The low abundance of circulating tumour DNA (ctDNA) in plasma samples makes the analysis of ctDNA biomarkers for the detection or monitoring of early-stage cancers challenging. Here we show that deep methylation sequencing aided by a machine-learning classifier of methylation patterns enables the detection of tumour-derived signals at dilution factors as low as 1 in 10,000. For a total of 308 patients with surgery-resectable lung cancer and 261 age- and sex-matched non-cancer control individuals recruited from two hospitals, the assay detected 52–81% of the patients at disease stages IA to III with a specificity of 96% (95% confidence interval (CI) 93–98%). In a subgroup of 115 individuals, the assay identified, at 100% specificity (95% CI 91–100%), nearly twice as many patients with cancer as those identified by ultradeep mutation sequencing analysis. The low amounts of ctDNA permitted by machine-learning-aided deep methylation sequencing could provide advantages in cancer screening and the assessment of treatment efficacy.

Background: Detection of cancer in its early stages is likely to be the most effective way to improve clinical outcome. In patients with cancer, a portion of cell-free DNA (cfDNA) in the blood stream is tumor derived, providing an opportunity to analyze the cancer genome in real time noninvasively. MERMAID is a retrospective multicenter case-control study to investigate early tumor signatures in blood using different platforms. Methods: The study was conducted among 452 surgery-resectable patients with lung cancer (N=180), colorectal cancer (N=210), liver cancer (N=62), and 290 age-/sex- matched non-cancer controls. Patients who were recognized to have anemia, autoimmune diseases, treated with neoadjuvant therapy were excluded from the study. The non-cancer controls were recruited with the criteria of showing no clinical symptoms or history of cancer at time of administration. The participants were divided into subgroups and analyzed by ultra-deep mutation sequencing (HS-UMI), droplet digital PCR (ddPCR), and deep methylation sequencing (ELSA-seq) singly or in combination. Results: Overall, the highest accuracy was achieved using ELSA-seq and will be reported in full. Highly similar classification results were obtained in training and test sets, with the area under the curve (AUC) value ranging from 0.90-0.97. The specificity for each cancer type ranged from 96-99%, and the average sensitivities with 95CI were lung cancer (61%, 53-70%), colorectal cancer (82%, 76-87%), and hepatocellular cancer (91%, 80-96%). Moreover, incorporation of somatic variants and epigenetic alterations improved the overall accuracy. Conclusions: This study highlighted the potential of machine learning-aided deep methylation sequencing as a sensitive ctDNA profiling approach for early cancer detection. Further investigation in large-scale clinical studies is ongoing. Citation Format: Bingsi Li, Chenyang Wang, Jiayue Xu, Shuai Fang, Fujun Qiu, Jing Su, Huiling Chu, Han Han-Zhang, Xinru Mao, Hao Liu, Xianling Liu, Wei Zhang, Heng Zhao, Zhihong Zhang. Multiplatform analysis of early-stage cancer signatures in blood [abstract]. In: Proceedings of the AACR Special Conference on Advances in Liquid Biopsies; Jan 13-16, 2020; Miami, FL. Philadelphia (PA): AACR; Clin Cancer Res 2020;26(11_Suppl):Abstract nr A06.