Engineered Proteins Will Lead to "Synthetic Biology"

Duke University Medical Center biochemists have developed a computational method to design proteins that can specifically detect a wide array of chemicals from TNT to brain chemicals involved in neurological disorders. In a paper in the May 8, 2003, issue of the journal Nature, they demonstrate the breadth of their design method, and also that such sensor proteins can be re-incorporated into cells to activate cellular signaling and genetic pathways.

The researchers said their achievement constitutes an important step toward a new technology of “synthetic biology,” in which scientists construct tailor-made organisms for a variety of tasks, including as “biological sentinels.” Such sentinels could find wide use in medical and environmental applications, said the paper’s senior author, Associate Professor of Biochemistry Homme Hellinga, Ph.D.

“You could imagine a bacterium engineered to respond to TNT by fluorescing that could be used to detect mines for humanitarian demining operations,” he said. “Also, one might develop a plant whose cells would change its color in response to a pollutant. Such bushes could be planted around chemical facilities to detect groundwater pollution from the facility.

“Such scenarios might sound like something out of science fiction, but they are entirely feasible over the long term using computational design approaches to create sensor proteins,” he said.

Hellinga is already developing a glucose-sensing protein that could be used as the basis for a monitor — and ultimately an artificial pancreas — for people with diabetes. Also, the TNT-sensing protein is intended to provide free-roaming underwater robots with the capability of “sniffing” a plume of TNT emanating from unexploded ordnance, tracking it to its source, to aid the U.S. Navy’s cleanup program.

The scientists’ research is supported by the Office of Naval Research, the Defense Advanced Research Projects Agency’s “Simulation of Bio-Molecular Microsystems” (SIMBIOSYS) program and the National Institutes of Health.

In the Nature paper, Hellinga and co-authors Loren Looger, Mary Dwyer and James Smith describe how they redesigned proteins from the gut bacterium E. coli. They used “periplasmic binding proteins” which are normally part of the bacterium’s chemical-sensing system by which it detects nutrients. Such protein receptors detect their target molecule via an “active site” with a precise complementary shape and binding properties that fits only that molecule, called a “ligand” — like a key fitting a lock. Proteins consist of long strings of amino acid units that fold upon themselves to form intricate globular structures.

Basically, the computational design process developed in Hellinga’s laboratory involves redesigning a normal protein’s “lock” to fit a very different molecular key. The computational process narrows down to a manageable few the vast number of possible mutations and their corresponding structures, to fit a particular molecule.

“We call these supra-astronomical calculations, because they can produce more possible combinations than there are particles in the known universe,” said Hellinga. “The proteins we work with have binding sites of about 15 amino acid residues. And each of these 15 residues could consist of any of the 20 known amino acids. And we do our calculations in three-dimensional space, so each amino acid is described by a number of different structures that we call rotamers. And we use about 6,000 rotamers to represent the 20 amino acids. So, that alone gives us 6,000 to the 15th power possible sequences.”

What’s more, when the researchers factor in the further geometric complexities of how the ligand molecule docks to the active site, said Hellinga, the number of possible designs skyrockets to 10 to the hundredth power possibilities.

However, the Duke biochemists have enhanced and applied a mathematical technique called dead end elimination — pioneered by mathematician Johan DeSmet of the University of Leuven in Belgium and his colleagues — to drastically narrow the possible structures. Using this approach, their system can narrow the candidates down to a reasonable number with several days’ worth of calculations on a 30-processor computer network. The high-speed network, called a Beowulf cluster, consists of linked processors of the type available on high-end commercial desktop computers.

In the Nature paper, the scientists described adapting E. coli proteins to detect three very different molecules of environmental and clinical importance.

— TNT, which is a nonbiological compound and carcinogen that the Navy is seeking to detect, as part of developing a TNT-sensing robot to aid its environmental cleanup effort.

— lactate, which is an indicator of metabolic stress in the body and also associated with certain cancers. Thus, a lactate-sensing protein could locate metabolic distress in the body. Also, lactate has a “left-handed” and “right-handed” molecular form. Demonstrating that an engineered protein could distinguish such “chiral” forms would be of great interest to pharmaceutical firms, said Hellinga, because they must purify drugs to eliminate unwanted chiral forms that can be highly toxic.

— serotonin, which is a neurotransmitter molecule that nerve cells use to trigger nerve impulses in neighboring cells. Fluctuating serotonin levels are associated with certain psychiatric disorders and elevated levels are indicative of certain bowel tumors.

Also, building a serotonin-sensing protein is a special design challenge, said Hellinga, because the molecule has distinct regions that are negatively or positively charged. “And frankly, we thought it was a rather dramatic achievement to build a receptor for a neurotransmitter from a bacterial protein,” said Hellinga.

“We basically tried to show that this is a very general design approach,” he said. “We demonstrate that, within reason, no matter what the chemical nature of the ligand, we can design an active site to specifically bind it.” Also to demonstrate the adaptability of their technique, the researchers started with a number of different E. coli proteins in their design effort.

They applied their computational method to automatically generate a collection of possible designs and to narrow them down through dead end elimination to arrive at a workable number of candidates.

Then, the scientists actually constructed a selection of the candidate proteins by genetically altering E. coli to produce them. To measure the proteins’ sensing capabilities, they used a technology that they had developed earlier for the periplasmic binding proteins, and engineered the molecules to incorporate a fluorescent molecule, called a fluorophore, attached such that the protein would only fluoresce if the target molecule had plugged into the protein.

The resulting sensor proteins proved in test tube studies to have high affinity and specificity for their target molecules, said Hellinga. For example, he said, both the TNT-sensing and serotonin-sensing proteins could not be fooled by “decoy” molecules that were very closely related to TNT or serotonin in structure. Also, the lactate-detecting protein preferentially bound only to one chiral form of the molecule, proving that the design system could produce such specific sensing proteins.

The researchers used a feedback technique, called “quantitative structure-activity relationship,” (QSAR) to apply data on molecular interactions in the experimental proteins to optimize the theoretical design calculations even further.

Finally, Hellinga and his colleagues sought to demonstrate that the sensor proteins could function in the machinery of living cells.

“We had shown that these proteins could function as sensors in the test tube by labeling them with fluorophores,” said Hellinga. “But we were also interested in putting them back into a biological system and showing that we could modulate the behavior of that biological system. It’s an important step toward developing what we call ’synthetic biology.’”

Thus, the scientists showed that they could insert either the TNT-binding protein or the lactose-binding protein into E. coli bacteria in such a way as to activate a “reporter gene,” which signaled that their molecule had become a functioning component of the bacterial genetic machinery. Such experiments proved the importance of their new design approach, said Hellinga.

“The potential implications of this technique are very broad,” he said. “Essentially what we have done is establish control over molecular recognition of small molecules in biology. Such recognition is central to the most basic processes in organisms, and we have demonstrated that not only can we design highly specific receptors, but we can put them into biological systems to control them.

“Also, we have shown that we can create sensors that are exquisitely sensitive, even to the chiral forms of molecules. Such sensors could prove invaluable to pharmaceutical companies, since many drugs are chiral — with one form being pharmacologically active and the other being either inactive or toxic.”

According to Hellinga, the Duke computational design system is far superior to the other major technique — called “directed evolution” — used to tailor proteins to bind specific molecules. In directed evolution, researchers create a large library of candidate proteins and screen them for affinity and specificity to a target molecule.

“On the one hand directed evolution is a very elegant method, in which they’ve essentially programmed nature to give the right answer,” said Hellinga. “But there are real limitations on how large a variety of possible protein structures can be surveyed. Our computational methods can overcome that limitation.”

In terms of commercial applications, he said, the university is exploring the launching of a company to market and support the design methodology and applications developed by Hellinga and his colleagues.