Lead generation of heat shock protein 90 inhibitors by a combination of fragment-based approach, virtual screening, and structure-based drug design

Article abstract of DOI:10.1016/j.bmcl.2011.08.001

Heat shock protein 90 (Hsp90) is a molecular chaperone which regulates maturation and stabilization of its substrate proteins, known as client proteins. Many client proteins of Hsp90 are involved in tumor progression and survival and therefore Hsp90 can be a good target for developing anticancer drugs. With the aim of efficiently identifying a new class of orally available inhibitors of the ATP binding site of this protein, we conducted fragment screening and virtual screening in parallel against Hsp90. This approach quickly identified 2-aminotriazine and 2-aminopyrimidine derivatives as specific ligands to Hsp90 with high ligand efficiency. In silico evaluation of the 3D X-ray Hsp90 complex structures of the identified hits allowed us to promptly design CH5015765, which showed high affinity for Hsp90 and antitumor activity in human cancer xenograft mouse models.

Full text of DOI:10.1016/j.bmcl.2011.08.001

Bioorganic & Medicinal Chemistry Letters 21 (2011) 5778–5783
Contents lists available at SciVerse ScienceDirect
Bioorganic & Medicinal Chemistry Letters
journal homepage: www.elsevier.com/locate/bmcl
Lead generation of heat shock protein 90 inhibitors by a combination
of fragment-based approach, virtual screening, and structure-based drug design
Takaaki Miura a, , Takaaki A. Fukami a, Kiyoshi Hasegawa a, Naomi Ono a, Atsushi Suda a, Hidetoshi Shindo a,
⇑
Dong-Oh Yoon b, Sung-Jin Kim b, Young-Jun Na b, Yuko Aoki a, Nobuo Shimma a, Takuo Tsukuda a,
Yasuhiko Shiratori a
a Kamakura Research Laboratories, Chugai Pharmaceutical Co., Ltd, 200 Kajiwara, Kamakura, Kanagawa 247-8530, Japan
b C&C Research Laboratories, 146-141 Annyeong-dong, Hwaseong-si, Gyeonggi-do 445-380, Republic of Korea
a r t i c l e i n f o
a b s t r a c t
Article history:
Heat shock protein 90 (Hsp90) is a molecular chaperone which regulates maturation and stabilization of
its substrate proteins, known as client proteins. Many client proteins of Hsp90 are involved in tumor pro-
gression and survival and therefore Hsp90 can be a good target for developing anticancer drugs. With the
aim of efficiently identifying a new class of orally available inhibitors of the ATP binding site of this pro-
tein, we conducted fragment screening and virtual screening in parallel against Hsp90. This approach
quickly identified 2-aminotriazine and 2-aminopyrimidine derivatives as specific ligands to Hsp90 with
high ligand efficiency. In silico evaluation of the 3D X-ray Hsp90 complex structures of the identified hits
allowed us to promptly design CH5015765, which showed high affinity for Hsp90 and antitumor activity
in a human cancer xenograft mouse model.
Received 4 June 2011
Revised 29 July 2011
Accepted 1 August 2011
Available online 6 August 2011
Keywords:
Virtual screening
Fragment-based approach
Hsp90 inhibitor
Ó 2011 Elsevier Ltd. All rights reserved.
The 90-kDa heat shock proteins (Hsp90) are ATP-dependent
molecular chaperones which are responsible for the stabilization
and maturation of their substrate proteins, known as client pro-
teins.1 The client proteins of Hsp90 include many oncology targets
such as p53 mutants, Raf1, Akt, and Bcr-Abl, which are important
for tumor generation and progression. Therefore, an inhibition of
Hsp90 could lead to an effective suppression of tumors by interfer-
ing with multiple oncogenic pathways simultaneously. Although
there once was concern about the adverse effect of a ubiquitous
expression of Hsp90 in both normal and tumor cells, it has since
been shown that a selective suppression of tumor cells was indeed
achievable with the discovery of selective Hsp90 inhibitors, gel-
danamycin and radicicol.2 When geldanamycin derivatives,
17AAG and 17DMAG, subsequently proceeded to clinical trials, it
stimulated the community to make significant efforts to discover
and develop a different class of Hsp90 inhibitors with better DMPK
profiles and antitumor activities.1,3
In this report, we describe our attempt to generate a new class
of orally available Hsp90 inhibitors by a combinatorial approach of
fragment screening, virtual screening and molecular modeling
with the assistance of X-ray structures of Hsp90–ligand complexes.
Several groups have reported the discovery of Hsp90 inhibitors by
virtual screening or fragment screening.4,5 Also, the combination of
fragment and virtual screening for Hsp90 inhibitor generation was
recently reported.6 In this study, we adopted the parallel approach
of fragment and virtual screening and the information obtained
from the two approaches was utilized directly with the aim of gen-
erating inhibitors more efficiently and effectively. In fragment
screening, only a small set of fragments selected for targeting the
ATP binding site was screened, and the potentially limited chemi-
cal space when searching in the fragment screening was compen-
sated for by conducting virtual screening, which explored for wider
chemical space. The binding of small molecule ligands to Hsp90
can be characterized by two important interactions7,8: (1) the
hydrogen bonding network mediated by the carboxylic side chain
of Asp93 and crystal waters; (2) two hydrophobic pockets formed
by Leu107-Phe138 and Ile96. Notably, the hydrogen bonding net-
work involving Asp93 and associated crystal waters is conserved
for all ATP site binders (Fig. 1). Thus, our strategy was to first ex-
ploit the Asp93 site in the fragment and virtual screening to iden-
tify initial binders and then, with the help of the X-ray structures of
Hsp90-inhibitor complexes, to design new inhibitors based on the
identified hits to gain additional interaction with the protein.
In humans, two isoforms of Hsp90, namely, Hsp90
known. In this study, the N-terminal domain (9–237) of human
Hsp90 was used for all biophysical analysis. The domain was ex-
a and b, are
a
pressed in Escherichia coli, and purified to homogeneity by stan-
dard column chromatography. For ATPase assay, full-length
HSP90a expressed as N-terminally GST-tagged form in E. coli,
was employed. The issue with the ATPase assay was its inherent
low sensitivity caused by the weak Hsp90 ATPase activity.
⇑
Corresponding author. Tel.: +81 467 47 6019; fax: +81 467 45 6824.
E-mail address: miuratka@chugai-pharm.co.jp (T. Miura).
0960-894X/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmcl.2011.08.001

T. Miura et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5778–5783
5779
Figure 1. Schematic drawing of the interaction between Hsp90a and ADP (a) and geldanamycin (b).
Figure 2. Chemical structures of representative hits. (a) Fragment hits; (b) 2-aminopyrimidine and its analogues; (c) hit compounds from hit expansion of fragment hits; (d)
hit compound from virtual screening. Kd, LEI, and IC50 values are shown together with chemical structure when available. LEI was calculated using the equation, LEI =
DG/(no.
of heavy atoms).12
In fragment screening, surface Plasmon resonance (SPR)-based
binding assay was employed to screen 164 purchased fragments
which possess a donor–acceptor pattern complimentary to the
hydrogen bonding network of the Asp93 site. A highly active and
stable surface of Hsp90 was prepared by coupling minimally bio-
tinylated Hsp90 to the streptavidine-coated surface of a sensor
chip (type SA, GE healthcare) (Supplementary Fig. 1). Nineteen
compounds were classified as active, showing more than 50% occu-
pancy of the immobilized Hsp90 at the screening concentration of
to Hsp90 (Supplementary Fig. 2).9 The second were 2D 1H–15N-
HSQC chemical shift perturbation experiment in which determin-
ing the chemical shift difference experienced by the HN-group of
each amino acid allowed us to map the binding sites of the frag-
ment hits (Supplementary Fig. 3).10 The analysis indicated that
all 13 fragments did indeed bind to the designated active site of
the protein. Interestingly, 8 out of 13 validated fragment hits pos-
sess 2-aminopyrimidine and 2-aminotriazine as core motifs, as
shown in Figure 2a. This finding prompted us to determine the
affinities of 2-aminotriazine and its analogues for Hsp90 so as to
prioritize them as a core motif for hit expansion by substructure
search. The dissociation constants of the compounds were deter-
mined by recording a series of 2D NMR NH-correlation spectra at
various concentrations of the test fragment and by following the
size of chemical shift changes of NH crosspeaks (Fig. 2b). The result
indicated that 2-aminopyrimidine, 2-aminotriazine and indazole
250 lM. Reproducibility and dose dependence of the binding re-
sponse narrowed down these 19 initial hits to 13 compounds.
Then, these 13 fragments were subjected to two different types
of NMR experiments to verify that they bound to the ATP pocket
of Hsp90. The first were 1D 1H displacement experiments that
examined if the addition of geldanamycin suppressed the line-
broadening of proton signals of a test fragment induced by binding

5780
T. Miura et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5778–5783
Figure 3. X-ray structures of N-terminal Hsp90
a
in complex with identified compounds. Detailed binding mode of the active site and the overall structure are shown. The
overall structures are colored according to the secondary structure with
a-helices in red and b-strands in cyan. The flexible loops consisting of residues 108–112, are colored
in yellow. (a and b) Fragment hit, 1; (c and d) fragment hit, 3; (e and f) a hit from virtual screening, 4.
showed better affinity among those tested. Therefore, we collected
from our in-house library, 710 compounds possessing these moie-
ties as the core structures. These compounds were screened using
molecules around the carboxylic group of Asp93 were also in-
cluded as a part of the binding site, as these three water molecules
were consistently found in the X-ray structures of Hsp90 in com-
plex with small molecules binding to the ATP site. Following the
docking by FlexX, the 3000 highest-scoring compounds were se-
lected for further analysis and each compound was visually in-
spected on the graphics as to whether they bound to the Asp93
site as designated and/or were chemically tractable. After this fil-
tering process, 1460 compounds remained and were tested for
the ATPase assay with full-length Hsp90
of a higher throughput of the ATPase assay. Notably, 16 com-
pounds showed IC50 values less than 100 M with the most potent
a
11 instead of SPR because
l
compounds, 2 and 3, having IC50 values of 24 and 9.6 lM, respec-
tively (Fig. 2c). Calculation of ligand efficiency index12 (LEI = À
DG/
(the number of heavy atoms in the ligand)) for fragment hits and
active compounds from the hit expansion indicated that good li-
gand efficiency was maintained for the hits from the expansion,
although the highest LEI was found for a hit fragment, 1, with a
LEI of 0.54.
their inhibitory activity against Hsp90
100
M.11 Seventy-three compounds showed an inhibitory activity
larger than 30% and an outstanding activity was observed for com-
pound 4 with an IC50 value of 6.9 M and LEI of 0.43 (Fig. 2d).
a by ATPase assay at
l
l
Our second approach for identifying Hsp90 inhibitors was vir-
tual screening using a docking tool, FlexX.13 From an in-house
source, 435,000 compounds were used to dock for the screening.
The program CORINA14 was employed to convert these compounds
in the docking library from 2D to 3D and the complex structure be-
tween Hsp90 and geldanamycin (PDB entry: 1YET) was used as a
template. The active site was defined as a sphere consisting of
the atoms within 10 Å from geldanamycin. In addition, three water
In order to obtain detailed structural information on the binding
mode of the compounds, crystallization was performed15 and the
X-ray crystal structures of the complex with Hsp90 were solved
for 1, 3, and 4.16 As expected, the 2-aminopyrimidine and triazine
moieties of these compounds are accommodated in the Asp93 site
in the same manner but also differences were found among these
structures in the conformation of residues Ile110-Gly114. This part
consists of a central portion of a long a-helix in the structure of 3,

T. Miura et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5778–5783
5781
Figure 4. Detailed analyses of compound 4. (a) The GRID analysis for compound 4. The location of DRY probe is shown in cyan. (b) SAR to evaluate the effect of the ortho-
substitution at the left phenyl ring. (c) Detailed binding mode of compound 6 to N-terminal-Hsp90
a
.
while it shows the significant rearrangement in the structures of 1
and 4, resulting in ‘semi-open’ and ‘open’ conformations, respec-
tively (Fig. 3). Such a conformational flexibility of this part of the
molecule was reported previously.17 It would be possible to exploit
Combining all the information obtained together, we designed
and synthesized the ‘hybrid’ compound CH5015765 which has 2-
aminotriazine as a core and an S-methyl group at the C-4 posi-
tion and an ortho-Cl-tricyclic group at the C-6 position (Fig. 5a).
Remarkably, this compound showed significantly improved bind-
ing affinity over the hit compounds from fragment screening and
virtual screening, with a Kd value of 3.4 nM and LEI of as high as
0.50. No significant change was found for the inhibitory activity
of the ATPase assay, but this was due to an inherent low sensi-
tivity of the assay system. The binding mode of the compound
was not altered, as revealed by the X-ray structure analysis
(Fig. 5b).16 In addition, in vitro cell growth inhibition (CGI) assay
using HCT116 cells showed that this compound had CGI activi-
a hydrophobic pocket created by the formation of
a-helix in the
structure of 3 to design more potent inhibitors, but we focused
on 1 and 4, as the shapes of the binding pocket of 1 and 4 are more
similar to each other and therefore it was expected to be more
straightforward to design new inhibitors by merging the structural
information on 1 and 4. In addition, the slower dissociation of 4
with 0.10/s when compared to 3 (>1/s), was preferred as an initial
hit (data not shown). A fast dissociation observed for 3 suggested
that the formation of a-helix in the residues Ile110-Gly114 would
not be a rate-limiting step, but that this portion would be in the
very rapid equilibrium between the different conformations.
In order to search for additional interaction points for increas-
ing potencies of 4, the binding site of the obtained X-ray structure,
was analyzed using the program GRID (Molecular Discovery,
http://moldiscovery.com).18,19 A regular grid was built over the
rectangular box defined by the size of compound 4 as found in
the complex structure of 4 and Hsp90. Then for each grid, the inter-
action energy of the DRY probe, which represents steric and hydro-
phobic interactions, was computed using the GRID force field. The
GRID descriptor thus calculated revealed the small hydrophobic
space formed by Leu107 and Phe138 close to the tricyclic group
of 4, which could be filled with methyl group or halogen atom
(Fig. 4a). In fact, model compounds 5–7, showed that an introduc-
tion of methyl group or chlorine atom at the ortho-position of the
phenyl ring led to a remarkable improvement of the affinity over
the non-substituted form, 5, with the Kd values 11 or seven times
smaller than that of 5, respectively (Fig. 4b). Concomitantly, LEI
was increased from 0.39 to 0.45 or 0.44. The X-ray structure of 7
confirmed that the introduced chloro atom occupies the small
hydrophobic pocket as expected (Fig. 4c).16
ties with an IC50 value of 0.46 lM. In order to confirm that the
observed CGI was due to the suppression of Hsp90 and not
due to the general toxic effect, the down-regulation of the
Hsp90 client proteins, Her2, Raf1 and pERK, was measured at
0.5, 5, and 50 lM of the compound (Fig. 5c). The analysis of
the HCT116 cells treated with the inhibitor demonstrated that
such down-regulation was indeed induced in a manner depen-
dent on the inhibitor concentrations. PK profile of this compound
indicated some oral bioavailability in mouse (4.5%), despite the
high plasma clearance of 19.5 mL/min/kg, which encouraged us
to measure in vivo antitumor activity of this compound
(Fig. 5d). Remarkably, when CH5015765 was administered orally
to mice, it exhibited 54% inhibition of tumor volume at a dose of
400 mg/kg in a human stomach cancer NCI-N87 xenograft mod-
el, without any loss of body weight (Fig. 5e.).
Noteworthy is that, when LEI of the inhibitors is plotted against
molecular weight, LEI is actually not only maintained, but rather
improved over the course of the modification (Fig. 6). This was
uncommon as, in general, LEI decreases or, at most, stays at a com-
parable value as the molecular weight of compounds increases
during compound optimization.20 This clearly demonstrates that

5782
T. Miura et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5778–5783
Figure 5. Biological activities of CH5015765. (a) Chemical structure of CH5015765 and its in vitro activities. (b) Binding mode of CH5015765 in Hsp90. (c) Effects of
CH5015765 on expression of Her2, Raf1, pERK and ERK. Actin was used as a loading control. (d) PK profile of CH5015765 in mouse. (e) In vivo antitumor activity of
CH5015765 in human stomach cancer NCI-N87 xenograft mouse model.
our combined approach was extremely robust in efficiently gener-
ating a good lead compound.
In summary, we discovered a new class of orally available
Hsp90 inhibitor with in vivo antitumor activity, by combining
the information derived from the hits from the two types of screen-
ing and molecular modeling/X-ray analysis. The information from
the screenings was more directly used to design the new inhibitor
in our study. This combined and direct use of the information from
the different approaches was a key for success in quickly develop-
ing the potent inhibitor. Further modification of CH5015165 to im-
prove its physicochemical property and oral bioavailability, has
been carried out and the work will be reported elsewhere as a sep-
arate paper.
Acknowledgments
We thank Kenichi Kawasaki, Susumu Komiyama, Kazuhiro
Ohara, Yasue Nagata, Osamu Kondoh, Kiyoaki Sakata, Toshikazu
Yamazaki and all people who contributed to this work.
Figure 6. A plot of LEI values to number of heavy atoms for the series of Hsp90
inhibitors in this study. Symbols corresponding to fragment hit 1 and virtual
screening hit 4 are colored in green; CH5015765 in red.

T. Miura et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5778–5783
5783
10. Shuker, S. B.; Hajduk, P. J.; Meadows, R. P.; Fesik, S. W. Science 1996, 274, 1531.
11. Hsp90 ATPase assay was performed in a 96-well plate. 100 l of the reaction
mixture containing 30–100 g full-length Hsp90 and 10 mM ATP in 50 mM
Supplementary data
a
l
l
a
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.bmcl.2011.08.001.
tris(hydroxymethyl)aminomethane–HCl, pH 7.6, 10 mM KCl, 5 mM MgCl2, was
incubated for 1 h at 37 °C. Tested compounds were prepared in DMSO as 100Â
stock and added to the reaction mixture at a final DMSO concentration of 1%.
After the incubation, the amount of inorganic free phosphate produced by the
References and notes
ATPase activity of Hsp90a, was quantitated by the addition of 100 ll malachite
green solution and the subsequent measurement of the absorbance at 655 nM
using a microplate reader.
12. Hopkins, A. L.; Groom, C. R.; Alex, A. Drug Discovery Today 2004, 9, 430.
13. Rarey, M.; Kramer, B.; Lengauer, T.; Klebe, G. J. Mol. Biol. 1996, 261, 470.
14. Gasteiger, J.; Rudolph, C.; Sadowski, J. Tetrahedron Comp. Method 1990, 3, 537.
15. Crystals were grown by the hanging drop vapor diffusion technique. Drops
1. Duerfeldt, A. S.; Blagg, B. S. J. Bioorg. Med. Chem. Lett. 2010, 20, 4983.
2. Whitesell, L.; Mimnaugh, E. G.; Costa, B. De.; Myers, C. E.; Neckers, L. M. Proc.
Natl. Acad. Sci. U.S.A. 1994, 91, 8324.
3. Taldone, T.; Gozman, A.; Maharaj, R.; Chiosis, G. Curr. Opin. Pharmacol. 2008, 8,
370.
were prepared by mixing 1
solution with of reservoir solution, containing 18–27% (w/v)
Polyethyleneglycol 3350, 0.2–0.4 M MgCl2, 0.1 M Tris–Cl (pH 8.0), and were
incubated at 277 K against 400 of the reservoirs for 1–2 weeks. X-ray
ll of 10–20 mg/mL protein and 1 mM inhibitor
4. Porter, J. R.; Fritz, C. C.; Depew, K. M. Curr. Opin. Chem. Biol. 2010, 14, 412.
5. Murray, C. W.; Carr, M. G.; Callaghan, O.; Chessari, G.; Congreve, M.; Cowan, S.;
Coyle, J. E.; Downharm, R.; Figueroa, E.; Frederickson, M.; Graham, B.;
McMenamin, R.; O’Brien, M. A.; Patel, S.; Phillips, T. R.; Williams, G.;
Woodhead, A. J.; Woolford, A. J. J. Med. Chem. 2010, 53, 5942.
6. Brough, P. A.; Barril, X.; Borgognoni, J.; Chene, P.; Davies, N. G.; Davis, B.;
Drysdale, M. J.; Dymock, B.; Eccles, S. A.; Garcia-Echeverria, C.; Fromont, C.;
Hayes, A.; Hubbard, R. E.; Jordan, A. M.; Jensen, M. R.; Massey, A.; Merrett, A.;
Padfield, A.; Parsons, R.; Radimerski, T.; Raynaud, F. I.; Robertson, A.; Roughley,
S. D.; Schoepfer, J.; Simmonite, H.; Sharp, S. Y.; Surgenor, A.; Valenti, M.; Walls,
S.; Webb, P.; Wood, M.; Workman, P.; Wright, L. J. Med. Chem. 2009, 52, 4794.
7. Stebbins, C. E.; Russo, A. A.; Schneider, C.; Rosen, N.; Hartl, F. U.; Pavletich, N. P.
Cell 1997, 89, 239.
1 ll
ll
diffraction datasets were collected at the beamlines of BL-6B and AR-NW12A in
Photon Factory, KEK and X10SA in Swiss Light Source.
16. The atomic coordinates and structure factors for the N-terminal domain of
Hsp90a in complex with 1, 3, 4, 7, and CH5015765, have been deposited in the
Protein Data Bank under accession code, 3B24, 3B25, 3B26, 3B27, and 3B28,
respectively.
17. Wright, L.; Barril, X.; Dymock, B.; Sheridan, L.; Surgenor, A.; Besweick, M.;
Drysdale, M.; Collier, A.; Massey, A.; Davies, N.; Fink, A.; Fromont, C.; Aherne,
W.; Boxall, K.; Sharp, S.; Workman, P.; Hubbard, P. E. Chem. Biol. 2004, 11, 775.
18. Goodford, P. J. J. Med. Chem. 1985, 28, 849.
19. Boobbyer, D. N.; Goodford, P. J.; McWhinnie, P. M.; Wade, R. C. J. Med. Chem.
1989, 32, 1083.
8. Obermann, W. M. J.; Sondermann, H.; Russo, A. A.; Pavletich, N. P.; Hartl, U. H. J.
Cell. Biol. 1998, 143, 901.
9. Miura, T.; Klaus, W.; Ross, A.; Sakata, K.; Masubuchi, M.; Senn, H. Eur. J.
Biochem. 2001, 268, 4833.
20. Reynolds, C. H.; Tounge, B. A.; Bembenek, S. C. J. Med. Chem. 2008, 51, 2432.