Two solutions for the same problem: Multiple binding modes of pyrrolidine-based HIV-1 protease inhibitors

Article abstract of DOI:10.1016/j.jmb.2011.04.052

Structure-based drug design is an integral part of industrial and academic drug discovery projects. Initial lead structures are, in general, optimized in terms of affinity using iterative cycles comprising synthesis, biological evaluation, computational methods, and structural analysis. X-ray crystallography commonly suggests the existence of a single well-defined state, termed binding mode, which is generally assumed to be consistent in a series of similar ligands and therefore used for the following optimization process. During the further development of symmetrically disubstituted 3,4-amino-pyrrolidines as human immunodeficiency virus type 1 protease inhibitors, we discovered that, by modification of the P1/P1' moieties of our lead structure, the activity of the inhibitors towards the active-site mutation Ile84Val was altered, however, not being explainable with the initial underlying structure-activity relationship. The cocrystallization of the most potent derivative in complex with the human immunodeficiency virus type 1 protease surprisingly led to two different crystal forms (P2₁2 ₁2₁ and P6₁22). Structural analysis revealed two completely different binding modes; the interaction of the pyrrolidine nitrogen atom with the catalytic aspartates remains as the only similarity. The study presented clearly demonstrates that structural biology has to escort the entire lead optimization process not to fail by an initially observed binding orientation.

Full text of DOI:10.1016/j.jmb.2011.04.052

doi:10.1016/j.jmb.2011.04.052
J. Mol. Biol. (2011) 410, 745–755
Contents lists available at www.sciencedirect.com
Journal of Molecular Biology
journal homepage: http://ees.elsevier.com.jmb
Two Solutions for the Same Problem: Multiple Binding
Modes of Pyrrolidine-Based HIV-1 Protease Inhibitors
Andreas Blum†, Jark Böttcher†, Stefanie Dörr, Andreas Heine,
Gerhard Klebe and Wibke E. Diederich
⁎
⁎
Institut für Pharmazeutische Chemie, Philipps-Universität Marburg, Marbacher Weg 6, 35032 Marburg, Germany
Received 26 January 2011;
received in revised form
20 April 2011;
Structure-based drug design is an integral part of industrial and academic
drug discovery projects. Initial lead structures are, in general, optimized in
terms of affinity using iterative cycles comprising synthesis, biological
evaluation, computational methods, and structural analysis. X-ray crystal-
lography commonly suggests the existence of a single well-defined state,
termed binding mode, which is generally assumed to be consistent in a
series of similar ligands and therefore used for the following optimization
process. During the further development of symmetrically disubstituted
3,4-amino-pyrrolidines as human immunodeficiency virus type 1 protease
inhibitors, we discovered that, by modification of the P1/P1′ moieties of our
lead structure, the activity of the inhibitors towards the active-site mutation
Ile84Val was altered, however, not being explainable with the initial
underlying structure–activity relationship. The cocrystallization of the most
potent derivative in complex with the human immunodeficiency virus type
1 protease surprisingly led to two different crystal forms (P212121 and
P6122). Structural analysis revealed two completely different binding
modes; the interaction of the pyrrolidine nitrogen atom with the catalytic
aspartates remains as the only similarity. The study presented clearly
demonstrates that structural biology has to escort the entire lead
optimization process not to fail by an initially observed binding orientation.
© 2011 Elsevier Ltd. All rights reserved.
accepted 21 April 2011
Edited by M. F. Summers
Keywords:
HIV-1 protease inhibitors;
structure-based drug design;
pyrrolidine-based inhibitors;
crystallography;
deviating binding modes
Beddell et al., who utilized the three-dimensional
structure of hemoglobin for ligand development.1
Since then, the number of protein structures
publicly available in the Protein Data Bank (PDB)
has increased exponentially, resulting in more than
70,000 entries to date.2 Due to the steadily growing
amount of available protein structures and meth-
odological advances in structural analysis, SBDD
has meanwhile become an integral part of drug
discovery projects.3
Introduction
The first successful application of structure-based
drug design (SBDD) was reported in 1976 by
*Corresponding authors. E-mail addresses:
klebe@mailer.uni-marburg.de;
wibke.diederich@staff.uni-marburg.de.
The analysis of a protein–ligand complex gives
detailed information about protein–ligand interac-
tions. Once an initial lead has been identified and the
target protein is accessible for structural analysis, it
can further be optimized using iterative cycles
comprising synthesis, biological evaluation, compu-
tational methods, and structural analysis.4 Structural
† A.B. and J.B. contributed equally to this work.
Abbreviations used: SBDD, structure-based drug
design; PDB, Protein Data Bank; HIV, human
immunodeficiency virus; SAR, structure–activity
relationship; HIV-1, human immunodeficiency virus type
1; vdW, van der Waals; DMSO, dimethyl sulfoxide; MS,
mass spectrometry; ESI, electrospray ionization.
0022-2836/$ - see front matter © 2011 Elsevier Ltd. All rights reserved.

746
Multiple Binding Modes of HIV-1 Protease Inhibitor
analysis using X-ray crystallography usually sug-
gests the existence of a single well-defined state,
anticipated as a unique binding mode, which is then
utilized to propose the most promising modifica-
tions of the initial scaffold to further optimize a given
lead. Numerous successful projects of SBDD have
been published, leading at best to the development
of an approved drug as, for example, in the case of
the human immunodeficiency virus (HIV) protease
inhibitor tipranavir.5
However, the iterative process can face a lot of
obstacles and surprises, and numerous examples of
a nonunique behavior within a series of similar
ligands have been described, of which only the
following three should be mentioned as represen-
tative examples: a reversed binding mode was
detected by X-ray crystallography in the due
course of the SBDD process for nonpeptidic HIV
protease inhibitors. In the reported case, alkylation
of the inhibitor's amide functionalities resulted in a
completely different occupation of individual spec-
ificity subpockets.6 The small energy deviations of
such opposing binding modes have been investi-
gated in detail (e.g., by the pH dependency of the
binding mode observed in the case of trypsin
inhibitor complexes).7 An extraordinary observa-
tion was made in the case of protein kinase C: the
inhibitor binds in different orientations to the two
crystallographically independent monomers pre-
sent in the asymmetric unit.8 Observations like
those mentioned above suggest that the occurrence
of several binding orientations within a series of
similar compounds or even for one certain ligand is
not a rare case. Such phenomena can be indicated,
for example, from a complex and not straightfor-
ward structure–activity relationship (SAR), which
cannot be explained using the broadly accepted
hypothesis that similar ligands bind in a similar
fashion.9
Fig. 1. Schematic representation of the conserved
binding mode of the pyrrolidine-3,4-bis-N-benzyl-sulfon-
amides. Hydrogen bonds are indicated by broken lines.
imposed C2 symmetry is occupied by substituents at
the core skeleton of the inhibitor that mutually
correspond by topological symmetry. The benzyl
moieties occupy the S1 and S1′ subpockets, whereas
the sulfonamide side chains of the inhibitor address
the S2 and S2′ subpockets. The N-benzyl-substituted
derivatives were additionally analyzed with respect
to their susceptibility towards the active-site mu-
tants Ile50Val (PRI50V) and Ile84Val (PRI84V), and
revealed promising properties towards the Ile84Val
mutation located at the borders of the S1/S2′ and
S2/S1′ subpockets, respectively. Particularly, com-
pounds bearing polar substituents in the S2/S2′
pockets showed a significantly improved affinity for
the mutation of isoleucine by valine (e.g., compound
3S,4S)-3,4-bis[benzyl-(4-carbamoylbenzenesulfonyl)
amino]pyrrolidine (2)).11 This observation was
further analyzed by X-ray crystallography of the
corresponding complexes of two derivatives, to-
gether with the protease variants (PDB IDs: 2R38,
2R3T, 2R3W, and 2R43), and could finally be
explained by improved van der Waals (vdW)
contacts, enhanced polar interactions mediated by
an interstitial water molecule, and relief of steric
strain likely given in the wild-type complexes.
Recently, we reported our design and synthesis of
C2-symmetric 3,4-disubstituted pyrrolidines as a
new class of human immunodeficiency virus type
1 (HIV-1) protease inhibitors.10 Our structure-
guided optimization was based on the initially
observed cocrystal structure of the N-benzyl-
substituted inhibitor (3S,4S)-3,4-bis[benzenesulfo-
nylbenzylamino]pyrrolidine (1) (PDB ID: 2PQZ;
Fig. 1 and Table 1).
Interestingly enough, determination of the bind-
ing affinities for the PRI84V and PRI50V mutants of
inhibitors bearing smaller N-alkyl substituents
((3S,4S)-3,4-bis-(allyl-benzenesulfonyl-amino)-pyr-
rolidine (3), (3S,4S)-3,4-bis-[benzenesulfonyl-(2-
methyl-allyl)-amino]-pyrrolidine (4), (3S,4S)-3,4-
bis-[benzenesulfonyl-(3-methyl-but-2-enyl)-amino]-
pyrrolidine-hydrochloride (5)) revealed a different
selectivity profile (Table 1). In order to further
elucidate this surprising difference in the SAR, we
designed a new small series of inhibitors. Inhibitor 5,
In the due course of the project, additional
cocrystal structures of the further developed in-
hibitors were determined to elucidate their interac-
tions explaining the fully consistent SAR. The
conserved binding mode observed in the five crystal
structures (PDB IDs: 2PQZ, 2QNP, 2PWR, 2PWC,
and 2QNN) is depicted in Fig. 1. All complexes
between the topologically C2-symmetric inhibitors
and the sequentially C2-symmetric protein adopt an
unsymmetric binding mode with respect to the
ligand. However, each subpocket related by an

Multiple Binding Modes of HIV-1 Protease Inhibitor
747
Table 1. Ki values of the inhibitors towards the HIV protease
Ki [μM]
PRI50V
11
P2/P2′
P1/P1′
PRWT
2.2
PRI84V
1.1
1
2
3
4
5
6
7
8
9
0.26
12
1.4
NIa
340
10
0.036
84
75
53
1.6
5.8
0.48
0.69
2.3
13
1.9
2.0
7.3
11
0.91
4.9
3.3
2.9
a
NI=Ki N500 μM.
comprising N-dimethylallyl substituents, which
showed the best potency of the series, was picked
as a reference. Its properties were further varied by
decorating the sulfonamide benzyl group with para-
substituents of different physicochemical properties,
such as polar carboxylate and amide groups, and
nonpolar bromine and cyano groups. Based on the
previously observed binding modes, this moiety
was thought to occupy the S2 and S2′ pockets.
(3-methyl-but-2-enyl)-amino]-pyrrolidinium trifluor-
oacetate (9) or with the corresponding nitrile moiety
in (3S,4S)-3,4-bis-[(4-cyano-benzenesulfonyl)-3-methyl-
but-2-enyl)-amino]-pyrrolidinium trifluoroacetate (8)
did not lead to an improvement in any of the cases.
In comparison to PRWT and PRI84V binding, all
compounds within this series reveal a reduced
affinity for PRI50V, similar to the previously studied
series of N-benzyl-substituted derivatives. In contrast
to the N-benzyl-substituted inhibitors, no significant
improvement in affinity for PRI84V is observed for all
the inhibitors comprising 3,3-dimethylallyl substitu-
ents. The substitution with polar substituents in the
para-position of the sulfonyl aryl moieties improves
the binding affinities for this mutation in the case of
the N-benzyl derivatives. However, in the case of
dimethylallyl-substituted inhibitors, particularly for
the most potent derivative 6, no change in the relative
affinity difference can be observed when compared to
the unsubstituted parent structure 5. Compared with
the corresponding benzyl derivative 2, which showed
a 7-fold improvement with respect to PRI84V, the now
4-fold decrease can hardly be explained regarding
only the steric demand of the different N-alkyl
moieties. To rationalize these observations in struc-
tural terms and in order to compare with the known
binding mode of 2, we crystallized compound 6 in
complex with the wild-type HIV protease and
consecutively determined its crystal structure.
Results
Biological evaluation
Determination of affinities for the protease variants
gave rise to a SAR that surprisingly showed no clear
preference for one type of substituents (Table 1).
Whereas (3S,4S)-3,4-bis-[(4-carbamoyl-benzene-
sulfonyl)-(3-methyl-but-2-enyl)-amino]-pyrrolidinium
trifluoroacetate (6) bearing a polar amide functionality
and (3S,4S)-3,4-bis-[(4-bromo-benzenesulfonyl)-
(3-methyl-but-2-enyl)-amino]-pyrrolidinium trifluor-
oacetate (7) bearing a hydrophobic bromide substituent
gain similarly in their affinities for PRWT and PRI84V
,
substitution with a carboxylate functionality in
(3S,4S)-3-[(4-carbamoyl-benzenesulfonyl)-(3-methyl-
but-2-enyl)-amino]-4-[(4-carboxy-benzenesulfonyl)-

748
Multiple Binding Modes of HIV-1 Protease Inhibitor
Fig. 2. The Fo −Fc densities for the
ligands are displayed at a σ level of
2.0 as a blue mesh. (a) Space group
P212121 (green; color coded by atom
type). (b) Space group P6122 (blue;
color coded by atom type).
Structural analysis
Binding mode in the orthorhombic form
In the orthorhombic space group, inhibitor 6
exhibits a binding mode closely related to the one
observed previously for all our pyrrolidine-3,4-bis-
N-benzyl-sulfonamides. The endocyclic amino func-
tionality addresses the catalytic dyad and forms a
hydrogen-bond network with aspartic acid residue
25A (2.8 Å/2.8 Å) and aspartic acid residue 25B
(2.8 Å/2.9 Å). Flap interactions are established via
hydrogen bonding of only one of the inhibitor's
sulfonyl oxygen atoms to the main-chain NH of
Ile50A (2.6 Å). The second sulfonyl oxygen atom of
this sulfonamide group remains uninvolved in any
polar interaction (Fig. 3).
Also, the subsite occupancy is comparable to the
previously determined complexes. Whereas the
dimethylallyl substituents occupy the S1/S1′ sub-
pockets establishing contacts with Gly27A, Gly48A,
Gly49A, Leu23B, Pro81B, Val82B, Ile84B (S1) and
Leu23A, Gly27B, and Ile84A (S1′), the S2/S2′ pockets
are addressed by the phenyl moieties of 6 forming
numerous vdW contacts with Ala28A, Val32A,
Ile47A, Gly48A (S2); and Ile50A, Ala28B, Gly48B,
Ile84B (S2′). The amide functionalities form hydro-
gen bonds with the main-chain NH of Asp30A
(2.6 Å) and Asp30B (2.7 Å), and additionally in the
S2′ pocket to the corresponding Asp30B side chain
(2.8 Å).
Crystals of PRWT in complex with 6 were obtained
by cocrystallization of the enzyme with an inhibitor
concentration of 100 μM and a final dimethyl
sulfoxide (DMSO) concentration of 10%, following
our standard protocol. Crystals grew within a week
and showed a deviating appearance: cubic shaped
at a precipitant concentration of 3.5 M NaCl and
needle-like shaped at a concentration of 3 M NaCl.
Whereas the cubic crystals exhibited space group
P212121, the needle-shaped crystal form corre-
sponded to space group P6122. Both structures
could be determined with a resolution of 1.65 Å,
and ligand geometries were clearly visible in the
Fo −Fc omit map at a σ level of 2 (Fig. 2a and b).
However, two completely different binding modes
were observed and will be described separately:
the geometry in space group P212121 is referred to
as orthorhombic binding mode, whereas the
binding mode in P6122 is named hexagonal
binding mode (Table 2).
Table 2. Crystallographic data
P212121
P6122
Resolution (Å)
Cell dimensions a, b, c (Å)
25–1.65
52.0, 57.6,
61.4
25–1.65
62.5, 62.5,
82.3
The Cα superposition of this structure with the
crystal structure of the corresponding N-benzyl-
substituted inhibitor (PDB ID: 2PWR) reveals that
the overall binding mode is only slightly affected by
the exchange of the N-benzyl moiety with the
N-dimethylallyl moiety, which is reflected by a
root-mean-square deviation (RMSD) of 0.53 between
the Cα atoms of the complexes and by an RMSD of
0.60 Å for the identical ligand atoms of 2 and 6. The
ligand is deeply buried in the binding pocket (90% of
its solvent-accessible surface is buried).
Highest-resolution shell (Å)
Number of measured reflections
Number of independent reflections
Completeness (%)a
1.68–1.65
99,069
1.68–1.65
80,672
22,526
11,774
98.8 [89.4]
26.1 [2.7]
5.2 [35.0]
10–1.65
18.3; 20.2
23.9; 26.2
97.8 [96.2]
23.2 [4.3]
8.0 [36.8]
10–1.65
22.4; 23.4
27.4; 29.7
30.5
I/σa
Rsym (%)a
Resolution in refinement (Å)
Rcryst (FN4σFo; Fo)
Rfree (FN4σFo; Fo)
Mean B-factor (Å2) (peptide chain A; 18.3; 17.3
peptide chain B)
Main chain (Å2) (peptide chain A;
peptide chain B)
15.6; 15.5
21.2; 19.4
27.5
33.9
Side chain (Å2) (peptide chain A;
peptide chain B)
Binding mode in the hexagonal form
Ligand (Å2)
32.4
19.2
26.1
38.2
—
33.5
Cl (Å2)
The binding mode observed in the hexagonal
space group deviates dramatically from the previ-
ously described binding mode, in agreement with all
examples yet observed for our pyrrolidine-based
HIV protease inhibitors. In the hexagonal case, the
complex of the C2-symmetric protease and the C2-
Water (Å2)
Ramachandran plot
Most favored geometry (%)
Additionally allowed (%)
96.2
3.8
94.9
5.1
Values in brackets refer to the highest-resolution shell.

Multiple Binding Modes of HIV-1 Protease Inhibitor
749
Fig. 3. Crystal structure of 6
(green; color coded by atom type)
in complex with HIV protease
(P212121). The protein backbone
trace is schematically illustrated in
wheat, and the catalytic Asp25A
and Asp25B, Ile50A, Ile50B,
Asp30A, and Asp30B are displayed
in gray (color coded by atom type).
Hydrogen bonds are indicated by
broken lines.
symmetric inhibitor 6 fully agrees with the C2
symmetry (Fig. 4).
substituent (3.0 Å/3.2 Å) of chains A and B,
respectively. The S2 and S2′ pockets are now
occupied by the dimethylallyl moieties establishing
vdW contacts with Ala28, Val32, Ile47, and Ile50 of
chains A and B. Comparison of the complex with the
corresponding complex of the N-benzyl-substituted
inhibitor using a Cα superposition reveals a high
level of similarity of the protein structures reflected
by an RMSD of 0.60 Å, but a high dissimilarity for
the identical ligand atoms (RMSD=5.4 Å). Similarly
to the orthorhombic case, the ligand is deeply buried
in the binding pocket (86% of its solvent-accessible
surface is buried).
In this space group, the asymmetric unit comprises
one monomer of HIV-1 protease. The functional
dimer is generated via a 2-fold crystallographic
symmetry. For purposes of comparison, amino acids
of protein chain A and their symmetry equivalents
will be arbitrarily referred to as 1A–99A and 1B–99B,
respectively. According to the imposed symmetry,
any ligand interactions with the A and B chains are
identical. Similar to the orthorhombic case, the
catalytic aspartates Asp25A and Asp25B are
addressed by the endocyclic pyrrolidine nitrogen of
6 (2.8 Å/2.9 Å). However, this is the only conserved
feature shared by the two binding modes. In contrast
to the orthorhombic mode, each sulfonyl group of 6
forms a direct hydrogen bond with Ile50A and
Ile50B, respectively, with one sulfonyl oxygen atom
of each sulfonamide group (3.0 Å). The para-
carboxamido-benzenesulfonamide moieties occupy
the S1 and S1′ pockets forming vdW interactions
with Leu23, Gly27, Gly49, Pro81, Val82, and Ile84,
and polar interactions with Arg8, with the amide
Discussion
Investigation of the inhibitory potency of a series of
pyrrolidine-based inhibitors against mutant HIV
protease variants resulted in a remarkably different
selectivity profile of the N-allyl-substituted derivatives
compared to the N-benzyl-substituted initial lead
structure. To elucidate the reasons for this opposing
Fig. 4. Crystal structure of 6
(light blue; color coded by atom
type) in complex with HIV protease
(P6122). For purposes of compari-
son, amino acids of protein chain A
and their symmetry equivalents
will be arbitrarily referred to as
1A–99A and 1B–99B, respectively.
The protein backbone trace is sche-
matically illustrated in wheat, and
the catalytic Asp25A and Asp25B,
Ile50A, Ile50B, Arg8A, and Arg8B
are displayed in gray (color coded
by atom type). Hydrogen bonds are
indicated by broken lines.

750
Multiple Binding Modes of HIV-1 Protease Inhibitor
behavior especially against the Ile84Val mutant, we
synthesized a small series of inhibitors based on the
most potent derivative 5 of the initial series.
also under biologically relevant conditions. This
binding mode reveals an inhibitor orientation much
closer to that found for all approved peptidomimetic
inhibitors, which are all known to exhibit suscepti-
bility towards the I84V substitution (Fig. 5b and d).
The commonly accepted assumption that similar
ligands bind in a similar fashion can definitely be
ruled out in this case, at least considering the
crystalline state. Regarding the two different bind-
ing modes, it is not evident which orientation will be
preferred under biological conditions. The nature of
the substituent and the different binding pocket
shape of the mutant enzymes might induce different
preferences, resulting in an overall nonuniform
SAR. Whether the adopted space groups impose
special symmetry constraints onto the deviating
binding modes or whether the preferred binding
mode is reinforced by the crystal packing cannot be
resolved at this point. A comparison of protein–
protein contacts due to crystal packing, as well as
the additional observation of two chloride ions
present in the P212121 complex but missing in the
P6122 packing, which exhibit at least a distance of
10 Å to the closest ligand atom, however, does not
satisfactorily explain why the binding mode ob-
served in the orthorhombic case should not be
possible in the hexagonal P6122 space group and
vice versa. Both binding modes seem to possess a
very close-by energy content; however, they obvi-
ously trigger spontaneously by crystallization.
However, the nonuniform SAR obtained within
this series, comprising substituents with different
physicochemical properties in the para-position of the
inhibitor's sulfonamide aryl groups, did not match the
activity pattern expected from the binding mode of
the N-benzyl derivatives. For a detailed analysis of
this phenomenon and for a comparison with the
corresponding N-benzyl derivative 2, 6 in complex
with HIV protease was selected for crystal structure
determination. To our surprise, protein crystals were
obtained in two crystal forms under very similar
crystallization conditions. Both exhibited a space
group deviating from the former series, in which 10
complexes of N-benzyl-substituted inhibitors
revealed identical crystal packing in P21212.
The analysis of the complex in space group
P212121 revealed a binding mode closely resembling
the one observed previously for all examples of the
N-benzyl series. Due to the similarity of the surface
complementarities observed for this orthorhombic
binding mode to those previously determined for 2,
particularly in the contact area to Ile84 (Fig. 5a
and c), a convincing explanation for the deviating
selectivity profile is difficult to provide.
However, the structural evidence found in the
second hexagonal crystal form suggests that a
different binding mode can possibly be adopted
Fig. 5. Crystallographically observed binding modes in HIV protease in equal orientation and size. (a–d) Surface
representation of the ligand (color coded by atom type) and the corresponding protein structure (wheat); view from the
top with the flap region clipped off. Ile84 is highlighted in hot pink. (a) Orthorhombic binding mode of 6. (b) Hexagonal
binding mode of 6. (c) Binding mode of 2 (PDB ID: 2PWR). (d) Binding mode of amprenavir (PDB ID: 1HPV). (e)
Superposition of the ligand orientations observed in the corresponding orthorhombic (green) and hexagonal (blue) crystal
structures of 6 aligned by a Cα fit of the protein coordinates.

Multiple Binding Modes of HIV-1 Protease Inhibitor
751
swap from one to the other in order to escape
resistance development. A similar concept has been
followed during the development of TMC125-
R165335 (etravirine) and TMC120-R147681 (dapivir-
ine), two novel allosteric site inhibitors of the reverse
transcriptase with remarkable resistance profile.12
The authors describe such an advantage to overcome
resistance development by the ability to bind the
target enzyme in multiple conformations and to more
efficiently escape structural modifications of the
resistance-mutated protein.
Summary and Conclusion
Recently, we developed N-benzyl-substituted
HIV protease inhibitors and investigated in detail
their behavior toward the active-site mutants PRI50V
and PRI84V. The observation of an overall increase in
affinity for PRI84V was, among others, attributed to
relief of steric strain present in the wild-type
complexes. The series 3–9 presented in this contri-
bution was intended to elucidate whether smaller
hydrophobic residues might be better suited to
address the S1 and S1′ pockets appropriately.
However, an inconsistent SAR demanded the
determination of a cocrystal structure of at least
one derivative bearing such modifications. Remark-
ably, two different crystal structures of the most
potent derivative 6, resulting from two different
crystal forms found serendipitously, could be
determined. Remarkably, they could be grown
under very similar crystallization conditions, exhi-
biting only a tiny concentration difference in the
precipitating NaCl concentration. Both crystal forms
exhibited comparable diffraction qualities; in each
case, a well-defined but distinct inhibitor orientation
could be identified. To our surprise, the complexes
of identical overall compositions experience differ-
ent protein–inhibitor interactions. Only the interac-
tions with the catalytic dyad are similar, along with
a comparable overall burial of the ligands. The
observation of two different binding modes for 6
suggests that also other members of this series could
occur with more than one binding mode, thus
explaining the nonconsistent SAR within the series.
Are the results obtained by this study a peculiar,
rarely occurring phenomenon or are they of funda-
mental importance also for other structure-based
drug discovery projects? There might be some
statistical evidence for the widely accepted opinion
that similar ligands bind in similar fashion; however,
this is by no means a “truism.” The question remains:
How often does a similar behavior, as found in this
study, remain unnoticed in a congeneric ligand
series? Thoroughly and exhaustively performed
structural biology has to be an integral part of drug
discovery projects and has to escort the entire
optimization process. The same holds for biological
evaluation, which has to be interpreted in a very
careful mannerto identify possible changes in binding
orientations during lead optimization. In fact, the
observation of a different binding mode for members
of a series might even provide unexpected opportu-
nities to optimize ligands towards a particular
binding profile. Furthermore, such ligands might
exhibit a remarkable advantage to escape resistance
development. Likely, both binding modes are close in
energy. Mutational changes can affect and influence
either of the two modes and can change the energetic
balance. However, an inhibitor that is capable of
binding in two close-by alternative modes might
Experimental
Synthesis
Inhibitors 1–5 have been synthesized as previously
described.10 Inhibitors 6–9 were prepared from
the common intermediate (3S,4S)-3,4-bis-[(4-bromo-benze-
nesulfonyl)-(3-methyl-but-2-enyl)-amino]-pyrrolidine-1-car-
boxylic acid t-butyl ester (12) following the sequence
outlined below. The reaction of 3,4-diamino-pyrrolidine
(10) with 4-bromo-phenylsulfonyl chloride renders the
corresponding sulfonamide (3S,4S)-3,4-bis-(4-bromo-benze-
nesulfonylamino)-pyrrolidine-1-carboxylic acid t-butyl ester
(11), which is then alkylated with 3,3-dimethylallyl bromide,
thus leading to the key intermediate 12 (Fig. 6).
Deprotection of 12 under acidic nonaqueous conditions
renders 7. Substitution of the 4-bromo substituent employ-
ing Zn(CN)2,Pd[P(C6H5)3]4 under microwave irradiation
required prolonged reaction times and high temperatures,
which resulted in a concomitant cleavage of the BOC-
protecting group and thus directly furnished inhibitor 8.13
Intermediate 8 was further utilized for the preparation of
the inhibitors 6 and 9. Saponification of the nitrile
functionality in 8 under basic conditions and microwave
irradiation yielded the corresponding acid 9. Mild
hydrolysis with 30% hydrogen peroxide in DMSO/
K2CO3 gave rise to the carboxamido derivative 6.14
General
Reported yields refer to the analytically pure product
obtained by column chromatography. Flash chromatog-
raphy was performed using silica gel 60 (0.04–0.063 mm)
purchased from Macherey-Nagel. For reversed-phase
chromatography, prepacked RP-18 columns (Redisep-C-
18, 13 g) were used. Commercially available solvents and
reagents were used without further purification. Micro-
wave reactions were performed in the laboratory micro-
wave system Discover (LabMate) from CEM. All proton
and carbon NMR spectra were recorded on a Jeol Eclipse+
Spectrometer. 1H NMR spectra were referenced to CDCl3
(7.26 ppm) or DMSO-d6 (2.50 ppm). 13C NMR spectra
were referenced to CDCl3 (77.00 ppm) or DMSO-d6
(39.70 ppm). The values of chemical shifts (δ) are given
in parts per million, and coupling constants (J) are given in
hertz (br=broad; ps=pseudo; s=singlet; d=doublet;
t=triplet; q=quartet; sm=symmetric multiplet; m=mul-
tiplet). Mass spectra were obtained from a double-

752
Multiple Binding Modes of HIV-1 Protease Inhibitor
Fig. 6. Preparation of the inhibitors. The BOC-protected 3,4-diamino-pyrrolidine was prepared from D-(−)-tartaric acid,
as previously described.10 (a) 4-Br-PhSO2Cl, NEt3, CH2Cl2, 65%. (b) 3,3-Dimethylallyl bromide, K2CO3, CH3CN, 0–20 °C,
79%. (c) CH2Cl2/CF3COOH 1:1, 74%. (d) Zn(CN)2, Pd(PPh3)4, dimethylformamide, μW, 51%. (e) 30% H2O2, K2CO3,
DMSO, 38%. (f) NaOH, H2O, μW, 46%.
focusing sector field Micromass VG-Autospec spectrom-
eter. Combustion analyses were determined on a Vario
Micro Cube (Elementar Analysen GmbH). Infrared
spectra were obtained on a Nicolet FT-IR Spectrometer
510 P.
(3S,4S)-3,4-bis-(4-Bromo-benzenesulfonylamino)-
pyrrolidine-1-carboxylic acid t-butyl ester (11)
A solution of triethylamine (1.87 g, 18.5 mmol, 2.4 Eq)
was added to an ice-cold solution of 4-bromo-

Multiple Binding Modes of HIV-1 Protease Inhibitor
753
benzenesulfonyl chloride (4.33 g, 17.0 mmol, 2.2 Eq) in
dichloromethane (40 mL), followed by the dropwise
addition of 10 (1.64 g, 7.71 mmol, 1.0 Eq) dissolved in
dichloromethane (5 mL), with the temperature main-
tained. The reaction mixture was allowed to reach
ambient temperature and was stirred for an additional
12 h. The reaction was quenched with water and washed
three times with 10 mL of an aqueous solution of HCl
(1%). The combined organic layers were dried over
MgSO4, filtered, and concentrated under reduced pres-
sure. Column chromatography (3:1 t-butyl methyl ether/
hexanes containing 0.1% of triethylamine) rendered 11
(3.23 g, 65%): 1H NMR (400 MHz, CDCl3) δ=7.86–7.62
(brd, 8H), 5.43 (brs, 1H), 5.26 (brs, 1H), 3.72–3.44 (brs,
4H), 3.08–2.94 (brs, 2H), 1.39 (s, 9H); 13C NMR
(100 MHz, CDCl3) δ=153.2, 140.2, 132.5, 128.7, 128.6,
126.6, 78.9, 56.8, 56.0, 48.7, 48.3, 28.1; mass spectrometry
(MS) [electrospray ionization (ESI)] m/z (%): 662 (100)
[M+Na]+; analysis calculated for C21H25Br2N3O6S2: C
39.45, H 3.94, N 6.57; found: C 39.87, H 4.21, N 6.62.
2.3 Hz, 4H), 4.84 (brt, J=6.6 Hz, 2H), 4.66 (brt, J=6.1 Hz,
2H), 3.79 (d, J=6.6 Hz, 4H), 3.26–3.19 (sm, 2H), 3.01–2.92
(sm, 2H), 1.64 (s, 6H), 1.56 (s, 6H); 13C NMR (100 MHz,
2
DMSO-d6) δ=158.7 (q, JC,F =31.6 Hz), 139.7, 135.3, 132.5,
129.2, 127.1, 120.6, 117.2 (q, 1JC,F =298.1 Hz), 56.1, 42.4, 42.0,
25.5, 17.7; MS (ESI) m/z (%): 676 (34) [M+H]+; analysis
calculated for C26H33Br2N3O4S2·CF3COOH: C 42.60, H
4.34, N 5.32; found: C 42.88, H 4.63, N 5.22.
(3S,4S)-3,4-bis-[(4-Cyano-benzenesulfonyl)-
(3-methyl-but-2-enyl)-amino]-pyrrolidinium
trifluoroacetate (8)
Zinc cyanide (130 mg, 1.1 mmol, 1.0 Eq) and tetrakis
(triphenylphosphine)palladium (38 mg, 0.033 mmol,
0.03 Eq) were added to a solution of 12 (854 mg,
1.1 mmol, 1.0 Eq) in dimethylformamide (5 mL) under a
positive stream of argon. The reaction vessel was sealed,
and the reaction mixture was heated in a laboratory
microwave for 30 min at 175 °C and 300 W. After having
cooled down to ambient temperature, the reaction mixture
was diluted with t-butyl methyl ether (30 mL) and washed
twice with water (20 mL) and brine. The organic layer was
dried over MgSO4, filtered, and concentrated under
reduced pressure. Reversed-phase column chromatography
of the oily residue using a gradient of water/CH3CN
containing 0.1% of CF3COOH (100% water within 5 min,
linear gradient to 100% CH3CN within 55 min, then
additional 10 min of pure CH3CN) afforded 8 (385 mg,
51%) as its trifluoroacetate salt: 1H NMR (500 MHz,
CD3OD) δ=8.02 (psd, J=8.7 Hz, 4H), 7.97 (psd, J=8.7 Hz,
4H), 4.86–4.79 (m, 4H), 3.95 (dd, J=16.3 Hz, 7.3 Hz, 2H), 3.87
(dd, J=16.4 Hz, 6.4 Hz, 2H), 3.54–3.48 (m, 2H), 3.29–3.25 (m,
2H), 1.70 (s, 6H), 1.60 (s, 6H); 13C NMR (100 MHz, DMSO-
(3S,4S)-3,4-bis-[(4-Bromo-benzenesulfonyl)-
(3-methyl-but-2-enyl)-amino]-pyrrolidine-1-
carboxylic acid t-butyl ester (12)
Anhydrous K2CO3 (1.10 g, 8.0 mmol, 1.6 Eq) was added
to a solution of sulfonamide 11 (3.20 g, 5.0 mmol, 1.0 Eq) in
acetonitrile (40 mL), and the resulting mixture was cooled
to 0 °C, upon which dimethylallyl bromide (1.79 g, 12.0
mmol, 1.2 Eq) was carefully added dropwise. The reaction
mixture was allowed to reach ambient temperature, and
stirring was continued for 12 h. After addition of 20 mL
of t-butyl methyl ether, the reaction mixture was filtered,
and the filtrate was concentrated under reduced pressure.
Column chromatography (5:1 hexanes/t-butyl methyl
2
d6) δ=158.8 (q, JC,F=33.6 Hz), 144.7, 135.8, 133.5, 127.8,
1
120.2, 117.8, 116.6 (q, 1JC,F=295.2 Hz), 115.5, 56.3, 42.4, 42.1,
ether) of the oily residue gave rise to 12 (3.08 g, 79%): H
25.4, 17.7; MS (ESI) m/z (%): 568 (100 [M+H]+; infrared
NMR (400 MHz, DMSO-d6) δ=7.81 (brd, J=8.7 Hz, 4H),
7.76–7.70 (brd, 4H), 4.89 (brs, 1H), 4.74 (brs, 1H), 4.61 (brs,
2H), 3.83 (brs, 2H), 3.72 (brs, 2H), 3.25 (dd, J=10.3 Hz,
8.2 Hz, 2H), 3.04–2.86 (brs, 2H), 1.62 (s, 6H), 1.56 (s, 6H),
1.35 (s, 9H); 13C NMR (100 MHz, DMSO-d6, rotamers)
δ=153.3, 140.2, 134.7, 132.3, 129.2, 126.8, 121.4, 121.0, 79.1,
56.9, 56.4, 44.4, 44.0, 41.8, 28.2, 25.5, 17.7; MS (ESI) m/z (%):
798 (100) [M+Na]+; analysis calculated for C31H41Br2N3-
O6S2: C 48.01, H 5.33, N 5.42; found: C 48.23, H 5.42, N 5.31.
(KBr): ν=2200 (CN) cm−1
.
(3S,4S)-3,4-bis-[(4-Carbamoyl-benzenesulfonyl)-
(3-methyl-but-2-enyl)-amino]-pyrrolidinium
trifluoroacetate (6)
A solution of 8 (171 mg, 0.25 mmol, 1.0 Eq) in DMSO was
cooled to 0 °C. An aqueous H2O2 solution (30%, 1.5 mL)
and K2CO3 (507 mg, 0.38 mmol, 1.5 Eq) were added to this
frozen mixture, and the reaction mixture was allowed to
slowly reach ambient temperature. Stirring was continued
for an additional 30 min. The reaction mixture was again
cooled to 0 °C and quenched by the addition of water
(50 mL), upon which the product precipitated. The
precipitate was collected by filtration, dissolved in a
small amount of methanol, and further purified by
reversed-phase chromatography using the same gradient
as described above, giving rise to 6 (68 mg, 38%): 1H
NMR (500 MHz, DMSO-d6) δ=9.01 (brs, 2H), 8.20 (brs,
2H), 8.07 (d, J=8.7 Hz, 4H), 7.87 (d, J=8.5 Hz, 4H),
7.63 (brs, 2H), 4.86 (pst, J=6.7 Hz, 2H), 4.73 (brt, J=6.2 Hz,
2H), 3.81 (brd, J=6.6 Hz, 4H), 3.27–3.19 (m, 2H), 3.01–2.92
(m, 2H), 1.63 (s, 6H), 1.54 (s, 6H); 13C NMR (100 MHz,
DMSO-d6) δ=166.5, 158.8 (q, 2JC,F =31.6 Hz), 142.6, 138.2,
(3S,4S)-3,4-bis-[(4-Bromo-benzenesulfonyl)-
(3-methyl-but-2-enyl)-amino]-pyrrolidinium
trifluoroacetate (7)
Neat trifluoroacetic acid (17.1 mg, 11.2 μL, 0.15 mmol,
1.0 Eq) was added to a solution of 12 (117 mg, 0.15 mmol,
1.0 Eq) in dichloromethane (3 mL), and the resulting
reaction mixture was stirred under a positive stream of
argon for 12 h at ambient temperature, upon which the
reaction mixture was concentrated in vacuo. Reversed-
phase column chromatography using a gradient of water/
CH3CN containing 0.1% of CF3COOH (100% water within
5 min, linear gradient to 100% CH3CN within 55 min, then
additional 10 min of pure CH3CN) afforded 7 (88 mg, 74%)
1
as its trifluoroacetate salt: H NMR (500 MHz, DMSO-d6)
1
δ=7.83 (dt, J=8.7 Hz, 2.5 Hz, 4H), 7.70 (dt, J=8.9 Hz,
135.3, 128.5, 127.2, 120.6, 117.1 (q, JC,F =299.0 Hz), 56.2,

754
Multiple Binding Modes of HIV-1 Protease Inhibitor
42.4, 42.0, 25.5, 17.7; MS (ESI) m/z (%): 604 (100) [M+H]+;
analysis calculated for C28H37N5O6S2·CF3COOH·H2O: C
48.97, H 5.48, N 9.52; found: C 49.12, H 5.44, N 9.31.
traacetic acid, and 1 mM DTT] at an HIV protease
concentration of 7 mg/mL. Crystals corresponded to
space group P212121 at a precipitant concentration of
3.5 M NaCl and to space group P6122 at 3 M NaCl. For
cryoprotection, both crystal forms were briefly soaked in
mother liquor containing 25% glycerol.
(3S,4S)-3-[(4-Carbamoyl-benzenesulfonyl)-
(3-methyl-but-2-enyl)-amino]-4-[(4-carboxy-
benzenesulfonyl)-(3-methyl-but-2-enyl)-amino]-pyrro-
lidinium trifluoroacetate (9)
Data collection, phasing, and refinement
An aqueous NaOH solution (25%, 2 mL) was added to a
solution of 8 (123 mg, 0.18 mmol, 1.0 Eq) in 2-propanol
(2 mL). The reaction vessel was sealed, and the reaction
mixture was heated in a laboratory microwave for 90 min
at 100 °C and 300 W. After having cooled down to ambient
temperature, the reaction mixture was neutralized using
aqueous HCl. The solution was concentrated in vacuo to a
residual volume of 0.5 mL. Reversed-phase column
chromatography using the same gradient as described
above afforded 9 (60 mg, 46%) as its trifluoroacetate salt:
1H NMR (500 MHz, DMSO-d6) δ=8.12 (d, J=8.5 Hz, 4H),
7.90 (d, J=8.5 Hz, 4H), 4.82 (t, J=6.5 Hz, 2H), 4.73 (t,
J=6.2 Hz, 2H), 3.82 (d, J=6.6 Hz, 4H), 3.24 (dd, J=11.3 Hz,
7.5 Hz, 2H), 2.98 (dd, J=11.1 Hz, 9.0 Hz, 2H), 1.63 (s, 6H),
1.53 (s, 6H); 13C NMR (100 MHz, DMSO-d6) δ=166.3,
The data sets were collected at the synchrotron BESSY II
(Berlin, Germany) on PSF beamline 14.2. Data were
processed and scaled with Denzo and Scalepack, as
implemented in HKL2000.16The structures were deter-
mined by the molecular replacement method with
Phaser;17 one monomer of the 1.5-Å structure of the HIV-
1 protease in complex with a pyrrolidine-based inhibitor
(PDB ID: 2PQZ) was used as the search model. Refinement
was continued with SHELXL-97;18 for each refinement
step, at least 10 cycles of conjugate gradient minimization
were performed, with restraints on bond distances, angles,
and B-values. Intermittent cycles of model building were
performed with the program Coot.19
2
158.9 (q, JC,F =31.6 Hz), 144.0, 135.4, 134.8, 130.2, 127.4,
Surface calculations
120.5, 117.2 (q, 1JC,F =298.1 Hz), 56.2, 42.4, 41.9, 25.4, 17.7;
MS (ESI) m/z (%): 606 (64) [M+H]+; analysis calculated for
C28H35N3O8S2·CF3COOH: C 50.06, H 5.04, N 5.84; found:
C 50.10, H 5.23, N 5.82.
The buried surface area was calculated with the
program MS developed by Connolly.20 The program
calculates the molecular surface from atom coordinates;
the probe radius was set to 1.4 Å.20
Kinetic assay
Enzymatic assays were performed in 172 μL of assay buffer
[100 mM 4-morpholineethanesulfonic acid, 300 mM KCl,
5 mM ethylenediaminetetraacetic acid, and 1 mg/mL bovine
serum albumin (pH 5.5)] by the addition of substrate
dissolved in 4 μL of DMSO, distinct inhibitor concentrations
dissolved in 4 μL of DMSO, and 20 μL of HIV-1 protease in
assay buffer, to a final volume of 200 μL (final DMSO
concentration, 4%). The fluorogenic anthranilyl-HIV protease
substrate (Abz-Thr-Ile-Nle-(p-NO2-Phe)-Gln-Arg-NH2) was
purchased from Bachem. The hydrolysis of the anthra-
nilyl-HIV protease substrate was recorded as the increase
in fluorescence intensity (excitation wavelength, 337 nm;
emission wavelength, 410 nm).15 The kinetic parameters of
PRWT (Km=14.6 μM), PRI50V (Km=139 μM), and PRI84V
(Km=70 μM) were determined by the Lineweaver–Burk
method. IC50 values were generated by a nonlinear
regression analysis from plots of vi/v0 versus inhibitor
concentration, in which vi is the velocity in the presence of
an inhibitor, and v0 is the velocity in the absence of an
inhibitor. Ki values were calculated from the following
equation: Ki=IC50/[1+(S/Km)].
PDB accession numbers
The coordinates have been deposited in the PDB‡ with
access codes 3CKT (orthorhombic) and 2ZGA (hexagonal).
Acknowledgements
We gratefully acknowledge financial support
from the joint DFG research projects KL1204/5-3
and DI827/3-2, and from the Fonds der Chemischen
Industrie (A.B.).
We acknowledge the support of the beamline staff
at BESSY II, and we thank the Bundesministerium
für Bildung und Forschung (05 ES3XBA/5) for
financing the travel costs to BESSY II. The pET11a
plasmid containing the HIV-1 protease gene was
kindly provided by Professor Helena Danielson
(University of Uppsala, Uppsala, Sweden).
Crystallization of HIV-1 protease
Crystals of HIV-1 protease in complex with 6 were
obtained by cocrystallization using the sitting-drop vapor
diffusion method. The well buffer [0.1
M 2-[bis(2-
hydroxyethyl)amino]-2-(hydroxymethyl)propane-1,3-diol
(pH 6.5) and 3–3.5 M NaCl] was mixed with 1 μL of protein
solution [50 mM NaAc (pH 6.5), 1 mM ethylenediaminete-
‡ http://www.rcsb.org/pdb/

Multiple Binding Modes of HIV-1 Protease Inhibitor
755
C2-symmetric HIV-1 protease inhibitors based on a
pyrrolidine scaffold. J. Med. Chem. 51, 2078–2087.
11. Böttcher, J., Blum, A., Heine, A., Diederich, W. E. &
Klebe, G. (2008). Structural and kinetic analysis of
pyrrolidine-based inhibitors of the drug-resistant
Ile84Val mutant of HIV-1 protease. J. Mol. Biol. 383,
347–357.
12. Das, K., Clark, A. D., Lewi, P. J., Heeres, J., deJonge,
M. R., Koymans, L. M. H. et al. (2004). Roles of
conformational and positional adaptability in struc-
ture-based design of TMC125-R165335 (etravirine)
and related non-nucleoside reverse transcriptase in-
hibitors that are highly potent and effective against
wild-type and drug-resistant HIV-1 variants. J. Med.
Chem. 47, 2550–2560.
13. Alterman, M. & Hallberg, A. (2000). Fast microwave-
assisted preparation of aryl and vinyl nitriles and the
corresponding tetrazoles from organo-halides. J. Org.
Chem. 65, 7984–7989.
14. Katritzky, A., Pilarski, B. & Urogdi, L. (1989).
Efficient conversion of nitriles to amides with basic
hydrogen peroxide in dimethyl sulfoxide. Synthesis,
12, 949–950.
References
1. Beddell, C. R., Goodford, P. J., Norrington, F. E.,
Wilkinson, S. & Wootton, R. (1976). Compounds
designed to fit a site of known structure in human
haemoglobin. Br. J. Pharmacol. 57, 201–209.
2. Bernstein, F. C., Koetzle, T. F., Williams, G. J. B.,
Meyer, E. F., Jr., Brice, M. D., Rodgers, J. R. et al. (1977).
The Protein Data Bank: a computer-based archival file
for macromolecular structures. J. Mol. Biol. 112,
535–542.
3. Hubbard, R. E. (2005). 3D structure and the drug-
discovery process. Mol. BioSyst. 1, 391–406.
4. Anderson, A. C. (2003). The process of structure-based
drug design. Chem. Biol. 10, 787–797.
5. Suvit Thaisrivongs, J. W. S. (1999). Structure-based
discovery of tipranavir disodium (PNU-140690E): a
potent, orally bioavailable, nonpeptidic HIV protease
inhibitor. Pept. Sci. 51, 51–58.
6. Reich, S. H., Melnick, M., Davies, J. F., II, Appelt, K.,
Lewis, K. K., Fuhry, M. A. et al. (1995). Protein
structure-based design of potent orally bioavailable,
nonpeptide inhibitors of human immunodeficiency
virus protease. Proc. Natl Acad. Sci. USA, 92,
3298–3302.
15. Toth, M. V. & Marshall, G. R. (1990). A simple,
continuous fluorometric assay for HIV protease. Int. J.
Pept. Protein Res. 36, 544–550.
7. Stubbs, M. T., Reyda, S., Dullweber, F., Möller, M.,
Klebe, G., Dorsch, D. et al. (2002). pH-dependent
binding modes observed in trypsin crystals: lessons
for structure-based drug design. ChemBioChem, 3,
246–249.
8. Gassel, M., Breitenlechner, C. B., Konig, N., Huber,
R., Engh, R. A. & Bossemeyer, D. (2004). The
protein kinase C inhibitor bisindolyl maleimide 2
binds with reversed orientations to different con-
formations of protein kinase A. J. Biol. Chem. 279,
23679–23690.
16. Otwinowski, Z. & Minor, W. (1997). Processing of
X-ray diffraction data collected in oscillation mode.
In Methods in Enzymology (Carter, C. W., ed.), Vol.
276, pp. 307–326. Academic Press, New York, NY.
17. Storoni, L. C., McCoy, A. J. & Read, R. J. (2004).
Likelihood-enhanced fast rotation functions. Acta
Crystallogr. Sect. D, 60, 432–438.
18. Sheldrick, G. M. & Schneider, T. R. (1997). SHELXL:
high-resolution refinement. In Methods in Enzymology
(Charles, W. C. J. & Robert, M. S., eds), Vol. 277,
pp. 319–343. Academic Press, New York, NY.
19. Emsley, P. & Cowtan, K. (2004). Coot: model-building
tools for molecular graphics. Acta Crystallogr. Sect. D,
60, 2126–2132.
9. Bostrom, J., Hogner, A. & Schmitt, S. (2006). Do
structurally similar ligands bind in a similar fashion?
J. Med. Chem. 49, 6716–6725.
10. Blum, A., Böttcher, J., Heine, A., Klebe, G. &
Diederich, W. E. (2008). Structure-guided design of
20. Connolly, M. (1983). Analytical molecular surface
calculation. J. Appl. Crystallogr. 16, 548–558.