Estimating Selection on Synonymous Codon Usage from Noisy Experimental Data

Edward W. Wallace, Edoardo M. Airoldi, D. Allan DrummondMolecular 2013 · DOI

Abstract

A key goal in molecular evolution is to extract mechanistic insights from signatures of selection. A case study is codon usage, where despite many recent advances and hypotheses, two longstanding problems remain: the relative contribution of selection and mutation in determining codon frequencies and the relative contribution of translational speed and accuracy to selection. The relevant targets of selection--the rate of translation and of mistranslation of a codon per unit time in the cell--can only be related to mechanistic properties of the translational apparatus if the number of transcripts per cell is known, requiring use of gene expression measurements. Perhaps surprisingly, different gene-expression data sets yield markedly different estimates of selection. We show that this is largely due to measurement noise, notably due to differences between studies rather than instrument error or biological variability. We develop an analytical framework that explicitly models noise in expression in the context of the population-genetic model. Estimates of mutation and selection strength in budding yeast produced by this method are robust to the expression data set used and are substantially higher than estimates using a noise-blind approach. We introduce per-gene selection estimates that correlate well with previous scoring systems, such as the codon adaptation index, while now carrying an evolutionary interpretation. On average, selection for codon usage in budding yeast is weak, yet our estimates show that genes range from virtually unselected to average per-codon selection coefficients above the inverse population size. Our analytical framework may be generally useful for distinguishing biological signals from measurement noise in other applications that depend upon measurements of gene expression.

Introduction: The Challenge of Measuring Silent Selection

In the genetic code, multiple "synonymous" codons can encode the same amino acid. While these mutations do not alter the resulting protein sequence, they are not evolutionarily neutral. Natural selection often favors specific codons to optimize the speed and accuracy of protein synthesis (translation). Quantifying this selection pressure is essential for understanding molecular evolution, but it relies heavily on gene expression data (mRNA levels), as highly expressed genes face stronger pressure for translational efficiency.

The Impact of Measurement Noise

A significant hurdle in this field is that mRNA measurements are notoriously noisy. While measurements within a single laboratory are typically highly correlated, correlations between different labs or experimental platforms can drop significantly, sometimes as low as 0.6. This "proxy error"—the discrepancy between experimental measurements and the actual translational environment that shaped the genome over millions of years—leads to attenuation bias. In standard models, this measurement error systematically pushes estimates of selection strength toward zero, causing researchers to significantly underestimate the influence of selection on "silent" DNA sites.

Comparison of gene expression correlation within and between laboratories, highlighting the significant noise present in cross-study data.

Biological Context: Speed vs. Accuracy

Selection on codon usage is primarily driven by the trade-off between translational speed and accuracy. Efficient codons are translated faster and with fewer errors, reducing the metabolic cost of protein production and the risk of protein misfolding. These mechanistic properties of the translational apparatus are the ultimate targets of selection.

Conceptual diagram of the trade-offs between translational speed and accuracy that drive selection on codon usage.

Methods: A Noise-Aware Bayesian Framework

The authors introduced a Bayesian hierarchical model to account for this inherent noise. Instead of treating observed mRNA levels as ground truth, the model treats "true" gene expression as a latent variable.

  1. The Population-Genetic Core: A multinomial logistic regression model relates codon frequencies to mutation bias and selection strength.
  2. The Noise Model: Observation error is modeled as lognormal, acknowledging that measurements are imperfect.
  3. The Latent Distribution: Underlying "true" expression levels are modeled using a log-asymmetric Laplace distribution to capture the long-tailed nature of gene expression.
  4. Inference: Markov Chain Monte Carlo (MCMC) sampling is used to simultaneously estimate selection coefficients, mutation biases, and the magnitude of the measurement noise itself.

Key Results

Applying this framework to Saccharomyces cerevisiae genomic and expression data yielded several transformative insights:

  • Corrected Selection Estimates: Selection coefficients inferred with the new method were 1.7x to 4.7x higher than those from noise-blind approaches, suggesting that previous methods significantly underestimated the strength of selection.
  • Convergence Across Datasets: Perhaps the most striking result is that while different expression datasets originally gave wildly different selection estimates, the noise-aware model caused these estimates to converge. This suggests the model is capturing a true biological signal rather than experimental artifacts.
  • Predictive Accuracy: The model’s inferred mutation biases showed a high correlation ($R^2 = 0.77$) with direct experimental measurements of mutation rates, providing strong external validation.
  • Evolutionary Interpretation: The study introduced SCU (Selection on Codon Usage) and mSCU (mean SCU), metrics that carry a direct evolutionary meaning related to fitness and effective population size ($N_e$), unlike traditional scores like the Codon Adaptation Index (CAI).

Discussion and Future Directions

This study demonstrates that accounting for measurement noise is essential for extracting biological signals from genomic data. The model identified nearly 250 genes where selection is strong enough to overcome random genetic drift (selection coefficient $> 1/N_e$).

While the framework provides more robust estimates, some limitations remain. The absolute values of selection coefficients still depend on estimates of total mRNA transcripts per cell, which vary between studies. Furthermore, the current model assumes selection is uniform across a gene's length and uses a simplified mutation model. Future iterations could incorporate site-specific selection (e.g., for proper protein folding or 5' end requirements) and more complex, context-dependent mutational signatures.

Summary updated

Abstract

A key goal in molecular evolution is to extract mechanistic insights from signatures of selection. A case study is codon usage, where despite many recent advances and hypotheses, two longstanding problems remain: the relative contribution of selection and mutation in determining codon frequencies and the relative contribution of translational speed and accuracy to selection. The relevant targets of selection--the rate of translation and of mistranslation of a codon per unit time in the cell--can only be related to mechanistic properties of the translational apparatus if the number of transcripts per cell is known, requiring use of gene expression measurements. Perhaps surprisingly, different gene-expression data sets yield markedly different estimates of selection. We show that this is largely due to measurement noise, notably due to differences between studies rather than instrument error or biological variability. We develop an analytical framework that explicitly models noise in expression in the context of the population-genetic model. Estimates of mutation and selection strength in budding yeast produced by this method are robust to the expression data set used and are substantially higher than estimates using a noise-blind approach. We introduce per-gene selection estimates that correlate well with previous scoring systems, such as the codon adaptation index, while now carrying an evolutionary interpretation. On average, selection for codon usage in budding yeast is weak, yet our estimates show that genes range from virtually unselected to average per-codon selection coefficients above the inverse population size. Our analytical framework may be generally useful for distinguishing biological signals from measurement noise in other applications that depend upon measurements of gene expression.
BibTeX
@inproceedings{wallace2013estimating,
  author = {Edward W. Wallace and Edoardo M. Airoldi and D. Allan Drummond},
  title = {{Estimating Selection on Synonymous Codon Usage from Noisy Experimental Data}},
  booktitle = {Molecular},
  year = {2013},
  doi = {10.1093/molbev/mst051},
}
Figure