12 June 2014
By Andrew Brown
To read this blog in Spanish, please click here
Recently, two papers have been published which, for the first time, have shown in the context of gene expression that epistasis, a phenomenon where one gene has an impact on others, has widespread consequences in humans.
Epistasis in humans has always been controversial. Ronald Fisher in the 1930s had already decided it was unimportant and the arguments continue to today. Every month brings a new heated Twitter discussion between scientists about it: the model organism people (believers in epistasis) against the human geneticists (non-believers). Adding piquancy to these recent findings, Peter Visscher is one of the senior authors on one of these new papers; and he has previously been seen as one of the modern heirs of Fisher: “No epistasis on my watch!” has always been his rallying cry.
To put this debate in context, the human genetic code is around 3 billion base pairs long. While for any two individuals the code is mostly identical, there are tens of millions of places (or loci) where differences have been observed. With some of these loci, we know that what you have in your DNA can affect you in meaningful ways: how tall you are, your body mass index or your risk of disease (geneticists call these properties traits).
Searches for these loci in the DNA are called genome-wide association studies. And the assumption behind genome-wide association studies has largely been that these loci act independently, both of the environment in which they are found (see my previous blog post), and also of other loci (this is assuming that epistasis doesn’t exist).
Often it is difficult to know how a genetic locus acts: why does it make us taller? The activity of genes begins to provide an explanation. If we see the locus is near a gene, and we observe that it causes the gene to make more of the messenger molecule mRNA (this is known as gene expression), we can begin to understand how it acts on, for example, height.
Expression quantitative trait loci (eQTL) studies are genome-wide association studies where the trait, the characteristic influenced by the gene, is defined as gene expression. Because this trait is so close to the DNA, these eQTL studies have been very successful in discovering loci. This also raises the possibility that, when looking at gene expression, we have the ability to observe genetic phenomena that would require much larger samples to discover if we were looking at a more distant trait. This is why gene expression was the context in which we began to search for epistasis.
Epistasis is when genetic loci do not work independently but rather in combination: the effect of one genetic locus depends on what an individual has inherited at another genetic locus. We noticed that if a particular locus has a different effect on trait depending on other factors, then this could be seen as a larger than expected spread in values in the population that share this locus (because one individual could inherit a combination that increases expression, another that decreases expression, such effects would contribute to the larger spread). Therefore, finding such loci could be a first step to finding epistasis.
In our latest study, we uncovered 508 loci that had such effects. We used them to discover 256 examples of epistasis; and of these, 57 replicated in another sample. These examples of epistasis explained up to 16 per cent of variance in gene expression.
So where does this leave the debate about epistasis? It’s very difficult to argue with Peter Visscher’s original conclusion: from what the maths says and what we know about genetic architecture I can’t conclude that the missing heritability (what we call the fraction of the genetic influence that we haven’t been able to tie down yet) lies in this hitherto ignored epistasis. But missing heritability is about one aspect of our knowledge of genetics, our ability to predict outcomes. And very few of the genetic results we get excited about are good for predicting outcomes. The reason we get excited about them is not the potential for prediction, but because they increase our knowledge of how life functions.
My own belief is that the results we have found on how genetics works on gene expression will filter up to how genetics works on traits like height. And if we want to understand how we function, then there’s another layer of complexity that up until now we’ve all been ignoring.
Andrew Brown is a Postdoctoral Fellow in the Human Genetics group at the Wellcome Trust Sanger Institute where he works on the EuroBATS project.
- Brown A, et al (2014) Genetic interactions affecting human gene expression identified by variance association mapping. eLIFE. doi: 10.7554/eLife.01381
- Hemani G, et al (2014) Detection and replication of epistasis influencing transcription in humans. Nature. doi: 10.1038/nature13005
- Hill W G, et al (2014) Data and Theory Point to Mainly Additive Genetic Variance for Complex Traits. PLOS Genetics. doi: 10.1371/journal.pgen.1000008