Interactions between proteins underlie nearly everything that happens inside a cell -- from reading DNA to communicating with the outside world.
Many of those proteins have very similar structures, yet somehow they locate and interact with only their specific partner. For years, scientists have been trying to model and design such interactions, with limited success.
Now, MIT researchers have developed a model, reported in this week's issue of Nature, that can be used to design new protein interactions and could help scientists create proteins for use in developing new drugs.
"The proteins we design now are not likely to become drugs or therapeutics, but can be used in order to figure out the basic mechanisms of these interactions, which could be extremely valuable," said Amy Keating, associate professor of biology and senior author of the paper being published in the April 16 issue of Nature.
Scientists who work in computational protein design, a relatively new field, try to design proteins that perform specific functions. Most of those functions involve binding to partner proteins, so one of the major challenges facing researchers is designing proteins that bind strongly to their intended target but not to other proteins with very similar structures.
In other words, proteins must have both high affinity for their intended targets and high specificity. Current computational design models are reasonably good at maximizing affinity, but are not equipped to introduce specificity, said Keating.
Keating and her team took a new approach to the problem. Their model first comes up with a protein predicted to have high affinity for the target, then runs through several additional iterations to improve predicted specificity.
"What our model does is systematically explore the tradeoffs between binding tightly to the target, and interaction specificity," said Keating.
The resulting protein may have lower affinity than the original protein, but it will have a better combination of affinity and specificity.
The new protein model is based on a method called cluster expansion, used in materials science to predict how metal atoms form a lattice structure.
Keating's team adapted the model to predict the binding energy that a given protein will have when folded to a pre-defined structure, based on its amino acid sequence. This offers a rapid shortcut for the usual, very time-consuming method of constructing an atomic model of the protein and then evaluating the energy of its interactions.
"There was a convergence of methods from a variety of different sources that we didn't foresee that allowed us to solve this longstanding problem," said Keating.
The MIT team focused on a family of transcription factor proteins known as bZIPs. The proteins activate DNA transcription when they are bound to a partner protein. Using the model, the researchers designed bZIP proteins that successfully bind to their partner but do not bind strongly to similar proteins.
The current work took advantage of simplifying properties of the bZIPs. Over the next five to 10 years, models to describe more complex protein interactions will become available. The new computational technique can be applied in conjunction with such models to generate specific designed proteins for diagnostic or therapeutic purposes.
Keating worked with Gevorg Grigoryan, a recent MIT PhD recipient and lead author of the paper, and Aaron Reinke, a current PhD student.
This work was funded by the National Institutes of Health.
A version of this article appeared in MIT Tech Talk on April 29, 2009 (download PDF).