Codon bias among synonymous rare variants is associated with Alzheimer's disease imaging biomarker

Jason E. Miller, Manu K. Shivakumar, Shannon L. Risacher, Andrew J. Saykin, SeungGeun Lee, Kwangsik Nho, Dokyoon Kim

Executive Summary

This research demonstrates that "silent" synonymous mutations—genetic variations that change a DNA sequence without altering the resulting protein's amino acid sequence—play a significant role in the pathology of Alzheimer’s disease (AD). By accounting for codon bias (the efficiency and frequency with which cells use specific genetic codes), the study identified several genes associated with AD neuroimaging biomarkers that standard genomic analyses typically overlook.

The Impact of Synonymous Variants in Alzheimer's Disease

In the search for the genetic drivers of Alzheimer’s disease, researchers traditionally focus on "non-synonymous" variants that alter a protein's structure. Synonymous variants have historically been dismissed as "silent" because the protein sequence remains unchanged.

However, the genetic code is "degenerate," meaning multiple codons can code for the same amino acid. Cells often prefer certain codons over others because some are translated more efficiently or accurately by the ribosome. This study addresses whether these "silent" variants impact disease risk by disrupting the timing of protein folding or mRNA stability. Furthermore, it tackles the "rare variant" problem: individual rare mutations often lack the statistical power for significance, requiring sophisticated "binning" strategies to uncover meaningful associations.

Mechanism of ribosome pausing and protein folding influenced by synonymous codon usage. Rare or non-optimal codons can cause the ribosome to pause, potentially affecting how the nascent protein folds.

Methodological Framework

The researchers employed a multi-stage computational pipeline to link genetics to brain structure using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI):

  1. Variant Annotation & Filtering: Using the Variant Effect Predictor (VEP), they identified synonymous rare variants (Minor Allele Frequency < 0.01) from whole-genome sequencing (WGS) data.
  2. Codon Bias Metrics: Variants were categorized based on Relative Synonymous Codon Usage (RSCU) (genomic frequency) and Optimality (tRNA interaction strength).
  3. Binning with BioBin: Rare variants were grouped by gene to increase statistical power, looking at the cumulative effect of mutations within functional units.
  4. Association Testing: The SKAT-O (Optimal Sequence Kernel Association Test) was used to correlate these gene bins with entorhinal cortical thickness, a brain region that thins early in AD progression.
The multi-stage computational pipeline used to link genetic variants to neuroimaging phenotypes, incorporating VEP annotation, codon bias filtering, and SKAT-O association testing.

Key Insights: Ribosome Pausing and Novel Candidates

A critical finding of the study is that functional filtering is essential. When testing all synonymous variants together, no significant associations were found. The signal only emerged when filtering for variants that specifically changed codon frequency or optimality.

This highlights the biological importance of ribosome pausing. "Non-optimal" codons are not inherently detrimental; they often cause the ribosome to pause, which can be essential for allowing complex proteins to fold correctly. Replacing these with "optimal" codons (or vice versa) can lead to misfolded proteins or altered expression levels. Through this lens, the study identified novel AD candidate genes like MLST8 (a regulator of cell growth) and RHOB (involved in the actin cytoskeleton) that were previously undetected.

Results and Future Directions

The study successfully identified several significant gene associations:

  • RSCU Associations: Genes such as MLST8 and RHOB were associated with decreased entorhinal cortical thickness, while others like FLG2 and CHD6 showed associations with increasing RSCU scores.
  • Optimality Associations: Five genes, including DTL and SERPINB5, showed significant associations when variants shifted from non-optimal to optimal codons.

While these results provide a powerful statistical link, the authors note that "wet lab" functional validation is required to prove these variants directly cause changes in protein folding. Future work will also need to account for tissue-specific codon usage and utilize larger cohorts to further validate these rare variant associations.

Summary updated

Abstract

Alzheimer's disease (AD) is a neurodegenerative disorder with few biomarkers even though it impacts a relatively large portion of the population and is predicted to affect significantly more individuals in the future. Neuroimaging has been used in concert with genetic information to improve our understanding in relation to how AD arises and how it can be potentially diagnosed. Additionally, evidence suggests synonymous variants can have a functional impact on gene regulatory mechanisms, including those related to AD. Some synonymous codons are preferred over others leading to a codon bias. The bias can arise with respect to codons that are more or less frequently used in the genome. A bias can also result from optimal and non-optimal codons, which have stronger and weaker codon anti-codon interactions, respectively. Although association tests have been utilized before to identify genes associated with AD, it remains unclear how codon bias plays a role and if it can improve rare variant analysis. In this work, rare variants from whole-genome sequencing from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort were binned into genes using BioBin. An association analysis of the genes with AD-related neuroimaging biomarker was performed using SKAT-O. While using all synonymous variants we did not identify any genomewide significant associations, using only synonymous variants that affected codon frequency we identified several genes as significantly associated with the imaging phenotype. Additionally, significant associations were found using only rare variants that contains an optimal codon in among minor alleles and a non-optimal codon in the major allele. These results suggest that codon bias may play a role in AD and that it can be used to improve detection power in rare variant association analysis.
BibTeX
@inproceedings{miller2018codon,
  author = {Jason E. Miller and Manu K. Shivakumar and Shannon L. Risacher and Andrew J. Saykin and SeungGeun Lee and Kwangsik Nho and Dokyoon Kim},
  title = {{Codon bias among synonymous rare variants is associated with Alzheimer's disease imaging biomarker}},
  booktitle = {Psb},
  year = {2018},
}
Figure