Engineering Success

Introduction

Our work in the project was conducted in a cycle of four parts: Design, Build, Test and Learn.

Design- Generating Toehold switch sequences using our Model.

Build - Inserting the generated sequence to GFP expressing plasmid.

Test - Determining the toehold switch effectiveness in vitro.

Learn - implementing new insights into the Models and the cycle repeats.

The results obtained in the Test part led us to new conclusions and the needed improvements implemented in the Model. Then we could repeat design with the improved Model, Build, Test and learn again.

In general, we can divide our work into 4 experiments, all of which undergo at least one engineering cycle. The first was reproducing a former study in eukaryotes with miRNA as the trigger. This experiment, being the first one, underwent the largest number of engineering cycles, as we tried to calibrate our systems. Eventually, the lab results led us to critical conclusions. Therefore, we changed the Model and added new features. For example, we added the prevention of stop codons and limited the sequence to single start codons which have to be in the reading frame.

The second experiment included an improved sequence, still based on the former study but with our optimizations by our Model. The results of the second experiment showed that we succeeded in improving the performance of the toehold and therefore was our first engineering success. It also led us to realize that miRNA might be too short for eukaryotic switches, which led us to designing the next experiment for mRNA triggers. The difficulties and length of experiments with mammalian cells also led us to design an experiment in yeast, which will be discussed later.

The third experiment was designing sequences from scratch using mRNA as a trigger. The lab results showed that we are getting high ON concentration (protein abundance in the presence of the mRNA) but also high OFF concentration (protein abundance in the absence of the mRNA). Therefore, we decided to design different structures, for example one that contains two hairpins instead of one in order to lower the OFF ratio. In parallel, as mentioned earlier, we learned that working with mammalian cells can be time consuming and cause many difficulties, so we designed an experiment in yeast. In this experiment we decided to check a few different conditions for validation of our models parameters.

---

First Experiment – Reproducing the Original Paper’s Results

In the beginning of our project, we learned about toehold switches and their implementation, specifically looking for implementation in eukaryotes and mammalian cells. While reviewing the literature, we came across a paper about synthetic toehold switch for microRNA detection in mammalian cells [1]. We contacted the author for consultation and even received the original plasmids (for detecting miR-155 or miR-21) used in the paper for our preliminary testing.

This experiment was the one that underwent the most engineering cycles, as we tried to calibrate our system, those cycles are summarized below

---

Engineering Cycle - 1

Design:

We designed an experiment for reproducing the results of this paper – detection of microRNA (miR) by a toehold switch fused to a reporter gene (GFP). Although the original work used miR mimics, our budget limitations led us to use plasmids for miR expression.

We used HEK293 cells to examine sequences’ specific expression in-vivo. For the negative control, we used untreated cells; for the OFF-state we transfected cells with the toehold-GFP plasmid but not with the trigger plasmid; for the ON-state we transfected the cell with both toehold-reporter and trigger plasmids. Cells are analyzed for GFP expression through FACS. All sequences are validated by Sanger sequencing.

Build:

With the original plasmids in our hands, we still had to build plasmids for our own toehold sequences and for the miR-155 and miR-21 expression plasmids. We decided to choose a backbone plasmid and clone our toehold sequences into it, a decision proved to be a major cost saver. However, for the original miR-155 toehold test we chose to use commercial plasmids for expression of miR-155 and miR-21.

Test:

We tested the miR-155 toehold in a wet lab experiment as designed in the Design stage. Initial results showed no expression levels in both ON and OFF conditions, as can be seen in figure 1.

Learn (and design again):

Since we couldn’t reproduce the results, we looked for flaws in our design. For the next round of experiment we decided to add a positive control condition (GFP expressing plasmid) and change the used concentration of plasmids.

---

Engineering Cycle - 2

The second cycle of the first experiment differs from the first one mainly in adding a positive control condition and calibration of concentration. The build and test stages are similar to the ones in the first cycle.

Learn (and design again):

Positive control shows that the transfection itself is working and that the issue isn’t lying in the protocol. Moreover, changing the ratio between toehold plasmids and trigger plasmids didn’t seem to improve the situation. All of that led us to speculate that there was something wrong with our glycerol stock, so we used a fresh new one in the next cycle. We also modified our positive control to have an open-structure segment in the length of the toeholds upstream to the GFP sequence (“open toehold-GFP”). In addition to HEK293 cells, we decided to check our system in HEPA cells, in case our HEK293 cells are damaged. We also decided to generate a new cloning backbone from the original plasmid used in the paper.

---

Engineering Cycle - 3

Build:

We wanted to use open toehold-GFP as our positive control, so we designed one with our algorithm and cloned it to our backbone plasmid. This sequence was designed to have an unfolded segment in a length of a toehold upstream to the GFP.

Test:

We tested our system in both HEK293 and HEPA1-6 cells, final parameters can be seen in figure 3.

This time, as can be seen in figure 4, we observe GFP expression in the positive control, OFF and ON conditions, where expression levels in the ON condition are lower than in the OFF condition.

Learn:

After receiving these results, we performed a few repeats which confirmed them. This led us to take a closer look at the plasmids. When analyzing the sequences (from the sequence reported in the article, which, as we validated, was similar to the one we used), we noticed the presence of a stop codon between the Kozak sequence and the GFP sequence and that there might be a reading frame discrepancy. These observations led us to implement algorithms for prevention of such phenomena into our model. We then proceeded to design our next experiment, based on our own, model based, generated toehold switches.

---

Second Experiment – Testing Our Model-Generated Toehold Switches

Our second experiment surrounded our own sequences. While it could also be divided into more than one cycle, it is summarized as one cycle with ongoing learning and adjustments.

In addition, before beginning this experiment, we reviewed the literature and learned that the localization of the ribosome binding site (Kozak sequence) in toehold switches can affect switch quality, so we decided to examine different localization conditions. Results for these experiments are better described in the Results section.

Design:

Sequences for our second experiment were designed by our software (see Model page). In this experiment, we wanted to check several variants produced by different conditions of our model, with variations in Kozak localization, linker sequence and more. Since the only varying part is the 5’ UTR of the mRNA, we reduced the cost of this experiment by ordering and cloning the 5’ UTR variations into our GFP-containing backbone plasmid. Cloning was validated by sequencing and PCR. The system we used is similar to the one used in the first experiment. When results weren’t as expected, we learned and modified our experimental procedure accordingly. For example, when receiving low yield of plasmids, we amplified them using PCR.

Build:

We ordered synthesized sequences from iGEM’s sponsors and cloned them into our GFP-containing backbone plasmid

Test:

Similar to the first experiment, we tested our sequences in HEK293 cells. This experiment included several repeats, with some calibration between each one. As low fold-changes were detected between the OFF and ON states, we tried to add controls to our experiments such as transfecting the cells with a non-interacting miRNA-21 plasmid instead of the trigger.

Learn:

While performing repeats for this experiment, we learned that plasmid load might affect our ON condition, since we are transfecting the cell with ~double the amount of plasmids than in the OFF condition. That caused us to try different normalization parameters. Moreover, we learned that OFF state expression is not low enough, so for our next mammalian experiment we decided to strengthen our switch by either extending the stem of our toehold switch or concatenating two hairpins in the 5’ UTR region. For the purpose of a longer stem, we realized that miR are not long enough for such designs, and since our end goal is to trigger translation by an endogenous trigger, we decided to take the leap into using mRNA as a trigger. In addition, we decided to design an experiment for yeast to minimize our expenses and cycle duration.

---

Third Experiment – Longer Stems and Two Concatenated Hairpins

Design:

As we learned from our second experiment, OFF state expression was higher than wanted. As a result, sequences for the third experiment were designed to strongly block expression in the OFF state, either by longer stems or by concatenating two hairpins. The need for a longer stem, as well as the goal of triggering expression by endogenous mRNAs, led us to design switches that are opened by mRNAs. Alongside using mRNA windows from the literature, we implemented an algorithm for finding unfolded areas that can be used as triggers. For this experiment we use mCherry’s mRNA as our trigger molecule and GFP as our reporter gene.

Build:

The ordered DNA sequences were cloned into our backbone plasmid similarly to the previous experiment.

Test:

As for the writing of these words, this experiment is still underway and we are pending results.

---

Yeast Experiment – Engineering Cycle

After learning that working with mammalian cells can be time consuming, we decided to design an experiment for yeast. In this experiment, similar to our third experiment in mammalian cells, we used mCherry mRNA as the trigger molecule and GFP as the reporter gene.

Design:

After making some yeast-suited adjustments to our model, we used it to generate several toehold switches for different conditions. For each condition, we checked the best ranked sequences and another, worse ranked one, for validating our model’s ranking metrics. We also decided to compare between two mCherry molecules, unoptimized and yeast-optimized. In addition, when designing the switches we examined their biophysical and thermodynamical properties. We also implemented an algorithm for preventing hybridization between the start of the reporter gene to the trigger binding domain. Cloning was validated by plate reader and/or PCR, as well as using selection by medium.

Build:

We used gap repair for cloning our segments into the plasmids and certain amino-acid poor medium for selection. First, we transformed two yeast strains, each with a different mCherry containing plasmid to be used as a trigger. Then, we transformed our toeholds into trigger-negative and trigger-positive yeast strains.

Test:

Trigger-positive yeast were tested using a plate-reader, for mCherry wavelengths to validate mCherry’s transformation. Then, we used a plate-reader to test for GFP wavelengths for both the OFF and ON conditions, as well with positive and negative controls.

Learn:

From this experiment, we learned that our model’s metrics are in line with wet lab results, as for two different conditions, the two highest ranked switches showed the best performances detected, similar to the positive GFP control. For the design of our next experiment, we plan on testing a few variations of each condition for further validation of our metrics.

---

Model Engineering Cycles

Throughout the competition, we learned from, designed, built, and tested our models several times, most of which had implications and called for adjustments. Instead of segmenting this into a few engineering cycles, we describe our model evolution in sections of eukaryotes and prokaryotes.

Eukaryotes:

The first design was basic, built upon the NUPACK package [2-5]. After learning from the results of the first experiment, we returned to designing an improved version that included several new features, such as reading frame check, pseudoknots prevention, kozak localization and more. After testing our improved model in the second experiment, we learned that we need to implement algorithms for finding differentiating mRNA molecules and finding unfolded segments within molecules. We also learned that we need to allow for a more flexible design in regard to toehold sizes. As a result of our human practices, we also implemented algorithms for translation optimization and prevention of immune response triggering.

Prokaryotes:

Our partnership with Aboa’s iGEM team led us to implement toehold design models for prokaryotes. The initial model was based on work in prokaryotes and supported one structure. After designing the first batch of sequences, we received simulation results from Aboa, which led to a joined brainstorming session. In this session we learned about a metric for evaluation of prokaryotic switches and the existence of another experimentally tested eukaryotic toehold structure. Following the new learned information, we returned to the design phase and implemented both the assessment metric and support for the second toehold structure. The second batch of generated switches resulted in a better simulated results, and we are currently waiting for Aboa’s wet lab results for further validation.

---

References

1. 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.
2. Wolfe, B. R., Porubsky, N. J., Zadeh, J. N., Dirks, R. M., & Pierce, N. A. (2017). Constrained multistate sequence design for nucleic acid reaction pathway engineering. Journal of the American Chemical Society, 139(8), 3134-3144
3. Wolfe, B. R., & Pierce, N. A. (2015). Sequence design for a test tube of interacting nucleic acid strands. ACS Synthetic Biology, 4(10), 1086-1100
4. Zadeh, J. N., Wolfe, B. R., & Pierce, N. A. (2011). Nucleic acid sequence design via efficient ensemble defect optimization. Journal of computational chemistry, 32(3), 439-452.
5. Dirks, R. M., Lin, M., Winfree, E., & Pierce, N. A. (2004). Paradigms for computational nucleic acid design. Nucleic acids research, 32(4), 1392-1403.