Why Mendel tells you not to drink alcohol when you are conceiving an Einstein

Posted on July 24, 2014


pregnant_woman_drinking_and_smokingDrinking alcohol during pregnancy is generally considered as not a very good idea. For one, it can cause fetal alcohol syndrome if you are serious about your drinking (reference). It has been associated with smaller babies. And smaller babies is not good (for the babies themselves, presumably for their mothers it is a benefit during labour that is hard to quantify for a man) because a Professor Barker tells us it leads to increased risk of a number of diseases at later stages of life. Because he came up with this, he got the hypothesis named after him – indeed, the Barker hypothesis (link).

In addition to these, alcohol use during or prior to pregnancy has also been linked to negative effects on cognition and school performance. As with other effects caused by alcohol however, there is the difficult aspect that this seems not to be a straightforward association in which a higher consumption of alcohol leads to a higher risk of whatever it was you were interested in, but instead that drinking some alcohol is actually good for you (or the baby in this particular case). So yeah, that’s a bit harder to explain.

We could, of course, go with the assumption that this is in fact true. And a little alcohol is good for the cognitive development of the apple in your eye, the heir to your throne and the receiver of your fortune. That’s good news and we all hope that that is true (for a number of other diseases as well). In fact, this is both bad “I could  join in the pub, again, because scientific evidence has shown that a little alcohol is good for you” and good “I should stick to the one beer on a school night, because it is harmful if I drink more (and probably not such a good idea for tomorrow’s 9 am meeting)”.

We could also assume that this is wrong, and that there is something else going on that may explain this apparent beneficial effect. The most likely explanation for this is residual confounding; the effect is really explained by another, unmeasured factor that also causes the disease (or reduces cognitive development) and is correlated to the exposure (eg drinking a bit of alcohol during pregnancy), or the use of alcohol or that of one of the included confounding factors is measured wrongly (biased) in such a way that it suggests that a little alcohol improves cognitive development; even though this is not the case. For example, children of affluent parents (better known as ‘rich kids’) generally show better outcomes for a variety of outcomes; especially those related to education. This is odd, because affluent parents are more often light-to-moderate drinkers than poorer parents; so their children should have worse cognitive development and should do worse at school.

If our second hypothesis is correct that is….




Alcohol consumption is notoriously difficult to measure. It is generally collected from self-reported consumption, and you can imagine that questions like “how often did you drink more than 1 unit in the last month, and between 2-4 units, etc…” is going to cause some problems. What, for example, do you do when you cannot quite remember how much you drank….other than that you were very likely in one of the higher categories last Saturday? Do you over-estimate….because you are 18 and alcohol is cool…or do you under-report because, let’s face it, the people asking do some kind of health study? Generally speaking, people will choose the second of the two options. Just to be on the safe side and don’t come across as irresponsible, or an alcoholic. Besides, by under-reporting you won’t run the risk of having to go to the ‘alcohol-support nurse’ or have a chat with your GP about your behaviour….so just to be on the safe side, you’ll under-report quite a lot more: You never know what value triggers a care-pathway! You get my point. So to get back to the story; you can imagine that this under-reporting (or even non-reporting of alcohol use) can be even larger for pregnant women than for the normal population, because we know it is harmful for the baby…everybody told you this…and even the Daily Mail says it (fact!), so it must be true (and you knew it as well when you took that glass of white wine, so better not report that). So how do you get around this when you are researcher doing a study to look at the effects of alcohol use during pregnancy on children? That’s not easy…

…but some of colleagues here in Bristol came up with a clever idea; they used what we know about the colour of pea plants (link). More specifically, the used Mendelian randomization (link) to get a marker of alcohol use that is not dependent on the reporting of the mothers.



Actually the idea is not only clever, it is also quite simple. What if you had an error in your gene that did not enable you to drink alcohol (you would, for example, feel so awful after a single small sip of beer- like a massive hangover, but one that got you straight into intensive care [like a peanut allergy])? Chances are you would never drink alcohol again…or if you are particularly hard-headed you would stop after trying it twice. So anyway, by the time you are 17 you would have stopped drinking. Now here is the trick…you would inherit that gene from your parents, but because this inheritance is only affected by the genetic material your mum and dad gave you, it is not influenced by anything else. So the great thing about this is that those rich kids mentioned above have the same chance as the poor kids to have that gene! And similarly, if you have a really healthy diet (you did eat all the green stuff on your plate when you were 5), which also has an effect on cognitive development and school performance, this should not cause your genes to change; or alternatively this gene should not make you eat more green stuff…because it was involved in alcohol, not greens, remember (although we’ll get back to this one at the end). So the beauty of this idea is that regardless of all the residual confounding and mis-reporting of alcohol use that may have affected your original study, this is now not that important anymore because you now have someone’s actual exposure to alcohol from knowing if someone has that particular gene that makes them ill when they drink.

So now the good news: this genetic variant does indeed exist (for those interested its an ADH1B variant (rs1229984))!

And then the bad news: although working by the mechanism described above, and shown in other studies to be predictive of lower use of alcohol during pre- and early pregnancy (see the references in the paper), it isn’t actually as strong as I described. So although having that variant leads to lower alcohol use because of side effects, that isn’t the same as having such bad side effects that you wouldn’t drink…you’d just drink less.

Still, if you would compare reported drinking in relation to an outcome measure between those with and without the genetic variant, you should see a difference in outcome compared to what you would see if you didn’t look at the genetic variation and grouped them all together. That is, and this is the crux of the story, if the genetic variant caused women to drink less during pregnancy (and is not related to other factors as discussed above), and drinking less alcohol was then in turn associated with a better outcome.  Or, in cool-speak:

We know A->C and we know A->B, so therefore B->C



And this is exactly what my colleagues did. The paper can be found here (link). And indeed, as shown in previous studies they showed that the more alcohol a mother drank (as reported by herself) before and during pregnancy, generally speaking the higher the child’s IQ at age 8 years and the better the school performance at age 11 (the only exception being that if the mother reporting drinking more than 7 units per week on average in the first trimester, that this was not such a brilliant idea).  Nothing new so far…

…but now add in this genetic variant that was identified. They compared IQ and school performance for the different levels of reported drinking between the carriers of the variant and those that had the “normal gene” and suddenly the IQ score at age 8 was not associated with alcohol consumption in the first trimester anymore. For the school performance at age 11, the results were even more interesting and showed better performance for children of those with the genetic variant, with the difference between the groups larger for higher self-reported drinking. In other words, these clever analyses at the same time indicated that (a) drinking alcohol in the first trimester is in reality not associated with improved IQ and school performance (if anything, it probably makes it worse), and (b) that women who drink during pregnancy under-report the quantity they drink, and they under-report more when in reality they drink more.


…caught out by science. How embarrassing!


There are a number of “buts” and “ifs” to this approach, but I will leave it to you to read the paper itself if you want to know more about this. If not, just remember not to drink when you are trying to get pregnant or during pregnancy, and you don’t need to worry about the validity of the study and about the use of this particular genetic variant to accurately measure differences in alcohol use (eg. the validity of the instrumental variable, in stats-speak).

What is quite important though, is that indeed the distribution of this genetic variant in the population does not depend on any confounding factors. This makes sense…why would it?

Just to be on the safe side, the authors have looked at this too, and I have copied part of the table from their paper, but I have added a column (the last one):


So as rightfully concluded by the author, there are no statistically significant differences between carriers of the genetic variant and the non-carriers for any of the confounding factors. I haven’t put the confidence intervals in, but the p-values reflect this.

Anyway, just for interest I added the column at the end, and although none of the individual factors statistically significantly differ, overall you may see the large number of “yesses” in that column. This may just be because of random variation (or noise), but it does suggest that the carriers of the genetic variant were just a little bit healthier than the non-carriers. The differences are not that large, but neither is the statistically significant better school performance of the children at age 11 of the carriers compared to the non-carriers (~2% better). So maybe this explains the observed benefit of non-drinking, and there really is no causal effect of drinking on cognitive development? More research…..etcetera.

But how interesting. If this is not just random variation that just by chance suggests a healthier lifestyle, than how can this be explained? Since as the authors describe, it seems very unlikely that your genetic make-up is determined by your lifestyle (it should be the other way around), but then how does having this rare allele, that leads to reduced alcohol use, also have some other beneficial aspects? Or, and this is probably more likely, people drinking less alcohol are in general a bit more health-conscience…maybe mediated through better education?

Excuse the pun, but “Food for thought”…


Posted in: Public Health