Clinically Useful Genetic Markers of Antidepressant Response: How Do We Get There From Here?

Abstract

Predicting who will respond to an antidepressant—and who will not—remains anurgent but unfulfilled need in psychiatry (1). Clinical indicators of response have a long history, but probably have low predictive value. This state of affairs has fueled an energetic search for biomarkers. Genetic markers are ideal biomarkers, since they do not suffer from reverse causation. Reverse causation arises when abiomarkervalue is actually the consequenceof the outcome, rather than its cause. Genetic markers that represent the inherited sequence of DNA cannot reflect the result of a health outcome. The best genetic markers would identify individuals most likely to benefit from a particular treatment and least likely to suffer adverse events. Years of work with candidate genes have yielded some reproduciblefindings, but their individual effects are small, with little impact on clinical decision making (2). Studies that use genome-wide genotyping technology have implicated additional genes (3), but well-replicated findings that can help guide clinical decision making remain elusive. In this issue, Schatzberg et al. (4) report that common genetic markers in ABCB1 provide some information about which patients are most likely to respond to certain selective serotonin reuptake inhibitors.ABCB1 encodes P-glycoprotein, which influences the efficiency with which certain antidepressants (and other drugs) cross the blood-brain barrier. Common variants in ABCB1 have been associated with antidepressant outcome in some previous studies (5), so replication in an independent sample was the next logical step. Ten common markers in and around ABCB1 were tested in 683 patients with major depressive disorder whowere treated for at least 2 weeks and an additional 576 patients who were treated for 8weeks as part of the International Study to Predict Optimized Treatment in Depression (iSPOT-D) trial. The samples were carefully ascertained and evaluated, with prospective measures of efficacy and side effect burden, using standard instruments. General and emotional cognition were further assessed with a battery of 13 tests. Antidepressants included escitalopram, sertraline, and extended-release venlafaxine, all of which are known to bind P-glycoprotein. One singlenucleotide polymorphism (SNP) upstream of ABCB1 was associated with remission and subjective side effects, and the direction of the effect differed among the medications tested. How far does this take us, and what’s next? Treatment assignmentwas randomized, but not blinded, and therewere no placebo controls and no direct assessments of adherence. Despite these limitations, this study represents one of the few replications in the field of antidepressant pharmacogenomics, and is thus a real step forward. Still, the effect sizes are too small to be useful in clinical decision making. It seems likely that antidepressant response, like depression itself, is a polygenic trait, influenced by numerous genes, each of small effect. Therefore, many more genetic markers will be needed before we can achieve clinically significant prediction. This will require much larger samples than have been studied so far. We now know that genome-wide association studies (GWAS) require very large sample sizes to identify genetic markers of polygenic traits. Samples in the tensof thousandsarenowroutine,yetnoGWASofantidepressant outcome to date has exceeded 2,000. As pharmacogenomics begins to bear fruit in psychiatry, we can hope for a renewed focus on the large sample sizes necessary for robust findings. Theheritability of antidepressant response puts anupper limit onwhatcan be predicted by genetic markers alone. Heritability measures how much of the individual differences in a trait can be explained by inherited genetic differences. This is typically estimated from twin studies, but these are generally impractical for treatment response studies, since twins would only rarely be treated for the same illness in the same way during the time frame of a study. An alternative approach directly compares genetic similarity (estimated from a genome-wide set of SNPs) with phenotypic similarityusingsophisticatedregressionmethods(6).Onestudy used this approach to estimate that antidepressant response was about 40% heritable, but with a large confidence interval (7). Other statistical models show that traits with this degree of heritability cannot be predicted verywell fromgeneticmarkers alone(8).Evenapanelofgeneticmarkersthatexplainsaquarter of the genetic variance—more than anyGWAS todate—gives an area under the curve of only about 64%.Thus, a combination of clinical data and genetic and other biomarkerswill be needed if we are to push predictability into the clinically useful range. Advances in cancer chemotherapy may light the way. We can already tailor treatment for breast cancer based on a Studies like that of Schatzberg et al. are encouraging because they suggest that the ultimate goal of clinically useful outcome prediction can be achieved.