Genome-scale screen for DNA methylation-based detection markers for ovarian cancer

Abstract

Background: The identification of sensitive biomarkers for the detection of ovarian cancer is of high clinical relevance for early detection and/or monitoring of disease recurrence. We developed a systematic multi-step biomarker discovery and verification strategy to identify candidate DNA methylation markers for the blood-based detection of ovarian cancer.

Methodology/principal findings: We used the Illumina Infinium platform to analyze the DNA methylation status of 27,578 CpG sites in 41 ovarian tumors. We employed a marker selection strategy that emphasized sensitivity by requiring consistency of methylation across tumors, while achieving specificity by excluding markers with methylation in control leukocyte or serum DNA. Our verification strategy involved testing the ability of identified markers to monitor disease burden in serially collected serum samples from ovarian cancer patients who had undergone surgical tumor resection compared to CA-125 levels. We identified one marker, IFFO1 promoter methylation (IFFO1-M), that is frequently methylated in ovarian tumors and that is rarely detected in the blood of normal controls. When tested in 127 serially collected sera from ovarian cancer patients, IFFO1-M showed post-resection kinetics significantly correlated with serum CA-125 measurements in six out of 16 patients.

Conclusions/significance: We implemented an effective marker screening and verification strategy, leading to the identification of IFFO1-M as a blood-based candidate marker for sensitive detection of ovarian cancer. Serum levels of IFFO1-M displayed post-resection kinetics consistent with a reflection of disease burden. We anticipate that IFFO1-M and other candidate markers emerging from this marker development pipeline may provide disease detection capabilities that complement existing biomarkers.

Figure 1. Schematic representation of the ovarian cancer marker discovery and verification pipeline.

The Infinium platform was used to screen 27,578 probes representing 14,489 individual gene loci. We used a systematic stepwise approach to eliminate probes that failed in any of the samples, probes that contained SNPs or repeat sequences, or probes with a beta value higher than 0.2 in any of the PBL samples. The remaining probes were ranked based on their difference between tumors and blood (see Materials and Methods), and the probes with higher DNA methylation in PBL than in any of the tumor samples were eliminated. The top 15 from the remaining 517 markers were transitioned to the MethyLight platform for further verification. All 15 markers passed an independent verification test performed on publically available TCGA ovarian cancer dataset and additional PBL samples. Three markers failed due to incompatibility issues related to the MethyLight platform, while another ten failed because they were methylated in normal PBL DNA (3 markers) or normal plasma (7 markers). Only one marker, IFFO1, was selected for further verification on patient samples using Digital MethyLight. (The asterisk indicates probes that failed in any of the samples, as well as those that included SNPs and repeat sequences).

Figure 2. Heat map representation of the marker selection process.

A, the 12,194 markers remaining after the elimination of the probes that failed in any of the samples, and of the probes containing SNPs or repetitive elements. Markers are ranked in an ascending order based on the mean DNA methylation β value of the two PBL samples. B, the 8,701 markers remaining after eliminating probes with DNA methylation β values≥0.2 in any of the two PBL samples. Probes were ranked in a descending order based on the difference in DNA methylation between the tumor with the lowest β value (T_L) and the PBL sample with the highest β value (PBL_H). C, the 517 markers with higher DNA methylation values in any of the tumor than in any of the PBL samples. The markers are ranked in a descending order based on the difference between the tumors and the PBL DNA methylation values. D, the top-ranked 15 markers that were transitioned to the MethyLight platform for further verification.

Figure 3. Dot plot display of the top 15-ranked marker distribution in two independent data sets of ovarian cancer samples and ten normal PBL samples.

The Infinium-derived β values (Y-axis) for the top 15-ranked markers were compared in the present study (PS) data set (41 ovarian cancers of mixed subtypes), the TCGA data set (284 serous ovarian cancers) and ten normal PBL samples. The horizontal lines represent the median values for each group.

Figure 4. Representation of the verification phase on the MethyLight platform of the top-ranked 15 DNA methylation markers.

Technical controls for the MethyLight (ML) platform led to the elimination of four markers (crossed gray boxes) due to design incompatibility, and failure to amplify the in vitro methylated DNA positive control for MethyLight reactions (M.SssI test). Eleven markers were tested in normal PBL samples using an excess of PBL DNA (50 ng). Markers with a cycle threshold (Ct) higher than 35 (blue boxes) in the two normal PBL samples were retained and markers with a Ct less than 35 (yellow boxes) were eliminated. MethyLight assays with Ct values<35 indicate appreciably detectable amounts of methylated DNA at these loci. Further testing in normal control plasma samples (100 µl) resulted in the elimination of seven of the eight remaining markers. One remaining marker, IFFO1, was tested in 15 ovarian cancers of different histological subtypes. The MethyLight results for the normal plasma control and the ovarian cancer samples are expressed as Percent of Methylated Reference (PMR). Blue boxes represent PMR values less than 10, yellow boxes indicate PMR values between 10 and 50, whereas red boxes signify PMR values higher than 50. The types of ovarian tumors used in the analysis are as follows: clear cell carcinomas (CC), mixed clear cell and endometrioid (CC/E), endometrioid (E), mucinous (M), and serous (S).

Figure 5. The performance of IFFO1-M marker in the baseline serum samples of ovarian cancer patients and disease-free control women.

A, IFFO1-M levels (expressed as the number of IFFO1-M methylated molecules detected in 1 ml sera) in the baseline samples of 16 patients and eight normal controls were determined by Digital MethyLight. The number of molecules in patient #1 is an approximation since counts higher than 15 hits/96-well plate/100 µl tested could reflect the presence of more than one molecule/well. The histological subtype of the tumors is indicated in parenthesis as follows: serous (S), mucinous (M), and endometrioid (E). The asterisks indicate the patient from whom samples were used in the subsequent longitudinal analysis. B, Receiver operating characteristic curve for IFFO1-M. AUC = area under the curve.

Figure 6. Comparison between the CA-125 and IFFO1-M performance in serially collected serum samples of nine ovarian cancer patients.

Blood collected from ovarian cancer patients at the time of surgery (baseline samples) and at subsequent follow-up visits was used to measure CA-125 and IFFO1-M levels. The CA-125 levels (gray bars) are expressed in units/ml of blood, and the IFFO1-M methylation levels (black bars) are expresses as number of detected molecules/ml of sera on the Y-axis. The methylation analysis was performed using Digital MethyLight in DNA extracted from 100 µl of serum. The number of weeks since the baseline sample was collected is represented on the X-axis. The horizontal dashed line set at 35 u/ml represents the normal cut off value for CA-125. All patients except patients #5 and #18 had elevated levels of CA-125 in the baseline samples (>35 u/ml). The arrow labeled S indicates the time of surgery and the arrow labeled R indicates the time of tumor relapse as determined by CA-125 and/or imaging techniques. Due to the large range of CA-125 values we restricted the Y-axis to a scale of 400 for both of the markers, and we indicated the measurements that exceeded this scale by an asterisk. The values for these determinations are listed in the Table S4.