Integrated structure-based activity prediction model of benzothiadiazines on various genotypes of HCV NS5b polymerase (1a, 1b and 4) and its application in the discovery of new derivatives

Bioorganic & Medicinal Chemistry 20 (2012) 2455–2478

Contents lists available at SciVerse ScienceDirect

Bioorganic & Medicinal Chemistry

journal homepage: w ww.elsevier.com/locate/bmc

Integrated structure-based activity prediction model of benzothiadiazines

on various genotypes of HCV NS5b polymerase (1a, 1b and 4)

and its application in the discovery of new derivatives

⇑

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

Department of Pharmaceutical Chemistry, Faculty of Pharmacy, Ain Shams University, Cairo 11566, Egypt

a r t i c l e i n f o

a b s t r a c t

Article history:

This work presents the ﬁrst structure-based activity prediction model for benzothiadiazines against

various genotypes of HCV NS5b polymerase (1a, 1b and 4).The model is a comprehensive workﬂow of

structure-based ﬁeld template followed by guided docking. The ﬁeld template was used as a pre-ﬁlter

and a tool to provide hits in good orientation and position. It was created based on detailed molecular

interaction ﬁeld analysis which includes Topomer CoMFA, grid independent analysis and Superstar. On

the other hand, Guided docking was used as a reﬁnement and assessment tool. It was actively directed

by two scores: Moldock score as an interaction descriptor (r²= 0.65) and a template similarity score as

a measure for accurate binding-mode compliance. The docking template was based on energy-based

pharmacophore analysis. The whole procedure was formulated and tweaked for both screening (ROC

of AUC = 0.91) and activity prediction (r²of 0.8) for the genotype 1a. In order to widen the model scope,

linear interaction energy was used as a tool for predicting activities of other genotypes based on the

docked ligand poses while mutation binding energy was used to investigate the effect of each amino acid

mutation in genotype 4. The model was applied for structure-based fragment hopping by screening a

library designed by reaction enumeration. A top scoring hit was used to generate a focused library such

that it has lower TPSA than the original class ligands and thus better pharmacokinetic properties. After

that, experimental validation was carried out by the synthesis of this library and its biological evaluation

Received 21 November 2011

Revised 25 December 2011

Accepted 5 January 2012

Available online 28 January 2012

Keywords:

HCV NS5b polymerase inhibitor

Topomer CoMFA

AMANDA

Guided docking

Homology

Library design

Field alignment

Fragment-hopping

which yielded compounds that exhibit EC₅₀ranging from 1.86 to 23 lM.

1. Introduction

RNA-dependent RNA polymerase is recognized as a key target for

therapeutic intervention mainly because it is not present in mam-

malian cells and offers a wide range of possibilities for the discov-

ery of new molecular entities as anti-HCV agents.^8–11Mechanistic

and structural studies of this enzyme have revealed the existence

of multiple allosteric binding sites, and in particular two thumb

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

have been identiﬁed to date.¹²According to the target site, the dif-

ferent 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). Out of

these different allosteric sites and their corresponding inhibitors,

we focused this study on palm I site and in particular on benzothi-

adiazines as one of the main palm I-NNI¹³(non-nucleoside inhibi-

tors) (see Figs. S1 and S2 in the Supplementary data and see full

account on the different allosteric sites in Section 1.1 HCV general

information).

The hepatitis C virus (HCV) is a member of the Flaviviridae

family. Chronic infection with this virus is associated with liver cir-

rhosis that often leads to hepatic failure and hepatocellular carci-

noma. Although the number of new infections has been

signiﬁcantly reduced by the introduction of reliable blood testing,

more than 170 million people worldwide are chronically infected

with HCV, which has become a global health threat and the main

cause of adult liver transplants in developed nations. There is no

effective therapy for HCV-associated chronic hepatitis up till

now. As a result, hepatitis C is considered a major public health

threat and there is a growing unmet medical need to discover no-

vel therapies.^1,2Employing HCV proteins as targets, directly acting

antiviral agents have been identiﬁed and collectively described as

‘speciﬁcally targeted antiviral therapy for HCV⁰(STAT-C).^3–5Among

the nonstructural proteins, NS3–4A protease, NS5B polymerase,

NS3 helicase and NS5A have been the object of intense research ef-

forts both by academia and pharmaceutical companies.^6,7NS5B

The main aims of this study can be outlined in these points:

1. The development for the ﬁrst time a structure-based activity

prediction model for the benzothiadiazine class which has

wider applicability domain¹⁴than that of the ligand based ones.

⇑

Corresponding author. Tel.: +20 224051120; fax: +20 24051107.

E-mail address: mhismaeel@yahoo.com (M.A.H. Ismail).

doi:10.1016/j.bmc.2012.01.031

2456

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

Table 1

on active inhibitors similarity. However, we did not want to leave

it unguided such that it may result in a plethora of irrelevant hits

that do not have the essential features required for activity. Thus,

we decided to support this model by knowledge retrieved from

literature concerning known inhibitors of this class. In another

words, we decided to generate a guided receptor-based activity

prediction model.

Mini review on the CADD attempts for benzothiadiazines and benzothiazines

Hendricks

et al.

Ab initio calculations were employed for conformational

analysis of a series of compounds^61,62The importance of

internal hydrogen bonding was highlighted as the main

method to achieve NS5b-bound conformations

Melagraki et al. built MLR QSAR model using lipophilicity,

HOMO energy, Ki2, KiInf0 and KiInf3 as descriptors.¹⁵This

was carried out for Substituted hydroxyquinolone-

benzothiadiazine and substituted naphthyridone-

benzothiadiazine derivatives

Melagraki

et al.

2. Design process

Pourbasheer

et al.

Pourbasheer et al.¹⁶built MLR and SVM models using seven

descriptors (hydration energy, PW4, GATS2m, GATS7m,

GATS5p, RDF095e and Mor30m) to study pyridazinone-

benzothiadiazine

Being a workﬂow, we decided to describe it all over ﬁrst then

give the results of each part. The structure-based activity predic-

tion model presented in this study consists of two elements:

structure-based ﬁeld template²³followed by guided docking.²⁴

The ﬁeld template (ﬁrst main component of this workﬂow) con-

sists of ﬁeld points which describe the van der Waals and elec-

trostatic (both positive and negative) minima and maxima²⁵that

surround molecules. It is considered a structure-based tool as it

takes the bound conformer as a template to construct ﬁeld

points while considering the protein as an excluded volume. In

addition, It consists of hotspots which represent nodes or ﬁeld

points of high relevance in ligand–protein interaction. These hot-

spots are treated in the ﬁeld template as constraints and are ex-

tracted using molecular interaction ﬁeld analysis as will be

described later. The ﬁeld template is used as a pharmacophore

pre-ﬁlter and a tool to align ligands in the binding site and pro-

viding them in a good orientation before docking. The second

main component of the workﬂow is the guided docking module.

It is a type of docking which incorporates chemical information

to actively guide the orientation of the ligand into the binding

site during the ligand sampling. It is used as an efﬁcient knowl-

edge-based strategy for binding afﬁnity estimation, ligand bind-

ing-mode prediction, virtual screening enrichment and further

ligand optimization.²⁴Herein, we used Moldock that is based

on evolutionary algorithm (EA)²⁶in sampling that is guided by

chemical information represented in the form of pharmacophore

template that reﬂects the essential features for interaction. More

speciﬁcally, it is directed by a consensus scoring function which

is divided into two:

Chen et al.

Wang et al.

Chen et al.¹⁷carried out both 2D-QSAR and 3D-QSAR using

MFA coupled with G/PLS to study Benzoisothiazoles dioxide,

benzothiophenes dioxide

Wang et al.¹⁸developed ligand-based and receptor-based

CoMFA and CoMSIA models for large number of

benzothiadiazines. Homology model was developed for

genotype 1a in this study. It was used to obtain receptor-

based alignment for CoMFA analysis via automated docking

procedure. Besides, binding mode was studied via Molecular

dynamics experiment

Li et al.

Li et al.²⁰used MD-simulation to study binding mode of

benzothiadiazines and other palm I classes. MM-GB/SA

calculations together with free energy decomposition were

deployed in this study

This wider scope is attributed to the usage of the knowledge of

the three dimensional structure of the biological target

obtained through x-ray crystallography. This model is

considered a receptor-based QSAR model which was mainly

inﬂuenced by the fact that previous prediction models by Mel-

agraki et al.¹⁵Pourbasheer et al.¹⁶Chen et al.¹⁷and Wang

et al¹⁸were ligand-based (Table 1) with limited applicability

domain.¹⁹They were likely able to predict only the activities

of benzothiadiazines with similar properties to that of the

training dataset. Additionally we realized that all the previous

structure-based models were used to predict binding mode

only.^18,20Applicability domain method deployed is explained

in Section 1.7 in the Supplementary data.

1. A score for ligand–protein interaction that correlates well with

activity. This score is used to optimize the ligand in the binding

site according to the interactions.

2. A template similarity score which reﬂects the compliance of the

inhibitor with the proper binding-mode (i.e., mapping with the

pharmacophore properly). This score applies similarity correc-

tions throughout the entire ligand incremental construction

process, thus affecting not only the relative ordering of the solu-

tions but also actively guiding the ligand placement such it

complies with the main reference of ligand–protein complex.

2. Extending the applicability of the model by being able to pre-

dict the activities of various genotypes. This is because the palm

I site is the least conserved among the different allosteric bind-

ing sites of HCV polymerase²¹which affects the activity of the

benzothiadiazine class leading to a drastic decrease in potency

against genotypes other than genotype 1. The effect is even

obvious in the subtypes of genotype 1 itself (10-fold decrease

in genotype 1a than 1b). Thus, we wanted to increase the scope

of this model to be able to predict the activities of various geno-

types. Here we included three genotypes: 1a,1b and 4.

3. Applying the model in virtual screening and identiﬁcation of

new derivatives of this speciﬁc class of benzothiadiazines which

may have better pharmacokinetic properties especially the

topological polar surface area (TPSA). This is mainly due to

the fact that this class suffers from poor bioavailability due to

the large polarity of the ligands that hinders the permeability.²²

Thus, the requirements of the guided docking are a pharmaco-

phore template and a scoring function which correlates well with

the activity. These requirements were extracted according to inno-

vative procedures applied in the workﬂow which is described

below.

Systematically, we constructed the aforementioned model and

applied it using an innovative workﬂow which is depicted in

Figures 1 and 2 and consists of the following steps:

The most important criterion of this work was the desire to ﬁnd

hits that avoid the common chemotype trap in ﬁnding benzothia-

dinzes derivatives. By this, we are pointing to the fact that most

patents depend on simple markush structure similarity. This is

very obvious where nearly most of the patents focused on systems

having in common a keto-enol system in the ring fused with the

benzothiadiazine (e.g. hydroxyquinolinone system). Thus, we

shifted to structure-based virtual screening that does not depend

1. Detailed SAR analysis was done using molecular interaction

ﬁeld analysis methods, molecular dynamics and energy-based

pharmacophore²⁷to extract hotspots and pharmacophore

features. As shown in Figure 1, SAR analysis had three roles.

It was used to construct structure-based ﬁeld template

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2457

Figure 1. The workﬂow of the construction and the application of the structure-based activity prediction model for the benzothiadiazine class.

Figure 2. The workﬂow used to construct and validate the structure-based activity prediction model of benzothiadiazines.

2458

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

(pharmacophore pre-ﬁlter), guided docking pharmacophore

template and aided in the designing of the virtual library which

was later screened.

compounds in order to characterize them according to their poten-

tial to act as ligands, binding a certain receptor. When used in this

manner, the MIF can be seen as computationally obtained descrip-

tor variables (‘molecular descriptors’), which represent properties

of the molecules (e.g., AMANDA³³) or contour maps (e.g., Topomer

CoMFA³⁴). Both AMANDA and Topomer CoMFA were carried out in

the ligand-based methods. The details of these analyses are listed

below:

3.1.1.1.1. Ligand based methods. The methods used here are

QSAR methods which deals mainly with the variations among

structures (it does not describe constant part) thus they were re-

stricted on two functionalities: studying SAR of the variable sub-

stituents and developing virtual screening QSAR models.

Topomer CoMFA and AMANDA analyses were deployed. These

are two different ﬁeld based techniques; the ﬁrst uses the C.sp3

and H+ probes to represent steric and electrostatic ﬁelds while

the second use different types of probes (O, N1, probes) which re-

ﬂect H-bond donor, H-bond acceptor and hydrophobic ﬁelds. In

this way, the information obtained from the two methods can be

complementary and not redundant as it covers different types of

ﬁelds. The two MIF analysis techniques are non-alignment based

and the combination of them, as far as we know, has not been pub-

licly utilized in literature unlike COMFA and COMSIA. These tools

were used to avoid alignment step which is the time consuming

step of the classic MIF analysis.

2. Construction of the structure-based activity prediction model

according to the following steps:

a. Generating a structure-based ﬁeld template using the ﬁrst

genotype 1a crystal structure recently released²⁸where

the bound ligand was used as a template to generate ﬁeld

points while the hotspots (extracted above) were used as

constraints.

b. Setting up the guided docking preferences to direct the

ligands properly by:

i. Constructing a pharmacophore template to be used for

guided docking. This was guided by the SAR analysis

done in step 1.

ii. Finding the scoring function that correlates well with

the activity of this class of inhibitors. This is depicted

in Figure 2. As shown, a group of ligands with reported

activities were aligned using the ﬁeld template in the

binding site then reﬁned by energy minimization and

scored using different scoring engines to ﬁnd the best

scoring function that correlates well with the activity.

c. Validation of the model statistically using cross validation

(to assess activity predication capability) and ROC analy-

sis (to assess screening capability).^29,30

Recursive partitioning³⁵was used in addition as a ﬁngerprint-

based decision tree. It was not used to extract hotspots but simply

aided in the rapid understanding of the AMANDA and Topomer

CoMFA results as will be described later.

The three aforementioned techniques were carried out using a

dataset of 98 benzothiadiazine inhibitors retrieved from litera-

ture^36–38(see Tables 2–6). The activities used were those of geno-

type 1a. The results of each technique are as following:

3.1.1.1.1.1.Topomer COMFA. Topomer CoMFA is a technique

introduced by Cramer.³⁴It delineates the need for alignment which

is mandatory for typical CoMFA analysis. For space saving; only ﬁg-

ures with important outcomes are shown. Topomer CoMFA has

both graphical and statistical results. Concerning the graphical re-

sults, It was used to construct stdev^⁄coeff contour maps to show

ﬁeld effects on the target features. The contour plots are beneﬁcial

to identify important regions where some changes in steric or elec-

trostatic ﬁelds can affect the biological activity. The maps gener-

ated depict regions having scaled coefﬁcients greater than 80%

(favored) or less than 20% (disfavored).

d. Extending the model applicability domain to describe the

different activities among various genotypes: linear inter-

action energy³¹method was used as a post-docking scor-

ing function to calculate binding energies of the ligands

against various genotypes (genotype 1a, 1b and 4).The

calculations were done using crystal structures of geno-

types 1a and 1b and a homological model for genotype 4.

3. A library of benzothiadiazines was enumerated on reaction

basis such that it complies with SAR analysis done in step 1.

4. Structure-based fragment hopping was carried out by virtual

screening of the generated library using the validated model

in order to ﬁnd new derivatives.

5. Construction of a focused library based on a top scoring hit such

that it has lower TPSA followed by its synthesis and biological

evaluation to validate model experimentally.

3. Results and discussion

3.1. Molecular modeling

CoMFA sterically favorable (green) contours are observed

(Fig. 3A) adjacent to the 3-methylbutyl group of compound side

chain (the hydrophobic tail substituted at hydroxyquinolinone N)

and this indicates the importance of the hydrophobic tail in this

site. It is surrounded by sterically unfavorable yellow contours

which point out that this site is a narrow pocket and should be

occupied by hydrophobic tails that are not too bulky. Another

group of CoMFA sterically favored green contours are present at

methansulfonamide substituent group which is also surrounded

by sterically unfavorable yellow contours indicating that there is

a limit for this side chain expandability (Fig. 3B). In the CoMFA

electrostatic contour maps, the red contours show favorable elec-

tronegative regions, and the blue contours show regions where

the electropositive charges are favored. A close inspection of the

electrostatic contour plots reveals that, for the tested molecules,

electropositive and electronegative pattern which maps perfectly

with aminosulfonamide (Fig. 3C) is more preferred surrounding

side chain at position-7 of the benzothiadiazine.

3.1.1. SAR analysis

SAR analysis was carried out in order to extract Hotspots and

the complementary pharmacophore features. This is a crucial step

in which knowledge related to this class is extracted. As described

earlier, this knowledge will be used to construct the structure-

based ﬁeld template, the guided docking pharmacophore template

and ﬁnally generating a virtual library based on this class of

compounds.

3.1.1.1. Hotspots extraction. Hotspots are highly relevant regions

for ligand interactions which can be used for accurately placing li-

gands in the binding site. They were extracted using molecular

interaction ﬁeld (MIF) analysis techniques which can be applied

by different methods to study the ligand–receptor interactions.

MIF can be computed in the receptor binding site in order to scan

regions of importance for interaction, as was originally described

by Goodford.³²This was carried out in the protein-based methods

below using Superstar program. It is also possible to proceed the

other way around and compute MIF in one or many small

Regarding the statistical results of Topomer CoMFA analysis, the

model was validated internally yielding a q²of 0.777 and r²of

0.864 with

3 optimum components (PLS components). This

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2459

Table 2

Structures and activities of N-1-heteroalkyl-4-hydroxyquinolone-3-yl-benzothiadiazines

O

S

OH

N

H

N

O

R

NH

R₁

Name FR

Structure

Observed ꢀlogIC₅₀

Topomer CoMFA

AMANDA

LIAISON

Genotype1a Genotype1b

pred pred

0.293282 1.002365

ꢀ0.91 ꢀ0.05 ꢀ0.66 ꢀ0.421 ꢀ0.656 ꢀ0.893 ꢀ0.775 ꢀ0.811 ꢀ0.52737

0.382967

R

R1 Genotype 1a Genotype1b PRED R1

R2

LV1

LV2

LV3

LV4

LV5

1

A1 Ph

H

ꢀ0.71

ꢀ0.83

ꢀ0.02

ꢀ1.02

ꢀ0.79

ꢀ0.91

ꢀ0.37

ꢀ1.55

ꢀ0.51

ꢀ0.19

ꢀ1.18

ꢀ0.68

0.02

0.88

0.32

1.12

ꢀ0.67

0.31

0.67

0.90

0.17

0.51

0.87

ꢀ0.07

ꢀ0.16

0.79

1

ꢀ0.94 ꢀ0.05 ꢀ0.69 ꢀ0.357 ꢀ0.634 ꢀ0.776 ꢀ0.437 ꢀ0.559

2

3

4

5

6

7

8

9

A1 2-BrC₆H₄

A1 3-BrC₆H₄

A1 4BrC₆H₄

A1 2-MeC₆H₄

T1 3-MeC₆H₄

A1 2-Thienyl

A1 2-Thiazolyl

A1 2-Furyl

A1 3-Furyl

A1 3-Me-thieny-2-yl

A1 5-Cl-thienyl-2-yl

A1 Pr

0.536115

1.064247

ꢀ0.6206

0.443742

0.138103

0.722524

0.455823

0.30788

ꢀ0.86 ꢀ0.05 ꢀ0.61 ꢀ0.456 ꢀ0.364 ꢀ0.258 ꢀ1.164 ꢀ1.296

ꢀ1.02 ꢀ0.05 ꢀ0.77 ꢀ0.305 ꢀ0.390 ꢀ0.536 ꢀ0.368 ꢀ0.384 ꢀ2.01536

ꢀ1.09 ꢀ0.05 ꢀ0.84 ꢀ0.432 ꢀ0.587 ꢀ0.667 ꢀ0.674 ꢀ0.867 ꢀ0.64675

ꢀ1.00 ꢀ0.05 ꢀ0.75 ꢀ0.407 ꢀ0.294 ꢀ0.238 ꢀ0.820 ꢀ0.879 ꢀ0.81434

ꢀ0.76 ꢀ0.05 ꢀ0.51 ꢀ0.494 ꢀ0.470 ꢀ0.405 ꢀ0.555 ꢀ0.586

0.032644

ꢀ0.76 ꢀ0.05 ꢀ0.51 ꢀ0.451 ꢀ0.510 ꢀ0.537 ꢀ0.744 ꢀ0.911 ꢀ0.35388

ꢀ0.72 ꢀ0.05 ꢀ0.47 ꢀ0.501 ꢀ0.403 ꢀ0.298 ꢀ0.466 ꢀ0.488 ꢀ0.71322

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

ꢀ0.48 ꢀ0.05 ꢀ0.23 ꢀ0.467 ꢀ0.434 ꢀ0.372 ꢀ0.527 ꢀ0.646

0.109668

0.975671

0.471498

0.370195

1.595055

0.929392

1.114931

0.376449

1.014823

1.452157

1.320084

0.198086

0.637255

0.992241

0.627329

1.033736

0.166939

1.429609

0.287409

1.1061

ꢀ0.79 ꢀ0.05 ꢀ0.54 ꢀ0.419 ꢀ0.686 ꢀ0.754 ꢀ0.712 ꢀ0.726 ꢀ0.47609

ꢀ0.84 ꢀ0.05 ꢀ0.59 ꢀ0.410 ꢀ0.234 ꢀ0.138 ꢀ0.328 ꢀ0.380 ꢀ0.57943

0.08 ꢀ0.05

0.09 ꢀ0.05

0.03 ꢀ0.05

ꢀ0.22 ꢀ0.05

ꢀ0.09 ꢀ0.05

ꢀ0.03 ꢀ0.05

0.33 ꢀ0.523 ꢀ0.294 ꢀ0.025 ꢀ0.058

0.006

0.035

1.02182

0.332452

0.601349

A1 Bu

A1 i-Bu

0.03

0.34 ꢀ0.435 ꢀ0.091

0.220

0.219

0.20

0.39

0.03

0.55

0.87

0.86

0.99

0.83

0.82

0.19

0.92

0.41

ꢀ0.25

ꢀ0.05

ꢀ0.17

0.96

0.82

1.2

0.28 ꢀ0.499 ꢀ0.200

0.089 ꢀ0.001 ꢀ0.111

A1 Neopentyl

A1 i-Pr

A1 Cyclopropyl

A1 Cyclohexyl

A1 Me

A1 Et

A1 Pr

0.03 ꢀ0.496 ꢀ0.377 ꢀ0.106 ꢀ0.013

0.265 ꢀ0.61384

0.16 ꢀ0.462 ꢀ0.660 ꢀ0.619 ꢀ0.497 ꢀ0.293

0.22 ꢀ0.482 ꢀ0.327 ꢀ0.106 ꢀ0.127 ꢀ0.004

0.16641

0.945155

0.797223

ꢀ0.40

ꢀ0.47 ꢀ0.05 ꢀ0.23 ꢀ0.437 ꢀ0.264 ꢀ0.068 ꢀ0.107 ꢀ0.113

Me ꢀ0.80

0.05 ꢀ0.05

0.12 ꢀ0.05

0.14 ꢀ0.05

0.06 ꢀ0.05

ꢀ0.16 ꢀ0.05

0.30 ꢀ0.524 ꢀ0.204

0.37 ꢀ0.414 ꢀ0.232 ꢀ0.072 ꢀ0.135 ꢀ0.206 ꢀ0.09094

0.39 ꢀ0.495 ꢀ0.426 ꢀ0.296 ꢀ0.357 ꢀ0.268 0.423538

0.31 ꢀ0.381 ꢀ0.296 ꢀ0.254 ꢀ0.418 ꢀ0.442 ꢀ0.25025

0.09 ꢀ0.442 ꢀ0.569 ꢀ0.497 ꢀ0.395 ꢀ0.345 0.19382

0.130

0.054 ꢀ0.037 ꢀ1.00277

Et

Pr

ꢀ0.39

ꢀ0.28

ꢀ0.35

i-Pr ꢀ0.81

A1 Me

Ph

ꢀ1.22

ꢀ0.89 ꢀ0.05 ꢀ0.64 ꢀ0.455 ꢀ0.496 ꢀ0.474 ꢀ0.617 ꢀ0.726 ꢀ0.91748

A1

Cyclobutyl

Cyclopentyl

Cyclohexyl

Cycloheptyl

4-Pyrnayl

0.56

0.25 ꢀ0.05

0.29 ꢀ0.05

0.25 ꢀ0.05

0.06 ꢀ0.05

ꢀ0.09 ꢀ0.05

0.50 ꢀ0.511 ꢀ0.399 ꢀ0.198 ꢀ0.258 ꢀ0.101

0.955955

0.13

0.45

0.54 ꢀ0.468 ꢀ0.350 ꢀ0.170 ꢀ0.182 ꢀ0.015 ꢀ0.87332

0.50 ꢀ0.391 ꢀ0.148

0.042

0.045

0.099

0.58855

ꢀ0.02

0.66

0.35

0.31 ꢀ0.477 ꢀ0.370 ꢀ0.216 ꢀ0.286 ꢀ0.225

0.382967

0.534419

0.964247

1.168749

ꢀ0.67

0.16 ꢀ0.420 ꢀ0.424 ꢀ0.426 ꢀ0.625 ꢀ0.821

number of components identiﬁed in the LOO (leave one out) cross-

validation is used in the ﬁnal non-cross-validated PLS run (see re-

sults in Tables 2–6).

This model can be used for screening, especially R-group

screening, as the compounds were divided into 2 across the bond

joining ring B and C into R1 and R2 (see Section 5). However, this

was not the aim of Topomer CoMFA here; instead, it was used

for SAR analysis and extracting the hotspots represented in some

ﬁeld points of highly relevance in interactions.

3.1.1.1.1.2. Pentacle GRIND. As the name implies, this MIF anal-

ysis depends on GRIND which is short for Grid-independent

descriptors.^39–41It is described in details in Section 1.2 of the Sup-

plementary data. The data set of compounds (98 compounds) was

used as an input for Pentacle (see Tables 2–4 and 6). Default probes

(DRY, O, and N1) were used obtaining three auto-correlograms

(DRY–DRY, O–O, and N1–N1) and three cross-correlograms

(DRY–O; DRY–N1; O–N1). The DRY probe represents hydrophobic

interaction, O is sp2 carbonyl oxygen and has hydrogen-bond

acceptor properties, and N1 is a neutral ﬂat NH and has hydro-

gen-bond donor properties. The 6 types of GRIND variables were

used in PLS (partial leas square) to give a model.

In general, a Pentacle model gives graphical (Fig. 4) as well sta-

tistical results (Table 7). Simply we used the numerical values of

the PLS coefﬁcients to select the GRIND able to correlate with bio-

logical activity (Table 8), whereas the distances were used to visu-

alize the pair of nodes that have been used to assign a value to the

GRIND variable (Fig. 4). Inspection of this table conﬁrms the impor-

tance of the hotspots at the regions of the methansulfonamide at 7-

position, hydrophobic tail and the enolic oxygen. It is obvious that

pentacle did not add much to SAR analysis regarding the substitu-

ents (same information as Topomer CoMFA. However, the model

can be used for rapid virtual screening of the whole ligands which

can add to the R-screening in Topomer CoMFA. Practically, the

optimum number of PLS components (latent variables, LV) was

chosen by monitoring changes in the model’s predictivity index

evaluated by applying the cross-validation procedure available in

Pentacle (see Fig. S3 in Supplementary data). It was three just like

Topomer CoMFA.

3.1.1.1.1.3. Recursive partitioning. Recursive partitioning is a

technique to classify members of the population (the training set

here) according to several dichotomous dependent variables

(ﬁngerprints) in a decision tree. A total of 4 tree(s) were created.

Table 3

Structures and activities of N-1-benzyl and N-1-(3-methylbutyl)-4-hydroxy-1,8-naphthyridon-3-yl benzothiadiazine analogs containing substituents on the aromatic ring

O

S

R₇

OH

N

H

R₅

N

O

R₁

Name

FR

Structure

Observed ꢀlogIC₅₀

Topomer CoMFA

AMANDA

LV3

LIAISON

R1

R5

R7

Genotype 1a

Genotype1b

PRED

R1

R2

ꢀ0.69

LV1

LV2

LV4

LV5

Genotype1a pred

Genotype1b pred

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

A1

Benzyl

3-Methylbutyl

i-Pentyl

H

ꢀ0.76

ꢀ1.26

ꢀ0.92

ꢀ1.41

ꢀ0.83

ꢀ1.22

ꢀ0.79

0.09

ꢀ1.25

ꢀ0.05

0.41

ꢀ0.21

ꢀ0.70

0.44

0.03

0.74

1.34

0.20

0.85

0.51

0.92

0.33

1.0

ꢀ0.32

1.11

1.26

0.63

1.01

0.12

ꢀ0.06

1.38

ꢀ0.06

1.18

1.37

0.031

0.07

1.24

1.15

1.096

1.79

1.29

1.60

1.63

1.61

1.46

1.15

2.15

2.52

2.30

2.22

ꢀ0.94

ꢀ1.37

ꢀ0.63

ꢀ1.16

ꢀ0.63

ꢀ1.20

ꢀ0.76

ꢀ1.12

ꢀ0.08

ꢀ0.51

0.23

ꢀ0.05

ꢀ0.47

0.26

ꢀ0.412

ꢀ0.438

ꢀ0.194

ꢀ0.008

ꢀ0.395

ꢀ0.451

ꢀ0.388

ꢀ0.416

ꢀ0.477

ꢀ0.455

ꢀ0.242

0.774

0.025

0.670

1.350

0.412

0.969

1.471

1.308

0.432

ꢀ0.752

ꢀ0.932

ꢀ0.705

ꢀ0.071

ꢀ0.843

ꢀ0.886

ꢀ0.766

ꢀ0.853

ꢀ0.828

ꢀ0.726

ꢀ0.719

0.911

ꢀ0.901

ꢀ1.187

ꢀ0.968

ꢀ0.388

ꢀ1.061

ꢀ1.095

ꢀ0.980

ꢀ1.084

ꢀ0.883

ꢀ0.782

ꢀ0.876

0.593

ꢀ0.872

ꢀ1.138

ꢀ0.856

ꢀ0.537

ꢀ1.102

ꢀ1.082

ꢀ1.018

ꢀ1.057

ꢀ0.708

ꢀ0.754

ꢀ0.689

0.670

ꢀ0.143

ꢀ0.089

0.288

0.034

0.400

1.141

0.461

0.797

1.290

1.212

1.050

0.366

1.665

ꢀ0.824

ꢀ1.118

ꢀ0.771

ꢀ0.611

ꢀ1.237

ꢀ1.165

ꢀ1.073

ꢀ1.000

ꢀ0.353

ꢀ0.827

ꢀ0.503

0.765

ꢀ0.96343

ꢀ0.95527

ꢀ0.2196

ꢀ0.82273

ꢀ0.40993

ꢀ0.52608

ꢀ0.82401

ꢀ1.79239

0.231515

ꢀ0.84502

1.146922

0.208935

0.090485

0.00103

0.535334

1.029653

1.044727

1.73724

ꢀ0.80414

0.990781

0.908638

2.120819

0.127902

0.342872

2.08722

0.235235

0.140864

0.618475

0.32232

OMe

H

OMe

ꢀ0.69

0.17

OH

H

Me

ꢀ0.26

0.27

0.417148

0.337008

0.231433

ꢀ0.33675

0.959745

0.186936

1.681376

0.844165

0.762435

0.800711

1.26938

1.510461

1.490862

2.088696

0.145144

1.383639

1.42696

Me

H

Br

ꢀ0.30

0.14

ꢀ0.22

ꢀ0.05

ꢀ0.47

0.26

Br

H

OMe

H

OMe

0.17

H

OH

ꢀ0.14

ꢀ0.11

H

–OCH₂CH₂CH₃

–OCH₂CO₂t-Bu

–OCH₂COOH

–OCH₂CONMe₂

–OCH₂CONHMe

–OCH₂CONH₂

–-OCH₂CH₂NH₂

–OCH₂CN

–NH₂

0.49

0.52

ꢀ0.440

ꢀ0.174

0.562

ꢀ0.505

ꢀ0.314

0.400

0.129

ꢀ0.24

0.22

ꢀ0.21

0.25

ꢀ0.262

0.471

0.38

0.41

ꢀ0.173

ꢀ0.314

ꢀ0.019

0.168

1.213

0.739

1.049

1.165

0.939

1.034

0.195

1.571

1.536

1.679

2.005

0.36

0.39

0.648

0.284

0.44

0.47

1.356

0.934

0.080

1.536

1.143

1.384

0.545

1.145

1.303

0.823

0.184

1.422

1.024

1.086

0.289

1.338

0.44

0.48

0.30

0.33

H

0.28

0.31

0.637

0.964

1.217

0.654

0.689

0.519

0.750

0.817

i-Pentyl

H

–NH₂CH₂CN

–NH₂CH₂CONH₂

–NHCOCF₃

–NHSO₂Ph

–NHSO₂iPr

–NHSO₂(CH₂)₃CH₃

–NHSO₂Me

0.65

0.69

2.363365

0.888252

1.006582

2.330182

2.215868

2.53373

0.88

0.91

0.04

1.39

2.10

1.66

1.03

1.06

1.73

1.77

1.94

1.97

0.17

0.789

1.327

1.551

0.885

1.410

1.520

1.195

1.635

1.725

2.19691

2.65758

2.99897

1.56

1.59

2.70

1.56

1.59

2.869289

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2461

Table 4

Structures and activities of N-alkyl-4-hydroxyquinolon-3-yl-benzothiadiazine

O

H

N

H

N

S

O

OH

N

O

S

R

O₂

O

N

H

N

Name

Structure

R

FR

Observed –logIC₅₀

Topomer CoMFA

AMANDA

LV3

Liaison

Genotype1a Genotype1b

pred pred

Genotype1a

Genotype1b

PRED

R1

R2

LV1

LV2

LV4

LV5

59

60

61

62

63

64

65

66

67

68

69

CH₃

A1

1.20

0.72

1.06

1.14

1.46

1.24

1.68

1.21

1.06

1.02

1.30

1.18

1.10

1.13

1.15

1.10

1.11

1.09

1.21

1.13

1.16

1.18

1.13

1.14

1.12

0.17

0.817

0.938

1.305

1.311

1.115

1.520

1.388

1.777

1.470

1.496

1.288

1.551

1.182

1.219

1.111

1.407

1.255

1.387

2.111

1.033

0.743

1.107

1.520

0.796

1.081

0.989

1.259

1.334

1.335

1.850

0.689

0.779

0.839

1.725

0.778

1.028

1.038

1.319

1.414

1.408

1.475

0.633

0.717

0.984

2.005

1.057

0.962

0.740

1.078

1.617

0.948

1.613

0.703

1.017

0.777

CH₂CH₂Cl

CH₂CHCH₂

CH₂CCH

CH₂CH₂CN

CH₂Ph

CH₂CH₂NH₂

CH₂CO₂CH₂CH₃

CH₂CH₂OCH₃

CH₂CH₂OCH₂Ph

CH₂COOH

Table 5

Structures and activities of N-alkyl-4-hydroxyquinolon-3-yl-benzothiadiazine

X

O

H

N

S

N

Y

Z

OH

N

S

O₂

N

H

N

O

Name

Structure

FR

Observed ꢀlogIC₅₀

Topomer CoMFA

R2

AMANDA

LV3

Liaison

Genotype1a Genotype1b

pred pred

X

Y

Z

Genotype1a Genotype1b PRED R1

LV1

LV2

LV4

LV5

70

71

72

73

74

75

76

77

78

H

COOEt

COOH

CONH₂

CONHCH₂CONH₂

H

COOMe

COOH

CONHMe A1 1.37

A1 1.32

A1 1.70

1.45 1.49 0.17 1.773 0.705 0.964 0.837 1.265

1.42 1.45 0.17 1.327 1.462 1.331 1.386 1.457

1.48 1.51 0.17 1.648 0.984 1.628 1.663 1.620

1.47 1.50 0.17 1.470 1.410 1.550 1.486 1.128

1.17 1.20 0.17 1.836 1.659 1.964 1.858 1.649

1.23 1.26 0.17 1.066 1.517 1.348 1.359 1.122

1.24 1.27 0.17 1.651 1.323 1.198 0.856 0.667

1.19 1.22 0.17 1.744 1.460 1.666 1.478 1.334

1.19 1.22 0.17 2.079 1.023 1.826 1.574 1.209

CONH₂

A1 1.74

A1 0.88

A1 0.94

A1 1.37

A1 0.89

H

COOMe

The best tree was the ﬁrst tree with a cross-validated ROC score of

0.746 (see Fig. S4 in Supplementary data).

addition, it points out that ring A, as pyridine, is more active than

benzene. Regarding aminosulfonamide, it should be short in gen-

eral indicating that it binds in a narrow pocket as increasing bulk-

iness decreases activity relatively. The pocket hydrophilic nature

prefers polar expandability than those hydrophobic substituents

with aromatic ring.

The compounds were divided according to their activity into

bins ranging from 1 to 10. The tree begins by dividing the whole

series into two major categories: those with amino sulfonamide

and those without. Compounds with amino sulfonamide are more

active: Activity-bin was 6 and above. The tree also indicates that

the isoamyl moiety is the best hydrophobic tail in this class. In

Speciﬁcally here, the recursive partitioning was used in order to

aid the interpretation of MIF Analysis by pointing out to those

Table 6

Structures and activities of N-alkyl-4-hydroxyquinolon-3-yl-benzothiadiazine

R₁

N

O

H

N

S

OH

N

S

R₂

O₂

N

H

N

O

Name

Structure

FR

Observed –logIC₅₀

Topomer CoMFA

R2

AMANDA

LV3 LV4

Liaison

R1

R2

Genotype 1a Genotype1b PRED R1

LV1

LV2

LV5

Genotype1a Genotype1b

pred

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

H

CH₂CH₂Ph

Ch₂Ph

Ph

A1

1.74

2.26

1.39

0.97

1.17

1.41

2.00

1.49

1.70

2.00

1.82

1.24

1.47 1.50 0.17

1.83 1.86 0.17

1.40 1.43 0.17

1.65 1.68 0.17

1.61 1.65 0.17

1.68 1.71 0.17

1.61 1.65 0.17

1.65 1.68 0.17

1.49 1.53 0.17

1.85 1.88 0.17

1.84 1.87 0.17

1.67 1.71 0.17

1.550 1.280 1.060 1.142 1.031

1.111 1.782 1.862 1.975 1.809

1.172 1.811 1.938 2.027 1.722

0.862 1.414 1.454 1.603 1.456

1.040 1.327 1.218 1.196 1.363

0.831 1.400 1.405 1.497 1.499

1.197 1.504 1.518 1.561 1.630

1.654 2.176 2.000 1.688 1.945

1.259 1.155 1.669 1.591 1.914

1.914 1.319 1.612 1.615 2.056

1.215 1.428 1.475 1.703 1.576

1.465 1.713 1.651 1.741 1.701

CH₂CH₂OH

Cyclohexyl

Cyclopentyl

CH₂CH₂NH₂

4-piperidinyl

CH₂CH₂CONH₂A1

4-MeOC₆H₄CH₂A1

3-MeOC₆H₄CH₂A1

2-MeOC₆H₄CH₂A1

Piperidinyl

Pyrrolidinyl

Azetidinyl

A1

1.29 1.34679

1.57

1.62

1.35

1.62

1.75

1.39 0.17 1.319 1.297 1.575 1.673 1.484 1.69243 2.067777

1.65 0.17 0.736 1.081 1.252 1.525 1.277

1.78 0.17 0.881 1.408 1.374 1.597 1.702

O

H

N

S

N

O

OH

N

S

O₂

O

N

H

94

A1

0.92

1.06

1.09 0.17 0.929 1.505 1.662 2.037 2.004

N

O

H

N

S

NH₂

OH

N

S

O₂

N

H

95

A1

2.05 2.55

1.81

1.85 0.17 1.081 1.033 0.957 1.154 0.821 1.04576 1.521574

N

O

H

N

H

N

S

OH

N

S

O₂

N

H

96

A1

1.85

1.92

1.96 0.17 1.202 1.925 1.870 1.780 2.304

N

O

H

N

S

NH₂

OH

N

S

O₂

N

H

97

A1

2.28 3.4

2.22

1.85 0.57 1.267 1.703 1.624 1.510 1.618 2.424

2.44256

N

O

HN

O

H

N

H

N

S

OH

N

S

O₂

N

H

98

A1

2.06

2.33

1.96 0.57 1.210 2.149 2.036 1.769 2.155

N

O

HN

2464

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

Figure 3. Topomer CoMFA results. (A) Green contours favor steric bulk. Adding substituents that reach into these areas can improve activity. It is generally well illustrated

above that increasing bulkiness in this position increase activity but to certain limit indicated by the yellow contour, (B) Contour maps at the hydrophobic tail indicate that

the sub pocket is relatively narrow as the yellow contour limits the extending of the substituent, (C) Blue contours indicate favorable electropositive groups represented

idealistically by NH groups, while red contours represent favorable electronegative regions (sulfonyl oxygen), (D) The linear relationship between experimental (x-axis) and

predicted activity (y-axis) yielding Q2 of 0.777 and R2 of 0.864.

important features that should be taken into consideration. In

addition, it facilitates picking up those compounds which include

these features. The way it is used to aid interpretation is exempli-

ﬁed in Figure S5 of the Supplementary data. Finally, the model val-

idation using confusion matrix and ROC analysis was done in

Supplementary data (1.3 recursive partitioning).

benzothiadiazine and benzothiazines in Section 1.8 in the Supple-

mentary data to conﬁrm the importance of the features corre-

sponding to the hotspots and pharmacophore features.

3.1.1.2. Pharmacophore elements extraction. Pharmacophore

features were extracted using molecular dynamics and energy-

based pharmacophore and this is described in the ligand–protein

approach below.

3.1.1.1.2. Protein-based method. SuperStar⁴²technique was

used here. It is given in Section 1.4 in the Supplementary data. In

this work, it was used to investigate the protein binding site in or-

der to assign the position of the hotspots contributing in the favor-

able interactions that comply with those obtained from the ligand

based MIF analysis. Actually, SuperStar (see Section 1.4 in Supple-

mentary data for full analysis) was used together with 2D interac-

tion diagram (Fig. 5) (generated after passing the complex through

a MD production run) to highlight the main pockets of the binding

site (pockets I, II and III) and the amino acids which form them (Ta-

ble 9): pocket I consists of Phe415, Met414 and Gly410, pocket II

consists of Pro197 and Leu384 and ﬁnally pocket III consists of

Ser288, Ser556, Asn291 and Asp318. Besides, it was used mainly

to solve the issue of structural water molecules in the binding site

which are essential for binding. This is shown by the pharmaco-

phore points (Figs. S6, S7 and S10 in Supplementary data) that indi-

cate the importance of two structural water molecules that should

be included in the docking study (Table 9 and Fig. 5). The ﬁrst

water molecule interacts with both the ligand sulfonyl oxygen

and Ser556 (acts as a bridge) while the second water interacts with

both NH of Gly449 and the ligand enolic oxygen and similarly acts

as another bridge. Superstar analysis, as mentioned before, is given

in details in the Supplementary data in Section 1.4. In addition, we

provided the total 2D interaction diagrams of all the complexes of

3.1.1.2.1. Ligand–protein complex method. Energy-optimized

pharmacophores (E-pharmacophores)²⁷are based on mapping of

the energetic terms of the Glide XP scoring function onto atom cen-

ters. Ligand-receptor complex was reﬁned using Glide XP. After

that, the energy terms computed were mapped onto the molecule

atoms to guide the pharmacophore generation. The features are

energy weighed because each feature is the sum of the Glide XP

energies of the atoms forming it (Fig. 6). The energies of each site

are used for ranking where the most favorable sites are selected

to generate the hypothesis. GLide XP was chosen to be energeti-

cally dissected as it accounts for more complex energy terms than

traditional molecular mechanics or empirical scoring functions

such as ChemScore.⁴³These extra features (e.g., hydrophobic

enclosure term and hydrophobic interactions motifs detection)

are well illustrated in our case. For example, hydrophobic enclo-

sure is detected when lipophilic ligand atoms are enclosed on

opposite sides by lipophilic protein atoms, producing a large ener-

getic reward as a consequence of the release of unfavorable water

molecules that were previously present in the hydrophobic envi-

ronment. Pockets I and II represent examples of this enclosure. Fur-

thermore, Glide XP recognizes common hydrophobic interaction

motifs such as

p

ꢁ ꢁ ꢁ

p

and

p

ꢁ ꢁ ꢁcation interactions.⁴⁴This was also

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2465

Figure 4. AMANDA MIF analysis for the benzothiadiaizines as HCV polymerase inhibitors: (A) PLS pseudo-coefﬁcients plot showing the importance of single descriptors to

explain Y-variable: positive values of a coefﬁcient indicates direct correlation to the Y and negative ones indicate an inverse correlation to the Y. Large values mean a strong

impact and low values mean a marginal impact. In this special plot, variables 238 and 362 are highlighted showing a high positive value while variable 299 shows high

negative value, (B) Var 299 (Dry-N1 grind descriptor) with negative coefﬁcient shows the negative impact of aromatic ring as hydrophobic tail on the activity, (C) Var 362 (O-

N1 grind descriptor) showing importance of H-bond donor NH group of methansulfonamide and H-bond acceptor feature of the enolic group, (D) Var 238 (DRY-O grind

descriptor) with positive coefﬁcient showing importance of the hydrophobic aromatic group (ring A) and the H-bond donor of the NH of the methansulfonamide.

Table 7

depicted in the interaction diagram of 3HHK complex which shows

Statistical results of the AMANDA model

the

pꢁ ꢁ ꢁcation interaction between MET414 and ring A and that be-

tween PHE193 and ring D (Fig. 5).

LV

SSX

SSXacc

SDEC

SDEP

R2

R2acc

Q2acc

In this study E-pharmacophore together with 2D interaction

diagrams were deployed to analyze molecular dynamics simula-

tion snapshots along production trajectory and the outcome can

be summarized as following:

1

2

3

4

5

57.11

5.71

4.53

4.07

2.22

57.11

62.83

67.36

71.43

73.66

0.62

0.47

0.42

0.38

0.32

0.64

0.52

0.50

0.51

0.54

0.65

0.15

0.04

0.03

0.04

0.65

0.80

0.84

0.87

0.90

0.63

0.76

0.77

0.76

0.74

(a) SSX: percentage of the X sum of squares explained by this LV; (b) SSXacc:

accumulative percentage of the X sum of squares explained by the model; (c)

SDEC: standard deviation error of the calculations. It is an index of model ﬁtting

on the training set; the lower the better; (d) SDEP: standard deviation error of

the predictions. It is an index of the model predictive ability obtained by cross-

validation; the nearer to SDEC the better. (e) R²: contribution of the current LV

to the coefﬁcient of determination (r²) of the model; (f) R²acc: coefﬁcient of

determination (r²) of the model. It is an index of model ﬁtting on the training

set; the nearer to 1.00 (theoretical maximum) the better; (g) Q²acc: equivalent to

r²but obtained from cross-validation. It is an index of the model predictive

power. It is obvious that the best descriptive model is the one with three latent

variables as q²reached a ceiling level where further increase in latent variables

decrease q².

1. The important information extracted using this method is

related to the main scaffold (constant part). This information

cannot be obtained via ligand-based methods which describe

difference in activity as a function of the structure variability.

2. The enolic oxygen proved to be essential feature as it forms H-

bond with Tyr448 directly and indirectly via water bridge with

Gly449 (Fig. 6).

3. Met414 is important in the formation of sigma–pi interaction

with ring A.

4. Phe193 is important in the formation of sigma–pi interaction

with ring D.

Table 8

AMANDA important grind variables which are illustrated in Figure 5

Grind variable

Fig. Importance

Var 238 (DRY-O grind descriptor).The distance range 5D

of this variable is 14.8–15.2 A

Var 362 (O-N1 grind descriptor) The distance range 5C

of this variable is 10.8–11.2 A

NH of the Methanesulfonamide group and hydrophobic aromatic feature (Ring A) are important hotspots.

Postive PLS coefﬁcient indicates direct correlation with activity

NH of the methansulfonamide group and the enolic group are important features which correlate

positively with activity

Var 299 (dry-N1 grind descriptor) The distance

range of this variable is 12.4–12.8 A

5B

The bulky groups as a hydrophobic tail have negative impact on the activity. The effect of this variable is

negative one

2466

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

Figure 5. 2D ligand–protein interaction diagram showing detailed analysis of the binding site. The importance of water molecules in the interaction is shown: structural

water acts as a bridge with Gly449 while the other acts as a bridge with Ser 556. Hydrophobic interaction of the isoamyl tail with Pro 197 and Leu 384 (greasy green residues)

is obvious. Two important pi sigma interactions are depicted: Met 414 with ring A and Phe193 with ring D. Three hydrophobic pockets are red boxed. Polar amino acids are

green-boxed. Structural water molecules are blue-boxed.

Table 9

Binding pockets of palm site I

Region in protein

Pocket I

Components

Description

Complementary feature in drug

Method of detection

Consists of residues Phe415,

Met414, Asn411 and Gly410,

close to the surface of the enzyme sub-pocket

A hydrophobic

shallow and wide

The benzo moiety (ring-A) of the benzothiadiazine

E-pharmacophore ring

aromatic hydrophobic

feature and Superstar

contours

Pocket II

Pocket III

Consists of residues Pro197 and

Leu384

Deep and narrow

hydrophobic sub-

pocket

The hydrophobic tail at attached to the ring A

The methansulfonamide at position 7 of ring D

E-pharmacophore

hydrophobic

feature,Ligand-based

methods and Superstar

Ligand based methods and

2D-interaction diagram

Consists of Ser288, Ser556,

Asn291 and Asp318.

Relatively narrow

hydrophilic

subpocket

Polar amino acid

residues

Tyr448,Gly449 and Ser 556

Involved in polar

interactions with

inhibitor

The enol moiety forms Hbond directly with Tyr 448 Superstar, 2D interaction-

and indirectly via water bridge with Gly449

The sulfonyl oxygen and ser 556 are bridged by

water molecule

map and E-pharmacophore.

Structural water

molecules

Two water molecules Figure 5

Forms water

bridge between

ligand and the

protein

The interaction between sulfonyl oxygen and serine Superstar and interaction-

556 is bridged by a structural water molecule

The oxygen atom of the enol moiety forms Hbond

with a structural water molecule, which itself forms

H-bond to the backbone NH of Gly449

map

5. Sulfonyl oxygen proved to be important in the formation of H-

bond with Ser556 via water bridge.

3.1.2. Generation of the structure-based activity prediction

model

The structure-based activity prediction model is composed of

two components: ﬁeld template followed by guided docking. This

model was validated after that and its scope was extended to be

able to predict the activity of various genotypes.

The energy-based pharmacophore method was preferred to

contact-based methods⁴⁵where accuracy can be achieved by

passing the complex through a nanosecond MD production. The

E-pharm was extracted along the trajectory and thus, each phar-

macophore feature represents the average energy across trajectory.

Molecular dynamics was performed by replicating literature

results while shortening production time to 1 ns instead of 10.²⁰

To sum up, we summarized the results obtained in Table 9.

3.1.2.1. Field template. Field template is a structure-based phar-

macophore that is used before docking as shown in the workﬂow.

It uses ﬁeld points which describe the van der Waals and electro-

static (both positive and negative) minima and maxima that

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2467

Figure 6. Energy optimized pharmacophore. It is clear that hydrophobic tail and the enolic oxygen of ring B are essential features. Green spheres represent hydrophobic

features; the red sphere represents hydrogen-bond acceptor feature and the orange circle represents ring aromatic feature. The table shows the energy weight of each feature.

surround molecules. The ﬁeld points are generated using the ex-

tended electron distribution (XED) force ﬁeld⁴⁶which explicitly

represents electron anisotropy as an expansion of point charges

around each atom, whereas in classical molecular mechanics

charges are placed on the centers of atoms. They are calculated

by moving different probes on a grid of points placed above the

Van der Waals molecular surface. As depicted in Figure 7, ﬁeld

points are color coded (see Supplementary data for the key in

Section 1.5) and their radius represents the depth of the energy

well.

Field template has two main roles in the workﬂow:

First, it was used to align ligands in the binding site with respect

to the main bound conformer of the reference ligand such that they

are placed in a good orientation and position. This is like providing

initial good poses to the docking engine to speed up convergence

(i.e., reaching a good solution very fast). In other words, docking

Figure 7. Field template was used as a guide for ligand placement. Small blue spheres indicate H-bond acceptor; Yellow spheres represent hydrophobic ﬁelds while red

spheres indicate positive electrostatic ﬁeld. Constraints were applied on some important hotspots as indicated in the ﬁgure.

2468

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

will be restricted on just pose reﬁnement and assessment. Field

template was chosen as a best choice for this task and was pre-

ferred to any geometrical features structural-based pharmaco-

phore for the following reasons:

2. Goldscore, Chemscore ASP, CHEMPLP.⁵²This was done using

Gold 5 suite.

3. Moldock score⁵³using Moldock optimizer in Molegro Virtual

Docker.

4. Glide score⁴⁴using Schrodinger Glide XP reﬁnement.

a. Success of MIF analysis in describing the activity with good

correlation suggested the suitability of using ﬁeld points in

alignment.

b. The fact that structure-based pharmacophore (based on geo-

metrical features) cannot keep the RMSD between ligands to

a minimum. On the other hand, the ideal tool for accurate

ligand alignment, when the compounds share the same scaf-

fold, is the ﬁeld alignment. It is one of the best choices to

supply aligned ligands for COMFA analysis.

Correlation matrix was generated and Moldock score proved to

correlate well with activity yielding r²of 0.62. This result is not

commonly found in literature as scoring functions are normally

not measured in chemical units and are not likely to correlate with

the activity. They are mainly used to achieve good enrichment or

ranking in screening. Despite this fact, this is sometimes achieved

in some cases.⁵⁴

c. Molecules are aligned based on their molecular ﬁelds, not on

their structure. The interaction between a ligand and a pro-

tein involves electrostatic ﬁelds and surface properties (e.g.,

hydrogen bonding, hydrophobic surfaces and so on). Two

molecules which both bind to a common active site tend

to make similar interactions with the protein and hence

have highly similar ﬁeld properties.

3.1.2.2. Knowledge-based docking. Guided docking is a knowl-

edge-based strategy which is guided by both pharmacophore fea-

tures and ligand–protein interaction. Moldock was used in this

study as an example for this type of docking. The inputs of the

docking engine can be listed as follows:

1. Pharmacophore template and this was obtained as a function of

the bound conformer using the energy-based pharmacophore

and molecular dynamics results. It is used to guide docking

depending on the template similarity score which evaluates

compliance of the docked poses with the pharmacophore

template.

2. Interaction score and we choose Moldock score (variant of PLP)

since it correlates well with activity.

3. Aligned ligands which emerge from the ﬁeld template in a good

orientation and position thus restricting docking role on reﬁne-

ment and assessment.

Second, it was used in ﬁnding the best scoring function which

correlates well with the activity.

In order to carry out these two functionalities, we had to build

the ﬁeld template ﬁrst. This template needs the reference bound li-

gand to extract ﬁeld points and a protein to be used as an excluded

volume. The process of building the template was carried out in

two steps:

a The bound conformer and the protein were prepared as fol-

lowing: the crystal structure of genotype 1a was used

because the MIF analysis showed good correlation with the

reported activities for this genotype. Previously, Homological

model of Genotype 1a¹⁸was used in studying of benzothiadi-

azines to inspect binding mode. However, the ﬁrst 3D struc-

ture of the subtype 1a NS5B polymerase in complex with an

inhibitor (a novel sulfone-benzodiazepine derivative) was

publicly disclosed (PDB ID 3HKW) recently.²⁸In order to

use it in this study, we had to insert the benzothiadiaizine

in the palm I site of this deposited crystal. Thus, alignment

was done using 3HHK (genotype 1b with benzothiadiazine

ligand) as a template using 3DMA program to align the

sequences of the two protein structures according to their

3D structural similarity. 3HHK bound conformer was then

copied to 3HKW binding site and clashing waters were

removed. It was obvious that water molecules responsible

for forming water bridges (with SER556 and GLY449) were

present in genotype 1a crystal. The complex was then opti-

mized using LigX algorithm.

Guided docking was used here for general and speciﬁc reasons.

In general, guided docking is used because it applies similarity

corrections (according to the pharmacophore supplied) through-

out the entire ligand incremental construction process, thus

affecting not only the relative ordering of solutions but also ac-

tively guiding the ligand docking. In the end, the use of a similar-

ity-weighted docking score aims at promoting target ligand

orientations after the binding mode observed for the reference

structure and penalize those diverging from the observed binding

mode. The ﬁnal best docking solutions should result in docking

orientations of the ligand that provide a balance between main-

taining favorable interactions (guided by docking score) with the

receptor without deviating too much from the observed binding

mode of the reference structure used (guided by template similar-

ity score).

Guided docking was used speciﬁcally in this case because:

1. Palm I site is wide and can accept the ligand in many different

orientations which differ drastically from the ideal pattern.

2. The rotatable bond joining ring B and C allows ﬂipping of the

hydrophobic tail, especially when it is bulky, from the narrow

pocket II to the wider pocket I.

b The ﬁeld template was generated using 3HHK bound con-

former and 3HKW protein as excluded volume while using

the extracted hotspots as constraints (Fig. 7).

After we generated the ﬁeld template, we used it to ﬁnd the

suitable scoring function. The template was used to align the li-

gands of the dataset (98 compounds). This type of alignment takes

place on a ligand level and avoids clashing with the protein using a

soft potential but it does not take interactions with protein into

consideration. Therefore; an optimization step was needed to re-

ﬁne the ligands after alignment. After that, the ligands were scored

to assess ligand binding. The scoring functions used were:

The algorithm used in this study for the guided docking is based

on an evolutionary algorithm (EA).²⁶(See Supplementary data for

details concerning the algorithm and why it was used in Section

1.6). The EA used by the Moldock optimizer was guided by the pose

score which consists of: Moldock score and a Template similarity

score which is normalized to ꢀ500.

The uses of guided docking can be tailored here either for vir-

tual screening or activity prediction. Virtual screening can be car-

ried out using the combined score (pose score) which is the sum

of the interaction and similarity scores. This is to obtain ﬁnal best

docking solutions with a balance between maintaining favorable

1. LigScore1,⁴⁷LigScore2,⁴⁷PLP1,⁴⁸PLP2,⁴⁸Jain,⁴⁹PMF,⁵⁰LUDI1,⁵¹

LUDI2⁵¹and LUDI 3.⁵¹These were evaluated after in situ ligand

minimization using CHARMm (Accelrys Discovery Studio).

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2469

interactions with the receptor without deviating too much from

the binding mode described by the template. On the other hand,

activity can be predicted using the interaction score as it is the

one which correlates with the activity. This is carried out by dock-

ing using the combined score as a guide during sampling and ﬁnal-

ly predicting the activity using the interaction score.

production to achieve a reasonable result. Thus, we shifted to a less

exhaustive computational method which is the linear interaction

energy.⁵⁷It was used to calculate binding energy of the poses

retrieved from docking thus it was used as a rescoring or a post-

docking scoring engine. It is a QSAR method that depends on

ﬁnding the correlation between the binding energy and several

energy terms which includes electrostatic and van der waals inter-

actions (see Section 1.9 in the Supplementary data for detailed

description of the method).

3.1.2.3. Validation. Validation was carried for both prediction and

screening capabilities. For screening validation, ROC analysis^29,30

was done using 1000 diverse ligands from zinc database as decoys

(supplied with Sybyl). It was repeated twice: once using the com-

bined score and the other using the interaction score. The com-

bined score proved superiority in screening (AUC = 0.914 if

compared to Interaction score which have AUC = 0.748) as it guar-

antees that ligands which lack essential chemical features will be

penalized and will have lowered overall score (see Fig. S12 in the

Supplementary data).

3.1.2.4.1. Genotype 1a/1b proﬁling. First, we investigated the

actual correlation between the observed activities of both geno-

types using a group of 61 compounds with reported activities on

both genotypes^36–38(see Tables 2, 3 and 6). The correlation be-

tween the activity of the benzothiadiazines of genotype 1a and

1b is shown in Figure S15 in the Supplementary data. It is clear

from that graph that the impact of this difference is almost con-

stant (about 0.7–1.2 log unit for ꢀlog (IC₅₀) of 1b more than

1a)²¹and the activity of 1b is generally greater than 1a.This trig-

gered us to investigate genotype 1b using Topomer CoMFA analysis

but the results were not good as genotype 1a (r²= 0.722,

q²= 0.467).This may explain why CoMFA studies in literature were

never carried out using genotype 1b.Thus we shifted to linear

interaction energy method to predict the activity difference be-

tween genotype 1a and 1b. These two genotypes have only two dif-

ferent amino acids (415 and 446) in the palm I allosteric site (Table

10). Genotype 1b has Glutamine 446 instead of glutamic acid (in

genotype 1a) and Tyrosine 415 instead of phenylalanine (in geno-

type 1a). Due to this small difference in the binding site between

the two genotypes, we carried out point mutations in the binding

site of genotype 1a crystal (as a way of expressing genotype natural

polymorphism) and applied LIA equation trained using genotype

1a on the genotype 1b. This was done in order to minimize vari-

ance between the two genotypes and decrease time required for

convergence in minimization simulation employed during LIA

optimization step.

For prediction capability validation, correlation study was done

using the Moldock score and the activity in the form of ꢀlog (IC₅₀

)

after increasing the number of iterations of the EA to 10,000. The

analysis yielded r²of 0.8 (see Fig. S13 in the Supplementary data)

which is superior to the previous correlation using energy minimi-

zation for the aligned ligands (r²= 0.62).This is a strong indication

that multiple parameter optimization deployed in the EA

algorithm can yield better results.

Model extrapolative capability was tested using ligands which

do not belong to the benzothiadiazine class to evaluate the model

extrapolative capabilities and compare it with ligand-based MIF

analysis described early in the study. 1,5 Benzodiazepine was used

as a test (pdb 3HKW). It acts by the same mechanism on the allo-

steric site (palm I)⁵⁵and its activity is around 10 nM (i.e., ꢀlog

(IC₅₀) = 2). Using AMANDA as an example for ligand-based MIF

analysis, the activity predicted for this compound was ꢀ0.218

using 3 latent variables. Using Moldock optimizer, on the other

hand, as an example for guided docking; the activity predicted

was 1.8 with RMSD of 1.5. As a result, one can conclude that the

receptor-based model can be extrapolated to other series without

losing predictability or pose accuracy. This proved that structure-

based activity prediction models have wider application than that

of ligand based ones.

Practically, this method was done using Schrodinger 2009⁵⁸

Glide for reﬁnement of the ligand poses in the binding site of geno-

type 1a crystal employing XP mode. Following this, the LIE energy

terms were calculated using Liasion. Finally, Strike was used to

make a multiple linear regression (MLR yielding a q²of 0.7 using

leave one out (LOO) for cross validation while neural network

regression yielded q²of 0.75. The equation was employed using

3HKW (genotype1a). After that, the equation was applied by carry-

ing out the two point mutations on the two amino acids 415 and

446 and the activities were predicted. It was generally observed

that the difference between the activities of one ligand in the

two genotypes ranges from 0.7 to 1.2 log units which were consis-

tent with the general ﬁnding above (see Fig. S16 in the Supplemen-

tary data). Liaison results are also given in the tables of the training

data set.

3.1.2.4. Extending model applicability. In this workﬂow, it was

important to have a valid procedure to predict the activity differ-

ence among various genotypes.²¹This is because palm I site is

the least conserved among the other allosteric sites of HCV and

thus will have effect on the activity of its inhibitors. For instance,

the model was used to describe genotype 1a; however, in biologi-

cal assay genotype 1b is used as a ﬁrst choice because the activity

level is generally higher for the same ligand.²¹Another example is

the genotype 4 which is another variant of HCV. It shows high

mutation in the binding site amino acids and is more likely to ex-

hibit dramatic change in the activity. In addition, it is the most pre-

valent genotype in our country (Egypt). Generally, activity

variation can be studied by modeling when using more accurate

methods for binding energy prediction like FEP (free energy pertur-

bation).⁵⁶The FEP is the ideal method to be used in this case but it

needs many nanoseconds of explicit solvation molecular dynamics

3.1.2.4.2. Genotype 4. Another approach was followed to study

genotype 4 due to the big variation in the binding site region (Table

10). The approach is divided into two main aspects:

1. Studying the effect of each amino acid variation by calculating

mutation binding-energy and detect the destabilizing effect if

exists.

Table 10

Palm I allosteric site amino acids variation among different genotypes

Residue position

410

411

412

413

414

415

446

447

448

449

450

451

551

555

556

1b

1a_2

4a_26

G

—

N

—

I

—

I

—

M

—

V

Y

F

—

Q

E

D

I

—

M

Y

—

G

—

A

—

V

C

—

T

F

—

Y

—

A

S

—

G

2470

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

Table 11

formed between Met414 and ring A (see Section 1.13 in the Sup-

plementary data).

Mutation binding energy calculated for interacting amino acids of palm I site for

genotype 4

Up till now, we have created two main components which are

the detailed SAR analysis and the development of the virtual

screening and activity prediction models. The integrity of the

workﬂow and the non-redundancy of its components are discussed

in details in Section 1.14 in the Supplementary data. What is left is

the way that this model can be applied for new inhibitors design or

discovery. This model can be used in designing new ligands by

many methods:

Residue

Mutated to

Mutation energy(kcal/mol)

Effect

MET414

GLN446

ILE447

TYR555

SER556

VAL

ASP

MET

ALA

GLY

0.91

0.42

0.07

ꢀ0.16

0.11

Destabilizing

Neutral

The destabilizing effect of MET414 is shown.

1. Virtual screening of commercial databases (not used in this

study but was validated using decoys sets).This can be used

for discovery of benzothiadiaizine analogues.

2. Virtual screening of virtual libraries that are built while taking

into consideration the essential structural requirements (this

takes into consideration the detailed SAR analysis provided by

the workﬂow).

The main variations in genotype 4 (M414V, Q446D, I447M,

A450V, C451T, Y555A, S556G) were studied using an approach

implemented in Accelrys discovery studio which depends on calcu-

lating free energy of binding of AB complex either mutated or not.

The protocol used was described in the Supplementary data in Sec-

tion 1.11.

The method was applied on the amino acids which interact di-

rectly with the benzothiadiazines (Table 11). The results comply

well with the literature data about the effect of mutation on the

activity of the benzothiadiazines: It is obvious that MET414 plays

an important role interacting with pocket I and its mutation, in nat-

ural polymorphism of genotypes, is deleterious on the activity.²¹

3. De-novo ligand design directed by the correlating scoring func-

tion (PLP).

Herein, we will give a simple experimental example based on

method-2 where we will try to ﬁnd a new template that can be fur-

ther exploited by SAR analysis.

2. Creating a homological model using genotype 1b (3H98) as a

template and applying the LIA equation derived from genotype

1a on it. The Homological model was based on the alignment of

the sequence of genotype 4 with that of genotype 1b (see

Fig. S17 in the Supplementary data). The template chosen was

3H98. Details are mentioned in Section 5.

3.1.3. Virtual library generation

As mentioned above, we have chosen virtual library screening

method to apply model on. This approach of designing new deriv-

atives of benzothiadiazine is based on fragment hopping. The de-

signed ligands can be prioritized after that by evaluating their

activity using the generated model. Initially, a combinatorial vir-

tual library was built using library design module in discovery stu-

dio that is based on reaction enumeration. The reaction was

selected according to a literature method in which aldehydes react

with orthanilamides in the presence of sodium bisulﬁte at 150 °C

(Fig. 8).⁵⁹

The reactants were selected such that they are synthetically fea-

sible and meet the essential SAR analysis. They were searched

using Sciﬁnder due to its wide scope where it includes compounds

synthesized in literature and thus avoiding limitations of using the

commercially available substances which cover a narrow range of

synthetically feasible compounds. Initially, the aldehydes were

searched such that: the N-substituted alkyl group at position 3

with respect to the aldehyde group should exist (Fig. 8B), usage

of dashed bonds in Sciﬁnder to avoid speciﬁc bond type and

searching is by substructure search to get diverse solutions. The re-

sults were retrieved in the form of pdf (portable document format)

ﬁle which was converted into structure data (sd) ﬁle using a

chemical OCR (Clide Pro).⁶⁰

The homological model of genotype 4 was aligned with geno-

type 1a crystal structure in order to copy the docked poses from

1a to 4. Reﬁnement of these poses was done using Glide XP. After

that, Liaison parameters were calculated and substituted in the

general equation created using genotype 1a. After that, an equation

was derived by applying MLR (multiple linear regressions) relating

the predicted activity on genotype 4 (by applying LIA model on the

homology model) and the genotype 1b reported activity. The dif-

ference in the activities can be illustrated as follows:

⁄

neg1bꢀ1.70595

Predicted genotype 4 = 0.994659

the binding energy of genotype 1b

where neg1b is

It is obvious that the general trend observed is the decrease in

activity by about 2 log units (it means 100-fold decrease in activ-

ity).This is somehow consistent with the results published.²¹

Further analysis was carried out on the effect of the mutation of

genotype 4 (Met414 to valine) using molecular dynamics and ra-

dial distribution function of the sigma-pi interaction that can be

Figure 8. Reaction A is used to build a combinatorial virtual library. It is a one pot reaction between an aldehyde and an orthanilamide. The aldehyde was searched by query B

using Sciﬁnder to build the virtual library. Dashed lines are used to indicate unspeciﬁed bonds and the aldehyde group is placed at 3-position with respect to N.

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2471

Additional constraints were applied on the aldehydes retrieved

by applying post-ﬁlters. The ﬁrst ﬁlter applied was a one which re-

moves aldehydes having less than two hydrogen-bond acceptors.

This was based on the fact that the hydrogen bond acceptor feature

which exists in the quinolinone (oxygen of the keto-enol) system is

essential for interaction with Tyr448 and Gly449 residues. This was

added to the second acceptor which exists by default taking into

consideration that the retrieved hits are aldehydes (i.e., carbonyl

group of the aldehyde). The ﬁltered hits were further modiﬁed

by substituting the alkyl group at N by isoamyl group in order uni-

fy the hydrophobic tail among the candidates.

The second ﬁlter applied was a pharmacophore ﬁlter which

keeps only aldehydes which have the proper shape and orientation

of hydrogen bond acceptor. This was done using Accerlys Discov-

ery studio catalyst module. First, the quinolinone substructure

was used as a template for creating the shape query and assigning

the proper position to the other essential features. Based on the

reaction selected to make the focused library, the benzothiadizine

and quinolinone were dissected along the bond joining the ring B

and C such that aldehyde group was added to the quinolinone.

After that, an aldehyde custom feature was created such that it

maps C and O atoms of the aldehyde where the number of hydro-

gen atoms suitable for substitution at C was assigned one. Finally,

the query was created as shown in the Figure 9a manually such

that A represents the hydrogen-bond acceptor feature, B represents

the aldehyde feature, C represents the hydrophobic tail and D rep-

resents shape of the ligand. The ﬁltered aldehydes were used to

construct the virtual library by reacting them with orthanilamides

that were selected from HCV literature such that they have a short

polar substituent at position 7 (e.g., methansulfonamide).

Figure 10. Docking of the hit found using virtual screening in the palm I site.

Docking was performed on the homological model of genotype 4 which shows a

mutation of Met414 to Val414. The methansulfonamide can form H-bond with

Asn291 and Asp318.

II. Simple molecular dynamics simulation pointed out to the for-

mation of an internal hydrogen bond between indole OH and

benzothiadiazine N2. This gives some rigidity to the structure

and increase propensity of the desired conformation.

III. Indole benzene ring is aromatic and can form sigma–pi

interaction with Met414.

IV. The methansulfonamide group forms a network of H-bonds

with Asn291 and Asp318 that can compensate loss of the

activity due to Met414 mutation in genotype 4.

V. Topological polar surface area (TPSA) of the ligand is 146

which is near the desired range. This may improve the phar-

macokinetic properties of this class which is known to suffer

from poor bioavailability due to the large PSA.

VI. The aldehyde fragment, which formed the hit, showed per-

fect mapping with the pharmacophore and perfect align-

ment with the reference as shown in the Figure 9b.

3.1.4. Virtual screening

The Virtual library was screened using the model and one of the

top hits (an indole based hit) was selected to be further explored

through experimental validation (another hits were found but we

have chosen to explore the hit which is considered in itself a

new scaffold if we considered that we are rescaffolding the quinoli-

none). The hits retrieved and methods deployed are exempliﬁed in

Section 1.15 in the Supplementary data.

Further assessment of this compound was speciﬁcally carried

out by synthesis and biological evaluation which proved its activity

(IC₅₀against polymerase enzyme was 0.3 lM).

This mentioned hit was selected because of the following rea-

sons (Fig. 10):

3.1.5. Focused library generation

In order to validate the model experimentally, a small focused

library of ligands was synthesized based on the hit retrieved and

was further biologically evaluated. The library members are listed

I. Hydroxyl group at the 4 position of the indole makes H-bond

directly with Tyr448 and indirectly with both Gly449 and

Asp 446 via water molecules.

Figure 9. (A) The manual pharmacophore used to ﬁlter aldehydes which were used to construct the virtual library. Feature A represents H-bond acceptor, feature B

represents aldehyde, Feature C represents hydrophobic feature which corresponds to the tail and Feature D represents the shape feature; (B) mapping of the hit with the

quionlinone shows good superposition of the main features.

2472

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

in Scheme 3 as compounds III (a–o) and Scheme 4 as compounds

IV (a–f). The library was created while aiming at further decreasing

the topological polar surface (TPSA) in order to increase the bio-

availability, therefore, the hydroxyl group on the 4-position of

the indole ring was omitted depending on the following theories.

We will deal with classic keto-enol system of 4-hydroxyquinolin-

2(1H)-one as a reference. The keto-enol system seems to be

necessary for the formation of internal hydrogen bonding with

the two nitrogen atoms of the 1,1 dioxo-4H-1,2,4-benzothiadiazine

(Fig. S22 in Supplementary data). This was validated in literature

using ab-initio technique.^61,62The internal hydrogen bonding re-

sulted in a 20° tilt between the two systems (hydroxyquinolinone

and benzothiadiazine) (Fig. 5) that leads to the positioning of the

hydrophobic tail (e.g., isoamyl tail) in the correct way necessary

for interaction with the hydrophobic pocket (pocket II) amino acid

residues. Despite the above facts, the ligands designed below were

based on bypassing the internal hydrogen bonding depending on a

Hartree–Fock conformational sampling of the ligand. It proved that

the ideal conformer, which can achieve proper binding, is a local

minimum that only differs by 0.9 Kcal/mol from the global mini-

mum which can interact similarly but less efﬁcient (due to the ﬂip-

ping pattern) (Fig. 11) (see Section 1.16 in the Supplementary data

for complete details).

mide was protected by fusion with urea to give 2H-1,2,4-benzo-

thiadiazin-3(4H)-one 1,1-dioxide which was nitrated, further

de-protected and reduced to give 2,5-diaminobenzenesulfona-

mide.This was alkylated using methansulfonyl chloride to

give 2-amino-5-methanesulfonylamino-benzenesulfonamide. This

pathway has many advantages to the other procedures mainly

used to synthesize this important intermediate (see Supplemen-

tary data for details of other procedures of I-c synthesis in Sec-

tion 1.17): It is very economical, yet effective modiﬁed method,

it avoids drastic reagents like chlorosulfonyl isocayante, harsh

conditions like Friedel–Crafts acylation at ꢀ40 °C. In addition;

it avoids sulfonation reaction of p-nitroanline which can lead to

polymerization reaction if conditions are slightly uncontrolled.⁶³

2-Amino-5-bromo-pyridine-3-sulfonamide, an analogue to

orthanilamide in Scheme 4, was prepared starting from 2-amino-

pyridine. This was carried out by brominating 2-aminopyridine

to give 2-amino-5-bromopyridine which was chlorosulfonated

and further reacted with ammonia to give 2-amino-5-bromo-pyr-

idine-3-sulfonamide.

Substituted indole carbaldehydes were synthesized according

Scheme 2. The reaction starts by indoles formylation via Vilsme-

ier–Haack reaction using DMF and phosphorus oxychloride mix-

ture for in situ formation of the formylating agent. This was

followed by N-alkylation using various alkyl halides. Substituted

5-bromo indole carbaldehydes were synthesized starting by in-

doles via bromination in three steps. First, indole was converted

into Sodium Indoline-2-sulfonate using sodium bisulﬁte. Second,

it was acetylated using acetic anhydride to give sodium 1-acety-

lindoline-2-sulfonate.Third; bromination was carried out on this

protected indole followed by deprotection using sodium hydrox-

ide. The resulting 5-bromo indole was further formylated and

substituted as above.

Further validation was done by embedding the synthesized

compounds in a library comprising of 1000 zinc decoys and 98 ac-

tive compounds used in the study. The library was screened using

the combined workﬂow and it was found that the synthesized

compounds are in the top 10% of the ranked library together with

the actives.

3.2. Synthesis

The synthesis of the target compounds was carried out using

a one pot reaction of orthanilamide⁵⁹(or one of its derivatives)

and the corresponding aldehyde. This reaction is efﬁcient for

easy building up of combinatorial libraries of this class of com-

pounds. Orthanilamides were synthesized according to Schemes

1 and 4. 2-Aminobenzenesulfonamide (I-a) was prepared starting

from the commercially available o-nitrobenzensulfonylchloride

which was reacted with ammonia to give the amide that was

further reduced by catalytic hydrogenation. Compound (I-b)

was synthesized via direct bromination of the orthanilamide

(I-a) using bromine/acetic acid while compound (I-d) was syn-

thesized via direct iodination using NIS (N-iodosuccinimide).

Compound (I-c) was prepared according to a modiﬁed procedure

starting from the compound (I-a). In this procedure, orthanila-

The synthesis of the target compounds III (a–o) was carried

out using the one pot reaction of orthanilamide (or one of its

derivatives) and the corresponding aldehyde. This was done by

heating the mixture in dimethylacetamide in the presence of so-

dium bisulﬁte at 150 °C. On the other hand, the target com-

pounds IV (a–f) were synthesized via modiﬁed procedure

which uses pressure tube in addition to the aforementioned

procedure.

3.3. Biological evaluation

Biological evaluation was carried out for the synthesized com-

pounds using replicon assay against genotype 1b (Table 12).⁶⁴It

was found that biological activities of the above mentioned

Figure 11. Conformational analysis of the indole derivative revealing local and global minima. Local minimum is at angle 159 while global minimum is at angle 22. The

difference between local and global minimum is 0.9 Kcal/mol. Analysis was carried out by Spartan 2010 using Hartee-Fock 3-21G in vacuum while letting the torsional angle

to span 180°.

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2473

Scheme 1. Reagents and conditions (a) NH₃; (b) Pd/C, NH₂NH₂, reﬂux,2 h; (c) Br₂, glacial acetic acid; (d)NH₂CONH₂,180 °C; (e) HN0₃/H₂SO₄, 20 h; (f) 50% H₂S0₄,130 °C; (g) (i)

Pd/C, H₂, THF, 5 h, (ii) CH₃SO₂Cl, acetonitrile, pyridine, 20 h; (h) NIS, CHCl₃, reﬂux, 24 h.

Scheme 2. Reagents and conditions (a) (i) POCl₃, DMF, 1 h, 0 °C; (ii) alkyl halide, NaH, DMF,4 h; (b) (i) NaHSO₃, (ii) acetic anhydride, (iii) Br₂; (c) (i) POCl₃, DMF, 1 h, 0 °C; (ii)

alkyl halide, NaH, DMF, 4 h.

compounds comply with the model described. The most active li-

gands were having isoamyl substituent as a hydrophobic tail and

methansulfonamide in the pocket III. Methansulfonamide capabil-

ity to form extensive hydrogen bonding network with pocket III

amino acids is illustrated in Figure S26 in the Supplementary

data. Additionally, in this investigation we explored the effect of

substituting methansulfonamide with halogen and see the effect

of halogen bonding (see Figs. S27 and S28 in the Supplementary

data). It was obvious from the biological data that these com-

pounds are of equivalent potency but with higher toxicity. To

conﬁrm the activity mechanistically, ligand III-m was selected

for polymerase enzymatic inhibition assay and its IC₅₀was found

system and that 4-hydroxy substituted indole was more potent

due to the importance of a hydrogen bond acceptor at this posi-

tion. This concluded one of the main aims of the article which is

drifting away from the chemotype trap using guided structure-

based models.

The assay was performed, in addition, on genotype 4 and all

compounds showed almost no activity except those with meth-

ansulfonamide group that showed moderate inhibition (60%) at

10 lM. The reason can be due to the hydrogen bonding network

which is formed by the methansulfonamide group that compen-

sates the loss of the activity due to Met414 mutation. This phe-

nomenon actually conﬁrms the computational studies carried

out on the mutation effect where the molecular dynamics (using

radial distribution function on the sigma–pi interaction between

to be 1.2

lM. In our opinion, the most important conclusion is

that indole proved to be a good alternative for the quionolinone

2474

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

Scheme 3. Reagents and conditions (a) NaHSO₃, DMA, 150 °C, 2 h.

Scheme 4. (a) Br₂, acetic acid; (b) (i) ClSO₃H, 160 °C, 3 h, (ii) NH₃; (C) NaHSO₃, DMA, 150 °C, pressure vessel.

Table 12

tion binding energy method which was carried out using

Abagyan and Tortov algorithm proved the same concept.

Allover, one can deduce that the model was used in this study to

retrieve ligands which are likely to be active because of the

pharmacophoric and binding pattern similarity they share with

the main active compounds.

Biological data of various synthesized compounds using replicon assay

against genotype 1b

Compound

CC₅₀

(l

M)

EC₅₀(lM)

III-a

III-b

III-d

III-e

III-f

III-i

III-j

III-o

V-a

>50

37.2

49.2

>50

16.6

21.9

>50

ND

4.15

8.01

1.86

19.7

18.4

9.9

4. Conclusion

7.7

In summary, we provided a comprehensive workﬂow to study

benzothiadiazines as important HCV NS5b polymerase inhibitors

acting on palm I allosteric site. The workﬂow is a multipurpose

one that can be used for:

25.4

2.96

5.89

20.2

7.07

V-b

V-d

V-e

V-f

1. SAR analysis of this class using a combined approach by apply-

ing Topomer CoMFA (q²= 0.77), AMANDA (same q2) and recur-

sive partitioning (ROC = 0.746).

ring A and Val414) proved the less frequent interaction (see

Section 1.13 in the Supplementary data). Additionally, the muta-

2. Binding mode prediction using Guided docking.

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2475

3. Activity estimation using PLP scoring function which was

proved to correlate well with activity (q²= 0.65 and after opti-

mization of the whole protocol gave q²= 0.8).

5.1.1.2. Compounds II (a–j). Procedures of synthesis are men-

tioned in Supplementary data in Section 1.20.2.2 according to

Scheme II.

4. Virtual screening enrichment by using consensus of both inter-

action and template similarity scores which was validated by

ROC analysis (AUC = 0.91).

5. Genotypes proﬁling using linear interaction energy to calculate

binding energy of the ligand with different genotypes and

mutation binding energy calculation.

5.1.1.3. 2-Amino-5-bromopyridine. 28.2 g (0.3 moles) of 2-ami-

nopyridine were dissolved in 50 ml of acetic acid. The solution is

cooled to below 20° by immersion in an ice bath, and 48 g

(15.4 ml, 0.3 moles) of bromine dissolved in 30 ml of acetic acid

is added dropwise with vigorous stirring over a period of 1 h. Ini-

tially the temperature is maintained below 20°. After half the bro-

mine solution has been added, it is allowed to rise to 50° to delay as

long as possible the separation of the hydrobromide of 2-amino-5-

bromopyridine. At 50° the hydrobromide usually begins to crystal-

lize when about three-quarters of the bromine has been added.

When addition of bromine is completed, the mixture is stirred

for 1 h and is then diluted with 75 ml of water to dissolve the

hydrobromide. The contents of the ﬂask are transferred to a

500 ml beaker and are neutralized, with stirring and cooling,

by the addition of 120 ml of 40% sodium hydroxide solution.

2-Amino-5-bromopyridine, contaminated with some 2-amino-3,5-

dibromopyridine,was ﬁltered and dried. The 2-amino-3,5-dib-

romopyridine is removed from the product by washing with three

500-ml. portions of hot petroleum ether. The yield of 2-amino-5-

bromopyridine, is 32–34.7 g (62–67%). mp 134 °C (literature⁶⁵

mp 132–135°C)

6. R-screening using validated Topomer CoMFA and general ligand

screening using pentacle model or recursive partitioning.

Extensive validation was carried out for this model using both

statistical and experimental methods. Statistically, cross valida-

tion was used throughout this study to validate different correla-

tions. Experimentally, model was validated by using it as a tool

for fragment-hopping. A virtual library, enumerated on reaction

basis, was created such that it complies with SAR analysis. It

was further screened and one of the top scoring hits (indole-

based) was further investigated by considering it a template for

focused library creation. Synthesis of this focused library was fol-

lowed by biological evaluation which proved the activity of the

indole-based derivatives (most active ligand showed activity of

1.86 lM).

Regarding synthesis, one step reaction was adopted to easy syn-

thesize this class of compounds and a facile synthesis of the 2-ami-

no-5-methanesulfonylamino-benzenesulfonamide was provided

in details due to its importance as an intermediate in synthesis

of HCV inhibitors.

5.1.1.4. 2-Amino-5-bromo-pyridine-3-sulfonamide. It was pre-

pared according to the literature method.⁶⁶

In our opinion, this workﬂow can be used generally to study

congeneric series of drugs. It uses ﬁeld-alignment followed by

guided docking. The protocol described provides a mean to gener-

ate a receptor-based QSAR model which relies on the ﬁeld align-

ment in decreasing the RMSD between ligands in the binding site

to the degree that permits ﬁnding a correlating scoring function.

It is worth to validate this workﬂow on other datasets as its appli-

cability domain is generally wider than that of the ligand-based

approaches.

5.1.2. Target compounds synthesis

5.1.2.1. Compounds III (a–o). They were synthesized according to

Scheme III. This was carried out using the general procedure pro-

posed by Imai et al.⁵⁹

General procedure: sodium hydrogen sulﬁte (0390 g 3.75 mmol)

is added to a solution of o-aminobenzenesulfonamide (0.430 g,

2.5 mmol) and the aldehyde (2.5 mmol) in dimethyl acetamide

(3 ml). The mixture is heated with stirring at 150 °C for 2 h and

then poured into water (500 ml). A precipitate forms which is col-

lected and dried (recrystallized is generally from ethanol or any

appropriate alcohol).

5. Experimental data

5.1.2.1.1. 3-(1-Allyl-1H-indol-3-yl)-7-bromo-4H-pyrido[2,3-e]

[1,1270 2,4] thiadiazine 1,1-dioxide (IV-a). Yield: 50%. ¹H NMR

(300 MHz, DMSO) d = 11.58 (s, 1H), 8.48 (d, J = 3.2, 1H), 8.39 (dt,

J = 6.6, 2.7, 1H), 7.82 (dt, J = 7.8,2,1H), 7.76–7.20 (m, 6H), 6.21–

6.00 (m,1H), 5.35–5.12 (m,2H), 4.98(dq, J = 5.5, 2.3, 1.9, 2H). ¹³C

NMR (75 MHz, DMSO) d = 141.9, 136.1, 134.6, 133.3, 129.2, 127.3,

124.2, 122.8, 120.5, 119.8, 117.6, 114.5, 110.6, 100.2, 49.3. HRMS

calcd for C₁₈H₁₅N₃O₂S: 337.0885; found: 337.0873. Anal. Calcd

for C₁₈H₁₅N₃O₂S: C, 64.08; H, 4.48; N, 12.45. Found: C, 64.12; H,

4.46; N, 12.43.

5.1.2.1.2. 3-(1-Propyl-1H-indol-3-yl)-4H-1,2,4-benzothiadiazine

1,1-dioxide (III-b).Yield: 60%. ¹H NMR (300 MHz, DMSO) d = 11.70

(s, 1H), 8.51 (d, J = 2.5, 1H), 8.39–8.30 (m, 1H), 7.98–7.83 (m, 3H),

7.67 (d, J = 7.6, 1H), 7.51 (dd, J = 8.8,2.4,1H), 7.32 (ddd, J = 7.2,5,

1.9,2H), 4.28 (t, J = 7.3, 2H), 1.89 (q, J = 7.5, 2H), 0.91 (t, J = 7.4,3H).

¹³C NMR (75 MHz, DMSO) d = 141.7, 136.7, 134.4, 127.3, 127.9,

125.2, 124.9, 122.3, 120.9, 119.4, 117.8, 110.1, 98.1, 52.0, 21.0,

11.9.HRMS calcd for C₁₈H₁₇N₃O₂S: 339.1041; found: 339.1053.

Anal. Calcd for C₁₈H₁₇N₃O₂S: C, 63.70; H, 5.05; N, 12.38. Found: C,

63.73; H, 5.02; N, 12.35

Molecular modeling and biological testing experimental proce-

dures are mentioned in Sections 1.20.1 and 1.20.3 in the Supple-

mentary data.

5.1. Synthesis

General methods. Starting materials were either commercially

available or prepared as reported previously in the literature, un-

less otherwise noted. Solvents and reagents were used without fur-

ther puriﬁcation. Reactions were monitored by TLC, performed on

silica gel glass plates containing F-254 indicator (Merck).Visualiza-

tion on TLC was achieved by UV. Proton and carbon NMR spectra

were recorded on a Bruker ARX-300. Chemical shifts (ä) are re-

ported in ppm downﬁeld from internal TMS standard or from sol-

vent references. Mass spectra were recorded on API-SCIEX 2000.

HRMS were recorded using an Agilent MSD-TOF (G1969A) con-

nected to an Agilent 1100 HPLC system. Melting points were deter-

mined on a Stuart melting point apparatus (Stuart Scientiﬁc,

Redhill, UK) and are uncorrected.

5.1.2.1.3.3-(1-Isopropyl-1H-indol-3-yl)-4H-1,2,4-benzothiadiazine

1,1-dioxide (III-c). Yield: 65%. ¹H NMR (300 MHz, DMSO)

d = 11.40 (s, 1H), 8.54 (d, J = 1.4, 1H), 8.35 (m, 1H), 7.95–7.62 (m,

3H), 7.54–7.23 (m, 4H),4.90 (m, 1H), 1.37 (d, J = 6.0, 6H). ¹³C

NMR (75 MHz, DMSO) d = 142.3, 136.9, 135.8, 135.1, 133.1, 129.2,

127.1, 125.2, 124.1, 120.5, 120.2, 119.3, 116.9, 112.4, 97.8, 53.9,

5.1.1. Intermediates synthesis

5.1.1.1. Compounds I (a–o). Procedures of synthesis are men-

tioned in Supplementary data in Section 1.20.2.1 according to

Scheme I.

2476

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

18.8.HRMS calcd for C₁₈H₁₇N₃O₂S: 339.1041; found: 339.1032.

Anal. Calcd for C₁₈H₁₇N₃O₂S: C, 63.70; H, 5.05; N, 12.38. Found:

C, 63.73; H, 5.03; N, 12.36

(dd, J = 7.4,1.5),7.62–7.38 (m,3H),7.13–7.09 (m,2H), 4.03 (t, J = 7.3,

2H), 1.79 (p, J = 7.1, 2H), 1.31 (m, 2H), 0.93 (t, J = 7.3, 3H). ¹³C

NMR (75 MHz, DMSO) d = 143.6, 135.1, 135.4, 134.9, 139.6, 126.5,

127.1, 123.9, 123.1, 118.9, 118.1, 112.3, 112.6, 97.6, 49.8, 28.7,

20.7, 14.0. HRMS calcd for C₁₉H₁₈Br(79)N₃O₂S: 431.0303; found:

5.1.2.1.4.

3-(1-Butyl-1H-indol-3-yl)-4H-1,2,4-benzothiadiazine

1,1-dioxide (III-d). Yield: 56%. ¹H NMR (300 MHz, DMSO)

d = 11.58 (s, 1H), 8.52 (d, J = 1.4, 1H), 8.36 (dt, J = 6.3, 1.7, 1H),

7.87–7.62 (m, 3H), 7.58–7.23(m, 4H), 4.33 (t, J = 7.0, 2H), 1.84 (p,

J = 7.1, 2H), 1.32 (p, J = 7.4, 2H), 0.93 (td, J = 7.5, 1.3, 3H).¹³C NMR

(75 MHz, DMSO) d = 142.2, 136.4, 135.6, 127.8, 127.7, 124.8,

124.8, 123.4, 121.6, 115, 118.6, 110.4, 98.1, 49.8, 28.7, 20.7, 14.0.

HRMS calcd for C₁₉H₁₉N₃O₂S: 353.1198; found: 353.1188. Anal.

Calcd for C₁₉H₁₉N₃O₂S: C, 64.57; H, 5.42; N, 11.89. Found: C,

64.59; H, 5.40; N, 11.88

431.0310; calcd for

C₁₉H₁₈Br (81) N3O2S:433.0283 found

433.0273. Anal. Calcd for C₁₉H₁₈BrN₃O₂S: C, 52.78; H, 4.20; N,

9.72. Found C, 52.80; H, 4.18; N, 9.70.

5.1.2.1.11. 7-Bromo-3-(1-propyl-1H-indol-3-yl)-4H-1,2,4-benzo-

thiadiazine 1,1-dioxide (III-k). ¹H NMR (300 MHz, DMSO) d =

11.70 (s, 1H), 8.51 (d, J = 2.5, 1H), 8.39–8.30 (m, 1H), 7.71–7.36

(m, 5H), 7.15 (d, J = 7.4, 1H), 4.20 (t, J = 7.3, 2H), 1.93 (q, J = 7.3,

2H), 0.98 (t, J = 7.3, 3H). ¹³C NMR (75 MHz, DMSO) d = 141.9,

136.4, 134.7, 131.2, 129.2, 127.8, 125.1, 122.9, 121.5, 120.6,

5.1.2.1.5. 3-(1-Benzyl-1H-indol-3-yl)-4H-1,2,4-benzothiadiazine

1,1-dioxide (III-e). Yield: 45%. ¹H NMR (300 MHz, DMSO)

d = 11.64 (s, 1H), 8.59 (d, J = 3.6, 1H), 8.39 (dd, J = 6.0, 3.5, 1H),

7.82 (m,1H), 7.75–7.58(m,2H), 7.57–7.23 (m, 9H) 5.59 (d, J = 3.5,

2H). ¹³C NMR (75 MHz, DMSO) d = 141.9, 137.1, 136.1, 134.5,

131.4, 128.5, 127.9, 127.1, 124.2, 122.7, 120.4, 119.8, 117.6,

110.8, 100.4, 52.6. HRMS calcd for C₂₂H₁₇N₃O₂S: 387.1041; found:

387.1049. Anal. Calcd for C₂₂H₁₇N₃O₂S: C, 68.20; H, 4.42; N, 10.85.

Found: C, 68.23; H, 4.40; N, 10.83.

5.1.2.1.6. 3-[1-(3-Methylbutyl)-1H-indol-3-yl]-4H-1,2,4-benzothi-

adiazine 1,1-dioxide (III-f). Yield: 60%. ¹H NMR (300 MHz, DMSO)

d = 11.58 (s, 1H), 8.51 (d, J = 1.4, 1H), 8.35 (dt, J = 6.3, 1.7, 1H), 7.98

(dd, J = 7.5, 1.5, 1H), 7.63–7.36 (m, 4H), 7.24–7.08 (m, 2H), 0.98 (d,

J = 7, 6H), 1.56 (m, 2H), 1.74 (m, 1H), 4.16 (m, 2H). ¹³C NMR

(75 MHz, DMSO) d = 143.1, 137.3 (s), 136.1, 129.1, 127.1, 126.3,

125.3, 128.9, 121.2, 118.9, 118.2, 110.4, 97.9, 48.9, 36.9, 27.1,

22.7.HRMS calcd for C₂₀H₂₁N₃O₂S: 367.1354 found: 367.1345.

Anal. Calcd for C₂₀H₂₁N₃O₂S: C, 65.37; H, 5.76; N, 11.44. Found:

C, 65.39; H, 5.74; N, 11.42

119.2, 110.4, 98.1, 52.0, 21.0, 11.9.HRMS calcd for

C₁₈H₁₆-

Br(79)N₃O₂S: 417.0147; found 417.0143, C₁₈H₁₆Br(81)N₃O₂S:

419.0126; found: 419.0134. Anal. Calcd for C₁₈H₁₆BrN₃O₂S: C,

51.68; H, 3.86; N, 10.05. Found: C, 51.69; H, 3.84; N, 10.03.

5.1.2.1.12. 7-Iodo-3-(1-propyl-1H-indol-3-yl)-4H-1,2,4-benzothi-

adiazine 1,1-dioxide (III-l). Yield: 60%. ¹H NMR (300 MHz, DMSO)

d = 11.58 (s, 1H), 8.52 (d, J = 1.4, 1H), 8.41 (d, J = 1.7,1H), 8.36 (dt,

J = 6.3, 1.7, 1H), 7.78 (dd, J = 7.5,1.5,1H), 7.63–7.36 (m, 3H), 6.96

(d, J = 7.5,1H), 4.26 (t, J = 7.3,2H),1.90 (q, J = 7.5,2H), 0.90(td,

J = 7.4,2.3,3H).¹³C NMR (75 MHz, DMSO) d = 148.7, 141.9, 134.7,

134.2, 127.8, 125.2, 124.6, 122.9, 121.2, 120.6, 110.4, 103.3, 98.1,

52.0, 21.0, 11.9. HRMS calcd for C₁₈H₁₆IN₃O₂S:465.0008; found:

465.0015. Anal. Calcd for C₁₈H₁₆IN₃O₂S: C, 46.46; H, 3.47; N,

9.03. Found: C, 46.43; H, 3.49; N, 9.05.

5.1.2.1.13. N-{3-[1-(3-Methylbutyl)-1H-indol-3-yl]-1,1-dioxido-

4H-1,2,4-benzothiadiazin-7-yl}methansulfonamide (III-m). Yield:

46%. ¹H NMR (300 MHz, DMSO) d = 11.55 (s, 1H), 9. 21 (s, 1H), 8.52

(d, J = 1.4, 1H), 8.36 (dt, J = 6.3, 1.7, 1H), 7.63–7.29 (m, 4H), 7.21 (d,

J = 1.5, 1H), 7.11 (d, J = 7.5, 1H), 4.30 (m, 2H), 1.89–1.58 (m, 3H),

0.91(d, J = 6.2,6H). ¹³C NMR (75 MHz, DMSO) d = 141.4, 135.5,

134.7, 128.6, 127.8, 125.2, 123.3, 122.9, 120.9–120.4 (m), 113.8,

110.4, 98.1, 49.7, 42.9, 36.3, 27.1, 22.7.HRMS calcd for C₂₁H₂₄N₄O₄S₂:

460.1239; found:460.1224. Anal. Calcd for C₂₁H₂₄N₄O₄S₂: C, 54.76;

H, 5.25; N, 12.16. Found: C, 54.77; H, 5.23; N, 12.14

5.1.2.1.7. 3-(1-Benzyl-2-methyl-1H-indol-3-yl)-4H-1,2,4-benzothi-

adiazine 1,1-dioxide (III-g). Yield: 55%. ¹H NMR (300 MHz, DMSO)

d = 11.58 (s, 1H), 8.39 (dt, J = 6.6, 2.7, 1H), 7.80–7.2(m, 12H), 5.59

(s, 1H), 2.48 (m, 3H). ¹³C NMR (75 MHz, DMSO) d = 143.3, 140.9,

136.9, 136.1, 134.6, 129.0, 128.4, 128.0, 127.1, 124.2, 123.8,

120.3, 119.8, 119.4, 117.6, 103.8, 103.0, 47.9, 14.2. HRMS calcd

for

C

₂₃H₁₉N₃O₂S:401.1198; found: 401.1185. Anal. Calcd for

5.1.2.1.14. N-[3-(1-Benzyl-1H-indol-3-yl)-1,1-dioxido-4H-1,2,4-

benzothiadiazin-7-yl]methansulfonamide (III-n). Yield: 54%. ¹H

NMR (300 MHz, DMSO) d = 11.64 (s, 1H), 9.13 (s,1H), 8.59 (d,

J = 3.6, 1H), 8.39 (dd, J = 6.0, 3.5, 1H), 7.69–7.55 (m,1H), 7.43–7.18

(m, 9H), 7.10(d, J = 7.5, 1H), 5.56 (s, 2H), 3.22 (s, 3H). ¹³C NMR

(75 MHz, DMSO) d = 141.9, 137.1, 135.5, 134.4, 131. 3,

128.7–128.3(m), 127.9, 127.0, 123.3, 122.7, 120.6–120.2 (m),

113.8, 110.8, 100.4, 52.6, 42.9. HRMS calcd for C₂₃H₂₀N₄O₄S₂:

480.0926; found: 480.0943. Anal. Calcd for C₂₃H₂₀N₄O₄S₂: C,

57.48; H, 4.19; N, 11.66. Found: C, 57.49; H, 4.17; N, 11.65

C₂₃H₁₉N₃O₂S: C, 68.81; H, 4.77; N, 10.47. Found: C, 68.84; H,

4.75; N, 10.46.

5.1.2.1.8. 3-(1-Benzyl-5-bromo-1H-indol-3-yl)-4H-1,2,4-benzothi-

adiazine 1,1-dioxide (III-h). Yield: 70%. ¹H NMR (300 MHz, DMSO)

d = 11.64 (s, 1H), 8.59 (d, J = 3.6, 1H), 8.39 (dd, J = 6.0, 3.5, 1H), 7.99

(dd, J = 7.5, 1.5, 1H), 7.84 (d, J = 7.5, 1H), 7.71–7.51 (m,2H), 7.40–

7.08 (m,7H), 5.59 (d, J = 3.5, 2H). ¹³C NMR (75 MHz, DMSO)

d = 141.8, 137.3, 136.3, 135.4, 134.6, 132.7, 129.6, 128.9, 127.8,

127.7, 124.3, 124.6, 122.7, 119.5, 117.7, 112.6, 110.9, 99.0, 52.6.

HRMS calcd for C₂₂H₁₆Br(79)N₃O₂S: 465.0147; found: 465.0135;

calcd for C₂₂H₁₆Br(81)N₃O₂S: 467.0126; found: 467.0135. Anal.

Calcd for C₂₂H₁₆BrN₃O₂S: C, 56.66; H, 3.46; N, 9.01. Found: C,

56.68; H, 3.45; N, 9.00.

5.1.2.1.9. 3-[5-Bromo-1-(3-methylbutyl)-1H-indol-3-yl]-4H-1,2,4-

benzothiadiazine 1,1-dioxide (III-i). Yield: 55%. ¹H NMR (300 MHz,

DMSO) d = 11.58 (s, 1H), 8.52 (d, J = 1.4, 1H), 8.36 (dt, J = 6.3, 1.7,

1H), 7.99 (dd, J = 7.4,1.5), 7.72–7.48 (m, 3H), 7.23–7.08 (m, 2H),

4.16 (m, 2H), 1.74 (m, 1H),1.56 (m, 2H), 0.98 (d, J = 7, 6H). ¹³C

NMR (75 MHz, DMSO) d = 142.3, 136.5, 135.6, 134.6, 130.1, 127.1,

126.6, 124.2, 122.9, 119.8, 117.6, 111.9, 111.6, 97.4, 49.7, 36.3,

27.1, 22.7.HRMS calcd for C₂₀H₂₀Br(79)N₃O₂S: 445.0460; found:

5.1.2.1.15.

N-[3-(5-Bromo-1-butyl-1H-indol-3-yl)-1,1-dioxido-

4H-1,2,4-benzothiadiazin-7-yl]methansulfonamide (III-o). Yield:

60%. ¹H NMR (300 MHz, DMSO) d = 11.58 (s, 1H), 9.21 (s,1H),

8.52 (d, J = 1.4, 1H), 8.36 (dt, J = 6.3, 1.7, 1H), 7.67 (d, J = 1.2, 2H),

7.46–7.34 (m, 1H), 7.28 (d, J = 1.5, 1H), 7.12 (d, J = 7.5, 1H), 4.31

(t, J = 7.7, 2H), 1.82 (p, J = 7.8, 2H), 1.31 (tq, J = 7.7, 6.5, 2H), 0.89

(t, J = 6.6, 3H). ¹³C NMR (75 MHz, DMSO) d = 141.3, 136.1, 136.4,

130.1, 128.9, 126.4, 123.9, 123.1, 121.4, 121.5, 112.9, 110.9,

111.6, 97.4, 49.8, 42.9, 28.7, 20.7, 14.0. HRMS calcd for:

C₂₀H₂₁Br(79)N₄O₄S_2:524.0188; found:524.0174;calcd for C₂₀H₂₁-

Br(81)N₄O₄S₂:526.0167; found: 526.0175. Anal. Calcd for

C₂₀H₂₁BrN₄O₄S₂: C, 45.72; H, 4.03; N, 10.66. Found: C, 45.74; H,

4.01; N, 10.64

445.3615,calcd for

C₂₀H₂₀Br(81)N₃O₂S: 447.0439; found:

447.0446. Anal. Calcd for C₂₀H₂₀BrN₃O₂S: C, 53.82; H, 4.52; N,

9.41. Found: C, 53.85; H, 4.50; N, 9.40.

5.1.2.1.10. 3-(5-Bromo-1-butyl-1H-indol-3-yl)-4H-1,2,4-benzothi-

adiazine 1,1-dioxide (III-j). Yield: 60%. ¹H NMR (300 MHz, DMSO)

d = 11.58 (s, 1H), 8.51(d, J = 1.5, 1H), 8.34 (dt, J = 6.2, 1.6, 1H), 7.81

5.1.2.2. Compounds IV (a–j). They were synthesized according to

the following general procedure.

Sodium hydrogen sulﬁte (0.390 g, 3.75 mmol) is added to a

solution of 2-amino-5-bromo-pyridine-3-sulfonamide (0.59 g,

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

2477

2.5 mmol) and the aldehyde (2.5 mmol) in dimethyl acetamide

(3 ml). The mixture is heated with stirring at 150° for 2 h and then

poured into water (500 ml). A precipitate forms which is collected

and dried (recrystallization is generally from ethanol or any appro-

priate alcohol).

C

₁₉H₁₉Br(79)N₄O₂S: 446.0412; found: 446.0408, calcd for

₁₉H₁₉Br(81)N₄O₂S: 448.0392; found: 448.0388. Anal. Calcd for

₁₉H₁₉BrN₄O₂S: C, 51.01; H, 4.28; N, 12.52. Found: C, 51.03; H,

4.26; N, 12.50.

5.1.2.2.1 3-(1-Allyl-1H-indol-3-yl)-7-bromo-4H-pyrido[2,3-e][1,

2, 4]thiadiazine 1,1-dioxide (IV-a).

Supplementary data

Yield: 50%. ¹H NMR (300 MHz, DMSO) d = 11.69 (s, 1H), 8.54 (s,

1H), 8.35 (m, 1H), 8.12 (m, 1H), 7.72 (m, 1H), 7.42–7.25 (m, 3H),

6.21–6.00 (m,1H), 5.35–5.12 (m,2H), 4.98 (dq, J = 5.5, 2.3, 1.9,

2H). ¹³C NMR (75 MHz, DMSO) d = 152.0, 141.7, 141.3, 140.5,

134.5, 133.3, 129.2, 127.4, 122.8, 121.2, 120.5, 118.7, 114.5,

110.6, 100.2, 49.3.HRMS calcd for C₁₇H₁₃Br(79)N₄O₂S: 415.9943;

found: 415.9934, calcd for C₁₇H₁₃Br(81)N₄O₂S: 417.9922; found

417.9934. Anal. Calcd for C₁₇H₁₃BrN₄O₂S: C, 48.93; H, 3.14; N,

13.43. Found: C, 48.95; H, 3.12; N, 13.41.

Supplementary data associated with this article can be found, in

the online version, at doi:10.1016/j.bmc.2012.01.031.

References and notes

1. Ikegashira, K.; Oka, T.; Hirashima, S.; Noji, S.; Yamanaka, H.; Hara, Y.; Adachi, T.;

Tsuruha, J.-I.; Doi, S.; Hase, Y.; Noguchi, T.; Ando, I.; Ogura, N.; Ikeda, S.;

Hashimoto, H. J. Med. Chem. 2006, 49, 6950.

2. Gopalsamy, A.; Chopra, R.; Lim, K.; Ciszewski, G.; Shi, M.; Curran, K. J.; Sukits, S.

F.; Svenson, K.; Bard, J.; Ellingboe, J. W.; Agarwal, A.; Krishnamurthy, G.; Howe,

A. Y. M.; Orlowski, M.; Feld, B.; O’Connell, J.; Mansour, T. S. J. Med. Chem. 2006,

49, 3052.

3. Lange, C. M.; Sarrazin, C.; Zeuzem, S. Aliment Pharmacol. Ther. 2010, 32, 14.

4. Parﬁeniuk, A.; Jaroszewicz, J.; Flisiak, R. World J. Gastroenterol. 2007, 13, 5673.

5. Thompson, A. J.; McHutchison, J. G. J. Viral Hepat. 2009, 16, 377.

6. Schinazi, R. F.; Coats, S. J.; Bassit, L. C.; Lennerstrand, J.; Nettles, J. H.; Hurwitz, S.

J. Handb. Exp. Pharmacol. 2009, 189, 25.

7. Flisiak, R.; Parﬁeniuk, A. Expert Opin. Investig. Drugs 2010, 19, 63.

8. Legrand-Abravanel, F.; Nicot, F.; Izopet, J. Exp. Opin. Investig. Drugs 2010, 19,

963.

5.1.2.2.2.

7-Bromo-3-(1-propyl-1H-indol-3-yl)-4H-pyrido[2,3-

e][1,2,4]thiadiazine 1,1-dioxide (IV-b). Yield: 56%. ¹H NMR

(300 MHz, DMSO) d = 11.58 (s, 1H), 8.48 (s, 1H), 8.39 (m, 1H),

8.02 (m, 1H), 7.59 (m, 1H), 7.51–7.23 (m, 3H), 4.15 (t,J = 7.1, 2H),

1.93 (q, J = 7.3,2H), 0.96 (td, J = 7.3,2.3,3H). ¹³C NMR (75 MHz,

DMSO) d = 151.4, 141.8, 142.3, 140.5, 134.7, 127.8, 125.2, 122.9,

121.2, 120.6, 118.7, 110.4, 98.1, 52.0, 21.0, 11.9.HRMS calcd for

C₁₇H₁₅Br(79)N₄O₂S: 418.0099; found: 418.0080, calcd for

C₁₇H₁₅Br(81)N₄O₂S: 420.0079;found: 420.0092. Anal. Calcd for

C₁₇H₁₅BrN₄O₂S: C, 48.70; H, 3.61; N, 13.36; Found: C, 48.72; H,

9. Li, H.; Shi, S. T. Future Med. Chem. 2010, 2, 121.

10. Deore, R. R.; Chern, J. W. Curr. Med. Chem. 2010, 17, 3806.

11. Watkins, W. J.; Ray, A. S.; Chong, L. S. Curr. Opin. Drug Discov. Devel. 2010, 13,

441.

3.60; N, 13.34.

12. Talele, T. T. Curr. Bioact. Compd. 2008, 4, 86.

5.1.2.2.3. 7-Bromo-3-(1-isopropyl-1H-indol-3-yl)-4H-pyrido[2,3-

e][1,2,4]thiadiazine 1,1-dioxide (IV-c). Yield: 75%. ¹H NMR

(300 MHz, DMSO) d = 11.64 (s, 1H), 8.52 (s, 1H), 8.27 (m, 1H),

8.13 (m, 1H), 7.63 (m, 1H), 7.46–7.22 (m, 3H), 4.90 (m, 1H), 1.37

(d, J = 6.0, 6H). ¹³C NMR (75 MHz, DMSO) d = 152.0, 141.7, 141.3,

140.5, 137.3, 132.5, 128.7, 123.9, 121.2, 120.5, 118.6, 111.9, 97.8,

54.1 18.8. HRMS calcd for C₁₇H₁₅Br(79)N₄O₂S: 418.0099; found

418.0080, calcd for C₁₇H₁₅Br(81)N₄O₂S: 420.0079; found:420.0065.

Anal. Calcd for C₁₇H₁₅BrN₄O₂S: C, 48.70; H, 3.61; N, 13.36. Found:

C, 48.71; H, 3.60; N, 13.35.

13. Barreca, M. L.; Iraci, N.; Manfroni, G.; Cecchetti, V. Future Med. Chem. 2011, 3,

1027.

14. Gombar, V. K. 6036349; March 14, 2000.

15. Melagraki, G.; Afantitis, A.; Sarimveis, H.; Koutentis, P. A.; Markopoulos, J.;

Igglessi-Markopoulou, O. Bioorg. Med. Chem. 2007, 15, 7237.

16. Pourbasheer, E.; Riahi, S.; Ganjali, M.; Norouzi, P. Mol. Divers. 2011, 15, 645.

17. Chen, K.-X.; Xie, H.-Y.; Li, Z.-G. Acta Chim. Slov. 2009, 56, 684.

18. Wang, X.; Yang, W.; Xu, X.; Zhang, H.; Li, Y.; Wang, Y. Curr. Med. Chem. 2010, 17,

2788.

19. Weaver, S.; Gleeson, M. P. J. Mol. Graph Model 2008, 26, 1315.

20. Li, T.; Froeyen, M.; Herdewijn, P. J. Mol. Model. 2010, 16, 49.

21. Pauwels, F.; Mostmans, W.; Quirynen, L. M.; van der Helm, L.; Boutton, C. W.;

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

A. J. Virol. 2007, 81, 6909.

22. Palm, K.; Stenberg, P.; Luthman, K.; Artursson, P. Pharm. Res. 1997, 14, 568.

23. Cheeseright, T.; Mackey, M.; Rose, S.; Vinter, A. Exp. Opin. Drug Disc. 2007, 2,

131.

24. Fradera, X.; Mestres, J. Curr. Top. Med. Chem. 2004, 4, 687.

25. Cheeseright, T.; Mackey, M.; Rose, S.; Vinter, A. J. Chem. Inf. Model. 2006, 46,

665.

26. Thomsen, R.; Christensen, M. H. J. Med. Chem. 2006, 49, 3315.

27. Salam, N. K.; Nuti, R.; Sherman, W. J. Chem. Inf. Model. 2009, 49, 2356.

28. Nyanguile, O.; Devogelaere, B.; Vijgen, L.; Van den Broeck, W.; Pauwels, F.;

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

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

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

A.; Fanning, G. C. J. Virol. 2010, 84, 2923.

5.1.2.2.4.

7-Bromo-3-(1-butyl-1H-indol-3-yl)-4H-pyrido[2,3-

e][1,2,4]thiadiazine 1,1-dioxide (IV-d). Yield: 60%. ¹H NMR

(300 MHz, DMSO) d = 11.70 (s, 1H), 8.49 (s, 1H), 8.34 (m, 1H),

8.14 (m, 1H), 7.62 (m, 1H), 7.63–7.36 (m, 3H), 4.12 (t, J = 7.3, 2H),

1.78 (p, J = 7.1, 2H), 1.31 (p, J = 7.4, 2H), 0.93 (t, J = 7.3,3H). ¹³C

NMR (75 MHz, DMSO) d = 152.2, 141.9, 141.5, 140.7, 134.9, 127.6,

124.9, 123.1, 122.2, 120.7, 118.8, 110.4, 98.1, 49.8, 28.7, 20.7,

14.0.HRMS calcd for

432.0243 calcd for

C

₁₈H₁₇Br(79)N₄O₂S: 432.0256 found:

₁₈H₁₇Br(81)N₄O₂S: 434.0235;found:

434.0225. Anal. Calcd for C₁₈H₁₇BrN₄O₂S: C, 49.89; H, 3.95N,

12.93. Found: C, 49.92; H, 3.92; N, 12.91.

29. Kirchmair, J.; Markt, P.; Distinto, S.; Wolber, G.; Langer, T. J. Comput. Aided Mol.

Des. 2008, 22, 213.

30. Triballeau, N.; Acher, F.; Brabet, I.; Pin, J. P.; Bertrand, H. O. J. Med. Chem. 2005,

48, 2534.

31. Aqvist, J.; Medina, C.; Samuelsson, J. E. Protein Eng. 1994, 7, 385.

32. Goodford, P. J. J. Med. Chem. 1985, 28, 849.

33. Duran, A.; Martinez, G. C.; Pastor, M. J. Chem. Inf. Model. 1813, 2008, 48.

34. Cramer, R. D. J. Med. Chem. 2003, 46, 374.

5.1.2.2.5.

3-(1-Benzyl-1H-indol-3-yl)-7-bromo-4H-pyrido[2,3-

e][1,2,4]thiadiazine 1,1-dioxide (IV-e). Yield: 66%. ¹H NMR

(300 MHz, DMSO) d = 11.52 (s, 1H), 8.53 (s, 1H), 8.35(m, 1H),

8.12 (m, 1H), 7.72 (m, 1H), 7.69–7.55 (m, 1H), 7.43–7.20 (m, 7H),

5.56 (s, 2H). ¹³C NMR (75 MHz, DMSO) d = 152.0, 141.7, 141.3,

140.5, 137.1, 134.4, 131.4, 128.5, 127.9, 127.0, 122.7, 121.2,

120.4,

118.7,

110.8,

100.4,

52.6.

HRMS

calcd

for

35. Rusinko, A.; Farmen, M. W.; Lambert, C. G.; Brown, P. L.; Young, S. S. J. Chem. Inf.

Comput. Sci. 1999, 39, 1017.

C

₂₁H₁₅Br(79)N₄O₂S::

466.0099;

found:

466.0088,

calcd

36. Pratt, J. K.; Donner, P.; McDaniel, K. F.; Maring, C. J.; Kati, W. M.; Mo, H.;

Middleton, T.; Liu, Y.; Ng, T.; Xie, Q.; Zhang, R.; Montgomery, D.; Molla, A.;

Kempf, D. J.; Kohlbrenner, W. Bioorg. Med. Chem. Lett. 2005, 15, 1577.

37. Krueger, A. C.; Madigan, D. L.; Jiang, W. W.; Kati, W. M.; Liu, D.; Liu, Y.; Maring,

C. J.; Masse, S.; McDaniel, K. F.; Middleton, T.; Mo, H.; Molla, A.; Montgomery,

D.; Pratt, J. K.; Rockway, T. W.; Zhang, R.; Kempf, D. J. Bioorg. Med. Chem. Lett.

2006, 16, 3367.

38. Rockway, T. W.; Zhang, R.; Liu, D.; Betebenner, D. A.; McDaniel, K. F.;

Pratt, J. K.; Beno, D.; Montgomery, D.; Jiang, W. W.; Masse, S.; Kati, W.

M.; Middleton, T.; Molla, A.; Maring, C. J.; Kempf, D. J. Bioorg. Med. Chem.

Lett. 2006, 16, 3833.

for C₂₁H₁₅Br(81)N₄O₂S: 468.0079; found: 468.0098. Anal. Calcd

for C₂₁H₁₅BrN₄O₂S: C, 53.97; H, 3.24; N, 11.99. Found: C, 53.99;

H, 3.23; N, 11.97.

5.1.2.2.6. 3-(1-Isopentyl-1H-indol-3-yl)-7-bromo-4H-pyrido[2,3-

e][1,2,4]thiadiazine 1,1-dioxide (IV-f). Yield: 43%. ¹H NMR

(300 MHz, DMSO) d = 11.69 (s, 1H), 8.50 (s, 1H), 8.35 (m, 1H),

8.21 (m, 1H), 7.65 (m, 1H), 7.42–7.22 (m, 3H), 4.21 (m, 2H), 1.72

(m, 1H), 1.54 (m, 2H),0.96 (d, J = 7,6H). ¹³C NMR (75 MHz, DMSO)

d = 152.0, 141.7, 141.3, 140.5, 134.7, 127.8, 125.2, 122.9, 121.2,

120.6, 118.7, 110.4, 98.1, 49.7, 36.3, 27.1, 22.7.HRMS calcd for

39. Cruciani, G.; Ed. Molecular Interaction Fields: Applications in Drug Discovery and

ADME Prediction.; 2006, 27: Wiley-VCH GmbH & Co. KGaA, 2006.

2478

M. A. H. Ismail et al. / Bioorg. Med. Chem. 20 (2012) 2455–2478

40. Kastenholz, M. A.; Pastor, M.; Cruciani, G.; Haaksma, E. E.; Fox, T. J. Med. Chem.

2000, 43, 3033.

41. Fontaine, F.; Pastor, M.; Sanz, F. J. Med. Chem. 2004, 47, 2805.

42. Boer, D. R.; Kroon, J.; Cole, J. C.; Smith, B.; Verdonk, M. L. J. Mol. Biol. 2001, 312,

275.

K.; Vandercruyssen, G.; Van der Helm, L.; Cleiren, E.; Mostmans, W.; Lory, P.;

Pille, G.; Van Emelen, K.; Fanning, G.; Pauwels, F.; Lin, T. I.; Simmen, K.;

Raboisson, P. Bioorg. Med. Chem. Lett. 2009, 19, 2492.

56. Guerois, R.; Mendes, J.; Serrano, L. Protein Science Encyclopedia; Wiley-VCH

GmbH & Co: KGaA, 2008.

43. Eldridge, M. D.; Murray, C. W.; Auton, T. R.; Paolini, G. V.; Mee, R. P. J. Comput.

Aided Mol. Des. 1997, 11, 425.

57. Zhou, R. H.; Friesner, R. A.; Ghosh, A.; Rizzo, R. C.; Jorgensen, W. L.; Levy, R. M. J.

Phys. Chem. B 2001, 105, 10388.

44. Friesner, R. A.; Murphy, R. B.; Repasky, M. P.; Frye, L. L.; Greenwood, J. R.;

Halgren, T. A.; Sanschagrin, P. C.; Mainz, D. T. J. Med. Chem. 2006, 49, 6177.

45. Wolber, G.; Langer, T. J. Chem. Inf. Model. 2004, 45, 160.

46. Vinter, J. G. J. Comput. Aided Mol. Des. 1994, 8, 653.

47. Mayo, S. L.; Olafson, B. D.; Goddard, W. A. J. Phys. Chem. 1990, 94, 8897.

48. Gehlhaar, D. K.; Verkhivker, G. M.; Rejto, P. A.; Sherman, C. J.; Fogel, D. B.; Fogel,

L. J.; Freer, S. T. Chem. Biol. 1995, 2, 317.

58. Owsley, D. C.; Bloomﬁeld, J. J. Synthesis 1977, 118.

59. Imai, Y.; Sato, S.; Takasawa, R.; Ueda, M. Synthesis 1981, 35.

60. Valko, A. T.; Johnson, A. P. J. Chem. Inf. Model. 2009, 49, 780.

61. Hendricks, R. T.; Fell, J. B.; Blake, J. F.; Fischer, J. P.; Robinson, J. E.; Spencer, S. R.;

Stengel, P. J.; Bernacki, A. L.; Leveque, V. J. P.; Pogam, S. L.; Rajyaguru, S.; Najera,

I.; Josey, J. A.; Harris, J. R.; Swallow, S. Bioorg. Med. Chem. Lett. 2009, 19, 3637.

62. Hendricks, R. T.; Spencer, S. R.; Blake, J. F.; Fell, J. B.; Fischer, J. P.; Stengel, P. J.;

Leveque, V. J. P.; LePogam, S.; Rajyaguru, S.; Najera, I.; Josey, J. A.; Swallow, S.

Bioorg. Med. Chem. Lett. 2009, 19, 410.

49. Jain, A. N. J. Comput. Aided Mol. Des. 1996, 10, 427.

50. Muegge, I. J. Med. Chem. 2005, 49, 5895.

51. Böhm, H.-J. J. Comput. Aided Mol. Des. 1994, 8, 243.

52. Korb, O.; Stutzle, T.; Exner, T. E. J. Chem. Inf. Model. 2009, 49, 84.

53. Yang, J. M.; Chen, C. C. Proteins 2004, 55, 288.

54. Guo, J.; Hurley, M. M.; Wright, J. B.; Lushington, G. H. J. Med. Chem. 2004, 47,

5492.

63. Poshkus, A. C. Product R&D 1971, 10, 72.

64. Paeshuyse, J.; Leyssen, P.; Mabery, E.; Boddeker, N.; Vrancken, R.; Froeyen, M.;

Ansari, I. H.; Dutartre, H.; Rozenski, J.; Gil, L. H.; Letellier, C.; Lanford, R.;

Canard, B.; Koenen, F.; Kerkhofs, P.; Donis, R. O.; Herdewijn, P.; Watson, J.; De

Clercq, E.; Puerstinger, G.; Neyts, J. J. Virol. 2006, 80, 149.

55. McGowan, D.; Nyanguile, O.; Cummings, M. D.; Vendeville, S.; Vandyck, K.; Van

den Broeck, W.; Boutton, C. W.; De Bondt, H.; Quirynen, L.; Amssoms, K.;

Bonfanti, J. F.; Last, S.; Rombauts, K.; Tahri, A.; Hu, L.; Delouvroy, F.; Vermeiren,

65. Fox, B. A.; Threlfall, T. L. Org. Synth. 1964, 44, 34.

66. Johnson, J. A.; Lloyd, J.; Finlay, H.; Jiang, J.; Neels, J.; Dhondi, N. K.; Gunaga, P.;

Banerjee, A.; Adisechan, A. WO2011028741A1; 2011, 495pp.

Article Doi

Integrated structure-based activity prediction model of benzothiadiazines on various genotypes of HCV NS5b polymerase (1a, 1b and 4) and its application in the discovery of new derivatives

DOI: 10.1016/j.bmc.2012.01.031

Source and publish data:

Authors:

Article abstract of DOI:10.1016/j.bmc.2012.01.031

Full text of DOI:10.1016/j.bmc.2012.01.031

Products guided by the article

R&D Labs maybe for 1370714-01-4

Relevant to this article

Hot Product