Ruohan Zhang, Xinrong Yang, Guangming Li, Yinan Deng, J T Liu, Gao H, Jie Zhao, Jianwen Cheng, X L Zhao, Yi Yang, Zhen Wu, 谷双振, Yang Wu, Zhongying Ma, Liu Y, Yan Kang, Guangpeng Zhou, Hui Li, Yonghong Zhang, Xiaoliang Han, Jia Fan, Jian Zhou, 窦科峰, 陶开山
BACKGROUND AND AIMS: Existing imaging and serum-marker assays miss many early liver cancers, especially in high-risk chronic liver disease carriers. We aimed to create a highly accurate, non-invasive, methylation-based liquid biopsy for early detection. METHODS: We used a comprehensive, multi-platform, multi-cohort strategy for marker discovery, starting with methylation profiling of hepatocellular carcinoma samples from TCGA and in-house cohorts. From 30 initial candidates, nine highly liver-specific methylation markers were shortlisted, and three optimal cfDNA markers (RNF135, CHFR, PAX5) were selected to develop a robust diagnostic model, tuned in a training set (N = 280) and locked in an internal testing set (N = 124). The model was then validated in a prospective, large-scale trial conducted at four geographically distinct Chinese centres. RESULTS: The clinical trial included 1097 participants from two groups, (i) a diagnosing group (N = 646) that prospectively enrolled individuals without prior diagnostic results and represented a real-world high-risk population, and (ii) a diagnosed group recruited after pathology confirmation. Overall, the model achieved 94.43% (95% confidence interval, 92.12-96.09%) sensitivity and 95.16% (92.78-96.78%) specificity for liver cancer, with stage-I sensitivity of 93.10% (89.78-95.40%). Within the diagnosing group, overall sensitivity was 93.99% (91.28-95.90%), and for the 267 stage-I cases, it reached 92.88% (89.15-95.39%). As for specificity, it remained high across confounders: 92.78% (85.84-96.46%) in cirrhosis, 91.74% (85.46-95.45%) in other-cancer interference samples. Besides, the model outperformed the traditional liver cancer biomarker AFP and showed changes in methylation signals before and after surgery, suggesting a possible role in perioperative monitoring. Each centre independently reported sensitivities and specificities exceeding 90%, demonstrating robust geographic performance. CONCLUSIONS: Using a systematic marker-discovery pipeline and a multi-centre prospective cohort, we developed a methylation-based liquid biopsy that reliably detects early liver cancer in high-risk populations. CLINICAL TRIAL NUMBER: Chictr.org identifier: ChiCTR2400092883. KEY POINTS: Three cfDNA methylation markers, RNF135, CHFR and PAX5, were identified for liver cancer liquid biopsy. A three-marker diagnostic model based on qMSP was established for highly accurate non-invasive detection of liver cancer. The LC-HMC model achieved 94.43% sensitivity and 95.16% specificity in the clinical trial. The model detected stage-I liver cancer with a sensitivity of 93.10%.