A Common Mechanism Underlying Promiscuous Inhibitors from Virtual and High-Throughput Screening

High-throughput and virtual screening are widely used to discover novel leads for drug design.
On examination, many screening hits appear non-drug-like:  they act noncompetitively, show
little relationship between structure and activity, and have poor selectivity. Attempts to develop
these peculiar molecules into viable leads are often futile, and much time can be wasted on
the characterization of these “phony” hits. Despite their common occurrence, the mechanism
of action of these promiscuous molecules remains unknown. To investigate this problem, 45
diverse screening hits were studied. Fifteen of these were previously reported as inhibitors of
various receptors, including β-lactamase, malarial protease, dihydrofolate reductase, HIV Tar
RNA, thymidylate synthase, kinesin, insulin receptor, tyrosine kinases, farnesyltransferase,
gyrase, prions, triosephosphate isomerase, nitric oxide synthase, phosphoinositide 3-kinase,
and integrase; 30 were from an in-house screening library of a major pharmaceutical company.
In addition to their original targets, 35 of these 45 compounds were shown to inhibit several
unrelated model enzymes. These 35 screening hits included compounds, such as fullerenes,
dyes, and quercetin, that have repeatedly shown activity against diverse targets. When tested
against the model enzymes, the compounds showed time-dependent but reversible inhibition
that was dramatically attenuated by albumin, guanidinium, or urea. Surprisingly, increasing
the concentration of the model enzymes 10-fold largely eliminated inhibition, despite a 1000-fold excess of inhibitor; a well-behaved competitive inhibitor did not show this behavior. One
model to explain these observations was that the active form of the promiscuous inhibitors
was an aggregate of many individual molecules. To test this hypothesis, light scattering and
electron microscopy experiments were performed. The nonspecific inhibitors were observed to
form particles of 30−400 nm diameter by both techniques. In control experiments, a well-behaved competitive inhibitor and an inactive dye-like molecule were not observed to form
aggregates. Consistent with the hypothesis that the aggregates are the inhibitory species, the
particle size and IC50 values of the promiscuous inhibitors varied monotonically with ionic
strength; a competitive inhibitor was unaffected by changes in ionic strength. Unexpectedly,
aggregate formation appears to explain the activity of many nonspecific inhibitors and may
account for the activity of many promiscuous screening hits. Molecules acting via this
mechanism may be widespread in drug discovery screening databases. Recognition of these
compounds may improve screening results in many areas of pharmaceutical interest.