You are on page 1of 11

This is an open access article published under an ACS AuthorChoice License, which permits

copying and redistribution of the article or any adaptations for non-commercial purposes.

Article

pubs.acs.org/JCTC

Rational Design of Methodology-Independent Metal Parameters


Using a Nonbonded Dummy Model
Yang Jiang,† Haiyang Zhang,‡ and Tianwei Tan*,†

Beijing Key Lab of Bioprocess, College of Life Science and Technology, Beijing University of Chemical Technology, Beijing 100029,
China

Department of Biological Science and Engineering, School of Chemistry and Biological Engineering, University of Science and
Technology Beijing, 100083 Beijing, China
*
S Supporting Information

ABSTRACT: A nonbonded dummy model for metal ions is


highly imperative for the computation of complex biological
systems with for instance multiple metal centers. Here we
present nonbonded dummy parameters of 11 divalent metallic
cations, namely, Mg2+, V2+, Cr2+, Mn2+, Fe2+, Co2+, Ni2+, Zn2+,
Cd2+, Sn2+, and Hg2+, that are optimized to be compatible with
three widely used water models (TIP3P, SPC/E, and TIP4P-EW).
The three sets of metal parameters reproduce simultaneously the
solvation free energies (ΔGsol), the ion−oxygen distance in the
first solvation shell (IOD), and coordination numbers (CN) in
explicit water with a relative error less than 1%. The main sources
of errors to ΔGsol that arise from the boundary conditions and
treatment of electrostatic interactions are corrected rationally,
which ensures the independence of the proposed parameters
on the methodology used in the calculation. This work will be of great value for the computational study of metal-containing
biological systems.

1. INTRODUCTION coordination geometry simultaneously by introducing a


Metal ions in the biomolecular system have multiple functions 12−6−4 Lennard-Jones (LJ) potential.4 However, the coordi-
and play an important role in the catalytic process.1 For nation numbers (CNs) of V2+, Mn2+, Cd2+, Sn2+, and Hg2+ were
computational studies of metalloenzymes, one of the most not modeled accurately. The nonbonded dummy atom model
challenging tasks is to develop suitable force field para- makes a good balance between the former two models. It
meters for metal ions, especially for divalent metal ions. was originally presented by Åqvist and Warshel,11 and then
Three strategies for metal modeling have been proposed: the improved by many researchers.5−9,13 This model was originally
bonded,2 nonbonded point charge,3,4 and nonbonded dummy designed for the divalent metal ions with CN equal to six.
atom models.5−9 The bonded model has predefined covalent The metal ion has a metal core (MC) surrounded by six dummy
bonds or harmonic restrains between the metal and ligands and atoms (D) to form an octahedron (Figure 1). There are
can reproduce a highly accurate crystal structure. The force field covalent bonds (b0) between MC and the dummy atoms, but
parameters of the metal−ligand complex are usually generated no covalent bond between the dummy atom and the ligand,
and optimized by quantum mechanics (QM) calculations.2 which hence allows ligand exchange. Recently, Duarte et al.8
However, once the ligand is altered, the bonded model pa-
rameters need be reoptimized to match the new metal site.
Moreover, it does not allow ligand exchange during the
simulation. The point charge model is simply described as a
point charge surrounded by van der Waals (vdW) potentials.
Only the parameters of vdW potentials need to be optimized.3,4,10
Although the point charge model allows ligand exchange, it will
be inadequate when the simulation system has a complex metal Figure 1. Geometry of the nonbonded dummy atom model. The
center (two or more metal ions in one active site) or transition metal center (MC) with a charge of n-6δ bond covalently to six
dummy atoms (D) with a charge of + δ to form an octahedron.
metals are present.6,11 Recently, Li and Merz have presented the
Particle Mesh Ewald (PME)12 compatible point charge model
parameters for 16 divalent metal ions, and these parameters Received: February 29, 2016
reproduce the experimental solvation free energy and the Published: May 16, 2016

© 2016 American Chemical Society 3250 DOI: 10.1021/acs.jctc.6b00223


J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

have optimized dummy model parameters of seven divalent we used Duarte’s parameters as the reference.8 The values of
metal ions, including Mg2+, Mn2+, Fe2+, Co2+, Ni2+, Zn2+, Ca2+, force constants (Kb for bonds and Kθ for angles), equilibrium
using the OPLS-AA force field, and Liao et al.9 have provided bond length (b0) and angle (θ0), atomic mass and charge of
the dummy atom model of Cu2+ that includes the Jahn−Teller Duarte’s model were still used here. The vdW parameters
effect. for each metal type were set to be the vdW radius (R) and the
Till now, all the proposed dummy atom models, even most 12−6 LJ potential well depth (ε). The nonbonding interaction
of the nonbonded models, were optimized to reproduce the potential function of two atoms i and j in the AMBER ff03 force
experimental solvation free energies (ΔGsol) provided by field used in the present study has the following form:23
Noyes,14 Rosseinsky,15 or Marcus.16 These three sets of ΔGsol
were derived from the so-called “conventional” values calculated Uijnonbond(rij) = UijvdW(rij) + Uijele(rij)
using the tabulated thermodynamics data from National Bureau ⎡⎛ ⎞12 ⎛ R ij ⎞6 ⎤ qiqj
R ij
= εij⎢⎜⎜ ⎟⎟ − 2⎜⎜ ⎟⎟ ⎥⎥ + k

of Standards (NBS), followed by adding the proton hydration
free energy and then converted to the “absolute” values. The
⎣⎝ ij ⎠ ⎝ rij ⎠ ⎦
r rij
proton hydration free energy cannot be directly measured by (1)
experiments, so different methods were used to obtain this value. where rij is the distance between atom i and j in Å, qi is the
Marcus derived the proton hydration free energy based on the charge of atom i in e, and k is a constant equaling 332. Rij and εij
extra-thermodynamic assumption “TATB hypothesis”,17 which satisfy the Lorentz/Berthelot mixing rules:
is different from the methods reported by Rosseinsky15 and
Noyes,14 leading to deviations between the three sets of ΔGsol. ⎧ 1
Later on, another proton hydration free energy was provided by ⎪ R ij = (R ii + R jj) = R i + R j
⎨ 2
Tissandier et al.18 using the cluster pair approximation method, ⎪ε = ε ε = ε ε
which was considered to be accurate and recommended by ⎩ ij ii jj i j (2)
several reports.19−21 However, no revised experimental solvation
free energies for the divalent metal ions based on Tissandier’s Note that Ri here is the same as Rmin/2 in many other papers.
proton hydration free energy have been reported and used as the 2.2. Experimental Values Used in This Work. The
reference for force field development, except for our previous experimental values of the ion-oxygen distance (IOD) and
work13 in which we demonstrated a remarkable advantage of the coordination number (CN) for each metal were obtained
using the revised experimental solvation free energy for the Mg2+ from several works24−27 (shown in Table S1). The experimental
dummy model in refining the parameter. In the present work, value of the solvation free energy ΔGexp sol was obtained from
the experimental solvation free energies of the divalent metal Marcus’s work16 and then revised by the accurate value of
ions taken from Marcus16 are revised using Tissandier’s proton the proton hydration free energy recommended by many
hydration free energy (details were shown in Supporting reports.18−21 The proton hydration energy (−265.9 kcal/mol)
Information, Table S1). Considering that Cu2+ has the Jahn− has been corrected by converting the 1-atm gas phase standard
Teller effect and has been optimized well before,9 we take state ΔG sol
o
to the 1 M solution standard state ΔG sol *
the remaining 11 ions, namely, Mg2+, V2+, Cr2+, Mn2+, Fe2+, (by subtracting ΔGo→* = 1.9 kcal/mol).19 Note that, Marcus’s
Co2+, Ni2+, Zn2+, Cd2+, Sn2+, and Hg2+, to develop the dummy solvation free energies also have been corrected to the 1 M
atom models for TIP3P, SPC/E, and TIP4P-EW water models, solution standard state, but the correction seems to have the
respectively. wrong direction in his work16 (ΔGo→* was added to ΔGosol).
In addition, a free energy correction is introduced into the Thus, for the absolute solvation free energy provided by Marcus
parameter generation procedure, including the correction for * ), the conventional ionic solvation free energy ΔGconv
(ΔGMarcus sol
use of periodic boundary conditions with the PME method can be derived by
and the correction for an improper summation scheme.22 conv
ΔGsol *
= ΔGMarcus − ΔGo →* − zΔG H (3)
More details on the free energy corrections are presented in
section 2.5. Considering these corrections makes the calculated where ΔG*Marcus is under 1 M solution standard state, z is the net
free energy independent of the simulation methodology such as charge of the ion (2 for divalent ions), and ΔGH is the proton
the finite system, the lattice-sum, and the cutoff-based methods.22 hydration free energy (−252.3 kcal/mol from Marcus). Then
That is, the optimized models in this work can reproduce the our revised metal ionic solvation free energy can be derived
experimental solvation free energies using any of the above three by using the accurate proton hydration free energy (ΔGH =
simulation methodologies just by adding the corresponding −265.9 kcal/mol) and then corrected to the 1 M solution
corrections.22 The major strength of our models is that they can standard state:
reproduce accurate coordination geometries without high-level exp
QM calculations and without additional modifications of the van ΔGsol conv
= ΔGsol + zΔG H − ΔGo →* (4)
der Waals (vdW) potential such as the 12−6−4 LJ potential The revised experimental solvation free energies for the divalent
proposed by Li and Merz,4 especially for V2+, Cd2+, Sn2+, and metal ions are shown in Table S1.
Hg2+, whose CN values cannot be reproduced correctly by a 2.3. General Approach of Molecular Simulations. All
nonbonded model. Unlike the point charge metal parameters simulations were performed using the AMBER14 suite28 with
presented by Li and Merz that are only compatible with the PME the all-atom ff03 force field.29 A 1000 step energy minimization
simulations, the models in this work are in principle compatible was performed to eliminate improper contacts in each system.
with almost all the simulation systems. After energy minimization, each system was heated gradually
from 0 to 298 K. During the heating steps, position restraints
2. MATERIALS AND METHODS were imposed on the metal ions with a force constant of
2.1. Force Field Used in This Work. To generate the 10.0 kcal/mol/Å2. Each system was then equilibrated at the
dummy atom model parameters of the 11 divalent metal ions, constant temperature (298 K) and pressure (1 bar) conditions
3251 DOI: 10.1021/acs.jctc.6b00223
J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

via the Langevin dynamics (the collision frequency is 1.0 ps−1), equilibration simulation; RI is the ionic radius, which is defined
with a coupling constant of 0.2 ps for both parameters. All to be the distance value where the ion-oxygen radial distribution
the production simulations were performed at 298 K and 1 bar function (RDF) begins to accumulate;34 ξLS is a constant and
using a time step of 2 fs. Electrostatic interactions were approximately equals to −2.837297.22,33,34
calculated using the PME algorithm. The cutoff distance for van The second term ΔGϕcorr is the correction mainly for the
der Waals interactions was 10.0 Å. The SHAKE algorithm30 error in the potential at the ionic site due to its evaluation
was applied to the bonds involving hydrogen. using an improper summation scheme.22,35 In this study, we
2.4. Two-Step Thermodynamic Integration. The used the formula provided by Kastenholz et al.22 to estimate
solvation free energy ΔGsol of metal ion M2+ was calculated this term:
using the two-step thermodynamic integration (TI) method.
The first step was perturbing nothing to the metal atom M ⎛ 4πRI 3 ⎞ ργ 4πRI 3
ϕ
ΔGcorr = −q⎜1 − ⎟ − q (16.8 − 26/RI)
(with no charge) in water using the soft-core potential ⎝ 3L3 ⎠ 6ε0 3L3
(“appearing” part) in Amber11 suite.31 For the parameter (6)
dVSC
generation, dλ
was calculated from five independent −3
where ρ is the solvent number density (0.033 Å for all the TI
simulations at λ = 0.04691, 0.23076, 0.50000, 0.76923, and simulation systems); γ is the quadrupole-moment trace of the
0.95308, including a 400 ps energy equilibration and a 100 ps solvent molecule:22
dV
production. For the model testing, dλSC was calculated from Ns

nine independent simulations at λ = 0.01592, 0.08198, 0.19331, γ= ∑ qiri 2


i (7)
0.33787, 0.50000, 0.66213, 0.80669, 0.91802, and 0.98408. The
dV
parameters of soft-core potential were set to default. dλSC was where qi is the partial charge of atom i in the solvent molecule;
obtained from the production simulation. The free energy ri is the distance between the M site and atom i of the solvent
change ΔGvdW SC
was then calculated using the Gaussian molecule; Ns is the total atom number of the solvent molecule.
dV The quadrupole-moment traces for the three different water
quadrature formulas. The dλSC plots for the metal ions cal-
culated using the final parameters for TIP3P water were shown models used in this work were shown in Table S2. Note that
in Figure S1. Others are similar to these and not shown. the last term of ΔGϕcorr corrects the error due to the possible
The second step was charging the metal M from 0 e to +2 e presence of a constraint of vanishing average potential over
dU the computational box and depends on the potential jump at
by a parameter λ. For the parameter generation, dλ was the ion−solvent interface (for the PBC system with the LS
calculated from five independent simulations at λ = 0.04691, scheme).22,35,36 It has a negligible magnitude because of RI ≪ L
0.23076, 0.50000, 0.76923, and 0.95308, including a 400 ps in the present case. We still retain this negligible term here to
energy equilibration and a 100 ps production. For the model keep consistent with the previous work.13
dU
testing, dλ was calculated from nine independent simulations The correction technology for the charging free energy of the
at λ = 0.01592, 0.08198, 0.19331, 0.33787, 0.50000, 0.66213, multiatomic system has been proposed by Reif et al.21 and
dU
0.80669, 0.91802, and 0.98408. dλ was obtained from the Rocklin et al.,37 which can only be estimated through numerical
methods. Because of the small volume with symmetrical charge
production simulation. The free energy change ΔGMTI0→M2+
distribution of the dummy model, the shift in ΔGPBC corr due to the
was then calculated using the Gaussian quadrature formulas.
dU different charge distribution of our dummy model compared
The dλ plots for the metal ions calculated using the final with the point charge model can be ignored. Moreover, because
parameters for TIP3P water were shown in Figure S1. Others of the eliminated self-potential, ΔGPBCcorr is less than 2 kcal/mol
are similar to these and are not shown. with the ionic radius less than 2.5 Å, leading the small effect of
2.5. Free Energy Correction. The correction for the raw the ΔGPBC corr . Therefore, although the dummy model contains
ionic charging free energies has been discussed and provided more than one charged atom, the above correction technology
by many reports.21,22,32−34 Two different correction terms for the monatomic ion, as a fast estimation, is still adapted to
should be added to the calculated solvation free energy. our dummy model.
The first term ΔGPBC corr is the correction for using the periodic Note that there is no need to add the correction for the
boundary condition (PBC) with the PME approach. PME is truncated vdW interaction, because Amber has already added
one of the most popular lattice-sum (LS) methods. For the this correction into its calculation protocol (see the keyword
PME approach, the ion−ion self-potential is already included in “vdwmeth” in Amber’s manual, http://ambermd.org/doc12/
the instantaneous electrostatic potential of the ion.33 This is dif- Amber14.pdf). Note that there is also no need to add the
ferent from the general LS method. So the ion−ion self-potential correction for standard state ΔGo→* here, because our revised
term should be eliminated from the general finite-size correction. experimental solvation free energy has been already corrected
Thus, the correction for using PBC with the PME approach is into the 1 M solution standard state. Therefore, the solvation
given by free energy was finally calculated by
PBC q2 ξLS cal
ΔGsol SC
= ΔGvdW TI
+ ΔG M PBC
+ ΔGcorr ϕ
+ ΔGcorr
ΔGcorr =− 0
→ M2 + (8)
8πε0εs L
2.6. Parameters Generation Process. The vdW param-
q2 1 − εs−1 ⎡ 4π ⎛ RI ⎞2 16π 2 ⎛ RI ⎞5⎤ eters R and ε for each divalent metal ion and each water
+ ⎢ ⎜ ⎟ − ⎜ ⎟ ⎥
8πε0 L ⎣ 3 ⎝L⎠ 45 ⎝ L ⎠ ⎦ (5) model were generated by following the process similar to that
provided by Li et al.3 The parameter space was defined to be
where q is the ionic charge (+2 e); εs is the solvent permittivity R ∈ [0.30 Å, 1.50 Å] and ε ∈ [10−3 kcal/mol, 200 kcal/mol],
(78.0 for water); L is the box boundary length estimated from the and then divided into forty-two discrete points, which were
3252 DOI: 10.1021/acs.jctc.6b00223
J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

the combinations of R = 0.30, 0.50, 0.75, 1.00, 1.25, 1.50 Å and combination that can both reproduce the experimental solvation
ε = 10−3, 10−2, 10−1, 1, 10, 100, 200 kcal/mol. At each point, free energy and the coordination geometry.
the dummy atom model was built with the specific vdW 2.7. Model Testing. For each of the parameters generated
parameter combinations. The mass of the metal center of each from the previous section, the relative errors of the reproduced
dummy atom model was set to be 6.3 (the choice of mass has ΔGsol and IOD values related to the respective experimental
a limited influence on the simulated solvation free energy and values were calculated after longer and more careful MD and TI
the coordination geometry of the metal ion−water system3). simulations than those in the parameter generation process. For
The dummy atom model at each point was then solvated in MD simulations, 1000-step energy minimization, 60 ps heating,
three cubic boxes with a total of 1075 water molecules (the 500 ps equilibration, and 10 ns production were performed. For TI
box length L is 31.65 Å) parametrized by TIP3P, SPC/E, and simulations, more λ values were used as described in section 2.4.
TIP4P-EW model, respectively. IODRDF were calculated using the 10 ns production trajectories.
The solvation free energy ΔGcalsol for each system was cal- To test other relevant properties that were not directly
culated using the accurate “two-step” thermodynamic integra- parametrized, we calculated the water exchange rate of the
tion (TI) and then added by the free energy corrections. The present Mg2+ model and compared it with those provided by
CN value was defined to be the integral value (N[g(r)]) of the Allnér,39 Duarte,8 and Li and Merz.4 The Li and Merz 12−6−4
metal−oxygen radial distribution function (g(r)) from 0 Å to potential parameters were generated using AmberTools15,40
the distance value (dm) at the minimum between the two peaks in which the bug in calculating the 12−6−4 parameter in
of the RDF: AmberTools14 has been fixed. To calculate the water exchange
rate, we followed the procedure described by Allnér et al.39 The
CNRDF = ∑ Δn(rj) brief calculation scheme is as follows: The potential of mean
0 ≤ r j ≤ dm (9)
force (PMF) was calculated along the reaction coordinate,
which is the distance (x) between the metal ion and the water
where rj is the MC-O distance for the RDF; Δn(rj) is the oxygen atom from 1.6 to10.0 Å, using the umbrella sampling.
unnormalized oxygen atom number on the sphere surface The interval between 1.6 and 6 Å was 0.1 Å with a force
with rj as the radius. The RDF was calculated using the constant of 150 kcal/(mol·Å2), and the interval between 6 and
trajectory produced by a 1 ns production MD simulation with 10 Å was 0.5 Å with a force constant of 10 kcal/(mol·Å2),
the interval 0.01 Å. The IOD value for each system was leading to a total of 53 simulation windows. For each window,
derived by fitting the first pick of RDF using the quadratic the simulation included 1000-step energy minimization with a
function and described in detail by Li et al.3 The IOD value force constant of 500 kcal/(mol·Å2), 100 ps equilibration, and
was then defined to be the distance at the first peak of the 1 ns production. The PMF curves were then generated using the
RDF and named IODRDF. All the scanned data are shown in weighted histogram analysis method41 with a tolerance of 10−5. An
Tables S3−S5. entropic contribution, S = 2 RT ln(x), to the average constraint
After the three quantities for all the discrete points within the force due to the free rotation of the solute−solute connecting
parameter space were calculated, the surface of the ΔGsol and vector was subtracted out from the PMF values.39 The rate
the IODRDF can be reproduced by using the 2D cubic spline constant k was then calculated according the following equation:39
interpolation in MATLAB.38 For each metal ion, the contour †
k = Ae−ΔG / RT (10)
curves which represent the values equaling the experimental †
ΔGsol value and the experimental IOD value can be both found where ΔG is the free energy of activation estimated from the
and then plotted on one graph with −lg(ε) as the horizontal PMFs as the energy difference between the global minimum and
axis and R as the vertical axis. The intersection point of these the global maximum, R is the ideal gas constant, and T is the
two contour curves was then searched within the parameter temperature. A is the pre-exponential factor and calculated by
space by finding the zero point for the difference function 1 E″
of the two contour curves fitted using the 1D cubic spline inter- A=
polation in MATLAB. The intersection was the target parameter 2π μ (11)

Figure 2. ΔGsol (a) and IOD (b) surfaces for the parameter generation with TIP3P water model. The parameter space is shown as the x−y plane
with R ∈ [0.3000, 1.5000] and −lg(ε) ∈ [−2.3010, 3.0000].

3253 DOI: 10.1021/acs.jctc.6b00223


J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

where E″ is the second derivative of the PMF calculated by of Duarte et al.8 We chose the E. coli Glyoxalase I (GlxI,
fitting a quadratic function to the bottom data points, and μ is PDBID: 1F9Z42) to perform our testing. The origin metal ion
the reduced mass of the atom pair.39 was Ni2+ and we then replaced it with Co2+, Cd2+, and Zn2+.
To demonstrate the simulation performance of our para- The ion parameters of Li-Merz 12−6−44 were also tested.
meters in an actual biomolecule system, we followed the study The MD simulation setup can be found in section 2.3, and the
12−6−4 parameters were generated using AmberTools15.40
Note that, in this study, we did not modify the partial charge
on the coordinated oxygen atoms of Glu56 and Glu122. Each
simulation included a 5000-step energy minimization, 60 ps
heating, 500 ps equilibration, and 20 ns production.
2.8. Coordination Number Optimization. For the
models that failed to reproduce the experimental CN values,
we reoptimized the MC-D bond length (b0) and the partial
charge on D (δ). First, the b0-δ parameter space was scanned at
b0 = 0.8, 0.9, 1.0, 1.1, 1.2, and 1.3 Å, and δ = 0.3, 0.4, 0.5, 0.6,
and 0.7 e, with several fixed R and ε values in the TIP3P water.
We found that increasing the b0 and δ values can increase the
IOD threshold that determines the CN value to be six or not.
Hence, the b0-δ combinations that show the IOD threshold
exceeding the target experimental IOD values were selected.
Figure 3. Intersection points of contour curves for the parameter They are b0 = 1.1 Å and δ = 0.7 e for Cd2+, b0 = 1.3 Å and
generation with TIP3P water model. The parameter space is shown as
the x−y plane within R ∈ [0.3000, 1.5000] and −lg(ε) ∈ [−2.3010,
δ = 0.7 e for Sn2+, and b0 = 1.2 Å and δ = 0.6 e for Hg2+. Second,
3.0000]. The contour curves of ΔGsol and IODRDF are shown by solid the same parameter generation procedure was applied on these
and dotted lines, respectively. ΔGsol contour curves of Hg2+ and Mn2+ three b0-δ combinations with TIP3P, SPC/E, and TIP4P-EW
are overlapped because of the same free energy values, and thus only water models. For Cd2+, the R-ε space was defined to be
the curve for Mn2+ is given here. R ∈ [0.75 Å, 1.50 Å] and ε ∈ [0.1 kcal/mol, 100 kcal/mol].

Table 1. Reproduced Coordination Geometries and Solvation-Free Energiesa of Models for the 11 Divalent Metal Ions Selected
from the Parameter Generation Process for TIP3P Water (Parameters with Bad Performances Were Marked in Bold)
revised ΔGexp
sol IODexp IODRDF RI ΔGTI
M0→M2+ ΔGSC
vdW ΔGcorr ΔGcal
sol

metal ions kcal/mol Å CN Å CNRDF Å kcal/mol


Mg2+ −468.2 2.090 6 2.088 6.0 1.93 −433.68 0.70 −33.96 −466.9
V2+ −467.0 2.210 6 2.189 6.0 2.08 −412.53 −23.06 −33.90 −469.5
Cr2+ −473.0 2.080 6 2.079 6.0 1.94 −437.69 0.49 −33.95 −471.2
Mn2+ −451.5 2.192 6 2.186 6.0 2.07 −408.84 −10.91 −33.90 −453.6
Fe2+ −470.6 2.114 6 2.113 6.0 1.99 −431.23 −5.01 −33.93 −470.2
Co2+ −488.5 2.106 6 2.103 6.0 1.96 −440.35 −15.28 −33.94 −489.6
Ni2+ −504.1 2.061 6 2.065 6.0 1.93 −452.48 −14.98 −33.96 −501.4
Zn2+ −498.1 2.098 6 2.096 6.0 1.98 −444.77 −19.05 −33.94 −497.8
Cd2+ −450.3 2.301 6 2.279 6.6 2.12 −394.25 −22.77 −33.88 −450.9
Sn2+ −387.0 2.620 6 2.594 8.6 2.39 −338.27 −16.38 −33.77 −388.4
Hg2+ −451.5 2.410 6 2.388 7.9 2.21 −382.48 −36.02 −33.85 −452.4
a
The detail values of free energy calculation and correction are shown in the table. RI is the ionic radius that is derived from the RDF plot and used
ϕ
to calculate the free energy correction. ΔGcorr is the sum of ΔGPBCcorr and ΔGcorr calculated using eq 5 and 6.

Table 2. Reproduced Coordination Geometries and Solvation Free Energiesa of the CN-Optimized Models for Cd2+, Sn2+,
and Hg2+
revised ΔGexp
sol IODexp IODRDF RI ΔGTI
M0→M2+ ΔGSC
vdW ΔGcorr ΔGcal
sol

metal ions kcal/mol Å water model Å CNRDF Å kcal/mol


Cd2+ −450.3 2.301 TIP3P 2.301 6.0 2.16 −412.57 −4.42 −33.35 −450.3
SPC/E 2.300 6.0 2.16 −412.44 0.45 −37.17 −449.2
TIP4P-EW 2.301 6.0 2.18 −401.27 −7.81 −41.61 −450.7
Sn2+ −387.0 2.620 TIP3P 2.623 6.0 2.50 −340.02 −13.96 −33.09 −387.1
SPC/E 2.611 6.0 2.47 −343.46 −7.27 −36.93 −387.7
TIP4P-EW 2.610 6.0 2.47 −333.33 −12.46 −41.38 −387.2
Hg2+ −451.5 2.410 TIP3P 2.411 6.0 2.29 −399.99 −18.84 −33.26 −452.1
SPC/E 2.410 6.0 2.27 −402.18 −13.69 −37.09 −453.0
TIP4P-EW 2.415 6.0 2.30 −384.75 −26.95 −41.52 −453.2

a
The detail values of free energy calculation and correction are shown in the table. RI is the ionic radius that is derived from the RDF plot and used
ϕ
to calculate the free energy correction. ΔGcorr is the sum of ΔGPBCcorr and ΔGcorr calculated using eq 5 and 6.

3254 DOI: 10.1021/acs.jctc.6b00223


J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

For Sn2+, the R-ε space was defined to be R ∈ [1.25 Å, 1.75 Å]


and ε ∈ [0.1 kcal/mol, 100 kcal/mol]. For Hg2+, the R-ε space
was defined to be R ∈ [0.75 Å, 1.50 Å] and ε ∈ [0.1 kcal/mol,
200 kcal/mol]. Note that the water box here was smaller than
that in the original procedure in section 2.6 and contained
580 water molecules (the box length L is 26.03 Å) in order to
speed up the parameter generation procedure. Finally, all
the parameters for each water model and each metal ion were
optimized and tested to check the simulation performance.
2.9. Model Refining. For the bad models, whose relative
errors of the reproduced ΔGsol and IOD values exceed 1%,
a parameter refining process should be performed on them.
The parameter refining process was described in detail in our
previous work.13 The process starts from defining an initial
hunting zone according to the initial parameter combination.
Through step 1, parameters that can produce IOD values all
close to the experimental IOD value after energy minimizations
are retained. Through step 2, parameters that can the produce
IODRDF value close to the experimental IOD value after 10 ns
MD simulations are retained. Through step 3, parameters
that can produce a solvation free energy close to the revised
experimental value can be retained. The parameter combination
with the closest calculation values to the experimental values is
finally selected to terminate the process; otherwise a narrower
hunting zone according to the eligible parameter combinations
produced by step 2 is defined and a new iteration starts. Here,
we used these bad model parameters as the initial values to start
the process. The threshold values for step 1 and 2 were set
to be 0.01 Å for the first iteration. Fortunately, the refining
processes found the optimal parameters after one step iteration.

3. RESULTS
3.1. Parameter Development. For the development
procedure of the force field parameters, the vdW parameter
Rmin/2 (written as R for convenience hereinafter) and ε for
each model need to be optimized; the other parameters were
taken from our previous work13 (note that the MC-D bond
length b0 was 0.9 Å and the partial charge δ was 0.5 e). The
solvation free energies (ΔGsol), ion-oxygen distances in the first
hydration shell (IOD), and CN values for the parameter
combinations at the discrete points within the parameter space
were calculated. The ΔGsol surface and the IOD surface for
each water model were both generated, and the two surfaces for Figure 4. Relative errors of the calculated values of IODRDF (red) and
solvation free energy (blue) for the 11 divalent metal ions simulated
the TIP3P water model were shown in Figure 2; the surfaces using the parameters for (a) TIP3P water, (b) SPC/E water, and
for SPC/E and TIP4P-EW water models are given in Figure S2 (c) TIP4P-EW water models before being refined. The experimental
and Figure S3, respectively. The ΔGsol surface for the TIP4P- values of IOD and solvation free energy of the 11 divalent metal ions
EW model showed slightly higher values than those for the (can be found in Table 1) are used as the standard. The threshold
TIP3P and SPC/E models. All the IOD surfaces display similar value 1% is represented by a dotted line in each plot. For TIP4P-EW,
landscapes. Contour curves that represent the experimental the relative errors of IODRDF for V2+ and Ni2+ exceed the threshold,
values of the ΔGsol and the IOD values of the 11 divalent metal leading us to perform a parameter refining process for these two metal
ions were then plotted on the parameter space, and the one for ions with TIP4P-EW water.
TIP3P water was shown in Figure 3; the curves for SPC/E
and TIP4P-EW water models are given in Figure S4. The six-coordinated divalent metal ions with the IOD values
intersection points of the ΔGsol and the IOD contour curves exceeding 2.30 Å. The CN values for the metal ions with
were clearly marked and then transformed to the parameters of IOD lower than 2.30 Å were all reproduced accurately. Now,
the 11 divalent metal ions. the IOD threshold that determines the CN value to be six was
The generated parameters were then tested to produce the 2.30 Å. The b0-δ space with several fixed R and ε values were
ΔGsol, IOD and CN values by more careful calculations then scanned to find appropriate b0-δ combinations that can
described in Section 2.7. The above values are shown in Table 1 keep CN being six for Cd2+, Sn2+, and Hg2+. It was found
for TIP3P water, Table S6 for SPC/E water, and Table S7 that increasing b0 and δ values can increase the above IOD
for TIP4P-EW water. For the current parameters, the CN threshold. Finally, three different b0-δ combinations were found:
values of Cd2+, Sn2+, and Hg2+ were all produced incorrectly, b0 = 1.1 Å and δ = 0.7 e for Cd2+, b0 = 1.3 Å and δ = 0.7 e
indicating somewhat a failure of reproducing CN values for the for Sn2+, and b0 = 1.2 Å and δ = 0.7e for Hg2+. For each
3255 DOI: 10.1021/acs.jctc.6b00223
J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

Table 3. Final Optimized Force Field Parameters of the 11 Divalent Metal Ions for TIP3P, SPC/E, and TIP4P-EW Water
Models
Force Field Parametersa
b0
bond typeb Kb Cd Sn Hg others
MCDi 800.0 1.1 1.3 1.2 0.9
angle typec Kθ θ0
DiMCDi 125.0 180.0
DiMCDj≠i 125.0 90.0
Mass, Charge and Nonbonding Parametersd
TIP3P SPC/E TIP4P-EW
atom type mass charge R ε R ε R ε
MC Mg 6.31 −1.00 0.9203 0.2394 1.2392 0.0096 0.6036 5.6131
V 32.94 −1.00 0.6775 93.2484 0.7010 88.2098 0.6360 144.5440
Cr 34.00 −1.00 0.8480 0.3014 1.1758 0.0116 0.5274 9.9596
Mn 36.94 −1.00 0.7646 25.6237 0.8147 14.5482 0.6966 43.1992
Fe 37.85 −1.00 0.6857 9.6775 0.8254 1.2306 0.5568 24.5709
Co 40.93 −1.00 0.5348 64.8829 0.5651 34.4805 0.4734 107.8194
Ni 40.69 −1.00 0.4152 83.3241 0.3997 34.5862 0.3280 201.3724
Zn 47.39 −1.00 0.4895 100.7481 0.5004 57.3981 0.4523 187.4485
Cd 94.40 −2.20 1.0228 4.9444 1.2711 0.3978 0.8844 11.6724
Sn 100.70 −2.20 1.4313 12.9508 1.5131 5.3147 1.3786 12.2698
Hg 182.60 −2.20 0.9903 41.4921 1.0285 22.9838 0.9106 74.5833
atom type mass charge R ε
Di Cd 3.00 0.70 1.3882 1 × 10−8
Sn 3.00 0.70
Hg 3.00 0.70
Others 3.00 0.50
a
The atom types of the metal core and the dummy atom is MC and Di. The dummy atoms have three types and named D1, D2, and D3. bUb = Kb(b − b0)2,
where Kb is in kcal·mol−1 Å−2 and b0 is in Å. cUθ = Kθ(θ − θ0)2, where Kθ is in kcal·mol−1 rad−2 and θ0 is in degrees. dThe nonbonding parameters
R and ε follow the Lorentz/Berthelot combination rule. The unit of R, i.e. Rmin/2 in many other papers, is Å. The unit of ε is kcal/mol.

combination, the R and ε values were reoptimized for TIP3P, discussed in detail. Here, we performed simple tests for the
SPC/E, and TIP4P-EW water models through the same present dummy atom models to show their simulation
parameter generation procedure described above. All the performance of reproducing the water exchange rate constant
new scanned models for Cd2+, Sn2+, and Hg2+ can correctly and the geometry of a metal center in a biological system.
reproduce the experimental values of CN, as well as the ΔGsol The water exchange rate constants of Mg2+ calculated using
and the IOD values (see Table 2). the present dummy model, the Duarte dummy model, the
After CN optimization, for the two parameter sets of TIP3P Allnér model, and the Li and Merz 12−6−4 model were shown
and SPC/E, the relative errors for ΔGsol and IOD are all less in Table 5. The calculated water exchange rate constants were
than 1% (Figure 4a and Figure 4b), indicating a remarkable compared with the experimental value (6.7 × 105 s−1) provided
accuracy of ΔGsol and IOD surfaces from the parameter genera- by Bleuzen et al.43 The comparison indicates that the dummy
tion process. However, for the TIP4P-EW water, the relative atom models (the present model and the Duarte model) could
errors of IOD for V2+ and Ni2+ exceed 1% (Figure 4c), estimate a higher energy barrier (ΔG†), leading to a lower
indicating that some deviations exist on the surfaces around the water exchange rate constant. On the contrary, the Li and Merz
points of V2+ and Ni2+. We further performed the parameter 12−6−4 models could estimate a lower energy barrier (ΔG†),
refining process proposed in our previous work13 for V2+ and leading to a higher water exchange rate constant. The Allnér
Ni2+ with TIP4P-EW water to find the optimal parameter model, which is exactly optimized for reproducing the water
combinations. The parameter combinations for V2+ and Ni2+ exchange rate, produced the best result (6.87 × 105 s−1).
produced by the parameter generation process for TIP4P-EW Among the remainder of the models, the Duarte model for
water were already close to the optimal points and were TIP3P water (3.65 × 104 s−1), the present model for TIP4P-
therefore used as initial values of the parameter refining process. EW water (4.36 × 104 s−1), and the Li and Merz 12−6−4
After only one round of iteration, the optimal parameter model for SPC/E water (5.18 × 106 s−1) produced better
combinations were found. The details were given in Table S8. results than the others.
At this time, all the parameters of the 11 divalent metal ions for For testing in a biology system, we followed the work
the three water models were completely developed (given in provided by Duarte et al.,8 where their parameters were tested
Table 3). They can accurately reproduce the experimental ΔGsol in E. coli Glyoxalase I (GlxI). Here, we tested the present
and IOD values with relative errors less than 1%. All these parameters of Ni2+, Zn2+, Co2+, and Cd2+ with TIP3P water in
models can reproduce the experimental CN values to be six. the E. coli. GlxI system. All the tested models show an excellent
3.2. Parameter Testing. In our previous work,13 the performance of reproducing the metal center geometry (except
advantage and drawback of the dummy atom model have been the CN for Zn2+, see Table S9a for Ni2+, Table S9b for Co2+,
3256 DOI: 10.1021/acs.jctc.6b00223
J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

Table S9c for Cd2+, and Table S9d for Zn2+): our calculated
average distances are mostly within 0.1−0.2 Å of the
experimental value. Moreover, unlike the necessary modifica-
tion of the partial charge on the coordinated oxygen atoms of
Glu residues (otherwise, Glu will artificially doubly coordinate
to the metal center by using Duarte parameters) performed in
the Duarte’s work8, the present models produced the metal
center with monodentate Glu residues without the modifica-
tion. We also tested the simulation performance of Li−Merz
12−6−4 parameters4 in the E. coli GlxI system. Without the
modification of the partial charge on the coordinated oxygen
atoms of Glu residues, the metal center showed a similar
geometry with that produced by the present model for
Ni2+, Co2+, and Zn2+, indicating that the Li−Merz 12−6−4
parameters can reproduce the metal center geometry at the
same level of the present model, which can also be achieved
by taking into account the side-chain polarization when using
the Duarte model. However, because of the poor performance
of reproducing the CN in water (CN = 7.5−7.8, shown in
Figure 5b), the Li−Merz 12−6−4 model for Cd2+ cannot
produce the correct CN in the E. coli GlxI system, even when
we modified the Glu side-chain charge distribution as that
performed in the work of Duarte8 (shown in Table S9c).
Note that similar to the results shown in Duarte’s work, the
substitution of the original Ni2+ with Zn2+ in our test also failed
to produce the reported metal center geometry with five ligands
(PDBID: 1FA5)42 during the simulation time, indicating the
complex deactivation mechanism of metal substitution is still
difficult to be studied using such MD simulations.

4. DISCUSSION Figure 5. (a) Relative errors of the calculated values of IODRDF for
Our dummy atom models have several advantages. Although Mg2+, Mn2+, Fe2+, Co2+, Ni2+, Zn2+ simulated with the parameters for
the present model is developed based on the Duarte model,8 TIP3P water from Duarte’s work (blue) and the present work (red);
the improvements can be clearly pointed out: (I) The present (b) calculated CN values for V2+, Mn2+, Cd2+, Sn2+, Hg2+ simulated
with the parameters for TIP3P, SPC/E, TIP4P-EW, water models
work used the two-step TI calculation for the solvation free
from Li and Merz and the present work.
energy, leading to the calculated vdW part within the range
from −36 kcal/mol to +2 kcal/mol. This is more accurate than
those in the Duarte work, where the vdW part (called “cavity” present model cannot accurately reproduce the water exchange
term) was set to be a constant value of +2.5 kcal/mol.8 (II) The rate constant. However, because of the failure in reproducing
present parameter sets for Cd2+, Hg2+, and Sn2+ have different the rate constant using all the tested models except for the
charge distributions and different geometries with the original Allnér model, the present model for TIP4P-EW water can be
Duarte dummy models.8 Moreover, the relative errors of the acceptable after a fashion, as well as the Duarte model for TIP3P
most IOD values in the Duarte work8 (for TIP3P water) are water and the Li and Merz 12−6−4 model for SPC/E water.
bigger than those in this work (see Figure 5a), indicating an People who want to use the present models to calculate the water
improved accuracy of our parameter sets generated by the exchange rate constant must take caution.
comprehensive search within the parameter space. For the pa- By testing several metal ions in the biomolecule system,
rameters developed by Li and Merz,4 the relative errors of E. coli GlxI, we demonstrated the actual simulation performance
CN values are bigger than those in the present work (see of the present models. Similar with the results in the previous
Figure 5b), especially for V2+, Mn2+, Cd2+, Sn2+, and Hg2+. work,13 the present models can keep the monodentate
For our final optimized parameters, the CN values were all coordination state of Glu side chains without modification of
produced correctly, indicating a remarkable advantage of the charge on the coordinated oxygen atoms of Glu residues,
reproducing CN values for the six-coordinated divalent metal which cannot be achieved by directly using the Duarte model.
ions. But for the point charge models provided by Li and Merz, The Li−Merz 12−6−4 model can reproduce the metal center
the CN values of the six-coordinated divalent metal ions can geometry at the same level of the present model (except for the
be accurately reproduced only for Mg2+, Cr2+, Fe2+, Co2+, Ni2+, tested Cd2+ parameters), which can also be achieved by taking
and Zn2+.4 The produced ΔGsol for the dummy atom models in into account the side-chain polarization when using the Duarte
the present work have a slightly larger deviation (unsigned model. This indicates that the present model has the proper
average error is between 1.1 and 1.3 kcal/mol, shown in Table 4) effect, which is equivalent to the Li−Merz 12−6−4 model
than those for the models of Li and Merz (unsigned average with the ion-induced dipole interaction4 and the Duarte
error is about 0.4 kcal/mol);4 but all the relative errors of our model taking into account the side-chain polarization,8 on the
models were controlled under 1%. octahedral metal center. However, for V2+, Mn2+, Cd2+, Sn2+,
Furthermore, by testing the Mg2+ parameters to calculate and Hg2+ in the Li−Merz 12−6−4 model, their calculated
the water exchange rate constant, we demonstrated that the geometries in the biomolecule may be incorrect because of
3257 DOI: 10.1021/acs.jctc.6b00223
J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

Table 4. Simulated IOD, CN, and ΔGsol for the Final Optimized Parameter Sets in Table 3
TIP3P SPC/E TIP4P-EW
IOD (Å) CN ΔGsol (kcal/mol) IOD (Å) CN ΔGsol (kcal/mol) IOD (Å) CN ΔGsol (kcal/mol)
Mg 2+
2.088 6.0 −466.9 2.088 6.0 −471.6 2.092 6.0 −465.7
V2+ 2.189 6.0 −469.5 2.207 6.0 −468.6 2.202 6.0 −469.2
Cr2+ 2.079 6.0 −471.2 2.079 6.0 −474.7 2.079 6.0 −472.5
Mn2+ 2.186 6.0 −453.6 2.193 6.0 −452.4 2.189 6.0 −453.5
Fe2+ 2.113 6.0 −470.2 2.109 6.0 −469.9 2.109 6.0 −470.7
Co2+ 2.103 6.0 −489.6 2.107 6.0 −488.9 2.106 6.0 −488.6
Ni2+ 2.065 6.0 −501.4 2.054 6.0 −502.4 2.051 6.0 −506.8
Zn2+ 2.096 6.0 −497.8 2.093 6.0 −497.6 2.108 6.0 −498.1
Cd2+ 2.301 6.0 −450.3 2.300 6.0 −449.2 2.301 6.0 −450.7
Sn2+ 2.623 6.0 −387.1 2.611 6.0 −387.7 2.610 6.0 −387.2
Hg2+ 2.411 6.0 −452.1 2.410 6.0 −453.0 2.415 6.0 −453.2
average error −0.003 0 0.0 −0.003 0 −0.6 −0.002 0 −0.6
SD 0.007 0 1.6 0.003 0 1.5 0.006 0 1.5
unsigned average error 0.004 0 1.2 0.003 0 1.3 0.005 0 1.1

Table 5. Water Exchange Rate Constantsa (k) of Mg2+ Calculated Using the Present Dummy Model, The Duarte Dummy
Model,8 the Allnér Model,39 and the Li and Merz 12−6−4 Model4
Mg2+ model water model A (s−1) ΔG⧧ (kcal/mol) k (s−1)
Allnér TIP3P 1.68 × 1013
10.07 6.87 × 1005
this work TIP3P 1.99 × 1013 15.51 8.35 × 1001
Duarte TIP3P 1.90 × 1013 11.88 3.65 × 1004
Li and Merz 12−6−4 TIP3P 1.71 × 1013 7.04 1.19 × 1008
this work SPC/E 2.03 × 1013 15.93 4.23 × 1001
Li and Merz 12−6−4 SPC/E 1.71 × 1013 8.89 5.18 × 1006
this work TIP4P-EW 1.82 × 1013 11.75 4.36 × 1004
Li and Merz 12−6−4 TIP4P-EW 1.77 × 1013 7.84 3.16 × 1007
expb 9.9 6.70 × 1005
a
The water exchange rate constant is calculated according the scheme described in section 2.7. bThe experimental value of the water exchange rate
constant of Mg2+ is provided by Bleuzen et al.43

their bad performance of reproducing the CN in water4 octahedral dummy model is inadequate to simulate the metal
(shown in Figure 5b). People who want to use V2+, Mn2+, ions with the CN less than 6 (such as Be2+ with a CN of 4, or
Cd2+, Sn2+, and Hg2+ in the Li−Merz 12−6−4 model must take Zn2+ with a CN of 5).
caution.
The usage of our models is convenient as well. With the 5. CONCLUSIONS
preparation file (.prepi) and the force field modification file This work provided three sets of nonbonded dummy atom
(.frcmod) of the Amber force field provided in the Supporting model parameters of 11 common divalent metal cations (Mg2+,
Information, the dummy atom model can be directly loaded V2+, Cr2+, Mn2+, Fe2+, Co2+, Ni2+, Zn2+, Cd2+, Sn2+, and Hg2+)
into the target system without any additional modifications optimized for the TIP3P, the SPC/E, and the TIP4P-EW water
to the potentials. Note that the model parameters of Mg2+ for models. These three sets of parameters were generated through
TIP3P water is not identical to that provided in our previous a parameter generation procedure using the experimental
work13, because of using different parameter generation solvation free energies revised by an accurate proton hydra-
approaches. But both of them can produce the coordination tion free energy (−265.9 kcal/mol). All the parameters can
geometry and solvation free energy with the relative error accurately reproduce the experimental values of metal−oxygen
under 1%. Therefore, both the two models of Mg2+ are well- distance, coordination number (especially for V2+, Cd2+, Sn2+,
optimized and applicable. and Hg2+) and solvation free energy in water. Moreover, the
On the other hand, the CN values produced during calculated solvation free energies for all the models are
the parameter space scanning can be greater than six (see independent of the simulation methodology. The models in
section 3.1), indicating the capacity of our model to accurately this work provide a good choice for researchers to study these
reproduce the CN values of the nonhexacoordinate metal ions. divalent metal cations in silico with exchangeable ligands for
studying the metal-containing biological systems.


For example, the Ca2+ parameters of the current octahedral
model for TIP3P water is [R = 1.1729 Å; ε = 6.7992 kcal/mol]
with the MC-D bond length of 0.9 Å and the dummy atom ASSOCIATED CONTENT
charge of 0.5 e. This set of parameters can produce the CN *
S Supporting Information

of Ca2+ to be 8.0, IOD to be 2.47 Å, and ΔGsol to be The Supporting Information is available free of charge on the
−390.42 kcal/mol, where the experimental CN44 is 8, IOD44 is ACS Publications website at DOI: 10.1021/acs.jctc.6b00223.
2.46 Å, and ΔGsol is −390.6 kcal/mol (revised according to Supporting tables and figures for the calculation details,
the method described in section 2.2). However, the current and the examples for the model preparing files (PDF)
3258 DOI: 10.1021/acs.jctc.6b00223
J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

■ AUTHOR INFORMATION
Corresponding Author
(17) Marcus, Y. The Thermodynamics of Solvation of Ions. Part 4.
Application of the Tetraphenylarsonium Tetraphenylborate (TATB)
Extrathermodynamic Assumption to the Hydration of Ions and to
*E-mail: twtan@mail.buct.edu.cn. Properties of Hydrated Ions. J. Chem. Soc., Faraday Trans. 1 1987, 83,
Funding 2985−2992.
(18) Tissandier, M. D.; Cowen, K. A.; Feng, W. Y.; Gundlach, E.;
All the simulations were supported by CHEMCLOUDCOM-
Cohen, M. H.; Earhart, A. D.; Coe, J. V.; Tuttle, T. R. The Proton’s
PUTING. This work was supported by the National Basic Absolute Aqueous Enthalpy and Gibbs Free Energy of Solvation from
Research Program of China (973 program) (2013CB733600), Cluster-Ion Solvation Data. J. Phys. Chem. A 1998, 102, 7787−7794.
the National Nature Science Foundation of China (19) Camaioni, D. M.; Schwerdtfeger, C. A. Comment on “Accurate
(21436002,21390202), and the China Postdoctoral Science experimental values for the free energies of hydration of H+, OH-, and
Foundation (2015M580993). H3O+. J. Phys. Chem. A 2005, 109, 10795−10797.
Notes (20) Kelly, C. P.; Cramer, C. J.; Truhlar, D. G. Aqueous Solvation
The authors declare no competing financial interest. Free Energies of Ions and Ion-Water Clusters Based on an Accurate


Value for the Absolute Aqueous Solvation Free Energy of the Proton.
J. Phys. Chem. B 2006, 110, 16066−16081.
REFERENCES (21) Reif, M. M.; Hünenberger, P. H.; Oostenbrink, C. New
(1) Williams, R. J. Metal Ions in Biological Systems. Biol. Rev. 1953, Interaction Parameters for Charged Amino Acid Side Chains in the
28, 381−412. GROMOS Force Field. J. Chem. Theory Comput. 2012, 8, 3705−3723.
(2) Hu, L.; Ryde, U. Comparison of Methods to Obtain Force-Field (22) Kastenholz, M. A.; Hünenberger, P. H. Computation of
Parameters for Metal Sites. J. Chem. Theory Comput. 2011, 7, 2452− Methodology-Independent Ionic Solvation Free Energies from
2463. Molecular Simulations. II. The Hydration Free Energy of the Sodium
(3) Li, P.; Roberts, B. P.; Chakravorty, D. K.; Merz, K. M., Jr Rational Cation. J. Chem. Phys. 2006, 124, 224501.
Design of Particle Mesh Ewald Compatible Lennard-Jones Parameters (23) Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Merz, K.
for + 2 Metal Cations in Explicit Solvent. J. Chem. Theory Comput. M.; Ferguson, D. M.; Spellmeyer, D. C.; Fox, T.; Caldwell, J. W.;
2013, 9, 2733−2748. Kollman, P. A. A Second Generation Force Field for the Simulation of
(4) Li, P.; Merz, K. M., Jr Taking into Account the Ion-Induced Proteins, Nucleic acids, and Organic Molecules. J. Am. Chem. Soc.
Dipole Interaction in the Nonbonded Model of Ions. J. Chem. Theory 1995, 117, 5179−5197.
Comput. 2013, 10, 289−297. (24) Marcus, Y. Ionic Radii in Aqueous Solutions. Chem. Rev. 1988,
(5) Pang, Y. P. Novel Zinc Protein Molecular Dynamics Simulations: 88, 1475−1498.
Steps Toward Antiangiogenesis for Cancer Treatment. J. Mol. Model. (25) Smirnov, P.; Trostin, V. Structural Parameters of Close
1999, 5, 196−202. Surroundings of Sr2+ and Ba2+ Ions in Aqueous Solutions of Their
(6) Oelschlaeger, P.; Klahn, M.; Beard, W. A.; Wilson, S. H.; Warshel, Salts. Russ. J. Gen. Chem. 2011, 81, 282−289.
A. Magnesium-Cationic Dummy Atom Molecules Enhance Repre- (26) Miyanaga, T.; Watanabe, I.; Ikeda, S. Amplitude in EXAFS and
sentation of DNA Polymerase β in Molecular Dynamics Simulations: Ligand Exchange Reaction of Hydrated 3d Transition Metal
Improved Accuracy in Studies of Structural Features and Mutational Complexes. Chem. Lett. 1988, 17, 1073−1076.
Effects. J. Mol. Biol. 2007, 366, 687−701. (27) Ohtaki, H.; Radnai, T. Structure and Dynamics of Hydrated
(7) Saxena, A.; Sept, D. Multisite Ion Models that Improve Ions. Chem. Rev. 1993, 93, 1157−1204.
Coordination and Free Energy Calculations in Molecular Dynamics (28) Case, D. A.; Berryman, J. T.; Betz, R. M.; Cerutti, D. S.;
Simulations. J. Chem. Theory Comput. 2013, 9, 3538−3542. Cheatham, T. E., III; Darden, T. A.; Duke, R. E.; Giese, T. J.; Gohlke,
(8) Duarte, F.; Bauer, P.; Barrozo, A.; Amrein, B. A.; Purg, M.; Åqvist, H.; Goetz, A. W.; Homeyer, N.; Izadi, S.; Janowski, P.; Kaus, J.;
J.; Kamerlin, S. C. L. Force Field Independent Metal Parameters Using Kovalenko, A.; Lee, T. S.; LeGrand, S.; Li, P.; Luchko, T.; Luo, R.;
a Nonbonded Dummy Model. J. Phys. Chem. B 2014, 118, 4351−4362. Madej, B.; Merz, K. M.; Monard, G.; Needham, P.; Nguyen, H.;
(9) Liao, Q.; Kamerlin, S. C. L.; Strodel, B. Development and Nguyen, H. T.; Omelyan, I.; Onufriev, A.; Roe, D. R.; Roitberg, A.;
Application of a Nonbonded Cu2+ Model That Includes the Jahn−
Salomon-Ferrer, R.; Simmerling, C. L.; Smith, W.; Swails, J.; Walker,
Teller Effect. J. Phys. Chem. Lett. 2015, 6, 2657−2662.
R. C.; Wang, J.; Wolf, R. M.; Wu, X.; York, D. M.; Kollman, P. A.
(10) Panteva, M. T.; Giambaşu, G. M.; York, D. M. Comparison of
AMBER, version 14; University of California: San Francisco, 2014.
Structural, Thermodynamic, Kinetic and Mass Transport Properties of
(29) Duan, Y.; Wu, C.; Chowdhury, S.; Lee, M. C.; Xiong, G.; Zhang,
Mg2+ Ion Models Commonly Used in Biomolecular Simulations. J.
Comput. Chem. 2015, 36, 970−982. W.; Yang, R.; Cieplak, P.; Luo, R.; Lee, T. A Point-Charge Force Field
(11) Åaqvist, J.; Warshel, A. Free Energy Relationships in for Molecular Mechanics Simulations of Proteins Based on
Metalloenzyme-Catalyzed Reactions. Calculations of the Effects of Condensed-Phase Quantum Mechanical Calculations. J. Comput.
Metal Ion Substitutions in Staphylococcal Nuclease. J. Am. Chem. Soc. Chem. 2003, 24, 1999−2012.
1990, 112, 2860−2868. (30) Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H. J. Numerical
(12) Darden, T.; York, D.; Pedersen, L. Particle Mesh Ewald: An N · Integration of the Cartesian Equations of Motion of a System with
log (N) Method for Ewald Sums in Large Systems. J. Chem. Phys. Constraints: Molecular Dynamics of n-alkanes. J. Comput. Phys. 1977,
1993, 98, 10089−10092. 23, 327−341.
(13) Jiang, Y.; Zhang, H.; Feng, W.; Tan, T. Refined Dummy Atom (31) Steinbrecher, T.; Mobley, D. L.; Case, D. A. Nonlinear Scaling
Model of Mg2+ by Simple Parameter Screening Strategy with Revised Schemes for Lennard-Jones Interactions in Free Energy Calculations. J.
Experimental Solvation Free Energy. J. Chem. Inf. Model. 2015, 55, Chem. Phys. 2007, 127, 214108.
2575−2586. (32) Hummer, G.; Pratt, L. R.; García, A. E. Ion Sizes and Finite-Size
(14) Noyes, R. M. Thermodynamics of Ion Hydration as a Measure Corrections for Ionic-Solvation Free Energies. J. Chem. Phys. 1997,
of Effective Dielectric Properties of Water. J. Am. Chem. Soc. 1962, 84, 107, 9275−9277.
513−522. (33) Darden, T.; Pearlman, D.; Pedersen, L. G. Ionic Charging Free
(15) Rosseinsky, D. Electrode Potentials and Hydration Energies. Energies: Spherical versus Periodic Boundary Conditions. J. Chem.
Theories and Correlations. Chem. Rev. 1965, 65, 467−490. Phys. 1998, 109, 10921−10935.
(16) Marcus, Y. A Simple Empirical Model Describing the (34) Warren, G. L.; Patel, S. Hydration Free Energies of Monovalent
Thermodynamics of Hydration of Ions of Widely Varying Charges, Ions in Transferable Intermolecular Potential Four Point Fluctuating
Sizes, and Shapes. Biophys. Chem. 1994, 51, 111−127. Charge Water: An Assessment of Simulation Methodology and Force

3259 DOI: 10.1021/acs.jctc.6b00223


J. Chem. Theory Comput. 2016, 12, 3250−3260
Journal of Chemical Theory and Computation Article

Field Performance and Transferability. J. Chem. Phys. 2007, 127,


064509.
(35) Kastenholz, M. A.; Hünenberger, P. H. Computation of
Methodology-Independent Ionic Solvation Free Energies from
Molecular Simulations. I. The Electrostatic Potential in Molecular
Liquids. J. Chem. Phys. 2006, 124, 124106.
(36) Reif, M. M.; Hünenberger, P. H. Computation of Methodology-
Independent Single-Ion Solvation Properties from Molecular Simu-
lations. III. Correction Terms for the Solvation Free Energies,
Enthalpies, Entropies, Heat Capacities, Volumes, Compressibilities,
and Expansivities of Solvated Ions. J. Chem. Phys. 2011, 134, 144103.
(37) Rocklin, G. J.; Mobley, D. L.; Dill, K. A.; Hünenberger, P. H.
Calculating the Binding Free Energies of Charged Species Based on
Explicit-Solvent Simulations Employing Lattice-Sum Methods: An
Accurate Correction Scheme for Electrostatic Finite-Size Effects. J.
Chem. Phys. 2013, 139, 184103.
(38) MATLAB, version 2011b; The MathWorks, Inc.: Natick, MA,
2011.
(39) Allnér, O.; Nilsson, L.; Villa, A. Magnesium Ion−Water
Coordination and Exchange in Biomolecular Simulations. J. Chem.
Theory Comput. 2012, 8, 1493−1502.
(40) Case, D. A.; Berryman, J. T.; Betz, R. M.; Cerutti, D. S.;
Cheatham, T. E., III; Darden, T. A.; Duke, R. E.; Giese, T. J.; Gohlke,
H.; Goetz, A. W.; Homeyer, N.; Izadi, S.; Janowski, P.; Kaus, J.;
Kovalenko, A.; Lee, T. S.; LeGrand, S.; Li, P.; Luchko, T.; Luo, R.;
Madej, B.; Merz, K. M.; Monard, G.; Needham, P.; Nguyen, H.;
Nguyen, H. T.; Omelyan, I.; Onufriev, A.; Roe, D. R.; Roitberg, A.;
Salomon-Ferrer, R.; Simmerling, C. L.; Smith, W.; Swails, J.; Walker,
R. C.; Wang, J.; Wolf, R. M.; Wu, X.; York, D. M.; Kollman, P. A.
AmberTools, version 15; University of California: San Francisco, 2015.
(41) Kumar, S.; Rosenberg, J. M.; Bouzida, D.; Swendsen, R. H.;
Kollman, P. A. The Weighted Histogram Analysis Method for Free-
Energy Calculations on Biomolecules. I. The Method. J. Comput.
Chem. 1992, 13, 1011−1021.
(42) He, M. M.; Clugston, S. L.; Honek, J. F.; Matthews, B. W.
Determination of the Structure of Escherichia coli Glyoxalase I
Suggests a Structural Basis for Differential Metal Activation.
Biochemistry 2000, 39, 8719−8727.
(43) Bleuzen, A.; Pittet, P.-A.; Helm, L.; Merbach, A. E. Water
Exchange on Magnesium (II) in Aqueous Solution: a Variable
Temperature and Pressure 17O NMR Study. Magn. Reson. Chem.
1997, 35, 765−773.
(44) Jalilehvand, F.; Spångberg, D.; Lindqvist-Reis, P.; Hermansson,
K.; Persson, I.; Sandström, M. Hydration of the Calcium Ion. An
EXAFS, Large-Angle X-ray Scattering, and Molecular Dynamics
Simulation Study. J. Am. Chem. Soc. 2001, 123, 431−441.

3260 DOI: 10.1021/acs.jctc.6b00223


J. Chem. Theory Comput. 2016, 12, 3250−3260

You might also like