Mendelian randomisation to test causal relationships between diurnal preference and mental health

Several MR approaches were employed, first, the standard one-sample instrumental variable analyses using the GRS_s was performed in the unrelated subset [33]. One-sample MR uses one dataset in the instrumental variable analysis to yield the causal estimate of the risk factor (here diurnal preference) on the outcome (here depression). This method enables sensitivity analyses to be easily performed but requires the unrelated sample as it cannot account for relatedness in the model. One-sample MR uses the two-stage least-squares regression estimator to predict the levels of morningness per genotype and then regress the mental health outcome against the predicted value. First, an unconfounded estimate of diurnal preference variation was estimated by taking the association between being a morning person and the diurnal preference GRS. The mental health outcome was then used as the dependent variable in a logistic regression (binary outcome) or ordered logistic (ordinal outcome) model.

Second, we investigated the causal relationship using two-sample MR (this uses two different study samples to estimate the instrument-risk factor and instrument-outcome associations) in the larger group of related individuals. The variants were extracted from our UK Biobank BOLT-LMM [34] GWAS summary data for the mental health outcomes. The variant-chronotype associations were taken from the primary GWAS of diurnal preference with the betas for both the 339 and 108 coming from 23&Me. Four different two-sample MR methods were used that follow different assumptions. Inverse variance weighted (IVW) MR [35] is a weighted regression of the chronotype variant-chronotype association against the chronotype variant-mental health/wellbeing association, with the intercept constrained to zero. Using the multiplicative random-effects IVW model accounts for balanced horizontal pleiotropy. However, further methods were performed to help account for horizontal pleiotropy. These included the MR-Egger analyses [36], which essentially performs the weighted regression without a constrained intercept, therefore allowing for unbalanced horizontal pleiotropy. MR-Egger assumes that the pleiotropic effects are independent of the variant-exposure effects (InSIDE assumption) and therefore weighted median MR which is robust to horizontal pleiotropy and does not rely on the InSIDE assumption was also used. This approach bases the overall estimate on the weighted median variant estimate [37], but does require that 50% or more of the instruments are valid. Finally, a penalised weighted median was calculated where outlying variants are penalised.

The results from MR analyses may represent a valid causal effect estimate under the condition of three core assumptions:

The genetic instrument needs to robustly associate with the exposure (‘relevance’);

There should be no joint causal influence affecting the exposure instrument and the outcome (‘independence’);

The instrument must not affect the outcome through any mechanism other than through the exposure (‘exclusion restriction’).

Using the MR power calculator (https://shiny.cnsgenomics.com/mRnd/) we have demonstrated at p = 0.05 with our sample size (for depressive symptoms) we have >99% power [38].