Multi-ancestry genome-wide association study of dry eye disease in the Million Veteran Program

Abstract

Purpose : Dry eye disease (DED) is a common condition characterized by pain and dryness of the ocular surface. Twin studies have estimated a substantial genetic heritability of DED, explaining 30-40% of the variance. However, the difficulty of phenotyping DED has limited the application of large-scale genome-wide association studies (GWAS). The VA Million Veteran Program (MVP) biobank, comprising over 650,000 participants with genetic data linked to electronic medical records (EMRs), provides a unique opportunity to study the genetic basis of DED.

Methods : We devised a case-control algorithm based on ICD-9/10 codes and prescription records, validated by manual chart review at 3 different VA eye clinics. As applied to MVP participants, our algorithm identified 48,794 cases and 29,224 controls. We classified individuals into ancestry groups using the Harmonized Ancestry and Race/Ethnicity method. Logistic mixed model GWAS analyses were performed in each group (European, African, Hispanic, and East Asian) using SAIGE (v1.3.0), adjusting for age, sex, and 10 ancestry principal components, followed by multi-ancestry meta-analysis assuming fixed effects. Heritability was estimated from European ancestry GWAS summary statistics using linkage disequilibrium score regression (LDSC).

Results : Our chart review yielded a positive predictive value of 98% (147/150) and a negative predictive value of 93% (139/150). Female sex, age, African ancestry, and Charlson comorbidity index were associated with DED (p<0.001). We observed associations with comorbid conditions, such as depression (OR=2.7 [2.6-2.8]) and sleep apnea (OR=2.3 [2.2-2.4]). In the multi-ancestry meta-analysis, a novel common variant locus at 3p21.31 (rs9855433, near ZNF445) reached genome-wide significance. A phenome-wide association scan for this variant in FinnGen revealed directionally consistent pleiotropic associations with depression, sedative-hypnotic medication use, and chronic pain conditions. Using LDSC, we derived a significant common variant heritability estimate of 12% (s.e. 1.4 %) on the liability scale, assuming a prevalence of 25%.

Conclusions : We developed and validated a DED phenotyping algorithm that can be applied to other EMR-linked biobanks. In addition to identifying the first GWAS locus for DED, we attained a significant overall genetic signal, motivating expanded GWAS meta-analyses and the development of polygenic risk score models.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.