USING REPEATED MEASURES TO IMPROVE THE PRECISION AND POWER OF GENOME-WIDE ASSOCIATION STUDIES (GWAS)

Genome Wide Association Studies (GWAS) have been vital to understanding the genetics of complex traits. However, the majority of GWAS use data from only one occasion, even in longitudinal studies with repeated assessments. Traits like depression are often subject to measurement error. Using a more precise and longitudinal phenotype of depression could reduce measurement error and increase power and precision in depression GWAS, further enhancing understanding of the genetics of depression.

We used data from the UK Biobank (max n=462,566) on the PHQ-2, a measure of depressive symptoms assessed up to 8 occasions over approximately 17 years. We tested a GWAS baseline model (a traditional cross-sectional GWAS that took the first observed assessment) against four other longitudinal GWAS models: 1) the mean of all the assessments, 2) a structural equation model (common factor model), 3) a precision-weighted shrinkage model and 4) a genomicSEM model. We also conducted analysis across multiple ancestries and performed out of sample polygenic prediction.

All longitudinal GWAS models outperformed the GWAS baseline model in European ancestries, with the most powerful model being the precision-weighted shrinkage model which identified 169 genome wide significant single nucleotide polymorphisms (SNPs). Importantly, this precision-weighted shrinkage method had an increase of 34 more lead SNPs (121% increase), 107 more independent significant SNPs (173% increase), 50 more mapped genes (35% increase) and a greater SNP heritability (35% increase), compared to the baseline model. Polygenic prediction into an external cohort also explained a greater proportion of variance (17% increase). There were no GWAS lead SNPs identified in the South Asian and African baseline models, but 2 and 7 novel lead SNPs in the precision-weighted shrinkage methods, respectively.

Leveraging repeated information within GWAS appears to improve power and precision to detect novel biological underpinnings in depression. This is likely due to a reduction in measurement error and increased power. These methods can be applied to other noisy traits within psychiatric genetics and could be useful for detecting novel loci in smaller studies.