視網膜蛋白質的結構預測和動力學模擬

Abstract

膜蛋白在細胞組成和生理等許多方面上扮演一個非常重要的角色，然而將其結晶進而利用X-ray、電子繞射等實驗技術來研究其摺疊動力學以及獲得高解析的結構之研究是非常困難，近代由於電腦運算技術的大躍進使得電腦模擬成為研究蛋白質結構、功能及摺疊力學的一個非常方便的工具。 視網膜蛋白質是一個由200多個胺基酸和7個α-helices二級結構所組成的蛋白質。因此，我們建構一個二階段的模型來預測膜蛋白的native structure，在第一階段，我們利用一個粗粒化的蒙地卡羅模型來模擬膜蛋白的α-螺旋之間的交互作用，進而得到一個膜蛋白native-like的近似結構，在此一階段我們可以得到在脂質膜中間平面(LMP)的均方根結構偏差(RMSD)，我們稱作RMSD-LMP，當每一個膜蛋白在第一階段所得到的最低能量的結構，其RMSD-LMP對bacteriorhodopsin (BR)、 halorhodopsin (HR)、 sensory rhodopsin II (SRII)依序為1.22 Å、1.64 Å、1.20 Å。 在第二階段，我們利用Amber套裝軟體建構一個all-atom的分子動力學模擬系統，更進一步的將我們在第一階段所得到的近似結構，在一個原子尺寸的模型中包含蛋白質所有原子的情況，使其結構更能接近native structure，而我們也得到蛋白質在骨幹上所有原子的RMSD，對BR而言，RMSD從第一階段的3.99 Å被進一步的改進到2.64 Å，對HR而言，RMSD從第一階段的2.59 Å被進一步的改進到1.89 Å，對SRII而言，RMSD從第一階段的3.12 Å被進一步的改進到1.92 Å。而我們也利用此一模擬模型成功的預測出SRI的結構。

Membrane proteins play a crucial role in many cellular and physiological processes, but the knowledge of their high resolution structures and folding mechanism is very limited due to the difficulties in determining their structures experimentally. Recently, remarkable advances in computer simulations offer a convenient tool to study these problems. The retinal proteins contain more than 200 amino acids and a retinal molecule and form a seven helix structure. Here we employed a two-step approach to predict the native structures and study the folding dynamics of membrane proteins by using off-lattice coarse-grained Monte-Carlo simulations and all-atom molecular dynamics simulations. In particular, we have applied this approach to predict the structure of retinal proteins found in Halobacterium salinarum membranes. At the first stage, the lowest energy structure with a small root mean square deviation (RMSD) at the lipid midpoint plane (RMSD-LMP) can be obtained. The RMSD-LMP is 1.22 Å for bacteriorhodopsin (BR), 1.64 Å for halorhodopsin (HR), and 1.20 Å for sensory rhodopsin II (SRII). At the second step, the predicted structures are further refined in an all-atom model using Amber force field. The overall RMSD of backbone atoms from the X-ray structures can be reduced from 3.99 Å to 2.64 Å for BR, from 3.12 Å to 1.92 Å for SRII, and from 2.59 Å to 1.89 Å for HR. After successfully predicting the native structures of BR, HR, and SRII by combining Monte-Carlo simulations and molecular dynamics simulations, we further predict the native structure of sensory rhodopsin I.