【作者】 孙晓慧；

【导师】 朱益民；

【作者基本信息】 浙江大学 ， 流行病与卫生统计学， 2019， 博士

【摘要】 研究背景和目的结直肠癌(colorectal cancer,CRC)是常见的消化道恶性肿瘤之一。世界范围内,结直肠癌的发病率和死亡率居常见恶性肿瘤的第三位和第四位。在我国,结直肠癌的发病率和死亡率正不断地增加,已然成为重大公共卫生问题。由于早期诊断对结直肠癌患者预后的判定、治疗方案的选择有着重要的意义。因此,寻找出能够早期发现结直肠癌的肿瘤标志物显得尤为重要。结直肠癌的发生发展是一个涉及多因素、多步骤的复杂过程,除与生活方式、饮食习惯等环境因素相关外,其主要发病机制涉及到一系列遗传学和表观遗传学的积累改变。DNA甲基化是重要的表观遗传调节机制之一,并与结直肠癌的发生发展密切相关。基因启动子区域的异常甲基化导致肿瘤抑癌基因的转录沉默和原癌基因的转录激活,可促进结直肠癌的发生发展。研究DNA甲基化的异常改变可以为发现结直肠癌相关原癌基因或抑癌基因提供新线索,对于阐明结直肠癌发生的分子机制有着重要的意义。同时,亦可为结直肠癌的诊断、治疗和预后等方面提供新思路。因此,为了系统地探索结直肠癌的甲基化分子标志物,本研究首先基于全基因组DNA甲基化和表达谱筛选结直肠癌相关分子标志物,随后对筛选的标志物进行结直肠癌鉴别诊断模型和预后指数模型的构建。探讨DNA甲基化异常对结直肠癌发生发展的影响,及作为分子标志物的应用前景。材料和方法筛选阶段收集了 2006-2012年于浙江大学医学院附属邵逸夫医院就诊并接受手术的6例CRC患者肿瘤组织和癌旁配对的正常组织标本,运用甲基化免疫共沉淀测序(Methylated DNA Immunoprecipitation Sequencing,MedIP-seq)和转录组测序(RNA Sequencing,RNA-seq)进行基因甲基化水平和转录水平的检测。根据测序结果,选取了甲基化和表达存在差异的基因(差异基因)。随后,利用癌症和肿瘤基因图谱(The Cancer Genome Atlas,TCGA)中45对肿瘤组织和配对的癌旁正常组织的表达和甲基化数据对差异基因进行一期验证。二期验证阶段收集了浙江台州医院于2006-2011年经病理确诊的122例CRC患者的临床资料和组织标本,并进行随访。运用多重目的区域甲基化富集测序技术(MethylTarget)进行CpG位点甲基化的检测。最终挑选出甲基化水平差异显著的基因和CpG位点进行后续研究。应用荧光定量PCR和MethylTarget来检测经甲基化转移酶抑制剂5-氮杂-2’-脱氧胞苷(5-Aza-2’-deoxycytidine,5-Aza-dC)处理前后,基因的表达和甲基化程度在结直肠癌细胞系中改变情况。采用脂质体转染法,构建敲除、过表达目的基因的细胞系;采用细胞计数试剂盒(Cell Counting Kit-8,CCK-8)和细胞克隆实验检测细胞增殖,采用Transwell小室实验检测细胞迁移和侵袭能力。采用套索(Least absolute shrinkage and selection operator,LASSO)回归进行鉴别诊断模型的构建,以模型中有意义的预测变量β值计算甲基化风险评分(methylation risk score,MRS),并利用GEO数据对MRS进行验证。运用受试者工作特征曲线(receiver operating characteristic curve,ROC)曲线方法计算 ROC 曲线下面积(Area Under Curve,AUC)和不同切点值的灵敏度和特异度及约登指数。采用Kaplan-Meier法拟合生存曲线,Log-Rank检验比较生存率的差别。采用Cox比例风险回归模型进行多因素生存分析,根据Cox 比例风险回归模型给出的回归系数计算每个患者的预后指数(prognostic index,PI);使用时间依赖性ROC曲线下面积(Area Under Curve(time),AUC(t))、整体鉴别指数(Integrated Discrimination Improvement,IDI)和净重分类指数(Net Reclassification Improvement,NRI)对模型预测效果进行评估。研究结果第一部分基于全基因组DNA甲基化和表达谱筛选结直肠癌相关分子标志物通过全基因组测序和差异分析后,我们筛选了基因表达和启动子区甲基化水平存在明显差异的53个基因,其中21个基因的表达和甲基化呈现负相关。利用TCGA数据对上述基因进行验证,发现13个基因验证的结果与测序结果一致。我们随后在122例CRC患者的癌组织与匹配的癌旁组织中对13个基因启动子区的特征甲基化位点进行验证。最终,我们发现C14orf159、CADM3、CNRIP1、GRIA4、GSTM2、NRXN1、PRKG2和RSP02在肿瘤组织中表达下降、甲基化水平升高,而CEACAM6、GRHL2、SRPRB在肿瘤组织中表达上升、甲基化水平降低。在经过 5-Aza-dC 处理 CRC 细胞系后,CADM3、CEACAM6、GRIA4、NRXN1、PRKG2和RSP02的甲基化水平明显降低,而其中CEACAM6、GRIA4、NRXN1的mRNA表达水平在HT29、DLD1、SW620和HCT8中均有明显的升高(P<0.05)。随后,我们在敲除CNRIP1和GSTM2的DLD1和HCT8细胞系中发现,细胞的增殖能力增强。而CNRIP1和GSTM2的表达改变未发现对CRC细胞迁移和侵袭能力产生影响。第二部分基于差异DNA甲基化位点的结直肠癌鉴别诊断模型的构建和评估我们利用LASSO回归模型从差异CpG位点中筛选了 9个结直肠癌特征位点(cg 15442728、cg06907680、cg26813458、cg08157672、cg20342184、cg23559689、cg12647497、cg01508023和cg14733048),基于这9个独立特征位点的甲基化水平构建了结直肠癌诊断模型。随后,针对独立特征位点和鉴别模型我们利用GEO数据库(GSE48684和GSE42752)进行了验证,发现9个位点的甲基化水平在癌组织和正常组织间存在明显的差异(FDR<0.05)。在GSE48684和GSE42752中该模型的AUC分别为0.966和0.993,对结直肠癌肿瘤组织和正常组织的鉴别准确性较高。第三部分基于差异DNA甲基化位点的结直肠癌预后指数模型的构建和评估我们利用Cox比例风险回归模型,调整年龄、性别、TNM分期和放化疗情况后,发现4个CpG位点的甲基化水平与预后相关。随后在逐步Cox回归分析的基础上,构建了基于cg25639415和cg12754421甲基化水平的预后指数模型,并对模型的效能进行评估和验证。经验证发现预后指数对结直肠癌患者1年、3年和5年生存情况的区分能力较好,AUC(t)分别为0.613、0.659和0.770。对患者5年生存情况预测时,该模型的IDI和NRI较TNM分期改善效果明显(P=0.013)。将预后指数与TNM分期联合构建联合预测指数(combining predictor,CP)评判患者生存情况,发现在评判患者5年生存情况时,联合预测指数比单用TNM分期可增加11.26%的曲线下面积。结论C14orf1 59、C ADM3、CEAC AM6、CNRIP1、GRHL2、GRIA4、GSTM2、PRKG2、RSP02、NRXN1和SRPRB的启动子区甲基化水平影响其基因的表达,并与CRC的发生发展存在一定关系,是潜在的生物标志物。利用9个CpG位点的甲基化水平所构建结直肠癌鉴别诊断模型,对肿瘤组织和正常组织的鉴别诊断效能良好,模型拟合较优。基于2个CpG位点构建的预后指数模型,在评判结直肠癌患者术后5年内生存情况时具有较高的准确性;联合TNM分期与预后指数对患者的预后评价更加全面和准确。更多还原

【Abstract】 Background and ObjectivesColorectal cancer(CRC)is one of the common malignancy of the digestive tract.Worldwide,the incidence and mortality of colorectal cancer ranks third and fourth in common malignancies.With the advancment of living standards and changes in diet,the incidence and mortality of colorectal cancer are increasing continuously,which has become a major public health problem in China.As early diagnosis and prompt treatment are favorable for patients’ prognosis,it is particularly important to identify and find biomarkers to detect CRC at an early stage.The occurrence and progression of CRC is considered as a complex progress involving multiple factos,such as lifestyle,diet habits and other environmental factors.Except them,its main pathogenesis involves a series of genetic and epigenetic changes.DNA methylation as one of the important epigenetic regulation mechanisms,has been found to play an important role in CRC.Abnormal methylation in gene promoter regions can lead to transcriptional silencing of tumor suppressors or transcriptional activation of proto-oncogenes.Therefore,studying on DNA methylation may give hints to discover new CRC-realated genes,and elucidate the molecular mechanism of CRC.Moreover,it can provide new ideas for the diagnosis,treatment and prognosis of CRC.In present study,we focused on DNA methylation biomarkers and explored their impact on the development of CRC and applications in early diagnosis asmolecular markers.Materials and MethodsIn discovery stage,6 cases of CRC patients with tumor tissue and adjacent normal tissues were collected from the Sir Run Run Shaw Hospital of School of Medicine from 2006 to 2012.Methylated immunoprecipitation sequencing(MedIP-seq)and transcriptome sequencing(RNA-seq)were used to detect gene methylation levels and transcription levels.The differential genes were then validated using the expression and methylation data of 45 pairs of tumor tissues and matched para-cancer normal tissues in the Cancer and Tumor Gene Map(TCGA).Afterward,we used multiple target region methylation enrichment sequencing(MethylTarget)to test the methylation level of specific CpG sites in 122 subjects with a diagnosis of CRC in Zhejiang Taizhou Hospital from 2002 to 2011.Finally,genes and CpG sites with significant differences in methylation levels were selected for follow-up studies.Fluorescence quantitative PCR and MethylTarget were used to detect the changes in gene expression and methylation status in colorectal cancer cell lines before and after treatment with 5-aza-2’-deoxycytidine(5-Aza-dC).CCK8 was used to detec cell proliferation and transwell assay was performed to measure migration and invasion.LASSO Logistic regression was used to dimension reduction and construct a model for discriminating CRC based on methylation levels.The methylation risk score(MRS)was calculated according to the meaningful predictor β value in the model,and was further validated in GEO data.Area under the curve(AUC),sensitivity,specificity and Youden index were computed based on the receiver operating characteristic(ROC)curve analysis.The survival curve was fitted by Kaplan-Meier method,and the difference in survival rate was compared by Log-Rank test.Multivariate survival analysis was performed using Cox proportional hazard regression model.The prognostic index(PI)of each patient was calculated according to the regression coefficient given by Cox proportional hazard regression model.The model prediction effect was evaluated using the area under the time-dependent ROC curve(AUC(t)),overall discrimination index(IDI),and net weight classification index(NRI).ResultsPart I Genome-wide methylation and expression profiling identify methylation-associated genes in colorectal cancerAfter genome-wide sequencing and analyzing,we screened 53 genes with differential methylation and mRNA expression.According to the common phenomenon that levels of methylation in the promoter usually inversely associated with the gene expression at mRNA level,21 genes were selected for the further analyses.We then performed the validation on differential levels of regional methylation and expression of these 21 screened genes between the tumor and normal tissues with TCGA data,and 13 genes were identified as candidate genes.Next,we performed the validation at specific CpG sites of these 13 candidate genes in additional 122 Chinese CRC patients.We found that the expression of C14orf159,CADM3,CNRIP1,GRIA4,GSTM2,NRXN1,PRKG2 and RSP02 were high in tumor tissues,while low methylation in their promoters.CEACAM6,GRHL2 and SRPRB were down-regulated in tumor tissues with high methylation levels.After treatment with 5-Aza-dC,the methylation levels of CADM3,CEACAM6,GRIA4,NRXN1,PRKG2 and RSP02 were decreased,while CEACAM6,GRIA4 and NRXN1 exhibited higher expression in the four CRC cell lines.Finally,we selected CNRIP1 and GSTM2 to explore their biological function in CRC cell lines and found that CNRIP1-deficient and GSTM2-deficient could promote cell proliferation.While,we did not find their impact on cell migration and invation.Part II A methylation-based assay for differentiating normal and tumor tissues of colorectal cancerAfter LASSO regression,nine independent CRC-related CpG sites(cg1 5442728,cg06907680,cg26813458,cg08157672,cg20342184,cg23559689,cg12647497,cg01508023 and cg14733048)were screened out.Based on these nine CpGs,a discriminating model was constructed for distinguishing tumor tissue from normal tissue.Subsequently,it was verified in the GEO database(GSE48684 and GSE42752).Using this model,we obtained AUC values of 0.966 and 0.993 for GSE48684 and GSE42752,respectively.Part Ⅲ A methylation-based predictor for prognosis of colorectal cancerUsing the Cox proportional hazard regression model,methylation levels of four CpG sites were found to be associated with CRC prognosis,after adjusting for age,gender,TNM stage,and chemoradiotherapy.Then based on the stepwise Cox regression analysis,a prognostic index model based on cg25639415 and cg12754421 methylation level was constructed,and the efficacy of the model was further evaluated and verified.We found that prognostic index(PI)has a better ability to distinguish between 1-,3-,and 5-year survival of colorectal cancer patients,with AUC(t)of 0.613,0.659,and 0.770,respectively.The IDI and NRI of the model were significantly improved compared with the TNM stage(P=0.013).A combined predictive predictor(CP)was construced by combining PI and TNM stage to predict patient’s survival.It was found that CP can increase the AUC(t)by 11.26%than the TNM stage alone when predicting patient’s 5-year survival.ConclusionAfter multiple validation,we identified 11 genes which were aberrantly regulated by DNA methylation in tumors and showed as promising biomarkers for CRC.The promoter region methylation may be one of the mechanisms of CEACAM6,GRIA4,NRXN1 silence in CRC.GSTM2 and CNRIP1,which exhibited high methylation levels in cancerous tissues,were mainly responsible for cell proliferation and acted as potential tumor suppressor genes in CRC.Using methylation levels of nine CpG sites,we developed a discriminating CRC model,which had high sensitivity and specificity for discriminating normal and tumor tissues and might have potential practical value.The prognostic index we established has a high accuracy in predicting the 5-year survival of patients with colorectal cancer.The combined prognosis predictor is more comprehensive and accurate to predict patients’ survival.更多还原