A REPEATED MEASURES GWAS IN RELATIVES: DETECTING AGE-VARYING GENETIC EFFECTS ON MATERNAL DEPRESSION IN MOBA

Abstract

Background

Genome-wide association studies (GWAS) of depression typically rely on case-control analyses, using lifetime history (LTH) of a diagnosis as the primary outcome. Longitudinal cohorts provide opportunities to understand how genetic effects on depression vary across time and contexts. We aimed to harness repeated measures in a single sample to minimise phenotypic heterogeneity and improve statistical power to detect genetic effects on depression.

Methods

We used data from mothers who participated in the Norwegian Mother, Father and Child study (MoBa). We restructured the Hopkins Symptoms Checklist Depression (SCL-D) scores for 76,044 mothers by age, producing an accelerated longitudinal design with 418,159 depression observations across the lifecourse (age range: 16, 60 years old). We used continuous-time item response theory models to minimise measurement error, and Empirical Bayes Estimates with Simultaneous Correction (SCEBE) to conduct repeated measures GWAS (GWASRM). We estimated genetic effects on depression symptoms at age 30, and the linear change in age (variant by age interaction), which were combined to compute age-specific effects at ages 20 to 50 (GWASt=20..50). For comparison, we performed a GWAS of LTH depression assessed in the same sample (GWASLTH). Summary statistics from GWASRM and GWASLTH were compared using heritability and genetic correlations. We also performed 10-fold-leave-one-out GWASRM and GWASLTH, leaving 10 test datasets for polygenic index (PGI) analyses. The efficacy and robustness of SCEBE was further assessed via comparison with linear mixed effects models applied to a subset of the data. The impact of selective attrition on GWASRM is also currently under investigation.

Results

SCEBE is an efficient and robust method for estimating genetic variant main effects and variant-by-age interactions, with perfect overlap with effects estimated using a linear mixed model. Following GWASRM, there was one independently significant (p < 5e-8) locus associated with depression symptoms at age 30 and an additional locus at age 40. The SNP-based heritability (h2SNP) of GWASt ranged from 0.01 to 0.05. Following GWASLTH there was one independently significant locus associated with lifetime history of depression, and the h2SNP estimate was 0.15. The age-specific PGI (PGIt) was associated with higher SCL-D symptoms (β=4.21, 95% CI: 6.27, 2.14; p=6.68e-5). The PGILTH was also associated with SCL-D symptoms (β=3.73, 95% CI: 2.78, 4.68).

Discussion

Despite the anticipated gains in statistical power from a repeated measures approach in a homogenous sample, we did not detect substantial improvement in genome-wide signal for depression. These results might be due to etiological heterogeneity associated with depression and unobserved confounding. For example, medication status was not available across these time-points and was not included in these models. Furthermore, both initial and continued participation are associated with depression symptoms - and individuals with the greatest burden from depression are less likely to participate and more likely to drop out of studies earlier. Work is ongoing to assess the role of confounding from medication status and to adjust repeated GWASRM for selective participation and attrition.