Circulating Cell-Free DNA Methylation Profiles in the Early Detection of Ovarian Cancer: A Scoping Review of the Literature

Simple Summary There are limited non-invasive methods for detecting epithelial ovarian cancer despite early detection and treatment dramatically increasing survival. As alterations in serum or plasma cell-free (cf)DNA methylation occur early in cancer development, they are promising biomarkers for ovarian cancer. Our literature review includes 18 studies depicting a wide array of gene targets and techniques. The data suggest a good performance of these cfDNA methylation tests, with accuracies up to 91% in detecting ovarian cancer in serum or plasma. Abstract Epithelial ovarian cancer is the most lethal gynecologic malignancy and has few reliable non-invasive tests for early detection or diagnosis. Recent advances in genomic techniques have bolstered the utility of cell-free DNA (cfDNA) evaluation from peripheral blood as a viable cancer biomarker. For multiple reasons, comparing alterations in DNA methylation is particularly advantageous over other molecular assays. We performed a literature review for studies exploring cfDNA methylation in serum and plasma for the early diagnosis of ovarian cancer. The data suggest that serum/plasma cfDNA methylation tests have strong diagnostic accuracies for ovarian cancer (median 85%, range 40–91%). Moreover, there is improved diagnostic performance if multiple genes are used and if the assays are designed to compare detection of ovarian cancer with benign pelvic masses. We further highlight the vast array of possible gene targets and techniques, and a need to include more earlier-stage ovarian cancer samples in test development. Overall, we show the promise of cfDNA methylation analysis in the development of a viable diagnostic biomarker for ovarian cancer.


Introduction
Epithelial ovarian cancer (OC) is the most lethal gynecologic malignancy with a 5-year survival rate under 50%. Although disease confined to the ovaries is associated with a greater than 90% survival rate, the majority of patients who present with distant metastasis have a dismal prognosis [1,2]. There is also an increasing age-standardized incidence rate of ovarian cancer worldwide [3], while mortality rates have remained relatively constant [2].
Most epithelial OCs have high-grade serous histology, an aggressive subtype with few known risk factors, the majority of which present with advanced-stage disease and have a

Results
After removal of duplicates, 480 article titles were screened for thematic relevance over the two database searches. Subsequently, 91 articles underwent abstract review. We selected 24 primary research studies and 22 review articles for full-text review. After full-text review, 18 primary research studies were included for review, including 8 studies abstracted from previously published review articles. Full-text primary research articles were excluded from final analysis for evaluating methylation in sources that were not serum-or plasma-based cfDNA, including leukocyte-derived DNA from peripheral blood (n = 6), primary ovarian tissue (n = 5), the cervix (n = 2), and the endometrium (n = 1). A flowchart is shown in Figure 1. However, despite increasing investigation on DNA methylation biomarkers for OC, their use is still novel, and a comprehensive review does not exist. We thus conducted a scoping review of the literature for cfDNA methylation alterations in serum/plasma for early detection of ovarian cancer. Specifically, we compiled methodological and outcome data from the included studies to compare and summarize the state of research in this emerging research landscape.

Results
After removal of duplicates, 480 article titles were screened for thematic relevance over the two database searches. Subsequently, 91 articles underwent abstract review. We selected 24 primary research studies and 22 review articles for full-text review. After fulltext review, 18 primary research studies were included for review, including 8 studies abstracted from previously published review articles. Full-text primary research articles were excluded from final analysis for evaluating methylation in sources that were not serum-or plasma-based cfDNA, including leukocyte-derived DNA from peripheral blood (n = 6), primary ovarian tissue (n = 5), the cervix (n = 2), and the endometrium (n = 1). A flowchart is shown in Figure 1.

Study Characteristics
The 18 articles were published between 2004 and 2018 (Table 1). Overlapping investigators were noted in three pairs of studies [27][28][29][30][31][32], with one pair using samples from the same patient and control populations [29,30] and another pair using the same control population [31,32]. The case-control study design was utilized by all studies except for one case series [33] and two that analyzed the diagnostic potential of their chosen methylation targets with prospective samples [34,35]. Most studies used serum samples for their
Subsequent methylation analysis was mostly conducted using methylation-specific polymerase chain reaction (MSP) followed by gel electrophoresis for comparison, with some studies reporting slight PCR modifications. Other techniques employed include two studies using the self-developed "MethDet" technique to analyze their products via customdesigned methylation assay microarrays [27,28], two studies utilizing sequencing-specific information for methylation quantification, and one study using a methylation-sensitive high-resolution melting analysis assay for quantitative estimation of methylation [31].

Target Genes
All studies except three had preselected genetic targets for methylation detection based on prior studies or literature reviews. The remaining three studies performed discovery analysis to determine the most suitable candidate genes: Two studies used their own microarray of 56 promotor fragments to determine which genes were differentially methylated between clinical cohorts [27,28], and one study performed genome-wide methylation analysis with subsequent validation of a small subset of the best candidate markers [35].
A total of 26 genes or gene families were evaluated by the various publications comparing OC with non-OC specimens ( Table 2). Eleven studies evaluated the methylation profiles of single genes, while seven studies evaluated a panel of genes (median 4, range 3-7). Most gene targets were tumor suppressors, and 10 were utilized by more than one study. The gene included most often was RASSF1A, a tumor suppressor involved in numerous apoptosis and cell cycle checkpoint functions. OPCML, encoding for a plasma membrane protein involved in cell-to-cell recognition and adhesion, was the next commonly targeted gene. Most studies evaluated methylation in the promotor region of the gene; two studies [34,38] alluded to, but did not explicitly state, use of the promotor region(s), while two studies [30,46] did not specify where methylation occurred in their chosen gene(s).

Diagnostic Performance
As summarized in Table 1 and Supplemental Table S1, studies had wide ranges for the performance(s) of the targeted genes and utilized an array of controls and comparisons. All but one study reported relevant statistical results or provided enough data to extrapolate some statistical measures. Studies evaluated serum/plasma methylation alterations in differentiating OC patients from healthy controls (39%, 7/18 studies), benign ovarian masses (11%, 2/18 studies), or both (39%, 7/18 studies). Two studies only compared serum/plasma with paired tissue methylation [33,42].
For all studies, sensitivities ranged from 16 to 97.8% (median 80%), specificities ranged from 50 to 100% (median 86.7%), and accuracies ranged from 40 to 91.2% (median 84.8%). In general, the positive predictive value (PPV) of each study's reported methylation tests in determining OC was higher than the negative predictive value (NPV). The average accuracies tended to be higher when the methylation profiles of a panel of genes were evaluated rather than a single gene (89.9% vs. 74.5%, p = 0.07) and when using both benign ovarian masses and healthy controls as the comparison arm against OC instead of just using healthy controls (85.4% vs. 64.8% p = 0.006; Figure 2). Six studies found improved discriminatory abilities with serum/plasma from advanced-stage OC patients compared to those with early stages [27,29,30,[43][44][45], but most reported no correlation with clinicopathologic characteristics. Whether serum or plasma was used for cfDNA extraction and the subsequent bisulfite conversion kit or methylation analysis technique employed did not lead to significant differences in test characteristics. Wang et al. reported the best overall diagnostic performance in their two studies, which used the same patient population but had slightly different gene targets for methylation analysis: The earlier study [29] looked at a panel of genes (RUNX3, TFPI2, and OPCML (accuracy 90.7%) in differentiating OC patients from healthy controls or patients with benign ovarian masses, while their later study [30] narrowed down their analysis to OPCML (accuracy 91.2%). When comparing individual genes (Supplemental Table S1), OPCML also had the greatest overall sensitivity and specificity. the best overall diagnostic performance in their two studies, which used the same patient population but had slightly different gene targets for methylation analysis: The earlier study [29] looked at a panel of genes (RUNX3, TFPI2, and OPCML (accuracy 90.7%) in differentiating OC patients from healthy controls or patients with benign ovarian masses, while their later study [30] narrowed down their analysis to OPCML (accuracy 91.2%). When comparing individual genes (Supplemental Table 1), OPCML also had the greatest overall sensitivity and specificity.

Discussion
To our knowledge, this study represents the first scoping review of serum/plasma cfDNA methylation targets for the early diagnosis of OC. We included 18 studies depicting how epigenetic alterations could be used to diagnose OC patients of differing stages and histologies with diagnostic accuracies up to 91.2% (median 84.8%, range 40-91%), in line with prior meta-analyses of using cfDNA in general for OC [13,14]. Better diagnostic performance was appreciated for studies that included benign masses as comparison arms instead of only healthy controls and when multiple genes were evaluated for methylation status instead of a single gene.
The inclusion of benign masses is a clinically relevant comparison given the clinical dilemma of undiagnosed pelvic masses. The improved accuracy noted when these benign masses were used as a comparison group is perhaps a statistical result of studies including benign masses having larger overall cohort sizes or due to a real epigenetic difference between the cfDNA methylation patterns of benign and malignant ovarian masses. Unfortunately, the nuances behind this differentiation could not be elucidated in our current review and warrant further evaluation. Studies also reported a wide range of genetic targets, including classic tumor suppressor genes such as BRCA1 and PTEN and OC-specific tumor suppressors such as RASSF1A and OPCML. Since most genes were only included in one or two studies, our ability to conduct gene-level analysis from this review was very limited. Moreover, as no single gene has been identified as being predominantly methylated for ovarian cancer tissues [47], including a panel of methylated gene targets would appropriately be more suitable. This is concordant with both our results and studies in ovarian and other cancers showing that utilizing multimodal methods and algorithms provides improved screening and diagnostic performances [7,9,48]. Indeed, non-invasive methylation-based biomarker tests that are currently commercially available for various cancers tend to be panels that target multiple genes [17,49,50].
A few studies in our review directly compared the performance of cfDNA methylation status with CA125 and found better detection using methylation methods [29,30,34,45]. For example, Widschwendter et al. [35] stratified their screening population by CA125 levels and found improved accuracy of their methylation test when only analyzing healthy controls with normal CA125. Prospective trials will ultimately be needed to determine the comparative effectiveness of these tests; future studies would also benefit from utilizing a combination of traditional biomarker, ultrasonography, and epigenetic profiles to further enhance early diagnosis potential.
Given the known challenges in obtaining sufficient cfDNA for adequate analysis from healthy individuals and those with early-stage cancers [12,51], it was promising to see that all but one study included patient samples from early-stage OC. However, samples were still predominantly from stage III and IV patients. This expectedly led to better discriminatory abilities of the methylation tests with advanced-stage patients, although two studies [30,34] found improved detection of early-stage OC using methylation markers as compared to CA125. Although small yields of cfDNA extractable from peripheral blood will continue to be a barrier to the validation of methylation methods, technological advancements will assist in overcoming this technical hurdle; in clinical practice, larger volume blood draws can also offset the low cfDNA yields. Until then, other techniques currently being explored include methylation analysis of bodily fluids closer to the ovaries such as endometrial lavages, endocervical brushings, Papanicolaou smears, and vaginal fluids [4,9].
Overall, the wide range of reported sensitivities and specificities for the varying studies is possibly a function of their assorted techniques, gene targets, and human samples used. Five separate commercially available bisulfite conversion kits were used likely due to their availability for each research group, which may have uncertain downstream effects on analysis given the differences in recovery or conversion rates between the kits [52]. Furthermore, while serum samples tend to produce greater cfDNA yields than plasma, they have poorer fidelity in maintaining mutation frequencies because of dilution with non-cancer cfDNA [53,54]. However, we did not note better diagnostic performances for certain methods or detect a difference between those studies using serum or plasma.
As such, the main limitation of this review lies in the heterogeneity of the included studies, which precludes robust conclusions from being drawn. This is a function of the relative novelty of epigenetic evaluation of cfDNA, as evident by most of the included studies being published in the past decade and the lack of standardized methylation analysis methods. Many of the included studies were exploratory and retrospective with small sample sizes, but the two with prospective screening samples showed promising results with accuracies above 80% [34,35]. We may have also missed relevant studies from excluding non-English publications, such as one Chinese language study that was included in a previous systematic review on cfDNA [55]. However, our current review ultimately provides the most current evaluation of the state of methylation science for OC and demonstrates exciting future possibilities of serum/plasma cfDNA methylation in assisting with diagnosis and early detection.

Search Strateg
A systematic search was performed following the criteria of Preferred Reporting Items for Systematic Reviews and Meta Analyses-Extension for Scoping Reviews (PRISMA-ScR) [56] in PubMed's Medline and Elsevier's Embase and Scopus. Articles were required to be in English, with no limitation on the date of publication. MeSH terms were used for PubMed but were unavailable for use in Elsevier searches. The initial query was made on July 8 2020 and a second broader search was performed on September 10 2020 utilizing the same databases with a slight variation in search terms that removed the requirement for biomarkers (Table 3). Table 3. Review search terms. Two searches were conducted in PubMed's Medline and Elsevier's Embase and Scopus. MeSH terms were used only in Medline. Terms were combined using Boolean logic statements: Within-column terms were combined with "OR"; between-column terms were combined using "AND". The asterisk (*) represents a wildcard symbol used to broaden the search to all terms with the previous stem.

Inclusion and Exclusion Criteria
Articles were required to be primary, original research evaluating the DNA methylation profiles of cell-free DNA in peripheral blood for patients with OC. Due to the relatively novel nature of the research, we placed limited restrictions on the study size, the presence of controls, techniques utilized, or the diagnostic performance of the described epigenetic marker. Excluded publication types included conference abstracts, letters, or editorials without complete methods or data. Review articles identified through the systematic searches were excluded if they did not contain original research but examined for additional studies.

Data Extraction
Eligible articles were examined for basic study characteristics including publication year, sample size(s) and source for cases and controls, clinicopathologic information for the OC cases (histological subtype and tumor stage, if available), method for DNA methylation analysis, and gene target(s). Key findings including diagnostic performance markers for the genetic biomarkers were also retrieved; this included any available sensitivities, specificities, positive and negative predictive values (PPV and NPV), accuracies or areas under the ROC (receiver operating characteristic) curve, and associated p-values. Finally, a summary section highlighted the given study's serum or plasma DNA methylation analysis in differentiating or detecting OC along with any comparison groups. If the study did not provide explicit summary statistics, extrapolated summaries were calculated based on available data based on the following calculations: Sensitivity  . A data extraction sheet was developed with the above variables and actively refined during a first-pass review of the articles. All articles were then reviewed a second time to ensure that the included information was consistent. Any ambiguities in the retrieved data were resolved by consensus amongst the co-authors.
Diagnostic data were compiled and analyzed with Microsoft Excel, Version 1908. Mean ± standard deviation, median, and range were calculated as continuous variables. Comparisons were made with independent sample t-tests for continuous variables. All analyses assumed a two-sided 5% level of significance.

Conclusions
In recent years, multi-cancer screening blood tests utilizing cfDNA are gaining attention and funding but suffer from detection rates around 50% and lack of large prospective clinical data [25,57]. Commercial tests for specific cancers such as colorectal and liver are also coming to fruition, with many using DNA methylation techniques [50]. For ovarian cancer, a minimally invasive diagnostic marker would be invaluable. However, despite ample attempts at studying biomarkers and algorithms, efforts have fallen short in identifying a reliable strategy for the detection of ovarian cancer [4]. As such, the nature and time course of DNA epigenetic alterations make methylation status a unique and robust biomarker compared to other liquid biopsies specifically for early cancer detection. Our review shows the current breadth of research in this topic and suggests improved diagnostic performance of methylation tests when comparing OC with benign pelvic masses and when including multiple genes in the analysis. We also highlight the vast array of possible gene targets and techniques and a need to include more earlier-stage OC samples in test development. Overall, we show the promise of cfDNA methylation analysis as a viable diagnostic biomarker for OC.