A conundrum of pleiotropy: GWAS-derived pleiotropy is weakly negatively associated with co-expression network centrality in Arabidopsis thaliana

Abstract

Pleiotropy, the influence of a single gene on multiple phenotypic traits, is a fundamental feature of genetic systems but remains difficult to quantify at genome-wide scales. Genome-wide association studies (GWAS) provide one avenue to characterize pleiotropy through multi-trait genetic associations, while functional genomic approaches such as gene co-expression networks offer indirect predictions of gene importance and pleiotropic potential based on network centrality statistics. However, the relationship between GWAS-derived pleiotropy and network-based measures of gene importance remains unclear. Here, we quantified gene-level pleiotropy in Arabidopsis thaliana using gene-trait associations from the AraGWAS Catalog and AraPheno databases and compared them to previous estimates of pleiotropy based on gene co-expression networks. GWAS pleiotropy was measured using the Hill number of order q = 1, an entropy-based metric adapted from ecological diversity theory. Across 2,179 genes, pleiotropy exhibited a right-skewed distribution, with most genes influencing relatively few traits and a smaller subset exhibiting broad multi-trait effects. Contrary to expectations, pleiotropy was weakly negatively correlated with multiple measures of co-expression network centrality, including degree, strength, closeness, and betweenness. Additional analyses showed that GWAS-associated genes were enriched within the co-expression network, exhibited broader expression across tissues, and were modestly enriched among conserved BUSCO genes. Together, these findings reveal a conundrum: although genes identified by GWAS are broadly expressed and integrated into functional networks, the most pleiotropic genes occupy relatively peripheral positions within those networks. This unexpected pattern suggests that GWAS-derived pleiotropy captures a distinct aspect of gene function than co-expression network centrality.