We used a new strategy to identify blood-based candidate DNA methylation markers of ovarian cancer and to verify their potential to detect recurrent disease. We sought to circumvent some of the limitations associated with the biomarker development process 
. To our knowledge this is the most extensive marker discovery study to date for ovarian cancer since most of the previous studies have relied on the use of candidate markers or limited screens.
One problem with biomarkers identified by high throughput technologies is their lack of sufficient specificity 
. The use of a genome-scale screening approach presented us with the challenge of defining a clear marker selection strategy that would emphasize both marker sensitivity and specificity and help us prioritize among the hundreds of potential biomarkers. In most studies of epigenetic biomarkers for blood-based detection of cancer, marker specificity is initially inferred from normal vs. tumor tissue comparisons. We emphasized specificity of blood-based detection by directly comparing tumors from ovarian cancer patients to blood DNA from women without ovarian cancer, and eliminating markers found to be methylated in blood from age-matched healthy controls. We included ovarian cancer samples from four different ovarian cancer subtypes in the analysis (Table S1
) to maximize marker sensitivity for detection for all of these types of ovarian cancer. During the verification process we counter-screened our markers with large quantities of PBL DNA and then with both serum- and plasma-derived DNA to exclude markers with low specificity.
DNA methylation markers generally suffer from poor clinical sensitivity 
. In order to emphasize sensitivity in our marker selection strategy we used very stringent criteria that required consistently higher DNA methylation in all tumors than in any of the normal blood samples. We anticipated that this approach would enrich for markers with a high prevalence of DNA methylation in ovarian cancers, which in turn, would translate into a higher sensitivity for detection of ovarian cancer in patient blood than for markers with a lower frequency of tumor DNA methylation.
Promising biomarkers emerging from large-scale discovery efforts have often performed poorly when tested on independent validation samples 
due in part to lack of randomization of case and control blood samples at baseline in observational diagnostic studies 
. Diagnostic biomarker studies usually rely on subject selection on the basis of diagnosis, which can result in baseline differences between cases and controls 
. This can lead to false-positive identification of disease-associated markers. Population-based cohort studies with prediagnostic blood samples can also be an excellent source for nested case-control comparisons (also referred to as a prospective-specimen collection, retrospective-blinded-evaluation (PRoBE) design 
. However, the number of incident cases in cohorts is usually limiting and the high demand for these precious samples generally precludes their use for early-stage candidate DNA methylation biomarker evaluation. Also, since ovarian cancer is a disease with low incidence, the use of a prospective population-based study for early-phase ovarian cancer markers validation is not very practical.
In this study we tested an alternative verification scheme that evaluates the marker's correlation with a validated marker, CA-125, known to be associated with disease status in post-resection serially collected blood samples. This within-subject approach circumvents some of the drawbacks of traditional case-control designs, since each case serves as her own genetically matched control. The information regarding the ability of our top candidate marker, IFFO1-M, to measure disease status was extracted from the temporal pattern across many serial samples for each subject (8–21 samples per patient). The comparison to the validated biomarker CA-125 lent further support to the conclusion that IFFO1-M is measuring disease status.
The rapid decrease in the serum levels of IFFO1-M in all nine patients in the weeks immediately following surgery provides compelling evidence that IFFO1-M serum levels reflect tumor burden. In many cases, IFFO1-M closely tracked CA-125 serum levels in the post-resection serum specimens. IFFO1-M rose at the time of disease recurrence in three of the nine patients in a similar manner to CA-125, and even outperformed CA-125 in one additional patient in which CA-125 never increased over its normal values, despite disease relapse (, Patient #18).
CA-125 surpassed IFFO1-M performance in three of the patients (#2, #14, and #21). This however, could be a direct consequence of the small volume of serum (100 µl) used for the DNA methylation analyses, and better performance of IFFO1-M should be expected in future studies using larger volumes of sera. In combination, CA-125 and IFFO1-M corresponded to relapse in seven of the eight patients with recurrent disease, suggesting that IFFO1-M may complement CA-125 in monitoring residual disease.
Despite increased efforts to identify new protein-based ovarian cancer biomarkers, CA-125 still remains the best marker for detecting early disease, up to three years in advance of the clinical diagnosis in some patients 
and to monitor disease recurrence. This is the first time that a DNA methylation marker has been shown to have a concordant behavior with a protein marker with recognized clinical use. The analysis of post-resection serially collected samples may provide an effective method to evaluate whether a candidate marker has the potential to detect recurrent disease prior to the onset of symptoms or clinical evidence of disease and to help in the triage process of candidate markers that could be advanced for further analysis in valuable samples from larger population-based studies.
In conclusion, we have described here the potential of a new strategy to discover and verify candidate DNA methylation markers for detection of ovarian cancer, and we characterized a new marker, IFFO1-M, that can help enhance the performance of CA-125 in monitoring disease status. We expect this marker and any additional candidate markers emerging from this discovery pipeline to have a great chance of success in future validation stages of the marker development process.