Multivariate genetic architecture reveals testosterone-driven sexual antagonism in contemporary humans

Abstract

Sex difference (SD) is ubiquitous in humans despite shared genetic architecture (SGA) between the sexes. A univariate approach, i.e., studying SD in single traits by estimating genetic correlation, does not provide a complete biological overview, because traits are not independent and are genetically correlated. The multivariate genetic architecture between the sexes can be summarized by estimating the additive genetic (co)variance across shared traits, which, apart from the cross-trait and cross-sex covariances, also includes the cross-sex-cross-trait covariances, e.g., between height in males and weight in females. Using such a multivariate approach, we investigated SD in the genetic architecture of 12 anthropometric, fat depositional, and sex-hormonal phenotypes. We uncovered sexual antagonism (SA) in the cross-sex-cross-trait covariances in humans, most prominently between testosterone and the anthropometric traits - a trend similar to phenotypic correlations. 27% of such cross-sex-cross-trait covariances were of opposite sign, contributing to asymmetry in the SGA. Intriguingly, using multivariate evolutionary simulations, we observed that the SGA acts as a genetic constraint to the evolution of SD in humans only when selection is sexually antagonistic and not concordant. Remarkably, we found that the lifetime reproductive success in both the sexes shows a positive genetic correlation with anthropometric traits, but not with testosterone. Moreover, we demonstrated that genetic variance is depleted along multivariate trait combinations in both the sexes but in different directions, suggesting absolute genetic constraint to evolution. Our results indicate that testosterone drives SA in contemporary humans and emphasize the necessity and significance of using a multivariate framework in studying SD.

Keywords: B matrix; additive genetic (co)variance matrix; genetic correlations; human complex traits; sex difference.

Fig. 1.

G_mf matrix in males and females. (A) Schematic showing G_mf genetic covariance matrix and its constituents G_m, G_f, B, and B^T for two traits in males and females. (B) G_mf correlation matrix for all the 12 traits in both the sexes. The male cross-trait genetic correlations (G_m/r_ctM) are on the top left and female cross-trait genetic correlations (G_f/r_ctF) are on the bottom right. The cross-sex genetic correlations (r_mf) are along the red diagonals of the top right and bottom left squares of the B and B^T matrix. The off-diagonal genetic correlations in the B and B^T (red squares) are the cross-sex-cross-trait genetic correlations (r_mf^ct). Genetic correlations which are not significantly different from zero are marked with crosses, except the diagonal of the B and B^T which are tested for significantly different from 1. The size of the yellow and purple squares is according to the strength of the correlations.

Fig. 2.

Genetic correlations of all 12 traits with testosterone and CBAT. (A and B) cross-trait genetic correlations of testosterone and CBAT with anthropometric and fat depositional traits in males and females (r_ct), respectively. (C and D) Cross-sex-cross-trait genetic correlations of testosterone and CBAT with other traits (r_mf^ct). (E) Cross-sex genetic correlations (r_mf) of all 12 traits.

Fig. 3.

Random selection skewers and genetic constraint to the evolution of sexual dimorphism. (A) Angle between the predicted responses to concordant selection vectors (Δz_m and Δz_f) in males and females, respectively, constraining B to be zero, and after the inclusion of B (Δz_mB and Δz_fB)_. The thin lines depict 1,000 response vectors, and the thick arrow is the median response. (B) Similar angles between male and female predicted responses before and after inclusion of B, but to antagonistic selection vectors. Here, B constrains the male and female evolutionary response vectors to evolve close to each other by reducing the angle between them. The length of the vectors is just for representation purpose.

Fig. 4.

Venn diagram showing polygenic overlap between testosterone, TFR, AFR, and other anthropometric traits (height, weight, WC, and HIP) in males and females. The number of variants is shown in thousands (with SE). Mauve circles depict unique variants for trait 1, and the brown circle depicts that of trait 2. Shared variants between two traits are shown in gray. r_g depicts the genetic correlation between traits from the MIXeR output. Shades of red denote positive genetic correlations, whereas shades of blue denote negative genetic correlations. Panels represent shared variants between anthropometric traits and (A) testosterone, (B) TFR, and (C) AFR.

Fig. 5.

Genome-wide significant SNPs between testosterone and BMI in males and females. (A) Genome-wide SNPs are plotted as colored horizontal lines along chromosomes (X axis) at specific positions depicted in mega-base-pairs (Y axis), which are significant in 1) both in BMI and testosterone, 2) only in BMI, and 3) only in testosterone. The vertical black lines represent the length of the specific chromosomes. The SNPs to the right of each black line are for males and that to the left of each line are for females. The red horizontal lines show the SNPs that reach the genome-wide significance in both the traits (class 1). (B) LocusZoom plots showing variants around the SULT1A1 gene on chromosome 16 for male and female testosterone. The labeled SNPs are 3 of the variants that are identified as shared causal variants on chromosome 16 between testosterone and BMI in males in the colocalization analysis. The same variants are way below the significance threshold for female testosterone. The violet diamond is the lead SNP in the region.