DNA Methylation as a Robust Classifier of Epithelial Ovarian Cancer

In this issue of Clinical Cancer Research, Bodelon and colleagues report that tumor-derived DNA methylation profiles stratify epithelial ovarian cancers (EOC) into four subgroups associated with distinct histopathology, genome stability, transcriptomes, and outcomes (1). Investigators assessed DNA methylation from 162 archival formalin-fixed, paraffin embedded (FFPE) ovarian epithelial tissues retrieved from the Polish Ovarian Cancer Study or the Surveillance Epidemiology and End Results Residual Tissue Repository. Tumor DNA was subjected to Infinium HumanMethylation 450 BeadChIP array profiling to identify differential DNA methylation between EOCs. The five major EOC histologic subtypes were represented among experimental samples: high and low grade serous, endometroid, mucinous and clear cell, as well as tumors of low malignant potential. Clustering based upon the 1,000 most differentially methylated regions segregates tumors into four subgroups that reveal insights into disease etiology, cell-of-origin, genome stability, and patient outcome. Because these insights are not easily garnered from existing EOC classification methods, DNA methylation–based EOC classification harbors potential to impact diagnoses and clinical decision making akin to the impact observed for other cancers (2).

Previous attempts to stratify EOCs beyond histopathologic classifications were largely conducted through transcriptional profiling (3). Bodelon and colleagues, recapitulate previously identified transcriptional subgroups with this EOC cohort using 39 genes assayed through NanoString. Expression subgroups associate with DNA methylation subgroups; however, DNA methylation exhibits several advantages over RNA as a biomarker (1). DNA methylation is more stable and readily recovered from archival patient-derived material with no need to modify existing clinical practices of tissue preservation or storage (2). This encompasses extraction of methylated DNA from frozen or FFPE tissue, or from plasma.

One of the distinguishing features of this study is the use of the same platform to measure DNA methylation profiles across the major EOC histology subgroups. In using the same platform, the authors discover that histologically similar tumors segregate into different DNA methylation subgroups, which may relate to distinct cell-of-origin populations within a given histology subgroup (1). Indeed, cancer DNA methylomes are the culmination of cell-of-origin combined with distinct somatically acquired alterations in DNA methylation (2, 4). In segregating more closely based upon cell-of-origin, DNA methylation–based classification reveals insights into tumor etiology and pathogenesis otherwise missed from analysis of histology alone (1).

Associations with overall survival (OS) also distinguish EOC classification by DNA methylation from classification by pathology or RNA expression. DNA methylation subgroups comprised predominantly of a single histologic subtype exhibit expected morbidity. For instance, 88% of DNA methylation cluster 1 is comprised of high-grade serous cancers (HGSC). Accordingly, DNA methylation cluster 1 exhibits the poorest OS relative to other subgroups. However, concordance between methylation subgroups and histologic subtype is 67%, and some tumors within a given EOC histologic subtype segregate into different DNA methylation clusters and exhibit OS more closely associated with the DNA methylation cluster rather than histology (1). Differences in OS revealed through DNA methylation profiling highlights the ability to uncover pathogenesis with more detail relative to histology alone. Indeed, this relationship between DNA methylation subgroups and OS is also observed in brain malignancies and is being used to stratify brain tumors for distinct courses of clinical management (2).

Associations between DNA methylation subgroups and patient outcome reveal the beginning of a framework for personalized medicine and precision oncology based upon epigenetic profiling. Translating associations between molecular biomarkers and patient outcome is readily evident within the management of gynecologic malignancies. For instance, given the significant increased risk of ovarian cancer in women harboring a germline BRCA1/2 mutation, risk-reducing surgery is recommended after childbearing (3). BRCA1/2 mutations in tumors serve as a biomarker of homologous recombination repair (HRR) deficiency, and thus direct treatment of HGSCs with PARP inhibitors (PARPi) further impair DNA repair and promote genome instability that leads to synthetic lethality (3). Bodelon and colleagues note that DNA methylation subgroups are associated with distinct genome instability index (GII) measures determined by array comparative genomic hybridization (1). If GII can serve as a surrogate measure of HRR deficiency, DNA methylation profiles should be explored as a putative biomarker for PARPi treatment (Fig. 1).

EOC stratification based upon DNA methylation offers new insights into EOC biology and holds the potential to improve patient management. EOC remains the most common cause of gynecologic cancer–related death with a persistent high death-to-incidence rate mainly due to the late diagnosis at advanced stage (3). Previous efforts to intercept EOCs at earlier stages have explored different screening strategies with no methods yet validated in routine practice. Cancer antigen 125 (CA125) levels in patient plasma are used to monitor patients with ovarian cancer for therapeutic response and disease resurgence; although monitoring CA125 levels fails to improve outcome (3). Likewise, the more recently incorporated human epididymis protein 4 plasma–derived biomarker does not exhibit sufficient performance. Beyond protein biomarkers, targeted deep sequencing of cell-free DNA (cfDNA) for EOC-enriched mutations, such as TP53 mutations, has been suggested as an alternative approach for early detection. However, the threshold of detection for TP53 mutations from plasma-derived cfDNA exceeds that of an early stage invasive ovarian cancer or typical serous tubal intraepithelial carcinoma precursor lesion (5). Thus, contemporary liquid biopsies lack a sensitive and specific biomarker of initial ovarian malignancy.

The discovery that DNA methylation distinguishes EOC subgroups regardless of histologic type merits exploring DNA methylation as the elusive liquid biopsy biomarker for early stage EOC. EOC DNA methylation subgroups stratify tumors based upon cancer methylomes unique to cell-of-origin, which is also the basis upon which cell-free methylated DNA immunoprecipitation and high-throughput sequencing (cfMeDIPseq) identifies tumor type. Because cfMeDIP is a capture-based method, cancer-specific methylomes are detected when circulating tumor DNA comprises <0.01% of total cfDNA (4). Thus, the DNA methylation subgroups derived here should be explored using cfMeDIP to determine whether insights into EOC etiology and outcome can be determined at low tumor burden stages, which in turn will determine whether cfMeDIP serves as an early detection tool for EOC.

Immediate next steps for this discovery must include independent validation with a larger EOC cohort that includes all major EOC histologic groups. Other groups demonstrate how DNA methylation–based classification can dictate cancer care from diagnosis to treatment. Combining DNA methylation profiling with machine learning can further stratify patients within DNA methylation subgroups to reveal prognostic insights, and match DNA methylation subgroups with therapy response or existing genetic biomarkers (2, 4). From this perspective, it is possible that DNA methylation profiles determined during low disease burden may identify the presence of common EOC-associated mutations in TP53 or BRCA1/2.

Identification of associations between DNA methylation subgroups and existing genetic biomarkers or potential GII measures of HRR deficiency will expedite translation of DNA methylation profiles into clinical decision making for PARPi selection in EOC. Such associations may also reveal onset of therapy resistance mechanisms that include BRCA mutation reversions that restore HRR function to diminish PARPi efficacy. Beyond PARPi, DNA profiling can reveal epigenetic vulnerabilities that can be targeted with epigenetic therapies to improve selection of patients who will benefit from emerging immunotherapies. With frameworks from other cancers in place, gynecologic malignancy management is poised for the emergence of epigenetics-based precision oncology.

S. Lheureux is a consultant/advisory board member for AstraZeneca and Merck. D.D. De Carvalho reports receiving commercial research grants from Nektar Therapeutics and Pfizer, and is the inventor of patents related to plasma cell-free DNA methylation profiling. No potential conflicts of interest were disclosed by the other author.

All authors acknowledge support from the Ontario Institute for Cancer Research through funding provided by the Government of Ontario. This work was supported by the Canadian Institutes of Health Research (CIHR) New Investigator Salary Award (201512MSH-360794-228629), a Helen M Cooke professorship from the Princess Margaret Cancer Foundation, and a Canada Research Chair (950-231346) to D.D. De Carvalho. C.A. Ishak is supported by a CIHR Postdoctoral Fellowship (MFE-164724).