Comparing Drug Images and Repurposing Drugs with BioGPS and FLAPdock: The Thymidylate Synthase Case

Article abstract of DOI:10.1002/cmdc.201600121

Repurposing and repositioning drugs has become a frequently pursued and successful strategy in the current era, as new chemical entities are increasingly difficult to find and get approved. Herein we report an integrated BioGPS/FLAPdock pipeline for rapid

Full text of DOI:10.1002/cmdc.201600121

DOI: 10.1002/cmdc.201600121
Full Papers
Comparing Drug Images and Repurposing Drugs with
BioGPS and FLAPdock: The Thymidylate Synthase Case
[b]
[a]
[a]
[a]
[a]
Lydia Siragusa, Rosaria Luciani, Chiara Borsari, Stefania Ferrari, Maria Paola Costi,
[c]
[a, d]
Gabriele Cruciani, and Francesca Spyrakis*
Repurposing and repositioning drugs has become a frequently
pursued and successful strategy in the current era, as new
chemical entities are increasingly difficult to find and get ap-
proved. Herein we report an integrated BioGPS/FLAPdock
pipeline for rapid and effective off-target identification and
drug repurposing. Our method is based on the structural and
chemical properties of protein binding sites, that is, the ligand
image, encoded in the GRID molecular interaction fields (MIFs).
Protein similarity is disclosed through the BioGPS algorithm by
measuring the pockets’ overlap according to which pockets
are clustered. Co-crystallized and known ligands can be cross-
docked among similar targets, selected for subsequent in vitro
binding experiments, and possibly improved for inhibitory po-
tency. We used human thymidylate synthase (TS) as a test case
and searched the entire RCSB Protein Data Bank (PDB) for simi-
lar target pockets. We chose casein kinase IIa as a control and
tested a series of its inhibitors against the TS template. Ellagic
acid and apigenin were identified as TS inhibitors, and various
flavonoids were selected and synthesized in a second-round
selection. The compounds were demonstrated to be active in
the low-micromolar range.
Introduction
With the threat of a future drugs shortage, techniques aimed
at drug repositioning and polypharmacology rationalization
are becoming more and more popular. Antimicrobials are
facing worrisome failures due to the spread and proliferation
not only lead to unexpected applications or to multi-target
drugs, but also to prevent adverse effects from occurring in
clinical trials, or worse, when drugs have already reached the
market. In addition, drug repurposing (using already approved
and safe drugs for new targets), would save money, time and
resources. A number of methodologies have appeared in
recent years with the aim of identifying off-targets, predicting
side effects (SEs) and possibly finding new applications for al-
ready known molecules.
[
1–3]
of multidrug resistant bacteria.
Cancer research has to deal
with resistance to chemotherapy and molecular targeted thera-
[
4]
pies and, in general, drugs are withdrawn from the market or
during clinical phases because of toxic effects. In this scenario,
[
5]
so far from Ehrlich’s magic bullet, the development and dis-
covery of drugs that are able to bind more than one single
Notable efforts have been made to develop in vitro assays
and standardize procedures to determine the pharmacological
profile of drug candidates. Even if experimental techniques can
provide robust information, assays remain challenging and
[
6]
target, acting as a magic shotgun, is very desirable. Polyphar-
macology has emerged as the next paradigm of drug discov-
[
7–12]
ery,
and off-target prediction is a key issue in bioinformat-
[13]
ics and drug design. The identification of unknown targets can
costly, and computational methods currently represent a val-
uable strategy to be pursued in combination with in vitro anal-
[
a] Dr. R. Luciani, Dr. C. Borsari, Dr. S. Ferrari, Prof. M. P. Costi, Dr. F. Spyrakis
Department of Life Sciences, University of Modena and Reggio Emilia,
Via Campi 103, 41125 Modena (Italy)
yses. Different methodologies have been developed for off-
[14,15]
target identification and drug repositioning.
While initial
[
16–20]
approaches were based on sequence comparison,
in the
E-mail: francesca.spyrakis@unipr.it
last decade more exhaustive compound-based approaches
were developed, starting from the assumption that similar
chemicals should be able to bind similar pockets. Statistical
and canonical correlation analyses were thus applied to link
the ligand chemical space to the targets and to the possible
[
b] Dr. L. Siragusa
Molecular Discovery Limited,
2
15 Marsh Road, Pinner Middlesex London HA55NE (UK)
[
c] Prof. G. Cruciani
Department of Chemistry, Biology and Biotechnology,
University of Perugia, Via Elce di Sotto 8, 06123 Perugia (Italy)
[11,21–25]
related SEs.
At the same time, a number of phenotypic-
[
d] Dr. F. Spyrakis
and pathway-based methods were released, combining drug-
disease relationships, clinically known SEs, gene-disease-drug
Current address: Department of Food Science, University of Parma, Viale
delle Scienze 17A, 43124 Parma (Italy)
Supporting information [a list of the 317 selected cavities along with
their PDB and UniProt codes and co-crystallized ligands; a detailed de-
scription of the pocket connections and Volsurf-based analysis of the
pockets; and docking poses of active flavonoids within TSpw] and the
ORCID identification number(s) for the author(s) of this article can be
found under http://dx.doi.org/10.1002/cmdc.201600121.
connections and drug-drug interactions, in a knowledge based
[26–30]
perspective.
Also, different databases were constructed,
and are currently available, to detect and predict relationships
between drugs, target, side effects and biological path-
[31–33]
ways.
More recently, approaches combining both chemical
ChemMedChem 2016, 11, 1 – 15
1
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
and target information have been reported showing that
a drug action is often unspecific, and underlying the necessity
of combining biology and chemistry to provide reliable molec-
The extension of a MIF depends on the generating groups and
on the energy associated with a possible interaction with
these groups. The larger the MIF produced by one or more res-
idues, the higher the probability to find a complementary
group in that region, i.e., a hydrophobic group if the pocket
residue was hydrophobic or a H-bond donor or acceptor
moiety if the residue was bearing a H-bond acceptor or donor
side-chain respectively. Comparing the MIFs means comparing
the chemical and geometrical properties as well as the en-
coded energetics. Recent applications have demonstrated
BioGPS’ capability of predicting off-target effects, classifying
protein families, justifying polypharmacology, and rationalizing
[
34]
ular explanations for complex SEs. Ligand binding site com-
parison and protein–ligand docking have been also successful-
ly integrated and applied for drug repurposing, side effect pre-
[
35–38]
diction and polypharmacology applications.
Approaches directly comparing protein pockets have also
been recently proposed, based on the assumption that similar
[
39–43]
binding sites can be targeted by similar ligands,
and that
the structural and chemical information encoded into the bind-
ing pockets guide the recognition between macromolecules
[
44–47]
[51,52,55]
and ligands.
SMAP is a fast method for ligand binding site
selectivity between sub-families.
comparison, using a shape description only based on Ca
Here we propose a specific pipeline (Figure 1) for investigat-
ing the biological space around a given target, identifying off-
targets and, eventually, repurposing known drugs. The inte-
grated approach includes a first BioGPS virtual screening (VS)
step for the selection of the most similar pockets to the tem-
plate within the RCSB Protein Data Bank (PDB), and then the
docking of the ligands co-crystallized with the off-targets
within the template binding site, using the FLAPdock algo-
[
48]
[45]
[44]
[46]
atoms.
SiteEngine
Cavbase,
and FuzCav
represent
each pocket residue as a series of pseudocenters, encoding
the physicochemical properties essential for molecular interac-
tions. These rule-based methods generally produce models in
which all pseudocenters are equally considered, without
paying attention to the residue environment. Slightly different
strategies are applied by PocketFEATURE, which compares pro-
tein sites by one or more microenvironments, described as oc-
currence of atoms, residues, and biochemical and biophysical
[56]
rithm implemented in FLAP.
The pipeline is composed of four main steps: 1) data collec-
tion, 2) cavities comparison and selection, 3) ligand docking,
and 4) in vitro binding experiments.
[
49]
properties, and by ProBIS, which is able to detect structurally
similar sites as patterns of physicochemical properties on the
[
50]
protein surface.
1) Data collection. The first step consists of the selection of
the protein template 3D structures and of the database of pro-
teins to compare with it. For each protein structure the co-
crystallized ligand and the cavity containing it (the binding
site) are detected.
In this mazy plethora of options, structure-based methods
can provide mechanistic indications about the off-targets se-
lection and about the related occurrence of side effects. More-
over, the possibility of finely tuning the effect of an approved
or candidate drug toward a new target often relies on the
modulation of their interactions. Structural information is, in
fact, essential to understand and ameliorate the interaction of
a compound with the binding site of a potential new target.
Trying to simplify as much as possible the off-target search,
and going back to the chemical and physical principles of pro-
tein–ligand interactions, we developed the BioGPS (Global Po-
2) Cavities comparison. The template pocket is compared
with all of the other pockets by using the BioGPS algorithm.
Cavities in the database are ranked according to their similarity
[
51]
sitioning System in Biological Space) algorithm,
based on
the chemical/structural comparison of protein binding
[
51,52]
sites.
BioGPS represents and compares pockets according
to their ligand image and not by any other rule-based residue
[
44]
feature. Pockets are described by their molecular interaction
[
53,54]
fields (MIFs, calculated by GRID
), that is, the shape, the hy-
drophobic regions, the H-bond donor and acceptor hotspots
a ligand would encounter upon entering the cavity. The
BioGPS similarity score quantifies the geometrical and chemical
similarity of multiple pockets upon alignment of their corre-
sponding MIFs, and gives valuable clues about the structural
correlation of proteins, even when belonging to distant and di-
verse families. Only GRID MIFs and multivariate statistical analy-
[
55]
sis are used to compare and cluster protein families. No se-
quence-related information, ligand similarity or side effect rela-
tionship is needed. With respect to other methods based on
pseudocenters, which only represent the hydrophobic or H-
bond donor/acceptor nature of the residues lining a cavity,
BioGPS considers the MIFs generated by those residues accord-
ing to their environment and, thus, the energetics of a pocket.
Figure 1. Workflow of the drug-repurposing approach based on the BioGPS/
FLAPdock integrated technology.
&
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
2
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
to the template. The algorithm’s capability of retrieving pock-
ets belonging to the same template protein family is evaluated
by means of enrichment analysis. An established similarity
threshold (Global Product >0.7) is used for selecting similar
cavities. To avoid sites not completely occupied by ligands,
only cavities containing at least 50% of the co-crystallized
ligand volume are retained. Once cavity hits are selected, co-
crystallized ligands are extracted and filtered according to their
volume similarity with respect to known template inhibitors.
Results and Discussion
Thymidylate synthase can be present in various configurational
states: 1) the apo inactive form, 2) an inactive form complexed
with peptides binding the homodimer interface, 3) an active
form complexed with dUMP, and 4) an active form complexed
with dUMP and mTHF or an antifolate drug (ternary complex).
TS undergoes significant conformational rearrangements upon
dUMP binding. In particular, in the native unbound form the
181–197 loop, containing the catalytic cysteine, is rotated
~1808 with respect to the binary/ternary complex. Conse-
quently, the catalytic Cys195 thiol group is 10 ꢁ away from the
active site, confirming the inactivity of the enzyme in this con-
formational state. On the contrary, upon dUMP binding the
3) Ligand docking. The extracted ligands are docked within
the template binding site with the FLAPdock algorithm. The
most promising ligands are selected according to the FLAP S-
score value, to their pseudo-MIFs complementarity with the
pocket MIFs, and to the number of hydrogen bonds formed
with the residues lining the cavity.
[63]
enzyme assumes the closed active conformation.
Looking
4
) In vitro binding experiments. We used human thymidylate
for new unknown TS ligands, we focused on the protein
closed active conformation in presence of the dUMP substrate
(Figure 2).
synthase (TS) as a template. Human and bacterial TS catalyzes
the reductive methylation of 2’-deoxyuridine-5’-monophos-
phate (dUMP) to 2’-deoxythymidine-5’-monophosphate
(dTMP), using 5,10-methylenetetrahydrofolate as one-carbon
methyl donor (mTHF). The reaction evolves through the forma-
tion of a covalent bond between dUMP and the catalytic
Cys195, the entrance of mTHF into the binding site and the
transfer of a methyl group to dUMP, thus transforming it into
[
57]
dTMP. After its release, dTMP is phosphorylated by two suc-
cessive steps to 2’-deoxythymidine-5’-triphosphate (dTTP), an
essential precursor for DNA synthesis. This pathway is the sole
intracellular de novo source of dTTP. It follows that human TS
represents a good pharmacological target for anticancer drugs
and antimicrobials. Nucleotide- and folate-like inhibitors are, in
fact, used in cancer chemotherapy because of the cytotoxic ef-
[
58–60]
fects of thymidylate depletion.
Microbial TSs have also
been demonstrated to be suitable targets for antimicrobial
[
61,62]
agents.
The TS binding site was used as a template to screen the
entire PDB, looking for similar cavities. Human and bacterial TS
were reasonably identified as the most similar proteins, being
followed by apparently diverse candidates such as kinases, pro-
teases, phosphodiesterases, nuclear receptors, and chaperones,
among others. Statistical and network analyses were used to
rationalize the investigated biological space and the connec-
tion, in terms of similarity, among the selected pockets. The re-
sults illustrate the strength of this approach in that it is able to
automatically and quickly identify similarity between the same
and different protein families.
Figure 2. Ribbon representation of human TS. The binding site of one mo-
nomer is represented as yellow mesh lines; dUMP is shown as orange
capped sticks.
In the PDB many different apo forms of wild type human TS
have been deposited. There are a few binary complexes of the
inactive form co-crystallized with peptides binding at the
homodimer interface, while there are no binary complexes for
the active form with the dUMP substrate. Only four ternary
complexes, that is, PDB ID: 1HVY (hTS+dUMP+raltitrexed),
To identify possible new TS ligands among known chemicals,
ligands co-crystallized in the most similar pockets to the TS
template were subsequently docked with the FLAPdock algo-
rithm in the TS cavity. Interestingly, the most promising mole-
cules were inhibitors of casein kinase IIa (CKIIa), which
emerged as a TS off-target. Two co-crystallized ligands and
a series of related flavonoids were tested in vitro for inhibition
activity toward TS. Six compounds inhibited TS in the low mi-
cromolar range, thus supporting the potential of the BioGPS
approach for drug repurposing campaigns.
1I00
(hTS+dUMP+raltitrexed);
1JU6
(hTS+dUMP+
LY231514); 1JUJ (hTS+dUMP+LY231514) are present. Accord-
ing to the resolution and the backbone completeness, 1HVY
[63]
was selected for modelling the TS binding site. The antifo-
late raltitrexed was removed from the crystallographic struc-
ture, leaving only the dUMP and thus limiting the search to
the pocket normally occupied by mTHF or other antifolate
drugs. A careful structural and energetic analysis of binding
site water molecules was also performed, to identify structural
waters able to modify the cavity shape and the possible inter-
action with ligands. Waters 435 and 622, mediating the interac-
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
3
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
tion of the protein with raltitrexed and dUMP, respectively, and
thus contributing to the complex stabilization, were identified.
The Fixpdb tool implemented in BioGPS was used to calculate
the GRID OH2 probe energy of the aforementioned waters
within the binding pocket. The two selected molecules
ꢀ
1
showed an energy value lower than ꢀ8 kcalmol , a strong
evidence of their importance in complex formation and stabili-
zation. Thus, two different human TS structures were finally
modelled: 1) TS with dUMP and without any water molecules
(
(
TSp) and 2) TS with dUMP and two bridging water molecules
wat435 and wat622; TSpw). The corresponding pockets were
automatically identified by using the FLAPsite tool and used as
templates to perform the following screening and docking ex-
periments, with the perspective that the incoming ligands
might displace any water (TSp) or retain and exploit the exist-
ing ones (TSpw).
First step: screening the pockets
The two TS pockets were used as templates to screen the PDB
with BioGPS. In particular, 90025 pockets were screened,
which correspond to all protein cavities present in the PDB
and co-crystallized with a ligand at the moment the database
was downloaded (September 2014).
Figure 3. BioGPS virtual screening statistics. a) Distribution of the similarity
Global Product score calculated by the BioGPS virtual screening. TSp and
TSpw, colored red and cyan respectively, were used as template pockets.
Overlap areas are colored grey. b) Enrichment curves calculated for the two
virtual screenings. All the pockets in the dataset belonging to TS were con-
sidered actives, all the other pockets inactives. The curves identifying the
VSs performed using TSp and TSpw as templates are colored purple and
cyan, respectively.
BioGPS superposes cavities by aligning their GRID MIFs. Tem-
plate MIFs were compared with cavity database MIFs, calculat-
ing for each pairwise comparison a set of nineteen FLAP
scores, representing the similarity of the match (see the Experi-
mental Section for further details on BioGPS and FLAP). In par-
ticular, the Global Product (GlobP), the product of the four
principal FLAP scores, was used to evaluate the degree of simi-
larity between the templates and the candidate cavities. The
GlobP score ranges from 0, for no superposition, to 1, for a per-
fect pocket overlap complete with identical interactions, and
provides a global evaluation of both geometric and chemical
similarity. Thus, the 90025 candidate cavities were ranked ac-
cording to the GlobP score for both TSp and TSpw. The distri-
butions of the GlobP scores for TSp and TSpw are reported in
Figure 3a, in purple and cyan respectively.
1336 cavities, relating to 606 unique proteins, and 4349 cavi-
ties, relating to 1513 unique proteins, were selected as hits for
TSp and TSpw, respectively (Figure 4). 1208 cavities (583 pro-
teins) were selected as common hits for both templates. To
consider both binding cases (water displacement [TSp] and
water-mediated binding [TSpw]), we retained all of the 4476
identified pockets, corresponding to 1536 proteins. To further
The plot suggests that the water molecules enhance the
promiscuity of the cavity. Indeed, using TSpw as a template,
the algorithm identifies a higher number of cavities with
higher similarity, that is, having higher GlobP (grey and cyan
bars). On the contrary, when TSp is used as template fewer
cavities presented such a high GlobP value. To demonstrate
the robustness of the procedure, the GlobP score was used to
evaluate the BioGPS performance in retrieving protein binding
sites that belong to the same family of the query (259 TS pock-
ets over 90025 pockets). The PDB entries of thymidylate syn-
thases were used to enrich the remaining cavities in the data-
base and the GlobP was used again as the ranking score. Fig-
ure 3b reports the enrichment curves obtained for each of the
two templates. In both cases about 80% of TS binding sites
were retrieved within the first 20% of the screened database,
confirming the strength of the BioGPS approach.
Figure 4. Cavity hits, and corresponding proteins, identified by the BioGPS
VS. The hits identified for the TSp and the TSpw templates are respectively
colored purple and cyan. The total number of hits is colored blue, while the
green bars correspond to the hits number filtered for the ligand fraction
volume.
To select the most similar cavities to the TS, a GlobP thresh-
[51]
old equal to 0.7 was set, according to previous analysis.
&
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
4
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
decrease this large number of cavities and focus on those
most similar to the templates, we considered only binding
sites containing a fraction of ligand volume (FV)>0.5, that is,
occupied by at least 50% of the ligand. This filtering was ap-
plied to discard pockets not completely occupied by the co-
crystallized ligand. Supposing to have a cavity A similar to
a cavity B, if cavity B contains only a fraction of its cognate
ligand, the transferability of that ligand to cavity A might be
misleading, because only a small part of the ligand is comple-
[
64,65]
mentary to that pocket.
After applying the filtering proce-
dure based on FV, we obtained overall 3770 cavities and 1297
proteins (Figure 4). These cavities contained 1361 different li-
gands, with some of them present in different pockets. Irrele-
vant compounds such as solvents, that is, ethylene glycol, glyc-
erol, or prosthetic groups were discarded. The remaining li-
gands were compared, in terms of volume, to known human
TS inhibitors retrieved from the ChEMBL database (see The Ex-
perimental Section for further details). Those having a volume
higher than the smallest TS ligand and lower than the largest
one were retained, to consider compounds in the same
volume range of known TS inhibitors and in the same drugga-
ble space. This led to a decreased set of 283 ligands belonging
to 135 proteins and 317 pockets (Table S1).
Figure 5. BioGPS-network. Similarity network of the 317 selected cavities
built according to the cavities pair Global Product estimated by BioGPS.
Each node represents a cavity and each edge between two nodes repre-
sents the similarity between them (the shorter the distance, the higher the
similarity). The different protein classes are color-coded according to the
legend. Singletons are reported below. Hit proteins, that is, containing the
ten selected ligands reported in Table 1, Figure 6, are identified by triangles.
potentially repurposed for phosphodiesterases (PDEs) and vice
versa, while it would be harder, in principle, to relocate PDE li-
gands in nuclear receptors (see the Supporting Information for
a detailed description of the pocket distribution).
Connecting the pocketome
To determine how the selected pockets are connected and to
identify how many unique pocket types we are dealing with,
we performed a connectivity analysis on the 317 hits, con-
structing a pairwise similarity matrix. We then built a binding
From a polypharmacology perspective, pocketomes could
be easily analyzed and investigated for multi-target therapies
or for unpredicted and unknown side effects. This underlines
the versatility of the BioGPS approach and the extent of the
possible related applications. We further characterized and sep-
arated the cavities according to their morphological and chem-
[
66,67]
site similarity network using Cytoscape,
where each node
represents a pocket and pockets are connected and clustered
if they have a GlobP higher than 0.4. The higher the GlobP for
a cavity pair, the closer the two cavities will be in the graph,
that is, more similar from a structural, chemical and energetic
point of view.
[68]
ical properties using Volsurf. Results are reported in the Sup-
porting Information (Figure S1).
The high color/cluster correspondence reported in Figure 5
shows how BioGPS is not only able to detect pocket similari-
ties, but also to cluster proteins according to their binding site
properties. In our study pockets are only described in terms of
four molecular interaction fields, that is, the shape, the hydro-
phobicity, the H-bond donor and acceptor character. These
seemingly simple properties that encode the energetic descrip-
tion of the cavities through GRID force field and the subse-
quent FLAP GlobP global similarity description are able to rep-
resent pockets to such an extent that proteins belonging to
the same family are easily picked up and clustered together.
No sequence or overall structure architecture is considered,
which means that binding site similarity is enough to classify
proteins into their corresponding families. This analysis has the
main advantage of depicting the connections among different
protein classes, regardless the similarity they have with the
original template, TS in the present case. In a drug repurposing
perspective, for instance, we could try to exchange ligands
among nuclear receptors and chaperones or nuclear receptors
and oxidoreductases, but with more difficulty among chaper-
ones and oxidoreductases. Chaperone ligands could also be
Second step: docking cognate ligands
From the 317 cavities we retrieved 283 co-crystallized different
ligands (Table S1). The ligand dataset was prepared considering
tautomers and protomers using MoKa and docked within TSp
and TSpw with FLAPdock, the docking tool implemented in
[56]
FLAP. About 80% of the compounds was first removed ac-
cording to the FLAP S-score. Only molecules with a score value
higher than 0.90 were retained, based on previous observa-
tions [data not shown]. The retained compounds were visually
inspected and further filtered according to the complementari-
ty of the pocket MIFs with the ligand pseudo-MIFs, and to the
number of hydrogen bonds formed with the residues lining
the cavity. According to the literature and to studies we previ-
ously performed, the visual inspection of the results by trained
operators can, indeed, strongly improve the success rate of
[69–72]
screening and docking experiments.
The ten most promising molecules were selected and are re-
ported in Figure 6 and Table 1. Ligands were retrieved from
casein kinase IIa, LFA-1 integrin, heat shock protein 90, b-secre-
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
5
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
The ligands, quite different from each other in terms of
shape, volume and chemical properties, present, predictably,
diverse interactions with the pocket residues. Only the p–p
contact with dUMP is maintained overall, with the exception of
E2M ligand (Figure 6c). Among the ten selected compounds,
ellagic acid and apigenin presented the most promising dock-
ing pose. In particular, ellagic acid makes hydrogen bonds with
Arg50, Asn112 and Tyr258, while apigenin is hydrogen
bonded to a pyrimidine carbonyl and to the hydroxy group of
the deoxy-ribose dUMP. In addition, slightly different apigenin
poses within the TS pockets show the formation of contacts
with the aforementioned residues, thus underlining the plasti-
city of the complex and the possibility of making several and
different interactions. Both ligands are involved in a p–p inter-
action with the dUMP pyrimidine moiety (Figure 6a,b). The
other molecules also mainly contact Arg50, Asn112, Tyr258
and dUMP. Additional hydrogen bonds are formed with Phe80,
Glu87, Ile108, Asp218, Asn226, Ala312 (Figure 6).
Interestingly, both ellagic acid and apigenin are known in-
hibitors of CKIIa. CKII is an attractive anti-neoplastic and antivi-
ral target, essential for cell viability and with many cellular tar-
[73]
gets. The catalytic subunits of CKII, a and a’, are constitu-
tively active, either alone or in combination with the b subunit,
a necessary property for the continuous need to phosphorylate
its numerous targets, but also potentially dangerous in neo-
[74]
plastic pathologies and viral infections.
Biological evaluation of ligands and synthesis of second-
round compounds
As mentioned, ellagic acid and apigenin were the most inter-
esting compounds, and are both inhibitors of CKIIa, having
[75]
IC =40 nm and K =740 nm, respectively. We therefore ex-
50
i
perimentally evaluated them for inhibition toward our tem-
plate TS and performed enzyme kinetics on both of them.
TS is a double substrate enzyme and competition with re-
spect to dUMP or with respect to the folate cofactor can be
performed. Following the computational model, dUMP was
considered the fixed substrate and used at saturating concen-
tration, while the folate cofactor was used as the limiting sub-
strate for the competition kinetic. Results are reported in
Figure 6. Docking poses of the ten best selected compounds within the hTS
binding pockets as predicted by FLAPdock. The PDB ligand code and the
ligand name are reported as follows: a) REF, ellagic acid, 2,3,7,8-tetrahydroxy-
chromeno[5,4,3-cde]chromene-5,10-dione; b) AGI, apigenin, 5,7-dihydroxy-2-
Table 2, first part. Ellagic acid showed a K value of 16 mm,
(4-hydroxyphenyl)-4H-chromen-4-one; c) E2M, cis-4-{[2-({4-[(1E)-3-morpholin-
i
4
-yl-3-oxoprop-1-en-1-yl]-2,3-bis(trifluoromethyl)phenyl}sulfanyl)phenoxy]me-
while apigenin only showed poor inhibition of 6% at 100 mm.
Considering that apigenin did not fulfill the H-bond potential
of TS pockets (Figure 6b) and that flavonoids are known CKII
inhibitors, we extended the analysis to other purchased and
in-house synthesized flavonoids (compounds 1–4, Table 2,
second part; see The Experimental Section for synthesis de-
tails). Moreover, both compounds are known to be promiscu-
ous inhibitors and we wanted to further validate the TS inhibi-
tion through a second round of compound selection, similar to
those selected but with better properties. Docking simulations
were used to guide this second selection, with the aim of
maintaining the same orientation within the binding site but
increasing the number of hydrogen bonds formed with the
residues lining the cavity. We ended up with eleven com-
pounds, among which five showed relevant inhibition activity.
thyl}cyclohexanecarboxylic acid; d) H71, 8-[(6-iodo-1,3-benzodioxol-5-
yl)thio]-9-[3-(isopropylamino)propyl]-9H-purin-6-amine; e) 0VA, N-[N-(4-
amino-3,5-dichlorobenzyl)carbamimidoyl]-3-(4-methoxyphenyl)-5-methyl-1,2-
thiazole-4-carboxamide; f) CLM, chloramphenicol, 2,2-dichloro-N-[(1R,2R)-2-
hydroxy-1-(hydroxymethyl-2-(4-nitrophenyl)ethyl]acetamide; g) APJ, N -1H-
benzimidazol-5-yl-N -(3-cyclopropyl-1H-pyrazol-5-yl)pyrimidine-2,4-diamine;
h) 3ND, (3S,4R)-N-(7-chloro-1-oxo-1,4-dihydroisoquinolin-6-yl)-4-(4-chlorophe-
nyl)pyrrolidine-3-carboxamide; i) 2D3, methyl 3-isoxazol-5-yl-5-methyl-1H-
pyrazole-4-carboxylate; j) 37D, methyl 5-furan-2-yl-3-methyl-1H-pyrazole-4-
carboxylate. dUMP is shown in orange color sticks. Residues involved in hy-
drogen bonding the ligands are labeled. When present, water molecules are
displayed as red spheres. Complexes a)–h) were obtained by docking in
TSpw, complexes i),j) by docking in TSp.
2
4
tase 1, chloramphenicol acetyltransferase 3, serine/threonine-
protein kinase/endoribonuclease IRE1 and Rho associated pro-
tein kinase.
&
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
6
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
Table 1. Structures and properties of the most promising compounds selected by the BioGPS/FLAPdock integrated approach.
PDB code
Structure
Co-crystallized protein (PDB ID)
REF
casein kinase IIa (2ZJW)
casein kinase IIa (3AMY)
AGI
E2M
LFA-1 binding domain (3E2M)
H71
Hsp90 (2FWZ)
0VA
b-secretase 1 (4FSE)
CLM
APJ
chloramphenicol acetyltransferase 3 (4CLA)
Ser/Thr protein kinase/endoribonuclease IRE1 (3FBV)
3ND
Rho-associated protein kinase (3NDM)
2
3
D3
7D
Hsp90 (2YE8)
Hsp90 (3HZ1)
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
7
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
Table 2. Structures and inhibition values for the compounds tested against hTS.
[a,b]
Name
M
r
[Da]
Structure
i
IC50/K [mm]
ellagic acid
302.19
270.24
238/16
apigenin
1440/94
morin
302.24
286.24
286.24
304.25
290.27
286.24
302.24
45/2.9
52/3.4
NI
fisetin
datiscetin
taxifolin
catechin
kaempferol
quercetin
NI
NI
57/3.7
NI
compound 1
compound 2
256.25
256.25
101/6.6
NI
&
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
8
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
Table 2. (Continued)
[a,b]
Name
M
r
[Da]
Structure
i
IC50/K [mm]
compound 3
compound 4
272.25
286.24
273/18
61/4.0
[
61]
[
a] Standard error within ꢁ20% of the given value. [b] NI: no inhibition at the solubility limit.
In particular, seven were purchased from vendors and four
compounds 1–4) were synthesized in house, as reported in
Scheme 1. Among the acquired compounds, morin, fisetin, and
kaempferol presented an inhibition effect in the low micromo-
in vitro analyses and chemical synthesis, thus demonstrated to
be fundamental for compound selection and optimization.
Both ellagic acid and apigenin are known inhibitors of other
(
[78]
proteins apart from CKII. According to Drugbank, ellagic acid
is active against carbonic anhydrase 1, 2, 3, 4, 5A mitochondri-
al, 5B mitochondrial, 6, 7, 9, 12, 14, CKIIa, cAMP-dependent
protein kinase a, protein kinase a and b type, tyrosine-protein
kinase SYK, cytochrome P450 1 A1 and 2E1. No specific infor-
mation is reported for apigenin in Drugbank, while ChEMBL re-
lar range (K : 2.9, 3.4, 3.7 mm, respectively). Compounds 1 and 4
i
showed a K <10 mm (K of 6.6 and 3.3 mm, respectively), com-
i
i
pound 3 a poor inhibition of 8% at 25 mm, while 2 could not
be studied because of the low solubility. According to ChEMBL,
known TS inhibitors present K values ranging from low nano-
i
[
75–77]
[75]
molar to high micromolar values.
Our hits, having, in gen-
ports a number of possible targets. Among these we find al-
eral inhibition values in the low micromolar range represent
good potential starting point for the development and optimi-
zation of more potent inhibitors. As well as ellagic acid, they
all behaved as competitive inhibitors of folic acid, nevertheless
most of them presented poor water solubility and we had to
test very low inhibitor concentration in the 10–100 mm range.
Docking poses of the active compounds within the TS bind-
ing site are reported in Figure S2. The presence of more polar
groups on the molecules allowed the formation of additional
hydrogen bonds within the target pocket, which might explain
their higher activity. While apigenin was able to form a few hy-
drogen bonds (in the pose shown in Figure 6b only the dUMP
substrate is contacted) the five active flavonoids interact with
Arg50, Glu87, Asn112, Asp218, Val223, Tyr258 and Ala312.
This higher protein–ligand structural complementarity can in-
crease the inhibitors’ potency but also reduce the promiscuous
character of these molecules. Docking simulations, along with
dehyde dehydrogenase 1A1, cytochrome P450 2C9, 2C19, 2D6
and 3A4, DNA polymerases, ERBB1, MAP kinase ERK2 and
p38a, acetylcholinesterase, cholinesterase, b-secretase 1, pyru-
vate kinase, protein kinase Ca, Rho-associated protein kinase 2,
tyrosine-protein kinase SYK, phosphodiesterase 5A, and others.
Despite their rather promiscuous profile, for both ellagic acid
and apigenin TS has not been previously reported as a poten-
tial target. In addition morin, fisetin and kaempferol, according
to ChEMBL and PubChem did not count TS, neither human or
bacterial, among their possible targets.
Pockets exchange
The above described results suggest that TS and CKIIa pockets
present similar properties. We compared the pockets MIFs and
reported their superimposition in Figure 7. The pockets shape
and dimension present a certain degree of similarity, with vol-
Scheme 1. A) Synthesis of compounds 1–3: a) SOCl
1m in dry DMC), dry DMC, 08C!RT.
2 2 2 3
, EtOH, RT; B) Synthesis of compound 4: a) NaOH (3m), EtOH, RT; b) H O , NaOH (1m), EtOH, RT; c) BBr
(
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
9
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
their ligand image, that is, the volume, the shape and the
chemical features a ligand encounters once entering into a spe-
cific pocket. Pockets are represented and compared according
to their GRID molecular interaction fields, also encoding the
energetics of the pockets, differently from many other ap-
proaches. Once similar pockets are identified, co-crystallized li-
gands, or any other known inhibitor can be cross-docked be-
tween the template and the queries, or among the related
queries to look for new targets and applications.
We ran this pipeline for the specific TS case, identifying
Hsp90, the estrogen receptor, the vitamin D3 receptor, differ-
ent kinases, transferarses and phosphodiesterases and others
as possible related targets. We selected some ligands from
CKIIa, ellagic acid and flavonoids, and found them to be inhibi-
tors of the TS template in the low micromolar range.
The specific case described here supports the applicability of
the BioGPS/FLAPdock integrated pipeline. Our aim is not to
propose ellagic acid or flavonoids as new potential hTS inhibi-
tors, but to illustrate the possibility of identifying similar pro-
teins in a new, fast and automatic way, and subsequently re-
purposing known drugs or ligands for specific proteins. As pre-
viously described, given a template the identification of the
most similar pockets and, consequently, of the possible off-tar-
gets is totally automatic. In the pipeline we reported this corre-
sponds to the first BioGPS step. After the selection of possible
related targets, ligands can be exchanged by docking simula-
tions, then tested and improved by means of in silico analysis,
in vitro analysis, and chemical synthesis. This second step
allows the rationalization of the protein–ligand interaction, the
potential improvement of the complex stability and the iden-
tification/development of more potent and specific inhibitors.
The potential of this pipeline is extremely large. Apart from
the identification of new inhibitors among known ligands for
a specific target, BioGPS has a variety of applications and pos-
sible uses. Pocketomes can be easily and rapidly analyzed for
identifying targets likely responsible for unpredicted side ef-
fects. Further, the similarity of targets involved in specific path-
ways or over-expressed in pathological conditions can be in-
vestigated for designing multi-target therapies.
Figure 7. Superimposition of hTS and CKII binding sites. a) Shape superim-
position, TS pocket (PDB ID: 1HVY) is shown as yellow surface and CKIIa
(PDB ID: 2ZJW) pocket as magenta mesh lines. b),c),d) Superimposition of
hydrophobic, H-bond donor and H-bond acceptor molecular interaction
fields, respectively; TS MIFs are shown as solid surfaces while CKIIa MIFs as
mesh lines.
3
umes of 1390 and 1675 ꢁ , for TS and CKIIa, respectively (Fig-
ure 7a). The hydrophobic MIFs show the higher level of over-
lap, in particular in the region where the hydrophobic core of
the selected ligands is involved in p–p interactions with the
dUMP in TS. A lower superimposition can be observed for the
H-bond donor and H-bond acceptor group. Clearly, a total
overlap of pockets’ chemical and geometrical properties is not
expected, in particular when dealing with such large cavities.
Ligands can differently adapt and occupy only portions of the
pockets. For instance we have previously reported the similari-
2
+
ty of the ERa and the SERCA (sarcoplasmic reticulum Ca ion
channel ATPase) cavity, which is a known off-target for selec-
The main advantage is represented by the algorithm’s capa-
bility of depicting the real structural and energetic scenario of
a protein binding site, totally independent of any other protein
or ligand-related information, apart from the pocket definition
(which is itself automatically defined). The simplicity of the
pocket search, the rapid and semi-automated procedure,
makes it a promising and valuable tool for modeling polyphar-
macology, drug repurposing and side effects.
[
79]
tive estrogen receptor modulators. The two superposed cav-
ities showed only a 62% overlap of the volume. Again the
highest similarity scores were detected for the hydrophobic
[
51]
and the H-bond donor MIFs. Also it must be remembered
that proteins and ligands are flexible and so is the image that
ligands produce in their binding site. There is no expectation
that ligands will be totally complementary toward the new
target, but recognize at least a part of the image they pro-
duced in the original pocket.
Experimental Section
Conclusions
Cavity identification: The FLAPsite algorithm is used for the iden-
tification of cavities in 3D protein structures. By embedding the
protein structure into a 3D grid with a spatial resolution of 1.0 ꢁ,
[51]
Recognizing the potential of drug repurposing strategies, we
have presented here the BioGPS/FLAPdock approach for off-
target identification and repositioning applications. Given
a protein cavity this approach is able to automatically search
the PDB and identify the most similar binding sites in terms of
the algorithm identifies pocket points using the GRID probe H
[53]
(
shape). For each point a buriedness-index is calculated. Points
with a buriedness-index lower than a specific threshold are discard-
ed. Two morphological operations, erosion, and dilation, are ap-
&
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
10
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
plied to the remaining points for removing small anomalies and
connecting areas. The hydrophobic probe DRY is used to prioritize
hydrophobic cavities usually targeted by drugs. The FLAPsite pro-
cedure was applied to 1) human TS (PDB ID: 1HVY) with dUMP,
without any water molecule (TSp); 2) human TS (PDB ID: 1HVY)
with dUMP and two bridging water molecules (wat435 and
wat622; TSpw); 3) all PDB protein structures co-crystallized with
a ligand. Only cavities containing a ligand were selected for the fol-
lowing steps of the pipeline: (1–2) two TS template cavities and
cluding the Global Sum and Global Product, FLAP also calculates
a Distance Score, representing the overall similarity derived from
a combination of all calculated similarity scores computed for the
candidates and the template, that is, the protein binding site. The
Global Product score was considered for hit selection in this study;
it ranges between 0 and 1, where the higher the score the more
similar are the two entities.
Hit (pockets) selection: The Global Product score was set to 0.7 as
restrictive threshold for selecting the most similar cavities to the
hTS templates; 4476 cavities were selected. Then, for each cavity
the FV was calculated as the ratio of the ligand volume contained
within a cavity over the total volume of the ligand. Only cavities
with FV>0.5 were selected. We thus obtained 3770 cavities, from
which 1361 co-crystallized ligands were extracted. Irrelevant com-
pounds such as solvents, that is, ethylene glycol, glycerol, or pros-
thetic groups were discarded. The remaining ligands were com-
pared with a set of known human TS inhibitors, retrieved from the
(
3) 90025 dataset cavities.
Virtual screening with BioGPS: We collected from the PDB the
structures of all proteins co-crystallized with a ligand. The Fixpdb
tool was used for processing the protein residues, solvent mole-
cules, co-crystallized ligands, cofactors and ions contained in the
PDB protein structures. All nucleic acids, ligands and water mole-
cules co-crystallized with the protein were removed, while cofac-
tors were retained (i.e., NAD, FAD, GSH). Additionally, to retain ions
involved in interactions with the protein residues, a defined GRID-
[75]
ChEMBL database. A bioactivity search was performed and 354
+
2
+2
+2
+2
inhibitors with known K toward human TS were retrieved. Mole-
energy threshold for Cu , Fe , Zn , Mg was applied. Binding
sites were then detected by using the FLAPsite algorithm (90025
binding sites; September 2014). MIFs were calculated for each
binding site and stored in a database. The BioGPS technology was
used to compare two TS pocket templates against 90.025 MIFs
cavities dataset. The BioGPS algorithm compares binding sites by
i
cules likely designed to displace the dUMP from the binding site
were discarded. The remaining 328 molecules were analyzed with
Volsurf to calculate their volume. Their minimum and maximum
volumes were used as cut-offs. Unknown ligands (extracted form
cavities) having a volume higher than the smallest hTS inhibitor
and lower than the biggest one were retained. This process yielded
283 ligands belonging to 135 proteins and 317 pockets (Table S1).
We thus confined the research to ligands occupying the same
druggable space of known TS inhibitors. This filter was applied to
increase the probability of success and to create a focused and re-
duced dataset to perform in vitro analyses and validate the pipe-
line. Discarded ligands will be further analyzed to investigate the
effect of enlarging the chemical space on the compounds activity.
[51,55]
means of their MIF similarity,
and exploits the technology im-
[56]
plemented in FLAP. FLAP (Fingerprints for Ligands and Proteins)
is a virtual screening algorithm developed and licensed by Molecu-
lar Discovery Ltd. (www.moldiscovery.com). Several VS campaigns
have been successfully performed with FLAP and are reported in
[
69,70,80–85]
the literature,
tionalization.
as well as binding mode prediction and ra-
[
86,87]
Initially, the approach uses the GRID force field to
evaluate the type, strength and direction of the interactions that
a cavity is capable of making. The GRID probes H, DRY, O, and N1
are used to compute the shape, the hydrophobic interactions, the
H-bond donor interactions and the H-bond acceptor interactions
respectively for each cavity considered in the analysis. Because
a simple comparison of the entire MIF areas might be computa-
tionally expensive, the algorithm decreases the information by se-
lecting a number of representative points, called hotspots, propor-
tional to the energy and the volume of each MIF. All possible com-
binations of four hotspots (called quadruplets) are generated and
Connectivity analysis: Network graphs reported in Figure 5 and
[66]
S1 were built with Cytoscape. In the BioGPS network (Figure 5)
the GlobP score was used to define the distance among the 317
pockets. To calculate a GlobP value for each pocket pair, a BioGPS
“
all vs. all” VS approach was adopted and each cavity was com-
pared and scored against the others. Because the distance is
meant to be lower than a threshold value to connect two objects
in a network graph, pocket pairs were joined when having
a 1ꢀGlobP value <0.6, that is GlobP>0.4. To avoid over-connec-
tions between TS and the other cavities in the graph (all cavities
identified as similar from the first BioGPS VS had at least GlobP>
[56]
stored in a fingerprint, named Common Reference Framework.
The BioGPS algorithm compares two cavities by comparing such
Common Reference Frameworks in a pairwise manner. This ap-
proach searches for the largest number of favorable quadruplet su-
perpositions. When the quadruplets of template and candidate
cavities match the feature types H (shape), DRY (hydrophobic), O
0
.7 with respect to TS), stricter thresholds were set. 1ꢀGlobP<0.2
(GlobP>0.8) and 1ꢀGlobP<0.05 (GlobP>0.95) were the thresh-
olds considered to connect TS with the other pockets and to itself,
respectively.
(
H-bond donor), N1 (H-bond acceptor) within certain distances (the
Molecular docking: FLAPdock is a docking approach implemented
tolerance is fixed to <1 ꢁ), the algorithm overlaps their 3D cavity
structures with a specific orientation, according to the matching
quadruplet, and then calculates the MIF similarity of the overlap-
ping areas. The final superposition is called a “solution” and is
quantitatively scored by considering the corresponding MIFs simi-
larity, summarized in nineteen different scores. FLAP first calculates
scores representing the degree of volume overlap for each of the
probes (and of the corresponding generated MIFs) being used indi-
vidually, that is, H, DRY, O and N1, and then combines these scores
in order to produce probe-combination scores. In addition FLAP
calculates two Global scores, the Global Sum, which is produced
by summing all the scores of the individual probes together, and
the Global Product (GlobP), produced by multiplying all the scores
of the individual probes together. Once the Probe scores for the in-
dividual probes and their combinations have been calculated, in-
[56]
[53]
in the software FLAP, based on GRID MIF similarities,
com-
bined with classical energetics. FLAPdock follows a molecular frag-
mentation approach, subsequent placement of each fragment in
the site of the target, followed by incremental construction of the
molecule. At each phase of the docking, a number of solutions are
generated, scored, and a subset retained for the subsequent
phase. A set of poses for the starting fragment is generated using
FLAP quadruplet alignment of the fragment conformer atom quad-
ruplets, and the receptor site GRID MIF minimal points. In this way
hundreds of thousands of poses are typically generated for this
starting fragment. As a first step, the poses are scored using
a weighted sum of the FLAP field similarities, including shape,
donor, acceptor, and hydrophobic similarity. A second scoring step
calculates Lennard-Jones and dielectric corrected Coulombic ener-
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
11
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
getic terms for this subset; the solutions are then ranked according
to the combined score, subjected to RMS clustering, and the best
scoring pose in each cluster retained. Internal validation (re-dock-
ing of X-ray ligands) has shown that one of the top five poses con-
tains a pose within 2.0 ꢁ of the X-ray position in more than 90%
of the cases [unpublished results]. Recent simulations of covalent
docking demonstrated to provide reliable results in agreement
with experimental data [work in preparation]. The most promising
compounds were ultimately selected according to the FLAP S-
score value, to the number of hydrogen bonds made with the sur-
rounding residues and to the complementarity of the pocket MIFs
with the ligand pseudo-MIFs. Pseudo-MIFs correspond to the pro-
jection of the MIFs on the atoms that generate them. Ligand tauto-
mer and protomer enumeration was performed with MoKa 2.6
prior to the docking simulations. Docking analyses were performed
within both TSp and TSpw.
(100, 50, 25 or 10 mm) were prepared, decreasing the concentra-
tion until opalescence in the solution and/or scattering effects dis-
appeared. Thus, inhibition assays were performed at the com-
pounds’ maximum solubility. Given the impossibility of gradually
increasing the compounds’ concentration and experimentally de-
termining the IC50 value, IC50 and K were calculated from the inhib-
i
[90]
ition percentage as reported. Values might be underestimated.
Chemistry: The following compounds were purchased from
Sigma–Aldrich: ellagic acid (CAS: 476-66-4), apigenin (CAS: 520-36-
5
), morin hydrate (CAS: 654055-01-3), fisetin (CAS: 345909-34-4),
datiscetin (CAS: 480-15-9), taxifolin (CAS: 480-18-2), (+)-catechin
CAS: 154-23-4), kaempferol (CAS: 520-18-3), quercetin (CAS: 117-
9-5). In Table 2 the molecular weight of morin, fisetin, quercetin is
(
3
reported as anhydrous basis. All the reagents and solvents used for
the synthesis of compounds 1–4 were purchased from Sigma–Al-
drich and used without further purification. Silica gel plates (Merck
1
13
Protein expression and purification: Human TS was cloned in the
F254) were used for thin-layer chromatography. H and C NMR
spectra were recorded on a Bruker FT-NMR AVANCE 400. Chemical
shifts (d scale) are reported in parts per million downfield from tet-
ramethylsilane as internal standard. Splitting patterns are designat-
ed as follows: s, singlet; d, doublet; t, triplet; q, quadruplet; m,
multiplet; brs, broad singlet; and dd, double doublet. Silica gel
Merck (60–230 mesh) was used for column chromatography. Melt-
ing points were determined with a Stuart SMP3 and they are un-
corrected. Mass spectra were obtained on a 6520 Accurate-Mass Q-
TOF LC–MS. The synthetic procedures for the synthesis of com-
pounds 1–4 are reported in Scheme 1. Compounds 5 and 6 are
the intermediates for the synthesis of compound 4.
[88]
pQE80 L system, as reported. The recombinant protein was ex-
pressed in DH5a Escherichia coli strain. The expression vector
codes for a hexa-histidine tag at the N-terminus of the gene prod-
uct, designed to facilitate the purification of the recombinant pro-
tein through immobilized metal affinity chromatography. Bacteria
(
DH5a/pQE80L) were grown in LB medium containing (2 L) ampicil-
ꢀ1
lin (50 mgmL ). The solution was centrifuged at 378C, 120 rpm,
until the OD600nm reached a value of 0.6. TS expression was induced
adding isopropyl-b-d-thiogalactopyranoside (IPTG, 1 mm) and incu-
bating the culture for 4 h, 378C, 120 rpm. Cells were then centri-
fuged at 4000 rpm for 30 min at 48C. The cell pellet was suspend-
ed in buffer A (20 mm NaH PO , 30 mm NaCl, 20 mm imidazole,
2
4
General procedure for the synthesis of hydroxylated flavanones
pH 7.5) containing Completeꢂ (protease inhibitor) and sonicated in
an ice bath. The broken cells were centrifuged for 40 min at
(1, 2, and 3): To a stirred mixture of 2’,5’-dihydroxyacetophenone
(0.300 g, 1.97 mmol) and the appropriate aldehydes (1 equiv) in ab-
1
2000 rpm, 48C, and the pellet was discarded. The supernatant
solute EtOH (2 mL), thionyl chloride (120 mL) was added dropwise
over 5 min. The reaction was stirred at room temperature for 6 h.
EtOH and excess thionyl chloride were removed under reduced
pressure on a rotary evaporator. Column chromatography was car-
ried out to purify the desired product (eluent system: cyclohexane/
ethyl acetate 9.8/0.2).
was treated with streptomycin (10%), stirred for 10 min at 48C and
centrifuged for 30 min, 12000 rpm, 48C. The discarded pellet and
the supernatant were filtered (0.8/0.45 mm filters) and loaded on
a Ni-HTP column pre-equilibrated with buffer A. The enzyme was
eluted with buffer B (20 mm NaH PO , 30 mm NaCl, 1m imidazole,
2
4
pH 7.5). The fractions with enzyme were collected, pooled and
loaded on a HiTrap Desalting column to change the buffer with
6-Hydroxy-2-(4-hydroxyphenyl)chroman-4-one (1): was isolated
[89]
1
buffer C (20 mm NaH PO , 30 mm NaCl, pH 7.5). Only fractions
as a yellow solid with 31% yield; mp: 2308C. H NMR (CD OD,
2
4
3
with detected enzymatic activity were collected.
400 MHz): d=7.33 (d, 2H, J2’,3’/5’,6’ =8.6 Hz, H-2’+H-6’), 7.22 (d, 1H,
J5,7 =3.0 Hz, H-5), 7.04 (dd, 1H, J7,8 =8.9 Hz, J7,5 =3.0 Hz, H-7), 6.90
Enzymatic activity and inhibition assays: TS enzymatic activity
was measured spectrophotometrically (Beckman DU640) by moni-
toring the absorbance increase at 340 nm, for 3 min during the oxi-
dation reaction of the substrate THF to 7,8-dihydrofolate. KM values
were determined for both mTHF and dUMP varying the substrate
concentrations. The concentration ranges for KM were 2–80 mm for
mTHF and 3–150 mm for dUMP. Values of kcat and specific activity
were determined by varying the enzyme concentration (0.04–
(
5
1
d, 1H, J8,7 =8.9 Hz, H-8), 6.84 (d, 2H, J3’,2’/5’,6’ =8.6 Hz, H-3’+H-5’),
.34 (dd, 1H, J2,3b =13.3 Hz, J2,3a =2.8 Hz, H-2), 3.08 (dd, 1H, J3b,3a
=
7.0 Hz, J3b,2 =13.3 Hz, Hb-3), 2.74 ppm (dd, 1H, J =17.0 Hz,
3a,3b
13
J3a,2 =2.8 Hz, Ha-3); C NMR (CD OD, 100 MHz): d=193.46, 157.50,
1
3
55.63, 151.51, 130.09, 127.59 (2C), 124.55, 120.77, 118.76, 114.90
(2C), 109.95, 79.45, 44.00 ppm; ESI-HRMS calcd for C H O [M+
15
13
4
+
H] 257.0808, found 257.0805.
0
.3 mm). The reaction mixture contained 50% of assay buffer (TES,
6-Hydroxy-2-(3-hydroxyphenyl)chroman-4-one (2): was isolated
as a yellow solid with 26% yield; mp: 2408C. H NMR (DMSO,
400 MHz): d=9.51 (brs, 1H, 3’-OH), 9.42 (brs, 1H, 6-OH), 7.20 (dd,
1H, J5’,4’ =8.2 Hz, J5’,6’ =7.4 Hz, H-5’), 7.12 (d, 1H, J5,7 =2.9 Hz, H-5),
7.04 (dd, 1H, J7,8 =8.8 Hz, J7,5 =2.9 Hz, H-7), 6.95 (d, 1H, J8,7 =8.8 Hz,
H-8), 6.91 (m, 2H, H-6’+H-2’), 6.76 (d, 1H, J4’,5’ =8.2 Hz, H-4’), 5.47
1
N-[tris(hydroxymethyl)methyl]-2-aminoethanesulfonic acid (100
mm), MgCl2 (50 mm), formalin (13 mm), EDTA (2 mm), pH 7.4, b-
mercaptoethanol (150 mm)), the enzyme (0.1 mm), mTHF (50 mm),
dUMP (120 mm) and water to 800 mL. The reaction was initiated
when dUMP was added to the reaction mixture. The selected com-
pounds were evaluated against recombinant hTS and the inhibi-
tion percentage was determined for a 10–100 mm compound con-
centration range. The molecules (10 mm) were solubilized in
DMSO. The inhibition percentage was determined upon evalua-
tion of the differential optic depth (DOD)/min ratio. It was not pos-
sible to perform a detailed study of the inhibition activity for all
the compounds because of their poor aqueous solubility. Samples
(dd, 1H, J2,3b =12.7 Hz, J2,3a =2.6 Hz, H-2), 3.10 (dd, 1H, J3b,3a
=
16.8 Hz, J3b,2 =12.7 Hz, Hb-3), 2.76 ppm (dd, 1H, J =16.8 Hz,
3a,3b
13
J3a,2 =2.6 Hz, Ha-3); C NMR (DMSO, 100 MHz): d=192.17, 157.90,
154.81, 152.03, 141.11, 129.99, 124.98, 121.33, 119.46, 117.43,
115.69, 113.79, 110.39, 79.08, 44.22 ppm; ESI-HRMS calcd for
[89]
+
C H O [M+H] 257.0808, found 257.0808.
15
13
4
&
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
12
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
2
-(3,4-Dihydroxyphenyl)-6-hydroxychroman-4-one (3): was isolat-
(DMSO, 100 MHz): d=172.58, 154.46, 148.82, 147.92, 146.30,
1
ed as an orange solid with 26% yield; mp: 2208C. H NMR (DMSO,
00 MHz): d=9.35 (brs, 1H, OH), 8.90 (brs, 2H, OH), 7.11 (d, 1H,
J5,7 =3.0 Hz, H-5), 7.02 (dd, 1H, J7,8 =8.8 Hz, J7,5 =3.0 Hz, H-7), 6.91
d, 1H, J8,7 =8.8 Hz, H-8), 6.90 (m, 1H, H-2’), 6.75 (m, 2H, H-5’+H-
’), 5.35 (dd, 1H, J2,3b =12.8 Hz, J2,3a =2.6 Hz, H-2), 3.10 (dd, 1H,
J3b,3a =16.8 Hz, J3b,2 =12.8 Hz, Hb-3), 2.68 ppm (dd, 1H, J
145.50, 137.72, 123.35, 122.95, 122.52, 120.36, 119.95, 116.03,
+
4
115.66, 107.27 ppm; ESI-HRMS calcd for C H O [M+H]
15
11
6
287.0550, found 287.0552.
(
6
=
3a,3b
13
Acknowledgements
1
6.8 Hz, J3a,2 =2.6 Hz, Ha-3); C NMR (DMSO, 100 MHz): d=192.52,
1
54.99, 151.90, 146.01, 145.62, 130.49, 124.93, 121.26, 119.42,
We thank Molecular Discovery Ltd. for supplying BioGPS and
FLAPdock software for the virtual screening and docking analy-
ses. We also acknowledge AIRC 2015, grants IG10474 and
IG16977 given to M.P.C. We thank Dr. Simon Cross for critically
reading and revising the manuscript.
118.27, 115.77, 114.74, 110.36, 79.22, 44.13 ppm; ESI-HRMS calcd for
+
C H O [M+H] 273.0757, found 273.0759.
1
5
13
5
(E)-3-(3,4-Dimethoxyphenyl)-1-(2-hydroxy-5-methoxyphenyl)-
prop-2-en-1-one (6): To a solution of 2’-hydroxy-5’-methoxyaceto-
phenone (0.724 g, 4.36 mmol) and 3,4-dimethoxybenzaldehyde
(
0.724 g, 4.36 mmol) in EtOH (12 mL), an aqueous solution of
NaOH (3m, 4.4 mL) was added. The reaction was stirred at room
temperature overnight. The reaction mixture was cooled in an ice-
water bath and acidified to pH 2 with concentrated HCl (37%). The
solid formed was filtered and re-crystallized from EtOH to obtain
Keywords: drug discovery
screening · thymidylate synthase · cross docking
· drug repurposing · virtual
compound 6 as a red solid (0.781 g, 57% yield); mp: 114–1158C.
[
1] D. Farina, F. Spyrakis, A. Venturelli, S. Cross, D. Tondi, M. P. Costi, Curr.
Med. Chem. 2014, 21, 1405–1434.
1
3
H NMR (CDCl , 400 MHz): d=12.29 (brs,1H, OH), 7.92 (d, 1H, JB,A
=
1
2
5.7 Hz, H ), 7.49 (d, 1H, J =15.7 Hz, H ), 7.43 (d, 1H, J
6’,4’
.9 Hz, H-6’), 7.31 (dd, 1H, J6,5 =8.3, J6,2 =2.0 Hz, H-6), 7.20 (d, 1H,
=
[2] F. Spyrakis, Curr. Drug Targets 2015, 17, 972–973.
[3] D. Tondi, S. Cross, A. Venturelli, M. P. Costi, G. Cruciani, F. Spyrakis, Curr.
Drug Targets 2015, 17, 983–1005.
B
A,B
A
J2,6 =2.0 Hz, H-2), 7.18 (dd, 1H, J4’,3’ =9.3, J4’,6’ =2.9 Hz, H-4’), 7.01 (d,
J3’,4’ =9.3 Hz, H-3’), 6.95 (d, 1H, J5,6 =8.3 Hz, H-5), 4.01 (s, 3H,
[
4] C. Holohan, S. Van Schaeybroeck, D. B. Longley, P. G. Johnston, Nat. Rev.
Cancer 2013, 13, 714–726.
13
-OCH ), 3.98 (s, 3H, -OCH ), 3.88 ppm (s, 3H, -OCH ); C NMR
3
3
3
[
[
[
5] B. Witkop, Proc. Am. Philos. Soc. 1999, 143, 540–557.
6] A. L. Hopkins, Nature 2009, 462, 167–168.
7] G. V. Paolini, R. H. Shapland, W. P. van Hoorn, J. S. Mason, A. L. Hopkins,
Nat. Biotechnol. 2006, 24, 805–815.
(
CDCl , 100 MHz): d=193.27, 157.84, 151.90, 151.65, 149.36, 145.87,
3
1
27.61, 123.61, 123.17, 119.87, 119.25, 117.83, 113.54, 111.22,
110.50, 101.01, 56.25, 56.09, 56.07 ppm; ESI-HRMS calcd for
+
C H O [M+H] 315.1227, found 315.1232.
1
8
19
5
[8] M. A. Yildirim, K. I. Goh, M. E. Cusick, A. L. Barabasi, M. Vidal, Nat. Bio-
technol. 2007, 25, 1119–1126.
2
-(3,4-Dimethoxyphenyl)-3-hydroxy-6-methoxy-4H-chromen-4-
[
9] A. L. Hopkins, Nat. Chem. Biol. 2008, 4, 682–690.
one (5): An aqueous solution of H O (30%, 620 mL) was added to
2
2
[
[
10] T. I. Oprea, J. Mestres, AAPS J. 2012, 14, 759–763.
an ice-cold suspension of compound 6 (0.781 g, 2.48 mmol) in
EtOH (12 mL) and 1m NaOH (5 mL). The mixture was allowed to
warm to room temperature and was stirred for 4 h. Then the reac-
tion mixture was cooled in an ice-water bath and acidified to pH 2
with concentrated HCl (37%). The solid formed was filtered and re-
11] Z. Simon, A. Peragovics, M. Vigh-Smeller, G. Csukly, L. Tombor, Z. Yang,
G. Zahoranszky-Kohalmi, L. Vegner, B. Jelinek, P. Hari, C. Hetenyi, I. Bitter,
P. Czobor, A. Malnasi-Csizmadia, J. Chem. Inf. Model. 2012, 52, 134–145.
[12] A. S. Reddy, S. Zhang, Expert Rev. Clin. Pharmacol. 2013, 6, 41–47.
[13] J. Bowes, A. J. Brown, J. Hamon, W. Jarolimek, A. Sridhar, G. Waldron, S.
Whitebread, Nat. Rev. Drug Discovery 2012, 11, 909–922.
[14] T. T. Ashburn, K. B. Thor, Nat. Rev. Drug Discovery 2004, 3, 673–683.
15] C. R. Chong, D. J. Sullivan, Jr., Nature 2007, 448, 645–646.
16] A. Bairoch, Nucleic Acids Res. 1991, 19, 2241–2245.
17] A. Bairoch, B. Boeckmann, Nucleic Acids Res. 1994, 22, 3578–3580.
crystallized from EtOH to obtain compound 5 as pale-yellow
powder (0.522 g, 64% yield); mp: 1908C. H NMR (CDCl , 400 MHz):
1
3
[
[
[
d=7.88 (dd, J6’,2’ =1.9 Hz, J6’,5’ =8.6 Hz, H-6’),7.84 (s, 1H, J2’,6’
.9 Hz, H-2’),7.55 (d, 1H, J5,7 =3.0 Hz, H-5), 7.51 (d, 1H, J8,7 =9.1 Hz,
H-8), 7.30 (dd, 1H, J7,8 =9.1 Hz, J7,5 =3.0 Hz, H-7), 7.14 (brs, 1H, OH),
.01 (d, 1H, J5’,6’ =8.6 Hz, H-5’), 3.98 (s, 3H, -OCH ), 3.93 (s, 3H,
=
1
[18] A. J. Bleasby, D. Akrigg, T. K. Attwood, Nucleic Acids Res. 1994, 22, 3574–
3577.
7
3
13
-OCH ), 3.88 ppm (s, 3H, -OCH ); C NMR (CDCl , 100 MHz): d=
[19] W. R. Pearson, Methods Enzymol. 1990, 183, 63–98.
[20] O. Roche, P. Schneider, J. Zuegge, W. Guba, M. Kansy, A. Alanine, K.
Bleicher, F. Danel, E. M. Gutknecht, M. Rogers-Evans, W. Neidhart, H.
Stalder, M. Dillon, E. Sjogren, N. Fotouhi, P. Gillespie, R. Goodnow, W.
Harris, P. Jones, M. Taniguchi, S. Tsujii, W. von der Saal, G. Zimmermann,
G. Schneider, J. Med. Chem. 2002, 45, 137–142.
3
3
3
1
1
5
3
72.70, 156.42, 150.69, 150.28, 148.82, 145.07, 137.51, 124.09,
23.79, 121.42, 121.11, 119.58, 110.91, 110.68, 103.83, 56.01, 55.96,
+
5.92 ppm; ESI-HRMS calcd for C H O [M+H] 329.1020, found
18
17
6
29.1024.
2
-(3,4-Dihydroxyphenyl)-3,6-dihydroxy-4H-chromen-4-one (4):
[21] N. Atias, R. Sharan, J. Comput. Biol. 2011, 18, 207–218.
[
22] S. Mizutani, E. Pauwels, V. Stoven, S. Goto, Y. Yamanishi, Bioinformatics
012, 28, i522–i528.
To a stirring solution of compound 5 (0.500 g, 1.52 mmol) in anhy-
drous dichloromethane (30 mL) under nitrogen at 08C, boron tri-
bromide in dichloromethane (1.0m, 13.7 mL, 13.7 mmol) was
added. The mixture was allowed to warm to room temperature
and stirred for 2 days. The reaction mixture was then cooled to
2
[
[
23] E. Pauwels, V. Stoven, Y. Yamanishi, BMC Bioinf. 2011, 12, 169.
24] Y. Yamanishi, M. Kotera, M. Kanehisa, S. Goto, Bioinformatics 2010, 26,
i246–54.
[
25] Y. Yamanishi, M. Kotera, Y. Moriya, R. Sawada, M. Kanehisa, S. Goto, Nu-
cleic Acids Res. 2014, 42, W39–45.
0
8C and MeOH (10 mL) was added. The reaction mixture was con-
centrated in vacuo. Water (10 mL) was added, the reaction was so-
nicated for 1 h and then left to stand. The solid was filtered to col-
[26] V. I. Pꢃrez-Nueno, M. Souchet, A. S. Karaboga, D. W. Ritchie, J. Chem. Inf.
Model. 2015, 55, 1804–1823.
lect compound 4 as a red solid (0.440 g, quantitative yield); mp:
[27] M. Campillos, M. Kuhn, A. C. Gavin, L. J. Jensen, P. Bork, Science 2008,
1
3
21, 263–266.
3
2
9
368C (dec). H NMR (DMSO, 400 MHz): d=7.73 (d, 1H, J8,7
=
[
28] J. Scheiber, B. Chen, M. Milik, S. C. Sukuru, A. Bender, D. Mikhailov, S.
Whitebread, J. Hamon, K. Azzaoui, L. Urban, M. Glick, J. W. Davies, J. L.
Jenkins, J. Chem. Inf. Model. 2009, 49, 308–317.
.2 Hz, H-2’), 7.58 (dd, 1H J=2.2, J=8.5 Hz, H-6’), 7.57 (d, 1H, J=
.1 Hz, H-8), 7.35 (d, 1H, J=3.0 Hz, H-5), 7.23 (dd, 1H, J7,8 =9.1 Hz,
13
J7,5 =3.0 Hz, H-7), 6.90 ppm (d, 1H, J=8.5 Hz, H-5’); C NMR
[
29] L. Yang, P. Agarwal, PLoS One 2011, 6, e28025.
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
13
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
[
30] F. Cheng, W. Li, Z. Wu, X. Wang, C. Zhang, J. Li, G. Liu, Y. Tang, J. Chem.
[66] P. Shannon, A. Markiel, O. Ozier, N. S. Baliga, J. T. Wang, D. Ramage, N.
Amin, B. Schwikowski, T. Ideker, Genome Res. 2003, 13, 2498–2504.
[67] Z. Zhang, M. G. Grigorov, Proteins Struct. Funct. Bioinf. 2006, 62, 470–
478.
[68] G. Cruciani, M. Pastor, W. Guba, Eur. J. Pharm. Sci. 2000, 11, S29–39.
[69] F. Spyrakis, B. Cellini, S. Bruno, P. Benedetti, E. Carosati, G. Cruciani, F. Mi-
cheli, A. Felici, P. Cozzini, G. E. Kellogg, C. B. Voltattorni, A. Mozzarelli,
ChemMedChem 2014, 9, 1501–1511.
[70] F. Spyrakis, R. Singh, P. Cozzini, B. Campanini, E. Salsi, P. Felici, S. Raboni,
P. Benedetti, G. Cruciani, G. E. Kellogg, P. F. Cook, A. Mozzarelli, PLoS One
2013, 8, e77558.
[71] W. Xu, J. Lim, C. Y. Goh, J. Y. Suen, Y. Jiang, M. K. Yau, K. C. Wu, L. Liu,
D. P. Fairlie, J. Chem. Inf. Model. 2015, 55, 2079–2084.
Inf. Model. 2013, 53, 753–762.
[
[
31] R. B. Altman, Nat. Genet. 2007, 39, 426.
32] M. Kanehisa, S. Goto, M. Furumichi, M. Tanabe, M. Hirakawa, Nucleic
Acids Res. 2010, 38, D355–60.
[33] M. Kuhn, M. Campillos, I. Letunic, L. J. Jensen, P. Bork, Mol. Syst. Biol.
2010, 6, 343.
[
[
34] M. Duran-Frigola, P. Aloy, Chem. Biol. 2013, 20, 594–603.
35] R. L. Chang, L. Xie, P. E. Bourne, B. O. Palsson, PLoS Comput. Biol. 2010,
6, e1000938.
[
[
36] T. Hart, S. Dider, W. Han, H. Xu, Z. Zhao, L. Xie, Sci. Rep. 2016, 6, 20441.
37] S. J. Ho Sui, R. Lo, A. R. Fernandes, M. D. Caulfield, J. A. Lerman, L. Xie,
P. E. Bourne, D. L. Baillie, F. S. Brinkman, Int. J. Antimicrob. Agents 2012,
4
0, 246–251.
[72] W. Xu, A. J. Lucke, D. P. Fairlie, J. Mol. Graphics Modell. 2015, 57, 76–88.
[73] L. A. Pinna, F. Meggio, Prog. Cell Cycle Res. 1997, 77–97.
[74] S. Sarno, S. Moro, F. Meggio, G. Zagotto, D. Dal Ben, P. Ghisellini, R. Bat-
tistutta, G. Zanotti, L. A. Pinna, Pharmacol. Ther. 2002, 93, 159–168.
[75] A. P. Bento, A. Gaulton, A. Hersey, L. J. Bellis, J. Chambers, M. Davies,
F. A. Kruger, Y. Light, L. Mak, S. McGlinchey, M. Nowotka, G. Papadatos,
R. Santos, J. P. Overington, Nucleic Acids Res. 2014, 42, D1083–D1090.
[76] C. Chichester, D. Digles, R. Siebes, A. Loizou, P. Groth, L. Harland, Drug
Discovery Today 2015, 20, 399–405.
[
38] S. L. Kinnings, L. Xie, K. H. Fung, R. M. Jackson, P. E. Bourne, PLoS
Comput. Biol. 2010, 6, e1000976.
39] V. J. Haupt, S. Daminelli, M. Schroeder, PLoS One 2013, 8, e65894.
40] F. Moriaud, S. B. Richard, S. A. Adcock, L. Chanas-Martin, J. S. Surgand,
M. B. Jelloul, F. Delfaud, Briefings Bioinf. 2011, 12, 336–340.
41] J. Bostrçm, A. Hogner, S. Schmitt, J. Med. Chem. 2006, 49, 6716–6725.
42] Y. Hu, J. Bajorath, Drug Discovery Today 2013, 18, 644–650.
43] K. Park, S. Lee, H. S. Ahn, D. Kim, Mol. Biosyst. 2009, 5, 844–853.
44] S. Schmitt, D. Kuhn, G. Klebe, J. Mol. Biol. 2002, 323, 387–406.
45] A. Shulman-Peleg, R. Nussinov, H. J. Wolfson, J. Mol. Biol. 2004, 339,
[
[
[
[
[
[
[
[77] F. A. Kruger, A. Gaulton, M. Nowotka, J. P. Overington, Bioinformatics
2015, 31, 776–778.
6
07–633.
[78] D. S. Wishart, C. Knox, A. C. Guo, S. Shrivastava, M. Hassanali, P. Stothard,
Z. Chang, J. Woolsey, Nucleic Acids Res. 2006, 34, D668–D672.
[79] L. Xie, J. Wang, P. E. Bourne, PLoS Comput. Biol. 2007, 3, e217.
[80] L. Dellafiora, M. Marchetti, F. Spyrakis, V. Orlandi, B. Campanini, G. Cru-
ciani, P. Cozzini, A. Mozzarelli, Bioorg. Med. Chem. Lett. 2015, 25, 4297–
4303.
[81] G. Muratore, L. Goracci, B. Mercorelli, A. Foeglein, P. Digard, G. Cruciani,
G. Palu, A. Loregian, Proc. Natl. Acad. Sci. USA 2012, 109, 6247–6252.
[82] F. Spyrakis, P. Benedetti, S. Decherchi, W. Rocchia, A. Cavalli, S. Alcaro, F.
Ortuso, M. Baroni, G. Cruciani, J. Chem. Inf. Model. 2015, 55, 2256–2274.
[83] L. Goracci, S. Buratta, L. Urbanelli, G. Ferrara, R. Di Guida, C. Emiliani, S.
Cross, Eur. J. Med. Chem. 2015, 92, 49–63.
[
[
[
46] N. Weill, D. Rognan, J. Chem. Inf. Model. 2010, 50, 123–135.
47] K. Yeturu, N. Chandra, BMC Bioinf. 2008, 9, 543.
48] SMAP: Software for ligand binding site comparison, The City University of
New York, http://compsci.hunter.cuny.edu/~leixie/smap/smap.html.
49] T. Liu, R. B. Altman, PLoS Comput. Biol. 2011, 7, e1002326.
50] J. Konc, D. Janezic, Bioinformatics 2010, 26, 1160–1168.
51] L. Siragusa, S. Cross, M. Baroni, L. Goracci, G. Cruciani, Proteins Struct.
Funct. Bioinf. 2015, 83, 517–532.
[
[
[
[
52] L. Siragusa, F. Spyrakis, L. Goracci, S. Cross, G. Cruciani, Mol. Inf. 2014,
33, 446–453.
[
[
[
53] P. J. Goodford, J. Med. Chem. 1985, 28, 849–857.
54] E. Carosati, S. Sciabola, G. Cruciani, J. Med. Chem. 2004, 47, 5114–5125.
55] V. Ferrario, L. Siragusa, C. Ebert, M. Baroni, M. Foscato, G. Cruciani, L.
Gardossi, PLoS One 2014, 9, e109354.
[84] L. Goracci, M. Ceccarelli, D. Bonelli, G. Cruciani, J. Chem. Inf. Model.
2013, 53, 1436–1446.
[85] R. Navakauskienꢃ, M. Mori, M. Christodoulou, A. Zentelytꢃ, B. Botta, L.
Dalla Via, F. Ricci, G. Damia, D. Passarella, C. Zilio, N. Martinet, MedChem-
Comm 2016, 7, 1245–1255.
[56] M. Baroni, G. Cruciani, S. Sciabola, F. Perruccio, J. S. Mason, J. Chem. Inf.
Model. 2007, 47, 279–294.
[
[
57] C. W. Carreras, D. V. Santi, Annu. Rev. Biochem. 1995, 64, 721–762.
58] E. Chu, M. A. Callender, M. P. Farrell, J. C. Schmitz, Cancer Chemother.
Pharmacol. 2003, 52, 80–89.
[86] C. G. Fortuna, C. Bonaccorso, A. Bulbarelli, G. Caltabiano, L. Rizzi, L. Gor-
acci, G. Musumarra, A. Pace, A. Palumbo Piccionello, A. Guarcello, P.
Pierro, C. E. Cocuzza, R. Musumeci, Eur. J. Med. Chem. 2013, 65, 533–
545.
[87] G. Muratore, B. Mercorelli, L. Goracci, G. Cruciani, P. Digard, G. Palu, A.
Loregian, Antimicrob. Agents Chemother. 2012, 56, 6009–6013.
[88] D. Cardinale, G. Guaitoli, D. Tondi, R. Luciani, S. Henrich, O. M. Salo-
Ahen, S. Ferrari, G. Marverti, D. Guerrieri, A. Ligabue, C. Frassineti, C.
Pozzi, S. Mangani, D. Fessas, R. Guerrini, G. Ponterini, R. C. Wade, M. P.
Costi, Proc. Natl. Acad. Sci. USA 2011, 108, E542–9.
[89] E. Carosati, G. Sforna, M. Pippi, G. Marverti, A. Ligabue, D. Guerrieri, S.
Piras, G. Guaitoli, R. Luciani, M. P. Costi, G. Cruciani, Bioorg. Med. Chem.
2010, 18, 7773–7785.
[90] R. Z. Cer, U. Mudunuri, R. Stephens, F. J. Lebeda, Nucleic Acids Res. 2009,
37, W441–W445.
[
[
[
59] K. Azijli, I. A. van Roosmalen, J. Smit, S. Pillai, M. Fukushima, S. de Jong,
G. J. Peters, I. V. Bijnsdorp, F. A. Kruyt, Cancer Chemother. Pharmacol.
2014, 73, 1273–1283.
60] E. Carosati, A. Tochowicz, G. Marverti, G. Guaitoli, P. Benedetti, S. Ferrari,
R. M. Stroud, J. Finer-Moore, R. Luciani, D. Farina, G. Cruciani, M. P. Costi,
J. Med. Chem. 2012, 55, 10272–10276.
61] M. P. Costi, A. Gelain, D. Barlocco, S. Ghelli, F. Soragni, F. Reniero, T.
Rossi, A. Ruberto, C. Guillou, A. Cavazzuti, C. Casolari, S. Ferrari, J. Med.
Chem. 2006, 49, 5958–5968.
[
[
[
[
62] D. Tondi, A. Venturelli, S. Ferrari, S. Ghelli, M. P. Costi, J. Med. Chem.
2005, 48, 913–916.
63] J. Phan, S. Koli, W. Minor, R. B. Dunlap, S. H. Berger, L. Lebioda, Biochem-
istry 2001, 40, 1897–1902.
64] G. Ausiello, P. F. Gherardini, E. Gatti, O. Incani, M. Helmer-Citterich, BMC
Bioinf. 2009, 10, 182.
65] S. Pꢃrot, O. Sperandio, M. A. Miteva, A. C. Camproux, B. O. Villoutreix,
Drug Discovery Today 2010, 15, 656–667.
Received: February 29, 2016
Revised: June 8, 2016
Published online on && &&, 0000
&
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
14
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

FULL PAPERS
The TS test case: We describe the inte-
grated and innovative BioGPS/FLAPdock
pipeline for rapid and effective compari-
son of protein cavities, off-target iden-
tification, and drug repurposing. Struc-
tural, chemical, and energetic properties
of the cavities are simply encoded in
the corresponding GRID molecular inter-
action fields. BioGPS discloses pocket
similarity, and cross-docking experi-
ments identify drugs for potential repur-
posing.
L. Siragusa, R. Luciani, C. Borsari,
S. Ferrari, M. P. Costi, G. Cruciani,
F. Spyrakis*
&& – &&
Comparing Drug Images and
Repurposing Drugs with BioGPS and
FLAPdock: The Thymidylate Synthase
Case
ChemMedChem 2016, 11, 1 – 15
www.chemmedchem.org
15
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ