As the name suggests, pharmacogenetics is the intersection of the fields of pharmacology and genetics. Simply stated, pharmacogenetics is the study of how genetic variations affect the ways in which people respond to drugs. These variations can manifest themselves as differences in the drug targets or as differences in the enzymes that metabolize drugs. A difference in the target will usually lead to differences in how well the drug works, whereas differences in metabolizing enzymes can result in differences in either efficacy or toxicity. It's also possible that genes not directly involved in a particular pathway could end up being predictive of clinical outcomes.

Although pharmacogenomics has the potential to radically change the way health care is provided, it is only in its infancy. In the future, pharmacogenomics could find uses along the entire drug discovery and development timeline, all the way from target discovery and validation to late-stage clinical trials. Beyond that, pharmacogenomic tests could find their way into the doctor's office as a means to get the right medicine to the right patient at the right time.

While genetics and genomics are often used synonymously, pharmacogenetics is more focused in scope than and is viewed as a subset of pharmacogenomics, which encompasses factors beyond those that are inherited.

Some people believe that pharmacogenomics will lead to the stratification of diseases into genetically defined categories. Kenneth J. Conway, president of Millennium Predictive Medicine in Cambridge, Mass., uses the example of breast cancer.

	MULTIPLE GENOMES Just looking at a group of people is proof that there isn't just one human genome. Identifying SNPs requires looking for differences in DNA sequences across an ethnically diverse panel of individuals.
	Pharmacogenomics will lead to the stratification of diseases into genetically defined categories.

"WE ALREADY know that if we sample tumor tissues from 100 different women, those tissues would have a molecular makeup that would break up into different categories. In essence, those patients [each] have a different disease, but we just happen to be calling it the same thing--breast cancer," Conway says. "We think we're going to subdivide diseases. Once we get people with the right disease diagnosis, the disease definitions are going to change from 'You have breast cancer' to 'You have molecular profile A, B, C, or D.' The treatments of those diseases are going to be different."

Douglas Dolginow, senior vice president of pharmacogenomics at Gene Logic in Gaithersburg, Md., also believes that diseases will be subdivided using genomic information. "Rather than just looking under the microscope and saying [the sample] looks like squamous cell carcinoma, [we'll recognize] that squamous cell carcinoma group may be five or six different genetic or genomic classifications of disease, each with a different outcome and each with a different treatment," he says.

For pharmacogenetics to be effective, markers must be found that indicate the connection between drug response and genetic makeup. The markers that companies are pursuing most diligently are known as single-nucleotide polymorphisms, or SNPs (pronounced "snips"). SNPs, which are defined only in relation to a population, are variations in DNA at a single base that are found in at least 1% of the population.

A group called the SNP Consortium--a partnership of pharmaceutical and technology companies, academic research centers, and the Wellcome Trust--is working to publish a high-density SNP map of the human genome. Although the consortium's original goal was to map 300,000 SNPs, it already has a publicly available map of more than a million SNPs. The SNP database, which is maintained by Cold Spring Harbor Laboratory, can be accessed at http://snp.cshl.org/db/snp/map.

Orchid BioSciences, located in Princeton, N.J., has collaborated with the SNP Consortium on two projects. It has been the primary commercial company working to confirm many of the SNPs in the public database. "The very fact that people report the existence of a SNP is usually a result of sequencing several people or of in silico [computational] alignment of sequences from several people," points out Scott L. Rakestraw, acting general manager of the Pharmaceutical Value Creation business at Orchid. "The way one confirms a SNP is by pulling together an ethnically diverse panel of DNA and then assaying for the presence or absence of that SNP. In about 50% of the cases, we find that putative SNPs cannot be confirmed."

The second project that Orchid is conducting for the SNP Consortium concerns allele frequency determination. (Alleles are alternative forms of a gene.) "Once you confirm that a SNP does exist, the next question you want to ask is, 'What is the frequency of occurrence of that SNP within the members of ethnically diverse populations?' " Rakestraw says. "We query the presence or absence of a particular SNP over a large panel of ethnically diverse DNA profiles. We count how many times we see the occurrence of the SNP within each of the populations, divide by the total, and that's the allele frequency." Orchid released the first group of allele frequency data generated from 10,000 SNPs to the consortium in late June. Orchid is using its early access to these data to develop assays and panels of potentially valuable SNPs for its pharmaceutical and academic medical research customers.

ANOTHER FIRM, Incyte Genomics in Palo Alto, Calif., has a database of more than 100,000 polymorphisms, says Alan J. Schafer, vice president of genetics. Whereas the public efforts have gathered SNPs from across the genome, Incyte has focused on only the portions of the genome that code for proteins, what Schafer calls the "business end of the genome." In addition, Incyte offers a data set related to ADME (absorption, distribution, metabolism, and excretion) genes and SNPs, which are associated with how the body responds to drugs.

Meanwhile, CuraGen, in New Haven, Conn., focuses its activities "on those genes that we know historically you can turn into drugs," explains Michael P. McKenna, vice president of collaborative research. "The reality is that even with the substantial amount of public information [available], when it gets down to a gene or a cluster of genes around a disease area that you're really interested in, you have to go in, you have to get all the SNPs from that gene, and you have to be able to do the association studies with the right clinical population," he says. About 8,000 genes are appropriate targets for drugs, according to McKenna.

	VIEW A LARGER VERSION (98k)
	FAMILY TREE Identifying genetic causes for common diseases requires the use of a large group of distantly related individuals. Here, deCODE Genetics uses the Icelandic genealogy to determine how a group of asthma patients are related going back 11 generations.
	AT YOUR SERVICE Orchid provides high-throughput genotyping services to pharmaceutical, biotechnology, and academic customers through its MegaSNPatron facilities.

Sequenom, based in San Diego, uses mass spectrometric methods to study SNPs. "The advantage of using a mass spectrometer as a detector is that the mass spectrometer is a self-validating instrument. If the assay works the first time you run it, you basically know that it works and know that it's valid," says Charles R. Cantor, Sequenom's chief scientific officer. Cantor was one of the driving forces at the Department of Energy behind the human genome project.

Sequenom's scientists are interested in changes in the frequency of SNPs as the population ages. "We take advantage of the fact that most human diseases are late-onset. Age is a major risk factor," Cantor says. "If young people are carrying a harmful variation, they're still well, whereas an old person carrying that same variation has a very high chance that he's been made sick or killed by it. You make the prediction that variations that are harmful to health should decline in frequency as a function of age in the healthy population."

So, Sequenom designs assays to look for every public-domain SNP in people's DNA, Cantor says. Each assay is specific for a single SNP. "We have a database of 1.7 million designed assays. Of those, we've run about 50,000 public-domain SNPs, and we've run another 50,000 private-domain SNPs, in terms of actual physical assays that have been performed in collaboration with Incyte, one of our partners," he says. Sequenom's assays work on the first pass 94% of the time. "We throw away the ones that fail. With these numbers, it isn't even worth looking at them again." The first 10,000 assays were performed in a collaboration with the National Cancer Institute, and the results were published earlier this year [Proc. Natl. Acad. Sci. USA, 98, 581 (2001)].

One percent of genes appears to show an age-dependent frequency in SNPs, Cantor says. He suspects that only 200 to 400 genes will be involved in disorders that affect a major population. Finding these genes in healthy people, however, gives no indication of what the diseases actually are. "After we find them in the healthy population, we have to go back and look at biochemically stratified populations or clinically stratified populations," he explains. "The advantage is that instead of having to do all the genes with these tricky populations, we only have to do 200 to 400. We can pay a lot more attention to the details."

Instead of focusing on single SNPs, Genaissance Pharmaceuticals in New Haven, and Variagenics in Cambridge, Mass., look for SNPs that travel in groups, known as haplotypes, and work together to cause a particular drug response. Haplotypes are discovered by sequencing DNA, just like SNPs are. Then Genaissance uses proprietary software to reconstruct the haplotypes.

"WE FIND THAT a given gene has about 14 or 15 SNPs, but since you have two copies of the gene--one from your father and one from your mother--the SNP information is unphased. If you can pull that apart and look at the haplotypes, which are really the two chromosomes, you have a much more accurate view. Essentially you have the complete view of what two versions of the gene you actually have," says Richard S. Judson, senior vice president of informatics at Genaissance.

Genaissance has taken its haplotype technology into the clinic. The company recently finished recruiting for a trial in which it is looking for genetic associations in the patient response to four of the major statin drugs for lowering cholesterol. "We are essentially running a trial that has the same parameters as the Phase III trials that got these drugs approved--the same number of people, the same inclusion criteria, the same end points. We're taking a very deep genetic look at the patients we recruited."

Earlier work by Genaissance demonstrated that haplotypes for the ₂-adrenergic gene, not individual SNPs, were predictive of asthma patients' response to albuterol [Proc. Natl. Acad. Sci. USA, 97, 10483 (2000)].

Variagenics uses a technology called NuCleave, which incorporates mass spectrometry as the detection method, for looking at haplotypes. "We amplify a small region of DNA around a SNP, then fragment the DNA and measure the mass of the fragments that are produced. In a way, we are determining the sequence around the SNP," says Colin W. Dykes, vice president of research.

Not everyone, however, believes that SNPs are the way to go. "People have to get over this tendency to talk incessantly about SNPs in the context of pharmacogenomics," says Kari Stefansson, president and CEO of deCODE Genetics in Reykjavik, Iceland. "SNPs are just one set of markers. They are relatively uninformative markers that are, however, very easy to automate."

Expression profiling--in which mRNA levels are measured to determine which genes are turned on at a given time--can also be used in pharmacogenomics. Compugen, of Tel Aviv, Israel, and Jamesburg, N.J., has been working to build an accurate picture of the "transcriptome," its president, Eli Mintz, says. Many genes actually code for several mRNA transcripts and therefore several proteins. The transcript that is produced depends on how the gene is spliced.

"There will be variations between people at many different levels," Mintz says. "We try to identify all the splice variants that are out there and look at SNPs that are specific for these splice variants and for regulation patterns that are specific for these variants in different individuals. This is an early-stage research project."

This work builds on Compugen's LEADS technology, which is a set of software programs and algorithms that take DNA sequences from expressed sequence tags and other genomic data as inputs. The software can identify putative SNPs.

Pharmacogenetics is primarily being applied to pharmacokinetics; that is, how the body processes a drug. However, DNA Sciences, based in Fremont, Calif., focuses on pharmacodynamics, or how the drug affects the disease, in addition to pharmacokinetics, says Hugh Y. Rienhoff Jr., chairman and CEO. He differentiates DNA Sciences from other companies by saying it is looking at "the heritable components that influence disease susceptibility and disease course and therapeutic response in a given disease."

"Pharmacodynamics is very influenced by what type of disease you have," Rienhoff says. For example, diabetes and infectious hepatitis are diseases with different types that respond very differently to different classes of drugs. "If you didn't have the distinction between type I and type II diabetes, which we didn't have 60 years ago, your success in managing those diseases or discovering their treatments would be much more difficult," Rienhoff says.

DNA Sciences is focusing most heavily on cancer, particularly breast, colon, and prostate cancer, in which genetics plays a large role in disease susceptibility. "We have every reason to believe that identifying susceptibility markers is going to be useful for the clinician and for the patient managing his or her health," Rienhoff says. "If I were a 30-year-old male, and I knew that I had a one-in-five chance of developing colon cancer because of a genetic susceptibility test, I would--and my physician would probably counsel me to--get a colonoscopy sooner rather than wait until I was 50."

	TO THE POINT Advancing pharmacogenomics requires high-throughput instrumentation such as Sequenom's SpectroPOINT sample spotter, which can transfer nearly 4,000 samples in a single run.
	"We can know within minutes exactly how everyone is related to everyone else.".

THINKING THAT breast cancer markers such as BRCA1 and BRCA2 only provide information about susceptibility does the markers a disservice, Rienhoff believes. "You would think a susceptibility marker is something you wouldn't test in somebody who already has the disease, but in fact it might be the right thing to do," he says. "It gives you a better sense of the kind of disease they have. Obviously, breast cancer caused by BRCA1 is a different kind of cancer than cancers not caused by BRCA1 mutations."

DNA Sciences is interested in finding common causes of diseases. In genetic parlance, that means they are looking for "low penetrance" genes. "Penetrance is essentially a measure of the probability of your getting the disease over a lifetime, given that you have a certain genetic variant. The higher the penetrance, the rarer the disease." Lower penetrance genes are more likely to be involved as a common cause of the disease, because they are associated with a risk rather than a certainty of developing a disease.

"If you're looking for moderately penetrant genes," Rienhoff says, "the best way is to look in families, whether they be gigantic families such as the large ones [maintained in a population database that] we study through the University of Utah or small families, which are represented by some of our academic collaborations."

In the case of breast cancer, these pedigrees from the family population database are used to determine how breast cancer patients are related. Families--which Rienhoff actually calls kinships because they are larger than the typical notion of family--contain 1,000 to 5,000 members. "We can pick out those kinships that have an elevated frequency of cancer across all the individuals in that kinship, suggesting that the gene came from a common ancestor," Rienhoff says. "Those are the genes that we think are most important for clinical practice, because they have the widest utility. They address a much larger proportion of the genetic risk in the population."

Another company is also taking a family-oriented approach, with a family tree that stretches back more than a millennium. Based in Iceland, deCODE Genetics has access to the genealogy of the entire nation. "Our strength lies in our population approach to human genetics," points out deCODE's Stefansson.

"THE SCARCE resource in human genetics is access to a good population," Stefansson says. "When you're studying human genetics, you're studying the information that goes into the making of man and how that information flows from one generation to the next. To be able to do that well, you have to know the population structure. We can basically take the list that includes everyone in the country or 2,000 people with schizophrenia. We can know within minutes exactly how everyone is related to everyone else, which is key for being able to study the genetics of anything in a sensible manner."

For example, deCODE has used the Icelandic family tree to look at people who are taking statins. Approximately 10,000 people in Iceland take statins, but about 2,000 of those don't respond. The list of patients who don't respond can be run through the genealogy database. "I can tell you that they are related to each other, and we can get families that have a structure that allows us to map a gene that indicates a lack of response to statins."

In another example, deCODE has developed a test that predicts with more than 90% accuracy whether an asthma patient will respond to steroids. "That's very important because steroids have significant side effects," Stefansson says.

Some companies are looking at other markers, too. For example, Millennium Predictive Medicine and CuraGen both have integrated platforms that they use to look at changes in RNA expression and protein expression. "You need very sophisticated systems that can do that analysis in all those three different mediums [DNA, RNA, protein] and then correlate across all three to give you the right target," Millennium's Conway says.

"We recognized a while ago," CuraGen's McKenna says, "that by using an integrated genomics platform, you can understand these complex biological problems like drug response using the same tools that you use to understand disease and identify targets in disease."

CuraGen's platform is made up of four main technologies: SeqCalling, GeneCalling, PathCalling, and SNPCalling. SeqCalling generates expressed sequence information, using DNA sequencing technology; GeneCalling is an mRNA profiling tool used to determine expression differences between tissues; PathCalling identifies protein-protein interactions on a global scale; and SNPCalling identifies genetic associations through genotyping. "These four processes pretty much cover the chemical matter that you can interrogate inside the cell in a high-throughput manner," McKenna says. "Using that platform, we've been able to go through the entire genome. We've been essentially sorting the best targets and putting validation information around them."

Pharmacogenomics can be used in many ways during drug discovery and development. For example, it can identify variations in drug targets and ensure that companies are screening against the most common variant. However, pharmacogenomics is most likely to be used during the clinical development process.

"A typical scenario is that a company would come to us with data from a Phase II study," Variagenics' Dykes says. "The study may indicate that their compound was extremely effective in some patients, but overall the efficacy data might be disappointing. They have to decide whether to proceed into Phase III or can the compound. We would analyze their Phase II data, identifying markers or SNPs or haplotypes that would identify the patients in whom the drug did work. They could take the drug into Phase III studies based on demonstrating superior efficacy in genetically selected patients."

Pharmaceutical companies will be able to use pharmacogenetics to help differentiate their drugs in a crowded marketplace. "Say you're developing the fifth drug in some market," Judson says. "You want to know: 'How am I going to fit myself into that very crowded, very complicated market?' We can go find markers for nonresponse to the drugs that are out there. This is a group of people that form an underserved market. 

AT ORCHID, the task of identifying new assays falls to the Pharmaceutical Value Creation business. "We are very clinically focused," Rakestraw says. "We are the ones who take some of those millions of random SNPs out there, conduct clinical trials with them, and see if we can understand the impact of genetic diversity on the clinical outcomes."

Incyte, meanwhile, is looking for pharmacogenomic markers for osteoarthritis, which encompasses about half of all the cases of arthritis. In this disease, which affects primarily women older than 45, cartilage in the joints breaks down. "One of the reasons drugs haven't been developed is that the progression time is long, so the clinical trials are long," says Incyte's Schafer. The progression is monitored by taking X-rays of the joints and measuring the thinning of the distance between the bones. Clinical trials are expensive because they require many patients and a long time.

"It turns out that there are people who are known as fast progressers," Schafer says. "In a period of a year or so they'll have a significant degradation. If you could find a marker that would identify fast progressers, it would reduce the number of individuals that you would have to have and the length of time you would need in a clinical trial. We're looking at several hundred genes that we think are candidate genes to be involved in the development of this disease." If trials could be first performed with fast progressers to make sure a drug worked, larger clinical trials including slow progressers would become more feasible.

Pharmacogenomics will probably be most successful in areas such as oncology, where many therapies are available but each one works for only a small percentage of patients. Getting the treatment right the first time can be important, even life-saving. "If a patient has cancer and is going to be treated with a drug for an extended period before you find out if the drug has worked, and if you know that a given drug is only going to work in 10% of patients, there's a very strong reason for a physician to do a test to try to get the drug right from day one," Variagenics' Dykes says.

"It's not unusual for there to be 70% nonresponders for a chemotherapeutic drug. Clearly, cancer is an area that's ripe for understanding the pharmacogenetics," Orchid's Rakestraw says. "It's also tough because there's no single definition of cancer. Even within a particular type of cancer, like colon cancer, you can find extreme molecular variability."

Toxicity is another important application of pharmacogenomics. One such test, developed by researchers at St. Jude Children's Research Hospital in Memphis, identifies patients who have mutations in the thiopurine S-methyltransferase gene, which is involved in the metabolism of mercaptopurine drugs. Children who metabolize these drugs too slowly can literally overdose on their treatment.

In a collaboration with Bayer, CuraGen is analyzing compounds for toxicity early in the process, right after high-throughput screening. "Very early in preclinical research, you can try to eliminate compounds that are the most egregious offenders with regard to safety and help direct the medicinal chemistry resources toward those compounds that have the highest probability of being a success in the long run," McKenna says.

In order for pharmacogenomics to advance, people must start treating it as a serious field, Stefansson says. Also, they must stop acting as if it is separate from the rest of genetics. "The fact of the matter is that you will need a population to study that has the same qualities that you need to study the genetics of any human trait. Let's face that."

Technology must also advance if the field is to move forward. A platform that can do large amounts of genotyping inexpensively is required, Judson says. "Right now, the cheapest real way to do genotyping costs about a dollar per SNP. If you want to look at 100,000 SNPs, it's $100,000 a patient, which just isn't practical. You need to get that down to a penny [per SNP]. Then suddenly there are a lot of interesting experiments that you can do that you can't do today," he says. Technology that is inexpensive could be available in as early as three, but no later than 10, years from now, he predicts.

Cantor concurs that costs need to decrease. "These are very expensive studies," he notes. "If you make a mistake in study design, you may not be able to afford to recover from it." But, Dolginow points out, "as the technology costs come down, which they continue to do, that improves access and possibilities to do the billions of dollars of work that has to be done."

In addition to technological improvements, gaps in the knowledge of the human genome need to be filled. "Certainly the completion of the human genome provides the starting material upon which an analysis of the variation in that genome will continue to proceed," says Denis M. Grant, senior director of pharmacogenetics at Orchid. "Upon that framework of genome variation, we'll need to superimpose the observed variations that we see in clinical drug responses and in toxicity to drugs."

"One critical thing," Judson says, "is to have the human genome project really finished. There are still big gaps in terms of mapping genes onto the chromosomes, he points out. "We don't even know what all the genes are. If we had all the genes, there's a lot of work that could be done. There are a lot of experimental techniques that you could run in a very high-throughput way."

Advancing pharmacogenomics will also require well-annotated samples on which to do the necessary research, Millennium's Conway says. He also believes that pharmaceutical and biotech companies will have to form partnerships with diagnostics companies so that pharmacogenomics is "available at some reasonable price and some reasonable time frame so it can be not just a scientific breakthrough but a commercial success."

IN ORDER for pharmacogenomics to take hold, pharmaceutical companies, physicians, and patients are going to have to understand how it will benefit them. That is, they need to see the "value proposition," as Rakestraw puts it. "One way you can show the consumer side of the market a value proposition is to actually come up with very effective pharmacogenetic relationships that help them avoid adverse drug reactions or help them be put on the right drug the first time," he says. Focus groups indicate, Rakestraw says, that people will be willing to pay to have their pharmacogenetic profiles assessed when the technology becomes readily available and then they will use that information to make health care decisions with their doctors.

Along those same lines, CuraGen's McKenna comments that "the most important thing is to prove to the industry that [pharmacogenomics] will have a positive overall impact on the cost of drug development. If you can shave 20% off your costs and in the end reinvest those dollars into more projects, you end up having 20% more drugs on the market. That's a huge positive impact on the industry."

One drug already on the market requires administration of a test before it can be prescribed. The test is not a true pharmacogenetic test, because it measures protein expression in a tumor rather than the underlying genetic makeup of the patient, but it shows the power of such tests. The drug is trastuzumab, sold under the trade name Herceptin by Genentech, South San Francisco.

Herceptin is an engineered monoclonal antibody that is used as a treatment for metastatic breast cancer. It is only prescribed for patients whose tumors overexpress the human epidermal growth factor receptor (HER2) protein--about 25 to 30% of women with breast cancer.

The company knew very early that it would develop a test to be used in conjunction with prescribing Herceptin, says Neil Cohen, a company spokesman. HER2 overexpression had been linked to cancer prior to clinical development, and Herceptin was engineered specifically for HER2.

Because of the test, Genentech was able to streamline the Phase III clinical trials, using only patients who overexpressed HER2. Herceptin was approved with data from two Phase III clinical trials, one with 469 patients and one with 222 patients. Because so few breast cancer patients overexpress HER2, a clinical trial that did not selectively enroll patients would have needed to be much larger to show statistically significant efficacy, Cohen says.

The protein overexpression is identified using immunohistochemical (IHC) testing. However, measurements of amplification of the HER2 gene using fluorescence in situ hybridization (FISH) appear to be a better predictor of patient survival rates. In a retrospective study, gene amplification was measured in tumor tissue from 458 out of 469 patients in one of the Phase III studies. The women who were identified by the FISH measurement survived 50% longer (27 months) when they received Herceptin in combination with chemotherapy relative to women who received chemotherapy alone (18 months). In contrast, the IHC test resulted in a 24% survival advantage. Genentech does not provide the IHC test itself. Instead, it collaborated with the Danish diagnostics company DAKO, which developed and markets the test.

Herceptin's sales were $188 million in 1999, its first full year on the market. That figure has been growing--to $276 million in 2000 and $160 million in the first half of 2001. Obviously, the need for a test has not slowed the demand for Herceptin.--CELIA HENRY

Chemical & Engineering News
Copyright © 2001 American Chemical Society