DNA Methylation Status of Key Cell-Cycle Regulators Such As CDKNA2/P16 and CCNA1 Correlates with Treatment Response to Doxorubic

Published OnlineFirst October 7, 2014; DOI: 10.1158/1078-0432.CCR-14-0297

Clinical Cancer Personalized Medicine and Imaging Research

DNA Methylation Status of Key Cell-Cycle Regulators Such as CDKNA2/p16 and CCNA1 Correlates with Treatment Response to Doxorubicin and 5-Fluorouracil in Locally Advanced Breast Tumors

Jovana Klajic1,2,3, Florence Busato4, Hege Edvardsen3, Nizar Touleimat4, Thomas Fleischer2,3, Ida Bukholm5,6, Anne-Lise Børresen-Dale2,3, Per Eystein Lønning7,8,Jorg€ Tost4, and Vessela N. Kristensen1,2,3

Abstract Purpose: To explore alterations in gene promoter methylation as a potential cause of acquired drug resistance to doxorubicin or combined treatment with 5-fluorouracil and mitomycin C in human breast cancers. Experimental Design: Paired tumor samples from locally advanced breast cancer patients treated with doxorubicin and 5-fluorouracil-mitomycin C were used in the genome-wide DNA methylation analysis as discovery cohort. An enlarged cohort from the same two prospective studies as those in the discovery cohort was used as a validation set in pyrosequencing analysis. Results: A total of 469 genes were differentially methylated after treatment with doxorubicin and revealed a significant association with canonical pathways enriched for immune cell response and cell-cycle regulating genes including CDKN2A, CCND2, CCNA1, which were also associated to treatment response. Treatment with FUMI resulted in 343 differentially methylated genes representing canonical pathways such as retinoate biosynthesis, gai signaling, and LXR/RXR activation. Despite the clearly different genes and pathways involved in the metabolism and therapeutic effect of both drugs, 46 genes were differentially methylated before and after treatment with both doxorubicin and FUMI. DNA methylation profiles in genes such as BRCA1, FOXC1, and IGFBP3, and most notably repetitive elements like ALU and LINE1, were associated with TP53 mutations status. Conclusion: We identified and validated key cell-cycle regulators differentially methylated before and after neoadjuvant chemotherapy such as CDKN2A and CCNA1 and reported that methylation patterns of these genes may be potential predictive markers to anthracycline/mitomycine sensitivity. Clin Cancer Res; 1–10. 2014 AACR.

Introduction

1Division of Medicine, Department of Clinical Molecular Biology and Initially implemented for locally advanced breast cancers, Laboratory Science (EpiGen), Akershus University Hospital, Lørenskog, primary (presurgery) systemic therapy is nowadays increas- 2 Norway. K.G. Jebsen Center for Breast Cancer Research, Institute for ingly used for operable breast carcinomas. Multiple studies Clinical Medicine, Faculty of Medicine, University of Oslo, Oslo, Norway. 3Department of Genetics, Institute for Cancer Research, OUS Radiumhos- have correlated a pathologic complete response to primary pitalet Montebello, Oslo, Norway. 4Laboratory for Epigenetics and Envi- chemotherapy with improved prognosis (1). However, ronment, Centre National de Genotypage, CEA—Institut de Genomique, although factors like hormone receptor status, HER2 over- Evry, France. 5Department of Surgery, Akerhus University Hospital, Oslo, Norway. 6Institute for Clinical Medicine, Faculty of Medicine, University expression, and gene expression signatures do correlate with of Oslo, Norway. 7Section of Oncology, Institute of Clinical Science, pathologic complete response (2, 3), the predictive power University of Bergen, Bergen, Norway. 8Department of Oncology, Hauke- land University Hospital, Bergen, Norway. of each parameter is modest, and the cause of drug resistance remains poorly understood (4). TP53 mutation status Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/). is another strong prognostic factor, being associated with worse prognosis after treatment with anthracycline as well Corresponding Author: Vessela N. Kristensen, Faculty of Medicine, Insti- tute for Clinical Medicine, University of Oslo, 0372, Oslo, Norway. Phone: as mitomycin-containing chemotherapy (5, 6). 47-22-78-13-75/47-920-68-432; Fax: 47-22-78-13-95; E-mail: Beside genetic abnormalities, epigenetic alterations [email protected] may contribute to breast carcinogenesis and tumor doi: 10.1158/1078-0432.CCR-14-0297 growth. A large number of genes have been shown to be 2014 American Association for Cancer Research. inactivated in breast cancer through gene silencing by

www.aacrjournals.org OF1

Klajic et al.

Discovery cohort (genome-wide screening) Translational Relevance Paired tumor samples from locally advanced breast can- We investigated treatment-associated changes in cer were taken before and after neoadjuvant treatment. The the DNA methylation profiles of locally advanced study comprised 19 tumors treated with doxorubicin and breast cancer patients treated with doxorubicin and 14 tumors treated with FUMI. These patients were admitted 0 5 fluorouracil-mitomycin C, prior and after neoadju- to the Haukeland University Hospital in Bergen, Norway vant chemotherapy and validates these profiles in a between 1991 and 2001 (5, 6). DNA from nine normal larger cohort. We identify and validate methylation breast tissues were collected at the Akershus University patterns of key cell-cycle regulators such as CDKN2A Hospital (Lørenskog, Norway) and included as a control and CCNA1 as potential predictive markers to anthra- of DNA methylation background to which tumor-specific cycline/mitomycine sensitivity. We report on the asso- methylation events are compared with. ciation of differentially methylated genes before and after treatment with canonical pathways enriched for Validation cohort cell-cycle regulating genes and immune cell response. An enlarged cohort comprising 85 paired tumors treated WedemonstrateanassociationbetweenTP53 muta- with doxorubicin and 39 paired tumors treated with FUMI tion status and DNA methylation levels in repetitive from the same two prospective studies as those in the elements like ALU and LINE1 and suggest that meth- discovery cohort was used as a validation set. Samples that ylation levels of these genes together with TP53 status were used for 27K arrays were also among those in valida- could be of value in predicting response to chemo- tion cohort. Clinical and molecular parameters were avail- therapy treatment. We also show an association of able for all samples. DNA from fresh-frozen tumor tissues CDKN2A to molecular subtypes and suggest that the was used in both cohorts. Study design is shown in Fig. 1. Luminal B subtype could be used in defining a group, which will benefit from chemotherapy treatment. Classification of Treatment Response The protocols enrolled patients between 1991 and 2001. Response to therapy was classified by the UICC system DNA methylation. Such findings include genes involved commonly applied at that time (11) and not the more in cell-cycle regulation (CDKN2A, CCND2), DNA repair recent RECIST system (12). Responses were classified as CR (MGMT, BRCA1, MLH1, GSTP1), regulation of cell tran- (complete response-complete disappearance of all tumor scription (HOXA5), hormone and receptor-mediated cell lesions), PR (partial response reduction 50% in the sum of signaling (ER and THRb), apoptosis, and tissue invasion all tumor lesions), PD (progressive disease-increase in the and metastasis (RASSF1A, RARb,TWIST,HIN1,CDH1; diameter of any individual tumor lesion by 25%), and ref. 7). Epigenetic silencing of these genes by DNA meth- StbD (stable disease; the condition between PR and PD). ylation is frequent and, in contrast with genetic mutation, Similar to what has been done for previous reports on these reversible (8, 9), making them potential candidates for materials (5, 6), we compared PD tumors as nonresponders pharmacologic manipulation, for example, through tar- with the combined group of tumors classified as StbD/PR as geted inhibition of DNA methyltransferases. Experimen- responders. tal studies have indicated drug-induced alterations in gene promoter methylation as a potential cause of TP53 analysis acquired drug resistance (10). Mutations in the TP53 gene were analyzed with use of The aim of this study was to explore alterations in gene genomic DNA and the temporal temperature gradient gel promoter methylation as a potential cause of acquired drug electrophoresis (TTGE) strategy (5, 6). DNA fragments resistance to doxorubicin or combined treatment with 5- covering exons 2–11 were amplified, all with a GC clamp fluorouracil (5-FU) and mitomycin C (FUMI) in human on one of the primers, and submitted to analysis by TTGE. breast cancers. To do so, we took advantage of our tissue Mutations of the TP53 gene were correlated with response to bank providing unique access to paired tumor samples chemotherapy with use of the c2 method, including Yates collected from the same patients before and after primary correction for a limited number of observations. In addi- treatment with doxorubicin or FUMI. We used the Illumina tion, the differences in the distribution of TP53 mutation Infinium 27K Human DNA Methylation BeadChip to assess among patients revealing a PD and the remaining groups to which degree the genome-wide methylation status is were analyzed with the use of Fisher exact test. associated with treatment response. Methylation assays Materials and Methods Tumor and normal DNA were isolated using standard Patient cohorts phenol/chloroform procedure. Of note, 500 ng of DNA was All tumor samples were collected from two prospective bisulphite converted using the EpiTect 96 Bisulfite Kit studies described in detail elsewhere (5, 6). Each patient (Qiagen GmbH). Effective bisulphite conversion was provided written informed consent, and the protocols were verified by absolute quantification assay using Applied approved by the regional ethical committee. Biosystem7900HT/7900HTFast Real Time PCR System

OF2 Clin Cancer Res; 2014 Clinical Cancer Research

DNA Methylation and Chemotherapy Response

Locally advanced breast cancer

Discovery cohort Validation cohort

Figure 1. Study design. 5-Fluouracil 5-Fluouracil Doxorubicin and Doxorubicin and Mitomycin C Mitomycin C

19 paired 14 paired tumor 85 paired tumor 39 paired tumor tumor samples samples samples samples

Genome-wide screening Pyrosequencing

Amplification and pairs of primers specific for either con- Subsequently, quantile normalization was performed for verted or unconverted DNA. Aliquote of 4 mL of bisulphate- between sample normalization. After preprocessing, seven converted DNA was used to perform genome-wide DNA paired samples treated with doxorubicin were excluded methylation profiling of 33 paired tumor samples and nine because of low-quality DNA methylation profiles, resulting normal breast tissues using the Illumina HumanMethyla- in 26 paired tumor samples that were further analyzed. Beta tion27 BeadChip. All steps were performed according to the values were used for further analysis. Infinium protocol. Methylation data are deposited in Gene Genes were termed differentially methylated in the dis- Expression Omnibus (GEO) repository under accession covery cohort when medians between the two selected number GSE59724 and may be viewed at http://www. groups (jmedian_gpA - median_gpBj) showed a difference ncbi.nlm.nih.gov/geo/query/acc.cgi?acc¼GSE59724. Quan- of at least 15% methylation. The methylation index of titative DNA methylation analysis of bisulphate-treated samples (Z-score) was calculated as follows: (methylation DNA was performed in 125 tumors by pyrosequencing level of each sample mean value of methylation levels)/ (13) on 22 selected genes. Thirteen genes (ABCB1, APC, SD of methylation levels. Then the sum for the genes was BRCA1, CCNA1, CCND2, CDH1, CDKN2A, ESR1, FOSL1, calculated giving one single value (Z-score) for each sample. GSTP1, IGFBP3, MGMT, and RARB2) were chosen for In the validation cohort, differences in the distribution of validation from the genome-wide analysis and seven methylation between the two selected groups were assessed (FOXC1, FOXC2, HMLH1, IGF2, PPP2R2B, PTEN, and by the nonparametric Mann–Whitney test (on parameters RASSF1A) were selected from previous reports from us and with two categories) or the Kruskal–Wallis test analysis on others for differential DNA methylation in breast cancer or parameters with more than two categories. All obtained P breast cancer cell lines. Global DNA methylation levels were values were corrected with the Bonferroni correction meth- assessed analyzing the repetitive LINE1 and ALU sequence od in which the P values are multiplied by the number of families as proxy. comparisons.

Statistical analysis Ingenuity pathway analysis Dataset preprocessing. All targets that contained a "zero" Data were also analyzed by Ingenuity Pathway Analysis value in one sample for methylated and unmethylated (IPA; ref. 14). Core analyses were performed to determine signals were removed (n ¼ 10). Second, intrasample nor- which genes from our dataset were present in already malization, which consists of color bias and background deﬁned canonical pathways. The signiﬁcance of the asso- level correction, was performed using the lumi R package. ciation between our dataset and the canonical pathways are

www.aacrjournals.org Clin Cancer Res; 2014 OF3

Klajic et al.

assessed by: (i) the ratio of the number of molecules from cantly higher than normal controls both before and after our gene list that map to a canonical pathway; (ii) Fisher treatment with doxorubicin, whereas methylation levels of exact test P value, P < 0.05; and (iii) FDR, FDR < 0.05. nonresponders were close to those in normal controls (Fig. 2A). Differentially methylated genes and their func- Results tions are listed in Supplementary Table S1c. These 1,310 Differentially methylated genes in tumors before and genes were then used as input for an IPA. The top significant after treatment with doxorubicin and FUMI revealed by canonical pathways showed a significant association with array screen two canonical pathways (Supplementary Table S2c). Again, Genome-wide DNA methylation analysis was performed communication between innate and adaptive immune cells on 12 and 14 paired tumor samples from locally advanced came on top, as in the case of most differentially methylated breast cancer taken before and after neoadjuvant treatment genes found when comparing before and after treatment with doxorubicin and FUMI. with doxorubicin. A total of 486 CpGs representing 469 genes were found Differentially methylated genes were also observed in differentially methylated (jmedian_gpA - median_gpBj > responders versus nonresponders for 1,220 genes repre- 15% of methylation) before and after treatment with doxo- sented by 1,424 CpGs in patients treated with 5-FU/mito- rubicin; out of these, 230 genes had higher methylation mycin and 546 genes had higher methylation levels in level before and 239 genes after treatment. The DNA meth- responders compared with nonresponders. Compared with ylation status was also compared before and after treatment the normal controls, methylation levels of responders were with FUMI identifying 380 CpGs corresponding to 343 significantly higher before treatment, whereas methylation genes to be differentially methylated (215 genes had higher levels of nonresponders were close to those in normal methylation level before and 128 genes higher methylation controls. Interestingly after treatment, in contrast with what level after treatment). Among these, 469 and 343 genes were was observed for doxorubicin, methylation levels of non- those involved in regulation of cell transcription, immu- responders were significantly higher compared with both, noregulatory and inflammatory processes, cell-cycle regu- responders and normal controls (Fig. 2B). Differentially lators, growth factors, and tumor suppressor genes methylated genes and their functions are listed in Supple- (Supplementary Table S1a and S1b). mentary Table S1d. All differentially methylated 1,220 Next, IPA was applied to the lists of differentially meth- genes were then used as input for IPA. The top significant ylated genes before/after treatment with both drugs sepa- canonical pathways are shown in Supplementary Table S2d. rately. Interestingly, the pathway analysis of the 469 genes Canonical pathway gai signaling involved in the G-protein differentially methylated before and after treatment with cascade was one also found among the most differentially doxorubicin revealed a significant association with the methylated genes before and after treatment with FUMI. following canonical pathways: LXR/RXR activation, com- When comparing the differentially methylated genes munication between innate and adaptive immune cells associated to treatment response to both drugs, the meth- (Supplementary Fig. S1), and the role of cytokines in ylation status of 333 genes was associated to response to mediating communication between immune cells (Supple- both treatment types. The full list of genes is given in mentary Table S2a). IPA of the 343 differentially methylated Supplementary Table S4, including CD33, CD40, ESR2, genes before and after treatment with FUMI revealed a and WT1 as well as CCNA1, CCND2, and CDKN2A, which significant association with canonical pathways such as notably were also found differentially methylated after each retinoate biosynthesis, gai signaling (Supplementary Fig. treatment. Pathway analysis on 323 genes revealed three S2), LXR/RXR activation, and cardiac b-adrenergic signaling significant canonical pathways: cAMP-mediated signaling, (the four significant canonical pathways are given in Sup- G-protein-coupled receptor signaling, and Gai signaling plementary Table S2b). (Supplementary Table S5). Despite the clearly different genes and pathways involved in the metabolism and therapeutic effect of both drugs, 46 Validated genes differentially methylated before and genes were differentially methylated before and after treat- after treatment ment with both drugs. The full list of the 46 genes and their Validation experiment was performed using the highly functions are given in Supplementary Table S3a and S3b. quantitative pyrosequencing method in 85 paired tumors treated with doxorubicin and 39 paired tumors treated DNA methylated genes from array screen and response with FUMI from the same two prospective studies as those to treatment with doxorubicin and FUMI in the discovery cohort. Of the 22 genes selected for A total of 1,468 CpGs representing 1,310 genes were validation, 13 and six genes remained significantly dif- found differentially methylated (jmedian_gpA med- ferentially methylated either before or after treatment ian_gpBj > 15% of methylation) in tumors progressing on with doxorubicin and FUMI, respectively (Table 1 and therapy (PD) compared with tumors retaining a stable 2). After correction for multiple testing (Bonferroni cor- disease or an objective response after treatment with doxo- rection), the observed differences in DNA methylation rubicin. Out of these, 639 genes had higher methylation levels before and after treatment with doxorubicin were levels in responders compared with nonresponders before still significant for all discovered genes except for GSTP1 treatment. Methylation levels of responders were signifi- (markedinTable1)andforallinitiallyfoundgenes

OF4 Clin Cancer Res; 2014 Clinical Cancer Research

DNA Methylation and Chemotherapy Response

A Doxo before treatment Doxo after treatment 800 700 P = 0.001 700 P = 0.001 600

600 500 500

400 400 -Methylation index Methylation index

300 300

Normal Responders-before Nonresponders-before Normal Responders-after Nonresponders-after

FUMI before treatment FUMI after treatment B 600 550

P = 2.9e–4 500

500 450 P = 0.002

400 400 350 P = 0.033 Methylation index Methylation index

300 300

250

Normal Responders-before Nonresponders-before Normal Responders-after Nonresponders-after

Figure 2. Boxplots illustrating significant association between methylation status of response groups and normal breast tissue controls before and after treatment with doxorubicin and FUMI. A, significantly higher methylation levels in responders compared with nonresponders and normal controls are observed before and after treatment with doxorubicin. B, significantly higher methylation levels in responders compared with nonresponders, and normal controls before treatment with FUMI. Significantly higher methylation levels in nonresponders compared with responders and normal controls after treatment. before and after treatment with FUMI. Three genes levels between responders and nonresponders were signif- (CCNA1, CCND2,andCDKN2A)wereconsistently icant for all initially observed genes except for APC, CCND2, reported with the same results in the doxorubicin discov- and PTEN before treatment and CDKN2A, CCNA1, and ery and validation cohorts, whereas CDKN2A was found PTEN after treatment with FUMI (marked in the Table 2). both in the FUMI discovery and validation cohort. Six genes (ABCB1, BRCA1, CCNA1, CCND2, CDKN2A, and RARB) were discovered in the genome-wide analysis and Validated genes associated with response to confirmed by pyrosequencing. treatment The DNA methylation status of cell-cycle regulatory Pyrosequencing analysis confirmed 15 and 13 genes genes, CCNA1, CCND2,andCDKN2A was associated to differentially methylated either before or after treatment treatment response for both doxorubicin and FUMI (Fig. between responders and nonresponders in the doxorubicin 3). These genes were significantly differentially methyl- and FUMI-treated tumors, respectively (Table 2). After ated before/after treatment with doxorubicin and before/ correction for multiple testing (Bonferroni correction), after treatment between responders. Lower levels of meth- differences in DNA methylation levels between responders ylation were observed after treatment in the responder and nonresponders were still significant for all initially groups. Changes of methylation levels in nonresponders found genes except for ABCB1 and FOXC1 before treatment were significant for CDKN2A, and levels for all three genes and CCND2, IGF2, and LINE1 after treatment with doxo- were close to those in normal controls (especially after rubicin (marked in Table 2). Five of these genes (CCNA1, treatment). In general, similar patterns of methylation CCND2, CDKN2A, FOSL1, and APC) were also discovered levels were observed before and after treatment with in the array analysis and then validated by pyrosequencing FUMI. The main difference was higher levels of methyl- in the larger cohort. After correction for multiple testing ation of nonresponders compared with responders of the (Bonferroni correction), differences in DNA methylation CCNA1 gene.

www.aacrjournals.org Clin Cancer Res; 2014 OF5

Klajic et al.

Table 1. Differentially methylated genes before Table 2. Differentially methylated genes before and after neoadjuvant chemotherapy with both and after neoadjuvant chemotherapy in drugs responders versus nonresponders for both drugs Before/after Before treatment After treatment Before (% of After (% of Gene methylation) methylation) P (responders vs. (responders vs. nonresponders) nonresponders) Doxorubicin Gene PP ALU 17.9 20.3 2.3e–19 BRCA1 87.2 85.8 0.001 Doxorubicin APC – – CCNA1 26.8 15.6 4.7e–4 2.3e 11 1.5e 9 ABCB1 a – CCND2 15.4 8.4 5.3e–11 0.006 2.5e 10 BRCA1 – – CDH1 11.4 7.5 6.1e–18 1.3e 11 2e 10 CCNA1 – FOSL1 9.1 1.7 6.1e–18 2.1e 4 CCND2 – a HMLH1 4.2 3.6 8.3e–15 2.3e 8 0.05 CDH1 – – IGF2 45.5 61.4 2.7e–11 1.8e 11 3e 9 CDKN2A – – LINE1 51 40.6 7.1e–6 4.1e 11 7.1e 8 FOSL1 – MGMT 8.8 2.7 1.1e–17 1.1e 11 FOXC1 a – CDKN2A 49.7 30.4 1.8e–6 0.008 7.5e 10 FOXC2 – – PTEN 10.8 5.4 1.7e–5 5.3e 12 2.1e 10 IGF2 a RASSF1A 27.7 12.1 7.1e–5 0.014 LINE1 a GSTP1 11.3 4.6 0.013a 0.019 PPP2R2B – – FUMI 2.1e 7 1.2e 8 PTEN – – ALU 40.1 19.6 6.3e–9 1.3e 11 8.3e 8 RARB – – BRCA1 4.3 7.2 1.6e–4 5.1e 11 2.7e 10 CDH1 5.2 8 2.7e–5 FUMI APC a – HMLH1 5.9 6.7 1.2e–7 0.01 1.7e 6 ABCB1 – – CDKN2A 41 31.9 2.6e–4 2.4e 4 1.8e 7 BRCA1 – – PTEN 2.9 4.3 1.7e–4 3.2e 9 1.1e 8 CCNA1 6.6e–5 0.038a a Not significant after Bonferroni correction. CCND2 0.05a CDKN2A 3e-7 0.001a CDH1 3.8e–6 Correlation with TP53 mutations and gene expression FOSL1 0.033 profiles FOXC1 5.5e–5 2.2e–6 We compared the observed DNA methylation profiles FOXC2 7.2e–8 6.5e–7 with the TP53 mutations status before and after treatment in PPP2R2B 5.3e–8 4.8e–7 both patient cohorts. In the patient cohort treated with PTEN 0.006a 0.017a FUMI, TP53-mutated tumors had significantly lower DNA RARB 9.8e–8 3.2e–6 methylation levels both before and after treatment in three a fi genes: ABCB1, CCND2, and PPP2R2B (Fig. 4A). In addition, Not signi cant after Bonferroni correction. six genes were significantly differentially methylated between TP53 wild-type and mutated tumors before treatment only (Supplementary Table S6). In the doxorubicin- after treatment in the Luminal B subtype. Significant differ- treated cohort, tumors with TP53 mutations had signifi- ences in methylation levels were observed between Luminal cantly lower DNA methylation levels in CCND2 and A and ERBB2 enriched subtype both before/after treatment. RASSF1A before treatment (P ¼ 0.023 and 0.035). In all cases, the methylation levels were lower after treatment We then compared the observed methylation patterns but the fold change of methylation levels was strongest in the with well-defined tumor subclasses. Interestingly, only Luminal B and a fraction of the basal-like subtype. The CDKN2A, which was found associated both with degree of changes were less among tumors belonging to the Luminal methylation before and after treatment as well as with A subtype (Fig. 4B). treatment response to FUMI and doxorubicin, was also differentially methylated between different molecular sub- Discussion classes, both before/after doxorubicin treatment. However, The use of neoadjuvant chemotherapy for patients with although the methylation levels of CDKN2A in Luminal A large and locally advanced breast cancer before surgery has and Luminal B did not differ significantly before treatment, become an established practice allowing shrinkage of the the methylation level of this gene was significantly decreased primary tumor before surgery. Whether the mechanisms of

OF6 Clin Cancer Res; 2014 Clinical Cancer Research

DNA Methylation and Chemotherapy Response

CDKN2A CCND2 CCNA1

P = 4.1e–11 P = 1.8e–6 P = 7.1e–8 P = 5.3e–11 P = 2.3e–8 P = 4.7e–4 P = 2.1e–4 Doxo 10 20 30 40 50 60 70 10 Average DNA methylation (%) Average DNA methylation (%) Average DNA methylation (%) 020304050 0 203040506070 10 before–after before–after before–after before–after before–after before–after before–after before–after before–after Normal All tumors Responders Nonresponders Normal All tumors Responders Nonresponders Normal All tumors Responders Nonresponders

CDKN2A CCND2 CCNA1 P = 3e–7 P = 2.6e–4

P = 6.6e–5 P = 0.08 P = 0.042 P = 0.001 FUMI 10 10 10 Average DNA methylation (%) Average DNA methylation (%) Average DNA methylation (%) 0 203040506070 0 203040506070 020304050

before–after before–after before–after before–after before–after before–after before–after before–after before–after Normal All tumors Responders Nonresponders Normal All tumors Responders Nonresponders Normal All tumors Responders Nonresponders

Figure 3. Average percentage of DNA methylation levels for CDKN2A, CCND2,andCCNA1 gene in doxorubicin and 5-FUMI-treated breast cancers. For each gene, difference in methylation between all samples before/after, between responders before/after, and between nonresponders before/after and normal samples are shown. primary and acquired resistance are similar is still unknown. metastatic potential of metastatic tumor cells (24). In the The few studies conducted on potential associations present study, TLR4 was highly methylated in responders between gene expression profiles and chemoresistance have compared with nonresponders, implying that silencing of not provided definite answers (15, 16). Genes are expressed this gene by promoter methylation may predict better to influence whole "pathways"; thus, it is likely that resis- prognosis. Furthermore, studies are warranted to address tance is not related to either individual factors or expression, whether epigenetics changes in other genes acting up or but rather, should be considered in the context of "func- downstream in this pathway may be related to outcome as tional cascades" (17). well. Functional pathways may be inactivated not only by In the present study, we identified genes differentially genetic, but also by epigenetic events. These may either be methylated before and after neoadjuvant chemotherapy de novo or acquired in response to therapy, mutation effects, involved in cell-cycle control such as CDKN2A, CCNA1, or cellular selection. Here, we took advantage of the fact that and CCND2. To our knowledge, this is the first report on the we have at hand before and after treatment samples from DNA methylation level of these genes and treatment two prospective studies. We found genes differentially response in locally advanced breast cancer treated with methylated between responders and nonresponders work- doxorubicin and FUMI. CDKN2A promoter methylation ing in the same pathway. Notably, these relate to immune has been detected in breast and many other human cancers response and cell-cycle control. Emerging evidence indi- (25). We previously showed that hypermethylation of cates that many chemotherapeutic agents exert immune CDKN2A is a late event during breast carcinogenesis, lead- stimulatory effects (18, 19). Studies of immunogenic cell ing to its inactivation in late stage breast cancers (26). death after chemotherapy revealed that doxorubicin is Expression of CCND2 is lost in the majority of breast involved in processes which lead to expression of toll-like cancers, and a mechanism underlying this loss is the pro- receptor 4 (TLR4) by dendritic cells (20, 21). It was also moter hypermethylation of cyclin D2 (27). In addition, shown that TLR4 expressed on tumor cells contributes to hypermethylation of CCNA1 has been correlated to breast tumor progression by promoting tumor cell proliferation, cancer progression (28). Again, further studies are needed to (22, 23), and that silencing of TLR4 could reduce the validate these findings. A study on the DNA methylation

www.aacrjournals.org Clin Cancer Res; 2014 OF7

Klajic et al.

ABCB1-TP53 CCND2-TP53 A P = 0.036 P = 0.041

P = 0.044 P = 0.017 Average DNA methylation (%) DNA methylation Average (%) DNA methylation Average 01020304050 01020304050

wt-before mut-before wt-after mut-after wt-before mut-before wt-after mut-after

P = 0.045

35 PPP2R2B-TP53 B P = 0.009 Doxo before/after CDKN2A

P = 0.005

P = 0.036 Average DNA methylation (%) DNA methylation Average (%) DNA methylation Average 20 30 40 50 60 0 5 10 15 20 25 30 before–after before–after before–after before–after before–after wt-before mut-before wt-after mut-after Luminal A Luminal B ERBB2 Basal Normal-like

Figure 4. A, boxplots illustrating signiﬁcant association between methylation status and TP53 mutation status. For each gene, difference in methylation between wild-type and TP53-mutated tumors before and after treatment is given. B, boxplots illustrating the differential methylation of CDKN2A between different molecular subtypes before and after treatment with doxorubicin.

profiles in breast cancer cell lines and their doxorubicin- show transcriptional activation of normally silent SINEs. resistant counterpart showed that methylation levels of Activation of repeats resulting from combined lack of TP53 some genes had a directional tendency when acquiring function and DNA methylation was followed by induction doxorubicin resistance (10). Some genes (ABCB1, APC, of the type I IFN signaling pathway (29). Loss of TP53 GSTP1) were becoming hypomethylated and some function and hypomethylation occur naturally in tumors, (BRCA1, CDH1, ESR1) got hypermethylated in breast can- and transcription of normally silent and heavily methylated cer cells resistant to the drug. In the present study, meth- sequences that represent repeats has been shown to occur in ylation levels of ABCB1 and APC were lower in nonrespon- tumors (30, 31). In present study, we showed that ALU ders compared with the responders, whereas the methyla- (SINEs) and LINE1 had significantly lower methylation tion level of CHD1 was higher in nonresponders compared levels in TP53-mutated tumors compared with wild-type. with the responders. Identification of these processes in tumors could be used for In the present study, we observed lower levels of meth- diagnosis of specific tumor types and stages of progression ylation in TP53-mutated tumors compared with wild-type of the given tumor. for some genes both before and after treatment with both Studies on different expression subtypes in breast cancer drugs. It was previously shown that TP53 mutations, togeth- have shown that different subtypes have a different under- er with DNA methylation, play a role in keeping large lying biology strongly influenced by TP53 mutation status families of interspersed (SINEs) and tandem repeats tran- which is reflected in the methylation profile. Thus, basal- scriptionally dormant (29). The authors demonstrated that like tumors contain in general a low degree of DNA meth- TP53-mutated cells treated with a DNA demethylating agent ylation but frequently harbor TP53 mutations (32).

OF8 Clin Cancer Res; 2014 Clinical Cancer Research

DNA Methylation and Chemotherapy Response

Interestingly Luminal B tumors and some of the basal-like Authors' Contributions tumors had significantly lower methylation levels after Conception and design: A.-L. Børresen-Dale, V.N. Kristensen Development of methodology: N. Touleimat, J. Tost treatment. Patients with luminal B tumors were among the Acquisition of data (provided animals, acquired and managed patients, better responders so patients of this subtype might benefit provided facilities, etc.): F. Busato, N. Touleimat, I.R.K. Bukholm, P.E. from treatment with doxorubicin. Lønning, J. Tost, V.N. Kristensen Analysis and interpretation of data (e.g., statistical analysis, biosta- tistics, computational analysis): J. Klajic, F. Busato, H. Edvardsen, Conclusion T. Fleischer, V.N. Kristensen Writing, review, and/or revision of the manuscript: J. Klajic, F. Busato, We identified and validated key cell-cycle regulatory H. Edvardsen, T. Fleischer, A.-L. Børresen-Dale, P.E. Lonning, J. Tost, V.N. genes differentially methylated before and after neoadju- Kristensen CDKN2A CCNA1 Administrative, technical, or material support (i.e., reporting or orga- vant chemotherapy such as and .Thus, nizing data, constructing databases): V.N. Kristensen methylation patterns of these genes may be potential Study supervision: J. Tost, V.N. Kristensen predictive markers to anthracycline/mitomycine sensitiv- Other (experimental work): J. Klajic Other (supervisor of the PhD thesis of J. Klajic): V.N. Kristensen ity. Furthermore, we identified genes differentially methylated between TP53-mutated and wild-type tumors LINE1 ALU Acknowledgments among which are and elements and suggested The authors thank Grethe I. Grenaker Alnæs for technical support in that methylation levels of these genes together with TP53 performing the laboratory experiment. status could be of value in predicting response to chemotherapy treatment. However, further studies are nec- Grant Support essary requiring additional patient cohorts treated with This work was supported by grant no. 193387/V50 from the Norwegian doxorubicin and FUMI to confirm predictive value of all Research Council (to V.N. Kristensen and A.-L. Borresen-Dale) and by the Norwegian Cancer Society (Grant no.4196163832; to V.N. Kristensen). genes mentioned above. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked Disclosure of Potential Conflicts of Interest advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate J. Tost reports receiving commercial research support from DBV Tech- this fact. nologies and has provided expert testimony for Roche Diagnostics, travel expenses to scientific conferences. No potential conflicts of interest were Received February 10, 2014; revised September 25, 2014; accepted disclosed by the other authors. September 25, 2014; published OnlineFirst October 7, 2014.

References 1. Lonning PE. Molecular basis for therapy resistance. Mol Oncol 12. Therasse P, Arbuck SG, Eisenhauer EA, Wanders J, Kaplan RS, Rubin- 2010;4:284–300. stein L, et al. New guidelines to evaluate the response to treatment in 2. Kaufmann M, von MG, Bear HD, Buzdar A, McGale P, Bonnefoi H, solid tumors. European organization for research and treatment of et al. Recommendations from an international expert panel on the cancer, national cancer institute of the United States, national cancer use of neoadjuvant (primary) systemic treatment of operable institute of Canada. J Natl Cancer Inst 2000;92:205–16. breast cancer: new perspectives 2006. Ann Oncol 2007;18: 13. Tost J, Gut IG. DNA methylation analysis by pyrosequencing. Nat 1927–34. Protoc 2007;2:2265–75. 3. Prat A, Perou CM. Deconstructing the molecular portraits of breast 14. IngenuitySystems. Ingenuity Pathway Analysis. [Accessed July 16 cancer. Mol Oncol 2011;5:5–23. 2014]. Available from: http://www ingenuity com/products/ipa. 4. Lonning PE, Knappskog S. Mapping genetic alterations causing che- 15. Buchholz TA, Stivers DN, Stec J, Ayers M, Clark E, Bolt A, et al. Global moresistance in cancer: identifying the roads by tracking the drivers. gene expression changes during neoadjuvant chemotherapy for Oncogene 2013;32:5315–30. human breast cancer. Cancer J 2002;8:461–8. 5. Geisler S, Lonning PE, Aas T, Johnsen H, Fluge O, Haugen DF, et al. 16. Chang JC, Wooten EC, Tsimelzon A, Hilsenbeck SG, Gutierrez MC, Influence of TP53 gene alterations and c-erbB-2 expression on the Elledge R, et al. Gene expression profiling for the prediction of ther- response to treatment with doxorubicin in locally advanced breast apeutic response to docetaxel in patients with breast cancer. Lancet cancer. Cancer Res 2001;61:2505–12. 2003;362:362–9. 6. Geisler S, Borresen-Dale AL, Johnsen H, Aas T, Geisler J, Akslen LA, 17. Lonning PE. Genes causing inherited cancer as beacons to identify the et al. TP53 gene mutations predict the response to neoadjuvant mechanisms of chemoresistance. Trends Mol Med 2004;10:113–8. treatment with 5-fluorouracil and mitomycin in locally advanced breast 18. Emens LA, Machiels JP, Reilly RT, Jaffee EM. Chemotherapy: friend or cancer. Clin Cancer Res 2003;9:5582–8. foe to cancer vaccines? Curr Opin Mol Ther 2001;3:77–84. 7. Jovanovic J, Ronneberg JA, Tost J, Kristensen V. The epigenetics of 19. Lake RA, Robinson BW. Immunotherapy and chemotherapy–a prac- breast cancer. Mol Oncol 2010;4:242–54. tical partnership. Nat Rev Cancer 2005;5:397–405. 8. Jones PA, Baylin SB. The fundamental role of epigenetic events in 20. Obeid M, Tesniere A, Ghiringhelli F, Fimia GM, Apetoh L, Perfettini JL, cancer. Nat Rev Genet 2002;3:415–28. et al. Calreticulin exposure dictates the immunogenicity of cancer cell 9. Sandhu R, Rivenbark AG, Coleman WB. Enhancement of chemother- death. Nat Med 2007;13:54–61. apeutic efficacy in hypermethylator breast cancer cells through tar- 21. Apetoh L, Ghiringhelli F, Tesniere A, Obeid M, Ortiz C, Criollo A, et al. geted and pharmacologic inhibition of DNMT3b. Breast Cancer Res Toll-like receptor 4-dependent contribution of the immune system to Treat 2012;131:385–99. anticancer chemotherapy and radiotherapy. Nat Med 2007;13:1050–9. 10. Boettcher M, Kischkel F, Hoheisel JD. High-definition DNA methylation 22. Huang B, Zhao J, Li H, He KL, Chen Y, Chen SH, et al. Toll-like profiles from breast and ovarian carcinoma cell lines with differing receptors on tumor cells facilitate evasion of immune surveillance. doxorubicin resistance. PLoS ONE 2010;5:e11002. Cancer Res 2005;65:5009–14. 11. Hayward JL, Carbone PP, Heusen JC, Kumaoka S, Segaloff A, Rubens 23. Szczepanski MJ, Czystowska M, Szajnik M, Harasymczuk M, Boy- RD. Assessment of response to therapy in advanced breast cancer. Br iadzis M, Kruk-Zagajewska A, et al. Triggering of Toll-like receptor 4 J Cancer 1977;35:292–8. expressed on human head and neck squamous cell carcinoma

www.aacrjournals.org Clin Cancer Res; 2014 OF9

Klajic et al.

promotes tumor development and protects the tumor from immune 28. Starlard-Davenport A, Kutanzi K, Tryndyak V, Word B, Lyn-Cook B. attack. Cancer Res 2009;69:3105–13. Restoration of the methylation status of hypermethylated gene pro- 24. Hua D, Liu MY, Cheng ZD, Qin XJ, Zhang HM, Chen Y, et al. Small moters by microRNA-29b in human breast cancer: a novel epigenetic interfering RNA-directed targeting of Toll-like receptor 4 inhibits therapeutic approach. J Carcinog 2013;12:15. human prostate cancer cell invasion, survival, and tumorigenicity. Mol 29. Leonova KI, Brodsky L, Lipchick B, Pal M, Novototskaya L, Chenchik Immunol 2009;46:2876–84. AA, et al. p53 cooperates with DNA methylation and a suicidal inter- 25. Herman JG, Merlo A, Mao L, Lapidus RG, Issa JP, Davidson NE, et al. feron response to maintain epigenetic silencing of repeats and non- Inactivation of the CDKN2/p16/MTS1 gene is frequently associated coding RNAs. Proc Natl Acad Sci U S A 2013;110:E89-E98. with aberrant DNA methylation in all common human cancers. Cancer 30. Ting DT, Lipson D, Paul S, Brannigan BW, Akhavanfard S, Coffman EJ, Res 1995;55:4525–30. et al. Aberrant overexpression of satellite repeats in pancreatic and 26. Klajic J, Fleischer T, Dejeux E, Edvardsen H, Warnberg F, Bukholm I, other epithelial cancers. Science 2011;331:593–6. et al. Quantitative DNA methylation analyses reveal stage dependent 31. Li TH, Kim C, Rubin CM, Schmid CW. K562 cells implicate increased DNA methylation and association to clinico-pathological factors in chromatin accessibility in Alu transcriptional activation. Nucleic Acids breast tumors. BMC Cancer 2013;13:456. Res 2000;28:3031–9. 27. Evron E, Umbricht CB, Korz D, Raman V, Loeb DM, Niranjan B, et al. 32. Holm K, Hegardt C, Staaf J, Vallon-Christersson J, Jonsson G, Olsson Loss of cyclin D2 expression in the majority of breast cancers is H, et al. Molecular subtypes of breast cancer are associated with associated with promoter hypermethylation. Cancer Res 2001;61: characteristic DNA methylation patterns. Breast Cancer Res 2010;12: 2782–7. R36.

OF10 Clin Cancer Res; 2014 Clinical Cancer Research

DNA Methylation Status of Key Cell-Cycle Regulators Such as CDKNA2/p16 and CCNA1 Correlates with Treatment Response to Doxorubicin and 5-Fluorouracil in Locally Advanced Breast Tumors

Jovana Klajic, Florence Busato, Hege Edvardsen, et al.

Clin Cancer Res Published OnlineFirst October 7, 2014.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-14-0297

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2014/10/08/1078-0432.CCR-14-0297.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/early/2014/11/22/1078-0432.CCR-14-0297. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.