【作者】 胡荣静；

【导师】 张瑞生； Michel Delamar；

【作者基本信息】 兰州大学 ， 分析化学， 2009， 博士

【摘要】 获得性免疫缺陷综合征(艾滋病)是由人类免疫缺陷病毒(human immunodeficiencyvirus，HIV)通过破坏人体的免疫体统而导致的一系列症状。从HIV病毒开始在人类中传播，世界上大约有25 000 000人死于艾滋病。当前，40 300 000左右的人是艾滋病毒的携带者。自然界中，艾滋病毒有两种存在类型：艾滋病毒1型(HIV-1)和艾滋病毒2型(HIV-2)。其中，以HIV -1的影响最为广泛。艾滋病的流行严重危害人类的安全，艾滋病药物的研究是世界的热点问题。艾滋病毒逆转录酶(reverse transcriptase，RT)在HIV的生命周期中起到至关重要的作用，它利用病毒RNA基因组作为模板合成DNA，通过人体免疫系统中扮演重要角色的T细胞来实现自身的复制，从而产生新的病毒。所以，逆转录酶已经作为一个重要的靶点应用于抗艾滋病毒药物的研发。逆转录酶抑制剂(reverse transcriptase inhibitors，RTIs )能够抑制逆转录酶的功能，阻止病毒双链DNA的合成，从而阻断HIV病毒的增殖。根据其抑制剂机理，逆转录酶抑制剂分为两种：核苷类／核苷酸类逆转录酶抑制剂(nucleoside/nucleotide reverse transcriptase inhibitors，NRTIs/ NtRTIs)和非核苷类逆转录酶抑制剂(non-nucleoside reverse transcriptase inhibitors，NNRTIs)。近年来，计算机辅助药物的发现和设计已经被成功的应用于很多研究项目。在抗HIV病毒药物的研究中，二维定量构效关系(2dimension-quantitative structure-activityrelationship，2D-QSAR)、三维定量构效关系(3D-QSAR)和分子对接(Molecular Docking)已经被广泛应用。分子动力学(Molecular Dynamics)模拟在HIV领域的应用主要体现在研究HIV -1蛋白质酶、逆转录酶、整合酶等酶与其抑制剂之间的关系。同时，量子化学方法也被引入这个领域。本论文采用定量构效关系、分子对接和分子动力学多种技术和手段，研究逆转录酶抑制剂的分子结构与其生物活性之间的关系，以及抑制剂与逆转录酶的作用机理。目的在于比较不同抑制剂与逆转录酶的不同作用模式，获得有关分子的何种结构特征能够有效地提高其抗HIV病毒的生物活性的信息，用于辅助设计和合成更高效的抗HIV病毒抑制剂。论文第一章，我们对艾滋病、HIV-1逆转录酶及其抑制剂做了些介绍，同时，描述了计算机辅助药物设计方法及其应用。第二章对本论文中使用的各种方法做了详细描述。论文第三章研究了一系列嘧啶核苷类逆转录酶抑制剂，采用多元线性回归(multiplelinear regression，MLR)、支持向量机(support vector machine，SVM)和投影寻踪回归(projection pursuit regression，PPR)方法，以几何、静电和量子化学参数建立了定量结构-生物活性关系模型。MLR产生的线性模型，其训练集的相关系数(R~2)和均方误差(mean square error，MSE)分别为0．729和0．36；其测试集的R~2和MSE分别为0．662和0．42。SVM和PPR方法用于建立非线性模型，同样的训练集，SVM和PPR得到相关系数R~2分别为0．850和0．841，MSE分别为0．22和0．21；SVM和PPR测试集的R~2分别是0．830和0．840，MSE分别为0．27和0．30。预测结果表明，SVM和PPR方法优于MLR。建立的模型也许对新的嘧啶核苷类逆转录酶抑制剂设计起到一定帮助。第四章中，我们采用相似的方法对一系列2-氨基-6-苯磺酰基苄腈及其类似物非核苷逆转录酶抑制剂进行研究。用拓扑描述符、几何描述符和量子化学描述符描述分子结构特征。通过比较多元线性回归(MLR)、多元自适应样条回归(multivariate adaptiveregression splines，MARS)、径向基函数神经网络(radial basis function neural networks，RBFNN)、广义回归神经网络(general regression neural networks，GRNN)、投影寻踪回归(PPR)和支持向量机方法(SVM)六种不同方法，分别对抗HIV-1活性数据集和HIV-1 RT抑制活性数据集建立不同的QSAR模型。结果表明PPR和SVM模型具有最好的预测能力。为了深入探讨药物与蛋白质的相互作用关系，继二维定量关系以后，我们又对这些抑制剂进行三维定量构效关系、分子对接和分子动力学的研究。首先，通过分子对接方法将抑制剂与RT的活性位点相结合，找出两者的作用模式和抑制剂最合理的构象。然后，采用比较分子力场分析(comparative molecular field analysis，CoMFA)和比较分子相似性分析(comparative molecular similarity indices analysis，CoMSIA)方法建立基于配体和受体的预测模型。CoMFA和CoMSIA模型的交互检验系数q~2分别为0．723和0．760。CoMFA和CoMSIA等势图和受体的三维结构重叠图可以帮助我们更清楚的了解RT蛋白与抑制剂的相互作用，以及抑制剂的结构特征对活性的影响。比如，2-氨基-6-苯磺酰基苄腈及其类似物的B环C-3位和C-5位如果引入较大的或者疏水性较强的基团有利于提高分子的生物活性；在化合物A环C-2位置引入氢键给体基团，且能够与RT蛋白残基Lys 101形成氢键，则有利于提高分子的生物活性。然后，我们对数据集中三个代表不同类结构化合物的分子进行水溶剂体系的分子动力学模拟研究。采用分子力学／泊松-波尔兹曼表面积方法(Molecular MechanicsPoisson-Boltzmann Surface Area，MM-PBSA)和分子力学／广义博恩表面积方法(MolecularMechanics Generalized Born Surface Area，MM-GBSA)，用MD模拟的轨道计算体系的结合能。并且，分析了三个不同代表性分子与RT蛋白的作用模式、氢键作用，三个体系的结合自由能，以及重要氨基酸残基对结合自由能的贡献。结果表明由于抑制剂含有不同的含硫官能团，导致抑制剂与残基的相互作用不同，因为不同类抑制剂与与RT的结合模式也不同。第五章中，我们对一系列硫代氨基甲酸酯类非核苷类逆转录酶抑制剂进行2D-QSAR和3D-QSAR研究。在2D-QSAR研究中，我们采用MLR、RBFNN、PPR、SVM和最小二乘支持向量机(least squares support vector machines，LS-SVM)方法建立QSAR模型。启发式方法用于选择描述符。70个化合物选择56个作为训练集建模。PPR方法建立的模型具有最好结果：训练集和测试集的R~2分别为0．873和0．755。基于同样的分子构型和子集划分，又对这一系列分子进行基于配体的3D-QSAR研究。CoMFA模型的交叉检验系数q~2为0．701(5个主成分)，最好的CoMSIA (SH)模型的q~2q为0．672。为了获取更多的抑制剂与RT受体的相互作用关系信息，我们进行了分子对接研究。对接结果显示绝大多数配体与蛋白质Lys 101形成氢键。同时，其他的作用力，如范德华力也对受体-配体的结合有重要影响。然后，对其中53个具有闭环邻苯二甲酰亚胺结构的化合物的对接构象进行基于受体的3D-QSAR研究。53个化合物分为49个化合物的训练集和10个化合物的测试集。通过训练集建立的CoMFA模型的q~2为0．488，CoMSIA模型(SHD)的q~2为0．642。经过分析，化合物的立体场、静电场、疏水场和氢键给体场的部分特征对生物活性有较大影响。更多还原

【Abstract】 Acquired immune deficiency syndrome (AIDS) is a set of symptoms resulting from thedamage to the human immune system caused by the human immunodeficiency virus (HIV).About 25 million people worldwide have died from this infection since the start of theepidemic, and 40.3 million people around the world are currently living with HIV/AIDS.There are two types of HIV: HIV-1 and HIV-2. The predominant virus is HIV-1.The prevalence of AIDS has been a big problem and endangering human life, therefore,the study of AIDS drugs becomes a current event in the world. In the infection process, thereverse transcriptase (RT) is very important. RT converts the single-stranded HIV RNA todouble-stranded HIV DNA which contains the instructions HIV needs to use a T-cell’s geneticmachinery to reproduce itself. Hence, RT is one of the major targets for the treatment ofAIDS.Reverse transcriptase inhibitors (RTIs) block reverse transcriptase’s enzymatic functionand prevent completion of synthesis of the double-stranded viral DNA, thus preventing HIVfrom multiplying. There are two forms of RT inhibitors according to their inhibitorymechanism: nucleoside/nucleotide (analog) reverse transcriptase inhibitors(NRTIs/ NtRTIs)and non-nucleoside reverse transcriptase inhibitors (NNRTIs).Computer-aided drug discovery and design have proven successful in many recentresearch programs. 2D-QSAR and 3D-QSAR have been used in the studies of anti-HIVinhibitors, as well as Molecular docking studies. The application of Moelcular Dyanmics inthe field of HIV-1 has involved studies of HIV-1 protease, reverse transcriptase, integrase andenzymes with their inhibitors. Some studies were performed with different QuantumChemistry methods.The thesis uses several techniques (QSAR, Molecular Docking and Molecular dyanmics)to correlate molecular structure features to their bioactivity, and to study the interaction modebetween reverse transcriptase and their inhibitors. We aim at comparing the different actionmode between RT and their different inhitor families, obtaining more information aboutwhich molecular features are favorable to activity, and aiding to design and synthesize highlyactive ant-HIV inhibitors. In Chapter 1,we present a general introduction of AIDS, HIV-1 reverse trascriptase, theirinhibitors and a brief description of the methods in Computer-Aided Drug Design (CADD). InChapter 2, we present the methods that are used in this work in a detailed way.In Chapter 3, we study a set of pyrimidine nucleosides RT inhibitors and establishquantitative structure-activity relationships (QSAR) using a comprehensive set of geometrical,electrostatic and quantum-chemical molecular descriptors, by multiple linear regression(MLR), support vector machine (SVM) and projection pursuit regression (PPR) methods.MLR yields a linear model withadetermination coefficient (R~2) and mean square error (MSE)of 0.729 and 0.36 for the training set and of 0.662 and 0.42 for the test set, respectively. SVMand PPR methods that we used to construct non-linear prediction models, lead to a better R~2of 0.850 (SVM) and 0.841 (PPR) and MSE of 0.22 (SVM) and 0.21 (PPR) for the sametraining set, together with R~2 of 0.830 (SVM) and 0.840 (PPR) and MSE of 0.27 (SVM) and0.30 (PPR) for the same test set, respectively. The prediction results of the SVM and PPRmodels are better than those of MLR. These models might help designing new pyridinenucleosides inhibitors with enhanced activity.In Chapter 4, we analyze in a similar way another series of HIV-1 reverse transcriptaseinhibitors: 2-amino-6-arylsulfonylbenzonitriles and their thio and sulfinyl congeners. We usetopological and geometrical, as well as quantum mechanical energy-related and chargedistribution-related descriptors to describe the structural features. We compare six techniques:multiple linear regression (MLR), multivariate adaptive regression splines (MARS), radialbasis function neural networks (RBFNN), general regression neural networks (GRNN),projection pursuit regression (PPR) and support vector machine (SVM) to establish QSARmodels for two data sets:anti-HIV-1 activity and HIV-1 reverse transcriptase binding affinity.Our results show that PPR and SVM models provide a powerful capacity of prediction.This 2D-QSAR analysis is completed with two more approaches: 3D-QSAR, relying onmolecular docking, and molecular dynamics in order to examine into more detail thedrug-protein interaction. Docking simulations are employed to position the inhibitors into theRT active site to determine the most probable binding mode and most reliable conformations.Then we develop comparative molecular field analysis (CoMFA) and comparative molecularsimilarity indices analysis (CoMSIA) approaches, using a complex receptor-based and ligand-based alignment procedure and different alignment modes to obtain highly reliable andpredictive CoMFA and CoMSIA models with cross-validated q~2 value of 0.723 and 0.760,respectively. The CoMFA and CoMSIA contour maps with the 3D structure of the target (thebinding site of RT) inlaid allow us to better understand the interaction between the RT proteinand the inhibitors and the structural requirements for inhibitory activity against HIV-1.Forinstance, we show that for 2-amino-6-arylsulfonylbenzonitriles inhibitors to have appreciableinhibitory activity, bulky and hydrophobic groups in 3-and 5-position of the B ring arerequired. Moreover, H-bond donor groups in 2-position of the A ring to build up H-bondingwith the Lys101 residue of the RT protein are also favorable to activity.We then perform dynamics (MD) simulations in water environment on the RTcomplexes with one representative of each of 3 series of inhibitors:2-amino-6-arylsulfonylbenzonitriles, and their thio and sulfinyl congeners. MolecularMechanics Poisson-Boltzmann Surface Area (MM-PBSA) and Molecular MechanicsGeneralized Born Surface Area (MM-GBSA) are applied to calculate the binding free energybased on the obtained MD trajectories. We carry out a comparison of interaction modes,binding free energy, contributions of the residues to the binding free energy and H-bonds withthe average structures. Our results show that there exist different interaction modes betweenRT and ligands due to the different sulfur functional groups in the inhibitors and to specificinteractions with some residues.In Chapter 5, we study a series of O-(2-phthalimidoethyl)-N-substituted thiocarbamatesand their ring-opened congeners as non-nucleoside HIV-1 reverse transcriptase inhibitors,using 2D-QSAR and 3D-QSAR methods. In 2D-QSAR studies, we used and comparedseveral methods: MLR, RBFNN, PPR, SVM and LS-SVM to build QSAR models. Thedescriptors were selected by heuristic method. Among the 70 compounds, we selected 56 asthe training set. The best results were generated by PPR with a square correlation coefficientR~2 of 0.873 for the training set and 0.755 for the test set. Based on the same conformations ofthe compounds, we then performed ligand-based 3D-QSAR studies, with a cross-validated q~2of 0.701 (with 5 components) in CoMFA and 0.672 (with 6 components) in CoMSIA (SH).In order to obtain more information about the RT receptor interaction with theseinhibitors we performed further studies based on molecular docking. Our results indicate that an H-bond between Lys101 of the protein and ligands exists in most cases. Moreover, someother interactions, such as Van der Waals force, exist contributing to the binding affinitybetween receptor and ligand. The docking conformations of 59O-(2-phthalimidoethyl)-N-substituted thiocarbamates were generated to carry outreceptor-based 3D-QSAR studies. 53 compounds were divided into training set (43) and testset(10). With the training set, a q~2 of 0.488 was obtained in CoMFA, while a higher value ofq~2: 0.642 was obtained in CoMSIA with three SHD descriptors. The steric, electrostatic,hydrophobic and H donor features of the compounds make much contribution to thebioactivity of the inhibitors.更多还原