Thermal energy transport and signal transduction of biomolecular machines: Molecular dynamics simulation study of proteins

概要

学位報告４

別紙４
報告番号

※

甲

主

第

号

論

文

の

要

旨

論 文 題 目 Thermal energy transport and signal transduction of
biomolecular machines: Molecular dynamics simulation study of
proteins
(生体分子機械の熱エネルギー輸送と情報伝達：タンパク質の分子動力学シ
ミュレーション)
氏
名 WANG Tingting

論 文 内 容 の 要 旨
Proteins are crucial macromolecules in living organisms, performing various essential
functions. Thermal energy transport is among the essential biophysical properties of proteins,
but its relationship with protein structures, dynamics, and functions is still elusive.

The structures of folded proteins are highly inhomogeneous, giving rise to an anisotropic and
non-uniform flow of thermal energy through different types of pathways and interactions. To
illustrate such transport nature of proteins, I developed a theoretical framework for analyzing
the local thermal transport properties of thermal energy based on the autocorrelation function
formalism using equilibrium molecular dynamics (MD) simulations. In addition, advanced
machine learning-based methods were utilized to illustrate the contributing factors of
thermal transport in protein and the structural characterization of intrinsically disordered
regions.

First, the overall thermal conductivity of an α-helical protein (HP36) was calculated as 0.26
 0.01 W/(m K) based on the linear response theory using equilibrium MD simulations. The
local heat transport properties along peptide chains were studied using a linearhomopolymer-like model, where the entire protein was divided into small pieces by residues
and heat flows only within each residue (intra-) and between two peptide-bonded residues

(inter). Short-range cross-correlation corrections were introduced and employed to correct the
model for redistributing the overestimated contributions from partial heat currents to the
total heat current of the model due to the non-neglectable independent fluctuations between
residues. The model reproduced the exact value of the protein thermal conductivity, derived
from the total heat current, within 1% error. The local thermal transport calculation indicates
that intra-residue thermal transport has a dominant contribution to the overall heat current
and residue-wise thermal conductivity demonstrated distinct residue-type dependence: their
values decreased in a order of charged, polar, and hydrophobic residues.

Second, the local thermal transport properties of non-bonded contacts in proteins were
studied by introducing a derived physical quantity of inter-residue thermal conductivity
(λinter). As a results, λinter of non-bonded contacts exhibited a decreasing trend on interaction
types: hydrogen bonding > π-stacking > electrostatic > hydrophobic. Furthermore, the λinter
values are proportional to the hydrogen bond occurrence probability ( PHB), a measure of the
average number of hydrogen bonds during MD simulations. The random forest regression
prediction model was used to investigate the non-linear relationship between λinter and
various static and dynamical properties of the protein. The contact distance, variance in
contact distance, and PHB were found to be the three most important factors.
As a case study, the allosteric signaling process of the oxygen sensor domain of BjFixL protein
was investigated by constructing the vibrational energy exchange network (EEN) model. In
addition, a hybrid approach of deep learning algorithms, MD simulation, small angle X-ray
scattering (SAXS), and double electron-electron resonance electron paramagnetic resonance
(DEER/EPR) spectroscopy, was developed to characterize the ensemble of a chloroplast
protein, CP12, which contains intrinsically disordered regions (IDRs).
In summary, a theoretical framework of site-selective heat current analysis for analyzing the
local thermal transport was developed based on the linear response theory using equilibrium
MD simulations. The role of thermal energy transport in the signaling mechanism was
investigated by using the energy exchange network model. Furthermore, machine learning
methods were employed to solve complicated problems, such as the non-linear relationship
between heat transfer pathways and static and dynamical properties of protein and the
structural characterization of proteins with IDRs and their complexes in structural biology.