Institute for Protein Research, Osaka University
(From the above) Yasushige Yonezawa, Shusuke Yamanaka,
Hiromitsu Shimoyama, Hideki Yamazaki and Haruki Nakamura
RIKEN, Computational Science Research Program
Ikuo Fukuda

　Our group is attempting to elucidate the mechanism of functions of biological macromolecules, proteins and nucleic acids, at multiple levels including molecular, atomic and electronic levels using molecular simulation. Actual molecules in living matter show thermally fluctuating actions at room temperature, and the fluctuation has the essential roles on molecular functions. Our purpose is to elucidate the mechanism of such biological molecules including their electronic state while simulating the thermal fluctuation in details.

　For this achievement, we have developed a multiscale simulation program, Platypus (PLATform for dYnamic Protein Unified Simulation ), which couples quantum mechanics (QM) with molecular dynamics (MD) which were developed individually. Quantum mechanics to calculate electronic state is a computation method essential for handling enzymatic functions in simulation, and here, computation by the Hartree-Fock (HF) method, the density functional theory (DFT), CASSCF and CIS are available. Molecular dynamics is a simulation method, which reproduces the thermal fluctuation of molecules. Platypus is an integral of the program which realizes massive parallelism with the original codes, which has accelerated performance of up to 8192 parallels in the computations of both HF and DFT in the development so far, and particularly, it shows good scalability up to 1000 to 2000 parallels. In addition, Platypus loads the MD computation with an independently developed coarse graining model (CGM), with which the multiscale molecule simulation of QM/MD/CGM can be implemented. Moreover, we are developing an algorithm which performs efficient sampling with the QM/MD computation.

　As one of the applications of Platypus , we introduce research on the cis-trans isomerization mechanism of the proline residue, which is related to the signal molecular control of various life activities. In a study of small peptides including proline in water (reference 1), the transition state between cis and trans forms the conventionally known pyramid as well as the inverse pyramid, and it was firstly shown that this state fluctuates between these 2 forms (Figure 1). From a study of the Pin1 isomerase, which accelerates the cis-trans isomerization of the proline residue, we succeeded in grasping the mechanism, in which the strain occurring by binding of the peptide including proline to the enzyme active site stabilizes the energy in the transition state so as to accelerate isomerization.

　In the molecular dynamics simulation, the time for computation of interatomic long-range forces not based on chemical bonds between molecules accounts for a large amount. By parallelizing this part, the computation time can be considerably shortened. The Ewald method is a general method most often used, but it is known that this Ewald method is not suitable for parallelization of a very large system. We are therefore pursuing research to revise the long-range force potential recently developed by Wolf et al., and develop our original Force Switching-Wolf method (FSw-Wolf method), by which a consistent and steady simulation can be performed instead of the Ewald method. We succeeded in showing by simulation of a fused sodium salt (reference 2 and Figure 2) and a short peptide in water that the FSw-Wolf method can realize the computation at a precision comparable to that of the Ewald method. The FSw-Wolf method is not only a very convenient algorithm, but also it considers only the force between atoms at a relatively short distance. Thus, it is very suitable for large scale parallelized molecular dynamics simulation. Moreover, it is considered that this method will become a key method which can realize highspeed parallel computation by combining with other algorithms.

Figure1: Free energy landscape of cis-trans isomerization of a peptide including a proline residue, which was elucidated by QM/MD multiscale simulation. A transition state largely fluctuating between cis and states is observed.

Figure2: The energy error between the FSw-Wolf method and the Ewald method, plotted as a function of the cutoff distance and the parameter α in the FSw-Wolf method. This shows that the FSw-Wolf method can reproduce an energy very close to that in the Ewald method.

Reference 1. Yonezawa Y., Nakata K., Sakakura K., Takada T., Nakamura H., J. Am. Chem. Soc. 131(12), 4535-4540.
Reference 2. Fukuda I., Yonezawa Y., Nakamura H., J. Phys. Soc. Jpn. 77(11), 114301, 2008.