p53 downregulates the Fanconi anaemia DNA repair pathway

Germline mutations affecting telomere maintenance or DNA repair may, respectively, cause dyskeratosis congenita or Fanconi anaemia, two clinically related bone marrow failure syndromes. Mice expressing p53Δ31, a mutant p53 lacking the C terminus, model dyskeratosis congenita. Accordingly, the increased p53 activity in p53Δ31/Δ31 fibroblasts correlated with a decreased expression of 4 genes implicated in telomere syndromes. Here we show that these cells exhibit decreased mRNA levels for additional genes contributing to telomere metabolism, but also, surprisingly, for 12 genes mutated in Fanconi anaemia. Furthermore, p53Δ31/Δ31 fibroblasts exhibit a reduced capacity to repair DNA interstrand crosslinks, a typical feature of Fanconi anaemia cells. Importantly, the p53-dependent downregulation of Fanc genes is largely conserved in human cells. Defective DNA repair is known to activate p53, but our results indicate that, conversely, an increased p53 activity may attenuate the Fanconi anaemia DNA repair pathway, defining a positive regulatory feedback loop.

I nherited bone marrow failure syndromes are a set of clinically related yet heterogeneous disorders in which at least one haematopoietic cell lineage is significantly reduced. Among them, Fanconi anaemia (FA) and dyskeratosis congenita (DC) are caused by germline mutations in key cellular processes, that is, DNA repair and telomere maintenance, respectively 1 .
We recently found that p53 D31/D31 mice, expressing a mutant p53 lacking its C-terminal domain, die rapidly after birth with a complete set of features of the telomere syndrome DC, including aplastic anaemia, pulmonary fibrosis, oral leukoplakia, skin hyperpigmentation, nail dystrophy and short telomeres 2 . Loss of the p53 C terminus increases p53 activity in mouse embryonic fibroblasts (MEFs) and in most tested tissues 2,3 , and p53 D31/D31 MEFs exhibited decreased messenger RNA (mRNA) levels for 4 out of 10 genes implicated in telomere syndromes (Dkc1, Rtel1, Tinf2 and Terf1). Nutlin, a drug that prevents the Mdm2 ubiquitin ligase from interacting with p53, allowed to confirm that p53 activation leads to the downregulation of these four genes. These data revealed that p53 plays a major role in telomere metabolism.
We previously focused on the potential p53-mediated regulation of genes mutated in DC (Dkc1, Rtel1 and Tinf2) or implicated in aplastic anaemia, a milder form of telomere syndrome (Terf1) 4 . As a striking evidence for the clinical relevance of our mouse model, patients with severe DC who carry mutations affecting PARN, a negative regulator of p53, were recently shown to exhibit decreased DKC1, RTEL1 and TERF1 mRNA levels 5 . Importantly, however, tens of proteins are thought to be involved in the regulation of telomeres (reviewed in ref. 6). Thus, it remained possible that the impact of p53 on telomere-related genes was underestimated in our previous study. Here we tested whether p53 affects the expression of 42 additional genes implicated in telomere metabolism, and found 7 genes that are downregulated in p53 D31/D31 cells. Importantly, some of these p53-regulated genes are involved in the FA DNA repair pathway. This was particularly intriguing because Rtel1, one of the four telomere-related genes we previously found regulated by p53, encodes a Fancj-like helicase 7 . These observations led us to evaluate whether p53 regulates more genes belonging to the FA pathway, and whether p53 D31/D31 cells exhibit characteristic features of FA cells. We found that murine p53 downregulates 12 Fanc genes, that human p53 downregulates 9 FANC genes and that the capacity to repair DNA interstrand crosslinks is attenuated upon p53 activation. These data reveal an unexpected role for p53 in downregulating the FA DNA repair pathway, which may help to understand the pathological processes implicated in FA, and suggest therapeutic strategies against tumour cells that retain a functional p53 pathway.
For a gene to be a good candidate, we considered that the mean (from three to four independent experiments) of its mRNA levels in unstressed WT cells should fall between the means measured in p53 À / À and p53 D31/D31 cells; and that the means for the three genotypes should be statistically different according to an analysis of variance. Out of the 42 genes, 7 fulfilled these criteria: Blm, Dek, Fancd2, Fen1, Gar1, Recql4 and Timeless ( Fig. 1a; Supplementary Fig. 1). Because RECQL4 was shown to be downregulated by p53 in human cells 27 , the lower Recql4 mRNAs in p53 D31/D31 cells were not surprising. The decreased mRNA levels for the six other genes were not anticipated however. To specifically assay for a p53-dependent regulation, we next compared the effects of Nutlin, a drug that activates p53 by preventing its interaction with the ubiquitin ligase Mdm2. Results clearly indicated that p53 activation leads to the downregulation of these genes (Fig. 1b).
Importantly, the finding that p53 downregulates Gar1, which encodes a component of the telomerase complex, strengthened our previous conclusion that p53 plays a significant role in telomere biology. However, Fancd2 appeared as the gene whose expression was most markedly affected by p53 activation (Fig. 1b). This was surprising because, even if primary cells from patients with a FANCD2 mutation may exhibit telomere dysfunction 28 , these patients are diagnosed with FA, a syndrome primarily characterized by defects in DNA repair. This led us to further analyse the p53-dependent regulation of Fancd2. We first verified that the relative decrease in Fancd2 mRNA levels were observed in vivo, in bone marrow cells (BMCs) from p53 D31/D31 mice (Fig. 1c). We next tested whether the p53-dependent regulation of Fancd2 detected by quantitative PCR had an impact on Fancd2 protein levels. Lower Fancd2 protein levels were observed in unstressed p53 D31/D31 cells compared with unstressed p53 À / À or WT cells, and Nutlin treatment led to a decrease in Fancd2 proteins only in WT and p53 D31/D31 MEFs, in complete agreement with quantitative PCR data ( Fig. 1d; Supplementary Fig. 2).
p53 activation leads to increased E2F4 binding at Fancd2. The p53-mediated downregulation of many genes requires the cdk inhibitor p21, and occurs through the recruitment, upon p53 activation, of E2F4 repressive complexes at their promoters 29,30 . Notably, this mechanism would account for the p53-dependent regulation of cell cycle genes whose promoters contain CDE/CHR regulatory motifs [31][32][33] . Consistent with this mechanism, p53 activation had no effect on Fancd2 mRNA levels in p21 À / À cells (Fig. 2a), and chromatin immunoprecipitation (ChIP) experiments with an antibody against E2F4 indicated increased E2F4 binding at the Fancd2 promoter in Nutlin-treated WT cells, compared with unstressed WT or Nutlin-treated p53 À / À cells ( Fig. 2b; Supplementary Fig. 3). Of note, ChIP assays for E2F4 binding at the Fancd2 promoter could not be performed in p53 D31/D31 MEFs because their accelerated senescence 2 prevented the recovery of sufficient amounts of chromatin, but it is likely that the p53/p21/E2F4 pathway operates similarly in p53 D31/D31 cells. We next identified a candidate CDE/CHR motif in the Fancd2 promoter, and mutation of the CDE element (typically bound by E2F4) abolished the Nutlin-dependent repression of this promoter in NIH-3T3 cells (Fig. 2c), independently of cell cycle dynamics (Supplementary Fig. 4). Thus, although the expression of Fancd2 is known to vary during the cell cycle 34 , the differences in Fancd2 mRNA levels observed between WT and p53 D31/D31 MEFs would not simply result from differences in G1/S ratios 2 . Rather, our results indicate that p53 activation promotes the recruitment of E2F4 at the Fancd2 gene, and that E2F4 plays a major role in the repression of Fancd2.
In the experiments above, p53 activation resulted from a treatment with Nutlin, a molecule that acts as a specific Mdm2 inhibitor. We next tested whether similar results could be obtained in response to DNA damage, by evaluating the effects of doxorubicin, a clastogenic anticancer agent. Doxorubicin treatment led to decreased Fancd2 mRNA and protein levels in WT and p53 D31/D31 cells, but not p53 À / À MEFs ( Supplementary  Fig. 5a,b). Furthermore, we observed increased E2F4 binding at the Fancd2 promoter in doxorubicin-treated WT cells, compared with unstressed WT or doxorubicin-treated p53 À / À cells ( Supplementary Fig. 5c). Thus, both Nutlin and doxorubicin lead to p53 activation and consecutive Fancd2 downregulation.
Interestingly, the Blm and Fen1 genes, also downregulated by p53 (Fig. 1b), respectively, encode an helicase that associates with Fanc proteins in a multienzyme complex 35 , and an endonuclease stimulated by a Fanc protein 36 . Furthermore, Rtel1, one of the four telomere-related genes we previously found regulated by p53 (ref. 2), encodes a Fancj-like helicase 7 . This led us to further evaluate the impact of p53 activation on the FA DNA repair pathway.
p53 downregulates many Fanc genes. Because the expression levels of four FA genes had been tested in our previous experiments-Fancc, Fancd2, Fancp/Slx4 and Fancq/Ercc4 ( Fig. 1; Supplementary Fig. 1), we next compared, in unstressed p53 À / À , WT and p53 D31/D31 cells, mRNA levels for the 15 remaining FA genes. Strikingly, 11 were less expressed in p53 D31/D31 cells (Fig. 3a). Again, Nutlin was used to confirm the p53-mediated downregulation of these genes (Fig. 3b) Fig. 6), and p53 activation correlated with an increased binding of E2F4 near the transcription start site of each of these Fanc genes (Fig. 3c). We next used the sequence of six functional CDE/CHRs to define a positional frequency matrix, which was then used to search in silico for candidate CDE/CHRs near the E2F4-binding sites identified in ChIP assays. Using this approach, candidate CDE/CHR motifs were identified for 9 out of the 11 tested Fanc genes, with the best candidate motifs for Fanci and Fancr ( Fig. 4a; Supplementary Fig. 7). These data led us to further analyse the p53-mediated regulation of Fanci and Fancr. We first verified that the relative decreases in Fanci and Fancr mRNA levels were observed in vivo, in BMCs from p53 D31/D31 mice ( Supplementary Fig. 8). We then found that p53 activation leads to decreased Fanci and Fancr protein levels ex vivo ( Fig. 4b; Supplementary Fig. 9). Luciferase assays next showed that mutating the CDE site in each candidate CDE/CHR abolished the Nutlin-dependent repression of the Fanci and Fancr promoters (Fig. 4a,c).
We also observed that a 24-h long treatment with doxorubicin led to decreased Fanci and Fancr mRNA, and protein levels in WT and p53 D31/D31 cells, but not p53 À / À MEFs ( Supplementary  Fig. 10a). Furthermore, the nine other Fanc genes downregulated by p53 on Nutlin treatment were also downregulated in a p53-dependent manner on treatment with doxorubicin ( Supplementary Fig. 10b). We then searched for confirmation of our results by analysing the data recently reported by Younger et al., who performed a genomic analysis that integrated transcriptome-wide expression levels, genome-wide p53-binding profiles and chromatin state maps to characterize the regulatory role of p53 in response to DNA damage 37 . Although this approach was designed to identify direct p53 targets, we reasoned that genes regulated by p53 indirectly, via p21/E2F4, might also be detected in their transcriptome-wide expression data. These experiments were performed on p53 À / À and WT MEFs, treated or not with doxorubicin for 6 h (ref. 37), and our previous time-course experiments with Nutlin suggested that 6 h might be sufficient to observe a partial p53-mediated trancriptional downregulation 2 . Thus, we extracted the data of Younger et al.
(Gene Expression Omnibus # GSE55727) to analyse the expression of the 12 Fanc genes that we had found downregulated by p53. In agreement with our results, this analysis showed that doxorubicin led to an overall decrease in the expression of Fanc genes in WT, but not p53 À / À MEFs ( Supplementary Fig. 11).
Transcriptome data mining was also used to find whether the downregulation of Fanc genes could correlate with p53 activation in haematopoietic cells. The Homeobox (Hox) transcription factors are important regulators of normal and malignant haematopoiesis, because they control proliferation, differentiation and self-renewal of haematopoietic cells. We analysed the data of Muntean et al. (Gene Expression Omnibus # GSE21299), who immortalized murine BMCs by transduction with Hoxa9-ER cells in the presence of tamoxifen (4-OHT), and observed that they undergo myeloid differentiation 5 days after 4-OHT withdrawal 38 . We found this differentiation to correlate with an induction of genes known to be transactivated by p53 (Cdkn1A/p21, Mdm2 and Fas), and with the downregulation of Fanc genes ( Supplementary Fig. 12).
In sum, we found that 12 genes of the FA DNA repair pathway are downregulated by p53 via a p21/E2F4 pathway, and identified CDE/CHR motifs that are crucial for this regulation for three of these genes. Importantly, the genes are downregulated by p53 in response to Mdm2 inhibition or DNA damage, or on haematopoietic cell differentiation, and encode proteins involved in all parts of the FA DNA repair pathway, that is, proteins that , and ChIP data on the right. ChIP assay for E2F4 binding was performed in Nutlin-treated p53 À / À MEFs, and untreated or Nutlin-treated WT MEFs, with an antibody against E2F4 or rabbit IgG as a negative control. Immunoprecipitates were quantified using real-time PCR, fold enrichment was normalized to data over an irrelevant region, then E2F4 binding at Fancd2 in untreated WT cells was given a value of 1.    (a) A comparison of p53 À / À , wild-type and p53 D31/D31 cells suggests a potential p53-dependent regulation for 11 additional genes of the Fanconi anaemia (FA) DNA repair pathway. mRNAs for the indicated Fanc genes were quantified as described in Fig. 1a, in four independent experiments. For 11 of the tested genes, mean mRNA levels were intermediate in WT cells compared with p53 À / À and p53 D31/D31 cells, with statistical significance by one-way analysis of variance (ANOVA). (b) The 11 genes are downregulated on murine p53 activation. mRNAs for the indicated Fanc genes were quantified in untreated or Nutlin-treated MEFs. Results from three independent experiments. (c) Increased E2F4 binding at several Fanc promoters upon p53 activation. ChIP assay for E2F4 binding was performed in untreated or Nutlin-treated WT MEFs, with an antibody against E2F4 or rabbit IgG as a negative control. Immunoprecipitates were quantified using real-time PCR, fold enrichment was normalized to data over an irrelevant region, and then E2F4 binding in untreated WT cells was given a value of 1. Data are from two to three independent ChIP experiments, each quantified in triplicates. Below the ChIP data are represented, as in Fig. 2b, sequences around the TSS for each gene, putative E2F4-binding sites (lollipops), and primers used for ChIP assays (arrows). In all figures, means þ s.e.m. are shown; ***Pr0.001, **Pr0.01, *Pr0.05, NS, not significant by ANOVA or Student's t-tests.  [39][40][41][42] . Together, these data suggested an important role for p53 in regulating the FA pathway.
p53 activation attenuates the repair of specific DNA lesions. A typical feature of FA cells is their inability to repair DNA interstrand crosslinks, as evidenced by an increased frequency of chromosomal aberrations, and more specifically tri-and quadri-radial chromosomes, after exposure to mitomycin C (MC) 39 . We compared the effects, on WT and p53 D31/D31 cells, of a 48-h treatment with 50 nM MC. Such a treatment procedure was previously reported to differentially affect WT MEFs and MEFs with an impaired FA pathway 43 . Interestingly, we found that this procedure led to a rather subtle induction of p53 (suggested by a limited increase in p21 transactivation), which correlated with a twofold decrease in Fancd2 mRNA expression in p53 D31/D31 MEFs, but no significant alteration of Fancd2 mRNA levels in WT cells (Supplementary Fig. 13). We next determined the frequencies of all types of chromosomal aberrations, or of radial chromosomes, in WT and p53 D31/D31 cells before or after treatment with MC. In untreated cells, no significant difference was found between the two genotypes. Strikingly, however, chromosomal aberrations, and particularly radial chromosomes, were more frequent in p53 D31/D31 cells after treatment with MC, consistent with a decreased capacity to repair interstrand crosslinks in the mutant cells (Fig. 5a). Accordingly, chromosomes with sister chromatid exchanges were also more frequent in MC-treated p53 D31/D31 cells than in WT cells (Fig. 5b). These results suggested that the FA DNA repair pathway is attenuated in p53 D31/D31 cells, presumably because these cells exhibit an increased p53 activity. Consistent with this, p53 D31/D31 cells exhibited a decreased capacity to form Rad51 foci and an increased sensitivity to MC, and the pretreatment of cells with Nutlin appeared to further impact on these cellular phenotypes (Fig. 5c,d). Further evidence that the decreased DNA repair in p53 D31/D31 cells resulted from increased p53 activity (rather than a loss of the p53 CTD per se) came from analysing Mdm2 þ / À Mdm4 þ /DE6 MEFs. These MEFs express a WT p53 protein, but exhibit an increased p53 activity due to lower levels of p53 inhibitors 44,45 . Like p53 D31/D31 MEFs, Mdm2 þ / À Mdm4 þ /DE6 cells were more sensitive than WT cells to MC (Supplementary Fig. 14). In sum, a defective FA DNA repair pathway is known to activate p53 (ref. 46), but these results indicate that an increased p53 activity might reduce the expression of several FA genes and attenuate the FA DNA repair pathway. Taken together, these data indicate the existence of a positive regulatory feedback loop (Fig. 6).
Human p53 also regulates FA genes. We next tested whether the FA genes that were found regulated by murine p53 were similarly regulated in human cells. We compared human primary WT cells with p53-deficient cells, and observed that out of the 12 p53-regulated FA genes identified in mouse cells, 9 are also downregulated upon p53 activation in human MRC5 cells: FANCA, FANCB, FANCD1, FANCD2, FANCI, FANCJ, FANCM, FANCR and FANCT (Fig. 7a). Interestingly, one of these genes, FANCB, was recently identified as one of 210 genes most likely to be downregulated by p53 in a E2F4-dependent manner 33 . Furthermore, candidate CDE/CHR motifs could be found for each of these genes ( Supplementary Fig. 15a), and the CDE/CHRs in Fancd2, Fanci and Fancr were highly conserved in the human FANC homologous genes ( Supplementary Fig. 15b). Consistent with this, we next found that human p53 activation leads to increased E2F4 binding at the FANCD2, FANCI and FANCR promoters (Supplementary Fig. 16a), and that mutation of the CDE/CHRs in these promoters abolished their p53-dependent regulation (Supplementary Fig. 16b). The p53-dependent downregulation of FANC genes could also be observed in response to DNA damage in MRC5 cells ( Supplementary Fig. 17), and we verified that the CDE/CHR motif in FANCD2 is important for its DNA damage-induced downregulation ( Supplementary Fig. 18). In addition, the data mining of a transcriptome-wide analysis were again consistent with our results (Supplementary Fig. 19). BLM, DEK, FEN1, TIMELESS and RECQL4 were also downregulated in human cells upon p53 activation, further indicating an overall conservation of the regulatory pathways identified in murine cells ( Supplementary Fig. 20). Further evidence of this conservation was obtained using the Oncomine software (www.oncomine.org). Tumour samples from the Australian Ovarian Cancer Study revealed that the p53 pathway is functional in low-grade ovarian serous tumours, but frequently lost in high-grade ovarian carcinomas. Evidence for this first came from using a transcriptomic signature of p53 target genes 47 . Formal demonstration was later obtained by TP53 sequencing, which identified p53 mutations in 0% of low-grade serous tumours 48 and 96.7% of high-grade carcinomas 49 . We analysed the transcriptome data of Anglesio et al. 47 , who characterized 90 ovarian samples from the Australian Ovarian Cancer Study, including 60 high-grade adenocarcinomas. As expected, the expression of genes activated by p53 (CDKN1A/ p21, MDM2, DDB2 and SESN1) was decreased in high-grade tumours. On the opposite FANCD2, and other genes known to be repressed by E2F4 in a p53-dependent manner (BIRC5, CDC6 and CDC25C), were more expressed in high-grade tumours (Fig. 7b). Increased FANCD2 expression also correlated with increases in the expression of other FA genes (FANCA, FANCI, FANCJ, FANCR and FANCT), as well as additional genes regulated by p53 in our experiments (BLM, FEN1 and TIMELESS; Fig. 7b). Similar results were obtained when we analysed data from liver cancers (Supplementary Fig. 21) and adrenocortical tumours ( Supplementary  Fig. 22), providing evidence that human p53 downregulates several genes of the FA pathway in many tissues, and that loss of p53 function leads to an increased expression of FANC genes in advanced human cancers.
We next found that Nutlin sensitized human primary WT cells, but not their p53-deficient counterparts, to MC (Fig. 7c). Likewise, the sensitivity to MC of human cancer cells expressing a WT p53 was markedly increased by Nutlin (Fig. 7d), suggesting a potential therapeutic relevance of our findings.

Discussion
In this report, we further analysed the consequences of a deletion of the p53 carboxy-terminal domain. Our previous analysis indicated that most p53 D31/D31 mice exhibit a full set of features characteristic of DC. At the molecular level, the increased p53 activity in p53 D31/D31 MEFs correlated with the downregulation of four genes implicated in telomere syndromes: Dkc1, Rtel1, Terf1 and Tinf2 (ref. 2). Here we show that several other genes involved in telomere metabolism are downregulated in p53 D31/D31 cells: Blm, Dek, Fancd2, Fen1, Gar1, Recql4 and Timeless, strengthening the notion that p53 plays a major role in the regulation of telomere metabolism. Importantly, some of these genes are involved in DNA repair, and we next found p53 D31/D31 cells to exhibit decreased mRNA levels for 11 additional genes mutated in FA, and a reduced capacity to repair DNA interstrand crosslinks. Because DC and FA are both inherited bone marrow failure syndromes in humans, these new findings raised the possibility that an attenuated FA pathway might contribute to the bone marrow failure that affects p53 D31/D31 mice. Importantly, however, mice carrying knocked out alleles of Fanc genes exhibit little or no haematological abnormalities in the absence of additional stress 50 (for example, aldehyde-mediated DNA damage 51,52 ), whereas aplastic anaemia occurs spontaneously in mouse models of telomere dysfunction (for example, Pot1b À / À mTR þ / À mice 53 ) and in p53 D31/D31 mice 2 . Furthermore, p53 D31/D31 mouse cohorts of mixed genetic backgrounds previously indicated that a gene linked to the Agouti locus, on chromosome 2, had an impact on their survival 2 . None of the Fanc genes maps on chromosome 2, whereas mRNA levels for Rtel1, located 26 cM away from Agouti, affected the survival of mutant mice 2 . Rtel1 encodes a Fancj-like helicase that might participate in DNA repair 54 , but that mainly acts as a dominant regulator of telomere length 55 . Accordingly, Rtel1 is mutated in telomere syndromes, including severe DC [56][57][58] and pulmonary fibrosis 59 . Together, these data indicate that telomere dysfunction most likely plays a predominant role in the aplastic anaemia that affects p53 D31/D31 mice.
Interestingly, aplastic anaemia is not the only clinical trait shared by patients with FA and DC: abnormal skin pigmentation,   Figure 6 | A simplified model of p53 regulation by a bipolar feedback system may account for the attenuated DNA repair capacity of p53 D31/D31 MEFs. Defects in DNA repair activate p53 (1), and activated p53 downregulates several FA genes (2), which would attenuate the FA pathway and cause partial defects in DNA repair (3), hence defining a positive-feedback loop (in green). In unstressed WT MEFs, this positivefeedback loop would be efficiently counterbalanced by the negativefeedback loop (in red) between p53 and its major inhibitor, the ubiquitin ligase Mdm2 (4 and 5). In p53 D31/D31 MEFs however, the p53 D31 protein is more abundant, indicating that its interaction with Mdm2 is decreased 2 . Hence, the p53/Mdm2 negative-feedback loop is enfeebled in p53 D31/D31 MEFs (or in Mdm2 þ / À Mdm4 þ /DE6 MEFs), which would lead to a stronger p53/FA positive-feedback loop and thus to a reduced capacity to repair mitomycin C-induced DNA lesions. In both WT and p53 D31/D31 cells, Nutlin specifically affects the p53/Mdm2 negative-feedback loop, which would further increase the cellular sensitivity to mitomycin C.   Fig. 23). Because the p53 pathway is affected by single-nucleotide polymorphisms in many genes including TP53, MDM2, MDM4 and CDKN1A 68 , we further presume that the strength of the regulatory loops that affect p53, telomere-related and FA genes should vary among humans, and that this might contribute, in patients with identical disease-causing mutations, to the variability in clinical overlap between these syndromes. Independently, our data also provide a rationale for the combination of Nutlin with therapeutic agents inducing DNA interstrand crosslinks, to efficiently kill cancer cells that retain a functional p53 pathway.
Cell cycle assays. Log phase cells, treated or not with Nutlin, were incubated for 24 h, then pulse-labeled for 1 h with bromo-deoxy uridine (BrdU) (10 mM), fixed in 70% ethanol, double stained with fluorescein isothiocyanate anti-BrdU and propidium iodide, and sorted using a LSRII cytometer. Data were analysed using FlowJo.
Luciferase expression assays. To construct the Luciferase reporter plasmids, we cloned a 2-kb fragment (for Fancd2) or 1-kb fragment (for Fanci, Fancr, FANCD2, FANCI or FANCR) centred around the transcription start site upstream of the firefly luciferase gene in a pGL3-basic vector (Promega), or a variant fragment generated by PCR mutagenesis of the putative CDE/CHR motif (details on request). Next, 10 6 NIH-3T3 cells were transfected using lipofectamine 2000 by 3 mg of a Fanc-luciferase reporter plasmid and 30 ng of renilla luciferase expression plasmid (pGL4.73, Promega) for normalization, and treated or not with 10 mM Nutlin 3a or 0.5 mg ml À 1 doxorubicin. Transfected cells were incubated for 24 h, then trypsinized, resuspended in 75 ml culture medium with 7.5% FBS and transferred into a well of an optical 96-well plate (Nunc). The dual-glo luciferase assay system (Promega) was used according to the manufacturer's protocol to lyse the cells and read firefly and renilla luciferase signals. Results were normalized, then the average luciferase activity in cells transfected with a WT Promoter and not treated with Nutlin were assigned a value of 1.
Metaphase spread preparation and analyses. Cells were plated in duplicate, then untreated or treated with 50 nM MC for 48 h, and treated with 0.1 mM nocodazole for 3 h to arrest cells in metaphase. Cells were submitted to hypotonic shock (75 mM KCl), fixed in a (3:1) ethanol/acetic acid solution, dropped onto glass slides and air-dried slides were stained with Giemsa to score for chromosome aberrations. To analyse sister chromatid exchanges, cells plated in duplicate and treated or not with MC were, 1 h after plating, treated with 10 mM (BrdU 1/3 BrdC) for 48 h, then metaphase spreads were prepared as above. Air-dried slides were stained with 10 mg ml À 1 Hoescht 33258 for 20 min, submitted to ultraviolet at 365 nm while heated at 55°C during 30 min, then stained with Giemsa. Images were acquired using a Zeiss Axiophot (X63) microscope.
Immunofluorescence. Cells were spread onto coverslips, treated or not with Nutlin 10 mM, then MC 0.1 mg ml À 1 for 1 h, and left to recover for 12 h. Twenty-four hours after Nutlin treatment, cells were fixed and permeabilized. Coverslips were incubated with a Rad51 antibody (Ab-1 Calbiochem) for 1 h at 37°C in a humid chamber, then with secondary Alexa Fluo anti-rabbit antibody (Invitrogen). Slides were mounted in Vectashield with 0.2 mg ml À 1 4,6-diamidino-2-phenylindole. Images were captured on a Zeiss Axioplan2 microscope using equal exposure times for all images.
Cellular sensitivity to mitomycin C. Cells were seeded into wells of a 96-well plate (500 cells per well, in triplicates). After adhesion, cells were treated or not with Nutlin 2.5 mM for 24 h, then with MC for 48 h at 0, 0.01, 0.1 and 1 mg ml À 1 . Cells were then counted using the CyQUANT kit (Life technologies) and a microplate reader according to the supplier's recommendations.