REVIEWS

Structure-based discovery of
antibacterial drugs
Katie J. Simmons*, Ian Chopra* & Colin W. G. Fishwick*
Abstract | The modern era of antibacterial chemotherapy began in the 1930s, and the next
four decades saw the discovery of almost all the major classes of antibacterial agents that
are currently in use. However, bacterial resistance to many of these drugs is becoming an
increasing problem. As such, the discovery of drugs with novel modes of action will be vital
to meet the threats created by the emergence of resistance. Success in discovering
inhibitors using high-throughput screening of chemical libraries is rare. In this Review
we explore the exciting opportunities for antibacterial-drug discovery arising from
structure-based drug design.

Pharmacophore The treatment of bacterial infections through the The determination of complete bacterial genome
A set of structural features in a administration of chemotherapeutic agents, which sequences and the parallel development of other tech-
molecule that are recognized began in the 1930s, was one of the most profound med- niques such as proteomics inspired a new genomics-
at a receptor site and are ical advances of the twentieth century. The origins of based approach to drug discovery from the mid 1990s9.
responsible for the biological
activity of the molecule.
almost all of the antibacterial drugs in use today lie in By March 2009, crystal structure data were available for
empirical screening programmes to identify inhibitors more than 600 individual proteins derived from bacte-
on the basis of their ability to prevent bacterial growth ria (see the Protein Data Bank and TargetDB databases).
and can be traced to the so-called golden period of Many companies sought to identify novel antibacterial
antibacterial-drug discovery between the 1940s and agents from high-throughput screening (HTS) cam-
1970s1,2. Subsequent development of these drugs, or paigns using purified enzyme targets that were vali-
agents derived from them, has produced an impres- dated by genomic approaches as being essential for the
sive global reduction in the burden of disease caused by organism. It was thought that the era of exploiting novel
bacterial infection. natural products or continually modifying existing com-
Unfortunately, the widespread emergence of resist- pounds into improved analogues had passed and that
ance to antibiotics in pathogenic bacteria over the past novel agents directed against previously unexploited tar-
30 years is now a serious threat to global public health gets would be identified10. The major investment dedi-
and could undermine the major advances achieved cated to the genomic approach for antibacterial-drug
in the treatment of infection39. Paradoxically, as the discovery reflected the optimism about its likely success.
problems accompanying the emergence of resistance to Small-molecule screening approaches had successfully
existing drugs increase, there has been a decline in the identified promising lead compounds in other thera-
discovery and development of new antibacterials. The peutic areas, such as treatments for cancer, diabetes and
reasons for this situation are complex 1,35 but, in part, asthma11, and consequently the compound collections
reflect technical difficulties associated with the identifi- used for HTS of bacterial targets were largely composed
cation of suitable novel compounds for development as of small synthetic molecules12. However, the success
*Antimicrobial Research candidate antibacterials. rate of the concerted genomic and HTS initiatives has
Centre, University of Leeds.
In the past, many successful antibacterial agents were been extremely low, and new strategies are required in

School of Chemistry,
University of Leeds. sourced from empirical screening of natural products order to develop the next generation of antibiotics6.

Institute of Molecular and or of synthetic chemical libraries6,7. However, in recent The time may now be right to consider fresh
Cellular Biology, University of years empirical screening has not returned suitable approaches to antibacterial-drug discovery. The power
Leeds, Leeds, LS2 9JT, UK. pharmacophores for development. Indeed, in the past of structure-based drug discovery (SBDD) (FIG. 1) has
Correspondence to C.W.G.F.
e-mail:
40 years only two new structural types, daptomycin and been demonstrated most clearly by the discovery of new
[email protected] linezolid, have been introduced to the clinic following their therapeutics for HIV/AIDS, a case in which structural
doi:10.1038/nrmicro2349 discovery using empirical screening methods8. knowledge of the HIV protease enabled the successful

nATUre reVIewS | Microbiology VolUMe 8 | jUly 2010 | 501

2010 Macmillan Publishers Limited. All rights reserved

REVIEWS

a b during ligand binding. This reflects the fact that, unlike

300,000 compounds Define protein the structure of the protein in the crystal, in solu-
structure tion the protein is flexible and can undergo considerable
vHTS screening
conformational changes on ligand binding.
Identify region Until recently, most SBDD programs relied on a sin-
Compound Hit identification of binding
selection gle high-resolution protein crystal structure. However,
as this is only a snapshot of the protein frozen in one
10 compounds Dock atoms or
fragments to binding sites form, it can cause problems when designing ligands,
Enzyme assay as the biologically active form of the protein may be
conformationally different to the crystal form. Many
Active compounds Connect fragments studies emphasize the importance of allowing for pro-
tein and ligand flexibility when performing SBDD2527.
SAR studies and Hit optimization
analogue synthesis Evaluate complete Unfortunately, modelling of molecular flexibility, espe-
skeletons cially for the protein, drastically increases the compu-
ter time required, often making it prohibitive in terms
Figure 1 | Protocols for high-throughput screening docking and
Nature de novo
Reviews design.
| Microbiology
of time and technical capacity. Programs and techniques
a | A typical selection process for a compound purchased using virtual high-throughput
screening (vHTS). b | A typical procedure for de novo design using programs such as such as SlIDe28, Flexe29,30 and MCSAPCr (multicon-
SPROUT. SAR, structureactivity relationship. formation simulated annealingpseudo-crystallographic
refinement)31 can be used to model protein flexibility,
and most programs can now model ligand flexibility as
design and development of five protease inhibitors that well. A good example of the large degree of conforma-
are now commercially available drugs1315. Further suc- tional flexibility seen during substrate binding is pro-
cesses that owe their origins to SBDD are drugs such as vided by MurA, which is involved in the synthesis of
nelfinavir (Viracept; ViiV Healthcare )16 and amprenavir bacterial peptidoglycan. High-resolution crystal struc-
(Agenerase; GlaxoSmithKline) (compounds 14 and 15; tures of MurA have been solved in the apo form and
see Supplementary information S1 (figure))17 for AIDS, also with bound substrate and substrate analogues. The
zanamivir (relenza; GlaxoSmithKline)18 for influenza, inhibitor T6361 binds to MurA and blocks the confor-
the cyclooxygenase 2 (CoX2; also known as PTGS2) mational change that is normally induced by substrate
inhibitors celecoxib (Celebrex; Pfizer)19 and rofecoxib binding. The crystal structure of the MurAT6361
(Vioxx; Merck although this was later withdrawn complex shows that the protein adopts a substantially
owing to safety concerns) (compounds 16 and 17; see different conformation to that seen in the structure of
Supplementary information S1 (figure))20. the MurAsubstrate crystal3234. This example demon-
Although the potential of SBDD in antibacterial-drug strates the challenge of predicting the conformation
discovery has yet to be fully realized, there are many of the ligand-binding pocket, even if knowledge of the
validated molecular targets already available (FIG. 2; dynamic motion of the protein is available. This situa-
TABLE 1). The growing number of validated targets for tion is not unusual, creating a challenge for both virtual
Algorithm which structural information has been obtained makes HTS and de novo SBDD. Several methods are emerg-
A finite sequence of this approach increasingly attractive21. In this review, we ing to overcome these difficulties but have yet to be
instructions; an explicit, describe the use of SBDD for the development of novel completely evaluated35.
step-by-step procedure for antibacterial agents. Although the Protein Data Bank is rapidly expanding,
solving a problem, often used
for calculation and data
there is a substantial gap between the number of struc-
processing. The principles underpinning SBDD tures available in the database and the number of known
The starting point for all structure-based design work, gene sequences. TABLE 2 gives a cross section of the
Scoring function whether ligand- or protein-based, is the choice of a suit- number of individual protein targets for which structural
A fast, approximate
able target. An antimicrobial-drug target should be essen- data is available versus the number of genomic orFs for
mathematical method that is
used to predict the strength of tial, have a unique function in the pathogen and exhibit an eight prominent bacterial pathogens.
the non-covalent interaction activity that can be altered by a small molecule. Tools are For cases in which the crystal structure of the par-
(also referred to as the binding emerging to prioritize targets on the basis of their predicted ticular bacterial protein is not available, construction
affinity) between two molecules suitability for SBDD. A recent study correlated the charac- of a homology model is often possible, provided that a
following docking. Scoring
functions are normally
teristics of protein-binding pockets with the frequency of crystal structure is available for a protein with substantial
parameterized (or trained) the binding ligands identified by nuclear magnetic reso- sequence similarity to the protein of interest. Approaches
against a data set consisting of nance (nMr)-based screening. This has lead to the crea- to producing these models vary from purely ab initio
experimentally determined tion of an algorithm that can predict the suitability of the methods36 based on only physical and chemical principles
binding affinities between
binding pocket on the basis of the characteristics that can to models based on sequence and structural information.
molecular species that are
similar to the species that the be identified from high-resolution protein structures, Advanced homology models use experimentally deter-
user wishes to predict. For such as the rigidity of the binding site and its hydrophobic mined structures to predict the conformation of another
predictions of proteinligand character. Such algorithms will allow future researchers protein that has a similar amino acid sequence. Homology
affinities, the tertiary structure to focus their efforts in drug discovery on proteins that modelling involves four steps: fold assignment, sequence
of the protein, the active
conformation of the ligand
are more likely to yield high-affinity ligands2224. alignment, model building and model refinement. Several
and the binding mode must when choosing a target enzyme, it is important to computer packages are available to perform this process
be known. be aware of the conformational variations of a protein automatically; for example, the SwISS-MoDel software

502 | jUly 2010 | VolUMe 8 www.nature.com/reviews/micro

2010 Macmillan Publishers Limited. All rights reserved

 REVIEWS

HO Br
NH2 Br
Cl O
O Br Br
OH
H
HN N
N Br
O H Br CO2 H
H N N HN N O
OH OH CH3
O
CH3 HN
OH O
HO O OH
OH
1 SB-284485 (PS) 2 DdlB inhibitor (CW) 3 MetRS inhibitor (PS) 4 MurD inhibitor (CW)
IC50 = 4 nM49 Ki = 218 M54 IC50 = 8 nM57 IC50 = 10 M63

OH NO2
H SO3H
N O
Cl N N O
O
Cl H3C OH
S N SH O S CO2H N
H2N
O O N NH2 HO
Cl Cl CH3
HO N O O
Cl
5 MurF inhibitor (CW) 6 Chorismate mutase inhibitor 7 AccD5 inhibitor 8 CTX-M -lactamase inhibitor
IC50 = 63 M63 Ki = 5.7 M64 Ki = 13.1 M65 Ki = 8 M66,67

H
N HO
N O
H3C
O
NH O
O S O S
NH O N N
O O O H
O O
2 O O
O NH
O NH3 + HO P
OH
O HO2C

9 DNA gyrase inhibitor (DR) 10 MurD inhibitor (CW) 11 DdlB inhibitor (CW) 12 VanA inhibitor (CW)
MNEC = 0.03 M69 IC50 = 0.7 M77 Ki = 12 M78 IC50 = 224 M79

Figure 2 | inhibitors designed using structure-based drug discovery. The structures of some of the key compounds
that have been designed using structure-based drug design techniques. Abbreviations givenNature in parentheses
Reviews | denote the
Microbiology
molecular target for the compounds: cell wall synthesis (CW), DNA replication (DR), protein synthesis (PS) or RNA
synthesis (RS). DdlB, d-alanined-alanine ligase; IC50, half-maximal inhibitory concentration; Ki, inhibition constant; MetRS,
methionyl-tRNA synthetase; MNEC, maximal non-effective concentration; VanA, vancomycin resistance protein A.

is a fully automated homology-modelling server avail- Methods for SBDD

able through the exPASy Proteomics Server 3740. CASP Three main methods are available to assist in the
(Critical Assessment of Techniques for Protein Structure identification of new putative ligands on the basis of
Prediction) is a competition involving protein structure structural information.
prediction that has taken place biannually since 1994. The In the first approach, termed substrate- and known
competition provides users of structure prediction serv- inhibitor-inspired design, the structures of substrates,
ers with an opportunity to assess the quality of the various known inhibitors or cofactors for the particular tar-
methods and servers41,42. get enzyme are modified to become inhibitors by
maximizing complementary interactions in the target
Table 1 | Targets and classes of antibacterial drugs site1315,43,44. In the second approach, databases con-
taining the structures of small molecules are docked
Drug target Drug classes
into a region of interest in silico and scored according
Cell wall synthesis -Lactams, bacitracin, cycloserine, fosfomycin and to their predicted interactions in the target site. Many
glycopeptides programs are available to perform such virtual HTS,
Cell membrane integrity Daptomycin and polymyxins all of which have different docking algorithms and dif-
Nucleotide biosynthesis Sulfonamides and trimethoprim ferent scoring functions. The third important approach
involves de novo design of inhibitor scaffolds. Fragments
DNA replication Quinolines, nitrofurans and nitroimidazoles
of molecules are positioned in chosen sites in the target
RNA synthesis Rifamycins protein and then linked in silico to give complete mol-
Protein synthesis Aminoglycosides, chloramphenicol, fusidic acid, ketolides, ecules. These molecules can then be scored and ranked
macrolides, oxazolidinones, streptogramins, tetracyclines for factors such as their predicted binding affinity,
and mupirocin their molecular complexity or their expected synthetic

nATUre reVIewS | Microbiology VolUMe 8 | jUly 2010 | 503

2010 Macmillan Publishers Limited. All rights reserved

REVIEWS

Table 2 | Structural genomics in bacterial pathogens

Species Total number Number of orFs Number of Number of solved
of orFs* cloned purified proteins structures in PDb
Escherichia coli 5,402 4,562 1,974 233
Pseudomonas aeruginosa 2,230 2,262 791 92
Haemophilus influenza 1,191 959 264 20
Staphylococcus aureus 2,043 1,762 465 35
Streptococcus pneumoniae 1,340 1,174 364 48
Enterococcus faecalis 2,363 2,106 580 63
Mycobacterium tuberculosis 2,315 1,094 341 112
Helicobacter pylori 841 537 173 14
PDB, Protein Data Bank. *Data taken from the genome database PEDANT (14 October 2009) . Data taken from TargetDB (14 October
2009) using the built-in query tools and requesting the information under each of the column headings for each species listed.

accessibility (FIG. 1). The resulting designed inhibitor prominent docking programs currently available for
scaffolds are then synthesized in the laboratory and sub- virtual screening are AutoDock, Glide (Schrdinger),
jected to biological evaluation. recent examples of these GolD (The Cambridge Crystallographic Data Centre,
three approaches in the context of antibacterial-drug Cambridge, UK), DoCK and eHiTS (SimBioSys Inc.,
discovery are discussed below. Toronto, Canada). A selection of the most successful
programs applied to the discovery of potential new
Substrate- and known inhibitor-inspired design antibacterials is discussed below (see also Supplementary
This approach uses the structural modification of known information S2 (table) for additional resources).
biologically active substrates and natural product-based
inhibitors45. For example, SB-219383 (compound 18; see AutoDock. AutoDock is a suite of automated docking
Supplementary information S1 (figure)) is a potent and tools designed to predict how small molecules, such as
specific inhibitor of bacterial tyrosyl-trnA synthetase substrates or drug candidates, bind to a receptor of known
(TyrrS) and was originally identified from the fer- three-dimensional structure. AutoDock consists of two
mentation broth of Micromonospora sp. nCIMB 40684 main programs: autodock performs the docking of the
(REFS 4648). To simplify the chemical structure of the ligand to a set of grids describing the target protein, which
molecule, the bicyclic ring was cleaved to yield a com- are pre-calculated by autogrid. In addition, a graphical
pound that retained potent TyrrS inhibition, and the front-end tool, AutoDockTools, is available to set up, visu-
addition of a butyl ester group led to improved potency. alize and analyse the results of dockings performed using
Another simpler molecule, SB-284485 (compound 1; AutoDock.
FIG. 2) was also derived without losing inhibitory activ- The most recent version of the software, AutoDock 4.0,
ity 49, therefore providing an excellent template for uses the AMBer force field as well as a free-energy scoring
further structural modifications. function based on a linear-regression analysis and a diverse
set of proteinligand complexes with known inhibition
Virtual screening approaches constants.
The technique of molecular docking has been well AutoDock uses a genetic algorithm to generate a range
established for some time. However, recent advances in of docking poses that can be clustered according to their
Force field both hardware and software algorithms have made pos- energetic similarity. Several studies have shown that in
The functional form and
sible the rapid docking of very large collections of small docking calculations the most populated clusters of the
parameter sets used to
describe the potential energy molecules into the chosen molecular target. The speed docked-ligand conformation are better predictors of
of a system of particles. of some of these programs is such that up to 100,000 the native state than the lowest-energy cluster 5053.
molecular structures can be docked per day when using A recent example of the application of AutoDock
Free energy a cluster of parallel processors. However, as with all to antibacterial-drug discovery is the structure-based
The calculated difference
between the internal energy of
docking algorithms, the scoring function that is used virtual screening of the UK national Cancer Institute
a system and the product of its to assess the validity of specific docking poses is para- (nCI) diversity set of 2,000 compounds using a
absolute temperature and mount, and each program has its own unique scoring crystal structure of d-alanined-alanine ligase (DdlB)
entropy. function. These scoring functions will necessarily place from Escherichia coli, a key enzyme in peptidoglycan
different weightings on the various factors involved in biosynthesis. Docking results were obtained as a list of
Linear-regression analysis
Any approach to modelling the ligand binding. As none of these functions is considered compounds ranked according to their mean estimated
relationship between one or faultless, a consensus scoring approach is the best way binding affinity to the protein. This was determined by
more variables denoted Y and to identify potential lead molecules. Consensus scoring the calculated average free energy of binding for the most
one or more variables denoted uses several scoring functions to predict binding affinity. populated cluster of docked poses. Using this approach,
X, such that the model
depends linearly on the
If a compound is predicted to bind tightly to a chosen the top 130 compounds were tested in an in vitro assay
unknown parameters to be protein using several docking algorithms, this provides for inhibition of E. coli DdlB and several hits were
estimated from the data. higher confidence in the prediction. Among the most identified. Three of these hits have novel scaffolds; two

504 | jUly 2010 | VolUMe 8 www.nature.com/reviews/micro

2010 Macmillan Publishers Limited. All rights reserved

 REVIEWS

1 specific hydrophobic pockets. on the basis of this

Ligands are divided information, a four-point pharmacophore model was
into rigid fragments N constructed using Catalyst. This model was used as
and connecting
flexible chains HN O
a search query against the diverse compound collec-
2 tion from ChemDiv, which contains around 250,000
NH2 Rigid dock: compounds. A total of 461 molecules from the Catalyst
Each fragment is docked
independently everywhere search were identified as potential hits. These were
in the receptor docked into the S. aureus MetrS structure and re-
scored using the ligandFit scoring function56. of the
HN NH2 O 31 compounds that were subsequently selected for bio-
NH2
N logical testing, 22 displayed greater than 50% enzyme
H2N NH
N inhibition at a concentration of 100 M, which is a hit
HN NH
O rate of 71%. The most potent inhibitors, compounds
HN H2N 3 and 21, are shown in FIG. 2 and Supplementary infor-
NH2
O
N
mation S1 (figure), respectively. This method provides
H2N an efficient way of finding new leads from a known
N
HN NH O active compound and compares favourably to random
NH2 biological assaying of large compound libraries, which
give hit rates of between 0.1% and 0.5%, on average57.

eHiTS. This is a recently developed virtual HTS soft-

ware package that takes individual compounds from a
3 Reconnected-ligand pose
large library and calculates the optimal conformation
Pose match:
A fast graph-matching that each of these ligands can adopt in a targeted pro-
O
algorithm finds
N
tein cavity. The program then calculates a score for each
all matching solutions 4
structure according to the geometries of the ligand and
to reconstruct Local energy optimization:
the original molecules The structure is optimized the complementarities of surface points on the recep-
in the receptor tor and ligand. Complementary surface points receive a
positive score, whereas repulsive surface points receive
a penalty score. Additional terms are used in the final
5 scoring function to further reflect all factors involved
HN
Ranking: in binding, such as steric clashes, depth of the cavity,
NH2 Structures are ranked based
on scoring functions solvation, conformational-strain energy of the ligand,
intramolecular interactions in the ligand, and entropy
Figure 3 | The eHiTS docking strategy. In eHiTS, ligands are divided into rigid loss due to frozen rotatable bonds5860.
Nature Reviews
fragments and connecting flexible chains (step 1). These fragments | Microbiology
are docked
eHiTS takes a unique approach to the docking
individually into the binding site of the target receptor (step 2) and a fast
graph-matching algorithm finds all matching solutions to reconstruct the original
problem, having an innovative docking algorithm and
molecule (step 3), which can then be optimized (step 4), scored and ranked (step 5). a novel system for the scoring function. The approach
involves breaking ligands into rigid fragments and the
connecting flexible chains and then docking each rigid
of these (compounds 19 and 20; see Supplementary fragment to every possible place in the cavity (FIG. 3).
Toxicity information S1 (figure)) are competitive inhibitors, recent examples of the use of eHiTS include screens
The degree to which a competing with ATP, with inhibition constant (Ki) for inhibitors of the Mycobacterium tuberculosis shiki-
substance is able to damage values in the low micromolar range, whereas the third mate kinase (AroK), inhibitors of TyrrS and inhibitors
an exposed organism.
(compound 2; FIG. 2) inhibited the enzyme in a non- of MurD and MurF.
ADME competitive manner. Additionally, compounds 2 and 20 M. tuberculosis AroK catalyses the phosphorylation
Absorption, distribution, possessed some antimicrobial activity and are therefore of shikimate to shikimate-3-phosphate. An eHiTS
metabolism and excretion. An promising hits for further optimization54. screening protocol was used recently to identify poten-
acronym in pharmacokinetics tial inhibitors of AroK61. Screening compounds were
and pharmacology describing
the disposition of a
Catalyst. Discovery Studio (Accelrys) provides a suite of extracted from the FAF-Drugs (Free ADME/tox filter-
pharmaceutical compound software that has many applications, including SBDD, ing) collection to give 214,492 compounds following
in an organism. toxicity prediction, protein modelling and virtual HTS. filtering using ADMe and toxicity filters. FAF-Drugs is
The built-in virtual HTS screening tool, Catalyst, has been an online service that allows users to process their own
Log P
used in the development of novel antibacterial agents. compound collections through simple ADMe/tox-
The partition coefficient (P),
usually expressed as a log A high-throughput approach to crystallography filtering rules such as molecular mass, polar surface area,
value, is a measure of the identified several small-molecule inhibitors of the log P or number of rotatable bonds. From this collection
hydrophobic character of Staphylococcus aureus methionyl-trnA synthetase of over 200,000 compounds, docking in the AroK active
a substance, such that (MetrS), an enzyme that catalyses the highly specific site using eHiTS identified 644 small molecules that
P = concentration of the
substance in octanol /
attachment of methionine onto cognate methionine- were predicted to bind more tightly to AroK than the
concentration of the specific trnA55. The identified inhibitors all interact natural substrate and that are potential inhibitors with
substance in water. with an important amino acid, Asp51, and with two half-maximal inhibitory concentration (IC50) values in

nATUre reVIewS | Microbiology VolUMe 8 | jUly 2010 | 505

2010 Macmillan Publishers Limited. All rights reserved

REVIEWS

a O OH
scoring compounds for each enzyme were selected for
Lys48 biological evaluation63. For MurD, 4 of the top 50 com-
Thr36
pounds showed IC50 values of below 250 M (3 of these
are shown here as compounds 4, 13 and 21; FIGS 2,4a and
Ser415
Supplementary information S1 (figure)).
Thr321 O O OH
only one of the compounds selected from the
OH OH
MurF screen (compound 5; FIGS 2,4b) showed signifi-
cant inhibitory activity (that is, had residual activity
13
at 250 M below their IC50 values). The lower hit rate
Phe422 Thr16 for MurF might be attributed, in part, to the com-
pounds being assayed against E. coli MurF rather than
Leu15 Streptococcus pneumoniae MurF, the crystal structure of
which was used to perform the eHiTS screening runs.
b
OH UNITY. UnITy, a module in the SyByl (Tripos)
Cl N N
molecular modelling software, is a search and analysis
Phe31 Cl
system for exploring chemical and biological databases.
Ile139
S N SH It can be used for locating compounds that match a
Leu45
Cl Cl
pharmacophore or fit a receptor site. UnITys two-
Cl
dimensional searching capabilities offer exact, substruc-
ture and similarity searching. Conformationally flexible
Tyr135 5 three-dimensional searching rapidly finds molecules
Asn326 that can satisfy queries regardless of the conformation
Thr330 Leu367 stored in a database. Structural queries may be based
Leu360
on complete molecular structures, molecular fragments,
Asn328 Pro329
pharmacophore models or the receptor site.
Using UnITy, a virtual screen was carried out for
inhibitors of M. tuberberculosis chorismate mutase64,
an enzyme that catalyses the conversion of chorismate
Figure 4 | inhibitors designed using eHiTS. a | An inhibitor of MurD (compound 13),
to prephenate in the tyrosine and phenylalanine bio-
discovered using eHiTS, docked in the MurD active site. The Nature Reviews
inhibitor | Microbiology
binds in the same
pocket as the substrate (not shown) and makes key interactions with the labelled residues.
synthesis pathway. Starting from a known inhibitor
The MurD backbone is shown as green ribbons. b | An inhibitor of MurF (compound 5), of a homologous enzyme, a three-dimensional phar-
discovered using eHiTS, docked in the active site of the enzyme. The inhibitor interacts macophore search of a database of 15,000 compounds
with key labelled residues, and the protein backbone is shown as green ribbons. was performed. of the 15 highest-scoring molecules,
4 demonstrated inhibition in the enzyme assay. The most
potent molecule (compound 6) is shown in FIG. 2.
the low-micromolar range. The top 200 compounds were
examined in further detail using the graphical user inter- DOCK. DoCK is an open-source molecular-docking
face CheVi (SimBioSys) to determine key interactions software package that is frequently used in SBDD.
and shape complementarity with the binding pocket of Historically, the DoCK algorithm addressed rigid-body
the enzyme. docking using a geometric-matching algorithm to super-
In another study, a series of 340 potential inhibi- impose the ligand onto a negative image of the binding
tors of TyrrS were docked into the active sites of both pocket. Important features that improved the algorithms
the human and staphylococcal TyrrS proteins using ability to find the lowest-energy binding mode have been
eHiTS version 5.3. This study sought to find molecules added over recent years, including force field-based
that bind more strongly to the staphylococcal enzyme scoring, on-the-fly optimization, an improved matching
than to the human enzyme. Therefore, the scores for algorithm for rigid-body docking and an algorithm for
each ligand docked into the two separate active sites flexible-ligand docking. In one study using DoCK, over
were compared, and the ligands with the greatest differ- 4 million compounds were screened for predicted bind-
ence in predicted binding affinity for the two enzymes ing to the biotin- or propionyl CoA-binding pockets of
were then examined in further detail. This led to the AccD565, an essential M. tuberculosis acyl CoA carboxy-
identification of ten potential inhibitors with a stronger lase carboxyltransferase subunit. one of the nine top-
affinity for staphylococcal TyrrS than for the human scoring compounds identified by DoCK (compound 7;
protein. These compounds have not yet been screened FIG. 2) had an IC50 of 10 M against the enzyme. DoCK
against the bacterial enzyme, but this approach would has also been used for a fragment-based screening pro-
seem to offer potential for the design of selective gramme against the -lactamase CTX-M-9 (REFS 66,67).
antibacterial agents62. A fragment subset of 67,489 compounds were docked
On-the-fly optimization The Mur enzymes are essential for steps in peptido- into the active site of the enzyme, and compounds
A fast, dynamic procedure
used to make a system or
glycan biosynthesis in bacteria63. eHiTS was used to screen were selected from the top of the ranking list. From
design as effective or functional 1,990 compounds from the nCI diversity set using crystal the 69 fragments investigated (which included com-
as possible. structures of MurD and MurF as targets, and the 50 top- pound 8; FIG. 2), 10 exhibited IC50 values in the micromolar

506 | jUly 2010 | VolUMe 8 www.nature.com/reviews/micro

2010 Macmillan Publishers Limited. All rights reserved

 REVIEWS

DNA gyrase
range a hit rate of 14.5%. The same group also carried to experimentally determined binding constants of
An enzyme that unwinds DNA out a similar study looking for fragment inhibitors of proteinligand complexes70.
so that the DNA can duplicate. the -lactamase AmpC68. A library of 137,639 fragments In silico screening for potential inhibitors of DNA gyrase
from the ZInC database were docked, and 48 fragments was performed using lUDI and Catalyst to screen the
that were highly ranked were purchased, 23 of which had available-chemicals database (ACD) and part of the roche
Ki values ranging from 0.7 to 9.2 mM. The 48% hit rate compound inventory (a total of 350,000 compounds).
compared very favourably with lead-like docking and lUDI was used to dock small needle molecules into the
HTS against the same enzyme68. binding site of the gyrase. needle screening is a technique
that can be used to identify low-molecular-mass inhibitors
De novo design using protein structures (with a molecular mass of <300 daltons) that can pene-
De novo ligand design has been continually improving trate into deep and narrow channels and subpockets. The
since its invention in the early 1990s, reflecting both the lUDI searches led to the identification of around 150 weak
tremendous developments in computational power and inhibitors (including molecules 2325; see Supplementary
the continual improvement of efficient computational information S1 (figure)). X-ray crystallography verified
algorithms (FIG. 1). Most of the de novo design tools binding of the needles to the ATP-binding site. Following
follow a similar pipeline in terms of operation. The structureactivity relationship (SAr) studies to probe
most widely used de novo design programs are lUDI the structural requirements of the validated needles, a series
and SProUT (Keymodule ltd, leeds, UK) (discussed of indazole-based inhibitors was identified, with the most
below), SkelGen (De novo Pharmaceuticals ltd), Flux 69, potent compound (compound 9; FIG. 2) being ten times
GAnDI and BoMB (Cemcomco, Madison, Connecticut, as active as novobiocin, with a maximal non-effective
USA). Several of these de novo methods have been used concentration (MneC) of 0.03 M70.
to design novel antibacterial agents.
SPROUT. SProUT uses a fragment-joining technique
LUDI. lUDI, another module in Discovery Studio, con- to generate structures that fit the steric and electronic
structs possible new ligands for a given protein of known constraints of a specific receptor site or pharmacophore
three-dimensional structure. This approach is based on hypothesis. During structure generation, atoms or frag-
rules about energetically favourable non-bonded-contact ments of molecules are placed at each of the target sites
geometries between functional groups of the protein and are then linked to produce molecular skeletons.
and the ligand; these rules are derived from a statistical Molecular fragments are represented by templates, in
analysis of the crystal packing of organic molecules. which atoms are labelled purely by their hybridization
Small fragments are docked into the proteins binding state and are represented by vertices, and bonds are
site in such a way that hydrogen bonds and ionic interac- labelled as single, double, triple or aromatic. Several
tions can be formed with the protein and hydrophobic actual molecular fragments can be produced from each
pockets are filled with lipophilic groups derived from template by replacing the vertices with any element that
the ligands. The program can then append further frag- can adopt the appropriate hybridization state.
ments onto a previously positioned molecular core. It is The structure generation phase joins templates
also possible to link several fragments together, through together to produce skeletons. each skeleton represents
bridge fragments, to form a complete molecule. All puta- several molecules, because each component template can
tive ligands retrieved or constructed by lUDI are then represent several molecular fragments. Skeleton genera-
scored using a simple scoring function that was fitted tion begins by selecting a template and positioning it at a
target site so that it satisfies the steric requirements asso-
ciated with that particular site. new molecular templates
can then be added to further target sites, chosen by the
user. These docked molecular templates are then joined
using spacer templates, again under the full control of
Arg255 the user. once skeletal generation is complete, the result-
Mg2+ O O ing structures can be clustered and sorted using a range
O
of parameters specified by the user. These include overall
O
molecular properties such as molecular complexity and
Gly15 NH3+
estimated binding affinity, as well as more detailed filter-
11 ing options such as the presence or absence of certain
molecular features. In this way, unwanted structures can
be identified and discarded, leaving only structures that
Glu15 meet the requirements of the user 7177.
Tyr216 SProUT was used to design a series of novel mac-
rocyclic inhibitors of the bacterial cell wall biosynthesis
enzyme MurD78. SProUT revealed that the binding
cavity contains a hydrophobic pocket that is not used
Figure 5 | SProUT-designed inhibitor modelled in d-alanined-alanine ligase. The by the natural substrate. Simplification of the MurD sub-
inhibitor (compound 11) interacts with key labelled residues Nature Reviews | Microbiology
in the d-alanined-alanine strate generated models of possible macrocyclic inhibi-
ligase (DdlB) of Escherichia coli, the backbone of which is in green. tors that omit much of the sugar portion of the natural

nATUre reVIewS | Microbiology VolUMe 8 | jUly 2010 | 507

2010 Macmillan Publishers Limited. All rights reserved

REVIEWS

design is now commonplace using this approach, and

such inhibitors provide an excellent starting point for
further optimization.
Ile243 His244
H3C O
SPROUT-LeadOpt. SProUT-leadopt (Keymodule) is
Phe241 specifically designed to aid in the structural optimization
O process following the discovery of a lead molecule. Unlike
S SProUT, SProUT-leadopt is not a de novo design tool;
ADP
HN O
rather, it produces structures that are similar to known
OH
O P O
lead molecules and that have improved predicted bind-
OH
ing affinities. Core extension allows the user to explore
Arg317 Gly311 N different regions of a receptors binding site by adding
monomers to a core molecule. SProUT-leadopt uses a
O set of synthetic-chemistry rules to identify reactive func-
12 tional groups on a core molecule. It then systematically
generates extended structures in a combinatorial fashion
by carrying out virtual synthetic chemistry using common
synthetic reactions and a database of monomer structures.
Figure 6 | inhibitor of vancomycin resistance protein A, designedNature
using SProUT.
Reviews | Microbiology Monomer replacement uses retrosynthetic fragmenta-
The inhibitor (compound 12) binds to vancomycin resistance protein A (VanA) (green
tion to identify monomers that are present in one or
backbone) through the same pocket as the phosphate substrate. Key interactions with
labelled VanA residues are shown.
more bound ligands. new molecular structures are then
created by replacing these monomers with structurally
related molecules taken from a monomer library. The
substrate. A series of macrocyclic molecules based on power of this approach was recently shown by the devel-
these SProUT-designed molecular skeletons were syn- opment of inhibitors of dihydroorotate dehydrogenase82.
thesized and evaluated as inhibitors of E. coli MurD. Although the enzyme is not a bacterial target protein, this
It was found that compound 10 (FIG. 2) exhibited the work demonstrates the power of SProUT-leadopt in the
highest affinity for MurD. development of new lead scaffolds.
The SProUT software suite was also used to design
a novel cyclopropyl-based inhibitor of the bacte- Future perspectives
rial enzyme DdlB79. enzymological evaluation of the Undoubtedly, structure-based approaches for the iden-
designed inhibitor (compound 11; FIG. 2) showed that tification of new inhibitors of both classical and novel
this molecule (as a diastereomeric mixture) inhib- bacterial target proteins will increase in future years.
its the activity of DdlB with an apparent Ki value of However, it should also be remembered that improved
12.5 ( 0.l) M (FIGS 2,5). understanding of the molecular basis by which existing
The Van enzymes are responsible for resistance to van- antibacterial agents act may provide insights for new
comycin in S. aureus and Enterococcus spp. SProUT was SBDD approaches. For example, the structural basis of
used in conjunction with the X-ray crystal structure of the the interaction between the bacterial ribosome and sev-
Enterococcus faecium d-alanined-lactate ligase (VanA) eral established antibiotics, such as tetracycline, strep-
to devise new VanA inhibitors based on a hydroxyethyl- tomycin and chloramphenicol, has been elucidated in
amine template that was designed to mimic the tetrahe- recent years8385.
dral reaction intermediates80. The most active compound In addition to ribosomeantibiotic structures, several
(compound 12; FIG. 2) had an IC50 of 224 M against high-resolution crystal structures have been obtained
VanA. owing to the similar topology of the active sites that show the binding of the antibiotics rifampicin
of VanA and DdlB, the set of generated compounds was and myxopyronin to bacterial rnA polymerase86,87.
also tested against E. coli DdlB. Compound 12 exhibited These structures are helpful not only for determining
an IC50 value of 110 M against DdlB (FIGS 2,6). the mechanism of action of these antibiotics but also
recent reports from groups in the pharmaceutical for the development of novel inhibitors or, in a ligand-
industry have commented on the difficulty in identi- based approach, for searching for structurally simplified
fying inhibitors of bacterial rnA polymerase, a cru- analogues of these macrocyclic, natural product-based
cial enzyme in the construction of rnA chains from inhibitors. The binding position of the macrocycles in
DnA templates, using traditional HTS approaches9. the crystal structures of rnA polymerase can also be
SProUT was used to create the first de novo- used to identify suitable binding pockets in this large
designed rnA polymerase inhibitors (for example, enzyme. Indeed, it may be possible to design molecules
compound 26; see Supplementary information S1 that simultaneously inhibit two or more functional sites
(figure))81 using the X-ray crystal structure of rnA in the enzyme, which could minimize the potential for
polymerase from Thermus aquaticus. Preliminary the development of resistance.
experimental data indicate that these molecules do The growing global burden of bacterial resistance
inhibit bacterial rnA polymerase. to established antibacterial drugs is creating important
The production of inhibitors that are active in the but unmet medical needs, and in some cases isolates
low-micromolar concentration range after pure de novo have been described that are resistant to all previously

508 | jUly 2010 | VolUMe 8 www.nature.com/reviews/micro

2010 Macmillan Publishers Limited. All rights reserved

 REVIEWS

appropriate chemotherapeutic agents. In recent years, its infancy. However, the technology described in this
HTS of chemical libraries has dominated the search review may provide the springboard for future, more
for new antibacterial-drug leads. Unfortunately this successful attempts to discover the new classes of anti-
has not been successful, and a consensus is now emerg- bacterial drugs that will be required to fight infection in
ing that new approaches are required9. SBDD is still in the twenty-first century and beyond.

1. Chopra, I., Hesse, L. & ONeill, A. J. Exploiting current 23. Villoutreix, B. O. et al. Free resources to assist 39. Kopp, J. & Schwede, T. Automated protein structure
understanding of antibiotic action for discovery of new structure-based virtual ligand screening experiments. homology modeling: a progress report.
drugs. J. Appl. Microbiol. 92, S4S15 (2002). Curr. Protein Pept. Sci. 8, 381411 (2007). Pharmacogenomics 5, 405416 (2004).
2. Knowles, D. J. C. New strategies for antibacterial drug This report provides links for in silico ADME/tox 40. Schwede, T., Kopp, J., Guex, N. & Peitsch, M. C.
design. Trends Microbiol. 5, 379383 (1997). prediction tools, compound collections, some SWISS-MODEL: an automated protein homology-
3. Spellberg, B. et al. The epidemic of antibiotic-resistant ligand-based methods, characterization and modeling server. Nucleic Acids Res. 31, 33813385
infections: a call to action for the medical community simulation of three-dimensional targets, (2003).
from the Infectious Diseases Society of America. Clin. and homology modelling tools, as well as 41. Jones, T. A. & Kleywegt, G. J. CASP3 comparative
Infect. Dis. 46, 155164 (2008). links to software for druggable-pocket modeling evaluation. Proteins 37 (Suppl. 3), 3046
4. Talbot, G. H. et al. Bad bugs need drugs: an update predictions, active-site comparisons, analysis (1999).
on the development pipeline from the Antimicrobial of macromolecular interfaces, protein-docking 42. Tramontano, A., Leplae, R. & Morea, V. Analysis and
Availability Task Force of the Infectious Diseases tools to help identify binding pockets, and assessment of comparative modeling predictions in
Society of America. Clin. Infect. Dis. 42, 657668 protein ligand docking and scoring methods. CASP4. Proteins 45, 2238 (2001).
(2006). 24. Cheng, A. C. et al. Structure-based maximal affinity 43. Chan, D. C. M., Laughton, C. A., Queener, S. F. &
5. Chopra, I. et al. Treatment of health-care-associated model predicts small-molecule druggability. Nature Stevens, M. F. G. Structural studies on bioactive
infections caused by Gram-negative bacteria: a Biotech. 25, 7175 (2007). compounds. 34. Design, synthesis, and biological
consensus statement. Lancet Infect. Dis. 8, 133139 25. Carlson, H. A. & McCammon, J. A. Accommodating evaluation of triazenyl-substituted pyrimethamine
(2008). protein flexibility in computational drug design. Mol. inhibitors of Pneumocystis carinii dihydrofolate
6. Overbye, K. M. & Barrett, J. F. Antibiotics: where Pharmacol. 57, 213218 (2000). reductase. J. Med. Chem. 44, 25552564 (2001).
did we go wrong? Drug Discov. Today 10, 4552 An explanation of the need to accommodate an 44. Varney, M. D. et al. Crystal structure-based design and
(2005). ensemble of protein configurations in drug design synthesis of benz[cd]indole-containing inhibitors of
7. Newman, D. J., Cragg, G. M. & Snader, K. M. The and the computational methods available for thymidylate synthase. J. Med. Chem. 35, 663676
influence of natural products upon drug discovery. generating and manipulating that dynamic (1992).
Nat. Prod. Rep. 17, 215234 (2000). information. 45. Schmid, M. B. Seeing is believing: the impact of
8. Shlaes, D. M. An update on tetracyclines. Curr. Opin. 26. Jain, A. N. Surflex-Dock 2.1: robust performance from structural genomics on antimicrobial drug discovery.
Investig. Drugs 7, 167171 (2006). ligand energetic modeling, ring flexibility, and Nature Rev. Microbiol. 2, 739746 (2004).
9. Payne, D. J., Gwynn, M. N., Holmes, D. J. & knowledge-based search. J. Comput. Aided Mol. Des. 46. Berge, J. M. et al. Inhibitors of bacterial tyrosyl tRNA
Pompliano, D. L. Drugs for bad bugs: confronting the 21, 281306 (2007). synthetase: synthesis of four stereoisomeric analogues
challenges of antibacterial discovery. Nature Rev. 27. Zhao, Y. & Sanner, M. F. FLIPDock: docking flexible of the natural product SB-219383. Bioorg. Med.
Drug Discov. 6, 2940 (2007). ligands into flexible receptors. Proteins 68, 726737 Chem. Lett. 10, 18111814 (2000).
GlaxoSmithKline share their experience of (2007). 47. Berge, J. M. et al. Synthesis and activity of analogues
evaluating more than 300 genes and 70 HTS 28. Schnecke, V., Swanson, C. A., Getzoff, E. D., Tainer, of SB-219383: novel potent inhibitors of bacterial
campaigns over a period of 7 years; they analyse J. A. & Kuhn, L. A. Screening a peptidyl database for tyrosyl tRNA synthetase. J. Antibiot. (Tokyo) 53,
what they have learned and how this has influenced potential ligands to proteins with side-chain flexibility. 12821292 (2000).
their antibacterials strategy. Proteins 33, 7487 (1998). 48. Stefanska, A. L. et al. SB-219383, a novel tyrosyl
10. Fernandes, P. Antibacterial discovery and development 29. Claussen, H., Buning, C., Rarey, M. & Lengauer, T. tRNA synthetase inhibitor from a Micromonospora sp.
the failure of success? Nature Biotech. 24, FlexE: efficient molecular docking considering protein I. Fermentation, isolation and properties. J. Antibiot.
14971503 (2006). structure variations. J. Mol. Biol. 308, 377395 (Tokyo) 53, 345350 (2000).
11. Golebiowski, A., Klopfenstein, S. R. & Portlock, D. E. (2001). 49. Brown, P. et al. Synthetic analogues of SB-219383.
Lead compounds discovered from libraries. Curr. Opin. 30. Claussen, H., Buning, C., Rarey, M. & Lengauer, T. in Novel C-glycosyl peptides as inhibitors of tyrosyl tRNA
Chem. Biol. 5, 273284 (2001). Rational Approaches to Drug Design: Proceedings of synthetase. Bioorg. Med. Chem. Lett. 11, 711714
12. Chan, P. F., Holmes, D. J. & Payne, D. J. Finding the the 13th European Symposium on Quantitative (2001).
gems using genomic discovery: antibacterial drug Structure-Activity Relationships (eds Hltje, H.-D. & 50. Prasad, J. C., Goldstone, J. V., Camacho, C. J., Vajda,
discovery strategies the successes and the challenges. Sippl, W.) 324333 (Heinrich-Heine Universitt, S. & Stegeman, J. J. Ensemble modeling of substrate
Drug Discov. Today Ther. Strateg. 1, 519527 Dsseldorf, 2001). binding to cytochromes p450: analysis of catalytic
(2004). 31. Ota, N. & Agard, D. A. Binding mode prediction for differences between CYP1A orthologs. Biochemistry
13. Dorsey, B. D. et al. L-735,524: the rational design of a a flexible ligand in a flexible pocket using multi- 46, 26402654 (2007).
potent and orally bioavailable HIV protease inhibitor. conformation simulated annealing pseudo 51. Kllblad, P., Mancera, R. L. & Todorov, N. P.
J. Med. Chem. 37, 34433451 (1994). crystallographic refinement. J. Mol. Biol. 314, Assessment of multiple binding modes in ligand-
14. Erickson, J. et al. Design, activity and 2.8 crystal 607617 (2001). protein docking. J. Med. Chem. 47, 33343337
structure of a C2 symmetrical inhibitor complexed to 32. Skarzynski, T. et al. Structure of UDP- (2004).
HIV-1 protease. Science 249, 527533 (1990). N-acetylglucosamine enolpyruvyl transferase, an This study quantifies the effect of considering
15. Roberts, N. A. et al. Rational design of peptide-based enzyme essential for the synthesis of bacterial multiple docking solutions on the success rate of
HIV proteinase inhibitors. Science 248, 358361 peptidoglycan, complexed with substrate obtaining the crystallographic binding mode.
(1990). UDP-N-acetylglucosamine and the drug fosfomycin. 52. Limongelli, V. et al. Ensemble-docking approach on
16. Kaldor, S. W. et al. Viracept (nelfinavir mesylate, Structure 4, 14651474 (1996). BACE-1: pharmacophore perception and guidelines for
AG1343): a potent, orally bioavailable inhibitor 33. Eschenburg, S., Kabsch, W., Healy, M. L. & drug design. ChemMedChem 2, 667678 (2007).
of HIV-1 protease. J. Med. Chem. 40, 39793985 Schonbrunn, E. A new view of the mechanisms of 53. Kozakov, D., Clodfelter, K. H., Vajda, S. & Camacho,
(1997). UDP-N-acetylglucosamine enolpyruvyl transferase C. J. Optimal clustering for detecting near-native
17. Highleyman, L. New anti-HIV drugs in development. (MurA) and 5-enolpyruvylshikimate-3-phosphate conformations in protein docking. Biophys. J. 89,
BETA 12, 78 (1999). synthase (AroA) derived from X-ray structures of their 867875 (2005).
18. McCauley, J. Relenza. Curr. Biol. 9, R796R796 tetrahedral reaction intermediate states. J. Biol. 54. Kovac, A. et al. Discovery of new inhibitors of
(1999). Chem. 278, 4921549222 (2003). d-alanine:d-alanine ligase by structure-based virtual
19. Stratton, M. S. & Alberts, D. S. Current application of 34. Eschenburg, S. et al. A novel inhibitor that suspends screening. J. Med. Chem. 51, 74427448 (2008).
selective COX-2 inhibitors in cancer prevention and the induced fit mechanism of UDP-N-acetylglucosamine 55. Finn, J., Stidham, M., Hilgers, M. & Kedar, G. C.
treatment. Oncology (Williston Park) 16, 3751 enolpyruvyl transferase (MurA). J. Biol. Chem. 280, Identification of novel inhibitors of methionyl-tRNA
(2002). 1407014075 (2005). synthetase (MetRS) by virtual screening. Bioorg. Med.
20. Mardini, I. A. & FitzGerald, G. A. Selective inhibitors of 35. DeWeese-Scott, C. & Moult, J. Molecular modeling Chem. Lett. 18, 39323937 (2008).
cyclooxygenase-2: a growing class of anti-inflammatory of protein function regions. Proteins 55, 942961 56. Venkatachalam, C. M., Jiang, X., Oldfield, T. &
drugs. Mol. Interv. 1, 3038 (2001). (2004). Waldman, M. LigandFit: a novel method for the
21. Schmid, M. B. Crystallizing new approaches for 36. Yang, J. S., Chen, W. W., Skolnick, J. & Shakhnovich, shape-directed rapid docking of ligands to protein
antimicrobial drug discovery. Biochem. Pharmacol. E. I. All-atom ab initio folding of a diverse set of active sites. J. Mol. Graph. Model. 21, 289307
71, 10481056 (2006). proteins. Structure 15, 5363 (2007). (2003).
This article describes the use of structural 37. Arnold, K., Bordoli, L., Kopp, J. & Schwede, T. The 57. Dove, A. Drug screeningbeyond the bottleneck.
genomics, in the form of SBDD, to overcome the SWISS-MODEL workspace: a web-based environment Nature Biotech. 17, 859863 (1999).
bottlenecks in the antimicrobial pipeline, such as for protein structure homology modelling. 58. Zsoldos, Z. eHiTS: exhaustive flexible ligand docking
identifying good leads. Bioinformatics 22, 195201 (2006). with customizable scoring function tailored to protein
22. Halgren, T. A. Identifying and characterising binding 38. Bordoli, L. et al. Protein structure homology modeling families. SimBioSys Inc. [online], http://www.simbiosys.
sites and assessing druggability. J. Chem. Inf. Model. using SWISS-MODEL workspace. Nature. Protoc. 4, ca/science/presentations/acs-2004/abstract.html
49, 377389 (2009). 113 (2009). (2004).

nATUre reVIewS | Microbiology VolUMe 8 | jUly 2010 | 509

2010 Macmillan Publishers Limited. All rights reserved

REVIEWS

59. Zsoldos, Z., Johnson, A. P., Simon, A., Szabo, I. & 72. Gillet, V., Johnson, A. P., Mata, P., Sike, S. & Williams, Acknowledgements
Szabo, Z. eHiTS: novel algorithm for fast, exhaustive P. SPROUT: A programme for structure generation. Work in the authors laboratories is supported by grants
flexible ligand docking and scoring. SimBioSys Inc. J. Comput. Aided Mol. Des. 7, 127153 (1993). from the UK Medical Research Council (MRC), the UK
[online], http://www.simbiosys.com/science/ 73. Gillet, V. J., Myatt, G., Zsoldos, Z. & Johnson, A. P. Biotechnology and Biological Sciences Research Council
presentations/acs-2002/abstract.html (2002). SPROUT, HIPPO and CAESA: Tools for de novo (BBSRC), Heart Research UK and the European Communitys
60. Zsoldos, Z., Reid, D., Simon, A., Sadjad, B. S. & structure generation and estimation of synthetic Seventh Framework Programme FP7 (under grant agreement
Johnson, A. P. eHITS: an innovative approach to the accessibility. Perspect. Drug Discov. Des. 3, 3450 HEALTH-F4-2007-201924, EDICT Consortium).
docking and scoring function problems. Curr. Protein (1995).
Pept. Sci. 7, 421435 (2006). 74. Gillet, V. J. et al. SPROUT: recent developments in the Competing interests statement
61. Segura-Cabrera, A. & Rodriguez-Perez, M. A. de novo design of molecules. J. Chem. Inf. Comput. The authors declare no competing financial interests.
Structure-based prediction of Mycobacterium Sci. 34, 207217 (1994).
tuberculosis shikimate kinase inhibitors by high- 75. Law, J. M. S. et al. Validation of the SPROUT de novo
throughput virtual screening. Bioorg. Med. Chem. design program. J. Mol. Struct. 666, 651657 DATABASES
Lett. 18, 31523157 (2008). (2003). Entrez Genome Project:
62. Eitner, K. et al. eHiTS-to-VMD interface 76. Mata, P. et al. SPROUT: 3D structure generation http://www.ncbi.nlm.nih.gov/genomeprj
application. The search for tyrosine-tRNA ligase using templates. J. Chem. Inf. Comput. Sci. 35, Enterococcus faecium | Escherichia coli | Mycobacterium
inhibitors. J. Chem. Inf. Model. 47, 695702 479493 (1995). tuberculosis | Staphylococcus aureus |
(2007). 77. Zsoldos, Z., Szabo, I., Szabo, Z. & Johnson, A. P. Streptococcus pneumoniae | Thermus aquaticus
63. Turk, S. et al. Discovery of new inhibitors of the Software tools for structure based rational drug UniProtKB: http://www.uniprot.org
bacterial peptidoglycan biosynthesis enzymes MurD design. J. Mol. Struct. 666, 659665 (2003). AccD5 | AmpC | AroK | DdlB | MurA | MurD | MurF | VanA
and MurF by structure-based virtual screening. Biorg. 78. Horton, J. et al. Macrocyclic inhibitors of bacterial cell
Med. Chem. 17, 18841889 (2009). wall biosynthesis enzyme MurD. Bioorg. Med. Chem. FURTHER INFORMATION
64. Agrawal, H., Kumar, A., Bal, N. C., Siddiqi, M. I. & Lett. 13, 15571560 (2003). Colin W. G. Fishwicks homepage:
Arora, A. Ligand based virtual screening and 79. Besong, G. E. et al. A de novo designed inhibitor of http://www.chem.leeds.ac.uk/People/Fishwick.html
biological evaluation of inhibitors of chorismate d-Ala-d-Ala ligase from E. coli. Angew. Chem. Int. Ed. ACD:
mutase (Rv1885c) from Mycobacterium tuberculosis Engl. 44, 64036406 (2005). http://cds.dl.ac.uk/cds/datasets/orgchem/suppliers.html
H37Rv. Bioorg. Med. Chem. Lett. 17, 30533058 80. Sova, M. et al. Design and synthesis of new AutoDock: http://autodock.scripps.edu
(2007). hydroxyethylamines as inhibitors of d-alanyl-d-lactate BOMB: http://www.cemcomco.com/GenMol.html
65. Lin, T. W. et al. Structure-based inhibitor design of ligase (VanA) and d-alanyl-d-alanine ligase (DdlB). CheVi: http://www.simbiosys.ca/chevi/index.html
AccD5, an essential acyl-CoA carboxylase Bioorg. Med. Chem. Lett. 19, 13761379 Discovery Studio:
carboxyltransferase domain of Mycobacterium (2009). http://accelrys.com/products/discovery-studio/index.html
tuberculosis. Proc. Natl Acad. Sci. USA 103, 81. Agarwal, A. K., Johnson, A. P. & Fishwick, C. W. G. DOCK: http://dock.compbio.ucsf.edu
30723077 (2006). Synthesis of de novo designed small-molecule eHiTS: http://www.simbiosys.ca/ehits/index.html
66. Babaoglu, K. et al. Comprehensive mechanistic inhibitors of bacterial RNA polymerase. Tetrahedron ExPASy Proteomics Server: http://expasy.org
analysis of hits from high-throughput and docking 64, 1004910054 (2008). FAF-Drugs: http://bioserv.rpbs.univ-paris-diderot.fr/Help/
screens against -lactamase. J. Med. Chem. 51, 82. Davies, M. et al. Structure-based design, synthesis FAFDrugs.html
25022511 (2008). and characterization of inhibitors of human and GANDI: http://www.biochem-caflisch.unizh.ch/download
67. Morandi, S., Morandi, F., Caselli, E., Shoichet, B. K. & Plasmodium falciparum dihydroorotate Glide: http://www.schrodinger.com/products/14/5
Prati, F. Structure-based optimization of cephalothin- dehydrogenases. J. Med. Chem. 52, 26832693 GOLD:
analogue boronic acids as -lactamase inhibitors. (2009). http://www.ccdc.cam.ac.uk/products/life_sciences/gold
Biorg. Med. Chem. 16, 11951205 (2008). 83. Schlunzen, F. et al. Structural basis for the PENDANT: http://pedant.gsf.de
68. Teotico, D. G. et al. Docking for fragment inhibitors of interpretation of antibiotics with the peptidyl Protein Data Bank: http://www.pdb.org/pdb/home/home.do
AmpC -lactamase. Proc. Natl Acad. Sci. USA 106, transferase centre in eubacteria. Nature 413, SkelGen: http://www.denovopharma.com/page2.
74557460 (2009). 814821 (2001). asp?PageID=484
69. Fechner, U. & Schneider, G. Flux (1): a virtual 84. Pioletti, M. et al. Crystal structures of complexes of SPROUT: http://www.keymodule.co.uk/SPROUT.html
synthesis scheme for fragment-based de novo design. the small ribosomal subunit with tetracycline, edeine SPROUT-LeadOpt:
J. Chem. Inf. Model. 46, 699707 (2006). and IF3. EMBO J. 20, 18291839 (2001). http://www.keymodule.co.uk/SPROUT-LeadOpt.html
70. Boehm, H. J. et al. Novel inhibitors of DNA gyrase: 85. Franceschi, F. & Duffy, E. M. Structure-based drug SYBYL: http://www.tripos.com/index.php?family=modules,
3D structure based biased needle screening, hit design meets the ribosome. Biochem. Pharmacol. 71, SimplePage,,,&page=sybyl_sybyl
validation by biophysical methods, and 3D guided 10161025 (2006). TargetDB: http://targetdb.pdb.org
optimization. A promising alternative to random 86. Mukhopadhyay, J. et al. The RNA polymerase switch ZINC: http://zinc.docking.org
screening. J. Med. Chem. 43, 26642674 region is a target for inhibitors. Cell 135, 295307
(2000). (2008). SUPPLEMENTARY INFORMATION
71. Boda, K. & Johnson, A. P. Molecular complexity 87. Belogurov, G. A. et al. Transcription inactivation See online article: S1 (figure) | S2 (table)
analysis of de novo designed ligands. J. Med. Chem. through local refolding of the RNA polymerase All liNkS Are AcTive iN THe oNliNe PDF
49, 58695879 (2006). structure. Nature 457, 332338 (2009).

510 | jUly 2010 | VolUMe 8 www.nature.com/reviews/micro

2010 Macmillan Publishers Limited. All rights reserved