ESOC 2023 | Mendelian randomization shows a causal effect of valproate for ischemic stroke prevention
Ernst Mayerhofer, MD, Massachusetts General Hospital, Harvard University, Boston, MA, outlines his work looking into the association between valproate use and incident ischemic stroke. A number of observational studies have noted a decreased risk of ischemic stroke amongst valproate users, but confounding by indication bias prevents the determination of a causal association. To overcome this, Dr Mayerhofer used a mendelian randomization framework for assessing causal drug effects. Genetic valproate response scores were derived for over 2000 valproate users, based on drug response genome-wide association data. As the genetic scores were not known at the time of drug initiation, the data is essentially blinded by genetics. The team then compared stroke occurrence based on genetic score, to test if genetically predicted valproate response was associated with reduced stroke risk. As the genetic score is randomly distributed amongst the population, a causal association can be detected. It was demonstrated that a higher genetic score was associated with lower ischemic stroke risk in valproate users. The same relationship was not seen in valproate non-users, minimizing the risk of pleiotropic effects of the variants. Dr Mayerhofer comments on the promise of this approach for addressing many other clinical questions, such as the efficacy of a repurposed drug or the true association of rare side effects with the use of a given agent. This interview took place during the European Stroke Organisation Conference (ESOC) in Munich, Germany.
These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.
Transcript (edited for clarity)
So histone deacetylase 9 is a gene, HDAC9 for short, is a gene that has been the first that was associated with large artery stroke in genome wide association studies and it has been well replicated in every GWAS ever since. So people started to look into this gene and into the protein and found out that valproate, that is an old antiepileptic drug, non-specifically inhibits all HDACs. There are a lot of histone deacetylase proteins, they are involved in epigenetic modification, so in histone deacetylation as the name suggests...
So histone deacetylase 9 is a gene, HDAC9 for short, is a gene that has been the first that was associated with large artery stroke in genome wide association studies and it has been well replicated in every GWAS ever since. So people started to look into this gene and into the protein and found out that valproate, that is an old antiepileptic drug, non-specifically inhibits all HDACs. There are a lot of histone deacetylase proteins, they are involved in epigenetic modification, so in histone deacetylation as the name suggests. And so there has never been a trial and different people started to look into observational data and compared valproate users with non-valproate users or valproate users with users of other antiepileptic drugs. But the main limitation of such studies is that causal drug effects can only be shown in clinical trials where you randomize people into two different groups. From observational data the main problem is that you have a huge confounding by indication bias because people that use valproate, they get a prescription for valproate, they get this for a reason. And so epilepsy is common in stroke patients and people who had a stroke have a much higher risk for recurrent stroke than those without a stroke. So this confounding by indication, there are other strong examples, for instance statins for heart disease. If you look into who gets statins and compare people who get statins that protect them from heart disease and compare them to non-statin users and you will see that statin users have a much higher risk for myocardial infarction. So some of these studies suggest an increased risk of stroke for valproate and some of those that did it a little bit better with matching by risk factors and so on suggested a decreased risk. But really causality can only be shown in a trial. So these are the shortcomings of current data.
This is a framework that we have developed in the past and used for another purpose as well. And it’s a Mendelian randomization framework for assessing causal drug effects. So the concept is the following: if we have a drug response genome wide association study, that is a study that tries to discover specific genetic markers that are associated with better response to a certain drug. If you have that, if you have identified genetic markers, then you can construct and calculate a genetic score for every individual that you have genetic data and that predicts how well this individual will respond to a specific drug in this case valproate. So if we now have a cohort of drug users from observational data, if they have genetic data too, then for each of those individuals we can calculate this genetic score and each of those people will have a slightly different genetic score. And one assumption here is that at the time of the prescription, neither the physician nor the patient knew anything about their genetic score. So we do this after the data was collected. So effectively at the time of prescription, both the patients and the physicians were blinded to the genetic score.
And now as a thought experiment, if we look at the two most extremes, we have people that have very high genetic score that have they take a standard dose of the drug and they have a very high response to the drug. A huge effect. Okay. On the other very extreme are people that have a very low genetic score. They take a standard dose of the drug, and they have a very poor response to the drug, a poor effect of the drug. Now, because this is in principle now a blinded experiment, blinded by genetics, nobody knew at the time of prescription how the patient will respond to the drug and all of those got the drug. Now, if the genetic score is associated with the outcome of interest and we did this in this case for valproate for ischemic stroke. If people with a better genetic response to valproate have a lower incidence of ischemic stroke, then this must be a causal effect of the drug simply because there cannot be any difference, any bias in risk factors or age or dose by the genetic score because nobody knew the genetic score and the genetic score is randomly distributed in the population.
So we did that in the UK Biobank and I can just show the results now, and we found that the effect of the genetic score, so the association between the genetic score and the ischemic stroke, indeed had a hazard ratio of 0.75. So higher genetic predisposition to valproate response was associated with a lower risk of ischemic stroke and that was only shown in valproate users and not in valproate non-users, further confirming that this is a valproate specific effect. And if we look at the absolute incidence so this is follow up over almost 14 years and this is stratified by genetic score tertile, then we can see that the lowest tertile has the highest absolute risk for ischemic stroke and the highest genetic score tertile, so those who have the highest effect of the drug have the lowest absolute risk for ischemic stroke. But all of these people are in the in the therapeutic range for valproate.
There is a clinical trial now in Cambridge in the UK that tests valproate for recurrent ischemic stroke prevention. Valproate is not the best drug that we have. I mean, it’s a little dirty drug we have to monitor serum levels and it has a lot of side effects. So it’s not the drug that we want to use, especially for primary prevention. It could have a potential dual use for post-stroke epilepsy. But we know its mechanism of action pretty well, and probably the mechanism of action that prevents stroke is HDAC inhibition. So we could use this information to design a drug that inhibits HDAC for slowing of atherosclerosis.
So our framework I mean, this is an interesting result, but we’re really excited about the framework that we have also applied in the past to another question and hopefully many people will also apply this this concept. It’s really powerful and it can give you two insights. First, you can try drug repurposing in silico before running a trial with any patient if you have the data and this data is becoming more and more available. So for repurposing. And the other thing is if you want really to understand if rare side effects that are supposedly an effect of the drug, if they’re really caused by the drug because clinical trials are never powered for rare side effects, they are always powered for the primary endpoint. And you can ask this question using this framework. So all you need is a drug response GWAS, but there are thousands of them around and yeah, we’re very excited about the concept that hopefully gets reapplied by other researchers.
Read more...
