Paper I
Phenotypically independent profiles relevant to mental health are genetically correlated
We initially set out to create a set of profiles with phenotypes relevant for mental health. For this we accessed the self-administered questionnaire for each participant with respective data available in UK Biobank. The self-administered questionnaire consists of about 140 items in total covering general mental health, risk factors, and psychiatric symptoms. However, for our analyses we only analyzed a subset of about 43 questions presented to all participants after removing follow-up questions, questions on short-term temporal experiences, and questions with more than 10% missing data. We imputed missing data using the kNN approach with k = 3 and z-normalized the data. We then decomposed this dataset into 13 independent components using an ICA algorithm, each reflecting a different domain relevant for mental health as shown in panel A in Figure 25.1. The loadings from each individual on these 13 components served as the input for the GWAS.
The ICA analysis confirmed that even in an undiagnosed population sample, there is sufficient phenotypic variance to capture relevant domains of mental health, such as psychosis (IC2), anxiety, depression, and mental distress (IC3), and addiction and mania (IC11). It also suggests that there will be sufficient genetic variance to run a GWAS successfully.
We performed a GWAS on each of the 13 independent components separately. In line with common practice at the time, we included only individuals with White European ancestry resulting in a dataset of 117,611 individuals aged 47 to 80 years (mean: 64, SD: 7.66) at time of the mental health asssessment, and the dataset comprised of 56.2% females.
We compared the GWAS summary statistics to a set of publicly available summary statistics on a number of psychiatric disorders and traits. We used the LDSC toolbox to calculate the genetic correlation between each of the 13 independent components and each of the 9 psychiatric disorders and traits, 117 comparisons in total. The result of these genetic correlations are displayed in panel B in Figure 25.1.
We found that for most of the psychiatric disorders and traits the independent components that closely reflected the features of that given disorder showed the strongest genetic correlation. For example, SCZ showed the strongest genetic correlation with IC2 (capturing items related to psychosis), ANX correlated most strongly with IC3 (capturing items related to anxiety, depression, and mental distress), and BIP correlated most strongly with IC11 (capturing items related to addiction and mania).
When comparing the genetic and phenotypic correlation of each of the independent components to each other we found widespread genetic correlation despite the absence of phenotypic correlation, as illustrated in panel C in Figure 25.1. Approximately half of the genetic correlations in this analysis survived multiple testing correction (51%, p < FDR). IC3 (reflecting anxiety, depression, and mental distress) correlated with all but two other independent components, and no independent component had no significant genetic correlations with other components.
These genetic correlations in the absence of phenotypic correlation indicate that the shared genetics between psychiatric disorders may reflect shared biology that extends beyond the common symptoms shared between traits relevant for mental health.