Fragment-based discovery of hepatitis C virus NS5b RNA polymerase inhibitors

Article abstract of DOI:10.1016/j.bmcl.2008.03.056

Non-nucleoside inhibitors of HCV NS5b RNA polymerase were discovered by a fragment-based lead discovery approach, beginning with crystallographic fragment screening. The NS5b binding affinity and biochemical activity of fragment hits and inhibitors was de

Full text of DOI:10.1016/j.bmcl.2008.03.056

Available online at www.sciencedirect.com
Bioorganic & Medicinal Chemistry Letters 18 (2008) 2990–2995
Fragment-based discovery of hepatitis C virus NS5b
RNA polymerase inhibitors
Stephen S. Antonysamy,a Brandon Aubol,a Jeff Blaney,a Michelle F. Browner,b
b
b
a,
a
*
´
Anthony M. Giannetti, Seth F. Harris, Normand Hebert, Jo¨rg Hendle,
Stephanie Hopkins,a Elizabeth Jefferson,a Charles Kissinger,a Vincent Leveque,b
a
a
b
a
´
David Marciano, Ethel McGee, Isabel Najera, Brian Nolan,
Masaki Tomimoto,a Eduardo Torresa and Tobi Wrighta
aMedicinal Chemistry, SGX Pharmaceuticals, Inc., 10505 Roselle Street, San Diego, CA 92121, USA
bRoche Palo Alto, 3431 Hillview Avenue, Palo Alto, CA 94304, USA
Received 21 December 2007; revised 14 March 2008; accepted 18 March 2008
Available online 22 March 2008
Abstract—Non-nucleoside inhibitors of HCV NS5b RNA polymerase were discovered by a fragment-based lead discovery
approach, beginning with crystallographic fragment screening. The NS5b binding affinity and biochemical activity of fragment hits
and inhibitors was determined by surface plasmon resonance (Biacore) and an enzyme inhibition assay, respectively. Crystallo-
graphic fragment screening hits with ꢀ1–10 mM binding affinity (KD) were iteratively optimized to give leads with ꢀ200 nM bio-
chemical activity and low lM cellular activity in a Replicon assay.
ꢀ 2008 Elsevier Ltd. All rights reserved.
Hepatitis C virus is the leading cause of chronic liver dis-
ease throughout the world. Patients infected with HCV
are at risk of developing cirrhosis of the liver and subse-
quent hepatocellular carcinoma, and hence HCV is one
of the major reasons for liver transplantation. HCV
NS5b polymerase has been the focus of many drug dis-
covery efforts,1 but has proven to be a difficult target as
evidenced by the small number of clinical candidates tar-
geting this enzyme to date.2
graphic fragment screening also provides unambiguous
proof of binding to the target site and reveals the details
of the hits binding mode, providing clear direction for
how they may be optimized into more potent lead com-
pounds. Cocrystal structures of HCV NS5b RNA
dependent RNA polymerase (NS5b) with small mole-
cule inhibitors have been previously reported.5 We
developed experimental protocols for growing NS5b
crystals that were suitable for determining inhibitor
cocrystal structures by soaking.
Fragment-based lead discovery has been recently re-
viewed.3 Crystallographic fragment-based screening
seemed particularly well suited for a challenging target
such as HCV polymerase, since previous work has
shown that the high screening hit rate for such small,
low MW compounds can often identify novel, tractable
hits that can be optimized into potent leads with good
‘drug-like’ properties.4 In particular, crystallographic
fragment screening can identify hits that may be too
weak to be identified in biochemical assays. Crystallo-
Fragments in the screening library were chosen to be
consistent with ‘lead-like’ properties,6 and also to in-
clude two or more substituents to facilitate analog syn-
thesis. The fragment library had a MW range of 100–
220D and an average MW of 160.7 About half of the
fragments contain an aromatic bromine, both as a con-
venient substituent for synthesis and to facilitate deter-
mination of bound hits by collecting X-ray data at the
bromine anomalous dispersion wavelength. Ninety-six
mixtures of 10 fragments (5 mM concentration each)
were incubated with the NS5b protein crystals for
24 h. Datasets were obtained for 89 of 96 mixtures at
Keywords: HCV; NS5b; Fragment screening; Protein crystallography.
*
˚
˚
a typical resolution of 1.7 A (range 1.6–2.1 A). Twenty
individual hits were identified, 17 of which were bound
Corresponding author. Tel.: +1 858 558 4850; fax: +1 858 558
0642; e-mail: nhebert@sgxpharma.com
0960-894X/$ - see front matter ꢀ 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmcl.2008.03.056

S. S. Antonysamy et al. / Bioorg. Med. Chem. Lett. 18 (2008) 2990–2995
2991
at an allosteric site in the thumb domain.8 In most cases
the hit was unambiguously identified directly from the
mixture screen. Other hits with less well-defined electron
density were identified by soaking the individual com-
pounds from the mixture.
bromophenyl occupies a hydrophobic pocket in the
floor of the binding site. In a biochemical assay, which
measures the inhibition of H UTP incorporation into
RNA using a poly A RNA template,11 the IC50 value
of fragments discovered in the crystallographic screen
was in the 200–500 lM range, or was too weak to be
measured. The binding affinity of 1 and 4 to the NB5b
protein was in the millimolar range.12
3
Surface Plasmon Resonance (SPR) technology as imple-
mented with BIAcore instruments was applied initially
to detect protein–protein interactions, followed by pro-
tein–ligand interactions with high affinity (e.g., sub-
lM). Recent hardware and software improvements have
enabled the determination of binding affinities of smal-
ler, weaker ligands, such that SPR analysis can be used
to detect fragment binding. Using this technique we
were able to measure equilibrium binding affinities
(KD) for fragments and elaborated molecules with mod-
est affinity.
In parallel, the biochemical assay was used to identify
hits from the same fragment library screened at high
concentration. None of the biochemical hits, which
had IC50 values ranging from 2 to >100 lM, were ob-
served to bind to the NS5b protein crystals in soaking
experiments. SPR analysis of the biochemical hits also
failed to detect any binding affinity.
The initial optimization efforts were focused on 1 which
offered a number of possibilities for elaboration. The
goal was to increase the affinity and biochemical activity
by creating additional interactions in the binding pocket
between R501 and S476 and Y477 (Fig. 2).
The fragments found to bind to the thumb site had a
number of common features (Fig. 1), most notably the
frequent presence of an aryl or heteroaryl bromide, and
a preponderance of carboxylic acids. The hits tolerated
a range of substituents meta and para to the bromine.
Interestingly, the corresponding chloro substituted ana-
logs were not observed crystallographically.
A second class of fragments was also discovered in the
initial crystallographic fragment screen (3 and 4 are rep-
resentative examples). In this case, the bromoaryl moi-
ety binds in the pocket as in 1, but the amide carbonyl
oxygen interacts with the S476 backbone NH in the car-
boxylic acid binding site. The succinic acid electron den-
sity is well defined, with the carboxylate interacting with
the S476 hydroxyl on the surface of the protein. Since
this is a surface residue, it was unclear how much bind-
ing affinity was due to this interaction and whether it
was critical for binding. An initial library of anthranila-
mides 7 was made using a diverse selection of primary,
secondary, and aromatic amines (Scheme 1). A second
set of amines, distinct from the first, was selected and
coupled to 5-bromo anthranilic acid 1, and then treated
with succinic anhydride to give the succinamide prod-
ucts 8 (Scheme 1).
A number of hits were observed to interact with the car-
boxylic acid binding site (S476 and Y477) in a manner
analogous to the published Shire inhibitor 6.9,10 For
example, 5 is involved in two hydrogen bonds to S476
and Y477 backbone NHs. These interactions were also
observed with other carboxylic acids. In the case of 2,
the carboxylic acid binds to S476 and Y477 while the
Figure 1. Some of the fragment hits observed at Thumb site in NS5b
crystal structures, and Reference inhibitor 6 from reference 2.
Figure 2. Crystal structure of fragments 1 and 4 bound to NS5B.

2992
S. S. Antonysamy et al. / Bioorg. Med. Chem. Lett. 18 (2008) 2990–2995
non-ideal behavior such as aggregation, poor solubility,
or non-stoichiometric binding, which was presumed to
reflect non-specific inhibition (e.g., 7b and 7c).
Based on these initial results, we prepared additional
analogs of 7 to probe the binding site further. An ex-
panded set of R1, including substituted pyrrolidines,
piperidines, piperazines, and azepines, as well as bicyclic
amines, was coupled to 5-bromoanthranilic acid 1. The
second set of compounds retained the most active 3,5-
cis-dimethylpiperidine at R1, coupled to a series of 5-
and 6-substituted 3-bromo or 3-chlorobenzoic acids.
These were designed to probe the steric and electronic
preferences in the hydrophobic pocket, as well as
attempting to extend into the groove. Biochemical
screening of these second round arrays revealed no com-
pounds with greater inhibition than compound 7e.
Scheme 1. Reagents and condition: (a) 1, R1 Amine (1.2 equiv),
HATU (1.2 equiv), DIEA (1.2 equiv), DMF; (b) Succinic anhydride
(1.2 equiv) Toluene, 90 ꢁC.
The initial set of compounds was screened and a number
of derivatives showed biochemical inhibition in the 17–
500 lM range, with comparable SPR binding affinities
(Table 1). From this initial screening set, it was clear that
secondary amides were less active than tertiary, and that
cyclic amides were preferred over acyclic amides. The
measured binding affinities were generally in agreement
with the biochemical results. SPR analysis also allowed
us to identify and exclude compounds that displayed
Additional analogs of 7e with substitution ortho to the
aniline provided other insights. A number of small sub-
stituents were tolerated in crystal structures (Cl, CH3)
but all were less potent than 7e. To provide insight for
this lack of improvement, a number of compounds were
soaked into NS5b crystals. These crystal structures
showed that the 3,5-dimethylpiperidine amide makes a
number of critical interactions with the protein. The
amide carbonyl is twisted out of the plane of the aryl
group and interacts with the guanidine of R501
(Fig. 3). Both methyl groups on the piperidine appear
to be necessary, since removing one or both methyls
(e.g., 7f and 7g) reduces activity. The dimethyl piperi-
dine also seems to have optimal interactions with the
floor of the binding site, since no other amines profiled,
including substituted morpholines and thiomorpholines,
offered any improvement. Pyrrolidines, tetrahydroiso-
quinolines, azepines, and piperazines also did not im-
prove activity beyond 7e. Crystal structures reveal that
although several pyrrolidine derivatives (e.g., 7c) bind
to the site, their binding mode is different from 7e. In
these compounds, the amide carbonyl oxygen no longer
Table 1. RNA polymerase inhibition of 5-bromoanthranilic acid
derivatives
Compound
R1
IC50 (lM)
KD (lM)
7a
>500
nd
7b
178
nd
7c
7d
7e
7f
7g
7h
7i
57
132
ins
210
14
17
170
192
nd
nd
36
>500
>500
190
7j
>500
nd
Values are means of duplicate experiments. KD was measured by
Biacore. Standard deviations are 50% of the reported values.
Figure 3. Crystal structures of 7c and 7e bound to NS5b.

S. S. Antonysamy et al. / Bioorg. Med. Chem. Lett. 18 (2008) 2990–2995
2993
interacts with R501, but rather hydrogen bonds to the
backbone NH of S476. In this conformation, there is a
large portion of the site unfilled. In compounds with a
substituent ortho to the aniline, the aromatic ring ro-
tates to accommodate the substituent. This prevents
the ring from pushing down into the pocket to the same
extent as the unsubstituted analogs.
As can be seen in Table 2, incorporation of the succinic
acid group into the preferred compound 7e to provide
compound 8e showed a 17-fold improvement over the
parent, and was found to maintain the interactions of
the succinamide fragment 4. Interestingly, introduction
of the succinate into other anthranilamides (e.g., 8a–c)
led to a marked decrease in biochemical activity. We
therefore focused on determining which of the interac-
tions made by the succinamide moiety were responsible
for the gain in potency in going from 7e to 8e.
Table 2. RNA polymerase inhibition of succinate derivatives
Truncation of the terminal acid and replacement with an
acetamide or sulfonamide (8f and 8m), or ester forma-
tion (8g) were found to be deleterious. Various amides
were also tested. The dimethyl amide 8h and the mor-
pholine amide 8i were 2.5- and 4-fold more active than
the carboxylic acid, respectively. Examination of the
cocrystal structures of acid 8e (Fig. 4) showed that the
carbonyl adjacent to the aniline nitrogen interacts with
the backbone NH of S476 as in fragment 4, while the
carboxylic acid binds to the S476 hydroxyl and the imid-
azole of H475. In the case of the morpholine amide 8i,
the interaction of the anilide carbonyl with S476 was un-
changed, but the morpholine stacks with the imidazole
ring instead of forming a hydrogen bond. The primary
amide 8j was 18-fold less active than the morpholine
amide 8i. A 1,4-dicarbonyl arrangement appears to be
preferred, since glutarate 8k and malonamide 8l were
both 10-fold weaker than parent 8e. It is likely that a
1 or 3 atom linker between the carbonyls does not allow
the terminal group to interact productively with S476
and H475.
Compound R1
R2
IC50 KD
(lM) (lM)
8a
8b
>500 nd
>500 nd
8c
8e
533
1.0
nd
0.31
8f
8g
8h
8i
11
1.7
At thispoint we focusedourefforts in three directions: first,
optimizing the stacking interaction with H475; second,
reducing the rotational freedom of the succinate moiety;
and third, exploring other 1,4-dicarbonyl arrangements.
2.1
0.4
0.25
4.3
9
ins
Starting from 7e, ketoamides such as 9a–c were obtained
by acylation of the aniline (Scheme 2). The results
0.30
0.46
9.4
8j
8k
nd
8l
11
nd
nd
8m
7.5
Values are means of duplicate experiments. KD were measured by
Biacore. Standard deviations are 50% of the reported values.
ND, not determined; Ins, insoluble.
Figure 4. Crystal structure of 8e and 8i bound to NS5b.

2994
S. S. Antonysamy et al. / Bioorg. Med. Chem. Lett. 18 (2008) 2990–2995
Table 3. RNA polymerase inhibition of succinate derivatives
Compound
R2
IC50 (lM)
KD (lM)
Scheme 2. Reagents: (a) 7e, RCOOH (1.2 equiv), HATU (1.2 equiv),
DIEA (1.2 equiv), DCM/DMF; (b) TFA/DCM, then, RCOOH (1.2
equiv), HATU (1.2 equiv), DIEA (1.2 equiv), DCM/DMF; (c) COCl2,
DCM, NaHCO3, then, RNH2.
9a
0.44
0.97
showed that the terminal amide could be replaced with
an aryl ketone, and provide similar biochemical activity
(Table 3).
9b
0.54
0.59
0.60
1.85
1.04
0.46
9c
Replacement of C2 of the succinate with nitrogen main-
tained potency as can be seen comparing the urea deriv-
ative 10a with its direct analog 8h. Interestingly,
compounds such as 10b–d, which replaced the terminal
amide with an aromatic or heteroaromatic ring were
also active, indicating that the carbonyl is not essential,
but can be replaced by another group capable of p
stacking with the H475. The results of compounds 9
and 10 are consistent with the notion that the stacking
interaction with H475 is providing increased activity,
and that this is possible with a range of different substit-
uents. Since these interactions occur in a solvent exposed
region there is little steric discrimination to prevent
groups differing widely in size from binding efficiently.
As can be seen in the urea series, both electron-rich
and electron-poor rings interact favorably in this region.
10a
10b
10c
1.0
0.80
0.35
0.36
10d
11a
0.17
0.13
0.37
0.54
Inspired by the positive results with 9 and 10, a series of
amino acid amide derivatives (11a–e) were also pre-
pared. Glycine derivatives 11a–c, which replaces a car-
bon in 9 with nitrogen, showed similar biochemical
potencies as well as binding affinities. Of the many cyclic
analogs prepared (not shown), only proline analogs such
as 11d and 11e exhibited less than 10 lM potency. These
observations indicate that analogs with restricted rota-
tion between the carbonyl moieties derived their potency
from a dependency on the conformation of the side
chain.
11b
11c
11d
0.43
0.25
0.35
0.57
1.02
1.28
Based on both biochemical activity and binding affinity,
a set of compounds was selected for testing in the Rep-
licon assay (Table 4). Compounds with non-ideal behav-
ior in the SPR assay were discarded.
The most active was the proline analog 11e, possessing
an EC50 of 3.7 lM. The correlation of the replicon activ-
ity of these analogs with the SPR affinity and with the
biochemical activity is observed, although there are out-
liers in both cases. Nonetheless, both assays taken to-
gether were good predictors of Replicon activity.
Interestingly, these results demonstrate that compounds
without a charged group can interact strongly with the
carboxylic acid binding site, S476 and Y477.
11e
1.03
0.15
Values are means of duplicate experiments. KD were measured by
Biacore. Standard deviations are 50% of the reported values.
resulting in compounds with nanomolar affinity using
structure-based design. SPR analysis, using BIAcore
technology, was critical in the early stages of this work,
as most of the initial fragments showed undetectable
In summary, this work demonstrates that fragments
with millimolar binding affinity discovered by crystallo-
graphic screening can be improved over 1000-fold

S. S. Antonysamy et al. / Bioorg. Med. Chem. Lett. 18 (2008) 2990–2995
2995
Chem. 2005, 5, 751; Oprea, T. I.; Davis, A. M.; Teague, S. J.;
Leeson, P. D. J. Chem. Inf. Comput. Sci. 2001, 41, 1308.
7. Blaney, J.; Nienaber, V.; Burley, S. K. In Fragment-based
Approaches in Drug Discovery; Jahnke, W., Erlanson, D.,
Eds.; Wiley-VCH: Weinheim, Germany, 2006; p 215.
8. For a description of the structure of NS5b see Bressanelli,
S.; Tomei, L.; Roussel, A.; Incitti, I.; Vitale, R. L.;
Mathieu, M.; De Francesco, R.; Rey, F. A. Proc. Natl.
Acad. Sci. U.S.A. 1999, 96, 13034; Ago, H.; Adachi, T.;
Yoshida, A.; Yamamoto, M.; Habuka, N.; Yatsunami,
K.; Miyano, M. Structure 1999, 7, 1417.
9. Chan, L.; Reddy, T. J.; Proulx, M.; Das, S.; Pereira, O.;
Wang, W.; Siddiqui, A.; Yannopoulos, C. C.; Poisson, C.;
Turcotte, N.; Drouin, A.; Alaoui-Ismaili, M. H.; Bethell,
R.; Hamel, M.; L’Heureux, L.; Bilimoria, D.; Nguyen-Ba,
N. J. Med. Chem. 2003, 46, 1283.
10. Chan, L.; Pereira, O.; Reddy, t. J.; Das, S. K.; Poisson, C.;
Courchesne, M.; Proulx, M.; Siddiqui, A.; Yannopoulos,
C. G.; Nguyen-Ba, N.; Roy, C.; Nasturica, D.; Moinet, C.;
Bethell, R.; Hamel, M.; L’Heureux, L.; David, M.;
Nicholas, O.; Courtemanche-Asselin, P.; Brunette, S.;
Bilimoria, D.; Bedard, J. Bioorg. Med. Chem. Lett. 2004,
14, 797.
Table 4. Replicon activity for selected analogs
Compound
Replicon activity EC50 (lM)
8e
8h
22.8 1.5
6.1 1.0
7.4 1.6
4.3 0.3
7.9 0.4
5.6 0.5
4.9 0.7
6.9 0.6
4.6 0.9
5.5 0.4
5.7 0.7
8.7 0.1
12.9 2.9
3.7 0.2
8i
9a
9c
10a
10b
10c
10d
11a
11b
11c
11d
11e
Values are the mean of at least n = 2 experiments.
biochemical activity. SPR analysis also helped diagnose
biochemical false positives during fragment optimiza-
tion; several of these analogs did not bind to the enzyme
based on both SPR and crystallographic analysis. Using
these tools, neutral inhibitors of NS5b were discovered
and shown to possess interesting cellular activity. Fur-
ther progress on this series will be reported in due
course.
11. Each 50 ll enzymatic reaction contains 8.4 lg/mL poly-
A:oligo U16 (template:primer), 200 nM NS5B570n-BK
enzyme, 2.1 lCi of tritiated UTP (Perkin Elmer catalog
No. TRK-412; specific activity: 30–60 Ci/mmol; stock
solution
concentration
from
7.5 · 10À5 M
to
20.6 · 10À6 M), 1 lM ATP, CTP, and GTP, 40 mM
Tris–HCl, pH 8.0, 2–40 mM NaCl, 4 mM DTT (dithio-
threitol), 4 mM MgCl2, and 5 ll of compound serial
diluted in DMSO. Reaction mixtures are assembled in
MADVN0B 96-well filter plates (Millipore Co.) and
incubated for 2 h at 30 ꢁC. Reactions are stopped by the
addition of 10% (v/v) trichloroacetic acid and incubated
for 40 min at 4 ꢁC. Reactions are filtered, washed with 6
reaction volumes of 10% (v/v) trichloroacetic acetic acid, 2
reaction volumes of 70% (v/v) ethanol, air dried, and 25 ll
of scintillant (Microscint 20, Perkin-Elmer) is added to
each reaction well. The amount of light emitted from the
scintillant is converted to counts per minute (CPM) on a
Topcountꢂ plate reader (Perkin-Elmer, Energy Range:
Low, Efficiency Mode: Normal, Count Time: 1 min,
Background Subtract: none, Cross-talk reduction: Off).
12. NS5B570n-BK enzyme was minimally biotinylated and
captured onto a streptavidin surface to a density of
ꢀ6000 RU. The compounds, supplied at 10 mM in
100% DMSO, were diluted in running buffer containing
50 mM HEPES, 600 mM NaCl, 10 mM MgCl2, 5 mM
DTT, pH 8.0, to a concentration of 5%. All com-
Acknowledgments
We gratefully acknowledge the support of Roche Palo
Alto, Dr. David Myszka of the University of Utah,
Dr. John Cashman of the Human Biomolecular Re-
search Institute, San Diego, CA.
References and notes
1. Gordon, C. P.; Keller, P. A. J. Med. Chem. 2005, 48, 1;
Tan, S.-L.; Pause, A.; Shi, Y.; Sonenberg, N. Nat. Rev.
Drug Disc. 2002, 1, 867.
2. Hirashima, S.; Suzuki, T.; Ishida, T.; Noji, S.; Yata, S.;
Ando, I.; Komatsu, M.; Ikeda, S.; Hashimoto, H. J. Med.
Chem. 2006, 49, 4721.
3. Erlanson, D. A.; McDowell, R. S.; O’Brien, T. J. Med.
Chem. 2004, 47, 3463; Hartshorn, M. J.; Murray, C. W.;
Cleasby, A.; Frederickson, M.; Tickle, I. J.; Jhoti, H.
J. Med. Chem. 2005, 48, 430; Hajduk, P. J.; Greer, J. Nat.
Rev. Drug Disc. 2007, 6, 211.
4. Nienaber, V. L.; Richardson, P. L.; Klighofer, V.; Bouska,
J. J.; Giranda, V. L.; Greer, J. Nat. Biotechnol. 2000, 18,
1105; Rees, D. C.; Congreve, M.; Murray, C. W.; Carr, R.
Nat. Rev. Drug Disc. 2004, 3, 660.
pounds were run in
a 3-fold dilution series using
50 mM HEPES, 600 mM NaCl, 10 mM MgCl2, 5 mM
DTT, 5% DMSO, pH 8.0 as the running buffer. Each
compound was flowed across the HCV surface and a
streptavidin control surface. For each dataset, the
responses at equilibrium (t = 60 s) were plotted versus
the compound concentration. Data points were fit to a
simple binding isotherm to obtain the interaction
affinity. More potent compounds exhibiting observable
5. The PDB ID codes for complexes discussed in this paper
are: 3CIZ, 3CJ0, 3CJ2, 3CJ3, 3CJ4, 3CJ5.
kinetics were fit to
a 1:1 kinetic model. All data
reduction and fitting was done with the Scrubber II
software package (BioLogic Software, Pty., Australia).
6. Schuffenhausen, A.; Ruedisser, S.; Marzinzik, A. L.; Jahnke,
W.; Blommers, M.; Selzer, P.; Jacoby, E. Curr. Top. Med.