Nature holds no patent on protein design, but scientists have struggled to create new proteins that aren’t based on already-existing ones. Being able to design proteins that can bind to specific, chosen targets would open up a world of possibilities, including medicines targeting previously undruggable pathways and chemical sensors that detect specific molecules.
Now, researchers report in Nature that they’ve achieved the task, creating a 110-amino-acid protein, unlike any in nature, that binds a specific molecule, activating it to glow green in cells (Nature 2018, DOI: 10.1038/s41586-018-0509-0). “We can now build an unlimited number of protein structures from scratch and optimize them for the function that we really want,” says David Baker of the University of Washington, Seattle, who led the work.
Scientists have had some success designing proteins with a coiled α-helical structure, but β-sheets—flattish strands held together by hydrogen bonds—have proved a stumper because the strands tend to clump together. Yet natural protein binding sites often occur within a three-dimensional structure called a β-barrel—a β-sheet twisted to form an open end where the molecule binds and a closed end that stabilizes the structure.
Baker’s team, too, worked on the problem for several years without success, until they realized that the geometry of the hydrogen bonds that held these structures together created strain, making the β-barrel shape unstable.
The researchers found that introducing bumps called glycine kinks and β bulges into their computational model created stable β-barrel structures. “You basically have to relieve this strain by introducing asymmetry,” Baker says.
The team then developed a new computational tool to optimize where their desired small molecule could bind and what side-chains the protein would need to hold onto the molecule. The tool predicted 56 β-barrel structure sequences that could potentially bind a molecule called DFHBI, which fluoresces green only when bound to a protein. Of these, 20 folded into monomeric proteins when synthesized, and two of those activated DFHBI.
The team further optimized their design with the help of screens for DFHBI binding in yeast cells and computational modeling of the structures. This round produced three better versions. The researchers were able to crystallize two of these proteins, and found them to be strikingly similar to their computationally predicted shape, Baker says. All three performed swimmingly when expressed in Escherichia coli, yeast, and mammalian cells, switching DFHBI into its glowing green state.
“This is one of the most impressive and sophisticated examples of computational protein design, which highlights the potential of this approach for protein engineering,” says Ingemar André, who designs proteins at Lund University. The challenge going forward, he says, is to automate the procedure and to make it amenable to binding more complex molecules.
The computational tools the researchers developed are available to the research community. The team is now working to refine the approach, experimentally testing large numbers of protein for stability and function.