Translation rate is controlled by coupled trade-offs between site accessibility, selective RNA unfolding and sliding at upstream standby sites

The ribosome’s interactions with mRNA govern its translation rate and the effects of post-transcriptional regulation. Long, structured 5′ untranslated regions (5′ UTRs) are commonly found in bacterial mRNAs, though the physical mechanisms that determine how the ribosome binds these upstream regions remain poorly defined. Here, we systematically investigate the ribosome’s interactions with structured standby sites, upstream of Shine–Dalgarno sequences, and show that these interactions can modulate translation initiation rates by over 100-fold. We find that an mRNA’s translation initiation rate is controlled by the amount of single-stranded surface area, the partial unfolding of RNA structures to minimize the ribosome’s binding free energy penalty, the absence of cooperative binding and the potential for ribosomal sliding. We develop a biophysical model employing thermodynamic first principles and a four-parameter free energy model to accurately predict the ribosome’s translation initiation rates for 136 synthetic 5′ UTRs with large structures, diverse shapes and multiple standby site modules. The model predicts and experiments confirm that the ribosome can readily bind distant standby site modules that support high translation rates, providing a physical mechanism for observed context effects and long-range post-transcriptional regulation.


Ribosomal Platform Interactions with Structured Standby Sites Control Translation Initiation Rate
The ribosomal platform, which makes contact with structured mRNAs at its surface, is one of the three compartments of 16S ribosomal RNA, a scaffold for 30S ribosomal subunit. Six small ribosomal proteins at the platform surface create positively charged surface and bind non-specifically to the negatively charged phosphate backbone of mRNA via electrostatic interactions. In addition, the anti-SD of 16S ribosomal RNA, which is located on the platform surface, stabilizes the mRNA through hydrogen binding ( Figure S1A). Although about 30 nt of mRNA around the start codon has to be unfolded and fed into the mRNA channel to position the start codon at the P-site (2), the upstream 5' UTR can remain folded and in contact with the ribosomal platform surface. The efficient binding of structured mRNAs to the ribosomal platform is a prerequisite for their translation. Therefore, understanding the interactions between ribosomal platform and structured mRNAs is necessary to accurately control their translation rate.
The previously developed biophysical model of translation initiation, RBS Calculator v1.0, (14) accounts for five molecular interactions between mRNA and 30S ribosomal subunit. The total Gibbs free energy change during translation initiation, total ΔG , is the quantitative measure of its rate ( Figure S1B).
The model accurately predicted the translation initiation rates of 132 mRNAs with different Shine-Dalgarno sequences, inhibitory mRNA structures, and 5' coding sequences. However, v1.0 of the biophysical model was not able to accurately calculate the ribosome's binding free energy total ΔG for mRNAs that contain structures upstream of the Shine-Dalgarno sequence in the After-bound state. The predicted translation initiation rate of an example, RBS-91, was 26.25-fold greater than its measured fluorescence level (3387.86 au vs. 129.08±39.93 au on the same proportional scale), equivalent to neglecting a 7.26 kcal/mol binding free energy penalty ( Figure S1C). In contrast, the previous model can accurately predict the translation initiation rates of mRNAs where mRNA structures overlap with the Shine-Dalgarno sequence, but they are primarily unstructured in the After-bound state (RBS-44 example shown in Figure S1C).
The v1.0 of the biophysical model assumed that a small portion of standby site, which was defined as a 4 nt region upstream of the 16S rRNA binding, must be unstructured and that the energy to 3 unfold this region would lower the translation initiation rate. However, based on this and other data, it was clear that this assumption was false, and that the interactions controlling the ribosomal platform's ability to bind standby sites were not correctly modeled. Therefore, it was important to develop a new biophysical model to accurately calculate the ribosomal platform's interactions with standby sites, denoted by standby ΔG . Figure S1. Translation Initiation Begins with Ribosomal Platform's Binding to Structured 5' UTRs 4 (A) The 16S ribosomal RNA of 30S subunit from wild type E. coli is shown with its three compartments. Six small ribosomal proteins create the ribosomal platform surface to bind to mRNAs. mRNA (black color) passes across this surface and enters the mRNA channel of 16S rRNA (PDB entry 3J0U). (B) Total Gibbs free energy change during translation initiation, total ΔG , controls the translation initiation rate. (C) Previous model (14) inaccurately calculated total ΔG of 5' UTRs with structured standby sites. Measured fluorescence (light yellow bar) is compared to predicted translation initiation rate (blue bar). RBS-91 and RBS-44 are from Ref. 14.

The Relationship Between Ribosomal Platform Binding and mRNA Stability
We measured the fluorescent protein expression levels and mRNA levels from structured mRNAs with standby site modules containing either short, medium, and long distal or proximal binding site lengths, and compared these measurements to those from an unstructured mRNA (see Methods in main text). In all cases, the structured mRNAs contain a hairpin with a 15 nt height. We observed a 9.6-fold reduction in fluorescence for a short proximal binding site (P = 4 nt), compared to 5.4-fold reduction in mRNA level (Figure S2.B). Similarly, we observed a 10.1-fold reduction in fluorescence for a short distal binding site (D = 5 nt), compared to a 3.3-fold reduction in mRNA level (Figure S2.D). In contrast, fluorescence levels of a medium length proximal binding site (P = 12 nt) or a medium length distal binding site (D = 12 nt) were 1.3-fold and 2.5-fold lower than an unstructured mRNA, compared to a 2.4-fold and 1.3-fold reduction in mRNA level, respectively. From this data, it appears that the hairpin itself does not significantly decrease the mRNA levels by destabilization. When shortening the proximal or distal binding site lengths, the fluorescent protein expression levels decrease more than the mRNA levels, showing that an inaccessible standby site module is primarily affecting translation rate. Though, there is likely a small amount of coupling between decreases in translation rate and decreases in mRNA stability, due to the protective effects of having ribosomes actively elongate a mRNA to prevent the binding of RNAses.
Long proximal (P = 24 nt) and distal (D = 20 nt) binding sites caused further reduction in fluorescence level (3.3-and 4.2-folds), comparable to the reduction in their mRNA levels (5.8-and 2.7folds, respectively) ( Figure S2B and Figure S2D). These reductions in fluorescence level and mRNA levels cannot be explained by the proposed biophysical model of the ribosomal platform's interactions with standby sites. Instead, these changes are likely due to increased RNAse binding to the long, unstructured proximal or distal binding sites, causing increased mRNA degradation and lower mRNA levels. The error in the biophysical model predictions is larger when long proximal or distal binding sites (greater than 15 nt) are utilized in mRNAs, due to the confounding effects of increased mRNA degradation (see Figure S6).

Branched Structured 5' UTRs Control the Ribosomal Platform's Binding According to their Effective Hairpin Height
Branched mRNA structures can control the ribosomal platform's distortion energy penalty differently than the normal mRNA structures. Because of the three dimensional orientations of the hairpin stems, we propose that a two-branched structure ( Figure S3A) can bend over the neighboring proximal or distal single stranded RNA regions in three different ways ( Figure S3B). It can bend over the single stranded RNA using its base hairpin stem together with one of its branched hairpin stems (stem I or II) (top and bottom configurations in Figure S3B) or only by using the base hairpin stem (middle configuration in Figure S3B). At each configuration, a different length of single stranded RNA is sequestered, resulting in different available RNA surface areas (A S ). We assume that the probability of each configuration is equal. Therefore, the effective A S for a two-branched structure is the average of three available RNA surface areas, according to formula: , where the average branched hairpin height, , is calculated as .
In order to calculate H 1 , H 2 , H 3 , the branched hairpin stems (stems I or II) are counted as an extra nucleotide on the multi-branched loop ( Figure S3CD). In general, because H 2 is less than H 1 and H 3 , the average branched hairpin height is shorter than the maximum hairpin height. This leads to a larger available RNA surface area and therefore lower ribosomal platform's distortion energy penalty.
Importantly, this proposed formulation of average branched hairpin height can be expanded to higher number of branched stems. For example, the average branched hairpin height of a three-branched structure is calculated as , where calculation of individual hairpin heights follows the same procedure shown in Figure S3CD. We critically tested this proposed formulation, by predicting