Background

According to Green et al. (2014) [1], the optimal length of RNA to be detected by a toehold switch is around 30 bp (complementary region at below figure). In other words, a target RNA with 1000 bp in length can have 970 possible switches with different performance, which is governed by their structures and thermodynamic parameters. We aim at providing a workflow and platform of modeling to help users design the switches by reducing the manual processing and increasing the hit-rate of finding a good switch.

We adopted the toehold switch design from the original paper (below figure). Our toehold switch contains 15 nts “toehold domain”, 18 nts stem and a loop that contains the RBS B0034. A 21 nts linker sequence plus an mRFP reporter sequence is present downstream the toehold switch. The linker is used to separate the coding sequence in the toehold switch and the reporter to prevent interference of protein folding.

RNA thermodynamic Modelling

Assumption 1: Switch MFE (Minimum Free Energy) correlates with the expression leakage

● Switch MFE is the minimum Gibbs free energy that a toehold switch could have among all the possible structures.
● Expression leakage is a phenomenon where the reporter (i.e. Red Fluorescent Protein in our project) is expressed in the absence of trigger RNA. The level of leakage can be measured as:

To activate toehold switch, an amount of energy is needed to open the toehold switch hairpin. The Switch MFE reflects the difficulty for the toehold switch unwinding process. We assume that the more negative the Switch MFE, the harder for the unwinding to take place, and hence a lower leakage.

● ΔG_RBS-Linker is the Gibbs free energy of the RNA sequence starting from the RBS to the linker in the switch-trigger duplex.
● The duplex expression is the reporter expression of the switch-trigger dimer RNA.
After the switch RNA hairpin is unwound after binding to the trigger RNA, a switch-trigger dimer RNA would be formed. The RBS-linker region of the MFE structure of this dimer RNA should have minimal base pairs. This makes it easier to unwind the RNA for this region, allowing ribosomes to bind to the RBS and move along the RNA for translation of the RFP reporter gene to occur. ΔG_RBS-Linker reflects the difficulty for the unwinding process of the RBS-linker region. It is assumed that the more negative the ΔG_RBS-Linker , the harder it is for the unwinding to take place, leading to lower translation rates. Thus, the duplex expression would be reduced.

Assumption 3: ΔMFE correlates with the reporter expression of Switch-Trigger duplex

● ΔMFE is defined as MFE of the switch-trigger duplex RNA minus (switch RNA MFE + trigger RNA MFE).
Since ΔMFE = –RTlnK, where:
R = Gas constant
T = Temperature
K = Equilibrium constant
Therefore, we assume that the more negative the MFE difference is, the higher the switch-trigger duplex RNA concentration compared to that of the switch RNA when in equilibrium.

Consequently, increased equilibrium concentrations of the switch-trigger duplex RNA would provide an increased number of active mRNAs for the translation of the reporter RFP.

(THE toehold domain base pairing modelling would be covered by the switch RNA suboptimal structure modelling part)

Screening by our software

To minimize the manpower on screening of the switches, we constructed an online toehold switch design program. Apart from basic thermodynamic parameters, it also screens for other factors(please visit our software page . Ultimately, the program generated a list of possible toehold switch sequences according to many different free energy parameters using the ViennaRNA package [2]. The graph below shows 394 possible H5 toehold switches generated by our software. The assumptions motioned earlier stated 3 very important parameters for the selection of switch candidates with the greatest possible performances: Switch MFE, ΔG_RBS-linker, and ΔMFE. We applied these parameters to our switch selection process: We first chose the switches that with the highest ΔG_RBS-linker (-3.8 kcal/mol). Among those switches, we chose the 3 switches with the lowest switch MFE and the highest ΔMFE.

Figure 2: ΔG_RBS-linker of switch candidates generated from an example RNA sequence input by our software

Thermodynamic parameters of chosen influenza switches

Switch Candidate	ΔGRBS-linker	MFE Switch	ΔMFE
H5-1	-3.8	-19.1	+32.1
H5-2	-3.8	-21.1	+34.8
H5-3	-3.8	-24.0	+37.9
H7-1	-3.8	-24.5	+41.4
H7-2	-3.8	-17.8	+34.2
H7-3	-3.8	-16.3	+26.3
N1-1	-3.8	-17.1	+34.4
N1-2	-3.8	-17.1	+24.0
N1-3	-3.8	-19.4	+25.9
N9-1	-3.8	-22.2	+34.5
N9-2	-3.8	-17.9	+30.8
N9-3	-3.8	-21.6	+27.0
PB2-1	-3.8	-12.5	+34.6
PB2-2	-3.8	-24.7	+38.0
PB2-3	-3.8	-16.0	+28.2

To improve Chang Gung University(CGU)’s oral cancer toehold switch, we employed the screening function of our program. A toehold switch was chosen that is predicted to outperform the CGU’s switch according to the MFE RBS-Linker.

Thermodynamic parameters of original and improved switches from CGU

Switch Candidate	ΔGRBS-linker	MFE Switch	ΔMFE
CGU’s SAT switch	-9.2	-17.4	+46.8
New SAT switch	-4.1	-24.5	+23.1

Suboptimal Structure Modelling

Assumption: Accessibility of toehold domain correlates with the performance of switch

● The toehold domain (first 15 nts) of the switch RNA is crucial for the binding of the trigger RNA to the switch RNA to initiate the switch-unwinding process:

Toehold domain must have minimal paired bases in the switch RNA to ensure the successful binding of this domain with the complementary sequence in the trigger RNA, which allows the unwinding of the switch RNA and permits translation of the reporter protein, RFP, to occur.

Our program can only calculate the structure with the minimal free energy (MFE) for each target RNA region to reduce calculation workload. In reality, different conformations of RNAs with the same sequence co-exist in solution, and the concentrations of those populations are determined by their structures and free energy. Therefore, we manually checked the predicted structures and equilibrium concentrations of the ten suboptimal structures of each influenza switches with the lowest MFEs on the web tool developed by ViennaRNA package [2]. Then we predicted the performance of each influenza switches and compared with the experimental results.

The table below shows the different suboptimal structures of each switch RNA sequence:

Thermodynamic parameters of suboptimal structures of chosen influenza switches

Switch Candidate	Suboptimal structures	Expected fluorescence	Experimental fluorescence(relative)	Conclusion
H5-1	1, 6, 7, 2	Low	11.676366 (Low)	True Negative
H5-2	1, 2, 3, 4, 5	Low	38.24207 (Low)	True Negative
H5-3	1, 2, 4, 3	Low	13.34770467 (low)	True Negative
H7-1	1, 2, 3, 4	Low	10.82038167 (Low)	True Negative
H7-2	1, 3, 2, 7	Low	8.577606333 (Low)	True Negative
H7-3	1, 2	High	637.066 (High)	True Positive
N1-1	1, 2	High	204.7757333 (High)	True Positive
N1-2	1, 2	Low	9.325934 (Low)	True Negative
N1-3	1, 2	High	9.148810667 (Low)	False Positive
N9-1	1, 2, 3	High	166.49639 (High)	True Positive
N9-2	1, 2, 6	High	604.2225333 (High)	True Positive
N9-3	1	High	10.587811 (Low)	False Positive
PB2-1	1, 2	High	8.756591333 (Low)	False Positive
PB2-2	1, 3	Low	24.68932333 (Low)	True Negative
PB2-3	1, 2, 3, 5	High	368.0475667 (High)	True Positive

In the table, each suboptimal structure is labelled by numbers 1 to 10, where structure 1 has the lowest free energy (the MFE structure), and structure 10 has the highest free energy. In the "Suboptimal structures" column, only suboptimal structures with non-negligible equilibrium concentrations are considered and shown. The equilibrium concentrations were calculated by the ViennaRNA web server [2].
For example, for H5-3 switch:
We checked the status of the toehold domain for each suboptimal structure from the ViennaRNA web server output [2] to see if it was open or closed. In the table, the structures with an open toehold domain are shown in

green

, and ones with a closed toehold domain are shown in

red

.
For example, for H5-3 switch:

The expected fluorescence(determined by the ratio of suboptimal structures with open/closed toehold domains) exhibited by E. coli co-transformed with the switch and trigger RNA after 12 hours was compared with the actual observed fluorescence during the experiment.

From the table above, we observed that only 3 out of 15 of our predictions were incorrect, showing promising accuracy of this prediction method. None of the 3 incorrect predictions were false negatives, indicating that the suboptimal structure prediction method should be mostly applied to filtering out switch sequences with low predicted performances.

Comparing thermodynamic parameters with experimental result

Correlation of switch MFE (Minimum Free Energy) and expression leakage

The plot shows some kind of weak positive correlation between expression leakage and the MFE of a switch, with an outlier N9-2. This suggests that the MFE of a switch might not have a large contribution to the expression leakage.

Correlation of switch MFE (Minimum Free Energy) and the expression by Switch-Trigger duplex

Although the switch MFE correlates with the expression leakage, this plot showed that the switches with lower MFE also have lower RFP signal in the presence of trigger. This suggests that a low MFE of the switch could also be a hindrance to detection that it increases the energy input to the system (activation energy), so it takes more energy for switch-trigger dimer formation to occur. This could lower the expression level of the switch-trigger dimer.

Correlation of MFE difference and the expression by Switch-Trigger duplex

We initially thought that the larger the MFE difference between the sum of the MFE of switch and trigger and that of the switch-trigger duplex, the more favourable the duplex structure will be. However, there seems to be a weak positive correlation between the expression level of the switch+trigger dimer and the MFE difference of a switch. This could be because that a low MFE difference could indicate a more similar conformation of the switch/trigger RNA to the switch+trigger dimer RNA. This could lead to less conformational changes in the dimer formation process, making the process faster. Thus, a small MFE difference could lead to an increased expression level of the switch+trigger dimer.

Correlation of toehold domain base pair and the expression by Switch-Trigger duplex

The plot shows some kind of negative correlation between RFP signal in the presence of trigger and the number of base pairs in the toehold domain. Some exceptions were observed, they are N1-3, N9-3, N9-1. H5-3 and N1-2. This supports the findings from the suboptimal structure modelling.



References:
[1] Green AA, Silver PA, Collins JJ, Yin P. Toehold switches: De-novo-designed regulators of gene expression. Cell. 2014;159(4):925–39.
[2] Hofacker IL. RNA secondary structure analysis using the Vienna RNA package. Curr Protoc Bioinformatics [Internet]. 2009;Chapter 12:Unit12.2. Available from:  http://www.ncbi.nlm.nih.gov/pubmed/19496057
[3]J. N. Zadeh, C. D. Steenberg, J. S. Bois, B. R. Wolfe, M. B. Pierce, A. R. Khan, R. M. Dirks, N. A. Pierce. NUPACK: analysis and design of nucleic acid systems. J Compute Chem, 32:170–173, 2011.