Published on Web 03/11/2003
Discovery of a Potent Small Molecule IL-2 Inhibitor through Fragment
Assembly
Andrew C. Braisted,* Johan D. Oslob,* Warren L. Delano, Jennifer Hyde, Robert S. McDowell,
Nathan Waal, Chul Yu, Michelle R. Arkin, and Brian C. Raimundo
Sunesis Pharmaceuticals, 341 Oyster Point BouleVard, South San Francisco, California 94080
Received January 20, 2003; E-mail: braisted@sunesis.com; joslob@sunesis.com
Chart 1. IL-2/IL-2RR Antagonists: 1 (Ro26-4550), 2, and 3a
Protein-protein binding interfaces are considered to be improb-
able sites for high-affinity small molecule ligands; yet this target
class represents the majority of therapeutically important targets.1
These binding surfaces are typically large and featureless, and they
lack the well-defined pockets or mechanism-based contacts that
confer binding energy to enzyme inhibitors. While leads against
an enzyme target can be improved by incorporating functionality
known to be important on the basis of substrate preferences,
protein-protein targets do not offer such opportunities. Conse-
quently, it is difficult to envision how a small ligand can bind this
target class with high affinity.
Fragment assembly methods hold great promise because they
access very large numbers of potential fragment combinations
without requiring the synthesis of each individual compound.
a Fragment 4 is one of the hits selected by the IL-2 L72C mutant.
Ideally, a set of small fragments (<200 MW) is screened, and only
those fragments with affinity for the target are combined, thereby
increasing the efficiency of the search process. Two significant
challenges encountered in fragment assembly are the identification
of low-affinity fragments and the determination of how to link
different fragments productively. A number of fragment assembly
methods have been described including SAR by NMR, dynamic
libraries, and virtual screening.2 We have developed an approach
called tethering3 that can rapidly identify low-affinity fragments
that bind to specific sites on a target protein. Herein we describe
the first application of this strategy to generate a high-affinity
inhibitor against a protein-protein target.
Binding of the cytokine interleukin-2 to its receptor induces T-cell
proliferation and is an important therapeutic target for immune
disorders.4 Our interest in this target was stimulated by a report of
a small molecule (Ro26-4550, 1, Chart 1) that binds to IL-2 and is
a low micromolar antagonist of IL-2RR binding (IC50 ) 3 µM).5
We initiated a medicinal chemistry program using structure-based
design and parallel synthesis that generated a novel lead series (2,
3). However, this series also reached a low micromolar affinity
plateau, and efforts to identify new binding interactions were
unsuccessful.6
Analysis of the X-ray structure of 1 binding at the “hot spot” of
IL-2 revealed that the protein is adaptive and able to undergo
significant rearrangements which create the small molecule binding
site.7 This observation refutes the perception that protein-protein
interactions are flat and featureless and suggests that the surface
of IL-2 could present additional nonobvious binding sites capable
of binding a small molecule with high affinity. The adaptive nature
of the site creates an additional challenge though, because accurate
structure-based predictions are more difficult.
We applied tethering with the goal of improving the affinity of
the lead compound 2. Ten individual cysteine mutations were
designed to search the perimeter of the IL-2 “hot-spot” based on
alanine scanning data and a crystal structure of 1 bound to IL-2.
The mutants were then screened against a library of 7000 disulfide-
containing fragments. Analysis of the screening results showed that,
while most regions selected few if any fragments, one region,
accessible by two different cysteine mutants (Y31C and L72C, cf.
Figure 1), selected a number of structurally related fragments.
Intriguingly, this is the same region that was found to be structurally
adaptive, undergoing rearrangements upon binding of different
ligands.7
Statistical analysis of the fragments identified in this region
indicated that the mutants Y31C and L72C preferentially selected
small aromatic carboxylic acids (see Supporting Information).
Molecular modeling suggested that the fragments could occupy
a deep hydrophobic cavity within the adaptive region. The fact that
tethering requires covalent attachment of the fragments greatly
facilitates computational prediction of the fragments’ binding sites,
even within a region of the protein that can adopt multiple low-
energy conformations. An overlay of the modeled tethering hits
with a crystal structure of analogue 3 bound to IL-26 (Figure 1)
suggested that productive merging of fragments onto 2 might be
achieved by linking these fragments to the dichloro-phenyl ring of
2. On the basis of these observations, we designed a focused set of
20 compounds (Chart 2), incorporating the chemical functionalities
(specifically aromatic carboxylic acids) that had been selected by
tethering. Eight of the 20 compounds inhibited IL-2/IL-2RR binding
at submicromolar concentrations, demonstrating a 5-50-fold
improvement in potency over the starting scaffold. All of the tight-
binding inhibitors contained a carboxylic acid. Thus, the acidic
functionality identified by tethering was clearly required for the
improved binding. For example, the benzoic acid derivative 7
displays an IC50 of 0.20 µM, representing a 15-fold improvement
over 2. In contrast, control compounds 6 and 8 are at least 10-fold
less potent than 2 and 200-fold less potent than 7.
Analogue 13 was the most active compound with an IC50 of 60
nM. Characterization by surface plasmon resonance showed that
this compound binds to IL-2 with a 1:1 stoichiometry and a Kd )
100 nM. Compound 13 was also found to have an EC50 ) 3 µM
9
3714
J. AM. CHEM. SOC. 2003, 125, 3714-3715
10.1021/ja034247i CCC: $25.00 © 2003 American Chemical Society