Kurki Milla, Nesterenko Alexey M, Alsaker Nicolai E, M Ferreira Tiago, Kyllönen Sami, Poso Antti, Bartos Piia, Miettinen Markus S
School of Pharmacy, University of Eastern Finland, 70211 Kuopio, Finland.
Computational Biology Unit, Department of Informatics, University of Bergen, 5008 Bergen, Norway.
J Phys Chem B. 2024 Dec 19;128(50):12483-12492. doi: 10.1021/acs.jpcb.4c04719. Epub 2024 Dec 9.
Atomistic molecular dynamics (MD) simulations are a much-used tool for investigating the structure and dynamics of biomembranes with atomic resolution. The validity of the representations obtained is determined by the accuracy and realism of the MD model (force field). Here, we evaluated the proprietary OPLS4 force field of Schrödinger, Inc. against atomic-resolution experimental data, and compared its performance to CHARMM36, one of the best-performing openly available force fields. As a benchmark, we used high-resolution nuclear magnetic resonance (NMR) order parameters for C-H bonds─directly and reliably calculable from MD simulations─measured in phosphatidylcholine (PC) lipid bilayers under varying hydration conditions. Comparisons were made with two dehydration data sets: for saturated (1,2-dimyristoylphosphatidylcholine, DMPC) lipid bilayers from the literature and for unsaturated (1-palmitoyl-2-oleoylphosphatidylcholine, POPC) lipid bilayers measured here. Our findings indicate that OPLS4 reproduces the structure and dehydration response of PC-lipid bilayers fairly well, even slightly outperforming CHARMM36. Both models' main inaccuracies appear in (1) the order parameter magnitudes in the glycerol backbone and unsaturated carbon segments and (2) the qualitatively differing structural response of the PC headgroup to dehydration compared to experiments. In summary, this work underscores the importance of independent validation for (proprietary) force fields and highlights the striking similarities and nuanced differences between OPLS4 and CHARMM36 in describing biomembranes.
原子分子动力学（MD）模拟是一种常用工具，用于以原子分辨率研究生物膜的结构和动力学。所获得表征的有效性取决于MD模型（力场）的准确性和真实性。在此，我们根据原子分辨率实验数据评估了Schrödinger公司专有的OPLS4力场，并将其性能与CHARMM36（性能最佳的公开可用力场之一）进行了比较。作为基准，我们使用了在不同水合条件下，在磷脂酰胆碱（PC）脂质双层中测量的C-H键的高分辨率核磁共振（NMR）序参数——可直接且可靠地从MD模拟中计算得出。我们与两个脱水数据集进行了比较：一个是文献中饱和（1,2-二肉豆蔻酰磷脂酰胆碱，DMPC）脂质双层的数据集，另一个是此处测量的不饱和（1-棕榈酰-2-油酰磷脂酰胆碱，POPC）脂质双层的数据集。我们的研究结果表明，OPLS4能较好地再现PC脂质双层的结构和脱水响应，甚至在性能上略优于CHARMM36。两种模型的主要不准确之处在于：（1）甘油主链和不饱和碳段中的序参数大小；（2）与实验相比，PC头基对脱水的结构响应在性质上存在差异。总之，这项工作强调了对（专有）力场进行独立验证的重要性，并突出了OPLS4和CHARMM36在描述生物膜方面的显著相似性和细微差异。
