"Warning," blares the voice on the television. "This drug may cause drowsiness, dizziness, blurred vision, rapid heartbeat, shortness of breath, upset stomach, hallucinations, loss of sexual function, or death. Do not take if you are pregnant or plan to become pregnant." Sound familiar? Ads for antidepressants warn of the possible risk of suicidal thoughts, while commercials for erectile dysfunction medications notify users of the possibility for uncomfortably prolonged erections. Caveats like these should be recognizable to anyone who has ever taken so much as an aspirin. As much as medications help combat health problems and — in some cases — save lives, no medication or treatment is free from the risk of often annoying, sometimes debilitating side effects.
According to a recent blog post by Muin Khoury, director of the Centers for Disease Control and Prevention's Office of Public Health Genomics, an estimated 82 percent of Americans take at least one medication and about 29 percent take five or more drugs. Adverse drug reactions are responsible for about 700,000 emergency room visits and 120,000 hospitalizations per year, and cost about $3.5 billion annually, Khoury adds.
The US Food and Drug Administration recently reversed a decision to allow the use of the cancer drug Avastin to treat breast cancer. It was originally approved in 2004 to treat metastatic colon cancer and non-small-cell lung cancer, and its maker, Roche, had won accelerated approval from FDA in 2008 to add the breast cancer indication to the drug's label. But FDA officials reconsidered their decision when several studies came out suggesting Avastin created too big a risk of serious adverse effects with a minimal impact on breast cancer itself. Roche says a certain group of breast cancer patients — those with high levels of VEGF-A — respond very well to the drug, and plans to do studies on these so-called "super responders" to determine why they react differently to Avastin than other patients.
Different models, same purpose
The promise of pharmacogenomic research is neatly illustrated by the Avastin response example. Several researchers are attempting to find polymorphisms, genes, or mutations that may signal whether a patient will respond well to a drug or will suffer serious or debilitating side effects, whether that patient should be prescribed lower dosages or higher dosages, or whether something in that patient's environment could affect the way the drug works.
Different researchers use different approaches, but their purpose is the same — to find genetic factors that influence drug reactions. At the University of Chicago, Eileen Dolan and her team have built cell-based models to identify genetic variants that may put people at risk for toxicity from anti-cancer therapeutics. "I built these models because it's very, very difficult — almost impossible — to get a large cohort of individuals that are treated with the same chemotherapeutic agents at the same dosage to do pharmacogenomic discovery, to get at which patients are having an adverse drug reaction and which are not," she says. "These models allow you to look at, in a discovery way, millions of SNPs because we're using the International HapMap samples. And it allows us then to use a pharmacological phenotype, such as chemotherapeutic-induced cytotoxicity, or chemotherapeutic-induced apoptosis, and then we identify those genetic variants that are associated." Some of the cell lines are sensitive to pharmacological effects while others are resistant, allowing Dolan and her team to screen for the responsible SNPs.
She has found that about 20 percent to 35 percent of the pharmacological phenotype has a genetic component, and has identified SNPs associated with various reactions that people of different ethnic backgrounds have to certain medications. People of African descent tend to be more sensitive to anti-metabolite compounds, whereas people with Asian ancestry tend to be more sensitive to platinating agents, like cisplatin or carboplatin.
Dolan's biggest challenge, however, is clinically recapitulating what's happening in the cell lines. "That's really important to know because if we can figure out that the same pathways that are important in neuropathy are present in the cell lines and can identify, for example, paclitaxel-induced neuropathy genetic variants that are important for that particular phenotype, that would be very valuable," she says. "But we still don't know if the models are best for a particular toxicity or another toxicity." And because the cells are EBV-transformed lymphoblastoid cell lines, EBV affects the expression of certain genes; variables like that have to be taken into account to understand their effect on the genotype-phenotype relationship.
Dolan's view is that the cell line models could serve as a replication set for genome-wide association studies done within clinical populations. "Some groups are also doing GWAS … and they run into problems when they submit work to be published. They'll sometimes get a reviewer saying, 'You need to repeat that.' Well, that costs millions and millions of dollars to do," she says. "My view is that anytime there's a GWAS where we can get our hands on the data or collaborate with the individual group that generated the data, we ask: 'Are the SNPs that you found in this study enriched in our cell line study using the same drug?' And we've been finding consistently that you do see an enrichment of our SNPs." Such replication could serve as validation of results from GWAS, she adds.
At St. Jude Children's Research Hospital, Mary Relling does her research on patients enrolled in clinical studies. Relling is also trying to implement testing for genetic variants that are already known to be important in a clinical setting, so her group uses those data in real patient care — to adjust and modify the medications prescribed to patients.
"Most polymorphisms differ substantially, at least in frequency, by racial background — race, ethnicity — and sometimes that translates into differences in frequency of adverse effects among racial and ethnic backgrounds," Relling says. "Of course, one of the advantages of a genetic approach is that genetic variants will tend to be important regardless of what ancestral background they occur in. We really hope that you won't necessarily have to consider race or ethnicity. You'll just have to use genomics to decide if a patient is high risk or not. The idea is that you would, to some extent, personalize the way you would give medicines to people based on their genetic background."
Relling also studies non-genetic variables that affect a patient's sensitivity to adverse drug reactions. To do so, Relling does most of her work in the context of patients who are enrolled in clinical trials. "Leukemia patients who are enrolled in St. Jude clinical trials, or leukemia patients enrolled in Children's Oncology Group clinical trials — where you've collected all the information about their age, their other drugs, their disease severity, their kidney function, their liver function, their diet — using all that information plus genetics, you try to figure out what are the most important variables affecting whether someone has an adverse drug effect or not," she says.
Relling's challenge, however, lies in not having enough patients that are treated in clinical trials that also include germline DNA collection in their protocols. "In the childhood leukemia world, that's not as big a disadvantage for us because the vast majority of patients with that disease are treated in clinical trials and do have germline DNA collected," she says. "But in general in the field, most patients are not treated in clinical trials, and even those that are don't have DNA collection built into them, so that's probably the biggest thing that's slowing down research." It wouldn't take much extra effort to fix this, she adds, "just an incremental increase in money."
Although research is ongoing and there are already several pharmacogenomic tests available for physicians to use, it remains to be seen whether these tests are clinically useful.
One test that has been recommended for use in the clinic, says the CDC's Khoury, is the test for HLA-B*5701 to identify patients at risk of side effects from treatment with the HIV drug abacavir. Other tests, like those that help physicians determine the correct dosages of warfarin, could have a big impact on public health, he adds. But for most of the tests, there is still too little evidence showing their clinical utility.
CDC and the National Institutes of Health are trying — through various working groups — to change this. NIH's Pharmacogenomics Research Network, which the agency has funded since 2000, is tasked with understanding how people's genes can affect their responses to different medicines. As part of their work, the researchers participating in the PGRN collect data on the associations between human genetics and drug reactions in a database called the Pharmacogenomics Knowledge Base, or PharmGKB. In it, the researchers curate data on genotype and phenotype, annotate gene variants, and review the literature for new studies on gene-drug relationships.
"One big gap is that we haven't yet written very good, peer-reviewed, updatable guidelines that link genetic variation with real clinical decision-making and real, strong recommendations on how to adjust or change medications based on genetics," St. Jude's Relling says. "And so as part of the PGRN, we're working very hard to write these gene-drug pair guidelines that will give very specific recommendations that can then be implemented for use in the clinic, that will help clinicians make the right decision based on genetic variation."
Several other agencies and researchers have set up databases that serve a similar purpose. The National Cancer Institute runs the Cancer Adverse Events Reporting System, an open-source database that collects reports of adverse reactions to cancer drugs and treatment regimens. And at the University of Alberta, researcher David Wishart runs the DrugBank database, which he developed with his colleague Craig Knox in 2006. DrugBank — which is offered to the public as a free resource — is a curated repository of information on thousands of drugs, including their targets, interactions, and side effects. Knox and Wishart published the third version of the database, DrugBank 3.0, in Nucleic Acids Research in January, and Wishart says they plan to continue to update it with relevant information.
At CDC, Khoury says one of the mandates of his office is to keep track of advances in pharmacogenomics. To that end, the agency also maintains several of its own databases. The Human Genome Epidemiology Navigator, or HuGE Navigator, is a searchable program for information on genetic associations and human genome epidemiology. There are now "more than 60,000 entries on everything that has to do with gene-disease associations, and probably about 5 percent of that literature is on pharmacogenomics," Khoury says. The other, newer database run by CDC is the Genomic Applications in Practice and Prevention Knowledge Base, or GAPPKB, which provides information on any test based on genomic research that comes on the market for use by the public, including any pharmacogenomic tests. CDC has also commissioned the Evaluation of Genomic Applications in Practice and Prevention working group, or EGAPP, to review the literature and "develop evidence-based guidance on the use of genomic testing in practice," Khoury says. And the agency curates an online evidence collection at PLoS Currents: Evidence on Genomic Tests to provide physicians and the public with information on the validity of currently available pharmacogenomic tests, and other genomics-based diagnostics. "We're trying to follow the field very closely, and see what the gaps in knowledge are, and educate the rest of CDC and work with the programs to implement those that are ready for implementation and integrate them into public health programs," Khoury says. "We're also working with stakeholders and partners like NIH to plug the holes in our knowledge."
Eventually, researchers and regulators alike say pharmacogenomics will find its way into the clinic, and that it could become one of many genomics-based weapons in the physician's arsenal against disease. Chicago's Dolan says she thinks pharmacogenomics is not too far off from being used in the clinic. "I think the implementation is really getting the clinicians to buy into it and understand what it means, so I think the implementation side is that we need to bring these tests to a point where you just tell a clinician some genetic result, you put it into some kind of computer program that tells them which SNPs should get which drug and which shouldn't," she adds.
Physicians are busy people, Dolan says, and when they are multi-tasking and dealing with patients, the pharmacogenomic portion of their routine should be simplified to the point where an individual's genotype can be easily used to determine what dose that patient gets, and what side effects the physician should watch for. "One of the problems that we're running into is that when you have multiple genetic effects that are playing into a particular phenotype, and the contribution may be small for each one, then we really need to understand the pharmacogenomics signature, and then figure out should the dosage be altered because of that particular signature that a person carries," Dolan says. "The implementation will be interesting. It's starting to happen."
And the advent of cheaper and faster sequencing technology — which is also making its way into the clinic — will hasten the use of pharmacogenomics in the clinic, she adds. "I think once people are sequenced, you don't have the issues of where you genotype somebody today and five years from now you find a different set of SNPs that weren't genotyped in that individual before, so now you need to re-evaluate," Dolan says.
Before clinical implementation of pharmacogenomics really takes hold, however, an electronic medical record system is needed. "You have enough data on all these variables, and that you can efficiently warn clinicians when they're about to prescribe a high-risk drug to a patient that is at high risk for an adverse drug effect," St. Jude's Relling says. "You're going to need to have computing tools — computerized physician support — to help you write rules to warn clinicians and prevent them from making a bad prescribing decision."
CDC's Khoury remains a bit reserved about how ready pharmacogenomics is for the clinic at this point, but nevertheless seems enthused about its promise. "I think there's going to be a lot of new discoveries, both in adverse effects as well as effectiveness of drugs," he says, adding that it's nevertheless important not to "short-change" the necessary step of proving clinical efficacy before a test is used. "The technology is promising, but not there yet, in general. … We really need not rush into implementing it before we study it along the way to figure out if it can do more good than harm at the population level," Khoury adds. "The point is to use it when the science dictates or it matures enough, and that's going to be a challenge."