Molecular Modeling of the Human P-glycoprotein Transporter Protein

像许多跨膜蛋白一样，尽管付出了很多努力，但通过X射线结晶学确定P-gp的结构已被证明是难以捉摸的。这源于难以形成足够高质量的晶体，以维持蛋白质不同部分的天然物理化学环境。因此，代替直接的实验确定，我们正在努力将所有可用的(通常是间接的)实验数据与基于物理化学的数学方法相结合，以产生一个或多个物理上真实的结构模型。幸运的是，三十多年的研究已经提供了关于P-gp的丰富信息，从中我们可以闪耀出结构信息。在广泛的范围内，我们知道该蛋白由两个同源结构域组成，每个结构域都有一个六段跨膜成分和一个核苷酸结合成分。到目前为止，关于P-gp结构信息最好的两个来源是同源细菌蛋白Sav1866的X射线晶体结构和人P-gp的低分辨率冷冻电子显微镜。此外，细菌脂翻转酶MSBA的修正X射线结构有望很快公布，它与P-gp的关系甚至比Sav1866更密切。因此，我们工作的一个主要重点是利用细菌蛋白的晶体结构作为模板来建立人P-gp的同源模型。虽然另一个小组最近发表了这样的模型，但它只基于P-gp和Sav1866序列的简单比对，不幸的是，这对于跨膜片段没有很好的定义。相反，我们的努力更深入地研究了密切相关的MDR蛋白家族和ABC转运蛋白超家族中的残基保守模式。这些信息有助于预测哪些残基暴露在膜的核心层和头基层，哪些残基排列在孔内，哪些位于两个跨膜域的界面上。我们目前正在开发同源家族和超家族的宏大序列比对。这一结果还将使相关突变模式的确定成为可能，这有助于识别蛋白质三维结构中邻近的残基群。最后，我们将检查所得到的模型与所有实验数据的一致性，例如定点突变、自然发生的多态和交叉连接数据的影响。如果基于Sav1866模板的模型无法解释实验结果，我们将搜索替代构象以使其符合要求。另一个主要的焦点是从电子显微镜获得的密度图开发模型。这样做的好处是，结构数据直接来自目标蛋白人类P-gp，而不是来自可能在结构上不同的细菌同系物。这包括这样一个事实，即人P-gp的两个跨膜结构域在序列上不同，因此围绕大约两个孔的轴不对称(如显微照片中所观察到的)，而细菌同系物只包含一个结构域，从而在膜上形成完全对称的同源二聚体。在获得了与跨膜电子密度相适应的通用螺旋坐标后，我们几乎完成了确定氨基酸序列与蛋白质骨架结构的比对的计算机软件。这相当于一个庞大计算规模上的典型蛋白质穿线过程。该程序建立了氨基酸片段到跨膜螺旋的每一种可能的组合，并根据一系列物理化学和实验确定的标准对其进行评分。这些包括每种氨基酸残基的适当环境，保守残基的聚集，以及实验中的交联距离限制，一个必须克服的困难是必须扫描的相对大量的排列。这是通过开发聪明的算法来克服的，该算法将所需的计算量减少到当今技术能力的实际范围内。随着模型的发展，我们将与细胞生物学实验室内的其他小组密切合作，特别是Di Xia博士、Suresh Ambukar博士和Michael Gottesman博士的部门，进行实验测试、X射线结构确定和抑制剂开发。

Like many transmembrane proteins, determination of the structure of P-gp by X-ray crystallography has proven elusive despite much effort. This stems from the difficulty of forming sufficient-quality crystals that maintain the native physiochemical environments for the different parts of the protein. Thus, in lieu of direct experimental determination, we are striving to integrate all available (often indirect) experimental data with physiochemically-based mathematical methods to produce one or more physically realistic models of the structure. Fortunately, over three decades of study has provided a wealth of information about P-gp from which we can gleam structural information. On a broad scale, it is known that the protein is composed of two homologous domains, each with a six-segment transmembrane component and a nucleotide-binding component. To date, the best two sources of structural information about P-gp are an X-ray crystal structure of the homologous bacterial protein Sav1866, and low-resolution cryo-electron micrographs of human P-gp. In addition, the corrected X-ray structure of the bacterial lipid flippase MsbA is expected soon, which is even more closely related to P-gp than is Sav1866. Thus, one major focus of our work is to develop a homology model of human P-gp using the crystal structures of the bacterial proteins as templates. While another group has recently published such a model, it was only based on a simple alignment of the P-gp and Sav1866 sequences, and unfortunately, this is not well defined for the transmembrane segments. Rather, our efforts go deeper into examining the patterns of residue conservation within the family of closely related MDR proteins and the superfamily of ABC transporters. This information helps predict which residues are exposed to the core and headgroup layers of the membrane, which residues line the pore, and which are at the interfaces of the two transmembrane domains. We are currently in the process of developing a grand sequence alignment of homologous families and the superfamily. The results of this will also enable the determination of patterns of correlated mutations, which help identify groups of residues that are proximal in the 3-dimensional structure of the protein. Finally, we will examine the resultant models for consistency with all the experimental data, such as the effects of site-directed mutagenesis, naturally occurring polymorphisms, and cross-linking data. Where the model based on the Sav1866 template fails to explain the experimental results, we will search for alternate conformations that bring it into compliance. The other major focus is to develop models from the density maps obtained from electron microscopy. This has the advantage that the structural data is directly from human P-gp, the target protein, and not from a bacterial homolog, which likely differs in structure. This includes the fact that the two transmembrane domains of human P-gp are different in sequence, and thus are asymmetrical around the approximate two-fold axis of the pore (as observed in the micrographs), while the bacterial homologs only contain one domain, and thus form perfectly symmetrical homodimers in the membrane. Having obtained coordinates of generic helices fitted to the transmembrane electron density, we have nearly completed computer software to determine the alignment of the amino acids sequences onto the backbone structure of the protein. This amounts to a typical Protein Threading procedure on a grand computational scale. The program builds every possible combination of amino acid segment to transmembrane helix and scores it according to a series of physiochemical and experimentally-determined criteria. These include the proper environment for each type of amino acid residue, the clustering of conserved residues, and experimental cross-linking distance constraints, One difficulty that had to be surmounted was the relative enormous number of permutations that had to be scanned. This was overcome by developing clever algorithms to reduce the required amount of computations to within the practical scope of present-day technological capabilities. As the models develop, we will be working closely with other groups within the Lab of Cell Biology, especially Dr.s Di Xia's, Suresh Ambudkar's, and Michael Gottesman's sections, to undertake experimental tests, X-ray structure determination, and inhibitor development.