DNA methylation changes in epithelial ovarian cancer histotypes.

Two predominant global methylation patterns in cancer, including EOC, have emerged. These are (1) hypermethylation of promoters leading to gene silencing facilitating cancer formation, and (2) hypomethylation of highly repeated DNA sequences leading to aberrant expression of oncogenic genes. In keeping with the paradigm that each EOC histological type represents different disease, hypermethylation patterns are histotype specific. To highlight the differences in these patterns, and stress the importance of analyzing these tumors in a histology-informed fashion, DNA hypermethylation will be reviewed by histological type. Table 1 summarizes global and focal DNA methylation events observed for each histological type and the common genetic features of these tumors. In contrast to hypermethylation, histological type appears to play no major role in global DNA hypomethylation patterns. This is in keeping with the hypothesis that DNA hypomethylation is a ubiquitous feature of cancers [ 36 ]. In all EOC histologies, hypomethylation of repeat elements (satellite DNA, LINEs, ALu) increases in tissues from non-neoplastic precursors toward cancers [ 37 – 42 ]. Increasing hypomethylation is associated with advanced stage, grade, and poor prognosis [ 38 , 42 ]. With few exceptions, there is no correlation between DNA hypomethylation and hypermethylation events [ 38 , 43 ]. Clinicopathologic features of HGS and LGS have been reviewed previously [ 44 ]. Genetic features of HGS tumors are mutations, germline and somatic, in homologous recombination genes, typically BRCA1 and BRCA2, and near ubiquitous somatic mutation in TP53 as demonstrated by the TCGA and others [ 45 – 47 ]. Many HGS tumors have complex karyotypes suggestive of a period of massive genomic instability [ 48 , 49 ]; however, they have very few focal mutational events relative to many other cancers [ 50 ]. Epigenetically, the defining DNA methylation pattern is a lack of hypermethylation relative to the other EOC histotypes. The TCGA surveyed 519 HGS tumors using Illumina HumanMethylation27k Beadchips and found 168 genes that were aberrantly hypermethylated and had reduced gene expression [ 49 ]. REB25, AMT, CCL21, and SPARCL1 were noted for being frequently targeted for methylation. Patch et al. similarly studied 80 HGS tumors using Illumina HumanMethylation450k Beadchips, and found 433 hypermethylated genes with reduced gene expression [ 48 ]. Surprisingly, only 5 genes (ALDH1A3, AMT, LONRF2, NPDC1, and SLC16A5) were concordant (1–3% agreement) between these two studies, despite their larger sample sizes and similar methodological and analytical approach. One explanation for this is that TCGA hypothesized that epigenetic silencing of different genes occurs at varying frequencies in tumors, and thus focused their methylation analyses on those 10% of tumors with the highest levels of DNA methylation per CpG site, effectively comparing the 90th percentile of tumors to the mean of fallopian tube samples [ 49 ]. Meanwhile, Patch et al. looked for hyper-methylation in all of their HGS tumors versus fallopian tube samples, and effectively screened for sites that, on average, are hyper-methylated in tumors [ 48 ]. Only hyper-methylation events that are frequent in HGS tumors are likely to be found by both approaches, and it is possible that relatively few of these exist. Many smaller studies of fewer HGS samples and CpG sites, often focusing on known or suspected tumor suppressor genes (TSGs), have likewise reported lists of hypermethylated sites (ranging from 50 to 500) [ 39 , 51 – 53 ]. Again there is poor overlap in the reported genes, and inconsistency in the estimated frequency of tumors found to be hypermethylated. At least two studies of hypermethylation of TSGs and ovarian cancer related genes found that HGS tumors are not hypermethylated relative to normal fallopian tube [ 39 , 51 ]. Clustering analyses performed based on DNA methylation data in HGS tumors have failed to yield stable clusters that distinguish tumors based on biological and clinical characteristics such as tumor vs normal [ 54 ], primary versus recurrent disease [ 52 ], and expression signatures [ 49 ]. These data suggest that when hypermethylation occurs, it may do so quite randomly, and driver hypermethylation events may not play a major role in HGS biology. LGS EOC is genetically and epigenetically different from HGS. In this category of tumor, mutations to genes in the Ras pathway are common (KRAS, BRAF, NRAS, ERBB2, and PTEN); mutations to homologous recombination genes and TP53 are not [ 44 , 55 ]. Relative to HGS, these tumors are chromosomally stable, and thought to arise from benign or borderline tumors, progress through the more proliferative serous borderline tumor (SBT), and finally transform into invasive LGS [ 56 ]. Epigenetically, LGS tumors are closely related to SBTs, and distinct from HGS. The overall pattern is one of more hypermethylation in LGS tumors [ 52 ]. Shih et al. compared the methylation profiles of LGS, SBT, and HGS tumors using Illumina’s GoldenGate arrays (1505 CpGs), and found MAPK4, HOXA9, AATK, WNT5A, and GFI1 to be frequently hypermethylated in LGS versus SBTs; TUBB3, TSG101, HDAC6, DBC1 and GPATC3 were frequently hypomethylated [ 52 ]. Ziller et al. subsequently compared LGS and SBT methylation patterns using Illumina HumanMethylation27k Beadchips [ 57 ], and found 383 hypermethylated genes and 340 hypomethylated genes. Only one gene, NF3, was hypermethylated in LGS versus SBT. In the future, more studies with greater numbers of LGS samples and genome-wide methylation data will be needed to replicate these patterns. Ovarian clear cell (CC) carcinoma has distinct clinical characteristics [ 58 ]. Unlike HGS, development is associated with endometriosis [ 11 ], and the hypothesized precursor tissues are endometrium and endometriosis. Genetic features of this histotype are germline mutations in mismatch-repair (MMR) genes (MLH1, MSH, MSH6, PSM2) in familial cases (part of Lynch Syndrome), and somatic mutations in ARID1A, PTEN, and PIK3CA in sporadic cases [ 54 , 59 ]. Microsatellite instability (MSI) is typically observed in MMR deficient CC tumors. Unlike HGS EOC, chromosomal instability is not a feature of CC. Epigenetically, this histotype is characterized by wide-spread CpG Island promoter hyper-methylation [ 54 , 59 ], reminiscent of CIMP signatures described in clear cell renal cell carcinoma [ 60 ], colorectal carcinoma [ 49 ], and breast tumors [ 61 ]. We estimate that 71% of CC tumors have this CIMP signature ( Table 1 ) [ 59 , 62 ]. To our knowledge, a comparison of the CpG sites hypermethylated in CC-EOC versus other CIMP tumors has not been made. In keeping with a global pattern of hypermethylation of CGIs, CC-EOCs have greater hypermethylation of TSGs and ovarian-cancer related genes relative to HGS tumors and precursor tissues (normal endometrium and fallopian tube) [ 39 , 51 ]. And, earlier stage (I and II) tumors may actually exhibit greater TSG hypermethylation than later stage (III and IV) tumors. Thus, CIMP seems to be an early event in CC-EOC carcinogenesis, but also a dynamic one, and may be passively lost or actively reversed during tumor evolution. It is hypothesized that these widespread changes in DNA methylation arise due to the oxidative stress and/or inflammation experienced in the unique CC-EOC tumor microenvironment (i.e. endometrial tissues exposed to the ovary surface or in the oxygen free and radical enriched environment of endometriotic cysts) [ 54 , 63 , 64 ]. Determining driver epimutations in CIMP-associated cancers is an active area of study and most events are presumed to be “passengers”. In CC EOC, synchronous hypermethylation of genes (N 50) enriched in the estrogen receptor alpha pathway (ex ESR1, ESR2, WT1, TGIF2), and hypomethylation of genes enriched in the HNF1 pathway (ex HNF1B, HNF1a, SERPINA6) [ 59 , 65 ], have suggested that these are pathways of importance. However, the cancer-specificity of these patterns needs to be confirmed as CC precursor tissues ( Table 1 ) have not been comprehensively examined. CIMP status and HNF1B expression levels are strongly positively correlated in EOC (P b 10 −16 ) [ 59 ], but a hypothesis for causation is not forthcoming. Plausibly, these two molecular features are a byproduct of the CC-EOC microenvironment, overexpression of HNF1B being a mechanism by which CC-EOC precursors manage and survive their stressful environment [ 64 ], and CIMP arising in that environment by mechanisms that are not yet known. Endometrioid ovarian carcinoma (EN-EOC) shares similarities with CC-EOC, particularly an association with endometriosis ( Table 1 ). Like CC, germline mutations in MMR genes are observed in a subset of familial EN-EOC cases. Indeed, EOCs associated with Lynch Syndrome are predominantly of the EC histotype. Somatic mutations in ARID1A, PTEN, and PIK3CA are noted in sporadic cases [ 54 , 59 ]. Unlike CC, EC-EOC can be further divided into high-grade and low-grade disease, the former being HGS-like in its clinical characteristics and molecular features, and the latter being more CC-like [ 66 ]. This pattern may be recapitulated in the epigenetic profiles of ECs. Genome-wide DNA methylation datasets clearly support that some EN tumors cluster with CC-EOCs [ 65 ] and exhibit a CIMP signature ( Table 1 ), while others do not [ 62 , 67 ]. Targeted gene studies have also shown that similar to CC-EOC, LG and low-stage (I and II) EN-EOCs have greater TSG hypermethylation than high grade and high stage (III, VI) ENs [ 39 ]. Indeed, high grade EN-EOCs were indistinguishable from HGS in their lack of hypermethylation of TSGs. It is tempting to speculate that the patterns will be similar to CC-EOC; however, the EN-EOC histotype does not overexpress HNF1B [ 59 ]. Finally, ovarian endometrioid carcinomas share many molecular genetic features with their uterine counterparts [ 66 ]. Kolbe et al. extended these findings including similar patterns of DNA methylation (based on Illumina HumanMethylation27k Beadchips). They found that ovarian or endometrial EN tumors are highly similar to each other, displaying wide spread DNA methylation aberrations (affecting N 1000 genes), while HGS tumors are more similar to normal ovarian and endometrial tissues [ 67 ]. They conclude that the number of hypermethylated sites in EN-EOC suggests an underlying defect in DNA methylation pathways. No reports specifically addressing patterns of DNA hypermethylation in MC EOC have been published. However, it is common for analyses performed on ‘all’ EOCs to have some MC tumors (typically b 10) [ 65 , 68 , 69 ]. These studies demonstrate that while MC tumors may not have a CC-EOC like CIMP signature ( Table 1 ), they experience more frequent hypermethylation than HGS tumors. Indeed, when subjected to unsupervised consensus clustering, MC tumors group predominantly with CCC tumors [ 65 ], albeit as a subcluster. These results suggest that MC tumors acquire distinct hypermethylation events [ 70 ] which may in part reflect differences in the carcinogenic process, and/or differences in precursor tissue.

DNA methylation changes are also hypothesized to play a role in driving acquired resistance to chemotherapy. Platinum (carboplatin) combined with a taxane (paclitaxel) is the primary therapeutic regimen in EOC, regardless of histological type [ 71 , 72 ]. Carboplatin is incorporated into DNA, inducing formation of adducts and eliciting mismatch repair which, in turn, activates apoptosis [ 73 ]. Taxanes stabilize tubulin, leading to defective spindle formation, cell cycle arrest at G2/M, and apoptosis [ 74 , 75 ]. After an initial response to therapy, a majority of patients relapse, due to development of resistance to the therapy [ 76 ]. Resistance to chemotherapeutic regimens may be intrinsic; i.e. DNA repair mechanisms are impaired. Or, it may be acquired de novo due to the acquisition of mutations or other alterations such as DNA methylation during the course of therapy. While it is now recognized that there are specific histological subtypes of EOC that are distinct both genetically and epigenetically, many of the studies on DNA methylation and response to therapy do not consider this in the analysis. Many studies cite mainly serous tumors (~ 80–60%), but include all histotypes in the analysis [ 77 – 79 ], others do not specify histotype [ 80 – 82 ], while others cite no differences between histotypes but have very small numbers of samples with which to conclude this [ 83 – 85 , 160 ]. There are studies that have focused on the CC and ENEOC histotypes [ 86 – 88 ] and HGS EOC [ 48 ], and it is hoped that future studies continue to consider histotype in their analyses. However, given the variability we are unable to review resistance to chemotherapy by histotype, with a few exceptions. In addition, the approaches used to analyze this range from in vitro work that may show that efficacy of a drug is modulated by the methylation status of a gene, to assessing methylation and resistance to chemotherapy and outcome, a more indirect measure but one that is clinically applicable. There are several reports using genome wide approaches to examine methylation and response to therapy or outcome. Patch et al. characterized 114 chemoresistant HGS EOC from 92 patients using whole genome and transcriptome sequencing, Illumina 450K methylation arrays, and copy number analysis [ 48 ]. They reported several molecular events associated with acquired resistance, some of which involved DNA methylation. They noted a reversion of a methylated BRCA1 in one patient. They also describe 94 probes that had N 10% median methylation difference between the sensitive and chemoresistant groups, including probes in the promoter region of COLA1A1, AURKC, EPHA1, MAPKBP3, HOXB9, IGFL1, and SOX13. Bauerschlag et al. [ 77 ] used a 27 K methylation array to profile 20 mostly HGS advanced EOC cases. Longer survival was associated with hypermethylation in TMC5, PTPRN and GUCY2C, and hypomethylation in GREB1, TGIF and TOB1. Lum et al. [ 89 ] examined promoter methylation in 15 refractory/resistant and 12 late sensitive patient samples, using a custom 280 K feature microarray, identifying 296 genes of which 60% were methylated in resistant and 40% methylated in sensitive tumors. They tested these candidate genes in an in vitro carboplatin resistance assay using shRNA, identifying 19 genes (GSK3B, DOK2, APRT, OXSR1, CENBP, FZD1, ESRRA, HRIP3, GTF2b, SGPL1, GABPA, TWIST1, MDH1, NR2E1, NR3C2, SOZ9, UNG, and ZIC1) that distinguish between resistant and sensitive cell lines (HOSE 63 an immortalized human ovarian surface epithelium cell line, SKOV3 an E/CC EOC cell line and CAOV3 a serous EOC cell line). Wei et al. [ 78 ] profiled two sets of mostly HGS tumors, 19 in set 1 and 44 in set 2, using a CpG Island microarray. One hundred and eighty two methylated loci, in 40 genes, classified samples with reduced progression-free survival. The genome wide approach is agnostic, in that it is hypothesis generating rather than hypothesis testing. The methods employed in analysis in each of the reports above differ. Patch et al. is the largest study, while Lum et al. carried out further testing of candidate methylated genes and Wie et al. used two classification modules previously used in microarray gene expression analyses. These may account for the poor overlap in the genes significantly associated with resistance, as does the relatively small sample numbers used. What is notable however, is that these studies rarely find any of the commonly studied tumor suppressor genes or oncogenes in their top lists. Many studies have taken a targeted approach to studying acquired resistance, focusing on genes relevant to platinum and taxane, including DNA repair, apoptosis and cell cycle regulation. Many of these studies are limited by small sample numbers, admixture of histological subtypes, a lack of validation, and a paucity of data regarding expression. Methylation of BRCA1 is well documented, occurring in 9–20% of serous EOCs [ 47 , 85 , 90 – 92 ], and is reported to be associated with improved response to treatment in some [ 48 , 85 , 93 – 95 ] but not all studies [ 48 , 49 ]. These tumors may have an improved response to DNA damaging chemotherapy [ 96 , 97 ] and PARP inhibitors [ 98 , 99 ]. In contrast, events that restore BRCA1 expression are associated with acquired resistance to therapy [ 48 , 100 , 101 ]. Reports targeting genes that have activity within the mechanism underlying response to chemotherapy include MLH1, RASSF1A and PLK1. Methylation of MLH1 (involved in MMR, see Table 1 ) has been associated with resistance to platinum in EOC cell lines [ 102 – 104 ] and EOC [ 83 ], and is acquired during primary chemotherapy in some studies [ 81 , 105 ], but not others [ 94 , 95 ]. Defective MMR may result in a defective cisplatin induced G 2 arrest and loss of apoptotic response [ 104 , 106 ]. MMR may be present in the primary tumor, particularly the endometrial subtype [ 107 – 111 ] ( Table 1 ). Methylation of RASSF1A is relatively common and is seen across all EOC histologies [ 112 ], however it is most common in CC EOC [ 113 , 114 ]. RASSF1A binds to tubulin and stabilize spindles [ 115 ] and thus down regulation by methylation may mediate resistance to taxanes. However, while RASSF1A methylation has been shown to affect the response of cells exposed to taxol in vitro [ 75 ], it generally has not been associated with a clinical response to chemotherapy or survival [ 85 , 86 , 94 , 113 ]. PLK2 is a serine–threonine kinase which positively regulates progression through G 2 and is reported to be overexpressed in many cancers [ 116 ]. In EOC, methylation of PLK2 is associated with transcriptional silencing and may be associated with a risk of relapse following chemotherapy [ 117 , 118 ]. Other genes in which methylation has been implicated in response to therapy by targeted studies include HIN1, HERV-K, DLEC1 and CDH1 in CC EOC [ 88 , 113 ], and ESR1 [ 80 ], HSulf-1 [ 119 ], FBOXO32 [ 120 ], ABCA1 [ 82 ], SFRP [ 79 ], CABIN1 [ 121 ], TGBFB1 [ 122 , 123 ] and HOXA11 [ 84 ] in EOC. Many of the studies cited represent the only report for a given gene and while they generally present a clear rationale for the analyzed genes, when analyzed in the context of response seen in EOC patients, the data are less compelling. Several groups have used EOC cell lines to investigate epigenetic mechanisms underlying response to therapy [ 75 , 89 , 103 , 120 , 124 – 127 , 161 ]. Three studies used genome wide approaches [ 89 , 125 , 126 ], while the others targeted EZH2 [ 124 ], MLH1 [ 103 ], miRNA-145 [ 127 ] and RASSF1A [ 75 ]. Using the cell line A2780, that is sensitive to platinum therapy, Li et al. exposed cells to several cycles of cisplatin and performed array based methylation and gene expression analyses to analyze levels of methylation with each cycle [ 126 ]. They showed increasing methylation during development of resistance, with upregulation of the DNA methyltransferases DNMT1 and DNMT3B. Pathway analysis performed on 55 genes identified three down regulated pathways: cell adhesion molecules (ITFAV, CLDN11, NEO1, CDH2), tight junction (CLDN11, PPP2R4, INADL), PPAR signaling pathway (CPT1A, SLC27A6) and leukocyte transendothelial migration (CLDN11) and upregulation of six pathways including PIK3R3, PDGFRA, E2F1, and TGFBR2 genes. Treatment of resistant cells with decitabine, a DNA methyltransferase inhibitor (DNMTi) causing reduced DNA methylation genome-wide, was associated with increasing sensitivity to cisplatin. Zeller et al. [ 125 ] used the same cell line and a resistant line (2780-Ccp70), and performed chemoresensitization with decitabine and/or a histone deacetylase inhibitor (HDACi) to identify epigenetic changes in A2780-cp70. They measure DNA methylation using HumanMethylation27k Beadchips and gene expression using the Affymetrix U133 Plus 2.0 Array, and noted that only a fraction of the genes silenced in resistant cells were re-expressed following treatment with decitabine (ARHGD1B, PSMB9, HSPA1A, ARMCX2, MEST, FLNC, MLH1, MDK, GLUL, COLO1A, FLNA, NTS). With the addition of anHDACi, two more genes were re-expressed (SERPIMB2 and HIST1H2BF). Following up in three independent pairs (pre-therapy and at relapse) of in vivo derived serous EOC cell lines they confirmed acquired methylation of six of the 13 genes in the relapsed cell lines. Lastly, using seven pairs of matched primary and relapsed EOC tumors, 8/13 genes were found to have acquired methylation and experienced gene silencing at relapse (no histology noted). Three genes (ARMCX2, MEST, and MLH1) were found to have increased methylation in all three analyses. It should be noted that A7280 is not representative of HGS EOC; it most likely has an EN-OC origin [ 128 ]. Zeller et al. [ 125 ] did validate their findings in additional cell lines and samples, importantly, cell lines most relevant to HGS, the most common EOC histotype. As well, interpretation of these in vitro assays should be carried out with caution, as the tumor microenvironment may influence the response to therapies. As with genome wide studies of EOC above, there is little overlap although MLH1 was identified. Plumb et al. [ 103 ] examined MLH1 methylation status in MMR deficient A2789/cp70 xenografts treated with decitabine. Treatment sensitized the xenografts to chemotherapeutic agents, and showed a decrease in methylation of MLH1. Kassler et al. [ 75 ] used EOC cell lines A547 (does not express RASSF1A) and UCI-107 (taxol sensitive) [ 129 ] to examine the role of methylation of RASSF1A in taxol resistance. They show that loss of RASSF1A reduces the ability of taxol to promote microtubule polymerization. They also noted changes in RASSF1A expression in mRNA isolated from a small number of HGS tumors (n = 20); non-responders have reduced gene expression. However, as noted above, survival is not associated with methylation of this gene. SKOV3/PTX and A2780/PTX cell lines and xenografts, both CC histotype [ 128 ], were used to examine miR-145 methylation, which is down-regulated in many types of cancer including EOC [ 127 ]. Restoration of miR-145 expression by treatment with decitabine increased sensitivity to paclitaxel. Use of cell lines allows in depth analysis of the effects of methylation of specific genes, and also efficacy of decitabine and HDACi in restoring expression of a given gene. A2780 is the most commonly used; this and others such as SKOV3 do not have the molecular makeup of HGS EOC, but more closely resemble CC and EC EOC. While it may be argued that the response to a given therapy might be similar in different cell types, the genetic and epigenetic characteristics may also influence the response. Care should be taken to be cognizant of the different histologies. Translating findings from cell lines to clinical EOC will require greater numbers of samples. As expressed in Patch et al., in HGS EOC, the heterogeneity and marked adaptability of the cancer genome indicates that overcoming resistance to therapy will require many different approaches. This is probably true also for the other histotypes. Development of resistance is likely a complex process, and probably accounted for more than by a single gene.

Epigenomics is making its way into the clinical care of patients. Non-invasive methods to detect tumor DNA in the circulation system, “liquid biopsies”, are a very active area of development [ 130 ]. Circulating cell free DNA (cfDNA) shed from cells into the blood stream, and circulating intact tumor cells (CTC), are present in most cases with advanced neoplasia [ 131 , 132 ] and to a lesser extent in cases with localized disease [ 130 , 133 ]. Studies on EOC plasma methylation biomarkers have generally targeted specific genes, and a relatively small number of samples. Methylation of RASSF1A and BRCA1 promoters has been reported in plasma of EOC patients [ 134 – 136 ]. However, these genes have also been found methylated in some controls [ 137 – 140 ], highlighting the need for larger studies [ 130 ]. A clinical phase III trial examined methylation of MLH1 in plasma and tumor before carboplatin/taxoid chemotherapy and at relapse in 138 patients with EOC (mixed stages and histotypes not specified) [ 81 ]. Methylation increased at relapse; 25% had acquired methylation at relapse, which was associated with a poor overall survival. A multiplex methylation-specific PCR assay was used to assess the methylation status of seven genes (APC, RASSF1A, RUNX3, CDH1, TFP12, SRFP5, OPCML) in 202 serum samples comprising 62 healthy individuals, 53 benign ovarian tumor patients, 46 advanced stage EOC cases (no histotype) and 17 early stage cases. A methylation index, the number of methylated and unmethylated genes, of ≥ 1 was predictive of early stage EOC with 85.3% sensitivity and 90.5% specificity. Studies of SLIT2 [ 141 ] and OPMCL [ 142 ] have also been published. Both describe the presence of methylated genes in the serum of EOC patients, of mixed histologies and grade. Using a genome-wide approach, Teschendorff et al. [ 143 ] profiled peripheral blood DNA in 131 pre- and 113 post-treatment EOC patients (primary tumors) and 274 healthy controls. They identified age related and cancer specific DNA methylation signatures, both of which predicted tumor presence. Gene set enrichment analysis revealed four main categories of genes: hypomethylated with age-CpGs in REST targets and developmental genes, hypomethylated with age-CpGs in hematopoiesis and lymphoid–myeloid differentiation genes, hypermethylated in cancer-CpGs enriched in T-cell activation and NK mediated cytotoxicity, and hypomethylated in cancer-CpGs in cell adhesion and HOXA9 regulatory pathways. The authors noted that the non-island CpGs were more predictive of cancer, supporting that future studies should consider broad coverage of CpGs in the genome. This result is also supported by the recent observation that gene body methylation can alter gene expression [ 144 ]. Overall these studies are generally encouraging, but await validation by others. Additionally, establishing clear methylation patterns in the different histologies and in other cancers is needed as is the methylation profiles of the normal tissues that give rise to EOC.

Genetic regulation of the DNA methylation is a growing area of interest, in large part to address whether the non-coding heritable risk variants identified by genome-wide association studies (GWAS) exert their effects by altering important regulatory mechanisms [ 145 ]. In EOC, GWAS have identified 22 genomic loci associated with risk [ 6 , 7 , 9 , 59 , 146 – 151 ]; the majority of associated variants are in non-coding regions. Most of the top risk variants identified by these GWAS have been evaluated for their association with DNA methylation in cis (1-Mb region), using TCGA data for HGS tumors [ 49 ], and/or in data from 100 to 300 tumors representing a cross section of all EOC histotypes [ 68 ]. There is strong evidence that genetic regulation of the DNA methylation is a mechanism behind altered risk in the 17q11 region containing HNF1B [ 59 , 151 ]. In this region, independent signals (SNPs) are found to confer risk in different EOC histotypes. Specifically, HGS risk alleles were found to increase risk and associate with greater HNF1B-promoter methylation, while CC risk alleles decreased risk but were not associated with HNF1B expression [ 59 ]. This suggests a different mechanism of action for risk SNPs in these two EOC histotypes. At least two other EOC risk loci harbor SNP-CpG (genetic-DNA methylation) associations that also affect gene expression, including 8q21 (CHMP4C) and 10p12 (SKIDA1, MLLT10) [ 151 ]. Studies of genetic regulation of DNA methylation in EOC risk regions are ongoing. Prior reports only examined top risk SNPs, but fine-mapping efforts are generating more complete lists of candidate causal SNPs and these are being evaluated for their SNP-CpG associations. It is possible that EOC risk variants affect DNA methylation in trans (beyond 1 Mb); this had not yet been pursued. Finally, methylation quantitative trait loci (mQTL) studies that assess all CpG-SNP combinations genome-wide (i.e. in cis and trans) have now been reported for several normal tissues, including the brain [ 152 ], lung [ 153 ], and pancreatic islets [ 154 ], but not in precursor tissues relevant to EOC (see Table 1 ), and not in EOC tumors. Genome-wide mQTLs, although computationally and statistically challenging, hold the promise of agnostically evaluating the role that common genetic variants play in DNA methylation and regulation of gene expression. Somatic mutations in IDH1, IDH2, H3F3A, TET1, TET2 and/or DNMT3A have been found to explain distinct DNA methylation patterns observed in subgroups of glioblastoma [ 155 , 156 ] and leukemias [ 157 , 158 ]. No such pattern has been reported for EOC, although, relatively few of the rarer histotypes have been sequenced. Overall, the genetic component of variable DNA methylation, including somatic mutations and common germline variants, are only beginning to be understood for EOC. Datasets large enough (number of samples) and comprehensive (genome wide sequencing and DNA methylation) to properly address these questions are confined at present to the HGS histotype (TCGA). Future work to genetically and epigenetically profile CC, EN, and MC tumors, and compare DNA methylation patterns, may reveal novel insights.

Epithelial ovarian cancer (EOC) is a heterogeneous cancer with at least four distinct histological types: serous (70% cases), endometrioid (EN) (11%), clear cell (CC) (12%), and mucinous (MC) (3%) [ 1 ]. Serous and EN tumors can be further divided in high-grade (HG) and low-grade (LG) tumors. By histological type, and grade, these tumors have different genetic [ 2 – 9 ] and epidemiologic [ 10 – 12 ] risk factors, precursor lesions [ 11 , 13 ], pattern of spread, expression signatures [ 14 – 16 ], response to platinum-taxane based treatment, and patient outcome [ 17 ]. Migration of precursor lesions from disparate organ sites to the ovary is one proposed model for how these disparate histotypes arise [ 18 ]. The precursor sites are thought to be: HGS from fallopian tube or ovarian surface epithelium, EN and CC from endometrium or endometriosis, and MC from endocervix or intestinal mucosa [ 18 ]. A diagnosis of EOC remains ominous even in this time; 5-year survival is poor (b 40%) as most cases (N 70%) are diagnosed with advanced stages of disease [ 19 ]. Despite histological differences, treatment is similar, and resistance to the first line therapeutic agents, paclitaxel and carboplatin, is common. A connection between aberrant DNA methylation patterns and human cancers was first noted in 1983 [ 20 ]. Since that time, there has been a growing appreciation of its role in cancer development and prognosis. For example, aberrant DNA methylation-associated transcriptional silencing is widely observed in cancers [ 21 ]. In EOC, studies have shown that DNA methylation changes are an early step in carcinogenesis and could represent a mechanism of disease, one that if understood, could be targeted or addressed in some way. DNA methylation is the process by which methyl groups are added to cytosine nucleotides in DNA, typically in the C 5 position (5mC) in the context of CpG dyads; however, there is a growing appreciation that this is not the only context for DNA methylation [ 22 ]. Between 60% and 90% of all CpG loci are methylated in humans [ 23 ], and these wide-spread modifications have important roles in maintaining stable tissue-specific gene expression, transcriptional repression of repetitive elements in the genome, inactivation of one X-chromosome in women, and allele-specific expression. The tissue specification of DNA methylation patterns is particularly relevant to studies of EOC. The different EOC precursor tissues, coming from different cell lineages, mesothelium (ovarian surface epithelium) and Müllerian (fallopian tube) [ 24 ], are anticipated to have different DNA methylation patterns. A salient example are the HOXA genes, which are differentially expressed in ovarian surface epithelium, endometrium, and endocervix and control differentiation of the Müllerian ducts into fallopian tubes, uterus, and cervix [ 25 ]. The extent of the DNA methylation differences in the EOC precursor tissues is not yet known. DNA methylation patterns are established and maintained in tissues by a family of proteins known as DNA methyltransferase (reviewed previously in [ 26 – 28 ]). Relevant to aging and cancer, these patterns become less consistent with age, either through passive means (failure of maintenance methylation) or active demethylation activity. Approximately 70% of human promoters feature CpG rich sequence (spanning on average 1000 bp) [ 29 ], which is typically not methylated. These CpG rich sequences have been annotated to the genome using different algorithms, and are referred to as CpG Islands (CGIs). Methylation of CGI’s is particularly important, as it is often associated with changes in gene expression. Importantly, the platforms most commonly used to survey DNA methylation focus on CGI regions, including Illumina’s HumanMethylation450k and HumanMethylation27k Beadchips and GoldenGate panels [ 30 ]. Next generation sequencing approaches can provide an agnostic scan of DNA methylation, but are still relatively rare in the literature. Some cancers, including tumors from colorectal, breast, and gliomas are uniquely characterized by wide-spread CGI methylation; these are often referred to as having a CGI methylation phenotype (CIMP) (reviewed by Hughes et al. [ 31 ]). Here we review alterations in DNA methylation in EOC, focusing whenever possible on the differences observed in the four major histotypes. We also examine the roles of DNA methylation in determining response to therapy, again focusing whenever possible on the differences observed in the histotypes. As epigenetics is making its way into clinical care, we review the application of cell free DNA methylation to EOC diagnosis and care. Finally, we comment on recurrent limitations in the DNA methylation in EOC literature, which can and should be addressed to mature this field. Recent related reviews include Balch et al., Barton et al., Asadollahi et al., and Koukoura et al. [ 32 – 35 ].

Several recurrent limitations are present in the DNA methylation literature for EOC. In our opinion, these are: (1) the admixing of histotypes in DNA methylation analyses, (2) the lack of, or mismatching, of precursor tissues, (3) small sample size, and (4) lack of replication. The admixing of histotypes could result in aberrant DNA methylation patterns, and their association with other molecular features, being missed. Correctly pairing tumors to their relevant precursor tissue is particularly critical for DNA methylation analyses, as epigenetic patterns are highly tissue specific [ 159 ]. In EOC, correctly specifying the relevant normal tissue is difficult owing to it being a heterogeneous disease with ambiguous precursors. Moving forward, multiple possible precursor tissues could be included in analyses to ensure that tissue-specific and cancer-specific methylation patterns can be distinguished from one another. Small sample sizes make replication particularly important, however, for the most part DNA methylation changes have not been extensively replicated. More rigorous replication is needed, particularly of DNA methylation biomarker being considered for clinical use. Future work to genetically and epigenetically profile more EOC tumors, particularly the CC, EN, and MC histotypes, will likely reveal (and confirm) novel DNA methylation patterns that inform disease etiology, and provide DNA methylation biomarkers that may be used clinically.