Team:Wageningen UR/Model/integration
Model integration
We identified two modules of our Mantis system suitable to mathematical modeling.
Signal Transduction
We use E. coli’s native Cpx signal transduction system to internalize the signal created by antigen binding. As explained here [Link to Bart result intro] this system depends on several protein-protein interactions. We tested which one of these interactions was most suitable to link to BiFC, both in the wet lab [Link to Bart result page] and in silico [Link to Sabine result page]. Here we show how we used and combined the output of both projects!
We went through several iterations between the model and the lab work which are explained in more detail in the collapsed boxes below. First, we modeled three possible systems to visualize antigen binding with the Cpx system, and found out which parameters we had to optimize in the lab to get the strongest signaling. We took these recommendations to the lab and found wich candidate system is indeed suitable for antigen visualization, and under which conditions this system works optimally.
We used the data gathered in the lab to fit with our computer model, to find out in what other ways our system can be optimized. We found several parameters that could improve the speed of our signaling, and we combined with the “Fluorescent Protein” project to test this hypothesis in the lab.
Phase 1: Initial system modeling
Firstly, we targeted to find out which protein-protein interactions in the Cpx pathway were most suited to connect to BiFC. Three candidate systems were analyzed, which were based on either CpxA-CpxR phosphotransfer or CpxR-CpxR dimerization (figure 1). While the constructs for the wet lab were created, initial in silico tests were ran. We defined protein expression and signal activation through addition of stress as the most important to test; these would later also be tested in the wet lab.
We quickly found out that in theory, CpxR-CpxR dimerization (figure A) was the most promising protein interaction to experiment with in the lab.
Results of all three setups can be seen at the (link) modeling page. Although the CpxA-CpxR setup is dependent on the right CpxA and antigen levels in the sample, it seems that the CpxR-CpxR setup is not dependent on any protein concentration. It even shows that the maximum reached YFP concentration is limited by CpxR, which can be increased in the lab. The strongest signal will be obtained when CpxR expression and Cpx activation are maximized (figure B).
We went into the lab to test these propositions!
Phase 2: Testing model propositions in the lab
To test these hypotheses, we created constructs in which we coupled split eYFP halves to CpxA and CpxR respectively and placed them under control of the inducible araC/pBAD promoter. We transformed E. coli K12 with these constructs. The Cpx system was activated with the known activator KCl in different concentrations to mimic different antigen levels at t = 20 min. Here (link) an extensive overview of the performed experiments can be found!
We quickly found out that the systems based on CpxA-CpxR interaction did not generate a clear fluorescent signal, which matches the prediction of the model! Furthermore, we show that visualizing CpxR dimerization with BiFC is indeed a viable option. Because we put the CpxR-eYFP-termini construct under control of the inducible araC/pBAD promoter we were able to test hypothesis 1: The strongest signal will be obtained when CpxR expression is maximized. We found out that this is indeed true. You can check this result here (Link Bart results).
We then set to test hypothesis 2: The strongest signal will be obtained when Cpx activation is maximized. In figure 3 we show that this is also true! We mimic antigen binding by adding a known activator of the Cpx pathway, and by increasing its concentration the fluorescence intensity also rises!
We take this data back to the lab to further improve our computer model!
Phase 3: Fitting the computer model to the lab data
To learn about the characteristics and potential of the system build in the lab, the laboratory data was compared to the model (Figure D). A parameter set was found which gave the most similar YFP production to the data found in the Lab. By doing this, we found that the best fitting set showed a relatively good score for fluorescence intensity, but it scored fairly low on the speed score. This means that there is still room for improvement!
Of course, to improve the current signaling system we want to know what parameters should be changed in order to increase the YFP signal and the response speed rate. To assess this, each parameter of the best fitting set was varied between 0.001 and 1000, keeping the other parameters constant. The effect of this change on the YFP score and speed score was calculated.
This figure shows that the YFP score has more possibility for large improvements than the speed score. Both scores could be profoundly improved by increasing the kinetic rates for antigen binding by the affibody and CpxR-eYFPc phosphorylation (check out all the used parameters here (sabine page). Given the iGEM time limits, we could not manage to adapt these in the lab. What could be improved is the necessary fluorophore maturation time. In the next box we explain how we combined several projects to examine this!
Phase 4: Implementing faster maturation
We aim to improve the response time of our visualization system. As stated, our model shows that this can be done by using a faster maturing fluorescent protein. During our “Fluorescent Protein” project we tested a number of fluorescent proteins, of which mVenus showed the shortest maturation time. Furthermore mVenus is designed to have a fast and efficient maturation time [1], exactly what we need! We show here (link demo) how this experiment was designed.
The results show that usage of mVenus over eYFP as a reporter protein increases the produced fluorescent signal some ten times! Unfortunately, the background signal also increases a lot, which means we lose a lot of specificity. We hypothesize that the maturation rate of mVenus is too high, which means that many non-specific interaction become irreversible, which leads to high fluorescent signals, even when no activator is present. This means that mVenus is not a suitable candidate to visualize antigen binding.
During this project, more reporter proteins were tested. Unfortunately we didn’t have time to test, nor model these. At this moment, we recommend testing sfGFP as a reporter for antigen binding. We found out sfGFP is thermostable, it is efficient in maturating at high temperatures, while still being one of the fastest and brightest reporters we tested.
Conclusions
Three strategies were tested to find the optimal visualization method of antigen binding using the Cpx pathway. In the end, it was clear that fusion of eYFP-termini to CpxR and measuring its dimerization shows clear fluorescence within two hours after activation, even in low concentrations! This shows that we found a potentially usable method of rapidly and specifically measuring antigens in blood.
During this “wet lab” project, we constantly implemented results from the computer modeling to improve the experiments performed in the lab. As it turns out, the “wet lab” and modeling results are very similar! Check out how we integrated the lab work with modeling here!(link to model integration)
Furthermore, we had time to integrate this visualization module with other wet lab projects: check out here(link to demo page) how we combined this project with the Signal Transduction and Fluorescent Protein experiments!
Quorum sensing
We wanted to incorporate quorum sensing to make Mantis more robust and increase sensitivity to be able to sense very low antigen concentrations. By using quorum sensing, we aim to convert the antigen concentration into a singaling molecule concentration. Every biosensor cell in our Mantis system will then be able to detect the signaling molecule. To do this, we need a quorum sensing system maybe link to background info on LuxR, LuxI, AHL, [LuxR-AHL], dimer, etc that can self-activate, but only does so upon induction. As a starting point we took the BBa_K1913005 biobrick. This contains the LuxI and LuxR proteins from Alivibrio fischeri, each with their native promoter. It also contains the GFP protein under control of a promoter identical to the promoter of LuxI. LuxI can auto-activate itself while GFP can report on the amount of auto-activation.
Previous experiments have shown that this construct produces fluorescence even at low cell densities. This suggests that leaky expression of LuxI and high basal levels of LuxR will result in a significant amount of GFP production. The goal is to construct a variation on this part that no longer has spontaneous activation, but can be induced by an outside mechanism, for example a two-component system.
Existing biobrick
Natalia says something about her initial BBa_K1913005 experiments
Modeling insights
In paralel to the lab experiments, we developed a model of our quorum sensing system. The global parameter space was explored showing the important role of certain parameters in preventing spontaneous auto-activation.
The model showed that in simulations, a low rate for the cellular AHL degradation is associated with spontaneous auto-activation. The role of using aiiA to combat leaky expression has been explored by previous iGEM teams; ETH
The next step was to add aiiA to the spatial model of the entire system. This includes both the quorum sensing system, as well as the lytic mechanism responsible for producing a fluorescent signal. In the spatial model, parameter sets that result in spontaneous auto-activation of the system were simulated after including a production term of aiiA. Now, cells induced by antigen detection stop producing aiiA. In the model, aiiA is able to suppress auto-activation. Then, when a signal is detected and aiiA production is halted, levels of aiiA will drop until that specific cell starts auto-activation spontaneously, similar to how all cells behaved before addition of aiiA. Auto-activation of a small number of cells would eventually lead to auto-activation of all cells in the system. After adding aiiA to our model, we simulated a model where cells are able to communicate the presence of antigen, resulting in a population-wide signal. But in the absence of antigen, cells remain inactive and no fluorescence is generated.
We were able to tune the kinetic parameters of aiiA so that aiiA would prevent spontaneous auto-activation caused by leaky expression while not preventing the detection of low levels of AHL produced by neighboring cells.| Table 3: A system with negatively regulation of LuxR that self-activated spontaneously was modified by adding aiiA or increasing the degradation rate of LuxR, resulting in the desired behavior. | |||
|---|---|---|---|
| Score | Animation, no antigen | Animation, antigen | |
| Original set | 99.97 | link | link |
| Original Set & aiiA | 0.059 | link | link |
| Original Set & increased LuxR deg. | 0.084 | link | link |
These modeling results support the strategy of using aiiA to tune the sensitivity of the quorum sensing construct and indicate that in principle the functionality we want to engineer can be obtained using only the components we are using.
Lab implementation
Natalia says something about how after adding aiiA, the construct produces significant fluorescence after inducing the system.
References
- Nagai, T., Ibata, K., Park, E. S., Kubota, M., & Mikoshiba, K. (2001). A variant of yellow fluorescent protein with fast and efficient maturation for cell-biological applications. Nature Biotechnology, 20, 1585–1588.
