The phenotypes were corrected for age and sex, and standardized to z-scores in each adult and adolescent dataset separately. We used a two-tailed 90% Winsorisation33 to adjust the z-scores of four individuals in the adult dataset with absolute values greater than 4.17, the (100 – 5/3248)th percentile of the standard normal distribution based on Bonferroni correction, and combined the z-scores in both adult and adolescent datasets for the combined dataset of height (Supplementary Fig. 1e).
