Systematic single-variant and gene-based association testing of thousands of phenotypes in 394,841 UK Biobank exomes

Konrad J. Karczewski, Matthew Solomonson, Katherine R. Chao, Julia K. Goodrich, Grace Tiao, Wenhan Lu, Bridget M. Riley-Gillis, Ellen A. Tsai, Hye In Kim, Xiuwen Zheng, Fedik Rahimov, Sahar Esmaeeli

Read Systematic single-variant and gene-based association testing of thousands of phenotypes in 394,841 UK Biobank exomes by Konrad J. Karczewski, Matthew Solomonson, Katherine R. Chao, Julia K. Goodrich, Grace Tiao, Wenhan Lu, Bridget M. Riley-Gillis, Ellen A. Tsai, Hye In Kim, Xiuwen Zheng, Fedik Rahimov, Sahar Esmaeeli in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc9903662.txt.

Introduction

Karczewski and co-authors report systematic association tests across thousands of traits in UK Biobank exomes. Genome-wide studies found many common variants, but rare coding variants stayed less explored. Exome data from population biobanks allow a broad scan of rare variant effects. The team analyzed 394,841 people with exome sequence data.

They ran single-variant and gene-based tests for 4,529 phenotypes. Discovery tracked closely with variant frequency. It also tracked with measures of harm and natural selection. The authors highlight biological findings that the data reveal. All results are released as a public resource. A Genebass browser lets users explore rare-variant findings quickly. The work helps link rare coding changes to health and disease.

Catalog record

CREATORKonrad J. Karczewski, Matthew Solomonson, Katherine R. Chao, Julia K. Goodrich, Grace Tiao, Wenhan Lu, Bridget M. Riley-Gillis, Ellen A. Tsai, Hye In Kim, Xiuwen Zheng, Fedik Rahimov, Sahar Esmaeeli
COLLECTIONPAPERS
FORMATTXT
FILE SIZE93 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC9903662

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Systematic single-variant and gene-based association testing of thousands of phenotypes in 394,841 UK Biobank exomes By Konrad J. Karczewski, Matthew Solomonson, Katherine R. Chao, Julia K. Goodrich, Grace Tiao, Wenhan Lu, Bridget M. Riley-Gillis, Ellen A. Tsai, Hye In Kim, Xiuwen Zheng, Fedik Rahimov, Sahar Esmaeeli Published in: Cell Genomics DOI: 10.1016/j.xgen.2022.100168 ABSTRACT Summary Genome-wide association studies have successfully discovered thousands of common variants associated with human diseases and traits, but the landscape of rare variations in human disease has not been explored at scale. Exome-sequencing studies of population biobanks provide an opportunity to systematically evaluate the impact of rare coding variations across a wide range of phenotypes to discover genes and allelic series relevant to human health and disease. Here, we present results from systematic association analyses of 4,529 phenotypes using single-variant and gene tests of 394,841 individuals in the UK Biobank with exome-sequence data. We find that the discovery of genetic associations is tightly linked to frequency and is correlated with metrics of deleteriousness and natural selection. We highlight biological findings elucidated by these data and release the dataset as a public resource alongside the Genebass browser for rapidly exploring rare-variant association results. INTRODUCTION Coding variation has been the most readily interpretable class of genomic variation since the development of the gene model and mapping of the human genome. As such, it has facilitated the mapping and interpretation of variants with immediate clinical importance such as the American College of Medical Genetics actionable variant list.1 More recently, exome sequencing has yielded the discovery of specific causal variants for hundreds of rare diseases, particularly dominant acting de novo variants for severe diseases.2 As the sample sizes of exome sequencing datasets continue to grow, so do the opportunities to identify associations between rare variants and phenotypes (both complex traits and diseases). In complex diseases, identifying causal genetic factors for a given disease can provide direct insight into the potential for therapeutic avenues. For instance, gain-of-function variants in PCSK9 have been demonstrated to increase low-density lipoprotein (LDL) levels and thus risk for cardiovascular disease.3 Accordingly, loss-of-function (LoF) variants are protective for cardiovascular disease,4 and less than 15 years after the discovery of this effect, therapeutic approaches to inhibit PCSK9 have been brought to market.5 Deeply phenotyped biobanks present a unique opportunity to simultaneously analyze multiple diseases and traits within a single cohort, enabling the discovery of new disease genes with therapeutic potential at a large scale, such as the identification of rare variants in ANGPTL7 that protect against glaucoma.6 The UK Biobank is a collection of approximately 500,000 participants with standardized, detailed phenotypic data7 on which genome-wide association studies (GWASs) have been run extensively. The UKB Exome Sequencing Consortium, a partnership between the UKB and 8 biopharma companies, generated exome sequences for this cohort,8 and recent studies have used the exome sequence data to explore various aspects of rare-variant associations, including novel biological signals for type 2 diabetes9 and cardiometabolic traits,10 as well as cross-phenotype analyses that identify new hits for a variety of traits.11, 12, 13 Here, we describe results from a systematic, large-scale rare-variant association analysis of 4,529 phenotypes, release these full sets of summary statistics in a results browser, and explore the role of natural selection and allele frequency on rare-variant associations. RESULTS Generating high-quality exome data for rare-variant associations We built an end-to-end pipeline for read mapping, processing, joint variant calling, quality control (QC), and mixed-model association analysis and applied this pipeline to 454,697 individuals with exome sequence data from the UK Biobank. The read mapping and processing pipeline adopted the GATK Best Practices pipeline (GRCh38), and the resulting variants (gVCF files) were joint called using a scalable implementation in Hail (Figure S1).14 We processed a set of 4,529 phenotypes including 1,233 quantitative traits as well as 3,296 binary traits with at least 200 cases, which included 725 disease endpoints based on Internal Classification of Diseases (ICD)-10 codes (Figure S2). After performing QC in a similar but augmented (e.g., array concordance; see Supplemental information) manner as for the Genome Aggregation Database (gnomAD),15 we generated a high-quality dataset of 450,953 individuals (Figures S3–S5; Table S1) including related individuals. This included 394,841 individuals of European ancestry in which we find 23,880,790 high-quality variants (Figure S6). For each of the 19,407 protein-coding genes, we considered up to four functional annotation categories: predicted LoF (pLoF), missense (including low-confidence pLoF variants and in-frame insertions or deletions [indels]), synonymous, and the combination pLoF or missense group, resulting in 8,074,878 variants and 75,767 groups for association testing (i.e., one group per gene and functional annotation category). Creating a high-quality set of rare-variant associations We performed group tests using the mixed-model framework SAIGE-GENE,16 which includes single-variant tests and gene-based burden (mean), SKAT (variance), and SKAT-O (hybrid variance/mean) tests (Figure S7). In total, we performed up to 8,074,878 single-variant tests and 75,767 group tests for each of 4,529 phenotypes (Figure 1). Additionally, we generated 314 heritable random phenotypes to test the asymptotic properties of the mixed-model association testing framework (Figures S8 and S9) and to determine empirical p value thresholds for Type I error control. Based on this analysis, for each phenotype, in addition to QC criteria defined below, we consider genome-wide p value thresholds of 2.5 × 10−7 for SKAT-O tests, 6.7 × 10−7 for burden tests, and 8 × 10−9 for single-variant tests (see Supplemental information and Figure S10), corresponding to approximately 0.05 expected false positives per phenotype.Figure 1Quality control (QC) of rare-variant association tests(A–C) The number of phenotypes (A), variants (B), and groups (i.e., gene-annotation pairs; C) before and after QC.(D and E) After QC, the number of variants (D) and genes (E) are broken down by annotation and frequency bin (alternate allele frequency [AF] for variants, cumulative AF [CAF] for genes). We performed extensive QC on these summary statistics (Figure 1; Table S2), including a minimum of two variants per group test, a minimum coverage of 20×, and a minimum expected allele count (frequency × n_cases) of 50 for the summary statistics, respectively, as well as genomic control (lambda GC) for each phenotype and each gene (Figures S11–S15). Further, we pruned to a set of 3,819 high-quality independent phenotypes encompassing 677 continuous traits and 3,142 binary traits, including 708 ICD codes (Figures 1A and S16; Table S2). We confirmed the robustness of our results by comparing them wiht a previous large-scale study of height (Tables S3–S5; Figure S17) and red blood cell phenotypes (Table S6), for which our analysis replicates the majority of associations with consistent direction of effect.17,18 We filtered to 263,696 variants, including 6,117 pLoF variants, 155,705 missense variants, and 101,874 synonymous variants with at least one phenotype having an expected allele count (cohort frequency × n_cases) over 50 (Figure 1B). For group tests, we filter to a high-quality set of 54,647 gene tests with at least 20× coverage (Figure S13) and at least one phenotype with expected allele count ≥50 for pLoF (7,296 genes), missense (15,943), synonymous (16,014), and pLoF or missense (15,394) (Figure 1C). Using these criteria, we identified a total of 68,623 and 6,403 associations meeting our p value threshold with a mean of 18.0 and 1.7 associations per phenotype for single-variant tests and group tests, respectively

Read the complete text