1995 ACA Annual Meeting, Montreal, Quebec

Protein Structure-Based Drug Design Session (Chair: Christophe Verlinde)
and Related Posters

Leslie Kuhn, Department of Biochemistry, Michigan State University and
Christophe Verlinde, Biomolecular Structure Center, University of Washington

Christophe Verlinde (U. Washington) began this exciting, diverse 
session with a crisp overview of the state of structure-guided drug 
design.  He marked the major challenges facing drug design as pre-assessing 
drug toxicity, flexibility, and binding energy.  The comprehensive  
overview covered many currently used drug scaffold libraries as 
well as computer methods for matching pharmacophores and building, 
linking, and docking drug candidates.  Verlinde's results focused
on inhibitors developed to halt potentially lethal trypanosomiasis 
(African sleeping sickness) by blocking the cofactor site of 
glyceraldehyde-3-phosphate dehydrogenase (GAPDH).  Targeting the
cofactor rather than active site came from the bright observation 
that human and trypanosomal GAPDH active sites were highly conserved, 
but the cofactor site differed enough to allow a parasite-specific 
inhibitor.  A 45-fold selective inhibitor was achieved by modifying 
substituents on the cofactor's adenosine ring, and novel inhibitors  
identified by DOCK shape-template searches of the Available Chemicals 
Database (ACD). 

X. Chen (UCSF) presented another important application of DOCK:  
identifying non-peptidyl drugs for malaria, which kills a million 
people each year. The malarial protease falcipain was modelled 
based on papain's structure and a loop dictionary, and its active site 
used as a DOCK template for searching the ACD.  Exhaustive molecular 
graphics screening led to identification of an oxalic hydrazide 
with 10 micromolar inhibition.  QSAR and structure-guided redesign 
led to 150 nanomolar inhibitors of even chloroquine-resistant malarial 
strains.  It's exciting that homology modelling can provide a 
good enough template for successful inhibitor design, and the results 
show the benefits of attacking a hard problem from all sides.
For the loop modelling problem, a poster by Sucha Sudarsanam (Immunex) 
impressively simplified the task by constructing loops using a 
dipeptide database.  He demonstrated that choosing dipeptide phi,psi 
values consistent with adjacent residues' preferences and selecting
loops with ends overlapping the known structure leaves only a few 
loop conformations, which are typically close to the correct one. 
 
Next, Trevor Hart (U. Alberta) took us for a non-random walk through 
ligand docking using BOXSEARCH.  A Monte Carlo/simulated annealing  
algorithm testing a number of random initial ligand positions, 
BOXSEARCH  has a novel way of resolving steric clashes. Its "floating" 
algorithm uses Boltzmann probability to accept or reject steric 
overlaps based on overlap energy.  Genetic algorithms are being 
explored to increase searching efficiency.
 
The structural basis for antibiotic specificity and resistance were the
theme of Osnat Herzberg's (CARB) talk.  Mutants and structures 
for class A and C  beta-lactamases, which have similar structure but  
different specificity (penicillin vs. cephalosporin), imply that the 
two classes have different mechanisms for beta-lactam cleavage. Mutating 
just two residues allowed hydrolysis of previously resistant 
3rd-generation cepholosporins, a sobering result.  Natalie Strynadka's
(U. Alberta) poster presented the structure of a class A beta-lactamase in 
complex with BLIP (beta-lactamase inhibitory protein).  This mysterious 
protein binds a spectrum of beta-lactamases with very high affinity, 
producing an enormous  protein:protein interface including BLIP phenyl 
and carboxyl groups mimicking those in penicillin G.  A surprising case 
of a protein mimicking a previously-known small molecule inhibitor!  
24 interfacial water molecules mediate binding and may provide plasticity
for binding different beta-lactamases.  To win against bacteria and 
viruses, it's clear that we need to stay one step ahead of their mutations.  
Insightful strategies for beating drug resistance in HIV protease (HIVP) 
were presented in Paul Ala's poster (DuPont Merck).  While it seems
natural to use symmetric inhibitors for an active site formed by two 
identical subunits,  there is a definite advantage to asymmetry:  then
active-site mutations can only break interactions with half the inhibitor.  
Developing larger inhibitors can also increase binding and reduce the effect 
of any single protein mutation.

Pat Weber (Schering Plough) used combinatorial libraries to discover
L- and D-peptidyl ligands for streptavidin as reversibly-binding 
alternatives to biotin.  For biotechnology applications, such as tagging 
and affinity purification, controlling binding affinity can be much more 
powerful than having a very tight binder (the usual goal of drug design). 
Scott Dixon (SmithKline Beecham) then presented a computational analog of
combinatorial libraries by using a genetic algorithm to "cross over"
and evaluate new combinations of 2-dimensional molecular structures
as ligand candidates, with  penalties being incorporated to disfavor
hard-to-synthesize compounds. Using nonbonded interactions as an evaluator,
flexible ligands were docked in the protein active site making the 
assumption that active-site bound water molecules were not displaced.  
This brought to mind Michael Raymer's earlier genetic algorithms talk, 
which demonstrated that ligand-displaced and ligand-mediating water 
molecules can be predicted and appropriately  included.  Dixon is also 
analyzing the relationship between binding affinity and structural 
characteristics such as buried surface area, solvation, and mobility  
for 240 protein complexes.  
 
Charlie Bugg (Biocryst) gave an inspiring overview of 15 years of pioneering
research on purine nucleotide phosphorylase, which is required for T-cell 
immunity.  It was revealing that the crystal structures not only gave insight 
in how to improve the affinity of leads, but also indicated which parts of the 
structure should not be touched for improving important physicochemical 
properties such as solubility.  Starting from leads in the milli- to micromolar 
range, several nanomolar inhibitors were obtained.  Importantly, a number of 
these compounds are in Phase I and II clinical trials for T-cell mediated 
diseases.  This great session was closed by Richard Schevitz (Eli Lilly),
who described the iterative design of a selective non-pancreatic phospholipase
A2 inhibitor.  A take-home message from Richard's talk was that examining
the structure and using common sense are the key to success.  After all, 
it looks as if it's difficult for computer programs to beat the human brain!  
(And genetic algorithms and combinatorial libraries remind us that combining  
two brains is better than one.)