AUTHOR=Li Kang , Song Yi , Qin Ling , Li Ang , Jiang Sanjie , Ren Lei , Zang Chaoran , Sun Jianping , Zhao Yan , Zhang Yonghong 

TITLE=A CpG Methylation Signature as a Potential Marker for Early Diagnosis of Hepatocellular Carcinoma From HBV-Related Liver Disease Using Multiplex Bisulfite Sequencing

JOURNAL=Frontiers in Oncology

VOLUME=Volume 11 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2021.756326

DOI=10.3389/fonc.2021.756326

ISSN=2234-943X

ABSTRACT=Background: Aberrant methylation of CpG sites severed as epigenetic marker for building diagnostic, prognostic, and recurrence models for hepatocellular carcinoma (HCC). 
Methods: Using Illumina 450K and EPIC Beadchip, we identified 34 CpG sites in peripheral blood mononuclear cell (PBMC) DNA that were differentially methylated in early HCC versus HBV-related liver diseases (HBVLD). We employed multiplex bisulfite sequencing (MBS) based on next-generation sequencing (NGS) to measure methylation of 34 CpG sites in PBMC DNA from 654 patients that were divided into a training set (n = 442), a test set (n = 212). Using training set, we selected and built a six-CpG-scorer (including cg14171514, cg07721852, cg05166871, cg18087306, cg05213896, and cg18772205), applying least absolute shrinkage and selection operator (LASSO) regression. We performed multivariable analyses of four candidate risk predictors (including six-CpG-scorer, age, sex, AFP level), using 20 times imputation of missing data, non-linearly transformed and backwards feature selection with logistic regression. The final model’s regression coefficients were calculated according to “Rubin's Rules”. The diagnostic accuracy of model was internally validated with 10000 bootstrap validation dataset, and then applied to the test set for validation.
Results: The area under the receiver operating characteristic curve (AUROC) of the model was 0.81(95% CI, 0.77-0.85) and it showed good calibration, decision curve analysis. Using enhanced bootstrap validation, adjusted C-statistics and adjusted brier score was 0.809 and 0.199, respectively. The model also showed AUROC value of 0.84 (95% CI 0.79-0.88) of diagnosis for early HCC in test set. 
Conclusions: Our model basing on six-CpG-scorer was a reliable diagnosis tool for early HCC from HBVLD. The usage of MBS method can realize large-scale detection of CpG sites in clinical diagnosis of early HCC and benefit the majority of patients.