An integrative in silico methodology for the identification of modulators of macrophage migration inhibitory factor (MIF) tautomerase activity

Article abstract of DOI:10.1016/j.bmc.2010.05.010

Macrophage migration inhibitory factor (MIF) is a major proinflammatory cytokine that has been increasingly implicated in the pathogenesis of several inflammatory, autoimmune, infectious and oncogenic diseases. Accumulating evidence suggests that the tautomerase activity of MIF plays a role in modulating some of its intra- and extra-cellular activities. Therefore, the identification and development of small-molecule inhibitors targeting the catalytic activity of MIF has emerged as an attractive and viable therapeutic strategy to attenuate its function in health and disease. Herein we report a novel virtual screening protocol for the discovery of new inhibitors of MIF's tautomerase activity. Our protocol takes into account the flexibility and dynamics of the catalytic site by coupling molecular dynamics (MD) simulations aimed at modeling the protein's flexibility in solution to (i) docking with FlexX, or (ii) docking with FlexX and pharmacophoric filtering with Unity. In addition, we applied in parallel a standalone docking using the new version of Surflex software. The three approaches were used to screen the ChemBridge chemical library and the inhibitory activity of the top-ranked 333 compound obtained from each approach (1000 compound in total) was assessed in vitro using the tautomerase assay. This biochemical validation process resulted in the identification of 12 novel MIF inhibitors corresponding to a 1.2% hit rate. Six of these hits came from Surflex docking; two from FlexX docking with MD simulations and four hits were identified with MDS and pharmacophore filtering with minimal overlap between the hits from each approach. Six hits were identified with IC₅₀ values lower than 10 μM (three hits with IC₅₀ lower than 1 μM); four were shown to be suicide inhibitors and act via covalent modification of the N-terminal catalytic residues Pro1. One additional inhibitor, N-phenyl-N-1,3,4-thiadiazol-2-yl-thiourea, (IC ₅₀ = 300 nM) was obtained from FlexX docking combined to pharmacophoric filtering on one of the eight MD structures. These results demonstrate the power of integrative in silico approaches in the discovery of new modulator of MIF's tautomerase activity. The chemical diversity and mode of action of these compounds suggest that they could be used as molecular probes to elucidate the functions and biology of MIF and as lead candidates in drug developments of anti-MIF drugs.

Full text of DOI:10.1016/j.bmc.2010.05.010

Bioorganic & Medicinal Chemistry 18 (2010) 5425–5440
Contents lists available at ScienceDirect
Bioorganic & Medicinal Chemistry
journal homepage: www.elsevier.com/locate/bmc
An integrative in silico methodology for the identification of modulators
of macrophage migration inhibitory factor (MIF) tautomerase activity
Farah El Turk a, , Bruno Fauvet a, , Hajer Ouertatani-Sakouhi a, Adrien Lugari b, Stephane Betzi b,
a,
Philippe Roche b, Xavier Morelli b, , Hilal A. Lashuel
*
*
a Laboratory of Molecular Neurobiology and Neuroproteomics, Brain Mind Institute, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland
b Centre National de la Recherche Scientifique (CNRS) & Aix-Marseille Universités, Institut de Microbiologie de la Méditerranée (IMM), Laboratoire Interactions et Modulateurs de
Réponses (UPR3243), 31 Chemin Joseph Aiguier, 13402 Marseille Cedex 20, France
a r t i c l e i n f o
a b s t r a c t
Article history:
Macrophage migration inhibitory factor (MIF) is a major proinflammatory cytokine that has been increas-
ingly implicated in the pathogenesis of several inflammatory, autoimmune, infectious and oncogenic dis-
eases. Accumulating evidence suggests that the tautomerase activity of MIF plays a role in modulating
some of its intra- and extra-cellular activities. Therefore, the identification and development of small-
molecule inhibitors targeting the catalytic activity of MIF has emerged as an attractive and viable ther-
apeutic strategy to attenuate its function in health and disease. Herein we report a novel virtual screening
protocol for the discovery of new inhibitors of MIF’s tautomerase activity. Our protocol takes into account
the flexibility and dynamics of the catalytic site by coupling molecular dynamics (MD) simulations aimed
at modeling the protein’s flexibility in solution to (i) docking with FlexX, or (ii) docking with FlexX and
pharmacophoric filtering with Unity. In addition, we applied in parallel a standalone docking using the
new version of Surflex software. The three approaches were used to screen the ChemBridge chemical
library and the inhibitory activity of the top-ranked 333 compound obtained from each approach
(1000 compound in total) was assessed in vitro using the tautomerase assay. This biochemical validation
process resulted in the identification of 12 novel MIF inhibitors corresponding to a 1.2% hit rate. Six of
these hits came from Surflex docking; two from FlexX docking with MD simulations and four hits were
identified with MDS and pharmacophore filtering with minimal overlap between the hits from each
Received 9 March 2010
Revised 29 April 2010
Accepted 4 May 2010
Available online 13 May 2010
approach. Six hits were identified with IC50 values lower than 10
lM (three hits with IC50 lower than
1
l
M); four were shown to be suicide inhibitors and act via covalent modification of the N-terminal cat-
alytic residues Pro1. One additional inhibitor, N-phenyl-N-1,3,4-thiadiazol-2-yl-thiourea, (IC50 = 300 nM)
was obtained from FlexX docking combined to pharmacophoric filtering on one of the eight MD struc-
tures. These results demonstrate the power of integrative in silico approaches in the discovery of new
modulator of MIF’s tautomerase activity. The chemical diversity and mode of action of these compounds
suggest that they could be used as molecular probes to elucidate the functions and biology of MIF and as
lead candidates in drug developments of anti-MIF drugs.
Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction
pivotal role in the regulation of inflammation and in host innate
and adaptive immune responses.1–5 Through a range of intra-
Macrophage migration inhibitory factor (MIF) is among the first
cytokines discovered.1,2 Initially identified as an inhibitor of the
random migration of macrophages as well as a major player in de-
layed-type hypersensitivity, MIF was later demonstrated to play a
and extra-cellular activities, the expression and secretion of MIF
modulate immune and inflammatory responses in health and dis-
ease.6–9 MIF plays a major role in macrophage activation and
phagocytosis,3,10–13 T-cell proliferation14 and antibody production
by B-cells14; it also promotes cell growth,15,16 and counter-regu-
lates glucocorticoid-induced cytokine suppression.17,18 Neutraliza-
tion of MIF activity using immunological, pharmacological and/or
genetic deletion approaches has shown beneficial results in a wide
range of inflammatory, infectious and oncogenic disease mod-
els,19–24 thereby underlying the crucial role that MIF plays in mod-
ulating the pathogenesis of these disorders and its potential as a
therapeutical target.6,23
Abbreviations: MIF, macrophage migration inhibitory factor; wt, wild-type; SEC,
size exclusion chromatography; LS, Light Scattering; MALDI, matrix-assisted laser
desorption ionization; MD, molecular dynamics; HTS, high-throughput screening;
F-Score, FlexX Score.
* Corresponding authors.
E-mail addresses: morelli@ifr88.cnrs-mrs.fr (X. Morelli), hilal.lashuel@epfl.ch
(H.A. Lashuel).
These authors contributed equally.
0968-0896/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmc.2010.05.010

5426
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
Unlike other cytokines, MIF functions as an enzyme and exhib-
its two catalytic activities: a keto-enol tautomerase activity (sub-
strates: -dopachrome and (hydroxy)-phenylpyruvate) (Fig. 1A
approaches. The majority of the early classes of MIF inhibitors
(e.g., acetaminophen, isoxazolines, and cinnamates) represent
mainly analogues and derivatives of its non-physiological sub-
strates.33,36,41–43
Until recently, the lack of a robust high-throughput tautomer-
ase assay, as a result of instability of its substrates, has precluded
any large-scale efforts to screen large chemical libraries. Therefore,
high-throughput screening (HTS) efforts have focused primarily on
structure-based computer assisted approaches (in silico).44 The
majority of the virtual docking methodologies applied to MIF were
based on a single X-ray structure of MIF from the PDB database35,44
and the dynamics of the catalytic pocket in solution was not taken
into account.
We hypothesized that a better understanding of the flexibility
of the catalytic site and how it contributes to ligands binding
would facilitate computer assisted drug design of small-molecules
inhibitors of MIF activity and would enhance the success rates of in
silico screening approaches. To test this hypothesis, we developed
an original in silico protocol for the discovery of novel inhibitors
targeting MIF’s active site. This virtual high-throughput screening
protocol takes into account the molecular dynamic parameters of
MIF as well as the structural and chemical properties of its catalytic
pocket. Our protocol combines three robust molecular modeling
methodologies: (1) high-throughput docking of virtual compounds
libraries with BioSolveIT’s FlexX software version 3.0.2 applied on
an ensemble of conformations generated by Molecular Dynamic
(MD) simulations with CHARMM; (2) a dynamic pharmacophore-
based hit selection with Tripos’s Unity software; in order to elim-
inate irrelevant docking poses and enforce the selection of ligands
interacting with key residues within the catalytic pocket of MIF
and; (3) a standalone docking with the latest Surflex (version
2.4) software using the geomX option that takes into account li-
gand flexibility during molecular docking. Our MD simulations
demonstrate that the active site of MIF is quite flexible and sug-
gested that the dynamic analysis of the protein enzymatic pocket
could provide insights that would facilitate the design of novel
MIF inhibitors. To validate the virtual screening protocol, we
performed a biochemical experimental assay on hit candidates ob-
tained in silico on a subset of the ChemBridge library correspond-
ing to the best-ranked 1–3% of ligands according to their docking
scores from (i) FlexX coupled to MD simulations, (ii) a combination
of FlexX coupled to MD simulations and Pharmacophore Filtering
D
and B) and thiol oxido-reductase activity (substrates: insulin and
2-hydroxyethyldisulfide (HED)). X-ray crystallography, NMR and
biophysical solution studies demonstrate that MIF exists predom-
inantly as a homotrimer.25–28 Structure-function relationship stud-
ies demonstrated that the keto-enol isomerisation activity of MIF is
catalyzed by the N-terminal proline residue (Pro1), and involves
residues from two adjacent subunits (Fig. 1C).29 Pro1 exhibits an
unusual pKa of 5.6 0.1 and is thus assumed to be un-protonated
under biological conditions, which allows it to serve as a catalytic
base.30 MIF shares a striking structural similarity with three bacte-
rial isomerases: 4-oxalocrotonate tautomerase (4-OT), 5-carboxy-
methyl-2-hydroxymuconate isomerases (5-CHMI) and chorismate
mutase.31 Like MIF, these enzymes are homotrimers (4-OT is a tri-
mer of homodimers) and belong to the tautomerase super family
characterized by a b-a-b motif and a catalytic amino-terminal pro-
line.31 Mutation of the catalytic Pro1 residue in MIF, 4-OT, and 5-
CHMI abolishes the tautomerase activity of these enzymes.31,32
Accumulated evidence suggest that the catalytic activities of
MIF play a role in modulating some of its intra- and extra-cellular
activities33,34 and its interactions with other proteins.35 Small-
molecule inhibitors of MIF tautomerase activity, such as ISO-1
((S,R)-3-(4-hydroxyphenyl)-4,5-dihydro-5-isoxazole acetic acid
methyl ester), were shown to protect against MIF proinflammatory
effects in experimental models of sepsis,36 and in mouse models of
endotoxemia.36 Inhibition of MIF tautomerase activity by ISO-1
inhibits the migration, invasion and anchorage-independent
growth of human cancer cell lines in vitro.34,37,38 Moreover, cata-
lytically inactive MIF mutants were unable to induce superoxide
production in neutrophils,30 and displayed reduced ability to en-
hance the matrix metalloproteinases (MMPs) mRNA level in syno-
vial fibroblasts from individuals with rheumatoid arthritis.7
Increased MMPs levels in serum and synovial tissue of a rheuma-
toid arthritis patient has been proposed as a marker of joint dam-
age and systemic inflammation.39,40 Based on these findings,
targeting the catalytic activity of MIF has emerged as an attractive
and viable therapeutic strategy to attenuate its function in health
and disease. The availability of extensive X-ray structural data on
MIF has facilitated and accelerated the design and discovery of
active site small-molecule inhibitors using structure-based
Figure 1. MIF is a keto-enol tautomerase. (A), (B): Schematic mechanism of MIF-catalyzed keto-enol tautomerase of D-dopachrome (A) and phenylpyruvate (B), showing the
role of the catalytic residue Pro1 as general acid/base. (C) X-ray structure of the MIF catalytic site. The van-der-Waals structure is shown in gray transparency; MIF monomers
are shown in cartoon representation with a different color for each (green, purple and brown). The labeled key residues involved in substrate binding and catalysis are shown
in yellow-carbon stick representation.

F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
5427
Table 1
(PF), or (iii) Surflex. Validation of our combined in silico protocol
resulted in the discovery of three novel potent inhibitors of MIF
Comparison of the available crystal structures of MIF in the PDB, and the eight
conformations obtained by MD simulations and used for pharmacophoric filtering
tautomerase activity, confirmed with IC50s 6 1
inhibitors were also validated with IC50 ranging from
>200 M. The mechanisms of action of these inhibitors were de-
lM. Nine other
Structures compared
RMSD mean (Å)
RMSD Stdev (Å)
1
to
l
1GD0/X-ray
1GD0/MDS
0.186
0.759
0.053
0.126
fined using a battery of biochemical and biophysical methods.
Our results suggest that coupling these different approaches devel-
oping an integrative in silico protocol should facilitate the identifi-
cation and discovery of novel classes of MIF inhibitors.
2. Results
For well-characterized protein targets, in silico virtual screening
provides rapid and low-cost complementary approaches to screen
large small-molecule chemical database to identify potential inhib-
itors of proteins activity in vitro and in vivo. Our in silico method-
ology combines the power of three robust computational
approaches: (1) molecular docking methods for a fast sampling
of the database in the enzyme structure(s), (2) MD simulations
which take into account the protein dynamics and; (3) pharmaco-
phore filtering that allows the modeler to pilot the sampling using
chosen features of the enzyme pocket. Each of these different com-
ponents was first separately optimized and validated using known
X-ray structures of MIF complexed with known inhibitors.
The ligand-free X-ray structure 1GD0 is very similar to structures in
which ligands were co-crystallized (1CA7, 1GCZ, 1LJT, 1MFI, 2OOZ),
but the RMSD between 1GD0 and the MD structures is significantly
higher (p = 2.1 ꢀ 10ꢁ7).
functions were available and tested: F-Score (FlexX’s scoring func-
tion), D-Score, G-Score, PMF-Score, and Chem-Score. The first step
in optimizing the virtual screening process was thus to select the
scoring function that best correlated the calculated score with the
crystallographic structures available. To this end, we selected en-
tries from the PDB in which MIF was co-crystallized with a known
substrate (4-hydroxyphenylpyruvate, PDB 1CA729) or inhibitors
(7-hydroxy-2-oxo-chromene-3-carboxylic acid ethyl ester, PDB
1GCZ,44 3-(4-hydroxyphenyl)-4,5-dihydro-5-isoxazole-acetic acid
methyl ester, PDB 1LJT,41 2-fluoro-3-(4-hydroxyphenyl)-2E-prope-
neoate, PDB 1MFI,42 and 4-hydroxybenzaldehyde o-(3,3-dimeth-
ylbutanoyl)-oxime, PDG 2OOZ49). We then re-docked each ligand
into its corresponding pocket, consisting of the same ensemble of
residues as specified in Materials and Methods (Table 2). The best
poses were those having an RMSD of less than 2.0 Å with respect
to the X-ray coordinates. For each receptor-compound complex,
we scored the best 30 poses according to each function mentioned
above, and chose as the best scoring function the one that gave the
best scores to the poses with the lowest RMSDs. We found that F-
Score gave higher scores to the poses that were most similar to
the crystal structure (data not shown).
Although scoring is a critical part of molecular docking, its main
purpose in our protocol was to generate reasonable conformations
of screened small-molecules inside the catalytic pocket of MIF.
From the database of poses computed by FlexX, candidate inhibi-
tors were selected through a pharmacophore-based filtering step.
A receptor-defined model of MIF’s tautomerase site was built with
the Unity software. Briefly, it consists of a ‘negative image’ of the
catalytic pocket’s features, such as hydrogen bond donors and
acceptors, in the form of 3D spatial constraints into which screened
compounds must fit to be selected as hits. The residues used to de-
fine the pharmacophore are detailed in Materials and Methods and
shown in Figure 2. We then optimized the tolerance of these spatial
features, that is, the allowed distance from an ideal pose. The best
tolerance is a compromise between the ability of the pharmaco-
phore to reject false positives and its capacity to retain as many true
positives as possible. To determine this value, we built a series of
pharmacophores with tolerances ranging from 0.5 to 2.0 Å for each
of the 55 MIF conformers obtained by MD simulations. The 55 con-
formers were first docked with the 5 positive control ligands (men-
tioned above) and 200 true negatives to generate 30 ꢀ 205 = 6150
poses. Each generated model (6150 poses) was then filtered
through the 55 pharmacophores corresponding to tolerances from
2.1. Molecular dynamics simulations
Structural models derived from X-ray diffraction of protein
crystals yield static coordinates. Therefore, in order to account
for the structural flexibility of a protein in solution, and to obtain
a reasonable ensemble of MIF conformations, we performed MD
simulations of MIF in explicit aqueous solvent by applying the
CHARMM force field45–48 and using the 1GD0 X-ray structure as
a starting template. Eight independent simulations were per-
formed with different conditions that varied according to temper-
ature (300 or 333 K), presence or absence of crystallographic water
molecules, and solvation (see Materials and Methods section and
Supplementary Table S1). Our objective was to obtain the highest
diversity possible of reasonable MIF conformations to be used as
receptors in docking experiments. Coordinates were saved every
500 ps for subsequent analysis leading to a total of 32,000 confor-
mations for the 16 ns simulations. The overall protein structure
was stable during the time course of the trajectory, with an average
root mean square deviation over CA atoms of 1.05 0.24 Å. In order
to reduce the number of conformations to a reasonable set to be
used as receptors in high-throughput docking calculations, we per-
formed a clustering analysis of the MD ensemble. On the basis of
initial visual inspection of the trajectories, we found that in a
majority of the MD snapshots, the backbones and side chains of
residues surrounding (Lys32, Phe113) and inside (Pro1) the active
site blocked the entrance of the active site, thus preventing any
docking. Therefore, the clustering analysis was aimed at separating
the ‘open’ from ‘closed’ conformations. The criteria used were a set
of distances described in Supplementary Table S2. We obtained 78
structures fulfilling these criteria, which we visually inspected in
detail. From these, we selected 55 diverse MIF structures that were
used to optimize the virtual HTS protocol (Table 1).
2.2. Optimization of the virtual screening protocol
Whereas MD simulations were used to model the protein’s flex-
ibility in solution, docking was used to account for the possibility
that bound ligands may adopt conformations distinct from their
minimum free energy states. Docking also served the purpose of
ranking the ligands according to a scoring function. Several scoring

5428
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
Table 2
Structures and binding modes of positive control ligands that were co-crystallized with MIF
Ligand
Structure
Residue:atom
p-Hydroxyphenylpyruvate (1CA7)
N97:OD1, N97:HD21, N97:HD22, K32:HZ1-3, I64:HN
Chromene (1GCZ)
N97:OD1, N97:HD21, N97:HD22, K32:HZ1-3, I64:HN
ISO-1 (1LJT)
N97:OD1, N97:HD21, N97:HD22, K32:HZ1-3
2-FHC (1MFI)
N97:OD1, N97:HD21, N97:HD22, K32:HZ1-3, P1:N
OXIM-6 (2OOZ)
N97:OD1, N97:HD21, N97:HD22, Y95:OH, Y36:O
For each ligand, the key protein residues interacting with these compounds are shown, along with the respective atoms involved.
A
B
M101
K32
I64
N97
I64
P1
Y36
A38
C
D
5 True Posiꢀves
Tolerance (Å)
Receptor Structure
Figure 2. Optimization of the pharmacophore-based filter. (A) Three-dimensional representation of the spatial constraints derived from key residues of MIF’s catalytic pocket.
Acceptor sites are shown in blue, donor sites are shown in red. (B) List of all residues/atoms used to build the pharmacophore model. (C) Optimization of the tolerance of the
pharmacophoric filter. The ratio between the proportions of true positive ligands retrieved over the fraction of mistakenly negative ligands selected is represented for each
tolerance value, showing an optimum at 0.8 Å followed by a decrease in specificity. (D) Analysis of the pharmacophore’s performance for each of the MIF structures used in
the virtual screening. Each bar represents the number of positive ligands retrieved by each structure (the scale is represented by a vertical bar on the figure). Tolerance values
from 0.6 to 1.4 Å are mapped to the bar colors. Note that at tolerance 0.8 Å all five of the true positives are retrieved with the eight MD structures.

F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
5429
0.5 to 2 Å. The lowest tolerance for which all five controls were re-
trieved with the lowest number of true negatives was selected. Fig-
ure 2C shows the number of retrieved true positive hits as a
function of pharmacophore tolerance. With this procedure, we ob-
tained an optimal tolerance of 0.8 Å. This step allowed us to further
reduce the number of MIF conformers to be used in the final virtual
screening. At the optimal tolerance of 0.8 Å, we found that eight MD
structures (with their corresponding pharmacophore) were suffi-
cient to retrieve the set of five structurally diverse control ligands
(Fig. 2D). These eight structures (Fig. 3) were chosen as templates
for our screening protocol, in addition to the X-ray structure. The
remaining structures (47) required higher tolerances and were
discarded.
tautomerase pocket of the protein. After eliminating duplicates
(i.e., conformers of the same compound), we obtained 5185 unique
potential hits.
Previous studies by our group have demonstrated robust results
using Surflex as docking software. After a validation process that
consisted of checking that Surflex could retrieve all the correct
poses (RMSD <2 Å) from the single docking with the different X-
ray structures (1CA7, 1GCZ, 1LJT, 1MFI and 2OOZ), Surflex high-
throughput docking was carried out on MIF apo-structure (PDB
code 1GD0) using the same amino acids as described for the FlexX
docking to generate the ‘Surflex protomol’ (Surflex priority area of
docking) with standard parameters (threshold 0.5, no bloat). It is
noteworthy that at this stage only a very small overlap was ob-
served between the 333 hits first selected by Surflex and those se-
lected either from FlexX on the nine representative structures or
those selected from FlexX on the nine representative structures fil-
tered by the pharmacophore approach. This validates the necessity
of coupling the best 1% from each approach into a global in silico
selection that should increase the chemical diversity of our result-
ing hits.
2.3. Virtual high-throughput screening
With the optimized in silico pipeline (F-score as scoring func-
tion, 1GD0 + eight MD protein conformers and tolerance ratio of
0.8 Å), we proceeded to screen for novel candidate MIF inhibitors
from the Chembridge library, comprising ꢂ500,000 compounds.
We reduced this number to 33,900 by applying a restriction of
30 or fewer non-hydrogen atoms, based on the observation that
the binding pocket can only accommodate low molecular weight
compounds and that known MIF inhibitors always fulfill this
condition.35 Docking experiments also showed that larger mole-
cules are generally bound outside the active site (data not
shown).
The 33,900 molecules database was docked against each of the
eight ‘optimal’ MIF conformers obtained by MD simulations, plus
the 1GD0 structure (Fig. 4). The 30 best poses for each compound
were selected according to the FlexX scoring function (F-Score)
resulting in ꢂ9,000,000 structures (33,900 ꢀ 9 structures ꢀ 30
poses). Each generated model was then subjected to our pharma-
cophore-based filtration using the optimized tolerance value of
0.8 Å on the nine templates (1GD0 + eight MD structures). We con-
sidered as potential hits the structures matching at least two pro-
tein acceptor sites and one protein H-bond donor site, with the
condition that one of these matched features must be at the
interior of the catalytic site (Fig. 2A and B). At this stage, our objec-
tive was to reject poses that were docked onto the rim of the
2.4. Experimental HTS evaluation of the best 1000 compounds
obtained using our integrative in silico protocol
To validate our in silico protocol(s) and to test the inhibitory ef-
fects of the highly-ranked molecules identified, we chose the best-
scored 1000 molecules to be experimentally validated via the HTS
protocol employing the tautomerase assay recently developed in
our laboratory.50 The 1000-compounds set consisted of the follow-
ing: (i) best 333 scored-molecules selected using the ‘Docking
standalone’ (Surflex) protocol; (ii) best 333 scored-molecules se-
lected using the ‘Docking + MDS’ (FlexX score + CHARMM), but
not retrieved by Unity filtering; and (iii) best 334 scored-molecules
selected using the ‘Docking + MDS + Unity Pharmacophore Filter-
ing’ (FlexX score + CHARMM + Unity).
D-dopachrome substrate
was selected for the screening. As a positive control, we used ben-
zyl isothiocyanate (BITC), which we recently reported as a potent
inhibitor of MIF that leads to complete inhibition of MIF tautomer-
ase activity at 10
were determined through calculations of the intra- and inter-plate
l
M.52 Robustness and reproducibility of the assay
Figure 3. MIF tautomerase catalytic site exhibits high conformational flexibility. (A) Orientation of key residues of the active site in the eight MD conformations
superimposed on the X-ray structure (1GD0). The van-der-Waals structure is shown in gray transparency; MIF monomers are shown in cartoon representation with a
different color for each (green, purple and brown). (B) Active site shown for the eight MD conformations. Each conformation is presented in a different color (from red to blue).
Chains B and C surfaces are represented in green and pale orange, respectively.

5430
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
Figure 4. Workflow of the virtual HTS. The ChemBridge library was first filtered down to compounds with fewer than 30 non-hydrogen atoms based on previous observations
that larger ligands would have less chance to fit into MIF’s active site. Protein flexibility was taken into account in FlexX docking by using different conformations derived
from MD simulations. In Surflex, one receptor structure is used (1GD0), and ligand flexibility is taken into account by defining the geomX option. The best 333–334
compounds from each category, corresponding to a total of 1000 compounds were validated by experimental HTS.
variations using single-point Z0 measurements, which yielded a Z0
of P0.9 for the assay.
(Fig. 5, Table 3) and 9 others with IC50s ranging from 1 to
>200 M (Table 3).
l
The 1000 compounds were screened at two different concentra-
tions in 384 multi-well plates; the assay was performed as recently
described in Ouertatani-Sakouhi et al.50 Briefly, the protein was
2.5. Mode of action of MIF antagonists
incubated with 10 or 100
DMSO percentage 6 1.5%). The reaction was then initiated by the
l
M of compound for an hour at RT (final
To assess the mode of action of the validated hits (i.e., excluding
hits 5, 9, and 14) obtained in our screen, we performed a series of
biochemical and biophysical studies to determine whether the
inhibitors exert their effects by one of the following mechanism:
(1) covalent modification of the N-terminal catalytic Pro1 residue;
(2) direct binding to the active site; (3) alteration of the tertiary
structure of MIF, or; (4) disruption of the trimer. Each hit com-
pound was incubated with MIF (1/1 molar ratio) at rt for one hour
and the reaction mixtures were evaluated by MALDI/TOF–TOF
mass spectrometry, size exclusion chromatography, light scatter-
ing and SDS–PAGE analysis. None of the compounds tested caused
disruption of the trimer or affected the solubility of MIF (data not
shown). To determine whether the selected compounds covalently
modify MIF, the reaction mixtures were analyzed by MALDI/TOF–
TOF tandem mass spectrometry. Potent hits obtained by Surflex
standalone docking, hits 1 (IC50 = 50 nM) and 3 (IC50 = 80 nM), as
addition of
D
-dopachrome. It is noteworthy that DMSO at 62%
has no effect on MIF tautomerase activity.50 In the absence of
MIF, the absorbance of the substrate at 475 nm is around 0.9–1.
Upon addition of MIF, the absorbance decreases to ꢂ0.1 within
3 min. Inhibitor potencies were measured by recording the absor-
bance at the end point (2 min). Hit compounds showing greater
than 15% inhibition at 100 lM were selected and validated manu-
ally to eliminate the false positives (Table 3). A further validation
was performed, in order to determine whether some of the con-
firmed inhibitors could act via an aggregation-based, non-specific
mechanism. The assay was adapted from Feng and Shoichet.51
Briefly, it consists of comparing the IC50s measured in absence
and presence of low amounts of non-ionic detergent. Compounds
that show a strong reduction in inhibition upon addition of deter-
gent are likely to be aggregators and are not counted as computa-
tional successes, as they are not true binders. We chose to use 0.1%
v/v Triton X-100 as detergent in both assay buffer and enzyme
stock solution, after we showed this concentration of Triton has
no effect on MIF’s tautomerase activity (data not shown). The re-
sults are shown in Figure 5. Addition of detergent fully reversed
inhibition in the case of hits 5, 9, and 14, indicating a probable
mechanism based on a phase transition of these inhibitors. Among
the remaining compounds, three were confirmed with
well as hits 4 (IC50 = 2.8 lM) and 6 (IC50 = 6.9 lM) obtained by
docking with FlexX on MD structures, resulted in a shift in the
molecular mass of MIF (Fig. 7) that was consistent with a single
modification of MIF by these compounds. To identify the exact res-
idue modified by these inhibitors, the modified MIF species were
subjected to tryptic digestion and peptide mapping by MALDI/
TOF–TOF mass spectrometry. In every case, we observed a molec-
ular weight shift corresponding to a modified N-terminal fragment
comprising residues PMFIVNTNVPR (MW = 1287.6 Da) (Fig. 7). MS/
MS analyses and sequencing of the modified peptide fragment
IC50s 6 1 lM (hit 1: 50 nM, hit 2: 300 nM, and hit 3: 80 nM)

F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
5431
Table 3
A summary of the inhibitors identified in our screening, including their code numbers, chemical names and structures, determined IC50 values and structures of the validated
inhibitors
Hit
names
Origin
Catalog
ID
Compound name
MW (g/
mol)
IC50
inhibition (%)
(
l
M) or max.
Structure
1
Surflex 7956004 6-Phenyl-3-pyridazinyl 2-thiophenecarboxylate
282.3
0.05 lM
2
3
Unity
5484950 N-Phenyl-N0-1,3,4-thiadiazol-2-ylthiourea
236.3
251.3
0.3 lM
Surflex 5320163 S-1,3-Benzoxazol-2-yl O-butyl thiocarbonate
0.08
l
M
4
5
6
FlexX
6205546 N-[(2-Bromobenzoyl)oxy]-4-chlorobenzamide
354.6
403.4
253.3
2.8
l
M
5-({2-Nitro-4-[(trifluoromethyl)sulfonyl]phenyl}thio)-1,3,4-
thiadiazole-2(3H)-thione
Surflex 6367607
Detergent-sensitive
FlexX
7291474 1-(2-Methoxybenzoyl)-1H-1,2,3-benzotriazole
Methyl {[4-methyl-5-(2-thienyl)-4H-1,2,4-triazol-3-
6.9 lM
7
8
Surflex 9049826
269.3
204.2
18.6
45.2
l
M
yl]thio}acetate
Surflex 5113769 3-Acetyl-7-hydroxy-2H-chromen-2-one
l
M
9
FlexX
6598745 N-(2-Furylmethyl)-7-nitro-2,1,3-benzoxadiazol-4-amine
260.2
233.2
Detergent-sensitive
10
Surflex 5315866 Ethyl 2-cyano-3-(2,4-dihydroxyphenyl)acrylate
62.1 lM
11
12
13
14
15
Unity
Unity
5185917 N-(1H-1,2,3-Benzotriazol-1-ylmethyl)-2-pyridinamine
225.3
263.8
263.2
222.2
266.3
69% (200 lM)
7847308 N-(2-Chlorophenyl)-N0-2-pyridinylthiourea
1.04
57.6
l
M
M
3-[(2,4-Difluorophenyl)amino]-1-(5-methyl-2-furyl)-2-propen-
1-one
Surflex 7509861
l
FlexX
Unity
5140738 3-Nitroso-2-phenylindolizine
Detergent-sensitive
6150973 3-Benzoyl-7-hydroxy-2H-chromen-2-one
64.2 lM
revealed that modification by these compounds takes place exclu-
sively at the N-terminal proline residue.
level of tertiary structure; this mixture elutes from the SEC column
as two peaks, each of which corresponds to the MIF trimer.28 More
recently we showed, using NMR, SEC, and light scattering, that
covalent modification of MIF by isothiocyanates (ITCs) alters the
tertiary structure of the trimer and results in a shift in the elution
profile of MIF, causing it to elute earlier as a single peak.52 To deter-
mine the effect of the selected hit compounds on MIF tertiary and
quaternary structures, the reaction mixtures (MIF + hits) were
2.6. Covalent modification of MIF by hits 1, 3, 4 and 6 alter the
tertiary structure and SEC elution profile of the trimer
Previous studies from our laboratory demonstrated that MIF ex-
ists as a mixture of two different trimeric states that differ at the

5432
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
Figure 5. Dose–response curves (mean S.D.; n = 4) obtained by monitoring the end point absorbance of the
reaction. Hit compounds were incubated with 500 nM wt huMIF for 2 h at room temperature, at concentrations either from 24 nM to 50
200 M (hits 6–15), followed by the addition of 533 M of -dopachrome., in the presence (dotted lines) or absence (solid lines) of 0.1% v/v Triton X-100. The absorbance at
D-dopachrome keto form, that is, after 2 min of enzymatic
l
M (hits 1–5), or from 98 M to
l
l
l
D
475 nm was then monitored for 2 min. When applicable, IC50 values are given as pairs, the first and second one being measured in absence and in presence of detergent,
respectively. N.D., not determined.
examined by analytical size exclusion chromatography coupled to
a multiangle light scattering detector. We observed that inhibition
and covalent modification of MIF with hits 1, 3, 4 and 6 led to a ma-
jor shift in the MIF elution profile, causing it to elute as a single
peak with an earlier elution time (between 28 and 30 min) than
that of unmodified MIF, which eluted as two overlapping peaks
with elution times of 30.5 and 32.5 min. These changes in the
elution profile of MIF are similar to those we observed after

F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
5433
Figure 6. Best docking poses of validated hits identified by Surflex: hit 1 (A), hit 3 (B), and pharmacophore-based hit selection with Tripos’s Unity software, hit 2 (C). Chains B
and C are represented in green and gold, respectively. The compounds are represented in the best score most populated conformational state.
incubation of the protein with benzyl isothiocyanate (BITC) and
other isothiocyanate-based inhibitors of MIF52 (Fig. 8). It is note-
worthy, that the degree of peak shift correlated well with the po-
tency of the inhibitor and with the extent of MIF modification by
each inhibitor as determined by mass spectrometry. MIF incubated
with hits 1 (IC50 = 50 nM) and hit 3 (IC50 = 80 nM) eluted at
similar properties as that of MIF,53 we observed mass shifts of
109 and 100 Da upon incubation with the hits 1 and 3, respec-
tively. These findings are expected and consistent with the highly
nucleophilic properties of Pro1 in MIF and 4-OT.
3. Discussion
ꢂ28 min, whereas MIF incubated with hit 6 (IC50 = 11.86
l
l
M)
M)
eluted ꢂ29.5 min, and MIF incubated with hit 4 (IC50 = 2.8
MIF has emerged as a major therapeutic target in a wide range
of inflammatory, autoimmune and oncogenic diseases. Current
therapeutic strategies for neutralizing MIF activity are focused on
immunoneutralization using anti-MIF antibodies or on pharmaco-
logical approaches aimed at targeting the catalytic site. Although
both strategies have shown beneficial effects in different experi-
mental models of inflammatory, infectious and autoimmune dis-
eases,34,54–56 the two strategies target different biochemical and
cellular activities of MIF. Anti-MIF antibodies are effective in neu-
tralizing MIF activity by blocking its interaction with other pro-
teins and receptors. However, the tautomerase activity of MIF is
not significantly affected upon formation of the MIF-antibody com-
plex (Ouertatani-Sakouhi H and Lashuel H, unpublished results).
By contrast, several studies have shown that small-molecule inhib-
itors of MIF tautomerase activity, while targeting the active site,
can induce significant tertiary structure changes in the trimer such
that antibody recognition of MIF and its receptor binding proper-
ties are altered.35 These findings suggest that different active site
inhibitors may have different effects on the intra- and extra-cellu-
lar activities of MIF, depending on how they modify the tertiary
and/or quaternary structure of the protein.
eluted at ꢂ30 min. Other hits (e.g., hit 3) did not affect MIF’s elu-
tion profile, despite their potency, suggesting that their binding
does not affect MIF’s tertiary and conformational structure.
In summary, we have validated a total of 12 hits, out of 1000
compounds tested, and corresponding to a 1.2% hit rate. Six of
these hits come from Surflex docking, two from FlexX docking with
MDS and four hits were identified with MDS and Unity (Tables 3
and 4). In another experimental HTS of the Maybridge database
using the same experimental protocol, we validated 15 hits out
of 14,400 compounds tested, corresponding to a 0.104% hit rate.
We also screened the FDA library and have validated 2 hits out
of 1400 compounds, corresponding to a 0.19% hit rate. It should
be noted here that the FDA library consists of compounds that re-
ceived FDA approval; these are smaller than those found in the
Maybridge library and thus it is not surprising that we validated
more hits in a blind test from this library. We should thus expect
a 0.104% hit rate from a blind screening of the Chembridge library.
It follows that when comparing our integrative in silico approach
versus with blind experimental screening the global enrichment
factor is around 12 (Table 4). This result demonstrates the high
productivity of in silico procedures in such biological systems.
The majority of MIF active site inhibitors developed to date
were identified by screening substrate analogues or derivatives
of known MIF substrates and inhibitors.33,36,41–43 The lack of a ro-
bust high-throughput assay, as a result of the low stability of
known MIF substrates, has precluded the screening of large chem-
ical libraries. Computational and molecular docking approaches
provide unique opportunities for high-throughput screening and
exploration of large chemical landscapes, including chemical
spaces that are not yet accessible by synthetic chemistry, with
minimal time and resources. To identify diverse classes of inhibi-
tors, several groups employed computational approaches and
docking of chemical libraries to the tautomerase active site. In
2002, Orita et al. performed a virtual screening of one million com-
pounds via molecular docking with the DOCK4.0.1 docking program.
Five hundred twenty-four compounds were then selected, pur-
chased and tested in vitro57; fourteen inhibitors with Ki values
2.7. Specificity of the suicide inhibitors
To determine whether covalent modification of MIF’s Pro1 res-
idue by hits 1, 3, 4, and 6 was specific to MIF, and possibly requires
a specific binding to the active site, we incubated 10
l
M of each
M)
inhibitor with (1) four structurally unrelated proteins (at 10
l
that contain lysine and/or cysteine residues: RNAse A, lysozyme,
chymotrypsinogen A and ubiquitin (Supplementary Fig. S4C–F);
and (2) the bacterial MIF homolog 4-oxalocrotonate tautomerase
(4-OT), (Supplementary Fig. S4B). The compound–protein mixtures
were incubated at room temperature for 1 h in MIF D-dopachrome
assay buffer. None of the four inhibitors (hits 1, 3, 4, and 6) was ob-
served to modify RNAse A, lysozyme, chymotrypsinogen A and
ubiquitin as determined by mass spectrometry (Supplementary
Fig. S4). These findings support the specificity of these molecules
for MIF and suggest that covalent modification of Pro1 by these
compounds is likely to be mediated by their binding to the
catalytic site and/or the high nucleophilicity of Pro1. In the case
of 4-OT, which shares a catalytic N-terminal proline residue with
ranging between 0.038 and 7.4
ner and colleagues identified 4-iodo-6-phenylpyrimidine, 4-IPP
(IC50 = 5 M), as a suicide inhibitor of MIF tautomerase activity,
via virtual screening using Ludi (Accelrys software).34 To discover
lM were validated. Recently, Win-
l

5434
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
Figure 7. MALDI/TOF-TOF mass spectrometric analyses carried out on modified protein and trypsin-digested samples to map the binding site of modification. Wt huMIF
alone (A); wt huMIF incubated with compound 1 (B), compound 3 (C), hit 6 (D) and hit 4 (E). The theoretical mass of the unmodified tryptic peptide containing Pro1 residue
(PMFIVNTNVPR) is 1287.5 g/mol. The fragments of the compounds that remain covalently bound to Pro1 are shown encircled in red in the respective structures.
new MIF-CD74 interaction antagonists targeting the tautomerase
active site, Cournia et al. performed virtual HTS by docking 2.1
million compounds, from Maybridge and Zinc databases that had
been pre-filtered for properties based on Lipinski rules. Molecular

F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
5435
Figure 8. Hits’ effects on the oligomeric and tertiary states of MIF. MIF alone or MIF incubated with inhibitor, were injected onto analytical size exclusion column (Superdex
75 10/30), and analyzed by light scattering upon elution. MIF elutes as two peaks at 30.5 and 32.5 min. Covalent inhibitors lead to a shift in MIF’s elution profile: BITC, hits: 1,
3, 6 and 4 lead to MIF’s earlier elution as one peak; hits 11 and 13 lead to a change in the ratio Peak 1/Peak 2; all other inhibitors do not affect MIF’s elution profile.
Table 4
Comparative evaluation of hit rates obtained for experimental HTS and by the in silico mixed methodology approach
Protocol
Library
Tested compounds
Hits
Hit rate (%)
0.105
Enrichment factor
Experimental HTS
Experimental HTS
Maybridge
FDA
ChemBridge
14,400
1040
333
334
333
15
2
12
1
0.192
1.2
1.84
In silico combined approach
MDS/FlexX
MDS/UnityÒ
Surflex-Dock
2
4
6
0.6
1.2
1.8
11.43
5.71
11.43
17.14
The enrichment factor for the in silico protocol combining three parallel strategies was increased by a factor of 11.43 compared to the Maybridge experimental screening.
structures were docked and scored using the Glide docking pro-
gram. Twenty six molecules were then selected by visual inspec-
tion of the 1200 top-scoring molecules, and eleven were
confirmed as MIF tautomerase inhibitors effective in the micromo-
allel with a standalone docking using the new version of Surflex
software (2.4), and the X-ray structure of ligand-free MIF according
to the PDB 1GD0 coordinates. Given the small chemical overlap be-
tween the compounds selected from these three approaches, we
decided to select the top-ranked 1% identified by each method to
increase the chances of obtaining highly potent hits.
An in silico HTS of Chembridge library was performed using
the protocol described above. Finally, validation of the hits using
tautomerase assays was then performed on the 333 best-scored
molecules obtained by FlexX docking on MD structures, as well
as on the 334 top-scored molecules obtained by FlexX and by
the Unity pharmacophore, and the 333 top-scored Surflex docked
compounds, for a total of 1000 compounds. This biochemical val-
idation process resulted in the identification of 12 novel MIF
inhibitors using FlexX + MDS (2), or FlexX + MDS + Unity (4) and
Surflex (6). These results suggest that coupling these different
approaches developing an integrative in silico protocol should
facilitate the identification and discovery of novel classes of
MIF inhibitors. Three hits were identified using Surflex Docking
lar range, with four compounds having IC50 as low as 0.5–5 l
M.35
These results highlight the potential of the virtual screening ap-
proach for the discovery of novel and potent inhibitors of MIF cat-
alytic and biological activities.
Until now, in silico-based approaches applied to MIF have used
only one set of X-ray coordinates corresponding to a particular PDB
entry.34,35,44,57 Therefore, the dynamics of the MIF catalytic pocket
in solution have not previously been taken into consideration.
Nonetheless, NMR studies from our laboratory (Ouertatani-Sak-
ouhi et al., 2010 and El-turk et al., 2010) and our MD simulations
performed on MIF showed high dynamic fluctuations of the cata-
lytic pocket (Fig. 3 and Table 1). We hypothesized that a better
understanding of the flexibility of the catalytic site and how it con-
tributes to ligands binding would facilitate computer assisted drug
design of small molecule inhibitors of MIF activity and would en-
hance the success rates of in silico screening approaches. To test
this hypothesis, we sought to develop an in silico protocol that al-
lows better consideration of these parameters when screening
large databases of chemical compounds for more potent and di-
verse inhibitors of MIF tautomerase activity. Towards this goal,
we generated a novel and integrative virtual screening protocol
that combines MD simulations aimed at modeling the protein’s
flexibility in solution, which we coupled to (i) docking with FlexX,
or (ii) docking with FlexX and pharmacophoric filtering with Unity.
Eight MD structures were selected based on their ability to retrieve
all known inhibitors at the best tolerance; this tolerance (0.8 Å)
represents a compromise between the ability of the pharmaco-
phore to reject false positives and its capacity to retain as many
true positives as possible (Fig. 2). This protocol was applied in par-
with IC50 values lower than 10
lM (two hits with IC50 lower than
1
lM: hits 1 and 3, Fig. 6A and B); these two were shown to be
suicide inhibitors that act via covalent modification of the N-ter-
minal catalytic residue Pro1, which is essential for MIF tautomer-
ase activity. The activity of these compounds mirrors that of
other Pro1 modifiers, N-acetyl-p-benzo-quinone imine (NAPQI,
acetaminophens),33 4-iodo-6-phenylpyrimidine (4-IPP)34 and iso-
thiocyanates.52 Interestingly, studies from our group have shown
that only catalytically active trimers undergo modification by
these compounds at Pro1 in vitro and in the cell.52,58 These stud-
ies highlight the reactivity of the N-terminal proline, Pro1, and
provide a basis of novel strategies for specific detection and
quantification of catalytically active MIF and targeting of its cat-
alytic activities.

5436
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
One additional inhibitor, N-phenyl-N0-1,3,4-thiadiazol-2-
a potent tautomerase inhibitor44 and co-crystallized with MIF
ylthiourea, (IC50 = 300 nM) was obtained from FlexX docking
combined with pharmacophoric filtering on one of the eight MD
structures (Fig. 6C). When the validation of the hits from the differ-
ent approaches outlined above was carried out at higher com-
(PDB code: 1GCZ). Orita et al.44 demonstrated that compound 8
has a Ki of 4
lM; whereas our studies showed that it inhibits MIF
with an IC50 of 45.2
lM. Plausible explanations for this discrepancy
are the fact that our IC50 calculations were based on end point
absorbance (i.e., absorbance of the substrate after 2 min reaction
with the enzyme), whereas Ki values in Orita et al. were calculated
from the initial velocity of reaction of the substrate with the
enzyme. Moreover, our experimental design is based on the use
of higher amounts of substrate and enzyme (MIF, 500 nM versus
pound concentration (100
Standalone docking with Surflex allowed the identification of five
new hits with IC50 between 18.6 and 62.1 M. Other active inhib-
itors (hits 11, 12, and 15 which have IC50 values of >200, 1.04, and
64.2 M, respectively) were obtained from docking + pharmaco-
lM), we identified nine additional hits.
l
l
phore filtering combined with three MD simulation conformations.
Our hits can be clustered according to their chemical classes.
Compounds 1, 3, 6, 10 and 11 belong to chemical classes that do
not share any similarity with previously known inhibitors. Notably
all of these hits have conjugated ring systems that fit inside the ac-
tive site and are stabilized with aryl–aryl interactions provided by
Tyr95 and/or Phe113. Moreover, the specificity of interaction is of-
ten provided by hydrogen bonding and salt bridge interactions
involving at least two of the following residues: Lys32, Ile64,
Pro1, Tyr95, Asn97. Compounds 2 and 12 belong to the ‘phenyl-
thiourea’ chemical family. Figure 9B illustrates the interaction of
the computed best pose hit 2 within the catalytic site of MIF. As
with previously known inhibitors, hydrogen bonds and salt bridges
involving Ile64 and Lys32 seem to contribute significantly to the
binding energy. Similarity searches carried out on the 1000
screened compounds identified five other molecules belonging to
the same family. All five compounds were inactive and represent
structural analogues of compound 2 (Supplementary Fig. S3). Un-
like compound 2, the aromatic rings of these compounds contain
additional substitutions on the phenyl or pyridine rings that ap-
pear to introduce sterical effects that distort the interactions ob-
served with compound 12 and interfere with occupancy of the
active site.
Compound 4 is predicted to bind to the enzymatic pocket in a
manner similar to the known inhibitors. Edge-to-face aryl–aryl
interactions, as well as hydrogen bonds involving Lys32, Ile64
and Pro1, participate in the stabilization of compound 4 within
the pocket. Two inactive close structural analogues that differ with
respect to the substitution of one of the phenyl rings were identi-
fied among the 1000 compounds screened. Changing the type of
halogen atoms or its substitution position, as well as the incorpo-
ration of additional substitutions (e.g., Cl instead of Br) appears
to be sufficient to abolish the activity of compound 4, highlighting
the specificity with which this inhibitor interacts within the active
site (Supplementary Fig. S3).
320 nM in Orita et al.’s study; substrate: 533
250–500
M used by Orita et al.44). Compound 15, which is also
a coumarin derivative and has not been reported previously,
showed a higher IC50 (64.2 M). Analysis of the 1000 screened
lM versus
l
l
compounds demonstrated the existence of nine other inactive cou-
marin analogues (Supplementary Fig. S3). Examination of the com-
puted best poses of hits 8 and 15 demonstrated the importance of
the hydroxyl at position 7, which interacts with Asn97 through
hydrogen bonding and stabilizes the molecule within the active
site (Fig. 9), consistent with previous studies by Orita et al.44
The inhibition curves shown in Figure 5 are unusually steep for
some compounds (e.g., 1, 2, 3, 4, 5, 6, 9, and 11); this could be
indicative of compound aggregation at particular concentrations,
resulting in false positive inhibition. To examine this possibility,
we performed the concentration-dependant inhibition assays with
the 15 initially confirmed hits, and determined that hits 5, 9, and
14 are most likely promiscuous aggregators, and were discarded
from the enrichment factor calculations. Inhibition of MIF by these
molecules was abolished in the presence of 1% of detergents
(Fig. 5). The basis underlying the sharp inhibition curves has been
discussed extensively by Shoichet59; in our case the steepness of
the curve for some of the validated inhibitors could be explained
by the facts that MIF has three catalytic sites and/or the high en-
zyme concentrations used in our assays (500 nM). This concentra-
tion of MIF was selected to reduce the effects of enzyme adsorption
to the walls of microtiter plates. This may lead to the effects ob-
served in the case of high [enzyme]/Kd ratios. We believe that this
is the basis for the high Hill coefficients we observed for the com-
pounds (hits 1, 2, 3, 4, 6, and 11) which we validated as non-aggre-
gating inhibitors.
Molecular docking experiments predicted that the inhibitors 1,
3, 4 and 6 interact with the active site of MIF. Experimentally they
proved to act as suicide inhibitors, thereby casting doubts about
counting them as computational successes. However, we showed
that all four inhibitors only react with MIF’s catalytic Pro1 residue,
which requires that they fit and adopt the correct orientation with-
in the catalytic pocket of MIF. It is also plausible that the high reac-
tivity of Pro1 may be sufficient for covalent modification, although
the efficiency of covalent modification suggest that both of these
Despite the fact that most of inhibitors discovered in this study
do not share structural similarities with known inhibitors of MIF,
our in silico protocol identified two coumarin derivatives
(compounds 8 and 15), one of which was previously identified as
Figure 9. Calculated docking poses of coumarin and phenylthiourea inhibitors. (A) Hit 15 (coumarin derivative). (B) Hit 2 (phenylthiourea derivative). Polar interactions are
shown as black dotted lines; aryl–aryl contacts are pictured in gray.

F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
5437
factors contribute to the potency of these inhibitors. Furthermore,
uncertainties about their specificity were ruled out experimentally.
We showed that the four inhibitors failed to modify other proteins,
suggesting that they modify MIF in a specific manner. (Supplemen-
tary Fig. S4C–F). We chose proteins that contain potentially nucle-
ophilic groups such as cysteines and lysines, which would
therefore have been modified by hits 1, 3, 4, and 6 if they were
not specific for MIF. The finding that hits 1 and 3 covalently bind
to 4-oxalocrotonate tautomerase (Supplementary Fig. S4B) does
not dismiss them. As we have shown that these molecules react
with MIF’s Pro1 residue (Fig. 7), it is not surprising that these com-
pounds modify 4-OT. Like MIF, these enzymes are homotrimers (4-
OT is a trimer of homodimers) and belong to the tautomerase super
sampling, eight independent simulations were carried out using dif-
ferent initial conditions, which are summarized in Supplementary
TableS1. The system was solvated with a pre-equilibrated solvation
box (edge length around 76 Å) consisting of TIP3P water molecules
where periodic boundary conditions were applied. Crystallographic
water molecules were included in the initial model in four of the
eight conditions; all crystal water molecules were removed in the
remaining conditions (Supplementary Table S1). Chloride and so-
dium ions were added to ensure a neutral system. Unfavorable con-
tacts were removed by a short energy minimization with conjugate
gradient and ABNR. Electrostatic interactions were treated using the
particle-mesh Ewald summation method, and we used the switch
function for the van der Waals energy interactions with cuton, cut-
off and cutnb values of 9, 11 and 13 Å, respectively. The Shake algo-
rithm60 was applied to all hydrogen-containing bonds and a 1 fs
integration step was used. The system was heated gradually to the
desired temperature (300 or 333 K), followed by a further equilibra-
tion step (150 ps). During these two early steps, harmonic con-
straints were applied to protein heavy atoms. The constraint
harmonic constant (k) was equal to 1 and 0.1 kcal/mol/Å2 for the
backbone and side chains, respectively. The production phase was
performed without any constraints. Snapshots of the coordinates
were saved every 500 steps (0.5 ps) leading to about 4000 instanta-
neous conformations for each trajectory. A typical 2 ns simulation
required about three weeks CPU time on a 3.5 GHz processor. Tra-
jectories were analyzed using a combination of CHARMM, in house
Perl or Matlab scripts, and VMD. Overall <RMSD> variations were
computed with CHARMM after superimposition of the CA atoms.
family characterized by a b-a-b motif and a catalytic amino-termi-
nal proline.31 Mutation of the catalytic Pro1 residue in MIF, 4-OT,
and 5-CHMI abolishes the tautomerase activity of these en-
zymes.31,32 Together, all these data suggest that the covalent hits
1, 3, 4 and 6 we present here are novel, specific irreversible mod-
ulators of MIF’s tautomerase function.
4. Conclusion
Using our integrative virtual screening approach we identified
and validated 12 novel inhibitors of MIF tautomerase activity.
The chemical diversity and mode of action of these compounds
suggest that they could be used as molecular probes to elucidate
the functions and biology of MIF and as potential lead candidates
in drug developments of anti-MIF drugs. Taken together, these
findings confirm the importance of MD simulations for the virtual
screening, since FlexX used on the X-ray coordinates of MIF alone
would not have identified these inhibitors. Moreover, the fact that
the compounds obtained using the three different approaches be-
long to different chemical classes and exhibit different modes of
action underscores the importance of coupling different computa-
tional approaches for optimizing screening protocols. The combi-
nation of two docking softwares with molecular dynamic
simulations and pharmacophore filtering demonstrates one more
time the power of in silico procedures as a complementary ap-
proach in the discovery of novel inhibitors of enzymes and other
biological systems. Combining the eight structures obtained from
MD simulations with docking with either FlexX or Surflex and
pharmacophore filtering with Unity can now be applied to very
large databases of small compounds in the hope to discovering
new scaffolds inhibiting MIF activity.
5.2. Ligand library
Virtual screening methods were applied to the NINDS library of
FDA-approved molecules (1040 compounds), the HitFinder diverse
subset of the Maybridge database (14,400 compounds), and the
Chembridge library (ꢂ500,000 compounds obtained from Chem-
Bridge Corporation). The last library, in 2D Structure Definition File
format (SDF), was transformed into a 3D database using the CONCORD
software (implemented in Tripos’ Sybyl 8.0). Default parameters
were employed for the conversion and all anilinic nitrogen were
designed in a planar conformation. The library was then filtered
using the Tripos SELECTOR module to retain only structures with
30 or fewer non-hydrogen atoms, resulting in a final dataset of
33,900 compounds. Positive control ligands were obtained from
MIF X-ray structures where the protein had been co-crystallized
with substrate (PDB code 1CA7) or with tautomerase inhibitors
(PDB codes 1GCZ, 1LJT, 1MFI, and 2OOZ). Coordinates of all ligands
were then refined using the energy minimization module of Sybyl
(MMFF94 force field).
5. Methods
5.1. Molecular dynamics simulations
5.3. Docking
All MD simulations were performed in explicit solvent with peri-
odic conditions with the CHARMM package version 33b1 and
CHARMM force field version 22 with CMAP correction.45–48 The ini-
tial coordinates were taken from the ligand-free crystal structure
solved at a resolution of 1.5 Å (PDB ID: 1GD044). Non-protein deriv-
atives such as citric acid and SO42- ions were not taken into account.
C-terminal His-tag residues were removed from the PDB file. Orien-
tation of the side chains of Asn, Gln, and His residues were checked
using the in-house VMD plugin checksidechains (http://www-
s.ks.uiuc.edu/Research/vmd/plugins/) and using the WHAT IF web
interface (http://swift.cmbi.kun.nl/). Ionization states of titratable
residues were determined according to empirical pKa and Pois-
son–Boltzmann calculations. Residues His40 and His62 were as-
signed HSE types. All hydrogens were considered explicitly in the
calculation. Coordinates of missing hydrogen atoms were added
using the hbuild algorithm in CHARMM. To improve conformational
Protein–ligand complexes were simulated using FlexX version
3.0.2 (BioSolveIT, Germany), and Surflex version 2.4 implemented
in Sybyl 8.0 from Tripos, Inc.61 The MIF residues that defined the
active site in our docking studies are: (Subunit C) Pro1, Met2,
Thr30, Gly31, Lys32, Pro33, Pro34, Tyr36, Ile37, Ala38, Ser60,
Leu61, His62, Ser63, Ile64, Gly65, Lys66, Ile67, Asp100, Met101,
Ala103, Val106, Gly107, Trp108, Phe113, and (Subunit B), Phe49,
Gly50, Val94, Tyr95, Ile96, Asn97. Formal charge assignments for
both receptors and ligands were made using the default parame-
ters of FlexX, except for Pro1 of MIF, which is assumed to be neu-
tral under physiological conditions because of its pKa of 5.6–6.62
For each compound docked, we saved the 50 best poses according
to FlexX scoring function (F-Score63). Additionally, each pose was
evaluated using functions from the CScore module of Sybyl.64 This
consensus score (CScore) integrates a number of popular scoring

5438
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
functions to rank ligand’s affinity for the active site of the receptor.
D_score and G_score are two Force-Field derived scoring functions.
D_score, derived from DOCK score in the FlexX implementation,
uses steric and electrostatic terms based on the AMBER Force-
Field,65 whereas G_score, derived from GOLD score in the FlexX
implementation, is a sum of the hydrogen bonding stabilization
energy (calculated from van der Waals energy for the ligand and
conformers) and a pair wise dispersion potential between ligand
methyl ester (4 mM) for 5 min at room temperature and then
placed directly on ice.
6.4. Multi-well plate in vitro screening of the 1000 best-scored
molecules identified by virtual HTS
To validate our in silico screening protocol, we constructed a
small library of 1000 compounds consisting of (i) the 333 highest
ranking molecules obtained from docking with Surflex (Docking
standalone with the geomX option toggled on); (ii) the 333 high-
est ranking molecules obtained from Docking with FlexX using
different structures of the catalytic site (Docking + MDS); and
(iii) the 334 highest ranking molecules obtained upon Docking
with FlexX + Unity Pharmacophore filtering (PF) on the different
structures of the catalytic site as well (Docking + MDS + PF). The
and protein that describes the hydrophobic binding energy.66,67
A
knowledge-based scoring function, PMF score, exploits structural
information of known complexes and converts it into distance-
dependent Helmholtz free energies.68 Finally, two empirical scor-
ing functions were implemented: ChemScore consists of a term
that estimates lipophilic contact energy, a metal–ligand binding
contribution, an empirical form for hydrogen bonds, and a penalty
for ligand flexibility,69 whereas FlexX score considers the number
of rotatable bonds in the ligand, hydrogen bond interaction, ion
pairing, aromatic interactions, and lipophilic contact energy.70
The final choice of the best scoring function was optimized on
the basis of the docking results of the positive control ligands
(known X-ray structures). The Surflex docking was carried out on
MIF apo-structure (PDB code 1GD0) using the amino acids de-
scribed above to generate the Surflex protomol (Surflex active site
definition) with standard parameters (threshold 0.5, no bloat). One
third of randomly picked compounds within the ChemBridge fil-
tered database (11,300 compounds) were screened using the Geom
X option from Surflex-Dock version 2.4. According to Surflex-Dock
scoring function, the 333 best-docked compounds were selected
for further analysis (top 3%).
compounds were dissolved to prepare
10 mM in 100% DMSO, in three 384-well plates (384-well Corning
clear plates). To perform the tautomerase assay in high-
a stock solution of
a
throughput format, wt huMIF (final concentration: 500 nM) was
added to 384-well plates (Nunc, Black wall, clear bottom) con-
taining potassium phosphate buffer (50 mM), EDTA (0.5 mM), at
pH6, and the compound of interest at a final concentration of
10 lM or 100 lM (final percentage of DMSO lower than 1.5%).
Wt huMIF and the compound were mixed and incubated for
30 min at rt. The reaction was initiated by adding freshly pre-
pared
D-dopachrome methyl ester substrate (prepared as de-
scribed above). The plates were centrifuged for 2 min at
2000 rpm to remove air bubbles and the absorbance was mea-
sured at 475 nm using a Tecan Safire II reader. BITC, which was
previously shown to completely inhibit MIF at 10 l
M,52 was used
5.4. Pharmacophore-based search
as a positive control. The screening assay was carried out in trip-
licate. Inhibitors’ potencies were measured by recording the
absorbance at the end point, that is, absorbance of the substrate
2 min after initiation of reaction with the enzyme.
A pharmacophore model of MIF’s active site was built with the
UNITY component of Sybyl 8.0. Spatial constraints were built for
the following residues: (1) Acceptor sites: Pro1-A (N atom lone
pair), Tyr36-A (carbonyl O), Ile64-A (carbonyl O), Asn97-B (side
chain amide), and Met101-B (carbonyl O); (2) Donor sites: Lys32-
A (side chain ammonium), Ala38-A (amide N), His62-A (amine N),
and Ile64-A (amide N). Ligand structures were selected as hits if
they presented at least two donor atoms matching the receptor’s
‘Acceptor site’ constraints and one acceptor atom fitting into a ‘Do-
nor site’ query. The tolerances of the spatial pharmacophoric con-
straints (i.e., the allowed deviation from a perfect match) were
optimized over a range of 0.5–2.0 Å, using known X-ray structures.
6.5. Validation and IC50 measurements of in vitro screening hits
Primary hits obtained as described above were subjected to a
validation assay in 384-well plates under the same conditions as
for HTS. Buffer, enzyme and substrate solutions were added to
the wells using an EL406 dispenser robot (BioTek Instruments).
The validation assay was performed in five replicates. IC50 calcu-
lations were done using the same microtiter plate-based assay
and were carried out in triplicate. Compounds were incubated
with 500 nM wt huMIF for 2 h at room temperature, at concen-
trations ranging from 24 nM to 50
lM (hits 1–5), or 98 to
6. Experimental section
6.1. Chemicals
200 M (hits 6–15), in absence or presence of 0.1% v/v Triton
l
X-100 in both buffer and enzyme solutions, in order to check
for compound aggregation. Mixing was carefully done below
the surface of the liquid in order to minimize the formation of
bubbles. Relative inhibitions were determined from absorbance
measurement at 475 nm after 2 min of enzymatic reaction. IC50
values were determined by nonlinear regression in Matlab 7.5
(MathWorks).
Compounds were purchased from ChemBridge Corporation
(www.chembridge.com). Miscellaneous chemicals were purchased
from Sigma–Aldrich Chemicals, unless otherwise stated and are of
the highest grade commercially available. Trypsin was purchased
from Promega.
6.6. MIF aggregation studies
6.2. Expression and purification of MIF
To determine if the validated hits induce the aggregation of MIF,
Wild type human MIF (wt huMIF) was expressed and purified as
described previously.28
wt huMIF (ꢂ15
at concentrations ranging from 15
than 15 M), for 1 h at room temperature. The MIF-compound
mixture was then filtered with a 0.2 m centrifugal filter (Milli-
l
M) was incubated in 1X PBS with each compound
l
M to the IC50 value (if higher
l
6.3. Synthesis of the
D-dopachrome methyl ester substrate for
l
keto-enol tautomerase assay
pore) and the retentate was solubilized in 0.1% SDS. Both the
supernatant and the retentate were analyzed on a 15% SDS–PAGE.
Two controls of protein incubated without compound, or with
ebselen (a molecule that induces MIF aggregation50) were used.
The
D-dopachrome methyl ester was freshly prepared by addi-
tion of periodate (NaIO4) (8 mM) to
L
-3,4-dihydroxyphenylalanine

F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
5439
6.7. MALDI/TOF–TOF MS molecular mass measurements for
examining possible protein modifications
the distances between key residues to select structures where the
catalytic pocket has a suitable conformation for ligand docking; as
well as the mass spectrometry data for the incubation of the cova-
lent hits with five different proteins) associated with this article
can be found, in the online version, at doi:10.1016/
j.bmc.2010.05.010.
To determine whether any of the identified inhibitors cova-
lently modifies MIF, mass spectrometric analysis of wt huMIF incu-
bated with each inhibitor was performed by matrix-assisted laser
desorption ionization (MALDI) MS using a linear positive ion mode
on ABI 4700 (PCF Lausanne). Linear mode calibration was applied
References and notes
with a single point using human
Sample preparation: 10 M of huMIF protein was incubated for
1 h at rt with 10 M of inhibitor; then 1 L of MIF-inhibitor mix-
ture was mixed with 4 L of matrix solution, which was prepared
a
-synuclein (MH+: 14461.19).
1. Bloom, B. R.; Bennett, B. Science 1966, 153, 80.
2. David, J. R. Proc. Natl. Acad. Sci. U.S.A. 1966, 56, 72.
l
l
l
3. Weiser, W. Y.; Pozzi, L. M.; David, J. R. J. Immunol. 1991, 147, 2006.
4. Pozzi, L. A.; Weiser, W. Y. Cell Immunol. 1992, 145, 372.
5. Bernhagen, J.; Calandra, T.; Mitchell, R. A.; Martin, S. B.; Tracey, K. J.; Voelter,
W.; Manogue, K. R.; Cerami, A.; Bucala, R. Nature 1993, 365, 756.
6. Calandra, T.; Echtenacher, B.; Roy, D. L.; Pugin, J.; Metz, C. N.; Hultner, L.;
Heumann, D.; Mannel, D.; Bucala, R.; Glauser, M. P. Nat. Med. 2000, 6, 164.
7. Onodera, S.; Kaneda, K.; Mizue, Y.; Koyama, Y.; Fujinaga, M.; Nishihira, J. J. Biol.
Chem. 2000, 275, 444.
8. Morand, E. F.; Leech, M.; Bernhagen, J. Nat. Rev. Drug Disc. 2006, 5, 399.
9. Santos, L. L.; Morand, E. F. Wien. Med. Wochenschr. 2006, 156, 11.
10. Bernhagen, J.; Mitchell, R. A.; Calandra, T.; Voelter, W.; Cerami, A.; Bucala, R.
Biochemistry 1994, 33, 14144.
11. Calandra, T.; Bernhagen, J.; Mitchell, R. A.; Bucala, R. J. Exp. Med. 1994, 179,
1895.
12. Herriott, M.; Jiang, H.; Stewart, C.; Fast, D.; Leu, R. J. Immunol. 1993, 150, 4524.
13. Onodera, S.; Suzuki, K.; Matsuno, T.; Kaneda, K.; Takagi, M.; Nishihira, J.
Immunology 1997, 92, 131.
l
by dissolving 14 mg of sinapinic acid (SA) in 0.1% TFA/acetonitrile
mixture (1:1). One microliter of the sample-matrix mixture was
then deposited on the MALDI plate.
6.8. Protein digestion analysis to identify the residue(s) modified
by the suicide compounds
To identify the exact residues being covalently modified by MIF
inhibitors, 200
with 30 M of the covalent inhibitor (e.g., hits 1, 3, 4 and 6). Diges-
tion with trypsin (1:50 protein weight) was then performed over
night at 37 °C, with and quenched with 1 L of formic acid. Di-
lL of wt huMIF (30 lM) is incubated at rt for 1 h
l
l
14. Bacher, M.; Metz, C. N.; Calandra, T.; Mayer, K.; Chesney, J.; Lohoff, M.; Gemsa,
D.; Donnelly, T.; Bucala, R. Proc. Natl. Acad. Sci. U.S.A. 1996, 93, 7849.
15. Yang, Y.; Degranpré, P.; Kharfi, A.; Akoum, A. J. Clin. Endocrinol. Metab. 2000, 85,
4721.
gested samples were analyzed by MALDI MS/MS on a 4700 MAL-
DI/TOF–TOF instrument, as described in the previous paragraph.
16. Mitchell, R. A.; Metz, C. N.; Peng, T.; Bucala, R. J. Biol. Chem. 1999, 274, 18100.
17. Lolis, E. Curr. Opin. Pharmacol. 2001, 1, 662.
18. Van Molle, W.; Libert, C. Eur. J. Immunol. 2005, 35, 3396.
19. Awandare, G. A.; Martinson, J. J.; Were, T.; Ouma, C.; Davenport, G. C.;
Ong’echa, J. M.; Wang, W.; Leng, L.; Ferrell, R. E.; Bucala, R.; Perkins, D. J. J. Infect
Dis. 2009, 200, 629.
20. Bando, H.; Matsumoto, G.; Bando, M.; Muta, M.; Ogawa, T.; Funata, N.;
Nishihira, J.; Koike, M.; Toi, M. Jpn. J. Cancer Res. 2002, 93, 389.
21. Bucala, R.; Donnelly, S. C. Immunity 2007, 26, 281.
22. Calandra, T.; Roger, T. Nat. Rev. Immunol. 2003, 3, 791.
23. Cvetkovic, I.; Stosic-Grujicic, S. Int. Immunopharmacol. 2006, 6, 1527.
24. Lue, H.; Kleemann, R.; Calandra, T.; Roger, T.; Bernhagen, J. Microbes Infect.
2002, 4, 449.
25. Sun, H. W.; Swope, M.; Cinquina, C.; Bedarkar, S.; Bernhagen, J.; Bucala, R.;
Lolis, E. Protein Eng. 1996, 9, 631.
26. Sun, H. W.; Bernhagen, J.; Bucala, R.; Lolis, E. Proc. Natl. Acad. Sci. U.S.A. 1996, 93,
5191.
27. Muhlhahn, P.; Bernhagen, J.; Czisch, M.; Georgescu, J.; Renner, C.; Ross, A.;
Bucala, R.; Holak, T. A. Protein Sci. 1996, 5, 2095.
28. El-Turk, F.; Cascella, M.; Ouertatani-Sakouhi, H.; Narayanan, R. L.; Leng, L.;
Bucala, R.; Zweckstetter, M.; Rothlisberger, U.; Lashuel, H. A. Biochemistry 2008,
47, 10740.
6.9. Effects of compounds on the tertiary and quaternary struc-
ture of MIF
To determine whether the MIF inhibitors identified in our
screen act by inducing changes in the tertiary and/or quaternary
structure of MIF, MIF-inhibitor mixtures were analyzed by static
light scattering coupled with size exclusion chromatography. Mea-
surements were carried out at room temperature on a DAWN HEL-
EOS multi-angle light scattering detector (Wyatt Technology Corp,
Santa Barbara CA) online with a size exclusion chromatography
column (Superdex 75 10/300). The system is also equipped with
UV and refractive index detector. The buffer used for equilibration
of the gel filtration column was 20 mM Tris, 150 mM NaCl, 0.05%
w/v NaN3, pH 7.4. A volume of 100
lL of MIF samples (20–30 lM
in 1X PBS), incubated with 10 M of inhibitors for 1 h at rt, was in-
l
jected onto the column and eluted at a flow rate of 0.4 mL/min.
Absolute MWs were determined using ASTRA version 5.3 from
Wyatt Technologies.
29. Lubetsky, J. B.; Swope, M.; Dealwis, C.; Blake, P.; Lolis, E. Biochemistry 1999, 38,
7346.
30. Swope, M.; Sun, H.; Blake, P.; Lolis, E. EMBO J. 1998, 17, 3534.
31. Poelarends, G. J.; Veetil, V. P.; Whitman, C. P. Cell Mol. Life Sci. 2008, 65, 3606.
32. Whitman, C. P. Arch. Biochem. Biophys. 2002, 402, 1.
33. Senter, P. D.; Al-Abed, Y.; Metz, C. N.; Benigni, F.; Mitchell, R. A.; Chesney, J.;
Han, J.; Gartner, C. G.; Nelson, S. D.; Todaro, G. J.; Bucala, R. Proc. Natl. Acad. Sci.
U.S.A. 2002, 99, 144.
34. Winner, M.; Meier, J.; Zierow, S.; Rendon, B. E.; Crichlow, G. V.; Riggs, R.;
Bucala, R.; Leng, L.; Smith, N.; Lolis, E.; Trent, J. O.; Mitchell, R. A. Cancer Res.
2008, 68, 7253.
35. Cournia, Z.; Leng, L.; Gandavadi, S.; Du, X.; Bucala, R.; Jorgensen, W. L. J. Med.
Chem. 2009, 52, 416.
36. Al-Abed, Y.; Dabideen, D.; Aljabari, B.; Valster, A.; Messmer, D.; Ochani, M.;
Tanovic, M.; Ochani, K.; Bacher, M.; Nicoletti, F.; Metz, C.; Pavlov, V. A.; Miller,
E. J.; Tracey, K. J. J. Biol. Chem. 2005, 280, 36541.
37. Rendon, B. E.; Roger, T.; Teneng, I.; Zhao, M.; Al-Abed, Y.; Calandra, T.; Mitchell,
R. A. J. Biol. Chem. 2007, 282, 29910.
38. Meyer-Siegler, K. L.; Iczkowski, K. A.; Leng, L.; Bucala, R.; Vera, P. L. J. Immunol.
2006, 177, 8730.
39. Choi, H.; Lee, Y.; Lee, S.; Hong, S.; Hahm, D.; Choi, S.; Yang, H.; Yoo, M.; Kim, K.
Arthritis Res. Ther. 2009, 11.
40. So, A.; Chamot, A.; Peclat, V.; Gerster, J. Rheumatology 1999, 38, 407.
41. Lubetsky, J. B.; Dios, A.; Han, J.; Aljabari, B.; Ruzsicska, B.; Mitchell, R.; Lolis, E.;
Al-Abed, Y. J. Biol. Chem. 2002, 277, 24976.
Acknowledgments
Financial support for these studies was provided by the Swiss
Federal Institute of Technology Lausanne (HAL, FE, HO), and a grant
from the Swiss National Science Foundation (HAL, FE-310000-
110027). We thank Dr. Marc Moniatte, Mr. Diego Chiappe, Mr. Jon-
athan Paz Montoya and Mr. Jérome Vialaret from the proteomic
facility (http://pcf.epfl.ch) for their assistance and advices. We also
thank Dr. Gerardo Turcatti, Dr. Damiano Banfi, Dr. Marc Chambon
and M. Miquel Busquets Lopez from the screening facility (http://
bsf.epfl.ch) for their help to perform the in vitro screening and val-
idation of the hits. We finally thank Prof. Brian Shoichet and Mrs.
Allison Doak from the University of California, San Francisco for
their very kind assistance about the detergent-based assay.
Supplementary data
42. Taylor, A. B.; Johnson, W. H., Jr.; Czerwinski, R. M.; Li, H. S.; Hackert, M. L.;
Whitman, C. P. Biochemistry 1999, 38, 7444.
43. Zhuang, S.; Zou, J.; Jiang, Y.; Mao, X.; Zhang, B.; Liu, H.; Yu, Q. J. Med. Chem.
2005, 48, 7208.
Supplementary data (includes two tables describing the 8 MD
simulation conditions used in the in silico study, and representing

5440
F. E. Turk et al. / Bioorg. Med. Chem. 18 (2010) 5425–5440
44. Orita, M.; Yamamoto, S.; Katayama, N.; Aoki, M.; Takayama, K.; Yamagiwa, Y.;
Seki, N.; Suzuki, H.; Kurihara, H.; Sakashita, H.; Takeuchi, M.; Fujita, S.; Yamada,
T.; Tanaka, A. J. Med. Chem. 2001, 44, 540.
45. MacKerell, A. D. J.; Bashford, D.; Bellott, M.; Dunbrack, R. L. J.; Evanseck, J. D.;
Field, M. J.; Fischer, S.; Gao, J.; Guo, H.; Ha, S.; Joseph-McCarthy, D.; Kuchnir, L.;
Kuczera, K.; Lau, F. T. K.; Mattos, C.; Michnick, S.; Ngo, T.; Nguyen, D. T.;
Prodhom, B.; Reiher, W. E. I.; Roux, B.; Schlenkrich, M.; Smith, J. C.; Stote, R.;
Straub, J.; Watanabe, M.; Wiórkiewicz-Kuczera, J.; Yin, D.; Karplus, M. J. Phys.
Chem. 1998, 102, 3586.
54. Fischer-Valuck, B. W.; Barrilleaux, B. L.; Phinney, D. G.; Russell, K. C.; Prockop,
D. J.; O’Connor, K. C. Biotechnol. Lett. 2009.
55. Peng, H.; Yang, Z. K.; Hou, J. Zhonghua Gan Zang Bing Za Zhi 2008, 16, 918.
56. Tanino, Y.; Makita, H.; Miyamoto, K.; Betsuyaku, T.; Ohtsuka, Y.; Nishihira, J.;
Nishimura, M. Am. J. Physiol. Lung Cell Mol. Physiol. 2002, 283, L156.
57. Orita, M.; Yamamoto, S.; Katayama, N.; Fujita, S. Curr. Pharm. Des. 2002, 8, 1297.
58. Brown, K. K.; Blaikie, F. H.; Smith, R. A.; Tyndall, J. D.; Lue, H.; Bernhagen, J.;
Winterbourn, C. C.; Hampton, M. B. J. Biol. Chem. 2009.
59. Shoichet, B. K. J. Med. Chem. 2006, 49, 7274.
46. Mackerell, A. D., Jr. J. Comput. Chem. 2004, 25, 1584.
60. Ryckaert, J. P.; Ciccotti, G.; Berendsen, H. J. J. Comput. Phys. 1977, 23, 327.
61. Jain, A. J. Med. Chem. 2003, 46, 499.
62. Stamps, S. L.; Fitzgerald, M. C.; Whitman, C. P. Biochemistry 1998, 37, 10195.
63. Rarey, M.; Wefing, S.; Lengauer, T. J. Comput. Aided Mol. Des. 1996, 10, 41.
64. Clark, R. D.; Strizhev, A.; Leonard, J. M.; Blake, J. F.; Matthew, J. B. J. Mol. Graphics
Modell. 2002, 20, 281.
65. Kuntz, I. D.; Blaney, J. M.; Oatley, S. J.; Langridge, R.; Ferrin, T. E. J. Mol. Biol.
1982, 161, 269.
66. Jones, G.; Willett, P.; Glen, R. C.; Leach, A. R.; Taylor, R. J. Mol. Biol. 1997, 267,
727.
47. Mackerell, A. D., Jr.; Feig, M.; Brooks, C. L., 3rd. J. Comput. Chem. 2004, 25, 1400.
48. Buck, M.; Bouguet-Bonnet, S.; Pastor, R. W.; MacKerell, A. D., Jr. Biophys. J. 2006,
90, L36.
49. Crichlow, G. V.; Cheng, K. F.; Dabideen, D.; Ochani, M.; Aljabari, B.; Pavlov, V. A.;
Miller, E. J.; Lolis, E.; Al-Abed, Y. J. Biol. Chem. 2007, 282, 23089.
50. Ouertatani-Sakouhi, H.; El-Turk, F.; Fauvet, B.; Cho, M. K.; Le Roy, D.; Dewor,
M.; Roger, T.; Bernhagen, J.; Calandra, T.; Zweckstetter, M.; Lashuel, H. A.
J. Biol.Chem. 2010, submitted for publication.
51. Feng, B. Y.; Shoichet, B. K. Nat. Protoc. 2006, 1, 550.
52. Ouertatani-Sakouhi, H.; El-Turk, F.; Fauvet, B.; Roger, T.; Le Roy, D.; Karpinar, D.
P.; Leng, L.; Bucala, R.; Zweckstetter, M.; Calandra, T.; Lashuel, H. A.
Biochemistry 2009, 48, 9858.
53. Stivers, J. T.; Abeygunawardana, C.; Mildvan, A. S.; Hajipour, G.; Whitman, C. P.;
Chen, L. H. Biochemistry 1996, 35, 803.
67. Jones, G.; Willett, P.; Glen, R. J. Mol. Biol. 1995, 245, 43.
68. Muegge, I.; Martin, Y. C.; Hajduk, P. J.; Fesik, S. W. J. Med. Chem. 1999, 42, 2498.
69. Eldridge, M. D.; Murray, C. W.; Auton, T. R.; Paolini, G. V.; Mee, R. P. J. Comput.
Aided Mol. Des. 1997, 11, 425.
70. Rarey, M.; Kramer, B.; Lengauer, T.; Klebe, G. J. Mol. Biol. 1996, 261, 470.