3950

J. Phys. Chem. B 2010, 114, 39503964

Computational Modeling of Carbohydrate-Recognition Process in E-Selectin Complex:

Structural Mapping of Sialyl Lewis X onto Ab Initio QM/MM Free Energy Surface
Toyokazu Ishida*
Research Institute for Computational Sciences (RICS), National Institute of AdVanced Industrial Science and
Technology (AIST), 1-1-1 Umezono, Tsukuba, 305-8568, Japan, and Core Research for EVolutional Science
and Technology (CREST), Japan Science and Technology Agency (JST), Kawaguchi 332-0012, Japan
ReceiVed: June 23, 2009; ReVised Manuscript ReceiVed: December 16, 2009

To advance our knowledge of carbohydrate recognition by lectins, we propose a systematic computational

modeling strategy to identify complex sugar-chain conformations on the reduced free energy surface (FES).
We selected the complex of E-selectin with sialyl Lewis X (denoted E-selectin/SLex complex) as a first target
molecule. First, we introduced the reduced 2D-FES that characterizes conformational changes in carbohydrate
structure as well as the degree of solvation stability of the carbohydrate ligand, and evaluated the overall free
energy profile by classical molecular dynamics simulation combined with ab initio QM/MM energy corrections.
Second, we mapped flexible carbohydrate structures onto the reduced QM/MM 2D-FES, and identified the
details of molecular interactions between each monosaccharide component and the amino acid residues at the
carbohydrate-recognition domain. Finally, we confirmed the validity of our modeling strategy by evaluating
the chemical shielding tensor by ab initio QM/MM-GIAO computations for several QM/MM-refined geometries
sampled from the minimum free energy region in the 2D-FES, and compared this theoretical averaging data
with the experimental 1D-NMR profile. The model clearly shows that the binding geometries of the E-selectin/
SLex complex are determined not by one single, rigid carbohydrate structure but rather by the sum of averaged
conformations fluctuating around the minimum free energy region. For the E-selectin/SLex complex, the major
molecular interactions are hydrogen bonds between Fuc and the Ca2+ binding site in the carbohydraterecognition domain, and Gal is important in determining the ligand specificity.
1. Introduction
Carbohydrates are the most widely distributed naturally
occurring organic compounds on Earth. Molecular interactions
between proteins and carbohydrates are ubiquitous in a wide
variety of physiological and pathological processes.1-3 For
example, specific molecular recognition of carbohydrate/sugar
chains (or glycan in general) by relevant proteins plays a key
role in cell-cell interactions such as innate immune response,
metabolism, fertilization, inflammation, and cell growth. Characterizing the molecular basis of carbohydrate recognition based
on detailed three-dimensional (3D) structures is crucially
important for a variety of classes of protein-carbohydrate
complexes.4,5 X-ray crystallography studies have played a central
role in the exponential increase in structural determinations of
proteins and their complexes with various ligands. In contrast,
high-resolution structural studies of protein-carbohydrate complexes continue to be a major challenge.1,2 The main problem
that prevents direct application of crystallographic techniques
is the complexity of flexible carbohydrate ligands. The variety
of monosaccharide components, linkage types, and hydroxyl
group modifications lead oligosaccharides to more potential
structural diversity than is seen for other classes of biomolecules.6,7
Furthermore, carbohydrates often function in biological processes as glycoconjugates with various proteins and lipids. The
chemical and conformational heterogeneity of such carbohydrate
ligands generally prevents crystallization. As a result, protein
crystal structures can be readily obtained in the absence of
carbohydrate ligands or in complexes with unnatural ligands.
* Phone: +81-29-861-5206.
toyokazu.ishida@aist.go.jp.

Fax:

+81-29-851-5426.

E-mail:

Accordingly, most of our knowledge of protein-carbohydrate

interactions is based on information from 3D structures of
homologue proteins with small ligands.1,2,4,5 In such cases,
nuclear magnetic resonance (NMR) spectroscopy is the primary
tool for determining molecular structures.8,9 Although NMR can
be used directly to determine solution-phase structures, interpretation of NMR spectra of oligosaccharides is usually difficult
because most protons are in a similar chemical environment.
And even in the same target protein complex, different NMR
measurements often predict slightly different conformations of
carbohydrate ligands.
In view of these difficulties in conducting experimental studies
on glycoconjugate complexes, we look to theoretical/computational modeling to explore and predict molecular interactions
between proteins and carbohydrates. Knowledge of these
interactions is necessary, for example, for rational drug design
of specific molecular inhibitors.9-13 In this paper, we report a
systematic computational modeling strategy based on ab initio
quantum mechanical/molecular mechanical (QM/MM) computations combined with all-atom molecular dynamics (MD)
simulations. By introducing a free energy surface that characterizes the conformational changes of the carbohydrate structure
as well as the degree of solvation energy gain of the carbohydrate ligand, we determine the binding conformation of carbohydrates inside the protein complex and clarify the essence of
the protein-carbohydrate interactions. As a first target system
for this purpose, we selected the selectin-carbohydrate complex.
Proteins that bind carbohydrates/sugar chains are generally
called lectins. Many types of known lectins are categorized into
major groups.1,2 Among these, selectins are C-type lectins that
require Ca2+ ions for the carbohydrate-recognition domain

10.1021/jp905872t 2010 American Chemical Society

Published on Web 01/15/2010

Carbohydrate-Recognition Process in E-Selectin Complex

J. Phys. Chem. B, Vol. 114, No. 11, 2010 3951

twofold: (1) practical determination of the free energy surface
(FES) that characterizes sugar chain binding by means of ab
initio QM/MM computations; (2) structural mapping of the
carbohydrate conformations on the FES and practical analysis
of molecular interactions in the E-selectin/SLex complex as a
demonstration of typical lectin-carbohydrate recognition.

Figure 1. Chemical structure of sialyl Lewis X; the four sugar

components are N-acetyl neuramic acid (Neu5Ac), galactose (Gal),
N-acetyl glucosime (GlcNAc), and fucose (Fuc). Arabic numerals
indicate the numeration of the atoms in the carbohydrate structure. The
dihedral angles of the glycoside linkages are defined as follows: In
Neu5Ac(2-3)Gal, is C1-C2-O2-C3 and is C2-O2-C3-H3;
in Gal(1-4)GlcNAc, is H1-C1-O1-C4 and is C1-O1-C4-H4;
and in Fuc(1-3)GlcNAc, is H1-C1-O1-C3 and is
C1-O1-C3-H3.

(CRD)14 and mediate the adhesion of leukocytes to the vascular

endothelial surface during the early stages of an inflammatory
response.15-18 Three selectins have been identified: E (endothelial), P (platelet), and L (leukocyte) selectins. All three have
a common structural motif consisting of a C-type CRD, an
epidermal growth-factor-like domain, complement control protein modules, and a transmembrane-anchoring motif with a short
cytoplasmic domain. These selectins mediate cell adhesion by
means of molecular recognition of, and interactions with,
carbohydrates. Selectin ligands require the carbohydrate motif
of the so-called sialyl Lewis X (SLex),19,20 a tetrasaccharide
composed of four sugar unitssN-acetylneuramic acid (Neu5Ac),
galactose (Gal), N-acetylglucosamine (GlcNAc), and fucose
(Fuc)swith the following linkage (Figure 1): R-Neu5Ac-(2-3)-D-Gal-(1-4)-(R-L-Fuc-(1-3))--D-GlcNAc. Although all selectins show a significant degree of sequence homology among
themselves and share a common CRD that is responsible for
the SLex binding, their affinities and recognition patterns for
target ligands differ slightly.21-25
Because of the biological importance of selectins during cell
adhesion, many experimental studies have been reported.26-35
However, because of the complexity and flexibility of the
carbohydrate chain inside the selectin complex, an atomistic
picture of molecular recognition remains elusive. Differences
among the three selectins have not yet been clarified. Few
computational studies have been reported,36,37 and most previous
studies discuss only limited topics. For example, Pichierri and
Matsuo performed semiempirical molecular orbital (MO) computations for the E-selectin/SLex complex to assess the charge
state of the SLex carboxyl group: analysis of MO, electrostatic
potential surfaces, and conformation of the sugar chain revealed
that the neutral form of SLex has the necessary characteristics
for carbohydrate recognition.36 Veluraja and Margulis performed
MD simulations to determine the conformational dynamics of
SLex in aqueous solution as well as in the E-selectin complex,
and suggested that two lysine residues at the binding site (Lys99
and Lys111) may play important roles in sugar chain binding
because of their large configurational flexibility.37
Fortunately, the 3D structure of the E-selectin/SLex complex
has recently been determined by X-ray crystallography,31
enabling us now to prepare a reasonable molecular model to
investigate molecular interactions on the CRD.31 To obtain an
atomistic picture of the carbohydrate-recognition process, which
is the origin of tethering and rolling processes, we examined
the molecular mechanism of carbohydrate recognition in the
E-selectin/SLex complex by means of a systematic computational modeling procedure. The major focus of this paper is

2. Method and Computational Details

2.1. General Outline of Carbohydrate Modeling. A major
difficulty in modeling glycoconjugates is that only limited
structural data is available for the carbohydrate binding region,
and it is typically difficult to identify flexible and multiple
conformations of carbohydrate chains on the CRD using
crystallographic data.13 In this work, to overcome these difficulties, we employ the following modeling strategy.
(1) First, we define the binding mode of the SLex motif on
the CRD of selectin. Although sugar chain conformations bound
to CRDs are generally necessary for carbohydrate recognitions,
what an actual binding conformation looks like is not clear. In
the present model, we define the binding state as the structural
aVerage of flexible carbohydrate conformations around the
minimum free energy region in the reduced 2D-FES. We define
the reduced 2D-FES by two characteristic parameters (or
reaction coordinates): the solute coordinate that describes
conformational changes in the carbohydrate chain (for example,
dihedral angles between glycosidic linkages and rotational angles
of hydroxyl groups) and the solvation coordinate that describes
molecular interactions between the SLex ligand and the surrounding environment (including the protein and aqueous
solvent). In this Article, solvation means the generalized
solute-solvent interaction, in which surrounding protein and
aqueous solvent are equally considered in the computational
modeling.
(2) Next, we calculate the free energy profile in reduced 2D
coordinate space. For this purpose, we perform classical MD
simulations to explore the necessary phase space because direct
ab initio QM/MM MD simulations are computationally too
demanding for a system of this size. For the parameters of the
SLex carbohydrate structure, we first calculate charges derived
from electrostatic potential (ESP) to optimize the MM parameters so as to reproduce the ab initio QM/MM results.38,39 To
better reproduce the QM/MM calculation results, we employed
several numbers of SLex conformations extracted from the MD
simulations. For each conformer, we have performed ab initio
QM/MM computations to determine ESP charges. Finally, by
averaging each parameter set, we have obtained a reasonable
ESP set which can reproduce similar QM/MM results for the
selected SLex conformers. After obtaining several nanosecondorder MD trajectories, we perform ab initio QM/MM computations for the entire solvated selectin complex, evaluate the QM
energy along the MD trajectories, and replace the MM energy
with the QM/MM energy.
(3) Then, we characterize the SLex conformations on the CRD
of E-selectin as an ensemble average of structures in the free
energy minimum region. In the subsequent structural analysis,
we focus on two important experimental parameters;40 the
binding epitope that characterizes molecular recognition between
amino acid residues and each monosaccharide unit of SLex, and
the bioactive conformation that describes the SLex structure on
the CRD of E-selectin. For the latter purpose, we discuss the
distribution of the two dihedral angles and between
glycosidic linkages.
(4) Finally, to confirm the validity of the modeling strategy,
we compare calculation results with several experimental

3952

J. Phys. Chem. B, Vol. 114, No. 11, 2010

Ishida

observations. Among available experimental data, we focus on

the proton chemical shifts of the SLex structure because NMR
measurements are the standard experimental method for studying
glycoprotein complexes to date.8,9 To compute reliable chemical
shifts, we perform ab initio QM/MM geometry optimizations
for several solvated selectin complexes extracted from the
minimum energy region of the FES, and perform gaugeincluding atomic orbital (GIAO) calculations within the QM/
MM methodology for each optimized structure. We then
compare computational modeling results with available experimental data to determine the essence of carbohydrate recognition
by selectins as well as the reliability of our systematic
computational modeling strategy, and to better understand the
known structural deviations between X-ray crystallography31 and
NMR measurements.26-29
2.2. Structural Mapping of Carbohydrate Conformation
on the QM/MM Free Energy Surface. Throughout this paper,
we employ the ab initio QM/MM method in which the total
energy of the system is defined by41-47

Etotal ) EQM + EQM/MM + EMM

(1)

where EQM and EMM are the molecular energies in the QM and
MM regions and EQM/MM is the interaction energy between the
QM and MM regions. In the present QM/MM model, we use
the electrostatic embedding form, so the interaction energy
between the QM and MM regions is defined by
elec
vdw
strain
EQM/MM ) EQM/MM
+ EQM/MM
+ EQM/MM

F(RQM, RMM) ) -kBT ln Z

) -kBT ln

-[EQM(RQM)+EQM/MM(RQM,RMM)+EMM(RMM)]

(5)

dRQM dRMM

Recently, computations of QM/MM free energy have been

highly debated.51-54 In previous papers, we demonstrated a
computational strategy for enzymatic reactions that combines
ab initio QM/MM calculations with constrained MD-FEP
simulations.48-50 Here, we introduce another practical technique
for ligand binding processes within the same ab initio QM/
MM framework.
Because direct evaluation of the above equation is impractical
for almost all chemical applications, we introduce appropriate
operators to reduce the dimension of the free energy hypersurface. In this study, we map QM geometry onto the reduced 2DFES. At a first step, we focus on a SLex geometry stabilized by
the surrounding aqueous protein environment and extract a
i
) from the total free
particular QM configuration (denoted RQM
energy space as follows:

i
F(RQM
, RMM) ) -kBT ln

dR

QM(RQM

i
- RQM
)

e-[EQM(RQM)+EQM/MM(RQM,RMM)+EMM(RMM)] dRMM
) -kBT ln [e-EQM(RQM)
i

dR
)

(2)

MMe

i ,R
-[EQM/MM(RQM
MM)+EMM(RMM)]

i
)
EQM(RQM

kBT ln

dR

MMe

i ,R
-[EQM/MM(RQM
MM)+EMM(RMM)]

(6)

elec
is the electrostatic interaction energy, evaluated
where EQM/MM
by the ab initio MO method as follows:48-50

elec
EQM/MM
)

MM

-ZMM
D|
| +
|r - RMM |
ZQMZMM
R - RMM
QM QM

(3)

where ZQM and ZMM are the nuclear charges of QM atoms and
MM point charges, RQM and RMM are the QM and MM nuclear
coordinates, and D is the density matrix in which and run
vdw
is the nonbonded van der Waals
over atomic orbitals. EQM/MM
strain
is the steric interaction energy
interaction energy, and EQM/MM
at the boundary region between QM and MM regions. These
three EQM/MM terms are necessary to properly describe the QM/
MM interactions, and the last two are calculated using the
strain
includes the bonded interaction
empirical force field. EQM/MM
energy at the boundary, calculated from the MM bond, bend,
torsion, and improper torsion energies.
On the basis of this electrostatic embedding QM/MM scheme,
we now evaluate the free energy profile that describes sugar
chain binding. Formally, the configurational partition function
of the total QM/MM system is

Z)

e-[E

QM(RQM)+EQM/MM(RQM,RMM)+EMM(RMM)]

dRQM dRMM
(4)

and the free energy of the total QM/MM system is

The first term in this equation is the electronic energy of the

QM region (that is, the intramolecular, conformational energy
of the SLex sugar chain); the second term is the set of free energy
factors for creating an appropriate surrounding environment for
stabilizing particular QM geometries. As a second step, we
evaluate the second free energy term of the above equation. In
our 2D-FES model, this free energy contribution corresponds
to the reorganization energy for creating an ideal environment
that stabilizes a particular SLex conformation. To account for
this free energy cost, we consider the operator described by the
collective reaction coordinate (E - E(RMM)), where E(RMM)
is the interaction energy between the QM region and the
surrounding MM environment, determined by the RMM configuration. The second term of the above equation then becomes

F2nd(RMM) )

1
ZMM

dRMM(E - E(RMM))
e-[EQM/MM(RQM,RMM)+EMM(RMM)]
i

(7)

where ZMM is the normalization factor defined by

ZMM )

dRMMe-[E

i
QM/MM(RQM,RMM)+EMM(RMM)]

(8)

In summary, the final form of the free energy expression is given

by the following relationship:

Carbohydrate-Recognition Process in E-Selectin Complex

i
i
F(RQM
, E(RMM)) = EQM(RQM
)1
kBT ln
dRMM(E - E(RMM))
ZMM

e-[EQM/MM(RQM,E(RMM))+EMM(RMM)]
i

J. Phys. Chem. B, Vol. 114, No. 11, 2010 3953

shielding tensor within the framework of QM/MM methodology.49,57 The shielding tensor is defined as the second-order
response to the magnetic field and the nuclear magnetic moment:

(9)

Because this equation is a formal expression, for practicality,

we first determine the particular forms of the reaction coordinates. In this case, we introduce two kinds of reaction
coordinates to represent a reduced 2D-FES: the solvation
coordinate that describes the degree of solvation stability of the
SLex ligand, and the solute coordinate that describes the
conformational changes of the SLex sugar chain.
To select the solvation coordinate (which corresponds to the
second term in eq 9), we consider the interaction energy between
the carbohydrate ligand and surrounding environment (including
protein and solvent molecules). This selection is rather straightforward and gives a clear physical picture. In the present
theoretical model, this energy term directly corresponds to the
QM/MM interaction energy (electrostatic interaction evaluated
by ab initio QM/MM computations and van der Waals interaction evaluated by MM force field treatment). To select the solute
i
in the
coordinate (which corresponds to the selection of RQM
first term of eq 9), however, is not as straightforward. For
example, for enzymatic reactions that accompany formation or
breakdown of particular chemical bondings, selection of the QM
solute coordinate is obvious, and we can define the solute
coordinate as the breaking/forming bond distances.48-50 However, for carbohydrate recognition, selection is not obvious
because there are no covalent bond formations. In our preliminary calculations, we could not identify the particular modes
of dihedral angle that describe crucial conformational changes
in carbohydrate structure (see later discussion). Therefore, for
the QM solute coordinate, we select the intramolecular energy
evaluated by ab initio QM/MM calculations. Direct application
of ab initio QM/MM MD simulations may be possible; however,
nanoscale simulations of target protein systems are at present
too computationally demanding. Considering the system size
of the QM region and the total atomic numbers of the solvated
protein complex, we employ another efficient approach to
evaluate eq 9. To collect the MD trajectory, we first perform
classical MD simulations on a several-nanosecond time scale.
After obtaining numbers for enough sample configurations, we
perform ab initio QM/MM calculations to evaluate the QM/
MM 2D-FES based on eq 9. Due to the complexity of the SLex
sugar chain, different conformations have nearly the same
intramolecular energy; that is, E(RiQM) E(RjQM). It is meaningless to distinguish among different structures that have the same
energy even though they show small structural differences.
Therefore, for the solute coordinate, we finally select the
potential of mean force profile along the selected solute
coordinate

F(E(RQM)) ) -kBT ln F(E(RQM)) + C

(10)

where F is the probability distribution of the QM solute energy

and C is a constant. We then estimate the 2D-FES using QM/
MM sampling calculations. For practicality in computing the
free energy factors in eq 9, we use a histogram-based approach
to evaluate the probability density.55,56 The actual computational
procedure is described in a later section.
For calculations of NMR chemical shifts, here we briefly
summarize the computational procedure to evaluate the chemical

k )
)

2E(B, mk)
Bmk

B)0,mk)0

H
+
D
Bmk

P H
B mk

(11)

where B is the external magnetic field and mk is the nuclear

magnetic moment of atom k. We employ the gauge-including
atomic orbital (GIAO) approach to calculate eq 11.58-60 In
GIAO, the shielding tensor is calculated by means of the fielddependent basis functions as follows:

|(B) ) exp -

ie
(B R) r |(0)
2cp

(12)

where |(0) is the usual field-independent basis function and

R is the center of the basis function.
We limit computations to the Hartree-Fock wave function
to describe the QM region. Because we consider only the
electrostatic effect caused by the ESP charges assigned on each
MM atomic site, the perturbation effect from the MM environment is introduced by means of the density matrix D and its
derivative. D is determined from the converged QM/MM wave
function, and the derivative of D with respect to the magnetic
field B is solved via the coupled perturbed Hartree-Fock
(CPHF) equation. The QM/MM correction to the CPHF equation
is given by

elec
|(B)
)
(B)|HQM + HQM/MM
B
B
(B)
elec
|(B) + (B)|HQM +
|HQM + HQM/MM
B
(B)
HQM
elec
HQM/MM
|
(13)
+ (B)|
| (B)
B
B

where HQM is the one-electron part of the QM Hamiltonian and

elec
is the electrostatic interaction between QM and MM
HQM
regions. With these slight modifications, the chemical shielding
tensor can be calculated within the QM/MM framework.
2.3. Computational Details. All QM/MM and MM (modeling and simulations) calculations reported here are based on
the original code reported earlier.48-50 The ab initio QM/MM
program was developed based on the HONDO package,61 and
MM modeling/simulation routines were added to the ab initio
MO computation part. For the MM region, we employed the
AMBER (parm.96/parm.99)62,63 and GLYCAM64 parameter sets
for force field calculations.
2.3.1. QM/MM Structural Refinement of X-ray Crystallographic Data. Because relatively well-resolved X-ray crystal
data have been recently published for E-selectin31 (an apparently
rare case in the field of glycoprotein complexes), we performed
ab initio QM/MM structural refinements on reference structural
data for the following discussion. The procedure to obtain a
reasonable chemical model is summarized as follows. The initial
coordinates of the lectin-carbohydrate complex were adopted

3954

J. Phys. Chem. B, Vol. 114, No. 11, 2010

Ishida

Figure 2. (a) Initial solvated protein model used in ab initio QM/MM structural refinement. Green sticks represent the sialyl Lewis X bound to
the carbohydrate-recognition domain; yellow sticks represent surrounding solvent molecules which were added to prevent unphysical structural
deformation of the naive lectin-carbohydrate interaction. (b) Superimposed structures of the QM/MM-refined model with the original X-ray
coordinates. Green and blue sticks represent the original X-ray data for E-selectin and sialyl Lewis X, respectively; purple and red sticks represent
the QM/MM optimized geometries of the protein and sugar chain, respectively. (c, d) Detailed hydrogen-bond geometries in the Fuc and Gal units,
respectively, determined by QM/MM refinement (geom0 in Table 1).

from X-ray crystal data for E-selectin lectin/EGF domains

complexed with SLex, determined at 1.5 resolution (PDB code
1G1T). Hydrogen atoms were added to the E-selectin/SLex
complex in a standard manner; all polar residues (Asp, Glu,
Lys, and Arg) were assumed to be in the ionized state, and all
His were assumed to be in the protonated form (both 1 and 2
positions were protonated), because all are located on the
solvent-exposed surface and are surrounded by crystal water.
For the carboxyl group of Neu5Ac in SLex, the protonated form
was assumed because of the experimental implications of the
recent X-ray structural study.31,36 Our initial model considered
all crystal water and Ca2+ ion. On the basis of these definitions,
the total charge of the system was exactly zero.
Before QM/MM geometry optimization, we removed unfavorable steric contact by rough MM energy minimizations.
Because the SLex ligand is weakly attached to the protein surface
of the CRD in this case, we first generated a reasonable solvated
structure so as not to destroy naive contact between the flexible
SLex and the selectin. The initial structure of the selectin
complex was solvated in a sphere of TIP3P water molecules65
with 35 radius centered on the center of mass of the complex.
Any water molecules that came within 3.0 of the lectin
complex were removed; the resultant system contained 6300
water molecules. We performed NVT Monte Carlo (MC)
simulations at 303 K in the standard Metropolis manner to obtain
reasonable solvation configurations.66 Only water molecules
were moved randomly with an acceptance probability of 40%,
and 30 M configurations were generated to achieve thermal
equilibrium. The system temperature was decreased gradually

to below 250 K, and the resultant solvation geometry was

adopted for the starting structure of the ab initio QM/MM
geometry optimizations. In all QM/MM calculations, the QM
region was limited for simplicity to only the SLex ligand, so a
steric strain term that links the QM/MM boundaries in eq 2
was not required. We used the restricted Hartree-Fock (RHF)
wave function with the 6-31G* basis set. However, considering
the system size of the QM region, we first optimized the entire
solvated protein complex at the QM/MM RHF/3-21G/AMBER
level. Finally, we refined the molecular geometry at the QM/
MM RHF/6-31G*/AMBER level for the detailed description.
Convergence criteria used in ab initio QM/MM optimizations
were as follows: maximum gradient less than 5 10-4 au/bohr
in the QM region; rms gradient less than 1 10-2 kcal/mol -1
in the MM region. No cutoff was introduced between QM and
MM interactions.
2.3.2. Free Energy Surface by Ab Initio QM/MM Combined
with MD Sampling. For MD simulations to construct the 2DFES, we used the same initial protein geometry but considered
a larger solvation structure. The selectin-carbohydrate complex
was solvated in a sphere of TIP3P water molecules with 42
radius centered on the center of mass of the complex: the
resultant system contained 11000 water molecules. After
preparing a reasonable condition for the solvent configuration
by NVT MC simulations, we relaxed the solvated selectin
complex completely by performing MD simulations for 1 ns.
We employed the Nose-Hoover-chain (NHC) thermostat
method to generate the NVT ensemble,67 and maintained the
system temperature at 303 K by attaching the five chains of the

2.82

2.72

2.67

3.56

2.48

4.58

5.52

5.53

2.94

3.16

5.59

2.83

2.57

3.30

2.8

2.9

4.0

3.6

2.5

3.8

5.3

5.4

4.0

3.2

4.3

3.2

2.6

3.0

2.93

2.45

2.82

2.44

3.73

2.97

9.58

2.78

4.48

6.36

7.79

7.50

4.07

5.83

6.63

geom
1

2.83

2.46

2.80

3.73

3.67

3.14

11.72

3.87

2.45

5.36

6.82

6.74

3.78

6.71

6.90

geom
2

2.83

2.47

2.81

5.51

3.40

2.79

11.95

4.11

2.48

4.15

5.38

7.01

4.88

6.89

7.71

geom
3

2.98

2.44

2.84

4.72

3.65

2.86

9.60

3.07

2.47

4.62

6.01

6.44

4.02

5.46

6.33

geom
4

2.86

2.46

2.75

4.49

3.80

2.89

10.43

2.84

2.50

5.02

6.42

6.95

4.41

7.13

8.03

geom
5

3.03

2.46

2.79

4.54

3.61

3.14

11.82

3.03

2.50

4.66

6.03

6.27

3.89

6.20

6.72

geom
6

2.80

2.45

2.80

4.32

3.89

2.90

10.92

2.87

2.43

4.77

6.26

6.28

4.25

7.79

8.18

geom
7

2.79

2.45

2.82

4.84

3.61

3.09

11.39

4.02

2.47

4.71

6.04

6.78

4.03

5.78

4.98

geom
8

2.92

2.47

2.79

4.74

3.89

3.14

11.69

3.27

2.45

4.59

6.19

6.19

4.76

5.72

7.80

geom
9

2.83

2.45

2.82

4.54

4.01

2.82

9.73

2.78

2.45

4.72

6.23

6.47

5.23

7.94

8.32

geom
10

2.81

2.45

2.86

5.30

3.71

2.83

8.27

2.94

2.53

3.92

5.63

6.01

4.44

5.66

6.54

geom
11

2.85

2.43

2.80

4.98

3.82

2.81

8.89

3.06

2.45

5.13

6.48

6.67

4.23

7.83

7.68

geom
12

2.91

2.49

2.76

4.33

3.99

2.94

4.65

2.84

2.44

4.23

5.43

5.46

3.74

5.55

6.69

geom
13

2.91

2.47

2.83

4.15

3.73

3.21

2.76

5.20

2.49

5.03

6.71

6.62

3.96

6.01

7.27

geom
14

2.78

2.45

2.80

4.02

3.69

3.57

4.18

3.05

2.52

5.21

6.84

6.84

3.73

7.08

7.27

geom
15

2.83

2.42

2.88

4.87

3.44

3.26

2.71

3.53

2.50

4.93

6.42

7.00

3.70

6.71

6.44

geom
16

2.80

2.45

2.79

3.75

3.70

3.41

2.68

3.33

2.47

5.30

7.07

6.94

3.92

7.87

7.75

geom
17

2.79

2.45

2.80

3.75

3.80

2.86

4.81

3.20

2.54

5.46

6.67

6.86

4.02

8.26

9.27

geom
18

2.82

2.45

2.81

4.42

3.68

3.18

2.71

3.52

3.77

4.85

6.36

6.62

3.71

8.90

10.28

geom
19

2.87

2.45

2.75

4.51

3.76

3.49

2.69

3.02

3.31

5.15

6.96

6.83

3.95

8.42

9.77

geom
20

54.4

41.0

14.6

37.4

23.9

34

16

41

22

19.7

40.1

15.7

2.6

-12 -5.9

25.3

51.7

16.1

33.9

3.1

43.9

63.5

26.1

37.4

5.1

30.9

56.1

14.8

42.0

2.0

22.6

54.8

17.3

43.0

9.3

21.2

49.8

7.7

42.0

3.0

25.5

59.5

14.2

36.1

1.6

28.6

56.0

10.4

35.3

-4.3

33.2

66.9

20.1

29.6

30.5

55.0

18.2

32.2

27.0

53.8

19.2

30.8

23.5

58.5

18.9

37.0

-29.4 -27.4 -24.0 -3.8

25.0

45.5

12.1

39.2

11.5

20.2

48.4

13.9

28.8

12.6

12.9

46.3

8.5

32.3

-1.9

5.8

46.8

13.0

37.1

-1.9

21.2

47.4

11.2

32.6

2.4

25.5

48.4

19.3

44.3

14.9

15.9

46.7

21.9

39.1

30.0

21.8

50.5

18.1

33.8

26.4

-65 -62.6 -67.2 -75.0 -58.7 -71.0 -69.9 -71.6 -67.6 -75.7 -58.7 -64.4 -64.5 -61.7 -58.6 -66.5 -65.0 -56.5 -69.8 -34.3 -24.6 -39.4

2.63

geom
0

2.5

exp.

sampled and quenched structures from MD trajectoryb

a
Original experimental data from X-ray coordinates (exp.) and QM/MM-refined geometry of X-ray coordinates (geom0). b QM/MM optimized structures sampled from the minimum free energy
region of the reduced 2D-FES. c Interatomic distances between two heavy atoms in both E-selectin and sialyl Lewis X. d Dihedral angles at the glycoside linkages.

O(carboxyl,Neu5Ac)
-OH(Tyr48)c
O(carboxyl,Neu5Ac)
-NE(Arg97)c
O4(hydroxyl,Gal)
-OH(Tyr94)c
O6(hydroxyl,Gal)
-OH(Tyr94)c
O6(hydroxyl,Gal)
-OE1(Glu92)c
O6(hydroxyl,Gal)
-OE2(Glu92)c
O6(hydroxyl,Gal)
-OE1(Glu107)c
O6(hydroxyl,Gal)
-NZ(Lyz111)c
O(carbonyl,GlcNAc)
-NH(Arg108)c
O2(hydroxyl,Fuc)
-ND(Asn83)c
O3(hydroxyl,Fuc)
-ND(Asn105)c
O3(hydroxyl,Fuc)
-OE1(Glu107)c
O4(hydroxyl,Fuc)
-OE1(Glu80)c
O4(hydroxyl,Fuc)
-OE2(Glu80)c
O4(hydroxyl,Fuc)
-ND(Asn82)c
Neu5Ac(2-3)
-Gal, d
Neu5Ac(2-3)
-Gal, d
Gal(1-4)
-GlcNAc, d
Gal(1-4)
-GlcNAc, d
Fuc(1-3)
-GlcNAc, d
Fuc(1-3)
-GlcNAc, d

geometrical
parameter

X-raya

TABLE 1: Characteristic Geometrical Parameters (Hydrogen Bond Distances/Dihedral Angles in Glycoside Linkages) in 20 QM/MM-Refined Structural Models

Carbohydrate-Recognition Process in E-Selectin Complex

J. Phys. Chem. B, Vol. 114, No. 11, 2010 3955

3956

J. Phys. Chem. B, Vol. 114, No. 11, 2010

thermostat with the thermostat mass corresponding to ) 0.5

ps.68 We also employed the reversible reference system propagation (rRESPA) algorithm extended to the non-Hamiltonian
NHC system.69,70 Long-range nonbonded forces (electrostatic
and van der Waals) were integrated on a long time scale (2.0
fs); short-range bonded forces (bond, bend, torsion, and improper
terms) were integrated on a short time scale (0.25 fs). No cutoff
was introduced for nonbonded interactions in all simulations.
As noted in section 2.1, we performed MD simulations at the
classical all-atom MM level to save computational time. Before
MD simulations, the partial charges on the MM atomic sites of
the SLex motif were reoptimized to reproduce the QM/MM
interaction energy analysis.38,39 After collecting the MD trajectory for more than 5 ns, we performed QM/MM single-point
energy calculations for each sampled geometry for a total of
25000 QM/MM computations. For practicality, we limited
calculations to the QM/MM RHF/3-21G/AMBER level, which
is sufficient to construct the present reduced 2D-FES because
the major interaction energy between the carbohydrate and
surrounding environment is mainly electrostatic, and computations at the QM/MM RHF/3-21G/AMBER and QM/MM RHF/
6-31G*/AMBER levels produce nearly the same results (see
later discussion). After collecting sufficient QM/MM sampling
data, we analyzed the intramolecular electronic energy of the
SLex motif and the interaction energy between QM and MM
regions, and reconstructed the 2D-FES based on eq 9.
2.3.3. Shielding Tensor by Ab Initio QM/MM-GIAO Calculations. After constructing the QM/MM 2D-FES, we further
refined several numbers of the SLex geometries from the
minimum free energy region. By selecting 20 configurations of
SLex, we performed an annealing procedure to obtain solvated
selectin complexes. The sampled geometries from the minimum
free energy region have a time interval of longer than 100 ps
along the time evolution of the MD trajectories. The procedure
for this process is the same as that for X-ray structural refinement
as described above. Each solvated cluster contains the E-selectin/
SLex complex surrounded by 6000 water molecules extracted
from the whole solvated protein. For these large systems, we
performed ab initio QM/MM-GIAO calculations to predict
NMR chemical shifts.
In these calculations, we focus only on the chemical shifts
of the hydrogen bonding interaction between each sugar unit
and amino acid residue of selectin. Again, for simplicity, we
limited the QM region to only the SLex motif, and employed
the 6-31G* basis set for all QM/MM-GIAO computations. No
cutoff was introduced between QM and MM interactions. For
reference compounds, we chose tetramethylsilane (TMS) for1H
and13C chemical shifts. Relative shifts were evaluated using the
chemical shifts of the reference compounds optimized at RHF/
6-31G*.
3. Results and Discussion
3.1. QM/MM Refinement of X-ray Crystal Structure.
First, for reference in the following discussion, we summarize
the geometrical parameters of the E-selectin/SLex complex
bound to the CRD as determined by ab initio QM/MM structural
refinements. The calculated protein complex surrounded by a
solvent sphere is shown in Figure 2a. Backbone structures of
the E-selectin/SLex complex overlapped with the original X-ray
coordinates are shown in Figure 2b. As can be seen from Figure
2, there are no apparent structural deviations between original
X-ray and QM/MM-refined coordinates because we added
sufficient water molecules to prevent geometrical deformation
of the naive carbohydrate binding structure.

Ishida

Figure 3. Comparisons of results obtained by classical MM (x-axis)

and ab initio QM/MM (y-axis) models: (top) Electrostatic interaction
energies between the sialyl Lewis X ligand and the surrounding
environment (E-selectin and aqueous solvent). (bottom) Sialyl Lewis
X dipole moments. In both figures, red circles represent ab initio QM/
MM RHF/3-21G results and blue squares represent QM/MM RHF/631G* results. The rms deviations between 3-21G and 6-31G* basis
sets are as follows: 2.5 kcal/mol for the electrostatic interaction energy
and 0.3 D for the dipole moment.

Next, we consider the binding conformation of SLex at the

CRD. Characterization of molecular recognition is not a simple
problem, and here, we simply mention the hydrogen-bond
distances between each monosaccharide unit and amino acid
residue as one characteristic index. For the major hydrogen
bonds in the CRD, the detailed hydrogen bonding patterns of
QM/MM structures (two important interaction sites with Gal
and Fuc units) are shown in Figure 2c,d. The original X-ray
coordinates31 show the carboxyl group of Neu5Ac to be in close
contact with the side chains of Tyr48/Arg97. For Gal, the 4-OH
and 6-OH groups form hydrogen bonds with the hydroxyl group
of Tyr94 and the carboxyl side chain of Glu92, respectively.
Fuc is a member of the coordination ligand of the Ca2+ ion,
and is simultaneously surrounded by four amino acid residues
(Glu80, Asn82, Asn83, and Glu107). In comparing this and the
QM/MM-refined crystal structures, we observe that the overall
geometries of the two models are similar, although they deviate
slightly at several points. For the QM/MM-refined structure,
the Gal unit is relatively distorted inside the protein surface and
Fuc is more tightly bound to the Ca2+ binding site. Accordingly,
small changes in several hydrogen bonding distances are evident,
as summarized in Table 1. However, differences between the
two structures are small, which implies that ab initio QM/MM

Carbohydrate-Recognition Process in E-Selectin Complex

structural optimizations work well to refine well-defined crystallographic data.
Finally, we briefly mention the electronic structure of SLex.
For a reference of the following analyses, we calculated the
dipole moment of SLex at the conformation observed in the
X-ray structure; it shows 8.2 D at the QM/MM RHF/6-31G*/
AMBER level.
3.2. 2D Free Energy Profiles Obtained by MM and QM/
MM Models. As described in section 2.3.2, we performed all
atom MD simulations in a classical manner using an empirical
force field, and reoptimized the partial charges on each atomic
site of SLex to reproduce the QM/MM energy component
analysis. To confirm the validity of this modeling procedure,
we now consider the major differences observed between the
MM and QM/MM treatments.
In our 2D-FES, one reaction coordinate is the solvation
coordinate that describes the interaction energy between QM
(SLex) and MM (surrounding protein and aqueous solvent)
regions. Figure 3 (top) shows these calculated electrostatic
interaction energies for several hundred sampling structures. Due
mainly to the huge computational cost of directly applying ab
initio QM/MM computations to large numbers of sampled
structures, we employed the 3-21G basis set. To verify the
validity of using a set of such reduced size, we also performed
computations at the QM/MM-RHF/6-31G* level for the same
number of structures, and found the QM/MM electrostatic
interaction energies to be nearly the same. In comparing both
of these QM/MM energies with the MM charge model, we
observe a systematic deviation especially in the region where
QM/MM electrostatic interactions are large. This is reasonable,
since QM/MM treatment principally describes electronic polarization and enhanced polarity induces a large QM/MM
electrostatic interaction energy, whereas the point-charge approximation cannot generally describe this effect properly. Note
that, in contrast to earlier QM/MM simulations of simple organic
molecules in which the van der Waals parameters were adjusted
to reproduce experimental data,71 here we focus only on
electrostatic interactions and reoptimize the partial atomic
charges to reproduce the QM/MM results because the interaction
energy between protein and each sugar unit is mainly electrostatic. In our 2D-FES, the other reaction coordinate is the solute
coordinate that describes changes in the SLex conformation. For
this coordinate, we cannot directly compare the intramolecular
energies of the two theoretical methods, since each has different
energy units and meanings. Instead, we compare the dipole
moment of the solute molecule in both methods, since our
preliminary analysis clearly shows that changes in intramolecular
energy (equal to the conformational changes of SLex) are
roughly proportional to the dipole moment of the solute SLex.
Figure 3 (bottom) shows computed solute dipole moments for
hundreds of sampled structures. Again, both theoretical methods
produce nearly the same dipole moments within the range 2-10
D but deviate slightly where the dipole moments are small and
large, mainly due to loss of the collective polarization effect in
the MM point-charge approximation. In summary, these results
warrant the present approach to evaluating the 2D-FES by
mapping the classical trajectory onto the ab initio QM/MM
trajectory.
A next step is to consider the calculated 2D-FES based on
pure MM-MD results. In this model, the reduced FES is
characterized by two elements. Computational energy component analyses clearly reveal that the intramolecular energy of
the solute is roughly proportional to the total dipole moment of
SLex. This implies that, when the SLex ligand has a stable

J. Phys. Chem. B, Vol. 114, No. 11, 2010 3957

Figure 4. (top) Reduced 2D free energy surface obtained by the

classical MM treatment. The x-axis is the solvation coordinate (in kcal/
mol) defined by solute (sialyl Lewis X) and solvent (protein and aqueous
solvent) interactions; the y-axis is the solute coordinate (in kcal/mol)
defined by conformational changes in the carbohydrate chain. The z-axis
shows the relative free energy within these definitions (kcal/mol).
(bottom) Contour map of the 2D free energy surface. The x-axis is the
solvation coordinate (kcal/mol), and the y-axis is the solute coordinate
(kcal/mol). The contour space is 0.5 kcal/mol in both cases.

conformation with small intramolecular energy, it tends to have

a small dipole moment itself (and, therefore, less solvation
energy gain is obtained). On the other hand, to attain large
solvation stability, SLex must expend additional energy cost to
deform its molecular conformation to enhance the solute dipole,
and be solvated efficiently inside the polar protein environment.
Thus, the present model clearly implies that the binding stability
of the SLex ligand is determined by a detailed balance between
the degree of solvation energy and the energy penalty exacted
by conformational changes of SLex, and as a result, the essential
free energy profile should be characterized by at least two
reaction coordinates. On the basis of this definition, the
calculated 2D-FES and its contour map are shown in Figure 4
at the top and bottom, respectively. As can be seen from these
figures, the overall FES profile has a shallow energy minimum,
where various numbers of SLex configurations can coexist.
Next, we look at the same free energy profile determined by
ab initio QM/MM energy corrections performed by MD
trajectory-mapping. By adding the energy corrections (that is,
by replacing the MM energies with QM/MM data), we can
reconstruct the same free energy profile in reduced 2D coordinate space. The calculated QM/MM 2D-FES is shown in
Figure 5 (top), and its contour map is in Figure 5 (bottom).
The overall energy profile obtained by ab initio QM/MM
computations is similar to that obtained by MM; however, we

3958

J. Phys. Chem. B, Vol. 114, No. 11, 2010

Figure 5. (top) Reduced 2D free energy surface obtained by performing

ab initio QM/MM energy corrections. The x-axis is the solvation
coordinate (in kcal/mol) defined by solute (sialyl Lewis X) and solvent
(protein and aqueous solvent) interactions; the y-axis is the solute
coordinate (in kcal/mol) defined by conformational changes in the
carbohydrate chain. The z-axis shows the relative free energy within
these definitions (kcal/mol). (bottom) Contour map of the 2D free
energy surface. The x-axis is the solvation coordinate (kcal/mol), and
the y-axis is the solute coordinate (kcal/mol). The contour space is 0.5
kcal/mol in both cases.

observe an apparent difference in the energy scale of the

solvation coordinate when we compare both methods of the
energy profile. Due to the polarization effect of the solute QM
molecule, large amounts of QM/MM electrostatic interaction
energies are achieved, and as a result, the solvation coordinate
shows a larger negative energy scale. In addition, QM treatment
extracts less energy penalty to change the SLex conformation,
and the solute coordinate that describes the conformational
stability of SLex allows a wider energy scale than does the
related MM energy range in Figure 4.
In summary, these free energy analyses clearly demonstrate
that the binding conformation of the E-selectin/SLex complex
is well characterized by a broad free energy landscape with a
shallow energy minimum. To clearly capture the essence of
carbohydrate/glycan binding (or recognition), it is necessary now
to analyze the structural parameters of the carbohydrate
conformations observed in the free energy minimum region.
3.3. Structural Analyses of SLex Conformation on the 2DFES. On the basis of the 2D-FES, we next analyze the structure
of the SLex sugar chain. In this geometrical survey, we define
the binding conformation of the selectin complex as the sum
of the flexible SLex conformations coexisting in the free energy
minimum region. For practicality, we selected the following
energy regions (or windows) in which to sample the SLex
structures: solvation coordinate in the range from -360 to -290

Ishida
kcal/mol; solute coordinate in the range from 90 to 130 kcal/
mol. These selections are based on the MM 2D-FES results
shown in Figure 4. In this selected energy window, about 60%
of the total sampled structures from the MD trajectories can be
assigned as effective binding conformations. In the following
sections, we employ these sampling structures to analyze the
geometrical parameters that characterize the carbohydrate
bindings.
First, we determine the hydrogen bond distances between the
SLex motif and E-selectin protein. Figure 6 summarizes
representative hydrogen bond parameters between each monosaccharide unit and amino acid residue at the CRD. In the figures
four panels, the x-axis shows the sampling number from the
MD trajectory in accordance with the time evolution order. In
comparing the sampled structures with the known X-ray
geometry, we observe a few structural differences especially in
the Gal interaction site. While the 4-OH of Gal forms hydrogen
bonds with the hydroxyl group of Tyr94 and the carboxyl group
of Glu92 in the X-ray structure, the 6-OH of Gal, instead of
4-OH, bonds to the side chain of either Glu107 or Lys111 in
most MD-sampled structures. When the 6-OH of Gal forms a
stable hydrogen bond with the side chain of Glu107, hydrogen
bonds with Lys111 occasionally form (see Figure 6a). On the
other hand, when the 6-OH of Gal binds to the side chain of
Lys111, the other hydrogen bond with Glu107 disappears in
most simulation times. In short, the hydrogen bonds of Gal
couple loosely with target amino acid partners inside the CRD.
For the structural parameters of Fuc with E-selectin, we observe
basically the same trend between the X-ray and the sampled
structures. At the Ca2+ binding site, Fuc coordinates to the Ca2+
binding ligand and forms several tight and stable hydrogen
bonds with Glu80, Asn82, and Asn83, as summarized in Figure
6c,d. One slight difference is a weak coupling with the side
chain of the Glu107 residue; more than 70% of the sampled
structures form no apparent hydrogen bond with Glu107. In
contrast to the apparent hydrogen bonds formed between Gal/
Fuc and protein, we cannot observe any characteristic hydrogen
bonds between Neu5Ac and E-selectin, although the original
X-ray structure shows hydrogen bonds between the carboxyl
group of Neu5Ac and the protein. In the present model, the
acetyl group of GlcNAc occasionally forms a weak hydrogen
bond with the side chain of Arg108, as shown in Figure 6b.
Next, to clearly characterize the geometrical parameters
between the theoretical models and the X-ray structure, we
conduct ab initio QM/MM structural refinements for the selected
20 structures sampled from the minimum free energy region.
In this calculation, we explicitly consider the surrounding
aqueous environment of the E-selectin/SLex complex. After
annealing typical snapshots sampled from the minimum energy
region, which has at least an 100 ps time interval along the
MD trajectory, we perform ab initio QM/MM geometry
optimizations as described in section 2. The major geometrical
parameters for each structural model are listed in Table 1. As
clearly shown in these numerical data, most hydrogen bond
distances are within a reasonable range, as summarized in the
sampling data (see Figure 6). In checking each structure
carefully, we observe slight structural differences at several
points; for example, hydrogen bonding between Fuc and Glu107,
between GlcNAc and Arg108, and so on. These points are
consistent with the results summarized in Figure 6. In summary,
these annealed and quenched QM/MM structures seem to be
representative geometrical models observed in the free energy
minimum region. As a reference, a typical optimized geometry
is shown in Figure 7. In comparing both QM/MM-optimized

Carbohydrate-Recognition Process in E-Selectin Complex

J. Phys. Chem. B, Vol. 114, No. 11, 2010 3959

Figure 6. Interatomic distances: (a) between the 6-hydroxyl group of Gal and the side chains of Glu107/Lys111; (b) between the carbonyl oxygen
of NHAc in GlcNAc and the side chain of Arg108; (c) between the 2- and 3-hydroxyl groups of Fuc and the side chains of Asn83/Glu107; (d)
between the 4-hydroxyl group of Fuc and the side chains of Glu80/Asn82. In all graphs, the x-axis shows the number of sampled structures
extracted from the MD trajectory as a function of the time evolution order; the y-axis shows the interatomic distance between two heavy atoms (in
angstroms). In several cases, these parameters represent the hydrogen-bond distances between each monosaccharide unit and related amino acid
residues of the protein.

Figure 7. (left) QM/MM-refined geometry of the original X-ray coordinates at the carbohydrate-recognition domain (CRD) of selectin. (right) One
typical QM/MM optimized geometry (model no. 14 in Table 1) at the CRD, sampled and quenched from the minimum free energy region in the
2D-FES. Purple-labeled stick models represent amino acid residues that come into contact with the sialyl Lewis X; green balls represent Ca2+ ion.

Figure 8. Scatter plots of the dihedral angles (/) in the following glycosidic linkages: (a) Neu5Ac(2-3)-Gal; (b) Gal(1-4)-GlcNAc; (c)
Fuc(1-3)-GlcNAc.

models between the X-ray refined structure (left) and the

quenched model (right), we observe an apparent difference in
the molecular interaction between Gal and Glu107/Lys111. Also,

the position of Neu5Ac is shifted toward the solvent-accessible

region in the quenched models. As for these structural differences especially in the local binding site, possible reasons are

2.17

4.39

3.61

4.10

3.40

4.13

3.84

3.42

5.01
3.86
4.17
4.41
3.29
3.21
3.63
4.17

3.69

3.83

3.74

3.45

3.87

3.74

4.87
3.78
3.71
3.57
4.33
1.31
1.37
1.02

2.29

3.45

3.80

3.71

3.88

3.67

3.94

3.93

4.57
3.91
4.03
3.68
3.16
3.06
4.01
4.46

3.96

3.53

4.04

3.19

3.87

3.90

4.81
3.68
4.01
3.80
4.48
1.55
0.97
1.22

4.96
3.84
4.41
3.97
4.81
1.50
0.97
1.02

4.15

3.81

2.90

3.82

3.45

4.13

4.56
3.56
4.44
3.49
3.40
3.78
3.65
3.93

3.35

4.16

4.11

3.28

3.63

3.34

3.72

2.73

1.44

geom
2

4.87
3.68
3.84
3.73
4.23
1.02
1.22
1.30

3.97

4.05

3.28

3.76

3.66

3.91

4.92
3.81
3.65
3.75
3.50
3.64
3.74
3.90

3.34

3.93

4.53

3.04

3.82

3.47

3.47

2.64

1.97

geom
3

4.79
3.51
3.96
3.90
4.44
1.33
1.14
1.13

3.71

3.96

2.83

4.02

3.61

3.81

4.19
3.59
3.98
3.47
3.37
3.74
3.63
3.85

3.76

3.87

3.90

3.26

4.07

3.33

3.22

2.37

2.01

geom
4

4.80
3.60
3.90
3.96
4.19
1.26
1.15
1.54

3.68

3.84

3.23

3.59

3.41

3.95

4.65
3.57
3.93
4.48
3.41
3.99
3.56
4.29

3.42

3.77

3.82

3.10

3.74

3.73

4.54

2.00

2.01

geom
5

4.67
3.50
3.98
3.82
4.63
0.90
1.28
1.18

3.88

3.81

3.47

3.60

3.47

4.08

4.30
3.47
3.86
4.95
3.51
3.57
3.53
4.01

3.58

3.83

3.51

2.91

3.81

3.26

4.36

2.24

1.81

geom
6

4.73
3.76
3.90
3.98
4.46
1.08
1.44
0.89

3.73

3.86

3.29

3.75

3.59

3.68

4.57
3.56
3.96
4.96
3.49
3.55
3.77
4.28

3.59

4.04

3.96

2.76

3.84

3.38

4.38

2.21

1.74

geom
7

4.81
3.66
3.74
3.71
4.47
1.74
0.92
1.04

4.13

3.57

3.41

4.07

3.90

3.62

4.62
3.78
4.30
3.47
2.91
3.55
3.80
4.31

3.49

4.11

4.39

3.18

3.31

3.76

4.01

2.52

1.78

geom
8

geom0 indicates the QM/MM-refined geometry of X-ray coordinates (see Table 1).

1.79

geom
1

2.15

geom
0a

5.06
3.89
3.96
3.84
4.63
1.65
0.71
1.17

3.63

3.89

3.45

4.19

4.05

3.92

4.30
3.75
4.10
3.32
3.62
3.79
3.76
4.13

3.38

3.80

4.03

3.32

4.01

3.51

3.11

2.30

1.81

geom
9

4.98
3.57
3.77
3.64
4.38
1.06
1.04
1.38

4.00

3.62

3.35

3.89

3.63

3.71

4.23
4.31
4.01
3.73
3.32
3.14
3.79
3.95

3.87

4.02

4.18

3.51

4.12

3.83

3.34

2.38

1.98

geom
10

4.72
3.55
4.04
3.64
4.32
1.10
1.11
1.34

3.86

3.71

3.40

3.92

3.83

4.23

4.51
3.61
4.26
3.59
3.22
3.54
3.70
4.33

4.14

4.06

4.08

3.44

4.29

3.65

3.12

2.73

1.66

geom
11

4.99
3.65
3.89
3.93
4.52
1.55
0.91
0.95

4.03

3.83

2.78

3.71

3.65

3.72

4.29
3.45
3.78
4.18
3.38
3.68
3.72
4.06

3.85

4.14

3.69

3.74

5.15

3.31

3.53

2.28

1.77

geom
12

5.17
3.61
3.83
3.70
4.48
1.50
0.90
0.98

3.88

3.79

3.26

4.14

3.80

4.25

4.39
3.46
4.06
3.63
3.11
3.60
3.93
3.98

3.65

3.60

4.19

3.57

3.30

3.45

3.81

2.51

2.03

geom
13

4.66
3.57
4.75
3.91
5.03
0.78
1.20
1.36

4.11

3.64

3.07

4.03

3.54

4.07

4.48
3.36
3.80
4.12
3.47
3.50
3.47
4.02

3.54

3.78

3.63

4.19

5.02

3.31

4.13

2.00

1.68

geom
14

4.98
3.85
4.18
3.81
4.71
1.40
1.21
0.83

4.08

3.67

3.21

4.00

3.28

3.73

4.74
3.50
3.79
4.07
3.50
3.51
3.53
4.07

3.63

3.82

4.22

3.69

4.72

3.58

3.74

1.83

1.97

geom
15

4.78
3.74
4.16
3.80
4.67
1.37
1.28
0.89

3.63

3.61

3.07

4.04

3.23

3.62

4.84
3.64
3.81
4.92
3.03
3.69
3.55
4.09

4.13

3.89

4.13

3.84

4.47

3.12

4.16

2.05

1.48

geom
16

4.99
3.62
4.31
3.79
4.57
1.21
1.40
0.94

3.72

3.62

3.35

3.69

3.78

3.73

4.34
3.47
3.83
4.81
3.38
3.51
3.52
4.23

3.75

4.14

3.54

3.27

3.73

3.51

4.48

2.32

1.91

geom
17

4.80
3.77
4.24
4.03
4.65
1.42
1.19
1.27

3.63

4.15

2.81

4.13

3.49

3.77

4.41
3.61
3.52
3.89
3.27
3.66
3.71
3.95

3.50

3.77

3.48

3.16

4.51

3.38

3.76

2.14

1.94

geom
18

4.82
3.71
4.31
3.82
4.86
1.25
1.04
1.41

4.02

3.74

3.09

3.88

3.66

3.71

3.94
3.72
4.18
3.52
2.99
3.45
3.56
4.38

3.61

3.59

3.01

4.03

3.38

3.43

4.49

2.45

1.68

geom
19

4.68
3.57
4.39
3.83
4.76
1.53
1.04
1.35

4.23

3.80

3.17

4.00

3.60

3.84

4.26
3.61
3.05
3.65
3.25
3.56
3.68
3.98

4.07

3.80

3.17

3.81

4.56

3.50

3.44

2.63

1.83

geom
20

4.86
3.67
4.06
3.82
4.56
1.30
1.13
1.15

3.89

3.79

3.19

3.90

3.62

3.86

4.48
3.63
3.92
4.02
3.32
3.58
3.66
4.10

3.65

3.90

3.89

3.42

4.08

3.47

3.86

2.33

1.81

average

J. Phys. Chem. B, Vol. 114, No. 11, 2010

H3ax
(Neu5Ac)
H3eq
(Neu5Ac)
H4
(Neu5Ac)
H5
(Neu5Ac)
H6
(Neu5Ac)
H7
(Neu5Ac)
H8
(Neu5Ac)
H9
(Neu5Ac)
H9
(Neu5Ac)
H1 (Gal)
H2 (Gal)
H3 (Gal)
H4 (Gal)
H5 (Gal)
H6 (Gal)
H6 (Gal)
H1
(GlcNAc)
H2
(GlcNAc)
H3
(GlcNAc)
H4
(GlcNAc)
H5
(GlcNAc)
H6
(GlcNAc)
H6
(GlcNAc)
H1 (Fuc)
H2 (Fuc)
H3 (Fuc)
H4 (Fuc)
H5 (Fuc)
H6 (Fuc)
H6 (Fuc)
H6 (Fuc)

proton type

calculated proton chemical shifts (in ppm)

TABLE 2: Calculated Proton Chemical Shifts of Each QM/MM Optimized Structure by the Ab Initio QM/MM-GIAO RHF/6-31G*/AMBER Level (in ppm)

3960
Ishida

Carbohydrate-Recognition Process in E-Selectin Complex

J. Phys. Chem. B, Vol. 114, No. 11, 2010 3961

Figure 9. Left panel: Superimposed geometries of the sialyl Lewis X at the carbohydrate-recognition domain. Ten QM/MM optimized structures
of SLex are superimposed onto one reference structure. Only the SLex structures are shown by the stick model. Right panels: (top) Experimental
NMR profile reported in ref 29. This original figure was provided by Prof. Thomas Peters. (bottom) Computed 1D proton chemical shift profile of
the SLex ligand. This profile was calculated by the simple average of 20 QM/MM-refined geometries summarized in Tables 1 and 2.

(1) an inaccuracy of the current force field for carbohydrate

complex, (2) a treatment of the highly charged system as well
as the protonation state, and (3) an insufficient statistical
sampling of the available phase space. However, a major reason
for these structural deviations is interaction between the hydroxyl
groups of SLex and the bulk aqueous phase, which is missing
in the X-ray crystal environment.
Next, we consider another important parameter that characterizes complex sugar chain conformations; the conventional
dihedral angles at the glycosidic linkages between each monosaccharide unit. For all of the sampled structures obtained from
the minimum free energy region, distributions of each dihedral
angle are summarized in Figure 8. Comparing these three types
of distributions, we observe that the glycosidic linkage between
Neu5Ac and Gal shows a broader distribution than do the other
two linkages, clearly implying that the Neu5Ac unit is relatively
flexible and has more free space to move. This point is consistent
with the geometrical analysis, which shows that there are no
apparent hydrogen bond linkages between proteins. The optimized values of each dihedral angle inside the 20 optimized
complex structures are listed in Table 1. As can be seen from
these results, there are slight deviations from the mean value
for each glycosidic linkage, although all data are obtained from
the QM/MM-refined geometries. Again, these results clearly
indicate that SLex does not have a tight and rigid conformation
at the CRD but a rather flexible conformation that allows small
structural transitions. When we check the details in three parts
of glycosidic linkages, the angles between Neu5Ac and Gal
show the largest deviations from the mean values, consistent
with the distribution summarized in Figure 8.
3.4. 1D-NMR Profile of E-Selection/SLex Complex. By
introducing FES in reduced 2D coordinate space and defining
the binding conformation as the ensemble average in the
minimum free energy region, we obtain a set of computational
models of the SLex sugar chain. To validate the reliability of
this method, we check the NMR chemical shifts on each atomic
site of SLex, which characterize the complex sugar chain
conformations. In principle, we could perform ab initio QM/
MM-GIAO calculations for all collected trajectories and estimate
the desired chemical shifts by averaging ab initio data on each

atomic site. However, the large computational costs involved

make this impractical. Also, it is generally recognized that the
(downfield proton) chemical shifts correlated more directly with
the C-H (or N-H and O-H) distances than with the surrounding environment that polarizes the QM region.49,57 These earlier
observations suggest that chemical shifts computed using
geometries determined from empirical force fields may reflect
a systematic deviation from ab initio based results. Therefore,
in this paper, we employ the optimized QM/MM geometries
listed in Table 1 to compute the proton chemical shifts; such
calculated proton chemical shifts for 20 structures are listed in
Table 2. In comparing the available experimental data for the
E-selectin/SLex complex,29 we focus only on selected protons
attached to the carbohydrate carbon skeleton.
As can be seen from these data, deviations (or fluctuations)
from the mean values of each chemical shift are mostly small,
typically in the range 0.2-0.4 ppm. To determine the correlation
between structural fluctuation and chemical shift, we superimpose the QM/MM-refined geometries onto one typical reference,
as shown in Figure 9 (left). As this figure clearly shows, the
binding conformation of SLex is clearly not tight and rigid but
flexible especially in the solvent-exposed region. As discussed
in section 3.3, Fuc forms stable hydrogen bonds with the Ca2+
binding site and Gal interacts moderately with the protein. In
reflection to these trajectory analyses, Fuc and Gal show small
structural fluctuations, while Neu5Ac and GlcNAc show
relatively larger structural motions.
By averaging calculated chemical shifts for QM/MM-refined
geometries, we obtain the 1D-NMR spectral profile shown in
Figure 9 (right panels, bottom). In comparing theoretical results
with available experimental information in Figure 9 (right
panels, top),29 we observe that the calculated 1D-NMR profile
qualitatively reproduces recent experimental results within the
accuracy of the method employed (ab initio QM/MM-RHF/631G* level). Although most proton chemical shifts are reasonably assigned by QM/MM-GIAO averaging, a few points
deviate slightly; in particular, the calculated proton peaks for
H1F and H3Neq are not as far downfield as the experimental
values, although the deviations between computations and
experiments are small (0.2-0.3 ppm). These deviations are

3962

J. Phys. Chem. B, Vol. 114, No. 11, 2010

Ishida

Figure 10. Sialyl Lewis X conformations on the 2D free energy surface. In region A, SLex has energetically unstable conformations with large
dipole moments. In region C, SLex has energetically stable conformations with small dipole moments, resulting in a small solvation energy gain.
In region B, as a result of detailed balance between the relative stability of the SLex conformation and the solvation energy gain, energetically
favorable conformations are observed. Several representative geometries sampled from the MD trajectory are shown in the three regions as
superimposed structures.

observed especially when monosaccharide units are exposed to

the solvent-accessible region and have a relatively high flexibility in its structure. In summary, the calculated 1D-NMR
profile reasonably reproduces the experimental trend, and this
result clearly confirms the validity of our systematic computational modeling strategy.
3.5. Discussion: Insight into Experimental Observation.
In this paper, we proposed a systematic computational modeling
strategy that identifies complex carbohydrate conformations on
the CRD of selectin. As noted in the Introduction, the molecular
basis of lectin-carbohydrate interactions is not well understood.
Although experimental studies to determine the E-selectin/SLex
binding structures have been reported,26-31 problems remain
regarding ligand-binding affinities among selectin families.10,15-17
Therefore, to obtain detailed insight into carbohydrate recognition, finally we consider the characteristic findings clarified by
the present model.
To capture the flexible SLex conformation inside the CRD,
we first introduced the reduced 2D-FES defined by two reaction
coordinates. Analysis of molecular geometries sampled in this
2D-FES clarifies the relationship between sugar-chain conformation and solvation stability, as shown in Figure 10. The
present model determines the strength of carbohydrate recognition (that is, the affinity of sugar-chain binding) by means of
two conflicting factors: the degree of solute-solvent (in this
case, SLex and solvated protein environment) interaction and
the conformational stability of the solute SLex itself. Although
SLex is a short-chain sugar, it is flexible enough to transform
its structure by deforming its monosaccharide carbon skeleton
and rotating its dihedral angles around the glycosidic linkages
and around the many hydroxyl groups attached to each
monosaccharide unit. As a result, many configurations of SLex

can be observed on the reduced 2D-FES. To stably bind to the

CRD of selectin, SLex must gain a large amount of solvation
energy, which it can do if it has a large dipole moment. Indeed,
it has such a large dipole moment when all of its hydroxyl
groups are directed (rotated) in the same direction with parallel
orientation (Figure 10, region A). This extended conformation
lends it a large amount of intramolecular energy (that is, the
conformation of SLex is energetically unfavorable). In contrast,
in its relaxed, stable conformation, each hydroxyl group orients
in the opposite direction with antiparallel orientation (Figure
10, region C), and it has only a small dipole moment and thus
less solvation stability. Favorable SLex binding conformations
are observed in the middle point between regions A and C,
where the orientation of hydroxyl groups is distorted and the
glycosidic linkages are flexible, allowing conformational changes
around the minimum free energy region.
As noted in previous sections, we identified the average
binding conformations shown in Figure 10, region B, as actual
molecular structures in many experimental studies. In comparing
hydrogen-bond distances from the X-ray structure and the QM/
MM-refined models (see Table 1), we observe that molecular
interactions between the Fuc unit and Ca2+ binding site are
similar; the 3- and 4-hydroxyl groups of Fuc are coordinated
to the Ca2+ ion and form hydrogen bonds with several amino
acid residues (Glu80, Asn82, and Asn83), which are members
of other coordination ligands of Ca2+ ion. For other molecular
interactions, we observe several deviations between the X-ray
structure and theoretical models. In the X-ray structure, Gal
forms moderate hydrogen bonds with Glu92 and Tyr94; in the
QM/MM-optimized model, it forms hydrogen bonds with
Glu107 and Lys111. For Neu5Ac, there are no important
hydrogen bonds in the binding conformation, although the X-ray

Carbohydrate-Recognition Process in E-Selectin Complex

structure shows close contact with Tyr48 and Arg97.31 These
results are consistent with the distributions of dihedral angles
at the glycosidic linkage summarized in Figure 8. The present
model clearly shows the flexibility of the Neu5Ac unit, which
is located on the solvent-exposed region. Thus, our present
theoretical study reasonably supports the proposition that the
Fuc unit is key to electrostatic binding of carbohydrates,31 and
also suggests that variation among other selectins is determined
by interactions with other sugar units.
In comparing calculated and experimental 1D NMR profiles,
we observe that most proton chemical shifts are in good
agreement.29 NMR experiments are usually performed in the
range 500-600 MHz and observed profiles are formulated by
long-time averaging.8,9,11 To replicate this ensemble averaging
in our theoretical model, we compute chemical shifts by
averaging the sampled structures extracted from the minimum
free energy region on the 2D-FES. Again, consistency between
theoretical and experimental profiles confirms the validity of
our modeling strategy. On the basis of saturation transfer
difference (STD) NMR results, Rinnbauer et al. suggested that
H4 and H6 protons of Gal are in close contact with the protons
of E-selectin.29 This is consistent with our theoretical model
where these protons are surrounded by several amino acid
residues; H4 proton is close to the side chains of Tyr94 and
Arg97, and H6 is close to the side chains of Tyr94, Asn105,
and Lys111. Also, the STD-NMR profile predicts that H2 and
H4 protons of Fuc interact with the side chains of amino acid
residues in the binding pocket, and the C6 methyl group points
away from the surface of the binding pocket. These experimental
observations are confirmed by the present modeling results
where H2 and H4 protons are located around the Ca2+ binding
site and C6 methyl is oriented away from the binding site. In
addition, proton shifts deviate from experimental results for
H3Neq, H1GN, and H1F, in that the experimental shifts are slightly
further downfield than the calculated shifts. One reason for this
deviation may be SLex structural fluctuations in the solventaccessible region, because all of these protons are located on
the solvent-exposed site, distant from the tight binding position.
However, the deviations are generally small (0.3-0.4 ppm),
considering the level of computation, and the model gives a
reasonable profile.
In summary, the modeling strategy presented in this paper
predicts reasonable binding conformations. Comparison of
experimental and theoretical results verifies that the computational model provides insight into deviations observed in several
experimental studies.
4. Conclusion
In this Article, we investigated the details of carbohydrate
recognition in the complex of E-selectin with sialyl Lewis X
by employing a systematic computational modeling strategy.
We defined the reduced 2D free energy surface that characterizes
changes in carbohydrate conformation as well as the degree of
solvation stability. Then, we estimated the overall profile of the
free energy surface by MD simulations combined with ab initio
QM/MM energy corrections. By mapping the complex carbohydrate structures onto the QM/MM 2D-FES, we successfully
characterized the details of molecular interactions in the local
binding structure. In our theoretical analysis, the carbohydrate
conformation is defined by both parameters: (a) distribution of
the glycosidic linkage between each sugar unit and (b) rotational
orientation of each hydroxyl group attached to the four sugar
units. When we focus only on the glycosidic linkage, the major
conformation seems to be limited. However, as clarified in this

J. Phys. Chem. B, Vol. 114, No. 11, 2010 3963

study, a more important parameter in this system is the
orientation of several hydroxyl groups attached to the sugar ring.
By the combinations of both parameters, the carbohydrate chain
can actually hold many configurations which are energetically
nearly degenerate, and can bind to the target lectin with nearly
the same binding energy.
As for the carbohydrate recognition, the major findings are
summarized as follows. (1) In the Neu5Ac unit, there are no
apparent hydrogen bonds around the carboxyl group, and this
unit is more flexible along the glycosidic linkage than are any
of the other three sugar units. (2) In the Gal unit, the 4-OH
group is always located near the hydroxyl group of Tyr94,
although no apparent hydrogen bonds are formed. The major
molecular interactions are the hydrogen bonds between the 6-OH
group and the side chains of Glu107 and Lys111. (3) In the
GlcNAc unit, there is no apparent close molecular contact
between the hydroxyl groups of the sugar unit and the target
amino acid residues. However, the side chain of the NHAc group
is relatively flexible, and occasionally forms a hydrogen bond
with the side chain of Arg108. (4) In the Fuc unit, rather stable
hydrogen bonds form because of the three types of hydroxyl
groups. The 2-OH group can form a moderately strong hydrogen
bond with the side chain of Asn83, the 3-OH group occasionally
forms a hydrogen bond with Glu107, and the 4-OH group
always forms tight and stable hydrogen bonds with Glu80 and
Asn82. Although several local site interactions differ from the
static picture predicted from the X-ray crystal coordinates, the
estimated chemical shift profile computed from an average of
20 QM/MM optimized geometries shows reasonable agreement
with the results of recent saturation transfer difference NMR
experiments. These results clearly demonstrate the validity of
the present calculation procedure. In our theoretical analysis,
no particular forms of solvent interaction are observed. This
means that most water molecules act as a bulk aqueous
environment.
One of the most important findings clarified in this study is
that the binding geometries of carbohydrates with selectin are
determined not by one single, rigid sugar-chain conformation
structure but rather by the average of several conformations that
fluctuate around the minimum free energy region. The present
systematic computational modeling strategy thus clarifies the
mechanism of selectin specificity and provides a systematic way
to explore the atomistic details of carbohydrate recognition.
Acknowledgment. The author is grateful to Prof. Kazuo
Kitaura for his continuous encouragement and also wishes to
thank Prof. Thomas Peters for providing us the original NMR
data. This work was supported by a Grant-in-Aid for Young
Scientists from MEXT, KAKENHI (no. 19750021), and the
Next Generation Super Computing Project, Nanoscience Program, MEXT, Japan. Part of the numerical calculations were
performed at the Computer Center of the Institute for Molecular
Science (IMS). Finally, the author would like to thank Clive S.
Langham and Braian K. Purdue for reading the manuscript
carefully, and the anonymous reviewers for their useful comments.
References and Notes
(1) Varki, A., Cummings, R. D., Esko, J. D., Freeze, H. H., Stanley,
P., Bertozzi, C. R., Hart, G. W., Etzler, M. E., Eds. Essentials of
Glycobiology, 2nd ed.; Cold Spring Harbor Laboratory Press: New York,
2008.
(2) Taylor, M. E.; Drickmer, K. Introduction to Glycobiology, 2nd ed.;
Oxford University Press: New York, 2006.
(3) Dwek, R. A. Chem. ReV. 1996, 96, 683720.
(4) Rini, J. M. Annu. ReV. Biophys. Biomol. Struct. 1995, 24, 551
577.

3964

J. Phys. Chem. B, Vol. 114, No. 11, 2010

(5) Weis, W. I.; Drickamer, K. Annu. ReV. Biochem. 1996, 65, 441
473.
(6) Woods, R. J. Curr. Opin. Struct. Biol. 1995, 5, 591598.
(7) Imberty, A. Curr. Opin. Struct. Biol. 1997, 7, 617623.
(8) Duus, J. .; Gotfredsen, C. H.; Bock, K. Chem. ReV. 2000, 100,
45894614.
(9) Wormald, M. R.; Petrescu, A. J.; Pao, Y.-L.; Glithero, A.; Elliott,
T.; Dwek, R. A. Chem. ReV. 2002, 102, 371386.
(10) Simanek, E. E.; McGarvey, G. J.; Jablonowski, J. A.; Wong, C.H. Chem. ReV. 1998, 98, 833862.
(11) Vliegenthart, J. F. G., Woods, R. J., Ed. NMR Spectroscopy and
Computer Modeling of Carbohydrates; ACS Symposium Series 930;
American Chemical Society: Washington, DC, 2006.
(12) Woods, R. J. Glycoconjugate J. 1998, 15, 209216.
(13) DeMarco, M. L.; Woods, R. J. Glycobiology 2008, 18, 426440.
(14) Drickamer, K. Curr. Opin. Struct. Biol. 1999, 9, 585590.
(15) Vestweber, D.; Blanks, J. E. Physiol. ReV. 1999, 79, 181213.
(16) Ley, K. Trends Mol. Med. 2003, 9, 263268.
(17) Ehrhardt, C.; Kneuer, C.; Bakowsky, U. AdV. Drug DeliVery ReV.
2004, 56, 527549.
(18) Kneuer, C.; Ehrhardt, C.; Radomski, M. W.; Bakowsky, U. Drug
DiscoVery Today 2006, 11, 10341040.
(19) Phillips, M. L.; Nudelman, E.; Gaeta, F. C. A.; Perez, M.; Singhal,
A. K.; Hakomori, S.; Paulson, J. C. Science 1990, 250, 11301132.
(20) Lin, Y.-C.; Hummel, C. W.; Huang, D.-H.; Ichikawa, Y.; Nicolaou,
K C.; Wong, C.-H. J. Am. Chem. Soc. 1992, 114, 54525454.
(21) Varki, A. Proc. Natl. Acad. Sci. U.S.A. 1994, 91, 73907397.
(22) Brandley, B. K.; Kiso, M.; Abbas, S.; Nikrad, P.; Srivasatava, O.;
Foxall, C.; Oda, Y.; Hasegawa, A. Glycobiology 1993, 3, 633641.
(23) Asa, D.; Raycroft, L.; Ma, L.; Aeed, P. A.; Kaytes, P. S.; Elhammer,
A. P.; Geng, J.-G. J. Biol. Chem. 1995, 270, 1166211670.
(24) Sanders, W. J.; Katsumoto, T. R.; Bertozzi, C. R.; Rosen, S. D.;
Kiessling, L. L. Biochemistry 1996, 35, 1486214867.
(25) Koenig, A.; Jain, R.; Vig, R.; Norgard-Sumnicht, K. E.; Matta,
K. L.; Varki, A. Glycobiology 1997, 7, 7993.
(26) Scheffler, K.; Ernst, B.; Katopodis, A.; Magnani, J. L.; Wang, W. T.;
Weisemann, R.; Peters, T. Angew. Chem., Int. Ed. Engl. 1995, 34, 1841
1844.
(27) Poppe, L.; Brown, G. S.; Philo, J. S.; Nikrad, P. V.; Shah, B. H.
J. Am. Chem. Soc. 1997, 119, 17271736.
(28) Harris, R.; Kiddle, G. R.; Field, R. A.; Milton, M. J.; Ernst, B.;
Magnani, J. L.; Homans, S. W. J. Am. Chem. Soc. 1999, 121, 25462551.
(29) Rinnbauer, M.; Ernst, B.; Wagner, B.; Magnani, J.; Benie, A. J.;
Peters, T. Glycobiology 2003, 13, 435443.
(30) Graves, B. J.; Crowther, R. L.; Chandran, C.; Rumberger, J. M.;
Li, S.; Huang, K.-S.; Presky, D. H.; Familletti, P. C.; Wolitzky, B. A.; Burns,
D. K. Nature 1994, 367, 532538.
(31) Somers, W. S.; Tang, J.; Shaw, G. D.; Camphausen, R. T. Cell
2000, 103, 467479.
(32) Beauharnois, M. E.; Lindquist, K. C.; Marathe, D.; Vanderslice,
P.; Xia, J.; Matta, K. L.; Neelamegham, S. Biochemistry 2005, 44, 9507
9519.
(33) Simonis, D.; Fritzsche, J.; Alban, S.; Bendas, G. Biochemistry 2007,
46, 61566164.
(34) Torgersen, D.; Mullin, N. P.; Drickamer, K. J. Biol. Chem. 1998,
273, 62546261.
(35) Bouyain, S.; Rushton, S.; Drickamer, K. Glycobiology 2001, 11,
989996.
(36) Pichierri, F.; Matsuo, Y. Bioorg. Med. Chem. 2002, 10, 27512757.
(37) Veluraja, K.; Margulis, C. J. J. Biomol. Struct. Dyn. 2005, 23, 101
111.
(38) Breneman, C. M.; Wiberg, K. B. J. Comput. Chem. 1990, 11, 361
373.
(39) Bayly, C. I.; Cieplak, P.; Cornell, W. D.; Kollman, P. A. J. Phys.
Chem. 1993, 97, 1026910280.

Ishida
(40) Haselhorst, T.; Weimar, T.; Peters, T. J. Am. Chem. Soc. 2001,
123, 1070510714.
(41) Field, M. J. J. Comput. Chem. 2002, 23, 4858.
(42) Gao, J.; Truhlar, D. G. Annu. ReV. Phys. Chem. 2002, 53, 467
505.
(43) Warshel, A. Annu. ReV. Biophys. Biomol. Struct. 2003, 32, 425
443.
(44) Ryde, U. Curr. Opin. Chem. Biol. 2003, 7, 136142.
(45) Friesner, R. A.; Guallar, V. Annu. ReV. Phys. Chem. 2005, 56, 389
427.
(46) Riccardi, D.; Schaefer, P.; Yang, Y.; Yu, H.; Ghosh, N.; Prat-Resina,
X.; Konig, P.; Li, G.; Xu, D.; Guo, H.; Elstner, M.; Cui, Q. J. Phys. Chem.
B 2006, 110, 64586469.
(47) Senn, H. M.; Thiel, W. Angew. Chem., Int. Ed. 2009, 48, 1198
1229.
(48) (a) Ishida, T.; Kato, S. J. Am. Chem. Soc. 2003, 125, 1203512048.
(b) Ishida, T.; Kato, S. J. Am. Chem. Soc. 2004, 126, 71117118.
(49) Ishida, T. Biochemistry 2006, 45, 54135420.
(50) Ishida, T. J. Chem. Phys. 2008, 129, 125105.
(51) (a) Rosta, E.; Klahn, M.; Warshel, A. J. Phys. Chem. B 2006, 110,
29342941. (b) Rosta, E.; Haranczyk, M.; Chu, Z. T.; Warshel, A. J. Phys.
Chem. B 2008, 112, 56805692.
(52) (a) Lu, Z.; Yang, W. J. Chem. Phys. 2004, 121, 89100. (b) Hu,
H.; Lu, Z.; Yang, W. J. Chem. Theory Comput. 2007, 3, 390406.
(53) (a) Rod, T. H; Ryde, U. J. Chem. Theory Comput. 2005, 1, 1240
1251. (b) Kaukonen, M.; Soderhjelm, P.; Heimdal, J.; Ryde, U. J. Chem.
Theory Comput. 2008, 4, 9851001.
(54) (a) Takahashi, H.; Matubayashi, N.; Nakahara, M.; Nitta, T.
J. Chem. Phys. 2004, 121, 39893999. (b) Takahashi, H.; Ohno, H.; Kishi,
R.; Nakano, M.; Matubayashi, N. J. Chem. Phys. 2008, 129, 205103.
(55) King, G.; Warshel, A. J. Chem. Phys. 1990, 93, 86828692.
(56) Kumar, S.; Rosenberg, J. M.; Bouzida, D.; Swendsen, R. H;
Kollman, P. A. J. Comput. Chem. 1995, 16, 13391350.
(57) Cui, Q.; Karplus, M. J. Phys. Chem. B 2000, 104, 37213743.
(58) Ditchfield, R. Mol. Phys. 1974, 27, 789807.
(59) Wolinski, K.; Hinton, J. F.; Pulay, P. J. Am. Chem. Soc. 1990, 112,
82518260.
(60) Helgaker, T.; Jaszunski, M.; Ruud, K. Chem. ReV. 1999, 99, 293
352.
(61) Dupuis, M.; Marquez, A.; Davidson, E. R. HONDO Quantum
Chemistry Program Exchange (QCPE), Indiana University, Bloomington,
IN.
(62) Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Merz, K. M.,
Jr.; Ferguson, D. M.; Spellmeyer, D. C.; Fox, T.; Caldwell, J. W.; Kollman,
P. A. J. Am. Chem. Soc. 1995, 117, 51795197.
(63) Kollman, P.; Dixon, R.; Cornell, W.; Fox, T.; Chipot, C.; Pohorille,
A. In Computer Simulation of Biomolecular Systems; van Gunsteren, W. F.,
Weiner, P. K., Wilkinson, A. J., Eds.; Kluwer Academic Publishers: Leiden,
The Netherlands, 1997; Vol. 3.
(64) Woods Group (2005-2009) GLYCAM Web. Complex Carbohydrate Research Center, University of Georgia, Athens, GA (http://www.glycam.com).
(65) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.;
Klein, M. L. J. Chem. Phys. 1983, 79, 926935.
(66) Frenkel, D.; Smit, B. Understanding Molecular Simulation; Academic Press: London, 1996.
(67) Martyna, G. J.; Klein, M. L.; Tuckerman, M. J. Chem. Phys. 1992,
97, 26352643.
(68) Cheng, A.; Merz, K. M., Jr. J. Phys. Chem. 1996, 100, 19271937.
(69) Tuckerman, M.; Berne, B. J.; Martyna, G. J. J. Chem. Phys. 1992,
97, 19902001.
(70) Martyna, G. J.; Tuckerman, M. E.; Tobias, D. J.; Klein, M. L. Mol.
Phys. 1996, 87, 11171157.
(71) (a) Freindorf, M.; Gao, J. J. Comput. Chem. 1996, 17, 386395.
(b) Gao, J. J. Phys. Chem. B 1997, 101, 657663.

JP905872T