Design

Toehold switches are programmable synthetic riboregulators that regulate translation in response to a specific trigger RNA.

We have designed genetic constructs which contain toehold switch sensors that are activated in the presence of specific microRNAs (miRNAs). The toehold switches regulate translation of the reporter protein GFP, allowing for fluorometric analysis to obtain the level of a target miRNA in a body fluid.

Our toehold switches are based off the design parameters delineated by Green et al.^[1] and Pardee et al.^[2] We designed two series of sequence specific toehold switches to sense our target miRNAs – hsa-miR-15b-5p and hsa-miR-27b-3p, and propose novel riboregulators with improved specificity and reaction kinetics.

We designed our first series of toehold switches to have very low leakage. However, these switches were incapable of discriminating between miRNA sequences with close sequence homology. We therefore designed a second series of toehold switches, which our modeling showed are capable of achieving single base mismatch specificity.

The software package NUPACK is capable of predicting the secondary structure of one or more interacting nucleic acid strands.^[3] We used NUPACK to model our toehold switch prototypes, enabling us to obtain toehold switches that fulfil the desired parameters discussed below.

First series toehold switches

Toehold switches contain several predetermined regions: the RBS (ribosome binding site), the start codon, the 21 nucleotide (nt) linker sequence between the hairpin structure and the protein coding sequence. We developed a very simple software tool that sped up the design process by automating these regular structures.

We designed two sets of toehold switches for characterisation of two different reporter proteins, GFP and luciferase, to establish the most suitable reporter protein for our system. We tested GFP to determine if fluorescent proteins would be suitable for our sensor. Fluorescent proteins have a wide spectral variety and strong signal output,^[4] making them suitable for a multiplexing assay. On the other hand, luciferases are much more sensitive and have faster maturation times.^[5] Although we also wanted to test luciferase we found experimentally that it couldn’t be inserted into our plasmid.

We used the RBS (AGAGGAGA) and linker sequence (AACCUGGCGGCAGCGCAAAAG) used by Green et al (2014). The RBS is strong and lacks a secondary structure. The linker sequence ensures correct formation of the toehold switch in both the off and on state. In the off state, this linker sequence does not bind to the toehold region, and in the on state it does not bind with the RBS and start codon. Consequently, we used this linker sequence to allow for better comparison of dynamic ranges achieved with previously designed toehold switches.

The target miRNA that acts as the trigger RNA binds to the trigger binding site. The trigger binding site is the sequence found at the 5’ end of the toehold switch and is the reverse complement of the target miRNA sequence. The trigger binding site includes the toehold region and lower stem. Because mature miRNAs are only ~22nt in length, the entire miRNA sequence binds to the trigger binding site. Therefore, the trigger binding site and the region complementary to the trigger binding site in the stem are pre-determined sequences.

The binding of the trigger miRNA is initiated in the toehold region. The toehold region is a linear region found at the 5’ end of the trigger binding site that is complementary to the 3’ end of the trigger miRNA (miRNAs bind to the 5’ end of the toehold switch in a 3’-5’ direction).

Toehold length

Dynamic range in previous toehold switches increased with an increasing toehold region length, up to 12nt.^[1] Longer toehold regions result in more stable miRNA-toehold duplexes, significantly reducing the risk of spontaneous dissociation of the miRNA from the toehold region.^[6]

We designed our toehold switches to contain longer toehold regions, preferably 12nt in length. However, this was not possible to achieve with one of the switches, for reasons discussed in the Homologs section.

Base of the stem

The miRNA completes a branch migration with the base of the stem, resulting in strand displacement. Our switches’ toehold regions are 9 and 12nt long and the miRNAs are 22 and 21nt long respectively. The miRNAs can therefore complete a branch migration with a stem of 13 and 9nt.

The longer the stem, the more thermodynamically stable it is in the off state. Therefore more energy is required to unravel the stem, reducing leakage in the off state. The branch migration is completed just before the mini-loop which contains the start codon. This results in a smaller secondary structure around the RBS in the on state, that can easily be melted by the ribosome.

Mini-loop

The start codon is left largely unpaired. This ensures that there is no pairing with the start codon in the on state. This is important as the miRNA is too short to be able to displace a sequence bound to the start codon. However, we placed a weak wobble-pair in the mini-loop to increase stem stability in the off state. This wobble-pair breaks apart in the on state as the surrounding secondary structure is unraveled during the branch migration, so the on-state activity is not affected.

Top of the stem

Reduced G-C content in the top of the stem increases dynamic range. This is due to the decreased free energy required for the ribosome to melt the stem in the on state, as fewer bonds per nucleotide need to be broken. Therefore, we designed our switch to have only one G-C base pair at the top of the stem to ensure correct loop formation, whilst allowing for a high dynamic range to be achieved.

System leakage

Our models demonstrated that toehold switches needed to have very low levels of leakage to reduce background levels of fluorescence as we are working with small quantities of miRNAs. For more information, see our mass action kinetics model.

To reduce leakage we decreased the size of the loop containing the RBS and eliminated the downstream refolding domain present in the original toehold switches by Green et al.

Loop size

Decreasing loop size encourages stem formation and thus toehold switch formation.

Decreasing loop size also decreases leakage in the off state. This is due to the reduced accessibility of the RBS to ribosomes. However, on state activity is also decreased due to reduction in loop mediated docking.

Therefore, we designed our switches to contain a loop region of just 12-15nt long. Furthermore, we had an adenine rich pre-RBS sequence in the loop, which increases translational efficiency.^[7] This also ensured no base-pairing occurred within the loop and aided in correct formation of the smaller secondary structure in the on state.

Eliminating downstream refolding

As bases are exposed during branch migration, RNA refolding can occur. Smaller secondary structures that are able to form as a result decrease the energetic barrier to further strand displacement. This increases the rate of stem unraveling and thus switch activation. However, it reduces the free energy required to unravel the stem, increasing leakage in the off state. We therefore ensured that our sensors lack the downstream refolding domain.

Homologs

We used miRBase to find mature miRNAs with similar sequences to our target miRNAs - hsa-miR-15b-5p and hsa-miR-27b-3p.^[8][9] For brevity we’ve removed the hsa-miR prefix and the 3p and 5p suffix from miRNAs discussed here onwards.

We found that both miRNAs each had one homolog that demonstrated significant sequence similarity. 15a-5p is similar to 15b-5p and 27a-3p is similar to 27b-3p.

In our NUPACK simulations, the first series of toehold switches for 27b were unable to prevent the homolog 27a binding as 27a only differs from 27b by one nucleotide. Conversely, we were able to dramatically reduce the effects of the homolog 15a for the sensor for 15b by altering the kinetic barriers of strand displacement.

Proximal mismatches are mismatches adjacent to the toehold region. They markedly reduce the probability of successful strand displacement. Proximal mismatches cause a homologous miRNA to immediately enter a significantly energetically less-favourable state to begin branch migration as it must enclose a mismatch.

This reduces the rate of transition of the homologous miRNA from being bound only to the toehold region to also being partially bound to the stem. This causes the homologous miRNA to spend more time bound only to the toehold region. This increases the probability of the homologous miRNA detaching from the toehold region. Short toehold regions are particularly prone to this as they form fewer bonds resulting in a more transient toehold binding. Additionally, even if the homologous miRNA does partially bind to the stem, the branch point is more likely to move back down the stem due the unstable mismatch in the newly-formed duplex.^[6] We used these principles to significantly decrease toehold switch activation by 15a.

The homolog 15a contains three consecutive mismatches at its 3’ end and one mismatch centrally located in the miRNA. To position the central mismatch at the base of the stem (a proximal mismatch), the stem length was increased to 13nt. As a result, the first codon after the start codon was a stop codon. Therefore, we removed the first nucleotide of the stop codon, resulting in a stem length of 12nt. Consequently, the miRNA has an overhang of one nucleotide. This left a toehold region 9nt long.

Since 15a has three mismatches at its 3’ end, it can only bind to 6nt of the toehold region. The short toehold region results in a less stable miRNA-toehold duplex. Additionally, 15a is forced to spend a considerable amount of time weakly bound to only the toehold region due to the proximal mismatch. The combination of these factors drastically increases the probability of 15a detaching from the toehold. The number of toehold switches activated by 15a is therefore made negligible.

The increased stem length also makes the toehold switch more stable, which reduces leakage. However, it also makes it harder to complete strand displacement, thus reducing on state activity. Therefore we used a slightly larger loop size of 15nt to counteract the increased stem stability.

A further strategy to increase the specificity of the first series of toehold switches involves using CRISPR C2c2. C2c2 is a CRISPR effector protein that can be programmed to cleave single stranded RNA with single base mismatch specificity.^[10][11] It may therefore be possible to use CRISPR C2c2 to cleave homologous miRNAs to prevent them from activating the toehold switch. However, miRNAs may be too short to be targeted by C2c2 and the additional use of CRISPR C2c2 increases the cost of the system.

Validating our NUPACK modeling

Green et al. found that the term ∆GRBS-Linker is the best indicator of switch performance.^[1] The ∆GRBS-Linker is the free energy of the structure formed by the sequence immediately after the trigger binding site through to the end of the 21nt linker. Typically, the smaller the |∆GRBS-Linker| is, the higher the dynamic range. We found that our |∆GRBS-Linker| values were quite small, so we therefore expected to achieve quite large dynamic ranges.

Our experiments back this up. See our results page for more information.

Expressing our toehold switch constructs

To express our toehold switch constructs, we used the inducible promoter BBa_K808000. This was necessary as we found that the transcripts produced from our constructs were toxic to E. coli, making successful amplification impossible. Therefore, we chose BBa_K808000, a promoter with low levels of leakage, to express our parts, allowing for amplification in E. coli, although we still had some complications. In future designs, we recommend using a promoter that has even lower leakage than BBa_K808000 (e.g. BBa_K1067007) to allow for successful amplification of the toehold switch constructs before use in the cell free system.

Design summary

Second series toehold switches

Our sensors must be able to distinguish between the target miRNAs and homologs. Therefore, we designed a second series of switches that can distinguish between two strands which differ by just one nucleotide. Our second series of toehold switches utilize a mechanism that colocalizes several RNA molecules to activate a single toehold switch^[13].

Toehold switches are typically activated by a single trigger RNA. However, it is possible to divide the trigger RNA sequence into separate RNA molecules. The RNA molecules contain hybridization domains that can be used to join them together to form the complete trigger RNA sequence. As a result, the toehold switch sequence will only be activated when all the separate RNA molecules combine.

The anti-miRNAs could be produced in the cell free system alongside the toehold switches. To conserve resources in the cell free system, an inducible promoter should be used for expressing the anti-miRNA sequence. We recommend using the same inducible promoter for both the anti-miRNA and toehold switch for ease. Since, the toehold switch transcripts are toxic in E. coli, this promoter must have very low leakage. For reference, we found BBa_K808000 to be too leaky when expressing our part BBa_K2206007.

Anti-miRNA design

The anti-miRNA can be designed to bind to either the 5’ or 3’ end of the target miRNA. The end chosen depends on the nucleotides that differ between the target miRNA and its homologs. If the differing nucleotides are in the 5’ end of the miRNAs, then the anti-miRNA binds to the 5’ end and vice versa. This is because the miRNA to anti-miRNA hybridization step is the source of the single base mismatch specificity of the new toehold switches.

The only sequence constraint imposed upon the anti-miRNA is the hybridization domain, which must be perfectly complementary to the 12nt of the target miRNA that it binds to. The rest of the anti-miRNA can be programmed to contain any sequence of any length.

Toehold switches are designed based on a specific trigger RNA sequence. The non-hybridization region of the anti-miRNA has no sequence constraints, so the region of the toehold switch complementary to it is fully programmable.

The anti-miRNA can either bind to the 3’ or 5’ end of the target miRNA. This determines the exposed sequence and thus how the toehold switch can be designed.

1. If the anti-miRNA binds to the toehold region, it allows for a toehold region of any sequence and length. A programmable toehold region can be designed such that it does not form a secondary structure, and the linker does not bind to it. It also allows for high G-C content, ensuring highly stable binding of the anti-miRNA, reducing the probability of the anti-miRNA detaching.
2. If the anti-miRNA binds to the base of the stem, it allows for a stem of any sequence and length. The length and G-C content of the stem could then be chosen to reduce leakage. The length of stem unraveled could also be changed to allow for a larger part of the teohold region to be unraveled, providing a lower |∆GRBS-Linker|.

For both the second series switches the anti-miRNA is exposed at its 3’ end. The exposed anti-miRNA sequence therefore binds to the toehold region. This exposed sequence is 12nt long and contains all guanines to ensure stable binding to the toehold region. This also prevents secondary structure formation within the toehold region and with the linker sequence.

Design summary

The second series of toehold switches were forward engineered using similar design principles employed in the first series of toehold switches.

Molecular beacons are able to distinguish between their targets more easily than linear probes (such as anti-miRNA) as there is a wider range of temperatures where the fluorescence from the target RNA (solid) is at a maximum whilst for the homologous RNAs (dashed) it is at a minimum.^[17]

Our changes

A strong secondary structure in the 5’ region directly adjacent to the RBS would significantly decrease translation by preventing contacts between the ribosome’s 30S subunit and the pre-RBS region.^[18]

Therefore, a programmable hairpin loop immediately upstream of the RBS, that changes conformation in response to an arbitrary RNA sequence, could be used to regulate translation of a fluorescent or luminescent reporter protein.

The loop of the hairpin structure should be complementary to part or all of the target miRNA. When the target miRNA binds to the loop, the arms of the hairpin structure are pried apart. This relieves the secondary structure immediately upstream of the RBS, allowing for translation of a reporter protein. Consequently fluorescence or luminescence intensity is indicative of miRNA concentration.

Due to the complete lack of sequence constraints, it is possible to have no secondary structure around the start codon, increasing on state activity. Furthermore, the lack of branch migration results in more favourable reaction kinetics. Finally, the high specificity allows for distinction between miRNAs with close sequence homology.

We believe that we could achieve substantially larger fold changes with these new riboregulators in comparison to our first series of toehold switches. The larger fold change and use of luciferase could allow for much miRNA lower concentrations of miRNAs to be measured. Therefore, these new riboregulators would be well suited for sensitive and specific miRNA quantification.

Implementation of the sensors

Constructs containing a toehold switch and reporter protein can be embedded into paper along with transcription and translation machinery from E. coli by freeze-drying, resulting in a cheap and safe test that can even be used in LEDCs. A cell free sensor also allows for easy addition of miRNAs. Each strip of paper could contain multiple wells, each containing a toehold switch for a specific miRNA. However, toehold switches have very low levels of crosstalk^[1], which make it possible to create a multiplexing assay. By having multiple switches that each regulate the production of a different reporter protein with a distinct emission peak in one well, the efficiency of the test would increase drastically.

See our silver human practices page to learn more about the clinical implementation of our project.

References

1. ↑ ^{1.0 1.1 1.2 1.3} 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.
2. ↑ Pardee, K., Green, A. A., Takahashi, M. K., Braff, D., Lambert, G., Lee, J. W., ... & Daringer, N. M. (2016). Rapid, low-cost detection of Zika virus using programmable biomolecular components. Cell, 165(5), 1255-1266.
3. ↑ Zadeh, J. N., Steenberg, C. D., Bois, J. S., Wolfe, B. R., Pierce, M. B., Khan, A. R., ... & Pierce, N. A. (2011). NUPACK: analysis and design of nucleic acid systems. Journal of computational chemistry, 32(1), 170-173.
4. ↑ (n.d.). The Advantages of Fluorescent Proteins over Luciferase for In Vivo Imaging. Retrieved October 25, 2017, from https://www.uvp.com/pdf/FP129.pdf
5. ↑ Zhang, Y., Phillips, G. J., & Yeung, E. S. (2008). Quantitative imaging of gene expression in individual bacterial cells by chemiluminescence. Analytical chemistry, 80(3), 597-605.
6. ↑ ^{6.0 6.1} Machinek, R. R., Ouldridge, T. E., Haley, N. E., Bath, J., & Turberfield, A. J. (2013). Programmable energy landscapes for kinetic control of DNA strand displacement. Nature communications, 5, 5324-5324.
7. ↑ Vimberg, V., Tats, A., Remm, M., & Tenson, T. (2007). Translation initiation region sequence preferences in Escherichia coli. BMC molecular biology, 8(1), 100.
8. ↑ Kozomara, A., & Griffiths-Jones, S. (2013). miRBase: annotating high confidence microRNAs using deep sequencing data. Nucleic acids research, 42(D1), D68-D73.
9. ↑ Griffiths-Jones, S., Grocock, R. J., Van Dongen, S., Bateman, A., & Enright, A. J. (2006). miRBase: microRNA sequences, targets and gene nomenclature. Nucleic acids research, 34(suppl_1), D140-D144.
10. ↑ Abudayyeh, O. O., Gootenberg, J. S., Konermann, S., Joung, J., Slaymaker, I. M., Cox, D. B., ... & Severinov, K. (2016). C2c2 is a single-component programmable RNA-guided RNA-targeting CRISPR effector. Science, 353(6299), aaf5573.
11. ↑ Gootenberg, J. S., Abudayyeh, O. O., Lee, J. W., Essletzbichler, P., Dy, A. J., Joung, J., ... & Myhrvold, C. (2017). Nucleic acid detection with CRISPR-Cas13a/C2c2. Science, eaam9321.
12. ↑ Machinek, R. R., Ouldridge, T. E., Haley, N. E., Bath, J., & Turberfield, A. J. (2013). Programmable energy landscapes for kinetic control of DNA strand displacement. Nature communications, 5, 5324-5324.
13. ↑ Green, A. A., Kim, J., Ma, D., Silver, P. A., Collins, J. J., & Yin, P. (2016, September). Ribocomputing devices for sophisticated in vivo logic computation. In Proceedings of the 3rd ACM International Conference on Nanoscale Computing and Communication (p. 11). ACM.
14. ↑ Baker, M. B., Bao, G., & Searles, C. D. (2011). In vitro quantification of specific microRNA using molecular beacons. Nucleic acids research, 40(2), e13-e13.
15. ↑ Tyagi, S., & Kramer, F. R. (1996). Molecular beacons: probes that fluoresce upon hybridization. Nature biotechnology, 14(3), 303-308.
16. ↑ Shore, D., Langowski, J., & Baldwin, R. L. (1981). DNA flexibility studied by covalent closure of short fragments into circles. Proceedings of the National Academy of Sciences, 78(8), 4833-4837.
17. ↑ Bonnet, G., Tyagi, S., Libchaber, A., & Kramer, F. R. (1999). Thermodynamic basis of the enhanced specificity of structured DNA probes. Proceedings of the National Academy of Sciences, 96(11), 6171-6176.
18. ↑ Malmgren, C., Engdahl, H. M., Romby, P., & Wagner, E. G. (1996). An antisense/target RNA duplex or a strong intramolecular RNA structure 5'of a translation initiation signal blocks ribosome binding: the case of plasmid R1. Rna, 2(10), 1022-1032.