Discovery of novel DNA methylation biomarkers for non-invasive sporadic breast cancer detection in the Latino population  nica Cappetta1 , Lucıa Fernandez1, Lucıa Brignoni1, Nora Artagaveytia2, Carolina Bonilla3,4, Mo Miguel Lo pez5,6, Manel Esteller7,8,9,10 , Bernardo Bertoni1 and Marıa Berdasco5,6 1 Departamento de Gene tica, Facultad de Medicina, Universidad de la Rep ublica, Montevideo, Uruguay 2 Departamento B asico de Medicina, Facultad de Medicina, Universidad de la Repu blica, Montevideo, Uruguay 3 Departamento de Medicina Preventiva, Facultad de Medicina, Universidad de S~ ao Paulo, Brazil 4 Population Health Sciences, Bristol Medical School, University of Bristol, UK 5 Cancer Epigenetics and Biology Program (PEBC), Bellvitge Biomedical Research Institute (IDIBELL), Barcelona, Spain 6 Epigenetic Therapies Group, Experimental and Clinical Hematology Program (PHEC), Josep Carreras Leukaemia Research Institute (IJC), Badalona, Spain 7 Cancer Epigenetics Group, Cancer and Leukemia Epigenetics and Biology Program (PEBCL), Josep Carreras Leukaemia Research Institute (IJC), Badalona, Spain 8 Physiological Sciences Department, School of Medicine and Health Sciences, University of Barcelona, Spain 9 Centro de Investigacio n Biome dica en Red Cancer (CIBERONC), Madrid, Spain 10 Institucio Catalana de Recerca i Estudis Avancßats (ICREA), Barcelona, Spain Keywords Human diversity is one of the main pitfalls in the development of robust biomarker; breast cancer; DNA methylation; worldwide biomarkers in oncology. Epigenetic variability across human human diversity; non-invasive test populations is associated with different genetic backgrounds, as well as variable lifestyles and environmental exposures, each of which should be *Correspondence M. Cappetta. Laboratorio de Epidemiologıa investigated. To identify potential non-invasive biomarkers of sporadic genetica, Departamento de Gene tica, breast cancer in the Uruguayan population, we studied genome-wide DNA Facultad de Medicina, Universidad de la methylation using Illumina methylation arrays in leukocytes of 22 women Repu blica. Gral. Flores 2125, Montevideo, with sporadic breast cancer and 10 healthy women in a case–control study. Uruguay We described a panel of 38 differentially methylated CpG positions that Tel: (+598) 29243414 was able to cluster breast cancer patients (BCP) and controls, and that also E-mail: monicac@fmed.edu.uy recapitulated methylation differences in 12 primary breast tumors and their M. Berdasco. Josep Carreras Leukaemia Research Institute (IJC), IJC Building, matched normal breast tissue. Moving forward, we simplified the detection Campus ICO-Germans Trias i Pujol, Ctra de method to improve its applicability in a clinical setting and used an inde- Can Ruti, Camı de les Escoles s/n 08916 pendent well-characterized cohort of 80 leukocyte DNA samples from BCP Badalona, Barcelona and 80 healthy controls to validate methylation results at specific cancer-re- Tel: (+34) 93 557 28 00 lated genes. Our investigations identified methylation at CYFIP1 as a novel E-mail: mberdasco@carrerasresearch.org epigenetic biomarker candidate for sporadic breast cancer in the Urugua- yan population. These results provide a proof-of-concept for the design of (Received 18 September 2020, revised 19 October 2020, accepted 23 October 2020), larger studies aimed at validating biomarker panels for the Latin American available online 19 November 2020) population. doi:10.1002/1878-0261.12842 Abbreviations BMI, body mass index; CpGDMs, differentially methylated CpG sites; MS-HRM, methylation-sensitive high-resolution melting. Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 473 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. Blood methylation in breast cancer in Latino population M. Cappetta et al. studies were carried out in Europe, Asia, and USA 1. Introduction and must be validated in each population, especially in Breast cancer is a complex and heterogeneous disease those with admixed genetic ancestry like the Latino caused by the interactions of both genetic and non-ge- population [13]. netic factors. Age, gender, and family history are the In the current study, we aimed to identify novel can- major factors for breast cancer. The known high-risk didate sporadic breast cancer epigenetic biomarkers in inherited mutations in breast cancer susceptibility peripheral blood from the Uruguayan population. genes, such as BRCA1, BRCA2, P53, PTEN, CHEK2, Consequently, we analyzed genome-wide DNA methy- and ATM, together only explain 1.5–3% of all breast lation signatures in leukocytes from sporadic BCP and cancer cases. Meanwhile, known variants with low- healthy women in a well-defined discovery cohort in a penetrance risk to breast cancer only represent a pre- case–control study. Epigenetic biomarkers observed in dictive accuracy of 60% [1]. Therefore, genetic risk specific candidate genes were replicated using a large factors are not enough to evaluate risk of breast can- validation cohort. cer. DNA methylation is a key process involved in the regulation of gene expression. Interestingly, DNA 2. Materials and methods methylation is potentially modifiable and is related to age, the strongest breast cancer risk predictor [2]. 2.1. Study population Alterations in DNA methylation patterns, both at the global genomic level and loci-specific, have been suc- For genome-wide DNA methylation profiling analysis, cessfully explored as molecular biomarkers in cancer 24 DNA samples from peripheral blood leukocytes of management [3]. In our previous work, we reported patients with sporadic breast cancer and 12 DNA sam- global DNA hypomethylation in leukocytes of spo- ples from unaffected controls were selected from a radic breast cancer patients (BCP) compared with group of individuals originally recruited in a previous healthy controls, supporting the potential use of DNA study described in Bonilla et al. [14]. After a quality methylation in leukocytes as a biomarker for cancer control analysis, a discovery cohort composed of 22 [4]. Additionally, we found a negative correlation patients and 10 controls were included in the methyla- between African ancestry and global DNA methyla- tion study (Table S1). To validate candidate CpG tion in cancer patients, suggesting that the ancestral sites, we selected a large and independent validation genome structure generated by the admixture process cohort of the same previous study consisting of 80 in the Uruguayan population influences DNA methy- DNA leukocyte samples from sporadic BCP and 80 lation patterns [4]. This underscores the importance of DNA leukocyte samples from healthy controls, paired searching for population-specific DNA methylation by age, socioeconomic status, and educational level markers for sporadic breast cancer. (Table S2). The procedures followed were approved by In addition, most studies that leverage DNA methy- the ethics committee of the Facultad de Medicina of lation as potential biomarkers for cancer use primary the Universidad de la Rep ublica, Uruguay (Reference tumor tissues. However, a reliable biomarker should number: 071140-000303-12). After obtaining written meet criteria such as acceptable costs and feasibility in informed consent from all participants of the study, preventive medicine to stratified population according peripheral blood was drawn for DNA extraction and to risk scores or to detect cancer earlier. Detection of participants answered an interview-based questionnaire CpG methylation changes in non- or minimal invasive to record medical and epidemiological information. All liquid biopsies including blood cells, saliva, or urine human samples included in the study were handled in would increase translation of molecular evidence into accordance with the tenets of the Declaration of Hel- clinical practice [3]. Several studies have investigated sinki. peripheral blood DNA methylation biomarkers in dif- All cases were originally selected according to the ferent cancer types including head and neck, breast, following criteria: women with breast cancer over lung, bladder, gastric cancer, prostate, colorectal, and 35 years with no personal history of cancer and no ovarian cancers [5–10]. Few studies have attempted to first-degree family history of breast and/or ovarian investigate the role of loci-specific DNA methylation cancer, and non-consanguineous parents. All patients in leukocytes as a marker of breast cancer, most of were sampled at the time of diagnosis and prior to sur- them by candidate gene approaches, and did not use a gery and/or any therapy. The control group was validation set to confirm their results (reviewed in Ref. women over 35 years of age, with normal mammo- [11,12]). However, the majority of the genome-wide grams, no first- or second-degree family history of 474 Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. M. Cappetta et al. Blood methylation in breast cancer in Latino population cancer, and unrelated to any other project participant. bias of dye in the array was corrected using the methy- Controls were selected in the same hospitals as lumi R package [16] from Bioconductor [17]. The patients. For inclusion in this study, all participants methylation percentage of a CpG site was reported as were required to: (a) have a normal hemogram at the a beta value ranging between 0 (no methylation) and 1 time of sampling, and (b) have previous data of global (full methylation). Once the normalized beta values DNA methylation quantification by HPLC and indi- were obtained, a second phase of quality control was vidual genetic ancestry [4]. Breast cancer cases and performed. Β values with a P-value detection > 0.01 controls were age- and individual genetic ancestry- and 23631 CpG sites exhibiting SNP with a frequency paired. of > 1% (1000 Genomes Project Consortium 2010) in Genetic ancestry of all participants of this study was the probe sequence were removed from subsequent previous determined by 59 ancestry informative mark- analyzes. Since all DNA samples were from women, ers (AIMs) selected from the AIMs panel for Hispanic the X chromosome probes were not removed. As a population described by Fejerman et al. [15]. Genotyp- result of the quality control process, we included in ing and individual admixture analysis were described subsequent analysis 459 999 probes (CpG sites) in 22 in Cappetta et al. [4]. women with breast cancer and 10 control women. The Of all the epidemiological information collected Infinium methylation data are available in the Gene from the participants of the discovery and validation Expression Omnibus (GEO) database under the acces- cohorts, the following variables were analyzed in this sion number GSE148663. study: age, sex, genetic ancestry, body mass index (BMI), smoking status (self-reported), and tumor char- 2.3. Independent public cohorts for in silico acteristics (Table S1,S2). studies Methylation data for the independent European BCP 2.2. Genome-wide methylation analysis were obtained from the public data set GSE52865 [18]. Infinium HumanMethylation450 BeadChips were used This data set provided methylation from breast tumor to analyze DNA methylation on a genome-wide scale tissue samples and normal breast tissue of European in the discovery cohort. DNA was extracted from women hybridized to the Illumina HumanMethyla- whole peripheral blood by standardized salting out tion450 BeadChip array. We analyzed only 24 samples methods. Five hundred nanogram of DNA per sample corresponding to breast tumor tissue samples and their was first bisulfite treated using the Zymo EZ-96 DNA- paired adjacent normal breast tissue from this data set. methylation kit (Zymo Research, Orange, CA, USA). Clinical characteristics of these samples were provided Next, about 200 ng of bisulfite-converted DNA was in Table S3. used for hybridization on the HumanMethylation450 DNA methylation and clinical data of 735 BCP of BeadChip (Illumina, San Diego, CA, USA) according The Cancer Genome Atlas (TCGA) database were to the manufacturer’s protocol. Pre-processing and ini- abstracted from GDC Data Portal of the National tial quality assessment of hybridizations was per- Cancer Institute (https://portal.gdc.cancer.gov/). We formed using the Illumina BeadStudio METHYLATION analyzed methylation data of 735 breast primary MODULE software, version 3.2 (Illumina, Inc.). Negative tumors and 89 normal breast tissues. control bead types were used to obtain an estimation of the background intensity level that was subtracted 2.4. Statistical analysis from the loci probe signals. Moreover, the distribution of intensities at negative controls was used to compute All statistical analyses were carried out using the R detection P-values which were assigned to each probe programming language (http://www.r-project.org/) and as a measure of the signal-to-noise ratio. As quality no parametrical analyses. To identify consistently dif- criteria, probes showing a detection P-value greater ferentially methylated CpG sites (CpGDMs) between than 0.001 in at least 6% of the samples and samples BCP and controls, Wilcoxon rank sum test was per- with a detection P-value greater than 0.001 in more formed for normalized beta values of each group. The than 10% of the probes were considered defective. P-values were adjusted for multiple testing using false Thousand nine hundred and forty-seven probes (CpG discovery rate estimation (FDR), and those CpGs with sites evaluated) did not reach these criteria and were P-value < 0.05 were selected and termed ‘CpGDMs’. therefore excluded from further analysis. Beta values Cluster analysis of the selected CpGDMs was per- were then computed from methylated and unmethy- formed using unsupervised hierarchical clustering with lated signals from each microarray assay, and potential complete linkage and Euclidean distance as a measure Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 475 Blood methylation in breast cancer in Latino population M. Cappetta et al. of similarity between samples. We then used the Visualization, and Integrated Discovery (DAVID v6.8) pvclust package in R using multiscale bootstrap resam- [24] and Kyoto Encyclopedia of Genes and Genomes pling (n = 10 000) to define statistically significant (KEGG database) were used for an analysis of molecu- samples clusters [19]. Furthermore, in order to explore lar pathways. In the DAVID analysis, the set of genes the similarity of methylation data between BCP and represented on the Illumina HumanMethylation450 unaffected control groups, an analysis of multidimen- array was used as the reference set and the set of sional scaling was performed using variable methyla- CpGDMs composed the gene set tested. Using the mani- tion values (b value) of CpGDMs of each individual. fest file provided by Illumina, we classified CpGDMs Possible confounding effects on methylation status according to their position relative to CpGs islands were evaluated using generalized linear regression (Island, Shore, Shelf or Open sea) or relative to repeti- models including age, genetic ancestry, smoking status, tive elements; or their genomic compartments feature and leukocyte cell composition as covariables. The es- (Promoter, TSS, Exon, intron, intergenic region). The timateCellCounts function of the minfi Bioconductor genomic location of the CpGDMs was compared to the package was utilized to determine the proportions of distribution of the CpGs in the whole methylation data the white blood cell types (CD4+ and CD8+ T cells, set. P values were computed using Fisher’s exact test to CD56+ NK cells, B cells, monocytes, and granulo- determine over- or under-representation of the cytes) in each sample [20]. CpGDMs. Age, BMI, and genetic ancestry were analyzed as a continuous variable and were assessed using Student’s 2.6. Bisulfite-treated DNA sequencing t-test, while the remaining parameters of the study were considered qualitative variables. Smoking status We amplified by PCR four CpGDMs using bisulfite- was categorized as yes/no considering the individual as treated DNA from four BCP and four controls. PCR a smoker whether it was at the time of sampling and/ amplification reactions in a final volume of 15 µL, or in the past. Data obtention was not possible for all containing: 109 PCR buffer EcoStart (Ecogen), individuals included in the study. 50 mM MgCl2 (Ecogen, Barcelona, Spain), 2 mM Association between clusters in BCP and epidemio- dNTPs; 1 µM specific primers to amplify the gene logical or clinical covariates (tumor stage, tumor histo- sequence and 3 U of DNA polymerase enzyme logical type, hormone receptors) was assessed using (DNA polymerase EcoStart; Ecogen). Primers were Kruskal–Wallis test and Fisher’s exact test. designed to amplify the selected CpGDM region and Overall survival for BCP from the TCGA database flanking sequences of the transcription start site of was evaluated using Kaplan–Meier analysis grouping the corresponding gene (Table S4). PCR amplified patients with methylation values above and below the sequences were visualized by electrophoresis and median, and a proportional hazard Cox regression extracted from the agarose gel using the QIAquick model adjusted for the methylation level of the candi- Gel Extraction Kit (Qiagen, Hilden, Germany). The date biomarker evaluated and age at diagnosis of the extracted DNA was cloned into competent disease. Escherichia coli bacteria (NovaBlue SinglesTM) using The receiver operating characteristic (ROC) curve the pGEM-T vector. Estimation of methylation sta- was plotted with R package pROC version 1.16.1 [21], tus of each CpG site was performed by automated to estimate the discriminatory power of methylation at sequencing of 10 colonies of each study sequence the candidate region of the CYFIP1 gene. The area (Applied Biosystems, Waltham, MA, USA). The aver- under the ROC curve (AUC) was calculated, and the age methylation throughout the region assessed for DeLong method was used to calculate the 95% confi- each gene between patients and controls was com- dent interval (CI) for AUC. pared using Student’s t-test. 2.5. Enrichment analyses of biological pathways 2.7. Methylation-sensitive high-resolution and common sequence features melting (MS-HRM) assay The genes associated with CpGDMs sites were mapped DNA obtained from blood samples was subjected to to discern their relation to cancer by gene ontology anal- bisulfite modification using the EZ DNA methylation ysis using Genes to Systems Breast Cancer Database gold kit (Zymo Research) following the manufacturer’s (G2SBC, http://www.itb.cnr.it/breastcancer/) [22] and protocol. Primers used for amplifying flanking regions searching in the COSMIC database (https://cancer.sa of cg14024502, cg26568226, cg04890607, cg09580608, nger.ac.uk/cosmic/) [23]. The Database for Annotation, cg01229567, cg19246761 and cg24840062 are listed in 476 Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. M. Cappetta et al. Blood methylation in breast cancer in Latino population Table S4. PCR amplification and MS-HRM assay To identify CpGDMs between BCP and healthy were performed on Eco Illumina real-time PCR. The controls, we applied a Wilcoxon rank sum test, deter- final volume of each reaction system was 10 µL, mining 77 CpGDMs positions after correction for including 5 µL of 29 Epitec HRM PCR Master Mix multiple testing (Table 1). This panel of identified (Qiagen), 10 ngr of sodium bisulfite-modified template CpGDMs was able to cluster BCP and healthy con- DNA, and 0.4 mM of each forward and reverse pri- trols separately using a hierarchical cluster approach mer. A series of methylated DNA standards (100%, (Fig. 1A). These results indicate the existence of CpG 75%, 50%, 25%, 15%, 10%, 5%, 2.5% and 0% methylation differences at specific sequences between methylated DNA) were constructed by mixing univer- cancer patients and controls at the leukocyte level, sal unmethylated (0% methylated) and methylated which can be easily visualized by unsupervised classifi- (100% methylated) human whole genomic DNA sam- cation techniques and could function as a breast can- ples (Qiagen). Fluorescence of each sample was nor- cer signature in blood. Since we detected 3 defined malized as a differential signal against unmethylated subclusters among the patient samples group DNA control. Area under the curve from the normal- (Fig. 1A), we analyzed whether these subclusters were ized, difference curves was used to generate a standard associated with tumor characteristics or epidemiologi- curve and determine the degree of methylation of each cal variables. However, no association was found in DNA sample [25,26]. samples from cancer patients between age (P = 0.868), To identify differential methylation in flanking smoking status (P = 0.852), genetic ancestry (Euro- regions of candidate CpGDMs between patients with pean P = 0.064; African P = 0.675; Native American breast cancer and controls in the validation cohort, P = 0.898), tumor stage (P = 0.716), histological type Wilcoxon rank sum test was performed for methyla- of tumor (P = 0.187), hormone receptors (ER P = 1; tion data from MS-HRM analysis of each group. Pos- PR P = 0.112; Her2 P = 0.494) and the subgroups sible confounding effects on methylation status in derived from cluster analysis. candidate genes were evaluated using generalized linear It has been shown that age, genetic ancestry, smok- regression models including age, BMI, smoking status, ing status, and possible disease-related cell heterogene- and genetic ancestry. ity in blood may act as potential cofounders when investigating DNA methylation differences between cases and controls. Adjusting our analysis for these 3. Results epidemiological variables and the predicted cell-type proportion in leukocytes, 38 of the CpGDMs still showed significant differences between BCP and 3.1. Blood DNA methylation profiling reveals a healthy women (P < 0.05 adjusted, Table 1). This panel of differentially methylated CpGs that panel of 38 candidate CpGDMs sites also groups sam- discriminates sporadic breast cancer patients and ples by separating cancer patients from healthy healthy controls women, using cluster analysis (Fig. 1B). We applied a case–control study to describe the key Most of the candidate CpGDMs (37 of 38) were less genomic sequences involved epigenetically in the sus- methylated in the leukocyte DNA from cancer patients ceptibility to sporadic breast cancer in the Uruguayan compared to controls, with only one CpGDMs hyper- population. Consequently, we performed genome-wide methylated. Out of 38 CpGDMs, 29 were associated DNA methylation profiling in a discovery cohort using with 28 different genes: 11 in gene promoters and 18 DNA from leukocytes of women with sporadic breast in gene bodies (Fig. 1C). The remaining nine cancer (n = 22) and healthy women as control CpGDMs mapped to intergenic regions. Considering (n = 10). Comparison of mean methylation values of density and regional composition of CpG, the majority all CpGs sites analyzed between BCP and control (81.6%) of CpGDMs was located outside CpG-rich groups showed a high correlation across all CpGs regions (CpG islands), with 34.2% located in CpG (Spearman, r2 = 0.997, P < 2.2 9 1016), indicating shores flanking the islands (Fig. 1C). The genomic dis- that global DNA methylation patterns in all samples tribution of the CpGDMs in relation to gene context are very similar (Fig. S1A). In addition, to avoid spu- (promoter, UTR, 1st exon, body gene, or intergenic rious relationships due to technical or sampling proce- region) was not different compared to the whole array dures we applied hierarchical clustering of samples CpG distribution (Fisher exact test, P = 0.480) using the methylation values of random 45 000 CpG (Table S5). Gene ontology analysis revealed a func- sites and we were unable to cluster samples according tional enrichment of candidate CpGDMs in biological to their disease status (Fig. S1B). processes associated with signal transduction Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 477 Blood methylation in breast cancer in Latino population M. Cappetta et al. Table 1. Features of 77 differentially methylated CpG (CpGDM) among BCP and unaffected control group. CpG: identification of the probe (from array). CHR: chromosome. TSS1500: 1500 bp of the start site of transcription. TSS200: 200 bp of the start site of transcription. D b value Genomic CpG island (cancer patients - P value P value CpG CHR UCSC REFGENE context context controls) (FDR) adjusted* cg01015663 1 TCEA3 gene body Open sea 0.0766 0.0494 0.0123 cg01229567 1 MIB2 TSS1500 Shore 0.1184 0.0494 0.0417 cg04400047 1 UBIAD1 TSS1500 Shore 0.0526 0.0494 0.0343 cg06432479 1 TAL1 gene body Island 0.0657 0.0494 0.0300 cg10159215 1 LPPR5 TSS200 Island 0.1014 0.0498 0.1404 cg14024502 1 MAP3K6 TSS1500 Shore 0.008 0.0498 0.0275 cg15452381 1 HIVEP3 5’UTR Open sea 0.0196 0.0498 0.0095 cg16727538 1 C1orf213 Shore 0.0714 0.0494 0.0271 cg19246761 1 MIB2 TSS1500 Shore 0.1116 0.0494 0.0353 cg26251270 1 GPX7 TSS1500 Shore 0.0571 0.0494 0.0197 cg26750487 1 CR1L TSS200 Island 0.0848 0.0494 0.9985 cg02537909 2 COBLL1 gene body Island 0.0604 0.0494 0.9979 cg02738156 2 LINC00487 intergenic Open sea 0.0785 0.0494 0.0807 cg09408768 2 KCNJ3 TSS200 Island 0.0621 0.0494 0.0635 cg17165836 2 AFF3 gene body Open sea 0.0701 0.0494 0.0687 cg24130711 2 CCDC85A gene body Shore 0.0541 0.0494 0.0596 cg26175971 2 CYP27A1 gene body Island 0.052 0.0494 0.1145 cg26874367 2 intergenic Open sea 0.0904 0.0494 0.0220 cg01615258 3 intergenic Shore 0.0521 0.0494 0.0202 cg01814969 3 QARS gene body Shore 0.0468 0.0498 0.9991 cg15450445 3 intergenic Open sea 0.0141 0.0143 0.9991 cg24840062 3 CDCP1 gene body Open sea 0.1073 0.0494 0.0218 cg25616514 3 intergenic Open sea 0.0315 0.0498 0.0399 cg03002688 4 intergenic Open sea 0.0598 0.0498 0.0137 cg18860310 4 SLC10A6 gene body Open sea 0.05 0.0494 0.0946 cg26994377 4 intergenic Open sea 0.0587 0.0494 0.4133 cg00608540 5 TRIM7 gene body Open sea 0.0388 0.0498 0.0319 cg01204911 5 ARHGAP26 gene body Open sea 0.0406 0.0494 0.0949 cg06527989 5 UNC5A gene body Shore 0.0735 0.0494 0.0291 cg07475151 5 LOC100268168 gene body Open sea 0.1005 0.0494 0.0648 cg11953913 5 C5orf32 5’UTR Open sea 0.0542 0.0494 0.6035 cg21550107 5 intergenic Open sea 0.0261 0.0494 0.9987 cg02174359 6 MRPS18A gene body Shore 0.0494 0.0494 0.9963 cg04334016 6 CNKSR3 gene body Open sea 0.0324 0.0494 0.3819 cg09639771 6 intergenic Open sea 0.0442 0.0494 0.9975 cg02377685 7 GBX1 TSS200 Island 0.0645 0.0494 0.9980 cg03761471 7 ZYX TSS1500 Shore 0.0661 0.0494 0.1001 cg04153882 7 WIPI2 gene body 0.0659 0.0494 0.1738 cg15602580 7 SDK1 gene body Shore 0.0553 0.0494 0.0258 cg18967180 7 DENND2A gene body 0.068 0.0494 0.9978 cg21252523 7 intergenic Shore 0.0394 0.0494 0.0345 cg27403098 7 KIAA1908 gene body 0.0119 0.0498 0.1343 cg14681767 8 ARHGEF10 5’UTR Island 0.0913 0.0494 0.0718 cg17588094 8 intergenic Shelf 0.0167 0.0498 0.9987 cg05616472 9 EHMT1 gene body 0.0475 0.0494 0.9983 cg14223444 9 TSTD2 5’UTR Shore 0.0367 0.0498 0.0253 cg01311537 10 C10orf128 gene body Open sea 0.026 0.0494 0.0350 cg08560387 10 TSPAN14 5’UTR 0.1142 0.0498 0.0174 cg14770293 10 intergenic Island 0.0763 0.0494 0.9992 cg24168991 10 ITPRIP 5’UTR Shelf 0.0308 0.0494 0.0198 cg03611487 11 LOC100126784 gene body Shelf 0.0486 0.0494 0.0149 cg10141801 11 GUCY2E TSS200 Open sea 0.0456 0.0494 0.9599 478 Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. M. Cappetta et al. Blood methylation in breast cancer in Latino population Table 1. (Continued). D b value Genomic CpG island (cancer patients - P value P value CpG CHR UCSC REFGENE context context controls) (FDR) adjusted* cg13100962 11 intergenic Open sea 0.0544 0.0498 0.0233 cg17679104 11 STK33 TSS1500 Island 0.0612 0.0494 0.0351 cg22623080 11 AMOTL1 TSS200 Island 0.0437 0.0498 0.0314 cg23460961 11 intergenic Open sea 0.0752 0.0494 0.0227 cg04890607 12 HMGA2 gene body Open sea 0.1331 0.0498 0.0356 cg27292547 12 intergenic Open sea 0.0539 0.0498 0.0339 cg11398020 13 KLF5 gene body Island 0.0766 0.0494 0.0667 cg01972418 14 PAX9 TSS1500 Island 0.0386 0.0498 0.0272 cg18581173 15 CT62 TSS1500 Shore 0.0673 0.0494 0.9997 cg20172862 15 intergenic Island 0.0309 0.0498 0.9981 cg24359188 15 BUB1B TSS200 0.0758 0.0494 0.1524 cg26568226 15 CYFIP1 5’UTR Island 0.0733 0.0494 0.0470 cg04470044 16 WFDC1 gene body Shelf 0.0259 0.0498 0.0274 cg10155261 16 LOC23117 gene body Open sea 0.0499 0.0494 0.0581 cg26591162 16 SRL 3’UTR Open sea 0.0149 0.0494 0.9956 cg07777703 17 TUBG2 gene body Shore 0.0491 0.0498 0.9984 cg07848706 17 intergenic Island 0.0722 0.0494 0.9987 cg08960549 17 intergenic Open sea 0.0198 0.0494 0.9984 cg09580608 17 GNA13 1stExon Island 0.0852 0.0494 0.0142 cg22163463 17 PITPNM3 gene body Shore 0.056 0.0494 0.0489 cg01823541 19 GNG7 5’UTR Island 0.0744 0.0498 0.0641 cg03363633 19 TYROBP TSS1500 Open sea 0.0515 0.0494 0.0185 cg22313519 19 KIAA1683 TSS1500 0.0386 0.0498 0.0496 cg20477147 20 NPEPL1 gene body Shore 0.0712 0.0494 0.0211 cg26468205 20 PCMTD2 TSS200 Island 0.015 0.0494 0.0487 Bold values indicate adjusted P value < 0.05. *P value adjusted by age, genetic ancestry, smoking status, and cellular heterogeneity. (GO:0007165) (Fisher exact test, P = 0.024) and regu- To validate methylation status determined by the lation of stem cell population maintenance microarray, we analyzed in four patients and four con- (GO:2000036) (Fisher exact test, P < 0.01). trol methylation status of four candidate CpGDMs and Among the candidate CpGDMs, we found eight their flanking regions by bisulfite-treated DNA sequenc- genes with differential methylation previously associ- ing. Selection criteria were CpGDMs located in regula- ated with breast cancer susceptibility and pathology in tory regions of genes previously associated with breast the G2SBC and COSMIC Databases (AMOTL1, cancer. Hypermethylation of CYFIP1 gene (only CDCP1, CYFIP1, MAP3K6, MIB2, SDK1, TAL1, CpGDM hypermethylated in patients) was observed in and TYROBP) (Table S6). Broadening the search to patients with breast cancer (P = 2.22 9 1012), while genes previously associated with other cancer types in MAP3K6 and MIB2 gene promoters showed a slight Cancer census, we identified three other genes overlap- hypomethylation in BCP (Fig. S2). ping with CpG sites identified as candidate CpGDMs Next, we studied whether methylation at leukocytes (HMGA2, GNA13, and STK33) (Table S6). Except could recapitulate concomitant methylation changes in CYFIP1, all CpGDMs in identified cancer genes are in the primary tumor. Because methylation data on pri- average hypomethylated in leukocytes of BCP com- mary breast tissues were not available for the Urugua- pared to healthy women. Furthermore, five CpGDMs yan BCP and there are no public methylation profiling that show an average methylation difference > 10% data of Latin American patients, an independent pub- between patients and controls mapped to four different lic cohort of European BCP was used to validate genes: CDCP1, MIB2, TSPAN14, and HMGA2, three in silico the CpG differential methylation previously of which were previously described as cancer genes identified in blood (Table S3). We analyzed methyla- (Table 1). tion data of the 38 CpGDMs panel in primary breast Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 479 Blood methylation in breast cancer in Latino population M. Cappetta et al. 480 Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. M. Cappetta et al. Blood methylation in breast cancer in Latino population Fig. 1. Definition of genome-wide CpG profiles in sporadic BCP. (A) Hierarchical heatmap clustering of 77 CpGDMs in 22 BCP (purple) and 10 healthy controls (blue) analyzed on methylation array. CpGDM were ordered by the difference in mean betas values between patients and controls. Methylation level is color coded (green: lowest methylation level; red: highest methylation level). (B) Hierarchical heatmap clustering of 38 selected candidate CpGDMs in 22 BCP (purple) and 10 healthy women (blue). Candidate CpGDMs were ordered by the difference in mean betas values between patients and controls. Methylation level is color coded (green: lowest methylation level; red: highest methylation level). (C) Genomic distribution of 38 candidate CpGDMs regarding their respective location to genes and CpG context. (D) Hierarchical heatmap clustering of 38 candidate CpGDMs in 12 primary breast tumor samples (T, purple) and their matched normal breast tissues (N, blue) analyzed on methylation array. Methylation level is color coded (green: lowest methylation level; red: highest methylation level). tumors samples and their paired-normal breast tissue CYFIP-1 (cg26568226), HMGA2 body gene of this independent cohort. We found that the 38 (cg04890607), 1st exon of GNA13 (cg09580608), MIB2 CpGDMs previously identified in blood were capable promoter region (cg01229567, cg19246761), and intron of clustering tumor breast samples and normal tissues 1 of CDCP1 (cg24840062). To validate the differen- separately using hierarchical cluster analysis (Fig. 1D). tially methylated regions selected, we used a larger and More interestingly, methylation levels of 38 CpGDMs independent Uruguayan leukocyte cohort of 80 spo- were able to separate primary tumor tissue from nor- radic BCP and 80 healthy controls using MS-HRM mal breast tissue in patients with familial breast cancer PCR assay (Table S2). CYFIP1 and CDCP1 were dif- mutation in the BRCA2 gene (pairs N4/T4, N5/T5, ferentially methylated in patients compared to controls N11/T11) (Fig. 1D, Table S3). In addition, 9 of the 38 in the validation cohort (Wilcoxon test, P < 0.01 and candidate CpGDMs showed differential methylation P < 0.05 respectively, Table 2 and Fig. 2). Methylation between tumor tissues and paired healthy tissues in the analysis of 147 bp containing nine CpG sites in the European cohort (Wilcoxon test, FDR adjusted P- evaluated region of the 50 UTR of CYFIP1 showed sig- value < 0.05; Table S7). nificant consistent DNA hypermethylation in the can- cer patients compared with healthy controls (Fig. 2), while analysis of 124 bp containing two CpG sites in 3.2. Validation of aberrant CpG methylation at intron 1 of CDCP1 showed significant hypomethyla- the CYFIP1 gene as a sporadic breast cancer tion in cancer patients. Next, we analyzed whether candidate biomarker methylation levels in these regions were associated With the aim to simplify the CpGDM panel for a with some epidemiological variables. Adjusting our more suitable use in a clinical environment, we selected analysis for age, smoking status, and genetic ancestry seven candidate CpGDMs sites located at regulatory with methylation levels in evaluated regions of the six sites of genes previously reported as associated with candidate genes selected, CYFIP1 still showed a signif- cancer and/or having an average methylation differ- icant difference between BCP and healthy women ence > 10% between patients and controls. These CpG (P = 6.1 9 106, Table 2). sites correspond to the following selected regions: In order to study CYFIP1 as a candidate sporadic MAP3K6 promoter region (cg14024502), 50 UTR of breast cancer biomarker in blood, we evaluated the Table 2. DNA methylation in candidate genes evaluated in the independent validation cohort. TSS1500: 1500 bp of the start site of transcription. Methylation breast Methylation controls Gene Genomic context cancer (mean) (mean) P value P value adjusted* MAP3K6 TSS1500 0.717 0.710 0.915 0.799 CYFIP1 50 UTR 0.39 0.10 7.29 3 107 6.1 3 106 HMGA2 Body 0.469 0.433 0.277 0.515 GNA13 1st exon 0.0035 0.0035 0.878 0.947 MIB2 TSS1500 0.024 0.030 0.100 0.128 CDCP1 Body 0.807 0.758 0.044 0.088 Bold values indicate adjusted P value < 0.05. *P value adjusted by age, smoking status, and genetic ancestry. Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 481 Blood methylation in breast cancer in Latino population M. Cappetta et al. Fig. 2. Validation of methylation differences at candidate genes in an independent cohort (validation cohort). DNA methylation level of MAP3K6, CYFIP1, HMGA2, GNA13, MIB2, and CDCP1 genes in leukocytes from 80 BCP and 80 healthy controls. The boxes represent the interquartile ranges, and the lines across the boxes indicate the median value. Statistically significant differences between cancer patients and healthy controls were determined using Wilcoxon rank sum test (*P < 0.05; **P < 0.01). strength to predict BCP against controls using ROC Table S8, Fig. S3). This result suggested a potential curves in the validation cohort. Methylation of candi- functional role of CYFIP1 methylation in breast can- date CpGDM and the flanking region in the 50 UTR of cer development. Next, we wondered whether CYFIP1 CYFIP1 showed a good predictive ability with an area could be contributing to the overall survival of under the ROC curve (AUC) of 0.732 (95% CI, patients with breast cancer. Since we do not have fol- 0.649–0.815) (Fig. 3). low-up data on Uruguayan patients, we studied overall With the aim to elucidate the possible functional survival in BCP from the TCGA database in relation role of our candidate biomarker in breast cancer, we to CYFIP1 methylation in primary samples and age at analyzed in silico methylation data of CYFIP1 diagnosis of the pathology (Fig. S4). The results of (cg26568226) in tissue samples from BCP from TCGA these analysis (Cox Regression, P value = 0.681) sug- Program. Comparing 735 primary breast tumors with gest that 50 UTR CYFIP1 methylation would not be a 89 healthy breast tissues, we detected significant differ- prognostic marker for BCP. ential methylation of the candidate CpG site located in CYFIP1 (Wilcoxon test, P value = 9.3 9 104, 4. Discussion Patients with localized breast cancer have a 5-year sur- vival rate of 98%. However, if it is diagnosed after metastasis, the survival rate drops dramatically to 27%. These results mark the benefit of screening and early detection, and the vital importance of finding new markers to supplement mammography results. For sporadic breast cancer, a variety of changes in DNA methylation were detected both in primary can- cer samples [27] and in blood of BCP [11,13,28]. In spite of the preclinical data, a meta-study to find stan- dard candidate markers with potential clinical use has not been performed. These difficulties may be attribu- table to differences in the subject populations and tumor pathologies, but most likely due to low power in each of these studies. Genetic ancestry should also be considered especially in populations with a high Fig. 3. ROC curve assessing the discriminative power of the admixed genetic background like Latino ones [4,29]. methylation in 50 UTR region of CYFIP1 for validation cohort. Area In this line, our study described for the first time under the curve (AUC) of 0.732 (95% CI: 0.649–0.815), with an DNA methylation profiling in leukocytes from spo- optimal cutoff point of 14.28% of methylation (red point). radic BCP in a Latin American population. 482 Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. M. Cappetta et al. Blood methylation in breast cancer in Latino population Accordingly with the described global hypomethyla- with < 8% differences between the two groups. Despite tion in primary breast cancer tissues and in peripheral the small change in magnitude, the integration of multi- blood of BCP [4], we have also detected almost exclu- ple epigenetic biomarkers as a predictive signature and/ sively hypomethylated CpGDMs in leukocytes taken or in combination with genetic markers could be of from cancer patients. Each tissue has a unique epige- high translational value [9]. netic signature, which often reflects its differential Importantly, among the CpGDMs panel detected in function [30,31]. Although the greatest variation in leukocytes, CpG sites in genes previously associated DNA methylation is observed between tissues, with breast cancer are described, including Cytoplasmic interindividual differences in DNA methylation of FMR1 interacting protein 1 (CYFIP1), reinforcing the internal tissues are correlated with blood cells for a utility of this approach in the search for biomarkers group of CpG sites [27,32–34]. This reinforces the associated with breast cancer in peripheral blood. hypothesis that variation in DNA methylation at a CYFIP1 may play an important role in the occurrence systemic level in many tissues could be associated with and development of cancers. Loss of CYFIP1 expres- predisposition to certain diseases, allowing to detect sion has been found in a number of human cancers, differential DNA methylation in blood associated with including breast cancer, colon cancer, lung cancer, breast cancer several years before diagnosis [35,36]. bladder cancer, cutaneous squamous cell carcinoma, Furthermore, in studies like our work that use breast nasopharyngeal carcinoma, and acute lymphoblastic cancer cases recruited at diagnosis we must consider leukemia [38–41]. CYFIP1 expression was correlated the possibility that the cancer itself is causing the epi- with tumor progression in epithelial cancers and it genetic changes detected in the blood DNA, including raised the possibility that loss of CYFIP1 might corre- circulating tumor DNA. Unfortunately, primary tumor late with clinical outcome [39]. Specifically, CYFIP1 samples from the Uruguayan BCP were not available. would play a role in the suppression of breast cancer As a closer approach, we confirmed that the panel of cell migration/invasion and metastasis [42], although 38 CpGDMs detected in leukocytes could be validated its suppressive role has been contradicted in other in breast cancer tissue of women of European origin, studies [43]. In sum, the functional role of CYFIP1 in which supports the sensitivity of our current approach tumor development is still unclear and controversial. using non-invasive specimens. Finally, we are aware that our study should be com- To our knowledge, DNA methylation blood plemented with validation of the proposed candidate biomarkers associated with breast cancer have not on large sample size. Although methylation of candi- been previously described in Latin American countries. date CpGDM and flanking region in 50 UTR of If we compared the panel of 38 CpGDMs detected in CYFIP1 in blood showed a good predictive ability of the Uruguayan population with CpGDMs detected in sporadic BCP against healthy women, additional epi- blood samples of European BCP (GSE37965 [37], demiological factor information including genetic fac- GSE51057 [36]), none of our candidate CpGDMs tors and age and is needed to evaluate its potential coincides with these panels (data not shown). The dif- value as an independent biomarker. The cost of the ferent genomic context in admixed populations such as methylation array-based technologies limits its use as a Uruguay as well as different lifestyles could determine screening tool, but the identification of a panel of a differences in epigenetic cancer biomarkers detected. limited number of CpG sites as a cancer biomarker This reinforces the need for each population to detect would allow evaluation of it with other less expensive their own markers associated with breast cancer. technologies. Although most CpGDMs detected have average dif- ferences between patients and controls < 10%, these dif- 5. Conclusions ferences are consistent. Blood biomarker identification has the challenge of blood cell-specific events that can- In summary, this work represents the first study in not be entirely excluded, and the marginal methylation Latin America that describes the search for epigenetic levels introduced by circulating tumor DNA. Therefore, markers in peripheral blood in a well-characterized alterations are expected to present changes of small cohort of patients with sporadic breast cancer. magnitude between cancer patients and controls. Even Although not yet adequate for use in clinical settings, in the study of Heyn et al. [37], in which genetic noise the description of a panel of 38 CpGDM associated is removed and other sources of confounders are with breast cancer in the discovery sample and the val- reduced by analyzing identical twin pairs discordant for idation of CYFIP1 as candidate biomarker in a larger breast cancer, they detected 403 differentially methy- sample demonstrates the potential of blood DNA lated sites in blood DNA between discordant twins, all methylation for development of non-invasive Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 483 Blood methylation in breast cancer in Latino population M. Cappetta et al. applications for detection of sporadic breast cancer References biomarker in a Latin American population. Future 1 Wacholder S, Hartge P, Prentice R, Garcia-Closas M, studies should be aimed at continued exploration of Feigelson HS, Diver WR, Thun MJ, Cox DG, blood DNA methylation biomarkers using prospective Hankinson SE, Kraft P et al. (2010) Performance of studies. common genetic variants in breast-cancer risk models. N Engl J Med 362, 986–993. 2 Horvath S, Zhang Y, Langfelder P, Kahn RS, Boks Acknowledgements MPM, van Eijk K, Ophoff RA (2012) Aging effects on The authors are indebted to the participating patients DNA methylation modules in human brain and blood and unaffected controls whose generosity and coopera- tissue. Genome Biol 13, R97. tion have made this study possible. They also acknowl- 3 Berdasco M & Esteller M (2019) Clinical epigenetics: edge all the researchers of the Susan Komen seizing opportunities for translation. Nat Rev Genet 20, Foundation research project (POP13620) and MSc. 109–127. 4 Cappetta M, Berdasco M, Hochmann J, Bonilla C, Valentina Colistro for their contributions to statistical Sans M, Hidalgo PC, Artagaveytia N, Kittles R, analysis. Martınez M, Esteller M et al. (2015) Effect of genetic This study was supported by grants from the Insti- ancestry on leukocyte global DNA methylation in tuto de Salud Carlos III co-funded by European cancer patients. BMC Cancer 15, 434. Regional Development Funds (ERDF/FEDER) a way 5 Langevin SM, Koestler DC, Christensen BC, Butler to build Europe (PI10/02267 to MB), Agencia Nacio- RA, Wiencke JK, Nelson HH, Houseman EA, Marsit nal de Investigaci on e Innovaci on (ANII) CJ & Kelsey KT (2012) Peripheral blood DNA (FMV_3_2011_1_5904 to MC), and Programa de Methylation profiles are indicative of head and neck Desarrollo de las Ciencias B asicas (PEDECIBA, to squamous cell carcinoma: an epigenome-wide MC and BB). association study. Epigenetics 7, 291–299. 6 Li J, Chen Y, Yu H, Tian J, Yuan F, Fan J, Liu Y, Zhu L, Wang F, Zhao Y et al. (2017) DUSP1 promoter Conflict of interest methylation in peripheral blood leukocyte is associated The authors declare no conflict of interest. with triple-negative breast cancer risk. Sci Rep 7, 43011. 7 Marsit CJ, Koestler DC, Christensen BC, Karagas MR, Houseman EA & Kelsey KT (2011) DNA methylation Data Accessibility array analysis identifies profiles of blood-derived DNA The DNA methylation microarray data reported in methylation associated with bladder cancer. J Clin this article can be found at Gene Expression Ommibus Oncol 29, 1133–1139. (GEO) under the accession number GSE148663. 8 Barry KH, Moore LE, Liao LM, Huang WY, Andreotti G, Poulin M, Berndt SI (2015) Prospective study of DNA methylation at LINE-1 and Alu in Author contributions peripheral blood and the risk of prostate cancer. Prostate 75, 1718–1725. MC and MB conceived and designed the project. MC, 9 Onwuka JU, Li D, Liu Y, Huang H, Xu J, Liu Y, LF, and ML performed epigenetic experiments, and Zhang Y & Zhao Y (2020) A panel of DNA analyzed and interpreted the experimental data. LB methylation signature from peripheral blood may and CB contributed in the selection of samples and in predict colorectal cancer susceptibility. BMC Cancer 20, epidemiological data analysis. NA provided the human 1–11. samples and clinical expertise. ME provided feedback 10 Sun H, Zhou H, Zhang Y, Chen J, Han X, Huang D, on the epigenetic analyses and the manuscript and pro- Ren X, Jia Y, Fan Q, Tian W et al. (2018) Aberrant vided data. BB contributed in the statistical analysis methylation of FAT4 and SOX11 in peripheral blood and provided feedback on manuscript; and MC and leukocytes and their association with gastric cancer risk. MB wrote the manuscript. All authors read and J Cancer 9, 2275–2283. approved the manuscript. 11 Guan Z, Yu H, Cuk K, Zhang Y & Brenner H (2019) Whole-blood DNA methylation markers in early detection of breast cancer: a systematic literature Peer Review review. Cancer Epidemiol Biomarkers Prev 28, 496–505. The peer review history for this article is available at 12 Tang Q, Cheng J, Cao X, Surowy H & Burwinkel B https://publons.com/publon/10.1002/1878-0261.12842. (2016) Blood-based DNA methylation as biomarker for 484 Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. M. Cappetta et al. Blood methylation in breast cancer in Latino population breast cancer: a systematic review. Clin Epigenetics 8, analysis. In Epigenetics Protocols (Tollefsbol TO, ed), pp. 115. 55–71. Humana Press, Totowa, NJ. 13 Johansson A & Flanagan JM (2017) Epigenome-wide 26 Wojdacz TK, Dobrovic A & Hansen LL (2008) association studies for breast cancer risk and risk Methylation-sensitive high-resolution melting. Nat factors. Trends Cancer Res 12, 19–28. Protoc 3, 1903–1908. 14 Bonilla C, Bertoni B, Hidalgo PC, Artagaveytia N, 27 Avraham A, Cho SS, Uhlmann R, Polak ML, Ackermann E, Barreto I, Cancela P, Cappetta M, Sandbank J, Karni T, Pappo I, Halperin R, Vaknin Z, Ega~na A, Figueiro G et al. (2015) Breast cancer risk Sella A et al. (2014) Tissue specific DNA methylation in and genetic ancestry: a case-control study in Uruguay. normal human breast epithelium and in breast cancer. BMC Womens Health 15, 11. PLoS One 9, e91805. 15 Fejerman L, Romieu I, John EM, Lazcano-Ponce E, 28 Michailidou K, Beesley J, Lindstrom S, Canisius S, Huntsman S, Beckman KB, Perez-Stable EJ, Gonzalez Dennis J, Lush MJ, Maranian MJ, Bolla MK, Wang Burchard E, Ziv E & Torres-Mejıa G (2010) European Q, Shah M et al. (2015) Genome-wide association ancestry is positively associated with breast cancer risk analysis of more than 120,000 individuals identifies 15 in Mexican women. Cancer Epidemiol biomarkers Prev new susceptibility loci for breast cancer. Nat Genet 47, 19, 1074–1082. 373–380. 16 Du P, Kibbe WA & Lin SM (2008) lumi: a pipeline for 29 Apprey V, Wang S, Tang W, Kittles RA, Southerland processing Illumina microarray. Bioinformatics 24, WM, Ittmann M, Kwabi-Addo B (2019) Association of 1547–1548. genetic ancestry with DNA methylation changes in 17 Gentleman RC, Carey VJ, Bates DM, Bolstad B, prostate cancer disparity. Anticancer Res 39, 5861–5866. Dettling M, Dudoit S, Ellis B, Gautier L, Ge Y, Gentry 30 Fernandez AF, Assenov Y, Martin-Subero JI, Balint B, J et al. (2004) Bioconductor: open software Siebert R, Taniguchi H, Yamamoto H, Hidalgo M, development for computational biology and Tan AC, Galm O et al. (2011) A DNA methylation bioinformatics. Genome Biol 5, R80. fingerprint of 1628 human samples. Genome Res 22, 18 Stefansson OA, Moran S, Gomez A, Sayols S, Arribas- 407–419. Jorba C, Sandoval J, Hilmarsdottir H, Olafsdottir E, 31 Armstrong DA, Lesseur C, Conradt E, Lester BM & Tryggvadottir L, Jonasson JG et al. (2015) A DNA Marsit CJ (2014) Global and gene-specific DNA methylation-based definition of biologically distinct methylation across multiple tissues in early infancy: breast cancer subtypes. Mol Oncol 9, 555–568. implications for children’s health research. FASEB J 28, 19 Suzuki R & Shimodaira H (2006) Pvclust: an R 2088–2097. package for assessing the uncertainty in hierarchical 32 Avraham A, Sandbank J, Yarom N, Shalom A, Karni T, clustering. Bioinformatics 22, 1540–1542. Pappo I, Sella A, Fich A, Walfisch S, Gheber L et al. (2010) 20 Jaffe AE & Irizarry RA (2014) Accounting for cellular A similar cell-specific pattern of HOXA methylation in heterogeneity is critical in epigenome-wide association normal and in cancer tissues. Epigenetics 5, 41–46. studies. Genome Biol 15, R31. 33 Cho YH, Yazici H, Wu HC, Terry MB, Gonzalez K, 21 Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Qu M, Dalay N & Santella RM (2010) Aberrant Sanchez JC, M€ uller M (2011) pROC: an open-source promoter hypermethylation and genomic package for R and S+ to analyze and compare ROC hypomethylation in tumor, adjacent normal tissues and curves. BMC Bioinformatics 12, 77. blood from breast cancer patients. Anticancer Res 30, 22 Mosca E, Alfieri R, Merelli I, Viti F, Calabria A & 2489–2496. Milanesi L (2010) A multilevel data integration resource 34 Slieker RC, Bos SD, Goeman JJ, Bovee JV, Talens RP, for breast cancer study. BMC Syst Biol 4, 76. Van Der Breggen R, Suchiman HE, Lameijer EW, 23 Forbes SA, Beare D, Bindal N, Bamford S, Ward S, Putter H, van den Akker EB et al. (2013) Identification Cole CG, Jia M, Kok C, Boutselakis H, De T et al. and systematic annotation of tissue-specific (2016) COSMIC: high-resolution cancer genetics using differentially methylated regions using the Illumina the catalogue of somatic mutations in cancer. Curr 450k array. Epigenet Chromatin 6, 1–12. Protoc Hum Genet 91, 10.11.1-10.11.37. 35 Severi G, Southey MC, English DR, Jung C, Lonie A, 24 Huang DW, Sherman BT, Zheng X, Yang J, Imamichi McLean C, Tsimiklis H, Hopper JL, Giles GG & T, Stephens R, Lempicki RA (2009) Extracting Baglietto L (2014) Epigenome-wide methylation in biological meaning from large gene lists with DAVID. DNA from peripheral blood as a marker of risk for Curr Protoc Bioinformatics Chapter 13: Unit 13.11.27, breast cancer. Breast Cancer Res Treat 148, 665–673. 13.11.1–13.11.13. https://doi.org/10.1002/0471250953.b 36 van Veldhoven K, Polidoro S, Baglietto L, Severi G, i1311s27 Sacerdote C, Panico S, Mattiello A, Palli D, Masala G, 25 Candiloro ILM, Mikeska T & Dobrovic A (2011). Close- Krogh V et al. (2015) Epigenome-wide association tube PCR methods for locus-specific DNA methylation Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 485 Blood methylation in breast cancer in Latino population M. Cappetta et al. study reveals decreased average methylation levels years Fig. S2. Bisulfite sequencing of CYFIP1, MAP3K6 before breast cancer diagnosis. Clin Epigenetics 7, 67. and MIB2 genes in sporadic breast cancer and con- 37 Heyn H, Carmona FJ, Gomez A, Ferreira HJ, Bell JT, trols. Sayols S, Ward K, Stefansson OA, Moran S, Sandoval Fig. S3. DNA methylation level of candidate CpGDM J et al. (2013) DNA methylation profiling in breast site in CYFIP1 gene (cg26568226) in 735 primary cancer discordant identical twins identifies DOK7 as breast tumours and 89 healthy breast tissues from The novel epigenetic biomarker. Carcinogenesis 34, 102–108. Cancer Genome Atlas Program (TCGA). 38 Chen X, Qin L, Li P & Mo W (2016) Cyfip1 is Fig. S4. Overall survival analysis of 735 breast cancer downregulated in acute lymphoblastic leukemia and patients from The Cancer Genome Atlas (TCGA) may be a potential biomarker in acute lymphoblastic database in relation to the methylation in our candi- leukemia. Tumor Biol 37, 9285–9288. date CpGDM in CYFIP1. 39 Silva JM, Ezhkova E, Silva J, Heart S, Castillo M, Table S1. Clinical and epidemiological characteristics Campos Y, Castro V, Bonilla F, Cordon-Cardo C, of sporadic breast cancer patients (SBC) and healthy Muthuswamy SK et al. (2009) Cyfip1 is a putative invasion suppressor in epithelial cancers. Cell 137, controls (C) analyzed on 450K Human Methylation 1047–1061. BeadChip platform (discovery cohort). 40 Dziunycz PJ, Neu J, Lefort K, Djerbi N, Freiberger Table S2. Clinical and epidemiological characteristics SN, Iotzova-Weiss G, French LE, Dotto GP & of sporadic breast cancer patients and healthy controls Hofbauer GF (2017) CYFIP1 is directly controlled by used as an independent sample to validate candidate NOTCH1 and down-regulated in cutaneous squamous CpG sites (validation cohort). cell carcinoma. PLoS One 12, 1–13. Table S3. Clinical characteristics of tumors corre- 41 Shi X, Chen X, Li WC, Mo LJ, Lin ZY, Li YY, Yang sponding to paired breast tissues (Normal/Tumor) of Z & Mo WN (2018) Low expression of Cyfip1 may be an independent European cohort of breast cancer a potential biomarker in nasopharyngeal carcinoma. patients. Neoplasma 65, 292–295. Table S4. Primers used in bisulfite DNA sequencing 42 Chang JW, Kuo WH, Lin CM, Chen WL, Chan SH, and MS-HRM for the validation of methylation status Chiu MF, Chang IS, Jiang SS, Tsai FY, Chen CH in selected CpGDMs. et al. (2018) Wild-type p53 upregulates an early onset Table S5. Comparison of the genomic distribution of breast cancer-associated gene GAS7 to suppress 38 CpGDMs to the expected distribution according to metastasis via GAS7–CYFIP1-mediated signaling all printed CpG in the 450K HumanMethylation pathway. Oncogene 37, 4137–4150. array. 43 Teng Y, Bahassan A, Dong D, Hanold LE, Ren X, Table S6. CpGDMs overlapping genes previously asso- Kennedy EJ, Cowell JK (2016) Targeting the WASF3- ciated with cancer (Cancer Gene Census and G2SBC CYFIP1 complex using stapled peptides suppresses Database). cancer cell invasion. Cancer Res 76, 965–973. Table S7. Overlapping CpGDMs detected in the blood of breast cancer patients from Latino population and Supporting information in primary tissues from an independent European cohort of breast cancer patients. Additional supporting information may be found Table S8. In silico methylation analysis of selected online in the Supporting Information section at the end candidate CpGDMs in breast primary tumors of the article. (n = 735) and normal breast tissues (n = 89) of Fig. S1. Unsupervised analysis of CpG methylation in TCGA database. breast cancer and healthy controls. 486 Molecular Oncology 15 (2021) 473–486 ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd.