The structure of psychiatric comorbidity without selection and assortative mating

Abstract

The widespread comorbidity observed across psychiatric disorders may be the result of processes such as assortative mating, gene-environment correlation, or selection into population studies. Between-family analyses of comorbidity are subject to these sources of bias, whereas within-family analyses are not. Because of Mendelian inheritance, alleles are randomly assigned within families, conditional on parental alleles. We exploit this variation to compare the structure of comorbidity across broad psychiatric polygenic scores when calculated either between-family (child polygenic scores) or within-family (child polygenic scores regressed on parental polygenic scores) in over 25,000 genotyped parent-offspring trios from the Norwegian Mother Father and Child Cohort study (MoBa). We fitted a series of factor models to the between- and within-family data, which consisted of a single genetic p-factor and a varying number of uncorrelated subfactors. The best-fitting model was identical for between- and within- family analyses and included three subfactors capturing variants associated with neurodevelopment, psychosis, and constraint, in addition to the genetic p-factor. Partner genetic correlations, indicating assortative mating, were not present for the genetic p-factor, but were substantial for the psychosis (b= 0.081;95% CI [0.038,0.124]) and constraint (b= 0.257;95% CI [0.075,0.439]) subfactors. When average factor levels for MoBa mothers and fathers were compared to a population mean of zero we found evidence of sex-specific participation bias, which has implications for the generalisability of findings from cohort studies. Our results demonstrate the power of the within-family design for better understanding the mechanisms driving psychiatric comorbidity and their consequences on population health.

Maternal depression and the polygenic p factor: A family perspective on direct and indirect effects

“…Like all population studies, The Norwegian Mother, Father, and Child study suffers from selection biases, although evidence suggests that the impact of this selection bias is largely restricted to prevalence estimates obtained using the dataset and not to associations between exposures and outcomes (Nilsen et al, 2009). Only genotyped complete trios were included, however we show in separate analyses that the average polygenic p factor burden in these trios is similar to the population average (Ayorech et al, 2023). We indexed depression liability through genomic data rather than symptoms or clinical diagnoses, which partially buffers some of the selection biases as those with high genetic liability but who would not reach symptom criteria still can provide valuable information for risk prediction.…”

“…Exploratory factor analysis for the polygenic p factor was performed in Mplus 8.5 with geomin orthogonal bi-factor rotation (Browne, 2001;Muthén and Muthen, 2017;Muthén and Muthén, 1998). Further details on the construction of the polygenic p factor including model fit statistics and comparisons across mental health indices within MoBa and between MoBa and the population average, have been published elsewhere (Ayorech et al, 2023;Rosenström et al, 2019). With this approach we obtained a separate polygenic p factor for mothers, fathers, and children, which we could jointly model to separate direct (mother) and indirect (father and child) genetic effects on mothers depression, while simultaneously controlling for the shared genetic relatedness between family members.…”

“…Our exploratory factor analysis with a bifactor solution revealed a single general psychopathology factor for which each of the eleven PGS loaded significantly (Table 2). Full details regarding computation of the polygenic p factor and model selection can be found in the Supplementary Materials and online (Ayorech et al, 2023).…”

“…Our trio polygenic p design controls for shared genes between family members to reveal each member's unique genetic effect. An important implication of this design is that any observed indirect effects are corrected for many of the biases that plague typical behavioural genetic studies including assortative mating and population stratification (Ayorech et al, 2023;Torvik et al, 2022). Assortative mating refers to non-random mate choice that results in similarities between partners that are larger than chance (Torvik et al, 2022).…”

“…Evidence of indirect effects on mothers' depression have been demonstrated using an adoption design (McAdams et al, 2015), and now here with DNA alone. Importantly, compared to measured symptoms of general mental health, our polygenic p factor is not prone to reporter or selection bias and is normally distributed (Ayorech et al, 2023).…”