Combined inhibition of EZH2 and ATM is synthetic lethal in BRCA1-deficient breast cancer

Previously, we reported that EZH2 is highly expressed in tumors associated with a BRCAness profile. Here, the association of EZH2 mRNA expression with BRCA1-mutant breast cancer was further evaluated using the TCGA BRCA whole exome sequencing data set (n = 526) confirming the increased expression of EZH2 in BRCA1-mutant (n = 18), compared to BRCA1-wild type breast cancer (n = 508) (p = 0.0003; Fig. 1A, left panel). Significantly increased Ezh2 expression in female BRCA1-deficient K14cre; Brca1^F/F; Trp53^F/F (KB1P) mouse mammary tumors compared to BRCA1-proficient K14cre; Trp53^F/F (KP) mammary tumors was further corroborated by RNA sequencing of tumor tissue (p = 0.0156; Fig. 1A, right panel).

To study the role of EZH2 in BRCA1-mutant breast cancer, we made use of previously described murine tumor cell lines derived from KB1P (clones KB1P-G3 and KB1P-B11) or KP (clones KP-3.33 and KP-6.3) mammary tumors [29]. First, inhibition of the EZH2 methyltransferase activity by GSK126 was assessed, showing a reduced H3K27me3 level by immunoblotting (Fig. 1B). We then determined the IC₅₀ value of GSK126 for all cell clones, showing that 3- to 11-fold lower concentrations of GSK126 were required for a 50% reduction in viability in BRCA1-deficient compared to BRCA1-proficient cells, indicating that BRCA1-deficient mammary tumor cells are more sensitive to EZH2 inhibition (Fig. 1C; KB1P-G3: IC₅₀ = 13.4 µM; KB1P-B11: IC₅₀ = 17.6 µM; KP-3.33: IC₅₀ = 44.9 µM; KP-6.3: IC₅₀ = 148.9 µM).

In order to identify synthetic lethal partners of EZH2 that are specifically effective in BRCA1-deficient breast cancer, we used BRCA1-deficient and -proficient mammary tumor cell lines to perform a high-throughput drug screen in combination with GSK126. We analyzed a panel of 27 distinct compounds targeting main oncogenic signaling pathways, such as DNA repair and cell cycle checkpoint signaling, and included small-molecule inhibitors that are in preclinical testing for cancer treatments. Single agent profiles for IC₅₀ determination were assessed in serial dilutions for all compounds in BRCA1-deficient (KB1P-G3 and KB1P-B11) and BRCA1-proficient (KP-3.33 and KP-6.3) cells (Additional file 1: Fig. S1 and Additional file 8: Table S1). Each compound was then profiled for synergistic interaction with GSK126, using 6 × 6 concentration combinations (vehicle control, 5 single concentrations per compound, 25 combinations). For analysis of potential synergy, the Bliss independence model was used with a score of  >15 indicative of a synergistic compound interaction (Additional file 8: Table S1). Synergy scores were further analyzed based on the difference between BRCA1-deficient and BRCA1-proficient cell lines. Synergistic compound screening highlighted the two Ataxia telangiectasia mutated (ATM) inhibitors AZD1390 [30] and KU60019 [31] as the best hits with the highest difference in synergy scores between BRCA1-deficient and -proficient cells (Fig. 1D). Synergistic effects were visualized using the web-application tool SynergyFinder [32]. Heatmaps for GSK126/AZD1390 and GSK126/KU60019 combination treatments are depicted in Fig. 1E. Visualized synergy heatmaps and synergy scores for the remaining compounds in combination with GSK126 are shown in Additional file 2: Fig. S2 and Additional file 8: Table S1, respectively. Since the highest synergy score was observed for AZD1390 (Bliss synergy score: 26.6 in KB1P-G3, 22.9 in KB1P-B11), this ATM-kinase specific inhibitor was prioritized for further analysis in combination with GSK126.

ATM together with ATR (ataxia telangiectasia and Rad3 related) are DNA damage sensor kinases that, upon induction of DNA damage, phosphorylate checkpoint kinase 1 and 2 (CHK1/2), respectively [33]. This results in G1-S and G2-M cell cycle checkpoint activation and initiation of DNA repair. ATM is a central molecule in the signaling cascade of DSBs. In the absence of an intact DNA repair mechanism, collateral ATM-deficiency leads to increased genomic instability [34].

The ATM inhibitor AZD1390 was able to inhibit downstream ATM signaling as shown by reduced Kap1 (S824) phosphorylation at various doses (0.1–2 µM) in BRCA1-deficient and BRCA1-proficient cells (Fig. 2A). However, AZD1390 showed 4- to 6-fold lower IC₅₀ values for BRCA1-deficient cells than BRCA1-proficient cells, indicating that BRCA1-deficient cells are more sensitive to ATM inhibition (Fig. 2B; KB1P-G3: IC₅₀ = 6.70 µM; KB1P-B11: IC₅₀ = 6.8 µM; KP-3.33: IC₅₀ = 26.9 µM; KP-6.3: IC₅₀ = 42.90 µM).

In addition to genetic context, the choice of compound concentration is crucial to achieve maximal drug synergy and to avoid over- or under-treatment resulting in unspecific toxicity or response failure [35]. Therefore, a concentration optimization experiment was performed to determine the compound concentration with the highest synergy between GSK126 and AZD1390. BRCA1-deficient and -proficient mouse mammary tumor cells were treated with increasing concentrations of one compound in the absence or presence of a fixed concentration of the other compound and vice versa (Additional file 3: Fig. S3). Single agent GSK126 treatment at intermediate concentrations induced moderate cytotoxicity in BRCA1-deficient cells, but strong cytotoxicity with the highest concentration of 10 µM (Additional file 3: Fig. S3A, panels in second row). The cytotoxic effect could be potentiated by adding a fixed concentration of AZD1390, while this concentration had no cytotoxic effect as a single treatment (Additional file 3: Fig. S3B). Treatment with increasing concentrations of single agent AZD1390 slightly decreased BRCA1-deficient cell viability, while no cytotoxic effect was observed in BRCA1-proficient cells (Additional file 3: Fig. S3B). When introducing GSK126 at a fixed concentration, BRCA1-deficient cells were sensitized to the cytotoxic effect of AZD1390, while BRCA1-proficient cells remained unaffected by this combination. As was evident from the IC₅₀ determination, high concentrations of single GSK126 or AZD1390 treatment also induced toxicity in BRCA1-proficient cells (Figs. 1C, 2B), however, this cytotoxicity was not attributed to a synergistic effect but rather to single agent-induced toxicity. This experiment revealed maximal synergistic interaction between 7.5 µM GSK126 and 3 µM AZD1390 in BRCA1-deficient cells (Additional file 3: Fig. S3A, panels in third row and Fig. S3B, panels in fourth row). Importantly, this combination treatment also induced slight cytotoxicity in BRCA1-proficient cells. We therefore used 7.5 µM GSK126 and 2 µM AZD1390 in subsequent validation experiments, which is below the half-maximal inhibitory concentration of both compounds. First, the inhibitory effect on cell viability was validated in an independent cell viability measurement using CellTiter-Glo (Fig. 2C). GSK126 and AZD1390 single agent treatment induced 17% and 7% reduction of viability in BRCA1-deficient cells, respectively. Remarkably, the combination of both drugs displayed 93% cytotoxicity in BRCA1-deficient cells (Fig. 2C), while no significant toxicity was observed in BRCA1-proficient cells, confirming that the reduction in cell viability was triggered by context-specific synergistic activity of GSK126 and AZD1390 in BRCA1-deficient cells.

Combined EZH2/ATM inhibitor treatment was subsequently validated using different functional assays. Assessment of clonogenic survival revealed a substantial decrease in the number of colonies by combined GSK126/AZD1390 treatment in BRCA1-deficient mouse mammary tumor cells, compared to vehicle control, as well as to single GSK126 or AZD1390 treatment (Fig. 3A). Quantification of colony formation showed that combined GSK126/AZD1390 treatment caused a 79–81% decrease in colony-forming units in BRCA1-deficient cells compared to vehicle control (KB1P-G3: p < 0.0001, KB1P-B11: p = 0.002), a 73–86% decrease compared to single GSK126 treatment (KB1P-G3: p = 0.0001, KB1P-B11: p = 0.01), and a 50–57% decrease compared to single AZD1390 treatment (KB1P-G3: p = 0.0001, KB1P-B11: p = 0.001), but not in BRCA1-proficient cells, indicating a robust synergistic effect in BRCA1-deficient cells (Fig. 3B). Continuous live-cell imaging using IncuCyte analysis revealed that growth inhibition in BRCA1-deficient cells through combined GSK126/AZD1390 treatment occurred between 30–40 h after induction of treatment, resulting in 68–79% growth reduction after 120 h in BRCA1-deficient cells, compared to vehicle control (KB1P-G3: p < 0.0001, KB1P-B11: p < 0.0001) (Fig. 3C), while growth was not affected in BRCA1-proficient cells. Single agent treatment showed a mild growth inhibitory effect in BRCA1-deficient cells, which was not statistically significant compared to vehicle control.

To investigate whether combined EZH2/ATM inhibition induces apoptosis in BRCA1-deficient mouse mammary tumor cells, we measured the percentage of Annexin V/propidium iodide double-positive cells by flow-cytometry (Fig. 3D). Indeed, the percentage of apoptotic cells drastically increased by combined GSK126/AZD1390 inhibition compared to vehicle control or single agent treatment in BRCA1-deficient cells (KB1P-G3: 41.8% by GSK126/AZD1390 compared to 9.1% by vehicle (p < 0.0001), 15.1% by GSK126 (p < 0.0001) and 12.3% by AZD1390 (p < 0.0001); KB1P-B11: 41.7% by GSK126/AZD1390 compared to 10.1% by vehicle (p < 0.0001), 13.3% by GSK126 (p < 0.0001) and 10.8% by AZD1390 (p < 0.0001)). In contrast, the percentage of apoptotic cells in BRCA1-proficient cells did not change significantly upon combined EZH2/ATM inhibition compared to single agent treatment or vehicle control (KP-3.33: 7.2% by GSK126/AZD1390 compared to 5.3% by vehicle (p = 0.17), 6.3% by GSK126 (p = 0.78) and 6.7% by AZD1390 (p = 0.95); KP-6.3: 8.5% by GSK126/AZD1390 compared to 4.2% by vehicle (p = 0.0004), 4.2% by GSK126 (p = 0.0004, showing a statistically significant, however, biologically insignificant difference) and 6.6% by AZD1390 (p = 0.14) (Fig. 3D). Together, these results indicate that the synergistic cytotoxicity of combined EZH2/ATM inhibition in BRCA1-deficient mouse mammary tumor cells is mediated by apoptosis.

To investigate the cytotoxic effect of combined EZH2 and ATM inhibition in human BRCA1-mutant breast cancer, we treated the human TNBC cell lines SUM149 (BRCA1-mutant) and CAL120 (BRCA1-wild type) with GSK126 and AZD1390, and analyzed cell viability using CellTiter-Glo and clonogenic survival (Fig. 4A, B). SUM149 cells were specifically sensitive to combined GSK126/AZD1390 inhibition leading to a significantly reduced cell viability by 77–79%, compared to vehicle control (p = 0.02) or single agent treatment (GSK126: p = 0.02; AZD1390: p = 0.03), while no effect was observed in CAL120 cells (Fig. 4A), suggesting that the cytotoxic effect by EZH2/ATM inhibition might be specific to BRCA1 mutation status in human breast cancer cells. Again, assessment of clonogenic survival revealed a significant decrease in the number of colonies by combined GSK126/AZD1390 treatment in BRCA1-mutant TNBC cells, compared to vehicle control, as well as to single GSK126 or AZD1390 treatment (Fig. 4B). Quantification of colony formation revealed that combined GSK126/AZD1390 treatment caused a 78% decrease in colony-forming units in BRCA1-mutant SUM149 cells compared to vehicle control (SUM149: p < 0.015), a 69% decrease compared to single GSK126 treatment (SUM149: p = 0.004), and a 57% decrease compared to single AZD1390 treatment (SUM149: p = 0.012), but not in BRCA1-wild type CAL120 cells, corroborating a robust synergistic effect also in human BRCA1-mutant TNBC cells (Fig. 4B).

To confirm that the observed treatment effect is due to combined EZH2/ATM inhibition and not caused by any potential off-target effect, BRCA1-deficient and -proficient mouse mammary tumor cells were treated with the structurally distinct EZH2 and ATM inhibitors ZLD1039 [36] and KU60019 [31], respectively (Fig. 4C, D). Combined EZH2/ATM inhibition by GSK126/KU60019 or ZLD1039/AZD1390 treatment reduced the viability of BRCA1-deficient cells compared to vehicle control by 80–90% (KB1P-G3: p < 0.0001, KB1P-B11: p = 0.017), and by 42–59% (KB1P-G3: p = 0.0005, KB1P-B11: p < 0.0001), respectively, whereas growth of BRCA1-proficient cells was not significantly affected by either combination. These data indicate that the cytotoxic effects of the drug combinations are resultant of the combined EZH2/ATM inhibition and not due to unspecific compound effects.

To further establish the role of EZH2 inhibition in sensitizing BRCA1-deficient cells to ATM inhibition and to exclude off-target drug activity, we made use of a doxycycline (Dox)-inducible shRNA-mediated knockdown system [37] allowing stable and inducible EZH2 knockdown in BRCA1-deficient (KB1P-G3-shEZH2) and BRCA1-proficient (KP-3.33-shEZH2) cells. Following seven days of Dox-induced EZH2 knockdown, a strong decrease in EZH2 protein expression and global H3K27 trimethylation was observed in shEZH2-expressing cells but not in cells expressing a Dox-inducible random shRNA (shRandom) sequence (Fig. 4E). Next, clonogenic growth was assessed upon Dox-induced EZH2 knockdown alone or in combination with the ATM inhibitor AZD1390 (Fig. 4F, G). In BRCA1-deficient cells, EZH2 knockdown combined with AZD1390 induced 64% reduction in clonogenic growth compared to vehicle control (KB1P-G3-shEZH2: p < 0.0001), while shRandom-expression in combination with AZD1390 had no effect (Fig. 4F, G, upper two panels). In BRCA1-proficient cells, a mild effect was observed upon EZH2 knockdown in combination with AZD1390, however, this effect was not statistically significant (KP-3.33-shEZH2: p = 0.12; Fig. 4F, G, lower two panels).

We hypothesized that the observed increase in apoptotic cell death in BRCA1-deficient cells by combined EZH2/ATM inhibition was driven by higher levels of DNA damage. Immunoblotting showed inhibition of phospho-ATM and downstream phospho-Kap1 after AZD1390 single and combination treatment with GSK126, which is the expected effect of AZD1390-mediated ATM inhibition (Additional file 5: Fig. S5). However, immunoblotting is not sufficiently sensitive to detect slight changes in phosphorylated nuclear histone H2AX (yH2AX)yH2AX foci [38]. We used immunofluorescence staining as a more sensitive technique to detect yH2AX changes, together with the previously established automated high-throughput microscopy [39] to improve accuracy for counting of yH2AX foci formation and to assess the level of DNA damage and repair ability in BRCA1-deficient and BRCA1-proficient mouse mammary tumor cells treated with GSK126 and AZD1390 (Fig. 5A, B). The number of yH2AX foci per cell, which correspond with sites of DSBs, was highest in KB1P-G3 and KB1P-B11 cells treated with combined GSK126/AZD1390, showing a two to threefold increase, compared to vehicle control or single agent treatment (Fig. 5C). We also detected a slight, but statistically significant increase of yH2AX foci per cell upon combined GSK126/AZD1390 treatment in KP-3.33 and KP-6.3 cells, which, however, does not correlate with increased cytotoxicity (Fig. 3). Cisplatin (1 µM) served as a positive control and brought about a two to threefold increased number of yH2AX foci per cell as a result of DNA adduct formation. As expected, the number of detected cells per field was lowest in GSK126/AZD1390 and cisplatin-treated BRCA1-deficient cells (Additional file 4: Fig. S4). These results indicate that combined EZH2/ATM inhibition induces increased DNA damage in BRCA1-deficient mouse mammary tumor cells.

To test the effect of combined EZH2/ATM inhibition in vivo, mice bearing tumor fragments derived from KB1P donor mice (Fig. 6A) were treated with GSK126 and AZD1390. Dose optimization was performed prior to survival analysis (Additional file 6: Fig. S6). For drug treatment (Fig. 6), mice were treated with vehicle, GSK126 (150 mg/kg, by daily intraperitoneal injection), AZD1390 (20 mg/kg, by oral gavage twice daily, 5 days on, 2 days off) or the combination of GSK126 and AZD1390 for 28 consecutive days, as evaluated previously [14, 30]. We observed a modest effect on tumor growth by single GSK126 or AZD1390 treatment. However, after combined GSK126/AZD1390 therapy tumor growth was reduced compared to the single agents (Fig. 6B). As a result of the treatment, progression-free survival was significantly increased by single agent treatment using GSK126 compared to vehicle control (p = 0.002), as wells as with AZD1390 treatment compared to vehicle control (p = 0.039). The longest progression-free survival was observed in the GSK126/AZD1390 combination therapy cohort, which was significantly increased compared to vehicle (p = 0.0001), and to single agent treatment with AZD1390 (p = 0.015) (Fig. 6C). Drug treatment was well tolerated in vivo and showed no toxicity to the animals (Additional file 7: Fig. S7). Immunohistochemistry analysis of GSK126 and AZD1390 treated tumors confirmed effective target inhibition by reduced H3K27me3 and by phosphorylated ATM levels, respectively (Fig. 6D). The results of double agent treatment in vivo suggest a combination effect by GSK126/AZD1390 treatment and support our in vitro findings of increased sensitivity in BRCA1-deficient mammary tumor cells.

Breast invasive carcinoma patient data from The Cancer Genome Atlas Breast Cancer (TCGA BRCA) were retrieved using the UCSC Cancer Browser (https://​genome-cancer.​ucsc.​edu/​). For graphical view of genomic data, whole-exome sequencing (n = 526) data were analyzed using the Xena browser at the cBioPortal (http://​www.​cbioportal.​org). A detailed description of data generation and instructions can be viewed on https://​xenabrowser.​net [61].

Gene expression was analyzed in KB1P and KP mouse mammary tumors. Illumina TruSeq mRNA libraries were generated and sequenced with 50–65 base single reads on a HiSeq 2500 using v4 chemistry (Illumina Inc., San Diego). The resulting reads were trimmed using Cutadapt (version 1.15) to remove any remaining adapter sequences and to filter reads shorter than 20 bp after trimming to ensure good mappability. The trimmed reads were aligned to the GRCm38 reference genome using STAR (version 2.6.1a [62]). Gene expression counts were generated by feautureCounts (version 1.5.0-p1 [63]) using genome definitions from Ensembl GRCm38 version 76. Normalized expression values were obtained by correcting for differences in sequencing depth between samples using DESeq median-of-ratios approach [64] and the log-transforming normalized counts. To statistically test the differences between the groups, ANOVA and pairwise t-test with multiple testing correction was used. The RNA sequencing data reported in this study are available in the NCBI GEO database (GSE182448).

Murine tumor cell lines were generated from individual tumors arising in female KB1P or KP mice as described previously [29]. Established cell lines were cultured at 37 °C with 5% carbon dioxide under low oxygen conditions (3%) in DMEM/F12 medium (Thermo Fisher, Cat#31331028) supplemented with 10% FCS, 1% Penicillin–Streptomycin (5000 U/mL, Thermo Fisher, Cat#12140122), 5 mg/mL insulin (Sigma, Cat#53003-018), 5 ng/mL epidermal growth factor (Thermo Fisher, Cat#I6634), and 5 ng/mL cholera toxin (Sigma-Aldrich Israel (Cat#C8052).

KB1P-G3 and KP-3.33 cells stably expressing shEzh2 were cultured in DMEM/F12 medium supplemented with 10% FCS, 1% Penicillin–Streptomycin, 5 µg/mL insulin, 5 ng/mL epidermal growth factor, and 5 ng/mL cholera toxin. shRNA expression was induced with 100 ng/ml doxycycline (Dox) (Sigma-Aldrich, Cat#D9891) in culture medium. Medium of uninduced cells was supplemented with the respective volume of PBS.

Human SUM149 cells were cultured in DMEM/F12 medium supplemented with 10% FCS, 1% Penicillin–Streptomycin, 5 µg/mL insulin, and 1 µg/mL hydrocortisone (Sigma, Cat#0315). Human CAL120 cells were cultured in RPMI1640 (Thermo Fisher, Cat#12633012) supplemented with 10% FCS, 1% Penicillin–Streptomycin.

All cell lines were tested negative for Mycoplasma contamination upon thawing using a PCR Mycoplasma Test Kit (AppliChem, Cat#A3744).

The following compounds were purchased from Selleckchem: AZD1390 (ATM inhibitor, Cat#S8680), AZD7762 (CHK1/2 inhibitor, Cat#S1532), BI2536 (PLK1 inhibitor, Cat#S1109), BKM120 (PI3K inhibitor, Cat#S2247), crizotinib (ROS1 inhibitor, Cat#S1068), dinaciclib (CDK inhibitor, Cat#S2768), gefitinib (EGFR inhibitor, Cat#S1025), GSK126 (EZH2 inhibitor, Cat#S7061), JQ1 (BET inhibitor, Cat#S7110), KU60019 (ATM inhibitor, Cat#S1570), KU60648 (DNA-PK inhibitor, Cat#S8045), LDC67 (CDK9 inhibitor, Cat#S7461), MK1775 (Wee1 inhibitor, Cat#S1525), olaparib (PARP inhibitor, Cat#S1060), palbociclib (CDK4/6 inhibitor, Cat#S1579), panobinostat (HDAC inhibitor, Cat#S1030), PF477736 (CHK1 inhibitor, Cat#S2904), purvalanol A (CDK1/2 inhibitor, Cat#S7793), RO3306 (CDK1 inhibitor, Cat#S7747), selisitat (SIRT1 inhibitor, Cat#S1541), selumetinib (MEK inhibitor, Cat#S1008), senexin A (CDK8/19 inhibitor, Cat#S8520), TH287 (MTH1/NUDT1 inhibitor, Cat#S7631), THZ1 (CDK7 inhibitor, Cat#S7549), VE822 (ATR inhibitor, Cat#S7102), venetoclax (BCL2 inhibitor, Cat#S8048). NSC663284 (Cdc25 inhibitor, Cat#383907-43-5) was purchased from Cayman Chemical, PF3644022 (MK2 inhibitor, Cat#B5549) from ApexBio, ZLD1039 (EZH2 inhibitor, Cat#AOB9716) from AOBIOUS. All compounds were dissolved in DMSO (Carl Roth, Cat#A994.2) at a concentration of 10 mM. Equal amounts of DMSO added to the cell culture medium served as vehicle control.

Optimal seeding density per cell line was derived from growth curves performed prior to screening experiments. Cells from subconfluent cell culture dishes were filtered through a cell strainer and viable cells were counted with Countess™ II FL Automated Cell Counter (Invitrogen) to allow optimal exponential growth (log phase) during the whole experiment. Cells were plated into 384-well plates (KB1P and KP cells: 500 cells/well, SUM149 and CAL120 cells: 1000 cells/well) in 30 µl complete culture medium. The compounds were added after 24 h by using the TECAN D300e digital dispenser (HP). After 72 h of treatment, cell viability was assessed by measuring ATP content in each well using CellTiter-Glo Reagent (Promega, Cat#G7573) 1:1. Luminescence intensity was measured using a plate reader (Tecan Infinite M1000 Pro) and normalized to intensities of control wells.

Before screening for drug synergy, single agent effects of all compounds were profiled on the KB1P and KP mouse mammary tumor cell lines for 10 different concentrations in two-fold serial dilutions, ranging from 20 nM to 20 µM and assessed by cell viability measurement using CellTiter-Glo. The concentration-effect relationship (IC₅₀ values) was determined by logistic interpolation using GraphPad Prism. See Additional file 8: Table S1 for IC₅₀ values.

Next, in a pre-screen for drug synergy, GSK126 was tested against each compound in two-fold serial dilutions to determine the optimal drug concentrations for synergy analysis using a 6 × 6 matrix for drug combinations (5 concentrations for each compound and DMSO control). For the main synergistic combination screen, 5 representative concentrations titrated around the previous determined IC₅₀ values were used for each compound. The compounds were then profiled in combination with GSK126 using the 6 × 6 matrix layout, and cells were treated as described above See Additional file 8: Table S1 for compound concentrations. Cell viability was detected after 72 h using CellTiter-Glo Reagent. See Additional file 9: Table S2 for normalized cell viability data used for calculation of synergy scores.

Analysis of synergy scores was determined using the Bliss independence model calculating the difference between observed and expected compound effects [65] as described before [66]. Briefly, single agent effects of compounds A and B (α^A, α^B) at concentrations Cx^A and Cy^B were used to calculate the expected effect for additive compound interactions (α_exp = α^A + α^B − α^A * α^B), and subsequently compared to the effect observed under combination of both compounds (α_obs). The delta score (Δα = α_obs − α_exp) calculated as the difference between observed and expected effects over the full dose–response matrix characterizes the synergistic effect. A score of > 15 was used as threshold indicating a synergistic compound interaction [66]. Synergistic effects were visualized using the web-application tool SynergyFinder version 2.0 [32].

Cells were seeded at a density of 25,000 cells (KB1P and KP cells) or 50,000 cells (SUM149 and CAL120) per well into 6-well cell culture plates. After overnight incubation, cells were continuously treated with GSK126 (7.5 µM) or with AZD1390 (2 µM) alone or in combination for 7 days. DMSO was used as vehicle control.

After the treatment period, colonies were fixed with methanol on ice, stained with 0.5% crystal violet (Sigma-Aldrich, Cat#HT90132) and imaged using Carl Zeiss Stemi 2000-C Stereo microscope equipped with a CCD camera (Zeiss) at 0.65× magnification. Colonies were counted on whole surface area using the Clono-counter software and values were normalized to control wells [67].

Cells were plated into 384-well plates (500 cells/well) in 30 µl complete culture media. After overnight incubation, cells were treated with GSK126 (7.5 µM), AZD1390 (2 µM) or the combination of both. DMSO served as vehicle control. Cells were allowed to grow for 120 h. Phase-contrast images were automatically acquired by IncuCyte FLR (Essen Bioscience) from the incubator at 4-h intervals. Proliferation was monitored by analyzing the occupied area (% confluence) of cell images over time by IncuCyte software (Essen Bioscience).

Whole-cell lysates were prepared in RIPA lysis buffer (50 mM Tris–HCl pH 7.5, 100 mM NaCl, 0.1% SDS, 0.5% Sodium deoxycholate) supplemented with 1 mM PMSF, 1% SDS, and phosphatase inhibitors (1x PhosphoSTOP Phosphatase Inhibitor Cocktail, Roche Diagnostics, Cat#4906845001). Samples were lysed on ice for 20 min, sheared by passing cells through a G25 needle (B. Braun, Hypodermic Needle-Pro®, Cat#4658304), cleared by centrifugation (14,000 × g, 15 min), and quantified using Pierce™ Bradford protein assay kit (Thermo Scientific, Cat#23200). Lysates were boiled 5 min at 95 °C with 6x reducing Laemmli buffer (0.12 M Tris pH 6.8, 47% glycerol, 12% SDS, 0.6 M DTT, 0.06% bromophenol blue). Samples were separated on a polyacrylamide gel and transferred to PVDF membranes using the mini wet/tank blotting system (Bio-Rad). After blocking with 5% BSA in Tween-20/PBS, membranes were probed with primary antibodies prepared in blocking solution overnight at 4 °C on a roller, followed by incubation with horseradish peroxidase-conjugated secondary antibody in blocking solution for 1 h at room temperature and ECL detection (Thermo Fisher, Cat#34096) by the ChemiDoc XRS + system (Bio-Rad). Primary and secondary antibodies used for immunoblotting are listed in Additional file 8: Table S1. Quantitative analysis of protein expression relative to GAPDH was done using Image Lab software (Bio-Rad).

Cell lines (150,000 cells/well) were seeded into 6-well plates at 40–60% confluency. After overnight incubation, medium was replaced with growth medium containing single inhibitors GSK126 (7.5 µM) or AZD1390 (2 µM) or in combination for 48 h. Culture medium was collected, cells were trypsinized, washed with ice-cold PBS, and incubated with Annexin-V (BD Bioscience, Cat#556420) and propidium iodide (PI, Carl Roth, Cat#CN74, 5 µg/ml) in antibody binding buffer (2.5 mM CaCl₂, 10 mM HEPES (pH 7.4), 140 mM NaCl, 20% accutase solution (Sigma-Aldrich, Cat#A6964), 70% PBS). Annexin-V/propidium iodide staining was detected by flow cytometry (Beckman Coulter, Gallios Flow Cytometer). The fraction of apoptotic cells was quantified as the Annexin-V, PI, and double positive stained populations using the Kaluza analysis software (Beckman Coulter).

Cells (5000 cells/well) were seeded into 96-well imaging plates (Greiner Bio-One, µclear, Cat#655090). After overnight incubation, cells were treated with single or combined agents of GSK126 (7.5 µM) and AZD1390 (2 µM). Treatment with 1 µM cisplatin was used as positive control. For pre-extraction of nonchromatin-bound proteins, after 48 h of treatment cells were incubated with sucrose buffer (25 mM HEPES pH 7.5, 50 mM NaCl, 1 mM EDTA, 3 mM MgCl₂, 300 mM sucrose, 0.5% TritonX-100) for 2 min, followed by fixation in 4% paraformaldehyde for 10 min at RT. After 1 h of blocking (5% BSA, 2% normal goat serum, 0.1% TritonX-100, 0.05% Tween-20) at RT, cells were incubated with primary and secondary antibodies as listed in Additional file 8: Table S1. After three final washes, cells were covered with PBS and stored at 4 °C until scanning. Quantitative high-throughput microscopy was performed similarly as described before [66]. We used a Thermo Fisher Cellomics CellInsight CX7 LZR High Content Analysis (HCA) Platform with laser light source to scan stained cell models in 96-well imaging plates. 2 × 2 binned images (1104 × 1104 pixels) were acquired with a 20× (0.4 NA) Achroplan objective using the laser-based autofocus and analyzed using the Spotdetector V4.1 Bioapplication of the Cellomics software package (Version 6.6.2, Built 8533). Cell nuclei were identified by Hoechst 34580 staining in background corrected images (3D surface fitting) according to the object identification parameters size: 100–1500 μm², ratio of perimeter squared to 4π area: 1–5, length-to-width ratio: 1–5, average intensity: 500–8000, total intensity: 2x105–5x107. Foci were identified within the nuclear region using the Box method with a value of 3. Object selection parameters for foci were 1–30 μm², ratio of perimeter squared to 4π area: 1–5, length-to-width ratio: 1–5, average intensity: 500–16,000, total intensity: 3x102–1x106.

This study is compliant with all relevant ethical regulations regarding animal research. All animal experiments were approved by the Animal Ethics Committee of The Netherlands Cancer Institute (Amsterdam, the Netherlands) and performed in accordance with the Dutch Act on Animal Experimentation. Generation of conditional K14cre; Brca1^F/F; Trp53^F/F (KB1P) and K14cre; Trp53^F/F (KP) breast cancer mouse models were described previously [68, 69]. Orthotopic transplantations, tumor monitoring, and tissue sampling were performed as described before [70]. Briefly, donor KB1P tumor fragments were transplanted into the fourth mammary fat pad of 8-weeks old FVB females and treatments began upon reaching tumor outgrowth of approximately 100 mm³ (100%). Maximal tolerable dose (MTD) for the combined therapy with GSK126 and AZD1390 was determined prior to the intervention study using FVB females. GSK126 (Syncom) was reconstituted in 20% Captisol (CyDex Pharmaceuticals) and brought to a pH of 4.5 with 10 M potassium hydroxide, to create a working stock of 15 mg/mL. AZD1390 (Selleckchem) was dissolved at 40 mg/ml in ethanol and then dropwise added to 0.5% (w/v) HPMC, 0.1% (w/v) Tween-80 to achieve a final concentration of 2 mg/ml. Tumor-bearing mice were blindly randomized into four treatment groups and treated with vehicle (daily intraperitoneal injection), GSK126 (150 mg/kg, daily intraperitoneal injection), AZD1390 (bi-daily 20 mg/kg, by oral gavage for 5 days on, 2 days off) or a combination of GSK126 and AZD1390. Animals were treated for 28 consecutive days and mammary tumor volume (mm) was quantified by caliper measurements using the following formula: 0.5 × length × width². For a better comparability, the tumor volume at day x was normalized to the initial tumor volume at treatment start (day 0) and defined as relative tumor volume (RTV). The endpoint of this study was reached when tumor size was 10 times the RTV, defined as progression-free survival (PFS). Animals were euthanized by CO₂ when tumors extended a volume of 1500 mm³ or when severe side effects were observed. All procedures were carried out by animal technicians in a blinded fashion.

A small cohort of mice (n = 12) was sacrificed after 7 days of single agent or combined treatment with GSK126 and AZD1390, and KB1P tumors were harvested for immunohistochemistry analysis (IHC). IHC staining of EZH2, H3K27me3, and phosphorylated ATM was performed using formalin-fixed paraffin-embedded tumor tissue. The following monoclonal antibodies were used for immunohistochemistry: EZH2 (Rb, Cell signaling, Cat#5246, 1:200), H3K27me3 (Rb, Abcam, Cat#ab6002, 1:100) and phosphorylated ATM (phospho S1981, Rb, Abcam, Cat#ab81292, 1:400) overnight at 4 °C. These antibodies were extensively tested for target specificity [14]. All slides were digitally processed using the Aperio ScanScope (Aperio, Vista, CA, USA) and captured using ImageScope software version 12.0.0 (Aperio).

Statistical analysis was performed using GraphPad Prism Version 8 and 9. Mann–Whitney U test, one-way ANOVA with Tukey multiple comparison testing or Kaplan–Meier survival testing with log-rank comparison were used as indicated in the figure legends. p values below 0.05 were considered statistically significant. Significance levels are indicated as: *—p ≤ 0.05; **—p ≤ 0.01; ***p ≤ 0.001; nonsignificant levels are not labeled. Number of experimental replicates are indicated in the figure legends.