The primary analysis was conducted in each cohort separately using regression analysis, assuming additive genetic effects and accounting for dependence among family member when appropriate. We accounted for principal components where necessary in order to prevent population stratification as well as the assumption of homogeneity within samples of European ancestry. These results were gathered together and used to conduct a fixed effects weighted Z meta-analysis (to allow for differences in phenotype scaling across the participating studies) using METAL [53]. We have previously shown that the association between single-slice VAT and SAT measurements at the L4/L5 level is highly correlated with volumetric measurements (r = 0.95–0.99) [54]. This approach weights the signed Z statistics from each study by its sample size to obtain a weighted sum, from which a p value is obtained. We applied the genomic control correction to control type I error rates. SNPs that reached a meta-analysis P value≤5×10−8 were considered to be genome-wide significant [55].
