discovery of novel non acidic mpges 1 inhibitors by virtual screening with a multistep protocol

Nohaa, Katrin Fischerb, Andreas Koeberleb, Ulrike Garschab, Oliver Werzb, Daniela Schustera,⇑ a Computer Aided Molecular Design CAMD Group, Institute of Pharmacy/Pharmaceutical Chemistry

Discovery of novel, non-acidic mPGES-1 inhibitors

by virtual screening with a multistep protocol

Stefan M Nohaa, Katrin Fischerb, Andreas Koeberleb, Ulrike Garschab, Oliver Werzb, Daniela Schustera,⇑

a

Computer Aided Molecular Design (CAMD) Group, Institute of Pharmacy/Pharmaceutical Chemistry, University of Innsbruck, Innrain 80/82, A-6020 Innsbruck, Austria

b

Chair of Pharmaceutical/Medicinal Chemistry, Institute of Pharmacy, University of Jena, Philosophenweg 14, D-07743 Jena, Germany

a r t i c l e i n f o

Article history:

Received 18 February 2015

Revised 13 May 2015

Accepted 19 May 2015

Available online xxxx

Keywords:

Inflammation

mPGES-1

Virtual screening

3D pharmacophore

Kruskal–Wallis test

a b s t r a c t

Microsomal prostaglandin E2synthase-1 (mPGES-1) inhibitors are considered as potential therapeutic agents for the treatment of inflammatory pain and certain types of cancer So far, several series of acidic

as well as non-acidic inhibitors of mPGES-1 have been discovered Acidic inhibitors, however, may have issues, such as loss of potency in human whole blood and in vivo, stressing the importance of the design and identification of novel, non-acidic chemical scaffolds of mPGES-1 inhibitors Using a multistep virtual screening protocol, the Vitas-M compound library (1.3 million entries) was filtered and 16 predicted compounds were experimentally evaluated in a biological assay in vitro This approach yielded two molecules active in the low micromolar range (IC50values: 4.5 and 3.8lM, respectively)

Ó 2015 The Authors Published by Elsevier Ltd This is an open access article under the CC BY license (http://

creativecommons.org/licenses/by/4.0/)

1 Introduction

In the arachidonic acid cascade, the activity of the cytosolic

phospholipase A2 is required for the release of arachidonic acid

(AA), a critical precursor molecule for pro-inflammatory mediators

Different enzymatic pathways convert AA into distinct eicosanoids

These mediators regulate various physiological processes and also

trigger multiple effects in various human diseases.1,2Among these

mediators, prostaglandin E2(PGE2) is well recognized as critical

bioactive molecule High PGE2levels, typically occurring in

inflam-mation, are relevant for swelling, fever, and inflammatory pain,

and thus, pharmacological inhibition of PGE2biosynthesis is

con-sidered a promising opportunity for the treatment of inflammatory

pain, for example, in rheumatic diseases.3Additionally, PGE2

syn-thesis is important in tumor growth and cancer progression.4–6

PGE2is produced from the cyclooxygenase (COX)-derived

prosta-glandin H2 (PGH2) by PGE2 synthases (PGES) (EC 5.3.99.3).7

Among the three PGES isoenzymes, the microsomal PGES-1

(mPGES-1) displays a unique role because its expression is induced

in the inflammatory response, similar to COX-2.8

By inhibiting mPGES-1 as the terminal synthase in PGE2

biosynthesis, mPGES-1 inhibitors are considered very promising

regarding their side effect profile.9,10 The application of other

anti-inflammatory agents, such as unspecific COX inhibitors,

traditional nonsteroidal anti-inflammatory drugs (NSAIDs), or COX-2-selective inhibitors (coxibs), is associated with side effects concerning, among others, the renal function and effects on the gastrointestinal tract.11 In contrast, during prolonged inhibition

of mPGES-1 in dogs, pronounced effects on the renal function were not observed.12So far, there is no mPGES-1 inhibitor available for clinical use, although data from pre-clinical studies stressed the relevance of mPGES-1 inhibitors as potentially therapeutic agents Therefore, the development of mPGES-1 inhibitors is highly relevant.13

A series of mPGES-1 inhibitors is reported in the literature, of which several comprise an acidic functionality, such as an oxicam template,14a sulfonamide group15or a carboxylic acid moiety.16,17

Unfortunately, acidic molecules suppressing mPGES-1 activity may have inferior potency in human whole blood seemingly due to unspecific plasma protein binding.14,16 This suggests that the design and identification of novel, non-acidic chemical scaffolds

is warranted As an overview, non-acidic chemical scaffolds of mPGES-1 inhibitors, which were reported so far, are shown in

Previously, we reported the discovery of acidic mPGES-1 inhibi-tors using a pharmacophore-based virtual screening approach Using this screening protocol, acidic inhibitors from synthetic libraries were discovered The most potent inhibitors exhibited

IC50values in the sub-micromolar range.17Additionally, mPGES-1 inhibitors with comparable potency from Lichen species were dis-covered using the previously reported pharmacophore model.28 http://dx.doi.org/10.1016/j.bmc.2015.05.045

0968-0896/Ó 2015 The Authors Published by Elsevier Ltd.

This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ).

⇑ Corresponding author Tel.: +43 (0)512 507 58253; fax: +43 (0)512 507 5269.

E-mail address: Daniela.Schuster@uibk.ac.at (D Schuster).

Contents lists available atScienceDirect

Bioorganic & Medicinal Chemistry

j o u r n a l h o m e p a g e : w w w e l s e v i e r c o m / l o c a t e / b m c

Comparably, other groups have reported virtual screening

approaches to find novel mPGES-1 inhibitors For instance,

Rörsch et al applied a multistep ligand-based virtual screening

protocol to discover novel and non-acidic mPGES-1 inhibitors.29

In addition, several active compounds were discovered by applying

docking-based screening strategies, of which some even elicited

high potency.30–33 Docking-based virtual screening campaigns

towards mPGES-1 have been facilitated as a 3D electron

crystallog-raphy structure was reported in 2008.34In 2013, a high-resolution

X-ray crystal structure of mPGES-1 has been resolved.35 Very

recently an X-ray crystal structure of mPGES-1 with a

co-crystal-lized ligand has been reported.36

In this study, a novel concept for the validation of the 3D

phar-macophore model was applied using the Kruskal–Wallis test.37

This test was suggested as a robust investigation of the

discrimina-tory power of distinct virtual screening methods, and was

previ-ously used for the comparative assessment of docking and

scoring functions.38,39 The analysis with the Kruskal–Wallis test

is characterized as less artifact-prone and also enables a post hoc

test, rendering this analysis an attractive method in the validation

also for pharmacophore-based virtual screening.38,39

2 Materials and methods

2.1 Study design

In brief, we consecutively performed forward filtering, using

2D similarity screening, and pharmacophore-based virtual

screening The most interesting molecules which were retained

thereof, accounting in addition pharmacophore fit evaluation

and diversity clustering, were submitted to molecular docking

Finally, this protocol was applied to prospective virtual screening

of the Vitas-M library (http://www.vitasmlab.com/) The hit-list

was visually inspected to select compounds for a biological

eval-uation to discover novel and non-acidic mPGES-1 inhibitors

2.2 Software specifications The computational studies were performed on a workstation running Microsoft Windows 7, which was employed for the work with the molecular modeling package Discovery Studio version 3.540and PipelinePilot 8.0.1.41In parallel, the calculations for the work with Maestro suite 9.2.11242were performed on a worksta-tion running OpenSuse 12.1 The statistical evaluaworksta-tion was per-formed within Microsoft Excel 2010 and its add-in Analyse-it Method Evaluation version 2.26.43

2.3 Validation 2.3.1 Concept

We assessed the discriminatory power of the 3D pharma-cophore model by following the workflow reported by Seifert

et al.38,39In this work, the discriminatory power of docking and scoring functions was assessed by ANOVA (analysis of variance)

or a nonparametric version of it, that is, the Kruskal–Wallis test.37

Because this concept can also be useful for the development of 3D pharmacophore models, this analysis was included in the model validation and conducted as an extension to the validation with benchmarking experiments So a validation set, set_1, was assem-bled and used for screening experiments with the hypotheses The

Figure 1 Chemical series of non-acidic mPGES-1 inhibitors are depicted with 2D structures.

Figure 2 Overview of the virtual screening protocol.

statistical evaluation of the results was accomplished with the

Kruskal–Wallis test and a post hoc test Furthermore,

benchmark-ing experiments were conducted by screenbenchmark-ing a second validation

set, set_2, and calculating well-established performance metrics

2.3.2 Validation sets and calculations

Set_1 comprised highly active (IC5060.5lM), medium active

(IC50: 0.5–5lM), and confirmed inactive molecules (IC50>5lM)

from several congeneric series of non-acidic mPGES-1 inhibitors,

with 14 molecules in each group It consisted, in total, of 42

mole-cules For more details on set_1, seeSupporting information In the

validation, we screened set_1, followed by the statistical

evalua-tion of the results obtained thereof with the Kruskal–Wallis test

Furthermore, we included in this analysis Bonferroni’s post hoc

test, employing the confirmed inactive molecules in the post hoc

test as control group, and accounting the results of this evaluation

significant with p <0.1

Additionally, we assembled set_2 by combining highly active

and medium active inhibitors of set_1, resulting in 28 active

mole-cules, and a virtual library of 12,775 putatively inactive molecules

(‘decoys’).44,45Afterwards, the virtual screening results from set_2

were used to calculate the percent yield (%Y) and the goodness of

hit-list (GH)-score, which follow Eqs.1 and2 Furthermore, the

enrichment factor (EF, Eq.3) was calculated to compare the

enrich-ment in the x% of the ranked hit-list to the ratio of actives to decoys

in the entire validation set We calculated the EF1%and the EF0.5%

which may attain a maximal value of 100 or 200, respectively.46

%Y ¼TP

GH ¼ TP  ð3A þ nÞ

4nA

 1 n  TP

N  A

ð2Þ

EFx%¼TPx%=nx%

TP true positives, active molecules retained in the hit-list

n number of hits found by the method

A actives, all active molecules

N all molecules, active molecules and the decoy set

2.4 Forward filtering

First, to evaluate the enrichment obtained by employing 2D

similarity screening, set_2 was utilized for virtual screening with

2D fingerprints Later, in prospective virtual library screening 2D

fingerprints were applied with adjusted and optimized settings

and further filters: (i) a filter to focus on molecules with aqueous

solubility level P2, and (ii) Veber rules47 and Lipinski’s

Rule-of-5.48These filters were applied by performing respective protocols

(‘ADMET Descriptors’ and ‘Filter by Lipinski and Veber Rules’) with

default settings within PipelinePilot, while 2D similarity screening

was performed within Discovery Studio with the protocol ‘Find

Similar Molecules by Fingerprints’ The 2D similarity screening

was performed with SciTegic fingerprints, representing a type of

combinatorial/circular fingerprints.49,50 In the virtual screening

campaign, the Vitas-M library was filtered which was downloaded

in version September 2013 (http://www.vitasmlab.com/,

1,305,485 entries)

2.5 Conformational analysis

Prior to the hypotheses generation process, the conformational

model of the training set compounds was generated using

Discovery Studio with the more exhaustive ‘BEST’ quality51and a maximum number of 255 conformations per molecule

All compound libraries used for validating the pharmacophore models and in the prospective virtual library screening were con-verted into 3D multi-conformational databases using ‘CAESER’ quality52 with a maximum number of 100 conformations per molecule

2.6 Pharmacophore modeling and virtual screening The 3D pharmacophore models were generated employing the HipHop algorithm within Discovery Studio, which is available as protocol ‘Common Feature Pharmacophore Generation’ This algo-rithm elucidates the pharmacophore hypotheses in a so-called

‘pruned exhaustive search’ All two feature hypotheses, which are feasible in 3D space and are defined by the molecules in the train-ing set as well as respective conformers as input, are generated This is followed by the generation of more complex models utiliz-ing the hypotheses retrieved in the previous step The procedure is stopped when the generation process is exhausted, and then the best-ranked hypotheses are reported.53,54Before hypotheses are created in a running, so-called principal values have to be assigned,

to imply the influence a training set molecule has on the hypothe-sis generation process Furthermore, the value for ‘Maximum Omitted Features’ (MOF) of a training set compound may be adjusted; when the value is set to 1, this implies that a ‘partial fit’ of this training set molecule is allowed Additionally, the chem-ical feature types used for model building were set to include hydrogen-bond donor (HBD), hydrogen-bond acceptor (HBA), hydrophobic interaction (HYD), aromatic ring (RA), hydrophobic aromatic, and hydrophobic aliphatic (Hal)

Furthermore, pharmacophore-based virtual screening was per-formed by applying the ‘best flexible search’ algorithm In this algorithm, the database entries are pre-filtered according to the absence or presence of all required chemical features needed for pharmacophore fitting in a rapid procedure This is followed by a more rigorous procedure attempting to match the atoms of the 3D database molecule conformations to the features of the 3D search query During the latter, small modifications of molecule’s conformation are allowed to enable and optimize the fitting of the molecule into the model As limit for the optimized fitting into the model, the molecule’s strain energy is accounted during this procedure.55

2.7 Diversity clustering

To enhance the chemical diversity among the fitting molecules, diversity clustering was performed within Discovery Studio by applying the protocol ‘Cluster Ligands’ and using default settings

to return 500 clusters We collected the ‘cluster centers’ and sub-mitted these compounds to molecular docking

2.8 Molecular docking Docking in the context of prospective virtual library screening was conducted employing Glide within the graphical interface, the Maestro suite, and as described previously.56 Following the preparation of the protein and the organic molecules, molecular docking was performed and hit-list ranking was achieved employ-ing Glide in ‘standard precision’ (SP) mode

For the binding pose predictions, the recently reported X-ray crystal structure of mPGES-1 with the co-crystallized ligand (PDB code: LVJ; PDB accession code: 4bpm)36was utilized for the dock-ing study The macromolecule 3D coordinates were downloaded from the PDBe57 providing a web portal (http://www.ebi.ac.uk/

interactions as 3D pharmacophore as well as for visualization

purposes

2.9 Induction of mPGES-1 in A549 cells, isolation of

microsomes, and determination of mPGES-1 activity

Human A549 cells were treated and prepared as described.59

Briefly, cells (2  106/20 ml DMEM/High glucose (4.5 g/L) medium

containing FCS (2%, v/v)) were incubated for 16 h at 37 °C, 5% CO2

The culture medium was replaced by fresh medium, interleukin-1b

(1 ng/ml) was added, and cells were incubated for another 72 h

Then, cells were detached with trypsin/EDTA, washed with PBS

and frozen in liquid nitrogen Ice-cold homogenization buffer

(0.1 M potassium phosphate buffer pH 7.4, 1 mM

phenylmethyl-sulfonyl fluoride, 60lg/ml soybean trypsin inhibitor, 1lg/ml

leu-peptin, 2.5 mM glutathione, and 250 mM sucrose) was added After

15 min on ice, cells were resuspended and sonicated (3  20 s) The

homogenate was subjected to differential centrifugation (10,000g

for 10 min and at 174,000g for 1 h at 4 °C) The pellet

(microso-mal fraction) was resuspended in 1 ml homogenization buffer,

and protein concentration was determined by the Coomassie

pro-tein assay The microsomal membranes were then diluted in

potas-sium phosphate buffer (0.1 M, pH 7.4) containing 2.5 mM

glutathione (100ll total volume) and test compounds or vehicle

(DMSO) were added After 15 min, enzymatic PGE2 formation

was initiated by addition of 20lM PGH2 (final concentration)

After 1 min at 4 °C, the reaction was terminated with 100ll of stop

solution (40 mM FeCl2, 80 mM citric acid and 10lM of 11b-PGE2),

PGE2was separated by solid phase extraction on reversed phase

(RP)-C18 material using acetonitrile (200ll) as eluent, and

ana-lyzed by RP-HPLC (30% acetonitrile aqueous + 0.007% TFA (v/v),

Nova-PakÒC18 column, 5  100 mm, 4lm particle size, flow rate

1 ml/min) with UV detection at 195 nm 11b-PGE2 was used as

internal standard to quantify PGE2formation

3 Results and discussion

3.1 Validation of the 2D search query

In brief, various runs showed that the best results were

achieved with ECFP4 fingerprints with a Tanimoto coefficient

P0.25 and a multiple-molecule query consisting of compounds

1, 3, and 4, which represented the most potent and chemically

diverse molecules also later employed in the creation of the

phar-macophore model (Fig 3).20,25,26 In this combination, a high

enrichment was achieved in the screening experiments using set_2 (%Y = 48.28, GH = 0.49, EF1%= 67.87, EF0.5%= 121.46) 3.2 Pharmacophore model validation

In order to select the best-performing model for the virtual screening campaign, the hypotheses were validated thoroughly The synthetic compounds 1–4 and the natural product arzanol (5) from the medicinal plant Helichrysum italicum were selected

as training set (Fig 3) for pharmacophore modeling.20,25,26,60We assigned compound 1 a principal value of 2 (highly active), while the remaining compounds in the training set 2–5 were assigned

a principal value of 1 (active,Table 1) Furthermore, a different binding mode was assumed in case of 5, and so the value for MOF was adjusted to 1, while this value was left unmodified among the other compounds in the training set This adjustment implies that a ‘partial fit’ of all compounds, except compound 5, was not allowed during the hypothesis generation process Among the 10 computed hypotheses, Hypo01 and Hypo06 showed the most promising results They returned 35.7% of the highly active inhibitors, respectively, and 57.1% (Hypo01) or 50.0% (Hypo06) of the medium active inhibitors, while all con-firmed inactive molecules were discarded Accounting in the eval-uation the Kruskal–Wallis’ statistic, good model quality was achieved in case of Hypo01 and Hypo06 (Table 2) The differences between these two most promising 3D pharmacophore models were quite subtle Hypo01 consisted of one RA, one HBA, one HBD, and two Hal, while Hypo06 comprised basically the same fea-tures, apart from one HYD in the position of the RA The statistical evaluation with the post hoc test showed, that in case of Hypo01, confirmed inactive molecules were separated from medium active inhibitors (p = 0.0029) and highly active inhibitors (p = 0.0523) by their geometrical fit score, displaying the best results among all hypotheses The results from the screening experiments of set_1 with Hypo01 are available in Supporting information In the

Figure 3 Training set compounds for pharmacophore modeling.

Table 1 Training set for HipHop model Compound IC 50 (lM) Class Reference

1 0.0091 Highly active 26

benchmarking experiments, Hypo01 together with Hypo07 to

Hypo10 showed the most promising results when the GH-score

was accounted (GH P0.15) The EF-values prioritized Hypo01

together with Hypo06 to Hypo10 Especially, when EF0.5% was

accounted, Hypo01 outperformed the other hypotheses, while

Hypo01 also attained the best results in terms of%Y Together,

the results pointed towards superior model quality in case of

Hypo01 (Table 2) Hypo01, the best-performing pharmacophore

model is shown inFigure 4

3.3 Virtual screening using an external library

Finally, to experimentally validate the virtual screening

proto-col, the Vitas-M library (http://www.vitasmlab.com/) was filtered

Out of the initial compound library, which comprised 1,305,485

molecules, 1,020,417 molecules (78.2%) and 974,991 molecules

(74.7%) were retrieved in the forward filtering by applying the

fil-ter on the aqueous solubility and the Lipinski and Veber rules,

respectively Then, out of the compound library which was

returned from these previous filters, a focused library of 18,057

molecules (1.4%) was retrieved by the application of 2D similarity

screening Afterwards, pharmacophore-based virtual screening

was performed returning 8079 fitting molecules (0.6% of the initial

Vitas-M library) We only considered molecules for further

pro-cessing, which attained high fit-values (pharmacophore fit P2.5)

This reduced the number of remaining molecules to 1857 (0.14%

of the initial library) To enhance the structural diversity of the hits

to be biologically tested, these molecules were clustered The 500

molecules (0.04% of the Vitas-M collection), retrieved as cluster centers were submitted to molecular docking After re-scoring by molecular docking, the hit-list was visually inspected Finally, 20 molecules were selected of which 17 were available and acquired for biological evaluation in the cell-free mPGES-1 activity assay Among the tested molecules (10lM, final concentration), compounds 6 and 7 were revealed as efficient inhibitors of mPGES-1-mediated PGE2 synthesis (Table 3) Concentration– response analysis for compounds 6 and 7 revealed IC50values of 4.5 and 3.8lM (Fig 5), potencies that are close to those of the ref-erence inhibitor MK-886 (IC50= 2.4lM) in a comparable test sys-tem.61 In addition, compounds 8 and 9 significantly suppressed mPGES-1 activity at 10lM, but less than 50% (Table 3) Thus,

IC50values were not determined The other 13 molecules out of the 17 acquired ones (which are shown inSupporting information) failed to inhibit mPGES-1 at 10lM or were not determined

In order to confirm that the most potent compounds 6 and 7 are specific for mPGES-1, we tested on one hand if reduced mPGES-1

Table 2

Results from the theoretical validation of the 3D pharmacophore models

%Y GH EF 1% EF 0.5% Kruskal–Wallis’ statistic p

Hypo01 4.45 0.15 32.14 50.00 9.78 0.0075

Hypo02 1.54 0.12 17.86 35.71 5.04 0.0806

Hypo03 1.55 0.12 17.86 35.71 5.65 0.0593

Hypo04 0.99 0.10 14.29 21.43 5.30 0.0705

Hypo05 0.99 0.10 14.29 14.29 5.30 0.0705

Hypo06 2.36 0.12 25.00 35.71 8.36 0.0153

Hypo07 2.58 0.15 28.57 35.71 5.37 0.0682

Hypo08 1.37 0.16 21.43 42.86 4.89 0.0869

Hypo09 1.58 0.15 39.29 35.71 3.74 0.1538

Hypo10 1.61 0.15 32.14 28.57 2.99 0.2238

Figure 4 Depiction of Hypo01; chemical features are color-coded: light blue-Hal;

orange-RA; magenta-HBD; green-HBA.

Table 3 Summary of results from the experimental evaluation for the four novel molecules, suppressing mPGES-1 activity, among the 16 tested molecules

Compound Chemical structure Remaining

activity at

10lM (%)

IC 50

(lM)

0 25 50 75

compd 7

[µM]

Figure 5 Inhibition of mPGES-1 by compounds 6 and 7 Data are given as mean ± SEM; n = 3.

activity is simply due to irreversible or nuisance inhibition, and on

the other hand, we analyzed if they also affect related enzymes

within the AA cascade Wash-out experiments revealed reversible

mPGES-1 inhibition (Fig S1), and triton-X100 failed to abolish the

mPGES-1 inhibitory activity of 6 and 7 (Fig S2), thus, excluding

unspecific interference Moreover, 6 and 7 were poor inhibitors

of isolated COX-2 (IC50>10lM, Fig S3) However, 7 effectively

inhibited 5-lipoxygenase in a cell-free assay with an IC50value of

about 5lM, whereas 6 was less active (IC50>10lM, Fig S4)

Note that dual inhibition of mPGES-1 and 5-lipoxygenase is

com-mon to many structural classes of inhibitors of natural synthetic

origin.62 Together, we conclude that 6 and 7 rather specifically

interact with mPGES-1 and inhibit its activity

To obtain a more profound insight on potential binding modes,

the virtual screening hits were submitted to molecular docking

Accounting the docking poses of compounds 6 and 7 (Fig 6A),

these novel bioactive molecules were predicted to be involved in

hydrogen-bonding to Ser127, which is assumed to serve as a key

residue in the catalytic activity.35 Furthermore, a hydrogen-bond

was predicted to be formed between 6 and Gln36 Interestingly,

both inhibitors were predicted to adopt a conformation, where

these ligands complement the surface of the binding site adjacent

to glutathione In case of 6, a substituted benzene ring was

pre-dicted to be orientated close to Phe44 and Leu39, which formed

a hydrophobic contact to this ligand moiety In case of 7, the

ben-zene ring, which was predicted to bind next to Phe44, has a

chlo-rine substituent attached, which could be involved in a

hydrophobic interaction with Leu39 The substituted benzene

moi-eties of 6 and 7, which were oriented towards the opposite site of

the mPGES-1 binding site, were predicted to be embedded in a

hydrophobic site formed by mainly aromatic or hydrophobic

amino acids (e.g., Tyr28, Tyr130, Thr131, Ala31, and Ile32) In

com-parison, LVJ (Fig 6B), which has inhibitory potency on mPGES-1

activity in the low nanomolar range, adopts a position and

orienta-tion in the binding pocket which is slightly shifted towards mainly

hydrophobic amino acids (e.g., Ala123, Val128, and Leu132), while

several hydrogen-bonds are formed to the key residue Ser127 and

other residues (e.g., Gln36 and His53)

4 Discussion

We herein report the discovery of four novel molecules sup-pressing mPGES-1 activity, of which the two most active ones showed the desired activity in the low micromolar range When only regarding the most active compounds, a hit rate of 12.5% (two virtual screening hits out of 16 tested molecules) was achieved Interestingly, this is comparable to other studies, in which prospective virtual library screening for the discovery of novel and non-acidic mPGES-1 inhibitors was conducted For instance, He et al attained good results by employing a molecular dynamics simulation to obtain an altered conformation of the 3D-structure of mPGES-1, which was modified towards an active state conformation and utilized in a docking-based screening strategy Following the in silico approach, 21 molecules of 142 tested mole-cules showed the desired activity in the experimental evaluation (hit rate: 14.8%).32 Furthermore, Rörsch et al applied basically ligand-based methods in the search for mPGES-1 inhibitors Following the experimental evaluation of 17 molecules, three novel bioactive molecules were discovered and for one of those

an IC50value was determined, showing that this compound exerts potency in the sub-micromolar range.29Fortunately, very recently the 3D-structure of mPGES-1 with a co-crystallized inhibitor became available.36 We therefore utilized this 3D-structure in a docking study of the two most active molecules, yielded in this study, compounds 6 and 7, in order to surmise binding modes of these novel mPGES-1 inhibitors We thereby predicted that these molecules are accommodated nicely in the site adjacent to the cofactor glutathione, and could exhibit molecular interactions to the key residue Ser127 Together, the compounds 6 and 7 showed inhibitory potency on mPGES-1 activity in the low micromolar range, making them interesting starting points for optimization efforts

Basically, virtual screening techniques are usually validated by screening (a) validation set(s) in benchmarking experiments In cases like this study, where very similar models perform with quite comparable results, an additional validation with the Kruskal– Wallis test can be helpful in the selection of the screening model,

Figure 6 (A) Predicted binding modes, shown for the two most active molecules yielded in the virtual screening campaign, compounds 6 (gray) and 7 (magenta) (B) In comparison, the binding mode of the highly potent inhibitor LVJ is depicted, following the minimization and 3D pharmacophore creation within LigandScout Glutathione (GSH); chemical-features are color-coded: red arrow—HBA; green arrow—HBD, yellow sphere—HYD; the poses are shown with receptor-binding surface (color-coded by aggregated hydrophilicity/hydrophobicity: blue/gray, respectively).

especially as this test is considered to serve as robust investigation

of the model quality.38,39

5 Conclusion

In summary, a multistep virtual screening protocol is presented,

which included a novel concept in the validation of the 3D

pharma-cophore Following a virtual screening campaign the results of the

experimental evaluation confirmed the protocol quality, while the

two most active molecules which inhibited mPGES-1 in a cell-free

mPGES-1 activity assay, compounds 6 and 7, may serve as

promis-ing startpromis-ing points for further optimization The results may be

considered as a case study; however, the modified concept applied

in the pharmacophore model validation may be useful for further

studies on other targets

Acknowledgments

We are grateful for the financial support by the Tyrolean

Science Foundation (TWF), granted in 2011 (ID 134311) and the

Austrian Science Fund (FWF, S10711)

Supplementary data

Supplementary data associated with this article can be found, in

the online version, athttp://dx.doi.org/10.1016/j.bmc.2015.05.045

References and notes

1 Idborg, H.; Olsson, P.; Leclerc, P.; Raouf, J.; Jakobsson, P J.; Korotkova, M.

Prostaglandins Other Lipid Mediat 2013, 107, 18

2 Leclerc, P.; Idborg, H.; Spahiu, L.; Larsson, C.; Nekhotiaeva, N.; Wannberg, J.;

Stenberg, P.; Korotkova, M.; Jakobsson, P J Prostaglandins Other Lipid Mediat.

2013, 107, 26

3 Korotkova, M.; Jakobsson, P J Nat Rev Rheumatol 2014, 10, 229

4 Radmark, O.; Samuelsson, B J Intern Med 2010, 268, 5

5 Kamei, D.; Murakami, M.; Nakatani, Y.; Ishikawa, Y.; Ishii, T.; Kudo, I J Biol.

Chem 2003, 278, 19396

6 Nakanishi, M.; Gokhale, V.; Meuillet, E J.; Rosenberg, D W Biochimie 2010, 92,

660

7 Haeggstrom, J Z.; Rinaldo-Matthis, A.; Wheelock, C E.; Wetterholm, A.

Biochem Biophys Res Commun 2010, 396, 135

8 Murakami, M.; Naraba, H.; Tanioka, T.; Semmyo, N.; Nakatani, Y.; Kojima, F.;

Ikeda, T.; Fueki, M.; Ueno, A.; Oh, S.; Kudo, I J Biol Chem 2000, 275, 32783

9 Xu, D.; Rowland, S E.; Clark, P.; Giroux, A.; Cote, B.; Guiral, S.; Salem, M.;

Ducharme, Y.; Friesen, R W.; Methot, N.; Mancini, J.; Audoly, L.; Riendeau, D J.

Pharmacol Exp Ther 2008, 326, 754

10 Hara, S.; Kamei, D.; Sasaki, Y.; Tanemoto, A.; Nakatani, Y.; Murakami, M.

Biochimie 2010, 92, 651

11 Castellsague, J.; Riera-Guardia, N.; Calingaert, B.; Varas-Lorenzo, C.;

Fourrier-Reglat, A.; Nicotra, F.; Sturkenboom, M.; Perez-Gutthann, S.; Safety of Non

Steroidal Anti-Inflammatory Drugs, P Drug Saf 2012, 35, 1127

12 Salazar, F.; Vazquez, M L.; Masferrer, J L.; Mbalaviele, G.; Llinas, M T.; Saez, F.;

Arhancet, G.; Salazar, F J Am J Physiol Renal Physiol 2014, 306, F68

13 Friesen, R W.; Mancini, J A J Med Chem 2008, 51, 4059

14 Wang, J.; Limburg, D.; Carter, J.; Mbalaviele, G.; Gierse, J.; Vazquez, M Bioorg.

Med Chem Lett 2010, 20, 1604

15 Bylund, J.; Annas, A.; Hellgren, D.; Bjurstrom, S.; Andersson, H.; Svanhagen, A.

Drug Metab Dispos 2013, 41, 634

16 Riendeau, D.; Aspiotis, R.; Ethier, D.; Gareau, Y.; Grimm, E L.; Guay, J.; Guiral,

S.; Juteau, H.; Mancini, J A.; Methot, N.; Rubin, J.; Friesen, R W Bioorg Med.

Chem Lett 2005, 15, 3352

17 Waltenberger, B.; Wiechmann, K.; Bauer, J.; Markt, P.; Noha, S M.; Wolber, G.;

Rollinger, J M.; Werz, O.; Schuster, D.; Stuppner, H J Med Chem 2011, 54,

3163

18 Cote, B.; Boulet, L.; Brideau, C.; Claveau, D.; Ethier, D.; Frenette, R.; Gagnon, M.;

Giroux, A.; Guay, J.; Guiral, S.; Mancini, J.; Martins, E.; Masse, F.; Methot, N.;

Riendeau, D.; Rubin, J.; Xu, D.; Yu, H.; Ducharme, Y.; Friesen, R W Bioorg Med.

Chem Lett 2007, 17, 6816

19 Giroux, A.; Boulet, L.; Brideau, C.; Chau, A.; Claveau, D.; Cote, B.; Ethier, D.;

Frenette, R.; Gagnon, M.; Guay, J.; Guiral, S.; Mancini, J.; Martins, E.; Masse, F.;

Methot, N.; Riendeau, D.; Rubin, J.; Xu, D.; Yu, H.; Ducharme, Y.; Friesen, R W.

Bioorg Med Chem Lett 2009, 19, 5837

20 Koeberle, A.; Haberl, E M.; Rossi, A.; Pergola, C.; Dehm, F.; Northoff, H.;

Troschuetz, R.; Sautebin, L.; Werz, O Bioorg Med Chem 2009, 17, 7924

21 Bruno, A.; Di Francesco, L.; Coletta, I.; Mangano, G.; Alisi, M A.; Polenzani, L.; Milanese, C.; Anzellotti, P.; Ricciotti, E.; Dovizio, M.; Di Francesco, A.; Tacconelli, S.; Capone, M L.; Patrignani, P Biochem Pharmacol 2010, 79, 974

22 Wu, T Y.; Juteau, H.; Ducharme, Y.; Friesen, R W.; Guiral, S.; Dufresne, L.; Poirier, H.; Salem, M.; Riendeau, D.; Mancini, J.; Brideau, C Bioorg Med Chem Lett 2010, 20, 6978

23 Chiasson, J F.; Boulet, L.; Brideau, C.; Chau, A.; Claveau, D.; Cote, B.; Ethier, D.; Giroux, A.; Guay, J.; Guiral, S.; Mancini, J.; Masse, F.; Methot, N.; Riendeau, D.; Roy, P.; Rubin, J.; Xu, D.; Yu, H.; Ducharme, Y.; Friesen, R W Bioorg Med Chem Lett 2011, 21, 1488

24 Abdel-Magid, A F ACS Med Chem Lett 2012, 3, 703

25 Rorsch, F.; Buscato, E.; Deckmann, K.; Schneider, G.; Schubert-Zsilavecz, M.; Geisslinger, G.; Proschak, E.; Grosch, S J Med Chem 2012, 55, 3792

26 Shiro, T.; Takahashi, H.; Kakiguchi, K.; Inoue, Y.; Masuda, K.; Nagata, H.; Tobe,

M Bioorg Med Chem Lett 2012, 22, 285

27 Pelcman, B.; Olofsson, K.; Schaal, W.; Kalvins, I.; Katkevics, M.; Ozola, V.; Suna,

E PCT Int Appl WO 2007042816, 2007.

28 Bauer, J.; Waltenberger, B.; Noha, S M.; Schuster, D.; Rollinger, J M.; Boustie, J.; Chollet, M.; Stuppner, H.; Werz, O ChemMedChem 2012, 7, 2077

29 Rorsch, F.; Wobst, I.; Zettl, H.; Schubert-Zsilavecz, M.; Grosch, S.; Geisslinger, G.; Schneider, G.; Proschak, E J Med Chem 2010, 53, 911

30 Hamza, A.; Zhao, X.; Tong, M.; Tai, H H.; Zhan, C G Bioorg Med Chem 2011, 19,

6077

31 Park, S J.; Han, S G.; Ahsan, H M.; Lee, K.; Lee, J Y.; Shin, J S.; Lee, K T.; Kang,

N S.; Yu, Y G Bioorg Med Chem Lett 2012, 22, 7335

32 He, S.; Li, C.; Liu, Y.; Lai, L J Med Chem 2013, 56, 3296

33 Lauro, G.; Strocchia, M.; Terracciano, S.; Bruno, I.; Fischer, K.; Pergola, C.; Werz, O.; Riccio, R.; Bifulco, G Eur J Med Chem 2014, 80, 407

34 Jegerschold, C.; Pawelzik, S C.; Purhonen, P.; Bhakat, P.; Gheorghe, K R.; Gyobu, N.; Mitsuoka, K.; Morgenstern, R.; Jakobsson, P J.; Hebert, H Proc Natl Acad Sci U.S.A 2008, 105, 11110

35 Sjogren, T.; Nord, J.; Ek, M.; Johansson, P.; Liu, G.; Geschwindner, S Proc Natl Acad Sci U.S.A 2013, 110, 3806

36 Li, D.; Howe, N.; Dukkipati, A.; Shah, S T.; Bax, B D.; Edge, C.; Bridges, A.; Hardwicke, P.; Singh, O M.; Giblin, G.; Pautsch, A.; Pfau, R.; Schnapp, G.; Wang, M.; Olieric, V.; Caffrey, M Cryst Growth Des 2014, 14, 2034

37 Kruskal, W H.; Wallis, W A J Am Stat Assoc 1952, 47, 583

38 Seifert, M H J Chem Inf Model 2006, 46, 1456

39 Seifert, M H.; Kraus, J.; Kramer, B Curr Opin Drug Disc Dev 2007, 10, 298

40 Discovery Studio Modeling Environment, version 3.5; Accelrys Software: San Diego, CA, 2005–2012

41 PipelinePilot, version 8.0.1; SciTegic: San Diego, CA, 2001–2010

42 Maestro Suite, version 9.2.112; Schrödinger, LLC: New York, NY, 2011

43 Analyse-it Method Evaluation for Microsoft Excel, version 2.26; Analyse-it Software, 1997–2013

44 Schuster, D.; Laggner, C.; Steindl, T M.; Palusczak, A.; Hartmann, R W.; Langer,

T J Chem Inf Model 2006, 46, 1301

45 Schuster, D.; Markt, P.; Grienke, U.; Mihaly-Bison, J.; Binder, M.; Noha, S M.; Rollinger, J M.; Stuppner, H.; Bochkov, V N.; Wolber, G Bioorg Med Chem.

2011, 19, 7168

46 Braga, R C.; Andrade, C H Curr Top Med Chem 2013, 13, 1127

47 Veber, D F.; Johnson, S R.; Cheng, H Y.; Smith, B R.; Ward, K W.; Kopple, K D.

J Med Chem 2002, 45, 2615

48 Lipinski, C A.; Lombardo, F.; Dominy, B W.; Feeney, P J Adv Drug Delivery Rev.

1997, 23, 3

49 Rogers, D.; Hahn, M J Chem Inf Model 2010, 50, 742

50 Heikamp, K.; Bajorat, J Future Med Chem 2012, 4, 1945

51 Smellie, A.; Teig, S L.; Towbin, P J Comput Chem 1995, 16, 171

52 Li, J.; Ehlers, T.; Sutter, J.; Varma-O’brien, S.; Kirchmair, J J Chem Inf Model.

2007, 47, 1923

53 Barnum, D.; Greene, J.; Smellie, A.; Sprague, P J Chem Inf Comput Sci 1996, 36,

563

54 Fechner, N.; Hinselmann, G.; Wegner, J K In Handbook of Chemoinformatics Algorithms; Faulon, J.-L., Bender, A., Eds.; Chapman & Hall/CRC: Boca Raton, FL, 2010; pp 89–143

55 Seidel, T.; Ibis, G.; Bendix, F.; Wolber, G Drug Discovery Today Technol 2010, 7, e221

56 Schaible, A M.; Traber, H.; Temml, V.; Noha, S M.; Filosa, R.; Peduto, A.; Weinigel, C.; Barz, D.; Schuster, D.; Werz, O Biochem Pharmacol 2013, 86, 476

57 Gutmanas, A.; Alhroub, Y.; Battle, G M.; Berrisford, J M.; Bochet, E.; Conroy, M J.; Dana, J M.; Fernandez Montecelo, M A.; van Ginkel, G.; Gore, S P.; Haslam, P.; Hatherley, R.; Hendrickx, P M.; Hirshberg, M.; Lagerstedt, I.; Mir, S.; Mukhopadhyay, A.; Oldfield, T J.; Patwardhan, A.; Rinaldi, L.; Sahni, G.; Sanz-Garcia, E.; Sen, S.; Slowley, R A.; Velankar, S.; Wainwright, M E.; Kleywegt, G J Nucleic Acids Res 2014, 42, D285

58 LigandScout, version 3.1; Inte: Ligand GmbH: Maria Enzersdorf, 1999–2013

59 Koeberle, A.; Zettl, H.; Greiner, C.; Wurglics, M.; Schubert-Zsilavecz, M.; Werz,

O J Med Chem 2008, 51, 8068

60 Bauer, J.; Koeberle, A.; Dehm, F.; Pollastro, F.; Appendino, G.; Northoff, H.; Rossi, A.; Sautebin, L.; Werz, O Biochem Pharmacol 2011, 81, 259

61 Claveau, D.; Sirinyan, M.; Guay, J.; Gordon, R.; Chan, C C.; Bureau, Y.; Riendeau, D.; Mancini, J A J Immunol 2003, 170, 4738

62 Koeberle, A.; Werz, O Curr Med Chem 2009, 16, 4274