Engineering Success

Summaries of how the iterative engineering design process informed various subprojects


Many components of Cellucoat went through iterations of the design cycle before coming together in the finished product you see today. After feedback on the difficulty of protein secretion from certain bacteria, we decided to integrate our recombinant proteins through a co-culture between E. coli and K. xylinus, as supported our proof of concept with GFP. To mitigate the costs associated with by 60% BC, our team made supplemented our bacterial media with fruit waste media. To further alleviate the shortcomings of raw BC, we incorporated PHB and used unaxial testing to further improve its mechanical strength. To support the functionalization of our BC with antimicrobial peptides, we developed INN to take FASTA sequences of antimicrobial peptides and predict its properties.

Nisin Production and Characterization

We successfully cloned our composite part, NisQ-His (BBa_K4437001), into a pSB1A3 backbone, but had difficulty with protein expression because our expected band size (7kDa) was so small. We pivoted our strategy and successfully cloned NusA-NisQ-His (BBa_K4437002) into the Xpress expression vector (BBa_K3945014), forming a new composite part GST-His-NusA-NisQ-His (BBa_K4437003) which allowed us to successfully produce protein from BL21 E. coli. We visualized bands at 91kDa on our SDS-PAGE and Western blot gels which corresponded to the expected band size of GST-His-NusA-NisQ-His.

Figure 1.1 Western blot of the whole cell lysate of GST+NusA+NisQ auto-induced in overnight express. A his-tagged positive control was also included. The protein ladder used was the Novex sharp pre-stained protein standard. The antibodies used were Mouse Anti-HIS-tag mAb (Abcam) for the primary antibody and Goat Anti-Mouse:HRP (Abcam) for the secondary antibody.

Our future steps will involve cleaving NisQ from NusA and GST using enterokinase, and purifying our protein with Ni-NTA. Once purified, we planned to run a Kirby-Bauer antimicrobial test against B. subtilis and compare zone of inhibition sizes to our characterization tests with bought nisin.

We successfully characterized nisin against both Gram-positive bacteria and several fungi strains specific to fruit. Our partnership with U of Alberta yielded results for nisin’s effectiveness against fungi, which showed that nisin could not inhibit the growth of Mucor ramanpianus, Penicillium claviforme, Rhizopus, Penicillium duclouxii, Penicillium commune, or Trichoderma viride. Based on this information, we decided to no longer classify nisin as antifungal. Our characterization continued with our series of Kirby-Bauer disc diffusion tests, where we tested nisin against other common antimicrobial agents (thymol and phenol) against B. subtilis. Nisin effectively inhibited growth, and while not to the same extent as thymol and phenol, a visible zone of inhibition was present.

Figure 1.2. Nisin’s effectiveness against B. subtilis. Red arrow indicates the zone of inhibition, showing that nisin can effectively inhibit bacterial growth.

Figure 1.3. Average zone of inhibition sizes for Kirby Bauer disc diffusion test over time results against B. subtilis. B. subtilis was plated at time 0.00 hours and incubated at 37oC, and treatment rounds were applied every 2 hours for 6 hours total. Lab-grown BC paper discs were autoclaved and soaked in either bought nisin (20,000 - 40,000 IU/mL), thymol, phenol, or water and placed in a separate quadrant of the plate. Diameter measurements (n=3) were taken 24 hours later using Fiji ImageJ software.

For our application as a preservative packaging, we also tested nisin’s effectiveness against B. subtilis over time. We demonstrated that nisin could inhibit B. subtilis after 2, 4, and 6 hours of growth at optimal conditions (ie. 37oC on LB agar) - and because fruit transportation trucks are often much cooler, thus slowing down bacterial growth, we predict that our packaging would be effective for much longer. We also demonstrated our packaging’s success in preventing deterioration on fruit, as we set up a spot test with our lab-grown bacterial cellulose soaked in nisin, on the surface of grapes. We recorded a timelapse for 12 days and found that our packaging noticeably protected the grape’s surface against shriveling.

Using a Co-culture Between E. coli and K. xylinus to Integrate Recombinant Proteins

In the field of molecular biology Escherichia coli (E. coli) is one of the ideal model organisms especially when it comes with dealing with recombinant proteins. However, even with all of the advantages, one of the greatest disadvantages E. coli has is its difficulty with secreting proteins. Dr. Brianne Burkinshaw, a professor in the department of biochemistry at the University of Calgary confirmed the difficulty with the secretion of proteins from E. coli. Initially, we had designed our project in which E. coli would release our antimicrobial protein however, after talking to the difficulties involved due to the complexity of E. coli secretion systems we proceed to look for a new way of integration for our recombinant protein. We chose a co-culture system.

A co-culture is when two bacteria are able to grow together in the same media. In our co-culture system we have K. xylinus producing bacterial cellulose (BC), and E. coli producing our recombinant protein. In this system E. coli integrates itself within the BC fibers. Therefore, instead of secreting the protein we decided to lyse the E. coli cells using an autoclave at 121ºC and 15psi. This releases our antimicrobial protein -Nisin- into the BC fibers.

To model this we decided to use green fluorescence protein (GFP) which has been inserted into E. coli. This recombinant GFPE. coli was grown in a co-culture with K. xylinus to visualize the attachment and the effects of sterilization using an autoclave to the protein and BC. Initial tests using GFP E. coli included making a co-culture system inside a test tube. The BC produced in the test tube were green as outlined in Figure 2.1.

Figure 2.1. Preliminary co-culture in a test tube with K. xylinus producing bacterial cellulose and GFP E. coli expressing green fluorescence protein.

To further analyze the distribution of the GFP E. coli on the BC we proceeded to set up our co-culture systems in a larger surface area such as a petri dish as shown in Figure 2.2. However, the BC sample was very thin and the concentration of GFP was hard to visualize. Henceforth, another approach we took to visualizing GFP integrated into BC was folding it to better see the actual concentration of GFP integrated into the BC.

Figure 2.2. Figure 2: Co-culture in a petri dish. Petri dish on the left is a control with bacterial cellulose containing no GFP E. coli. Petri dish on the right contains GFP E. coli integrated within the bacterial cellulose fibres.

Figure 2.3. Folded bacterial cellulose produced in a co-culture.

This allowed us to reach the conclusion that protein producing E. coli -GFP E. coli in our case- is able to integrate itself within the BC fibers and remain there.

However, our BC serves the purpose of a fruit packaging and customers do not want bacterial cellulose with bacteria still present on it. For this reason all bacteria needs to be removed from the packaging. We used the autoclave to sterilize the BC sample which effectively removed all bacterial cells present on the BC by lysing them. This means our GFP E. coli would have been lysed by the GFP would theoretically remain on the BC. This process of using an autoclave to lyse E. coli cells containing our recombinant protein is deemed effective for our antimicrobial protein- Nisin. Advantageously, Nisin is a thermostable protein which has a denaturation temperature of 160ºC.

In doing so we were able to show the co-culture is an adequate way of integrating recombinant proteins secreted by E. coli.

Figure 2.4. A control bacterial cellulose which has not been autoclaved and does not have any GFP E. coli present. On the right is an autoclaved bacterial cellulose sample consisting of lysed E. coli cells and GFP remaining on the BC.

Cost of Bacterial Cellulose Production

One major factor that has limited the BC industry from making headway is it being capital intensive. This was the motivation for us utilizing the fruit waste media as through HP, we discovered that for Cellucoat to be implemented as a produce packaging, it has to be cost-attractive. To determine if our modified media made from fruit waste reduced the production costs of Cellucoat, we conducted a cost analysis. We performed the analysis by calculating the cost of producing BC based on the costs of the media for several iterations of the experiments.

Here is the summary of the cost analysis on a lab scale:

Figure 3.1. Summary of a cost analysis performed on several monoculture experiments for the production of bacterial cellulose

From the analysis, we discovered that glucose makes up about 33% of media costs. Hence, if we could make the glucose cost zero, we would reduce production costs by 33%. However, after running experiments from this analysis we realized that we were able to reduce the cost of production by 60% instead of 33%. This was from modified media consisting of 45% FWM and 55% HS. We can also hypothesize that since this media still contains HS media, reducing the amount of HS and maximizing BC yield could help us further reduce the costs of BC production. Ultimately, reducing the costs would make bolster the affordability of Cellucoat.

We also conducted a financial analysis to determine the costs needed to implement Cellucoat in the real world. Through this analysis, we found that the total start-up capital needed to launch Cellucoat is $19,201,035.56. More detailed information on the breakdown of our costs can be found on the Entrepreneurship page.

BC and PHB Mechanical Properties

After conducting HP interviews and testing the mechanical properties and strength of the BC our team was planning on using as the material for Cellucoat, it was found that BC is very weak. For comparison, when less than 2 mm thick, is 500x weaker than plastic of the same thickness, and has a strength akin to that of paper.

To rationalize the benefit of adding PHP to BC to create a composite that has improved mechanical properties, our team decided to use pure solid PHB that came as granules and dissolve it into a 0.25M solution with 100% Acetic acid. 1.5 mL of the 0.25 g PHB per 1 mL of acetic acid solution (0.25 g/mL) was poured into BC palettes that have been culturing for three weeks at 30 degrees and being fed using a 24 hour schedule with 4 mL of HS media. After both samples have grown, they were autoclaved, purified using NaHCO3 solution and air dried.

After drying, the samples were cut to size for uniaxial tests. The uniaxial testing looked at the tensile strength, stiffness, and stretchability of the BC and BC and PHB composites.

Figure 4.1: The ultimate tensile strength (MPa) of pure BC and composite PHB-BC.

Figure 4.2: The Young’s Modulus of pure BC and composite PHB-BC.

Figure 4.3: The maximum elongation of pure BC and composite PHB-BC.

Based on these results, BC showed an ultimate tensile strength 3.780MPa higher than PHB-BC, which means that BC only could withstand a larger force before tearing. The Young’s Modulus of pure BC was also 758.2MPa higher than composite PHB-BC. This value indicates that pure BC is the stiffer of the two polymers, and is less flexible. Finally, the maximum elongation of composite PHB-BC was 0.02868 higher than pure BC, meaning the PHB and BC composite was more flexible and able to be stretched. The reduction in strength of the PHB BC composite can be attributed to the uneven distribution of PHB within the BC due to the ex situ method used to incorporate PHB and acid hydrolysis causing the BC to break down when exposed to 100% glacial acetic acid.

Taken together, these results contribute to the iteration of the engineering design cycle. With this test, it has revealed the need for PHB as an additive to BC to improve its stretchability and flexibility. Furthermore, these tests have also demonstrated the need for the use of a co-culture to recombinantly integrate PHB into BC, as dissolving the hydrophobic PHB granules in acid solutions severely compromises the integrity of the material.

Using Involution to Predict Protein Properties

As we started exploring ways to functionalize our BC, we realized that it would be helpful to have a quick way to check whether proposed inserts actually had the antimicrobial functions we wished to use them for, given that not all proposed inserts had been experimentally verified. We compiled a user guide to different antimicrobial peptide prediction models so that the wet lab could begin preliminary experiments while we constructed our own network. We aimed to investigate whether involution was an effective way to make predictions about biological data, specifically FASTA sequences.

Once our network was complete, it was able to predict that the nisin fragment we aimed to clone and insert in the wet lab indeed retained its antimicrobial properties. This was then experimentally validated by Kirby-Bauer tests.

Figure 5.1. Prediction of antimicrobial properties of the proteins in the nisin fragment inserted into our BC.

Our INN accurately made predictions about antimicrobial properties when given a FASTA sequence, performing with the same accuracy as the convolutional model we based it on. Detailed information about the performance of the INN and the metrics used to evaluate it are discussed on the INN page.

Figure 5.2. Comparison of sensitivity, specificity, and accuracy between CNN and INN

Figure 5.3. Comparison of Maxwell correlation coefficient (MCC) and area under the receiver operating characteristic (AUROC) between CNN and INN

These results are promising, but in following the engineering design process, we know that our work is not complete, and we can extend it in future iterations; we’d like to attempt hyperparameter tuning again and construct the network using a different method than data reshaping to see whether this can make the network even more accurate. As well, we’d like to collect data on the computational load of our network to examine how its performance matches up to other models. In following the iterative engineering process, we’ve taken the first steps to apply a new machine learning technique in the synthetic biology context, laying the framework for more investigation into the uses of involution and its performance.