Soft clustering analysis of primary open angle glaucoma loci and their association with multiple endophenotypes uncovers pathophysiological and cellular mechanisms of disease

Abstract

Purpose : Primary open angle glaucoma (POAG) is a heterogeneous disease for which the underlying molecular mechanisms and their associated traits are not well defined. We applied a soft clustering method, Bayesian nonnegative matrix factorization (bNMF), to genome-wide association study (GWAS) data to identify novel sets of known POAG genetic loci that cluster with endophenotypes potentially significant to POAG. Such clusters propose different mechanistic pathways and traits that could be causal to POAG, which we further investigated for gene enrichment in ocular cell types and biological pathways.

Methods : bNMF clustering emulates physiological relevance as individual genes can belong to more than one pathway, and accounts for trait direction of effect on POAG risk in the prepared standardized association matrix. bNMF was performed using 133 independent POAG variants from the largest European and cross-ancestry genome-wide association study (GWAS) meta-analyses of POAG. Endophenotypes were selected based on known or putative pathogenic mechanisms of POAG, and available GWAS summary statistics. Genes were mapped to POAG loci via e/sQTL colocalization analysis, and tested for single cell expression enrichment in anterior segment (AS), optic nerve head and retina, and enrichment in biological processes.

Results : Of 100 POAG-related traits identified, 33 met the inclusion criteria. bNMF identified 4 clusters with 5 to 39 POAG variants each (converging to N=4, 83.8% of 1000 iterations). Each cluster was represented by unique trait and direction of effect combinations. The strongest effects on POAG risk resulted from decreased plateletcrit, diastolic blood pressure and urate serum levels (cluster 2), and increased intraocular pressure and decreased corneal hysteresis (cluster 3). Genes in POAG loci of each cluster showed significant enrichment (FDR<0.1) in specific cell types and pathways, e.g. pericyte (AS) and Müller glia (retina) (cluster 2), and ciliary and iris fibroblasts (AS) and retinal astrocytes (cluster 3).

Conclusions : Soft clustering identified several trait clusters and cell types associated with POAG risk, suggesting functionally distinct disease subtypes and offering a genetically-informed path towards personalized diagnosis and treatment.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.