Ultra-sensitive detection of hepatocellular carcinoma (HCC) with methylation signal enrichment of ctDNA and hepatitis B virus (HBV).

4136 Background: Aberrant methylation patterns in cell-free DNA (cfDNA) have been identified as effective biomarkers for HCC early detection, with circulating tumor DNA (ctDNA) from HCC patients exhibiting distinct methylation signatures. Additionally, HBV infection and the associated methylation alterations are closely linked to the development and progression of both cirrhosis and HCC. In this study, we utilize an ultra-sensitive Methylation Anchor Probe for Low Signal Enrichment (MAPLE) to enrich HCC-related methylation signals in ctDNA, as well as those from HBV genomes. By integrating these signals with a machine learning model, we achieve improved discrimination between HCC patients and non-cancer controls, while reducing false positives in individuals with cirrhosis. Methods: Whole blood samples were collected from 246 participants, including 96 HCC patients, 123 healthy controls, and 27 cirrhosis individuals. cfDNA was extracted from plasma, followed by enzymatic conversion and library preparation. Targeted hybrid capture was performed using a custom-designed panel that enriched methylation signals associated with HCC and HBV CpG islands. The final libraries were sequenced using next-generation sequencing (NGS). A machine learning model was developed, incorporating methylation features derived from both the human genomic regions and HBV CpG islands. Participants were randomly divided into training and test sets at a 3:1 ratio, with the training set undergoing 5-fold cross-validation for model optimization. To assess model robustness, 40 resampling iterations were conducted to evaluate performance in distinguishing HCC patients across various stages from non-cancer individuals. Results: Among all participants, 39.8% tested positive for HBV. Incorporating methylation features from the HBV genome into the model improved sensitivity for detecting early-stage HCC in HBV-positive individuals and enhanced accuracy in distinguishing early-stage HCC from cirrhosis. Analysis of selected HBV methylation features revealed hypermethylation in HCC patients compared to individuals with cirrhosis and healthy controls. The final machine learning model achieved a specificity of 97.6% (96.2%–97.9%). Sensitivities for detecting HCC across all stages were: I: 76.4% (73.5%–79.4%), II: 94.6% (92.0%–97.3%), III: 99.5% (98.8%–100.0%), and IV: 100.0% (100.0%–100.0%). For distinguishing cirrhosis, the model demonstrated a specificity of 81.9% (77.6%–86.3%). Conclusions: Using the ultra-sensitive MAPLE technique, we developed a novel panel that enriches methylation signals from both the human and HBV genomes. This assay significantly improved sensitivity for detecting early-stage HCC. By incorporating HBV genome features, we further enhanced the accuracy of distinguishing early-stage HCC from cirrhosis in HBV-positive individuals.