Abstract
Depression and anxiety are highly prevalent and comorbid psychiatric traits that cause considerable burden worldwide. Previous studies have revealed substantial genetic overlap between depression, anxiety, and neuroticism. Here, we use factor analysis and genomic structural equation modelling (Genomic SEM) to investigate the genetic factor structure underlying 28 items assessing these traits in over 400,000 individuals from the UK Biobank. We identified two distinct, genetically correlated factors: depressive symptoms (DEP) and anxiety symptoms (ANX). Multivariate GWAS on these factors identified 89 independent variants for DEP (29 novel) and 102 for ANX (71 novel). These findings replicated at high rates (72-78%) in an independent 23andMe cohort of ~1.9 million individuals. Our results provide novel insight into the genetic architecture of these disorders, identifying trait-specific associations with metabolic and cardiovascular phenotypes.
Introduction
Depression and anxiety are often described as the "common colds" of psychiatry due to their high prevalence and frequent co-occurrence. Despite their global impact, genetic research has faced a significant hurdle: phenotypic heterogeneity. A diagnosis of "Major Depressive Disorder" is an umbrella for diverse symptoms like insomnia, guilt, and fatigue. Furthermore, both disorders overlap significantly with neuroticism—a personality trait reflecting emotional instability.
While large-scale studies have identified over 100 risk loci for depression, anxiety research has remained chronically underpowered. This study seeks to move past broad clinical labels by using individual symptom data to untangle the shared genetic architecture of these traits and boost the statistical power to identify novel risk genes.
Methods
The researchers employed a multi-stage statistical pipeline to analyze data from the UK Biobank (UKB), the 23andMe cohort, and the Australian Twin Registry.
- Exploratory Factor Analysis (EFA): Genetic correlations were analyzed across 28 individual survey items from the PHQ-9 (depression), GAD-7 (anxiety), and EPQR-S (neuroticism).
- Genomic Structural Equation Modelling (Genomic SEM): Rather than using a simple sum of symptoms, the team modeled "latent factors"—the underlying genetic signals driving symptom groups. This identified two primary factors: DEP (depressive symptoms) and ANX (anxiety symptoms).
- Multivariate GWAS: A genome-wide association study was conducted on these latent factors, leveraging the massive sample size of neuroticism data to "boost" the signal for anxiety and depression.
- Pairwise Analysis (
gwas-pw): A Bayesian approach was used to distinguish between shared and trait-specific genomic regions.
Key Insights and Results
Neuroticism as a Genetic Hybrid
One of the study's most striking findings is that neuroticism did not form its own distinct genetic factor. Instead, its items partitioned between the two primary factors: "mood swings" and "loneliness" loaded onto the DEP factor, while "worry" and "nerves" loaded onto the ANX factor. This suggests that neuroticism is essentially a measure of sub-clinical depression and anxiety rather than a separate genetic construct.
Identification of Novel Loci
By leveraging the genetic overlap with neuroticism through the latent factor approach, the authors significantly increased the discovery of risk loci:
- Depression: 89 independent variants identified (61 genomic loci, 29 of which are novel).
- Anxiety: 102 independent variants identified (73 loci, 71 of which are novel).
These results showed high robustness, with approximately 72% of depression variants and 78% of anxiety variants replicating in the 23andMe cohort of ~1.9 million individuals.
Metabolic vs. Cardiovascular Divergence
While the two traits share roughly 80% of their genetic liability, the unique regions identified are biologically telling. Depression-specific genomic regions were associated with hypertriglyceridemia (high blood fats). In contrast, anxiety-specific regions were linked to blood pressure phenotypes and shared pathways with schizophrenia and autism.
Improved Prediction
Polygenic Risk Scores (PRS) derived from these latent factors were significantly more effective at predicting symptoms in independent samples compared to standard sum-score methods, showing a 185% to 237% increase in explained variance.
Discussion and Future Work
This study demonstrates that moving from broad clinical diagnoses to symptom-level genetic modeling can unlock significant discovery power. The findings suggest that current diagnostic boundaries may not perfectly align with underlying biology, particularly regarding the role of neuroticism.
However, the study is limited by its reliance on population-based cohorts like the UK Biobank, which may not capture the full severity of clinical psychiatric cases. Additionally, the questionnaires used primarily focused on psychological symptoms, potentially missing the genetic architecture of somatic symptoms (e.g., physical pain). Future work applying this "bottom-up" genetic taxonomy to a broader range of psychiatric traits could lead to a more biologically-based diagnostic system in psychiatry.
Summary updated