New Understanding of Life and Its Processes: Structure of Proteins and DNA

spacer spacer

Photo: Jeffry Madura Jeffry Madura, Duquesne University.


New simulations help revise thinking about antifreeze proteins.

From simple questions, marvelous and useful things sometimes transpire. In the 1950s, Canadian scientist P. F. Scholander wanted to know how Arctic fish can swim in water colder than the freezing point of fish blood. Eventually, his experiments showed that the blood of some northern fish contains "antifreeze." In the late 1960s, animal biologist Arthur DeVries investigated several Antarctic fish and isolated a protein that accounted for the antifreeze effect. By the 1990s, biotech companies began staking out patent rights.

So far biologists have found antifreeze proteins in several species of Arctic and Antarctic fish, plants and the yellow mealworm beetle. While there are several variant types of AFPs, different in structure, all show similar ability to interact with incipient ice crystals and inhibit crystal growth. Depending on the type of AFP, this interaction lowers water's freezing point two to five degrees Celsius. AFPs are hundreds of times more effective at lowering the freezing point than any other known water solute, such as salt.

Commercial possibilities are bounded only by the imagination. Scientists have already used AFPs to produce laboratory strains of transgenic tobacco, tomatoes and potatoes with improved resistance to frost damage. Genetically engineered salmon bred to resist "superchill" could be a boon to commercial fish hatcheries in the North Atlantic. AFPs may eventually safely prevent freezer burn in ice cream, stop leafy vegetables like celery and lettuce from going limp in the fridge, reduce highway damage from de-icing with salt, and protect frozen blood from the effects of thawing.

"Because there's water in blood and you're moving through the temperature from freezing to melting," says Duquesne University biochemist Jeffry Madura, "ice-crystals form and inevitably damage some red-blood cells. Antifreeze proteins could minimize the size and shape of ice crystals and make it possible to recover a higher proportion of preserved red blood cells."

What's not known is how AFPs work. "What we think," says Madura, who has done AFP-related research for a decade, "is it kinetically inhibits ice growth. It modifies the shape of the ice crystal, and that modification inhibits further crystal growth." A prominent theory holds that this antifreeze effect arises from hydrogen bonds between the proteins and the ice. Recent experimental work by Madura's research colleague, A.D.J. Haymet at the University of Houston, however, refutes this hydrogen-bond formulation.

With Haymet's experiments as background, Madura last year turned to the CRAY T3E at the Pittsburgh Supercomputing Center, which he used to carry out the most realistic simulations to date of AFPs interacting with ice and water. His results support Haymet's findings and suggest new understanding of what happens as these proteins work their wizard-like effects.

Simulation of the ice-water interface region.

This snapshot represents a simulation of the ice-water interface region. Water molecules are shown as spheres centered on the oxygen atom with bulk liquid regions (red) to the left and right and bulk ice (blue) in the center. The left-hand interfacial region is highlighted with color gradation to illustrate layered variation in average diffusion coefficient through this region.

Download a larger version of this image.

Testing the Hydrogen-Bond Theory

For about ten years, scientists interested in AFPs have understood that these proteins accumulate where ice crystals form and that their presence inhibits crystal growth. The question of interest has been what happens in this interaction that accounts for the inhibiting effect? Aside from pure science, the payoff is the ability to make synthetic AFPs that, more cheaply than proteins from living organisms, could be used in a range of applications.

The hydrogen-bond theory arose from experiments looking at how the AFP from winter flounder aligns with the ice-crystal surface. This surface is defined by an ordered "lattice" of water molecules in fixed relation to each other. At the crystal face, the flounder AFP aligned itself so that the spacing of a repeated amino-acid, threonine, matched the lattice spacing of the crystal.

"There were threonine residues 16.7 Angstroms apart," says Madura, "and if we looked at the bipyramidal ice face, we found lattice positions every 16.7 Angstroms. Threonine forms hydrogen bonds with water, and people said, ah-ha, there's a match. It's a lock-and-key relationship."

As a way of testing this theory, Haymet and his colleagues replaced the threonines in the winter flounder AFP with valine, a "hydrophobic" amino-acid — that is, it doesn't form hydrogen bonds with water. In effect, he turned off the hydrogen bonding and tested to see if this mutant protein would also inhibit ice growth. His experiments showed no change. The modified AFP also had the antifreeze effect.

To further investigate this result, which negates the hydrogen-bond theory, Madura undertook a "molecular dynamics" simulation, a computational approach which, in effect, puts the molecules together and sets them in motion to go about their biochemical business while recording what happens. For Madura, the challenge was to set up his simulation to mimic the protein-ice interaction as realistically as possible.


Snapshot of the winter flounder antifreeze protein.

Snapshot of the winter flounder antifreeze protein simulated at the ice-water interface. The protein is a helical backbone (blue). Oxygen atoms (red) indicate ice and water molecules.

Download a larger version of this image.

Ice-Water Realism

Most prior AFP simulations have represented AFPs in interaction with ice alone, no liquid water. For a number of reasons, to include water significantly complicates the computational chore. Not only does it mean accounting for a system with more atoms, but also simulation of two different phases in equilibrium with each other. With computational practicality as a major consideration, excluding the water has been an acceptable approximation. The protein-ice interaction appeared to be the crux of the mater, and since the results in general have supported the hydrogen-bond theory, the approximation appeared to work.

Studies by Haymet, however, as well as computations by Madura show that ice crystals form in a complex region of transition between solid and liquid water. "The ice-water interface is very complex," says Madura. "You don't go from liquid water to solid water. There's an interfacial region that slowly changes from bulk liquid water to entirely solid. I can't just put the protein at the ice surface, because how do I define where the ice surface is?"

With access to the massively parallel capability of the CRAY T3E, Madura constructed a simulation that included ice and water in a transitional region, about 15 Angstroms across — with 2,841 molecules of water in the ice phase and 5,484 molecules of liquid water — with a smoothly varying transition in density between liquid and solid. With the winter flounder AFP, the simulation included 25,000 atoms, larger than any prior ice-protein-water simulation.

In contrast to prior simulations, Madura also accounted for the long-range electrostatic interaction among the atoms with an accurate formulation called Ewald summation. Also in contrast to prior simulations, he imposed no constraints on the ice molecules. "In other simulations," explains Madura, "the ice lattice is constrained so that it can't fall apart." For the first time, Madura represented the full ice-water interface with no imposed limits on the interactions.

His first step was to allow the ice-water interface to find its natural equilibrium between solid and liquid. This was a major computation by itself, requiring weeks using 64 CRAY T3E processors. After this preliminary step, Madura included the protein and collected data from 510 picoseconds (trillionths of a second) of simulated molecular dynamics, more than twice as long as prior studies.

The results support Haymet's experimental findings, and they show that accurate simulations of this interaction depend upon realistic representation of the ice-water interface. Although the winter flounder AFP forms hydrogen bonds, there's no significant gain in hydrogen bonds as the protein moves from water to the ice-water region, indicating that some other interaction accounts for the protein's ability to inhibit ice-crystal growth. Madura believes that the answer will come from more simulations that look in detail at what happens in this transitional region. "It's vital that any alternative hypothesis account for the fact that the protein interacts with water molecules in between liquid and crystal form."

Among the next steps, Madura will simulate valine-modified winter flounder AFP to parallel Haymet's experiments. If the computational result tracks with Haymet's laboratory work, the hydrogen-bond theory would seem to be conclusively disproved. To go on to more detailed understanding of what happens, says Madura, longer simulation times are needed, and he expects PSC's Terascale Computing System to make that possible. With more computing power, Madura also envisions an even more realistic and challenging simulation that would place an AFP in ice and water along with a cellular membrane.

"These studies," he says, "are really the beginning of looking at how proteins interact with interfaces. The next step is to take a phospho-lipid interface, a cellular membrane, and simulate what happens when ice grows in a cell. How does the protein prevent ice-crystals from destroying that lipid interface? We're slowly working toward understanding what happens at the cellular level when freezing takes place."

Download PDF Fishy Proteins
from Projects in Scientific Computing 2001
Researchers Jeffry Madura, Duquesne University
Hardware CRAY T3E
Software User-developed code.
Related Material
on the Web
Madura Research, Duquesne Chemistry and Biochemistry

Pranav Dalal, Jared Knickelbein, A.D.J. Haymet, Frank D. Sšnnichsen & Jeffry D. Madura, "Hydrogen bond analysis of Type 1 antifreeze protein in water and the ice/water interface," Physical Chemistry Communications 7, 1-5 (2001).

Jeffry D. Madura, Kelli Baran & Andrzej Wierzbicki, "Molecular recognition and binding of thermal hysteresis proteins to ice," Journal of Molecular Recognition 13, 101-113 (2000).

Writing: Michael Schneider
HTML Layout/Coding: R. Sean Fulton
Graphic: PSC Logo
© Pittsburgh Supercomputing Center (PSC)
Revised: November 1, 2001