【作者】 王璇；

【导师】 张树义；

【作者基本信息】 华东师范大学 ， 生物化学与分子生物学， 2015， 硕士

【摘要】 基因表达的调控是一个复杂的过程,受到多种表观遗传学修饰和非编码基因的影响。其中组蛋白修饰和小RNA(miRNA)分别在基因转录前和转录后的调控过程中扮演着重要的角色。虽然许多研究致力于分别探索表观遗传修饰和miRNA对基因调控过程的作用,但是两者之间的协同作用尚缺乏综合的研究。在本研究中,我们利用最近发表的全基因组的组蛋白修饰数据和高质量的miRNA靶基因的数据,系统的研究了miRNA和表观遗传学调控的组合作用的关系。这些数据包括了miRTarBase数据库收集的miRNA与其靶基因的关系,以及ENCODE project等工作测定的不同细胞系中多种组蛋白修饰的ChIP-Seq的数据。我们选取了具有代表性和数据质量较好的CD4+ T cell细胞系作为首要研究对象,来观察miRNA靶基因和组蛋白修饰之间的关系。结果显示相比不被miRNA靶向的基因,miRNA的靶基因有着独特的组蛋白修饰模式。大多数受到miRNA调控的基因,启动子区域的39种组蛋白修饰中的大部分显著的区别于非靶基因的启动子区域,大部分的显著差异的特征是在miRNA的靶基因启动子区域组蛋白修饰信号更高,而只有少数组蛋白的甲基化在非靶基因启动子区域修饰信号更高。这些结果表明组蛋白修饰有着在miRNA靶基因上聚集的倾向。基于这一发现,我们提出一种机器学习的方法来预测一个基因是否是某个miRNA的靶基因。这种监督分类的方法(SVM模型)以已知确定性较高的miRNA的靶基因作为训练集,随机选取相同数目非靶基因作为负数据集,分析靶基因特有的组蛋白修饰组合信息,进而对全基因组水平的组蛋白修饰数据进行分类,可以将所有基因分为某个miRNA的靶基因或者非靶基因。我们在三个细胞系中验证了这种方法有更好的敏感性和特异性。相比传统上基于基因3’UTR序列互补而设计的miRNA的靶基因的预测,该方法给出了新颖的可能的靶基因。另外对预测的靶基因的序列分析显示,miRNA可能结合在靶基因的编码区域或者内含子上。最后,我们发现该方法预测的miRNA的靶基因有着明显的功能富集。相比基于序列互补性预测出的miRNA的靶基因,SVM预测给出的靶基因对应的GO富集程度和KEGG通路富集程度都显著较高。这可能意味着基于表观遗传学修饰富集程度来预测的niRNA的靶基因在执行相似的生物学功能。本研究是首次利用全基因组的数据来研究miRNA和组蛋白修饰的协同作用,研究结果对于理解基因的表达调控的复杂性提供了新的参考。更多还原

【Abstract】 Regulation of gene expression is a relatively complex process, which could be impacted by various epigenetic modification and non-coding RNAs. Both histone modifications and microRNAs (miRNAs) play pivotal role in gene expression regulation. Although numerous studies have been devoted to explore the gene regulation by miRNA or epigenetic regulations, their coordinated actions have not been comprehensively examined.In this work, we systematically investigated the combinatorial relationship between miRNA and epigenetic regulation by taking advantage of recently published whole genome-wide histone modification data and high quality miRNA targeting data. Those data include mirTarBase high-confident miRNA-mRNA targeting relationship and a number of ChIP-Seq based different type histone modification collected in several cell lines. We observed relationship between miRNA and histone modification primarily in CD4+ T cell, data in which contains the most complete histone modification type.The results showed that miRNA targets have distinct histone modification patterns compared with non-targets in their promoter regions. Most of 39 histone modifications of miRNA targets have significantly different signal in promoter regions compared to non-target genes. Only few methylations of histone proteins have higher signal in non-targets genes while other histone modification have higher miRNA target genes. Those results indicate a concentration of histone modification pattern in miRNA target genes.Based on this finding, we proposed a machine learning approach to fit predictive models on the task to discern whether a gene is targeted by a specific miRNA. We found a considerable advantage in both sensitivity and specificity in diverse human cell lines. The SVM model takes high confidence miRNA target genes as training set. After screen miRNA target specific histone modification pattern, SVM give predictions of target genes of each miRNA. Compared to other sequenced bases software result, SVM give a number of novel miRNA target as well as potential new miRNA target sites located in gene intron and coding regions.Finally, we found that our predicted miRNA targets are consistently annotated with Gene Ontology terms, which may imply miRNA target genes predicted by SVM model likely located in similar biological processes.Our work is the first genome-wide investigation of the coordinated action of miRNA and histone modification regulations, which provides a guide to deeply understand the complexity of transcriptional regulation.更多还原