ChapterPDF Available

Computer simulations study of biomolecules in non-aqueous solutions.



Pure organic solvents or their mixtures with water are commonly used as artificial media for biotechnological application and for fundamental research on protein stability and folding mechanism. The molecular interactions between these environments and biological molecules are complex and variegated, and only recently we started to shade a light on their nature. Molecular modeling methods are effective tools to address the atomistic detail of these processes. In particular, Molecular Dynamics simulation is one of the most powerful and versatile tools to investigate the solvation of complex molecules. In the last few years, the number of publications on peptide and protein simulations in non-natural environments has rapidly increased. These studies are providing important information to shed light on the nature of non-aqueous solvent effects. In this chapter, the achievements and the future prospects in this field of computational biochemistry are reviewed by summarizing the most important theoretical results published in the last 15 years.
Computer Simulations Study of Biomolecules
in Non-Natural Medium
Danilo Roccatano1
Jacobs University Bremen
Campus Ring 1, 28759 Bremen (Germany)
Pure organic solvents or their mixtures with water are commonly used as artificial media for
biotechnological application and for fundamental research on protein stability and folding mechanism.
The molecular interactions between these environments and the biomolecules are complex and
variegated, and only recently we started to shade a light on their nature. A powerful tool to address the
atomistic detail of these processes is the use of molecular modeling methods. In particular, Molecular
Dynamics simulation is one of the most powerful and versatile tools to investigate the solvation of
complex molecules. In the last few years, the number of publications on peptide and protein simulations
in non-natural environments has rapidly increased. These studies are providing important information to
shed light on the nature of non-aqueous solvent effects. In this chapter, the achievements and the
future prospects in this field of computational biochemistry are reviewed by summarizing the most
important theoretical results published in the last 15 years.
Keywords: organic solvents and cosolvents, preferential solvation, protein stability, peptide
folding, peptide aggregation, enzymatic catalysis.
1 Address correspondence to this author at School of Engineering and Science, Jacobs University Bremen,
Campus Ring 1, 28759 Bremen, Germany; Tel: +49-421-200-3144; Fax: +49-421-200-3249; E-mail:
Computer simulations provide detailed atomic information of processes that would be difficult
(or even impossible) to study experimentally. One of those processes is the solvation
mechanism of biomolecules in aqueous and non-aqueous environments. Among the different
computational methods, Molecular Dynamics simulation has been used to study the effect of
solvent ever since its early infancy [1]. In contrast to the slow conformational dynamics of the
protein, solvent molecules can equilibrate faster around the protein surface and, in the time
scale of the simulation, relevant properties such as preferential solvation, diffusion constants
or reorientation dynamics, can be quantitatively estimated [2-4].
Among the common solvents, water was, for obvious reasons, the first solvent
molecule to be studied using computer simulation methods. Since then, many different models
have been proposed and used in simulations of biomolecular systems [5-8]. The systematic
study of the effect of non-aqueous solvents on protein thermodynamics properties begin in the
middle 1950 [2]. These studies thrived the interest to study proteins and nucleic acids in these
milieus with the aim to prove the effect of non-natural solvent conditions on the stability and
folding, and hence to use these knowledge for biotechnological applications [10-12]. For the
numerous solvents used in these experimental studies, although not in copious abundance as
for water, different computational models are available in the literature. Nevertheless, the use
of these solvents for simulation studies of proteins still remains very limited. Two main reasons
can be adducted to justify this lack of interest. The larger size and complexity of some of these
solvents, compared to water, can challenge the modelling of the interaction with biological
molecules and the required computational resources to conduct simulations in the presence of
macromolecules. The other reason is the lack of experimental data at atomic level on the
interaction of the solvent with the protein that can be used to validate simulation results. In the
past few years, availability of low-cost high performing computer clusters and advancement of
numerous experimental techniques boasted the number of theoretical studies of biomolecules
in pure organic solvent and their mixtures with water. This growing interest is also sustained
by the perspective of unrevealing the Holy Grail [3] of the protein folding mechanism with the
obviously tremendous implication in biomedical and biotechnological applications.
Actually, non-aqueous environment can stabilize/destabilize thermodynamically or
kinetically the folding of both secondary structure forming peptides and proteins. However, it is
not yet clear how biomolecules change their structural, dynamical and functional properties in
the presence of stabilizing or destabilizing solvents molecules and theoretical models could
offer important clues to understand the nature of the phenomenon. Furthermore, the
increasing use of enzymes in organic synthesis, as green and efficient replacement of
traditional synthetic approaches, has urged the improvement of catalytic activity of enzymes in
non-aqueous conditions. In this way, the reaction yield can be increased as effect of the
excess of substrate (when the solvent is also the substrate of the enzyme) and/or of increased
solubility of both reagents and products. The studies of Klibanov and co-workers have
recognized steadily that many enzymes can still be active in anhydrous or semi-anhydrous
conditions [4, 5] as well as in the presence of hydrophobic solvents. In addition,
crystallographic and spectroscopic studies of enzymes in non-aqueous solvents have
reinforced the hypothesis on the conservation of the aqueous structure of the enzyme in these
conditions [6, 7]. Interestingly, although it is still matter of debate, the conservation of the
solution three-dimensional structure has been also suggested for the extreme case of protein
in vacuum by the combination mass-spectrometry experiments and molecular dynamics
simulations [18-20].
In the case of mixture of organic solvents with or without water, the effect of the
cosolvents on the stability and activity of the protein become even more unpredictable and
dependent by the nature and concentration of the cosolvents.
The complexity of the matter is evident by the fact that despite the large amount of
experimental data accumulated over the years [14, 15, 21], a general unifying theory on
solvent effects on biomolecules still does not exist. Recently, we begin to shed light on the
microscopic nature of this complex phenomenon thanks to the synergy of novel experimental
methods and extensive and refined theoretical models based on MD simulations.
This chapter is an update and extension of a previous review on the topics published
by the same author in 2008 [8]. It contains a review of the progress made in the last 15 years
on protein and peptide simulations in organic solvent and their mixture with water by
addressing the following questions:
1) What is the state of art of MD simulations of protein and peptide in non-natural solvent
2) How results from the theoretical studies can give support to experimental data of
protein solvation?
3) What kind of solvation mechanism has been proposed on the bases of these
theoretical simulations for the most common solvents?
4) In retrospect, what can we expect from this growing sector of computational chemistry
and biochemistry?
The chapter is divided into two parts. The first part will provide a general introduction of
the MD techniques and of the current theories of cosolvent effects on biomolecular systems. It
is followed by a summary of current knowledge on structure, thermodynamic, dynamic and
functional properties of peptides and proteins in non-aqueous solvents. The second part of the
chapter will provide a summary of simulations studies of biomolecules in the most common
organic solvents (see Table I for a list). Finally, in the perspective section, a short outline of
the future development of MD simulations in this field will be given.
Comprehensive introductions on the MD techniques can be found in several reviews
[23-26] and textbooks [27-31]. In this section, the basic principles and the most common
methods to analyse the solvation properties of macromolecules from MD trajectories are
shortly introduced.
Molecular dynamics simulation is a useful tool for exploring the atomic properties of
large biomolecular systems. It is based on classical mechanics description of atomic motion.
The trajectory, e.g. the position in space and time, for a given atom i can be derived from the
integration of the Newton’s equation
where t is the time, ri, the position vector and mi the mass of the ith atom, respectively. The
force (Fi) acting on the ith atom, and generated by the interaction with the N-1 atoms of the
system, is obtained by the potential energy, V(r1,r2,r3, rN) that is a function of the
coordinates of the interacting atoms. The latter is a mathematical expression modeling, in
approximate manner, interactions among atoms in the system. In this approximation, atoms
are normally considered spherical particles without an internal structure. The missing
electronic degrees of freedom are implicitly included in the potential energy V(r1,r2,r3, rN)
that approximate, using analytical functions, the quantum nature of atomic interactions. The
use of analytical functions determines an enormous speeding up of the intermolecular
interactions calculation allowing the simulations of large molecular systems. In fact, the
absence of the electronic degree of freedom makes the required calculation proportional only
to the square of number of atoms but not to the 3rd or 5th power of number of electrons in the
system as for quantum mechanics methods [9, 10]. In this way, MD simulations can be used
to study systems with thousand to million atoms for time scale of nanoseconds to
microseconds [32-34]. The payback of this simplification is the restriction encountered when
studying electron-mediated process (like chemical reactions). Nevertheless, for the
investigation of solvation dynamics, this approximation does not reduce the precious value of
the possible outcomes.
The potential energy function is usually called empirical force field since it contains
several parameters (e.g. force constants, reference distances and angles, partial charges, van
der Waals parameters) that are normally derived from experimental data. In Fig. (1), a
; i=1N
schematic representation of the most common energetic terms contributing to the force field is
reported. The different force fields currently available for MD simulations share similar
functional forms. The energetic terms can be grouped as bonded and non-bonded one. The
former is usually represented by harmonic function describing bond and bond angles
vibrations, and by cosine functions for torsional interaction terms. For non-bonded
interactions, the van der Waals energetic contributions are normally modelled using the
Lennard-Jones function; while for the electrostatic ones, a Columbic potential with fixed partial
atomic charges is generally used.
As mentioned before, force field parameters are evaluated from experimental data or
from quantum mechanical calculations on small model molecules. Force constants for the
harmonic potential of bond and bond angle interactions are usually obtained from
Infrared/Raman spectroscopy and/or from QM calculations [11]. Partial charges are derived
from molecular electron densities obtained from ab-initio calculations. Finally, the σ and ε
parameters for the Lennard-Jones potential (see Fig. (1)) are derived from experimental data
or by fitting energy potential derived from QM calculations [11]. For solvent models, the initial
set of non-bonded parameters is usually adjusted to reproduce the physico-chemical
properties of the pure liquid. Density, enthalpy of vaporization, heat capacities and free energy
of solvation are the properties typically targeted in the optimization process. This optimization
can be extended to mixtures with a water model and their properties are normally investigated
over a complete range of concentrations (see, for example, Refs. [36-38]). For amino acids or
nucleic acids, the force field optimization relies on both thermodynamics and structural data
from crystallography, NMR and other spectroscopic techniques. An example of force field
parameterization based on thermodynamics data is the recent improvement of the GROMOS
force field (GROMOS 53A6 and 54A7 versions) [39, 40]. The new force field parameters were
first optimized to reproduce the thermodynamic properties of pure liquids of a range of small
polar molecules and solvation free enthalpies of amino acid analogous in cyclohexane. The
partial charges were then further modified to reproduce also the hydration free energies of the
same analogous [39, 40]. The new force field resulted in an improved accuracy with respect
the older one (G43A1) in simulations of proteins in solution [12].
A very important approximation underlying force field optimization procedure is the
transferability. For biomolecules, this assumption states that the parameters of a force field,
obtained by optimization of each isolated amino acids, are suitable also for the simulation of
the same amino acids in proteins. Fortunately, this approximation works in most of the case
very well. Indeed, this is quite surprising if one considers that the same amino acid into
proteins can be surrounded by completely diverse and not homogeneous environment
conditions than in the bulk solvent. The fortunate occurrence of this robustness in the force
field model makes MD simulations a wonderful tool for theoretical study of complex molecular
systems. Different types of force fields have been developed. These force fields consist of
libraries of parameters optimized for different classes of macromolecules (usually proteins or
nucleic acids) and for different organic solvent molecules. Commonly used force fields are the
following: GROMOS [13, 14] , OPLS [43, 44], CHARMM [45, 46] and AMBER [47-49].
The success of MD simulations is also strongly connected to the exponential increase
of computer performance. In this way, the upper limit of simulation scale is continuously
pushed towards larger systems [15] and longer simulation time [16]. Unfortunately, not all the
processes involving solvent diffusions have time scale accessible to the current performance
of computer simulations study [17]. For example, diffusion of small molecules into active sites
[53-55] (or in other protein cavities and tunnels) or across membrane [18] can be rare events
even for nanoseconds time scale. For these slow diffusive processes, different MD techniques
can be applied to speed up them in the time scale of the simulation. Three examples are the
replica exchange molecular dynamics [19], the steered molecular dynamics simulations [20]
and umbrella sampling methods [59, 60] (see also Ref. [21] for a general review on these
methods). The first method is based on multiple simulations of the same system but at
different temperatures with the possibility of switching the generated conformations between
the different runs according to an assigned Boltzmann criteria [19]. In this way, the
biomolecule is able to overcome higher conformational barriers and populate larger regions of
its conformational space. The original method and the variants have gained popularity
especially for peptide simulations [61-64]. In fact, for these small molecules, an extended
analysis of conformational space allows to study solvent effects from a thermodynamic
perspective [22].
Steered molecular dynamics is used to apply an external force on the system to move
a subset of atoms in a given direction. In this sense, it can be used to emulate atomic force
microscopy experiments and to perform in-silico nano-manipulation experiments. The method
is normally used by applying and a pulling velocity in a given direction by connecting single
atoms or molecules to a harmonic spring the elongation of which return the value of the mean
force exercised by the overall system on the moving atoms along the pulling direction [20].
The mean force acting on the pulled atoms is measured by the extension of a spring and it is
used to qualitatively and quantitatively calculate the potential mean force (PMF) of the process
[31, 59]. The method can be used to speed up the diffusion of ligands or solvent molecules
into buried active site [53, 65], along channel [23], membrane [67, 68] and estimate free
energy barrier of the processes [24].
Accurate determination of the PMF can be calculated using the umbrella sampling method
[25, 26]. In this method, a set of N separate MD simulations, in which a harmonic potential
(umbrella potential),
( )
is applied between the centre of mass of the target molecule and a reference group of atoms
are performed. This restrains the molecule at a distance ξi with a force constant of Ki. In each
simulation, the value of the distance ξi is changed from a maximum value corresponding to the
molecule in the bulk solvent phase to a minimum one at which the molecule is in the solvation
shell of the solute. From each simulation a histogram is calculated, representing the
probability distribution (Pi) along the reaction coordinate (ξ) biased by the umbrella potential.
The PMF from umbrella sampling histograms is then calculated using the weighted histogram
analysis method (WHAM) [27]. This method computes PMF based on all the MD trajectories
obtained along the reaction coordinate (ξ) sampling points. PMF are usually calculated along
distances but also angular coordinates can be used. Each trajectory provides, at sampled
position i along this coordinate, a distribution histogram (Pi) as result of the large set of
observations from the simulation. The PMF is calculated by weighting the contribution of all
these distributions at each point
PMF(ξi)= –kBT log(Pi), (3)
where kB is the Boltzmann constant and T is the temperature. The integral of the PMF curves
provides the free energies of the process under consideration. Depending by the length of the
sampling, the method can provide free energy estimations in quantitative agreement with
experimental data [26].
Another way to speed up the simulation of large biomolecules is the use of coarse-
grained MD methods. These methods consist in removing computational intensive atomic
degrees of freedom of the system by either clumps together group of atoms or using special
potential that provide a mean field effect of the removed atoms or molecules [10]. The mean
field is normally used to obtain implicit solvent models [72, 73]. However, in this review, only
MD simulations in explicit solvent will be considered.
Analysis of MD trajectories
After a MD simulation, trajectories of the atoms (coordinates and velocities over the
simulation time) are analysed to extract relevant structural and dynamical properties of the
simulated system. Thereafter, these results are directly or indirectly compared with
experimental data from different sources (see Fig. (2)). For structural properties, the root
mean square deviation about a reference structure, the radius of gyration, the secondary
structure and the clustering analysis of the generated conformations are some of the most
useful indicators for characterizing the folding state of peptides and proteins. The visualization
of single conformations or the entire MD trajectory is also a fast and effective way to analyze
the overall structural behaviour of the system. For this purpose, many programs for molecular
visualization (for an updated list of them see at are available. All
molecular structures showed in the figures of this review are generated using the public
domain programs VMD ( and ChemSketch
Dynamical properties are normally analysed using root mean squared fluctuations, and
principal component analysis [28]. The latter technique is used to separate the different
independent modes of the protein motion and evidences large scale slow motions from fast
local fluctuations of the molecule [28]. Readers can find more detailed information on the
different type of analyses of MD trajectories in other reviews [23, 74, 75]. Here, some of the
most common methods used to analyse solute/solvent interactions are shortly described.
Solvent Structure
At the atomic level, the dynamics of the liquid state is characterized by a continuous
diffusive tumbling of molecules. Despite the apparent chaotic motion of system, the presence
of atomic centred interaction potentials generates, on average, well-defined distance (and
angular) based structural correlations among the molecules. These correlations are measured
using different experimental techniques (e.g. X-ray diffraction, neutron scattering [29]). Using
the statistical mechanics theory of liquid, the experimental information can be mathematically
related to the solvent organization around a given atomic centre in term of pair correlation
functions. These functions describe the positional correlation of the centre a about the centre
b where the two centres are atoms as well as molecular centre of masses, and they are
related to intermolecular interaction potentials by the following expression [30]:
gab (r
1,r2)=Nb(Nb1) dr3rNa e
U(r,,rNa )
U(r,,rNa )
where the integral at the denominator is the partition function of the system and the one at the
nominator excludes coordinates r1 and r2 of the two centres (a, b) from the integration. For
homogenous and isotropic system as liquid solvents, only relative distances between two
molecules are meaningful. In this way, the Eqn. 4 is reduced to just a function of the distance
r between the two centres:
gab (r)=
where dr is the thickness of a spherical sector centred on the reference centre a,
local density of b in the spherical sector of thickness dr and radius r, and
the total density of
the liquid. The resulting distribution function gab(r) describes the local organization of the
molecules b around the center a in the system (see Fig. (3) for a schematic description).
Since, the definition of g(r) implies that 2
g(r)dr is proportional to the probability of
finding an atom in the volume element dr at the distance r from a given atom, for a
homogenous and isotropic fluid, n(dr) = 4
g(r)r2dr represents the average number of atoms (or
molecules) within a shell of radius r and thickness dr that surrounds the atom (see Fig. (3)).
The centres a and b can be single atoms of the centre of mass of a molecule.
However, for large solutes, as proteins, the organization of the solvents with respect the
centre of mass of the molecules provides little clues to the local solute solvation. Therefore, it
is more useful and informative to analyse the organization of solvent molecules around group
of atoms as each amino acids. In this case, the g(r) functions of the solvent are calculated with
respect representative atoms (usually the Cα atom) or about the centre of mass of the
Density distribution of solvent structure around proteins
A more convenient description of the solvent distribution of around large and complex
solute is the use of the spatial solvent density distribution. This function is obtained by
calculating the average number (number density) of water molecules present in each cell of a
three-dimensional grid fixed to the reference frame of the protein. By cumulating the number
of solvent molecules present in each grid point along the trajectory and, then, by averaging for
the values for the number of frames, volumetric density data of the space surrounding the
protein are obtained [31]. In Fig. (4), an example of spatial water density distribution
surrounding a peptide is shown. On each slice of the volumetric data, the local density in the
form of three-dimensional elevation is reported [32]. From these data, it is possible to localize
preferential hydration sites and correlate them with, for example, position of crystallographic
waters [77, 79, 80]. In this way, MD simulations data can complement information from
crystallographic [33] and NMR data [34] to identify ligand binding site on protein surfaces [35].
Residence time and diffusion coefficients
Dynamical properties of solvent molecules close to protein surfaces can be measured
using a wide range of techniques. Multidimensional NMR techniques are the most common
methods to measure the solvent relaxation times and diffusion coefficients [17, 36]. In 2-D
NMR experiments, a pair of spins (e.g., a protein hydrogen and a selected paramagnetic
nucleus of the solvent molecules) can interact by dipolar coupling if the distance between
them is less than 0.5 nm. The decay of the excitation will depend, among other things, on the
exchange of the bound solvent molecule with the bulk phase. This relaxation process can be
studied only if it is slower than the characteristic rotational correlation time (molecular
tumbling) of the protein in solution. For a protein of small to medium size, this time is about
few nanosecond [37]. This restriction limits the investigation to only closely bound (internal)
solvent molecules excluding more mobile ones located on the protein surface [17]. However,
surface solvation, occurring at shorter time scale (less than tens of picoseconds), can still be
addressed by magnetic relaxation dispersion [4, 52] and by time-resolved spectroscopic
techniques [17] both providing relaxation dynamics with femtoseconds to picoseconds
resolution [4, 86].
Both short and long time scale relaxation processes can be directly calculated from
MD trajectories. The residence time of solvent molecules in protein solvation shells provides
precious information about the solvation mechanisms and dynamics. These data can be
obtained from MD trajectories in different ways [38]. A common approach is to use the survival
time correlation function [87, 88]:
where the probability function, p
,j (t, t + t') is a binary function that adopts a value of one if the
solvent molecule labelled j has been in the referred solvation shell around site a from time t' to
time t + t', without escaping between this time interval (or leaving during this interval the shell
for a time not longer than a t*), and zero otherwise. The value of p
,j(t, t + t') is averaged over
time and over all solvents molecules. Since this function is a time correlation function, P
,j(t =0)
gives the average number of solvent molecules belonging to the solvation shell of site (i.e. the
coordination number), and P
,j(t) gives the average number of solvent molecules that still
remain in the solvation shell after a time t from the time they first entered the shell. The
relaxation trend of P
,j(t) provides information about the local dynamics of the solvation
molecules. The value of P
,j(t) is usually approximated to single or double exponential function
is the relaxation time constant and the subscript symbols s and l referring to a short
and long relaxation time, respectively. The relaxation time for the water shows a non-
exponential behaviour that can be fitted using the Kohlraush-Williams-Watt stretched
exponential function [39] :
This function is commonly used to describe relaxation process involving diffusion in
corrugated self-similar energy landscapes. The self-similarity of the protein surface provides
an interpretative key to this anomalous diffusion process [39]. An example of residence time
study for non-aqueous solvents (methanol, DMSO and chloroform) on protein surface is given
by Kovacs et al. [40].
Hydrogen bond dynamics between solvent molecules and proteins is also used to
estimate the residence time. In this case, the function h(t), having value 1 or 0 according to the
presence or absence of a hydrogen bond between the solvent molecule and the protein
surface, is used. The correlation function of h(t) is then defined as :
where the brackets indicate the average along the trajectory. The fitting of CH(t) by an
exponential function will provide the relaxation time for the hydrogen bond network [41]. This
approach is used only for molecules having donor or acceptor atoms (for example alcohols or
urea) forming hydrogen bonds.
Another description of the solvent mobility on the protein surface is the calculation of
its self-diffusion coefficient. This dynamical property is directly correlated to diffusivity
measurements from NMR and other spectroscopic techniques [17]. The diffusion coefficient
on the protein surface from MD simulation trajectories is calculated using the Einstein relation
6Dt =1
[ ]
where ri(t) is the coordinate vector of the particle ith at time t, ri(0) the coordinate vector of the
particle ith at time t=0 and N the total number of particles [30]. The effect of the different
regions of the protein surface on solvent diffusion coefficient can be also calculated with
respect the nodes of a grid in a reference frame moving with the protein. In each grid node
(with indices u, v, w), the diffusion coefficient is then given by
6Duvw =1
r (t2)!
r (0) 2!
r (t1)!
r (0) 2
( )
where the values of t1 and t2 are fixed (for water molecules) to 1 ps and 2 ps, respectively, on
the assumption that a normal diffusion regime is reached by the solvent molecules after 1 ps
of simulation [31, 42]. The average contact time of solvent with proteins is of the order of few
tens of picoseconds. Therefore, several nanosecond long MD trajectories are required for a
good estimation of the diffusion from simulations [43].
Besides translational diffusion, rotational diffusion provides important information on
the orientation of the molecule with respect to the protein surface. For polar molecules, this
property is calculated by the autocorrelation function of the dipole moment of the solvent
where Pl is the Legendre polynomial of the order l, µ is the unit dipolar vector and the brackets
indicate the average along the trajectory [38]. The studies of diffusion properties of water on
protein surfaces have shown anomalous behaviour due to the corrugated nature of this protein
region (for a detailed discussion see [38, 39]).
Several MD studies on water diffusion around proteins and nucleic acids have been
reported in literature [79, 80, 87, 92, 94]. Contrary, the dynamics properties of non-aqueous
solvents around biomolecules have been less extensively analysed (see examples [4, 90, 91,
Preferential solvation
The variegated molecular surfaces of peptides and proteins create complex selective
interactions with the solvent and the different cosolutes present in the same solution. This
different interaction propensity is referred to as preferential binding” or “preferential solvation
[44]. This is a thermodynamics property and it is related to the activity coefficient of the
cosolvent molecules in the proximity of the solute surface. At equilibrium, it is related to the
variation of the chemical potential of the protein when transferred from the water to the
is the transfer free energy of the protein from pure water into the mixed solvent
system, m is the molarity, and subscripts X and P identify the cosolvent and the protein,
respectively. The first term in the relation can be determined experimentally for mixture of
water and cosolvent. It measures the variation of the chemical potential at mp0, thus it
describes the perturbative effect of the solvent on the water structure. The second term is the
“preferential binding” coefficient:
This coefficient measures the excess binding of the cosolvent on the protein surface. Using
the molecular statistical mechanics theory of liquids, it is possible to connect this coefficient
with the effective number of cosolvent molecules present on the protein surface:
XP =nx
In this expression, n indicates the number of specific molecules named in the subscripts, and
the superscripts S and B indicate the regions (S = surface, B= bulk) where the molecules are
located. Baynes and Trout have proposed a practical approach to calculate the values of ΓXP
from MD simulations [43] based on two approximations introduced by Smith {Pierce, 2008
#942;Smith, 2004 #941}. They defined the instantaneous preferential binding coefficient as:
XP =mX
where ΓXP(t’) is the instantaneous preferential binding coefficient at the simulation time t’ as:
XP =nx
The number of molecules present in the two regions are evaluated from the simulation by
using radial distribution functions of the water gw(r) and cosolvent molecules gX(r) with respect
to protein centres (e.g. side chains, backbone atoms). In this way, the relation (14) can be
transformed as:
XP =mX
XP =nx
( )
where the last integral is calculated relative to a surface patch region of the protein and until a
distance r* from the surface (in the bulk region) at which gx(r*)=gW(r*)=1 and the contribution to
ΓXP become null. Baynes and Trout have applied their method to the RNAse T1 and RNAse A
with glycerol and urea as cosolvents [43]. The results of their studies showed that the effect of
cosolvents is mainly given by the first solvation shell but also the second shell can give a
significant contribution to the preferential binding coefficients [43]. Furthermore, for the two
cosolvents preferential binding is larger on more hydrophobic regions of the protein.
The preferential solvation can be also evaluated as local concentration of cosolvent.
The concentration (% v/v) of the cosolvent molecules around the peptide residues, named
local cosolvent concentration (LCC), is evaluated from the cumulative number of water nw(r)
and cosolvent nc(r) molecules within a distance r from the Cα's of each residue using the
following relation:
[ ]
where Vmc and Vmw are the average excluded volumes for cosolvent and water molecules,
respectively. In Fig. (5), an example of local density distribution of HFIP around the peptide
melittin is shown [45].
Thermodynamics properties
Computer simulations are also used to calculate thermodynamics properties of protein
and peptides in solution [98-101]. However, accurate determinations of thermodynamic data
rely on multiple and extended simulations and, for protein, even small one, this task can
require expensive calculations [46]. For peptides, the situation is considerably better because
their size is smaller compared to proteins and, therefore, the required computation effort is
more affordable. Nevertheless, exhaustive sampling of the conformation space of small
peptides can still be a challenging issue [75, 102].
An alternative approach is the use of the extended Kirkwood-Buff theory of mixture
solutions by combining experimental and theoretical data {Shimizu, 2006 #323;Smith, 2004
#49;Pierce, 2008 #942}. This theory, based on the statistical mechanics, connects
thermodynamics properties to atomic structural correlation functions by the Kirkwood-Buff
integrals [47]:
Gij =4
r2gij (r)1
[ ]
These integrals contain pair distribution functions gij(r) of each atomic species (i) with respect
to other ones (j) present in solution. The gij(r) functions are estimated using Monte Carlo or
MD simulations. Once the value of Gij is known then it is possible to derive thermodynamics
properties as, for example, partial molar volume, activity coefficients and free energy of
transfer [47]. This approach has been used to garner insights into the nature of
cosolvent/water mixtures of different osmolytes. The method has been used to relate the
different terms of the free energy to structural properties and to study the enthalpy-entropy
compensation in the solvation of small solutes by solvent mixtures [48, 49]. Another example
of application is the study of the solvation free energy of the Leu-enkephalin, a opioid
pentapeptide of sequence YGGFL [50]. In this study, the salting-out effect has been
reproduced and quantified by calculating the population shift to the unfolded state of the
peptide conformation in the presence of NaCl 2M [50].
The three-dimensional reference interaction site model (3D-RISM) is another recently
developed statistical mechanics approach based on the integral equation theory of molecular
liquids [51]. It also combines MD simulations and the integral equation theory to predict
thermodynamics properties of solutes in solution. The method successfully predicted the
hydration structure and thermodynamic quantities of different peptides and proteins [51].
Recently, it has also been applied to study cosolvent (urea and glycerol) effects on the partial
molar volume change of Staphylococcal nuclease associated with pressure denaturation [52].
Non-aqueous solvents can be classified according to their effects on the protein and,
for mixtures, on the water structure (see Fig. (6)). In the first case, we have two general
Compatible solvents. This class represent all solvents and/or cosolutes that can
stabilize folded protein structure. These types of molecules are also called compatible
osmolytes because they have an important role in the regulation of osmotic balance in
the cell. Actually, they are present at high concentrations in living cells and the function
is to protect them by external protein denaturating agents (pressure, temperature,
salts). Typical examples of this class of solvents or cosolvents are polyols (e.g.,
glycerol), sugars (e.g., trehalose), polymers (e.g., polyethylene glycol), amino acids
(e.g. arginine or glycine betaine) .
Incompatible solvents. This class of molecules comprises all protein denaturants as,
for example, urea and guanidinium cation.
The second classification is based on the effect of solvents on the water structure:
1) Chaotropic solvents. Water structure perturbing solvents are called chaotropes
(from Greek: cháos (χάος), confusion and tropos (τρόπος), to turn) that means
order-breakers. Those solvents interfere with the formation of the hydrogen bond
network among water molecules.
2) Kosmotropic solvents. Water structure promoting solvents are called
kosmotropes (from Greek kósmos (κόσµος), order and tropos (τρόπος), to turn) that
means order-makers. Those solvents are able to promote ordering in the hydrogen
bond networks of water.
The protein stabilization/destabilization mechanism(s) of both solvent types is not yet
completely clarified. Experimental techniques usually provide indirect clue on the molecular
process. For this reason, molecular simulations are used to complement the experimental
data and provide realistic models of the molecular mechanisms.
Pure solvents
The proposed solvation mechanism of proteins in pure organic solvents requires the
(partial) dehydration of the molecule and the substitution of water molecules with the non-
aqueous solvent [4]. However, in these conditions, the protein can fully retain its structural and
functional properties only if lubricant water molecules (mainly structural water molecules) are
conserved in the protein structure. The importance of structural water molecules was recently
evidenced by the recent electrospray ionization mass spectrometry experiments [53]. The
results of these experiments and of MD simulations studies [18] have shown that proteins,
evaporated in vacuum by matrix-assisted laser desorption/ionization (MALDI) techniques,
retain their solution/crystal-like three-dimensional structure if they conserve part of the
hydration water molecules [110, 111].
It was also proposed that pure organic solvents could influence the protein dynamics by
constraining the conformational space accessible to the molecule [4]. This hypothesis is based
on the so-called ligand memory effect that consists in an enhanced catalytic activity of the
lyophilized protein in the presence of substrate upon solubilisation in organic solvent [4]. The
proposed mechanism for this effect supposes that the substrate binds the protein in the
lyophilized state forming an activated conformation state. Upon the solubilisation, the reduced
conformational flexibility in the organic solvent constraints the protein in this reactive state
enhancing the substrate binding affinity and activity [4]. A recent MD simulation study [54]
have partially evidenced the role of the water as lubricant for catalytic part of the proteins
although the results are still far being conclusive.
In this review, I will limit to report the most relevant results of protein in pure solvents from
computer simulations perspectives. The reader interested to have more general information
can refer to several excellent reviews published on this topics [4, 5, 55].
Mixed solvents
As reported in the previous paragraphs, the differential interaction of the protein with
the components of mixture is described in terms of preferential solvation. Preferential solvation
can be studied with different experimental techniques such as vapour pressure osmometry
[56], time-resolved fluorescence measurement [115], ultrafast two-dimensional infrared (2D-
IR) spectroscopy [57], X-ray crystallography, and time-relaxation NMR spectrometry [4, 52].
The latter technique measures the local solvent concentration in contact with the magnetized
nuclei. X-ray diffraction studies can also provide information on the preferential binding of
cosolvent molecules in the crystal structure. This approach is also used to localize preferential
binding of small molecules for drug design [117, 118]. In fact, the binding sites localized in the
crystal structure are usually the same as in solution (see for example [119-122]).
Preferential solvation can affect structural, dynamical and functional properties of
proteins and peptides in solutions (see Fig. (7)). For the sake of simplicity, the different effects
on biomolecular systems are classified into three intertwined groups: 1) effects on stability and
folding; 2) effects on aggregations; 3) effects on enzymatic activity. For the last one, we
consider more specifically preferential solvent binding that can interfere with enzymatic
Effect on stability and folding
Despite the large amount of experimental data cumulated in the last decades, the
mechanism by which organic solvents or cosolvents stabilize or destabilize proteins or
peptides remains a matter of debate [4, 55]. The chemical environment is an important factor
in driving protein folding. Many proteins and peptides, being poorly soluble in water, are stable
in their folded state only inside the lipid bilayer of cellular membranes. The different
mechanisms, formulated to explain the stabilizing effect, advocate direct and indirect effects.
The latter ones are generally due to the nature of interaction between the solvent and water.
These interactions can either reinforce the hydrogen bond network of the water (kosmotropic
solvents) or destabilize them (chaotropic solvents). Recently, it has been shown that
traditionally chaotropic molecules (e.g. urea [58]) does not necessarily perturb the water
structure and the claimed indirect mechanism for their action seems not to be a valid option.
Direct effects are owing to the binding of solvent molecules on the surface of peptides
or proteins. The direct interactions can favour the stabilization/destabilization of the secondary
structure of the biomolecule in different ways. The enhancement of stabilizing interactions
(e.g. hydrogen bonds) by coating the biopolymer with a cosolvent layer, the reduction of the
dielectric constant on molecular surface and the consequent stabilization of hydrophobic
interactions between side chains, are some of the most reasonable explanations [59].
To find clues to support these hypotheses, many experimental studies on biopolymers
have been performed in aqueous mixtures with different organic cosolvents. Cosolvents such
as DMSO, acetonitrile, DMF and short chain alcohols (in particular fluorinated ones), are
commonly used as stabilizing or denaturing agents. Their stabilizing properties are a
consequence of the peculiar physico-chemical properties of the mixtures they form with water.
For example, computer simulations of the fluorinated alcohol TFE and HFIP (both promoting
stabilization effects on isolated α-helix or β−hairpins forming peptides in solution) have
provided strong evidence in support of a coating mechanism as one of the most important
stabilizing factor [97, 124].
For proteins, although several MD simulation studies in pure solvents have been
performed [60, 61], relatively smaller number of publications have addressed effects on the
stability or enzymatic reactivity in non-natural solvent mixtures. Cosolvents can indeed
dramatically affect also the structural properties of proteins and, therefore, theoretical
investigations can help to understand their role in the protein stability and catalysis.
Protein aggregation
Peptide and protein destabilizing cosolvents are used to induce protein misfolding [62].
This process is accompanied by the formation of highly structured thread-like aggregates
termed amyloid or amyloid-like fibrils. The study of protein aggregation is very important for
understanding a variety of human diseases that are now thought to be associated with
formation of highly organized protein aggregates [62]. Among cosolvents, fluorinated alcohols
(TFE and HFIP) highly promote fibrils formation by promoting change and/or stabilization of β-
strands conformations in the secondary structure of different proteins [63].
Several MD studies have been conducted to investigate the misfolding and
aggregation mechanism [64]. Isolated amyloid beta and prion H1 peptides have been studied
in fluorinated alcohols to gain insights on the effect of solvent induced secondary structure
stabilizations [65]. However, computational studies of peptide aggregation are by far quite
limited to pure water solutions [66].
The following general conclusions summarize the experimental achievements in non-
aqueous enzymology: 1) single or clusters water molecules in the interior or on the protein
surface are very important for enzymatic catalysis; 2) polar solvents tend to strip these water
molecules off, thereby reducing the catalytic activity; 2) more hydrophobic solvents tend on the
contrary to improve the catalytic activity by retaining the molecules bound to the proteins
cause of their hydrophobicity.
The main interpretation of these phenomena is related to the increased rigidity of the
enzyme having most of the lubricant water molecules removed. In pure hydrophobic solvents
(e.g. hexane), the structure of the enzyme is retained but the flexibility can be reduced. This
effect promotes the ligand-induced memory effect (see above): the protein remains kinetically
trapped in a configuration favourable for the catalytic function of the enzyme [4]. On the other
hand, mixtures of polar organic solvents with water can be very complex and, as consequence
of the point 2 process, deplete the catalytic activity and denature the protein [67]. Therefore,
for these solvents a delicate equilibrium between the amount of water and cosolvents is
required for tuning both catalysis and protein stability.
Theoretical studies of the protein catalytic activity in pure organic solvents support the
experimental evidences on the role of these solvents in changing enzymatic activity, and of
conserved water layers in maintaining the catalytic efficiency of the enzyme [60]. Combination
of classical MD simulations with QM methods [125, 132] has been used for these studies. In
this approach, the active site geometries of enzyme-substrate complex generated using
classical MD simulations in explicit solvents, are used to study the mechanism of reaction by
QM methods. This approach has been successfully used to study, for example, the
regioselectivity of subtilisin in DMF [68] or more recently the γ-Chymotrypsin in acetonitrile
media [69].
Finally, MD simulations studies of enzymes in water/cosolvent mixtures have provided
molecular evidences in support of other solvent mediated effects, as water stripping [70],
reduced protein motilities and the pH memory effect [71, 72].
The solvents used in biochemical and protein bioengineering studies are numerous.
They have different properties and they are selected according to their effects on protein
stability and activity. In the latter case, the polarity is used to enhance hydrophobic condition
of the solution that in turn increases the solubility of reagents and products.
As mention before, computer simulation studies of biomolecules in non- aqueous
environments is limited by the complexity of solvent structure and the amount of physical data
available to assess and refine the quality of the solvent model. In Table I, a list of organic
solvents used to study using MD simulations their effects on protein or peptides is reported.
The most investigated solvents and cosolvents are: methanol, DMSO, fluorinated alcohols,
urea, trehalose, glycerol, TMAO and more recently ionic liquids. The other solvents like
tetrahydrofuran, ethanol, acetone, DMF, acetonitrile are also studied but so far in less
extensive way.
Methanol and ethanol
Pure methanol and mixture with water are commonly used to increase hydrophobicity
of aqueous solution and to solubilise the structure of hydrophobic amino acid rich peptides
and proteins. It was one of the first non-aqueous solvents after water to be deeply studied with
MD not only for the availability of experimental data but also for the advantage due to its low
density that permits to reduce the number of solvent molecules per unit of volume compared
to water and, therefore, to extend the time length of MD simulations. This property is
particularly useful to study peptide folding mechanisms on long simulations time scale [73].
Pure methanol is also used to mimic membrane environment and study membrane
soluble peptides. For example, membrane soluble α-helix forming peptides (e.g. alamethicin
[138-140], melittin [138, 141, 142] and pulmonary surfactant lipoprotein C [90, 143]) have
been studied in pure methanol. For these peptides, simulations have shown that methanol
molecules have a slight tendency to stabilize the α-helix contents compared to pure water. An
alcohol chain length dependence of Melittin secondary structure stabilization was also
observed using very short time scale simulation in pure methanol, ethanol, n-propanol and n-
butanol and at different temperatures [74]. These results are consistent with the larger
stabilization of α-helix observed when peptides are embedded in lipid bilayers [139, 142, 144,
An interesting comparative MD NMR study of the hormone [Val5]-Angiotensin in 25%
(v/v) water/methanol has been recently published [75]. The results of the simulations
confirmed the micro-heterogeneity of the methanol/water mixture observed in the NMR
experiments. In this condition, most peptide hydrogen atoms are preferentially solvated by
interactions with methanol molecules. The peptide BBA5, a family of peptides designed to
exhibit a ββα fold in water, was also recently studied in methanol/water solutions using a
combination of NMR and molecular dynamics. The results evidenced the dual tendency of
methanol to reinforce the secondary structure of the peptide and also, by weakening
hydrophobic core interactions, to expand the hydrophobic core [76]. A comparative MD
simulation study of the antimicrobial hlF1-11 derived from the first 11 N-terminal amino acids
of the human lactoferrin protein in different solvent has been also recently reported [77]. The
peptide has been studied in pure methanol as well as in the membrane mimetic 4:4:1
methanolchloroformwater mixture. Paschek al. have study the effect of the changing
methanol/water mixture composition on the shift of the amide I band of a 20 amino acid alpha-
helical AK peptide [78].
Several simulation studies of non-natural peptides (in particular β-peptides) in methanol
solution have been published [150-158] (see also reviews [73, 79]). They are mainly focused
on the dynamic behaviour (for example, the reversible folding) of these peptides in methanol
and other non-aqueous environments. Fluorinated β-peptides have been also recently
investigated with MD simulations in pure methanol [80].
Pure ethanol and mixture with water have been used to study proteins [160, 161] and
different peptides [162-164]. Fioroni et al. have performed an interesting study by combining
NMR data and MD simulation of a tetra-peptide in EtOH/water 30 % v/v. The results of
simulations confirm the experimental evidences that ethanol molecules (in contrast to TFE
molecules) do not show direct interactions with the peptide surface [81]. Lousa et al. have
recently reported an interesting comparative MD study of two proteases (pseudolysin and
thermolysin) in water and in mixtures of water/ethanol at 25 % (v/v). The MD simulations were
extended up to 1 µs. The results of the simulations evidenced a differential stability (with
pseudolysin more stable then thermolysin) in ethanol/water mixtures of the two proteins in
good agreement with experimental data. The different stability seems related to a higher
preferential interaction of ethanol molecules with the thermolysin than with the pseudolysin.
Fluorinated alcohols
Fluorinated alcohols such as HFIP and TFE are commonly used to study
conformational states of small peptides and proteins [82]. These molecules have the capability
of increasing the content of secondary structure (α-helix, β-hairpins) in proteins and secondary
structure forming peptides [82]. The stabilizing effects of these alcohols change according to
their size and chemical structure [82]. Recent studies [83, 84] have demonstrated a stronger
secondary structure stabilizing effect of HFIP in comparison to TFE. One reason for this effect
is related to the low polarity of the solvent [85]. Among the possible mechanisms, preferential
solvation of the folded state by fluorinated alcohols is the most accredited [86]. According to
this hypothesis, fluorinated alcohols act within the context of a pre-existing helix-coil
equilibrium, and the preferential interaction of TFE with the folded state shifts the equilibrium
toward more structured conformations [86]. Alternative mechanisms have been proposed to
explain the stabilizing effect, for instance reinforcement of hydrogen bonds between carbonyl
and amidic NH groups by the removal of water molecules in the proximity of the solute [87]
and/or lowering of the dielectric constant of the medium [82]. Furthermore, the low polarity of
the cosolvent allows formation of micro-clusters in aqueous solution. The evidence for micro-
heterogeneity in TFE/HFIP water mixtures comes primarily from NMR and FTIR spectroscopic
studies [167, 172-175]. In addition, SAXS studies indicate that cluster formation is a function
of the concentration of both TFE [83] and HFIP [176, 177]. In both solvents, the micro-
heterogeneity (see for the TFE Fig. (9)) reaches a maximum value at a cosolvent
concentration of 30% v/v [166, 167], that is the concentration normally used for the study of
biopolymers. HFIP and TFE clusters in water have been proposed to assist the folding of
secondary-structure elements by providing a solvent matrix that promotes hydrophobic
interactions between amino acid side chains [83, 88]. This hypothesis has found experimental
support in several NMR studies involving small peptides in TFE and HFIP mixtures with water
[4, 165, 179, 180].
Different models of fluorinated solvents are available for studying biomolecules in
solutions [37, 181, 182]. We have proposed models of TFE and HFIP specifically optimized to
reproduce the physico-chemical properties of the neat liquid and mixtures with water [89, 90].
For the HFIP model, recent X-ray diffraction, small angle neutron scattering and NMR
spectroscopy measurements have further validated the good quality of this model [91]. In
particular, our model was able to predict the structural transition between the two regimes of
predominant solvent clusterization observed in the experiments [91]. Both the TFE and HFIP
models have been used for simulations of secondary structure forming peptides. In particular,
α-helix forming peptide melittin was studied in mixtures at 30% v/v of TFE [59] and HFIP [45]
with water. Melittin is a 26 amino acids amphiphilic peptide from honey bee venom, which is
largely unstructured in water but adopts a α-helical structure upon addition of TFE [92] or
HFIP [93, 94]. The circular dichroism ellipticity at 222 nm reaches a maximum at around 10%
v/v for a mixture of HFIP/water, with a plateau between 10 and 40% v/v [83]. This stabilizing
effect has been theoretical investigated using comparative simulations in water and in 30% v/v
TFE/water mixture [59]. These studies have provided for the first time a theoretical support to
the experimental hypothesis of the coating effect by the fluorinated solvents on the structure of
α-helix forming peptides in solution [165, 179, 180]. The coating of the peptide surface
reduces its interaction with water, stabilizing the helical conformation. A similar study on
Melittin in 30% v/v HFIP/water mixture showed that the main mechanism to limit the unfolding
of the peptide in water is the preferential solvation of the peptide surface by cosolvent
molecules (see Fig. (5)) [94].
The stabilization effect induced by fluorinated solvents has also been shown in other
MD simulation studies of different peptides [129, 185-193] and proteins [194-196].
Another broadly used organic solvent in non-aqueous enzymology and protein
bioengineering [95] is the dimethylsulfoxide. For example, DMSO/water mixtures are used to
increase the solubility of hydrophobic protein substrates. Pure DMSO or in its mixture with
water is also used as cryo-protective agent for X-ray crystallography, to study protein or
peptide folding, to enhance membrane permeability, and to promote cell fusion. DMSO solvent
is also a good and convenient solvent to mimic the lipid bilayer hydrophobic environment.
Several studies of peptides and proteins have been performed in pure DMSO and
mixtures with water (see Table I and II). The simulation results have been compared with
experimental data to provide insight on the structure and folding mechanism of peptides and
proteins (see also [73] for a review).
The heme domain of Cytochrome P450 BM-3 from Bacillus megaterium [198] has
been used as protein model for theoretical and experimental investigations in DMSO/water
mixture. The study of hP450 in aqueous solution has been investigated by various research
groups using MD simulations [199-204] (see also the comprehensive review on P450 BM3
enzyme by Whitehouse et al. [96]). Most of these studies have been focused on the access of
natural substrates into the active site cavity [97, 98]. In a recent article, we have analyzed the
structural and dynamical properties of hP450 in pure water and in a 14 % v/v DMSO/water
mixture on nanosecond time scale using MD simulations. We have found that DMSO, at this
concentration, does not unfold the protein during the simulation time. However, the structural
and dynamical behaviour of the protein shows significant differences in regions important for
regulating catalytic activity. In particular, DMSO tends to modify the structure in
correspondence to helices F and G, especially in the FG loop. In pure water, these regions
showed less variation form the crystal structure but larger fluctuations [99]. Furthermore,
DMSO molecules distributed on the protein surface (see Fig. (10)) and preferentially
associated to concave protein regions with no evident correlation with properties of surface
residues. This might be due to the size and dipolar aprotic character of the solvent, which
tends to maximize the interaction with amino acids present in cavities and grooves of the
protein surface. In addition, the DMSO diffusion into the active site was kinetically very slow; in
a simulation time of 15 ns, none of the DMSO molecule was able to access the active site
tunnel. Longer simulations and/or use of enhanced sampling of the conformational space of
P450 BM-3 might allow the study this diffusion process. The effect of DMSO molecules on the
active site of P450 BM-3 heme domain was also investigated to elucidate the possible
mechanisms of cosolvent induced inactivation [100]. In fact, different experimental clues have
suggested that the protein inactivation may be determined by a possible interaction between
DMSO molecules with residues in the P450 BM-3’s active site. Therefore, MD simulations with
different DMSO molecules positioned inside the access channel to the active site of the
enzyme were performed. The results of these simulations supported the hypothesis that
activity inhibition might result from the displacement of the heme coordinating water by DMSO
molecules [100]. Experimental measurements of protein activity have shown that the
displacement mechanism is less effective for the wild type than for the F87A mutant. The
F87A mutation replaces the bulky phenyl group of F87 with a methyl increasing the DMSO
accessibility to the heme iron (see Fig. (11)).
Recently, these hypotheses, formulated from the results of MD simulations, have been
validated by the determination of the X-ray structure for the hP450 wild type and its F87A
mutant at co-crystallized at different DMSO concentrations [101, 102]. The presence of a
DMSO molecule coordinating the heme iron at the 6th position and the absence of structural
unfolding in the crystal structure clearly indicate an inactivation mechanism of P450 BM-3
mediated by the direct interaction of the DMSO with the heme iron.
Urea and Guanidinium salts
Urea and Guanidinium molecules are common denaturating cosolvents for proteins.
The nature of their effect is still controversial and direct and indirect mechanisms have been
proposed for these molecules [207].
Concerning the indirect, chaotropic mechanism, recent studies have shown that urea
molecules, even at large concentration typically used for protein denaturation, do not perturb
the water structure as proposed by the indirect mechanism [58]. These results indicate that a
direct interaction mechanism of urea with the protein is more likely to occur. The direct
mechanism hypothesis is based on the amphiphilic nature of urea that permits interactions
with hydrophobic and hydrophilic parts of biomolecules. This dual effect of the urea is
supported by results of several MD studies on peptides and proteins (see Table II and III). The
preferential interaction of urea, with more hydrophobic solvent exposed amino acids of
proteins, increases the polarity of the protein surface that would weaken protein stabilizing
hydrophobic interactions [208-211]. On the other hand, the charged urea amino groups can
compete in the hydrogen bonds with pivotal residues for the protein structure with the
consequence of altering their structural functions and thereby compromising the stability of the
protein structure [104, 212-216].
Gmd+ ion is normally used as chloride or tiocianate salt. As for urea, the mechanism
of protein denaturation induced by these salts is still unclear, though in the last few years a lot
of experimental and theoretical efforts have been dedicated to solve this hurdle [207].
Experimental SANS studies of GdmSCN showed very poor hydration of both Gdm+ and
tiocianate, supporting a direct interaction mechanism [103]. In addition, differently from Urea,
Gmd+ cations have a low tendency to form aggregate in solution. Gdm+ ion are characterized
by a rather hydrophobic surfaces that may interact with similar protein surfaces to enable
protein denaturation [104] and a positive charge that can form strong ion pair with negatively
charged carboxylate groups of side chain amino acids. These effects have been investigated
with simulation of different proteins and peptides (see Table II and III).
Glycerol, trehalose, glycine betaine and TMAO
These molecules are stabilizing osmolytes. They are present in living organism and
their function is to protect them from extreme conditions such as dehydration and/or freezing.
In particular, trehalose is a non-reducing disaccharide composed of two D-glucopyranose
units in α,α(11) linkage (see Fig. (8)). It is naturally abundant, especially in the hemolymph
of insects and into the cells of other organisms, which are subjected to the anhydrobiosis
phenomenon. Trehalose under this condition plays a role in promoting a protecting action on
the cellular membrane and on proteins. The mechanism of this stabilization effect is yet to be
clarified. Three hypotheses have been suggested based on experimental data available [105]:
1) water-replacement hypothesis: trehalose substitute the hydration water molecules and
stabilize the protein with hydrogen bonds; 2) water-layer: the sugar forms a protecting layer
that prevent the dehydration of the water hydration layer and 3) mechanical-entrapment
hypothesis: trehalose forms a glass surrounding the protein and trapping it in a stable
conformational state. Several molecular dynamics simulations studies have been reported in
the literature on different proteins to prove the three different hypotheses. MD Simulations of
carboxy-myoglobin in glassy state (89% w/w trehalose/water) and lysozyme, in moderately
concentrated solution, (see Table II) have evidenced on short simulation time scales (up to 2.5
ns) a reduced mobility of the enzymes supporting the third hypothesis.
Polyols (as glycerol, sorbitol) are also very important kosmotropic cosolvents. In
particular, glycerol is frequently used as cryoprotectant. Also for these osmolytes, several MD
simulations of different peptides and proteins have been performed (see Table II and III). As
an example, for cytochrome c in glycerol/water mixtures, the results of these studies evidence
the presence of a damping effect on molecular fluctuations induced by the glycerol [106]. In
addition, a reduction of the protein compressibility in good agreement with experimental data
was observed. This effect has been ascribed to the decoupling of the protein motion from the
solvent motion induced by the presence of glycerol. In fact, it suppresses the exchange
between water molecules bind to the protein surface and those in the hydration shell. In this
way, volume fluctuations of the protein core become disconnected from those of the solvation
shell [106]. A probable reason for this effect is the increased viscosity of the solvent in
solvation shell due to the presence of glycerol. This hypothesis is consistent with experimental
and theoretical evidences of the viscosity effect on protein and peptide dynamics [91, 96, 221-
225]. For a recent account on the viscosity effect on protein folding see also the review by
Hagen [107].
TMAO is another common natural osmolyte or chemical chaperone. It accumulates in the
intracellular environment at relatively high concentration, in response to osmotic or hydric
stress produced by the exposure to high salinity level, hydrostatic pressure, and anhydrobiosis
processes. MD simulations have evidenced the direct interaction between TMAO and protein
as possible mechanism of stabilization [108]. In this study, the protein γ-chymotrypsin inhibitor
was simulated in a solution of TMAO 4 M, urea 8 M and water. In the presence of TMAO, the
protein reduced considerably its mobility and retains its initial fold compared to a 8M
urea/water simulation [108]. The simulations evidenced a tendency of TMAO to reinforce
water-water and water-urea hydrogen bond interactions. In this way, urea does not interact
with the protein surface and, therefore, does not favour the exposure of hydrophobic internal
residues that can initiate the protein unfolding. A similar trend was observed in a MD study on
the prion protein [109].
Betaines and in particular the glycine betaine (N,N,N-trimethylglycine) are also
common natural osmolytes that help cells to survive osmotic stresses. So far few MD studies
have been reported on the effect of this osmolytes on peptides and proteins [110].
Finally, the behaviour of these osmolytes has been also generalized in the framework
of a statistical mechanic model that considers only the interaction of osmolytes with the protein
backbone. The model captures many common features of different osmolytes [111],
reinforcing the relevance of the direct effects in the osmolyte stabilization mechanism.
Ionic liquids
Biocatalysis in conventional organic solvents often suffers from reduced activity,
selectivity or stability of the enzyme [95]. Furthermore, organic solvents are normally toxic and
not environmentally friendly. For these reasons, the search for greener solvents is rapidly
becoming of utterly importance in industrial biocatalysis. Since the last few years, the interest
in ionic liquid as green solvent for biotechnological applications is growing exponentially [112].
This novel class of solvents composes of an organic cation (mostly an alkyl-substituted
imidazolium or pyridinium cation) and an anion, which can be organic and inorganic. They are
not flammable, thermally stable and do not have a measurable vapour pressure, which makes
them interesting from safety and environmental standpoints. Moreover, their physico-chemical
properties (for example polarity, hydrophobicity or miscibility) can be tailored to specific
applications. These capabilities are particularly attractive for protein bioengineering to
overcome environmental and safety hurdles associated with the use of conventional organic
solvents/water solutions. Therefore, it is not surprising rapid raising of the study and
applications in enzymatic catalysis [112].
The modelling studies of pure ionic liquids and their water solution are also rapidly
increasing [232-238], though MD studies of protein in ionic liquids are still quite limited [239-
242]. This emerging field is undoubtedly an interesting area of theoretical investigation.
However, the peculiar properties of ionic liquids will probably require more accurate treatment
of the electrostatic interactions in the simulations.
Other solvents
Hexane, DMF, ACN, diisopropyl ether, 3-pentanone, toluene, chloroform and carbon
tetrachloride are typical solvents for bioengineering applications [5]. They are also used in MD
studies of peptides and proteins to understand effects of organic solvents on protein stability
and activity [60]. Several MD simulation studies of different enzymes in these solvents are
reported in literature (see Table III). The general outcome of these simulations has shown that
they reduce protein flexibility (especially in the loop regions) with the retention of their initial
conformations. This result is in agreement with the hypothesis, based on experimental
observations that organic solvents tend to stiffen protein structure by reducing conformational
dynamics (see the first part of the Chapter). This effect is strongly related to the number of
water molecules nested in the enzyme structure [136, 137]. A comprehensive MD simulation
study on this effect for hexane, 3-pentanone, ethanol, diisopropyl ether, and ACN on cutinase
(a serine protease from Fusarium solani pisi) has been recently reported [113]. For each
solvent, different protein hydration levels ranging from 5% to 100 % (weight of water/weight of
protein) have been considered. The results of this study evidenced a solvent dependence of
the structural and dynamical properties of the protein during the simulations. The structure of
this enzyme maintains its native-like conformation until a certain extent of hydration beyond
that the protein structure start to unfold [136, 160]. This effect shows the importance of water
as lubricant for the protein function. As the flexibility of the protein is recovered, cosolvent
molecules promote stabilizing interactions with hydrophobic regions of the protein that
otherwise would not be exposed in pure water and, thereby, shifting the conformational
equilibrium toward the unfolded state. In addition, the polarity of the cosolvent influences
clustering properties of water molecules. In fact, non-polar solvents favour water segregation
on the protein surface with formation of large clusters. On the contrary, polar solvents promote
the formation of small water clusters loosely bound to the protein. Finally, water molecules
distribution in specific regions of the protein surface, seems to be unrelated to the type of
organic solvent used in the simulation [113].
Pure hexane, octane, decane, chloroform and carbon tetrachloride have also been
used to study the effect of hydrophobic environments on protein stability [134, 152, 243-247].
The interface with water of the same hydrophobic solvents has been used as a model for
membrane/water interfaces to study membrane permeation by peptides or hydrophobic
surface-induced peptide folding [248-252].
Finally, an interesting MD study of the Candida Antarctica Lipase B in supercritical
carbon dioxide has been very recently published [114]. This is a promising green-chemistry
solvent for many enzyme-catalyzed chemical reactions [115] but still little is known on the
molecular mechanism that trigger stability of some enzymes in such unconventional
At the end of this survey on computer simulation studies of proteins and peptides in
non-aqueous or in cosolvent/water mixtures, the questions posed in the Introduction should
have found a clear answer. In particular, we have introduced the MD simulations method and
evidenced its powerful capabilities to complement experimental techniques by providing
detailed information of biomolecular solvation phenomena at molecular model. Indeed, we
have also learnt that several aspects of non-aqueous enzymology related to the peptide and
protein dynamics and preferential solvation start to clarify with the support of MD simulation
studies. Still, many fascinating unsolved questions remain to be addressed and the study of
biomolecules in non-aqueous solvents using computational method is a growing field with of
fundamental (e.g. understanding the nature of protein stability and folding) and applied
research (protein bioengineering). Therefore, future endeavours (and this bring us to the
answer of the other question posed in the Introduction) are directed to deeper understanding
of cosolvent effects on molecular structures and to study new important solvent classes. In
this respect, for example, green solvents, as ionic liquids and polymers, are a promising area
of future theoretical investigations [10-12]. For this purpose, the constant increase of computer
performances is extending the time scale of MD simulations to get closer those of experiments
(microseconds to milliseconds) providing more accurate model of these phenomena. In
addition, more accurate force fields parameterized to better describe thermodynamics
properties of the solvents in different conditions are constantly under development [254, 255].
Some of these new force fields include improved treatment of electrostatic interactions by
directly accounting the atomic polarizability that will significantly improve the quality of both
solvent and protein models [256-258].
ACN = Acetonitrile.
AMBER = Assisted Model Building with Energy Refinement.
BAGUA = acyclic butylpentamethylguanidinium.
BMIM = 1-butyl-3-methylimidazolium.
CHARMM = Chemistry at HARvard Molecular Mechanics.
CI = γ-chymotrypsin inhibitor.
DCGUA = cyclic decyltrimethylguanidinium.
DMF = Dimethyl formamide.
DMSO = Dimethyl sulfoxide.
EMIM = Ethylmethyl imidazolium.
EtOH = Ethanol.
FTIR = Fast Transform InfraRed.
GROMOS = GROningen Molecular Simulation
HFIP = 1,1,1,3,3,3-hexafluoro-propan-2-ol.
hP450 = heme domain of cytochrome monooxygenase P450 BM-3
MD = Molecular dynamics.
MCGUA = cyclic tetramethylguanidinium.
MeOH = Methanol.
MOEMIM = 1-monoethoxy-3-methylimidazolium.
NMR = Nuclear magnetic resonance.
OPLS = Optimized Potentials for Liquid Simulations.
QM = Quantum mechanics.
SAXS = Small-Angle X-ray Scattering.
SANS = Small-Angle Neutron Scattering.
TFE = 2,2,2-Trifluoroethanol.
THF = Tetrahydrofuran.
TMAO = Trimethylamine-N-oxide.
[1] Vangunsteren, W. F. and Karplus, M.; (1981) Nature, 293, 677-678.
[2] Rees, E. D. and Singer, S. J.; (1956) Archives of Biochemistry and Biophysics, 63, 144-
[3] Berendsen, H. J. C.; (1998) Science, 282, 642-643.
[4] Klibanov, A. M.; (2001) Nature, 409, 241-246.
[5] Carrea, G. and Riva, S.; (2000) Angew. Chem. Int. Ed., 39, 2226-2254.
[6] Frizpatrick, P. A., Steinmetz, C. U., Ringe, D. and Klibanov, A. M.; (1993) Proc. Natl.
Acad. Sci. USA, 90, 8653-8657.
[7] Schmitze, J. L., Stern, L. J. and Klibanov, A. M.; (1998) Proc. Natl. Acad. Sci. USA,
95, 12918-12923.
[8] Roccatano, D.; (2008) Curr Protein Pept Sci, 9, 407-426.
[9] Leach, A. R.; (2001) Molecular Modelling. Principles and Applications (Prentice Hall).
[10] Berendsen, H. J. C.; (2007) Simulating the physical world : hierarchical modeling from
quantum mechanics to fluid dynamics (Cambridge ; New York: Cambridge University
[11] Dinur, U. and Hagler, A. T.; (1991) in, 2 ( VCH Publishers, New York).
[12] Villa, A., Fan, H., Wassenaar, T. and Mark, A. E.; (2007) J. Phys. Chem. B, 111, 6015-
[13] Soares, T. A., Daura, X., Oostenbrink, C., Smith, L. J. and van Gunsteren, W. F.;
(2004) J. Mol. NMR, 30, 407-422.
[14] Oostenbrink, C., Villa, A., Mark, A. E. and van Gunsteren, W. F.; (2004) J. Comp.
Chem., 25, 1656-1676.
[15] Freddolino, P. L., Arkhipov, A. S., Larson, S. B., McPherson, A. and Schulten, K.;
(2006) Structure, 14, 437-449.
[16] Duan, Y. and Kollman, P. A.; (1998) Science, 282, 740-744.
[17] Halle, B.; (2004) Phil. Trans. R. Soc. Lond. B, 359, 1207-1224.
[18] Samanta, S., Hezaveh, S., Milano, G. and Roccatano, D.; (2012) J Phys Chem B, 116,
[19] Okamoto, Y.; (2004) J. Mol. Graph. & Mod., 22, 425-439.
[20] Sotomayor, M. and Schulten, K.; (2007) Science, 316, 1144-1148.
[21] Adcock S. A. and McCammon, J. A.; (2006) Chem. Rev., 106, 1589-1615.
[22] Takeda, T. and Klimov, D. K.; (2009) Biophysical Journal, 96, 442-452.
[23] Jensen, M. O., Yin, Y., Tajkhorshid, E. and Schulten, K.; (2007) Bioph. J., 93, 92-102.
[24] Jarzynski, C.; (1997) Phys. Rev. Lett., 78, 2690-2693.
[25] Jarzynski, C.; (1997) Physical Review Letters, 78, 2690.
[26] Hub, J. S., de Groot, B. L. and van der Spoel, D.; (2010) Journal of Chemical Theory
and Computation, null-null.
[27] Kumar, S., Bouzida, D., Swendsen, R. H., Kollman, P. A. and Rosenberg, J. M.; (1992)
Journal of Computational Chemistry, 13, 1011-1021.
[28] Roccatano, D.; (2007) in Nanostructured Soft Matter: Experiment, Theory, Simulation
and Perspectives (Springer ).
[29] Finney, J. L. and Soper, A. K.; (1994) Chem. Soc. Rev., 1, 1-10.
[30] Allen, M. P. and Tildesley, D. J.; (1989) Computer Simulation of Liquids (Oxford
University Press).
[31] Makarov, V., Pettitt, B. M. and Feig, M.; (2002) Acc. Chem. Res., 376-384.
[32] Roccatano, D., Amadei, A., Di Nola, A. and Berendsen, H. J. C.; (1999) Protein Sci., 8,
[33] Mattos, C. and Ringe, D.; (1996) Nature Biotech., 14, 595-599.
[34] Liepinsh, E. and Otting, G.; (1997) Nature Biotech., 15, 264-268.
[35] English, A. C., Groom, C. R. and Hubbard, R. E.; (2001) Prot. Eng., 14, 47-59.
[36] Otting, G.; (1997) Prog. NMR Spectrosc., 31, 259-285.
[37] Serdyuk, I. N., Zaccai, N. R. and Zaccai, G.; (2007) Methods in molecular biophysics :
structure, dynamics, function (Cambridge ; New York: Cambridge University Press).
[38] Bizzarri, A. R. and Cannistraro, S.; (2002) J. Phys. Chem. B, 106, 6617-6633.
[39] Rocchi, C., Bizzarri, A. R. and Cannistraro, S.; (1998) Phys. Rev. E, 57, 3315-3325.
[40] Kovacs, H., Mark, A. E. and vanGunsteren, W. F.; (1997) Proteins: Struct. Funct. and
Genet., 27, 395-404.
[41] Dirama, T. E., Carri, G. A. and Sokolov, A. P.; (2005) J. Chem. Phys., 122, -.
[42] Makarov, V., Feig, M., Andrews, B. K. and Pettit, B. M.; (1998) Bioph. J., 75, 150-158.
[43] Baynes, B. M. and Trout, B. L.; (2003) J. Phys. Chem. B, 107, 14058-14067.
[44] Scharnagl, C., Reif, M. and Friedrich, J.; (2005) BBA-Proteins Proteom., 1749, 187-
[45] Roccatano, D., Fioroni, M., Zacharias, M. and Colombo, G.; (2005) Protein Sci., 14,
[46] Snow, C. D., Sorin, E. J., Rhee, Y. M. and Pande, V. S.; (2005) Ann. Rev. Biophys.
Biomol. Struct., 34, 43-69.
[47] Newman, K. E.; (1994) Chem. Soc. Rev., 23, 31-40.
[48] Shimizu, S. and Matubayasi, N.; (2006) Chem. Phys. Lett., 420, 518-522.
[49] Chitra, R. and Smith, P. E.; (2001) J. Chem. Phys. B, 105, 11513-11522.
[50] Aburi, M. and Smith, P. E.; (2004) J. Phys. Chem. B, 108, 7382-7388.
[51] Imai, T., Kovalenko, A. and Hirata, F.; (2006) Mol. Simulat., 32, 817-824.
[52] Yamazaki, T., Imai, T., Hirata, F. and Kovalenko, A.; (2007) J. Phys. Chem. B, 111,
[53] Wyttenbach, T. and Bowers, M. T.; (2009) Chemical Physics Letters, 480, 1-16.
[54] Branco, R. J. F., Graber, M., Denis, V. and Pleiss, J.; (2009) Chembiochem, 10, 2913-
[55] Halling, P. J.; (2004) Phil. Trans. R. Soc. Lond. B, 359, 1287-1297.
[56] Courtenay, E. S., Capp, M. W., Anderson, C. F. and Record, M. T.; (2000)
Biochemistry, 39, 4455-4471.
[57] King, J. T., Arthur, E. J., Brooks, C. L., III and Kubarych, K. J.; (2012) Journal of
Physical Chemistry B, 116, 5604-5611.
[58] Rezus, Y. L. A. and Bakker, H. J.; (2006) Proc. Natl. Acad. Sci. USA, 103, 18417-
[59] Roccatano, D., Colombo, G., Fioroni, M. and Mark, A. E.; (2002) Proc. Natl. Acad. Sci
USA, 99, 12179-12184.
[60] Colombo, G. and Carrea, G.; (2002) J. Biotechnol., 96, 35-42.
[61] Daggett, V.; (2006) J. Mol. Biol., 106, 1898-1916.
[62] Chiti, F. and Dobson, C. M.; (2006) Annu. Rev. Biochem., 75, 333-366.
[63] Chiti, F., Webster, P., Taddei, N., Clark, A., Stefani, M., Ramponi, G. and Dobson, C.
M.; (1999) Proc. Natl. Acad. Sci. USA, 96, 3590-3594.
[64] De Mori, G. M. S., Meli, M., Monticelli, L. and Colombo, G.; (2005) MINI-Rev. Med..
Chem., 5, 353-359.
[65] Daidone, I., Simona, F., Roccatano, D., Broglia, R. A., Tiana, G., Colombo, G. and Di
Nola, A.; (2004) Proteins: Struct. Funct. and Bioinformatics, 57, 198.
[66] Ma, B. Y. and Nussinov, R.; (2006) Curr. Opin. Chem. Biol., 10, 445-452.
[67] Griebenow, K. and Klibanov, A. M.; (1996) J. Am. Chem. Soc., 118, 11695-11700.
[68] Pieraccini, S., Sironi, M. and Colombo, G.; (2006) Chem. Phys. Lett., 418, 373-376.
[69] Zhu, L. J., Yang, W., Meng, Y. Y., Xiao, X. C., Guo, Y. Z., Pu, X. M. and Li, M. L.;
(2012) Journal of Physical Chemistry B, 116, 3292-3304.
[70] Yang, L., Dordick, J. S. and Garde, S.; (2004) Biophys. J., 87, 812-821.
[71] Micaelo, N. M., Teixeira, V. H., Baptista, A. M. and Soares, C. M.; (2005) Bioph. J.,
89, 999-1008.
[72] Soares, C. M., Teixeira, V. H. and Baptista, A. M.; (2003) Bioph. J., 84, 1628-1641.
[73] van Gunsteren, W. F., Bakowies, D., Baron, R., Chandrasekhar, I., Christen, M., Daura,
X., Gee, P., Geerke, D. P., Glättli, A., Hünenberger, P. H., Kastenholz, M. A.,
Oostenbrink, C., Schenk, M., Trzesniak, M., van der Vegt, N. F. A. and Yu, H. B.;
(2006) Angew. Chem. Int. Ed., 45, 4064-4092.
[74] Liu, H. L. and Hsu, C.-M.; (2003) Chem. Phys. Lett., 375, 119-125.
[75] Gerig, J. T.; (2012) Journal of Physical Chemistry B, 116, 1965-1973.
[76] Hwang, S., Shao, Q., Williams, H., Hilty, C. and Gao, Y. Q.; (2011) Journal of
Physical Chemistry B, 115, 6653-6660.
[77] Fornili, S. L., Pizzi, R. and Rebeccani, D.; (2010) International Journal of Peptide
Research and Therapeutics, 16, 223-231.
[78] Paschek, D., Puehse, M., Perez-Goicochea, A., Gnanakaran, S., Garcia, A. E., Winter,
R. and Geiger, A.; (2008) Chemphyschem, 9, 2742-2750.
[79] Daura, X.; (2006) Theor. Chem. Acc., 116, 297-306.
[80] Gattin, Z. and van Gunsteren, W. F.; (2009) Journal of Physical Chemistry B, 113,
[81] Fioroni, M., Diaz, M. D., Burger, K. and Berger, S.; (2002) J. Am. Chem. Soc., 124,
[82] Buck, M.; (1998) Q. Rev. Biophys., 31, 297-355.
[83] Hong, D.-P., Hoshino, M., Kuboi, R. and Goto, Y.; (1999) J. Am. Chem Soc., 121,
[84] Wood, S. J., Maleeff, B., Hart, T. and Wetzel, R.; (1996) J. Mol. Biol., 256, 870-877.
[85] Liu, Y. and Bolen, D. W.; (1995) Biochemistry, 34, 12884-12891.
[86] Jasanoff, A. and Fersht, A.; (1994) Biochemistry, 33, 2129-2135.
[87] Luo, P. and Baldwin, R. L.; (1997) Biochemistry, 36, 8413-8421.
[88] Reiersen, H. and Rees, A. R.; (2000) Protein Eng., 13, 739-743.
[89] Fioroni, M., Burger, K., Mark, A. E. and Roccatano, D.; (2000) J. Phys. Chem. B, 104,
[90] Fioroni, M., Burger, K., Mark, A. E. and Roccatano, D.; (2001) J. Phys. Chem. B, 105,
[91] Yoshida, K., Yamaguchi, T., Adachi, T., Otomo, T., Matsuo, D., Takamuku, T. and
Nishi, N.; (2003) J. Chem. Phys., 119, 6132-6142.
[92] Hirota, N., Mizuno, K. and Goto, Y.; (1998) J. Mol. Biol., 275, 365-368.
[93] Hirota, N., Mizuno, K. and Goto, Y.; (1997) Protein Sci., 6, 416-421.
[94] Gerig, J. T.; (2004) Biophys. J., 86, 3166-3175.
[95] Wong, T., Arnold, F. and Schwaneberg, U.; (2004) Biotechnol. and Bioeng., 85, 351-
[96] Whitehouse, C. J., Bell, S. G. and Wong, L. L.; (2012) Chem Soc Rev, 41, 1218-1260.
[97] Winn, P. J., Luedemann, S. K., Gauges, R., Lounnas, V. and Wade, R. C.; (2002) Proc.
Natl. Acad. Sci. USA, 99, 5361-5366.
[98] Luedemann, S. K., Lounnas, V. and Wade, R. C.; (2000) J. Mol. Biol., 303, 797-811.
[99] Roccatano, D., Wong, T. S., Schwaneberg, U. and Zacharias, M.; (2005) Biopolymers,
78, 259-267.
[100] Roccatano, D., Wong, T. S., Schwaneberg, U. and Zacharias, M.; (2006) Biopolymers,
83, 467-476.
[101] Kuper, J., Wong, T. S., Roccatano, D., Wilmanns, M. and Schwaneberg, U.; (2007) J.
Am. Chem. Soc., 129, 5786-5787.
[102] Kuper, K., Tee, K. L., Wilmanns, M., Roccatano, D., Schwaneberg, U. and Wong, T.
S.; (2012) Acta Crystallographica Section F, in press.
[103] Mason, P. E., Neilson, G. W., Dempsey, C. E., Barnes, A. C. and Cruickshank, J. M.;
(2003) Proc Natl Acad Sci U S A, 100, 4557-4561.
[104] Mason, P. E., Neilson, G. W., Enderby, J. E., Saboungi, M. L., Dempsey, C. E.,
MacKerell, A. D., Jr. and Brady, J. W.; (2004) J Am Chem Soc, 126, 11462-11470.
[105] Lins, R. D., Pereira, C. S. and Hünenberger, H.; (2004) Proteins: Struct. Funct. and
Bioinformatics, 55, 177-186.
[106] Scharnagl, C., Reif, M. and Friedrich, J.; (2005) Bioph. J., 89, 64-75.
[107] Hagen, S. J.; (2010) Current Protein & Peptide Science, 11, 385-395.
[108] Bennion, B. J. and Daggett, V.; (2004) Proc. Natl. Acad. Sci. USA, 101, 6433.
[109] Bennion, B. J., DeMarco, M. L. and Daggett, V.; (2004) Biochemistry, 43, 12955-
[110] Ma, L., Pegram, L., Record, M. T. and Cui, Q.; (2010) Biochemistry, 49, 1954-1962.
[111] Street, T. O., Bolen, D. W. and Rose, G. D.; (2006) Proc. Natl. Acad. Sci. USA, 103,
[112] van Rantwijk, F., Madeira Lau, R. and Sheldon, R. A.; (2003) Trends in Biotech., 21,
[113] Micaelo, N. M. and Soares, C. M.; (2007) FEBS Lett., 274, 2424-2436.
[114] Silveira, R. L., Martinez, J., Skaf, M. S. and Martinez, L.; (2012) Journal of Physical
Chemistry B, 116, 5671-5678.
[115] Lozano, P.; (2010) Green Chemistry, 12, 555-569.
[116] Roccatano, D., Sbardella, G., Aschi, M., Amicosante, G., Bossa, C., Di Nola, A. and
Mazza, F.; (2005) Journal of Computer-Aided Molecular Design, 19, 329-340.
[117] Glattli, A., Chandrasekhar, I. and van Gunsteren, W. F.; (2006) Eur. Biophys. J.
Biophy., 35, 255-267.
[118] Mason, P. E., Brady, J. W., Neilson, G. W. and Dempsey, C. E.; (2007) Biophysical
Journal, 93, L4-L6.
[119] Liu, F. F., Dong, X. Y. and Sun, Y.; (2008) Journal of Molecular Graphics &
Modelling, 27, 421-429.
[120] Dauber-Osguthorpe, P., Roberts, V. A., Osguthorpe, D. J., Wolff, J., Genest, M. and
Hagler, A. T.; (1988) Proteins: Struct. Funct. and Genet., 4, 31-47.
[121] Tieleman, D. P., Sansom, M. S. P. and Berendsen, H. J. C.; (1999) Bioph. J., 76, 40-49.
[122] Sessions, R. B., Gibbs, N. and Dempsey, C. E.; (1998) Biophys. J., 74, 138-152.
[123] Daura, X., Gademann, K., Jaun, B., Seebach, D., van Gunsteren, W. F. and Mark, A.
E.; (1999) Angew. Chem. Int. Ed., 38, 236-240.
[124] van der Spoel, D. and Berendsen, H. J. C.; (1997) Bioph. J., 72, 2032-2041.
[125] Niggli, D. A., Ebert, M.-O., Lin, Z., Seebach, D. and van Gunsteren, W. F.; (2012)
Chemistry-a European Journal, 18, 586-593.
[126] Wang, D., Jaun, B. and van Gunsteren, W. F.; (2009) Chembiochem, 10, 2032-2041.
[127] Dwyer, D. S.; (1999) Biopolymers, 49, 635-645.
[128] De Loof, H., Nilsson, L. and Rigler, R.; (1992) J. Am. Chem. Soc., 114, 4028-4035.
[129] Bodkin, M. J. and Goodfellow, J. M.; (1996) Biopolymers, 39, 43-50.
[130] Diaz, M. D., Fioroni, M., Burger, K. and Berger, S.; (2002) Chem. Eur. J., 8, 1663-
[131] Johnston, J. M., Cook, G. A., Tomich, J. M. and Sansom, M. S. P.; (2006) Bioph. J., 90,
[132] van Buuren, A. R. and Berendsen, H. J. C.; (1993) Biopolymers, 33, 1159-1166.
[133] Villa, A., Mark, A. E., Saracino, G. A. A., Cosentino, U., Pitea, D., Moro, G. and
Salmona, M.; (2006) J. Phys. Chem. B, 110, 1423-1428.
[134] Wei, G. and Shea, J.-E.; (2006) Biophys J. , 1638-1647.
[135] Tobi, D., Elber, R. and Thirumalai, D.; (2003) Biopolymers, 68, 359-369.
[136] Caballero-Herrera, A., Nordstrand, K., Berndt, K. D. and Nilsson, L.; (2005) Bioph. J.,
89, 842-857.
[137] Klimov, D. K., Straub, J. E. and Thirumalai, D.; (2004) Proc. Natl. Acad. Sci. USA,
101, 14760-14765.
[138] Liu, F. F., Ji, L., Dong, X. Y. and Sun, Y.; (2009) Journal of Physical Chemistry B,
113, 11320-11329.
[139] Zhang, Z., Zhu, Y. and Shi, Y.; (2001) Biophys. Chem., 89, 145-162.
[140] Knecht, V., Mohwald, H. and Lipowsky, R.; (2007) Journal of Physical Chemistry B,
111, 4161-4170.
[141] Dos Santos Cabrera, M. P., Arcisio-Mirandaa, M., Broggio Costa, S. T., Konno, K.,
Ruggiero, J. R., Procopio, J. and Neto, J. R.; (2008) Journal of Peptide Science, 14,
[142] Duarte, A. M. S., van Mierlo, C. P. M. and Hemminga, M. A.; (2008) Journal of
Physical Chemistry B, 112, 8664-8671.
[143] Yang, C., Li, J., Li, Y. and Zhu, X.; (2009) Journal of Molecular Structure-Theochem,
895, 1-8.
[144] Mehrnejad, F., Khadem-Maaref, M., Ghahremanpour, M. M. and Doustdar, F.; (2010)
Journal of Computer-Aided Molecular Design, 24, 829-841.
[145] Chen, R. and Mark, A. E.; (2011) European Biophysics Journal with Biophysics
Letters, 40, 545-553.
[146] Mehrnejad, F., Ghahremanpour, M. M., Khadem-Maaref, M. and Doustdar, F.; (2011)
Journal of Chemical Physics, 134.
[147] Shao, Q., Fan, Y., Yang, L. and Gao, Y. Q.; (2012) Journal of Chemical Physics, 136.
[148] Hatfield, M. P. D., Murphy, R. F. and Lovas, S.; (2010) Journal of Physical Chemistry
B, 114, 3028-3037.
[149] Wei, H. Y., Shao, Q. A. and Gao, Y. Q.; (2010) Physical Chemistry Chemical Physics,
12, 9292-9299.
[150] Tran, H. T., Mao, A. and Pappu, R. V.; (2008) Journal of the American Chemical
Society, 130, 7380-7392.
[151] Levitt, M., Hirshberg, M., Sharon, R. and Daggett, V.; (1995) Comp. Phys. Comm., 91,
[152] Alonso, D. O. V. and Daggett, V.; (1995) J. Mol. Biol., 247, 501.
[153] Lousa, D., Baptista, A. M. and Soares, C. M.; (2012) Journal of Chemical Information
and Modeling, 52, 465-473.
[154] Micaelo, N. M. and Soares, C. M.; (2008) Journal of Physical Chemistry B, 112, 2566-
[155] Calflish, A. and Karplus, M.; (1999) Structure, 7, 477.
[156] Tirado-Rives, J., Orozco, M. and Jorgensen, W. L.; (1997) Biochemistry, 36, 7313-
[157] Toba, S., Hartsough, D. S. and Merz, K. M.; (1996) J. Am. Chem. Soc., 118, 6490-
[158] Bennion, B. J. and Daggett, V.; (2003) Proc. Natl Acad. Sci. USA, 100, 5142-5147.
[159] Lindgren, M. and Westlund, P. O.; (2010) Biophysical Chemistry, 151, 46-53.
[160] Zhang, N., Liu, F. F., Dong, X. Y. and Sun, Y.; (2012) Journal of Physical Chemistry
B, 116, 7040-7047.
[161] Colombo, G., Toba, S. and Merz, K. M.; (1999) J. Am. Chem. Soc., 121, 3486-3493.
[162] Zheng, Y. J. and Ornstein, R. L.; (1996) J. Am. Chem. Soc., 118, 4175-4180.
[163] Lousa, D., Baptista, A. M. and Soares, C. M.; (2011) Protein Science, 20, 379-386.
[164] Lousa, D., Cianci, M., Helliwell, J. R., Halling, P. J., Baptista, A. M. and Soares, C. M.;
(2012) Journal of Physical Chemistry B, 116, 5838-5848.
[165] Flöck., D., Daidone, I. and Di Nola, A.; (2004) Biopolymers, 75, 491-496.
[166] Curtis, J. E., Dirama, T. E., Carri, G. A. and Tobias, D. J.; (2006) J. Phys. Chem. B,
110, 22953-22956.
[167] Vagenende, V., Yap, M. G. S. and Trout, B. L.; (2009) Journal of Physical Chemistry
B, 113, 11743-11753.
[168] Rezaei-Ghaleh, N., Amininasab, M. and Nemat-Gorgani, M.; (2008) Biophysical
Journal, 95, 4139-4147.
[169] Shukla, D., Schneider, C. P. and Trout, B. L.; (2011) Journal of the American Chemical
Society, 133, 18713-18718.
[170] Trodler, P. and Pleiss, J.; (2008) Bmc Structural Biology, 8.
[171] Klaehn, M., Lim, G. S. and Wu, P.; (2011) Physical Chemistry Chemical Physics, 13,
[172] Trodler, P., Schmid, R. D. and Pleiss, J.; (2009) Bmc Structural Biology, 9.
[173] Fogolari, F., Corazza, A., Varini, N., Rotter, M., Gumral, D., Codutti, L., Rennella, E.,
Viglino, P., Bellotti, V. and Esposito, G.; (2011) Proteins-Structure Function and
Bioinformatics, 79, 986-1001.
[174] Wei, H. Y., Yang, L. J. and Gao, Y. Q.; (2010) Journal of Physical Chemistry B, 114,
[175] Haberler, M., Schroeder, C. and Steinhauser, O.; (2011) Physical Chemistry Chemical
Physics, 13, 6924-6938.
[176] Diaz-Vergara, N. and Pineiro, A.; (2008) Journal of Physical Chemistry B, 112, 3529-
[177] Fraga, A. S., Esteves, A. C., Micaelo, N., Cruz, P. F., Brito, R. M. M., Nutley, M.,
Cooper, A., Barros, M. M. T. and Pires, E. M. V.; (2012) International Journal of
Biological Macromolecules, 50, 323-330.
[178] Stumpe, M. C. and Grubmuller, H.; (2009) Biophysical Journal, 96, 3744-3752.
Figure Captions
Fig. (1). Description of the energetic terms present in the force field for biomolecular
simulations. The molecules in the picture are: at the centre, the Ace-Ala-NH2; on the right,
a TFE molecule and on the top left side, a water molecule. Only polar hydrogen atoms are
Fig. (2). Summary of the main sources of experimental data of solvent effects that can be
used to compare with the results of MD simulations of biomolecular systems.
Fig. (3). Correspondence of radial distribution function of a TFE/water mixture with the
physical distribution of molecules in solution. Rings in yellow and cyan correspond to
section views of two spherical sectors centred on the blue atom.
Fig. (4). Example of representation of the volumetric data of water density surrounding a
β-hairpin forming peptide in solution. On each slice of the volumetric data, the local density
in the form of three-dimensional elevation is shown. In the centre of the figure, the
average structure of the peptide is reported. The analysis was performed on a 5 ns long
MD simulation [32].
Fig. (5). Calculation of the local concentration of cosolvent around a peptide. On the left:
averaged local concentration of HFIP (LHC, %v/v) around each residue of the peptide
Melittin [45]. On the right: density distribution of HFIP molecules around the Melittin
calculated from the last 5 ns of the simulations in Ref. [45].
Fig. (6). Classification of solvent according to their effect on the protein structure.
Fig. (7). Summary of the effects of non-natural solvent environment on biomolecules. The
different effects are intertwined.
Fig. (8). Structures of some of the solvent molecules commonly used in MD simulations of
biomolecular systems. A) hexane; B) carbon tetrachloride; C) chloroform; D) acetonitrile; E)
2,2,2-trifluoroethanol; F) 1,1,1,3,3,3-hexafluoro-propan-2-ol; G) ethanol; H) methanol; I)
dimethyl sulfoxide; J) urea; K) trimethylamine-N-oxide; L) trehalose; M) glycerol; N) dimethyl
Fig. (9). Snapshots of conformations sampled every 100 ps from a simulation of a 30% (v/v)
TFE/water mixture box. The water molecules are shown in cyan and TFE in red colors.
Fig. (10). Top views of P450 BM-3 heme domain conformation taken from a
DMSO/water MD simulation [116]. Left view: secondary structure representation. Right
view: DMSO surface accessible area (probe radius 0.24 nm) representation. In both
cases, the protein structure is coloured according to the local DMSO concentration
(LDC, calculated at 0.5 nm from the Ca atom of each residue). The lowest and highest
DMSO concentration correspond to regions represented in red and blue colours,
respectively [116].
Fig. (11). Interactions of DMSO with the active site of hP450. (A) Active sites of the crystal
structure (pdb-code: 2J1M) crystallized in 28 % v/v DMSO/water mixture [101]. A DMSO
molecule is coordinating to the heme iron. Anomalous scattering data have shown that the
coordination occurs via the sulphur atom [101]. (B) Snapshot from the simulation of the F87A
mutant of hP450 in DMSO/water mixture [100]. The figure shows two out of three DMSO
molecules that are present in the active site. The DMSO molecules approach the iron and
displace the coordinating water molecule from its position.
Table I: Summary of the most common solvents used to study different properties of proteins
and peptides using MD simulations.
Carbon tetrachloride
Ionic liquids
TABLE II: Selection of peptides studied in different organic solvent and mixtures. Time indicates the
longest MD simulation reported in the paper.
Peptide sequence
30 %HFIP/water
GmdCl 3 M
onium Sulphate
0.1 M,
GmdCl 3 M
β-hairpin from
Protein G B1
β-hairpin from
Protein G B1
Trehalose 0-
0.26 M
CVFF [120]
peptide II
1: β3-
2: tethered
Mixed α/β
1: (β2-(S)-F)H(β2-(S)-L)(β3-
2: Aib(β3-(S)-Y)Aib(β3-(S)-
Fluorinated β-
1: H-(β-hVal)(β-hAla)(β-
2: H(β-hVal)(β-hAla)(β-
3: H(β-hVal)(β-hAla)(β-hLeu)-
50% TFE/water
30% TFE/water
forming peptide
30% TFE/water
30% TFE/water
Prion H1 peptide
30% TFE/water
H1 helix from
the mouse prion
Urea 3.94M
GdmCl 3.05 M
80% HFIP/water
640 a)
Blocked valine
Urea 8M
of C-peptide
Urea 8M
Urea 8M
Aβ16-22 and Aβ40
Trehalose (0-
0.18 mol/L)
Urea 8M
Binding B18
30 % TFE/water
25% methanol/
Angiotensin II
BB5 Peptide
50% methanol/
Peptide hlF1-11
Pure methanol,
4:4:1 methanol
AK peptide
30% TFE/water
Peptide sMTM7
Aβ(1-42) peptide
Magainin 2
Various water
mixtures of
Urea, GmdCl,
aurein 1.2
50 % TFE/water
maculatin 1.1
50 % TFE/water
citropin 1.1
50 % TFE/water
caerin 1.1
50 % TFE/water
Magainin 2
5 M)
BB5 Peptide
Various 30-40%
(v/v) water
Mixture of TFE,
MeOH, Glycerol
and DMSO
GmdCl 3,6 M
Urea 4,8 M
Urea 1.9-5.8 M
32500 a)
Amber (ff94)
GmdCl 2M
Urea 2 M
Amber (ff99)
Urea 2M
Glycine betaine
Peptide 1
Urea 5 M
Urea 5M
Polyglycine and
G15 and (GS)8
Urea 8M
a) Replica exchange MD simulation study, in this case the cumulative time is reported.
TABLE III: Selection of proteins studied in different organic solvent and mixtures. Time indicates the
longest MD simulation reported in the paper.
Force field
[136, 137]
60% MeOH/water
ENCAD [151]
Pure diisopropyl
ether and water
Pure 3-pentanone
and water mixtures
Pure EtOH and
water mixtures
25 % water/EtOH
25 % water/EtOH
Pure ACN and
water mixtures
Pure Hexane and
water mixtures
[136, 137, 160]
Pure [BMIM][PF6]
and [BMIM][NO3]
Urea 8 M
[155, 156]
Urea 8 M
Urea 4 M
8 M Urea
10 M Urea
8 M Urea;
8M Urea/1 M
P450 BM3
14% (v/v)
[54, 55]
cytochrome c
60% (v/v) Glycerol/
Subtilisin BPN’
Subtilisin BPN’
Subtilisin BPN’
[135, 286]
Subtilisin Carlsberg
Subtilysin Carlsberg
Subtilysin Carlsberg
Subtilysin Carlsberg
25% (v/v)
water glass
[223, 290]
Glycerol glass
5.87 M
30 % TFE/water
α-Chymotrypsin A
Candida Antartica
Lipase B
Candida Antartica
Lipase B
Hexane, tert-butyl
ether, methanol,
tert- butyl alcohol
Candida Antartica
Lipase B
Candida Antartica
Lipase B
Burkholderia cepacia
β-2 microglobulin
26% TFE/water
Urea 8M
Protein L
Urea 10 M
Villin headpiece
protein HP-35 and
A doubly norleucine-
substituent mutant
Urea 5 M
Human zinger Finger
r mixtures
2.36-4.41 M
isomerase from
Trypanosoma cruzi
Cardosin A
10, 90 % (v/v)
Cold Shock protein
Bc-CsP from Bacillus
Urea 8M