Rudolf Uher, MD
Professor, Department of Psychiatry, Dalhousie University School of Medicine, Halifax, Nova Scotia
Dr. Uher has disclosed that he has no relevant financial or other interests in any commercial companies pertaining to this educational activity.
TCPR: Dr. Uher, you have been a co-investigator on the major studies of pharmacogenetics in psychiatry. It’s a complicated field, and I thought you could explain how the research is actually done.
Dr. Uher: The basic goal of this kind of research is to try to find an association between a genetic variant and the clinical response to a particular medication. Given that there are dozens of medications to choose from for any given disorder, it would be very helpful if we could do a genetic test that would tell us which drug is the best for a particular patient.
TCPR: All right, so we all have 23 pairs of chromosomes and many thousands of genes that might play a role in medication response. How does one go about solving this problem?
Dr. Uher: Traditionally, we’ve done these studies using the candidate gene method. The way this works is that you start by identifying a small number of promising candidates for genes that might be related to drug response. For example, the serotonin transporter is the target molecule of SSRIs, so a logically related gene would be the serotonin transporter gene, which encodes the protein that is responsible for serotonin reuptake. Different people have different gene variants that affect how much of the transporter is produced.
TCPR: And how might variations in the amount of serotonin transporter protein affect the response to antidepressants?
Dr. Uher: Let’s assume that SSRIs work by blocking the reuptake of serotonin back into the neuron. They do this by disabling the serotonin transporter protein, making it harder to “vacuum” up excess serotonin and clear it from the synapse. That extra serotonin presumably helps to improve mood in patients with depression. If a person has a gene variant that causes the brain to make more serotonin transporters, that means the person’s brain supplies more vacuum cleaners to clear out serotonin. If that person took an SSRI, it might not be as effective because all those extra transporters would counteract the serotonin reuptake inhibition. Conversely, if a person produces less transporter protein, the SSRI would be able to block most of it, leaving more serotonin in synapses and therefore leading to a better antidepressant response. This is a simplistic explanation, but it’s the basic idea.
TCPR: Interesting. Are there any other gene variants that might affect antidepressant response?
Dr. Uher: Another example is cytochrome P450 genes (CYP genes) for liver enzymes that break down antidepressants. If there are too few of the enzymes, blood levels of the drug would go up, and this might cause more side effects. On the other hand, if there are too many enzymes, blood levels would be low, and there might not be enough response.
TCPR: What kinds of studies have been done with candidate genes?
Dr. Uher: There have been hundreds of studies done with various candidate genes to predict antidepressant response or toleratibility. Typically, you do genetic tests on a group of patients who have taken a particular antidepressant, you assess the response, and you do statistical analysis to see if there is any correlation between having certain gene variants and antidepressant response.
TCPR: And what have these studies concluded?
Dr. Uher: Unfortunately, thus far, there is no consistent effect of any of the genes on either antidepressant response or side effects. The two largest pharmacogenetic studies, of response and side effects, both found equally negative results. One was STAR*D, and the other was GENDEP. Speaking as both a psychiatrist and a researcher involved in these trials, it was a huge disappointment. When we were planning the analyses, we thought these genes were safe bets. We predicted not only that we would find effects of these genes, but also that we would find out about other genes that would be promising. In fact, we ended up disconfirming the effects of the genes we had assumed would be shoo-ins.
TCPR: That’s surprising—it seems that we often hear about findings showing some kind of association.
Dr. Uher: Yes. This was confusing, because there were many positive results, but none of them have been consistently replicated. We think the reason for this is that investigators typically test multiple candidate genes, but they tend to publish only the results that were positive. This selective reporting makes it appear that there are many positive results in the literature, but you are not seeing all the negative results. And if you run many different tests, it’s possible that you will get some positive findings by chance alone. It’s appearing very unlikely that variations in candidate genes have any real effects on response to antidepressants.
TCPR: Can you give us a bit more detail on the STAR*D and GENDEP studies?
Dr. Uher: STAR*D had genetic data on approximately 1,400 patients with depression, and GENDEP collected genetic data from over 800 patients with depression (Am J Psychiatry 2013;170(2):207–217. doi:10.1176/appi.ajp.2012.12020237). Both studies found important non-genetic predictors of outcome—for example, patients with symptoms of loss of interest, reduced activity, and anxiety were more likely to have poor outcomes. But the genetic results were largely disappointing. Both STAR*D and GENDEP examined comprehensive arrays of candidate genes, but could not replicate positive findings of previous smaller studies. STAR*D found a strong association in the serotonin receptor 2A gene, but this did not replicate in GENDEP. GENDEP found an association in the serotonin transporter gene, but it did not unequivocally replicate in STAR*D. In addition, both studies examined the CYP genes and found that they were unrelated to treatment outcomes or side effects.
TCPR: So what is your take on the commercial genetic tests that have been marketed? They certainly claim that their tests are predictive of both response and side effects.
Dr. Uher: I’ve looked at the studies done by one of the companies, Assurex, which markets the Genesight test. This test is based on a small number of candidate gene variants—the same ones that did not show replicable effects in large independent studies. The company did publish their own studies where they randomized patients to an active condition, which got the test, or to a control condition, which did not. They showed that people who get the test are slightly more likely to get better, but interpreting the meaning of this result is difficult. What they are doing in their studies is comparing a complex intervention against nothing.
TCPR: How does that make it trickier to interpret?
Dr. Uher: Let’s think of how they recruited patients. They approached patients and essentially asked, “Would you be willing to be in a study? In it, you will either get a test that will predict the best antidepressant for you based on your genetic makeup, or you will not get the test.” The people who got the test, and their psychiatrists, received a colorful printout that gave recommendations about which antidepressants would be best. Even if the testing that generates the recommendations is not valid, simply being in the active group generates a potential placebo effect. The treatment group feels like something special is being done for them. On the other hand, the patients assigned to the control group heard about the possibility of this cutting-edge gene test and they knew they were not getting it, so that generates disappointment, which can lead to potentially higher scores on depression scales.
TCPR: Sounds like there was a potential for biased results.
Dr. Uher: Exactly. It’s also important to keep in mind that only one of the Genesight studies was blinded. This study was small (49 patients total) and it did not meet the standards for a well-designed clinical trial, because the researchers knew which patients were getting the test, which can lead to measurement bias (Winner JG et al, Discovery Med 2013;16(89):219–227).
TCPR: So the evidence was low-quality and not convincing. What do you conclude?
Dr. Uher: If you do something special for your patients, it will be on average beneficial. Although the genetic tests being marketed do nothing helpful for prediction, at least they are not harming anyone. The psychiatrists ordering the tests feel this is something special for their patients, and the patients feel that they are being given something special. “Genetic test” is an impressive-sounding phrase, and there is also a novelty effect. There’s an old adage, “Drugs are most effective when they are first launched.” For example, when Prozac was first launched, it seemed to work for everyone. People get excited by new technologies, and this boosts the placebo effect.
TCPR: All right, so there is likely an expectancy effect here. Are there any other reasons why people would be impressed with these tests?
Dr. Uher: Yes. The printout listing the genetic results includes other information, including general guidelines about which antidepressants are good options for patients with depression. When doctors see such guidelines at the point of care, they tend to practice more rationally. They end up appropriately changing treatment more often when the first medication doesn’t work well, and this in itself leads to more effective practice. We have seen this in studies using measurement-based guidelines, which give the doctors tools to measure response and guidelines for what to prescribe in different circumstances. It’s likely that any effect of the genetic tests works simply by providing doctors with commonsense treatment guidelines—and not via any genetic prediction.
TCPR: We’ve talked about gene candidate studies. What other methods are being used to find predictive genes?
Dr. Uher: The current standard of genetic research is the genome-wide association study (GWAS). About nine years ago, we began to have the molecular technology and computational power to measure and analyze a very large number of variants across genes. Using these techniques, we can get information about gene variants in the entire genome. Testing all of the genes in this way is a huge advantage, because we no longer have to guess on a small number of candidate genes for our focus. This means that we can get genetic testing on patients who have responded to drugs and search for any possible gene-response association. (Editor’s note: For more information about GWAS, see https://www.genome.gov/20019523/.)
TCPR: And what sorts of findings have these studies yielded?
Dr. Uher: The first genome-wide significant finding came from the GENDEP study. We found a gene variant that predicted response to nortriptyline (Uher R et al, Pharmacogenomics J 2009;9(4):225–233. doi:10.1038/tpj.2009.12). It was on chromosome 6, and it is in the UST-1 gene. This gene is important for determining where newly generated neurons go in adulthood—it helps with adult neurogenesis. Unfortunately, over the past few years we still have not been able to replicate that finding, meaning it could still be a fluke. The problem is that since relatively few people use nortriptyline these days, it’s hard to recruit enough subjects to do that kind of study.
TCPR: Any other GWAS findings?
Dr. Uher: Yes. Another study got published in September 2016 based on data collected through the consumer genomics test “23andMe” (Li QS, Transl Psychiatry 2016;6(9):e889. doi:10.1038/tp.2016.171). They did a GWAS of 40,000 people who were treated with antidepressants at some point in their lives. Nearly 10,000 were treated with SSRIs, and they had no significant findings. However, in one of the smaller analyses of 4,000 people who had taken Wellbutrin, they found one association: a weakly predictive variant in an area on a chromosome that was not a gene itself, but was between two other genes. The odds ratio was 1.35, meaning that having this variant increases the odds of achieving remission on Wellbutrin 1.35 times. It’s a small effect size, and one may debate whether it should be considered statistically significant, because they conducted 12 different analyses and this was the only one that was positive. I am looking forward to seeing the results of replication, which should be easier to get for Wellbutrin, since it is commonly used.
TCPR: So overall, the results have been pretty underwhelming.
Dr. Uher: The message is clear that there is not a single variant that will predict response to antidepressants, but that response will be predicted by variants of numerous genes. The good news is that, given our more sophisticated research methods, we can now estimate what proportion of response from antidepressants is likely to be genetically determined, and it’s about 42%.
TCPR: How is that number determined, given that no particular gene has been associated with response?
Dr. Uher: While no single variant shows a signal that is strong enough to be predictive of response, the number of weak signals is larger than what would be expected by chance. There is a statistical procedure (called genome-wide complex trait analysis) that allows us to estimate from the numerous weak signals how large the overall genetic contribution will likely be. From these data, we believe a large portion of antidepressant response is genetically based, and eventually we will figure out a genetic test to predict that. But we’re not there yet.