基于基因表达谱的结直肠癌分子分型和预后评估

发布时间：2018-08-31 09:02

【摘要】：研究背景及目的结直肠癌是世界范围内常见的恶性肿瘤之一,其发病率和死亡率分别位于恶性肿瘤的第三位和第四位。我国结直肠癌的死亡率也高居于恶性肿瘤死亡率的第四、五位之间,并呈增加的趋势,已严重危害人类的生命健康。结直肠癌在分子形成机制和病理学形态上都呈现高异质的特点,这对结直肠癌的诊断、治疗方案的选择、预后的评估是一个巨大的挑战。从分子生物学的角度看,肿瘤是由于某些染色体上的DNA损伤致使细胞内基因异常表达,导致细胞生长失控、缺乏分化而异常增生的一类复杂遗传性疾病。研究肿瘤基因表达谱、选取特征信息基因是解释肿瘤发生发展机制,寻找治疗靶点和预后标记的最直接手段。因此本研究旨在研究结直肠癌肿瘤组织与正常组织的基因表达谱,挑选差异表达基因,构建结直肠癌鉴别诊断模型;将非监督聚类分析与遗传学分析相结合,对结直肠癌进行分型,并分析不同分型的基因表达模式及其与临床病理指标的相关性;尝试用预后指数对结直肠癌患者进行预后评估,并建立预后指数分级,比较预后指数分级与经典TNM分期在预后评判上的差别;尝试构建TNM分期和预后指数分级联合预测模型进行预后评估。材料与方法收集2006年9月到2012年2月在某医院确诊并接受手术治疗的127例结直肠癌患者,收集其术后切除的结直肠癌组织和正常组织标本,同时收集包括性别、手术时的年龄、肿瘤发生的位置、手术时局部浸润情况、有无淋巴结转移和远处转移等资料。选取6对结直肠癌肿瘤组织和正常组织进行转录测序,在差异表达的基因中,选取表达差异倍数大于10的97个基因,然后进行扩大样本验证,最终挑选出差异表达显著、方向与测序一致的75个基因进行后续研究。采用寿命表法计算累积生存率,Kaplan-Meier法进行单因素生存分析;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;采用Logistic回归模型对二分类观察结局数据进行多因素分析;采用χ2检验或者Fisher's精确检验进行组间构成比比较;采用Wilcoxon秩和检验对独立非正态数据进行比较分析;采用ROC曲线分析评价LASSO回归模型和Logistic回归预测模型的特异度和灵敏度;采用ROC曲线分析评价预后指数在生存评判中的价值;采用非监督聚类方法进行分型研究。研究结果本研究发现与RNA-seq结果一致的75个结直肠癌肿瘤组织和正常组织差异表达基因中有13个基因的表达与结直肠癌患者预后相关,CPNE8、LOC646627、CDKN2A、ATP6V1A、CA1、SCARA5、BEST4、SCNN1B、KLF9 的高表达对预后不利,DNMT3B、ANLN、DNMT1、DNMT3A的高表达对病人预后有利。将以上13个基因纳入到Cox比例风险回归模型进行多因素分析后发现有5个基因与患者预后独立相关:DNMT3B的高表达对预后独立有利,LOC646627、SCARA5、CDKN2A、ATP6V1A的高表达对预后独立不利。将LASSO分析应用于多元线性回归模型,挑选出18个结直肠癌特征基因(MLH1、PLOD3、TGM2、ATP6V1A、SQLE、MET、S100P、MT1M、BEST4、CA7、LOC646627、ANPEP、P2RX1、FOXF2、GAB3、ABI3BP、SCARA5、ADAMDEC1),利用这18个基因构建的结直肠癌鉴别诊断模型可以将肿瘤组织和正常组织区分开,特异度为96.85%,灵敏度为98.43%,准确度达97.6%。本研究分别用75个差异表达基因、13个预后相关基因、5个预后独立相关基因对127例结直肠癌患者进行聚类分析,结果发现用13个基因和5个基因都可以将患者分成预后不同的两类,但当校正年龄、性别及TNM分期之后,发现5个基因的聚类结果仍与患者预后相关,其中DNMT3B在第一分型中高表达,SCARA5、LOC646627、CDKN2A在第二分型中高表达,并且第一分型的预后优于第二分型。利用Cox比例风险回归模型提供的回归系数计算127例患者的预后指数,对患者进行预后评估。发现用5个预后独立相关基因计算的预后指数(PI-5gene),对患者1年、3年、5年生存情况的评判与13个预后相关基因计算的预后指数(PI-13gene)无异,但优于TNM分期。PI-5gene评判患者1年、3年、5年生存情况的曲线下面积分别为0.719、0.772、0.772。将PI-5gene与TNM分期联合构建联合预测因子评判患者1年、3年、5年生存情况,比单用TNM分期分别增加20.98%、29.51%、26.77%的曲线下面积。将预后指数分级加入到TNM分期预后预测模型中,通过不基于风险等级的重分类改善指数(cfNRI)评价预后指数分级的实际生存预测能力,发现预后指数分级的加入可以显著改善模型的1年、3年、5年预后预测能力(P0.001),cfNRI 分别为 0.381、0.507 和 0.465。研究结论利用LASSO筛选出的结直肠癌特征基因,同时构建结直肠癌鉴别诊断模型,对肿瘤组织和正常组织的区分具有较高的准确性,并且该模型的特异度、灵敏度及准确度均高于传统的Logistic回归预测模型。基于结直肠癌肿瘤组织和正常组织的差异基因表达谱,用非监督聚类方法可以对结直肠癌进行分型研究,分型结果对结直肠癌的发生发展具有一定的解释力。我们建立的预后指数在评判结直肠癌患者术后1年、3年、5年生存情况时具有较高的准确性;在此基础上建立的预后指数分级较TNM分期在预后评判上有更高的准确性;联合TNM分期与预后指数分级对患者的预后评价更加全面和准确。
[Abstract]:Background and Objective Colorectal cancer is one of the most common malignant tumors in the world. The morbidity and mortality of colorectal cancer rank the third and fourth in malignant tumors respectively. Colorectal cancer is characterized by high heterogeneity in molecular formation mechanism and pathological morphology, which is a great challenge to the diagnosis, treatment and prognosis of colorectal cancer. The study of tumor gene expression profiles and the selection of characteristic information genes are the most direct means to explain the mechanism of tumorigenesis and development and to find therapeutic targets and prognostic markers. Differentially expressed genes were used to construct the differential diagnosis model of colorectal cancer; unsupervised clustering analysis and genetic analysis were combined to classify colorectal cancer, and the gene expression patterns of different types and their correlation with clinicopathological indexes were analyzed; prognostic index was used to evaluate the prognosis of colorectal cancer patients, and prognostic index was established. Methods 127 patients with colorectal cancer who were diagnosed and treated in a hospital from September 2006 to February 2012 were collected and their postoperative colorectal resection colorectal cancer were collected. Six pairs of colorectal cancer tissues and normal tissues were transcribed and sequenced, and the differentially expressed genes were selected. Seventy-seven genes larger than 10 were selected and validated by enlarged sample. Seventy-five genes with significant difference in expression and consistent direction and sequence were selected for follow-up study. The regression coefficients given by the case risk regression model were used to calculate the prognostic indices of each patient; Logistic regression model was used to analyze the multivariate analysis of binary observation outcome data; _2 test or Fisher's exact test was used to compare the composition ratio among groups; Wilcoxon rank sum test was used to compare and analyze the independent non-normal data; and R. OC curve analysis was used to evaluate the specificity and sensitivity of LASSO regression model and Logistic regression prediction model; ROC curve analysis was used to evaluate the value of prognostic index in survival assessment; unsupervised clustering was used to classify colorectal cancer tissues. The results of this study showed that 75 colorectal cancer tissues and normal groups were consistent with the results of RNA-seq. The overexpression of 13 genes was associated with the prognosis of colorectal cancer patients. The overexpression of CPNE8, LOC646627, CDKN2A, ATP6V1A, CA1, SCARA5, BEST4, SCNN1B, KLF9 was unfavorable to the prognosis of colorectal cancer patients. The overexpression of DNMT3B, ANLN, DNMT1, DNMT3A was beneficial to the prognosis of colorectal cancer patients. Five genes were independently associated with prognosis: high expression of DNMT3B was independently associated with prognosis, and high expression of LOC646627, SCARA5, CDKN2A, ATP6V1A was independently associated with prognosis. LASSO analysis was applied to multiple linear regression model to select 18 colorectal cancer characteristic genes (MLH1, PLOD3, TGM2, ATP6V1A, SQLE, MET, S100P, MT1M, BEST4, C). A7, LOC646627, ANPEP, P2RX1, FOXF2, GAB3, ABI3BP, SCARA5, ADAMDEC1 were used to construct a differential diagnosis model for colorectal cancer. The specificity, sensitivity and accuracy of the model were 96.85%, 98.43% and 97.6% respectively. Seventy-five differentially expressed genes, 13 prognostic-related genes and 5 prognosis-related genes were used in this study. Cluster analysis of 127 patients with colorectal cancer by independent related genes showed that 13 genes and 5 genes could be used to classify the patients into two groups with different prognosis. However, when adjusted for age, sex and TNM stage, the clustering results of 5 genes were still related to the prognosis of patients. DNMT3B was highly expressed in the first type, SCARA5 and LOC were high in the first type. 646627, CDKN2A was highly expressed in the second subtype, and the prognosis of the first subtype was better than that of the second subtype. PI-5 gene was superior to TNM staging in predicting 1 year, 3 years, and 5 years survival. The sub-curve scores of PI-5 gene and TNM staging were 0.719, 0.772 and 0.772, respectively. The prognostic index classification was added to the TNM prognostic prediction model. The actual survival prediction ability of the prognostic index classification was evaluated by the non-risk-based reclassification improvement index (cfNRI). It was found that the prognostic index classification could significantly improve the one-year, three-year, and five-year survival prediction ability of the model. The annual prognostic predictive ability (P 0.001) and cfNRI were 0.381, 0.507 and 0.465, respectively. Conclusion LASSO screening of colorectal cancer characteristic genes, and the construction of colorectal cancer differential diagnosis model, tumor tissue and normal tissue with high accuracy, and the specificity, sensitivity and accuracy of the model are higher than the traditional ones. Logistic regression model. Based on the differential gene expression profiles of colorectal cancer tissues and normal tissues, unsupervised clustering method can be used to classify colorectal cancer. The typing results have some explanatory power for the occurrence and development of colorectal cancer. The prognostic indices were more accurate than TNM staging, and the prognostic evaluation combined with TNM staging and prognostic index grading was more comprehensive and accurate.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R735.34

【相似文献】