Вы находитесь на странице: 1из 17

A molecular approach to bioseparations: Proteinprotein and

proteinsalt interactions
R.A. Curtis , L. Lue
School of Chemical Engineering and Analytical Science, The University of Manchester, PO Box 88, Sackville Street, Manchester M60 1QD UK

Abstract
Many bioprocess separations involve manipulating the solution conditions to selectively remove or concentrate a target protein. The
selectivity is determined by the thermodynamics of the protein solution, which are governed, at the molecular level, by proteinprotein
interactions. Thus, to optimise these processes, one must understand the precise nature of these interactions, how they are affected by
the control parameters (e.g., temperature, ionic strength, pH), and how they relate to the solution properties of interest (e.g., protein
solubility, protein stability, crystal nucleation rates). Recently, studies of proteinprotein interactions have been motivated by the discovery
of a crystallisation window, which places bounds on the proteinprotein attraction that is required for crystallisation to be possible. This
review focuses on experimental and theoretical studies of proteinprotein interactions and discusses the current gaps in understanding these
forces. The rst models for these interactions had been based on DLVO theory, which has proven useful for understanding proteinprotein
interactions in dilute electrolyte solutions. However, DLVO theory does not include such important effects as anisotropic interactions,
solvation forces, and specic ion effects. Various approaches have been developed to include these phenomena although they are still not
well understood. One outstanding issue concerns specic ion effects, which play a crucial role in biological systems. An initial step in
understanding these effects might be to rst rationalize proteinsalt interactions.
2005 Elsevier Ltd. All rights reserved.
Keywords: Proteins; Phase equilibria; Crystallisation; Colloidal phenomena; Intermolecular interactions; Molecular thermodynamics

1. Introduction
At the molecular level, proteinprotein and proteinsalt
interactions control the structure and thermodynamics of
protein solutions. Consequently, knowledge of these interactions, how they are affected by the solution conditions,
and how they impact the bulk properties of the solution,
can lead to key insights in designing and optimising many
bioprocesses, including protein salting-out (Melander and
Horvath, 1977; Coen et al., 1995), protein partitioning in
aqueous two-phase polymer-based systems (Haynes et al.,
1991, 1993), protein chromatography (Melander and
Horvath, 1977; Vailaya and Horvath, 1998; Oberholzer and
Lenhoff, 1999), protein formulation (Chi et al., 2003) and
Corresponding author. Tel.: +441613064401; fax: +441613064399.

E-mail address: r.curtis@manchester.ac.uk (R.A. Curtis).

the re-folding of proteins from inclusion bodies (Ho


et al., 2003). A wealth of experimental and theoretical
studies have been directed at understanding these interactions. Proteinprotein interactions have been probed using a
wide variety of experimental techniques, such as static and
dynamic light scattering, osmometry, and small angle neutron scattering. The resulting experimental data can be used
as inputs in molecular thermodynamic models to predict
protein phase equilibria, protein crystallisation, and protein aggregation (Vilker et al., 1981; Haynes et al., 1992;
Taratuta et al., 1990). Proteinsalt interactions, which are
accessible through dialysis equilibrium experiments via the
preferential interaction parameter (Lee et al., 1979), can also
be used to rationalise the behaviour of protein solutions.
The rst theoretical descriptions of protein salting-out were
based on modelling the effect of salt on the solvation free
energy of the protein (Melander and Horvath, 1977). These

theories were also used to calculate the retention time of


proteins in hydrophobic interaction chromatography. More
importantly, the inuence of salt on protein stability can
be understood in terms of differences in the interactions
between the salt and the protein in the folded and unfolded
states (Timasheff, 1991).
Recently, one of the main motivations of understanding
proteinprotein interactions is based on the discovery of a
crystallisation window for protein solutions. Only when the
attractions between proteins fall within the crystallisation
window can high quality crystals be obtained on reasonable
time scales. Much of the work in this area has focused on
the connection between proteinprotein interactions and the
phase diagram and the relationship between protein crystallisation rates and the location on the protein phase diagram.
The majority of proteinprotein interaction models are derived from colloid science, where proteins are modelled as
charged, dielectric spheres. These models, based on DLVO
theory (Verwey and Overbeek, 1948), provide an adequate
description for proteins in dilute aqueous electrolyte solutions, where the interactions are dominated by long-ranged
electrostatic forces. Despite the general success of DLVO
theory, there are many situations where it is unable to provide even a proper qualitative description of protein solution behaviour. One important example is protein saltingout, which is the precipitation of protein with the addition
of salt. The ability of an ion to induce protein precipitation
is strongly dependent on its type. The salting-out effectiveness is correlated with the ions position in the Hofmeister
series (Hofmeister, 1888). DLVO theory is unable to quantitatively predict when salting-out will occur and, in addition,
does not predict any dependence on salt type.
The remainder of this review is organised as follows. First,
we discuss the use of osmometry as a tool to probe the
general behaviour of protein solutions. Then in Section 1.2,
the concept of the potential of mean force is introduced,
which provides a precise denition of proteinprotein interactions. The relationship between the pmf and the structure and thermodynamics of the protein solution is given by
McMillanMayer theory, which is described in Section 1.3.
Section 2 provides a discussion of the general phase behaviour of aqueous protein solutions, its dependence on
the form of the proteinprotein potential of mean force,
and the crystallisation window. Section 2.1 reviews DLVO
theory and subsequently discusses its limitations in modelling proteinprotein interactions in moderately concentrated electrolyte solutions. As mentioned earlier, one of
its major shortcomings is it cannot account for specic
ion effects, which are discussed in Section 2.2. It has also
been found that the proteinprotein interactions are probably anisotropic; weak proteinprotein interactions should
be treated more like molecular recognition processes where
the averaged proteinprotein interactions are only given by
a few highly attractive congurations. These are discussed
in Section 2.3. Furthermore, due to the short-ranged nature
of most proteinprotein interactions, solvation forces need

to be considered. Solvation forces are a major driving force


in protein folding and protein complex formation and are
covered in Section 2.4.
Proteinprotein interactions result from overlapping layers of solvent perturbation around the protein surfaces. Consequently, in order to understand proteinprotein interactions
in salt solutions, it is rst necessary to understand the effect
of salt on the protein solvation free energy. In Section 3, we
provide a general discussion of proteinsolvent interactions.
Much of what is known about proteinsalt interactions has
been determined from extrapolating the results of solubility
studies for model compounds in salt water. These studies are
summarised in Section 3.1. Section 3.2 describes the preferential interaction parameter and its applications, such as in
predicting the effect of salt on protein stability (covered in
Section 3.3) and protein salting-out (covered in Section 3.4).
Because models of protein salting-out have been based on
proteinprotein and on proteinsalt interactions, in Section
3.5 we attempt to show the link between the two types of interactions, thereby gaining some insight into the Hofmeister
effects in proteinprotein interactions.
Finally, we conclude the review in Section 4 with a general discussion of the current outstanding problems and future directions in the molecular thermodynamics of protein
solutions.
1.1. Osmotic pressure
One key measurement that probes the thermodynamic behaviour of a protein solution is membrane osmometry. In this
experiment, a salt solution containing protein is placed in a
chamber that is separated from a second chamber containing
only the salt solution. The two chambers are separated by a
membrane that is impermeable to the protein, but permeable
to the water and the salt. At equilibrium, the pressure of the
protein solution is higher than that of only the electrolyte
solution. This difference in pressure is known as the osmotic
pressure, , which can be expressed as an expansion in the
experimentally accessible protein mass concentration, cp ,

= 1/Mp + B2 (T , w , s )cp
cp RT
+ B3 (T , w , s )cp2 + .

(1)

Here, R is the ideal gas constant, B  2 and B  3 are the second and third osmotic virial coefcients, respectively, T is
absolute temperature, and Mp is the molecular weight of the
protein. The chemical potential of component i is denoted
by i , where i = w for water, i = s for salt, and i = p for
protein.
Experiments such as static light scattering or small angle neutron scattering have also been used to probe the
osmotic pressure of protein solutions. More recently, a highthroughput chromatographic method has been used to measure interactions between proteins in a mobile phase and
proteins attached to a stationary phase (Tessier et al., 2002a).

Relating these measurements to values of B2 requires care as


the immobilized proteins must be at a very low surface coverage to avoid multi-body interactions (Teske et al., 2004).
A less direct method of measuring protein behaviour is dynamic light scattering, where hydrodynamic and thermodynamic contributions to protein behaviour are probed.
1.2. The potential of mean force
An extremely useful concept in interpreting and modelling the behaviour of protein solutions is the potential of
mean force (pmf). Physically, the n-body potential of mean
force is the work (free energy) required to bring n protein
molecules from innite separation to a xed conguration
in the solution. During this process, all the other molecules
in the system (e.g., water, salt, etc.) are allowed to freely
move. In general, the pmf is a many-body interaction that
depends on the relative positions and orientations of the protein molecules and is averaged over all positions of solvent
molecules. In most approximations, however, it is treated
simply as a pairwise additive potential.
The potential of mean force can be related to the structure
of the solution. The two-body contribution to the potential
of mean force, w2 , is directly related to the pair correlation
function, g2 , by McMillan and Mayer (1945)
w2 (r, 1 , 2 ; T , w , s )
= kB T ln g2 (r, 1 , 2 ; T , w , s ),

(2)

where r is the distance between the centres of mass of the


protein molecules, 1 represents the orientation (parameterised by Euler angles or quaternions) of protein 1, and 2
is the orientation of protein 2, and kB is the Boltzmann constant.
Specifying the potential of mean force xes the thermodynamic behaviour of the system. In other words, once the
potential of mean force is known, the structure and the thermodynamics of the system can be determined, in principle.
The potential of mean force depends on the solution conditions. Changing such parameters as the temperature, salt
concentration, salt type, etc. will alter the pmf and, consequently, the thermodynamics of the system. Therefore, being
able to describe the proteinprotein pmf and its dependence
on solution conditions is crucial if one wants to understand
the behaviour of protein solutions.
1.3. Relating pmf to thermodynamic properties
In most protein solutions, water makes up the majority
of the species present in the system. From a theoretical perspective, one could treat all species in the system explicitly, on an equivalent basis. However, the interactions between water molecules are quite complex and play a major
role in determining the behaviour of the solution. Although
formal methods are available to handle these complexities,
the approximations which must be made to make practical

calculations lead to theories that are either extremely computationally difcult or that are not sufciently accurate.
On the other hand, the thermodynamic properties of pure
water and water/salt solutions are well known through experimental measurements. In addition, one can usually make
reasonable approximations for the interactions between solute particles (e.g., proteins) immersed in water/salt solutions
(the potential of mean force). These facts lead to the idea
of treating the solvent as a continuum background and explicitly dealing with only the solute particles. McMillan and
Mayer (1945) provided a theoretical framework for dealing
with these types of systems.
McMillanMayer theory formalises the analogy between
the behaviour of a solute interacting through a solvent (e.g.,
proteins in aqueous salt solutions) and molecules interacting
in a vacuum. As a result, free energy models and equation
of state descriptions of the behaviour of real uids can be
directly used to describe the behaviour of proteins in a solution. The interaction potential between proteins is replaced
by the potential of mean force. Similarly, the various thermodynamic properties of the uid are replaced by properties
of the solution; for example, the uid pressure is replaced
by the osmotic pressure. The virial expansion for the osmotic pressure can be expressed in terms of protein number
concentration, p :

= p + B2 (T , w , s )2p
kB T
+ B3 (T , w , s )3p + ,

(3)

where Bi is the ith coefcient in the virial expansion. The


virial coefcients reect the effective interaction between
protein molecules. For instance, B2 is directly related to the
two-body potential of mean force by McMillan and Mayer
(1945)

B2 =

1
2


dr d1 d2



w2 (r, 1 , 2 ; T , w , s )
exp
1 .
kB T

(4)

This is precisely the same relation found between the second


virial coefcient of a dilute gas and the intermolecular potential. The osmotic second virial coefcient is related to the
experimentally obtained value by B2 = B2 (NA /Mp2 ) where
NA is Avogadros number.
The phase behaviour of uids has parallels for solutions. Below its critical temperature, a uid may exhibit a
vapourliquid phase transition. Within the McMillanMayer
framework, this corresponds to a phase transition between
two liquid phases: one phase dilute in protein, and the other
phase rich in protein.

2. Protein phase diagrams and proteinprotein


interactions
The protein phase diagram is fundamental in understanding and manipulating the behaviour of proteins. Many diseases, such as neurodegenerative diseases (Koo et al., 1999),
cataracts (Pande et al., 2001), and sickle-cell disease (San
Biagio and Palma, 1991), are linked to protein aggregation/phase separation. Furthermore, the efciency of a significant number of biotechnologically relevant operations (e.g.,
protein salting-out, aqueous polymer-based two-phase separations, protein crystallisation and formulation, and protein
renaturation) is highly dependent on the relative location on
the phase diagram.
The phase diagrams of protein solutions exhibit a broad
region of liquidsolid coexistence, within which, there is
a metastable area of liquidliquid coexistence, one liquid
phase dilute in protein and the other concentrated in protein
(see Fig. 1). This has been observed for lysozyme (Taratuta
et al., 1990; Ishimoto and Tanaka, 1977; Broide et al., 1996;
Muschol and Rosenberger, 1997; Grigsby et al., 2001) for crystallins (Schurtenberger et al., 1989; Broide et al., 1991;
Berland et al., 1992) and for hemoglobin (San Biagio and
Palma, 1991). The quantitative details of the phase diagram,
such as the temperature dependence of the protein solubility, is controlled by the precise nature of the proteinprotein
potential of mean force. This, in turn, depends on such properties as the protein size, shape, charge, or on the solution
conditions such as the salt concentration or type, pH, or
polymer concentration.
The phase diagrams of different proteins at different solution conditions are, in general, quite quantitatively different.
However, Guo et al. (1999) discovered that, if the protein
phase diagram is recast in terms of B2 instead of temperature, the phase diagrams of many disparate systems could be
collapsed on a single universal phase diagram. Expressed

Fig. 1. Schematic protein phase diagram where the optimal crystallisation


region is found near the metastable liquidliquid equilibrium.

in these terms, the liquid branch of the liquidsolid equilibrium becomes independent of the solvent; once B2 is known,
the approximate position on the phase diagram is known.
One important application of this universal phase diagram is in protein crystallisation. Determining the optimal
conditions under which to crystallise proteins is especially
important. It has been found that there are only small regions
on the phase diagrams where crystals can be obtained on
reasonable time scales (George and Wilson, 1994). To obtain crystals, one must be located in a window on the phase
diagram bound by 8104 < B2 <2104 ml mol/g2 .
If B2 > 2 104 ml mol/g2 , the solubility of the protein is too high to form quality crystals, whereas if B2 <
8 104 ml mol/g2 , the proteinprotein interactions are
too sticky, and the proteins will stick together randomly
without specic order as is required to form protein crystals.
Although being within the crystallisation window is a necessary condition for crystallisation, it does not guarantee that
high quality crystals will be obtained. Consequently, many
investigations have focused on determining the relationship
between the shape of and location on the phase diagram and
protein crystallisation.
Protein phase diagrams have similarities to those of colloids immersed in polymer solutions (Ilett et al., 1995;
Lekkerkerker et al., 1992; Gast et al., 1983). In these systems, phase separation occurs because there is an effective
attractive interaction between the colloids in the presence
of the polymer. This interaction results from an osmotic
pressure difference between two colloids when polymer
molecules are excluded from between them (Asakura and
Oosawa, 1954). The phase diagram of these systems depends on the range of the osmotic attraction potential, which
scales with the radius of gyration of the polymer, Rg . When
Rg /dcol > 0.3, where dcol is the diameter of the colloid,
the region of liquidliquid coexistence occurs along with
a region of liquidsolid coexistence. For smaller values of
Rg /dcol , the region of liquidliquid coexistence becomes
metastable to that of the liquidsolid region. This general
behaviour has been observed in computer simulations and
theoretical calculations using model potentials, including
the Yukawa potential (Hagen and Frenkel, 1994; Lomba and
Almarza, 1994) and the square-well uid (Daanoun et al.,
1994). In all these studies, the position of the liquidliquid
binodal relative to the liquidsolid coexistence is determined by the range of the potential; as the range of the
potential decreases, the liquidliquid binodal moves further
below the liquidsolid equilibrium.
The same simplied potential models used to reproduce
colloid/polymer phase diagrams have been used to describe protein phase diagrams. The model potentials used
to describe protein phase diagrams need to be sufciently
short-ranged so that the predicted liquidliquid transition is
metastable to the liquidsolid transition as observed with
protein solutions (Haas and Drenth, 1998; Asherie et al.,
1996; Rosenbaum et al., 1996; Poon, 1997). An adhesive
hard-sphere model has been shown to capture much of the

physics (Baxter, 1968). This potential is taken in the limit


of short range where only two parameters are needed to
characterise the potential, the hard sphere diameter and a
stickiness parameter. In this work, the location of the crystallisation window on the phase diagram was determined
and found to be located around the metastable liquidliquid
coexistence curve. The presence of the liquidliquid critical
point might be very important for protein crystallisation
because density uctuations are enhanced in this region
lowering the free energy barrier to the formation of critically sized nuclei (ten Wolde and Frenkel, 1997; Talanquer
and Oxtoby, 1998). However, most of the conditions within
the crystallisation window correspond to temperatures below that at the liquidliquid critical point indicating that
the presence of the light branch of the liquidliquid binodal
curve is important for crystallisation (Haas and Drenth,
1998; Vliegenthart and Lekkerkerker, 2000). The enhancement of nucleation just outside the light branch of the
binodal might occur due to the formation of nuclei from
small droplets of density corresponding to the dense branch
of the binodal (Galkin and Vekilov, 2000). Below the crystallisation window, the liquidliquid binodal is wider and
most conditions will fall within the binodal where the formation of large droplets of high-density is favoured over
the formation of the crystalline nuclei.
Experimentally, it has been found that the value of B2
at the critical temperature, Tc , depends on the solution
conditions (Rosenbaum et al., 1999). For instance the
value of B2 (Tc )/dp3 (where dp is the effective protein diameter) for lysozyme in 0.75 M ammonium sulphate is
about 5 whereas that for lysozyme in 5% w/v sodium
chloride is 12. Because it is very difcult to crystallise
lysozyme in sulphate solutions, whereas chloride solutions
yield lysozyme crystals, it was suggested that nucleation
can occur more easily when there is a larger distance between the liquidsolid line and the liquidliquid critical
point. This effect cannot be captured with models based
on centrosymmetric potentials that give B2 (Tc )/dp3 6,
where this value is insensitive to the form of the potential
(Vliegenthart and Lekkerkerker, 2000). Thus, the relative
position of the crystallisation window with respect to the
liquidliquid critical point is independent of the details of
model centrosymmetric potentials. Because the enhancement of nucleation rates has been attributed to the proximity
of the liquidliquid binodal, it is unlikely that these models
can be used to explain differences in crytallisability for
solutions that fall within the window.
The dependence of B2 (Tc )/dp3 on solution conditions can
be explained using anisotropic potentials (Haas et al., 1999;
Sear, 1999; Curtis et al., 2001; Kern and Frenkel, 2003). For
instance, protein phase diagrams have been modelled using
Wertheims theory for the uid phases, where the protein
molecules are modelled as hard spheres with sticky squarewell sites located on the surfaces. The value of B2 (Tc )/dp3 increases with the number of sticky sites included in the model.
These results are in semi-quantitative agreement of simula-

tion results where it was found that B2 (Tc )/dp3 decreases as


the directionality of the model potential is increased. The
nding that proteinprotein interactions are better described
using anisotropic interactions is not surprising because it is
these interactions that are required to x the orientation of
the protein in the crystal. Other work has shown that models
based on anisotropic interactions provide better predictions
of the width of the liquidliquid binodal (Lomakin et al.,
1999), the entropy of crystallisation (Curtis et al., 2001) and
protein crystal nucleation rates (Dixit and Zukoski, 2000).
The formation of protein crystals can also be interrupted
by a gel transition at high protein concentrations (Muschol
and Rosenberger, 1997). This region of the phase diagram
should be avoided because the formation of the gel arrests
crystallisation, much more so than the formation of the
dense liquid phase. The formation of a gel occurs when protein molecules form a space spanning network, with protein
molecules trapped in cages formed by other proteins. These
cages are much tighter for short-ranged potentials because
protein molecules want to maximise attractive interactions
in the gel. Consequently, as the range of the potential is
shortened, the gel phase becomes more stable with respect to
the dense liquid phase and eventually the liquidliquid critical point can be submerged behind the gel transition (Fof
et al., 2002).
2.1. DLVO theory: Proteins as colloids
Determination of B2 can give valuable information about
the protein phase diagram and protein phase separations.
However, B2 is very difcult to measure, consequently it has
been studied for only a few different proteins and, in many
cases, for only a limited range of solvent conditions. Furthermore, in cases where only a small amount of protein is available, it is difcult to measure B2 . For this reason, it is useful
to understand the molecular origins of the proteinprotein
interactions. In the next section, we cover measurements of
the proteinprotein interactions and descriptions of these interactions using models that include excluded volume, electrostatics, dispersion forces, and solvation forces.
The rst models (Vilker et al., 1981; Haynes et al., 1992)
of proteinprotein interactions were based on DLVO theory
(Verwey and Overbeek, 1948), which was originally developed for modelling the behaviour of lyophobic colloids. In
this approach, the proteins are considered to be hard, polarizable spheres of diameter, dp , with a net charge, Q, uniformly distributed over their surface. Water is treated as a
structureless, continuum dielectric medium with dielectric
constant, . The advantage of using DLVO theory is that
there is an analytical solution for the two-body potential of
mean force, w2 , which is given by the sum of three terms, an
excluded volume interaction, wex , an electric double-layer
force, welec , and a Hamaker dispersion interaction, wdisp ,
w2 (r) = wex (r) + wdisp (r) + welec (r).

(5)

Due to the assumptions in modelling the electric-double


layer force, DLVO theory is applicable for solutions with
very low salt concentrations, although it is often applied to
solutions with greater salt concentrations.
The excluded volume interaction, wex , is simply given by
a hard sphere potential, which is innite for congurations
where protein molecules overlap and vanishes otherwise.
The diameter of the protein corresponds to the diameter of a
sphere that has approximately the same volume of the crystalline protein, which can be determined from the crystal
coordinates. In addition, the potential includes a layer of impenetrable solvent surrounding the protein of thickness, ,
which controls the distance of closest approach for the proteins. Because the proteinprotein interactions are strongest
for the closest separations, B2 calculations are very sensitive to the choice of .  has been approximated as about
which corresponds roughly to the diameter of a water
3 A,
molecule. This value has also been determined from protein diffusion coefcient measurements using dynamic light
scattering (Eberstein et al., 1994).
The electric double-layer force, welec , arises from the electrostatic interactions between the colloids (e.g., proteins) and
electrolytes present in the solution. Within DLVO theory, the
description of this electrostatic interaction is based on the
PoissonBoltzmann equation where the salt ions are treated
as point charges. There are different forms of the doublelayer potential, based on the various approximations used to
solve the PoissonBoltzmann equation. The most common
approximation is based on low electrolyte concentration and
low surface potential expansion, and is given by Vilker et al.
(1981)
welec (r) =

Q2 e(rdp )
,
r (1 + dp /2)2

(6)

where  is the inverse Debye length which controls the range


of attraction. For Eq. (6) to be valid, dp   1, this corresponds to about an ionic strength of between 0.01 M and
0.1 M. The net charge on the protein can be determined using the intrinsic pK a s of the individual amino acids via the
HendersonHasselbach equation or, alternatively, using the
hydronium ion titration data. The latter method is preferred
because pK a s are perturbed due to the surrounding protein
and the presence of salt. It has been assumed the charge is
uniformly distributed over the surface of the protein. As this
is not necessarily the case with proteins, it is sometimes necessary to include corrections due to the asymmetric charge
distributions of the proteins. These include a chargedipole
interaction and a dipoledipole interaction. These latter interactions are generally attractive (Coen et al., 1995; Vilker
et al., 1981; Haynes et al., 1992; Velev et al., 1998).
The Hamaker dispersion potential is given by
AH
wdisp (r) =
6


2
s2 4
2
+ ln
,
+ 2
s2
s 4
s2

where s = 2r/dp , AH is the Hamaker constant and determines the magnitude of the proteinprotein dispersion interaction. Because all proteins have similar compositions they
also have similar Hamaker constants. These values are on
the order of 3 kB T (Roth et al., 1996). This dispersion potential is based on a continuum approximation which breaks
down for very small separations where the potential becomes
innitely attractive at contact. Due to this divergence, the
amount of dispersion attraction included in the model is very
sensitive to the choice of . As a rst approximation, the dispersion interaction is independent of ionic strength because
the time scale for electronic uctuations is much smaller
than the time needed for the reorganisation of the salt ions.
The pH and ionic strength trends of the proteinprotein
interactions as predicted using DLVO theory are based on
the electric double-layer force. Proteinprotein interactions
become more attractive as the salt concentration is increased
due to screening of the repulsive double-layer force, which
is also reduced when the net charge of the protein is decreased by changing the pH toward the isoelectric pH of the
protein. These qualitative trends predicted by DLVO theory
have been veried for small proteins such as lysozyme (see
Fig. 2a, where lysozyme pI 11) (Eberstein et al., 1994;
Velev et al., 1998; Muschol and Rosenberger, 1995; Tardieu
et al., 1999), chymotrypsin (Coen et al., 1995; Haynes et
al., 1992), chymotrypsinogen (Velev et al., 1998), ribonuclease A (Tessier et al., 2002a; Boyer et al., 1996, 1999),
-crystallins (Bonnete et al., 1997), BSA (Vilker et al.,
1981; Tessier et al., 2002a), subtilisin (Pan and Glatz, 2003),
ovalbumin (Curtis et al., 2002a), and myoglobin (Tessier
et al., 2002a), and for larger proteins including apoferritin
(Petsev et al., 2000), aspartyl transcarbamase (Budayova
et al., 1999), urate oxidase (Bonnete et al., 2001), and crystallins (Finet and Tardieu, 2001). Under the solution
conditions where the electric double-layer force is small,
the proteinprotein interactions are determined by the excluded volume of the protein and the attractive Hamaker
dispersion interaction. However, as discussed below, there
are some forces that are not accounted for by DLVO
theory, which are believed to be signicant, especially
when long-range double-layer forces are screened and the
molecular/chemical nature of the protein surface cannot be
ignored.
The applicability of DLVO theory has been further tested
by comparing the model parameters (dp , AH , Q, and )
t to experimental data with physically realistic values of
those parameters. For example, Eberstein et al. (1994) used
dynamic light scattering to study aqueous sodium chloride
solutions with lysozyme in a pH 4.2, 0.10 M sodium acetate buffer. The concentration of sodium chloride in the
solutions was varied from 0.05 to 1.4 M and the experimental data were t to a potential of mean force model
given by Eq. (5). A Hamaker constant of 7.7 kB T and a net
charge of 6.4e gave the best ts of the model to the experimental data. Similar results were obtained by Muschol
and Rosenberger (1995) who investigated tting the data

2
B2 (10-4 mL mol/g )

40

(a)

30

20

-5

10

-10

-15

-10

(b)

-20
0

0.1

0.2

0.3

0.4

0.5
2

(c)

(d)

0
-1

-2

-2

-4

-3
-6

-4

-8

-5
0

0.2

0.4

0.6

0.8

0.5

1.5

2.5

Ionic Strength (M)


Fig. 2. Protein osmotic second virial coefcients plotted as a function of ionic strength in various aqueous electrolyte solutions. (a) Lysozyme in solutions
of sodium chloride at pH 4.5 (circles), pH 6 (squares), or pH 9 (triangles) (Velev et al., 1998). (b) Myoglobin at pH 9 (triangles) (Tessier et al., 2002a),
BSA at pH 6.2 (circles) (Asanov et al., 1997), ovalbumin at pH 7 (squares) (Curtis et al., 2002a) in solutions of ammonium sulfate. (c) Solutions of
lysozyme at pH 4.5 with potassium isothiocyanate (open circle) (Curtis et al., 2002a), with sodium isothiocyanate (closed circle), with sodium nitrate
(squares), with sodium chloride (diamonds), or with sodium acetate (triangles) (Bonnete et al., 1999). (d) Lysozyme in solutions of magnesium bromide
at pH 7.8 (circles) (Tessier et al., 2002b) and ovalbumin in solutions of magnesium chloride at pH 7 (squares) (Curtis et al., 2002a).

using different expressions for the electric double-layer potential. They found that the t parameters were very sensitive to the pmf model indicating that it is difcult to assign
physical meaning to the t parameters, Q and AH . Nevertheless, other studies have obtained reasonable ts of experimental B2 data to DLVO theory for solutions of ovalbumin (AH = 3 kB T , Curtis et al., 2002a), of BSA (AH =
3 kB T , Moon et al., 2000), and of subtilisin (AH = 5.1 kB T ,
Pan and Glatz, 2003).
Velev et al. (1998) compared lysozymelysozyme interactions to those for chymotrypsinogen. In agreement with
previous work, the lysozymelysozyme interactions could
be accurately modelled using the pmf model given by Eq.
(5). However, at a pH greater than 5.3, chymotrypsinogen
interactions are more attractive at low salt concentration than
at high salt concentration indicating the presence of attractive electrostatic interactions. This effect was attributed to
the large dipole moment of chymotrypsinogen. The extra
electrostatic attraction was included in the model by using
dipoledipole and dipolecharge potentials. These latter interactions were also used to describe the pH dependence of
the proteinprotein interactions in aqueous electrolyte solutions of chymotrypsin (Coen et al., 1995; Haynes et al.,
1992) and of BSA (Vilker et al., 1981).
One of the major approximations in DLVO theory is
the use of the PoissonBoltzmann equation to describe

the electrostatic interactions in the colloidal system. The


PoissonBoltzmann theory treats the ions in solution as
point charges, which interact with each other and with the
protein (colloid) only through a mean electrostatic potential.
The theory does not account for effects such as correlations
between ions, non-electrostatic interactions between ions
(e.g., excluded volume interactions, dispersion interactions,
etc.), and image charge interactions between the electrolytes
and the colloidal particles.
Correlations between ions can lead to behaviour that
is qualitatively different from that predicted by the
PoissonBoltzmann equation. One important example is
the electrostatic contribution to the potential of mean
force between two similarly charged surfaces. Bell and
Levine (1958) demonstrated that models based on the
PoissonBoltzmann equation always lead to repulsive interactions between similarly charged surfaces. However, by
including ionion correlations within the hypernetted chain
integral equation, Patey (1980) and later Kjellander and
Marcelja (1984), demonstrated the theoretical possibility
of an attraction with sufciently high surface charges on
the colloidal particles. Computer simulation studies on the
restricted primitive model (Torrie and Valleau, 1980, 1982;
Guldbrand et al., 1984; Valleau et al., 1991) conrmed the
presence of an attractive interaction. Computer simulations
have indicated that the potential of mean force between

two similarly charged hard spheres immersed in an electrolyte solution can be attractive (Gronbech-Jensen et al.,
1998; Wu et al., 1999). More surprisingly, the potential of
mean force between two oppositely charged sphere can be
repulsive (Wu et al., 2000). This effect has signicant implications to the stability and phase behaviour of colloidal
solutions and cannot be captured by the PoissonBoltzmann
equation.

2.2. Specic ion effects: The Hofmeister series


One of the deciencies of the PoissonBoltzmann equation is that ions are treated as point charges and consequently
lose their identity. However, many types of biological interactions depend on the specic ion type. The effect of the ion
is usually related to its position in an empirical series, based
on some measurable effect. One such series is the Hofmeister
series (Hofmeister, 1888), which was originally developed
as a ranking of the salting-out effectiveness of various ions
for globular proteins. The effectiveness is much stronger
for anions, where this series is given by in decreasing order
2

SO2
4 > HPO4 > CH3 COO > Cl > Br > I > SCN
+
+
+
and that for cations is given by Li > Na K > NH+
4
> Mg2+ . Another series that characterises specic ion effects is the lyotropic series (Israelachvili, 1992), where
ions are ranked according to their interactions with water.
High lyoptropic series ions interact with water strongly
leading to large hydration numbers. These ions are termed
kosmotropes because the water molecules surrounding the
ions are structured relative to bulk water. Alternatively, low
lyotropic series ions are termed chaotropes because water
is less structured around these ions. The lyotropic series
is correlated with measurements such as the JonesDole
viscosity B coefcients or molal surface tension increments, both of which reect the afnity of the ion for water
(Collins and Washabaugh, 1985).
In general, the Hofmeister series is similar to the lyotropic
series. However, there are exceptions; for example, multivalent cations are kosmotropes (high lyotropic series ions),
but have low salting-out effectiveness and are consequently
low on the Hofmeister series.
As salt concentration is raised within the salting-out
region, the protein solubility decreases as a result of saltinduced proteinprotein attractions. In this region, the solubility also depends on the type of salt according to itsposition
in the Hofmeister series. Two distinctive trends have been
observed. Generally, if the pH of the solution is above the pI
of the protein, traditional salting-out behaviour is observed
where the effectiveness of the ion at reducing the protein
solubility increases with the ions position in the ascending
Hofmeister series. One explanation for this phenomena is
that the salt ions sequester water molecules and prevent
them from forming favourable hydrogen bonds with the
protein surface (Collins, 2004). Consequently, proteins prefer to form intermolecular interactions between themselves

instead of with water resulting in an effective proteinprotein


attraction. This type of behaviour is illustrated in Fig. 2b,
where B2 is plotted versus ionic strength for ovalbumin (Curtis et al., 2002a), myoglobin (Tessier et al., 2002a), or BSA
(Tessier et al., 2002a; Asanov et al., 1997) in aqueous solutions of ammonium sulphate, a popular salting-out agent.
If the pI of the protein is above the pH of the solution, the
effectiveness of the anion at inducing proteinprotein attraction follows the reverse lyotropic series (Pan and Glatz,
2003; Rieskautt and Ducruix, 1989; Finet et al., 2004).
This reverse lyotropic series effect has been investigated
thoroughly, for solutions containing lysozyme (Curtis et
al., 2002a; Bonnete et al., 1999; Rieskautt and Ducruix,
1989) (see Fig. 2c), where it was found that the chaotropic
anion, SCN , is an excellent crystallisation agent. On the
other hand solutions containing the kosmotropic anion,
SO2
4 , rarely yielded lysozyme crystals. Differences in
lysozymelysozyme interactions as a function of salt type
are observed at low salt concentration ( 0.05 M). This is
in contrast to traditional salting-out behaviour where the
effect of the salt type on protein solubility is not observed
until much greater salt concentrations ( 1.0 M) (Curtis
et al., 2002a). The reverse lyotropic series effect has been
attributed to the formation of an insoluble lysozymeSCN
complex due to strong binding interactions between the
chaotropic anions and the positively charged arginine groups
on the lysozyme surface (Rieskautt and Ducruix, 1989).
In some cases, proteins are salted-in by concentrated
salt solutions. For aqueous sodium chloride solutions containing apoferritin, as salt concentration is initially raised,
the proteinprotein interactions become more attractive due
to screening the electric double-layer repulsion (Petsev et al.,
2000). However, the proteinprotein attraction is at a maximum for a salt concentration of 0.15 M. Above this value, the
proteinprotein interactions become much more repulsive.
Similar effects have been observed from proteinprotein
interaction measurements for solutions containing either
lysozyme or ovalbumin dissolved in aqueous solutions containing magnesium salts (see Fig. 2d) (Curtis et al., 2002a;
Tessier et al., 2002b; Grigsby et al., 2000) and for solutions
of malate dehydrogenase dissolved in various salt solutions
(Costenaro et al., 2002). Also, similar trends have been observed in solutions of divalent cations with solubility studies
of lysozyme (Broide et al., 1996; Grigsby et al., 2001; Benas et al., 2002), of BSA or of
-lactoglobulin (Arakawa et
al., 1990). In these studies, the protein solubility has a minimum at intermediate salt concentrations. These effects have
been attributed to repulsive hydration forces which occur
when heavily hydrated cations bind to negatively charged
surfaces (Israelachvili, 1992). The origin of these forces
is still not clear although it is expected that they originate
from the energy required to disrupt the hydrogen bonding
network around charged or polar surfaces. The strength and
the range of the hydration force was found to scale with
the hydration number of the bound cation which correlates
with the position of the cation in the lyotropic series. It is

interesting to note that the binding of well-hydrated ions


to the protein surface leads to repulsive hydration forces,
whereas the binding of chaotropic anions leads to attractive
forces.
For a few large oligomeric proteins, proteinprotein attraction is not observed over the entire range of concentrated salt solutions. This observation is attributed to the
non-compact structure of these proteins, resulting in a reduced proteinprotein dispersion interaction because the dielectric constant of the protein interior is similar to that of
water (Budayova et al., 1999; Finet and Tardieu, 2001).
Although many different phenomena have been observed
to follow the Hofmeister series, the physical mechanism that
leads to the series is still poorly understood (Kunz et al.,
2004). This is partly due to the sheer diversity of the phenomena. Recently, it has been proposed that the specic
ion effects on proteinprotein interactions may be reconciled by rst understanding the effect of salt on the surface tension of water (Piazza and Pierno, 2000). Salts have
positive surface tension increments due to the desorption of
ions from the airwater interface; ions experience repulsive
image charge interactions at the airwater interface due to
the relatively low dielectric constant of air. The effect of
salt on the surface tension of water was rst captured by
Onsager and Samaras (1934) by incorporating the image
charge interactions into the PoissonBoltzmann equation.
In general, however, the actual surface tension increment is
strongly dependent on the type of salt and not just its valency, an effect that cannot be explained simply by image
charges (which only distinguish ions by their valency). Alternatively, specic ion effects can be captured in models by
including dispersion interactions between ions and between
ions and interfaces (Bostrm et al., 2001; Tavares et al.,
2004). Specic ion effects also become more pronounced
at high salt concentrations, where electrostatic interactions
become highly screened and dispersion interactions become
dominant.
Bostrm and coworkers have used dispersion interactions
to model the inuence of salt type on the net charge of
lysozyme. The dispersion interactions between the protein
and the ions become more attractive with increasing ion polarisability leading to preferential adsorption of the more
polarisable ions (Bostrm et al., 2003). Note that anions
are generally more polarisable than cations, and the polarizability of an ion generally increases with its position
in the descending order of the lyotropic series. This preferential interaction of the protein with various ions can
lead to either an enhancement or a displacement of H+
ions from the surface of the protein. The fractional dissociation of acid and base groups on the surface of the
protein, and thus the net charge of the protein, is determined by the surface pH. Incorporating these effects into the
PoissonBoltzmann equation, Bostrm et al. were able to
explain the observed variation of the surface charge of eggwhite lysozyme with salt concentration, as well as with salt
type.

2.3. Anisotropic interactions


Under most conditions relevant to bioseparations, or in
physiological salt solutions, the electric double-layer potential will be effectively screened and the proteinprotein interactions will be generally short-ranged. In this case, the
details of the proteinprotein interactions will depend on the
chemical nature of the protein surface and on the protein
shape and surface roughness. Most likely, the interactions
will be anisotropic due to the protein surface heterogeneity;
it is these interactions that are important in xing the orientation of a protein molecule in the crystal.
Recent investigations have focused on the effect of surface
anisotropy on the proteinprotein interactions using more
realistic models for proteins where the shape of the protein
is determined from the crystal structure coordinates (Neal
et al., 1999; Asthagiri et al., 1999; Elcock and McCammon,
2001). The key nding was that the net proteinprotein interactions are determined from a few highly attractive orientations marked by a high degree of surface complementarity
much like the types of interactions involved in molecular
recognition events between proteins (Neal et al., 1999). The
primary contribution to the energy of the highly attractive
congurations comes from the dispersion force; electrostatic
interactions contribute to the energy of these interactions,
but do not determine the different orientations that contribute
to B2 . It was postulated that attractive electrostatic interactions occur due to the charge asymmetry that occurs in
the dominant congurations. The importance of anisotropic
interactions were conrmed by showing that the substitution of a single amino acid located in a crystal contact can
lead to large changes in values of B2 for bacteriophage T4
lysozyme (Chang et al., 2000). Here, the orientations of the
proteins in the crystals have been identied as the attractive
congurations sampled by the protein in the solution.
The all-atomistic models of proteinprotein interactions
are very sensitive to the surface properties of proteins. In
the model used by Elcock and McCammon, the value of
B2 is very sensitive to the protonation states of the charged
residues located on the complementary surfaces; these protonation states could change upon complexation. This sensitivity of the model to the protein surface properties is expected because slight alterations in protein surface charge
(Chang et al., 2000) or surface hydrophobicity (Curtis et al.,
2002b; Ho and Middelberg, 2003) can lead to large changes
in the values of B2 . These changes cannot be accounted for
by using pmf models based on centrosymmetric potentials,
where the large changes in values of B2 would be accounted
for by unrealistic changes to the t parameters such as AH ,
Q, or .

2.4. Solvation forces


Because of the similarities between weak proteinprotein
interactions and the specic proteinprotein interactions

that stabilise biologically relevant complexes, recent work


has focused on using methods similar to those used in
protein docking algorithms for calculation of the weak
proteinprotein interactions (Elcock and McCammon,
2001). One of the interactions that is not included in DLVO
theory, but is a major driving force in protein complex formation, is the hydrophobic interaction (Horton and Lewis,
1992; Jones and Thornton, 1996; Janin and Chothia, 1990).
The hydrophobic interaction is a type of solvation force.
The structure of water is perturbed near the protein surface
where water molecules cannot hydrogen bond freely with
other water molecules. A solvation force occurs when the
layers of solvent perturbation overlap (Israelachvili, 1992).
This force could be either repulsive or attractive depending
on the chemical nature of the interacting surfaces. If the interacting surfaces contain non-polar groups, then the force
is referred to as hydrophobic and is generally attractive. On
the other hand, the term, hydration forces, is used to refer
to the solvation interaction between surfaces containing polar or charged groups. These forces are generally repulsive
because energy is required to dehydrate polar or charged
groups.
Thermodynamically, the interaction between a solute and
the solvent is given by the solvation free energy, which is
the work to transfer a solute from an ideal gas phase into
the solvent. A solutesolute solvation force is related to the
change in the solvation free energy of a solute due to the
proximity of a second solute. If the solutesolvent interaction is sufciently short-ranged, the solvation force can be
approximated by the area that has been removed from the
solvent due to the solutesolute interaction multiplied by
the solvation free energy of the solute. This is known as the
solvent accessible surface area (SASA) approach, which is
commonly used in protein folding and protein docking algorithms (Horton and Lewis, 1992; Chothia, 1974; Eisenberg and McLachlan, 1986). Within the SASA method, the
solvation free energy, Gsolv , is approximated by

i Ai .
(7)
Gsolv =
i

The sum is over all different atomic groups, i, Ai is the area


of group i that is exposed to the solvent, and i is the atomic
solvation parameter of group i. The atomic solvation parameters are determined by tting model compound solubility
data, where it is assumed that the free energy to bury a surface group is similar to the free energy of transferring that
group from a non-polar phase to water. The applicability of
Eq. (7) is based on the additivity approximation which states
that the solvation free energy of the entire protein is given by
the sum of the solvation free energies of the surface groups
(Eisenberg and McLachlan, 1986; Hermann, 1972). This approximation was based on the results of solubility studies
of alkane molecules where it was found that the solvation
free energies scaled with the alkane chain length. However,
because the interaction between polar groups and the surrounding water molecules is longer-ranged then one solvent

layer, the additivity approximation might not necessarily be


valid (Wang and BenNaim, 1997; Chalikian et al., 1994).
There is also evidence that the interaction between surfaces
with small hydrophobic patches is different from that with
large hydrophobic patches. The crossover between the two
types of interactions occurs at a length scale of approximately 1 nm (Lum et al., 1999).
The SASA approach has been included in models for
calculating weak proteinprotein interactions (Elcock and
McCammon, 2001; Curtis et al., 2002b; Ho and Middelberg, 2003). In the pmf model of Elcock and McCammon
(Elcock and McCammon, 2001), both the short-range dispersion force and the solvation force were modelled using a
SASA potential with a surface-averaged solvation parameter. This surface free energy parameter was t to the B2 data
for solutions of lysozyme in sodium chloride. Interestingly,
the t parameter was similar to the surface free energy obtained from transferring apolar compounds from nonpolar
solvents into water. An alternative approach to model solvation forces was used by Asthagiri et al. (1999), where a hybrid Hamaker/LennardJones model was used to calculate
the short-range attractive interactions. Here, the Hamaker
dispersion potential was used to describe the proteinprotein
interaction for large separations. For close separations, a
LennardJones potential is used to calculate the interactions between groups on opposite protein molecules with
The cuta surface-to-surface separation of less than 3 A.
off corresponds to excluding one solvent molecule from between the closely separated surfaces.

3. Proteinsolvent interactions
As discussed above, the interface between the protein and
the solvent plays an important role in the stability of the
protein. If the interaction between the solvent and protein
is favourable, the protein will prefer to remain in solution
surrounded by the solvent. However, if the interaction between the protein and the solvent is unfavourable, the protein molecules will prefer to form proteinprotein contacts
resulting in an effective proteinprotein attraction.
The solvation free energy of a compound can be decomposed into the sum of two contributions
Gsolv = Gcav + Gint ,

(8)

where Gcav is the work to create a cavity in the solvent, and


Gint is the work of turning on the interactions between the
solute and the solvent molecules.
The work required to create very large cavities (large compared to molecular length scales) can be readily estimated
from the bulk properties of the solvent. For cavity sizes much
less than 100 nm, this work is given by
Gcav A ,

(9)

where is the surface tension of the solvent, and A is the


surface area of the cavity. This form leads one to consider
that the solvation of the protein molecules can, in principle,
be related directly to the surface area of the protein and
the macroscopic surface tension of the solvent. From this, it
would follow that the effect of salt on Gcav is related to the
surface tension increment of the salt.
However, the use of Eq. (9) is not well justied when the
size of the cavity becomes of the order of the correlation
lengths present in the solvent. The work of cavity formation is then related to the probability that there will spontaneously be no solvent molecules in a volume of the shape of
the cavity. According to scaled-particle theory (Reiss et al.,
1959; Pierotti, 1965; Stillinger, 1973) or information theory
(Hummer et al., 1996, 1998), this probability is related to
the volume of the cavity, not the exposed surface area of
the cavity as given in Eq. (9). Lum and coworkers (1999)
were able to develop a theory that successfully interpolates
between the small cavity expressions given by information
theory and the limiting large cavity expression given by the
macroscopic surface tension. Thus, the work to form a microscopic cavity, Gcav , can be interpreted as a curvaturedependent surface tension,
Gcav = RA ,

(10)

where R is a factor to correct for the curvature of the interface. Note that Eq. (10) must be applied with care, as the effect of salt on the microscopic surface free energy, R , is not
necessarily related to its effect on the bulk surface tension,
, as the molecular origins of the microscopic and macroscopic surface tensions are different. However, assuming the
two effects are related has proven to be a good starting point
in interpreting the properties of proteins.
3.1. Solubility studies of model compounds
Most knowledge of proteinsalt interactions has been determined from solubility/partitioning studies of model compounds where the effects of salt on the different atomic
groups of the protein can be clearly delineated (Baldwin,
1996). The effectiveness of an ion at salting-out alkanes is
correlated with the molal surface tension increment of the
salt, and thus, the effectiveness follows the lyotropic series. In addition to salting-out non-polar groups on a protein, ions also interact favourably with the peptide groups.
The effectiveness of ion salting-in increases with the position of the ion in the descending order of the Hofmeister
series (Robinson and Jencks, 1965; Nandi and Robinson,
1972; Schrier and Schrier, 1967). Salting-in is primarily due
to favourable electrostatic interaction between the ion and
the large dipole moment of a peptide bond; however, the
cause of ion specicity is not clear (Baldwin, 1996). Some
believe that the iondipole interaction is specic (Nandi
and Robinson, 1972), whereas others believe the iondipole
interaction is non-specic, and the specicity arises from

unfavourable interactions of the ions with nearby non-polar


groups (Hamabata and von Hippel, 1973). The latter explanation can be used to understand the salting-in interaction
of chaotropic anions. However, divalent cations also salt-in
the peptide group, although it is expected that these ions
will interact unfavourably with non-polar groups due to the
high surface tension increment of these ions. The salting-in
interaction of divalent cations has been attributed to the hydrogen bonding interactions of the heavily hydrated cation
with the peptide dipole (Collins, 2004). As a consequence
of this interaction, divalent cations are high on the lyotropic
series, but low on the Hofmeister series.
In addition, there are specic interactions between the salt
ions and the charged groups of peptides/proteins. These interactions are stronger between similar ions (i.e., where both
ions are either kosmotropes or both are chaotropes) then between dissimilar ions (Collins, 1997). Thus, the binding ability of monovalent anions to protein charges should follow
the reverse Hofmeister series because the positively charged
guanidinium and imizadolium groups are chaotropic. However, this electrostatic interaction is stronger if the ion is divalent versus monovalent, as with the case of SO2
4 (Nandi
and Robinson, 1972). The opposite trend is observed for the
interaction of cations with the protein charged groups. The
positively charged carboxylate ions form very strong interactions with kosmotropic cations, such as Na+ or the divalent cations Mg2+ or Ca2+ . The strong interactions between
the divalent cations and the charged groups of proteins play
very important biological roles.
3.2. Preferential interaction parameter measurements
The solvent environment directly surrounding a protein
is different than that in the bulk solution, due to preferential interactions between the protein surface and the
various solvent components. The difference of the salt composition in the region next to the protein surface and of
that in the bulk solution is experimentally determinable in
terms of the proteinsalt preferential interaction parameter,
(jms /jmp )T ,w ,s , where mi is the molality of component i (Timasheff, 1998; Timasheff and Arakawa, 1988).
Favourable proteinsalt interactions are characterised by a
positive value of (jms /jmp )T ,w ,s . For this case, there is
an excess of salt near the surface of the protein, relative
to the bulk solution. If (jms /jmp )T ,w ,s < 0, the solvent
layer around the protein is depleted of salt. The perturbation
of the protein chemical potential by the addition of salt is
related to the preferential interaction parameter by






jGsolv
js
jms
=
, (11)
jms T ,P ,mp
jmp T , , jms T ,P ,mp
w

where P is pressure. Accordingly, the protein chemical potential is determined from the proteinsalt interactions. If
the proteinsalt interactions are favourable, the addition of
salt will decrease the solvation free energy of the protein.

Whereas, if the salt is preferentially excluded from the domain of the protein, addition of salt will raise the protein
solvation free energy.
The consequences of Eq. (11) are similar to the predictions
of the Gibbs adsorption isotherm. According to the Gibbs
adsorption isotherm, the change in the macroscopic surface
tension of an airsalt/water interface is given by the amount
of adsorption or desorption of the ions at the interface. Salts
have positive surface tension increments because they are
excluded from the interface due to the formation of image
charges. Arakawa and Timasheff (1982,1984a) (Arakawa et
al., 1990) have shown that the preferential exclusion of ions
from the proteinsolvent interface can also be predicted from
the salt molal surface tension increment. They determined
the curvature correction factor, R, by equating measured
proteinsalt preferential interaction parameters to the values
predicted using Eqs. (10) and (11)


jms
jmp

calc
T ,w ,s

= R

NA A(j /jms )T ,P ,mp


(js /jms )T ,P ,mp

(12)

For solutions containing strongly excluded salts (such as


Na2 SO4 ), the curvature correction factor ranges from R=0.5
to 0.7. The obtained values of R are in semi-qualitative agreement with the values predicted using the theory of Lum and
coworkers (1999). Eq. (12) is based on the assumption that
the surface free energy of a microscopic cavity is related to
the macroscopic surface tension. The result that the exclusion of the salt from a microscopic protein surface can be
predicted from the macroscopic surface tension increment
is intriguing and needs further studies. Factors that might
account for this effect are the inuence of image charges on
the macroscopic and microscopic surface tensions of aqueous salt solutions.
Eq. (12) provides an upper bound on the preferential exclusion of the salt. The difference between the calculated
value of the proteinsalt preferential interaction parameter
and the measured value is attributed to preferential binding
of the salt to the protein. The amount of compensation or
binding is found to follow the reverse lyotropic series; the
monovalent anion SCN has a large preferential binding
to proteins, whereas Cl is weakly excluded (Arakawa and
Timasheff, 1982) from proteins. Strong proteinsalt preferential binding is also observed with solutions of divalent
cations such as Mg2+ or Ca2+ , even though salts of these
ions tend to have high surface tension increments (Arakawa
and Timasheff, 1984a; Arakawa et al., 1990). These results
are in agreement with the solubility data of model compounds where the preferential interactions are usually attributed to the interaction of the salt ion and the peptide bond.
Another nding is that the preferential interactions are additive. For example, generally, MgSO4 is preferentially excluded from protein surfaces, whereas MgCl2 is only weakly
excluded. The Mg2+ ion can overcome the weak exclusion
of the Cl ions, however, it cannot overcome the strong preferential exclusion of the kosmotropic SO2
4 anion. While the

preferential exclusion does not depend on the pH or the salt


concentration, the preferential binding interactions do. The
preferential binding of MgCl2 to proteins increases with increasing salt concentration and increasing pH indicating that
the charge density of the protein affects the binding. This
behaviour has also been observed with solutions of ArgHCl
and lysozyme or BSA (Kita et al., 1994).
3.3. Protein stabilisation
Measurements of preferential interaction parameters have
provided valuable insights into various reactions including
protein stabilisation, protein solubility, self-assembly of subunit systems, binding of ligands to proteins, and protein
complex formation (Timasheff, 1998). These investigations
are based on the Wyman linkage relation which relates the
equilibrium constant for a reaction to the preferential interactions between the solvent and the reacting system and between the solvent and the product (Wyman, 1964; Tanford,
1969; Aune et al., 1971). When applied to protein stabilisation by salt, the Wyman linkage relation is given by






j ln K uf
jms d
jms n
=

,
js
jmp T , ,
jmp T , ,
T ,P ,mp
w

(13)
where K uf is the equilibrium constant for the protein unfolding reaction, and superscripts d and n denote the denatured
(unfolded) and native (folded) protein. Protein stabilisation
is determined by the difference of the preferential interactions of the salt with the native state and with the denatured
state. Stabilisation is favoured if the preferential binding of
the salt to the native state is greater or the preferential exclusion is lesser than that of the unfolded protein.
The proteinsalt interactions can only be measured with
respect to one of the states; generally the interactions are
measured with the native state and then inferred for the denatured state (Timasheff, 1998). Preferential exclusion of
the salt due to the surface tension increment effect is a nonspecic interaction between the salt and the protein surface;
if this effect dominates the proteinsalt interaction, the salt
stabilises the protein because the unfolded state presents a
larger surface than the folded state, thus more salt is excluded from the unfolded state. However, in cases where
preferential exclusion is counteracted by preferential binding of the salt to the protein surface, the solvation interactions of the denatured protein cannot be inferred from those
of the native protein. This is because proteinsalt binding
interactions depend on the chemical nature of the protein
surface which is different for the unfolded and the folded
protein. For these cases, both protein stabilisation or protein
destabilisation are possible. For instance, most guanidinium
salts destabilise proteins due to extensive binding between
the guanidinium ions and the peptide groups exposed upon
protein unfolding. For this reason, salts such as GuaHCl
and GuaSCN are used as denaturants. However, GuaSO4 is

a protein stabiliser because the exclusion of the SO2


4 ion
overcomes the binding interactions of the guanidinium ion
(Arakawa and Timasheff, 1984b).

However, salts such as MgCl2 can have a salting-in effect because preferential binding to the protein will favour
solubilization.

3.4. Protein salting-out


3.5. Linking proteinprotein and proteinsalt interactions
In the same manner that salts prevent proteins from exposing surface to the solvent, they can encourage burying the
surfaces in the formation of proteinprotein contacts like the
ones that occur in protein precipitates. For this reason, strong
salting-out agents are good stabilisers of protein structure.
Note that this correlation is not as strong for weakly interacting salts, because in this case the chemical nature of the
protein surface that is buried is different from that surface
which is exposed during unfolding.
Melander and Horvath (1977) were the rst to link protein
salting-out to the molal surface tension increment. In their
work, the precipitated phase was treated as a pure phase and
proteinprotein interactions were neglected, in which case
the protein solubility is determined from the equilibrium
condition
ig

cp = lp = p + Gsolv + RT ln Sp ,

(14)

where superscripts c, l, and ig are used to denote the solid


phase, the liquid phase and the standard state ideal gas, respectively, and Sp is protein solubility. According to this
model, the solubility of the protein is determined from the
variation of the protein solvation free energy with respect to
salt concentration. At low salt concentration, salting-in occurs because there are favourable interactions between the
charged protein and its electric double layer. As salt concentration is increased further, the solubility goes through a
maximum, after which the solubility decreases because the
salt raises the solvation free energy of the protein according
to its molal surface tension increment.
Experimentally, it has been found that protein crystals
contain a considerable amount of solvent. Consequently,
the protein solubility is determined by the difference of the
proteinsalt interactions in the uid phase and those in the
solid phase according to





jp c
d ln Sp
1
jGsolv l

.
=

d ms
RT
jms T ,P ,mp
jms T ,P ,mp
(15)
Arakawa et al. (1990) determined (jp /jms )cT ,P ,mp
from experimental measurements of protein solubility
and of (jms /jmp )lT , , . They found that variations in
w

(jp /jms )cT ,P ,mp parallelled changes in (jGsolv /jms )lT ,P ,mp
indicating that the proteinsalt interactions in the liquid
phase were representative of those interactions in the solid
phase. The absolute values of the proteinsalt interactions
in the precipitated phase were smaller because there is less
surface exposed to the solvent in this phase, consequently
salting-out always occurs if salt is preferentially excluded.

Protein stabilisation and protein solubility can be understood in terms of proteinsalt interactions. However,
these interactions are not currently included in pmf models. Timasheff and Arakawa (1988) proposed an expression
for the proteinprotein association based on the Wyman
linkage relation for a monomerdimer equilibrium where
the difference in the preferential interactions of the salt
with the dimer and with the monomers determines the
effect of salt on the proteinprotein association constant.
Because a protein dimer has less solvent exposed surface
than that of two monomers, the absolute magnitude of
(jms /jmp )T ,w ,s will be larger for the monomers than for
the dimer. Consequently, preferential exclusion of the salt
from the protein surface will favour protein association.
On the other hand, a preferential proteinsalt binding will
result in an effective proteinprotein repulsion, where the
protein prefers to be surrounded by the solvent rather than
making a proteinprotein contact.
Using the Wyman linkage relation for protein association
provides a method for linking preferential interaction parameters to proteinprotein interactions, but is only useful when
the end-states are well delineated (e.g., a monomerdimer
equilibrium). A more appropriate approach to linking these
interactions might be by using the solvent accessible surface
area potential where the proteinsalt preferential interaction
parameters can be linked to the atomic solvation parameters using Eqs. (7) and (11). This approach was followed
by Curtis et al. (2002a) who t a surface-averaged atomic
solvation parameter to B2 data as a function of salt concentration for lysozyme in solutions of (NH4 )2 SO4 or NaCl,
and for ovalbumin in solutions of KSCN or (NH4 )2 SO4 .
In all cases, semi-quantitative agreement was obtained between the salt increment of the t atomic solvation parameter
and the molal surface tension increment of the salt indicating this method could provide a rst approximation at connecting experimentally measured proteinsalt interactions to
proteinprotein interactions.
It is interesting to compare the proteinprotein interactions of ovalbumin and of lysozyme in solutions of
KSCN. Very little attraction is observed between ovalbumin
molecules in concentrated solutions of KSCN, whereas the
lysozymelysosyme interactions are very attractive in solutions dilute in KSCN. The difference in these behaviours is
related to the binding afnity of SCN to protein surfaces
(Curtis et al., 2002a). If binding of the ion to the protein
surface is weak, the potential of mean force is determined
by the work to remove the solvent. When the ion is SCN ,
the force is only weakly attractive due to the balance in the
preferential exclusion of the salt with the anion preferential

binding. However, if the ion binding is strong, the force


is determined from the potential of mean force between
the proteinanion complexes. When the bound anion is
SCN , the forces between the complexed proteins are more
attractive than those between the uncomplexed proteins.
Here, the positively charged residues are identied as the
strong ion-binding sites and the peptide groups as the weak
ion-binding sites.
For protein solutions containing divalent cations, preferential binding of cations to proteins occurs over broad ranges
of pH and salt concentration. This preferential binding is
associated with repulsive hydration forces; these forces are
repulsive because the binding of strongly hydrated cations
to proteins leads to increased protein hydration.
Two general trends can be deduced from the results
of preferential interaction parameter measurements. If the
measured preferential interaction parameter is only due
to the preferential exclusion of the salt, the effect of the
ion is to dehydrate the protein surface resulting in an
effective proteinprotein attraction, this effect has been
correlated with the surface tension increment of the salt.
However, if the protein surface chemistry is altered by ion
binding, the salting-out behaviour follows the reverse lyotropic series. In this case, the binding of heavily hydrated
cations to the protein surface is associated with repulsive
proteinprotein interactions, whereas the interactions between proteinchaotropic anion complexes are attractive.

4. Conclusions
Recent studies of proteinprotein interactions have been
focused at determining the origin of the crystallisation window. Analysis of protein phase diagrams and those generated using model potentials have shown that the this window
surrounds a metastable liquidliquid binodal, near which the
crystallisation rate is enhanced due to large density uctuations. However, crystallisation is not guaranteed for locations
in the window. One possible reason is that the relative position of the liquidliquid binodal to that of the window is not a
constant, but depends on the form of the proteinprotein potential. This effect has not been accounted for by using centrosymmetric potentials. However, it has been shown that the
relative position of the binodal in the crystallisation window
depends on the anisotropy of the proteinprotein potential.
The next step in developing a more accurate crystallisation
diagnostic is to link the model anisotropic potentials to the
actual proteinprotein interactions. However, currently, the
molecular origins of proteinprotein interactions are poorly
understood.
One of the major challenges in understanding protein
protein interactions is to explain the specic ion effect. The
proteinprotein interactions are determined by the balance
between the preferential exclusion and the preferential binding of the salt to the protein. Both these proteinsalt interactions are dominated by Hofmeister effects. For instance,

the strength of the anion interactions with the peptide dipole


follows the reverse Hofmeister series, while the interaction
between the charged groups on the protein surface and the
counter-ions depends on the nature of each ion in the interacting pair. The effect of ion binding on the proteinprotein
interactions also depends on the nature of the ion. The binding of high lyotropic series cations to the protein surface is
linked to repulsive hydration forces whereas the binding of
chaotropic anions to the protein surface is associated with
attractive proteinprotein interactions.
By including ionprotein and ionion dispersion potentials in their model, Bostrm and coworkers have been able
to rationalize various specic ion effects, such as the dependence of the protein surface charge on salt type. However,
a substantial part of the inuence of salt on proteinprotein
interactions is related to the interaction of the ion with water. For instance, the salting-out effect of kosmotropic salts
has been attributed to the ability of the ions to prevent water from forming favourable hydrogen bonds with the polar groups on the protein surface and, more generally, to
the structure-making ability of the ion. In these cases, it
is likely that water cannot be treated as a continuum, but
its discrete nature needs to be taken into account. Because
the salting-out of proteins is often correlated with the molal surface tension increment of the salt, understanding the
molecular origin of surface tension might give insight into
the origins of proteinprotein interactions and their relation
to the Hofmeister series. A recent issue of Current Opinion
in Colloid and Interface Science (2004, vol. 9) was devoted
to uncovering Hofmeister effects in a whole range of experiments. Once these interactions are determined, it should be
possible to include these effects in more realistic models of
the proteinprotein interactions.
References
Arakawa, T., Timasheff, S.N., 1982. Preferential interactions of proteins
with salts in concentrated-solutions. Biochemistry 21 (25), 6545.
Arakawa, T., Timasheff, S.N., 1984a. Mechanism of protein salting in and
salting out by divalent-cation saltsbalance between hydration and
salt binding. Biochemistry 23 (25), 5912.
Arakawa, T., Timasheff, S.N., 1984b. Protein stabilization and
destabilization by guanidinium salts. Biochemistry 23 (25), 5924.
Arakawa, T., Bhat, R., Timasheff, S.N., 1990. Preferential interactions
determine protein solubility in 3-component solutionsthe MgCl2
system. Biochemistry 29 (7), 1914.
Asakura, S., Oosawa, F., 1954. Interaction between two bodies immersed
in a solution of macromolecules. Journal of Chemical Physics 22, 1255.
Asanov, A.N., DeLucas, L.J., Oldham, P.B., Wilson, W.W., 1997.
Interfacial aggregation of bovine serum albumin related to
crystallization conditions studied by total internal reection
uorescence. Journal of Colloid and Interface Science 196 (1), 6273.
Asherie, N., Lomakin, A., Benedek, G.B., 1996. Phase diagram of colloidal
solutions. Physical Review Letters 77 (23), 4832.
Asthagiri, D., Neal, B.L., Lenhoff, A.M., 1999. Calculation of short-range
interactions between proteins. Biophysical Chemistry 78 (3), 219.
Aune, K.C., Goldsmith, L.C., Timasheff, S.N., 1971. Dimerization of
alpha-chymotrypsin. 2. Ionic strength and temperature dependence.
Biochemistry 10 (9), 1617.

Baldwin, R.L., 1996. How Hofmeister ion interactions affect protein


stability. Biophysical Journal 71 (4), 2056.
Baxter, R.J., 1968. Percus-Yevick equation for hard spheres with surface
adhesion. Journal of Chemical Physics 49 (6), 2770.
Bell, G.M., Levine, S.M., 1958. Statistical thermodynamics of
concentrated colloidal solutions. 2. General theory for the double layer
forces on a colloidal particle. Transactions of the Faraday Society 54,
785.
Benas, P., Legrand, L., Ries-Kautt, M., 2002. Strong and specic effects of
cations on lysozyme chloride solubility. Acta Crystallographica Section
D-Biological Crystallography 58, 1582.
Berland, C.R., Thurston, G.M., Kondo, M., Broide, M.L., Pande, J., Ogun,
O., Benedek, G.B., 1992. Solidliquid phase boundaries of lens protein
solutions. Proceedings of the National Academy of Sciences of USA
89 (4), 1214.
Bonnete, F., Malfois, M., Finet, S., Tardieu, A., Lafont, S., Veesler, S.,
1997. Different tools to study interaction potentials in gamma-crystallin
solutions: relevance to crystal growth. Acta Crystallographica Section
D-Biological Crystallography 53, 438.
Bonnete, F., Finet, S., Tardieu, A., 1999. Second virial coefcient:
variations with lysozyme crystallization conditions. Journal of Crystal
Growth 196 (24), 403.
Bonnete, F., Vivares, D., Robert, C., Colloch, N., 2001. Interactions in
solution and crystallization of aspergillus avus urate oxidase. Journal
of Crystal Growth 232 (14), 330.
Bostrm, M., Williams, D.R.M., Ninham, B.W., 2001. Surface tension
of electrolytes: specic ion effects explained by dispersion forces.
Langmuir 17 (15), 4475.
Bostrm, M., Williams, D.R.M., Ninham, B.W., 2003. Specic ion effects:
the role of co-ions in biology. Europhysics Letters 63 (4), 610.
Boyer, M., Roy, M.O., Jullien, M., 1996. Dynamic light scattering study
of precrystallizing ribonuclease solutions. Journal of Crystal Growth
167 (12), 212.
Boyer, M., Roy, M.O., Jullien, M., Bonnete, F., Tardieu, A., 1999. Protein
interactions in concentrated ribonuclease solutions. Journal of Crystal
Growth 196 (24), 185.
Broide, M.L., Berland, C.R., Pande, J., Ogun, O.O., Benedek, G.B., 1991.
Binary-liquid phase separation of lens protein solutions. Proceedings
of the National Academy of Sciences of the United States of America
88 (13), 5660.
Broide, M.L., Tominc, T.M., Saxowsky, M.D., 1996. Using phase
transitions to investigate the effect of salts on protein interactions.
Physical Review E 53 (6B), 6325.
Budayova, M., Bonnete, F., Tardieu, A., Vachette, P., 1999. Interactions
in solution of a large oligomeric protein. Journal of Crystal Growth
196 (24), 210.
Chalikian, T.V., Sarvazyan, A.P., Breslauer, K.J., 1994. Hydration and
partial compressibility of biological compounds. Biophysical Chemistry
51 (23), 89.
Chang, R.C., Asthagiri, D., Lenhoff, A.M., 2000. Measured and calculated
effects of mutations in bacteriophage T4 lysozyme on interactions in
solution. ProteinsStructure Function and Genetics 41 (1), 123.
Chi, E.Y., Krishnan, S., Randolph, T.W., Carpenter, J.F., 2003. Physical
stability of proteins in aqueous solution: mechanism and driving forces
in non-native protein aggregation. Pharmaceutical Research 20 (9),
1325.
Chothia, C., 1974. Hydrophobic bonding and accessible surface-area in
proteins. Nature 248 (5446), 338.
Coen, C.J., Blanch, H.W., Prausnitz, J.M., 1995. Salting-out of aqueous
proteins-phase-equilibria and intermolecular potentials. A.I.Ch.E.
Journal 41 (4), 996.
Collins, K.D., 1997. Charge density-dependent strength of hydration and
biological structure. Biophysical Journal 72 (1), 65.
Collins, K.D., 2004. Ions from the Hofmeister series and osmolytes: effects
on proteins in solution and in the crystallization process. Methods 34,
300.

Collins, K.D., Washabaugh, M.W., 1985. The Hofmeister effect and the
behavior of water at interfaces. Quarterly Reviews of Biophysics 18
(4), 323.
Costenaro, L., Zaccai, G., Ebel, C., 2002. Link between proteinsolvent
and weak proteinprotein interactions gives insight into halophilic
adaptation. Biochemistry 41 (44), 13245.
Curtis, R.A., Blanch, H.W., Prausnitz, J.M., 2001. Calculation of phase
diagrams for aqueous protein solutions. Journal of Physical Chemistry
B 105 (12), 2445.
Curtis, R.A., Ulrich, J., Montaser, A., Prausnitz, J.M.,
Blanch, H.W., 2002a. Proteinprotein interactions in concentrated
electrolyte solutionsHofmeister-series effects. Biotechnology and
Bioengineering 79 (4), 367.
Curtis, R.A., Steinbrecher, C., Heinemann, A., Blanch, H.W., Prausnitz,
J.M., 2002b. Hydrophobic forces between protein molecules in aqueous
solutions of concentrated electrolyte. Biophysical Chemistry 98 (3),
249.
Daanoun, A., Tejero, C.F., Baus, M., 1994. Van-der-Waals theory for
solids. Physical Review E 50 (4), 2913.
Dixit, N.M., Zukoski, C.F., 2000. Crystal nucleation rates for particles
experiencing short-range attractions. Applications to proteins. Journal
of Colloid and Interface Science 228 (2), 359.
Eberstein, W., Georgalis, Y., Saenger, W., 1994. Molecular-interactions
in crystallizing lysozyme solutions studied by photon-correlation
spectroscopy. Journal of Crystal Growth 143 (12), 71.
Eisenberg, D., McLachlan, A.D., 1986. Solvation energy in protein folding
and binding. Nature 319 (6050), 199.
Elcock, A.H., McCammon, J.A., 2001. Calculation of weak
proteinprotein interactions: the pH dependence of the second virial
coefcient. Biophysical Journal 80 (2), 613.
Finet, S., Tardieu, A., 2001. Alpha-crystallin interaction forces studied
by small angle X-ray scattering and numerical simulations. Journal of
Crystal Growth 232 (14), 40.
Finet, S., Skouri-Panet, F., Casselyn, M., Bonnete, F., Tardieu, A., 2004.
The Hofmeister effect as seen by SAXSs in protein solutions. Current
Opinion in Colloid and Interface Science 9 (12), 112.
Fof, G., McCullagh, G.D., Lawlor, A., Zaccarelli, E., Dawson, K.A.,
Sciortino, F., Tartaglia, P., Pini, D., Stell, G., 2002. Phase equilibria
and glass transition in colloidal systems with short-ranged attractive
interactions: application to protein crystallization. Physical Review E
65 (31), 031407/1.
Galkin, O., Vekilov, P.G., 2000. Control of protein crystal nucleation
around the metastable liquidliquid phase boundary. Proceedings of
the National Academy of Sciences of the United States of America 97
(12), 6277.
Gast, A.P., Hall, C.K., Russel, W.B., 1983. Polymer-induced phase
separations in non-aqueous colloidal suspensions. Journal of Colloid
and Interface Science 96 (1), 251.
George, A., Wilson, W.W., 1994. Predicting protein crystallization from a
dilute-solution property. Acta Crystallographica Section D-Biological
Crystallography 50, 361.
Grigsby, J.J., Blanch, H.W., Prausnitz, J.M., 2000. Diffusivities of
lysozyme in aqueous MgCl2 solutions from dynamic light-scattering
data: effect of protein and salt concentrations. Journal of Physical
Chemistry B 104 (15), 36453650.
Grigsby, J.J., Blanch, H.W., Prausnitz, J.M., 2001. Cloud-point
temperatures for lysozyme in electrolyte solutions: effect of salt type,
salt concentration and pH. Biophysical Chemistry 91 (3), 231.
Gronbech-Jensen, N., Beardmore, K.M., Pincus, P., 1998. Interactions
between charged spheres in divalent counterion solution. Physica A
261, 74.
Guldbrand, L., Jnsson, B., Wennerstrm, H., 1984. Electrical doublelayer forcesa Monte-Carlo study. Journal of Chemical Physics 80,
2221.
Guo, B., Kao, S., McDonald, H., Asanov, A., Combs, L.L., Wilson, W.W.,
1999. Correlation of second virial coefcients and solubilities useful
in protein crystal growth. Journal of Crystal Growth 196 (24), 424.

Haas, C., Drenth, J., 1998. The proteinwater phase diagram and the
growth of protein crystals from aqueous solution. Journal of Physical
Chemistry B 102 (21), 4226.
Haas, C., Drenth, J., Wilson, W.W., 1999. Relation between the solubility
of proteins in aqueous solutions and the second virial coefcient of
the solution. Journal of Physical Chemistry B 103 (14), 2808.
Hagen, M.H.J., Frenkel, D., 1994. Determination of phase diagrams for
the hard-core attractive Yukawa system. Journal of Chemical Physics
101 (5), 4093.
Hamabata, A., von Hippel, P.H., 1973. Model studies on effects of
neutral salts on conformational stability of biological macromolecules.
2. Effects of vicinal hydrophobic groups on specicity of binding of
ions to amide groups. Biochemistry 12 (7), 1264.
Haynes, C.A., Carson, J., Blanch, H.W., Prausnitz, J.M., 1991.
Electrostatic potentials and protein partitioning in aqueous 2-phase
systems. A.I.Ch.E. Journal 37 (9), 1401.
Haynes, C.A., Tamura, K., Korfer, H.R., Blanch, H.W., Prausnitz,
J.M., 1992. Thermodynamic properties of aqueous alpha-chymotrypsin
solutions from membrane osmometry measurements. Journal of
Physical Chemistry 96 (2), 905.
Haynes, C.A., Benitez, F.J., Blanch, H.W., Prausnitz, J.M., 1993.
Application of integral-equation theory to aqueous 2-phase partitioning
systems. A.I.Ch.E. Journal 39 (9), 1539.
Hermann, R.B., 1972. Theory of hydrophobic bonding. 2. Correlation
of hydrocarbon solubility in water with solvent cavity surface-area.
Journal of Physical Chemistry 76 (19), 2754.
Ho, J.G.S., Middelberg, A.P.J., 2003. The inuence of molecular variation
on protein interactions. Biotechnology and Bioengineering 84 (5), 611.
Ho, J.G.S., Middelberg, A.P.J., Ramage, P., Kocher, H.P., 2003. The
likelihood of aggregation during protein renaturation can be assessed
using the second virial coefcient. Protein Science 12 (4), 708.
Hofmeister, F., 1888. Zur lehre von der wirkung der salze. Archiv for
Experimentelle Pathologie und Pharmakologie 24, 247.
Horton, N., Lewis, M., 1992. Calculation of the free-energy of association
for protein complexes. Protein Science 1 (1), 169.
Hummer, G., Garde, S., Garca, A.E., Pohorille, A., Pratt, L.R., 1996. An
information theory model of hydrophobic interactions. Proceedings of
the National Academy of Science USA 93, 8951.
Hummer, G., Garde, S., Garca, A.E., Paulaitis, M.E., Pratt, L.R.,
1998. Hydrophobic effects on a molecular scale. Journal of Physical
Chemistry B 102 (51), 10469.
Ilett, S.M., Orrock, A., Poon, W.C.K., Pusey, P.N., 1995. Phase behavior
of a model colloidpolymer mixture. Physical Review E 51 (2), 1344.
Ishimoto, C., Tanaka, T., 1977. Critical behavior of a binary mixture of
protein and salt water. Physical Review Letters 39 (8), 474.
Israelachvili, J., 1992. Intermolecular and Surface Forces. Academic Press,
London.
Janin, J., Chothia, C., 1990. The structure of proteinprotein recognition
sites. Journal of Biological Chemistry 265 (27), 16027.
Jones, S., Thornton, J.M., 1996. Principles of proteinprotein interactions.
Proceedings of the National Academy of Sciences of the United States
of America 93 (1), 13.
Kern, N., Frenkel, D., 2003. Fluiduid coexistence in colloidal systems
with short-ranged strongly directional attraction. Journal of Chemical
Physics 118 (21), 9882.
Kita, Y., Arakawa, T., Lin, T.-Y., Timasheff, S.N., 1994. Contribution of
the surface free energy perturbation to proteinsolvent interactions.
Biochemistry 33 (50), 15178.
Kjellander, R., Marcelja, S., 1984. Correlation and image charge effects
in electric double layers. Chemical Physics Letters 112, 49.
Koo, E.H., Lansbury, P.T., Kelly, J.W., 1999. Amyloid diseases: abnormal
protein aggregation in neurodegeneration. Proceedings of the National
Academy of Sciences of the United States of America 96 (18), 9989.
Kunz, W., Lo Nostro, P., Ninham, B.W., 2004. The present state of affairs
with Hofmeister effects. Current Opinion in Colloid and Interface
Science 9 (12), 1.

Lee, J.C., Gekko, K., Timasheff, S.N., 1979. Measurements of preferential


solvent interactions by densimetric techniques. Methods in Enzymology
61 (Part H), 26.
Lekkerkerker, H.N.W., Poon, W.C.K., Pusey, P.N., Stroobants, A., Warren,
P.B., 1992. Phase behavior of colloid + polymer mixtures. Europhysics
Letters 20 (6), 559.
Lomakin, A., Asherie, N., Benedek, G.B., 1999. Aeolotopic interactions
of globular proteins. Proceedings of the National Academy of Sciences
of USA 96 (17), 9465.
Lomba, E., Almarza, N.G., 1994. Role of the interaction range in the
shaping of phase-diagrams in simple uidsthe hard-sphere Yukawa
uid as a case-study. Journal of Chemical Physics 100 (11), 83672.
Lum, K., Chandler, D., Weeks, J.D., 1999. Hydrophobicity at small and
large length scales. Journal of Physical Chemistry B 103 (22), 4570.
McMillan, W.G., Mayer, J.E., 1945. The statistical thermodynamics of
multicomponent systems. Journal of Chemical Physics 13, 276.
Melander, W., Horvath, C., 1977. Salt effects on hydrophobic interactions
in precipitation and chromatography of proteins: an interpretation of
the lyotropic series. Archives of Biochemistry and Biophysics 183 (1),
200.
Moon, Y.U., Curtis, R.A., Anderson, C.O., Blanch, H.W., Prausnitz,
J.M., 2000. Proteinprotein interactions in aqueous ammonium sulfate
solutions Lysozyme and bovine serum albumin (BSA). Journal of
Solution Chemistry 29 (8), 699.
Muschol, M., Rosenberger, F., 1995. Interactions in undersaturated and
supersaturated lysozyme solutions: static and dynamic light scattering
results. Journal of Chemical Physics 103 (24), 10424.
Muschol, M., Rosenberger, F., 1997. Liquidliquid phase separation
in supersaturated lysozyme solutions and assocd. precipitate
formation/crystallization. Journal of Chemical Physics 107 (6), 1953.
Nandi, P.K., Robinson, D.R., 1972. The effects of salts on the free
energies of nonpolar groups in model peptides. Journal of the American
Chemical Society 94 (4), 1308.
Neal, B.L., Asthagiri, D., Velev, O.D., Lenhoff, A.M., Kaler, E.W.,
1999. Why is the osmotic second virial coefcient related to protein
crystallization? Journal of Crystal Growth 196 (24), 377.
Oberholzer, M.R., Lenhoff, A.M., 1999. Protein adsorption isotherms
through colloidal energetics. Langmuir 15 (11), 3905.
Onsager, L., Samaras, N.T., 1934. The surface tension of DebyeHckel
electrolytes. Journal of Chemical Physics 2, 528.
Pan, X., Glatz, C.E., 2003. Solvent effects on the second virial coefcient
of subtilisin and solubility. Crystal Growth & Design 3 (2), 203.
Pande, A., Pande, J., Asherie, N., Lomakin, A., Ogun, O., King, J.,
Benedek, G.B., 2001. Crystal cataracts: human genetic cataract caused
by protein crystallization. Proceedings of the National Academy of
Sciences of the United States of America 98 (11), 6116.
Patey, G.N., 1980. Models for strongly polar liquidsthe inuence of
molecular polarizability. Journal of Chemical Physics 72, 5763.
Petsev, D.N., Thomas, B.R., Yau, S.T., Vekilov, P.G., 2000. Interactions
and aggregation of apoferritin molecules in solution: effects of added
electrolytes. Biophysical Journal 78 (4), 2060.
Piazza, R., Pierno, M., 2000. Protein interactions near crystallization:
a microscopic approach to the Hofmeister series. Journal of
PhysicsCondensed Matter 12 (8A), A443.
Pierotti, R.A., 1965. Aqueous solutions of nonpolar gases. Journal of
Physical Chemistry 69 (1), 281.
Poon, W.C.K., 1997. Crystallization of globular proteins. Physical Review
E 55 (3B), 3762.
Reiss, H., Frisch, H.L., Lebowitz, J.L., 1959. Statistical mechanics of
rigid spheres. Journal of Chemical Physics 31, 369.
Rieskautt, M.M., Ducruix, A.F., 1989. Relative effectiveness of various
ions on the solubility and crystal-growth of lysozyme. Journal of
Biological Chemistry 264 (2), 745.
Robinson, D.R., Jencks, W.P., 1965. Effect of concentrated salt solutions
on activity coefcient of acetyltetraglycine ethyl ester. Journal of the
American Chemical Society 87 (11), 2470.

Rosenbaum, D., Zamora, P.C., Zukoski, C.F., 1996. Phase behavior of


small attractive colloidal particles. Physical Review Letters 76 (1), 150.
Rosenbaum, D.F., Kulkarni, A., Ramakrishnan, S., Zukoski, C.F., 1999.
Protein interactions and phase behavior: sensitivity to the form of the
pair potential. Journal of Chemical Physics 111 (21), 9882.
Roth, C.M., Neal, B.L., Lenhoff, A.M., 1996. Van der Waals interactions
involving proteins. Biophysical Journal 70 (2), 977.
San Biagio, P.L., Palma, M.U., 1991. Spinodal lines and Flory-Huggins
free-energies for solutions of human hemoglobins HbS and HbA.
Biophysical Journal 60 (2), 508.
Schrier, E.E., Schrier, E.B., 1967. Salting-out behavior of amides and
its relation to denaturation of proteins by salts. Journal of Physical
Chemistry 71 (6), 1851.
Schurtenberger, P., Chamberlin, R.A., Thurston, G.M., Thomson, J.A.,
Benedek, G.B., 1989. Observation of critical phenomena in a
proteinwater solution. Physical Review Letters 63 (19), 2064.
Sear, R.P., 1999. Phase behavior of a simple model of globular proteins.
Journal of Chemical Physics 111 (10), 4800.
Stillinger, R.H., 1973. Structure in aqueous solutions of non-polar solutes
from the standpoint of scaled-particle theory. Journal of Solution
Chemistry 2, 141.
Talanquer, V., Oxtoby, D.W., 1998. Crystal nucleation in the presence of
a metastable critical point. Journal of Chemical Physics 109 (1), 223.
Tanford, C., 1969. Extension of theory of linked functions to incorporate
effects of protein hydration. Journal of Molecular Biology 39 (3), 539.
Taratuta, V.G., Holschbach, A., Thurston, G.M., Blankschtein, D.,
Benedek, G.B., 1990. Liquidliquid phase separation of aqueous
lysozyme solutions: effects of pH and salt identity. Journal of Physical
Chemistry 94 (5), 2140.
Tardieu, A., Le Verge, A., Malfois, M., Bonnete, F., Finet, S., Ries-Kautt,
M., Belloni, L., 1999. Proteins in solution: from X-ray scattering
intensities to interaction potentials. Journal of Crystal Growth 196
(24), 193.
Tavares, F.W., Bratko, D., Blanch, H.W., Prausnitz, J.M., 2004. Ionspecic effects in the colloidcolloid or proteinprotein potential of
mean force: role of salt-macroion van der Waals interactions. Journal
of Physical Chemistry B 108 (26), 9228.
ten Wolde, P.R., Frenkel, D., 1997. Enhancement of protein crystal
nucleation by critical density uctuations. Science (Washington, DC)
277 (5334), 1975.
Teske, C.A., Blanch, H.W., Prausnitz, J.M., 2004. Measurement
of lysozymelysozyme interactions with quantitative afnity
chromatography. Journal of Physical Chemistry B 108 (22),
74377444.
Tessier, P.M., Vandrey, S.D., Berger, B.W., Pazhianur, R., Sandler,
S.I., Lenhoff, A.M., 2002a. Self-interaction chromatography: a
novel screening method for rational protein crystallization. Acta
Crystallographica Section D-Biological Crystallography 58, 1531.

Tessier, P.M., Lenhoff, A.M., Sandler, S.I., 2002b. Rapid measurement


of protein osmotic second virial coefcients by self-interaction
chromatography. Biophysical Journal 82 (3), 16201631.
Timasheff, S.N., 1991. Solvent effects on protein stability. Current Opinion
in Structural Biology 2 (1), 35.
Timasheff, S.N., 1998. Control of protein stability and reactions by
weakly interacting cosolvents: the simplicity of the complicated.
Advances in Protein Chemistry, vol. 51, Linkage Thermodynamics of
Macromolecular Interactions. p. 355.
Timasheff, S.N., Arakawa, T., 1988. Mechanism of protein precipitation
and stabilization by co-solvents. Journal of Crystal Growth 90 (13),
39.
Torrie, G.M., Valleau, J.P., 1980. Electrical double-layers 1. Monte-Carlo
study of a uniformly charged surface. Journal of Chemical Physics 73
(11), 5807.
Torrie, G.M., Valleau, J.P., 1982. Electrical double-layers 4. Limitations
of the Gouy-Chapman theory. Journal of Physical Chemistry 86 (16),
3251.
Vailaya, A., Horvath, C., 1998. Retention in reversed-phase
chromatography: partition or adsorption? Journal of Chromatography
A 829 (12), 1.
Valleau, J.P., Ivkov, R., Torrie, G.M., 1991. Colloidal stability: the forces
between charged surfaces in an electrolyte. Journal of Chemical Physics
95 (1), 520.
Velev, O.D., Kaler, E.W., Lenhoff, A.M., 1998. Protein interactions in
solution characterized by light and neutron scattering: comparison of
lysozyme and chymotrypsinogen. Biophysical Journal 75 (6), 2682.
Verwey, E.J.W., Overbeek, J.T.G., 1948. Theory of the Stability of
Lyophobic Colloids. Elsevier, Amsterdam.
Vilker, V.L., Colton, C.K., Smith, K.A., 1981. The osmotic-pressure of
concentrated protein solutionseffect of concentration and pH in saline
solutions of bovine serum albumin. Journal of Colloid and Interface
Science 79 (2), 548.
Vliegenthart, G.A., Lekkerkerker, H.N.W., 2000. Predicting the gasliquid
critical point from the second virial coefcient. Journal of Chemical
Physics 112 (12), 5364.
Wang, H.W., BenNaim, A., 1997. Solvation and solubility of globular
proteins. Journal of Physical Chemistry B 101 (6), 1077.
Wu, J.Z., Bratko, D., Blanch, H.W., Prausnitz, J.M., 1999. Monte-Carlo
simulation for the potential of mean force between ionic colloids in
solutions of asymmetric salts. Journal of Chemical Physics 111 (15),
7084.
Wu, J.Z., Bratko, D., Blanch, H.W., Prausnitz, J.M., 2000. Interaction
between oppositely charged micelles or globular proteins. Physical
Review E 62 (4), 5273.
Wyman, J., 1964. Linked functions and reciprocal effects in
hemoglobina 2nd look. Advances in Protein Chemistry 19,
223.

Вам также может понравиться