Journal of Chromatography A, 1216 (2009) 86788686

Contents lists available at ScienceDirect

Journal of Chromatography A
journal homepage: www.elsevier.com/locate/chroma

Molecular modeling of Protein A afnity chromatography

Matteo Salvalaglio, Laura Zamolo, Valentina Busini, Davide Moscatelli, Carlo Cavallotti
Dipartimento di Chimica, Materiali ed Ingegneria Chimica G. Natta, Politecnico di Milano, Via Mancinelli 7, 20131 Milan, Italy

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

Article history: The properties of the complex between fragment B of Protein A and the Fc domain of IgG were investi-
Available online 18 April 2009 gated adopting molecular dynamics with the intent of providing useful insight that might be exploited to
design mimetic ligands with properties similar to those of Protein A. Simulations were performed both
Keywords: for the complex in solution and supported on an agarose surface, which was modeled as an entangled
mAbs structure constituted by two agarose double chains. The energetic analysis was performed by means of
Molecular dynamics
the molecular mechanics Poisson Boltzmann surface area (MM/PBSA), molecular mechanics generalized
IgG
Born surface area (MM/GBSA), and the linear interaction energy (LIE) approaches. An alanine scan was
Ligand
Chromatography
performed to determine the relative contribution of Protein A key amino acids to the complex interaction
Agarose energy. It was found that three amino acids play a dominant role: Gln 129, Phe 132 and Lys 154, though
also four other residues, Tyr 133, Leu 136, Glu 143 and Gln 151 contribute signicantly to the overall
binding energy. A successive molecular dynamics analysis of Protein A re-organization performed when
it is not in complex with IgG has however shown that Phe 132 and Tyr 133 interact among themselves
establishing a signicant interaction, which is disrupted upon formation of the complex with IgG
and thus reduces consistently their contribution to the proteinantibody bond. The effect that adsorbing
fragment B of Protein A on an agarose support has on the stability of the proteinantibody bond was
investigated using a minimal molecular model and compared to a similar study performed for a synthetic
ligand. It was found that the interaction with the surface does not hinder signicantly the capability
of Protein A to interact with IgG, while it is crucial for the synthetic ligand. These results indicate that
ligandsurface interactions should be considered in the design of new synthetic afnity ligands in order
to achieve results comparable to those of Protein A right from the ligand design stage.
2009 Elsevier B.V. All rights reserved.

1. Introduction
The pharmaceutical industry has recently demonstrated great
interest towards the use of biomacromolecules, such as proteins,
peptides, and oligonucleotides as therapeutic agents [1,2]. In this
framework, a role of primary importance is played by monoclonal
antibodies (mAbs). These molecules can be used both for diagnos-
tic purposes (due to their high specicity towards a large number
of antigens) and for therapeutic use (to treat several diseases). As
a result, a large-scale production (in the order of grams per litre) is
required to be able to satisfy the mAbs world demand. At present,
several mAbs (about 20) have been approved for sale by the US Food
and Drug Administration (FDA) and it is estimated that about 300
are under development [3,4]. In the last 3 years the value of the
mAb market has doubled, passing from US$ 10.3 109 in 2004 [3]
to 20.6 109 in 2006, and it keeps growing at such a rate that it has
been recently depicted as a new gold rush [5]. While the produc-
tion technologies of mAbs have been developed reaching results
Corresponding author.
E-mail address: carlo.cavallotti@polimi.it (C. Cavallotti).
that can satisfy the market requirements (enhancing the process
productivity to the gram per litre range), the huge potential of this
class of molecules remains mostly limited by the difculties related
to the scaling up of downstream processing technologies [5,6]. Due
to the complex structure of these immunoglobulins along with the
presence of a huge number of different compounds in the produc-
tion media, the purication is the crucial step of the whole process.
The most utilized purication methodologies are precipitation, ion-
exchange chromatography, and afnity chromatography [7]. This
last procedure is one of the most used as rst recovery step in the
downstream processing of mAbs [8] and requires an appropriate
ligand, able to bind mAbs with high yield and selectivity.
Before further discussing the afnity chromatography of mAbs,
it is necessary to have a molecular understanding of the nature of
the interaction between mAbs and afnity ligands. Since human
antibodies are classied into ve different families (named A, D,
E, G and M), for sake of simplicity in this discussion we will refer
to the G class (IgG), which is the most common. IgG and, more in
general, monoclonal antibodies are large globular proteins charac-
terized by a complex structure in which two variable domains (Fab)
are linked through disulde bonds with a constant region: the Fc
domain. The Fc domain is divided in two equal parts, which rela-

0021-9673/$ see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.chroma.2009.04.035
 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686
Fig. 1. Molecular structure of an antibody with highlighted variable and constant
domains.
tive motion is restricted both by several oxidized cysteinecysteine
bonds and by a carbohydrate region located in proximity of the con-
junction between Fab and Fc domain. The X-ray structure of IgG is
sketched in Fig. 1.
Purication by an afnity method can be performed following
two different approaches: the rst is based on the specicity of
antigen binding (in this case the purication process involves the
bond formation between the ligand and the Fab domain of the anti-
body), while the second targets the constant part of the antibody:
the Fc fragment. The purication of mAbs by means of selective
Fc ligands is the most promising methodology for the large-scale
production of therapeutic mAbs. In fact, this procedure does not
require to establish any interaction with the active part of IgG, the
Fab domain, which is thus left intact and can retain its properties.
The ligands most adopted to bind selectively IgG are Protein A and
Protein G, which are able to establish highly selective interactions
with the Fc domain of IgG in a region known as consensus binding
site (CBS) [9], which is located at the hinge between the CH2 and
CH3 domains of the Fc fragment (see Fig. 1).
Although these natural receptors possess all the features neces-
sary to be used in antibody renement systems, they have some
limitations for large-scale purication processes, which is nec-
essary for therapeutic purposes. In fact both Protein A and G
present high costs of manufacturing and exhibit several operation
problems, mainly due to contamination issues, involving both the
separation of ligands from the culture broth and the chromatogra-
phy steps [1,2,10,11]. For these reasons, new synthetic ligands [12]
and tailored peptides [9,1315] have been researched to decrease
the costs related to the separation process; such ligands need to
be stable to sterilization, inexpensive and non-toxic in order to
compete with natural receptors. Most of the afnity ligands avail-
able in commerce [1417] have been developed to mimic some key
chemical features of the binding sites of Protein A and G.
The development of an efcient synthetic ligand is however a
difcult task made complex by the huge number of parameters
that inuence its performances. In fact it is known that the bind-
ing process between ligands and mAbs is determined not only by
the ligand itself, but also by the nature of the support [18,19] and
by the spacer arm adopted to connect ligand and support [1921].
The experimental determination of the extent to which ligand,
support and spacer contribute to establish a binding interaction
with the mAb is extremely complex. An interesting alternative,
which became available in recent years, is to try to extract this
information from molecular simulations. Molecular models offer
in fact the possibility to investigate from an atomistic point of
view the nature of the interaction between proteins and ligands,
thus supplying information that can be difcultly accessed from an
8679
experimental standpoint. Though molecular modeling has not been
used extensively in the eld of afnity chromatography, mainly
because of the difculty of treating this complex system at a suf-
cient level of accuracy, recently several publications in which this
powerful tool has been used to address problems of interest for the
chromatography community have appeared. To cite some recent
examples, we remember the work of Xu and Lenhoff on protein
adsorption isotherms [22], the molecular dynamics (MD) study of
ion-exchange materials performed by Liapis and co-workers [23],
the density functional theory (DFT)/MD investigation of the chi-
ral stationary phase performed by Nita and Cann [24], and the
studies of peptide adsorption in hydrophobic chromatography of
Makrodimitris et al. [25,26]. Molecular modeling has also been
recently applied to investigate the interaction between antibod-
ies and ligands or peptides, either with the intent of determining
free binding energies [27], to develop rational design criteria for
mutants of antibodies [28,29], or to investigate the physicochem-
ical properties of proteinligand complexes [30,31]. Despite its
importance, only a few computational studies are available in the
literature about the properties of the complex between Protein A
or G and IgG [12,3234], and they have mostly been focused on
the elucidation of the structure of the complex [33,34], or on the
extraction of raw binding energy data based on scoring function to
be adopted for the design of new IgG ligands [12,32]. While several
experimental investigations based on X-ray data are available in the
literature [9,35,36], a detailed energetic investigation of the nature
of the binding site between Protein A and IgG is currently missing.
In this work we investigated the nature of the binding between
fragment B of Protein A and IgG, which we consider as a model mAb,
using molecular dynamic simulations with the aim of elucidating
the relative contribution of each Protein A amino acid to the for-
mation of the bond with the mAb. Successively, since in previous
works we have shown that the interaction of the support with the
ligand can affect signicantly the interaction with IgG, we stud-
ied the interaction between Protein A supported on a molecular
model of the agarose surface and IgG. Results of simulations were
then compared with those of the solvated Protein AIgG complex
to get an insight on how the presence of a surface inuences the
Protein AIgG binding process. Finally, we compared the structural
and energetic properties of the Protein AIgG complex with those
of the complex between a synthetic ligand supported on agarose
and IgG.
The molecular modeling results provide several useful insights
that can be exploited for the design of synthetic ligands. The
detailed energetic description of the binding moieties on fragment
B of Protein A can in fact highlight the chemical features needed to
synthesize an efcient mimetic ligand. The extension of the inves-
tigation to supported systems can offer also useful information
about the interplay between the support and the afnity ligand.
In fact, the supported afnity ligand accessibility (either natural
or synthetic) is strongly inuenced by the ligandsupport interac-
tions, as we have shown in our previous works. All this information,
obtained modeling the system at the molecular scale, can thus be
adopted for the development of a rational approach to the design of
synthetic ligand based purication systems for mAbs downstream
processing.
2. Molecular model and theoretical basis
2.1. Molecular models
Three different molecular models were considered to investi-
gate the nature of the interaction between Protein A and IgG. The
rst model is the Protein AIgG system in solution, the second is
the Protein AIgG system with Protein A adsorbed on an agarose
8680 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686
Fig. 2. Agarose molecular model used in the simulations of the supported systems.
support, and the third a model of IgG in complex with a synthetic
ligand, namely A2P, supported on an agarose substrate.
The A2P system is the same described in our previous publica-
tion [19] and is composed of four parts: the ligand, a spacer, the
agarose support, and the full Fc domain of IgG. Support, spacer and
ligand are covalently bonded and bind the protein through elec-
trostatic and van der Waals forces. The energetic and structural
parameters of the A2P complex here reported are the same dis-
cussed in our previous paper [19], to which we refer for the details
of the simulations. Here we report however some unpublished data
that are of interest for the direct comparison with the Protein AIgG
system.
The molecular model adopted to investigate the interaction
between IgG and Protein A is the X-ray structure determined by
Deisenhofer, deposited in the pdb with code 1FC2 [35]. The Deisen-
hofer structure consists of fragment B of Protein A, just about one
fth of the whole protein, and by half Fc domain of the antibody.
Clearly, this is only a small portion of the whole Protein AIgG
complex. However, despite its relatively limited size, this complex
is generally considered as a good model of the complex between
Protein A (or G) and IgG [9,35], as it comprehends all the IgG bind-
ing site and most of its surroundings. In order to reproduce the
steric constraint acting on half Fc domain when covalently bound
to its symmetric partner in the whole Fc macrostructure, we have
imposed a distance restraint on the two terminal residues of the
antibody. In this way we could avoid the relative bending of the
CH2 and CH3 domains. Moreover, since the separation of the mAbs
by means of afnity chromatography is carried out in a pH range
next to pH 7, we have considered the His residues not protonated
for all the simulations.
The model of the Protein AIgGsupport complex is composed
by three parts: the agarose support, fragment B of Protein A and the
(half) Fc domain of IgG. The structure of the Protein AIgG molecular
model is the same used to study the complex in solution. To study
the effect of the support, fragment B of Protein A has been covalently
bound to the agarose support by means of an ester bond between
a glucose monomer unit and the backbone of the peptidic ligand.
The molecular model used to represent the agarose support is
composed of two double helixes involved in an entangled structure,
as described in detail in [18]. This structure was obtained by means
of a 30 ns equilibration MD run starting from two parallel double
helixes. The initial coordinates for a single double helix structure
were taken from the Protein Data Bank (PDB ID 1AGA [37]) and the
MD run was performed using an explicit solvation shell of TIP3P [38]
water molecules. After 15 ns the system converged to the structure
reported in Fig. 2, which was maintained for the successive 15 ns.
The adopted molecular model of Protein A supported on an
agarose surface should be regarded as an extremely simplied
model of the real agaroseProtein A system because of the limited
size of agarose and since we considered only a fragment of Protein A.
However, the purpose of this simulation is to investigate whether
the adsorption of Protein A on an agarose support can hinder its
capability to interact with IgG. In fact in our previous works [18,19]
we found that this was the case for the considered synthetic ligand.
It is thus interesting to check whether a complex peptide with a
well dened secondary structure can be adsorbed on a surface and
contemporarily maintain its capability to interact with an antibody
dispersed in solution or if its activity is somehow impeded by the
interaction with the surface. From this viewpoint, the molecular
model here adopted should be sufcient to give an answer on this
important point, since the single fragment B of Protein A will be
more inuenced by the interaction with the agarose surface then
the whole Protein A.
2.2. Simulation approach
All the simulations were performed in explicit water in a cubic
box using a non-bonded cutoff of 10 . Periodic boundary condi-
tions (PBC) were applied and long range electrostatic interactions
were evaluated using the particle mesh Ewald (PME) method [39].
MD simulations were performed following a ve-step protocol
consisting of:
(i) 2000 cycles minimization to remove any possible unfavor-
able contact between solvent and solute; in this step the
complex is restrained with a harmonic potential of the form
k(x)2 , where x is the displacement and k the force constant
(k = 500 kcal mol1 2 );
(ii) 1500 cycles minimization without restraints;
(iii) simulated annealing of 20 ps at constant volume to raise the
temperature of the system from 0 to 300 K; in this phase, a weak
restraint is imposed on the complex (k = 10 kcal mol1 2 ) to
avoid wild structural uctuations;
(iv) 100 ps run at constant pressure (equilibration phase) to allow
the water density to relax;
(v) MD run for a standard period of 1018 ns at constant pressure
and temperature (300 K).
In all the simulation steps the temperature has been controlled
with a Langevin dynamics algorithm, with a collision frequency
 = 1 ps1 . The pressure in the equilibration phase is controlled by
means of a weak coupling Berendsen scheme. The SHAKE algorithm
was used to restrain all the covalent bonds involving hydrogen,
which allowed using a time step of 2.0 fs [40]. Structures to be used
for successive analysis were extracted every 500 fs.
Binding free energies between protein and ligands were deter-
mined using the linear interaction energy (LIE), the molecular
mechanics Poisson Boltzmann surface area (MM/PBSA), and the
molecular mechanics generalized Born surface area (MM/GBSA)
approaches. The fundamental hypothesis on which these methods
rest and their consequent accuracy is discussed in the following, as
it is relevant to the interpretation of the results.
As reported in the literature [41], the LIE approach is based
on the main assumption that the free energy of interaction of a
molecule with the surrounding environment is linearly correlated
both to electrostatic (ELE) and van der Waals (VdW) interactions.
This assumption is useful in order to evaluate the free energy differ-
ence between two different states, free and bound, of the solute. In
the free state the surrounding environment is constituted only by
water molecules, while in the bound state the environment com-
prehends also the receptor.
According to LIE theory, the free binding energy can then be
computed as
vdw  V vdw ) + (V ele  V ele )
G = (Vbound (1)
free bound free
 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686
Table 1
Optimal and values for the evaluation of G contributions as a function of charge
and of total number of hydroxyl groups [41].
Charge No. of OH groups
=
/ 0 0.5
0 0 0.43
0 1 0.37
0 2 0.33
where the VdW an ELE terms contained in Eq. (1) are the non-
bonded energies of the adopted force eld and and are linear
scoring parameters. Since we are dealing with a very complex
solute, it was necessary to explicitly consider its auto-interactions
in the calculation of Vele and VVdW . This was obtained decom-
posing the electrostatic and the van der Waals potentials as
Vse = Vse (L L) + Vse (L E) + Vse (E E)
where e stands for ELE or VdW, s for bound or free, L for ligand and
E for environment.
The ELE and the VdW terms of the force eld were computed by
means of sums over all the non-bonded pairs, which were divided
into subgroups.
A fundamental assumption of the adopted LIE model is that the
internal re-organization of the environment can be considered neg-
ligible: in other words we are assuming a lock and key behavior of
the system in which the non-bonded potential of interaction of the
atoms making part of the environment is constant. This assumption
can be expressed as
e e
Vbound (E E) Vfree (E E)
The introduction of Eq. (3) in Eq. (2) leads to the simplication
of the (E E) terms and thus to the following expression:
 TSgas contribution for the Protein AIgG complex is 28 kcal/mol.
V e  = Vbound
e e
 Vfree e
 = (Vbound e
(L L) + Vbound (L E)
e e e e
+Vbound (E E))(Vfree (L L)+Vfree (L E) + Vfree (E E))
 
= V e (L L) + V e (L E)
where the superscript e refers to electrostatic or van der Waals
interactions, L to ligand and E to environment. The Ve (L E) term
is directly related to the interactions that take place between lig-
and and environment during the binding process, while Ve (L L)
is determined by the reorganization of the ligand upon passing
from the bound to the free state. This auto-interaction term can
be neglected for small ligands such as A2P but can become impor-
tant for peptidic ligands such as Protein A or other complex ligands
[42]. The lock and key assumption was found reasonable for systems
solvated in water [43,44], while it does not hold inside a protein,
for example for the binding site of an enzyme. The consensus bind-
ing site of IgG is characterized by the capability to establish both
electrostatic and hydrophobic interactions, which might involve a
certain re-organization of the amino acids involved in the forma-
tion of the bond with the ligand. It is thus difcult to estimate a
priori whether Eq. (3) holds for this system and this aspect must be
considered as a source of uncertainty for the calculations reported
in the next section.
The and coefcients of Eq. (1) are dependent on the chemical
properties of the solute, and in particular on the number of OH
groups. A recent survey has proposed the values reported in Table 1
[41].
The and values reported in Table 1 have been determined
for small molecules with a semi-empirical approach, and thus can-
not be straightforwardly applied to Protein A, which is a large and
complex solute. Thus, in order to evaluate the order of magnitude
of G with a simple approach we have calculated an upper and a
8681
lower G limit using respectively the maximum and the minimum
values for the and coefcients reported in Table 1.
Binding free energies of the Fc domain of IgG1 with the frag-
ment B of Protein A were also determined using the MM/PBSA
0.181 and MM/GBSA approaches. These theoretical approaches, origi-
0.181 nally proposed by Kollman and co-workers [45], are based on the
0.181 assumption that binding free energies of a ligand L with a protein P
0.181
can be determined as the sum of different contributions as:
G = EMM  + GSol  TSGas (5)
where EMM is molecular energy change in the gas phase for the
reaction of dissociation of the complex, and is determined by the
sum of three contributions:
EMM  = EInt  + EEl  + EVdW  (6)
(2) which correspond to changes in internal energy (composed by bond
stretching, angular deformation and torsional energy), electrostatic
energy (expressed as Coulomb interaction between point charges),
and van der Waals energy (expressed through Lennard Jones poten-
tials). GSol is the difference between the solvation free energy
of the complex (GSol LP) and that of the ligand (GSol L) and
IgG(GSol P). Solvation free energies were implicitly determined
using either the PoissonBoltzmann approach or generalized Born
theory. Finally, TSgas is the entropy difference between complex,
ligand and protein and was here calculated only considering the
rotational and translation contributions, which were determined
through statistical thermodynamics making reference to the cor-
responding molecular partition functions. The vibrational entropy
contribution has thus been neglected, mainly because of the dif-
(3)
culty of estimating it accurately for these large proteins. This is
likely to cause an uncertainty of a few kcal/mol in the free bind-
ing energies estimated using these computational approaches. The
The energetic terms appearing in Eq. (6) were averaged over the
 structures that the ligandprotein complex can assume during the
MD simulations.
MD simulations were performed using the Amber 8 computa-
(4) tional suite [46], all structures reported in this work were produced
using either PyMol 1.3 [47] or VMD 1.8.2 [48]. The parm94 [49]
force eld was adopted to model proteins and ligands while the
glycam 04 [50] and TIP3P [38] force elds were used for agarose
and water, respectively. The interaction energies between solute
and environment were determined with the Anal program of the
Amber 8 computational suite.
3. Results and discussion
3.1. Modeling the Protein AIgG interaction
The interaction between fragment B of Protein A and the human
IgG1 Fc fragment was investigated with the aim of calculating the
free binding energy of the complex and to determine which are the
residues of Protein A that contribute most to the binding energy. A
better understanding of the nature of the bond between Protein A
and IgG can in fact help to guide the design of improved synthetic
ligands for mAbs. As mentioned, the starting structure for the MD
simulations was the crystal structure determined by Deisenhofer
[35] and deposited in the Protein Data Bank as 1FC2. The 1FC2 struc-
ture is composed by one of the twin CH3CH2 domains of IgG1 and
by fragment B of Protein A, which is the Fc binding domain of Protein
A, and is reported in Fig. 3.
The bond between fragment B of Protein A and IgG involves the
hinge region located between the CH2 and CH3 domains of IgG and
several amino acids of Protein A. As previously reported [9], the
region between the CH2 and CH3 domains of IgG is known to be
8682 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686

Fig. 4. Free binding energy of the IgGProtein A complex calculated using the
MM/PBSA MM/GBSA and the LIE approach ( = 0.181, = 0.5) and reported as a
function of simulation time.

Table 2
Free binding energies calculated using the
MM/PBSA, MM/GBSA and LIE approaches and
averaged over the whole simulation time span.

G [kcal/mol]
Fig. 3. Structure of the Protein AFc complex used as staring point for the simula- MM/PBSA 7.4
tions. MM/GBSA 5.6
LIE 9.2
Exp [12] 9.8
able to bind several proteins beyond Protein A, such as the domain
C2 of Protein G, the rheumatoid factor, and the neonatal Fc recep-
ods and also with the available experimental data. The calculated
tor. The physicalchemical properties of the CBS that make it able to
free binding energies averaged over the whole simulation time span
coordinate several proteins are: a capability to form strong hydro-
are reported in Table 2.
gen (H) bonds, a high degree of solvent accessibility, an exposed
Since the LIE approach is sensible to the and values adopted
hydrophobic surface, a high adaptability and mobility of the con-
in Eq. (1), we analyzed in detail the effect of their variation between
sensus region, and the capability to change the relative orientation
the maximum ( = 0.181, = 0.5) and minimum ( = 0.181, = 0.37)
of the CH2 and CH3 domains. What is most interesting is that the
values reported in Table 1, which provides thus a sort of uncertainty
Fc binding region can coordinate different residues of proteins that
bar. The value reported in Table 2 is the average between those two
greatly differ in their secondary structure. This suggests that, on the
limits.
basis of a deep understanding of the chemical and physical prop-
Electrostatic (Vele ) and van der Waals (VVdW ) contributions
erties of this region, it might be possible to devise a set of rules
to the interaction energy are reported separately in Fig. 5. Both the
to be used in the synthesis of molecules with improved binding
components of the total energy of interaction are of the same order
properties [9,14,15,51,52].
of magnitude, which means that the binding process is driven by
The conformational evolution of the Protein AIgG complex was
studied through a 18 ns MD simulation in explicit water. A similar
simulation was performed for Protein A in solution to determine its
conformational evolution when IgG is not present.
The free binding energy between protein and antibody was then
calculated as a function of simulation time using the MM/PBSA,
MM/GBSA, and LIE approaches with standard = 0.181 and = 0.5
parameters. The results are summarized in Fig. 4.
It can be noted that the free binding energies predicted adopt-
ing the three different methods differ signicantly in the rst 2 ns
of simulation, after which they converge to a constant value. Taking
as a reference the free energy proles determined employing the
MM/PBSA and MM/GBSA methods, it can be observed that the LIE
free energy shows signicant uctuations at the 10th and 16th ns.
While the initial incoherence between the MM-based methods and
the LIE approach can be ascribed to the equilibration of the solvent
molecules (that are explicitly considered by the LIE approach and
substituted with continuum models in the other cases), these uc-
tuations are most likely due to an instantaneous violation of the
lock and key hypotheses expressed with Eq. (3). Despite of this, the
LIE free binding energy averaged over the whole simulation time Fig. 5. Protein A VdW and electrostatic energy change contributions in the solvated
span is coherent with the averages obtained with the other meth- complex.
 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686
Fig. 6. Protein A binding surface: the carbon atoms of non-polar and polar residues
are reported in green and yellow, respectively. The surface is coloured on the basis
of the charge of the ligand, the zone depicted in red is negatively charged while the
zone in blue is positively charged. It can be noted that the central part of the ligand
surface is uncharged, but surrounded by polar and charged groups.
both polar and non-polar interactions established by specic moi-
eties present on the Protein A molecular surface with the consensus
binding site.
The structure of fragment B of Protein A with labelling of key
amino acids in sketched in Fig. 6. The amino acids that contribute
most to the bond are those located on the binding interface, which
is composed by two of the three alpha helixes that constitute frag-
ment B of Protein A, and include all residues between Gln 129 and
His 137 and between Glu 143 and Lys 154.
As reported in literature [9], Protein A and the Fc fragment of IgG,
are stabilized mostly by hydrophobic interactions. This observation
is in agreement with the nature of the CBS, which comprehends
a hydrophobic pocket on the surface of the protein. The residues
composing the IgG CBS are Ile 253, Ser 254, Met 252, Met 423, Tyr
326, His 435, Asn 434, His 433, Arg 255, and Glu 380. The charged
amino acids (Arg 255, Glu 380) are placed around the hydrophobic
knob formed by Ile 253 and Ser 254, and their presence can lead to
the establishment of some polar and hydrophilic interactions. The
structure of the binding interface of fragment B of Protein A, shown
in Fig. 6, presents all the characteristics needed to be a good comple-
mentary partner of the IgG consensus binding site. In particular the
binding interface has a central non-polar region, where are placed
the hydrophobic Phe 132 and Tyr 133 residues that are considered
in the literature [12] to be among those contributing mostly to the
bond, surrounded by polar groups that can interact with charged
and polar groups on the Fc surface.
To determine the amino acid relative contribution to the Protein
AIgG bond we performed an alanine scan analysis. The alanine
scanning computational technique consists in replacing, one by one,
all the amino acid residues suspected to play a role in the inter-
8683
Fig. 7. Variation of free binding energies calculated replacing key amino acids of
Protein A with alanine residues using the MM/GBSA protocol.
molecular bond with an Ala residue, which is known to interact only
very slightly with other amino acids, and in re-calculating the bind-
ing energy at the MM/GBSA level. The changes in binding energy
with respect to the reference case are reported in Fig. 7.
The results reported in Fig. 7 highlight that three amino acids
play a dominant role: Gln 129, Phe 132 and Lys 154, though also
four other residues, Tyr 133, Leu 136, Glu 143, and Ile 150 contribute
signicantly to the overall binding energy. The Protein AIgG bond
can thus be rationalized in terms of two main binding sites: the rst
is positioned in the CH2 domain and is characterized by hydropho-
bic interactions between Phe 132, Leu 136, Ile 150 (Protein A), and
the IgG hydrophobic knob constituted by Ile 253 and Ser 254 (IgG),
and by one electrostatic interaction between Lys 154 and Thr 256.
The second site is positioned of the CH3 domain and is dominated
by electrostatic interactions between Gln 129 and Tyr 133 (Protein
A) and His 433, Asn 434, and His 435 (IgG). The interaction estab-
lished by Phe 132 and Tyr 133 was taken as model by Li et al. [12]
to develop a well known family of Protein A mimetic molecules.
However, Phe 132 and Tyr 133 are two hydrophobic amino acids
that occupy neighbor positions on the Protein A main chain and
are thus likely to form strong intra-molecular interactions in the
absence of IgG. This was conrmed analyzing the conformational
evolution of Protein A in solution, which was performed starting
from the conformation assumed by Protein A when in complex
with IgG. The simulations clearly showed that it took less than one
ns to the two residues to reach a structure, depicted in Fig. 8b, in
which a interaction is established and then maintained for the
remaining simulation time. The protein re-organization energy is
not accounted in the alanine scanning analysis, which is thus likely
to signicantly overestimate the importance of the second binding
site. To have a quantitative estimate of this overestimation, we cal-
culated the Phe 132 and Tyr 133 interaction energy adopting the
LIE protocol for each amino acid, thus explicitly accounting for the
ligand re-organization energy. Interestingly, we found that the free
interaction energies predicted by the LIE approach for Phe 132 and
Tyr 133 are 0.39 and 0.21 kcal/mol, which makes their contribution
to the overall Protein AIgG interaction energy almost negligible.
This computational result thus seems to question the motivation
behind the choice made by Li et al. to adopt Phe 132 and Tyr 133
as model structures to build Protein A mimetics, a family of ligands
from which derives also the ligand considered in this work, A2P. This
computational prediction is however contradicted by the fact that
the designed ligands exhibited good IgG binding properties. The
solution to this apparent dilemma is that Li and co-workers adopted
a triazine scaffold to bind the two functional groups that mimic
the Tyr and Phe amino acids, which actually prevents their mutual
8684 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686
Fig. 8. (a) Phe 132 and Tyr 133 relative orientation in the crystal structure and (b)
after 1 ns of MD simulation of Protein A conformational evolution in water; (c) struc-
ture of the A2P synthetic ligand, which design was inspired to the Phe 132 and Tyr
133 Protein A amino acid residues.
interactions for steric reasons, thus recovering the self-interaction
energy that is lost, according to our calculations, in Protein A. This
is evidenced in Fig. 8c, where the A2P ligand structure is com-
pared to that of the Protein A residues to which it was inspired.
This appears thus as an example where the mimicry of a natural
proteinprotein interaction of low energy inspired the design of a
ligand which, thanks to an internal structure different form that of
a peptide, could have a relatively high interaction energy with the
target protein.
3.2. Support effect: comparison of a Protein A-based system with
supported synthetic ligand
Despite the growing economic relevance of the mAbs purica-
tion market, there is not yet a viable synthetic alternative to Protein
Fig. 9. Initial and stabilized (5 ns) conguration of the supported Protein
AIgGagarose complex.
A afnity purication. In the previous section we have investigated
in detail the properties of the bond between Protein A and IgG,
while in our previous publication [19] we have studied the interac-
tion between the A2P synthetic ligand, which structure is sketched
in Fig. 8c, and the same IgG fragment. One of the main ndings of
our previous work is that the interaction between A2P and mAb
is signicantly inuenced by the interaction of the ligand with
the material on which it is adsorbed. In order to investigate the
inuence of the support material on Protein A interaction with IgG
we have developed a minimal molecular model in which we have
attached fragment B of Protein A to the same agarose support model
adopted to study A2P. The intent is to determine whether part of the
success of Protein A is that it is able to retain its capability to inter-
act with IgG even when it is adsorbed on a support material. From
this standpoint, the minimal structural model adopted for Protein
A (we are considering only fragment B of Protein A, i.e. one fth
of the molecule) is likely to enhance any critical aspect related to
the establishment of an interaction with the support surface that
might rival that with IgG. Also, considering only a portion of Protein
A allows us to adopt a reduced molecular model for the support.
The simulations were started from an initial structure of the
IgGProtein Aagarose complex that was determined by attaching
in proximity of the consensus binding site the Fc fragment of IgG
to a relaxed structure of the Protein Asupport complex, previously
obtained through a 1 ns MD simulation of this system. Two snap-
shots of the initial and nal structure of the simulated complex are
reported in Fig. 9.
To compare the interaction of IgG and A2P with that of Protein
A we reported in Fig. 10 the time evolution of the change of van
der Waals (Vvdw ) and electrostatic interaction energies (Vele ) of
the ligand with the environment between a state in which it is sup-
ported on agarose and in complex with IgG and a state in which IgG
is not present. The interaction energy estimation thus involves two
different MD simulations for each complex and explicitly accounts
for the possibility of the ligand to form favourable interactions with
the support. The analysis of the electrostatic interaction energy data
evidences a markedly different behaviour of the A2P ligand with
respect to Protein A: while A2P starts from positive electrostatic
 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686
Fig. 10. Comparison of the evolution of electrostatic and van der Waals interaction
energies between Protein A and the synthetic A2P ligand supported on agarose and
half Fc domain of IgG.
interaction energies and rapidly reaches a steady situation in which
its electrostatic interaction with IgG is globally neutral, Protein A
electrostatic interaction with IgG is not only always negative, but
even slightly larger than that calculated in the previous section for
the complex in solution and reported as a straight line in Fig. 10a.
As expected from the results of previous simulations performed in
the absence of a support, both ligands establish negative van der
Waals interaction energies with IgG, though Protein A interaction
energies are also in this case signicantly larger than those calculate
for A2P and approach the average value calculated for its complex
with IgG in solution, reported also in this case as a straight line in
Fig. 10b. It is important to remark that the interaction energies for
the A2PIgG complex in water are larger, in absolute value, than
those reported in Fig. 10, which refer to the supported complex.
Thus, the low interaction energy of A2P with IgG can mostly be
ascribed to the electrostatic and van der Waals interaction energies
it establishes with the support, which are not balanced by those
formed when in complex with IgG. This is not the case for Protein
Fig. 11. Time evolution of the distances between the two agarose units that interact most with IgG and the centre of mass of the CH3 domain of IgG for the IgGProtein A and
IgGA2P complexes.
8685
A. In the case of the Protein AIgG complex in fact the interaction
with the support determines a slight decrease of the van der Waals
interaction energy with respect to that calculated for the complex
in solution, which is however counterbalanced by an increase of
the electrostatic interaction energy, thus indicating that Protein A
activity towards IgG is not signicantly inuenced by the interac-
tion with the support. A structural analysis of the Protein Aagarose
relative orientations conrms this conclusion. In fact the study of
the MD simulation results revealed that, in presence of IgG, Protein
A assumes a conformation in which one of its alpha helixes, the
same that in the solvated complex does not participate to the bond
with IgG, is interacting with the agarose surface while the other two
are free to interact with the antibody. This structural conformation
can be directly observed in Fig. 9b.
To further investigate the differences between the IgG complex
with the supported fragment B of Protein A and the supported
A2P ligand, we report in Fig. 11 the time evolution of the distances
between the two agarose units that establish the most signicant
interactions with IgG and the centre of mass of the CH3 domain
of IgG. It is interesting to observe that there is no signicant dif-
ference between the two systems, thus indicating that IgG will
move similarly towards the surface independently from the ligand
adopted, until it reaches an average distance characterized by the
establishment of interactions between some antibody amino acids
and agarose, that is maintained for the rest of the simulation.
4. Conclusions
The properties of the complex between fragment B of Protein
A and the Fc domain of IgG were investigated both when it is
immersed in solution and when the ligand is adsorbed on a sup-
port material adopting molecular dynamics simulations. The results
were compared with those of a previous computational study of the
interaction between a synthetic ligand with the Fc domain of IgG.
The results here presented are based on several assumptions,
which had to be taken in order to make the investigated molec-
ular system tractable from a computational standpoint. The most
important approximations are: the consideration of only a portion
of the involved proteins, and in particular of a single fragment of
Protein A and of the Fc domain of IgG; the limited time frame of the
simulations, which allowed to sample only a portion of the poten-
8686 M. Salvalaglio et al. / J. Chromatogr. A 1216 (2009) 86788686

tial energy surface; the limited size of the agarose surface model; [8] S. Hober, K. Nord, M. Linhult, J. Chromatogr. B 848 (2007) 40.
the description of the solvent as a pure water environment, thus [9] W.L. DeLano, M.H. Ultsch, A.M. de Vos, J.A. Wells, Science 287 (2000) 1279.
[10] A.C.A. Roque, C.S.O. Silva, M.A. Taipa, J. Chromatogr. A 1160 (2007) 44.
neglecting the effect of salt and other biomolecules that are usually [11] K. Huse, H.J. Bohme, G.H. Scholz, J. Biochem. Biophys. Methods 51 (2002)
present in real feedstocks and that are known to affect the separa- 217.
tion process. While such assumptions make the molecular model [12] R.X. Li, V. Dowd, D.J. Stewart, S.J. Burton, C.R. Lowe, Nat. Biotechnol. 16 (1998)
190.
here presented more close to an ideal system than to a faithful [13] H. Yang, P.V. Gurgel, R.G. Carbonell, J. Peptide Res. 66 (2006) 120.
description of a real chromatographic process, some conclusions [14] G. Fassina, M. Ruvo, G. Palombo, A. Verdoliva, M. Marino, J. Biochem. Biophys.
can be drawn that can help to interpret what already is known Methods 49 (2001) 481.
[15] G. Fassina, A. Verdoliva, M.R. Odierna, M. Ruvo, G. Cassini, J. Mol. Recognit. 9
or suspected for this type of interactions, as well as to shed new (1996) 564.
light on the properties of this important protein-protein complex. [16] A.R. Newcombe, C. Cresswell, S. Davies, K. Watson, G. Harris, K. ODonovan, R.
In particular we found that: Francis, J. Chromatogr. B 814 (2005) 209.
[17] K. Sproule, P. Morrill, J.C. Pearson, S.J. Burton, K.R. Hejnaes, H. Valore, S. Lud-
vigsen, C.R. Lowe, J. Chromatogr. B 740 (2000) 17.
(i) The interaction energy between Protein A and IgG is deter- [18] V. Busini, D. Moiani, D. Moscatelli, L. Zamolo, C. Cavallotti, J. Phys. Chem. B 110
mined both by electrostatic and van der Waals interactions. (2006) 23564.
(ii) The alanine scan analysis evidenced that the key residues [19] L. Zamolo, V. Busini, D. Moiani, D. Moscatelli, C. Cavallotti, Biotechnol. Prog. 24
(2008) 527.
involved in the proteinantibody bond are: Gln 129, Phe 132 [20] P. DePhillips, I. Lagerlund, J. Farenmark, A.M. Lenhoff, Anal. Chem. 76 (2004)
and Lys 154, followed by Tyr 133, Leu 136, Glu 143 and Gln 151. 5816.
(iii) The LIE energetic analysis suggests that the role played by Phe [21] S.Y. Suen, S.Y. Lin, H.C. Chiu, Ind. Eng. Chem. Res. 39 (2000) 478.
[22] X.K. Xu, A.M. Lenhoff, J. Phys. Chem. B 112 (2008) 1028.
132 and Tyr 133 might be overemphasized by alanine scanning [23] E. Riccardi, J.C. Wang, A.I. Liapis, J. Phys. Chem. B 112 (2008) 7478.
calculations, since their interaction energy with the antibody [24] S. Nita, N.M. Cann, J. Phys. Chem. B 112 (2008) 13022.
is counterbalanced by the strong intra-molecular interaction [25] K. Makrodimitris, E.J. Fernandez, T.B. Woolf, J.P. OConnell, Anal. Chem. 77 (2005)
1243.
they develop among themselves in solution. The A2P syn- [26] K. Makrodimitris, D.L. Masica, E.T. Kim, J.J. Gray, J. Am. Chem. Soc. 129 (2007)
thetic ligand, which design was inspired from these two amino 13713.
acids, maintains intact its capability to interact because the [27] J.G. Chen, R.X. Wang, M. Taussig, K.N. Houk, J. Org. Chem. 66 (2001) 3021.
[28] Y.M. Pan, D.Q. Gao, C.G. Zhan, J. Am. Chem. Soc. 130 (2008) 5140.
self-interaction of the two functional groups of which it is com- [29] R. Barderas, J. Desmet, P. Timmerman, R. Meloen, J.I. Casal, Proc. Natl. Acad. Sci.
posed, which mimic Phe 132 and Tyr 133, is sterically hindered. U.S.A. 105 (2008) 9029.
(iv) Differently from what found for the complex between IgG [30] C. Arcangeli, C. Cantale, P. Galef, G. Gianese, R. Paparcone, V. Rosato, J. Biomol.
Struct. Dyn. 26 (2008) 35.
and the A2P synthetic ligand, the adsorption of fragment B
[31] J. Morll, J. Neumann, K. Blank, U. Steinbach, E.M. Puchner, K.E. Gottschalk, H.E.
of Protein A on an agarose support material does not limit its Gaub, J. Mol. Biol. 381 (2008) 1253.
capability to interact with the antibody. [32] J. Sengupta, P. Sinha, C. Mukhopadhyay, P.K. Ray, Biochem. Biophys. Res. Com-
mun. 256 (1999) 6.
[33] L. Jendeberg, M. Tashiro, R. Tejero, B.A. Lyons, M. Uhlen, G.T. Montelione, B.
We believe that the results here reported can be exploited to Nilsson, Biochemistry 35 (1996) 22.
develop an efcient strategy for the development of Protein A [34] K. Kato, L.Y. Lian, I.L. Barsukov, J.P. Derrick, H.H. Kim, R. Tanaka, A. Yoshino, M.
mimetic ligands, which, in our opinion, would be most success- Shiraishi, I. Shimada, Y. Arata, G.C.K. Roberts, Structure 3 (1995) 79.
[35] J. Deisenhofer, Biochemistry 20 (1981) 2361.
ful if it combined systematic ligand development methods, such as [36] E.O. Saphire, P. Parren, R. Pantophlet, M.B. Zwick, G.M. Morris, P.M. Rudd, R.A.
screening of combinatorial libraries, with computational tools sim- Dwek, R.L. Staneld, D.R. Burton, I.A. Wilson, Science 293 (2001) 1155.
ilar to those presented in this work. Our results however suggest [37] S. Arnott, A. Fulmer, W.E. Scott, I.C.M. Dea, R. Moorhouse, D.A. Rees, J. Mol. Biol.
90 (1974) 269.
that it is imperative, in the search for a new ligand, that support [38] W.L. Jorgensen, J. Chandraskhar, J.D. Madura, R.W. Impey, M.L. Klein, J. Phys.
and ligand are optimized together right from the initial in silico or Chem. 79 (1983) 926.
in vitro screening of the combinatorial library. [39] D.A. Case, T.E. Cheatham, T. Darden, H. Gohlke, R. Luo, K.M. Merz, A. Onufriev,
C. Simmerling, B. Wang, R.J. Woods, J. Comput. Chem. 26 (2005) 1668.
[40] J.P. Ryckaert, G. Ciccotti, H.J.C. Berendsen, J. Comput. Phys. 23 (1977) 320.
Acknowledgements [41] T. Hansson, J. Marelius, J. Aqvist, J. Comput. Aided Mol. Des. 12 (1998) 27.
[42] D. Moiani, C. Cavallotti, A. Famulari, C. Schmuck, Chem. Eur. J. 14 (2008) 5207.
[43] M. Almlof, J. Aqvist, A.O. Smalas, B.O. Brandsdal, Biophys. J. 90 (2006) 433.
This work has been performed as part of the Advanced Inter-
[44] M. Almlof, B.O. Brandsdal, J. Aqvist, J. Comput. Chem. 25 (2004) 1242.
active Materials by Design (AIMs) project, supported by the Sixth [45] J. Wang, P. Morin, W. Wang, P.A. Kollman, J. Am. Chem. Soc. 123 (2001) 5221.
Research Framework Program of the European Union (NMP3-CT- [46] D.A. Case, T.A. Darden, T.E. Cheatham III, C.L. Simmerling, J. Wang, R.E. Duke,
R. Luo, K.M. Merz, B. Wang, D.A. Pearlman, M. Crowley, S. Brozell, V. Tsui, H.
2004-500160).
Gohlke, J. Mongan, V. Hornak, G. Cui, P. Beroza, C. Schafmeister, J.W. Caldwell,
W.S. Ross, P.A. Kollman, AMBER 8, University of California, San Francisco, CA,
References 2004.
[47] W.L. DeLano, The PyMOL Molecular Graphics System, DeLano Scientic, San
Carlos, CA, 2002.
[1] K. Keller, T. Friedmann, A. Boxman, Trends Biotechnol. 19 (2001) 438.
[48] W. Humphrey, A. Dalke, K. Schulten, J. Mol. Graphics 14 (1996) 33.
[2] A.C.A. Roque, C.R. Lowe, M.A. Taipa, Biotechnol. Prog. 20 (2004) 639.
[49] W.D. Cornell, P. Cieplak, C.I. Bayly, I.R. Gould, K.M. Merz Jr., D.M. Ferguson, D.C.
[3] S.J. Kim, Y. Park, H.J. Hong, Mol. Cells 20 (2005) 17.
Spellmeyer, T. Fox, J.W. Caldwell, P.A. Kollman, J. Am. Chem. Soc. 117 (1995)
[4] D. Filpula, Biomol. Eng. 24 (2007) 201.
5179.
[5] K. Maggon, Curr. Med. Chem. 14 (2007) 1978.
[50] K.N. Kirschner, R.J. Woods, Proc. Natl. Acad. Sci. U.S.A. 98 (2001) 10541.
[6] A.A. Shukla, B. Hubbard, T. Tressel, S. Guhan, D. Low, J. Chromatogr. B 848 (2007)
[51] Y.D. Clonis, J. Chromatogr. A 1101 (2006) 1.
28.
[52] C.R. Lowe, Curr. Opin. Chem. Biol. 5 (2001) 248.
[7] D. Low, R. OLeary, N.S. Pujar, J. Chromatogr. B 848 (2007) 48.