By Elizabeth Gardner
Sept 9, 2002 | The drug development process has historically been funnel-shaped: Hundreds of possibilities pour in, but scant few winners dribble out.
The objective of Structural GenomiX Inc. (SGX) — one of a handful of young companies pioneering high-throughput protein-structure determination — is to reshape the process into more of a cylinder: higher-quality leads going in, and a steady stream of winners emerging.
To do that, the San Diego company is depending on a homegrown Oracle-based laboratory information management system (LIMS) to identify high-quality leads as early as possible in each new project. Combining this database with a heavily roboticized protein-crystallization operation and unmatched access to high-throughput X-ray facilities at a major national laboratory, SGX is poised to solve dozens of protein structures and test thousands of drug leads per year. Such a goal would have been essentially impossible just a few years ago.
"We can work with medicinal chemists testing compounds," says Sean Buchanan, the company's director of bioinformatics. "They start off with a compound that's effective but has problems. They need to test derivatives and see if they're on the right track. They can generate a
new series of compounds every couple of weeks, and we can test them immediately and show the chemists if they're being led down a red-herring path. What's new is to be able to do it at such capacity."
The SGX game plan rests on the proteomics equivalent of an ace up the sleeve: its very own beamline
|Researchers employ state-of-the-art synchrotron radiation tools for studying protein structures
at one of the most brilliant sources of X-rays in the world (see "X-ray Specs,"
right). When fully operational, SGX' outpost at Argonne National Laboratory should be able to collect data on as many as 50 crystals a day and solve thousands of structures a year.
In July 2000, SGX entered into an agreement with the U.S. Department of Energy, which owns Argonne, to fund development of protein-crystallography beamlines. The company spent $8 million on the first beamline, which began operation last year and is now ramping up to full speed. It plans to build two additional beamlines, which will give it by far the best synchrotron access of any company in the field.
Kevin D'Amico, vice president of synchrotron research, estimates that current projects will yield a dozen solved protein structures with as many as 100 co-crystals (representing individual drug leads) for each one. "Our pipeline has a lot of controls and procedures in place to supply feedback early in the process," he says. "We want to identify the blind alleys as soon as possible, before we get too far along in them. In some instances, the structural work merely validates the conclusions, because it doesn't come early enough in the process. The whole goal is to try to turn it around so that structural information can help drive the decisions."
SGX is one of several companies bringing automation and quick turnaround to the problem of solving protein structures and testing small molecules for their drug potential. Others include its close neighbor Syrrx Inc., just a few miles away across Interstate 5, and Astex Technology Ltd. in Cambridge, England. All are making rapid progress, but SGX has something of an edge, according to analyst James Reddoch of Banc of America Securities. "Growing protein crystals has been more of an art form up until now, and Structural GenomiX has found a really fast, industrialized way to do it. And they have a privileged relationship with Argonne for their beamline facility."
SGX (www.stromix.com) was founded in 1998 by two of the brightest lights in protein crystallography, Wayne Hendrickson and Barry Honig, both of Columbia University. Funded by $7 million in seed
|Running the Cycle
|The route from gene to drug via high-throughput protein structure determination could ultimately yield thousands of structures a year.
capital, the company's initial mission was to solve novel protein structures as quickly as possible, creating a database to license to drug discovery companies. Two more rounds of venture funding brought total backing to $85 million. And SGX scored a major personnel coup late last year by recruiting star structural biologist Stephen Burley away from an endowed professorship at New York's Rockefeller University to be its chief scientific officer.
The business model made a solid database not just a piece of the company's infrastructure but a key product, and it presented an appealing challenge to Howard Hackworth, SGX senior database architect. "They said, 'You can build it the way you want it from the ground up,'" he explains. "Because we were going to be a database company, we needed all the data about experiments, good and bad. I tried to translate business rules into a database structure. It was very difficult to figure out what we were trying to keep track of. I wanted to steer clear of having it be an electronic lab book, and I just wanted to include the pertinent information."
One of the biggest initial challenges was getting the research staff to appreciate the importance of putting the pertinent data directly into the database, instead of letting it languish in a lab book, says Jon Christopher, crystallographic programmer-analyst and designer of the HTML interface that now makes the LIMS a favorite tool in the SGX laboratories. "We knew from the beginning that we had to do it, but it wasn't clear to everyone," he says. "For the people at the bench, it was just another thing they had to do. It was extra work without a reward."
Now, the database remembers all the "recipes" the researchers have used previously in the alchemistic art of trying to crystallize entities that don't naturally care to be crystallized. "It's easier to use the database now than to design the experiments by hand," Christopher says. "You just tell it what you want the results to be, and it will tell the robot how to make it for you."
Database development went fairly smoothly until a change in the business plan occurred. The idea of selling data gave way to a model in which SGX would collaborate with pharmaceutical companies, first solving structures for specific drug targets, then co-crystallizing and testing lead compounds to speed drug discovery. "The market for 'bio-information' had [decreased] because the industry was getting more sophisticated and more interested in doing the work itself," says CEO Tim Harris. "We could see that the database model wasn't going to fly. And we were getting more selective about the proteins we went after. We wanted to concentrate on the ones where we thought we could really make a difference, either on our own or with partners." In January the company announced its first two major drug discovery agreements, with Millennium Pharmaceuticals Inc. of Cambridge, Mass., and Aventis Pharmaceuticals Inc., the U.S. division of Frankfurt, Germany-based Aventis Pharma AG.
The database faced major upheaval. It had been designed assuming that SGX would handle every step of the research, from the initial PCR (polymerase chain reaction) through production of the protein crystal to the annotation of the structure. Suddenly the company's partners were providing samples that had already gone through part of the process, and there was no way to enter them into the database.
"It wasn't designed so that you could start in the middle," Hackworth remembers. "I got a call from the lab: 'I can't put information in the database.' OK! Emergency meeting! The database changed pretty dramatically after that. That's why we didn't go with a third-party product."
The system designers seek to remove as much human intervention as possible — but not before they fully understand what they're trying to automate. "We have a philosophy of modular automation," says Christopher. "First we implement a process, then we figure out how to automate it. We don't want to design a great big thing up front."
The Argonne facility is connected by a T1 line to SGX' main computer system in San Diego, and looks like just another node on the network. "We can call up a record of what crystals we have, then control the instrumentation at [Argonne] from San Diego," Christopher says.
The system also minimizes data entry. "When I started [at the beginning of 2002], the database
|Structural GenomiX' LIMS
|Hardware, software, staff and more.
didn't have the capability to enter codes into the [beamline] data collection system, so I entered them manually," explains Kanagalaghatta R. Rajashankar, a senior scientist who works at Argonne. "They were complicated: capitals, lower case, numbers, and letters. They [were time-consuming] to enter, and there was plenty of possibility for human error. Now the system files all the names in a table, and you just have to click on them."
The process promises to become even more automated when the beamline acquires a system for changing the crystals automatically during a series of studies. "We'll have a system in-house by the end of the year," D'Amico says. "By this time next year, we hope to be able to put a tray of samples in and go away for the weekend."
Within the next year, Harris expects to attract additional partners, close another round of private financing, and move quickly toward developing a complete in-house drug discovery capability. The company was scheduled to open a large chemistry facility at its San Diego campus. "We've already gotten insights into how particular compounds bind to particular kinases, and we have a clear idea of which compounds we want to chase down."
Elizabeth Gardner is a Chicago-area writer. E-mail her at firstname.lastname@example.org.