Harrison I. W. Wright

Most heritability studies have focused on common genetic variation and have largely excluded non-European or highly admixed populations. Here, we estimated the whole-genome heritability (h ² _WGS ) of seven complex traits using 73,600,953 autosomal and chromosome X variants for 198,931 unrelated individuals from three genetic similarity groups. Adjusting for population stratification, we showed that rare variants (minor allele frequency2 _WGS . We used simulations to show how functionally-stratified heritability can be inflated due to linkage with missing or lower-quality variants. Finally, after accounting for confounding effects, we found substantial heritability enrichment from variants in non-coding RNA exons (maximum fold enrichment, FE=8.7, P =3.56e-7), UTRs (23.0, P =3.46e-30), distal regulatory regions (33.8, P =3.26e-42), and protein-coding exons (58.0, P =7.78e-15). Our study provides a framework for investigating the genetic architecture of complex human phenotypes using diverse whole-genome sequencing data and highlights the contribution of rare and non-coding genetic variants to the heritability of complex traits.

GWAS have generally focused on common variants from genotyping arrays or rare protein-coding variants from exome sequencing. Here, we use whole-genome sequencing data to evaluate the contribution to and architecture of rare non-coding variants for three commonly studied anthropometric traits: height, BMI and waist-hip ratio adjusted for BMI. Analysing 447,461 individuals in the UK Biobank for discovery and 225,515 individuals in All of Us for replication, we identify 90 rare and low-frequency single variant associations, including two independent rare variants upstream of IGF2BP2 that substantially reduce waist-hip ratio adjusted for BMI, but have distinct effects on other adiposity traits. We further identify 135 coding variant aggregates. For example, UBR3 protein-truncating variants are associated with a 2.7 kg/m2 increase in BMI. We additionally identify 51 non-coding variant aggregate associations, including one in the 5’UTR of FGF18 associated with up to 6 cm effects on height. We show that 97% of rare variant associations occur near GWAS-identified loci, demonstrating convergence of rare and common variant associations. Finally, we show that ultra rare variants explain a small fraction of heritability compared to common variants for these traits, that heritability is largely shared across ancestries, and that it concentrates around common variant loci.

Whole genome sequencing in diverse population-scale biobanks offers new insights into the genetic architecture of complex traits from rare and non-coding variants. However, rare single variant and aggregate associations are often confounded by linkage disequilibrium and haplotype structure, resulting in large numbers of false-positive associations. Previous methods that rely on reference panels or linkage disequilibrium-matrices to determine conditional independence in meta-analyses do not scale to very rare variants, which may be observed in only one biobank and can exhibit long-range haplotypes. Here, we implement a federated approach to perform iterative conditional meta-analysis on individual-level genotype and phenotype data across biobanks while adhering to data sharing policies. We applied our methodology to a meta-analysis of LDL-C in 614,375 individuals from UK Biobank and All of Us, encompassing six genetic ancestry groups. After conditioning, only 4.3% of significantly associated rare single variants and 6.9% of aggregates remained statistically independent. The proportion of significant aggregates that remained independent after conditioning was higher for coding-based tests than non-coding. We further validate that our approach effectively suppresses false-positive associations using simulations centred on the LDLR locus. We identify allelic series of variants associated with reduced LDL-C, including loss-of-function variants in DNAJC13 and variants in the 3-prime untranslated region of LDLR . Our results highlight that federated conditioning can distinguish independent rare variant signals from linkage and haplotype structure artifacts in multi-ancestry meta-analyses across separate biobanks.

Population-scale whole-genome sequencing data will contain many individual-level genotype errors, even after allele-level quality control (QC). We establish the need for genotype-level QC using UK Biobank (N=490,726) and All of Us v8 (N=414,830), where we remove up to 100 million (∼9%) additional low-quality variants. We demonstrate reduced false positive rate in downstream genetic association studies, highlight the power of parent-offspring trios for QC, and illustrate the need for sex-specific X-chromosome filtering. We provide a QCed All of Us v8 dataset in plink -pgen format, and an efficient pipeline for QC and conversion from VCF to plink- pgen for UK Biobank.