Oily film

The cell membrane is made up of a double layer of oily molecules known as lipids. In living cells, the lipid bilayer is studded with cholesterol molecules and membrane-spanning proteins such as ion channels. All of these are in constant motion, jostling and bobbing like a flock of rubber ducks in a tub. At the same time, they bump into, repel, and interact with a sea of solvent molecules on both sides of the membrane.

In a watery solution, lipids will naturally assemble into sheets to avoid interacting with the water. The interactions between cell membrane lipids and surrounding water molecules also help stabilize the membrane against the stresses incurred when cells swell and shrink. For this reason, the interactions of every atom in the system can potentially affect the behavior of the entire cell membrane.

Voth and Ayton began their simulations by replicating experiments done in the laboratory on real membranes. On NCSA's Platinum Linux cluster, the researchers built a virtual version of the simplest cell membrane possible: a bilayer made solely of DMPC, a type of lipid molecule. Their atomic-scale model showed the membrane swelling in response to a decrease in exterior osmotic pressure. The flexibility of their virtual membrane turned out to be very similar to previously measured experimental values, confirming their model was quite accurate.

But just as the range of a standard camera lens is too limited to capture the panorama seen by the eye, the molecular dynamics simulation was limited by computing capabilities to just small snippets of membrane. The same constraints meant the simulation followed the membrane for only a few nanoseconds--a period too brief to observe longer-lasting effects playing out in their entirety. To catch any effects that ripple across larger portions of the cell surface, the researchers would have to expand the model's scale.

To do this, Voth and Ayton applied the techniques of statistical mechanics. A very broad and diverse field, statistical mechanics can take impossibly large problems and reduce them to a manageable size. But the cost of reducing the problem is that some information is lost. The trick is determining which information is critical to keep in raw form and which information can be averaged to simplify the model. "If you average too much, you average away everything that's interesting," Voth says. "That's the art in this technique." In the case of cell membranes, the most critical information turned out to be the membrane's flexibility.