Вы находитесь на странице: 1из 9

Fluid Phase Equilibria 472 (2018) 85e93

Contents lists available at ScienceDirect

Fluid Phase Equilibria


j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / fl u i d

Using molecular dynamics simulations to predict the effect of


temperature on aqueous solubility for aromatic compounds
lica Fierro b, Loreto M. Valenzuela a, c, Jose
Raimundo Gillet a, Ange rez-Correa a, *
 R. Pe
a
Department of Chemical and Bioprocess Engineering, School of Engineering, Pontificia Universidad Cato lica de Chile, Avda. Vicun~ a Mackenna 4860,
Santiago, Chile
b lica de Chile, Avda, Vicun
Institute of Biological and Medical Engineering, School of Engineering, Pontificia Universidad Cato ~ a Mackenna 4860, Santiago,
Chile
c
Bioorganic and Molecular Modeling Lab, Organic Chemistry Department, Faculty of Chemistry, Pontificia Universidad Cato lica de Chile, Avda. Vicun
~a
Mackenna 4860, Santiago, Chile

a r t i c l e i n f o a b s t r a c t

Article history: Since polyphenols are highly bioactive, their separation and purification are active research topics. In the
Received 20 February 2018 optimization of such processes, the trend of solubility at different conditions is of major interest,
Received in revised form although obtaining experimental solubility data for polyphenols is lengthy and costly. A simple and
7 May 2018
previously explored thermodynamic integration (TI) procedure was applied to estimate the Gibbs hy-
Accepted 11 May 2018
Available online 12 May 2018
dration free energies at different temperatures for benzoic acid, (þ)-catechin and toluene. These hy-
dration energies were compared with experimental solubilities. An extensive test with 10 replications
comprising a total of 200 molecular dynamics (MD) TI simulations was conducted. Linear trends were
Keywords:
Polyphenols
observed for the estimated Gibbs hydration free energy at different temperatures. Electrostatic contri-
Solid-liquid extraction butions showed a strong effect for the studied polyphenols due to hydrogen bonds. Lennard-Jones (LJ)
Solubility contributions to hydration free energies were small, being more significant for toluene due to cavity
Hydration free energy formation. The estimated Gibbs free energy of hydration presented an exponential trend for benzoic acid
Molecular dynamics simulations and (þ)-catechin when compared to experimental solubility data. These trends were fitted to an
Thermodynamic integration empirical expression. To understand the physical meaning of the empirical fitting parameters, the effect
Solvation of temperature in the sublimation enthalpy and in the entropy of the solid solute should be estimated.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction for the optimization of these processes. Obtaining solubility data of


polyphenols by means of experiments is complex, therefore,
Polyphenols are a well-studied and broad family of bioactive extraction processes of polyphenols are commonly optimized by
natural compounds found in plants and known for their antioxi- using the experimental response surface method [2,3], which is
dant, vasodilatory, anti-carcinogenic, anti-inflammatory, immune- highly demanding in terms of cost and time. Reliable predictions of
stimulating, anti-allergic and antiviral properties [1]. There is the effects of temperature and pressure on the solubility of poly-
much interest in industry and academia for extracting these com- phenols would simplify, accelerate, and reduce the costs of the
pounds from their natural sources to use them for the preparation development of new natural products.
of new nutraceuticals, functional food ingredients, pharmaceuti- To accurately predict solubility, the non-ideality in the mixture
cals, and cosmetic products. Extraction processes can be optimized should be considered. The most commonly used methods for this
if the solubility of the desired compound in the extraction solvent is task are activity coefficient models [4e8], nevertheless, these
known at different temperatures and pressures. Specially, the trend models are limited since they require experimental solubility data
of solubility at different conditions provides sufficient information as input [9]. Quantitative structure-property relationship (QSPR)
methods have been widely used as well. They are based in statis-
tical relations between experimental solubility data and molecular
properties such as electronic and topological evaluations, hydro-
* Corresponding author. philicity/hydrophobicity calculations, molecular surface area,
E-mail addresses: rgillet@uc.cl (R. Gillet), afierroh@uc.cl (A. Fierro), lvalenzr@
among other properties [10e13]. However, these computational
rez-Correa).
ing.puc.cl (L.M. Valenzuela), perez@ing.puc.cl (J.R. Pe

https://doi.org/10.1016/j.fluid.2018.05.013
0378-3812/© 2018 Elsevier B.V. All rights reserved.
86 R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93

methods need experimental solubility data for many compounds


and do not consider the exact interaction between solute and sol-
vent. Molecular dynamics (MD) and Monte Carlo (MC) simulations
have proven to be more suitable computational tools for the study
and prediction of solubility. Unlike activity coefficient and QSPR
methods, MD and MC consider a detailed description of the atomic
and molecular interaction without simplifications and without Fig. 1. Benzoic acid (C7H6O2), (þ)-catechin (C15H14O6) and toluene (C7H8) molecules.
requiring experimental solubility data.
Two MD and MC methods have been applied for solubility
compound which presents structural similarity with benzoic acid,
predictions [9,14e30]. In the chemical potential route (CPR), solu-
however, its aqueous solubility is much lower.
bility is determined as the solute concentration in solution at which
The remaining sections of this paper are organized as follows:
the chemical potential of the solution is equal to that of the crystal
Section 2 contains computational details and explains the methods
solid [14e26]. In the alternative direct coexistence method (DCM),
used in our study. In Section 3 we present and discuss our results.
the solution is brought into contact with the solid crystal, and the
Our findings are summarized in Section 4.
solubility corresponds to the equilibrium solute concentration
[27e30]. The main limitation of DCM and CPR methods, is the
2. Methods and computational details
scarcity of accurate force fields available for crystalline compounds
[14]. In addition, computing absolute solid free energies of molec-
2.1. Software packages
ular solids, although straightforward, is cumbersome [14]. More-
over, the obtained solubility values strongly depend on the force
MD simulations were carried out in the NAMD (NAnoscale
field used [15]. For example, several research groups employing
Molecular Dynamics) v2.12 simulation package [40]. To perform
different CPR or DCM methods have reported widely different NaCl
thermodynamic integration (TI) [38] calculations, we employed the
aqueous solubilities at ambient conditions [31]. To improve MD
TI method [41] included in the NAMD v2.12 package. VMD (Visual
solubility calculations, the estimation of solid phase properties
Molecular Dynamics) v1.9.2 [42] was used to create initial system
should be improved and more accurate force field parametrizations
configurations and to represent and study molecular structures, as
for crystalline compounds should be derived.
well as molecular motions of the systems.
Two decades ago, Lipinski et al. [32] reviewed solubility pre-
diction methods, and suggested that solubility is the result of the
complex interplay between different energy contributions such us 2.2. Models
(1) the energy required to disarray the solute crystal lattice to bring
its molecules into solution (generally referred to as sublimation free All force field parameters, except for the atomic point charges of
each solute, were taken from the CHARMM force field [43]. This
energy, Dsub G), (2) the energy associated to the interactions be-
tween solute and solvent molecules in solution, and (3) the energy force field has been extensively used to model pharmaceutical and
natural compounds. Point charges were calculated quantum-
required to form a cavity in the solvent, into which the solute
molecule is hosted. In this study we applied MD simulations to mechanically at the Hartree-Fock 6-31G* basis set [44] with the
Spartan’10 v1.1.0 software [45]. Mol2 archives containing the
analyze the effect of temperature on solubility, focusing on the
latter two contributions, which are accounted for by the hydration calculated charge values (support information) were extracted
from the software, and then, these charge values were inserted into
free energy term (Dhyd G).
We studied the effect of temperature variations in hydration the input files.
Water molecules were represented using the TIP3P model [46],
free energy and compared the calculated trends with experimental
solubility data. There are several studies that used MD and MC which is probably the most used water model for biomolecules
simulations to calculate hydration free energies. For example, simulations. In addition, TIP3P simulations provided accurate pre-
Garrido et al. [33] calculated hydration free energies and 1-octanol dictions of hydration free energy for a wide set of compounds [35],
solvation free energies to predict partition coefficients, which along and was also used to infer relative solubilities [37]. Moreover, since
with the experimental melting point of a compound, can be used to TIP3P is the default solvation model in NAMD, it is straightforward
predict its solubility [34]. Mobley et al. [35] applied a simple MD to apply.
method using the explicit TIP3P (Three-Point Transferrable Inter- CHARMM force-fields use harmonic bonds and angles of the
molecular Potential) water model [36] to predict the hydration free form
energies of 504 neutral compounds. Chebil et al. [37] also applied X
MD with the explicit TIP3P water model and solvation free energy Ebond ¼ kbi ðbi  b0i Þ2 (1)
i2bonds
calculations, to estimate the solubilities of quercetin on different
solvents using a single experimental solubility. In our study we
and
applied MD simulations with the explicit TIP3P water model and
the thermodynamic integration (TI) method [38] to calculate hy- X 2
Eangle ¼ kqi ðqi  q0i Þ (2)
dration free energies at different temperatures. In addition, we i2angles
compared the calculated hydration free energies with the experi-
mental aqueous solubility of a set of compounds. The impact of the where kbi and kqi are force constants for the bonds and angles har-
hydration free energy term in the variation of solubility with monic potentials, respectively. b0i and q0i are equilibrium bond
temperature has not been analyzed before using MD simulations. lengths and angles. A truncated Fourier series models dihedral
To validate our results, each MD simulation was replicated 10 times, angles of the form
yielding more than 200 TI-MD simulations. Three compounds were
X
chosen for this study, (þ)-catechin, benzoic acid and toluene Edih ¼ kfi ½1 þ cosðni fi  di Þ (3)
(Fig. 1). (þ)-Catechin and benzoic acid are part of the polyphenol i2dih
family [39]; (þ)-catechin is a flavonoid while benzoic acid is a
precursor of the phenolic acids [1]. Toluene is an aromatic where kfi are force constants, besides ni , fi and di are dihedral
R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93 87

period, angle, and phase, respectively. Short-range and long-range Langevin piston method, a period of 100 fs and a decay of 150 fs
interactions are modeled using 12-6 Lennard-Jones (LJ) potentials were selected. LJ interactions were truncated using a cutoff radius
and Coulomb potentials, of the form of 1.2 nm. Electrostatic interactions were calculated using the Par-
8 2 9 ticle Mesh Ewald summation [55] with a grid size equal to the size
!12 !6 3
X X< Rmin Rmin qi qj = of each simulation box.
εij 4 5þ
ij ij ij
ELJþC ¼  (4) In the TI simulations, 50 different l values uniformly distributed
: rij rij 4pε0 rij ;
i j>i between 0 and 1 were used. Each l step was simulated through 0.1
ns and the energy was measured each 20 fs (2$105 ns), summing
where εij and Rmin
ij are LJ parameters, qi and qj are partial charges 5000 energy measurements for each l step. The last 20,000 fs (1000
and rij is the distance between atoms i and j. Unlike-atom in- energy measurements) of each l step were collected, and the en-
teractions are computed using geometric mean combining rules for ergy was averaged. Along with the average energies, standard de-
the εij parameters, and average combining rules for the Rmin ij viations for the energy measurements were calculated in each l
parameters, step and were later used to calculate the error from propagation of
pffiffiffiffiffiffiffi uncertainty. Following the methodology of Garrido et al. [56], LJ
εij ¼ εi εj (5) interactions were interpolated using a soft-core potential with a
soft-core parameter of 0.5.
Rmin
i þ Rmin
j
This procedure was implemented for all compounds at each
Rmin
ij ¼ (6) temperature. The TI procedure was repeated 10 times to assure the
2
consistency and repeatability of the results. Since 20 simulations
In the CHARMM forcefield, 1e4 intramolecular non-bonded were required for each replica (9 temperatures for benzoic acid, 6
interactions are handled by scaling down the electrostatic inter- temperatures for (þ)-catechin and 5 temperatures for toluene), a
action by a factor of 1.0. Regarding LJ interactions, a secondary set of total of 200 TI-MD simulations were conducted in a multiprocessor
parameters is included in the force field for specific atoms. In such computer with 12 processing nodes.
cases, 1e4 intramolecular LJ interactions are modified by using the
secondary LJ parameters. In the remaining cases, 1e4 LJ interactions 2.4. Output analysis
are scaled down by a factor of 1.0 since their spatial relationship is
already influenced by the internal terms, including the dihedral The curves of energy versus lambda retrieved from the simu-
term [47]. lations were integrated numerically using a modified version of the
namd_ti.pl perl script that can be found in the NAMD website [57].
2.3. Simulation details The most common approach for integrating the free energy
gradient in TI is the trapezoidal rule, nevertheless, it has been
For the three compounds, isobaric aqueous solubility data at shown to lead to systematic errors [58e60]. Conversely, namd_ti.pl
different temperatures was retrieved from literature to subse- script uses cubic spline interpolation for the numerical integration.
quently compare it to the simulation results. Cuevas-Valenzuela In addition, we added to the script a feature to compute standard
et al. [48] measured benzoic acid and (þ)-catechin solubilities at deviations of the free energies. To do this, we applied the general
different temperatures and atmospheric pressure (101.325 kPa), formula for propagation of uncertainty [61], which defines the
while Miller & Hawthorne [49] measured toluene solubilities at uncertainty df of a function f of n independent variables
different temperatures and 50 bar (5 MPa). The experimental data ðx1 ; x2 ; … ; xn Þ as follows:
is presented as plots of solubility against temperature in supporting sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
information. In this work, simulations were set up to reproduce the  2  2
vf vf
same conditions as experimental references, particularly the pres- df ¼ dx1 þ … þ dxn (7)
vx1 vxn
sure and temperature, and in such conditions, the Gibbs hydration
free energy was estimated for each compound. Finally, the calculated solvation free energies averaged from the
MD simulations were conducted using the NAMD v2.12 simu- 10 replicas were compared with experimental solubility data. In
lation package. To build the systems, each solute was submerged in addition, to assess the variation of cavity formation terms at
a pre-equilibrated solvent box of 3 nm of side, which included different temperatures, hydration shells of each compound were
approximately 800 water molecules. This was done using the Sol- analyzed by means of radial distribution functions (RDF) at the
vate 1.5 plugin [50], which is incorporated in VMD (Visual Molec- equilibrium state prior to each TI run.
ular Dynamics) v1.9.2 package [42]. Firstly, each system was pre-
equilibrated to reach temperatures and pressures in accordance 3. Results and discussion
to the mentioned experimental conditions. To achieve this, each
system was subjected to an energy minimization of 0.05 ns using 3.1. Hydration free energy at different temperatures
the conjugate gradient algorithm [51], followed by an NPT equili-
bration for 1 ns. Afterwards, hydration free energies were esti- Calculated average hydration free energies and their errors for
mated for each compound at each of the conditions following the TI each replica were compared. The results from the 10 repetitions
procedure described by Garrido et al. [52], whom were able to were gathered into average values along with their respective er-
accurately predict hydration free energies without any experi- rors, and their trends were analyzed in graphs of hydration free
mental information and without any fitting of parameters. energy against temperature. None experimental measurements of
MD simulations were performed using periodic boundary con- absolute hydration free energy were found in literature for the
ditions in all directions, and Newton's equations of motion for all selected compounds at the selected conditions (i.e. 0.1 MPa for
the species were integrated using the leap-frog dynamic algorithm benzoic acid and (þ)-catechin; 5 MPa for toluene), therefore the
[53] with a timestep of 2 fs (2$106 ns). Temperature and pressure accuracy of our hydration free energy calculations could not be
were kept constant using Langevin dynamics and the Langevin validated directly against experimental data. Toluene's hydration
piston method, respectively [54]. A frictional constant of 1 ps1 was free energy has been reported to take a value of 0.77 kcal mol1
used for the temperature control Langevin method. For the (3.22 kJ mol1) [62e64] at ambient conditions (293 K, 0.1 MPa),
88 R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93

(a) 278 283 294 300 308 318 327 337 341 278 283 294 300 308 318 327 337 341
100 100

LJ Contribution to ΔhydG (kJ∙mol-1)


El. Contribution to ΔhydG (kJ∙mol-1)
0 0
-100 -100
-200 -200
-300 -300
-400 -400
-500 -500
-600 -600
-700 -700
-800 -800
Temperature (K) Temperature (K)
R1 R2 R3 R1 R2 R3

278 283 298 308 325 331 278 283 298 308 325 331
(b)100 100

LJ Contribution to ΔhydG (kJ∙mol-1)


0 0
El. Contribution to ΔhydG (kJ∙mol-1)

-100 -100
-200 -200
-300 -300
-400 -400
-500 -500
-600 -600
-700 -700
-800 -800
Temperature (K) Temperature (K)
R1 R2 R3 R1 R2 R3

298 323 373 423 473 298 323 373 423 473
(c)100 100
LJ Contribution to ΔhydG (kJ∙mol-1)
El. Contribution to ΔhydG (kJ∙mol-1)

0 0
-100 -100
-200 -200
-300 -300
-400 -400
-500 -500
-600 -600
-700 -700
-800 -800
Temperature (K) Temperature (K)
R1 R2 R3 R1 R2 R3

Fig. 2. Average electrostatic and LJ contributions to the Gibbs hydration free energies calculated for benzoic acid (A), (þ)-catechin (B) and toluene (C) at each simulation tem-
perature for three replicas. Error bars correspond to the errors calculated by propagation of uncertainty.

while we calculated a value of 47.78 kJ mol-1 at ambient temper- significant variations with changing temperatures, although these
ature and 5 MPa. Even though these need to be corrected by the contributions present slightly more variability between replicas than
pressure difference, our estimations were higher than the usual those of electrostatic contributions.
hydration free energy values, which might well be explained by the Fig. 3shows for each compound the trends of electrostatic
water model that we used. Nevertheless, our interest was esti- contributions, LJ contributions, and Gibbs hydration free energies
mating the relative differences between hydration free energies at calculated at different simulation temperatures and averaged from
different temperatures rather than obtaining accurate predictions.
Fig. 2 shows the results of three replicas along with their errors;
the remaining results are tabulated in our supporting information. Table 1
We also include global error estimates from the 10 individual replicas Hydrogen bond donor/acceptor groups count, polarizability and dipole moment
at each temperature as supporting information, to verify the consis- of each solute compound. Properties were determined using the software
Spartan ’10 v1.1.0.
tency between both global errors and the errors from individual
replicas (see Table 1 e in supporting information). The figure shows Compound Hydrogen bond Polarizability Dipole moment
highly consistent simulations, with small variations between donor/acceptor count
different replicas and low standard deviations. Statistically significant Benzoic acid 1/1 48.76 2.16 debye
differences are observed in electrostatic contributions calculated at (þ)-Catechin 5/6 60.21 0.82 debye
different temperatures. LJ contributions show non-statistically Toluene 0/0 47.84 0.27 debye
R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93 89

the 10 replicas. Errors determined by propagation of uncertainty as donors or acceptors of hydrogen bonds in the studied poly-
are presented with bars. phenols; these interactions have large contributions to the free
In the CHARMM force field, hydrogen bonds are included in the energy. Accordingly, hydrogen bonds donor and acceptor groups
electrostatic interaction term along with coulombic attractions and were counted for each compound (Table 1) using the software
repulsions. LJ term accounts for the weak dipole attraction between Spartan ’10 v1.1.0 [45]. It was observed that higher donor/acceptor
distant atoms and the hard-core repulsion between neighbor atoms counts lead to more pronounced trends in the electrostatic con-
that is related to the formation of cavities. It is clear from Fig. 3 that tributions against temperature. Moreover, the donor/acceptor
the trend in hydration free energy is basically of electrostatic origin. count informed that toluene does not act neither as an acceptor nor
Clear trends were observed for benzoic acid and (þ)-catechin, donor of hydrogen bonds, which was reflected in its small elec-
while a small variation was observed for toluene. This might be trostatic contribution variation. This small variation might be due
explained by the presence of functional groups that usually behave to a dipole moment (Table 1) effect in electrostatic contributions,

(a) 0 (a) 1.2


-200
Elec. Contribution to ΔhydG

1
-400

g(r) (g ∙ cm-3)
0.8
-600 278 K
(kJ ∙ mol-1)

283 K
0.6 294 K
-800
300 K
toluene 0.4 308 K
-1000
benzoic acid 318 K
-1200 0.2 327 K
catechin 337 K
-1400 341 K
0
270 330 390 450
Temperature (K) 0 2 4 6 8 10
r (Å)
(b) 0 (b) 1.2
-200
LJ Contribution to ΔhydG

1
-400
g(r) (g ∙ cm-3)

0.8
(kJ ∙ mol-1)

-600

-800 0.6
278 K
toluene 283 K
-1000 0.4
benzoic acid 298 K
-1200 308 K
catechin 0.2 325 K
-1400 331 K
270 330 390 450 0
Temperature (K) 0 2 4 6 8 10
r (Å)
(c) 0
(c) 1.2
-200
ΔhydG (kJ ∙ mol-1)

1
-400
g(r) (g ∙ cm-3)

0.8
-600

-800 0.6
toluene
-1000 0.4 298 K
benzoic acid 323 K
-1200 373 K
catechin 0.2 423 K
-1400 473 K
270 330 390 450 0
Temperature (K) 0 2 4 6 8 10
r (Å)
Fig. 3. Calculated electrostatic (Elec.) and LJ contributions, and total Gibbs free energy
of hydration (DhydG) at different temperatures for benzoic acid (a), (þ)-catechin (b) and Fig. 4. Solute (excluding hydrogens) e water oxygen RDFs calculated from the equi-
toluene (c). Error bars correspond to the standard deviation of each result, calculated librium state prior to each TI run for benzoic acid (a), (þ)-catechin (b) and toluene (c)
by propagation of uncertainty. Dotted lines correspond to adjusted linear trends to at each temperature. Noticeable variations in the height of toluene's RDFs illustrate a
each data set. temperature-dependence of its hydration shell size.
90 R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93

which was minor compared to the effect of hydrogen bonds. Starting from the thermodynamic definition of solid-liquid
Toluene and benzoic acid show a small, but statistically signifi- equilibrium and considering an infinitely diluted system, the sol-
cant, increasing trend in their LJ contributions with temperature, ubility of a compound can be related to its hydration free energy
while (þ)-catechin does not show any trend. These contributions using the following expression,
are due to vdW attractive dispersion interactions and cavity for-  
mation terms. Van der Waals (vdW) attractive dispersion in- n1 ðT; pÞf2s ðT; pÞ Dhyd G
x2 ¼ exp  (11)
teractions are related to the polarizability of each molecule RT RT
(Table 1). Contrary to the observed trend in LJ contributions,
(þ)-catechin presents the highest polarizability value, while where n1 is the intensive molar volume of the solvent, f2s ðT; pÞ is the
toluene and benzoic acid present similar lower values. Therefore, fugacity of the solid (compound 2) at pressure p and temperature T
vdW attractive dispersion interactions seems not significant and LJ and R is the universal gas constant. Equations (10) and (11) look
contribution variations would mainly be due to cavity formation. To n ðT;pÞf s ðT;pÞ
1,
similar, with parameters A and B equivalent to 1 RT2 and RT
assess the relative magnitude of the variation in the cavity forma-
respectively. However, we compared the solubility data with the
tion contributions at different temperatures, the hydration shells of Dhyd G
each compound were analyzed using radial distribution functions term RT at different temperatures (supporting information) and
observed that parameter B is not equivalent to RT 1 . Moreover, the
(RDF) of oxygen atoms in the water molecules surrounding the
solutes (Fig. 4). In this case, trajectories of the pre-equilibration s
fugacity of the solid, f2 ðT; pÞ, should present significant variations
runs were considered rather than the trajectories of TI runs, since with temperature that might be expressed linearly (in parameter A)
the hydration shell is formed at the equilibrium state which pre- or exponentially (in parameter B). Prausnitz et al. [66] derived the
cedes the solute annihilation. following expression for the fugacity of a solid or a liquid,
RDFs in Fig. 4 show a noticeable temperature-dependence in the
 !
hydration shell size of toluene, while no significant variation is ns p  psat
observed for the studied polyphenols. Since the free energy of fis ðT; pÞ ¼ psat
i exp i
(12)
RT
cavity formation in water is proportional to the surface area of the
cavity [65], it can be inferred that the cavity formation term
where psat
i is its vapor pressure at temperature T, ns is the molar
dependence with temperature is stronger for toluene than for
volume of the solid at temperature T and pressure p, and R is the
benzoic acid and (þ)-catechin. Therefore, variations in LJ contri-
universal gas constant. This expression is valid only at conditions
butions at different temperatures were mainly associated to a
far from the critical point, where the solid phase can be considered
varying cavity formation term, which was only important for
as incompressible. Also, it is required that the saturation pressure of
toluene (non-polar).
the solid is lower than 1 bar, which normally is the case of com-
pounds which behave as solids or liquids at normal conditions. The
3.2. Comparison between the MD results and experimental vapor pressure of a solid or liquid at temperature T can be deter-
solubility data mined using the Clausius-Clapeyron relation,
   
Fig. 5 compares the calculated hydration free energies at DHis ðTÞ 1 DSsi ðTÞ
psat
i ¼ exp  þ (13)
different temperatures with experimental solubility data [48,49] R T R
(Fig. 5).
Exponential shapes were observed for the Gibbs free energy of where DHis ðTÞ and DSsi ðTÞ are the sublimation enthalpy and entropy
hydration of (þ)-catechin and benzoic acid (Fig. 5), while no trend of the solid, respectively at temperature T and R is the universal gas
was shown for toluene (supporting information). Calculated values constant. Substituting equations (12) and (13) in equation (11), the
showed small errors and the differences in the trends were statis- following expression is derived,

  

0 DHs ðTÞ DSs ðTÞ 1


ns ðT;pÞ pexp  2
R
1
T
þ 2
R
   
DH2s ðTÞ DSs2 ðTÞA
exp@
Dhyd G 1
RT  RT  R T þ R n1 ðT; pÞ
x2 ¼ (14)
RT

tically significant. As discussed in Section 3.1, the marked trends Therefore, to understand the physical meaning of parameters A
that are observed in polyphenol's results are due to an important and B and, moreover, to predict their values, expressions of the
effect in electrostatic contributions at different temperatures, variation of sublimation enthalpies and entropies of the solid at
which are strongly related to hydrogen bonds. These trends were different temperatures are needed. It seems that these values do not
fitted to an exponential function, need to be calculated accurately to get good estimations of the sol-
ubility variation of polyphenols with temperature, since we observed
  that inaccuracies in the Dhyd G calculations did not prevent us for
x ¼ A exp  B$Dhyd G (10) obtaining good estimations of its variations with temperature.

where A and B are fitting parameters. To understand the physical 4. Conclusions


meaning of these parameters, we will now examine a similar
equation that has been proposed [18] for the relation between In this article, we have reviewed the state of art in solubility
solubility and hydration free energy. prediction from MD simulations and observed that a current
R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93 91

0.003
benzoic acid
catechin
0.0025

Solubility (mol ∙ mol-1)


0.002 y = 2E-06e-0.006x y = 1E-05e-0.009x
R² = 0.872 R² = 0.9694
0.0015

0.001

0.0005

0
-1300 -1100 -900 -700 -500 -300
ΔhydG (kJ ∙ mol-1)

Fig. 5. Calculated free energies of solvation (DhydG) vs. experimental solubility at different temperatures. Benzoic acid and (þ)-catechin show exponential trends (adjusted; dotted
lines). Toluene is not shown since no trend was observed. Horizontal error bars correspond to the standard deviations of the free energy calculations, obtained by propagation of
uncertainty. Vertical error bars correspond to the reported errors in each experimental reference.

limitation in this field is the inaccuracy of solid state simulation kf Dihedral angle constant (kJ mol1)
results. As a result, a method for studying the effect of temperature n Dihedral period
in solubility by means of hydration free energy calculations has p Pressure (kPa)
been proposed. psat Vapor pressure (kPa)
Even though our hydration free energy predictions were not q Partial charge in units of jej
accurate, our results were robust, and, in the case of polyphenols, r Interatomic distance (Å)
were able to yield exponential fittings to the relation between R Ideal gas constant (kJ mol1 ∙ K1)
solubility and hydration free energy, just as it is proposed in ther- Rmin Lennard-Jones parameter (Å)
modynamic models. We observed that the fitting parameters might T Temperature (K)
be related to the sublimation enthalpy and entropy of the solid x mole fraction
solute. A complementary study comprising solid state simulations
should be carried out to understand the physical meaning of these Greek letters
parameters. There, the effect of temperature in the sublimation Dhyd G Gibbs hydration free energy (kJ mol1)
enthalpy and entropy of the solid solute should be studied. Even if DHs ðTÞ Sublimation enthalpy of the solid (kJ mol1)
the results of such study were not accurate (which is common in DSs ðTÞ Sublimation entropy of the solid (kJ mol1 ∙ K1)
the field of solid state simulations), the retrieved trends might be d Dihedral phase (rad)
robust and meaningful, just as we observed in this work. ε Lennard-Jones parameter (kJ mol1)
ε0 Permittivity of vacuum (C2$N1 ∙ m2)
Acknowledgements q Covalent angle (rad)
q0 Harmonic equilibrium angle (rad)
This research was part of a master thesis and was funded by l Coupling parameter for thermodynamic integration
project “ANILLO ACT 1105/2012” and FONDECYT 1161375. We are (scalar)
grateful to Andres Mejia for recommendations and suggestions m Chemical potential (kJ mol1)
during the course of this work; and to the National Institutes of n Intensive molar volume (m3 ∙ mol1)
Health for the use of the software NAMD v2.12. p Pi number
f Dihedral angle (rad)

Appendix A. Supplementary data


Abbreviations
LJ Lennard-Jones
Supplementary data related to this article can be found at
MC Monte Carlo
https://doi.org/10.1016/j.fluid.2018.05.013.
MD Molecular Dynamics
NAMD NAnoscale Molecular Dynamics
List of symbols NPT Ensemble: fixed number, pressure and temperature
RDF Radial distribution function
Å angstrom (1 Å ¼ 0.1 nm) TI Thermodynamic Integration
b Bond length (Å) TIP3P Three-Point Transferrable Intermolecular Potential
b0 Harmonic bond equilibrium length (Å) vdW van der Waals
f2s ðT; pÞ fugacity of the solid (kPa) VMD Visual Molecular Dynamics
G Gibbs free energy (kJ mol1)
H Hamiltonian function References
kb Harmonic bond constant (kJ ∙ Å2 ∙ mol1)
kq Harmonic angle constant (kJ rad2 ∙ mol1) [1] M. Leopoldini, N. Russo, M. Toscano, The molecular basis of working
92 R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93

mechanism of natural polyphenolic antioxidants, Food Chem. 125 (2011) [26] M. Lsal, W.R. Smith, J. Kolafa, M. Lı, Molecular simulations of aqueous elec-
288e306 doi:https://doi.org/10.1016/j.foodchem.2010.08.012. trolyte Solubility : 1. The expanded- ensemble osmotic molecular dynamics
[2] C.L. Roriz, L. Barros, M.A. Prieto, P. Morales, I.C.F.R. Ferreira, Floral parts of method for the solution phase molecular simulations of aqueous electrolyte
Gomphrena globosa L. as a novel alternative source of betacyanins: opti- Solubility : 1, The Expanded-Ensemble Osmotic Molecular Dynamics Me 2
mization of the extraction using response surface methodology, Food (2005) 12956e12965, https://doi.org/10.1021/jp0507492.
Chem. 229 (2017) 223e234 doi:https://doi.org/10.1016/j.foodchem.2017. [27] H.M. Manzanilla-Granados, H. Saint-Martín, R. Fuentes-Azcatl, J. Alejandre,
02.073. Direct coexistence methods to determine the solubility of salts in water from
[3] K.H. Wong, G.Q. Li, K.M. Li, V. Razmovski-Naumovski, K. Chan, Optimisation of numerical simulations. Test case NaCl, J. Phys. Chem. B 119 (2015)
Pueraria isoflavonoids by response surface methodology using ultrasonic- 8389e8396, https://doi.org/10.1021/acs.jpcb.5b00740.
assisted extraction, Food Chem. 231 (2017) 231e237 doi:https://doi.org/10. [28] J.R. Espinosa, J.M. Young, H. Jiang, D. Gupta, C. Vega, E. Sanz, P.G. Debenedetti,
1016/j.foodchem.2017.03.068. A.Z. Panagiotopoulos, On the calculation of solubilities via direct coexistence
[4] E. Sheikholeslamzadeh, S. Rohani, Solubility prediction of pharmaceutical and simulations: investigation of NaCl aqueous solutions and Lennard-Jones bi-
chemical compounds in pure and mixed solvents using predictive models, Ind. nary mixtures, J. Chem. Phys. 145 (2016), https://doi.org/10.1063/1.4964725.
Eng. Chem. Res. 51 (2012) 464e473, https://doi.org/10.1021/ie201344k. [29] J. Kolafa, Solubility of NaCl in water and its melting point by molecular dy-
[5] G.M. Wilson, Vapor-liquid equilibrium. XI. A new expression for the excess namics in the slab geometry and a new BK3-compatible force field, J. Chem.
free energy of mixing, J. Am. Chem. Soc. 86 (1964) 127e130, https://doi.org/ Phys. 145 (2016), https://doi.org/10.1063/1.4968045.
10.1021/ja01056a002. [30] J.L. Aragones, E. Sanz, C. Vega, Solubility of NaCl in water by molecular
[6] H. Renon, J.M. Prausnitz, Local compositions in thermodynamic excess func- simulation revisited, J. Chem. Phys. 136 (2012), https://doi.org/10.1063/
tions for liquid mixtures, AIChE J. 14 (1968) 135e144 doi:https://doi.org/10. 1.4728163.
1002/aic.690140124. [31] I. Nezbeda, F. Mou cka, W.R. Smith, Recent progress in molecular simulation of
[7] D.S. Abrams, J.M. Prausnitz, Statistical thermodynamics of liquid mixtures: a aqueous electrolytes: force fields, chemical potentials and solubility, Mol.
new expression for the excess Gibbs energy of partly or completely miscible Phys. 114 (2016) 1665e1690, https://doi.org/10.1080/00268976.2016.
systems, AIChE J. 21 (1975) 116e128, https://doi.org/10.1002/aic.690210115. 1165296.
[8] A. Klamt, Conductor-like screening model for real solvents: a new approach to [32] C.A. Lipinski, F. Lombardo, B.W. Dominy, P.J. Feeney, Experimental and
the quantitative calculation of solvation phenomena, J. Phys. Chem. 99 (1995) computational approaches to estimate solubility and permeability in drug
2224e2235 doi:https://doi.org/10.1021/j100007a062. discovery and development settings, Adv. Drug Deliv. Rev. 23 (1997) 3e25.
[9] J.Y. Seyf, A. Haghtalab, A junction between molecular dynamics simulation [33] N.M. Garrido, A.J. Queimada, M. Jorge, E.A. Macedo, I.G. Economou, 1-Octanol/
and local composition models for computation of solid-liquid equilibrium-A water partition coefficients of n-Alkanes from molecular simulations of ab-
pharmaceutical solubility application, Fluid Phase Equil. 437 (2017) 83e95, solute solvation free energies, J. Chem. Theor. Comput. 5 (2009) 2436e2446,
https://doi.org/10.1016/J.FLUID.2016.12.021. https://doi.org/10.1021/ct900214y.
[10] J. Huuskonen, Estimation of aqueous solubility for a diverse set of organic [34] D. Alantary, S. Yalkowsky, Comments on prediction of the aqueous solubility
compounds based on molecular topology, J. Chem. Inf. Comput. Sci. 40 (2000) using the general solubility equation (GSE) versus a genetic algorithm and a
773e777, https://doi.org/10.1021/ci9901338. support vector machine model, Pharmaceut. Dev. Technol. 0 (2017) 1e2,
[11] I.V. Tetko, V.Y. Tanchuk, T.N. Kasheva, A.E.P. Villa, Estimation of aqueous https://doi.org/10.1080/10837450.2017.1321663.
solubility of chemical compounds using e-state indices, J. Chem. Inf. Comput. [35] D.L. Mobley, C.I. Bayly, M.D. Cooper, M.R. Shirts, K.A. Dill, Small molecule
Sci. 41 (2001) 1488e1493, https://doi.org/10.1021/ci000392t. hydration free energies in explicit solvent: an extensive test of fixed-charge
[12] D.M. Balthasar, Estimation of aqueous solubility of organic molecules by the atomistic simulations, J. Chem. Theor. Comput. 5 (2009) 350e358, https://
group contribution approach. Application to the study of biodegradation, doi.org/10.1021/ct800409d.
J. Chem. Inf. Comput. Sci. 32 (1992) 474e482. [36] W.L. Jorgensen, J. Chandrasekhar, J.D. Madura, R.W. Impey, M.L. Klein, Com-
[13] C. Lamanna, M. Bellini, A. Padova, G. Westerberg, L. Maccari, Straightforward parison of simple potential functions for simulating liquid water, J. Chem.
recursive partitioning model for discarding insoluble compounds in the drug Phys. 79 (1983) 926e935, https://doi.org/10.1063/1.445869.
discovery process, J. Med. Chem. 51 (2008) 2891e2897, https://doi.org/ [37] L. Chebil, C. Chipot, F. Archambault, C. Humeau, J.M. Engasser, M. Ghoul,
10.1021/jm701407x. F. Dehez, Solubilities inferred from the combination of experiment and
[14] L. Li, T. Totton, D. Frenkel, Computational methodology for solubility predic- simulation. Case study of quercetin in a variety of solvents, J. Phys. Chem. B
tion: application to the sparingly soluble solutes, J. Chem. Phys. 146 (2017), 114 (2010) 12308e12313 doi:https://doi.org/10.1021/jp104569k.
https://doi.org/10.1063/1.4983754. [38] D.L. Beveridge, F.M. Dicapua, Free energy via molecular simulation: applica-
[15] A.L. Benavides, J.L. Aragones, C. Vega, Consensus on the solubility of NaCl in tions to chemical and biomolecular systems, Annu. Rev. Biophys. Biophys.
water from computer simulations using the chemical potential route, J. Chem. Chem. 18 (1989) 431e492 doi:https://doi.org/10.1146/annurev.bb.18.060189.
Phys. 144 (2016), https://doi.org/10.1063/1.4943780. 002243.
[16] D.S. Palmer, J.L. McDonagh, J.B.O. Mitchell, T. Van Mourik, M.V. Fedorov, First- [39] V. Neveu, J. Perez-Jimenez, F. Vos, V. Crespy, L. du Chaffaut, L. Mennen,
principles calculation of the intrinsic aqueous solubility of crystalline druglike C. Knox, R. Eisner, J. Cruz, D. Wishart, A. Scalbert, Phenol-Explorer: an online
molecules, J. Chem. Theor. Comput. 8 (2012) 3322e3337, https://doi.org/ comprehensive database on polyphenol contents in foods, Database 2010
10.1021/ct300345m. (2010) bap024. doi:https://doi.org/10.1093/database/bap024.
[17] Z. Mester, A.Z. Panagiotopoulos, Mean ionic activity coefficients in aqueous [40] J.C. Phillips, R. Braun, W. Wang, J. Gumbart, E. Tajkhorshid, E. Villa, C. Chipot,
NaCl solutions from molecular dynamics simulations, J. Chem. Phys. 142 R.D. Skeel, L. Kale , K. Schulten, Scalable molecular dynamics with NAMD,
(2015), https://doi.org/10.1063/1.4906320. J. Comput. Chem. 26 (2005) 1781e1802 doi:https://doi.org/10.1002/jcc.20289.
[18] A.S. Paluch, E.J. Maginn, Predicting the solubility of solid phenanthrene: a [41] M. Bhandarkar, A. Bhatele, E. Bohm, R. Brunner, F. Buelens, C. Chipot, et al.,
combined molecular simulation and group contribution approach, AIChE J. 59 NAMD User's guide namd, version 2.10b1 (accessed June 9, 2017), http://
(2013) 2647e2661 doi:https://doi.org/10.1002/aic.14020. www.ks.uiuc.edu/Research/namd/2.10b1/ug.pdf, 2014.
[19] M. Ferrario, G. Ciccotti, E. Spohr, T. Cartailler, P. Turq, Solubility of KF in water [42] W. Humphrey, A. Dalke, K. Schulten, VMD: visual molecular dynamics, J. Mol.
by molecular dynamics using the Kirkwood integration method, J. Chem. Graph. 14 (1996) 33e38 doi:https://doi.org/10.1016/0263-7855(96)00018-5.
Phys. 117 (2002) 4947e4953, https://doi.org/10.1063/1.1498820. [43] K. Vanommeslaeghe, E. Hatcher, C. Acharya, S. Kundu, S. Zhong, J. Shim,
[20] Z. Mester, A.Z. Panagiotopoulos, Temperature-dependent solubilities and E. Darian, O. Guvench, P. Lopes, I. Vorobyov, A.D. Mackerell, CHARMM general
mean ionic activity coefficients of alkali halides in water from molecular force field: a force field for drug-like molecules compatible with the CHARMM
dynamics simulations, J. Chem. Phys. 143 (2015), https://doi.org/10.1063/ all-atom additive biological force fields, J. Comput. Chem. 31 (2009) 671e690
1.4926840. doi:https://doi.org/10.1002/jcc.21367.
[21] M.J. Schnieders, J. Baltrusaitis, Y. Shi, G. Chattree, L. Zheng, W. Yang, P. Ren, [44] P.C. Hariharan, J.A. Pople, The influence of polarization functions on molecular
The structure, thermodynamics, and solubility of organic crystals from orbital hydrogenation energies, Theor. Chim. Acta 28 (1973) 213e222 doi:
simulation with a polarizable force field, J. Chem. Theor. Comput. 8 (2012) https://doi.org/10.1007/BF00533485.
1721e1736. [45] Wavefunction, Spartan ’10, 2010.
[22] A.S. Paluch, S. Jayaraman, J.K. Shah, E.J. Maginn, A method for computing the [46] W.L. Jorgensen, J. Chandrasekhar, J.D. Madura, R.W. Impey, M.L. Klein, Com-
solubility limit of solids: application to sodium chloride in water and alcohols, parison of simple potential functions for simulating liquid water, J. Chem.
J. Chem. Phys. 133 (2010) 0e13, https://doi.org/10.1063/1.3478539. Phys. 79 (1983) 926e935 doi:https://doi.org/10.1063/1.445869.
[23] F. Mou cka, I. Nezbeda, W.R. Smith, Chemical potentials, activity coefficients, [47] L. Kale, R. Skeel, M. Bhandarkar, R. Brunner, A. Gursoy, N. Krawetz, J. Phillips,
and solubility in aqueous NaCl solutions: prediction by polarizable force fields, A. Shinozaki, K. Varadarajan, K. Schulten, NAMD2: greater scalability for
J. Chem. Theor. Comput. 11 (2015) 1756e1764, https://doi.org/10.1021/ parallel molecular dynamics, J. Comput. Phys. 151 (1999) 283e312, https://
acs.jctc.5b00018. doi.org/10.1006/jcph.1999.6201.
[24] F. Mou 
cka, M. Lísal, J. Skvor, J. Jirs
ak, I. Nezbeda, W.R. Smith, Molecular simu- [48] J. Cuevas-Valenzuela, A.  Gonza lez-Rojas, J. Wisniak, A. Apelblat, J.R. Pe
rez-
lation of aqueous electrolyte solubility. 2. Osmotic ensemble Monte Carlo Correa, Solubility of (þ)-catechin in water and water-ethanol mixtures within
methodology for free energy and solubility calculations and application to Nacl, the temperature range 277.6-331.2K: fundamental data to design polyphenol
J. Phys. Chem. B 115 (2011) 7849e7861, https://doi.org/10.1021/jp202054d. extraction processes, Fluid Phase Equil. 382 (2015) 279e285 doi:https://doi.
[25] F. Mou cka, M. Lísal, W.R. Smith, Molecular simulation of aqueous electrolyte org/10.1016/j.fluid.2014.09.013.
solubility. 3. alkali-halide salts and their mixtures in water and in hydro- [49] D.J. Miller, S.B. Hawthorne, Solubility of liquid organics of environmental in-
chloric acid, J. Phys. Chem. B 116 (2012) 5468e5478, https://doi.org/10.1021/ terest in subcritical ( hot/liquid ) water from 298 K to 473 K, J. Chem. Eng. Data
jp301447z. 45 (2000) 315e318 doi:https://doi.org/10.1021/je990190x.
R. Gillet et al. / Fluid Phase Equilibria 472 (2018) 85e93 93

[50] E. Caddigan, J. Cohen, J. Gullingsrud, J. Stone, VMD User's guide. http://www. polynomial fitting techniques, J. Comput. Chem. 32 (2011) 134e141, https://
ks.uiuc.edu/Research/vmd/, 2003. doi.org/10.1002/jcc.21609.
[51] S.J. Watowich, E.S. Meyer, R. Hagstrom, R. Josephs, A stable, rapidly [60] M. Jorge, N.M. Garrido, A.J. Queimada, I.G. Economou, E.A. Macedo, Effect of
converging conjugate gradient method for energy minimization, J. Comput. the integration method on the accuracy and computational efficiency of free
Chem. 9 (1988) 650e661 doi:https://doi.org/10.1002/jcc.540090611. energy calculations using thermodynamic integration, J. Chem. Theor. Com-
[52] N.M. Garrido, A.J. Queimada, M. Jorge, I.G. Economou, E.A. Macedo, Molecular put. 6 (2010) 1018e1027, https://doi.org/10.1021/ct900661c.
simulation of absolute hydration Gibbs energies of polar compounds, Fluid Phase [61] J.R. Taylor, Chapter 3. Propagation of Uncertainties, en: An Introd. to error
Equil. 296 (2010) 110e115 doi:https://doi.org/10.1016/J.FLUID.2010.02.041. Anal. study uncertainties Phys. Meas, University Science Books, New York,
[53] W.F. Van Gunsteren, H.J.C. Berendsen, A leap-frog algorithm for stochastic 1997, pp. 45e79.
dynamics, Mol. Simulat. 1 (1988) 173e185 doi:https://doi.org/10.1080/ [62] R. Wolfenden, L. Andersson, P.M. Cullis, C.C.B. Southgate, Affinities of amino
08927028808080941. acid side chains for solvent, Water, Biochemistry 20 (1981) 849e855, https://
[54] S.E. Feller, Y. Zhang, R.W. Pastor, B.R. Brooks, Constant pressure molecular doi.org/10.1021/bi00507a030.
dynamics simulation: the Langevin piston method, J. Chem. Phys. 103 (1995) [63] D. Sitkoff, K.A. Sharp, B. Honig, Accurate calculation of hydration free energies
4613e4621 doi:https://doi.org/10.1063/1.470648. using macroscopic solvent models, J. Phys. Chem. 98 (1994) 1978e1988,
[55] U. Essmann, L. Perera, M.L. Berkowitz, T. Darden, H. Lee, L.G. Pedersen, https://doi.org/10.1021/j100058a043.
A smooth particle mesh Ewald method, J. Chem. Phys. 103 (1995) 8577e8593 [64] L. Michielan, M. Bacilieri, C. Kaseda, S. Moro, Prediction of the aqueous sol-
doi:https://doi.org/10.1063/1.470117. vation free energy of organic compounds by using autocorrelation of molec-
[56] N.M. Garrido, A.J. Queimada, M. Jorge, I.G. Economou, E.A. Macedo, Molecular ular electrostatic potential surface properties combined with response surface
simulation of absolute hydration Gibbs energies of polar compounds, Fluid analysis, Bioorg. Med. Chem. 16 (2008) 5733e5742, https://doi.org/10.1016/
Phase Equil. 296 (2010) 110e115, https://doi.org/10.1016/j.fluid.2010.02.041. J.BMC.2008.03.064.
[57] namd2_ti.pl, (s. f.). http://www.ks.uiuc.edu/Research/namd/utilities/(accessed [65] H.S. Ashbaugh, M.E. Paulaitis, Effect of solute size and solute-water attractive
June 9, 2017). interactions on hydration water structure around hydrophobic solutes, J. Am.
[58] S. Bruckner, S. Boresch, Efficiency of alchemical free energy simulations. II. Chem. Soc. 123 (2001) 10721e10728, https://doi.org/10.1021/Ja016324k.
Improvements for thermodynamic integration, J. Comput. Chem. 32 (2011) [66] J. Prausnitz, R. Lichtenthaler, E. de Azevedo, Molecular Thermodynamics of
1320e1333, https://doi.org/10.1002/jcc.21712. Fluid-phase Equilibria, third ed., Prentice Hall, Upper Saddle River, 1998.
[59] C. Shyu, F.M. Ytreberg, Accurate estimation of solvation free energy using

Вам также может понравиться