Structure-based de novo design of ligands using a three-dimensional model of the insulin receptor

Article abstract of DOI:10.1016/j.bmcl.2004.01.064

For the first time, a three-dimensional model of the insulin receptor is used in the de novo design of novel ligands that potentially mimic interactions of insulin at its receptor. Compound 4 competed with insulin as seen in autophosphorylation assays and inhibited up to 68% of IR autophosphorylation at 300 μM of 4 in 3T3IR cells induced by 1 nM insulin. This model provides a basis for the design of potent insulin receptor ligands.

Full text of DOI:10.1016/j.bmcl.2004.01.064

Bioorganic & Medicinal Chemistry Letters 14 (2004) 1407–1410
Structure-based de novo design of ligands using a
three-dimensional model of the insulin receptor§
Christopher Tan,a,b Lianhu Wei,a,c F. Peter Ottensmeyer,d Ira Goldfine,f
Betty A. Maddux,f Cecil C. Yip,e Robert A. Bateya,b,* and Lakshmi P. Kotraa,b,c,
*
aMolecular Design and Information Technology Center, University of Toronto, 19 Russell Street, Toronto, Ontario, Canada
bDavenport Research Laboratories, Department of Chemistry, University of Toronto, 80 St. George Street, Toronto, Ontario, Canada
cDepartment of Pharmaceutical Sciences, Leslie Dan Faculty of Pharmacy, University of Toronto, Toronto, Ontario, Canada
dOntario Cancer Institute, Princess Margaret Hospital, 620 University Ave., Toronto, Ontario, Canada
eBanting and Best Department of Medical Research, University of Toronto, Ontario, Canada
fUniversity of California at San Francisco, San Francisco, California, USA
Received 7 September 2003; revised 22 January 2004; accepted 22 January 2004
Abstract—For the first time, a three-dimensional model of the insulin receptor is used in the de novo design of novel ligands that
potentially mimic interactions of insulin at its receptor. Compound 4 competed with insulin as seen in autophosphorylation assays
and inhibited up to 68% of IR autophosphorylation at 300 mM of 4 in 3T3IR cells induced by 1 nM insulin. This model provides a
basis for the design of potent insulin receptor ligands.
# 2004 Elsevier Ltd. All rights reserved.
Anumber of metabolic diseases, including diabetes
which is reaching epidemic proportion in North Amer-
ica, are the results of abnormal interactions between
insulin and its receptor.1 Since insulin has to be admi-
nistered by injection in the treatment of these diseases,
the development of ingestable insulin mimetics has been
of great interest in recent years. Several attempts were
made to design molecules that can affect the insulin
receptor, either its extracellular ligand-binding sites or
at its intracellular tyrosine kinase domain.1,2 Past
attempts used either crystallographic information avail-
able on insulin, its mutational data, or the X-ray crystal
structures of the intracellular components of the insulin
receptor. Due to the absence of the structural informa-
tion on the interactions of insulin at its receptor such as
an X-ray crystal structure, a structure-based drug design
against this receptor has not been possible. Addition-
ally, such a design process against receptors is not as
trivial as it is against an enzyme due to the complexities
involved in the receptor signaling processes. Here, we
disclose the refinement of a three-dimensional model of
insulin–insulin receptor complex and its use in an insu-
lin receptor-based de novo strategy to design molecules
that interfere with the interactions of insulin at its
receptor site.
The quaternary structure of the complex of insulin–
insulin receptor was recently constructed using electron
cryomicroscopy (EM).3,4 An atomic model was built by
fitting the X-ray crystallographic and NMR structures
of various domain substructures of the insulin receptor
and that of insulin into the above EM structure.3
Through this exercise, a contiguous and complete three-
dimensional model of complex of the insulin–insulin
receptor was derived.
The insulin-binding site on the insulin receptor was
identified in the electron microscopic reconstruction
using gold-labeled insulin, and the X-ray crystal struc-
ture of insulin was docked into the binding pocket in
such a way that various interactions between the amino
acids in the receptor and the amino acids in insulin are
complementary.3,4 Herein we describe the refinement of
a model of the insulin–insulin receptor complex, de
novo design of ligands, their synthesis and their ability
to bind the insulin-binding site.
Keywords: Insulin receptor; De novo ligand deign.
§Supplementary data associated with this article can be found, in the
online version at, doi: 10.1016/j.bmcl.2004.01.064
* Corresponding authors. Tel.: +1-416-978-5058 (R.A.B.); tel.: +1-
416-978-2882; fax: +1-416-978-8511 (L.P.K.); e-mail: rbatey@chem.
utoronto.ca; p.kotra@utoronto.ca.
0960-894X/$ - see front matter # 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmcl.2004.01.064

1408
C. Tan et al. / Bioorg. Med. Chem. Lett. 14 (2004) 1407–1410
Initially, the three-dimensional model of the extra-
cellular portion of the insulin–insulin receptor complex
was refined including disulfide bonds for computer
modeling experiments. This model was subjected to
energy-minimization, and the interactions between
insulin and the insulin receptor were dissected.5 This
three-dimensional model was considered as a valid
starting point not only to understand the interactions
between insulin and its receptor, but also as a template
pharmacophore to design nonpeptidic insulin receptor
ligands. An analysis of the Connolly surface com-
plemented by individual amino acid characteristics in
the insulin-binding pocket revealed noticeable interac-
tions between the insulin receptor and its ligand, insulin,
that could be exploited for drug design. It is noteworthy
here that although insulin receptor is a homodimer, the
interactions between the insulin receptor and insulin are
not symmetric. The obvious challenge is how to develop
small molecule ligands that bind to the insulin receptor,
for which the endogenous ligand, insulin, has a large
surface of interactions.4
Initially, relevant hydrophobic and ionic pockets on the
extracellular domain of the insulin receptor were identi-
fied to characterize the binding site(s) for ligand design.
These regions were contiguously mapped to create the
binding pocket on the monomer 1 (regions 1 and 2 in
Fig. 1A) spanning towards monomer 2 (region 3). This
region of the binding pocket has three residues from
insulin, TyrB16, GluB21 and ArgB22, which were seen
as ‘direct interaction regions’ with the insulin receptor
(Fig. 1A).
Following the above criteria, core structure I (Chart 1)
was selected from our compound repository that can be
readily derivatized at various positions. This core struc-
ture can carry three substitutions on the hydantoin ring
providing an ability to attach three different ‘functional
groups’ that can confer hydrophobic, cationic or anio-
nic interactions at the binding regions, as needed. The
lead template was docked into the binding site of the insu-
lin receptor, from which compound 4 emerged as a repre-
sentative ligand with appropriate side-chain substitutions
extending into the hydrophobic (region 1, Fig. 1A), catio-
nic (region 2) and anionic (region 3) pockets at the
binding site. Specifically, compound 4 was docked into
the binding pocket of insulin receptor using the Sybyl
software program.6 The three side chains of compound
4 docked into the binding pocket of the insulin receptor
were positioned to mimic the interaction sites of insulin
residues TyrB16, GluB21 and ArgB22. This docked
complex was then subjected to energy minimization to
obtain a reasonable model of the complex of compound
4 with the insulin receptor (Fig. 1B).7
Based on this model, a series of compounds 1–9 were
designed and synthesized to understand the structure
activity relationships. This series of compounds was
designed to investigate the fidelity of the identified
binding sites, by variation of the side-chain lengths
attached to the hydantoin core, and the role of phenyl
group versus a naphthyl group, to arrive at a reasonable
lead compound. The structural diversity in the above set
of nine compounds also spans the conformational space
to overcome the lack of information on the dynamic nat-
ure of the receptor-binding site, although at the expense
of introducing entropic penalties into the molecules.
Figure 1. (A) Stereo view of the Connolly water accessible surface at
the binding pocket of insulin receptor. The lipophilic character of the
amino acid residues is mapped on to the surface (hydrophilic in blue,
and hydrophobic in dark green to brown). The amino acid residues
from insulin, TyrB16, GluB21 and ArgB22 (at the identified hydro-
phobic, cationic and anionic regions, respectively) are shown by the
block arrows 1, 2 and 3, respectively. The backbone of the insulin
receptor is shown by line representation in cyan and the three residues
from insulin are shown as capped sticks in orange. (B) Stereo view of
the energy-minimized model of a complex of compound 4 bound in
the binding pocket of the refined model of the insulin receptor.
Compound 4 is represented by a capped-sticks model color coded
according to atom types (C: white, N: blue, O: red). Region 1 of the
Connolly surface is shown by hydrophilic potential properties (brown:
hydrophobic, blue: hydrophilic) and surface on the regions 2 and
3 are colored by electrostatic potential properties (purple-blue:
electronegative, red-brown: electropositive).
Chart 1.

C. Tan et al. / Bioorg. Med. Chem. Lett. 14 (2004) 1407–1410
1409
Phthalimides 10 (n=2 or 3) were prepared by the
addition of b-alanine or g-aminobutyric acid to phthalic
anhydride (Scheme 1). Compounds 10 were first con-
verted to their respective acid chlorides and then, sub-
jected to a palladium-catalyzed coupling reaction with a
zinc homoenolate to obtain 6-phthalimido-4-oxo-hex-
anoic acid ethyl ester or 7-phthalimido-4-oxo-heptanoic
acid ethyl ester 11 (n=2 or 3, respectively).8,9 Phthalyl
and ethyl groups on compounds 11 were hydrolyzed
and the amine groups were protected as benzyl carba-
mates to yield 12.10 Compounds 12 were subjected to
Bucherer–Bergs hydantoin formation to yield 13 (R¼H,
n=2 or 3).11 Interestingly, when the free acid 12 was
protected as a benzyl ester, hydantoin formation
was unsuccessful, instead resulting mainly in amidation
of the benzyl ester. The carboxylic acid functional group
in 13 (R¼H) was protected as a methyl ester 13 (R¼Me)
and subjected to alkylation at the N-3 position using
Mitsunobu conditions12 to yield the corresponding fully
protected hydantoin derivatives. Deprotection of the Cbz
protecting groups then afforded the target hydantoin
derivatives 1–9 as their hydrochloride salts.
Figure 2. Evaluation of binding ability of compound 4 in competition
with insulin (in vivo autophosphorylation of insulin stimulated 3T3IR
cellsÆ2 h incubation with compound 4).
‘binding ability’ of these compounds to the insulin
receptor as seen by the decrease in the phosphorylation
of insulin receptor when activated with insulin, its nat-
ural ligand. Compound 4 showed an observable and
significant binding to the insulin receptor as concluded
from its ability to decrease the autophosphorylation of
insulin receptor in the presence of insulin, at concentra-
tions of 4 from 50 mM to 300 mM (Fig. 2). At a con-
centration of 150 mM of compound 4, insulin-induced
autophosphorylation levels were inhibited by 36% and
19% at 1 nM and 10 nM of insulin concentrations,
respectively. At a higher concentration of 300 mM of
compound 4, insulin-induced autophosphorylation was
inhibited by 68% and 36% at 1 nM and 10 nM con-
centrations of insulin, respectively. This indicates that
compound 4 may be interacting or competing with
insulin at the insulin binding site and inhibiting agonis-
tic activity of insulin. Other compounds 1–3 and 5–9 did
not show observable and/or significant effects up to 250
mM concentration (data not shown here). All the tested
compounds have different side chain lengths connecting
the functional moieties, and the differences in these side
chains presumably do not permit favorable structural
interactions to allow binding at the insulin receptor-
binding pocket. The conformational flexibilities intro-
duced into the methylene groups of the three substitu-
tions on the hydantoin core in compound 4 may have
resulted in modest potency of compound 4.
Whilst we recognized the high risk nature of the model
built using the EM topological mapping fit with X-ray
crystal structures of individual domains, and the uncer-
tainties associated with structure-based design of mole-
cules against receptors, the initial design process based
on the model of the complex of insulin and the insulin
receptor was expected to provide a lead-like structure. It
was also obvious to us that the discovery of an agonist
or an antagonist will need a better understanding of the
structure–activity relationship study based on this three-
dimensional model. Moreover, due to the large size of
this membrane receptor, we utilized only the extra-
cellular region of the insulin receptor containing the
insulin interacting domains.
With the above premises in mind, compounds 1–9 were
evaluated for their autophosphorylation activity, at
various concentrations of insulin, in a cell-based assay
using 3T3IR cells expressing insulin receptor.13 If the
designed compounds were agonists, then one would see
a synergistic activity with insulin, whilst if they are sim-
ply binding at the insulin-binding site, the autophos-
phorylation activity would be lowered compared to that
using insulin alone. Our primary focus was on the
In conclusion, compound 4, designed using an insulin-
insulin receptor model, exhibited optimal chain lengths
amongst a set of designed compounds based upon a
hydantoin core structure, antagonizing the binding of
insulin. While it is necessary to optimize the structures
of such binders to the insulin receptor, the above results
show that the model of the complex of insulin receptor-
insulin is a valid starting point for structure-based drug
discovery efforts to design insulin mimetics.
Acknowledgements
Scheme 1. Reagents. (a) AcOH, appropriate amino acid, Á; (b)
SOCl2, Á and then, Zn(CH2CH2CO2Et)2, Pd(PPh3)2Cl2, C6H6/Et2O;
(c) HCl/HCO2H/H2O, Á followed by ClCO2CH2C6H5, K2CO3, Et2O,
H2O, followed by (NH4)2CO3, KCN, EtOH, H2O, Á; (d) four steps,
see Supporting Information.
This research was supported by a grant from the
Canadian Institutes of Health Research-Proof of Prin-
ciple program and by a grant from the American

1410
C. Tan et al. / Bioorg. Med. Chem. Lett. 14 (2004) 1407–1410
Diabetes Association. L.P.K. is a recipient of Rx&D
HRF-CIHR research career award. R.A.B gratefully
acknowledges receipt of a Premier’s Research Excel-
lence Award. Infrastructure grant from Ontario Inno-
vation Trust for the Molecular Design and Information
Technology Center is acknowledged.
Atotal of 75,240 water molecules were added during
solvation. During the energy minimization, the cutoff
˚
distance for non-bonded interactions was set at 12.0 A.
8. (a) Wang, J.; Scott, I. Tetrahedron Lett. 1997, 38, 739. (b)
Rasolonjatovo, I.; Sarfati, S. R. Nucleosides & Nucleotides
1998, 19, 2021.
9. Nakamura, E.; Aoki, S.; Sekiya, K.; Oshino, H.; Kuwa-
jima, I. J. Am. Chem. Soc. 1987, 109, 8056.
10. Kim, S.; Lee, I. I.; Kim, Y. C. J. Org. Chem. 1985, 50,
560.
References and notes
11. (a) Bergs, H. Ger. Pat. 566,094, 1929. (b) Bucherer, H. R.;
Barsch, H. J. Prakt. Chem. 1934, 140, 151. (c) Abshire,
C. J.; Planet, G. J. Med. Chem. 1972, 15, 226.
1. (a) Balasubramanyam, M.; Mohan, V. J. Biosci. 2001, 26,
383. (b) Nuss, J. M.; Wagman, A. S. Ann. Rep. Med.
Chem. 2000, 35, 211.
12. Mitsunobu, O. Synthesis 1981, 1.
2. (a) Liu, K.; Xu, L.; Szalkowiski, D.; Li, Z.; Ding, V.;
Kwei, G.; Huskey, S.; Moller, D. E.; Heck, J. V.; Zhang,
B. B.; Jones, A. B. J. Med. Chem. 2000, 43, 3487. (b)
Schlein, M.; Ludvigsen, S.; Olsen, H. B.; Andersen, A. S.;
Danielsen, G. M.; Kaarshlom, N. C. Biochemistry 2001,
40, 13520 and references cited therein.
13. The ligand binding abilities of compounds 1–9 were car-
ried out in in vivo cell systems using the following proce-
dure: 3T3HIR Cells were grown in 96-well plates for 24 h,
then media was changed to DMEM without serum. After
18 h, media was removed and the test compounds at var-
ious concentrations were added to wells at 37 ꢀC and the
cells were incubated for 2 h. Next, insulin (0–100 nM in
DMEM, 0.1% BSA) was added and incubated for 10 min.
After incubation, cells were washed and solubilized in 200
mL solubilization buffer (50 mM Hepes, pH 7.6, 150 mM
NaCl, 1% Triton X-100, 1 mM PMSF). Lysate (100 mL)
was added to wells coated with MA-20, and allowed to
bind for 2 h. After washing, anti-phosphotyrosine anti-
body-horseradish peroxidase conjugate (PY20HRP, Santa
Cruz Biotechnology) was added for 2 h, then washed
again. Signal was detected using TMB peroxidase sub-
strate (3,30,5,50-tetramethylbenzidine and H2O2 in a citric
acid buffer, KPL, Gaithersburg, MD). Several concentra-
tions of the test compounds were considered starting from
1 mM to 300 mM and the experiments were performed in
triplicate. The autophosphorylation was measured by
optical density measurements at 451 nm. Aplot of the
optical density (ÁOD, difference between the sample and
the blank reflecting IR autophosphorylation) versus var-
ious insulin concentrations (in nM) used for stimulation
of 3T3 IR cells is constructed to analyze the binding abil-
ity of each test compound to the insulin receptor. Each
data point is an average of the triplicate values ÆSD.
3. Luo, R. Z.; Beniac, D. R.; Fernandes, A.; Yip, C. C.;
Ottensmeyer, F. P. Science 1999, 285, 1077.
4. (a) Ottensmeyer, F. P.; Beniac, D. R.; Luo, R. Z.; Yip,
C. C. Biochemistry 2000, 39, 12103. (b) Yip, C. C.;
Ottensmeyer, F. P. J. Biol. Chem. 2003, 278, 27329.
5. Energy minimizations were carried out using Amber ver-
sion 6 suite of programs (UCSF). Visualization, three-
dimensional structural analysis and design were carried
out using Sybyl version 6.7 suite of programs (Tripos,
Inc., USA). The surface construction was carried out by
MOLCAD module available in Sybyl suite of programs.
6. The docking exercise was performed either manually or
using Sybyl suite of software using the ‘DOCK’ procedure
available as part of this software. First, compound 4 was
constructed and was energy-minimized using default para-
meters in Sybyl. This energy-refined structure was then used
to dock into the binding pocket on the insulin receptor.
7. The electrostatic potential charges, bond lengths and
angles for 4 were obtained using Gaussian98 density
functional theory method (B3LYP/6-31G*). The complex
of 4 and insulin receptor was immersed in a box of TIP3P
˚
water with 8 A thickness from the surface of the protein.