Two questions. There are two main questions that we would like to answer with genetics in the field of epilepsy. First, are there genetic risk factors for epilepsies and if so, what are they? Secondly, are there genetic factors that help us understand how patients react to treatment, i.e. are there genes that predispose to response to antiepileptic drugs or that might be associated with side effects? While we have made much progress in answering the first question by identifying many epilepsy genes, there have been few answers for the second question, the field of pharmacogenomics. Now, a recent study in Human Molecular Genetics looks at potential genetic risk factors for the response to antiepileptic drugs in newly treated epilepsy. This is a study that needed to be performed and that we were waiting for.
What we can expect. Prior to discussing the study by Speed and colleagues, let’s quickly review what we can expect from a genetic study that looks at potential genetic risk factors for drug response. If we look at genetic risk factors for epilepsy, we know one thing for sure: There is no such thing as THE epilepsy gene. If there were a common and very strong risk factor for epilepsy, we would have found it already. What we have learned is that genetic risk factors for epilepsy are either strong and rare or common and weak. An example of strong and rare risk factors are de novo mutations in epileptic encephalopathies. These mutations have a very strong effect on the disease, but are very rare. At the other end of the spectrum are common genetic variants such as Single Nucleotide Polymorphisms (SNPs), which are common in the general population, but which might also represent mild risk factors for epilepsy. These variants typically increase the risk by a factor of not more than 1.5. Between both extremes are genetic risk factors that are called rare variants. Epilepsy-associated microdeletions represent an example of these rare variants. We have already learned that the identification of these risk factors is a statistical nightmare. In summary, the spectrum of genetic variants predisposing to disease that we can reasonably identify is quite limited.
The promise of pharmacogenomics. When we approach the genetics of treatment response, there are two main differences to conventional gene discovery, i.e. the hunt for disease genes. First, we don’t know whether the phenotype that we are looking at is genetic at all. While we use population studies or twin studies to assess the genetic burden for a given disease, we don’t have this opportunity for so-called secondary phenotypes like drug response. There is basically no epidemiological data powerful enough to give us an idea whether drug response in epilepsy runs in families or is more similar in identical than in non-identical twins. Therefore, if we embark on gene discovery in epilepsy pharmacogenomics, we are entering uncharted territory. The second difference is a more positive observation. The basic restrictions for the spectrum of disease genes does not apply to genes for treatment response. Genetic factors modifying a phenotype can be common and strong in principle. This observation has already been made for other phenotypes. For example, the odds ratio (OR) of a particular HLA genotype for liver toxicity due to the antibiotic flucloxacillin is ~45, and the OR of DDRGK1 variants as risk factors for side effects for interferon and ribavirin therapy for chronic hepatitis C is ~33. These effect sizes are unheard of for common genetic variants as risk factors for diseases. Basically, genetic risk factors modifying a disease can be common and strong and thereby be useful in clinical practice.
A GWAS for drug response. Speed and colleagues performed a genome-wide association study in newly diagnosed epilepsy, looking at a cohort of roughly 900 patients who were followed prospectively. Their cohort was split into a discovery and confirmation cohort for the genetic studies, which is common practice in the field. Genome-wide association studies investigate possible association between a phenotype and several hundred thousand common genetic variants (SNPs) spread across the genome. The phenotype in the study by Speed and colleagues was seizure remission after 12 months of treatment, which was present in 60-80% of patients included in their study. The authors identified three genetic markers that were associated with treatment response in the proximity of the PTPRD, ARHGAP11B and GSTA4 genes. None of these genes has previously been implicated in epilepsy and they do not represent intuitive candidates. PTPRD and ARHGAP11B may have some role in neuronal development, while GSTA4 encodes an enzyme involved in conjugation, one step of biotransformation that might be potentially relevant to antiepileptic drugs. Further analysis including clinical parameters and particular antiepileptic drugs did not reveal further information regarding additional candidate genes. All three genetic variants had an odds ratio between 1.7 and 2.7, which is relatively high compared to the known common genetic variants that predispose to epilepsy. However, the effect size of these variants fell behind the general expectations of “high risk” variants that might be immediately useful in clinical practice.
Prediction versus association. The publication by Speed and colleagues is the first study that takes a sufficiently powered approach to find genetic factors associated with response to AED treatment. While the study discovered three promising variants, their study could also exclude that major common genetic variants exist that explain a significant proportion of the genetic liability to treatment response. To a certain extent, this is a disappointing finding, suggesting that the genetics of drug response might be as complex as the genetics of epilepsy itself. However, looking for genes associated after stringent correction for multiple testing might be asking for too much. It would be interesting to see the predicitive power of a set of 10-100 genetic variants, a technique that does not necessary require that all variants are well-established risk factors by themselves. Likewise, it would be interesting to know whether the risks conferred by the three variants are additive, as the combined effect of these three variants may result in an almost 10-fold chance of treatment response, again approaching the range of values that may be clinically relevant. In either case, the study by Speed and colleagues has generated a unique dataset that may be explored further.