Design

We imposed the following criteria on our mRNA sensor design:

• orthogonality
• easy integration with strand displacement circuits
• ability to amplify a signal
• ease of sensor generation

We decided to build a proof-of-concept mRNA sensor for eBFP2 (BBa_K779300). With our design principles, any mRNA could be sensed using an easily generated sensor construct.

The way we implemented this sensor is by considering the mRNA sequence and choosing within it two consecutive abstract domains: the toehold (T) and the sensing (S) domains. These domains mirror the toehold and recognition domains in strand displacement, so provide an interface to any strand displacement circuit. Specifically, the chosen toehold and sensing domain will then directly interface with an intermediate gate:output complex using toehold-mediated strand displacement, with signal amplification achieved using a fuel strand.

Illustration of an mRNA sensor. A toehold region (T) in the mRNA can bind to the toehold on a gate:output complex. Branch migration can then occur on the domain S, resulting in a free output signal strand. Using fuel, the mRNA can be released from the gate, enabling it to react with another gate:output complex.

However, this abstraction needs to take into account the secondary structure of mRNA: some regions are more accessible (less basepairing), while others are strongly basepaired and thus unavailable to be directly sensed by mechanisms that involve basepairing to the mRNA (Kertesz et al. 2007).

An example software rendered secondary structure of eBFP2 (BBa_K779300) showing various secondary structure elements (e.g. stems and loops)

Much work has been done to generate these secondary structures computationally (see nupack.org), and to find accessible regions predicted to be miRNA binding sites within mRNAs (e.g. PITA). We leveraged these algorithms to identify potential toehold and sensing domains within mRNAs by looking for regions with different levels of accessibility.

Once suitable domains have been chosen using modeling of the secondary structure, we can rank them by their orthogonality to other transcribed RNAs. These sequences are available in online databases (e.g. mRNA data for HEK 293 cells). For strand displacement, the rate limiting step is the binding of a strand to a gate:output complex using the toehold. After this step, the process of strand displacement is sensitive to nucleotide mismatches, with early mismatches being more disruptive than later mismatches due to the nature of branch migration (Qian et al. 2011). Thus, we can use a weighted Hamming distance between a candidate domain and every subsequence in the transcriptome. This method is similar to identifying orthogonal sequences for protein-DNA interactions as previously published (Silver et al. 2012).
Care has to be taken to make sure that sensors in a circuit are orthogonal as well, but the same method can be applied to any sequences that are introduced into a cell.

Four highlighted domains of eBFP2 with various secondary structures we chose to sense as a proof-of-concept. Blue-filled circles represent toehold nucleotides, green-filled nucleotides represent sensing domain nucleotides. The border of a nucleotide's circle represents the probability of the nucleotide being in the given state (red - very likely, blue - very unlikely). These are closeups of the mRNA structure above.

Modeling

Reaction Kinetics

Modeling of mRNA secondary structure was done using NUPACK. Strand displacement reaction kinetics were simulated using Visual GEC.

We used Visual GEC extensively to develop and test computational models of the strand displacement systems we designed, including the mRNA sensor, the NOT gate (see Processing), and a hypothetical NOT sensor which combines the functionalities of both. This enabled us to optimize the properties of our systems, and examine their kinetics, before performing in vitro experiments with actual DNA or RNA.

The syntax of Visual GEC code is straightforward: indicate when to plot and what to plot, indicate kinetic rates, and indicate allowable reactions. The software then calculates how reactions will proceed. While Visual DSD, a similar tool which is specialized for simulating DNA strands, would also have worked for our purposes, Visual GEC gave finer control of reaction rates. The kinetic constants were based on work published by Lulu Qian and Erik Winfree (Science 2011).

mRNA Sensor

The first task is to create a simulation of an mRNA sensor. The strands involved in the design are the mRNA input itself, the ROX:quencher fluorescent reporter complex, the annealed gate:output complex, and the fuel strand which accelerates the reaction rate. The visual GEC code describing this system is below.

This compiles to a visual depiction of the reaction network, and simulating a kinetic run outputs a prediction of the level of each product.

We can then specifically compare the levels of the output molecule -- which is fluorescence of the unquenched ROX-bound RNA -- and observe the difference in output level for different concentrations of input.

green curve: 9 nanomolar (i.e. 90%) input strand red curve: 1 nM (10%) input strand both run with 100nM fuel strand, 12 nM ROX:quencher reporter complex, 10 nM gate:output complex

Inverting mRNA Sensor (NOT sensor)

Next, we designed an inverting mRNA sensor, in the hope that creating an effective NOT sensor could simplify potential circuits by removing the need to cascade a NOT gate with an mRNA sensor.

The NOT sensor includes the familiar mRNA strand, fluorescent reporter complex, gate:output complex, and fuel strands. In addition, it contains a new strand called the "sensor" strand, which displaces the gate:output but is competitively bound by mRNA. Thus, the presence of an mRNA strand inhibits the sensor, preventing output signal. The absence of an mRNA strand allows the sensor strand to displace the gate:output signal, creating a fluorescent signal. This creates NOT logic. These reactions are described in the Visual GEC code.

Running this code, we can compare the output that results from various concentrations of mRNA input. The shape of this transfer function reveals the NOT logic. The NOT sensor system was designed with a steep dropoff as opposed to a steady gradient, so that the circuit as a whole would behave more reliably.

<--TODO>

In vitro tests of the functionality of the NOT sensor strands are in progress.

In vitro results

We chose 4 domains in eBFP2 with various predicted secondary structure properties to test out in vitro (see design considerations above). eBFP2 (BBa_K779300) mRNA was produced by PCRing on a T7 promoter and terminator . The resulting template was then used for in vitro transcription . After purification and quantification on a NanoDrop 2000C, the transcribed mRNA, corresponding gate:outputs and fuel strands along with a fluorescent reporter were added to wells in a 96-well plate and the fluorescence was measured on a plate reader.

For proof-of-concept studies we chose DNA as nucleic acid for the gate:output, fuel, and reporter complexes, as these results will mirror results with RNA-based strands, as shown by foundational experiments. However, the thermodynamics, kinetics and steady states will be different between DNA and RNA strands. We expect mRNA to produce less output than a corresponding DNA input mimic (comprising of just the toehold and sensing domains) in the same amount.

Results indicate that there is a difference between the 4 domains we chose, and that the output signal from mRNA is less than the signal from DNA mimics.

In vitro results from a DNA-based mRNA sensor. Graphed here is the fold increase in RFU - comparing the pre-input fluorescence to the fluorescence after (TODO) hours. Baseline (light gray) reflects no input. DNA input mimic (dark gray) is a DNA oligonucleotide with the same sequence as the toehold and sensing domains in mRNA (black).
Comparing mRNA as an input to DNA as an input. Graphed here is the ratio of fold increase fluorescence comparing mRNA inputs to DNA inputs. This indicates the possible completion level due to mRNA inputs. As described above, we expect this to be <1.