Computational Approaches for RNA StructureFunction Determination

< < < < < < < < < < < < < < < < < < < < & >HDV</ &gt；我们的RNA结构分析工作台StructureLab和我们的RNA结构预测大规模并行遗传算法（MPGAfold）被应用于丁型肝炎病毒（HDV）的研究。HDV是一种与乙型肝炎病毒（HBV）相关的病毒。它在HBV存在下复制。HDV合并HBV会增加肝脏疾病的严重程度，并增加发生肝癌的可能性。丁型肝炎病毒产生一种蛋白质，即丁型肝炎抗原，它有两种形式，短形式和长形式。先前我们通过MPGAfold展示了HDV的厄瓜多尔菌株（ES）获得了两个对功能至关重要的二级结构。当HDV RNA达到分支构象时，它被编辑，将停止密码子变成色氨酸。随后，病毒转变为复制所必需的线性形式，导致翻译更长的抑制病毒合成的肽。有时，RNA绕过分支形式，达到线性复制形式，避免了编辑，导致HDV复制所需的肽更短。最近MPGAfold显示秘鲁HDV株（PS）与ES具有不同的折叠特征。ES更容易获得编辑结构。我们的合作者John Casey通过实验验证了这一点，并表明ES与编辑蛋白腺苷脱氨酶的结合效率低于PS。这些结果表明，HDV菌株在编辑状态和复制状态的形成之间保持了微妙的平衡。与NCI-CCR的Nancy Colburn博士合作，我们表征了受致癌翻译因子eIF4E调节的mrna的分子特性。真核生物的帽依赖翻译是由起始因子eIF4E介导的，其存在是核糖体募集所必需的。eIF4E在人类癌症中经常过表达，在动物模型中靶向eIF4E已显示出积极的治疗效果。激活eIF4E诱导编码癌蛋白的特定mrna的翻译增加。这种激活可能是由于AKT信号传导产生的促肿瘤信号。出现了一个问题，即什么特征会导致对eIF4E的过度响应。为了确定这些特征，我们分析了两个数据集，一个来自过表达eIF4E蛋白的细胞，另一个来自致癌AKT通路激活的细胞。通过比较由于eIF4E过表达而从单体转移到主动翻译多核糖体部分的mrna与不转移的mrna，可以确定在增加eIF4E暴露后，哪些转录本上调、下调或保持不变。我们的计算分析表明，上调的mrna平均具有更短的3' utr，更高的G+C含量，并且在开始密码子之前和停止密码子周围的RNA二级结构略多。microrna的结合位点也明显减少，而microrna是对eIF4E浓度升高高度敏感的mrna的肿瘤抑制因子。我们还设计了一个基于G+C含量、3' UTR长度和总长度的分类器，能够预测所提供数据集中的多染色体运动。< < < & >TCV</ & > 3' utr的细胞和病毒mrna含有通过增强翻译在基因表达中起作用的元件，其机制尚不清楚。为了确定这些元件的功能，我们使用了一种简单的模型病毒，萝卜皱病毒（TCV）。TCV以帽独立的方式翻译，包含一个3‘近端区域，与5’ UTR一起协同增强翻译。与马里兰大学的安妮·西蒙教授合作，我们正在破译这个3'元素的功能。我们使用mpgfold和Structurelab来鉴定一系列的发夹和一个假结，这些发夹和假结已经被遗传学证实。利用RNA二级结构信息和RNA2D3D（我们用于RNA 3D结构测定的分子建模软件），我们预测了一系列的三个发夹和两个假结在结构上类似于tRNA，这是自然界中发现的第一个内部tRNA样结构。利用这些信息，我们提出该元件的翻译增强可能涉及核糖体结合。该元件通过与60S核糖体亚基相互作用而与核糖体结合，这是首次发现与大亚基相互作用。从生物化学角度确定，这种trna样元素是一个结构开关的主要部分，它将模板从一个被翻译的转换为一个被复制的。Musashi1 （Msi1）是一种高度保守的RNA结合蛋白，在神经系统的干细胞维持和发育中起作用。有很好的证据表明Msi1与肿瘤形成有关。我们的合作者，德克萨斯大学健康科学中心的Luiz Penalva博士正在使用一种高通量方法来识别一组靶mrna，并阐明它们在干细胞维持、细胞分化和肿瘤发生中的作用。鉴定出与Msi1优先相关的mrna。本课题组应用计算数据挖掘技术，在Msi1靶基因的3' UTR中寻找调控信号和结构基序。我们建立了Msi1在这些由Msi1调控的mrna的3' UTR中的结合序列的两个模型。在这两个模型中，我们的程序EDscan发现了不同的RNA结构，这些结构高度稳定且具有统计学意义。实验证实它们与神经rna结合蛋白Msi1相互作用。由此可见，rna结合蛋白Msi1的调控与两个模型中显示的保守结合序列和结构信号密切相关。<P> < > < < > < &gt；伪能量最小化&lt；/b&gt；基于热力学过程的模拟算法通常将单个RNA序列的折叠自由能最小化，以预测其二级结构。额外使用来自多个序列比对的协方差分数可以提高这些预测的准确性。我们与新泽西理工学院的Jason Wang开发了一种算法RSpredict，它可以预测一组对齐序列的一致二级结构，该算法结合了动态规划原理和协变分数。我们可以相当准确地从一组对齐序列中预测出一致的二级结构。<P> <b&gt；结合核磁共振和SAXS&lt；/b&gt；通过核磁共振、x射线晶体学或其他实验技术来确定大的3D RNA结构一直是一个非常困难的问题。我们的团队与CCR的王云兴团队一起开发了一种方法，该方法将核磁共振（NMR）和小角度x射线散射技术与一个名为G2G的软件包相结合，以确定主要由a型螺旋组成的大型rna的整体结构。类a型螺旋构成了x射线晶体学确定的大部分结构，因此它们是RNA结构的主要组成部分。因此，确定螺旋的方向和螺旋轴周围的旋转以及螺旋的相对全局位置[摘要截短为7800个字符]

<b>Applications RNA Structure prediction and analysis</b> <P> <b>HDV</b> Our massively parallel genetic algorithm for RNA structure prediction (MPGAfold) and StructureLab, our RNA structure analysis workbench, were applied in a study of the Hepatitis Delta virus (HDV). HDV is a virus associated with the Hepatitis B virus (HBV). It replicates in the presence of HBV. HDV with HBV increases the severity of liver disease and enhances the likelihood of developing liver cancer. HDV produces one protein, the hepatitis delta antigen, which has two forms, the short and the long form. Previously we showed, with the use of MPGAfold, that the Ecuadorian strain (ES) of HDV attains two secondary structures that are crucial for functionality. The HDV RNA is edited when it attains a branched conformation, changing a stop codon into a tryptophan. Later, the virus changes into a linear form which is necessary for replication, leading to the translation of a longer peptide which inhibits viral synthesis. At times the RNA bypasses the branched form and attains the linear replication form, avoiding editing, resulting in a shorter peptide required for HDV replication. Recently, MPGAfold indicated that the Peruvian strain (PS) of HDV had different folding characteristics than ES. ES attained the editing structure more readily. Our collaborator John Casey verified this with experiments and showed that ES binds to its editing protein adenosine deaminase less efficiently than PS. These results showed that HDV strains maintain a delicate balance between the formation of the editing and replication states. <P> <b>eIF4E</b> In collaboration with Dr. Nancy Colburn from NCI-CCR we characterized the molecular properties of mRNAs that are regulated by the oncogenic translation factor eIF4E. Cap-dependent translation in eukaryotes is mediated by the initiation factor eIF4E, whose presence is required for ribosomal recruitment. eIF4E is often over-expressed in human cancer and targeting eIF4E in animal models has shown positive therapeutic effects. Activation of eIF4E induces increased translation of specific mRNAs that encode oncoproteins. This activation can be due to tumor promoting signals originating from AKT signaling. A question arises as to what characteristics enable over-responsiveness to eIF4E. To establish these features, we analyzed two datasets, one from cells over-expressing the eIF4E protein and the other from cells activated by the oncogenic AKT pathway. By comparing mRNAs that shift from monosomes into actively translating polyribosome fractions as a result of eIF4E overexpression, with mRNAs that do not shift, one can determine those transcripts that are up-regulated, down-regulated or remain about the same after increased eIF4E exposure. Our computational analysis showed that up-regulated mRNAs have on average shorter 3' UTRs, higher G+C content and slightly more RNA secondary structure before the start codon and around the stop codon. There is also apparent diminution of binding sites for microRNAs that are known to be tumor suppressors for mRNAs that are highly responsive to increased eIF4E concentration. We also designed a classifier based on G+C content, 3' UTR length and total length that is capable of predicting polysomal movement in the provided datasets. <P> <b>TCV</b> 3' UTRs of cellular and viral mRNAs harbor elements that function in gene expression by enhancing translation using as unknown mechanisms. To determine the function of these elements we used a simple model virus, Turnip crinkle virus (TCV). TCV is translated in a cap-independent fashion and contains a 3' proximal region that together with the 5' UTR synergistically enhances translation. In collaboration with Professor Anne Simon, from the University of Maryland, we are deciphering the function of this 3' element. We used MPGAfold and Structurelab to identify a series of hairpins and one pseudoknot that have been confirmed genetically. Using this RNA secondary structural information with RNA2D3D, our molecular modeling software for RNA 3D structure determination, we predicted that a series of three hairpins and two pseudoknots structurally resembled a tRNA, the first internal tRNA-like structure found in nature. Using this information, we proposed that translational enhancement by the element might involve ribosome binding. The element was found to bind ribosomes by interaction with the 60S ribosomal subunit, the first such interaction with the large subunit discovered. It was biochemically determined that this tRNA-like element is a major part of a structural switch that converts the template from one that is translated to one that is replicated. <P> <b>Musashi</b> Musashi1 (Msi1) is a highly conserved RNA binding protein with functions in stem cell maintenance and development of the nervous system. There is good evidence that links Msi1 to tumor formation. A high-throughput approach is being used by our collaborator, Dr. Luiz Penalva at the University of Texas Health Science Center to identify a group of target mRNAs and to elucidate their participation in stem cell maintenance, cell differentiation and tumorigenesis. mRNAs preferentially associated with Msi1 were identified. Our group applied computational data mining to find the regulatory signal and structural motif in the 3' UTR of Msi1 targeted genes. We developed two models for the binding sequence of Msi1 in the 3' UTR of these mRNAs regulated by Msi1. In both models distinct RNA structures which are highly stable and statistically significant were found with our program EDscan. These were experimentally confirmed to interact with the Neural RNA-binding protein Msi1. It thus appears that the regulation of the RNA-binding protein Msi1 is closely correlated with a conserved binding sequence and a structureal signal that is indicated in the two models. <P> <b>RNA Structure Prediction and Analysis Software:</b> <P> <b>Pseudo energy minimization</b> Simulation algorithms that are based on thermodynamic processes often minimize the free energy of folding of single RNA sequences to predict their secondary structures. The additional use of covariance scores derived from multiple sequence alignments can improve the accuracy of these predictions. We developed with Jason Wang at the New Jersey Institute of Technology, an algorithm, RSpredict, that predicts the consensus secondary structure of a set of aligned sequences that combines the principles of dynamic programming with covariation scores. We can predict fairly accurately a consensus secondary structure from the set of aligned sequences. <P> <b>Combining NMR and SAXS</b> The determination of large 3D RNA structures by NMR, X-ray crystallography or other experimental techniques has been a very difficult problem. Our group with Yun-Xing Wang's group in CCR, has developed a methodology that combines techniques from Nuclear Magnetic Resonance (NMR) and Small Angle X-ray scattering with a software package called G2G to determine the global architecture of large RNAs consisting mostly of A-form helices. A-form-like helices comprise a large percentage of the structures determined by X-ray crystallography thus they are a predominant building block for RNA structure. Therefore the determination of the orientation and the rotation of helices around their helical axes and the relative global positions of the he [summary truncated at 7800 characters]