Thread by @arbelharpak

In work led by Carrie Zhu, we consider some puzzles that arise in evaluating autosomal gene-by-sex interaction (GxSex) in complex human traits.
We argue that polygenic GxSex may be pervasive but largely missed by current approaches.

biorxiv.org/content/10.110…

1/n

Should we polygenic effects to depend on context? In other organisms, where the environment can be controlled / manipulated, we know gene-by-environment interaction is the rule, not the exception. 2/n

What about sex as the context? A lot of traits are sexually dimorphic; but should we expect the causes to be sex-specific genetic effects? And to be found in existing, common genetic variation? Evidence from GWAS has been largely underwhelming. 3/n

We hypothesize that GxSex may be pervasive but largely missed if it acts primarily through sex differences in the magnitude of the effect of many genetic variants (“amplification”), rather than differences in the identity of causal variants or the direction of their effect. 4/n

(throughout this thread, “female” or "male" are shorthands referring to people in the UK Biobank with XX or XY chromosomes, respectively. This approach entails some misclassification with respect to sex and moreso with respect to gender.) 5/n

The classic statistical approach for searching for genetic interactions is to search for loci with large interaction effects. With this approach, the multiple hypothesis burden will kill polygenic amplification signals. 6/n

More recently, for complex traits, the focus has been on using genetic correlations as a litmus test for whether there are differences in genetic effects among sexes / environments / genetic backgrounds. 7/n

But correlations are scaleless and can therefore miss what we call “amplification” signals entirely. Consider the extreme of systematic amplification: All genetic effects are multiplied by a constant in one sex / environment compared with the other. Genetic correlation = 1.
8/n

Genetic correlations between males and females are typically close to 1.
9/n

A first piece of evidence that amplification may be pervasive is that, for many of the traits we examined, the effect of a polygenic score on trait values is significantly larger in one sex compared with the other. 10/n

But we also saw other patterns that suggest that GxSex can show up in multiple forms (perhaps reflecting different pathways) for the same trait. This motivated us to quantify m-f genetic covariance directly, and model it as a mixture of genetic covariance relationships. 11/n

The mixture weights we estimate tell stories. Testosterone is a good example to start with, thanks to a series of papers (e.g., by @technopolymath, @elenabernabio, @emflynn4 and their colleagues)
elifesciences.org/articles/58615
nature.com/articles/s4158…
nature.com/articles/s4143…
... 12/n

...which set our expectation for the comparison of M and F effects on testosterone. We can decompose these expectations in terms of relative magnitude & correlation. 13/n

Magnitude:
The majority of variants should have much greater effect in males.

Correlation:
We expect a class of genetic effects acting through largely independent pathways alongside a smaller class of effects via shared pathways.

Our mixture weights fit these expectations.

While testosterone is unusual in several ways, a consistent pattern (about half of the traits we analyzed) was amplification in one sex. e.g.: We estimated 69% of genetic effects on hip circumference to be larger in females than males, compared with only 6% F<M effects.

Across traits, the amplification signals we estimate correlate strongly with phenotypic variance.
This is consistent with our hypothesis of amplification directly translating to sex differences in genetic variance, and therefore in phenotypic variance. 16/n

One mechanism that may underlie GxSex is a cue or exposure that modulates the magnitude of genetic effects, and varies in its distribution between the sexes. 17/n

A natural candidate is testosterone. We find subtle evidence that T indeed modulates the magnitude of effects on traits, but, interestingly, in a sexually-antagonistic manner. 18/n

This also made us wonder whether the same modulators that act on the magnitude of genetic effects act on environmental effects. 19/n

One way in which this could happen: Genetic and environmental factors both regulate a core gene, whose effect size is modulated in a sex-specific manner. 20/n

Consider the example of skeletal muscle. Resistance exercise can be thought of as an environmental effect as it upregulates multiple skeletal muscle genes (like IGF-1) which are in turn is involved in muscle growth. However, after resistance exercise at similar intensities, ...

...upregulation is sustained in males, while levels return sooner to the resting state in females. It is plausible that sex-specific modulators (like T) set the magnitude of effect of core genes like IGF-1 and thus of both its genetic and environmental regulators. 22/n

Finally, we discuss the implications of pervasive polygenic GxSex for natural selection. Despite admittedly more tentative empirical conclusions in this part, this is one of my favorite parts of the work; so I’ll try and dedicate a separate thread to this on a different day. 23/n

Work was led by Carrie Zhu, who started it as an undergrad (!) and the very 1st person to join my lab. I'd like to say how proud I am of her accomplishment; but mostly I'm just lucky; not sure myself how she went from no prior experience in the field to this, and within 1 year.

Carrie will be presenting a poster on her work at #BOG22—please check it out if you are around!

Co-authors @jmillercole, and tweeterless Matthew Ming and Mark Kirkpatrick made key contributions. We also got very helpful feedback from very generous colleagues. 25/25