The sensing part of our project exists out of the two-component TtrS/R sensing system. To have more control over the expression of the TtrR and TtrS proteins, we wanted to make use of chemically controlled promoters. In the literature, we discovered that the pTac and the pLtetO-1 have been used by researchers . These promoters are induced by IPTG and anhydrotetracycline (tetracycline derivative), and produce TtrS and TtrR respectively. Since plasmids including these promoters have been used in previous studies and were commercially available from Addgene, these plasmids have been used in our project.
To reproduce the results from the research paper as described in Results, we had to determine the concentrations of all the inducers required for the test. Our supervisors advised us to make use of a university standard IPTG concentration of 0.1 mM and a series of tetrathionate (Ttr) concentrations, which is in line with the research paper (0.1 - 1000 µM). For the last inducer, anhydrotetracycline, it was more difficult to determine the optimal concentration, since this compound was not in stock at our university. Nevertheless, we found that doxycycline (dox), another tetracycline derivative, was present. Our supervisors advised us to search for a concentration often used and apply that concentration to obtain a full dose-response curve. We found an article  that uses a concentration of 250 ng/mL dox to induce the pLtetO-1 promoter inside BL21 (DE3) cells.
The building of the sensing part was relatively easy. We ordered the required plasmids from Addgene (pKD227 and pKD233.7-3), purified the plasmids, and co-transformed them into BL21 (DE3) cells. The initial co-transformation failed, however, the second one succeeded, giving us altered bacteria which should behave as the sensing part. To test this, we prepared all of the remedies as well as any additional materials we would need.
To test the system, we made eight different cultures. The cultures consisted of five samples induced with different concentrations of tetrathionate, one negative control without tetrathionate induction, and two time-dependent samples. These time-dependent samples were induced with IPTG and dox at the same time as the rest of the samples, however, Ttr was added later. In addition to these samples, we also measured the emission of the buffer. The results showed that all eight samples emitted similar amounts of sfGFP, and the buffer emitted much less. Therefore, we concluded that the induction of the cells and the expression of sfGFP was successful and the right plasmids were present, however, the bacteria did not respond to Ttr as expected under these conditions.
We went back to the aforementioned research paper  and checked the concentrations of inducers used in their experiments. It appeared that they used an IPTG concentration of 0.01 mM and anhydrotetracycline was not used at all for their sensing system, since the background expression was sufficient. Based on those statements, we composed an experiment making eight different samples altering the IPTG concentrations between 0.1 and 0.01 mM, the tetrathionate concentration between 0 and 1000 µM, and the dox concentration between 0 and 250 ng/mL.
Test and Learn Again
We put the aforementioned setup to the test, as can be read about in Results. It was concluded that samples induced with 250 ng/mL dox, with and without tetrathionate, produced similar quantities of sfGFP. The other samples, without dox, only expressed sfGFP when induced with 1000 µM Ttr and (nearly) none when no Ttr was present. From these results, we concluded that the dox concentration should be further analyzed and possibly adjusted. We set up two experiments testing several dox concentrations ranging from 0.1 ng/mL to 250 ng/mL and tested the efficiency of the sensing part of our project. From this experiment, it turned out that concentrations of 250 ng/mL and 100 ng/mL were indeed overexpressing TtrR and that there was no difference between samples induced by IPTG concentrations under 10 ng/mL. Therefore, the advisors suggested using no doxycycline for the final design of our project, as supported by the research paper.
Daeffler, K., Galley, J., Sheth, R., Ortiz‐Velez, L., Bibb, C., Shroyer, N., Britton, R. and Tabor, J., 2017. Engineering bacterial thiosulfate and tetrathionate sensors for detecting gut inflammation. Molecular Systems Biology, [online] 13(4), p.923.
Mazumder M, Brechun K, Kim Y, Hoffmann S, Chen Y, Keiski C et al. An Escherichia coli system for evolving improved light-controlled DNA-binding proteins. Protein Engineering Design and Selection [Internet]. 2015 [cited 3 July 2021];28(9):293-302. Available from: https://academic.oup.com/peds/article/28/9/293/1605339
In our project, we want to image the ARG gas vesicles with ultrasound. Therefore, we needed to know which ultrasound machine and probe to use, how to handle them, and which script to employ. We looked into two articles that described ARG and tried to find two scripts that could be utilized to make ultrasound images and to collapse the gas vesicles [1,2]. Unfortunately, the scripts that were used by these particular researchers were not available and they did not respond to our requests. This was no problem since our university contains an ultrasound department, and already possesses a lot of standard research scripts we could use.
During the meeting with Hans-Martin Schwäb, an ultrasound expert at our university, we discovered that the article supplied a lot of detailed information about the settings and the setup of the ultrasound script and equipment. He told us to use a vertical sweep of the phantom, containing gas vesicle producing bacteria, with an L22-14V Verasonics probe. Some details about the research paper are further explained in Human Practices segment the Science.
After researching the paper and talking to ultrasound experts, we started with the building phase. During a new meeting with Hein de Hoop, another expert at our university, we got familiar with the ultrasound script and we were able to change most of the settings of the Verasonics standard ultrasound script to match our desires. To test this script, we imaged a standard phantom from the PULS/e lab at the university, adapting these settings to see if the settings work. In addition, the setup was constructed by clamping the transducer above the plate and making good contact between the probe and the phantom using a Phosphate-Buffered Saline (PBS) buffer.
The next day, we tested the ultrasound setup and the script on our own samples, which were all induced with different IPTG concentrations and therefore, contained different concentrations of ARG vesicles. A clear transition between the liquid PBS buffer and the solid phantom was visible, however, a lot of background signal was visible as well (Figure 1). Furthermore, it was not clear to us whether we saw the PBS, the gas vesicles, or something else. So we went back to Hein de Hoop to help us out.
After explaining our problem to Hein, he suggested clamping the transducer at an angle above the plate. The white stripes we saw on the images were probably created by the reflections of the sound waves at the bottom of the phantom. When angling the transducer, the reflections on the flat bottom of the phantom will bounce away from the transducer. Since the gas vesicles are round/ellipse shaped, the ultrasound signals hit the gas vesicles perpendicular, reflecting the signal back to the transducer. The noise reflections of the table will therefore not be visible on the images, while the vesicles will be visible.
Accordingly, we changed the setup of the ultrasound and tested new phantoms with the vertical transducer and by angling the transducer. This time, Hans-Martin Schwäb joined us in the lab to check the results and to help us configure some minor settings, which had to be set manually each time. The angled images indeed showed better results, than the vertical images. During this measurement session, Hans-Martin also explained a little bit about post-image processing and he pointed out that we should save the raw data instead of the images themselves. Not only is the quality of raw data higher than the quality of the images, image processing is also easier with raw data. We implemented this right away.
We had another meeting with Hans-Martin to design a script that could read the data and could calculate the difference between the images before and after the collapse of the gas vesicles (using the in-phase Quadrature, IQ). That way we could remove all background signals (e.g. from the agar) and it should be able to visualize only the gas vesicles. All latter ultrasound measurements have been performed with this script, this processing, and this setup. For all results please look at Results segment ARG1 reporter system.
Figure 1: The difference in before and after collapse images, imaged with ultrasound with on the left with a straight probe and on the right with an angled probe.
Bourdeau, R., Lee-Gosselin, A., Lakshmanan, A. et al. (2018). Acoustic reporter genes for noninvasive imaging of microorganisms in mammalian hosts. Nature 553, 86–90. https://doi.org/10.1038/nature25021
Lakshmanan, A., Lu, G., Farhadi, A. et al. (2017) Preparation of biogenic gas vesicle nanostructures for use as contrast agents for ultrasound and MRI. Nat Protoc 12, 2050–2080. https://doi.org/10.1038/nprot.2017.081
During the development of our kinetic model, we ran through several iterations of the design cycle. Our supervisor, Tom de Greef, supported us during this process. Tom de Greef is an Associate Professor of Synthetic Biology in the Department of Biomedical Engineering and specialized in the modeling and simulation of biomolecular networks. Tom gave us the advice to use the Matlab add-on SimBiology and pointed out that it was important to keep the design cycle in mind during the development of a computational model. So instead of trying to build the entire model in one step, it is important to make intermediate, simplified models. These simplified models can be tested and validated before expansion. In this way, we were able to correct our mistakes and learn from them before moving on to a more complicated model.
After deciding on our concept in the early stages of our project, we determined what type of model we wanted to develop. We researched previous iGEM teams and found that the most useful model for our project would be a kinetic model of the TtrS/R two-component system. We set up a meeting with prof.dr.ir. Tom de Greef to discuss the main components of our model and how we want to achieve these. We established that our output should consist of a dose-response curve, from which we can derive how much GFP can be obtained for different amounts of tetrathionate. Later, this model can be expanded to derive the number of gas vesicles that are produced.
We started by searching the literature for the chemical reactions that take place in the transcription and translation of our two-component system, and the signaling cascade induced by tetrathionate. Various versions were made and in each version, more chemical reactions were added, thus expanding the model to be more extensive and more realistic. We decided on building the following five versions, each containing the components from the previous version and an additional extension.
A model consisting of only the two-component sensing pathway, and the activation of GFP. This version contains a constant concentration of TtrS and TtrR, no delaying steps, and no degradations are taken into account.
Expanded with the transcription and translation of TtrS and TtrR under the same promoter. The activation of TtrS/R transcription is modeled in one step, and no delaying steps or degradations are taken into account in this version.
Expanded with a more elaborate activation of TtrS/R transcription modeled in multiple steps.
Taking into account delaying steps and degradation reactions.
Including the activation of acoustic reporter genes and the formation of gas vesicles.
After the chemical reactions were established, we had to search for reaction rates. We did this by looking at similar models and searching for experimental data in the literature. In addition, initial concentrations had to be determined for some of the species. Some of these concentrations could be retrieved from the lab protocols, such as the doses of tetrathionate and IPTG that are added. Other concentrations had to be calculated using values found in literature, such as the number of ribosomes or RNA polymerase per E. coli cell. After the reactions, rates, and initial concentrations were determined, we could start building the model.
As mentioned before, the program that was used to build our kinetic model is the MATLAB add-on SimBiology. For this program, it is not necessary to calculate the ordinary differential equations manually, as this is all done automatically. In this program, we entered the chemical reactions, rates, and concentrations that were previously determined. Naturally, we only entered the values necessary for the reactions that we were modeling in that particular version. A more in-depth description of our method can be found on Model, segment method.
After building the models, we used the extra function of SimBiology called SimBiology Analyzer to simulate the model. With this program, you can simulate the expression of each species in the model over time. For each version, we analyzed several things to validate the accuracy of the model. Firstly, we checked whether the simulation only contained positive, logical values. We validated whether the protein and mRNA expression over time could be justified when looking at the chemical reactions that are happening in the TtrS/R system. In addition, we validated whether the total amount of TtrR and TtrS stayed stable, by summing the concentrations of all proteins either bound or unbound. The total concentration TtrR and TtrS should show a linear increase over time at first, and eventually stabilize. Further, the effect of the degradation reactions was checked by comparing the versions including degradation with the versions excluding degradation.
Testing of the versions often resulted in errors or inaccuracies which had to be resolved in order to expand the model with extra reactions. When validation failed, the first step was to trace and analyze the error. Errors could often be resolved by adjusting the simulation settings. Inaccuracies in the models could have different origins, such as inaccurate rate constants, initial concentrations, or reactions. Thus, when unrealistic protein or mRNA expression was found in one of the simulations, our initial action response was to delve back into the literature to improve our input parameters. If we could not find the origin of the error, we contacted our supervisor for additional support. This way we were able to solve the problems that occurred in each version, before moving on to the next version.
All results of this model can be found at Model, segment results. Right now, TtrS and TtrR transcription and translation are modeled equally. For future work, the model could be extended by including a more realistic transcription and translation of TtrS and TtrR on two different plasmids. Further, the parameters used in the model could be improved by conducting experimental measurements to determine the unknown reaction rates instead of estimating them. This way, the simulation of the system will become more accurate.