You are on page 1of 48

The Art and Practice of Structure-Based Drug

Design: A Molecular Modeling Perspective


Regine S. Bohacek, Colin McMartin, and Wayne C. Guida*
Research Department, Pharmaceuticals Division, Ciba-Geigy Corporation,
556 Morris Avenue, Summit, New jersey 07901

I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
...................................................................... 6
rase Inhibitors ................................................. 6
gs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
C. HIV-1 Protease Inhibitors .................................................... 12
1. Design of Nonpeptide Cyclic Ureas that are Potent, Bioavailable Inhibitors
of HIV Protease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2. Optimization of a Coumarin Lead from Random Screening I .............. 15
3. Optimization of a Coumarin Lead from Random Screening I1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4. Optimization of a Lead Obtained by Screening Renin Inhibitors . . . . . . . . . . . . . . . . . . . . . . . . 18
5. Optimization of a Penicillin Lead from Random Screening . . . . . . . . . . . . . . . . . . . . . . . 20
6. Optimization of the C-Terminus of a Known Inhibitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
7. Design of a Macrocyclic Inhibitor Based on a Known Protease Inhibitor 25
D. NEP Inhibitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
E. Dual Inhibitors of ACE and NEP .............................................. 29
F. PNP Inhibitors . . . . ................................ 32
G. Sialidase Inhibitors ......................................................... 34
H. Thymidylate Synth ...................... 35
1. Optimization of ors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2. De novo Design of Thymidylate Synthase Inhibitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3. Discovery of Thymidylate Synthase Inhibitors Aided by 3-D Database Searching . . . . . . . . . 38
111. Discussion . . . . . . . . . . . . . . . . . . . . ............................ 39
A. Optimization of Lead Compounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1. Improvement of Complementarity to the Binding 41
2. Improvement of Conformational Properties . . . . . 42
3. Improvement of Bioavailability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
B. Finding New Leads . . . . .................................................. 42
1. High Throughput Scr 42
2. Computer-Based Screening of Structural Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3. Design of New Leads . . . . . . . . . . . . . . . . .................... 44
4. Potency Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
IV. Conclusions . . . . . . . . . . . . . . . . . . . . . . ........................... 46
References .................................................................... 47

I. INTRODUCTION
The conceptual basis for structure-based drug design was formulated 100 years ago by
Emil Fischer.1 His "lock and key" hypothesis is a constantly recurring leitmotif in mod-
ern drug design, and yet, it is only recently that we have been in possession of detailed
descriptions of the "locks", i.e., the biochemical targets to which potent and selective
*To whom all correspondence should be addressed.
We dedicate this review to the memory of Dr. George deStevens who encouraged us to write it and who
was extremely supportive during its preparation.
Medicinal Research Reviews, Vol. 16, No. 1, 3-50 (1996)
0 1996 John Wiley & Sons, Inc. CCC 0198-6325/96/010003-48
4 BOHACEK, MCMARTIN, AND GUIDA

drugs (the "keys") must fit. Success has been achieved in the past by exploring the shape
of the "keys," i.e., by structure/activity analysis of a series of experimental test ligands.
However, this approach can yield only limited information about the "lock," and a large
amount of work is required to achieve useful results by such experimental trial and error.
Three-dimensional structures obtained by X-ray crystallography and NMR spectroscopy
contain a quantity of information that is orders of magnitude greater than that which can be
obtained by building pharmacophores, even when a well studied set of ligands is available.
This increased information introduces the possibility of directly designing drugs based on a
detailed model of the target binding site, a possibility that raises a number of questions:

1. Is high-resolution information about a target binding site really helpful in the


design of new drugs?
2. What changes are likely to occur in the rate of drug discovery and in the quality
of the products reaching the market?
3. What techniques can fully utilize the new types of information?
4. What are the limitations of this approach?

In spite of initial skepticism, the design of novel, potent therapeutic agents based
upon 3-dimensional structural information is now a reality. With increasing frequency,
the structure of the target binding site or that of a closely related analog is available prior
to embarking upon a drug design project. Thus, structure-based drug design (i.e., the
design of drugs based on the 3-dimensional structure of a biomacromolecule) has
emerged as a powerful tool for drug discovery. Recent reviews*-4 have presented numer-
ous examples in which the utility, not simply the potential, of structure-based drug
design has been amply demonstrated. An account of this activity has even appeared in
the popular literature.5
Of course, one must ask, how does one measure the success of a project involving
structure-based drug design? One measure might be a reduction in the number of
compounds synthesized and tested in order to produce a clinical candidate. Often,
however, the mere existence of a candidate drug in human clinical trials is taken as an
end-point for structure-based drug design, which some believe is proof of the validity of

Regine S. Bohacek received her Ph. D. in Physical Chemistry from Rutgers University in 1987. Her thesis
dealt with a configurational analysis of a synthetic polymer using the statistical mechanical methods developed by
Flory. Shehasapplied thesemethods in acomputerprograinforde novostructure-baseddrugdesign. Dr. Bohacek
is currently engaged in computer-aided moleculnr design at Ciba in Summit, N . 1.Her interests include: methods
forestimatingenzyme inhibitor bindingenergies, thedesign of new drugs based on the3-dimensional structureof
binding sites or inhibitor templates, and zinc metallo-proteases and their inhibitors.
Colin McMartin is a senior scientist in the Biophysical Chemistry group at Ciba in Summit, N . 1. where
he is engaged in the development and application of computer methods for drug design. Previously, he led a
group of biologists and biochemists investigating pharmacokinetics, metabolism, and bioavailability of pep-
tides at Ciba in Horsham, Sussex, U . K. He received his B. A. in Chemistry with honors from Cambridge
University in the U . K. His interests include: development of new structure prediction methods, methods for
estimating enzyme inhibitor binding energies, and the design of new drugs based on the three-dimensional
structure of binding sites andlor inhibitor templates.
Wayne C. Guida is Executive Director of Core Drug Discovery Technologies at Ciba in Summit, N . J.
Previously, he had been an Associate Professor of Chemistry at Eckerd College in St. Petersburg, Florida. In
1985-1986, while on sabbatical leave, he was a Senior Research Fellow at Columbia University. He is one of
the co-authors of the MacroModel molecular modeling program, which was developed in Professor Clark Still's
laboratory at Columbia. Prior to joining the faculty at Eckerd College, he received his Ph. D. degree in Organic
Chemistryfrom the University of South Florida and was a post-doctoral fellow at Duke University. His recent
research has focused on structure-based drug design and computational methods for conformational analysis.
STRUCTURE-BASED DRUG DESIGN 5

1 Clone and express protein

Generate Structure

Purified Protein

3-D Database
Libraries Compound
Searching
1 I

Combinatorial
Drug Design
Libraries

in viva in vivo
Potency

Combinatorial
Drug Design
Libraries

1
[Drug)
Figure 1. Drug discovery and design flow chart.

this novel approach. Although we would argue whether this is the appropriate end-
point, it is a fact that the number of new chemical entities in the clinic for which
structure-based drug design has played a key role continues to grow.5 The list includes
medications for T-cell lymphoma, psoriasis, tumor suppression, viral infection (includ-
ing HIV), and glycoma.
It is clear that structure-based drug design plays an important role in drug discovery.
In fact, a new paradigm for drug discovery and design is beginning to emerge in which
contributions from rational drug design (which includes structure-based methods), tra-
ditional screening and synthesis, and the newer combinatorial approaches are coupled.
In Fig. 1, we illustrate the synergy among these various techniques. This review will
demonstrate that structure-based drug design has been successfully used at every step,
from lead finding to final optimization, in the discovery of novel drugs. For example,
searching of 3-dimensional databases for molecules that can dock to a target binding site
has already contributed to lead finding. The optimization of these and other leads has
been accomplished using structure-based methodology. Moreover, structure-based drug
design techniques have contributed to in vivo potency optimization by serving as a guide
for those modifications to a molecule that might enhance bioavailability without destroy
ing in vitro potency. In addition, new methods are now emerging that will make even
6 BOHACEK, MCMARTIN, AND GUIDA

more puissant use of the information inherent in a high-resolution structure of a binding


site. For example, computer-based de novo structure-based drug design techniquesh-**
have already begun, and will continue to contribute to the generation of novel lead
candidates. We are confident that structure-based drug design techniques will, for the
foreseeable future, play a role of increasing importance in drug discovery.
Prior to embarking upon a structure-based drug design project, it is necessary to have
available either the 3-dimensional structure of the target biomacromolecule or the 3-di-
mensional structure of a closely related analog. If the structure of an homologous protein
is available, for example, it can either be used directly, or converted to a model of the
target by using homology modeling techniques.3 Advances in both X-ray crystallogra-
phy and NMR spectroscopy have yielded a number of target structures of therapeutic
relevance over the past few years. Nonetheless, molecular modeling is necessary to use
this information for drug design. In some cases, the contribution of molecular modeling
has been largely confined to visualization of a ligand complexed to the target binding site
using computer graphics. Whereas, in other cases, computational chemistry techniques
have made a significant contribution to the design activities. Examples exist where
computer-assisted molecular modeling has played an essential role in the design of poten-
tial ligands that are both sterically and chemically compatible with the binding site of a
target biomacromolecule. It is the combination of molecular modeling, structural biolo-
gy, and chemical synthesis that has allowed structure-based drug design to be reduced
to practice.
In this review, we describe the art and practice of structure-based drug design by
focusing on the design techniques that were actually used for its successful application
to drug discovery. We analyze sixteen selected recent examples of the design of com-
pounds based on the coordinates of a target binding site (or those of a related analog)
that have resulted from either protein X-ray crystallographic analysis or NMR spec-
troscopic analysis. We identify the procedures that were used in these examples, and
evaluate their role in the design process. We conclude with a discussion of emerging
techniques, such as prediction of potency and de novo drug design, which, we believe,
will allow the enormous potential of structure-based drug design to be fully realized.
Finally, we point out that structure-based drug design is a multifaceted endeavor that
requires a collaborative venture among scientists from a variety of disciplines including
computational chemistry, synthetic organic chemistry, medicinal chemistry, molecular
biology, X-ray crystallography, and NMR spectroscopy; each activity contributes to the
success of a structure-based drug design project. Nonetheless, in this review we high-
light the actual and potential role of computational chemistry and molecular modeling.

11. EXAMPLES
A. Carbonic Anhydrase Inhibitors (Merck)
Recently reported work on inhibitors of carbonic anhydrase illustrates the iterative use
of experimental and computational techniques.3 The results indicate the importance of
conformational strain energy for determining free energies of binding.
The conversion of carbon dioxide to bicarbonate by carbonic anhydrase is a critical
step in the secretion of aqueous humor. Thus, inhibitors of this enzyme can lower ocular
pressure and may be of benefit in glaucoma. The starting point for the further develop-
ment of compounds was MK-927. This compound has high water solubility and an
appropriate partition coefficient to enable rapid penetration into the ocular tissues after
topical application.
X-ray structures for complexes of carbonic anhydrase I1 with the two enantiomers of
STRUCTURE-BASED DRUG DESIGN 7

MK-927

MK-927 have been determined, and this was the starting point for compound optimiza-
tion. The S-enantiomer (MK-417) has a Ki of 0.61 nM and is about 100 times more potent
than the R-enantiomer (Ki = 71 nM). Since the two isomers will have the same solvation
energies, it is clear that this difference must result from differences in conformational
energy of the structure docked in the enzyme binding site. The docking geometries of
the two isomers are shown in Fig. 2. This figure shows how the shape of the binding site
strongly determines the docking geometries. The sulfonamide is rigidly locked in a very
small pocket at the back of the site. The ring system and side chain lie across a larger
cavity, but the positions of the sulfone sulfur and oxygen atoms and the iso-butyl methyl
groups are strongly determined by way they fit with precision into cusp-like niches in
the cavity as indicated by the accessible surface (see Fig. 2).
Before designing compounds, the experimental data was subjected to computational
analysis and an hypothesis was formed that would explain observed differences in
potency. While the contacts of the two inhibitors with the enzyme are rather similar, the
internal geometries have small but energetically significant differences. Energy differ-
ences due to internal geometry were computed using ab initio molecular orbital quantum

R-enantiomer
S-enantiomer(MK-417)
Ki = 71 nM
Ki = 0.61 nM

Conformational Energy Difference of R- vs. S-enantiorner = 2 KcaVmol


8 BOHACEK, MCMARTIN, AND GUIDA

Figure 2. Stereo images illustrating the docking of two isomers of MK-927; (A) S-enantiomer in green; (8)
R-enantiomer (pink) and S-enantiomer (green). The figure was prepared for illustration purposes by docking
the compounds into a structure of a carbonic anhydrase complex obtained from the Brookhaven Protein Data
Bank.12.13 The mesh in this and other figures shows an accessible surface and indicates where ligand atoms in
good van der Waal's contact with the binding site should lie.I4 Mesh colors show the ideal properties of ligand
atoms: yellow: hydrophobic; red: hydrogen-bond accepting; blue: hydrogen-bond donating. For accurate pic-
tures of the X-ray results for the MK-927 compound, the reader is referred to the original study.3
STRUCTURE-BASED DRUG DESIGN 9

TABLE I

Computed
Ki Dihedral Angle Energy Difference
Compound nM T (degrees) (Kcal/mol)

0.37 140 0.0

5.5 175 1.5

2.0 153 0.5

15.3 168 1.5

mechanics calculations (3-21 G*, Gaussian 88). The dihedral angles of the bond connec-
ting the sulfonamide group to the thiophene ring ( N - M - S ) were 150" for the
S-enantiomer and 170" for the R-enantiomer, the difference resulting in an energy for the
S isomer that was more favorable by about 1 kcal/mol. A similar energy difference was
found comparing the anti geometry for the N-C bond of the iso-butyl side chain of the
S isomer to the gauche geometry of the R enantiomer. The combined energy difference of
approximately 2 Kcal/mol accounts for an appreciable part of the observed potency ratio
of 1OO:l.
This conclusion is important in a general sense, since there has been some controversy
over whether strain energies should be considered as significant determinants of poten-
cy. This study is unusual, since other factors that might influence potency are relatively
10 BOHACEK, MCMARTIN, AND GUIDA

similar for the two molecules. Examples where it is possible to cleanly isolate one of the
multiple factors that determine potency and study it with reliable experimental and
computational methods are fairly rare but essential for the further development of de-
pendable structure-based design methodology (see Discussion).
Both isomers are docked in a pseudo-axial conformation. Ab initio quantum mechanics
calculations show that this conformation has an energy of about 1 Kcal/mol higher than
the pseudo-equatorial conformation. Ab initio quantum mechanics calculations also
showed that if a methyl group is introduced in the 6 position, the conformational
preference is removed. This modification was, therefore, made and, at the same time, to
conserve the physicochemical properties of the molecule, two methyl substituents were
removed from the iso-butyl side-chain.
The methyl group in the 6 position (Table I) introduces a second chiral center resulting
in four possible stereoisomers, all of which were synthesized and assayed. The com-
pounds differed in potency, the most potent offering a modest improvement over the
starting structure. X-ray analysis showed that each compound had a slightly different
docked conformation, the major differences being in the N-M--S dihedral angle.
The data is summarized in Table I.
Once again, evaluation of conformational strain energy by ab initio quantum mechan-
ics calculations provided an explanation for the difference in potencies. The two trans
isomers had a calculated energy difference of 1.5 kcal/mol due to changes in the NSCS
dihedral, which could account for most of the 20-fold difference in potency. The cis
isomers had a smaller difference in potency and the calculated energy difference was
1.0 kcal/mol.
Further series of compounds were synthesized to explore the effects of varying chain
length at the 6 position and the nitrogen. The results have been reviewed previously.3
Although the gain in potency reported in the carbonic anhydrase work is modest, the
results show the value of repeated experimental structure determinations at each cycle of
drug optimization for confirming hypotheses and further extending our knowledge of
the factors that determine potency. The use of optical antipodes is very elegant, because
it eliminates factors between the molecules such as solvation energy and, thus, focuses
attention on differences in docking mode and internal conformational energy.

B. Cyclosporin Analogs (Harvard University)


Cyclosporin A, 1, is an immunosuppressant, which, by binding to two proteins simul-
taneously, inhibits the proliferation of T lymphocytes. Thus Cyclosporin A has found
widespread use in the clinic for organ transplantation. The goal of a study undertaken
by Alberg and Schreiber15 was to design a cyclosporin A (CsA) analog with increased
binding affinity to both cyclophilin and calcineurin by conformationally restricting por-
tions of CsA.
The structure of CsA bound to cyclophilin has been determined by NMR.16.17 The
structure of CsA in the unbound state has also been determined in chloroform solution by
NMR18.19 and in the solid state by X-ray crystallography.18Interestingly, it was found that
the bound conformation differs markedly from both of these unbound conformations.20J
Similar findings have been reported for another immunosuppressant, FK506. The
structure of FK506 complexed to the FK506 binding protein (FKBP) has been determined
by X-ray crystallography to 1.7 A.22 This bound conformation of FK506 is very different
from the unbound conformation both in the solid state23 and in chloroform solution.24
However, more recently, the conformation of an FK506 analog in aqueous solution has
STRUCTURE-BASED DRUG DESIGN 11

1 CyclosporinA

been shown to be similar to the bound conformation of FK506.25 Apparently, the bound
conformation of FK506 and its close analogs must preexist to some extent in aqueous
solution. Based on the X-ray crystal structure of a CsA/Fab complex26 and kinetic evi-
dence,27 it has been suggested that a similar situation occurs for cyclosporin.
The conformationally restricted cyclosporin analog described here was carefully de-
signed so that any modifications would not interfere with either the binding to cyclo-
philin or calcineurin. The residues in contact with cyclophilin were known from the
X-ray structure of CsA bound to cyclophilin. Those residues that do not affect binding to
calcineurin had previously been identified by the synthesis of cyclosporin analogs. The
modifications of cyclosporin that do not affect immunosupressive activity28,29 are in-
ferred to be in positions not interfering with the binding to calcineurin. Therefore, the
segments of CsA chosen for alteration were those not in contact with cyclophilin and
those that can be modified without loss of immunosupressive activity and, thus, pre-
sumed not to be in contact with calcineurin. Two residues of CsA, L-Ala7-D-Alas, met
these criteria and were selected for modification. The conformation of the backbone of
these residues is significantly different in the bound and solution conformations. The
program CAVEAT30 from Cambridge Structural Database31 was used to identify a rigid
core structure that would fix the positions of the amine substituent and a carboxyl
moiety of L-Ala7-D-Ala8 in the same relative orientations as found in the cyclophilin
bound structure. CAVEAT suggested a bicyclic heterocycle (2), and this heterocycle was
incorporated into cyclosporin A resulting in TCsA (3).1
The binding affinity of the conformationally restrained analog (3), TCsA, to cyclophilin
(Ki = 2 nM) was three times better than that of cyclosporin (1,K i = 6 nM). Since this
modified portion of the molecule is not expected to bind to cyclophilin, it is hypothe-
sized that the increase in binding energy is entirely due to entropic effects. The affinity
12 BOHACEK, MCMARTIN, AND GUIDA

Y
H

3 TCsA

of the complex of the conformationally constrained cyclosporin analog and cyclophilin


with calcineurin is 2-3 times that of the cyclosporin-cyclophilin complex for calcineurin,
i.e., the K, is 78 nM for the constrained analog vs. 198 nM for cyclosporin.
This work shows that the structure of the ligand bound to the receptor can provide
valuable information for the rational design of more potent ligands. The authors re-
marked that the relatively small increases in potency indicate the complexity of the
problems of modeling and predicting the effects that conformational constraints will
have on binding affinity. However, we speculate that a possible reason for the small
increase in potency could also be attributed to the hypothesis that the bound conforma-
tion of CsA is similar to the conformation in aqueous solution (as is the case for FK506)
and that rigidifying a conformation that is already stabilized in solution may not gain
much in binding free energy.

C. HIV-1 Protease Inhibitors


HIV-1 protease is an aspartic protease that is encoded in the HIV viral genome. The
substrate specificity of HIV-1 protease is not clear; the enzyme makes a number of highly
specific cleavages with the gag and gag-pol polyprotein at sites spanning remarkably
heterogeneous amino-acid sequences and no consensus sequence for HIV protease has
yet been deduced.
The protease is essential for proper virion assembly and maturation, and inactivation
of HIV-1 protease by inhibition or mutation has been shown to lead to the production of
immature, noninfectious viral particles. Therefore, HIV-1 protease has become a major
target in the quest for an effective agent to combat the HIV virus.
Undoubtedly, the structure most used for the design of enzyme inhibitors is that of
HIV protease. The number of enzymelinhibitor complexes for which the 3-dimensional
structure has been determined continues to climb and is, at this date, estimated to
STRUCTURE-BASED DRUG DESIGN 13

exceed 200. Molecular modeling has been used extensively in an attempt to discover
drugs effective in halting this dreaded viral infection.
In this review, we will discuss only a few of the most recent examples (up to November
1994) illustrating some of the different techniques that have been successfully imple-
mented in this area.

1. Design of Nonpeptide Cyclic Ureas that are Potent, Bioavailable lnhibitors


of HIV Protease (DuPont-Merck)
A variety of compounds that display excellent inhibition of HIV protease has been
designed. However, most of these retain substantial peptidic character and are not
sufficiently bioavailable. The goal of an investigation at DuPont-Merck was the design of
smaller inhibitors that would retain potency with improved pharmacokinetic profiles.32
The strategy employed was that 1) by including a moiety in the inhibitor that would
form the same interactions as the water molecule found in the high-resolution HIV
protease crystal structures, and 2) by designing more conformationally constrained,

TABLE I1
HIV Protease Inhibitors - DuPont-Merck

6 4.7

7 0.31

0.27
14 BOHACEK, MCMARTIN, AND GUIDA

cyclic inhibitors, it would be possible to design smaller inhibitors that would retain
potency but display improved oral biooavailability. From the extensive HIV inhibitor
SAR, it was known that C2-symmetric diols imparted significant potency. Therefore,
scientists at DuPont-Merck also wanted to include this feature in their design. Because
there was no crystallographic information about how diols bind to the protease when the
project was initiated, they used the X-ray structure of an hydroxyethylamine inhibitor
bound to HIV-1 protease33 and used the distance geometry program DGEOMN to devel-
op several pharmacophores. From this model, they constructed a simple 3-dimensional
database query specifying the positions of hydrophobic groups that bind to the S1 and
S1' subsites of the enzyme and either one or two hydrogen bond donors or acceptors to
bind to the catalytic aspartates. This query was used to search a subset of the Cambridge
Structural Database,31 which had been converted for use with MACCS-3D,35 yielding
structure 4 (Table 11). MACCS3D allows the user to search a 3-dimensional database to
locate structures that meet user-defined geometrical constraints. Not only did structure 4
show that a tetra-substituted benzene ring could properly position substituent groups to
interact with aspartates 25 and 25', it also possessed an additional oxygen that matched
the position of the key structural water molecule. However, because a benzene ring
might not be the ideal scaffold to properly position its substituents to reach the enzyme's
hydrophobic pockets, molecular modeling was used to optimize the type of ring in a
stepwise fashion from a benzene ring, to a substituted cyclohexanone, to the larger
cycloheptanone, and finally to a 7-membered cyclic urea, 5. Additional modeling was
performed to determine the ideal stereochemistry and conformation for substituents
attached to the cyclic urea with and without substituents on the nitrogens. The confor-
mational preferences were later confirmed from small molecule X-ray crystallography.
The actual potencies of cyclic ureas with various stereochemistry correlated well with
those predicted by the model. In the final structure, the N-substituted cyclic urea has
substituents with stereochemistry predicted to be optimal by the model: 4R, 5S, 6S, and
7R. This structure appears to be an ideal, rigid scaffold to position groups into the S1,
Sl', S2, and S2' pockets as well as to place the hydroxyl groups to interact with the
catalytic aspartates. In addition, the model showed the urea carbonyl oxygen occupying
the position of the key structural water molecule. As predicted by the model, compound
6 binds to the enzyme with high affinity ( K ; = 4.7 nM).
However, compound 6 exhibited only modest antiviral activity. Molecular modeling
indicated that there was room for a larger group in the S2/S2' subsite and, therefore,
naphthalene substituents were introduced. This compound, 7, showed a 10-fold im-
provement in potency ( K ; = 0.31 nM) probably due to the increased hydrophobic con-
tacts. Unfortunately, this enhanced enzyme inhibitory activity did not translate into
improved antiviral activity, presumably because of the lipophilicity of the naphthalene
groups. Additional modeling suggested the incorporation of a hydrogen bonding group.
The naphthalene rings were replaced by p-hydroxymethylphenyl substituents resulting
in compound 8. Compound 8 combines excellent potency against HIV protease ( K , =
0.27 nM), and against viral replication, with good oral bioavailability. This compound is a
specific HIV protease inhibitor showing no inhibition of renin, pepsin, or cathepsin D.
This specificity is hypothesized to stem in part from the observation that these enzymes
do not have a structural water analogous to that found in HIV protease and mimicked by
cyclic urea 8. The X-ray crystal structures of ten cyclic ureas complexed to HIV protease
have been determined. The crystal structure of compound 7 complexed to HIV protease
was refined to the highest resolution (1.8 A, final R factor = 19.5%), and is depicted in
STRUCTURE-BASED DRUG DESIGN 15

TABLE I11
HIV Protease Inhibitors - Upjohn
~~

Compound Potency

9 (warfarin) IC,, = -30 pM

10 K, = 1 pM

11 Ki = 0.5 p M

OH \

12
K, = 0.038 p M

OH \

the article.32 The X-ray structure shows that 7 forms interactions with the enzyme pre-
dicted by molecular modeling.

2. Optimization of a Coumarin Lead from Random Screening 1 (Upjohn)


The group at Upjohn36 began this study with a lead from targeted screening. A set of
5000 dissimilar compounds from the Upjohn compound library were screened for HIV
protease activity. Compound 9 (Table III), warfarin, was identified as a weak inhibitor
-
(ICs0 30 pM). Based on this finding, a search for compounds similar to 9 was con-
ducted, resulting in compound 10, which has a Kiof 1 FM. Both compounds 9 and 10
have already been used as therapeutic agents in humans and exhibit high oral bio-
availability and low clearance, making these compounds especially promising leads.
To begin the optimization and design process, the X-ray crystal structure of compound 10
complexed with HIV protease was determined to 2.5 8,resolution. The inhibitor was found
to exist in two orientations related by a 180"rotation. The C-4 hydroxyl group was shown to
hydrogen bond with the two catalyticaspartic acid residues, whereas the two oxygen atoms
of the lactone formed hydrogen bonds with the backbone HN of Ile 50 and Ile 50'. The
a-ethyl group and the a-phenyl ring were found near the S1 and S2 subsites, respectively.
16 BOHACEK, MCMARTIN, AND GUIDA

Analysis of this enzyme inhibitor complex showed that the position of the benzene
ring of the coumarin moiety does not readily allow for the addition of substituents that
will extend into the S2' subsite. Therefore, the fused benzene ring was expunged and
side chains were added to the remaining 4-hydroxy-2-pyrone ring yielding compound
11. The X-ray crystal structure of 11 complexed to HIV protease was determined to 2.3 A
resolution and showed that, despite replacement of the coumarin benzene ring, the
hydrogen bonding pattern of 11 remained the same as that of compound 10. Also, as
anticipated, the phenethyl group at C-6 was located near the S2' subsite. Next, an ethyl
group was added at the position alpha to C-6 (compound 12). The ethyl group was
designed to occupy the S1' pocket and to move the phenyl ring even closer into the S2'
subsite. This compound can be viewed as having pseudosymmetric substitutions with
the four substituents occupying the S2, S1, Sl', and S2' pockets. These modifications led
to substantial improvements in the binding affinity. Compound 12 was found to inhibit
HIV-1 protease with a K , of 38 nM. This compound is a mixture of four stereoisomers, all
of which were isolated. All were found to inhibit HIV, the best with a K j of 14 nM.
Compound 12 also inhibits HIV-2 ( K j = 32 nM), has excellent properties such as selec-
tivity for HIV protease, antiviral activity in HIV-lIIIBinfected MT4 and H9 cells, good oral
bioavailability, and is synthesized in three chemical steps from readily available starting
materials. Compound 12, U-96988, has entered phase I clinical trials.

3 . Optimization of a Coumarin Lead from Random Screening 11


(Parke-Davis/NCI-Frederick Cancer Research and Development Center)
Another example of the use of high-throughput screening, molecular modeling, X-ray
crystallography, and synthesis to discover potent HIV protease inhibitors was carried
out jointly at Parke-Davis and at the NCI-Frederick Cancer Research and Development
Center.37
A nonpeptide, low micromolar ( K j = 1.0 FM) coumarin derivative, 13 (Table IV), was

TABLE IV
HIV Protease Inhibitors - Parke-Davis/NCI

Compound IC, (PM)

13 &.XI 2.3

14 1.7

15 0.52
STRUCTURE-BASED DRUG DESIGN 17

TABLE V
HIV Protease Inhibitors - Parke-Davis/NCI

Compound Ki (CLM)
on

16 1.1

17 0.7

?"

0.051

discovered using high-throughput screening techniques. Compound 13 was modeled in


the binding site of HIV using the X-ray structure of HIV-1 protease complexed with the
inhibitor, MVT-101. Aut~dock,~* a computer program designed to dock molecules into
binding sites, was used to position compound 13 into the active site. The position of this
ligand was further optimized using energy minimization. In this study, the movable
flaps of HIV were "presented in a closed position" and simulations were carried out with
and without the conserved water molecules, H20301and H,OCAT. The crystal structure
of 13 complexed with HIV protease was subsequently determined to 3.0 A resolution,
revealing two binding modes, one of which was similar to the compound modeled
without the water molecules.
Three parts of the molecule were modified to optimize the structure: the side chain
(especially around the phenoxy oxygen), the phenyl group, and the coumarin ring.
Since, according to the crystal structure of the complex of 13 with HIV protease, the
phenoxy oxygen did not appear to interact with any polar enzyme atoms, a series of
analogs were synthesized involving modifications of this atom and the length of the
chain. However, it was found that an oxygen occupying a position near the phenoxy
oxygen of 13, i.e., 14, was necessary for potency. It was concluded that the oxygen
probably forms a water mediated hydrogen bond with the enzyme. The GRID program39
was used to suggest possible positions of this water molecule, and a water positioned to
hydrogen bond both with the NH of Asp29 and the inhibitor oxygen was postulated.
Modifications of the chain length did not lead to more active compounds. The X-ray
crystal structure showed that the phenyl group interacts with ArglO8, and changing the
chain length would alter this apparently optimal contact. Attempts to improve the inter-
actions with Arg 108 were not successful. Substitutions at the 7 position of the coumarin
ring were explored resulting in the most potent inhibitor, 15, which has a hydroxyl
group at that position. Again, the model did not show any interactions between this
18 BOHACEK, MCMARTIN, AND GUIDA

hydroxyl group and polar enzyme atoms. Using GRID, the position of a water molecule
to mediate a hydrogen bond between the hydroxyl and the carbonyl of Gly148 could be
identified. In conclusion, the authors indicated that the particular binding mode of this
lead made it difficult to extend the structure to form additional favorable contacts with
the enzyme.
More recently, another study from Parke-Davis and NC140 described the optimization
of structure 16 (Table V), another compound identified by random screening. To predict
the binding mode of compound 16, a number of docking experiments were conducted
using molecular dynamics as implemented in the program SYBYL (Version 5.5, available
from Tripos Associates, St. Louis, MO) and the Monte Carlo-based docking program,
Autodock.38 In order to optimize the position of the P1’ phenyl group to better fit into
the S1‘ enzyme subsite, analogs with 1-3 carbons between the sulfur and the benzene
ring were synthesized. A 2-3 fold enhancement in binding affinity was observed. Com-
pound 17, with a one carbon spacer, has a Ki of 0.7 pM. The binding mode of 17 was also
studied using the molecular modeling techniques mentioned above. Subsequently, the
X-ray structure of 17 complexed to the enzyme was determined and confirmed the
modeling results. The lactone moiety in the pyrone structure was found to displace
water molecule-301 (the critical water molecule usually found in the crystal structure of
HIV protease/inhibitor complexes). The lactone formed hydrogen bonds with Ile50 and
Ile150. The enol hydroxyl interacted with the two catalytic aspartic acid residues, Asp25
and Asp125.
Subsequent molecular modeling studies indicated that a tether could be added to the
6-aryl ring to reach the S1 subsite. Therefore, an analog with a OCH,COOH substituent at
the para position of the benzene ring located at the 6 position of the pyrone ring, which
would position a negatively charged group to interact with Arg 8, was designed and
synthesized. This compound, 18, was found to inhibit HIV protease with a Kiof 0.051 KM.

4. Optimization of a Lead Obtained by Screening Renin lnhibitors (Merck)


Screening of renin inhibitors, which had previously been prepared at Merck, identi-
fied compound 19 (Table VI) as a potent inhibitor of HIV-1 protease (Ki= 1.0 nM).
Although a potent inhibitor, 19 was not effective in inhibiting viral infection in H9 cell
cultures. Therefore, 19 was modified resulting in two series of compounds. The best of
the first series was 20.41Compound 20 is a very potent HIV inhibitor (1C5, = 0.03 nM)
with no renin activity and the ability to prevent the spread of viral infections in cell
cultures. Compound 21 was the best compound in the second series42 with an IC,, of
0.3 nM. It has antiviral activity similar to 20.
Although 20 and 21 are very potent inhibitors, they lack appropriate solubility and a
good pharmacokinetic profile and, therefore, a new study was undertaken to design
inhibitors to improve these properties.43 By incorporating the P1’ group of a Roche
inhibitor (22), it was hypothesized that the new compound (23) would be more water
soluble and that the conformational restrained decahydroisoquinoline would decrease
the entropy change upon binding to the enzyme.
The design process began with the X-ray structure of 22 complexed with HIV pro-
tease. Compound 21, the lead compound, was modeled into the active site using 22 as
a guide. The position of 21 was optimized using energy-minimization. Next, compound
23 was modeled to see if it would occupy the same space and form the same impor-
tant interactions as the potent known inhibitors 21 and 22. An excellent superimposi-
STRUCTURE-BASED DRUG DESIGN 19

TABLE VI
HIV Protease Inhibitors - Merck

H
19 I 1.0
N
L364,505 Boc-PhePhr'
7 0

20 0.03

d
21 0.3
L685,4324

22
Ro-31-8959

23 7.8

24 0.56
L-735,524
20 BOHACEK, MCMARTIN, AND GUIDA

tion was found, and compound 23 was synthesized. Compound 23 was found to inhibit
HIV with an IC,, of 7.8 nM, but was weak in inhibiting the spread of viral infections in
the cellular assay. However, because 23 possessed a favorable pharmacokinetic profile,
further optimization was carried out to modify the physical properties while maintaining
potency. The result of this optimization was L-735,524 (24), a compound undergoing
extensive human clinical trials (Phase 11). The crystal structure of a closely related analog
of 24 complexed with HIV protease was determined to 2.15 A resolution and confirmed
the modeling studies.

5. Optimization of a Penicillin Lead from Random Screening (Glaxo)


Screening at Glaxo led to the discovery of a crude sample of a penicillin dimer from
which, through an “artifact of the purification process,” a potent inhibitor, 25 (Table VII),
with an IC,, of 60 nM was discovered.& This compound was subsequently optimized
leading to several different series of potent inhibitors with good properties.45.46 The
X-ray crystal structure of a close analog of compound 25, in which the methyl esters
were replaced by ethyl amides, was determined to 2.5 A resolution47 and was used by
one of the groups at G l a ~ to o ~design
~ more potent inhibitors with lower molecular

TABLE VII
HIV Protease Inhibitors - Glaxo

Compound Potency

26

27 IC,, = 0.15 p,M

(continued)
STRUCTURE-BASED DRUG DESIGN 21

TABLE VII
(Continued)

Compound Potency

28 IC, > 170 KM

29 IC, = 0.082 FM

30 IC, = 1.9 pM

31 IC, = 4.6 p M
K, = 0.25 nM

32 IC, = 3.8 nM
22 BOHACEK, MCMARTIN, AND GUIDA

weight and, hence, an improved in uivo profile. Therefore, monomeric analogs of com-
pound 25 were designed.
The monomer of 25, compound 26, was found to be inactive against the enzyme, and
scientists at Glaxo speculated that at least one hydroxyl group to interact with the
catalytic aspartic acids and a hydrophobic group to bind to the S1 subsite would be
necessary for activity. A series of compounds that incorporated these groups was syn-
thesized resulting in compound 27 (1C5, = 0.15 pM). The significant difference in poten-
cy between 27 and 28, which differed only in the configuration of the P1' benzyl group,
was rationalized by inspection of the model. The "active" R enantiomer showed that the
benzyl group could fit into the S1' pocket, whereas for the "inactive" S isomer this was
not possible, These modeling results were later confirmed by the X-ray structure of 27
bound to the enzyme determined to 2.2 A r e s ~ l u t i o n . ~ ~
Inspection of the crystal structure of the analog of 25 complexed to HIV protease
suggested the importance of inhibitor carboxamide carbonyl oxygens in the center of the
molecule, which hydrogen bond with H,O 301, which in turn hydrogen bonds with the
NH of Ile 50 and Ile 50'. Compounds 27 and 28 cannot fully partake of these interactions,
since they have only one carbonyl group in this part of the binding site. From the crystal
structure, it was apparent that an additional carbonyl group separated by a 4-atom
spacer would be ideal. Therefore, a statine isostere was incorporated into the computer
model of 27. The carbonyl group was positioned in the model to form the desired
hydrogen bonds and the second amide group was adjusted to form potential hydrogen
bonds with the amide NH and the carbonyl of Gly27'. The structure was then subjected
to energy minimization in the static enzyme structure, which included the crystal-
lographic waters using the Insight/Discover molecular modeling software (available
from Biosym Technologies, Inc., San Diego, CA). Subsequently, compounds 29 and 30,
where synthesized and, 29, with an IC,, = 0.082 pM, did exhibit improved potency.
From the model, it was further postulated that a lipophilic amine could reach the S2'
subsite. A series of compounds were synthesized with a variety of hydrophobic groups
in the P2' position. The best of these were 31 and 32. Although both are potent HIV
inhibitors with an IC,, of 4.6 nM and 3.8 nM, respectively, only compound 31 exhibited
good antiviral activity in vim. Compound 31 had excellent selectivity, thus achieving the
goal of a potent, low-molecular-weight HIV protease inhibitor.

General Structure of Statine Analogs

These compounds are interesting in that they differ from the statine analogs. Statine
analogs have the S stereochemistry at the asymmetric center from which the P1 side
chains are attached. In compounds 31 the attachment is in the R configuration and the
lipophilic group binds not to the S1 subsite but to the S1' subsite. The crystal structure of
STRUCTURE-BASED DRUG DESIGN 23

TABLE VIII
HIV Protease Inhibitors - Glaxo

Compound
0

33, 34 3.1 pM
1.5 pM

22
RO-31-8959

35 9.0 nM
JG-365
Ac-Ser-Leu-AsnHH
OH
0 -He-Val-OMe

36 23 nM

32 complexed with HIV protease was subsequently determined and "confirms the for-
mation of the major interactions designed to optimize potency."48
Although very potent, these compounds displayed a poor pharmacokinetic profile and,
therefore, further optimization was undertaken.49The first targets of this new study were
compounds 33 and 34 (Table VIII). The modeling of these compounds was based on the
X-ray structure of a closely related analog of the earlier penicillin derivative, 25, complexed
to HIV protease. In addition, the position of the proline amide was modeled using the P1'
proline of JG365(35)50fromthe structure of JG365bound to synthetic HIV-1 pr0tease.~3The
biological results for 33 and 34 were disappointing (IC50of 3.1 and 1.5 pM, respectively)
when compared to JG365 (35), which has an IC,, of 0.5 nM, indicating that the interactions
that 33 and 34 make with the enzyme must be far from optimal.
Increasing the ring size of the P1' group by substituting a pipecolic acid for the proline did
not result in the expected increase in potency. Modeling the pipecolic acid analogs in the
manner described above did reveal that the pipecolic acid did not form any major
hydrophobic interactions with the enzyme. The decahydroisoquinoline moiety in the P1'
site suggested by the Roche inhibitor, Ro-31-8959, (22), was incorporated resulting in
24 BOHACEK, MCMARTIN, AND GUIDA

compound 36. In the model, improved hydrophobic interactions were observed between
the decahydroisoquinoline group and the enzyme, which did result in a substantial
increase in potency. Compound 36 was found to inhibit HIV protease with an IC,, of 23 nM.
Compounds 33, 34 and 36 were also modeled using the binding mode of the Roche
inhibitor, 22, as described in the communication of Krohn and co-workers.51 As in the
case when the JG365 structure was used as a template, the corresponding groups of 36
and the Roche inhibitor formed similar interactions with the enzyme. Unfortunately, co-
crystallization studies with these types of inhibitors were unsuccessful and, therefore,
the modeling results could not be verified.
Although further optimization was carried out using molecule modeling, compound
36 remained the most potent of the series in the enzyme and cellular assays. Unfor-
tunately, the pharmacokinetic profile of 36 was still poor, and the further development of
the penicillin-derived HIV protease inhibitors was abandoned.

6 . Optimization of the C-Terminus of a Known lnhibitor (Agouron)


This work52 is an example of the iterative use of molecular modeling and X-ray crystal-
lography at Agouron for the design of novel C-terminal inhibitors. The lead compound

TABLE IX
HIV Protease Inhibitors - Agouron

Compound K,

0.0018 p M

Q
38 1.67 pM

39

40 0.033 pM
STRUCTURE-BASED DRUG DESIGN 25

for this work was compound 37 (Table IX), a known inhibitor, which was modeled in
HIV using the X-ray structure of HIV-1 protease complexed with the inhibitor, MVT-101.
Using this structure, the C-terminal Val-Val methyl ester, which occupies the S2' and S3'
subsites of the protein, was replaced by a diphenylhydramine amide derivative in which
the two phenyl groups fill the 52' and S3' side-chain pockets. The compound, 38, was
synthesized and found to have a Kiof 1.67 pM. The crystal structure of 38 complexed
with HIV protease was solved. This structure compared favorably with the modeled
structure. A more elaborate inhibitor was designed, 39, which replaces the P3' phenyl
group with an indole ring designed so that the NH moiety of the indole would form a
hydrogen bond with the carbonyl oxygen of Gly 48. This compound, synthesized as a
mixture of diastereomers, was found to be eight times more potent than compound 38
with a Ki of 0.20 pM. Again, the X-ray structure was determined and compared to the
model structure. The indole was found to bind to the S3' subsite; however, the indole
NH did not make the predicted hydrogen bond but instead formed a water-mediated
H-bond to the Gly 27 carbonyl oxygen. At this point, the N-terminal alanines were
replaced by a moiety reported by Roche, i.e., an asparagine-2-quinoline-carbonylsubsti-
tuent.
Further optimization was carried out to see if substituents on the P2' phenyl group
could be designed to use the extra space revealed by the X-ray structure. Placing a
methyl or trifluoromethyl group in the meta position resulted in compound 40, which
showed small enhancements in binding affinity.
Subsequently, in order to test the free-energy perturbation method for computing
binding energies, calculations were performed on these compounds. Good correlation
was found between the calculated results and those experimentally determined.53

7. Design of a Macrocyclic Inhibitor Based on a Known


Protease Inhibitor (Marion Merrell Dow)
Screening of the Marion Merrell Dow protease inhibitor library resulted in the discov-
ery of 41 (Table X), a difluorostatone in which hydration of the fluorinated ketone is
believed to generate mimics of the tetrahedral intermediate. The lead compound was
optimized using traditional medicinal chemistry methods (no reported use of the struc-
tural information)54,55resulting in 42, a potent HIV-1 protease inhibitor with a Kiof 5 nM
and an improved selectivity index (ratio of EC,, for inhibition of cellular growth of MT,
cells infected with HIV-1 to toxicity).
A study was initiated to further optimize 42 using molecular modeling and crystal-
lographic data.56 The goal of this study was to increase the binding affinity of 42 through
additional favorable enzyme/ligand interactions, decreased molecular weight, and intro-
duction of conformational constraints designed to cause the solution conformation to be
similar to the bound conformation. Compound 42 was modeled in the HIV protease
binding site using the structure of HIV protease bound to MVT-101.57 Compound 42 was
docked using the inhibitor MVT-101 as a guide. After adjustments of bad van der Waals
interactions by hand, the complex was subjected to 50 ps of molecular dynamics using
the Insight/Discover software (available from Biosym Technologies, Inc., San Diego,
CA). This was followed by solvating the complex with two concentric shells of water.
The positions of the water were optimized using energy minimization followed by
subsequent molecular dynamics. The resulting structure shows phenyl rings A and B,
outside of the enzyme cavity in solvent, engaging in unfavorable hydrophobic/
hydrophilic interactions. It was postulated that these rings could be eliminated and that
the P1 phenyl group could be connected to the P3 portion of the molecule by formation
26 BOHACEK, MCMARTIN, AND GUIDA

TABLE X
HIV Protease Inhibitors - Marion Merrel Dow

Compound Potency (K,)

41 0.6 p,M

0.

42 5 nM
MDL 73,669

43 20 nM

of a macrocycle, compound 43. This idea had been successfully applied to the design of
macrocyclic potent pepsin inhibitors.58 Compound 43 was modeled in the active site of
HIV protease using the protocol described above, subsequently synthesized, and it was
found to inhibit HIV with a K iof 20 nM. NMR studies were performed to determine if
the macrocyclic portion of a close analog of 43 would show a solution conformation
similar to the bound conformation predicted by the model. Parts of the molecule in the
model of the complex corresponded well to the NMR data, whereas other sections of the
molecule could not be explained by the information determined by NMR studies. In
addition, in uacuo molecular dynamics simulations were performed on 43 in order to
qualitatively define its dynamical behavior. Subsequently, the X-ray crystal structure of
43 complexed to HIV was determined, and it was reported that those parts of the
molecule that had been characterized by NMR and found to be stable in the dynamics
simulations correspond well to X-ray structure, while remaining parts of the molecule
appeared to be a time average of two conformations defined by the dynamics calcula-
tions.
The authors do not address reasons for the lack of significant increase in binding
affinity of the macrocycle vs. the lead compound. We note, however, that Professor
Daniel Rxh has suggested that if a linear inhibitor already spends considerable time in
solution in a conformation close to the binding conformation, then introducing confor-
mational constraints may not lead to large gains in binding energy.59
STRUCTURE-BASED DRUG DESIGN 27

D. NEP Inhibitors-Neutral Endopeptidase, 24.11 (Ciba)


In this example, the 3-dimensional structure of the target binding site was not known
and, therefore, the structure of the homologous enzyme, thermolysin, was used to
successfully design novel potent NEP inhibitors.
Atrial natriuretic factor (ANF) is a cardiac hormone that has been shown to play an
important role in regulation of electrolyte levels and in the suppression of renin and
aldosterone secretion.60-62 It has been shown that infusion of ANF produces rapid
natriuresis, diuresis, and lowering of blood pressure.63,@The rapid metabolism of this
28 amino acid peptide makes ANF itself unsuitable as a dmg.65 However, with the
discovery of neutral endopepidase (NEP) 24.11 as the major enzyme responsible for the
clearance of ANF,66,67 inhibitors of NEP have been designed to potentiate the beneficial
effects of ANF by retarding its degradation.
NEP 24.11 is a zinc metalloprotease that catalyzes the hydrolysis of an amide bond on
the amino side of a hydrophobic residue. Pozsgay et a1.68 suggested that the optimal
substrate would have a P2-Pl-Pl' sequence of Phe-Gly(or A1a)-Phe(or Leu), and
Hersh and Morihara found that Leu would be ideal at the P2' site.69
The 3-dimensional structure of NEP 24.11 has not as yet been determined. However,

TABLE XI
NEP Inhibitors - Ciba

Compound G o

44 TLN: 9.1 nM (Ki)

45
HS.,. ..q? 0 COOH
NEP: 3.0 nM
TLN: 2.5 p M

46 NEP: 0.8 nM
CGS 25155

RCO0"
47 NEP: 1.0 nM
CGS 26670 TLN: 68 nM

HS'
28 BOHACEK, MCMARTIN, AND GUIDA

the X-ray crystal structure of thermolysin, a zinc metalloprotease, which is similar to


NEP in the primary sequence of the active site, substrate specificity, and inhibitor SAR,
has been successfully used as a model for NEP. In earlier work,70 the group at Pfizer
used thermolysin in the initial stages of the design of candoxatrilat, whose prodrug
candoxatril is under clinical evaluation as a potential therapy for congestive heart failure.
Later, the X-ray crystal structure of a candoxatrilat analog complexed to thermolysin was
determined and verified the model.71
Here we summarize the more recently published work carried out at Ciba. MacPher-
son, et al.72 designed novel, potent, orally active NEP inhibitors using the X-ray struc-
ture of thermolysin (see Table XI). Examination of the crystal structure of thermolysin
complexed with a transition state inhibitor, 44, showed that the two leucine side chains
occupied the S1’ and S2‘ subsites.73J4 These two subsites from a connected cavity in this
enzyme. This observation suggested that the side chains could be connected to form a
macrocycle, resulting in structure 45. Molecular modeling was then used to test this
hypothesis by docking the four stereoisomers of compound 45 into the binding site of
thermolysin. The docking was carried out by using Monte Carlo-like perturbation with
energy minimization to search for low energy ~onformations.~5 All but the R,R isomer
had low energy conformations, which formed good interactions with the enzyme. Upon
synthesis, the compounds were found to be very potent inhibitors of NEP and some-
what weaker inhibitors of thermolysin. The S,R isomer inhibited NEP and thermolysin
with an IC,, of 3 nM and 2.5 pM, respectively.
Compound 45 was subsequently optimized for oral availability resulting in CGS25155,
46, an orally active, novel, potent (IC, = 0.8 nM) NEP inhibitor. The crystal structure of
this compound in thermolysin was subsequently determined, revealing that this com-
pound has two binding modes, one of which is very similar to that predicted by molecu-
lar modeling76 (see Fig. 3).
In another study at Ciba, benzo-fused macrocycles were designed to inhibit NEP. To
prioritize the synthesis of these difficult-to-preparecompounds, macrocyclic compounds

Figure 3. CGS 25155, 46, an orally active potent NEP inhibitor (IC, = 0.8 nM) docked using molecular
modeling techniques in the binding site of thermolysin.
STRUCTURE-BASED DRUG DESIGN 29

containing 10- to 15-membered rings were first modeled in thermolysin. The model
showed that an 11-membered macrocycle, 47, formed the most favorable interactions
with thermolysin. Compounds with 12- and 13-membered rings also appeared favorable
in the model. The 11-, 12-, and 13-membered macrocycles were subsequently synthe-
sized and found to be potent inhibitors of NEP and thermolysin. In fact, the 11-
membered macrocycle inhibited thermolysin with an IC,, of 68 nM, making it one of the
most potent thiol thermolysin inhibitors yet reported.
Clearly, the ideal situation is to have available the 3-dimensional structure of the actual
target enzyme. However, this example illustrates that the structure of an homologous
enzyme that is sufficiently similar to the actual target can be a useful model for structure-
based drug design.

E. Dual Inhibitors of ACE and NEP (Ciba)


The design of dual ACE/NEP inhibitors at Ciba provides an additional example of a
structure-based drug design project in which the 3-dimensional structures of the target
enzymes have not as yet been reported. In this study, an homologous enzyme was used
to simulate one of the target enzymes and a 3-dimensional inhibitor composite template
was used as a model for the other.
ACE (angiotensin-converting enzyme, EC 2.4.15.1)catalyzes the hydrolysis of angio-
tensin I, a decapeptide, resulting in formation of angiotensin 11, a potent vasoconstrictor.
Angiotensin I1 is also involved in the release of the sodium-retaining steroid, al-
dosterone. Both factors result in an increase in blood pressure. The inhibition of these
processes by ACE inhibitors is one of the most effective therapeutic approaches for the
treatment of hypertension and congestive heart f a i l ~ r e . ~As
7 pointed out in Section D,
inhibition of NEP has been shown to potentiate the beneficial effects of the peptide ANF
and, therefore, NEP inhibitors are being investigated clinically for their ability to induce
ANF-like effects.78 Recently, preclinical results with combinations of ACE and NEP in-
hibitors have indicated possible synergistic effects.79-82 Therefore, a number of research
groups have been working to exploit some of the similarities of the two enzymes for the
design of a single compound capable of inhibiting both enzymes.
Like NEP and thermolysin, ACE is a zinc metalloprotease. ACE is a dipeptidyl carb-
oxypeptidase, usually cleaving the C-terminal dipeptide residues of its substrates. ACE
is rather nonspecific, although it does not favor substrates with terminal dicarboxylic
acids or a proline in the penultimate position.83
Scientists at Ciba have used an X-ray crystal structure of thermolysin as a model for
NEP and a 3-dimensional ACE inhibitor composite template to aid in the design of dual
ACE/NEP inhibitors.84 Initially, two potent ACE inhibitors previously discovered at Ciba
were combined to form a composite ACE inhibitor template, 48 (see Table XII), defining
the P1’ and P2’ regions of the inhibitor. This template was used during the early phases
of the project to model macrocyclic dual ACE/NEP inhibitors.85 As inhibitors were de-
signed that incorporated residues designed to bind in the S1 subsite, a more comprehen-
sive template was required. Therefore, four conformationally constrained, potent ACE
inhibitors were used to enlarge the initial template. A zinc atom was added to the zinc
chelating group of each inhibitor. The computer program, TFIT,86 was used to super-
impose all four inhibitors. This program identified low-energy conformations, which
allowed the zinc, the hydrophobic groups, the amido carbonyl, and the C-terminal
carboxylic acid group of each inhibitor to be superimposed onto the equivalent groups of
the other inhibitors.
30 BOHACEK, MCMARTIN, AND GUIDA

TABLE XI1
ACEiNEP Inhibitors - Ciba

Compound IC50

48
Initial ACE Template HS q COOH

49
Thiorphan
HS f N -coon ACE: 0.8 pM
NEP: 6 nM

46 ACE: 29.7 pM
CGS 25155 NEP: 0.8 nM

50 ACE: 6.1 nM
Benazaprilat 0 NEP: inactive
0

51 ACE: <50 nM
NEP: 2.7 nM
HS

52 ACE: 40 nM
NEP: 48 nM
HS

COOH
STRUCTURE-BASED DRUG DESIGN 31

A single molecule capable of inhibiting these two enzymes must possess those func-
tional groups required for tight binding to each enzyme. The models predict that the
conformation that an inhibitor must adopt to bind to ACE differs from the one required
for binding to NEP. Thus, an inhibitor must be able to adopt a conformation so that it can
bind to the S1, Sl', and S2' subsites of ACE and to the Sl', S2', and S3' subsites of NEP.
In the investigation at Ciba, inhibitors were proposed possessing conformationally re-
stricted sections designed to bind within certain areas in the binding site of each en-
zyme. These structural units were connected by a flexible hinge, allowing the entire
molecule to adopt conformations complementary to the active site of each enzyme. The
3-dimensional computer models were used as an aid in this design process.
The design strategy is outlined in Table XII. Comparison of two potent NEP inhibitors,
thiorphan 49 and the macrocycle 46, suggests that NEP can tolerate two different posi-
tions for the C-terminal carboxylic acid. The situation is different in ACE; thiorphan is a
moderate inhibitor of ACE; but, the macrocycle has little binding affinity for ACE, indi-
cating that the position of the C-terminal carboxylic acid is much more critical in ACE.
However, there are potent ACE inhibitors such as benazeprilat, 50, in which the zinc
chelating group is further removed from the terminal C-terminal carboxylic acid than in
thiorphan. Therefore, combining thiorphan and benazeprilat led to compound 51. To
further improve the potency of 51, a fused proline residue was incorporated into the
C-terminal portion of the molecule as suggested by the original template (48). Since
P-thiols such as 51 usually display poor pharmacokinetic properties, and we also wanted
to improve the in vivo stability of the thiol, the thiol was moved from the p to the a
position. a-Thiols had been previously shown to inhibit ACE.87 Thus, these two mod-
ifications led to compound 52.When docked and optimized using energy minimization
in the thermolysin model, this molecule formed good interactions with the enzyme
atoms. Thiol 52 could also be readily superimposed onto the ACE inhibitor template.
When synthesized, 52 was found to be a potent inhibitor of ACE (IC5" = 40 nM) and NEP

Figure 4. Tricyclic a-thiol dual ACD/NEP inhibitor, 52, bound to thermolysin. In pink is the conformation
predicted by molecular modeling. In green is the conformation subsequently determined by X-ray crystallogra-
phy. Relevant atoms of the thermolysin binding site are shown in darker colors.
32 BOHACEK, MCMARTIN, AND GUIDA

(IC50= 48 nM) and, to a lesser extent, of thermolysin (IC50= 1.6 pM). Scientists at both
Marion Merrell Dow8* and Bristol Myers Squibb89-91 have also reported on potent dual
ACE/NEP inhibitors containing the a-thiol functionality.
Because no 3-dimensional structure of an a-thiol bound to a zinc metalloprotease had
been reported, there was some concern about the reliabilityof the modeling of compound 52
in thermolysin. Therefore, the X-ray crystal structure of thermolysin complexed with 52
was determined at a resolution of 1.9 A. The close agreement between the conformation
predicted by molecular modeling to that determined experimentally (see Fig. 4) validated
the model and supported the hypothesis for the binding mode of 52 in NEP.M

F. PNP Inhibitors (Ciba, BioCryst, UAB, SRI)


In a collaborative venture between scientists at BioCryst Pharmaceuticals, Ciba Phar-
maceuticals, the University of Alabama at Birmingham, and Southern Research Institute,
inhibitors of the enzyme purine nucleoside phosphorylase (PNP) were designed using
structure-based techniques.92-96 Since this work has previously been reviewed in this
journal,97 we describe it here only briefly.
PNP is essential for purine salvage and catabolism, and is an important modulator of
T-cell proliferation. The enzyme catalyzes the reversible phosphorolysis of the purine
nucleosides guanosine and inosine (and their 2' deoxy analogs) and inhibitors of PNP
are potentially useful as therapeutic agents in host-graft rejection following organ trans-
plantation, and in T-cell proliferative disease such as T-cell lymphoma.
0

HOOH
HO O H

+ +

Using the X-ray crystal structure of the human PNP/ guanine complex determined to
2.8 A resolution, it was possible to obtain potent inhibitors for this enzyme. Neverthe-
less, one of the initial obstacles in the design of PNP inhibitors was the inability to
accurately predict the bound geometry of potential inhibitors prior to their synthesis.
Since all of the compounds synthesized in this investigation were guanine analogs, this
"docking" problem was reduced to the determination of the conformational preferences
of groups attached to the purine ring of these analogs in the PNP binding site. By using
Monte Carlo/energy minimization conformational search procedures,75,94 low-energy
conformers of potential inhibitors could be located in the PNP binding site. Synthesis of
the most promising compounds and X-ray crystal structure determination of the com-
plexes confirmed that, in all cases, one of the lowest energy conformers compared
STRUCTURE-BASED DRUG DESIGN 33

Figure 5. Low energy conformers of 9-cyclohexylmethyl-9-deazaguanine in the binding site of PNP. A phos-
phate ion is shown in purple. The pastel-colored molecules show several of the possible docking modes
discovered using a search procedure based on Monte Carlo torsion perturbation with energy minimization.

favorably with the X-ray derived structure. It was found that the Monte Carlo/ energy
minimization method as implemented in the MacroModel/ BatchMin software98 was
useful in the design of novel PNP inhibitors, since it provided a means for estimating
how potential (but unsynthesized) inhibitors might dock to the enzyme binding site.
This information provided a basis for estimating the relative binding affinities of these
inhibitors using the complementarity surface14 (see the Discussion section).
One might assume that a disadvantage of the Monte Carlo/ energy minimization
procedure is that it generates a collection of low-energy conformers rather than a single
conformer (i.e., the structure observed crystallographically).However, this ensemble of
conformers effectively maps the space available to molecules bound to the active site of
an enzyme and has proven to be useful for the design of PNP inhibitors. For example, by
comparing the low energy conformers of 9-cyclohexylmethyl-9-deazaguanine(a rela-
tively potent PNP inhibitor) in the binding site of PNP, the adamantyl analog 9-(2-
adamantylmethyl)-9-deazaguanine was suggested. This molecule was synthesized and
was observed to inhibit PNP with good binding affinity (see Fig. 5).
Recently, the Monte Carlo /energy minimization method has been extended and the
studies mentioned above have been repeated so that orientational as well as conforma-
tional sampling was performed.99 Thus, the molecule is allowed to translate and rotate
during the Monte Carlo step. The sampling is quite efficient, and the geometries located
(from poorly docked starting structures) compare well with crystallographically derived
models. This procedure can be used for the complete docking (translational, rotational,
plus conformational sampling) of ligands in binding sites.
34 BOHACEK, MCMARTIN, AND GUIDA

9-cyclohexylmethyl-9-deazaguanine 9-(2-adamantylmethyl)-9deazaguanine

The PNP design effort resulted in a number of 9-deazaguanine analogs that include
the ones shown below, which possess IC,,s in the 5-50 nM range. Using molecular
modeling techniques, molecules (see below) were designed to take advantage of the
major binding sites in PNP discovered using X-ray crystallography. The 9-deazaguanine
moiety forms key hydrogen bonds within the purine binding site, including one be-
tween Asn-243 and the N-7 hydrogen, which is only present in 9-deaza analogs. Substi-
tuents at R, were designed to fill the hydrophobic binding site and substituents at R,
were designed to fill the hydrophobic binding site and substituents at R, were designed
to interact with the phosphate binding site. By incorporating these design features, some
of the most potent PNP inhibitors yet discovered were produced.

R, = Aryl, Cycloalkyl, and


Heterocyclic Substituents

R2 = H, CH,CH2OH, CH,CN, CH2COOH


R2
R1

G. Sialidase Inhibitors (Monash University, CSIRO, Glaxo)


The design of inhibitors of the enzyme sialidase by investigators at Monash University
(Australia), CSIRO (Australia), and Glaxo (UK) provides another example of the success-
ful application of structure-based drug design techniques to the design of enzyme inhibi-
tors with therapeutic relevance. 100 Sialidase, which is also known as neuraminidase,
cleaves terminal a-ketosidically linked sialic acids from glycoproteins, oligosaccharides,
and glycolipids. It is expressed on the surface of the influenza virus and is thought to
play a role in the exit of viral particles from infected cells. It is also believed to aid in the
movement of the virus through the mucus lining of the respiratory tract. Thus, interest
in the design of potent inhibitors for this enzyme arises from the belief that these
compounds would serve as antiviral agents for influenza.
Based upon the X-ray crystal structure of sialidase (2.8 A structure of influenza virus
A/Tokyo/3/67 sialidase complexed with sialic acid and sialic acid analogs101,102),the most
potent influenza virus sialidase inhibitors yet reported were designed. An investigation
of the binding site of the enzyme using standard molecular graphics techniques and
using the GRID computer program39 aided in the design of novel inhibitors with excel-
STRUCTURE-BASED DRUG DESIGN 35

H OH

Sialic Acid (N-Acetyl-neuraminic acid)

lent binding affinities. GRID allows the user to explore the binding site of an enzyme
with various “functional groups” in order to locate energetically favorable binding re-
gions for these groups. A carefully parameterized molecular mechanics force field is
used to estimate the binding energy, displayed as contours, of the various probes. Thus,
with the aid of GRID, a protonated primary amine probe was used to suggest that
replacement of a hydroxyl group on the previously known inhibitor Neu5Ac2en (53)by
an amino substituent would yield a more potent analog. This molecule (54) was synthe-
sized and it was found to e an excellent sialidase inhibitor [Ki = 50 nM; A/Tokyo/3/67
(N2) sialidase]. Replacement of the hydroxyl group with a guanidino substituent pro-
vided an even more potent compound 155, Ki = 0.2 nM; A/Tokyo/3/67 (N2) sialidase].

It
H3C- C N -

R2
& \
COO‘
R,= H OH

CH20H

This example stresses the utility of the structure-based approach. The availability of
the X-ray crystal structure of sialidase allowed extremely potent inhibitors to be de-
signed in essentially one step using a previously known inhibitor as a lead.

H. Thymidylate Synthase Inhibitors


The enzyme thymidylate synthase (TS) catalyzes the conversion of deoxyuridylate
monophosphate to thymidylate monophosphate (dTMP) via a methylation reaction in
which 5,lO-methylenetetrahydrofolateis involved as a co-factor. TS is involved in the
sole, rate-limiting pathway for the biosynthesis of dTMP.
Inhibitors of TS have been shown to possess broad spectrum activity as antiprolifera-
tives and antitumor agents, and, thus, have become attractive candidates for cancer
36 BOHACEK, MCMARTIN, AND GUIDA

2-03m-u OH

durn
0

-
TS

FH4 -
I
OH

dTMP

chemotherapy. TS has also represented an attractive target for structure-based drug


design, since structural information about the TS binding site has been available for
some time now.

1. Optimization of Thymidylate Synthase lnhibitors (Agouron)


The X-ray structures of E . coli TS in its apo form and as the ternary complex were
determined several years ago to moderately high resolution103-106 (2.0-2.5 A), and scien-
tists at Agouron Pharmaceuticals have been able to utilize protein crystallography and
molecular modeling in an iterative fashion to design potent inhibitors of this enzyme,
yielding at least one compound that is currently in phase I1 clinical trials as an anti-
tumor agent.3~107
The initial goal of the effort to obtain potent, bioavailable inhibitors of TS involved the
structure-based modification of known TS inhibitors such as 56 (see Table XIII). This
compound possesses excellent potency (Ki= 8 nM; human TS) but, nonetheless, has
liabilities. It is actively transported into cells, which may lead to drug resistance that has
been observed for a number of classical antifolates such as 56.Thus, the optimization of
lipophilic analogs of 56 was sought in order to afford inhibitors that would cross the cell
membrane by passive diffusion. Structure-based lead optimization was employed, using
crystallographic analysis, of inhibitors complexed to the E . Coli enzyme, which was
available in sufficient quantities for iterative studies. Synthesis of 57, in which the
p-COGlu moiety had been removed, yielded a compound that possessed a Kiof 2.2 p M
against human TS and, thus, had lost 2.4 orders of magnitude in binding affinity.
Examination of the X-ray crystal structure of 58, the closely related 2-amino analog of 56,
bound to E. coli TS revealed the existence of a hydrophobic binding site near the meta
position of the phenyl substituent of 58.Synthesis of 59 produced an inhibitor with a Ki
of 0.39 pM against human TS and its binding to the E . coli enzyme was as predicted.
Aided by crystallographic analysis of the TS binding site, it was postulated that a large
hydrophobic substituent, which also possessed an electron withdrawing group, posi-
tioned at the para position of the phenyl group of 57 would lead to enhanced binding.
Using semi-empiricalMO calculations (AMl/MNDO) to search conformational space for
appropriate low-energy conformers in vucuo, it was postulated that compound 60 should
not incur appreciable strain energy upon binding to TS. Compound 60 was synthesized
and it possessed a K iof 13 nM against human TS. Thus, potency was nearly restored to
that of the initial lead compound.
STRUCTURE-BASEDDRUG DESIGN 37

TABLE XI11
Thymidylate Synthase Inhibitors - Agouron

Compound Ki

56

COGlu

57 2.2 p M

58

59 0.39 uM

60 13 nM

I
SOzPh

2. De novo Design of ThyrnidyIate Synthase Inhibitors (Agowon)


In a second example, which the authors have described as de novo design,3,107,10*
potent TS inhibitors were designed for the cofactor binding site using the crystal struc-
ture of E . coli TS complexed with 5-fluoro-2’-deoxyuridylate.The GRID39 software was
used to locate hydrophobic regions in the binding site by employing a methyl probe. It
was observed that a naphthalene ring would achieve good overlap between the aromatic
ring and the GRID contour map. Using the naphthalene ring as a template, hydrogen
bonding groups were envisioned at the 1-and 8-positions to provide chemical comple-
mentarity for an aspartic acid in the binding site and for a bound water molecule. This
and further analysis lead to the design of a benz[cd]indole ring substituted at the 6-posi-
38 BOHACEK, MCMARTIN, AND GUIDA

TABLE XIV
Thymidylate Synthase Inhibitors - Agouron
~

Compound Ki

1.6 FM

62 34 nM

63 2 nM

tion with a disubstituted nitrogen (see Table XIV). A benzyl group was used as one of the
substituents and it was substituted at the 4-position with a piperazine sulfonamide in
order to impart water solubility to the molecule. Analysis of the binding site indicated
that the piperazine ring should be oriented toward solvent. Synthesis of compound 61
afforded a TS inhibitor that possessed a K j of 1.6 pM against human TS. X-ray analysis
indicated that it was bound to the enzyme in a fashion similar to the one predicted by the
modeling studies. Improvements in the binding affinity of this lead inhibitor were made
using iterative structure-based drug design techniques to afford compound 62 and 63
with K j s of 34 nM and 2 nM (human TS), respectively.

3 . Discovery of Thymidylate Synthase lnhibitors Aided by 3 - 0 Database Searching (UCSF)


Another example109 of the application of structure-based drug design techniques to
the discovery of TS inhibitors involves database searching using the DOCK computer
program*10-”* developed in Tack Kuntz’s laboratory at the University of California, San
Francisco, and the commercial software MACCS-3D35 (available from MDL Information
Systems, Inc., San Leandro, CA). MACCS-3D allows one to search a 3-dimensional
database in order to find structures that meet user-imposed 3-dimensional constraints
(e.g., distances between key functional groups). The DOCK program can also be used to
search 3-dimensional databases, but it operates by locating structures that are compati-
ble with a target binding site whose 3-dimensional structure is known. The DOCK
program can be used to ”score” potential inhibitors with respect to their approximate
binding affinity.
For the design of TS inhibitors, .DOCK was employed to search the Fine Chemicals
Directory (which is distributed by MDL Information Systems, Inc. and is now referred to
as the Available Chemicals Directory) to locate molecules with potential affinity for the
STRUCTURE-BASED DRUG DESIGN 39

Phenolphthalein
Sulisobenmne
ICso = 15 PM

TS binding site. The X-ray crystal structure of L. casei TS determined to 2.3 8, resolution
was used for the docking experiments performed with DOCK. The MACCS3D program
was then used to search the Fine Chemicals Directory to locate structures that were
similar to sulisobenzone, one of the molecules that had been found by DOCK. DOCK
was then used to dock and score each of the molecules located by MACCS3D. Using this
methodology, phenolphthalein analogs were found that inhibit TS in the low micromo-
lar range. It was suggested that these compounds could serve as potential leads for the
synthesis of novel, more potent analogs.
Interestingly, when the X-ray structure of the sulisobenzone/TS complex was deter-
mined to 2.5 /! resolution in two different buffer solutions, two different binding modes
were observed, which depended upon the buffer. Both structures differed from the one
suggested by DOCK and revealed an alternate binding region within the active site of
TS. The composite binding region was used to dock and evaluate the compounds found
as a result of the MACCS3D search. When the X-ray structure of the phenolphthalein/TS
complex was determined, it was found that the binding mode of phenolphthalein was
similar to the one suggested by DOCK.

111. DISCUSSION
The examples covered in this and other reviewsz-4 show that structure-based design is
a fertile approach to drug discovery that has enormous potential. It should be pointed
out that the literature in this area is already very extensive, and, as a result, reviews have
not yet appeared that are fully representative of a rapidly growing field. The numerous
examples covered in the present review, drawn only from the most recent literature,
suggest a state of exponential growth for this approach to drug discovery.
It is clear that a large number of successful new approaches are currently being devel-
oped in which structural information is employed for the design of drugs. We have the
impression of an area of science in rapid transition, which offers considerable scope and
opportunity to scientists who are comfortable working at the interface between disci-
plines and have imagination and initiative. At present, there is no single well-beaten
path to success. Rather, there are a number of different ways in which experimental
data, or in some cases hypotheses, about the interactions between potential drugs and
their host binding site are being used to catalyze, inspire, and focus the search for
biologically active compounds. In this sense, the practice of structure-based drug design
is somewhat of an art and we have attempted in this review to provide a flavor of the
state of the art in this endeavor.
40 BOHACEK, MCMARTIN, AND GUIDA

The process of drug discovery can be conveniently divided into two stages: (1) the
discovery of novel lead compounds and (2) the optimization of these compounds. A
suitable lead will typically have a binding affinity in the micromolar range and should
also be chemically accessible so that modifications can readily be made. Formerly, in the
absence of the structure of the bound ligand, the easiest leads to convert to a bioavailable
drug were small and not too flexible. For example, it would have been very difficult to
optimize a linear hexapeptide lead compound. Where high-quality structural informa-
tion is available, this task becomes approachable. Once the structure of the complex of a
ligand with a binding site is known, smaller analogs with very different physicochemical
properties can be designed. Some examples are described in Section IIC of this review.
Lead optimization is usually necessary to obtain compounds that are sufficiently po-
tent and have appropriate bioavailability characteristics. Leads may also require optimi-
zation to improve selectivity, although generally, as more potent drugs with highly
optimized binding characteristics are developed, they can be expected to be more selec-
tive. Since lower doses will be required, potent compounds are less likely to have side-
effects arising from actions unrelated to the mechanism of action.
Although the distinction between lead discovery and lead optimization is somewhat
arbitrary, it will serve in this discussion to distinguish between two different aspects of
structure-based design:
1. The use of structural information to suggest and plan a series of small changes
intended to optimize the structure of a weakly active compound. The structural
information may help to eliminate steps with little chance of success and may
encourage or justify the spending of resources on more difficult chemical syn-
theses of compounds that may have improved properties.
2 . The design or discovery of radically different molecules to provide new leads.
Ideally, these should be capable of optimization to highly potent compounds.
Within the drug industry it is also important that they should have sufficient
novelty to be patentable.

A. Optimization of Lead Compounds


Structure-based optimization must be based on a 3-dimensional model of a docked
starting compound. This model can be hypothetical or based on computational or experi-
mental methods.
At present, optimization of a lead compound is more reliable when there is an experi-
mental structure of the complex of that compound with the binding site. In fact, once a
round of optimization has occurred (compounds made and tested for potency), it is often
useful to obtain experimental structures for new compounds to be used as leads for
further optimization. Even more informative results can be obtained using a series of
compounds of differing potency for experimental structure determination so that accu-
rate docking modes can be obtained for each compound. From these results, it may be
possible to learn in detail the factors that seem to be responsible for potency (see, for
example, Section IIA on carbonic anhydrase, and Section IIH-3 on thymidylate syn-
thase). These studies may not be essential for a specific optimization task, but they are of
tremendous value for the advancement of the science of structure-based drug design.
In addition, the experimental results obtained at high resolution can give information
about ordered water molecules and about small but potentially important changes in the
enzyme binding site. They may also reveal a totally unexpected change of docking mode
(see Sections IID and IIH-3), which potentially could be utilized for further optimization.
STRUCTURE-BASED DRUG DESIGN 41

Ideally, computational methods would provide the information currently gained by itera-
tive cycles of synthesis and structure determination. This would greatly accelerate the
optimization process by reducing the number of compounds that need to be synthesized
and the time-consuming process of further structure determinations. There are, in fact, a
considerable number of methods for docking.113 Reasonable results have been obtained
in some cases. However, in general, these methods cannot be relied upon to always
produce the correct result. Despite this drawback, it can be very helpful to perform
docking studies to select promising candidates from a variety of molecules that require
lengthy synthesis. A good example, from our laboratories, of the effective use of a
thorough docking study is work on benzo-fused macrocyclic NEP inhibitors (see Section
IID), where the study helped to decide which of a large series of cyclic structures to
synthesize. A similar example has also been described in Section IIF of this review on
docking of PNP inhibitors. We believe computer docking methods are most reliable
when the binding site is relatively rigid and the docking geometry is largely determined
by steric complementarity. These methods are less effective at predicting the correct
geometry when a number of alternative docking geometries of low energy is possible.
Although experimental docked geometries may differ from those obtained from a
computational study, it is probable that, whenever a computation shows a binding mode
with high steric and chemical complementarity, the molecule will indeed bind. A lower
energy binding mode may be possible, but the computational method can still be ex-
pected to act as an effective filter for active compounds. For example, as described in
Section IIH-3, a molecule (sulisobenzone) selected by DOCK as a candidate for inhibition
of thymidylate synthase was subsequently shown by X-ray crystallography to have a
binding mode different than the one predicted using DOCK. In fact, two binding modes
were found, depending on the buffer, and a composite of both of the complexes was
then used for further database searching resulting in inhibitors with IC,,s in the low
p.M range.
A number of considerations are relevant to the process of optimization. They are
discussed here in approximate order of ease and reliability. In general, the effects of
changes that lower conformational energy, or that avoid bad van der Waals contacts, can
be predicted with reasonable confidence. Those based on nonbonded interactions in-
volving electrostatic and solvation energies are difficult to predict and less accurate.

1 . Improvement of Complementarity to the Binding Site


Steric fit to a binding site is one of the easiest factors to estimate. Van der Waals contact
distances are known with high accuracy and are, in fact, assumed as invariants (along
with bond angles and bond lengths) in the process of structure refinement from experi-
mental data. Chemical complementarity can be reliably assessed at the level of simple
scoring of hydrogen bonding and hydrophobic contacts.14 However, subtle differences
due to charge distribution in the molecule or to solvation effects are more difficult to
predict. A reasonable approach is to use broad complementarity criteria to design a
series of conipounds that can be made and experimentally tested.
Improvements in complementarity based on small changes can usually be proposed
by simply looking at the docked complex on a high-resolution graphics workstation. We
have found the display of a solvent-accessible surface of the binding site, represented as
a grid, to be very helpful in this process. The surface can also be colored to indicate
chemical complementarity. A coloring algorithm based on experimental data has proven
to be reliable in our hands.14 A different approach, which has been used successfully in a
42 BOHACEK, MCMARTIN, AND GUIDA

number of cases, is the application of the program GRID39 to identify and graphically
display positions in the binding site where a variety of functional and hydrophobic
probe groups can be expected to bind (see Sections IIG and IIH-2). Although it can be
argued that these methods are not essential for drug design, one should not underesti-
mate their ability to guide and stimulate the creative process by clearly displaying oppor-
tunities for improving the binding affinity of the ligand.

2. lmprovement of Conformational Properties


Where the docked structure is strained or has a high degree of flexibility, it may be
possible to introduce changes that make the molecule less strained or more rigid. The
likely effects of these changes, which are estimated by computational conformational
analysis of the starting compound and the modified compounds, can usually be pre-
dicted with some confidence. Typically, molecular mechanics is used for this purpose.
In cases where the force-field parameters are inadequate (e.g., with unusual functional
groups, highly conjugated systems, or strained rings) it may be desirable to perform
quantum mechanics calculations (see, for example, Sections IIA and IIH-1). Semi-
empirical and/or ab initio calculations are now accessible using a workstation of modest
cost.

3 . Improvement of Bioavailability
It may be necessary to change physicochemical properties such as octanol/water parti-
tion coefficient and water solubility in order to improve bioavailability. Usually, it is
important to achieve this with minimal loss in potency. The structure of a docked com-
plex can be used: (a) by designing potency-neutral changes in parts of the molecule in
contact with solvent or (b) selecting parts of the molecule that, although in contact with
the site, do not seem to be important for binding. A number of examples of successful
optimization of bioavailability have been cited in this review (see Section IIC).

B. Finding New Leads


A number of different approaches are available. To provide perspective, we start with
screening methods that do not require structure information.

1 . High Throughput Screening of Coinpound Libraries


Compounds are tested experimentally for ability to inhibit binding or an enzyme
catalyzed reaction. The process is indicated in Fig. 6. The advantage of this method is
that it can be applied in the absence of any structural knowledge and that "hits," once
validated, are known with confidence to be active. With the development of combina-
torial libraries, it is becoming possible to screen large numbers of compounds. However,
as indicated in Fig. 6, the universe of potential molecules is very large, and it is by no
means certain that libraries of synthetic compounds, whether combinatorial or not, will
adequately sample this universe.
Experience has shown that it is often possible to obtain lead compounds through high
throughput screening. It is not clear, however, that this process will always find leads or
that it will find the best leads in terms of ease of synthesis, ease of optimization,
bioavailability, and selectivity.
STRUCTURE-BASED DRUG DESIGN 43

Figure 6. Schematic illustration of a search for lead compounds by screening. Molecules are assumed to be
arranged so that similar molecules are close to each other. A subset of molecules containing up to 30 C, N, 0,
and S atoms may have more than 1060 members=. Contours show regions where the molecules are potent
inhibitors for a specific (in this case hypothetical) binding site.

2. Computer-Based Screening of Structural Databases


This process, exemplified by the program DOCK,"0-11* can be used to limit the
number of candidates and, thus, accelerate the physical screening process or to find a
small number of compounds that can be purchased or synthesized and tested for po-
tency. It uses the structure of a target binding site and a database with the 3-dimensional
structures of a collection of molecules.

aAlthough the number of possible molecules is difficult to estimate accurately, simple considerations show
that it must be very large! Consider growing a linear molecule an atom at a time and choosing a carbon,
nitrogen, oxygen, or sulfur atom at each position. Some of these atoms can be doubly or triply bonded, but not
all combinations of atoms are chemically stable, and some multiple bonds will only be possible in nonlinear
structures, i.e., a C=O group. Assuming a very approximate average choice multiplicity of 6, then 630 or 2 X
1023 molecules could be grown containing 30 atoms. Now consider the ways of introducing branching or
cyclization into the resulting structure. Closure of rings with three or more atoms involves selecting two atoms
to form a bond and could be achieved in 30'28/2 ways. Making a branched molecule could be achieved by
choosing a point to cut the chain and a point in the first part of the chain to attach to the cut end of the second
part of the chain (i.e., 302 ways). Not all atoms can be joined in this way. However, this will be offset by the fact
that when stereochemical considerations are introduced, the number of possibilities will be expanded. Based
upon these considerations, approximately 1040 molecules with u p to four rings and 10 branch points could be
produced from each linear chain, resulting in a very approximate estimate of 1065 molecules in total. Although
this is a rough estimate, it seems likely that when all the different possible combinations of ring closure and
branching are taken into account, the true number will be well in excess of 1060 and will rise steeply with
increasing molecular weight.
44 BOHACEK, MCMARTIN, AND GUIDA

3. Design of New Leads (de novo Design)


New leads can be designed interactively at the workstation. As mentioned in Section
IIIA on lead optimization, the use of a graphical representation of the surface of the
binding site can be very helpful for this purpose. It is, however, difficult to design highly
novel structures in this way, because it is very hard to visualize the effects of large
changes without building and optimizing the proposed structures. Furthermore, a very
large number of changes is possible. This process can be handled very effectively using
completely automated computer algorithms for de novo design. The method with which
we are most familiar is our own program GrowMol,6 which uses a complementarity grid
based on the complementarity surface14 described earlier. This guides the growth of
molecules to achieve a high degree of complementarity with the binding site. This
method produces very large numbers of unique structures and has, in fact, revealed the
wealth of potential inhibitors available in even a quite constricted site such as that of
thermolysin.6
Figure 7 shows how, with suitable rules for growing the molecule an atom (or func-
tional group) at a time, and a probabilistic rejection of new atoms based on complemen-
tarity, it may in the future be possible to search the entire universe of small organic
molecules for potential drugs.
De nozm computer programs differ somewhat in their philosophy, with some biased
towards producing only one or a few compounds intended to be highly potent whereas

:one containing a11 possible molecules


srranged vertically according lo size,
msifioned so fhaf similar molecules
B r a close together.

Pofenf molecules

complemenfary

Per
molecule Zone 01 growth -
molecules wifh high
sferic and chemical
complemenW i l y
to binding site

Figure 7. Algorithm for de nouo drug design. Atom-by-atom growth of molecules, complementary to a target
binding site, searches the universe of organic molecules for potential drugs.
STRUCTURE-BASED DRUG DESIGN 45

other programs produce a wide range of compounds. The latter approach provides the
synthetic chemist with a variety of choices, and enables compounds to be selected based
on synthetic accessibility and on subjective factors such as likelihood of achieving bio-
availability. Of course, rules might be devised that will further filter out the output
structures to avoid difficult syntheses or compounds unlikely to be bioavailable.
An alternative approach to structure-based design is to explore possible ways of con-
necting fragments selected and docked by the user. The NEWLEAD program, for exam-
ple, accomplishes this readily and is easy to use.114

4. Potency Prediction
A major requirement for the effective design and optimization of lead compounds is
the prediction of potency. With a reliable predictor, we will be able to scan the large
number of structures produced by de novo programs and select those having potencies
better than a threshold value. Potency prediction should also increase the probability of
forecasting accurate docking geometries for inhibitors and eliminate some of the itera-
tions in the process of optimization. Indeed, an ultimate goal would be to dispense with
the distinction between lead discovery and optimization, and design potent novel mole-
cules in a single step.
A large number of factors are important for determining the binding free energy,
including:

1. Interaction energies (van der Waals and electrostatic) with the binding site
2. Solvation free energy changes
3. Loss of conformational entropy
4. Internal (strain) energy of the bound molecule

For a given series of molecules in a specific binding site, some of these factors may be
much more important than others. For example, we found that for a series of thermo-
lysin inhibitors, a simple complementarity score, based on the number of hydrogen
bonding and hydrophobic contacts, gave a very high correlation with potency.6~14Al-
though the scoring method is simple, the parameters defining hydrophobic contacts and
hydrogen bonds were critical to the success of the method and were based on a careful
study of complementarity in proteins.
Very different results were obtained in the work, cited earlier on carbonic anhydrase
(Section IIA), which clearly shows that internal strain energy is the predominant factor
for the series of compounds reported.3
In another example reported recently,115 a high degree of correlation was achieved
between the calculated intermolecular interaction energy (van der Waals plus electrosta-
tic energies), for a series of HIV-1 protease inhibitor complexes and the observed i n vitro
potency. The inhibitors had been subjected to energy minimization in the HIV protease
binding site. This methodology was subsequently used to predict the potency of poten-
tial HIV protease inhibitors in advance of their synthesis and testing for biological
activity.
Each of these methods accounts for only a subset of the terms listed above, which
account for the total free energy of binding and can be used with confidence only when a
learning set of compounds is available. A method is needed that takes into account all
the factors listed above and can be applied without further parameterization to any
inhibitor in any binding site.
46 BOHACEK, MCMARTIN, AND GUIDA

Computational methods that involve free energy simulations,1*6,117which explicitly


account for the factors mentioned above, are poten t i d y capable of reproducing accurate
relative binding free energies. However, at present, the range of drug design problems
to which they can be successfully applied appears to be quite limited. Free energy
perturbation calculations, for example, are computationally demanding and the CPU
time required to ensure convergence can be excessive. Furthermore, the method is
usually restricted to small differences between the pair of molecules for which one
wishes to compute the difference in binding free energy. Adequate conformational sam-
pling must be performed, if flexible ligands and/or large structural perturbations are
involved, and this task can be prohibitive for all but the simplest of systems. Nonethe-
less, these methods are continuously being improved. A recent publication,lls for exam-
ple, indicates how the conformational sampling problem might be overcome by combin-
ing Monte Carlo sampling with molecular dynamics, using a continuum solvation
model.
There have been only a few reports of the application of free energy simulation
methodology to structure-based drug design. One exception is the work by Reddy, et
al.53 in which the free energy perturbation approach was employed for the design of
HIV-1 protease inhibitors. Recently, a potentially more rapid semi-empirical method for
estimating absolute free energies of binding based on data collected during molecular
dynamics simulations has been described. *I9 The method was applied to five endo-
thiapepsin inhibitors with moderate structural diversity. It lacks the rigorous theoretical
basis of free energy simulation methods and it remains to be seen whether it will be
generally applicable to other systems.
A recent study120 investigated the use of multiple scoring factors to predict potencies
of a number of inhibitors for a large number of enzymes. The range of enzymes for
which a single scoring algorithm was able to predict potency was impressive. The
enzymes included examples in which the ligand was free, as well as those where the
ligand was bound to a metal ion. The standard deviation of the correlation was approx-
imately 1.5 log units, corresponding to a factor of 30. The prediction of relative potencies
of similar compounds can be expected to be much better. This method, which is very
rapid, should provide a useful basis for prioritizing compounds prior to their synthesis.
These results are encouraging and we anticipate that it will be possible to improve
upon the accuracy of such methods further. A method with an accuracy of a factor of ten
within a 95% confidence limit would provide a very powerful tool for computer-based
design of new ligands.

IV. CONCLUSIONS
The art and practice of structure-based drug design is rapidly evolving into a legiti-
mate scientific discipline. We use the term "scientific discipline" with confidence since
the best work in this area currently involves:

1. Precise and reproducible experimentation


2. Use of sophisticated theoretical models, which can be rigorously subjected to
exact experimental verification

The scientific foundations for this approach were established many years ago with the
development of X-ray and NMR techniques for the accurate determination of structures
of macromolecular complexes and with the development of computational methods for
molecular mechanics and quantum mechanics. However, the basic requirement for the
STRUCTURE-BASED DRUG DESIGN 47

successful application of these techniques to drug discovery is a well equipped, multi-


disciplinary team of actively collaborating scientists. Many pharmaceutical companies
invested heavily in this area at a time when this approach was as yet unproven. We are
now seeing the results of this investment (see Section 11). With some notable exceptions,
our impression is that most of the leading-edge research in structure-based molecular
design is currently being carried out in industry.
While structure-based design is clearly proving to be of value in many drug discovery
projects, the full potential of this approach has not yet been realized. At present, suc-
cessful applications rely heavily on repeated cycles of potency measurement and experi-
mental structure determination. In this process, the full potential of computational mo-
lecular modeling methods has not yet been reached. Improved computational methods
are necessary to optimally utilize available structural information. Methods for predict-
ing docked geometries, conformational changes in the binding site, and binding affini-
ties of novel molecules can be expected to improve. Once these operations become
sufficiently reliable, the process of lead finding and optimization can be expected to be
truly revolutionized.
Based on past and present accomplishments, further advances in the art and practice
of structure-based drug design are expected. These advances will enable us to very
rapidly obtain novel, highly potent compounds for a target binding site, once its 3-di-
mensional structure is known.

REFERENCES
1. E. Fischer, Ber. Dtsch. Chem. Ges. 27, 2985 (1894).
2. J. W. Erickson and S. W. Fesik, in Annual Reports in Medicinal Chemistry M. C. Venuti, Ed., Academic Press,
Inc., New York, 1992, p. 271.
3. J. Greer, J. W. Erickson, J. J. Baldwin, and M. D. Varney, 1. Med. Chem. 37, 1035 (1994).
4. C. L. M. J. Verlinde and W. G. J. Hol, Structure 2, 577 (1994).
5. C. E. Bugg, W. M. Carson, and J. A. Montgomery, Scientific American 269, 60 (1993).
6. R. S. Bohacek and C. McMartin, 1. Am. Chem. SOC. 116, 5560 (1994).
7. V. J. Gillet, W. Newell, P. Mata, G. Myatt, S. Sike, Z. Zsoldos, and A. P. Johnson, 1. Chem. I f . Comput. Sci.
34, 207 (1994).
8. D. A. Pearlman and M. A. Murcko, 1. Comp. Chem. 14, 1184 (1993).
9. S. H. Rotstein and M. A. Murcko, 1. Camp.-Aided Mol. Design 7, 23 (1993).
10. S. H. Rotstein and M. A. Murcko, 1. Med. Chem. 36, 1700 (1993).
11. Y. Nishibata and A. ltai, 1. Med. Chem. 36, 2921 (1993).
12. E. E. Abola, F. C. Bernstein, S. H. Bryant, T. F. Koetzle, and J. Weng, in Data Commission offhe Int’l Union of
Crystallography, F. H. Allen, G. Bergerhoff, and R. Sievers, Eds., Int’l Union of Crystallography, Bonn,
Cambridge, Chester, 1987, p. 107.
13. F. C. Bernstein, T. F. Koetzle, G. J. 8 . Williams, E. F. Meyer, M. D. Brice, J. R. Rodgers, 0. Kennard,
T. Shimanouchi, and M. Tasumi, 1. Mol. B i d . 112, 535 (1977).
14. R. S. Bohacek and C. McMartin, 1. Med. Chem. 35, 1671 (1992).
15. D. G. Alberg and S. L. Schreiber, Science 262, 248 (1993).
16. C. Weber, G. Wider, B. von Freyberg, R. Traber, W. Braun, H. Widmer, and K. Wuthrich, Biochemistry 30,
6563 (1991).
17. S. W. Fesik, R. T. Gampe Jr., H. L. Eaton, G. Gemmecker, E. T. Olejniczak, P. Neri, T. F. Holzman, D. A.
Egan, R. Edalji, R. Simmer, R. Helfrich, J. Hochlowski, and M. Jackson, Biochemistry 30, 6574 (1991).
18. H.-R. Loosli, H. Kessler, H. Oschkinat, H.-P. Weber, T. J. Petcher, and A. Widmer, Helu. Chim. Acta. 68,
682 (1985).
19. H. Kessler, M. Kock, T. Wein, and M. Gehrke, Helu. Chim. Acta. 73, 1818 (1990).
20. W. L. Jorgensen, Science 254, 954 (1991).
21. K. Wuthrich, B. von Freyberg, C. Weber, G. Wider, R. Traber, H. Widmer, and W. Braun, Science 254,953
(1991).
22. G. D. Van Duyne, R. F. Standaert, P. A. Karplus, S. L. Schreiber, and J. Clardy, Science 252, 839 1991).
48 BOHACEK, MCMARTIN, AND GUIDA

23. T. Taga, H. Tanaka, T. Goto, and S. Tada, Acta. Cryst. C43, 751 (1987).
24. D. F. Mierke, P. Schmieder, P. Karuso, and H. Kessler, Helv. Chim. Acfa. 74, 1027 (1991).
25. A. M. Petros, J. R. Luly, H. Liang, and S. W. Fesik, 1. A m . Chem. Soc. 115, 9920 (1993).
26. D. Altschuh, 0. Vix, B. Rees, and J.-C. Thierry, Science 256, 92 (1992).
27. J. L. Kofon, P. Kuzmic, V. Kishore, G. Gemmecker, S. W. Fesik, and D. H. Rich, 1. Am. Chem. Soc. 114,
2670 (1992).
28. N. H. Sigal, F. Dumont, P. Durette, J. J. Siekierka, L. Peterson, D. H. Rich, 8. E. Dunlap, M. J. Staruch,
M. R. Melino, S. L. Koprak, D. Williams, B. Witzel, and J. M. Pisano, 1. Exp. Med. 173, 619 (1991).
29. A. A. Patchett, D. Taub, 0. D. Hensens, R. T. Goegelman, L. Yang, F. Dumont, L. Peterson, and N. H.
Sigal, 1. Antibiot. 45, 94 (1992).
30. G. Lauri and P. A. Bartlett, 1. Computer-Aided. Mol. Design 8, 51 (1994).
31. F. H. Allen, J. E. Davies, J. J. Galloy, 0.Johnson, 0. Kennard, C. F. Macrae, E. M. Mitchell, G . F. Mitchell,
J. M. Smith, and D. G. Watson, 1. Chem. In!. Comput. Sci. 31, 187 (1991).
32. P. Y. S. Lam, P. K. Jadhav, C. J. Eyerman, C. N. Hodge, Y. Ru, L. T. Bacheler, J. L. Meek, M. J. Otto, M. M.
Rayner, Y. N. Wong, C.-H. Chang, P. C. Weber, D. A. Jackson, T. R. Sharpe, and S. Erickson-Viitanen,
Science 263, 380 (1994).
33. A. L. Swain, M. M. Miller, J. Green, D. L. Rich, J. Schneider, S. 8. H. Kent, and A. Wiodawer, Proc. Natl.
Acud. Sci. U . S . A. 87, 8805 (1990).
34. J. M. Blaney, G. M. Crippen, A. Dearing, and J. S. Dixon, Indiana University-Quantum Chemistry
Program Exchange (QCPE # 590), Bloomington, IN.
35. 0. F. Guner, D. W. Hughes, and L. M. Dumont, 1. Chem. I f . Comput. Sci. 31, 408 (1991).
36. S. Thaisrivongs, P. K. Tomich, K. D. Watenpaugh, K.-T. Chong, W. J. Howe, C.-P. Yang, J. W. Strohbach,
S. R. Turner, J. P. McGrath, M. J. Bohanon, J. C. Lynn, A. M. Mulichak, P. A. Spinelli, R. R. Hinshaw, P. J.
Pagano, J. B. Moon, M. J. Ruwart, K. F. Wilkinson, B. D. Rush, G. L. Zipp, R. J. Dalga, F. J. Schwende,
G. M. Howard, G. E. Padbury, L. N. Toth, Z . Zhao, K. A. Koeplinger, T. J. Kakuk, S. L. Cole, R. M. Zaya,
R. C. Piper, and P. Jeffrey, 1. Med. Chem. 37, 3200 (1994).
37. E. A. Lunney, S. E. Hagen, J. M. Domagala, C. Humblet, J. Kosinski, B. D. Tait, J. S. Warmus, M. Wilson,
D. Ferguson, D. Hupe, P. J. Tummino, E. T.Baldwin, T. N. Bhat, B. Liu, and J. W. Erickson, 1. Med. Chem.
37, 2664 (1994).
38. D. S. Goodsell and A. J. Olson, Proteins: Struct., F u n d . , Genet. 8, 195 (1990).
39. P. J. Goodford, /. Med. Chem. 28, 849 (1985).
40. J. V. N. V. Prasad, K. S. Para, E. A. Lunney, D. F. Ortwine, J. 8 . Dunbar, D. Ferguson, P. J. Tummino, D.
Hupe, B. D. Tait, J. M. Domagala, C. Humblet, T. N. Bhat, 8. Liu, D. M. A. Guerin, E. T. Baldwin, J. W.
Erickson, and T. K. Sawyer, 1. A m . Chem. Soc. 116, 6989 (1994).
41. J. P. Vacca, J. P. Guare, S. J. deSolms, W. M. Sanders, E. A. Giuliani, S. D. Young, P. L. Darke, J. Zugay,
I. S. Sigal, W. A. Schleif, J. C. Quintero, E. A . Emini, P. S. Anderson, and J. R. Huff, 1. Med. Chem. 34,1225
(1991).
42. T. A. Lyle, C. M. Wiscount, J. P. Guare, W. J. Thompson, P. S. Anderson, P. L. Darke, J. A. Zugay, E. A.
Emini, W. A. Schleif, J. C. Quintero, R. A. F. Dixon, I. S. Sigal, and J. R. Huff, 1. Med. Chem. 34, 1228
(1991).
43. B. D. Dursey, R. B. Levin, S. L. McDaniel, J. P. Vacca, J. P. Guare, P. L. Darke, J. A. Zugay, E. A. Emini,
W. A. Schleif, J. C. Quinterno, J. H. Lin, I.-W. Chen, M. K. Holloway, P. M. D. Fitzgerald, M. G. Axel,
D. Ostovic, P. S. Anderson, and J. R. Huff, 1. Med. Chem. 37, 3443 (1994).
44. D. C. Humber, N. Cammack, J. A. V. Coates, K. N. Cobley, D. C. Orr, R. Storer, G. G. Weingarten, and
M. P. Weir, 1. Med. Chem. 35, 3080 (1992).
45. D. C. Humber, M. J. Bamford, R. C. Bethell, N. Cammack, K. Cobley, D. N. Evans, N. M. Gray, M. M.
Hann, D. C. Orr, J. Saunders, B. E. V. Shenoy, R. Storer, G. G. Weingarten, and P. G. Wyatt, 1. Med. Chem.
36, 3120 (1993).
46. D. S. Holmes, R. C. Bethell, N. Cammack, I. R. Clemens, J. Kitchin, P. McMeekin, C. L. Mo, D. C. Orr,
8. Patel, I. L. Paternoster, and R. Storer, 1. Med. Chem. 36, 3129 (1993).
47. A. Wonacott, R. Cooke, F. R. Hayes, M. M. Hann, H. Jhoti, P. McMeekin, A. Mistry, P. Murray-Rust,
0. M. P. Singh, and M. P. Weir, 1. Med. Chem. 36, 3113 (1993).
48. H. Jhoti, 0. M. P. Singh, M. P. Weir, R. Cooke, P. Murray-Rust, and A. Wonacott, Biochem. 33, 8417
(1994).
49. J. Kitchin, R. C. Bethell, N. Cammack, S. Dolan, D. N. Evans, S. Holman, D. S. Holmes, P. McMeekin,
C. L. Mo, N. Nieland, D. C. Orr, J. Saunders, B. E. V. Shenoy, 1. D. Starkey, and R. Storer, 1. Med. Chem.
37, 3707 (1994).
50. D. H. Rich, J. Green, M. V. Toth, G. R. Marshall, and S. B. H. Kent, 1. Med. Chem. 33, 1285 (1990).
STRUCTURE-BASED DRUG DESIGN 49

51. A. Krohn, S. Redshaw, J. C. Ritchie, B. J. Graves, and M. H. Hatada, J. Med. Chem. 34, 3340 (1991).
52. M. D. Varney, K. Appelt, V. Kalish, M. R. Reddy, J. Tatlock, C. L. Palmer, W. H. Romines, B.-W. Wu, and
L. Musick, J. Med. Chem. 37, 2274 (1994).
53. M. R. Reddy, M. D. Varney, V. Kalish, V. N. Viswanadhan, and K. Appelt, j . Med. Chem. 37, 1145 (1994).
54. D. Schirlin, S. Baltzer, V. Van Dorsselaer, F. Weber, C. Weill, J. M. Altenburger, B. Neises, G. Flynn, J. M.
Remy, and C. Tarnus, Biorg. Med. Chem. Lett. 3, 253 (1993).
55. D. Schirlin, V. Van Dorsselaer, C. Tamus, A. S. Tyms, S. Baltzer, F. Weber, J. M. Remy, T. Brennan, R. Farr,
and D. Janowick, Biorg. Med. Chem. Lett. 4, 241 (1994).
56. B. L. Pologar, R. A. Farr, D. Friedrich, C. Tamus, E. W. Huber, R. J. Cregge, and D. Schirlin, J. Med. Chem.
37, 3684 (1994).
57. M. Miller, J. Schneider, B. K. Sathyanarayana, M. V. Toth, G. R. Marshall, L. Clawson, L. Selk, S. B. H.
Kent, and A. Wlodawer, Science 246, 1149 (1989).
58. Z. Szewczuk, K. L. Rebholz, and D. H. Rich, Int. 1. Peptide Protein Res. 40, 233 (1992).
59. R. A. Wiley and D. H. Rich, Med. Res. Reviews 13, 327 (1993).
60. B. M. Brenner, B. J. Ballermann, M. E. Gunning, and M. L. Zeidel, Physiol. Rev. 70, 665 (1990).
61. P. Needleman, E. H. Blaine, J. E. Greenwald, M. L. Michener, and C. 8. Saper, Ann. Rev. Pharmacol.
Toxicol. 29, 23 (1989).
62. R. E. Lang, T. Unger, and D. Ganten, I. Hypertension 5, 255 (1987).
63. A. E. G. Raine, J. G. Firth, and J. G. G. Ledinham, Clin. Sci. 76, 1 (1989).
64. J. P. Bussien, J. Biollaz, B. Waeber, J. Nussberger, G. A. Turini, H. R. Brunner, F. Brunner-Feber, H. J.
Gomez, and E. S. Otterbein, 1. Curdiovasc. Pharmacol. 8, 216 (1986).
65. T. G. Yandle, A. M. Richards, M. G. Nicholls, R. Cuneo, E. A. Espiner, and J. H. Livesey, Life Sci. 38,1822 (1986).
66. J. L. Sonnenberg, Y. K. Sakane, A. Y. Jeng, J. A. Koehn, J. A. Ansell, L. P. Wennogle, and R. D. Ghai,
Peptides 9, 173 (1988).
67. A. J. Kenny and S. L. Stephenson, FEBS Lett. 232, 1 (1988).
68. M. Pozsgay, C. Michaud, M. Liebman, and M. Orlowski, Biochemistry 25, 1292 (1986).
69. L. 8 . Hersh and K. Morihara, 1. Bid. Chem. 261, 6433 (1986).
70. J. C. Danilewicz, P. L. Barclay, I. T. garnish, D. Brown, S. F. Campbell, K. James, G. M. R. Samuels, N. K.
Terrett, and M. J. Wythes, Biochem. Biophys. Res. Commun. 164, 58 (1989).
71. D. R. Holland, P. L. Barclay, J. C. Danilewicz, B. W. Matthews, and K. James, Biochemistry 33, 51 (1994).
72. L. J. MacPherson, E. K. Bayburt, M. P. Capparelli, R. S. Bohacek, F. H. Clarke, R. D. Ghai, Y. Sakane, C. J.
Berry, J. V. Peppard, and A. J. Trapani, j . Med. Chem. 36, 3821 (1993).
73. P. A. Bartlett and C. K. Marlowe, Science 235, 569 (1987).
74. H. M. Holden, D. E. Tronrud, A. F. Monzingo, L. H. Weaver, and B. W. Matthews, Biochemistry 26,8542
(1987).
75. W. C. Guida, R. S. Bohacek, and M. D. Erion, 1. Comp. Chem. 13, 214 (1992).
76. R. S. Bohacek, J. Priestle, and M. Gruetter, unpublished results.
77. L. H. Opie, Angiotensin Converting Enzyme Inhibitors, Wiley-Liss, New York, 1992.
78. A. M. Richards, I. G. Crozier, T. Kosoglou, M. Rallings, E. A. Espiner, M. G. Nicholls, T. G. Yandle,
H. Ikram, and C. Frampton, Hypertension 22, 119 (1993).
79. K. B. Margulies, M. A. Perrella, L. J. McKinley, and J. C. Burnett, 1. Clin. Invest. 88, 1636 (1991).
80. I. Pham, W. Gonzales, A.-I. K. El Amrani, M.-C. Fourni&Zaluski, M. Philippe, I. Laboulandine, B. P.
Roques, and J.-B. Michel, 1. Pharmacol. Exp. Ther. 265, 1339 (1993).
81. A. A. Seymour, M. M. Asaad, V. M. Lanoce, K. M. Langenbacher, S. A. Fennell, and W. L. Rogers, 1.
Phurmacol. Exp. Ther. 266, 872 (1993).
82. C. Krulan, R. D. Ghai, R. W. Lappe, and R. L. Webb, FASEB J. 7, A247 (1993).
83. A. A. Patchett, and E. H. Cordes, in Adv. Enzymol. A. Meister, Ed., John Wiley & Sons, New York, 1985.
84. R. S. Bohacek, S. DeLombaert, C. McMartin, J. Priestle, and M. Gruetter, unpublished results; manu-
script in preparation.
85. G. Ksander, R. S. Bohacek, R. de Jesus, A. Yuan, Y. Sakane, C. Berry, R. Ghai, and A. J. Trapani,
unpublished results, manuscript in preparation.
86. C. McMartin and R. Bohacek, unpublished results.
87. D. W. Cushman, H. S. Cheung, E. F. Sabo, and M. A. Ondetti, Biochemistry 16, 5484 (1977).
88. G. A. Flynn, D. W. Beight, S. Mehdi, J. R. Koelh, E. L. Giroux, J. F. French, P. W. Hake, and R. C. Dage, 1.
Med. Chem. 36, 2420 (1993).
89. N. G. Delaney, J. C. Banish, R. Neubeck, S. Natarajan, M. Cohen, G. C. Rovnyak, G. Huber,
N. Murugesan, R. Girotra, E. Sieber-McMaster, J. A. Robl, M. M. Asaad, H. S. Cheung, J. E. Bird,
T. Waldron, and E. W. Petrillo, Bioorg. Med. Chem. Lett. 4, 1783 (1994).
50 BOHACEK, MCMARTIN, AND GUIDA

90. J. A, Robl, L. M. Simpkins, 1. Stevenson, C. Sun, N. Murugesan, J. C. Barrish, M. M. Asaad, J. E. Bird,


T. R. Schaeffer, N. C. Trippodo, E. W. Petrillo, and D. C. Karanewsky, Bioorg. Med. Chem. Lett. 4, 1789
(1994).
91. J. A. Robl, L. M. Simpkins, R. Sulsky, E. Sieber-McMaster, J. Stevenson, Y. F. Kelly, C. Sun, R. J. Misra,
D. E. Ryono, M. M. Asaad, J. E. Bird, N. C. Trippodo, and D. C. Karanewsky, Bioorg. Med. Chem. Lett. 4,
1795 (1994).
92. W. C. Guida, R. D. Elliott, H. J. Thomas, J. A. Secrist, Y. S. Babu, C. E. Bugg, M. D. Erion, S. E. Ealick,
and J. A. Montgomery, J. Med. Chem. 37, 1109 (1994).
93. M. D. Erion, S. Niwas, J. D. Rose, S. Ananthan, M. Allen, J. A. Secrist, Y. S. Babu, C. E. Bugg, W. C.
Guida, S. E. Ealick, and J. A. Montgomery, J. Med. Chem. 36, 3771 (1993).
94. J. A. Montgomery, S. Niwas, 1. D. Rose, J. A. Secrist, Y. S. Babu, C. E. Bugg, M. D. Erion, W. C. Guida,
and S. E. Ealick, J. Med. Chem. 36, 55 (1993).
95. J. A. Secrist, S. Niwas, J. D. Rose, Y. S. Babu, C. E. Bugg, M. D. Erion, W. C. Guida, S. E. Ealick, and J. A.
Montgomery, 1. Med. Chem. 36, 1847 (1993).
96. S. E. Ealick, Y. S. Babu, C. E. Bugg, M. D. Erion, W. C . Guida, J. A. Montgomery, and J. A. Secrist, Proc.
Nutl. Acud. Sci. U . S . A. 88, 11540 (1991).
97. J. A. Montgomery, Med. Res. Reviews 13, 209 (1993).
98. F. Mohamadi, N. G. J. Richards, W. C. Guida, R. Liskamp, M. Lipton, C. Caufield, G. Chang, T.
Hendrickson, and W. C. Still, I. Comp. Chem. 11, 440 (1990).
99. W. C. Guida, and 1. Kolossvary, unpublished observations.
100. M. von Itzstein, W.-Y. Wu, G. B. Kok, M. S. Pegg, 1. C. Dyason, B. Jin, T. V. Phan, M. L. Smythe, H. F.
White, S. W. Oliver, P. M. Colman, J. N. Varghese, D. M. Ryan, J. M. Woods, R. C. Bethell, V. J. Hotham,
J. M. Cameron, and C. R. Penn, Nature 363, 418 (1993).
101. J. N. Varghese, J. McKimm-Breschkin, J. 8. Caldwell, A. A. Kortt, and P. M. Colman, Proteins: Struct.,
Funct., Genet. 14, 327 (1992).
102. A. K. J. Chong, M. S. Pegg, N. R. Taylor, and M. von Itzstein, Eur. I. Biochem. 207, 335 (1992).
103. J. S. Finer-Moore, W. R. Montfort, and R. M. Stroud, Biochemistry 29, 6977 (1990).
104. D. A. Matthews, K. Appelt, S. J. Oatley, and N. H. Xuong, 1. Mol. Biol. 214, 923 (1990).
105. D. A. Matthews, J. E. Villafranca, C. A. Janson, W. W. Smith, K. Welsh, and S. Freer, J. Mol. Bid. 214,937
(1990).
106. W. R. Montfort, K. M. Perry, E. B. Fauman, J. S. Finer-Moore, G. F. Maley, L. Hardy, F. Maley, and R. M.
Stroud, Biochemistry 29, 6964 (1990).
107. K. Appelt, R. J. Bacquet, C. A. Bartlett, C. L. J. Booth, S. Freer, M. A. M. Fuhry, M. R. Gehring, S. M.
Herrmann, E. F. Howland, C. A. Janson, T. R. Jones, C.-C. Kan, V. Kathardekar, K. K. Lewis, G. P.
Marzoni, D. A. Matthews, C. Mohr, E. W. Moomaw, C. A. Morse, S. J. Oatley, R. C. Ogden, M. R. Reddy,
S. H. Reich, W. S. Schoettlin, W. W. Smith, M. D. Varney, 1. E. Villafranca, R. W. Ward, S. Webber, S. E.
Webber, K. M. Welsh, and J. White, J. Med. Chem. 34, 1925 (1991).
108. M. D. Varney, G. P. Marzoni, C. L. Palmer, J. G. Deal, S. Webber, K. M. Welsh, R. J. Bacquet, C. A.
Bartlett, C. A. Morse, C. L. J. Booth, S. M. Herrmann, E. F. Howland, R. W. Ward, and J. White, j . Med.
Chem. 35, 663 (1992).
109. 8. K. Shoichet, R. M. Stroud, D. V. Santi, I. D. Kuntz, and K. M. Perry, Science 259, 1445 (1993).
110. I. D. Kuntz, J. M. Blaney, S. J. Oatley, R. Langridge, and T. E. Ferrin, J. Mol. Biol. 161, 269 (1982).
111. R. L. DesJarlais, R. P. Sheridan, G. L. Seibel, J. S. Dixon, I. D. Kuntz, and R. Venkataraghavan, J. Med.
Chem. 31, 722 (1988).
112. B. K. Shoichet, D. L. Bodian, and I. D. Kuntz, J. Comp. Chem. 13, 380 (1992).
113. I. D. Kuntz, E. C. Meng, and B. K. Shoichet, Acc. Chem. Res. 27, 117 (1994).
114. V. Tschinke and N. C. Cohen, J. Med. Chem. 36, 3863 (1993).
115. M. K. Holloway, J. M. Wai, T. A. Halgren, P. M. D. Fitzgerald, J. P. Vacca, B. D. Dorsey, R. B. Levin, W. J.
Thompson, J. Chen, S. J. deSolms, N. Gaffin, A. K. Ghosh, E. A. Giuliani, S. L. Graham, 1. P. Guare,
R. W. Hungate, T. A. Lyle, W. M. Sanders, T. J. Tucker, M. Wiggins, C. M. Wiscount, 0. W. Woltersdorf,
S. D. Young, P. L. Darke, and J. A. Zugay, J. Med. Chem. 38, 305 (1995).
116. P. A. Kollman, Chem. Rev. 93, 2395 (1993).
117. P. A. Kollman, Current Opinion in Structural Biology 4, 240 (1994).
118. F. Guarnieri and W. C. Still, J. Comp. Chem. 15, 1302 (1994).
119. J. Aqvist, C. Medina, and J.-E. Samuelsson, Protein Engineering 7, 385 (1994).
120. H.-J. Boehm, J. Computer-Aided Mol. Design 8, 243 (1994).

You might also like