Phenome-Wide Association of APOE Alleles in the All of Us Research Program

Ehsan Khajouei, Valentina Ghisays, Ignazio S. Piras, Kiana L. Martinez, Marcus Naymik, Preston Ngo, Tam C. Tran, Joshua E. Denny, Travis J. Wheeler, Matthew J. Huentelman, Eric M. Reiman, Jason H. KarnesmedRxiv 2024 · DOI

TL;DR

This study leverages the "All of Us" research program's diverse dataset to conduct a phenome-wide association study (PheWAS) of APOE alleles. It confirms known associations with conditions like Alzheimer's disease and hyperlipidemia while uncovering unique associations across different sexes and ancestries, highlighting the importance of diverse populations in genetic research.

What Problem Does This Paper Address?

Genetic variations in the Apolipoprotein E (APOE) gene are well-established risk factors for various diseases, particularly Alzheimer's disease (AD), cardiovascular conditions, and metabolic disorders. However, most previous studies have been limited to European ancestry populations, failing to capture the full spectrum of APOE's influence across diverse groups and its differential impact based on sex and ancestry. This lack of diversity hinders our understanding of how genetic predispositions interact with sociocultural factors and exacerbates existing health inequalities. This paper aims to address this gap by using a phenome-wide association study (PheWAS) approach on a large, diverse cohort to explore APOE-associated phenotypes across various demographic groups.

Methods

The researchers employed a phenome-wide association study (PheWAS) approach. This involved:

  • APOE Genotyping: APOE alleles were determined for participants using whole genome sequencing data.
  • Phenotype Extraction: Electronic Health Record (EHR) data was used to derive 2,318 distinct phenotypes, mapped using the Phecode system.
  • Statistical Analysis: Logistic regression was used to test associations between APOE variants (treated as an ordinal variable based on AD risk hierarchy) and the derived phenotypes.
  • Covariate Adjustment: Analyses were adjusted for age, sex at birth, EHR length, and 16 genetic principal components to account for population stratification.
  • Subgroup Analyses: The study performed analyses stratified by ancestry (seven groups) and sex, and also included sensitivity analyses adjusting for social determinants of health (SDOH).
  • Bonferroni Correction: A stringent Bonferroni correction was applied to account for multiple comparisons, ensuring statistical significance.
  • Heterogeneity Testing: The study assessed heterogeneity of effect sizes across different sex and ancestry groups.

Datasets

The primary dataset used in this study was the "All of Us" Research Program dataset. This dataset includes:

  • Genomic Data: Short-read whole genome sequencing (srWGS) data for a substantial portion of participants.
  • Electronic Health Record (EHR) Data: Longitudinal EHR data from over 50 healthcare providers, standardized using the OMOP Common Data Model.
  • Participant Demographics: Information on age, sex assigned at birth, and inferred genetic ancestry.

The "All of Us" program is a large, diverse biobank aiming to enroll at least one million people living in the U.S. The data used in this study was from release version 7 (CDRv7), containing data on 413,450 participants.

Key Insights and Ideas

  • Diversity Matters: The study powerfully demonstrates that APOE's influence on health phenotypes is not uniform across different ancestries and sexes, underscoring the critical need for diverse datasets in genetic research.
  • Beyond Alzheimer's: While confirming APOE's strong link to Alzheimer's disease and related dementias, the PheWAS approach revealed a broader spectrum of associated conditions, including metabolic and cardiovascular diseases.
  • Ancestry-Specific Associations: Unique associations were identified in specific ancestral groups, such as transient retinal arterial occlusion in the European ancestry group and first-degree atrioventricular block in the American Admixed/Latino group, suggesting distinct genetic or environmental interactions.
  • Sex-Specific Nuances: While many associations were similar between sexes, some differences emerged, with females showing reduced odds of certain noninflammatory disorders and males showing increased odds of cardiovascular conditions.
  • SDOH Impact: Preliminary analysis suggested that social determinants of health had a limited direct impact on APOE-associated disease risk, though this area warrants further comprehensive investigation.

Key Results

  • Overall Cohort: 17 significant associations were identified after Bonferroni correction.
  • Increased Risk: Hyperlipidemia (OR 1.15), Alzheimer's disease (OR 1.55), and dementias were strongly associated with APOE alleles.
  • Reduced Risk: Fatty liver disease (OR 0.93) and chronic liver disease showed reduced odds.
  • Sex-Specific Findings:
    • Females showed reduced odds for noninflammatory disorders of vulva and perineum (OR 0.89).
  • Males showed increased odds for atherosclerotic heart disease (OR 1.05) and ischemic heart disease (OR 1.05).
  • Ancestry-Specific Findings:
    • European ancestry: Transient retinal arterial occlusion (OR 1.31).
  • American Admixed/Latino ancestry: First-degree atrioventricular block (OR 1.74).
  • Heterogeneity: Significant heterogeneity in effect sizes was observed for hyperlipidemia and mild cognitive impairment across ancestry groups.

Limitations and Future Work

  • EHR Data Accuracy: The study acknowledges limitations in the accuracy of EHR-derived phenotypes, particularly for complex conditions like neurodegenerative diseases, which may lead to underestimated odds ratios.
  • Sample Size for Subgroups: Smaller ancestry groups (EAS, MID, SAS) had reduced statistical power, limiting the number of phenotypes analyzed and potentially missing significant associations. This also applied to a lesser extent for AFR, OTH, and AMR.
  • Site-Specific Patterns: Differences in billing practices, data upload, and clinical assessments across different sites could have introduced confounding factors.
  • SDOH Comprehensiveness: While adjusted for some SDOH factors, the investigation was not exhaustive, and more detailed SDOH data could provide deeper insights into biological versus sociocultural influences.
  • Rare Alleles: The study lacked the power to detect associations with the extremely rare APOE ε1 allele.

Future Work: The authors suggest the need for replication of hypothesis-generating results, particularly for noninflammatory disorders of the vulva and perineum and AV block. Increasing sample sizes for smaller ancestry groups is also crucial for more robust findings. Further research into the complex interplay of APOE, sex, ancestry, and SDOH is recommended. The development of scalable blood-based biomarker endophenotypes could also help improve diagnostic accuracy for conditions like AD.

Summary updated

Abstract

We replicate extensive phenotypic associations with <i>APOE</i> alleles in a large, diverse cohort, despite limitations in accuracy for EHR-derived phenotypes. We provide a comprehensive catalog of <i>APOE</i>-associated phenotypes and present evidence of unique phenotypic associations by sex and ancestry, as well as heterogeneity in effect size across ancestry.
BibTeX
@misc{khajouei2024phenome,
  author = {Ehsan Khajouei and Valentina Ghisays and Ignazio S. Piras and Kiana L. Martinez and Marcus Naymik and Preston Ngo and Tam C. Tran and Joshua E. Denny and Travis J. Wheeler and Matthew J. Huentelman and Eric M. Reiman and Jason H. Karnes},
  title = {{Phenome-Wide Association of <i>APOE</i> Alleles in the <i>All of Us</i> Research Program}},
  year = {2024},
  doi = {10.1101/2024.09.04.24313010},
  howpublished = {medRxiv},
}
Figure