The Resistance Part II:
Fighting HIV Resistance At Home and in the Laboratory
By Jason Socrates Bardi
"All must
work together or the Body will go to pieces."
—Aesop,
The Belly and The Members, circa 600 B.C.
It was early summer, lunchtime at the Scripps Research Institute (TSRI),
and the monthly meeting of scientists who are funded by a program project
grant from the National Instuitutes of Health called Drug Design Cycle
Targeting HIV-Protease Drug Resistance was about to begin.
There were over a dozen people in the room from all parts of the TSRI
campus and beyond—organic chemists, molecular biologists, computer
scientists, protein chemists, and cell biologists.
TSRI Associate Professor Bruce Torbett, one of the investigators in
the room, delighted at being able to have biologists at the same table
with modelers, crystallographers, and chemists. "It forces us to think
differently," he says.
At 5 minutes after the hour, TSRI Professor John Elder rushed in and
grabbed an open seat across from the other biologists. "I think I'll sit
on the computational side today," he said.
The meeting lasts over an hour. One person describes the latest "phage
display" experiments. Another discusses cloning a mutant only to find
it to be a wild type. There were reports on data mining; shuttle and expression
vectors; the chemistry and synthesis of small molecule HIV inhibitors;
selecting D-amino acid peptides as inhibitors; and an upcoming conference
in Washington, D.C.
Also during the meeting, Molecular Biology Professor Arthur Olson discusses
the status of the FightAIDS@Home project. FightAIDS@Home is a distributed
computer project that is integral to the research consortium that Olson
leads.
"Scripps is a highly collaborative place," says Olson. "That's why I've
stayed here for 20 years."
Computing for a Cause
Twenty years ago, AIDS was still a relatively unknown disease. TSRI
was at that time called the Research Institute of the Scripps Clinic,
and there was no Department of Chemistry. The institute had no computational
research program at all, Olson adds. There were not even any computers,
outside of the few in administrative offices and those that were hooked
up to one scientific instrument or another.
Over the years, computers and chemists have both arrived at TSRI, and
now Olson directs a program project grant that has brought these two groups
together with biologists to take a multidisciplinary approach to addressing
the problem of HIV protease drug resistance.
"Nobody knows how to do everything, so you really need these kinds of
collaborations," says Olson.
Olson's FightAIDS@Home project is one of a growing number of "distributed
computing" projects that seek to make use of the vast untapped computational
resource that exists in the form of personal home computers. In fact,
Olson's was the first such project involving biomedical research.
"The idea is that with a large enough computing source, you can look
at all the mutations that may arise during the evolution of drug resistance,"
says Olson.
The procedure is simple. Any person with a computer and an internet
connection can sign up by logging onto http://fightaidsathome.scripps.edu/
and downloading the program—the "client" in computer software parlance.
Once the client is installed on the individual's computer, it is designed
to conduct some set of calculations that may be one tiny part of a larger
computation.
The computations basically take a known or candidate drug and simulate
docking it into the HIV protease enzyme—or one of many mutant forms
of this enzyme.
The client, designed by the company Entropia, Inc.ª, runs so that the
computations take place without disturbing normal computer use. The program
runs when the machine is not in use, and runs until the computation is
finished—usually after several dedicated hours of computing time.
The process is further made unobtrusive by what is known as pull scheduling.
In this system, when the client finishes with one computation, the program
waits until the user connects to the internet. Then the program wakes
up, sends the results to the server at TSRI, and requests another job.
The server then sends another computation.
"It's as simple as that," says Olson.
AutoDock and the Source Code
FightAIDS@Home was originally managed by Entropia, but it is now a nonprofit
venture managed by TSRI. TSRI investigators are now in direct communication
with the users.
In May, investigators at TSRI sent out an email to some 30,000 previous
FightAIDS@Home users. Each email went to an individual who has expressed
an interest in donating some of his/her computer time for the cause, and
invited him/her to upload the new client application and continue under
TSRI's management.
At a meeting a few days later, a research associate describes how in
the first week of the sign-up they had received a flood of emails from
countries in Europe, Japan, and even Turkey. In the first 48 hours of
TSRI-managed operations, they had received 1,000 emails. And by the end
of May, there were about 2,500 people who had uploaded the new client.
"We're thrilled about this," says Olson
The real advantage of local management of the project, says Olson, is
that the investigators have access to the source code. This means that
Olson and the other investigators can build improvements directly into
their docking software.
The docking software they use is called AutoDock, and it was designed
by Olson's group in the 1980s. This software basically takes a computer
representations of a protein like the HIV protease and assays how well
it binds to a computer representation of a flexible molecule like any
protease inhibitor.
Originally the TSRI investigators had to supply Entropia with the source
code for AutoDock, which locked them into using the one version of AutoDock
they had when the project started.
They can now use different variations of the AutoDock code for different
computations, and they can also upgrade the FightAIDS@Home client as the
AutoDock software improves.
In fact, improvements to the AutoDock software have been made nearly
every year since the program first came out in 1989. Version 4 of the
software—the next major release—is currently waiting in the
wings for beta-testing.
"It has a lot of enhanced functionality," says Olson.
All Mutants Great and Small
Now that the project has moved to TSRI, Olson and his colleagues have
done a lot of thinking about how best to schedule and run calculations.
This has resulted in redundant jobs to multiple clients and statistical
analysis to make sure that the data returned is not corrupted—if
someone were to unplug a computer during the middle of a calculation,
for instance.
"We don't have to worry about losing a job or two," says Olson.
They have also done a lot of thinking about the sort of jobs they want
to send to people who are donating their computer time.
Originally, the questions were rather simple. Investigators took the
known structures of the wild type and mutant proteases and docked clinically
approved drugs to them to see if they could detect resistance.
"In fact, we could detect resistance," says Olson.
Now they are asking somewhat more complicated questions, such as whether
they can predict the resistance in a mutant protease without a known structure.
Not all the structures of the mutants that arise clinically in patients
who take HIV drugs have been solved.
Investigators can expand the types of calculations that are used, changing
the way in which the computer models the proteins and the inhibitors.
They can increase the grid size—something akin to resolution—and
make more accurate computations. They can make parts of the modeled molecules
more or less flexible as they see fit as flexibility directly affects
the complexity of a calculation.
The goal now is to fill in a large matrix of potential drugs and known
and potential HIV protease mutants in order to see which drugs are the
most robust against which mutants.
The first step is to figure out which computations to run. A vast number
of possible permutations of mutant and inhibitor combinations exist, so
it would be impossible to run them all. Olson and his colleagues focus
on mutations to the 10 amino acids in the protease monomer that have direct
contact with the substrate or inhibitor in the binding site.
This still takes quite a bit of computing power.
So Olson and his colleagues try to rationally decide which computations
are most important to run first. These, accordingly, are put higher up
in the queue to be solved by FightAIDS@Home computers.
Rik Belew, who is a professor in the Department of Cognitive Science
at the University of California, San Diego, and a member of the research
consortium, is an expert in machine learning. Belew and his graduate student
Chris Rosen developed a computational optimization scheme called co-evolution
to find the best candidate mutant and inhibitor combinations to study.
The scheme roughly looks for the best inhibitor for the best mutant
by using a simple docking of candidate molecules, assigning a numeric
score for how well the molecules fit together. From that, ideally, comes
information on which are the best inhibitors.
"His goal is to take a look at the matrix and see which results will
give us the most information about the entire system," says Olson. "We're
not going to be able to do all possible drugs against all mutants, so
we try to use the most informative mutants to test the [most promising]
drugs."
These computations can then guide the experimental biologists and chemists,
who can make the mutant proteins, synthesize the inhibitors, and test
whether they bind as expected.
All this information is then fed to the chemists who can use it to design
inhibitors to test and to the biologists who can create the mutant proteases
and design the experiments in which to test them. For instance, Elder
can take whatever mutants are deemed to be the most informative and express
them. Then, if what the computational group has predicted is not true
at all or if it is, the team will get his experimental feedback
The computational group can also interact with Elder and Torbett, who
are co-principal investigators on one of the four projects on the program
project grant, and look at the viral dynamics that take place in tissue
culture. Elder and Torbett have developed a panel of tissue culture assays
with various mutant forms of HIV—with wild type and drug-resistant
proteases.
"Before we try to predict viral dynamics in a patient, we may try to
predict it in tissue culture," says Olson.
The Path of Most Resistance
Torbett and Elder are interested in the molecular biological consequences
of mutations to the HIV protease as a result of drug therapy, and their
project follows from work that they previously carried out with TSRI Professor
Chi-Huey Wong who is the Ernest W. Hahn Professor and Chair in Chemistry.
Torbett used the TL3 protease inhibitor that was developed by Elder
and Wong to develop protease-resistant mutants by treating HIV-infected
cells with the drug and forcing mutants to arise. Torbett's lab isolated
the protease genes after each mutation arose, and developed a chronological
library of mutant forms of the HIV protease.
"We end up getting a mutation path of changes, from a little resistant
to very resistant," says Torbett. In the laboratory, Torbett has generated
a "supermutant" form of protease that has changed a half dozen animo acids
and is resistant to protease inhibitors.
Along this path of most resistance, they put the various mutant forms
into expression systems and asked how these sequential changes affect
the biochemistry of the protein.
What they have observed is that the initial changes are, not surprisingly,
directed against the particular inhibitor being used. This initial inhibitor
selects for those mutants that have arisen spontaneously and can resist
it.
However, over time, the protease continues to mutate.
One of the important observations that Torbett and other scientists
have made is that the later mutants exhibit broad resistance against a
number of drugs. This robust cross-resistance gives rise to the dangerous
multiple-drug-resistant strains of HIV that have been emerging since the
advent of antiretroviral therapy in the last decade.
"Why does the protease become cross-resistant when treated with a single
drug?" asks Torbett.
Substrates and Aptamers
Torbett and Elder are also asking how the mutations to the viral protease
affect the binding of the protease to substrate and how these changes
map to structural changes that occur when HIV protease mutates.
Mutant forms of the protease are all different in terms of how they
bind to substrate, and Torbett and his laboratory use a technique called
phage display to see the range of structures to which the HIV protease
can bind as it accumulates mutations.
Phage display is a method of generating large libraries of variant forms
of a particular protein—in this case, the HIV protease substrate.
In the technique, a protein is fused to a viral coat protein of the phage—a
filamentous virus that infects bacteria. Then the virus is allowed to
reproduce in culture, where it copiously makes new copies of itself.
Potentially, more than a billion variants of substrate can be generated
in this way, and the substrate preferences of HIV protease can be checked
against this library.
Torbett also uses RNA aptamers—a type of structural probe made
out of RNA—to probe the outside structure of the protease to see
how it has changed in response to the mutations. These aptamers are also
generated in great numbers as well.
All of this is aimed at giving the investigators some idea of possible
targets than will not mutate.
"If you can force the virus into a corner where it can replicate but
at low efficiency, it will not be able to grow very well," says Torbett.
"If the immune system's intact, then it should be able to control the
virus.
Next week: Chemical Approaches
Go back to News & Views Index
|
