Tumor cell lines were generated from individual tumors arising in female mice with a KB1P or KP genotype as described previously 23. Established cell lines were cultured at 37°C with 5% carbondioxide in DMEM-F12 medium (Gibco, Carlsbad, CA, USA) supplemented with 10% FCS, 50 U/ml penicillin, 50 μg/ml streptomycin (Gibco, Carlsbad, CA, USA), 5 μg/ml insulin (Sigma, St. Louis, MO, USA), 5 ng/ml epidermal growth factor (Invitrogen, Carlsbad, CA, USA) and 5 ng/ml cholera toxin (Gentaur, Brussels, Belgium).

One million KB1P3.12 cells were electroporated (3 μF, 0.8 kV) with the bacterial artificial chromosome (BAC) clone RP11-812O5 containing the complete human BRCA1 gene and regulatory sequences. The RP11-812O5 BAC was obtained from the Children's Hospital and Research Center at Oakland, CA, USA 24. The vector backbone was modified by insertion of the pgk/EM7 Neo/KanR-positive selection cassette pCEI1 25 into the sacBII gene by bacterial homologous recombination in Escherichia coli SW102 26. After selection with 300 μg/ml Geneticin (Gibco-BRL, Carlsbad, CA, USA) for two weeks, clones were picked and checked for the presence of the BAC by PCR for exon 11 of human BRCA1 27.

Stem cell medium (SCM) containing defined growth factors as described by 2829 was freshly prepared each time and DZNep (5 μm) or dimethyl sulfoxide (DMSO; no-drug control) was added. Cells were trypsinized, which was inactivated with 10% serum and subsequently washed with PBS to remove the serum, and resuspended in SCM. Cells were filtered to obtain single cells and 40,000 cells, counted with a Casy counter (Schaerfe Systems, Reutlingen, Germany), were plated out in ultra-low binding plates with a flat bottom (Corning Incorporated, Corning, NY, USA). Sphere formation was checked every day and cell culture images were obtained after 72 hours using a Zeiss Axiovert 25 microscope (Carl Zeiss MicroImaging, Goettingen, Germany) with 10× objective on a Sony Cybershot (Sony corporation, Tokyo, Japan).

Human breast tumor tissue samples were obtained from the pathology archive of the Netherlands Cancer Institute. BRCA1-mutation status was determined by routine DNA diagnostics. The basal-like and luminal status was determined using expression data to classify the tumors according to the intrinsic gene set as described 30.

Paraffin-embedded tumor samples were sectioned (4 μm) and deparaffinized by treating twice with xylene for 10 minutes each and subsequently hydrated in 100%, 80%, and 70% ethanol. Antigen retrieval was performed by boiling samples in 10 mM sodium citrate for one minute at 900 W and 15 minutes at 250 W in the microwave, followed by 20 minutes gradual cooling at room temperature. Slides were blocked in 5% normal goat serum in PBS. Mouse tissue was additionally permeabilized with 0.25% Triton prior to blocking. The samples were incubated overnight with a mouse monoclonal antibody against EZH2 (1:100, BD Biosciences San Jose, California, USA). Immunodetection was performed by diaminobenzidine (DAB) oxidation using the Powervision system (ImmunoLogic, Duiven, The Netherlands). Samples were dehydrated in 70%, 80%, and 100% ethanol. Once immunostained, slides were digitally scanned with a Scanscope (Aperio Technologies, Vista, CA, USA) and the amount of EZH2-expressing epithelial cell nuclei were counted in a blinded manner by two observers (SAJ and JP) independently. For statistical analysis, the Wilcoxon test was used to compare the percentage of EZH2-positive nuclei between two samples and P < 0.05 was considered statistically significant.

Cells were scraped from subconfluent plates and lysed in RIPA buffer (150 mM NaCl, 10 mM Tris-HCl pH 7.5, 1% Triton X100, 1% DOC, 0.1% SDS, 1 mM EDTA) Equal amount of protein (10 μg) was loaded and separated by electrophoresis on NuPAGE Novex 4 to 12% SDS-PAGE (Invitrogen, Carlsbad, CA, USA) and transferred to nitrocellulose membranes. The blot was blocked with TBST (0.1% Tween 20) containing 5% BSA (Sigma-Aldrich, St. Louis, MO, USA) and incubated with primary antibodies for two hours at room temperature. After washing with TBST, the membrane was incubated with a horseradish peroxidase (HRP)-conjugated secondary antibody and the signal was detected with enhanced chemiluminescence substrate (GE Healthcare, Amersham, UK). Following antibodies were used: mouse monoclonal against EZH2 (1:20, provided by K. Helin), β-tubulin (1:10000, Sigma-Aldrich, St. Louis, MO, USA), anti-mouse IgG-HRP (1:10000, Biosource International, Camrillo, CA, USA).

Total RNA was extracted using Trizol Reagent (Invitrogen, Carlsbad, CA, USA) and 1 μg per sample was treated with DNase. Reverse transcription was performed using the SuperScript™ First-Strand Synthesis System for RT-PCR (Invitrogen, Carlsbad, CA, USA; according to manufacturer's protocol). The generated cDNA was analyzed using SYBR Green (Taqman universal PCR master mix, Applied Biosystems, Foster City, CA, USA), performed on an ABI Prism 7000 SDS (Applied Biosystems, Foster City, CA, USA). Product accumulation was evaluated using the comparative C_T method (DDC_T), with Hprt levels as internal control for normalization. The following primers were used (all for mouse transcripts): Hprt FW: CTGGTGAAAAGGACCTCTCG; Hprt RV: TGAAGTACTCATTATAGTCAAGGGCA, Ezh2 FW: AAAGACCCTGAATGCAGTCGC; Ezh2 RV: TGATCCAGAACTTCATCCCCC.

Expression values (Log2 ratio) of Ezh2 in 21 KB1P and 32 KP mammary tumors were obtained from oligonucleotide microarrays representing 18,173 genes. Methods for RNA extraction, RNA amplification, microarray hybridization, and data processing are described by Liu and colleagues 12. For comparison of EZH2 gene expression (Log10 ratio) signatures between mouse and human breast tumors, we used the expression profiles of 96 human breast tumors: 18 ER-negative BRCA1 tumors, 34 tumors with a good prognosis signature and 44 tumors with a poor prognosis signature categorized by the human 70-genes signature dataset 31.

DZNep was provided by YU Qiang (Genome Institute of Singapore) and trichostatin A (TSA) was obtained from Sigma-Aldrich (St. Louis, MO, USA). Both compounds were solved in DMSO and stored at -20°C. Before the growth inhibition assays, growth curves were made of all cell lines to determine the amount of cells that ensure exponential growth during a five-day culture period. For cell viability analysis, subconfluent dishes were trypsinized and filtered to obtain single cells. Cells were counted with a Casy counter and appropriate amounts of cells were typically plated out in 96-well plates on day 0. Drugs were added in twofold serial dilutions on day 1 in triplicate. DMSO was used as a no-drug control. Both drugs were left on the cells for 120 hours. On day 5 cell viability was measured using the Cell Titer Blue assay (Promega, Madison, WI, USA). Cell Titer Blue reagent was added directly to the cells (10 μl/well) and incubated for two hours at 37°C. Fluorescence at 560_ex/590_em nm was measured using a Tecan infinite m200 plate reader (Tecan, Salzburg, Austria). After correction for medium only and no-drug controls, data points were fitted to a sigmoidal dose-response curve with variable slope using GraphPad Prism Version 5.00 (GraphPad Software, San Diego, CA, USA): Y = 100/(1 + 10^((LogIC50-X)*HillSlope))). At least three independent experiments were used to determine the half maximal inhibitory concentration (IC₅₀) values for each drug/cell line combination.

The SMARTpool small interfering RNA (siRNA) targeting Ezh2 and the non-targeting control were purchased from Dharmacon (Lafayette, CO, USA). Prior to all knock-down experiments optimal transfection conditions were determined for all cell lines. Cells were plated on day 0 and either transfected with 2 μM siRNA using DharmaFECT transfection reagent according to manufacturer's protocol (Dharmacon, Lafayette, CO, USA) or supplied with 5 μM DZNep on day 1. For protein and RNA analysis cells were harvested 48, 72 and 96 hours after transfection. The effect on cell growth was quantified using a Cell Titer Blue cell viability assay as described above. Cells were plated on day 0 in a density to allow exponential growth during the whole experiment and either transfected with siRNA or treated with 5 μM DZNep on day 1. Fluorescence was recorded 24, 48, 72 and 96 hours after transfection. Cell culture images were obtained using a Zeiss Axiovert 25 microscope with 10× objective on a Sony Cybershot. In all cases, data are presented from at least three independent experiments.

To define the molecular changes associated with BRCA1-deficient breast cancer, we previously compared BRCA1-deficient (KB1P) mammary tumors derived from our conditional K14cre;Brca1^F/F;p53^F/F mouse model for hereditary breast cancer with BRCA1-proficient mammary tumors (KP) derived from K14cre;Brca1^w.t/w.t;p53^F/F mice. Gene expression microarray analysis showed that KB1P tumors expressed markers of basal-like breast cancer, for example p63 and keratin 5, compared with the KP tumors 1232. Strikingly, the polycomb repressor EZH2 is also higher expressed in BRCA1-deficient tumors than in BRCA1-proficient control tumors (Figure 1a). Whereas there is heterogeneity in the BRCA1-proficient group, virtually all BRCA1-deficient tumors display increased Ezh2 expression, suggesting that in the absence of BRCA1 increased levels of EZH2 may be required. To determine whether the increase in mRNA levels translates to higher EZH2 protein expression, we analyzed tissue sections from both KB1P and KP tumors by immunohistochemistry (Figure 1b). We indeed found that BRCA1-deficient mouse mammary tumors have higher EZH2 protein levels than control tumors, also indicated by the higher percentage of tumor cells with EZH2 expression above background (77% in KB1P tumors versus 11.5% in KP tumors; Wilcoxon P < 0.029; Figure 1c).

Next, we determined whether EZH2 is also overexpressed in human BRCA1-deficient breast cancer. Recently, we and others showed that EZH2 levels are high in breast tumors with a poor prognosis 161833. Tumors from BRCA1-mutation carriers belong to this group of aggressive breast cancer, and accordingly EZH2 mRNA levels are also high in human BRCA1-deficient tumors (Figure 2a). Consistent with our previous observation that EZH2 mRNA and protein levels correlate relatively well 16, we observed increased EZH2 protein levels in human BRCA1-deficient tumor sections compared with other breast tumors (Figure 2b). To allow direct comparison, we simultaneously performed immunohistochemistry for EZH2 on sections from luminal A, basal-like and BRCA1-mutated breast tumors. As previously reported 161833, we found EZH2 levels to be higher in basal-like tumors compared with luminal A-type tumors (Figure 2b). By the same detection criteria, EZH2 levels in the BRCA1-deficient tumors were at least as high as in the sporadic basal-like tumors (no significant difference, Wilcoxon P < 0.164) and significantly increased compared with luminal A-type tumors (Wilcoxon P < 0.002, Figure 2c).

To determine whether the increased EZH2 levels are functionally relevant in the BRCA1-deficient tumor cells, we made use of cell lines that were derived from KB1P and KP mouse mammary tumors 23. In total, we used a panel of three BRCA1-deficient and three BRCA1-proficient cell lines, all derived from individual primary tumors. Although DZNep is a selective inhibitor of EZH2 function, some effects on other epigenetic marks, such as H4K20 methylation, have been reported 22. To ascertain whether observed effects with DZNep are due to its inhibition of EZH2 specifically, we included siRNAs targeting Ezh2 in our experiments.

Quantitative PCR demonstrated efficient knockdown of Ezh2 mRNA levels in both BRCA1-proficient and BRCA1-deficient cells (Figure 3a). DZNep does not affect Ezh2 mRNA levels, as reported previously 22. However, treatment with either Ezh2-specific siRNAs or DZNep resulted in significant reduction of EZH2 protein levels in both KB1P and KP cells (Figure 3b). Forty-eight hours after treatment there was visible toxicity in KB1P cells when EZH2 levels were reduced by either DZNep or siRNAs targeting Ezh2, but not in KB1P cells treated with non-targeting control siRNAs (Figure 3c). In contrast, there was no apparent effect of reduced EZH2 levels, either by Ezh2 knock-down or DZNep treatment, in BRCA1-proficient tumor cells. A more quantitative assessment of the effect of the treatments by a growth inhibition assay revealed that there is some adverse affect of DZNep in the KP cells. However, this effect of DZNep is unrelated to EZH2, as knock-down of Ezh2 does not inhibit the growth of these cells (Figure 3d, blue lines). Possibly, this is due to the effect of DZNep on H4K20 or other methylation events. In contrast, KB1P cells are severely affected by reduced EZH2 levels, as demonstrated by a strong growth inhibition of KB1P cells treated with siRNAs targeting Ezh2 (Figure 3d, red lines). In BRCA1-deficient cells, treatment with DZNep inhibited growth even more effectively than knock-down of Ezh2, which could be due to a more effective depletion of EZH2 by DZNep than that achieved by siRNAs, or due to possible effects of DZNep on other epigenetic marks. Nevertheless, DZNep shows remarkable selectivity in inhibiting BRCA1-deficient tumor cells compared with BRCA1-proficient tumor cells (Figures 3c, d).

To better quantify the difference in sensitivity to DZNep between KB1P and KP cells, we performed a dose-response curve (Figure 4a). Strikingly, the average IC₅₀ for BRCA1-deficient cells is 163 nM, whereas an almost 19-fold higher dose is required for 50% growth inhibition in BRCA1-proficient cells (average IC₅₀ of 2944 nM, P < 0.0001, t-test; Table 1).

To exclude the possibility that KB1P cells are in general more sensitive to epigenetic inhibitors we tested the effect of the histone deacetylase-inhibitor TSA in the same growth inhibition assay (Figure 4b). TSA affected KB1P and KP cell lines to a similar extend (average IC₅₀ of 26 nM vs 29 nM) showing no significant difference (P = 0.5; t-test, Table 1). When the cell lines were grown under non-adherent conditions, DZNep also inhibited sphere-formation, suggesting that there is no subpopulation of BRCA1-deficient cells that is resistant to DZNep treatment [see Additional data file 1]. However, in vivo experiments should demonstrate whether targeting EZH2 inhibits all tumor-initiating potential.

As loss of BRCA1 function results in genomic instability, we wanted to establish whether the dependence on EZH2 is a direct consequence of Brca1 loss, or whether this is a secondary effect caused by mutations accumulated during the tumorigenic process. To test this, we re-introduced a BAC clone encompassing the complete human BRCA1 gene into a BRCA1-deficient cell line (KB1P-3.12) and derived several clones that were shown to re-express BRCA1 (Figure 5a). These cells became less sensitive to cisplatin treatment indicating that the introduced BRCA1 is functional (data not shown). Of note, we did not observe a decrease in EZH2 levels in the reconstituted cell lines, indicating that BRCA1 does not directly influence Ezh2 expression (Figure 5b). However, treatment with DZNep reduces EZH2 levels to a similar extent in all cell lines (Figures 3b and 5d, and data not shown).

Interestingly, when the reconstituted cell lines were treated with DZNep, we observed a substantial rescue from DZNep-induced cell death (Figure 5c). The IC₅₀ values for DZNep in the BRCA1-reconstituted cells lines were more similar to the IC₅₀ values of the KP cells, than those of KB1P cells (average IC₅₀ of 1391 nM, Table 2). This shows that sensitivity of BRCA1-deficient cells to DZNep is mainly due to a loss of BRCA1 function, and not due to secondary mutations. This also indicates that there is a synthetic lethal effect; the effect of targeting one gene (e.g. EZH2) becomes deleterious specifically in the absence of another gene (e.g. BRCA1) 34.

Although the reconstituted cells become over eight-times more resistant to DZNep, the rescue is not complete. This could be due to differences in mouse and human BRCA1 or to technical issues with achieving the correct amount of BRCA1 expression. Alternatively, additional mutations could play a minor role in the sensitivity to DZNep.

In summary, we have demonstrated that DZNep selectively inhibits BRCA1-deficient but not BRCA1-proficient mammary tumor cells, and that this effect is mainly due to the fact that BRCA1-deficient cells are dependent on EZH2, whereas BRCA1-proficient cells are not.