Trade-offs between tRNA abundance and mRNA secondary structure support smoothing of translation elongation rate

Abstract

Translation of protein from mRNA is a complex multi-step process that occurs at a non-uniform rate. Variability in ribosome speed along an mRNA enables refinement of the proteome and plays a critical role in protein biogenesis. Detailed single protein studies have found both tRNA abundance and mRNA secondary structure as key modulators of translation elongation rate, but recent genome-wide ribosome profiling experiments have not observed significant influence of either on translation efficiency. Here we provide evidence that this results from an inherent trade-off between these factors. We find codons pairing to high-abundance tRNAs are preferentially used in regions of high secondary structure content, while codons read by significantly less abundant tRNAs are located in lowly structured regions. By considering long stretches of high and low mRNA secondary structure in Saccharomyces cerevisiae and Escherichia coli and comparing them to randomized-gene models and experimental expression data, we were able to distinguish clear selective pressures and increased protein expression for specific codon choices. The trade-off between secondary structure and tRNA-concentration based codon choice allows for compensation of their independent effects on translation, helping to smooth overall translational speed and reducing the chance of potentially detrimental points of excessively slow or fast ribosome movement.

Figure 1.

Trade-offs between mRNA secondary structure and tRNA abundance can potentially smoothen overall elongation rate. (a) mRNA secondary structure and cognate aa-tRNA abundance each affect elongation speed. (b) By trading-off the negative effect of one factor with the positive effect of the other, a more constant elongation rate can be maintained along a transcript, reducing the chance of excessively slow or fast regions.

Figure 2.

Distributions of predicted codon translation times for protein coding regions of mRNAs in the Saccharomyces cerevisiae and Escherichia coli genomes. Upper plots (red) show regions with high mRNA secondary structure and lower plots (blue) regions with low mRNA secondary structure. All analyzed sequences had a minimal region length of 20 bp. Green vertical lines and the value in top right corner denote the median value for the distribution. The shifts in codon translation times are statistically significant in both cases, see Table 1 for details.

Figure 3.

Elongation speed and secondary structure propensity are anti-correlated as exemplified by the aceB gene in Escherichia coli. Codon elongation times are calculated using the method of Zhang et al. (4). Light gray vertical bars represent the individual codon elongations times and the red line the smoothed translational profile. Dips correspond to regions of slow translation. The dotted horizontal line represents a threshold under which translational pausing is thought to occur (4). mRNA secondary structure predicted using Vienna RNAFold is shown by the blue line. Lower values correspond to stronger secondary structure. The first and last 17 codons are removed from our analysis (dark shaded regions).

Figure 4.

Codon usage for protein coding regions of mRNAs with high and low mRNA secondary structure across the Saccharomyces cerevisiae and Escherichia coli genomes. Separate plots are displayed for each amino acid, with upper plots (red) showing regions with high mRNA secondary structure and lower plots (blue) regions with low mRNA secondary structure. Each plot displays the normalized usage of each codon for the associated amino acid on a scale from 0 to 0.08 and codons are sorted from left to right in terms of predicted translational times, see Supplementary Table S1 (stop codons are excluded), fastest (left) to slowest (right). Notice a clear bias toward faster codons in regions with high secondary structure that is not always related to G/C content in the second and third positions (e.g. in S. cerevisiae see the CCA codon for proline and GAA codon for glutamic acid, both of which display a strong bias in highly structured regions even though other synonymous codons with greater G/C content exist they are, however, predicted to be the fastest translated). Horizontal grid lines are positioned at intervals of 0.01. Number in top left corner of each plot denotes the percentage of individual codons shown in the plot in relation to the total number of codons contained within the total associated low or high mRNA secondary structure regions. For example in E. coli, codons coding for alanine in low secondary structure regions correspond to 6.6% of all codons in low structured regions, while the same codons in high secondary structure regions correspond to 13.7% of all codons in high structured regions.

Figure 5.

A strong selection pressure operates on the Escherichia coli genome to select for trade-offs between tRNA abundance and mRNA secondary structure. Time difference is calculated as the difference of the median predicted codon translation time between high and low structured regions. Greater negative values are evidence that these factors trade-off their contributions to smoothen translational speed. The time difference for E. coli is marked on the axis. The clear separation between both distributions and the native E. coli genome is significant in all cases.

Figure 6.

Synonymous codon choice in a synthetic gene effects expression level. Time difference is calculated as in Figure 5 as the difference in predicted codon translation time between high and low structured regions. Greater negative values correspond to stronger trade-offs between tRNA abundance and mRNA secondary structure which exhibits a significant negative correlation (R = −0.5908, P = 0.0005) with the absolute protein expression. The red line represents a linear least squares regression fit (a = −0.0869); points are colored (red = low, green = high) and labeled in relation to the codon adaptation index (CAI) of each variant.