标题：基于DNA甲基化的分子亚型可预测肝细胞癌患者的预后

      作者：杨阮阮，李锦忠，龚晓兵    暨南大学附属第一医院感染科，广东 广州 510632

      卷次： 2020年31卷22期

      【摘要】 目的 研究甲基化位点联合基因表达预测肝细胞癌患者预后的作用。方法 基于癌症基因组图谱数据库(TCGA)的424例肝细胞癌样本，通过COX比例风险回归分析筛选出具有独立预后能力的CpG位点，然后利用一致性聚类方法进行肿瘤分型，对差异CpG位点相应的启动子基因进行功能富集分析，最后根据分型差异分析构建甲基化位点的预后模型，以及Suivival、ROC曲线判断风险模型的临床应用价值。结果 在癌组织和正常组织中成功筛选出具有独立预后能力的2 248个差异甲基化位点，利用一致性聚类方法得到7个肿瘤亚组，亚组间的预后差异有统计学意义(P<0.05)，亚组间临床特征如TMN分期、年龄、病理分期(Stage)和组织学分级(Grade)各不相同，这种临床差异和预后也密切关联。对差异甲基化位点相应的启动子基因进行功能富集分析，它们主要涉及RNA运输、细胞周期、p53信号通路和剪接体。多变量Cox回归构建肿瘤预后风险模型：风险评分=4.98×cg05489143-21.18×cg09600437+3.50×cg19165652+4.59×cg19569208+11.08×cg22732432+5.07×cg22958262-16.02×cg24153171+4.75×cg2545598。根据该模型计算，随着风险评分的增高，高风险组的生存时间虽然没有明显下降，但是死亡率明显升高。另外，高低风险两组的预后具有显著差异，同时，ROC曲线的AUC值为0.822。结论 在癌组织和癌旁组织中成功筛选出2 248个差异甲基化位点并通过一致性聚类方法得到7个肿瘤亚组，根据分型差异分析构建甲基化位点的预后模型，该模型可以很好地预测患者的存活率。
      【关键词】 DNA甲基化；肝细胞癌；TCGA数据库；预后预测模型；生物信息学
      【中图分类号】 R735.7 【文献标识码】 A 【文章编号】 1003—6350（2020）22—2870—07

Prediction of prognosis in patients with hepatocellular carcinoma based on molecular subtypes of DNAmethylation.
YANG Ruan-ruan, LI Jin-zhong, GONG Xiao-bing. Department of Infectious Disease, the First AffiliatedHospital of Jinan University, Guangzhou 510632, Guangdong, CHINA
【Abstract】 Objective To study the effect of methylation sites combined with gene expression analysis in pre-dicting the prognosis of patients with hepatocellular carcinoma. Methods We use the Cox proportional hazard regres-sion analysis to screen the CpG sites with independent prognostic function from 424 cases of hepatocellular carcinomasamples based on The Cancer Genome Atlas (TCGA), and then the tumor classification was performed by consistencyclustering method. Next, promoter genes of different CpG sites were analyzed by the functional enrichment analysis. Fi-nally, the prognosis model of methylation sites was constructed according to the difference analysis of genotypes, andthe clinical application value of the risk model was judged by Suivival and ROC curves. Results A total of 2 248 differ-entially methylated sites with independent prognostic function were successfully screened out from cancer and normaltissues. Seven tumor subgroups were obtained by consistency clustering method. The prognosis between the subgroupswas significantly different (P<0.05). The clinical features such as TMN staging, age, pathological stage, and histologicalgrading were different among subgroups, and the difference were closely related to prognosis. According to the function-al enrichment analysis, we found that promoter genes of different methylation sites were mainly involved in RNA trans-port, cell cycle, p53 signal pathway, and spliceosome. Multivariate Cox regression was used to construct the risk modelof tumor prognosis: risk score=4.98 × cg05489143-21.18 × cg09600437 + 3.50 × cg19165652 + 4.59 × cg19569208 + 11.08 ×cg22732432+5.07×cg22958262-16.02×cg24153171+4.75×cg2545598. According to the model, with the increase of riskscore, the survival time of high-risk group did not significantly decrease, but the mortality rate increased significantly. Inaddition, there was significant difference in the prognosis between the two groups, and the AUC value of ROC curve was0.822. Conclusion A total of 2 248 differentially methylated sites were successfully screened from cancer tissues andadjacent tissues, and 7 tumor subgroups were obtained by consistency clustering method. The prognosis model relatedto methylation sites was constructed by the difference analysis of genotypes, which can well predict the survival rateof patients.
      【Key words】 DNA methylation; Hepatocellular carcinoma; TCGA database; Prognosis prediction model; Bioin-formatics