Positively Charged Residues Are the Major Determinants of Ribosomal Velocity

Both for understanding mechanisms of disease and for the design of transgenes, it is important to understand the determinants of ribosome velocity, as changes in the rate of translation are important for protein folding, error attenuation, and localization. While there is great variation in ribosomal occupancy along even a single transcript, what determines a ribosome’s occupancy is unclear. We examine this issue using data from a ribosomal footprinting assay in yeast.


Both for understanding mechanisms of disease and for the design of transgenes, it is important to understand the determinants of ribosome velocity, as changes in the rate of translation are important for protein folding, error attenuation, and localization. While there is great variation in ribosomal occupancy along even a single transcript, what determines a ribosome’s occupancy is unclear. We examine this issue using data from a ribosomal footprinting assay in yeast. While codon usage is classically considered a major determinant, we find no evidence for this. By contrast, we find that positively charged amino acids greatly retard ribosomes downstream from where they are encoded, consistent with the suggestion that positively charged residues interact with the negatively charged ribosomal exit tunnel. Such slowing is independent of and greater than the average effect owing to mRNA folding. The effect of charged amino acids is additive, with ribosomal occupancy well-predicted by a linear fit to the density of positively charged residues. We thus expect that a translated poly-A tail, encoding for positively charged lysines regardless of the reading frame, would act as a sandtrap for the ribosome, consistent with experimental data.


Ribosomes do not synthesize protein at a constant rate along transcripts, and changes in translation speed can have knock-on consequences for the expression of that protein, even altering its folding or subcellular localization. It has long been thought that RNA-level features modulate translation rates, whether by delays incurred through the presence of codons that require relatively rare tRNAs, or by regions of mRNA folding that physically impede ribosomal progression. We find on the contrary that it is not RNA-level features but positive charges in the already translated protein that most retard ribosomes, possibly by interacting with the negatively charged ribosomal exit tunnel. We show that positive charge explains the sites where ribosomes stall most commonly within transcripts. We also show why, if protein charge were not considered, one could be misled into suspecting a role for non-optimal codons. Finally, we observe that the poly-A tail provides a massively positively charged terminus no matter in which frame it is translated. A missed stop codon or frameshifting would then lead to a stalled ribosome, which is consistent with experimental data.


While it is known that there is great variation in ribosomal velocity along even a single transcript [1], what determines how fast a transcript (or part thereof) is processed is unresolved. Resolving this issue is important for understanding causes of disease and for the generation of transgenes, as changes in the local translation rate along mRNAs have been implicated in the regulation of protein folding [2], error attenuation processes such as no-go decay in yeast [3], transcription attenuation in bacterial systems [4], and correct protein localization [5],[6].

For some time it has been hypothesized [7]–[10], and commonly assumed (e.g., [11],[12]), that codons matching rare tRNAs slow ribosomes along transcripts due to differential tRNA availability. The supposition is that codons corresponding to less abundant tRNAs are translated at slower rates as the ribosome must pause while the appropriate tRNA becomes available. This, for example, is held up to explain the usage of codons specified by the most abundant tRNAs in the most highly expressed genes [13],[14]. Although the notion that rare codons must stall ribosomes is commonplace, recent work has started to undermine the supposition that differential usage of synonymous codons will significantly alter the rate of ribosomal translocation within a transcript under normal conditions [15]–[17]. Indeed, much of the evidence cited as support for an effect on translational speed is questionable (see Note S1) and many of the patterns attributed to selection for translational speed are better explained in terms of selection on codon usage for translational accuracy [18]–[21].

Codon usage, however, is not the only potential factor affecting elongation speed. Double-stranded mRNA hairpin or pseudoknot structures are thought to impede progress of the ribosome [22],[23]. The generality of this during elongation, however, is unclear, as other studies [24] suggest that the ribosome can more readily melt moderately stable secondary structures once initiation has taken place.

While the above factors consider ribosomal velocity to be modulated by properties of the mRNA, much less attention has been paid to the possibility that the resultant protein might impact translation rates. However, recent experimental work on recombinant peptides has shown that positive charges on the newly synthesized peptide might slow ribosomes [25],[26]. This is conjectured to be owing to an electrostatic interaction between the cation in the emerging polypeptide and the negatively charged exit tunnel of the ribosome [25],[26]. Following on from this, it has been suggested that positive charges, codon usage bias, and transcript folding play a role in ribosomal stalling at 5′ transcript ends [27],[28].

Here we ask not whether certain features can sometimes modulate translation speed along a transcript (e.g., when grossly overrepresented in transgenes; see Note S1), but if they do as evolved in endogenous genes when expressed at “normal” levels, and to what extent. Ribosomally protected mRNA footprints from an experimental Saccharomyces cerevisiae dataset [29] enable us to profile the location of ribosomes across the S. cerevisiae transcriptome. Under the assumption that ribosomal densities inversely reflect ribosomal velocity [30],[31], we independently examine the effects of codon usage, mRNA folding, and positive charge on ribosomal speed throughout endogenous yeast genes. We show that positive charges in the nascent peptide slow the ribosome along transcripts in an additive manner in vivo, and that this slowing effect cannot be accounted for by mRNA structure, and even far surpasses that (if any) induced by codon usage bias. Within transcripts, those regions with the highest ribosomal occupancy are those most likely to be just downstream of positively charged residues. The cation sandtrap effect has potential relevance for the evolution of the poly-A tail, specifying as it does a series of positively charged amino acids if translated.


While some recent work on nucleotide-resolution ribosomal footprint data [29] has claimed that codon usage plays a role in slowing ribosomes [27],[28], another study that examined the same footprint data, filtered for noise, contradicts this claim [16]. Here we reanalyze the same dataset using both stringent mapping to reduce false-positive footprints (see Methods, “Ribosomal Density Data” for further comments on this and previous studies) as well as a novel normalization method to detect any accrual of ribosomal density, on average across transcripts, after putative ribosome-slowing features.

Neither Clusters of Nor Consecutive Rare Codons Tend to Slow Ribosomes

Ribosomal footprint data [29] allow us to examine changes in the rate of translation given the assumption that the slower a ribosome travels along a given portion of a transcript, the more likely it is to be found there at any point in time [30],[31]. In the case of codon usage, we expect to see any possible ribosomal stalling centered over the rare codon(s) while the ribosome awaits a tRNA to enter its A-site. Hence to examine the effect of a sequence feature such as rare codons on the speed of translation, we calculate the relative change in stringently mapped ribosomal densities that occurs within a single transcript as ribosomes begin to translate regions of transcript enriched for rare codons (see Methods and Figure 1). To this end, within each transcript we compared the ribosomal occupancy at codon positions (rpos) in the vicinity of clusters of rare codons (rpos) to the average ribosomal occupancy of the 30 codons immediately preceding the first rare codon in the cluster (rprec30). We then averaged the relative increase or decrease in ribosomal occupancy across transcript sections aligned by rare codon clusters. A mean rpos/rprec30 after the clusters >1 indicates a denser sampling of ribosomal footprints on average and hence slowing at that codon position, while a mean rpos/rprec30<1 denotes sparser ribosomal coverage, consistent with acceleration.


Figure 1. Visual overview of our plotting analyses.

A feature of one codon encoding a positive charge as a potential slower of translation elongation is considered as an example. The feature of interest (here the encoded charge) must be surrounded by no other codons encoding positive charges for 30 codons in both directions so as to not interfere with our measurement of slowing due to the single encoded charge we have identified. (A) We start with footprint data, which we have stringently mapped to the codons surrounding the encoded positive charge of interest on the mRNA in which the encoded charge resides. We first count the ribosomal footprints mapping to each codon position in this area. We take the average of the ribosomal footprint counts among the 30 codons preceding (the start of) the feature. We consider the average footprint counts of these preceding 30 codons (rprec30) to reflect the baseline speed at which ribosomes are translating before they reach the encoded charge. We then divide the ribosomal footprint counts in each of the 61 codon positions in this section of the mRNA by rprec30 to measure whether they are more densely or sparsely covered with ribosomal footprints in a given codon position relative to the density before the feature. Note the ratios prior to x = 0 will tend to center around 1 as they will have been normalized by a value likely close to their own. We calculate these relative ratios separately for every feature cluster in every mRNA we identify as suitable for our analysis. (B) To ask whether there is a trend in slowing upon the translation of the feature of interest (the single positive charge in this example), we align all of the mRNAs with the feature of interest by (the start of) the feature. We determine the average relative change in ribosomal density upon translation of the feature by averaging each of the ratios calculated in (A) for each aligned codon surrounding the feature. It is these mean ratios we consider when we calculate the slowing effect (if any) of a given feature. The degree of slowing due to a feature is a function of both the magnitude of the footprint buildup on any one codon as well as the length along the mRNA that the buildup extends. We hence calculate the slowing due to the feature (here the single positive charge) by summing the area between the line y = 1, which represents the baseline speed (see A) and the mean relative ratios between the start of the feature at x = 0 and the point where the means cross y = 1 again (highlighted purple area). If the line does not intersect with y = 1 again by the end of the window (x = 30), the entire area under the curve from x = 0 to x = 30 was used. We do not consider codons at x>30 as there may be positive charges encoded in this downstream region that we do not wish to interfere with our measurements. In some cases, not slowing but speeding will occur, indicated by ratios that are less than 1 (not shown). In this case, we calculate the degree of speeding similarly, by summing the area between the mean ratios and y = 1.

In our main analysis we make use of the tAI (range 0–1) as a measure of codon optimality as this metric uniquely reflects the tRNA pool. The tAI of a sequence is defined as the geometric mean of the relative adaptiveness of its constituent codons to the tRNA pool available in that organism [32]. A higher tAI indicates the codon has a high abundance of decoding isoacceptor tRNAs and, according to the codon usage hypothesis of translational speed, should be translated faster on account of its ready coupling with an aminoacylated tRNA. A lower tAI conversely indicates a codon that is matched by a low number of tRNAs and is therefore putatively slowly translated and nonoptimal. Here we define “rare” codons to be those in the lowest quartile of tAI values (Methods, “The Average Effect of Codon Usage on Ribosomal Densities”) (see also Figures S1, S2, S3 and Table S1 for analysis of rare codons defined according to genomic frequency).

Our results show inconsistent trends in ribosomal occupancy after rare codon clusters when all clusters of a given size are aligned and the average increase in ribosomal density after the cluster (here uncontrolled for covariates) is plotted (Figure 2A). This inconsistency is still apparent when we consider rare codons to be not those with a low tAI but those that are genomically infrequent (Figure S1). If there is any slowing due to rare codons, we should expect an increase in the amount of slowing along the mRNA as the number of rare codons increases. However, no such trend is evident (Figure 3A). This lack of influence of rare codon usage on ribosomal speed is not owing to a covariance between rare codon clusters and expression levels (Table S2). Shifting the location of the “preceding 30 codons” we use to normalize footprint values slightly upstream, to accommodate the 5′ portion of the ribosome potentially slowed over a rare codon, still detects no slowing due to codon usage (Figure S4).


Figure 2. Clusters of rare codons do not tend to slow ribosomes.

The first of the number of nonoptimal codons indicated always occurs at x = 0, and the rest, if any, may be found at points up to and including the codon indicated by the second arrowhead. The mean rpos/rprec30, or relative change in ribosomal occupancy, at each position across aligned transcripts ± s.e.m. is plotted. The horizontal at y = 1 represents the null expectation that positive charges do not alter ribosomal speed—that is, that ribosomes are, on average, as frequently present before the rare codon cluster as after it. The three-rare codon plot in (B) is plotted with different axes as it is an outlier. Some residual slowing is observed near x = −30 on all plots due to slowing elements (e.g., positive charges) that may be encoded just upstream (x<−30). (A) All genes with rare codon clusters. (B) Genes with rare codon clusters that have 0 or 1 positive charges coded for in the last 30 codon positions plotted. These plots represent the net effect of tAI on ribosomal density, with the bulk of the effect of positive charge removed. (C) Genes with rare codon clusters that have two or more positive charges in the last 30 codon positions plotted.



Figure 3. Positive charges show an additive (linear) trend in slowing ribosomes, but rare codons do not.

The degree of slowing is a function of both the magnitude of ribosomal density and the length of transcript the slowing covers. Therefore, to measure any trend in the ability of either positive charges or codon clusters to slowing, the area between the curves depicting the average relative change in ribosomal density (rpos/rprec30) and the y = 1 null in Figure 2A, Figure S5A, and Figure 5, whether positive or negative, was summed between x = 0 (the beginning of the cluster) and the point where the plotted values intersect with y = 1 again, regardless of where the last charge in the cluster is (see Figure 1 for further explanation of the area under the curve). A positive value for the area under the curve indicates ribosomal slowing after the feature in question, while a negative value reflects faster movement. (A and B) Regression of area under curve∼size of cluster, slope p = 0.45 and 0.33, respectively. (C) Regression of area under curve∼size of cluster gives a slope of 2.81 (p = 0.020), r2 = 0.93. To achieve such a regression slope in the set of genes used is significantly nonrandom (p = 0.011, Note S3).

As it has been postulated that tandem nonoptimal codons may more strongly inhibit progression of the ribosome than scattered rare codons [33],[34], we also investigated whether consecutive rare codons (adjacent codons, each from the lowest quartile of tAI values) may be affecting ribosomal velocity. Examining changes in ribosomal densities after pairs, triplets, and so forth of rare codons, however, also indicates that consecutive rare codons do not systematically slow ribosomes (Figure S5 and Figure 3B). We achieve similar findings when defining rare codons according to their genomic frequency (Figure S2).

If the above results are correct, then we should also find that codon usage cannot explain ribosomal slowing when we compare sites within a given mRNA. Upon locating the highest and lowest ribosomal occupancy portions within a given mRNA, we determined whether the denser region was associated with a putative ribosome-slowing feature: lower tAI, or more rare codon pairs or rare 6-mers (two adjacent in-frame codons that, as a pair, come from the lowest 10% of all 6-mers within the genome) (see Methods, “The Relative Contributions of Charge, Folding, and Codon Usage to Extremes of Slowing Within Transcripts”). Considering all transcripts, the most slowly translated region within an mRNA in fact tends to be comprised of more optimal codons or fewer rare pairs, suggesting low codon optimality does not cause slowing (Table 1A,B and Figure S6). These results are not affected if we consider suboptimal codons to be those that are genomically infrequent (Table S1 and Figure S3). Nor do we find that transcript similarity to the yeast Kozak sequence can explain slowing within these regions (Figure S7 and Table S3). Additionally, as the difference in ribosomal occupancy between the two intra-transcript windows increases (and hence the presumed difference in the inferred ribosomal velocities between the two windows grows all the more), the already low proportion of transcripts for which tAI, genomic infrequency, or presence of rare pairs could possibly explain ribosomal pausing in fact decreases (Table 1A,B and Table S1). In other words, in transcripts that have the greatest differences in ribosomal densities along their length (as inferred from the highest and lowest ribosomal occupancy windows), and hence that contain the greatest degree of internal slowing relative to maximum translation speed, the most ribosomally occluded windows are even more likely to be comprised of more optimal codons. This indicates that not only is low codon optimality incapable of explaining ribosomal slowing in general, it is even less capable of explaining the greatest relative slowing within a transcript.


Table 1. Only positive charge is systematically capable of explaining ribosomal slowing, including the severest slowing.

 We note that the decrease in the ability of codon usage to explain slowing in the upper quantiles (Table 1A) is simply a side effect of differential amino acid usage between the two windows. When we control for differential amino acid content between the two windows, we no longer see the decrease in the ability of codon usage to explain slowing, but codon usage still remains unable to explain the slowing that is observed in any of the quantiles (Table S4). Thus, in addition to the above finding that codon usage becomes less able to explain slowing as the degree of slowing grows (as deduced from observed transcripts), this amino-acid-controlled analysis suggests that even if amino acid sequence had evolved in any other way, codon usage would still not be a factor in the slowing of ribosomes.

It is possible that codon usage could have different effects during different times of cell cycle if tRNA levels fluctuate [35]. We do not, however, detect a systematic influence of codon usage on ribosomal speed even under amino acid starvation conditions (Figures S8, S9, S10 and Table S5) when presumably tRNA charging levels are lower, making codon usage potentially more rate-limiting [36],[37].

RNA Structure on Average Increases Ribosomal Occupancy Marginally

If neither codon usage nor consecutive rare codons can explain variation in ribosomal speed, then what can? As it has been suggested that transcript structure can impede ribosomes along the length of the transcript [28], we next investigated whether RNA structure might be the major contributor to slowing.

We used empirically determined (rather than computationally predicted) RNA structure data (PARS values, see Methods, “The Average Effect of Transcript Structure on Ribosomal Densities”) [38]. S. cerevisiae protein-coding sequences were scanned for stretches whose average PARS value was 0 or negative (and hence tending to be single-stranded), which were immediately followed by a block of codons whose average PARS value was positive (i.e., with propensity for double-strandedness). The general contribution of folding to slowing was examined by calculating the relative change in ribosomal density (rpos/rprec30) at each position of the identified region of a transcript, where rprec30 is the average ribosomal occupancy in the single-stranded block. We then take the average of this ratio across transcripts aligned by identified blocks of structure.

The method is similar to that used above with codons, but with one complication. In the case of codon usage, we have a prior expectation that any ribosomal pausing should occur while the ribosome is positioned over the “slow” codon. It is not immediately clear, however, where along the transcript we should expect any structure-induced pausing to take place. After translating an unstructured span of mRNA, will the ribosomal active site be able to get very close to the first double-stranded ribonucleotide it meets before it is finally slowed, or might pausing take place more 5′ if the ribosome progression is sterically occluded at some distance upstream? We investigated both hypotheses.

We cannot immediately distinguish between the possibilities that mRNA folding has an effect on ribosomal progression either upon or upstream of the folded ribonucleotides in question, as some degree of pausing is observed in both cases (Figure 4). But how strong is this slowing effect? Could mRNA folding account for the bulk of the variance in ribosomal speed observed along transcripts? We find, again comparing the slowest and fastest translated regions within a given mRNA, that not only is secondary structure incapable of systematically explaining the slowest regions of translation, but the presence of secondary structure decreases as the difference between the ribosomal density (i.e., difference in translation speed) of the two intra-transcript windows increases (Table 1C). Hence we conclude something other than mRNA folding must be responsible for the greatest slowing within transcripts.


Figure 4. Ribosomes travelling along single-stranded RNA are not greatly retarded upon traversal into double-stranded structures.

PARS values >0 denote structured mRNA, <0 single-stranded. All averages plotted (± s.e.m.) are calculated across transcripts aligned by blocks of mRNA structure. The slowing of ribosomes (rpos/rprec30>1) relative to the preceding 30 codons starting from both the beginning of double-stranded structure (A) and 10 codons upstream of the same regions of double-stranded structure (B) are shown. In both cases, there is a degree of translational pausing observed upon the transition into folded mRNA, although some of this slowing may be caused by the presence of two or more positive charges encoded in the folded area (0≥x≤30).

Positively Charged Amino Acids Additively Slow Ribosomes on Endogenous Yeast Transcripts

We performed a parallel version of the codon cluster analysis to look for changes in ribosomal density after differently sized clusters of encoded positive charges (see Methods, “The Average Effect of Positive Charge on Ribosomal Densities”), calculating the average relative change in ribosomal density within a transcript (rpos/rprec30) after positively charged residues (lysine, arginine, or histidine) are added to a nascent peptide chain. The effect, note, should be a stalling after the codon specifying the charged amino acid as the stalling process is hypothesized to be an interaction between the charged amino acid and the charged exit tunnel [26],[39].

We find that a single positive charge will slow the ribosome relative to the preceding sequence (Figure 5), regardless of whether the codon encoding the residue is A/G- or C-rich (Figure S11). Our findings show that at maximum (in real transcripts), ribosomes are more than twice as likely to be found at a given region of the transcript as before the addition of the cation to the polypeptide (Figure 5). The higher the density of positive charges in a peptide, the proportionally greater the effect (Figure 3C), in agreement with experimental findings that increasing the number of positive charges locally correspondingly increases ribosomal dwell time [26]. Our estimation of charge-induced pausing is conservative since some ribosomal density after charges is not included in the analysis if the mean ribosomal occupancy of the 30 codons preceding a charged cluster is 0 for a given transcript (our method in this case would require division by 0).


Figure 5. Positive charges slow ribosomes.

The first of the positive charges indicated always occurs at x = 0, and the rest, if any, may be found at points up to and including the codon indicated by the second arrowhead. rpos/rprec30 is the ribosomal occupancy at position x normalized by the average occupancy of the 30 codons preceding the encoded positively charged cluster within the same transcript. The mean rpos/rprec30, or average relative change in ribosomal occupancy, at each position across aligned transcripts ± s.e.m. is plotted. The horizontal at y = 1 represents the null expectation that positive charges do not alter ribosomal speed; in other words, that ribosomes which translate in positive-charge free peptides are, on the average, as frequently present before the charge cluster as after it. 

We can also test whether charge is responsible for slowing by noting that the pKa, and hence overall net charge, of histidine is lower than that of either arginine or lysine at physiological pH. Thus we should expect a weaker slowing effect due to histidine residues being added to the polypeptide. When we re-calculate the slowing effect after a single positive charge (as shown in Figure 5, first panel), but separate the single charges according to whether or not they are histidine, we indeed observe that histidine causes weaker slowing (Figure S12). The slowing effect after a single histidine residue, as calculated using the area under the curve method, is anywhere from 25%–78% (95% CI) of the slowing found after a single lysine or arginine. As histidine is used much less frequently than either of the other positively charged residues, we consider slowing after single positive charges to be the best comparator due to the larger sample sizes available. When we separate larger positive charge clusters according to their histidine content (at least one histidine in the two- or three-charge clusters, and at least two histidines in the four- or five-charge clusters), we note that the slowing due to the histidine-enriched group is always lesser than that after the histidine-free group (Figure S12).

If charge is a major determinant of ribosomal slowing, then it should be capable of explaining the regions of greatest translational pausing within transcripts (see Methods, “The Relative Contributions of Charge, Folding, and Codon Usage to Extremes of Slowing within Transcripts”). We find this is indeed the case. Of all the putative slowing features we consider, only positive charge is more often associated with the higher occupancy window within each transcript (Note S2). Breaking the comparisons into quantiles according to the magnitude of difference in ribosomal occupancy between each pair of windows further reveals that positive charge is the feature most often responsible for not just slowing when comparing between transcripts, but the greatest magnitude of slowing within any given mRNA. As the difference in ribosomal occupancy between the two windows increases, the window with the higher ribosomal occupancy tends increasingly to be the one with more positive charges (Table 1D). In fact the only clearly significantly overused amino acid in the higher occupancy windows is lysine, which is positively charged (Figure S13). This increase in ribosomal occupancy cannot be explained by physiochemical properties of other amino acids, namely hydropathy, negative charge, or polarity (Tables S6, S7, S8, S9). We note that even when both windows in a transcript have the same number of charges each, there is no predominant influence of tAI, rare codon pairs, or RNA structure on ribosomal slowing (Tables S10, S11, S12).

The Effect of Positive Charge Is Not Explained by Covariance with Codon Usage or mRNA Folding

The positive charge effects seen above could potentially be explained as covariate to codon usage bias, were, for example, codons specified by rare tRNAs especially abundant near those specifying positively charged residues. Given the absence of evidence for codon usage bias to affect translation rates, this now seems unlikely. To nonetheless test whether this is the case, we examined patterns of codon usage in the vicinity of positive charges similarly to the manner in which we investigated changes in ribosomal occupancy after positively charged clusters above. Thus if nonoptimal codon usage were causing the slowing patterns after encoded positive charges observed in Figure 1, we should see, on average, a relative decrease in tAI in those sites with elevated ribosomal occupancy. Contrary to this expectation, however, the trend for ribosomal occupancy to increase after positive charges (Figure 5) is independent of patterns of codon usage (Figure S14).

It is also possible the slowing effects observed after positive charge clusters in Figure 5 occur ancillary to mRNA secondary structure, as such structure may have some slowing effect (Figure 4). Again we allow for mRNA folding to impede the flow of ribosomes starting either locally or 10 codons upstream (in the case that local double-strandedness creates a structure within the transcript that sterically occludes ribosomes from progressing further toward codons within the folded structure). We find that patterns of transcript secondary structure near positive charge clusters are unable to explain the pausing after translation of positive charges (Figure S14). Hence we argue that mRNA folding cannot explain the slowing seen in Figure 5, which is perhaps not surprising given its apparently weak effect on the whole (Figure 4).

Covariance with Positive Charge Does Explain Some of the Slowing Observed After RNA-Level Features

Given that positive charge slows ribosomes, we should expect that some of the (relatively weaker and/or inconsistent) ribosomal slowing at rare codon clusters or transcript secondary structure might in fact be due to the presence of uncontrolled-for positive charge. We find this to be the case. When groups of rare codons that are followed by either a lesser or greater number of positive charges are plotted separately, it is clear that rare codon clusters do not in and of themselves slow ribosomes (Figure 2B) but that the apparent (yet unsystematic) slowing in Figure 2A is in fact due to the presence of positive charge after some of the codon clusters (Figure 2C). Similarly, sorting by the number of positive charges present after a cluster reveals that some of the slowing observed at structured regions of transcript is likely due to previously unaccounted-for positive charge (Figure 4).


We find that codon usage and transcript secondary structure do not substantially affect ribosomal velocities systematically across endogenously occurring transcripts. Although it has been suggested that amino acid starvation might increase the ability of codon usage to modulate ribosomal speed [37], we find no such effect upon examination of ribosomal footprints taken from amino-acid-starved yeast (Figures S8, S9, S10 and Table S5). We do not, however, wish to assert that codon usage and RNA structure can never affect translation rates. Certain secondary structure configurations may substantially impact ribosomal flow. As regards codon usage, if we return to the original logic by which codon usage was thought to affect translation rates, we can both see where the prior logic was misleading and in turn can predict when codon usage should slow ribosomes.

The classical logic supposes that because common codons are specified by abundant tRNAs, the waiting time for the ribosome to capture the necessary tRNA must be lower for “optimal” or common codons. The key parameter, however, to determine waiting time is not the absolute tRNA abundance (as often considered) but the tRNA availability. We note, similarly to Qian et al. [16], that if codons are used in proportion to tRNA availability [40], then this could dampen any pausing effect, since rare codons matching rare tRNAs will not be as rate-limiting as if they were used more often. Put differently, if highly abundant transcripts all require the same tRNA, then this acts as a drain on the availability of that tRNA. This can be described in terms of supply and demand economics. In the case of rare codons in lowly expressed transcripts, the supply (the pool of tRNA) is small and the demand (number of codons requiring that tRNA at any given time) low. For a common codon in an abundant transcript, the supply (tRNA pool) is large but the demand is also large.

We can then imagine an equilibrium situation in which the ribosome waiting time is the same for all codons as the demand (absolute codon abundance in transcripts) and supply of tRNAs are balanced. This is consistent with our observation that, under normal growth conditions, codon usage does not predict ribosome occupancy. However, the same model can predict that under abnormal conditions, we might see an effect as the situation has been forced far out of supply–demand equilibrium. Greatly overexpressing a transcript rich in rarely used codons should slow the ribosome as the demand for the rare tRNAs now exceeds supply. Likewise, we expect that gross modification of tRNA pools should have gross effects on translational speed as the system has been shifted away from the demand–supply equilibrium. This distinction between normal (equilibrium) and experimentally forced (nonequilibrium) conditions makes good sense of the prior literature, where reports of an effect of codon usage on translational velocity involved experimentally forced conditions (for review, see Note S1).

Further evidence that the impact of codon/tRNA abundance is buffered comes from the report that some codons whose aminoacyl-tRNAs are selected either intrinsically rapidly or slowly by the ribosome have either low or high tRNA concentrations within the cell, respectively [41], suggesting that intrinsic differences in the translation speeds of certain codons are not accentuated but rather compensated for. The evidence for codon usage/tRNA buffering indirectly suggests either that some property other than speed causes selection on codon usage (e.g., accuracy of translation [18]–[21]) or that selection for speed occurs when the demand–supply balance is perturbed, for example when selection acts on growth rates and favor duplications of tRNAs. That codon usage also has little or no effect on ribosome velocity in mammals [15] as well as yeast is then, in retrospect, perhaps not so unexpected.

Our results are consistent with the interaction of the cations in the protein with the ribosomal exit tunnel [25],[26], a model supported by the stalling being displaced from the location on the mRNA of the codons specifying the positive charge. Our results also indicate that positive charge, more than other chemical or biophysical properties of amino acids (see Tables S6, S7, S8, S9), is key. While some highly conserved amino acid sequences have been shown to interact with the ribosomal tunnel to stall translation in order to regulate the specific gene product they control (see, e.g., [42]–[44]), our results suggest a fundamental feature of proteins that slows ribosomes regardless of sequence context (either the local amino acid sequence or the gene in which they reside) and without the addition of trans acting factors.

A general slowing of translation due to positive charge has ramifications for the evolution of the poly-A tail. If translated, the poly-A tails results in a long run of positively charged lysines. This is expected to stall run-on ribosomes [39]. This stalling may glue the aberrantly translated peptide to the ribosome, preventing potentially toxic products from diffusing into the cell and/or permit tagging of the peptide in the nascent chain–ribosome complex with a signal for degradation, as observed [26],[39].

Our results are consistent with translation of poly-A tails stalling ribosomes. Extrapolating the linear trend for larger clusters of positive charges to additively slow ribosomes (reported in Figure 3C), we note that a poly-A tail of 80 consecutive adenines (∼27 lysines) in yeast [45] should slow translation at least 4-fold more than that observed in clusters of six or more positive charges (Figure 3C), probably halting it. This is in line with experimental work showing that while nonstop mRNAs without poly-A tails are efficiently translated [46], translation of polyadenylated mRNAs lacking stop codons or full 3′UTRs is repressed after initiation [47]. Similarly, inserting a poly-A tract into a coding sequence represses translation post-initiation, but not on account of rapid mRNA decay [39]; a similar finding was reported for 3′ poly-A tails [48]. Recently, it was shown that translation of 12 consecutive basic amino acids inserted into a reporter gene causes not only translation arrest but degradation of the polypeptide [49].

Why is the tail poly-lysine if any positive charge will do? The reason is likely to be found at the DNA sequence level. Of all codons encoding positive charges, only lysine possesses a codon that is a triplet repeat of a single nucleotide (AAA) and therefore may be added simply and sequentially by a single enzyme. Moreover, the triplet repeats form a homogenous run of adenines, meaning that positive charges will still be added to the nascent chain (and hence stall ribosomes) no matter how the stop codon is missed, be it by failure to interpret the stop when in-frame or owing to frame-shifting. This may have less relevance in species with long 3′UTRs, in which an alternative stop may be found with the UTR, but in the ancestor in which the poly-A tail evolved, if 3′UTRs were short, then this sandtrap for ribosomes may have been of considerable benefit.

It is noteworthy that bacteria, which for the most part lack poly-A tails, have an alternative mechanism (tmRNA) to tag and destroy proteins resulting from frameshifting or stop codon readthrough [50]. Stalling initiated by positive charges resulting from translation of poly-A tails in eukaryotes and tmRNA system in prokaryotes may be functionally equivalent modes of error correction [51].


  1. Randall LL, Josefsson LG, Hardy SJ (1980) Novel intermediates in the synthesis of maltose-binding protein in Escherichia coli. Eur J Biochem 107: 375–379.
  2. Siller E, DeZwaan DC, Anderson JF, Freeman BC, Barral JM (2010) Slowing bacterial translation speed enhances eukaryotic protein folding efficiency. J Mol Biol 396: 1310–1318.
  3. Doma MK, Parker R (2006) Endonucleolytic cleavage of eukaryotic mRNAs with stalls in translation elongation. Nature 440: 561–564.
  4. Yanofsky C (1981) Attenuation in the control of expression of bacterial operons. Nature 289: 751–758.
  5. Chartrand P, Meng XH, Huttelmaier S, Donato D, Singer RH (2002) Asymmetric sorting of ash1p in yeast results from inhibition of translation by localization elements in the mRNA. Mol Cell 10: 1319–1330.
  6. Mariappan M, Li X, Stefanovic S, Sharma A, Mateja A, et al. (2010) A ribosome-associating factor chaperones tail-anchored membrane proteins. Nature 466: 1120–1124.
  7. Ikemura T (1981) Correlation between the abundance of Escherichia coli transfer RNAs and the occurrence of the respective codons in its protein genes: a proposal for a synonymous codon choice that is optimal for the E. coli translational system. J Mol Biol 151: 389–409.
  8. Kimchi-Sarfaty C, Oh JM, Kim IW, Sauna ZE, Calcagno AM, et al. (2007) A “silent” polymorphism in the MDR1 gene changes substrate specificity. Science 315: 525–528.
  9. Anderson WF (1969) The effect of tRNA concentration on the rate of protein synthesis. Proc Natl Acad Sci U S A 62: 566–573.
  10. Gouy M, Gautier C (1982) Codon usage in bacteria—correlation with gene expressivity. Nucleic Acids Res 10: 7055–7074.
  11. Thanaraj TA, Argos P (1996) Ribosome-mediated translational pause and protein domain organization. Protein Sci 5: 1594–1612.
  12. Cortazzo P, Cervenansky C, Marin M, Reiss C, Ehrlich R, et al. (2002) Silent mutations affect in vivo protein folding in Escherichia coli. Biochem Biophys Res Commun 293: 537–541.
  13. Grantham R, Gautier C, Gouy M, Jacobzone M, Mercier R (1981) Codon catalog usage is a genome strategy modulated for gene expressivity. Nucleic Acids Res 9: r43–r74.
  14. Bennetzen JL, Hall BD (1982) Codon selection in yeast. J Biol Chem 257: 3026–3031.
  15. Ingolia NT, Lareau LF, Weissman JS (2011) Ribosome profiling of mouse embryonic stem cells reveals the complexity and dynamics of mammalian proteomes. Cell 147: 789–802.
  16. Qian W, Yang J-R, Pearson NM, Maclean C, Zhang J (2012) Balanced codon usage optimizes eukaryotic translational efficiency. PLoS Genet 8: e1002603 .
  17. Li GW, Oh E, Weissman JS (2012) The anti-Shine-Dalgarno sequence drives translational pausing and codon choice in bacteria. Nature 484: 538–541.
  18. Akashi H (1994) Synonymous codon usage in Drosophila melanogaster: natural selection and translational accuracy. Genetics 136: 927–935.
  19. Stoletzki N, Eyre-Walker A (2007) Synonymous codon usage in Escherichia coli: selection for translational accuracy. Mol Biol Evol 24: 374–381.
  20. Precup J, Parker J (1987) Missense misreading of asparagine codons as a function of codon identity and context. J Biol Chem 262: 11351–11355.
  21. Warnecke T, Hurst LD (2010) GroEL dependency affects codon usage-support for a critical role of misfolding in gene evolution. Molecular Systems Biology 6: 340.
  22. Wen JD, Lancaster L, Hodges C, Zeri AC, Yoshimura SH, et al. (2008) Following translation by single ribosomes one codon at a time. Nature 452: 598–603.
  23. Somogyi P, Jenner AJ, Brierley I, Inglis SC (1993) Ribosomal pausing during translation of an RNA pseudoknot. Mol Cell Biol 13: 6931–6940.
  24. Kozak M (1986) Influences of mRNA secondary structure on initiation by eukaryotic ribosomes. Proc Natl Acad Sci U S A 83: 2850–2854.
  25. Lu J, Kobertz WR, Deutsch C (2007) Mapping the electrostatic potential within the ribosomal exit tunnel. J Mol Biol 371: 1378–1391.
  26. Lu J, Deutsch C (2008) Electrostatics in the ribosomal tunnel modulate chain elongation rates. J Mol Biol 384: 73–86.
  27. Tuller T, Carmi A, Vestsigian K, Navon S, Dorfan Y, et al. (2010) An evolutionarily conserved mechanism for controlling the efficiency of protein translation. Cell 141: 344–354.
  28. Tuller T, Veksler-Lublinsky I, Gazit N, Kupiec M, Ruppin E, et al. (2011) Composite effects of gene determinants on the translation speed and density of ribosomes. Genome Biol 12: R110.
  29. Ingolia NT, Ghaemmaghami S, Newman JR, Weissman JS (2009) Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. Science 324: 218–223.
  30. Tuller T, Waldman YY, Kupiec M, Ruppin E (2010) Translation efficiency is determined by both codon bias and folding energy. Proc Natl Acad Sci U S A 107: 3645–3650.
  31. Bulmer M (1991) The selection-mutation-drift theory of synonymous codon usage. Genetics 129: 897–907.
  32. dos Reis M, Savva R, Wernisch L (2004) Solving the riddle of codon usage preferences: a test for translational selection. Nucleic Acids Research 32: 5036–5044.
  33. Kane JF (1995) Effects of rare codon clusters on high-level expression of heterologous proteins in Escherichia coli. Curr Opin Biotechnol 6: 494–500.
  34. Varenne S, Baty D, Verheij H, Shire D, Lazdunski C (1989) The maximum rate of gene expression is dependent on the downstream context of unfavourable codons. Biochimie 71: 1221–1229.
  35. Frenkel-Morgenstern M, Danon T, Christian T, Igarashi T, Cohen L, et al. (2012) Genes adopt non-optimal codon usage to generate cell cycle-dependent oscillations in protein levels. Mol Syst Biol 8: 572.
  36. Brackley CA, Romano MC, Thiel M (2011) The dynamics of supply and demand in mRNA translation. PLoS Comput Biol 7: e1002203 .
  37. Elf J, Nilsson D, Tenson T, Ehrenberg M (2003) Selective charging of tRNA isoacceptors explains patterns of codon usage. Science 300: 1718–1722.
  38. Kertesz M, Wan Y, Mazor E, Rinn JL, Nutter RC, et al. (2010) Genome-wide measurement of RNA secondary structure in yeast. Nature 467: 103–107.
  39. Ito-Harashima S, Kuroha K, Tatematsu T, Inada T (2007) Translation of the poly(A) tail plays crucial roles in nonstop mRNA surveillance via translation repression and protein destabilization by proteasome in yeast. Genes Dev 21: 519–524.
  40. Ikemura T (1982) Correlation between the abundance of yeast transfer RNAs and the occurrence of the respective codons in protein genes. Differences in synonymous codon choice patterns of yeast and Escherichia coli with reference to the abundance of isoaccepting transfer RNAs. J Mol Biol 158: 573–597.
  41. Curran JF, Yarus M (1989) Rates of aminoacyl-tRNA selection at 29 sense codons in vivo. J Mol Biol 209: 65–77.
  42. Nakatogawa H, Ito K (2002) The ribosomal exit tunnel functions as a discriminating gate. Cell 108: 629–636.
  43. Bhushan S, Meyer H, Starosta AL, Becker T, Mielke T, et al. (2010) Structural basis for translational stalling by human cytomegalovirus and fungal arginine attenuator peptide. Mol Cell 40: 138–146.
  44. Fang P, Spevak CC, Wu C, Sachs MS (2004) A nascent polypeptide domain that can regulate translation elongation. Proc Natl Acad Sci U S A 101: 4059–4064.
  45. Brown CE, Sachs AB (1998) Poly(A) tail length control in Saccharomyces cerevisiae occurs by message-specific deadenylation. Mol Cell Biol 18: 6548–6559.
  46. Meaux S, Van Hoof A (2006) Yeast transcripts cleaved by an internal ribozyme provide new insight into the role of the cap and poly(A) tail in translation and mRNA decay. RNA 12: 1323–1337.
  47. Inada T, Aiba H (2005) Translation of aberrant mRNAs lacking a termination codon or with a shortened 3′-UTR is repressed after initiation in yeast. EMBO J 24: 1584–1595.
  48. Akimitsu N, Tanaka J, Pelletier J (2007) Translation of nonSTOP mRNA is repressed post-initiation in mammalian cells. EMBO J 26: 2327–2338.
  49. Dimitrova LN, Kuroha K, Tatematsu T, Inada T (2009) Nascent peptide-dependent translation arrest leads to Not4p-mediated protein degradation by the proteasome. J Biol Chem 284: 10343–10352.
  50. Gillet R, Felden B (2001) Emerging views on tmRNA-mediated protein tagging and ribosome rescue. Mol Microbiol 42: 879–885.
  51. Bengtson MH, Joazeiro CA (2010) Role of a ribosome-associated E3 ubiquitin ligase in protein quality control. Nature 467: 470–473.
  52. Percudani R, Pavesi A, Ottonello S (1997) Transfer RNA gene redundancy and translational selection in Saccharomyces cerevisiae. J Mol Biol 268: 322–330.
  53. Karolchik D, Hinrichs AS, Furey TS, Roskin KM, Sugnet CW, et al. (2004) The UCSC Table Browser data retrieval tool. Nucleic Acids Res 32: D493–D496.
  54. R Development Core Team (2005) R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.
  55. Lu J, Deutsch C (2005) Folding zones inside the ribosomal exit tunnel. Nat Struct Mol Biol 12: 1123–1129.
  56. Yonath A, Leonard KR, Wittmann HG (1987) A tunnel in the large ribosomal subunit revealed by three-dimensional image reconstruction. Science 236: 813–816.
  57. Cavener DR, Ray SC (1991) Eukaryotic start and stop translation sites. Nucleic Acids Res 19: 3185–3192.

Leave a Reply

Positively Charged Residues Are the Major Determinants of Ribosomal Velocity

Both for understanding mechanisms of disease and for the design of transgenes, it is important to understand the determinants of ribosome velocity, as changes in the rate of translation are important for protein folding, error attenuation, and localization. While there is great variation in ribosomal occupancy along even a single transcript, what determines a ribosome’s occupancy is unclear. We examine this issue using data from a ribosomal footprinting assay in yeast.

Scroll to top