Improved cyclodextrin-based receptors for camptothecin by inverse virtual screening

Article abstract of DOI:10.1002/chem.200700661

We report the computeraided optimization of a synthetic receptor for a given guest molecule, based on inverse virtual screening of receptor libraries. As an example, a virtual set of β-cyclodextrin (β-CD) derivatives was generated as receptor candidates for the anticancer drug camptothecin. We applied the two docking tools AutoDock and GlamDock to generate camptothecin complexes of every candidate receptor. Scoring functions were used to rank all generated complexes. From the 10% top-ranking candidates nine were selected for experimental alidation. They were synthesized by reaction of heptakis-[6-deoxy-6-iodo]-β-CD with a thiol compound to form the hepta-substituted β-CDs. The stabilities of the camptothecin complexes obtained from solubility measurements of five of the nine CD derivatives were significantly higher than for any other CD derivative known from literature. The remaining four CD derivatives were insoluble in water. In addition, corresponding mono-substituted CD derivatives were synthesized that also showed improved binding constants. Among them the 9-H-purine derivative was the best, being comparable to the investigated hepta-substituted β-CDs. Since the measured binding free energies correlated satisfactorily with the calculated scores, the applied scoring functions appeared to be appropriate for the selection of promising candidates for receptor synthesis.

Full text of DOI:10.1002/chem.200700661

FULL PAPER
DOI: 10.1002/chem.200700661
Improved Cyclodextrin-Based Receptors for Camptothecin by Inverse
Virtual Screening
Andreas Steffen,[a] Carolin Thiele,[b] Simon Tietze,[c] Christian Strassnig,[b]
Andreas Kämper,[a] Thomas Lengauer,[a] Gerhard Wenz,*[b] and Joannis Apostolakis*[c]
Abstract: We report the computer-
aided optimization of a synthetic recep-
tor for a given guest molecule, based
on inverse virtual screening of receptor
libraries. As an example, a virtual set
of b-cyclodextrin (b-CD) derivatives
was generated as receptor candidates
for the anticancer drug camptothecin.
We applied the two docking tools
validation. They were synthesized by
reaction of heptakis-[6-deoxy-6-iodo]-
b-CD with a thiol compound to form
the hepta-substituted b-CDs. The sta-
bilities of the camptothecin complexes
obtained from solubility measurements
of five of the nine CD derivatives were
significantly higher than for any other
CD derivative known from literature.
The remaining four CD derivatives
were insoluble in water. In addition,
corresponding mono-substituted CD
derivatives were synthesized that also
showed improved binding constants.
Among them the 9-H-purine derivative
was the best, being comparable to the
investigated hepta-substituted b-CDs.
Since the measured binding free ener-
gies correlated satisfactorily with the
calculated scores, the applied scoring
functions appeared to be appropriate
for the selection of promising candi-
dates for receptor synthesis.
A
utoDock and GlamDock to generate
camptothecin complexes of every can-
didate receptor. Scoring functions were
used to rank all generated complexes.
From the 10% top-ranking candidates
nine were selected for experimental
Keywords: bioinformatics
puter chemistry · cyclodextrins ·
drug delivery molecular
recognition
· com-
·
Introduction
make them interesting candidates for diagnostic, therapeu-
tic, analytical, and separation purposes.[1] Their three-dimen-
sional structure can be more stable at high temperatures
and non-physiological pH conditions. Their comparably low
molecular weights and their better tolerability by the human
immune system make them interesting candidates for the
delivery of drugs. On the other hand, the development of
synthetic receptors is hampered by tedious synthesis, by
time consuming trial-and-error searches, and problems with
solubility in water.[2] In this paper we demonstrate how the
computational technique of virtual screening can support
synthetic receptor design.
Synthetic receptors are molecules that specifically bind
guest molecules. In general, they cannot rival proteins in
terms of binding affinity and specificity. However, they do
exhibit numerous advantages over natural receptors that
[a]A. Steffen, Dr. A. Kämper, Prof. Dr. T. Lengauer
Computational Biology and Applied Algorithmics
Max-Planck-Institut für Informatik, Stuhlsatzenhausweg 85
66123 Saarbrücken (Germany)
[b]C. Thiele, C. Strassnig, Prof. Dr. G. Wenz
Organische Makromolekulare Chemie, Saarland University
Geb. C4.2, 66123 Saarbrücken (Germany)
Fax : (+49) 681 302 3909
Virtual screening with protein-ligand docking tools, such
as FlexX,[3] Gold,[4] Glide,[5] AutoDock,[6] or Dock,[7] is well
established in the field of computer-aided drug design for
identification of novel ligands for a given protein target.[8] In
general, these docking tools consist of two components:
1) the conformational sampling of the guest molecules
within the receptor binding site and 2) the scoring function
for ranking the different conformations of the complex. In
contrast to time-consuming molecular dynamic simulations,
docking tools are designed to be fast and to provide solu-
tions in the range of seconds to minutes. This efficiency
E-mail: g.wenz@mx.uni-saarland.de
[c]S. Tietze, Dr. J. Apostolakis
Institute for Informatics
Research and Educational Unit for Bioinformatics
and Practical Informatics
Ludwig-Maximilians-Universität, Amalienstr. 17
80333, Munich (Germany)
Supporting information for this article is available on the WWW
under http://www.chemeurj.org/ or from the author.
Chem. Eur. J. 2007, 13, 6801 – 6809
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
6801

allows for virtual screening. Virtual screening can be under-
stood as the virtual analog of high-throughput screening.[9]
A library of candidate molecules are docked against the
target and ranked according to the predicted binding affini-
ty. The most promising molecules are subsequently submit-
ted to experimental validation. Several successful studies
have been reported in which virtual screenings identified
relevant ligands leading to new promising drugs.[10]
De Jong et al. reported the identification of novel ligands
for a given synthetic receptor by application of the docking
tool Dock,[7] originally developed for drug design.[11] Dock-
ing was performed by placement of energetically minimized
ligand structures into a b-CD dimer. Despite the fact that
conformational flexibilities of both the receptor and the
ligand had not been taken into account, nine out of 30 pro-
posed ligands were found to bind to the receptor with high
affinity. In this work we address the opposite problem of
looking for synthetic receptors that bind a given ligand with
high affinity, since this task is highly relevant for the com-
plexation and controlled delivery of drugs.
ubility and stability.[14] Some attempts have been made to
circumvent these difficulties by means of pharmaceutical
formulations and inclusion in cyclodextrins (CDs).[14–19]
CDs and their derivatives are an interesting class of or-
ganic hosts[20,21] that have been applied in various pharma-
ceutical formulations.[22] Their hydrophobic cavity combined
with a hydrophilic exterior designates their application for
solubilizing small hydrophobic molecules, such as drugs, in
water.[23] In this way, CDs can increase the plasma level of
the complexed drugs and thus increase their therapeutic
effect. Recently, we demonstrated that CDs are able to rec-
ognize the thicknesses of included monomeric[24] and poly-
meric[25,26] guest molecules. Selectivities and affinities of
CDs can be increased by chemical modifications of the
CDs.[27]
In this study we focus on the development of b-CD deriv-
atives specifically tailored for solubilization of camptothecin.
We chose a straightforward synthesis procedure for regiose-
lective modification of b-CD by nucleophilic displacement
reactions of 6-O-iodo or 6-O-tosyl-b-CDs with various
thiols.[28] First, a virtual library of candidate receptors was
generated. Each candidate of this library was then docked
onto camptothecin. This approach is referred to as inverse
virtual screening as the docking direction is inverted[29] with
respect to common virtual screenings in drug design.[9] Simi-
lar to the normal virtual screening scenario, we applied scor-
ing functions to rank the different candidates. Top-ranking
candidate receptors were synthesized and experimentally
tested (see Scheme 2).
Camptothecin (Scheme 1) was chosen here as the ligand
for the design of a tailored receptor by means of computa-
tional methods, as it represents a promising class of antineo-
Scheme 1. Structure of camptothecin and schematic receptor design by
attachment of one or seven building blocks to b-CD.
Scheme 2. Design of the study.
plastic agents that exhibit a broad spectrum of activity
against several types of cancer, including colorectal and
ovarian cancer.[12] Camptothecin inhibits topoisomerase I,
which is a nuclear enzyme involved in the relaxation of
DNA during cell replication and transcription.[13] Unfortu-
nately, its high therapeutic potential is hampered by low sol-
Results
A virtual library (1846 entities) of 6-O-mono- and 6-O-
hepta-substituted b-CD derivatives was generated from the
b-CD core and thiol building blocks. The structure of the
complex between camptothecin and the different derivatives
6802
www.chemeurj.org
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Chem. Eur. J. 2007, 13, 6801 – 6809

Computer Chemistry
FULL PAPER
Table 1. Building blocks selected by virtual screening of corresponding b-CD derivatives.
was predicted and the deriva-
tives were ranked according to
the score of the complex (see
IUPAC Name
CAS-No.
mono
hepta
Method[a]
derivative
derivative
Computational
Methods).
1
2
3
4
5
6
7
8
9
1-methyltetrazole-5-thiol
2-aminoethanethiol
13183-79-4
60-23-1
68-11-1
3375-50-6
79-42-5
96-27-5
107-96-0
50-44-2
2637-34-5
21
22
23
24
25
26
27
28
29
11
12
13
14
15
16
17
18
19
AD
GD
GD
GD
GD
GD
GD
AD
AD, GD
[Scores are used as heuristic es-
timates of DG8. By convention,
the lower the score, the more
favorable the interaction; thus,
the first rank corresponds to
the complex exhibiting the
lowest score.]Usually, protein-
ligand docking tools explore
the conformational space of li-
gands, for example, drug mole-
2-mercaptoacetic acid
2-mercaptoethanesulfonate
2-mercaptopropanoic acid
3-mercaptopropane-1,2-diol
3-mercaptopropanoic acid
9H-purine-6-thiol
pyridine-2-thiol
[a]AD =AutoDock, GD=GlamDock.
cules, while treating the protein as rigid (in its crystal struc-
ture conformation) during the simulated binding. For our
work this simplification was not appropriate. The conforma-
tions of the virtually generated b-CD derivatives had to be
generated during the docking process. Since the so-called
flip-flop hydrogen bonds between secondary hydroxyl
groups of neighboring glucose units restrict the flexibility of
the b-CD core,[30] only the side-chains were considered as
flexible in the present study, whereas the b-CD core was
kept as given in the crystal structure (see Computational
Methods). Conversely, camptothecin was described by a
single conformation, since it is rather rigid. Consequently,
we performed an inverse docking in which the receptor con-
formation was optimized in the field of the rigid ligand.[29,31]
The virtual screening was performed through the use of
two docking tools, namely AutoDock[6,32–34] and Glam-
Dock.[35] AutoDock uses a grid-based energy evaluation pro-
cedure based on the Amber force field. Four algorithms are
implemented for optimizing the complex conformation. The
Lamarckian genetic algorithm was chosen for our study,
since it had been shown to be most effective and reliable.[6]
The current version of GlamDock relies on a Monte Carlo
procedure based on matching of functional groups of the
ligand with favorably interacting probes in the binding site
combined with local minimization.[36,37] The scoring function
for the optimization b-CD derivatives conformations is a
continuous-gradient approximation to the docking version
of ChemScore.[38,39] For ranking the compounds we addition-
ally introduced a size-penalizing term (see Computational
Methods). GlamDock has been validated on benchmark sets
of literature data. It performed comparably or better than
the state-of-the-art methods on the Kellenberger data set.[40]
For experimental verification only those compounds were
considered that were found by at least one docking tool
within the top 10% of the respective ranking lists. All po-
tential candidates were visually inspected. Promising b-CD
derivatives were selected for synthesis and further experi-
mental investigation. Furthermore, the building blocks for
synthesizing the b-CD derivative had to be commercially
available. AutoDock favored b-CD derivatives with aromat-
ic and hydrophobic side-chains, whereas GlamDock in com-
bination with the size penalty mainly suggested derivatives
forming hydrogen bonds to camptothecin (Table 1). The
predicted affinity scores for heptakis derivatives were gener-
ally more favorable than for the corresponding mono deriv-
atives with both docking programs.
Nine heptakis-b-CD derivatives were synthesized by nu-
cleophilic displacement reactions in good yields (see
Scheme 3). Four of them (11, 15, 18, 19) were insoluble in
water. For a closer investigation of the molecular interac-
tions and for obtaining an estimate of binding affinities of
all substituents, we also synthesized the corresponding mono
derivatives 21–28 (see Scheme 4), which were all soluble in
water. Furthermore the heptakis-substituted thiosulfate b-
CD (compound 20) was synthesized. This molecule had an
unfavorable predicted binding energy and served as a nega-
tive test.
The binding constants K of all synthesized b-CD deriva-
tives were determined from the solubility isotherms. The in-
crease of the solubility of camptothecin with increasing con-
centration of the CD derivatives is demonstrated in
Figure 1. The binding constants K were derived from the
slope and the solubility of camptothecin.[19] To assure com-
parability we additionally measured the binding constants of
the native b-CD, hydroxypropyl-b-CD (HP-b-CD) and ran-
domly methylated b-CD (RDM-b-CD), which were already
investigated by Kang et al.[19] It should be noted that the
value of K obtained for RDM-b-CD (186mꢀ1) significantly
differed from its literature value (909.7mꢀ1).[19] This differ-
ence might be caused by different experimental protocols
and different substitution patterns of the randomly methy-
lated b-CDs.
Out of the nine synthesized receptors from the virtual
screening, five exhibit binding constants K clearly superior
to the ones of the native b-CD and the two other known
CD derivatives[19] (see Table 2 and Figure 1). Heptakis-[6-
deoxy-6-(2-sulfanylethanesulfonic acid)]-b-CD (14) showed
the highest value of K with 7496mꢀ1. Since receptors 11, 15,
18, and 19 were insoluble in water, the corresponding mono
derivatives 21–29 were investigated. Among them, mono-[6-
deoxy-6-(6-sulfanyl-9H-purine)]-b-CD (28) showed the
strongest binding affinity with 3629mꢀ1, which is in the
range of the hepta-substituted CD derivatives. As predicted,
the negative test example (compound 20) exhibits a compa-
rably low binding affinity with a K value of 370mꢀ1.
Chem. Eur. J. 2007, 13, 6801 – 6809
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemeurj.org
6803

G. Wenz, J. Apostolakis et al.
reasons. Remarkably, one 6-sul-
fanyl-9H-purine building block
in compound 28 leads to an ex-
ceptionally strong stabilization
of DDG8=ꢀ7 kJmolꢀ1.
In Figures 2 and 3 the Auto-
Dock and the GlamDock affini-
ty scores, respectively, are plot-
ted versus the experimentally
determined values of the bind-
ing affinities. For compounds
11, 15, 18, and 19 no binding
free energy could be experimen-
tally determined due to insolu-
bility in water. The correlation
coefficient for AutoDock is r=
0.57 (residual standard error of
the regression 2.4 kJmolꢀ1), for
GlamDock r=0.82 (residual
standard error of the regression
1.6 kJmolꢀ1). Compound 14 is
an evident outlier for both
docking tools, but particularly in
the case of AutoDock. If this
compound is omitted, the corre-
lation coefficient for AutoDock
increases to 0.78 (residual stan-
Scheme 3. Synthesis of hepta substituted CD derivatives. a) 1) PPh3, I2, DMF 2) CH3ONa, CH3OH; b) 2,
NH4HCO3, DMF/H2O; c) 1, NEt3, DMF; d) 4, NEt3, DMSO; e) 1) methyl ester of 7, NEt3, DMF 2) NaOH; f)
1) methyl ester of 5, NEt3, DMF 2) NaOH; g) 6, NEt3, DMF; h) 1) methyl ester of 3, NEt3, DMF 2) NaOH; i) dard error of the regression
1.7 kJmolꢀ1).
9, NEt3, DMF; j) 8, NEt3, DMF; k) sodium thiosulfate, DMSO.
Discussion
Due to the flexibility and larger size of artificial receptors,
virtual screening of receptors (inverse screening) is in gener-
al more complex than virtual screening of ligands.[11] For a
given complex the predicted binding free energy DG8
(score) consists in principle of three components [Eq. (1)]in
which DGoR is the change of energy in the receptor molecule,
DGoL is the change of energy in the ligand upon complexa-
tion, and DGoRL is the interaction energy of the complex.
DGꢁðpredictedÞ ¼ DGoR þ DGoL þ DGoRL
ð1Þ
In ligand screening the larger receptor structure is nor-
mally treated as rigid and thus DGoR cancels. The estimated
binding energy of the system depends only on the interac-
tion energy between ligand and receptor DGoRL and to a
small extent on the change of the internal energy of the flex-
ible ligand DGoL.
Figure 1. Dependence of the solubility of camptothecin (CPT) on the
concentration of the b-CD derivatives.
Comparison of the binding free energies DG8 for the ma-
jority of the mono-substituted b-CD derivatives 21–27 and
29 with unsubstituted b-CD shows a stabilization energy of
around DDG8=ꢀ2 kJmolꢀ1 caused by one building block.
The same comparison for the hepta-substituted b-CDs 12–
In inverse screening the receptors were treated flexible,
whereas the guest molecule was kept rigid (DGoL =0). Due
to the large size of the receptor the change of its internal
energy (DGoR) contributes predominantly to the binding
energy. Self-inclusion of the receptor can lead to low-energy
conformations of the receptors with little interaction to the
guest molecule. This is shown in Figure 4, which depicts a
17 results in DDG8=ꢀ
C
6804
www.chemeurj.org
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Chem. Eur. J. 2007, 13, 6801 – 6809

Computer Chemistry
FULL PAPER
the size of the receptor remains
and leads to better scoring of
unspecifically interacting hydro-
phobic receptors. The second
approach reduces this problem,
but is highly empirical and re-
quires the definition of more or
less arbitrary weights for the
size term. Finally, in the last
option, conformations as shown
in Figure 4 are explicitly forbid-
den, even though they may cor-
respond to the most probable
structure of the complex.
In the current work, we
chose two different combina-
tions of these approaches. In
the screening with GlamDock
we used approaches 1) and 2)
by explicitly adding a term pe-
nalizing the size of the receptor
for the ranking, and only used
the intermolecular interaction
energy DGoRLfor scoring. For
AutoDock we used approach
3), as the sampling region of
the receptor is limited in such a
way that camptothecin is
always within the binding cleft
of the derivatives.
Scheme 4. Synthesis of mono substituted CD derivatives. a) Tos-Cl, NaOH, ACN/H2O; b) 2, NH4HCO3, DMF/
H2O; c) 1, NEt3, DMF; d) 4, NEt3, DMF; e) 1) methyl ester of 7, NEt3, DMF 2) NaOH; f) 1) methyl ester of
5, NEt3, DMF 2) NaOH; g) 6, NEt3, DMF; h) 1) methyl ester of 3, NEt3, DMF 2) NaOH; i) 9, NEt3, DMF; j)
8, NEt3, DMF.
Overall, the results show that
complex with a predicted favorable score. Camptothecin lies
on top of the receptor. One of the hydrophobic side chains
is buried in the CD cavity and leads to a favorable internal
energy (DGoR), which compensates the poor intermolecular
interactions (DGoRL). This leads to well-scoring complexes
that show little interaction between ligand and receptor.
Furthermore, the average interaction of a system increases
quadratically with the number of its atoms, and therefore re-
ceptors with large substituents are generally scored more fa-
vorably than smaller receptors.
these two approaches have their own advantages and disad-
vantages. The AutoDock approach led to the selection of re-
ceptors that were highly hydrophobic, and could therefore
not be measured. On the other hand, it also led to the iden-
tification of compound 28, which is the only mono derivative
that can rival the heptakis-substituted derivatives in terms
of binding affinity. The GlamDock approach proposed re-
ceptors with smaller and more hydrophilic side chains in
general, which show improved binding affinity compared to
b-CD. Furthermore the scores correlate reasonably well
with the experimental binding affinities. It is interesting to
note that one derivative (compound 14) appears to be an
outlier for both scoring functions. Both, AutoDock and
GlamDock significantly underpredict its binding affinity.
Nevertheless, in spite of the uncertainties of structure pre-
diction, and the modeling itself, the overall results suggest
that at least the tendency of binding affinity is reproduced.
For AutoDock a residual standard error of 9.11 kJmolꢀ1 was
reported in literature for a set of 30 protein–ligand com-
plexes.[6] Furthermore, with regression methods a cross-vali-
dated correlation coefficient of 0.89 and a standard devia-
tion of 2.38 kJmolꢀ1 were reported for a set of 218 com-
plexes between b-CD and different guest molecules (CO-
DESSA-PRO descriptors).[41] This correlation is better than
those achieved in the present work (r=0.82), while the aver-
age error is even higher than in the one here (1.6 kJmolꢀ1).
There are at least three different approaches to address
this type of problem within the paradigm of fast virtual
screening:
1) Score the complexes only according to the interaction
between receptor and ligand DGRo L
.
2) Add a size-dependent term to the ranking function,
which simply depends on the number of atoms, to penal-
ize large complexes.
3) Constrain the docking to allow only conformations with
the camptothecin in the binding site of the receptor.
In the first approach it is important to consider the intra-
molecular receptor energy DGoR during the conformational
sampling to avoid physically unreasonable conformations of
the receptor. However, the proportionality of interactions to
Chem. Eur. J. 2007, 13, 6801 – 6809
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemeurj.org
6805

G. Wenz, J. Apostolakis et al.
Table 2. Binding constants K and binding free energies DG8 for camptothecin in 0.02m HCl.
chains leads to higher binding
affinity. This effect is illustrated
in Figure 5 in which the hydro-
phobic parts of the cysteaminyl
side chains of compound 12
show a good shape complemen-
tarity and hydrophobic interac-
tions to the camptothecin ring
system (dashed green lines). In
addition, camptothecin is also
able to interact specifically by
forming directional hydrogen
bonds. The complex exhibits
three intermolecular hydrogen
bonds (dashed red lines) of the
ammonium groups to hydrogen
bond acceptor atoms of camp-
tothecin.
Compound
K [mꢀ1
]
DG8 [kJmolꢀ1
]
b-CD
HP-b-CD
RDM-b-CD
202Æ30
223Æ32
ꢀ13.2Æ0.5
ꢀ13.4Æ0.4
ꢀ12.9Æ0.2
–
ꢀ21.0Æ0.3
ꢀ18.0Æ0.3
ꢀ22.1Æ0.7
–
186Æ12
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
heptakis-[6-deoxy-6-(1-methyl-5-sulfanyl-tetrazole)]-b-CD
heptakis-[6-deoxy-6-(2-aminoethylsulfanyl)]-b-CD
heptakis-[6-deoxy-6-(2-sulfanyl acetic acid)]-b-CD
heptakis-[6-deoxy-6-(2-sulfanylethanesulfonic acid)]-b-CD
heptakis-[6-deoxy-6-(2-sulfanylpropanoic acid)]-b-CD
heptakis-[6-deoxy-6-(3-sulfanylpropane-1,2-diol)]-b-CD
heptakis-[6-deoxy-6-(3-sulfanylpropanoic acid)]-b-CD
heptakis-[6-deoxy-6-(6-sulfanyl-9H-purine)]-b-CD
heptakis-[6-deoxy-6-(2-sulfanylpyridine)]-b-CD
heptakis-[6-deoxy-6-sulfanylsulfonyloxysodium)]-b-CD
mono-[6-deoxy-6-(1-methyl-5-sulfanyltetrazole)]-b-CD
mono-[6-deoxy-6-(2-aminoethylsulfanyl)]-b-CD
mono-[6-deoxy-6-(2-sulfanyl acetic acid)]-b-CD
mono-[6-deoxy-6-(2-sulfanylethanesulfonic acid)]-b-CD
mono-[6-deoxy-6-(2-sulfanylpropanoic acid)]-b-CD
mono-[6-deoxy-6-(3-sulfanylpropane-1,2-diol)]-b-CD
mono-[6-deoxy-6-(3-sulfanylpropanoic acid)]-b-CD
mono-[6-deoxy-6-(6-sulfanyl-9H-purine)]-b-CD
mono-(2-mercaptopyridine)-b-CD
insoluble
4821Æ572
1450Æ177
7496Æ2002
insoluble
4106Æ475
3134Æ364
insoluble
insoluble
370Æ48
ꢀ20.6Æ0.3
ꢀ19.9Æ0.3
–
–
ꢀ14.7Æ0.3
ꢀ15.2Æ0.3
ꢀ15.4Æ0.3
ꢀ15.4Æ0.3
ꢀ15.0Æ0.3
ꢀ15.0Æ 0.3
ꢀ15.6Æ0.4
ꢀ15.7Æ0.3
ꢀ20.3Æ1.1
ꢀ16.0Æ0.2
465Æ55
498Æ69
493Æ61
431Æ56
419Æ53
531Æ79
569Æ68
On the other hand, polar in-
teracting groups pay a relatively
high desolvation penalty in
aqueous solution and are most
probably not the main driving
3629Æ1567
641Æ53
Figure 2. Predicted binding free energies (AutoDock) plotted versus the
experimental binding free energies. The mono derivatives are depicted
by filled circles, the heptakis derivatives are shown as filled triangles,
except compound 14 shown as open triangle.
Figure 3. Predicted binding free energies (GlamDock) plotted versus the
experimental binding free energies. The mono derivatives are depicted
by filled circles, the heptakis derivatives are shown as filled triangles,
except compound 14 shown as open triangle.
Regression methods are not applicable for our situation as
no training data for generating the regression was available.
However, the comparison suggests that our results on this
system are at the upper bound of what can be achieved with
simple modeling approaches.
To exemplify the interactions involved in the complex for-
mation of camptothecin and the described b-CD derivatives,
we show predicted complex structures of compounds 12 and
18 in Figures 5 and 6, respectively. The molecular structure
of camptothecin offers several possibilities for intermolecu-
lar interactions. The large hydrophobic area of camptothecin
facilitates dispersive interactions. Consequently, an enlarge-
ment of the hydrophobic CD cavity by hydrophobic side-
force behind complex formation for the regarded system.
Additionally we could show that aromatic building blocks,
for example, purine in compound 28 and, to a smaller
extent, pyridine in compound 29 increase complex stability.
This result might be best explained by the occurrence of p-
stacking (dashed pink line) between camptothecin and the
heterocycle (see Figure 6).
In general, hydrophobic interactions are the main driving
force behind affinity, while polar interactions are more re-
sponsible for the specificity of the interaction. While a gen-
eral size effect can be observed in the data, specific effects
are evident, as mono-substituted compounds exist that bind
better than heptakis-substituted compounds and vice versa.
6806
www.chemeurj.org
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Chem. Eur. J. 2007, 13, 6801 – 6809

Computer Chemistry
FULL PAPER
Figure 4. GlamDock docking result for a candidate receptor (heptakis-{6-
deoxy-6-[3-(trifluoromethyl)benzenesulfanyl]}-b-CD) and camptothecin
with a predicted low binding free energy. Hydrogen atoms are omitted
for clarity.
Figure 6. The figure shows the generated complex structure of compound
28 to camptothecin (AutoDock). The dashed pink line depicts a possible
p-stack interaction. Hydrogen bonds are shown as dashed red lines.
Apolar hydrogen atoms are omitted for clarity.
to the work of de Jong et al.,[11] who described the identifi-
cation of new ligands for a given CD host. Our results indi-
cate that inverse virtual screening can support the identifica-
tion of improved receptors for a given ligand and might
open up novel possibilities for the tailored design of drug
delivery systems. Finally, we would like to point out that this
approach is not limited to CD derivatives. The rules for gen-
erating the virtual library of hosts can be arbitrarily expand-
ed to other host–guest systems of interest.
Experimental Section
Computational methods: The structures of all compounds were saved in
the MOL2 file format. For the preparation of the camptothecin structure
the crystal structure of an iodoacetyl derivative of camptothecin was ob-
tained from the Cambridge Structural Database (CSD)[42] (ID:
CAMPTC10).[43] The iodoacetyl group was replaced by a hydrogen atom
in order to construct the unmodified camptothecin molecule. Missing hy-
drogen atoms were added with SYBYL 6.7. Subsequently a force-field
optimization was performed with the MMFF94s force field[44] until gradi-
Figure 5. The figure shows the generated complex structure of compound
12 to camptothecin (GlamDock). Hydrogen bonds are depicted by
dashed red lines, strong hydrophobic interactions are shown as dashed
green lines. Apolar hydrogen atoms are omitted for clarity.
Compound 28 binds better than all other mono-substituted
derivatives and better than some heptakis-substituted com-
pounds. It is recognized by both affinity predictions as the
best of the mono-derivatized compounds. Inversely, an ex-
tension of the cavity does not necessarily result in an in-
creased binding affinity of the complex. The heptakis-substi-
tuted thiosulfate b-CD derivative (compound 20), for exam-
ent convergence (0.005 kcalmolꢀ1 ꢀ1).
Š
For the preparation of the b-CD core structure, the crystal structure of b-
CD with resolved deuterium positions, derived from neutron diffraction
data, was obtained from the CSD (ID: BUVSEQ03).[45] The D atoms
were changed to H atoms. All remaining atom types were inspected and
corrected according to the SYBYL atom type rules if necessary. Water
molecules present in the crystal structure were completely removed.
ple, exhibits
a rather weak binding free energy of
Building blocks containing one thiol group were extracted from the
Sigma-Aldrich catalog (605 entities) and the ZINC database[46] (318 enti-
ties) by means of standard substructure search interfaces allowing a mo-
lecular weight range of 0–200 gmolꢀ1 in order to limit the sizes. Low-
energy conformations were generated with CORINA,[47] and reasonable
protonation states for pH 7 were assigned to all compounds, that is,
acidic groups were deprotonated and aliphatic amines were protonated.
ꢀ14.62 kJmolꢀ1 as predicted by both docking tools.
Conclusion
We have investigated a rational optimization approach to
synthetic receptor design. Our approach is complementary
The virtual library (1846 entities) of 6-O-mono- and 6-O-hepta-substitut-
ed b-CDs was generated from the b-CD core and each of the thiol build-
Chem. Eur. J. 2007, 13, 6801 – 6809
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemeurj.org
6807

G. Wenz, J. Apostolakis et al.
ing blocks with the help of a PYTHON script. Each building block was
transformed in three-dimensional space such that its thiol group was su-
perimposed onto either one or all seven primary oxygen atoms of the b-
CD structure. The oxygen atoms were virtually substituted by the sulfur
atoms of the building blocks. Excess hydrogen atoms were removed and
stirring for 3 d at 608C under N2 the product was worked up as described
above.
General procedure for the synthesis of heptakis-6-O-b-CD derivatives
(acidic thiol building blocks): A solution of heptakis-[6-deoxy-6-iodo]-b-
CD[51] (2.86 g, 1.5 mmol) in DMF (20 mL) was mixed with triethylamine
(3.66 mL, 26.3 mmol) and the methyl ester of the thiol compound
(26.3 mmol). After stirring for 3 d at 608C under N2 the product was con-
centrated in vacuo, precipitated by addition of ethanol or acetone, and
filtered. This product was stirred in 1m NaOH (50 mL) for 18 h and fur-
ther purified by nanofiltration.
ꢀ
the lengths and the angles a of the C S bonds were adjusted to standard
values (see Scheme 5). Rotatable torsion angles were optimized during
docking.
General procedure for the synthesis of mono-6-O-b-CD derivatives
(acidic thiol building blocks): A solution of mono-[6-deoxy-6-(p-toluol-
sulfonyl)]-b-CD[52] (1.93 g, 1.5 mmol) in DMF (20 mL) was mixed with
triethylamine (2.10 mL, 15.0 mmol) and the methyl ester of the thiol
compound (15.0 mmol). After stirring for 3 d at 608C under N2 the prod-
uct was worked up as described above.
Determination of binding constants: The binding constants K and the
corresponding binding free energies DG8 of the camptothecin complexes
were determined with the solubility method described by Higuchi and
Lach[53] and Kang et al.[19] Solutions of the CD derivatives (0–6 mm) in
0.02m HCl (5 mL) were stirred with an excess of camptothecin at 258C
for 18 h. The concentration of dissolved camptothecin was determined
spectrophotometrically after filtration from an extinction coefficient of
camptothecin e(370 nm, water/DMSO 1:1 v/v)=42282mꢀ1 cmꢀ1. Due to
the overlap of the absorbances of camptothecin and b-CD derivative 28
in the aqueous phase, an aqueous camptothecin solution (5 mL) was ex-
tracted with trichloroethane/trifluoroacetic acid 75:1 v/v (5 mL). The
camptothecin concentration in the organic phase was determined spectro-
photometrically with e(370 nm, trichloroethane/TFA 75:1 v/v)=
13490mꢀ1 cmꢀ1 taken as being equal to the original concentration in the
aqueous phase. The solubility of camptothecin was plotted versus the
concentration of the CD derivative (see Figure 1). The binding constant
K was calculated from the zero-point solubility in the solvent [G]0 2.11Æ
Scheme 5. Schematic drawing of the virtual synthesis of the b-CD library.
Docking and scoring was performed with AutoDock 3.05 and Glam-
Dock 1.0 on a cluster of thirty IntelP4 Xeon 3.06 GHz CPUs. The confor-
mational search was performed for the b-CD derivatives, whereas camp-
tothecin was kept rigid. For AutoDock the grid maps were generated for
camptothecin with AutoDock tools[48] by defining a 50”50”50 Š cube
around camptothecin with grid spacing of 0.375 Š. The number of ener-
getic evaluations was set to five million, the number of GA runs to 100,
and the maximal possible number of torsions to 30. For all other parame-
ters default values were used. For GlamDock the docking protocol con-
sisted of five single docking runs each consisting of 650 Monte Carlo min-
imization (MCM) steps, with 15 steps of Levenberg–Marquardt[49] mini-
mization in torsion space[50] at each MCM step. A maximum of 40 poses
were finally post-minimized by 150 steps of Levenberg–Marquardt.
0.24 mm and the slope
B
according to Equation (2).[53] Binding free
For AutoDock docking and ranking was performed based on the overall
score for AutoDock (Dock Score). In the case of GlamDock the scoring
function for docking considers the internal energies of the receptors,
whereas for ranking a size penalizing variant of the fitness score for
GlamDock without internal energy was used. GlamDock does not con-
strain the receptor to dock around the ligand. It explicitly allows confor-
mations where the ligand lies on top of the CD ring (see Figure 4). Such
conformations are mainly stabilized by the internal energy of the recep-
tor, which on the average scales quadratically with its size (the number
of atoms). The size penalty has the effect of identifying more specifically
interacting complexes; however, this is a clearly empirical approach of
ranking the screening results and does not necessarily correlate with
binding affinity. The reason for the size penalty was that initially both vir-
tual screening results contained mainly large hydrophobic receptors on
top ranks.
energy was calculated according to DG8=ꢀRT lnK.
B
K ¼
ð2Þ
ð1ꢀBÞ½Gꢂ0
Acknowledgements
We thank A. Engelke and J. Büch for technical support. Financial sup-
port from the Deutsche Forschungsgemeinschaft through the projects
KA 1804/1 and AP 101/1 is gratefully acknowledged.
Materials: Compounds 2, 6, 7, 8, 9 and DMF (absolute) were purchased
from Fluka, 1 and 5 from Aldrich, 4 from Sigma, 3 from Merck, campto-
thecin from TCI Europe. b-CD, randomly methylated b-CD (RDM-b-
CD, DS=1.8 per glucose unit) and hydroxypropyl-b-CD (HP-b-CD,
DS=0.9 per glucose unit) were donated by Wacker. The purification of
the synthesized compounds was performed by nanofiltration with a Berg-
hof BM-5 membrane (cut-off molecular weight 500 Da) and MilliQ
water. Syringe filters from Roth (CME, 0.22 mm) were used for separa-
tion of insoluble materials before UV measurement.
[1]T. Schrader, A. D. Hamilton, Functional Synthetic Receptors, Wiley-
VCH, Weinheim, 2005.
[2]G. V. Oshovsky, D. N. Reinhoudt, W. Verboom, Angew. Chem. 2007,
119, 2418; Angew. Chem. Int. Ed. 2007, 46, 2366.
[3]M. Rarey, B. Kramer, T. Lengauer, G. Klebe, J. Mol. Biol. 1996, 261,
470.
[4]G. Jones, P. Willett, R. C. Glen, A. R. Leach, R. Taylor, J. Mol. Biol.
1997, 267, 727.
[5]R. A. Friesner, J. L. Banks, R. B. Murphy, T. A. Halgren, J. J. Klicic,
D. T. Mainz, M. P. Repasky, E. H. Knoll, M. Shelley, J. K. Perry,
D. E. Shaw, P. Francis, P. S. Shenkin, J. Med. Chem. 2004, 47, 1739.
[6]G. M. Morris, D. S. Goodsell, R. S. Halliday, R. Huey, W. E. Hart,
R. K. Belew, A. J. Olson, J. Comput. Chem. 1998, 19, 1639.
[7]T. J. A. Ewing, I. D. Kuntz, J. Comput. Chem. 1997, 18, 1175.
[8]N. Brooijmans, I. D. Kuntz, Annu. Rev. Biophys. Biomol. Struct.
2003, 32, 335.
General procedure for the synthesis of heptakis-6-O-b-CD derivatives
(non-acidic thiol building blocks): A solution of heptakis-[6-deoxy-6-
iodo]-b-CD[51] (2.86 g, 1.5 mmol) in DMF (15 mL) was mixed with tri-
A
After stirring for 3 d at 608C under N2, the product was concentrated in
vacuo, precipitated by addition of ethanol or acetone, filtered, dried in
vacuo, and further purified by nanofiltration.
General procedure for the synthesis of mono-6-O-b-CD derivatives (non-
acidic thiol building blocks): A solution of mono-6-O-(p-toluolsulfonyl)]-
b-CD[52] (1.93 g, 1.5 mmol) in DMF (20 mL) was mixed with triethyl-
[9]B. K. Shoichet, Nature 2004, 432, 862.
A
[10]G. Klebe, DrugDiscovery Today 2006, 11, 580.
6808
www.chemeurj.org
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Chem. Eur. J. 2007, 13, 6801 – 6809

Computer Chemistry
FULL PAPER
[11]M. R. de Jong, R. M. Knegtel, P. D. Grootenhuis, J. Huskens, D. N.
Reinhoudt, Angew. Chem. 2002, 114, 1046; Angew. Chem. Int. Ed.
2002, 41, 1004.
[34]D. Q. Wei, R. Zhang, Q. S. Du, W. N. Gao, Y. Li, H. Gao, S. Q.
Wang, X. Zhang, A. X. Li, S. Sirois, K. C. Chou, Amino Acids 2006,
31, 73.
[12]C. H. Takimoto, J. Wright, S. G. Arbuck, Biochim. Biophys. Acta
Gene Struct. Expression 1998, 1400, 107.
[13]W. J. Slichenmyer, E. K. Rowinsky, R. C. Donehower, S. H. Kauf-
mann, J. Natl. Cancer Inst. 1993, 85, 271.
[14]B. B. Lundberg, Anti-Cancer DrugDes. 1998, 13, 453.
[15]T. G. Burke, A. E. Staubus, A. K. Mishra, H. Malak, J. Am. Chem.
Soc. 1992, 114, 8318.
[35]a) M. Karasz, R. Koerner, M. Marialke, S. Tietze, J. Apostolakis,
Proceedings of the 15th European Symposium on Quantitative Struc-
ture Activity Relationships & Molecular Modeling (Turkey), 2004,
p. 493; b) S. Tietze, J. Apostolakis, J. Chem. Inf. Model.,
DOI:10.1021/ci7001236.
[36]R. Abagyan, M. Totrov, D. Kuznetsov, J. Comput. Chem. 1994, 15,
488.
[16]B. Ertl, P. Platzer, M. Wirth, F. Gabor, J. Controlled Release 1999,
61, 305.
[37]J. Apostolakis, A. Pluckthun, A. Caflisch, J. Comput. Chem. 1998,
19, 21.
[17]S. M. Sugarman, Y. Y. Zou, K. Wasan, K. Poirot, R. Kumi, S. Reddy,
R. PerezSoler, Cancer Chemother. Pharmacol. 1996, 37, 531.
[18]R. Cortesi, E. Esposito, A. Maietti, E. Menegatti, C. Nastruzzi, Int.
J. Pharm. 1997, 159, 95.
[19]J. Kang, V. Kumar, D. Yang, P. R. Chowdhury, R. J. Hohl, Eur. J.
Pharm. Sci. 2002, 15, 163.
[38]M. D. Eldridge, C. W. Murray, T. R. Auton, G. V. Paolini, R. P. Mee,
J. Comput.-Aided Mol. Des. 1997, 11, 425.
[39]M. L. Verdonk, J. C. Cole, M. J. Hartshorn, C. W. Murray, R. D.
Taylor, Proteins: Struct. Funct. Genet. 2003, 52, 609.
[40]E. Kellenberger, J. Rodrigo, P. Muller, D. Rognan, Proteins: Struct.
Funct. Bioinf. 2004, 57, 225.
[20]G. Wenz, Angew. Chem. 1994, 106, 851; Angew. Chem. Int. Ed. Eng-
l.Angew. Chem. Int. Ed. 1994, 33, 803.
[21]M. V. Rekharsky, Y. Inoue, Chem. Rev. 1998, 98, 1875.
[22]M. E. Davis, M. E. Brewster, Nat. Rev. DrugDiscovery 2004, 3,
1023.
[41]A. R. Katritzky, D. C. Fara, H. Yang, M. Karelson, T. Suzuki, V. P.
Solovꢁev, A. Varnek, J. Chem. Inf. Comput. Sci. 2004, 44, 529–541.
[42]F. H. Allen, Acta. Crystallogr. Sect. B. 2002, 58, 380.
[43]A. T. McPhail, G. A. Sim, J. Chem. Soc. B 1968, 923.
[44]T. A. Halgren, J. Comput. Chem. 1999, 20, 730.
[45]V. Zabel, W. Saenger, S. A. Mason, J. Am. Chem. Soc. 1986, 108,
3664.
[46]J. J. Irwin, B. K. Shoichet, J. Chem. Inf. Comput. Sci. 2005, 45, 177.
[47]J. Sadowski, J. Gasteiger, Chem. Rev. 1993, 93, 2567.
[48]M. F. Sanner, J. Mol. Graphics Modell. 1999, 17, 57.
[49]A. S. Deo, I. D. Walker, J. Intell. Robot. Syst. 1995, 14, 43.
[50]C. Bystroff, Protein Eng. 2001, 14, 825.
[23]K. A. Connors, Chem. Rev. 1997, 97, 1325.
[24]A. Müller, G. Wenz, Chem. Eur. J. 2007, 13, 2218.
[25]G. Wenz, B. H. Han, A. Müller, Chem. Rev. 2006, 106, 782.
[26]G. Wenz, C. Gruber, B. Keller, C. Schilli, T. Albuzat, A. Müller,
Macromolecules 2006, 39, 8021.
[27]T. Kitae, T. Nakayama, K. Kano, J. Chem. Soc. Perkin Trans. 2 1998,
207–212.
[28]V. A. Karginov, S. M. Hecht, N. Fahmi, K. Aliben (Pinnacle Pharma-
ceuticals, USA; Innovative Biologics, Inc.; Advanced Biosystems,
Inc.) WO 2006001844, 2006, [Chem. Abstr. 2006 144:82369].
[29]A. Kämper, J. Apostolakis, M. Rarey, C. M. Marian, T. Lengauer, J.
Chem. Inf. Model. 2006, 46, 903.
[51]P. R. Ashton, R. Kçniger, J. F. Stoddart, D. Alker, V. D. Harding, J.
Org. Chem. 1996, 61, 903.
[52]J. Defaye, S. Crouzy, N. Evrard, H. Law (Centre National De La Re-
cherche Scientifique, Fr.), WO 9961483, 1999 [Chem. Abstr. 1999
132:24077].
[30]C. Betzel, W. Saenger, B. E. Hingerty, G. M. Brown, J. Am. Chem.
Soc. 1984, 106, 7545.
[31]A. Steffen, A. Kämper, T. Lengauer, J. Chem. Inf. Model. 2006, 46,
1695.
[32]J. P. Rogers, A. E. Beuscher, M. Flajolet, T. McAvoy, A. C. Nairn,
A. J. Olson, P. Greengard, J. Med. Chem. 2006, 49, 1658.
[33]D. A. Evans, S. Neidle, J. Med. Chem. 2006, 49, 4232.
[53]T. Higuchi, J. L. Lach, J. Am. Pharm. Assoc. 1954, 43, 349.
[54]F. Guillo, B. Hamelin, L. Jullien, J. Canceill, J. M. Lehn, L. Derober-
tis, H. Driguez, Bull. Soc. Chim. Fr. 1995, 132, 857.
[55]B. Ekberg, L. I. Andersson, K. Mosbach, Carbohydr. Res. 1989, 192,
111.
Received: May 2, 2007
Published online: July 3, 2007
Chem. Eur. J. 2007, 13, 6801 – 6809
ꢀ 2007 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemeurj.org
6809