION
The complex network of DNA repair system in mammals counteracts genome insult and maintains genomic stability. DNA repair failure increases mutations frequency and affects cell-cycle regulation, promoting tumorigenesis; in fact, inactivating mutations in DNA repair pathways are frequent in cancer. On the other hand, DNA repair defects provide new therapeutic opportunities to treat cancer through DNA-damageinducing radiation and chemotherapies. Identification of synthetic interactions, i.e., negative genetic combination of mutations in two or more genes, leading to severe slow growth or lethality compared to single mutants,

Research Paper
has grown in popularity in the DNA damage field, with the finding that poly(ADP-ribose) polymerase (Parp-1) inhibitors show synthetically lethal effects in homologous recombination (HR) defective tumors [1].
PolyADP-ribosylation acts in DNA-damage repair response and maintenance mechanisms of genomic stability. Several DNA repair pathways, including baseexcision repair and double strand break (DSB) repair, involve Parp functions. Inhibition of Parp-1 results in an increase in single strand breaks (SSBs) that then collapse replication forks into toxic one-ended DSBs [1,2], which are substrates for HR [3,4]. In addition, through enzymedependent chromatin remodeling and enzyme-independent motif recognition, Parp-1 also plays important roles in regulating gene expression [5]. Furthermore, Parp-1 binding motifs may be readily found in promoter elements of DNA repair genes, linking DNA repair and transcription functions of Parp-1 [5]. The critical role of Parp-1 in maintenance of genomic stability is reflected in its frequent upregulation in cancer [6,7], and by hypersensitivity of Parp-1 null animals toward the mutagenic effect of DNA damaging agents [8].
The Rad54 gene encodes a dsDNA-dependent ATPase of the Swi2/Snf2 family [9], which interacts directly with Rad51 [10,11], and stimulates its DNA exchange activity [12]. Rad54 promotes chromatin remodeling, Rad51 displacement from double-stranded DNA, binds Holliday junctions and drives their branch migration [13]. Although disruption of other genes involved in HR leads to embryonic lethality, adult Rad54 −/− mice are viable and fertile, providing a valuable model to study the effects of attenuated HR. Interestingly, several point mutations in conserved regions of the Rad54 gene family have been found in primary tumors (e.g., breast and colon carcinomas and lymphomas) [14,15], and approximately half of epithelial ovarian cancers have alterations in genes regulating HR repair [16].
Most DNA repair pathways are complex, involving many proteins working in discrete consecutive steps. Therefore, their balanced and coordinated expression is important to avoid erroneous repair that might result from excessive base removal and DNA cleavage. In addition, as DNA damage is a likely factor in promoting central nervous system (CNS) pathophysiology, DNA repair may be an important mechanism for maintenance of normal physiological function. In this study, to investigate the functional relationship between the HR factor Rad54 and Parp-1 in development and oncogenesis in vivo, we introduced single or combined Rad54 and Parp-1 inactivating mutations in a well characterized cancer model, the Ptc1 heterozygous mice. Mice in which one copy the Ptc1 gene has been inactivated (Ptc1 +/-) are developmentally nearly normal, but show a marked predisposition to tumor development, including medulloblastoma (MB), a frequent pediatric malignant tumor of the cerebellum [17,18]. Our previous work showed that neonatal irradiation dramatically increases the frequency of MB [19,20]. We also showed that combined loss of Ptc1 and Rad54 or of Ptc1 and Parp-1 increased radiation induced MB [21,22]. Here, we analyzed the effects of combined loss of Rad54 and Parp-1 in the Ptc1 mouse model of spontaneous and radiation-induced cancer, highlighting novel synthetic lethal interactions during development and tumorigenesis, and dissecting the underlying cellular and molecular mechanisms.

Rad54, Parp-1 and Ptc1 mutant mice
Defects in DNA damage signaling are frequently associated with neurodegeneration, neurodevelopmental disease and brain tumors. We therefore sought to determine the oncogenic potential of combined genetic disruption of Rad54 and Parp-1 in the CNS by knocking out these genes in Ptc1 +/mice. To this aim, Rad54 +/-/ Parp-1 +/-/Ptc1 +/and Rad54 +/-/Parp-1 +/-/Ptc1 +/+ mice were intercrossed, and the progeny genotyped. For simplification, wt genotypes are hereafter omitted from the label, unless differently specified. Mutants for Rad54 Irrespective of Ptc1 genotype, no mice with compound Rad54/Parp-1 inactivation were found at weaning, showing strong synthetic lethal interaction between Rad54 and Parp-1 genes in vivo, in agreement with existing in vitro data. To look for haploinsufficient interactions, compound mutant mice with varying gene dosage of Rad54 and Parp-1 were analyzed for survival and cerebellum tumorigenesis, with or without irradiation. Because only Ptc1 +/mice are prone to MB, the effects of Rad54 and Parp-1 inactivation was evaluated only in mice with Ptc1 +/genotype.
Loss of the normal Ptc1 allele is the causative event in MB development in Ptc1 +/mice [24]. To test if single or combined Rad54 and Parp-1 deficiency alters the typical chr-13 LOH pattern in MB, we performed microsatellite analysis with a set of markers spanning the length of chr-13, where Ptc1 gene is located ( Figure S1). Within the sample size, no obvious differences were observed in Ptc1 LOH pattern in single or compound Rad54 and Parp-1 mutants.

Combined inactivation of Rad54 and Parp-1 causes growth retardation and perinatal lethality
The complete lack of Rad54 -/-/Parp-1 -/null mice in our triple cross at weaning age prompted us to investigate the effect of this genetic interaction at earlier developmental times. Analysis of the genotype distribution in the offspring at postnatal day 1 (P1) revealed that, of 764 newborns derived from heterozygous mating, only 10 and 13 in wt and Ptc1 +/genetic backgrounds, respectively, were Rad54 -/-/Parp-1 -/compound mutants, far below the 24 mice expected based on a Mendelian inheritance ratio of 25%, although this trend reached statistical significance only in Ptc1 +/+ mice (Table 1). In addition, double knockout Rad54 -/-/Parp-1 -/pups at P1 and P7 showed significantly smaller body size compared to relative littermates (Figure 2A and 2C). In fact, inactivation of both genes significantly reduced body weight at P1 (by 34% in wt and 17% in Ptc1 +/genotypes) and P7 (by 37% in wt and 20% in Ptc1 +/genotypes), showing that combined loss of Rad54 and Parp-1 causes severe growth retardation (Figure 2A-2D). Moreover, 45% and 29% of the double mutants on Ptc1 +/and wt background, respectively, dyed during the first fortnight of life and none survived to weaning age (data not shown).

Parp-1 on cerebellum morphogenesis
To further clarify the underlying mechanisms of synthetic lethal interactions of Rad54 and Parp-1 perturbation, we focused on the cerebellum, from which MB originates. The well-characterized developmental processes and stereotypical foliation pattern make this organ particularly amenable to the study of factors affecting development, leading to abnormal morphology or function. By measuring the cross-sectional area of developing cerebellum at P1 and P7 we show that, in line with generalized growth delay, significant decrease in size was observed in cerebella from Rad54 -/-/Parp-1 -/mutants at P1 and P7 relative to Ptc1 +/+ mice (by 36% at P1, P = 0.0035, and 31% at P7, P = 0.0078; Figure 3). Similarly decreased cerebellum area was observed in double null mice with Ptc1 +/genotype (by 34% at P1, P = 0.0007 and 43% at P7, P = 0.0012) ( Figure 3A-3F). In keeping with growth delay, cerebella from Rad54 -/-/Parp-1 -/-/Ptc1 +/mice showed shallow principal fissures at P1 relative to Ptc1 +/mice (asterisks in Figure 3C). In addition, delayed lobularization was evident at P7 ( Figure 3E and 3F).

Proliferation, DNA damage and cell death in CGNPs of cerebellum in Rad54 and Parp-1 double mutants
We next analyzed the cellular mechanisms underlying synthetic lethal interactions between Rad54 and Parp-1 by focusing on the cerebellum of Ptc1 +/mutants. The neonatal cerebellum undergoes rapid postnatal development, occurring through a burst of proliferation of granule cell precursors (GCPs) in the external granule layer (EGL) on the cerebellar surface. Notably, highly proliferating cells are particularly susceptible to endogenous and exogenous DNA damage, due to the presence of numerous replication forks that may stall and eventually undergo DSBs after collision with DNA damage [25,26]. This makes this cellular population suitable for investigating DNA repair-related mechanisms responsible for the growth delay observed. Quantification of Ki-67 staining in the EGL revealed a high proliferation rate in Parp-1 -/-/Ptc1 +/mice ( Figure 4A-4E), which is in keeping with the higher spontaneous MB tumorigenesis detected in Parp-1 -/mice. Similarly high growth rates were not observed in Rad54 -/-/Parp-1 -/-/Ptc1 +/mice, suggesting that persistent SSBs formation due to Parp-1 inhibition, promoting collapse of replication forks and DSBs which would normally be repaired by HR, cannot be repaired effectively in the absence of functional Rad54.
We therefore evaluated whether concomitant inactivation of Rad54 and Parp-1 enhances DNA damage and apoptosis in the neural precursor population of the EGL of double-compared to single-mutants or wt mice. To this aim, P1 brains were immunostained with a DSB (γ-H2AX), or an apoptotic marker (cleaved caspase-3), and the staining in the EGL was quantified. Consistent with its recognized role in resolution of SSB and stalled replication forks, inactivation of Parp-1, but not Rad54, resulted in a significant 3-fold increase of spontaneous DSBs compared to Ptc1 +/mice (P = 0.0013). A 4-fold increase (P = 0.0010) was also detected in Rad54 -/-/Parp-1 -/-/Ptc1 +/mutants ( Figure 4F-4J). However, despite similar amount of DSBs induced by inactivation of Parp-1 and by combined inactivation of Parp-1 and Rad54, we detected large differences in positivity for cleaved caspase-3 among the different mutants ( Figure 4K-4O). Mice with combined inactivation of Rad54 and Parp-1 3-fold, P = 0.003) mice. These findings confirm that Parp-1 inactivation causes an excess of spontaneous DSBs in neural cells, which can be partly repaired if Rad54 is active, but result in high apoptotic levels in GCPs with concurrent Rad54 inactivation.
In the neonatal cerebellum, the proliferation phase is followed by differentiation processes, consisting in cell cycle exit and inward migration of postmitotic GCPs to the internal granule layer (IGL), which causes progressive EGL disappearance and formation of the IGL of mature granule neurons. To evaluate the effects of combined inactivation of Rad54 and Parp-1 on differentiation processes, we measured the areas of the EGL and IGL in the cerebellum of Ptc1 +/mice with single or combined inactivation of Rad54 or Parp-1 at P7, and calculated their ratio. Despite drastic decrease in cerebellum size, no modification of the EGL/IGL ratio was detected in mice with combined inactivation of Rad54 and Parp-1 relative to other genotypes, indicating no overt alterations in the balance between proliferation (EGL) and differentiation (IGL) in the cerebellum caused by single or combined inactivation of the genes under investigation ( Figure 4P). However, quantification of the density of neurons labeled by the NeuN marker of mature neurons in fixed regions of the IGL (crest number VIII, red square in Figure 4R and 4S), revealed a significant decrease in the density of NeuN+ mature neurons in Parp-1 -/-/Ptc1 +/-(43.5%, P = 0.0126) and Rad54 -/-/Parp-1 -/-/Ptc1 +/mice (45.7%, P = 0.0042), suggestive of depletion of granule neurons ( Figure 4Q and 4S).

Synthetic lethality affects the expression of DNA repair related gene
The robust synthetic lethal effect of combined Rad54 and Parp-1 inactivating mutations led us to investigate the underlying molecular mechanisms by addressing the modification of the expression profile of 84 genes with established roles in DNA damage signaling. We carried out this analysis in the P1 cerebellum of Ptc1 +/mice with single or combined inactivation of Rad54 and/ or Parp-1 (Rad54 -/-/Ptc1 +/-, Parp-1 -/-/Ptc1 +/and Rad54 -/-/ Parp-1 -/-/Ptc1 +/-) and results were compared with those from Ptc1 +/mice using a pathway-based PCR expression array. In Rad54 -/-/Ptc1 +/mice 14 genes were significantly downregulated (16.7%) and 8 upregulated (9.5%); in Parp-1 -/-/Ptc1 +/mice, 22 genes were significantly downregulated (26%) and 6 (7%) upregulated; in Rad54 -/-/ Parp-1 -/-/Ptc1 +/mice, 10 genes (12%) were downregulated and 16 (19%) upregulated (S1Table). The magnitude of many of these changes was small. However, as DNA repair genes are expressed at basal level in mammals, detection of small changes does not necessarily mean a negligible biological significance, but rather that even slight deregulation may significantly modify DNA repair capacity. We also performed a hierarchical clustering analysis that produced different groups of genes (colored bars, left side) based on similar fold-change patterns among the three genotypes (Rad54  Table S1).

Parp-1 -/compound mutants
Based on results of bioinformatic analysis, we analyzed p53 expression in protein extracts from P1cerebellum by immunoblotting. Although no significant differences were detected between Rad54 and Parp-1  single and compound mutants in phosphorylation of p53 at Ser18 [27], we observed a significant increase in total p53 expression (3-fold) only in compound mutants, indicating marked p53 activation ( Figure 6A), which is consistent with persistent genetic damage. We also show that p53 stabilization did not occur at transcriptional level, as no changes in the expression of p53 were detected in the array ( Figure 5A and Table S1), suggesting, instead, posttranslational modification. Immunoblotting for p21 showed a trend towards increased expression that, however, was not significant ( Figure 6A).

DISCUSSION
The DNA repair system functions as a critical tumor suppressor network to preserve the integrity of the genome preventing malignancy, and DNA repair defects are frequently encountered in human cancer. Therefore, gaining insights into the mechanism of action of DNA repair genes is pivotal to development of novel tumorspecific treatment strategies targeting deregulated DNA repair [28]. A major obstacle to achieving synthetic lethal therapies is a lack of insight into the phenomenon in cancer cells, and insufficient knowledge of the molecular mechanisms influencing synthetic lethal interactions.
We earlier investigated in vivo genetic interactions between Parp-1 and Ptc1, and found that combined inactivation of the two genes is not critically important for normal development and survival of the compound mutant mice; however, increased genetic instability caused by abrogation of Parp-1 was able to promote MB formation in Ptc1 +/mice by fostering accumulation of genetic defects and facilitation of loss of the remaining normal Ptc1 allele [22]. Despite the critical role of the HR system for embryonic development, attenuated HR by deletion of Rad54 also did not result in lethality in Ptc1 +/mice, implying redundancy in function of proteins involved in HR; however, in previous work, by exposing double Rad54/Ptc1 mutants to 2 Gy, we detected significantly increased MB tumorigenesis, possibly through the generation of viable but genetically rearranged neural cell precursors due to repair of DNA lesions by error-prone mutagenic pathways [21]. An analogous increase was not seen here, likely due to the lower dose used and relative radioresistance of Rad54 -/mice [23].
Former in vitro studies have shown that Parp-1 inhibitors induce synthetic lethality in cells that are defective in HR [1,2,29]. An extensive body of literature has since documented the impact of Parp inhibitors on synthetic lethality in cells that are defective in HR due to loss of HR-associated proteins, and their use as singleagent therapies acting through the concept of synthetic lethality [30]. The results shown here demonstrate for the first time that Parp-1 and Rad54 are genetically linked and that they share overlapping functions that are essential in mouse development. We observed not only that the combination is active in vivo, but could also clarify that Parp-1 is not strictly necessary for HR during early embryogenesis, as combined deletions of Parp-1 and Rad54 caused significantly reduced birth rate and growth retardation -resulting in early death of Rad54 -/-/Parp-1 -/pups -not full embryonic lethality.
Since evaluation of the effects of simultaneous Rad54 and Parp-1 inactivation on MB development was prevented by the lack of viable compound mutants, we investigated the inactivation of DNA repair genes in the context of Ptc1 tumor suppression by examining mice with complete abrogation of either Rad54 or Parp-1 function, in combination with heterozygosis of Parp-1 or Rad54, respectively. Attenuation of Rad54 in the context of Parp-1 abrogation had synthetic lethal effects in dosage-independent manner. In contrast, inactivation of one copy of Parp-1 in the setting of Rad54 abrogation www.impactjournals.com/oncotarget had two distinct outcomes depending on treatment, i.e., no effect on spontaneous tumor development, or decreased induction of MB induced by radiation. A further increase in Parp-1 inactivation (i.e., both copies) caused a shift to synthetic lethality (i.e., loss of viable progeny), demonstrating that combined defects in Rad54 and Parp-1 cause synthetic lethal interactions in a Parp-1-gene dosage dependent manner, and highlighting the delicate balance that must be hit when targeting essential pathways.
Important to our understanding of the underlying mechanisms of this novel genetic link, immunohistochemical analysis revealed a trend to high degree of granule cell proliferation in the EGL of Parp-1 null mice (Parp-1 -/-/Ptc1 +/-), which decreased significantly in age-matched double mutants (Rad54 -/-/Parp-1 -/-/Ptc1 +/-). This was accompanied by substantial delay in neuronal differentiation, as shown by sparse NeuN staining and limited presence of committed GCPs in both groups. Given the concurrent high levels of spontaneous DSBs in both Parp-1 null mutants (Parp-1 -/-/Ptc1 +/and Rad54 -/-/ Parp-1 -/-/Ptc1 +/-), the low apoptotic levels observed only in Parp-1 -/-/Ptc1 +/correlate well with the high spontaneous rate of MB in these mice, and suggest that the cells can survive with a mutation in either repair pathways, not both, which causes cells to die.
Driven by the results of the bioinformatic analysis converging on p53 signaling, we investigated the expression of p53 protein in P1-cerebellum and MEFs, detecting significant accumulation of p53 protein in both. Because p53 phosphorylation is classically considered as the first crucial step of p53 stabilization, driven by several kinases, including ATM and DNA-PKcs, we analyzed phosphorylation of p53 at Ser18 [27] in P1 cerebella and did not detect significant differences among genotypes, despite accumulation of p53 only in double mutants. However, it is known that multiple phosphorylation sites exist, acting cooperatively for stimulating p53 activity [38]. Also, phosphorylation may not be a universal requirement for p53 stabilization, especially in vivo, as several studies have established, shifting attention on Mdm2 regulation of p53 [39]. All these factors, along with differences between in vitro and in vivo conditions, may concur to explain the differences in the expression pattern of p53/p53 Ser18 observed between MEFs and cerebellum.
Our results of p53 expression in P1-cerebellum and MEFs are in line with the fact that deficiency in several DNA repair genes such as DNA Lig4, XRCC4 and Rad51 cause increased level of genotoxic stress, triggering a p53-dependent DNA damage response leading to embryonic lethality [40][41][42]. More generally, mutations in over 150 cell-essential genes from different pathways result in recurrent phenotypes, which involve neural apoptosis and developmental delay in zebrafish and share p53 upregulation, suggesting that p53 might function as a developmental checkpoint triggered by various stressors, including genetic defects [43]. In agreement, mice with constitutively active p53 are generally not viable, due to inhibition of proliferation -via cell-cycle arrest, quiescence, senescence and differentiation -and/ or induction of apoptosis [44]. The nature and intensity of damage, as well as, the level of p53 expression and its interaction with specific proteins, may determine the choice among the different cellular responses that overall are cell-type specific. Besides p53-dependent neuronal apoptosis, our findings that MEFs from compound Rad54/Parp-1 mutant mice display severe p53-mediated accelerated senescent phenotypes, and that features of senescence were detected in several neonatal tissues, strongly argue for senescence as the mechanism underlying synthetic lethal interactions of Rad54 and Parp-1 inactivation in vivo. Notably, many of the DNA repair pathways found deregulated in the cerebellum of Rad54 -/-/Parp-1 -/-/Ptc1 +/mutants have been reported to cause senescence; overexpression of Pttg1, the most upregulated gene in our settings, induces senescence in a p53-dependent fashion in normal fibroblasts [45], and chemical inhibition of DNA-PKcs leads to p53-dependent accelerated senescence after irradiation in vivo and in vitro [46].
In summary, our data contribute to the understanding of the interplay of Rad54 and Parp-1 in vivo by showing that their combined genetic disruption caused a marked growth delay culminating in a perinatal lethal phenotype. We identified accumulation of DNA damage and enhanced apoptotic cell death as a driving mechanism for growth retardation during development of the cerebellum in Rad54/Parp-1 compound mutants. Mechanistically, NHEJ was selectively repressed through downregulation www.impactjournals.com/oncotarget of DNA-PKcs. Concomitant downregulation of Atm was also identified, directly involving transcriptional repression of PI3K signaling in synthetic lethality caused by Rad54 and Parp-1 deficiency. Finally, combined inactivation of Rad54 and Parp-1 caused accumulation of p53 and p21, which resulted in enhanced senescence in MEFs and several mouse tissues from Rad54 -/-/Parp-1 -/-/Ptc1 +/mutants. The p53-depedent apoptotic/senescent phenotypes is also potentially tumor protective; in fact, inactivation of one copy of Parp-1 in Rad54 -/-/Ptc1 +/mutants suppressed radiation-induced MB tumorigenesis, identifying haplosufficient interactions also during neoplastic growth. Investigating the functional interplay between the HR factor Rad54 and Parp-1 in vivo during development and oncogenesis may help delineate the molecular mechanisms of lethal synthetic interactions and guide more successful clinical translation.

Animal treatment and irradiation
Mice were housed under conventional conditions with food and water available ad libitum and a 12-h light cycle. For long-term study, mice were irradiated at postnatal day 1 (P1) with 1 Gy of X rays. Irradiation was performed using a Gilardoni CHF 320 G X-ray generator (Gilardoni S.p.A., Mandello del Lario, Italy) operated at 250 kVp, 5mA, Half-Value Layer = 1.6 mm Cu (additional filtration of 2.0 mm Al and 0.5 mm Cu) [50]. The size of the experimental groups was chosen to ensure statistical relevance. Randomized methods were applied and experiments were performed blinded. All experiments were conducted according to the directive 2010/63/EU of the European Parlament. Animal studies were approved and permission was issued by "Ministero della Salute" (Approval number is 986/2015-PR).

Histological analysis and tumor quantification
Mice were observed daily for their lifespan. Upon decline of health (i.e., severe weight loss, paralysis, ruffling of fur, or inactivity), they were killed and autopsied. Normally appearing and tumor bearing brains were fixed in 10% buffered formalin or snap frozen in liquid nitrogen. Autopsied mice were also checked for the presence of any other tumor, and all tumor masses were collected and processed for histological analysis using standard methods. MB incidence was expressed as the percentage of mice with tumors.

Microsatellite analysis at the Ptc1 locus in MB
DNA from MBs and normal tissue was extracted using Wizard SV Genomic DNA Purification System (Promega Corporation, Madison, WI). Polymerase chain reaction amplifications were performed as described previously. About 10-14 microsatellites spanning the length of chromosome (chr) 13 were used to examine tumor DNA in comparison with the corresponding genomic DNA [51].

Morphometric analysis
Cerebella external granule layer (EGL) and internal granule layer (IGL) total cross sectional areas were measured in at least 3 samples. The imaging software NIS-Elements BR 4.00.05 (Nikon, Instruments Europe B.V., Italy) was used for morphometric analyses.

Preparation of mouse embryonic fibroblast (MEFs)
MEFs (two independent embryos per genotype) were generated from embryos at gestational day 13.5 (E13.5) according to standard protocols [52]. Briefly, the whole embryo was disaggregated using a mechanical disaggregator system (Medimax, CTSV, Bruino, TO, Italy) and grown in DMEM with 15% fetal bovine serum (FBS), 100 units/ml penicillin, 100 mg/ml streptomycin, and 2mM L-glutammine, in a humidified incubator with 5% CO2. All experiments were performed with cells that had been subjected to 2 passages.

Sulforhodamine B (SRB) assay
For the growth curve SRB assay was performed as described previously [53]. The plate was analysed at 510 nm optical density in the VICTOR X3 Multilabel Plate Reader (PerkinElmer, Waltham, MA). Doubling time (Dt) was calculated with the formula: Dt = t log2 / (log n1 -log n0), where n0 is the number of cell at time t0, n1 is the number of cells at time t1 and t is the time in exponential growth phase required by cells to increase their number from n0 to n1 (t = t1 -t0).

SA-β-Gal staining
Cells were cultured for 24 h, and stained using the Senescence β-Galactosidase Staining Kit (Cell Signalling Technology). SA-β-Gal staining of MEFs was performed following manufacturer's instructions. The Percentage of cells expressing SA-β-Gal was quantified by inspecting at least 1000 cells in at least two different cell lines for each genotype.

mRNA extraction and reverse transcription
Three to 7 cerebella were pooled for each genotype. mRNA was extracted with RNeasy Microarray Tissue kit (Qiagen, Germany) in accordance with manufacturer's instructions. Reverse transcription was accomplished with the RT2 First Strand kit (Qiagen, Germany) as suggested by manufacturer.

RT2 Profiler PCR Array
RT 2 RNA QC PCR Array (SaBiosciences, Qiagen, Germany) was used to test for RNA quality and inhibitors of RT-PCR analysis. For quantitative comparison of mRNA levels, real-time PCR was performed using RT 2 Profiler PCR Arrays Mouse DNA Damage (SaBiosciences, Qiagen). For each genotype, two assays were carried out. Gene expression was related to the mean expression of all five housekeeping genes included in the array. Only Ct values <35 were included in the calculations. Data presented are averages of two independent experiments.
Prkdc and Atm genes were further validated by SYBR Green real-time PCR (SaBiosciences, Qiagen). The PCR primers for mouse Atm (PPM03454C-200 RT² qPCR Primer Assay ) and Prkdc (PPM03711E-200 RT² qPCR Primer Assay) were purchased from SABiosciences, Qiagen and real-time PCR were performed according to manufacturer's instructions. Data represent the average of three independent experiments.