In 1951, Hans Eysenck conducted the first modern twin study of neuroticism. Using the classical twin method, he estimated that the heritability of neuroticism was .81, and concluded, “neurotic disposition is to a large extent hereditarily determined.” In the seventies, when Arthur Jensen was galvanizing the genetics of intelligence, he usually worked with a heritability of .8.
This week, a preprint describing a massive meta-analysis of sibling GWAS dropped. The estimate of the direct genetic effect heritability of neuroticism was .081, down from Eysenck’s estimate by an order of magnitude. And scanning across behavioral phenotypes, this heritability was on the high side. Depression (depression!) was .015; drinks per week was .027; income was .024. These are heritabilities, mind you, upper limits on what we could expect any real world polygenic score to accomplish, and it would take millions of sibling participants to accomplish that. Cognitive performance was a little higher, at .188, for example, but the corresponding polygenic score accounted for essentially zero variance. An excellent summary and analysis has been posted by Sasha Gusev, here. Sasha argues (the fine points are over my statistical genetic head) that even these estimates are biased upward. He places the direct heritability of cognitive ability at closer to .12.
I don’t like it when hereditarians dance in the end zone when some positive genetic finding is published, and I am not here to declare some kind of environmentalist victory. In fact, I have always been supportive of the idea that the heritability of behavioral differences are more than trivially non-zero. My efforts have been about understanding the implications of the heritability of behavior, not trying to erase it. In all the writing I have done about the heritability of behavior, it never occurred to me to say that once we figured out how to estimate SNP heritability very very carefully, we would realize that heritability is much lower than we thought, barely different from zero for a wide range of behavioral phenotypes.
But people who are younger than I am have to project themselves back thirty years, when the great wave of twin studies was suggesting that at least half the variance in everything we cared about was “explained by” genetics. Jensen was in his heyday, declaring Headstart a waste of time and making his notorious racial claims. The Bell Curve was published. Closer to the behavioral genetic mainstream, I think of Thomas Bouchard, someone I have always disagreed with but have a great deal of respect for. Tom was (he is retired) a fierce proponent for the importance of heritability for understanding behavioral differences. (A taste of Bouchard giving the Galton lecture is here.) I remember his presidential address at BGA, putting up slide after slide of MZ and DZ correlations for ability, intelligence, and social attitudes, proclaiming, “Look at those heritabilities!” What would he say now? I can only imagine what the estimates of the direct heritability of social attitudes (which were his speciality) would look like in SNPs.
Again, I am not declaring victory here, because I am not sure what I want to say. Are we really ready to accept that once appropriate familial controls are in place, genetics is an unimportant factor in how many drinks people take every week? Because on Eysenckian or Bouchardian terms, that is what the data are saying. In the years between 1980 and now, how many review articles on the genetics of deepression or drinking or neuroticism or you name it have begun, “X is a substantially heritabile phenotype.” Does all that now need to be reconsidered? Given the imprecision we always work with in the behavioral sciences, I don’t really care if h2 = .02 is significantly different from zero— 99% of the time you could ignore it.
We are, once again, at a remarkable unmarked crossroads in understanding the genetics of behavioral differences.
In my personal opinion it is still important to figure out *why* there is a heritability gap:
-- If it is GxE (or AxC in ACE model language) inflating the twin estimates -- that's a lot of GxE! -- and we could imagine re-running all of the studies in Turkheimer & Waldron 2000 using polygenic score interactions to try to understand the interactive effect.
-- If it is gross EEA violations (I know this is an unpopular view) then we may need to re-evaluate some of the null findings from classical BG studies in twins, as we've effectively had a group of defiers present in the analysis this whole time. It also tells us something interesting about how parents (and society) treat kids when they look identical.
-- If it is a massive contribution of rare variation that GWAS is missing, that has major implications for population genetics and the way we think about selection on common traits.
Would it be fair to say that the gap between classic heritability estimates and direct genetic effects reflects the messy, complicated, and looping interactions between genes (one's own genes and family's) and environment? Cause it doesn't look like we can attribute all of it to "environment" alone in a manner that makes no reference to genes at all. I am not quite sure what's next given these revelations of direct genetic effects, especially since I am outsider to the field. But it's not clear to me that a direct h2 of 2% means that we can simply ignore "genetics." Maybe we can ignore the person's DNA for that purpose, but our scientific understanding of what's going on would still be referring to all the indirect genetic effects at the population level. No?