Complex Formation and Function of Estrogen Receptor α in Transcription Requires RIP140

RIP140, also known as nuclear receptor interacting protein 1 (Nrip1), is a transcriptional coregulator for a large number of transcription factors (reviewed in ref. 1). RIP140 is widely expressed, playing crucial roles in metabolic control in adipose tissues (2), skeletal (3), and cardiac muscle (4), as well as in liver (5). RIP140 is also involved in circadian rhythm regulation (6) and is essential for female fertility (6). Recently, we found RIP140 to be required for normal mammary gland development, where it functions as an essential component in estrogen signaling (7). RIP140 has a dual role as a transcriptional regulator, acting as both a corepressor for catabolic genes in metabolic tissue (3, 8) and a coactivator for inflammatory genes (9). RIP140 functions as a coactivator for ERα-responsive genes in mammary gland formation (7).

RIP140 was first identified as a cofactor for ERα in breast cancer cell lines (10, 11), but its putative function in ERα biology remained elusive. Circumstantial evidence suggests that RIP140 may have an important role in breast cancer, as RIP140 expression is significantly decreased in basal-like breast cancers as compared with luminal tumors and RIP140 has been implicated in controlling cell proliferation, potentially by regulating the activity of E2F transcription factors (12). In addition, RIP140 expression has also been found elevated in ductal carcinomas in situ (13). ERα regulates the expression of many components involved in ERα transcription complex formation, including GATA3 (14), XBP1 (15), FOXA1 (15, 16), and GREB1 (17). Analogous to these findings, estrogen treatment of breast cancer cells directly induces RIP140 expression (18).

ERα belongs to the superfamily of nuclear hormone receptors. When activated by estrogen, ERα drives the expression of a large number of target genes, leading to breast tumor cell proliferation. Ligand binding results in receptor dimerization, chromatin association, and the recruitment of a large set of coregulators required for transcription of target genes (19). For ERα to associate with the chromatin, it requires the action of so-called “pioneer factors” to enable chromatin accessibility. Putative pioneer factors for ERα include FOXA1 (16), PBX1 (20), and AP-2γ (21), all directly affecting ERα chromatin binding and functionality in breast cancer (reviewed in ref. 22).

ERα is expressed in 75% of all breast cancers, where it drives cell proliferation and tumor growth. Because of the action of ERα, most treatment options for luminal breast cancer revolve around blocking the activity of this key transcription factor. This is achieved either by competitive inhibitors for estrogen binding (such as tamoxifen) or by blocking estrogen synthesis (through aromatase inhibitors). Tamoxifen binds the same ligand-binding pocket on the receptor as estrogen does, but alters receptor conformation to block coactivator binding and consequently cell proliferation (23). Unfortunately, resistance to endocrine treatment is common. ERα can acquire additional agonistic features through altered kinase activities (24–26) or cofactor overexpression (27, 28).

In this study, we investigated the role of RIP140 in breast cancer and its functional relationship with ERα. RIP140 was shown to be a key component in the ERα transcription complex and required for estrogen-dependent transcriptional regulation and breast cancer cell proliferation. In addition, a downstream gene signature for RIP140-regulated genes enabled the identification of breast cancer patients with a poor outcome after tamoxifen treatment, illustrating the clinical implications of our findings.

MCF7 cells were cultured in DMEM supplemented with 10% FBS and 1% antibiotics. For hormone depletion, cells were grown in phenol red-free DMEM with 5% charcoal-treated FBS and 1% antibiotics for 3 days. For transient transfection of siRIP140, cells were transfected with Lipofectamine 2000 (Invitrogen) with 67 nmol/L siRIP140 or siControl (NBS Biological), and assayed 48 hours after transfections, unless stated differently. RIP140 overexpression experiments were performed using PEI (polyethylenimine; Polysciences, Inc.). For Western blotting, antibodies were used for Ki67 (m7240; DAKO), HSP90 (sc-7947; Santa Cruz Biotechnology), MCM4 (559544; BD Biosciences), and PLK1 (sc-17783; Santa Cruz Biotechnology). Cell lines were obtained from the ATCC, authenticated through STR profiling, and used at low passage after receipt from the vendor. Cells were cultured for less than 6 months after validation.

Chromatin immunoprecipitations (ChIP) were performed as described before (29). For each ChIP, 10 μg of antibody was used and 100 μL of Protein A magnetic beads (Invitrogen). The antibodies used were ERα (SC-543; Santa Cruz Biotechnology), RIP140 (6D7; ref. 5), total Histone 3 (generous gift from Fred van Leeuwen, the Netherlands Cancer Institute, Amsterdam, the Netherlands; ref. 30), H3K4me1 (ab8895; Abcam), H3K4me2 (07-030; Millipore), and H3K4me3 (ab8580; Abcam). For the sequential chromatin immunoprecipitation (re-ChIP) assay, chromatin was eluted with dithiothreitol (10 mmol/L), as described before (31). All quantitative PCR reactions were carried out in duplicates from two independent biologic replicates, averaged, and expressed relative to the input signal. Primer sequences are shown in Supplementary Table S1. Solexa sequencing and bioinformatics is fully described in the Supplementary Methods. ChIP-seq read count and number of aligned reads are shown in Supplementary Table S2.

Total RNA was extracted with TRIzol. RNA for microarray analysis was purified using QIAGEN RNeasy Mini columns. The expression of target genes was determined using SYBR Green Reagent and gene-specific primers (Supplementary Table S1). Relative expression levels were normalized to L13 ribosomal unit. Complete description of gene expression analyses is described in the Supplementary Methods.

MCF7 cells were seeded under hormone-depleted conditions and transfected with siRIP140 or siControl. The next day, cells were trypsinized, reseeded in a clear-bottom 384-well plate (IncuCyte) or 6-well plates (MTT assay), and hormones were added at the concentrations indicated. For MTT analyses, standard protocols were used (Millipore). For Incucyte analyses, cell proliferation was determined by plate confluency and measured in time using an Incucyte Life Cell Imaging Device (Essen Bioscience). Four wells were measured per condition.

All genomic data are deposited at ArrayExpress, with accession numbers E-MTAB-2576 (ChIP-seq) and E-MTAB-2577 (microarray data).

To identify the RIP140 genomic binding profile, we conducted ChIP-seq experiments using an antibody against endogenous RIP140 (Fig. 1). MCF7 cells were hormone-deprived for 3 days, after which the cells were treated for 3 hours with estradiol (E₂) or vehicle control, a time point previously shown to be optimal for cofactor binding (32), as compared with the earlier time point used for analysis of ERα binding (33). Importantly, chromatin binding of ERα and cofactors after 3 hours of E₂ treatment is still proximal to primary E₂-responsive genes, including GREB1, XBP1, and CCND1 (32). For each condition, two independent replicates were generated and only peaks found in both replicates were considered. As exemplified at the AREG locus, RIP140 chromatin binding was actively induced by estrogen treatment and overlapped with ERα-binding sites (Fig. 1A). Despite the fact that RIP140 is an ERα-responsive gene, it is unlikely that the increased RIP140 chromatin binding after E₂ exposure results from elevated RIP140 protein levels, as Western blot analysis showed that RIP140 levels were not elevated after 3 hours of E₂ treatment (Supplementary Fig. S1). In addition, exogenous overexpression of RIP140 (for Western Blot analysis, see Supplementary Fig. S1) did not increase RIP140 chromatin interactions, both under vehicle and E₂ conditions (Supplementary Fig. S2). In hormone-deprived conditions, 12,729 RIP140-binding events were found and following estrogen treatment this increased to 21,934 RIP140-binding sites, of which 17,985 sites (82%) are shared with ERα (Fig. 1B). The hormone-independent RIP140 chromatin-binding events are increased in intensity after hormonal treatment (Fig. 1C). In total, 19,210 (50%) ERα-binding sites are not cobound by RIP140 and this is not due to peak calling threshold issues (Fig. 1C, bottom). Note that peaks unique for RIP140 in E₂-treated cells (Fig. 1B) do show low ERα signal, indicating that false negative peaks are observed.

ERα is rarely found at promoters (∼5% of ERα binding occurs at promoters; ref. 16) and the vast majority of ERα chromatin interactions occurs at distal enhancers that are mainly found within a 20-kb window from the transcription start site of responding genes (34). RIP140 has a comparable genomic distribution as ERα (Fig. 1D); chromatin-binding events of RIP140 (regardless of whether they are cobound with ERα or not), are enriched at introns and distal intergenic regions, representing enhancer regions. Similar to ERα, approximately 5% of RIP140-binding events are found at promoters. To identify mechanisms of protein–DNA interactions, motif enrichment analyses were performed for the different categories of binding sites. The top 3 enriched motifs are shown for each subset, with their corresponding P values included (Fig. 1E). In addition, the top 20 enriched motifs are depicted in a heatmap visualization, ranked on z-score of the shared ERα/RIP140 sites (Fig. 1F; entire list see Supplementary Table S3). For all binding site subsets, ESR1 motifs were enriched. For the RIP140 unique binding sites, weak ERα signal is still observed (Fig. 1F), explaining the observed ESR1 motifs at these regions. As expected, forkhead motifs were found, representing binding sites for the ERα pioneer factor FOXA1 (16, 33, 35). FOXA1 motifs were not observed for regions only bound by RIP140. Instead, motifs underlying RIP140-binding events (either or not shared with ERα) were enriched for the alternative ERα pioneer factor AP-2 (21) and RARA. Cumulatively, most RIP140 chromatin-binding sites are affected by estrogen treatment and shared with ERα in breast cancer cells, being enriched for distinct transcription factor motifs at these sites.

We found a substantial overlap between RIP140 and ERα chromatin-binding patterns in breast cancer cells (Fig. 1). Next, we analyzed the binding relationship between ERα, RIP140, and other proteins known to be involved in the ERα transcription complex (Fig. 2). To this end, we compared ERα and RIP140 with binding profiles of the p160 family members SRC1, SRC2, and SRC3, p300 and CBP (32), as well as FOXA1 (33) and GATA3 (36), all of which were previously mapped in our laboratories under similar experimental conditions. As ERα is mainly found at enhancers, the enhancer-selective histone modification H3K4me1 (37) was also analyzed. For all ChIP conditions, cross-comparisons were generated, and the percentage of sites shared with any of the interaction partners was determined. The vast majority of ERα and RIP140-binding sites are shared with p300, CBP, and FOXA1. The overlap with the p160 proteins, GATA3 and H3K4me1 was significantly lower. Importantly, the shared chromatin-binding events between RIP140 and ERα are strongly associated with all factors, as compared with sites bound by RIP140 or ERα alone.

Next, we analyzed whether RIP140 and ERα are part of the same physical complex. Cells were hormone-deprived and treated with estrogen for 3 hours. ChIP was performed for ERα (Fig. 2B, top), after which a re-ChIP on the isolated chromatin was performed for RIP140. The reciprocal re-ChIP was also conducted (Fig. 2B, bottom), and for all conditions, IgG was included as a control (Fig. 2B). Quantitative PCR (qPCR) was performed for ten shared ERα/RIP140-binding sites adjacent to well-annotated estrogen-dependent genes, including RARA, GREB1, and PGR (primer sequences are provided in Supplementary Table S1). For these well-defined chromatin-binding sites, enrichment was found over IgG control, which was further enhanced by E₂ treatment, indicating that RIP140 and ERα co-occupy the DNA together as an E₂-induced complex.

As our findings suggest that RIP140 is an intrinsic part of the ERα transcriptional complex, we tested the effect of RIP140 knockdown on RNA Polymerase II recruitment (Fig. 2C). MCF-7 cells were transfected with siControl or siRNA targeting RIP140. The efficiency of the knockdown is shown on protein and mRNA levels (Supplementary Fig. S3). ChIP-qPCR primers were designed on the basis of chromatin-binding sites shared by ERα and RIP140: enhancer regions for RARA, PGR, XBP1, and GREB1, representing ERα-responsive genes. For all these sites, RIP140 knockdown resulted in a significant decrease of RNA Polymerase II recruitment in E₂-treated cells, indicating that RIP140 contributes to ERα activity at these sites (Fig. 2C). ChIP-qPCR for Histone 3 as well as epigenetically modified H3K4me1, H3K4me2, and H3K4me3 did not show significant alterations of these epigenetic histone marks after silencing of RIP140 (Fig. 2D).

Knockdown of RIP140 abrogated E₂-induced expression of a selection of typical ERα-responsive genes, including CCND1, AREG, and COX2 (Supplementary Fig. S3C). To assess the global role of RIP140 on E₂-induced gene expression, hormone-deprived MCF7 cells were transfected with siControl or siRIP140 and subsequently treated for 6 hours with E₂ (Fig. 3A). Six biologic replicates were generated for siControl and siRIP140 and 743 probes were differentially expressed when knocking down RIP140 (FDR < 0.01), which correspond to 726 unique transcripts. Of these, 506 transcripts were downregulated and 220 transcripts were found to be upregulated after knocking down RIP140. Ingenuity Pathway Analysis was performed, identifying β-estradiol (P = 4.65E−31) and ESR1 (P = 8.45E−22) as the top upstream regulators, where signaling cascades were predicted to be inhibited following RIP140 knockdown. The most enriched gene networks were centred around ESR1, where direct ERα-responsive genes were found to be affected. These include classic ERα-upregulated genes, including GREB1 (Fig. 3B), CCND1, MYC, and XBP1, all of which were inhibited when RIP140 was silenced (Supplementary Table S4). Quantitative real-time PCR (qRT)-PCR was used to validate the effects of siRIP140 on expression of a selection of ERα-upregulated genes: MYC, XBP1, GREB1, CCND1, FOS, RARA, AREG, and PgR (Fig. 3C). RIP140 was included as a control of the knockdown. The selected genes were co-occupied by both ERα and RIP140, as identified through re-ChIP experiments (Fig. 2B) and were all downregulated following siRIP140, indicating that RIP140 functions as a coactivator for these genes. HES1 and TFF1 (pS2) were not differentially regulated by siRIP140 in the microarray experiments and no differential mRNA expression was found by qPCR following RIP140 silencing, indicating that effects of RIP140 depletion are not fully generalizable on a genome-wide scale. ERα downregulates its own expression levels in MCF7 cells (32) and was upregulated after knocking down RIP140, indicating that RIP140 functions as a corepressor in this case (Supplementary Fig. S3C).

As the vast majority of RIP140 chromatin-binding events were shared with ERα binding, and RIP140 was required for RNA Polymerase II recruitment to a subset of ERα-responsive genes, we determined what proportion of RIP140-affected genes are normally estrogen regulated (Fig. 3D). For this, we used a publicly available dataset of all differentially expressed genes (P < 0.01) in hormone-deprived MCF7 cells following 6 hours of E₂ treatment (32), resulting in 1,812 downregulated genes and 1,821 upregulated genes in response to E₂. In total, 506 of these genes were downregulated after RIP140 knockdown, of which, 262 (52%) are E₂-upregulated genes. In contrast, of 220 genes upregulated after siRIP140, 147 (67%) are downregulated by E₂. Cumulatively, these data indicate that RIP140 enables ERα action both at genes that are upregulated as well as those that are downregulated following E₂ treatment.

RIP140, just like ERα, has a bivalent function and knocking down RIP140 results in both up- and downregulation of responsive genes. To elucidate potential transcriptional regulators that may underlie this divergent action of RIP140, RIP140 chromatin-binding sites were identified that were proximal (<20 kb from the transcription start site) to E₂-responsive genes that were RIP140-dependent. Genes up- and downregulated by E₂ treatment were analyzed separately, resulting in 382 (for E₂-upregulated genes) and 141 (for E₂-downregulated genes) RIP140 chromatin-binding sites (Fig. 3E). The RIP140 binding sites proximal to genes E₂-downregulated were most strongly enriched for ESR1 motifs, whereas an additional enrichment of AP-2 and SP1 motifs was found for the RIP140 sites proximal to E₂-upregulated genes.

In line with these results, ESR1 motifs at RIP140-binding sites were more frequently observed proximal to genes upregulated by RIP140 knockdown (Fig. 3F) that were downregulated after E₂ treatment (Fig. 3G). In contrast, genes with proximal ESR1/AP-2 motifs were mostly downregulated after RIP140 knockdown (Fig. 3F) and upregulated after E₂ treatment (Fig. 3G).

RIP140 plays an essential role in ERα complex formation and responsive gene expression. Next, we tested whether RIP140 expression is required for E₂-induced cell proliferation (Fig. 4A–C). Hormone-deprived MCF7 cells were transfected with siRIP140 or siControl, treated with E₂, and cellular confluence was assessed. Knocking down RIP140 decreased hormone-induced MCF7 cell proliferation (Fig. 4A). These data were validated using an independent assay (Fig. 4B), where siRIP140- or siCntrl-transfected cells were cultured for 1 week before analysis, in the presence of E₂, tamoxifen, fulvestrant, or vehicle control. A significant inhibition of cell proliferation was observed for vehicle and E₂-treated cells, again implicating the involvement of RIP140 in cell proliferation. Knockdown of RIP140 did not affect cell numbers under tamoxifen and fulvestrant conditions. As tamoxifen and fulvestrant block ERα-dependent cell proliferation, these data indicate that RIP140 is involved in cell proliferation rather than cell survival. These data were further confirmed by Western blot analysis of cell proliferation markers MCM4, PLK1, and Ki67/MIB1 (Fig. 4C), with HSP90 as loading control (38, 39). Cells were transfected with siCntrl or siRIP140 and hormone-deprived for 2 days. Subsequently, cells were treated with E₂ or vehicle control for 4 days, after which cells were lysed and processed. As expected, E₂ treatment increased levels of the cell proliferation markers in siCntrl-transfected cells, while this response to E₂ was not observed after siRIP140.

To test for any clinical implications of RIP140 action, genes under RIP140 control were used to generate a gene classifier that was tested in multiple cohorts of patients with breast cancer. As RIP140 is an intrinsic part of the ERα complex and dictates E₂ responsiveness, siRIP140-affected genes could also be directly affected under tamoxifen-resistant conditions, suggesting that differential expression of these direct target genes in breast tumors could function as a hallmark for tamoxifen unresponsiveness. All genes differentially expressed following siRIP140 (Fig. 3A) were tested for correlation with distant metastasis-free survival in a cohort of 263 tamoxifen-treated patients, diagnosed with primary ERα-positive breast cancer (40). Analyses of clinicopathologic parameters are provided in Supplementary Table S5. As visualized in a heatmap (Fig. 4D), unsupervised clustering on the basis of our gene signature identified two distinct subgroups of patients. Directionality of gene expression by siRIP140 in the MCF7 cell line (Fig. 3A) is highlighted in a red/blue color table on the left side of the heatmap. ERα-mediated (32) and RIP140-affected (Fig. 3A) gene expression involves both activation as well as inhibition of activity. In line with these data, no clear subgroups of directional gene expression were identified on the basis of unsupervised hierarchical clustering. Functionally, the two largest gene clusters were enriched for networks involving ERα (geneset A) and NFκB (geneset B) related networks (Supplementary Fig. S4).

Subsequently, distant metastasis-free survival (DMFS) after adjuvant tamoxifen treatment was determined (Fig. 4E), and significantly differential DMFS was found between the two largest patient subgroups [P = 0.028; HR = 1.84; 95% confidence interval (CI), 1.06–3.2]. Patients with a poor outcome (cluster 2) correlate with active RIP140-responsive genes (RIP-140 upregulated genes are increased, RIP140-downregulated genes are decreased), whereas good outcome (cluster 1) patients generally represent cases with a less active RIP140-responsive gene signature (RIP140-upregulated genes are decreased, RIP140-downregulated genes are increased; Fisher exact test, P = 0.001; Supplementary Fig. S5).

On the basis of the hierarchical clustering analyses, these patient groups could be further segregated into smaller subsections, further identifying subgroups of patients with differential DMFS (Supplementary Fig. S6). These data were validated using a second but smaller cohort of patients with tamoxifen-treated breast cancer (ref. 41; Supplementary Fig. S7), and classification of patients with tamoxifen-treated breast cancer over differential outcome was borderline significant (P = 0.056; HR = 1.73; 95% CI, 0.98–3.06). No correlation with survival was found for a cohort of 209 patients with ERα-positive breast cancer who did not receive any adjuvant endocrine treatment (ref. 42; P = 0.393; HR = 1.23; 95% CI, 0.77–1.96; Fig. 4F).

Cumulatively, we found that RIP140 is required for E₂-driven cell proliferation of the breast cancer cell line MCF7, and the RIP140-specific gene signature could selectively identify patients with tamoxifen-treated breast cancer with a poor outcome.

Even though RIP140 was originally identified in the breast cancer cell line MCF7, its biologic role in breast cancer and its effects on ERα function in this context has remained largely elusive. Here, we show that RIP140 plays a crucial role in ERα transcription complex formation, being essential for RNA Polymerase II recruitment and ERα-mediated breast cancer cell proliferation. Its physiologic role in ERα-mediated breast tumours is further strengthened by the identification of a RIP140-based gene expression classifier that could successfully stratify patients with breast cancer on survival after adjuvant tamoxifen treatment.

RIP140 is critically involved in ERα function, and we showed that RIP140/ERα complexes bind enhancer regions for the RIP140 gene itself. While most ERα genomics analyses have been performed following 45 minutes of E₂ treatment (33), we previously found coregulator/chromatin interactions, on a genome-wide scale, to be more robust after 3 hours of E₂ treatment (32). While E₂ treatment does increase RIP140 levels, no differential RIP140 protein levels were observed after 3 hours of E₂ exposure, nor did exogenous overexpression of RIP140 further increase chromatin interactions under vehicle or E₂ treatment. These data not only illustrate direct E₂ dependence of RIP140 chromatin interactions, but also suggest a higher mode of regulation where RIP140 action provides a positive feedback loop in an E₂-dependent fashion, further upregulating its own activity at later time point of hormone exposure.

RIP140 is an atypical transcriptional regulator and depending on the physiologic context (hormonal status, genomic environment, and transcription factor interaction repertoire) it can function either as a repressor or as an activator of transcription. In breast cancer, ERα is known to behave similarly, downregulating as many genes as it upregulates after 6 hours of hormonal treatment (32). This bivalent action of ERα is facilitated by RIP140, as the response to E₂ treatment (for both up and downregulated genes) was dampened after RIP140 knockdown. Recently, we found that RIP140 has a similar bivalent function in normal mammary gland development (7). Analogous to many other transcriptional coregulators, RIP140 is shared by multiple transcription factors, including ERβ (43) and E2F1 (12). RIP140 inhibits E2F1 function and is highly expressed in luminal breast cancer (12). As E2F1 is one of the major transcription factors mediating ERα action (44), the complexity of RIP140 functional behavior in breast cancer involves tight regulation and interplay between multiple transcription factors, further highlighting the complexity of the bivalent role of RIP140 in transcriptional regulation.

This bivalent and complex role of RIP140 in transcriptional regulation complicates interpretation of its role in clinical outcome. RIP140-based gene expression classifiers cannot focus on directionality of gene expression, but should rather emphasize “physiologic RIP140-mediated gene expression directionality”; a matter of attention also when generating expression signatures based of ERα action. In addition, even though ChIP-seq analyses do enable the identification of genesets directly responsive to ERα/RIP140 action, this complex is also critically involved in the expression of other transcription factors, including Myc (45). Analyzing merely directly responsive genes as determined through ChIP-seq represents an underestimation of the biological complexity of the system, and we therefore studied the full spectrum of genes differentially expressed after siRIP140. Yet, as the vast majority of RIP140 chromatin-binding sites were involved in ERα complexes, patient classification is likely biased towards ERα function.

RIP140 was required for ERα activity and E₂-driven cell proliferation, and we therefore tested a possible correlation of RIP140-action in ERα-dependent tumor cell proliferation. Being a transcriptional regulator in ERα-action, antiestrogen treatment outcome may be correlated with expression of these downstream genes. Using two distinct cohorts, the gene expression classifier successfully identified patients with breast cancer with a poor outcome after adjuvant tamoxifen treatment, whereas no correlation was observed in ERα-positive patients who did not receive adjuvant endocrine treatment. Future research using randomized clinical trial samples is needed to elucidate whether our classifier is merely a prognostic marker, or that it also has a possible predictive component.

Cumulatively, we illustrate that RIP140 in breast tumor cells, analogous to its role in mammary gland development (7), facilitates ERα activity and dictates the estrogen response in breast cancer cells. As such, RIP140 is a pivotal player in endocrine treatment response in breast cancer and provides a gene expression classifier for breast cancer patient survival.

No potential conflicts of interest were disclosed.

Conception and design: M. Rosell, J. Nautiyal, M.G. Parker, W. Zwart

Development of methodology: M. Rosell, A. Poliandri, W. Zwart

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Rosell, S. Stelloo, J. Nautiyal, J.S. Carroll, M.G. Parker, W. Zwart

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Rosell, E. Nevedomskaya, S. Stelloo, J.S. Carroll, M.G. Parker, W. Zwart

Writing, review, and/or revision of the manuscript: M. Rosell, S. Stelloo, L.F.A. Wessels, M.G. Parker, W. Zwart

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Stelloo, J.H. Steel

Study supervision: M. Rosell, L.F.A. Wessels, M.G. Parker, W. Zwart

The authors thank Gordon Brown and Suraj Menon from Cancer Research UK and University of Cambridge and Sander Canisius from NKI for bioinformatics support. The authors also thank James Hadfield and the entire genomic core facility at Cancer Research UK, University of Cambridge for help with Illumina sequencing.

This work was supported by grants from the Dutch Cancer Society and the Netherlands Organisation for Scientific Research (NWO).

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.