Call it the Poste genomic era. Credited with kicking off the pharmacogenomics age with the first genomics database subscription and collaboration — a $125 million deal in 1993 between SmithKline Beecham and Human Genome Sciences — George Poste admits that colleagues thought he was nuts. Since then, pharmas have joined the frenzy, but their efforts could be charitably described as works in progress. Critics say that's a plain sign that pharmacogenomics will never truly get off the ground.
This comes as no surprise to the former president of R&D and chief science and technology officer at SmithKline. "This is far more complex than putting a person on the moon. Then, all the pieces were available, and it was just a question of putting it in the right sequence. All the technology was proven," Poste says. "What we're trying to do is like trying to go to the moon without understanding gravitational fields and planetary orbits."
Indeed. Pharmacogenomics is more than just plugging in your sequencers and genotyping instruments — it means rethinking the discovery and development steps that pharmas have fine-tuned for more than 100 years. "They should be looking at how to reinvent the process and not be hung up on … saving the process," says Michael Liebman, director of computational biology at the University of Pennsylvania Cancer Center, who previously worked for pharmas Roche and Wyeth. The mission: retooling the way pharmas operate with strategically placed bioinformatics, genomics, and proteomics technologies. McKinsey consultants estimate an investment ranging from $100 million to $300 million each year. But pharmacogenomics advocates say this will result in higher-quality targets and an understanding of how they function in various populations. The ultimate goal is what Poste calls "predict-and-prevent" medicine.
Recent challenges to the tried-and-true ways of drug development have rendered pharmas more amenable to change. These include a stunning target drought — PricewaterhouseCoopers measured the reservoirs at 500 targets across the entire industry, while outsiders proclaimed that their beloved genome sequence could yield a dam-bursting 25,000 — coupled with rapidly rising costs of drug development (and not-so-rapidly rising drug sales).
Despite the hefty price tags and largely unproven technologies of SNP typing and population studies, Boston Consulting Group vice president Peter Tollman points out, pharmacogenomics looked pretty good. After all, it promised targets already linked to human disease and immediate knowledge of disease or target frequency in a population. And if the gamble paid off, pharmacogenomics could save anywhere from $130 million to $420 million per drug with smaller clinical trials and potentially shorter lead time.
So pharmas have eagerly (according to them) or reluctantly (according to skeptics) cobbled together the pieces of pharmacogenomics. In fact, they've done so much that Poste declares, "Genomics has totally revolutionized discovery and the validation of targets for discovery."
But has it? The patchy way most companies add technology suggests a series of hastily fought skirmishes — hardly a revolution. Many seem to have been badgered into it: "They're investing more out of fear of lockout [and] fear that their competitors could develop a significant patent estate," Tollman says. And it shows. Pharma's efforts to assimilate pharmacogenomics, though noteworthy, are disjointed and not clearly planned out. Talk to the head honchos at these companies, and you're unlikely to find two compatible definitions of pharmacogenomics, let alone strategies to accomplish it.
With conditions like these piggybacked on all the problems inherent with ever-evolving genomics technologies, it's no wonder that pharmas are nervous about bringing the first fruits of their pharmacogenomics research to the FDA in upcoming years. (The first drug that could be considered pharmacogenomic for its targeted population, Herceptin, was developed through standard processes, and only after approval was gene-based testing introduced.)
New technologies mean new data, and new data mean new risks during approval. Larry Lesko, a director at FDA's Center for Drug Evaluation and Research, says his side is concerned, too. "The agency does not have a lot of experience with [pharmacogenomic] data," he says. "The interpretation of the information is going to be challenging."
There is hope. FDA and PhRMA are cosponsoring a workshop this month at the University of Maryland for scientists and clinicians dealing with pharmacogenomics or regulations. The goal is to hash out potential problems about what kinds of data will have to be submitted to get new therapeutics to market. Arrays in particular have been a bone of contention so far, and pharma reps are eager to find out just how far back into research the FDA will dig. That'll provide an indication of where it's safe to use new technologies and where to stick with old standbys.
Forward and Upward
The real impact of pharmacogenomics won't be felt, some say, until it's brought to the very front of discovery. Most pharmas now hold off on pricey pharmacogenomic treatment — it's more expensive than high-throughput screening, according to Liebman — until clinical trial design, when they've had the target in hand for several years.
"They're putting it after the fact," Liebman says. He argues that stratifying the populations and diseases should happen in discovery, leading to higher-quality targets. "If they put it up front, it will better guide what they will do."
Not everyone agrees. Dalia Cohen, global head of functional genomics at Novartis, says it's "much, much too early" to bring pharmacogenomics that far upstream. "If you identify a target, you have to learn a lot about it before you can start to determine which disease populations could be addressed," she says. But working pharmacogenomics into development has had its own effect on discovery: as Novartis feeds the results of its clinical studies back to its research team, "we are basically mining the data to determine whether or not [it] can lead to the discovery of new targets," Cohen adds.
Lee Babiss, vice president of preclinical R&D at Hoffman-La Roche, says his company starts singling out promising targets — ones that seem particularly linked to genetic variation — for the pharmacogenomics track about 18 to 24 months before clinical trials. The process starts with a comprehensive literature search to find out everything that's known about the target "with respect to genetics and allelic variation and genetics in other species," he says. Then Roche launches in-house projects to study related SNPs and check the target's molecular pathway and surrounding proteins.
While some envision pharmacogenomics eventually applied to every target, Nicholas Dracopoli, executive director of clinical discovery technologies at Bristol-Myers Squibb, doesn't see the point. "It's safe to say that pharmacogenomics approaches are now considered on everything that's being developed," he says of BMS's strategy, "but in many cases you don't really need it."
Using pharmacogenomics for statins, for instance, would be a waste, Dracopoli says. "They basically work pretty well in everybody, and you have an immediate test of how well it's working — you measure the lipid level." But diseases such as cancer, where prescribing drugs by trial and error greatly increases the patient's risk, will get high priority for pharmacogenomics development.
As the development side gears up its pharmacogenomics efforts, discovery teams are bracing their side with various genomics technologies.
Most pharmas start with partnerships before breaking out the checkbook and bringing in new tools. Stephen Faraci, assistant director of Pfizer's genomics heavyweight Discovery Technology Center, says, "You have upfront collaborations [to make] sure it works before you bring it inside. This isn't a quick hit. Something like genomics, it's going to take a few years" to choose a tool, bring it inside, and get it up and running.
A couple of years' time is not uncommon to get a new technology settled. Dan Burns, vice president of discovery genetics at GlaxoSmithKline, says when the company bought its supply of ABI Prism 3700s, it took nine to 12 months from the first sequencer's arrival to the day everyone was trained, the informatics in place, and the instruments performing reliably. And that's with ABI, which "has the most experience" at putting genomics technologies in pharma, Burns points out.
In the time it takes to set up an instrument, no doubt there'll be advances in the field. The market "is so young and evolving so quickly, you run a risk by tying yourself to one technology," Burns says. Companies will have to "stay fast, loose, and nimble." Not exactly the image of big pharma.
The continually changing platforms have deluged the market, Poste says. "How do you choose between 27 different array companies or 50 proteomics companies when they're all claiming to be doing the same thing?"
Obsolescence is a tremendous concern, according to Patrice Milos, who became pharmacogenomics manager when Pfizer created the position in 1996. "We're not settling on … one technology." She rattles off some of the company's assets: an alliance with Genaissance for gene resequencing and finding SNP markers; internal sequencing with the ABI 3700; SNP discovery with the Transgenomic Wave System and with Sequenom's mass spec; allele frequency in pooled samples with Pyrosequencing technology; and proprietary fluorescence-based robots for some routine analysis.
In addition to staying current, pharmas bite their nails over compatibility — among technology platforms as well as data. Roche, for example, has an extensive chip-supply deal with Affymetrix. Though Babiss liked what he saw of CuraGen's gene profiling technology, the data it produces isn't easily reconcilable with data from Affy's chips, so it's unlikely he'll use it. "I don't know which is right or wrong," Babiss says, "but we have to be consistent."
Outsourcing allows pharmas to ease up on resources dedicated to each technology. Babiss is preparing to send SNP discovery outside, but he'll have to keep resource-intensive proteomics work, including 2D gel and mass spec development, in-house. "The assays [for SNP discovery] are fairly simple and robust and very amenable to outsourcing," he explains. "On the proteomics side, the technology is still being developed."
Cost, especially in genotyping, is yet another consideration. Though it's reasonable to look at SNPs on certain candidate genes, "if you're routinely genome-scanning tens of thousands of SNPs in an individual," BMS's Dracopoli says, "that's not feasible under current budgets." Poste believes successful assimilation will depend on costs falling from the current dollar or so per genotype to a tenth of a cent.
…Or Pipe Dream?
Detractors say the field's ever-shifting bleeding edge means technology is being introduced off the cuff, with no big-picture plan. "On the surface, it might look a bit uncoordinated," Babiss acknowledges.
This gives fuel to those who believe pharmacogenomics will never really be assimilated into discovery and development. William Haseltine, CEO of Human Genome Sciences, considers the must-have technologies so many toys to placate scientists. "You're not seeing those tools being seriously integrated into big pharmaceutical companies," he says. He argues that the premise of personalized medicine is ludicrous: pharmas "can barely develop one drug for one disease," he scoffs. "The idea that you can rescue a drug that has adverse effects on a subset of the population is fanciful."
But proponents are eager to stamp out the notion that status quo is the way to go. Babiss points to the number of drugs withdrawn from market and to a JAMA study showing that most drugs are just 30 to 50 percent effective, and dismisses doomsday fears of market fragmentation. "If you look at any particular drug that a manufacturer has, the amount of the market that they have for that disease is relatively small," he says. With pharmacogenomics-based drugs, however, "you'll capture much more of [the market] because you'll have much greater efficacy and you can charge a premium for it."
The Great Beyond
With the yardstick that this industry uses to measure success — getting a drug to market — insiders generally concur that Glaxo will be the first to bring a pharmacogenomics-based therapeutic to the FDA.
"They've done more to push this forward, they've spent more money on it," says Alan Williamson, who launched the SNP Consortium.
It comes closest, observers say, to meshing pharmacogenomics into the traditional process. "We have both approaches going," says Wayne Anderson, Glaxo's director of exploratory clinical target genetics. Scientists are mining the genome database and using classic techniques to find better targets and susceptible genes. "We then start to apply pharmacogenomics all the way through [development], from early Phase I now through first time in man," he adds.
"Gene expression arrays to SNP genotyping — you name the technique and we've probably done it, are doing it, or are partnering [to do it]," Glaxo's Burns says. Among the technologies: 3700s, Affy chips, and genotyping with Luminex and ABI.
A good chunk of its investment went to bioinformatics, which Burns calls the "linchpin" to everything else. With a genetics team of about 600 and an estimated 150 to 200 in bioinformatics, Glaxo's commitment is significant. (Another major pharma counted its bioinformatics group at around 40 or 60.)
At least some of Glaxo's headway is due to the lead time it got from Poste's groundbreaking HGS deal.
Now CEO of consulting firm Health Technology Networks, through which he counsels six pharmas, Poste pounced on genomics before most people saw it coming. "George was one of the original thought leaders in this area," says Viaken Systems' CSO Richard Hamer, who used to work for Aventis.
At home in Scottsdale, Ariz., Poste spends his little free time driving through the desert in his silver Hummer. Looking back at his decision to make the HGS deal, he says it was no great leap. "It was so obvious that Craig Venter's work with ESTs was going to give us the first major insights into genome organization … and dramatically expand targets," he says.
After the $125 million SmithKline-HGS deal, other companies came around. The next year saw a $70 million genomics collaboration between Millennium Pharmaceuticals and Roche. By 1996, Poste arranged for four pharmas to join the existing HGS deal: Merck, Schering-Plough, Synthelabo, and Takeda Chemical Industries.
"Any time you have advanced technologies that radically alter an industry, you'll have companies who see it and embrace it, other companies that eventually realize it's a competitive advantage and embrace it, and the third group will wonder what the hell happened," says Poste, who left SmithKline in 1999. He acknowledges the good argument for not being a pioneer — avoiding early risk — but says anyone coming in after the first couple of waves will miss out entirely.
(Debate rages about which companies rank where. Hamer says the first screen is who joined the SNP Consortium: "those are all the real players." An informal survey suggests that Glaxo leads the pack, followed by Pfizer, AstraZeneca, and Bristol-Myers Squibb; insiders place Roche and Merck, not a consortium member, at the end of any list of companies with viable programs.)
"A number of companies that were late into genomics were basically taken to the cleaners," Poste adds, citing $450 million deals, such as Bayer's contract with Millennium, in which pharmas emerged with the promise of just a handful of targets.
To Market, To Market
Glaxo's aggressive push might win it the booby prize in this race: the onus of getting the first purely pharmacogenomics-based drug through FDA approval. Williamson says, "[Pharmas] can't afford to be anything other than thorough."
Or smart. Worried about showing the FDA data from unfamiliar sources, such as gene arrays, some pharmas are hoping to use genomics technologies and then pull the same results out of traditional techniques. "We wouldn't just take an array experiment with 20,000 or 30,000 genes and dump all that data with the regulatory authorities," Dracopoli says. "We would use that data to identify the set of genes that we believe to be predictive" — and, if possible, convert those genes to a standard assay the FDA has seen before.
There's good reason for keeping arrays in the background. Pharmas "may find additional things that they don't know how to interpret yet that may come back later and be significant," explains Liebman. "It makes the generalized use of the arrays in an FDA submission a risk that has to be considered. If you can come up with a specialized array, you're going to be better off."
FDA's Lesko encourages pharmas to fork over pharmacogenomics data now to give the agency a chance to get used to it. Other possible implications for pharmacogenomics-based drugs: testing the false predictive power of the accompanying molecular diagnostics; examining the drug in nonresponders to make sure it doesn't harm people who take it without proper testing; and making sure genetic markers are prevalent enough in the population that it's worth releasing the drug.
So far, the FDA has issued no guidelines, but the upcoming workshop may shed light on what it expects. Meanwhile, Lesko is planning a course for reviewers at CDER. Speakers such as Glaxo's Allen Roses and Carol Reed of Genaissance Pharmaceuticals "will bring everybody up to speed," Lesko hopes.
In all likelihood, Poste says, the FDA won't see a full submission until 2005. (Simple math: the earliest alliance in 1993 plus the shortest average development time, 12 years.) Faraci's estimate is a more conservative 2010 for "the earliest possible."
That leaves plenty of time for the industry to come to terms with pharmacogenomics. "In the early '80s people were having these discussions: 'Boy, do we need to bring in molecular biology?' Now it's an integral part of every point of R&D," says Glaxo's Burns. "Ten or 15 years from now, pharmacogenomics ... will be at the same point molecular biology is now."