Too much scientific research turns up in the media as hyped, misleading, or just plain wrong. As scientists we are often guilty of exaggerating the importance of our work. Here is a short blog where we try and balance things up with some thoughts on latest claims. We try and stick to stories on which we are qualified to comment, such as those related to diabetes or genetics.
Height, BMI and socioeconomic status: a Mendelian randomisation study in the UK Biobank
Below you can find more illustrations and data from our paper in the British Medical journal “Height, BMI and socioeconomic status: a Mendelian randomisation study in the UK Biobank”, including the principle of Mendelian randomisation and addressing the potential caveats to our interpretation. For more details on Mendelian randomisation, visit the experts’ website at the MRC Integrative Epidemiology Unit website just up the road in Bristol. You will note that we randomly spell randomisation with a z or s.
This diagram shows the principle and why it is analagous to a randomised controlled trial. A version of the FTO gene that makes people a little fatter occurs in 40% of human genomes. The alternative version, which makes people a little thinner, occurs in the other 60% of human genomes. There is therefore a 40% chance that you will inherit the “fat” version of the gene from one parent, and a 40% chance you will inherit it from your other parent. This random process means 16% of people (40% x 40%) will be a little fatter for no other reason than the two copies of the “fat” version of the gene they inherited. Likewise, 60% x 60% = 36% of people will be a little thinner. Everyone else has one copy of each version of the gene, and their body mass index is , on average, in between that of the other two groups. All of this means that any one study will consist of 3 groups of individuals randomized to 3 different groups based on their genetic BMI.
We can then use the genetic variation to ask whether or not a higher BMI leads to things like higher glucose levels and higher cholesterol levels. And sure enough it does, and that fits with what we already knew. Being a little fatter causes your blood sugar and cholesterol to go up a little. Likewise your blood pressure and many other things and the genetic randomisation test confirms that.
But now we can be more sophisticated. Instead of using one genetic variant in the FTO gene, we can use 70 because last year a paper was published describing these genetic variants that alter BMI. Some of these variants have stronger effects on BMI than others, and we can use this to plot genetic effect on BMI against genetic effect on an “outcome” trait. Here is an obvious proof of principle example using data from the UK Biobank. It shows the effects of the 70 variants on BMI on the X axis, plotted against their effects on type 2 diabetes on the Y axis. You can see very nicely how the larger the effect of the genetic variant on BMI, the larger effect on type 2 diabetes. Because the X axis is based on genetic BMI only, it means the line we draw between the points has not been influenced by type 2 diabetes leading to higher BMI, or confounding factors, such as poverty and lack of education leading to both higher BMI and type 2 diabetes. Instead we can say that higher BMI leads to type 2 diabetes – it is a cause, in the same way we can only truly claim that a new drug works by performing a randomised controlled trial.
So here is the new bit. Instead of testing BMI genetic variants against something obvious like type 2 diabetes, we tested them against measures of your socio-economic status.
This allowed us to take away one of these arrows and just ask “is there a link from higher BMI to altered status, as measured by things such as your income?”.
Here is the randomisation experiment again, shown with just the FTO gene variant again for clarity. In fact we used this genetic variant and another 68.
Here is the plot of the 70 BMI genetic variants. BMI against household income. It is not very interesting. There is a trend from higher BMI, as measured by your randomised genetics, towards lower income, but it is far from certain. This is using all 120,000 people we had available from UK Biobank
And here is where it becomes interesting. There is no effect in men. But in women, when we plot the effects of their genetic variants on BMI, we see that on the Y axis, there is a clear association with lower income. Being an overweight woman, on average, leads to a lower income.
Now, we address some of the concerns.
Genuine positive associations exist between BMI raising alleles and higher BMI, and lower wealth and higher BMI.
If a study is depleted of people of higher BMI relative to the general population (as UK Biobank is likely to be), then this induces a false association between BMI raising alleles and higher wealth This is the opposite to what was observed in Tyrrell et al, where BMI raising alleles were associated with lower wealth
Let’s assume that BMI and wealth & education have completely independent effects on the likelihood of participating in the study (unlikely to be true but they will be partially independent)
Let’s also assume that the UK Biobank healthy cohort effect means everyone in the study has a BMI below 30 (represented by the line). This is a gross exaggeration for illustrative purposes, but there will be a truncation effect compared to the general population.
Now look at the group of overweight people – those with BMIs 25 to 30 kgm2. Wealthy, better educated people in this range will have more BMI raising alleles than the less wealthy and less well educated.
BMI raising alleles thus become associated with higher education when we analyse everybody. This is the opposite to what was observed in Tyrrell et al, where BMI raising alleles were associated with lower wealth
Higher BMI may have resulted in lower wealth and education in the parents of the UK Biobank participants . If so BMI raising alleles and high BMI would be passed down through genes and environment, and lower wealth would be passed down through the parental environment.
If this effect was present, which seems highly plausible, then it still represents a causal effect from higher BMI to lower wealth, only in previous generations. However, obesity was much rarer in the first 70-80 years of the 20th century, when these effects would have had to have taken place, so any of these “dynastic” effects would have had to have reflected relatively lower levels of BMI, which is possible, but:
For a “dynastic” effect to have had a strong effect on our results, it would have had to have influenced daughter’s BMI more than son’s BMI.