Genotyping is going grand scale. "You're definitely seeing the emergence of industrialized genotyping facilities," says Jack Ball, senior vice president and general manager of Orchid BioSciences' life sciences group. "It's here now for sure."
What do you need to know before you set up a large-scale SNP shop? Those who've done it say that the main thing to remember is that the genotyping field is rapidly growing and far from mature. "We're orders of magnitude from where we want to be, but also orders of magnitude ahead of where we were a few years ago," says David Altshuler, director of medical and population genetics at Whitehead's Center for Genome Research.
Here's what the been-there, done-that experts have to say about putting together an industrial-scale genotyping facility.
Check Your Bank Account
The main considerations? "Money, money, and money," according to Ball. Jeff Olson, senior director of molecular biology at Variagenics, says it'll cost hundreds of thousands of dollars for a small facility doing medium throughput. For high throughput, plan on an investment in the millions.
According to CTO Jerry Vovis of Genaissance Pharmaceuticals, a rule of thumb is about $1 per genotype. With an estimated 3 million SNPs in the genome, that adds up. "Now that there is a need to do high-throughput, high-scale genotyping, cost must come down," he says. "I'm talking 10-fold, not 20 or 50 percent."
The tiniest things — oligos, in this case — are significant, says Scott Peterson, a faculty member at the Institute for Genomic Research who's been planning the genotyping component of TIGR's new Pathogen Functional Genomics Resource Center. Though prices have come down from about 50 cents a base five years ago, they're still around 18 cents a base now, Peterson says. "You end up using maybe as many as five different oligos to generate information about any one site."
Start Small, Stay Modular
The most widely agreed-upon tip for going industrial scale: don't.
At least, don't start that way. According to Michael Boyce-Jacino, Orchid's CTO, "One of the strategic issues is, do you design on paper a facility with a million-a-day genotyping and design each component of it and then go out and deal with that? Or do you design a subsection of it, say 100,000 a day, and then prove that, set it in place?" Orchid began with one instrument doing 25,000 genotypes per day. After adding instruments and scaling up production, the company reached 750,000 per day and expected to top the million mark at the beginning of this year.
Peterson has also been dealing with this. The TIGR center will have to start small because no one's sure how many people will actually need to do microbial genotyping. "We can easily assume a low-throughput demand," he says, "but we're also having to plan for the eventuality that there will be high-throughput demand."
Key to the scaling-up concept is a modular approach. Orchid added a few modules here, a few there. That enabled it to avoid building a dedicated facility. "We've been able to essentially move this facility a couple of times during its development," Boyce-Jacino says. "The modular approach has given us the ability to grow with the market."
Altshuler says you should keep things so modular that you don't even tie your LIMS to the technology. "Build it with the knowledge that you may want to rip it out in a couple of years."
Know Your Customers
Just who are your users? And what kind of studies will they want to do? If you can't answer these questions, you're likely to wind up with an expensive genotyping complex that serves few customers.
You might study "a limited number of loci to document many strains," Peterson says. Or you could be "looking at thousands of different sites in a few strains."
accuracy is another issue. Clients with genome-wide projects might not mind getting just, say, 85 percent of the SNPs in the samples, says Variagenics' Olson — but you can bet that won't work for a clinical trial.
And if pharma is your targeted market, you'll need to keep good manufacturing practice rules in mind for your program. "As the marketplace is moving to true clinical genotyping, it has to have very tightly controlled processes in place," Boyce-Jacino says.
But, Vovis warns, becoming compliant with these regulations means a much bigger investment than just setting up a factory. So keep outsourcing in mind: Variagenics sends its regulated clinical-trial studies to a CRO.
Never Settle on a Technology
Remember this about your facility: "It's never completed," Altshuler says. Not only do technologies evolve (or become obsolete) at a frightening rate, but your own technology needs will change as you cater to new customers.
"Our expectation is that, at least for the foreseeable future, there's no technology that will stay stable and will stay at the cutting edge of the field for a very long time," Altshuler adds.
For Orchid, this means using many platforms at once. For association studies with relatively few samples, Ball says Affymetrix chips work best. For medium throughput, he relies on Luminex bead technology, and Orchid uses a proprietary system for ultra-high-throughput demands.
Variagenics also uses several platforms — having two or three on hand virtually guarantees finding all of the SNPs, Olson says.
Mark Daly, a fellow in computational biology who's had a hand in setting up Whitehead's last two genotyping centers, says the center is testing several technologies, most based on mass spec. But chip-based technologies "may be coming on line to do genotyping soon," he says. Ideally, "you want to have a technology development end," Altshuler says — one that will keep an eye out for better multiplexing, DNA isolation or purification, and improved robotics. "Even if you're not inventing new technologies, you want to be constantly looking … and be in a position to change over."
Secure Your Assays
"The majority of the time," Olson says, "assay development takes more time and effort than running the samples." Moral: don't scrimp on the assay program.
As there's no public source of assay design information, having the unique set of primers to detect each SNP on hand is crucial. "That's the biggest strategic barrier stopping the mom-and-pop genotyping," Boyce-Jacino adds.
It's okay to be picky about your assays. For one thing, "you want to use a method where both a positive and a negative result will be inherent in the output," Peterson says. "So if there's ambiguity in either one, that raises a red flag." Also, the more sensitive the assay, the more amenable it is to multiplexing.
As for what you're looking for with the assays, Vovis says haplotypes are the way to go. "If you're doing individual SNPs, that's more theoretical than practical," he says. To be truly high-throughput, he argues, you've got to get to the hap level.
Automate and Simplify
Keep basic factory principles in mind. A one-way flow prevents confusion: bring the samples in at one end and get the data out at the other, Ball says. Orchid and Variagenics found it useful to take a large space and divide it into pre-PCR, PCR, and genotyping sections.
The fewer the steps, the lower the chance of error, Altshuler points out. "You want things that are simple, steps that can be automated by robots."
Peterson says it's the same for algorithms. You want the kind of assurance where "the computer can basically make sure all the data falls into the [right] category." The less human intervention, the better.
And don't forget to build in equipment redundancy, says Vovis, "so that you don't suffer downtime."
Go the Distance for Informatics
"You have to make a commitment to informatics and biocomputing when you're setting up a facility like this," Daly says. "There's no way you can short yourself on that."
Unfortunately, that doesn't mean buying the best bioinformatics solution for genotyping. "I don't think there's anything out there that strikes me as particularly brilliant," Peterson says.
Homegrown solutions abound. Whitehead has hired a large bioinformatics staff to build custom software, LIMS, and databases. TIGR, too, plans to build its own SNP database. And Vovis and Olson argue that there's just no topping an internally designed LIMS.
Orchid designed most of its informatics in-house, as well, though it relied on an outside firm for processes that might fall under GMP or CLIA regulation, including the LIMS. Having an internal bioinformatics team also enables Orchid to customize its work to interface with pharmaceutical companies' databases.