After months of missed deadlines, the US Food and Drug Administration this March finally released its Guidance for Industry on when and how drug companies should submit genomics data, and how the agency will evaluate it.
The halting process of composing the guidelines – originally planned for launch in summer 2003 – mirrors the slow realization of pharmacogenomics' much-hyped promise. Understanding the genetics of drug response and toxicity can guide the development of effective, targeted drugs; witness Herceptin, Gleevec, and Erbitux. The FDA recently approved the first automated lab test allowing doctors to test a patient's cytochrome P450 genotype to determine the optimum dosage of certain medications.
Yet the field still is far from the ideal of personalized medicine for all, largely because it's producing massive amounts of information that scientists are still struggling to interpret. Companies have been wary of releasing genomics data, given what it might reveal about...
HUNDREDS OF MILLIONS IN SAVINGS
Although the terms are sometimes used interchangeably, pharmacogenetics is the study of how inherited DNA variations, typically in just a few genes, affect drug metabolism or toxicity. Pharmacogenomics is a broader term encompassing all the technologies used in high throughput screening.
Current, yet controversial, estimates peg the cost of bringing a new drug to market at around $880 million over 15 years. Whatever the price, pharmacogenetics could help bring it down. According to a 2001 report by The Boston Consulting Group, genetics could save drug companies as much as $420 million and 1.6 years of development time.
One key benefit of pharmacogenomics is that it allows researchers to start with human populations, "as opposed to starting with data from animal models that come from directed perturbations like a gene knockout or overexpression that doesn't always replicate what you're trying to do with human populations," Martin says.
Roche uses pharmacogenomics in the very beginning of drug discovery to identify drug targets by conducting case-control association studies, where researchers bin a population otherwise very similar in terms of age, ethnicity, gender, and so on, into two pools, one that has a disease and one that does not.
"For instance, we are using case-control studies with Partners HealthCare in Cambridge, Mass., to bring together networks from Massachusetts General Hospital and Brigham and Women's Hospital, and identified roughly 2,000 people, about half with type 2 diabetes," Martin explains. He and his colleagues are now looking for single nucleotide polymorphisms (SNPs) in genes associated with the disease. "We put all that information together to see if they tend to cluster in certain pathways or operate in certain tissues. This can break a complex disease like diabetes into more discrete subphenotypes like pancreatic beta-cell defects or peripheral insulin resistance. And then you can begin to ask about tractable drug targets."
Pharmaceutical and biotech companies often work with genetic research firms such as Perlegen and deCODE for whole genome analysis. Perlegen partners with Pfizer, Johnson & Johnson, AstraZeneca, Bristol-Myers Squibb, GlaxoSmithKline, and Eli Lilly and Co., while deCODE collaborates with Roche, Wyeth, and Pharmacia.
"We look across millions of SNPs across the whole genome for associations with differential drug response," says Perlegen vice president of alliance management Paul Cusenza. "We might find 20 different genes that each have a small effect towards differential drug response, but collectively, when analyzed with a diagnostic, might drop the percentage of adverse events from 4% to 2%, so 90% of people can use it instead of 80%."
A LIMITED MARKET?
One of pharmacogenomics' pitfalls lies in its nature of reaching only a limited segment of a patient population. "If a drug only applies to 10% of a population, your market is vastly restricted," says Gregory Petsko of Brandeis University in Waltham, Mass. "To some extent, this is not as bad as it sounds. That 10% is virtually guaranteed to respond to the drug, so you have a very good chance of being the treatment of choice, and set your price to cover some of the development costs." Conducting trials with smaller groups of patients more likely to respond to a drug could also cut costs, he adds.
Pharma might be leery of pharmacogenomics because "pharma is so big now, it needs huge returns to cover gigantic operating costs and debt from mergers, so they need returns that are historically unprecedented. But there are no end of biotech companies that would love a return of $300 million to $400 million a year, and just step into that breach," Petsko said.
Another fear is that the targeted approach which pharmacogenomics takes "can limit your target population, so you might miss other applications, limit the marketing potential for the drug," said Marie Vodicka, senior director of biologics and biotechnology at the industry group Pharmaceutical Research and Manufacturers of America. "You have to weigh that against the benefit of having much greater efficacy or safety. So approval might be faster and easier, but for a smaller market."
With exploratory analysis for markers and trends, pharmacogenomics will also incrementally add work to clinical trials. While this will translate to increased costs and time, "if you look at the portfolio of compounds as a whole, the biggest issue is the attrition rate. If you could improve the fact that only about 10% of compounds going into Phase I make it to approval, you could bring more to market, add value to the entire portfolio. That's a significant advantage," explains Nicholas Dracopoli, Bristol-Myers Squibb vice president of clinical discovery technologies.
WHO GETS HIRED
Companies look for a combination of three sets of skills in pharmacogenomics – knowledge of the drug development process in the pharmaceutical and biotech industries, a background in human genetics and molecular and cellular biology related to human disease, and an experience with diagnostics, says Paul Waring, senior director of pathology and diagnostics at Genentech. "Few people have a combination of all three, so there is a great opportunity for those who do to advance quickly in their career."
Computer and statistics skills also are essential. "We're sifting through oceans of data – on SNPs or proteins or metabolites – to find patterns to begin projects. That takes people who are comfortable with large data sets who can recognize a certain kind of distribution and know which statistical methods use to pull out the pattern," Roche's Martin says. "So we're not just looking for people who have a background in the statistics we all had in skills, but things like data mining and study design, not only data analysis."
"We are looking for people who can both understand and straddle biology and computer science," says Perlegen vice president of human resources Karen Haynes.
Martin says he also looks for communication skills in prospective pharmacogenomics researchers. "It's a multidisciplinary field, and you're collaborating with a lot of different project teams in the organization. So getting your message across to a lot of different people is really very important. And in a fast-paced environment like we have with pharmacogenomics, you really have to be able to describe pretty complex ideas in a graspable way fairly quickly. You might only have 10 slides to get that across."
Given how pharmacogenomics is an emerging field, "we look for an aptitude for continuous learning, and for contributing innovation from their own discipline, be it biology or computer science," Haynes says.
The ability to make quick decisions also is key, says Martin. "We can't be too wedded to any one approach or pathway. You have to be ready to drop something if it just doesn't work, and not be doctrinaire. We need people who can pick up new skills on the fly."
"Those in drug development contemplating a career in this area, might choose to move to diagnostics for two or three years, and then combine the drug and diagnostic processes," Genentech's Waring says.
A pharmacogenomics researcher can find a home anywhere, Martin asserts. "These are people who can think on their feet, who are comfortable with high-resolution questions about individual objects such as genes or SNPs, and can therefore communicate with disease experts, but are pretty adaptable and can think in terms of distributions of data, designing studies without consideration of the underlying objects of study, of patterns in the data as a whole. You want that flexibility. These are people you want to have around."