Eukaryotic Toehold Design for Differential Translation
Our project aims at constructing a platform for the design of mRNA sequences in which translation is dependent on the existence of other specific mRNA elements in the cell. To attest to our project’s success in advancing toward this goal, it was important for us to show that we could improve current eukaryotic toeholds, specifically by using an mRNA molecule as a trigger (in contrast to previous studies which utilized microRNA as trigger). We were able to use for our comparisons the plasmid used in Wang and colleagues’ work [1] and examine different toehold designs both in mammalian cells and in yeast. Results are further described in the results section. We were able to attain specific GFP expression in yeast in two types of toehold design: the classical toehold design reported in bacteria [2], where the start codon and Kozak sequence are found inside the stem-loop; and a toehold in which the start codon and Kozak are found downstream to the stem-loop, which also worked apparently better than Wang et al.’s toehold in the mammalian cell line (see Results).
We believe our results demonstrate our program’s ability to produce toehold sequences working in eukaryotes. Moreover, to our understanding, these are the only results reported showing the usage of mRNA to trigger toeholds of this structure in eukaryotes (see Figure below). Moreover, sequences ranked the highest when using our program for our toehold assessment indeed presented the highest fold-increases when tested in-vivo providing evidence of our software’s strength (Supplementary – cycles’ sequences).
Figure 1: Fluorescence results for the second cycle in yeast
Sequences 1 and 6, predicted to be the optimally active sequences in each category (Supplementary – cycles’ sequences) show relatively high fold-increase. Significance between mCherry treated and respective toehold-only wells was assessed via student’ t-test assuming unequal variance. Graph further described in Results/Figure 6.
(* p-value < 0.05, ** p-value < 0.01, *** p-value < 0.001, **** p-value < 0.0001.)
References
- Wang, S., Emery, N. J., & Liu, A. P. (2019). A novel synthetic toehold switch for microRNA detection in mammalian cells. ACS synthetic biology, 8(5), 1079-1088.
- 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.
Possible Triggers Based on Cancer Related Mutations
As part of our POC, we wanted to show that cancerous cells could be differentiated from normal cells in a way that will enable us to generate a toehold switch that will open only within the cancer cells and not in the corresponding healthy tissue. We desired to find mutations that appear in a large number of patients, and use the segment that contains those mutations as our trigger molecule. By the courtesy of Dr. Yoram Zarai from the start-up OncoDecipher, we got access to a large amount of restricted data from the NIH. The data contained every mutation that was found in 787 genes of 8300 patients suffering from various types of cancer. The distribution of the number of patients by cancer types is presented in figure 1.
Figure 1: - Distribution of number of patients by cancer type
At first, we were looking for the most common mutations across all genes. We tried to pinpoint a mutation that: 1) it is common to many cancer patients. 2) It induces in the cancer cell an mRNA window that is significantly different than the windows appear in the mRNA from the healthy tissue.
To validate point 2) above, we specifically computed the hybridization free energy between the reverse complementary sequence of the cancerous segment (the candidate for being the trigger binding site of the switch) and the corresponding wild-type sequence, to make sure it is not too strong (i.e. to make sure that the switch could not be opened by the wild-type mRNA sequence).
We first restricted our search for deletion and insertion mutations, hoping to find as long mutations as possible in a row. We found out that all the mutation segments (50 mutations) shared by more than 150 patients were no longer than 3 nucleotides, except for one mutation sequence that spanned over 6 nucleotides which is shared by 779 patients. We assumed that a deletion of up to 3 nucleotides will not make a significant change so we decided to analyze the effect of this mutation. We did so by taking 30-nucleotide windows around the mutation and comparing the hybridization MFE of the trigger binding site candidate with the mutant window and the wild-type window. The maximal MFE difference between those structures across all windows was 12 kcal/mol. We suspected that this difference is not significant enough and decided to expand our search, hoping to find a better window.
Subsequently, we started to search for several mutations that are close to each other in the sequence, wishing to find a 30-40 nucleotides window that contains at least two different mutant sequences. We assumed that several mutant sequences on different but close positions in the window, even if their length is smaller than 6, could make that window a good candidate for serving as our trigger. If this sequence is different enough, in a way that the hybridization with the corresponding wild-type window will be weak enough, the switch will not get opened by the wild-type sequence. We found out that the case that we were looking for exists in PABPC1 gene – there are 2 deletion mutations, one spans over 10 nucleotides and the other over 2, that are found in a distance of 2 nucleotides between each other. A total of 92 patients share the co-occurance of these two mutations. Those mutations are detailed in table 1.
Table 1: Details on chosen mutations
We analyzed this case and discovered that our assumption was right – those mutations were significant enough to differentiate the cells, according to our analysis. At first, we took 30-nucleotides windows around the mutations and looked for the biggest MFE difference between the structures (mutant window – trigger binding site candidate & wild type window – trigger binding site candidate) across all windows. We calculated the MFE of the structures using Vienna RNA[1]. We found out that for the best window, the MFE of the structure with the wild-type window was -20.27 kcal/mol, whereas the MFE of the structure with the mutant window was -54.89 kcal/mol.
The dot-bracket notations of the best window are presented in figures 2 and 4. In addition, the secondary structure of the windows with the trigger binding site candidate are presented in figures 3 and 5.
Figure 2: Dot-bracket notation of wild-type window and trigger binding site candidate, predicted by Vienna RNA
Figure 3: RNA secondary structure of wild-type window and trigger binding site candidate, predicted by foRNA
The green strand represents the complementary mutant sequence, whereas the red strand represents the wild-type sequence. It is worth mentioning that the beginning of the switch sequence (from the 5’ end) is the most crucial part for opening the structure by hybridization to another molecule. In that aspect, the results above are promising because the beginning of what would be the trigger binding site remains open, which implies that the wild-type sequence will not lead to an opening of the switch structure.
Figure 4: Dot-bracket notation of mutant window and trigger binding site candidate, predicted by Vienna RNA
Figure 5: RNA secondary structure of mutant window and trigger binding site candidate, predicted by foRNA
Following those results, we have decided to generate a toehold switch sequence based on this mutant window in order to validate this finding. The toehold switch sequence and secondary structure are presented in figure 6.
Figure 6: - Toehold switch generated based on mutant window;
Toehold sequence: AAAGGUUGAGUUUAGGGGGGGGUGGAAUAUGAGGAUUCGUUACAGCCACCAUGGGAUUUUUAUAUUUUGCC
We checked the hybridization between our switch and the mutant window, that serves as our trigger molecule, and saw that there is an optimal binding, the trigger binds entirely to the switch, in the right positions, and by that opening it for translation. The dot-bracket notation and secondary structure of this complex is shown in figure 7.
Figure 7: Dot-bracket notation and secondary structure of toehold + mutant window
In contrast, when checking the hybridization with the wild-type window, we see that the wild-type window does not bind completely to the switch, leaving it closed. The dot-bracket notation and secondary structure of this complex is shown in figure 8.
Figure 8: Dot-bracket notation and secondary structure of toehold + wild-type window
We also looked at the difference in the MFE between those 2 complexes. The MFE of the complex with the wild-type window is -31.23 kcal/mol, whereas the MFE of the complex with the mutant window is -47.29 kcal/mol.
These results strengthen our assumptions that cancer cells could be distinguished from normal cells based on mutation segments that are close to each other in the sequence, in a way that enables us to design a toehold switch that will be activated by the mutant cells but not by the normal cells.
Since we got access to this data at a late stage of the project, we still have a lot more analysis that we plan to do. We intend to take into account every mutation, not only deletions and insertions, and analyze the data by cancer type. We also wish to test the toehold switch we designed and many more that will be generated based on our future analysis in a wet lab, in order to get a stronger validation of our results.
References
- Lorenz, R., Bernhart, S. H., Höner zu Siederdissen, C., Tafer, H., Flamm, C., Stadler, P. F., & Hofacker, I. L. (2011). ViennaRNA Package 2.0. Algorithms for molecular biology, 6(1), 1-14.