Genetic Association Analyses of Longitudinal Cognitive Changes Related to Alzheimer’s Disease in Diverse Populations

Xian Wu, Khine Zin Aung, Erin L. Abner, Peter T. Nelson, David W. Fardo, Yuriko KatsumataAlzheimer 2024 · DOI

TL;DR

This study investigates genetic associations with cognitive decline in Alzheimer's Disease (AD) across different ancestral groups. By employing a Tobit model to account for the ceiling effect in cognitive tests, researchers identified specific genetic variants linked to initial cognitive performance and longitudinal changes in non-Hispanic Whites, African Americans, and Hispanics, highlighting the importance of population-specific genetic risk factors.

What Problem Does This Paper Address?

Late-onset Alzheimer's Disease (LOAD) is known to have a significant genetic component. However, recent research suggests that the genetic risks for LOAD might differ across various populations. To better understand these population-specific risks, this study focuses on analyzing the association between genetic variants and cognitive changes over time, as measured by the Mini-Mental State Examination (MMSE). A key challenge with the MMSE is its "ceiling effect," where the maximum score of 30 can obscure subtle cognitive declines in individuals who consistently score at the top. Traditional linear models struggle with this, leading the authors to use a more appropriate statistical approach.

Methods

The researchers utilized a mixed-effects Tobit model, a statistical method that is particularly well-suited for data with a ceiling effect. This model was applied to analyze the association between 82 AD-related single nucleotide variants (SNVs) and MMSE scores. To identify distinct genetic ancestry (GA) groups within their dataset, they employed Principal Component Analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP). Association analyses were then conducted separately for each GA group, adjusting for factors like age, sex, and education. A stringent Bonferroni correction was used to account for multiple testing, setting the significance threshold at P = 6.10 × 10⁻⁴.

Datasets

  • Phenotype Data: National Alzheimer's Coordinating Center (NACC) Uniform Data Set (UDS) - September 2022 data freeze.
  • Genotype Data: Alzheimer's Disease Genetics Consortium (ADGC).

The study analyzed genetic variants previously identified by a genome-wide association study (Bellenguez et al., 2022).

Key Insights and Ideas

  • Population-Specific Genetic Risks: The most significant takeaway is that genetic associations with cognitive function in AD are not uniform across all populations. Different genetic variants appear to play a role in different ancestral groups.
  • Tobit Model Superiority: The use of a Tobit model is crucial for accurately capturing cognitive changes, especially when scores are at or near the maximum. This approach likely reveals associations that might be missed by standard linear models.
  • BIN1 as a Common Factor (but not entirely): While BIN1 showed an association with initial cognitive performance in both non-Hispanic Whites and African Americans, its role in Hispanics was less pronounced for initial performance, suggesting nuances even within shared variants.
  • LILRB2's Unique Role: The identification of LILRB2 specifically associated with initial cognitive performance in Hispanics points to a potentially unique genetic influence in this population.

Key Results

The study analyzed three main genetic ancestry (GA) groups: non-Hispanic White (NHW, n=15,112), African American (AA, n=2,392), and Hispanic (n=1,159).

  • Initial Cognitive Performance (ICP):
    • In NHW, 18 genes (including CR1, SORT1, PRKD3, BIN1, MME, EED, and ABCA7) were associated with ICP.
  • In AA, only BIN1 was associated with ICP.
  • In Hispanics, LILRB2 was associated with ICP.
  • Change in MMSE Scores Over Time:
    • CR1 was associated with changes in MMSE scores in NHW.
  • NCK2 was associated with changes in MMSE scores in AA.
  • EED was associated with changes in MMSE scores in Hispanics.

Limitations and Future Work

  • Sample Size for AA and Hispanic Groups: While the study included these groups, their sample sizes are considerably smaller than the NHW group, which may limit the power to detect all relevant genetic associations.
  • Focus on Specific SNVs: The study focused on 82 pre-selected AD-related SNVs. Future work could expand to a genome-wide analysis to discover novel genetic associations.
  • Further Validation: The authors suggest that novel genetic risks may vary across GA groups and that these findings will be further revealed in future studies, implying a need for replication and expansion of these analyses.

Summary updated

Abstract

Abstract Background Late‐onset Alzheimer’s disease (LOAD) is highly heritable. 1 Recent research suggested that population‐specific LOAD genetic risks may exist. 2 The Mini‐Mental State Examination (MMSE; a measure of global cognitive function) has been commonly used to monitor AD‐related cognitive changes. MMSE raw scores have a strong ceiling effect (upper limit = 30 points). In a previous study, we demonstrated that the Tobit model utilizing ceiling information for estimation, is a superior approach compared to the linear model. 3 In this study, we aim to employ Tobit modeling to investigate associations of single nucleotide variants (SNVs) with MMSE scores to identify population‐specific genetic risks. Method The phenotype data were drawn from the National Alzheimer’s Coordinating Center (NACC) Uniform Data Set (UDS) September 2022 data freeze. The genotype data were obtained from the Alzheimer’s Disease Genetics Consortium (ADGC). Alongside self‐reported race/ethnicity, we applied the principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) to identify genetic‐ancestry (GA) groups (Fig. 1). Within each GA group, we performed association analyses of 82 AD‐related SNVs identified by a genome‐wide association study (Bellenguez et al., 2022), on MMSE using mixed‐effect Tobit modeling, adjusting for age at baseline, sex, education, and 2‐4 PC scores. We utilized Bonferroni correction to set P = 6.10 × 10 −4 for multiple testing. Result Three GA groups were included in analyses: non‐Hispanic White (NHW, n = 15,112), African American (AA, n = 2,392), and Hispanic (n = 1,159). Genetic variants were associated with MMSE shown in Table 1. At baseline, in NHW, genes CR1, SORT1, PRKD3, BIN1, INPPSD, MME, UNC5CL, CD2AP, TMEM106B, SPDYE, USP6NL, MS4A4A, EED, SLC24A4, SPPL2A, MAF, PLCG2, and ABCA7 were associated with initial cognitive performance (ICP). In contrast, only BIN1 showed an association with ICP in AA, while LILRB2 was associated with ICP in Hispanics. Moreover , CR1 in NHW, NCK2 in AA, and EED in Hispanics were associated with change in MMSE scores over time. Conclusion Our findings underscore the importance of considering population‐specific genetic variants when studying AD‐related longitudinal cognitive changes. Novel genetic risks may vary across GA groups and be further revealed in future studies.
BibTeX
@inproceedings{wu2024genetic,
  author = {Xian Wu and Khine Zin Aung and Erin L. Abner and Peter T. Nelson and David W. Fardo and Yuriko Katsumata},
  title = {{Genetic Association Analyses of Longitudinal Cognitive Changes Related to Alzheimer’s Disease in Diverse Populations}},
  booktitle = {Alzheimer},
  year = {2024},
  doi = {10.1002/alz.095497},
}
Figure