We sought to create a cell-free diagnostic kit for the detection of plant pathogens. Our kit uses toehold-switch based sensing of the pathogen's genome to facilitate the detection. The translation of a reporter protein is activated only in the presence of a conserved sequence in a pathogen’s genome.

Our system consists of two major components: a cell-free expression system and a sensor plasmid, which contains the toehold switch regulating the translation of the reporter protein. In our wet lab efforts, we successfully produced a viable cell-free expression system and assembled the sensor plasmids. We successfully combined the functionalities of these major components, although further testing and optimization would be needed. We also showcased our system capable of toehold switch-based sensing and managed to prove the functionality of a novel sensor plasmid. We produced a promising system, which could be transformed into a functional detection kit with further optimization.

Design of novel toehold switches

First we started by aligning ten barley yellow dwarf virus whole genome entries from GenBank. These aligned genomes were screened for at least 36-nucleotide long conserved sequences. For each of them, we designed multiple A- and B-series toehold switches (Pardee et al, 2016), which are modified from original toehold switches to reduce translational leakage. Toehold switches are RNA sequences in the 5’-untranslated region of mRNA, which only allow the translation of a downstream protein coding sequence in the presence of their specific trigger. The B-series switches exhibit lower background signal, ideal for diagnostic kits, while A-series switches produce higher output signal, ideal for colorimetric assays. We wanted to test these different design principles to see which would better suit the needs of our kit.

Any toehold switches containing in-frame stop codons were disregarded from further consideration. We used a model adapted from Ma et al. (2018) to estimate the performance of the designed toehold switches (see our Modeling page). Three of the highest scoring sensors from each design series were taken into laboratory testing. We also collaborated with team TAU and their project TrigGate, who provided us alternative designs for each of our best toeholds. We ranked them with our model, and picked the highest scoring A-series and B-series toeholds for lab tests.

The toehold switch scores predicted by our model are lower than anticipated, as we did not yet have an optimized design algorithm when creating these. This issue was revised in later iterations of the design algorithm (see Model and Partnership). The theoretical maximum score is 54, yet this is almost impossible to reach in practise. In our experience based on evaluating viable toehold switches from literature, a score over 25 is to be pursued, but the scores rarely exceed 32. The modeling results for each toehold switch selected for laboratory testing is presented in Table 1.

Table 1. Results of computational modeling of toehold switches. We assigned a score for each toehold switch with our model, based on their deviations from ideal structures (see Model). Higher scores represent higher estimated performances.
Toehold identifier Score Toehold identifier Score
ABOA A70 16.41 ABOA B39 13.19
ABOA A92 15.40 ABOA B46 14.08
ABOA A95 15.39 ABOA B69 11.99
TrigGate A70 6.20 TrigGate B46 24.07

Successful Golden Gate assembly

Each sensor plasmid consists of the plasmid backbone, a promoter, a toehold switch, a gene encoding a reporter protein, and a terminator. pOdd1 was used as the plasmid backbone. We used the iGEM Type IIS standard fusion sites in our assembly, with slight modifications. The fusion site between the reporter coding sequence (cds) and the terminator was changed from GCTT to GGTT, as the standard junction site shared three consecutive nucleotides with the CGCT junction site between the terminator and the backbone. This way, the probability of incorrect assembly was lowered. Also, the toehold switch and reporter protein are joined with scarless assembly, but the junction site corresponds to the Type IIS standards. The promoter, the toehold switches, and the terminator were synthesized with flanking sequences producing these junction sites. The reporter genes were amplified from plasmids with PCR using primers with their corresponding flanking sites, so the amplified product also has these sites. An illustration of the complete sensor plasmid can be seen in Figure 1.

A picture showing the structure of the toehold switch plasmids used in our project. The sites for promoter, toehold switch, reporter gene, terminator and antibiotic resistance can be seen from the picture.

Figure 1.
Map of the sensor plasmids assembled with golden gate assembly. The kanamycin resistance was provided in the backbone, while other parts were added in the assembly. The modularity of our system is based on the exchangeable toehold switch in the sensor plasmid.

We assembled 14 sensor plasmids in total. A sensor plasmid with ꞵ-galactosidase as the reporter for each toehold and sensor plasmids containing our toehold switches and mScarlet-I as the reporter were assembled. In the assembly, an RFP gene is cleaved from the plasmid backbone and the remaining junction sites of the plasmid are not compatible so, in theory, a plasmid without inserts or the RFP gene cannot transform bacteria. This makes it possible to visualize the successfully transformed colonies as white, as they do not possess the RFP gene. To confirm the success of the assembly, we tested the presence of the insert in the white colonies with colony PCR. We used primers that attached to the linker sequence of the toehold sensor, as well as the terminator sequence. A representative result of the colony PCR reactions is shown in Figure 2.

A picture of agarose gel electrophoresis for Golden Gate assembled sensor plasmids with m-Scarlet-I as the reporter. The gel contains six wells. All six sample wells show bright bands around 700-800 base pairs. No other bands are shown.

Figure 2.
Results of colony PCR reactions. Lane 1 was loaded with Quick-Load® Purple 1 kb Plus DNA Ladder (NEB, #N0550) and the molecular weight standards are marked. Lanes 2-7 contain colony PCR reactions performed for our novel toehold sensor plasmid with mScarlet-I as the reporter. The electrophoresis was performed in a 1 % agarose TAE gel with 100 V for 40 minutes.

In Figure 2 we can see that each sample in lanes 2-7 contains a PCR product with lengths between 700 and 800 bp. We expected PCR products of 762 bp. We can conclude that the assembly was successful, as we were able to amplify a sequence containing parts of three of the four inserts. All other constructs were confirmed in a similar fashion before being used in cell-free reactions.

Validating the measurement technique

The activity of the produced β-galactosidase is assessed by measuring the absorbance at 420 nm, resulting from the breakdown of ONPG (2-nitrophenyl β-D-galactopyranoside) to 2-nitrophenyl and D-galactose. We planned to test the viability of our cell-free expression reactions by measuring the activity of β-galactosidase in vivo. We used an inducible pET-15b-derived expression plasmid containing the lacZ gene and induced its expression in BL21 (DE3) cell culture. ONPG was added to this cell culture and the activity of β-galactosidase was monitored by measuring absorbance at 420 nm. Absorbance was shown to increase over time, resulting from the breakdown of ONPG. This result validates our measurement technique for subsequent reactions.The results are presented in Figure 3.

A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) in BL21 (DE3). The absorbance values start from 0 at time point zero, and rise up to a little below 0,20. Error bars are also shown.

Figure 3.
In vivo activity of β-galactosidase. The activity of β-galactosidase in BL21 (DE3) cell cultures was monitored by measuring absorbance at 420 nm, resulting from the breakdown of ONPG. The graph shows the increase of absorbance as the function of time. Each data point represents the average of four parallel reactions while error bars represent standard deviation.

Demonstration of CFE viability

Cell-free expression (CFE) systems consist of all the necessary components needed for protein synthesis, but contain no live cells. There are three major components in a CFE system: protein components of transcription and translation, smaller functional molecules, such as NTPs and amino acids, and reaction buffers designed to maximize protein yield. To drive down the cost of making a CFE system, we opted to obtain the protein components from crude extracts. On top of this, we decided to use maltodextrin instead of using the traditional energy source 3-phosphoglyceric acid, as it has previously been proven to produce similar protein yields (Arce et al, 2021). In many applications of cell-free systems, the cells are lysed by bead-beating. The bead-beating tubes are single-use, which increases the cost of this method. Therefore, we tested two different cheaper methods for the cell lysis, sonication and autolysis. Sonication is a standard procedure in many protein extraction methods and the machine can be used multiple times, driving down the per-use cost (see Design).

In autolysis, the cells contain a plasmid that produces phage lambda endolysin. When exposed to freeze-thaw cycles, the inner membrane is disrupted and the endolysin gains access to the bacterial cell wall and degrades it, lysing the cell (Didovyk et al, 2017) (see Design).

We produced cell lysates by sonication (sCFE) and autolysis (aCFE) and compared their protein production capabilities in CFE systems. The functionality of the SCFE system was measured by its ability to produce β-galactosidase from the same expression plasmid as previously. The measurement setup was validated as before.

As we also wanted to test out mScarlet-I as the reporter in our toehold switches, we tested its expression from an expression plasmid in both lysates with otherwise identical CFE systems. Due to technical limitations, we measured the time required for each reaction to surpass the upper detection limit of our plate reader (10 000 A.U. over baseline). The sCFE systems required 41 ± 5,7 minutes, while the aCFE required 90 ± 48 minutes. The results from all of these measurements are presented in Figure 4.

A. A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) for autolysis (aCFE) and sonication (sCFE) samples. The absorbance values start from 0 at time point zero, aCFE rising to around 0,10, while sCFE rises to absorbance over 0,25 at time point 40, and then decreases to around 0,15 at time point 180. The error bars for both of the measurements are quite big, especially at time point 100 where they overlap with each other. B. A graph of fluorescence intensity in cell-free reaction as a function of time. Two samples for both autolysis (aCFE) and sonication (sCFE)c are shown. The sCFE samples reach 10 000 A.U. at around 40 min, while aCFE samples reach 10 000 A.U. at around 50 min and 120 min.

Figure 4.
Comparison of different CFE systems. (A) The activity of β-galactosidase in CFE systems produced with autolysis (aCFE) or sonication (sCFE) was monitored by measuring the change of absorbance at 420 nm, resulting from the breakdown of ONPG. This graph shows the absorbance over time. Each data point represents the average of three independent reactions and error bars represent standard deviation. (B) Comparison of mScarlet-I production capabilities in CFE systems produced by autolysis and sonication. This graph shows the fluorescence intensity over baseline at 569/593 nm. The results are from two parallel measurements for both CFE systems.

In both measurements, sCFE exhibited more rapid signal production. These results suggest that sonication is a superior method for producing CFE systems, although repeating these tests is required to confirm this conclusion. However, absorbance in sCFE decreases after reaching a peak value, which could result from degradation of the colorimetric product, 2-nitrophenyl in the reaction. This would require further optimization of the colorimetric CFE reaction. Based on these measurements, we opted to use sCFE systems in all our subsequent reactions.

Confirming toehold functionality in sCFE system

Next, we proceeded to test the sensors’ detection capabilities in the sCFE system. We used the 27B sensor from Pardee et al. (2016) (Addgene #75006) as our control to see if a proven toehold sensor could be activated in our system. The sensor produces β-galactosidase when activated, so we assessed its activity by measuring the absorbance at 420 nm as previously. We tested the activity in the presence of its specific ssDNA trigger. As we observed the toehold to be functional in our sCFE system, we proceeded to analyze the functionality of our novel toehold sensor plasmids, with β-galactosidase as the reporter, and compare their performance to the control teohold sensor. The results are shown in Figure 5.

A. A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) in sample 27B with and without trigger. The absorbance values start from 0 at time point zero, sample 27B with trigger rising up to around 0.12 and sample 27B without trigger rising up to around 0,05 at time point 180. Differences between the samples start to accumulate around time point 30. The error bars for 27B with trigger are big, especially in time point 180 where the error bars go from 0.05 up to 0.20. B. A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) in sample A70 with and without trigger. The absorbance values start from 0 at time point zero, sample A70 without trigger rising to around 0.04, sample A70 without trigger rising up to around 0.08 at time point 180. The samples start to deviate from each other at around 30 minutes. Both of the samples have some error margin, the error bars of sample A70 with trigger almost reaching the error bars of A70 without sample. C. A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) in sample B46 with and without trigger. The absorbance values start from 0 at time point zero, sample B46 without trigger rising up to around 0.07 while B46 with trigger rises up to around 0.04 at time point 180. The samples start to deviate from each other at around time point 100. The error bars for B46 without trigger are notably large, B46 with trigger also shows some error. D. A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) in sample TG70 with and without trigger. The absorbance values start from 0 at time point zero, TG70 without trigger rising up to around 0.11 and TG70 rising up to 0.02. The samples deviate from each other especially at time point 110, where the absorbance values of TG70 with trigger drop drastically. The error bars for TG70 with trigger are large. E. A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) in sample TG46 with and without trigger. The absorbance values start from 0 at time point zero, TG46 without trigger rising up to around 0.12 and TG46 with trigger rising up to around 0.18. No error bars are shown for either of the samples. F. A bar chart showing ON/OFF-ratios of A70, B46, TG70, TG46 and 27B. A70 and B46 show a ratio around 0.5. TG70 shows a ratio of below 0.25. TG46 shows a ratio of nearly 1.5, while 27B shows a ratio of nearly 2.5.

Figure 5.
Toehold switch performance in sCFE system. sCFE reactions containing toehold sensor plasmids with β-galactosidase as the reporter protein were monitored by measuring the absorbance at 420 nm. Cell-free reactions containing (A) the control toehold sensor plasmid and (B-E) our novel toehold sensor plasmids were measured in the presence or absence of their specific trigger ssDNA. Graphs A-E show the absorbance over time in the reactions. Each data point represents the average of two parallel reactions and error bars represent standard deviation. Graph E shows only the average of two independent reactions as representative results. Graph F shows the ON/OFF ratios for each toehold switch based on these reactions, calculated as the ratio of end point signal of reactions with and without trigger. The average of the two parallel reactions were used in the calculation.

The established 27B toehold sensor plasmid showed clear distinction in absorbance produced in the presence and absence of its specific ssDNA trigger. The reactions containing the trigger produced a 2.46-fold end point signal compared to the reactions where the trigger was left out (ON/OFF-ratio). This result validates the sCFE system for being capable of toehold switch-based sensing. However, the ON/OFF-ratio measured in these tests was lower than in published CFE systems (Arce et al, 2021; Pardee et al, 2016), which suggests that our sCFE system is not yet optimized for this purpose.

Our novel toehold sensor plasmids did not match the performance of the control toehold switch in our measurements. Toehold switches A70, B46 and TG70 were not observed to exhibit trigger-dependent signal production, as their signal production was lower in the presence of the trigger; with ON/OFF-ratios of 0.56, 0.56 and 0.17, respectively. Our toehold sensor B46 was seen to exhibit trigger-specific signal production, as it had an ON/OFF-ratio of 1.45 in our tests. These results are still limited and further tests would be needed to confirm functionality. On top of this, the remaining eight of the assembled toehold sensor plasmids would require testing to draw further conclusions of the functionality of our system.

As we gained experimental data about the functionalities of the toehold switches, we assessed the performance prediction capability of our model. We plotted the ON/OFF-ratios of our toehold switches as the function of their scores assigned by our model. A clear correlation between the predicted and observed performance can be seen, which suggests our model to be viable in predicting the best performing toehold switches in identical conditions. However, we do not have enough data to credibly draw any conclusions. The results are presented in Figure 6.

A graph showing the ON/OFF-ratio as a function of score. Four points are presented in the graph, and a linear fit line is drawn between the points. The second point is exactly on the line, while other points deviate a bit.

Figure 6.
Correlation of predicted and measured performance of toehold switches. The graph shows the ON/OFF-ratio, measured from two independent reactions with and without the trigger, over the score predicted by our model.

As we had demonstrated the viability of our sCFE system for toehold-based sensing using ssDNA as the trigger, we wanted to test the system for RNA sensing. Our goal was to enable sensing of NASBA amplicons, which would be produced from the sample to be analyzed (see Proposed Implementation). To test this, we sought to analyze the control toehold sensor’s ability to sense RNA sequence designed to mimic NASBA amplicons. However, the control toehold switch was not observed to exhibit RNA trigger-dependent signal production. The result of this measurement is shown in Figure 7.

A graph of absorbance in 420 nm measures in cell-free reaction as a function of time (min) in sample 27B with and without sample. The absorbance values start from 0 at time point zero, and go up to approximately 0.15 in both samples. The sample 27B with trigger, seems to have a bit higher absorbance especially from 100 minutes to 180 minutes, where the measurement stops. The error bars for both samples at this time are very high.

Figure 7.
Assessment of RNA-sensing in sCFE. The production of β-galactosidase is measured by the increase of absorbance at 420 nm, resulting from breakdown of ONPG. The absorbance is measured from reactions in the presence and absence of RNA trigger. Absorbance is presented as the function of time, with each data point representing the average of two independent reactions and error bars representing standard deviation.

These results suggest that our sCFE system inhibits the sensing of RNA sequences. However, additional testing would be required, as no concrete conclusions can be drawn from the result. Here we have only measured two reactions, which exhibited high variability, seen as the high values of standard deviations.


We succeeded in producing a viable CFE system capable of toehold-based sensing. We managed to create functional toehold sensors and demonstrated their effectiveness in trigger-dependent translation. However, our goal of RNA trigger sensing was not showcased and based on our test reactions, the performance of our toehold sensors was not on par with the control sensor. The data set is limited and more information would need to be gathered to speculate the system’s functionality in more detail. Still, we gained valuable information and were able to pinpoint deficiencies in our system, which could be revised in ensuing revisions.

Our toehold switches were not demonstrated to exhibit high trigger-dependency in translational activity. This could stem from insufficient performance of the designed toehold switches. Our design algorithm was not optimized when creating these toehold switches and in later iterations, we were able to produce better performing toehold switches, according to our model. We also gained some experimental data about our model’s viability. With further testing of the assembled toehold sensors and testing of new toehold switches produced by updated versions of our algorithm and team TAU’s TrigGate, we would further optimize the toehold-mediated sensing. On top of the toehold switch, other parts of the sensor plasmid would need optimization as well.

The control toehold switch also displayed lower translational specificity for its trigger, which suggests that our sCFE inhibits the sensing capabilities of toehold switches. Again, further testing with this toehold sensor as well as other established sensors would be necessary for examining this suggestion. Our sCFE system was produced with crude extract from sonication, while established toehold sensing in CFE systems have been conducted in systems produced with crude extract from bead-beating (Arce et al, 2021) or with PURE (protein expression using recombinant elements) systems (Green et al, 2017; Pardee et al, 2016). On top of this, the extract-based CFE was produced with a strain deficient of multiple nucleases. As a result, the established systems inherently contain lower amounts of nucleases, which can impair the stability of the sensor DNA and RNA. The sensor DNA was linearized before input in these systems, which could also impact the transcriptional activity and therefore the amount of sensor RNA in the reaction.

Sensing of RNA triggers was not observed in our sCFE system. This was also documented in published CFE systems. This could be caused by faulty produced RNA-triggers as we were unable to confirm correct transcription products due to unavailability of necessary components of RNA analysis. Our system could also impair the stability of the RNA trigger, which would hinder signal production. However, we did not perform enough experiments for us to draw credible conclusions.

We also observed a decrease in the colorimetric signal in sCFE reactions. This was not the case in previously established CFE systems used for toehold-mediated sensing (Pardee et al, 2016; Ma et al, 2018; Arce et al, 2021). As we’d visualize our system to produce a colorimetric signal, degradation of the colorimetric product impairs the viability of the system. This could be solved by optimizing the CFE system to increase the stability of 2-nitrophenyl. This issue could also be solved by changing the substrate, for example to chlorophenol red-β-D-galactopyranoside (CPRG), which was used in previous systems. This could also produce a signal more easily detected visually, as the color change is more dramatic with CPRG (yellow to dark red) than in ONPG (colorless to yellow).

Despite these challenges, our detection system shows promise for its capability. Overcoming the issues and optimization of all parts of the system would provide a powerful detection system, which could help fight the global issues of hunger and food security.


  • Arce, A., Guzman Chavez, F., Gandini, C., Puig, J., Matute, T., Haseloff, J., Dalchau, N., Molloy, J., Pardee, K., & Federici, F. (2021). Decentralizing Cell-Free RNA Sensing With the Use of Low-Cost Cell Extracts. Frontiers in Bioengineering and Biotechnology, 9, 744. https://doi.org/10.3389/FBIOE.2021.727584/BIBTEX

  • Didovyk, A., Tonooka, T., Tsimring, L., & Hasty, J. (2017, August 21). Rapid and Scalable Preparation of Bacterial Lysates for Cell-Free Gene Expression. ACS Synthetic Biology, 6(12), 2198–2208. https://doi.org/10.1021/acssynbio.7b00253

  • Green, A. A., Kim, J., Ma, D., Silver, P. A., Collins, J. J., & Yin, P. (2017). Complex cellular logic computation using ribocomputing devices. Nature 2017 548:7665, 548(7665), 117–121. https://doi.org/10.1038/nature23271

  • 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. https://doi.org/10.1016/J.CELL.2014.10.002

  • Ma, D., Shen, L., Wu, K., Diehnelt, C. W., & Green, A. A. (2018). Low-cost detection of norovirus using paper-based cell-free systems and synbody-based viral enrichment. Synthetic Biology, 3(1). https://doi.org/10.1093/SYNBIO/YSY018

  • Pardee, K., Green, A. A., Takahashi, M. K., Braff, D., Lambert, G., Lee, J. W., Ferrante, T., Ma, D., Donghia, N., Fan, M., Daringer, N. M., Bosch, I., Dudley, D. M., O’Connor, D. H., Gehrke, L., & Collins, J. J. (2016). Rapid, Low-Cost Detection of Zika Virus Using Programmable Biomolecular Components. Cell, 165(5), 1255–1266. https://doi.org/10.1016/J.CELL.2016.04.059