Computational simulations show how slight structural change in a ubiquitous protein translates to dramatically altered flexibility of the cellular skeleton, fundamental knowledge that could help the search for new anti-cancer drugs

Actin — so named because it activates muscle cells — is the most abundant protein by mass in the body. Basically, where there’s life there’s actin. In evolutionary terms, it’s highly conserved — present to some degree in the most simple to complex species of “eukaryotic” organisms — any living thing made of cells with a nucleus.

PHOTO: Gregory Voth
Greg Voth, University of Chicago

PHOTO: Jim Pfaendtner
Jim Pfaendtner, University of Washington

Perhaps foremost among actin’s important functions is structural engineering. As bones are to the body, actin is to cells. Just beneath the cellular membrane, actin forms networks of filaments that are a primary component of the cytoskeleton, an undergirding molecular structure that maintains cellular shape, protects the cell and — with its ability to dynamically extend and contract — enables cellular motion.

“One of the most important things in cell biology is the behavior of the cytoskeleton and what proteins do to drive this behavior,” says University of Chicago chemist Greg Voth. “We’ve learned a lot over the years about actin, how it forms polymers and then changes its flexibility properties and depolymerizes. This cycle is what cells use to create motion, change shape and to do many other things, and it’s at the core of cellular behavior.”

The fundamental processes by which individual actins link with each other (polymerize) to form filaments and, in turn, break apart (depolymerize) have been found to be involved in various cancers, in particular breast cancer, and better understanding could point the way to new cancer therapies. “There’s a lot of interest in cancer that goes along with this kind of fundamental cellular biology,” says Voth. “People would like to find drugs that target actin and certain actin binding proteins in cancer cells. Our work isn’t to design drugs, but you can’t design drugs of this kind until you understand these fundamental steps.”

Over the past few years, with an extensive series of computational simulations, Voth and biomolecular engineer Jim Pfaendtner of the University of Washington have uncovered previously unknown details of these fundamental processes. Using TeraGrid resources at Texas, Tennessee and Pittsburgh, they first looked at how one small part of actin’s structure changes during metabolism — a question raised by laboratory experiments. Their results — reported in Proceedings of the National Academies of Sciences (August 2009) — clarified this question, and pointed the way toward further work.

They followed with very large-scale simulations involving multiple actins combined into filaments, looking at complicated interactions between actin filaments and other biomolecules. Their findings from this work —Journal of Molecular Biology (February 2010) and PNAS (April 2010) — show that these interactions, which involve the same small part of actin they first looked at, radically change the properties of the filaments, making them structurally more flexible — in effect softer — initiating the structural change that leads to the breaking apart of the filaments.

“We’ve been able to show,” says Voth, “how structural changes of the protein at a relatively small scale, things that you can’t see in typical structural biology experiments, translate at a larger scale to global changes in the flexibility of the actin filaments that can lead them to sever and break into pieces. We’ve been able to use these big calculations to reveal at the molecular level what’s giving rise to some of the experimentally observed properties.”

Collaboration Makes Things Happen

In all this work, Pfaendtner and Voth collaborated closely with molecular biologists — Thomas Pollard and Enrique De La Cruz at Yale — who have done pioneering experimental studies of actin. “Our computational work,” says Pfaendtner, “complements experiment, which often offers hypotheses that are based on static pictures, sometimes of low resolution. With our simulations, we can look at the movement of individual atoms and say, ‘Yes, this is happening, or no, this isn’t happening,’ and then corroborate our findings with our experimental collaborators.”

Following from discussions with Pollard, Voth and Pfaendtner attacked the question of how actin’s structure differs when it’s bound with ATP (adenosine triphosphate) from when the ATP reacts to form ADP (adenosine diphosphate), a fundamental reaction of cellular metabolism. The question involved one small loop — called the “DB loop,” a length of eight amino-acid groups — within actin. Some studies showed that the DB loop, which is disordered in ATP-actin, forms a helix in ADP-actin. Other work didn’t show the helix.

Actin & the DB Loop
Actin is a three-part protein (a trimer — blue, red, green), shown with the binding pocket, where ATP binds, and the DB loop (labeled α and β respectively). The inset shows the DB loop in both its disordered and helical form.

From other actin research, it was apparent that the reaction of actin-bound ATP to ADP leads to softening of the actin filaments, and the larger, interesting question was whether a change from disordered to helical structure in the DB loop might be a switch for this overall softening. Voth and Pfaendtner’s simulations show that a helical DB loop is the stable, low-energy form for ADP-actin and provide convincing evidence that the helix forms as part of the ATP to ADP reaction.

The main tool for these simulations was NAMD, a molecular dynamics (MD) program developed to run efficiently on massively parallel systems. MD simulations track the movements in time of each atom in a molecule — 5,700 atoms in one actin protein (50,000 atoms including the water environment), and NAMD has been a powerful tool in protein studies. “Development of this code and the associated force-field we use has been a boon to this field,” says Pfaendtner. “We do things routinely now that not long ago we couldn’t have imagined doing.”

For the DB loop simulations, Voth and Pfaendtner used the Ranger system at TACC, where PSC scientist Phil Blood, working through the TeraGrid ASTA program (Advanced Support for TeraGrid Applications) had optimized NAMD to run 20 to 30-percent faster than was previously possible, using up to 2,000 processors simultaneously. Even with NAMD and Ranger, however, it wouldn’t have been feasible to get useful results, because the folding of the DB loop into a helix occurs over a microsecond or longer of biological time, too much, even with the most powerful tools, to fully simulate with MD.

Pfaendtner addressed this problem with an innovative approach called “metadynamics” — a powerful, mathematically sophisticated algorithm that is applied, in effect, on top of MD and which makes it possible to arrive at the stable, free-energy structure of a folding event without the complete atom-by-atom MD. Through an NSF-funded international collaboration, Pfaendtner worked in the research group of acclaimed physicist Michele Parrinello, at the Swiss Federal Institute of Technology (ETH Zurich), which developed metadynamics.

Through this collaboration, Pfaendtner applied metadynamics to actin simulation, which enabled the finding that the DB loop’s transformation from unstructured to helical is part of the ATP to ADP reaction. “We have this picture now,” says Pfaendtner. “It had been suggested by experiment, but the experiments were conflicting, and we were able to clarify the details.”

It remains to be seen to what extent the DB loop presents a target for drug therapies aimed at stopping cancer.

The next question, says Pfaendtner, was “What does it mean for the mechanical properties of the cells that this loop folds?” With Kraken at NICS and BigBen at PSC, Pfaendtner and Voth built on groundbreaking 2005 work by Voth and Jhih-Wei Chu (University of California, Berkeley) and simulated 13 actin proteins linked as an actin filament, about 500,000 atoms, for a total of about 500 nanoseconds, very large simulations. With these studies, which modeled actin filaments in several different configurations, the researchers interpreted experimental data showing that a result of the ATP to ADP reaction is softening of the filament. Their data, furthermore, taken together with the prior simulations, indicate that folding of the DB loop causes the filament to reduce its “persistence length,” in effect a softening of the filament.

Actin Filaments
Actin filaments in standard form (left) form a double helix (yellow/blue strands). When ADF (green) binds with actin (right), the structure changes.

The next round of simulations, this time with BigBen at PSC, included a protein called cofilin or ADF (actin depolymerization factor) for short. “We were again looking at actin interactions,” says Pfaendtner, “but trying to understand at the molecular level ‘How do these filaments actually fall apart?’” Experiments had shown filament softening related to binding with ADF, but the how and why of it weren’t understood. Once again, with very large simulations of 13-actin filaments, Voth and Pfaendtner added new detail to the picture of the actin polymerization-depolymerization cycle.

Perspective Going Forward

Their results show that ADF-binding changes the orientation of the DB loop, moving it away from the central axis of the filaments, which in effect, loosens contacts between adjacent actins, hence an overall softening. “TeraGrid has enabled this experiment-based collaboration,” says Pfaendtner, “and through it we’ve seen the experimental community becoming enthusiastic about what we can do with simulations.”

Through several years of synergy between experiments and simulation, this work has arrived at new understanding about a small structure within actin, the DB loop, as a major player in regulating cytoskeletal dynamics and structure. It remains to be seen to what extent this loop may present a target for drug therapies aimed at stopping cancer, which are, in effect, cells whose growth process has gone out-of-control.

“Ultimately,” says Voth, “we want to develop multi-scale models and take all these molecular properties and put them into simpler models to understand these phenomena at the cellular scale, but we’ve been really surprised to find how relatively small changes at the molecular scale propagate to properties of the cytoskeleton. This is very fundamental knowledge that we’re getting from these simulations.”

© Pittsburgh Supercomputing Center, Carnegie Mellon University, University of Pittsburgh
300 S. Craig Street, Pittsburgh, PA 15213 Phone: 412.268.4960 Fax: 412.268.5832

This page last updated: May 18, 2012