Team:ULaval/Results
Results
Dry lab
The design of toehold riboswitches is critical for their success as biosensors. Our project includes comprehensive modeling approaches that allowed us to gain insight into the properties of toehold riboswitches (see Model)and the development of a tool that simplifies their production (see Software).In doing so, we performed several in silico tests that allowed us to evaluate the efficiency of our tools and find areas of design that could be further optimized.
Toehold design tool
Our design tool reliably produces toeholds with an appropriate secondary structure (see Software).However, we performed several rigorous tests that would allow us to evaluate the properties of riboswitch - trigger pairs and would be the best options to detect a particular gene of interest. We used the NUPACK suite (Zadeh et al., 2011) to model how our designed toeholds would bind to their intended targets. This allowed us to answer the following questions:
- How does the accessibility of the trigger sequence within the target RNA molecule affect the capacity of the toehold to bind to it?
- Is it actually more favorable for the toehold to bind to the trigger sequence than to remain unbound?
- Are our toeholds predicted to bind specifically to the trigger RNA? Could they bind elsewhere within the target RNA molecule?
To address the first question, we systematically design toehold riboswitches for trigger sequences with different accessibilities. We defined accessibility as the number of unpaired bases in the 30 bp trigger sequence in the predicted secondary structure for the target mRNA molecule and selected only those that had 10 or more unpaired bases. We then used NUPACK to model the binding between each of the designed toeholds and the target mRNA molecule. We show that the accessibility of the trigger within the target mRNA does not have an impact on the free energy of binding (Figure 1). In fact, we see a small trend towards greater accessibility resulting in a less favorable binding, but the correlation between the variables is very low, as implied by the low R2 value. This observation is contrary to our expectations, but factors involved might include the small sample size for these regions with high accessibility or sequence properties like low GC content that cause binding to them to be weaker.
The results of simulating how the toeholds bound to their corresponding triggers allowed us to address our other two questions. By comparing the free energies of the bound states to those of the unbound states, we show that the bound state is more favorable since for all our toeholds it results in a more negative free energy (Figure 2A). Next, we compared the position where the toehold binds to the target molecule to the actual position of the trigger to evaluate. We observed that a very high percentage (78%) of our riboswitches bind perfectly to the target, whereas a small percentage of them bind imperfectly to it (Figure 2B). These results further suggest that our tool can produce toehold riboswitches that bind to their targets. Since the percentage of toeholds that are predicted to bind perfectly is so high, we would suggest discarding all the others to ensure the best results.
3D models
Toehold riboswitches had never been modeled in 3D. We set out to do so because it would allow us to study their structural dynamics and use this new knowledge to improve their design. We started by generating 3D models from the secondary structure of the toehold riboswitch produced by Green et al. (2014) that had the highest ON/OFF ratio. This led to two alternative models, one obtained with the RNAComposer tool (Purzycka et al., 2015; Popenda et al., 2012) and the other with Rosetta (Das et al., 2010; Cheng et al., 2015) through a collaboration with Team iGEM Concordia. We validated these models using MOLProbity (Chen et al., 2010), which showed that they had an overall good quality (see Model).
Once we had 3D structures for the toehold riboswitch, we mounted a system for molecular dynamics simulations with the CHARMM-GUI server (Lee et al., 2016; Jo et al., 2008) and the NAMD simulation engine (Phillips et al., 2005). This would allow us to explore the dynamics of our structural model and to evaluate the flexibility of the ends of the molecule. We ran two simulations starting from our RNAComposer model: a computationally tractable one using an implicit solvent and a computationally intensive one using explicit solvent. Briefly, the simulation with implicit solvent uses the Born generalization to derive an approximation of the effect of solvation on each atom in the toehold’s structure, while the simulation with explicit solvent actually uses distance measurements to water molecules to calculate the magnitude of their interactions with the toehold (Anandakrishnan et al., 2015). In the simulations, we observe qualitatively that under an implicit solvent the toehold riboswitch starts unfolding after the first 40 ns (Movie 1) but not in the explicit solvent simulation (Movie 2). We analyzed the number of hydrogen bonds formed in the hairpin throughout the simulations using VMD built-in tools (Humphrey et al., 1996). In the explicit solvent simulation, the number of hydrogen bonds stays closer to the theoretical optimal expectation, which suggests a stable hairpin (Figure 3). This confirms the advantages of using an explicit solvent to obtain more accurate results from simulations. However, it also suggests that there could be potential for design improvements to prevent leaky expression caused by a spontaneous unfolding of the toehold riboswitch. We studied which hydrogen bonds were detected in most of the simulation and which were lost to find potential sites for design improvement (Figure 4). We found that the most stable hydrogen bonds were mediated by GC pairings at the base and the top of the hairpin, while the least stable ones were located closer to the AUG mismatch near the center of the hairpin. This result suggests that even more stable structures could be obtained by selecting sequences that have a higher GC content or ensuring that the last pairing at the top of the hairpin is a GC pair. Nevertheless, an excessively stable structure could bind inefficiently to the trigger sequence, so this potential trade-off between preventing leaky expression and target recognition. Further simulations will allow us to provide insight into this design aspect and further improvements to our workflow.
Wet lab
Switch cloning and testing
The first goal of this project was to demonstrate that we could produce efficient and specific ToeHold switches. To do so, we first designed switches targeting genes in E. coli. This design choice was so that once the switches were designed, a simple transformation and fluorescence measurement in vivo could allow us to assess switch efficiency. However, we had a lot of trouble with the initial cloning and had to change our destination vector, resulting in complications for this kind of test. Furthermore, our initial designs often targeted essential genes in E. coli. This was suboptimal, as the expression of the ToeHold could encourage the formation of RNA duplexes in the cell, resulting in lowered expression of essential proteins and a subsequent lowering in cell fitness. Our final design relied on the plasmid pNZ123, a small plasmid with a rolling circle replication origin and a chloramphenicol resistance.
As the switch that we decided to move forward with targeted ampicillin resistance, we had to prepare a co-transformation of the switch-containing plasmid and another plasmid with a compatible replication origin and an ampicillin resistance gene. At this point, we needed a specific strain of E. coli, BL21 DE3, as it possesses an insertion from the lysogenic phage T7, and can express its polymerase, which is necessary for ToeHold expression. Sadly, producing chemically competent cells for this strain proved more complex than planned, and the in vivo part was dropped.
We then decided to move forward with the cell-free characterization of our part BBa_K3026001. We placed the plasmid harboring this part in myTXTL cell-free expression system from ArborBiosciences, along with a constant amount of the total mRNA sample extracted from bacterial cultures. As we wanted to replicate the detection apparatus from our tool, we decided to do four biological (BR) and three technical replicates for each biological ones, to a total of 12 replicates. For each biological replicate, we did an mRNA-less reaction to confirm the absence of signal leakage, and also did a control with only myTXTL and mRNA, to confirm that signal was not from autofluorescent proteins produced from the mRNA.
This figure clearly demonstrates that fluorescence was detected in only samples with both the ToeHold switch and the mRNA from ampicillin-resistant E. coli. All negative controls show no fluorescence whatsoever. This data confirms that this part works as intended and that our ToeHold design workflow produces functioning ToeHolds. However, there is a lot of variation between BR and even among technical replicates. For the design of our tool, this confirmed the design choice to incorporate several detection chambers, and treat as much of the initial sample as possible.
Sequencing of the clones used for the BR also showed that the part sequence was as designed for the entire length of the sequence, further confirming the efficiency and specificity of this part.
With these results, we knew that the sequence constraints we used for our switch design was efficient, and we moved forward to create more switches. Sadly, due to time and monetary constraints, we could not characterize other parts we designed further. Fluorescence reference curve for part characterization is available at the bottom of the page. For the future, we intend to test more switches that we designed to confirm our design and tool further, as well as benchmark our switches more thoroughly to assess the proper detection limit for the various switches. While there is still optimization work that needs to be done, the initial results with the tested switch show that we are on the right track and that our workflow is efficient.
Characterization of already available parts
As we were doing most of our experiments in cell-free media, we looked for parts that had been characterized in such a media as points of reference. However, we found very few resources regarding this kind of system, especially for fluorescent protein expression. Therefore, we chose to select a part that was initially designed for in vivo use, and express it into our cell-free system to compare fluorescence levels and efficiency of expression. We chose parts BBa_K567018 and BBa_I746907 for this experiment. We produced fluorescence curves for these two proteins and made them available for the IGEM community to use as a comparison point between in vivo and cell-free expression of fluorescent proteins.
Figure 7, A and B: Fluorescence curves for the two pre-existing characterized parts.
Hardware
The only hardware we were able to produce from scratch was the microfluidics circuits. This was a very complex part, as we had no prior experience with microfluidics. We are very proud that we were able to produce this device and test liquid flow through it. The device contains two layers, once where the sample will flow, and another where valves are present to control sample flow.
Now that the circuit is ready, the next step will be to integrate all the sample treatment parts into the chip and set up the automatic sample treatment. Finally, creating the A.D.N. device itself and incorporate the cartridge system in the device. This part will require the help of more engineers, but we are hopeful that the logic of our design is sound, and that the only hurdles left are technical and monetary ones.
Anandakrishnan, Ramu, Aleksander Drozdetski, Ross C. Walker, and Alexey V. Onufriev. 2015. “Speed of Conformational Change: Comparing Explicit and Implicit Solvent Molecular Dynamics Simulations.” Biophysical Journal 108 (5): 1153–64.
Cheng, Clarence Yu, Fang-Chieh Chou, and Rhiju Das. 2015. “Modeling Complex RNA Tertiary Folds with Rosetta.” Methods in Enzymology 553 (February): 35–64.
Chen, Vincent B., W. Bryan Arendall 3rd, Jeffrey J. Headd, Daniel A. Keedy, Robert M. Immormino, Gary J. Kapral, Laura W. Murray, Jane S. Richardson, and David C. Richardson. 2010. “MolProbity: All-Atom Structure Validation for Macromolecular Crystallography.” Acta Crystallographica. Section D, Biological Crystallography 66 (Pt 1): 12–21.
Das, Rhiju, John Karanicolas, and David Baker. 2010. “Atomic Accuracy in Predicting and Designing Noncanonical RNA Structure.” Nature Methods 7 (4): 291–94.
Green, Alexander A., Pamela A. Silver, James J. Collins, and Peng Yin. 2014. “Toehold Switches: De-Novo-Designed Regulators of Gene Expression.” Cell 159 (4): 925–39.
Humphrey, W., A. Dalke, and K. Schulten. 1996. “VMD: Visual Molecular Dynamics.” Journal of Molecular Graphics 14 (1): 33–38, 27–28.
Jo, Sunhwan, Taehoon Kim, Vidyashankara G. Iyer, and Wonpil Im. 2008. “CHARMM-GUI: A Web-Based Graphical User Interface for CHARMM.” Journal of Computational Chemistry 29 (11): 1859–65.
Lee, Jumin, Xi Cheng, Jason M. Swails, Min Sun Yeom, Peter K. Eastman, Justin A. Lemkul, Shuai Wei, et al. 2016. “CHARMM-GUI Input Generator for NAMD, GROMACS, AMBER, OpenMM, and CHARMM/OpenMM Simulations Using the CHARMM36 Additive Force Field.” Journal of Chemical Theory and Computation 12 (1): 405–13.
Phillips, James C., Rosemary Braun, Wei Wang, James Gumbart, Emad Tajkhorshid, Elizabeth Villa, Christophe Chipot, Robert D. Skeel, Laxmikant Kalé, and Klaus Schulten. 2005. “Scalable Molecular Dynamics with NAMD.” Journal of Computational Chemistry, Springer Seri, 26 (16): 1781–1802.
Popenda, Mariusz, Marta Szachniuk, Maciej Antczak, Katarzyna J. Purzycka, Piotr Lukasiak, Natalia Bartol, Jacek Blazewicz, and Ryszard W. Adamiak. 2012. “Automated 3D Structure Composition for Large RNAs.” Nucleic Acids Research 40 (14): e112.
Purzycka, K. J., M. Popenda, M. Szachniuk, M. Antczak, P. Lukasiak, J. Blazewicz, and R. W. Adamiak. 2015. “Chapter One - Automated 3D RNA Structure Prediction Using the RNAComposer Method for Riboswitches1.” In Methods in Enzymology, edited by Shi-Jie Chen and Donald H. Burke-Aguero, 553:3–34. Academic Press.
Zadeh, Joseph N., Conrad D. Steenberg, Justin S. Bois, Brian R. Wolfe, Marshall B. Pierce, Asif R. Khan, Robert M. Dirks, and Niles A. Pierce. 2011. “NUPACK: Analysis and Design of Nucleic Acid Systems.” Journal of Computational Chemistry 32 (1): 170–73.