Early detection of hepatocellular carcinoma with methylation and fragmentation signatures of circulating tumour DNA: a prospective, multicentre, case-control, observational study.

<bold>Background: </bold>Hepatocellular carcinoma is one of the deadliest cancers worldwide. Early detection has been shown to enable more effective treatments, thus decreasing morbidity and mortality. Non-invasive cancer detection via circulating tumour DNA (ctDNA) has emerged as a promising approach to monitor the molecular changes in liver tumour cells. We aimed to use ctDNA methylation and fragmentation signals to develop a blood-based assay for hepatocellular carcinoma early detection, named HcSeer.<bold>Methods: </bold>A targeted methylation sequencing panel was designed to integrate 1601 liver cancer-informative methylation markers, on the basis of in-house data and data from the public databases The Cancer Genome Atlas and Gene Expression Omnibus. Various methylation features were constructed from methylation sequencing data, including methylation haplotype load and methylated haplotype fraction. Low-pass whole-genome sequencing data from plasma samples were also analysed at a mean sequencing depth of 10×. Fragmentomic features, such as end motif, breakpoint motif, fragmentation size ratio, and copy number variation, were extracted from whole-genome sequencing data. A two-step deep neural network model was built to classify cancer and healthy samples with selected features of both types. The robustness of entire approach was verified with a 3× cross-validation by randomly splitting samples into training set and test set at a 2:1 ratio.<bold>Findings: </bold>In the discovery phase, a case-control study was designed to develop the HcSeer assay for the early detection of hepatocellular carcinoma. A total of 401 participants were recruited (200 healthy individuals and 201 patients with hepatocellular carcinoma). Most patients with cancer were at early stages standardised by Chinese liver cancer staging (109 [54%] at stage I and 25 [12%] at stage II). The classification model of HcSeer assay was built and cross-validated to achieve an average area under the curve of 0·99 (sensitivity 94% [189 of 201; 95% CI 90-97%] and specificity 96% [192 of 200; 92-98%]). The detection accuracy was observed to increase with cancer stages, with 91% (99 of 109; 95% CI 83-95%) sensitivity for stage I, 96% (24 of 25; 77-100%) for stage II, 98% (43 of 44; 86-100%) for stage III, and 100% (23 of 23, 84-100%) for stage IV. The validation cohort is ongoing, with the aim of reaching 510 plasma samples from multiple centres, including a full spectrum of liver diseases and age-matched healthy controls. The validation phase is expected to be completed in June, 2022.<bold>Interpretation: </bold>We have developed the HcSeer assay to combine the signatures of DNA methylation and genome-wide fragmentome. In a case-control study, we showed its feasibility to detect early-stage hepatocellular carcinoma with high accuracy. We propose it as a potential aid for non-invasive diagnostics of hepatocellular carcinoma. The performance of the assay is being validated in an independent sample set collected through a multicentre study, which was approved by the Ethical Committee of Zhongshan Hospital affiliated to Fudan University (number B2020-299R).<bold>Funding: </bold>This study was supported by the National Key Research and Development Program of China (grant number 2019YFC1315800). [ABSTRACT FROM AUTHOR]