Read about how we leveraged the engineering design cycle to change the direction of our project and make a better project.
Following our decision to create a project on the resistance to antibiotics and the usage of some type of substitute as a solution for this problem, we started our designing process by brainstorming ideas about alternatives to antibiotics. We came across antimicrobial peptides (AMPs) which have proven to be effective against bacteria, viruses, fungi, and even cancer cells 1 . These peptides can be naturally found in animals and plants, as they are part of the innate immune response 2 . AMPs are a sophisticated way of fighting infections since compared to antibiotics, pathogenic bacteria have a far smaller likelihood of growing resistant against AMPs with time 2 3 . Deeper research into this topic showed us, that AMPs are a broad topic with many different AMP classes and origins. However, even if many AMPs are known, most of them are not very well characterized. This led us to the decision that we wanted to make a platform to create and test AMPs. However, in the laboratory and especially clinical use most AMPs have shown to be rather unstable and thus not suitable as a substitute to antibiotics 4 5 . We had to face this challenge and needed to come up with a strategy to increase the AMPs’ stability to use them in-vitro . A discussion with Prof. Groß gave us the idea of using cyclotides in our process. Cyclotides are plant-derived peptides with a cyclic structure. Cyclotides are known to be highly stable through their special structure which comprises disulfide bonds inside the circular structure of the peptides 6 . However, further reading showed us, that the antimicrobial activity of cyclotides often is accompanied by hemolytic activity, which makes cyclotides unsuitable for medical use. This led us to the idea to integrate none-cyclotide AMPs, whose antimicrobial activity has been shown, into a none-hemolytic cyclotide scaffold to make these AMPs more stable and easier to handle. We decided to use Golden Gate cloning for generating the constructs since this is a preferably fast method, as described in more detail in our cloning results part . In general, we generated plasmids containing the AMP-cyclotide constructs, the green fluorescent protein (GFP), and the cyclizing enzyme asparagine endopeptidase (AEP). We designed a workflow to clone the constructs, express them in Nicotiana benthamiana , extract the cyclotides and purify them via an integrated His-tag and test their antimicrobial activity. We designed our workflow in a way to be a suitable platform for the fast and efficient testing of different AMPs.
Design Iteration #1
Our first goal in Drylab was finding potential new antimicrobial peptides (AMPs) in the form of cyclotides. We first tried different methods of finding amino acid sequences like the ones we found in common databases and other resources.
It was important that the new findings contain the characteristic sequence of cysteine residues at predefined positions relative to each other, which are required to preserve the typical loop structures found in cyclotides.
Other residues between the cysteines should be allowed to vary.
Genome Mining for New AMPs
Using search engines like BLAST provided by NCBI wasn’t fruitful, as its results only provided us with sequences already available to us.
This makes sense, as BLAST, by using a so-called seed and extend approach, splits up the sequence provided by the user into keywords and searches its database for sequences featuring similar keywords (called seeds).
The sequences containing those seeds are then further analyzed by comparing the amino acids surrounding the seeds to the input sequence. If the similarity score crosses a certain threshold, they are returned as search results. Therefore, BLAST doesn’t give particular importance to the cysteine residues that are important for the loop structures compared to other amino acids at other positions, which meant that BLAST wasn’t the suitable algorithm for our use case.
Soon afterward, we learned about Hidden Markov Models (HMM) and their use in bioinformatics to find similar sequences based on a multiple sequence alignment (MSA).
Using HMM in combination with plant genomes we expected to contain still undiscovered AMP sequences, we finally found promising results.
y using an MSA as input, HMMs evaluate each residue according to its frequency at specific positions of the MSA.
As our selection of cyclotides, when aligned properly, all have cysteine residues at the same position, the HMM favors amino acid sequences that are similarly structured regarding cysteine residues.
Other positions in the MSA, which are not as relevant for the loop structures, have more variation in amino acids, therefore allowing more variation at those positions in the search results.
After some research, we decided to use HMMER, a sequence homolog search tool using HMMs, provided by hmmer.org, using Cucurbitaceae, Fabaceae, Solanaceae, Rubiaceae, and Violaceae genomes as a database and the MSA of the amino-acid sequences of circulin B, cycloviolacin O2 as well as cyclopsychotride A as an HMM.
This approach finally gathered plenty of useful results.
Design Iteration #2
Due to the hemolytic activity of cyclotides and the consequent risk for our Wetlab members, we decided to change our goal mid-project.
Instead of finding new AMPs and testing their antimicrobial activity, we decided to tackle the inherent stability issues of most non-cyclic AMPs by integrating (or grafting) their amino-acid sequence into the sequence of a cyclotide scaffold.
We expected the cyclic scaffold to stabilize the resulting construct while conserving its antimicrobial activity without potential harmful hemolytic activities.
For the Drylab part of the project, this meant that instead of finding new AMPs, we would now focus on testing potential cyclotide-AMP constructs before the Wetlab would start expressing and testing them manually.
Broken down, we had to find a method to test the stability of the construct as well as its interaction with a membrane.
After some research and meetings, it was clear we had to find a way to predict a tertiary structure from our amino-acid sequence, which we then had to use in simulations that allowed conclusions about the above-mentioned properties.
Derived from the designing process, our experimental design had four major steps: cloning of our vectors, AMP expression in Nicotiana , peptide extraction and purification, and finally antimicrobial activity testing. We started with the cloning process in competent E. coli cells where we generated the desired plasmids, as described in more detail in the cloning results part. After purification of our final vectors , we transformed Agrobacteria with the respective plasmids for agroinfiltration. We used the agrobacteria to infiltrate the leaves of tobacco plants for transient protein expression. As we integrated the green fluorescent protein (GFP) into the plasmid, we could check the leaves for fluorescent signals as a first protein expression control. Thereafter, proteins were mechanically extracted by pestling under liquid nitrogen 7 . Since our constructs were labeled by a His-tag, our next step after extraction was to purify the cyclotide with affinity purification. In order to gain higher concentrations, we use acetone precipitation. The final step in our project was to test the antimicrobial activity in different assay conditions.
Predicting the Tertiary Structure
Right around the time we started working towards our new goal, AlphaFold 2.0, developed by DeepMind and EMBL-EBI, was released to the public.
Before its release, AlphaFold 2.0 made headlines regarding its then unmatched accuracy in predicting the fold and therefore tertiary structure of proteins in the Critical Assessment of Protein Structure Prediction (CASP) competition in November 2020.
As the capacity of our personal computers was limited, we had to use AlphaFold Colab , a simplified version of AlphaFold 2.0.
Most of the results looked promising, but all lacked the essential connection between the first and last amino acid.
We guess that AlphaFold 2.0 misses such a bond due to the rareness of proteins featuring a cyclic structure.
To fix this, we wrote a program that would cyclize the structures for our purposes (see Model ).
We integrated testing and controls steps in nearly every step of our protocol. At the beginning of our process, we tested whether the constructed plasmids were correctly expressed in E. coli . For this, we used restriction digest and sequencing of our purified vectors. Furthermore, selective pressure of different antibiotics allows only correct clones to grow and we used blue-white screening in some cloning steps, as described in the cloning results section. After knowing, that the cloning worked, we needed to test our next major step, the peptide expression. Therefore, in the next step, we needed to check if the protein expression of our plasmid was successful in the transfected tobacco leaves. We designed our vectors in a way to easily allow this testing. GFP was part of the plasmid, so we were able to examine the leaves under a fluorescent microscope and verified the fluorescence intensity. Only leaves showing green fluorescence were harvested for protein extraction and purification, as described in the Expression in Nicotiana results section. Afterwards we checked with SDS-PAGE and western blot with an anti-His antibody whether our protein of interest, the cyclotide, was expressed. In the western blots, we used untransfected leaves as a negative control and a protein with His-tag as a positive control. For additional controls, testing the expression of GFP with an anti-GFP antibody or the expression of the AEP with an anti-HA antibody would also be possible. This could give us some hints if the proteins of our plasmid were expressed in the plant. In addition, we tested the antimicrobial activity of our constructs to determine if the AMPs that are incorporated into the cyclic structures are still functional. We used microdilution assays and radial diffusion assays to determine the antimicrobial properties. We conducted a series of experiments with different media since we were aware that activity can be hidden by the used media. For positive controls, we used the synthesized AMPs KR-12 and CHEN. In this way, we could determine if there is a change in their antimicrobial activity when incorporated into our cyclotide construct. In addition, we needed the control peptides to adapt the assays to our samples. We also included untransfected plants and plants transfected with p19 as negative controls. This was necessary to exclude that the leaves themselves have any antimicrobial activity. Moreover, we wanted to test the stability of our constructs since we expected them to be more stable compared to the single AMPs. However, due to lack of time, we could not do any stability analysis.
To test the construct for stability and its capability to interact with the membrane, we had to run a molecular dynamics (MD) simulation.
No one on our team had any experience regarding simulations or in using Gromacs 8 9 10 , our simulation software of choice.
However, we were able to reach out to several experts, including Dr. Thiel from the biomedical informatics department of the university hospital Tübingen, as well as team IISERK from Kolkata (see Partnership ).
Each of them helped us greatly towards understanding more about MD and Gromacs specifically.
Thanks to Dr. Thiel, we also were granted access to the supercomputer BinAC in order to perform some of our simulations.
Generally speaking, the simulations take place within a virtual box containing the predicted protein structure and a so-called solvent (e.g., water molecules).
Those simulations can be performed at different levels of precision.
An all-atom simulation describes a simulation where the trajectory of each atom is being computed (which is computationally intensive). In contrast, a coarse-grained simulation sums up groups of atoms to single entities, for which the trajectories are being calculated (can even be computed on a consumer-grade computer).
Please refer to our experiment wiki entry for more general information about MD simulations using Gromacs (see Experiments ).
In order to get the desired data about our constructs, we had several simulation setups:
- A coarse-grained simulation containing the construct and a membrane
- An all-atom simulation containing the construct and a membrane, provided by team IISERK from Kolkata
The dual setup was to get an idea about which construct performs best through coarse-grained simulations before running the all-atom simulations on BinAC which takes way longer.
Simulating Lipid to Protein Interaction
Apart from using Gromacs for simulation purposes, we also used AutoDock
Given a protein (in our case one of our constructs) and a ligand (a phospholipid), AutoDock perceives their positioning relative to each other as “genes”: values describing their conformation, orientation, and translation which can undergo random mutations which may increase their “fitness”, a score representing the total interaction energy of ligand and protein.
Initially, multiple instances of the ligand are being generated, each with a different set of genes.
Based on their fitness score, their chance to “mate” with another instance and create a new instance with a genetic set derived from its parents is computed.
The offspring can undergo some random mutation, altering its positioning relative to the protein.
Through this procedure, instances with a bad fitness die out, while those with higher interaction energy replicate and have a chance to increase their score by mating with others further. 12
Using AutoDock showed us which construct binds best to a phospholipid, and therefore which has the best chance to disrupt a membrane.
However, we also knew that it only shows the binding between a construct and a free phospholipid, not between the construct and a membrane.
This means that the simulated interaction can only happen once the construct has perforated the membrane.
To conclude which construct binds best to the membrane (or is best at initializing the perforation), the results from MD were more relevant.
We continuously had to adapt our project idea and workflow, due to new information and realizations from the lab. With new ideas, our design – build – test cycle had to start again and new experiments were performed in the lab until new results forced us to integrate new adaptations. When we started with the practical work, we were aware of the fact, that one usually needs more time than expected because of failure in experiments. However, the first experiments we did worked remarkably well. We never had issues with the Golden Gate cloning system and generated our plasmids very fast. However, after the first enthusiasm, we faced many challenges, while we performed our experiments in the lab. In general, we learned a lot about how to work in a team and what should be done if an experiment is not working or results are not conclusive. Good planning when starting with the lab work is essential because we needed to do as many experiments as possible in a short time.
The cloning and expression in tobacco processes worked as planned. The first challenge we faced was that the extraction and purification process showed the desired results in neither the western blots nor the antimicrobial assays. We started extensive troubleshooting to determine what could be the problem and how we could make our system work, which is described in detail in the Extraction and Purification results (Verlinkung Troubleshooting results extraction and purification) section. In short, we discussed eight potential problems that might be responsible for our protocol failure: (A) Problems with the blotting procedure, (B, C) Expression of the peptide or the AEP in plants did not work, (D) Low expression levels, (E) Inefficient crushing of the leaves, (F) Unsuitable extraction buffer, (G) Problems with the His-tag recognition and (H) Binding of our peptide to plastic. After defining our potential problems and finding possibilities for testing if they are a problem, we were able to exclude some of them, like problems in the blotting procedure by including a positive control. We noticed, that we could not directly test for some of our potential problems, like the expression of the peptide in plants, as all our testing strategies required a successful extraction beforehand. For other potential problems, we did extensive testing in the lab but were still not able to solve the problem. We considered the extraction buffer to be the most likely problem and tested four more buffer compositions, but none of them showed the desired results.
A highly important fact we learned was including controls. This is crucial for verifying the results. For example, in the first western blots, we did not include positive controls. This made the interpretation of results difficult since we were not sure if we had no protein of interest in the samples or if the antibody failed. Controls were also crucial for our assays since we needed a positive control to test if our assays were suitable for our AMPs, and negative controls to ensure that the tobacco extract itself did not possess any antimicrobial activity.
We also noticed how important it is, to find advisors that are experts in our field of studies. Our team did a lot of research and we read countless papers, but in the end, the advice and input we got from our PIs were essential to shape our work. As our project included different steps, we consulted different experts. Dr. Kolukisaoglu was our primary PI and helped to shape the whole project and was especially experienced in working with plants. Prof. Brötz-Oesterheld especially helped us with the antimicrobial assays. Prof. Groß allowed us to do the acetonitrile extraction in his lab, amongst other things. When our project took a new direction, we also looked for more experts that could help us with the new ideas. What also helped our project a lot was the excursion to Nomad, where we not only got ideas for the implementation of our project but also got advice on how to specifically change experiments in the lab. All in all, our project would not have been possible without the input of so many experts and we want to thank all of them for their help.
When our lab work started, we also learned how essential good communication in the team is. Our Wetlab team included many members and we took turns in working in the lab. Therefore, experiments one person started were often continued by another person. Good documentation of the lab work was essential to make this possible. In our weekly Wetlab meetings we discussed the results from the last week and planned the next week, so everyone was up to date. However, when experiments did not work as planned, the people in the lab had to decide how to solve these problems. As our team members had different levels of experience in working in the lab and also a different level of knowledge about our project, it was important that other team members were available when severe problems came up. In the end, we really learned how to organize ourselves as a team.
To sum up, we had to tackle many issues regarding our project. Of course, we were not able to solve every single problem especially due to the lack of time. However, we did a lot of troubleshooting and discussion and found suitable solutions for some issues. In addition, we have several ideas about how the project could continue. And what is even more important, we as a team gained a lot of knowledge in the methods we used. We learned how to plan, execute and analyze experiments on our own, which is crucial to be a successful scientist. In addition, we trained our soft skills like organize ourself and others and working together as a team. The iGEM competition provides an opportunity to develop yourself further, especially outside your professional field. We, iGEM Team Tübingen, can proudly claim that every single team member seized this opportunity!
Analysis of Gromacs Simulations
After finishing the simulations, we used VMD 13 to analyze the constructs regarding their interaction with the membrane and learn from the data we collected. We analyzed the parameters Root Mean Squared Deviation, Solvent Accessible Surface, and Hydrogen Bonds (for more details, see Model ).
1 Harris, Frederick; Dennison, Sarah R.; Phoenix, David A. (2009): Anionic antimicrobial peptides from eukaryotic organisms. In Current protein & peptide science 10 (6), pp. 585–606. DOI: 10.2174/138920309789630589.
2 Boparai, Jaspreet Kaur; Sharma, Pushpender Kumar (2020): Mini Review on Antimicrobial Peptides, Sources, Mechanism and Recent Applications. In Protein and Peptide Letters 27 (1), pp. 4–16. DOI: 10.2174/0929866526666190822165812.
3 Zhu, Meng; Liu, Peng; Niu, Zhong-Wei (2017): A perspective on general direction and challenges facing antimicrobial peptides. In Chinese Chemical Letters 28 (4), pp. 703–708. DOI: 10.1016/j.cclet.2016.10.001.
4 Koehbach, Johannes; Gani, Jurnorain; Hilpert, Kai; Craik, David J. (2021): Comparison of a Short Linear Antimicrobial Peptide with Its Disulfide-Cyclized and Cyclotide-Grafted Variants against Clinically Relevant Pathogens. In Microorganisms 9 (6), p. 1249. DOI: 10.3390/microorganisms9061249.
5 Gunasekera, Sunithi; Foley, Fiona M.; Clark, Richard J.; Sando, Lillian; Fabri, Louis J.; Craik, David J.; Daly, Norelle L. (2008): Engineering stabilized vascular endothelial growth factor-A antagonists: synthesis, structural characterization, and bioactivity of grafted analogues of cyclotides. In Journal of medicinal chemistry 51 (24), pp. 7697–7704. DOI: 10.1021/jm800704e.
6 Craik, David J. (2010): Discovery and applications of the plant cyclotides. In Toxicon : official journal of the International Society on Toxinology 56 (7), pp. 1092–1102. DOI: 10.1016/j.toxicon.2010.02.021.
7 Poon, Simon; Harris, Karen S.; Jackson, Mark A.; McCorkelle, Owen C.; Gilding, Edward K.; Durek, Thomas et al. (2018): Co-expression of a cyclizing asparaginyl endopeptidase enables efficient production of cyclic peptides in planta. In J Exp Bot 69 (3), pp. 633–641. DOI: 10.1093/jxb/erx422.
8 E. Lindahl, B. H. (2001). GROMACS 3.0: A package for molecular simulation and trajectory analysis. Journal of Molecular Modeling, 306-317.
9 H. Bekker, H. B. (1993). Gromacs: A parallel computer for molecular dynamics simulations. Physics computing, 252–256.
10 H.J.C. Berendsen, D. v. (2001). GROMACS: A message-passing parallel molecular dynamics implementation. Computer Physics Communications, 306-317.
11 Morris, G. M. (2009). Autodock4 and AutoDockTools4: automated docking with selective receptor flexiblity. Journal of Computational Chemistry, 2785-9271.
12 Garrett M. Morris, D. S. (1999). Automated docking using a Lamarckian genetic algorithm and an empirical binding free energy function. Journal of computational chemistry, 1639-1662.
13 William Humphrey, A. D. (1996). VMD - Visual Molecular Dynamics. Journal of Molecular Graphics, 33-38.