Scaffold hopping and optimization towards libraries of glycogen synthase kinase-3 inhibitors

Article abstract of DOI:10.1016/S0960-894X(02)00169-5

Using a virtual screening strategy based on a methodology derived from the CATS molecular descriptor, a novel compound class with inhibitory activity against the GSK-3 enzyme was identified through scaffold hopping. These compounds were readily synthesize

Full text of DOI:10.1016/S0960-894X(02)00169-5

Bioorganic & Medicinal Chemistry Letters 12 (2002) 1525–1528
Scaffold Hopping and Optimization towards Libraries of Glycogen
Synthase Kinase-3 Inhibitors
Lars Nærum, Leif Nørskov-Lauritsen* and Preben H. Olesen
˚
Medicinal Chemistry Research, Novo Nordisk A/S, Novo Nordisk Park, DK-2760Ma løv, Denmark
Received 22 November 2001; revised 31 January 2002; accepted 25 February 2002
Abstract—Using a virtual screening strategy based on a methodology derived from the CATS molecular descriptor, a novel com-
pound class with inhibitory activity against the GSK-3 enzyme was identified through scaffold hopping. These compounds were
readily synthesized, either by solid-phase or solution-phase chemistry. Compounds with inhibitory activity below 1 mM were iden-
tified. # 2002 Elsevier Science Ltd. All rights reserved.
Glycogen Synthase Kinase-3 (GSK-3) is a protein-serine
kinase implicated in the hormonal control of several
regulatory proteins. It was first discovered by virtue of
its ability to phosphorylate and inactivate glycogen
synthase, the regulatory enzyme of glycogen synthesis in
mammals.1,2 Since then a number of other substrates
have been identified, implicating the enzyme in the reg-
ulation of several physiological processes. A number of
different chemical compounds have recently been found to
inhibit the GSK-3 kinase, for example, indirubins,3 paul-
lones,4 maleimide derivatives5,6 and the marine sponge
hymenialdisine.7 These compounds belong to highly
different structure classes and show varying degrees of
kinase specificity.
structures with significantly different molecular back-
bones.8 In this study, we have compared 32 different
virtual libraries against all 47 GSK-3 hits derived from
HTS. Using the in-house capability of making ‘cherry-
picked’ library designs within a virtual scope based on
verified chemistry and immediately available reagents all
virtual compounds were readily available from solid-
phase parallel synthesis.
CATS2 Analysis and Compound Library Design
In traditional similarity analysis based on 2D finger-
prints, it is often assumed that compounds above a
Tanimoto similarity of 0.85 relative to a known active
reference are likely to be active.9 Using this technique,10
we found no structural similarity between any of the 47
GSK-3 hits and all 137,847 enumerated molecules from
the virtual scopes as shown in Figure 1. Thus, no virtual
molecules could be selected for synthesis by this approach.
Compounds that specifically inhibit GSK-3 activity may
be useful in the treatment of diabetes. To find new
inhibitors of the GSK-3 kinase the Novo Nordisk
library of compounds was screened for GSK-3 activity
using standard HTS technology. The results derived
from these measurements were single point determin-
ations representing the amount of GSK-3 activity left
within each sample after treatment with the test com-
pound. Forty-seven active compounds were identified as
a source for a drug discovery project. However, only
few of these hit compounds were suitable for further
optimization. To find better hits, a scaffold hopping
approach was attempted. Scaffold hopping can be
defined as the identification of isofunctional molecular
Recently, Chemically Advanced Template Search
(CATS) has been developed as a technique for virtual
screening.8 This approach calculates similarity between
molecules by comparing the topological pattern of
pharmacophore features assigned to atom pairs. An
analogous (CATS2) methodology has been imple-
mented in our laboratory combining components from
the SPL10 and Perl11 computer scripting languages.
Differences from the original implementation are mainly
related to the definition and assignment of the different
pharmacophore atom features within a molecule.12
These pharmacophore atom features are Positive,
*Corresponding author. Fax: +45-4466-3450; e-mail: lnl@
novonordisk.com
0960-894X/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved.
PII: S0960-894X(02)00169-5

1526
L. Nærum et al. / Bioorg. Med. Chem. Lett. 12 (2002) 1525–1528
Table 1. CATS2 similarity searching results within the virtual scope
derived from 32 libraries
Library
number
Compounds
in virtual
scope
Compounds
examined
by CATS2a
Hits
found using
CATS2b
V13432
V2
V3
V4
V5
V6
V7
V8
V9
V10
V11
V12
V13
V14
V15
V16
V17
V18
V19
V20
V2155,1 5593
V22
V23
V24
V25
V26
V27
V28
V29
V30
3279
7800
400
17,640
5775
28,729
9044
12,960
768
1178
1892
2160
11,880
9450
84,000
55,200
3283
861,224
10,952
11,070
6
874
400
0
0
0
0
0
3989
2094
3549
40210
6008
333
1003
1881
2047
10,000
8622
5118
8573
1376
10,000
7142
10,000
0
0
0
0
0
0
821
0
0
0
Figure 1. Lack of similarity between the 47 HTS hits and all poten-
tially synthetically accessible compounds available from 32 virtual
libraries.
Negative, Acceptor, Donor and Hydrophobic. Addi-
tionally, we have chosen to reduce the overall lipophilic
contribution to the molecular mapping by excluding
0
0
0
25
0
lipophilic–lipophilic atom pairs.
1963
26,312
18,000
2891708
5148
7128
113,680
1681
2352
1068
4416
3288
0
0
0
Computationally, all virtual libraries were compared
simultaneously against each of the 47 HTS hits by using
the CATS2 approach (Table 1). Based upon the hit rate
found after virtual screening, the library number V14
was selected. Then, the 821hits found by CATS2 were
further reduced to 324 by conventional fragment-based
0
2118
6984
10,000
1680
23410
69
0
109
0
2D fingerprint diversity analysis in order to represent as
many different molecular fragments as possible within
the final library design.
,248,884V311
V32
9269
0
42,840
73
0
aFiltering criteria were Mr<500 and a maximum of 10,000 randomly
selected compounds.
bA Euclidian distance <0.25 in descriptor space was used as cut-off.
A combinatorial library was synthesized according to
Scheme 1and tested in the GSK-3 assay as a single-
point determination at 30 mM concentration. IC50
values were only determined for inhibitors showing less
than 60% activity left. Compounds with activity above
100% may be activators of GSK-3, but this issue
remains unexplored. Several compounds from this
‘cherry-picked’ library, derived from 14 R1, 8 X and 41
R2 fragments, displayed significant inhibition (Fig. 2)
and the most active compound (1) is shown in Table 2.
whole molecule and similarity is really measured in
terms of dissimilarity. The two molecules shown are
similar because they have less global dissimilarity and
hence one cannot specify any structural part as being
particularly similar to another. Nor can one infer a
common binding mode.
A crude SAR obtained by evaluation of this library
indicates that the R2 fragment (Scheme 1) as a 2-sub-
stituted 5-(1-pyridyl)-[1,3,4]-oxadiazol moiety is impor-
tant for activity. Additionally, a one-carbon linker
between the sulfur atom and a phenyl ring seems to be
important. In order to further optimize the compounds
derived from the library, a computational HQSAR10
It is tempting to analyze the resulting molecule in terms
of recognizable fragments from the parent HTS hit
(Table 2) but, even though one can visually identify
certain common molecular features, it must be stressed
that CATS generates composite descriptors for the
Scheme 1. Synthetic route towards the libraries.14

L. Nærum et al. / Bioorg. Med. Chem. Lett. 12 (2002) 1525–1528
1527
model was established using standard set-up and vali-
dation. This model can predict the activities of virtual
compounds within the scope of library V14.
A general procedure for the synthesis was used.15 More
than 100 compounds were prepared by this method and
tested in the GSK-3 kinase assay. The most important
SAR information extracted from these data is given in
Tables 3 and 4. Early on, it became clear that the R1-
NHCO fragment within the solid-phase synthesized
library was of limited importance for the activity. To
reduce the number of flexible bonds and the molecular
weight, for the compounds prepared in solution phase,
the R1-NHCO fragment was replaced by smaller sub-
stituents directly attached to an aromatic ring.
With this information in hand, a new compound library
including the compounds predicted as the most active
was synthesized according to Scheme 1and tested in the
GSK-3 kinase assay. This ‘cherry-picked’ compound
library was biased towards compounds containing the
oxadiazol-pyridyl moiety and comprised of 96 com-
pounds derived from 21 R1, 1 X and 15 R2 fragments.
This library included an increased number of active com-
pounds, but unfortunately no compound was found to be
more potent or efficacious compared to the first compound
library (Fig. 2). However, this library of compounds con-
firmed the importance of the oxadiazol-pyridyl moiety to
obtain active compounds.
As seen from the data in Table 3, a 4-pyridyl moiety is
crucial for activity. A sulfur or oxygen in the five-mem-
bered heterocycle gives compounds with similar potency,
whereas an H donor at the same position is detrimental
for the activity. The size and bulkiness of the substituent
on the aromatic ring seems to be of minor importance.
Additionally, the optimum substitution pattern for the
aromatic ring was examined. As shown in Table 4, com-
pounds with substituents in the 3-position consistently
attained the best inhibitory activity.
Further Synthesis and SAR
The new compound series was well suited for single
array synthesis of compound libraries in solution phase.
Table 3.
Compd
Positional isomer
Y, m
GSK-3
IC50 (mM)13
2
3
4
5
6
7
4-Pyridyl
4-Pyridyl
4-Pyridyl
3-Pyridyl
2-Pyridyl
4-Pyridyl
1
S,11
O,18.0
NH,1
S,1
S,1
O,0
>250
>250
>250
29
Figure 2. Hit rate and activities measured from the library containing
324 compounds (white) and the second optimized library containing 96
compounds (black). Low activity values represent active compounds.
Table 4.
Table 2. The most potent library compound (1) and the correspond-
ing HTS hit
Compd
n
Z
GSK-3
IC50 (mM)13
8
9
1H
2
6.7
H
29
10
11
12
13
14
15
16
17
13-Cl
12-Cl
13-I
14-I
13-F
13-COOCH
13-COOH
1
0.82
1
0.39
2
98
1
.4
1
1.1
3
Compd
GSK-3 IC50 (mM)13
235
Wa
>250
1
1.2
aW, 3-(5-methyl-1,2,4-oxadiazolyl).

1528
L. Nærum et al. / Bioorg. Med. Chem. Lett. 12 (2002) 1525–1528
10. Software is developed and available from Tripos Inc., St.
Louis, MO, USA.
Conclusion
11. Available from www.perl.org
Using a virtual screening strategy based on the CATS2
molecular descriptor, a novel compound class with
inhibitory activity on the GSK-3 enzyme was identified by
scaffold hopping where the conventional 2D fingerprint-
based similarity method fails. These compounds were
readily available either by solid-phase or solution-phase
chemistry. Compounds with activities below 1 mM were
identified. Some of these compounds are currently being
evaluated in secondary screening assays.
12. Atom features as defined by the Sybyl Line Notation (SLN)
language are: Acceptor N[not¼NH]|O[not¼OH], Donor Het
[is¼HetH], Negative O[is¼O(H)Hev¼Het], Positive N[(is¼N
(Any)(Any)Any&not¼N-Hev¼O)], and LipophilicC[not=Cꢀ
N,CꢀO,CꢀS=O,CꢀP¼O]|S[not¼SH, SꢀO].
13. Inhibition of GSK-3 by a test compound was evaluated
using human GSK-3b and a glycogen synthase derived sub-
strate with the following amino acid sequence: YRR-
AAVPPSPSLSRHSSPHQS(PO4)EDEEE-NH2.
In
brief,
GSK-3b was incubated with 32 mM substrate and varying
concentrations of test compound in a buffer containing 0.1
mM 33P-labeled ATP, 10 mM magnesium acetate, 8 mM
MOPS pH 7.0, 0.2 mM EDTA, 0.1% dithiothreitol and
0.03% Triton-X100 for 60 min at room temperature. The
reaction was performed using 96-well filter plates. The reac-
tion was terminated by filtration followed by addition of 25 mL
2% phosphoric acid to each well. All wells were then washed
three times in 0.5% phosphoric acid to remove unreacted 33P-
labeled ATP, dried and radioactivity was counted in a Packard
topcounter. Dose–response profiles were generated, and the
IC50 value for inhibition of GSK-3 by the test compound was
calculated.
Acknowledgements
The authors wish to thank Leif Christensen for the library
productions, Florencio Zaragoza Dorwald for the solid-
phase chemistry development and Anders Robert Sør-
ensen for providing the biological test results.
References and Notes
14. Step A: To (4-formyl-3-methoxyphenoxy)ethyl-poly-
styrene (60 mg or less, 0.55 mmol/g) are added in the order
given a solution of the primary amine (0.6 mmol) in a mixture
of NMPꢁ(10.6 mL) and water (50 mL), NaCNBH3 (0.6 mL of a
1. Embi, N.; Rylatt, D. B.; Cohen, P. Eur. J. Biochem. 1980,
107, 519.
2. Rylatt, D. B.; Aitken, A.; Bilham, T.; Condon, G. D.;
Embi, N.; Cohen, P. Eur. J. Biochem. 1980, 107, 529.
3. Leclerc, S.; Garnier, M.; Hoessel, R.; Marko, D.; Bibb,
J. A.; Snyder, G. L.; Greengard, P.; Biernat, J.; Wu, Y. Z.;
Mandelkow, E. M.; Eisenbrand, G.; Meijer, L. J. Biol. Chem.
2001, 276, 251.
4. Leost, M.; Schultz, C.; Link, A.; Wu, Y. Z.; Biernat, J.;
Mandelkow, E. M.; Bibb, J. A.; Snyder, G. L.; Greengard, P.;
Zaharevitz, D. W.; Gussio, R.; Senderowicz, A. M.; Sausville,
E. A.; Kunick, C.; Meijer, L. Eur. J. Biochem. 2000, 267, 5983.
5. Smith, D. G.; Buffet, M.; Fenwick, A. E.; Haigh, D.; Ife, R. J.;
Saunders, M.; Slingsby, B. P.; Stacey, R.; Ward, R. W. Bioorg.
Med. Chem. Lett. 2001, 11, 635.
6. Coghlan, M. P.; Culbert, A. A.; Cross, D. A. E.; Corcoran,
S. L.; Yates, J. W.; Pearce, N. J.; Rausch, O. L.; Murphy, G. J.;
Carter, P. S.; Cox, L. R.; Mills, D.; Brown, M. J.; Haigh, D.;
Ward, R. W.; Smith, D. G.; Murray, K. J.; Reith, A. D.; Holder,
J. C. Chem. Biol. (London) 2000, 7, 793.
7. Meijer, L.; Thunnissen, A. M. W. H.; White, A. W.; Gar-
nier, M.; Nikolic, M.; Tsai, L. H.; Walter, J.; Cleverley, K. E.;
Salinas, P. C.; Wu, Y. Z.; Biernat, J.; Mandelkow, E. M.;
Kim, S. H.; Pettit, G. R. Chem. Biol. 2000, 7, 51 .
8. Schneider, G.; Neidhart, W.; Giller, T.; Schmid, G. Angew.
Chem. Int. Ed. 1999, 38, 2894.
1mol L
solution in THF), and acetic acid (0.12 mL). The
mixture is shaken at room temperature for 5–10 h. Wash with
methanol (1ꢂ1.5 mL) and NMP (2ꢂ1.5 mL). Step B: To the
resin-bound amine (40 mg, 0.04 mmol) a solution of a halo
acid (0.8 mmol) in DCP (0.7 mL)+NMP (0.7 mL) is added,
followed by the addition of a mixture of DIC (0.062 mL, 0.4
mmol) and toluene (0.1mL). The resulting mixture is shaken
at room temperature for 5 h. Step C: After washing (2ꢂ1.5 mL
NMP) a solution of the thiol (0.8 mmol) in NMP (1.4 mL) is
added, followed by the addition of DIPEA (0.15 mL). The
mixture is shaken for at least 12 h. Step D: Cleavage in neat
TFA for 2–20 h.
15. To a solution of the thiol compound16 (1equiv) in dime-
thylformamide the appropriate benzylhalide (1equiv) and
powdered potassium carbonate (3 equiv) were added. The
reaction was monitored by the disappearance of the yellow
colour or by TLC. The reaction time was dependent upon the
reactivity of the benzyl halide used and was generally in the
range from 1min to 1h. Aqueous workup and neutralization
of the reaction mixture with 1N hydrochloric acid and iso-
lation of the compounds by filtration gave products in excel-
lent yield and purity. NMR and LC–MS data for the
compounds were in accordance with the structures given.
16. Kubota, S.; Koida, Y.; Kosaka, T.; Kirino, O. Chem.
Pharm. Bull. 1970, 18, 1696.
9. Matter, H. J. Med. Chem. 1997, 40, 1219.