Cycle 1
Design
To create an adhesive platform called CoPlat, we selected six adhesive proteins as the source of adhesion.
These proteins consist of natural adhesive proteins, such as mussel protein (Mfp3, Mfp5), and barnacle
protein (Bamcp20K-1, cp19K, Mrcp20K, Aacp19K). We connected each of these adhesive proteins with a GS
linker to the membrane protein gene CsgA, allowing the adhesive proteins to anchor on the cell membrane of
Escherichia coli and create CoPlat. Finally, we added the lac promoter for future protein expression. To
prove the following three points, we designed several experiments:
1. Whether the adhesive proteins form on the membrane.
2. Whether CoPlat can maintain adhesion and compare the adhesion of different adhesive proteins to select
the most suitable CoPlat adhesive protein.
3. Whether the adhesive proteins have the ability to capture antibodies or enzymes and similar substances.
Establishing an adhesive standard that can serve as a reference for whether a protein possesses adhesion.
Build
1. Cloning result
2. SDS result
After cloning, we induced the bacteria with 200 μM IPTG for 12 hours to express the proteins. However, when performing SDS-PAGE, we did not observe protein expression. After extensive research and consulting with experts, we speculated that this could be due to the likelihood of membrane proteins having low expression levels, making it difficult to detect whether the target protein has been expressed through SDS-PAGE.
Since SDS-PAGE could not confirm protein expression, we decided to directly confirm the production of the target protein through functional assays. In all of our tests, we assumed that the target gene could reach the cell membrane, and none of our tests resulted in cell disruption.
Test
1. Flushing test
To confirm that CoPlat can adhere to a flat surface, we spread bacterial liquid onto glass slides and cultured it for 30 minutes, allowing the bacteria to attach using adhesive proteins. We then rinsed the slides and compared the results of slides with and without adhesive proteins under a microscope.
We observed that in the presence of bacteria with adhesive proteins, there were more residual bacteria on the slides, demonstrating the effectiveness of adhesive proteins, with the Mrcp20k sequence having the highest adhesion.
2. Viscosity test
In order to recheck whether the CoPlat possesses adhesion, we used a rheometer to measure the shear forces
of different proteins and attempted to quantitatively assess the adhesion of each adhesive protein based
on the results. Additionally, we analyzed which adhesive protein exhibited the highest adhesion strength.
A. First Test: Except for Mrcp20k, no comparison of adhesion of adhesive proteins was observed, and we
suspected that this was due to not switching the vector to a plasmid with higher expression capability,
resulting in lower protein expression. Therefore, we replaced the vector from the originally lower
expression pSB1A3 with pSB3K3 and conducted the viscosity test again.
B. Second Test: It was observed that bacteria with adhesive proteins exhibited higher adhesion compared to the control group(pSB3K3-J04450), with cp19k, Bamcp20k-1, and Mrcp20k possess adhesion.
3. Modified ELISA
Furthermore, we wanted to prove our CoPlat have ability to capture antibodies or related substances, we
tried modifying the ELISA test. We attempted to culture labeled antibodies with CoPlat for 30 minutes
and
added TMB, allowing adhesive proteins that successfully captured antibodies to show color.
A. First Test: To prevent the blocking buffer from being stuck by adhesive proteins, which would hinder
successful TMB color development, we omitted the blocking step and used the Eppendorf method for
testing.
However, we found that the pure PB solution without any adhesive proteins showed brighter than the other
groups, which was contrary to our initial assumption. We suspected that the antibodies adhered to the
walls of the Eppendorf tube during the incubation period, causing TMB coloration.
B. Second Test: To improve the experiment, we transferred the Modified ELISA test to a new Eppendorf tube after removing the antibodies and conducted the test again.
From the coloration results, it can be seen that E. coli with adhesive protein genes can capture labeled antibodies, allowing TMB to show color. In this graph, we can infer that CoPlat with cp19k, Bamcp20k-1, and Mrcp20k have adhesion. This result corresponds to the rheometer test, which indicates that the ability to capture antibodies may relate to the adhesion strength from adhesive protein of CoPlat.
Learn
1. Since all our tests do not conduct cell disruption, yet the function remained normal, it can be
inferred that the natural adhesive recombinant proteins are successfully produced and correctly anchor on
the membrane of E.coli.
2. Based on the results from test1~3, cp19k, Bamcp20k-1, Mrcp20k exhibit adhesion, with Mrcp20k showing
the best performance.
3. From the results of the Modified ELISA test, it is evident that our adhesive proteins can immobilize
antibodies or related substances.
4. We established an adhesion standard that can be used to analyze whether the sample has adhesion.
Cycle 2
Design
We are looking for the possibility to get a better adhesive protein, that is stronger, more adhesive, and easier to produce, so we made an attempt to design a classifier that can find proteins with general adhesive, but its adhesion is not recorded in the protein databases.
Build
To find the potentially adhesive proteins, we built a classifier based on a long short-term memory network(LSTM) with one hot encoding and trained it with the training data we mentioned above.
Test
The model reached a poor accuracy of 0.4.
Learn
We realized that LSTM does not work well with insufficient training data. Therefore, we gave up on the deep learning methods and wanted to build the classifier with a simpler machine learning model.
Cycle 3
Design
To make our classifier more reliable, we delved into the ESM encoding method to determine how to best represent protein sequences effectively. In the previous cycle, we realized that the quantity of training data collected was insufficient to support a deep learning model such as LSTM. As a result, we selected support vector machine as our machine learning method.
To validate the predicted potential adhesive proteins and compare their adhesion with existing adhesive
proteins, we designed biobricks similar to Cycle 1. Specifically, we connected each of the predicted four
potential adhesive proteins (Nid1, ecpA, epd2, zig-4) to the membrane protein gene CsgA using GS linkers,
allowing them to form on the cell membrane of Escherichia coli to create a CoPlat based on the adhesive
proteins. Finally, we added the lac promoter for future protein expression.
Since we established adhesive protein standards in Cycle 1, we attempted to compare the adhesion and
functional tests with existing adhesive proteins. We aimed to determine whether our predicted potential
adhesive proteins were correct. Therefore, we designed functional tests similar to those in Cycle 1.
Build
Once we have a well-defined problem and the dataset in place, we will transition to the build phase. Here,
we will focus on data preprocessing, ensuring that the dataset is clean, any missing values are handled
appropriately, and the ESM-encoded features are processed as needed, such as normalization or scaling.
Next, we will proceed with the implementation of the SVM classifier using a suitable framework,
integrating the ESM encoding to convert protein sequences into feature vectors. Training the model will
involve splitting the dataset into training and validation sets, fine-tuning hyperparameters, and closely
monitoring the model's performance.
Test
1. Flushing Test
To confirm that CoPlat, based on potential adhesive proteins, can adhere to a flat surface, we spread bacterial liquid onto glass slides and cultured it for 30 minutes, allowing the bacteria to attach using potential adhesive proteins. We then rinsed the slides and compared the results of slides under a microscope with and without our predicted adhesive proteins.
We observed that there were more residual bacteria on the slides in the presence of bacteria with adhesive proteins, demonstrating that potential adhesive proteins have adhesion comparable to existing proteins.
2. Viscosity test
In order to recheck whether the CoPlat possesses adhesion, we used a rheometer to quantitatively measure the shear forces of potential adhesive proteins. Additionally, we analyzed which potential adhesive protein exhibited the highest adhesion strength.
We observed that bacteria with potential adhesive proteins exhibited higher adhesion compared to the control group, with adhesion strength in the following order: zig-4 > epd2 > Nid1 > ecpA.
3. Modified ELISA
Furthermore, we wanted to prove our CoPlat have ability to capture antibodies or related substances, we tried modifying the ELISA test. We attempted to culture labeled antibodies with potential adhesive proteins for 30 minutes and added TMB, allowing adhesive proteins that successfully captured antibodies to produce color.
From the results, it can be inferred that zig-4 has adhesion, similar to existing adhesive proteins. However, the other predicted adhesive proteins cannot be verified whether they have adhesion through the modified ELISA test results.
Learn
1. Since all our tests do not conduct cell disruption, yet the function remained normal, it can be
inferred that the potential adhesive recombinant proteins are successfully produced and correctly anchor
on the membrane of E. coli.
2. From the test 1~3 results, it is the evident that potential adhesive proteins possess adhesion, with
adhesion strength in the following order: zig4 > epd2 > Nid1 > ecpA.
3. From the results of the Modified ELISA test, it is the evident that our potential adhesive proteins can
immobilize antibodies or related substances.
Cycle 4
Design
After confirming that CoPlat can be expressed on the membrane, we attempted to link functional proteins to
CoPlat, giving it not only adhesion but also the ability to express additional functions. Therefore, we
used a “10 amino acid flexible protein domain linker” to connect our functional protein to the CoPlat
designed in Cycle 1. The reason we did not use GS linker was to avoid excessively repetitive sequences,
which could complicate sequence synthesis. Additionally, we linked our functional protein before and after
CoPlat, and we attempted to verify which arrangement forms a more stable structure and exhibits better
functionality. We used GFP as a verification method for our functional protein. If GFP can successfully
express green fluorescence, it verifies that the functional protein is functioning normally on CoPlat. To
prove the following three points, we designed several experiments for confirmation:
1. The functional protein can successfully link with CoPlat and be expressed on the cell membrane.
2. It can maintain a certain level of adhesion.
3. The functional protein can function normally.
4. Which arrangement forms a more stable structure and exhibits better functionality.
(In the following text, M5G and GM5 represent the functional protein linked to CoPlat.)
Build
Cloning result: While performing the cloning steps, we consistently observed deletions in M5G during
sequencing. This unsuccessful cloning of M5G can be attributed to the gene mutation, which also proves
that GM5 have more stable structure than M5G.
Test
1. Flushing Test
To confirm that CoPlat can adhere to a flat surface after adding the functional protein, we spread bacterial liquid onto glass slides and cultured it for 30 minutes, allowing the bacteria to attach. We then rinsed the slides and compared the results of slides under a microscope with and without GM5.
We observed that in the presence of bacteria with GM5, there were more residual bacteria on the slides, demonstrating that CoPlat can still maintain adhesion after adding the functional protein.
2. Viscosity Test
In order to recheck whether the CoPlat linked to the functional protein possesses adhesion, we used a rheometer to measure the shear forces of different proteins. We attempted to quantify the adhesion of each adhesive protein based on the results and compared it with existing adhesive proteins to determine if the adhesion matches.
It can be observed that CoPlat with GM5 exhibited higher adhesion compared to the control group, confirming that CoPlat can still maintain adhesion after adding the functional protein.
3. Fluorescence Test
By comparing the green fluorescence intensity at the 545 nm wavelength between pure GFP and GM5, we determined whether the functional protein's activity is reduced due to its connection to CoPlat.
From the data, it is the evident that GM5 has normally expressed the GFP gene. However, its fluorescence intensity is lower than that of the pure GFP strain, indicating that CoPlat still slightly affects the functional protein's functionality.
Learn
1. From the viscosity and flushing test results, it can be concluded that CoPlat, when connected to the
functional protein, can still express itself on the membrane normally and does not affect the inherent
adhesion of CoPlat.
2. From the fluorescence test results, it is evident that CoPlat allows the functional protein to maintain
its functionality normal when linked to the functional protein using the linker.
3. The functional protein linked before CoPlat forms a much more stable structure, which will be the final
arrangement of our CoPlat.