Projects in Scientific Computing: New Understanding of Life and Its Processes

Simulating an HIV enzyme may help shut down HIV replication.

It will soon be 20 years since acquired immunodeficiency syndrome, AIDS, sprang into medical awareness. It remains one of the principal threats to human life and health worldwide. Over 20 million people live with AIDS, and recent data for the United States alone indicate 40,000 new infections annually.

The culprit is human immunodeficiency virus, HIV, an infinitesimally small particle of proteins and ribonucleic acid that invades immune-system cells, where it reproduces and gradually breaks down the body's ability to fend off disease. Limited success in treating AIDS has been achieved with drugs that slow down HIV's ability to reproduce. Many of these target reverse transcriptase, an HIV enzyme — carried into host cells by the virus. RT does what amounts to a copy-and-paste of HIV's single-stranded RNA to form double-stranded DNA, which assimilates into the host-cell DNA. By this maneuver, the host-cell's genetic machinery gets co-opted to reproduce HIV particles.

"Reverse transcriptase is very important in the biology of AIDS," says Pittsburgh Supercomputing Center scientist Marcela Madrid. "If you can inhibit RT from working, you can cure AIDS." RT may be HIV's Achilles heel, but the potent drug cocktails that attack it haven't been a cure-all. HIV has a protean ability to mutate and outmaneuver drugs, and for this reason especially scientists seek detailed knowledge of exactly how RT goes about its business of transcribing the HIV genome from RNA to DNA.

Madrid's special expertise is using supercomputers to simulate the structure and movement of biological molecules. In recent work, she collaborated with a team of HIV researchers. Using the CRAY T3D and T3E at Pittsburgh, she did a series of large-scale "molecular dynamics" simulations of RT, one of the largest enzymes ever simulated this way.

Along with extending the limits for this kind of study, this work represents an important step in understanding RT: The results confirm insights from crystallographic studies about how the enzyme changes shape under different circumstances and resolve a question about the low-energy form of the structure. Even more importantly, they show that computer simulations have a major role to play in helping to develop drugs that shut down HIV replication.

Thumbs Up/Thumbs Down

Since the late 1980s, several teams of structural biologists have focused on RT, using x-ray crystallographic methods to determine the 3D structure of the enzyme. Because of the implications for AIDS, this has been high-profile work. In early stages, it included sending samples of RT on the space shuttle to crystallize at zero gravity.

In 1992, a team led by Thomas Steitz at Yale achieved the first success. Another team, led by Eddy Arnold at Rutgers, followed soon after with a structure of RT bound with double-stranded DNA, in effect a snapshot of RT in action. In the next few years, the Arnold team produced a series of variant RT structures, including drug-resistant mutant versions and others of RT linked with inhibitors, compounds that block the enzyme's ability to link with RNA and DNA and carry out its role. Ultimately, this body of work led the way to RT-inhibitor drugs, the best known of which is AZT, which has helped many HIV-infected people. Arnold believes that because of this work newer, more effective drugs are on the way.

In the middle 1990s, a member of Arnold's team, Alfredo Jacobo-Molina, who was instrumental in their first success, continued his work on RT at the Monterrey Institute of Technology in Mexico. Through a collaborative program between Monterrey and Carnegie Mellon, Jacobo-Molina visited Pittsburgh, where he met Madrid and explained an interesting feature of RT's structure.

You can visualize the 3D structure of the active subdomain of RT as analogous to a hand — with palm, fingers and thumb. Arnold's group had produced two complementary RT structures — with and without DNA — that revealed a significant difference. With the hand grasping DNA, the thumb is extended and open, making space for the DNA to fit into the palm. Without DNA, the thumb is closed.

This joint-like flexibility of the thumb region has implications for HIV transcription and for the possibility of designing drugs to stop it. Only in the closed-thumb position, it's believed, can RT slide along a single strand of RNA to find the starting place for transcribing and building DNA. Some biologists believe that one class of RT-inhibitor drugs works by lodging in the palm of the active subdomain, like sand in a gear, locking the thumb in open position, so the enzyme can't function normally.

A New Movie: How the Thumb Closes

fig1 -- thumb-down structure
This ribbon representation of the RT enzyme shows the hand with fingers (blue), palm (red) and thumb (green), with and without DNA (yellow).

Download larger version (190K)

From crystallographic studies, researchers in effect had still photographs of a molecular hand in two positions. What they lacked was knowledge of how the thumb moved, a molecular movie, which is what Madrid's specialty, molecular dynamics could provide. "It's difficult to predict how hinging motions occur in proteins," says Arnold, "especially with a molecule of this size."

The researchers also realized that molecular dynamics could help resolve another question: which of these structures is the "native state" of the enzyme when not linked with another molecule — its low-energy, resting state. "There's been controversy," says Madrid, "about the native structure for unliganded RT. Some crystallographers say the correct native state is with the thumb up."

fig2 -- MD rearrangement
This image from the simulations shows the thumb subdomain in motion toward the fingers, which remain stationary. The structure is traced as it moves starting at one picosecond (red), 20 picoseconds (turquoise) and 30 picoseconds (blue).

Download larger version (229K)

Could supercomputers simulate the shift from thumb-open to thumb-closed position? That was the question for Madrid, who recognized it as a daunting challenge, due to the immensity of the enzyme — more than 1,000 amino acids. But worth a try.

Arnold and Jianping Ding provided coordinates from a recent high-resolution (2.8 angstroms) RT structure, obtained from a crystal that included DNA. For the simulation, Madrid removed the DNA coordinates, so the starting structure was thumbs-up. With a total of nearly 10,000 atoms, (requiring 10,0002 calculations for each time-step), the computation was too large to include water molecules, which would surround RT in the living, cellular environment.

Madrid simulated a full nanosecond of movement (a billionth of a second), at picosecond time-slices — a freeze-frame snapshot every trillionth of a second. The software (AMBER) imparts random velocities to the atoms to start the simulation, but the structure can get stuck in energy states known as local minima. To get reliable results, Madrid ran eight simulations, each of which required 41 hours of computing time on 16 CRAY T3E processors. For six of the eight, the structure closed its thumb, strongly suggesting this is the native state.

fig5 -- sim compared to crystal structure
The closed-thumb structure at the end of the simulation (green) compares well with the unliganded crystal structure (red).

Download larger version (276K)

"It was extremely exciting," says Jacobo-Molina, "to see that the simulations backup what we saw experimentally. These calculations describe a huge movement, an entire subdomain rearrangement, for a very large enzyme. Because of the capabilities required, there are few places in the world that can carry out a simulation of this magnitude."

Perhaps most importantly, Madrid's results show that high-performance computing has a vital role to play in AIDS research. Arnold found himself pleasantly surprised that simulations could yield informative results. "It identifies portions of the protein that are involved in this huge motion, which suggests that experiments can further probe this idea. My long-term view is that this kind of simulation will become more and more informative as we learn to do it better."

Madrid sees her results as a first step. A necessary next step is to include surrounding water molecules. In further work, she expects to investigate how RT moves with and without attached inhibitor drugs. "This will allow us to test hypotheses about how drugs function, and to help evaluate what approaches may be worth further research."

Researchers Marcela Madrid, Pittsburgh Supercomputing Center
Alfredo Jacobo-Molina, Monterrey Technological Institute
Eddy Arnold, Rutgers University
Jianping Ding, Rutgers University
Hardware CRAY T3E
Software AMBER
Related Material
on the Web
AMBER (Assisted Model Building and Energy Refinement).
HIV-1 Reverse Transcriptase Tutorial (requires Chemscape Chime plug-in).
References M. Madrid, A. Jacobo-Molina, J. Ding & E. Arnold, "Major Subdomain Rearrangement in HIV-1 Reverse Transcriptase Simulated by Molecular Dynamics," Proteins: Structure, Function, and Genetics 35: 332-37 (1999).
Writing: Michael Schneider
HTML Layout/Coding: R. Sean Fulton

© Pittsburgh Supercomputing Center (PSC), Revised: June 21, 1999