Thread by @molly_przew

I keep reading versions of the statement that genetic differences within a human population have no implications for genetic differences between populations, alongside strong statements about what we know about the causes of differences within a population. 1/n

It got me thinking about how that distinction is not well defined, for many reasons: principally because (i) “populations” are an abstraction and not a real thing and (ii) that abstraction hides huge amounts of environmental variability… 2/n

...that correlate in unknown ways with genetic ancestry. Also bc (iii) it’s often unclear if people are talking about direct genetic effects only, or all the signals that can be picked up in a genome-wide association study (GWAS)
3/n

Consider the claim that polygenic scores (PGS) can be useful (if noisy) predictors within a population but don’t “port” well to other populations. What is a population here? 4/n

Population genetics offers operational definitions of populations with which we model genetic variation: a set of individuals that draw their allele frequencies from the same set of distributions, say (e.g., genetics.org/content/155/2/…). 5/n

But most of the time, we don’t mean something all that precise. As the quip goes “An academic discipline is a set of individuals who agree not to question the same assumption” For population genetics, that assumption is a population. 6/n

In the real world, the distinction between “within” and “between” populations is illusory. There is a giant family tree of humanity, and people who share more ancestral paths through it than others, and more similar environments than others. 7/n

We can try slicing and dicing it into bins, but in fact, as @magnusnordborg writes: “Natural populations are never homogeneous, … 8/n

and it is therefore misleading to imply there is a qualitative difference between ‘within-population’ and ‘between-population’ comparisons…” elifesciences.org/articles/45380 9/n

What people often mean to emphasize, I think, when they make the distinction “within” vs “between population” is something different: It’s that learning about genetic effects on human traits is incredibly difficult, because humans grow up & live in distinct environments. 10/n

And between-group comparisons are often hopelessly confounded by these environmental differences. 11/n

Quoting @magnusnordborg again: “With respect to confounding by population structure, the key qualitative difference is between controlling the environment experimentally, and not doing so… 12/n

…Once we leave an experimental setting, we are effectively skating on thin ice, and whether the ice will hold depends on how far out we skate.” 13/n

Since humans are never compared in an experimental setting, nothing ensures that environmental effects are identical across groups. 14/n

Critically, however, the limits imposed by not controlling environmental effects also apply to comparisons among individuals within the study group, not just between groups. Hence the problem of environmental confounding in GWAS: science.org/doi/pdf/10.112… 15/n

So it’s not about whether comparisons are made within vs between hypothetical populations. It’s about all the confounders that exist in the absence of the ability to do experiments and how well we can measure and control for them. 16/n

These considerations motivate the recent turn from a standard GWAS--which picks up indirect effects, assortative mating, and population stratification effects—back to family designs: science.org/doi/full/10.11…. 17/n

Family designs such as sib GWAS can help tease apart direct genetic effects from all these other factors. (Precisely because family designs only include direct genetic effects, their prediction accuracy can be much lower than that of standard GWAS.) biorxiv.org/content/10.110… 18/n

The fuzziness around the concept of population also has other implications, for example for how we interpret heritability, a concept that is defined within a population. 19/n

Evolutionary and quantitative geneticists learn about heritability as informing the expected increase in a trait in response to selection in one generation. 20/n

Applied to humans, heritability estimates allow us to ask how the current variance in a trait is partitioned under some model, say how much of the variance in some trait maps to genic vs intergenic regions. 21/n

Without controlled environments or experiments, heritability estimates are not informative about causes of trait variation, as Lewontin pointed out long ago: ncbi.nlm.nih.gov/pmc/articles/P… 22/n

Moreover, the model under which heritability is estimated is one in which genetic and environmental effects are assumed to be separable (in addition to which there may be a specific interaction term). 23/n

Whether that model is truly applicable to traits that arise mostly (entirely?) from interactions with the social environment, such as behavior, is not clear to me. 24/n

So a population is a convenient abstraction that can obscure a key issue: When interpreting trait differences, we are forced to make comparisons among humans who experience different sets of environments without being able to control for them well. 25/25