Models are a useful tool to gain insight on real life problems without using as many resources as typical wet lab experiments. Models provide a prediction to how a real life system will work based on justified approximations that are used to simplify the problem. However, due to these approximations, models at best are just educated guesses to how the real life system actually works. As with any science experiment, it is important to keep in mind these uncertainties when examining results and drawing conclusions.
With our model, we wanted to look more into the characteristics that define a good sensor. There are many important qualities to consider when building a sensor, but due to time restraints, we focused on the following three.
- Specificity describes the sensor's ability to detect the analyte of interest from a mixture of compounds. Specificity is a very important characteristic of a biosensor as a non-specific sensor will induce a false positive.
- Sensitivity or the limit of detection is defined as the minimum amount of analyte that can be detected. This is important when detecting small amounts as false negatives can occur if the sensor is not sensitive enough to detect the very small amounts.
- Range illustrates the minimum and maximum concentrations the sensor is able to detect.
We examined these characteristics using Hill plots, protein-ligand docking analysis, and protein mutagenesis. By examining the specificity, sensitivity, and range of our biosensor to our two metabolites of interest kynurenic acid (KYNA) and tryptamine (TA), we were able to gain a better understanding of the functionality of our biosensor.
To improve our detection system, we looked into how to make it more specific for our target ligands, kynurenic acid and tryptamine. In order to achieve this, we conducted thorough research on the binding pocket of the human Aryl hydrocarbon Receptor (AhR), the main protein of our detection system. AhR is also found in human cells and it functions as a ligand-activated transcription factor. It has been well characterized to bind to a variety of different substrates, which indicates that AhR may not be so specific for our target ligands (Larigot et al., 2018). Read more about GutLux’s detection mechanism in Design. By modeling the binding pocket, we could for example use computational protein mutagenesis tools to make the binding pocket chemically preferable for our target ligands and thus, improve the specificity of our GutLux biosensor.
Unfortunately, there are no reliable full AhR 3D-models available, since the full-length protein has not been successfully isolated and crystallized before (Mosa et al., 2021). However, the ligand binding domain, also referred to as the PAS-B domain, has been well characterized through homology modeling. Therefore, we decided to use only the PAS-B domain for further analysis. We also reached out to Maria Sammalkorpi, a research group leader at Aalto University. Her group is focusing on computationally modeling soft materials and her expertise is in molecular modeling. She suggested continuing our research with only the PAS-B domain, because if the full protein model is not reliable, problems will arise. We obtained the sequence from UniProt, and submitted it to trRosetta, a database based protein structure prediction algorithm. The confidence of the predicted model was very high, with an estimated TM-score of 0.871. We also tried using SWISS-MODEL to model the 3D-structure of the AhR PAS-B domain. The results were very similar to trRosetta, so we decided to proceed with the results from trRosetta.
To gain insights on the binding pocket, we conducted a docking analysis on the AutoDock Vina software. Molecular docking is used to predict the orientation and binding of a molecule to another molecule and in our case, the ligand to a protein. There are also many more softwares to use for molecular docking, but we decided to use AutoDock Vina, because it was suggested by the Aalto-Helsinki 2020 iGEM team, for its ease of use.
The goal for this docking analysis was to find out the exact amino acid residues affecting the ligand binding that would be good candidates for mutagenesis. To achieve this, we used three different AhR binding ligands: kynurenic acid (KYNA), tryptamine (TA) and indole-3-acetic acid (IAA). We examined these three ligands because KYNA and TA are our metabolites of interest and indole-3-acetic acid is a very well-known compound that would be easily available for us in case we wanted to perform wet-lab experiments.
In addition to our docking analysis, we did lots of literature research to determine other promising sites for possible mutations. During our research, we came up with two study papers, one by Bisson et al. (2019) and one by Szöllősi et al. (2016), which state that the main energy contributors for the binding between AhR and 2,3,7,8-tetrachlorodibenzodioxin (TCDD), an environmental pollutant and another important ligand of AhR, are Thr289, His291, Thr296, Ile298, Leu315, Ile325, His326, Cys333, Met340 and Gln383.
After visualizing the protein structure, we could start preparing the protein and the ligand files for the actual docking analysis. All the instructions can be found here. First, the PDB file needed to be cleaned and relaxed. The protein, AhR PAS-B, was relaxed using Rosetta software. We used Rosetta because we were planning on conducting a protein mutagenesis with the same software. The cleaning is done by removing water molecules and adding only polar hydrogens and Kollman charges. Cleaning is an important part of this process, because water molecules could affect the ligand docking by blocking the ligand from the pocket region, for example. This process was done with AutoDock MGLTools, which has a very user-friendly graphic interface. Next, the ligands were also prepared for docking with MGLTools. After the preparation step, both the protein and the ligands were saved in .pdbqt format.
The next step is to set up a grid box around the binding pocket. The grid box is used to define in which space the dockings for the ligands will be searched for. Because the binding pocket was already known and stated by a lot of literature (Bisson et al., 2009; Szöllősi et al., 2016), we could easily set up the grid box around the binding pocket, and thus, we did not have to perform blind docking. The final step is to run docking with desired parameters in AutoDock Vina. After running, Vina will give an output file in .pdbqt format, which contains a variety of dockings, depending on the given parameters.
For the analysis of the docking results, we used PyMOL to visualize the success of the docking. As can be seen in Video 1, there is a clear binding pocket where kynurenic acid is docked. For kynurenic acid, we got estimated binding affinities ranging between -15.4 and -12.9 kcal/mol, which is a good result. However, when visually estimating the results, we noticed that the two docking results with highest affinities are not located inside the binding pocket, so we decided to discard those results. After reviewing the docked molecule, we got a list of amino acid residues affecting ligand binding. We found out that the following residues affect kynurenic acid binding: Thr289, His291, Gly321 and His337.
For tryptamine, the estimated binding affinities were not as high as for kynurenic acid. The values were from -6.2 to -4.6 kcal/mol, which is a good result too. In this case, the results with highest affinities were located within the binding pocket. We found out that the following residues affect kynurenic acid binding: Phe287, Thr289, His291, Leu315, Ile325 and Gln383.
Lastly, for indole-3-acetic acid, the binding affinity values were ranging from -6.4 to -4.5 kcal/mol, which was approximately the same as tryptamine. Also, the best docked molecules resulting in the highest affinities were located inside the binding pocket. We found out that the following residues affect kynurenic acid binding: His291, His337 and 383Q.
As a final result, we constructed Table 1, in which we can see what residues are the main energy contributors for the binding between AhR and the ligand. Based on this table, we can set out to examine the possibility to redesign the binding pocket, in order to achieve increased specificity to our target ligands.
After successfully conducting a docking analysis, we decided to look more into protein mutagenesis. We decided to use Rosetta software to predict mutations to the binding pocket of our target protein, because it was suggested by Aalto-Helsinki 2020 team. Rosetta includes algorithms for enzyme design, de novo protein design, ligand docking and protein structure prediction. The goal for this was to increase the binding affinity between AhR PAS-B and our target ligands and thus, make the binding pocket chemically preferable for those molecules.
Due to the docking analysis, we gained valuable information on the amino acid residues affecting ligand binding in AhR protein. By computationally creating mutations to these amino acids, we could possibly optimize the binding pocket for kynurenic acid and tryptamine. Based off our docking analysis and literature research, we planned to create mutations to the following amino acids: Thr289, His291, Thr296, Ile298, Leu315, Gly321, Ile325, His326, Cys333, His337, Met340 and Gln383. We also decided not to allow any amino acids to be mutated into cysteine, as it has a tendency to create sulphur bonds, which due to the nature of the model, would cause almost all of the amino acids to mutate into cysteine. Read more about design options here.
Unfortunately, we ran out of time and did not have a chance to fully conduct protein mutagenesis. However, conducting the computational experiment based on these premises is one of our short-term future prospects by which we would surely gain very valuable insight. Since modeling can only provide educated guesses, these results are tested in the wet lab to see if the predicted mutations increase the protein specificity as expected.
Sensitivity and Range of Biosensor
To better understand our biosensor’s limitations and feasibility, we examined the sensitivity, and range of our biosensor using the Hill equation. By examining the Hill equation, we get a better understanding of how our biosensor reacts in a range of different ligand concentrations and if our sensor is sensitive enough to detect the metabolite concentration levels typically found in the gut. The Hill equation, shown below in Figure 1, depicts the fraction of bound receptors as a function of ligand concentration.
Literature values of 1.4 μM and 0.2 mM for KA of kynurenic acid and tryptamine, respectively, were used for the analysis (DiNatale et al., 2010; Nguyen and Bradfield, 2008). For the Hill coefficients, no specific values were found for kynurenic acid and AhR binding or tryptamine and AhR binding. However, one source showed the Hill coefficients of different ligands to AhR binding (Hoffman et al., 2019). As a result the Hill plots were plotted for these different possible Hill coefficients. Figure 2 shows the Hill plots for kynurenic acid and tryptamine plotted on a semi-log graph. The curves were plotted for a range of possible Hill coefficients on a semi-log graph.
As we can see from the graphs, as the Hill coefficient increases the sigmoidicity of the curve increases creating a sharp inflection point. For a more switch-like, on-and-off biosensor, a higher Hill coefficient is desired. However, in our case, when we are wanting to quantify the concentration of our metabolite of interest, which correlates to the output or fraction of receptor bound, we hope for a smaller Hill coefficient. With a smaller Hill coefficient we achieve a greater range of concentrations that can be detected and quantified.
To examine whether our range of detection corresponded to expected metabolite concentrations in the gut, we compared the Hill plots with literature values of kynurenic acid and tryptamine that have been reported in the gut (Paluszkiewicz et al., 2008; Kuc et al., 2008; Saraf et al., 2017). For kynurenic acid, the curve of the graph falls nicely in between the reported minimum and maximum kynurenic acid concentrations, indicating the sensor should be sensitive enough to detect kynurenic acid. For tryptamine, the curve is to the left of the minimum tryptamine concentration. This would suggest that the sensor would quickly get saturated as maximum output is reached, and no quantification of the metabolite can be done.
However, just like with any model it is important to keep in mind the limitations, simplifications made and the key differences between the model and real life application. One limitation of the Hill plot analysis is that it is a static model. It only looks at the sensor's response with respect to a specific concentration without any time dependence. In the application of our biosensor, with real-time in vivo measurements, time is a very important factor. One specific case where time plays an important role in the sensitivity of the sensor is with diffusion and mass transport of the ligand across the semipermeable membrane that separates our biological detection mechanism from the gut luminal contents.
Our biological detection mechanism measures the ligand concentration in the capsule. Given infinite time, this concentration will equal the ligand concentration in the gut, which is the target measurement. In reality, the ligand concentration exposed to our detection mechanism will vary with time depending on diffusivity of the ligand and permeability of the membrane. The next step would be to analyze a dynamic model that examines the significance of the time dependence and possible concentration differences across the membrane that may result.
1. Bisson, W. H., Koch, D. C., O'Donnell, E. F., Khalil, S. M., Kerkvliet, N. I., Tanguay, R. L., Abagyan, R., & Kolluri, S. K. (2009). Modeling of the aryl hydrocarbon receptor (AhR) ligand binding domain and its utility in virtual ligand screening to predict new AhR ligands. Journal of medicinal chemistry, 52(18), 5635–5641. https://doi.org/10.1021/jm900199u
2. DiNatale, B. C., Murray, I. A., Schroeder, J. C., Flaveny, C. A., Lahoti, T. S., Laurenzana, E. M., Omiecinski, C. J., & Perdew, G. H. (2010). Kynurenic acid is a potent endogenous aryl hydrocarbon receptor ligand that synergistically induces interleukin-6 in the presence of inflammatory signaling. Toxicological sciences : an official journal of the Society of Toxicology, 115(1), 89–97. https://doi.org/10.1093/toxsci/kfq024
3. Hoffman, T. E., Acerbo, E. R., Carranza, K. F., Gilberto, V. S., Wallis, L. E., & Hanneman, W. H. (2019). Ultrasensitivity dynamics of diverse aryl hydrocarbon receptor modulators in a hepatoma cell line. Archives of toxicology, 93(3), 635–647. https://doi.org/10.1007/s00204-018-2380-z
4. Kuc, D., Zgrajka, W., Parada-Turska, J., Urbanik-Sypniewska, T., & Turski, W. A. (2008). Micromolar concentration of kynurenic acid in rat small intestine. Amino acids, 35(2), 503–505. https://doi.org/10.1007/s00726-007-0631-z
5. Larigot, L., Juricek, L., Dairou, J., & Coumoul, X. (2018). AhR signaling pathways and regulatory functions. Biochimie open, 7, 1–9. https://doi.org/10.1016/j.biopen.2018.05.001
6. Mosa, F.E.S., El-Kadi, A.O.S., & Barakat, K. (2021). Targeting the Aryl Hydrocarbon Receptor (AhR): A Review of the In-Silico Screening Approaches to Identify AhR Modulators. IntechOpen. DOI: 10.5772/intechopen.99228.
7. Nguyen, L. P., & Bradfield, C. A. (2008). The search for endogenous activators of the aryl hydrocarbon receptor. Chemical research in toxicology, 21(1), 102–116. https://doi.org/10.1021/tx7001965
8. Paluszkiewicz, P., Zgrajka, W., Saran, T., Schabowski, J., Piedra, J. L., Fedkiv, O., Rengman, S., Pierzynowski, S. G., & Turski, W. A. (2009). High concentration of kynurenic acid in bile and pancreatic juice. Amino acids, 37(4), 637–641. https://doi.org/10.1007/s00726-008-0183-x
9. Saraf, M. K., Piccolo, B. D., Bowlin, A. K., Mercer, K. E., LeRoith T., Chintapalli, S. V., Shankar, K., Badger, T. M., & Yeruva L. (2017). Formula diet driven microbiota shifts tryptophan metabolism from serotonin to tryptamine in neonatal porcine colon. Microbiome 5(77). https://doi.org/10.1186/s40168-017-0297-z
10. Szöllősi, D., Erdei, Á., Gyimesi, G., Magyar, C., Hegedűs, T. (2016). Access Path to the Ligand Binding Pocket May Play a Role in Xenobiotics Selection by AhR. PLOS ONE 11(1): e0146066. https://doi.org/10.1371/journal.pone.0146066