A Speculative Explanation of the Missing Heritability Problem
1. Statement of the problem
Here are three incidents that describe the peculiar problem modern behavior genetics is facing.
The first involves me. In a post a week or two ago about racial-hereditarianism, I needed an example of a question on which I disagreed with the majority of the BG community. I somewhat randomly picked alcohol abuse, saying that while most BG members would say that genetics is crucial to understanding alcoholism, I would say it is relatively unimportant (with a parenthetical note that there are qualifications). I got immediate pushback on this, both here and on the bad site (which, yes, I still check occasionally to see what people are saying about me, especially with the book out). Do I really think genes are unimportant for alcohol use? The twin and adoption studies are pretty unanimous, after all, and even more: who doesn’t know a family in which multiple individuals, across generations, struggle with alcohol?
I agree. As I will make clear below, I don’t think the twin studies were wrong, I do know families like that, and I agree that the reasons are, at least in a very general sense, genetic. But there is another side to it. In modern SNP-based genomics, the heritability of alcohol use is usually under .1, and often substantially less than that. Polygenic scores for alcohol use basically don’t work at all. That fact has a very important implication: if you don’t have an identical twin or extensive phenotypic data about your first degree relatives, modern genomics has essentially nothing to say about your alcohol use. And using the phenotypes of your first degree relatives is kind of cheating, because your relationship with them is not purely genetic. Simply saying, “biological relatives resemble each other, and we assume that the basis for this resemblance is genetic” is just a solipsistic way of dressing up old-fashioned phenotypic behavioral science in genetic language. We have known to pay attention to family history forever.
My bottom line is that this is a real problem. How can twin heritabilities and SNP futility both be true? Two other anecdotes:
In another Bluesky thread, I got into a discussion with the anti-hereditarian theorist Jay Joseph.
Jay Joseph on My Book
One of the ongoing traditions in the nature-nurture debate is absolute rejection of the idea that genes have anything at all to do with human behavior. Back in the day there was Lewontin, Rose, and Kamin’s Not in Our Genes, for example. This point of view, I have always thought, was often more a reaction against hereditarian overreach than an accurate …
I presented Jay with a thought experiment that I often use with anti-hereditarians, and it involved alcohol abuse. Suppose you are planning to adopt a child, and you are informed that both of the child’s biological parents suffered from severe alcoholism. Would this be a cause for concern, and if so would any of the reasons for concern be genetic? I think the answer to both questions is clearly yes. Jay denied it, but I think he missed the point. Jay’s denial was based on the fact that it is hard to prove that the risk is genetic in origin, because in human beings genetic and environmental risks are always confounded. That’s true, but the question remains: do you really think the genetic risk is plausibly zero? I don’t.
The other story runs in the opposite direction. I was recently chatting with a friend who is a very prominent psychiatric geneticist. I was asking him/her about the recent meta-analysis of heritability within families of parents and siblings. The heritabilities that were reported were just alarmingly low. Depression, for example, had a heritability of .015. What did s/he make of that? Did it do anything to shake her/his confidence that depression is a largely genetic disorder?
No, my friend said, because Peter Visscher had published a couple of papers that showed where the missing heritability between SNPs and twins can be found. And yes, Peter Visscher has written a lot of papers about missing heritability, and I am not qualified to criticize them from the point of view of statistical genetics, but they don’t solve the fundamental problem. You can say, some of the heritability is to be found in rare variants, or epistasis, or whatever, but that still does not give you the ability to make meaningful predictions or explanations about individual people in the absence of twins or first degree relatives. It is a little like a search for a hiker lost in the wilderness. The head of search and rescue steps up to the podium and says, good news, we know where the hiker is: he is out there somewhere in the woods. That’s great, but let me know when you can actually find him.
This is the problem I want to offer an answer to: how can twin heritabilies of major phenotypes be larger than SNP heritabilities by an order of magnitude, and what is this trying to tell us about the relation between genes and complex behavior. Neither of the easy answers— the twin studies were just wrong, or it is just a solvable technical problem in molecular genetics— are going to get us there. I suggest we need a major shift in our thinking about how genetic differences become correlated with complex phenotypes. More soon.
As far as I'm aware the most direct paper from Visscher that touches on this question is Robinson et al. 2017 (https://pubmed.ncbi.nlm.nih.gov/28692066/) which localizes a significant portion of heritability to gene-environment interactions and contains fairly unequivocal claims: "Taken together, population studies imply that estimates of h2 for BMI are systematically inflated in classical twin studies." (in the introduction) and "Therefore, we suggest that genotype–age and genotype–environment effects may contribute to the inflation of BMI heritability estimates in classical twin studies, in combination with stronger common environment effects in close relatives than more distant ones." (in the conclusion). I think the field has generally not grappled with the implication of substantial gene-environment interactions and what this means for how we think about and predict complex traits.
I'd be interested in learning if there's a heritable overlap or synergy between alcoholism and diabetes (either type 1 or type 2.) "Alcoholism" may not be just one phenomenon, for that matter. Perhaps there's one variant with a link to impaired sugar metabolism, and another independent of that interaction.
Alcohol use as an example is particularly intriguing, because the "environmental" influence of social and cultural (including subcultural) factors is also widely held to be prominent. (I'd say that's indisputable.) Including legacy and historical factors: the acceptability of alcohol use in social environments and as a cultural practice is mutable over time, and there are examples to be found where a change in social attitudes has asserted itself quite swiftly and decisively.
Alcohol use is also a practice where dramatic examples can be found, ranging from social groups and cultures where alcohol is forbidden, to the opposite extreme of common resort to volume drinking of distilled liquor in a social milieu where the option of moderation is rarely if ever pursued, or even considered as an individual choice. There are also often notable culture-dependent differences in consumption habits between males and females.
So there's a lot of grist worth studying, both in terms of sampling population genetics and researching the relative importance of sociocultural factors.