UKB

The UKB is a large-scale population-based cohort study with deep phenotypic and genetic data from half a million participants aged 40–69 yr from the United Kingdom³³. For genetic data, approximately 488,000 participants were genotyped with more than 800,000 markers. The genotype platforms, genetic quality controls and imputation procedures were detailed in a previous study³³. In the current analysis, we included 438,870 participants who were genetically defined as ‘white-British’ ancestry^8,32. SNPs with minor allele frequency (MAF) > 0.01 and imputation quality score > 0.8 were kept in association analysis.

The detailed definitions of phenotypic data, including glaucoma, VCDR and IOP, were described in our previous studies^8,9,32,34. Briefly, glaucoma cases were ascertained from ICD-10 diagnosis, record-linkage data from local general practitioners and self-reported previous diagnosis; controls were defined as participants who self-reported having no eye disease (UKB phenotypic data downloaded in March 2020). In our association analysis, we kept 11,239 glaucoma cases and 137,621 controls of European ancestry. We ran generalized mixed models in SAIGE (v.0.29.6)³⁵ and adjusted for age, sex and the first ten genetic principal components. In the SAIGE analysis, generalized linear mixed models with two steps were fitted to account for unbalanced case–control ratios and sample relatedness. The first step (fitNULLGLMM) was used to estimate variance component and model parameters. The second step (SPAGMMATtest) performed single variant score tests with saddlepoint approximation based on logistic mixed models³⁵.

The VCDR measurements of optical nerve head photographs were based on CNN models trained on clinical assessments⁹. Both VCDR and vertical disc diameter from approximately 70,000 UKB fundus images were used to train CNN models. In our previous work, we have shown that AI-based measurements were more accurate and increased GWAS power of genetic discovery. In the current study, we performed GWASs in 68,240 participants with AI labeling VCDR. The association tests were conducted using linear mixed models (BOLT-LMM v.2.3 (ref. ³⁶)) adjusting for vertical disc diameter, age, sex and the first ten principal components.

The IOP GWAS in UKB was based on corneal-compensated IOP measurements in 103,914 participants^8,32. Linear mixed models were performed in BOLT-LMM (v.2.3) adjusting for age, sex and the first ten principal components.