Dong Sijia S, Goddard William A, Abrol Ravinder
Materials and Process Simulation Center, California Institute of Technology, Pasadena, CA, United States.
Materials and Process Simulation Center, California Institute of Technology, Pasadena, CA, United States.
Methods Cell Biol. 2017;142:173-186. doi: 10.1016/bs.mcb.2017.07.009. Epub 2017 Sep 11.
G protein-coupled receptors (GPCRs) are membrane proteins critical in cellular signaling, making them important targets for therapeutics. The activation of GPCRs is central to their function, requiring multiple conformations of the GPCRs in their activation landscape. To enable rational design of GPCR-targeting drugs, it is essential to obtain the ensemble of atomistic structures of GPCRs along their activation pathways. This is most challenging for structure determination experiments, making it valuable to develop reliable computational structure prediction methods. In particular, since the active-state conformations are higher in energy (less stable) than inactive-state conformations, they are difficult to stabilize. In addition, the computational methods are generally biased toward lowest energy structures by design and miss these high energy but functionally important conformations. To address this problem, we have developed a computationally efficient ActiveGEnSeMBLE method that systematically predicts multiple conformations that are likely in the GPCR activation landscape, including multiple active- and inactive-state conformations. ActiveGEnSeMBLE starts with a systematic coarse grid sampling of helix tilts/rotations (~13 trillion transmembrane domain conformations) and identifies multiple potential active-state energy wells, using the TM3-TM6 intracellular distance as a surrogate activation coordinate. These energy wells are then sampled locally using a finer grid in conformational space to find a locally minimized conformation in each energy well, which can be further relaxed using molecular dynamics (MD) simulations. This method, combining homology modeling, hierarchical complete conformational sampling, and nanosecond scale MD, provides one of the very few computational methods that predict multiple candidates for active-state conformations and is one of the most computationally affordable.
G蛋白偶联受体（GPCRs）是细胞信号传导中至关重要的膜蛋白，使其成为治疗学的重要靶点。GPCRs的激活是其功能的核心，在其激活过程中需要多种构象。为了合理设计靶向GPCR的药物，获得GPCRs沿其激活途径的原子结构集合至关重要。这对结构测定实验来说极具挑战性，因此开发可靠的计算结构预测方法很有价值。特别是，由于活性状态构象的能量高于非活性状态构象（稳定性较差），它们难以稳定。此外，计算方法通常在设计上偏向于最低能量结构，从而遗漏了这些高能量但功能重要的构象。为了解决这个问题，我们开发了一种计算效率高的ActiveGEnSeMBLE方法，该方法系统地预测GPCR激活过程中可能出现的多种构象，包括多种活性和非活性状态构象。ActiveGEnSeMBLE首先对螺旋倾斜/旋转进行系统的粗网格采样（约13万亿个跨膜结构域构象），并使用TM3 - TM6细胞内距离作为替代激活坐标来识别多个潜在的活性状态能量阱。然后在构象空间中使用更精细的网格对这些能量阱进行局部采样，以在每个能量阱中找到局部最小化的构象，该构象可使用分子动力学（MD）模拟进一步优化。这种方法结合了同源建模、分层完全构象采样和纳秒级MD，是极少数能够预测活性状态构象多个候选结构的计算方法之一，也是计算成本最低的方法之一。
