Pharmacogenomics Lurches Forward

PREDICTIVE POWER:© 2004 Massachusetts Medical SocietyThis analysis of gene expression ranks 36 genes on the basis of their predictive power (univariate z score), with a negative score associated with longer overall survival and a positive score associated with shorter overall survival. The dashed lines represent an absolute univariate z score of ± 1.5. The prediction model is based on the weighted expression of six genes in the equation shown. (N Engl J Med, 350:1828–37, 2004.)Me

Mike May
Aug 1, 2004

© 2004 Massachusetts Medical Society

This analysis of gene expression ranks 36 genes on the basis of their predictive power (univariate z score), with a negative score associated with longer overall survival and a positive score associated with shorter overall survival. The dashed lines represent an absolute univariate z score of ± 1.5. The prediction model is based on the weighted expression of six genes in the equation shown. (N Engl J Med, 350:1828–37, 2004.)

Medicinal drugs save patients, but compounds fail in many cases and even cause death in others. According to the US Food & Drug Administration's Center for Education and Research on Therapeutics, adverse drug reactions kill 100,000 Americans each year, registering as the fourth leading cause of death.1 In the United States at least, prescribed drugs are more lethal than AIDS or automobile accidents. The problem is, no physician can know for certain...


Although much of the hype for so-called personalized medicine began as the human genome sequence came within reach, pharmacogenomics actually preceded sequencing efforts. Richard Weinshilboum, professor in the departments of molecular pharmacology and experimental therapeutics and medicine at the Mayo Clinic in Rochester, Minn., says, "There is some lack of appreciation that pharmacogenomic concepts date back to mid-20th century." In the late 1950s, for example, scientists discovered that some people inherit a genetic mutation for cholinesterase that causes the slow breakdown of succinylcholine, a muscle relaxant.2 Single genes control many traits that can be approached on an individual basis.

Even a more recently heralded poster child for pharmacogenomics, Iressa, an epidermal growth factor receptor (EGFR) inhibitor has variable effects based on a single gene. Iressa battles non-small-cell lung cancer (NSCLC), which accounts for 85% of all lung cancers. Unfortunately, the drug fails more often than it works. Moreover, different populations experience varying efficacies with Iressa, which works in about 25% of NSCLC cases in Japan, but only about 10% of cases in the United States. Because of those numbers, Matthew Meyerson of the Dana-Farber Cancer Institute in Boston and his colleagues suspected a genetic factor in Iressa's efficacy.

To explore that hypothesis, Meyerson and his team looked for mutations in EGFR in 119 NSCLC tumors from both Japan and the United States. They found activating EGFR mutations in 15 of 58 (25%) tumors from Japan, and just 1 of 61 (~1.5%) US tumors.3 Another group from Harvard Medical School and Massachusetts General Hospital had similar findings.4 From this research, Meyerson says, "The concept has been established that an effective strategy for cancer treatment is to inhibit proteins that have cancer-specific mutations." In the case of NSCLC, the right mutation means the difference between a powerful therapy and no response. "The finding that patients with EGFR mutations respond dramatically to Iressa," says Meyerson, "will allow physicians to select patients who will most benefit from treatment."


More complex gene-response relationships have begun to emerge. Some drugs, for example, demand precise attention to dosage. Warfarin, the most common oral anticoagulant, is prescribed to two million Americans, and it causes severe bleeding in some patients. "We've got to get the dose just right," says Brian F. Gage, associate professor of medicine at Washington University in St. Louis, "and that is highly variable. I might be taking care of a football player with a blood clot after orthopedic surgery, and he might need less warfarin than a little old lady who just had a hip fracture."5

Several genes play a role in warfarin's metabolism. Polymorphisms in CYP2C9, a gene for cytochrome P450, cause about 30% of patients to be slow warfarin metabolizers, which could result in high blood concentrations.6 Moreover, Johannes Oldenburg's research group at the University of Würzburg in Germany recently showed that polymorphisms in the vitamin K epoxide reductase multiprotein complex (VKOR) affect warfarin metabolism in rats.7 Oldenburg's team showed that mutations in one of the complex's subunits, VKORC1, confer warfarin resistance in some human disorders. Overexpression of the wild-type protein made rats sensitive to the treatment.

"The bottom line," says Gage, "is that we want to make this drug's dosing safer, instead of using trial and error." In a study of 48 orthopedic patients, Gage showed that testing for CYP2C9 polymorphisms does provide a better starting point for the warfarin dose, which would achieve stable blood levels more quickly than trial-and-error dosing. In an upcoming study, Gage will genotype for both CYP2C9 and VKORC1 when prescribing warfarin before orthopedic surgery.

Multiple genes participate in most forms of cancer, too. At one time, large-B-cell lymphoma, the most common form of adult lymphoma, proved universally lethal. Today, chemotherapy cures about half the patients with this cancer. But Ronald Levy, professor of medicine and chief of the division of oncology at Stanford University Medical Center, wants a better result. As the treatment outcome varies considerably in patients with similar characteristics, including age and tumor stage, Levy suspected a genetic contribution. To zero in on the genes related to large-B-cell lymphoma, Levy turned to microarray studies from a number of other labs that measure gene expression from biopsy samples. He narrowed the focus to a panel of 36 genes deemed predictive for drug response.8 "We are looking at a phenotype, a behavior pattern, that has no single manifestation," says Levy. "Even if a single gene were found to be the master controller, it may not be measured accurately, and therefore multiple ways of measuring its action add precision."

From 1975 to 1995, Levy and his team collected tumor samples from patients who had just been diagnosed with large-B-cell lymphoma. Using quantitative PCR, Levy and his team measured the expression of all 36 genes, and then compared the findings to the survival of 66 patients who were all receiving the same regimen of chemotherapy. Analysis showed that six genes related to blood production, inflammation, metastasis, protein production, and immune system regulation correlated most closely with patients' survival.

In some cases, genetic tests can already be used in diagnosing chemotherapy. DakoCytomation in Copenhagen, Denmark, makes EGFR-pharmDx that determines if a colon tumor overexpresses EGFR. If it does, then the patient with that particular colorectal cancer can be treated with Erbitux from ImClone in New York. Other diagnostic-drug combinations lie just ahead.


Even as the data continue to grow, pharmacogenomics faces significant challenges in moving to general use. "We need an informatics structure for physicians to use pharmacogenomics," says Russ B. Altman, associate professor of genetics and medicine and director of the bioinformatics training program at Stanford University Medical Center. "Without that, it will be hard to deliver useful knowledge." But the costs involved may thwart those efforts.

Heiner Fangerau of the Institute for the History of Medicine at Heinrich-Heine University in Dusseldorf, Germany, says, "Pharmacogenomics sounds good at first, but if there is just one person in a million who could benefit from a pharmacogenomic test, then it may not be affordable to have the other 999,999 people undergoing the blood test prior to therapy." Fangerau says that clinical applications of pharmacogenomics remain rare, but he thinks that psychiatry could benefit from this field very soon.

Others agree, including Gustavo Turecki, director of suicide studies at McGill University in Montreal: "In psychiatry, the phenotype is complex, and there are lots of variations in response to treatment." For example, psychiatrists treat bipolar disorder with lithium. "If it works well, the patient will be normal," says Turecki. A good response to lithium runs in families, and that response could come from having the "right" genes, including the serotonin transporter gene (SERTPR). Alessandro Serretti, a psychiatrist at the University Vita-Salute in Milan, Italy, says, "SERTPR variants influence lithium efficacy through their control of serotonin transporter transcription. That is, some subjects have less serotonin transporter in their brain. Therefore, their serotonin pathway is unbalanced and less responsive to the stabilizing effect of lithium."

So far, pharmacogenomics offers promise to only those with the right genes, or the right mutations. Nonetheless, many scientists see broader benefits ahead. Says Weinshilboum: "Pharmacogenomics is an unprecedented opportunity to use high-throughput technology to dissect a multitude of genetic pathways to enhance drug therapy."

Mike May