Team:Hamburg/Model
We build a new model for protein synthesis controlled by RNA regulation using a two trigger AND-gate.
We predicted the performance of our part BBa_K2970008 in a genetic circuit setup.
We used insights gained from our model to construct the part (antibiotic resistance + trigger/gate) and used similar constitutive promoters for optimal protein synthesis.
We predicted structure of modified RNA gate and Chloramphenicol acetyltransferase (CAT) to test influence on activity.
Modelling is important for synthetic biology since the functionality and behavior of synthetic circuits are often difficult to characterize by experiments. Models are designed to predict these functionalities and can help with design decision without the necessity of much experimental data.
For our project we focussed on translation regulation by RNA, which is a promising alternative to standard translational control by transcription factors ( See Design ). We based our project on studies by Green et al. [1], who identified/constructed toehold switches (in the following called control gates), which can be controlled by variable input parameters (in the following called triggers). We focussed our project on an AND-Gate with two triggers, which both have to be present for induction of translation. Their experimental data indicated an increase in translation for the induced protein synthesis compared to the basal synthesis by a factor up to 1000.
First, we needed to find out whether the gate acts as expected in ourcase using distinct promoters. Additionally, we introduced minor changes to the gate sequence and the expression gene. Therefore, we needed to verify that the modifications to the sequences of the gate and the chloramphenicol acetyltransferase (CAT) do not cause negative effects on our expression system.
Thus, we simulated protein synthesis in silico, and did different structure predictions. We designed BBa_K297008 as a gate/trigger test part to characterize the system and to determine the strength of activation. BBa_K297008 consists of two inducible promoters (T7, pBAD strong) for trigger expression and one constitutive promoter (BBa_J23100) for the gate, which controls the measurable expression of GFP ( Figure 1). We assumed that the production capacity of BBa_K2970008 is strongly influenced by the ratio of expressed Trigger I/II mRNA and gene of interest (GOI) mRNA. The objective was to optimize the ratio of mRNA expressions and determine the best set of promoters. Our model can predict the expression of the GOI based on constitutive promoter strength and PTG/arabinose induction, respectively. Our aim was to select promoters with best signal/noise ratio to have the best control over our riboswitch.
iBioSim [2] is a java-based desktop application designed to create and analyze genetic circuit models. Cellular agents can be simulated and the kinetics of genetic circuits predicted. In classical genetic circuits, promoters are regulated by proteins. Although iBioSim can be used for transcription and translation of mRNA, regulatory functions of mRNA are neglected and not supported. Since the logic of BBa_K2970008 can be purely explained on RNA level, iBioSim can not be used directly.
For our model we needed to simulate all RNAs as protein-level species behaving like RNA. Therefore, we modified the stoichiometry of production, a value, which normally describes the number of proteins produced from one mRNA molecule before degradation, to 1.
In our genetic circuit, translation of the GOI is induced by the trigger mRNA I/II complex. The promoter of the GOI was set up to be activated by this trigger complex. As long as the trigger complex is not formed, translation is repressed and only basal expression occurs. This can be contributed to transcription termination before transcription of the GOI,due to the inaccessibility of the ribosome binding site, which is shielded by the toehold switch. Translation activation was achieved by binding of the trigger mRNA I/II complex to the promoter of the GOI open the stem loop and making the RBS accessible, subsequently inducing translation of the GOI (Figure 2).
Despite being a work around, the model should perform essentially in the same way as the real construct, if correct parameters are used.
To be accurate and determine suitable parameters, models need numerous experimental data sets. For the constitutive Anderson promoter family, data for the relative strength in comparison to each other is available in the iGEM part registry [3]. Additionally we characterised the expression of BBa_J23100 and BBa_J23102 and measured similar expression rate s(Results/Link).. Sequencing of the promoters showed that three mutations were introduced in the promoter sequence of BBa_J23102. Since we modified BBa_J23102 we used the data for the relative promoter strength directly from the registry. The T7 [4] and pBAD strong [5] promoters are not as well characterized and comparable data to the Anderson promoters are missing. Additionally, they are inducible promoters and strongly dependent on the context, in which they are expressed. Since there are more variables to be considered, like the concentration of the inducers and the induction time needed before protein synthesis starts, the complexity of the model drastically increases.
Furthermore, there is no experimental data for the interaction and formation of the Trigger I/II mRNA complex. Even though many values for parameters used in this model can not be reliably determined and the model will not be completely accurate, it should give us an idea of how our part could behave in a real context. The input values for variables used in this model are displayed in Table 1.
Table 1: Input values for the kinetic model of BBa_K2970008 Text, TExt BBa_K2970008 was modelled using a Runge-Kutta-Fehlberg simulation [10]. The simulation was run for 1500-time units with addition of IPTG and arabinose after 500-time units. IPTG and arabinose were set up to be degraded to mimic consumption by the cell. The amount of both triggers, trigger-mRNA complex and GOI were calculated over time (Figure 3A). For simplification, in a first simulation induction of T7 and pBAD strong promoter were simulated as equally induced so that Trigger I and Trigger II mRNA concentrations were the same. The model predicted just a small basal expression for the triggers and the GOI. Upon IPTG/arabinose addition, concentration of both triggers should subsequently increase and a trigger/mRNA complex should be formed. Also, was shown that the expression of the GOI is induced after trigger/mRNA complex formation. Since inducers are degraded, expression of trigger mRNA declines until the basal expression rate is reached. GOI protein stability was set arbitrarily and the GOI therefore also degrades over time.
We investigated the case with similar promoter strength so far. The advantage of inducible promoters is that expression rates can be accurately adjusted by variation of the inducer concentration. The ratio of expressed Trigger I to Trigger II is likely to be a key factor for expression rates of the GOI. Therefore, we modified the strength of the inducible promoters in a second simulation to determine the best ratio for expression. We analyzed the ratio in a range from 0.1 to 1.8 relative strength of promoter 1 (controlling Trigger I transcription) compared to promoter 2 (controlling Trigger II transcription). To measure GOI translation a signal/noise ratio was determined, based on the ratio of basal expression of GOI before addition of the inducer and the peak GOI level after induction (Figure 3B). The simulation data showed that the expression of GOI is optimal at similar promoter strength.
The simulation was simplified since the inducer concentration was fixed and just the promoter strength adjusted. This assumption does not describe the logic of BBa_K2970008 perfectly since the strength of T7 and pBAD strong promoters are fixed and just the concentration of inducer is variable. It would have been more precise to model different inducer concentrations and fixed promoter strength. However, we were not able to collect experimental data for the comparison of promoter strength dependent on inducer concentration. With our model we analyzed all possible promoter strength ratios, but were not able to determine absolute concentrations.
For the application of our logic in a real-life context we simplified the part and used constitutive promoters with the same expression rate for the triggers, since our model suggested the best control of GOI expression using promoters with similar strength.
We wanted to test BBa_K2970008 in another context and chose an antibiotic resistance as target. For more information see Design page (Link). Therefore, we needed to modify the part to some degree. First we divided the three important features of BBa_K2970008 (Trigger I, Trigger II and gate + GOI) to different plasmids, since we wanted to establish a triple transformation method. The parts themselves should behave in the same manner, regardless on how many plasmids they are distributed. However, it has to be accounted for different plasmid numbers in the cell, because the prior statement is only true as long as the plasmid numbers are similar. The main influence for plasmid numbers is the size and the kind of the plasmid. Since all three plasmids are derived from a pSB1A3 vector and have roughly the same size, major differences were not expected and this variable was excluded from the model.
As our prior modelling indicated, we could use constitutive promoters for trigger expression and chose BBa_J23102 from the Anderson collection. Chloramphenicol acetyltransferase (CAT) was selected for the antibiotic resistance. All other model interaction remained unchanged (Figure 4).
The modified BBa_K2970008 was modelled using a Runge-Kutta-Fehlberg simulation [10]. Simulation was run for 1500-time units. All promoters are constitutive and expressed from the beginning. The amount of both triggers and GOI over time were calculated. Since coding sequences for the triggers and gate/CAT were situated on different vectors, three different cases were simulated to mimic transformation. First, we simulated a transformation with only two out of three plasmids, consequently removing one of the triggers from the simulation. As a result, in either cases just one of the triggers is expressed and there are just basal CAT expression levels to be expected (Figure 5A+ B). Consequently, bacteria cultivated in media containing chloramphenicol will not survive. On the contrary, simulations predicted a high expression rate of CAT in the case that all three plasmids were transformed, so that bacteria will survive despite the chloramphenicol treatment (Figure 5C). The simulation showed that CAT levels approach an equilibrium asymptotically, in which the same amount of CAT is translated as degraded.
Furthermore, we wanted to know whether the strength of the promoter upstream of the gate in relation to the promoters translating the triggers is also an important factor. Our first assumption was that a stronger translation of the gate compared to the trigger was preferable, since in this case enough gate mRNA would always be present. To test this hypothesis we modified the promoter strength of the gate/CAT from 0 – 100 % of the strength of the trigger promoters (Figure 5D) and found no significant differences in CAT translation unless transcription levels of the gate are drastically decreased to less than 20 % strength compared to the trigger expression. The resulting curves for CAT expression look similar to the previously calculated results, but the maximum level of protein synthesis is reduced. This shows that the logic is stable to changes, which can be applied if necessary.
We then tested the system in E. coli DH5𝛼 cells and showed that a CAT synthesis occurs, when all three plasmids are present. Unfortunately, the basal expression rate of the gate was too high, so that antibiotic resistance was noticeable even for transformation with only the gate plasmid. Our experiments indicated that the resistance was reduced compared to the triple transformation. Nevertheless very high antibiotic concentrations would be necessary to distinguish between triple transformation and single transformation with gate plasmid (Results/Link).
As our model indicated, reduction of promoter strength of the gate leads to reduced CAT expression. In order to make our system work in a better antibiotic range, a reduction of promoter strength in the gate expression would be necessary. We used the strongest members of the Anderson promoter family so far, thus there is much scope for change. By reducing the promoter’s capacity, we should be able to make our system work with more standard antibiotic concentrations. The promoter J23105 (five times lower expression [9]), J23114 (10 times less compared to BBa_J23100 [9]) were suitable candidates to decrease CAT production and improve the behaviour of our part. Until now, we were not able to conduct further experiments to prove this hypothesis.
We changed the gate by introducing additional structure elements, which could influence expression. Our experiments have indicated that the gate is not as tightly regulated as hoped and we observed leaky basal expression of the GOI. This strong leakage was not yet reported by Green et al. and we were worried that our modification might jeopardize the usability of the gate. So we conferred with Dr. Alexander Green from the Arizona State University. He supported our idea that the added secondary structure would probably decrease translation. Furthermore, he pointed out the possibility that duplicated sequences could generate an RBS/start codon pair that would cause leakage. With the particular sequence we used this possibility nevertheless seems unlikely. Additionally we tested only one possible modification to optimise the gate since all our experiments were conducted with the same sequences of triggers and gate RNA. It will be beneficial for future iGEM projects to optimise the gate sequence for better control of the translation system, since we encountered obstacles we have not yet solved. Furthermore, it would be interesting to determine the resistance of the gate sequence towards mutations. This will become important, if our expression system is expressed for a longer time period. A computational approach for model-based decisions for modification of the gate sequence could be promising. As we showed, it is easily possible to model the structure of modified gate sequences and calculate the stability and interaction with the triggers. With Monte-Carlo simulation it is possible to find a potentially optimized gate. We did not simulate an optimized RNA gate, because we did not have the possibilities to implement our findings in the lab. All methods to establish modified gate sequences, like a random mutagenesis approach, would have been too time-consuming or expensive for our team to realize. Moreover, it is unlikely that all calculated RNA-gates work better. We kept on the safe side by using just one modification of a known RNA-gate. But we strongly recommend future teams, when trying to optimise our project, to calculate RNA gates, if their resources will allow it. Compared to the known functional gates found by Green et al. [1] (Figure 6A) we repeated the linker sequence of the RNA gate . Repetitive sequences are known to form secondary structures. Our goal was to decrease the basal expression rates by including additional secondary structures, thus optimizing our gate expression. To confirm that the secondary structure was formed we predicted the secondary structure using the RNAfold Web Server [11]. We compared the secondary structure of the original gate sequence with our gate and were particularly interested in the energy prediction of both structures. The shapes of both molecules are similar. Both structures include a stem loop structure hiding the RBS and the AUG start codon (Figure 6 B+C). In our structure we predicted an additional, longer loop structure at the 3’-end of the RNA. Consequently the minimum free energy prediction calculated an energy of -26.00 kcal/mol for our model compared to -19.60 kcal/mol for the original gate. As a result, our structure is more stable and the inhibitory effect of our closed gate might be increased. We changed the gate by introducing additional structure elements, which could influence expression. Our experiments have indicated that the gate is not as tightly regulated as hoped and we observed leaky basal expression of the GOI. This strong leakage was not yet reported by Green et al. and we were worried that our modification might jeopardize the usability of the gate. So we conferred with Dr. Alexander Green from the Arizona State University. He supported our idea that the added secondary structure would probably decrease translation. Furthermore, he pointed out the possibility that duplicated sequences could generate an RBS/start codon pair that would cause leakage. With the particular sequence we used this possibility nevertheless seems unlikely. Additionally we tested only one possible modification to optimise the gate since all our experiments were conducted with the same sequences of triggers and gate RNA.
It will be beneficial for future iGEM projects to optimise the gate sequence for better control of the translation system, since we encountered obstacles we have not yet solved. Furthermore, it would be interesting to determine the resistance of the gate sequence towards mutations. This will become important, if our expression system is expressed for a longer time period. A computational approach for model-based decisions for modification of the gate sequence could be promising. As we showed, it is easily possible to model the structure of modified gate sequences and calculate the stability and interaction with the triggers. With Monte-Carlo simulation it is possible to find a potentially optimized gate. We did not simulate an optimized RNA gate, because we did not have the possibilities to implement our findings in the lab. All methods to establish modified gate sequences, like a random mutagenesis approach, would have been too time-consuming or expensive for our team to realize. Moreover, it is unlikely that all calculated RNA-gates work better. We kept on the safe side by using just one modification of a known RNA-gate. But we strongly recommend future teams, when trying to optimise our project, to calculate RNA gates, if their resources will allow it.
Due to the design of our gate, where the target gene sequence could not start directly behind the start codon, we expressed a modified chloramphenicol acetyltransferase (CAT) with additional 19 residues at the N-terminus of the protein (CAT+19). To determine, whether the CAT+19 still can degrade chloramphenicol we did some structure prediction.
Since we looked at modifications of a known protein sequence and not a random amino acid sequence, we performed homology structure modelling with SWISS-MODEL online tool [12] rather than de novo structure modelling. The first step in modeling is a sequence alignment to find a suitable template for the model (Figure 7). Fortunately, we found with chloramphenicol acetyltransferase I from E. coli (3u9f) in complex with chloramphenicol a protein structure in the database which could be used as a template.
By predicting the structure of CAT+19 we discovered that the N-terminus is located at the verge of the protein (Figure 8A), spatially divided from the active site were chloramphenicol binds the protein. CAT catalyzes the transfer of an acetyl group from acetyl-CoA to chloramphenicol forming chloramphenicol 3-acetate and CoA [13]. The active form is a homotrimer of three CAT chains. Our structure predicts that CAT+19 still forms a trimer. We therefore conducted a structure alignment of CAT+19 trimer with CAT (I) trimer, using the MatchMaker tool of UCSF Chimera [14] (Figure 8B). At the first glance there were not many relevant differences visible. So we took a closer look at one of the chloramphenicol binding sites. Four residues F44, S165, L177, V189 (for CAT+19) were important to coordinate the chloramphenicol via hydrophobic interactions and hydrogen bonds. Additionally H212 works as proton acceptor during the transfer reaction. (Figure 9C). The difference between the two models were minimal and an effect on the acetyltransferase function is unlikely because important residues have similar positions.
The position of the additional amino acids at the N-terminus could not be well determined by homology modelling because they only have a minor influence on the overall protein structure and therefore are missing in our model.This should not affect the model negatively since the overall influence is minimal. We were nevertheless interested to see whether the additional amino acids would form structures. Thus, we first predicted the secondary structure with PSIPRED 4.0 (Predict Secondary Structure) from the PSIPRED workbench online tool provided by the University College London (UCL) [15] with sequence data as input variables.
So far, we could show that the structural influence of the additional residue can be neglected. Yet, another matter still needed to be considered, namely that the additional sequence mainly consisted of hydrophobic residues. It therefore cannot be ruled out that these residues will make the protein less soluble and hereby influence the protein function. Since we were unsure how to estimate the importance of this effect we turned to. Prof. Dr. Andrew Torda for advise. Asgroup leader for biomolecular modelling at the University Hamburg, he supported us throughout the whole modelling process and helped us interpret our data.. Together, we came to the conclusion that it was very unlikely that the minor modification would influence the protein function. Nevertheless we needed to confirm the functionality of the protein in vivo, so we assembled a plasmid with our modified CAT+19 and transformed DH5𝛼 cells. After transformation we could show that these cells still survived incubation in LB-medium with chloramphenicol (34 µg/mL). Therefore, we are confident that CAT+19 is still fully functional and can be used in the experiments for our gate expression. Using structure prediction we demonstrated that the applied sequence modifications of the RNA gate and the chloramphenicol acetyltransferase do not influence the functionality of neither gate nor CAT, and confirmed the antibiotic resistance for CAT+19 with additional in vivo experiments. RNA gate and CAT+19 can therefore be used in our project. By simulating BBa_K2970008 we found out, how to use the Ribot Collection (BBa_K2970003, BBa_K297004, BBa_K2970006) to express CAT. Constitutive promoters for the expression of the triggers have been predicted to be most effective for this application, we therefore choose BBa_J23100 as a constitutive promoter for the gate expression. During testing in the lab we observed unexpected CAT synthesis due to the leakage of the gate. After looking at our simulation data, we would recommend using promoters for the gate expression which are less strong to reduce the basal expression rate. We have not yet been able to test this hypothesis. RNA regulation is a promising tool for synthetic biology. During our project we realised that it can be complicated to apply this method to translation regulation. Modelling, in many ways, helped us in finding the right strategies to overcome problems and making our project a success.
[1] Green, A. A., Silver, P. A., Collins, J. J., & Yin, P. (2014). Toehold switches: De-novo-designed regulators of gene expression. Cell, 159(4), 925–939. https://doi.org/10.1016/j.cell.2014.10.002 [2] Watanabe, L. et al. i B io S im 3: A Tool for Model-Based Genetic Circuit Design. ACS Synth. Biol. 8, 1560–1563 (2019). [3] Part- BBa_J23100 Available at: http://parts.igem.org/Part:BBa_J23100 (Accessed at 15.10.2019) [4] iGEm Part-BBa_J64997 (T7 promoter); Available at: http://parts.igem.org/Part:BBa_J64997 (Accessed at 15.10.2019) [5] Part:BBa_K206000 : Available at: http://parts.igem.org/Part:BBa_K206000 (Accessed at 15.10.2019) [6] Politi, N. et al. Half-life measurements of chemical inducers for recombinant gene expression. J. Biol. Eng. 8, 5 (2014). [7] Selinger, D. W., Saxena, R. M., Cheung, K. J., Church, G. M. & Rosenow, C. Global RNA half-life analysis in Escherichia coli reveals positional patterns of transcript degradation. Genome Res. 13, 216–23 (2003). [8] set arbitrarily [9] relative data of protein expression; Available at http://parts.igem.org/Part:BBa_J23100 (Accessed 17.10.2019) [10] Fehlberg, E. Klassische Runge-Kutta-Formeln vierter und niedrigerer Ordnung mit Schrittweiten-Kontrolle und ihre Anwendung auf Wärmeleitungsprobleme. Computing 6, 61–71 (1970). [11] Gruber, A. R., Lorenz, R., Bernhart, S. H., Neuböck, R. & Hofacker, I. L. The Vienna RNA websuite. Nucleic Acids Res. 36, W70-4 (2008). Available at: http://rna.tbi.univie.ac.at/cgi-bin/RNAWebSuite/RNAfold.cgi (Accessed: 20.10.2019) [12] Waterhouse, A., Bertoni, M., Bienert, S., Studer, G., Tauriello, G., Gumienny, R., Heer, F.T., de Beer, T.A.P., Rempfer, C., Bordoli, L., Lepore, R., Schwede, T. SWISS-MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res. 46, W296-W303 (2018).; Available at https://swissmodel.expasy.org/ (Accessed 20.10.2019) [13] Biswas, T., Houghton, J. L., Garneau-Tsodikova, S. & Tsodikov, O. V. The structural basis for substrate versatility of chloramphenicol acetyltransferase CATI. Protein Sci. 21, 520–530 (2012) [14] Pettersen, E.F., Goddard, T.D., Huang, C.C., Couch, G.S., Greenblatt, D.M., Meng, E.C., and Ferrin, T.E. "UCSF Chimera - A Visualization System for Exploratory Research and Analysis." J. Comput. Chem. 25(13):1605-1612 (2004). [15] Available at PSIPRED Workbench provided by University College London (UCL) http://bioinf.cs.ucl.ac.uk/psipred/ (Accessed: 18.10.2019) Model
Achievements
Introduction
In silico simulation of gene expression
BBa_K2970008 – an RNA-based AND gate
Idea
Model of RNA/RNA interaction using iBioSim
Results: Gate test expression GFP (BBa_K2970008)
Model for CAT expression of Ribot Collection (BBa_K2970003, BBa_K297004, BBa_K2970006)
Design
Results
Input from experiment
Structure prediction
Structure prediction of gate RNA sequence
Structure prediction of modified chloramphenicol acetyltransferase (CAT)
Conclusions
References: