High-density wafers at the ready, Perlegen takes whole-genome scanning by storm
By Meredith W. Salisbury
When Greg Brandeau joined Perlegen Sciences this past March, one of his chief concerns was about recruitment. How would he convince people to come to a six-month-old company with unproven technology in an already crowded SNP-finding market? Turns out, it hasn’t been a problem, says Brandeau. “Lots of people hear about our story and say, ‘Oh, hey, I want to come do that.’”
And what a story it is. Perlegen, the Santa Clara, Calif.-based Affymetrix spinoff with a staff of 50 and $100 million in private funding, has captured the industry’s attention. Led by the unlikely triumvirate of a shrimp farmer, computer animation exec, and med school prof, the company says it has made possible what everyone’s been drooling for: whole genome scanning on a reasonable budget. Perlegen’s got exclusive access to Affy’s next-generation wafer (think chips on steroids) and plans to scan 50 human genomes at single-base resolution in the next year or so, with an eye toward making its mark in association studies.
“We have developed a new sort of microscope, if you will,” Brandeau says. “And we’re going to look at information that no one has been able to see before.”
But confidence has never guaranteed success. Perlegen’s up against competitors who have more than just funding — they already have customers and revenue. If nothing else, Perlegen’s ready to give them a run for their money.
Shoot for the moon
Perlegen got its start as a think-tank experiment at Affymetrix, when CEO Stephen Fodor set his sights on improving his company’s gene-chip arrays. Affy’s traditional chips are made by dicing up a five-inch-square wafer into 50, 100, or even more pieces with up to 400,000 probes on each. What if the wafers were left whole? What if the probes could be made even smaller? What if cost weren’t an issue, and you could just invent the best DNA plate possible?
“His dream was to get enough DNA onto these wafers that you could get a whole genome on to scan all the bases,” says Perlegen CEO Brad Margus. So in January 2000, Fodor recruited David Cox, medical professor and co-director of the Stanford Genome Center, to take a sabbatical and see what he could do with what was known in-house as the genome scanning project. It wasn’t just a matter of squishing more DNA on a substrate. “If you make the wafers with these very small features, you still have to have a way to scan them,” Margus says. “And you have to deal with sample preparation because there’s so much more going on.”
By October 2000, pie-in-the-sky met reality. The technical kinks were worked out, and Affy found itself the proud parent of a wafer carrying 60 million probes. For less than $2 million (including wafers, reagents, and labor), scientists could thoroughly analyze 300 wafers containing an entire human genome — and presumably find all the SNPs and haplotypes within it.
Reducing the problem
The theory is simple. With an established genome sequence to work from, the wafer technology allows users to compare other, unsequenced genomes base by base, finding all of the differences among them. (Perlegen uses the public sequence, and isn’t worried about possible errors. “If you’re cruising along and you come to a T in the human genome project reference and that was a mistake, it doesn’t bother us at all because we’re going to look at 50 more [genomes],” Margus says. “The template doesn’t have to be perfect.”) The resulting catalog of SNPs found across, say, 50 genomes, could be the definitive guide to genomic variations — and the importance of that isn’t lost on the Perlegen crew.
After finding what could be eight or 10 million SNPs, they’ll focus on the more common ones — occuring in more than 10 percent of the population. That ought to narrow the field to about four million SNPs. Those, in turn, will be sifted through to find the relevant haplotypes, or blocks of SNPs. “If you knew the right blocks,” Margus says, “you could [look for] two SNPs instead of 15.” So if you find SNPs 1 and 2 together, you could accurately predict that the genome in question also has SNPs 3, 4, and 5. By choosing SNPs representative of each haplotype, Perlegen expects to narrow the search field to just 300,000 or 400,000 SNPs that will give a reliable interpretation of all the important variations in each genome.
It certainly beats out the established candidate gene approach, in which scientists have to choose a few genes of interest and then poke and prod to see what they come up with. That method “requires an up-front hypothesis, doesn’t consider new genes that may not yet be discovered, and completely ignores the 97 percent of the genome that is not represented by genes,” says Thane Kreiner, vice president of corporate affairs for Affymetrix.
“It’s very frustrating,” says Margus. “Pharmas can spend millions or billions and not hit the genes that cause the disease.”
Hold on or let go?
Promising as the technology was, the wafer left Affy in a tricky situation. The company had assured analysts it would reach profitability by 2002, and pursuing the genome-scanning project would require unexpected millions of dollars. “I don’t think Affymetrix wanted to readjust Wall Street’s expectations and say they wouldn’t be profitable for another four or five years,” Margus says.
Another consideration was sowing the seeds for future revenue. Michael King, an analyst with Robertson Stephens, points out that the service and product business is a tough way to make money. “In the world of genomics, the way to great riches is through an ownership interest or a royalty stream,” he says.
Execs debated setting up a tracking stock, Margus adds, but those hadn’t seen much approval among investors. So Affy’s solution was to spin out Perlegen, which comes from the Latin “perlego” — “to scan.”
The spinoff got a three-year exclusive license to the wafers. And by April, it had rounded up $100 million in private financing, including money from Affy founder Alejandro Zaffaroni. The parent company still owns about 53 percent of Perlegen.
Kreiner calls it a “win-win” structure. “Affymetrix recognizes further economies of scale in its manufacturing by selling wafers to Perlegen, and also gains access to the patterns of SNPs [found],” he says. Affy expects to be able to design better-targeted chips within a few years by using the haplotype patterns discovered.
‘Fire hose of data’
One of Perlegen’s earliest victories came in the person of David Cox, who surprised the industry by changing his Stanford sabbatical to an indefinite leave of absence so he could stay on as the company’s scientific director. CEO Brad Margus left his frozen food company and came aboard with a compelling story of his own (see sidebar), while CIO Greg Brandeau added his years of IT experience at Pixar, the animation studio responsible for movies such as A Bug’s Life and Toy Story 2.
Meanwhile, the technology was making headlines of its own. “We figure one person in our lab can do, in one day, three wafers,” Margus says. That person “can read as many bases of DNA as over 100 [ABI Prism 3700s] can run in 24 hours.” They call it Perlegen’s “fire hose of data.”
Postdocs emerging from DNA studies have probably looked at 20 kilobases in three years, Margus estimates. “In their first week working for us, we can give them three million bases north of that and three million bases south of that region to look at.”
But excited employees could only take Perlegen so far. To really get attention, they would have to produce serious data. To that end Perlegen chose to focus on chromosome 21, scanning it across 20 individuals and submitting the data for publication. Margus and others have already shown that data to some industry VIPs, including NHGRI director Francis Collins, who has indicated that he’s considering Perlegen as a partner to build a haplotype map.
Out of the woodwork
With publication in the works, Perlegen plows ahead. But it’s far from alone in its quest. Known competitors Orchid, Pyrosequencing, and Sequenom already have proven SNP-finding and -scoring technologies, and Genaissance and Variagenics have been in the hap business for a couple of years now. Orchid’s pharma customers include Bristol-Myers Squibb and AstraZeneca, while Pyrosequencing has sold more than 80 of its systems and pulled in $4.9 million last year. Perlegen, on the other hand, is still finalizing a partnership that was expected to be announced back in April.
The formidable Celera is eyeing the field too. The company boasts that it’s already located more than two million SNPs, and recently announced a plan to sequence 50 more human genomes to locate disease-related SNPs and haplotypes. Celera president Craig Venter says the major difference is that his company will do it better by focusing on SNPs in coding regions instead of throughout the genome. “Most of the SNPs associated with human disease are found in the first intron,” Venter says.
Well, Margus argues, then why did Celera sequence the whole genome in the first place? With Celera’s more expensive process, he believes that cost is the main factor for being so choosy about SNP location. “If you can’t afford to do the whole genome it would make sense that you start by focusing on the genes,” Margus says. But with haplotypes that stretch across genes and possibly functional conserved regions embedded in junk DNA, “clearly the whole genome is critical.”
Despite Venter’s insistence, the industry isn’t quite ready to back the genes-only plan as superior. Perlegen’s wafer-scanning promises to be cheaper than Celera’s capillary-gel-based sequencing, and many people are concentrating on price. “How is Celera going to compete on a cost per genotype?” says Robertson Stephens’ King. If the cost were feasible without chips, he reasons, “everyone would be doing it on gels.” Also, while Perlegen has shown that haplotypes can be found using ordered chips, there’s no similar argument for the ease of discovering haplotypes with gels.
King contends that Celera’s head start lies in its sequence data, which is arguably a bit more complete than the public project’s. “Whatever advantage Celera might have had by having the complete database, it won’t really matter,” he says. Celera still has to overcome the slower processing of gels, and King believes that will put the companies on an even playing field — or give the edge to Perlegen.
The real paydirt
That edge may be enough. “The real objective for Perlegen is just to have a head start,” Margus says. By having exclusive access to the wafers for three years, scientists should be able to scan the genomes and have SNP and haplotype data in hand well before anyone else can even use the technology.
Armed with the haplotypes, Perlegen believes it could trump just about any other association study — from other genomics companies or pharmas. “If you want to look at four million SNPs and 6,000 people, that comes to a billion-dollar cost for an association study,” Margus says. “And that’s just too much.” Having narrowed the field to a few hundred thousand SNPs, the Perlegen crew should be able to do association studies for less money and in less time.
“Our value is doing those association studies,” Margus says. He has no interest in developing a database business. Instead, Perlegen will seek partners among pharmas, research institutes, and other companies for the studies. And eventually, the company will venture into its own discovery. “We’ll do much more of the downstream biology ourselves once we find the targets,” Margus adds. “But we definitely have a long way to go to build the expertise for that.”
With reporting in San Francisco by Jackie Cohen
It’s not business, it’s personal
Perlegen CEO Brad Margus “has a sense of urgency,” says Anne Bowdidge, director of investor and public relations at Affymetrix — and it has nothing to do with his company’s race against Celera. For Margus, the personal implications of whole-genome scanning and finding haplotypes far outweigh the professional.
Margus, former president of a Florida frozen shrimp company, learned in 1993 that two of his children have the rare, incurable disease ataxia-telangiectasia. A-T combines neurodegeneration with cancer and immune deficiency. Margus hired molecular genetics scholars to tutor him and formed a nonprofit research organization that raised $7 million for research into the mutated gene that causes A-T. Finally, he left his company to join biotech.
His dedication ultimately won the attention of Perlegen’s scientific director, David Cox, who left Stanford University School of Medicine to take his first private-sector job at the company. It was Cox who recruited Margus. “We’d like to see more genomics companies apply their work to real things rather than just hope,” Cox says. “With Brad as CEO, not many hours go by each day without being reminded of that.”
Despite the triumphs there is still no cure for A-T, which has crippled the Margus boys, Quinn and Jarrett, ages 10 and 11.
— Jackie Cohen
Too much information
Perlegen’s high-throughput approach has taxed its computer systems from the start. The major bottleneck is storage, says CIO Greg Brandeau. He estimates the company will generate “half a petabyte of information in the next 18 months or so.” To deal with this, Brandeau went shopping, and came back with 15 terabytes of online storage and Sun’s nearline hierarchical management system — “a tape jukebox.” With nearline, Brandeau says, “You don’t get your files back as quickly, but you can actually afford to keep them.”
Another issue, of course, was network speed. Brandeau settled on equipment that offers gigabit Ethernet for Perlegen’s 100 to 200 CPUs. The processor power will eventually be upgraded, but that expansion can wait — at least for now.