Insomnia and the genetic overlap of sleep with depression and anxiety: a genomic structural equation modelling study

Abstract

Insomnia, depression and anxiety co-occur and share genetic liability, but a genetic correlation cannot show whether that liability is spread across the sleep domain or concentrated in insomnia, a diagnostic criterion for depression and generalized anxiety disorder. We applied genomic structural equation modelling to ten genome-wide association studies and fitted a four-factor model of sleep, internalizing (depression and anxiety), psychotic-bipolar and metabolic liability. A model linking the sleep and internalizing domains only through their factor correlation was inadmissible, and we compared, under one decision rule, each one-parameter link that could repair it: domain-wide links, an insomnia-specific cross-loading, and residual covariances representing shared diagnostic content, of which three were admissible. The insomnia cross-loading fitted better than the only admissible domain-wide alternative ({Delta}AIC = 13.8; better in 98.6% of 500 Monte Carlo draws) and than shared content ({Delta}AIC = 96.9), and placing it on psychotic-bipolar or metabolic liability was inadmissible. Under this model the insomnia pathway carried 56.3% of the sleep-internalizing covariance (95% CI 45.6-67.0%). A second sleep pathway improved fit ({Delta}AIC = -33.3), but with three sleep indicators its location was not identifiable, and across the equivalent models the insomnia share ranged from 0% to 127.0%. At the variant level, no internalizing lead departed from depression-anxiety proportionality after correction, and internalizing leads replicated in FinnGen, with partial sample overlap, whereas sleep leads did not (60.4%; P = 0.17). How concentrated the overlap is will require finer sleep phenotypes to establish. Among single-link models, insomnia is the best-supported location of the overlap.