Silent substitutions predictably alter translation elongation rates and protein folding efficiencies

Abstract

Genetic code redundancy allows most amino acids to be encoded by multiple codons which are non-randomly distributed along coding sequences. An accepted theory explaining the biological significance of such non-uniform codon selection is that codons are translated at different speeds. Thus, varying codon placement along a message may confer variable rates of polypeptide emergence from the ribosome, which may influence the capacity to fold towards the native state. Previous studies report conflicting results regarding whether certain codons correlate with particular structural or folding properties of the encoded protein. This is partly due to different criteria traditionally utilized for predicting translation speeds of codons, including their usage frequencies and the concentration of tRNA species capable of decoding them, which do not always correlate. Here, we developed a metric to predict organism-specific relative translation rates of codons based on the availability of tRNA decoding mechanisms: Watson-Crick, non-Watson-Crick or both types of interactions. We determine translation rates of messages by pulse-chase analyses in living E. coli cells and show that sequence engineering based on these concepts predictably modulates translation rates in a manner that is superior to codon usage frequency, occur during the elongation phase, and significantly impacts folding of the encoded polypeptide. Finally, we demonstrate that sequence harmonization based on expression host tRNA pools, designed to mimic ribosome movement of the original organism, can significantly increase the folding of the encoded polypeptide. These results illuminate how genetic code degeneracy may function to specify properties beyond amino acid encoding, including folding.

In most organisms, 61 out of the 64 possible codon combinations are used to encode 20 different amino acids and thus, a single amino acid can be encoded by several (up to six) codons. The distribution of such synonymous codons along protein coding sequences is generally not uniform, suggesting that their properties are not entirely equivalent. This biased codon usage has been described, for example, in organisms where certain codons are more common than others within highly expressed genes (referred to as frequent or optimal codons)¹. Multiple theories have arisen to explain the biological significance of this biased codon selection, and most revolve around the notion that certain codons allow faster or more efficient translation while others result in slower rates^{2; 3}. These different rates of polypeptide emergence from the ribosome are hypothesized to influence its folding properties^{2; 3}. However, the factors that determine the rates at which different codons are translated have remained unclear, which has led to disagreements on whether or not changes in elongation rates have any influence on the properties of the encoded polypeptide². tRNA selection has been determined to be rate limiting for translation elongation in various models^{4; 5}, and thus it is likely that tRNA availability plays a critical role in determining translation elongation rates^{3; 6; 7}. Remarkably, in every organism examined to date, there are considerably fewer than 61 different tRNA species (Fig. 1a), as certain tRNAs are capable of decoding more than one synonymous codon⁸. Thus, there are essentially two modes by which a particular tRNA molecule can decode a codon: (1) through strict Watson-Crick (WC) base pairing in all three positions of the codon:anticodon interaction and (2) through non-WC base pairing at the third position of the codon (referred to as a “wobble” interaction)⁹. Previous studies have suggested that the speeds of decoding of these two mechanisms may be different with wobble-based decoding resulting in slower rates^{10; 11; 12}. Although the precise reasons for such rate differences are currently unclear, it is possible that they may reflect differences in dissociation rates between A-site tRNA and the mRNA after codon:anticodon binding, with wobble-type interactions displaying higher dissociation rates³. Importantly, direct determination of translation elongation rates based on these mechanisms for actual full-length polypeptides and their effect on protein folding are lacking.

Figure 1. Incorporation of tRNA gene information and nature of codon:anticodon base pairing allows the prediction of relative translation elongation rates.

(a) Predicted gene content for tRNAs capable of decoding the standard genetic code according to gtrnadb.ucsc.edu⁸ is plotted for each codon in histogram form (as indicated) by each domain of life in different patterns (as indicated). The length of each box represents the extent to which genes for tRNAs capable of decoding the corresponding codon are present in a domain. For Met or Trp, 100% of genera examined in each domain are predicted to contain a single species of tRNA genes to decode these codons (and thus the length of these bars corresponds to “100% exclusivity”). (b) Predicted relative protein synthesis rates (see main text and Supplementary Methods) in E. coli for Luc sequences lacking codons decoded by wobble-based tRNA interactions (Luc_fast, orange), containing the most frequent E. coli codons (Luc_cbf, blue) or the unmodified firefly coding sequence (Luc_WT, gray).

We began by predicting relative codon translation speeds based on E. coli tRNA gene information (gtrnadb.ucsc.edu)⁸ and values derived from previously measured rates of select individual codons in vivo that allow the rate comparison between WC and wobble decoded codons^{10; 12}. We developed a formula that incorporated these parameters (see methods) and utilized it to generate predicted relative translation speed profiles of any mRNA in any organism of known tRNA gene content (Fig. 1b). To determine whether WC-based decoding is indeed faster than that mediated by wobble interactions, we reasoned that complete elimination of wobble decoding along an mRNA molecule would result in a detectable enhancement of the translation rate of the encoded protein (Fig. 1b). Thus, we employed DNA synthesis to engineer a bacterial expression construct for the model protein firefly luciferase (Luc) in which every amino acid is encoded by a synonymous codon read by a WC-pairing tRNA anticodon (Luc_fast) (Supplementary Materials and Methods and Supplementary Fig. 1), directly measured its translation rate by pulse-chase analysis in live E. coli cells¹³, and compared it to that of the wild type sequence (Luc_WT) (Fig. 2a), (their respective mRNAs accumulated to similar levels; Supplementary Fig. 2). Interpretation of our pulse-chase experiments using a method that utilizes theoretical constant elongation rates to calculate protein synthesis rates¹³ reported a speed of 9.8 amino acids per second (aa/s) for Luc_WT and 19.2 aa/s for Luc_fast, as judged by least sum of squares analysis. However, our predictions (Fig. 1b) suggest that elongation rates are not constant along the mRNAs of our Luc constructs. Thus, incorporation of theoretical variable rates in the interpretation of our pulse-chase data (see supplementary materials) would be expected to yield a better fit. Indeed, for Luc_WT, calculations that utilized variable rates led to a least square value that was lower than any value that could be obtained using constant rates, reflecting a considerably better fit (Supplementary Fig. 3). It is likely that the experimental methodologies utilized in this study may not be of sufficiently high resolution to reveal the finer details associated with regional variations in ribosome movement along mRNAs and thus the differences detected between constant and variable theoretical rates may actually be considerably greater than demonstrated here. For Luc_fast, we were unable to improve the fit beyond the best-fit constant rate by using variable elongation rates. This was not unexpected since much of the variability that exists in elongation rate along the Luc_WT sequence was removed in the Luc_fast sequence by replacing slow with fast codons, which yield a more constant, fast speed profile (Fig. 1b). Regardless, the observed ~2-fold average increment in rate is of very similar magnitude to the one predicted by our metric (~1.7-fold; Fig. 1b). Thus, WC-decoding appears to confer faster translation relative to wobble-based interactions. Frequent codons as determined by biased codon usage patterns have traditionally been considered “fast”, while rare ones have been predicted to be “slow”^{1; 2}. However, in every genome examined to date (gtrnadb.ucsc.edu)⁸, several of the most frequently utilized codons have no cognate tRNA genes and must rely on Wobble-based decoding (Fig. 1a and Supplementary Fig. 4). To address this discrepancy, we designed a Luc construct composed exclusively of the most frequently utilized codons in E. coli regardless of the number of tRNA genes capable of decoding those codons (Luc_cbf) and compared its translation rate to that of Luc_fast. Calculations based on constant elongation rates determined that translation of Luc_cbf occurred at 14.3 aa/s (Fig. 2a) and, probably because of reasons similar to Luc_fast (see above), there was no improvement in fit when variable rates were considered (Supplementary Fig. 3). An intermediate rate between that of of Luc_fast and Luc_WT is not unexpected, as a considerable fraction of the most frequent codons in E. coli correspond to codons decoded by WC tRNAs (Supplementary Fig. 4) and thus Luc_cbf is indeed predicted to be translated at rates intermediate between Luc_WT and Luc_fast (Fig. 1b). Attempts to determine the translation rate of a Luc sequence engineered to contain mostly wobble codons and thus be translated more slowly (Luc_slow) (Supplementary Fig. 1) were unsuccessful because protein production was extremely limited, and precluded unambiguous identification of the full length Luc band (Supplementary Fig. 5), probably as a result of marked ribosome sequestration along this recombinant mRNA^{3; 11; 14}. These results show that WC-based codon:anticodon interactions lead to faster ribosome movement along an mRNA molecule in vivo and constitute a more accurate basis for predicting translation elongation rates than codon frequency per se.

Figure 2. Avoidance of wobble-based interactions during mRNA decoding results in acceleration of translation elongation rates in vivo.

(a) Pulse-chase analyses (left panels) in live E. coli cells synthesizing recombinant Luc from the indicated constructs and plots (right panels) depicting the appearance of incorporated [³⁵S]methionine in full length Luc produced from the indicated constructs (colored dots), curves for the theoretical appearance of methionines with four calculated constant translation rates of the indicated constructs (colored lines) and calculated theoretical appearance of methionines according to our predicted variable rates (x symbols), which demonstrate that Luc_fast is translated faster than Luc_WT and Luc_cbf. (b) Pulse-chase analyses (left panels) and plots (right panels) as in a, for the Luc_WT-fast and Luc_WT-cbf constructs, as indicated, demonstrating that the observed effects on rates are not due to changes in translation initiation.

In order to ensure that effects associated with translation initiation were not responsible for our observed effects on translation acceleration^{15; 16}, we engineered Luc_fast and Luc_cbf sequences in which their first 50 nucleotides were identical to Luc_WT, to yield Luc_WT-fast and Luc_WT-cbf (Supplementary Fig. 1 and Fig. 2b) and determined their translation rates to be 17.4 aa/s and 14.5 aa/s, very similar to those of their Luc_fast and Luc_cbf counterparts, respectively. Thus, we believe that the observed acceleration of translation is due to increased polypeptide elongation rates.

It has been previously demonstrated that decreased translation elongation rates enhance the folding efficiency of Luc upon expression in E. coli¹⁷, and therefore we hypothesized that translation acceleration would result in the opposite effect. Thus, we measured enzymatic activity as an indication of acquisition of the native state and determined the fractional accumulation of the soluble (presumably folded) and aggregated (misfolded) species of protein produced from the wild-type and engineered Luc sequences (which all contain identical amino acid sequences) (Supplementary Fig. 1). At similar levels of total recombinant protein accumulation, the activity of the protein from the Luc_fast construct is less than half of that from Luc_WT (Fig. 3) and protein from Luc_cbf displays intermediate levels (Fig. 3). Consistently, a greater amount of protein partitioned into the aggregated fraction when translated from Luc_fast, with Luc_cbf yielding again intermediate levels (Fig. 3). Thus, it appears that, at least for Luc, increments in overall translation elongation rates correlate with decrements in folding efficiency.

Figure 3. Synonymous sequence based acceleration influences the folding of the encoded polypeptide.

Specific activities of protein products identical in primary sequence produced from Luc_WT, Luc_fast and Luc_cbf, as indicated (top panel). The value of the protein from Luc_WT was set to 100. Error bars represent S.E.M. SDS-PAGE of total (T), soluble (S) and insoluble (P) recombinant protein produced in E. coli from the indicated sequence-engineered constructs (bottom panel).

tRNA gene content differs significantly between bacteria and eukaryotes¹⁸ (Fig. 1a and Supplementary Fig. 4). Thus, for a given mRNA sequence, the mode (WC- vs. wobble-based) by which a particular codon is decoded may differ depending on whether the mRNA is being translated in a eukaryotic or a bacterial cytosol. For example, in E. coli (Supplementary Fig. 4), there are no tRNA genes that decode the GAG codon (glutamic acid) by strict WC base-pairing. This codon must rely on wobble-based decoding by tRNAs produced from the four GAA tRNA genes present in that organism. In contrast, D. melanogaster contains six GAA tRNA genes and thus a GAA codon will be decoded by strict WC base-pairing tRNAs in addition to being decoded by wobble-based interactions from tRNAs produced by the 19 GAG tRNA genes. Thus, a GAA codon would be expected to be a “slow codon” in E. coli but a “fast codon” in D. melanogaster. If the relative translation elongation rates are calculated for the same mRNA sequence using our algorithm described above that takes into account these parameters (see methods), one would expect that the profiles would be considerably different depending on whether bacterial vs. eukaryotic tRNA gene contents were utilized. When such profiles are generated for Luc using tRNA gene data from E. coli or D. melanogaster (as the organism closest to the firefly available in the database; gtrnadb.ucsc.edu⁸), we find that this is indeed the case (Fig. 4a). We suggest that these profiles reflect differences in local rates of ribosome elongation along an mRNA in each particular organism, consistent with the finding that ribosome movement along natural mRNAs is likely not uniform^{7; 11}, but rather punctuated by regions of acceleration and deceleration. It has been well established that translation elongation rates of eukaryotic ribosomes are generally slower than those of mesophilic bacteria (3-8 vs. 12-20 aa/s, respectively)^{19; 20; 21}. By using mutant E. coli ribosomes that translate at slower rates (more similar to those of eukaryotes) we previously showed that a general reduction in elongation rates resulted in a reproducible, yet marginal increase in the folding efficiency of Luc¹⁷. Here we show that the general increase in translation rate of Luc_fast results in a converse decrease in its folding efficiency (Fig. 3). However, as mentioned above, we expect that in the insect, the ribosome will not move at a constant speed along the Luc mRNA, but rather increase and decrease its speed as it encounters stretches of fast and slow codons (reflected in Fig. 4b, middle panel, by the profile’s peaks and valleys, respectively). We propose that these variations in speed (a sort of ribosomal “rhythm”) have been optimized throughout evolution to precisely orchestrate the emergence rates of each segment of the nascent polypeptide to fold or interact with molecular chaperones as it exits the ribosome. Thus, we reasoned that if we were capable of recreating these naturally occurring variations during expression of Luc in the heterologous E. coli cytosol, we might be able to mimic the natural rhythm that the ribosome follows in the insect, and thus increase its folding efficiency. Since the tRNA gene content of the firefly is not currently available, we utilized the tRNA gene content of D. melanogaster as the closest insect with a sequenced genome to conduct our Luc engineering. We created a Luc sequence (Luc_re; Supplementary Fig. 1) in which fast codons in D. melanogaster (decoded by WC interactions; Supplementary Fig. 4) were substituted with synonymous fast codons in E. coli (also translated by WC; Supplementary Fig. 4) and similarly for slow (wobble-based) codons in each organism (Supplementary Fig. 4; see methods). We expressed Luc_re in E. coli (which encodes an identical polypeptide to all our other Luc constructs; Supplementary Fig. 1) and analyzed its folding efficiency (Fig. 4). At similar levels of accumulation, the protein produced from Luc_re was more than twice as active and considerably more soluble than that from Luc_WT (Fig. 4b) although the predicted average (global) translation rates were very similar for both sequences (Fig. 4a). These results suggest that segmental variations in elongation rate can considerably influence the folding of the encoded polypeptide, even if these do not significantly alter the overall time that the ribosome spends along the mRNA. We thus propose that sequence engineering directed to mimic the ribosome rhythm of the original host may constitute a valuable strategy for production of recombinant proteins in heterologous systems.

Figure 4. Mimicking eukaryotic tRNA population via synonymous sequence engineering of mRNA enhances folding efficiency of recombinant proteins in bacterial host.

(a) Plots of predicted relative translation elongation rates for Luc_WT when expressed in E. coli (top panel) or D. melanogaster (middle panel) and the harmonized Luc_re sequence when expressed in E. coli (bottom panel). (b) Specific activities (top panel) and solubility analysis (bottom panel) of protein products identical in primary sequence produced from Luc_WT and Luc_re, as in Figure 3.

Our findings suggest that the genetic code has the capacity to regulate the rates of protein synthesis and folding. They support the notion that not all proteins fold via simple two-state mechanisms, but rather follow particular pathways throughout their available conformational space, influenced by the regional rates by which their nascent segments emerge unidirectionally from the ribosome. Although our predictions and experimental findings have captured principal features of the coupling between translation and folding, our model is likely oversimplified. For example, it is well known that post-transcriptional tRNA modifications can substantially influence codon:anticodon interactions²² (particularly in eukaryotes¹⁸) and that codons neighboring the A-site may influence elongation rates²³, offering additional levels of speed modulation. Nevertheless, we believe that our study provides insight into how so-called silent polymorphisms may result in human disease²⁴ and how variations in tRNA concentrations impact cellular proteostasis in a wide variety of developmental²⁵ and disease states²⁶.

Abbreviations

aa

amino acid

aa/s

aa per second

A-site

acceptor tRNA site

Luc

luciferase from the firefly P. pyralis

Luc_cbf

Luc sequence in which each aa is encoded by its most frequent E. coli codon

Luc_fast

Luc sequence in which each aa is encoded by our predicted fastest E. coli codon

Luc_re

Luc sequence in which each aa is encoded by an E. coli codon of matching predicted speed to a D. melanogaster codon

Luc_WT

wild type Luc sequence

Luc_WT-cbf

Luc sequence with the first 50 coding nucleotides identical to Luc_WT and the rest identical to Luc_cbf

Luc_WT-fast

Luc sequence with the first 50 coding nucleotides identical to Luc_WT and the rest identical to Luc_fast

P

pellet or insoluble fraction of a lysate

S

supernatant or soluble fraction of a lysate

T

total lysate

WC

Watson-Crick base pairing