Abstract
Epigenetic modifications, such as DNA methylation, are widely studied in cancer as they are stable and easy to measure genome wide. DNA methylation changes have been used to differentiate benign from malignant tissue and to predict tumor recurrence or patient outcome. Multiple genome wide DNA methylation studies in breast and prostate cancers have identified genes that are differentially methylated in malignant tissue compared with non-malignant tissue or in association with hormone receptor status or tumor recurrence. Although this has identified potential biomarkers for diagnosis and prognosis, what is highlighted by reviewing these studies is the similarities between breast and prostate cancers. In particular, the gene families/pathways targeted by DNA methylation in breast and prostate cancers have significant overlap and include homeobox genes, zinc finger transcription factors, S100 calcium binding proteins, and potassium voltage-gated family members. Many of the gene pathways targeted by aberrant methylation in breast and prostate cancers are not targeted in other cancers, suggesting that some of these targets may be specific to hormonal cancers. Genome wide DNA methylation profiles in breast and prostate cancers will not only define more specific and sensitive biomarkers for cancer diagnosis and prognosis but also identify novel therapeutic targets, which may be direct targets of agents that reverse DNA methylation or which may target novel gene families that are themselves DNA methylation targets.
Introduction
Numerous diseases are associated with abnormal hormonal regulation including breast and prostate cancers. Breast cancer is a heterogeneous disease that can be classified in many ways. The most primary classification is by hormone receptor status, namely estrogen receptor α (ERα) and progesterone receptor (PR), with more recent recognition that the androgen receptor (AR) is also important. Hormone receptor status of breast cancer is critical because it directs clinical decisions about adjuvant hormone therapy. ERα-positive tumors are responsive to anti-estrogenic treatments, including selective ER modulators like tamoxifen or aromatase inhibitors, which have significantly increased the odds of survival from this disease. However, ∼30% of women have ERα-negative tumors, and currently, there are no established hormone therapies for the clinical management of women with ERα-negative breast cancers.
Prostate cancer is also a heterogeneous disease and, if diagnosed early, is potentially curable by surgery and/or radiotherapy. However, up to 20% of patients will relapse with metastatic disease within 5–10 years. The main therapy for patients with locally advanced or metastatic disease targets androgen production and its receptor. These therapies result in an initial period of tumor regression; however, eventually, prostate cancers become unresponsive and progress to the castrate-resistant state. Current therapies for advanced prostate cancer are not curative and there are limited treatment options for patients that develop castrate-resistant prostate cancer (CRPC).
Although the role of hormone receptors in breast and prostate cancers is well established and utilized for therapeutic treatments, new targets are needed for hormone-resistant disease in breast and prostate cancers. Epigenetic modifications are critical events in tumorigenesis and, in comparison to somatic mutations, occur at a much higher frequency. Epigenetic modifications are reversible changes to the genome and therefore are excellent therapeutic targets. Epigenetic modifications like DNA methylation are also stable and technically easy to assess making them ideal biomarkers for diagnosis and prognosis. It is well established from candidate gene studies that epigenetic modifications play an important role in hormone-dependent cancers (Chin et al. 2011, Huang et al. 2011, Jeronimo et al. 2011, Huynh et al. 2012); however, it is essential to expand our knowledge of epigenetic modifications that may occur commonly in hormonally responsive disease states. In this review, we focus on genome wide DNA methylation studies in breast and prostate cancers and in particular the overlap in aberrantly methylated genes in these hormonally regulated cancers.
DNA methylation
Unlike mutations that alter the DNA sequence, epigenetic modifications regulate gene expression via chromatin remodeling. The most studied epigenetic modification is DNA methylation, which is critical during embryogenesis, imprinting, and X chromosome inactivation (Jaenisch & Bird 2003, Salozhin et al. 2005, Nafee et al. 2008). DNA methylation involves the transfer of a methyl group from methyl donor S-adenosylmethionine to the 5′ carbon of cytosine. This occurs predominantly at CpG. CpG islands, which are clusters of CpGs (>55%), are frequently found within gene promoter regions. Methylation of promoter-associated CpG islands is often linked to gene repression, either through a direct or indirect influence on the chromatin structure resulting in chromatin condensation (Chan et al. 2000, Nakao 2001, Salozhin et al. 2005).
DNA methylation and cancer
DNA methylation is important in maintaining the integrity of the genome, and so it is hardly surprising that alterations in DNA methylation are commonly associated with disease. Numerous studies have supported the important role of DNA methylation in carcinogenesis (reviewed in Baylin & Jones (2011)), have shown that targeting aberrant DNA methylation is a promising cancer therapeutic (Baylin & Jones 2011), and demonstrated that DNA methylation can act as a disease biomarker (Jovanovic et al. 2010, Huang et al. 2011, Jeronimo et al. 2011, Chiam et al. 2012, Sandoval & Esteller 2012). Alterations in DNA methylation have been observed in multiple cancers, with global hypomethylation occurring simultaneously with gene-specific promoter hypermethylation (Esteller 2005, Karpinets & Foy 2005, Baylin & Ohm 2006). Global hypomethylation has been linked to activation of proto-oncogenes and chromosomal instability (Hake et al. 2004, Szyf et al. 2004, Szyf 2005). In contrast, gene-specific hypermethylation is associated with inactivation of genes involved in DNA repair, cell cycle regulation, apoptosis, and tumor suppression (Esteller 2005, Miyamoto & Ushijima 2005, Baylin & Ohm 2006).
Methods
This is a systematic review of genome wide DNA methylation studies in breast and prostate cancers, as reported on PubMed at May 2013. All studies with an unbiased genome wide DNA methylation analysis conducted in human breast and/or prostate cancer tissue, or cell lines derived from these tissues, were included. Cell line studies were included only if they analyzed at least two independent cell lines. We define ‘unbiased’ as techniques with no selection of genomic regions analyzed based on function, structure, or any other knowledge of genes or regions. Techniques that analyze only gene promoter CpG islands are included, although we acknowledge that these methods are selective by their nature. Genome wide techniques are those that assess methylation on a global scale, usually at least 10 000 CpG sites, and where the methylation status can be linked to a genomic location. Gene expression and chromosome copy number analysis of the same samples will be discussed, where available. Studies investigating genome wide changes in gene expression after treatment with demethylating agents, but without corresponding genome wide DNA methylation data, were excluded on the basis that these studies include both primary (direct) and secondary (indirect) effects of demethylation. The search term ‘DNA methylation profiling cancer breast’ retrieved 180 references and ‘DNA methylation profiling cancer prostate’ retrieved 56 references on 6th May 2013. Some of the papers reviewed here were not retrieved by these search terms in PubMed but were identified from reviewing the selected papers. In total, 22 breast cancer and ten prostate cancer studies fitted our genome wide, unbiased criteria. The techniques used for genome wide DNA methylation analysis in breast or prostate cancer have been comprehensively reviewed elsewhere (Laird 2010) and therefore will not be discussed in this review.
Limitations and biases of genome wide DNA methylation techniques
Many techniques for genome wide methylation analysis do not assess DNA methylation equally across the entire genome, instead focusing on CpGs within islands or gene promoters. In breast cancer, Novak et al. (2008), using methyl DNA immunoprecipitation (IP) followed by a promoter array, found that hypermethylation (38%) was more frequent than hypomethylation (6%) in differentially methylated regions (DMRs) between tumor vs normal. However, methyl DNA IP is biased for hypermethylated DNA. Similarly, Ordway et al. (2007) found that hypermethylation is more frequent than hypomethylation in breast cancer compared with non-malignant tissue. Ordway et al. (2007) digested DNA with McrBC, which has loose site specificity, allowing for the analysis of regional DNA methylation density. Samples were hybridized to OGHAv1.0 microarray (Ordway et al. 2006), which predominantly covers promoter regions and transcription start sites. Both these studies predominantly analyzed promoter regions rather than analyzing equally across the entire genome. Hu et al. (2005) used methylation-specific digital karyotyping (MSDK) to identify that in breast, tumor stromal and epithelial cells were hypomethylated relative to normal stromal and epithelial cells. The MSDK technique is biased for CpG islands, but regions analyzed are distributed across the genome rather than clustered around transcription start sites. Friedlander et al. (2012) reported hypermethylation in CRPC relative to benign prostate using the Illumina Infinium HumanMethylation27 BeadChip. This was more pronounced for CpG islands but was common across all areas of the genome analyzed, although the HumanMethylation27 BeadChip predominantly assesses CpGs in gene proximal promoters. Similarly, Kim et al. (2011c) observed overall hypermethylation in cancer using the same platform. In contrast, in the same study, ALU elements were hypomethylated in cancer compared with non-malignant tissue (Kim et al. 2011c). These findings highlight the bias that can be introduced to DNA methylation analysis by different methodologies employed and the requirement for consideration of the technique used when interpreting genome wide DNA methylation results.
DNA methylation correlates with gene expression in breast and prostate cancers in genome wide studies
Historically, candidate gene studies have demonstrated a correlation between DNA methylation and gene expression. Genome wide DNA methylation studies have supported this correlation, particularly the association between CpG island DNA hypermethylation and gene repression (Li et al. 2009, Van der Auwera et al. 2010, Dedeurwaerder et al. 2011, Fang et al. 2011, Kamalakaran et al. 2011, Sproul et al. 2011, Sun et al. 2011). Flanagan et al. (2010), using familial breast cancers and BRCA1/2-mutated cancers, combined DNA methylation profiles that alone predicted BRCA status, with gene expression and copy number variation (CNV) and found that genes with reduced expression were more likely to be in genomic regions with loss of heterozygosity and/or high levels of DNA methylation. Using a panel of breast cancer cell lines, it has also been shown that the combination of gene dosage, allelic status, and DNA methylation explains more gene expression changes than either genomic element alone (Chari et al. 2010). This highlights that in future studies, combining DNA methylation profiling with CNV and gene expression will be an effective tool to facilitate the identification of critical genes involved in tumorigenesis.
Similarly, in prostate cancer, Kim et al. (2011a) demonstrated that gene promoter methylation was associated with gene under-expression, irrespective of the presence of a CpG island, while coding exon methylation was associated with over-expression. A study by Friedlander et al. (2012) also supported the correlation between DNA methylation and gene expression, in this case also integrating chromosome copy number across metastatic CRPC compared with primary cancer and benign prostate tissue. Together, these studies in breast and prostate cancers demonstrate that the functional effects of gene silencing, initially identified for a small subset of candidate genes, can be identified by genome wide DNA methylation studies. Thus, future genome wide DNA methylation studies can provide an additional avenue to identify novel genes silenced by DNA methylation in tumorigenesis.
Genome wide DNA methylation in breast cancer
DNA methylation profiles discriminate tumor from non-malignant breast tissue
Genome wide DNA methylation has been extensively applied to discriminate between tumor and normal or histologically non-malignant breast tissue. One of the first genome wide DNA methylation studies in breast cancer developed MSDK to assess epithelial, myoepithelial, and stromal fibroblasts from normal and cancer breast tissues (Hu et al. 2005; Table 1). Although there were few differences between tumor and normal myoepithelial cells, DNA hypomethylation frequency was higher in tumor epithelial and stromal cells when compared with normal cells of the same subtype (Hu et al. 2005). Based on DNA methylation profiles, Ordway et al. (2007) identified 220 loci that differentiated non-malignant from tumor tissue, with hypermethylation more common than hypomethylation (Table 1). Validation was performed for 53 hypermethylated loci in 16 infiltrating ductal carcinomas and 25 normal or benign breast tissues (Ordway et al. 2007). Sixteen DMRs were also identified with >95% specificity between breast cancer (103 tumors) and non-malignant breast tissue and blood (Ordway et al. 2007), suggesting that tissue type was not critical for methylation detection. Differential methylation in tumors was not significantly associated with age, hormone receptor status, or family history (Ordway et al. 2007). Differential methylation of one marker, GHSR, distinguished cancer from normal with 90% sensitivity and 96% specificity (Ordway et al. 2007). In future, combining multiple markers with high specificity and sensitivity into a biomarker panel may assist in developing a clinical test to accurately differentiate breast tumor tissue from non-malignant.
Summary of published genome wide DNA methylation studies in breast cancer
References | Samples | Methods | Findings |
---|---|---|---|
Hu et al. (2005) | Normal and tumor myoepithelial, epithelial, stromal cells | MSDK | DNA hypomethylated in tumor epithelial and stromal cells when compared to normal cells |
Immunomagnetic separation of cells from fresh tissue | |||
Ordway et al. (2007) | Nine IDC; matched NM | DNA methylation-dependent DNA fractionation; McrBC digested DNA hybridized to OGHAv1.0 microarray (Ordway et al. 2006) | 220 loci distinguished tumor from NM. qPCR validation identified 16 DMR, independent of tumor stage and ER status. GHSR DNA methylation distinguished IDC from normal/benign (90% sensitivity and 96% specificity) |
Validation 1: 16 IDC, 25 NM, 19 normal blood | |||
Validation 2: 103 IDC, 104 NM, 25 normal blood | |||
Novak et al. (2008) | 16 BCa, 5 NM, 4 pure cell populations from tumor and normal; frozen tissue | MeDIP; Affymetrix GeneChip Human Promoter 1.0R Array | 3506 cancer-specific DMR: 2033 hypermethylated, 1473 hypomethylated. ∼90% of DMRs aberrantly methylated in >33% of tumors |
Li et al. (2009) | MCF-7 | MMSDK; Array CGH (CytoChip v2.0 BlueGnome); Gene expression Human U133 Plus 2.0 (Affymetrix) | Significant methylation differences in introns and LINE1. Significant correlation between methylation of promoter CpG islands and gene expression |
MDA-MB-231 | |||
Fresh cells | |||
Tommasi et al. (2009) | Six matched DCIS and NM breast tissue. Frozen tissue | MIRA-assisted microarray analysis. MBD2b/MBD3L1 IP; Human CpG island array (Agilent) | Greater than 100 significant CpG islands altered in at least 3/6 DCIS examined. 1/3 of the CpG islands were associated with homeobox superfamily members |
Chari et al. (2010) | HCC38, HCC1008, HCC1143, HCC1395, HCC1599, HCC1937, HCC2218, BT474, MCF7, and MCF10A (normal) | Infinium HumanMethylation27 BeadChip (Illumina); U133 Plus 2.0 gene expression (Affymetix); CNV (Chari et al. 2006) | Combined DNA copy number, loss of heterozygosity and DNA methylation to identify genes disrupted in BCa |
Flanagan et al. (2010) | Familial cancers: 11 BRCA1, 8 BRCA2, 14 non-BRCA | MeDIP and GeneChip Human Promoter 1.0R Array (Affymetrix); expression and CNV (Waddell et al. 2010) | Methylation profiling predicted BRCA mutation status. Significant number of genes with LOH had increased DNA methylation and decreased expression |
Macrodissected frozen tissue | |||
Validation set: 14 BRCA1, 13 BRCA2, and 20 non-BRCA | |||
Macrodissected FFPE tissue | |||
Li et al. (2010) | 12 ER+PR+ Bca | Infinium HumanMethylation27 BeadChip (Illumina) | 93 hypermethylated; 55 hypomethylated sites when comparing ERPR+ to ERPR− BCa. Hierarchical clustering distinguished ERPR+ from ERPR−. Significant network: inflammatory response and connective tissue disorders |
12 ER−PR− Bca | |||
Frozen tissue | |||
Validation: 31 ER+PR+ and 39 ER−PR− Bca | |||
Frozen tissue | |||
Ruike et al. (2010) | HMEC, MCF7, MDA-MB-213, SKBR3, T47D, Hs578T, MDA-MB-453, HMC-1-8, MRK-nu-1 | MeDIP; Sequencing on Illumina Genome Analyzer | Compared to HMEC there was greater than fourfold increase in CpG island methylation in BCa cell lines |
Van der Auwera et al. (2010) | Ten normal breast tissues, 19 IBC and 43 non-IBC. Frozen tissue | Infinium HumanMethylation27 BeadChip (Illumina) | Comparing breast tumors with normal: 1353 CpG loci (1134 genes) were significantly different; 77% were hypermethylated in tumor vs normal. Differentially methylated genes were related to focal adhesion, extracellular matrix receptor interaction, pathways in cancer, cytokine–cytokine receptor interaction and ether lipid metabolism. Unsupervised hierarchical cluster analysis separated BCa into low and high methylation. High-methylation group enriched for BCa from patients with distant metastasis and poor prognosis |
Gene expression by HGU133 plus 2.0 array (Affymetrix) | |||
Dedeurwaerder et al. (2011) | 119 tumors, 4 normal. Frozen tissue | Infinium HumanMethylation27 BeadChip (Illumina) | 6309 CpGs differentially methylated between normal and tumors. 2985 loci differentially methylated between normal and tumor identified two clusters that correlated with tumor grade and ER status. Using 55 genes with strong anti-correlation between methylation and expression status in univariate analysis, 32 were significant prognostic markers; 13/32 were immune related |
Validation set 117 tumors, 8 normal. Frozen tissue | Affymetrix HG133 Plus 2.0 GeneChip for expression | ||
Fackler et al. (2011) | 103 BCa, 6 NM distant to BCa. Frozen tissue. Fifteen epithelium-enriched organoids from normal tissue | Infinium HumanMethylation27 BeadChip (Illumina) | Unsupervised hierarchical cluster analysis identified two clusters; one enriched for ER+ and one enriched for ER− BCa. Identified 100 CpGs associated with recurrence, these were enriched for homeobox genes |
Fang et al. (2011) | 39 BCa; microdissected FFPE or frozen tissue | Infinium HumanMethylation27 BeadChip (Illumina); Human Genome U133A 2.0 expression (Affymetrix) | Methylation clustered into two groups: HR+ or HR+/−. Methylator phenotype (HR+ group) associated with good outcome. Differentially methylated genes enriched for PRC2 targets |
Validation set n=132 BCa; microdissected FFPE or frozen tissue | |||
Hill et al. (2011) | 39 primary BCa 4 matched tumor/NM pairs | Infinium HumanMethylation27 BeadChip (Illumina); 5-aza validated by PCR in T47D, MDA-MB-231, HCC1937 cells | 264 hypermethylated genes in CpG islands, involved in cell adhesion, cell proliferation, cell cycle, Wnt signaling pathway. High-methylation group associated with relapse and ER/PR+. Low-methylation group associated with triple-negative tumors. DNA methylation clusters associated with ER/PR status (P=0.001), tumor relapse (P=0.035), and lymph node metastasis (P=0.042). Nine genes associated with relapse-free survival |
Kamalakaran et al. (2011) | 108 tumors, 11 adjacent NM. Frozen tissue. 83 tumors used as discovery set and 25 as validation set | MOMA | Increased HOXA family methylation in luminal tumors compared to NM and basal or HER2 subtypes. 2259 loci determined patient prognosis; 921 remained significant in multivariate Cox regression analysis |
Expression data (Naume et al. 2007) | |||
Kim et al. (2011b) | Ten cancers mixed, ten adjacent NM mixed | MeDIP, whole-genome amplification, followed by hybridization on Human CpG island microarray (Agilent) | 972 CpGs hypermethylated and 209 CpGs hypomethylated (1043 genes). IPA: ‘cellular development, embryonic development, tissue development’, ‘cellular movement, cancer, neurological disease’, ‘cell death, embryonic development, cardiovascular disease’. STAT1 may be master regulator of multiple gene transcripts relevant to BCa |
Validation: MSP and qPCR expression in cell lines treated with 5-aza | |||
Sproul et al. (2011) | 19 BCa cell lines, normal HMEC, normal breast tissue, 47 BCa. Frozen tissue | Infinium HumanMethylation27 BeadChip (Illumina) | In cell lines, hypermethylation of CpGs near transcription start associated with gene silencing. 120 genes divided cell lines into epithelial and mesechymal lineages and clustered BCa cases |
Expression data (Neve et al. 2006) | |||
Sun et al. (2011) | MDA-MB-231, BT20, T47D, MCF7, ZR75.1, MDA-MB-468, BT474, MCF10A (normal) | Illumina bisulfite modified deep sequencing; Illumina mRNA-seq | 149 genes with differential CpG island methylation within 5 kb of the 5′ end of the gene and for which mRNA abundance was inversely correlated with methylation status. ER+ and ER− cell lines and tumors clustered according to methylation |
Validation set: 129 BCa gene expression | |||
Faryna et al. (2012) | Two in situ BCa, eight invasive BCa, ten non-malignant tissue (one from an affected patient). Frozen tissue | MBD2-Fc IP; Human CpG island microarrays (Agilent) | Comparing tumor to NM, 214 CpG islands hypermethylated in >6 of 10 microarrays. TAC1 hypermethylated in all ten BCa. Significant overrepresentation of homeobox genes and genes involved in transcription. Correlation with expression and aberrant methylation in early BCa (in situ). Correlations between DNA methylation of BCAN and tumor grade, NXPH1 or KCTD8 with tumor proliferation. Association between SIM1 or KLF11 methylation and metastasis-free survival |
Validation set 1: 45 tumors and 11 controls | |||
Validation set 2: 43 tumors and 8 controls | |||
Kim et al. (2012) | DNA pooled from three BCa and from three match-paired NM adjacent tissue. Frozen tissue | MeDIA using MBD2bt; whole-genome amplification, hybridization on Human CpG island microarray (Agilent) | Several hundred genes differentially methylated between tumor and adjacent NM tissue. Several filtering processes to identify six candidate genes: AXIN2, LHX2, NFIX, OTP, PTGER4, and WT1 |
Cancer Genome Atlas N (2012) | 802 BCa | Infinium HumanMethylation27 or HumanMethylation450 BeadChip (Illumina); gene expression (Agilent custom 244K); SNP (Affymetrix 6.0); miRNA (Illumina GAIIx or HiSSeq 2000); Exome sequencing Agilent SureSelect All Exome v2.0 kit or Nimblegen SeqCap EZ Human Exome v2.0 | 574 probes used to cluster tumors into five methylation groups. Group 3 was hypermethylated and enriched for luminal B subtype. Group 5 was low DNA methylation, basal-like subtype, and high-frequency TP53 mutation |
122 adjacent NM | Supervised analysis compared group 3 to groups 1, 2, and 4. 4283 genes differentially methylated, 1899 genes differentially expressed. 490 genes hypermethylated and lower expression, including genes involved in ‘extracellular region part’ and ‘Wnt signaling pathway’ | ||
Frozen tissue | |||
Zheng & Zhao 2012) | 30 BCa cell lines | DMH, Human CpG island array (Agilent) | No correlation between methylation and molecular subtype |
MSDK, methylation-specific digital karyotyping; MMSDK, modified MSDK; MOMA, methylation oligonucleotide microarray analysis; PRC2, polycomb complex 2; 5-aza, 5-aza-2′-deoxycytidine; BCa, breast cancer; CNV, copy number variation; CGH, comparative genomic hybridization; DCIS, ductal carcinoma in situ; DMH, differential methylation hybridization; DMR, differentially methylated regions; ER, estrogen receptor; FFPE, formalin fixed, paraffin embedded; GSEA, gene set enrichment analysis; HMEC, human mammary epithelial cells; HR+, hormone receptor-positive (ER+PR+); HR−, hormone receptor-negative (ER−PR−); IBC, inflammatory breast cancer; IDC, infiltrating ductal carcinoma; IP, immunoprecipitation; IPA, Ingenuity Pathway Analysis; miRNA, microRNA; LOH, loss of heterozygosity; MBD2bt, truncated methyl DNA binding domain; MeDIA, methylated DNA isolation assay; MeDIP, methyl DNA immunoprecipitation using a 5MeC MAB; MIRA, methylated CpG island recovery assay; MSP, methylation-specific PCR; NM, non-malignant; PR, progesterone receptor. If the information was available, details of the type of tissue used were also included.
Novak et al. (2008) utilized a promoter array after methyl DNA IP with 5-methyl-cytosine antibody (MeDIP) and found more hypermethylated (2033) than hypomethylated (1473) DMRs in cancer compared with normal breast tissue (Table 1). A similar study compared ductal carcinomas in situ to non-malignant tissue with a CpG island array after IP (Tommasi et al. 2009; Table 1). In this study, >100 CpG islands were significantly altered in tumor compared with non-malignant tissue. These islands were positively methylated in >50% of cancer samples. This demonstrates that not only can tumor tissue be differentiated from non-malignant using DNA methylation but also that hypermethylation can easily be distinguished from hypomethylation.
Van der Auwera et al. (2010) used Illumina Infinium HumanMethylation27 BeadChip to analyze normal breast tissues from ten healthy individuals and compared this to 62 breast tumor samples (19 were inflammatory breast cancer; Table 1). The DNA methylation profiles divided the samples into three groups based on high, intermediate, and low DNA methylation levels, with the normal samples having low DNA methylation levels. When comparing DNA methylation between normal and tumor samples, 1352 CpG loci (1134 genes) were differentially methylated (Van der Auwera et al. 2010). For 77% of these CpG loci, there was significantly greater methylation in tumors compared with normal. Another study using the same technology found 6309 CpGs differentially methylated between 119 tumors and four normal breast tissue samples (Dedeurwaerder et al. 2011; Table 1).
Kamalakaran et al. (2011) identified several hundred differentially methylated loci between 11 adjacent non-malignant breast tissues and 108 tumors (Table 1). Kim et al. (2011b) pooled DNA from ten cancers and ten non-malignant matched adjacent tissues and identified 1181 differentially methylated CpGs (corresponding to 1043 genes) with the vast majority (972) hypermethylated (Table 1). Hill et al. (2011) found 291 probes (264 genes) hypermethylated in breast cancer (n=39) compared with non-malignant breast tissue (n=4) after removal of imprinted genes and X chromosome genes (Table 1). Additional studies have also compared tumor to non-malignant tissue and the number of genes identified that discriminates the two depends on the filtering or analyses utilized. For example, Kim et al. (2012) used several filtering processes to identify six genes, whereas Faryna et al. (2012) identified 214 CpG islands but only one CpG island (TAC1) was methylated in all ten cancer samples. The identification of thousands of loci with altered DNA methylation between non-malignant and tumor tissue using genome wide methylation analyses raises several possibilities for future study. The most pertinent of these are i) loci with aberrant DNA methylation can be combined to generate a panel of biomarkers and ii) genes previously unrecognized as methylation targets can be identified in an unbiased manner, providing potential insight into the biology underlying the disease process.
DNA methylation profiles and breast cancer hormone receptor status or subtype
Several studies have investigated whether genome wide DNA methylation profiling can cluster breast cancers into hormone receptor status (ER/PR positive or negative) or subtype (luminal A or B, basal or HER2). DNA methylation profiles can differentiate hormone receptor-positive breast cancers from hormone receptor-negative cases (Li et al. 2010, Dedeurwaerder et al. 2011, Fang et al. 2011, Hill et al. 2011, Sun et al. 2011), although a study by Van der Auwera et al. (2010) did not find any association between hormone receptor status and DNA methylation profile (Table 1). It is important to note that there were several methodological differences between these studies, such as how the clustering analysis was performed and whether information from all CpG sites were used or only highly methylated CpG sites. These methodological differences may explain the different results reported. Likewise, the patient cohorts were different between these studies and so this may contribute to the differing outcomes. The majority of genome wide DNA methylation studies have found that ER+PR+ tumors have higher levels of DNA methylation compared with ER−PR− tumors (Li et al. 2010, Fackler et al. 2011, Fang et al. 2011, Hill et al. 2011). Li et al. (2010) found 148 altered CpG sites (93 hypermethylated and 55 hypomethylated) in ER+PR+ breast cancers relative to ER−PR− tumors (Table 1). Fackler et al. (2011) identified 40 CpG probes that had an overall specificity of 89% and sensitivity of 90% for classifying ER+ from ER− tumors. Hill et al. (2011) used cluster analysis to show that ER+PR+ tumors had high methylation, whereas triple-negative breast cancers had low methylation (Table 1). Genome wide DNA methylation studies in breast cancer cell lines have also shown clustering according to hormone receptor status based on DNA methylation levels (Sun et al. 2011). Most critically, these genome wide DNA methylation studies demonstrate that an adequately powered study of appropriate clinical samples should identify methylation differences based on hormone receptor status. With additional future study, this may serve as a basis for the development of an improved clinical test to identify the hormone status of breast cancers.
Several studies have also assessed DNA methylation and its association with breast cancer subtypes. A study profiling 30 breast cancer cell lines did not find any correlation between methylation and breast cancer subtypes, although the basal cell type was associated with two DNA methylation clusters (Zheng & Zhao 2012). Dedeurwaerder et al. (2011) identified six DNA methylation clusters and HER2-positive tumors, basal-like and luminal A cancer each belonged to a cluster, but three clusters were not associated with a predominant molecular subtype. Kamalakaran et al. (2011) also performed DNA methylation cluster analysis and found that one cluster was predominantly luminal A (22/30 samples), the second cluster was highly correlated with basal-like (7/8 samples), and the third cluster contained a mixture of subtypes. Recently, The Cancer Genome Atlas assessed DNA methylation profiles in 802 breast cancers and identified that the group with high DNA methylation was associated with the luminal B subtype while the group with low methylation was associated with the basal-like subtype (Table 1; Cancer Genome Atlas N 2012). Future studies need to include a more detailed investigation of the methylation differences between breast cancer subtypes to determine whether there is a methylation signature that can identify breast cancer subtypes. It is possible that methylation analysis will need to be combined with other genome wide analyses to correctly cluster breast cancer subtypes. It is also possible that DNA methylation subtypes are different to the subtypes identified by gene expression and may provide additional information that assists in the clinical setting. Further research is required to delineate these options and determine how subtypes identified by DNA methylation profiling differ to subtypes identified by gene expression.
DNA methylation profiles and breast cancer prognosis
Identifying biomarkers for breast cancer prognosis by DNA methylation profiling have also been the subject of intensive investigations. One of the first biomarker studies found that when breast cancer samples were divided into high and low DNA methylation levels, the high group was enriched for patients with a poor prognosis (Van der Auwera et al. 2010). Using a clustering approach, Fang et al. (2011) found that the group of breast cancers with high DNA methylation levels was ER/PR+ and that this group has a lower risk of relapse and better overall survival. By selecting genes with a strong anti-correlation between DNA methylation and expression status, univariate Cox regression analysis identified 32 genes that were significant prognostic markers in breast cancer (Dedeurwaerder et al. 2011). Further filtering of these genes identified 11 genes associated with immunity, and high expression of these genes was associated with a good clinical outcome; 10/11 were independent prognostic markers by multivariate analysis (Dedeurwaerder et al. 2011). Using a genome wide DNA methylation approach on 82 breast cancers, 100 CpGs were associated with recurrence; homeobox genes were enriched in this set of hypermethylated loci (Fackler et al. 2011). Focusing just on homeobox genes, 60 homeobox loci were associated with recurrence, and these were an independent predictor of recurrence (Fackler et al. 2011). It is important to note that this study design was biased toward hypermethylated loci.
Using cluster analysis of genome wide DNA methylation data to divide breast cancers into three groups, high, intermediate, and low DNA methylation levels, the high DNA methylation group was associated with relapse but there was no association with disease-free survival with any of the groups (Hill et al. 2011). However, when the probes/genes associated with a single clinical feature (hormone receptor status, tumor relapse, and lymph node status) were assessed individually using Fisher's exact tests with false discovery rate correction, nine genes were identified that were significantly associated with relapse-free survival (Hill et al. 2011). This demonstrates the importance of dividing cancers into appropriate groups for analysis. Kamalakaran et al. (2011) identified 2259 loci that divided patients into good and poor prognosis groups; 921 of these remained significant after multivariate and Cox regression analysis. Faryna et al. (2012), although with a small samples size, identified methylation of two genes associated with metastasis-free survival. Further study should be directed at determining the minimal number of loci required to effectively stratify patients by prognosis. Extensive future work is also required to follow on from all these prognostic studies to validate these observations in additional clinical cohorts. In addition, biological characterization of relevant loci is required to understand how methylation at particular loci contributes to or influences breast cancer prognosis.
Genome wide DNA methylation in prostate cancer
DNA methylation profiles that discriminate tumor from non-malignant prostate tissue
Eight genome wide DNA methylation studies have compared non-malignant and tumor tissue. Chung et al. (2008) used methylated CpG island amplification coupled with representational difference analysis of prostate cancer cell lines (pooled DU145, PC3, and LNCaP) and identified eight genes with promoter hypermethylation (Table 2). These genes were validated in prostate cancer compared with non-malignant tissue (n=20) and were re-expressed by the DNA methylation inhibitor 5-aza-2′-deoxycytidine (5-aza). FAM84A (NSE1) and SPOCK2 hypermethylation was the optimal biomarker combination in differentiating cancer from non-malignant tissue (sensitivity, 80%; specificity, 95%; and accuracy, 96%; Chung et al. 2008). Kim et al. (2011d) combined genome wide DNA methylation and gene expression in pooled LNCaP-FGC, DU-145, and PC-3 cells compared with RWPE-1 non-malignant cultured prostate cells (Table 2). Three genes exhibited concordant methylation and expression changes, had a CpG island, and enhanced expression after 5-aza treatment. In clinical samples, EFEMP1 methylation was the optimal marker to differentiate prostate cancer from benign prostatic hyperplasia (sensitivity, 95.3% and specificity, 86.6%). EFEMP1 gene expression was also reduced in an independent prostate cancer cohort (Kim et al. 2011d). Kron et al. (2009) performed differential methylation hybridization (DMH) of prostate cancer samples (Table 2). Of the genes hypermethylated in prostate cancer, HOXD3 and BMP7 were validated by 5-aza re-expression in vitro in DU145 cells (Kron et al. 2009). These studies identified several candidate genes with differential DNA methylation and gene expression between cancerous and non-malignant tissue. Future functional studies of these genes will clarify their role in prostate tumorigenesis.
Summary of published genome wide DNA methylation studies in prostate cancer.
References | Samples | Methods | Findings |
---|---|---|---|
Chung et al. (2008) | DU145, PC3, LNCaP, 20 matched microdissected PCa–NM pairs | MCA-RDA; sequenced 198 random clones | 34 promoter-associated CpG islands. Filtered through cell line panel; eight genes hypermethylated in PCa vs NM, re-expressed by 5-aza in vitro |
Kron et al. (2009) | 3+3 or 4+4 PCa (n=10 each; frozen tissue) | DMH; microarray with 7776 CpG islands (Yan et al. 2001) | 27/100 top methylated genes PCa vs NM (lymphocytes) homeobox or T-box genes. 3+3 vs 4+4 – 23/100 top methylated genes are homeobox genes |
Ref: lymphocytes | |||
Coolen et al. (2010) | LNCaP, DU145, PC3, MDA-2A, PrEC | MeDIP-chip; GeneChip Human Promoter 1.0R arrays (Affymetrix) | 47 regions of LRES in PCa, based on gene expression in clinical cohorts and cell lines after 5-aza treatment. Genome wide methylation reported for seven LRES |
Kim et al. (2011a) | LNCaP, PrEC 6 adjacent NM, 2 NM, 5 PCa, 4 metastatic CRPC | MethylPlex NGS (Illumina); Human GE 44K microarray (Agilent) | CpG island methylation increased with progression (benign 13%, localized 19%, and metastatic 22%). 2481 regions with cancer-specific differential methylation |
Kim et al. (2011d) | LNCaP-FGC, DU-145, PC-3, RWPE-1 | Infinium HumanMethylation27 BeadChip and Human HT-12 Gene Expression BeadChip (Illumina) | 106 genes hypermethylated in PCa cell lines compared to RWPE-1. Filtered by greater than twofold difference in gene expression in PCa cell lines, presence of a CpG island, re-expression with 5-aza. Four candidates validated in clinical cohort |
Validation: 97 BPH, 106 PCa; frozen tissue | |||
Kobayashi et al. (2011) | 95 PCa 86 adjacent NM (includes 70 matched pairs); microdissected frozen tissue | Infinium HumanMethylation27 BeadChip (Illumina) | 87 CpG biomarkers for NM vs PCa, 83 hypermethylated in PCa. 69 CpGs associate with recurrence. Gleason grade not distinguishable by methylation |
Kim et al. (2011c) | Pooled matched PCa and NM (n=12 patients); microdissected FFPE tissue | Infinium HumanMethylation27 BeadChip (Illumina) | General hypermethylation in PCa; ALU-specific hypomethylation in cancer. 972 CpGs hypermethylated; 209 CpGs hypomethylated (1043 genes) |
Friedlander et al. (2012) | Metastatic CRPC n=15 (14 individuals) seven liver and eight soft tissue | Infinium HumanMethylation27 BeadChip (Illumina); aCGH Human Genome microarray and Whole-Human Gene Expression 44K microarray (Agilent) | 513 probes with differential methylation between CRPC and BPH. Aberrant methylation more frequent (10.5%) than CNV (2.1%). Sixteen genes methylated and copy loss in the same tumor in ≥66% of samples. Pathways of genes structurally and epigenetically altered: androgen biosynthesis, p53 pathway, IGF1–protein kinase B signaling cascade |
Mahapatra et al. (2012) | 40 matched NM; 198 PCa, macrodissected frozen tissue; divided into: non-recurrence (75), recurrence (123), clinical recurrence (80), BCR (43), systemic recurrence (36), and local recurrence (44) | Infinium HumanMethylation27 BeadChip (Illumina) | Differentially methylated genes for: tumor vs normal, 147; recurrence vs no recurrence, 75; clinical recurrence vs biochemical recurrence, 16; systemic recurrence vs local recurrence, 68 |
Yang et al. (2013) | Five normal (no cancer), four histologically NM (with PCa) | MeDIP-chip; ENCODE HG18 DNA methylation arrays (Roche NimbleGen) | 615 (537 hypomethylated+78 hypermethylated) probes differentially methylated in NM tissue (from PCa tissues) compared to normal prostate tissue. DNA methylation changes seen in NM tissue adjacent to tumor |
Microdissected tissue | |||
Validation: 26 histologically NM tissue from men with PCa |
5-aza, 5-aza-2′-deoxycytidine; aCGH, array comparative genomic hybridization; BCR, biochemical recurrence; BPH, benign prostatic hyperplasia; CRPC, castrate-resistant prostate cancer; DMH, differential methylation hybridization; FFPE, formalin fixed, paraffin embedded; GE, gene expression; LRES, long-range epigenetic silencing; MCA-RDA, methylated CpG island amplification coupled with representational difference analysis; MeDIP, methyl DNA immunoprecipitation using a 5MeC MAB; NGS, next-generation sequencing; NM, non-malignant; PCa, prostate cancer; PrEC, cultured non-malignant prostate epithelial cell line. If the information was available, details of the type of tissue used were also included.
Kobayashi et al. (2011) identified 87 CpG biomarkers in a large clinical prostate cancer cohort, almost all of which were hypermethylated in cancer (83/87; Table 2). Kim et al. (2011c) identified 1034 differentially methylated genes in prostate cancer (Table 2). The highest functional network identified from these genes was ‘Cancer, Cellular Movement, Hematological System Development and Function’, with TNFα as the central mediator. Despite methylation discovery being performed with a single pooled sample each for tumor and non-malignant, of ten genes with promoter hypermethylation selected for expression analysis in an independent patient cohort (n=17), most genes showed reduced expression, supporting the methylation results reported by Kim et al. (2011c).
Coolen et al. (2010) mined public gene expression databases to identify candidate regions of long-range epigenetic silencing (LRES; Table 2). Regions were classified as candidates for LRES if they i) contained four or more consecutive repressed or silent genes in two prostate cancer clinical data sets; ii) essentially lacked upregulated probes, and iii) had probes that were upregulated in at least 2/4 prostate cancer cell lines after 5-aza (Coolen et al. 2010). Genome wide methylation in LNCaP and PrEC was analyzed by MeDIP-chip, and results were presented for seven candidate regions. Hypermethylation was identified for a number of individual genes and the HOX-A and SERPINB gene families, with limited validation in clinical samples (Coolen et al. 2010). Kim et al. (2011a) identified 2481 cancer-specific methylation events in clinical cancer (Table 2) and coupled this with gene expression. Methylation, 5-aza re-expression, and chromatin mapping data from LNCaP and PrEC cells, together with clinical methylation and expression data, also demonstrated that promoter methylation regulates alternative transcript expression for the known cancer-related genes RASSF1, NDRG2, and APC (Kim et al. 2011a). A recent study not only identified differential DNA methylation changes between non-malignant and tumor tissue but also identified DNA methylation changes in histologically non-malignant tissue adjacent to tumor (Yang et al. 2013). This highlights the critical nature of sample choice, particularly with respect to non-malignant tissue for comparison. It also suggests that some DNA methylation changes are seen very early in the tumorigenesis process, or alternatively ‘spread’ from the tumor into the surrounding tissue, potentially promoting a suitable environment for tumor growth. Future studies should investigate these possibilities. The studies discussed here in general included limited validation of DNA methylation; both in terms of clinical cohorts and relationship with other genome wide data such as chromosome copy number, gene expression, and re-expression by demethylating agents. Future work should include more systematic validation of the genome wide DNA methylation analyses currently reported.
DNA methylation profiles in CRPC
Kim et al. (2011a) analyzed DNA methylation of four metastatic prostate cancer samples; however, results for these samples were grouped with other cancers making analysis difficult. Friedlander et al. (2012) integrated genome wide chromosome copy number, gene expression, and DNA methylation across metastatic CRPC compared with primary cancer and benign prostate. Sixteen genes had methylation and copy loss in the same tumor in 66% or more of samples, and methylation changes were more common than copy number alteration in CRPC. Moreover, androgen biosynthesis, the p53 pathway, and the insulin-like growth factor 1–protein kinase B signaling cascade were identified as pathways of genes frequently altered by aberrant copy number or DNA methylation changes (Friedlander et al. 2012). Androgen biosynthesis is a key mechanism for maintaining androgen signaling following androgen deprivation therapy (Cai & Balk 2011), and loss of p53 plays an important role in regulating AR activity across the genome (Guseva et al. 2012). Together, this re-iterates the potential of DNA methylation to identify novel therapeutic targets in advanced prostate cancer.
DNA methylation profiles in prostate cancer progression and recurrence
Kim et al. (2011a) reported that total promoter-associated CpG island methylation increases in parallel with prostate cancer progression but did not correlate this increased methylation with any specific individual genomic loci. The first genome wide DNA methylation study to identify individual CpGs associated with prostate cancer progression was reported by Kron et al. (2009), who performed DMH of pure primary gleason pattern 3 (PP3) and PP4. From the top 100 differentially methylated genes, BMP7 and HOXD3 were selected for analysis as potential biomarkers based on their putative and known biological functions. Further research could investigate the functional role of differentially methylated genes identified in this study and also identify whether combinations of these could act as biomarkers of disease progression.
Two subsequent studies have identified panels of differentially methylated CpGs associated with progression or recurrence in prostate cancer clinical cohorts. Kobayashi et al. (2011) identified 69 CpGs associated with time to biochemical recurrence. These CpGs were located in the promoters of both novel and known cancer-related genes; further study is required to investigate the function of genes without a previously identified role in cancer. In the same study, Gleason grade could not be distinguished by methylation profile. Similarly, Mahapatra et al. (2012) found 75 genes differentially methylated between recurrence and no recurrence, 68 genes differentially methylated between systemic recurrence and local recurrence, and 16 genes differentially methylated between clinical recurrence and biochemical recurrence. Overall, there were fewer genes methylated in patients with clinical recurrence compared with patients with only biochemical recurrence (Mahapatra et al. 2012). Validation of 14 genes in an independent cohort (n=20 per group) supported the genome wide data. However, the cohorts in each of these studies were small and validation is required in larger independent cohorts.
Similarities between DNA methylation profiles in breast and prostate cancers
We have performed a meta-analysis of the currently reported genome wide DNA methylation studies in breast and prostate cancers. We identified genes or gene families (HUGO Gene Nomenclature, http://www.genenames.org/) commonly targeted for aberrant DNA methylation as those that were identified in at least four different studies (Table 3). In performing this meta-analysis, we utilized the gene lists provided with the publication. We note that this analysis was biased by differences between studies, the biggest difference being whether data was just genome wide DNA methylation or whether genes were identified by other filtering processes such as combining DNA methylation with gene expression or CNV. From 22 breast cancer studies, we identified 425 genes that were commonly subject to aberrant methylation. Ninety-five genes were commonly aberrantly methylated in prostate cancer. The lower number of genes identified in prostate cancer is most likely due to the lower number of genome wide DNA methylation studies currently published for prostate cancer compared with breast cancer. The combined genes (466 unique genes) were analyzed by Ingenuity Pathway Analysis (IPA) Ingenuity Systems (www.ingenuity.com) and identified Embryonic Development, Organismal Development and Cancer as the top network (Fig. 1).
Genes identified as differentially methylated in breast and prostate cancers.
Breast cancer | Prostate cancer | |||||
---|---|---|---|---|---|---|
Genes | NM vs T | Prognosis | HR status | Cell line | NM vs T | Recurrence |
Homeobox gene/gene family | ||||||
HOXA | J, K, M | B | ||||
HOXB | E, L, M | B | B | 3 | ||
HOXC | E, L, M | 2, 4 | 8 | 3, 7 | ||
HOXD | D, E, J, L, M | 4 | 3, 7 | |||
IRX1 | D, L, M | 4 | 7 | |||
ISL | E, L, M | I | ||||
LBX | L | 4 | 7 | |||
LHX | D, G, H, L, M | 4 | 7, 8 | |||
MNX1 | 4 | 7 | ||||
NKX | D, L, M | B, C | 4 | 1 | 7 | |
PAX/OTX | D, E, L, M | |||||
POU | D, E, G, M | B, C | 8 | |||
SIX6 | D, E, M | 4 | 7, 8 | |||
TBX | E, M | I | ||||
VAX1 | 4 | 7 | ||||
Gene family | ||||||
ADAMTSL | E, G, K, M | B, E | ||||
B3G | E, J, M | B, E | 6 | |||
CD | E, K, M | A | B, I | 5 | 6 | 8 |
CDH | E, M | 6 | 6 | |||
CDKN | E, G, L, M | I | 4, 5 | 6, 7 | ||
COL | E, K, M | B | E, I | |||
CXC | G, K, M | C | 6 | |||
CYB | E | I | 6, 8 | |||
ELF | C | 6, 8 | ||||
EPH | E, K, M | 6 | ||||
FGF | M | I | 6, 8 | |||
FOX | D, E, L, M | 8 | ||||
GAS | M | B | 6, 8 | 8 | ||
GPR | E, G, M | A, C | 5 | 6 | ||
HIST1 | E, J, M | 6 | ||||
IL | E, M | C | I | |||
ITG | M | C | B, I | 6 | ||
KCN | D, E, M | B, C | I | 6 | ||
KLF | D, E, G, L, M | |||||
NEURO | E, M | 4 | 8 | |||
PCDH | D, E, G, L, M | |||||
PPP | L, M | A | I | 2 | 6 | |
PRDM | D, F, K, L, M | |||||
RAB | D, E, G, K, M | C | 6, 8 | |||
RASSF | E, M | 4, 5 | ||||
RUNX | E, M | C | B | 4, 5 | 7 | 8 |
S100 | M | C | 6 | 6 | ||
SEMA | E, L, M | I | ||||
SFRP | E, M | C, E | ||||
SLC | D, E, F, G, J, K, M, | A, C | 6, 8 | 6, 8 | ||
ST8 | E, L, M | C | ||||
SYN | E, J, M | I | 6 | 8 | ||
TMEM | D, E, M | I | 6 | 6, 8 | ||
TRIM | D, E, G, M | C | I | |||
ZBTB | J, K, M | I | ||||
ZNF | D, E, G, J, K, L, M | C, E | E, I | 6, 8 | 6, 8 | |
Gene | ||||||
ACADL | E, M | E | B | |||
APC | E, M | 4, 5 | ||||
GSTP1 | E, M | C | 4, 5 | 6, 8 | ||
HIF3A | E | 6, 8 | ||||
GPR180 | E, M | E | E | |||
MAT1A | E, M | E | 6 | |||
NFIX | D, G, H, K | |||||
RECK | E, M | C, E | B | |||
TNFRSF10D | E, M | E | 4 | 8 | ||
WT1 | D, H, L, M | 4, 5 | 7 |
Cell line comparison, NM vs T; HR, hormone receptor status; NM, non-malignant; T, tumor. A, Dedeurwaerder et al. (2011); B, Fackler et al. (2011); C, Fang et al. (2011); D, Faryna et al. (2012); E, Hill et al. (2011); F, Hu et al. (2005); G, Kamalakaran et al. (2011); H, Kim et al. (2012); I, Li et al. (2010); J, Novak et al. (2008); K, Ordway et al. (2007); L, Tommasi et al. (2009); M, Van der Auwera et al. (2010). 1, Chung et al. (2008); 2, Coolen et al. (2010); 3, Friedlander et al. (2012); 4, Kim et al. (2011a); 5, Kim et al. (2011c); 6, Kobayashi et al. (2011); 7, Kron et al. (2009); 8, Mahapatra et al. (2012).
Surprisingly, there were 54 genes in common between the breast and prostate cancer gene lists (Fig. 2). A small number of these genes identified as aberrantly methylated in both breast and prostate cancers have previously been shown to be methylated in breast, prostate, and other cancers: APC, CDKN2A, GSTP1, RASSF1, RUNX3, S100A2, SFRP1, and WT1 (Table 4). However, a large number of these genes have not been previously shown to be methylated in breast and prostate cancers or cancer in general, including genes from the SLC family. Interestingly, the only study to investigate both breast and prostate cancers identified four genes to be hypermethylated in both cancers; one of these four genes was a member of the SLC family (Chung et al. 2008). Further study is required to determine the functional effects of these proteins on prostate and breast cancer and to validate these results in other clinical cohorts. Using the Gene Functional Classification Tool from the DAVID Bioinformatics Resources, there was an enrichment in this gene list for homeobox genes (19/54; Table 4; Huang da et al. 2009a,b). There was also enrichment of genes on chromosome 7 and this is due to the homeobox A cluster on 7p15.2 (HOXA1, HOXA10, HOXA11, HOXA13, HOXA4, HOXA5, HOXA7, and HOXA9). Interestingly, despite clearly identified CpG islands (Fig. 3), there is limited evidence for aberrant methylation of these enriched genes in cancers other than breast or prostate, identifying them as potential hormonal cancer-specific genes. This requires future study to clarify. Homeobox genes are transcription factors known to have an important key role in early embryonic development, but more recently, they have been associated with disease including cancer.
Genes aberrantly methylated in breast and prostate cancers.
APC | FOXE3 | HOXA1 | HOXD4 | MAT1A | SIX6 |
CD48 | GAS6 | HOXA10 | HOXD9 | NEUROG1 | SLC34A2 |
CD8A | GAS7 | HOXA11 | IRX1 | NKX2–5 | SLC9A3 |
CDH13 | GPR150 | HOXA13 | ITGA11 | POU3F3 | SYN2 |
CDH5 | GPR62 | HOXA4 | KCNK9 | RAB34 | TMEM74 |
CDKN1C | GPR83 | HOXA5 | KCNQ1 | RASSF1 | TNFRSF10D |
CDKN2A | GSTP1 | HOXA7 | KLK10 | RUNX3 | WT1 |
CXCL1 | HIF3A | HOXA9 | LBX1 | S100A2 | ZNF154 |
FGF6 | HIST1H3A | HOXD3 | LHX9 | SFRP1 | ZNF80 |
Homeobox genes are in bold.
Genes commonly subject to aberrant DNA methylation in breast cancer
Numerous breast cancer genome wide DNA methylation studies identified homeobox genes as differentially methylated between non-malignant and tumor or associated with hormone receptor status or breast cancer prognosis (Ordway et al. 2007, Novak et al. 2008, Tommasi et al. 2009, Van der Auwera et al. 2010, Fackler et al. 2011, Fang et al. 2011, Hill et al. 2011, Kamalakaran et al. 2011; Table 3). Other gene families that were commonly identified as differentially methylated between non-malignant and tumor included transcription factors (FOX, KLF, PRDM, ZBTB, and ZNF) and gene families involved in cell transport of proteins or vesicles (RAB and SLC) or involvement in cell adhesion (CDH and PCDH) (Table 3; Hu et al. 2005, Ordway et al. 2007, Novak et al. 2008, Tommasi et al. 2009, Van der Auwera et al. 2010, Hill et al. 2011, Kamalakaran et al. 2011). The ST8 family of ovarian cancer tumor suppressor genes was commonly identified in non-malignant vs tumor and prognosis as was the RECK tumor suppressor gene (Table 3). IPA of the genes methylated in breast cancer identified some biological functions, such as posttranslational modification and lipid metabolism, that were only identified in breast cancer and not in prostate cancer. This may be due to the differences in the gene list size between breast and prostate cancers. In terms of the canonical pathways, the top ones for breast cancer-specific genes were transcriptional regulatory network in embryonic stem cells and caveolar-mediated endocytosis signaling. Given that the pathways and gene families do not appear to have a strong link to hormone metabolism or signaling, it is likely that these genes are not drivers of cancer but rather are secondary events that occur as part of the tumorigenic process.
Genes identified as commonly targeted by DNA methylation in prostate cancer
There was substantial overlap between different studies of genome wide DNA methylation in prostate cancer (Table 3). GSTP1 frequently exhibits prostate cancer-specific gene promoter methylation (Goering et al. 2012), and this was reflected in the studies reviewed here (Table 3). The homeobox or T-box gene families were commonly identified in cancer compared to non-malignant and during progression (Kron et al. 2009). Despite the limitations of this study introduced using pooled lymphocytes as a reference, many homeobox genes were also identified in other studies (Kim et al. 2011a, Friedlander et al. 2012). Where studies had more similarities, there was greater overlap. For example, in prostate cancer vs non-malignant, 16/25 genes identified by Mahapatra et al. (2012) overlapped with those identified by Kobayashi et al. (2011). Similarly, there were many common genes in studies where cell lines were used. This suggests that the degree of variation in methylation between samples is sufficiently small that genome wide DNA methylation analysis techniques can reliably and robustly detect methylation changes. The degree of similarity observed also reflects the variability within each disease state, with less overlap identified between studies of prostate cancer progression and recurrence than for cell line or cancer vs non-malignant. IPA identified biological functions associated with genes methylated in prostate cancer (Inflammatory Response and Immune Cell Trafficking) that were not evident with genes from breast cancer. The top canonical pathways for prostate cancer were bladder cancer signaling and Wnt/β-catenin signaling.
Future directions
While this review demonstrates that genome wide DNA methylation analysis in breast and prostate cancers can provide novel data and identify gene pathways not usually associated with these cancers, further study is required. First, it is clear that studies that integrate multiple data sources will produce the highest quality targets. Additional studies that integrate genome wide DNA methylation, gene expression, CNV, and analysis of other histone modifications are essential in prostate cancer, and to a lesser extent, breast cancer. These studies must be performed in appropriately powered clinical cohorts with the relevant patient samples. Studies of this nature should produce an unbiased, high-quality ‘roadmap’ of all sites in the genome that are under the influence of DNA methylation. Secondly, a meta-analysis of currently reported studies, using the raw data, should be conducted. While the meta-analysis presented here demonstrates that there are novel and overlapping genes/gene families subject to aberrant DNA methylation in breast and prostate cancers, careful comparison of this data with genome wide DNA methylation data from other cancers is essential. This will minimize variation between studies resulting from inconsistent statistical analyses and generate an unbiased list of candidate genes, some of which are likely to be breast and/or prostate specific, making them good candidates for being drivers of disease. Drivers of disease are also potential therapeutic candidates for late-stage breast or prostate cancer, for which there are currently limited therapeutic options.
Conclusion
Genome wide methylation studies in breast and prostate cancers have identified in an unbiased manner genes and gene families with aberrant methylation in non-malignant vs tumor, or with cancer progression or associated with hormone receptor status. This demonstrates the potential of these techniques to enhance our understanding of what controls gene expression in hormonally regulated cancers. Our analysis of the genes identified with aberrant methylation in breast and prostate cancers clearly showed that there was substantial overlap between genes identified across different studies, both within and between cancer types. The observed overlap between breast and prostate cancers may be a result of common underlying hormonal growth pathways, although this requires further analyses.
In conclusion, the studies reviewed here demonstrate the potential of identifying epigenetically regulated genes that are involved in cancer initiation or progression, and therefore biomarkers of prognosis and diagnosis, and that may be targets for epigenetic-modifying agents. However, further validation in additional clinical cohorts is required to confirm the targets of aberrant methylation in breast and prostate cancers. In addition, methylation results need to be linked to alterations in gene expression, chromosome copy number alterations, and other epigenetic marks, all of which work together to contribute to the transcriptional picture of a cell. Only with a fully integrated, genome wide analysis, can the downstream effects of these genetic and epigenetic changes on breast and prostate cancer cells be determined. In doing so, we will identify the key drivers in hormonal tumorigenesis and elucidate key targets for chemoprevention and therapeutics.
Declaration of interest
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the review reported.
Funding
This work was supported by grants from the Prostate Cancer Foundation of Australia Young Investigator Grant (T K Day, YIG03); The United States Department of Defense Prostate Cancer Research Program Training Fellowship (T K Day, PC080400); The University of Adelaide Barbara Kidman Fellowship (T K Day); W Bruce Hall Cancer Council of South Australia Research Fellowship (T Bianco-Miotto); Cancer Council SA and SAHMRI Beat Cancer Project (T Bianco-Miotto, APP1030945). Views and opinions of the authors do not necessarily reflect those of the USA Army, the Department of Defense or the Cancer Council of SA.
Acknowledgements
The authors gratefully acknowledge Dylan McCullough for assistance with this mansucript.
References
Baylin SB & Ohm JE 2006 Epigenetic gene silencing in cancer – a mechanism for early oncogenic pathway addiction? Nature Reviews. Cancer 6 107–116. (doi:10.1038/nrc1799).
Baylin SB & Jones PA 2011 A decade of exploring the cancer epigenome – biological and translational implications. Nature Reviews. Cancer 11 726–734. (doi:10.1038/nrc3130).
Cai C & Balk SP 2011 Intratumoral androgen biosynthesis in prostate cancer pathogenesis and response to therapy. Endocrine-Related Cancer 18 R175–R182. (doi:10.1530/ERC-10-0339).
Cancer Genome Atlas N Comprehensive molecular portraits of human breast tumours Nature 490 2012 61–70. (doi:10.1038/nature11412).
Chan MF, Liang G & Jones PA 2000 Relationship between transcription and DNA methylation. Current Topics in Microbiology and Immunology 249 75–86. (doi:10.1007/978-3-642-59696-4_5).
Chari R, Lockwood WW, Coe BP, Chu A, Macey D, Thomson A, Davies JJ, MacAulay C & Lam WL 2006 SIGMA: a system for integrative genomic microarray analysis of cancer genomes. BMC Genomics 7 324. (doi:10.1186/1471-2164-7-324).
Chari R, Coe BP, Vucic EA, Lockwood WW & Lam WL 2010 An integrative multi-dimensional genetic and epigenetic strategy to identify aberrant genes and pathways in cancer. BMC Systems Biology 4 67. (doi:10.1186/1752-0509-4-67).
Chiam K, Ricciardelli C & Bianco-Miotto T Epigenetic biomarkers in prostate cancer: current and future uses Cancer Letters 2012 [in press] doi:10.1016/j.canlet.2012.02.011).
Chin SP, Dickinson JL & Holloway AF 2011 Epigenetic regulation of prostate cancer. Clinical Epigenetics 2 151–169. (doi:10.1007/s13148-011-0041-7).
Chung W, Kwabi-Addo B, Ittmann M, Jelinek J, Shen L, Yu Y & Issa JP 2008 Identification of novel tumor markers in prostate, colon and breast cancer by unbiased methylation profiling. PLoS ONE 3 e2079. (doi:10.1371/journal.pone.0002079).
Coolen MW, Stirzaker C, Song JZ, Statham AL, Kassir Z, Moreno CS, Young AN, Varma V, Speed TP & Cowley M et al. 2010 Consolidation of the cancer genome into domains of repressive chromatin by long-range epigenetic silencing (LRES) reduces transcriptional plasticity. Nature Cell Biology 12 235–246. (doi:10.1038/ncb2023).
Dedeurwaerder S, Desmedt C, Calonne E, Singhal SK, Haibe-Kains B, Defrance M, Michiels S, Volkmar M, Deplus R & Luciani J et al. 2011 DNA methylation profiling reveals a predominant immune component in breast cancers. EMBO Molecular Medicine 3 726–741. (doi:10.1002/emmm.201100801).
Esteller M 2005 Aberrant DNA methylation as a cancer-inducing mechanism. Annual Review of Pharmacology and Toxicology 45 629–656. (doi:10.1146/annurev.pharmtox.45.120403.095832).
Fackler MJ, Umbricht CB, Williams D, Argani P, Cruz LA, Merino VF, Teo WW, Zhang Z, Huang P & Visvananthan K et al. 2011 Genome-wide methylation analysis identifies genes specific to breast cancer hormone receptor status and risk of recurrence. Cancer Research 71 6195–6207. (doi:10.1158/0008-5472.CAN-11-1630).
Fang F, Turcan S, Rimner A, Kaufman A, Giri D, Morris LG, Shen R, Seshan V, Mo Q & Heguy A et al. 2011 Breast cancer methylomes establish an epigenomic foundation for metastasis. Science Translational Medicine 3 75ra25. (doi:10.1126/scitranslmed.3001875).
Faryna M, Konermann C, Aulmann S, Bermejo JL, Brugger M, Diederichs S, Rom J, Weichenhan D, Claus R & Rehli M et al. 2012 Genome-wide methylation screen in low-grade breast cancer identifies novel epigenetically altered genes as potential biomarkers for tumor diagnosis. FASEB Journal 26 4937–4950. (doi:10.1096/fj.12-209502).
Flanagan JM, Cocciardi S, Waddell N, Johnstone CN, Marsh A, Henderson S, Simpson P, da Silva L, Khanna K & Lakhani S et al. 2010 DNA methylome of familial breast cancer identifies distinct profiles defined by mutation status. American Journal of Human Genetics 86 420–433. (doi:10.1016/j.ajhg.2010.02.008).
Friedlander TW, Roy R, Tomlins SA, Ngo VT, Kobayashi Y, Azameera A, Rubin MA, Pienta KJ, Chinnaiyan A & Ittmann MM et al. 2012 Common structural and epigenetic changes in the genome of castration-resistant prostate cancer. Cancer Research 72 616–625. (doi:10.1158/0008-5472.CAN-11-2079).
Goering W, Kloth M & Schulz WA 2012 DNA methylation changes in prostate cancer. Methods in Molecular Biology 863 47–66. (doi:10.1007/978-1-61779-612-8_4).
Guseva NV, Rokhlin OW, Bair TB, Glover RB & Cohen MB 2012 Inhibition of p53 expression modifies the specificity of chromatin binding by the androgen receptor. Oncotarget 3 183–194.
Hake SB, Xiao A & Allis CD 2004 Linking the epigenetic ‘language’ of covalent histone modifications to cancer. British Journal of Cancer 90 761–769. (doi:10.1038/sj.bjc.6601575).
Hill VK, Ricketts C, Bieche I, Vacher S, Gentle D, Lewis C, Maher ER & Latif F 2011 Genome-wide DNA methylation profiling of CpG islands in breast cancer identifies novel genes associated with tumorigenicity. Cancer Research 71 2988–2999. (doi:10.1158/0008-5472.CAN-10-4026).
Hu M, Yao J, Cai L, Bachman KE, van den Brule F, Velculescu V & Polyak K 2005 Distinct epigenetic changes in the stromal cells of breast cancers. Nature Genetics 37 899–905. (doi:10.1038/ng1596).
Huang Y, Nayak S, Jankowitz R, Davidson NE & Oesterreich S 2011 Epigenetics in breast cancer: what's new? Breast Cancer Research 13 225. (doi:10.1186/bcr2925).
Huang da W, Sherman BT & Lempicki RA 2009a Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Research 37 1–13. (doi:10.1093/nar/gkn923).
Huang da W, Sherman BT & Lempicki RA 2009b Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nature Protocols 4 44–57. (doi:10.1038/nprot.2008.211).
Huynh KT, Chong KK, Greenberg ES & Hoon DS 2012 Epigenetics of estrogen receptor-negative primary breast cancer. Expert Review of Molecular Diagnostics 12 371–382. (doi:10.1586/erm.12.26).
Jaenisch R & Bird A 2003 Epigenetic regulation of gene expression: how the genome integrates intrinsic and environmental signals. Nature Genetics 33 (Suppl) 245–254. (doi:10.1038/ng1089).
Jeronimo C, Bastian PJ, Bjartell A, Carbone GM, Catto JW, Clark SJ, Henrique R, Nelson WG & Shariat SF 2011 Epigenetics in prostate cancer: biologic and clinical relevance. European Urology 60 753–766. (doi:10.1016/j.eururo.2011.06.035).
Jovanovic J, Ronneberg JA, Tost J & Kristensen V 2010 The epigenetics of breast cancer. Molecular Oncology 4 242–254. (doi:10.1016/j.molonc.2010.04.002).
Kamalakaran S, Varadan V, Giercksky Russnes HE, Levy D, Kendall J, Janevski A, Riggs M, Banerjee N, Synnestvedt M & Schlichting E et al. 2011 DNA methylation patterns in luminal breast cancers differ from non-luminal subtypes and can identify relapse risk independent of other clinical variables. Molecular Oncology 5 77–92. (doi:10.1016/j.molonc.2010.11.002).
Karpinets TV & Foy BD 2005 Tumorigenesis: the adaptation of mammalian cells to sustained stress environment by epigenetic alterations and succeeding matched mutations. Carcinogenesis 26 1323–1334. (doi:10.1093/carcin/bgi079).
Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, Zahler AM & Haussler D 2002 The human genome browser at UCSC. Genome Research 12 996–1006. (doi:10.1101/gr.229102).
Kim JH, Dhanasekaran SM, Prensner JR, Cao X, Robinson D, Kalyana-Sundaram S, Huang C, Shankar S, Jing X & Iyer M et al. 2011a Deep sequencing reveals distinct patterns of DNA methylation in prostate cancer. Genome Research 21 1028–1041. (doi:10.1101/gr.119347.110).
Kim JH, Kang HS, Kim TW & Kim SJ 2011b Differential methylation hybridization profiling identifies involvement of STAT1-mediated pathways in breast cancer. International Journal of Oncology 39 955–963. (doi:10.3892/ijo.2011.1075).
Kim SJ, Kelly WK, Fu A, Haines K, Hoffman A, Zheng T & Zhu Y 2011c Genome-wide methylation analysis identifies involvement of TNF-α mediated cancer pathways in prostate cancer. Cancer Letters 302 47–53. (doi:10.1016/j.canlet.2010.12.010).
Kim YJ, Yoon HY, Kim SK, Kim YW, Kim EJ, Kim IY & Kim WJ 2011d EFEMP1 as a novel DNA methylation marker for prostate cancer: array-based DNA methylation and expression profiling. Clinical Cancer Research 17 4523–4530. (doi:10.1158/1078-0432.CCR-10-2817).
Kim MS, Lee J, Oh T, Moon Y, Chang E, Seo KS, Hoehn BD, An S & Lee JH 2012 Genome-wide identification of OTP gene as a novel methylation marker of breast cancer. Oncology Reports 27 1681–1688. (doi:10.3892/or.2012.1691).
Kobayashi Y, Absher DM, Gulzar ZG, Young SR, McKenney JK, Peehl DM, Brooks JD, Myers RM & Sherlock G 2011 DNA methylation profiling reveals novel biomarkers and important roles for DNA methyltransferases in prostate cancer. Genome Research 21 1017–1027. (doi:10.1101/gr.119487.110).
Kron K, Pethe V, Briollais L, Sadikovic B, Ozcelik H, Sunderji A, Venkateswaran V, Pinthus J, Fleshner N & van der Kwast T et al. 2009 Discovery of novel hypermethylated genes in prostate cancer using genomic CpG island microarrays. PLoS ONE 4 e4830. (doi:10.1371/journal.pone.0004830).
Laird PW 2010 Principles and challenges of genomewide DNA methylation analysis. Nature Reviews. Genetics 11 191–203. (doi:10.1038/nrg2732).
Li J, Gao F, Li N, Li S, Yin G, Tian G, Jia S, Wang K, Zhang X & Yang H et al. 2009 An improved method for genome wide DNA methylation profiling correlated to transcription and genomic instability in two breast cancer cell lines. BMC Genomics 10 223. (doi:10.1186/1471-2164-10-223).
Li L, Lee KM, Han W, Choi JY, Lee JY, Kang GH, Park SK, Noh DY, Yoo KY & Kang D 2010 Estrogen and progesterone receptor status affect genome-wide DNA methylation profile in breast cancer. Human Molecular Genetics 19 4273–4277. (doi:10.1093/hmg/ddq351).
Mahapatra S, Klee EW, Young CY, Sun Z, Jimenez RE, Klee GG, Tindall DJ & Donkena KV 2012 Global methylation profiling for risk prediction of prostate cancer. Clinical Cancer Research 18 2882–2895. (doi:10.1158/1078-0432.CCR-11-2090).
Miyamoto K & Ushijima T 2005 Diagnostic and therapeutic applications of epigenetics. Japanese Journal of Clinical Oncology 35 293–301. (doi:10.1093/jjco/hyi088).
Nafee TM, Farrell WE, Carroll WD, Fryer AA & Ismail KM 2008 Epigenetic control of fetal gene expression. BJOG: an International Journal of Obstetrics and Gynaecology 115 158–168. (doi:10.1111/j.1471-0528.2007.01528.x).
Nakao M 2001 Epigenetics: interaction of DNA methylation and chromatin. Gene 278 25–31. (doi:10.1016/S0378-1119(01)00721-1).
Naume B, Zhao X, Synnestvedt M, Borgen E, Russnes HG, Lingjaerde OC, Stromberg M, Wiedswang G, Kvalheim G & Karesen R et al. 2007 Presence of bone marrow micrometastasis is associated with different recurrence risk within molecular subtypes of breast cancer. Molecular Oncology 1 160–171. (doi:10.1016/j.molonc.2007.03.004).
Neve RM, Chin K, Fridlyand J, Yeh J, Baehner FL, Fevr T, Clark L, Bayani N, Coppe JP & Tong F et al. 2006 A collection of breast cancer cell lines for the study of functionally distinct cancer subtypes. Cancer Cell 10 515–527. (doi:10.1016/j.ccr.2006.10.008).
Novak P, Jensen T, Oshiro MM, Watts GS, Kim CJ & Futscher BW 2008 Agglomerative epigenetic aberrations are a common event in human breast cancer. Cancer Research 68 8616–8625. (doi:10.1158/0008-5472.CAN-08-1419).
Ordway JM, Bedell JA, Citek RW, Nunberg A, Garrido A, Kendall R, Stevens JR, Cao D, Doerge RW & Korshunova Y et al. 2006 Comprehensive DNA methylation profiling in a human cancer genome identifies novel epigenetic targets. Carcinogenesis 27 2409–2423. (doi:10.1093/carcin/bgl161).
Ordway JM, Budiman MA, Korshunova Y, Maloney RK, Bedell JA, Citek RW, Bacher B, Peterson S, Rohlfing T & Hall J et al. 2007 Identification of novel high-frequency DNA methylation changes in breast cancer. PLoS ONE 2 e1314. (doi:10.1371/journal.pone.0001314).
Ruike Y, Imanaka Y, Sato F, Shimizu K & Tsujimoto G 2010 Genome-wide analysis of aberrant methylation in human breast cancer cells using methyl-DNA immunoprecipitation combined with high-throughput sequencing. BMC Genomics 11 137. (doi:10.1186/1471-2164-11-137).
Salozhin SV, Prokhorchuk EB & Georgiev GP 2005 Methylation of DNA – one of the major epigenetic markers. Biochemistry 70 525–532. (doi:10.1007/s10541-005-0146-8).
Sandoval J & Esteller M 2012 Cancer epigenomics: beyond genomics. Current Opinion in Genetics & Development 22 50–55. (doi:10.1016/j.gde.2012.02.008).
Sproul D, Nestor C, Culley J, Dickson JH, Dixon JM, Harrison DJ, Meehan RR, Sims AH & Ramsahoye BH 2011 Transcriptionally repressed genes become aberrantly methylated and distinguish tumors of different lineages in breast cancer. PNAS 108 4364–4369. (doi:10.1073/pnas.1013224108).
Sun Z, Asmann YW, Kalari KR, Bot B, Eckel-Passow JE, Baker TR, Carr JM, Khrebtukova I, Luo S & Zhang L et al. 2011 Integrated analysis of gene expression, CpG island methylation, and gene copy number in breast cancer cells by deep sequencing. PLoS ONE 6 e17490. (doi:10.1371/journal.pone.0017490).
Szyf M 2005 DNA methylation and demethylation as targets for anticancer therapy. Biochemistry 70 533–549. (doi:10.1007/s10541-005-0147-7).
Szyf M, Pakneshan P & Rabbani SA 2004 DNA demethylation and cancer: therapeutic implications. Cancer Letters 211 133–143. (doi:10.1016/j.canlet.2004.04.009).
Tommasi S, Karm DL, Wu X, Yen Y & Pfeifer GP 2009 Methylation of homeobox genes is a frequent and early epigenetic event in breast cancer. Breast Cancer Research 11 R14. (doi:10.1186/bcr2233).
Van der Auwera I, Yu W, Suo L, Van Neste L, van Dam P, Van Marck EA, Pauwels P, Vermeulen PB, Dirix LY & Van Laere SJ 2010 Array-based DNA methylation profiling for breast cancer subtype discrimination. PLoS ONE 5 e12616. (doi:10.1371/journal.pone.0012616).
Waddell N, Arnold J, Cocciardi S, da Silva L, Marsh A, Riley J, Johnstone CN, Orloff M, Assie G & Eng C et al. 2010 Subtypes of familial breast tumours revealed by expression and copy number profiling. Breast Cancer Research and Treatment 123 661–677. (doi:10.1007/s10549-009-0653-1).
Yan PS, Chen CM, Shi H, Rahmatpanah F, Wei SH, Caldwell CW & Huang TH 2001 Dissecting complex epigenetic alterations in breast cancer using CpG island microarrays. Cancer Research 61 8375–8380.
Yang B, Bhusari S, Kueck J, Weeratunga P, Wagner J, Leverson G, Huang W & Jarrard DF 2013 Methylation profiling defines an extensive field defect in histologically normal prostate tissues associated with prostate cancer. Neoplasia 15 399–408. (doi:10.1593/neo.13280).
Zheng S & Zhao Z 2012 DNA methylation profiling distinguishes three clusters of breast cancer cell lines. Chemistry & Biodiversity 9 848–856. (doi:10.1002/cbdv.201100354).