Developing Improved Methods for Modeling and Simulating Protein Structures

我们的第一个计算方法子项目涉及在分子建模和模拟中使用对称性约束。对于蛋白质或蛋白质组件由相同的亚基组成的情况，亚基通常具有相同的构象和与相邻亚基的相同的相互作用。二十多年来，我们已经将这一限制纳入我们的离子通道模型，并取得了相当大的成功。例如，我们的钾通道离子选择性区域的四重对称模型与随后确定的钾通道晶体结构的模型基本相同。然而，直到最近，我们还没有将对称约束纳入到自动算法中，也没有证明这样做实际上提高了结构模型的质量。我们的一位合作者Andriy Anishkin开发了一个程序，在分子动力学模拟过程中使用谐波约束来恢复对称性。通过分析KCSA、KirBac、NaK三个远缘离子通道的晶体结构，验证了该程序对离子通道同源建模的有效性。该项目的第一步是使用另外两个通道结构作为模板来开发每个通道的两个同源模型。接下来，对这六个模型以及三个晶体结构进行了分子动力学模拟，将蛋白质嵌入到一个显式的脂双层中，两侧和通道的孔洞中都有水和离子。在8纳秒的无约束模拟后，施加了对称约束。对于每个模型和晶体结构，这个过程再重复两次。然后，将没有分子动力学模拟、具有无约束模拟和具有对称性约束模拟的模型与原始结构进行比较，以确定哪一个更好。主要的发现是，尽管不受限制的分子动力学模拟并没有改进同源模型，但施加对称约束确实导致了实质性的改进。最大的改善发生在孔隙衬里片段，其中相邻亚基之间的相互作用广泛。我们开发离子通道同源模型的主要目的是充分了解它们的结构和药理，以便在基于结构的药物设计中利用这些模型。因此，值得注意的是，对称性限制大大改进了形成主要药物结合部位的孔区的模型。我们的第二个计算方法子项目涉及简化蛋白质的分子表示，以便可以更长时间地模拟蛋白质和多肽的性质。传统分子动力学模拟的一个主要局限性是它们只能在很短的时间内进行，通常不到十分之一微秒。这比大多数构象变化所需的时间要短得多，比组装淀粉样β结构所需的时间短许多个数量级。为了克服这一限制，我们团队的Sijung Yun博士一直在使用离散分子动力学，它的效率比传统分子动力学高约七个数量级。在波士顿大学做博士期间，他帮助开发了这个项目。为了在保持精度的同时提高效率，在离散分子动力学中引入了三种简化。首先，相互作用势被简化为离散的步骤。这极大地简化了模拟过程中所需的数学运算和计算次数。第二，由水分子引起的相互作用被有效势取代。这大大减少了通过消除水分子来模拟的原子数量。第三，我们使用四珠表示蛋白质中的一个残基，而不是表示残基的所有原子。这也减少了模拟原子的数量。在这里，云博士一直致力于改进参数，并测试在进行这些模拟时，众所周知的蛋白质晶体结构的保持情况。他还一直在使用这种方法来测试我们团队开发的淀粉样β六角体模型。淀粉样β六角体与阿尔茨海默病的神经毒性有关。我们的第三个计算方法子项目涉及开发规则晶格上的蛋白质结构模型。我们使用的晶格具有在我们的淀粉样β蛋白组装模型中出现的对称性类型。晶格方法应该允许我们检查和评估构建具有这种特定类型对称性的多肽组件的所有可能的方法。即使用简化的离散分子动力学，这种详尽的分析在计算上也是不可行的。这个项目刚刚开始，我们仍在评估该方法的可行性。

Our first computational methods subproject involves using symmetry constraints in molecular modeling and simulations. For those cases in which proteins or protein assemblies are composed of identical subunits, the subunits usually have identical conformations and identical interactions with neighboring subunits. We have incorporated this constraint into our models of ion channels for over two decades, with considerable success. For example our four-fold symmetric model of the ion selective region of potassium channels was virtually identical to those of subsequently determined potassium channel crystal structures. However, until recently, we have not incorporated symmetry constraints into an automated algorithm, nor demonstrated that doing so actually improved the quality of structural models. One of our collaborators, Andriy Anishkin, has developed a program that uses harmonic restraints to restore symmetry during molecular dynamic simulations. We have tested the utility of this program for homology modeling of ion channels by analyzing crystal structures of three distantly related ion channels named KcsA, KirBac, NaK. The first step of this project was to develop two homology models of each channel using the other two channel structures as templates. Next, molecular dynamics simulations of these six models, along with the three crystal structures, were performed with the proteins embedded in an explicit lipid bilayer with water and ions on each side and within the pore of the channels. After eight nanoseconds of unrestrained simulations, symmetry restraints were imposed. This process was repeated two more times for each model and crystal structure. The models developed without molecular dynamic simulations, with unrestrained simulations, and with symmetry-restrained simulations were then compared to the original structures to determine which was better. The major finding was that although unrestrained molecular dynamic simulations did not improve the homology models, imposition of symmetry restraints did lead to substantial improvement. The greatest improvement occurred for the pore lining segments, where interaction among adjacent subunits is extensive. Our principal purpose for developing homology models of ion channels is to understand their structure and pharmacology well enough to utilize the models in structure-based drug design. Thus, it is noteworthy that the symmetry restraints substantially improve models of the pore region that forms the principal drug binding sites. Our second computational methods subproject involves simplifying molecular representations of proteins so that properties of proteins and peptides can be simulated for much longer times. A major limitation of conventional molecular dynamic simulations is that they can be performed for only short periods of time, typically less than a tenth of a microsecond. This is substantially shorter than the time required for most conformational changes, and many orders of magnitude shorter than the time required for assembling amyloid beta structures. To overcome this limit, Dr. Sijung Yun of our group has been using discrete molecular dynamics that is about seven orders of magnitude more efficient than conventional molecular dynamics. He helped develop this program during his doctoral work at the University of Boston. Three simplifications are introduced in discrete molecular dynamics in order to increase efficiency while still preserving accuracy. First, interaction potentials are simplified into discrete steps. This greatly simplifies the mathematics and number of calculations required during the simulations. Second, interactions due to water molecules are replaced by effective potentials. This greatly reduces the number of atoms simulated by eliminating water molecules. Third, we used four-bead representation of a residue in a protein instead of representing all the atoms of the residue. This also reduces the number of simulated atoms. While here, Dr. Yun has worked to improve the parameters and test how well known protein crystal structures are maintained when subjected to these simulations. He has also been using this approach to test models of amyloid-beta hexamers developed by our group. Amyloid-beta hexamers are involved in neurotoxicity of Alzheimers disease. Our third computational methods subproject involves developing models of protein structures on regular lattices. We are using lattices that have the types of symmetry that occur in our models of amyloid beta assemblies. The lattice approach should allow us to examine and evaluate all possible ways of constructing peptide assemblies that have this specific type symmetry. Such exhaustive analysis is not computationally feasible even with simplified discrete molecular dynamics. This project has just been started and we are still evaluating the feasibility of the approach.