Abstract Introduction: Ovarian cancer (OC), often diagnosed at an advanced stage, lacks effective screening strategy. It is the most lethal gynecological cancer, with a 5-year survival ranging from 35-50% depending on the subtype. A clear correlation between stage and survival has been demonstrated; therefore, early detection represents the best hope for mortality reduction. This study interrogates the potential of ctDNA methylation status as an early diagnostic marker for OC. Methods: Panel Design: Public data sets such as TCGA and GEO databases (tumor=4,014, normal=352) and in-house generated functional methylome (5.5 million CpG sites, 50X) sequencing data (n=561) were used to profile methylation alterations in 12 common cancer types. The panel covers 225,369 CpG sites, spanning ~3.5Mb of human genome. Deep targeted bisulfite sequencing (ELSA-seq) was performed on plasma samples of 126 Chinese ovarian cancer patients, 59 patients with benign ovarian diseases, including but not limited to endometrial cyst, ovarian cyst and serous cystadenoma and 49 age-matched asymptomatic females to interrogate their methylation statuses with an average sequencing depth of 1,000x. The plasma samples were collected and processed from two centers independently, and randomly assigned into the training (n=123) or validation group (n=113) at 1:1 ratio. Samples that failed quality control specifications were removed from the downstream analysis and the diagnostic information was not disclosed for the validation group until all analyses were completed. Results: First, ovarian tissues samples from 22 OC (FIGO stage II-III) patients, 11 ovarian benign diseases patients and 11 healthy donors were profiled to derive markers that can differentiate the 3 groups. In this study, 45,753 OC-specific differentially methylated CpG sites/loci (DMLs) were selected using modified wald-test with an adjusted p-value <0.05 and fold-change>2, and subsequently grouped into 3,160 differentially methylated blocks (DMBs) based on the co-methylation levels and genomic distances of adjacent CpG sites. Blood samples from 63 patients with OC (4 stage I, 6 stage II and 53 stage III), 33 patients with benign ovarian disease and 27 healthy individuals were obtained to train the diagnostic classification model using a support vector machine (SVM)-based machine learning classifier. Subsequently, iterated 5-fold cross-validation were performed to gain a robust estimation of model performance, achieving a specificity of 91% for benign ovarian diseases and sensitivity of 25%, 50% and 87% for stage I, II and III OC, respectively at a specificity of 93% and area under curve (AUC) of 91.3%. The model was subsequently validated in a hold-out cohort, consisting of 22 healthy individuals, 26 patients with benign ovarian disease and 63 stage I-III OC patients with comparable clinical characteristics as the patients in the training cohort. The model yielded a specificity of 93% for benign disease and a sensitivity of 84% for OC at a specificity of 96% in the validation cohort, suggesting its robustness. Conclusion: Our findings demonstrated the potential of ctDNA methylation profiling, which can effectively distinguish ovarian cancer, benign diseases and healthy, to serve as a diagnostic tool in ovarian cancer early detection. The general concept can be further extended to other types of cancer diagnostics. Citation Format: Ning Li, Yu Zhang, Xin Zhu, Zhenjing Zhang, Jiayue Xu, Bingsi Li, Han Han-Zhang, Fujun Qiu, Shuai Fang, Hao Liu, Zhihong Zhang. Methylation profiling of circulating tumor DNA for the detection of ovarian cancer [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2302.
Read full abstract