On this page the main findings and interpretations of the results of our project can be found. In our project we aimed to achieve four different goals. These four proof of concepts are summarized in Figure 1.
Ammonia as the only nitrogen source
As we plan on feeding our GMO with ammonia extracted from the MOF, an essential part of our project is that our GMO should be able to grow in the presence of ammonia as its only nitrogen source. Therefore, we grew 6 different Saccharomyces strains under different ammonia concentrations to see if the presence of ammonia would affect their growth. In the end we were able to identify the best performing strains. The protocol can be found in the Experiments page.
As can be seen from Figure 2, there are 4 strains that outperformed in the experiment. These strains were chosen for further experimentation. The doubling time of the chosen strains can be seen in Table 1.
|0.05 g/L||5 g/L||7.5 g/L|
We learned that two of the strains that we were planning to use didn’t perform well when ammonia was the only source of nitrogen. This experiment helped us to exclude them for further tests as they weren’t suitable for our purpose (see Description page). Four strains proved to grow in different concentrations of ammonia, with strain ySB76 (S. cerevisiae) being the fastest grower overall. We could not observe a clear tendency that would suggest the optimal ammonia concentration for growing the cells. This result is relevant for our project, since the concentration of this nutrient doesn't necessarily need to be kept constant, which allows for more flexibility in ammonia concentration of the cell culture and thus more flexibility in how much ammonia the MOF should be able to capture. As we wanted to test different chassis in the optimization process, after this experiment we decided to express alpha amylase in yYS76, yYB77, ySB78 and ySB85, which correspond with 3 S. cerevisiae strains and 1 S. paradoxus strain.
For our second proof of concept, we wanted to demonstrate that it is possible to use Golden Gate Assembly to clone heterologous genes for alpha-amylase in Saccharomyces spp. Going from the gene of interest to the cassette plasmid expressed in Saccharomyces spp. is a long process. Throughout it there were some checkpoints that we used to confirm the success of the cloning: GFP screening; DNA concentration measurements with a Nanodrop; basic parts sequencing; and digestion of the final cassette plasmids.
The empty plasmid (pRS426__ConLS'-GFPdropout-ConRE'-URA3-2micron-Kan) used in the Golden Gate cassette plasmid assemblies contains a GFP dropout. This way, when the assembled plasmid is inserted correctly, the GFP insert will dropout and white colonies will be formed. Whenever the plasmid was not inserted correctly, the GFP dropout will still be present and colonies will appear green when examined under UV-light (Figure 3). With this simple technique we make sure that the chosen colonies have uptaken the assembled plasmid without the need of performing sequencing or digestion experiments prior to plasmid isolation. This GFP screening was used both for assembling our part plasmids and cassette plasmids.
GFP screening smoothened the process, considering the amount of samples that we needed to assemble. Generally, all the transformed E.coli with cassette plasmids (64 plates) showed a high efficiency of the Golden Gate assembly; a few countable green colonies were observed in contrast to hundreds of white colonies. One sample (SP007) didn’t show any white colonies, the sample was excluded for future experiments.
After purification of cassette plasmids amplified in E. coli, DNA concentrations were measured in order to assure a successful purification prior to Saccharomyces spp. transformations, results can be found in Table 2. These measurements were performed using a nanodrop spectrophotometer, measuring the DNA concentration at a spectrum of wavelengths.
|Sample number||ng/ul||A260/280 ratio||Sample number||ng/ul||A260/280 ratio|
The A260/280 ratio represents the purity of the DNA, an ideal sample would have a value of approximately 1.8. Due to large variation in DNA concentration and A260/280 ratio, 20 μl of each cassette plasmid was used in the Saccharomyces spp. transformation instead of 5 μl. Overall, the concentration of DNA in the samples after transformation and plasmid isolation were good and ensured the continuation with Saccharomyces spp. transformations.
To ensure that our ordered G-blocks for part 3 and part 3B we made were correct, we sequenced the basic parts (4 full alpha amylases and 4 alpha amylases without the native secretion peptide) assembled in the entry vector (more on this on the Engineering page). pYTK001 (entry vector) with Part 3 and 3b was sequenced using two primers flanking the gene, the vector is used to compose the cassettes plasmids in further experiments. All sequencing results are summarized in Table 3.
|Part||Alpha amylase Gene||Mutations||Comments|
|Part 3||From A. oryzae||-||Good quality sequence, 100% gene coverage.|
|From B. amyloliquefaciens||c.301C>Tp.T101I||The amino acid is not in the substrate binding site of the protein. 100 % gene coverage.|
|From B. licheniformis||c.690G>Tp.D231Y||Mutations in the flanking aspartate have shown a decrease in thermostability . No data about our mutation. 100% gene coverage.|
|From B. subtilis||-||Good quality sequence, 100% gene coverage.|
|Part 3b||From A. oryzae||-||Good forward sequencing, 63% gene coverage.|
|From B. amyloliquefaciens||-||Good forward sequencing, 52%|
|From B. licheniformis||-||Good forward sequencing, 66% gene coverage.|
|From B. subtilis||-||Good forward sequencing, 50% gene coverage.|
Generally, the results were quite successful and no major mutations were observed. However, two clear missense mutations were identified in alpha-amylases from B. amyloliquefaciens and B. licheniformis that may have an effect on the protein functionality. It would have been nice if we had the time to restart the cloning from the beginning, unfortunately, the cassette plasmids were already assembled and the time was very limited. In Part 3b full coverage of the gene couldn’t be achieved, repeating the sequencing was planned but sadly we didn’t have the time.
All in all, the mutations found will be taken into account for further results of those enzymes.
We used the restriction enzyme BsmBI to check for the correct ligation of the different constructs of the library. There are two sites for the enzyme BsmBI in the vector employed (pRS246) for our constructs, the sites are in part 1 and part 5, on both sides of the insert of interest (Figure 4). The use of BsmBI allows us to check for the size (Table 4) of the construct and assess the success of the assembly. The results of the digestion can be found in Figure 5.
|Sample||Expected sizes A (bp)||Expected sizes B (bp)||Sample||Expected sizes A (bp)||Expected sizes B (bp)|
After the digestion of all the constructs it can be seen that the expected sizes of the engineered plasmids match the sizes of the bands shown in the gel. The efficiency of the Golden Gate assembly is high. There are two samples in which a band of the size of the GFP-drop out is observed, SP012 and SP058, besides the expected size. This result may be due to contamination while pipetting. Nevertheless, the expected band can be observed, the samples were used in follow up experiments, consideration on alpha amylase activity will be taken into account. Additionally, some samples (SP006, SP011, SP022, SP023, SP044, SP046, SP047, SP052, SP053, SP055) show faint bands, likely evidencing that the concentration of DNA was low in the first place. The faint bands have the expected size, and considering the high efficiency in the rest of the digests we conclude that our Golden Gate assemblies and further screening was highly successful. No samples were excluded after digestion.
Alpha amylase production
For our third proof of concept, we wanted to show that our GMO does indeed show alpha-amylase activity after taking up our construct. We aimed to measure the alpha-amylase production of 64 samples listed at our Engineering page. This was tested with two main experiments. The principal one being an alpha amylase kit assay used to perform high throughput experiments and obtain quantitative data, needed for our model (see Model page). Additionally, a qualitative assay was employed to confirm the production of alpha-amylase in some of the samples. To understand more how this assay works, see the Experiments page.
Alpha-amylase starch breakdown
Functionality of alpha-amylase can qualitatively be proved by observing the breakdown of starch. The ability to break down starch is only available when amylase is produced, which in the case of S. cerevisiae and S. paradoxus doesn’t happen natively (Figure 6e). However, the assembled constructs should be able to break down starch. In order to confirm this, selective media plates without uracil (specific for the assembled constructs) were created with the addition of 1% starch. Several samples were cultured ON and treated before being plated on the starch plates, as described in the notebook (wk 39), after which pictures were taken (Figure 6) of the with/without iodine treated starch plates. Samples tested with this assay can be found in Table 5.
|SP001||ySB77||pPGK1||αMF + BS||tTDH1|
|SP010||ySB77||pHHF1||αMF + BS||tTDH1|
|SP021||ySB77||pPGK1||αMFΔ + BA||tTDH1|
|SP026||ySB76||pHHF1||αMF + BL||tTDH1|
|SP034||ySB76||pTEF1||αMF + BL||tTDH1|
|SP035||ySB76||pPAB1||αMF_no_EAEA + BL||tTDH1|
|SP051||ySB76||pRNR1||αMF_no_EAEA + BL||tTDH1|
|SP063||ySB76||pRNR2||αMF + BL||tTDH1|
As can be seen in Figure 6, a larger halo was observed in the cells’ lysate samples. Halos were also visible in most of the supernatant samples. This implies that alpha-amylase is present extra- and intracellular after cell lysis. We were able to confirm that this activity was only produced in the strains that had taken up the created constructs, as the non engineered strains didn’t show any halo at all. Overall, we see a difference in the alpha-amylase activity when samples were treated differently, suggesting that the secretion of the enzyme varies between constructs. More specifically to our project, we planned to isolate the protein from the supernatant, this way the processing of the cells is minimal and lysating is not directly necessary. Therefore, it was interesting to find that lysating some of the samples did produce a bigger halo, since this could change our proposed way of extracting alpha-amylase. Only few samples of the library were tested through this essay, therefore general conclusions can not be drawn, but confirmation of alpha amylase-activity could be assessed.
Alpha-amylase assay kit
We used a colorimetric assay to obtain quantitative data of the activity of the alpha amylase produced by the engineered yeast, allowing for comparison between samples and for quantitative data to use in our model.
The assay is based on the capacity of alpha amylase to cleavage ethylidene-pNP-G7, resulting in p-nitrophenol, a product that can be read at a wavelength of 405 nm, more details on the Experiments page. A testing round was first performed (Figure 7) in duplicates per sample to check how the experiment was done and identify possible steps that may need to be optimized. This round also helped to exclude samples SP003, SP004 and SP036 for further analysis as we were tight on reagents and the results weren’t promising.
During the high throughput experiments, we tested the capacity to produce functional alpha-amylase of all the samples which were not excluded from the subset of the combinatorial library (the excluded samples are listed on the Model page, section Accounting for lost samples). Activity of alpha-amylase in triplicates of each sample can be found in Figure 8. Supernatants from overnight cultures were incubated with ethylidene-pNP-G7 in 96 well plates and the appearance of p-nitrophenol was measured every 5 min during the first hour, then once after 2h and 3h. If alpha-amylase was in the supernatant, an increase of absorbance was expected due to the appearance of p-nitrophenol, an indirect measurement of alpha-amylase activity. Therefore, activity (y-axis of Figure 8) was measured as the increase of absorbance (quantity of p-nitrophenol) that appears per time unit (d[p-nitrophenol]/dt).
After having excluded the samples in which the variance between replicates deviated too much from the average variance, a table summarizing the mean activity of each sample can be found below. The different combinations from the subset library are also displayed in Table 6.
|Sample||Strain||Promoter||Secretion signal||Gene||Activity values (d[p-nitrophenol]/dt)|
Experiment were performed in triplicate to account for stochasticity, samples SP006, SP011, SP020, SP021, SP042 and SP059 were excluded, as the variance between replicates deviated from the mean variance, evidencing that the signal was more likely an artefact than actual alpha amylase activity. For the rest of the samples displayed in the graphs it can be concluded that there is a high variability in terms of how active alpha amylase is. Some constructs aren’t successful in expressing active alpha amylase in the supernatant (any sample with an activity value below 0,00001 was considered non active). However, other samples produce bigger amounts of p-nitrophenol and thus have more alpha-amylase activity in the supernatant.
Some general trends can be observed by looking at the graphs and the table, for instance, a relatively high alpha amylase activity is observed in samples that contain the alpha amylase gene from the host B. subtilis. The interpretation of such a big set of samples and identification of the best performers is an arduous task. AI technology could help getting deeper insights on our library and finding a better combination from the combinatorial library, more on this can be found in the model page.
Improving alpha amylase production
We used the Automated Recommendation Tool (ART) to model which samples from the combinatorial library discussed on the engineering page would yield a high alpha-amylase activity in the supernatant (exploitative recommendations). Furthermore, the ART also provides recommendations on which samples should be tested in a future DBTL-cycle in order to improve the performance of the model (exploratory recommendations). On this page, only the final results of the ART are discussed. How these results are generated and how to analyze them is discussed on the modeling page.
|Rank||strain||promoter||secretion||gene||Predicted ln(activity)||Predicted activity|
Table 7 shows that all exploitative recommendations contain the alpha-amylase encoding genes from B.subtilis. Furthermore, the αMF secretion peptide occurs in 29 out of 30 recommendations. It is also striking that 8 out of 10 recommendations contain either the pRNR2, pTDH3 or pPGK1 promoter, with the pPGK1 promoter occurring in 3 out of the top 5 recommendations. No clear pattern can be noticed in the recommendations regarding which Saccharomyces spp. should be used as a chassis.
|Rank||strain||promoter||secretion||gene||Predicted ln(activity)||Predicted activity|
The exploratory results of the ART are shown in Table 8. While these patterns might seem mostly random at first sight, there are some striking patterns that reflect the frequency and variance distributions shown in figures 7 and 8 discussed in the “Step 1: Creating a well-balanced subset of the combinatorial library” section on the modeling page. For instance the pTDH3 promoter which was shown to be underrepresented in the training data, is part of 8 out the 30 exploratorium recommendations. If all promoters were equally distributed across the recommendations, it should have occurred around 3 times. Also the αMF_no_EAEA secretion signal occurs more often (13 out of 30) than expected (around 7.5 out of 30). However, the ySB76 occurs a lot less (1 out of 30) than expected (around 7.5 out of 30) even though figure 8 on the modeling page shows it contributes less to the overall variance in the dataset than the other strains.
Based on the results from the ART, it is expected that a device containing the alpha-amylase encoding gene from B.subtilis and the αMF secretion signal will yield a high alpha-amylase activity in the supernatant. Furthermore, the exploratory recommendations show some correlations with the frequency and variance distributions of the part-variants shown on the modeling page.
Final discussion, finding the most active alpha amylase and future perspectives
Overall, we were able to successfully clone and functionally express alpha-amylase in Saccharomyces spp. using Golden Gate Assembly. However, the level of alpha-amylase activity did differ based on the way of extraction (lysising or not) and the combination of construct and chassis. Some results by themselves may not appear to be very conclusive, but bringing all of them together helped us to get a better insight on how to find the most active alpha-amylase.
When checking the activity of alpha-amylase the samples that holded the native sequence (Parts 3 and Parts 3b) from B. amyloliquefaciens and licheniformis, a general trend was observed, the samples didn’t perform very well (in terms of alpha-amylase activity). Combining these results together with the sequencing results, it turns out that those two samples had clear missense mutations that may affect the performance of them. Two possible scenarios could explain the lack of activity, 1. the deficient production of the enzyme or 2. the lack of functionality. Considering the experiments performed it is hard to tell what holds true in this case. Performing direct measurements like SDS gel electrophoresis to identify the presence of the protein would have given us more insights into where the problem was. Further experiments would be needed to check for the consequences of the mutations. Ideally, we would have avoided unknown mutations by an early detection of them, unfortunately, at the time we found them all the constructs of the library were already assembled and we didn’t have time to restart again.
Besides the engineering strategy, a very important part when expressing heterologous proteins is measuring the successful expression and activity of the protein. Using different assays as we did is a safe and insightful manner of testing the performance of your device. Far from being redundant, we draw different but complementary conclusions from both experiments. The halo test helped us understand the importance of the secretion of the enzyme. On the other hand, the high throughput experiment enabled for more precise comparison between samples and proved to be an efficient way of getting quantitative data from your devices.
In our attempt to screen as many combinations as possible to find the one that would achieve the highest yield of functional alpha-amylase, we combined in vitro high throughput experiments with a machine learning tool (ART). By doing so we expanded the search space, by not being just restricted to the actual combinations that were tested in the lab. When merging these two elements (experiments and machine learning), we encountered some aspects that needed to be taken into consideration. In our case, the subset of samples from the combinatorial library that are going to be evaluated experimentally needed to be thoroughly selected, trying to find an even and representative distribution of all the elements inside each category (as explained in the Model page). Another important aspect is the selection of a performance measure that is representative of what would constitute a “good” device. We selected the activity of alpha-amylase in the supernatant of the machine to be this performance measure as it represents what we aimed to achieve (an efficient production of alpha-amylase) and initial results showed that it can be accurately measured.
There are some improvements that need to be made in the experiments pipeline (see section Improvements in the Model page). In the next DBTL cycle those changes would need to be implemented in order to get more accurate results.
In general, we learnt that when expressing a new gene in a chassis, it is rather unpredictable how the different elements in the device would interact between each other. From our experience, we believe that by using this kind of technology more knowledge about the performance of your device can be acquired and the optimal results can presumably be obtained in fewer DBTL cycles.
- Declerck N, Machius M, Wiegand G, Huber R, Gaillardin C. Probing structural determinants specifying high thermostability in Bacillus licheniformis alpha-amylase. J Mol Biol. 2000 Aug 25;301(4):1041-57. doi: 10.1006/jmbi.2000.4025. PMID: 10966804.