Alan Dove

Determining protein structures remains one of the tightest bottlenecks in biology, but a new crop of technologies is starting to make structural biologists' lives a little bit easier.

Crystal gazing

or at least they try. Though it provides unparalleled structural detail, X-ray diffraction requires large, well-ordered crystals of a very pure sample, a tall order for any protein and a near impossibility for the complex membrane proteins that interest many biomedical researchers.

"Up until recently, there were many more review [articles] than there were structures," observes Michael Wiener, associate professor of biophysics at the University of Virginia (Charlottesville, VA). The chief problem is that growing protein crystals is a distressingly empirical exercise, with researchers typically setting up thousands of experiments to find the right conditions for a single crystallization. Because of the vast chemical diversity of proteins, conditions that work perfectly for one structure may fail completely for another.

The problem is even more complex with membrane proteins. "The entity that's being crystallized for a membrane protein is a protein-detergent complex ... so the properties of both the protein and the detergent influence crystallization," says Wiener.

Traditionally, crystallographers have used a brute-force approach to solve these problems, setting up larger and larger collections of experiments to find the right conditions. Laboratories increasingly use automated liquid handling systems and other robots to do this tedious work, but Wiener and his colleagues are hoping to simplify the problem instead. Borrowing a lesson from high-throughput screening efforts in the pharmaceutical industry, the researchers, under a NIH "Roadmap" grant, are developing strategies to eliminate bad leads earlier in the screening process.

"If there's 200,000 possible experiments that you could perhaps set up, wouldn't it be nice to know the 80 percent that you shouldn't, because of some physical property?" asks Wiener. To address that, the investigators are taking a three-pronged approach: identifying the protein-detergent interactions that have correlated with successful crystallization in the past, performing screens on specific protein and detergent combinations, and trying to engineer chaperone compounds to hold membrane proteins in conformations that improve crystallization.

While scientists like Wiener and his colleagues work on new strategies, other crystallographers are trying to improve old ones. For example, one of the oldest tricks to get a crystal to grow is the use of a nucleating agent, either a small seed of the desired crystal or some other substance that can help stabilize the core of a new lattice. Researchers have used everything from exotic salts to soot to try to induce crystal growth, but recent work at Imperial College in London and the University of Surrey shows that there are still more nucleants to be discovered.

The researchers, led by University of Surrey physics professor Richard Sear, developed a statistical model to describe the nucleation of protein crystals, then tested the model using Bioglass, a well-known gel-glass substance normally used to induce bone growth and wound healing. Though the scientists caution that the quest for a "universal nucleant" is not over, Bioglass does seem to prompt crystallization in a wider range of proteins than any other nucleant.

Figure 1. The New York Structural Biology Center, a major multi-institution center, utilizes the latest generation of high-field NMR equipment.

Sit or spin

The other major technique for solving high-resolution protein structures, nuclear magnetic resonance (NMR), relieves researchers of the need to grow protein crystals, but comes with its own set of limitations. To date, NMR structural studies have focused mainly on small, soluble proteins or protein fragments, as the procedure becomes considerably more difficult with molecules that are large or embedded in membranes.

"If you go from a protein with 100 amino acids to one with 500 amino acids, two things happen. One is the spectra are going to be more complicated, the other ... is that the protein will tumble more slowly in solution," says Stanley Opella, professor of chemistry and biochemistry at the University of California (San Diego, CA). The slower tumbling is especially troublesome, as it causes the peaks of the NMR spectrum to broaden, obscuring important structural details.

In a bit of technical jiu-jitsu, Opella and his colleagues are studying the structures of large membrane proteins by slowing down their tumbling even further to a dead stop. With the protein molecules all aligned in the same orientation on lipid bilayers, the researchers then use solid-state NMR in a high magnetic field to probe the protein-lipid structure.

Chad Rienstra, assistant professor of chemistry at the University of Illinois (Urbana-Champaign, IL) and his colleagues are also using solid-state NMR to study membrane-bound proteins, but they are addressing the tumbling problem differently. Rather than aligning the proteins in a single, uniform orientation, Rienstra spins the samples at a specific "magic angle," simulating a synchronized tumbling motion that makes all of the molecules in a sample appear symmetrical.

"We perform the magic angle spinning to simulate, in an experimental sense, the effect of this rapid tumbling that would naturally occur for a small molecule, so it gives us high resolution spectra even if the molecule is, on an atomic scale, immobile," says Rienstra.

Rienstra and Opella see their approaches as complementary. "Some problems we can do and some they can do, and both [techniques] are still in the early stages," says Opella. Nonetheless, major NMR suppliers like Bruker (Billerica, MA) and Varian (Palo Alto, CA) already sell appropriate equipment for both techniques off the shelf, so early adopters can try it themselves.

Figure 2. Technologies for magic-angle spinning solid-state NMR and resulting data from membrane proteins. (a) Varian BioMAS 3.2 mm probe design schematic. The cross section illustrates the mechanical and electrical components that surround the sample, packed in a 3.2 mm diameter rotor that spins up to 25,000 revolutions per second, with a coil specifically designed for membrane protein studies. (b) An example of data planes extracted from 3D experiment on the membrane protein DsbB, a disulfidebond forming enzyme from E. coli. Data were acquired at 750 MHz 1H frequency in the Rienstra laboratory at the University of Illinois at Urbana-Champaign by Ying Li and Deborah Berthold. (c) Picture of the DsbB membrane protein sample immediately prior to packing in the rotor for NMR experiments. Single crystals are not required to obtain such high quality magic-angle spinning NMR data. (d) Picture of a 1.6 mm diameter FastMASTM rotor (Varian), below a dime. New rotors such as these permit studies of membrane proteins at quantities of a milligram or less. Click to enlarge.

New NMR technologies aren't always that easily available, though, especially for researchers who want to use the latest generation of high-field instruments. As the magnets for these devices get larger, their structural resolution increases, but not as fast as their price, feeding a new trend toward "big science" consortia and shared multi-institution facilities. While that approach has become the norm in genomics, NMR spectroscopists are still divided over applying it to their own field.

"I think a center like ours does have the potential for letting people have much greater access to this kind of technology," says David Cowburn, head of physical biochemistry at the New York Structural Biology Center (NYSBC), which is funded by ten New York-area institutions and several Federal and state agencies.

Regional centers like the NYSBC can provide individual researchers with more powerful NMR equipment than they could obtain or maintain on their own, but some investigators say the upgrade isn't always worth the effort. "These facilities offer instrumentation that's a couple of steps up but the quality of data one can get on an instrument at his home institution might be almost as good," says Rienstra.

Pump up the volume

Whichever brand of physics they use, structural biologists traditionally take a serial approach to their projects, studying only one or a few proteins at a time. While that strategy may always yield the best biochemical understanding, it does little to break the bottleneck of proteome-scale analysis, where thousands or millions of interesting leads wait to be studied.

The Protein Structure Initiative (PSI), first funded by the NIH in 1999, was designed to address that problem with a massively parallel, factory-like strategy. "The idea was to develop new methods and techniques and to set up the structural genomics pipelines in which you could automate a lot of the steps," says John Norvell, chief of the Structural Genomics and Proteomics Technology branch at the NIH's National Institute of General Medical Sciences (Bethesda, MD).

While some initially criticized the PSI for focusing on low-hanging fruit, the results around 1,300 new structures and a slew of technical breakthroughs in the project's first five years now speak for themselves.

The PSI is now in its "production" phase, where the goal is to produce 3,000 more structures representing as many different protein families as possible. "Then you can use homology modeling methods to get structural information, or at least predictions, for all other members of [those] families," says Norvell. He adds that the combination of new structures and improved structure-predicting algorithms could give biologists as many as a million new high-resolution views of protein molecules. Perhaps that will be enough to keep us busy for another 300 years.