Finding the combination of miRNAs specific for RRMS

With this project we found an miRNA combination specific for RRMS. By building a model for the classification of expression data for RRMS patients and healthy controls, we found an miRNA combination that was able to distinguish between RRMS patients and healthy control. During interviews conducted in human practices, we found that it is quite important to compare not only RRMS miRNA expression data to healthy controls, but also to include so-called mimic diseases. Using the Human miRNA Disease Database (HMDD), we made a list with all diseases where the levels of each of the found miRNAs are dysregulated.¹ By analysing this list we found that the combination of the found miRNA was unable to distinguish between RRMS and mimic disease diabetes. We hypothesised that by finding a miRNA dysregulated only in diabetes and including a NOT gate in our diagnostic test platform, we could distinguish between RRMS and diabetes. This means that our diagnostic test platform will not give an output if this diabetes-specific miRNA is present.

With this approach we found an miRNA combination that allows to distinguish between RRMS, mimic diseases, and healthy controls. The number of miRNAs found in this project will be used as input for our model designing toehold circuits. In our final diagnostic test platform, the found miRNA will first be amplified by NASBA.

Implementing a threshold

With our modelling strategy, we managed to further optimise the existing toehold mediated strand displacement (TMSD) reaction from Qian & Winfree (2011) to reduce leaky background expression in the system’s dose-response curves.² Next, these optimised rates were used to predict experiments for a simpler TMSD, which was tested in our lab . Through the simulations we were able to find the best ratio of concentrations of the system input components to achieve a sharp threshold switch around a desired input concentration. Furthermore, a range of permissible parameters (e.g. binding affinities) were found for the simpler system. The nucleotides in a sequence influences the parameters of the system and thus knowing the permissible parameters could help with future sequence design in return. In comparing simulations of the more complex system from Qian & Winfree (2011) with the system engineered in our lab, we found that extra system components advantageously speed up the production of the system output and create a steeper slope in our dose-response curves. However, it comes at the cost of leaky background expression at low input doses.

Designing a toehold switch

For the proof of concept of the test, we wanted to see whether toehold switches were sensitive enough to detect the miRNA directly. To design the toehold switches we used an improved version of the SwitchMi Designer made by Uparis-BME iGEM team in 2021.³ Using NUPACK, we found that most SwitchMi Designer predicted toehold switches could not fully bind the miRNA.⁴ By altering the code, we obtained a tool which designs toehold switches with tighter binding to the miRNAs compared to the original software without taking away the key feature of the SwitchMi Designer: to find a hairpin base with two weak (A-T) and one strong (C-G) base pairs. We also replaced the conserved sequences of the toehold switches to match the shorter-(mi)RNA-detecting toehold switches designed by Pardee et al. (2016).⁵

Using this tool, toehold switches were designed for hsa-miR-484 that was found to be upregulated in RRMS by Regev et al. (2018).⁶ Using the miRNA sequence in the tool, we predicted three possible toehold switch sequences. Two of these switches were made and tested in the lab.

Designing a ribocomputing circuit

To apply toehold switch logic gates in our test, we need to find a circuit that is most suitable to diagnose MS. When all the MS correlated miRNA are essential to correctly diagnose MS, a circuit with many AND gates will be most optimal. Alternatively, when only some of the miRNAs are needed for an accurate diagnosis, OR gates work best. The circuit optimisation algorithm is able to apply the correct type of gate in both these situations. We used the algorithm to design a logic gate circuit with the miRNA found to be specific for MS.

Furthermore, a continuous ODE model was created representing the toehold switch and AND gate created by our team. This model is able to simulate the production of chlorophenol red over time. Data from wet-lab experiments were fitted against the ODE model to find accurate parameters for predicting chlorophenol red production over time. The model gave us insight in the way leakiness has an influence on our experiments. Furthermore, the fitted model can be used to optimise the system even further by changing parameters and predicting the effect these changes have, without performing wet lab experiments. Using this model we can study the dyanmics of the circuit in detail and learn everything about the behaviour of the toehold switch logic gate system.