A highly selective structure-based virtual screening model of Palm i allosteric inhibitors of HCV Ns5b polymerase enzyme and its application in the discovery and optimization of new analogues

European Journal of Medicinal Chemistry 57 (2012) 468e482

Contents lists available at SciVerse ScienceDirect

European Journal of Medicinal Chemistry

journal homepage: h ttp://www.elsevier.com/locate/ejmech

Original article

A highly selective structure-based virtual screening model of Palm I allosteric

inhibitors of HCV Ns5b polymerase enzyme and its application in the discovery

and optimization of new analogues

*

**

Amr H. Mahmoud , Dalal A. Abou El Ella, Mohamed A.H. Ismail, Khaled A.M. Abouzid

Department of Pharmaceutical Chemistry, Faculty of Pharmacy, Ain Shams University, 6 b abadeer zeitoun, Cairo 11566, Egypt

a r t i c l e i n f o

a b s t r a c t

Article history:

First structure-based activity prediction model of topologically diverse inhibitors of Palm I allosteric site

of HCV NS5b polymerase enzyme is reported here. The model is a workﬂow of structure-based

pharmacophore followed by guided docking. The pharmacophore was constructed using a novel

procedure which includes PLIF (protein ligand interaction ﬁngerprint), Hypogen, contact-based

pharmacophore and shape constraints. The guided docking was tweaked using both a scoring function

of high correlation with activity (ChemPLP) and essential pharmacophore features. Statistically, ROC

analysis for the workﬂow, deploying the novel technique of virtual decoys, yielded AUC of 0.947.

Experimentally, the model was used to screen Asinex GOLD database yielding a new hit with a different

scaffold which was further conﬁrmed by synthesis and biological evaluation.

Received 28 November 2011

Received in revised form

8 April 2012

Accepted 12 April 2012

Available online 25 April 2012

Keywords:

Hypogen

PLIF

GOLD

Knime

Muse

HCV

Virtual screening

Guided docking

Virtual decoys

1. Introduction

site inhibitors, allosteric site inhibitors and miscellaneous [4].

Mechanistic and structural studies have revealed the existence of

Hepatitis C virus (HCV) is one of the major causes of cirrhosis,

hepatocellular carcinoma and liver failure that lead to trans-

plantation [1]. Since its identiﬁcation in 1989 [2], it has been esti-

mated that more than 170 million people worldwide are infected by

HCV. The HCV genome encodes one single poly-protein of approx-

imately 3000 amino acids, depending on the HCV genotype. This

poly-protein is then proteolytically processed by viral and cellular

proteases to produce structural proteins (core, E1, E2 and p7) and

nonstructural proteins (NS2, NS3, NS4A, NS4B, NS5A, and NS5B) [3].

The NS5B RNA-dependent RNA polymerase is recognized as a key

target for therapeutic intervention [4e7] mainly because it does not

have a mammalian counterpart and offers a wide range of possi-

bilities for the discovery of new molecular entities as anti-HCV

agents. The inhibitors of this target are either nucleoside or non-

nucleoside inhibitors. The latter is further classiﬁed into active

multiple allosteric binding sites [8], and in particular two thumb

sites (thumb I and II) and three palm pockets (palm I, II and III) have

been identiﬁed to date. According to the target site, the different

inhibitors will be referred to as palm site I NNIs (PSI-NNIs), palm site

II NNIs (PSII-NNIs), palm site III NNIs (PSIII-NNIs), thumb site I NNIs

(TSI-NNIs) and thumb site II NNIs (TSII-NNIs)(Fig. 1).

Palm I site allosteric site is targeted by wide variety of inhibitors

that have topologically diverse scaffolds: acrylic acids, rhodanines,

benzothiadiazines, benzothiazines, benzoisothiazole dioxides,

proline sulfonamides, acyl pyrollodines, anthranilic acids and

benzodiazepines [9]. This site has been explored structurally in

literature in order to ﬁnd out the binding mode mainly [10] and not

for virtual screening purpose. Virtual screening on this site was

carried out previously adopting the ligand based 3D-QSAR method

using GOLPE [11]. Generally, ligand based methods suffer from the

common problem of limited applicability domain. They can’t

retrieve with reliability any ligands having information more than

that they were trained on. This triggered us to develop for the ﬁrst

time a multipurpose predictive structure-based computational

model for the inhibitors of this site (i.e. palm I). The model was

* Corresponding author. Tel.: þ202 01062656276.

** Corresponding author. Tel.: þ202 020122165624; fax: þ202 25080728.

E-mail

addresses:

amr.hamed@pharm.asu.edu.eg

(A.H.

Mahmoud),

khaled.abouzid@pharm.asu.edu.eg (K.A.M. Abouzid).

doi:10.1016/j.ejmech.2012.04.016

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

469

Fig. 1. HCV NS5b different allosteric sites with schematic diagrams of protein ligand interactions of each allosteric site. In this study, we focus on Palm I site. Picture was prepared

using Accelrys Discovery Studio 3 and Biosolveit Pose view.

designed while focusing on achieving targets: a-virtual screening

enrichment by having superior receiver operator characteristic

curves (ROCs) [12] [13]; b-predicting the activity of the inhibitors; c-

comprehensive SAR analysis of the site; d-binding mode prediction.

We tried to construct this model using an idea that depends on

enhancing the performance of hits ﬁnding by directing the search

through all the available information related to the palm I site. This

information can be translated into the essential chemical features

required for the activity and a ﬁt value which correlates well with

the activity of the ligands. One tool which can achieve this goal is

guided docking [14] which relies on multiple object optimization

using genetic [15] or evolutionary algorithm [16]. It uses both types

of information we mentioned: the chemical features (pharmaco-

phore restraints) and the correlating ﬁt value which both aim at

achieving a balance between favorable interactions with the

receptor without deviating from the ideal binding pattern [14].

Moreover, we were aiming at accelerating the process of optimi-

zation during docking sampling by initially providing the docking

engine with good solutions. In order to achieve this, we preceded

docking by structure-based pharmacophore which supplies the

docking engine with hits having both good orientation and

conformation. In this way, the convergence in the docking algo-

rithm can be achieved rapidly and without losing possible hits due

to exceeding the number of iterations during docking sampling.

Besides, pharmacophores (even if not structure-based) are nor-

mally used before docking in order to rapidly ﬁlter ligands which

don’t have the essential features.

interactions with minor need for energy reﬁnement in the actual

binding site.

As a second step, knowledge-based docking was to be applied

on the hits found by the pharmacophore. It should be guided by

two main criteria: 1-Fit value: a scoring function which correlates

well with the activity; 2-Soft constraints: provided by a geometrical

pharmacophore which is deduced from the above one. The

constraints should be applied in order to avoid poses generation

which are not satisfying the main features. Thus, we aimed to use

docking for reﬁnement of the pharmacophore-provided hits based

on the actual ligandeprotein interactions. This is unlike pharma-

cophores which are not based on actual interactions adjustment

even if structure-based. In addition, docking is used for the

assessment of the hits using a scoring function that correlates with

activity and at the same time provide a more accurate binding

mode prediction.

Ideally, the docking solutions should not deviate much from that

provided by the pharmacophore (i.e. small RMSD deviation) and

this was the aim of this workﬂow: obtaining quick solutions from

docking by initially providing it with good seeds as parents for the

GA (genetic algorithm) and thus achieving fast convergence.

However, this gave rise to an argument of whether docking is

actually important or the structure-based pharmacophore is

enough and should be followed by just scoring. This was concluded

from the results as will be shown.

In order to follow the results sequence, we provided the ﬂow of

the protocol illustrated in Fig. 2 and described it brieﬂy as follows:

Consequently, we planned the workﬂow (Fig. 2) to be used as

follows: A highly reﬁned structure-based pharmacophore of high

selectivity was to be constructed such that the hits found meet the

following criteria: 1-Satisify the essential features (those with high

probability or propensity in the diverse active inhibitors); 2-Have

accurate pose of very small RMSD, thus satisfying the essential

1 -Genertation of a highly reﬁned hybrid structure-based phar-

macophore using the following steps:

A -Superposition of 6 PDB complexes of HCV polymerase

inhibitors of allosteric site C and generating a PLIF [17,18]

(protein ligand interaction ﬁngerprint) to highlight

470

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

Fig. 2. Workﬂow used to study Palm I site diverse scaffolds. It consists of structure-based pharmacophore and guided docking modules.

common interactions (those having high probability)

between co-crystallized ligands and the proteins.

B -Applying different weights and tolerances to the extracted

features according to their importance using activity based

pharmacophore (Hypogen) [19].

C -Applying contact-based pharmacophore [20] to extract

excluded volumes which represents the regions where the

ligand will sterically clash with.

D -Applying shape constraints and adjusting its tolerance to

accommodate all ligands.

E -Validation of the pharmacophore by using virtual decoys

[21]of high relevance as a more accurate test for ROC

analysis.

bonds, ionic interactions and surface contacts are classiﬁed

according to the residue of origin, and built into a ﬁngerprint

scheme. It can be used to obtain the common features shared by

group of inhibitors by analyzing the ﬁngerprints. In order to classify

interaction ﬁngerprints, each amino acid residue of a protein is

broken down into categories, as shown in Fig. 3 where there are 6

types of interactions in which a residue may participate: side chain

hydrogen bonds (donor or acceptor), backbone hydrogen bonds

(donor or acceptor), ionic interactions, and surface interactions. The

most potent of each of these interactions in each category, if any,

are considered. The way the ﬁngerprint is calculated is explained in

the supplementary data in section 1.

Here in this study, the input data for the PLIF was six X-ray

crystal structures of six inhibitors which share the same binding

site (Palm I allosteric site of HCV polymerase NS5b genotype 1b)

and at the same time have diverse activities and scaffolds. These X-

2 -Selecting a suitable scoring function which can describe the

activity well.

3 -Utilizing knowledge-based docking with GA search engine

and provide it with the suitable ﬁt value (scoring function

selected in the previous step), pharmacophore constraints

while keeping the essential structural water molecules.

2. Results

2.1. Pharmacophore generation

As mentioned before, the aim of the structure-based pharmaco-

phore was to act as a fast substitute for docking by initially providing

good solutions with good pose accuracy. In addition, it performs the

normal pharmacophore function which is reducing the chemical

space and enriching the solutions with actives by acting as a pre-

ﬁlter that screens those ligands not satisfying essential features.

This dual functionality was done using a hybrid strategy as follows:

2.1.1. PLIF analysis

Fig. 3. PLIF (protein ligand interaction ﬁngerprint) way of binning. There are 6 types of

interactions in which a residue may participate: side chain hydrogen bonds (donor or

acceptor), backbone hydrogen bonds (donor or acceptor), ionic interactions, and

surface interactions. Each interaction is further divided into low or high.

The Protein Ligand Interaction Fingerprints (PLIF) tool is

a method for summarizing the interactions between ligands and

proteins using a ﬁngerprint scheme. Interactions such as hydrogen

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

471

Table 1

obvious. Molecular dynamics was carried out previously [10] and

results focused on hydrogen bonding only so we replicated the

simulation and stressed the other binding motifs (e.g. hydrophobic)

in the interaction diagrams (see section 2 in the supplementary

data) and mentioned them in Table 2.

Non-nucleoside inhibitors of diverse scaffolds acting on Palm I site and their protein

data bank codes.

Scaffold type

PDB codes

Acrylic acids

Rhodanines

1YVF, 1Z4U

2AWZ, 2AXO, 2AX1

Benzothiadiazines

2FVC, 2GIQ, 3HHK, 3BSA, 3BSC, 3CDE,

3BR9,3E51, 3CO9, 3CVK, 3H2L, 3H98, 3GYN, 3IGV

3CWJ, 3G86, 3H59

3D28, 3D5M, 3H5U, 3H5S, 2GC8, 2JC0, 2JC1

2GC8

2JC0,2JC1

2QE2, 2QE5

3GOL, 3CSO, 3GNV, 3GNW, 3HKW, 3HKY

2.1.2. Analysis of the binding mode shared by the 6 pdbs

We took 2GIQ benzothiadiazine bound conformer (Fig. 4) as

a reference in this analysis and used the four shared features to

analyze the binding mode:

1-Hydrophobic feature 1 is represented by Ring A in 2GIQ bound

conformer. This feature is essential to bind in the hydrophobic

pocket I which consists of amino acid residues Phe415, Met414,

Asn411 and Gly410.

Benzothiazines

Benzoisothiazoles dioxide

Proline sulfonamides

Acylpyrrolodines

Anthranilic acids

Benzodiazepines

ray crystals (3HKY, 1YVF, 2GC8, 2GIQ, 2JC0, 2QE5) were obtained

from the protein data bank [22] and represent benzodiazepines,

acrylic acids, proline sulfonamides, benzothiadiazines, acylpyrroli-

dines and anthranilic acids respectively. These classes and the PDB

codes corresponding to them are shown in (Table 1). Benzothia-

zines and benzoisothiazoles dioxide based inhibitors are excluded

since they are bioisosteres to benzothiadiazines and are likely not

to add more information to the structural requirements. We didn’t

select all the crystal structures and preferred to pick up a complex

of each category to simplify method of extraction of common

features and avoid over-representation and redundancy as all

crystal structures in each category bear the same scaffold.

Brieﬂy, the 6 X-ray structures were superposed and the ligand

interaction diagrams (Fig. 4) were created to extract common

interactions of high propensity (see section 2 of the supplementary

data for full 3D interaction diagrams). According tothis analysis, four

features were retrieved as common between the scaffolds (as shown

in Table 2). These features were mapped on the bound ligands as

shown in Fig. 5. Before we give detailed analysis, we saw it is

necessary to mention the importance of prior explicit solvation-

based molecular dynamics in this case because many of the inter-

actions were not obvious in the bound conformer state. For example,

sigmaepi interaction of Met414, hydrogen bonding with Gly449 and

generally the binding motifs of hydrophobic interactions were not

2-Hydrophobic feature 2 is represented by hydrophobic tail

attached to N1 of Ring B. This feature is essential for binding in the

hydrophobic pocket II which consists of amino acid residues

Pro197, Leu384 and Cys366 and Tyr415 .The latter showed speciﬁc

importance in pocket II where it forms piepi stacking as a common

hydrophobic interaction motif (2GC8 and 2JC0)(see section 2 in the

supplementary data for each bound conformer interactions).

3-Two nearby hydrogen bond acceptor features. These features

are essential for binding with polar amino acid residues

Tyr448,Gly449 and Ser556. The ﬁrst feature is represented by

Oxygen atom of hydroxyl group of 4-hydroxyquinolin-2(1H)-one

which forms H-bond with Tyr448 directly and indirectly via water

bridge with Gly449. The second feature is represented by oxygen

atom of sulfonyl group which forms indirect H-bond with Ser556

via water bridge.

The other bound ligands were analyzed the same way and

summarized in Table 3 (see section 2 of the supplementary data to

visulaize the interactions mentioned).

One important result regarding this analysis is that 2GC8 bound

ligand forms hydrogen bond with Gln446 and piepi stacking with

Tyr415. These 2-amino acids are mutated in genotype 1a where

gluatamine is mutated to glutamic acid and tyrosine is mutated to

phenylalanine. This hydrogen bonding is absent in case of mutation

and this may be the reason why the activity of palm I site inhibitors

Fig. 4. Schematic diagrams of protein ligand interaction for each scaffold at the palm I site. Red boxes show the features which are responsible for hydrophobic interaction with

pocket II while blue boxes show those responsible for hydrophobic interaction with pocket I. Tyrosine448 forms hydrogen bond with all ligands. Met414 is another important amino

acid which interacts hydrophobically with the ligand. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.)

472

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

Table 2

Common features of high propensity extracted using PLIF: 2 hydrophobic features and 2 hydrogen bond acceptors.

Feature

Complementary site Corresponding amino acids

Description

Depiction

Hydrophobic feature 1

Pocket I

Consists of residues Phe415, Met414, Asn411 and A hydrophobic shallow and wide s Depicted by blue dotted

Gly410, close to the surface of the enzyme.

ub-pocket

lines

Hydrophobic feature 2

Pocket II

Consists of residues Pro197, Leu384 and Cys366

Deep and narrow hydrophobic

sub-pocket

Depicted by red dotted

lines

Hydrogen bond acceptor Polar amino acid

feature 1 residue

Hydrogen bond acceptor Polar amino acid

feature 2 residue

Tyr448

Gly449

Involved in polar interactions

with inhibitor

Involved in polar interactions

with inhibitor

Depicted by purple circle

Depicted by orange circle

against genotype 1a is generally less than 1b [23]. This phenom-

enon can be extended to explain the decrease of the activity of

other scaffolds, for example benzothiadiazines. The above

mentioned hydrogen bond can be formed by the benzothiadiazines

in case the ligand is shifted slightly upwards away from the re-

ported bound conformer. This was proved by MD (molecular

dynamics) study which conﬁrmed the feasibility of formation of

this hydrogen bond. As a result, this ﬁnding may aid in under-

standing the effect of mutation of Gln446 which decreases the

activity of the compounds acting on this allosteric site [23,24].

using 6 ligands only which was not feasible without minimizing the

number of matching points in the feature used [19].

2.1.4. Hypogen

Using the information retrieved from the PLIF analysis, Hypogen

[19] was used to generate a pharmacophore using the custom-

made hydrogen bond acceptor and hydrophobic features while

allowing no conformations generation during the run. The co-

crystallized ligands were used in their bound conformation and

no conformation generation was needed. Hypogen was used as it

modiﬁes weights and tolerances of the features such that they are

correlated with the activity trend.

2.1.3. Custom feature generation

Based on the above analysis that indicates that there are two

hydrophobic features and two hydrogen bond acceptors in

common among the 6 inhibitors, we decided to generate a Hypo-

gen-based pharmacophore using these features. The main reason

beyond this was to be able to add different weights and tolerances

to the common features according to the activity of the inhibitors

and thus becoming an activity based query. Moreover, this enables

us to use the technology of Accelrys catalyst pharmacophore

functionalities like adding shape and excluded volume functional-

ities, screening using catalyst engine and ﬁnally integrates it with

gold docking engine which is supported in the Accelrys Discovery

studio environment. Some changes were needed however like the

need to modify the hydrogen bond acceptor feature. It was

customized such that it maps only oxygen type acceptors and

encompass only acceptor origin point (deleting both vector direc-

tion and projection point). The ﬁrst criterion which is the mapping

oxygen type acceptors only was done to increase selectivity of the

pharmacophore where it has been shown through SAR analysis by

Tedesco et al. [25] that the amino group (NH2), as an acceptor,

showed a drastic drop in activity. The second criterion which is

decreasing the number of the mapping points and restricting it to

the origin was done to comply with Hypogen module as it was built

Hypogen was selected as we were aiming here to study topo-

logically diverse inhibitors. It was not intended here to study

a congeneric series with same scaffold and minor change in

substituents (R-variation) which prefers methods like CoMFA and

CoMSIA.

Hypogen algorithm produces three cost values during genera-

tion of pharmacophore in order to assess its quality. The ﬁrst cost

value is the ﬁxed cost (also known as the ideal cost). It represents

the simplest model that ﬁts the data perfectly. The null cost, also

known as no correlation cost, represents the highest cost of

a pharmacophore with no features. One should expect that signif-

icant pharmacophore should have large difference between these

two values. The total cost is calculated for each pharmacophore and

should be close to the ﬁxed cost value.

These considerations should be taken into account if the

Hypogen model was constructed to estimate the activity. However,

in this study, it was used to be a part of hybrid structure-based

pharmacophore which aims at accurately placing the ligand in

the binding site and achieving a good ranking power. Speciﬁcally

Hypogen was selected to generate the pharmacophore instead of

HipHop(which doesn’t take activity into consideration) [19]

because it can achieve the purpose of accurately placing ligands

Fig. 5. Applying PLIF on 6 diverse scaffolds of Palm I site. The color-coded amino acids shown in the key are those amino acids interacting with the ligands. The ligands interact with

these amino acids by the matching features (have the same color as the interacting amino acids). (For interpretation of the references to color in this ﬁgure legend, the reader is

referred to the web version of this article.)

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

473

Table 3

Analysis of the 6 bound conformers found in the pdbs chosen to represent Palm I site. Analysis was carried out using features found of high propensity using PLIF analysis.

Scaffold

Hydrophobic feature 1

Hydrophobic feature 2

Ring C of 1YVF

Two nearby hydrogen bond acceptor features.

Acrylic acid

derivative (1YVF)

Ring A of 1YVF forms piesigma

interaction with Met414

O of the carbonyl directly attached to Ring

A H-bonds with Tyr448 while carboxylic

oxygen H-bonds with Gly449

Proline sulfonamide

derivative(2GC8)

Ring A (proline Ring)

Ring B forms piepi interaction

with Tyr415

O of sulfonamide group H-bonds with Tyr448

while carboxylic oxygen H-bonds with Gly449

and Gln446

Acyl pyrrolidine

derivative (2JC0)

Isobutyl group attached to Ring

B of 2JC0 bound conformer

Ring C forms piepi interaction

with Tyr415

O of carbonyl group attached to Ring C H-bonds

with Tyr448 while carboxylic oxygen H-bonds

with Gly449

Anthranilic acid

derivative(2QE5)

1,5-Benzodiazepine

derivative(3HKY)

Ring A of 2QE5 bound conformer

Ring C and attached methyl group

Ring B of 2QE5 bound conformer

Hydrophobic tail attached to Ring D

Oxygen of carboxylic group H-bonds with

Tyr448

Sulfonyl oxygen forms H-bond with both

Gly448 and Tyr449 while Ring E oxygen

forms H-bond with Ser556 via water bridge.

effectively. This can be explained as Hypogen was built on the

assumption that active molecules should map more features than

inactive and mapping should be perfect for most active ligands. In

another words, inactivity is attributed to the missing of important

feature or, most important in our case, improper orientation of the

feature in the space. The Hypogen model can serve this purpose by

properly placing the features in the space and assign different

weights and tolerances to them. Based on this fact, we selected the

best pharmacophore according to the pose accuracy. The best one

(we shall call it hypothesis 1 or shortly Hypo1) is shown in Table 4

where it was capable to minimize RMSD as much as possible and

was selected for further reﬁnement (Fig. 6). It is worth noting that

the pharmacophore has 5 features and not 4 which means that

there is an extra feature assigned by Hypogen engine. This extra

feature maps to an additional hydrophobic feature found in the

most active inhibitor which is the 1,5 benzodiazpine found in 3HKY

complex.

The best hypothesis (Hypo1) was modiﬁed by changing feature

weights manually till reaching a satisfactory RMSD while keeping

the features tolerances as they are. The motive beyond the weights

selection was the binding pocket analysis. The 2 hydrophobic

features were given the largest values (both of them were given 20)

since they act as an anchor for the ligands and represent the main

bulk of it. Tyr448 role in hydrogen bonding was shown to be more

prominent than Gly449 as it forms hydrogen bond directly with the

ligand and therefore was given a higher weight (was given a weight

of 10) while Gly449 mostly forms hydrogen bond via water bridge

(so was given a weight of 5).

In our opinion, multiple-parameter optimization can be used to

adjust the features weights, tolerances and positions such that it

achieves best activity correlation and best pose accuracy at the

same time.

2.1.5. Contact-based pharmacophore generation

This technique was developed by Wolber and Langer as a good

technique for screening of new compounds instead of computa-

tionally expensive docking [20]. The technique was implemented

already in ligandscout software [26]. This algorithm extracts

information according to certain rules depending on nearby contact

residues. It was used here to reﬁne Hypo2 by simply creating

excluded volume constraints based on the crystal of the largest

inhibitor present (3HKY). Thus, the pharmacophore created above

(Hypo2) was just clustered with the excluded volume. Shape

constraint was applied in addition to this pharmacophore using the

3HKY bound conformer shape since it was the biggest ligand while

decreasing tolerance from the default value 0.5 to 0.3 in order to

accommodate all the other ligands. The ﬁnal form of the pharma-

cophore (Hypo3) is depicted in (Fig. 7).

The pharmacophore (Hypo3) was employed such that ﬂexible

ﬁtting is used and only one feature is allowed to miss. The best

mapping was set to the one with the highest ﬁt value and not the

one with most features.

To rationalize these additional constraints (excluded volume

and shape constraint), we screened a library of highly diverse

ligands which is extracted from CAP database 2006 (chemicals

available for purchase). This library consists of 9607 diverse ligands.

The library was screened using Hypo3 with different features

conﬁgurations and the results were tabulated in Table 7.

Table 5 shows the new modiﬁed hypothesis (Hypo2) which

indicates a better pose accuracy as shown in Table 6.

Table 4

Hypogen unmodiﬁed hypothesis (Hypo1). First table shows 5 features of this

pharmacophore and their corresponding weights and tolerances. Second table

shows Hypogen different mappings for each bound conformer of the 6 X-ray crystal

structures. Fit represents ﬁt value, Est. represents estimated activity and act repre-

sents actual activity. Third table shows the three cost values of the selected phar-

macophore: Null, ﬁxed and total costs. Correlation value is also shown in the table.

Deﬁnition: Hydrophobic Hydrophobic Hydrophobic Or_1

Or_1

1.86857 2.73099

1.30 1.45

Act Uncrt

Weights:

Tolerances: 1.75

1.86857

1.30

1.86857

1.45

Name Fit

Cnf/Enan Mapping

Est

Err

3HKY

2GIQ

1YVF

2JC0

2QE5

2GC8

7.84

1

þ[27 39 1 34 17]

11

10 þ1.1

50 þ1.1

100 þ1.1

300 þ1.2

3

7.15

6.86

6.34

5.73

5.53

þ[22 * 4 12 28]

þ[28 23 * 4 17]

þ[2 20 32 11 *]

þ[* 21 8 * 3]

54

110

360

Fig. 6. Hypogen initial unmodiﬁed pharmacophore(Hypo1). The cyan-colored features

are hydrophobic while the green-colored features represent customized H-bond

acceptor features (mainly map oxygen type acceptors). (For interpretation of the

references to color in this ﬁgure legend, the reader is referred to the web version of this

article.)

1400 1600 ꢀ1.1

þ[8 * * 10 18]

2300 3100 ꢀ1.4

Null cost: 29.9068 Fixed cost: 50.5808 Total cost: 50.8555 Correl ¼ 0.998492

474

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

Table 5

Different pharmacophore features of Hypogen and their corresponding weights

after modiﬁcation (Hypo2).

Feature

Weight

Hydrophobic feature A

Hydrophobic feature B

Hydrophobic feature E

20

1

H-bond acceptor feature C (with Tyr448)

H-bond acceptor feature D (with either Gly449 or Ser556)

10

5

These results show the selectivity of the pharmacophores (the

lower the number of hits retrieved, the more selective the phar-

macophore). It is obvious that the common features shared among

the ligands of the 6 pdbs are four but these are not enough to attain

selectivity. Additionally, one can realize that each ligand has some

additional feature beside those four, especially in the most active

inhibitors. For example, as mentioned before, Hypogen assigned an

additional hydrophobic feature to the pharmacophore which is

found in the most active 1,5 benzodiazpine (3HKY). Another

example is the hydrogen bond acceptor feature found in benzo-

thiadiazines. It has been shown in the previous studies on this class

that there is an extra binding pocket (pocket III) [27]where one of

the strategies to increase both potency and the selectivity of this

class of inhibitors is achieved by substituting Ring D of the ben-

zothiadiazine with methansulfonamide which can form extra two

hydrogen bonds with both Asn291 and Asp316 (see section 3 of the

supplementary data for details).

Fig. 7. Structure-based 3D pharmacophore (Hypo3) as an alternative to docking. Gray

spheres represent excluded volumes, cyan spheres represent hydrophobic features,

and green spheres represent hydrogen bond acceptors of special type (restricted to

oxygen type acceptors only). Shape constraint is also shown. (For interpretation of the

references to color in this ﬁgure legend, the reader is referred to the web version of this

article.)

Conclusively, the pharmacophore (Hypo3) was tested so far for

both pose accuracy and selectivity. Additionally, as we shall see the

based pharmacophore before docking has many advantages: (a) it

reduces chemical space; (b) it provides docking with good solutions

which have correct orientations and conformations, thus time

required for sampling is reduced and convergence is achieved

rapidly. Thus, guided docking was applied as a reﬁnement tool for

initially good poses that provides a good binding prediction and at

the same time an assessment tool that rank the compounds

according to activity thus achieving an enhanced enrichment. We

didn’t sufﬁce with the structure-based pharmacophore only as it

doesn’t take the actual interactions into consideration and thus

can’t reﬁne pose inside binding site.

Guided docking and not normal docking was used in this

workﬂow although the hits were coming from the pharmacophore

and should be in good orientation. This is because of the of the

binding site nature which is wide enough to accommodate ligands

in different conformations which may not be suitable to illicit

activity. On the contrary, guided docking help to keep the good pose

in its place without altering it and restricts its function on the

energy reﬁnement. Direct approach [14] of this docking was

employed using chemical information explicitly to actively guide

the orientation of the ligand during sampling. This was achieved

using a scoring function which correlates well with activity while

applying constraints which penalize divergence from proper

pharmacophore has

a very good enrichment capability and

powerful ability to separate actives from decoys. This was tested in

the validation section (see below) and proved to be of very good

ROC (AUC ¼ 0.884). The predictive power of the selected model

should be, however, considered with care because the number of

the ligands used to build the Hypogen model was less than 16 and

the difference between ﬁxed and null cost is less than 40 [19]. This

was solved by following the pharmacophore with guided docking

engine supplied with a better correlating scoring function.

The success of this pharmacophore was based on selecting the

best of breed from different algorithms: (1) PLIF is considered

a method which can deal with many superposing complexes to

retrieve common features according to their propensity and

signiﬁcance; (2) Hypogen can adjust weight, tolerance and position

of different features to correlate well with activity; (3) The weight

variation and setting the selection criteria to the best ﬁt value

instead of number of features; (4) Hypothesis reﬁnement tech-

niques using shape constraints and excluded volumes.

2.2. Guided docking

Hits found by the Hybrid structure-based pharmacophore

search described above were subjected to Knowledge-based

docking using GOLD [15].As mentioned before, using a structure-

Table 7

The number of hits retrieved from CAP diverse using different conﬁgurations for the

Hypo3. The selectivity increases as the number of hits decreases.

Table 6

Pose accuracy of different inhibitors using the modiﬁed

hybrid pharmacophore (Hypo2).

Features used

Number of compounds

retrieved from CAP diverse^a

Bound conformer

RMSD^a

5 Features(maximum omitted features ¼ 1)

5 Features (maximum omitted features ¼ 1)

and excluded volume

2337

2242

2GC8

2GIQ

2JC0

1YVF

2QE5

3HKY

1.4

1.2

1.3

0.9

1.1

0.7

5 Features (maximum omitted features ¼ 1),

1601

431

excluded volume and shape

5 Features (maximum omitted features ¼ 0),

excluded volume and shape

a

RMSD ¼ Root Mean Square Distance.

CAP diverse ¼ Diverse subset of chemical available for purchase.

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

475

binding by adding a weighed penalizing energy term to the ﬁnal

score. Thus, sampling of both ligand conformational and orienta-

tional space provides a balance between maintaining favorable

interactions with receptor without deviating from the appropriate

binding mode.

(hydrophobic spatial constraints) were applied both on pocket I and

pocket II; (2) Interaction constraints were applied on Tyr448 and

Gly449 using NH of the backbone as H-bond donor feature. Docking

was applied using the 3HKY crystal structure while keeping

essential water molecules and removing those which clash with

other ligands. This was done by aligning all the crystal structures

and analyzing bound ligands interactions. In general, it was found

that there were some water molecules which clash with Ring D of

benzothiadiazine which has no complementary substructures in

the other bound conformers. These water molecules are displace-

able (not ﬁxed) as proved by experimental data.

Docking was validated by re docking study which conﬁrms

capability of GOLD in retrieval of the bound pose. In order to check

the reliability of the docking protocol, we performed docking

studies on the 6 NS5b polymerase complexes. The docking results

were evaluated by the comparison of the predicted docking poses

of the ligands and the experimental ones. RMSD between positions

of heavy atoms of the ligand in the experimental and calculated

structures are taken as a measure of reliability. The best docked

conformation was chosen by selecting the conformer with best

Chemscore. Random conformer of each bound ligand was docked

to prevent bias. Results are shown in Table 9.

GOLD was chosen as it employs genetic optimization for these

multiple parameters at the same time.

Regarding the usage of structure-based pharmacophore before

docking despite the fact that docking employs pharmacophore

constraints, the docking uses soft constraints which penalize

deviation only and will not ﬁlter those ligands which are violating

the main binding pattern. This was observed during docking opti-

mization runs where it was observed that some ligands can achieve

high docking score while not forming many essential interactions.

The wide nature of binding site allows many possible interactions

(other than essential ones) which can increase the score to the limit

it overcomes the penalty and becomes in the top 10% of ranked

database. In addition, it was observed that the number of hits

missed by docking increases in case of not applying structure-based

pharmacophore before it (see the validation study below).

Furthermore, the most important reason is that the pharmaco-

phore reduces chemical space rapidly unlike docking which applies

pharmacophore while taking interactions into consideration in

what is known by multiple parameter optimizations which can

takes much more time.

In general, the RMSD is very good even when using random

conformations because the ligand is guided by pharmacophore

which imposes bias in ligand placement. In addition, CHEMPLP

proved to be a good evaluator for activity having r²of 0.6 which is

sufﬁcient for ranking. To conﬁrm that, spearman rank coefﬁcient

was measured and was found to be 0.82.

Thus, ligand placement was guided by using:

2.2.1. Scoring function

In our studies regarding Benzothiadiazines (unpublished work);

it was found that Moldock score correlates well with activity

yielding a q²of 0.65. The scoring function, however, was not suit-

able to guide ligand placement alone but can be used to predict the

activity of the ligand when it is accurately placed in the binding site.

Thus, the Moldock score applicability was extended in this study to

evaluate ligands of diverse scaffolds acting on palm I site. Molegro

[16] was used evaluate the bound conformers of the 6 pdbs to ﬁnd

the correlation (Table 8). R²was found to be 0.745 which is

signiﬁcant enough to use this docking score for guiding sampling.

Moldock score is a variant of PLP [16]. Since we decided to use

GOLD so we tried to use an equivalent scoring function in the

software that is based on PLP so we selected CHEMPLP [28] to guide

ligand placement. It was checked for its correlation with activity

and was found to be true (see validation study below). In general, it

was found that Glide score [29], CHEMPLP and Moldock scores [16]

are suitable for evaluation of ligands at this site when they are

accurately placed (using constraints).

2.3. Virtual decoys and zinc decoys

Virtual decoy [21] dataset was generated in order to evaluate the

ability of the pharmacophore model, guided docking and combined

workﬂow to separate active ligands from the decoys. The decoys

normally are selected from the zinc database [30e32] such that they

are physically similar yet chemically dissimilar to the active ligands.

Here in this study, we used in addition the virtual decoys as a more

accurate test because they have high TPSA (topological polar surface

area) similarity to the active ligand (additional criteria not found in

zinc decoys). The virtual DUD of this dataset was created using the

strategy adopted by Wallach [21]. A virtual decoy set was generated

using Tripos MUSE [33]. This software deploys multiple-parameter

optimization technique to generate new ligands. It can generate set

of decoys for an active ligand. This method is highly practical and

reproducible. It is based on Knime [34] workﬂow and Evolutionary

algorithm which is applied in Tripos Muse [33]. Muse is a de-novo

building process which is dependent on Evolutionary Algorithm

Inventor (EAI) which was developed by Pearlman group [35]. It is

used to generate new chemical entities form a seed set or previous

generations. It can be used in conjunctionwith a single or composite

external scoring function. Evolutionary algorithm evolves improved

populations of structures by performing modiﬁcations on the

2.2.2. Constraints

Two types of gold constraints were applied, (a) interactions

constraints, which allow to deﬁne parts of the protein which must

form an interaction with the ligand, and (ii) spatial constraints,

which allow to deﬁne regions in space where some kind of ligand

atom must be present. Constraints were applied on the shared

interactions pattern only as follows: (1) Regional constraints

Table 9

Evaluation of GOLD by docking of the bound ligands in the 6 PDBs (using random

conformation) in 3HKY. RMSD is given with respect to the reference and GOLD PLP

score is given.

Table 8

Moldock scoring of the bound conformers of different PDBs after reﬁnement. The

activity of each complex is depicted in minus log units.

Bound conformer

Activity

Gold.PLP^a

RMSD^b

2GC8

2GIQ

2JC0

1YVF

2QE5

3HKY

ꢀ3.49136

ꢀ1.69897

ꢀ2.47712

ꢀ2

ꢀ46.0287

ꢀ48.3316

ꢀ59.9144

ꢀ60.7761

ꢀ47.5208

ꢀ80.856

1.2

Bound conformer

Activity in minus log units

Moldock score

0.43

0.34

0.6

1.4

0.34

2GC8

2GIQ

2JC0

1YVF

2QE5

3HKY

ꢀ3.49136

ꢀ1.69897

ꢀ2.47712

ꢀ2

ꢀ93

ꢀ109

ꢀ122

ꢀ124

ꢀ82

ꢀ3.21484

ꢀ1

ꢀ3.21484

ꢀ1

a

GOLD.PLP ¼ Gold piecewise linear potential scoring function.

RMSD ¼ Root Mean Square Distance.

ꢀ170

b

476

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

Fig. 8. Workﬂow for generating virtual decoys. Ligands are generated using evolutionary algorithm in Tripos muse such that they comply with speciﬁed MWT, number of hydrogen

bond acceptors, number of hydrogen bond donors, number of rotatable bonds and TPSA. The compliance is achieved by applying demerits to any deviation. In order to avoid

chemical similarity to the reference, a ﬁlter (N) was added which keeps molecules with Tanimoto coefﬁcient less than 0.9.

members of the previous generations (mutations) and swapping

successful substructures among different molecules (cross-overs).

In this process, molecules evolve via (survival of the ﬁttest) so as to

optimize the scoring function. The main advantages of using this

protocol described here are: (1) Using the integrated environment

of Knime which has many chemistry components from different

vendors. In this article, for instance, we used Tripos chemistry

nodes; (2) Multiple-parameter optimization uses EA where penal-

izing the outranged values can be done by Gaussian demerits; (3)

Dissimilarity is evaluated in this workﬂow using the desired

ﬁngerprints. Original MACCS can be easily deployed using MOE

(Molecular operating environment software) chemistry nodes

where Quasar module has ﬁngerprint MACCS node; (4) A fast

algorithm which converges quickly. The workﬂow is depicted in

Fig. 8 where it is explained in details the experimental part.

The 6 ligands were used to generate virtual compounds such

that they comply with following criteria (i) molecular weight of

ꢁ40 Da; (ii) same number of rotational bonds, HBDs (hydrogen

bond donors), and HBAs (hydrogen bond acceptors); (iii) cLogP of

ꢁ1.0 and (iv) TPSA ꢁ15. The decoys were generated such that each

ligand has 36 decoys [31].In addition, zinc decoys (1000 ligands)

were used as standard decoys for other screening methods

comparison (retrieved from Sybyl folder). Both the virtual decoys

and the zinc decoys are supplied as structural data ﬁles in the

supplementary data.

where selectivity is the rate of false negatives. Therefore, ROC

analysis describes sensitivity (Se) as a function of (1 ꢀ Sp) where Sp

is the speciﬁcity [12,13,32].

Se ¼ ðnumber of selected activesÞ=ðtotal number of activesÞ

¼ TP=ðTP þ FNÞ

Sp ¼ ðnumber of discarded inactivesÞ

=ðtotal number of inactivesÞ ¼ TN=ðTN þ FPÞ

TP is the number of true positives, FN is the false negatives, TN is

the true negatives, while FP is the false positives.

Virtual screening protocol is said to have a good discrimination

power when most of the active molecules are scored higher than

inactive molecules. Actually, an overlap always takes place where

some of the actives are scored lower than the decoys. This overlap

will lead to the prediction of false positives and false negatives

[12,13]. Ratio between actives and decoys will vary according to the

score taken as a threshold. That is why we use ROC analysis because

it considers all Se and Sp pairs for each score threshold. The

protocol is very simple. The ROC curve sets each active score as

a threshold, counting the number of decoys within the cutoff and

calculating corresponding pair of Se and Sp.

An ideal ROC curve continues as a horizontal straight line to the

upper-right corner where all actives and all decoys are retrieved,

2.4. Validation

Table 10

Comparison of AUC of ROC analysis using pharmacophore alone (Hypo3), guided

docking alone and the combined workﬂow. This was done using either virtual or

zinc decoys.

In order to assess workﬂow performance in screening, it was

carried out using the following scenarios: (1) Using GOLD with

constraints alone; (2) Using structure-based pharmacophore alone;

(3) Using combined workﬂow by docking hits retrieved by the

pharmacophore.

Type of decoys

Pharmacophore

alone (Hypo3)

Docking with

restraints alone

Combined

workﬂow

Virtual decoys

Zinc decoys

0.884

0.851

0.810

0.916

0.947

0.922

Assessment was done using ROC analysis. ROC curve is a plot of

true positives (sensitivity) against false positives (1-selectivity)

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

477

Fig. 9. ROC analysis carried out using virtual decoys. (A) Comparison of ROC curve using pharmacophore (Hypo3) alone (ﬁt value) and combined workﬂow (Gold.PLP Fitness) using

the virtual decoy as dataset. (B) ROC curve using GOLD with constraints only.

which corresponds to Se ¼ 1 and 1 ꢀ Sp ¼ 0. In contrast to that, the

ROC curve for a set of actives and decoys with randomly distributed

scores tends toward the Se ¼ 1 ꢀ Sp line asymptotically with

increasing number of actives and decoys.

Area under the ROC curve (AUC)is normally evaluated [13]. In an

optimal ROC curve an AUC value of 1 is obtained; however, random

distributions cause an AUC value of 0.5. Virtual screening that

performs better than a random discrimination of actives and

decoys retrieve an AUC value between 0.5 and 1, whereas an AUC

value lower than 0.5 represents the unfavorable case of a virtual

screening method that has a higher probability to assign the best

scores to decoys than to actives [12,13,32].

ROC analysis was carried out using Accelrys Discovery Studio 3

and results are shown in Table 10. ROC curve of the reﬁned phar-

macophore (hypo3) using both zinc and virtual decoys shows that

the pharmacophore has a good discrimination power. The AUC

against virtual decoys was 0.884 and that against zinc was

0.851.This is shown in Fig. 9A and Fig. 10A respectively and repre-

sented by ﬁt value. The combined workﬂow shows an excellent

ROC curve with AUC of 0.947 using virtual decoys and 0.922 using

zinc (Figs. 9and 10A respectively and represented by GOLD.

PLP.Fitness). This was considered excellent for us as the model was

built using inhibitors of highly variable activities and was not

restricted on highly active ones. Thus, its capability to separate

decoys even from moderately active ligands with that superior ROC

was a very good result. The superiority of the workﬂow ROC relies

on the fact that it doesn’t allow missing of ideal solutions before

maximum no of iterations of GA are exceeded. This is done by

supplying the GA engine with initial good solutions. The ideality of

these solutions is guaranteed by utilizing the structure-based

pharmacophore which should succeed in obtaining hits in a good

binding pose. The GA function in that case will be restricted on

reﬁning the hit according to a correlated ﬁt value.

It was observed that GOLD with constrains performs well when

used alone (AUC of 0.916) in case when it was screened against zinc

database (Fig. 10B) so further analysis of the results was carried out

to assess the performance of using it alone without structure-based

pharmacophore.

Using Hypo3 to screen Zinc decoys, 429 hits were found. They

were all successfully docked after that using GOLD. On the other

hand, docking zinc decoys using gold alone retrieved only 391

ligands. This means that GOLD failed to dock some ligands which it

successfully did using pharmacophore as a pre-ﬁlter. The common

ligands which were retrieved by gold alone and the combined

workﬂow were 229 ligands. This means that GOLD missed to dock

200 ligands, which it has already succeeded to do using the

workﬂow. This proposes that GOLD may fail to dock ligands when

applying pharmacophore constraints if the ligands were not in

a good conformation. This may lead to the loss of ligands which

may show good binding. The problem can be solved by increasing

Fig. 10. ROC analysis carried out using Zinc decoys. (A) Comparison of ROC curve using pharmacophore (Hypo3) alone (ﬁt value) and the combined workﬂow (Gold.PLP Fitness)

using the virtual decoy as dataset. (B) ROC curve using GOLD with constraints only. Schemes captions.

478

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

Scheme 1. Reagents and conditions: (a) o-ﬂuorobenzyl bromide,NaH,DMF, 24 h; (b) (i) Diethyl malonate,NaH,DMA, 1 h; (ii) DMA, 120 ^ꢂC, 17 h; (c) (i)150 ^ꢂC, 15 min; (ii)KOH,reﬂux,

3 h; (iii) HCl, pH ¼ 4.

worth to mention is the presence of the privileged substructure of

substituted quinolinone in this hit as it is found in many inhibitors

Table 11

Biological evaluation of III and IX using both EC₅₀and CC₅₀

.

of benzothiadiazine class. This encouraged us to further assess it

biologically. Additionally, the ligand found by virtual screening may

support the latest discovery of quinoline derivatives lacking

sulfonamide moiety at Ring C [38].

a

b

Compound

CC₅₀

EC₅₀

Modiﬁed hit (III)

Optimized hit (IX)

>50

m

g/ml

23.7

2.6

m

g/ml

m

g/ml

a

CC₅₀¼ 50% cytotoxicity concentration.

EC₅₀¼ 50% effective concentration.

b

2.6. Synthesis and biological evaluation

number of iterations of the GA engine but this will increase the

time, taking into consideration that we selected the max default

value for iteration number which should be slow.

The hit found via virtual screening of Asinex database was sub-

jected to modiﬁcation to decrease number of steps required for its

synthesis. We preferred N-arylation of the quinolinone to that of N-

alkylation. This is because of the synthetic feasibility as will be

shown. Regarding the N-alkyl derivative, ethyl esters of 1-alkyl-2-

oxo-4 hydroxyquinoline-3-carboxylic acids are synthesized via

multistep reaction as reported by Ukraintes et al. [39]. First, ethyl

ester of anthranilic acid is alkylated by alkyl halide. Second, N-alkyl

ethyl anthranilate was acylated using ethoxymalonyl chloride. Third,

the anilide is treated with aqueous KOH. Fourth, potassium salt is

treated with acid. N-aryl derivatives, however, can be prepared using

readily available diethyl malonate (compared to the expensive

lachrymatory ethyl malonyl chloride) and isatioic anhydride in two

steps reaction (see supplementary data section 5 for details). The

ﬁnal compound (III) can be prepared readily by fusion of Ethyl 1-(2-

ﬂuorobenzyl)-4-hydroxy-2-oxo-1,2-dihydroquinoline-3-

2.5. Virtual screening

Asinex GOLD [36]was screened using the combined workﬂow.

The top 10% of the ranked database was kept. Ligands retrieved

were not purchased but assessed chemically using Sylvia score [37].

Ligands with score below 5 (Sylvia score is up to 10 where low

scores represent good synthetic feasibility) were further selected

and retro synthetically dissected to assess availability of the start-

ing materials. Based on the visual inspection, synthetic feasibility,

and non-reported activity, (BAS 00525932) was selected. The IUPAC

name of this compound is: 2-(1-Butyl-4-hydroxy-2-oxo-1,2-

dihydro-quinolin-3-yl)-1H-quinazolin-4-one. One thing which is

Scheme 2. Reagents and conditions : (a) Thionyl chloride, CH₃OH,reﬂux; (b)Methansulfonyl chloride, pyridine; (c) i-Lithium hydroxide, THF, CH₃OH, H₂0, 65 ^ꢂC, overnight; ii-HCl;

(d) i-Thionyl chloride,ii-NH₄OH; (e)Fe/NH₄Cl,CH₃OH, H₂0; (f) i-150 ^ꢂC, 15 min; ii-KOH,reﬂux,3h; iii-HCl, pH ¼ 4.

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

479

carboxylate (II) with anthranilamide [40].The general scheme for the

4. Experimental data

compound preparation is depicted in Scheme 1 where (II) was

synthesized from alkylated isatoic anhydride (I) after fusion with

diethyl malonate.

4.1. PLIF

Biological evaluation of the compound was performed on

genotype 1b using replicon system [41] and the result is shown in

Table 11. It was used as a lead compound for further optimization.

One suggestion which was carried out to optimize this ligand is the

substitution of 5-position with polar group like meth-

ansulfonamide which is known to form H-bond network in Palm I

pocket III of benzothiadiazines (Asn291 and Asp318) (see section 3

in supplementary data). This was done according to Scheme 2. It

depends on the fusion of 2-amino-5-(methylsulfonamido) benza-

mide (VIII) with Ethyl1-(2-ﬂuorobenzyl)-4-hydroxy-2-oxo-1,2-

dihydroquinoline-3-carboxylate (II). The ﬁrst was prepared

according to the literature method [42]starting with 5-amino-2-

nitrobenzoic acid. The synthesis of this starting material (VIII)

was based on protecting 5-amino-2-nitrobenzoic acid carboxylic

acid by esteriﬁcation to yield methyl 5-amino-2-nitrobenzoate (IV).

The ester was alkylated by methansulfonyl chloride to give methyl

5-(methylsulfonamido)-2-nitrobenzoate (V) which was further

deprotected to give 5-(methylsulfonamido)-2-nitrobenzoic acid

(VI). The carboxylic acid is then converted to amide 5-(methyl-

sulfonamido)-2-nitrobenzamide (VII) and reduced to give 2-

6 Co-crystallized HCV NS5b Palm I inhibitors were loaded and

prepared in MOE (Molecular operating environment software). The

complexes were superposed using the Biopolymer protein panel.

The ligX algorithm was applied to reﬁne interactions and 2D ligand

interaction diagrams were computed for the 6 complexes. A data-

base was created to include the complexes and PLIF was used to

compute the interactions and generate ﬁngerprints. After that,

a pharmacophore was constructed using query generation panel

according to the propensity of the features.

4.2. Custom features

Accelrys Discovery Studio 3 was used to build up a catalyst type

query. Acceptor feature was customized to have oxygen atom as the

only type of acceptors. In addition, vector and project were deleted.

Map atom tool was used so as to map oxygen atom. The feature was

added to catalyst dictionary in order to be used for screening.

4.3. Hypogen

amino-5-(methylsulfonamido)benzamide

compound (IX) potency increased 10 folds where it showed an EC₅₀

of 2.6 uM/mL.

(VIII).

The

ﬁnal

Accelrys Discovery Studio 3 3D-QSAR pharmacophore genera-

tion tool was used. Only two features were allowed: H-bond

acceptor (custom one) and hydrophobic feature. The variable

weight and tolerance options were set to true.

3. Conclusion

4.4. Contact-based pharmacophore

In this study, we presented a highly selective workﬂow of

structure-based pharmacophore and guided docking that can be

used for virtual screening of HCV NS5b polymerase inhibitors of

palm I allosteric site.

Accelrys Discovery Studio 3 was used to generate a receptor-

ligand pharmacophore. 3HKY was used as a crystal structure.

Number of pharmacophores was set to 10 such that minimum

features were 4 and maximum features were 6. The excluded

volume was extracted from the pharmacophore and clustered with

the Hypogen pharmacophore after adjusting its position in the

binding site. This was carried out as following: the pharmacophore

was mapped to 3HKY bound ligand in the binding site; similar

pharmacophore (will call it template pharmacophore) was con-

structed manually in the binding site using the ligand as

a template; pharmacophore comparison was performed using

template pharmacophore as a reference and Hypogen pharmaco-

phore as input. Finally, shape constraint was added using shape

query and minimum similarity tolerance was set to 0.3.

Regarding the pharmacophore, we presented a novel protocol to

generate a highly reﬁned pharmacophore using a combination of

different techniques: PLIF, Hypogen, shape constraints and contact-

based pharmacophore. The generated pharmacophore was able to

deﬁne the essential features required for selective activity which

are the 2 hydrogen bond acceptors and the three hydrophobic

features. It was used as a pre-ﬁlter to knowledge-based docking

providing it with good solutions as a starting point. In addition, it

proved to have very good capability in screening as an alternative to

docking. The encompassed PLIF analysis which was based on the

reﬁned complexes explained the effect of mutation of Tyr415 and

Gln446 on activity of inhibitors of this site.

Regarding docking, PLP proved to be a good scoring function that

correlates well with activity of the Palm I site. However, this can be

achieved when the pose evaluated is an accurate one. In order to

guarantee this accuracy of the ligand placement, the docking engine

was guided by pharmacophore during sampling process.

4.5. Gold

Gold 5 within Accelrys Discovery Studio 3 was deployed. Docking

was in 3HKYafter removing clashing waters. CHEMPLP was selected

as a scoring function. GA settings were adjusted to 100,000 opera-

tions. 2 regional hydrophobic features were used as constraints. 2

hydrogen bond acceptor features constraints were applied.

Allover, the workﬂow of structure-based pharmacophore fol-

lowed by knowledge-based docking (guided by both a correlating

scoring function and pharmacophore constraints) proved to be

superior in many aspects: the ROC analysis was excellent, the

number of hits missed was decreased, the convergence was very

quick and the accuracy of the predicted poses was very high.

The model was validated both statistically and experimentally.

Statistically, virtual decoys were employed and proved to be very

effective in ROC analysis study. Virtual decoys were generated using

a novel knime workﬂow based on Tripos muse which was given in

details. Experimentally, the model was used for screening of the

Asinex GOLD database, followed by synthesis of a modiﬁed hit and

its biological evaluation. Further knowledge-based optimization of

the hit based yielded a more potent inhibitor.

4.6. Muse

Tripos Muse 2 was used to generate 36 ligands as virtual decoy

set for each active ligand. The workﬂow in Fig. 8 is described. The

workﬂow starts by Node-A “MolstoScore” which represents those

molecules generated by Tripos muse according to EA. They are read

by SLN reader. They are designated by moltoscore to indicate that

they will be scored according to the speciﬁed criteria. Standardi-

zation of structures by ﬁlling valances and normalizing aromaticity

is done by Node-B. This is done by adding new column having

standardized molecules; therefore non-standard molecules are

480

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

ﬁltered using Node-C. The standardized molecules are evaluated by

calculating their molecular weight, number of hydrogen bond

acceptors and donors for the generated ligands using Node D,

number of rotatable bonds which is a substructure feature using

Node E and ﬁnally AlogP, total polar surface area and number of

Lipinski violations using ADME/TOX properties calculation in node

F. The last parameter (number of Lipinski violations) is calculated in

order to control ligands which are de-novo generated where those

violating the Lipinski rules can be ﬁltered using Lipinski ﬁlter in

node G. Regarding the reference ligand (active ligand for which we

want to create decoys), its properties should be entered using Muse

GUI where one can specify whether to allow decoys to strictly

comply with the ranges mentioned before (e.g. MWT of ꢁ40 Da) or

expand ranges if no solution can be provided. Controlling the

generated ligands to meet criteria of the decoys of the reference

ligand can be achieved using this workﬂow. The chemical dissim-

ilarity of the decoys (which can guarantee that they are not highly

similar to the actual reference ligand) is achieved by ﬁltering

compounds having Tanimoto coefﬁcients of less than 0.9. This is

carried out by standardizing the reference ligand just like the

generated ligands were standardized in Nodes (H, I, J). After that,

Nodes (KeL) are used to calculate unity ﬁngerprint for both refer-

ence and the generated ligands where Tanimoto similarity can be

calculated between them using Node M and allowing only ligands

with Tanimoto coefﬁcients less than 0.9 to be kept using node N.

Regarding the scoring of the generated ligands, it depends on the

input we provide in muse which is divided into a range and toler-

ance of each property to be controlled. This is enough to calculate

the score for each molecule using Gaussian normalizer for each

number and summing up all normalized scores at the end. For

instance, if we set the number of hydrogen bond donors to have

range 3e4 with tolerance 1 and weight 0.5, the number will be

utilized in the scoring function as follows: If the generated mole-

cule has number of hydrogen bond donors between 3 and 4, the

score will be zero (zero is the ideal score here) On the other hand, if

the number of hydrogen bond donors is less than 3 or greater than

4,score will be assigned according to the Gaussian bell curve which

has spread (S) of 1 (speciﬁed by tolerance) and asymptotically

approach value ꢀ1 (which is the lowest value that can be given for

this criterion)(see supplementary data section 4.1 for the curve).

This can also be done using Tripos score with avoidance in Tripos

muse and this is given in details in section 4.2 supplementary data.

otherwise noted. Solvents and reagents were used without further

puriﬁcation. Reactions were monitored by TLC, performed on silica

gel glass plates containing F-254 indicator (Merck). Visualization on

TLC was achieved by UV. Proton and carbon NMR spectra were

recorded on a Bruker ARX-300. Chemical shifts (ä) are reported in

ppm downﬁeld from internal TMS standard or from solvent refer-

ences. Mass spectra were recorded on API-SCIEX 2000. HRMS were

recorded using an Agilent MSD-TOF (G1969A) connected to an

Agilent 1100 HPLC system. Melting points were determined on

a Stuart melting point apparatus (Stuart Scientiﬁc, Redhill, UK) and

are uncorrected.

4.9.1. 1-(2-Fluoro-benzyl)-1H-benzo[d] [1,3]oxazine-2,4-dione (I)

It was prepared according to the literature [43]. To a solution of

isatoic anhydride (16.3 g, 10 mmol) in 200 ml of DMF, sodium

hydride (4.6 g, 110 mmol, 57% in mineral oil, washed with hexane)

was slowly added and the mixture was stirred for 1 h at room

temperature. 2-ﬂuorobenzyl bromide (20.79 g, 110 mmol) was

added and the reaction was allowed to stir for 18 h at room

temperature. About two thirds of the solvent were then evaporated

in vacuo and the residue poured into 250 mL of ice/water. A

precipitate formed which was ﬁltered off, washed with water and

dried. It was recrystallized from DCM. Yield: 60%; mp 151e154 ^ꢂC as

reported [43].

4.9.2. Ethyl 1-(2-ﬂuorobenzyl)-4-hydroxy-2-oxo-1,2-dihydroquino-

line -3- carboxylate (II)

Neat diethyl malonate (1.2 mL, 8.0 mmol) was added slowly to

a suspension of sodium hydride (60 percent in mineral oil, 352 mg,

8.8 mmol) in dimethylacetamide. The mixture was stirred at room

temperature until the evolution of hydrogen gas ceased, then the

mixture was heated to 90 ^ꢂC for 30 min and cooled to room

temperature. A solution of compound 1-(2-ﬂuoro-benzyl)-1H-

benzo[d] [1,3]oxazine-2,4-dione (8.8 mmol) in dimethylacetamide

was added slowly and heated overnight at 110 ^ꢂC. The mixture was

cooled to room temperature, poured into ice water, and acidiﬁed by

cold 10 percent HCl. The solids formed were ﬁltered, washed

several times by water, and dried at room temperature under

vacuum. Yield: 80%; ¹H NMR (300 MHz, DMSO)

1.4, 1H), 7.70e7.09 (m, 6H), 5.43 (s, 2H), 4.39 (q, J ¼ 7.5, 2H), 1.31 (t,

J ¼ 7, 3H). HRMS-FAB m/z [M^þ] calcd for C₁₉H₁₆FNO₄: 341.1063;

found: 341.1060.

d

¼ 8.11 (dd, J ¼ 7.3,

4.7. ROC analysis

4.9.3. 2-[1-(2-Fluorobenzyl)-4-hydroxy-2-oxo-1,2-dihydroquinolin-

3-yl] quinazolin-4(1H)-one (III)

ROC analysis was carried out in Discovery Studio. Scores were

set to ﬁt value in pharmacophore search while GOLD. PLP ﬁtness

was used in docking search and in the workﬂow. Pose grouping was

adjusted to name in docking analysis. Actives were given a control

value of 1. Because pharmacophore always ﬁlters the ligands, the

number of total records was used to avoid false results.

Synthesis was carried out using general procedure by

Ukrainets [40].

The mixture of 0.01 mol of the ethyl 1-(2-ﬂuorobenzyl)-4-

hydroxy-2-oxo-1,2-dihydroquinoline-3-carboxylate and 1.36

g

(0.01 mol) of anthranilamide is maintained at 150 ^ꢂC for 10e15 min.

The mixture is cooled prior to the addition of 20 ml of alcohol and

the careful trituration. The residue of the 2-carbamylanilide is

ﬁltered off, washed with alcohol, and dried. It is recrystallized

from DMF.

4.8. Virtual screening

Asinex GOLD was converted into catalyst database using fast

conformation method. Database was screened using ligand phar-

macophore mapping tool and not search 3d database because the

ﬁrst allows adjusting number of missing features. The missing

features were allowed to be one while choosing the best hit as the

one which ﬁts most features.

2-Carbamylanilide (0.01 mol) was mixed with 30 ml of the 10%

aqueous solution of KOH and boiled using a reﬂux condenser for

2.5e3 h. The mixture is cooled and acidiﬁed with HC1 to the pH 4.

The residue of the quinazolone (I), which was separated out, is

ﬁltered off, washed with water, dried, and recrystallized from

DMF.Yield:90%.¹H NMR (300 MHz, d₆DMSO)

d

¼ 8.13 (dd, 7.6, 1.6,

2H), 7.64e7.08 (m, 10H), 5.45 (s, 2H). ¹³C NMR (75 MHz, d₆DMSO)

4.9. Synthesis

d

¼ 165.34, 164.1, 163.12, 159.76, 158.22, 150.88, 138.28, 136.43,

133.50, 131.68, 129.46, 126.94e125.92 (m), 124.84, 123.19, 118.24,

117.21, 115.86, 115.52, 113.52, 104.48, 47.67. HRMS-FAB m/z [M^þ]

calcd for C₂₄H₁₆FN₃O₃: 413.1176; found: 413.1171.

General Methods. Starting materials were either commercially

available or prepared as reported previously in the literature, unless

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

481

4.9.4. 2-Amino-5-(methylsulfonamido) benzamide (VIII)

Appendix A. Supplementary material

It was synthesized according to Chin et al. [42].

Supplementary data associated with this article can be found, in

the online version, at doi:10.1016/j.ejmech.2012.04.016.

4.9.5. N-(2-(1-(2-ﬂuorobenzyl)-4-hydroxy-2-oxo-1,2-

dihydroquinolin-3-yl)-4-oxo-1,4-dihydroquinazolin-6-yl)

methansulfonamide (IX)

References

Synthesis was carried out using the general procedure applied

by Ukrainets [40] just as the previous compound. Yield: 45%,

HRMS-FAB m/z [M^þ] calcd for C₂₅H₁₉FN₄O₅S: 506.1060, found: 506,

[1] D. Lavanchy, The global burden of hepatitis C, Liver International 29 (Suppl. 1)

(2009) 74e81.

[2] Q.L. Choo, G. Kuo, A.J. Weiner, L.R. Overby, D.W. Bradley, M. Houghton,

Isolation of a cDNA clone derived from a blood-borne non-A, non-B viral

hepatitis genome, Science 244 (1989) 359e362.

[3] F. Penin, J. Dubuisson, F.A. Rey, D. Moradpour, J.M. Pawlotsky, Structural

biology of hepatitis C virus, Hepatology 39 (2004) 5e19.

[4] R.R. Deore, J.W. Chern, NS5B RNA dependent RNA polymerase inhibitors: the

promising approach to treat hepatitis C virus infections, Current Medicinal

Chemistry 17 (2010) 3806e3826.

[5] F. Legrand-Abravanel, F. Nicot, J. Izopet, New NS5B polymerase inhibitors for

hepatitis C, Expert Opinion on Investigational Drugs 19 (2010) 963e975.

[6] H. Li, S.T. Shi, Non-nucleoside inhibitors of hepatitis C virus polymerase:

current progress and future challenges, Future Medicinal Chemistry 2 (2010)

121e141.

1054. ¹H NMR (300 MHz, DMSO-d₆)

d

¼ 8.11 (dd, J ¼ 7.4, 1.6, 1H),

7.70e7.57 (m, 2H), 7.56e7.44 (m, 2H), 7.38 (m, 2H), 7.23 (m, 1H),

7.15e6.99 (m, 3H), 6.44 (s, 1H), 5.77 (s, 1H), 5.31 (s, 1H). ¹³C NMR

(75 MHz, DMSO-d₆)

d

¼ 165.43, 163.81, 163.02, 159.69, 158.25,

150.76, 140.48, 136.45, 134.65, 131.69, 129.41, 126.94e126.40 (m),

124.83, 123.17, 119.45, 118.35, 117.27, 115.86, 115.52, 115.14, 113.54,

104.47, 47.67, 42.90.

4.10. Biology

[7] W.J. Watkins, A.S. Ray, L.S. Chong, HCV NS5B polymerase inhibitors, Current

Opinion in Drug Discovery and Development 13 (2010) 441e465.

[8] T.T. Talele, Multiple allosteric pockets of HCV NS5B polymerase and its

inhibitors: a structure-based insight, Current Bioactive Compound 4 (2008)

86e109.

[9] M.L. Barreca, N. Iraci, G. Manfroni, V. Cecchetti, Allosteric inhibition of the

hepatitis C virus NS5B polymerase: in silico strategies for drug discovery and

development, Future Medicinal Chemistry 3 (2011) 1027e1055.

[10] T. Li, M. Froeyen, P. Herdewijn, Insight into ligand selectivity in HCV NS5B

polymerase: molecular dynamics simulations, free energy decomposition and

docking, Journal of Molecular Modeling 16 (2010) 49e59.

[11] I. Musmuca, A. Caroli, A. Mai, N. Kaushik-Basu, P. Arora, R. Ragno, Combining

3-D quantitative structure-activity relationship with ligand based and

structure based alignment procedures for in silico screening of new hepatitis

C virus NS5B polymerase inhibitors, Journal of Chemical Information and

Modelling 50 (2010) 662e676.

[12] J. Kirchmair, P. Markt, S. Distinto, G. Wolber, T. Langer, Evaluation of the

performance of 3D virtual screening protocols: RMSD comparisons,

enrichment assessments, and decoy selectionewhat can we learn from earlier

mistakes? Journal of Computer-Aided Molecular Design 22 (2008) 213e228.

[13] N. Triballeau, F. Acher, I. Brabet, J.P. Pin, H.O. Bertrand, Virtual screening

workﬂow development guided by the "receiver operating characteristic"

curve approach. Application to high-throughput docking on metabotropic

glutamate receptor subtype 4, Journal of Medicinal Chemistry 48 (2005)

2534e2547.

One day before addition of test compounds, human hepatoma

cells (Huh5.2) containing the hepatitis C virus genotype 1b I389luc-

ubi-neo/NS3-3⁰/5.1 replicon sub-cultured in Dulbecco’s modiﬁed

eagle’s medium [DMEM (Cat. N^ꢂ41965039) supplemented with 10%

fetal calf serum (FCS), 1% non-essential amino acids (11140035), 1%

penicillin/streptomycin (15140148) and 2% Geneticin (10131027);

Invitrogen]. Cell lines were grown for 3e4 days in 75 cm²tissue

culture ﬂasks (Techno Plastic Products), were harvested and seeded

in assay medium (DMEM, 10% FCS, 1% non-essential amino acids, 1%

penicillin/streptomycin) at a density of 6500 cells/well (100ml/well)

in 96-well tissue culture microtiter plates (Falcon, Beckton Dick-

inson for evaluation of anti-metabolic effect and CulturPlate, Perkin

Elmer for evaluation of antiviral effect). The microtiter plates were

incubated overnight (37 ^ꢂC, 5% CO2, 95e99% relative humidity),

yielding a non-conﬂuent cell monolayer. The evaluation of the anti-

metabolic as well as the antiviral effect of each compound is per-

formed in parallel. Four-step, 1-to-5 test compound dilution series

are prepared. Following assay setup, the microtiter plates are

incubated for 72 h (37 ^ꢂC, 5% CO2, 95e99% relative humidity).

[14] X. Fradera, J. Mestres, Guided docking approaches to structure-based design

and screening, Current Topics in Medicinal Chemistry 4 (2004) 687e700.

[15] G. Jones, P. Willett, R.C. Glen, A.R. Leach, R. Taylor, Development and valida-

tion of a genetic algorithm for ﬂexible docking, Journal of Molecular Biology

267 (1997) 727e748.

[16] R. Thomsen, M.H. Christensen, MolDock: a new technique for high-accuracy

molecular docking, Journal of Medicinal Chemistry 49 (2006) 3315e3321.

[17] T. Sato, T. Honma, S. Yokoyama, Combining machine learning and

pharmacophore-based interaction ﬁngerprint for in silico screening, Journal of

Chemical Information and Modeling 50 (2009) 170e185.

4.10.1. Assay protocols

(i) Luciferase assay

For the evaluation of antiviral effects, Viral RNA replication was

determined using Renilla luciferase assays (the reporter gene

luciferase from the ﬁreﬂy (Photinus pyralis) and the coding

sequence for ubiquitin were inserted upstream of the neo gene by

using standard recombinant DNA technologies). Assay medium is

aspirated and the cell monolayers are washed with PBS. The wash

[18] J.S. Duca, Recent advances on structure-informed drug discovery of cyclin-

dependent kinase-2 inhibitors, Future Medicinal Chemistry

1 (2009)

1453e1466.

[19] [In: IUL Biotechnol. Ser., 2000; 2]O.F. Guener (Ed.), Pharmacophore: Percep-

tion, Development, and Use in Drug Design, International University Line,

2000.

[20] G. Wolber, T. Langer, LigandScout: 3-d pharmacophores derived from

protein-bound ligands and their use as virtual screening ﬁlters, Journal of

Chemical Information and Modeling 45 (2005) 160e169.

[21] I. Wallach, R. Lilien, Virtual decoy sets for molecular docking benchmarks,

Journal of Chemical Information and Modeling 51 (2011) 196e202.

[22] H.M. Berman, J. Westbrook, Z. Feng, G. Gilliland, T.N. Bhat, H. Weissig,

I.N. Shindyalov, P.E. Bourne, The protein data bank, Nucleic Acids Research 28

(2000) 235e242.

buffer is aspirated, 25

ega) is added after which lysis is allowed to proceed for 5 min at

m

l of Glo Lysis Buffer (Cat. N^ꢂ. E2661, Prom-

room temperature. Subsequently, 50 ml of Luciferase Assay System

(Cat. N^ꢂ. E1501, Promega) is added and the luciferase luminescence

signal is quantiﬁed immediately (1000 ms integration time/well,

Saﬁre², Tecan). Relative luminescence units are converted to

percentage of untreated controls.

[23] F. Pauwels, W. Mostmans, L.M. Quirynen, L. van der Helm, C.W. Boutton,

A.S. Rueff, E. Cleiren, P. Raboisson, D. Surleraux, O. Nyanguile, K.A. Simmen,

ii) Viability assay

Binding-site identiﬁcation and genotypic proﬁling of hepatitis

polymerase inhibitors, Journal of Virology 81 (2007) 6909e6919.

C virus

[24] O. Nyanguile, B. Devogelaere, L. Vijgen, W. Van den Broeck, F. Pauwels,

M.D. Cummings, H.L. De Bondt, A.M. Vos, J.M. Berke, O. Lenz,

G. Vandercruyssen, K. Vermeiren, W. Mostmans, P. Dehertogh, F. Delouvroy,

S. Vendeville, K. VanDyck, K. Dockx, E. Cleiren, P. Raboisson, K.A. Simmen,

G.C. Fanning, 1a/1b Subtype proﬁling of nonnucleoside polymerase inhibitors

of hepatitis C virus, Journal of Virology 84 (2010) 2923e2934.

For the evaluation of anti-metabolic effects, the assay medium is

aspirated, replaced with 75 ml of a 5% MTS (Promega) solution in

phenol red-free medium and incubated for 1.5 h (37 ^ꢂC, 5% CO2,

95e99% relative humidity). Absorbance is measured at a wave-

length of 498 nm (Saﬁre², Tecan) and optical densities (OD values)

are converted to percentage of untreated controls.

[25] R. Tedesco, A.N. Shaw, R. Bambal, D. Chai, N.O. Concha, M.G. Darcy, D. Dhanak,

D.M. Fitch, A. Gates, W.G. Gerhardt, D.L. Halegoua, C. Han, G.A. Hofmann,

482

A.H. Mahmoud et al. / European Journal of Medicinal Chemistry 57 (2012) 468e482

V.K. Johnston, A.C. Kaura, N. Liu, R.M. Keenan, J. Lin-Goerke, R.T. Sarisky,

[35] M. Feher, Y. Gao, J.C. Baber, W.A. Shirley, J. Saunders, The use of ligand-based

de novo design for scaffold hopping and sidechain optimization: two case

studies, Bioorganic & Medicinal Chemistry 16 (2008) 422e427.

[36] http://www.asinex.com/Libraries_Gold_Platinum.html.

[37] http://www.molecular-networks.com/products/sylvia.

K.J. Wiggall, M.N. Zimmerman, K.J. Duffy, 3-(1,1-Dioxo-2H-(1,2,4)-benzothia-

diazin-3-yl)-4-hydroxy-2(1H)-quinolinones, potent inhibitors of hepatitis C

virus RNA-Dependent RNA polymerase, Journal of Medicinal Chemistry 49

(2006) 971e983.

[26] S. Boyd, LigandScout, Chemistry World 3 (2006) 69e70.

[27] X. Wang, W. Yang, X. Xu, H. Zhang, Y. Li, Y. Wang, Studies of benzothiadiazine

derivatives as hepatitis C virus NS5B polymerase inhibitors using 3D-QSAR,

molecular docking and molecular dynamics, Current Medicinal Chemistry 17

(2010) 2788e2803.

[28] O. Korb, T. Stützle, T.E. Exner, Empirical scoring functions for advanced pro-

teinꢀligand docking with PLANTS, Journal of Chemical Information and

Modeling 49 (2009) 84e96.

[38] D.K. Hutchinson, C.A. Flentge, P.L. Donner, R. Wagner, C.J. Maring, W.M. Kati,

Y. Liu, S.V. Masse, T. Middleton, H. Mo, D. Montgomery, W.W. Jiang, G. Koev,

D.W.A. Beno, K.D. Stewart, V.S. Stoll, A. Molla, D.J. Kempf, Hepatitis C NS5B

polymerase inhibitors: functional equivalents for the benzothiadiazine

moiety, Bioorganic & Medicinal Chemistry Letters 21 (2011) 1876e1879.

[39] L.V. Ukrainets, O.V. Gorokhova, S.G. Taran, P.A. Bezulyi, A.V. Turov,

N.A. Marusenko, O.A. Evtifeeva, 4-hydroxy-2-quinolones. 22.Synthesis and

biological

properties

of

1-alkyl(aryl)-2-oxo-3-carbethoxy-4-

[29] R.A. Friesner, R.B. Murphy, M.P. Repasky, L.L. Frye, J.R. Greenwood,

T.A. Halgren, P.C. Sanschagrin, D.T. Mainz, Extra precision glide: docking and

scoring incorporating a model of hydrophobic enclosure for proteinꢀligand

complexes, Journal of Medicinal Chemistry 49 (2006) 6177e6196.

[30] N. Huang, B.K. Shoichet, J.J. Irwin, Benchmarking sets for molecular docking,

Journal of Medicinal Chemistry 49 (2006) 6789e6801.

[31] J.J. Irwin, Community benchmarks for virtual screening, Journal of

Computer-Aided Molecular Design 22 (2008) 193e199.

[32] J.J. Irwin, B.K. Shoichet, ZINCea free database of commercially available

compounds for virtual screening, Journal of Chemical Information and

Modelling 45 (2005) 177e182.

hydroxyquinolines and their derivatives, Chemistry of Heterocyclic

Compounds 30 (1994) 829e836.

[40] I.V. Ukrainets, S.G. Taran, P.A. Bezuglyi, S.N. Kovalenko, A.V. Turov,

N.A. Marusenko, 4-hydroxy-2-quinolones. 18. Synthesis and antithyroid

activity of 1-R-2-oxo-3-(4-oxo-3H-quinazolin-2-yl)-4-hydroxyquinolines,

Chemistry of Heterocyclic Compounds 29 (1993) 1044e1047.

[41] G. Yang, M. Huang, Evaluation of compound activity against hepatitis C virus

in replicon systems, in: Current Protocols in Pharmacology, John Wiley &

Sons, Inc., 2001.

[42] E. Chin, J. Li, A.S.-T. Lui, F.X. Talamas (Eds.), Preparation of Pyrimidine

Derivatives as Antiviral Agents, F. Hoffmann-La Roche AG, Switz, 2010, p. 75.

[43] G.E. Hardtmann, G. Koletar, O.R. Pﬁster, The chemistry of 2H-3,1-benzoxazine-

2,4(1H)dione (isatoic anhydrides) 1. The synthesis of N-substituted 2H-

3,1-benzoxazine-2,4(1H)diones, Journal of Heterocyclic Chemistry 12 (1975)

565e572.

[33] F. Fontaine, M. Pastor, F. Sanz, Incorporating molecular shape into the

alignment-free grid-independent descriptors, Journal of Medicinal Chemistry

47 (2004) 2805e2815.

[34] http://www.knime.org/.

Article Doi

A highly selective structure-based virtual screening model of Palm i allosteric inhibitors of HCV Ns5b polymerase enzyme and its application in the discovery and optimization of new analogues

DOI: 10.1016/j.ejmech.2012.04.016

Source and publish data:

Authors:

Article abstract of DOI:10.1016/j.ejmech.2012.04.016

Full text of DOI:10.1016/j.ejmech.2012.04.016

Products guided by the article

R&D Labs maybe for 1256489-03-8

Relevant to this article

Hot Product