proteins
STRUCTURE O FUNCTION O BIOINFORMATICS
On the diversity of physicochemical
environments experienced by identical ligands
in binding pockets of unrelated proteins
Abdullah Kahraman,1,2* Richard J. Morris,3 Roman A. Laskowski,1 Angelo D. Favia,4
and Janet M. Thornton1
1 European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, Cambridgeshire CB10 1SD, United Kingdom
2 Swiss Federal Institute of Technology Zurich, Institute of Molecular Systems Biology, Wolfgang-Pauli Strasse 16,
8106 Zurich, Switzerland
3 John Innes Centre, Computational and System Biology, Norwich Research Park, Norwich NR7 7UH, United Kingdom
4 Istituto Italiano di Tecnologia (IIT), Drug Discovery and Development, Via Morego 30, 16163 Genoa, Italy
ABSTRACT
Most function prediction methods that identify cognate
ligands from binding site analyses work on the assumption of
molecular complementarity. These approaches build on the
conjectured complementarity of geometrical and physico-
chemical properties between ligands and binding sites so that
similar binding sites will bind similar ligands. We found that
this assumption does not generally hold for protein–ligand
interactions and observed that it is not the chemical composi-
tion of ligand molecules that dictates the complementarity
between protein and ligand molecules, but that the ligand’s
share within the functional mechanism of a protein deter-
mines the degree of complementarity. Here, we present for a
set of cognate ligands a descriptive analysis and comparison of
the physicochemical properties that each ligand experiences in
various nonhomologous binding pockets. The comparisons in
each ligand set reveal large variations in their experienced
physicochemical properties, suggesting that the same ligand
can bind to distinct physicochemical environments. In some
protein ligand complexes, the variation was found to correlate
with the electrochemical characteristic of ligand molecules,
whereas in others it was disclosed as a prerequisite for the bio-
chemical function of the protein. To achieve binding, proteins
were observed to engage in subtle balancing acts between elec-
trostatic and hydrophobic interactions to generate stabilizing
free energies of binding. For the presented analysis, a new
method for scoring hydrophobicity from molecular environ-
ments was developed showing high correlations with experi-
mental determined desolvation energies. The presented results
highlight the complexities of molecular recognition and
underline the challenges of computational structural biology
in developing methods to detect these important subtleties.
Proteins 2010; 00:000–000.
V
C 2009 Wiley-Liss, Inc.
Key words: molecular recognition; protein–ligand interac-
tion; electrostatics; hydrophobicity; noncomplementarity;
cognate ligand; protein function; redox potential.
V
C 2009 WILEY-LISS, INC.
INTRODUCTION
It is generally assumed that the molecular recognition
between protein receptors and ligand molecules requires
in addition to geometrical complementarity, a high
degree of physicochemical complementarity between both
binding partners.1 Shape complementarity in particular
plays a crucial role in the recognition process as it per-
mits both binding partners to approach each other and
form short-range noncovalent bonds.2,3 However, several
studies have reported that the geometrical complementar-
ity is often far from perfect.4–6 These studies revealed
that ligands sometimes are only partially enclosed by
their binding sites with the rest of the ligand exposed to
the solvent. Furthermore, even within the enclosed
region, contacts between protein and ligand often occur
at relatively few points, involving strong hydrogen bonds
or van der Waals interactions7,8 leaving unoccupied
space like a ‘‘buffer zone’’ around the remainder of the
ligand. The buffer zone is partially filled by water mole-
cules enhancing the complementarity of the ‘‘naked’’
binding pocket. It has also been suggested that the buffer
zones provide space for motions of protein, ligand, or
water molecules, thus avoiding the complete loss of the
ligand’s and binding site residue’s entropy.9
Physicochemical complementarity also contributes to
molecular recognition, in particular, the complementarity
Additional Supporting Information may be found in the online version of this article.
This work was performed at the European Bioinformatics Institute, Hinxton, UK.
Grant sponsor: BioSapiens Network of Excellence, through the European Commis-
sion within its FP6 Programme, under the thematic area ‘‘Life Sciences, Genomics
and Biotechnology for Health’’; Grant number: LHSG-CT-2003-503265; Grant
sponsor: EMBL (Predoctoral Fellowship).
*Correspondence to: Abdullah Kahraman, ETHZ, IMSB, HPT C75, Wolfgang-Pauli
Strasse 16, 8093 Zurich, Switzerland. E-mail: abdullah@imsb.biol.ethz.ch
Received 29 September 2008; Revised 21 September 2009; Accepted 22 September
2009
Published online in Wiley InterScience (www.interscience.wiley.com).
DOI: 10.1002/prot.22633
PROTEINS 1
 A. Kahraman et al.
derived from long-range electrostatic interactions is
believed to be the driving force for molecular interac-
tions. For example, proteins binding an adenine or gua-
nine were found to discriminate between both molecular
moieties on the basis of electrostatics.10 Copper zinc
superoxide dismutases, a protein family known to have
the highest reaction rate among enzymes, attract posi-
tively charged metal ions into their binding sites through
a highly negative electrostatic field,11 and DNA-binding
proteins attract negatively charged DNA molecules
through positively charged patches on their molecular
surface.12 Different studies have been performed to
quantify the electrostatic complementarity between bind-
ing partners. In particular, early works by Nakamura
et al.,13–15 Chau and Dean,16–18 and Naray-Szabo and
Nagy19,20 analyzed the electrostatic lock-and-key model
of single ligand- and protein-binding sites by mapping
the molecular electrostatic potential (MEP) of both bind-
ing partners on a set of reference points on their surfa-
ces. The MEPs on the reference points were eventually
compared following the host-on-guest and guest-on-
guest model. These studies came to the conclusion that
long-range effects of partial charges are essential for cre-
ating complementary electrostatic environments between
binding site and ligand, but that partial charges on pro-
tein and ligand lack face-to-face complementarity.
The same conclusions were drawn for protein–protein
interfaces.21
It has been suggested that physicochemical interactions
in natural complexes lack exact complementarity,22 most
likely to avoid irreversible ligand binding. Ledvina et al.
showed that some phosphate receptors, sulfate-binding
proteins, flavodoxin structures, and DNase proteins exert
an intense negative electrostatic potential (ESP) at their
binding sites despite binding a highly negative ligand.23
The proteins stabilize the anion charges by van der Waals
interactions and an extensive local hydrogen-bonding net-
work comprising main-chain NH groups and hydroxyl
side chains. The question remains open whether in their
evolutionary past these enzymes were binding ligands with
complementary ESPs. Similarly, Herschlag and coworkers’
study of the electrostatic and geometric complementarity
in the oxyanion hole of ketosteroid isomerase concluded
that electrostatic complementarily makes only a modest
contribution to the enzyme’s catalytic mechanism, which
is primarily driven by geometrical complementarity.24
Nakamura et al. listed four reasons for the absence of elec-
trostatic complementary in some protein–ligand com-
plexes: (1) the ligand interacts mainly with the solvent, (2)
hydrophobic interactions are strong, (3) dissociation of
the ionizable protein residues was incorrectly assigned,
and (4) additional ionic ligands were affecting the experi-
enced electrostatic energy of the ligand.13
To gain a wider perspective on physicochemical com-
plementarily, we have analyzed the physicochemical char-
acteristics of 100 protein-binding sites that bind either
2 PROTEINS
AMP, ATP, FAD, FMN, glucose, heme, NAD, phosphate,
or steroids (estradiol and dehydroepiandrosterone).6
Countless applications have been developed in the past
for analysing active and binding sites, mainly in the field
of rational drug design. Most of the programs such as
LigBuilder,25 MCSS,26 PocketFinder,5 and Q-Site-
Finder27 are derivatives of the program GRID.28 GRID,
developed by Goodford, analyses the physicochemical
properties of protein-binding sites by computing the
interaction energy between various spherical probes and
the atoms of the protein using a standard potential
energy function. Other programs provide different infor-
mation about a given binding site. For example,
GRASP,29 although primarily a powerful protein struc-
ture visualization software, has a built-in Poisson-Boltz-
mann Solver, which allows the calculation and visualiza-
tion of the ESP in and around the protein-binding site.
The program HINT30 calculates the hydrophobicity of a
molecule using experimental octanol/water partition
coefficients and constructs a hydropathy field or comple-
mentarity map for a protein binding site. The CASTp31
database is a repository of protein clefts and voids and
provides area and volume measurements for each cavity.
Our own laboratory has developed CleftXplorer, which
uses spherical harmonic expansion coefficients to analyti-
cally describe the shape and size of binding sites and
ligands molecules.6,32–34 To obtain a more complete
picture also of the physicochemical properties within
binding pockets, the shape descriptor in CleftXplorer was
extended by various physicochemical descriptors that
characterize electrostatic charged–charged interactions,
hydrophobic interactions, hydrogen bonds, and van der
Waals interactions (see Methods).
Despite differences in their algorithms, all methodolo-
gies that aim to identify a ligand that binds to a given
protein binding site or, indeed, identify what the cognate
ligand might be assume that ligands and binding sites ex-
hibit geometric and physicochemical complementarity. If
this assumption is correct, binding sites that bind the
same cognate ligand should share very similar properties.
Most studies so far have investigated the physicochemical
properties of individual protein–ligand complexes and
have tried to highlight the complementarity between
those (see references earlier and Refs. 35–37). Our study
sets itself apart by a detailed investigation and compari-
son of the physicochemical properties in sets of nonho-
mologous binding sites that all bind the same ligand,
thereby revealing a surprising diversity in the protein
environments that a single ligand is able to bind.
METHODS
Data set
The analysis in this manuscript was performed on the
same data set as published in Kahraman et al.6 The pro-
 Physicochemical Variation in Binding Pockets
tein structures in this data set were selected from the
protein quaternary structures (PQS)38 database, which
contains the supposed biological unit of each crystallo-
graphically determined entry in the Protein Data Bank39
(PDB). Only structures containing a cognate ligand were
considered, that is, where the bound ligand is known to
be involved in the function of the protein as given by EC
number40 or by the protein’s corresponding UniProt
entry.41 These structures were filtered such that, for any
specific ligand type, the proteins binding the ligand were
nonhomologous, as defined by the homology (H) level
in the CATH protein structure classification database.42
The final selection stage required that for each ligand
type at least five nonhomologous protein–ligand struc-
tures remained. The net result of this selection procedure
was nine ligand types—AMP, ATP, FAD, FMN, glucose,
heme, NAD, phosphate, and steroid molecules (estradiol
and dehydroepiandrosterone) and 100 protein-binding
pockets containing one of these ligands.
Nine protein structures in the data set had loop
regions in the proximity of their binding sites for which,
because of insufficient experimental data, no atom coor-
dinates were available in the PDB file. These loops
regions probably exhibited high flexibility. Flexible loop
regions can play a major role in binding different ligand
molecules into a binding site, often enclosing the ligand
in a cage-like structure. The flexible loop regions in our
data set however remained or became flexible after the
ligand had bound to the protein structure, preventing
them from being a stable integral part of the protein-
binding site. As the contribution of these loop regions to
the ligand’s binding free energy varies with their flexibil-
ity, we assumed that their contribution can be neglected
and therefore have not attempted to include them in the
physicochemical property calculations.
Mapping protein physicochemical properties
on ligand molecules
The computer program CleftXplorer6 was modified so
that in addition to describing a 3D shape of a binding
pocket or a ligand molecule using spherical harmonics, it
could describe physicochemical properties mapped onto a
dot surface of that shape. The physicochemical properties
included ESPs, hydrophobicity scores, hydrogen bond do-
nor, acceptor, and van der Waals potential energies. These
were computed using standard software packages, supple-
mented by own code as described later. For the purpose of
this manuscript, the physicochemical properties were not
mapped on the surface of a shape, but rather onto the atom
center coordinates of each ligand in the data set, thereby
following a similar approach to the GRID.28 CleftXplorer’s
algorithm is simple and consists of three steps:
1. Remove all crystallographically observed water mole-
cules from the protein structure file. The computed
ESP and hydrophobicity score implicitly account for
surrounding water molecules.
2. For each ligand, identify and remove those polypep-
tide chains and neighboring chemical compounds
(NCCs) that lie within 9 Å of any ligand atom. NCCs
correspond to HET groups such as metals, cofactors,
and coenzymes. The cutoff distance of 9 Å is in ac-
cordance with interaction ranges of nonbonded inter-
actions.43 This speeds up the calculations while hav-
ing negligible effects10,44 on the calculated properties
(data not shown).
3. At the center of each ligand atom, calculate the value
of each physicochemical property, using the atoms of
the retained protein chains and NCCs.
Electrostatic potential
The ESPs were calculated by solving the nonlinear
Poisson-Boltzmann equation45 using APBS46 (version
1.1). A grid size of 1 Å was chosen and the finite differ-
ence discretization technique47 was applied. Calculations
were run at room temperature with a counterion concen-
tration of 0.1M. The continuum solvent model was allo-
cated a dielectric constant of es 5 78. Protein and ligand
atoms as well as NCCs were assigned a dielectric constant
of e 5 4. Hydrogen atoms were added to the protein
structure using the program REDUCE48 (version 3.13),
which optimizes the protein’s hydrogen bond network,
while flipping ASN or GLN side-chain amides where
appropriate. Protonation states of histidine residues were
approximated using PROPKA49 at a pH given for the
mother liquor in the PDB or in the primary literature of
the crystal structure. Partial charges and atom radii from
the PARSE50 parameter set were assigned to all protein
atoms using the automated procedure of PDB2PQR51
(version 1.4.0). The reported potentials from APBS were
converted from kT/e to kcal/mol/e by applying the con-
version factor of 0.592.
The ligands in our data set as well as all NCCs were
assigned Pauling’s van der Waals radii (which are also
used in the PARSE parameter set). All ligands in our
data set were left uncharged in the binding sites, to
account for the reduction of the screening effect by the
ligand molecule. Partial charges for the NCCs were
obtained from the Merck molecular force field
(MMFF94)52 as implemented in an in-house modified
version of the Chemistry Development Kit (CDK, version
2.0.2)53 for Java. The correctness of the MMFF94 partial
charges was checked against the MMFF94 validation suite
(http://www.ccl.net/cca/data/MMFF94/). For simplicity,
metal ions were assigned partial charges equal to their
formal charge following the approach of MMFF94. Iron
ions in heme molecules were treated as an integral part
of the molecule and assigned a partial charge of 12e
PROTEINS 3
 A. Kahraman et al.
irrespective of their oxidation state. Hydrogens were
added automatically to the ligands and NCCs at pH 7
using OpenBabel54 (version 2.1.1) and in the few cases
where the automatic assignment failed were added man-
ually using PyMOL (version 1.0).55 Phosphate molecules
were always assumed to be unprotonated. The spatial
position of the hydrogen atoms was optimized within the
binding site using REDUCE.
Scoring the hydrophobic environment
The calculation of CleftXplorer’s hydrophobicity scores
follows the approach of HINT30 and the hydrophobicity
potential of Fauchère et al.56 in relating the hydropho-
bicity score to octanol–water partition coefficients
(log P) values. CleftXplorer differs in that it uses atomic-
based rather than fragment-based log P values. This
makes CleftXplorer applicable to a wider spectrum of
molecules especially those for which HINT has no frag-
ment parameters. The atomic log P values were calcu-
lated using a modified version of the XlogP57 algorithm
in the CDK. Modifications were required to add log P
values for charged molecules, as by default, log P values
are defined only for neutral molecules. Some of the
ligands in our data set, as well as some of the NCCs,
contain charged moieties (e.g., PO4 groups), which
increase the solubility of the molecule in water. If the
molecule is treated as neutral its hydrophobicity will be
overestimated. To reduce the log P value for charged
atoms, a correction factor of 21.083 was introduced that
corresponds to half of the zwitterion correction factor
used in the XlogP algorithm for the hydrophobicity cal-
culation of amino acids. Metal ions were assigned an ar-
bitrary log P 5 23, assuming that metal ions in solution
minimally interfere with the water network. For protein
atoms log P values were calculated considering each
amino acid within the tripeptide Gly-X-Gly, where X is
the amino acid of interest. The values were calculated rel-
ative to glycine, following the method of calculating the
Hansch p-constant,58 by subtracting the mean atomic
log P value for glycine atoms from each of the atomic
log P values giving an atomic log Prel value. Based on log
Prel, the energy of transfer from water to organic solvent,
DGlog P, was inferred following the approach of Eisenberg
and McLachlan59 by applying the Boltzmann equation
from statistical thermodynamics:
DG log P
¼ log P rel 3 2:30 3 RT ¼ log P rel 3 1:36 kcal=mol;
where R is the Boltzmann factor and T is the room tem-
perature in Kelvin. The XlogP-based ‘‘hydrophobicity
energy’’ correlates with the experimental free energy of
transfer59 with R2 5 0.92. A similar approach to the
molecular lipophilic potential (MLP)60,61 was used to
map the ‘‘hydrophobicity energy’’ of the protein to the
atom center coordinates of the ligand molecules with a
4 PROTEINS
sigmoid potential function as suggested by Levitt and
Brylinski et al.62,63:


f distrel ¼ 1
0:5
 2 4 6 8

3 7 3 distrel
9 3 distrel þ 5 3 distrel
distrel ;
where the value of distrel ranges from 0 to 1 and corre-
sponds to the distance between a ligand’s atom center
and the center of the protein or NCC atom divided by
the cutoff distance of 9 Å.
To calculate the total hydrophobic environment score
(HES) at a ligand’s atom center, it was further necessary
to sum the product between the distance function
f(distrel) and the energy of transfer DGlog P over all n
neighboring atoms. The following equation expresses this
simple sum and gives the final equation for HES:
X
n

HES ¼
log P
f distrel i 3 DGi :
i¼1
Other physicochemical properties
CleftXplorer calculates van der Waals potential energies
using the ‘‘buffered 14-7’’ potential equation from the
MMFF94 force field.64 Hydrogen bond donor and
acceptor potential energies are calculated according to
the knowledge-based potential energies from Kortemme
et al.65 These orientation-dependent hydrogen-bonding
potentials were derived from geometric features found in
a large set of high-resolution protein structures. For
more details, see the cited publications.
Average properties and their variation
Once the physicochemical properties of all proteins
in a ligand set were mapped onto their ligand mole-
cules, the physicochemical property scores from differ-
ent binding sites were averaged and their standard
deviation computed. For this purpose, an atom-map-
ping table was created mapping corresponding atoms
between ligand A and ligand B, where both were from
the same ligand set. In cases in which the molecular
moiety had rotational symmetry, atoms at similar spa-
tial positions were associated to each other, after the
molecules were manually aligned on the neighboring
moieties. Atoms in the phosphate ligand set were
mapped randomly as a unique mapping due to sym-
metry was not possible. However, the small size of
phosphate molecules should generally result in similar
physicochemical property scores over the whole mole-
cule.
In addition to the standard deviation, the relative
number of sign changes in the property scores for each
atom was calculated as a further means to assess the vari-
 Physicochemical Variation in Binding Pockets

ation of the physicochemical properties in binding sites.
The sign change ratio (SCR) is given by


max Niþ ; Ni

SCRi ¼ 1

Nitotal
for an atom i in a ligand. Ni1 is the number of positive
score values, Ni2 the number of negative score values,
and Nitotal is the size of the ligand set. The functional
range of SCR is between 0 and 0.5, with 0 denoting no
variation (all corresponding atoms have equal sign) and
0.5 indicating highest variation (one half of correspond-
ing atoms are positive, the other half negative).
NCCs in the calculation of physicochemical properties
was crucial as without them most proteins in our data
set lacked the physicochemical properties required to
bind their cognate ligand.66 Of the 100 ligands in our
dataset, 67 had 144 NCCs, of which 63 were metals,
mostly coordinated to AMP, ATP, or phosphate mole-
cules. For the remaining 33 ligands (mostly HEM, NAD,
and steroid), no NCC was found in the PQS entry, but
their presence in vivo can, of course, not be excluded.
Physicochemical properties of proteins in
ligand-binding sites
Electrostatic potential on ligands

Figure 1 shows examples of the ESP experienced by all

RESULTS
All calculations shown in this section were performed
with all five physicochemical properties; that is, ESPs,
hydrophobicity scores, hydrogen bond donor, acceptor,
and van der Waals potential energies. Hydrogen bond
patterns were found to vary most within all ligand sets,
whereas van der Waals potential energies were found to
vary hardly. Nevertheless, all results and discussions in
the following were focused on the ESP and the HES,
because of their importance and discriminative power in
molecular recognition events.
For each ligand, ESP and HES were calculated in the
presence of any NCC (see Methods). The inclusion of
ATP, NAD, and heme ligands in the data set. For the ESP
of the remaining ligand sets, see Figure X1 in Supporting
Information.
Adenosine-50 -triphosphate
An adenosine-50 -triphosphate (ATP) molecule consists
of an aromatic adenine ring, a ribose sugar, and a nega-
tively charged triphosphate group. For recognition and
binding, the protein is expected to provide a positive
ESP for complementarity to the aromatic ring and in
particular the triphosphate tail. Figure 1(a) confirms this
expectation showing positive ESPs for all ATP-binding
sites. An inspection of the ATP-binding sites revealed

Figure 1
Electrostatic potential that (a) ATP, (b) NAD, and (c) heme molecules ‘‘experience’’ within their binding sites. The ligands are horizontally ordered
from top left to bottom right with increasing positive potential. The potential is colored from red to white to blue for values below 25 kcal/mol/e
to 0 to values above 5 kcal/mol/e. The PQS-Id of the protein structure in which the ligand was found is given below each ligand. Note that to
make the visual comparison easier, each type of molecule is represented by a single conformer and may not be the conformer in the given PQS file.
The representative conformers are 1e2q, 1mi3, and 1po5 and have yellow labels. NAD molecules bound to oxidoreductases are labeled green; those
bound to transferases are labeled orange.

PROTEINS 5
 A. Kahraman et al.

that the high positive potentials on the ATP molecules

21.94  3.93

2.87  1.19
Steroid
(on average 11.33 kcal/mol/e, see Table I) were caused
primarily by metal ions that are coordinated to one or
two phosphates at ATP’s triphosphate tail. Metal ions
were found coordinated in all ATP-binding sites except

17.14  14.64
in the DNA ligase (1a0i) and the biotin carboxylase

23.65  4.24
(1dv2). The function of those metals varies but most

PO4
often they act as catalytic stabilisers for transition states
of the enzyme–substrate complex67,68 or as charge
neutralizers between charged protein and phosphate
groups.69,70 In particular, the latter function is impor-
0.03  6.41

0.39  1.22
NAD

tant with respect to the molecular recognition of ATP as

metal ions can shield negative charges between the pro-
tein and ATP and lock the otherwise loose triphosphate
tail of ATP to the protein.71,72 The charge neutralizing
23.11  8.30

2.30  1.46

effect can increase ATP’s-binding affinity by several

HEM

orders of magnitude, as in the case of the bovine cAMP-

Average and Standard Deviation of the Experienced Electrostatic Potential and Hydrophobicity Environmental Scores for Different Ligand Sets

dependent protein kinase.73 The mouse homologue of

the same protein, 1rdq, in our data set shows, after the
exclusion of the divalent magnesium ion from the ESP
27.17  4.65

20.45  1.83

calculation, negative repulsive potentials toward the ATP

GLC

molecule. Only when the metal ion is included in the

calculation, the repulsive forces are neutralized and
become attractive (see Fig. 2). Similar neutralizing effects
of metal ions were observed in the PQS structures 1a49,
4.81  7.44

0.00  1.39

1b8a, 1e8x, 1o9t, and 1tid.

FMN

Not all ATP molecules require a metal ion to bind to

their receptor protein. As mentioned earlier, ATPs in the
binding sites of the DNA ligase (1a0i) and the biotin car-
5.27  5.93

0.94  1.21

boxylase (1dv2) were found without a metal ion or any

FAD

other NCCs. Although they are important for the cata-

The lowest and highest average score values for each property are colored purple and green, respectively.

lytic reaction of these enzymes, magnesium ions are not

essential for ATP binding. A detailed inspection of their
binding sites revealed that the proteins created attractive
11.33  9.13

21.02  2.00

positive ESPs [see Fig. 1(a)] toward the ATP phosphate

ATP

tails through phosphate-binding motifs. These motifs are

characterized by positive electrostatic fields originating
from positively polarized N-termini of a-helices and/or
positively charged lysine and arginine side chain and/or
1.53  5.87

20.65  2.53

ionic hydrogen bonds (O2

HN) between backbone NH

AMP

groups and phosphate oxygen ions.74

Nicotinamide adenine dinucleotide

1.96  6.75

0.70  1.53

Nicotinamide adenine dinucleotide (NAD) molecules

All

consist of three functional groups, namely the mainly ar-

omatic adenosine moiety, the negatively charged pyro-
phosphate group, and the aromatic nicotinamide ring
Hydrophobicity Environmental

that can carry a formal positive charge when oxidized to

Physicochemical property

NAD1.75 The ESP that a NAD molecule experiences in

Electrostatic Potential

binding sites depends on the enzymatic function of the

NAD molecule. In our data set, the two most prominent
functions were the redox function in oxidoreductases
Score (HES)
(kcal/mol/e)

(1hex, 1ib0, 1jq5, 1mew, 1mi3_1, 1o04_1, 1qax, 1t2d,

Table I

2npx) and the group transfer function in transferases

(1ej2, 1og3, 1s7g, 1tox_1, 2a5f). As a redox partner in

6 PROTEINS
 Physicochemical Variation in Binding Pockets

Figure 2
Electrostatic potential of the mouse cAMP-dependent protein kinase (PQS-Id 1rdq) as experienced by the cognate ligand ATP (a) without and (b)
with two coordinated magnesium ions (green-colored spheres). The potential is colored from red to white to blue for values below 25 kcal/mol/e
to 0 to values above 5 kcal/mol/e.

NAD-dependent dehydrogenases, a NAD1 molecule sup- card the nicotinamide moiety after cleavage for which re-
ports the oxidation of a substrate by accepting a hydride pulsive forces are advantageous, adenylyltransferases must
group (H2). As a substrate in ADP-ribosyltransferase reac- tightly bind the substrate nicotinamide ribonucleotide to
tions, NAD1 molecules are cleaved into an ADP-ribosyl fuse it to an ATP substrate molecule leading to the reac-
and a nicotinamide group, where the former is transferred tion product NAD1.
to an arginine group within the active site as part of a For the bacterial NADH peroxidase (2npx) repulsive
posttranslational modification and the latter is released potentials were found for the adenine moiety of NADH
into the solvent. For dehydrogenases, the nicotinamide (2npx is the only protein in our data set having bound
moiety should experience negative ESP due to the H2 NADH. All other NAD proteins have bound the oxidized
group transfer, whereas in ribosyl-transfer reactions the NAD1). The negative electrostatic field is created by six
positively charged nicotinamide moiety should experience negatively charged glutamate and aspartate amino acids
repulsive positive ESP. Figure 1(b) shows the ESP of all in close proximity. An examination of the residues
NAD molecules in our data set ordered by the total ESP around the nicotinamide moiety revealed that the repul-
experienced by the molecules from top left to bottom sive potentials were overridden by aromatic interactions
right. NAD molecules that are functioning as oxidation to a hydrophobic valine and an isoleucine side chain [see
partners were labeled green, whereas molecules acting as Fig. 4(a)]. Similar aromatic interactions were found for
substrates in transferase reactions were labeled orange.
All NAD molecules bound to oxidoreductases con-
firmed the expectation and showed an attractive negative
ESP at the majority of their nicotinamide moiety with
the exception of 1ib0, a rat cytochrome b5 reductase,
which experienced repulsive positive ESP. As the reduc-
tase name indicates 1ib0 distinguishes itself by using the
substrate NADH as a reducing agent. To increase the af-
finity toward NADH over NAD1, the enzyme was found
to use the observed positive ESP over the nicotinamide
moiety as a repulsive force against NAD1. The positive
ESP was created mainly at a conserved binding motif
‘‘CGPPPM’’76 with three prolines located at a positively
polarized N-terminus of an adjoining a-helix (see Fig. 3)
The transferase structure of the ADP-ribosyltransferase
1og3 and partially the structure of the NAD-dependent
deacetylase 1s7g exerted as expected repulsive electroposi-
tive potentials toward the nicotinamide moiety of NAD1. Figure 3
The structure of the nicotinamide mononucleotide NAD1-binding site of the rat cytochrome b5 reductase. The molecular
adenylyltransferase 1ej2 however exerted attractive nega- surface around the NAD molecule is shown transparent and colored
according to the electrostatic potential from red to white to blue for
tive ESP. Closer inspections showed that the attractive values below 25 kcal/mol/e to 0 to values above 5 kcal/mol/e. The
ESP was necessary for the function of 1ej2, namely to adjacent FAD molecule is shown in green, whereas the three proline
synthesize NAD1 molecules. Although transferases dis- residues of the conserved ‘‘CGPPPM’’ motif are shown in yellow.

PROTEINS 7
 A. Kahraman et al.

Figure 4
(a) The repulsive electrostatic potential of NADH peroxidase toward the adenine moiety of NADH is compensated via p–CH interactions between
the adenine moiety of NADH (varicolored) and Ile 180 (green colored) and Val 242 (green colored). (b) ADP-ribosyltransferase stabilizes the
binding of NAD1 via a p–p interaction with Phe160 (green colored), p–NH interaction with the backbone amide of Ser148 (green colored), and an
internal hydrogen bond (black dashed line) between the amide group and the phosphate group.

the nicotinamide group of 1og3 [see Fig. 4(b)], where
p–p stacking interaction to a phenylalanine from one side
and an aromatic-backbone-amide interaction from the
other side provide the necessary binding free energy for
the nicotinamide moiety of the NAD molecule. The im-
portance of aromatic interactions in our data set will be
described later in the text. Similar observations were made
on adenine groups that were found to bind protein-bind-
ing sites primarily with intermolecular hydrogen bonds to-
gether with p–p stacking and cation–p interactions.77,78
well known to contribute strongly to the binding energy
of heme molecules.79–81
Among all 16 heme-binding pockets analyzed, the bo-
vine endothelial nitric oxide synthase (eNOS) (PQS-Id:
1d0c) was the only enzyme that exposes the Fe of its
heme ligand to a repulsive positive ESP. The positive ESP
was mainly created by a loop region on the proximal site
of the heme molecule (see Fig. 5). The loop region
includes the Cys 186, which coordinates Fe, and three ar-

Heme type B

The heme molecule consists of an aromatic porphyrin

ring system with a central coordinated iron ion and two
negatively charged propionate groups sitting ‘‘on top’’ of the
porphyrin ring system. Arginine residues form ionic bonds
to the propionate groups and anchor the heme molecule to
the protein structure, whereas histidine and less frequently
cysteine residues coordinate the heme’s iron ion and affect
the electrochemical character of the iron metal ion (Fe).
Based on these characteristics, one would expect posi-
tive ESP around the propionate groups and negative
potential in the porphyrin ring system around the iron
ion. Visually, this prediction seemed true for the PQS
structures 1d7c, 1dk0, 1iqc_1, 1naz, and 1np4. However,
the remaining heme-binding sites, in particular, 1d0c,
1po5, 1qla, and 2cpo, show repulsive positive electrostatic
fields around the heme’s porphyrin group, causing the
heme molecules to have a diverse range of ESP values
with an average of 23.11 kcal/mol/e and a standard devi- Figure 5
Binding site of the bovine endothelial nitric oxide synthase (eNOS)
ation of 8.30 kcal/mol/e (see Table I). Heme molecules
with the blue-colored electrostatic potential isosurfaces at 10 kcal/mol/e.
are bound by more than 20 different protein folds79 The heme molecule is shown in stick representation and on its
leading to a large diversity of environments around the proximal site are depicted the loop region with Cys 186 (yellow color)
molecules. In-depth inspection of the protein environ- coordinated to the heme’s iron ion (dashed line) and Arg 189 (red
color). The loop main-chain amine groups are also shown in stick
ments around the heme molecules shows that, like representation. NCCs, here the inhibitor 7-NIBr (PDB-ligand name
NAD’s nicotinamide moiety, repulsive ESPs are overrid- INE) and the crystallization agent acetate (PDB-ligand name INE), are
den by aromatic interactions. Aromatic interactions are shown in green color.

8 PROTEINS
 Physicochemical Variation in Binding Pockets

higher ESP on the Fe ion generally induces a higher redox

potential for the heme molecule (see Fig. 6 and Table II).
The quinol:fumarate reductase (QFR) of Wolinella succi-
nogenes (PQS Id: 2bs2) binds two heme molecules, namely
a heme at the proximal site of the protein structure (heme
bP) and a heme at a distal site (heme bD). Both heme mol-
ecules have distinct redox potential. For many years it was
unclear, which of the heme molecules corresponded to the
low-potential (heme bL) and high-potential (heme bH)
heme molecule. With the computational simulation of the
electrochemical redox titration behavior of both heme mol-
ecules using electrostatic calculations, Haas and Lancaster97
were able to identify the low-potential heme as heme bD
and the high-potential heme as heme bP. Using the above
relationship between ESP and redox potential, we were
able to confirm Haas and Lancaster’s finding with an ESP
of 218.84 kcal/mol/e for heme bD and a much higher ESP
Figure 6
of 28.02 kcal/mol/e for heme bP.
Plot showing the correlation between heme redox potential and the
electrostatic potential that was measured on the heme’s iron ion (see
also Table II).
Remaining ligand sets

AMP-binding sites often generate positive ESPs toward

their ligand’s phosphate group (see Fig. X1 in Supporting
ginine residues of which Arg 189 is located right on top Information). Five of the proteins, namely 1amu, 1ct9_1,
of Fe. Main-chain NH groups in the loop region that are 1jp4, 1qb8, and 1tb7, had at least one metal ion bound
pointing inward into the loop were further strengthening within 9 Å proximity inducing a positive potential on
the positive ESP. Although the observed repulsive ESP the phosphate tail. The AMP molecule in asparagine syn-
seemed counterintuitive, it demonstrated once more that thetase (12as) compensated the repulsive global negative
the complementarity between proteins and ligands was ESPs through a local network of eight hydrogen bonds.
driven by the ligand’s role in the enzyme function and The flavin moiety in FAD as well as in FMN experienced
not by its chemical composition. eNOS must bind a pos- both negative and positive potentials in our data set. The
itively charged L-Arg substrate molecule to perform its phosphate group in both ligands was however always
function, namely creating nitric oxide. A heme Fe iron attracted by positive potentials. In the riboflavin kinase
ion in a reduced oxidation state Fe(II) is thereby advan-
tageous as it increases the binding affinity toward L-Arg
by 80-fold compared with the oxidized state Fe(III).82 Table II
Therefore, there is an interest from the enzyme to stabi- Comparison Between the Electrostatic Potential (ESP) on the Heme’s
Iron Ion and Its Redox Potential Against the Standard Hydrogen
lize the Fe(II). The positive ESP on the Fe of eNOS is Electrode (Em)
thereby advantageous, as it stabilizes the less positively
charged Fe(II) state. Furthermore, the positive electro- PQS-Id ESP (kcal/mol/e) Em (mV) Reference
static field increases the redox potential of Fe making the 1naz 242.81 ?
electron transfer from a distant FMN molecule thermo- 1dk0 233.13 2550 88
dynamically favorable.83 Finally, the field stabilizes the 1np4 224.67 2259 89
1qpa 223.65 2130 90
negatively charged Cys 186, which again increases the re- 1sox 221.45 168 91
dox potential of the heme molecule. 1pp9 220.03 210 92
In fact, the redox potential of the heme’s iron ion is 1qhu 19.20 ?
one of the most important electrochemical characteristics 1iqc_1 217.81 1450 85
1d7c 216.49 1169 93
of heme molecules. Heme redox potentials can vary 1eqg 210.69 250 94
between 2550 mV (in the hemophore HasA84) and 2bs2 28.02 29 95
1450 mV (in peroxidases85) versus the standard hydro- 1po5 27.30 ?
1ew0 25.82 ?
gen electrode. Correlating the redox potential to protein
1gwe 25.36 ?
electrostatics might be obvious; however, such a correla- 2cpo 20.61 2140 96
tion is not straightforward as the redox potential is influ- 1d0c 3.55 ?
enced by many different factors such as the type of the
See also Figure 6.
heme’s axial ligands, heme burial, and local ?, No redox potential was found in the literature for the protein at near crystalli-
charges.80,86,87 Nevertheless, our calculations show that zation condition.

PROTEINS 9
 A. Kahraman et al.
(1p4m), the pyrophosphate of the product ADP located
next to FMN caused the repulsive negative potential at
the phosphate group. The negative potential on the aro-
matic ring of the same ligand was compensated by aro-
matic interactions.
Glucose-binding sites were electrostatically most neg-
ative in our data set with an average ESP of 27.17
kcal/mol/e (see Table I). In contrast, phosphate-binding
sites were most positive with an average potential of
17.14 kcal/mol/e (see Table I). Metal ions were found
within 9 Å proximity in nine cases (1a6q, 1e9g, 1ew2,
1h6l, 1ho5_1, 1l7m_1, 1lby, 1qf5, and 1tco) generating
positive potentials around the PO4 molecules. An inter-
esting exception was the binding site of the bacterial
dethiobiotin synthetase (1dak). This particular binding
site apparently completely lacked the expected physico-
chemical properties to bind a phosphate molecule. Not
only it did exert repulsive negative potentials, but also
it lacked any compensating polar interactions and/or
hydrogen bonds toward its ligand. Also, van der Waals
interactions could be ruled out as the phosphate mole-
cule was solvent accessible on up to 70% of its molec-
ular surface. We could not find evidence that the mol-
ecule has bound at a crystal contact interface. Unfortu-
nately, the structure factors for 1dak were available at
neither the PDB nor the Electron Density Server,98
preventing further investigations of the electron density
of the phosphate molecule at its current location. The
50% occupancy value given for the entire phosphate
molecule in the current crystal structure of 1dak might
indicate an alternative location of the phosphate mole-
cule in the protein structure. And finally, as for heme
molecules, the molecular recognition of steroid mole-
cules is driven by hydrophobic and aromatic interac-
tions99 that override the observed repulsive ESP in ste-
roid-binding sites in our data set.
Nonelectrostatic interactions between protein and ligand
Our results show that binding events are not solely
directed by classical electrostatic interactions between
partially charged atoms, as there are many examples of
noncomplementary ESPs between protein and bound
ligand. Thus, other nonelectrostatic forces acting between
ligand and protein must contribute and result in negative
binding free energies and lead to the binding of the
ligand to its binding site. With the exception of glucose
and phosphate, all our ligands possess at least one aro-
matic ring complex, and most of the variation in terms
of ESP is observed at these ring systems. Ligand moieties
that are highly charged, like the phosphate groups, expe-
rience rather constant complementary electrostatic fields.
Close inspections of the binding site ligand complexes in
our data set showed that almost all aromatic ring groups
undergo various forms of aromatic interactions. Most of
these interactions were of a hydrophobic nature toward
10 PROTEINS
aliphatic hydrocarbons, such as valine, leucine, or isoleu-
cine side chains, forming p–CH interactions.100,101 The
second most often observed aromatic interactions were
p–p interactions between aromatic ring groups mostly
from phenylalanine and tyrosine residues.102 Other aro-
matic interaction types were p–cation interactions
between aromatic rings and positively charged side chains
of arginine and lysine103 and p–HN interactions with
backbone amide groups that act as a hydrogen bond do-
nor toward an aromatic ring.104 Also observed were p-
proline interactions.105 Aromatic interactions, in partic-
ular p–CH and p–p, are in general dominated by van der
Waals dispersion and hydrophobic interactions.106,107
Electrostatic interactions play only a secondary role and
only contribute where aromatic interactions involve pro-
tein hydroxyl or amine groups that have a high electro-
static dipole moment.108 Accurate dispersion energy cal-
culations require sophisticated and computationally
demanding quantum chemistry calculations106 that are
beyond the scope of this work. Instead, we analyzed the
hydrophobicity of the surrounding protein residues and
NCCs and assessed the contribution of hydrophobic
interactions in overriding electrostatic repulsion between
protein and ligand.
Figure 7 depicts the HES (see Methods) for the bind-
ing sites in Figure 1, colored from magenta to white to
green for 21 to 0 to 1 HES for polar, neutral, and
hydrophobic environments, respectively. From Figure 7,
it is clear that aromatic ring systems are generally located
in a hydrophobic environment. In particular, the varia-
tion observed for ESP in heme-binding sites does not
exist for HES. All heme molecules are located entirely in
hydrophobic-binding pockets whether facing attractive or
repulsive ESP. Similar dominant HES were observed for
steroid molecules and flavin moieties (see Fig. X2 in Sup-
porting Information). The adenine moieties of ATP and
NAD were usually found in a hydrophobic environment,
whereas their charged phosphate groups tend to face a
polar environment. The NAD molecules (PQS Ids 1ibo,
1hex, 1mew, 1og3, 1qax, and 2npx) that were found to
experience repulsive electrostatic forces at their adenine
and nicotinamide moieties simultaneously form attractive
aromatic-hydrophobic interactions.
Average and variation of physicochemical properties
Despite the various conformations of a ligand in dif-
ferent protein-binding sites, one can calculate the average
ESP and HES and its standard deviation for each ligand
atom in the protein-binding pocket. The average atomic
ESP and HES were calculated for each ligand set and
plotted in Figures 8 and 9. Next to the figures are
depicted ESP and HES generated by the ligands in isola-
tion. In contrast to many single cases (see in particular
NAD and heme in Fig. 1), the average values of the phys-
icochemical properties show complementary characteris-
 Physicochemical Variation in Binding Pockets

Figure 7
Hydrophobicity experienced by (a) ATP, (b) NAD, and (c) heme molecules within their binding pockets. The level of experienced hydrophobicity is
given by the hydrophobic environment scores (HES) (see Methods) and is colored from magenta to white to green for polar environments with less
than 21 HES to 0 to hydrophobic environments with values above 1 HES. The ligands are ordered as in Figure 1. The PQS-Id of the protein
structure in which the ligand was found is given below each ligand. The representative conformer for each set has a yellow label.

tics between protein and ligand for the majority of the
cases. From Figure 8, it becomes evident that usually
electrostatically positive protein potentials are neutralized
by electrostatically negative ligand potentials and vice
versa.
A closer look at the ESP in Figure 8(a) reveals average
positive ESPs for all adenine groups and all phosphate
groups, whether as separate entities or part of larger mol-
ecules. Note that despite the similarity between AMP and
ATP, the latter is usually bound in pockets with higher
positive potential because of the larger number of metal
ions coordinated to the phosphate tail of ATP. The highly
hydrophobic heme and steroid molecules, as well as glu-
cose and the nicotinamide moiety in NAD, are mainly
surrounded by negative potential. Figure 9(a) shows that
steroids and the porphyrin core of heme molecules are
bound in very hydrophobic environments together with
the flavin moieties in FAD and FMN. Because of the
large number of p–CH and p–p interactions involving
adenine moieties in our data set, these moieties experi-
ence large HES despite being of moderate polar nature.
The comparison of the averaged ESPs from the pro-
teins [Fig. 8(a)] and the ligands [Fig. 8(b)] on each
ligand atom reveals a correlation of only R2 5 0.25. The
correlation for HES is higher with R2 5 0.66. Both num-
bers are relatively small and indicate rather little interde-
pendence between the protein’s or ligand’s physicochemi-
cal properties. However, they still exceed the correlation
between all 100 individual binding site/ligand complexes,
with R2 5 0.14 for ESP and R2 5 0.35 for HES, demon-
strating that the average physicochemical properties show
a higher complementarity than individual protein–ligand
pairs.
The standard deviation of the absolute ESP and the
HES is given in Table I. ESP shows large variations with
standard deviations several times larger than the average
value. The highest variation is seen for phosphate-bind-
ing pockets having a standard deviation of 14.64 kcal/
mol/e, whereas the lowest variation is seen in steroid-
binding pockets with 3.93 kcal/mol/e. The variation of
HES is generally less and is around four times smaller
than the average standard deviation of ESP.
Although the absolute values of ESP and HES have a
high standard deviation and thus are very variable, their
values are consistently of the same sign. Figure 10 shows
for each ligand atom the SCR (see Methods) for ESP and
HES. Cyan color indicates no variation (0%) for a ligand
atom with score values all with the same sign; the highest
variation (50%) is colored orange reflecting half positive
and half negative score values. According to Figure 10,
half of all ligand atoms in the data set experience ESP
and hydrophobicity scores of low SCR, whereas half ex-
perience high SCR. Hydrophobic moieties in ligand mol-
ecules often occupy equivalent hydrophobic environ-
ments, like the flavin groups in FAD, FMN or the por-
phyrin core in heme molecules or steroid molecules, but
have their electrostatic environments change. Similar
observations can be made for central phosphate groups
in FAD and NAD. Thus, the interactions are on average
complementary but vary widely in their magnitude.

PROTEINS 11
 A. Kahraman et al.

Figure 8
(a) Average electrostatic potential that each ligand set in the data set ‘‘experiences’’ within protein-binding sites. (b) Electrostatic potential
calculated for each ligand molecule in isolation. The electrostatic potential is colored from red to white to blue for values below 25 kcal/mol/e
to 0 to above 5 kcal/mol/e.

Figure 9
(a) Average hydrophobicity that each ligand set in the data set ‘‘experiences’’ within protein-binding sites. (b) Hydrophobicity of the ligand
molecule in isolation. The level of experienced hydrophobicity is given by the hydrophobic environment scores (HES) (see Methods) and is colored
here from magenta to white to green for polar environments with less than 21 HES to 0 to hydrophobic environments with values above 1 HES.

12 PROTEINS
 Physicochemical Variation in Binding Pockets

Figure 10
Variation of physicochemical property scores measured by the sign change ratio of the scores at each atom. Variation is colored from cyan to white
to orange for relative sign changes of 0–25–50%. 0% represents no score variation for the property. 50% illustrates highest variation with half of
the ligand set having positive and half of the ligand set having negative score values.

DISCUSSION ognition, including the contribution of hydrophobic

Our starting point for this work was to explore the va-
lidity of the assumption that binding sites, which bind
the same ligand, achieve their selectivity by providing a
similar physicochemical environment. To analyze this
ligand binding pocket relationship, we have performed
semiempirical calculations to compute and then compare
the electrostatic and hydrophobic environments in 100
protein binding sites. The binding sites were analyzed in
groups with each group member binding the same
ligand. The results reveal large variations in protein envi-
ronments experienced by each ligand. These large quanti-
tative differences indicate the absence of perfect physico-
chemical complementarity between binding sites and
ligands. A biochemical explanation for this variation
could often be deduced from the differing requirements
of proteins to carry out their function. Nevertheless, this
work gives evidence that—similar to the concept of con-
vergent evolution of gene and protein function—nature
has evolved multiple binding solutions for the same
ligand. For some proteins, it may not matter how the
ligand binds to its receptor, but merely that it binds and
so a diverse range of strategies of binding the same
ligand have evolved.
Regardless of the simplicity of our approach, we were
able to observe some general principles of molecular rec-
interactions to molecular recognition.109,110 To investi-
gate hydrophobic interactions, we have developed a new
hydrophobicity environment score that is based on a sta-
tistical thermodynamic description of atomic log P values.
In this work, the hydrophobicity was observed to vary rel-
atively little among each set of ligands. In contrast, the
ESP varies widely and is highly influenced by NCCs. The
importance of entropic energy contributions is also sup-
ported by the ‘‘buffer zone,’’ that is, the free space
between a binding pocket and its bound ligand,6 which
can stabilize entropically the protein–ligand complex by
allowing the ligand and the binding site residues to retain
some of their vibrational motion and flexibility.9 It
should be pointed out that the correlation of our com-
puted properties with experimental observations (e.g., re-
dox potentials) is modest at best. Herein, we have not
considered any kinetics that accompany protein–ligand
interaction or the binding specificity and discrimination
power of proteins toward their cognate ligands, which all
add a further level of complexity to the analysis of the
protein–ligand interaction in silico. It is well known that
conformational changes in particular in allosteric enzymes
have large effects on the molecular recognition of sub-
strate molecules and thus on the protein function. In fact,
recent results on in vivo proteins have shown that pro-
teins, in particular cell signaling proteins and transcrip-

PROTEINS 13
 A. Kahraman et al.
tion factors, are commonly partially or fully intrinsically
disordered.111 They usually undergo transitions from
their disordered conformation to an ordered structure
upon ligand binding, thereby adopting different isomeric
states. Experiments on antibody–antigen complexes
showed that each of the isomeric states could bind differ-
ent antigens with high specificity though with low affin-
ity.112 Furthermore, most allosteric and induced-fit
enzymes lack a correlation between binding affinity and
specificity because of the smaller magnitude of DDG
when compared with the free binding energy DG.113
Consequently, proteins often lack physicochemical com-
plementarity toward their ligands, which can reduce the
binding affinity but retain a high binding specificity.
Our observations confirm the results found by some
earlier studies that were made on limited datasets (see
references in the Introduction), yet the myth of exact
complementarity remains. The results of our analysis
have consequences for a number of applications, particu-
larly those related to function prediction and virtual
screening. Function prediction methods from structure
are frequently based on detecting similarities between
annotated and functionally unknown binding sites. The
very basis of these approaches is, however, challenged if
complementarity is not a given. Such methods must cope
with this variation and include a probabilistic term in
their similarity search that accounts for our observation
that the same ligand can bind in one binding site via
electrostatic interactions, in another via entropic contri-
butions or in a third via purely van der Waals interac-
tions. A workaround to the probabilistic description
could be the utilization of ‘‘3D consensus binding pro-
files.’’ Such profiles would represent the physicochemical
environment that a ligand on average experiences in vari-
ous proteins. A similarity search would then involve the
comparison of the physicochemical properties of a poten-
tial ATP-binding site against a set of 3D consensus bind-
ing profiles. Not only would this decrease the screening
time for a whole database, but it also might increase the
prediction accuracy, as the average properties tend to
have higher complementarity to the ligand properties
than single binding-site/ligand-complexes.
Despite well-developed theory and extensive progress
in molecular simulations, computational approaches to
calculate both enthalpic and entropic energies are still
limited in accuracy.109 The accurate computation of
these terms is, however, necessary to understand the fine
balance of binding forces underlying protein–ligand inter-
actions and to make reliable predictions of binding ener-
gies. Scoring functions implemented in docking applica-
tions typically use a rather simplistic model of molecular
recognition to allow the screening of a myriad binding
poses.114 As a result, the average accuracy of docking
applications lies within 1.5–2 Å root-mean-square devia-
tion in about 70–80% of cases. Most of these successes
are cases in which ligand and protein are relatively
14 PROTEINS
rigid.115 Our own docking calculations with a MM-
GBSA model116 on our data set showed unfavorable pos-
itive energies of up to 1170 kcal/mol/e for 20 protein–
ligand complexes (data not shown). The results in this ar-
ticle show that computational shortcuts based on comple-
mentarity can be dangerous and that more sophisticated
models would be required for accurate predictions of
binding energies and binding poses that take into account
all factors and forces currently known to contribute to
molecular binding. In particular, desolvation energies117
at the protein-binding site and entropy gain/loss of ligand
and binding site molecules109 will need addressing. This
complexity will provide a challenge for computational
biology, be it for the accurate calculation of binding free
energies or the derivation of more empirical approaches.
CONCLUSIONS
This manuscript expands on previous work on the ge-
ometrical variation in protein-binding pockets and
ligands6 by analyzing the same binding-pocket/ligand-
complexes with respect to their physicochemical proper-
ties. Our analysis of the binding pocket geometries
showed that binding pockets that bind the same ligand
had a greater variation in their shapes than can be
accounted for by the conformational variability of their
ligand. Here, we have continued our analysis on the
same binding sites and found that it is not the chemical
composition of a ligand molecule that determines the
degree of complementarity between protein and ligand,
but rather the ligand’s role within the function of the
protein. Our analyses show that the physicochemical
properties ligands experience when bound to different
binding pockets can vary significantly (Figs. 1 and 7 and
Table I). This high variation reflects large energy fluctua-
tions that are sometimes functionally necessary (Figs. 3
and 5). Some of the variation included even changes in
the sign of the potentials for corresponding atoms in a
ligand set (see Fig. 10). To overcome repulsive electro-
static interactions, proteins were observed to use NCCs
(see Fig. 2) and attractive aromatic interactions (see Fig.
4) to their ligands. Complementarity was observed to
some extent only for averaged properties in the ligand
set (Figs. 8 and 9). With our in-depth analysis, we were
also able to identify false binding sites and determine the
identity of electrochemically active molecules in proteins
(Fig. 6 and Table II).
ACKNOWLEDGMENTS
The authors thank Rafael J. Najmanovich for valuable
comments and discussion throughout the work and
James D. Watson for his valuable comments on early
drafts of this manuscript. All figures containing mole-
cules were rendered using PyMOL (W.L. DeLano, http://
pymol.sourceforge.net/).
 Physicochemical Variation in Binding Pockets
REFERENCES
1. Tsai C-J, Norel R, Wolfson HJ, Maizel JV, Nussinov R. Protein-
Ligand Interactions: Energetic Contributions and Shape Comple-
mentarity. In: Encyclopedia of Life Sciences. John Wiley & Sons,
Ltd: Chichester http://www.els.net/ [doi: 10.1038/npg.els.0001343].
2. Sotriffer C, Klebe G. Identification and mapping of small-molecule
binding sites in proteins: computational tools for structure-based
drug design. Farmaco 2002;57:243–251.
3. Fersht AR. Catalysis, binding and enzyme-substrate complementar-
ity. Proc R Soc Lond B Biol Sci 1974;187:397–407.
4. Liang J, Edelsbrunner H, Woodward C. Anatomy of protein pockets
and cavities: measurement of binding site geometry and implica-
tions for ligand design. Protein Sci 1998;7:1884–1897.
5. An J, Totrov M, Abagyan R. Pocketome via comprehensive identifi-
cation and classification of ligand binding envelopes. Mol Cell Pro-
teomics 2005;4:752–761.
6. Kahraman A, Morris RJ, Laskowski RA, Thornton JM. Shape varia-
tion in protein binding pockets and their ligands. J Mol Biol 2007;
368:283–301.
7. Smith RD, Hu L, Falkner JA, Benson ML, Nerothin JP, Carlson HA.
Exploring protein-ligand recognition with binding MOAD. J Mol
Graph Model 2006;24:414–425.
8. Babine RE, Bender SL. Molecular recognition of protein-ligand
complexes: applications to drug design. Chem Rev 1997;97:1359–
1472.
9. Boehm HJ, Klebe G. What can we learn from molecular recognition
in protein-ligand complexes for the design of new drugs? Angew
Chem Int Ed Engl 1996;35:2588–2614.
10. Basu G, Sivanesan D, Kawabata T, Go N. Electrostatic potential of
nucleotide-free protein is sufficient for discrimination between ade-
nine and guanine-specific binding sites. J Mol Biol 2004;342:1053–
1066.
11. Livesay DR, Jambeck P, Rojnuckarin A, Subramaniam S. Conserva-
tion of electrostatic properties within enzyme families and superfa-
milies. Biochemistry 2003;42:3464–3473.
12. Tsuchiya Y, Kinoshita K, Nakamura H. Structure-based prediction
of DNA-binding sites on proteins using the empirical preference of
electrostatic potential and the shape of molecular surfaces. Proteins
2004;55:885–894.
13. Nakamura H, Komatsu K, Nakagawa S, Umeyama H. Visualization
of electrostatic recognition by enzymes for their ligands and cofac-
tors. J Mol Graph 1985;3:2–11.
14. Nakamura H, Komatsu K, Umeyama H. Electrostatic complemen-
tarities between guest ligands and host enzymes. J Phys Soc Jpn
1985;54:3257–3260.
15. Tsuchiya Y, Kinoshita K, Nakamura H. Analyses of homo-oligomer
interfaces of proteins from the complementarity of molecular sur-
face, electrostatic potential and hydrophobicity. Protein Eng Des Sel
2006;19:421–429.
16. Chau PL, Dean PM. Electrostatic complementarity between proteins
and ligands. I. Charge disposition, dielectric and interface effects.
J Comput-Aided Mol Des 1994;8:513–525.
17. Chau PL, Dean PM. Electrostatic complementarity between proteins
and ligands. II. Ligand moieties. J Comput-Aided Mol Des 1994;8:
527–544.
18. Chau PL, Dean PM. Electrostatic complementarity between proteins
and ligands. III. Structural basis. J Comput-Aided Mol Des 1994;8:
545–564.
19. Naray-Szabo G, Nagy P. Electrostatic complementarity in molecular
aggregates. IX. Protein-ligand complexes. Int J Quantum Chem
1989;35:215–221.
20. Naray-Szabo G. Electrostatic complementarity in molecular associa-
tions. J Mol Graph 1998;7:76–81.
21. McCoy AJ, Chandana Epa V, Colman PM. Electrostatic complemen-
tarity at protein/protein interfaces. J Mol Biol 1997;268:570–
584.
22. Kangas E, Tidor B. Electrostatic complementarity at ligand binding
sites: application to chorismate mutase. J Phys Chem B 2001;105:
880–888.
23. Ledvina PS, Yao N, Choudhary A, Quiocho FA. Negative electro-
static surface potential of protein sites specific for anionic ligands.
Proc Natl Acad Sci USA 1996;93:6786–6791.
24. Kraut DA, Sigala PA, Pybus B, Liu CW, Ringe D, Petsko GA, Hers-
chlag D. Testing electrostatic complementarity in enzyme catalysis:
hydrogen bonding in the ketosteroid isomerase oxyanion hole. Plos
Biol 2006;4:501–519.
25. Wang RX, Gao Y, Lai LH. LigBuilder: a multi-purpose program for
structure-based drug design. J Mol Model 2000;6:498–516.
26. Miranker A, Karplus M. Functionality maps of binding sites: a mul-
tiple copy simultaneous search method. Proteins 1991;11:29–34.
27. Laurie AT, Jackson RM. Q-SiteFinder: an energy-based method for
the prediction of protein-ligand binding sites. Bioinformatics 2005;
21:1908–1916.
28. Goodford PJ. A computational procedure for determining energeti-
cally favorable binding sites on biologically important macromole-
cules. J Med Chem 1985;28:849–857.
29. Nicholls A, Sharp KA, Honig B. Protein folding and association:
insights from the interfacial and thermodynamic properties of
hydrocarbons. Proteins 1991;11:281–296.
30. Kellogg GE, Semus SF, Abraham DJ. HINT: a new method of em-
pirical hydrophobic field calculation for CoMFA. J Comput-Aided
Mol Des 1991;5:545–552.
31. Binkowski TA, Naghibzadeh S, Liang J. CASTp: computed atlas of
surface topography of proteins. Nucleic Acids Res 2003;31:3352–
3355.
32. Kahraman A, Morris RJ, Laskowski RA, Thornton JM. Variation of
geometrical and physicochemical properties in protein binding
pockets and their ligands. BMC Bioinformatics 2007;8:S1.
33. Morris RJ. An evaluation of spherical designs for molecular-like
surfaces. J Mol Graph Model 2006;24:356–361.
34. Morris RJ, Najmanovich RJ, Kahraman A, Thornton JM. Real
spherical harmonic expansion coefficients as 3D shape descriptors
for protein binding pocket and ligand comparisons. Bioinformatics
2005;21:2347–2355.
35. Gruber J, Zawaira A, Saunders R, Barrett CP, Noble ME. Computa-
tional analyses of the surface properties of protein-protein interfa-
ces. Acta Crystallogr D Biol Crystallogr 2007;63:50–57.
36. Gabb HA, Jackson RM, Sternberg MJ. Modelling protein docking
using shape complementarity, electrostatics and biochemical infor-
mation. J Mol Biol 1997;272:106–120.
37. Gerczei T, Asboth B, Naray-Szabo G. Conservative electrostatic
potential patterns at enzyme active sites: the anion-cation-anion
triad. J Chem Inf Comput Sci 1999;39:310–315.
38. Henrick K, Thornton JM. PQS: a protein quaternary structure file
server. Trends Biochem Sci 1998;23:358–361.
39. Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig
H, Shindyalov IN, Bourne PE. The Protein Data Bank. Nucleic
Acids Res 2000;28:235–242.
40. Bairoch A. The ENZYME database in 2000. Nucleic Acids Res 2000;
28:304–305.
41. Apweiler R, Bairoch A, Wu CH, Barker WC, Boeckmann B, Ferro
S, Gasteiger E, Huang H, Lopez R, Magrane M, Martin MJ, Natale
DA, O’Donovan C, Redaschi N, Yeh LS. UniProt: the Universal Pro-
tein Knowledgebase. Nucleic Acids Res 2004;32:115.
42. Pearl F, Todd A, Sillitoe I, Dibley M, Redfern O, Lewis T, Bennett
C, Marsden R, Grant A, Lee D, Akpor A, Maibaum M, Harrison A,
Dallman T, Reeves G, Diboun I, Addou S, Lise S, Johnston C, Sil-
lero A, Thornton JM, Orengo C. The CATH domain structure data-
base and related resources Gene3D and DHS provide comprehen-
sive domain family information for genome analysis. Nucleic Acids
Res 2005;33:247.
43. Mackerell AD, Jr. Empirical force fields for biological macromole-
cules: overview and issues. J Comput Chem 2004;25:1584–1604.
PROTEINS 15
 A. Kahraman et al.
44. Nielsen JE, McCammon JA. Calculating pKa values in enzyme
active sites. Protein Sci 2003;12:1894–1901.
45. Honig B, Nicholls A. Classical electrostatics in biology and chemis-
try. Science 1995;268:1144.
46. Baker NA, Sept D, Joseph S, Holst MJ, McCammon JA. Electro-
statics of nanosystems: application to microtubules and the ribo-
some. Proc Natl Acad Sci USA 2001;98:10037–10041.
47. Klapper I, Hagstrom R, Fine R, Sharp K, Honig B. Focusing of elec-
tric fields in the active site of Cu-Zn superoxide dismutase: effects
of ionic strength and amino-acid modification. Proteins 1986;1:47–
59.
48. Word JM, Lovell SC, Richardson JS, Richardson DC. Asparagine
and glutamine: using hydrogen atom contacts in the choice of side-
chain amide orientation. J Mol Biol 1999;285:1735–1747.
49. Li H, Robertson AD, Jensen JH. Very fast empirical prediction
and rationalization of protein pKa values. Proteins 2005;61:704–
721.
50. Sitkoff D, Sharp KA, Honig B. Accurate calculation of hydration
free-energies using macroscopic solvent models. J Phys Chem 1994;
98:1978–1988.
51. Dolinsky TJ, Nielsen JE, McCammon JA, Baker NA. PDB2PQR: an
automated pipeline for the setup of Poisson-Boltzmann electro-
statics calculations. Nucleic Acids Res 2004;32:W665–W667.
52. Halgren TA. Merck molecular force field. II. MMFF94 van der
Waals and electrostatic parameters for intermolecular interactions.
J Comput Chem 1996;17:520–552.
53. Steinbeck C, Han Y, Kuhn S, Horlacher O, Luttmann E, Willighagen
E. The chemistry development kit (CDK): an open-source Java
library for chemo-and bioinformatics. J Chem Inf Comput Sci
2003;43:493–500.
54. Guha R, Howard MT, Hutchison GR, Murray-Rust P, Rzepa H,
Steinbeck C, Wegner J, Willighagen E. The Blue Obelisk-interoper-
ability in chemical informatics. J Chem Inf Model 2006;46:991–998.
55. DeLano WL. The PyMOL molecular graphics system. San Carlos,
CA: DeLano Scientific; 2002.
56. Fauchère JL, Quarendon P, Kaetterer L. Estimating and representing
hydrophobicity potential. J Mol Graph 1988;6:203.
57. Wang RX, Gao Y, Lai LH. Calculating partition coefficient by atom-
additive method. Perspect Drug Discov Des 2000;19:47–66.
58. Fauchère JL, Pliska V. Hydrophobic parameters-pi of amino-acid
side-chains from the partitioning of N-acetyl-amino-acid amides.
Eur J Med Chem 1983;18:369–375.
59. Eisenberg D, McLachlan AD. Solvation energy in protein folding
and binding. Nature 1986;319:199–203.
60. Heiden W, Moeckel G, Brickmann J. A new approach to analysis
and display of local lipophilicity/hydrophilicity mapped on molecu-
lar surfaces. J Comput-Aided Mol Des 1993;7:503–514.
61. Kelly MD, Mancera RL. A new method for estimating the impor-
tance of hydrophobic groups in the binding site of a protein. J Med
Chem 2005;48:1069–1078.
62. Levitt M. A simplified representation of protein conformations for
rapid simulation of protein folding. J Mol Biol 1976;104:59–107.
63. Brylinski M, Konieczny L, Roterman I. Ligation site in proteins rec-
ognized in silico. Bioinformation 2006;1:127–129.
64. Halgren TA. Representation of van der Waals (vdW) interactions in
molecular mechanics force-fields—potential form, combination
rules, and vdW parameters. J Am Chem Soc 1992;114:7827–7843.
65. Kortemme T, Morozov AV, Baker D. An orientation-dependent
hydrogen bonding potential improves prediction of specificity and
structure for proteins and protein-protein complexes. J Mol Biol
2003;326:1239–1259.
66. Thompson D, Simonson T. Molecular dynamics simulations show
that bound Mg21 contributes to amino acid and aminoacyl adenyl-
ate binding specificity in aspartyl-tRNA synthetase through long
range electrostatic interactions. J Biol Chem 2006;281:23792–23803.
67. Gonzalez B, Pajares MA, Hermoso JA, Guillerm D, Guillerm G,
Sanz-Aparicio J. Crystal structures of methionine adenosyltransfer-
16 PROTEINS
ase complexed with substrates and products reveal the methionine-
ATP recognition and give insights into the catalytic mechanism.
J Mol Biol 2003;331:407–416.
68. Larsen TM, Benning MM, Rayment I, Reed GH. Structure of the
bis(Mg21)-ATP-oxalate complex of the rabbit muscle pyruvate ki-
nase at 2.1 A resolution: ATP binding over a barrel. Biochemistry
1998;37:6247–6255.
69. Zheng J, Knighton DR, ten Eyck LF, Karlsson R, Xuong N, Taylor
SS, Sowadski JM. Crystal structure of the catalytic subunit of
cAMP-dependent protein kinase complexed with MgATP and pep-
tide inhibitor. Biochemistry 1993;32:2154–2161.
70. Schmitt E, Moulinier L, Fujiwara S, Imanaka T, Thierry JC, Moras
D. Crystal structure of aspartyl-tRNA synthetase from Pyrococcus
kodakaraensis KOD: archaeon specificity and catalytic mechanism of
adenylate formation. EMBO J 1998;17:5227–5237.
71. Masuda S, Murakami KS, Wang S, Anders Olson C, Donigian J,
Leon F, Darst SA, Campbell EA. Crystal structures of the ADP
and ATP bound forms of the bacillus anti-sigma factor SpoIIAB in
complex with the anti-anti-sigma SpoIIAA. J Mol Biol 2004;340:941–
956.
72. Bilwes AM, Quezada CM, Croal LR, Crane BR, Simon MI. Nucleo-
tide binding by the histidine kinase CheA. Nat Struct Biol 2001;8:
353–360.
73. Armstrong RN, Kondo H, Granot J, Kaiser ET, Mildvan AS. Mag-
netic resonance and kinetic studies of the manganese(II) ion and
substrate complexes of the catalytic subunit of adenosine 30 ,50 -
monophosphate dependent protein kinase from bovine heart. Bio-
chemistry 1979;18:1230–1238.
74. Hirsch AK, Fischer FR, Diederich F. Phosphate recognition in struc-
tural biology. Angew Chem Int Ed Engl 2007;46:338–352.
75. Smith PE, Tanner JJ. Conformations of nicotinamide adenine dinu-
cleotide (NAD(1)) in various environments. J Mol Recognit 2000;
13:27–34.
76. Percy MJ, Crowley LJ, Boudreaux J, Barber MJ. Expression of
a novel P275L variant of NADH: cytochrome b(5) reductase
gives functional insight into the conserved motif important for
pyridine nucleotide binding. Arch Biochem Biophys 2006;447:59–
67.
77. Mao L, Wang Y, Liu Y, Hu X. Molecular determinants for ATP-
binding in proteins: a data mining and quantum chemical analysis.
J Mol Biol 2004;336:787–807.
78. Denessiouk KA, Rantanen VV, Johnson MS. Adenine recognition: a
motif present in ATP-, CoA-, NAD-, NADP-, and FAD-dependent
proteins. Proteins 2001;44:282–291.
79. Schneider S, Marles-Wright J, Sharp KH, Paoli M. Diversity and
conservation of interactions for binding heme in b-type heme pro-
teins. Nat Prod Rep 2007;24:621–630.
80. Reedy CJ, Gibney BR. Heme protein assemblies. Chem Rev 2004;
104:617–649.
81. Roberts SA, Montfort WR. Haem proteins. In: Encyclopedia of Life
Sciences. John Wiley & Sons, Ltd: Chichester http://www.els.net/
[doi: 10.1002/9780470015902.a0003054].
82. Presta A, Weber-Main AM, Stankovich MT, Stuehr DJ. Comparative
effects of substrates and pterin cofactor on the heme midpoint
potential in inducible and neuronal nitric oxide synthases. J Am
Chem Soc 1998;120:9460–9465.
83. Stuehr DJ. Enzymes of the L-arginine to nitric oxide pathway.
J Nutr 2004;134:2748S–2751S.
84. Caillet-Saguy C, Turano P, Piccioli M, Lukat-Rodgers GS, Czjzek M,
Guigliarelli B, Izadi-Pruneyre N, Rodgers KR, Delepierre M,
Lecroisey A. Deciphering the structural role of histidine 83 for
heme binding in hemophore HasA. J Biol Chem 2008;283:5960–
5970.
85. Bradley AL, Chobot SE, Arciero DM, Hooper AB, Elliott SJ. A
distinctive electrocatalytic response from the cytochrome c
peroxidase of Nitrosomonas europaea. J Biol Chem 2004;279:13297–
13300.
 Physicochemical Variation in Binding Pockets
86. Mao J, Hauser K, Gunner MR. How cytochromes with different
folds control heme redox potentials. Biochemistry 2003;42:9829–
9840.
87. Gunner MR, Honig B. Electrostatic control of midpoint potentials
in the cytochrome subunit of the Rhodopseudomonas viridis reaction
center. Proc Natl Acad Sci USA 1991;88:9151–9155.
88. Reedy CJ, Elvekrog MM, Gibney BR. Development of a heme pro-
tein structure electrochemical function database. Nucleic Acids Res
2008;36:D307–D313.
89. Andersen JF, Ding XD, Balfour C, Shokhireva TK, Champagne DE,
Walker FA, Montfort WR. Kinetics and equilibria in ligand binding
by nitrophorins 1–4: evidence for stabilization of a nitric oxide-fer-
riheme complex through a ligand-induced conformational trap.
Biochemistry 2000;39:10118–10131.
90. Choinowski T, Blodig W, Winterhalter KH, Piontek K. The crystal
structure of lignin peroxidase at 1.70 A resolution reveals a hydroxy
group on the cbeta of tryptophan 171: a novel radical site formed
during the redox cycle. J Mol Biol 1999;286:809–827.
91. Spence JT, Kipke CA, Enemark JH, Sunde RA. Stoichiometry of
electron uptake and the effect of anions and pH on the molybde-
num and heme reduction potentials of sulfite oxidase. Inorg Chem
1991;30:3011–3015.
92. Rich PR, Jeal AE, Madgwick SA, Moody AJ. Inhibitor effects on re-
dox-linked protonations of the B-hemes of the mitochondrial-Bc1
complex. Biochim Biophys Acta 1990;1018:29–40.
93. Hallberg BM, Bergfors T, Backbro K, Pettersson G, Henriksson G,
Divne C. A new scaffold for binding haem in the cytochrome do-
main of the extracellular flavocytochrome cellobiose dehydrogenase.
Structure 2000;8:79–88.
94. Tsai AL, Kulmacz RJ, Wang JS, Wang Y, Vanwart HE, Palmer G.
Heme coordination of prostaglandin-H synthase. J Biol Chem 1993;
268:8554–8563.
95. Lancaster CRD, Gross R, Haas A, Ritter M, Mantele W, Simon J,
Kroger A. Essential role of Glu-C66 for menaquinol oxidation indi-
cates transmembrane electrochemical potential generation by Woli-
nella succinogenes fumarate reductase. Proc Natl Acad Sci USA
2000;97:13051–13056.
96. Makino R, Chiang R, Hager LP. Oxidation-reduction potential
measurements on chloroperoxidase and its complexes. Biochemistry
1976;15:4748–4754.
97. Haas AH, Lancaster CRD. Calculated coupling of transmembrane
electron and proton transfer in dihemic quinol: fumarate reductase.
Biophys J 2004;87:4298–4315.
98. Kleywegt GJ, Harris MR, Zou JY, Taylor TC, Wahlby A, Jones TA.
The Uppsala electron-density server. Acta Crystallogr D Biol Crys-
tallogr 2004;60:2240–2249.
99. Wallimann P, Marti T, Furer A, Diederich F. Steroids in molecular
recognition. Chem Rev 1997;97:1567–1608.
100. Tsuzuki S, Fujii A. Nature and physical origin of CH/pi interaction:
significant difference from conventional hydrogen bonds. Phys
Chem Chem Phys 2008;10:2584–2594.
101. Tsuzuki S, Honda K, Uchimaru T, Mikami M, Tanabe K. The mag-
nitude of the CH/pi interaction between benzene and some model
hydrocarbons. J Am Chem Soc 2000;122:3746–3753.
102. Hunter CA, Lawson KR, Perkins J, Urch CJ. Aromatic interactions.
J Chem Soc Perkin Trans 2001;2:651–669.
103. Cauet E, Rooman M, Wintjens R, Lievin J, Biot C. Histidine-aro-
matic interactions in proteins and protein-ligand complexes: quan-
tum chemical study of X-ray and model structures. J Chem Theory
Comput 2005;1:472–483.
104. Levitt M, Perutz MF. Aromatic rings act as hydrogen bond accept-
ors. J Mol Biol 1988;201:751–754.
105. Toth G, Murphy RF, Lovas S. Stabilization of local structures by pi-
CH and aromatic-backbone amide interactions involving prolyl and
aromatic residues. Protein Eng 2001;14:543–547.
106. Luo R, Gilson HS, Potter MJ, Gilson MK. The physical basis
of nucleic acid base stacking in water. Biophys J 2001;80:140–148.
107. Marsili S, Chelli R, Schettino V, Procacci P. Thermodynamics of
stacking interactions in proteins. Phys Chem Chem Phys 2008;10:
2673–2685.
108. Tsuzuki S, Honda K, Uchimaru T, Mikami M, Tanabe K. Origin of
the attraction and directionality of the NH/pi interaction: compari-
son with OH/pi and CH/pi interactions. J Am Chem Soc 2000;122:
11450–11458.
109. Gilson MK, Zhou HX. Calculation of protein-ligand binding affin-
ities. Annu Rev Biophys Biomol Struct 2007;36:21–42.
110. Davis AM, Teague SJ. Hydrogen bonding, hydrophobic interactions,
and failure of the rigid receptor hypothesis. Angew Chem Int Ed
1999;38:736–749.
111. James LC, Tawfik DS. Conformational diversity and protein evolu-
tion—a 60-year-old hypothesis revisited. Trends Biochem Sci 2003;
28:361–368.
112. James LC, Roversi P, Tawfik DS. Antibody multispecificity mediated
by conformational diversity. Science 2003;299:1362–1367.
113. Schneider HJ. Ligand binding to nucleic acids and proteins: does se-
lectivity increase with strength? Eur J Med Chem 2008;43:2307–2315.
114. Jain AN. Scoring functions for protein-ligand docking. Curr Protein
Pept Sci 2006;7:407–420.
115. Sousa SF, Fernandes PA, Ramos MJ. Protein-ligand docking: current
status and future challenges. Proteins 2006;65:15–26.
116. Lyne PD, Lamb ML, Saeh JC. Accurate prediction of the relative
potencies of members of a series of kinase inhibitors using molecu-
lar docking and MM-GBSA scoring. J Med Chem 2006;49:4805–
4808.
117. Koehl P. Electrostatics calculations: latest methodological advances.
Curr Opin Struct Biol 2006;16:142–151.
PROTEINS 17