Content uploaded by Fumio Hirata
Author content
All content in this area was uploaded by Fumio Hirata on Jul 12, 2022
Content may be subject to copyright.
Comparison of DNA hydration patterns obtained using two distinct
computational methods, molecular dynamics simulation
and three-dimensional reference interaction site model theory
Yoshiteru Yonetani,1Yutaka Maruyama,2Fumio Hirata,2,a兲and Hidetoshi Kono1,3,a兲
1Computational Biology Group, Quantum Beam Science Directorate, Japan Atomic Energy Agency,
8-1 Umemidai, Kizugawa, Kyoto 619-0215, Japan
2Department of Theoretical and Computational Molecular Science, Institute for Molecular Science,
Okazaki 444-8585, Japan
3PRESTO, Japan Science and Technology Agency, 4-1-8 Kawaguchi, Saitama 332-0012, Japan
共Received 28 December 2007; accepted 12 March 2008; published online 9 May 2008兲
Because proteins and DNA interact with each other and with various small molecules in the
presence of water molecules, we cannot ignore their hydration when discussing their structural and
energetic properties. Although high-resolution crystal structure analyses have given us a view of
tightly bound water molecules on their surface, the structural data are still insufficient to capture the
detailed configurations of water molecules around the surface of these biomolecules. Thanks to the
invention of various computational algorithms, computer simulations can now provide an atomic
view of hydration. Here, we describe the apparent patterns of DNA hydration calculated by using
two different computational methods: Molecular dynamics 共MD兲simulation and three-dimensional
reference interaction site model 共3D-RISM兲theory. Both methods are promising for obtaining
hydration properties, but until now there have been no thorough comparisons of the calculated
three-dimensional distributions of hydrating water. This rigorous comparison showed that MD and
3D-RISM provide essentially similar hydration patterns when there is sufficient sampling time for
MD and a sufficient number of conformations to describe molecular flexibility for 3D-RISM. This
suggests that these two computational methods can be used to complement one another when
evaluating the reliability of the calculated hydration patterns. © 2008 American Institute of Physics.
关DOI: 10.1063/1.2904865兴
I. INTRODUCTION
When discussing the structural and energetic properties
of biological molecules such as proteins and nucleic acids,
one must consider not only the solute biomolecules but also
the solvent or surrounding water molecules.1Indeed, the
stable states of biological molecules are significantly affected
by the solvent conditions. A good example of this is the
structure of DNA, which exhibits structural transition among
different A, B, and Z forms, depending on the solvent and its
ionic strength. Although high-resolution x-ray crystal struc-
ture analyses have provided us with detailed pictures of hy-
dration states,2,3these pictures remain incomplete because 共i兲
the positions of hydrogen atoms cannot be determined by
x-ray diffraction; 共ii兲the hydrated structure is not viewed
under physiological conditions, i.e., the solution state is not
necessarily obtained through analysis of crystallized
samples, as the effect of crystal packing is not negligible;
and 共iii兲the hydrated structure may be affected by the salt
species and concentration. The positions of hydrogen atoms
can be determined through neutron crystal structure
analysis,2but so far the structures of only 19 biomolecules
have been solved using neutron diffraction 共as of Nov. 20,
2007兲.
An alternative approach to the structural analysis of the
hydration of biomolecules is to use computational methods,
which can be complementary to experimental measurements.
Such computational approaches are classified into two cat-
egories. One is many-particles simulation,4such as the mo-
lecular dynamics 共MD兲and Monte Carlo methods, in which
molecular configurations are generated according to New-
ton’s equations of motion or stochastic walking, respectively.
Another is a more theoretical approach based on the integral
equations for molecular distribution functions.5A good ex-
ample is reference interaction site model 共RISM兲theory.6
Both the simulation and theoretical approaches are capable
of providing structures of the hydration around a solute mol-
ecule but have distinct features: The simulation can be used
with any complex molecular system because all atomic po-
sitions are numerically derived. This approach has therefore
been widely used to analyze the hydration of proteins,7
nucleic acids,8,9and their complexes.10 One shortcoming of
the simulation approach is the large computational resource
needed to obtain a complete statistical ensemble. If atomic
configurations are not sufficiently sampled, a proper descrip-
tion of the statistical properties will not be achieved. For
example, conventional MD techniques require very pro-
longed 共typically approximately milliseconds兲simulations to
estimate the probability of finding water molecules within a
protein cavity completely isolated from the bulk solvent.11
a兲Authors to whom correspondence should be addressed. Electronic
addresses: hirata@ims.ac.jp and kono.hidetoshi@jaea.go.jp.
THE JOURNAL OF CHEMICAL PHYSICS 128, 185102 共2008兲
0021-9606/2008/128共18兲/185102/9/$23.00 © 2008 American Institute of Physics128, 185102-1
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
The theoretical approach does not suffer from this sampling
problem because the configuration integral is made over the
entire configuration space of the solvent at equilibrium with
a fixed solute conformation. Consequently, it naturally
samples solvent configurations that correspond to the con-
fined space within a biomolecule cavity. On the other hand,
the theoretical calculation of a fixed solute conformation is
not necessarily sufficient to clarify the physics of DNA hy-
dration because structural flexibility is an essential feature of
most biomolecules and will have non-negligible effects on
the behavior of the surrounding water. In fact, the effect of
structural conformation on the hydration characteristics of
proteins has been investigated in our recent work.12
The purpose of the present work is to compare two rep-
resentative computational approaches, MD simulation and
RISM theory, and to evaluate their applicability for descrip-
tion of a biomolecule’s hydration. Both MD and RISM ap-
pear to have the potential to describe the structural features
of hydration, and the calculation conditions 共i.e., trajectory
length and solute treatment兲required to describe the hy-
drated structures of biomolecules by using the two methods
will be clarified through the present work. We chose the
double helix B-form of DNA as a test case. MD simulation
has always been capable of providing three-dimensional
共3D兲distributions of molecules, but until about 10 years ago
RISM theory was limited to the calculation of a one-
dimensional 共1D兲distribution function 共e.g., radial distribu-
tion function兲. Since then, however, extension to the
3D-RISM method enabled RISM to be also applicable for
calculation of 3D distributions.6A DNA-water system is suit-
able for the evaluation of calculated hydrations because
DNA molecules composed of particular sequences are
known to have a well-defined hydration pattern called a
“water spine,”13 which is composed of highly ordered water
molecules aligned along the narrow minor groove. Moreover,
it appears that the surface localized water plays important
roles in various biological and chemical processes. For in-
stance, the affinities with which proteins and drugs bind to
DNA are more or less influenced by entropic contributions
from water molecules released from the DNA surface upon
their binding.14 Surface water also affects the electronic
properties of DNA. The rate of charge transfer along DNA is
known to depend on the DNA sequence, and in this case,
surface water is also thought to be a major determinant of the
electronic properties.
II. METHODS
A. Molecular models
To compare the MD and 3D-RISM methods, we had to
construct a system in which the conditions used for the two
calculations were consistent. So as much as possible, we
used the same system and the same conditions for both cal-
culations. A 12 bare pair 共bp兲fragment of B-form DNA
d共5⬘CGCGATATCGCG3⬘兲was prepared as the solute. We
used the force field parameters of AMBER PARM 99 共Ref. 15兲
to simulate the DNA molecule. This parameter set is known
to be suitable for reproducing the structure of B-DNA in MD
simulations.16 For water molecules, we used the TIP3P
model.17 One distinction of the water treatment between the
3D-RISM and MD methods was that we used an additional
Lennard–Jones term with the parameters =192.5 J/mol and
=0.4 Å for the water hydrogen site in order to converge the
RISM calculation.18 K+and Cl−ions were included in the
MD simulation system, whereas they were ignored in the
RISM calculations.
Molecular flexibility is usually considered in MD simu-
lations of biomolecules, but the solute molecule is usually
fixed in the RISM calculations. Consequently, MD and 3D-
RISM calculations give different results, as the former con-
siders solute flexibility and the latter does not. In order to
evaluate the effects of solute treatment and to make an accu-
rate comparison, we carried out four types of calculations
shown in Table I, which included normal MD 共MDflex兲,MD
with a position constraint to mimic the solute treatment in
3D-RISM 共MDrigid兲, conventional 3D-RISM 共RISMrigid兲, and
3D-RISM with consideration of the solute flexibility
共RISMflex兲. We will explain how we performed these calcu-
lations in the following sections.
B. MD simulation
In MD simulations, time-dependent development of the
atoms in a system is derived from the numerical solution of
Newton’s equations of motion, where forces acting on the
atoms come from intra- and intermolecular interactions.4Af-
ter this numerical procedure, statistical properties, such as
the spatial distribution of solvent molecules, are calculated
over the trajectory. To perform our MD simulations 共i.e.,
MDflex and MDrigid兲, the DNA-water system was prepared as
follows. We initially generated the B-form structure of the 12
bp DNA d共5⬘CGCGATATCGCG3⬘兲fragment using the
Nucgen module of AMBER7.19 Then, 5114 water molecules
were placed around the DNA, and a periodic boundary con-
dition with a truncated octahedral box about 60⫻60
⫻60 Å3in size was imposed. Counterions consisting of 39
K+and 17 Cl−also were added so as to realize both a physi-
ological salt concentration of 0.15Mand electrical neutrality.
Using these systems, we computed the time-dependent
development of the atoms by numerically solving the equa-
tions of motion. Integration of the equations was accom-
plished using the Leapfrog algorithm with a time increment
of 1 fs. At this point, to stabilize the numerical integration,
stretching motions of the covalent bonds involving hydrogen
atoms were removed using the SHAKE constraint.20 When
evaluating the interaction forces, van der Waals components
were truncated at 9 Å. Coulomb components were evaluated
using the particle mesh Ewald21 method to account for their
long-ranged nature, which does not require any truncation.
For MDrigid, we further imposed a positional restraint on the
TABLE I. Methods and solute treatments adopted in the present work.
Method
Solute treatment
Flexible Rigid
MD MDflex MDrigid
3D-RISM RISMflex RISMrigid
185102-2 Yonetani et al. J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
solute DNA atoms with a harmonic function of
500 kcal mol−1 Å−2 to suppress flexibility and mimic the
RISM calculation. The MD simulations were performed us-
ing the AMBER SANDER module.19
A 10 ns trajectory was generated in which the pressure
and temperature were set to 1 atm and 300 K, respectively,
using the weak coupling method.22 We calculated the spatial
distribution of the water molecules using the last 8 ns of
data. In this process, the system space was divided into cubic
boxes with an edge size of 0.5 Å, after which the probability
that water molecules are present within each box was calcu-
lated. Note that before this calculation, we eliminated trans-
lational and rotational displacement of the DNA and ob-
tained a water distribution relative to the solute DNA. This
elimination was carried out by applying the translational and
rotational transformation to every instantaneous configura-
tion, so that the instantaneous DNA structure could be fitted
to a reference DNA structure. We chose the instantaneous
structure that was most similar to the time-averaged structure
as the reference structure.
C. 3D-RISM theory
The 3D-RISM theory6is an integral equation theory
based on statistical mechanics used to obtain the molecular
distribution functions from the intermolecular potential func-
tions and thermodynamic conditions 共i.e., temperature and
density兲. This theoretical procedure has two steps. The first
step is to calculate the density pair correlation function in the
aqueous solution based on the 1D-RISM theory, which rep-
resents the microscopic structure of the distribution of sol-
vent molecules. In the second step, we immerse a DNA
molecule into the solvent and calculate the 3D distribution of
solvent atoms around the solute molecule based on the
3D-RISM theory. The following is a brief outline of the 3D-
RISM method.
The 3D-RISM integral equation for the 3D solute-
solvent site total and direct correlation functions, h
␥
uv共r兲and
c
␥
uv共r兲, is written as23–25
h
␥
uv共r兲=兺
␥
⬘
c
␥
⬘
uv共r兲*共
␥
⬘
␥
vv +
vc
␥
⬘
␥
vv 共r兲兲,共1兲
where
␥
⬘
␥
vv 共r兲=
␦
共r−l
␥
⬘
␥
vv 兲is the intramolecular matrix of sol-
vent molecules with site separations l
␥
⬘
␥
vv ,
␥
and
␥
⬘are inter-
action sites for solvent molecules denoted with superscripts u
and v, respectively,
vis the solvent number density, and ⴱ
means convolution in direct space. The radial site-site corre-
lation functions for pure solvent h
␥
⬘
␥
vv 共r兲were independently
obtained from the conventional 1D-RISM theory modified
with the dielectrically consistent bridge corrections of
Perkyns and Pettitt.26 In the context of 3D-RISM, this en-
sures a proper macroscopic dielectric constant for the solvent
around the solute.25,27
In addition to 3D-RISM equation 关Eq. 共1兲兴, the following
Kovalenko-Hirata 共KH兲closure equations,24,28 which include
corrections for the supercell periodicity artifact in both the
direct and total 3D site correlation functions, c
␥
uv共r兲and
h
␥
uv共r兲, are employed to obtain the correlation functions
g
␥
uv共r兲=
再
exp共
兲+⌬Q
␥
uvfor
艋0
1+
+⌬Q
␥
uvfor
⬎0,
冎
共2兲
=−u
␥
uv共r兲
kBT+h
␥
uv共r兲−c
␥
uv共r兲−⌬Q
␥
uv,
where g
␥
uv共r兲=h
␥
uv共r兲+1 is the 3D solute-solvent site distri-
bution function, the interaction potential u
␥
uv共r兲between sol-
vent site
␥
and the whole solute is calculated on the supercell
grid using the Ewald summation method, and
⌬Q
␥
uv=4
VcellkBTq0lim
k→0兺
␥
⬘
q
␣
k2共
␥
⬘
␥
vv 共k兲+
vh
␥
⬘
␥
vv 共k兲兲 共3兲
is the shift in the distribution functions due to the supercell
background for the solute with net charge q=兺
␣
q
␣
ucom-
prised of the partial site charges q
␣
u. The 3D solute-solvent
site direct correlation functions c
␥
uv共r兲calculated from 3D-
RISM equation 关Eq. 共1兲兴with the closure equation 关Eq. 共2兲兴
were corrected by subtracting the long-ranged electrostatic
asymptotic of the periodic potential u
␥
uv共r兲synthesized using
the Ewald summation, and adding back that of the single,
nonperiodic solute simply tabulated on the 3D grid within
the supercell.25,27
In conventional RISM calculations, the positions of the
solute atoms do not change. We refer to such a calculation as
RISMrigid, and the solute DNA structure used was the same
as that used for MDrigid. We calculated RISMflex, which took
into account the flexibility of DNA, by repeating the 3D-
RISM calculation for every instantaneous configuration of
the solute, which were obtained from the MDflex calculation,
and then averaging the resultant water distributions. These
calculations were performed under ambient thermodynamic
conditions of 300 K and 0.997 g/cm3, in accordance with
the MD condition.
III. RESULTS AND DISCUSSION
A. Checking the required conditions
for the comparison of MD and 3D-RISM
Before comparing the MD and 3D-RISM results, we de-
termined whether or not the quality of each result was satis-
factory. A major determinant of the quality of the MD results
was the length of the trajectory. If the possible atomic con-
figurations were not sufficiently sampled in the simulation,
we would not obtain a reliable molecular distribution. To
examine the trajectory length dependence of the spatial dis-
tribution of water oxygen, sampling times of 1, 2, 4, and 8 ns
were compared. In Fig. 1, regions colored in blue have a
water oxygen distribution 2.25 times denser than that of the
bulk water. At trajectory lengths below 4 ns, the oxygen dis-
tribution varied with the trajectory length. By contrast, the
distribution showed no remarkable changes between 4 and
8 ns, which means a trajectory of ⬃8 ns is long enough to
get a reliable water distribution.
The 3D-RISM method is not subject to sampling prob-
lems because the 3D-RISM-derived distribution function is a
statistical average obtained from the integral over all the sol-
vation configurations. A problem does occur, however, when
185102-3 DNA hydration patterns J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
considering the solute flexibility in the RISM calculation. In
conventional RISM calculations, solute atoms are fixed, so
the solute flexibility is not considered. To consider the effect
of solute flexibility, in the present work, we performed nu-
merous 3D-RISM calculations for different DNA conforma-
tions obtained from MDflex and averaged the results. To see
the dependence of the results on the number of conforma-
tions used, the spatial distributions of water oxygen were
calculated for four sets of DNA conformations with different
sizes 共10, 50, 1200, and 2400兲共see Fig. 2兲. Each set was
prepared by randomly choosing the indicated number of
snapshots from the MDflex trajectory. As shown in Fig. 2, the
results obtained using 1200 and 2400 conformations were in
good agreement, while that obtained with the set of 10 or 50
conformations largely differed from the others, suggesting
that 1200 solute conformations are sufficient for realization
of a 3D-RISM calculation that considers solute flexibility.
On the basis of the above examination, we compared
3D-RISM and MD by using a trajectory length of 8 ns for
MDflex and 2400 conformations for RISMflex.
B. Comparison of the MD and 3D-RISM results
In our comparison of MD and 3D-RISM, we first com-
pared two cases of MDflex and RISMrigid because these two
calculations treated the solute DNA in the conventional man-
ner for each method. Figure 3shows the distributions of
water oxygen obtained using different threshold density val-
ues, which are multiples of the density of the bulk water. We
found that RISMrigid produces a much larger solvent distri-
bution than MDflex 关Figs. 3共a兲and 3共d兲兴. At a threshold of
2.25, for example, the water is distributed along the minor
groove when calculated using MDflex, but is widely distrib-
uted over the entire DNA molecule when calculated using
RISMrigid. Upon increasing the threshold to 4.5, the hydra-
tion pattern seen with RISMrigid becomes comparable to that
of MDflex with a threshold of 2.25. The same tendency was
confirmed by viewing the solvent distribution in a slice plane
parallel to the DNA base pairs 共Fig. 4兲. With MDflex
关Fig. 4共a兲兴, regions of high density are concentrated at the
groove surface, whereas with RISMrigid 关Fig. 4共d兲兴, they are
widely distributed and are present around the phosphates of
the DNA backbone as well as around the groove surface.
The difference in the apparent hydration patterns de-
scribed above mainly reflects the difference in the treatment
of the solute flexibility, not a difference in the methodologi-
cal details. We confirmed this statement by comparing the
results obtained when the calculations were made with con-
sistent solute treatment. MDflex 关Figs. 3共a兲and Fig. 4共a兲兴and
RISMflex 关Figs. 3共c兲and 4共c兲兴, which both consider the solute
flexibility, yielded similar patterns of water density. Like-
wise, MDrigid 关Figs. 3共b兲and 4共b兲兴and RISMrigid 关Figs. 3共d兲
and 4共d兲兴, which treat the solute as fixed, also gave water
densities that were similar to one another. Thus, MD and
RISM produce nearly the same hydration profile when the
solute is treated in the same manner. Irrespective of the
FIG. 1. 共Color兲MD trajectory length dependence of the distribution of water oxygen around DNA. Shown in blue are regions in which the density of the water
oxygen is 2.25 times greater than that of the bulk water. All the figures except Fig. 4were created using gOpenMol 共Ref. 37兲.
FIG. 2. 共Color兲Variation in the distribution of the water oxygen around DNA, depending on the number of DNA conformations used in the RISMflex
calculation. Shown in yellow are regions in which the density of the water oxygen is 1.7 times greater than that of the bulk water.
185102-4 Yonetani et al. J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
method used, the flexible models 共MDflex and RISMflex兲gave
less dense and broader distributions of water oxygen than the
rigid models 共MDrigid and RISMrigid兲. This is because the
structural variability of a flexible solute increases possible
hydration patterns.
To assess the patterns of DNA hydration in more detail,
we compared the MDflex and RISMflex results while focusing
on three representative regions 共Fig. 5兲. The water density
thresholds were set at 2.25 for MD and 1.7 for RISM. The
value of 1.7 was chosen for RISM so that the resultant spa-
tial distribution of the water resembled the MD result ob-
tained with a density threshold of 2.25. This slightly lower
threshold is likely attributable to an approximation employed
in the current 3D-RISM calculation. It is known that 3D-
RISM theory has a tendency to broaden the resultant distri-
bution functions.11 Another noteworthy difference between
the MD and RISM results is that, with RISM, high-density
regions are continuous with neighboring high-density re-
gions. This is also attributable to the characteristic spread of
the distribution functions seen without a treatment of inho-
mogeneity near the solute molecule.29
Except for the differences mentioned above, the MD and
3D-RISM results illustrated in Fig. 5are very similar. In the
central region of the ATAT sequence 关Fig. 5共a兲兴, the hydra-
tion sites are aligned along the minor groove, which is the
characteristic hydration pattern for DNA 共so called “spine
hydration”兲. At one end of the DNA molecule, the CGCG
region 关Fig. 5共b兲兴, the minor groove is slightly wider than in
the central ATAT region and shows a hydration pattern dif-
ferent from that seen in the ATAT region. In the CGCG re-
gion, water hydration sites are split into two branches, which
is consistent with an observation made in an earlier simula-
FIG. 3. 共Color兲Distributions of water oxygen obtained using the MDflex 共a兲,MD
rigid 共b兲, RISMflex 共c兲, and RISMrigid 共d兲calculations. Shown are the
distributions of water oxygen obtained with the indicated density thresholds, which are multiples of the water oxygen density in the bulk water.
FIG. 4. 共Color兲Water oxygen distribution on a slice
plane parallel to the DNA base pairs: 共a兲MDflex,
共b兲MDrigid,共c兲RISMflex, and 共d兲RISMrigid. Water oxy-
gen densities expressed as multiples of that in the bulk
water are given in the color scale. The figure was
created by Chimera 共Ref. 38.兲
185102-5 DNA hydration patterns J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
tion study.8Compared to the minor groove, the major groove
exhibits a less dense distribution with both MD and 3D-
RISM 关Fig. 5共c兲兴. In the major groove, various network pat-
terns of water molecules become possible as a result of the
width of the groove. Consequently, water localization at the
groove surface becomes weaker. In this respect, the
3D-RISM and MD calculations produced consistent
hydration results.
Next, we compared the calculated distribution of water
molecules with the positions of water molecules determined
by x-ray crystallographic analysis.30 In Fig. 6, the experi-
mental water oxygen atoms are represented by five different
colors, depending on their B-factor. Within the crystal struc-
ture, highly localized water molecules 共blue, B-factor
⬍10 Å2兲are aligned along the minor groove. Both MD- and
3D-RISM-derived water distributions agreed well with the
water positions in the crystal structure, although there were
slight positional displacements. This small difference prob-
ably reflects the different sample conditions used 共i.e., solu-
tion for the calculations and crystal for the experiment兲or
imperfection of the interaction potential employed in the cal-
culations. Then, we analyzed patterns of hydration around
the phosphate moieties of the DNA backbone. At a density
threshold of 2.25, MD-derived hydration was not observed
关Fig. 6共a兲兴. As already shown in Fig. 3共a兲, upon reducing the
density threshold to ⬃1.7, the MD-derived hydration pat-
terns became visible and were similar to the RISM-derived
patterns 关Fig. 6共b兲兴, which indicates that the water distribu-
tion is less dense and broader in the backbone region than the
minor groove region. This tendency is consistent with the
crystal structure, where only a few oxygen positions with
large B-factors were observed in the backbone region.
The distribution of water hydrogen atoms together with
that of oxygen atoms is shown in Fig. 7. In the MD result
关Fig. 7共a兲兴, hydrogen atoms are distributed on two sides of
the oxygen distribution, indicating that a hydrogen-bond net-
work is formed with the surface water molecules. A particu-
larly clear picture of spine hydration is seen in the central
ATAT region 关Fig. 7共c兲兴: Water molecules in the first layer
form bridges between two acceptor atoms 关N共blue兲or O
共red兲兴 in diagonally opposing bases belonging to different
DNA strands, and the second layer of water molecules com-
prises a bridge that connects two neighboring water mol-
ecules in the first layer. Unfortunately, a similarly detailed
hydration picture could not be specified from the RISM cal-
culation 关Fig. 7共b兲兴. This problem may be related to the
broadening of the distribution functions caused by neglecting
the effect of inhomogeneity and employing a relatively large
grid size for the feasible calculations at present. Recently,
Ishizuka et al. have proposed a new integral equation theory
for inhomogeneous molecular fluids.29 We will report further
investigations on the hydration/solvation structures around
biomolecules including inhomogeneity with a finer grid size.
Finally, we mention the behavior of counterions around
DNA. Although ions were not treated as a subject of the
current MD-RISM comparison, they are expected to have a
profound effect on the DNA stability. As previously
recognized,31,32 it is not easy for MD simulation to explore
FIG. 5. 共Color兲Calculated hydration patterns in the mi-
nor groove of the central ATAT region 共a兲, the minor
groove of the CGCG end region 共b兲, the major groove
of the central ATAT region 共c兲.MD共blue兲and RISM
共yellow兲results are given in the left and central views,
respectively. MD and RISM results are superimposed in
the right view.
185102-6 Yonetani et al. J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
the properties of ions because of the increased difficulty with
their configuration sampling. Since the number of ions in-
volved in the calculation is very small compared to the water
number 共e.g., the present MD system has only 39 K+and 17
Cl−in ⬃5000 H2O兲, it becomes difficult to obtain reliable
statistics of ions. Furthermore, ions will not stay at specific
sites. This also makes sampling of the ion configurations
more difficult. Using the present 8 ns trajectories, we calcu-
lated 3D distributions of K+and Cl−ions in the same manner
to that of water molecules. The result was roughly similar to
the recent results31 which were derived from much longer
simulations of ⬃50 ns: Cations preferentially distributed
along the minor groove between the two DNA strands. How-
ever, it is apparent that the time length of 8 ns was too short
to obtain a reliable ion distribution. Because the present
DNA of CGCGATATCGCG has a palindromic sequence, the
FIG. 6. 共Color兲Stereo graphics in which the calculated
results are compared to experimental data. In the upper
共MD; blue兲and bottom 共RISM; yellow兲panels, the re-
spective density thresholds were 2.25 and 1.7. Oxygen
positions determined by x-ray crystallography 共Ref. 30兲
共PDB ID 287d兲are indicated by spheres colored ac-
cording to the experimental B-factors. Only water mol-
ecules located within 5 Å of the DNA surface atoms are
shown; those situated at a short distance 共⬍4Å兲from
other crystal packed DNA molecules are excluded.
FIG. 7. 共Color兲Distribution of water
hydrogen and oxygen. The oxygen
distribution is the same as in Fig. 5
and is colored blue for MD 共a兲and
yellow for RISM 共b兲. The hydrogen
distribution is shown in white with
both MD and RISM. Note that differ-
ent values are used for the density
thresholds in the MD and RISM calcu-
lations 共see the text for the details兲.共c兲
Closer inspection of the MD distribu-
tion in 共a兲. Red and blue spheres are
oxygen and nitrogen atoms in the
DNA bases, respectively, which act as
acceptors for hydrogen atoms in the
first layer water molecules.
185102-7 DNA hydration patterns J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
resultant ion distribution should have a corresponding sym-
metry if sufficient sampling is achieved. Contrary to this ex-
pectation, the present result from the 8 ns trajectories exhib-
ited an explicit deviation from such symmetry. A similar
deviation has been confirmed from a previous short-time
simulation.32 On the basis of the recent thorough
evaluations,31 time length of ⬃50 ns is required to remove
such an undesirable deviation. Accordingly, the time length
required for constructing a satisfactory 3D distribution of
ions will become several times as long as the current study.
In such a case, it would be necessary to carefully consider a
recently mentioned force-field problem on ions.33 A necessity
of readjusting ion parameters has been inferred from a recent
study using ionic solutions. Solving the spatial distribution of
DNA surrounding ions will be a good subject to illustrate the
advantage of the RISM approach, because RISM is free from
the sampling problem unlike MD simulation.
IV. CONCLUDING REMARKS
Using two distinct computational methods 共i.e., MD
simulation and 3D-RISM theory兲, the spatial distributions of
water molecules around a DNA fragment were calculated to
assess the respective capacities of the two methods to simu-
late patterns of biomolecule hydration. This study confirmed
that the results agree well when the solute treatment is con-
sistent. If that is not the case, however, MD and 3D-RISM
will yield differing results. We demonstrated that when as-
suming a solute DNA molecule to be rigid, as in conven-
tional RISM treatments, the solvent distribution near the sol-
ute is about twice as dense as that obtained with the more
realistic MDflex calculation. This suggests that we can never
obtain a true picture of the hydration of solute biomolecules
without considering their flexibility. To cope with this prob-
lem, some improvements in RISM theory based on combi-
nation with the simulation techniques have been proposed
and tested.34 We anticipate that the agreement between the
simulation and theoretical approaches will form the basis for
further methodological development.
The most advantageous aspect of computational ap-
proaches is that they can tackle subjects that are experimen-
tally difficult to observe. A good example is determination of
the positions of the hydrogen atoms of hydrating water,
which is quite difficult to achieve with x-ray crystallography
because of the small atomic scattering factor. By contrast,
with a computational approach, hydrogen atoms are readily
observable. We anticipate that in the near future much infor-
mation on the distribution of water hydrogen will be ob-
tained through neutron diffraction experiments carried out at
the Japan Proton Accelerator Research Complex 共J-PARC兲
facility. We look forward to comparing the present computa-
tional results with those experimental data. Another interest-
ing problem that can be computationally approached is the
sequence-dependent properties of DNA.35 To systematically
investigate the effect of sequence on hydration, measure-
ments from many samples with differing sequences are
needed. Computational approaches are particularly well
suited for such an investigation because they are not subject
to any of the experimental difficulties associated with crys-
tallization. We are currently performing an analysis of
sequence-dependent hydration patterns vis-à-vis the
conformational properties of DNA,36 which will be reported
elsewhere.
ACKNOWLEDGMENTS
This work was supported by a grant from Scientific
Research on Priority Area of “Water and Biomolecules,”
and in part by Grants-in-Aid for Scientific Research No.
18031042 共H.K.兲from Ministry of Education, Culture,
Sports, Science and Technology in Japan. Y.M. and F.H.
were supported by the grant from the Next Generation
Supercomputing Project, Nanoscience Program of the minis-
try. Y.Y. was supported by JSPS Research Fellowships for
Young Scientists.
1T. M. Raschke, Curr. Opin. Struct. Biol. 16,152共2006兲; P. Auffinger and
Y. Hashem, ibid. 17, 325 共2007兲.
2S. Arai, T. Chatake, T. Ohhara, K. Kurihara, I. Tanaka, N. Suzuki, Z.
Fujimoto, H. Mizuno, and N. Niimura, Nucleic Acids Res. 33, 3017
共2005兲.
3M. Nakasako, J. Mol. Biol. 289, 547 共1999兲; N. Niimura, S. Arai, K.
Kurihara, T. Chatake, I. Tanaka, and R. Bau, Cell. Mol. Life Sci. 63,285
共2006兲.
4M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids
共Clarendon, Oxford, 1987兲.
5J.-P. Hansen and I. R. McDonald, Theory of Simple Liquids 共Academic,
New York, 1986兲.
6F. Hirata, Molecular Theory of Solvation 共Kluwer, Dordrecht, 2003兲.
7V. Lounnas, B. M. Pettitt, and G. N. Phillips, Jr., Biophys. J. 66,601
共1994兲; J. Higo, H. Kono, N. Nakajima, H. Shirai, H. Nakamura, and A.
Sarai, Chem. Phys. Lett. 306, 395 共1999兲; J. Higo, H. Kono, H.
Nakamura, and A. Sarai, Proteins 40, 193 共2000兲.
8V. P. Chuprina, U. Heinemann, A. A. Nurislamov, P. Zielenkiewicz, R. E.
Dickerson, and W. Saenger, Proc. Natl. Acad. Sci. U.S.A. 88,593
共1991兲.
9P. Auffinger and E. Westhof, J. Mol. Biol. 300, 1113 共2000兲;305, 1057
共2001兲; M. Feig and B. M. Pettitt, ibid. 286, 1075 共1999兲.
10 C. Chen, B. W. Beck, K. Krause, and B. M. Pettitt, Proteins 62, 982
共2006兲.
11 T. Imai, R. Hiraoka, A. Kovalenko, and F. Hirata, Proteins 66,804
共2007兲.
12 T. Imai, Y. Harano, M. Kinoshita, A. Kovalenko, and F. Hirata, J. Chem.
Phys. 126, 225102 共2007兲; T. Yamazaki, T. Imai, F. Hirata, and A.
Kovalenko, J. Phys. Chem. B 111, 1206 共2007兲.
13 H. R. Drew and R. E. Dickerson, J. Mol. Biol. 151,535共1981兲.
14 P. L. Privalov, A. I. Dragan, C. Crane-Robinson, K. J. Breslauer, D. P.
Remeta, and C. A. S. A. Minetti, J. Mol. Biol. 365,1共2007兲.
15 J. Wang, P. Cieplak, and P. A. Kollman, J. Comput. Chem. 21,1049
共2000兲.
16 T. E. Cheatham III, P. Cieplak, and P. A. Kollman, J. Biomol. Struct.
Dyn. 16, 845 共1999兲.
17 W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L.
Klein, J. Chem. Phys. 79, 926 共1983兲.
18 B. M. Pettitt and P. J. Rossky, J. Chem. Phys. 77,1451共1982兲.
19 D. A. Case, D. A. Pearlman, J. W. Caldwell, T. E. Cheatham III, J. Wang,
W. S. Ross, C. L. Simmerling, T. A. Darden, K. M. Merz, R. V. Stanton,
A. L. Cheng, J. J. Vincent, M. Crowley, V. Tsui, H. Gohlke, R. J. Radmer,
Y. Duan, J. Pitera, I. Massova, G. L. Seibel, U. C. Singh, P. K. Weiner,
and P. A. Kollman, AMBER7, University of California, San Francisco,
2002.
20 J.-P. Ryckaert, G. Ciccotti, and H. J. C. Berendsen, J. Comput. Phys. 23,
372 共1977兲.
21 T. Darden, D. York, and L. Pedersen, J. Chem. Phys. 98, 10089 共1993兲.
22 H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and
J. R. Haak, J. Chem. Phys. 81, 3684 共1984兲.
23 D. Beglov and B. Roux, J. Phys. Chem. B 101, 7821 共1997兲;A.
Kovalenko and F. Hirata, Chem. Phys. Lett. 290,237共1998兲.
24 A. Kovalenko and F. Hirata, J. Chem. Phys. 11 0, 10095 共1999兲.
25 A. Kovalenko and F. Hirata, J. Chem. Phys. 11 2, 10391 共2000兲.
185102-8 Yonetani et al. J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
26 J. S. Perkyns and B. M. Pettitt, J. Chem. Phys. 97, 7656 共1992兲;Chem.
Phys. Lett. 190,626共1992兲.
27 A. Kovalenko and F. Hirata, J. Chem. Phys. 11 2, 10403 共2000兲.
28 A. Kovalenko and F. Hirata, J. Phys. Chem. B 103,7942共1999兲.
29 R. Ishizuka, S.-H. Chong, and F. Hirata, J. Chem. Phys. 128, 034504
共2008兲.
30 M. Shatzky-Schwartz, N. D. Arbuckle, M. Eisenstein, D. Rabinovich, A.
Bareket-Samish, T. E. Haran, B. F. Luisi, and Z. Shakked, J. Mol. Biol.
267,595共1997兲.
31 S. Y. Ponomarev, K. M. Thayer, and D. L. Beveridge, Proc. Natl. Acad.
Sci. U.S.A. 101, 14771 共2004兲; P. Varnai and K. Zakrzewska, Nucleic
Acids Res. 32,4269共2004兲.
32 M. A. Young, G. Ravishanker, and D. L. Beveridge, Biophys. J. 73, 2313
共1997兲.
33 P. Auffinger, T. E. Cheatham III, and A. C. Vaiana, J. Chem. Theory
Comput. 3, 1851 共2007兲.
34 M. Kinoshita, Y. Okamoto, and F. Hirata, J. Am. Chem. Soc. 120, 1855
共1998兲; A. Mitsutake, M. Kinoshita, Y. Okamoto, and F. Hirata, J. Phys.
Chem. B 108, 19002 共2004兲; T. Miyata and F. Hirata, J. Comput. Chem.
29, 871 共2008兲.
35 M. J. Arauzo-Bravo, S. Fujii, H. Kono, S. Ahmad, and A. Sarai, J. Am.
Chem. Soc. 127, 16074 共2005兲; D. L. Beveridge, G. Barreiro, K. Suzie
Byun, D. A. Case, T. E. Cheatham III, S. B. Dixit, E. Giudice, F. Lankas,
R. Lavery, J. H. Maddocks, R. Osman, E. Seibert, H. Sklenar, G. Stoll, K.
M. Thayer, P. Varnai, and M. A. Young, Biophys. J. 87, 3799 共2004兲.
36 Y. Yonetani, H. Kono, S. Fujii, A. Sarai, and N. Go, Mol. Simul. 33,103
共2007兲.
37 D. L. Bergman, L. Laaksonen, and A. Laaksonen, J. Mol. Graphics
Modell. 15, 301 共1997兲.
38 E. F. Pettersen, T. D. Goddard, C. C. Huang, G. S. Couch, D. M.
Greenblatt, E. C. Meng, and T. E. Ferrin, J. Comput. Chem. 25,1605
共2004兲.
185102-9 DNA hydration patterns J. Chem. Phys. 128, 185102 共2008兲
Downloaded 30 May 2008 to 133.48.169.109. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
