The National Institute for Computational Sciences

Tracking the Movements of HIV’s Primary Player

Scientists simulate the dynamics of the drug treatment target HIV-1 protease

By Katie Freeman

More than 33 million people worldwide are infected with HIV — roughly four times the population of New York City. By killing T cells, the white blood cells that fight infection, HIV knocks out the body’s immune defenses and progresses to AIDS. Development of drugs to fight HIV has been an important global objective since the virus was first identified in the early 1980s.

The biggest challenge to drug efficacy is the virus’s ability to mutate quickly and become resistant to treatment. Like the genetic codes passed from generation to generation in families, the genetic codes of HIV particles change through repeated replication. But unlike families, HIV particles will grandfather billions of generations a day, increasing the chances that subsequent generations have evolved considerably and will not respond to a drug in the same way as previous generations.

Chemist Carlos Simmerling and his team at Stony Brook University — Ph.D. students Fangyu Ding, Christina Bergonzo, and Miranda Shang—are using Kraken, the world’s fastest supercomputer managed by academia, to simulate the movement of the HIV-1 protease, an enzyme that aids in the replication of the most prevalent strain of the virus, HIV-1. The Kraken supercomputer housed at the National Institute of Computational Sciences (NICS) at Oak Ridge National Laboratory is the result of a National Science Foundation (NSF) Track 2 award of $65 million to the University of Tennessee and its partners.

“In terms of drug resistance, HIV is a great case to study,” Simmerling said. “More structure [determinations] have been done for this molecule than any other molecule in history. It’s the poster child.”

HIV-1 protease has been a popular target of anti-AIDS drug design because of its crucial role in virus replication.

“A protease is supposed to do the job of cutting up long proteins into short ones,” Simmerling said, adding “A protease molecule in a virus makes cuts in certain spots of the newly produced viral progeny so that viral particles can come in and cause infection. So it’s great to stop the protease so infection doesn’t occur.”

Drugs targeted against proteases send a small-molecule inhibitor to bind to the HIV-1 protease’s active site. The protease inhibitor uses two flaps to cover up the business end of the protease, preventing the protease from cutting or snipping other proteins. Scientists in the lab use a marker to tag flaps on the protease inhibitor. A bound inhibitor keeps the protease from doing its job. The result? Defective viral particles cannot infect other cells. When this happens, the spread of HIV is limited.

Quick change artist

Finding the “lock and key,” or the inhibitor that has a shape that perfectly matches the receptor, is a common practice in drug design. But as HIV-1 mutates, structural changes occur in the protease, and the inhibitor “key” no longer fits in the receptor “lock.”

“Now the problem we have is patients developing resistance to the drug,” Simmerling said. “What we’re trying to understand is what is going on when inhibitors bind and what changes happen in the behavior of the molecule when it becomes resistant to the drug.”

Simmerling is interested in the dynamics of the protease as a key to increasing the longevity of the interaction between protease and inhibitor. Observing the continuous dynamics of these molecules is nearly impossible in an experimental laboratory. Molecules must be kept at temperatures that do not always mirror the temperature of the molecules in nature. Also, the time scales of biology are different from real time. A significant movement can occur in a fraction of a second. Using Kraken, Simmerling is studying movement on the scale of femtoseconds, or quadrillionths of a second.

Average structures of drug-sensitive (LAI) and drug-resistant (MDR, V6) HIV-1 protease variants from simulations on kraken. Both of the drug-resistant forms adopt different conformations in the unbound state (before the drug binds), suggesting that it may be more difficult for the drug to force the resistant proteases to adopt the closed drug-bound form, contributing to drug resistance. The data provide a detailed model for experimental EPR data that suggest structural differences occur in solution. The simulation results may help in the design of new inhibitors that are able to bind to the mutant structures.

Scientists have been examining molecules in the lab since Kendrew and Perutz solved the first protein crystal structure. Scattering beams of X-rays by bouncing them off a crystal to create a three-dimensional image of atoms in a molecule has remained a primary method for constructing images of molecular structures, including HIV-1.

But X-ray crystallography has limitations, Simmerling said, “When molecules are flexible, they don’t crystallize very well, so we only see them in arrested states. Now we can take the [bound and unbound] arrested states and use [computer] simulations to move them through time.”

Imagine a flip book depicting a woman opening the door of her car, sitting in the driver’s seat, and closing the door. If you knew nothing about how car doors operated and looked at only the first and last pages of the book, you would not understand how the woman got inside the car. Similarly, X-ray crystallography images do not allow scientists to observe what structural changes occur as the inhibitor and protease molecules bind together; they can only see the before and after of molecule binding.

Simmerling and his team at Stony Brook are simulating molecule movement by calculating energy and forces at points on the molecule at different steps in time. In collaboration with several other labs, the team developed the AMBER software that uses quantum mechanics calculations to develop mechanical models that calculate the energy of not just one molecule but many. The simulation typically follows 50,000 to 100,000 atoms and moves along femtosecond- sized steps.

“There are so many of these calculations, and to get on the timescales of biology, we have to do these calculations a billion or more times,” Simmerling said. “That’s why we need these big computers.”The Kraken high-performance computing system, for instance, has a peak performance of 607 teraflops, or 607 trillion calculations per second.

In a lab experiment, scientists might go through many trials to match receptor and inhibitor. By testing these matches with simulations, they can cut down on lab time and resources.

To apply the team’s findings to real world drug research, Simmerling collaborates with biologists Gail Fanucci and Ben Dunn, experts in HIV research at the University of Florida.

Fanucci can confirm that a protease and inhibitor bind by observing tagged flaps at the binding site that open and close on an inhibitor. Simmerling and his team built physics models to replicate this action in supercomputer simulations, providing a model to help interpret the complex experimental data.

Since establishing that simulation results are similar to lab results, the team is looking beyond binding sites to other sites on the protease that might affect function.

“Now what we’re doing is moving tags to different places to see if we can come up with some common themes of how the virus is evading the drug, and other binding sites that might interfere with its function” Simmerling said.

Simulating intermediate steps of inhibitor-binding allows the team to look at all the cogs and wheels of the molecule and identify parts of the protease that facilitate structural changes. Ultimately, if the protease could be stopped from rearranging its structure as it performs its job, new treatments could go beyond the simple lock and key approach, reversing the course of drug resistance.