Fragment-based lead discovery: Screening and optimizing fragments for thermolysin inhibition

Article abstract of DOI:10.1002/cmdc.201000084

Fragment-based drug discovery has gained a foothold in today's lead identification processes. We present the application of in silico fragment-based screening for the discovery of novel lead compounds for the metalloendoproteinase thermolysin. We have chosen thermolysin to validate our screening approach as it is a well-studied enzyme and serves as a model system for other proteases. A protein-targeted virtual library was designed and screening was carried out using the program AutoDock. Two fragment hits could be identified. For one of them, the crystal structure in complex with thermolysin is presented. This compound was selected for structure-based optimization of binding affinity and improvement of ligand efficiency, while concomitantly keeping the fragment-like properties of the initial hit. Redesigning the zinc coordination group revealed a novel class of fragments possessing K_i values as low as 128 μm, thus they provide a good starting point for further hit evolution in a tailored lead design.

Full text of DOI:10.1002/cmdc.201000084

MED
DOI: 10.1002/cmdc.201000084
Fragment-Based Lead Discovery: Screening and
Optimizing Fragments for Thermolysin Inhibition
Lisa Englert, Katrin Silber, Holger Steuber, Sascha Brass, Bjçrn Over, Hans-Dieter Gerber,
Andreas Heine, Wibke E. Diederich, and Gerhard Klebe*[a]
Fragment-based drug discovery has gained a foothold in
today’s lead identification processes. We present the applica-
tion of in silico fragment-based screening for the discovery of
novel lead compounds for the metalloendoproteinase thermo-
lysin. We have chosen thermolysin to validate our screening
approach as it is a well-studied enzyme and serves as a model
system for other proteases. A protein-targeted virtual library
was designed and screening was carried out using the pro-
gram AutoDock. Two fragment hits could be identified. For
one of them, the crystal structure in complex with thermolysin
is presented. This compound was selected for structure-based
optimization of binding affinity and improvement of ligand ef-
ficiency, while concomitantly keeping the fragment-like proper-
ties of the initial hit. Redesigning the zinc coordination group
revealed a novel class of fragments possessing Ki values as low
as 128 mm, thus they provide a good starting point for further
hit evolution in a tailored lead design.
Introduction
In modern drug discovery, fragment-based lead discovery has
become increasingly popular[1] as it presents a promising alter-
native to conventional screening approaches. Consideration of
candidate molecule libraries that do not exceed a molecular
weight limit of ꢀ250 gmolꢁ1 not only leads to less complex
chemical structures but, most important, also assure that the
screened molecules form suitable interactions with protein
binding sites. This implies that a reduced number of com-
pounds, compared to traditional virtual or high-throughput
screening, have to be examined. Instead, chemical space cov-
erage and library diversity can be efficiently maximized. Al-
though fragments usually display only micromolar to millimo-
lar binding affinity, they provide excellent starting points for
lead generation. Fragment-to-lead optimization can be classi-
fied by two deviating concepts: Either, two or three fragments
detected at different sites in a binding pocket are linked,[2,3] or
an initial fragment hit is subsequently decorated to gain ligand
potency.[4,5] Usually, for fragments, a larger portion of atoms is
directly involved in interactions with the protein. Thus, frag-
ments are argued to bind more efficiently than conventional
screening hits. However, their low binding affinity makes it par-
ticularly difficult to detect them in standard biochemical
assays. Instead, several biophysical techniques, such as NMR,
SPR, or X-ray crystallography, have been deployed to overcome
this limitation.[6–9] Despite broad application of these experi-
mental approaches, in silico fragment screening still plays a
minor role.[10] Although there have been some case studies re-
ported that demonstrate virtual screening to be a powerful
tool to discover potent fragments,[11,12] this approach is still
often considered to be too unreliable using popular computer
tools. Indeed, facing the shortcomings of current docking algo-
rithms and scoring functions with respect to ranking and affini-
ty prediction of small fragments, more elaborate protocols
have to be evolved departing from routine virtual screening
strategies.[13]
Herein, we present a successful application of fragment-
based virtual screening followed by experimental confirmation
and subsequent hit optimization for the neutral metalloendo-
proteinase thermolysin (TLN, EC 3.4.24.27, 316 amino acids)
from Bacillus thermoproteolyticus.[14] We have selected this
target as it is well studied and serves as an established model
protein for which a significant number of X-ray structures has
been determined. Apart from a zinc ion, the enzyme contains
four calcium ions integrated in its tertiary structure that are im-
portant for its heat stability.[15] Its active site zinc ion catalyzes
the cleavage of peptide substrates prior to hydrophobic amino
acids, such as leucine or valine.[16] The enzyme has a large
rather rigid binding pocket composed of four subsites (S1, S2,
S1’, S2’).[17] The specificity pocket S1’ plays a major role in sub-
strate recognition and is hydrophobic in nature.
Additionally, the positively charged zinc ion with two formal
charges provides an exciting challenge for in silico fragment
screening. Docking of small molecules into a binding pocket
comprising a charged metal ion often suffers from artificially
exaggerated metal–ligand interactions.
[a] L. Englert, Dr. K. Silber, Prof. Dr. H. Steuber,+ Dr. S. Brass, B. Over,
H.-D. Gerber, Dr. A. Heine, Prof. Dr. W. E. Diederich, Prof. Dr. G. Klebe
Philipps-Universitꢀt Marburg, Institut fꢁr Pharmazeutische Chemie
Marbacher Weg 6, 35032 Marburg (Germany)
Fax: (+49)6421-28-28994
E-mail: Klebe@staff.uni-marburg.de
[+] Present address: Institut fꢁr Biochemie – Universitꢀt zu Lꢁbeck
Ratzeburger Allee 160, 23538 Lꢁbeck (Germany)
Supporting information for this article is available on the WWW under
http://dx.doi.org/10.1002/cmdc.201000084.
930
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2010, 5, 930 – 940

Fragment-based Screening
At first, we discovered initial hits by virtual screening and
confirmed their binding by a kinetic assay and crystal structure
analysis. We tried to evolve one detected hit not only by im-
proving its binding affinity and ligand efficiency,[18,19] but also
by simultaneously keeping its fragment-like properties.
Dock. Several validation runs considering four different charge
models were analyzed: a default charge of +2, a charge of
+1.4 according to a suggestion of the Shoichet group,[27]
a
charge of +0.5 implying that the charge at the zinc does not
exceed any negative charge in the binding pocket, and no
charge at all. As we were able to properly redock a considera-
ble number of test ligands in agreement with the native crystal
geometry using a charge of +0.5, we adjusted this value to
the zinc ion in our AutoDock runs.
Results and Discussion
Protein-targeted library design
As AutoDock achieved superior performance in reproducing
native geometries, it was finally selected as the docking
engine. Furthermore, the scoring function implemented in Au-
toDock 3.0 slightly outperformed the other functions in identi-
fying the native geometry of the redocked fragments. Never-
theless, being aware of the shortcomings of current scoring
functions, particularly with respect to fragments, we decided
to apply a consensus scoring model based on six functions in
parallel.[28] We selected the best 10% scored docking solutions
from each scoring function. Nevertheless, to only focus on rea-
sonable geometries, we postfiltered the selected solutions ac-
cording to our predefined pharmacophore hypothesis: 1) At
least one atom of a zinc-binding motif of the docked fragment
had to penetrate into a 3 ꢁ sphere encompassing the zinc ion;
2) to assure that only fragments with hydrophobic groups suf-
ficiently accommodating the S1’ pocket remained, we created
a pharmacophore sphere of 2 ꢁ placed at the position of the
ligand Cg–Leu side chain as found in the parent crystal struc-
ture of PDB code: 1TMN. Docked fragments sufficiently pene-
trating the S1’ pocket had to match with this sphere.
The World Drug Index (WDI), our in-house database of com-
mercially available compounds (CAC) and the “fragment-like”
subset from the ZINC database,[20] were assembled and
checked for competence to coordinate a metal ion, resulting in
a screening library with more than 900000 entries. Molecules
detected as too large with respect to a predefined size (MW
ꢀ250 Da) were split using the RECAP algorithm.[21] Further fil-
tering based on one-dimensional and two-dimensional proper-
ties was employed to create a database suitable for subse-
quent docking.
In a first filtering step, overall fragment properties such as
1) number of atoms, 2) molecular weight, and 3) the number
of rotatable bonds were considered. This reduced the number
of putative candidate molecules to 66949 entries.
Secondly, a pharmacophore hypothesis based on two dis-
tinct thermolysin-relevant features was generated for further
filtering: 1) Requirement of a Zn-binding group: Zn-nitrogen,
-oxygen or -sulfur contacting groups allowing for a distance to
the metal ion of 1.5 ꢁ to 2.0 ꢁ. All functional groups capable
of extending the coordination sphere defined by the active-
site residues His, His, Glu next to the zinc ion with tetra- or
pentavalent geometry were included. 2) Occupancy of the S1’-
specificity pocket: Studying known inhibitors and the size of
the hydrophobic pocket, a maximal size and possible branch-
ing of putative ligand groups were defined.
The obtained hits were visually inspected with respect to
the plausibility of the adopted ligand-bound conformation. Au-
toDock generated repeatedly particular docking solutions; this
fact was taken as a further figure-of-merit for our selection. Fi-
nally, 19 compounds were considered for experimental valida-
tion. Their chemical structures are listed in Table 1.
Only 187 candidates passed these rather stringent, knowl-
edge-based selection criteria. They were complemented by an
additional subset of 247 entries from the ZINC database that
corresponded to fragment-like properties. They were used
without further filtering.
Screening results
As fragments often display low micromolar to millimolar affini-
ty the determination of their inhibition constants by standard
bioassays is notoriously difficult. To estimate the potential of
discovered fragments for further development, the concept of
ligand efficiency has been suggested, which relates the free
energy of binding to the number of non-hydrogen atoms in a
fragment. Although binding affinity is low, fragments usually
exhibit high ligand efficiency. For fragments 9 and 15 we de-
termined an in vitro potency of Ki =1.7 mm and Ki =154 mm re-
spectively, which corresponds to a ligand efficiency of 0.27 kcal
and 0.33 kcal. To structurally characterize fragment binding, we
embarked on co-crystallization and soaking attempts. We pre-
pared soaking solutions containing the discovered fragments
at saturated concentration. Native thermolysin crystals not
treated with any ligand solution, usually contain the micromo-
lar autoproteolysis dipeptidic product Val–Lys in the S1’ pocket.
Yet, we succeeded in using such protein crystals for soaking
and for 4, 5, 7, and 10; some residual difference electron den-
sity could indeed be observed indicating that not solely the di-
Docking and post-filtering
To select the most appropriate docking tool we tested three
different docking programs: AutoDock 3.0,[22] FlexX 1.13,[23] and
Gold 3.0[24] in combination with several scoring functions by re-
docking a set of fragment-like ligands found in thermolysin
crystal structures. Six scoring functions were used: The parent
scoring functions implemented in AutoDock, FlexX, Gold and
ChemScore,[25] DrugScoreCSD, and a fragment-tailored variant of
the scoring function SFCscore.[26] Fragments usually exhibit
weak binding affinity. Thus, docking into a binding pocket con-
taining a doubly charged Zn2+ ion will inevitably place any
atom of a screening candidate bearing a negative partial
charge next to the metal ion. To avoid possible artificial inter-
action generated motifs, we had to adjust the charge at the
zinc ion during our docking and scoring attempts with Auto-
ChemMedChem 2010, 5, 930 – 940
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemmedchem.org
931

G. Klebe et al.
MED
density, and thus its binding geometry in complex with ther-
molysin could be determined (Figure 1a).
Table 1. Virtual screening hits.
Compd Structure
Compd Structure
Binding mode of 3-methylaspirin
Important interactions of 3-methylaspirin with thermolysin are
the coordination of the zinc ion and the occupation of the hy-
drophobic S1’ pocket (Figure 1b). 3-Methylaspirin coordinates
with its carboxylate group in bidentate fashion to the zinc ion
(2.2 ꢁ, 2.7 ꢁ) and forms an additional hydrogen bond to
Glu143 (2.7 ꢁ). As known from other thermolysin inhibitors,
the metal–carboxylate coordination seems to mimic the cata-
lytic transition state.[29,30] Further hydrogen bonds can be ob-
served between the ester carbonyl functionality of the ligand
and Arg203. In addition to a direct hydrogen bond (3.0 ꢁ), a
water-mediated hydrogen bond (2.9 ꢁ···H2O···2.9 ꢁ) between
carbonyl oxygen and the second guanidinium nitrogen of
Arg203 is observed. The aromatic ring is located in the mainly
hydrophobic S1’ pocket. Altogether, 49 van der Waals contacts
(ꢂ3.3 ꢁ) can be counted between 3-methylaspirin and thermo-
lysin. The 3-methyl group points towards the S2’ pocket. Most
thermolysin structures deposited in the Protein Data Bank
(PDB) represent a conserved conformation of all amino acids in
the binding site. However, our structure reveals a conforma-
tional transition upon binding: Asn112 moves out of the usual-
ly observed orientation thus creating additional space for the
aromatic ring of the ligand.
1
11
2
3
12
13
4
5
14
15
6
7
8
16
17
18
Based on the crystal structure of the fragment hit 9, we
reevaluated the performance of the different docking and scor-
ing tools in our screening campaign. As mentioned, the ligand
induces an adaptation of the Asn112 side chain, which swings
out of the position observed in the overwhelming part of all
determined complex structures of thermolysin in the PDB. As a
consequence, an extended space is available to host the
ligand. With the knowledge of our rather unique protein con-
former, we searched again through the deposited crystal struc-
tures and could only detect two other cases among 69 exam-
ples (PDB codes: 1 KS7 and 1 KR6) in which this Asn112 adopts
a similar conformation as in our complex. Thus the observed
induced-fit conformation with bound 9 seems most likely to
be energetically less favorable.
9
19
However, our entire docking screen had been performed
with the conformers most frequently observed in crystal struc-
tures of thermolysin. Considering the experimentally detected
binding mode of 9, a severe steric clash with the protein
would occur (Figure 1c). Accordingly, among our docking solu-
tions we can only expect solutions that approximate the ac-
tually observed geometry in the crystal structure. Nevertheless,
it is quite educational to analyze the different solutions of Au-
toDock with respect to the applied scoring functions (Fig-
ure 2a and b). DrugScore places a solution on first rank that
shows 9 coordinated to the zinc but oriented towards the S1
pocket. The scoring function implemented in AutoDock selects
a solution that interacts with the zinc and places the 3-methyl-
phenyl portion towards the S1’ pocket as most favorable,
being in agreement with the subsequently determined crystal
structure. However, the acetyl substituent is incorrectly placed
10
peptide cleavage product was present. Unfortunately, the affin-
ity of our screening hits was obviously too low to fully replace
the Val–Lys autoproteolysis product and to achieve sufficient
occupancy of the fragment to record an unambiguous com-
plex structure. To our disappointment, our most potent frag-
ment 15 was not sufficiently soluble to obtain a complex struc-
ture. Instead, the less potent but more soluble fragment 9 (3-
methylaspirin) exhibited a clearly visible difference electron
932
www.chemmedchem.org
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2010, 5, 930 – 940

Fragment-based Screening
Figure 1. Thermolysin in complex with 3-methylaspirin. a) Difference electron density (FoꢁFc) of thermolysin in complex with 3-methylaspirin is contoured at
2s in green (blue: S1’, red: S2’ pocket). Ligand atoms are colored by atom type, oxygen in red, carbon in yellow. The solvent accessible surface of thermolysin
is shown in beige; the zinc ion is shown as a sphere in blue. b) Interaction pattern found for the binding mode of 3-methylaspirin in thermolysin. Hydrogen-
bonding interactions are shown in orange, hydrophobic interactions in blue. Protein and ligand atoms are colored by atom type, protein carbon atoms in
green, ligand carbon atoms in yellow, oxygen in red, nitrogen in blue. c) Electron density for 3-methylaspirin (cyan) embedded in the crystal structure (PDB
code: 1TMN). Ligand atoms are colored by atom type: oxygen in red, carbon in gray. The solvent accessible surface of thermolysin is shown in beige, the zinc
ion shown as a sphere in blue. The ligand would clash with the protein at Asn112 if this residue did not adopt an alternative conformation. In docking how-
ever, the 1TMN structure was used.
towards the entrance of the cat-
alytic center. Best performance is
observed for the AutoDock geo-
metries post-ranked by our frag-
ment-tailored SFCscore.[26] The
latter function picks a solution
for the best scoring rank that ac-
tually corresponds convincingly
to the finally determined crystal
structure. All substituents are ori-
ented correctly, and, based on
this docking solution, correct
predictions of an optimization
strategy (for example, increasing
the hydrophobicity of the acetyl
substituent pointing into the S1’
pocket) would have been possi-
ble (Figure 2c and d).
The observed deviations be-
tween docking solution and
crystal structure clearly result
from the incorrect and neglected
induced-fit adaptation of the
protein. It is also interesting to
mention that, if starting with the
observed protein conformer in
the crystal structure with 9, all
docking programs and applied
scoring functions would have
suggested the correct solution
Figure 2. Docking solutions of 3-methylaspirin (9) in PDB code: 1TMN performed by AutoDock and ranked by dif-
for 9 as observed in the crystal
structure. This clearly indicates
that even small induced-fit adap-
tations of the protein are essen-
tial and represent one of the
major bottlenecks that easily
ferent scoring functions. The solvent accessible surface of 1TMN is colored beige. a) Solution (carbon atoms gray)
ranked highest by DrugScoreCSD; b) Best-ranked solution by the AutoDock scoring function; c) Best-ranked solu-
tion retrieved by the fragment-tailored SCFscore; d) Superposition of the best-ranked SCFscore docking solution
and the subsequently determined crystal structure (carbon atoms green). The orientation found in the crystal
structure induces an adaptation of Asn112, thus opening a modified conformation of the protein. The observed
deviation between crystal structure and docking solution clearly results as a consequence of the not correctly
considered adaptation of the protein.
ChemMedChem 2010, 5, 930 – 940
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemmedchem.org
933

G. Klebe et al.
MED
mislead docking and scoring attempts to suggest the incorrect
binding mode.
ality of 3-methylaspirin. Based on this assumption, the corre-
sponding propyl and cyclopropyl derivatives of 3-methylsalicyl-
ic acid were synthesized.
For the synthesis of these derivatives, commercially available
3-methylsalicylic acid 22 served as the starting point. In case of
3-methyl-2-(propionyloxy)benzoic acid 23, the reaction of 3-
methylsalicylic acid with propionic anhydride, also in presence
of catalytic amounts of sulfuric acid, lead to the desired com-
pound. The corresponding 2-[(cyclopropylcarbonyl)oxy]-3-
methyl-benzoic acid 24 was obtained by reaction of 3-methyl-
salicylic acid with cyclopropane carbonyl chloride under stan-
dard coupling conditions (Scheme 2).
Hit optimization
Although compound 15 exhibits an approximately tenfold
higher binding affinity compared to 3-methylaspirin 9, the
latter provides an excellent starting point for further optimiza-
tion of binding affinity and ligand efficiency. In the complex
with 9, the 3-methyl group performs extended surface con-
tacts with the protein next to the S2’ pocket. Accordingly, we
replaced the 3-methyl group by a 3-chloro substituent to
probe for the importance of hydrophobic interactions.
To get access to the corresponding 2-(acetyloxy)-3-chloro-
benzoic acid 21, we treated 3-chloro-2-hydroxy-benzoic acid
20 with acetic anhydride (Ac2O) in the presence of a catalytic
amount of sulfuric acid, as shown in Scheme 1.
Scheme 1. Synthesis of 21. Reagents and conditions: a) (CH3CO)2O, H2SO4,
neat, 1408C, 2 h, 43%.
Scheme 2. Synthesis of 23 and 24. Reagents and conditions: a) (C2H5CO)2O,
H2SO4, neat, 1208C, 5 h, 51%; b) cC3H5COCl, CH2Cl2, Et3N, DMAP, RT, 14 h,
29%.
The analogous 3-chloroaspirin 21 shows a threefold im-
provement in affinity. This increase might be attributed to dif-
ferent effects: First, the chloro substituent might experience
slightly better contacts with the protein. Furthermore, the elec-
tron-withdrawing effect of the chloro substituent possibly
shifts the pKa of the carboxylate group thus making this por-
tion a better coordinative anchor for zinc. Finally, our recent
studies on aromatic ligand portions accommodated in the S1
pocket of the serine protease thrombin clearly showed that a
meta-chloro-substituted phenyl group has an advantage over
the corresponding methyl derivative due to a more favorable
desolvation energy of the chloro derivative.[31] This effect,
which should give an advantage to any chloro aromatic
system compared to the corresponding methyl derivative,
might also contribute to the enhanced binding affinity of 3-
chloroaspirin 21. However, a major disadvantage—possibly re-
lated to the latter desolvation effects—is that the compound is
significantly less soluble in aqueous solution.
The crystal structures of thermolysin in complex with 23 and
24 revealed indeed more convincing ligand–protein contacts
with enhanced shape complementarity (see Supporting Infor-
mation). As intended, the elongated ester functionalities reach
deeper into the S1’ pocket, improving van der Waals contacts
between ligand and protein. For the ethyl derivative and for
the cyclopropyl derivative, a total of 52 and 61 van der Waals
contacts, respectively, were observed compared to 49 contacts
for the parent structure (3-methylaspirin). However, to our sur-
prise, no improvement in binding affinity compared to 3-meth-
ylaspirin could be achieved.
The three structure determinations of 3-methylaspirin 9 and
its ester derivatives 23 and 24 clearly demonstrate that Asn112
has to adopt an alternative conformation upon ligand binding.
Obviously, the size of the central phenyl ring of our fragment
is too large and provokes the observed induced-fit adaptation.
Probably, the required rearrangement is energetically rather
costly. As a consequence, we introduced a smaller ring system
replacing the aromatic six-membered ring to reduce the unfav-
orable interactions with Asn112. A smaller thiophene moiety
was considered to replace the phenyl core, a frequently
chosen bioisosterism in lead optimization.[33,34]
As described above, 3-methylaspirin interacts weakly in the
S1’-specificity pocket of thermolysin. Interestingly, although this
pocket is not satisfactorily addressed by the ligand, no crystal-
lographically ordered water or any other molecule could be
observed in the remaining space. As the nature of the pocket
is mainly hydrophobic, binding affinity should increase by at-
taching nonpolar substituents that penetrate deeper into this
pocket filling this unoccupied, suboptimally hydrated pocket.
Furthermore, this will provide better shape complementarity in
terms of van der Waals contacts with hydrophobic residues of
the binding site.[32] This design concept encouraged us to first
focus our optimization efforts on expanding the ester function-
The thiophene analogue of 3-methylaspirin 9 can be synthe-
sized in two steps starting with the commercially available 3-
amino-substituted ester 25, which is N-acylated with acetyl
chloride followed by saponification with lithium hydroxide
giving rise to the corresponding carboxylic acid after acidifica-
tion. In the present case, we prepared the amide derivative 27
934
www.chemmedchem.org
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2010, 5, 930 – 940

Fragment-based Screening
instead of the corresponding ester analogue of 9 as the 3-hy-
droxy-substituted analogue of 25 is not commercially available
(Scheme 3). Despite the reduced size of the central core, 27
did not show any advantage in enzyme inhibition (Table 2).
(Figure 3). To better fill this rather stringent space but simulta-
neously providing a scaffold capable of orienting molecular
portions either into the S1- or S1’- pockets, the zinc coordinat-
ing atoms should be directly incorporated into the central ring
Scheme 3. Preparation of acid 27. Reagents and conditions: a) CH3COCl,
DIPEA, DMAP, CH2Cl2, ꢁ608C!RT, 12 h, 80%; b) LiOH, MeOH/H2O (2:1),
08C!RT, 15 h, then H+, 79%.
Table 2. Selected and optimized screening hits.
Compd
Structure
Ki value [mm]
9
1700
Figure 3. Proposed binding mode of 3-hydroxy-2-methylquinazolin-4(3H)-
one 30c (carbon atoms: light blue), generated by docking, superimposed
with the crystal structure of 24 (carbons atoms: gray). Two conformers of
Asn112 are shown, the most frequently observed and probably energetically
more favorable geometry in orange, the one adopted in the crystal structure
with 24 in magenta. The docking solution suggests bidentate interaction of
the quinazolinone moiety with the zinc ion and a hydrogen bond with
Asn112 is proposed (green line). The phenyl moiety orients towards the S1
pocket, the 2-methyl group towards the S1’ pocket. Chemical expansions of
the fragment have to occur along the indicated vectors (red).
21
27
~570
n.i.[a]
system. Hydroxamate functionalities are known to exert strong
chelating power, particularly towards zinc.[29,35] Such bidentate
functionality can be incorporated into the core moiety as 3-hy-
droxy-quinazolinone. Docking of this fragment suggests zinc
complexation in bidentate fashion (Figure 3). The annulated ar-
omatic ring system would then be able to form p–p stacking
with Phe114 in the S1 pocket. The available space below
Asn112 would allow the carboxamide group of this residue to
return to its usually observed orientation in thermolysin com-
plexes. Furthermore, it could potentially form a hydrogen
bond to the central quinazolinone moiety. To address the hy-
drophobic S1’ pocket with additional substituents, attachment
at the 2-position appears most appropriate.
33
n.i.[a]
30a
930
30b
315
128
The new target scaffold can be synthesized starting from
the methylester of anthranilic acid 28. Considering a thiophene
instead of a phenyl ring as annulated aromatic system, the
ester of 3-amino-thiophene carboxylic acid 31 can be used as
the starting point. In the first step, the corresponding acylami-
no derivatives were synthesized via acylation of the amino
function. The cyclic hydroxamate was formed in a microwave-
assisted reaction using hydroxylamine hydrochloride in the
presence of a strong base such as potassium hydroxide
(Scheme 4). Possible mechanisms of the ring formation are de-
scribed and discussed in the literature in detail.[36–38]
30c
[a] No inhibition.
We therefore reconsidered the geometry of our central
moiety coordinating to the zinc ion. The crystal structures of 9,
23, and 24 indicate that the coordinating carboxylate group at
the phenyl ring requires too much of the space available be-
tween the zinc ion and the carboxamide group of Asn112
Biochemical testing of the novel fragments revealed an over-
all improvement of binding affinity along with convincing
ligand efficiency: 30a (Ki =930 mm), 30b (Ki =315 mm), and 30c
ChemMedChem 2010, 5, 930 – 940
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemmedchem.org
935

G. Klebe et al.
MED
Thermolysin was dissolved in 0.05m Tris/HCl buffer (pH 7.3), con-
taining DMSO (50% v/v) and cesium chloride (1.9m). The final pro-
tein concentration was 4.0 mm. Crystals were grown at 188C by
the sitting drop vapor diffusion method using water as reservoir
solution. Protein–ligand complex crystals were obtained by soaking
crystals in a solution of 0.1m Tris/HCl, 2 mm calcium chloride, and
10% DMSO. Ligand concentration was 50–100 mm for fragment
soaking. When low solubility was given, saturated solutions were
used. Crystals were soaked between one and two days before
freezing. For cryoprotection crystals were briefly soaked in cryo-
buffer (10 mm Tris/HCl, pH 7.3, 10 mm CaCl2, 5% DMSO, 20% glyc-
erol).
Data collection, phasing and refinement
Data for 3-methyl-2-(propionyloxy)benzoic acid 23 were collected
at 100 K using a Rigaku R-AXIS IV image plate detector with CuKa
radiation from an in-house Rigaku RU-H3R rotating anode. Data for
2-(acetyloxy)-3-methylbenzoic acid 9 and 2-[(cyclopropylcarbonyl)-
oxy]-3-methylbenzoic acid 24 were determined at the synchrotron
BESSYII in Berlin on PSF beamline 14.2 equipped with a MAR-CCD
detector. Data were processed and scaled with Denzo and Scale-
pack as implemented in HKL2000.[40] The coordinates of thermoly-
sin in complex with an N-carboxymethyl dipeptide inhibitor (PDB
code: 1TMN) were used after removal of ligand, metal ions and
water molecules for initial rigid-body refinement of the protein
atoms followed by repeated cycles of maximum likelihood, simu-
lated annealing and B-factor refinement using the CNS program
package.[41] The structure refinement was continued with SHELXL-
97,[42] for each refinement step at least ten cycles of conjugate gra-
dient energy minimization were performed with restraints on bond
distances, angles, and B values (Table 3). Intermittent cycles of
model building were done with the program COOT.[43] The coordi-
nates have been deposited in the PDB (http://www.rcsb.org/pdb/)
with codes 3FCQ, 3F2P, and 3F28. Ramachandran plot outlier
Thr26 participates in a g-turn.[44] Van der Waals contacts were
quantified with CONTACSYM.[45,46]
Scheme 4. Preparation of quinazolinone-based fragments 30a–c and 33. Re-
agents and conditions: a) (RCO)2O, neat, RT, 24 h–5 d, 77–87%; b) NH2OH·HCl,
KOH, MeOH, microwave assisted, 1208C, 150 W, 6 min, 31–41%;
c) (CH3CO)2O, neat, RT, 5 h, 87%; d) NH2OH·HCl, KOH, MeOH, microwave as-
sisted, 1208C, 150 W, 6 min, 42%.
(Ki =128 mm), with ligand efficiencies of 0.32 kcal, 0.34 kcal, and
0.36 kcal, respectively (Table 2).
Summary and Conclusions
Our concept of virtual screening to detect new fragment-like
inhibitors for thermolysin has resulted in two novel hits. Impor-
tant for the success was the design of a protein-targeted li-
brary. As fragment scoring is a special challenge for most of
the currently applied scoring functions, we employed consen-
sus scoring in combination with visual inspection of the dock-
ing solutions. The crystal structure for hit 9 (3-methylaspirin)
was taken as a starting point for structure-based optimization
of binding affinity and ligand efficiency.
We first concentrated on enhancing interactions between
the hydrophobic S1’-specificity pocket of thermolysin. Two de-
rivatives of 3-methylaspirin with longer, more hydrophobic
ester functionalities were synthesized. Crystal structures of
these two derivatives revealed that up to 12 additional van der
Waals interactions to the aliphatic amino acids of the specifici-
ty pocket are formed by the ester groups of these ligands. As
our lead compound induces a presumably unfavorable adapta-
tion of the protein for steric reasons, we replaced the central
phenyl moiety of methylaspirin with a smaller thiophene ring.
Subsequent design suggested a 3-hydroxy-quinazolinone scaf-
fold featuring a cyclic hydroxamate functionality to coordinate
to the zinc ion. Synthetically easily accessible fragments based
on this scaffold show convincing binding to the target protein
with good ligand efficiency. They will serve as prospective hits
to further evolve the detected fragments into potent leads.
Virtual screening
Two-dimensional connection tables from the World Drug Index
2001 (WDI) (http://scientific.thomson.com/products/wdi) and
900000 compounds from different commercial sources compiled
in our in-house screening library (Asinex, IBS, Specs, Sigma, Lead-
quest, Maybridge) were converted into three-dimensional struc-
tures using CORINA.[47] Phosphate, sulfonate and carboxylate func-
tions were assumed to be deprotonated, amines protonated. The
following filters were applied for reducing the initial number of
compounds in the database from over 900000 to 187.
WDI preselection
All diagnostics, vitamins, and drugs administered differently than
oral were discarded. A further selection criterion was molecular
weight (ꢂ) set to 70 gmolꢁ1 <ꢂ<750 gmolꢁ1. Compounds in
agreement with this were split using the RECAP algorithm. Scissile
atoms were labeled with an R group dummy atom. After all dupli-
cates were removed, 3976 molecules remained. Firstly the selector
compound filtering utility of SYBYL was applied to keep only mole-
cules with at least seven atoms, a molecular weight (ꢂ) with
70 gmolꢁ1 ꢀꢂꢀ250 gmolꢁ1 and not more than eight rotatable
bonds.
Experimental Section
Crystallization
Native thermolysin (purchased from Calbiochem) was crystallized
as described by Holmes and Matthews[39] with slight modifications.
936
www.chemmedchem.org
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2010, 5, 930 – 940

Fragment-based Screening
From the “fragment-like” subset of
the ZINC 7 database the subset for
docking to metals (247 com-
pounds), which already matches
our criteria, was taken without any
further filtering step.
Table 3. Data collection and refinement statistics for TLN in complex with 3-methylaspirin esters.
Crystal data
PDB code
3-methylaspirin
3-methylaspirin
ethyl ester 23
3-methylaspirin
cyclopropyl ester 24
9
3FCQ
3F2P
3F28
Data collection and processing
No. crystals used
Wavelength [ꢁ]
Space group
Unit cell parameters:
a, b [ꢁ]
Docking
1
1
1
0.9184
P6122
1.5418
P6122
0.9184
P6122
A validation with three different
docking
programs
FlexX 1.13,
Gold 3.0, and AutoDock 3.0 and six
different scoring functions, the
scoring functions implemented in
FlexX, Gold, AutoDock and addi-
tionally ChemScore and Drug-
ScoreCSD, and a fragment-trained
function variant of SFCscore was
carried out. The protein structure
was retrieved from PDB code:
1TMN. The binding site was de-
fined by all residues coinciding
within an 8 ꢁ distance around the
bound ligand in PDB code: 1TMN.
For docking with AutoDock 3.0,
the partitioning of the total charge
into individual atomic contribu-
tions was performed using the
Gasteiger–Marsili charges.[50] Polar
hydrogens were added with the
PROTONATE utility in AMBER.[51]
AMBER united-atom charges were
assigned as defined in the AMBER
force field.[52] The charge of the
zinc ion was reduced from +2 to
+0.5. Docking runs were per-
formed with the Lamarckian ge-
netic algorithm as implemented in
AutoDock 3.0, using an initial pop-
ulation of 50 individuals, a maxi-
mum number of 0.75ꢂ106 energy
92.7
92.8
92.5
c [ꢁ]
130.0
130.7
129.0
Diffraction data
Resolution range [ꢁ]
Unique reflections
R(I)sym [%][b]
Completeness [%]
Redundancy
50–1.75
30–1.95
30-1.68
33957 (1669)[a]
8.8 (49.5)[a]
100.0 (100.0)[a]
6.7 (7.1)[a]
24171 (1151)[a]
11.2 (49.5)[a]
96.9 (94.7)[a]
8.5 (8.7)[a]
37792 (1852)[a]
6.3 (48.1)[a]
99.8 (99.4)[a]
9.8 (6.4)[a]
I/s(I)
21.0 (3.9)[a]
19.3 (4.7)[a]
31.6 (3.1)[a]
Refinement
Resolution range [ꢁ]
Reflections used in refinement
Final R values:
Rfree (Fo; Fo >4sFo) [%][c]
Rwork (Fo; Fo >4sFo) [%][c]
Number of atoms (non-hydrogen)
Protein atoms
Water molecules
Ligand atoms
RMSD, angle [8]
RMSD, bond [ꢁ]
Ramachandran plot[d]
Most favored regions [%]
Additionally allowed regions [%]
Generously allowed regions [%]
Disallowed regions [%]
Mean B-factors [ꢁ2]
Protein atoms
10–1.75
33663
10–1.95
23943
10–1.68
37567
23.0; 20.4
17.6; 15.5
26.2; 24.1
18.4; 16.6
23.6; 22.1
18.9; 17.8
2439
277
13
1.8
0.008
2448
191
15
1.8
0.007
2432
137
16
1.9
0.009
87.8
11.5
0.4
87.8
11.5
0.4
88.5
10.4
0.7
0.4
0.4
0.4
13.7
11.5
25.4
21.9
19.7
20.0
27.9
19.5
22.3
20.1
32.4
25.8
Metals
Water molecules
Ligand atoms
evaluations,
a mutation rate of
[a] Values in parentheses are statistics for the highest-resolution shell. [b] R(I)sym =[ꢀhꢀi jIi(h)ꢁ<I(h)>j/
ꢀhꢀiIi(h)]ꢂ100, where <I(h)> is the mean of the I(h) observation of reflection h. [c] Rwork =ꢀhkl jFoꢁFc j/ꢀhkl jFo j
; Rfree was calculated as for Rwork, but on 5% of the data excluded from refinement. [d] From Procheck.
0.02, a crossover rate of 0.8, and
an elitism value of 1. Generated
ligand docking solutions differing
by RMSD <1 ꢁ were clustered to-
gether. Ten solutions were generat-
ed for each considered ligand.
Secondly, for pharmacophore generation, all zinc-coordinating
groups were extracted using ReliBase+.[48,49] Additionally, functional
For validation, a set of 20 ligands extracted from their parent ther-
molysin X-ray structures was used and the ligands were redocked
into the protein (PDB codes: 1HYT, 1TLP, 1QF2, 2TMN, 4TLN, 5TLN,
7TLN, 1 KRO, 1 KTO, 1PE5, 1PE7, 1PE8, 1QF0, 1QF1, 1TLP, 1TMN,
4TMN; 5TMN, 6TMN, 2TLX). AutoDock reproduced 72% of the
native geometries. Ligands exceeding fragment size were split ac-
cording to the RECAP algorithm. For validation only fragments
binding into the S1’ pocket and to the zinc ion were considered,
resulting in a final validation set of 32 ligands. For AutoDock a spe-
cial atom type R was introduced to prevent R group to protein in-
teractions. Parameters were chosen similar to the nonaromatic
carbon atom type and all attracting interaction points on the
energy grid were set to zero. The post docking filter was per-
formed with the rigid three-dimensional search algorithm in Unity.
Binding mode prediction of compound 30c was made with Auto-
Dock 3.0 using the same parameters as described above.
groups binding to the S1’-specificity pocket and the size of the
pocket itself were taken into account. The program Unity 4.0
(http://www.tripos.com) was used to perform a two-dimensional
connectivity screening. Only molecules containing the following
atoms were included: C, O, N, H, Cl, F, S. Finally, molecules compris-
ing functional groups not likely to be favorable for drugs were ex-
cluded (Table 4).
Table 4. Functional groups excluded from screening.
ChemMedChem 2010, 5, 930 – 940
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemmedchem.org
937

G. Klebe et al.
MED
J=7.7 Hz, 1H), 2.61 (q, J=7.7 Hz, 2H), 2.15 (s, 3H), 1.17 ppm (t, J=
7.5 Hz, 3H); 13C NMR (125 MHz, [D6]DMSO): d=172.2, 166.1, 148.9,
135.1, 131.8, 129.2, 125.7, 124.3, 27.1, 15.8, 9.0 ppm; MS (ES+) m/z
(%): 226 (91) [M+NH4]+, 434 (100)[2M+NH4]+, 642 (22) [3M+NH4]+;
HRMS (EI+): m/z [M]+ calcd for C11H12O4: 208.073559, found:
208.072392; Anal calcd for C11H12O4: C 63.45, H 5.81, found: C
63.43, H 5.85.
General synthesis
Reported yields refer to the analytically pure product obtained by
column chromatography or recrystallization. All proton and carbon
NMR spectra were recorded on a Jeol Eclipse +500 MHz spectrom-
eter (1H NMR: 500 MHz, 13C NMR: 125 MHz) or a Jeol ECX-400 spec-
trometer (1H NMR: 400 MHz, 13C NMR: 100 MHz). Chemical shifts (d)
are stated in parts per million (ppm) and were referenced to TMS
at 0.00 ppm. 13C NMR spectra were referenced to CDCl3
(77.00 ppm) or DMSO (39.70 ppm). Coupling constants (J) are
given in Hz. Abbreviations: bdd=broad doublet of doublets, bm=
broad multiplet, bs=broad singlet, d=doublet, m=multiplet,
sm=symmetric multiplet, q=quartet, s=singlet, t=triplet, ps=
pseudo.
2-[(Cyclopropylcarbonyl)oxy]-3-methylbenzoic acid (24): Cyclo-
propanecarbonyl chloride (2.0 mL, 22.0 mmol) and a catalytic
amount of DMAP were added to a stirred solution of 2-hydroxy-3-
methyl-benzoic acid 22 (3.04 g, 20.0 mmol) and Et3N (7.0 mL,
50.0 mmol) in CH2Cl2 (50 mL) under Ar. After stirring at RT for 14 h,
HCl (1n, 50 mL) was added and the organic layer was separated.
The aqueous layer was extracted twice with CH2Cl2 and the com-
bined organic layers were washed with brine, dried over MgSO4, fil-
tered, and concentrated in vacuo. Column chromatography
(CH2Cl2/MeOH/formic acid, 10:1:0.1) afforded 1.28 g (29%) of 24 as
Mass spectra were obtained from a double-focusing sectorfield
spectrometer typ 7070H by Vacuum Generators or by a double-fo-
cusing sectorfield spectrometer typ VG-AutoSpec by Micromass.
Combustion analyses were determined on a Vario Micro Cube by
Elementar Analysen GmbH (Hanau, Germany). Melting points were
determined on a Leitz HM-Lux apparatus and are uncorrected.
Flash column chromatography was performed using silica gel 60
(0.04–0.063 mm) purchased from Macherey–Nagel (Dꢃren, Germa-
ny). For reversed-phase chromatography, prepacked RP-18 columns
(Redisepꢄ C-18, 13 g) were used using a gradient of water/MeCN
containing 0.1% of TFA (100% water 5 min, linear gradient to
100% MeCN within 55 min, then additional 10 min of pure MeCN).
TLC was carried out using 0.2 mm aluminum plates coated with
silica gel 60 F254 by Merck and the products were visualized by UV
detection or iodine. Solvents and reagents that are commercially
available were used without further purification unless otherwise
noted. All moisture-sensitive reactions were carried out using
oven-dried glassware under a positive stream of argon. Reactions
under microwave conditions were performed in the laboratory-mi-
crowave-system Discover (LabMate) from CEM.
1
a white solid; H NMR (500 MHz, [D6]DMSO): d=13.00 (s, 1H), 7.72
(dd, J=7.7 Hz, 1.4 Hz, 1H), 7.51 (dd, J=7.5 Hz, 0.9 Hz, 1H), 7.26 (t,
J=7.6 Hz, 1H), 2.15 (s, 3H), 1.93–1.88 (m, 1H), 1.05–1.01 ppm (m,
4H); 13C NMR (125 MHz, [D6]DMSO): d=172.2, 166.1, 148.6, 135.0,
131.8, 129.1, 125.7, 124.7, 15.8, 13.0, 8.6 ppm; MS (ES+) m/z (%):
238 (100) [M+NH4]+, 458 (67) [2M+NH4]+; HRMS (EI+): m/z [M]+
calcd for C12H12O4: 220.073559, found: 220.073060; Anal calcd for
C12H12O4: C 65.45, H 5.49, found: C 65.43, H 5.56.
Methyl 3-(acetylamino)-4-methylthiophene-2-carboxylate (26): A
solution of acetyl chloride (0.43 mL, 6.0 mmol) in CH2Cl2 (5 mL) was
added dropwise to a stirred solution of 25 (0.856 g, 5.0 mmol),
N,N-diisopropylethylamine (1.11 mL, 6.5 mmol) and DMAP (0.061 g,
0.5 mmol) in CH2Cl2 (15 mL) at ꢁ608C. The reaction mixture was
slowly allowed to warm to RT and then stirred for a further 12 h.
After addition of a saturated aq NH4Cl, the aqueous layer was sepa-
rated, and extracted three times with tert-butyl methyl ether. The
combined organic layers were dried over MgSO4 and filtered. Re-
moval of the solvent in vacuo followed by column chromatogra-
phy (hexane/EtOAc, 1:1) gave 0.847 g (80%) of 26 as a white solid;
mp: 1088C; 1H NMR (400 MHz, CDCl3): d=8.74 (bs, 1H), 7.13 (d,
J=0.9 Hz, 1H), 3.86 (s, 3H), 2.23 (s, 3H), 2.20 ppm (d, J=0.9 Hz,
3H); 13C NMR (125 MHz, CDCl3): d=168.5, 163.5, 142.9, 135.9,
127.4, 117.0, 51.8, 23.7, 15.5 ppm; MS (EI) m/z (%): 213 (98) [M]+,
170 (33), 169 (100), 139 (57), 138 (80); Anal calcd for C9H11NO3S: C
50.69, H 5.20, N 6.57, S 15.04 found: C 50.76, H 5.14, N 6.39, S
14.78.
2-(Acetyloxy)-3-chlorobenzoic acid (21): Concentrated H2SO4
(1 mL) was added to a stirred solution of 3-chloro-2-hydroxy-ben-
zoic acid 20 (0.518 g, 3 mmol) in Ac2O (25 mL). The reaction mix-
ture was heated to 1408C and stirred at this temperature for 2 h.
The reaction mixture was allowed to cool to RT and poured into
H2O. After saturating with NaCl, the aqueous layer was extracted
three times with EtOAc and the combined organic layers were
washed with water and brine, dried over MgSO4, filtered, and con-
centrated in vacuo. Column chromatography (isohexane/EtOAc/
formic acid, 5:1:0.1) afforded 0.28 g (43%) of 21 as a white solid;
1H NMR (400 MHz, [D6]DMSO): d=13.40 (bs, 1H), 7.89 (dd, J=
8.0 Hz, 1.6 Hz, 1H), 7.82 (dd, J=8.0 Hz, 1.6 Hz, 1H), 7.41 (t, J=
8.0 Hz, 1H), 2.31 ppm (s, 3H); 13C NMR (100 MHz, [D6]DMSO): d=
168.2, 165.0, 146.4, 134.0, 130.4, 127.9, 127.3, 126.4, 20.5 ppm; MS
(EI) m/z (%): 214 (10, 35Cl) [M]+, 216 (4, 37Cl) [M]+, 174 (45), 172
(92), 155 (36), 154 (100); HRMS (EI+): m/z [M]+ calcd for C9H7ClO4:
214.003287, found: 214.002527; Anal calcd for C9H7ClO4: C 50.37, H
3.29, Cl 16.52 found: C 50.82, H 3.65, Cl 16.06.
3-(Acetylamino)-4-methylthiophene-2-carboxylic acid (27): An
aqueous solution (5 mL) of LiOH (6.2 mmol) was added dropwise
to an ice-cold solution of 26 (660 mg, 3.1 mmol) in MeOH (10 mL)
and the mixture was stirred for 15 h at RT. The reaction was
quenched by adding a 1% aq HCl and the aqueous layer was ex-
tracted three times with EtOAc. The combined organic layers were
dried over MgSO4, filtered, and concentrated in vacuo. The residue
was further purified by reversed-phase chromatography and yield-
ed 0.528 g (79%) of 27 as a white solid; mp: 1678C; 1H NMR
(400 MHz, [D6]DMSO): d=12.86 (bs, 1H), 9.48 (bs, 1H), 7.43 (s, 1H),
2.014 (s, 3H), 2.012 ppm (s, 3H); 13C NMR (125 MHz, [D6]DMSO): d=
168.3, 162.7, 140.8, 136.8, 126.2, 123.1, 23.0, 14.3 ppm; MS (EI) m/z
(%): 199 (97) [M]+, 180 (82), 164 (71), 155 (100), 138 (94); Anal
calcd for C8H9NO3S: C 48.23, H 4.55, N 7.03, S 16.09 found: C 47.96,
H 4.69, N 7.02, S 16.15.
3-Methyl-2-(propionyloxy)benzoic acid (23): Concentrated H2SO4
(1 mL) was added to a stirred solution of 2-hydroxy-3-methylben-
zoic acid 22 (3.04 g, 20.0 mmol) in propanoic anhydride (6.51 g,
50.0 mmol). The reaction mixture was heated to 1208C and stirred
at this temperature for 5 h. After cooling to RT, the reaction mix-
ture was placed in a refrigerator for several weeks after which 23
solidified. The solid was obtained by filtration, washed with ben-
zene, and finally dried under vacuum to give 2.13 g (51%) of 24 as
General procedure A for the preparation of acylamino deriva-
tives 29a–c, 32 using the corresponding anhydrides: Amines 28
or 31 were dissolved in the corresponding anhydride (7–15 equiv)
1
a white solid; H NMR (500 MHz, [D6]DMSO): d=12.99 (s, 1H), 7.75
(dd, J=7.7 Hz, 1.4 Hz, 1H), 7.52 (dd, J=7.7 Hz, 0.9 Hz, 1H), 7.26 (t,
938
www.chemmedchem.org
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2010, 5, 930 – 940

Fragment-based Screening
and stirred at RT for the time given. Subsequently, the reaction
mixture was poured into ice water and the resulting white precipi-
tate was isolated by filtration, washed several times with ice water
and dried under vacuum. If the product did not precipitate by
pouring into water, the ice-water mixture was extracted with
EtOAc (3ꢂ20 mL). The combined organic layers were washed with
brine (40 mL) and saturated aq NaHCO3 (40 mL), dried over MgSO4,
filtered, and concentrated in vacuo. The residue was further puri-
fied by column chromatography.
2.53 ppm (s, 3H); 13C NMR (125 MHz, [D6]DMSO): d=157.8, 154.4,
146.1, 134.1, 126.7, 126.2, 126.0, 121.5, 20.3 ppm; MS (EI) m/z (%):
176 (100) [M]+, 160 (15), 146 (77), 117 (12), 90 (12); HRMS (EI+):
m/z [M]+ calcd for C9H8N2O2 176.058578 found 176.058802.
2-Ethyl-3-hydroxyquinazolin-4(3H)-one (30b): The desired com-
pound was prepared according to the general procedure B using
0.155 g (0.75 mmol) of 29b. After reversed-phase chromatography
(CH3CN/H2O), 0.059 g (41%) of 30b were obtained as white solid;
mp: 1218C; 1H NMR (400 MHz, [D6]DMSO): d=8.12 (d, J=8.0 Hz,
1H), 7.79 (t, J=8.2 Hz, 1H), 7.65 (d, J=8.2 Hz, 1H), 7.50 (t, J=
8.0 Hz, 1H), 4.71 (bs, 1H, overlaid by water), 2.88 (q, J=7.3 Hz, 2H),
1.27 ppm (t, J=7.3 Hz, 3H); 13C NMR (125 MHz, [D6]DMSO): d=
158.0, 157.7, 146.1, 134.1, 127.0, 126.3, 126.0, 121.4, 25.9, 10.5 ppm;
MS (EI) m/z (%): 190 (100) [M]+, 174 (41), 173 (89), 160 (42), 119
(20); HRMS (EI+): m/z [M]+ calcd for C10H10N2O2 190.074228 found
190.074646.
General procedure B for the preparation of quinazolinones
30a–c, 33: Amides 29a–c or 32, KOH (0.25 g, 4.5 mmol) and hy-
droxylamine hydrochloride (0.15 g, 2.25 mmol) were dissolved in
MeOH (3 mL). The reaction was carried out in a microwave pres-
sure vessel under microwave irradiation (conditions: 6 min, 1208C,
150 W). After cooling to RT, the mixture was acidified with TFA to
pH 4 and the isolated product mixture was purified by reversed-
phase chromatography. As a major by-product, the corresponding
noncyclic carboxylic acid was obtained.
3-Hydroxy-2-isopropylquinazolin-4(3H)-one (30c): According to
the general procedure B using 0.166 g (0.75 mmol) of 29c, 0.050 g
(33%) of 30c was obtained after reversed-phase chromatography
(CH3CN/H2O) as a white solid; mp: 1428C; 1H NMR (400 MHz,
[D6]DMSO): d=11.55 (bs, 1H), 8.12 (d, J=8.0 Hz, 1H), 7.79 (t, J=
8.0 Hz, 1H), 7.65 (d, J=8.0 Hz, 1H), 7.50 (t, J=8.0 Hz, 1H), 3.47
(sept, J=6.9 Hz, 1H), 1.29 (s, 3H), 1.27 ppm (s, 3H); 13C NMR
(125 MHz, [D6]DMSO): d=160.5, 158.1, 146.2, 134.0, 127.3, 126.2,
159.9, 121.3, 30.0, 20.1 ppm; MS (EI) m/z (%): 204 (100) [M]+, 188
(51), 187 (91), 176 (98), 173 (51); Anal calcd for C11H12N2O2: C 64.69,
H 5.92, N 13.72 found: C 64.45, H 5.98, N 13.77.
Methyl 2-(acetamino)benzoate (29a): The desired compound was
prepared according to the general procedure A (stirring time: 24 h)
using 1.214 g (8.03 mmol) of 28 and 7 mL of Ac2O. After precipita-
tion, 1.35 g (87%) of 29a were obtained as colorless crystals; mp:
938C; 1H NMR (400 MHz, CDCl3): d=11.06 (bs, 1H), 8.70 (d, J=
8.7 Hz, 1H), 8.03 (dd, J=8.0 Hz, J=1.6 Hz, 1H), 7.54 (ps dt, J=8.7,
J=1.6 Hz, 1H), 7.08 (ps t, J=8.0 Hz, 1H), 3.93 (s, 3H), 2.24 ppm (s,
3H); 13C NMR (125 MHz, CDCl3): d=168.6, 167.5, 140.0, 134.0,
130.6, 123.2, 121.2, 117.8, 52.5, 24.7 ppm; MS (EI) m/z (%): 193
(100) [M]+, 151 (92), 120 (55), 119 (81), 82 (51); Anal calcd for
C10H11NO3: C 62.17, H 5.74, N 7.25 found: C 61.95, H 5.71, N 7.13.
Methyl 3-(acetylamino)thiophene-2-carboxylate (32): According
to the general procedure A (stirring time: 5 h) using 5.50 g
(35.0 mmol) of 31 and 50 mL of Ac2O, 6.1 g (87%) of 32 was ob-
tained after precipitation as a white solid; mp: 1008C; 1H NMR
(400 MHz, CDCl3): d=10.15 (bs, 1H), 8.13 (d, J=5.5 Hz, 1H), 7.47
(d, J=5.3 Hz, 1H), 3.89 (s, 3H), 2.23 ppm (s, 3H); 13C NMR
(125 MHz, CDCl3): d=167.5, 164.7, 144.8, 131.5, 122.2, 109.7, 51.8,
24.4 ppm; MS (ES+) m/z (%): 200 (23) [M+H]+, 222 (58) [M+Na]+,
399 (100) [2M+H]+, 421 (74) [2M+Na]+; Anal calcd for C8H9NO3S: C
48.23, H 4.55, N 7.03, S 16.09 found: C 48.17, H 4.62, N 7.10, S
16.13.
Methyl 2-(propionylamino)benzoate (29b): According to the gen-
eral procedure A (stirring time: 36 h) using 1.21 g (7.0 mmol) of 28
and 7 mL of propionic anhydride, 1.32 g (79%) of 29b was ob-
tained after extraction with EtOAc, followed by column chromatog-
1
raphy (hexane/EtOAc, 9:1) as colorless crystals; mp: 338C; H NMR
(400 MHz, CDCl3): d=11.01 (bs, 1H), 8.67 (dd, J=8.5 Hz, J=0.9 Hz,
1H), 7.96 (dd, J=8.0 Hz, J=1.6 Hz, 1H), 7.47 (ps dt, J=8.5 Hz, J=
1.6 Hz, 1H), 7.00 (ps dt, J=8.0 Hz, J=0.9 Hz, 1H), 3.86 (s, 3H), 2.41
(q, J=7.6 Hz, 2H), 1.21 ppm (t, J=7.6 Hz, 3H); 13C NMR (125 MHz,
CDCl3): d=172.8, 168.7, 141.6, 134.6, 130.7, 122.2, 120.2, 114.7,
52.2, 31.6, 9.5 ppm; MS (EI) m/z (%): 207 (100) [M]+, 151 (90), 146
(46), 120 (46), 119 (67); Anal calcd for C11H13NO3: C 63.76, H 6.32, N
6.76 found: C 63.62, H 6.32, N 6.71.
3-Hydroxy-2-methylthieno[3,2-d]pyrimidine-4(3H)-one (33): Ac-
cording to the general procedure B using 0.149 g (0.75 mmol) of
32, 0.057 g (42%) of 33 was obtained after reversed-phase chroma-
tography (CH3CN/H2O) as a white solid; mp: 2188C; 1H NMR
(400 MHz, [D6]DMSO): d=11.71 (bs, 1H), 8.13 (d, J=5.3 Hz, 1H),
7.31 (d, J=5.3 Hz, 1H), 2.52 ppm (s, 3H, overlaid by NMR solvent);
13C NMR (125 MHz, [D6]DMSO): d=155.4, 155.0, 154.6, 135.4, 124.9,
121.0, 20.2 ppm; MS (EI) m/z (%): 182 (100) [M]+, 166 (48), 165 (56),
152 (77), 124 (44).
Methyl 2-(isobutyrylamino)benzoate (29c): The desired com-
pound was prepared according to the general procedure A (stirring
time: five days) using 1.214 g (8.03 mmol) of 28 and 7 mL of isobu-
tyric anhydride. After extraction with EtOAc followed by column
chromatography (hexane/EtOAc, 9:1), 1.37 g (77%) of 29c were
1
obtained as colorless crystals; mp: 578C; H NMR (400 MHz, CDCl3):
d=11.14 (bs, 1H), 8.76 (d, J=8.7 Hz, 1H), 8.03 (dd, J=8.0 Hz, J=
1.4 Hz, 1H), 7.54 (ps t, J=8.7 Hz, 1H), 7.07 (ps t, J=8.0 Hz, 1H),
3.93 (s, 3H), 2.63 (sept, J=7.1 Hz, 1H), 1.30 ppm (d, J=7.1 Hz, 6H);
13C NMR (125 MHz, CDCl3): d=176.0, 168.7, 141.8, 134.6, 130.7,
122.1, 120.2, 114.7, 52.2, 37.4, 19.4 ppm; MS (EI) m/z (%): 221 (14)
[M]+, 207 (100), 151 (98), 120 (34), 119 (74); Anal calcd for
C12H15NO3: C 65.14, H 6.83, N 6.33 found: C 64.91, H 6.81, N 6.21.
Thermolysin assay
Kinetic inhibition data of thermolysin (Calbiochem) was deter-
mined fluorimetrically using the quenched fluorescent dipeptide
Dabcyl-Ser-Phe-EDANS (2-N-(4-[4’-N’,N’-(Dimethylamino)phenylazo]-
benzoyl-l-serinyl-l-phenylalanylamido)-N’’-ethylaminonaphthalene-
5-sulfonic acid; N-Zyme, BioTech GmbH, Darmstadt, Germany).
Cleavage of the substrate was followed by monitoring the change
in fluorescence at lem =525 nm (excitation wavelength: lex =
336 nm).[54] The following conditions were used: 100 mm Tris/HCl,
pH 7.5, 2 mm CaCl2, 4% DMSO at 258C and different inhibitor con-
centrations. Initial reaction rates were measured, and IC50 values
were calculated from replicate curves using GraFit 4 software.[55]
3-Hydroxy-2-methylquinazolin-4(3H)-one (30a):[36,38,53] According
to the general procedure B using 0.145 g (0.75 mmol) of 29a,
0.041 g (31%) of 30a was obtained after reversed-phase chroma-
tography (CH3CN/H2O) as a white solid; mp: 2108C; 1H NMR
(400 MHz, [D6]DMSO): d=8.12 (dd, J=8.0 Hz, J=1.4 Hz, 1H), 7.79
(ps t, J=8.0 Hz, 1H), 7.62 (d, J=8.0 Hz, 1H), 7.50 (t, J=8.0 Hz, 1H),
ChemMedChem 2010, 5, 930 – 940
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.chemmedchem.org
939

G. Klebe et al.
MED
[23] M. Rarey, B. Kramer, T. Lengauer, G. Klebe, J Mol Biol 1996, 261, 470–
489.
[24] G. Jones, P. Willett, R. C. Glen, A. R. Leach, R. Taylor, J. Mol. Biol. 1997,
267, 727–748.
[25] M. L. Verdonk, J. C. Cole, M. J. Hartshorn, C. W. Murray, R. D. Taylor, Pro-
teins Struct. Funct. Bioinf. 2003, 52, 609–623.
[26] C. A. Sotriffer, P. Sanschagrin, H. Matter, G. Klebe, Proteins Struct. Funct.
Bioinf. 2008, 73, 395–419.
The overall error of the measurement is estimated to be ꢃ40%.
Ligand efficiency (LE) based on Ki values were calculated according
to the following equation:
ꢁRTꢄln ðKiÞꢄ0:239
LE [kcal]=
½number of heavy atomsꢅꢄ1000
Fragment 15 was obtained from ChemBridge (http://www.
chembridge.com).
[27] J. J. Irwin, F. M. Raushel, B. K. Shoichet, Biochemistry 2005, 44, 12316–
12328.
Acknowledgements
[28] R. Wang, Y. Lu, S. Wang, J. Med. Chem. 2003, 46, 2287–2303.
[29] M. A. Holmes, B. W. Matthews, Biochemistry 1981, 20, 6912–6920.
[30] A. F. Monzingo, B. W. Matthews, Biochemistry 1984, 23, 5724–5729.
[31] B. Baum, M. Mohamed, M. Zayed, C. Gerlach, A. Heine, D. Hangauer, G.
Klebe, J. Mol. Biol. 2009, 390, 56–69.
[32] S. W. Homans, Drug Discovery Today 2007, 12, 534–539.
[33] P. H. Olesen, Curr. Opin. Drug Discovery Dev. 2001, 4, 471–478.
[34] L. M. Lima, E. J. Barreiro, Curr. Med. Chem. 2005, 12, 23–49.
[35] M. F. Browner, W. W. Smith, A. L. Castelhano, Biochemistry 1995, 34,
6602–6610.
We acknowledge the support of the beamline staff at Bessy II,
Berlin, Germany and thank the Bundesministerium fꢁr Bildung
und Forschung (05 ES 3XBA/5) for financing the travel costs to
Bessy II. We thank Robert Jaeger, Sanofi-Aventis, Frankfurt for
providing fragmented structures of the WDI.
[36] R. W. Leiby, N. D. Heindel, J. Org. Chem. 1977, 42, 161–162.
[37] D. J. Mazera, J. C. Gesser, J. R. Pliego, ARKIVOC 2007, xv, 199–214.
[38] D. S. Reddy, G. Shailaja, P. P. Reddy, P. S. N. Reddy, Synth. Commun. 1996,
26, 2843–2846.
Keywords: drug discovery
inhibitors · medicinal chemistry · virtual screening · X-ray
diffraction
·
fragment-based design
·
[39] M. A. Holmes, B. W. Matthews, J. Mol. Biol. 1982, 160, 623–639.
[40] Z. Otwinowski, W. Minor, in Methods in Enzymology, Vol. 276, (Eds.: C. W.
Carter, Jr., R. M. Sweet), Academic Press, New York, 1997, pp. 307–326.
[41] A. T. Brꢃnger, P. D. Adams, G. M. Clore, W. L. DeLano, P. Gros, R. W.
Grosse-Kunstleve, J. S. Jiang, J. Kuszewski, M. Nilges, N. S. Pannu, R. J.
Read, L. M. Rice, T. Simonson, G. L. Warren, Acta Crystallogr. Sect. D: Biol.
Crystallogr. 1998, 54, 905–921.
[1] G. Siegal, E. Ab, J. Schultz, Drug Discovery Today 2007, 12, 1032–1039.
[2] N. Howard, C. Abell, W. Blakemore, G. Chessari, M. Congreve, S. Howard,
H. Jhoti, C. W. Murray, L. C. Seavers, R. L. van Montfort, J. Med. Chem.
2006, 49, 1346–1355.
[3] D. F. Wyss, A. Arasappan, M. M. Senior, Y. S. Wang, B. M. Beyer, F. G. Njor-
oge, M. A. McCoy, J. Med. Chem. 2004, 47, 2486–2498.
[4] P. G. Wyatt, A. J. Woodhead, V. Berdini, J. A. Boulstridge, M. G. Carr, D. M.
Cross, D. J. Davis, L. A. Devine, T. R. Early, R. E. Feltell, E. J. Lewis, R. L.
McMenamin, E. F. Navarro, M. A. O’Brien, M. O’Reilly, M. Reule, G. Saxty,
L. C. Seavers, D. M. Smith, M. S. Squires, G. Trewartha, M. T. Walker, A. J.
Woolford, J. Med. Chem. 2008, 51, 4986–4999.
[42] G. M. Sheldrick, T. R. Schneider, in Methods in Enzymology, Vol. 277 (Ed.:
C. W. Carter, Jr., R. M. Sweet), Academic Press, New York, 1997, pp. 319–
343.
[43] P. Emsley, K. Cowtan, Acta Crystallogr. Sect. D: Biol. Crystallogr. 2004, 60,
2126–2132.
[5] M. Oblak, S. G. Grdadolnik, M. Kotnik, R. Jerala, M. Filipic, T. Solmajer,
Bioorg. Med. Chem. Lett. 2005, 15, 5207–5210.
[6] C. A. Lepre, J. M. Moore, J. W. Peng, Chem. Rev. 2004, 104, 3641–3676.
[7] T. Neumann, H. D. Junker, K. Schmidt, R. Sekul, Curr. Top. Med. Chem.
2007, 7, 1630–1642.
[44] B. W. Matthews, Macromolecules 1972, 5, 818–819.
[45] S. Sheriff, ImmunoMethods 1993, 3, 191–196.
[46] S. Sheriff, W. A. Hendrickson, J. L. Smith, J. Mol. Biol. 1987, 197, 273–
296.
[47] J. Gasteiger, C. Rudolph, J. Sadowski, Tetrahedron Comput. Methodol.
1990, 3, 537–547.
[48] M. Hendlich, A. Bergner, J. Gꢃnther, G. Klebe, J. Mol. Biol. 2003, 326,
607–620.
[49] J. Gꢃnther, A. Bergner, M. Hendlich, G. Klebe, J. Mol. Biol. 2003, 326,
621–636.
[50] J. Gasteiger, M. Marsili, Tetrahedron 1980, 36, 3219–3228.
[51] D. A. Case, D. A. Pearlman, J. W. Caldwell, T. E. Cheatham III, J. Wang,
W. S. Ross, C. L. Simmerling, T. A. Darden, K. M. Merz, R. V. Stanton, A. L.
Cheng, J. J. Vincent, M. Crowley, V. Tsui, H. Gohlke, R. J. Radmer, Y. Duan,
J. Pitera, I. Massova, G. L. Seibel, U. C. Singh, P. Weiner, P. A. Kollman,
AMBER 7, University of California, San Francisco, 2002.
[52] S. J. Weiner, P. A. Kollman, D. A. Case, U. C. Singh, C. Ghio, G. Alagona, S.
Profeta, P. Weiner, J. Am. Chem. Soc. 1984, 106, 765–784.
[53] K. Tanaka, K. Matsuo, A. Nakanishi, Y. Kataoka, K. Takase, S. Otsuki,
Chem. Pharm. Bull. 1988, 36, 2323–2330.
[8] M. J. Hartshorn, C. W. Murray, A. Cleasby, M. Frederickson, I. J. Tickle, H.
Jhoti, J. Med. Chem. 2005, 48, 403–413.
[9] H. Nordstrçm, T. Gossas, M. Hamalainen, P. Kallblad, S. Nystrom, H. Wall-
berg, U. H. Danielson, J. Med. Chem. 2008, 51, 3449–3459.
[10] P. J. Hajduk, J. Greer, Nat. Rev. Drug Discovery 2007, 6, 211–219.
[11] C. Rummey, S. Nordhoff, M. Thiemann, G. Metz, Bioorg. Med. Chem. Lett.
2006, 16, 1405–1409.
[12] S. D. Pickett, B. S. Sherborne, T. Wilkinson, J. Bennett, N. Borkakoti, M.
Broadhurst, D. Hurst, I. Kilford, M. McKinnell, P. S. Jones, Bioorg. Med.
Chem. Lett. 2003, 13, 1691–1694.
[13] H. F. Velec, H. Gohlke, G. Klebe, J. Med. Chem. 2005, 48, 6296–6303.
[14] S. A. Latt, B. Holmquist, B. L. Vallee, Biochem. Biophys. Res. Commun.
1969, 37, 333–339.
[15] H. Drucker, S. L. Borchers, Arch. Biochem. Biophys. 1971, 147, 242–248.
[16] K. Morihara, Biochem. Biophys. Res. Commun. 1967, 26, 656–661.
[17] B. W. Matthews, Acc. Chem. Res. 1988, 21, 333–340.
[18] A. L. Hopkins, C. R. Groom, A. Alex, Drug Discovery Today 2004, 9, 430–
431.
[54] S. Weimer, K. Oertel, H. L. Fuchsbauer, Anal. Biochem. 2006, 352, 110–
119.
[55] R. J. Leatherbarrow, GraFit Version 4, Erithacus Software Ltd., Horley, UK,
1998.
[19] R. A. Carr, M. Congreve, C. W. Murray, D. C. Rees, Drug Discovery Today
2005, 10, 987–992.
[20] J. J. Irwin, B. K. Shoichet, J. Chem. Inf. Model. 2005, 45, 177–182.
[21] X. Q. Lewell, D. B. Judd, S. P. Watson, M. M. Hann, J. Chem. Inf. Comput.
Sci. 1998, 38, 511–522.
Received: February 26, 2010
[22] G. M. Morris, R. S. Halliday, D. S. Goodsell, R. Huey, W. E. Hart, R. E. Belew,
A. J. Olson, J. Comput. Chem. 1998, 19, 1639–1662.
Published online on April 14, 2010
940
www.chemmedchem.org
ꢀ 2010 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2010, 5, 930 – 940