Multivariate, Multi-rater and Multi-age GWAS of aggression and attention problems

Genetic effects contribute to variation in childhood aggression (AGG) and attention (ATT) problems. Analyses of repeated observations of AGG and ATT in childhood reveal substantial rank order stability (.41 to .78), with a... [ view full abstract ]

Genetic effects contribute to variation in childhood aggression (AGG) and attention (ATT) problems. Analyses of repeated observations of AGG and ATT in childhood reveal substantial rank order stability (.41 to .78), with a significant portion of this stability attributable to genetic factors (65%) (Kan et al. JAACAP, 2013). Twin studies have shown that teacher self and parental ratings of AGG and ATT are genetically correlated, yet the rater specific variance also is heritable(Arsenault et al. J Child Psychol Psychiatry, 2003; Bartels et al. Behav Genet, 2003). GWA studies of AGG in children (Pappa et al. Neuropsychiatric Genetics, 2015; Middeldorp et al.  JAACAP, 2016) have considered measures of AGG and ATT in univariate analyses, as have most other GWA meta-analyses projects (either in children or in adults). Twin studies have been at forefront of analyzing multivariate and longitudinal data, but GWA studies have only made use of these approaches in a limited way. The utilization of multiple measures on the same subject in GWA studies is non-trivial as it inflates type-1 error.

To enable the multivariate genome wide study of the (developmental) genetic etiology of AGG and ATT, a multivariate meta-regression model is presented. Our approach relies on univariate GWA of a longitudinal phenotype within a cohort, possibly assessed by multiple raters and the phenotypic correlations between traits within cohort (Nivard et al. Schizophr Bull, 2017). The multivariate aspect of the analysis is performed at the meta-analysis stage, omitting the need for multivariate modeling in individual cohorts. Our method enables the inclusion of dependent measures into a single analysis without inflating type-1 error. Our method further allows for estimation of age, rater or cohort specific effects for SNPs. We present a GWAS of AGG and ATT in repeatedly measured children at different ages (range 3-18) and by multiple raters (mother, father, teacher, self).