BRCA2-Deficient Sarcomatoid Mammary Tumors Exhibit Multidrug Resistance

A major clinical problem in cancer therapy is resistance of tumors to all available therapies, a phenomenon called pan-resistance (1). After an initial response primary tumors and especially metastases do not respond anymore to treatment, including radiotherapy. The frequently used term “multidrug resistance” historically refers to resistance due to drug efflux transporters, but upregulation of these transporters cannot fully explain pan-resistance. Drug resistance is not only a problem for classical chemotherapeutics, but also for targeted therapeutics. Mechanisms can be drug-specific, such as imatinib resistance caused by mutations in or overexpression of the drug target BCR–ABL (2), or downregulation of Top1 or Top2 causing resistance to topoisomerase I or II poisons (3). The precise mechanisms that cause resistance of tumors to multiple classes of drugs are not fully understood. One mechanism that has been put forward to explain pan-resistance of various types of cancer is epithelial-to-mesenchymal transition (EMT; refs. 1, 4). During EMT cells lose epithelial characteristics and acquire mesenchymal characteristics. EMT is a physiologic process involved in, for example, embryogenesis and wound healing, but it has also been described for epithelial cancers when cells acquire a spindle-shaped (also called “mesenchymal” or “sarcomatoid”) morphology and lose expression of cell adhesion molecules. In vitro, EMT was observed in various cell lines that acquired resistance to chemotherapeutic agents and targeted inhibitors (4), and induction of EMT by recombinant TGFβ treatment led to resistance to tyrosine kinase inhibitors and cisplatin (5), suggesting a role of EMT in pan-resistance.

Breast cancer is a heterogeneous disease, which comprises various histologic and molecular subtypes. Among these is the subgroup of metaplastic breast cancer, a variant of triple-negative breast cancer, which includes several morphologic entities, including spindle-shaped tumor cells (6, 7). A molecular subtype that is frequently observed in metaplastic cancers is the claudin-low signature (8, 9). Because metaplastic cancers have a poor prognosis, we wondered whether EMT might contribute to poor drug response of these tumors.

To study the influence of EMT on pan-resistance, we made use of a unique mouse model of BRCA2-deficient breast cancer, that is, the K14cre;Brca2^F/F;p53^F/F mammary tumor model (10). Female K14cre;Brca2^F/F;p53^F/F mice develop mostly epithelial mammary carcinomas, but also mesenchymal carcinosarcomas are formed. Because the K14 promoter drives Cre expression only in epithelial cells (10), it is plausible that these mesenchymal mammary tumors originate from an EMT. The advantage of such an in vivo model is that no cell lines have to be used, which may poorly represent the original tumor (11). We and others have previously shown that the BRCA2-deficient mouse mammary tumors are sensitive to DNA damage–inducing drugs and PARP inhibitors due to the lack of error-free repair of double-strand DNA breaks by homologous recombination (12–16). In patients with BRCA2-deficient breast cancer, such an increased sensitivity was also observed after neoadjuvant therapy with DNA-damaging agents (17, 18) or PARP inhibitors (19). We investigated whether this drug sensitivity is diminished in BRCA2-deficient carcinosarcomas. For this purpose, we compared the responses of epithelial carcinomas and mesenchymal carcinosarcomas with chemotherapy drugs and PARP inhibitors. We found that BRCA2-deficient carcinosarcomas are multidrug resistant, which was, at least in part, due to high expression of the drug efflux transporters P-glycoprotein (Pgp) and breast cancer resistance protein (BCRP), which transport a wide range of chemotherapeutic and targeted agents. In addition, we found that an EMT-like gene-expression profile correlates with Pgp expression in multiple independent mouse mammary tumor datasets.

Tumors were generated in K14cre;Brca2^F/F;p53^F/F (KB2P) female mice (10) and samples were taken for histology, RNA isolation and cryopreservation. Orthotopic transplantation of Brca2^Δ/Δ;p53^Δ/Δ, p53^Δ/Δ, and p53^Δ/Δ;Abcb1a/b^−/− tumors in wild-type FVB/Ola129 F1 mice was performed as previously described (20). The tumor size was monitored at least three times a week by calliper measurements. The tumor volume was calculated with the following formula: 0.5 × length × width². Animals were sacrificed with CO₂, when the tumor reached a size of 1,500 mm³. Cre-mediated deletion of exon 11 in Brca2 and exons 2–10 in Trp53 was confirmed by genotyping PCR and Southern blotting as described previously (14, 21). All animal experiments were approved by the Animal Ethics Committee of the Netherlands Cancer Institute (Amsterdam, the Netherlands).

Upon tumor outgrowth to approximately 200 mm³ (100%) the mice were either left untreated or received one of the following treatments: olaparib (50 mg/kg i.p., daily for 28 days), topotecan (4 mg/kg i.p., days 0–4 and days 14–18), doxorubicin (Amersham Pharmacia Netherlands; 5 mg/kg i.v., days 0, 7, and 14), docetaxel (Aventis; 25 mg/kg i.v., days 0, 7, and 14) or cisplatin (Mayne Pharma; 6 mg/kg i.v., day 0). Mice with a relapsing tumor received another treatment cycle when the tumor was 100% of the original size at treatment start. For the resensitization experiment with tariquidar (Fig. 4B), the mice received only one treatment cycle and tumor outgrowth was monitored. Tariquidar (Avaant; 10 mg/kg i.p.) was administered 15 minutes before treatment with olaparib, docetaxel, or doxorubicin. AZD2461 (100 mg/kg orally; ref. 22) was administered daily for 28 days.

Staining of E-cadherin and vimentin was performed on formalin-fixed paraffin-embedded (FFPE) tissue. Samples were boiled in Tris-EDTA pH9.0 to retrieve the antigens. Furthermore, we used 3% H₂O₂ in methanol to block endogenous peroxidase activity, and 10% milk (E-cadherin) or 4% BSA plus 5% normal goat serum in PBS (vimentin) as blocking buffer. Primary antibodies (mouse anti–E-cadherin, BD Transduction Laboratories 610182, 1:400; rabbit anti-vimentin, Cell Signaling Technology 5741, 1:200) were diluted in 1.25% normal goat serum plus 1% BSA in PBS. For detection and visualization labeled polymer-HRP (horseradish peroxidase) anti-mouse and rabbit Envision (Dako K4007 and K4011), DAB (Sigma D5905), and hematoxylin counterstaining were applied.

CD31 staining was done on cryosections after acetone fixation for 10 minutes at −20°C. Then we applied 0.3% H₂O₂ in methanol, avidin–biotin block (Dako X0590) and serum-free protein block (Dako X0909). The primary (rat anti-CD31, BD Biosciences 550274; 1:1,000) and secondary antibodies (biotinylated rabbit-anti–rat IgG, Dako E0468, 1:300) were diluted in 1% BSA in PBS. For detection and visualization, we used streptavidin–HRP (Dako K1016, 10 minutes incubation at room temperature), DAB (Dako K3468), and hematoxylin counterstaining.

For visualization of perfused blood vessels, biotinylated Lycopersicon Esculentum (Tomato) Lectin (B1175 Vector Laboratories) was added to streptavidin-AF633 (S21375 Invitrogen) in sterile PBS and injected in the tail vein 15 minutes before sacrificing the mouse. For visualization of the Lectin-AF633 signal, FFPE slides were deparaffinized, rehydrated, incubated with DAPI for 5 minutes, and mounted in Vectashield (H-1000 Vector Laboratories). Images were taken with a Leica TCS SP5 (Leica Microsystems) confocal system, equipped with a 405-nm Diode laser and 633-nm HeNe laser system.

Total RNA was isolated with TRIzol (Invitrogen) from snap-frozen tumor samples. The RNA was processed according to the manufacturer's instructions for single channel 45K MouseWG-6 v2.0 BeadChips (Illumina). Background correction was performed using the bg.adjust method from the Bioconductor affy package (23). For normalization between arrays, the robust spline method was applied.

A published EMT signature (5) was converted from human to mouse, resulting in 239 epithelial genes (down after EMT) and 224 mesenchymal genes (up after EMT). Of these, 235 epithelial and 223 mesenchymal genes could be mapped to Ensembl gene identifiers using Ensembl Biomart (24).

Generation of gene-expression data from p53^Δ/Δ and Cdh1^Δ/Δ;p53^Δ/Δ mouse mammary tumors has been described (Klijn and colleagues; submitted for publication). Gene-expression datasets GSE3165 (25), GSE23938 (26), and GSE10885 (27) were downloaded from the Gene Expression Omnibus.

Expression data were mean-centered per gene or probe for each dataset. The EMT score of each tumor in all datasets is calculated by subtracting the mean log₂ expression of the epithelial genes from the mean of the log₂ expression of the mesenchymal genes. Supplementary Table S1 summarizes the number of signature genes represented on each platform. As a result, tumors with a mesenchymal gene-expression profile have a positive score and tumors with an epithelial profile have a negative EMT score. For the correlation between EMT score and Abcb1a, Abcb1b, and Abcg2 expressions, we used the Spearman correlation and significance was tested with the t distribution test.

Array comparative genomic hybridization (aCGH) data from Brca2^Δ/Δ;p53^Δ/Δ mouse mammary tumors were downloaded from Array Express (E-NCMF-34 and E-NCMF-35). The phenotype was determined previously (28), with one exception: a tumor with unknown phenotype was later classified as carcinosarcoma. Unsupervised hierarchical clustering was performed using the Euclidean distance method with complete linkage.

Semiquantitative levels of Abcb1a, Abcb1b, and Abcg2 mRNA were measured with reverse transcriptase multiplex ligation–dependent probe amplification (RT-MLPA). Reverse transcription of total RNA, hybridization, ligation, PCR amplification, and fragment analysis by capillary electrophoresis were performed as described previously (29). Gene-expression levels were normalized to the internal reference Actb (β-actin).

To study the effect of a mesenchymal morphology on therapy response, we made use of the K14cre;Brca2^F/F;p53^F/F mouse model (10). K14cre-mediated deletion of exon 11 of Brca2 and exon 2–10 of Trp53 in mammary epithelial cells (Supplementary Fig. S1) results in the development of mammary tumors with an average latency of 181 days. We used an established orthotopic transplantation model to study the response of each donor tumor to various chemotherapies (30). As described previously (10), the predominant histopathologic mammary tumor phenotype in K14cre;Brca2^F/F;p53^F/F mice is a carcinoma with well-defined tumor cell nests. These tumors express epithelial markers such as E-cadherin and are negative for vimentin, a fibroblast and mesenchymal cell marker (Fig. 1A, top). A second phenotype present in the group of 14 Brca2^Δ/Δ;p53^Δ/Δ (KB2P) mammary tumors used in this study, is a sarcomatoid tumor that has undergone a spindle cell metaplasia, characterized by bundles with elongated cells, absence of E-cadherin and expression of vimentin (Fig. 1A, bottom). This subtype is referred to as carcinosarcoma. In our tumor panel, we identified 10 carcinomas and four carcinosarcomas. In contrast with the clear histopathologic separation, the carcinosarcomas did not form a separate group at the genomic level when we tested aCGH data of a larger panel of Brca2^Δ/Δ;p53^Δ/Δ carcinomas and carcinosarcomas (28) by unsupervised hierarchical clustering (Supplementary Fig. S2). Thus, these tumors do not form a clearly separate subgroup within this model for BRCA2-mutated breast cancer at the DNA level.

Because of the spindle-shaped morphology of the carcinosarcomas, we suspected that these tumors have undergone EMT after the initial induction of the Brca2 and Trp53 mutations in an epithelial cell. To test this we, applied an EMT signature (5) to our panel of BRCA2;p53–deficient mouse mammary tumors. This signature has previously been shown to be associated with genes that are upregulated in gefitinib-resistant non–small cell lung cancers (5). In addition, the EMT index based on a subset of this signature, correlates with colorectal cancer subtypes (31). As expected, the two histologic subtypes were clearly separated in an unsupervised clustering analysis using this EMT signature (Fig. 1B). Compared with carcinomas, carcinosarcomas show higher expression of mesenchymal genes and lower expression of epithelial genes.

To investigate differential drug sensitivities, we tested the response of 10 KB2P carcinomas versus 4 KB2P carcinosarcomas to the MTD of the topoisomerase I inhibitor topotecan, the microtubule-stabilizing agent docetaxel, the topoisomerase II inhibitor doxorubicin or the cross-linking agent cisplatin. As shown in Fig. 2, most KB2P carcinomas responded to all drugs, even though they eventually acquired resistance to olaparib, topotecan, docetaxel, and doxorubicin. In contrast, the four carcinosarcomas did not respond well to olaparib, topotecan, docetaxel, and doxorubicin, but were still sensitive to cisplatin. In Fig. 1B the response to each drug is depicted per tumor and shows that a poor response is highly correlated with a mesenchymal gene-expression profile. Carcinomas that acquired drug resistance have all retained their epithelial state, as measured by histology and gene expression (see Supplementary Fig. S3, for the olaparib-resistant tumors).

In a mouse model for pancreatic ductal adenocarcinoma, the lack of response to gemcitabine was caused by a poor perfusion of the tumors (32). We therefore checked the presence of blood vessels in KB2P carcinomas and carcinosarcomas. Both subtypes showed blood vessels throughout the tumor (Fig. 3A). In the carcinomas, blood vessels are mainly present between the cell nests, whereas in carcinosarcomas blood vessels lay in between the tumor cells. The vessels are functional, as shown by the presence of i.v. injected, fluorescently labeled Tomato-Lectin (Fig. 3B), indicating that the drugs reach the tumor cells in both KB2P subtypes. These data are consistent with our observation that both carcinomas and carcinosarcomas respond to cisplatin.

Each drug for which we observed primary resistance in the carcinosarcomas, is known to be transported by drug efflux transporters: olaparib (30), docetaxel (33), and doxorubicin (34) by ABCB1 (also known as Pgp) and topotecan mainly by ABCG2 (also known as BCRP; ref. 35), whereas cisplatin has no strong affinity for any efflux transporter. This suggested to us that high expression of drug efflux transporters in KB2P carcinosarcomas could have contributed to their drug resistance phenotype. In the SAM analysis, expression of Abcb1b (which encodes Pgp together with Abcb1a) was indeed higher in treatment-naïve carcinosarcomas compared with the carcinomas (Supplementary Fig. S4). We tested expression of Abcb1a, Abcb1b, and Abcg2 in a semiquantitative manner by RT-MLPA (Fig. 4A). Abcb1a and Abcb1b were expressed at varying levels in the untreated carcinosarcomas, but three of four carcinosarcomas showed a higher expression than all carcinomas. All carcinosarcomas expressed increased levels of Abcg2.

To determine whether increased expression of Abcb1a and Abcb1b was causally related to the drug insensitivity, we tested the effect of the Pgp inhibitor tariquidar on therapy responses of carcinosarcomas derived from donors KB2P4 and KB2P6. Tumors were treated with tariquidar alone; olaparib, docetaxel, or doxorubicin alone; or the drug in combination with tariquidar. In addition, mice were treated with AZD2461, a novel PARP inhibitor with low affinity for Pgp (22). The effect of the combination therapy differed between the two donor tumors (Fig. 4B). KB2P4 tumors showed no effect of tariquidar on olaparib sensitivity, a small delay in outgrowth when docetaxel was combined with tariquidar, and a clear delay in tumor growth for doxorubicin plus tariquidar. All KB2P6 tumors responded well to the combination therapies of tariquidar with olaparib, docetaxel, or doxorubicin. Also, the response to AZD2461 was comparable with that of olaparib plus tariquidar. Taken together, these results show that Pgp contributed substantially to the low drug sensitivity of KB2P6 and, in the case of doxorubicin, of KB2P4.

In several in vitro studies, EMT has been linked to resistance to various classes of drugs (4). As we observed in our KB2P mouse model, a positive correlation between an “EMT-like” gene-expression pattern and expression of Abcb1a, Abcb1b, and Abcg2, we wondered whether this is also the case in other mouse mammary tumor models. To obtain a continuous value for EMT, we used an EMT score based on the EMT signature. The score is calculated by subtracting the average mean-centered log₂ expression of the epithelial genes from the average mean-centered log₂ expression of the mesenchymal genes. We used three different gene-expression datasets from mouse mammary tumors: one that was generated at the NKI and two publicly available datasets. The first one consists of 91 mammary tumors from K14cre or WAPcre-driven mouse mammary tumor models with conditional deletion of Trp53 alone (10) or in combination with Cdh1 (36, 37). Similar to the KB2P model, these models develop two main histopathologic tumor subtypes: carcinoma and carcinosarcoma. The EMT score of these tumors significantly correlated with Abcb1b and (to a lesser extend) with Abcb1a and Abcg2 (Fig. 5A). The second and third datasets are publicly available from Herschkowitz and colleagues (25) and Zhu and colleagues (26), and contain a collection of 13 and eight different genetically engineered mouse mammary tumor models, respectively. Although both datasets contain mostly tumors with a low EMT score, a positive correlation between EMT score and Abcb1b expression was still detected (Fig. 5B and C). The lack of correlation for Abcb1a and Abcg2 in these datasets is likely due to the low number of tumors with a high EMT score. Another possible explanation might be a poor sensitivity of the oligos on the expression arrays, as we have shown earlier that this affects the outcome (38).

Interestingly, in line with the KB2P data, the phenotype of p53^Δ/Δ tumors not only correlated with Abcb1b expression, but also with therapy response. Mice with a K14cre;p53^Δ/Δ (KP) carcinoma have an improved survival upon treatment with docetaxel or doxorubicin, compared with untreated control mice (Fig. 5D). KP carcinosarcomas, however, respond poorly to docetaxel and only moderately to doxorubicin. Mice with an Abcb1a/b^−/− KP tumor (KPA) showed a survival time in response to docetaxel or doxorubicin, which is in between that of the KP carcinomas and carcinosarcomas (Fig. 5D, docetaxel-treated sarcomatoid KP vs. KPA P = 0.022; doxorubicin-treated sarcomatoid KP vs. KPA P = 0.032). This indicates that also in BRCA2-proficient tumors, Pgp contributes to the poor response of carcinosarcomas. Ablation of Pgp does not fully sensitize these tumors to treatment, however, which is consistent with the tariquidar combination treatments in KB2P tumors shown in Fig. 4.

In this study, we investigated the role of EMT in anticancer drug sensitivity in the KB2P mouse model for BRCA2-deficient breast cancer. We found that a subset of the tumors has a mesenchymal, sarcomatoid phenotype, and gene-expression profile. These BRCA2-deficient carcinosarcomas do not respond to several DNA-damaging chemotherapeutics or the PARP inhibitor olaparib, and are therefore multidrug resistant. They are not pan- resistant, because they remain highly sensitive to cisplatin, due to the irreparable deletion inactivating the Brca2 gene, which compromises DNA repair by homologous recombination. We have previously shown that such a defect in homologous recombination by the irreparable inactivation of Brca1 cannot be overcome by any known mechanism of resistance to cisplatin (29). We show in our KB2P model and three other mouse mammary tumor datasets that an EMT-related transcriptional profile (indicated by a high EMT score) correlates with high expression of the Abcb1a and Abcb1b genes, which both encode the drug efflux transporter Pgp. Moreover, KB2P carcinosarcomas could be sensitized to olaparib, docetaxel, and doxorubicin by the Pgp inhibitor tariquidar. Taken together, these results indicate that EMT-associated multidrug resistance is in part driven by increased activity of drug efflux transporters. Similar to the KB2P tumors, p53^Δ/Δ carcinosarcomas with high Abcb1b expression responded poorly to docetaxel and doxorubicin, which could partially be reversed by removal of Pgp, indicating that the EMT-related resistance phenotype is independent of the BRCA2 status.

To date, only a few studies have investigated a link between EMT and drug transporter levels. Doxorubicin treatment can induce EMT in cultured breast cancer cells and upregulate efflux transporters, which is mediated by EMT transcription factors TWIST1 (39) and ZEB1 (40). Conversely, overexpression of SNAI1 in MCF7 cells results in increased Pgp levels after doxorubicin treatment (41), and in increased BCRP (ABCG2) levels (42). These studies also reported a positive correlation between SNAIL and Pgp (41), and between SNAIL and BCRP (42), respectively, in human breast cancer tissues.

In contrast with the strong evidence for a causal role of Pgp in primary and acquired resistance to chemotherapy and targeted agents in mice (20, 30, 43), the relevance of drug efflux transporters for therapy response in patients with breast cancer is still controversial. Increased Pgp mRNA or protein levels are in some, but not all, studies related to worse outcome (44). A complication in these studies is that Pgp (ABCB1) mRNA in tumor extracts may be derived from nontumor cells, such as macrophages, in the tumor microenvironment (43).

Over the last decades many clinical studies with transporter inhibitors have been conducted, with mostly negative results (34), and overall the impact of drug efflux transporters on patient outcome is likely to be small. This is not due to the inability of these transporters to cause resistance in real tumors. We have shown in the BRCA1-deficient mouse breast cancer model that modest levels of Pgp are sufficient to cause complete resistance to drugs used in the clinic, such as doxorubicin, docetaxel, topotecan, or olaparib (20, 30, 43). Obviously, transporter levels in human tumors are very low and transcriptional activation of the ABCB1 gene does not easily occur in these tumors (45). Indeed, the gene needs to be linked to a strong promoter by chromosomal rearrangements for a tumor to reach sufficient levels of Pgp to acquire drug resistance (46), and this may be a rare event.

We have considered the possibility that Pgp plays a role in a small subset of breast cancer, such as metaplastic breast cancer. However, we did not find any correlation with ABCB1 expression and a high EMT score in metaplastic tumors (Supplementary Fig. S5). This does not mean that Pgp could not play a role in some patients with acquired or secondary resistance. Unfortunately, matched samples of initially sensitive and subsequently drug-refractory tumors are hardly available from individual patients with breast cancer to address this issue.

Pgp contributes to multidrug resistance in KB2P carcinosarcomas, but our finding that KB2P4 is still insensitive to PARP inhibition and only modestly responsive to docetaxel and doxorubicin when Pgp is inhibited by tariquidar strongly suggest that other factors in the EMT program contribute as well. For example, the damage induced by PARPi and docetaxel in KB2P4 can still be compensated by another yet unknown mechanism, whereas doxorubicin-induced damage (47) is not compensated by EMT-related drug resistance in this tumor. Such EMT-associated factors could also contribute to the EMT-related drug resistance that is frequently observed in human cancer cells in vitro. In humans, metaplastic and claudin-low breast cancers are both associated with EMT and a triple-negative phenotype. Metaplastic breast cancers are often refractory to treatment and have a poor prognosis compared with other triple-negative breast cancers (48). The claudin-low subtype has a worse pCR rate than the basal-like group (8, 9). Several studies have demonstrated the predictive value of EMT markers (single or in combination) for prognosis and relapse-free survival (49–51), but which proteins in the EMT program eventually cause low drug sensitivity in general and to which drugs specifically still requires further investigation.

Even though several studies have shown that drug treatment, especially with doxorubicin, can induce EMT in vitro, we have not observed this phenomenon in any of our treated KB2P carcinomas. A possible explanation is that other resistance mechanisms are more easily activated, although it is not clear what these mechanisms might be, other than Pgp upregulation. Another option is that the carcinomas and carcinosarcomas arise from a different cell of origin, for example, a luminal or myoepithelial progenitor cell, respectively, in which K14 is expressed (10), and that they therefore do not easily switch from one type to another. This low plasticity is also illustrated by the small effect of Snai1 or Twist overexpression on the phenotype of KB2P tumor cell lines in vitro and in vivo (data not shown).

Our data show that sarcomatoid HR-deficient mammary tumors respond better to cisplatin than to the PARP inhibitor olaparib. This could potentially be clinically relevant, although two major differences between human and mouse tumors must be taken into account: (i) the controversial role of drug transporters in human cancers as described above and (ii) the large, irreversible deletion of exon 11 of Brca2 in this mouse model. Also for carcinomas, a side-by-side comparison of PARP inhibitors versus platinum drugs would be informative. In addition to BRCA1/2-mutated breast cancers also cancers with a BRCA-like CGH pattern would be useful to include. It has been shown that these are highly sensitive to platinum-based intensified chemotherapy (52) and it would be very interesting to investigate whether the same effect is achieved with PARP inhibition.

In summary, we show the usefulness of studying multidrug resistance in a realistic mouse model of BRCA2-deficient breast cancer. We found that enhanced expression of Pgp contributes to multidrug resistance associated with a sarcomatoid tumor phenotype. In addition, our data suggest that the correlation between a high EMT signature score and high expression of Pgp is a general phenomenon in mouse models of breast cancer. Pan-resistance is likely to be an accumulation of multiple mechanisms, and mouse models could be useful to unravel the different layers of resistance.

No potential conflicts of interest were disclosed.

Conception and design: J.E. Jaspers, J. Jonkers, S. Rottenberg

Development of methodology: J.E. Jaspers, S. Rottenberg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.E. Jaspers, C. Guyader, G. Xu, S. Rottenberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.E. Jaspers, A. Schlicker, L. Wessels, J. Jonkers, S. Rottenberg

Writing, review, and/or revision of the manuscript: J.E. Jaspers, A. Schlicker, C. Guyader, L. Wessels, P. Borst, J. Jonkers, S. Rottenberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.E. Jaspers, W. Sol, A. Kersbergen, S. Rottenberg

Study supervision: J. Jonkers, S. Rottenberg

The authors thank Arno Velds for analyzing the aCGH data and generating Supplementary Fig. S2 and Ewa Gogola and Xaoling Liu for assistance with genotyping and Southern blotting. The authors also thank Susan Bates from the NIH (Bethesda, MD) for providing tariquidar, and Mark O'Connor from AstraZeneca for providing olaparib and AZD2461.

This work was supported by grants from the Netherlands Organization for Scientific Research (NWO-Toptalent 021.002.104 to J.E. Jaspers; NWO-VIDI 91.711.302 to S. Rottenberg; and NWO-roadmap MCCA 184.032.303), Dutch Cancer Society (projects NKI 2007–3772, NKI 2009–4303, and NKI-2011–5220), the EU FP7 Project 260791-Eurocan-Platform, CTMM Breast Care, the Swiss National Science Foundation (grant 310030_156869), and the NKI-AVL Cancer Systems Biology Centre.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.