Molecular dynamics simulations and structure-based rational design lead to allosteric HCV NS5B polymerase thumb pocket 2 inhibitor with picomolar cellular replicon potency

Article abstract of DOI:10.1021/jm4004522

The design and preliminary SAR of a new series of 1H-quinazolin-4-one (QAZ) allosteric HCV NS5B thumb pocket 2 (TP-2) inhibitors was recently reported. To support optimization efforts, a molecular dynamics (MD) based modeling workflow was implemented, pro

Full text of DOI:10.1021/jm4004522

Subscriber access provided by ANDREWS UNIV
Article
Molecular dynamics simulations and structure-based rational
design lead to allosteric HCV NS5B polymerase thumb
pocket 2 inhibitor with picomolar cellular replicon potency.
Oliver Thomas Hucke, René Coulombe, Mégan Bertrand-Laperle, Christian Brochu, James Gillard, Marc-
André Joly, Serge Landry, Olivier Lepage, Martin Marquis, Ginette McKercher, Marc Pesant, Martin Poirier,
Pierre L. Beaulieu, Pierre R Bonneau, Montse Llinàs-Brunet, George Kukolj, and Timothy A. Stammers
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/jm4004522 • Publication Date (Web): 17 Jun 2013
Downloaded from http://pubs.acs.org on July 6, 2013
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Medicinal Chemistry is published by the American Chemical Society. 1155
Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.

Page 1 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Title: Molecular dynamics simulations and structureꢀ
based rational design lead to allosteric HCV NS5B
polymerase thumb pocket 2 inhibitor with picomolar
cellular replicon potency.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Authors: Oliver Hucke,* René Coulombe, Pierre Bonneau, Mégan Bertrand-Laperle, Christian
Brochu, James Gillard, Marc-André Joly, Serge Landry, Olivier Lepage, Montse Llinàs-Brunet,
Marc Pesant, Martin Poirier, Maude Poirier, Ginette McKercher, Martin Marquis, George
Kukolj, Pierre L. Beaulieu and Timothy A. Stammers
Authors address: Research and Development, Boehringer Ingelheim (Canada) Ltd., 2100 Cunard
Street, Laval, Quebec, H7S 2G5, Canada
ABSTRACT
The design and preliminary SAR of a new series of 1Hꢀquinazolinꢀ4ꢀone (QAZ) allosteric HCV
NS5B thumb pocket 2 (TPꢀ2) inhibitors was recently reported. To support optimization efforts, a
Molecular Dynamics (MD) based modeling workflow was implemented, providing information
on QAZ binding interactions with NS5B. This approach predicted a small but critical ligandꢀ
ACS Paragon Plus Environment
1

Journal of Medicinal Chemistry
Page 2 of 37
1
2
3
4
5
6
7
8
9
binding induced movement of a protein backbone region which increases the pocket size and
improves access to the backbone carbonyl groups of Val 494 and Pro 495. This localized
backbone shift was consistent with key SAR results and was subsequently confirmed by Xꢀray
crystallography. The MD protocol guided the design of inhibitors, exploiting novel Hꢀbond
interactions with the two backbone carbonyl groups, leading to the first thumb pocket 2 NS5B
inhibitor with picomolar antiviral potency in genotype (gt) 1a and 1b replicons (EC50 = 120 pM
and 110 pM, respectively), and with EC50 ≤ 80 nM against gt 2ꢀ6.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
ACS Paragon Plus Environment
2

Page 3 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
INTRODUCTION
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
With an estimated 2ꢀ3% of the world population (130ꢀ170 million people; 5 million in the US
alone) as chronic carriers of the Hepatitis C virus (HCV),1 the impact of this infectious agent on
public health is undeniable. Chronic HCV infection is a leading cause for hepatocellular
carcinoma and liver transplants and causes approximately 86000 deaths/year worldwide.2
Until recently, the standard of care (SoC) for HCV treatment combined pegylated interferonꢀα
(PegIFN) with the broad spectrum antiviral ribavirin (RBV). However, sustained viral responses
(SVR) for this therapy rarely exceeded 50% for genotype 1 (gt1) the most prevalent genotype in
North America, Europe and Japan. This limited efficiency, along with severe side effects,3
underline the need for more efficacious and better tolerated approaches for HCV treatment.
Small molecules that specifically target HCV proteins (direct acting antivirals, DAAs) have
now begun to address the need for improved treatment outcome and tolerability. An initial
landmark in this regard was the structureꢀbased peptidomimetic design of ciluprevir, the first
inhibitor of a viral protein, in this case the HCV NS3/4A protease, to show antiviral activity in
HCV infected patients.4 Following this discovery, the search for NS3/4A inhibitors became a
focus of HCV drug development for many pharmaceutical companies. Two agents from this
class, boceprevir5 from Merck and telaprevir from Vertex6 have been approved. Today’s new
SoC (NSoC) combines one such protease inhibitor with the previously used PegIFN/RBV
therapy. This new regimen improves SVR rates to 68ꢀ75% and shortens duration of treatment for
some gt1 patients,7 but still suffers from side effects associated with the PegIFN/RBV in addition
to those caused by the new drugs themselves. Second generation HCV protease inhibitors with
improved profiles are currently in clinical trials and are expected to reach market approval in the
ACS Paragon Plus Environment
3

Journal of Medicinal Chemistry
Page 4 of 37
1
2
near future (e.g. simeprevir8 from Johnson & Johnson and faldaprevir from Boehringer
Ingelheim9). However, these single DAAꢀbased therapies still require combination with
PegIFN/RBV in order to prevent rapid emergence of resistant virus. Future directions that could
avoid the use of PegIFN and RBV coꢀtherapy are therefore of high interest. Combining DAAs
with complementary modes of action is one potential route to achieve this goal, and more
recently multiple DAA combinations have indeed shown the potential to increase cure rates up to
>90% in the gt1 population.10
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
HCV protease inhibitor faldaprevir and HCV NS5B polymerase inhibitor BI 207127 that binds
to the allosteric thumb pocket 1 site (Fig. 1)are two DAAs from our portfolio currently under
phase 3 clinical investigation in an IFNꢀfree combination.11 The thumb domain of NS5B also
harbors a second allosteric pocket (thumb pocket 2, TPꢀ2), where binding of ligands can also
lead to inhibition of enzymatic function.13 Although the exact molecular mechanisms by which
allosteric NS5B pocket 1 and pocket 2 inhibitors disrupt RNA synthesis is not entirely defined,
biochemical experiments suggest that they inhibit at the initiation stage of the RNA
polymerization process by preventing the formation of a productive polymerase:RNA primerꢀ
template complex. It is generally assumed that inhibition of RNA synthesis occurs through
interference with essential enzyme conformational changes.13 Importantly, it has been shown in
vitro that TPꢀ1 and TPꢀ2 NS5B inhibitors are not mutually exclusive, opening the possibility for
new combinations of DAAs for HCV therapy.
We recently reported the fragmentꢀbased design of an anthranilic acidꢀbased series of thumb
pocket 2 NS5B inhibitors,15 exploiting structural information on a bound fragment hit in
conjunction with publicly available information on TPꢀ2 binders. Subsequently we were able to
use our knowledge of the anthranilic acid SAR and their binding mode to develop a structurally
ACS Paragon Plus Environment
4

Page 5 of 37
Journal of Medicinal Chemistry
1
2
3
4
novel, quinazolinone (QAZ) TPꢀ2 chemotype (e.g. 1, Figure 1) with a distinct physicochemical
5
6
profile.16
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Figure 1 Ribbon structure of NS5B with thumb pockets 1 and 2 highlighted by two inhibitors
(CPK representation). The depicted ligands are BI 207127 (TPꢀ1, orange; model based on an in
house Xꢀray structure with a related inhibitor) and QAZ type compound 1 (TPꢀ2, magenta) that
emerged from the anthranilic acid chemotype (exemplified by A) through structure based drug
design (SBDD).16 The NS5B palm domain is shown in red. It harbors the active site including
two magnesium ions. The thumb domain (blue) contains a GTP binding site in close proximity to
TPꢀ1, which may be implicated in de novo initiation. The finger domain with the Λ finger loops
(green) are forming a channel that holds the RNA in place and regulate replication.12
Herein, we describe the optimization of this quinazolinone TPꢀ2 chemotype by extending
proteinꢀligand interactions within the left hand side (LHS) region of the binding site, i.e. into the
surface cleft above Leu 489 towards Val 494. (Fig. 2). A modeling approach using molecular
ACS Paragon Plus Environment
5

Journal of Medicinal Chemistry
Page 6 of 37
1
2
3
4
5
6
7
8
9
dynamics (MD) simulations that captured binding site flexibility, revealed a small but significant
backbone conformational change that resulted in improved access to the carbonyl groups of Val
494 and Pro 495 and to extended hydrophobic interactions. The design of an analog exploiting
these new interactions resulted in one of the most potent NS5B inhibitors reported to date, with
picomolar potency in gt1 replicon assays and low nanomolar potency against gt 2ꢀ6.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Figure 2 Xꢀray structure of quinazolinone (QAZ) 1 bound to TPꢀ2 of HCV NS5B. This
chemotype was designed to maintain Hꢀbond interactions with the backbone NH groups of Ser
476 and Tyr 477 (green dashed lines).16 White dashed lines outline the LHS region of TPꢀ2 that
was targeted for further optimization of the QAZ chemotype.
RESULTS AND DISCUSSION
Using our established Xꢀray crystallographic protocols (see Experimental section), we were
able to determine the binding mode of a quinazolinone analog in complex with NS5B (Fig. 2).
Compound 1 shows the expected Hꢀbonds to the backbone NH groups of Ser 476 and Tyr 477
that are crucial for binding of the anthranilic acid series and other TPꢀ2 NS5B inhibitors, and that
ACS Paragon Plus Environment
6

Page 7 of 37
Journal of Medicinal Chemistry
1
2
the quinazolinone core was designed to maintain.16 The N1ꢀ2,4,6ꢀtrifluorobenzyl group anchors
the inhibitor by occupying a deep hydrophobic pocket while the (2ꢀtrifluoromethylꢀ)phenyl
moiety at the other end of the molecule contacts the side chain of Leu 497 and places its CF3
group in a small hydrophobic pocket. As depicted in Figure 2, a NS5B TPꢀ2 surface cleft extends
from the LHS portion of the inhibitor and can be engaged through additional substitutions on the
LHS phenyl ring. This mostly hydrophobic LHS region of TPꢀ2 was targeted to further improve
the potency of the quinazolinone NS5B inhibitors.
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
The addition of an amide group at position 3 of the LHS phenyl ring slightly improved
antiviral potency from 380 nM ( compound 1) to 180 nM (compound 2; Table 1),17 in spite of the
increased hydrophilicity associated with the amide group, which may have negatively affected
cell permeability. A more substantial gain in antiviral potency to the single digit nanomolar
range was achieved by adding a phenyl or pyridyl substituent to the amide nitrogen atom
(compounds 3, 4, 5 in Table 1). The Xꢀray structures of compounds 2 and 4 in complex with
NS5B were solved by crystal soaking (Figure 3A), revealing in both cases a new hydrogen bond
between the amide carbonyl group and the backbone NH of Leu 497. In fact, previous SAR and
corresponding inꢀhouse Xꢀray structures of NS5B:TPꢀ2 inhibitor complexes had shown that this
Hꢀbond interaction usually contributed to improved potency. Based on the high potency of 4 and
related structures, subsequent SAR focused on maintaining the interaction with Leu 497 for
productively exploring the LHS region of pocket 2. Interestingly, the LHS pyridine ring of
compound 4 also forms a Hꢀbond with the side chain of Arg 490 (Fig. 3A). However, compound
3, which lacks the pyridyl nitrogen hydrogen bond acceptor, has comparable potency to
compound 4 (Table 1), suggesting that the Hꢀbond interaction with the Arg 490 side chain does
not significantly contribute to activity. Indeed, multiple Xꢀray structures of NS5B reveal various
ACS Paragon Plus Environment
7

Journal of Medicinal Chemistry
Page 8 of 37
1
2
3
4
5
6
7
8
9
conformations of the side chain of Arg 490, and suggest that a potential gain in binding enthalpy
through a hydrogen bond to its guanidino function may be offset by a loss in receptor entropy.
Figure 3A also shows the high degree of conservation of the loop backbone structures of the Val
494 to Leu 497 region that was characteristic of all Xꢀray structures of NS5B:pocket 2 inhibitor
complexes solved to date in both the anthranilic acid as well as quinazolinone chemotypes. The
addition of the LHS amide group in compounds 2ꢀ5 resulted in a significant drop in apical to
basolateral (A → B) Cacoꢀ2 cell permeability (Table 1), an effect often associated with amide
groups and the presence of hydrogenꢀbond donors. In order to address this issue, an
azabenzoxazole heterocycle was designed to mask the amide moiety while maintaining the Hꢀ
bond acceptor in a position equivalent to the carbonyl oxygen atom of the amide group, suitable
for forming a Hꢀbond interaction with Leu 497.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Table 1 Early examples of LHS SAR exploration
F
F
F
O
N
F
O
R1
N
F
F
O
O
O
O
O
N
R1
H
N
N
N
H2N
N
H
N
H
N
H
Compound
1
2
3
4
5
6
EC50 a (nM)
380±90
180±100
2.5±0.4
5.7±0.6
6.8±0.7
3.3±1.4
CACOꢀ2b
9.2
1.9
0.84
0.26
1.2
3.0
(10ꢀ6 cm/s)
a HCV gt 1b replicon luciferase reporter assay (n≥2).18 b The Cacoꢀ2 permeability assay was
run with apical pH 6.0/basolateral pH 7.4.
ACS Paragon Plus Environment
8

Page 9 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
Prior to the synthesis of 6, we evaluated our hypothesis with a simple model derived from the
Xꢀray structure of NS5B with compound 4 bound to TPꢀ2 (Figure 3B). This tentative model
strongly suggested that formation of the Hꢀbond between the oxazole nitrogen atom of 6 and Leu
497 would inevitably lead to a clash between the pyridine ring of the azabenzoxazole moiety and
the backbone atoms of Pro 496 (Figure 3B). Attempts to resolve this clash through
minimizations of 6 in a flexible protein binding site led to shifts of the LHS part of the inhibitor
from the position shown in Figure 3B, away from Leu 497 and Pro 496, which caused a loss of
the Hꢀbond interaction. Unexpectedly, 6 maintained the high potency of compounds 3ꢀ5 and
some improvement in Cacoꢀ2 cell permeability was also achieved, as intended by the design
(Table 1). The discrepancy between the observed potency of 6, the available structural
information on this class of inhibitors and initial modeling results prompted the use of Xꢀray
crystallography to gain further experimental information on the binding mode and rationalize the
potency achieved with this class of compounds. Unfortunately, repeated attempts to soak 6 into
NS5B crystals using our established procedures (see Experimental section) were unsuccessful.
Severe crystal cracking was observed upon soaking of compound 6, and suggested possible
ligandꢀinduced protein conformational changes or disturbance of crystal packing. In the absence
of Xꢀray information, we turned to a molecular dynamics (MD) based modeling protocol that
accounted for protein flexibility to elucidate the binding geometry of 6. In this structure
prediction protocol, the thumb domain that harbors pocket 2 was considered (220 residues) with
all amino acid residues fully flexible, except those that are at the interface with the parts of NS5B
which were not considered in the simulation (38 restrained residues, 17% of the residues
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
ACS Paragon Plus Environment
9

Journal of Medicinal Chemistry
Page 10 of 37
1
2
3
4
5
6
considered; Figure 5 summarizes the simulation approach; See Experimental section and
supporting information for additional details).
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Figure 3, A. Superposition of the Xꢀray structures of the binding modes of 2 and 4. The LHS
amide carbonyl group forms a Hꢀbond with the backbone NH of Leu 497. The LHS pyridine ring
of 4 displaces a water molecules that, in the Xꢀray structure with 2, forms a Hꢀbond with the side
chain of Arg 490. (The Hꢀbond between 4 and Arg 490 does not contribute to antiviral potency,
as suggested by the comparable potency of compound 3); B. Xꢀray binding mode of 4 (different
perspective, otherwise identical to A) superposed with a first tentative model of the binding
mode of 6, obtained by modifying 4 in its bound conformation. A Hꢀbond between 6 and the
backbone NH of Leu 497 inevitably leads to a serious atom clash with Pro 496.
The binding mode of 6 predicted using the MD protocol is shown in Figure 4A, in
superposition with the Xꢀray structure of 4 bound to TPꢀ2. The overall binding geometries of
compounds 4 and 6 are highly similar, with the oxygen atom of the oxazole part of the LHS
bicyclic ring system of 6 mimicking the amide carbonyl oxygen atom of 4. However, no
hydrogen bond between 6 and the NH of Leu 497 is observed. The absence of this interaction is
due to a unique conformation of the loop region between Val 494 and Leu 497. This backbone
ACS Paragon Plus Environment
10

Page 11 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
region shifts relative to the usually observed crystallographic position by approximately 1 Å such
that the clash between Pro 496 and the leftmost pyridine ring of 6 is avoided. One consequence
of this conformational change is that the backbone NH of Leu 497 points away from the oxazole
ring, above the plane of the azabenzoxazole system, where there is space for interaction with a
water molecule. In addition to avoiding the clash with Pro 496 this alternate protein
conformation improves access to the backbone carbonyl groups of Pro 495 and Val 494 which
form favorable electrostatic contacts with two CꢀH groups of the pyridine ring in our model.
These are highly polarized due to the electron withdrawing effect of the nitrogen and oxygen
atoms of the heterocycle and are thus suitable for such polar interactions and likely contribute to
the improved potency of 6.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Figure 4 A. Binding mode of 6 as predicted by the MD protocol in comparison with compound 4
(Xꢀray structure): The binding of 6 induces a shift of the protein backbone between Val 494 and
Leu 497, leading to favorable CꢀH…O contacts with the backbone carbonyl groups of Val 494
and Pro 495 (arrows – distances are indicated). No Hꢀbond to Leu 497 is apparent for compound
6. Only the variable LHS portions of the two inhibitors are shown for clarity. The insets show
good overall agreement of the bound geometries of both compounds and the electrostatic surface
potential (ESP) of the azabenzoxazole group (ESP units: kcal/mol; Arrows indicate CH groups
ACS Paragon Plus Environment
11

Journal of Medicinal Chemistry
Page 12 of 37
1
2
3
4
5
6
7
8
9
contacting the carbonyls of Val 494 and Pro 495). B. Comparison of binding modes of 6
predicted by the MD simulation protocol (yellow) and observed in a subsequently obtained Xꢀ
ray structure (cyan). The root mean square deviation (rmsd) between the model and Xꢀray
structure amounts to 0.48 Å for the loop residues Val 494 – Leu 497 and to 0.31 Å for the ligand
atoms (rmsd based on nonꢀhydrogen atoms; only LHS portion of ligand shown, for clarity – all
ligand atoms were considered in the rmsd calculation). Superpositions were done using the Cα
atoms of the thumb domain.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
In an effort to acquire an Xꢀray structure of NS5B with compound 6 bound to TPꢀ2, a variety
of soaking conditions were screened and optimized to provide a 3.0 Å resolution Xꢀray model.
The superposition of the MD model with the experimentally derived Xꢀray structure shows
excellent agreement (Figure 4B): It fully confirmed the predicted bound geometry of the
inhibitor as well as the localized shift of the protein backbone. (Based on the relatively low
resolution of this Xꢀray structure an orientation of the azabenzoxazole group that is rotated by
180° can not be excluded. However, in this rotated orientation both nitrogen atoms would point
towards the protein, causing desolvation and electrostatic repulsion with the backbone carbonyl
of Pro 495. This orientation was therefore considered unlikely).
This concordance of prediction and experiment further validated the MD protocol and it was
subsequently used to guide chemistry efforts through structureꢀbased interpretation of SAR and
the creation and assessment of design ideas prior to synthesis. To allow a more efficient use of
this approach, different steps of the MD workflow were automated, such that simulations for
different hypothetical poses of the same inhibitor in TPꢀ2 or for a set of different inhibitors were
set up and submitted to a Linux cluster and results were retrieved in an automated fashion.
(Figure 5; see Experimental section for details). The MD protocol included distance restraints
ACS Paragon Plus Environment
12

Page 13 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
that favored the presence of the Hꢀbond interactions between the quiazolinone core and the
backbone NH groups of Ser 476 and Tyr 477. These restraints were based on previous SAR that
showed that these interactions were crucial for potency and they were at the heart of the structure
based design leading to the quinazolinone chemotype.16 The restraints were applied to focus the
MD sampling on the LHS region of the inhibitors that was the focus of ongoing synthesis efforts.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
ligandbindingmodes
(e.g.from docking)
Receptor+cmpd1
(X-ray)
MDinputfilesfor1
proteinligandcomplex
Cmpd1, modified
Receptor
F
F
F
inbindingsite
O
N
F
O
O
O
N
N
R
Preparationof
ligandinputfiles
F
N
1
Manually
(once/receptor)
Initial binding
modemodel
Preparationof
Inputfilesfor
proteinligandcomplex
F
Automated
Datatransfer
MDsimulation
Manually
Averagingof MDsnapshots,
Minimization
Coordinateaveraging,
Visualinspection:
LinuxCluster
MDsimulation:
1complex/node(8CPU’s)
Bindingmodeprediction
MDmodelof
bindingmode
A.
B.
Figure 5. Graphical overview of the MD simulation protocol that was used for predicting
structures of NS5B:thumb pocket 2 inhibitor complexes. A. General workflow for binding mode
prediction of quinazolinones with LHS structure variations. B. Overview of the automation of
the MD protocol (see Experimental section for details).
Since a hydrogen bond between the inhibitor and Leu 497 no longer appeared to be a
prerequisite for the design of highly potent inhibitors, subsequent efforts shifted to capitalizing
on the newly revealed interaction opportunities resulting from the ligandꢀinduced backbone
movement of the loop between Val 494 and Leu 497. The MD model of the binding mode of 6,
invoking CꢀH…O contacts between the pyridine ring and the carbonyl groups of Pro 495 and
ACS Paragon Plus Environment
13

Journal of Medicinal Chemistry
Page 14 of 37
1
2
3
4
5
6
7
8
9
Val 494, suggested that structural elements capable of forming hydrogen bonds could be
beneficial. We therefore designed an indole replacement (compound 7, Table 3) of the LHS
azabenzoxazole group of 6. Figure 6 shows the MD model for the corresponding binding mode,
according to which the indole NH of 7 is located within Hꢀbond distance (3.0 Å) from the
carbonyl oxygen atoms of both Pro 495 and Val 494, with a more favorable NꢀH…O angle for
Pro 495 (149°, as opposed to 125° for Val 494). As indicated by the superposition with the
binding geometry of 6 (Fig. 6A), positions and orientations of the LHS moieties agree very well,
and the indole NH occupies the position of the 6ꢀCH group of the azabenzoxazole of 6 that is
closest to the two targeted carbonyls. The side view in Figure 6B shows that the phenyl part of
the indole ring partially “slips under” the displaced loop region, forming close contacts
especially with Pro 496, again with the NH of Leu 497 pointing above the plane of the bicyclic
ring system. The two outermost carbon atoms of the pyrrole part of the indole nucleus extend
hydrophobic contacts with the bottom of the binding pocket, most importantly with the side
chain of Leu 489 (not shown).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
The utility of our MD prediction protocol for structureꢀbased drug design is illustrated by the
substantial 25ꢀfold gain in cell culture antiviral potency achieved with compound 7 (Table 2)
which is likely attributable to the new Hꢀbond interaction and the described hydrophobic
contacts. Indole containing inhibitor 7 is one of the most potent antiꢀHCV agents reported to date
that acts via allosteric NS5B inhibition. The antiviral potency of 7 was tested in HCV gt1a and
gt1b replicon assays, as well as gt2b, 3a, 4a, 5a, and 6a, using chimeric HCV 1b replicons that
incorporate NS5B polymerase genes from these genotypes as surrogates (Table 2). The activity
was in the picomolar range for the two most predominant genotypes 1a and 1b, approaching the
magnitude of the exceptional potencies that were recently reported by BMS for HCV NS5A
ACS Paragon Plus Environment
14

Page 15 of 37
Journal of Medicinal Chemistry
1
2
inhibitors.19 Notably, antiviral potency remained in the 1 – 80 nM range for all other tested
NS5B genotypes, including the gt3a.
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Table 2. Potency of 7 against various genotypes.
Replicon results
HCV 1b chimeric replicons with NS5B from different
genotypes
HCV genotype
EC50 a (nM)
0.12±0.02
0.11±0.02
NS5B genotype
EC50 b(nM)
1a
1b
2b
3a
4a
5a
6a
15±2
80±5
1.1±0.2
1.3±0.6
11±3
a Cellꢀbased replicon assays with polio virus internal ribosomal entry site upstream of luciferase
(see Experimental section). b Cross genotype cell based replicon assays.
SAR around compound 7 is summarized in Table 3 and can be rationalized using the model
and aforementioned interactions with the pocket. The isomeric indole analogue 9 is predicted to
be incapable of forming the corresponding hydrogen bond to Pro 495 (and Val 494). Rather, the
NH is expected to point towards Arg 490 and to be solvent exposed, while polarized pyrrole CꢀH
groups (Figure 7) are oriented towards the backbone carbonyls and extensive hydrophobic
interactions are maintained with the binding site. Conversely, the MD model for compound 10,
in which the pyrrole portion of the indole system was reduced to an amino group, predicts the
new Hꢀbond to be present, but the extended hydrophobic contacts with Leu 489 are partially lost.
Consistent with the predictions, compounds 9 and 10 lose approximately the same level of
potency compared to 7 as they each give up one of the two main contributing factors to potency
ACS Paragon Plus Environment
15

Journal of Medicinal Chemistry
Page 16 of 37
1
2
3
4
5
6
7
8
9
(i.e., the new Hꢀbond in case of 9 and part of the hydrophobic interactions in case of 10).
Compounds 13 and 14 which lack the complete pyrrole portion of the indole system and can
therefore neither form favorable polar interactions with the backbone carbonyl groups of Pro
495/Val 494 nor engage in extended hydrophobic contacts with the binding site, show an
additional ~ 30ꢀ to 45ꢀfold loss in potency. Compound 11 constitutes an interesting case with a
potency that is between that of the compound pairs 9/10 and 13/14. This is rationalized by the
fact that the paraꢀmethyl of 11 that replaces the amino group of 10 favorably extends the contact
with Leu 489 and can to some degree interact favorably with the Pro 495/Val 494 carbonyls due
to the positive electrostatic potential of the methyl protons pointing away from the pyridine
nitrogen atom and towards the carbonyl oxygen atoms (Fig. 7). The additional methyl group in
compound 12 extends hydrophobic contacts with Leu 489 compared to 10 resulting in a slight
gain in potency. In the case of benzimidazole 8 on the other hand, a hydrophobic CH moiety of 7
is replaced by a hydrophilic nitrogen atom that has to be exposed to the solvent to avoid
desolvation and that is unable to form optimal hydrophobic contacts with Leu 489. This may
explain the observed loss in potency from 130 to 600 pM.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Table 3. SAR of compounds related to 7
CF3
O
N
F
R1
O
N
F
F
H
N
H
N
H2N
N
H2N
N
N
R1
N
H
N
N
Compound
EC50 a (nM)
7
8
9
10
11
55±4
12
13
14
0.13d
0.6±0.2
10±6
10±2
6±2
450±92 310±45
ACS Paragon Plus Environment
16

Page 17 of 37
Journal of Medicinal Chemistry
1
2
Cacoꢀ2b
10.5
0.4
n.d.
4.1
n.d.
1.1
74/<0.1
>4.7
n.d.
n.d.
n.d.
n.d.
n.d.
n.d.
3
4
5
6
7
8
9
(10ꢀ6 cm/s)
Sol.c pH2/6.8 <0.1/
(ꢁg/ml)
<0.1/
<0.1
<0.1/
<0.1
0.3/
<0.1
180/
<0.1
<0.1
logD(pH 7.4)
5.1
4.7
5.0
3.9
4.2
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
a HCV gt 1b replicon luciferase reporter assay (n≥2).18 b The Cacoꢀ2 permeability assay was
c
run with apical pH 6.0/basolateral pH 7.4. Solubility measured on lyophilized, amorphous
solids using the 24 h shaking flask method. d n=2, identical values in both determinations
Figure 6. Binding mode of compound 7, a picomolar thumb pocket 2 NS5B inhibitor, as
predicted by the MD protocol. A. Superposition with the binding mode of 6, highlighting that the
NH group of the indole moiety of 7 is within Hꢀbond distance of the backbone carbonyl groups
of Pro 495 and of Val 494, replacing the favorable CꢀH…O contacts of 6. Highly similar
backbone conformations of the loop region between Val 494 and Leu 497 are observed with
compounds 6 and 7, i.e. both compounds lead to a similar displacement of the loop region
between Val 494 and Leu 497. B. Surface representation of the binding pocket. The atom clash
ACS Paragon Plus Environment
17

Journal of Medicinal Chemistry
Page 18 of 37
1
2
3
4
5
6
with Pro 496 (see Fig. 3B) is avoided by the backbone movement that leads to a placement of the
phenyl ring of the indole under the backbone portion of Pro 496.
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Figure 7. Computed electrostatic potential on the molecular surfaces of fragments mimicking the
LHS groups of selected QAZ’s (DFT at B3LYP/ccꢀpVTZ(ꢀf) level, software: Jaguar, version 7.6,
Schrödinger, Inc., New York, NY, 2009). Relative orientations correspond to predicted bound
conformations. Arrows indicate the surface regions that are in close proximity to the (partially
negative) backbone carbonyl oxygen atoms of Pro 495 and Val 494.
It is noteworthy that the picomolar inhibitor 7 displayed extremely low solubility (< 0.1
ꢁg/mL) at pH 2 and 6.8 (Table 3). This had prompted us to explore structures with a somewhat
lower molecular weight and lipophilicity. Compounds 10 and 11 reduce the molecular weight by
23 and 24 Da, respectively, and decrease the logD by approximately 1 unit lipophilicity. By
keeping the new hydrogen bond to Pro 495 / Val 494, 10 manages to maintain the cell culture
potency within the desirable range (< 10 nM) and shows low but measurable solubility at low
pH. As a second approach to reduce lipophilicity and molecular weight, we explored QAZs
lacking the orthoꢀtrifluoromethyl group of the LHS phenyl ring but taking advantage of the
ACS Paragon Plus Environment
18

Page 19 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
newly discovered interactions with Pro 495/Val 494 (desꢀCF3 compounds; Table 4). Prior to
synthesis, the potential of this strategy was evaluated through MD simulations. The results
clearly suggested that in order to reach the beneficial interactions with Pro 495/Val 494 the point
of substitution had to be moved from the meta to the para position (Figure 8): in absence of the
CF3 group, the position of the phenyl ring shifts in the binding pocket towards the vacated “CF3ꢀ
pocket” in order to occupy the available space (Figure 8B) – for the desꢀCF3 phenyl ring the
corresponding conformation relative to the QAZ core is in good agreement with the low energy
conformation in solution. However, it is noteworthy that the phenyl ring cannot fill the “CF3
pocket” in an optimal fashion, due to inadequate shape complementarity with the narrow
hydrophobic depression in the TPꢀ2 surface that is surrounded by the side chains of Leu 419,
Met 423 and Val 485 and that was optimally filled by the CF3 moiety. A drop in potency is
therefore expected, even if all the other interactions were maintained. This was in agreement
with previous SAR work that showed that omitting the CF3 group leads to a systematic ~ 10ꢀfold
loss in potency (data not shown). Another contributing factor to this loss in activity could be that
desꢀCF3 compounds are expected to pay a higher entropic cost upon binding to NS5B, as the
presence of the relatively bulky orthoꢀCF3 group limits the conformational space that is
accessible to the LHS phenyl moiety. For all these reasons, it was not surprising to see a
substantial loss in potency for compound 15 that shares the azabenzoxazole group of CF3ꢀ
containing analogue 6. However, it was gratifying that the loss in potency can be partially
compensated by using the LHS indole group of 7. Compound 16, which is predicted to form a
hydrogen bond with the carbonyl of Pro 495 (Figure 8) had EC50 = 4.6 nM. This result highlights
that carefully designed desꢀCF3 molecules that make use of the newly discovered Hꢀbond
interaction, can still achieve low nM potency. 16 is an example of a potent QAZ inhibitor with
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
ACS Paragon Plus Environment
19

Journal of Medicinal Chemistry
Page 20 of 37
1
2
3
4
5
6
7
8
9
reduced molecular weight (498 Da compared to 566 Da for 7) and decreased lipophilicity (logD
= 4.4 vs 5.1). Unfortunately this did not translate into substantially improved solubility, and the
desꢀCF3 series was not further pursued.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Table 4 Examples of two desꢀCF3 compounds. 15 and 16 use the LHS extensions of 6 and 7,
attached to the phenyl ring in the para position based on modeling results.
O
O
N
N
F
F
R1
F
H
N
R1
O
N
N
Compound
15
16
EC50 a (nM)
490±140
4.6±2.4
CACOꢀ2b
n.d.
3.5
(10ꢀ6 cm/s)
Sol.c pH 2/6.8
(ꢁg/ml)
n.d.
n.d.
<0.1/0.18
4.4
logD(pH 7.4)
a HCV gt 1b replicon luciferase reporter assay (n≥2).18 b The Cacoꢀ2 permeability assay was
c
run with apical pH 6.0/basolateral pH 7.4. Solubility measured on lyophilized, amorphous
solids using the 24 h shaking flask method.
ACS Paragon Plus Environment
20

Page 21 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Figure 8. Superposition of MD models for desꢀCF3 compound 16 and compound 7. A. Both
show similar positions of the indole moiety and are predicted to form a Hꢀbond to Pro 495. B.
The side view shows the repositioning of the LHS phenyl ring of 16 relative to 7 as a
consequence of the absence of the CF3 group, requiring the attachment of the indole in the para
instead of the meta position for desꢀCF3 compounds (only the QAZ core and the LHS part are
shown, for clarity). The small pocket that is occupied by the CF3 group is shown as a dotted line,
the surrounding amino acids are indicated.
CHEMISTRY
The N1ꢀ2,4,5ꢀtriflurobenzylquinazolinone scaffold was built from 5ꢀhydroxyꢀ2ꢀaminobenzoic
acid 17 as shown in Scheme 1. A reductive amination was used to introduce the 2,4,5ꢀ
triflurobenzyl group onto 17. The quinazolinone heterocycle 19 was formed by refluxing 18
with formamidine acetate in 2ꢀmethoxyethanol. Hydroxylquinazolinone 19 could not be directly
coupled with methylꢀ3ꢀfluoroꢀ2ꢀtrifluoromethylbenzoate to form biarylether 23 without
concomitant decomposition. The problem was circumvented by first reducing 19 to
hydroxyquinazolidinone 20 thereby making the phenolic oxygen more nucleophilic. This
ACS Paragon Plus Environment
21

Journal of Medicinal Chemistry
Page 22 of 37
1
2
3
4
5
6
7
8
9
allowed SNAR coupling to proceed at a lower temperature and 22 could be recovered in
acceptable yield. Saponification followed by oxidation afforded acid intermediate 23 that was
coupled with the appropriate amines to afford inhibitors 2ꢀ5.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Scheme 1. Synthesis of carboxamide inhibitors 2ꢀ 5.
O
O
O
N
F
O
N
F
HO
HO
HO
HO
N
NH
OH
NH2
OH
NH
F
F
F
b
c
a
17
18
19
20
F
F
F
F
CF3
O
CF3
O
CF3
O
O
O
F
O
O
O
NH
N
O
HO
21
d
N
F
F
N
F
F
e, f
22
23
F
F
g
2, 3, 4, 5
(a) 2,4,6ꢀtrifluorobenzaldehyde, NaCNBH3, AcOH, 0 oC to RT, 16 h; (b) formamidine acetate,
2ꢀmethoxyethanol, reflux, 4 h; (c) NaBH4, MeOH, RT, 16 h; (d) 20, Cs2CO3, DMF, 50 oC, 24 h;
(e) NaOH, MeOH/THF, RT, 4 d; (f) I2, Na2CO3, MeOH, RT, 3 h; (g) NH3 or aniline or
aminopyridine, HATU, NEt3, DMF, RT, 3 h
Aryl substituted inhibitors 6ꢀ16 were prepared as shown in Scheme 2. Hydroxyquinazolinone
19 was found to couple efficiently with 2ꢀbromoꢀ6ꢀfluorobenzotrifluoride 24 to form 25 without
the reduction/oxidation sequence required to prepare 23. Direct arylation of 25 with 4ꢀ
azabenzoxazole to synthesize 6 was achieved using conditions adapted from the literature20.
ACS Paragon Plus Environment
22

Page 23 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
SuzukiꢀMiyaura crossꢀcouplings were used to prepare inhibitors 7, 10ꢀ13 and pyridineꢀ
substituted inhibitor 14 was obtained from a Stille crossꢀcoupling. Alternatively, bromide 25
could be borylated with bis(pinacolato)diboron21 to provide boronic ester 26, that was
subsequently crossꢀcoupled with 6ꢀbromoindole to afford inhibitor 9. Linear inhibitors 15 and
16 were prepared from iodide 27 using similar protocols.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Scheme 2. Synthesis of aryl substituted inhibitors 6-16.
CF3
O
N
F
CF3
CF3
16
O
N
F
Br
F
HO
Br
O
6
N
N
b
24
7, 10, 11, 12, 13
14
F
F
c
a
19
25
d
F
F
e
O
N
F
O
B
O
O
N
8
F
f
26
F
O
N
F
O
N
15
b
F
I
c
27
F
(a) 23, Cs2CO3, DMF, 40 oC, 3 d; (b) 4ꢀazabenzoxazole, KOAc, TBAB, CuI, (tꢀBu3P)2Pd,
DMF, 170 oC, 10 min.; (c) boronic acid, aq. Na2CO3, (Ph3P)4Pd, DMF, 170 oC, 10 min.; (d) 2ꢀ
ACS Paragon Plus Environment
23

Journal of Medicinal Chemistry
Page 24 of 37
1
2
(tributylstannyl)pyridine, (Ph3P)4Pd, DMF, 80 oC, 18 h; (e) bis(pinacolato)diboron, KOAc,
(dppf)PdCl2, DMF, 95 oC, 9 h; (f) 6ꢀbromoindole, aq. Na2CO3, (Ph3P)4Pd, DMF, 120 oC, 15 min.
3
4
5
6
7
8
9
CONCLUSION
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Substantial gains in potency were realized in the quinazolinone series of NS5B thumb pocket 2
allosteric inhibitors by extending into the left hand side of the binding pocket, towards residues
Pro 495 and Val 494. A structureꢀbased design approach was successfully applied for this
purpose. However, the use of established crystallographic protocols was limited due to ligand
induced crystal cracking. A molecular dynamics based structure prediction workflow was
therefore established and used as the main tool to support optimization efforts. This method
predicted a shift of the protein backbone in the loop region between Val 494 and Leu 497,
somewhat enlarging the pocket and, most importantly, improving accessibility of the backbone
carbonyl groups of Pro 495 and Val 494. The design of a structure with the capacity to form Hꢀ
bonds with these acceptor groups yielded an inhibitor with picomolar potency against the most
prevalent HCV genotypes (gt1a EC50 = 120 pM and gt 1b EC50 = 110 pM) and double digit
nanomolar activity against genotypes 2ꢀ6, inclusively. Compound 7 is one of the most potent
anti-HCV agents known to date that acts via allosteric NS5B inhibition. The high level of
antiviral activity provides significant leeway to balance potency with ADMETꢀPK properties
such as solubilityꢀlimited oral exposure. The identification of QAZ analogs with improved
overall profiles and potential for further development will be the subject of future reports. The
success of the structure based design efforts guided by the MD predictions underlines that the
MD workflow may be used in a broader fashion as a general protocol for modeling the
interactions of NS5B with diverse thumb pocket 2 ligands.
ACS Paragon Plus Environment
24

Page 25 of 37
Journal of Medicinal Chemistry
1
2
3
4
EXPERIMENTAL SECTION
5
6
7
X-ray crystallography
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Xꢀray crystal structures of NS5B inhibitor complexes were obtained by soaking preformed apo
NS5B crystals into a solution of inhibitor (1mM). For some of the inhibitors of pocket 2, it was
found that the diffraction resolution diminished proportionally to the soaking time. Successful
coꢀstructure data were obtained by adjusting the time of soaking such that sufficient pocket 2
inhibitor occupancy occurred without leading to unusable crystals due to cracking. Adjustments
were made through trialꢀandꢀerror by varying crystal size, soaking time and compound
concentration. For example, the structure of compound 6 was obtained by reducing the soaking
time to 3 hours and the concentration to 0.5 mM. Longer soaking time lead to loss of diffraction,
shorter soaking time resulted in low occupancy and poor inhibitor electron density maps.
Crystals of apo NS5B polymerase were obtained using the hanging drop technique with a
Nextal plate (Qiagen). The protein was prepared from constructs of HCV NS5B 1bJ4 strain that
lack the Cꢀterminal 21 amino acids and encode for a Cꢀterminal hexaꢀhistidine tag.22 The protein
was purifed as previously described23 and concentrated to 8.6 mg/ml in a buffer containing 25
mM tris at pH 7.5, 300 mM NaCl, 10% glycerol and 2 mM TCEP. The crystallization solution
used was made of 21% PEG 5000 mme, 100 mM MES pH 5.4, 10% glycerol and 410 mM
ammonium sulfate. Hanging drops of 2 ꢁl were incubated at 11 ˚C and crystals were grown
within 5 days. For soaking experiments, single crystals were transferred to a solution made of
14% PEG 5000 mme, 74 mM tris pH 7,5, 500 mM NaCl, 14% glycerol, 1% PEG 20K and 1 mM
of compounds.
Diffraction data were collected at the Swiss Light Source (SLS) PX1 beamline or locally on a
Rigaku FRE generator equipped with a MAR345 detector. The structures were solved by
ACS Paragon Plus Environment
25

Journal of Medicinal Chemistry
Page 26 of 37
1
2
3
4
5
6
7
8
9
molecular replacement using our published apo structure (3MWV), Model refinement was
performed using the CNX software (Accelrys). The NS5B coꢀstructures with compound 1, 2, 4, 6
and 14 have been deposited in the Protein Data Bank (accession codes 4JTW, 4JTY, 4JTZ, 4JU1
and 4JU2 respectively).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Structure prediction of NS5B:ligand complexes using molecular dynamics simulations
The Xꢀray structure with compound 1 served as starting point for the preparation of the
receptor coordinates for MD simulations. To limit the number of atoms and computational cost,
only the thumb domain (S365ꢀS563) and part of the finger loop (S19ꢀA39) were considered.
Harmonic restraints (50 kcal/mol) were applied to residues that are involved in interactions with
the deleted protein parts, to mimic the dampening effect of these contacts on the flexibility of
this set of interface residues (38 of a total of 220 considered residues, i.e. 83% of the protein
were fully flexible in the simulations). The restrained residue closest to compound 1 was located
at a distance of 13 Å, confirming that the binding pocket and its surroundings were fully flexible.
In order to focus the sampling of the simulations on the LHS region of the ligand structure,
harmonic distance restraints (1 kcal/mol) were introduced to favor the conserved Hꢀbond
interactions in the RHS region between the quinazolinone core and the backbone NH groups of
Ser 476 and Tyr 477.
In summary, the MD simulation protocol comprisedthe following steps: 1. Minimization, first
with strong restraints (50 kcal/mol) on the solute (protein ligand complex) which were
successively released in 3 minimization cycles (last cycle without restraints on nonꢀinterface
residues); 2. Equilibration in 3 steps, with an overall simulation time of 250 ps: a.) slow heating
to the simulation temperature of 298 K (harmonic restraints of 5 kcal/mol on nonꢀinterface
ACS Paragon Plus Environment
26

Page 27 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
residues, 50 ps), b.) pressure adjustment to 1 atm (harmonic restraints of 1 kcal/mol on nonꢀ
interface residues, 100 ps), c.) final NVT equilibration (no restraints on nonꢀinterface residues,
100 ps); 3. Production: NPT simulation, simulation time = 1.25 ns, no restraints on nonꢀinterface
residues (total simulation time = 1.5 ns). AMBER8 24 was used for all simulations, employing the
ff03 force field,25 AM1BCC charges26 for the ligand in combination with the gaff force field.27
Water was considered explicitly (TIP3P model), using a truncated octahedron solvent box with
the distance from the box surface to the closest solute atoms set to 12 Å. The overall protein
charge was neutralized by introducing amino acid changes in solvent exposed areas distant from
the ligand binding site (K20D, K379D, R394A, K441D, R465A, K510D, K535D). This
approach was chosen to avoid extended equilibration times that are often needed for the proper
configurational sampling of ions.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
Starting models for new ligands were obtained by modification of 1 and minimization within
the binding site, using MOE. Resulting binding mode hypotheses were exported as mol2 files
and submitted to a semiꢀautomatic structure prediction protocol (Figure 5): Ligand coordinates
from a mol2 file were automatically combined with the receptor coordinates. The corresponding
pdb file was the basis for generating the AMBER input files, which were transferred to our
Linux cluster. There, the MD simulation for each protein ligand complex was conducted on 8
cores of one computing node in parallel using sander. Upon completion, the output data was
transferred back to the workstation of the computational chemist. This process was automated
through a Perl program. Output trajectories were visually inspected, along with the rmsd trace of
the ligand coordinates, to identify distinct relevant binding poses that were sampled in the
simulation. The corresponding coordinate sets of the trajectory were averaged, and the average
was minimized to obtain the final predicted MD binding mode model. The simulation of one
ACS Paragon Plus Environment
27

Journal of Medicinal Chemistry
Page 28 of 37
1
2
3
4
5
6
7
8
9
protein ligand complex took approximately 10 hours and was usually done overnight, allowing
for an overall turnaround time of ~ 24 hours for setting up a simulation, running it, doing the
analysis and providing the results to the medicinal chemistry team.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
The MD structure prediction protocol was validated prior to its application in the project by
applying it to compound 4, for which an Xꢀray structure existed. As described above, compound
1 (in its corresponding Xꢀray structure) was modified to compound 4 and the resulting initial
model of the binding mode was submitted to the MD protocol. In the resulting MD model, the
rmsd from the Xꢀray structure amounted to 0.8 Å (ligand coordinates, nonꢀhydrogen atoms; see
supporting information for a figure of the superposition of MD model and Xꢀray structure).
These validation results and the successful predictive application of the established MD protocol
in the project suggest that it can be used as a general modeling approach for the interaction of
NS5B with thumb pocket 2 inhibitors, under consideration of protein flexibility.
HCV replicon cell based assays.
Cell culture EC50 potency and inhibition of HCV RNA replication with HCV subgenomic 1b
and 1a replicons were assessed as previously described using the cellꢀbased 1b luciferase
reporter assay or replicon assays using RTꢀPCR for RNA quantification.18
Additional replicon assays used HCV subgenomic luciferase reporter replicons that
incorporated the polio virus internal ribosomal entry site upstream of luciferase, increased assay
sensitivity and provided EC50 for reference compounds that were consistent with the previously
established luciferase reporter replicon results.
Potencies for nonꢀ1 genotype were determined by the use of chimeric gt1b sub genomic
replicons that substituted the endogenous NS5B with those from other genotypes. Briefly, the
ACS Paragon Plus Environment
28

Page 29 of 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
entire NS5B portion obtained from patient samples of the listed genotypes (2b, 3a, 4a, 5a and 6a)
was introduced into a bicistronic gt1b NS5B shuttle vector. The shuttle vector (p712) featured a
luciferase reporter gene in the first cistron that was preceded by the HCV 5’UTR, whereas the
second cistron encoded the NS2 to NS5B portion of the HCV nonꢀstructural proteins, the HCV
3’UTR and 3 copies of the HDV ribozyme. In vitro synthesized replicon RNA for each replicon
or chimeric replicon was introduced into HuH7ꢀJ cells by electroporation to generate stable cell
lines. These cell lines were then exposed for 72 hours to various serial dilutions of polymerase
inhibitors or control drugs (interferon alpha or protease inhibitors). The potency and half
maximal effect at inhibiting HCV RNA replication was assessed by measuring the luciferase
levels or the number of copies of replicon RNA by realꢀtime PCR in the case of the genotype 6a
chimera.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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
59
60
ASSOCIATED CONTENT
Supporting Information Available.
1H NMR spectra for inhibitors 1-7, 9-11, 13-16; 13C NMR spectra for compounds 1, 3ꢀ5, 7, 9-
11, 13ꢀ16; 19F NMR spectra for compounds 1, 2, 14ꢀ16; HRMS data for cpds 1, 3ꢀ5, 7, 9-11, 13ꢀ
16; Additional details on the setup of the NS5B protein for MD simulations; Validation of the
simulation protocol through application to compound 4. This material is available free of charge
via the Internet at http://pubs.acs.org.
Corresponding Author
*Dr. Oliver Hucke, oliver.hucke@boehringerꢀingelheim.com; ++49 7351 / 54ꢀ99462
ACS Paragon Plus Environment
29