Crystal Illumination

Researchers use automation to boost crystallography efforts, resulting in more data in less time, from less protein

Laura Lane
Jan 18, 2004

Courtesy of Syrrx

Crystal mounting and alignment robot developed at Lawrence Berkeley National Laboratory for use at the Advanced Light Source. Integrated into a synchrotron beamline with intelligent control software, these systems enable fully automated data collection.

The field of protein crystallization is turning heads. Sure, the work generates pictures of subtle beauty, but there's a commercial angle, too, and drug developers have lately come to recognize the potential value of structural genomics. Indeed, deducing protein structure is "a key element for drug discovery," explains Lance Stewart, vice president of the BioStructures division of deCODE Genetics in Reykjavik, Iceland.

With structure in hand, pharmaceutical companies need not rely on guesswork to find promising new drugs. Instead, they can design, in a virtual in silico lab, molecules that will nestle snuggly in the active site of target proteins. Getting to that point, though, will take some doing.

That's because proteins are...


<p>Courtesy of T. Earnest, Lawrence Berkeley National Laboratory</p>

The Syrrx Agincort Robot; (right) enlarged view of an actual crystallization plate shows the size of the experiments relative to a penny. With drops 40-fold smaller than in conventional crystallization trials, far less protein is required, and the results are known sooner.

Finding appropriate crystallization conditions tends to be a huge, tedious bottleneck, so engineers and biologists have collaborated over the last few years to develop liquid-handling devices that can automate that process. Cued by a computer, these instruments dispense a unique set of fluids into each well of a microtiter plate, usually containing 96 wells, though some devices can handle up to 1,536. Microplates employ smaller volumes per experiment than do other reaction vessels, thereby reducing the total amount of purified protein required. These smaller volumes, in turn, can reduce crystallization time up to 10 times.1

One of the first automated screening devices, the IMPAX system, was developed in the early 90s through a collaboration of Douglas Instruments, Berkshire, UK, and the Imperial College London. The system employs the microbatch method, in which precipitant and protein solutions are injected into a droplet of oil, which prevents evaporation of the small volumes used. The precipitant solution reduces the solubility of the protein, driving it toward supersaturation and crystallization, explains Naomi Chayen, principal research fellow at ICL, who helped develop the technique and IMPAX.

"It's often not possible to work with large quantities," says Chayen, explaining the difficulty in obtaining large amounts of purified proteins. "With the microbatch method, you can set up numerous experiments as you only need to use small amounts of protein."

Improving on IMPAX, Douglas' newer Oryx 6 instrument dispenses oil on top of the precipitant and protein solution solutions that have already been injected into the well. This is more accurate than dispensing the oil first, says Patrick Shaw Stewart, the company's founder and CEO. Increased accuracy means that the Oryx 6 can work with protein volumes as low as 0.1 microliters, while IMPAX's lower limit is 0.3 microliters. And while IMPAX is limited to 48 solutions, the Oryx 6 can dispense more than 192 precipitant solutions per run.


Other instruments promote crystallization in other ways. In the vapor diffusion (VD) method, a drop of solution containing the protein and precipitant is usually placed either on a platform or on an inverted glass slide above a reservoir of precipitant solution. Because the concentration of precipitant in the reservoir is greater than in the protein solution, water evaporates out of the protein solution, causing supersaturation.

The Agincourt™ from San Diego-based Syrrx is a VD device that can process 3,000 experiments of different solutions in 96-well plates in one hour, says Kenneth Goodwill, Syrrx's director of business development. Using small volumes, usually 50 nanoliters, is so crucial to its drug-discovery efforts that the company has patented its proprietary Nano-volume Crystallization™ protocol.

Other companies also automate their VD methods. Middleton, Wis.-based Gilson's 925 PC Workstation can handle four 96-well plates in one batch with protein volumes as low as 1.0 microliters, while deCODE's MatrixMaker™ can handle 60 or more different stock solutions at once. Hudson, NH-based Matrix Technologies' Hydra® Plus-One System is capable of dispensing volumes as low as 100 nanoliters into 96- or 384-well plates.

Fluidigm of South San Francisco, Calif., uses another crystallization method, free interface diffusion (FID), in its Topaz™ Crystallizer.2 Central to this automated system is the Topaz Screening Chip, which automates the experimental setup of 144 FID experiments, says Jaime Bartels, Fluidigm's public relations manager. The palm-sized chip automatically divides the protein – an average of 12.5 nanoliters per experiment – and facilitates diffusion with the precipitant solution, leading to crystallization often within 24 hours.


Once the crystallization trials are running, digital imagers can analyze the plates and screen out the many wells that lack any evidence of crystal formation. With crystals in only a small percentage of wells, researchers can handle the volume of images that require manual analysis.3

"We rarely have more than one percent form crystals," DeTitta says. While he points out that the current automated image analysis systems aren't yet perfect, they still do a great service in reducing the field of thousands to a mere few.4

The RoboMicroscope II from RoboDesign International of Carlsbad, Calif., is one such instrument. The company classifies images into one of four categories: clear, precipitate, crystals, or other, says Tommy Bui, director of business development. The patented method, called CPXO, has achieved a 90% correlation in studies that pit the system against human eyes. Only a small fraction of trials produce crystals, Bui notes. "We're confident we can filter out the others, greatly reducing the number of images you have to look at."

Bernard Rupp, founder of the structural genomics group at the Lawrence Livermore National Laboratory in Livermore, Calif., has directed the development of CrysFind, a software application trained to detect differences in the images. Depending on user-defined parameters, CrysFind pinpoints the images that show the most promising crystals. The accuracy increases with higher quality pictures, according to the laboratory's Web site, which also advertises the software as available for licensing. Also in the automated microscope and data analysis market are deCODE with its Crystal Miner™ and Diversified Scientific of Birmingham, Ala., with its CrystalScore™.5


Once researchers fine-tune the formula for growing good-quality crystals of their target protein, they're ready to probe its structure with X-ray crystallography. Researchers fortunate enough to have access to one of the three third-generation synchrotrons in the world have an advantage, says Stephen Burley, chief scientific officer of Structural GenomiX (SGX) of San Diego. These synchrotrons emit X-ray beams that are much more concentrated, making them more intense and brighter than those of the second-generation synchrotrons, such as the one at Brookhaven National Laboratory in New York.

Most companies share the 80 beamlines at the third-generation synchrotron at the Argonne National Laboratory, located near Chicago. But SGX, having spent $7 million to build a dedicated beamline at Argonne, can perform crystallography at will. "The decision to make the investment gave us a strong competitive advantage," Burley says. "As soon as we get a crystal, we can collect data."

Plugging in at Argonne allows companies to obtain structures orders of magnitude faster than researchers using older synchrotrons. Third-generation synchrotrons also provide the edge in obtaining the structures of small crystals, Burley says. This allows for the elucidation of structures of the many proteins for which large crystals can't be acquired. "That's the difference between success and failure," he says.

The X-ray crystallography step has not been immune to the automation revolution. Hans Bartunik at the Max Planck Research Institute in Hamburg, Germany, has developed a device that picks up a frozen protein crystal from its storage unit and centers it exactly in line with the X-ray beam. After data collection, the device returns the crystal. Once the crystal is mounted and centered, the device also aligns the crystal at the proper angle so that the process is entirely automated. Similar devices have been developed at the Scripps Research Institute in San Diego, the European Synchrotron Radiation Facility in Grenoble, France, and Abbott, which actually patented the automated crystal-alignment technology.6

Automated analysis capabilities for screening out poor-quality crystals prior to data collection are in the late stages of development, says Bartunik, who is head of the protein dynamics group at Max Planck. "When you look at a dish, you might see a crystal, but it might not have the three-dimensional periodic structure required for diffraction," he says, explaining that his automated device will return the crystal to the plate if it doesn't initially show preliminary evidence of having a proper crystal structure.


While research in structural genomics is receiving a giant boost from the automated devices on the market, one sector is still stumping even the best crystallizers: membrane proteins. The vast majority of drugs target membrane proteins, such as receptors, making crystallization of membrane proteins a priority, says Martin Caffrey, a structural biologist at the University of Limerick, Ireland. Yet current crystallization methods don't work for membrane proteins that have hydrophobic membrane-embedded regions, which avoid the water that's central to the principles of vapor diffusion and microbatch.

So far, Caffrey has found that a combination of detergent, water, lipids, and protein can form a liquid-crystal phase, which he calls a cubic mesophase.1 Adding salt removes the water, forcing proteins to crystallize.

Now Caffrey's lab has automated the method, which requires finely manipulating microscopic volumes of a lipid/protein mesophase that is as viscous as toothpaste. The robot his lab has built can manipulate 50-nanoliter-volume mesophases, representing about 200 nanograms of protein.

"We're trying to understand the molecular mechanism as to how this mesophase method works," Caffrey says, and after three years of work, he has made good progress. Using lipids with 16 to 19 carbon atoms – the approximate length of the hydrophobic interior of a cell membrane – works best, he says. As for detergents, Caffrey has found that octyl glucoside and a few other nonionic detergents work particularly well. Nevertheless, "You can't assume anything in this business."

"A lipid's eye view of membrane protein crystallization in mesophases," Caffrey M, Curr Opin Struct Biol , 2000 Vol 10, 486-97


Though far less streamlined than, say, genomics, protein crystallization and crystallography is no longer the alchemy of yore. Instead, a kind of group wisdom pervades the field, with crystallizers taking copious notes and building databases such as the Protein Data Bank, which detail successful experimental conditions. Other databases also contain valuable data on failed conditions. "From our store of shared experiences, accumulated wisdom, and tabulated data have come rapid and efficient screening procedures and detailed recipes," writes crystallization expert Alexander McPherson in his book, Crystallization of Biological Macromolecules (Cold Spring Harbor Laboratory Press, 1998).

Bottlenecks will remain, of course, as few insiders believe crystallization conditions can ever be predicted by merely studying the amino acid sequence. Yet many are hopeful they may one day be able to decipher trends and categorize certain types of proteins, such as protein families, which will crystallize under a specific set of conditions.7

In the meantime, structure-based drug design has proven to be a good strategy for pharmaceutical companies, and not just for the small start-ups. Giants like Abbott, realizing the inefficiency of blind combinatorial chemistry approaches, are shunting more resources into their protein-structure efforts. As a more direct route to developing chemical compounds that can specifically bind to protein targets, drug makers can increase potency and decrease side effects. That, in turn, will decrease costs, and the time-to-market.

Laura Lane is a freelance writer in San Francisco.

<p>Figure 3</p>

Courtesy of Argonne National Laboratory

The Syrrx Agincort Robot; (right) enlarged view of an actual crystallization plate shows the size of the experiments relative to a penny. With drops 40-fold smaller than in conventional crystallization trials, far less protein is required, and the results are known sooner.

Interested in reading more?

Become a Member of

Receive full access to more than 35 years of archives, as well as TS Digest, digital editions of The Scientist, feature stories, and much more!
Already a member?