NEW YORK--No two pharmaceutical company bioinformatics directors are likely to give the same definition for the term pharmacogenomics, but all agree on one thing: it's going to make their jobs much more complicated. "I keep telling people that pharmacogenomics is what's going to keep me and a lot of people who are pretty young right now employed until retirement age," observed Douglas Bigwood, worldwide coordinator of bioinformatics at Bayer Pharmaceuticals in West Haven, Conn.
According to forward thinkers in drug discovery, a pharmacogenomics approach will enable drug companies to link data from genomic analysis, genotyping, and clinical trial studies to understand disease pathways and differential drug responses. SmithKline Beecham Chief Scientific Officer George Poste, widely known in the pharmaceutical industry as a champion for pharmacogenomics, has said the approach will enable pharmaceutical companies to create "the right drug for the right patient."
The possibilities pharmacogenomics holds for pharmaceutical companies are vast, including providing new drug targets, potentially recovering and reevaluating drugs that failed clinical trials, relabeling drugs for appropriate populations, and saving time and money by eliminating certain targets much earlier in the pipeline.
Michael Liebman, bioinformatics director at Wyeth-Ayerst Research in Radnor, Pa., explained that pharmacogenomics will "be used to analyze populations in advance of going into drug design so that early on you have the potential to identify what and how big the target population is for the drug you're designing or the target you're selecting. By bringing that information to bear early in the process you make the whole process much more efficient. You know up front what your intended customer base could be and you save things from going into trials, which are very expensive."
Rainer Fuchs, chief information officer at Ariad Pharmaceuticals's Hoechst-Ariad Genomics Center in Cambridge, Mass., explained one way in which pharmacogenomics may offer new drug targets. "You could find that 10 percent of a population of people who take a drug don't react to the treatment. You look at polymorphisms and find that there's a change somewhere in the gene that hasn't been implied in the disease process so far. That could now feed directly back into your drug discovery process and could give you a completely new drug target."
Success Hinges On Bioinformatics
The success of pharmaceutical companies in this area will hinge on bioinformatics, according to the experts. "It's not trivial to set up your system in a way that you can integrate clinical data, patient information, phenotypes, and so on with genetic information such as polymorphisms and gene sequences," Fuchs commented.
"The question is how do you bring that information into the global, earlier elements of drug design and discovery," added Liebman. The challenge for bioinformatics, he said, will be not just to recover compounds from clinical trial failure, but to integrate the whole information stream into the early stages of target identification and drug validation.
The trouble, as every pharmaceutical company bioinformaticist who talked to BioInform for this article remarked, is that clinical data have never been integrated with research data. "The major bioinformatics challenge is to get a handle on how to deal with clinical information and how to create the links between traditional bioinformatics systems and the corporate patient databases," observed Fuchs.
But technological challenges will be only half the battle, he continued. "The real problem for most companies is going to be more the organizational issues," said Fuchs. "Pharmacogenomics requires you to combine data that are normally generated and used in discovery research with data that are used at the other end of the spectrum in the clinical phase. On one hand, most bioinformatics people don't know anything about clinical data. But organizationally, at most companies these functions are completely separated."
Much Talk, Little Action
At the moment, most bioinformaticists admitted, there is more talk than action on the pharmacogenomics front. Only the most progressive companies are putting pharmacogenomics programs into place:
* Glaxo Wellcome's Research Triangle Park facility has spent the past year working toward getting into pharmacogenomics, according to Robin DeMent, US department head for bioinformatics.
* At SmithKline Beecham's research facility in Harlow, UK, Peter Goodfellow, director and senior vice-president of discovery worldwide, who designed Cambridge University's high-throughput genotyping and polymorphism detection facility, has established a genetic technologies department.
* Wyeth-Ayerst's Liebman said his bioinformatics team is working closely with the company's clinical trials group to begin assembling patient data, analyzing retrospectively where it may have been applied, and testing to find out ways that "we could use examples to enable us to go forward in a more proactive rather than reactive manner."
Liebman will also chair a conference in Washington in February that he said aims to get bioinformatics professionals thinking about medical informatics and bioinformatics integration.
Yet some in the field aren't sure pharmacogenomics will prove profitable. Said Fuchs, "What doesn't convince me so far is the economic analysis of this issue. How interesting is it to go for the subpopulation? I would argue that most of the big pharmaceutical companies would not be interested. Anything that would generate under $500 million a year is probably not really interesting. If you can go for only a subpopulation of 10 percent and you can get $50 million, that may not be interesting," he emphasized.
Others argued that the US Food and Drug Administration (FDA) will make debate over the economics of pharmacogenomics irrelevant. "Since the technology is advancing to the point where these populations will be identified and the information will be more readily accessible, FDA is going to start examining the integration of this information into the trials," predicted Liebman.
Indeed, FDA has already begun to study the issue. The Biotechnology Industry Organization, an industry trade association, has invited the entire FDA staff to a "training day" in Washington next month. There Poste and other pharmaceutical experts, including Glaxo Wellcome Chief Scientific Officer Allen Roses, will lead workshops explaining the science behind genomics in drug discovery.
Many bioinformaticists said they expect it's only a matter of time before they must comply with pharmacogenomics regulations from FDA. According to DeMent, "It would really surprise me if it wasn't within a year that FDA comes back and says okay, now we understand the science, we want to see how your computer systems handle it."
Hurdles To Overcome
Bioinformaticists who have begun to think about how they will integrate clinical and research data indicated four main hurdles they will face: scale, security, organizational issues, and regulations.
To begin with, DeMent said she expects to have to install much more powerful servers to handle increased data volumes. "There's going to be a huge surge in the volume of information to handle. Once you get into genotyping of multiple segments of DNA on multiple patients on multiple reactions with multiple drugs, you're talking about terabytes of information."
"We've just gotten to the point where we can handle hundreds of gigabytes," she noted.
Databases will also need to scale to meet increased demands. "Being able to track all these polymorphisms and hook all this in with metabolic pathway data will be a number-one challenge," said Bayer's Bigwood. "Most pharmaceutical companies are trying to build up databases of metabolic pathway information but they're highly incomplete. Metabolism is a very complex thing to start with, and then you have to start dealing with variations in all these genes that may have effects."
"With clinical trials and new drug applications, we're talking potentially hundreds of volumes of information," he added. "There can literally be up to a roomful of documentation of data."
Ken Fasman, director of research informatics at Astra Bioinformatics Center in Cambridge, Mass., said his company is exploring what it would take to do large-scale genotyping of patients involved in clinical trials. "The logistics of that, and the expense of that, are considerable," Fasman noted, adding that "integrating existing legacy systems for managing clinical trials with some of the newer things going on on the genomics side" is a concern. "It's going to be tricky," he admitted.
Cultural and organizational obstacles also hinder bioinformatics departments that aim to integrate global data.
"We will have to deal with data that are farther downstream than have been used up to now," said Bigwood. "Infrastructure issues are going to be important for feeding those data back into the process. There's also a requirement for getting the data up front for the discovery process because the number of polymorphisms in individuals is enormous. To be able to sample enough patients to get a handle on what the differences are among individuals can and will be very difficult in many areas," he concluded.
Added Fuchs, "It doesn't help that many pharmaceutical companies don't have centralized clinical trial databases and trials are often run in different countries. I don't think it's scientifically a particularly challenging problem, but it's a quite significant implementation issue."
Fuchs observed that getting different cultures within pharmaceutical companies to speak the same language will be another challenge. "At most companies, people in research hardly ever talk to people in the clinic, and they do speak different languages," he explained. "Getting these people together from a pure organizational point of view is a huge undertaking."
"Most of the clinical organizations in a company have huge informatics groups of hundreds of people or more," Fuchs continued. "But again, these folks speak a completely different language than the people in bioinformatics."
Clinical informaticists also use different database systems and face restrictions such as FDA requirements for validated computer systems and standard operating procedures for analyzing data that bioinformaticists haven't had to deal with, Fuchs noted.
But bioinformatics departments may soon begin to experience the rigors of FDA compliance. DeMent explained, "All the genotypes eventually used for regulatory submissions have to be generated under FDA's Good Laboratory Practices (GLP) regulations. In early research they don't have those kinds of restrictions, but once you've got something that you're putting into animal models or early clinical trials, there are very strong restrictions."
She added, "Researchers are struggling now to make the process they use, say, for genotyping comply and still generate the same quality of data and flexibility for research. Of course that puts restrictions on bioinformatics because we're building tools that are used either to collect, analyze, output, or integrate information across those areas. There are strong GLP restrictions about how computer systems can be developed, how the data get stored, and the security of that information when the data are generated for eventual submission purposes."
Fuchs said he can envision a time when FDA requires systems used by bioinformaticists to meet the same standards as those used by clinical informaticists. "It's hard to say what the details of any regulation would be. But now, if you set up a bioinformatics computer system, you don't care very much about documenting or things like that. If you work on the clinical end or the late-stage development end of a pharmaceutical pipeline your computer systems all have to be validated and your environment has to be set up to conform with laboratory practices." Fuchs said it's a completely different world, but one that bioinformaticists will have to get used to if they begin handling regulated data.
Such compliance issues could impair bioinformatics activity, DeMent argued. "You just cringe when you look at the requirements for validation. It's gotten better, but at least four years ago, if it took you six months to develop an analytical tool, it would take you at least that long to validate it before you could use it."
Data confidentiality is another issue that many bioinformaticists expect they will need to begin addressing. Remarked Fasman, "On the preclinical side we work mostly with anonymous data. We don't have to worry about specific data for specific patients or confidentiality. Most of the human genomic data we work with is not patient-specific."
On the clinical side things are different, he said, noting, "They have a lot of experience with setting up appropriately secure databases."
DeMent explained: "Let's say you want to use information specifically for pharmacogenetics analysis for a clinical trial. You have identifications for the patients, you know what their phenotypes are, their history, those sorts of things. If you want to end up using those same data for research efforts you will have to go through a whole process of nonidentifying that patient to make sure you cannot track that sample back to the individual."
Added Liebman, "You need to ask, is the material available for other kinds of analysis later on and then, what's the ethical responsibility for reporting observations that may be made to the patient, because that would have to breach confidentiality."
Where Vendors Fit In
With users facing such burdens, what will pharmacogenomics mean for bioinformatics vendors? They'll need to start thinking about many of the same issues, according to the bioinformaticists. Researchers at pharmaceutical companies will be looking for products that answer their needs for GLP validation, security, and scalability, as well as products that can help deal with the complexity of the data they are handling.
"It's really the magnitude and complexity that concerns me," Bigwood reiterated. "Being able to track this information and being able to evaluate it properly will go far beyond what the needs or capabilities are of bioinformatics today."
"The future will require a more complex set of methods," asserted Liebman. "We need the tools, both in terms of computational methods and database resources, to enable us to deal with the complexity of the biological system and its adaptability, rather than trying to simplify it," he continued. "This includes considering biological systems as integrated systems and processes, not individual reactions, and going beyond what we've learned from a molecular basis. Reactions need to be dealt with at a more engineering level as systems."
Bigwood added, "The technical needs really have to do with the complexities. We're talking about one or two magnitudes of complexity added on to the data we're already trying to manage. Obviously that will require different database designs, different ways of looking at the data, and probably more of an emphasis on tools for visualizing data or at least ways of automatically processing large amounts of data."
"We'll be well beyond the days where you can take Blast searches and interpret the results by eye," he concluded.