Team:Thessaloniki/Model

☰







Model






Overview


The model of our project provides the theoretical design and analysis of the toehold switches alongside their behavioral statistics and the mathematical modeling of the reactions occurring during their interaction with the chosen miRNA molecules as well as the experiments' modeling. By modeling the entire project (bottom-to-top approach), our team managed to make a targeted choice of specific sequences of toeholds and to save both time and funds. Furthermore, the simulations conducted by the dry lab’s software increased the reliability of the experimental results. Finally, the modeling of the experimental processes served as a guide for the organization of our wet lab’s experiments.



Bioinformatics Analysis


From the initial involvement our team had with the analysis of miRNAs that are differentially expressed in PDAC, we understood that we have to set a bioinformatic workflow that will confirm the literature. Our aim was to analyze gene expression datasets derived from PDAC patients, available at the NCBI GEO database. Three such datasets were retrieved and subjected to exploratory and differential gene expression analysis (between tumour and non-tumour samples) in R, using both CRAN and Bioconductor packages. The gene expression matrices were also merged to form a “supermatrix” to increase the sample size. Prior to downstream analysis, batch correction methods were applied on the merged matrix, using functions from the sva and limma libraries. The set of significantly differentially expressed genes was then used as input for GO and KEGG enrichment analysis in order to identify up-regulated and down-regulated pathways. Significantly differentially expressed genes will also be matched to drugs they are known to interact with in order to infer potential relationships with outcomes of clinical interest, such as response to treatment, adverse effects, survival etc.



Diagnostic Tool: Toehold Switches


Toehold switches are regulatory molecules that activate gene expression in response to the presence of RNA molecules with arbitrary sequences [1], [2]. Toehold switches can be used as a biosensor for endogenous RNAs, since they display a high level of orthogonality. [2] Orthogonality, programmability and a wide dynamic range are the features of toehold switches that render them ideal as a versatile and powerful platform for the regulation of translation. Therefore, the integration of toehold switches with an in vitro cell-free system facilitates the deployment of these molecules for the point-of-care (POC) diagnostics [1], [2].
As a riboswitch, the toehold consists of two components: i) a sensory domain and ii) a regulatory domain. The former specifically senses the environmental change (in our case the selected miRNA sequences presence or absence) while the latter is responsible for influencing the expression level of the downstream gene (in our case GFP). [3]



Toehold Design




So, our sequences follow the basic rules of design of toehold switches, with the difference that they include three extra subsequences and they are missing the pre-RBS area. The three extra subsequences "GUA", "GUGUGU" and "ACACAC", where added in order for the hairpin structure to be completed ("G-C", "A-U" pairs are bound together and "GUA-AUG" is a loop), while the pre-RBS area was omitted since it was unnecessary and lead to lower stability of the molecule. The previous changes were implemented due to the fact that the toehold switches were used for miRNA detection, which led to the designing of smaller toehold sequences than usual.

The way toehold sequences were designed enabled each miRNA sequence to correspond to multiple toeholds. Furthermore, different RBS and Linker sequences were used, in order for their effectiveness to be tested.



| 1-st and 2-nd generations:


 Concerning the use of RBS sequences, we decided to split our toehold switches in two groups (1-st and 2-nd generation) depending on the RBS sequence used each time.

The sequences of the two generations were respectively identical with the only difference being in the RBS area.
Finally, for the design (and for the analysis) of the toehold sequences we developed a software tool, utilizing Nupack and ViennaRNA in python. The sequences containing misplaced stop codons or restriction areas were automatically rejected.


Learn More:

Software



Toehold Analysis


| Troubleshooting

Our first approach was to built two Deep Neural Network Regressors, one based on [2] and its dataset and another one based on [5] and its dataset -referred here as “Green” and “Nicolaas” datasets respectively- as an attempt to predict the ON/OFF GFP fluorescence ratios of our Toehold switches. Unfortunately, due to the small number of examples in the “Green” dataset the R^2 score was near 0 and therefore inadequate, and though the metrics of the “Nicolaas” dataset were promising (R^2 near 0.5) we were not able to pass our toehold switches into the respective Regressor without a significant drop in the predictive ability of the model making it not viable as well. After our collaboration with the iGEM Patras team's dry-lab we turned to the Random Forest Regressor as it seemed to fit better to a dataset like Green's, one with so few examples. The score, though much better than the DNN approach on the same dataset, was still too low.

| Our solution

Therefore, the frustrating specialization of machine learning tools led us to develop our own software tool, as previously mentioned, to analyze the sequences. For this purpose, Nupack and ViennaRNA statistical models were utilized, making our software modeling tool compatible with both Windows and Linux devices, providing results from two different packages for comparison and finally generalizing the function of the model, making it possible for future iGEM teams to use it. Various values of the sequences’ properties were calculated:

  • ~delta-G free energy of the toehold switch sequence.
  • ~Frequency of the mfe (minimum free energy) structure of every toehold sequence.
  • ~delta-G free energy of the RBS-Linker of every toehold switch sequence.
  • ~Perfect matches: number of paired nucleotides between toehold sequence and miRNA. This part of the software has 2 outputs for ViennaRNA: 1) the number of the nucleotides, 2) the percentage of paired nucleotides compared to the predicted nucleotides in the design process.
  • ~delta-G binding of the toehold switch sequence.
  • ~Concentration dependency plot for every toehold sequence: plot that shows the relative concentrations of the complex miRNA-toehold, toehold and miRNA sequences.
  • ~PostScript images (.ps files) that depicts the complex of the toehold switch and miRNA binding.
  • ~Scalable Vector Graphics (.svg files) that depict the secondary structure of every toehold switch sequence.
  • ~Pair probability matrix: matrix containing the probabilities of pairs between two sequences.
  • ~Sampled structures values.
  • ~Partition function: algorithm for nucleic acid secondary structure including pseudoknots.



Evaluation

The design process of synthetic sequences of toehold switches resulted in 19 1-st generation sequences and three miRNA sequences (hsa-miR-143-3p, hsa-miR-30e-5p and hsa-miR-1246). In order for the suitable sequences to be selected, the results for each sequence of our software tool should be evaluated (since the selection concerned three different miRNA’s, the results were split into three groups respectively). The most important criterion was Gibbs Free Energy of the part of the sequence extended from RBS to the end of the Linker, since according to [2] it is the thermodynamic parameter with the greatest correlation to the functionality of the toehold sequences. The next criterion was based on the Binding Energy of the molecules; the more negative it becomes, the more the statistical probability of complex creation increases. Last but not least, the pair probability matrix, as well as perfect matches results, were considered during selection since they are the confirmation of miRNA’s successful binding to the corresponding location of the toehold switch.



Corresponding miRNAPart's Name on the RegistryFree Energy of the Toehold (kcal/mol)ΔG Binding Energy (kcal/mol)ΔG RBS-Linker Energy (kcal/mol)
ViennaRNANupackViennaRNANupackViennaRNANupack
hsa-miR-143-3pBBa_K3727001-28,6-24,6-18,38-23,2-4,4-6.358
BBa_K3727000-33,08-28,4-18,36-23.198-4,3-11.276
BBa_K3727002-27,99-24.2-12,88-19,22-4-6.143
BBa_K3727003-32,47-28-12,89-19,22-4-11.211
hsa-miR-30e-5pBBa_K3727005-25,62-21,4-15,77-23,09-6,3-7.598
BBa_K3727004-34,45-29,3-13,85-15.353-6,3-10.361
BBa_K3727007-26,73-21,6-18,06-22.174-6,3-7.589
BBa_K3727006-31,21-25,4-18,13-22.171-6,3-12.854
BBa_K3727009-29,34-25,4-18,93-21.189-6,3-7.589
BBa_K3727008-33,82-29,2-19,01-21.186-6,3-12.854
BBa_K3727011-28,48-25,3-15,49-19.058-6,3-7.598
BBa_K3727010-32,95-29,1-15,57-19.055-6,3-12.854
hsa-miR-1246BBa_K3727012-29,11-25-17,63-19.041-6,5-7.656
BBa_K3727013-33,59-28,8-17,55-19.043-6,4-11.064
BBa_K3727014-29,54-25,3-16,7-17.884-6-7.438
BBa_K3727015-34,02-28,8-16,62-17,78-5,9-10.815


Corresponding miRNAPart's Name on the RegistryPerfect_matches (percentage % number of bases binded)Design (paired + unpaired)
hsa-miR-143-3pBBa_K3727001(1,0 20)20
BBa_K3727000(1,0 20)20
BBa_K3727002(0,95 20)18
BBa_K3727003(0,95 20)18
hsa-miR-30e-5p BBa_K3727005(1,0 19)19
BBa_K3727004(1,0 19)19
BBa_K3727007(1,0 20)20
BBa_K3727006(1,0 20)20
BBa_K3727009(1,0 21)21
BBa_K3727008(1,0 21)21
BBa_K3727011(1,0 18)18
BBa_K3727010(1,0 18)18
hsa-miR-1246 BBa_K3727012(1,0 20)19
BBa_K3727013(1,0 20)19
BBa_K3727014(1,0 19)18
BBa_K3727015(1,0 19)18


Learn More:

Results



Final Dry Lab Results

After the evaluation process completion, the following toehold sequences were selected and ordered to be tested in iGEM Thessaloniki lab:
1-st Generation2-nd Generation
BBa_K3727001BBa_K3727000
BBa_K3727002BBa_K3727003
BBa_K3727005BBa_K3727004
BBa_K3727007BBa_K3727006
BBa_K3727009BBa_K3727008
BBa_K3727011BBa_K3727010
BBa_K3727012BBa_K3727013
BBa_K3727014BBa_K3727015

Concerning the generations (as previously mentioned) the results were surprisingly conclusive. The 1-st generation of toeholds didn’t satisfy the necessary thermodynamic conditions in order for the fluorescence to be a declarative factor of either the presence or the absence of any miRNA sequences. The 2-nd generation on the other hand, did satisfy all these conditions and was selected as the ideal group of toeholds to be used for the diagnostic tool. These results were later confirmed by our wet lab’s experiments.


Learn More:

Engineering Success
Results
Proof of Concept




Mass Reaction Kinetics

In this particular genre of mathematical modeling, the products of the reactions taking place within the biological system depend on the concentrations of the reacting elements. The modeling of such systems requires solving ordinary differential equations (ODEs), which correlate the concentrations of the reacting elements to those of the products. In our case, the solution of the ODEs was performed using MATLAB.

The system under study consists of our synthetic sequence, that is toehold switch, and its corresponding miRNA sequence. The miRNA sequence works as a trigger for the translation of the reporter gene to be initiated; a procedure that is suppressed if the synthetic sequence doesn’t interact with the trigger RNA.

The parameters used to describe the reaction rates of this particular system were taken from the team iGEM SASTRA Thanjavur 2019, since they studied the system toehold switch – miRNA as well.



| Interaction of the toehold switch with the corresponding miRNA


The initial model, which is the first step towards the study of the system, consists of four differential equations. Essentially, it concerns exclusively the interaction of the toehold switch with the corresponding miRNA and it refers to the final form of the complex suitable for translation. The reciprocity of the reactions of the system requires using two rates for each reaction, which are not identical since thermodynamically the molecules being bound to each other are favored. In order for the proper study to be done, another transitional situation is added between the closed hairpin and the linear arrangement of the complex. (as implemented by iGEM SASTRA Thanjavur 2019 and iGEM CLBS-UK 2017)



The diagram below demonstrates the Concetration of the Closed Toehold Switch (CTS), miRNA, Partially Bound Toehold Switch (PBTS) and Open Toehold Switch over time.



| Transfer process of DNA, translation process of the reporter gene and decay process of all the molecules of the system


The second model concerns the transfer process of DNA (which results to the initial form of the RNA synthetic molecules), the translation process of the reporter gene (in our case GFP) and finally the decay process (which includes all the molecules of the system: closed toehold switch, miRNA, open toehold switch). This model aims to the quantification of the concentration of the fluorescence protein and to the representation of the reaction occurring within the cell-free system for better accuracy to be achieved during the experiments.



The diagram below shows the concentration of the Closed Toehold Switch, Open Toehold Switch, miRNA and GFP. The initial concentration of miRNA is the same as DNA of Toehold Switch , 100nM.



| Quantification of fluorescence


The next and final step of the modeling is the quantification of the fluorescence provided by the system. As our system approaches with great accuracy the one of iGEM SASTRA Thanjavur’s 2019, we are allowed to use the parameters of their system for the measuring of fluorescence. In this last system, the maturity and decay of the GFP molecule are studied as well as the scaling factor, which represents the fluorescence’s quantity.



The diagram of the third model represents the intensity of the fluorescence that should be detected after a 2 hours period when the initial concentration of miRNA is 1,10,100 and 1000 nM.






See also:

SASTRA Thanjavur 2019 Model
CLBS-UK 2017 Model



Wet Lab Experiments Design

In order to organize our experiments, we used Benchling, an online platform that combines many of the necessary tools for the design and simulation of the experiments.

We used Benchling throughout the design of our experiments, for keeping notes of the lab work we conducted each day and keeping our results organized, creating and writing our protocols, visualizing our sequences and plasmids and finally, for performing sequence alignments. For our protocols and notebooks, you can check the pages Experiments and Notebook, respectively. However, the most important procedure that this tool came in handy for, was cloning. Specifically, we needed a tool for the simulation of the reactions involved in cloning, such as restriction digestions and ligation, to better design these experiments.

Learn More:

Experiments
Wet Lab Notebook




First of all, we used Benchling to create interactive maps of our g-Blocks and vector. After that, we used the digest tool to run a virtual digestion of each part to ensure that these sequences were compatible with the cloning method we selected -that is BioBrick Assembly. In the following figure, there is the map of the vector we used, as shown in Benchling.


Through the simulation of the digests, we wanted to ensure that there are no restriction sites of the selected enzymes (EcoRI and PstI) within our sequences, except for BioBrick prefix and suffix, respectively. This tool, also, helped us to predict the electrophoresis results, after the digestion of the vector. This was important because as it is clear in the image below, the two bands -the linearized vector and its insert- are of similar length and their lanes are very close and difficult to distinct. This led us to run the gel for a longer time than usually recommended.
For the simulation of the assembly of our parts with the vector, we used the Assembly Wizard tool. With this tool, we were able to select the vector’s backbone and our part, by picking the regions between the restriction sites and automatically create a new plasmid. This procedure could give us insights into the ligation of the two parts and mistakes in the design and orientation of the sequences that could impede the cloning procedure. Although we tested all the parts, here we attached only an example of a virtual plasmid created by the Assembly Wizard.


This was, also, of great importance, since we tried cloning many times without success. Having our assembly method tested, we made sure that there was no problem with the design of our parts and this made troubleshooting these experiments much easier.

Finally, we used Benchling’s alignment tool, to compare the results we got after the sequencing of the parts we were working with, with their initial sequence and determine if there were any mutations that would alter or hinder their function. For example, we used this procedure for iGEM Thessaly 2019 parts. You can find more information about it in our “Lab Book: iGEM Thessaly 2019 Parts”:

Wet Lab Notebook




Modeling the Device

Concerning our fluorometer device, we decided to model the hardware and mechanical part with CAD designs, which enable a bottom-to-top construction of the device.
  • Fig.1 Temperature Controller
  • Fig.2 3D CAD Design



A more analytical explanation of the desings is hosted on our Hardware page:


Hardware




Contribution to Future iGEM Teams

The software tool alongside the mass reaction kinetics model and our experiments’ model were developed to be generally used for any miRNA, RBS, Linker and Reporter Gene sequence selected. Therefore, we hope that these tools will be helpful for future iGEM teams which include toehold switches in their projects and search for an easy, fast way to design, evaluate and predict the behaviors of their toehold switches-miRNA system.


Learn More:

Resources Table



Summary


iGEM Thessaloniki’s 2021 Model consists of:



References