University of Edinburgh
Deciphering the causal effects of coding sequence features on multiple stages of gene expression across human cell types
Abstract
dc:description.abstractHuman gene expression is regulated by multiple coding sequence properties. GC3, CpG contents, and codon adaptation have emerged as key determinants of gene expression. However, in the human genome, these features are highly intercorrelated, making it challenging to determine their individual contribution to the different stages of gene expression across various human tissues. To estimate the precise contribution of each coding sequence property and isolate their effects, I used a synthetic library of 194 synonymous Green Fluorescence Protein (GFP) variants, with varying GC3, CpG contents, and Codon Adaptation Index (CAI), and analyzed their expression levels at multiple stages across several human cell lines originating from different tissues. The primary analysis revealed that GC3 content is the major determinant of the steady-state mRNA levels. CpG content negatively influences the mRNA stability and translation, while the codon adaptation positively affects them, independently of the GC3 content. These effects of coding sequence properties on gene expression were consistent across cell types. While steady-state mRNA and protein level patterns were highly correlated, translation efficiency showed moderate variability between cell lines, indicating that tissue-specific codon preferences might play a role in fine-tuning translation rates. To further study these effects of codon preferences on different stages of gene expression, I developed a new metric, the Codon Expression Coefficient (CEC), defined as Pearson’s correlation between codon frequencies and expression levels, and measures the effects of individual sense codons on gene expression. CECs revealed that the effects of codons were similar on steady-state mRNA and protein levels across different cell lines. In contrast, their effects on translation efficiency were not only variable from mRNA & protein levels, but they were also cell type specific. Next, I used machine learning models to predict GFP protein levels directly from their coding sequences. These models were trained using empirical GFP protein level data, codon frequencies, and nine additional sequence-derived features. Remarkably, a single Random Forest model, exclusively based on codon frequencies, successfully explained approximately 77% of the variance in the observed protein levels. Leveraging this codon frequency-based Random Forest model, I benchmarked several codon optimization tools, revealing that optimizers prioritizing host codon adaptiveness are the best-performing optimizers. Finally, using 194 synonymous GFP variants, I also uncovered a novel relationship between codon optimality and mRNA poly(A) tail lengths. I found that higher codon optimality is associated with shorter poly(A) tails. Interestingly, these mRNAs with shorter poly(A) tails are the most abundant, highly stable, and highly translated mRNAs. This revealed a new layer of gene regulation encoded within the genetic code. Overall, these findings disentangle the complex knot of regulatory grammar, where multiple coding sequence properties orchestrate to fine-tune the gene expression at transcriptional, post-transcriptional, and translational levels.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ahmad, Mohd
- Advisors dc:contributor.advisor
-
- Kudla, Grzegorz
- Wood, Andrew
Subjects
dc:subject × 5Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- https://doi.org/10.7488/era/7491