Educational attainment, structural brain reserve and Alzheimer's disease: a Mendelian randomization analysis

Abstract

Higher educational attainment is observationally associated with lower risk of Alzheimer's disease. However, the biological mechanisms underpinning this association remain unclear. The protective effect of education on Alzheimer's disease may be mediated via increased brain reserve. We used two-sample Mendelian randomization to explore putative causal relationships between educational attainment, structural brain reserve as proxied by MRI phenotypes and Alzheimer's disease. Summary statistics were obtained from genome-wide association studies of educational attainment (n = 1 131 881), late-onset Alzheimer's disease (35 274 cases, 59 163 controls) and 15 measures of grey or white matter macro- or micro-structure derived from structural or diffusion MRI (nmax = 33 211). We conducted univariable Mendelian randomization analyses to investigate bidirectional associations between (i) educational attainment and Alzheimer's disease; (ii) educational attainment and imaging-derived phenotypes; and (iii) imaging-derived phenotypes and Alzheimer's disease. Multivariable Mendelian randomization was used to assess whether brain structure phenotypes mediated the effect of education on Alzheimer's disease risk. Genetically proxied educational attainment was inversely associated with Alzheimer's disease (odds ratio per standard deviation increase in genetically predicted years of schooling = 0.70, 95% confidence interval 0.60, 0.80). There were positive associations between genetically predicted educational attainment and four cortical metrics (standard deviation units change in imaging phenotype per one standard deviation increase in genetically predicted years of schooling): surface area 0.30 (95% confidence interval 0.20, 0.40); volume 0.29 (95% confidence interval 0.20, 0.37); intrinsic curvature 0.18 (95% confidence interval 0.11, 0.25); local gyrification index 0.21 (95% confidence interval 0.11, 0.31)]; and inverse associations with cortical intracellular volume fraction [-0.09 (95% confidence interval -0.15, -0.03)] and white matter hyperintensities volume [-0.14 (95% confidence interval -0.23, -0.05)]. Genetically proxied levels of surface area, cortical volume and intrinsic curvature were positively associated with educational attainment [standard deviation units change in years of schooling per one standard deviation increase in respective genetically predicted imaging phenotype: 0.13 (95% confidence interval 0.10, 0.16); 0.15 (95% confidence interval 0.11, 0.19) and 0.12 (95% confidence interval 0.04, 0.19)]. We found no evidence of associations between genetically predicted imaging-derived phenotypes and Alzheimer's disease. The inverse association of genetically predicted educational attainment with Alzheimer's disease did not attenuate after adjusting for imaging-derived phenotypes in multivariable analyses. Our results provide support for a protective causal effect of educational attainment on Alzheimer's disease risk, as well as potential bidirectional causal relationships between education and brain macro- and micro-structure. However, we did not find evidence that these structural markers affect risk of Alzheimer's disease. The protective effect of education on Alzheimer's disease may be mediated via other measures of brain reserve not included in the present study, or by alternative mechanisms.

Keywords: Alzheimer’s disease; MRI; Mendelian randomization; brain reserve; educational attainment.

Figure 1

Directed acyclic graph illustrating the putative causal relationships examined in this study. Imaging-derived measures of brain structure are explored as potential mediators of the effect of EA (exposure) on Alzheimer’s disease (outcome). G_X represents the set of instrumental variables for EA. G_M represents the set of instrumental variables for brain imaging-derived phenotype_{1, …, N}. Parameter a represents the direct causal effect of EA on imaging-derived phenotype_{1, …, N}. Parameter b represents the direct causal effect of imaging-derived phenotype_{1, …, N} on Alzheimer’s disease. Parameter c’ represents the direct causal effect of EA on Alzheimer’s disease. Parameters a and c’ are estimated using the set of variants G_X, and parameter b is estimated using the set of variants G_M. Confounding variables are omitted from the diagram.

Figure 2

Overview of the brain imaging-derived phenotypes investigated in this study. Structural brain reserve was proxied by 15 brain imaging-derived phenotypes. These included 10 global cortical phenotypes (six macro-structural metrics derived from T₁-weighted structural MRI and four micro-structural metrics derived from diffusion MRI), two white matter micro-structural phenotypes (measured at 21 major white matter tracts across the whole brain), bilateral hippocampal volumes and the total volume of white matter hyperintensities in the brain (derived from T₂-weighted MRI).

Figure 3

MR estimates of the association between genetically proxied EA and Alzheimer’s disease. Estimates represent odds ratios (95% CIs) for late-onset Alzheimer’s disease per 1 SD increase in genetically predicted YOS (∼4.2 years).

Figure 4

MR estimates of bidirectional associations between EA and brain imaging-derived phenotypes. Left: Estimates from IVW MR of the association between genetically proxied EA and imaging-derived brain structure phenotypes. Estimates represent the SD change in the imaging phenotype per 1 SD increase in genetically predicted YOS (∼4.2 years). Right: Estimates from IVW MR of the association between genetically proxied brain structure phenotypes and EA. Estimates represent the SD change in YOS per 1 SD increase in genetically predicted levels of each imaging-derived phenotype. Estimates marked with an asterisk were significant after correction for multiple testing. The SNPs column represents the number of SNPs remaining after clumping for independence and data harmonization. There were no significant clumps for the FA (cortex) phenotype after LD pruning. CT = cortical thickness; FA = fractional anisotropy; HC = hippocampus; IC = intrinsic curvature; ICVF = intracellular volume fraction; LGI = local gyrification index; MC = mean curvature; MD = mean diffusivity; ODI = orientation dispersion index; SA = surface area; Vol = volume; WM = white matter; WMH = white matter hyperintensities.

Figure 5

MR estimates of bidirectional associations between brain imaging-derived phenotypes and Alzheimer’s disease. Left: Estimates from IVW MR of the association between genetically proxied imaging-derived phenotypes and Alzheimer’s disease. Estimates represent the odds ratio for late-onset Alzheimer’s disease per SD increase in genetically predicted levels of each imaging-derived phenotype. Right: Estimates from IVW MR of the association between genetically proxied Alzheimer’s disease and brain structure phenotypes. Estimates represent the average change in each imaging-derived phenotype (SD units) per doubling (2-fold increase) in the odds of genetically predicted late-onset Alzheimer’s disease. Estimates marked with an asterisk were significant after correction for multiple testing. The SNPs column represents the number of SNPs remaining after clumping for independence and data harmonization. There were no significant clumps for the FA (cortex) phenotype after LD pruning. CT = cortical thickness; FA = fractional anisotropy; HC = hippocampus; IC = intrinsic curvature; ICVF = intracellular volume fraction; LGI = local gyrification index; MC = mean curvature; MD = mean diffusivity; ODI = orientation dispersion index; SA = surface area; Vol = volume; WM = white matter; WMH = white matter hyperintensities.