Candidate DNA Methylation Drivers of Acquired Cisplatin Resistance in Ovarian Cancer Identiﬁed by Methylome and Expression Proﬁling

Oncogene (2012) 31, 4567–4576 & 2012 Macmillan Publishers Limited All rights reserved 0950-9232/12 www.nature.com/onc ONCOGENOMICS Candidate DNA methylation drivers of acquired cisplatin resistance in ovarian cancer identiﬁed by methylome and expression proﬁling

C Zeller1, W Dai1, NL Steele2, A Siddiq3, AJ Walley3, CSM Wilhelm-Benartzi1, S Rizzo4, A van der Zee5, JA Plumb6 and R Brown1,4

1Epigenetics Unit, Department of Surgery and Cancer, Imperial College London, London, UK; 2Beatson West of Scotland Cancer Centre, Glasgow, UK; 3Department of Genomics of Common Disease, School of Public Health, Hammersmith Hospital, London, UK; 4Section of Medicine, Institute of Cancer Research, Sutton, UK; 5Department of Obstetrics and Gynecology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands and 6Cancer Research UK Beatson Laboratories, Institute of Cancer Sciences, University of Glasgow, Glasgow, UK

Multiple DNA methylation changes in the cancer Keywords: drug resistance; cisplatin; DNA methylation; methylome are associated with the acquisition of drug ovarian cancer; DNMTi; HDACi resistance; however it remains uncertain how many represent critical DNA methylation drivers of chemoresistance. Using isogenic, cisplatin-sensitive/resistant ovarian cancer cell lines and inducing resensitizaton with demethylating agents, we aimed to identify consistent Introduction methylation and expression changes associated with chemoresistance. Using genome-wide DNA methylation Ovarian cancer often presents at an advanced stage with profiling across 27 578 CpG sites, we identified loci at limited chances for curative treatment (Ozols, 2004). 4092 genes becoming hypermethylated in chemoresistant Common treatment strategies include debulking surgery A2780/cp70 compared with the parental-sensitive A2780 in combination with chemotherapy, which usually cell line. Hypermethylation at gene promoter regions is consists of a platinum-based compound, such as often associated with transcriptional silencing; however, cisplatin/carboplatin and a taxane, for example, pacli- expression of only 245 of these hypermethylated genes taxel. Despite responses to first-line chemotherapy in a becomes downregulated in A2780/cp70 as measured by high proportion of patients, many will relapse and microarray expression profiling. Treatment of A2780/ eventually develop resistance to currently available cp70 with the demethylating agent 2-deoxy-50-azacytidine treatment options, making the acquisition of clinical induces resensitization to cisplatin and re-expression of 41 drug resistance one of the major challenges in ovarian of the downregulated genes. A total of 13/41 genes were cancer therapy and a limiting factor in patient survival consistently hypermethylated in further independent (Cannistra, 2004). cisplatin-resistant A2780 cell derivatives. CpG sites at 9 Platinum drugs are DNA cross-linking agents exert- of the 13 genes (ARHGDIB, ARMCX2, COL1A, FLNA, ing their effect mainly via the formation of intra-strand FLNC, MEST, MLH1, NTS and PSMB9) acquired adducts between adjacent guanosines that interfere with methylation in ovarian tumours at relapse following transcription and replication, eventually leading to cell chemotherapy or chemoresistant cell lines derived at the death (Kartalou and Essigmann, 2001; Wang and time of patient relapse. Furthermore, 5/13 genes Lippard, 2005). Several mechanisms have been sug- (ARMCX2, COL1A1, MDK, MEST and MLH1) gested to participate in conferring platinum-resistant acquired methylation in drug-resistant ovarian cancer- properties to a tumour cell such as genetic alterations in sustaining (side population) cells. MLH1 has a direct role genes involved in DNA repair, drug uptake, apoptosis, cell cycle control and IGF signalling pathways (John- in conferring cisplatin sensitivity when reintroduced into cells ONCOGENOMICS in vitro. This combined genomics approach has identified stone et al., 2002; Luqmani, 2005; Edwards et al., 2008; further potential key drivers of chemoresistance whose Broxterman et al., 2009; Eckstein et al., 2009). More expression is silenced by DNA methylation that should be recently, it has been proposed that, in addition to further evaluated as clinical biomarkers of drug resistance. genetic changes, aberrant epigenetic marks can critically Oncogene (2012) 31, 4567–4576; doi:10.1038/onc.2011.611; contribute to the acquisition of drug resistance (Sharma published online 16 January 2012 et al., 2010). In cisplatin-resistant cancer cells, in particular, multiple DNA methylation changes at promoter CpG islands and associated transcriptional Correspondence: Professor R Brown, Epigenetics Unit, Department of gene silencing have been reported (Teodoridis et al., Surgery and Cancer, Imperial College London, Hammersmith 2005; Dai et al., 2008; Chang et al., 2010). For instance, Campus, IRDB 4th Floor, London W12 0NN, UK. E-mail: [email protected] methylation at MLH1, a mismatch repair gene, is Received 23 May 2011; revised 29 October 2011; accepted 28 November acquired in about 25–35% of ovarian cancer patients 2011; published online 16 January 2012 following platinum-based chemotherapy and has been Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4568 shown to be associated with poor patient survival line model that consists of sensitive and matched (Strathdee et al., 1999; Gifford et al., 2004). Impor- isogenic platinum-resistant lines derived from A2780 tantly, reversal of MLH1 epigenetic silencing by cells by repeated exposures to increasing levels of demethylation or re-expression of the gene was demon- cisplatin (Anthoney et al., 1996). We performed strated to resensitise tumour cells to subsequent genome-wide DNA methylation profiling of resistant chemotherapeutic treatment in vitro and in vivo (Plumb A2780/cp70, A2780/MCP1 and A2780/MCP6 com- et al., 2000; Papouli et al., 2004; Steele et al., 2009). pared with non-selected A2780p5 and A2780p6 clones MLH1 might, therefore, represent one of the key genes employing Infinium HumanMethylation27 BeadArrays driving chemoresistance in ovarian cancer cell lines. that comprise 27 578 CpG sites across more than 14 000 However, although the role of methylation changes at genes. In a first step, differentially methylated genes MLH1 has been well characterized, for the majority of were extracted as showing a significantly increased or aberrant DNA methylation events it is not particularly decreased difference of beta (7Db7) corresponding to a clear whether they are associated with response to false discovery rate (FDR) o0.05 estimated from bio- chemotherapy or are just occurring by chance due to a logical replicates within the study (7Db7B0.1) between methylator phenotype or simply as random methylation A2780/cp70 versus A2780 at X1 associated CpG site events (Issa, 2004) during platinum selection or DNA (see Materials and methods). Using these criteria, we damage induction. In analogy to the concept of identified multiple methylation changes between A2780/ ‘driver and passenger’ mutations emerging during cp70 versus A2780 (Figure 1): X1 CpG sites at 4092 carcinogenesis, methylation changes could either repre- genes were hypermethylated, whereas only 1289 genes sent ‘drivers’ of chemoresistance based on their potential became hypomethylated following exposure to cisplatin, to provide the cell with a selective advantage or suggesting that hypermethylation occurs more fre- ‘passenger’ events, with no substantial impact on quently than hypomethylation during the process of chemosensitivity (Greenman et al., 2007). selection for acquired cisplatin chemoresistance. We hypothesized that, in analogy to driver mutations, there might be a subset of epigenetic changes that are causally associated with the acquisition of chemoresis- Methylation changes in cisplatin-resistant lines associate tance. In order to identify the proportion of epigeneti- with gene expression changes in only a subset of genes cally altered genes driving platinum resistance in ovarian In a second step, we examined the expression profiles of cancer, we have analysed acquired DNA methylation sensitive A2780 and resistant A2780/cp70 cell clones changes in a human ovarian cancer cell line model of using Affymetrix HG-U133 Plus 2.0 Arrays and cisplatin resistance and expression changes associated identified differentially expressed genes by applying with acquired resistance or following resensitization rank products analysis (Breitling et al., 2004). Aberrant with demethylating agents. We then examined the DNA methylation at CpG island (CGI) has been shown generality of the changes observed in ovarian tumours to critically affect gene expression and is strongly at relapse following chemotherapy. associated with transcriptional repression (Jones and Baylin, 2002; Esteller, 2008). We assumed that expression changes associating with hypermethylation in chemoresistant A2780/cp70 as compared with A2780 Results would likely represent silencing events associated with the drug-resistant phenotype. We identified a total of Cisplatin selects preferentially for DNA hypermethylation 1370 genes showing significant gene expression changes, in A2780-chemoresistant cell lines with 687 genes going up and 683 genes going down in To identify DNA methylation changes associated with resistant versus sensitive cell lines. We further filtered for changes in gene expression and differential cisplatin those genes where hypermethylation associated with chemosensitivity, we used the human A2780 ovarian cell reduced expression in chemoresistant A2780/cp70 cells.

Figure 1 Hypermethylation is prevalent in cisplatin-resistant A2780 cell clones. Scatter plots of b-values show differentially methylated loci between A2780-sensitive (average of A2780p5/A2780p6) versus -resistant derivatives A2780/cp70, A2780/MCP1 and A2780/MCP6 (yellow) as measured by Inﬁnium HumanMethylation27 BeadArrays. The differential methylation cutoff (red and green solid line) was estimated from the difference between biological replicates of cell lines PEO1 and PEO4 (blue) by controlling FDRo0.05. A full colour version of this ﬁgure is available at the Oncogene journal online.

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4569 Of the 4092 genes hypermethylated in A2780/cp70, 3823 within the list of 45 genes, suggesting that addition of (93%) genes are present on the HG-U133 Plus 2.0 the HDACi does not add substantially to the number of Array. Combined analysis revealed that only a small genes reactivated by DAC alone. We further compared proportion of changes in methylation correlated with DNA methylation profiles of A2780/cp70 cells following expression changes, with 245 genes becoming hyper- drug treatment with untreated A2780/cp70. Using a methylated and downregulated following selection for differential methylation cutoff of |Db|X0.1, we validated cisplatin resistance in A2780/cp70. that all of the 41 DAC-responsive genes became demethylated at X1 associated CpG site following Only a small proportion of hypermethylated genes that treatment with DAC. This substantiates that re-expres- are downregulated become re-expressed following sion of the 41 DAC-induced genes was truly mediated by treatment with decitabine DNA demethylation (Supplementary Figure S2). In order to identify DNA methylation changes associating with changes in gene expression and differential A small set of 13 genes are strong candidates as cisplatin chemosensitivity in this cell line model, we epigenetically silenced cisplatin-resistant drivers combined our data with expression profiles of A2780/ In order to identify drug resistance-associated methyla- cp70 cells treated with either the DNA methyltransfer- tion changes that were independent of clonal effects, we ase inhibitor (DNMTi) 2-deoxy-50-azacytidine (decita- further filtered for genes showing consistent hyper- bine, DAC), the histone deacetylase inhibitor (HDACi) methylation of reinduced genes in independently Belinostat or a combination of both at conditions and selected cisplatin-resistant A2780 derivatives. Out of time point known to induce chemosensitization in this the 4092 hypermethylated genes initially identified in cell line (Steele et al., 2009). We reasoned that genes A2780/cp70, 1824 genes were also hypermethylated in affecting drug sensitivity via hypermethylation-asso- the two independent cisplatin-resistant derivatives, ciated silencing in chemoresistant cells would become A2780/MCP1 and A2780/MCP6 (Figure 3). switched backed on following pharmacological inter- Among those 1824 hypermethylated genes, a small vention of methylation and/or deacetylation. DNMTi proportion of only 13 genes (ARHGDIB, PSMB9, treatment can function synergistically with an HDACi HSPA1A, ARMCX2, MEST, FLNC, MLH1, MDK, in restoring gene expression (Cameron et al., 1999), and GLUL, FLNA, NTS, COL1A1 and NEFL) became sequential exposure of cells to DNMTi and HDACi re-expressed in A2780/cp70 following treatment with might, therefore, lead to re-expression of additional DAC (Table 1, Figures 3 and 4a). Epigenetic silencing of genes. MLH1 has previously been shown to be a clinically Treatment with DAC resulted in re-expression of 41 relevant mechanism of acquired cisplatin resistance in out of the 245 genes that were hypermethylated in ovarian cancer (Gifford et al., 2004), and re-expression addition to being downregulated in A2780/cp70 cells of MLH1 confers increased cisplatin chemosensitivity in (Figure 2, Supplementary Table S1). A total of 45 genes xenograft models (Plumb et al., 2000). MLH1 was became switched back on following combined treatment identified among the list of 13 genes supporting our with DAC and the HDACi Belinostat. Exposure to approach and substantiating the notion that aberrant Belinostat alone reinduced the expression of only 10 methylation of a small set of genes can putatively drive genes that were downregulated in addition to being the acquisition of cisplatin resistance. hypermethylated in untreated A2780/cp70 cells. Com- Combined DAC and Belinostat treatment did not parison of the gene lists showed a big overlap between markedly increase the number of re-expressed genes, the 41 (DAC) and 45 (DAC þ Belinostat) with 14 commonly hypermethylated genes becoming re-expressed genes, with 40/41 genes being contained switched back on in the resistant cell line (ARHGDIB,

Figure 2 Re-expression of genes in chemoresistant A2780/cp70 cells following treatment with epigenetic remodelling agents. In all, 41 genes (DAC), 45 genes (DAC þ Belinostat) or 10 genes (Belinostat) become reactivated following treatment respectively. Venn diagrams show the overlap between number of genes being hypermethylated in untreated A2780/cp70 (green circle), being downregulated in untreated A2780/cp70 (pink circle) and genes with increased expression following drug exposure in A2780/cp70 as compared with A2780 cells (blue circle). A full colour version of this ﬁgure is available at the Oncogene journal online.

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4570

Figure 3 A small percentage of consistently hypermethylated genes is affected following resensitization with epigenetic remodelling agents. Out of 1824 genes hypermethylated in independent cisplatin-resistant A2780 derivatives, only 13 genes become re-expressed following DAC treatment, 14 genes become reactivated following combined DAC and Belinostat treatment and only 1 gene becomes re-expressed following Belinostat treatment. Venn diagrams show the overlap between genes being hypermethylated in chemoresistant A2780/cp70, A2780/MCP1 and A2780/MCP6 (green circle), being downregulated in untreated A2780/cp70 (pink circle) and genes with increased expression following drug exposure in A2780/cp70 as compared with A2780 cells (blue circle). A full colour version of this ﬁgure is available at the Oncogene journal online.

Table 1 DAC-induced genes being hypermethylated and downregulated in A2780/cp70 Gene symbol Gene name Location Biological Role DM CGI DM non-CGI

ARHGDIB Rho GDP dissociation inhibitor-b 12p12.3 GDP/GTP exchange Na Y ARMCX2 Armadillo repeat containing, X-linked 2 Xq21.3-q22.2 Development, tissue integrity Na Y COL1A1 Collagen, type I, alpha 1 17q21.3 Extracellular matrix component Y N FLNA Filamin A, alpha Xq28 Cytoskeleton reorganization, motility YN and migration, signal transduction FLNC Filamin C, gamma 7q32-q35 Actin cytoskeleton reorganization Y Y GLUL Glutamate-ammonia ligase 1q31 Glutamine synthesis Y Nb HSPA1A Heat shock 70 kDa protein 1A 6p21.3 Protein binding, stress response Y N MDK Midkine 11p11.2 Growth Y Y MEST Mesoderm-speciﬁc transcript homologue (mouse) 7q32 Development Y Y MLH1 MutL homologue 1, colon cancer, 3p22.3 DNA mismatch repair Y Y non-polyposis type 2 (Escherichia coli) NEFL Neuroﬁlament light polypeptide 8p21.2 Axoskeleton Nc Y NTS Neurotensin 12q21 Neuromodulator, growth Na Y PSMB9 Proteasome subunit (prosome, macropain), 6p21.3 Antigen processing N Y beta type, 9

Abbreviations: CGI, CpG island; DAC, 2-deoxy-50-azacytidine, decitabine; DM, differentially methylated; GDP, guanosine diphosphate; GTP, guanosine triphosphate; N, no; Y, Yes. aNo CGI present. bNo probe outside of CGI present. cNo probe in CGI present.

PSMB9, HSPA1A, ARMCX2, MEST, FLNC, MLH1, We further validated our array-based ﬁndings in the MDK, GLUL, NTS, COL1A1, NEFL, SERPINB2 and A2780 cell line model by using bisulphite pyrosequen- HIST1H2BF). Again, we observed a strong overlap cing and quantitative real-time PCR. We arbitrarily between genes being reinduced following DAC and chose the three genes MEST, FLNC and ARHGDIB combined DAC/Belinostat treatment. Only SERPINB2 that were commonly hypermethylated in independent and HIST1H2BF were exclusively DAC/Belinostat cisplatin-resistant cell lines (Figure 4) and became re- responsive, suggesting that synergistic demethylation expressed following DAC as well as combined DAC/ and acetylation are necessary for their reactivation. In Belinostat treatment. FLNC and MEST are both contrast, only one gene (GLUL) was being reinduced associated with CGIs, whereas ARHGDIB does not following Belinostat treatment alone in addition to contain a CGI. Acquired hypermethylation was con- being consistently hypermethylated across resistant ﬁrmed for MEST, FLNC and ARHGDIB in resistant A2780 cell lines. Interestingly, GLUL was also reacti- versus sensitive A2780 cell lines (Figure 4b). The vated following DAC and combined DAC/Belinostat observed differences in CpG methylation between treatment. However, the effect of Belinostat alone on sensitive and resistant lines at 39%, 42% and 27% GLUL expression was very subtle with a low fold corresponded to a Db of 0.38, 0.43 and 0.23 for MEST, change, suggesting that DAC-induced demethylation is FLNC and ARHGDIB, respectively (Figure 4). Hyper- still required for full restoration of gene activity. methylation of the three genes was associated with

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4571

MEST FLNC ARHGDIB 100 100 100

80 80 80

60 60 60

40 40 40

20 20 20 % CG methylation % CG methylation % CG methylation 0 0 0 A2780 A2780 A2780 A2780 A2780 A2780 sens res sens res sens res

MEST FLNC ARHGDIB 0 -100 -200 -300 -400 -500 -600

vs. A2780 -700 -800 -900 -1000 Fold reduction in A2780/cp70 -1100 Figure 4 Methylation and expression analysis of genes getting re-expressed following DAC treatment in A2780-resistant cell lines. (a) Heatmap of the 13 DAC-regulated genes showing increased methylation in resistant A2780/cp70, A2780/MCP1 and A2780/MCP6 cell clones as compared with A2780-sensitive lines (7Db7X0.1). (b) Validation of hypermethylation of DAC-regulated candidate genes in A2780-resistant cell lines by bisulphite pyrosequencing. Average methylation values across A2780-sensitive and -resistant cell lines are shown for three genes. Pyrosequencing assays show average methylation values across two CpG sites for MEST (CpG site no. 1 relates to Illumina ID cg08077673), seven CpG sites for FLNC (CpG site no. 6 relates to Illumina ID cg02661879) and three CpG sites for ARHGDIB (CpG site no. 2 relates to Illumina ID cg10925082), covering the differentially methylated CpG site shown in methylation profile (a). (c) Validation of downregulation of DAC-regulated candidate genes in A2780-resistant cell lines by quantitative real-time PCR. The fold reduction in A2780/cp70 cells versus A2780-sensitive cells is shown for three genes. A full colour version of this figure is available at the Oncogene journal online. downregulation in A2780/cp70-resistant cells as com- of the tumour with inherently drug-resistant CSCs pared with A2780 (Figures 4b and c). We also tested following chemotherapeutic treatment. We examined mRNA expression of the three genes in independent whether candidate drug-resistant drivers showed altered cisplatin-resistant A2780/MCP1 and A2780/MCP6 and methylation in CSCs obtained from an ovarian cancer found downregulation to be associated with hyper- cell line by comparing the b-values of the set of 13 genes methylation in these cell lines (data not shown). in independently sorted side populations (SPs) of IGROV1 cells (n ¼ 2) versus non-SPs. We have pre- Candidate drug-resistant genes commonly acquire viously shown that IGROV1 SPs have tumour stem cell- methylation in independent in vivo derived ovarian like properties, including the ability to form tumours in chemoresistant cells and in tumours at relapse NOD/SCID mice at low cell number, enhanced ability In order to address how the small set of epigenetically to grow as spheroids, ability to repopulate SP and non- inactivated candidate resistance drivers impacts on SP cells and expression patterns with significant enrich- clinical drug resistance, we evaluated their methylation ment for known stem cell markers (Rizzo et al., 2011). status in three independent pairs of in vivo derived We have also previously shown that IGROV1 SP are chemonaive and chemoresistant ovarian tumour cell more resistant to cisplatin and carboplatin than non-SP lines derived from patients before chemotherapy and at IGROV1 cells (Rizzo et al., 2011). Although we did the time of developing clinical resistance (Langdon attempt to isolate SPs from A2780, the numbers of SP et al., 1988). Using a difference of 7Db7X0.1 we cells present were too low to allow detailed phenotypic observed acquired methylation in 6 out of the 13 genes and molecular analysis. (ARHGDIB, ARMCX2, COL1A1, FLNA, MEST and Applying a 7Db7X0.1 we observed five genes as MLH1) in chemoresistant, post-chemotherapy cell lines having higher methylation in IGROV1 CSC populations (Table 2). (ARMCX2, COL1A1, MEST, MDK and MLH1; One hypothesis of how drug resistance evolves implies Table 2). Notably, four of these genes had acquired the expansion of the so-called tumour-initiating or methylation in in vivo derived cisplatin-resistant cell lines cancer-sustaining cells (CSCs; Rajasekhar et al., 2008). (ARMCX2, COL1A1, MEST and MLH1), supporting This population of cells is thought to share many enrichment of methylation at these loci in chemoresis- characteristics with normal stem cells including the tant cell populations. capacity to self-renew and to survive drug treatment We further evaluated the methylation status of the 13 (Dean et al., 2005). According to this paradigm, the candidate genes in clinical specimens using seven pairs acquisition of drug resistance results from repopulation of matched primary ovarian tumours and tumours at

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4572 Table 2 Acquired methylation (7Db7X0.1) of candidate drug-resistant drivers in drug-resistant in vivo derived ovarian cancer cell lines, SPs and tumours at relapse

Abbreviation: SP, side population. Values show 7Db7X0.1 between matched resistant and sensitive pairs of in vivo derived cell lines, independently sorted SPs versus non-SPs of IGROV1 cells and tumours at relapse versus matched primary tumours at presentation (seven pairs). Empty boxes indicate values of 0o7Db7 o0.1. All CpG sites being differentially methylated between A2780-sensitive and -resistant cell lines are depicted.

the time of relapse. Methylation of 8/13 genes was in the platinum-responsive phenotype. However, our acquired in tumours following platinum-based che- knowledge of whether these DNA methylation changes motherapy (Table 2). Three of these genes (ARMCX2, and aberrantly inactivated genes are actually driving MEST and MLH1) were found to have higher chemoresistance is limited. By integrating drug induced methylation in chemoresistant in vivo derived cell lines, resensitization, methylation and expression proﬁling in CSC populations and tumours at relapse, supporting the an isogenic, cisplatin-resistant ovarian cancer cell line clinical relevance of changes at these loci. model, we present evidence that the proportion of key DNA methylation and linked expression changes associating with response to chemotherapy is surpris- ingly small. Our results indicate that o1% of hyper- Discussion methylated genes in independent platinum-resistant ovarian cancer cell lines account for the acquisition of Substantial DNA methylation changes have previously platinum resistance. Indeed, by comparing methylation been reported to occur during the acquisition of changes in multiple independent cisplatin-resistant platinum resistance. Among those changes, hypermethy- derivatives of A2780, we identify as few as 13 key genes lation at CGI promoters and associated epigenetic that acquire methylation at associated CpG sites. These silencing are prevalent in various cell line models (Dai are strong candidates for drivers of resistance in this et al., 2008; Li et al., 2009; Chang et al., 2010) and are genetic background, although other genes may be more thought to, either alone or in combination with genetic important in other ovarian tumour genetic back- changes, account for the loss of expression of key genes grounds. However, we show that a high proportion of

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4573 the loci identified acquire methylation in ovarian that were hypermethylated in chemoresistant cell lines. tumours at relapse (8/13) and are differentially methy- One explanation for this discrepancy could be the lated in drug-resistant ovarian tumour sustaining (stem) method that we used for identification of differentially cells (5/13). Therefore, methylation at CpG sites at these expressed genes. We applied the rank products method genes are key candidate epigenetic drivers of acquired based on our aim to detect differences that consistently drug resistance independent of genetic background. rank highly across any number of replicate experiments Few previous studies established hypermethylation as and that are most likely to have biological significance. a marker of platinum-resistant ovarian cancer cell lines This form of analysis avoids identifying a fold change on an epigenome-wide basis (Dai et al., 2008; Li et al., that exceeds an arbitrary threshold. Conversely, this 2009). Although the mechanism of how drug-resistant method can identify subtle changes and these subtle cells accumulate methylation is unresolved, there are changes may be missed by other methods. clear examples showing an association between epige- Our subset of candidate drug-resistant drivers was netic silencing of specific genes and drug response. For identified using an isogenic cell line model. Although example, acquired MLH1 CGI hypermethylation and this model allows identifying changes ultimately linked associated gene inactivation are observed following to cisplatin exposure, detection of genes in identical platinum-based chemotherapy in ovarian cancer and is rather than in diverse genetic settings, as well as issues associated with poor clinical outcome (Gifford et al., surrounding in vitro selection, may be a limitation to the 2004). Re-expression of MLH1 in chemoresistant cell value of this gene set. However, we observed increased lines has been shown to partially restore their sensitivity methylation of 6 out of the 13 candidate genes in at least to subsequent cisplatin therapy (Plumb et al., 2000; one of the three cisplatin-resistant in vivo derived cell Papouli et al., 2004). Notably, our list of methylation- lines. We further observed acquired methylation of five dependent genes included MLH1, substantiating the out of these six genes in tumours at relapse. Although hypothesis that we have identified a specific subset of our set of matched tumour pairs before and after genes that drive the acquisition of drug resistance. It has chemotherapy was small due to the clinical challenges in to be noted, however, that full resensitization of cells to obtaining such matched samples, this nevertheless platinum-based drugs could not be achieved via MLH1 suggests that the acquisition of methylation at these re-expression alone in cell line models, in contrast to genes may be a common event occurring during reversal of methylation using demethylating agents that acquired cisplatin resistance and may be clinically fully restore chemosensitivity and induce re-expression relevant. Notably, the low frequency at which epigenetic of multiple genes (Plumb et al., 2000; Papouli et al., changes are acquired at candidate drug-resistant drivers 2004). These observations support the idea that the shows analogies to driver mutations that often present induction of additional repressed genes to MLH1 is vital at a low frequency (Wood et al., 2007) and might, for full restoration of drug sensitivity in these models. therefore, be missed within a small set of tumours. However, this also illustrates the challenges in demon- Interestingly, we also observed increased methylation at strating direct effects of a gene on chemosensitivity three (ARMCX2, MEST and MLH1) out of these five in vitro based on gene reintroduction, where the effects commonly hypermethylated candidate genes in drug- may be small or require inactivation of multiple genes. resistant sustaining (SP) cells isolated from the human Our results indicate that several methylation changes are ovarian tumour cell line IGROV1 as compared with the directly associated with chemosensitization induced by bulk of the tumour cells. Our observation could indicate demethylating agents combined with HDAC inhibitors. that hypermethylation and epigenetic silencing of this Similar observations have been reported in other subset of three genes are already present in stem cells. As tumour types (Chang et al., 2010); however, our data SPs are believed to contain the cells responsible for also reveal that epigenetic alterations associated with maintenance of long-term growth of ovarian cancer sensitivity to cisplatin occur at a few selected genes (Bapat et al., 2005; Szotek et al., 2006), regrowth of rather than at large numbers of loci in this cell line these cells could be contributing to manifesting a drug- model (Glasspool et al., 2006). resistant phenotype (Agarwal and Kaye, 2003). The group of epigenetically inactivated genes was Silencing of MLH1 leads to the loss of DNA upregulated by DAC and also the combined DAC and mismatch repair, which in turn has been suggested to Belinostat treatment, but remained unaffected by lead to platinum DNA damage tolerance due to Belinostat alone, with the exception of GLUL. Con- translesion synthesis, reduced replication stalling or sistent with previous reports, the inhibition of HDAC reduced signalling of cell death pathways (Stojic et al., did not add extensively to the number of re-expressed 2004; O’Brien and Brown, 2006; Yoshioka et al., 2006). genes, with only two additional genes (SERPINB2 and However, little is known about the function in drug HIST1H2BF) being reinduced (Cameron et al., 1999; resistance of the other genes found to commonly acquire Suzuki et al., 2002). Addition of an HDACi can methylation in tumours and cell lines at relapse and in markedly increase the level of gene re-expression such drug-resistant SP populations. ARMCX2 might have a as MLH1 in vitro and in vivo (Steele et al., 2009), but role in tumour suppression based on the presence of an may not convert further genes to re-express. Although armadillo repeat motif, which is found in other proteins other studies reported lack of methylation of genes fulfilling functions in cell proliferation, migration, responsive to HDACi and, partially, to DNMTi (Suzuki maintenance of tissue integrity and tumourigenesis, et al., 2002), our analysis specifically extracted genes and has been involved in development (Smith et al.,

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4574 2005). MEST, a maternally imprinted gene, has been b ¼ max(M, 0)/(max(M,0) þ max(U,0) þ 100) and reflects the implicated in embryonic growth and maternal behaviour methylation status of a specific CpG site. Subsequent analysis (Lefebvre et al., 1998), and loss of MEST imprinting has was carried out in R (version 2.10.1), http://www.r-project.org/. been reported in breast and lung cancers (Pedersen et al., The reproducibility of the BeadChips was evaluated using 2002; Nakanishi et al., 2004). Interestingly, it has biological (independent bisulphite modifications of two recently been shown that MEST is a negative regulator independently prepared DNAs) and technical replicates (bisulphite conversion of identical DNA) of matched chemo- of the Wnt pathway and that MEST knockdown might sensitive/chemoresistant ovarian cancer cell lines PEO1/PEO4. activate Wnt signalling (Jung et al., 2011). Epigenetic b-Values of biological and technical replicates of PEO1 and regulation of Wnt pathway genes has previously been PEO4 were highly correlated with r2 values 40.99 (Supple- shown to be associated with patient survival following mentary Figure S1). platinum-based chemotherapy and chemoresistance in Gene annotation was carried out using the Mar. 2006 (NCBI/ ovarian cancer (Dai et al., 2011; Peng et al., 2011). 36/hg18) assembly at the UCSC database (http://genome.ucs- Taken together, we have identified a key subset of c.edu/). Individual CpG sites located within a CGI as defined by genes potentially driving acquired drug resistance in Gardiner-Garden and Frommer (1987) were linked to an ovarian cancer from the large number of epigenetic associated gene if the CGI was within 2 kb distance from the changes occurring following chemotherapy. As well as respective transcription start site (Saxonov et al.,2006).Non- CGI-related CpG sites were linked to a gene if they were located providing novel insight into mechanisms of drug within 2 kb from the transcription start site. resistance, this has identified candidate biomarkers for For pairwise differential methylation analysis, the difference further evaluation in future clinical studies, including of two biological replicates of the ovarian cancer cell lines potential stratification biomarkers in clinical trials of PEO1 and PEO4 was calculated as the expected difference epigenetic therapies that reverse the acquired resistance (null distribution). The observed difference was calculated phenotype. from the paired samples. Using bootstrap resampling method, we extracted the same number of CpG sites (n ¼ 27 578) from the null distribution 200 times. Each time, a FDR was calculated using the formula below. The median of FDRs Materials and methods was used as an estimation of the probability that CpG sites with differential methylation above the selected cutoff were Cell lines identified by chance. This median FDR was set at 0.05 In vitro derived A2780p5, A2780p6, A2780/cp70, A2780/ corresponding to 7Db7X0.1. MCP1 and A2780/MCP6 cell lines (Anthoney et al., 1996) were obtained from Dr Robert F Ozols (Fox Chase Cancer # CpG sites over cutoff in null data Centre, Philadelphia, PA, USA). PEO1, PEO4, PEO14, FDR ¼ PEO23, PEA1 and PEA2 in vivo derived cell lines (Langdon # CpG sites over cutoff in real data et al., 1988) were obtained from Professor Hani Gabra (Imperial College). Isolation of SPs and non-SPs of IGROV1 In order to filter for differentially methylated CpG sites in the cells was performed as described previously (Rizzo et al., A2780 cell line model mean b-values across sensitive cell clones 2011). The identity of the cell lines was verified by DNA were used. genotyping. All cell lines were grown in RPMI-1640 supple- mented with glutamine (2 mM) and 10% fetal bovine serum. Bisulphite modification and pyrosequencing A total amount of 1 mg of genomic DNA was bisulphite modified using the EpiTect Bisulphite Kit (Qiagen, West Tumour samples Sussex, UK) according to the manufacturer’s instructions. Seven pairs of matched serous epithelial ovarian tumours Pyrosequencing primer sets covering differentially methylated before chemotherapy and at relapse were obtained from the CpG sites in the ARHGDIB, FLNC and MEST gene promoters University Medical Center Groningen. Appropriate ethical were ARHGDIB_PYRO_F: 50-biotin-TGGGAATAGAAGTG approval was obtained for all the collected samples. AGTGGTATAA-30, ARHGDIB_PYRO_R: 50-CCTATTCCTT TACACTACCTATCT-30, ARHGDIB_PYRO_S: 50-CAACAT Methylation profiling TCTTATCAATTAATAACAC-30, FLNC_PYRO_F: 50-TGGA All array-based methylation profiling was performed using the GGGAGAGAGAGTTAG-30, FLNC_PYRO_R: 50-biotin-CTT Infinium HumanMethylation27 BeadChip (Illumina, San ACCCACCCACTTAAAATACTCATTAC-30, FLNC_PYRO_S: Diego, CA, USA), except for DNA methylation profiling 50- AGAAGTTGGAGAGGA-30, MEST_PYRO_F: 50-biotin- following treatment with epigenetic remodelling agents, which GTGGGTTATATTAGTTTTAGGGGTAG-30, MEST_- was performed using Infinium HumanMethylation450 Bead- PYRO_R: 50-CCTTTCCAACCTCCAAAACTAACTAT-30 Chips (Illumina). A total amount of 1 mg genomic DNA was and MEST_PYRO_S: 50-AAATTATATAACTTTTATATTC bisulphite modified using the EZ DNA Methylation Kit TC-30. Pyrosequencing PCR was performed in duplicate for (Zymo Research, Irvine, CA, USA). In all, 200 ng of converted each sample in a 25 ml volume containing 0.2 ml Faststart Taq DNA was further processed to run BeadArrays according to polymerase (Roche, Welwyn Garden City, UK), 2.5 ml the manufacturer’s instructions. Each locus is represented by Faststart Buffer (Roche), 0.2 mM dNTPs (Applied Biosystems, fluorescent signals from two bead types corresponding to the Warrington, UK), 75 ng primers (each) and 1 ml of modified methylated (M) and unmethylated (U) alleles, respectively. DNA template using the following conditions: 95 1C for 6 min, The raw signals of unmethylated and methylated bead types 35 cycles of 95 1C for 30 s, 60 1C for 30 s, 72 1C for 30 s, were background corrected and computed into a b-value using followed by 72 1C for 5 min. Pyrosequencing of PCR products the BeadStudio software Methylation Module (version 1.0.5; was performed using the PyroGold Reagent Kit (Qiagen) Illumina). The b-value represents the ratio of the intensity of according to the manufacturer’s instructions. The methylation the methylated bead type to the combined locus intensity: percentage of CpG sites for individual genes was calculated by

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4575 using the Pyro Q-CpG software (version 1.0.9), Biotage Paisley, UK) and subsequently used as a template in (Uppsala, Sweden), and then averaged across sensitive and quantitative real-time experiments to amplify products for resistant A2780 derivatives, respectively. MEST, FLNC, ARHGDIB and GAPDH. Primer sequences were MEST_qRT_2F: 50-CGGCCATGGTGCGCCGAGAT- 0 0 Treatment of cells with DNMT inhibitor and/or HDAC inhibitor 3 , MEST_qRT_2R: 5 -ACGCAGCAAGCAGGGGCACG- 0 0 Cells were treated at 50% confluence with 0.1. mM 5-aza-20- 3 , FLNC_qRT_1F: 5 -GTGCCCAAGGTCGCTGGGTTA- 0 0 0 deoxycytidine (DAC) (Sigma-Aldrich, St Louis, MO, USA) or CA-3 ,FLNC_qRT_1R:5-TCCCAGGGCCATGCCCAC3 , 0 0.1 mM Belinostat or mock treated for 48 h. DAC was replaced ARHGDIB_qRT_1F: 5 -AACGCTGCTGGGAGATGGTC 0 0 after 24 h. For combined DAC/Belinostat treatment, DAC CTGT-3 , ARHGDIB_qRT_1R: 5 -ACCAGGGTGAGCCG 0 0 (0.1 mM) was added for 48 h, with Belinostat (0.1 mM) being GGTGACAA-3 , GAPDH_qRT_F: 5 -CCTGTTCGACAGT 0 0 added for an additional 24 h following the initial 24 h DAC CAGCCG-3 and GAPDH_qRT_R: 5 -CGACCAAATCCGT 0 treatment. Cells were harvested 96 h following drug removal. TGACTCC-3 . Quantitative real-time PCR was carried out in triplicate with the Biorad CFX96 Real-Time system using SYBR Green JumpStart Taq ReadyMix (Sigma-Aldrich) and RNA extraction and microarray analysis amplification parameters: 95 1C for 3 min, 42 cycles of 95 1C Total cellular RNA was isolated using the RNAeasy kit (Qiagen) for 10 s, 60 1C for 10 s and 72 1C for 30 s, followed by 95 1C for according to the manufacturer’s instructions. RNA quality was 10 s and temperature increments from 72–95 1C for 5 s. verified using the Agilent 2100 Bioanalyser (Agilent, Woking- Relative expression was determined by applying the compara- ham, UK). Expression profiling was carried out using the tive Ct method (Schmittgen and Livak, 2008) using GAPDH as Human Genome U133 Plus 2.0 Array (Affymetrix, High the internal control gene. Wycombe, UK) following the manufacturer’s recommendations. Three biological replicates were used for each sample. The raw expression data were normalized as previously described Conflict of interest (Irizarry et al., 2003). Subsequent analysis was performed in R (version 2.10.1). Data were log2 transformed, and signal intensity and statistical significance were established for each The authors declare no conflict of interest. transcript. Rank products (Breitling et al., 2004) was applied to identify differentially expressed genes using individual probes, Acknowledgements and significance was set at a FDR o0.05. We thank Lisa McMillan for her valuable contribution to the Data deposition initial analysis of the Affymetrix expression data. We also Methylation and expression profiling data are available in thank Nahal Masrour for preparation of samples and Kerra NCBI’s Gene Expression Omnibus (GEO accession number: Pearce for running the Infinium HumanMethylation450 GSE28648). BeadChips at the UCL Genomics Center, London. This work was supported by a Cancer Research UK (CZ, WD and RB) Reverse transcription and real-time PCR grant (C536/A6689), Imperial Experimental Cancer Medicine A total amount of 2 mg RNA was reverse transcribed into Centre, Imperial Biomedical Research Centre and Ovarian complementary DNA using SuperScript II (Invitrogen, Cancer Action (SR and CSMWB).

References

Agarwal R, Kaye SB. (2003). Ovarian cancer: strategies for over- Dai W, Teodoridis JM, Graham J, Zeller C, Huang TH, Yan P et al. coming resistance to chemotherapy. Nat Rev Cancer 3: 502–516. (2008). Methylation linear discriminant analysis (MLDA) Anthoney DA, McIlwrath AJ, Gallagher WM, Edlin AR, Brown R. for identifying differentially methylated CpG islands. BMC (1996). Microsatellite instability, apoptosis, and loss of p53 function Bioinformatics 9: 337. in drug-resistant tumor cells. Cancer Res 56: 1374–1381. Dai W, Teodoridis J, Zeller C, Graham JS, Hersey JM, Flanagan JM Bapat SA, Mali AM, Koppikar CB, Kurrey NK. (2005). Stem and et al. (2011). Systematic CpG islands methylation profiling of genes in progenitor-like cells contribute to the aggressive behavior of human the Wnt pathway in epithelial ovarian cancer identifies biomarkers of epithelial ovarian cancer. Cancer Res 65: 3025–3029. progression-free survival. Clin Cancer Res 17: 4052–4062. Breitling R, Armengaud P, Amtmann A, Herzyk P. (2004). Rank Dean M, Fojo T, Bates S. (2005). Tumour stem cells and drug products: a simple, yet powerful, new method to detect differentially resistance. Nat Rev Cancer 5: 275–284. regulated genes in replicated microarray experiments. FEBS Lett Eckstein N, Servan K, Hildebrandt B, Politz A, von Jonquieres G, 573: 83–92. Wolf-Kummeth S et al. (2009). Hyperactivation of the insulin-like Broxterman HJ, Gotink KJ, Verheul HM. (2009). Understanding the growth factor receptor I signaling pathway is an essential event causes of multidrug resistance in cancer: a comparison of for cisplatin resistance of ovarian cancer cells. Cancer Res 69: doxorubicin and sunitinib. Drug Resist Updat 12: 114–126. 2996–3003. Cameron EE, Bachman KE, Myohanen S, Herman JG, Baylin SB. Edwards SL, Brough R, Lord CJ, Natrajan R, Vatcheva R, Levine DA (1999). Synergy of demethylation and histone deacetylase inhibition et al. (2008). Resistance to therapy caused by intragenic deletion in in the re-expression of genes silenced in cancer. Nat Genet 21: BRCA2. Nature 451: 1111–1115. 103–107. Esteller M. (2008). Epigenetics in cancer. N Engl J Med 358: Cannistra SA. (2004). Cancer of the ovary. N Engl J Med 351: 1148–1159. 2519–2529. Gardiner-Garden M, Frommer M. (1987). CpG islands in vertebrate Chang X, Monitto CL, Demokan S, Kim MS, Chang SS, Zhong X genomes. J Mol Biol 196: 261–282. et al. (2010). Identification of hypermethylated genes associated with Gifford G, Paul J, Vasey PA, Kaye SB, Brown R. (2004). The cisplatin resistance in human cancers. Cancer Res 70: 2870–2879. acquisition of hMLH1 methylation in plasma DNA after

Oncogene Epigenetic drivers of resistance in ovarian cancer C Zeller et al 4576 chemotherapy predicts poor survival for ovarian cancer patients. Plumb JA, Strathdee G, Sludden J, Kaye SB, Brown R. (2000). Clin Cancer Res 10: 4420–4426. Reversal of drug resistance in human tumor xenografts by 20-deoxy- Glasspool RM, Teodoridis JM, Brown R. (2006). Epigenetics as a 5-azacytidine-induced demethylation of the hMLH1 gene promoter. mechanism driving polygenic clinical drug resistance. Br J Cancer Cancer Res 60: 6039–6044. 94: 1087–1092. Rajasekhar VK, Dalerba P, Passegue E, Lagasse E, Najbauer J. (2008). Greenman C, Stephens P, Smith R, Dalgliesh GL, Hunter C, Bignell G The 5th International Society for Stem Cell Research (ISSCR) et al. (2007). Patterns of somatic mutation in human cancer Annual Meeting, June 2007. Stem Cells 26: 292–298. genomes. Nature 446: 153–158. Rizzo S, Hersey JM, Mellor P, Dai W, Santos-Silva A, Liber D et al. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, (2011). Ovarian cancer stem cell-like side populations are enriched Scherf U et al. (2003). Exploration, normalization, and summaries following chemotherapy and overexpress EZH2. Mol Cancer Ther of high density oligonucleotide array probe level data. Biostatistics 10: 325–335. 4: 249–264. Saxonov S, Berg P, Brutlag DL. (2006). A genome-wide analysis Issa JP. (2004). CpG island methylator phenotype in cancer. Nat Rev of CpG dinucleotides in the human genome distinguishes two Cancer 4: 988–993. distinct classes of promoters. Proc Natl Acad Sci USA 103: Johnstone RW, Ruefli AA, Lowe SW. (2002). Apoptosis: a link 1412–1417. between cancer genetics and chemotherapy. Cell 108: 153–164. Schmittgen TD, Livak KJ. (2008). Analyzing real-time PCR data by Jones PA, Baylin SB. (2002). The fundamental role of epigenetic events the comparative C(T) method. Nat Protoc 3: 1101–1108. in cancer. Nat Rev Genet 3: 415–428. Sharma SV, Lee DY, Li B, Quinlan MP, Takahashi F, Maheswaran S Jung H, Lee SK, Jho EH. (2011). Mest/Peg1 inhibits Wnt signaling via et al. (2010). A chromatin-mediated reversible drug-tolerant state in regulation of LRP6 glycosylation. Biochem J 436: 263–269. cancer cell subpopulations. Cell 141: 69–80. Kartalou M, Essigmann JM. (2001). Recognition of cisplatin adducts Smith CA, McClive PJ, Sinclair AH. (2005). Temporal and by cellular proteins. Mutat Res 478: 1–21. spatial expression profile of the novel armadillo-related gene, Alex2, Langdon SP, Lawrie SS, Hay FG, Hawkes MM, McDonald A, during testicular differentiation in the mouse embryo. Dev Dyn 233: Hayward IP et al. (1988). Characterization and properties of 188–193. nine human ovarian adenocarcinoma cell lines. Cancer Res 48: Steele N, Finn P, Brown R, Plumb JA. (2009). Combined inhibition of 6166–6172. DNA methylation and histone acetylation enhances gene re- Lefebvre L, Viville S, Barton SC, Ishino F, Keverne EB, Surani MA. expression and drug sensitivity in vivo. Br J Cancer 100: 758–763. (1998). Abnormal maternal behaviour and growth retardation Stojic L, Brun R, Jiricny J. (2004). Mismatch repair and DNA damage associated with loss of the imprinted gene Mest. Nat Genet 20: 163–169. signalling. DNA Repair (Amst) 3: 1091–1101. Li M, Balch C, Montgomery JS, Jeong M, Chung JH, Yan P et al. Strathdee G, MacKean MJ, Illand M, Brown R. (1999). A role (2009). Integrated analysis of DNA methylation and gene expres- for methylation of the hMLH1 promoter in loss of hMLH1 sion reveals specific signaling pathways associated with platinum expression and drug resistance in ovarian cancer. Oncogene 18: resistance in ovarian cancer. BMC Med Genomics 2: 34. 2335–2341. Luqmani YA. (2005). Mechanisms of drug resistance in cancer Suzuki H, Gabrielson E, Chen W, Anbazhagan R, van Engeland M, chemotherapy. Med Princ Pract 14(Suppl 1): 35–48. Weijenberg MP et al. (2002). A genomic screen for genes Nakanishi H, Suda T, Katoh M, Watanabe A, Igishi T, Kodani M upregulated by demethylation and histone deacetylase inhibition et al. (2004). Loss of imprinting of PEG1/MEST in lung cancer cell in human colorectal cancer. Nat Genet 31: 141–149. lines. Oncol Rep 12: 1273–1278. Szotek PP, Pieretti-Vanmarcke R, Masiakos PT, Dinulescu DM, O’Brien V, Brown R. (2006). Signalling cell cycle arrest and cell death Connolly D, Foster R et al. (2006). Ovarian cancer side population through the MMR System. Carcinogenesis 27: 682–692. defines cells with stem cell-like characteristics and Mullerian Ozols RF. (2004). Advanced ovarian cancer: a clinical update on first- inhibiting substance responsiveness. Proc Natl Acad Sci USA 103: line treatment, recurrent disease, and new agents. J Natl Compr 11154–11159. Canc Netw 2(Suppl 2): S60–S73. Teodoridis JM, Hall J, Marsh S, Kannall HD, Smyth C, Curto J et al. Papouli E, Cejka P, Jiricny J. (2004). Dependence of the cytotoxicity of (2005). CpG island methylation of DNA damage response genes in DNA-damaging agents on the mismatch repair status of human advanced ovarian cancer. Cancer Res 65: 8961–8967. cells. Cancer Res 64: 3391–3394. Wang D, Lippard SJ. (2005). Cellular processing of platinum Pedersen IS, Dervan P, McGoldrick A, Harrison M, Ponchel F, Speirs anticancer drugs. Nat Rev Drug Discov 4: 307–320. V et al. (2002). Promoter switch: a novel mechanism causing biallelic Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, Leary RJ et al. PEG1/MEST expression in invasive breast cancer. Hum Mol Genet (2007). The genomic landscapes of human breast and colorectal 11: 1449–1453. cancers. Science 318: 1108–1113. Peng C, Zhang X, Yu H, Wu D, Zheng J. (2011). Wnt5a as a predictor Yoshioka K, Yoshioka Y, Hsieh P. (2006). ATR kinase activation in poor clinical outcome of patients and a mediator in chemoresis- mediated by MutSalpha and MutLalpha in response to cytotoxic tance of ovarian cancer. Int J Gynecol Cancer 21: 280–288. O6-methylguanine adducts. Mol Cell 22: 501–510.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Oncogene