Cortical structure and the risk for Alzheimer’s disease: a bidirectional Mendelian randomization study

The study addresses whether variations in cortical structure (surface area and cortical thickness in the whole cortex and 34 regions) causally influence the risk of Alzheimer’s disease (AD), and conversely, whether genetic liability to AD causally affects cortical structure. Observational neuroimaging studies have linked cortical atrophy in regions such as the hippocampus, entorhinal cortex, medial temporal lobe, precuneus, and orbitofrontal cortex with AD, and distinct atrophy patterns have been used to define AD subtypes. However, confounding (notably aging) and reverse causation limit causal inference from observational data. Given that AD neuropathology precedes clinical symptoms, clarifying whether cortical changes are causes or consequences of AD is important for understanding disease etiology and progression. The authors use bidirectional Mendelian randomization (MR) to infer causality between genetically proxied cortical structure and AD risk.

Prior research consistently reports regional cortical atrophy in AD, including hippocampus, entorhinal cortex, medial temporal regions, precuneus, and orbitofrontal cortex, and identifies heterogeneous atrophy patterns corresponding to AD subtypes (typical, limbic-predominant, hippocampal-sparing). Selective neuronal vulnerability has been proposed as a hallmark of neurodegenerative diseases, and posterior cortical atrophy is a recognized AD-related syndrome. Observational associations are susceptible to confounding by age-related brain changes and reverse causation, highlighting the need for approaches like MR that leverage genetic instruments to strengthen causal inference.

Design: Bidirectional two-sample Mendelian randomization (MR) to estimate causal effects between cortical structure phenotypes (surface area and thickness for whole cortex and 34 Desikan–Killiany regions) and Alzheimer’s disease (AD). Data sources: Cortical structure GWAS meta-analysis in 33,992 Europeans (23,909 from 49 ENIGMA cohorts and 10,083 UK Biobank). AD genetic data from a meta-analysis including clinically diagnosed AD and AD-by-proxy (parental diagnosis) with high genetic correlation to AD (r=0.81), totaling 10,880 AD cases and 383,378 controls across ADSP, IGAP, PGC-ALZ, and UK Biobank (overall mGWAS n≈455,258). Instrument selection: For cortical structure→AD analyses, SNPs associated with cortical traits at p<1×10^-6 were considered; for reverse (AD→cortical structure), genome-wide significant SNPs (p≤5×10^-8) were used. LD clumping was applied (reported as r² > 0.001 in the scheme) to obtain independent instruments; correlated SNPs were removed. Instrument strength was evaluated (F-statistics >10 targeted), and NO measurement error (NOME) assumption assessed (τ0). Harmonization aligned effect alleles across exposure and outcome summary statistics. Primary causal estimator: Inverse-variance weighted (IVW) method (fixed/random-effects as appropriate) under the assumption of valid instruments. Sensitivity analyses: MR-Egger regression (with non-zero intercept to detect directional pleiotropy), weighted median estimator (robust if up to 50% invalid instruments), MR-PRESSO global test with outlier removal to detect and correct for horizontal pleiotropy, leave-one-out analyses and single-SNP analyses to assess influence of individual instruments, and Cochran’s Q tests for heterogeneity. Power, R², and F-statistics were computed for key analyses. Outcomes were interpreted with Bonferroni correction across multiple cortical traits.

- Forward MR (cortical structure→AD): Suggestive association between smaller temporal pole surface area and higher AD risk: OR 0.95 (95% CI 0.90–0.997), p=0.04 per 1-SD increase in surface area; sensitivity analyses (MR-Egger, weighted median) showed consistent direction; no evidence of directional pleiotropy; rs68552426 strongly influenced this estimate. Suggestive associations for larger surface area of lateral orbitofrontal, supramarginal, and lingual with higher AD risk: lateral orbitofrontal OR 1.04 (1.01–1.08), p=0.022; supramarginal OR 1.05 (1.01–1.09), p=0.008; lingual OR 1.03 (1.004–1.06), p=0.024. The lateral orbitofrontal association was driven by SNP rs752248 (CNV2 gene) and rs132802348 (RRB8); removing rs752248 attenuated the association to non-significance (OR 1.03 (0.99–1.07), p=0.088). No heterogeneity detected for key regions by Cochran’s Q. - Forward MR (cortical thickness→AD): Suggestive association where greater lateral occipital thickness increased AD risk: β=0.022 (SE 0.008), p=0.014; MR-Egger consistent (β=0.022 (0.008), p=0.008) without pleiotropy evidence. Cuneus thickness findings elsewhere suggest increased thickness associates with decreased AD risk (consistent with abstract). - Reverse MR (AD liability→cortical surface area): Genetic liability to AD was suggestively associated with decreased surface area of precentral (β=-43.4, SE 21.3, p=0.042) and isthmus cingulate (β=-18.5, SE 7.3, p=0.011). Additional occipital lobe regions (cuneus, pericalcarine, lateral occipital, lingual) showed patterns suggestive of genetically predicted AD associating with increased cortical measurements in some analyses. - Multiple-testing: None of the associations survived Bonferroni correction. - Pleiotropy/heterogeneity: Generally no evidence of substantial horizontal or directional pleiotropy by MR-Egger intercept or MR-PRESSO global tests; little heterogeneity observed; some results were influenced by single SNPs near known AD-related genes (e.g., rs442495 near ADAM10).

The bidirectional MR analyses provide suggestive causal links between specific cortical structural features and AD risk, addressing the central question of whether observed atrophy patterns are causes or consequences of AD. The inverse association between temporal pole surface area and AD risk aligns with observed temporal atrophy in AD and suggests that reduced temporal pole integrity may play a role in disease susceptibility or early pathophysiology. Findings regarding cuneus thickness support the notion that posterior cortical changes are relevant in AD, consistent with posterior cortical atrophy phenotypes. Conversely, evidence that AD genetic liability may reduce precentral and isthmus cingulate surface areas implies that aspects of AD pathogenesis contribute to structural decline in motor and limbic-associated regions, which corresponds to reported motor impairments and limbic-predominant atrophy patterns in AD. Some unexpected patterns in occipital regions (e.g., increased thickness or surface area associated with AD liability or risk) may reflect early amyloid-related processes (e.g., space-occupying effects of plaques) or neurogenesis-related mechanisms, highlighting complex, stage-dependent structural responses in AD. Overall, while results are suggestive and did not withstand stringent multiple-testing correction, they reinforce region-specific cortical involvement in AD and support a partially bidirectional relationship between cortical morphology and AD risk.

Using large-scale GWAS and bidirectional two-sample MR, the study provides suggestive evidence that smaller temporal pole surface area and thinner cuneus are associated with higher AD risk, and that genetic liability to AD may causally reduce the surface area of the precentral gyrus and isthmus cingulate. These findings support region-specific cortical involvement in AD etiology and progression. Future research should include larger GWAS of cortical traits to strengthen instruments, explore additional neuroanatomical metrics beyond surface area and thickness, dissect AD subtypes and stage-specific effects, and perform functional studies to elucidate biological mechanisms linking these regions to AD.

- Statistical significance: No associations survived Bonferroni correction across multiple cortical traits; results should be considered suggestive. - Instrument strength and variance explained: Cortical trait instruments explain a small fraction of variance; some analyses used SNPs at p<1×10^-6 (less stringent than genome-wide significance), increasing risk of weak instrument bias. - Potential pleiotropy and SNP influence: Although formal tests did not indicate strong pleiotropy, several associations were sensitive to single SNPs (e.g., rs752248, rs68552426, rs442495), which may bias estimates. - Phenotype scope: Only cortical surface area and thickness were analyzed; other brain structural and microstructural measures were not assessed. - Generalizability: Analyses were restricted to individuals of European ancestry. - Complexity of AD phenotypes: Inclusion of AD-by-proxy and inability to stratify by AD subtypes (e.g., APOE ε4 status, atypical presentations) may obscure region-specific causal effects. - Sample size constraints for cortical GWAS may limit power to detect small effects and to apply stringent instrument thresholds.