GenomicSEM Modelling of Diverse Executive Function GWAS Improves Gene Discovery

Abstract

Previous research has supported the use of latent variables as the gold-standard in measuring executive function. However, for logistical reasons genome-wide association studies (GWAS) of executive function have largely eschewed latent variables in favour of singular task measures. As low correlations have traditionally been found between individual executive function (EF) tests, it is unclear whether these GWAS have truly been measuring the same construct. In this study, we addressed this question by performing a factor analysis on summary statistics from eleven GWAS of EF taken from five studies, using GenomicSEM. Models demonstrated a bifactor structure consistent with previous research, with factors capturing common EF and working memory- specific variance. Furthermore, the GWAS performed on this model identified 20 new genomic risk loci for common EF and 4 for working memory reaching genome-wide significance beyond what was found in the constituent GWAS, together resulting in 29 newly mapped EF genes. These results help to clarify the underlying genetic structure of EF and support the idea that EF GWAS are capable of measuring genetic variance related to latent EF constructs even when not using factor scores. Furthermore, they demonstrate that GenomicSEM can combine GWAS with divergent and non-ideal measures of the same phenotype to improve statistical power.

Fig. 1

Standardised results for the initial three-factor confirmatory factor analysis model of EF GWAS. *p <.05, ** p <.01, *** p <.001

Fig. 2

Standardised results for the accepted confirmatory factor analysis model of EF GWAS. *p <.05, ** p <.01, *** p <.001