Design of a synthetic polycistronic mRNA gene circuit based on bacteriophage MS2The bacteriophage MS2 regulates the synthesis of its RNA dependent RNA polymerase (RdRp) gene via a combination of two mechanisms, translational coupling and translational repression. In the coupling mechanism, translation of the RdRp gene is dependent on translation of the upstream coat protein gene. Work by Van Duin and colleagues demonstrated that disruption of a long distance RNA secondary structure interaction, the Min Jou interaction33, by ribosomal movement over the coat cistron in the viral mRNA results in the opening of the ribosomal binding site (RBS) for the RdRp gene (c.f. Fig. 1a and b). The Min Jou interaction results from the intermolecular pairing of two sequence elements in the MS2 viral genome, the Min Jou sequence (nucleotides 1427-1433) and the anti-Min Jou-sequence (nucleotides 1738-1744). If this interaction is present, RdRp synthesis is shut off as shown in the schematic in Fig. 1a. However, if this interaction is absent or disrupted by ribosome movement over the coat cistron, then RdRp synthesis is turned on as shown in Fig. 1b. In the translational repression mechanism, translation of the RdRp gene is blocked by the binding of a coat protein dimer to the 19 nucleotide translational repressor (TR) stem-loop, which encompasses the start codon for the RdRp gene (Fig. 1b). It is important to note that this repression is independent of the presence or absence of the Min Jou interaction, which allows the virus to permanently block further RdRp synthesis once sufficient coat protein is present in the cell. Thus, the overall mRNA secondary structure of the MS2 coat and RdRp genes, as illustrated in Fig. 1, enables the synthesis of the RdRp protein to be regulated via feedback from the overall coat protein dimer concentration in the cell.Figure 1Translational coupling and repression of the RNA dependent RNA polymerase gene in Bacteriophage MS2. (A) Cartoon diagram of a section of the secondary structure of the MS2 viral mRNA (nucleotides 1132-1813)18,19 during ribosome initiation on the coat start codon. Approximate locations of the start codons for the coat, lysis, and RdRp genes are labelled by green bars, while stop codons for maturation protein (MP) and coat protein are shown with red bars. The nucleotide sequence of the RdRp ribosome binding site (green dashed box) is shown on the right hand side along with the Min Jou interaction. (B) During ribosome synthesis of the coat protein, the Min Jou interaction is disrupted exposing the RdRp ribosome binding site to ribosomes (blue dashed box) with nucleotide sequence shown to the right. When high concentrations of coat protein dimers are present in the cell, they bind to the TR hairpin blocking subsequent ribosome initiations. (C) Gene circuit diagram illustrating how the translational coupling/repression mechanism impact on protein synthesis. When the Min Jou interaction is present (ON), only coat protein can be expressed due to the RBS for RdRp being hidden. When the Min Jou interaction is absent (OFF) both coat and RdRp can be synthesised, with coat protein translational repressing the RBS for the RdRp gene.Figure 2Design of the CTnano-cnt and CTnano mRNAs based on the Bacteriophage MS2 Coat/RdRp gene fragment. The 3′ end of the MS2 viral mRNA (nucleotides 1284-3569) containing the coat, lysis, and RdRp genes is used as a template for construction of the CTnano-cnt and CTnano mRNAs. Using this sequence fragment, the RdRp gene has been replaced by a sequence encoding the NanoLuc luciferase protein, and the lysis start codon has been removed. Locations of start/stop codons are labelled with green/red bars, and the Min Jou long-distance interaction is highlighted in blue.When constructing a polycistronic mRNA where the NLuc protein is translationally repressed/coupled to the MS2 coat protein, one could directly swap the sequence encoding the RdRp gene with the sequence encoding the NLuc gene (as depicted in Fig. 2). However, it is not clear if this will result in an overall change in the mRNA secondary structure such that the NLuc RBS and/or the Min Jou interaction are disrupted. I hypothesise that the overall secondary structure of the NLuc RBS, and its response to co-translational folding, will have substantial effects on the expression of the protein, thus making the choice of codons of the NanoLuc luciferase gene critical to efficient protein expression.To test this hypothesis, I have first constructed the control mRNA (CTnano-cnt) which was created by directly replacing the RdRp gene in bacteriophage MS2 with an mRNA sequence encoding the NanoLuc luciferase protein. To ensure that translation of the codon sequence is optimal, I have used a NanoLuc luciferase protein which has been codon optimised for expression in E. coli (Genbank ID MN834152). The thermodynamic minimum free energy fold of this mRNA when the Min Jou interaction is present (Fig. 3a and Supplementary Fig. 2) suggests that the TR stem-loop is present, and thus should have similar expression of the nano luciferase protein to that of RdRp in phage MS2. However, analysis of the NLuc expression in CTnano-cnt using my ribosome/folding kinetics model17 reveals that the NLuc RBS can re-fold into a more stable hairpin (left hand side of Fig. 3b) either after ribosome disruption of the Min Jou interaction, or during co-translational folding of the NLuc RBS. The presence of this hairpin is predicted to substantially reduce protein expression as indicated by its energetic barrier to melting.In order to prevent formation of this hairpin during co-translational folding, I have synonymously re-coded the NanoLuc luciferase gene in CTnano-cnt mRNA creating the CTnano mRNA (see “Methods” for procedure). The predicted secondary structure of the CTnano mRNA when the Min Jou interaction is present/absent (right hand side of Fig. 3a and b, respectively) shows that the minimum free energy fold now predicts the start codon being sequestered in the TR hairpin, which has a much lower energetic barrier to melting (c.f. \(\Delta G_m\) in Fig. 3b), and thus should result in higher expression levels of the NLuc protein.Figure 3Predicted secondary structures of the NLuc ribosome binding sites in the CTnano-cnt and CTnano mRNAs. (A) Predicted structure of the RBS when the Min Jou long distance interaction is present. Ribosome binding to the NLuc RBS should be blocked in both CTnano-cnt and CTnano due to a lack of a ribosome standby site. (B) Predicted structure of the NLuc RBS when the Min Jou interaction has been disrupted by ribosomal translation of the coat gene. The estimated energy barrier (\(\Delta G_m\) – kcal/mol) and the predicted average time (\(T_m\)) until the binding site for the 30S:PIC (blue line) becomes exposed are shown, along with the Shine-Dalgarno sequence (red line) and the anti-Min Jou-sequence (black line).Theoretical prediction of NLuc protein expressionIn order to theoretically estimate the levels of NLuc that would be produced from the CTnano-cnt and CTnano mRNAs, one must take into account the competition between ribosomes and coat proteins for the TR stem loop and must also calculate the time dependence on the opening of the ribosome stand-by site for the NLuc gene as a result of the translational coupling between the Coat and NLuc genes. This means that the production rate of NLuc will depend on the production rate of coat protein and the speed of the ribosome movement over the coat gene (the coupling effect), as well as the amount of coat proteins present and their ability to compete with free ribosomes for the TR stem-loop (the repression effect). As discussed in “Methods”, my computational ribosome/folding kinetics model17 is able to account for both the translational coupling as well as the competition between ribosomes and coat proteins for the TR stem-loop. This is since it predicts the co-translational folding of the CTnano mRNA and the rate of movement of the ribosome over the coat gene while also considering the translation of all ribosomes that would be present in the cellular environment, allowing for a calculation of the competition between free ribosomes and coat proteins to be estimated.Using the secondary structures predicted for the CTnano-cnt and CTnano mRNAs (Supporting Figs. 2 and 3, respectively), I simulate the protein expression from multiple copies of either the CTnano-cnt or CTnano mRNA in the presence of the \(\approx 1200\) background cellular mRNAs and 15000 active ribosomes that would be present in an exponentially growing E. coli cell. The CTnano-cnt and CTnano mRNAs are modelled as being continuously produced with rate \(\beta \) from a plasmid vector after induction with IPTG which occurs at time point \(t=0\). The production rate \(\beta =0.012\) \(\hbox {s}^{-1}\) gives the best fit to the experimental data (see below) and results in roughly 20 copies of the mRNA being present in the cell, on average, within 30 min of induction. Figure 4a and b show the results of coat protein and NLuc synthesis in both mRNAs over the course of 30 min. While the amount of coat protein produced in both cases are roughly identical as expected, the CTnano mRNA is predicted to produce approximately 4.6 times as much nano luciferase as the CTnano-cnt mRNA prior to repression by the coat protein.In addition to the simulation where CTnano and CTnano-cnt mRNAs code for the wild-type MS2 coat protein, I also perform a protein expression simulation of the mRNAs where the MS2 coat protein has the N55D mutation present. This coat protein mutant has been shown to be incapable of binding to the TR stem-loop34. Thus, these mutant mRNAs should not be able to repress NLuc synthesis and hence represent the maximum protein production rate that is possible. The computer simulations in Fig. 4c and d show that the maximum production rate of NLuc in the CTnano mRNA is roughly 20 times that of the CTnano-cnt mRNA, indicating that the re-coded CTnano mRNA has enhanced NLuc protein expression compared with the CTnano-cnt mRNA.Figure 4Computational prediction of protein expression in the CTnano-cnt and CTnano mRNAs. Protein expression of the MS2 coat and NanoLuc luciferase genes are simulated using the ribosome/RNA folding kinetic model17 assuming multiple copies of the mRNA being produced in the cell at rate \(\beta =0.012\) \(\hbox {s}^{-1}\). Time courses for the amount of MS2 coat protein and NLuc produced for; (A) CTnano-cnt mRNA, (B) CTnano mRNA, (C) CTnano-cnt mRNA with N55D mutation, and (D) CTnano mRNA with N55D mutation.Experimental measurement of NLuc protein expressionIn order to validate the theoretical predictions of NLuc production in the CTnano-cnt and CTnano mRNAs, I have constructed the plasmids pET-CTnano-cnt and pET-CTnano, where expression of the mRNAs are under the control of the T7 promoter (see “Methods”). Small cultures of E. coli cells containing one of the plasmids were grown in LB media and the resulting luminescence from a sample of cell culture was measured with the NanoGlo assay (Premega) as detailed in “Methods”. The raw data was normalised by the final \(A_{600}\) of the culture and the data was fitted to a sigmodial function (see supplementary figs. 5 and 6). The best fit peak luminescence for both the CTnano and CTnano-cnt mRNAs, along with the peak luminescence from mutant pET-CTnano and pET-CTnano-cnt plasmids containing one of the three coat mutants (N55D, T19*, and S37*), are shown in Fig. 5. As can be seen, there is a clear and significant increase in expression of NLuc in the CTnano mRNA when compared with the CTnano-cnt mRNA. Table 1 gives the luminescence / \(A_{600}\) measurements for each of the mRNAs.NLuc is translationally coupled to MS2 coat expressionTo demonstrate the theoretically predicted translational coupling in the CTnano and CTnano-cnt mRNAs, nonsense mutations can be introduced to the coat protein before and after the Min Jou sequence. It has been previously shown that nonsense mutations introduced prior to the Min Jou sequence prevents full expression of the MS2 RdRp protein, while nonsense mutations after increase RdRp expression33. Following these experiments that were done to demonstrate translational coupling in phage MS2, I have constructed the plasmids pET-CTnano-M3, which has the nonsense mutation T19* before Min Jou and pET-CTnano-M7, which has the alternative nonsense mutation S37* after Min Jou. Supporting Fig. 4 gives a cartoon diagram of the positions of these stop codons, relative to the Min Jou long-distance interaction. Table 1, and Fig. 5 show that for both CTnano and CTnano-cnt mRNAs, the T19* mutation results in a similar level of NLuc production compared to wild-type coat protein. Since the 19 amino acid N-terminal fragment of the coat protein that is produced by the T19* mutant is unable to bind to the TR stem-loop, NLuc expression would be expected to be at similar protein expression levels seen in the N55D mutant if the NanoLuc start codon was not coupled to the translation of the coat gene. The translational coupling of NLuc to coat protein is further supported by the S37* mutant, which results in much higher expression of NLuc in both CTnano and CTnano-cnt mRNAs due to disruption of the Min Jou interaction. Since the S37* mutant is also unable to bind the TR stem-loop, and disruption of the Min Jou interaction has occurred, we should expect that this mutant has NLuc expression greater then that of the wild-type. It should be noted that if translational coupling is working perfectly then we should expect that the relative luminescence of the mutants are T19* \(\le \) WT \(\le \) S37* \(\le \) N55D. However, the experiments reported here show that the T19* mutant has a luminescence roughly equal to or just slightly higher then that of the wild-type, slightly contradicting this ordering (cf. Fig. 5). One possible explanation for this observation is that since the Min Jou interaction is only \(\approx -8\) kcal/Mol, thermal fluctuations allow periodic disruption of the interaction allowing ribosomes to occasionally initiate on the NLuc gene. If translational coupling was not present, then we would expect to see experimentally that both the T19* and S37* mutants have similar luminescence to the N55D mutant, which is not the case here. This explanation, taken together with the experimental results, suggests that the NLuc gene is translationally coupled to the upstream MS2 coat protein.NLuc is translationally repressed by MS2 coat proteinFinally, in order to demonstrate experimentally that NLuc protein production is also repressed by MS2 coat protein in the CTnano and CTnano-cnt mRNAs, I have constructed the plasmid pET-CTnano-M2 which contains the N55D mutation to the coat protein. This coat protein mutant was previously shown to have essentially no affinity for the TR hairpin34, and thus will be unable to bind to the TR stem-loop and block initiation of the ribosome on the NLuc start codon. This is clearly demonstrated in the data for both the CTnano-cnt and CTnano mRNAs, which both have a dramatic increase in NLuc production, with a larger increase seen for the CTnano mRNA. Specifically, CTnano-M2 mRNA produces roughly 22 times as much NLuc when compared with CTnano containing the WT coat protein. Likewise, CTnano-cnt-M2 mRNA produces 6.8 times as much NLuc when compared with CTnano-cnt mRNA (cf. Fig. 5 and Table 1). These experimental results are consistent with NLuc production in these mRNAs being suppressed by MS2 coat protein.Figure 5Experimental measurements of peak luminescence produced by the CTnano-cnt and CTnano mRNAs and their mutants. Experimental luminescence values divided by the \(A_{600}\) of the culture are shown for the CTnano-cnt mRNA (left hand side) and CTnano mRNA (right hand side). Note, experimental data have been divided by \(10^4\) for ease of plotting.Comparison of theoretical and experimental measurementsThe experiment measures the intensity of light produced by the NanoLuc enzyme for a given sample of culture while the theoretical calculations predict number of NLuc proteins produced. To compare with experiment, I assume that NanoLuc enzyme follows Michaelis-Menten kinetics and that light production will be directly proportional to the number of enzymes. Thus, experimental measurements can be compared to theoretical ones by examining the ratios. In order to fit the model to the experimental data only two parameters need to be adjusted; the rate of mRNA synthesis from the plasmid (\(\beta \)), and the size of the footprint of the 30S pre-initiation complex (30S:PIC) on the mRNA during initiation. These are the only parameters which require adjusting in the model as the remaining parameters for ribosome kinetics have been fitted from experimental measurements and for RNA folding from Turner energy parameters17,20,32.It should be noted that it is very difficult to theoretically predict the rate of mRNA production from the plasmid as this will depend on the concentration of T7 polymerase in the cell, the number of copies of the plasmid, and mRNA degradation dynamics, amongst other factors. From a simplified perspective, the value of \(\beta \) essentially adjusts the total amount of NLuc produced in a certain time by the N55D mutant, as this mutant will be unable to repress NLuc synthesis. Thus, as \(\beta \) increases, there is a corresponding increase in the ratio of NLuc produced by the N55D coat mutant to that of the WT where the amount of NLuc will be “capped” due to repression by the coat protein. I have adjusted \(\beta \) such that the ratio of NLuc produced by the N55D mutant to that of the WT roughly matches the experimental observations (c.f. ratios in the bottom row of Fig. 6). This results in a value of \(\beta =0.012\) \(\hbox {s}^{-1}\), which corresponds to a theoretical prediction of roughly 20 mRNAs being present in a single cell (on average) 30 min post induction.Table 1 Experimental and theoretical measurements of protein expression in the CTnano-cnt and CTnano mRNAs and their mutants.The size of the 30S:PIC footprint on the mRNA during initiation determines how much of the RNA needs to become single stranded to expose the 30S:PIC binding site. Thus, if the binding site is sequestered in a more stable RNA secondary structure, the ribosome will take longer to initiate translation resulting in reduced expression of the protein. Again from a simple perspective, altering the footprint will essentially change the amount of the alternative hairpin (present in CTnano-cnt) that requires melting, and will thus impact the ratio of NLuc produced in CTnano-cnt mRNA verses the amount from the CTnano mRNA. I have found that the ideal footprint size during 30S:PIC interaction with the RBS corresponds to + 6 (− 14) nucleotides 3’ (5’) from the A nucleotide in the start codon. This footprint is consistent with the larger footprint of the full ribosome which was estimated by an RNAase protection assay on the phage MS2 lysis gene35. This gives the best fit to the experimental NLuc protein ratios in CTnano-cnt verses CTnano mRNAs (c.f. ratios in the right hand column of Fig. 6). Interestingly, this footprint corresponds to the amount of mRNA needed to expose an ideally positioned Shine-Dalgarno sequence through to the A site codon. From a comparison to experiment, it is clear that the theoretical melting time for the CTnano-cnt hairpin needs to be slightly longer, or the TR hairpin slightly faster, in order to better match the experimental results. The source of this error is likely due to two possibilities. One possibility is that there could be slight errors in the Turner nearest neighbour model for computing RNA base-pair stacking energies, which would impact on the predicted \(\Delta G_m\) values and hairpin melting times. Alternatively, there could be inherent errors from using a breadth-first search or greedy algorithm to identify transition paths17,36, which are used in the model to compute the estimated energy barriers (i.e. \(\Delta G_m\)) to RNA melting. However, the ability of the model to get the correct qualitative behaviour of the protein expression demonstrates the importance of considering the effects of ribosome movement and competition with other cellular proteins when translational repression is present.Figure 6Comparison of experimental and theoretical measurements of NLuc production in CTnano-cnt and CTnano mRNAs. (A) Experimental luminescence/\(A_{600}\) measurements and (B) theoretical predicted number of NLuc per cell for the CTnano-cnt and CTnano mRNAs containing either the wild-type (WT) MS2 coat protein or N55D coat protein mutant. Ratios between the WT and N55D measurements are shown in the bottom row while ratios between the CTnano-cnt and CTnano mRNAs are shown in the right column. Experimental measurements have been divided by a factor of \(10^4\).
