The Genetic Architecture of Alzheimer’s Disease Risk: A Genomic Structural Equation Modelling Study

Abstract

Background: Targeting modifiable risk factors may have a role in the prevention of Alzheimers disease. However, the mechanisms by which these risk factors influence Alzheimers risk remain incompletely understood. Genomic structural equation modelling can reveal patterns of shared genetic architecture that provide insight into the pathophysiology of complex traits. Methods: We identified genome-wide association studies for Alzheimers disease and its major modifiable risk factors: less education, hearing loss, hypertension, high alcohol intake, obesity, smoking, depression, social isolation, physical inactivity, type 2 diabetes, sleep disturbance and socioeconomic deprivation. We performed linkage disequilibrium score regression among these traits, followed by exploratory factor analysis, confirmatory factor analysis and structural equation modelling. Results: We identified complex networks of linkage disequilibrium among Alzheimers disease risk factors. The data were best explained by a bi-factor model, incorporating a Common Factor for Alzheimers risk, and three orthogonal sub-clusters of risk factors, which were validated across the two halves of the autosome. The first sub-cluster was characterised by risk factors related to sedentary lifestyle behaviours, the second by traits associated with reduced life expectancy and the third by traits that are possible prodromes of Alzheimers disease. Alzheimers disease was more genetically distinct and displayed minimal shared genetic architecture with its risk factors, which was robust to the exclusion of APOE. Conclusion: Shared genetic architecture may contribute to epidemiological associations between Alzheimers disease and its risk factors. Understanding the biology reflected by this communality may provide novel mechanistic insights that could help to prioritise targets for dementia prevention.

Genetic risk shared across 24 chronic pain conditions: identification and characterization with genomic structural equation modeling

“…31 For this validation, we used a split-genome approach. 32,39 We developed a model using the EFA-CFA procedure described above in odd autosomes (1, 3, … 21) and assessed the fit of the final CFA in even autosomes (2, 4, … 22). This split created 2 independent sets, given that SNPs are not correlated across chromosomes (Mendel's law of independent assortment).…”

Identification and characterization of genetic risk shared across 24 chronic pain conditions in the UK Biobank

“…As the EFA-CFA model is more flexible, we tested for overfitting [33] by repeating the analysis originally conducted on the whole dataset, using a split-genome approach [42,34]. In step 1, we applied the EFA-CFA procedure described above in odd autosomes (1,3,...21), and in step 2, we assessed the fit of the CFA from step 1 in even autosomes (2,4,...22).…”

Are human connectomes heritable?

“…Some advantages of SEMs are that the model can incorporate multivariate inputs, account for covariates such as sex and age, allow for incorporating more familial relationships such as step siblings and parents, and provide more accurate estimates of heritability [19]. SEMs have been used to identify heritability of diseases such as schizophrenia and Alzheimer's [20,21]. Although SEMs are extremely useful in human genetics [22], they exhibit several limitations.…”

“…The intraclass correlation (ICC) [9], which estimates the amount of variation that is common to the group as a proportion of the variation in the population, and Falconer's method [8], which is the difference in ICC in identical and fraternal twins of a given trait, have been used to show significant heritability in human personalities [10][11][12], anatomy [13][14][15][16], and intelligence [17,18]. Structural equation models (SEMs), a type of linear mixed model, have also been used to estimate heritability of phenotypically complex diseases like schizophrenia and Alzheimer's [19,20]. Some advantages of SEMs are that the model can incorporate multivariate inputs, account for covariates such as sex and age, incorporate familial relationships such as step siblings and parents, and provide more accurate estimates of heritability [21].…”