基于RCF算法的表位研究及一株鼠疫中和抗体表位鉴定

发布时间：2018-01-16 17:10

本文关键词：基于RCF算法的表位研究及一株鼠疫中和抗体表位鉴定　出处：《中国人民解放军军事医学科学院》2017年博士论文　论文类型：学位论文

【摘要】：抗体是在抗原和免疫系统的相互作用下,由B淋巴细胞转化的浆细胞产生的能与相应抗原发生特异性结合的免疫球蛋白。当抗体与这些抗原结合时,抗体上CDR区与抗原上的某个区域,即抗原决定簇(antigen determinant)结合。这里的抗原决定簇就是该抗体结合的表位。而结合表位的抗体CDR区上的氨基酸残基构成抗体上的配位。表位分为线性表位和构象表位。线性表位由一段连续的氨基酸位点组成;而构象表位通常由一些在抗原一级序列上离散,但是在空间结构中相互靠近的位点共同构成。所以,抗原分子结构的变化可能会显著影响抗体结合构象表位,但是对于线性表位影响并不大。表位是抗体最为重要的性质之一,也是研究者最想获得的抗体信息。通过抗体的表位,可以获得抗体发挥保护作用的机制,研究病原体的致病机制,并能够以表位为基础,反向研究激发保护性抗体的疫苗。目前主流的研究抗体表位的方法是通过实验来鉴定。这些实验方法有些对实验条件与设备要求高,有些方法工作量大,有些方法则成功率低。随着计算机性能不断增强以及模拟计算的方法不断成熟,出现了一些应用在生物领域的分子模拟方法,能够通过模拟对生物大分子进行研究。这些方法的主要特点是对实验条件和设备要求低,大量的计算由计算机完成,并能为实验设计提供明确指导,为实验现象提供合理的解释,越来越被研究人员所重视。本研究的主要目的是建立一种通过计算机建模,分子对接等分析方法,预测抗体表位,指导实验进行验证,快速简便的对抗体表位进行鉴定的方法。这种方法的特点是仅需要抗体序列与抗原的晶体结构,就能对抗体表位进行预测,不仅需要的时间短,而且也没有任何实验要求,门槛较低。当然,对于预测的结果,需要通过实验进行验证才能确定。但是,计算机的预测结果给了我们一个目标去设计实验,有了这个目标,通过简单的突变实验就能验证,大大降低了表位鉴定的难度。在本文中,我们设计的表位鉴定方法具体步骤为:(1)通过Discovery Studio软件,使用抗体的一级氨基酸序列建立模型,获得抗体的分子结构;(2)抗原的晶体结构大多已经通过X射线晶体衍射方法获得,因而在PDB数据库中下载相应的抗原晶体结构;(3)使用Discovery Studio软件的ZDock,对抗体分子结构和抗原分子结构进行分子对接;(4)使用Residues Contact Frequency(RCF)算法对ZDock分子对接结果进行分析,预测抗原抗体相互作用的关键氨基酸;(5)设计实验验证预测结果。我们在Docking Benchmark 5数据库中选取了22对抗原抗体作为测试集,对以上预测方法的有效性进行验证。首先,我们验证了Discovery Studio软件对抗体结构的预测。对这22个抗体进行分子建模,将模型与真实的分子结构进行对比。我们发现抗体建模的准确程度很高。然后,我们验证RCF算法对ZDock结果的预测分析。RCF是一种对ZDock结果进行统计分析,预测蛋白-蛋白相互作用关键氨基酸位点的方法。我们使用Perl语言在DS软件中的Workscript窗口中实现了RCF算法。我们根据抗体结构的特殊性,对RCF算法设计了三种优化:1.仅考虑抗体CDR区原子进行RCF分析;2.使用抗原抗体分子的夹角对ZDock预测的pose(复合物构象)进行筛选;3.依据抗原抗体分子的夹角为每个pose添加权重函数-cos。我们分析了RCF算法及三种优化在测试集中的22对抗原抗体表位配位预测中的表现,结果显示RCF算法及三种优化均能够一定程度上对相互作用的关键氨基酸位点进行预测,并且三种优化的RCF算法预测结果均好于未优化的RCF算法,但是三种优化之间区分并不明显。于是我们选择第一种优化的RCF算法进行后续预测。在验证了抗体分子建模和RCF优化算法预测分析的有效性后,我们将这一方法应用在具体的抗体表位分析上。F2H5是一株鼠疫杆菌F1蛋白的抗体,是前期实验室通过杂交瘤技术获得的一株具有完全保护效果的鼠源抗体。实验室前期完成了F2H5抗体的人源化。我们首先通过实验确认人源化的F2H5抗体与F1蛋白在Western Blot和ELISA中均能相互结合。我们使用了主流的鉴定表位的实验方法——合成F1蛋白重叠肽库的方法,进行F2H5抗体的表位鉴定。但是出乎意料的是,所有的多肽都不与F2H5抗体结合。所以,我们采取了模拟计算的方法对抗体的表位进行预测,并进行实验验证。首先,我们使用DS软件对F2H5抗体进行建模,从PDB数据库中下载了F1蛋白的晶体结构,选取了其中分辨率较高的五个结构,分别与F2H5的结构进行分子对接,获得了五个ZDockResults.dsv的结果文件。RCF优化算法对这五个结果进行分析。预测结果显示F1蛋白上F96和E105这两个氨基酸位点可能是F2H5的关键氨基酸。基于这一结果,我们对F1蛋白进行丙氨酸扫描,设计了F1蛋白95-111位氨基酸突变为丙氨酸的单点突变体。实验结果显示,F1-G104A、F1-E105A、F1-N106A三个F1蛋白突变体在ELISA与Western Blot中均不能与F2H5结合,同时F1-K101A、F1-N103A两个突变体与F2H5抗体结合的能力明显降低。我们通过RCF优化算法成功鉴定了F2H5抗体的表位。完成了对F2H5抗体表位的鉴定后,我们使用G104E105N106这个表位筛选ZDock对接后产生的pose,选定了筛选后的pose中ZRank打分最高者作为F2H5-F1的相互作用的分子模型。对这个选定的pose,我们在DS软件中使用分子动力学进行了进一步的优化。根据优化后的pose,我们计算了基于这个构象,F1上95-111位点丙氨酸突变后对复合物稳定性的影响,其结果与丙氨酸扫描的结果吻合。我们进一步分析了抗体上的重要氨基酸位点。在RCF优化算法中,预测出Y170和Y214非常关键。与此同时,我们又使用了一个分析抗体上氨基酸适合程度的算法——Amino Acid Interface Fitness(AIF),对F2H5抗体与F1的相互作用进行了分析,也发现了这两个位点是最为关键的位点,并且酪氨酸是这两个位点上最适合的氨基酸。基于之前确定的复合物结构,我们计算了抗体重链上CDR2和CDR3上氨基酸饱和突变后的突变能,选取了能量变化明显的20株突变体进行实验验证。通过实验发现,20株突变体中,预测亲和力减弱的11株突变抗体均不结合F1或亲和力下降明显,预测准确率为100%;预测亲和力增加的9株突变抗体中,5株能够与F1结合。在这5株能够与F1结合的突变抗体中,2株F2H5上CDRH3区218位单点突变抗体,F2H5-D218R和F2H5-D218Y的EC50较F2H5表现出了显著下降,说明获得了亲和力增强的突变体。我们建立了一个基于ZDock分子对接,RCF优化算法的计算机预测抗体表位的方法。通过这个方法,我们对一株鼠疫F1蛋白的抗体F2H5的表位进行了预测,并成功通过实验验证了该预测表位。又通过计算机分析预测了抗原抗体复合物结构以及一些能够引起抗体亲和力显著变化的抗体突变株,也通过实验获得了验证。我们的结果说明,将计算机辅助的方法应用于抗体表位研究中,是有效并且很有意义的。
[Abstract]:In the interaction of antigen antibody and immune system, transformed by B lymphocytes plasma cells which can occur specific immunoglobulin binding with corresponding antigen. When combined with these antibodies and antigen, antibody and antigen on the CDR area of a region, namely the antigenic determinant (antigen determinant). Here is the antigenic determinant antibody binding epitopes. Combined with amino acid residues of CDR antibody epitope on the composition of antibody coordination. The epitope is divided into linear epitopes and conformational epitopes. The linear epitope consists of a continuous amino acid; and conformational epitopes usually consists of some in a sequence of discrete antigen, but close to each other in the spatial structure of the site together. So, change the molecular structure of the antigen may affect the antibody binding epitope, but for linear epitopes did not affect the epitope is. One of the most important properties of antibodies, antibody information but also of the most want to get. The antibody epitope mechanism can obtain antibody may play a protective role in the pathogenic mechanism of pathogens, and to epitope based reverse stimulation on protective antibody vaccine. The current mainstream method of antibody table who is identified by experiments. The experimental methods of some experimental conditions and equipment requirements, some methods of workload, some methods are low success rate. With the method of computer simulation and enhances the performance of the continuously mature, the simulation method of molecular applications in the field of biology, can through the simulation study biological macromolecules. The main features of these methods is the experimental conditions and equipment requirement is low, a large number of calculation by computer, and can provide clear guidance for experimental design, for real Experimental results provide a reasonable explanation, becomes more and more important. The main purpose of this study is to establish a through computer modeling, molecular docking analysis, prediction of epitopes, direct experimental verification, simple and rapid method for identification of antibody epitopes. The characteristic of this method is the only crystal structure need antibody and antigen sequence, will be able to predict epitopes, requires not only a short time, and there is no experimental requirements, low threshold. Of course, for the forecast results, through the experiment can be determined. However, the computer prediction results gave us a goal to design a experiment. This goal, through the mutation experiment can verify the simple, greatly reduces the difficulty of epitope identification. In this paper, we design the epitope identification method comprises the following steps: (1) by Discovery Stu Dio software to establish the model of primary amino acid sequence using antibody, molecular structure of the antibody; (2) the crystal structure of antigen have mostly through X ray diffraction method, the crystal structure of the corresponding antigen download in the PDB database; (3) using the Discovery Studio software ZDock, molecular docking of antibody molecules the molecular structure and antigenic structure; (4) using Residues Contact Frequency (RCF) algorithm for ZDock molecular docking results, prediction of key amino acid antigen antibody interaction; (5) the design of experiments. The prediction results we selected 22 antigen antibody in Docking Benchmark 5 database as test set to verify the effectiveness of above forecasting methods. First, we verified the predicted Discovery Studio software on the antibody structure. Molecular modeling of these 22 antibodies, the model with the real molecules The structure were compared. We found that the degree of accuracy is very high. Then the antibody modeling, we verify the prediction and analysis of.RCF RCF algorithm on ZDock result is a statistical analysis of the results of ZDock, prediction of protein protein interactions of key amino acids. We use the Perl language in DS software in the Workscript window to realize RCF algorithm. We according to the special antibody structure, three kinds of optimization algorithms of RCF design: 1. only consider the atomic antibody CDR region RCF analysis; 2. using antigen antibody molecules of pose angle ZDock prediction (complex conformation) screening; antigen antibody molecules according to an angle of 3. for each pose adding weight function -cos. we analyzed the RCF algorithm and three kinds of optimization in the test set 22 antigen antibody epitope ligand in the prediction of performance, results show that the RCF algorithm and the three optimization can to a certain extent The key amino acid sites of interaction prediction, and RCF algorithm to predict the three optimization results are better than the RCF algorithm is not optimized, but between the three kinds of optimization distinction is not obvious. So we choose the first kind of optimized RCF algorithm. In the subsequent prediction of antibody was verified Zi Jianmo analysis and prediction of RCF optimization algorithm then, we apply this method in the specific epitope analysis on.F2H5 antibody of Yersinia pestis F1 protein was obtained by the laboratory, is a strain of hybridoma technology has completely protective effect of murine anti F2H5 antibody. The previous complete human source. We first confirmed through experiments the humanized F2H5 antibody and F1 protein were combined in Western Blot and ELISA. We use the experimental synthesis of F1 protein overlapping peptide library identification method of the mainstream of the square table Method, epitopes were identified with the F2H5 antibody. But surprisingly, all of the peptides were not with F2H5 antibody binding. Therefore, we take the calculation of antibody epitopes were predicted and verified by the experiment. First, we use DS software to model the F2H5 antibody, the crystal structure of F1 protein the download from the PDB database, selects five high resolution structures, respectively. Molecular docking and F2H5 structure, obtained five results of ZDockResults.dsv file.RCF optimization algorithm to analyze these five results. The prediction results showed that F1 protein on F96 and E105 of the two amino acid sites may be the key amino acid F2H5. Based on this result, we performed an alanine scan of F1 protein, F1 protein design 95-111 amino acid mutations for single point mutants of alanine. Experimental results show that F1-G104A, F1-E105A, F1-N1 06A three F1 protein mutants were not with F2H5 in ELISA and Western Blot combination, and F1-K101A, F1-N103A two and F2H5 mutant antibody binding was significantly reduced by the RCF algorithm. We successfully identified F2H5 antibody epitope. The completion of the F2H5 antibody epitope identification, we use the pose G104E105N106 this epitope screening ZDock docking generated after the molecular model was selected after screening by pose ZRank in the highest scoring as F2H5-F1 interaction. The selected pose, we use molecular dynamics in DS software is optimized further. Based on the optimized pose, we calculated based on the conformation F1, on the site of 95-111 alanine mutation on stability of the complex, the results with alanine scanning results. We analyzed the important amino acid sites on the antibody in RCF optimization. In the algorithm, Y170 and Y214 prediction is very important. At the same time, we also use the Amino Acid Interface Fitness algorithm is an analysis of the antibody level for amino acid (AIF), the interaction of F2H5 antibody and F1 were analyzed, but also found that the two sites are the most important sites, and tyrosine this is the most suitable amino acids on the two loci. The composite structure based on the determined before, we calculated the antibody heavy chain CDR2 and CDR3 amino acid mutations after saturation mutagenesis, selects the energy change of 20 mutants significantly in experiments. Experimental results show that the 20 mutants, weakened affinity prediction the 11 mutant antibodies were not combined with F1 or affinity decreased significantly, the prediction accuracy is 100%; the 9 predicted increased affinity mutant antibody, 5 strains can bind with F1. In these 5 strains mutation could be combined with F1 Antibody, 2 strains of F2H5 CDRH3 218 single point mutations F2H5-D218R and F2H5-D218Y antibody, EC50 F2H5 showed significantly decreased, indicating the enhanced affinity of the mutant. We set up a ZDock based molecular docking prediction method of antibody epitope RCF optimization algorithm of computer. By this method, we the plague F1 antibody strain F2H5 protein epitopes were predicted, and through experimental verification of the predicted epitopes. And through the computer to analyze and predict the structure of antigen antibody complexes and some can cause significant changes in antibody antibody affinity of the mutant strain, was also verified by experiments. Our results show that the computer aided method applied to the antibody epitope study, is effective and meaningful.

【学位授予单位】：中国人民解放军军事医学科学院
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R392

【相似文献】