2After a decade of genome-wide association studies (GWASs), fundamental questions in 3 human genetics are still unanswered, such as the extent of pleiotropy across the genome, the 4 nature of trait-associated genetic variants and the disparate genetic architecture across human 5 traits. The current availability of hundreds of GWAS results provide the unique opportunity 6 to gain insight into these questions. In this study, we harmonized and systematically analysed 7 4,155 publicly available GWASs. For a subset of well-powered GWAS on 558 unique traits, 8we provide an extensive overview of pleiotropy and genetic architecture. We show that trait 9 associated loci cover more than half of the genome, and 90% of those loci are associated with 10 multiple trait domains. We further show that potential causal genetic variants are enriched in 11 coding and flanking regions, as well as in regulatory elements, and how trait-polygenicity is 12 related to an estimate of the required sample size to detect 90% of causal genetic variants. 13Our results provide novel insights into how genetic variation contributes to trait variation. All 14 GWAS results can be queried and visualized at the GWAS ATLAS resource 15 (http://atlas.ctglab.nl). 16 across a variety of traits and domains in terms of SNP heritability and trait polygenicity (see 42 Fig. 1). 43 44 Catalogue of 4,155 GWAS summary statistics for 2,965 unique traits 45
Extended DataWe collected publicly available full GWAS summary statistics (last update 23 rd October 46 2018; see Methods). This resulted in 3,555 GWAS summary statistics from 294 studies. We 47 additionally performed GWAS on 600 traits available from the UK Biobank release 2 cohort 48 (UKB2; release May 2017) 12 , by selecting non-binary traits with >50,000 European 49 individuals with non-missing phenotypes, and binary traits for which the number of available 50 cases and controls were each >10,000 and total sample size was >50,000 (see Methods, 51 Supplementary Table 1-2). In total, we collected 4,155 52
Supplementary Information 1 andGWASs from 295 unique studies and 2,965 unique traits (see Supplementary Table 3 for a 53 full list of collected GWASs). Traits were manually classified into 27 standard domains 54 based on previous studies 13,14 . The average sample size across curated GWASs was 56,250 55 subjects. The maximum sample size was 898,130 subjects for a Type 2 Diabetes meta-56 analysis 15 . The 4,155 GWAS results are made available in an online database 57 (http://atlas.ctglab.nl). The database provides a variety of information per trait, including 58 SNP-based and gene-based Manhattan plots, gene-set analyses 16 , SNP heritability 59 estimates 17 , genetic correlations, cross GWAS comparisons and phenome-wide plots. 60For the present study, we restricted our analyses to reasonably powered GWASs (i.e. sample 61 size >50,000), to avoid including SNP effect estimates with relatively large standard errors 62 (see Methods). By selecting a GWAS with the largest sample size per trait, it resulted in 558 63...