Comparison of Adopted and Nonadopted Individuals Reveals Gene–Environment Interplay for Education in the UK Biobank

Polygenic scores now explain approximately 10% of the variation in educational attainment. However, they capture not only genetic propensity but also information about the family environment. This is because of passive gene–environment correlation, whereby the correlation between offspring and parent genotypes results in an association between offspring genotypes and the rearing environment. We measured passive gene–environment correlation using information on 6,311 adoptees in the UK Biobank. Adoptees’ genotypes were less correlated with their rearing environments because they did not share genes with their adoptive parents. We found that polygenic scores were twice as predictive of years of education in nonadopted individuals compared with adoptees (R2s = .074 vs. .037, p = 8.23 × 10−24). Individuals in the lowest decile of polygenic scores for education attained significantly more education if they were adopted, possibly because of educationally supportive adoptive environments. Overall, these results suggest that genetic influences on education are mediated via the home environment.

• Table S1: results from training education polygenic scores in 50,000 non-adopted individuals in the UK Biobank. • Table S2: full polygenic prediction results. • Table S3: full results from GxE (education polygenic score x adoption status) model. • Table S4: Comparison of adopted and non-adopted individuals for mean years of education per decile of polygenic score for education • Figure S1: genetic correlations between traits in LD Hub and years in education (plus standard errors) for adopted (red) and for non-adopted (blue) individuals. • Table S5: Genetic correlations with binary adopted/not adopted variable. • Figure S2: Comparing the variance explained in years of education by polygenic scores (PRS) for years of education, per year-of-birth grouping, for adoptees and non-adopted individuals. • Table S6: sample sizes by year-of-birth grouping.
Cheesman et al. Comparison of adopted and non-adopted individuals reveals gene-environment interplay for education in the UK Biobank Table S1: results from training education polygenic scores in 50,000 non-adopted individuals in the UK Biobank. Note: this p-value (10 -50 ) held for all of the standard set-values tested in PRSice (0.001,0.05,0.1,0.2,0.3,0.4,0.5,1    Cheesman et al. Comparison of adopted and non-adopted individuals reveals gene-environment interplay for education in the UK Biobank Figure S1:

genetic correlations between traits in the LD Hub database and years of education (plus standard errors) for adopted (red) and for non-adopted (blue) individuals. The figure only shows results for traits that showed significant genetic correlations with education in the sample of adoptees at p<0.05.
Due to the relatively small sample of adoptees, only 4 traits were significantly genetically correlated with education years in the adopted sample after correcting for multiple testing by using a stringent p-value of 0.000202 (0.05/247): years of schooling (2016), college completion, intelligence, and years of schooling (proxy). All traits in this figure also showed significant genetic correlations with education years in the non-adopted sample at p<0.000202, except height of females age 10 and males at age 12 (p=0.0004), Alzheimer's disease (p=0.0002) and difference in height between adolescence and adulthood (p=0.7265). None of the genetic correlations with education were significantly different between adoptees and non-adopted individuals at p<0.000202. education in the UK Biobank Note: These significant negative genetic correlations with educational attainment, age of motherhood, and poor physical health agree with evidence that adoptees are more likely than individuals from comparable origins to be born to young mothers, to have had a low birthweight, and to have received suboptimal obstetric care (Maughan et al. 1998), although the two groups were highly similar apart from these factors. The fact that adoptee status had a low heritability (6%; liability scale) means that interpretation of these genetic correlations should be cautious.
Cheesman et al. Comparison of adopted and non-adopted individuals reveals gene-environment interplay for education in the UK Biobank Figure S2: Comparing the variance explained in years of education by polygenic scores (PRS) for years of education, per year-of-birth grouping, for adoptees and non-adopted individuals. 95% CIs were obtained by bootstrapping with 1000 replications.
Any systematic differences between adoptive and biological families are likely to have changed over time. The variance explained by polygenic scores is more variable in the group of individuals reared by their biological relatives, but generally increases over time. The plot shows that prediction is stronger for adoptees in the youngest age group (although this increase was not significant). Due to increasing use of contraceptives across the 20th century, adoption became less related to young motherhood and more related to removal of children from high risk environments. Adoptions occurring later in the twentieth century were more likely to involve older children rather than babies (Maughan et al. 1998;Keating 2008). We speculate that these changes could mean that younger adoptees in our sample are less generalisable, more likely to have come from high risk environments, and more likely to have spent more time with their biological relatives, experiencing passive gene-environment correlation. However, other than the non-significant spike in the youngest age group of adoptees, polygenic prediction remains stable across cohorts (at ~4%). It could be argued that the variance explained by polygenic scores in adoptees provide a rough benchmark of 'direct' genetic influence remaining throughout periods of great social change.