Identification of Hot Spots within Druggable Binding Regions by Computational Solvent
Mapping of Proteins
Melissa R. Landon,†David R. Lancia, Jr.,‡Jessamin Yu,†Spencer C. Thiel,‡and Sandor Vajda*,§
Bioinformatics Graduate Program, Boston UniVersity, 24 Cummington Street, Boston, Massachusetts, Department of Biomedical Engineering,
Boston UniVersity, 44 Cummington Street, Boston, Massachusetts, SolMap Pharmaceuticals, Inc., 196 Broadway, 2nd Floor,
ReceiVed September 28, 2006
Here we apply the computational solvent mapping (CS-Map) algorithm toward the in silico identification
of hot spots, that is, regions of protein binding sites that are major contributors to the binding energy and,
hence, are prime targets in drug design. The CS-Map algorithm, developed for binding site characterization,
moves small organic functional groups around the protein surface and determines their most energetically
favorable binding positions. The utility of CS-Map algorithm toward the prediction of hot spot regions in
druggable binding pockets is illustrated by three test systems: (1) renin aspartic protease, (2) a set of previously
characterized druggable proteins, and (3) E. coli ketopantoate reductase. In each of the three studies, existing
literature was used to verify our results. Based on our analyses, we conclude that the information provided
by CS-Map can contribute substantially to the identification of hot spots, a necessary predecessor of fragment-
based drug discovery efforts.
Characterization of protein-ligand binding sites has been
focused recently on the identification of hot spots, subsites of
ligand binding regions of proteins that are major contributors
to the binding energy.1,2Druggable binding pockets generally
contain smaller regions that are crucial to the binding of
functional groups and, hence, are the prime targets in drug
design. As developed over the past decade and summarized in
a recent paper,3NMR-based screening of a variety of protein
targets with a large compound library demonstrates that the hot
spot regions bind a large variety of small molecules. Although
the binding of most compounds is weak, the authors also
determined that a relatively high hit rate is predictive of target
sites that are likely to bind inhibitors with high affinity. A more
direct approach to the identification of hot spots within the
NADPH binding region of ketopantoate reductase (KPR) was
recently described Ciulli et al.4The strategy involved the
breaking down of NADPH into smaller fragments and the
subsequent characterization of each fragment’s ligand efficiency
and binding specificity using isothermal titration calorimetry,
NMR spectroscopy, and inhibition studies. Despite the decreased
affinity of smaller chemical fragments in a binding pocket as
compared to larger molecules, the study by Ciulli et al.4made
apparent that most of the binding free energy comes from select
regions of the binding pocket and that the binding modality of
NADPH fragments in a binding pocket can be predictive of
Despite tremendous progress in methodology, the identifica-
tion of hot spot regions by NMR, X-ray crystallography, or
biophysical methods is still expensive and time-consuming. Here
we describe a computational alternative, namely, computational
solvent mapping (CS-Map), and show that we can reliably
reproduce the available experimental results,5-7hence providing
an effective alternative to experimental hot spot identification.
The CS-Map algorithm moves small organic functional groups
around the protein surface to find their most favorable binding
positions. The direct motivation for developing the method was
to replicate in silico the results obtained by the multiple solvent
crystal structures (MSCS) method introduced by Mattos and
Ringe.8-10The MSCS method involves the soaking of a protein
in a series of organic solvents, whereupon binding regions can
be determined based on the crystallization of the protein in each
solvent and subsequent superimposition of the X-ray structures
to identify regions where multiple solvent molecules bind on
the protein; the authors observed that these regions of aggrega-
tion typically correspond very well with known binding regions.
While numerous computational techniques exist for the iden-
tification of binding sites on proteins,11-16it appears that CS-
Map is the first algorithm shown to reliably reproduce the results
of such MSCS experiments. Previously, the CS-Map algorithm
was applied to a variety of well-characterized proteins5-7to
identify the regions of their binding pockets that are most
important for the binding of small molecules; these results were
in good agreement with those determined by MSCS and other
experimental approaches. The advantage of using CS-Map
versus other methods rests primarily in the utilization of a library
of solvent-like molecules as probes, making it analogous to the
MSCS approach or to other biophysical-based analyses of the
binding of small organic molecules to proteins.
We used three test systems in this study to determine the
effectiveness of CS-Map in the identification of hot spots for
proteins that are both important pharmaceutical targets and
whose binding pockets have been experimentally characterized.
The test systems include renin aspartic protease, the training
set of proteins used in the Abbott NMR-based screening study,
and KPR. Renin and the angiotensin converting enzyme (ACE)
are long-standing pharmaceutical targets for the treatment of
hypertension.17,18The cleavage by renin of its peptide substrate,
angiotensin, and subsequent formation of angiotensin I precedes
the modification of angiotensin I by ACE. While both renin
and ACE are viable drug targets for the treatment of hyperten-
sion, currently all hypertension therapeutics on the market are
ACE inhibitors.19Inhibitors of renin consist primarily of
* To whom correspondence should be addressed. Phone: +1-617-353-
4757. Fax: +1-617-353-7020. E-mail: email@example.com.
†Bioinformatics Graduate Program, Boston University.
‡SolMap Pharmaceuticals, Inc.
§Department of Biomedical Engineering, Boston University.
10.1021/jm061134b CCC: $37.00 © xxxx American Chemical Society
Published on Web 02/17/2007PAGE EST: 9.8
peptidomimetics, specifically those targeting the S2 and S3
subsites of the angiotensin binding pocket.20,21Despite their high
affinity and specificity for the binding pocket, peptidomimetic
inhibitors exhibit poor pharmacokinetics, largely attributed to
their hydrophilicity and peptide character. For this reason, the
development of drug-like inhibitors of renin has proven to be
In 2000, Novartis Pharmaceuticals in collaboration with
Speedel published a new class of renin inhibitors demonstrating
oral availability.22-24These inhibitors lacked the peptide-like
features of earlier classes of inhibitors, resulting in molecules
that were more drug-like. Additionally, these inhibitors bound
primarily in the S1 and S3 subsites of the binding pocket,
demonstrating a marked departure from earlier efforts. The
authors contended that crucial to the affinity of this class of
inhibitors were interactions made in a previously uncharacterized
region of the binding pocket, named S3SP due to its proximity
to the S3 subsite. Further modification of these molecules has
resulted in the development of aliskiren, a renin inhibitor now
in clinical trials. In our analysis of renin, we sought to predict
the preferential binding pose of aliskiren in terms of key subsite/
residue contacts, as compared to that of the peptidomimetic
inhibitors using CS-Map.
A training set of 23 proteins, comprising 28 binding sites,
was used in the NMR-based study of druggability published
by Fesik and co-workers at Abbott Pharmaceuticals.3The
authors demonstrated a very high correlation between the
frequency of screening hits for a particular binding pocket and
the ability of that pocket to bind lead-like molecules with high
affinity. For those binding regions that were deemed druggable
from the training set, we performed a residue-based analysis of
binding affinity in an effort to predict hot spots within each
Last, we predicted hot spots for E. coli KPR and compared
our results to those recently published by Ciulli et al.4KPR
plays an important role in the biosynthesis of vitamin B5 by
catalyzing the reduction of ketopantoate to pantoate. As vitamin
B5 is an essential nutrient for bacteria, inhibition of enzymes
involved in this pathway could result in novel antibacterial
agents.25,26Using a fragment growth strategy to rebuild the
structure of NADPH in its binding pocket on KPR, Abell and
co-workers exposed via biophysical methods two regions that
are primarily responsible for the binding efficiency of NADPH
on opposite sides of the binding pocket, namely, the regions
that bind the 2′-phosphate and the reduced nicotinamide groups.
To confirm their findings, the authors performed mutational
studies, where a residue located in each putative hot spot, Arg31
and Asn98, were mutated to alanines to demonstrate a loss of
binding affinity upon mutation of residues in these regions. In
the CS-Map based analysis of KPR, we sought to uncover these
same regions of the binding pocket to directly compare our
method to experimental fragment-based approaches for hot spot
CS-Map Algorithm. CS-Map entails a five-step process by
which the energetically favorable binding positions of a series of
small, solvent-like probes are determined for a protein.5-7The
regions in which the highest number of different probe clusters
overlap are then regarded as the hot spots predicted by the method.
Input for the algorithm is an X-ray structure of the protein of interest
in either apo or liganded form. All ligands are removed from bound
complexes prior to initiation of the algorithm. A brief description
of the four algorithmic steps utilized in this study is provided below;
earlier publications should be consulted for a more detailed
explanation. An additional subclustering step, described in previous
mapping studies, was not used in this analysis.
Step 1: Rigid Body Search. A series of fourteen organic probes,
shown in Figure 1, are used to define binding pockets using CS-
Map. Each probe is run separately to avoid clashes. A set of 222
initial probe positions is determined based on a placement algorithm,
where the protein is placed on a dense grid, followed by the
identification of buried and surface points. Surface points in the
vicinity of buried points are considered to be in a pocket-like area
and are used as initial probe positions to ensure the sampling of
more buried regions. Nonpocket-like surface points are then
clustered such that the number of cluster centers obtained is
equivalent to the remaining number of available probe positions; a
probe is then placed at each of these cluster centers.
A multistart simplex method27is used to move the probes to
energetically optimal positions, where a free energy score of the
probe-protein complex is determined by the equation ∆Gs) ∆Eelec
+ ∆Gdes+ Vexc. Here ∆Eelecdenotes the coulombic component of
the electrostatic energy, ∆Gdesdescribes the desolvation free energy,
and Vexcis an excluded volume penalty term that is set to zero if
the probe does not overlap with the protein. The electrostatic term
is calculated as the summation of the products of the charge of
each probe atom and the solvated protein’s electrostatic field at
that position; atom charges are assigned by the partial charges
employed in the Quanta program (http://www.accelyrs.com), while
the electric field term is calculated according to the finite difference
Poisson-Boltzmann method28implemented in the software package
CONGEN (http://www.congenomics.com). The desolvation term,
∆Gdes, is determined using the atomic contact potential model.29
Step 2: Free Energy Refinement and Final Docking. The rigid
body search described above yields over 6000 conformations of
each probe type in complex with the protein. A modified free energy
term is used to further minimize this complex, where now ∆G )
∆Eelec+ ∆Evdw+ ∆G*des. This free energy equation includes a
van der Waals energy term for the protein-ligand complexes,
∆Evdw, and a revised desolvation term, ∆G*des, that includes the
change in solute-solvent van der Waals interaction energy. The
ACE model30implemented in Charmm version 2731is used to
compute the electrostatic and desolvation contributions to the free
energy. Minimization of each protein-probe complex is performed
using the Newton-Raphson method also found in Charmm, where
the protein is held fixed while the probe atoms move freely. At
most, one thousand minimization steps are allowed, although most
complexes require far fewer steps to achieve convergence.
Step 3: Clustering, Scoring, and Ranking. Minimized probe
positions are clustered in an interaction-based fashion where a
contact vector is used to represent the interactions between the
probes and the protein. Each index in u represents a distinct residue
i, and u j(i) is set to 1 if a probe atom is in contact with residue i;
otherwise u j(i) ) 0. Contact is defined as a probe atom being within
6 Å of atoms of the residue under consideration. Distances are
calculated in a pairwise fashion between contact vectors of different
probe molecules, denoted u j and V j, using a normalized distance D
) 1 - u j‚V j/(|u j||V j|). The distance is zero if and only if all interactions
are the same and is equal to one if vectors u j and V j are orthogonal.
Contact vectors are then clustered based on their D-values to all
Figure 1. Set of 14 probes used for binding site identification by the
Journal of Medicinal ChemistryLandon et al.
other vectors. The clustering algorithm works by seeding a cluster
with the unclustered probe with the lowest free energy. Remaining
unclustered probes are searched to find the probe with the lowest
distance score D to the seed probe. If this score is below 0.35, the
probe is added to the cluster. The maximum allowable score for
addition to a cluster, 0.35, was chosen based on a qualitative
analysis. Once more than two probes populate the cluster, additional
probes are added to the cluster by checking that the average distance
score between the new probe and all existing members is below
0.35. If the new probe has an average score above 0.35, the probe
is rejected and a new search begins. After all probes are checked
for admittance to a cluster and no additional probes can be added,
a new cluster begins by repeating the process above. Clusters are
optimized after initial creation by reclustering the probes such that
if a probe could be moved from one cluster to another so that the
average D score would be lowered in both clusters, then the probe
is moved. Small clusters consisting of less than fifteen members
are excluded from consideration. For each of the remaining clusters,
we calculate the probability pi) Qi/Q, where the partition function
Q is the sum of the Boltzmann factors over all conformations, Q
) ∑jexp(-∆Gj/RT), and Qiis obtained by summing the Boltzmann
factors over the conformations in the ith cluster only. The cluster
average of a property x for the ith cluster is calculated by <x>i)
∑jpijxj, where pij) exp(-∆Gj/RT)/Qi, and the sum is taken over
the members of the ith cluster. Average free energy terms <∆G>,
<∆Eelec>, <∆Evdw>, and <∆G*des> are calculated for each cluster;
the average free energy term, <∆G>, is used to rank each cluster.
Step 4: Creation and Ranking of Consensus Sites. The five
to ten lowest average free energy clusters of each probe type are
used to create consensus sites, for example, regions of the protein
where clusters of multiple probe types are located. Superimposition
of these low-energy probe clusters on the protein structure allows
for identification of these consensus sites. Ranking of consensus
sites is based on both the total number of probe clusters that
comprise the site as well as the number of different probe types
represented by the clusters. For example, a consensus site containing
13 of the 14 different probe molecules is ranked higher than a site
with only 7 of the 14 probe molecules if the total number of clusters
is the same. If two sites share the same number of probe types,
then duplicate types within the consensus site are also considered
in the count.
Docking of Aliskiren. We used the GOLD (Genetic Optimiza-
tion for Ligand Docking) program32,33to generate conformations
of aliskiren in the peptide binding of renin. The input structure
was PDB code 1RNE, where the binding pocket was defined as
residues within 10 Å of the R carbon of Asp32. Rings, amides,
and nitrogens bound to sp2 carbons were allowed to rotate during
the optimizations. Ten conformations of aliskiren were generated
using GOLD and then compared to the published conformation.22
Detection of Hot Spots Using CS-Map and HBPLUS. The
HBPLUS program34created by Thornton and co-workers was
utilized to generate both nonbonded and hydrogen-bonded interac-
tions on an atomic level between the residues comprising the
binding pockets of the proteins used in our studies and the low
energy conformations of probe molecules resulting from the
mapping of each protein. For each protein, we defined its binding
pocket as those residues within 6 Å of a bound ligand (see the
Results and Discussion section for a description of each ligand).
Solvent probes that were assigned to the top ten lowest free energy
probe clusters (see Computational Methods section) and were
located in the binding region were used in our analysis. Subsequent
to the calculation of nonbonded and hydrogen-bonded interactions,
the number of atom interactions for each residue was tallied and
normalized by the total number of residue interactions to determine
the percentage of atom interactions for each residue relative to all
residues in the binding pocket. We considered residues accounting
for greater than four percent of the atom interactions in a pocket to
be potentially located in a hot spot. Separate interaction analyses
were performed between each protein and its bound ligand in a
similar manner for comparison.
Results and Discussion
Conformational Analysis of Aliskiren in the Binding
Pocket of Renin. Structural information for the renin-aliskiren
complex is not available currently from the PDB; however, a
figure from the original Novartis publication,22reproduced here
as Figure 2A, illustrates the position of aliskiren, shown in
purple, relative to a peptidomimetic inhibitor, shown in green.
It is apparent from this view that, while aliskiren occupies in
part the same region of the binding pocket as the peptidomimetic
inhibitor does, namely, the S1 and S1′ subsites, significant
deviations in the conformations of the two molecules exist in
the S3 and S2′ subsites and both molecules occupy subsites
uniquely. Specifically, the S2 and S4 subsites are occupied solely
by an imidazole functional group and sulfonyl group, respec-
tively, of the peptidomimetic inhibitor, while a hydrophobic
functional group of aliskiren resides in the S3SP subsite, a region
of the binding site previously unoccupied by inhibitors and
considered to be largely responsible for the affinity of aliskiren.
Docking studies of aliskiren in the renin binding pocket using
PDB code 1RNE were performed using the GOLD docking
program (see Computational Methods) to determine a confor-
mation of aliskiren that exhibited high similarity to the published
conformation. Details regarding key contacts and hydrogen-
bonding positions from the literature were used in conjunction
with qualitative information, specifically Figure 2A, to determine
the most appropriate docked conformation for use in our
analyses. Figure 2B represents the most similar docked con-
formation of aliskiren, colored in purple, in the binding pocket
of renin. The conformation of the same peptidomimetic inhibitor
shown in Figure 2A, for which structural information is
Figure 2. (A) Published figure from Novartis depicting the conformation of aliskiren with respect to a peptidomimetic. Reprinted from Biochemical
and Biophysical Research Communications (Volume 308; J. M. Wood et al.; Structure-Based Design of Aliskiren, a Novel Orally Effective Renin
Inhibitor; pages 698-705), ref 22. Copyright 2003, with permission from Elsevier. Structural data on the renin-aliskiren complex is currently not
available from the Protein Data Bank. (B) Resulting docked conformation of aliskiren, shown in purple, in the peptide-binding pocket of renin. The
relative conformation of the same peptidomimetic from (A) is shown in green. The docked conformation of aliskiren is qualitatively consistent with
the published figure. Key structural features, such as the occupation of the S3SP pocket, conjectured to be responsible for the affinity of aliskiren,
are preserved in the docked conformation.
Hot Spots within Druggable Binding RegionsJournal of Medicinal Chemistry C
available, is shown in green to allow for comparison of the
docked and experimental conformations. Only minor deviations
are apparent between the docked conformation of aliskiren and
the actual conformation. In particular, the regions of the binding
pocket that make significant contact with aliskiren, namely, the
S1, S3, and S2′ regions, are preserved in the docked structure
shown in Figure 2A. The conformation of the chain of aliskiren
that occupies the S3SP pocket is also preserved. The resulting
contact data between aliskiren and renin were used to compare
our mapping results to residue interactions made by aliskiren
and other peptidomimetic inhibitors of renin.
Identification of High Affinity Subsites in the Binding
Pocket of Renin. We mapped five renin structures to predict
favorable binding locations within the peptide binding pocket.
As shown in Figure 3A, all consensus sites that were found to
be located in the binding region are shown superimposed onto
the structure of PDB code 1RNE. Consensus sites are colored
in Figure 3A according to the structure from which they were
derived; structures utilized from the PDB for the study were
1BIL, 1BIM, 1HRN, 1RNE, and 2REN, where the latter
structure represents the unliganded form of renin, while the four
previous structures are bound by peptidomimetic inhibitors.
Given the confinement of all consensus sites to the same region
of the binding pocket, we concluded that small changes in the
conformation of residues in the binding region do not affect
the mapping results significantly.
Figure 3B is a close-up view of the consensus sites located
in the binding pocket. In this depiction, all consensus sites
resulting from the five structures are colored uniformly in light
blue. For clarity, only the probe cluster representatives that
comprise the consensus sites are displayed. The docked
conformation of aliskiren is shown in purple in Figure 3B to
allow for comparison to the mapping results. It is readily
apparent from this view that the consensus sites in the binding
region overlap significantly with the region occupied by
aliskiren, making contacts primarily in the S1, S3, and S2′
subsites. The absence of probe clusters in the S2 and S4 subsites
is illustrated in Figures 3C and 3D, where the four peptidomi-
metic inhibitors taken from the bound PDB structures that were
used for mapping are added in green. Figure 3D is a side view
of the binding region. Both Figures 3C and 3D show that while
each of the peptidomimetic inhibitors makes significant contacts
in the S2 and S4 regions neither aliskiren nor the CS-Map probes
do so to a visible extent. Interestingly, one of the densest regions
of probe molecules is found in the S3SP subsite, a region of
the binding pocket that was described in the Novartis publication
as being unique to the binding modality of aliskiren versus other
The rank(s) of the consensus site(s) occupying the different
subsites of the binding pocket are summarized in Table 1, where
the assignment of residues to a subsite was utilized from a
previous publication.35With the exception of the unbound
structure, the top ranked consensus sites occupy both the S1
and S3 subsites of the binding pocket for each structure; in the
case of the bound structure, the consensus site in the S3 pocket
is first. Conformational changes undergone by aspartyl proteases
upon ligand binding may account for the difference in mapping
results existing between the unbound and the bound structures,
in particular the change in shape of the S3 region of the active
Figure 3. Mapping results for five structures of renin. (A) The first and second ranked consensus sites resulting from the mapping of the different
structures are superimposed in the peptide binding pocket of renin, demonstrating the reproducibility of the results. Each color represents the results
of a distinctive protein. (B) Closer examination of the consensus sites depicted in (A), now all colored light blue and shown in relation to the
docked conformation of aliskiren, supports the importance of the S1, S2, S3, and S3SP subsites for ligand affinity. (C,D) The preferred binding
mode of aliskiren as compared to the peptidomimetics, shown in green, is confirmed by the mapping results. The S2 and S4 subsites are bound
preferentially by the peptidomimetics, but not by aliskiren or the CS-Map probes.
Table 1. Rankings of the Consensus Sites Present in the Subsites of the
Binding Pocket of Renin for Each of the Five Structures Mappeda
PDB/subsiteS4S3 S2 S1S1′
aThe number in parentheses indicates the number of probe clusters used
to create the consensus site.bNP ) not present.cBoth consensus sites have
the same number of clusters.
Journal of Medicinal ChemistryLandon et al.
Conversely, no significantly populated consensus site is
present in the S4 pocket, and only a single, low-ranked
consensus site is found in the S2 subsite. This analysis suggests
that the S2 and S4 subsites bind drug-like functional groups
with lower affinity than the other subsites of the peptide binding
pocket. Based on these results we can conclude that the S1,
S3, and S2′ subsites of the binding pocket are hot spots for
fragment binding and, within the S3 subsite, the S3SP region
displays particularly high affinity for small molecules.
Residue-Based Analysis of Hot Spots in the Renin Binding
Pocket. Subsequent to the characterization of hot spots within
the binding pocket of renin, we applied CS-Map to the
identification of specific residues that are crucial for ligand
affinity within the hot spot regions (see Computational Meth-
ods). Residues were defined as part of the binding pocket if
any of their atoms were within 6 Å of an atom of aliskiren. In
addition, calculating interactions for the probe molecules, we
determined interactions for both aliskiren and the four pepti-
domimetic inhibitors shown in Figure 3C,D for comparison. The
resulting residue-based interaction distributions are shown in
Figure 4, with the residues composing the binding pocket placed
in sequence order on the horizontal axis. Residue interactions
were calculated separately for each peptidomimetic inhibitor
and then averaged to create one value. A high level of agreement
exists between the distributions for aliskiren, shown in dark gray,
and the CS-Map probes, shown in black; in particular, atom
interactions are enriched in both distributions for residues Gly13
and Tyr75, located in the S3SP and S1 subsites, respectively.
However, while both aliskiren and the peptidomimetic inhibitors
interact significantly with the catalytic Asp32, located in the
S1 subsite, the residue was not highly interactive with CS-Map
probes; because both aliskiren and the mapping probes make
the highest level of interactions in the S1 subsite with Tyr75,
this may suggest that while Asp32 is necessary for catalysis, it
may contribute less to the binding affinity of ligands than
As compared to the high level of agreement existing between
aliskiren and the CS-Map probes with respect to the distribution
of residue interactions, the comparison of interaction distribu-
tions between the CS-Map probes and the peptidomimetics
yielded a very low level of correlation. Residues predicted to
be hot spots based on the analysis of interactions made with
the peptidomimetics, shown in Figure 4 as the light gray
distribution, were located primarily in the S2 subsite, such as
residues Ser76 and Ala218. Agreement between the peptido-
mimetics and aliskiren was only seen in the S1 region with the
two catalytic aspartic acids, Asp32 and Asp215. As a quantita-
tive measure of similarity, we calculated Pearson correlation
coefficients (R) in a pairwise fashion for the three distributions,
where values can range from -1 (perfectly anti-correlated) to
1 (perfectly correlated). Assuming that there are 20 residues in
the entire binding site, the threshold for a correlation coefficient
to be significant with a p-value of less than 0.01 is R ) 0.52.
The calculated R-value between the CS-Map probes and
aliskiren was 0.72, significantly higher than the R-value of 0.19
existing between the CS-Map probes and the peptidomimetic
inhibitors. The correlation between aliskiren and the peptido-
mimetics was an intermediate value of 0.53; the main reason
for the differences in correlation existing between the two
different inhibitor types when compared to the mapping results
is due to the affinity of the CS-Map and aliskiren for the S3SP
pocket as compared to peptidomimetic inhibitors. Strong
hydrophobic interactions in this region allow aliskiren to exhibit
high affinity despite its decrease in peptide-like character. The
residues predicted by CS-Map as being highly interactive can
serve as starting points for the development of high-affinity,
Characterization of Druggable Binding Pockets Using CS-
Map. Using those proteins comprising the druggable training
group from the Fesik study,3we sought to identify hot spots
and specific residue interactions important for ligand binding
for a variety of drug targets. Consisting of both druggable and
Figure 4. Distribution of atom-based residue interactions in the binding
pocket of renin for the CS-Map probes, aliskiren, and three different
peptidomimetics. In the case of the peptidomimetics, the average
number of interactions at each residue was utilized. The Pearson
correlation coefficient between the CS-Map probes and aliskiren is 0.72,
while that between the CS-Map probes and the peptidomimetics is 0.19.
Table 2. Summary of Results on the Mapping of Proteins Used as the Training Set in the Druggability Study Published by Fesik and Co-Workersa
proteinsite PDB ID druggable3
consensus site rank(s)
(number of clusters)
consensus site avg
cluster energy rank
1 (26), 1 (23),b2 (15)b
1 (26), 2 (18)b
1 (18), 1 (18)b
1 (33), 2 (19)
5 (10),c5 (9)d
9 (4),c9 (2)d
aThe PDB accession code indicates the protein structure used for mapping.bNMR structure.cClosed conformation of WPD loop.dOpen conformation
of WPD loop.
Hot Spots within Druggable Binding RegionsJournal of Medicinal Chemistry E
non-druggable proteins, this training set was used by Fesik and
co-workers to create a druggability index. In addition to hot
spot identification, we applied CS-Map toward the computa-
tional prediction of druggability based solely on protein
structure, using those structures from the druggability study for
which complete structural coordinates were available from the
Mapping results for the 12 proteins analyzed, consisting of
13 binding sites, are summarized in Table 2, including the rank
of the consensus site found in a binding region and the number
of probe clusters that were used to create the consensus site.
The last column of Table 2 indicates the average energy ranking
values of probes comprising the consensus sites found in the
binding region, where the highest ranked cluster of each probe
type was used to determine the average value and then estimated
to the nearest whole number.
The protein binding pockets determined to be non-druggable
in the experimental study conducted at Abbott include the bir3
binding region of survivin, the phosphatidylinositol (3,4,5)-
trisphosphate binding domain of protein kinase B/AKT, the PZD
domain of protein-95 (PSD95), and the WW domain of Pin1.
As shown by Table 2, the highest ranking consensus site
determined for any of the non-druggable proteins was fourth,
suggesting that a low consensus site ranking is predictive of
binding regions that are not druggable. Conversely, for each
druggable binding pocket from the study, the consensus sites
found in those regions were top ranked based on size and, in
most cases, top ranked based on average cluster energy. To
assess the sensitivity of our results to structure resolution, we
mapped NMR structures available for druggable proteins in the
training set shown in Table 2. Within binding regions, all
mapping results derived from NMR data are in good agreement
with those determined using crystallographic data. Analysis of
this set of proteins suggests that CS-Map can be used to predict
the druggability of a binding region when structural information
is available; that is, those binding regions with top-ranked or
near top-ranked consensus sites in terms of cluster membership
and/or average cluster energy, are characteristic of druggable
The sole exception in our study to both the above-defined
druggability criteria is PTP1B, where two distinct conformations
of the binding region were analyzed using CS-Map. These two
conformations of the binding region result primarily from the
positition of the WPD loop; the loop can either be “closed”,
where the catalytic phosphotyrosyl (pTyr) binding region forms
a small, deep pocket, or “open”, resulting in a shallower catalytic
pTyr binding region that is continuous with the secondary
phosphotyrosyl binding location. Representative structure of
each conformation was chosen from the PDB for our mapping
analysis. A closed conformation structure of PTP1B, PDB
1PTY, is shown in cartoon and surface representations in Figure
5A and C, respectively; the WPD loop is highlighted in magenta
in 5A and the bound phosphotyrosyl groups are outlined in green
in both 5A and 5C. Mapping results for 1PTY are shown
superimposed with the phosphotyrosyl groups in Figure 5C;
although this consensus site is low-ranked in terms of cluster
membership, the small and relatively polar probes, namely,
dimethyl ether, urea, ethanol, acetonitrile, acetone, and acetal-
dehyde, bind preferentially in the catalytic phosphotyrosyl
binding site with cluster energies that are always either first or
second ranked, where the top-ranked cluster is defined as that
with the lowest average free energy of its members. These
Figure 5. Mapping analysis of open and closed conformations of PTP1B. (A) Cartoon depiction of the open conformation of PTP1B (PDB ID
1PTY). Also shown in green are the bound catalytic (left most) and noncatalytic (right side) phosphotyrosyl (pTyr) groups. The closed conformation
of the WPD loop is highlighted in magenta. (B) Structure of the “open” conformation of PTP1B (PDB ID 1PH0), with the WPD loop in magenta
and the bound active site inhibitor in blue. (C,D) Surface representations of the active site of PTP1B derived from (A) and (B), respectively. Shown
bound in (C) are the pTyr groups in green and CS-Map probes in blue. In (D) the active site inhibitor is colored in blue and the CS-Map probes
are colored in yellow. The small, deep nature of the catalytic pTyr site in the closed structure (C) as compared to that of the open conformation (D)
is evident from these representations. Mapping results of the closed conformation (A,C) reveal a low-energy consensus site in the catalytic pTyr
binding pocket and a less favorable region on the far end of the second pTyr binding location. The consensus sites formed in the same regions of
the open conformation of PTP1B are not as energetically favorable, suggesting that a larger molecule may be needed to gain affinity from both
Journal of Medicinal ChemistryLandon et al.
probes are located deep in the catalytic pTyr binding site and,
consequently, are not visible in Figure 5C. Larger polar probes,
such as benzaldehyde and t-butanol, bind at slightly higher
energies. Additionally, larger hydrophobic probes, that is,
cyclohexane and benzene, are either low-ranked or absent in
the active site. Our analysis suggests that the size of a consensus
site is not always enough to label a binding region as druggable
or non-druggable, particularly when the pocket is small and
exhibits a strong preference for probes of a singular type.
A structure of the open conformation of PTP1B, PDB 1PH0,
is shown in cartoon representation in Figure 5B, where a surface
of the active site is depicted in Figure 5D. Highlighted in blue
in both representations is the bound active site inhibitor of 1PH0.
Mapping of 1PH0, illustrated in yellow in Figure 5D, reveals
consensus sites in locations proximal to those determined for
1PTY; however, while the number of probe clusters comprising
these consensus sites is equivalent to those derived for 1PTY,
the energy ranking decreases significantly, with average cluster
energy rank in the catalytic pTyr site falling from third to sixth
(see Table 2). Thus, our analysis of the open conformation
reveals only relatively weak hot spots, located at opposite ends
of the binding region as shown in Figure 5D. The change in
hot spot affinity upon conformational shift supports the finding
that larger inhibitors are necessary to bridge both affinity
regions. However, as a result of their size, these inhibitors
typically exhibit poor ADMET properties.37Our study of PTP1B
demonstrates that, unlike the analysis of renin, CS-Map is
sensitive to large changes in conformation of the binding region;
however, these changes may be reflective of a change in
compound properties necessary for inhibition.
Identification of Hot Spots within Druggable Binding
Pockets. For each druggable binding pocket defined in Table
2, we identified groups of highly interacting residues for hot
spot formation. Hot spots resulting from our mapping analyses
were then compared to published data to validate our findings;
we considered our results to be in agreement with experimental
data if residues that we predicted to be in hot spots were
implicated as being important in ligand binding events by
crystallographic, spectroscopic, or other biophysical methods.
Table 3 lists the residues that we predicted to be highly
interactive in ligand binding events using CS-Map; an asterisk
(*) indicates those residues for which we found literature evident
for our findings. For the majority of residues, we were able to
validate our predictions. Of those residues for which we could
not find reports of involvement in ligand binding, a few were
discovered to mitigate the binding process by hydrogen bonding
with water, such as H299 in MurA.41
A pharmaceutical target of particular interest in Table 2 is
the FK-506 binding protein (FKBP). Inhibition of FKBP leads
to the inhibition of calcineurin, serving as a mechanism for the
activation of immunosuppression in organ transplant patients.
An NMR-based screening study performed on FKBP, published
in Science a decade ago, identified several residues for which
chemical shifts were detected upon inhibitor binding;38a figure
from that publication is reproduced here as Figure 6A, where
the residues colored in cyan, yellow, and purple compose the
three hot spot regions uncovered in their analysis.
Figure 6B is our mapping analysis of FKBP performed using
structure 1FKJ from the PDB, with the first- and second-ranked
consensus sites colored blue and red, respectively. Superimposed
in green on the mapping results in Figure 6B is the bound
conformation of FK-506 for comparison. To illustrate the
similarities existing between the hot spots defined by the NMR
Figure 6. Mapping results for the FK-506 binding protein (FKBP). (A) Previously published NMR data where the presence of three hot spots in
the FK-506 binding pocket was determined. Reprinted with permission from Science (http://www.sciencemag.org) (Volume 274; S. B. Shuker et
al.; Discovering High-Affinity Ligands for Proteins: SAR by NMR; pages 1531-1534), ref 38. Copyright 1996 AAAS. The residues comprising
these hot spots are highlighted in cyan, yellow, and purple. The locations of the consensus sites correspond to the hot spots determined by the NMR
experiment. (B) Comparison of top two ranked consensus sites, shown in blue and red, respectively, resulting from the CS-Map algorithm. The
bound structure of FK-506 is shown in green.
Table 3. Hot Spots Predicted by CS-Map for the Training Set of
Druggable Proteins Listed in Table 2a
PDB ID predicted hot spot residues
FKBPFK-5061FKJ Y26*, F36*, F46*, Q53*,
V55*, I56*, R57*, W59*38
K14*, R15*, G16*, E17*,
Y18*, I19*, R23*, R25*39,40
Q10*, N11*, F12*, I37,
L54*, L57*, G58*, I61*,
Y67*, Q72*, H73*43
R120*, V122*, V161, S162*,
V163*, E188*, H299, T304,
R24*, A27, Y46*, C215*,
S216*, A217*, I219*,
N162, V163*, L164*,
A165*, L197*, V198*,
D191*, S192*, C193*, W217*,
G218*, G220*, C221*50
L132*, F153*, Y166*, K232,
I235*, Y257*, I259*, V286*,
Akt-PH IP3 1H10
aAn * indicates a residue for which existing literature supports the
Hot Spots within Druggable Binding RegionsJournal of Medicinal Chemistry G
analysis and the CS-Map-based approach, the residue color
scheme is kept the same for between Figure 6A and B; the first-
ranked consensus site lies within the large region defined by
the purple and yellow residues from the NMR study, and the
second-ranked consensus site occupies the hot spot outlined by
the cyan residues. A more detailed residue distribution is shown
in Figure 7, where the R-value between the CS-Map probes,
shown in black, and four inhibitors of FKBP, including FK-
506 and shown in gray, is 0.51. This correlation value was
determined between the probes and an average residue interac-
tion percentage from the four inhibitors; these inhibitors were
extracted from four bound structures of FKBP from the PDB,
namely, 1FKD, 1FKJ, 1QPF, and 1QPL. The highest percent-
ages of interactions for the CS-Map probes occur at residues
Tyr26, Val55, Ile56, and Trp59; the latter two are validated by
the NMR study, while the former two are highly interactive
with known inhibitors.
Interestingly, the distribution of residue interactions for the
known inhibitors is not in perfect agreement with the NMR
study; three of the residues identified in the NMR screen, Gln53,
Arg57, and Ile90, have very little to no interaction with the
four bound ligands. Both Gln53 and Arg57 make significant
contacts with the mapping probes; these differences suggest that
the identification of hot spots through high-throughput experi-
mental or computational methods provides insight into binding
affinities that cannot be surmised from crystallographic contact
Comparison of CS-Map to Experimental Fragment-Based
Approaches for Hot Spot Identification: A Case Study with
E. coli KPR. Through a series of NMR-based screening studies
of KPR with NADPH-derived analogues in conjunction with
calorimetric analysis of two single point mutants, Ciulli et al.
uncovered two hot spots in the NADPH binding region of KPR,
located in the regions that bind the 2′-phosphate and reduced
nicotinamide groups of NADPH.4Two point mutations of
residues within these regions, Arg31 and Asn98, were used to
confirm their findings. The fragmentation of NADPH performed
by the authors to create analogues is reproduced here as Figure
8A; the 2′-phosphate and reduced nicotinamide groups, high-
lighted in red and blue, respectively, have significantly higher
ligand efficiency values than the ?-phosphate ribose fragment,
shown in white. Ligand efficiency was defined as the change
in the dissociation constant, KD, relative to the change in the
number of heavy atoms upon addition of the fragment to the
starting fragment AMP, shown in green. Our goal was to
confirm the presence of the two hot spots that were described
in the study, as well as to predict the differences in ligand
efficiencies existing among different regions of the binding
The three structures of KPR made available by the PDB were
used for a mapping-based prediction of hot spots in the NADPH
binding region; these structures represent both the unbound
(1KS9) and cofactor/cofactor-analogue (1YJQ and 1YON,
respectively) bound conformations of KPR. Consensus sites
located in the NADPH binding region for structure 1YJQ are
shown in Figure 8B; the sites, colored in cyan, are superimposed
with the structure of NADPH, in green, to illustrate the
concentration of probe clusters at either end of the binding
Figure 7. Distribution of the atom-based residue interactions in the
FK-506 binding pocket for CS-Map probes and four FKBP inhibitors.
The interactions of four different inhibitors of FKBP, including FK-
506, were also calculated on the same residues for comparison.
Figure 8. Mapping analysis of the NADPH binding region of KPR from E. coli. (A) Previous experimental work revealed a range of ligand
efficiency values corresponding to distinct fragments of NADPH. The 2′-phosphate region, circled in red, and the reduced nicotinamide group,
highlighted in blue, exhibit significantly higher ligand efficiencies than the B-phosphate ribose group. (B) Mapping analysis of the NADPH-bound
structure of KPR from the PDB is predictive of two hot spots regions, indicate by the consensus sites in blue, located on opposite sides of the
binding region. (C) The atom-based residue distribution of CS-Map probes in the binding pocket, colored in black, corresponds to the ligand
efficient regions of the binding region shown in (A). Atom interactions are enriched among those residues that bind the red and blue regions of
NADPH defined in (A), whereas the region that binds the white fragment in (A) has a decreased average value of interactions per residue.
Journal of Medicinal Chemistry Landon et al.
region. Residues Arg31 and Asn98, used for mutational
confirmation of the hot spots determined in the experimental
study, are highlighted in magenta in Figure 8B to highlight the
presence of the probe clusters in these regions. Probe interactions
in the NADPH binding region were calculated and summed
across all three structures to create the residue distribution shown
in Figure 8C, where the distribution for the CS-Map probes is
outlined in black and the distribution for NADPH is outlined
in gray; an R-value of 0.6 exists between the two distributions,
suggesting that several key residues are predicted using CS-
To test this hypothesis, we assigned each residue in the
binding region to one of four groups based on the fragment of
NADPH shown in Figure 8A with which it primarily interacts;
this was determined using HBPLUS, where a residue was
assigned to a group if the majority of its interactions with
NADPH occurred primarily with the fragment representing that
group. The percentage of interactions for each residue compris-
ing the group was then normalized by the number of group
members to create a value analogous to ligand efficiency, where
higher values indicate a higher concentration of ligand interac-
tions in that region. As shown in Figure 8C, residues comprising
the groups that interact primarily with the 2′-phosphate and the
reduced nicotinamide moieties of NADPH, labeled as the red
and blue groups, respectively, in Figure 8C, have a drastically
increased average number of interactions per residue, 4.15 and
4.63%, than those residues comprising the group that interacts
primarily with the ?-phosphate ribose moiety of NADPH, the
white group. Residues interacting with the AMP analogue, the
green group, have an average interaction value comparable to
that of the red and blue groups. These residue-based efficiency
values derived from interactions between CS-Map probes and
KPR are consistent with the ligand efficiency values for the
NADPH fragments shown in Figure 8A; this analysis suggests
that in addition to providing spatial information concerning the
location of hot spots, CS-Map can also be used to predict ligand
efficiencies of fragments of larger molecules.
We have described the application of a computational method
for binding site prediction, the CS-Map algorithm, to the
identification of hot spots within druggable binding sites. The
success of this approach was demonstrated for a variety of
proteins on which various experimental analyses were performed
to arrive at similar results. As illustrated by the analysis of renin,
our method is capable of distinguishing regions that bind drug-
like molecules with high-affinity from those that bind nondrug-
like molecules, such as peptidomimetics. Based on our study
of druggable proteins from the NMR screen performed by Fesik
and co-workers, we determined that in addition to the identifica-
tion of residues that play crucial roles in ligand affinity, CS-
Map can also be used to assess the druggability of a binding
pocket based on the rank of consensus sites located within the
pocket. Interestingly, as noted in the Results and Discussion,
our analyses were consistent for structures derived either via
crystallographic or NMR methods. However, as was made
apparent by the analysis of open and closed conformations of
PTP1B, mapping results are not as robust to large changes in
the conformation of the binding site as they are to minor ones,
as was the case with renin. Given these results, the question of
whether CS-Map can be used to analyze homology models will
require further investigation. Last, the prediction of hot spots
within the NADPH binding pocket of KPR illustrated the
usefulness of CS-Map toward the prediction of ligand-efficient
binding regions as well as validating hot spots derived from
biophysical methods. Given these successes, we conclude that
this approach can be applied to the prediction of hot spots for
protein targets where only general druggability features are
currently known, as well as validate new protein targets where
druggability is not known. This, in turn, could provide a wealth
of information for fragment-based drug design efforts.
Acknowledgment. We gratefully thank Frank Guarnieri and
colleagues at SolMap Pharmaceuticals for numerous insightful
conversations regarding this work. This work was made possible
through Grants R01GM064700 and R41GM075473 from the
National Institute of General Medical Sciences.
(1) DeLano, W. L. Unraveling hot spots in binding interfaces: Progress
and challenges. Curr. Opin. Struct. Biol. 2002, 12, 14-20.
(2) Kortemme, T.; Baker, D. A simple physical model for binding energy
hot spots in protein-protein complexes. Proc. Natl. Acad. Sci. U.S.A.
2002, 99, 14116-14121.
(3) Hajduk, P. J.; Huth, J. R.; Fesik, S. W. Druggability indices for protein
targets derived from NMR-based screening data. J. Med. Chem. 2005,
(4) Ciulli, A.; Williams, G.; Smith, A. G.; Blundell, T. L.; Abell, C.
Probing hot spots at protein-ligand binding sites: A fragment-based
approach using biophysical methods. J. Med. Chem. 2006, 49, 4992-
(5) Dennis, S.; Kortvelyesi, T.; Vajda, S. Computational mapping
identifies the binding sites of organic solvents on proteins. Proc. Natl.
Acad. Sci. U.S.A. 2002, 99, 4290-4295.
(6) Kortvelyesi, T.; Dennis, S.; Silberstein, M.; Brown, L., III; Vajda,
S. Algorithms for computational solvent mapping of proteins. Proteins
2003, 51, 340-351.
(7) Silberstein, M.; Dennis, S.; Brown, L.; Kortvelyesi, T.; Clodfelter,
K.; Vajda, S. Identification of substrate binding sites in enzymes by
computational solvent mapping. J. Mol. Biol. 2003, 332, 1095-1113.
(8) Mattos, C.; Bellamacina, C. R.; Peisach, E.; Pereira, A.; Vitkup, D.;
Petsko, G. A.; Ringe, D. Multiple solvent crystal structures: Probing
binding sites, plasticity, and hydration. J. Mol. Biol. 2006, 357, 1471-
(9) Mattos, C.; Ringe, D. Locating and characterizing binding sites on
proteins. Nat. Biotechnol. 1996, 14, 595-599.
(10) Mattos, C.; Ringe, D. Proteins in organic solvents. Curr. Opin. Struct.
Biol. 2001, 11, 761-764.
(11) Bohm, H. J. The computer program LUDI: A new method for the
de novo design of enzyme inhibitors. J. Comput.-Aided Mol. Des.
1992, 6, 61-78.
(12) Goodford, P. J. A computational procedure for determining energeti-
cally favorable binding sites on biologically important macromol-
ecules. J. Med. Chem. 1985, 28, 849-857.
(13) Lawrence, M. C.; Davis, P. C. CLIX: A search algorithm for finding
novel ligands capable of binding proteins of known three-dimensional
structure. Proteins 1992, 12, 31-41.
(14) Miranker, A.; Karplus, M. Functionality maps of binding sites: A
multiple copy simultaneous search method. Proteins 1991, 11, 29-
(15) Laurie, A. T.; Jackson, R. M. Q-SiteFinder: An energy-based method
for the prediction of protein-ligand binding sites. Bioinformatics
2005, 21, 1908-1916.
(16) Cavasotto, C. N.; Orry, A. J.; Abagyan, R. A. Structure-based
identification of binding sites, native ligands, and potential inhibitors
for G-protein coupled receptors. Proteins 2003, 51, 423-433.
(17) Raman, V. K.; Lee, Y. A.; Lindpaintner, K. The cardiac renin-
angiotensin-aldosterone system and hypertensive cardiac hypertro-
phy. Am. J. Cardiol. 1995, 76, 18D-23D.
(18) Stanton, A. Potential of renin inhibition in cardiovascular disease.
JRAAS 2003, 4, 6-10.
(19) Staffileno, B. A. Treating hypertension with cardioprotective thera-
pies: The role of ACE inhibitors, ARBs, and beta-blockers. J.
CardioVasc. Nurs. 2005, 20, 354-364.
(20) Simoneau, B.; Lavallee, P.; Anderson, P. C.; Bailey, M.; Bantle, G.;
Berthiaume, S.; Chabot, C.; Fazal, G.; Halmos, T.; Ogilvie, W. W.;
Poupart, M. A.; Thavonekham, B.; Xin, Z.; Thibeault, D.; Bolger,
G.; Panzenbeck, M.; Winquist, R.; Jung, G. L. Discovery of non-
peptidic P2-P3 butanediamide renin inhibitors with high oral efficacy.
Bioorg. Med. Chem. 1999, 7, 489-508.
(21) Tong, L.; Pav, S.; Lamarre, D.; Simoneau, B.; Lavallee, P.; Jung, G.
Crystallographic studies on the binding modes of P2-P3 butanedia-
mide renin inhibitors. J. Biol. Chem. 1995, 270, 29520-29524.
Hot Spots within Druggable Binding RegionsJournal of Medicinal Chemistry I
(22) Wood, J. M.; Maibaum, J.; Rahuel, J.; Grutter, M. G.; Cohen, N. C.;
Rasetti, V.; Ruger, H.; Goschke, R.; Stutz, S.; Fuhrer, W.; Schilling,
W.; Rigollier, P.; Yamaguchi, Y.; Cumin, F.; Baum, H. P.; Schnell,
C. R.; Herold, P.; Mah, R.; Jensen, C.; O’Brien, E.; Stanton, A.;
Bedigian, M. P. Structure-based design of aliskiren, a novel orally
effective renin inhibitor. Biochem. Biophys. Res. Commun. 2003, 308,
(23) Allikmets, K. Aliskiren (Speedel). Curr. Opin. InVest. Drugs 2002,
(24) Rahuel, J.; Rasetti, V.; Maibaum, J.; Rueger, H.; Goschke, R.; Cohen,
N. C.; Stutz, S.; Cumin, F.; Fuhrer, W.; Wood, J. M.; Grutter, M. G.
Structure-based drug design: The discovery of novel nonpeptide
orally active inhibitors of human renin. Chem. Biol. 2000, 7, 493-
(25) Ciulli, A.; Abell, C. Biophysical tools to monitor enzyme-ligand
interactions of enzymes involved in vitamin biosynthesis. Biochem.
Soc. Trans. 2005, 33, 767-771.
(26) Begley, T. P.; Kinsland, C.; Strauss, E. The biosynthesis of coenzyme
A in bacteria. Vitam. Horm. 2001, 61, 157-171.
(27) Dennis, S.; Vajda, S. Semiglobal simplex optimization and its
application to determining the preferred solvation sites of proteins.
J. Comput. Chem. 2002, 23, 319-334.
(28) Gilson, M. K.; Honig, B. Calculation of the total electrostatic energy
of a macromolecular system: solvation energies, binding energies,
and conformational analysis. Proteins 1988, 4, 7-18.
(29) Zhang, C.; Cornette, J. L.; Delisi, C. Consistency in structural
energetics of protein folding and peptide recognition. Protein Sci.
1997, 6, 1057-1064.
(30) Schaefer, M.; Karplus, M. A Comprehensive analytical treatment of
continuum electrostatics. J. Phys. Chem. 1996, 100, 1578-1599.
(31) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.;
Swaminathan, S.; Karplus, M. CharmmsA program for macromo-
lecular energy, minimization, and dynamics calculations. J. Comput.
Chem. 1983, 4, 187-217.
(32) Verdonk, M. L.; Cole, J. C.; Hartshorn, M. J.; Murray, C. W.; Taylor,
R. D. Improved protein-ligand docking using GOLD. Proteins 2003,
(33) Jones, G.; Willett, P.; Glen, R. C.; Leach, A. R.; Taylor, R.
Development and validation of a genetic algorithm for flexible
docking. J. Mol. Biol. 1997, 267, 727-748.
(34) McDonald, I. K.; Thornton, J. M. Satisfying hydrogen bonding
potential in proteins. J. Mol. Biol. 1994, 238, 777-793.
(35) Rahuel, J.; Priestle, J. P.; Grutter, M. G. The crystal structures of
recombinant glycosylated human renin alone and in complex with a
transition state analog inhibitor. J. Struct. Biol. 1991, 107, 227-
(36) Sali, A.; Veerapandian, B.; Cooper, J. B.; Moss, D. S.; Hofmann,
T.; Blundell, T. L. Domain flexibility in aspartic proteinases. Proteins
1992, 12, 158-170.
(37) Liu, G. Protein tyrosine phosphatase 1B inhibition: Opportunities
and challenges. Curr. Med. Chem. 2003, 10, 1407-1421.
(38) Shuker, S. B.; Hajduk, P. J.; Meadows, R. P.; Fesik, S. W.
Discovering high-affinity ligands for proteins: SAR by NMR. Science
1996, 274, 1531-1534.
(39) Hu, Y.; Qiao, L.; Wang, S.; Rong, S. B.; Meuillet, E. J.; Berggren,
M.; Gallegos, A.; Powis, G.; Kozikowski, A. P. 3-(Hydroxymethyl)-
bearing phosphatidylinositol ether lipid analogues and carbonate
surrogates block PI3-K, Akt, and cancer cell growth. J. Med. Chem.
2000, 43, 3045-3051.
(40) Thomas, C. C.; Deak, M.; Alessi, D. R.; van Aalten, D. M. High-
resolution structure of the pleckstrin homology domain of protein
kinase b/akt bound to phosphatidylinositol (3,4,5)-trisphosphate. Curr.
Biol. 2002, 12, 1256-1262.
(41) Hajduk, P. J.; Dinges, J.; Schkeryantz, J. M.; Janowick, D.; Kaminski,
M.; Tufano, M.; Augeri, D. J.; Petros, A.; Nienaber, V.; Zhong, P.;
Hammond, R.; Coen, M.; Beutel, B.; Katz, L.; Fesik, S. W. Novel
inhibitors of Erm methyltransferases from NMR and parallel
synthesis. J. Med. Chem. 1999, 42, 3852-3859.
(42) Schluckebier, G.; Zhong, P.; Stewart, K. D.; Kavanaugh, T. J.; Abad-
Zapatero, C. The 2.2 A structure of the rRNA methyltransferase
ErmC' and its complexes with cofactor and cofactor analogs:
Implications for the reaction mechanism. J. Mol. Biol. 1999, 289,
(43) Kussie, P. H.; Gorina, S.; Marechal, V.; Elenbaas, B.; Moreau, J.;
Levine, A. J.; Pavletich, N. P. Structure of the MDM2 oncoprotein
bound to the p53 tumor suppressor transactivation domain. Science
1996, 274, 948-953.
(44) Skarzynski, T.; Mistry, A.; Wonacott, A.; Hutchinson, S. E.; Kelly,
V. A.; Duncan, K. Structure of UDP-N-acetylglucosamine enolpyru-
vyl transferase, an enzyme essential for the synthesis of bacterial
peptidoglycan, complexed with substrate UDP-N-acetylglucosamine
and the drug fosfomycin. Structure 1996, 4, 1465-1474.
(45) Liu, G.; Xin, Z.; Liang, H.; Abad-Zapatero, C.; Hajduk, P. J.;
Janowick, D. A.; Szczepankiewicz, B. G.; Pei, Z.; Hutchins, C. W.;
Ballaron, S. J.; Stashko, M. A.; Lubben, T. H.; Berg, C. E.;
Rondinone, C. M.; Trevillyan, J. M.; Jirousek, M. R. Selective protein
tyrosine phosphatase 1B inhibitors: Targeting the second phospho-
tyrosine binding site with noncarboxylic acid-containing ligands. J.
Med. Chem. 2003, 46, 3437-3440.
(46) Szczepankiewicz, B. G.; Liu, G.; Hajduk, P. J.; Abad-Zapatero, C.;
Pei, Z.; Xin, Z.; Lubben, T. H.; Trevillyan, J. M.; Stashko, M. A.;
Ballaron, S. J.; Liang, H.; Huang, F.; Hutchins, C. W.; Fesik, S. W.;
Jirousek, M. R. Discovery of a potent, selective protein tyrosine
phosphatase 1B inhibitor using a linked-fragment strategy. J. Am.
Chem. Soc. 2003, 125, 4087-4096.
(47) Dunten, P.; Kammlott, U.; Crowther, R.; Levin, W.; Foley, L. H.;
Wang, P.; Palermo, R. X-ray structure of a novel matrix metallo-
proteinase inhibitor complexed to stromelysin. Protein Sci. 2001, 10,
(48) Dhanaraj, V.; Ye, Q. Z.; Johnson, L. L.; Hupe, D. J.; Ortwine, D.
F.; Dunbar, J. B., Jr.; Rubin, J. R.; Pavlovsky, A.; Humblet, C.;
Blundell, T. L. X-ray structure of a hydroxamate inhibitor complex
of stromelysin catalytic domain and its comparison with members
of the zinc metalloproteinase superfamily. Structure 1996, 4, 375-
(49) Pavlovsky, A. G.; Williams, M. G.; Ye, Q. Z.; Ortwine, D. F.;
Purchase, C. F., II; White, A. D.; Dhanaraj, V.; Roth, B. D.; Johnson,
L. L.; Hupe, D.; Humblet, C.; Blundell, T. L. X-ray structure of
human stromelysin catalytic domain complexed with nonpeptide
inhibitors: Implications for inhibitor selectivity. Protein Sci. 1999,
(50) Hajduk, P. J.; Boyd, S.; Nettesheim, D.; Nienaber, V.; Severin, J.;
Smith, R.; Davidson, D.; Rockway, T.; Fesik, S. W. Identification
of novel inhibitors of urokinase via NMR-based screening. J. Med.
Chem. 2000, 43, 3862-3866.
(51) Crump, M. P.; Ceska, T. A.; Spyracopoulos, L.; Henry, A.; Archibald,
S. C.; Alexander, R.; Taylor, R. J.; Findlow, S. C.; O’Connell, J.;
Robinson, M. K.; Shock, A. Structure of an allosteric inhibitor of
LFA-1 bound to the I-domain studied by crystallography, NMR, and
calorimetry. Biochemistry 2004, 43, 2394-2404.
(52) Liu, G.; Huth, J. R.; Olejniczak, E. T.; Mendoza, R.; DeVries, P.;
Leitza, S.; Reilly, E. B.; Okasinski, G. F.; Fesik, S. W.; von Geldern,
T. W. Novel p-arylthio cinnamides as antagonists of leukocyte
function-associated antigen-1/intracellular adhesion molecule-1 in-
teraction. 2. Mechanism of inhibition and structure-based improve-
ment of pharmaceutical properties. J. Med. Chem. 2001, 44,
Journal of Medicinal ChemistryPAGE EST: 9.8Landon et al.