Genetic associations with BMI and weight cluster probabilities.

We fit regularised splines as detailed above to the 17,006 individuals with multiple observations of BMI and 17,035 individuals with multiple observations of weight from repeat assessment centre visits (Supp. Fig. 3). Soft cluster membership probabilities for these individuals were calculated, and the three logit-transformed π_i traits were carried forward for association testing with GWS variants associated with adiposity change in our discovery analyses. As above, we pruned out second-degree or closer related individuals and performed association analysis, adjusted for baseline trait, baseline age, (baseline age)², assessment centre, first 21 genetic PCs and genotyping array, via the linear regression framework implemented in PLINK¹²³. We repeated the same protocol within each self-identified ethnic group of individuals not of white British ancestry.