Somatic loss of WWOX is associated with TP53 perturbation in basal-like breast cancer

Inactivation of WW domain-containing oxidoreductase (WWOX), the gene product of the common fragile site FRA16D, is a common event in breast cancer and is associated with worse prognosis of triple-negative breast cancer (TNBC) and basal-like breast cancer (BLBC). Despite recent progress, the role of WWOX in driving breast carcinogenesis remains unknown. Here we report that ablation of Wwox in mammary tumor-susceptible mice results in increased tumorigenesis, and that the resultant tumors resemble human BLBC. Interestingly, copy number loss of Trp53 and downregulation of its transcript levels were observed in the Wwox knockout tumors. Moreover, tumors isolated from Wwox and Trp53 mutant mice were indistinguishable histologically and transcriptionally. Finally, we find that deletion of TP53 and WWOX co-occurred and is associated with poor survival of breast cancer patients. Altogether, our data uncover an essential role for WWOX as a bona fide breast cancer tumor suppressor through the maintenance of p53 stability.


Introduction
Breast cancer is the most common malignancy in women and second to lung carcinoma in cancer mortality 1 . One of the greatest advances in the last few years has been the molecular categorization of breast cancer based on gene expression profiles. Transcriptomic analyses of human breast tumors have led to classification of several molecular subtypes with distinctive gene profiles and clinical relevance [2][3][4] . These molecular subtypes are strongly associated with survival outcome, with the basallike subtype (BLBC) having the worst prognosis [2][3][4] . Identification of new molecular targets and modeling of BLBC would hence greatly enhance our understanding of this aggressive subtype and aid in better management.
The WW domain-containing oxidoreductase (WWOX) gene spans one of the most active common fragile sites in the human genome located at the long arm of chromosome 16: FRA16D 5,6 . WWOX is commonly altered in breast cancer [7][8][9] . In particular, it has been shown that WWOX protein levels are reduced or absent in triplenegative breast cancer (TNBC) and in BLBC [10][11][12][13][14][15] . Beside genomic rearrangements, hypermethylation of the regulatory region of WWOX has been documented in neoplastic but not in paired adjacent non-neoplastic tissues 16,17 . Importantly, restoration of WWOX expression inhibits breast cancer cell growth both in vitro and in vivo, further proposing a tumor suppressive function 17 . These observations led us to question whether WWOX possesses a driver role in tumor suppression in genetically engineered mouse models.
Existing evidence using animal models has indeed linked WWOX with tumor suppressive functions [18][19][20][21] . Modeling WWOX loss in drosophila revealed that WWOX expression is required for efficient removal of tumorigenic cells via TNFα/Egr-mediated cell death, which was shown to be dependent on caspase-3 activity 22 . Furthermore, a number of Wwox mutant mouse models have also suggested tumor suppressive roles for WWOX.
In particular, aged germline Wwox-heterozygous mice on mixed background developed higher incidence of spontaneous lung tumors and B-cell lymphomas 23,24 , and those on C3H genetic background developed mammary tumors with 50% penetrance 25 . Some of these tumors had retained the other wild type allele, suggesting haploinsufficiency of WWOX function 23,25 . Importantly, mammary tumors in Wwox-heterozygous C3H mice were mostly ER-negative and PR-negative, expressing CK-14, hence reminiscent of the common WWOX inactivation in TNBC and in particular BLBC 25 . Despite these lines of evidence, however, no proof was established linking somatic loss of WWOX in mammary epithelium with mammary tumor advantage.
Several studies have shown that the WWOX protein antagonizes tumorigenesis by promoting apoptosis 26,27 , rewiring metabolism 28,29 and more recently, maintaining genome integrity 30,31 . WWOX, mainly through its first WW domain 32,33 , interacts with partner proteins such as p53 family members 26,27,34 , DNA-damage checkpoint proteins 30,31,35 , and key metabolic and stress proteins 28,29 to mediate its tumor suppressor activities. While multiple mechanisms have been suggested to explain the tumor suppressive role of WWOX, there is currently no functional evidence for the exact role of WWOX in breast carcinogenesis.
We and others have previously reported that mammary gland epithelium (MGE)-specific Wwox deletion (Wwox ΔMGE ) in B6/129 mixed genetic background does not result in a mammary tumor phenotype 36,37 . We therefore set out to determine whether somatic alteration in Wwox drives BLBC development in mammary tumor-susceptible mice: C3H/HeJ mice (shortly named C3H). We found that inactivation of Wwox in mammary gland epithelium in C3H mice results in mammary tumor formation resembling BLBC in human, as revealed by histological and molecular characterization of these tumors. This is the first mouse model that enables to directly study the role of WWOX in BLBC. Our data show that tumors of Wwox knockout or Trp53 knockout are indistinguishable. At the molecular level, we show that WWOX loss results in reduced p53 activity either through destabilizing the genome, resulting in p53 loss and genomic instability, or through hindering p53 transcriptional function. We further demonstrate that deletion of WWOX and TP53 co-occurs in breast cancer. Altogether, our results reveal WWOX, the gene product of FRA16D, as a bona fide breast cancer tumor suppressor with important functions in maintaining genome stability.

Mice
Wwox ΔMGE mice (on B6/129 genetic background) 36 were back-crossed onto the C3H/HeJ mice (shortly named C3H), a mammary tumor-susceptible genetic background, for seven rounds (~99% purity) generating Wwox ΔMMTV . Despite the lack of exogenous mouse mammary tumor virus (MMTV), virgin and breeding C3H female mice may still develop some mammary tumors later in life 38 . Trp53 ΔMMTV mice were generated by crossing Trp53 loxP mice 39 with MMTV-Cre mouse (The Jackson Laboratory, #003553). Genotyping of Wwox, Trp53 and Cre was performed using primers as detailed in Supplementary Table 3. All mice related experiments were approved by The Hebrew University Institutional Animal Care and Use Committee.

Immunohistochemistry
Tissues were fixed in 4% formalin. Paraffin-embedded tissue sections were deparaffinized and rehydrated. Antigen retrieval was performed in 25 mM sodium citrate buffer PH 6.0 (for ER, PR, CK14, and gamma H2Ax) or EDTA buffer PH 8.0 (for WWOX) using pressurized chamber for 2.5 min. Endogenous peroxidase was blocked with 3% H 2 O 2 for 15 min. The sections were then incubated with blocking solution (CAS Block) for 30 min to reduce non-specific binding followed by incubation with the primary antibody. Slides were subsequently incubated with horseradish peroxidase-conjugated anti-rabbit or anti-mouse immunoglobulin antibody for 30 min. The enzymatic reaction was detected in a freshly prepared 3,3 diamminobenzidine using DAB peroxidase kit (Vector laboratories) for several min at room temperature. The sections were then counterstained with hematoxylin. Eight tumors were stained for WWOX, ER, and CK14. Six tumors were stained for PR and γH2Ax.

Isolation of primary mouse epithelial cells (MECs)
Mammary glands were isolated and minced from the indicated mice. For each gram tissue, 5 ml digestion mix [DMEM media, 10% fetal bovine serum, 1% pen-strep, 1:100 collagenase A (from stock 1.5 mg/ml) and 1:1,000 DNase-I (from 10 mg/ml stock)] was added and left for 1.5 h at 37°C under moderate shaking (50 × g). MECs were separated from fat by differential centrifugation (three times) at 700×g for 30″. MECs were washed twice with PBS. In order to get rid of red blood cells, red blood cells lysis buffer was used.

RNA extraction and RT-PCR
Total RNA was isolated using Tri-reagent. For RT-PCR, RNA (1 µg) was reversed transcribed using the QScript cDNA syntesis kit (Quantabio). Real-time PCR was done using SYBR Green PCR Master. Real-time PCR was performed using primers as indicated in Supplementary  Table 4.

Gene expression analysis
RNA from 4 Wwox ΔMMTV mammary tumors, 2 Trp53 ΔMMTV mammary tumors, 4 double-knockout (Wwox;Trp53 ΔMMTV ) tumors 3 normal (prior to tumor formation) Wwox ΔMMTV mammary epithelial cells (MECs) and 3 normal wild-type MECs was extracted and prepared for RNA sequencing. After poly-A cleanup, cDNA was synthesized. Libraries were made using KAPA Single-Indexed Adapter Kit (Illumina, Massachusetts, USA). Sequencing was performed using Next seq 500 (Illumina). Normalized gene expression values were log2-transformed and scaled by subtracting the gene expression means. Dendrograms were constructed using Euclidean distances and Complete linkage, and heat maps were generated with the "pheatmap" R package. Analysis was performed using all expressed genes: prior to the analysis, non-expressed genes (normalized expression value < 1 in all samples) were excluded, and expression levels were floored to 1. Data from double-knockout mice were processed with all other expression data, but their analysis is not included in the current study. For detailed bioinformatics analysis, see supplementary methods.

E-karyotyping
E-karyotyping analysis was performed as previously described 40,41 . Briefly, normalized gene expression values were log2-transformed, non-expressed genes (log2 expression value < 1 in all samples) were excluded, and expression levels were floored to 1. The median expression value of each gene across normal samples was subtracted from the expression value of that gene in each normal and tumor sample, in order to obtain comparative values. Genes were ordered by their chromosomal location, and CNA profiles were then generated with the CGH-Explorer software (http://heim.ifi.uio.no/bioinf/ Projects/CGHExplorer/), using the program's piecewise constant fit (PCF) algorithm. The following set of parameters was used: Least allowed deviation = 0.25; Least allowed aberration size = 30; Winsorize at quantile = 0.001; Penalty = 12; Threshold = 0.01. Moving average plots were generated using the CGH-Explorer moving average fit tool.
Comparison to other breast cancer mouse models Gene expression data of five different GEMM types (Wnt, Myc, PyMT, Her2, and P53) from two studies 42,43 (GSE23938 & GSE25488) were downloaded from the Gene Expression Omnibus (GEO) website (http://www. ncbi.nlm.nih.gov/geo). Gene expression values were log2-transformed, non-expressed genes (log2 expression value < 5.5 in > 20% of samples) were excluded, and expression levels were floored to 5. The median expression value of each gene across normal samples was subtracted from the expression value of that gene in each tumor sample, in order to obtain comparative values. Gene symbols were compared across the three datasets (the two previous studies and the current study), yielding a list of 5,897 genes present and expressed in all tumor samples. Batch effect was next removed using the COMBAT algorithm 44 . Unsupervised hierarchical clustering was performed on the batch-corrected expression values, using Euclidean distances and complete linkage. The Epithelial-Mesenchymal transition (EMT) gene set was downloaded from MSigDB 45 (http://software.broadinstitute. org/gsea/msigdb). An EMT score was determined for each tumor sample as the sum of the comparative expression values of the expressed EMT genes. The average EMT score of each type of tumor was computed.

Molecular subtype assignment
PAM50 centroid values were obtained from Parker et al. 46 , FPKM expression values of the PAM50 genes were log2-transformed, and the Spearman's rank correlation between each sample and each subtype centroid was calculated. The class with the highest correlated centroid was assigned to each sample. In one p53 tumor, correlation values were similar for two subtypes (basal and lumB), and this tumor was therefore defined as mixed.

Genomic DNA extraction and quantitative PCR
Tumors or normal MECs were minced using DNA lysis buffer (150 mM NaCl, 50 mM tris-HCL, 25 mM EDTA, 0.5% SDS and 5 µl proteinase K). After overnight incubation at 56°C, phenol chloroform was added followed by centrifugation for 10 min. The upper phase was transferred to a new tube and followed by chloroform and isopropyl alcohol treatment. DNA was washed with 70% ethanol and eluted using TE buffer. Genomic DNA realtime PCR was performed using primers as indicated in Supplementary Table 5.
Cell culture assays ATCC breast cancer cell line MCF7 cells were cultured as in RPMI-1640 media supplied with 10% FCS, 1% pen-strep and 1% glutamine. The cells were mycoplasma free.
Colony formation assay MCF7 cells were plated at a density of 500 cells in a 60mm 2 plate in triplicate. After 2 weeks the cells were fixed with 70% ethanol, stained with Giemsa and counted.

WWOX overexpression in knockout cells
To prevent the CAS9 cut of the WWOX after the restoration, WWOX expressing lentivirus was mutated using QuikChange XL Site-Directed Mutagenesis Kit, Agilent Technologies, CA, USA. We mutated WWOX, at the guide RNA-identical sequence (CAS9-target sequence). The result sequence has different codons but codes for the same amino acids of the original sequence.

Co-occurrence gene inactivation analysis
Gene-level mutation and copy number data of the METABRIC breast cancer dataset 47 were downloaded from cBioPortal (www.cbiopotal.org). The number of tumors with a perturbation (homozygous deletion or mutation) in either Trp53, Wwox or both genes was determined. The statistical significance of co-occurrence was determined using a two-sided Fisher's exact test.

Survival analysis
Survival data of the METABRIC breast cancer dataset 47 were downloaded from cBioPortal (www.cbiopotal.org). Tumors were separated into groups by their WWOX and p53 status: both genes WT, p53-perturbed (mutation, deletion or both), WWOX-perturbed (mutation, deletion or both), and both genes perturbed. Survival analysis was performed using the "survival" R package.

Statistical analysis
Results were expressed as mean ± SD or ± SEM. The Student's t-test was used to compare values of test and control samples. P < 0.05 indicated significant difference.

Results
Mammary-specific deletion of Wwox in C3H mice is associated with mammary tumorigenesis Germline aged Wwox-heterozygous mice on C3H mammary tumor-susceptible genetic background develop mammary tumors with 50% penetrance 25 . To test whether somatic loss of WWOX in C3H mice could facilitate mammary tumorigenesis, Wwox ΔMGE mice were backcrossed onto the C3H background for seven generations (N7/F1;~99% purity) and incidence of mammary tumor formation was evaluated; these newly generated mice were named Wwox ΔMMTV .
Monitoring of Wwox ΔMMTV mice revealed that the majority (14/17,~76%) developed mammary tumors with median latency of 270 days, while no mammary tumors were obtained in the control WT mice (Fig. 1a). Histological and pathological characterization of these Wwox ΔMMTV tumors revealed that they are invasive ductal carcinoma, Grade III, with occasional lung metastases (2/ 14) (Fig. 1b, Supplementary Fig. S1a). Deletion of the Wwox fl/fl alleles in these tumors was confirmed by immunohistochemistry using anti-WWOX antibody ( Fig. 1b and Supplementary Fig. S1b). Immunohistochemical staining for ER and PR revealed that 100% of the tumors are ER/PR-negative, while~60% stained positive for CK14 basal marker ( Fig. 1b and Supplementary  Fig. S1b). These findings are consistent with previously published data showing that WWOX loss in human is associated with TNBC and BLBC 9,12,14,25 . Altogether, these observations suggest that WWOX ablation on C3H mammary tumor-susceptible genetic background is associated with BLBC development. According to our knowledge, this is the first mouse model showing that somatic ablation of WWOX in adult mice is sufficient to drive mammary tumorigenesis.

Molecular characterization of Wwox ΔMMTV tumors
To molecularly characterize WWOX-deficient tumors, RNA from 4 Wwox ΔMMTV mammary tumors, 3 normal (prior to tumor formation) Wwox ΔMMTV mammary epithelial cells (MECs) and 3 normal wild-type MECs was extracted and analyzed using RNA sequencing. Differential gene expression analysis revealed that 5588 genes were differentially expressed in the mammary tumors compared to normal controls (Fig. 1c (Fig. 1d), whereas WT MECs (WT_N) and MECs isolated from Wwox ΔMMTV pre-transformed tissues (Wwox_KO_N) clustered together (Fig. 1d). The differentially expressed genes were enriched for cancer-related pathways, including extracellular matrix, receptor interactions, focal adhesion and adherent junctions (Supplementary Table 2).
Consistent with our immunohistochemical analysis (Fig. 1b), RNA sequencing showed significant decrease in estrogen receptor (Esr1) and progesterone receptor (Pgr) transcripts, as well as upregulation of basal markers mRNA including Ck14, cytokeratin 6 (Ck6), cytokeratin 17 (Ck17), caveolin 1 (Cav1), caveolin 2 (Cav2), αB- Crystallin (Cryab) and P-cadherin (Cdh3) 9,25 (Fig. 1e). Moreover, a significant reduction in RNA levels of Foxa1, known to repress the basal-like phenotype [48][49][50] , was observed (Fig. 1e), further confirming the BLBC nature of these tumors. A major molecular subtype of TNBC is the claudin-low/ mesenchymal-like subtype 51,52 , expressing low levels of tight junction proteins, including certain claudins and Ecadherin, and high levels of genes associated with epithelial-to-mesenchymal transition (EMT) 51 . Our RNA sequencing analysis showed an increase in expression levels of the majority of EMT markers but no decrease in expression levels of claudins in Wwox ΔMMTV tumors compared to MECs (Supplementary Fig. S1c). Moreover, in a comparison of EMT scores between Wwox ΔMMTV tumors and five other known mammary tumor mouse models, the Wwox ΔMMTV model ranked among the highest, close to those of previously published Wnt and Trp53 knockout models, known models to generate TNBC-like tumors ( Supplementary Fig. S1d). Altogether, both immunohistochemistry and RNA sequencing data suggest that mammary tumors formed in Wwox ΔMMTV mice resemble basal-like TNBC.
Wwox ΔMMTV tumors display hampered p53 expression and function Similar to alterations of WWOX in BLBC, mutations in TP53 were found in 88% of BLBC 52 . Moreover, conditional knockout of Trp53 using MMTV and WAP-Cre in C57BL/ 6 mice showed high percentage of tumor incidence, though at a later stage of life 53 . Using MMTV-Cre mice, high percentage (47%-100%) of Trp53 f/f mice developed mammary tumors with latency of 10-14.5 months 54 . These tumors were negative for both ER and PR and resulted in metastases in both liver and lung 54 . We therefore examined the expression of p53 and its target genes in Wwox ΔMMTV tumors. Interestingly, our RNA sequencing analysis revealed a significant downregulation of Trp53 levels (Pvalue = 0.0176) (Fig. 1e). To confirm this observation, an RT-PCR analysis was performed on additional mammary tumors obtained from Wwox ΔMMTV mice, showing that Trp53 levels were indeed downregulated in all tumors as compared to normal MECs or to archived tumors from Wwox wild-type or heterozygous mice 25 ( Supplementary  Fig. S1e). Additionally, Global Gene Set Enrichment Analysis (GSEA) showed that the p53 pathway is suppressed in the Wwox ΔMMTV tumors when compared to the normal samples (the genes that are upregulated when p53 is knocked-down are upregulated in the Wwox ΔMMTV tumors, and the genes that are downregulated when p53 is knocked-down are downregulated in the tumors) (Supplementary Fig. S1f). Intriguingly, in one of three normal MECs isolated from Wwox ΔMMTV mice, there was a significant decrease in Trp53 gene expression ( Supplementary  Fig. S1g), suggesting that p53 downregulation might take place already at an early stage prior to mammary tumor development. To confirm impairment of p53 function, we measured levels of its target gene, Cdkn1a, and found a significant downregulation in the majority of Wwox ΔMMTV tumors ( Supplementary Fig. S1h) (P-value < 0.0001). These results imply that somatic loss of WWOX in mammary epithelium results in a reduced p53 activity to drive BLBC development.
Wwox ΔMMTV and Trp53 ΔMMTV tumors share similar patterns of gene expression and genomic instability The previous observations prompted us to determine whether inactivation of Wwox resembles the effect of Trp53 inactivation in mammary gland epithelium. We therefore generated Trp53 mammary conditional knockout mice on B6/129 mixed background (named Trp53 ΔMMTV ) and examined mammary tumor formation. While Wwox ΔMGE mice did not develop tumors, we found that Trp53 ΔMMTV mice developed mammary tumors with latency of 262.5 days (P-value = 0.007) and 45% penetrance. Histological and pathological characterization of Trp53 ΔMMTV mammary tumors revealed an invasive ductal carcinoma -Grade III that displays ER/PR negative expression and CK14 positive expression ( Supplementary  Fig. S2a). These results suggest that deregulation of WWOX and p53 share similar functions in driving BLBC mammary tumor development. Indeed, when comparing Trp53 ΔMMTV tumors to Wwox ΔMMTV tumors, the two tumor groups shared similar upregulation of the basal markers ( Supplementary Fig. S2a and Fig. 2a) and enhanced expression of EMT markers (Fig. S2b). Interestingly, Trp53 ΔMMTV tumors showed no or very low levels of WWOX protein as assessed by immunohistochemistry ( Supplementary Fig. S2a). In contrast, Wwox RNA levels in these tumors were markedly upregulated as revealed by (see figure on previous page) Fig. 1 WWOX loss is associated with triple-negative breast cancer (TNBC) and particularly, the basal-like subtype. a Kaplan Meier analysis for Wwox ΔMMTV mice as compared to wild-type mice. b Histological characterization of mammary tumors in Wwox ΔMMTV and normal wild-type (WT) mice, using H&E staining and immunohistochemistry (anti-WWOX, anti-ER, anti-PR, and anti-CK14). Magnification bar represents 50 µm. c Volcano analysis for WT normal mammary tissue vs Wwox ΔMMTV tumor tissues. c Gene expression moving average plots along Trp53 locus in chromosome 11 of normal mammary tissues from WT mice (n = 3), normal mammary tissues from Wwox ΔMMTV mice (n = 3), mammary tumors from Wwox ΔMMTV mice (n = 4) and mammary tumors from Trp53 ΔMMTV mice (n = 2). These expression patterns suggest that three out of four Wwox ΔMMTV mice have lost a copy of chromosome 11, which includes the Trp53 gene. d Quantitative PCR performed on genomic DNA of WT tissue, one Trp53 ΔMMTV tumor and two Wwox ΔMMTV tumors, for normalization primers specific for Trim2 (didn't not affect by Wwox loss) were used. Error bars represent standard deviation RNA sequencing and qRT-PCR analyses ( Fig. 2a and data not shown), possibly suggesting a compensatory effect. These data suggest that mutual inactivation of both p53 and WWOX takes place during mammary tumor formation.
To further characterize these tumors, we used their gene expression profiles to classify them according to the PAM50 breast cancer intrinsic subtypes 46,55 . Four out of four WWOX-KO tumors and one out of two p53-KO tumors best matched the basal subtype (Supplementary Table 7). The other p53-KO tumor displayed a mix of basal-like and LumB expression signatures, consistent with previously published analyses of p53 mouse models 46,55 . These findings further confirm that WWOX loss is associated with BLBC formation.
We further examined whether the global gene expression profile of our Wwox ΔMMTV model is similar to those of previously published mammary-specific p53 knockout models. Indeed, the Wwox ΔMMTV model clustered together with the Trp53 KO models from a previous study (GSE23938) 42 (Supplementary Fig. S2c).
To better characterize our mammary tumor models, we performed expression based-karyotyping analysis (e-karyotyping) 40,41 to explore the chromosomal landscapes of the tumors (Fig. 2b). Expression-based analysis of copynumber changes in both Wwox ΔMMTV Trp53 ΔMMTV models showed that all tumors exhibit high prevalence of copy-number alterations (Fig. 2b). Consistent with chromosomal aberrations and genomic instability, we observed high levels of γH2AX staining, a surrogate marker of DNA double strand breaks and repair signaling, in tumors of both groups (Supplementary Fig. S2d). Altogether, Wwox ΔMMTV and Trp53 ΔMMTV mice developed mammary tumors with very similar histologies, gene expression profiles, and genome instability patterns.

Wwox ΔMMTV tumors are p53 deficient
The similar molecular features of the two models and the reduction of Trp53 mRNA levels in Wwox ΔMMTV tumors suggest that changes in p53 levels may occur due to DNA gene deletion or due to transcriptional attenuation. While all Wwox ΔMMTV tumors have significantly decreased Trp53 gene expression, our expression-based copy-number analysis suggested that three of the four Wwox ΔMMTV tumors analyzed presented a large-scale copy-number loss in a region that includes the Trp53 gene, which may explain its reduced expression and the overall similarity between the two models (Fig. 2c). We therefore directly addressed this question, by performing quantitative real-time PCR on genomic DNA extracted from the Wwox ΔMMTV and the Trp53 ΔMMTV tumors. DNA isolated from wild-type and normal Wwox ΔMMTV MECs was used as a control. As expected, Trp53 ΔMMTV samples had a deletion in exons 2-10 of the Trp53 gene, which were originally targeted by the loxP sites. In line with the e-karyotype analysis, we also noticed a significant decrease in Trp53 genomic DNA levels in Wwox ΔMMTV tumors, indicating that the reduction of Trp53 expression was indeed due to its genomic loss (Fig. 2d) P-value < 0.005). We also examined the DNA status of genes located upstream and downstream of Trp53 region, and found reduced genomic levels of Wrap53 (P-value < 0.001) (upstream), as well as of Atp1b2 (P-value < 0.001 for T2 and T3), Shbg (P-value < 0.001), Sat2 (P-value < 0.001) and Fxr2 (P-value < 0.001) genes (downstream) in the Wwox ΔMMTV tumors, further suggesting that loss of Wwox resulted in genetic loss of the p53 locus (Supplementary Fig. S2e-g). Consistent with this later observation, p53 expression level in Wwox ΔMMTV model was as low as in the Trp53 ΔMMTV model (Fig. 2a).

WWOX knockout in MCF7 cells using CRISPR
To further determine the impact of WWOX deletion, we used CRISPR technology to knockout WWOX in a WWOX-positive breast cancer cell line. Since TNBC cell lines express very low levels of WWOX 56 , we knocked-out WWOX in MCF7, an ER+ cell line that expresses high levels of WWOX and harbors wild-type p53 56,57 . Complete WWOX knockout was confirmed in several clones of WWOX-KO MCF7 cells by immunoblot analysis (Fig. 3a). Consistent with previously published data on WWOX knockdown and reduced hormone receptor levels 25 , WWOX-KO MCF7 cells displayed reduced ER and PR transcript levels ( Fig. 3b and c) (P-value < 0.05 and < 0.01 respectively). These cells also exhibited a significantly increased survival capability, as assessed by a colony formation assay (Fig. 3d). WWOX has been shown to play a direct role in the DNA damage response (DDR) 31 . To determine the consequence of WWOX deletion on DDR signaling, we tested checkpoint protein activation and p53 status in MCF7-KO cells upon ionizing radiation (IR). WWOX knockout clones showed reduced or no phosphorylation of both ATM and its target protein KAP1 upon DNA damage, suggesting an impaired DDR in these clones (Fig. 3e). Moreover, WWOX-deficient MCF7 cells exhibited reduced induction of nuclear p53 and decreased expression of p53 target genes, p21 (Fig. 3f-h) and PUMA (Supplementary Fig. S3a). However this reduction was not due to genomic loss. Importantly, WWOX over-expression, using CRISPR-untargetable WWOX mutant, rescued these phenotypes ( Supplementary  Fig. S3b-d), confirming that the observed changes were due to on-target perturbation of WWOX. Altogether, WWOX loss is associated with impaired p53 function, enhanced survival and impaired DDR in MCF7 cells.
Co-occurrence of WWOX and TP53 deletion in breast cancer WWOX and TP53 are both commonly perturbed in breast cancer. To determine the human relevance of combined alteration of WWOX and p53, we analyzed their mutual perturbation in the METABRIC dataset (www.cbioportal.org; 47,58 2509 patients samples). Surprisingly, we found that genetic alterations of WWOX were a rare event in this dataset of human patients, despite multiple evidence for WWOX protein expression reduction in clinical samples of breast cancer, and in TNBC in particular 12,13 . Despite their rareness, however, we found a very significant co-occurrence of WWOX homozygous deletion and TP53 homozygous deletion (p = 0.0001 in a two-tailed Fisher's exact test; Fig. 4a). Furthermore, while TP53 perturbation was associated with poorer survival as expected (Fig. 4b), genetic perturbation of WWOX was not associated with poorer survival of TP53-perturbed patients (Fig. 4b). These data therefore support the idea that p53 and WWOX cooperate in human breast cancer tumorigenesis, and that loss of WWOX promotes tumorigenesis through the perturbation of TP53.

Discussion
BLCB is a heterogeneous, highly aggressive and difficult to treat tumor type [59][60][61][62] . We found that WWOX loss is associated with worse prognosis in BLBC, and that its targeted deletion in murine mammary epithelium leads to mammary tumors resembling BLBC. In a previous study, we noted that WWOX haploinsufficiency in C3H background promotes BLBC 25 . These observations are confirmed in the current study using a new mouse model harboring floxed-Wwox alleles. While none of control WT mice on the C3H background developed mammary tumors, somatic loss of Wwox alleles in this genetic background promoted BLBC-like tumors (Fig. 1). Although developed in C3H background, these tumors clustered together with tumors from Trp53 ΔMMTV mice (in B6/129 background), suggesting that these tumors are very similar and that a functional crosstalk between WWOX and p53 is critical to antagonize BLBC. Our findings suggest that WWOX deletion by itself might not be enough for BLBC development, and that a second hit facilitated in the C3H background, like perturbations in p53 signaling or locus, is required. Indeed, WWOX inactivation was associated with Trp53 gene instability and/or impaired DNA-damage checkpoint protein activation. Deletion of Wwox or Trp53 resulted in aggressive BLBC-like tumors. RNA profiling of mammary tumors of the different mouse models revealed similar patterns, though not identical, of differential gene expression, highlighting upregulation of basal-cell markers and EMT genes. Furthermore, co-occurrence of mutations in the TP53 and WWOX tumor suppressor genes is commonly seen in patients. Overall, WWOX functions as a mammary tumor suppressor, likely through mediating genome stability in a p53-dependent mechanism, and its loss of function in mammary epithelium reproduces several important features of BLBCs.
TNBC and BLBC tumors commonly harbor deleted or mutated p53, and mouse modeling of Trp53 dysregulation is associated with BLBC tumor formation 53,55,60 . The WWOX gene has been originally cloned due to its common inactivation in breast cancer 5,6 . Several subsequent studies have shown that WWOX expression is reduced or absent in breast cancer, particularly TNBC and BLBC 9,13,14 , and that its overexpression in breast cancer cells suppresses tumor growth 17 and enhances apoptosis mediated by activation of the p53 family proteins 26,27,34 . These results suggested that WWOX could act as a tumor suppressor, though its localization within FRA16D questioned this function.
Common fragile sites (CFSs) are regions that appear in vitro as gaps or breaks in metaphase chromosomes of cells exposed to partial inhibition of DNA replication 63 . The significance of CFSs was recently highlighted, as many regions associated with these sites display homozygous and hemizygous deletions in human cancer 64,65 hence accusing them to be the "weakest link" of the genome and to be responsible for genomic instability 63,66,67 . Furthermore, it was suggested that some of these regions might act as early warning sensors for DNA damage since chromosomal aberrations within CFSs were detected in experimentally-induced skin hyperplasia and in early stages of lung cancer 68,69 . By contrast, emerging evidence has shown that gene products of CFSs, at least some, may play direct roles in the DNA damage response (DDR), hence questioning their exact role in the carcinogenesis process [70][71][72] . For example, the fragile histidine triad (FHIT) gene was recurrently found lost in a plethora of human tumors, including TNBC 13 , and its deregulated expression was associated with unstable genome (reviewed in the ref. 71 ). In addition, recent studies revealed that WWOX levels are induced upon DNA damage and that WWOX interacts with DNA damage checkpoint proteins ATM 31 and BRCA1 35 to regulate DNA repair in breast cancer cells. These results argue against a passenger role of WWOX in breast cancer. These recent observations defining WWOX as a direct player in the DDR 30,31,35 led us to pursue our hypothesis and test whether specific targeted deletion of Wwox in murine mammary epithelium results in mammary tumor formation. Our results provide clear evidence that WWOX loss results in spontaneous deregulation of the Trp53 gene locus, leading to impaired DNA damage response and mammary tumors resembling human BLBC. This is the first in vivo evidence linking somatic WWOX deregulation in a mammary cell with spontaneous mammary tumorigenesis.
We show that WWOX inactivation results in destabilization of the genome and p53 loss, although the direct mechanisms of this remains to be determined. It is possible that the p53 genomic region is not directly targeted, but that WWOX loss leads to mild genomic instability, and the cells that lose the Trp53 locus are strongly selected for. Should this be true, we would expect cells harboring both WWOX and p53 alterations to have a greater advantage in cell growth and transformation and hence greater and accelerated tumor formation. Our findings might suggest that both WWOX and p53 function in the same pathway, and hence their deregulation in cancer is expected to be mutually exclusive. Surprisingly, we found that co-occurrence of altered WWOX and TP53 is common in breast cancer, suggesting that WWOX may have additional functions other than the DDR. In line with this cooperation, combined WWOX and p53 loss has been also shown to be associated with aggressive osteosarcoma formation 73 .
Our findings in vitro also support an important role of WWOX on p53 function. Modeling WWOX knockout in MCF7 cells didn't affect the genomic locus of TP53 but rather had a significant transcriptional repression and reduced activity of p53 (Fig. 3f-h). Therefore, WWOX may also act as a positive regulator of p53 transcription, potentially through its ability to bind transcription factors via its WW domains 74 .
Our findings prompt us to propose a model (Fig. 4c) by which WWOX cooperates with p53 under stress conditions, leading to enhanced apoptosis as previously reported 26,75 , and this cooperation is part of the tumorigenesis barrier. During early stages of breast cancer development, WWOX is reduced or lost, as documented in hyperplasia and DCIS lesions 11 , either genetically or as a result of epigenetic silencing. WWOX deregulation leads to impaired DDR, and results in destabilization of the genome, thus leading to compromised function of key tumor suppressor genes, such as p53. This would result in further genomic instability, enabling cells to overcome the tumor barrier and embark on cancer evolution. WWOX could also modulate p53 and other important signaling molecules by affecting their transcription or transactivation function and hence impact the carcinogenesis process. Our model further supports a scenario where CFSs and their gene products might have far reaching roles in driving human malignancies.

GEO data availability
The RNA sequencing data in this publication have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE117387.