Understanding Biases in Ribosome Profiling Experiments Reveals Signatures of Translation Dynamics in Yeast

Abstract

Ribosome profiling produces snapshots of the locations of actively translating ribosomes on messenger RNAs. These snapshots can be used to make inferences about translation dynamics. Recent ribosome profiling studies in yeast, however, have reached contradictory conclusions regarding the average translation rate of each codon. Some experiments have used cycloheximide (CHX) to stabilize ribosomes before measuring their positions, and these studies all counterintuitively report a weak negative correlation between the translation rate of a codon and the abundance of its cognate tRNA. In contrast, some experiments performed without CHX report strong positive correlations. To explain this contradiction, we identify unexpected patterns in ribosome density downstream of each type of codon in experiments that use CHX. These patterns are evidence that elongation continues to occur in the presence of CHX but with dramatically altered codon-specific elongation rates. The measured positions of ribosomes in these experiments therefore do not reflect the amounts of time ribosomes spend at each position in vivo. These results suggest that conclusions from experiments in yeast using CHX may need reexamination. In particular, we show that in all such experiments, codons decoded by less abundant tRNAs were in fact being translated more slowly before the addition of CHX disrupted these dynamics.

Fig 1. Experiments with and without CHX pretreatment report different A-site occupancies.

(A) To measure how long a ribosome spends on average with each codon identity in its A-site, footprint sequencing reads (such as the boxed sequence on the top line) are mapped to yeast coding sequences and assigned to the codon position that the A-site of the ribosome was positioned over (red box in coding sequence). For each coding sequence, read counts at each position (shown below each codon) are divided by the average count across the coding sequence to produce relative enrichment values (shown on the bottom line). The average A-site occupancy of each codon identity (in this example, ACT) is then computed by averaging the relative enrichment at all occurrences of the codon (such as the bolded value) across all coding sequences. (B) Scatter plots comparing average A-site occupancies of all 61 non-stop codons between different pairs of experiments. A pair of experiments using CHX pretreatment (upper left) and a pair of experiments done without CHX pretreatment (lower right) report A-site occupancies with strong positive Pearson correlations, but these two internally consistent sets of values are strikingly different from each other (upper right). (C) Pearson correlations of average A-site occupancies of all 61 non-stop codons between representative experiments from many different studies in yeast, grouped by hierarchical clustering. Red indicates positive correlation and blue indicates negative correlation. Clustering separates experiments using CHX pretreatment (labeled in orange) from experiments done without CHX pretreatment (labeled in purple), confirming the generality of the conclusion in (B). Darker labels of each color correspond to those samples compared in (B). (D) Spearman rank correlations of codon identities’ A-site occupancies with the inverse of tRNA adaption index (tAI) in different experiments. A positive correlation represents translation dynamics in which codons decoded by less abundant tRNAs take longer to translate, while a negative correlation implies that codons decoded by less abundant tRNAs are counter-intuitively faster to translate. Experiments using the standard CHX pretreatment protocol report a negative correlation, experiments done without CHX pretreatment report a positive correlation, and three sets of experiments across a gradient of CHX concentrations produced by Gerashchenko [34] each interpolate between these two phenotypes.

Fig 2. CHX pretreatment affects A-site occupancies in a coherent concentration-dependent manner.

Columns correspond to a series of experiments by Gerashchenko [34] using a gradient of concentrations of CHX, starting from no CHX on the far left and increasing to 100 times the standard concentration on the far right. Each column plots the measured A-site occupancies of all 61 non-stop codons for that concentration on a log scale. The width of the line connecting each codon identity across concentrations is scaled by the codon’s net change from no CHX to 100x CHX. The ten codon identities with the largest net changes are labeled. Most notably, the codon identities with the highest enrichments in the experiment with no CHX undergo dramatic, concentration-dependent depletions over the course of the gradient.

Fig 3. CHX pretreatment produces patterns in ribosome density downstream of different codon identities.

(A) To measure how frequently ribosomes are observed with their A-site positioned at a particular offset upstream or downstream of a given codon identity, e.g. 7 codons downstream of an ACT (like the boxed footprint sequence on the top line), relative enrichments at each position are first calculated as in Fig 1A. The enrichment values at all positions located exactly 7 codons downstream of an ACT (such as the bolded value) across all coding sequences are then averaged. (B) Profiles of mean relative enrichments at a range of offsets around three arginine codons in our CHX-pretreatment experiment. Dashed vertical lines mark the boundaries of a canonical 28 nt footprint, and red, blue, and green vertical lines (corresponding to offsets of 0, +1, and +2) mark the A-, P-, and E-sites. Unexpected peaks of enrichment at downstream offsets outside of the dashed lines are observed. The magnitudes of peaks vary substantially between different codon identities encoding the same amino acid, but the horizontal extents of peaks are roughly the same across all codon identities. (C) Profiles of mean relative enrichments around a single codon identity (CGA) in experiments from different studies. There is no downstream peak in the no-CHX-pretreatment experiment of Weinberg et al. (red). Peaks are centered at different offsets in CHX-pretreatment experiments by different groups (green and blue) and become broader and lower when located farther downstream. (D) Profiles of mean relative enrichments around a single codon identity (CGA) in experiments by Gerashchenko [34] using different concentrations of CHX. With no CHX (red), there is a strong enrichment at the A- and P-sites and no downstream peak. With CHX (all other colors), there are depletions at the A- and P-sites and downstream peaks that become closer, narrower, and higher with increasing concentration.

Fig 4. A change in the relative elongation rates of codons produces downstream waves in simulation and analytical models.

(A) In a simulation of the translation of yeast coding sequences, the average relative elongation time of each codon identity was changed from the codon’s A-site enrichment in a no-CHX experiment to its A-site enrichment in a CHX experiment. Elongation was allowed to proceed for a short time after this change, then enrichments in the positions of ribosomes were analyzed as in Fig 3B. The resulting profiles of simulated mean enrichments qualitatively reproduce the downstream peaks in data from experiments using CHX. (B) A model of the translation of a hypothetical coding sequence consisting of a single slow codon surrounded by long stretches of identically faster codons on either side was analyzed. At t = 0 (arbitrary units), the relative speed of the slow codon was changed to be faster than its surroundings. Ribosome density at offsets around the formerly-slow codon is plotted at several time points after this change. Immediately after the change, there is a temporary excess of ribosomes positioned at the formerly-slow codon relative to the eventual steady state of the new dynamics. As these excess ribosomes advance along the coding sequence, a transient wave of increased ribosome density moves downstream and spreads out over time.

Fig 5. Changes in tRNA binding site enrichments between a pair of experiments with and without CHX are matched by areas of downstream waves in the CHX experiment.

For a pair of experiments with and without CHX by Jan [41], the sum of each codon identity’s changes in mean relative enrichment at the A-, P-, and E-sites between the two experiments (green area in insets) is plotted against the total excess or deficit of enrichment in the CHX experiment from 6 to 65 codons downstream (red area in insets). The area of each codon identity’s downstream peak is strongly predicted by changes in enrichment at the tRNA binding sites, consistent with the hypothesis that downstream peaks are transient waves caused by continued elongation with disrupted dynamics in the presence of CHX.

Fig 6. Waves move proportionally farther downstream with increasing CHX pretreatment time.

Each heatmap row shows mean relative enrichments around CGA in a different experiment from Jan et al. [41], with columns corresponding to different offsets. Experiments are separated by CHX pretreatment conditions: either no pretreatment (top group) or the duration and temperature of pretreatment (all other groups). At 30°C, pretreatment for 9 minutes (bottom group) produces waves that have moved slightly more than twice as far downstream as those produced by pretreatment for 4 minutes (middle group).

Fig 7. Downstream waves recover positive correlations of estimated elongation times with 1 / tAI.

In experiments using CHX pretreatment, the combined enrichment of each codon identity at the A-, P-, and E-sites before the introduction of CHX can be estimated by adding the net area of the codon’s downstream wave back to the total remaining enrichments at the three tRNA binding sites. This sum correlates positively with 1 / tAI in all CHX experiments (purple dots), recovering the positive correlations counterintutively absent at the tRNA binding sites alone in these experiments (green dots). Positive correlations are statistically significant (p < 0.05, one-tailed) in most experiments. This suggests that non-optimal codons were being translated less quickly than optimal codons in CHX experiments before the introduction of CHX disrupted these dynamics.

Fig 8. Downstream waves recover the expected effects of lacking tRNA modifications.

(A) Profiles of mean relative enrichments around AAA in two wild type experiments (black lines) and six experiments with different components of the mcm⁵s²U pathway deleted (red lines) from Zinshteyn [23]. All mcm⁵s²U deletion strains produce clearly increased waves downstream of AAA compared to wild type. Darker lines correspond to the experiments compared in (B). (B) Histograms of the net change in enrichment for each codon identity between uba4Δ and wild type at the A-, P-, and E-sites (green) or at the A- P, and E-sites plus 7 to 90 codons downstream (purple). AAA shows a modest increase in net enrichment at the tRNA binding sites, but a dramatically larger increase in net enrichment if the area of downstream waves is also taken into account. This suggests that AAA does take substantially longer to decode in vivo in uba4Δ than in wild type, but that most of this difference disappears during continued elongation in the presence of CHX.