Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
ROYAL INSTITUTE
OF TECHNOLOGY
Molecular Dynamics Simulations of
Biomimetic Carbohydrate Materials
Qiong Zhang
(张 琼)
Doctoral Thesis in Theoretical Chemistry and Biology
School of Biotechnology
Royal Institute of Technology
Stockholm, Sweden 2011
© Qiong Zhang, 2011
ISBN 978-91-7415-966-0
ISSN 1654-2312
TRITA-BIO-Report 2011:12
Printed by Universitetsservice US-AB,
Stockholm, Sweden, 2011
I
Abstract
The present thesis honors contemporary molecular dynamics simulation methodologies which
provide powerful means to predict data, interpret observations and widen our understanding
of the dynamics, structures and interactions of carbohydrate systems. With this as starting
point my thesis work embarked on several cutting edge problems summarized as follows.
In my first work the thermal response in crystal cellulose Iβ was studied with special
emphasis on the temperature dependence of the crystal unit cell parameters and the
organization of the hydrogen bonding network. The favorable comparison with available
experimental data, like the phase transition temperature, the X-ray diffraction crystal
structures of cellulose Iβ at room and high temperatures, and temperature dependent IR
spectra supported our conclusions on the good performance of the GLYCAM06 force field for
the description of cellulose crystals, and that a cautious parameterization of the non-bonded
interaction terms in a force field is critical for the correct prediction of the thermal response in
cellulose crystals.
The adsorption properties of xyloglucans on the cellulose Iβ surface were investigated in
my second paper. In our simulations, the interaction energies between xyloglucan and
cellulose in water were found to be considerably lower than those in vacuo. The van der
Waals interactions played a prevailing role over the electrostatic interactions in the adsorption.
Though the variation in one side chain did not have much influence on the interaction energy
and the binding affinity, it did affect the structural properties of the adsorbed xyloglucans.
The interaction of the tetradecasaccharide XXXGXXXG in complex with the hybrid
aspen xyloglucan endo-transglycosylase PttXET16-34 was studied in the third paper. The
effect of the charge state of the “nucleophile helper” residue Asp87 on the PttXET16-34
active site structure was emphasized. The results indicate that the catalysis is optimal when
the catalytic nucleophile is deprotonated, while the “helper” residue and general acid/base
residue are both protonated.
In my forth paper, the working mechanism for a redox-responsive bistable [2]rotaxane
based on an α-cyclodextrin ring was investigated. The umbrella sampling technique was
employed to calculate the free energy profiles for the shuttling motion of the α-cyclodextrin
ring between two recognition sites on the dumbbell of the rotaxane. The calculated free
energy profiles verified the binding preferences observed experimentally. The driving force
for the shuttling movement of the α-cyclodextrin ring was revealed by the analysis of the free
energy components.
keywords: molecular dynamics simulation, carbohydrate, cellulose, xyloglucan, cyclodextrin
II
Preface
The work presented in this thesis has been carried out at the Department of Theoretical
Chemistry and Biology, School of Biotechnology, Royal Institute of Technology, Stockholm,
Sweden and Laboratory for Advanced Materials and Institute of Fine Chemicals, East China
University of Science and Technology, Shanghai, P. R. China.
List of papers included in the thesis
Paper I A Molecular Dynamics Study on the Thermal Response of Crystalline Cellulose
Iβ
Zhang, Q.; Bulone, V.; Ågren, H.; Tu, Y.
Cellulose 2011, 18, 207–221.
Paper II The Adsorption of Xyloglucan on Cellulose: Effects of Explicit Water and Side
Chain Variation
Zhang, Q.; Brumer, H.; Ågren, H.; Tu, Y.
(submitted for publication)
Paper III Molecular Dynamics Simulations of a Branched Tetradecasaccharide Substrate
in the Active Site of a Xyloglucan endo-transglycosylase
Mark, P.; Zhang, Q.; Czjzek, M.; Brumer, H.; Ågren, H.
Mol. Sim. 2011 (accepted)
Paper IV Working Mechanism for a Redox Switchable Molecular Machine Based on
Cyclodextrin: A Free Energy Profile Approach
Zhang, Q.; Tu, Y.; Tian, H.; Zhao, Y-L.; Stoddart, J. F.; Ågren, H.
J. Phys. Chem. B 2010, 114, 6561–6566.
Comments on my contribution to the papers included
I, as the first author, took the major responsibility for papers I, II, and IV.
I, as the coauthor, was responsible for the calculations, discussion and revision of paper III.
III
List of main publications not included in the thesis
Paper I Molecular Dynamics Simulations of Local Field Factors
Zhang, Q.; Tu, Y.; Tian, H.; Ågren, H.
J. Chem. Phys. 2007, 127, 014501.
Paper II Molecular Dynamics Simulations of Polycarbonate Doped with Lemke
Chromophores
Zhang, Q.; Tu, Y.; Tian, H.; Ågren, H.
J. Phys. Chem. B 2007, 111, 10645-10650.
Paper III Electric Field Poled Polymeric Nonlinear Optical Systems: Molecular
Dynamics Simulations of Poly(methyl methacrylate) Doped with Disperse Red
Chromophores
Tu, Y.; Zhang, Q.; Ågren, H.
.J. Phys. Chem. B 2007, 111, 3591-3598.
Paper IV Photochromic Spiropyran Dendrimers: ‘Click’ Syntheses, Characterization,
and Optical Properties
Zhang, Q.; Ning, Z.; Yan, Y.; Qian, S.; Tian, H.
Macromol. Rapid Commun. 2008, 29, 193-201.
Paper V ‘Click’ Synthesis of Starburst Triphenylamine as Potential Emitting Material
Zhang, Q.; Ning, Z.; Tian, H.
Dyes and Pigments 2009, 81, 80-84.
Paper VI Dye-sensitized Solar Cells Based on Bisindolylmaleimide Derivatives
Zhang, Q.; Ning, Z.; Pei, H.; Wu, W.
Front. Chem. China 2009, 269-277.
Paper VII Modulation of Iridium(III) Phosphorescence via Photochromic Ligands: a
Density Functional Theory Study
Li, X.; Zhang, Q.; Tu, Y.; Ågren, H.; Tian, H.
Phys. Chem. Chem. Phys, 2010, 12, 13730-13736.
IV
Acknowledgements
It is my great pleasure to extend my sincere acknowledgements to all those who have
contributed in one way or another to the accomplishment of my thesis work in the past few
years.
First of all, I wish to express my deep felt gratitude toward my supervisor, Prof. Hans Ågren,
for giving me this precious opportunity to work in our department and introducing me to the
field of molecular modeling. I appreciate all the freedom that you gave to me on choosing the
projects of interest. I am indebted for your helpful advice, strong support, and constant
encouragement, without which I would never have completed the work.
I would like to give my heartfelt thanks to my supervisor, Prof. He Tian, in East China
University of Science and Technology, for leading me into the research realm, for
recommending me to study in this group and for all your support on both my work and life.
I am truly thankful to my co-supervisor Dr. Yaoquan Tu for your patient guidance, instructive
discussions and considerable help in many ways. The nice collaborations and constructive
discussions with Prof. Vincent Bulone, Prof. Harry Brumer and Dr. Pekka Mark on the
cellulose and xyloglucan related projects, Prof. J. Fraser Stoddart and Dr. Yan-Li Zhao on the
molecular machine project, and Dr. Marten Ahlquist on the on-going enzymatic catalysis
project are highly acknowledged. Great discussions with my friends and colleagues,
especially Xin Li, Yu Kang, Dr. Ying Zhang, Dr. Shilu Chen and Dr. Rongzhen Liao are
greatly appreciated.
I am sincerely grateful to Prof. Yi Luo for your valuable guidance and positive advice on both
my work and life, for sharing your own experience and your view of life and for being so
considerate and generous to all of us. I appreciate a lot Prof. Boris Minaev, Prof. Kersti
Hermansson Dr. Zilvinas Rinkevicius and Dr. Paweł Sałek for your excellent lectures. Prof.
Faris Gel’mukhanov and Prof. Fahmi Himo are especially acknowledged for your particular
interest in Chinese culture and for all the fun. I would also like to thank Dr. Ying Fu, Dr. Olav
Vahtras and Staffan Hellström for your strong support and kind help on the laboratory life.
I am grateful to all colleagues and friends in our department for all the fruitful discussions and
advice, the pleasant working environment, and the good fun. There are so many of you but
please allow me to write down your names here: Dr. Feng Zhang, Dr. Shilu Chen, Dr.
Rongzhen Liao, Dr. Yuejie Ai, Dr. JunJiang, Dr. Tiantian Han, Dr. Kai Fu, Dr. Wenhua
Zhang, Dr. Na Lin, Dr. Bin Gao, Dr. Guangde Tu, Dr. Jicai Liu, Dr. Kailiu, Dr. Yanhua Wang,
V
Dr. Hui Cao, Dr. Katia J. de Almeida, Dr. Kathrin Hopmann, Xin Li, Yu Kang, Lu Sun, Dr.
Ying Zhang, Dr. Sai Duan, Dr. Hao Ren, Xing Chen, Dr. Yuping Sun, Xiao Cheng, Dr.
Xiaofei Li, Dr. Keyan Lian, Dr. Ke Zhao, Dr. Weijie Hua. Dr. Qiang Fu, Fuming Ying, Dr.
Xifeng Yang, Dr. N. Arul Murugan, Dr. Kestutis Aidas, Bogdan Frecus, Xiangjun Shang,
Zhihui Chen, Dr. Junkuo Gao, Dr. Xiaohao Zhou, Ying Hou, Dr. Hongmei Zhong, Jiayuan Qi,
Chunze Yuan, Qiu Fang, Guangjun Tian, Li Li, Bonaman Xin Li, Hongbao Li, Xiuneng Song,
Yong Ma, Lili Lin, Quan Miao, Ying Wang and Yongfei Ji ...... Thank all of you for all those
happy, colorful, and unforgettable memories! Special thanks to my close friends – Feng,
Yuejie, Yu, Xin Li and Ying Zhang, for sharing happiness and sadness in our life, for the
strong support and great encouragement and for our forever friendship.
I would like to thank the China Scholarship Council for financial support and the Swedish
National Infrastructure for Computing (SNIC) for providing powerful computational
capability.
Finally, my wordless gratitude goes to my husband Zhijun and my parents for your constant
support, forever love and everything!
VI
Abbreviations
AFM Atomic force microscopy
AMBER Assisted Model Building with Energy Refinement
Asp Asparate
CD Cyclodextrin
CHARMM Chemistry at HARvard Molecular Mechanics
ESP Electrostatic potential
Glc Glucose
Glu Glutamate
LINCS Linear Constraint Solver
MD Molecular Dynamics
NPT Isothermal-isobaric (ensemble)
NVE Microcanonical (ensemble)
NVT Canonical (ensemble)
OPLS Optimized Potentials for Liquid Simulations
PBC Periodic boundary condition
PMF Potential of mean force
RESP Restrained ElectroStatic Potential
SMD Steered molecular dynamics
TTF Tetrathialfulvalene
TZ Triazole
WHAM Weighted Histogram Analysis Method
XET Xyloglucan endo-transglycosylase
XG Xyloglucan
VII
Contents
1 Introduction ...................................................................................................................... 1
2 Molecular Dynamics Algorithms...................................................................................... 9
2.1 Basic Principle – the Newton’s Equation of Motion .................................................. 9
2.2 Numerical Integration Algorithms ............................................................................10
2.2.1 The Verlet Algorithm ......................................................................................10
2.2.2 The Leap-frog Algorithm................................................................................11
2.2.3 The Velocity-Verlet Algorithm........................................................................11
2.3 Periodic Boundary Condition ...................................................................................12
2.4 Treatment of Non-bonded Interactions .....................................................................13
2.5 Constraint Algorithms ..............................................................................................15
2.6 Temperature Coupling and Pressure Coupling Methods............................................16
2.6.1 Temperature Coupling Method .......................................................................16
2.6.2 Pressure Coupling Method..............................................................................18
3 Force Field Models ..........................................................................................................21
3.1 The Potential Energy Function Form........................................................................21
3.2 Force Field Parameterization....................................................................................25
3.3 Force Fields for Carbohydrates ................................................................................27
4 Potential of Mean Force ..................................................................................................29
4.1 Definition of Potential of Mean Force ......................................................................29
4.2 Methods for the Potential of Mean Force Calculation ...............................................31
4.2.1 Umbrella Sampling.........................................................................................31
4.2.2 Steered Molecular Dynamics Simulations.......................................................33
5 Summary of Papers .........................................................................................................37
5.1 Thermal Response of Crystalline Cellulose Iβ (Paper I) ...........................................37
VIII
5.2 Adsorption of Xyloglucan on Cellulose (Paper II) ....................................................42
5.3 Carbohydrate-Enzyme Interaction (Paper III) ...........................................................45
5.4 Working Mechanism of an α-cyclodextrin Based Rotaxane (Paper IV).....................48
6 Conclusions and Outlook.................................................................................................55
1
Chapter 1
Introduction
Face the past with the least lamentation
Face the present with the least waste
With the most dream to face future
Nature, through billions of years of trial and error, has evolved with effective and marvelous
solutions to innumerable complicated real-world problems. Humans have learnt from and
have been inspired by a variety of phenomena in nature. As the industrial age advances into
the biological era, we are able to penetrate deeper into nature’s secrets with the aid of modern
scientific techniques, to understand and learn more from her elegant designs. Biomimetics is a
field of science that investigates biological structures and processes for their use as models for
the development of artificial systems. [1] Biomimetic approaches are promising in the
development of new high-performance and environment-friendly materials. In this thesis,
properties related to biomimetic carbohydrate materials are investigated by molecular
dynamics (MD) simulation technique in atomic detail.
Carbohydrates, like other biomolecules such as proteins and nucleic acids, play important
roles in living things. Carbohydrate is the common name of a class of organic compounds
including polyhydroxy aldehydes, polyhydroxy ketones, and substances which yield such
compounds upon hydrolysis. Carbohydrates gained their name since the simplest sugars have
the empirical formula (CH2O)n, where n is three or more. Carbohydrates are classified by size
into mono-, di-, oligo- and polysaccharides. The open-chain and the closed-ring forms of a
monosaccharide often coexist, the latter being more common. The most occurring
carbohydrate rings are six-membered rings (pyranoses) and five-membered rings (furanoses).
There are several stereocenters in a monosaccharide and it is assigned as D- or L-
Chapter 1 Introduction
2
configuration according to the orientation of the asymmetric carbon most distant from the
carbonyl group. Take glucose for example, in a standard Fischer projection when the hydroxyl
groups on C2, C4, and C5 are on the right it is a D-glucose, otherwise it is an L-glucose (Figure
1.1 upper). A monosaccharide can also be classified as α- and β- anomers on the basis of the
position of the hydroxyl group on the anomeric carbon (C1) relative to the CH2OH group
bound to C5. They are either on the opposite sides in the α-anomer or the same side in the
β-anomer (Figure 1.1 lower).
Figure 1.1 (upper) The D- and L-glucose. (lower) The α- and β-glucose.
The plant cell wall is a dynamic biocomposite material, comprised of load-bearing
paracrystalline cellulose fibrils embedded in a matrix of cross-linking non-cellulosic
polysaccharides, (glyco)proteins, and polyphenolics, which undergoes continuous changes
during tissue differentiation and plant growth. [2] The cell wall accommodates a variety of
mechanical requirements during plant growth and demonstrates physical and mechanical
properties comparable to those of the best synthetic materials. In addition, declining oil
resources impose a need for seeking sustainable and renewable substitutes for
petrochemical-based materials. The cell walls represent highly sophisticated composite
materials that can provide inspiration on the design and fabrication of carbohydrate-based
materials that combine environmental friendliness and biocompatibility with high
performance and increased functionality for applications in advanced packaging, electronic
devices, vehicles and sports equipment, etc. [1]
Cellulose is the main constituent of plant cell walls and is the most abundant natural
polymer on Earth. Each year, photosynthesis converts more than 100 billion metric tons of
CO2 and H2O into cellulose and other plant products. [3] Cellulose is a linear polymer of
(1→4)-linked D-glucosyl units (Figure 1.2). Due to the low density and excellent load-bearing
Chapter 1 Introduction
3
properties, cellulose becomes an important raw material for many industries, such as paper,
wood, textiles and many other materials. [4, 5]
In plants, the extended glucan chains in cellulose interact with one another by hydrogen
bonds to form cellulose sheets. These sheets are further stacked on top of each other via van
der Waals interaction and weak hydrogen bonds. Thereby, a network of up to an average of 36
cellulose chains makes up the crystalline core of the cellulose fibril. The crystalline and
paracrystalline (amorphous) cellulose core is associated with hemicellulose (mainly
xyloglucan) and other polymers to form linear structures of high tensile strength known as
microfibrils. Each microfibril is about 10 to 20 nm in diameter. These crystalline microfibrils
act as the main load-bearing components of the cell wall.
Cellulose Chain
Figure 1.2 The hierarchical structure of cellulose microfibril. Schematic representation of a cellulose chain
comprising atom labeling and the definition of different torsion angles: Φ: O5-C1-O4’-C4’; Ψ:
C1-O4’-C4’-C5’; ω: O5-C5-C6-O6; τ2: C3-C2-O2-HO2; τ3: C4-C3-O3-HO3; τ6: C5-C6-O6-HO6.
Cellulose is polymorph and adopts various possible packing arrangements, with either
parallel chains (cellulose I, IIII, and IVI) or antiparallel chains (cellulose II, IIIII, and IVII) (for
a review of cellulose see ref. [6]) The so-called native cellulose corresponds to cellulose I and
occurs in the form of two different allomorphs represented by two distinct crystal phases
designated as I and Iβ, showing triclinic and monoclinic symmetry, respectively. The
main structural difference between cellulose I and Iβ is the relative position of the chains to
each other. Allomorph Iβ is thermodynamically more stable and can irreversibly be
transformed from I by annealing. The proportions of I and Iβ phases vary with the origin of
cellulose. Cellulose in higher plants (woody tissue, cotton etc.) consists mainly of the Iβ phase
Chapter 1 Introduction
4
whereas primitive organisms (bacteria, algae etc.) are enriched in the I phase. [7], [8] The
structures of different forms of crystalline cellulose have been elucidated with the aid of the
synchrotron X-ray and neutron fiber diffraction techniques. [9, 10, 11, 12, 13]
The properties of most cellulose-based materials used industrially are modified by
chemical treatment during the manufacturing process, for example to enhance the brightness
and to increase the smoothness and hydrophobicity of paper surfaces. However, the hydrogen
bond network could be destroyed and hence the structural integrity and mechanical strength
of the cellulose microfibril will be weakened by extensive chemical modification on the
hydroxyl groups. [14, 15, 16] Hence, in applications where the superb load bearing properties of
cellulose are required, coating, instead of direct chemical modification, is used to alter the
surface and the bulk mechanical properties of cellulose. In biocomposites, cellulose is usually
blended in a man-made polymeric matrix to achieve good mechanical properties. [17] However,
one major problem often encountered is the poor interfacial interactions between the natural
fibers and the polymeric matrix.
In 2004, a novel chemoenzymatic approach for the efficient incorporation of chemical
functionality onto cellulose surfaces was developed in the laboratory of Wood Biotechnology
at KTH. [18] This creative idea is inspired by the organization of the plant cell wall in which
the combination of strength and interfacial stability is achieved by coating cellulose with
xyloglucan (XG) that can interact with other cell wall polymers and proteins. The main
principle of this biomimetic method is to perform chemo-enzymatic synthesis on XG instead
of the hydroxyl groups on the cellulose surface. The modification is realized by using a
transglycosylating enzyme xyloglucan endo-transglycosylase (XET, E.C.2.4.1.207) to join
chemically modified XG oligosaccharides to XG, which has a naturally high affinity to
cellulose. [18]
Therefore, understanding the structural properties of cellulose, the interaction details of
the cellulose-XG assembly and the catalytic details of the enzyme XET on the substrate XG is
essential to improve our knowledge of the ultra-structural organization and function of the
plant cell wall and can further inform the use of XG in applications for biomimetic cellulose
modification. Due to a paucity of experimental methods for high-resolution (atomic-level)
structural analysis of poorly-crystalline solid materials, [19] computer simulations provide a
powerful alternative to explore atomic-level interpretation.
Thermal behavior is one of the most fundamental properties of solid-state materials.
Progress in the investigation of the thermal behavior of neat cellulose Iβ has been made more
recently by X-ray diffraction [20, 21] and temperature-dependent IR measurements [22, 23] using
Chapter 1 Introduction
5
highly crystalline cellulose samples. Temperature-dependent IR measurements have been
performed to explore the temperature-dependent changes in hydrogen bonds in the crystal,
and to elucidate the transition process at the functional group level. [22] The thermal response
of crystalline cellulose I is further explored at the atomic level by MD simulation in Paper I.
Figure 1.3 Common structural motifs of xyloglucan and xylogluco-oligosaccharides: XXXG (x = 0, y = 0,
z = 0), XLXG (x = 1, y= 0, z = 0), XXLG (x = 0, y = 1, z = 0), XLLG (x = 1, y = 1, z = 0) and XXFG (x =
0, y = 1, z = 1). The arrow indicates the site for endo-glucanase cleavage.
Contemporary models of the primary cell wall of dicots and diverse monocots suggest
that the non-covalent network formed by insoluble cellulose chains and the non-crystalline
polysaccharide XG is a central contributor to wall strength and extensibility. [2, 24, 25] XG is a
complex branched polysaccharide consisting of a cellulose-like β-(1→4)-glucan main chain
regularly substituted with branching -(1→6) xylosyl residues. Branching is both extensive
and highly regular, with xylose substitution at 50% to 75% of the glucosyl units, depending
on the source. [26, 27] The branching chains on XG dramatically change the physical properties
of the polymer compared to cellulose; XG is highly water soluble and cannot form ordered
crystalline microfibrils like cellulose. [28] In XGs typical of many dicots, the repeating motif is
based on the XXXG heptasaccharide. Further galactosylation and fucosylation give rise to
other common oligosaccharide units such as XXLG, XLXG, XLLG and XXFG. A single-letter
linear notation is widely used to represent the complex branching pattern of XG (See Figure
1.3). [29]
XG molecules can adsorb onto the cellulose surface to form a sheath around the cellulose
microfibrils preventing them from direct aggregation. Some XG molecules are trapped in or
between the microfibrils. Other XG molecules are believed to span the gap between different
cellulose microfibrils, providing a flexible and strong network in the primary cell wall. The
Chapter 1 Introduction
6
excellent review by Brumer et al. [30] presented a detailed overview on the knowledge of XG
structures, solution properties, and the mechanisms of interaction of XGs with cellulose. It has
long been suggested that the XG-cellulose interaction may involve alignment of the
polysaccharide chains. [31] Early molecular modeling adopted this hypothesis, although studies
differed on whether interchain association is mediated by glucose-glucose ring stacking
interactions or side-on hydrogen bonds. [32, 33, 34] However, a contemporary MD simulation by
Hanus and Mazeau has seriously questioned the necessity of a parallel chain alignment in the
XG-cellulose interaction. [35] XGs with different starting orientations can bind to all five
unique crystal faces of cellulose Iβ with similar binding energies. [35] One potential concern
with the work of Hanus and Mazeau is that the binding simulations were performed in vacuo,
i.e. in the absence of solvating water molecules. In paper II, we have re-examined the
XG-cellulose Iβ interaction in the presence of explicit water molecules.
The cleavage and non-hydrolytic re-arrangement of XG are catalyzed in vivo by XETs.
Thus XETs have the potential to modify the cellulose XG network and are most likely
involved in the reconstruction of cell walls during the plant growth and development. [36] XET
is related to the retaining glycosyl hydrolases. It catalyzes in the first reaction step the
cleavage of a glycosidic bond, which leads to the formation of a covalent enzyme-glycosyl
intermediate. In the second step, XET transfers the enzyme-glycosyl intermediate to a
carbohydrate acceptor. [ 37 , 38 , 39 ] The interaction between the tetradecasaccharide
XXXGXXXG in complex with the hybrid aspen xyloglucan endo-transglycosylase
PttXET16-34 was studied in Paper III.
A molecular machine is defined as a number of discrete molecular components which
perform mechanical-like movements (output) in response to specific external stimuli (input). [40] The most complex molecular machines are found within cells. These include motor
proteins, such as kinesin, which moves cargo inside cells away from the nucleus along
microtubules, myosin, which is responsible for muscle contraction, and dynein, which
produces the axonemal beating of motile cilia and flagella. [41] Construction of artificially
molecular machines is inspired by these biological machines and has been an active area of
supra-molecular research.
Cyclodextrins (CDs) are one kind of attractive ring components in the construction of
molecular machines. CDs are a family of cyclic oligosaccharides with sugar molecules bound
in a ring. The most common CDs are made up of six, seven and eight D-glucopyranosyl
residues (named as α-, β-, and γ-CD respectively) linked by α-1,4 glycosidic bonds. The
glucose residues have the 4C1 (chair) conformation. CDs can be topologically represented as
Chapter 1 Introduction
7
toroids with the larger and the smaller openings of the toroid corresponding to the secondary
and primary hydroxyl groups, respectively. The working mechanism for a redox-responsive
bistable [2]rotaxane incorporating an α-CD was studied by MD simulations in Paper IV.
Figure 1.4 The structure of α-CD
9
Chapter 2
Molecular Dynamics Algorithms
MD simulation is a technique used to evolve the motions of nuclei in a system composed of
nuclei and electrons. In MD simulation, the rapid motion of the electrons is averaged out
assuming that they adjust instantaneously to the comparatively slow motion of the heavy
nuclei (Born-Oppenheimer approximation). Thereby, the Schrödinger equation can be
separated into a time-dependent equation for the nuclei and a time-independent equation for
the electronic degrees of freedom. In classical MD simulations, the quantum aspects of the
nuclear motion are also neglected so that classical mechanics can be used to describe the
motion of atoms. Each nucleus propagates according to the Newton’s equation of motion
under the potential provided by all the nuclei and electrons in the system, thereby the structure
and other properties of the whole system are calculated.
2.1 Basic Principle – the Newton’s Equation of Motion
For a system of N atoms, the potential energy, U(r1, r2,..., rn), is a function of the coordinates
of these N atoms. To a good approximation, nuclei behave as classical particles and their
dynamics can thus be simulated by numerically solving the Newton’s equation of motion:
2
2= i
i i
i
dUm
dt
rF
r (2.1)
with 2
2
i ii
d d
dt dt
r va (2.2)
Here U is the potential energy of the system at time t. Fi is the force on atom i. mi and ri
Chapter 2 Molecular Dynamics Algorithms
10
are the mass and the coordinate of atom i. ai and vi are the acceleration and the velocity of
atom i, respectively.
2.2 Numerical Integration Algorithms
To describe the atomic motion, there are several finite-difference methods developed for the
numerical integration of the Newton’s equation of motion. In mathematics, finite-difference
methods are numerical methods for approximating the solutions to differential equations by
approximating derivatives with finite difference equations. The most popular integration
methods in MD simulations are the Verlet algorithm [42] and its velocity-explicit variants,
leap-frog [43] and velocity Verlet algorithms. [44, 45]
2.2.1 The Verlet Algorithm
Give the position at time t r(t), the position at a short time (Δt) later is represented by the
Taylor expansion:
2 32 3 4
2 3
( ) ( ) ( )1 1( ) ( ) ( )
2! 3!i i i
i i
d t d t d tt t t t t t O t
dt dt dt
r r rr r (2.3)
Substitute Δt with –Δt in eq. (2.3), the position at a short time (Δt) earlier is written as:
2 32 3 4
2 3
( ) ( ) ( )1 1( ) ( ) ( )
2! 3!i i i
i i
d t d t d tt t t t t t O t
dt dt dt
r r rr r (2.4)
Adding eqs. (2.3) and (2.4) gives the recipe for updating the configuration from the
current position and acceleration, and the previous position in the Verlet algorithm:
22
2
( )( ) 2 ( ) ( ) i
i i i
d tt t t t t t
dt
rr r r
22 ( ) ( ) ( )i i it t t t t r r a
(2.5)
Due to its simplicity and moderate requirement on memory, the Verlet algorithm was
widely used. The algorithm is exactly reversible in time. As seen from the Taylor expansions,
The error in the atomic positions is of the order of O(Δt4). Moreover, the new positions are obtained
Chapter 2 Molecular Dynamics Algorithms
11
by adding a small number ai(t)Δt2 to a difference between two large numbers (2ri(t) −
ri(t−Δt)). This may further lead to truncation errors due to finite precision. The Verlet
algorithm has another disadvantage that velocities do not appear explicitly, which is a
problem in connection with generating ensembles with constant temperature. The velocities
can be obtained via the following formula with an error of the order of O(Δt2).
( ) ( )( )
2i i
i
t t t tt
t
r rv (2.6)
2.2.2 The Leap-frog Algorithm
The leap-frog algorithm made improvement on the Verlet algorithm with respect to the
numerical aspect and the explicit velocity. The equations for propagating the atoms in the
leap-frog algorithm are given by:
( ) ( ) ( )2
i i i
tt t t t t
r r v (2.7)
( ) ( ) ( )2 2
i i i
t tt t t t
v v a (2.8)
A disadvantage is that the position and velocity updates are out of phase by half a time
step. Therefore, the exact kinetic energy contribution to the total energy cannot be calculated
at the same time as the positions are defined. The velocities at time t can be computed from:
)
2
1()
2
1(
2
1)( ttttt iii vvv
(2.9)
2.2.3 The Velocity-Verlet Algorithm
The disadvantage of the leap-frog algorithm is remedied in the velocity-Verlet algorithm,
where the position and velocity are updated by:
21( ) ( ) ( ) ( )
2i i i it t t t t t t r r v a (2.10)
Chapter 2 Molecular Dynamics Algorithms
12
1( ) ( ) ( ( ) ( ))
2i i i it t t t t t t v v a a (2.11)
The leap-frog algorithm is utilized in the GROMACS program. [46, 47] The velocity-Verlet
algorithm has become available in the GROMACS program since version 4.5. [48, 49]
2.3 Periodic Boundary Condition
MD simulations are performed on a limited number of particles at the cost of a reasonable
computational resource. In modern MD simulations, the number of particles in the simulated
system can reach a million, however, it is still far small compared with that in a macroscopic
piece of matter. In addition, the ratio between the number of surface particles and the total
number of particles would be much larger than in reality, which would lead to unwanted
surface artifacts in an MD model.
Figure 2.1 Two-dimensional representation of periodic boundary condition.
To solve this problem, the periodic boundary condition (PBC) is commonly applied in
MD simulations, where the central simulated box is surrounded by the periodic images of
itself. The arrangement and movement of particles in the periodic images are exactly the same
as in the central simulated box, which means that the particles move out from the box on one
side would re-enter the box instantaneously from the opposite side. This also ensures the
constant number of particles in the simulated system. However, caution needs to be exercised
to minimize the artifacts caused by the artificial periodicity if one wishes to simulate
non-periodic systems, such as liquids or solutions. [50, 51] The box shapes are limited to
Chapter 2 Molecular Dynamics Algorithms
13
space-filling boxes to ensure seamless connection between boxes, such as triclinic unit cells,
which cover cubic, rhombic dodecahedra, and truncated octahedra, etc. GROMACS is based
on triclinic unit cells.
2.4 Treatment of Non-bonded Interactions
Under PBC, the interactions between particles in different boxes are allowed. The number of
pair-wise interactions is increased to an enormous quantity because not only the interactions
between “real” particles but also those between “real” and “image” particles have to be
considered. The calculations of the two non-bonded terms (electrostatic and van der Waals
interactions) make up the computationally most expensive part in the integration step. In the
classical MD simulation, the inter-atomic interaction is often described by an empirical
potential defined in the force field. Most commonly, the van der Waals interaction is described
by the Lennard-Jones potential and the electrostatic interaction is modeled by interaction
between point charges according to Coulomb’s law (for details please refer to Chapter 3).
Usually, in order to enhance the computational efficiency, a cut-off radius (rcut) of about
1.0-1.5 nm is introduced for the explicit calculation of short-range non-bonded interactions.
The minimum image convention states that only one – the nearest – image of each particle is
considered for short-range non-bonded interaction terms, which also implies that the rcut used
to truncate non-bonded interactions may not exceed half the shortest box vector. [49]
With regard to the Lennard-Jones potential for the van der Waals interactions, the use of
cut-off requires long-range correction only for the dispersion term (−4εσ6/r6), since the
repulsion term (4εσ12/r12) decays so fast with respect to the inter-atomic distance that it can be
safely neglected beyond rcut. For a plain cut-off, the long-range correction for the
Lennard-Jones potential takes the form of
62
, 6
1 44 ( )
2cut
LJ LRC
r
V N r g r drr
(2.12)
where ρ=N/V is the particle density. If one assumes that the radial distribution function g(r)
equals 1 beyond rcut, the long-range correction simply reduces to:[45, 49]
6
, 3
8
3LJ LRC
cut
NV
r
(2.13)
Chapter 2 Molecular Dynamics Algorithms
14
On the contrary, for the long-range electrostatic interactions, the use of a cut-off is known
to cause severe artifacts due to its slow 1/r decay. [52, 53] Such long-range artifacts are
prevented by the use of lattice summation methods such as the Ewald summation methods [54]
capable of consideration of the full electrostatics in an efficient manner.
The idea of the Ewald methods is that the original electrostatic potential generated by
point charges is split into two separate parts: one short-range term that sums quickly in real
space and one long-range term that sums quickly in reciprocal space. An appropriate
Coubomb cut-off distance is chosen to separate the short- and long-range electrostatic
interactions. Two neutralizing charge distributions with exactly the same quantity but opposite
charge centered at the same position as each point charge are introduced within the Coubomb
cut-off distance, thereby yielding exactly the original point charge distribution (Figure 2.2).
The real space includes the original point charges and the charge distribution (screening
potential) with opposite charges to the original point charges. Usually, the screening potential
is chosen as a Gaussian. An appropriate selection of the Gaussian potential will lead to an
electrostatic potential approaching zero at certain distance in the real space. Therefore, the
electrostatic potential generated by the screened point charges become short-ranged in the real
space (Vreal) and can be simply evaluated. While the reciprocal space includes the negative of
the above screening potential within the Coubomb cut-off distance and Gaussian
representations for all the other point charges beyond the Coubomb cut-off distance. The
electrostatic potential in the reciprocal space (Vrecip) can be computed efficiently by Fourier
transformation methods. In addition, the self-interaction (Vself) introduced by the screening
potential must be subtracted. Finally, the total electrostatic potential is composed by three
parts:
coul real recip selfV V V V (2.14)
Figure 2.2 Representation of Ewald summation
Chapter 2 Molecular Dynamics Algorithms
15
By suitably adjusting the Gaussian function, optimal convergence of Vreal and Vrecip may
be achieved. The original Ewald method scales as O(N3/2) and therefore is not efficient
enough for use in very large systems. The Particle Mesh Ewald (PME) method, [55, 56] in
which the atomic charges are interpolated on grid points and the fast Fourier transformation is
used for the calculation of the reciprocal sum, is even more efficient than the Ewald
summation and scales as O(NlogN).
2.5 Constraint Algorithms
In MD simulations, a proper choice of integration step is important in order to save
computational time without loss of accuracy. When Newton’s equation of motion is integrated,
the highest frequency that occurs in the simulation system is the limiting factor of the chosen
time step. Usually, the vibrations of bonds involving hydrogen atoms are of the highest
frequency. For example, the bond stretching frequency of an O-H bond is about 104 Hz, which
limits the time step to be around 0.5 fs (the integration time step should be about 1/20 of the
fastest period).
In order to increase the integration time step, constraint algorithms are often introduced in
MD to maintain certain degrees of freedom (such as bond lengths or bond angles), which
allows to increase the time step to a typical value of 2 fs. Constraints can be imposed by the
traditional SHAKE [57] algorithm and LINCS (Linear Constraint Solver) algorithm. [58, 59]
In the SHAKE algorithm, the equations of motion must fulfill Nc holonomic constraints:
2 2( ) ( ) 0,k a b abt t d r r
1,..., ck N (2.15)
where the distance between two atoms, a and b, is constrained to a prescribed distance of dab
in the kth constraint and Nc the total number of constraints. The equations of motion are
expanded by the introduction of a constraining force in the form of Lagrange multipliers.
While in the LINCS algorithm, a different set of constraint equations is used:
( ) ( ) 0,k a b abg t t d r r
1,..., ck N (2.16)
The LINCS algorithm is 3 to 4 times faster than the SHAKE algorithm and it can be easily
parallelized. [58]
Chapter 2 Molecular Dynamics Algorithms
16
2.6 Temperature Coupling and Pressure Coupling Methods
The Hamiltonian of an isolated system is time-independent, translationally invariant and
rotationally invariant. Integration of the equations of motion for such a system leads to a
trajectory mapping a microcanonical (NVE) ensemble. However, the NVE ensemble does not
correspond to the conditions under which most experiments are performed. In most cases, a
canonical ensemble (NVT) or an isobaric-isothermal ensemble (NPT) will be a more suitable
representation of the real system. To achieve this, the equations of motion need to be modified
with temperature coupling and pressure coupling methods.
Table 2.1 Different types of thermodynamics ensembles used in the MD simulations
ensemble type constant parameters
microcanonical ensemble (NVE) fixed number of particles N, volume V and energy E
canonical ensemble (NVT) fixed number of particles N, volume V and temperature T
isobaric-isothermal ensemble (NPT) fixed number of particles N, pressure P and temperature T
grand canonical ensemble (μVT) fixed chemical potential μ, volume V and temperature T
2.6.1 Temperature Coupling Method
The temperature of the system T is related to the kinetic energy K by:
2
1
| | /2 / 2N
i i df Bi
K m N k T
(2.17)
where Ndf is the total number of degrees of freedom in the system, vi the velocity of particle i,
and kB the Boltzmann’s constant. The temperature of the system T can thus be adjusted
through the regulation of the velocity of particles. This is achieved by either the weak
coupling scheme of Berendsen, [60] the extended ensemble Nosé-Hoover Scheme, [61, 62] or the
velocity rescaling scheme. [63]
Berendsen Temperature Coupling
The Berendsen coupling scheme was developed in 1984 by Berendsen. The system is
supposed to be coupled with first-order kinetics to an external heat bath with reference
Chapter 2 Molecular Dynamics Algorithms
17
temperature T0. The change in temperature is proportional to the temperature difference
between the heat bath and the system, with a coupling strength determined by the coupling
parameter τ, as defined by:
0T TdT
dt
(2.18)
The heat flow into or out of the system is achieved by scaling the velocity every step with
the scaling factor λ:
01 ( 1)Tt
T
(2.19)
The Berendsen coupling scheme is simple to implement and efficient to relax the system
to the target temperature from far-from-equilibrium conditions. However, it does not produce
a correct canonical ensemble due to the neglect of the stochastic contribution to the
temperature fluctuations on the microscopic scale. The Berendsen coupling scheme is usually
employed before the production run.
Nosé-Hoover Temperature Coupling
The weak Berendsen thermostat is extremely efficient for relaxing a system to the target
temperature, but it might be more important to probe a correct canonical ensemble once the
system has reached equilibrium which can be achieved by the extended-ensemble
Nosé-Hoover scheme. In the Nosé-Hoover scheme, an additional degree of freedom s is
introduced with conjugate momentum Pξ and effective mass parameter Q. The system
Hamiltonian is modified by a friction term in the equation of motion, as given by:
2
2
i i i
i
d d
dt m dt
r F r (2.20)
where ξ = Pξ/Q. The equation of motion for the friction parameter ξ is:
20
d 1( 1)
di i df B
i
m N k Tt Q
v (2.21)
Chapter 2 Molecular Dynamics Algorithms
18
2.6.2 Pressure Coupling Method
The pressure P of the system is related to the box volume Vbox, the kinetic energy K and the
virial by:
2 1
3 2ij ij
i jbox
P KV
r F (2.22)
where Fij is the force on particle i exerted by particle j, and rij is the vector going from particle
j to i: rij =ri – rj. Similar to the temperature coupling, the system can also be coupled to a
“pressure bath” via weak Berendsen pressure coupling [60] or extended ensemble
Parrinello-Rahman pressure coupling. [64, 65]
Berendsen Pressure Coupling
In the Berendsen algorithm, the system is coupled to an external pressure bath with reference
pressure P0. The coordinates and box vectors are rescaled every step. The system pressure is
regulated with first-order kinetics according to the equation:
0
P
P PdP
dt
(2.23)
where τP is a time constant. The volume of the box is scaled by a factor χ, and the coordinates
are scaled by χ1/3 assuming the system is isotropic and the simulation box cubic. χ is given by:
01 ( )P
tP P
(2.24)
Here, is the isothermal compressibility, the value of which only affects the time
constant of the pressure relaxation but not the average value of pressure itself, thus is usually
set to the value of water, 4.6×10−10 Pa−1. The Berendsen pressure coupling does not yield the
exact NPT ensemble.
Parrinello-Rahman Pressure Coupling
Within the Parrinello-Rahman barostat, the box matrix h, formed by arranging the box vectors
Chapter 2 Molecular Dynamics Algorithms
19
{a, b, c}, obeys the matrix equation of motion:
21 T 1
02( ) ( )box
dV
dt
hW h P P (2.25)
where W is a matrix parameter that determines the strength of the pressure coupling and how
the box can be deformed. The equation of motion for the particles becomes:
2
2;i i i
i
d d
dt m dt
r rM
F T
1 T T 1( )d d
dt dt
h hM h h h h (2.26)
The combined use of the Nosé-Hoover thermostat and Parrinello-Rahman barostat can
provide realistic fluctuations in temperature and pressure when the thermodynamic properties
of a system are requested.
21
Chapter 3
Force Field Models
As shown in eq. (2.1), the force on a particle is defined as the negative gradient of a potential
energy function U with respect to the position of the particle. A force field refers to the form
and parameters of mathematical functions used to describe the potential energy of a system of
particles. There are several assumptions in the force field models: (1) Molecules are
represented by a “ball and spring” model, in which atoms are of different size and “softness”
and bonds are of different length and “stiffness”; (2) The potential energy function is
described as a sum of individual functions for bond stretching, angle bending, torsional
energies, and non-bonded interactions; (3) The transferability of a force field enables the
parameters derived from either experimental or high-level computational data of the small set
of models to be applied to a much larger set of related models. [66]
3.1 The Potential Energy Function Form
The force field energy is expressed as a sum of bonded interactions, non-bonded interactions
and sometimes cross-terms. It is an effective pair-wise potential parameterized in such a way
that a significant proportion of the many-body effect can be implicitly incorporated through
parameterization. The energy of the bonded interactions is generally accounted for by bond
stretching (Vbond), angle bending (Vangle), dihedral angle torsion (Vtorsion), and improper
dihedral torsion (Vimproper) terms. The energy of the interactions between non-bonded atoms is
accounted for by van der Waals (VvdW) and electrostatic (Velec) terms. Cross-terms (Vcross) are
included in the second-generation force fields to achieve higher accuracy, describing the
coupling between the bonded terms.
Chapter 3 Force Field Models
22
FF bonded non-bonded crossV V V V
bonded bond angle torsion improperV V V V V
non-bonded vdW elecV V V
Figure 3.1 Schematic representations of the key components in the potential energy function. [66]
3.1.1 Bond Stretching
The simple harmonic potential is a reasonable approximation to the shape of the potential
energy curve near the minimum of the potential well for the bond stretching although it is less
accurate away from equilibrium. The functional form is:
2bond
1( ) ( )
2b
ij ij ij ijV r k r b (3.1)
The Morse potential is particularly useful for the description of potential energy curve of
a typical bond. It has an asymmetric potential well and a zero force at infinite distance. It is
capable of describing a wide range of behavior from equilibrium behavior to bond
dissociation. The functional form is:
2morse ( ) [1 ( ( ))]ij ij ij ij ijV r D exp r b (3.2)
with 2
ij
ij
k
D (3.3)
Here Dij is the well depth, bij is the reference bond length, and k is the stretching constant
of the bond. However, it is not so widely adopted in force fields since it is not so
computationally efficient to implement and for near equilibrium situations it is good enough
to use the harmonic approximations.
Chapter 3 Force Field Models
23
3.1.2 Angle Bending
The deviation of angles from their reference values can also be approximated by a harmonic
potential on the angle θijk:
0 2angle
1( ) ( )
2ijk ijk ijk ijkV k (3.4)
Sometimes the Urey-Bradley angle potential is used, where an additional harmonic
correction term on the distance between the end atoms i and k is added to the harmonic
potential on the angle θijk:
0 2 0 2angle
1 1( ) ( ) ( )
2 2UB
ijk ijk ijk ijk ijk ik ikV k k r r (3.5)
3.1.3 Proper Torsional Terms
The proper torsional angle potential function models the presence of steric barriers between
atoms separated by three covalent bonds. The motion associated with this term is a rotation
around the middle bond. This potential is periodic in a 360° rotation and is most commonly
expressed as a cosine series expansion. The torsional potential can take the periodic form of:
torsion ( ) [1 cos( )]2
nijkl ijkl n
n
VV n (3.6)
Here in each term Vn is the torsional rotation force constant, which indicates qualitatively
the relative barrier. n is the multiplicity. ϕn is the phase angle usually chosen so that terms
with positive Vn have minima at 180°.
3.1.4 Improper Torsional Terms
An improper torsion angle is a “virtual” torsion angle in which the four atoms are not bonded
in the sequence i-j-k-l. Improper torsional potentials are introduced to keep planar groups
planar or to maintain stereochemistry at chiral centers. A stiff harmonic improper term in the
angle χ or in the distance d (the equilibrium angle or distance for a planar structure is zero) is
usually employed:
Chapter 3 Force Field Models
24
2improper
1( )
2V k or 2
improper
1( )
2V d kd
(3.7)
3.1.5 Van der Waals Interaction
The essential feature of the van der Waals interaction is that the interaction energy is zero at
infinite inter-atomic distance rij; as rij is reduced, the energy decreases and passes a minimum;
then the energy rapidly goes up as rij decreases further. The van der Waals interaction can be
decomposed into two parts: the long-range dispersive interaction and the short-range
repulsive interaction. The dispersive force arises from the induced dipole-dipole interaction
due to the fluctuations in the electron clouds and the dispersion interaction approximately
varies as –1/rij6. The repulsive force is attributed to the repulsion between the electrons in the
internuclear region due to the Pauli principle. [66] There are different potential forms for the
description of the repulsion part.
The Lennard-Jones 12-6 function is the most widely used potential function for
describing van der Waals interaction.
12 6
LJ ( ) 4 ij ij
ij ij
ij ij
V rr r
(3.8)
where σ is the collision diameter (the separation for which the energy is zero) and ε the well
depth. The Lorentz-Berthelot mixing rules are applied for the evaluation of the cross-terms in
the non-bonded Lennard-Jones interactions:
1( )
2ij ii jj
ij ii jj
(3.9)
The twelfth power term is found to be quite reasonable for rare gases but is rather too
steep for other systems such us hydrocarbons. Nevertheless the 12-6 function is widely used
due to computational efficiency reason.
The Buckingham potential has a more flexible and realistic exponential function for the
repulsive term than the 12-6 potential.
Chapter 3 Force Field Models
25
bh 6( ) exp( ) ij
ij ij ij ij
ij
CV r A B r
r (3.10)
The combination rules for the Buckingham potential are:
1( )
2
ij ii jj
ij ii jj
ij ii jj
A A A
B B B
C C C
(3.11)
3.1.6 Electrostatic Interaction
The electrostatic interactions between atom pairs can be modeled by assigning partial atomic
charges to each atom or assigning a bond dipole moment to each bond. The atomic charge
model is easier to parameterize and is more commonly adopted in almost all force fields. The
interaction between point charges is given by Coulomb’s law:
elec
0
( )4
i j
ij
ij
q qV r
r (3.12)
3.2 Force Field Parameterization
Once the potential energy function form is determined, the parameters in the potential energy
function need to be derived. These parameters can be determined from fitting against
experimental data or quantum mechanics calculations. The use of quantum mechanics results
allows the determination of force field parameters for molecules where little or no
experimental data exist. The well-established macromolecule force fields include AMBER, [67,
68, 69] CHARMM, [70, 71] GROMOS [72, 73] and OPLS-AA.[74] The general AMBER force field [75] is compatible with AMBER and is designed for small organic molecules and drug
molecules. Each force field has its own protocol for parameter development.
Parameters for bond stretching and angle bending can be treated separately from the
others. [66] The reference bond distances and angles can be determined from crystallography.
The bond and angle force constants can be obtained from vibrational spectra or from quantum
Chapter 3 Force Field Models
26
mechanics calculations. These parameters can often be transferred from one force field to
another without modification with little effect on the results.
In contrast, the performance of a force field is quite sensitive to the parameters for
non-bonded and torsional contributions. These parameters are closely coupled and can
significantly affect each other. [66] Most force fields include non-bonded interactions for atom
pairs that are separated by three bonds or more. But the 1-4 interactions between atom pairs of
three bonds apart are in many cases scaled by a scaling factor between 0 and 1. For example,
in the AMBER force field, the scaling factors for the 1-4 electrostatic interaction and the 1-4
van der Waals interaction are 0.8333 and 0.5, respectively. In the GLYCAM force field, the
scaling factors are set to be unity. The use of scaling factors is for the purpose of
transferability of the force field parameters, especially the torsional terms. In the AMBER
force field, the torsional parameters depend solely on the atom types of the two atoms that
form the central bond, while in the GLYCAM force field, all atomic sequences have an
explicit defined set of torsional terms.
Van der Waals parameters are often obtained by fitting to experimental liquid properties
such as density, or fitting to the intermolecular interaction energy. The method of Monte Carlo
simulations on pure liquids to reproduce the experimental thermodynamic quantities proposed
by Jorgensen et al. [74] is used in both OPLS-AA and AMBER force field for deriving the van
der Waals parameters.
The atomic charges can be fitted to experimental liquid properties, X-ray crystallographic
electron density, or be fitted to electrostatic potential (ESP) calculated by quantum mechanics
methods. Within the ESP charge scheme, the ESP ϕesp arises from the nuclear charges and the
electron density in the form of:
esp
( ')( ) d '
| | '
Nnuca
a a
Z
rr r
R r | r - r | (3.13)
The fitting of the atomic charges Qa to the potential ϕesp is carried out by minimizing the
following error function under the constraint that the sum of the atomic charges Qa equals the
total molecular charge.
2points
esp
( )ErrF( ) ( )
| |
N Natomsa a
r a a
Rr
R r (3.14)
However, the ESP points used in fitting the charges must lie outside the van der Waals
Chapter 3 Force Field Models
27
surface of the molecule. [76] Hence “buried” charges (e.g. an sp3 carbon) tend to be poorly
determined because even the closest surface points at which the ESP is evaluated are
relatively far away. This problem has been resolved to some extent in the Restrained
ElectroStatic Potential (RESP) fitting scheme, [77] in which a penalty function is added in the
form of restraints on non-hydrogen atomic charges to target charges. The RESP scheme is
used in the AMBER force field. A more accurate modeling of the many-body effects requires
inclusion of polarizable effects. The average many-body effect is often approximately
accounted for by using a set of empirical charges which often correspond to a molecular
dipole moment 10–20% larger than that for the isolate molecule, or by fitting to e.g.
Hartree–Fock results with for instance 6-31G(d) basis set that are known to overestimate the
polarity.
Finally, the torsional angle parameters are usually fitted to the energy curve determined
from quantum chemical calculations for rotations around a bond, in combination with the van
der Waals parameters and partial charges. [66]
It is worth noting that each force field is devised under specific conditions and is
optimized for specific kind of molecules, such as biomolecules (nucleic acid, protein, lipid,
and carbohydrate), organic molecules, or inorganic molecules. It is crucial to choose an
appropriate force field for the studied system in order to get reasonable and reliable results
from MD simulations.
3.3 Force Fields for Carbohydrates
The inherent flexibility of carbohydrates, combined with the multitude of possibilities for
linkage and functionality, provides a unique challenging endeavor in the development of
carbohydrate force fields. [78] The growing interest in structural glycobiology has stimulated
the need to develop protein-consistent carbohydrate force fields. The above mentioned
macromolecular force fields AMBER, CHARMM, GROMOS and OPLS-AA usually can not
accurately reproduce the carbohydrate conformation, such as the experimental solution-phase
rotamer populations for the glycosidic linkages as well as the ω-angle. Therefore, special
carbohydrate parameter sets have been implemented into these macromolecular force fields,
including the GLYCAM series (complementary to AMBER), [79, 80, 81] and carbohydrate-tuned
versions of CHARMM, [82, 83, 84] GROMOS-45a4 [85] and OPLS-AA-ESI [86] force fields. The
current state of carbohydrate force fields has been reviewed by R. J. Woods. [87]
29
Chapter 4
Potential of Mean Force
4.1 Definition of Potential of Mean Force
Free energy is the underlying driving force behind a physical or chemical process of a system.
The free energy calculation is thus of supreme importance for our exploration of the
microscopic origin of the measurable macroscopic quantities. In statistic mechanics, for an
NPT ensemble, the Gibbs free energy G is related to the partition function ZN by:
1( , , ) ln ( , ) ln ( , )B N NG T P N k T Z T P Z T P
(4.1)
with β=1/kBT. The partition function of the system writes:
H( , )( , )N NN N
NZ T P d d e p xx p (4.2)
where H is the classical Hamiltonian for a system of N interacting particles. H is composed of
the kinetic energy and the potential energy. The kinetic energy depends on the momentum pi
of the particles, and the potential energy is a function of the positions of particles xN.
2
1
( , ) ( )2
NN N Ni
i i
Vm
p
p x x (4.3)
However, accurate calculation of the absolute free energy via eq. (4.1) is not possible for
a system consisting of thousands of atoms due to insufficient sampling during finite length
Chapter 4 Potential of Mean Force
30
simulations. The problem becomes much simplified if only the free energy difference is
required. For many practical applications, the change in free energy as a function of a reaction
coordinate, named potential of mean force (PMF), is of particular interest. The free energy of
the system for which a given observable ξ(x,p) is constrained to a generalized reaction
coordinate ξ is called constrained free energy. [88] A PMF is a constrained free energy. The
constrained free energy evaluates the impact of a subset of parameters on the free energy. The
constrained partition function Z(ξ) and constrained free energy G(ξ) are given by:
H( , )( ) ( ( , )- )N NN NZ d d e
p xx p x p (4.4)
1( ) ln ( )G Z
(4.5)
The probability (ρ(ξ)) to find the system in a state described by ξ is given by the ratio of
Z(ξ) and ZN. In this thesis, only when the reaction coordinate is a function of position, such as
the separation between two sets of particles, is considered. The ρ(ξ) thus takes the form of:
( )
( )
( ( )- )( )( )
N
N
N V
N VN
d eZ
Z d e
x
x
x x
x (4.6)
Then the free energy difference between two constrained states can be written as:
1 ( ) 1 ( )ln ln
( ') ( ')
ZG
Z
(4.7)
It can be seen that the relative probability of finding a particle changes exponentially with
the barrier height. In practice, when the barrier is as high as several multiples of kBT, the
distribution function ρ(ξ) is difficult to converge in a standard MD simulation within
reasonable computational time due to insufficient sampling. Several methods have been
proposed in order to overcome the insufficient sampling problem by guiding the system along
a predefined reaction coordinate, including the non-Boltzmann sampling techniques such as
umbrella sampling, [89] and steered MD (SMD) simulations. [90]
Chapter 4 Potential of Mean Force
31
4.2 Methods for Potential of Mean Force Calculations
4.2.1 Umbrella Sampling
The method of umbrella sampling was proposed by Torrie and Valleau. [89] The essential idea
of this method is to overcome the sampling problem by modifying the potential function with
a restraining potential so that the unfavorable states in the high-energy regions can be sampled
adequately. The external restraining potential is designed such that its effect on the underlying
free energy can be removed.
The restraining potential in the modified potential function can be seen as a perturbation
to the original one:
'( ) ( ) ( )N NiV V x x (4.8)
Usually, we use the restraining potential that is a function of coordinates only. The
external restraining potential ωi(ξ) usually takes a quadratic form:
21( ) ( )
2i ik (4.9)
Figure 4.1 Schematic representation of umbrella sampling technique.
The basic principle of umbrella sampling is illustrated in Figure 4.1. The reaction
coordinate is divided into a number of simulation windows with reference positions at ξi. The
restraining potentials are then chosen in such a manner that the sampling distribution is
Chapter 4 Potential of Mean Force
32
shifted along the reaction coordinate. The width of the simulation window is controlled by the
magnitude of the force constant k in the restraining potential. Sufficient overlap should be
guaranteed between the sampling distributions of adjacent simulation windows in order to
derive a continuous energy function from these simulations. The restraining potential should
not be too large either; otherwise the results may take too long time to converge.
Referring to eq. (4.6), the probability to find the particle at position ξ due to the
restraining potential ωi(ξ) in the ith window is given by:
( ( ) ( ))
( ( ) ( ))
( ( )- )( )
Ni
Ni
VN
biasi VN
d e
d e
x
x
x x
x
( )
( )( )
i
ii
e
e
(4.10)
Let Wbias(ξ) and W(ξ) denote the PMF associated with ξ when the simulation is
performed with or without the restraining potential, respectively, then we get:
( )
( )
( )1 1W( ) W ( ) ln ln
( )
i
i
bias ibiasi
e
e
(4.11)
Fi is named free energy constant associated with the biasing potential and is defined as:
( ) ( ) ( )i i iFe e d e (4.12)
where ρ(ξ) is the overall unbiased distribution function. Then we get:
W( ) W ( ) ( )biasi iF (4.13)
In order to get W(ξ), the unknown Fi needs to be resolved. To minimize the statistical
error, the Weighted Histogram Analysis Method (WHAM) [91] is widely used to determine Fi.
The WHAM equations express the optimal estimate for the unbiased distribution function ρ(ξ)
as a weighted sum over the individual unbiased distribution functions ρi(ξ). The following
WHAM equations are solved iteratively until self-consistency is achieved.
Chapter 4 Potential of Mean Force
33
( ( ) )
( ( ) ) ( ( ) )1 1
1 1
( ) ( ) ( )i i
i i i i
FN Nbiasi i
i iN NF Fi ij jj j
n e n
n e n e
( ) ( )i iFe d e
(4.14)
where N is the number of biased window simulations and ni is the number of counts in the ith
window.
4.2.2 Steered Molecular Dynamics Simulations
Figure 4.2 The schematic figure of SMD principle. The green (left) atom represents the SMD atom and the
red (right) atom represents the dummy atom. They are connected by a virtual harmonic spring.
SMD simulations apply external steering forces in the desired direction to accelerate
processes that otherwise are too slow due to energy barriers. SMD simulation is a technique
mimicking the principle of the atomic force microscopy (AFM). There are two types of SMD
simulations. In the first protocol, named constant force SMD, a constant force is applied to
one or more atoms; displacement is recorded throughout dynamics. In the second protocol,
called constant velocity SMD, a harmonic potential (spring) is used to induce motion along a
reaction coordinate (see Figure 4.2). The free end of the spring is moved at constant velocity,
while the atoms attached to the other end of the spring (SMD atoms) are subject to the
steering force. The force is determined by the extension of the spring and is calculated as:
Chapter 4 Potential of Mean Force
34
( ) ( ) ( )i i i
t U t U tx y z
F i j k
2
0
1( ) ( )
2U t k vt r r n
(4.15)
where r and r0 are the instantaneous position and the initial position of the SMD atoms, n is
the direction of pulling and k the force constant for the virtual spring.
The calculation of free energy from SMD simulations is based on multiple
non-equilibrium MD simulations. The Jarzynski’s equality [92] is typically applied to extract
equilibrium free energy difference from the non-equilibrium work distribution:
G We e (4.16)
where < > denotes an ensemble average of the work W from non-equilibrium simulations.
However, the average of exponential work appearing in Jarzynski’s equality is dominated by
the small work values which are very rare. Appropriate sampling of such rare trajectories is
thus required for an accurate estimate of free energy. Hence, Jarzynski’s equality is considered
practically applicable only for slow processes of which the fluctuation of work is comparable
to the temperature. Yet, in practice, a pulling speed of several orders of magnitude higher than
the quasi-static speed is employed in the simulation and only a small number of trajectories
can be sampled due to limited computing capability. To overcome this difficulty, approximate
formulas via the cumulant expansion of We were proved to be effective. [93, 94, 95]
2 32 32 3 2ln 3 2 ......
2! 3!We W W W W W W W
(4.17)
If the distribution of work W is Gaussian, the third and higher cumulants are identically
zero. [96] It has been shown that with a limited number of trajectories, the second order
cumulant expansion yields the most accurate estimate. [90]
22
2G W W W
2
2WW
(4.18)
Chapter 4 Potential of Mean Force
35
where σW is the standard deviation of the work.
The proper choice of the force constant k in eq. (4.15) is important in SMD simulations.
The force constant should be chosen to be large enough to ensure that the reaction coordinate
closely follows the constraint position. But too large a k is also not allowed since PMF
calculated with a larger k shows larger fluctuation which scales as Bkk T .[97]
37
Chapter 5
Summary of Papers
5.1 Thermal Response of Crystalline Cellulose Iβ (Paper I)
Thermal behavior is one of the most fundamental properties of solid-state materials. Progress
in investigations of the thermal behavior of neat cellulose Iβ has been made by both
experimental [20-23] and theoretical [98] approaches. Atomic MD simulations of the thermal
response in crystalline cellulose Iβ was initiated by Bergenstråhle et al. [98] with the
GROMOS 45a4 force field. However, a relatively large discrepancy in the predicted unit cell
parameters a and γ compared to the crystallographic data was found, which might lead to a
controversial prediction on the thermal transition mechanism. The limited capability to
describe the van der Waals and intersheet coulombic interactions between the united CH
atoms and oxygen by the GROMOS 45a4 force field might be the reason.
Figure 5.1 Projection of the ab base plane of the crystal model for cellulose Iβ. Reprinted with permission
from Springer.
Chapter 5 Summary of Papers
38
The reported good performance of the GLYCAM06 force field [99]motivated us to apply
the force field to explore the atomic details regarding the thermal response of cellulose Iβ in
paper I. The crystal model for cellulose Iβ (Figure 5.1), built from the coordinates reported by
Nishiyama et al., [10] is composed of 6 × 6 chains. Each chain was composed of eight glucose
units and was covalently bonded to its own periodic image across the simulation boxes so that
infinite cellulose chains were modeled. Simulations were performed at seven different
temperatures (298, 350, 400, 450, 475, 500 and 550 K) to investigate the thermal response
behavior of crystalline cellulose Iβ.
5.1.1 Low Temperature Structure
At first, the structural properties for cellulose Iβ at 298 K were evaluated and compared to the
crystallographic data (Table 5.1, Figure 5.2). The predicted density was 1603.4 kg/m3 at room
temperature, which is about 2.0% larger than the experimental value (1636 kg/m3) [10]
determined from the diffraction data. The values for the unit cell axes of a (intersheet), b
(interchain) and c (chain direction) at 298 K were calculated to be 0.763, 0.823 and 1.080 nm,
respectively. The predicted unit cell angles for α, β and γ were 89.99, 89.99 and 97.17°,
respectively. The deviations of these calculated unit cell parameters from the crystallographic
data were quite small, with the largest deviation (4.0%) occurring for c.
Table 5.1 Predicted unit cell parameters and experimental data
Unit cell axial length (nm) Unit cell axial angle (°)
a b c α β γ
298 K (this work) 0.763 0.823 1.080 89.99 89.99 97.17
500 K (this work) 0.811 0.828 1.078 90.00 89.96 98.33
Exp. room temperature[10] 0.778 0.820 1.038 90.0 90.0 96.5
Exp. 523K [21] 0.819 0.818 1.037 90.0 90.0 96.4
Figure 5.2 Deviations for the calculated unit cell parameters with respect to crystallographic data.
Chapter 5 Summary of Papers
39
An extensive comparison with the simulation data from other reports [98, 100, 101, 102] (see
Table 1 in paper I) allows an evaluation of different force fields. A remarkable advantage of
our results is manifested by the fact that the GLYCAM06 predicts the most accurate a and b
values among all the tested force fields. The a and b axes, corresponding to the inter-sheet and
inter-chain distances in cellulose Iβ, respectively, rely heavily on intermolecular non-bonded
interactions. The success of the GLYCAM06 in reproducing the a and b values should be
attributed to its partial atomic charge derivation scheme of fitting the partial atomic charges to
the quantum mechanical molecular ESP, especially the implementation of ensemble-averaged
charge sets, which has the advantage of reproducing intermolecular interaction energies. In
addition, the GLYCAM06 yields unit cell angles closest to the experimental values compared
to the other force fields. On the other hand, the slightly overestimated c axis by the
GLYCAM06 probably results from the bonded force field parameters, i.e. the reference values
for the bond stretching and angle bending parameters derived from gas phase quantum
mechanical calculations on small molecular fragments; the torsional terms derived from
density functional theory B3LYP/6-311++G(2d, 2p) calculations.
Another important structural characteristic of crystalline cellulose Iβ is its stabilizing
hydrogen bonding networks. The hydrogen bonds were analyzed based on geometric criteria.
At 298 K, two intrachain hydrogen bonds O2H2... O6, O3H3...O5 and one interchain (but
intrasheet) hydrogen bond O6H6...O3 were systematically detected on both the origin and
center chains (Figures 5.3, 5.4), which was consistent with the dominant hydrogen bonding
network determined experimentally. [10] On the whole, the structural properties of cellulose Iβ
at room temperature are satisfactorily reproduced by the GLYCAM06 force field.
5.1.2 Transition
The structure for cellulose Iβ was explored at different temperatures by the GLYCAM06 force
field. A sudden drop in density is an indication of a phase transition occurring between 475
and 500 K. In Figure 5.3, we can see that the a axis expanded slowly from 0.763 nm at 298 K
to 0.777 nm at 475 K, corresponding to a 1.9% expansion at 475 K from room temperature.
At 500 K, the a axis experienced a sudden increase of a 6.3% expansion from 298 to 500 K
(0.811 nm). On the other hand, the b axis showed anisotropic expansion by only about 0.6%
from 0.823 nm at 298 K to 0.828 nm at 500 K. On the contrary, the c axis showed slight
shrinkage at higher temperature. The monoclinic angle γ displayed a small variation in the
range 97.11–98.33° in our simulation. It is noteworthy that there were only small shifts
observed in the α and β angles during the whole heating process, indicating no chain sliding
Chapter 5 Summary of Papers
40
effect during the transition, which is in strong contrast with the previous simulation result. [98]
Figure 5.3 Unit cell axes a, b, c, unit cell angles α, β, γ, and density of cellulose Iβ as a function of
temperature.
5.1.3 High Temperature Structure
It has been revealed by X-ray experiments [20] and temperature-dependent IR measurements [22], [23] that the cellulose crystal will change into a high-temperature phase upon heating.
Actually, Wada et al. [21] have recently confirmed by synchrotron X-ray fiber diffraction that
the high-temperature phase had a staggered half of a glucose residue along the chain axis
between the sheets, similar to the structure of cellulose Iβ, which indicates that no
longitudinal chain slippage is involved during the transition. The unit cell of the cellulose
high-temperature phase at 523 K was determined to have a two-chain monoclinic structure
with dimensions of a = 0.819 nm, b = 0.818 nm, c = 1.037 nm, and γ = 96.4°.[21]
Chapter 5 Summary of Papers
41
Figure 5.4 (left) The low- and high-temperature structures for cellulose Iβ. Origin and center chains are
colored in red and blue, respectively (right). The corresponding hydrogen-bonding network for the low-
and high-temperature structures.
The high-temperature structure of cellulose Iβ predicted by the GLYCAM06 force field is
shown in Table 5.1 and Figure 5.4. The unit cell parameters calculated at 500 K are in
reasonable agreement with experiment. [21] At 500 K, the conformation of all hydroxymethyl
groups (ω O5-C5-C6-O6, defined in Figure 1.2) changed from tg (ω=180°) to either gt (ω=60°)
on origin chains or gg (ω=300°) on center chains. Meanwhile, change of the conformation of
hydroxyl groups also took place, especially for the hydroxyl groups at C2 and C6 positions. In
addition, in the high-temperature structure, all the origin chains rotated around the helix axis
by about 30°.
The change in the orientation of the hydroxymethyl and hydroxyl groups led to the
reorganization of the hydrogen-bonding network (Figure 5.4). At 500 K, the O2H2…O6
intrachain hydrogen bond completely disappeared and was substituted by the weaker
O2H2…O6’ interchain hydrogen bond. The number of the O3H3…O5 intrachain hydrogen
bonds in the center chains decreased by about 20%. The O6H6…O3’ interchain hydrogen
bonds collapsed completely at 500 K in the center chains and decreased to 28% of the original
population in the origin chains. Two new intersheet hydrogen bonds (Figure 5.5) emerged in
the high-temperature structure, namely O6H6…O2’, where only the primary OH groups at the
C6 position in the center chains act as donors, and O6H6…O4’, where 40% of the primary
OH groups at the C6 position in the origin chains act as donors.
Chapter 5 Summary of Papers
42
Figure 5.5 A representative example of the intersheet H-bonds in cellulose I at 500 K. Reprinted with
permission from Springer.
In summary, compared to the previously employed GROMOS 45a4 force field,
GLYCAM06 yields data in much better agreement with experimental observations, which
reflects that a cautious parameterization of the non-bonded interaction terms in a force field is
critical for the correct prediction of the thermal response in cellulose crystals.
5.2 Adsorption of Xyloglucan on Cellulose (Paper II)
In paper II, we studied the adsorption of XGs on the surface of cellulose, in particular the
effects of explicit water and side chain variation. The adsorption properties were studied in
detail for three XGs on cellulose Iβ 1-10 surface in aqueous environment, namely
GXXXGXXXG, GXXLGXXXG and GXXFGXXXG, which differed in the length and
composition of one side chain.
5.2.1 Interactions Between XGs and Cellulose Iβ 1-10 Surface
The instantaneous interactions between XGs and cellulose Iβ 1-10 surface were analyzed after
the equilibration states were reached. In Figure 5.6, the total instantaneous interaction energy
(Eint) between XGs and cellulose Iβ 1-10 surface with different starting orientations, the
electrostatic (Eelec) and van der Waals (EvdW) contributions, along with the standard deviations,
are depicted. The results showed that each XG was able to adsorb onto the cellulose surface
regardless of the orientation; the adsorption of XGs on cellulose was stabilized by both Eelec
and EvdW; in addition, both the backbone and side chain residues interacted with the cellulose
surface. These characteristics are in agreement with the results of Hanus et al., [35] where the
XG-cellulose interaction was modeled in vacuo. In our model, when water molecules were
included, the interaction energy between XG and cellulose averaged over all the orientations
was -126.1, -123.0, and -117.8 kcal/mol for GXXXGXXXG, GXXLGXXXG and
GXXFGXXXG, respectively, indicating a small impact on the interaction energy by variation
Chapter 5 Summary of Papers
43
in one side chain in our model. When the interaction energy per repeat unit was compared, it
was found that it was considerably lower in our solvated model than in Hanus et al.’s [35]
vacuo model. Moreover, a slightly prevailing role of EvdW in the total interaction energy
between cellulose and XG was revealed in our model, opposed to the previous conclusion [35]
on a 75% contribution of Eelec.
Figure 5.6 The electrostatic interaction energy Eelec, and the van der Waals interaction energy EvdW between
GXXXGXXXG (A), GXXLGXXXG (B) and GXXFGXXXG (C) and cellulose Iβ 1-10 surface with
different starting orientations, respectively, averaged over the last 5 ns for each run.
In order to investigate the effect of water, the water molecules in the solvated systems
with the largest interactions (GXXXGXXXG (A-[180°]), GXXLGXXXG (A-[0°]) and
GXXFGXXXG (A-[270°])) were removed and simulations were continued on these
desolvated systems. The interaction energies in vacuo were also found to be considerably
higher within the same force field, especially the electrostatic component Eelec. One factor that
contributes greatly to the higher Eint in vacuo is the larger number of hydrogen bonds formed
between XG and cellulose in vacuo than in water. By observation, in the solvated model, there
were several water molecules between XG and cellulose. Meanwhile, water mediated
hydrogen bonds were also present. Due to the absence of opportunities for hydrogen bonding
to water molecules, XG moved closer to the cellulose surface in vacuo accounting for the
larger interaction. Therefore, when interaction energies are considered, it is important to take
water molecules into account in the theoretical modeling of this system.
5.2.2 Structural Properties
The equilibrium structures of GXXXGXXXG (A-[180°]), GXXLGXXXG (A-[0°]) and
GXXFGXXXG (A-[270°]) adsorbed on the cellulose Iβ 1-10 surface, which are of the largest
interaction energies for each XG, are depicted in Figure 5.7. Two important structural
parameters are the conformation of the glycosidic bonds in the backbone and that of the side
chains. Analysis of the conformation around the glycosidic bonds revealed that for
Chapter 5 Summary of Papers
44
GXXXGXXXG, all the glycosidic bonds took ‘twist’ conformations, while for
GXXLGXXXG and GXXFGXXXG, half of them took ‘flat’ conformations. Thus, extension
of XG side chains with galactosyl and fucosyl residues appeared to favor flattened
conformations of the XG backbone upon adsorption. Analysis of the branching side chains
orientation relative to the main chain showed that all the side chains in GXXXGXXXG were
closer to the gg conformation, while for GXXLGXXXG and GXXFGXXXG, the side chains
in the vicinity of the galactosyl and fucosyl residues tended to assume the tg or gt
conformations, facilitating extended conformations of the di- and trisaccharide side chains,
which leads to favorable interactions of both backbone and side chain residues with the
cellulose surface.
Figure 5.7 The equilibrium structures of GXXXGXXXG (A-[180°]), GXXLGXXXG (A-[0°]) and
GXXFGXXXG (A-[270°]) on cellulose Iβ 1-10 surface. Water molecules are omitted for clarity. Glucosyls
are indicated in magenta, xylosyls in cyan, galactosyl in yellow and fucosyl in green.
5.2.3 Binding Affinity Evaluated by SMD Simulations
To evaluate the binding affinity of the cellulose surface to XGs, free energy profiles for the
desorption of each XG from the surface were derived from the SMD simulations via
Jarzynski’s equality. SMD simulations were performed for each XG-cellulose system with the
Chapter 5 Summary of Papers
45
largest interaction energy (GXXXGXXXG (A-[180°]), GXXLGXXXG (A-[0°]) and
GXXFGXXXG (A-[270°])) to generate representative results.
Figure 5.8 Free energy profiles for GXXXGXXXG (A), GXXLGXXXG (B) and GXXFGXXXG (C)
desorbing from the cellulose Iβ 1-10 surface over 25 ns of the SMD trajectory during which the pulling
distance of the pulled atom was increased by 2.5 nm at a constant rate of 0.1 nm/ns.
The calculated PMFs are depicted in Figure 5.8. The free energy difference ΔG is
calculated to be -31.4, -34.5 and -30.3 kcal/mol upon adsorption, corresponding to -3.5, -3.8,
and -3.4 kcal/mol/glucose, for GXXXGXXXG, GXXLGXXXG and GXXFGXXXG,
respectively. The calculated results fall well into the range of the experimental average
desorption energy of 1.82-5.12 kcal/mol/glucose determined from AFM measurements, [103]
which also to some extent justifies our SMD simulations. The free energy differences
demonstrate that the galactosyl and fucosyl substitutions on one side chain of XG does not
have significant effect on their binding affinity toward adsorption on cellulose. Nevertheless,
the characteristics of the three PMFs are different, indicating possibly different desorption
mechanisms for these XGs.
Therefore, our results addressed the importance of taking into account the water
environment in the theoretical model of the XG-cellulose assembly. In our model, variation in
one side chain did not have significant influence on the interaction energy between XGs and
cellulose and the binding affinity of XGs. However, the side chain variation did affect the
equilibrium structural properties of the adsorbed XGs.
5.3 Carbohydrate-Enzyme Interaction (Paper III)
In paper III, we have investigated by MD simulations tetradecasaccharide XXXGXXXG in
complex with hybrid aspen xyloglucan endo-transglycosylase PttXET16-34, with the
“nucleophile helper” residue, aspartate 87 (Asp87), either in protonated (conjugate acid) form
or deprotonated (conjugate basic) form. The “helper” residue, Asp87, lies in between and on
the same face of strand as the catalytic residues glutamate 85 (Glu85) and glutamate 89
Chapter 5 Summary of Papers
46
(Glu89).
Two separate simulations were carried out. The sidechain of the Glu89, which functions
as a catalytic acid residue to deliver a proton to the nucleofuge in the first chemical step of the
canonical retaining mechanism employed by GH16 enzymes, was modeled in the protonated
form in both simulations. Meanwhile, Glu85, which functions as the catalytic nucleophile,
was modeled in the deprotonated form. The “helper” residue Asp87 was alternately modeled
in either deprotonated (Simulation 1) or protonated form (Simulation 2).
Figure 5.9 Schematic representation of interactions in the active site cleft of the hybrid aspen xyloglucan
endo-transglycosylase PttXET16-34 with XXXGXXXG. Subsites are numbered from the scissile
glycosidic bond according to Davies, Wilson, and Henrissat. Reprinted with permission from Taylar &
Francis.
By and large, similar global conformations of the relaxed XXXGXXXG in the active site
cleft were yielded in these two simulations. Both trajectories suggest that the PttXET16-34
active site cleft consists of seven primary backbone Glc-binding subsites. Three positive
subsites extend toward the reducing end of the substrate from the catalytic machinery on
strand β6, and four negative subsites extend toward the non-reducing terminus (Figure 5.9). It
is clearly shown that there are many hydrogen bonds and van der Waals interactions involved
in the substrate recognition. The multiple interactions in the active site encourage
XXXGXXXG to adopt a bent conformation. These interactions have been analyzed as a
function of the time in the MD trajectories (see Tables 2 and 3 in Paper III). In addition,
general trend was indicated that a correlation might exist between the number of interactions
Chapter 5 Summary of Papers
47
and the low mobility of certain glycosyl residues with the contributions of these residues to
transition state binding and catalysis.
Careful analysis of the puckering angles in Glc(-1) revealed that in both simulations, the
undecorated Glc(-1) backbone glycosyl unit immediately flipped from the 4C1 chair starting
conformation into various boat and skew boat conformations (for example 1,4B, B2,5, B3,6, 1S3
and 1S5) after energy minimization and equilibration when all harmonic constraints were
removed. These boat-like conformations of Glc(-1) showed inter-conversions in both MD
trajectories, which are necessary pre-transition state conformations allowing access of the
catalytic nucleophile to C1 of the sugar ring Glc(-1).
Figure 5.10 Close-up of tetradecasaccharide XXXGXXXG in complex with PttXET16-34 from time steps
late in Simulations 1 and 2. Reprinted with permission from Taylar & Francis.
Although both simulations yielded similar global conformations of the relaxed
XXXGXXXG in the active site cleft, the charge state of Asp87 had a distinct influence on the
structure of the PttXET16-34 active site (Figure 5.10). To give a clear picture of the
difference in the active site, some critical distances at the catalytic site were monitored in the
last 10 ns in both simulations (Table 5.2).
When Asp87 was protonated in Simulation 2, a stable hydrogen bond was formed with
the catalytic nucleophile, Glu85. In addition, the 2-OH group of Glc(-1) was also found to be
hydrogen bonded to the nucleophile Glu85. Thereby, the Glu85 was precisely poised to
receive the anomeric carbon of Glc(-1) undergoing electrophilic migration from the departing
O4 atom of Glc(+1), leading to the formation of a covalent glycosyl-enzyme. However, in
Simulation 1, when Asp87 was deprotonated, the electrostatic repulsive interaction between
Asp87 and Glu85 pushed the catalytic nucleophile away from C1 of Glc(-1).
Chapter 5 Summary of Papers
48
Table 5.2 Inter-atomic distances (nm) and hydrogen bonds averaged in the last 10 ns of the two trajectories
of XXXGXXXG in complex with PttXET16-34. Standard deviations are given in parentheses.
Simulation 1 Simulation 2
Glu85-Oε2 and C1-Glc(-1) 0.64(0.04) 0.36(0.02)
Glu85-Oε2 and Asp87-Oε1 0.63(0.04) 0.27(0.01) 99.8a
Glu89-Oε1 and O4-Glc(+1) 0.36(0.03) 0.38(0.04) 11.9a
Glu89-Oε1 and Asp87-Oε2 0.26(0.01) 98.0a 0.46(0.06)
aThe occupation time of the hydrogen bond (%) in the analyzed part of the trajectory.
On the other hand, the electrophilicity of the catalytic Glu89 was also affected by the
protonation state of Asp87. In Simulation 2, the OH group in Glu89 was found to occasionally
form hydrogen bond to the glycosidic oxygen atom of Glc(-1), which is catalytic productive.
On the contrary, in Simulation 1, the OH group in Glu89 was stably hydrogen bonded to the
carboxylate oxygen in Asp87, which is not favorable for the catalysis.
Our results thus suggest that catalysis in the glycoside hydrolase family 16 is optimal
when the “helper residue” Asp87 is protonated to align the nucleophile, the catalytic
nucleophile is deprotonated for nucleophilic attack on the substrate and general acid/base
residue is in the protonated form to deliver a proton to the departing sugar.
5.4 Working Mechanism of an α-cyclodextrin Based Rotaxane (Paper IV)
Molecular machines have recently attracted intense interest on account of their potential
application opportunities in the fields of nanoscience and materials science.[104 , 105, 106]
Redox-responsive bistable [2]rotaxanes and [2]catenanes are of considerable interest because
the redox-responsive switching process is usually rapid and precise and can be controlled
within solid-state devices, as well as on surfaces.
In paper IV, the binding preference and the working mechanism for a redox-responsive
bistable [2]rotaxane (Figure 5.11) incorporating an α-CD ring was investigated theoretically.
In this rotaxane, the α-CD ring can be induced to move between a tetrathiafulvalene (TTF)
recognition site and a triazole (TZ) unit on the dumbbell on account of the redox properties of
the TTF unit. When the TTF unit is in neutral state, the α-CD ring prefers to stay at the TTF
unit (denoted as the α-CD@TTF isomer); when the TTF unit is oxidized, the α-CD ring
leaves the TTF unit and binds with the TZ unit instead (denoted as the α-CD@TZ isomer).
Chapter 5 Summary of Papers
49
Figure 5.11 Structure of the rotaxane. Reprinted with permission from American Chemical Society.
5.4.1 Potential of Mean Force
The binding preference of the α-CD ring is determined by the relative free energy between the
two isomers, i.e. the free energy difference between the α-CD@TTF isomer and the
α-CD@TZ isomer. The PMFs for the shuttling movement of the α-CD ring along the
dumbbell were calculated via the umbrella sampling approach. The reaction coordinate was
defined as the z-component difference between the center of mass (COM) of the α-CD ring
and the reference atom (the carbon atom in the TTF unit indicated with asterisk mark in
Figure 5.11) of the dumbbell.
The initial calculations on the PMFs for the shuttling motion of the α-CD ring along the
dumbbell for different states were carried out in vacuo. As shown in Figure 5.12 (left), the
α-CD ring was predicted to encircle (1) the TTF unit, (2) the tetraethyleneglycol chain, and (3)
the TTF unit, respectively, for the neutral, oxidized +1 and oxidized +2 states. Apparently,
these PMFs in vacuo are inadequate for the correct binding preference prediction for both
oxidized states, for which the α-CD ring is observed experimentally to encircle the TZ station.
Chapter 5 Summary of Papers
50
Figure 5.12 The change of PMF as a function of the position of the α-CD ring along the dumbbell for
neutral, oxidized +1, and oxidized +2 states in vacuo (left) and water (right). Note: the hydrogen atoms are
omitted in the snapshots of energy minima for clarity. Reprinted with permission from American Chemical
Society.
Apparently, the water environment cannot be neglected in the theoretical model of this
rotaxane. Therefore we calculated PMFs (Figure 5.12 right) in explicit water for the shuttling
motion of the α-CD ring along the dumbbell for different states. For the neutral state, it is
obvious that the most stable isomer is α-CD@TTF (ring at 0.61 nm), which is energetically
more favorable by about 2.6 kJ/mol compared to the α-CD@TZ isomer (ring at 1.87 nm).
This free energy difference is comparable to the experimentally measured value [107] of 5.5 kJ/
mol. In contrast to the neutral state, the α-CD ring prefers to stay on the TZ station in both
oxidation states, in good agreement with experimental observations. In the oxidized +1 state,
the α-CD@TZ isomer is more stabilized than the corresponding α-CD@TTF+• isomer by
about 13.7 kJ/mol, while in the oxidized +2 state, the α-CD@TZ isomer is stabilized by about
48.0 kJ/mol. Accordingly, the free energy barrier for the ring shuttling from the TTF unit to
the tetraethyleneglycol chain decreases from 18.0 kJ/mol (neutral state) to 5.6 kJ/mol (+1
Chapter 5 Summary of Papers
51
state) and this barrier completely vanishes for the oxidized +2 state. This means that the
response time for this rotaxane relies totally on the oxidization of the TTF unit and the
α-CD@TZ isomers in the oxidized states are very stable. While the PMFs in vacuo cannot
explain the experimental observations, the PMFs in water are in good agreement with the
observed binding preference, indicating that it is important to take into account the water
environment in theoretical calculations.
5.4.2 Energy Decomposition and Working Mechanism
Figure 5.13 Variation of (left) the intermolecular interaction ECD-dumbbell, ECD-H2O, and Edumbbell-H2O, and
(right) Eelec(CD-dumbbell) and EvdW(CD-dumbbell) with the position of the α-CD ring on the dumbbell for the neutral
(black), oxidized +1 (blue), oxidized +2 (red) states in water. The positions where the α-CD ring locates at
the TTF (0.2 nm < z < 0.6 nm) and TZ (1.4 nm < z < 1.90 nm) units are marked in cyan. Energies are
shown in unit of kJ/mol. Reprinted with permission from American Chemical Society.
The trajectories were further analyzed in order to elucidate the energy contributions
accounting for the working mechanism. The crucial components analyzed include
non-bonded interactions between the α-CD ring and the dumbbell, between the α-CD ring and
water, as well as between the dumbbell and water, denoted by ECD-dumbbell, ECD-H2O, and
Edumbbell-H2O, respectively. It is interesting to find that the non-bonded interactions between the
Chapter 5 Summary of Papers
52
α-CD ring and the dumbbell are consistently attractive for all the three states. Additionally,
this attractive interaction ECD-dumbbell, in which electrostatic contribution dominates, becomes
stronger when the TTF unit is oxidized. Apparently, the working mechanism for this rotaxane
cannot be explained exclusively by ECD-dumbbell. Careful analysis on the non-bonded
interactions between the rotaxane subcomponents with water, ECD-H2O and Edumbbell-H2O,
reveals that both these interactions in the oxidized states favor the binding of the α-CD ring at
the TZ unit, which overwhelm the opposing effect of the attractive interaction ECD-dumbbell,
leading to the correct binding preference predicted for the oxidized states.
5.4.3 Quantum Chemical Calculation in Vacuo
Figure 5.14 (left) The total energy for the inclusion complexation of TTF-OCH3, [TTF-OCH3]+·,
[TTF-OCH3]2+ into the α-CD cavity calculated at the B3LYP/6-31G(d) level in vacuo. z denotes the
distance between center of mass of the guest and the α-CD ring. (right) Optimized geometries for the
inclusion complexes.
Quantum chemical calculations on the inclusion complexes between the α-CD ring and the
neutral/oxidized model guest molecules TTF-OCH3, [TTF-OCH3]+·, and [TTF-OCH3]
2+ were
also carried out. The fully optimized geometries for the complexes calculated at the
Chapter 5 Summary of Papers
53
B3LYP/6-31G (d) level in vacuo show that both radical cation [TTF-OCH3]+· and dication
[TTF-OCH3]2+ can form stable inclusion complexes with the α-CD ring. Single point energy
calculations at the B3LYP/6-311++G(d, p) level for these fully optimized geometries further
demonstrate that the stabilization energy is increased from α-CD-TTF-OCH3 to
α-CD-[TTF-OCH3]+·, and α-CD-[TTF-OCH3]
2+ complexes, which is in disagreement with
experimental observations. These results suggest that quantum chemical calculation in vacuo
cannot correctly predict the interaction of the α-CD ring with the charged TTF units. It is also
reported that simple continuum solvation model is not sufficient to model the CD interaction
with some radical cations and anions, so that explicit solvent molecules should be taken into
account to describe the interaction properly. [108]
In summary, the free energy profiles calculated by MD simulations with explicit water
can correctly reproduce the experimentally observed binding preference for each state.
Analysis of the free energy components reveals that, for these α-CD based rotaxanes with
charged TTF units, the real driving force for the shuttling in the oxidized states is actually the
interactions between water and the rotaxane components, which overwhelm the opposing
effect of the attractive interactions between the α-CD ring and the charged dumbbell.
Chapter 6 Conclusions and Outlook
55
Chapter 6
Conclusions and Outlook
In this thesis, the atomic MD simulation technique has been applied for investigations of
structural and energetic properties of several carbohydrate related materials:
The atomic details of thermal induced transitions in crystal cellulose Iβ were studied by
MD simulations with the GLYCAM06 force field. The predicted crystal unit cell
parameters for the structures at low and high temperatures agreed well with the
experimental crystal structures of cellulose Iβ. The computed phase transition
temperature of 475-500 K also correlated well with the experimental one. The
reorganization of the hydrogen-bonding network along with the phase transition was in
qualitative agreement with temperature-dependent IR measurements.
The adsorption properties of XGs on the surface of cellulose Iβ, in particular the effects
of solvent molecules and side chain variation, were examined. The three XGs studied,
GXXXGXXXG, GXXLGXXXG and GXXFGXXXG, differ in length and composition
of one side chain. It is found that the interaction energies between XG and cellulose in
water were considerably lower than those in vacuo and the van der Waals interactions
played a prevailing role over the electrostatic interactions in the adsorption. In our model,
variation in one side chain did not have significant influence on the interaction energy
and the binding affinity. However, the side chain variation did affect the equilibrium
structural properties of the adsorbed XGs.
The interaction of tetradecasaccharide XXXGXXXG in complex with the hybrid aspen
xyloglucan endo-transglycosylase PttXET16-34 was analyzed. The effect of the charge
state of the nucleophile “helper residue” Asp87 on the PttXET16-34 active site structure
was emphasized. When Asp87 is protonated, the hydrogen bonding network formed led
to the correct positioning of the catalytic machinery of both the nucleophile Glu85 and
Chapter 6 Conclusions and Outlook
56
electrophile Glu89. Our results indicate that the catalysis in glycoside hydrolase family
16 is optimal when the catalytic nucleophile is deprotonated, while the “helper residue”
and general acid/base residue are both protonated.
The working mechanism for a redox-responsive bistable [2]rotaxane based on α-CD ring
was investigated. The free energy profiles were calculated employing an umbrella
sampling technique for the shuttling motion of the α-CD ring between a TTF recognition
site and a TZ unit along the dumbbell of the rotaxane for three oxidation states (0, +1, +2)
of the TTF unit. These calculated free energy profiles verified the experimentally
observed binding preference for each state. Analysis of the free energy components
revealed that, for the α-CD based rotaxanes with charged TTF units the real driving force
for the shuttling in the oxidized states is actually the interactions between water and the
rotaxane components, which overwhelm the attractive interactions between the α-CD ring
and the charged dumbbell.
The classical MD simulation method has proved to be a powerful tool to understand in
atomic detail the dynamics, structures and interactions of carbohydrate systems. It is a
powerful supplement to predict and interpret experimental observations. The quality of MD
simulations relies on two important aspects: the accuracy of the chosen force field and a
sufficient conformational sampling.
The development of force fields for carbohydrates has been a challenging endeavor.
Recent progress in carbohydrate force fields has promoted the computer simulations of
carbohydrates and protein-carbohydrate interactions to a new frontier. The cellulose crystal
structure can be predicted with an accuracy that can match, and even refine the experimental
precision. Improvements of carbohydrate force fields will be undertaken in order to further
increase the accuracy of simulations of cellulose based systems. With the current computer
technology, the simulation time and system size of atomic-scale simulations have been
practically limited to about 100 ns and a million of particles. However, the structural
flexibility and ruggedness of the free energy landscape for polysaccharides in condensed
phase pose a tough issue for conformational sampling. For these cases, enhanced sampling
techniques such as replica exchange molecular dynamics or its variants and metadynamics
simulations can offer a way forward. On the other hand, coarse-graining techniques, in which
the number of degrees of freedom is reduced and fine interaction details are eliminated, is also
a possible way to extend by orders of magnitude the time-scale and length-scale of molecular
modeling and enhance conformational sampling. The combined/hybrid quantum mechanics
and molecular mechanics calculations provides the opportunity to treat electronic excited
Chapter 6 Conclusions and Outlook
57
states, electron transfer processes and chemical reactions in large biomolecular systems with
high accuracy and efficiency. These are viable approaches for further studies of enzymatic
catalyses on carbohydrates. Multi-scale modeling is a key for further improved accuracy and
efficiency for simulations of carbohydrate and carbohydrate-protein interactions, and will,
together with dedicated experiments for validation and feedback, pave the way for future
breakthroughs in carbohydrate research.
59
References
[1] Teeri, T. T.; Brumer, H.; Daniel, G.; Gatenholm, P. (2007) Biomimetic Engineering of Cellulose-Based
Materials. Trends Biotech. Vol. 25, pp. 299-306.
[2] Cosgrove, D. J. (2005) Growth of the Plant Cell Wall. Nat. Rev. Mol. Cell. Biol. Vol. 6, pp. 850–861.
[3] Philipp, B. (1997) Cellulose-new prospects for an old polymer. Polymer News Vol. 22, pp. 205-208.
[4] Klemm, D.; Hans-Perter, S.; Heinze, T. (2002) In Biopolymers - Polysaccharides II. Vol. 6,
Steinbüchel A (ed) Wiley-VCH, Weinheim,
[5] Pérez, S.; Mazeau, K. (2005) In Polysaccharides, Structure and Functional Versatility, 2nd edn.
Dumitriu, S. (ed) Marcel Dekker, New York, pp.41-68.
[6] Sullivan, A. C. (1997) Cellulose: the Structure Slowly Unravels. Cellulose Vol. 4, pp. 173-207.
[7] Sugiyama, J.; Okana, T.; Yamamoto, H.; Horii, F. (1990) Transformation of Valonia cellulose crystals
by an alkaline hydrothermal treatment. Macromolecules Vol. 23, pp. 3196-3198.
[8] Clark, G. L.; Parker, E. A. (1937) X-Ray Diffraction Study of the Action of Liquid Ammonia on
Cellulose and Its Derivatives. J. Phys. Chem. Vol. 41, pp. 777-786.
[9] Langan, P.; Nishiyama, Y.; Chanzy, H. (2001) X-ray Structure of Mercerized Cellulose II at 1 Å
Resolution. Biomacromolecules Vol. 2, pp. 410-416.
[10] Nishiyama, Y.; Langan, P.; Chanzy, H. (2002) Crystal Structure and Hydrogen-Bonding System in
Cellulose Iβ From Synchrotron X-ray and Neutron Fiber Diffraction. J. Am. Chem. Soc. Vol. 124, pp.
9074-9082.
[11] Nishiyama, Y.; Sugiyama, J.; Chanzy, H.; Langan, P. (2003) Crystal Structure and Hydrogen -Bonding
System in Cellulose Iα From Synchrotron X-ray and Neutron Fiber Diffraction. J. Am. Chem. Soc. Vol.
125, pp. 14300-14306.
[12] Wada, M.; Chanzy, H.; Nishiyama, Y.; Langan, P. (2004) Cellulose IIII Crystal Structure and
Hydrogen Bonding by Synchrotron X-ray and Neutron Fiber Diffraction. Macromolecules Vol. 37, pp.
8548-8555.
[13] Wada, M.; Nishiyama, Y.; Langan, P. (2006) X-ray Structure of Ammonia-Cellulose I: New Insights
into the Conversion of Cellulose I to Cellulose III. Macromolecules Vol. 39, pp. 2947-2952.
[14] Bismarck, A.; Mohanty, A. K.; Aranberri-Askargorta, I.; Czapla, S.; Misra, M.; Hinrichsen, G.;
References
60
Springer, J. (2001) Surface Characterization of Natural Fibers; Surface Properties and Water Up-take
Behavior of Modified Sisal and Coir Fibers. Green Chem. Vol. 3, pp. 100-107.
[15] Feng, D.; Caulfield, D. F.; Sanadi, A. R. (2001) Effect of Compatibilizer on the Structure-Property
Relationships of Kenaf-Fiber/Polypropylene Composites. Polym. Compos. Vol. 22, pp. 506-517.
[16] Lu, X.; Zhang, M. Q.; Rong, M. Z.; Shi, G.; Yang, G. C. (2003) Self-Reinforced Melt Processable
Composites of Sisal. Compos. Sci. Technol. Vol. 63, pp. 177-186.
[17] Mohanty, A. K.; Misra, M.; Drzal, L. T. (2001) Surface Modifications of Natural Fibers and
Performance of the Resulting Biocomposites: An Overview. Compos. Interfaces Vol. 8, pp. 313-343.
[18] Brumer, H.; Zhou, Q.; Baumann, M. J.; Carlsson, K.; Teeri, T. T. (2004). Activation of Crystalline
Cellulose Surfaces Through the Chemoenzymatic Modification of Xyloglucan. J. Am. Chem. Soc. Vol.
126, pp. 5715-5721.
[19] Yarbrough, J. M.; Himmel, M. E.; Ding, S. -Y. (2009) Plant Cell Wall Characterization Using
Scanning Probe Microscopy Techniques. Biotechnol Biofuels Vol. 2, pp. 17.
[20] Wada, M. (2002) Lateral Thermal Expansion of Cellulose I and IIII Polymorphs. J. Polym. Sci.,
Polym. Phys. Vol. 40, pp. 1095-1102.
[21] Wada, M.; Hori, R.; Kim, U. J.; Sasaki, S. (2010) X-ray Diffraction Study on the Thermal Expansion
Behavior of Cellulose I and its High-Temperature Phase. Polymer Degrad. Stabil. Vol. 95, pp.
1330-1334.
[22] Watanabe, A.; Morita, S.; Ozaki, Y. (2006) Study on Temperature-Dependent Changes in Hydrogen
Bonds in Cellulose I by Infrared Spectroscopy with Perturbation-Correlation Moving-Window
Two-Dimensional Correlation Spectroscopy. Biomacromolecules Vol. 7, pp. 3164-3170.
[23] Watanabe, A.; Morita, S.; Ozaki, Y. (2006) Temperature-Dependent Structural Changes in Hydrogen
Bonds in Microcrystalline Cellulose Studied by Infrared and Near-Infrared Spectroscopy with
Perturbation-Correlation Moving-Window Two-Dimensional Correlation Analysis. Appl. Spectrosc.
Vol. 60, pp. 611-618.
[24] McCann, M. C.; Wells, B.; Roberts, K. (1990) Direct Visualization of Cross-Links in the Primary
Plant Cell Wall. J. Cell Sci. Vol. 96, pp. 323-334.
[25] Pauly, M.; Albersheim, P.; Darvill, A.; York, W. S. (1999) Molecular Domains of the
Cellulose/Xyloglucan Network in the Cell Walls of Higher Plants. Plant J. Vol. 20, pp. 629–639.
[26] Hoffman, M.; Jia, Z. H.; Pena, M. J.; Cash, M.; Harper, A.; Blackburn, A. R.; Darvill, A.; York, W. S.
(2005) Structural Analysis of Xyloglucans in the Primary Cell Walls of Plants in the Subclass
Asteridae. Carbohydr. Res. Vol. 340, pp. 1826–1840.
[27] Hsieh, Y. S. Y.; Harris, P. J. (2009) Xyloglucans of Monocotyledons Have Diverse Structures. Mol.
Plant Vol. 2, pp. 943–965.
[28] Fry, S.C. (1989). The Structure and Functions of Xyloglucan. J. Exp. Botany Vol. 40, pp. 1-11.
[29] Fry, S. C.; York, W. S.; Albersheim, P.; Darvill, A.; Hayashi, T.; Joseleau, J. P.; Kato, Y.; Lorences, E.
P.; Maclachlan, G. A.; McNeil, M.; Mort, A. J.; Reid, J. S. G.; Seitz, H. U.; Selvendran, R. R.; Voragen,
References
61
A. G. J.; White, A. R. (1993) An Unambiguous Nomenclature for Xyloglucan-derived
Oligosaccharides. Physiol. Plantarum. Vol. 89, pp. 1–3.
[30] Zhou, Q.; Rutland, M. W.; Teeri, T. T.; Brumer, H. (2007) Xyloglucan in Cellulose Modification.
Cellulose Vol. 14, pp. 625–641.
[31] Taylor, I. E. P.; Atkins, E. D. T. (1985) X-ray Diffraction Studies on the Xyloglucan from Tamarind
(Tamarindus indica) Seed. FEBS Lett. Vol. 181, pp. 300–302.
[32] Hayashi, T.; Ogawa, K.; Mitsuishi, Y. (1994) Characterization of the Adsorption of Xyloglucan to
Cellulose. Plant Cell Physiol. Vol. 35, pp. 1199-1205.
[33] Finkenstadt, V. L.; Hendrixson, T. L.; Millane, R. P. (1995) Models of Xyloglucan Binding to
Cellulose Microfibrils. J. Carb. Chem. Vol. 14, pp. 601-611.
[34] Levy, S.; Maclachlan, G.; Staehelin, L. A. (1997) Xyloglucan Sidechains Modulate Binding to
Cellulose During in Vitro Binding Assays as Predicted by Conformational Dynamics Simulations.
Plant J. Vol. 11, pp. 373–386.
[35] Hanus, J.; Mazeau, K. (2006) The Xyloglucan-Cellulose Assembly at the Atomic Scale. Biopolymers
Vol. 82, pp. 59-73.
[36] Carpita, N.C.; Gibeaut, D. M. (1993). Structural Models of Primary Cell Walls in Flowering Plants -
Consistency of Molecular Structure with the Physical Properties of the Walls During Growth. Plant J.
Vol. 3, pp. 1-30.
[37] Vocadlo, D. J.; Davies, G. J.; Laine, R.; Withers, S. G. (2001). Catalysis by Hen Egg-white Lysozyme
Proceeds via a Covalent Intermediate. Nature Vol. 412, pp. 835-838.
[38] Planas, A. (2000) Bacterial 1,3-1,4-β-D-glucanases:Structure, Function and Protein Engineering.
Biophys. Acta - Prot. Struct. Molec. Enzymol. Vol. 1543, pp. 361-382.
[39] Cheatham, T. E. (2004) Simulation and modeling of nucleic acid structure, dynamics and interactions.
Curr. Opin. Struct. Biol. Vol. 14, pp. 360-367.
[40] Ballardini, R.; Balzani, V.; Credi, A.; Gandolfi, M. T.; Venturi, M. (2001). Artificial Molecular-Level
Machines: Which Energy To Make Them Work? Acc. Chem. Res. Vol. 34, pp. 445–455.
[41] http://en.wikipedia.org/wiki/Molecular_machine
[42] Verlet, L. (1967) Computer “Experiments” on Classical Fluids. I. Thermodynamical Properties of
Lennard-Jones Molecules. Phys. Rev. Vol. 139, pp. 98–103.
[43] Hockney, R. W.; Goel, S. P.; Eastwood, J. (1974) Quiet High Resolution Computer Models of a
Plasma. J. Comp. Phys. Vol. 14, pp. 148–158.
[44] Swope, W. C.; Andersen, H. C.; Berens, P. H.; Wilson, K. R. (1982) A Computer-simulation Method
for the Calculation of Equilibrium-constants for the Formation of Physical Clusters of Molecules:
Application to Small Water Clusters. J. Chem. Phys. Vol. 76, pp. 637–649.
[45] Allen, M. P.; Tildesley, D. J. (1987) Computer Simulation of Liquids. Oxford University Press.
[46] Lindahl, E.; Hess, B.; van der Spoel, D. (2001) GROMACS 3.0: A Package for Molecular Simulation.
J. Mol. Mod. Vol. 7, pp 306–317.
[47] van der Spoel, D.; Lindahl, E.; Hess, B.; Groenhof, G.; Mark, A. E.; Berendsen, H. J. C. (2005)
References
62
GROMACS: Fast, Flexible and Free. J. Comput. Chem. Vol. 26, pp. 1701–1718.
[48] Hess, B.; Kutzner, C.; van der Spoel, D.; Lindahl, E. (2008) GROMACS 4: Algorithms for Highly
Efficient, Load-Balanced, and Scalable Molecular Simulation. J. Chem. Theory Comput. Vol. 4, pp.
435–447.
[49] van der Spoel, D.; Lindahl, E.; Hess, B.; van Buuren, A. R.; Apol, E.; Meulenhoff, P. J.; Tieleman, D.
P.; Sijbers, A. L. T. M.; Feenstra, K. A.; van Drunen R.; Berendsen, H. J. C. (2010) Gromacs User
Manual version 4.5.
[50] Bergdorf, M.; Peter, C.; Hünenberger, P. (2003) Influence of Cut-off Truncation and Artificial
Periodicity of Electrostatic Interactions in Molecular Simulations of Solvated Ions: A Continuum
Electrostatics Study. J. Chem. Phys. Vol. 119, pp. 9129–9144.
[51] Weber, W.; Hünenberger, P.; McCammon, J. (2000) Molecular Dynamics Simulations of a
Polyalanine Octapeptide under Ewald Boundary Conditions: Influence of Artificial Periodicity on
Peptide Conformation. J. Phys. Chem. B Vol. 104, pp. 3668–3675.
[52] Saito, M. (1994) Molecular Dynamics Simulations of Proteins in Solution: Artifacts Caused by the
Cutoff Approximation. J. Chem. Phys. Vol. 101, pp. 4055.
[53] Baumketner, A.; Shea, J. (2005) The Influence of Different Treatments of Electrostatic Interactions on
the Thermodynamics of Folding of Peptides. J. Phys. Chem. B Vol. 109, pp. 21322–21328.
[54] Ewald, P. (1921) Die Berechnung Optischer und Elektrostatischer Gitterpotentiale. Ann. Phys. Vol. 64,
pp. 253–287.
[55] Darden, T.; York, D.; Pedersen, L. (1993) Particle Mesh Ewald: An N·log(N) Method for Ewald Sums
in Large Systems. J. Chem. Phys. Vol. 98, pp. 10089–10092.
[56] Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.; Pedersen, L. G. (1995) A Smooth
Particle Mesh Ewald Method. J. Chem. Phys. Vol. 103, pp. 8577–8592.
[57] Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H. J. C. (1977). Numerical integration of the Cartesian
Equations of Motion of a System with Constraints: Molecular Dynamics of n-Alkanes. J. Comput.
Phys. Vol. 23, pp. 321–341.
[58] Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. G. E. M. (1997) LINCS: A linear constraint
solver for molecular simulations. J. Comp. Chem. Vol. 18, pp. 1463–1472.
[59] Hess, B. (2007). P-LINCS: A parallel linear constraint solver for molecular simulation. J. Chem.
Theory Comput. Vol. 4, pp. 116–122.
[60] Berendsen, H. J. C.; Postma, J. P. M.; DiNola, A.; Haak, J. R. (1984) Molecular Dynamics with
Coupling to an External Bath. J. Chem. Phys. Vol. 81, pp. 3684–3690.
References
63
[61] Nosé, S. (1984) A Molecular Dynamics Method for Simulations in the Canonical Ensemble. Mol. Phys.
Vol. 52, pp. 255–268.
[62] Hoover, W. G. (1985) Canonical Dynamics: Equilibrium Phase-Space Distributions. Phys. Rev. A Vol.
31, pp. 1695–1697.
[63] Bussi, G.; Donadio, D.; Parrinello, M. (2007) Canonical Sampling Through Velocity Rescaling. J.
Chem. Phys. Vol. 126, No. 014101.
[64] Parrinello, M; Rahman, A. (1981) Polymorphic Transitions in Single Crystals: A New Molecular
Dynamics Method. J. Appl. Phys. Vol. 52, pp. 7182–7190.
[65] Nosé, S.; Klein, M. L. (1983) Constant Pressure Molecular Dynamics for Molecular Systems. Mol.
Phys. Vol. 50, pp. 1055–1076.
[66] Leach, A. R. (2001) Molecular Modelling: Principles and Applications, Second Edition. Pearson
Education EMA.
[67] Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Merz, Jr. K. M.; Ferguson, D. M.; Spellmeyer,
D. C.; Fox, T.; Caldwell, J. W.; Kollman, P. A. (1995) A Second Generation Force Field for the
Simulation of Proteins, Nucleic Acids, and Organic Molecules. J. Am. Chem. Soc. Vol. 117, pp.
5179–5197.
[68] Wang, J. M.; Cieplak, P.; Kollman, P. A. (2000) How Well Does a Restrained Electrostatic Potential
(RESP) Model Perform in Calculating Conformational Energies of Organic and Biological Molecules?
J. Comput. Chem. Vol. 21, pp. 1049–1074.
[69] Duan, Y.; Wu, C.; Chowdhury, S.; Lee, M. C.; Xiong, G.; Zhang, W.; Yang, R.; Cieplak, P.; Luo, R.;
Lee, T.; Caldwell, J.; Wang, J.; Kollman, P. (2003) A point-charge force field for molecular
mechanics simulations of proteins based on condensed-phase quantum mechanical calculations. J.
Comput. Chem., Vol. 24, pp. 1999–2012.
[70] MacKerell, Jr., A. D.; Bashford, D.; Bellott, M.; Dunbrack, Jr. R. L.; Evanseck, J. D.; Field, M. J.;
Fischer, S.; Gao, J.; Guo, H.; Ha, S.; Joseph-McCarthy, D.; Kuchnir, L.; Kuczera, K.; Lau, F. T. K.;
Mattos, C.; Michnick, S.; Ngo, T.; Nguyen, D. T.; Prodhom, B.; Reiher, III W. E.; Roux, B.;
Schlenkrich, M.; Smith, J. C.; Stote, R.; Straub, J.; Watanabe, M.; Wiórkiewicz-Kuczera, J.; Yin, D.;
Karplus, M. (1998) All-Atom Empirical Potential for Molecular Modeling and Dynamics Studies of
Proteins. J. Phys. Chem. B Vol. 102, pp. 3586–3616.
[71] MacKerell, A. D.; Banavali, N.; Foloppe, N. (2000) Development and Current Status of the
CHARMM Force Field for Nucleic Acids. Biopolymers Vol. 56, pp. 257–265.
[72] Hermans, J.; Berendsen, H. J. C.; van Gunsteren, W. F.; Postma, P. M. (1984) A Consistent Empirical
Potential for Water-Protein Interactions. Biopolymers Vol. 23, pp. 1513-1518.
[73] Scott, W. R. P.; Hünenberger, P. H.; Tironi, I. G.; Alan E. Mark, A. E.; Salomon R. Billeter, S. R.;
References
64
Fennen, J.; Torda, A. E.; Huber, T.; Krüger, P.; van Gunsteren, W. F. (1999) The GROMOS
Biomolecular Simulation Program Package. J. Phys. Chem. A Vol. 103, pp. 3596–3607.
[74] Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. (1996) Development and Testing of the OPLS
All-Atom Force Field on Conformational Energetics and Properties of Organic Liquids. J. Am. Chem.
Soc. Vol. 118, pp. 11225–11236.
[75] Wang, J.; Wolf, R. M.; Caldwell, J. W.; Kollman, P. A.; Case, D. A. (2004) Development and Testing
of a General Amber Force Field. J. Comput. Chem. Vol. 25, pp. 1157–1174.
[76] Singh, U. C.; Kollman, P. A. (1984) An Approach to Computing Electrostatic Charges for Molecules.
J. Comput. Chem. Vol. 5, pp. 129-145.
[77] Bayly, C. I.; Cieplak, P.; Cornell, W.; Kollman, P. A. (1993) A Well-Behaved Electrostatic Potential
Based ethod Using Charge Restraints for Deriving Atomic Charges: The RESP Model. J. Phys. Chem.
Vol. 97, pp. 10269-10280.
[78] Vliegenthart, J. F. G.; Woods, R. J. Eds. (2006) NMR Spectroscopy and Computer Modeling of
Carbohydrates: Recent Advances, American Chemical Society, Washington, DC, pp. 235-257.
[79] Kirschner, K.N.; Woods, R. J. (2001) Solvent Interactions Determine Carbohydrate Conformation.
Proc. Natl. Acad. Sci. U. S. A. Vol. 98, pp. 10541–10545.
[80] Woods, R. J.; Dwek, R. A.; Edge, C. J.; Fraser-Reid, B. (1995) Molecular Mechanical and Molecular
Dynamical Simulations of Glycoproteins and Oligosaccharides. 1. GLYCAM-93 Parameter
Development. J. Phys. Chem. Vol. 99, pp. 3832–3846.
[81] Kirschner, K. N.; Yongye, A. B.; Tschampel, S. M.; González-Outeiriño, J.; Daniels, J. C. R.; Foley, B.
L.; Woods, R. J. (2008) GLYCAM06: a Generalizable Biomolecular Force Field. Carbohydrates. J.
Comput. Chem. Vol. 29, pp. 622-655.
[82] Hatcher, E.; Olgun Guvench, O.; MacKerell, A. D. Jr. (2009) CHARMM Additive All-Atom Force
Field for Aldopentofuranose, Methyl-aldopentofuranosides, and Fructofuranose. J. Phys. Chem. B Vol.
113, pp. 12466–12476.
[83] Guvench, O.; Greene, S. N.; Kamath, G.; Brady, J. W.; Venable, R. M.; Pastor, R. W.; Mackerell, A. D.
Jr. (2008) Additive Empirical Force Field for Hexopyranose Monosaccharides. J. Comput. Chem. Vol.
29, pp. 2543–2564.
[84] Guvench, O.; Elizabeth Hatcher, E.; Venable, R. M.; Pastor, R. W.; MacKerell, A. D. Jr. (2009)
CHARMM Additive All-Atom Force Field for Glycosidic Linkages between Hexopyranoses. J. Chem.
Theory Comput. Vol. 5, pp. 2353–2370.
References
65
[85] Lins, R. D.; Hunenberger, P. H. (2005) A New GROMOS Force Field for Hexopyranose-based
Carbohydrates. J. Comput. Chem. Vol. 26, pp. 1400–1412.
[86] Kony, D.; Damm, W.; Stoll, S.; Van Gunsteren, W. F. (2002) An Improved OPLS-AA Force Field for
Carbohydrates. J. Comput. Chem. Vol. 23, pp. 1416–1429.
[87] Fadda, E.; Woods, R. J. (2010) Molecular Simulations of Carbohydrates and Protein–Carbohydrate
Interactions: Motivation, Issues and Prospects. Drug Disc. Today Vol. 15, pp. 596-609.
[88] Chaikin, P. M.; Lupensky, T. C. (2000) Principles of condensed matter physics. Cambridge University
Press, Cambridge.
[89] Torrie, G. M.; Valleau, J. P. (1977) Non-physical Sampling Distributions in Monte-Carlo Free-energy
Estimation – Umbrella Sampling. J. Comput. Phys. Vol. 23, pp. 187-199.
[90] Park, S.; Khalili-Araghi, F.; Tajkhorshid, E.; Schulten, K. (2003) Free Energy Calculation from
Steered Molecular Dynamics Simulations using Jarzynski’s Equality. J. Chem. Phys. Vol. 119, pp.
3559-3566.
[91] Kumar, S.; Bouzida, D.; Swendsen, R. H.; Kollman, P. A.; J. M. Rosenberg. (1992) The Weighted
Histogram Analysis Method for Free-energy Calculations on Biomolecules. I. the Method. J. Comp.
Chem. Vol.13, pp. 1011-1021.
[92] Jarzynski, C. (1997) Equilibrium Free-energy Differences from Nonequilibrium Measurements: A
Master-equation Approach. Phys. Rev. E Vol. 56, pp. 5018-5035.
[93] Jarzynski, C. (1997) Nonequilibrium Equality for Free Energy Differences. Phys. Rev. Lett. Vol. 78,
pp. 2690-2693.
[94] Jensen, M. Ø.; Park, S.; Tajkhorshid, E.; Schulten, K. (2002) Energetics of Glycerol Conduction
through Aquaglyceroporin GlpF. Proc. Natl. Acad. Sci. U.S.A. Vol. 99, pp. 6731-6736.
[95] Hummer, G. (2001) Fast-growth Thermodynamic Integration: Error and Efficiency Analysis. J. Chem.
Phys. Vol. 114, pp. 7330.
[96] Marcinkiewicz, J. (1039) Sur une propriété de la loi de Gauss. Math. Z. Vol. 44, pp. 612-618.
[97] Balsera, M.; Stepaniants, S.; Izrailev, S.; Oono, Y.; Schulten, K. (1997) Reconstructing Potential
Energy Functions from Simulated Force-induced Unbinding Processes. Biophys. J. Vol. 73, pp.
1281-1287.
[98] Bergenstråhle, M.; Berglund, L. A.; Mazeau, K. (2007) Thermal Response in Crystalline I Cellulose:
A Molecular Dynamics Study. J. Phys. Chem. B Vol. 111, pp. 9138-9145.
References
66
[99] Spiwok, V.; Králová, B.; Tvaroška, I. (2010) Modelling of -D-glucopyranose Ring Distortion in
Different Force Fields: A Metadynamics Study. Carbohydr. Res. Vol. 345, pp. 530-537.
[100] Mazeau, K. (2005) Structural Micro-heterogeneities of Crystalline I cellulose. Cellulose Vol. 12,
pp. 339-349.
[101] Neyertz, S.; Pizzi, A.; Merlin, A.; Maigret, B.; Brown, D.; Deglise, X. (2000) A New All-atom Force
Field for Crystalline Cellulose I. J. Appl. Polym. Sci. Vol. 78, pp. 1939-1946.
[102] Reiling, S.; Brickmann, J. (1995) Theoretical Investigations on the Structure and Physical Properties
of Cellulose. Macromol. Theory Simul. Vol. 4, pp. 725-743.
[103] Morris, S.; Hanna, S.; Miles, M. J. (2004) The Self-assembly of Plant Cell Wall Components by
Single-molecule Force Spectroscopy and Monte Carlo Modeling. Nanotechnology Vol. 15, pp.
1296-1301.
[104] Balzani, V.; Credi, A.; Venturi, M. Molecular Devices and Machines Concepts and Perspectives for
the Nanoworld, Wiley-VCH, Weinheim, 2008.
[105] Balzani, V.; Credi, A.; Raymo, F. M.; Stoddart, J. F. (2000) Artificial molecular machines. Angew.
Chem. Int. Ed. Vol. 39, pp. 3349-3391.
[106] Tian, H.; Wang, Q. C. (2006) Recent Progress on Switchable Rotaxanes. Chem. Soc. Rev. Vol. 35, pp.
361-374.
[107] Zhao, Y-L.; Dichtel, W. R.; Trabolsi, A.; Saha, S.; Aprahamian, I.; Stoddart, J. F. (2008) A
redox-switchable alpha-cyclodextrin-based [2]rotaxane. J. Am. Chem. Soc. Vol. 130, pp.
11294-11296.
[108] Mu, T-W.; Liu, L.; Li, X-S.; Guo, Q-X. (2001) A Theoretical Study on the Inclusion Complexation
of Cyclodextrins with Radical Cations and Anions. J. Phys. Org. Chem. Vol. 14, pp. 559-565.