蛋白质配体结合位点柔性的系统分析及分子柔性对接方法的发展和应用

【作者】 李伟；

【导师】 黄牛；

【作者基本信息】 北京协和医学院 ， 生物化学与分子生物学， 2012， 博士

【摘要】 在药物靶标确定和三维结构已知的情况下,基于受体结构的虚拟数据库筛选方法利用分子对接技术自动地匹配受体结合腔穴和化合物数据库中的小分子三维结构,然后利用基于分子力场的能量函数或者经验性函数对分子对接的模式进行打分,进而选择与受体相互作用最好的一组化合物进行生物活性测试,从而大大节省了寻找先导化合物的费用和难度。尽管分子对接在先导化合物的寻找方面有许多成功的应用,但仍然存在很大的问题,其中忽略蛋白柔性常常是导致失败的重要原因。由于蛋白质受体和配体小分子在分子识别过程中,配体结合位点会发生不同程度的构象变化,如何通过计算化学的方法去评估和预测配体结合位点残基的构象柔性,缩小构象搜索的空间,一直是这个研究领域中具有挑战性的一个重要问题。幸运的是,随着结构生物学的发展,大量晶体结构的解析,为我们更好的理解分子识别和改进分子对接方法,提供了重要的资源和基础。因此,我们建立了一个专门用于研究配体结合位点柔性的关系型数据库Flex-Site DB (http://flexsi.te.huanglab.org.cn/flexsite),包括5,525个配体小分子结合位点,每个结合位点都包含多套蛋白晶体结构数据描述不同的蛋白构象,总共30,313套晶体结构数据。Flex-Site DB提供的信息有助于研究蛋白质配体结合位点的构象变化和蛋白质与配体之间的相互作用,以及两者之间的关系。同时,Flex-Site DB提供了配体结合位点的多个不同三维构象,结合位点残基的物理和化学的性质,以及蛋白质和配体相互作用的描述符。通过蛋白质晶体结构构象变化的统计分析和多重时间尺度蒙特卡洛构象搜索的方法,我们系统研究了配体结合位点残基的柔性状况。大部分的配体结合位点在结合了相似结构的化合物时,构象变化较小；在结合了结构差异较大的化合物时,有显著的构象变化,但主要是在残基的侧链水平。对于配体结合位点同一类型的极性残基,能够通过侧链与配体小分子形成氢键的残基,同侧链不与配体形成氢键的残基相比,更加刚性,突变率也更低。能与配体形成氢键相互作用的蛋白质残基侧链刚性较强的原因可能是,在与配体结合过程中,这类残基不发生较大构象变化,从而减少熵的损失。我们的结果显示,在先导化合物的发现和优化过程中,在大多数情况下只需要对少数几个柔性较强的残基进行侧链的构象搜索。并且,我们通过多重时间尺度蒙特卡洛构象搜索的方法预测配体结合位点的柔性残基的策略以及通过Flex-Site DB中鉴定到的柔性残基,能直接促进我们对于柔性对接方法的发展。除了利用Flex-Site DB中残基侧链柔性的信息,我们使用多重时间尺度的蒙特卡洛算法在没有配体结合的情况下,对结合位点的残基进行构象搜索,然后将搜索到的多个残基构象整合到D0CK3.5.54的柔性对接算法中来模拟“构象选择”的过程。此方法在10个不同的蛋白上进行了测试,总共做了20个对接。其中11个对接中都获得较好的结果,而刚性对接只有6个得到较好的结果。我们挑选了PIM1(proviral integration site in Moloney)激酶作为我们应用柔性对接方法来虚拟筛选的实例。PIM1激酶是与白血病和前列腺癌紧密相关的重要的药物靶点蛋白。柔性对接不仅较好地重现PIM1激酶配体小分子的结合模式,而且还能提高阳性化合物的富集率。柔性对接在PIM1激酶上虚拟筛选了小规模的LOPAC小分子化合物库,发现一个与PIM1形成很好相互作用的小分子硫利达嗪,是多巴胺受体D1/D2的选择性拮抗剂,并通过测定半数抑制率确认了它是PIM1和PIM3微摩尔级别的结构新颖的抑制剂。另外,基于“诱导契合”学说,我们改进了柔性对接策略,在柔性对接生成的蛋白质配体复合物中,对柔性残基侧链进行进一步的构象搜索和能量最小化。结果显示,残基侧链构象搜索和能量最小化,能够更好地重现复合物晶体结构中配体的结合模式,并且阳性化合物富集能力也有所提升。因此,基于我们对LOPAC化合物库筛选的经验以及对柔性对接方法的进一步提高,我们在PIM1激酶上筛选了更大规模的化合物库。最终我们挑选了70个具有较好结合模式的化合物,并在接下来的工作中会测试它们的生物学活性。更多还原

【Abstract】 Receptor-based virtual screening utilizes molecular docking method to place small molecular compounds into the receptor binding site, and rank the compounds according to their scores computed in force field-based or empirical-based scoring functions. Despite the successful applications of molecular docking, there still exists big challenges. Neglecting protein flexibility is known to cause errors in structure-based drug design. Treating binding-site flexibility in molecular docking is critical in certain cases, but not yet realized in rigorous fashion. Since protein-ligand binding will cause multiple levels of conformational change, computationally assessing ligand binding-site flexibility and reducing the sampling space are critical issues in structure-based drug design. Fortunately, the advance of the structural biology provides an opportunity for better understanding molecular recognition and improving the performance of docking programs. Here, we assembled a comprehensive, yet specialized relationship database, Flex-Site DB (free available at http://flexsite. huanglab. org. cn/flexsite), containing5,525unique ligand binding-site conformational ensembles derived from a total of30,313protein crystal structures. Flex-Site DB provides the community a rich resource for systematically investigating protein binding-site flexibility and protein-ligand interaction, including but not limited to binding-site conformational ensembles, the structural properties of binding-site residues and variety of descriptors of protein-ligand interaction. We mainly investigated the magnitude of protein conformational change at residue level via structure statistical analysis and Monte Carlo sampling method. Interestingly, many ligand binding pockets do not embody significant conformational rearrangement, especially among holo structures bound with structurally similar ligands. Remarkably, for the same residue type comparison, the binding-site residues forming hydrogen bond interaction with ligand can confer structural rigidity and low mutation propensity. It is likely that the specific protein-ligand interaction intends to be pre-organized to minimize the entropic loss upon ligand binding. Our analysis suggests that sampling a few predetermined binding-site residues may be adequate in majority of lead discovery and lead optimization applications. Nevertheless, the generated protein binding-site conformational ensembles and identified flexible residues in present study can directly facilitate our development of flexible-receptor docking algorithm.Besides exploring the residue flexibility information in Flex-Site DB to identify flexible residue in binding-site, we used the multiple "time step" Monte Carlo algorithm to sampling residue side chain conformational change without ligand and then used modified version of DOCK3.5.54for flexible molecular docking, based on "conformational selection" model. We benchmarked this method against10different proteins,20docking test cases.11out of20perform well in flexible docking while only6out of20perform well in rigid docking. We applied this method to screen the LOPAC database against PIM1(proviral integration site in Moloney) kinase, a drug target related with leukemia and prostate cancer. Interestingly, among the top ranked docking poses, a dopamine D1/D2receptor antagonist, Thioridazine, was identified to form favorable interactions with binding site residues. Subsequently, we experimentally validated Thioridazine as low micromolar inhibitor of PIM1kinase. One drug can simultaneously interact with G-protein coupled receptor and protein kinase, expanding our knowledge on drug off-target effect. In addition, we further improved our flexible docking protocol by sampling and minimizating the side chains along with the docked ligand based on "induced fit" theory. Encouragingly, after further side chain sampling and minimization, both docking poses and docking enrichment were improved. Therefore, we screened our in-house database containing200,000compounds against PIM1kinase using our improved flexible docking strategy. Finally, we selected70diverse compounds with good docking poses and will test their bioactivity.更多还原