Limited evidence for a general dimension of impulsivity across molecular genetic, transcriptomic, and neurogenomic levels of analysis

Impulsivity, broadly defined as a predisposition towards rapid and unplanned responses to stimuli, is a complex psychological construct implicated in numerous psychiatric conditions. Insights from neuroscience suggest that impulsivity can be ‘split’ into related-but-distinct facets rather than considered a unitary trait; however, recent advances in clinical psychology and psychiatry have revealed that many psychological constructs can be effectively ‘lumped’ into broader categories at multiple levels of analysis. Here, we conduct a series of multivariate genetic analyses to test the splitting versus lumping hypotheses in the genetic architecture of impulsivity. We used the Genomic SEM framework to conduct a multivariate analysis of eight impulsivity facets (Ns ∼ 125k), employing two distinct statistical models: (1) an “omnibus” model that provides an unstructured meta-analytic test of association, and (2) a common factor of impulsivity, where associations operate through a latent genetic factor. To gain a more comprehensive understanding of the complex biological underpinnings of impulsivity, we examined these models across molecular genetic, transcriptomic, and neurogenomic levels. To examine the validity of general impulsivity and compare it to facet-level genetic results, we conducted polygenic score analyses in two independent cohorts: the Adolescent Brain and Cognitive Development Study (N ∼ 5k) and the Vanderbilt University Medical Center biobank (N ∼ 72k). Results revealed widespread pleiotropy among impulsivity facets, with moderate-to-large genetic correlations observed for most trait pairs (mean rg = 0.434, range = -0.085-0.785). Genomic factor analysis suggested that the observed patterns of covariance could be closely approximated with a common factor model (χ2[17] = 410.883, CFI = 0.941, SRMR = 0.08). However, subsequent analyses revealed that the polygenic architecture of impulsivity was largely inconsistent with a model of general impulsivity. For example, although we observed inflation of the GWAS test statistics for the common factor model (λGC = 1.22, mean χ2 = 1.237), the substantially greater inflation of the QSNP (λGC = 1.607, mean χ2 = 1.568) and omnibus results (λGC = 1.592, mean χ2 = 1.535) suggested that there was an appreciable signal not captured by the common factor. This pattern was also observed in multivariate TWAS and genetic correlation analyses, where impulsivity facets again exhibited many trait-specific patterns of association, including highly heterogeneous signals in the default mode and visual networks. Finally, polygenic score results underscored the consequences of erroneously lumping impulsivity facets, as 67 links to health outcomes were lost when a general impulsivity polygenic score was used, including novel links between urgency traits and neurodevelopmental disorders. Through multivariate modeling, our study generates new insights into the genetic architecture of impulsivity, implicating novel risk genes and links to human health. We present results from two complementary statistical models that elucidate the question of lumping versus splitting in this domain. Collectively, our results suggest that a multidimensional approach is likely to be more informative when studying the heterogeneous biology of impulsivity.