In 2000, the US National Institute of General Medical Sciences of the National Institutes of Health funded the Protein Structure Initiative (PSI), a ten-year project to uncover the three-dimensional shapes of a wide range of proteins. The Joint Center for Structural Genomics (JCSG), based at The Scripps Research Institute in La Jolla, California, USA, is one of four large-scale centers involved in the production phase of the PSI. Four centers focus on high-throughput protein-structure determination, six specialized centers deal with difficult-to-solve proteins, such as membrane proteins, and two others provide new approaches to molecular modeling.

Ian Wilson is director of the one of four large production centers for the Protein Structure Initiative.

Ian Wilson, director of the JCSG, thinks the timing is perfect for the PSI centers to produce large numbers of new protein structures for the research community: “With more and more DNA sequences becoming available each day, the possibilities for the future of protein structure determination are tremendous.” A central goal of the PSI is to enable the prediction of three-dimensional structures for most proteins from knowledge of their corresponding DNA sequence. In principle, this can be done by inferring the structure of a protein based on the known structure of representative members of the protein's family. “Most of the big protein families have been mapped—but still for 70% of known families we have no structural data,” says Adam Godzik, of the Burnham Institute for Medical Research in La Jolla, and head of bioinformatics at the JCSG. This makes for a huge number of potential target proteins if one wants to have representatives from all families and therefore raises difficult questions: 'how do you choose which families to target and then which proteins within those families to obtain structures from?'.

Targeting the protein universe

“We are dealing with a continually expanding universe of proteins, so we had to have some rules about target selection,” says Wilson. For the PSI, seventy percent of the target protein families are communally selected through PSI's Target Selection Committee. “We all sit down and execute a draft to decide which families each center will get,” says Wilson. “By virtue of choosing particular families we avoid overlap, but also with this selection process each center can optimize specific targets within families for themselves,” says Godzik. Another 15% of target proteins are decided upon by each center, and the final 15% are community targets proposed by outside researchers.

Godzik says that it is most effective for individual centers to decide which proteins to go after within the families they have been assigned because each center relies on different 'reagent genomes'—large sets of genomic DNAs used to isolate homologous sequences. At JCSG, it is Godzik, along with his bioinformatics team, who is responsible for determining the specific proteins JCSG will work on. By aligning a protein family with all 100 genomes available at JCSG, they first identify all homologous proteins. Then, using their own software, they assign a crystallization score to each homologous gene identified within the family—a measure of the likelihood of success of the corresponding protein in the structure determination pipeline. “We take the ones that we predict to be most likely to succeed from this tool, and then we work our way down the list,” he says.

The GNF developed high-throughput bacteria fermentation system used by JCSG for large-scale protein production. (Courtesy of JCSG.)

Of course, there are those cases where nothing seems good. But thanks to the rapid advances in DNA sequencing technology and the explosion of fields like metagenomics (see Box 1), they have a very simple remedy to this situation: a continual update of potential targets in each family. “It is for this reason that we keep adding new genomes to our collection,” says Godzik.

Proteins for all

Any protein-structure-determination center is only as good as the quantity and quality of the proteins they generate. And during any given week of the year the Crystallomics core of the JCSG produces 25–50 new purified proteins. To attain this impressive yield, JCSG researchers were forced to alter some traditional methods for protein production.

Heath Klock, a JCSG research scientist, along with Scott Lesley, the Crystallomics core leader, and several colleagues devised the first step in the protein expression pipeline. “PIPE cloning is a ligation-free PCR-based cloning strategy that is amenable to cloning thousands of inserts into expression vectors in a very short period of time,” says Klock. PIPE, or polymerase incomplete primer extension, uses PCR primers with overlapping 5′ ends to amplify vectors and inserts that can then be mixed and transformed with the resulting colonies screened by colony PCR1.

For initial protein expression evaluation, JCSG scientists often rely on microexpression screening. “The real value here is that it gives you a peek ahead,” says Mark Knuth, the Crystallomics core manager. They use a Vertiga commercial shaker from Thompson Instrument Company adapted for 96-well deep blocks and capable of achieving high-density cultures rapidly. Protein expression and solubility are evaluated using small-scale purification by immobilized metal affinity chromatography. Amazingly, microexpression has a very high predictive value for successful expression in JCSG's large-scale fermentor, which is capable of generating 96 high-density cell cultures in 100-ml centrifuge tubes. “Even though we can put a lot of tubes through the large fermentor, it is still time wasted if you put things in there that you know are not going to work,” notes Knuth.

The protocols for purification of proteins after large-scale expression have evolved over time. “I think that one of the things that differentiate us from other centers is that we do higher-density fermentations, but we also do somewhat less purification,” says Knuth. For the majority of proteins that go to crystallization trials now, a purification on a nickel resin with subsequent cleavage by tobacco etch virus (TEV) protease and a reverse nickel purification is performed. During the early years of the PSI, JCSG worked on developing high-throughput secondary purification, but found that the material they got after these secondary purifications was equivalent, from a crystallization perspective, to the nickel resin–purified, TEV protease cleaved and then reverse nickel purified proteins.

After large-scale expression and purification, each protein is put through a series of assays before any attempt at obtaining crystals for structure determination. These tests include analytical sizing, mass spectrometry and SDS electrophoresis on a Bio-Rad Dodeca multigel runner to assess purity. But Knuth is quick to note that they usually rely on one test above others. “Analytical size exclusion chromatography has the greatest predictive leverage of how a protein will fare in the crystal trials.”

The JCSG Automated high-throughput crystallization platform, called CrystalMation, was built by Rigaku Automation. (Courtesy of JCSG.)

The crystallization trials

“For crystallization we usually set up the JCSG screen, which is four blocks of 96 conditions all done at two different temperatures,” says Marc Elsliger, the administrative core manager for the JCSG. Identifying these 384 conditions to test for crystallization did not happen overnight.

“These conditions were based on the testing of all available commercial screens,” says Elsliger. Over the early years of PSI-1, the JCSG tested conditions to identify which gave the best hits from a wide array of proteins during crystallization process. The fruits of this effort are not only available to the JCSG for their screens, but these conditions have also been commercialized recently through Qiagen as the JCSG Core Suites I–IV and JCSG+.

Protein crystallization trials at the JCSG rely on automated robotic systems. The original breakthrough system, developed at Syrrx/GNF, used nanoliter volumes to yield diffraction-quality crystals. They set up trials under all 384 conditions but then moved the plates into either 4 °C or 20 °C rooms for automated imaging over a 28-day period. The new JCSG CrystalMation system developed by Rigaku Automation is fully integrated with climate-controlled incubators. The introduction of the CrystalMation system boosts the capacity for crystallization trials at the JCSG to approximately 4,000 96-well plates per month.

But even with automated crystallization systems, Elsliger points out that two parts of the crystallization process proved difficult to automate. “The first is the scoring of the images to identify which of the drops contain usable crystals, and the second is the actual harvesting of the crystals.” He notes that it is challenging to image a crystal in a liquid drop because of the curvature of the drop, the requirement for optimal lighting and the fact that clear crystals in a clear liquid background cause problems for scoring crystals.

At JCSG, several people have been trained to extract around 2,500 crystals per month from the drops and place these into the aluminum crystal cassettes in which they make their final journey north to the Stanford Synchrotron Radiation Laboratory (SSRL) for X-ray diffraction.

Look mom, no hands!

Once the crystals reach the Structure Determination core at SSRL, headed by Ashley Deacon, they are ready to be placed into the Stanford Auto-Mounter (SAM) system. “We were a large part of the proof of concept that users could use this on a daily basis,” says Elsliger when discussing the JCSG's role in promoting the use of the automated beam line at SSRL.

Automated protein crystallography beamline at Stanford Synchrotron Radiation Laboratory, showing the 4-axis robot and the dispensing Dewar of the Stanford Auto-Mounting system. (Courtesy of JCSG.)

The only human involvement once the crystals reach SSRL is placing the aluminum crystal cassettes into a liquid nitrogen–cooled dispensing Dewar; then the SAM systems Epson ES553S 4-axis robot takes over, opening the Dewar, obtaining a crystal and then reliably centering each crystal with the X-ray beam. Each step of this process can be observed, remotely, by JCSG scientists in La Jolla. “The value is that you can run the beam line much more efficiently,” notes Elsliger. Using the SAM system, the beam line can operate 24 hours a day with only occasional human intervention when the cassettes in the dispensing dewar need to be changed. Remote operation of the beam line is not unique to the JCSG—the three other major production centers use this approach at other beam lines as well. Once the diffraction data have been obtained, automated analysis software interprets the crystallographic data for structure determination.

This also happens to be the point at which the JCSG has the most difficulty. “We take our biggest hit in going from a crystal hit to a usable dataset,” says Wilson, noting that only 50% of crystals are usually big enough to be sent to SSRL, and of those around 50% will diffract to sufficient resolution to determine the structure. But the bottom line numbers show that despite the obstacles, all the centers of the PSI are making considerable headway into the protein universe. To date the JCSG has deposited more than 530 new structures in the Protein Data Bank, and structures for 335 of the original list of 1,269 largest protein families given highest priority by the PSI in 2005 have already been solved—200 from the PSI alone. “This shows that the PSI has made quite a dent in these proteins families already,” says Wilson with a smile, “I find these numbers very encouraging for the future.” See Table 1.

Table 1 Suppliers guide: companies offering structural genomics products
Full size table